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 affiliates.
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)
46 #define __ex_clear(x, reg) \
47 ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
49 MODULE_AUTHOR("Qumranet");
50 MODULE_LICENSE("GPL");
52 static int __read_mostly bypass_guest_pf = 1;
53 module_param(bypass_guest_pf, bool, S_IRUGO);
55 static int __read_mostly enable_vpid = 1;
56 module_param_named(vpid, enable_vpid, bool, 0444);
58 static int __read_mostly flexpriority_enabled = 1;
59 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
61 static int __read_mostly enable_ept = 1;
62 module_param_named(ept, enable_ept, bool, S_IRUGO);
64 static int __read_mostly enable_unrestricted_guest = 1;
65 module_param_named(unrestricted_guest,
66 enable_unrestricted_guest, bool, S_IRUGO);
68 static int __read_mostly emulate_invalid_guest_state = 0;
69 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
71 static int __read_mostly vmm_exclusive = 1;
72 module_param(vmm_exclusive, bool, S_IRUGO);
74 static int __read_mostly yield_on_hlt = 1;
75 module_param(yield_on_hlt, bool, S_IRUGO);
78 * If nested=1, nested virtualization is supported, i.e., guests may use
79 * VMX and be a hypervisor for its own guests. If nested=0, guests may not
80 * use VMX instructions.
82 static int __read_mostly nested = 0;
83 module_param(nested, bool, S_IRUGO);
85 #define KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST \
86 (X86_CR0_WP | X86_CR0_NE | X86_CR0_NW | X86_CR0_CD)
87 #define KVM_GUEST_CR0_MASK \
88 (KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
89 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST \
90 (X86_CR0_WP | X86_CR0_NE)
91 #define KVM_VM_CR0_ALWAYS_ON \
92 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
93 #define KVM_CR4_GUEST_OWNED_BITS \
94 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
97 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
98 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
100 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
103 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
104 * ple_gap: upper bound on the amount of time between two successive
105 * executions of PAUSE in a loop. Also indicate if ple enabled.
106 * According to test, this time is usually smaller than 128 cycles.
107 * ple_window: upper bound on the amount of time a guest is allowed to execute
108 * in a PAUSE loop. Tests indicate that most spinlocks are held for
109 * less than 2^12 cycles
110 * Time is measured based on a counter that runs at the same rate as the TSC,
111 * refer SDM volume 3b section 21.6.13 & 22.1.3.
113 #define KVM_VMX_DEFAULT_PLE_GAP 128
114 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
115 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
116 module_param(ple_gap, int, S_IRUGO);
118 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
119 module_param(ple_window, int, S_IRUGO);
121 #define NR_AUTOLOAD_MSRS 1
122 #define VMCS02_POOL_SIZE 1
131 * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
132 * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
133 * loaded on this CPU (so we can clear them if the CPU goes down).
139 struct list_head loaded_vmcss_on_cpu_link;
142 struct shared_msr_entry {
149 * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
150 * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
151 * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
152 * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
153 * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
154 * More than one of these structures may exist, if L1 runs multiple L2 guests.
155 * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
156 * underlying hardware which will be used to run L2.
157 * This structure is packed to ensure that its layout is identical across
158 * machines (necessary for live migration).
159 * If there are changes in this struct, VMCS12_REVISION must be changed.
161 typedef u64 natural_width;
162 struct __packed vmcs12 {
163 /* According to the Intel spec, a VMCS region must start with the
164 * following two fields. Then follow implementation-specific data.
169 u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
170 u32 padding[7]; /* room for future expansion */
175 u64 vm_exit_msr_store_addr;
176 u64 vm_exit_msr_load_addr;
177 u64 vm_entry_msr_load_addr;
179 u64 virtual_apic_page_addr;
180 u64 apic_access_addr;
182 u64 guest_physical_address;
183 u64 vmcs_link_pointer;
184 u64 guest_ia32_debugctl;
187 u64 guest_ia32_perf_global_ctrl;
194 u64 host_ia32_perf_global_ctrl;
195 u64 padding64[8]; /* room for future expansion */
197 * To allow migration of L1 (complete with its L2 guests) between
198 * machines of different natural widths (32 or 64 bit), we cannot have
199 * unsigned long fields with no explict size. We use u64 (aliased
200 * natural_width) instead. Luckily, x86 is little-endian.
202 natural_width cr0_guest_host_mask;
203 natural_width cr4_guest_host_mask;
204 natural_width cr0_read_shadow;
205 natural_width cr4_read_shadow;
206 natural_width cr3_target_value0;
207 natural_width cr3_target_value1;
208 natural_width cr3_target_value2;
209 natural_width cr3_target_value3;
210 natural_width exit_qualification;
211 natural_width guest_linear_address;
212 natural_width guest_cr0;
213 natural_width guest_cr3;
214 natural_width guest_cr4;
215 natural_width guest_es_base;
216 natural_width guest_cs_base;
217 natural_width guest_ss_base;
218 natural_width guest_ds_base;
219 natural_width guest_fs_base;
220 natural_width guest_gs_base;
221 natural_width guest_ldtr_base;
222 natural_width guest_tr_base;
223 natural_width guest_gdtr_base;
224 natural_width guest_idtr_base;
225 natural_width guest_dr7;
226 natural_width guest_rsp;
227 natural_width guest_rip;
228 natural_width guest_rflags;
229 natural_width guest_pending_dbg_exceptions;
230 natural_width guest_sysenter_esp;
231 natural_width guest_sysenter_eip;
232 natural_width host_cr0;
233 natural_width host_cr3;
234 natural_width host_cr4;
235 natural_width host_fs_base;
236 natural_width host_gs_base;
237 natural_width host_tr_base;
238 natural_width host_gdtr_base;
239 natural_width host_idtr_base;
240 natural_width host_ia32_sysenter_esp;
241 natural_width host_ia32_sysenter_eip;
242 natural_width host_rsp;
243 natural_width host_rip;
244 natural_width paddingl[8]; /* room for future expansion */
245 u32 pin_based_vm_exec_control;
246 u32 cpu_based_vm_exec_control;
247 u32 exception_bitmap;
248 u32 page_fault_error_code_mask;
249 u32 page_fault_error_code_match;
250 u32 cr3_target_count;
251 u32 vm_exit_controls;
252 u32 vm_exit_msr_store_count;
253 u32 vm_exit_msr_load_count;
254 u32 vm_entry_controls;
255 u32 vm_entry_msr_load_count;
256 u32 vm_entry_intr_info_field;
257 u32 vm_entry_exception_error_code;
258 u32 vm_entry_instruction_len;
260 u32 secondary_vm_exec_control;
261 u32 vm_instruction_error;
263 u32 vm_exit_intr_info;
264 u32 vm_exit_intr_error_code;
265 u32 idt_vectoring_info_field;
266 u32 idt_vectoring_error_code;
267 u32 vm_exit_instruction_len;
268 u32 vmx_instruction_info;
275 u32 guest_ldtr_limit;
277 u32 guest_gdtr_limit;
278 u32 guest_idtr_limit;
279 u32 guest_es_ar_bytes;
280 u32 guest_cs_ar_bytes;
281 u32 guest_ss_ar_bytes;
282 u32 guest_ds_ar_bytes;
283 u32 guest_fs_ar_bytes;
284 u32 guest_gs_ar_bytes;
285 u32 guest_ldtr_ar_bytes;
286 u32 guest_tr_ar_bytes;
287 u32 guest_interruptibility_info;
288 u32 guest_activity_state;
289 u32 guest_sysenter_cs;
290 u32 host_ia32_sysenter_cs;
291 u32 padding32[8]; /* room for future expansion */
292 u16 virtual_processor_id;
293 u16 guest_es_selector;
294 u16 guest_cs_selector;
295 u16 guest_ss_selector;
296 u16 guest_ds_selector;
297 u16 guest_fs_selector;
298 u16 guest_gs_selector;
299 u16 guest_ldtr_selector;
300 u16 guest_tr_selector;
301 u16 host_es_selector;
302 u16 host_cs_selector;
303 u16 host_ss_selector;
304 u16 host_ds_selector;
305 u16 host_fs_selector;
306 u16 host_gs_selector;
307 u16 host_tr_selector;
311 * VMCS12_REVISION is an arbitrary id that should be changed if the content or
312 * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
313 * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
315 #define VMCS12_REVISION 0x11e57ed0
318 * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
319 * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
320 * current implementation, 4K are reserved to avoid future complications.
322 #define VMCS12_SIZE 0x1000
324 /* Used to remember the last vmcs02 used for some recently used vmcs12s */
326 struct list_head list;
328 struct loaded_vmcs vmcs02;
332 * The nested_vmx structure is part of vcpu_vmx, and holds information we need
333 * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
336 /* Has the level1 guest done vmxon? */
339 /* The guest-physical address of the current VMCS L1 keeps for L2 */
341 /* The host-usable pointer to the above */
342 struct page *current_vmcs12_page;
343 struct vmcs12 *current_vmcs12;
345 /* vmcs02_list cache of VMCSs recently used to run L2 guests */
346 struct list_head vmcs02_pool;
348 u64 vmcs01_tsc_offset;
350 * Guest pages referred to in vmcs02 with host-physical pointers, so
351 * we must keep them pinned while L2 runs.
353 struct page *apic_access_page;
357 struct kvm_vcpu vcpu;
358 unsigned long host_rsp;
361 bool nmi_known_unmasked;
363 u32 idt_vectoring_info;
365 struct shared_msr_entry *guest_msrs;
369 u64 msr_host_kernel_gs_base;
370 u64 msr_guest_kernel_gs_base;
373 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
374 * non-nested (L1) guest, it always points to vmcs01. For a nested
375 * guest (L2), it points to a different VMCS.
377 struct loaded_vmcs vmcs01;
378 struct loaded_vmcs *loaded_vmcs;
379 bool __launched; /* temporary, used in vmx_vcpu_run */
380 struct msr_autoload {
382 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
383 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
387 u16 fs_sel, gs_sel, ldt_sel;
388 int gs_ldt_reload_needed;
389 int fs_reload_needed;
394 struct kvm_save_segment {
399 } tr, es, ds, fs, gs;
402 u32 bitmask; /* 4 bits per segment (1 bit per field) */
403 struct kvm_save_segment seg[8];
406 bool emulation_required;
408 /* Support for vnmi-less CPUs */
409 int soft_vnmi_blocked;
411 s64 vnmi_blocked_time;
416 /* Support for a guest hypervisor (nested VMX) */
417 struct nested_vmx nested;
420 enum segment_cache_field {
429 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
431 return container_of(vcpu, struct vcpu_vmx, vcpu);
434 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
435 #define FIELD(number, name) [number] = VMCS12_OFFSET(name)
436 #define FIELD64(number, name) [number] = VMCS12_OFFSET(name), \
437 [number##_HIGH] = VMCS12_OFFSET(name)+4
439 static unsigned short vmcs_field_to_offset_table[] = {
440 FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
441 FIELD(GUEST_ES_SELECTOR, guest_es_selector),
442 FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
443 FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
444 FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
445 FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
446 FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
447 FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
448 FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
449 FIELD(HOST_ES_SELECTOR, host_es_selector),
450 FIELD(HOST_CS_SELECTOR, host_cs_selector),
451 FIELD(HOST_SS_SELECTOR, host_ss_selector),
452 FIELD(HOST_DS_SELECTOR, host_ds_selector),
453 FIELD(HOST_FS_SELECTOR, host_fs_selector),
454 FIELD(HOST_GS_SELECTOR, host_gs_selector),
455 FIELD(HOST_TR_SELECTOR, host_tr_selector),
456 FIELD64(IO_BITMAP_A, io_bitmap_a),
457 FIELD64(IO_BITMAP_B, io_bitmap_b),
458 FIELD64(MSR_BITMAP, msr_bitmap),
459 FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
460 FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
461 FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
462 FIELD64(TSC_OFFSET, tsc_offset),
463 FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
464 FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
465 FIELD64(EPT_POINTER, ept_pointer),
466 FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
467 FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
468 FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
469 FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
470 FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
471 FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
472 FIELD64(GUEST_PDPTR0, guest_pdptr0),
473 FIELD64(GUEST_PDPTR1, guest_pdptr1),
474 FIELD64(GUEST_PDPTR2, guest_pdptr2),
475 FIELD64(GUEST_PDPTR3, guest_pdptr3),
476 FIELD64(HOST_IA32_PAT, host_ia32_pat),
477 FIELD64(HOST_IA32_EFER, host_ia32_efer),
478 FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
479 FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
480 FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
481 FIELD(EXCEPTION_BITMAP, exception_bitmap),
482 FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
483 FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
484 FIELD(CR3_TARGET_COUNT, cr3_target_count),
485 FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
486 FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
487 FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
488 FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
489 FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
490 FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
491 FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
492 FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
493 FIELD(TPR_THRESHOLD, tpr_threshold),
494 FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
495 FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
496 FIELD(VM_EXIT_REASON, vm_exit_reason),
497 FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
498 FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
499 FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
500 FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
501 FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
502 FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
503 FIELD(GUEST_ES_LIMIT, guest_es_limit),
504 FIELD(GUEST_CS_LIMIT, guest_cs_limit),
505 FIELD(GUEST_SS_LIMIT, guest_ss_limit),
506 FIELD(GUEST_DS_LIMIT, guest_ds_limit),
507 FIELD(GUEST_FS_LIMIT, guest_fs_limit),
508 FIELD(GUEST_GS_LIMIT, guest_gs_limit),
509 FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
510 FIELD(GUEST_TR_LIMIT, guest_tr_limit),
511 FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
512 FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
513 FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
514 FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
515 FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
516 FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
517 FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
518 FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
519 FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
520 FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
521 FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
522 FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
523 FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
524 FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
525 FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
526 FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
527 FIELD(CR0_READ_SHADOW, cr0_read_shadow),
528 FIELD(CR4_READ_SHADOW, cr4_read_shadow),
529 FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
530 FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
531 FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
532 FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
533 FIELD(EXIT_QUALIFICATION, exit_qualification),
534 FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
535 FIELD(GUEST_CR0, guest_cr0),
536 FIELD(GUEST_CR3, guest_cr3),
537 FIELD(GUEST_CR4, guest_cr4),
538 FIELD(GUEST_ES_BASE, guest_es_base),
539 FIELD(GUEST_CS_BASE, guest_cs_base),
540 FIELD(GUEST_SS_BASE, guest_ss_base),
541 FIELD(GUEST_DS_BASE, guest_ds_base),
542 FIELD(GUEST_FS_BASE, guest_fs_base),
543 FIELD(GUEST_GS_BASE, guest_gs_base),
544 FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
545 FIELD(GUEST_TR_BASE, guest_tr_base),
546 FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
547 FIELD(GUEST_IDTR_BASE, guest_idtr_base),
548 FIELD(GUEST_DR7, guest_dr7),
549 FIELD(GUEST_RSP, guest_rsp),
550 FIELD(GUEST_RIP, guest_rip),
551 FIELD(GUEST_RFLAGS, guest_rflags),
552 FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
553 FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
554 FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
555 FIELD(HOST_CR0, host_cr0),
556 FIELD(HOST_CR3, host_cr3),
557 FIELD(HOST_CR4, host_cr4),
558 FIELD(HOST_FS_BASE, host_fs_base),
559 FIELD(HOST_GS_BASE, host_gs_base),
560 FIELD(HOST_TR_BASE, host_tr_base),
561 FIELD(HOST_GDTR_BASE, host_gdtr_base),
562 FIELD(HOST_IDTR_BASE, host_idtr_base),
563 FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
564 FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
565 FIELD(HOST_RSP, host_rsp),
566 FIELD(HOST_RIP, host_rip),
568 static const int max_vmcs_field = ARRAY_SIZE(vmcs_field_to_offset_table);
570 static inline short vmcs_field_to_offset(unsigned long field)
572 if (field >= max_vmcs_field || vmcs_field_to_offset_table[field] == 0)
574 return vmcs_field_to_offset_table[field];
577 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
579 return to_vmx(vcpu)->nested.current_vmcs12;
582 static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
584 struct page *page = gfn_to_page(vcpu->kvm, addr >> PAGE_SHIFT);
585 if (is_error_page(page)) {
586 kvm_release_page_clean(page);
592 static void nested_release_page(struct page *page)
594 kvm_release_page_dirty(page);
597 static void nested_release_page_clean(struct page *page)
599 kvm_release_page_clean(page);
602 static u64 construct_eptp(unsigned long root_hpa);
603 static void kvm_cpu_vmxon(u64 addr);
604 static void kvm_cpu_vmxoff(void);
605 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3);
606 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
608 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
609 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
611 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
612 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
614 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
615 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
617 static unsigned long *vmx_io_bitmap_a;
618 static unsigned long *vmx_io_bitmap_b;
619 static unsigned long *vmx_msr_bitmap_legacy;
620 static unsigned long *vmx_msr_bitmap_longmode;
622 static bool cpu_has_load_ia32_efer;
624 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
625 static DEFINE_SPINLOCK(vmx_vpid_lock);
627 static struct vmcs_config {
631 u32 pin_based_exec_ctrl;
632 u32 cpu_based_exec_ctrl;
633 u32 cpu_based_2nd_exec_ctrl;
638 static struct vmx_capability {
643 #define VMX_SEGMENT_FIELD(seg) \
644 [VCPU_SREG_##seg] = { \
645 .selector = GUEST_##seg##_SELECTOR, \
646 .base = GUEST_##seg##_BASE, \
647 .limit = GUEST_##seg##_LIMIT, \
648 .ar_bytes = GUEST_##seg##_AR_BYTES, \
651 static struct kvm_vmx_segment_field {
656 } kvm_vmx_segment_fields[] = {
657 VMX_SEGMENT_FIELD(CS),
658 VMX_SEGMENT_FIELD(DS),
659 VMX_SEGMENT_FIELD(ES),
660 VMX_SEGMENT_FIELD(FS),
661 VMX_SEGMENT_FIELD(GS),
662 VMX_SEGMENT_FIELD(SS),
663 VMX_SEGMENT_FIELD(TR),
664 VMX_SEGMENT_FIELD(LDTR),
667 static u64 host_efer;
669 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
672 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
673 * away by decrementing the array size.
675 static const u32 vmx_msr_index[] = {
677 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
679 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
681 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
683 static inline bool is_page_fault(u32 intr_info)
685 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
686 INTR_INFO_VALID_MASK)) ==
687 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
690 static inline bool is_no_device(u32 intr_info)
692 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
693 INTR_INFO_VALID_MASK)) ==
694 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
697 static inline bool is_invalid_opcode(u32 intr_info)
699 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
700 INTR_INFO_VALID_MASK)) ==
701 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
704 static inline bool is_external_interrupt(u32 intr_info)
706 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
707 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
710 static inline bool is_machine_check(u32 intr_info)
712 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
713 INTR_INFO_VALID_MASK)) ==
714 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
717 static inline bool cpu_has_vmx_msr_bitmap(void)
719 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
722 static inline bool cpu_has_vmx_tpr_shadow(void)
724 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
727 static inline bool vm_need_tpr_shadow(struct kvm *kvm)
729 return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
732 static inline bool cpu_has_secondary_exec_ctrls(void)
734 return vmcs_config.cpu_based_exec_ctrl &
735 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
738 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
740 return vmcs_config.cpu_based_2nd_exec_ctrl &
741 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
744 static inline bool cpu_has_vmx_flexpriority(void)
746 return cpu_has_vmx_tpr_shadow() &&
747 cpu_has_vmx_virtualize_apic_accesses();
750 static inline bool cpu_has_vmx_ept_execute_only(void)
752 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
755 static inline bool cpu_has_vmx_eptp_uncacheable(void)
757 return vmx_capability.ept & VMX_EPTP_UC_BIT;
760 static inline bool cpu_has_vmx_eptp_writeback(void)
762 return vmx_capability.ept & VMX_EPTP_WB_BIT;
765 static inline bool cpu_has_vmx_ept_2m_page(void)
767 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
770 static inline bool cpu_has_vmx_ept_1g_page(void)
772 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
775 static inline bool cpu_has_vmx_ept_4levels(void)
777 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
780 static inline bool cpu_has_vmx_invept_individual_addr(void)
782 return vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT;
785 static inline bool cpu_has_vmx_invept_context(void)
787 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
790 static inline bool cpu_has_vmx_invept_global(void)
792 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
795 static inline bool cpu_has_vmx_invvpid_single(void)
797 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
800 static inline bool cpu_has_vmx_invvpid_global(void)
802 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
805 static inline bool cpu_has_vmx_ept(void)
807 return vmcs_config.cpu_based_2nd_exec_ctrl &
808 SECONDARY_EXEC_ENABLE_EPT;
811 static inline bool cpu_has_vmx_unrestricted_guest(void)
813 return vmcs_config.cpu_based_2nd_exec_ctrl &
814 SECONDARY_EXEC_UNRESTRICTED_GUEST;
817 static inline bool cpu_has_vmx_ple(void)
819 return vmcs_config.cpu_based_2nd_exec_ctrl &
820 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
823 static inline bool vm_need_virtualize_apic_accesses(struct kvm *kvm)
825 return flexpriority_enabled && irqchip_in_kernel(kvm);
828 static inline bool cpu_has_vmx_vpid(void)
830 return vmcs_config.cpu_based_2nd_exec_ctrl &
831 SECONDARY_EXEC_ENABLE_VPID;
834 static inline bool cpu_has_vmx_rdtscp(void)
836 return vmcs_config.cpu_based_2nd_exec_ctrl &
837 SECONDARY_EXEC_RDTSCP;
840 static inline bool cpu_has_virtual_nmis(void)
842 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
845 static inline bool cpu_has_vmx_wbinvd_exit(void)
847 return vmcs_config.cpu_based_2nd_exec_ctrl &
848 SECONDARY_EXEC_WBINVD_EXITING;
851 static inline bool report_flexpriority(void)
853 return flexpriority_enabled;
856 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
858 return vmcs12->cpu_based_vm_exec_control & bit;
861 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
863 return (vmcs12->cpu_based_vm_exec_control &
864 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
865 (vmcs12->secondary_vm_exec_control & bit);
868 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
869 struct vmcs12 *vmcs12,
870 u32 reason, unsigned long qualification);
872 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
876 for (i = 0; i < vmx->nmsrs; ++i)
877 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
882 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
888 } operand = { vpid, 0, gva };
890 asm volatile (__ex(ASM_VMX_INVVPID)
891 /* CF==1 or ZF==1 --> rc = -1 */
893 : : "a"(&operand), "c"(ext) : "cc", "memory");
896 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
900 } operand = {eptp, gpa};
902 asm volatile (__ex(ASM_VMX_INVEPT)
903 /* CF==1 or ZF==1 --> rc = -1 */
904 "; ja 1f ; ud2 ; 1:\n"
905 : : "a" (&operand), "c" (ext) : "cc", "memory");
908 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
912 i = __find_msr_index(vmx, msr);
914 return &vmx->guest_msrs[i];
918 static void vmcs_clear(struct vmcs *vmcs)
920 u64 phys_addr = __pa(vmcs);
923 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
924 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
927 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
931 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
933 vmcs_clear(loaded_vmcs->vmcs);
934 loaded_vmcs->cpu = -1;
935 loaded_vmcs->launched = 0;
938 static void vmcs_load(struct vmcs *vmcs)
940 u64 phys_addr = __pa(vmcs);
943 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
944 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
947 printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
951 static void __loaded_vmcs_clear(void *arg)
953 struct loaded_vmcs *loaded_vmcs = arg;
954 int cpu = raw_smp_processor_id();
956 if (loaded_vmcs->cpu != cpu)
957 return; /* vcpu migration can race with cpu offline */
958 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
959 per_cpu(current_vmcs, cpu) = NULL;
960 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
961 loaded_vmcs_init(loaded_vmcs);
964 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
966 if (loaded_vmcs->cpu != -1)
967 smp_call_function_single(
968 loaded_vmcs->cpu, __loaded_vmcs_clear, loaded_vmcs, 1);
971 static inline void vpid_sync_vcpu_single(struct vcpu_vmx *vmx)
976 if (cpu_has_vmx_invvpid_single())
977 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
980 static inline void vpid_sync_vcpu_global(void)
982 if (cpu_has_vmx_invvpid_global())
983 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
986 static inline void vpid_sync_context(struct vcpu_vmx *vmx)
988 if (cpu_has_vmx_invvpid_single())
989 vpid_sync_vcpu_single(vmx);
991 vpid_sync_vcpu_global();
994 static inline void ept_sync_global(void)
996 if (cpu_has_vmx_invept_global())
997 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1000 static inline void ept_sync_context(u64 eptp)
1003 if (cpu_has_vmx_invept_context())
1004 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1010 static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
1013 if (cpu_has_vmx_invept_individual_addr())
1014 __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
1017 ept_sync_context(eptp);
1021 static __always_inline unsigned long vmcs_readl(unsigned long field)
1023 unsigned long value;
1025 asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1026 : "=a"(value) : "d"(field) : "cc");
1030 static __always_inline u16 vmcs_read16(unsigned long field)
1032 return vmcs_readl(field);
1035 static __always_inline u32 vmcs_read32(unsigned long field)
1037 return vmcs_readl(field);
1040 static __always_inline u64 vmcs_read64(unsigned long field)
1042 #ifdef CONFIG_X86_64
1043 return vmcs_readl(field);
1045 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
1049 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1051 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1052 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1056 static void vmcs_writel(unsigned long field, unsigned long value)
1060 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1061 : "=q"(error) : "a"(value), "d"(field) : "cc");
1062 if (unlikely(error))
1063 vmwrite_error(field, value);
1066 static void vmcs_write16(unsigned long field, u16 value)
1068 vmcs_writel(field, value);
1071 static void vmcs_write32(unsigned long field, u32 value)
1073 vmcs_writel(field, value);
1076 static void vmcs_write64(unsigned long field, u64 value)
1078 vmcs_writel(field, value);
1079 #ifndef CONFIG_X86_64
1081 vmcs_writel(field+1, value >> 32);
1085 static void vmcs_clear_bits(unsigned long field, u32 mask)
1087 vmcs_writel(field, vmcs_readl(field) & ~mask);
1090 static void vmcs_set_bits(unsigned long field, u32 mask)
1092 vmcs_writel(field, vmcs_readl(field) | mask);
1095 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1097 vmx->segment_cache.bitmask = 0;
1100 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1104 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1106 if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1107 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1108 vmx->segment_cache.bitmask = 0;
1110 ret = vmx->segment_cache.bitmask & mask;
1111 vmx->segment_cache.bitmask |= mask;
1115 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1117 u16 *p = &vmx->segment_cache.seg[seg].selector;
1119 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1120 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1124 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1126 ulong *p = &vmx->segment_cache.seg[seg].base;
1128 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1129 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1133 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1135 u32 *p = &vmx->segment_cache.seg[seg].limit;
1137 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1138 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1142 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1144 u32 *p = &vmx->segment_cache.seg[seg].ar;
1146 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1147 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1151 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1155 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1156 (1u << NM_VECTOR) | (1u << DB_VECTOR);
1157 if ((vcpu->guest_debug &
1158 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1159 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1160 eb |= 1u << BP_VECTOR;
1161 if (to_vmx(vcpu)->rmode.vm86_active)
1164 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1165 if (vcpu->fpu_active)
1166 eb &= ~(1u << NM_VECTOR);
1167 vmcs_write32(EXCEPTION_BITMAP, eb);
1170 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1173 struct msr_autoload *m = &vmx->msr_autoload;
1175 if (msr == MSR_EFER && cpu_has_load_ia32_efer) {
1176 vmcs_clear_bits(VM_ENTRY_CONTROLS, VM_ENTRY_LOAD_IA32_EFER);
1177 vmcs_clear_bits(VM_EXIT_CONTROLS, VM_EXIT_LOAD_IA32_EFER);
1181 for (i = 0; i < m->nr; ++i)
1182 if (m->guest[i].index == msr)
1188 m->guest[i] = m->guest[m->nr];
1189 m->host[i] = m->host[m->nr];
1190 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1191 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1194 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1195 u64 guest_val, u64 host_val)
1198 struct msr_autoload *m = &vmx->msr_autoload;
1200 if (msr == MSR_EFER && cpu_has_load_ia32_efer) {
1201 vmcs_write64(GUEST_IA32_EFER, guest_val);
1202 vmcs_write64(HOST_IA32_EFER, host_val);
1203 vmcs_set_bits(VM_ENTRY_CONTROLS, VM_ENTRY_LOAD_IA32_EFER);
1204 vmcs_set_bits(VM_EXIT_CONTROLS, VM_EXIT_LOAD_IA32_EFER);
1208 for (i = 0; i < m->nr; ++i)
1209 if (m->guest[i].index == msr)
1214 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1215 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1218 m->guest[i].index = msr;
1219 m->guest[i].value = guest_val;
1220 m->host[i].index = msr;
1221 m->host[i].value = host_val;
1224 static void reload_tss(void)
1227 * VT restores TR but not its size. Useless.
1229 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1230 struct desc_struct *descs;
1232 descs = (void *)gdt->address;
1233 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
1237 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
1242 guest_efer = vmx->vcpu.arch.efer;
1245 * NX is emulated; LMA and LME handled by hardware; SCE meaninless
1248 ignore_bits = EFER_NX | EFER_SCE;
1249 #ifdef CONFIG_X86_64
1250 ignore_bits |= EFER_LMA | EFER_LME;
1251 /* SCE is meaningful only in long mode on Intel */
1252 if (guest_efer & EFER_LMA)
1253 ignore_bits &= ~(u64)EFER_SCE;
1255 guest_efer &= ~ignore_bits;
1256 guest_efer |= host_efer & ignore_bits;
1257 vmx->guest_msrs[efer_offset].data = guest_efer;
1258 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1260 clear_atomic_switch_msr(vmx, MSR_EFER);
1261 /* On ept, can't emulate nx, and must switch nx atomically */
1262 if (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX)) {
1263 guest_efer = vmx->vcpu.arch.efer;
1264 if (!(guest_efer & EFER_LMA))
1265 guest_efer &= ~EFER_LME;
1266 add_atomic_switch_msr(vmx, MSR_EFER, guest_efer, host_efer);
1273 static unsigned long segment_base(u16 selector)
1275 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1276 struct desc_struct *d;
1277 unsigned long table_base;
1280 if (!(selector & ~3))
1283 table_base = gdt->address;
1285 if (selector & 4) { /* from ldt */
1286 u16 ldt_selector = kvm_read_ldt();
1288 if (!(ldt_selector & ~3))
1291 table_base = segment_base(ldt_selector);
1293 d = (struct desc_struct *)(table_base + (selector & ~7));
1294 v = get_desc_base(d);
1295 #ifdef CONFIG_X86_64
1296 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
1297 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
1302 static inline unsigned long kvm_read_tr_base(void)
1305 asm("str %0" : "=g"(tr));
1306 return segment_base(tr);
1309 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
1311 struct vcpu_vmx *vmx = to_vmx(vcpu);
1314 if (vmx->host_state.loaded)
1317 vmx->host_state.loaded = 1;
1319 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
1320 * allow segment selectors with cpl > 0 or ti == 1.
1322 vmx->host_state.ldt_sel = kvm_read_ldt();
1323 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
1324 savesegment(fs, vmx->host_state.fs_sel);
1325 if (!(vmx->host_state.fs_sel & 7)) {
1326 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
1327 vmx->host_state.fs_reload_needed = 0;
1329 vmcs_write16(HOST_FS_SELECTOR, 0);
1330 vmx->host_state.fs_reload_needed = 1;
1332 savesegment(gs, vmx->host_state.gs_sel);
1333 if (!(vmx->host_state.gs_sel & 7))
1334 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
1336 vmcs_write16(HOST_GS_SELECTOR, 0);
1337 vmx->host_state.gs_ldt_reload_needed = 1;
1340 #ifdef CONFIG_X86_64
1341 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
1342 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
1344 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
1345 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
1348 #ifdef CONFIG_X86_64
1349 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1350 if (is_long_mode(&vmx->vcpu))
1351 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1353 for (i = 0; i < vmx->save_nmsrs; ++i)
1354 kvm_set_shared_msr(vmx->guest_msrs[i].index,
1355 vmx->guest_msrs[i].data,
1356 vmx->guest_msrs[i].mask);
1359 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
1361 if (!vmx->host_state.loaded)
1364 ++vmx->vcpu.stat.host_state_reload;
1365 vmx->host_state.loaded = 0;
1366 #ifdef CONFIG_X86_64
1367 if (is_long_mode(&vmx->vcpu))
1368 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1370 if (vmx->host_state.gs_ldt_reload_needed) {
1371 kvm_load_ldt(vmx->host_state.ldt_sel);
1372 #ifdef CONFIG_X86_64
1373 load_gs_index(vmx->host_state.gs_sel);
1375 loadsegment(gs, vmx->host_state.gs_sel);
1378 if (vmx->host_state.fs_reload_needed)
1379 loadsegment(fs, vmx->host_state.fs_sel);
1381 #ifdef CONFIG_X86_64
1382 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1384 if (current_thread_info()->status & TS_USEDFPU)
1386 load_gdt(&__get_cpu_var(host_gdt));
1389 static void vmx_load_host_state(struct vcpu_vmx *vmx)
1392 __vmx_load_host_state(vmx);
1397 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1398 * vcpu mutex is already taken.
1400 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1402 struct vcpu_vmx *vmx = to_vmx(vcpu);
1403 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1406 kvm_cpu_vmxon(phys_addr);
1407 else if (vmx->loaded_vmcs->cpu != cpu)
1408 loaded_vmcs_clear(vmx->loaded_vmcs);
1410 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1411 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1412 vmcs_load(vmx->loaded_vmcs->vmcs);
1415 if (vmx->loaded_vmcs->cpu != cpu) {
1416 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1417 unsigned long sysenter_esp;
1419 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1420 local_irq_disable();
1421 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1422 &per_cpu(loaded_vmcss_on_cpu, cpu));
1426 * Linux uses per-cpu TSS and GDT, so set these when switching
1429 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
1430 vmcs_writel(HOST_GDTR_BASE, gdt->address); /* 22.2.4 */
1432 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1433 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1434 vmx->loaded_vmcs->cpu = cpu;
1438 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1440 __vmx_load_host_state(to_vmx(vcpu));
1441 if (!vmm_exclusive) {
1442 __loaded_vmcs_clear(to_vmx(vcpu)->loaded_vmcs);
1448 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
1452 if (vcpu->fpu_active)
1454 vcpu->fpu_active = 1;
1455 cr0 = vmcs_readl(GUEST_CR0);
1456 cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
1457 cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
1458 vmcs_writel(GUEST_CR0, cr0);
1459 update_exception_bitmap(vcpu);
1460 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
1461 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1464 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
1467 * Return the cr0 value that a nested guest would read. This is a combination
1468 * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
1469 * its hypervisor (cr0_read_shadow).
1471 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
1473 return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
1474 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
1476 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
1478 return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
1479 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
1482 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
1484 vmx_decache_cr0_guest_bits(vcpu);
1485 vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
1486 update_exception_bitmap(vcpu);
1487 vcpu->arch.cr0_guest_owned_bits = 0;
1488 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1489 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
1492 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1494 unsigned long rflags, save_rflags;
1496 if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
1497 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1498 rflags = vmcs_readl(GUEST_RFLAGS);
1499 if (to_vmx(vcpu)->rmode.vm86_active) {
1500 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1501 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
1502 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1504 to_vmx(vcpu)->rflags = rflags;
1506 return to_vmx(vcpu)->rflags;
1509 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1511 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1512 __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
1513 to_vmx(vcpu)->rflags = rflags;
1514 if (to_vmx(vcpu)->rmode.vm86_active) {
1515 to_vmx(vcpu)->rmode.save_rflags = rflags;
1516 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1518 vmcs_writel(GUEST_RFLAGS, rflags);
1521 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1523 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1526 if (interruptibility & GUEST_INTR_STATE_STI)
1527 ret |= KVM_X86_SHADOW_INT_STI;
1528 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1529 ret |= KVM_X86_SHADOW_INT_MOV_SS;
1534 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1536 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1537 u32 interruptibility = interruptibility_old;
1539 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1541 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1542 interruptibility |= GUEST_INTR_STATE_MOV_SS;
1543 else if (mask & KVM_X86_SHADOW_INT_STI)
1544 interruptibility |= GUEST_INTR_STATE_STI;
1546 if ((interruptibility != interruptibility_old))
1547 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1550 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
1554 rip = kvm_rip_read(vcpu);
1555 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1556 kvm_rip_write(vcpu, rip);
1558 /* skipping an emulated instruction also counts */
1559 vmx_set_interrupt_shadow(vcpu, 0);
1562 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1564 /* Ensure that we clear the HLT state in the VMCS. We don't need to
1565 * explicitly skip the instruction because if the HLT state is set, then
1566 * the instruction is already executing and RIP has already been
1568 if (!yield_on_hlt &&
1569 vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1570 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1573 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
1574 bool has_error_code, u32 error_code,
1577 struct vcpu_vmx *vmx = to_vmx(vcpu);
1578 u32 intr_info = nr | INTR_INFO_VALID_MASK;
1580 if (has_error_code) {
1581 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1582 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1585 if (vmx->rmode.vm86_active) {
1587 if (kvm_exception_is_soft(nr))
1588 inc_eip = vcpu->arch.event_exit_inst_len;
1589 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
1590 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
1594 if (kvm_exception_is_soft(nr)) {
1595 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1596 vmx->vcpu.arch.event_exit_inst_len);
1597 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1599 intr_info |= INTR_TYPE_HARD_EXCEPTION;
1601 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1602 vmx_clear_hlt(vcpu);
1605 static bool vmx_rdtscp_supported(void)
1607 return cpu_has_vmx_rdtscp();
1611 * Swap MSR entry in host/guest MSR entry array.
1613 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1615 struct shared_msr_entry tmp;
1617 tmp = vmx->guest_msrs[to];
1618 vmx->guest_msrs[to] = vmx->guest_msrs[from];
1619 vmx->guest_msrs[from] = tmp;
1623 * Set up the vmcs to automatically save and restore system
1624 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
1625 * mode, as fiddling with msrs is very expensive.
1627 static void setup_msrs(struct vcpu_vmx *vmx)
1629 int save_nmsrs, index;
1630 unsigned long *msr_bitmap;
1632 vmx_load_host_state(vmx);
1634 #ifdef CONFIG_X86_64
1635 if (is_long_mode(&vmx->vcpu)) {
1636 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1638 move_msr_up(vmx, index, save_nmsrs++);
1639 index = __find_msr_index(vmx, MSR_LSTAR);
1641 move_msr_up(vmx, index, save_nmsrs++);
1642 index = __find_msr_index(vmx, MSR_CSTAR);
1644 move_msr_up(vmx, index, save_nmsrs++);
1645 index = __find_msr_index(vmx, MSR_TSC_AUX);
1646 if (index >= 0 && vmx->rdtscp_enabled)
1647 move_msr_up(vmx, index, save_nmsrs++);
1649 * MSR_STAR is only needed on long mode guests, and only
1650 * if efer.sce is enabled.
1652 index = __find_msr_index(vmx, MSR_STAR);
1653 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
1654 move_msr_up(vmx, index, save_nmsrs++);
1657 index = __find_msr_index(vmx, MSR_EFER);
1658 if (index >= 0 && update_transition_efer(vmx, index))
1659 move_msr_up(vmx, index, save_nmsrs++);
1661 vmx->save_nmsrs = save_nmsrs;
1663 if (cpu_has_vmx_msr_bitmap()) {
1664 if (is_long_mode(&vmx->vcpu))
1665 msr_bitmap = vmx_msr_bitmap_longmode;
1667 msr_bitmap = vmx_msr_bitmap_legacy;
1669 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
1674 * reads and returns guest's timestamp counter "register"
1675 * guest_tsc = host_tsc + tsc_offset -- 21.3
1677 static u64 guest_read_tsc(void)
1679 u64 host_tsc, tsc_offset;
1682 tsc_offset = vmcs_read64(TSC_OFFSET);
1683 return host_tsc + tsc_offset;
1687 * Empty call-back. Needs to be implemented when VMX enables the SET_TSC_KHZ
1688 * ioctl. In this case the call-back should update internal vmx state to make
1689 * the changes effective.
1691 static void vmx_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1693 /* Nothing to do here */
1697 * writes 'offset' into guest's timestamp counter offset register
1699 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1701 vmcs_write64(TSC_OFFSET, offset);
1704 static void vmx_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment)
1706 u64 offset = vmcs_read64(TSC_OFFSET);
1707 vmcs_write64(TSC_OFFSET, offset + adjustment);
1710 static u64 vmx_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1712 return target_tsc - native_read_tsc();
1715 static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
1717 struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
1718 return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
1722 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1723 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1724 * all guests if the "nested" module option is off, and can also be disabled
1725 * for a single guest by disabling its VMX cpuid bit.
1727 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1729 return nested && guest_cpuid_has_vmx(vcpu);
1733 * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
1734 * returned for the various VMX controls MSRs when nested VMX is enabled.
1735 * The same values should also be used to verify that vmcs12 control fields are
1736 * valid during nested entry from L1 to L2.
1737 * Each of these control msrs has a low and high 32-bit half: A low bit is on
1738 * if the corresponding bit in the (32-bit) control field *must* be on, and a
1739 * bit in the high half is on if the corresponding bit in the control field
1740 * may be on. See also vmx_control_verify().
1741 * TODO: allow these variables to be modified (downgraded) by module options
1744 static u32 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high;
1745 static u32 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high;
1746 static u32 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high;
1747 static u32 nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high;
1748 static u32 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high;
1749 static __init void nested_vmx_setup_ctls_msrs(void)
1752 * Note that as a general rule, the high half of the MSRs (bits in
1753 * the control fields which may be 1) should be initialized by the
1754 * intersection of the underlying hardware's MSR (i.e., features which
1755 * can be supported) and the list of features we want to expose -
1756 * because they are known to be properly supported in our code.
1757 * Also, usually, the low half of the MSRs (bits which must be 1) can
1758 * be set to 0, meaning that L1 may turn off any of these bits. The
1759 * reason is that if one of these bits is necessary, it will appear
1760 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
1761 * fields of vmcs01 and vmcs02, will turn these bits off - and
1762 * nested_vmx_exit_handled() will not pass related exits to L1.
1763 * These rules have exceptions below.
1766 /* pin-based controls */
1768 * According to the Intel spec, if bit 55 of VMX_BASIC is off (as it is
1769 * in our case), bits 1, 2 and 4 (i.e., 0x16) must be 1 in this MSR.
1771 nested_vmx_pinbased_ctls_low = 0x16 ;
1772 nested_vmx_pinbased_ctls_high = 0x16 |
1773 PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING |
1774 PIN_BASED_VIRTUAL_NMIS;
1777 nested_vmx_exit_ctls_low = 0;
1778 #ifdef CONFIG_X86_64
1779 nested_vmx_exit_ctls_high = VM_EXIT_HOST_ADDR_SPACE_SIZE;
1781 nested_vmx_exit_ctls_high = 0;
1784 /* entry controls */
1785 rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
1786 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high);
1787 nested_vmx_entry_ctls_low = 0;
1788 nested_vmx_entry_ctls_high &=
1789 VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_IA32E_MODE;
1791 /* cpu-based controls */
1792 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
1793 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high);
1794 nested_vmx_procbased_ctls_low = 0;
1795 nested_vmx_procbased_ctls_high &=
1796 CPU_BASED_VIRTUAL_INTR_PENDING | CPU_BASED_USE_TSC_OFFSETING |
1797 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
1798 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
1799 CPU_BASED_CR3_STORE_EXITING |
1800 #ifdef CONFIG_X86_64
1801 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
1803 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
1804 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_EXITING |
1805 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1807 * We can allow some features even when not supported by the
1808 * hardware. For example, L1 can specify an MSR bitmap - and we
1809 * can use it to avoid exits to L1 - even when L0 runs L2
1810 * without MSR bitmaps.
1812 nested_vmx_procbased_ctls_high |= CPU_BASED_USE_MSR_BITMAPS;
1814 /* secondary cpu-based controls */
1815 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
1816 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high);
1817 nested_vmx_secondary_ctls_low = 0;
1818 nested_vmx_secondary_ctls_high &=
1819 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1822 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
1825 * Bits 0 in high must be 0, and bits 1 in low must be 1.
1827 return ((control & high) | low) == control;
1830 static inline u64 vmx_control_msr(u32 low, u32 high)
1832 return low | ((u64)high << 32);
1836 * If we allow our guest to use VMX instructions (i.e., nested VMX), we should
1837 * also let it use VMX-specific MSRs.
1838 * vmx_get_vmx_msr() and vmx_set_vmx_msr() return 1 when we handled a
1839 * VMX-specific MSR, or 0 when we haven't (and the caller should handle it
1840 * like all other MSRs).
1842 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1844 if (!nested_vmx_allowed(vcpu) && msr_index >= MSR_IA32_VMX_BASIC &&
1845 msr_index <= MSR_IA32_VMX_TRUE_ENTRY_CTLS) {
1847 * According to the spec, processors which do not support VMX
1848 * should throw a #GP(0) when VMX capability MSRs are read.
1850 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
1854 switch (msr_index) {
1855 case MSR_IA32_FEATURE_CONTROL:
1858 case MSR_IA32_VMX_BASIC:
1860 * This MSR reports some information about VMX support. We
1861 * should return information about the VMX we emulate for the
1862 * guest, and the VMCS structure we give it - not about the
1863 * VMX support of the underlying hardware.
1865 *pdata = VMCS12_REVISION |
1866 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
1867 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
1869 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1870 case MSR_IA32_VMX_PINBASED_CTLS:
1871 *pdata = vmx_control_msr(nested_vmx_pinbased_ctls_low,
1872 nested_vmx_pinbased_ctls_high);
1874 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1875 case MSR_IA32_VMX_PROCBASED_CTLS:
1876 *pdata = vmx_control_msr(nested_vmx_procbased_ctls_low,
1877 nested_vmx_procbased_ctls_high);
1879 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1880 case MSR_IA32_VMX_EXIT_CTLS:
1881 *pdata = vmx_control_msr(nested_vmx_exit_ctls_low,
1882 nested_vmx_exit_ctls_high);
1884 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1885 case MSR_IA32_VMX_ENTRY_CTLS:
1886 *pdata = vmx_control_msr(nested_vmx_entry_ctls_low,
1887 nested_vmx_entry_ctls_high);
1889 case MSR_IA32_VMX_MISC:
1893 * These MSRs specify bits which the guest must keep fixed (on or off)
1894 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
1895 * We picked the standard core2 setting.
1897 #define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
1898 #define VMXON_CR4_ALWAYSON X86_CR4_VMXE
1899 case MSR_IA32_VMX_CR0_FIXED0:
1900 *pdata = VMXON_CR0_ALWAYSON;
1902 case MSR_IA32_VMX_CR0_FIXED1:
1905 case MSR_IA32_VMX_CR4_FIXED0:
1906 *pdata = VMXON_CR4_ALWAYSON;
1908 case MSR_IA32_VMX_CR4_FIXED1:
1911 case MSR_IA32_VMX_VMCS_ENUM:
1914 case MSR_IA32_VMX_PROCBASED_CTLS2:
1915 *pdata = vmx_control_msr(nested_vmx_secondary_ctls_low,
1916 nested_vmx_secondary_ctls_high);
1918 case MSR_IA32_VMX_EPT_VPID_CAP:
1919 /* Currently, no nested ept or nested vpid */
1929 static int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1931 if (!nested_vmx_allowed(vcpu))
1934 if (msr_index == MSR_IA32_FEATURE_CONTROL)
1935 /* TODO: the right thing. */
1938 * No need to treat VMX capability MSRs specially: If we don't handle
1939 * them, handle_wrmsr will #GP(0), which is correct (they are readonly)
1945 * Reads an msr value (of 'msr_index') into 'pdata'.
1946 * Returns 0 on success, non-0 otherwise.
1947 * Assumes vcpu_load() was already called.
1949 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1952 struct shared_msr_entry *msr;
1955 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
1959 switch (msr_index) {
1960 #ifdef CONFIG_X86_64
1962 data = vmcs_readl(GUEST_FS_BASE);
1965 data = vmcs_readl(GUEST_GS_BASE);
1967 case MSR_KERNEL_GS_BASE:
1968 vmx_load_host_state(to_vmx(vcpu));
1969 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
1973 return kvm_get_msr_common(vcpu, msr_index, pdata);
1975 data = guest_read_tsc();
1977 case MSR_IA32_SYSENTER_CS:
1978 data = vmcs_read32(GUEST_SYSENTER_CS);
1980 case MSR_IA32_SYSENTER_EIP:
1981 data = vmcs_readl(GUEST_SYSENTER_EIP);
1983 case MSR_IA32_SYSENTER_ESP:
1984 data = vmcs_readl(GUEST_SYSENTER_ESP);
1987 if (!to_vmx(vcpu)->rdtscp_enabled)
1989 /* Otherwise falls through */
1991 vmx_load_host_state(to_vmx(vcpu));
1992 if (vmx_get_vmx_msr(vcpu, msr_index, pdata))
1994 msr = find_msr_entry(to_vmx(vcpu), msr_index);
1996 vmx_load_host_state(to_vmx(vcpu));
2000 return kvm_get_msr_common(vcpu, msr_index, pdata);
2008 * Writes msr value into into the appropriate "register".
2009 * Returns 0 on success, non-0 otherwise.
2010 * Assumes vcpu_load() was already called.
2012 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
2014 struct vcpu_vmx *vmx = to_vmx(vcpu);
2015 struct shared_msr_entry *msr;
2018 switch (msr_index) {
2020 vmx_load_host_state(vmx);
2021 ret = kvm_set_msr_common(vcpu, msr_index, data);
2023 #ifdef CONFIG_X86_64
2025 vmx_segment_cache_clear(vmx);
2026 vmcs_writel(GUEST_FS_BASE, data);
2029 vmx_segment_cache_clear(vmx);
2030 vmcs_writel(GUEST_GS_BASE, data);
2032 case MSR_KERNEL_GS_BASE:
2033 vmx_load_host_state(vmx);
2034 vmx->msr_guest_kernel_gs_base = data;
2037 case MSR_IA32_SYSENTER_CS:
2038 vmcs_write32(GUEST_SYSENTER_CS, data);
2040 case MSR_IA32_SYSENTER_EIP:
2041 vmcs_writel(GUEST_SYSENTER_EIP, data);
2043 case MSR_IA32_SYSENTER_ESP:
2044 vmcs_writel(GUEST_SYSENTER_ESP, data);
2047 kvm_write_tsc(vcpu, data);
2049 case MSR_IA32_CR_PAT:
2050 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2051 vmcs_write64(GUEST_IA32_PAT, data);
2052 vcpu->arch.pat = data;
2055 ret = kvm_set_msr_common(vcpu, msr_index, data);
2058 if (!vmx->rdtscp_enabled)
2060 /* Check reserved bit, higher 32 bits should be zero */
2061 if ((data >> 32) != 0)
2063 /* Otherwise falls through */
2065 if (vmx_set_vmx_msr(vcpu, msr_index, data))
2067 msr = find_msr_entry(vmx, msr_index);
2069 vmx_load_host_state(vmx);
2073 ret = kvm_set_msr_common(vcpu, msr_index, data);
2079 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2081 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
2084 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2087 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2089 case VCPU_EXREG_PDPTR:
2091 ept_save_pdptrs(vcpu);
2098 static void set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
2100 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
2101 vmcs_writel(GUEST_DR7, dbg->arch.debugreg[7]);
2103 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
2105 update_exception_bitmap(vcpu);
2108 static __init int cpu_has_kvm_support(void)
2110 return cpu_has_vmx();
2113 static __init int vmx_disabled_by_bios(void)
2117 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
2118 if (msr & FEATURE_CONTROL_LOCKED) {
2119 /* launched w/ TXT and VMX disabled */
2120 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2123 /* launched w/o TXT and VMX only enabled w/ TXT */
2124 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2125 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2126 && !tboot_enabled()) {
2127 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
2128 "activate TXT before enabling KVM\n");
2131 /* launched w/o TXT and VMX disabled */
2132 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2133 && !tboot_enabled())
2140 static void kvm_cpu_vmxon(u64 addr)
2142 asm volatile (ASM_VMX_VMXON_RAX
2143 : : "a"(&addr), "m"(addr)
2147 static int hardware_enable(void *garbage)
2149 int cpu = raw_smp_processor_id();
2150 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2153 if (read_cr4() & X86_CR4_VMXE)
2156 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2157 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
2159 test_bits = FEATURE_CONTROL_LOCKED;
2160 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
2161 if (tboot_enabled())
2162 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
2164 if ((old & test_bits) != test_bits) {
2165 /* enable and lock */
2166 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
2168 write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
2170 if (vmm_exclusive) {
2171 kvm_cpu_vmxon(phys_addr);
2175 store_gdt(&__get_cpu_var(host_gdt));
2180 static void vmclear_local_loaded_vmcss(void)
2182 int cpu = raw_smp_processor_id();
2183 struct loaded_vmcs *v, *n;
2185 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2186 loaded_vmcss_on_cpu_link)
2187 __loaded_vmcs_clear(v);
2191 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2194 static void kvm_cpu_vmxoff(void)
2196 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
2199 static void hardware_disable(void *garbage)
2201 if (vmm_exclusive) {
2202 vmclear_local_loaded_vmcss();
2205 write_cr4(read_cr4() & ~X86_CR4_VMXE);
2208 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2209 u32 msr, u32 *result)
2211 u32 vmx_msr_low, vmx_msr_high;
2212 u32 ctl = ctl_min | ctl_opt;
2214 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2216 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2217 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
2219 /* Ensure minimum (required) set of control bits are supported. */
2227 static __init bool allow_1_setting(u32 msr, u32 ctl)
2229 u32 vmx_msr_low, vmx_msr_high;
2231 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2232 return vmx_msr_high & ctl;
2235 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
2237 u32 vmx_msr_low, vmx_msr_high;
2238 u32 min, opt, min2, opt2;
2239 u32 _pin_based_exec_control = 0;
2240 u32 _cpu_based_exec_control = 0;
2241 u32 _cpu_based_2nd_exec_control = 0;
2242 u32 _vmexit_control = 0;
2243 u32 _vmentry_control = 0;
2245 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2246 opt = PIN_BASED_VIRTUAL_NMIS;
2247 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2248 &_pin_based_exec_control) < 0)
2252 #ifdef CONFIG_X86_64
2253 CPU_BASED_CR8_LOAD_EXITING |
2254 CPU_BASED_CR8_STORE_EXITING |
2256 CPU_BASED_CR3_LOAD_EXITING |
2257 CPU_BASED_CR3_STORE_EXITING |
2258 CPU_BASED_USE_IO_BITMAPS |
2259 CPU_BASED_MOV_DR_EXITING |
2260 CPU_BASED_USE_TSC_OFFSETING |
2261 CPU_BASED_MWAIT_EXITING |
2262 CPU_BASED_MONITOR_EXITING |
2263 CPU_BASED_INVLPG_EXITING;
2266 min |= CPU_BASED_HLT_EXITING;
2268 opt = CPU_BASED_TPR_SHADOW |
2269 CPU_BASED_USE_MSR_BITMAPS |
2270 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2271 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2272 &_cpu_based_exec_control) < 0)
2274 #ifdef CONFIG_X86_64
2275 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2276 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2277 ~CPU_BASED_CR8_STORE_EXITING;
2279 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2281 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2282 SECONDARY_EXEC_WBINVD_EXITING |
2283 SECONDARY_EXEC_ENABLE_VPID |
2284 SECONDARY_EXEC_ENABLE_EPT |
2285 SECONDARY_EXEC_UNRESTRICTED_GUEST |
2286 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2287 SECONDARY_EXEC_RDTSCP;
2288 if (adjust_vmx_controls(min2, opt2,
2289 MSR_IA32_VMX_PROCBASED_CTLS2,
2290 &_cpu_based_2nd_exec_control) < 0)
2293 #ifndef CONFIG_X86_64
2294 if (!(_cpu_based_2nd_exec_control &
2295 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2296 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2298 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2299 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2301 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2302 CPU_BASED_CR3_STORE_EXITING |
2303 CPU_BASED_INVLPG_EXITING);
2304 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
2305 vmx_capability.ept, vmx_capability.vpid);
2309 #ifdef CONFIG_X86_64
2310 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2312 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
2313 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2314 &_vmexit_control) < 0)
2318 opt = VM_ENTRY_LOAD_IA32_PAT;
2319 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2320 &_vmentry_control) < 0)
2323 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2325 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2326 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2329 #ifdef CONFIG_X86_64
2330 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2331 if (vmx_msr_high & (1u<<16))
2335 /* Require Write-Back (WB) memory type for VMCS accesses. */
2336 if (((vmx_msr_high >> 18) & 15) != 6)
2339 vmcs_conf->size = vmx_msr_high & 0x1fff;
2340 vmcs_conf->order = get_order(vmcs_config.size);
2341 vmcs_conf->revision_id = vmx_msr_low;
2343 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2344 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2345 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2346 vmcs_conf->vmexit_ctrl = _vmexit_control;
2347 vmcs_conf->vmentry_ctrl = _vmentry_control;
2349 cpu_has_load_ia32_efer =
2350 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2351 VM_ENTRY_LOAD_IA32_EFER)
2352 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2353 VM_EXIT_LOAD_IA32_EFER);
2358 static struct vmcs *alloc_vmcs_cpu(int cpu)
2360 int node = cpu_to_node(cpu);
2364 pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
2367 vmcs = page_address(pages);
2368 memset(vmcs, 0, vmcs_config.size);
2369 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
2373 static struct vmcs *alloc_vmcs(void)
2375 return alloc_vmcs_cpu(raw_smp_processor_id());
2378 static void free_vmcs(struct vmcs *vmcs)
2380 free_pages((unsigned long)vmcs, vmcs_config.order);
2384 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2386 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2388 if (!loaded_vmcs->vmcs)
2390 loaded_vmcs_clear(loaded_vmcs);
2391 free_vmcs(loaded_vmcs->vmcs);
2392 loaded_vmcs->vmcs = NULL;
2395 static void free_kvm_area(void)
2399 for_each_possible_cpu(cpu) {
2400 free_vmcs(per_cpu(vmxarea, cpu));
2401 per_cpu(vmxarea, cpu) = NULL;
2405 static __init int alloc_kvm_area(void)
2409 for_each_possible_cpu(cpu) {
2412 vmcs = alloc_vmcs_cpu(cpu);
2418 per_cpu(vmxarea, cpu) = vmcs;
2423 static __init int hardware_setup(void)
2425 if (setup_vmcs_config(&vmcs_config) < 0)
2428 if (boot_cpu_has(X86_FEATURE_NX))
2429 kvm_enable_efer_bits(EFER_NX);
2431 if (!cpu_has_vmx_vpid())
2434 if (!cpu_has_vmx_ept() ||
2435 !cpu_has_vmx_ept_4levels()) {
2437 enable_unrestricted_guest = 0;
2440 if (!cpu_has_vmx_unrestricted_guest())
2441 enable_unrestricted_guest = 0;
2443 if (!cpu_has_vmx_flexpriority())
2444 flexpriority_enabled = 0;
2446 if (!cpu_has_vmx_tpr_shadow())
2447 kvm_x86_ops->update_cr8_intercept = NULL;
2449 if (enable_ept && !cpu_has_vmx_ept_2m_page())
2450 kvm_disable_largepages();
2452 if (!cpu_has_vmx_ple())
2456 nested_vmx_setup_ctls_msrs();
2458 return alloc_kvm_area();
2461 static __exit void hardware_unsetup(void)
2466 static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
2468 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2470 if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
2471 vmcs_write16(sf->selector, save->selector);
2472 vmcs_writel(sf->base, save->base);
2473 vmcs_write32(sf->limit, save->limit);
2474 vmcs_write32(sf->ar_bytes, save->ar);
2476 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
2478 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
2482 static void enter_pmode(struct kvm_vcpu *vcpu)
2484 unsigned long flags;
2485 struct vcpu_vmx *vmx = to_vmx(vcpu);
2487 vmx->emulation_required = 1;
2488 vmx->rmode.vm86_active = 0;
2490 vmx_segment_cache_clear(vmx);
2492 vmcs_write16(GUEST_TR_SELECTOR, vmx->rmode.tr.selector);
2493 vmcs_writel(GUEST_TR_BASE, vmx->rmode.tr.base);
2494 vmcs_write32(GUEST_TR_LIMIT, vmx->rmode.tr.limit);
2495 vmcs_write32(GUEST_TR_AR_BYTES, vmx->rmode.tr.ar);
2497 flags = vmcs_readl(GUEST_RFLAGS);
2498 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2499 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2500 vmcs_writel(GUEST_RFLAGS, flags);
2502 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2503 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2505 update_exception_bitmap(vcpu);
2507 if (emulate_invalid_guest_state)
2510 fix_pmode_dataseg(VCPU_SREG_ES, &vmx->rmode.es);
2511 fix_pmode_dataseg(VCPU_SREG_DS, &vmx->rmode.ds);
2512 fix_pmode_dataseg(VCPU_SREG_GS, &vmx->rmode.gs);
2513 fix_pmode_dataseg(VCPU_SREG_FS, &vmx->rmode.fs);
2515 vmx_segment_cache_clear(vmx);
2517 vmcs_write16(GUEST_SS_SELECTOR, 0);
2518 vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
2520 vmcs_write16(GUEST_CS_SELECTOR,
2521 vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
2522 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
2525 static gva_t rmode_tss_base(struct kvm *kvm)
2527 if (!kvm->arch.tss_addr) {
2528 struct kvm_memslots *slots;
2531 slots = kvm_memslots(kvm);
2532 base_gfn = slots->memslots[0].base_gfn +
2533 kvm->memslots->memslots[0].npages - 3;
2534 return base_gfn << PAGE_SHIFT;
2536 return kvm->arch.tss_addr;
2539 static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
2541 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2543 save->selector = vmcs_read16(sf->selector);
2544 save->base = vmcs_readl(sf->base);
2545 save->limit = vmcs_read32(sf->limit);
2546 save->ar = vmcs_read32(sf->ar_bytes);
2547 vmcs_write16(sf->selector, save->base >> 4);
2548 vmcs_write32(sf->base, save->base & 0xffff0);
2549 vmcs_write32(sf->limit, 0xffff);
2550 vmcs_write32(sf->ar_bytes, 0xf3);
2551 if (save->base & 0xf)
2552 printk_once(KERN_WARNING "kvm: segment base is not paragraph"
2553 " aligned when entering protected mode (seg=%d)",
2557 static void enter_rmode(struct kvm_vcpu *vcpu)
2559 unsigned long flags;
2560 struct vcpu_vmx *vmx = to_vmx(vcpu);
2562 if (enable_unrestricted_guest)
2565 vmx->emulation_required = 1;
2566 vmx->rmode.vm86_active = 1;
2569 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2570 * vcpu. Call it here with phys address pointing 16M below 4G.
2572 if (!vcpu->kvm->arch.tss_addr) {
2573 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2574 "called before entering vcpu\n");
2575 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
2576 vmx_set_tss_addr(vcpu->kvm, 0xfeffd000);
2577 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2580 vmx_segment_cache_clear(vmx);
2582 vmx->rmode.tr.selector = vmcs_read16(GUEST_TR_SELECTOR);
2583 vmx->rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
2584 vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
2586 vmx->rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
2587 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2589 vmx->rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
2590 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2592 flags = vmcs_readl(GUEST_RFLAGS);
2593 vmx->rmode.save_rflags = flags;
2595 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2597 vmcs_writel(GUEST_RFLAGS, flags);
2598 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2599 update_exception_bitmap(vcpu);
2601 if (emulate_invalid_guest_state)
2602 goto continue_rmode;
2604 vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
2605 vmcs_write32(GUEST_SS_LIMIT, 0xffff);
2606 vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
2608 vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
2609 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
2610 if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
2611 vmcs_writel(GUEST_CS_BASE, 0xf0000);
2612 vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
2614 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.es);
2615 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.ds);
2616 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.gs);
2617 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.fs);
2620 kvm_mmu_reset_context(vcpu);
2623 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2625 struct vcpu_vmx *vmx = to_vmx(vcpu);
2626 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2632 * Force kernel_gs_base reloading before EFER changes, as control
2633 * of this msr depends on is_long_mode().
2635 vmx_load_host_state(to_vmx(vcpu));
2636 vcpu->arch.efer = efer;
2637 if (efer & EFER_LMA) {
2638 vmcs_write32(VM_ENTRY_CONTROLS,
2639 vmcs_read32(VM_ENTRY_CONTROLS) |
2640 VM_ENTRY_IA32E_MODE);
2643 vmcs_write32(VM_ENTRY_CONTROLS,
2644 vmcs_read32(VM_ENTRY_CONTROLS) &
2645 ~VM_ENTRY_IA32E_MODE);
2647 msr->data = efer & ~EFER_LME;
2652 #ifdef CONFIG_X86_64
2654 static void enter_lmode(struct kvm_vcpu *vcpu)
2658 vmx_segment_cache_clear(to_vmx(vcpu));
2660 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2661 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
2662 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
2664 vmcs_write32(GUEST_TR_AR_BYTES,
2665 (guest_tr_ar & ~AR_TYPE_MASK)
2666 | AR_TYPE_BUSY_64_TSS);
2668 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2671 static void exit_lmode(struct kvm_vcpu *vcpu)
2673 vmcs_write32(VM_ENTRY_CONTROLS,
2674 vmcs_read32(VM_ENTRY_CONTROLS)
2675 & ~VM_ENTRY_IA32E_MODE);
2676 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2681 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
2683 vpid_sync_context(to_vmx(vcpu));
2685 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
2687 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
2691 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2693 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2695 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
2696 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
2699 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
2701 if (enable_ept && is_paging(vcpu))
2702 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2703 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
2706 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2708 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2710 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
2711 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
2714 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2716 if (!test_bit(VCPU_EXREG_PDPTR,
2717 (unsigned long *)&vcpu->arch.regs_dirty))
2720 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
2721 vmcs_write64(GUEST_PDPTR0, vcpu->arch.mmu.pdptrs[0]);
2722 vmcs_write64(GUEST_PDPTR1, vcpu->arch.mmu.pdptrs[1]);
2723 vmcs_write64(GUEST_PDPTR2, vcpu->arch.mmu.pdptrs[2]);
2724 vmcs_write64(GUEST_PDPTR3, vcpu->arch.mmu.pdptrs[3]);
2728 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2730 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
2731 vcpu->arch.mmu.pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2732 vcpu->arch.mmu.pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2733 vcpu->arch.mmu.pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2734 vcpu->arch.mmu.pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2737 __set_bit(VCPU_EXREG_PDPTR,
2738 (unsigned long *)&vcpu->arch.regs_avail);
2739 __set_bit(VCPU_EXREG_PDPTR,
2740 (unsigned long *)&vcpu->arch.regs_dirty);
2743 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
2745 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
2747 struct kvm_vcpu *vcpu)
2749 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
2750 vmx_decache_cr3(vcpu);
2751 if (!(cr0 & X86_CR0_PG)) {
2752 /* From paging/starting to nonpaging */
2753 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
2754 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
2755 (CPU_BASED_CR3_LOAD_EXITING |
2756 CPU_BASED_CR3_STORE_EXITING));
2757 vcpu->arch.cr0 = cr0;
2758 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2759 } else if (!is_paging(vcpu)) {
2760 /* From nonpaging to paging */
2761 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
2762 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
2763 ~(CPU_BASED_CR3_LOAD_EXITING |
2764 CPU_BASED_CR3_STORE_EXITING));
2765 vcpu->arch.cr0 = cr0;
2766 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2769 if (!(cr0 & X86_CR0_WP))
2770 *hw_cr0 &= ~X86_CR0_WP;
2773 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2775 struct vcpu_vmx *vmx = to_vmx(vcpu);
2776 unsigned long hw_cr0;
2778 if (enable_unrestricted_guest)
2779 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST)
2780 | KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
2782 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON;
2784 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
2787 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
2790 #ifdef CONFIG_X86_64
2791 if (vcpu->arch.efer & EFER_LME) {
2792 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
2794 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
2800 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
2802 if (!vcpu->fpu_active)
2803 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
2805 vmcs_writel(CR0_READ_SHADOW, cr0);
2806 vmcs_writel(GUEST_CR0, hw_cr0);
2807 vcpu->arch.cr0 = cr0;
2808 __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
2811 static u64 construct_eptp(unsigned long root_hpa)
2815 /* TODO write the value reading from MSR */
2816 eptp = VMX_EPT_DEFAULT_MT |
2817 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
2818 eptp |= (root_hpa & PAGE_MASK);
2823 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
2825 unsigned long guest_cr3;
2830 eptp = construct_eptp(cr3);
2831 vmcs_write64(EPT_POINTER, eptp);
2832 guest_cr3 = is_paging(vcpu) ? kvm_read_cr3(vcpu) :
2833 vcpu->kvm->arch.ept_identity_map_addr;
2834 ept_load_pdptrs(vcpu);
2837 vmx_flush_tlb(vcpu);
2838 vmcs_writel(GUEST_CR3, guest_cr3);
2841 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
2843 unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
2844 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
2846 if (cr4 & X86_CR4_VMXE) {
2848 * To use VMXON (and later other VMX instructions), a guest
2849 * must first be able to turn on cr4.VMXE (see handle_vmon()).
2850 * So basically the check on whether to allow nested VMX
2853 if (!nested_vmx_allowed(vcpu))
2855 } else if (to_vmx(vcpu)->nested.vmxon)
2858 vcpu->arch.cr4 = cr4;
2860 if (!is_paging(vcpu)) {
2861 hw_cr4 &= ~X86_CR4_PAE;
2862 hw_cr4 |= X86_CR4_PSE;
2863 } else if (!(cr4 & X86_CR4_PAE)) {
2864 hw_cr4 &= ~X86_CR4_PAE;
2868 vmcs_writel(CR4_READ_SHADOW, cr4);
2869 vmcs_writel(GUEST_CR4, hw_cr4);
2873 static void vmx_get_segment(struct kvm_vcpu *vcpu,
2874 struct kvm_segment *var, int seg)
2876 struct vcpu_vmx *vmx = to_vmx(vcpu);
2877 struct kvm_save_segment *save;
2880 if (vmx->rmode.vm86_active
2881 && (seg == VCPU_SREG_TR || seg == VCPU_SREG_ES
2882 || seg == VCPU_SREG_DS || seg == VCPU_SREG_FS
2883 || seg == VCPU_SREG_GS)
2884 && !emulate_invalid_guest_state) {
2886 case VCPU_SREG_TR: save = &vmx->rmode.tr; break;
2887 case VCPU_SREG_ES: save = &vmx->rmode.es; break;
2888 case VCPU_SREG_DS: save = &vmx->rmode.ds; break;
2889 case VCPU_SREG_FS: save = &vmx->rmode.fs; break;
2890 case VCPU_SREG_GS: save = &vmx->rmode.gs; break;
2893 var->selector = save->selector;
2894 var->base = save->base;
2895 var->limit = save->limit;
2897 if (seg == VCPU_SREG_TR
2898 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
2899 goto use_saved_rmode_seg;
2901 var->base = vmx_read_guest_seg_base(vmx, seg);
2902 var->limit = vmx_read_guest_seg_limit(vmx, seg);
2903 var->selector = vmx_read_guest_seg_selector(vmx, seg);
2904 ar = vmx_read_guest_seg_ar(vmx, seg);
2905 use_saved_rmode_seg:
2906 if ((ar & AR_UNUSABLE_MASK) && !emulate_invalid_guest_state)
2908 var->type = ar & 15;
2909 var->s = (ar >> 4) & 1;
2910 var->dpl = (ar >> 5) & 3;
2911 var->present = (ar >> 7) & 1;
2912 var->avl = (ar >> 12) & 1;
2913 var->l = (ar >> 13) & 1;
2914 var->db = (ar >> 14) & 1;
2915 var->g = (ar >> 15) & 1;
2916 var->unusable = (ar >> 16) & 1;
2919 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
2921 struct kvm_segment s;
2923 if (to_vmx(vcpu)->rmode.vm86_active) {
2924 vmx_get_segment(vcpu, &s, seg);
2927 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
2930 static int __vmx_get_cpl(struct kvm_vcpu *vcpu)
2932 if (!is_protmode(vcpu))
2935 if (!is_long_mode(vcpu)
2936 && (kvm_get_rflags(vcpu) & X86_EFLAGS_VM)) /* if virtual 8086 */
2939 return vmx_read_guest_seg_selector(to_vmx(vcpu), VCPU_SREG_CS) & 3;
2942 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
2944 if (!test_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail)) {
2945 __set_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
2946 to_vmx(vcpu)->cpl = __vmx_get_cpl(vcpu);
2948 return to_vmx(vcpu)->cpl;
2952 static u32 vmx_segment_access_rights(struct kvm_segment *var)
2959 ar = var->type & 15;
2960 ar |= (var->s & 1) << 4;
2961 ar |= (var->dpl & 3) << 5;
2962 ar |= (var->present & 1) << 7;
2963 ar |= (var->avl & 1) << 12;
2964 ar |= (var->l & 1) << 13;
2965 ar |= (var->db & 1) << 14;
2966 ar |= (var->g & 1) << 15;
2968 if (ar == 0) /* a 0 value means unusable */
2969 ar = AR_UNUSABLE_MASK;
2974 static void vmx_set_segment(struct kvm_vcpu *vcpu,
2975 struct kvm_segment *var, int seg)
2977 struct vcpu_vmx *vmx = to_vmx(vcpu);
2978 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2981 vmx_segment_cache_clear(vmx);
2983 if (vmx->rmode.vm86_active && seg == VCPU_SREG_TR) {
2984 vmcs_write16(sf->selector, var->selector);
2985 vmx->rmode.tr.selector = var->selector;
2986 vmx->rmode.tr.base = var->base;
2987 vmx->rmode.tr.limit = var->limit;
2988 vmx->rmode.tr.ar = vmx_segment_access_rights(var);
2991 vmcs_writel(sf->base, var->base);
2992 vmcs_write32(sf->limit, var->limit);
2993 vmcs_write16(sf->selector, var->selector);
2994 if (vmx->rmode.vm86_active && var->s) {
2996 * Hack real-mode segments into vm86 compatibility.
2998 if (var->base == 0xffff0000 && var->selector == 0xf000)
2999 vmcs_writel(sf->base, 0xf0000);
3002 ar = vmx_segment_access_rights(var);
3005 * Fix the "Accessed" bit in AR field of segment registers for older
3007 * IA32 arch specifies that at the time of processor reset the
3008 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3009 * is setting it to 0 in the usedland code. This causes invalid guest
3010 * state vmexit when "unrestricted guest" mode is turned on.
3011 * Fix for this setup issue in cpu_reset is being pushed in the qemu
3012 * tree. Newer qemu binaries with that qemu fix would not need this
3015 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3016 ar |= 0x1; /* Accessed */
3018 vmcs_write32(sf->ar_bytes, ar);
3019 __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
3022 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3024 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3026 *db = (ar >> 14) & 1;
3027 *l = (ar >> 13) & 1;
3030 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3032 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3033 dt->address = vmcs_readl(GUEST_IDTR_BASE);
3036 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3038 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3039 vmcs_writel(GUEST_IDTR_BASE, dt->address);
3042 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3044 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3045 dt->address = vmcs_readl(GUEST_GDTR_BASE);
3048 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3050 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3051 vmcs_writel(GUEST_GDTR_BASE, dt->address);
3054 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3056 struct kvm_segment var;
3059 vmx_get_segment(vcpu, &var, seg);
3060 ar = vmx_segment_access_rights(&var);
3062 if (var.base != (var.selector << 4))
3064 if (var.limit != 0xffff)
3072 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3074 struct kvm_segment cs;
3075 unsigned int cs_rpl;
3077 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3078 cs_rpl = cs.selector & SELECTOR_RPL_MASK;
3082 if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
3086 if (cs.type & AR_TYPE_WRITEABLE_MASK) {
3087 if (cs.dpl > cs_rpl)
3090 if (cs.dpl != cs_rpl)
3096 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3100 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3102 struct kvm_segment ss;
3103 unsigned int ss_rpl;
3105 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3106 ss_rpl = ss.selector & SELECTOR_RPL_MASK;
3110 if (ss.type != 3 && ss.type != 7)
3114 if (ss.dpl != ss_rpl) /* DPL != RPL */
3122 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3124 struct kvm_segment var;
3127 vmx_get_segment(vcpu, &var, seg);
3128 rpl = var.selector & SELECTOR_RPL_MASK;
3136 if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
3137 if (var.dpl < rpl) /* DPL < RPL */
3141 /* TODO: Add other members to kvm_segment_field to allow checking for other access
3147 static bool tr_valid(struct kvm_vcpu *vcpu)
3149 struct kvm_segment tr;
3151 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3155 if (tr.selector & SELECTOR_TI_MASK) /* TI = 1 */
3157 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3165 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3167 struct kvm_segment ldtr;
3169 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3173 if (ldtr.selector & SELECTOR_TI_MASK) /* TI = 1 */
3183 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3185 struct kvm_segment cs, ss;
3187 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3188 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3190 return ((cs.selector & SELECTOR_RPL_MASK) ==
3191 (ss.selector & SELECTOR_RPL_MASK));
3195 * Check if guest state is valid. Returns true if valid, false if
3197 * We assume that registers are always usable
3199 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3201 /* real mode guest state checks */
3202 if (!is_protmode(vcpu)) {
3203 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3205 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3207 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3209 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3211 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3213 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3216 /* protected mode guest state checks */
3217 if (!cs_ss_rpl_check(vcpu))
3219 if (!code_segment_valid(vcpu))
3221 if (!stack_segment_valid(vcpu))
3223 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3225 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3227 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3229 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3231 if (!tr_valid(vcpu))
3233 if (!ldtr_valid(vcpu))
3237 * - Add checks on RIP
3238 * - Add checks on RFLAGS
3244 static int init_rmode_tss(struct kvm *kvm)
3248 int r, idx, ret = 0;
3250 idx = srcu_read_lock(&kvm->srcu);
3251 fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
3252 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3255 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3256 r = kvm_write_guest_page(kvm, fn++, &data,
3257 TSS_IOPB_BASE_OFFSET, sizeof(u16));
3260 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3263 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3267 r = kvm_write_guest_page(kvm, fn, &data,
3268 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3275 srcu_read_unlock(&kvm->srcu, idx);
3279 static int init_rmode_identity_map(struct kvm *kvm)
3282 pfn_t identity_map_pfn;
3287 if (unlikely(!kvm->arch.ept_identity_pagetable)) {
3288 printk(KERN_ERR "EPT: identity-mapping pagetable "
3289 "haven't been allocated!\n");
3292 if (likely(kvm->arch.ept_identity_pagetable_done))
3295 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
3296 idx = srcu_read_lock(&kvm->srcu);
3297 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3300 /* Set up identity-mapping pagetable for EPT in real mode */
3301 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3302 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3303 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3304 r = kvm_write_guest_page(kvm, identity_map_pfn,
3305 &tmp, i * sizeof(tmp), sizeof(tmp));
3309 kvm->arch.ept_identity_pagetable_done = true;
3312 srcu_read_unlock(&kvm->srcu, idx);
3316 static void seg_setup(int seg)
3318 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3321 vmcs_write16(sf->selector, 0);
3322 vmcs_writel(sf->base, 0);
3323 vmcs_write32(sf->limit, 0xffff);
3324 if (enable_unrestricted_guest) {
3326 if (seg == VCPU_SREG_CS)
3327 ar |= 0x08; /* code segment */
3331 vmcs_write32(sf->ar_bytes, ar);
3334 static int alloc_apic_access_page(struct kvm *kvm)
3336 struct kvm_userspace_memory_region kvm_userspace_mem;
3339 mutex_lock(&kvm->slots_lock);
3340 if (kvm->arch.apic_access_page)
3342 kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
3343 kvm_userspace_mem.flags = 0;
3344 kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
3345 kvm_userspace_mem.memory_size = PAGE_SIZE;
3346 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
3350 kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
3352 mutex_unlock(&kvm->slots_lock);
3356 static int alloc_identity_pagetable(struct kvm *kvm)
3358 struct kvm_userspace_memory_region kvm_userspace_mem;
3361 mutex_lock(&kvm->slots_lock);
3362 if (kvm->arch.ept_identity_pagetable)
3364 kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
3365 kvm_userspace_mem.flags = 0;
3366 kvm_userspace_mem.guest_phys_addr =
3367 kvm->arch.ept_identity_map_addr;
3368 kvm_userspace_mem.memory_size = PAGE_SIZE;
3369 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
3373 kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
3374 kvm->arch.ept_identity_map_addr >> PAGE_SHIFT);
3376 mutex_unlock(&kvm->slots_lock);
3380 static void allocate_vpid(struct vcpu_vmx *vmx)
3387 spin_lock(&vmx_vpid_lock);
3388 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3389 if (vpid < VMX_NR_VPIDS) {
3391 __set_bit(vpid, vmx_vpid_bitmap);
3393 spin_unlock(&vmx_vpid_lock);
3396 static void free_vpid(struct vcpu_vmx *vmx)
3400 spin_lock(&vmx_vpid_lock);
3402 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
3403 spin_unlock(&vmx_vpid_lock);
3406 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, u32 msr)
3408 int f = sizeof(unsigned long);
3410 if (!cpu_has_vmx_msr_bitmap())
3414 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3415 * have the write-low and read-high bitmap offsets the wrong way round.
3416 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3418 if (msr <= 0x1fff) {
3419 __clear_bit(msr, msr_bitmap + 0x000 / f); /* read-low */
3420 __clear_bit(msr, msr_bitmap + 0x800 / f); /* write-low */
3421 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3423 __clear_bit(msr, msr_bitmap + 0x400 / f); /* read-high */
3424 __clear_bit(msr, msr_bitmap + 0xc00 / f); /* write-high */
3428 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
3431 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy, msr);
3432 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode, msr);
3436 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3437 * will not change in the lifetime of the guest.
3438 * Note that host-state that does change is set elsewhere. E.g., host-state
3439 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3441 static void vmx_set_constant_host_state(void)
3447 vmcs_writel(HOST_CR0, read_cr0() | X86_CR0_TS); /* 22.2.3 */
3448 vmcs_writel(HOST_CR4, read_cr4()); /* 22.2.3, 22.2.5 */
3449 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
3451 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
3452 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3453 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3454 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3455 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
3457 native_store_idt(&dt);
3458 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
3460 asm("mov $.Lkvm_vmx_return, %0" : "=r"(tmpl));
3461 vmcs_writel(HOST_RIP, tmpl); /* 22.2.5 */
3463 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
3464 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
3465 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
3466 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
3468 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
3469 rdmsr(MSR_IA32_CR_PAT, low32, high32);
3470 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
3474 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
3476 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
3478 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
3479 if (is_guest_mode(&vmx->vcpu))
3480 vmx->vcpu.arch.cr4_guest_owned_bits &=
3481 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
3482 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
3485 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
3487 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
3488 if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
3489 exec_control &= ~CPU_BASED_TPR_SHADOW;
3490 #ifdef CONFIG_X86_64
3491 exec_control |= CPU_BASED_CR8_STORE_EXITING |
3492 CPU_BASED_CR8_LOAD_EXITING;
3496 exec_control |= CPU_BASED_CR3_STORE_EXITING |
3497 CPU_BASED_CR3_LOAD_EXITING |
3498 CPU_BASED_INVLPG_EXITING;
3499 return exec_control;
3502 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
3504 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
3505 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
3506 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
3508 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
3510 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
3511 enable_unrestricted_guest = 0;
3513 if (!enable_unrestricted_guest)
3514 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
3516 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
3517 return exec_control;
3521 * Sets up the vmcs for emulated real mode.
3523 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
3529 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
3530 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
3532 if (cpu_has_vmx_msr_bitmap())
3533 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
3535 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
3538 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
3539 vmcs_config.pin_based_exec_ctrl);
3541 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
3543 if (cpu_has_secondary_exec_ctrls()) {
3544 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
3545 vmx_secondary_exec_control(vmx));
3549 vmcs_write32(PLE_GAP, ple_gap);
3550 vmcs_write32(PLE_WINDOW, ple_window);
3553 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
3554 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
3555 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
3557 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
3558 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
3559 vmx_set_constant_host_state();
3560 #ifdef CONFIG_X86_64
3561 rdmsrl(MSR_FS_BASE, a);
3562 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
3563 rdmsrl(MSR_GS_BASE, a);
3564 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
3566 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
3567 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
3570 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
3571 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
3572 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
3573 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
3574 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
3576 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
3577 u32 msr_low, msr_high;
3579 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
3580 host_pat = msr_low | ((u64) msr_high << 32);
3581 /* Write the default value follow host pat */
3582 vmcs_write64(GUEST_IA32_PAT, host_pat);
3583 /* Keep arch.pat sync with GUEST_IA32_PAT */
3584 vmx->vcpu.arch.pat = host_pat;
3587 for (i = 0; i < NR_VMX_MSR; ++i) {
3588 u32 index = vmx_msr_index[i];
3589 u32 data_low, data_high;
3592 if (rdmsr_safe(index, &data_low, &data_high) < 0)
3594 if (wrmsr_safe(index, data_low, data_high) < 0)
3596 vmx->guest_msrs[j].index = i;
3597 vmx->guest_msrs[j].data = 0;
3598 vmx->guest_msrs[j].mask = -1ull;
3602 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
3604 /* 22.2.1, 20.8.1 */
3605 vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
3607 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
3608 set_cr4_guest_host_mask(vmx);
3610 kvm_write_tsc(&vmx->vcpu, 0);
3615 static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
3617 struct vcpu_vmx *vmx = to_vmx(vcpu);
3621 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
3623 vmx->rmode.vm86_active = 0;
3625 vmx->soft_vnmi_blocked = 0;
3627 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
3628 kvm_set_cr8(&vmx->vcpu, 0);
3629 msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
3630 if (kvm_vcpu_is_bsp(&vmx->vcpu))
3631 msr |= MSR_IA32_APICBASE_BSP;
3632 kvm_set_apic_base(&vmx->vcpu, msr);
3634 ret = fx_init(&vmx->vcpu);
3638 vmx_segment_cache_clear(vmx);
3640 seg_setup(VCPU_SREG_CS);
3642 * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
3643 * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4. Sigh.
3645 if (kvm_vcpu_is_bsp(&vmx->vcpu)) {
3646 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
3647 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
3649 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
3650 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
3653 seg_setup(VCPU_SREG_DS);
3654 seg_setup(VCPU_SREG_ES);
3655 seg_setup(VCPU_SREG_FS);
3656 seg_setup(VCPU_SREG_GS);
3657 seg_setup(VCPU_SREG_SS);
3659 vmcs_write16(GUEST_TR_SELECTOR, 0);
3660 vmcs_writel(GUEST_TR_BASE, 0);
3661 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
3662 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3664 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
3665 vmcs_writel(GUEST_LDTR_BASE, 0);
3666 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
3667 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
3669 vmcs_write32(GUEST_SYSENTER_CS, 0);
3670 vmcs_writel(GUEST_SYSENTER_ESP, 0);
3671 vmcs_writel(GUEST_SYSENTER_EIP, 0);
3673 vmcs_writel(GUEST_RFLAGS, 0x02);
3674 if (kvm_vcpu_is_bsp(&vmx->vcpu))
3675 kvm_rip_write(vcpu, 0xfff0);
3677 kvm_rip_write(vcpu, 0);
3678 kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
3680 vmcs_writel(GUEST_DR7, 0x400);
3682 vmcs_writel(GUEST_GDTR_BASE, 0);
3683 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
3685 vmcs_writel(GUEST_IDTR_BASE, 0);
3686 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
3688 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
3689 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
3690 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
3692 /* Special registers */
3693 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
3697 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
3699 if (cpu_has_vmx_tpr_shadow()) {
3700 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
3701 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
3702 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
3703 __pa(vmx->vcpu.arch.apic->regs));
3704 vmcs_write32(TPR_THRESHOLD, 0);
3707 if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
3708 vmcs_write64(APIC_ACCESS_ADDR,
3709 page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
3712 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
3714 vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
3715 vmx_set_cr0(&vmx->vcpu, kvm_read_cr0(vcpu)); /* enter rmode */
3716 vmx_set_cr4(&vmx->vcpu, 0);
3717 vmx_set_efer(&vmx->vcpu, 0);
3718 vmx_fpu_activate(&vmx->vcpu);
3719 update_exception_bitmap(&vmx->vcpu);
3721 vpid_sync_context(vmx);
3725 /* HACK: Don't enable emulation on guest boot/reset */
3726 vmx->emulation_required = 0;
3732 static void enable_irq_window(struct kvm_vcpu *vcpu)
3734 u32 cpu_based_vm_exec_control;
3736 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3737 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
3738 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3741 static void enable_nmi_window(struct kvm_vcpu *vcpu)
3743 u32 cpu_based_vm_exec_control;
3745 if (!cpu_has_virtual_nmis()) {
3746 enable_irq_window(vcpu);
3750 if (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
3751 enable_irq_window(vcpu);
3754 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3755 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
3756 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3759 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
3761 struct vcpu_vmx *vmx = to_vmx(vcpu);
3763 int irq = vcpu->arch.interrupt.nr;
3765 trace_kvm_inj_virq(irq);
3767 ++vcpu->stat.irq_injections;
3768 if (vmx->rmode.vm86_active) {
3770 if (vcpu->arch.interrupt.soft)
3771 inc_eip = vcpu->arch.event_exit_inst_len;
3772 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
3773 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3776 intr = irq | INTR_INFO_VALID_MASK;
3777 if (vcpu->arch.interrupt.soft) {
3778 intr |= INTR_TYPE_SOFT_INTR;
3779 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
3780 vmx->vcpu.arch.event_exit_inst_len);
3782 intr |= INTR_TYPE_EXT_INTR;
3783 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
3784 vmx_clear_hlt(vcpu);
3787 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
3789 struct vcpu_vmx *vmx = to_vmx(vcpu);
3791 if (!cpu_has_virtual_nmis()) {
3793 * Tracking the NMI-blocked state in software is built upon
3794 * finding the next open IRQ window. This, in turn, depends on
3795 * well-behaving guests: They have to keep IRQs disabled at
3796 * least as long as the NMI handler runs. Otherwise we may
3797 * cause NMI nesting, maybe breaking the guest. But as this is
3798 * highly unlikely, we can live with the residual risk.
3800 vmx->soft_vnmi_blocked = 1;
3801 vmx->vnmi_blocked_time = 0;
3804 ++vcpu->stat.nmi_injections;
3805 vmx->nmi_known_unmasked = false;
3806 if (vmx->rmode.vm86_active) {
3807 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
3808 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3811 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
3812 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
3813 vmx_clear_hlt(vcpu);
3816 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
3818 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
3821 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
3822 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
3823 | GUEST_INTR_STATE_NMI));
3826 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
3828 if (!cpu_has_virtual_nmis())
3829 return to_vmx(vcpu)->soft_vnmi_blocked;
3830 if (to_vmx(vcpu)->nmi_known_unmasked)
3832 return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
3835 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
3837 struct vcpu_vmx *vmx = to_vmx(vcpu);
3839 if (!cpu_has_virtual_nmis()) {
3840 if (vmx->soft_vnmi_blocked != masked) {
3841 vmx->soft_vnmi_blocked = masked;
3842 vmx->vnmi_blocked_time = 0;
3845 vmx->nmi_known_unmasked = !masked;
3847 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3848 GUEST_INTR_STATE_NMI);
3850 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
3851 GUEST_INTR_STATE_NMI);
3855 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
3857 return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
3858 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
3859 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
3862 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
3865 struct kvm_userspace_memory_region tss_mem = {
3866 .slot = TSS_PRIVATE_MEMSLOT,
3867 .guest_phys_addr = addr,
3868 .memory_size = PAGE_SIZE * 3,
3872 ret = kvm_set_memory_region(kvm, &tss_mem, 0);
3875 kvm->arch.tss_addr = addr;
3876 if (!init_rmode_tss(kvm))
3882 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
3883 int vec, u32 err_code)
3886 * Instruction with address size override prefix opcode 0x67
3887 * Cause the #SS fault with 0 error code in VM86 mode.
3889 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
3890 if (emulate_instruction(vcpu, 0) == EMULATE_DONE)
3893 * Forward all other exceptions that are valid in real mode.
3894 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
3895 * the required debugging infrastructure rework.
3899 if (vcpu->guest_debug &
3900 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
3902 kvm_queue_exception(vcpu, vec);
3906 * Update instruction length as we may reinject the exception
3907 * from user space while in guest debugging mode.
3909 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
3910 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3911 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
3922 kvm_queue_exception(vcpu, vec);
3929 * Trigger machine check on the host. We assume all the MSRs are already set up
3930 * by the CPU and that we still run on the same CPU as the MCE occurred on.
3931 * We pass a fake environment to the machine check handler because we want
3932 * the guest to be always treated like user space, no matter what context
3933 * it used internally.
3935 static void kvm_machine_check(void)
3937 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
3938 struct pt_regs regs = {
3939 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
3940 .flags = X86_EFLAGS_IF,
3943 do_machine_check(®s, 0);
3947 static int handle_machine_check(struct kvm_vcpu *vcpu)
3949 /* already handled by vcpu_run */
3953 static int handle_exception(struct kvm_vcpu *vcpu)
3955 struct vcpu_vmx *vmx = to_vmx(vcpu);
3956 struct kvm_run *kvm_run = vcpu->run;
3957 u32 intr_info, ex_no, error_code;
3958 unsigned long cr2, rip, dr6;
3960 enum emulation_result er;
3962 vect_info = vmx->idt_vectoring_info;
3963 intr_info = vmx->exit_intr_info;
3965 if (is_machine_check(intr_info))
3966 return handle_machine_check(vcpu);
3968 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
3969 !is_page_fault(intr_info)) {
3970 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3971 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
3972 vcpu->run->internal.ndata = 2;
3973 vcpu->run->internal.data[0] = vect_info;
3974 vcpu->run->internal.data[1] = intr_info;
3978 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
3979 return 1; /* already handled by vmx_vcpu_run() */
3981 if (is_no_device(intr_info)) {
3982 vmx_fpu_activate(vcpu);
3986 if (is_invalid_opcode(intr_info)) {
3987 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
3988 if (er != EMULATE_DONE)
3989 kvm_queue_exception(vcpu, UD_VECTOR);
3994 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
3995 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
3996 if (is_page_fault(intr_info)) {
3997 /* EPT won't cause page fault directly */
4000 cr2 = vmcs_readl(EXIT_QUALIFICATION);
4001 trace_kvm_page_fault(cr2, error_code);
4003 if (kvm_event_needs_reinjection(vcpu))
4004 kvm_mmu_unprotect_page_virt(vcpu, cr2);
4005 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
4008 if (vmx->rmode.vm86_active &&
4009 handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
4011 if (vcpu->arch.halt_request) {
4012 vcpu->arch.halt_request = 0;
4013 return kvm_emulate_halt(vcpu);
4018 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4021 dr6 = vmcs_readl(EXIT_QUALIFICATION);
4022 if (!(vcpu->guest_debug &
4023 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4024 vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
4025 kvm_queue_exception(vcpu, DB_VECTOR);
4028 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4029 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4033 * Update instruction length as we may reinject #BP from
4034 * user space while in guest debugging mode. Reading it for
4035 * #DB as well causes no harm, it is not used in that case.
4037 vmx->vcpu.arch.event_exit_inst_len =
4038 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4039 kvm_run->exit_reason = KVM_EXIT_DEBUG;
4040 rip = kvm_rip_read(vcpu);
4041 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4042 kvm_run->debug.arch.exception = ex_no;
4045 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4046 kvm_run->ex.exception = ex_no;
4047 kvm_run->ex.error_code = error_code;
4053 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
4055 ++vcpu->stat.irq_exits;
4059 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4061 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4065 static int handle_io(struct kvm_vcpu *vcpu)
4067 unsigned long exit_qualification;
4068 int size, in, string;
4071 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4072 string = (exit_qualification & 16) != 0;
4073 in = (exit_qualification & 8) != 0;
4075 ++vcpu->stat.io_exits;
4078 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4080 port = exit_qualification >> 16;
4081 size = (exit_qualification & 7) + 1;
4082 skip_emulated_instruction(vcpu);
4084 return kvm_fast_pio_out(vcpu, size, port);
4088 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4091 * Patch in the VMCALL instruction:
4093 hypercall[0] = 0x0f;
4094 hypercall[1] = 0x01;
4095 hypercall[2] = 0xc1;
4098 static int handle_cr(struct kvm_vcpu *vcpu)
4100 unsigned long exit_qualification, val;
4105 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4106 cr = exit_qualification & 15;
4107 reg = (exit_qualification >> 8) & 15;
4108 switch ((exit_qualification >> 4) & 3) {
4109 case 0: /* mov to cr */
4110 val = kvm_register_read(vcpu, reg);
4111 trace_kvm_cr_write(cr, val);
4114 err = kvm_set_cr0(vcpu, val);
4115 kvm_complete_insn_gp(vcpu, err);
4118 err = kvm_set_cr3(vcpu, val);
4119 kvm_complete_insn_gp(vcpu, err);
4122 err = kvm_set_cr4(vcpu, val);
4123 kvm_complete_insn_gp(vcpu, err);
4126 u8 cr8_prev = kvm_get_cr8(vcpu);
4127 u8 cr8 = kvm_register_read(vcpu, reg);
4128 err = kvm_set_cr8(vcpu, cr8);
4129 kvm_complete_insn_gp(vcpu, err);
4130 if (irqchip_in_kernel(vcpu->kvm))
4132 if (cr8_prev <= cr8)
4134 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4140 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4141 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4142 skip_emulated_instruction(vcpu);
4143 vmx_fpu_activate(vcpu);
4145 case 1: /*mov from cr*/
4148 val = kvm_read_cr3(vcpu);
4149 kvm_register_write(vcpu, reg, val);
4150 trace_kvm_cr_read(cr, val);
4151 skip_emulated_instruction(vcpu);
4154 val = kvm_get_cr8(vcpu);
4155 kvm_register_write(vcpu, reg, val);
4156 trace_kvm_cr_read(cr, val);
4157 skip_emulated_instruction(vcpu);
4162 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4163 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
4164 kvm_lmsw(vcpu, val);
4166 skip_emulated_instruction(vcpu);
4171 vcpu->run->exit_reason = 0;
4172 pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
4173 (int)(exit_qualification >> 4) & 3, cr);
4177 static int handle_dr(struct kvm_vcpu *vcpu)
4179 unsigned long exit_qualification;
4182 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
4183 if (!kvm_require_cpl(vcpu, 0))
4185 dr = vmcs_readl(GUEST_DR7);
4188 * As the vm-exit takes precedence over the debug trap, we
4189 * need to emulate the latter, either for the host or the
4190 * guest debugging itself.
4192 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4193 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
4194 vcpu->run->debug.arch.dr7 = dr;
4195 vcpu->run->debug.arch.pc =
4196 vmcs_readl(GUEST_CS_BASE) +
4197 vmcs_readl(GUEST_RIP);
4198 vcpu->run->debug.arch.exception = DB_VECTOR;
4199 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
4202 vcpu->arch.dr7 &= ~DR7_GD;
4203 vcpu->arch.dr6 |= DR6_BD;
4204 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
4205 kvm_queue_exception(vcpu, DB_VECTOR);
4210 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4211 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
4212 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
4213 if (exit_qualification & TYPE_MOV_FROM_DR) {
4215 if (!kvm_get_dr(vcpu, dr, &val))
4216 kvm_register_write(vcpu, reg, val);
4218 kvm_set_dr(vcpu, dr, vcpu->arch.regs[reg]);
4219 skip_emulated_instruction(vcpu);
4223 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
4225 vmcs_writel(GUEST_DR7, val);
4228 static int handle_cpuid(struct kvm_vcpu *vcpu)
4230 kvm_emulate_cpuid(vcpu);
4234 static int handle_rdmsr(struct kvm_vcpu *vcpu)
4236 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
4239 if (vmx_get_msr(vcpu, ecx, &data)) {
4240 trace_kvm_msr_read_ex(ecx);
4241 kvm_inject_gp(vcpu, 0);
4245 trace_kvm_msr_read(ecx, data);
4247 /* FIXME: handling of bits 32:63 of rax, rdx */
4248 vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
4249 vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
4250 skip_emulated_instruction(vcpu);
4254 static int handle_wrmsr(struct kvm_vcpu *vcpu)
4256 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
4257 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
4258 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
4260 if (vmx_set_msr(vcpu, ecx, data) != 0) {
4261 trace_kvm_msr_write_ex(ecx, data);
4262 kvm_inject_gp(vcpu, 0);
4266 trace_kvm_msr_write(ecx, data);
4267 skip_emulated_instruction(vcpu);
4271 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
4273 kvm_make_request(KVM_REQ_EVENT, vcpu);
4277 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
4279 u32 cpu_based_vm_exec_control;
4281 /* clear pending irq */
4282 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4283 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
4284 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4286 kvm_make_request(KVM_REQ_EVENT, vcpu);
4288 ++vcpu->stat.irq_window_exits;
4291 * If the user space waits to inject interrupts, exit as soon as
4294 if (!irqchip_in_kernel(vcpu->kvm) &&
4295 vcpu->run->request_interrupt_window &&
4296 !kvm_cpu_has_interrupt(vcpu)) {
4297 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
4303 static int handle_halt(struct kvm_vcpu *vcpu)
4305 skip_emulated_instruction(vcpu);
4306 return kvm_emulate_halt(vcpu);
4309 static int handle_vmcall(struct kvm_vcpu *vcpu)
4311 skip_emulated_instruction(vcpu);
4312 kvm_emulate_hypercall(vcpu);
4316 static int handle_invd(struct kvm_vcpu *vcpu)
4318 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4321 static int handle_invlpg(struct kvm_vcpu *vcpu)
4323 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4325 kvm_mmu_invlpg(vcpu, exit_qualification);
4326 skip_emulated_instruction(vcpu);
4330 static int handle_wbinvd(struct kvm_vcpu *vcpu)
4332 skip_emulated_instruction(vcpu);
4333 kvm_emulate_wbinvd(vcpu);
4337 static int handle_xsetbv(struct kvm_vcpu *vcpu)
4339 u64 new_bv = kvm_read_edx_eax(vcpu);
4340 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4342 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
4343 skip_emulated_instruction(vcpu);
4347 static int handle_apic_access(struct kvm_vcpu *vcpu)
4349 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4352 static int handle_task_switch(struct kvm_vcpu *vcpu)
4354 struct vcpu_vmx *vmx = to_vmx(vcpu);
4355 unsigned long exit_qualification;
4356 bool has_error_code = false;
4359 int reason, type, idt_v;
4361 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
4362 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
4364 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4366 reason = (u32)exit_qualification >> 30;
4367 if (reason == TASK_SWITCH_GATE && idt_v) {
4369 case INTR_TYPE_NMI_INTR:
4370 vcpu->arch.nmi_injected = false;
4371 vmx_set_nmi_mask(vcpu, true);
4373 case INTR_TYPE_EXT_INTR:
4374 case INTR_TYPE_SOFT_INTR:
4375 kvm_clear_interrupt_queue(vcpu);
4377 case INTR_TYPE_HARD_EXCEPTION:
4378 if (vmx->idt_vectoring_info &
4379 VECTORING_INFO_DELIVER_CODE_MASK) {
4380 has_error_code = true;
4382 vmcs_read32(IDT_VECTORING_ERROR_CODE);
4385 case INTR_TYPE_SOFT_EXCEPTION:
4386 kvm_clear_exception_queue(vcpu);
4392 tss_selector = exit_qualification;
4394 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
4395 type != INTR_TYPE_EXT_INTR &&
4396 type != INTR_TYPE_NMI_INTR))
4397 skip_emulated_instruction(vcpu);
4399 if (kvm_task_switch(vcpu, tss_selector, reason,
4400 has_error_code, error_code) == EMULATE_FAIL) {
4401 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4402 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4403 vcpu->run->internal.ndata = 0;
4407 /* clear all local breakpoint enable flags */
4408 vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
4411 * TODO: What about debug traps on tss switch?
4412 * Are we supposed to inject them and update dr6?
4418 static int handle_ept_violation(struct kvm_vcpu *vcpu)
4420 unsigned long exit_qualification;
4424 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4426 if (exit_qualification & (1 << 6)) {
4427 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
4431 gla_validity = (exit_qualification >> 7) & 0x3;
4432 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
4433 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
4434 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
4435 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
4436 vmcs_readl(GUEST_LINEAR_ADDRESS));
4437 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
4438 (long unsigned int)exit_qualification);
4439 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
4440 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
4444 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
4445 trace_kvm_page_fault(gpa, exit_qualification);
4446 return kvm_mmu_page_fault(vcpu, gpa, exit_qualification & 0x3, NULL, 0);
4449 static u64 ept_rsvd_mask(u64 spte, int level)
4454 for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
4455 mask |= (1ULL << i);
4458 /* bits 7:3 reserved */
4460 else if (level == 2) {
4461 if (spte & (1ULL << 7))
4462 /* 2MB ref, bits 20:12 reserved */
4465 /* bits 6:3 reserved */
4472 static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
4475 printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
4477 /* 010b (write-only) */
4478 WARN_ON((spte & 0x7) == 0x2);
4480 /* 110b (write/execute) */
4481 WARN_ON((spte & 0x7) == 0x6);
4483 /* 100b (execute-only) and value not supported by logical processor */
4484 if (!cpu_has_vmx_ept_execute_only())
4485 WARN_ON((spte & 0x7) == 0x4);
4489 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
4491 if (rsvd_bits != 0) {
4492 printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
4493 __func__, rsvd_bits);
4497 if (level == 1 || (level == 2 && (spte & (1ULL << 7)))) {
4498 u64 ept_mem_type = (spte & 0x38) >> 3;
4500 if (ept_mem_type == 2 || ept_mem_type == 3 ||
4501 ept_mem_type == 7) {
4502 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
4503 __func__, ept_mem_type);
4510 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
4516 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
4518 printk(KERN_ERR "EPT: Misconfiguration.\n");
4519 printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
4521 nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
4523 for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
4524 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
4526 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
4527 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
4532 static int handle_nmi_window(struct kvm_vcpu *vcpu)
4534 u32 cpu_based_vm_exec_control;
4536 /* clear pending NMI */
4537 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4538 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
4539 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4540 ++vcpu->stat.nmi_window_exits;
4541 kvm_make_request(KVM_REQ_EVENT, vcpu);
4546 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
4548 struct vcpu_vmx *vmx = to_vmx(vcpu);
4549 enum emulation_result err = EMULATE_DONE;
4552 bool intr_window_requested;
4554 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4555 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
4557 while (!guest_state_valid(vcpu)) {
4558 if (intr_window_requested
4559 && (kvm_get_rflags(&vmx->vcpu) & X86_EFLAGS_IF))
4560 return handle_interrupt_window(&vmx->vcpu);
4562 err = emulate_instruction(vcpu, 0);
4564 if (err == EMULATE_DO_MMIO) {
4569 if (err != EMULATE_DONE)
4572 if (signal_pending(current))
4578 vmx->emulation_required = 0;
4584 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
4585 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
4587 static int handle_pause(struct kvm_vcpu *vcpu)
4589 skip_emulated_instruction(vcpu);
4590 kvm_vcpu_on_spin(vcpu);
4595 static int handle_invalid_op(struct kvm_vcpu *vcpu)
4597 kvm_queue_exception(vcpu, UD_VECTOR);
4602 * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
4603 * We could reuse a single VMCS for all the L2 guests, but we also want the
4604 * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
4605 * allows keeping them loaded on the processor, and in the future will allow
4606 * optimizations where prepare_vmcs02 doesn't need to set all the fields on
4607 * every entry if they never change.
4608 * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
4609 * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
4611 * The following functions allocate and free a vmcs02 in this pool.
4614 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
4615 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
4617 struct vmcs02_list *item;
4618 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
4619 if (item->vmptr == vmx->nested.current_vmptr) {
4620 list_move(&item->list, &vmx->nested.vmcs02_pool);
4621 return &item->vmcs02;
4624 if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
4625 /* Recycle the least recently used VMCS. */
4626 item = list_entry(vmx->nested.vmcs02_pool.prev,
4627 struct vmcs02_list, list);
4628 item->vmptr = vmx->nested.current_vmptr;
4629 list_move(&item->list, &vmx->nested.vmcs02_pool);
4630 return &item->vmcs02;
4633 /* Create a new VMCS */
4634 item = (struct vmcs02_list *)
4635 kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
4638 item->vmcs02.vmcs = alloc_vmcs();
4639 if (!item->vmcs02.vmcs) {
4643 loaded_vmcs_init(&item->vmcs02);
4644 item->vmptr = vmx->nested.current_vmptr;
4645 list_add(&(item->list), &(vmx->nested.vmcs02_pool));
4646 vmx->nested.vmcs02_num++;
4647 return &item->vmcs02;
4650 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
4651 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
4653 struct vmcs02_list *item;
4654 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
4655 if (item->vmptr == vmptr) {
4656 free_loaded_vmcs(&item->vmcs02);
4657 list_del(&item->list);
4659 vmx->nested.vmcs02_num--;
4665 * Free all VMCSs saved for this vcpu, except the one pointed by
4666 * vmx->loaded_vmcs. These include the VMCSs in vmcs02_pool (except the one
4667 * currently used, if running L2), and vmcs01 when running L2.
4669 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
4671 struct vmcs02_list *item, *n;
4672 list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
4673 if (vmx->loaded_vmcs != &item->vmcs02)
4674 free_loaded_vmcs(&item->vmcs02);
4675 list_del(&item->list);
4678 vmx->nested.vmcs02_num = 0;
4680 if (vmx->loaded_vmcs != &vmx->vmcs01)
4681 free_loaded_vmcs(&vmx->vmcs01);
4685 * Emulate the VMXON instruction.
4686 * Currently, we just remember that VMX is active, and do not save or even
4687 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
4688 * do not currently need to store anything in that guest-allocated memory
4689 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
4690 * argument is different from the VMXON pointer (which the spec says they do).
4692 static int handle_vmon(struct kvm_vcpu *vcpu)
4694 struct kvm_segment cs;
4695 struct vcpu_vmx *vmx = to_vmx(vcpu);
4697 /* The Intel VMX Instruction Reference lists a bunch of bits that
4698 * are prerequisite to running VMXON, most notably cr4.VMXE must be
4699 * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
4700 * Otherwise, we should fail with #UD. We test these now:
4702 if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
4703 !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
4704 (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
4705 kvm_queue_exception(vcpu, UD_VECTOR);
4709 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4710 if (is_long_mode(vcpu) && !cs.l) {
4711 kvm_queue_exception(vcpu, UD_VECTOR);
4715 if (vmx_get_cpl(vcpu)) {
4716 kvm_inject_gp(vcpu, 0);
4720 INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
4721 vmx->nested.vmcs02_num = 0;
4723 vmx->nested.vmxon = true;
4725 skip_emulated_instruction(vcpu);
4730 * Intel's VMX Instruction Reference specifies a common set of prerequisites
4731 * for running VMX instructions (except VMXON, whose prerequisites are
4732 * slightly different). It also specifies what exception to inject otherwise.
4734 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
4736 struct kvm_segment cs;
4737 struct vcpu_vmx *vmx = to_vmx(vcpu);
4739 if (!vmx->nested.vmxon) {
4740 kvm_queue_exception(vcpu, UD_VECTOR);
4744 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4745 if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
4746 (is_long_mode(vcpu) && !cs.l)) {
4747 kvm_queue_exception(vcpu, UD_VECTOR);
4751 if (vmx_get_cpl(vcpu)) {
4752 kvm_inject_gp(vcpu, 0);
4760 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
4761 * just stops using VMX.
4763 static void free_nested(struct vcpu_vmx *vmx)
4765 if (!vmx->nested.vmxon)
4767 vmx->nested.vmxon = false;
4768 if (vmx->nested.current_vmptr != -1ull) {
4769 kunmap(vmx->nested.current_vmcs12_page);
4770 nested_release_page(vmx->nested.current_vmcs12_page);
4771 vmx->nested.current_vmptr = -1ull;
4772 vmx->nested.current_vmcs12 = NULL;
4774 /* Unpin physical memory we referred to in current vmcs02 */
4775 if (vmx->nested.apic_access_page) {
4776 nested_release_page(vmx->nested.apic_access_page);
4777 vmx->nested.apic_access_page = 0;
4780 nested_free_all_saved_vmcss(vmx);
4783 /* Emulate the VMXOFF instruction */
4784 static int handle_vmoff(struct kvm_vcpu *vcpu)
4786 if (!nested_vmx_check_permission(vcpu))
4788 free_nested(to_vmx(vcpu));
4789 skip_emulated_instruction(vcpu);
4794 * Decode the memory-address operand of a vmx instruction, as recorded on an
4795 * exit caused by such an instruction (run by a guest hypervisor).
4796 * On success, returns 0. When the operand is invalid, returns 1 and throws
4799 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
4800 unsigned long exit_qualification,
4801 u32 vmx_instruction_info, gva_t *ret)
4804 * According to Vol. 3B, "Information for VM Exits Due to Instruction
4805 * Execution", on an exit, vmx_instruction_info holds most of the
4806 * addressing components of the operand. Only the displacement part
4807 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
4808 * For how an actual address is calculated from all these components,
4809 * refer to Vol. 1, "Operand Addressing".
4811 int scaling = vmx_instruction_info & 3;
4812 int addr_size = (vmx_instruction_info >> 7) & 7;
4813 bool is_reg = vmx_instruction_info & (1u << 10);
4814 int seg_reg = (vmx_instruction_info >> 15) & 7;
4815 int index_reg = (vmx_instruction_info >> 18) & 0xf;
4816 bool index_is_valid = !(vmx_instruction_info & (1u << 22));
4817 int base_reg = (vmx_instruction_info >> 23) & 0xf;
4818 bool base_is_valid = !(vmx_instruction_info & (1u << 27));
4821 kvm_queue_exception(vcpu, UD_VECTOR);
4825 /* Addr = segment_base + offset */
4826 /* offset = base + [index * scale] + displacement */
4827 *ret = vmx_get_segment_base(vcpu, seg_reg);
4829 *ret += kvm_register_read(vcpu, base_reg);
4831 *ret += kvm_register_read(vcpu, index_reg)<<scaling;
4832 *ret += exit_qualification; /* holds the displacement */
4834 if (addr_size == 1) /* 32 bit */
4838 * TODO: throw #GP (and return 1) in various cases that the VM*
4839 * instructions require it - e.g., offset beyond segment limit,
4840 * unusable or unreadable/unwritable segment, non-canonical 64-bit
4841 * address, and so on. Currently these are not checked.
4847 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
4848 * set the success or error code of an emulated VMX instruction, as specified
4849 * by Vol 2B, VMX Instruction Reference, "Conventions".
4851 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
4853 vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
4854 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
4855 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
4858 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
4860 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
4861 & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
4862 X86_EFLAGS_SF | X86_EFLAGS_OF))
4866 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
4867 u32 vm_instruction_error)
4869 if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
4871 * failValid writes the error number to the current VMCS, which
4872 * can't be done there isn't a current VMCS.
4874 nested_vmx_failInvalid(vcpu);
4877 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
4878 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
4879 X86_EFLAGS_SF | X86_EFLAGS_OF))
4881 get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
4884 /* Emulate the VMCLEAR instruction */
4885 static int handle_vmclear(struct kvm_vcpu *vcpu)
4887 struct vcpu_vmx *vmx = to_vmx(vcpu);
4890 struct vmcs12 *vmcs12;
4892 struct x86_exception e;
4894 if (!nested_vmx_check_permission(vcpu))
4897 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
4898 vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
4901 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
4902 sizeof(vmptr), &e)) {
4903 kvm_inject_page_fault(vcpu, &e);
4907 if (!IS_ALIGNED(vmptr, PAGE_SIZE)) {
4908 nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
4909 skip_emulated_instruction(vcpu);
4913 if (vmptr == vmx->nested.current_vmptr) {
4914 kunmap(vmx->nested.current_vmcs12_page);
4915 nested_release_page(vmx->nested.current_vmcs12_page);
4916 vmx->nested.current_vmptr = -1ull;
4917 vmx->nested.current_vmcs12 = NULL;
4920 page = nested_get_page(vcpu, vmptr);
4923 * For accurate processor emulation, VMCLEAR beyond available
4924 * physical memory should do nothing at all. However, it is
4925 * possible that a nested vmx bug, not a guest hypervisor bug,
4926 * resulted in this case, so let's shut down before doing any
4929 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4932 vmcs12 = kmap(page);
4933 vmcs12->launch_state = 0;
4935 nested_release_page(page);
4937 nested_free_vmcs02(vmx, vmptr);
4939 skip_emulated_instruction(vcpu);
4940 nested_vmx_succeed(vcpu);
4944 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
4946 /* Emulate the VMLAUNCH instruction */
4947 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
4949 return nested_vmx_run(vcpu, true);
4952 /* Emulate the VMRESUME instruction */
4953 static int handle_vmresume(struct kvm_vcpu *vcpu)
4956 return nested_vmx_run(vcpu, false);
4959 enum vmcs_field_type {
4960 VMCS_FIELD_TYPE_U16 = 0,
4961 VMCS_FIELD_TYPE_U64 = 1,
4962 VMCS_FIELD_TYPE_U32 = 2,
4963 VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
4966 static inline int vmcs_field_type(unsigned long field)
4968 if (0x1 & field) /* the *_HIGH fields are all 32 bit */
4969 return VMCS_FIELD_TYPE_U32;
4970 return (field >> 13) & 0x3 ;
4973 static inline int vmcs_field_readonly(unsigned long field)
4975 return (((field >> 10) & 0x3) == 1);
4979 * Read a vmcs12 field. Since these can have varying lengths and we return
4980 * one type, we chose the biggest type (u64) and zero-extend the return value
4981 * to that size. Note that the caller, handle_vmread, might need to use only
4982 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
4983 * 64-bit fields are to be returned).
4985 static inline bool vmcs12_read_any(struct kvm_vcpu *vcpu,
4986 unsigned long field, u64 *ret)
4988 short offset = vmcs_field_to_offset(field);
4994 p = ((char *)(get_vmcs12(vcpu))) + offset;
4996 switch (vmcs_field_type(field)) {
4997 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
4998 *ret = *((natural_width *)p);
5000 case VMCS_FIELD_TYPE_U16:
5003 case VMCS_FIELD_TYPE_U32:
5006 case VMCS_FIELD_TYPE_U64:
5010 return 0; /* can never happen. */
5015 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
5016 * used before) all generate the same failure when it is missing.
5018 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
5020 struct vcpu_vmx *vmx = to_vmx(vcpu);
5021 if (vmx->nested.current_vmptr == -1ull) {
5022 nested_vmx_failInvalid(vcpu);
5023 skip_emulated_instruction(vcpu);
5029 static int handle_vmread(struct kvm_vcpu *vcpu)
5031 unsigned long field;
5033 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5034 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5037 if (!nested_vmx_check_permission(vcpu) ||
5038 !nested_vmx_check_vmcs12(vcpu))
5041 /* Decode instruction info and find the field to read */
5042 field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5043 /* Read the field, zero-extended to a u64 field_value */
5044 if (!vmcs12_read_any(vcpu, field, &field_value)) {
5045 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5046 skip_emulated_instruction(vcpu);
5050 * Now copy part of this value to register or memory, as requested.
5051 * Note that the number of bits actually copied is 32 or 64 depending
5052 * on the guest's mode (32 or 64 bit), not on the given field's length.
5054 if (vmx_instruction_info & (1u << 10)) {
5055 kvm_register_write(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
5058 if (get_vmx_mem_address(vcpu, exit_qualification,
5059 vmx_instruction_info, &gva))
5061 /* _system ok, as nested_vmx_check_permission verified cpl=0 */
5062 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
5063 &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
5066 nested_vmx_succeed(vcpu);
5067 skip_emulated_instruction(vcpu);
5072 static int handle_vmwrite(struct kvm_vcpu *vcpu)
5074 unsigned long field;
5076 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5077 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5080 /* The value to write might be 32 or 64 bits, depending on L1's long
5081 * mode, and eventually we need to write that into a field of several
5082 * possible lengths. The code below first zero-extends the value to 64
5083 * bit (field_value), and then copies only the approriate number of
5084 * bits into the vmcs12 field.
5086 u64 field_value = 0;
5087 struct x86_exception e;
5089 if (!nested_vmx_check_permission(vcpu) ||
5090 !nested_vmx_check_vmcs12(vcpu))
5093 if (vmx_instruction_info & (1u << 10))
5094 field_value = kvm_register_read(vcpu,
5095 (((vmx_instruction_info) >> 3) & 0xf));
5097 if (get_vmx_mem_address(vcpu, exit_qualification,
5098 vmx_instruction_info, &gva))
5100 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
5101 &field_value, (is_long_mode(vcpu) ? 8 : 4), &e)) {
5102 kvm_inject_page_fault(vcpu, &e);
5108 field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5109 if (vmcs_field_readonly(field)) {
5110 nested_vmx_failValid(vcpu,
5111 VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
5112 skip_emulated_instruction(vcpu);
5116 offset = vmcs_field_to_offset(field);
5118 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5119 skip_emulated_instruction(vcpu);
5122 p = ((char *) get_vmcs12(vcpu)) + offset;
5124 switch (vmcs_field_type(field)) {
5125 case VMCS_FIELD_TYPE_U16:
5126 *(u16 *)p = field_value;
5128 case VMCS_FIELD_TYPE_U32:
5129 *(u32 *)p = field_value;
5131 case VMCS_FIELD_TYPE_U64:
5132 *(u64 *)p = field_value;
5134 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
5135 *(natural_width *)p = field_value;
5138 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5139 skip_emulated_instruction(vcpu);
5143 nested_vmx_succeed(vcpu);
5144 skip_emulated_instruction(vcpu);
5148 /* Emulate the VMPTRLD instruction */
5149 static int handle_vmptrld(struct kvm_vcpu *vcpu)
5151 struct vcpu_vmx *vmx = to_vmx(vcpu);
5154 struct x86_exception e;
5156 if (!nested_vmx_check_permission(vcpu))
5159 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5160 vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
5163 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
5164 sizeof(vmptr), &e)) {
5165 kvm_inject_page_fault(vcpu, &e);
5169 if (!IS_ALIGNED(vmptr, PAGE_SIZE)) {
5170 nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
5171 skip_emulated_instruction(vcpu);
5175 if (vmx->nested.current_vmptr != vmptr) {
5176 struct vmcs12 *new_vmcs12;
5178 page = nested_get_page(vcpu, vmptr);
5180 nested_vmx_failInvalid(vcpu);
5181 skip_emulated_instruction(vcpu);
5184 new_vmcs12 = kmap(page);
5185 if (new_vmcs12->revision_id != VMCS12_REVISION) {
5187 nested_release_page_clean(page);
5188 nested_vmx_failValid(vcpu,
5189 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5190 skip_emulated_instruction(vcpu);
5193 if (vmx->nested.current_vmptr != -1ull) {
5194 kunmap(vmx->nested.current_vmcs12_page);
5195 nested_release_page(vmx->nested.current_vmcs12_page);
5198 vmx->nested.current_vmptr = vmptr;
5199 vmx->nested.current_vmcs12 = new_vmcs12;
5200 vmx->nested.current_vmcs12_page = page;
5203 nested_vmx_succeed(vcpu);
5204 skip_emulated_instruction(vcpu);
5208 /* Emulate the VMPTRST instruction */
5209 static int handle_vmptrst(struct kvm_vcpu *vcpu)
5211 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5212 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5214 struct x86_exception e;
5216 if (!nested_vmx_check_permission(vcpu))
5219 if (get_vmx_mem_address(vcpu, exit_qualification,
5220 vmx_instruction_info, &vmcs_gva))
5222 /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
5223 if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
5224 (void *)&to_vmx(vcpu)->nested.current_vmptr,
5226 kvm_inject_page_fault(vcpu, &e);
5229 nested_vmx_succeed(vcpu);
5230 skip_emulated_instruction(vcpu);
5235 * The exit handlers return 1 if the exit was handled fully and guest execution
5236 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
5237 * to be done to userspace and return 0.
5239 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5240 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
5241 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
5242 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
5243 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
5244 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
5245 [EXIT_REASON_CR_ACCESS] = handle_cr,
5246 [EXIT_REASON_DR_ACCESS] = handle_dr,
5247 [EXIT_REASON_CPUID] = handle_cpuid,
5248 [EXIT_REASON_MSR_READ] = handle_rdmsr,
5249 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
5250 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
5251 [EXIT_REASON_HLT] = handle_halt,
5252 [EXIT_REASON_INVD] = handle_invd,
5253 [EXIT_REASON_INVLPG] = handle_invlpg,
5254 [EXIT_REASON_VMCALL] = handle_vmcall,
5255 [EXIT_REASON_VMCLEAR] = handle_vmclear,
5256 [EXIT_REASON_VMLAUNCH] = handle_vmlaunch,
5257 [EXIT_REASON_VMPTRLD] = handle_vmptrld,
5258 [EXIT_REASON_VMPTRST] = handle_vmptrst,
5259 [EXIT_REASON_VMREAD] = handle_vmread,
5260 [EXIT_REASON_VMRESUME] = handle_vmresume,
5261 [EXIT_REASON_VMWRITE] = handle_vmwrite,
5262 [EXIT_REASON_VMOFF] = handle_vmoff,
5263 [EXIT_REASON_VMON] = handle_vmon,
5264 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
5265 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
5266 [EXIT_REASON_WBINVD] = handle_wbinvd,
5267 [EXIT_REASON_XSETBV] = handle_xsetbv,
5268 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
5269 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
5270 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
5271 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
5272 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
5273 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_invalid_op,
5274 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_invalid_op,
5277 static const int kvm_vmx_max_exit_handlers =
5278 ARRAY_SIZE(kvm_vmx_exit_handlers);
5280 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
5282 *info1 = vmcs_readl(EXIT_QUALIFICATION);
5283 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
5287 * The guest has exited. See if we can fix it or if we need userspace
5290 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
5292 struct vcpu_vmx *vmx = to_vmx(vcpu);
5293 u32 exit_reason = vmx->exit_reason;
5294 u32 vectoring_info = vmx->idt_vectoring_info;
5296 trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
5298 /* If guest state is invalid, start emulating */
5299 if (vmx->emulation_required && emulate_invalid_guest_state)
5300 return handle_invalid_guest_state(vcpu);
5302 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
5303 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5304 vcpu->run->fail_entry.hardware_entry_failure_reason
5309 if (unlikely(vmx->fail)) {
5310 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5311 vcpu->run->fail_entry.hardware_entry_failure_reason
5312 = vmcs_read32(VM_INSTRUCTION_ERROR);
5316 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
5317 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
5318 exit_reason != EXIT_REASON_EPT_VIOLATION &&
5319 exit_reason != EXIT_REASON_TASK_SWITCH))
5320 printk(KERN_WARNING "%s: unexpected, valid vectoring info "
5321 "(0x%x) and exit reason is 0x%x\n",
5322 __func__, vectoring_info, exit_reason);
5324 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked)) {
5325 if (vmx_interrupt_allowed(vcpu)) {
5326 vmx->soft_vnmi_blocked = 0;
5327 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
5328 vcpu->arch.nmi_pending) {
5330 * This CPU don't support us in finding the end of an
5331 * NMI-blocked window if the guest runs with IRQs
5332 * disabled. So we pull the trigger after 1 s of
5333 * futile waiting, but inform the user about this.
5335 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
5336 "state on VCPU %d after 1 s timeout\n",
5337 __func__, vcpu->vcpu_id);
5338 vmx->soft_vnmi_blocked = 0;
5342 if (exit_reason < kvm_vmx_max_exit_handlers
5343 && kvm_vmx_exit_handlers[exit_reason])
5344 return kvm_vmx_exit_handlers[exit_reason](vcpu);
5346 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5347 vcpu->run->hw.hardware_exit_reason = exit_reason;
5352 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
5354 if (irr == -1 || tpr < irr) {
5355 vmcs_write32(TPR_THRESHOLD, 0);
5359 vmcs_write32(TPR_THRESHOLD, irr);
5362 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
5366 if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
5367 || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
5370 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
5371 exit_intr_info = vmx->exit_intr_info;
5373 /* Handle machine checks before interrupts are enabled */
5374 if (is_machine_check(exit_intr_info))
5375 kvm_machine_check();
5377 /* We need to handle NMIs before interrupts are enabled */
5378 if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
5379 (exit_intr_info & INTR_INFO_VALID_MASK)) {
5380 kvm_before_handle_nmi(&vmx->vcpu);
5382 kvm_after_handle_nmi(&vmx->vcpu);
5386 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
5391 bool idtv_info_valid;
5393 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
5395 if (cpu_has_virtual_nmis()) {
5396 if (vmx->nmi_known_unmasked)
5399 * Can't use vmx->exit_intr_info since we're not sure what
5400 * the exit reason is.
5402 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
5403 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
5404 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
5406 * SDM 3: 27.7.1.2 (September 2008)
5407 * Re-set bit "block by NMI" before VM entry if vmexit caused by
5408 * a guest IRET fault.
5409 * SDM 3: 23.2.2 (September 2008)
5410 * Bit 12 is undefined in any of the following cases:
5411 * If the VM exit sets the valid bit in the IDT-vectoring
5412 * information field.
5413 * If the VM exit is due to a double fault.
5415 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
5416 vector != DF_VECTOR && !idtv_info_valid)
5417 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5418 GUEST_INTR_STATE_NMI);
5420 vmx->nmi_known_unmasked =
5421 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
5422 & GUEST_INTR_STATE_NMI);
5423 } else if (unlikely(vmx->soft_vnmi_blocked))
5424 vmx->vnmi_blocked_time +=
5425 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
5428 static void __vmx_complete_interrupts(struct vcpu_vmx *vmx,
5429 u32 idt_vectoring_info,
5430 int instr_len_field,
5431 int error_code_field)
5435 bool idtv_info_valid;
5437 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
5439 vmx->vcpu.arch.nmi_injected = false;
5440 kvm_clear_exception_queue(&vmx->vcpu);
5441 kvm_clear_interrupt_queue(&vmx->vcpu);
5443 if (!idtv_info_valid)
5446 kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
5448 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
5449 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
5452 case INTR_TYPE_NMI_INTR:
5453 vmx->vcpu.arch.nmi_injected = true;
5455 * SDM 3: 27.7.1.2 (September 2008)
5456 * Clear bit "block by NMI" before VM entry if a NMI
5459 vmx_set_nmi_mask(&vmx->vcpu, false);
5461 case INTR_TYPE_SOFT_EXCEPTION:
5462 vmx->vcpu.arch.event_exit_inst_len =
5463 vmcs_read32(instr_len_field);
5465 case INTR_TYPE_HARD_EXCEPTION:
5466 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
5467 u32 err = vmcs_read32(error_code_field);
5468 kvm_queue_exception_e(&vmx->vcpu, vector, err);
5470 kvm_queue_exception(&vmx->vcpu, vector);
5472 case INTR_TYPE_SOFT_INTR:
5473 vmx->vcpu.arch.event_exit_inst_len =
5474 vmcs_read32(instr_len_field);
5476 case INTR_TYPE_EXT_INTR:
5477 kvm_queue_interrupt(&vmx->vcpu, vector,
5478 type == INTR_TYPE_SOFT_INTR);
5485 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
5487 __vmx_complete_interrupts(vmx, vmx->idt_vectoring_info,
5488 VM_EXIT_INSTRUCTION_LEN,
5489 IDT_VECTORING_ERROR_CODE);
5492 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
5494 __vmx_complete_interrupts(to_vmx(vcpu),
5495 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
5496 VM_ENTRY_INSTRUCTION_LEN,
5497 VM_ENTRY_EXCEPTION_ERROR_CODE);
5499 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
5502 #ifdef CONFIG_X86_64
5510 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
5512 struct vcpu_vmx *vmx = to_vmx(vcpu);
5514 /* Record the guest's net vcpu time for enforced NMI injections. */
5515 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
5516 vmx->entry_time = ktime_get();
5518 /* Don't enter VMX if guest state is invalid, let the exit handler
5519 start emulation until we arrive back to a valid state */
5520 if (vmx->emulation_required && emulate_invalid_guest_state)
5523 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
5524 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
5525 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
5526 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
5528 /* When single-stepping over STI and MOV SS, we must clear the
5529 * corresponding interruptibility bits in the guest state. Otherwise
5530 * vmentry fails as it then expects bit 14 (BS) in pending debug
5531 * exceptions being set, but that's not correct for the guest debugging
5533 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5534 vmx_set_interrupt_shadow(vcpu, 0);
5536 vmx->__launched = vmx->loaded_vmcs->launched;
5538 /* Store host registers */
5539 "push %%"R"dx; push %%"R"bp;"
5540 "push %%"R"cx \n\t" /* placeholder for guest rcx */
5542 "cmp %%"R"sp, %c[host_rsp](%0) \n\t"
5544 "mov %%"R"sp, %c[host_rsp](%0) \n\t"
5545 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
5547 /* Reload cr2 if changed */
5548 "mov %c[cr2](%0), %%"R"ax \n\t"
5549 "mov %%cr2, %%"R"dx \n\t"
5550 "cmp %%"R"ax, %%"R"dx \n\t"
5552 "mov %%"R"ax, %%cr2 \n\t"
5554 /* Check if vmlaunch of vmresume is needed */
5555 "cmpl $0, %c[launched](%0) \n\t"
5556 /* Load guest registers. Don't clobber flags. */
5557 "mov %c[rax](%0), %%"R"ax \n\t"
5558 "mov %c[rbx](%0), %%"R"bx \n\t"
5559 "mov %c[rdx](%0), %%"R"dx \n\t"
5560 "mov %c[rsi](%0), %%"R"si \n\t"
5561 "mov %c[rdi](%0), %%"R"di \n\t"
5562 "mov %c[rbp](%0), %%"R"bp \n\t"
5563 #ifdef CONFIG_X86_64
5564 "mov %c[r8](%0), %%r8 \n\t"
5565 "mov %c[r9](%0), %%r9 \n\t"
5566 "mov %c[r10](%0), %%r10 \n\t"
5567 "mov %c[r11](%0), %%r11 \n\t"
5568 "mov %c[r12](%0), %%r12 \n\t"
5569 "mov %c[r13](%0), %%r13 \n\t"
5570 "mov %c[r14](%0), %%r14 \n\t"
5571 "mov %c[r15](%0), %%r15 \n\t"
5573 "mov %c[rcx](%0), %%"R"cx \n\t" /* kills %0 (ecx) */
5575 /* Enter guest mode */
5576 "jne .Llaunched \n\t"
5577 __ex(ASM_VMX_VMLAUNCH) "\n\t"
5578 "jmp .Lkvm_vmx_return \n\t"
5579 ".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t"
5580 ".Lkvm_vmx_return: "
5581 /* Save guest registers, load host registers, keep flags */
5582 "mov %0, %c[wordsize](%%"R"sp) \n\t"
5584 "mov %%"R"ax, %c[rax](%0) \n\t"
5585 "mov %%"R"bx, %c[rbx](%0) \n\t"
5586 "pop"Q" %c[rcx](%0) \n\t"
5587 "mov %%"R"dx, %c[rdx](%0) \n\t"
5588 "mov %%"R"si, %c[rsi](%0) \n\t"
5589 "mov %%"R"di, %c[rdi](%0) \n\t"
5590 "mov %%"R"bp, %c[rbp](%0) \n\t"
5591 #ifdef CONFIG_X86_64
5592 "mov %%r8, %c[r8](%0) \n\t"
5593 "mov %%r9, %c[r9](%0) \n\t"
5594 "mov %%r10, %c[r10](%0) \n\t"
5595 "mov %%r11, %c[r11](%0) \n\t"
5596 "mov %%r12, %c[r12](%0) \n\t"
5597 "mov %%r13, %c[r13](%0) \n\t"
5598 "mov %%r14, %c[r14](%0) \n\t"
5599 "mov %%r15, %c[r15](%0) \n\t"
5601 "mov %%cr2, %%"R"ax \n\t"
5602 "mov %%"R"ax, %c[cr2](%0) \n\t"
5604 "pop %%"R"bp; pop %%"R"dx \n\t"
5605 "setbe %c[fail](%0) \n\t"
5606 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
5607 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
5608 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
5609 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
5610 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
5611 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
5612 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
5613 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
5614 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
5615 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
5616 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
5617 #ifdef CONFIG_X86_64
5618 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
5619 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
5620 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
5621 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
5622 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
5623 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
5624 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
5625 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
5627 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
5628 [wordsize]"i"(sizeof(ulong))
5630 , R"ax", R"bx", R"di", R"si"
5631 #ifdef CONFIG_X86_64
5632 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
5636 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
5637 | (1 << VCPU_EXREG_RFLAGS)
5638 | (1 << VCPU_EXREG_CPL)
5639 | (1 << VCPU_EXREG_PDPTR)
5640 | (1 << VCPU_EXREG_SEGMENTS)
5641 | (1 << VCPU_EXREG_CR3));
5642 vcpu->arch.regs_dirty = 0;
5644 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
5646 asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
5647 vmx->loaded_vmcs->launched = 1;
5649 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
5651 vmx_complete_atomic_exit(vmx);
5652 vmx_recover_nmi_blocking(vmx);
5653 vmx_complete_interrupts(vmx);
5659 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
5661 struct vcpu_vmx *vmx = to_vmx(vcpu);
5665 free_loaded_vmcs(vmx->loaded_vmcs);
5666 kfree(vmx->guest_msrs);
5667 kvm_vcpu_uninit(vcpu);
5668 kmem_cache_free(kvm_vcpu_cache, vmx);
5671 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
5674 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
5678 return ERR_PTR(-ENOMEM);
5682 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
5686 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
5688 if (!vmx->guest_msrs) {
5692 vmx->loaded_vmcs = &vmx->vmcs01;
5693 vmx->loaded_vmcs->vmcs = alloc_vmcs();
5694 if (!vmx->loaded_vmcs->vmcs)
5697 kvm_cpu_vmxon(__pa(per_cpu(vmxarea, raw_smp_processor_id())));
5698 loaded_vmcs_init(vmx->loaded_vmcs);
5703 vmx_vcpu_load(&vmx->vcpu, cpu);
5704 vmx->vcpu.cpu = cpu;
5705 err = vmx_vcpu_setup(vmx);
5706 vmx_vcpu_put(&vmx->vcpu);
5710 if (vm_need_virtualize_apic_accesses(kvm))
5711 err = alloc_apic_access_page(kvm);
5716 if (!kvm->arch.ept_identity_map_addr)
5717 kvm->arch.ept_identity_map_addr =
5718 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
5720 if (alloc_identity_pagetable(kvm) != 0)
5722 if (!init_rmode_identity_map(kvm))
5726 vmx->nested.current_vmptr = -1ull;
5727 vmx->nested.current_vmcs12 = NULL;
5732 free_vmcs(vmx->loaded_vmcs->vmcs);
5734 kfree(vmx->guest_msrs);
5736 kvm_vcpu_uninit(&vmx->vcpu);
5739 kmem_cache_free(kvm_vcpu_cache, vmx);
5740 return ERR_PTR(err);
5743 static void __init vmx_check_processor_compat(void *rtn)
5745 struct vmcs_config vmcs_conf;
5748 if (setup_vmcs_config(&vmcs_conf) < 0)
5750 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
5751 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
5752 smp_processor_id());
5757 static int get_ept_level(void)
5759 return VMX_EPT_DEFAULT_GAW + 1;
5762 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
5766 /* For VT-d and EPT combination
5767 * 1. MMIO: always map as UC
5769 * a. VT-d without snooping control feature: can't guarantee the
5770 * result, try to trust guest.
5771 * b. VT-d with snooping control feature: snooping control feature of
5772 * VT-d engine can guarantee the cache correctness. Just set it
5773 * to WB to keep consistent with host. So the same as item 3.
5774 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
5775 * consistent with host MTRR
5778 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
5779 else if (vcpu->kvm->arch.iommu_domain &&
5780 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY))
5781 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
5782 VMX_EPT_MT_EPTE_SHIFT;
5784 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
5790 #define _ER(x) { EXIT_REASON_##x, #x }
5792 static const struct trace_print_flags vmx_exit_reasons_str[] = {
5794 _ER(EXTERNAL_INTERRUPT),
5796 _ER(PENDING_INTERRUPT),
5816 _ER(IO_INSTRUCTION),
5819 _ER(MWAIT_INSTRUCTION),
5820 _ER(MONITOR_INSTRUCTION),
5821 _ER(PAUSE_INSTRUCTION),
5822 _ER(MCE_DURING_VMENTRY),
5823 _ER(TPR_BELOW_THRESHOLD),
5833 static int vmx_get_lpage_level(void)
5835 if (enable_ept && !cpu_has_vmx_ept_1g_page())
5836 return PT_DIRECTORY_LEVEL;
5838 /* For shadow and EPT supported 1GB page */
5839 return PT_PDPE_LEVEL;
5842 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
5844 struct kvm_cpuid_entry2 *best;
5845 struct vcpu_vmx *vmx = to_vmx(vcpu);
5848 vmx->rdtscp_enabled = false;
5849 if (vmx_rdtscp_supported()) {
5850 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5851 if (exec_control & SECONDARY_EXEC_RDTSCP) {
5852 best = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
5853 if (best && (best->edx & bit(X86_FEATURE_RDTSCP)))
5854 vmx->rdtscp_enabled = true;
5856 exec_control &= ~SECONDARY_EXEC_RDTSCP;
5857 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
5864 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
5869 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
5870 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
5871 * with L0's requirements for its guest (a.k.a. vmsc01), so we can run the L2
5872 * guest in a way that will both be appropriate to L1's requests, and our
5873 * needs. In addition to modifying the active vmcs (which is vmcs02), this
5874 * function also has additional necessary side-effects, like setting various
5875 * vcpu->arch fields.
5877 static void prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
5879 struct vcpu_vmx *vmx = to_vmx(vcpu);
5882 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
5883 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
5884 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
5885 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
5886 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
5887 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
5888 vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
5889 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
5890 vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
5891 vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
5892 vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
5893 vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
5894 vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
5895 vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
5896 vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
5897 vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
5898 vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
5899 vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
5900 vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
5901 vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
5902 vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
5903 vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
5904 vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
5905 vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
5906 vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
5907 vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
5908 vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
5909 vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
5910 vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
5911 vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
5912 vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
5913 vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
5914 vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
5915 vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
5916 vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
5917 vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
5919 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
5920 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
5921 vmcs12->vm_entry_intr_info_field);
5922 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
5923 vmcs12->vm_entry_exception_error_code);
5924 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
5925 vmcs12->vm_entry_instruction_len);
5926 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
5927 vmcs12->guest_interruptibility_info);
5928 vmcs_write32(GUEST_ACTIVITY_STATE, vmcs12->guest_activity_state);
5929 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
5930 vmcs_writel(GUEST_DR7, vmcs12->guest_dr7);
5931 vmcs_writel(GUEST_RFLAGS, vmcs12->guest_rflags);
5932 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
5933 vmcs12->guest_pending_dbg_exceptions);
5934 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
5935 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
5937 vmcs_write64(VMCS_LINK_POINTER, -1ull);
5939 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
5940 (vmcs_config.pin_based_exec_ctrl |
5941 vmcs12->pin_based_vm_exec_control));
5944 * Whether page-faults are trapped is determined by a combination of
5945 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
5946 * If enable_ept, L0 doesn't care about page faults and we should
5947 * set all of these to L1's desires. However, if !enable_ept, L0 does
5948 * care about (at least some) page faults, and because it is not easy
5949 * (if at all possible?) to merge L0 and L1's desires, we simply ask
5950 * to exit on each and every L2 page fault. This is done by setting
5951 * MASK=MATCH=0 and (see below) EB.PF=1.
5952 * Note that below we don't need special code to set EB.PF beyond the
5953 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
5954 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
5955 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
5957 * A problem with this approach (when !enable_ept) is that L1 may be
5958 * injected with more page faults than it asked for. This could have
5959 * caused problems, but in practice existing hypervisors don't care.
5960 * To fix this, we will need to emulate the PFEC checking (on the L1
5961 * page tables), using walk_addr(), when injecting PFs to L1.
5963 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
5964 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
5965 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
5966 enable_ept ? vmcs12->page_fault_error_code_match : 0);
5968 if (cpu_has_secondary_exec_ctrls()) {
5969 u32 exec_control = vmx_secondary_exec_control(vmx);
5970 if (!vmx->rdtscp_enabled)
5971 exec_control &= ~SECONDARY_EXEC_RDTSCP;
5972 /* Take the following fields only from vmcs12 */
5973 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
5974 if (nested_cpu_has(vmcs12,
5975 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
5976 exec_control |= vmcs12->secondary_vm_exec_control;
5978 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) {
5980 * Translate L1 physical address to host physical
5981 * address for vmcs02. Keep the page pinned, so this
5982 * physical address remains valid. We keep a reference
5983 * to it so we can release it later.
5985 if (vmx->nested.apic_access_page) /* shouldn't happen */
5986 nested_release_page(vmx->nested.apic_access_page);
5987 vmx->nested.apic_access_page =
5988 nested_get_page(vcpu, vmcs12->apic_access_addr);
5990 * If translation failed, no matter: This feature asks
5991 * to exit when accessing the given address, and if it
5992 * can never be accessed, this feature won't do
5995 if (!vmx->nested.apic_access_page)
5997 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
5999 vmcs_write64(APIC_ACCESS_ADDR,
6000 page_to_phys(vmx->nested.apic_access_page));
6003 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
6008 * Set host-state according to L0's settings (vmcs12 is irrelevant here)
6009 * Some constant fields are set here by vmx_set_constant_host_state().
6010 * Other fields are different per CPU, and will be set later when
6011 * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
6013 vmx_set_constant_host_state();
6016 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
6017 * entry, but only if the current (host) sp changed from the value
6018 * we wrote last (vmx->host_rsp). This cache is no longer relevant
6019 * if we switch vmcs, and rather than hold a separate cache per vmcs,
6020 * here we just force the write to happen on entry.
6024 exec_control = vmx_exec_control(vmx); /* L0's desires */
6025 exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
6026 exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
6027 exec_control &= ~CPU_BASED_TPR_SHADOW;
6028 exec_control |= vmcs12->cpu_based_vm_exec_control;
6030 * Merging of IO and MSR bitmaps not currently supported.
6031 * Rather, exit every time.
6033 exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
6034 exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
6035 exec_control |= CPU_BASED_UNCOND_IO_EXITING;
6037 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
6039 /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
6040 * bitwise-or of what L1 wants to trap for L2, and what we want to
6041 * trap. Note that CR0.TS also needs updating - we do this later.
6043 update_exception_bitmap(vcpu);
6044 vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
6045 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
6047 /* Note: IA32_MODE, LOAD_IA32_EFER are modified by vmx_set_efer below */
6048 vmcs_write32(VM_EXIT_CONTROLS,
6049 vmcs12->vm_exit_controls | vmcs_config.vmexit_ctrl);
6050 vmcs_write32(VM_ENTRY_CONTROLS, vmcs12->vm_entry_controls |
6051 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
6053 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)
6054 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
6055 else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
6056 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
6059 set_cr4_guest_host_mask(vmx);
6061 vmcs_write64(TSC_OFFSET,
6062 vmx->nested.vmcs01_tsc_offset + vmcs12->tsc_offset);
6066 * Trivially support vpid by letting L2s share their parent
6067 * L1's vpid. TODO: move to a more elaborate solution, giving
6068 * each L2 its own vpid and exposing the vpid feature to L1.
6070 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
6071 vmx_flush_tlb(vcpu);
6074 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)
6075 vcpu->arch.efer = vmcs12->guest_ia32_efer;
6076 if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
6077 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
6079 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
6080 /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
6081 vmx_set_efer(vcpu, vcpu->arch.efer);
6084 * This sets GUEST_CR0 to vmcs12->guest_cr0, with possibly a modified
6085 * TS bit (for lazy fpu) and bits which we consider mandatory enabled.
6086 * The CR0_READ_SHADOW is what L2 should have expected to read given
6087 * the specifications by L1; It's not enough to take
6088 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
6089 * have more bits than L1 expected.
6091 vmx_set_cr0(vcpu, vmcs12->guest_cr0);
6092 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
6094 vmx_set_cr4(vcpu, vmcs12->guest_cr4);
6095 vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
6097 /* shadow page tables on either EPT or shadow page tables */
6098 kvm_set_cr3(vcpu, vmcs12->guest_cr3);
6099 kvm_mmu_reset_context(vcpu);
6101 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
6102 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
6106 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
6107 * for running an L2 nested guest.
6109 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
6111 struct vmcs12 *vmcs12;
6112 struct vcpu_vmx *vmx = to_vmx(vcpu);
6114 struct loaded_vmcs *vmcs02;
6116 if (!nested_vmx_check_permission(vcpu) ||
6117 !nested_vmx_check_vmcs12(vcpu))
6120 skip_emulated_instruction(vcpu);
6121 vmcs12 = get_vmcs12(vcpu);
6124 * The nested entry process starts with enforcing various prerequisites
6125 * on vmcs12 as required by the Intel SDM, and act appropriately when
6126 * they fail: As the SDM explains, some conditions should cause the
6127 * instruction to fail, while others will cause the instruction to seem
6128 * to succeed, but return an EXIT_REASON_INVALID_STATE.
6129 * To speed up the normal (success) code path, we should avoid checking
6130 * for misconfigurations which will anyway be caught by the processor
6131 * when using the merged vmcs02.
6133 if (vmcs12->launch_state == launch) {
6134 nested_vmx_failValid(vcpu,
6135 launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
6136 : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
6140 if ((vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_MSR_BITMAPS) &&
6141 !IS_ALIGNED(vmcs12->msr_bitmap, PAGE_SIZE)) {
6142 /*TODO: Also verify bits beyond physical address width are 0*/
6143 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
6147 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
6148 !IS_ALIGNED(vmcs12->apic_access_addr, PAGE_SIZE)) {
6149 /*TODO: Also verify bits beyond physical address width are 0*/
6150 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
6154 if (vmcs12->vm_entry_msr_load_count > 0 ||
6155 vmcs12->vm_exit_msr_load_count > 0 ||
6156 vmcs12->vm_exit_msr_store_count > 0) {
6157 if (printk_ratelimit())
6159 "%s: VMCS MSR_{LOAD,STORE} unsupported\n", __func__);
6160 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
6164 if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
6165 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high) ||
6166 !vmx_control_verify(vmcs12->secondary_vm_exec_control,
6167 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high) ||
6168 !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
6169 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high) ||
6170 !vmx_control_verify(vmcs12->vm_exit_controls,
6171 nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high) ||
6172 !vmx_control_verify(vmcs12->vm_entry_controls,
6173 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high))
6175 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
6179 if (((vmcs12->host_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
6180 ((vmcs12->host_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
6181 nested_vmx_failValid(vcpu,
6182 VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
6186 if (((vmcs12->guest_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
6187 ((vmcs12->guest_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
6188 nested_vmx_entry_failure(vcpu, vmcs12,
6189 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
6192 if (vmcs12->vmcs_link_pointer != -1ull) {
6193 nested_vmx_entry_failure(vcpu, vmcs12,
6194 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_VMCS_LINK_PTR);
6199 * We're finally done with prerequisite checking, and can start with
6203 vmcs02 = nested_get_current_vmcs02(vmx);
6207 enter_guest_mode(vcpu);
6209 vmx->nested.vmcs01_tsc_offset = vmcs_read64(TSC_OFFSET);
6212 vmx->loaded_vmcs = vmcs02;
6214 vmx_vcpu_load(vcpu, cpu);
6218 vmcs12->launch_state = 1;
6220 prepare_vmcs02(vcpu, vmcs12);
6223 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
6224 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
6225 * returned as far as L1 is concerned. It will only return (and set
6226 * the success flag) when L2 exits (see nested_vmx_vmexit()).
6232 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
6233 * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
6234 * This function returns the new value we should put in vmcs12.guest_cr0.
6235 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
6236 * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
6237 * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
6238 * didn't trap the bit, because if L1 did, so would L0).
6239 * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have
6240 * been modified by L2, and L1 knows it. So just leave the old value of
6241 * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
6242 * isn't relevant, because if L0 traps this bit it can set it to anything.
6243 * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
6244 * changed these bits, and therefore they need to be updated, but L0
6245 * didn't necessarily allow them to be changed in GUEST_CR0 - and rather
6246 * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
6248 static inline unsigned long
6249 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
6252 /*1*/ (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
6253 /*2*/ (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
6254 /*3*/ (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
6255 vcpu->arch.cr0_guest_owned_bits));
6258 static inline unsigned long
6259 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
6262 /*1*/ (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
6263 /*2*/ (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
6264 /*3*/ (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
6265 vcpu->arch.cr4_guest_owned_bits));
6269 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
6270 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
6271 * and this function updates it to reflect the changes to the guest state while
6272 * L2 was running (and perhaps made some exits which were handled directly by L0
6273 * without going back to L1), and to reflect the exit reason.
6274 * Note that we do not have to copy here all VMCS fields, just those that
6275 * could have changed by the L2 guest or the exit - i.e., the guest-state and
6276 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
6277 * which already writes to vmcs12 directly.
6279 void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
6281 /* update guest state fields: */
6282 vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
6283 vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
6285 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
6286 vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6287 vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
6288 vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
6290 vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
6291 vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
6292 vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
6293 vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
6294 vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
6295 vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
6296 vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
6297 vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
6298 vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
6299 vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
6300 vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
6301 vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
6302 vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
6303 vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
6304 vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
6305 vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
6306 vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
6307 vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
6308 vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
6309 vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
6310 vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
6311 vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
6312 vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
6313 vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
6314 vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
6315 vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
6316 vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
6317 vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
6318 vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
6319 vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
6320 vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
6321 vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
6322 vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
6323 vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
6324 vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
6325 vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
6327 vmcs12->guest_activity_state = vmcs_read32(GUEST_ACTIVITY_STATE);
6328 vmcs12->guest_interruptibility_info =
6329 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
6330 vmcs12->guest_pending_dbg_exceptions =
6331 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
6333 /* TODO: These cannot have changed unless we have MSR bitmaps and
6334 * the relevant bit asks not to trap the change */
6335 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
6336 if (vmcs12->vm_entry_controls & VM_EXIT_SAVE_IA32_PAT)
6337 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
6338 vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
6339 vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
6340 vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
6342 /* update exit information fields: */
6344 vmcs12->vm_exit_reason = vmcs_read32(VM_EXIT_REASON);
6345 vmcs12->exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6347 vmcs12->vm_exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6348 vmcs12->vm_exit_intr_error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
6349 vmcs12->idt_vectoring_info_field =
6350 vmcs_read32(IDT_VECTORING_INFO_FIELD);
6351 vmcs12->idt_vectoring_error_code =
6352 vmcs_read32(IDT_VECTORING_ERROR_CODE);
6353 vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
6354 vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6356 /* clear vm-entry fields which are to be cleared on exit */
6357 if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
6358 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
6362 * A part of what we need to when the nested L2 guest exits and we want to
6363 * run its L1 parent, is to reset L1's guest state to the host state specified
6365 * This function is to be called not only on normal nested exit, but also on
6366 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
6367 * Failures During or After Loading Guest State").
6368 * This function should be called when the active VMCS is L1's (vmcs01).
6370 void load_vmcs12_host_state(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
6372 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
6373 vcpu->arch.efer = vmcs12->host_ia32_efer;
6374 if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
6375 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
6377 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
6378 vmx_set_efer(vcpu, vcpu->arch.efer);
6380 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
6381 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
6383 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
6384 * actually changed, because it depends on the current state of
6385 * fpu_active (which may have changed).
6386 * Note that vmx_set_cr0 refers to efer set above.
6388 kvm_set_cr0(vcpu, vmcs12->host_cr0);
6390 * If we did fpu_activate()/fpu_deactivate() during L2's run, we need
6391 * to apply the same changes to L1's vmcs. We just set cr0 correctly,
6392 * but we also need to update cr0_guest_host_mask and exception_bitmap.
6394 update_exception_bitmap(vcpu);
6395 vcpu->arch.cr0_guest_owned_bits = (vcpu->fpu_active ? X86_CR0_TS : 0);
6396 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
6399 * Note that CR4_GUEST_HOST_MASK is already set in the original vmcs01
6400 * (KVM doesn't change it)- no reason to call set_cr4_guest_host_mask();
6402 vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
6403 kvm_set_cr4(vcpu, vmcs12->host_cr4);
6405 /* shadow page tables on either EPT or shadow page tables */
6406 kvm_set_cr3(vcpu, vmcs12->host_cr3);
6407 kvm_mmu_reset_context(vcpu);
6411 * Trivially support vpid by letting L2s share their parent
6412 * L1's vpid. TODO: move to a more elaborate solution, giving
6413 * each L2 its own vpid and exposing the vpid feature to L1.
6415 vmx_flush_tlb(vcpu);
6419 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
6420 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
6421 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
6422 vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
6423 vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
6424 vmcs_writel(GUEST_TR_BASE, vmcs12->host_tr_base);
6425 vmcs_writel(GUEST_GS_BASE, vmcs12->host_gs_base);
6426 vmcs_writel(GUEST_FS_BASE, vmcs12->host_fs_base);
6427 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->host_es_selector);
6428 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->host_cs_selector);
6429 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->host_ss_selector);
6430 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->host_ds_selector);
6431 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->host_fs_selector);
6432 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->host_gs_selector);
6433 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->host_tr_selector);
6435 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT)
6436 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
6437 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
6438 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
6439 vmcs12->host_ia32_perf_global_ctrl);
6443 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
6444 * and modify vmcs12 to make it see what it would expect to see there if
6445 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
6447 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu)
6449 struct vcpu_vmx *vmx = to_vmx(vcpu);
6451 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6453 leave_guest_mode(vcpu);
6454 prepare_vmcs12(vcpu, vmcs12);
6457 vmx->loaded_vmcs = &vmx->vmcs01;
6459 vmx_vcpu_load(vcpu, cpu);
6463 /* if no vmcs02 cache requested, remove the one we used */
6464 if (VMCS02_POOL_SIZE == 0)
6465 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
6467 load_vmcs12_host_state(vcpu, vmcs12);
6469 /* Update TSC_OFFSET if vmx_adjust_tsc_offset() was used while L2 ran */
6470 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
6472 /* This is needed for same reason as it was needed in prepare_vmcs02 */
6475 /* Unpin physical memory we referred to in vmcs02 */
6476 if (vmx->nested.apic_access_page) {
6477 nested_release_page(vmx->nested.apic_access_page);
6478 vmx->nested.apic_access_page = 0;
6482 * Exiting from L2 to L1, we're now back to L1 which thinks it just
6483 * finished a VMLAUNCH or VMRESUME instruction, so we need to set the
6484 * success or failure flag accordingly.
6486 if (unlikely(vmx->fail)) {
6488 nested_vmx_failValid(vcpu, vmcs_read32(VM_INSTRUCTION_ERROR));
6490 nested_vmx_succeed(vcpu);
6494 * L1's failure to enter L2 is a subset of a normal exit, as explained in
6495 * 23.7 "VM-entry failures during or after loading guest state" (this also
6496 * lists the acceptable exit-reason and exit-qualification parameters).
6497 * It should only be called before L2 actually succeeded to run, and when
6498 * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
6500 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
6501 struct vmcs12 *vmcs12,
6502 u32 reason, unsigned long qualification)
6504 load_vmcs12_host_state(vcpu, vmcs12);
6505 vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
6506 vmcs12->exit_qualification = qualification;
6507 nested_vmx_succeed(vcpu);
6510 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
6511 struct x86_instruction_info *info,
6512 enum x86_intercept_stage stage)
6514 return X86EMUL_CONTINUE;
6517 static struct kvm_x86_ops vmx_x86_ops = {
6518 .cpu_has_kvm_support = cpu_has_kvm_support,
6519 .disabled_by_bios = vmx_disabled_by_bios,
6520 .hardware_setup = hardware_setup,
6521 .hardware_unsetup = hardware_unsetup,
6522 .check_processor_compatibility = vmx_check_processor_compat,
6523 .hardware_enable = hardware_enable,
6524 .hardware_disable = hardware_disable,
6525 .cpu_has_accelerated_tpr = report_flexpriority,
6527 .vcpu_create = vmx_create_vcpu,
6528 .vcpu_free = vmx_free_vcpu,
6529 .vcpu_reset = vmx_vcpu_reset,
6531 .prepare_guest_switch = vmx_save_host_state,
6532 .vcpu_load = vmx_vcpu_load,
6533 .vcpu_put = vmx_vcpu_put,
6535 .set_guest_debug = set_guest_debug,
6536 .get_msr = vmx_get_msr,
6537 .set_msr = vmx_set_msr,
6538 .get_segment_base = vmx_get_segment_base,
6539 .get_segment = vmx_get_segment,
6540 .set_segment = vmx_set_segment,
6541 .get_cpl = vmx_get_cpl,
6542 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
6543 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
6544 .decache_cr3 = vmx_decache_cr3,
6545 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
6546 .set_cr0 = vmx_set_cr0,
6547 .set_cr3 = vmx_set_cr3,
6548 .set_cr4 = vmx_set_cr4,
6549 .set_efer = vmx_set_efer,
6550 .get_idt = vmx_get_idt,
6551 .set_idt = vmx_set_idt,
6552 .get_gdt = vmx_get_gdt,
6553 .set_gdt = vmx_set_gdt,
6554 .set_dr7 = vmx_set_dr7,
6555 .cache_reg = vmx_cache_reg,
6556 .get_rflags = vmx_get_rflags,
6557 .set_rflags = vmx_set_rflags,
6558 .fpu_activate = vmx_fpu_activate,
6559 .fpu_deactivate = vmx_fpu_deactivate,
6561 .tlb_flush = vmx_flush_tlb,
6563 .run = vmx_vcpu_run,
6564 .handle_exit = vmx_handle_exit,
6565 .skip_emulated_instruction = skip_emulated_instruction,
6566 .set_interrupt_shadow = vmx_set_interrupt_shadow,
6567 .get_interrupt_shadow = vmx_get_interrupt_shadow,
6568 .patch_hypercall = vmx_patch_hypercall,
6569 .set_irq = vmx_inject_irq,
6570 .set_nmi = vmx_inject_nmi,
6571 .queue_exception = vmx_queue_exception,
6572 .cancel_injection = vmx_cancel_injection,
6573 .interrupt_allowed = vmx_interrupt_allowed,
6574 .nmi_allowed = vmx_nmi_allowed,
6575 .get_nmi_mask = vmx_get_nmi_mask,
6576 .set_nmi_mask = vmx_set_nmi_mask,
6577 .enable_nmi_window = enable_nmi_window,
6578 .enable_irq_window = enable_irq_window,
6579 .update_cr8_intercept = update_cr8_intercept,
6581 .set_tss_addr = vmx_set_tss_addr,
6582 .get_tdp_level = get_ept_level,
6583 .get_mt_mask = vmx_get_mt_mask,
6585 .get_exit_info = vmx_get_exit_info,
6586 .exit_reasons_str = vmx_exit_reasons_str,
6588 .get_lpage_level = vmx_get_lpage_level,
6590 .cpuid_update = vmx_cpuid_update,
6592 .rdtscp_supported = vmx_rdtscp_supported,
6594 .set_supported_cpuid = vmx_set_supported_cpuid,
6596 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
6598 .set_tsc_khz = vmx_set_tsc_khz,
6599 .write_tsc_offset = vmx_write_tsc_offset,
6600 .adjust_tsc_offset = vmx_adjust_tsc_offset,
6601 .compute_tsc_offset = vmx_compute_tsc_offset,
6603 .set_tdp_cr3 = vmx_set_cr3,
6605 .check_intercept = vmx_check_intercept,
6608 static int __init vmx_init(void)
6612 rdmsrl_safe(MSR_EFER, &host_efer);
6614 for (i = 0; i < NR_VMX_MSR; ++i)
6615 kvm_define_shared_msr(i, vmx_msr_index[i]);
6617 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
6618 if (!vmx_io_bitmap_a)
6621 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
6622 if (!vmx_io_bitmap_b) {
6627 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
6628 if (!vmx_msr_bitmap_legacy) {
6633 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
6634 if (!vmx_msr_bitmap_longmode) {
6640 * Allow direct access to the PC debug port (it is often used for I/O
6641 * delays, but the vmexits simply slow things down).
6643 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
6644 clear_bit(0x80, vmx_io_bitmap_a);
6646 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
6648 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
6649 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
6651 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
6653 r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
6654 __alignof__(struct vcpu_vmx), THIS_MODULE);
6658 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
6659 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
6660 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
6661 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
6662 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
6663 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
6666 bypass_guest_pf = 0;
6667 kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull,
6668 VMX_EPT_EXECUTABLE_MASK);
6673 if (bypass_guest_pf)
6674 kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull);
6679 free_page((unsigned long)vmx_msr_bitmap_longmode);
6681 free_page((unsigned long)vmx_msr_bitmap_legacy);
6683 free_page((unsigned long)vmx_io_bitmap_b);
6685 free_page((unsigned long)vmx_io_bitmap_a);
6689 static void __exit vmx_exit(void)
6691 free_page((unsigned long)vmx_msr_bitmap_legacy);
6692 free_page((unsigned long)vmx_msr_bitmap_longmode);
6693 free_page((unsigned long)vmx_io_bitmap_b);
6694 free_page((unsigned long)vmx_io_bitmap_a);
6699 module_init(vmx_init)
6700 module_exit(vmx_exit)