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.
23 #include <linux/kvm_host.h>
24 #include <linux/module.h>
25 #include <linux/kernel.h>
27 #include <linux/highmem.h>
28 #include <linux/sched.h>
29 #include <linux/moduleparam.h>
30 #include <linux/mod_devicetable.h>
31 #include <linux/ftrace_event.h>
32 #include <linux/slab.h>
33 #include <linux/tboot.h>
34 #include "kvm_cache_regs.h"
40 #include <asm/virtext.h>
44 #include <asm/perf_event.h>
45 #include <asm/kexec.h>
49 #define __ex(x) __kvm_handle_fault_on_reboot(x)
50 #define __ex_clear(x, reg) \
51 ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
53 MODULE_AUTHOR("Qumranet");
54 MODULE_LICENSE("GPL");
56 static const struct x86_cpu_id vmx_cpu_id[] = {
57 X86_FEATURE_MATCH(X86_FEATURE_VMX),
60 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
62 static bool __read_mostly enable_vpid = 1;
63 module_param_named(vpid, enable_vpid, bool, 0444);
65 static bool __read_mostly flexpriority_enabled = 1;
66 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
68 static bool __read_mostly enable_ept = 1;
69 module_param_named(ept, enable_ept, bool, S_IRUGO);
71 static bool __read_mostly enable_unrestricted_guest = 1;
72 module_param_named(unrestricted_guest,
73 enable_unrestricted_guest, bool, S_IRUGO);
75 static bool __read_mostly enable_ept_ad_bits = 1;
76 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
78 static bool __read_mostly emulate_invalid_guest_state = true;
79 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
81 static bool __read_mostly vmm_exclusive = 1;
82 module_param(vmm_exclusive, bool, S_IRUGO);
84 static bool __read_mostly fasteoi = 1;
85 module_param(fasteoi, bool, S_IRUGO);
87 static bool __read_mostly enable_apicv = 1;
88 module_param(enable_apicv, bool, S_IRUGO);
90 static bool __read_mostly enable_shadow_vmcs = 1;
91 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
93 * If nested=1, nested virtualization is supported, i.e., guests may use
94 * VMX and be a hypervisor for its own guests. If nested=0, guests may not
95 * use VMX instructions.
97 static bool __read_mostly nested = 0;
98 module_param(nested, bool, S_IRUGO);
100 #define KVM_GUEST_CR0_MASK (X86_CR0_NW | X86_CR0_CD)
101 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST (X86_CR0_WP | X86_CR0_NE)
102 #define KVM_VM_CR0_ALWAYS_ON \
103 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
104 #define KVM_CR4_GUEST_OWNED_BITS \
105 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
106 | X86_CR4_OSXMMEXCPT)
108 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
109 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
111 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
114 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
115 * ple_gap: upper bound on the amount of time between two successive
116 * executions of PAUSE in a loop. Also indicate if ple enabled.
117 * According to test, this time is usually smaller than 128 cycles.
118 * ple_window: upper bound on the amount of time a guest is allowed to execute
119 * in a PAUSE loop. Tests indicate that most spinlocks are held for
120 * less than 2^12 cycles
121 * Time is measured based on a counter that runs at the same rate as the TSC,
122 * refer SDM volume 3b section 21.6.13 & 22.1.3.
124 #define KVM_VMX_DEFAULT_PLE_GAP 128
125 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
126 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
127 module_param(ple_gap, int, S_IRUGO);
129 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
130 module_param(ple_window, int, S_IRUGO);
132 extern const ulong vmx_return;
134 #define NR_AUTOLOAD_MSRS 8
135 #define VMCS02_POOL_SIZE 1
144 * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
145 * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
146 * loaded on this CPU (so we can clear them if the CPU goes down).
152 struct list_head loaded_vmcss_on_cpu_link;
155 struct shared_msr_entry {
162 * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
163 * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
164 * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
165 * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
166 * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
167 * More than one of these structures may exist, if L1 runs multiple L2 guests.
168 * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
169 * underlying hardware which will be used to run L2.
170 * This structure is packed to ensure that its layout is identical across
171 * machines (necessary for live migration).
172 * If there are changes in this struct, VMCS12_REVISION must be changed.
174 typedef u64 natural_width;
175 struct __packed vmcs12 {
176 /* According to the Intel spec, a VMCS region must start with the
177 * following two fields. Then follow implementation-specific data.
182 u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
183 u32 padding[7]; /* room for future expansion */
188 u64 vm_exit_msr_store_addr;
189 u64 vm_exit_msr_load_addr;
190 u64 vm_entry_msr_load_addr;
192 u64 virtual_apic_page_addr;
193 u64 apic_access_addr;
195 u64 guest_physical_address;
196 u64 vmcs_link_pointer;
197 u64 guest_ia32_debugctl;
200 u64 guest_ia32_perf_global_ctrl;
207 u64 host_ia32_perf_global_ctrl;
208 u64 padding64[8]; /* room for future expansion */
210 * To allow migration of L1 (complete with its L2 guests) between
211 * machines of different natural widths (32 or 64 bit), we cannot have
212 * unsigned long fields with no explict size. We use u64 (aliased
213 * natural_width) instead. Luckily, x86 is little-endian.
215 natural_width cr0_guest_host_mask;
216 natural_width cr4_guest_host_mask;
217 natural_width cr0_read_shadow;
218 natural_width cr4_read_shadow;
219 natural_width cr3_target_value0;
220 natural_width cr3_target_value1;
221 natural_width cr3_target_value2;
222 natural_width cr3_target_value3;
223 natural_width exit_qualification;
224 natural_width guest_linear_address;
225 natural_width guest_cr0;
226 natural_width guest_cr3;
227 natural_width guest_cr4;
228 natural_width guest_es_base;
229 natural_width guest_cs_base;
230 natural_width guest_ss_base;
231 natural_width guest_ds_base;
232 natural_width guest_fs_base;
233 natural_width guest_gs_base;
234 natural_width guest_ldtr_base;
235 natural_width guest_tr_base;
236 natural_width guest_gdtr_base;
237 natural_width guest_idtr_base;
238 natural_width guest_dr7;
239 natural_width guest_rsp;
240 natural_width guest_rip;
241 natural_width guest_rflags;
242 natural_width guest_pending_dbg_exceptions;
243 natural_width guest_sysenter_esp;
244 natural_width guest_sysenter_eip;
245 natural_width host_cr0;
246 natural_width host_cr3;
247 natural_width host_cr4;
248 natural_width host_fs_base;
249 natural_width host_gs_base;
250 natural_width host_tr_base;
251 natural_width host_gdtr_base;
252 natural_width host_idtr_base;
253 natural_width host_ia32_sysenter_esp;
254 natural_width host_ia32_sysenter_eip;
255 natural_width host_rsp;
256 natural_width host_rip;
257 natural_width paddingl[8]; /* room for future expansion */
258 u32 pin_based_vm_exec_control;
259 u32 cpu_based_vm_exec_control;
260 u32 exception_bitmap;
261 u32 page_fault_error_code_mask;
262 u32 page_fault_error_code_match;
263 u32 cr3_target_count;
264 u32 vm_exit_controls;
265 u32 vm_exit_msr_store_count;
266 u32 vm_exit_msr_load_count;
267 u32 vm_entry_controls;
268 u32 vm_entry_msr_load_count;
269 u32 vm_entry_intr_info_field;
270 u32 vm_entry_exception_error_code;
271 u32 vm_entry_instruction_len;
273 u32 secondary_vm_exec_control;
274 u32 vm_instruction_error;
276 u32 vm_exit_intr_info;
277 u32 vm_exit_intr_error_code;
278 u32 idt_vectoring_info_field;
279 u32 idt_vectoring_error_code;
280 u32 vm_exit_instruction_len;
281 u32 vmx_instruction_info;
288 u32 guest_ldtr_limit;
290 u32 guest_gdtr_limit;
291 u32 guest_idtr_limit;
292 u32 guest_es_ar_bytes;
293 u32 guest_cs_ar_bytes;
294 u32 guest_ss_ar_bytes;
295 u32 guest_ds_ar_bytes;
296 u32 guest_fs_ar_bytes;
297 u32 guest_gs_ar_bytes;
298 u32 guest_ldtr_ar_bytes;
299 u32 guest_tr_ar_bytes;
300 u32 guest_interruptibility_info;
301 u32 guest_activity_state;
302 u32 guest_sysenter_cs;
303 u32 host_ia32_sysenter_cs;
304 u32 vmx_preemption_timer_value;
305 u32 padding32[7]; /* room for future expansion */
306 u16 virtual_processor_id;
307 u16 guest_es_selector;
308 u16 guest_cs_selector;
309 u16 guest_ss_selector;
310 u16 guest_ds_selector;
311 u16 guest_fs_selector;
312 u16 guest_gs_selector;
313 u16 guest_ldtr_selector;
314 u16 guest_tr_selector;
315 u16 host_es_selector;
316 u16 host_cs_selector;
317 u16 host_ss_selector;
318 u16 host_ds_selector;
319 u16 host_fs_selector;
320 u16 host_gs_selector;
321 u16 host_tr_selector;
325 * VMCS12_REVISION is an arbitrary id that should be changed if the content or
326 * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
327 * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
329 #define VMCS12_REVISION 0x11e57ed0
332 * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
333 * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
334 * current implementation, 4K are reserved to avoid future complications.
336 #define VMCS12_SIZE 0x1000
338 /* Used to remember the last vmcs02 used for some recently used vmcs12s */
340 struct list_head list;
342 struct loaded_vmcs vmcs02;
346 * The nested_vmx structure is part of vcpu_vmx, and holds information we need
347 * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
350 /* Has the level1 guest done vmxon? */
353 /* The guest-physical address of the current VMCS L1 keeps for L2 */
355 /* The host-usable pointer to the above */
356 struct page *current_vmcs12_page;
357 struct vmcs12 *current_vmcs12;
358 struct vmcs *current_shadow_vmcs;
360 * Indicates if the shadow vmcs must be updated with the
361 * data hold by vmcs12
363 bool sync_shadow_vmcs;
365 /* vmcs02_list cache of VMCSs recently used to run L2 guests */
366 struct list_head vmcs02_pool;
368 u64 vmcs01_tsc_offset;
369 /* L2 must run next, and mustn't decide to exit to L1. */
370 bool nested_run_pending;
372 * Guest pages referred to in vmcs02 with host-physical pointers, so
373 * we must keep them pinned while L2 runs.
375 struct page *apic_access_page;
376 u64 msr_ia32_feature_control;
379 #define POSTED_INTR_ON 0
380 /* Posted-Interrupt Descriptor */
382 u32 pir[8]; /* Posted interrupt requested */
383 u32 control; /* bit 0 of control is outstanding notification bit */
387 static bool pi_test_and_set_on(struct pi_desc *pi_desc)
389 return test_and_set_bit(POSTED_INTR_ON,
390 (unsigned long *)&pi_desc->control);
393 static bool pi_test_and_clear_on(struct pi_desc *pi_desc)
395 return test_and_clear_bit(POSTED_INTR_ON,
396 (unsigned long *)&pi_desc->control);
399 static int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
401 return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
405 struct kvm_vcpu vcpu;
406 unsigned long host_rsp;
409 bool nmi_known_unmasked;
411 u32 idt_vectoring_info;
413 struct shared_msr_entry *guest_msrs;
416 unsigned long host_idt_base;
418 u64 msr_host_kernel_gs_base;
419 u64 msr_guest_kernel_gs_base;
422 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
423 * non-nested (L1) guest, it always points to vmcs01. For a nested
424 * guest (L2), it points to a different VMCS.
426 struct loaded_vmcs vmcs01;
427 struct loaded_vmcs *loaded_vmcs;
428 bool __launched; /* temporary, used in vmx_vcpu_run */
429 struct msr_autoload {
431 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
432 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
436 u16 fs_sel, gs_sel, ldt_sel;
440 int gs_ldt_reload_needed;
441 int fs_reload_needed;
446 struct kvm_segment segs[8];
449 u32 bitmask; /* 4 bits per segment (1 bit per field) */
450 struct kvm_save_segment {
458 bool emulation_required;
460 /* Support for vnmi-less CPUs */
461 int soft_vnmi_blocked;
463 s64 vnmi_blocked_time;
468 /* Posted interrupt descriptor */
469 struct pi_desc pi_desc;
471 /* Support for a guest hypervisor (nested VMX) */
472 struct nested_vmx nested;
475 enum segment_cache_field {
484 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
486 return container_of(vcpu, struct vcpu_vmx, vcpu);
489 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
490 #define FIELD(number, name) [number] = VMCS12_OFFSET(name)
491 #define FIELD64(number, name) [number] = VMCS12_OFFSET(name), \
492 [number##_HIGH] = VMCS12_OFFSET(name)+4
495 static const unsigned long shadow_read_only_fields[] = {
497 * We do NOT shadow fields that are modified when L0
498 * traps and emulates any vmx instruction (e.g. VMPTRLD,
499 * VMXON...) executed by L1.
500 * For example, VM_INSTRUCTION_ERROR is read
501 * by L1 if a vmx instruction fails (part of the error path).
502 * Note the code assumes this logic. If for some reason
503 * we start shadowing these fields then we need to
504 * force a shadow sync when L0 emulates vmx instructions
505 * (e.g. force a sync if VM_INSTRUCTION_ERROR is modified
506 * by nested_vmx_failValid)
510 VM_EXIT_INSTRUCTION_LEN,
511 IDT_VECTORING_INFO_FIELD,
512 IDT_VECTORING_ERROR_CODE,
513 VM_EXIT_INTR_ERROR_CODE,
515 GUEST_LINEAR_ADDRESS,
516 GUEST_PHYSICAL_ADDRESS
518 static const int max_shadow_read_only_fields =
519 ARRAY_SIZE(shadow_read_only_fields);
521 static const unsigned long shadow_read_write_fields[] = {
527 GUEST_INTERRUPTIBILITY_INFO,
539 CPU_BASED_VM_EXEC_CONTROL,
540 VM_ENTRY_EXCEPTION_ERROR_CODE,
541 VM_ENTRY_INTR_INFO_FIELD,
542 VM_ENTRY_INSTRUCTION_LEN,
543 VM_ENTRY_EXCEPTION_ERROR_CODE,
549 static const int max_shadow_read_write_fields =
550 ARRAY_SIZE(shadow_read_write_fields);
552 static const unsigned short vmcs_field_to_offset_table[] = {
553 FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
554 FIELD(GUEST_ES_SELECTOR, guest_es_selector),
555 FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
556 FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
557 FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
558 FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
559 FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
560 FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
561 FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
562 FIELD(HOST_ES_SELECTOR, host_es_selector),
563 FIELD(HOST_CS_SELECTOR, host_cs_selector),
564 FIELD(HOST_SS_SELECTOR, host_ss_selector),
565 FIELD(HOST_DS_SELECTOR, host_ds_selector),
566 FIELD(HOST_FS_SELECTOR, host_fs_selector),
567 FIELD(HOST_GS_SELECTOR, host_gs_selector),
568 FIELD(HOST_TR_SELECTOR, host_tr_selector),
569 FIELD64(IO_BITMAP_A, io_bitmap_a),
570 FIELD64(IO_BITMAP_B, io_bitmap_b),
571 FIELD64(MSR_BITMAP, msr_bitmap),
572 FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
573 FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
574 FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
575 FIELD64(TSC_OFFSET, tsc_offset),
576 FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
577 FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
578 FIELD64(EPT_POINTER, ept_pointer),
579 FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
580 FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
581 FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
582 FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
583 FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
584 FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
585 FIELD64(GUEST_PDPTR0, guest_pdptr0),
586 FIELD64(GUEST_PDPTR1, guest_pdptr1),
587 FIELD64(GUEST_PDPTR2, guest_pdptr2),
588 FIELD64(GUEST_PDPTR3, guest_pdptr3),
589 FIELD64(HOST_IA32_PAT, host_ia32_pat),
590 FIELD64(HOST_IA32_EFER, host_ia32_efer),
591 FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
592 FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
593 FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
594 FIELD(EXCEPTION_BITMAP, exception_bitmap),
595 FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
596 FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
597 FIELD(CR3_TARGET_COUNT, cr3_target_count),
598 FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
599 FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
600 FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
601 FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
602 FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
603 FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
604 FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
605 FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
606 FIELD(TPR_THRESHOLD, tpr_threshold),
607 FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
608 FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
609 FIELD(VM_EXIT_REASON, vm_exit_reason),
610 FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
611 FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
612 FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
613 FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
614 FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
615 FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
616 FIELD(GUEST_ES_LIMIT, guest_es_limit),
617 FIELD(GUEST_CS_LIMIT, guest_cs_limit),
618 FIELD(GUEST_SS_LIMIT, guest_ss_limit),
619 FIELD(GUEST_DS_LIMIT, guest_ds_limit),
620 FIELD(GUEST_FS_LIMIT, guest_fs_limit),
621 FIELD(GUEST_GS_LIMIT, guest_gs_limit),
622 FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
623 FIELD(GUEST_TR_LIMIT, guest_tr_limit),
624 FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
625 FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
626 FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
627 FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
628 FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
629 FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
630 FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
631 FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
632 FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
633 FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
634 FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
635 FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
636 FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
637 FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
638 FIELD(VMX_PREEMPTION_TIMER_VALUE, vmx_preemption_timer_value),
639 FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
640 FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
641 FIELD(CR0_READ_SHADOW, cr0_read_shadow),
642 FIELD(CR4_READ_SHADOW, cr4_read_shadow),
643 FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
644 FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
645 FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
646 FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
647 FIELD(EXIT_QUALIFICATION, exit_qualification),
648 FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
649 FIELD(GUEST_CR0, guest_cr0),
650 FIELD(GUEST_CR3, guest_cr3),
651 FIELD(GUEST_CR4, guest_cr4),
652 FIELD(GUEST_ES_BASE, guest_es_base),
653 FIELD(GUEST_CS_BASE, guest_cs_base),
654 FIELD(GUEST_SS_BASE, guest_ss_base),
655 FIELD(GUEST_DS_BASE, guest_ds_base),
656 FIELD(GUEST_FS_BASE, guest_fs_base),
657 FIELD(GUEST_GS_BASE, guest_gs_base),
658 FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
659 FIELD(GUEST_TR_BASE, guest_tr_base),
660 FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
661 FIELD(GUEST_IDTR_BASE, guest_idtr_base),
662 FIELD(GUEST_DR7, guest_dr7),
663 FIELD(GUEST_RSP, guest_rsp),
664 FIELD(GUEST_RIP, guest_rip),
665 FIELD(GUEST_RFLAGS, guest_rflags),
666 FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
667 FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
668 FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
669 FIELD(HOST_CR0, host_cr0),
670 FIELD(HOST_CR3, host_cr3),
671 FIELD(HOST_CR4, host_cr4),
672 FIELD(HOST_FS_BASE, host_fs_base),
673 FIELD(HOST_GS_BASE, host_gs_base),
674 FIELD(HOST_TR_BASE, host_tr_base),
675 FIELD(HOST_GDTR_BASE, host_gdtr_base),
676 FIELD(HOST_IDTR_BASE, host_idtr_base),
677 FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
678 FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
679 FIELD(HOST_RSP, host_rsp),
680 FIELD(HOST_RIP, host_rip),
682 static const int max_vmcs_field = ARRAY_SIZE(vmcs_field_to_offset_table);
684 static inline short vmcs_field_to_offset(unsigned long field)
686 if (field >= max_vmcs_field || vmcs_field_to_offset_table[field] == 0)
688 return vmcs_field_to_offset_table[field];
691 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
693 return to_vmx(vcpu)->nested.current_vmcs12;
696 static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
698 struct page *page = gfn_to_page(vcpu->kvm, addr >> PAGE_SHIFT);
699 if (is_error_page(page))
705 static void nested_release_page(struct page *page)
707 kvm_release_page_dirty(page);
710 static void nested_release_page_clean(struct page *page)
712 kvm_release_page_clean(page);
715 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu);
716 static u64 construct_eptp(unsigned long root_hpa);
717 static void kvm_cpu_vmxon(u64 addr);
718 static void kvm_cpu_vmxoff(void);
719 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3);
720 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
721 static void vmx_set_segment(struct kvm_vcpu *vcpu,
722 struct kvm_segment *var, int seg);
723 static void vmx_get_segment(struct kvm_vcpu *vcpu,
724 struct kvm_segment *var, int seg);
725 static bool guest_state_valid(struct kvm_vcpu *vcpu);
726 static u32 vmx_segment_access_rights(struct kvm_segment *var);
727 static void vmx_sync_pir_to_irr_dummy(struct kvm_vcpu *vcpu);
728 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx);
729 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx);
731 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
732 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
734 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
735 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
737 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
738 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
740 static unsigned long *vmx_io_bitmap_a;
741 static unsigned long *vmx_io_bitmap_b;
742 static unsigned long *vmx_msr_bitmap_legacy;
743 static unsigned long *vmx_msr_bitmap_longmode;
744 static unsigned long *vmx_msr_bitmap_legacy_x2apic;
745 static unsigned long *vmx_msr_bitmap_longmode_x2apic;
746 static unsigned long *vmx_vmread_bitmap;
747 static unsigned long *vmx_vmwrite_bitmap;
749 static bool cpu_has_load_ia32_efer;
750 static bool cpu_has_load_perf_global_ctrl;
752 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
753 static DEFINE_SPINLOCK(vmx_vpid_lock);
755 static struct vmcs_config {
759 u32 pin_based_exec_ctrl;
760 u32 cpu_based_exec_ctrl;
761 u32 cpu_based_2nd_exec_ctrl;
766 static struct vmx_capability {
771 #define VMX_SEGMENT_FIELD(seg) \
772 [VCPU_SREG_##seg] = { \
773 .selector = GUEST_##seg##_SELECTOR, \
774 .base = GUEST_##seg##_BASE, \
775 .limit = GUEST_##seg##_LIMIT, \
776 .ar_bytes = GUEST_##seg##_AR_BYTES, \
779 static const struct kvm_vmx_segment_field {
784 } kvm_vmx_segment_fields[] = {
785 VMX_SEGMENT_FIELD(CS),
786 VMX_SEGMENT_FIELD(DS),
787 VMX_SEGMENT_FIELD(ES),
788 VMX_SEGMENT_FIELD(FS),
789 VMX_SEGMENT_FIELD(GS),
790 VMX_SEGMENT_FIELD(SS),
791 VMX_SEGMENT_FIELD(TR),
792 VMX_SEGMENT_FIELD(LDTR),
795 static u64 host_efer;
797 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
800 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
801 * away by decrementing the array size.
803 static const u32 vmx_msr_index[] = {
805 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
807 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
809 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
811 static inline bool is_page_fault(u32 intr_info)
813 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
814 INTR_INFO_VALID_MASK)) ==
815 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
818 static inline bool is_no_device(u32 intr_info)
820 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
821 INTR_INFO_VALID_MASK)) ==
822 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
825 static inline bool is_invalid_opcode(u32 intr_info)
827 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
828 INTR_INFO_VALID_MASK)) ==
829 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
832 static inline bool is_external_interrupt(u32 intr_info)
834 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
835 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
838 static inline bool is_machine_check(u32 intr_info)
840 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
841 INTR_INFO_VALID_MASK)) ==
842 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
845 static inline bool cpu_has_vmx_msr_bitmap(void)
847 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
850 static inline bool cpu_has_vmx_tpr_shadow(void)
852 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
855 static inline bool vm_need_tpr_shadow(struct kvm *kvm)
857 return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
860 static inline bool cpu_has_secondary_exec_ctrls(void)
862 return vmcs_config.cpu_based_exec_ctrl &
863 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
866 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
868 return vmcs_config.cpu_based_2nd_exec_ctrl &
869 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
872 static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
874 return vmcs_config.cpu_based_2nd_exec_ctrl &
875 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
878 static inline bool cpu_has_vmx_apic_register_virt(void)
880 return vmcs_config.cpu_based_2nd_exec_ctrl &
881 SECONDARY_EXEC_APIC_REGISTER_VIRT;
884 static inline bool cpu_has_vmx_virtual_intr_delivery(void)
886 return vmcs_config.cpu_based_2nd_exec_ctrl &
887 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
890 static inline bool cpu_has_vmx_posted_intr(void)
892 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_POSTED_INTR;
895 static inline bool cpu_has_vmx_apicv(void)
897 return cpu_has_vmx_apic_register_virt() &&
898 cpu_has_vmx_virtual_intr_delivery() &&
899 cpu_has_vmx_posted_intr();
902 static inline bool cpu_has_vmx_flexpriority(void)
904 return cpu_has_vmx_tpr_shadow() &&
905 cpu_has_vmx_virtualize_apic_accesses();
908 static inline bool cpu_has_vmx_ept_execute_only(void)
910 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
913 static inline bool cpu_has_vmx_eptp_uncacheable(void)
915 return vmx_capability.ept & VMX_EPTP_UC_BIT;
918 static inline bool cpu_has_vmx_eptp_writeback(void)
920 return vmx_capability.ept & VMX_EPTP_WB_BIT;
923 static inline bool cpu_has_vmx_ept_2m_page(void)
925 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
928 static inline bool cpu_has_vmx_ept_1g_page(void)
930 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
933 static inline bool cpu_has_vmx_ept_4levels(void)
935 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
938 static inline bool cpu_has_vmx_ept_ad_bits(void)
940 return vmx_capability.ept & VMX_EPT_AD_BIT;
943 static inline bool cpu_has_vmx_invept_context(void)
945 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
948 static inline bool cpu_has_vmx_invept_global(void)
950 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
953 static inline bool cpu_has_vmx_invvpid_single(void)
955 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
958 static inline bool cpu_has_vmx_invvpid_global(void)
960 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
963 static inline bool cpu_has_vmx_ept(void)
965 return vmcs_config.cpu_based_2nd_exec_ctrl &
966 SECONDARY_EXEC_ENABLE_EPT;
969 static inline bool cpu_has_vmx_unrestricted_guest(void)
971 return vmcs_config.cpu_based_2nd_exec_ctrl &
972 SECONDARY_EXEC_UNRESTRICTED_GUEST;
975 static inline bool cpu_has_vmx_ple(void)
977 return vmcs_config.cpu_based_2nd_exec_ctrl &
978 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
981 static inline bool vm_need_virtualize_apic_accesses(struct kvm *kvm)
983 return flexpriority_enabled && irqchip_in_kernel(kvm);
986 static inline bool cpu_has_vmx_vpid(void)
988 return vmcs_config.cpu_based_2nd_exec_ctrl &
989 SECONDARY_EXEC_ENABLE_VPID;
992 static inline bool cpu_has_vmx_rdtscp(void)
994 return vmcs_config.cpu_based_2nd_exec_ctrl &
995 SECONDARY_EXEC_RDTSCP;
998 static inline bool cpu_has_vmx_invpcid(void)
1000 return vmcs_config.cpu_based_2nd_exec_ctrl &
1001 SECONDARY_EXEC_ENABLE_INVPCID;
1004 static inline bool cpu_has_virtual_nmis(void)
1006 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
1009 static inline bool cpu_has_vmx_wbinvd_exit(void)
1011 return vmcs_config.cpu_based_2nd_exec_ctrl &
1012 SECONDARY_EXEC_WBINVD_EXITING;
1015 static inline bool cpu_has_vmx_shadow_vmcs(void)
1018 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
1019 /* check if the cpu supports writing r/o exit information fields */
1020 if (!(vmx_msr & MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS))
1023 return vmcs_config.cpu_based_2nd_exec_ctrl &
1024 SECONDARY_EXEC_SHADOW_VMCS;
1027 static inline bool report_flexpriority(void)
1029 return flexpriority_enabled;
1032 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
1034 return vmcs12->cpu_based_vm_exec_control & bit;
1037 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
1039 return (vmcs12->cpu_based_vm_exec_control &
1040 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
1041 (vmcs12->secondary_vm_exec_control & bit);
1044 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12,
1045 struct kvm_vcpu *vcpu)
1047 return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
1050 static inline int nested_cpu_has_ept(struct vmcs12 *vmcs12)
1052 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_EPT);
1055 static inline bool is_exception(u32 intr_info)
1057 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1058 == (INTR_TYPE_HARD_EXCEPTION | INTR_INFO_VALID_MASK);
1061 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu);
1062 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
1063 struct vmcs12 *vmcs12,
1064 u32 reason, unsigned long qualification);
1066 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
1070 for (i = 0; i < vmx->nmsrs; ++i)
1071 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
1076 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
1082 } operand = { vpid, 0, gva };
1084 asm volatile (__ex(ASM_VMX_INVVPID)
1085 /* CF==1 or ZF==1 --> rc = -1 */
1086 "; ja 1f ; ud2 ; 1:"
1087 : : "a"(&operand), "c"(ext) : "cc", "memory");
1090 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
1094 } operand = {eptp, gpa};
1096 asm volatile (__ex(ASM_VMX_INVEPT)
1097 /* CF==1 or ZF==1 --> rc = -1 */
1098 "; ja 1f ; ud2 ; 1:\n"
1099 : : "a" (&operand), "c" (ext) : "cc", "memory");
1102 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
1106 i = __find_msr_index(vmx, msr);
1108 return &vmx->guest_msrs[i];
1112 static void vmcs_clear(struct vmcs *vmcs)
1114 u64 phys_addr = __pa(vmcs);
1117 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
1118 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1121 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
1125 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
1127 vmcs_clear(loaded_vmcs->vmcs);
1128 loaded_vmcs->cpu = -1;
1129 loaded_vmcs->launched = 0;
1132 static void vmcs_load(struct vmcs *vmcs)
1134 u64 phys_addr = __pa(vmcs);
1137 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
1138 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1141 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
1147 * This bitmap is used to indicate whether the vmclear
1148 * operation is enabled on all cpus. All disabled by
1151 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
1153 static inline void crash_enable_local_vmclear(int cpu)
1155 cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
1158 static inline void crash_disable_local_vmclear(int cpu)
1160 cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
1163 static inline int crash_local_vmclear_enabled(int cpu)
1165 return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
1168 static void crash_vmclear_local_loaded_vmcss(void)
1170 int cpu = raw_smp_processor_id();
1171 struct loaded_vmcs *v;
1173 if (!crash_local_vmclear_enabled(cpu))
1176 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
1177 loaded_vmcss_on_cpu_link)
1178 vmcs_clear(v->vmcs);
1181 static inline void crash_enable_local_vmclear(int cpu) { }
1182 static inline void crash_disable_local_vmclear(int cpu) { }
1183 #endif /* CONFIG_KEXEC */
1185 static void __loaded_vmcs_clear(void *arg)
1187 struct loaded_vmcs *loaded_vmcs = arg;
1188 int cpu = raw_smp_processor_id();
1190 if (loaded_vmcs->cpu != cpu)
1191 return; /* vcpu migration can race with cpu offline */
1192 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1193 per_cpu(current_vmcs, cpu) = NULL;
1194 crash_disable_local_vmclear(cpu);
1195 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1198 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1199 * is before setting loaded_vmcs->vcpu to -1 which is done in
1200 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1201 * then adds the vmcs into percpu list before it is deleted.
1205 loaded_vmcs_init(loaded_vmcs);
1206 crash_enable_local_vmclear(cpu);
1209 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1211 int cpu = loaded_vmcs->cpu;
1214 smp_call_function_single(cpu,
1215 __loaded_vmcs_clear, loaded_vmcs, 1);
1218 static inline void vpid_sync_vcpu_single(struct vcpu_vmx *vmx)
1223 if (cpu_has_vmx_invvpid_single())
1224 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
1227 static inline void vpid_sync_vcpu_global(void)
1229 if (cpu_has_vmx_invvpid_global())
1230 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1233 static inline void vpid_sync_context(struct vcpu_vmx *vmx)
1235 if (cpu_has_vmx_invvpid_single())
1236 vpid_sync_vcpu_single(vmx);
1238 vpid_sync_vcpu_global();
1241 static inline void ept_sync_global(void)
1243 if (cpu_has_vmx_invept_global())
1244 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1247 static inline void ept_sync_context(u64 eptp)
1250 if (cpu_has_vmx_invept_context())
1251 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1257 static __always_inline unsigned long vmcs_readl(unsigned long field)
1259 unsigned long value;
1261 asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1262 : "=a"(value) : "d"(field) : "cc");
1266 static __always_inline u16 vmcs_read16(unsigned long field)
1268 return vmcs_readl(field);
1271 static __always_inline u32 vmcs_read32(unsigned long field)
1273 return vmcs_readl(field);
1276 static __always_inline u64 vmcs_read64(unsigned long field)
1278 #ifdef CONFIG_X86_64
1279 return vmcs_readl(field);
1281 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
1285 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1287 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1288 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1292 static void vmcs_writel(unsigned long field, unsigned long value)
1296 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1297 : "=q"(error) : "a"(value), "d"(field) : "cc");
1298 if (unlikely(error))
1299 vmwrite_error(field, value);
1302 static void vmcs_write16(unsigned long field, u16 value)
1304 vmcs_writel(field, value);
1307 static void vmcs_write32(unsigned long field, u32 value)
1309 vmcs_writel(field, value);
1312 static void vmcs_write64(unsigned long field, u64 value)
1314 vmcs_writel(field, value);
1315 #ifndef CONFIG_X86_64
1317 vmcs_writel(field+1, value >> 32);
1321 static void vmcs_clear_bits(unsigned long field, u32 mask)
1323 vmcs_writel(field, vmcs_readl(field) & ~mask);
1326 static void vmcs_set_bits(unsigned long field, u32 mask)
1328 vmcs_writel(field, vmcs_readl(field) | mask);
1331 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1333 vmx->segment_cache.bitmask = 0;
1336 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1340 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1342 if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1343 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1344 vmx->segment_cache.bitmask = 0;
1346 ret = vmx->segment_cache.bitmask & mask;
1347 vmx->segment_cache.bitmask |= mask;
1351 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1353 u16 *p = &vmx->segment_cache.seg[seg].selector;
1355 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1356 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1360 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1362 ulong *p = &vmx->segment_cache.seg[seg].base;
1364 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1365 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1369 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1371 u32 *p = &vmx->segment_cache.seg[seg].limit;
1373 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1374 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1378 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1380 u32 *p = &vmx->segment_cache.seg[seg].ar;
1382 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1383 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1387 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1391 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1392 (1u << NM_VECTOR) | (1u << DB_VECTOR);
1393 if ((vcpu->guest_debug &
1394 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1395 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1396 eb |= 1u << BP_VECTOR;
1397 if (to_vmx(vcpu)->rmode.vm86_active)
1400 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1401 if (vcpu->fpu_active)
1402 eb &= ~(1u << NM_VECTOR);
1404 /* When we are running a nested L2 guest and L1 specified for it a
1405 * certain exception bitmap, we must trap the same exceptions and pass
1406 * them to L1. When running L2, we will only handle the exceptions
1407 * specified above if L1 did not want them.
1409 if (is_guest_mode(vcpu))
1410 eb |= get_vmcs12(vcpu)->exception_bitmap;
1412 vmcs_write32(EXCEPTION_BITMAP, eb);
1415 static void clear_atomic_switch_msr_special(unsigned long entry,
1418 vmcs_clear_bits(VM_ENTRY_CONTROLS, entry);
1419 vmcs_clear_bits(VM_EXIT_CONTROLS, exit);
1422 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1425 struct msr_autoload *m = &vmx->msr_autoload;
1429 if (cpu_has_load_ia32_efer) {
1430 clear_atomic_switch_msr_special(VM_ENTRY_LOAD_IA32_EFER,
1431 VM_EXIT_LOAD_IA32_EFER);
1435 case MSR_CORE_PERF_GLOBAL_CTRL:
1436 if (cpu_has_load_perf_global_ctrl) {
1437 clear_atomic_switch_msr_special(
1438 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1439 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1445 for (i = 0; i < m->nr; ++i)
1446 if (m->guest[i].index == msr)
1452 m->guest[i] = m->guest[m->nr];
1453 m->host[i] = m->host[m->nr];
1454 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1455 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1458 static void add_atomic_switch_msr_special(unsigned long entry,
1459 unsigned long exit, unsigned long guest_val_vmcs,
1460 unsigned long host_val_vmcs, u64 guest_val, u64 host_val)
1462 vmcs_write64(guest_val_vmcs, guest_val);
1463 vmcs_write64(host_val_vmcs, host_val);
1464 vmcs_set_bits(VM_ENTRY_CONTROLS, entry);
1465 vmcs_set_bits(VM_EXIT_CONTROLS, exit);
1468 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1469 u64 guest_val, u64 host_val)
1472 struct msr_autoload *m = &vmx->msr_autoload;
1476 if (cpu_has_load_ia32_efer) {
1477 add_atomic_switch_msr_special(VM_ENTRY_LOAD_IA32_EFER,
1478 VM_EXIT_LOAD_IA32_EFER,
1481 guest_val, host_val);
1485 case MSR_CORE_PERF_GLOBAL_CTRL:
1486 if (cpu_has_load_perf_global_ctrl) {
1487 add_atomic_switch_msr_special(
1488 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1489 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1490 GUEST_IA32_PERF_GLOBAL_CTRL,
1491 HOST_IA32_PERF_GLOBAL_CTRL,
1492 guest_val, host_val);
1498 for (i = 0; i < m->nr; ++i)
1499 if (m->guest[i].index == msr)
1502 if (i == NR_AUTOLOAD_MSRS) {
1503 printk_once(KERN_WARNING"Not enough mst switch entries. "
1504 "Can't add msr %x\n", msr);
1506 } else if (i == m->nr) {
1508 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1509 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1512 m->guest[i].index = msr;
1513 m->guest[i].value = guest_val;
1514 m->host[i].index = msr;
1515 m->host[i].value = host_val;
1518 static void reload_tss(void)
1521 * VT restores TR but not its size. Useless.
1523 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1524 struct desc_struct *descs;
1526 descs = (void *)gdt->address;
1527 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
1531 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
1536 guest_efer = vmx->vcpu.arch.efer;
1539 * NX is emulated; LMA and LME handled by hardware; SCE meaningless
1542 ignore_bits = EFER_NX | EFER_SCE;
1543 #ifdef CONFIG_X86_64
1544 ignore_bits |= EFER_LMA | EFER_LME;
1545 /* SCE is meaningful only in long mode on Intel */
1546 if (guest_efer & EFER_LMA)
1547 ignore_bits &= ~(u64)EFER_SCE;
1549 guest_efer &= ~ignore_bits;
1550 guest_efer |= host_efer & ignore_bits;
1551 vmx->guest_msrs[efer_offset].data = guest_efer;
1552 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1554 clear_atomic_switch_msr(vmx, MSR_EFER);
1555 /* On ept, can't emulate nx, and must switch nx atomically */
1556 if (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX)) {
1557 guest_efer = vmx->vcpu.arch.efer;
1558 if (!(guest_efer & EFER_LMA))
1559 guest_efer &= ~EFER_LME;
1560 add_atomic_switch_msr(vmx, MSR_EFER, guest_efer, host_efer);
1567 static unsigned long segment_base(u16 selector)
1569 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1570 struct desc_struct *d;
1571 unsigned long table_base;
1574 if (!(selector & ~3))
1577 table_base = gdt->address;
1579 if (selector & 4) { /* from ldt */
1580 u16 ldt_selector = kvm_read_ldt();
1582 if (!(ldt_selector & ~3))
1585 table_base = segment_base(ldt_selector);
1587 d = (struct desc_struct *)(table_base + (selector & ~7));
1588 v = get_desc_base(d);
1589 #ifdef CONFIG_X86_64
1590 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
1591 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
1596 static inline unsigned long kvm_read_tr_base(void)
1599 asm("str %0" : "=g"(tr));
1600 return segment_base(tr);
1603 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
1605 struct vcpu_vmx *vmx = to_vmx(vcpu);
1608 if (vmx->host_state.loaded)
1611 vmx->host_state.loaded = 1;
1613 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
1614 * allow segment selectors with cpl > 0 or ti == 1.
1616 vmx->host_state.ldt_sel = kvm_read_ldt();
1617 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
1618 savesegment(fs, vmx->host_state.fs_sel);
1619 if (!(vmx->host_state.fs_sel & 7)) {
1620 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
1621 vmx->host_state.fs_reload_needed = 0;
1623 vmcs_write16(HOST_FS_SELECTOR, 0);
1624 vmx->host_state.fs_reload_needed = 1;
1626 savesegment(gs, vmx->host_state.gs_sel);
1627 if (!(vmx->host_state.gs_sel & 7))
1628 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
1630 vmcs_write16(HOST_GS_SELECTOR, 0);
1631 vmx->host_state.gs_ldt_reload_needed = 1;
1634 #ifdef CONFIG_X86_64
1635 savesegment(ds, vmx->host_state.ds_sel);
1636 savesegment(es, vmx->host_state.es_sel);
1639 #ifdef CONFIG_X86_64
1640 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
1641 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
1643 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
1644 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
1647 #ifdef CONFIG_X86_64
1648 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1649 if (is_long_mode(&vmx->vcpu))
1650 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1652 for (i = 0; i < vmx->save_nmsrs; ++i)
1653 kvm_set_shared_msr(vmx->guest_msrs[i].index,
1654 vmx->guest_msrs[i].data,
1655 vmx->guest_msrs[i].mask);
1658 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
1660 if (!vmx->host_state.loaded)
1663 ++vmx->vcpu.stat.host_state_reload;
1664 vmx->host_state.loaded = 0;
1665 #ifdef CONFIG_X86_64
1666 if (is_long_mode(&vmx->vcpu))
1667 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1669 if (vmx->host_state.gs_ldt_reload_needed) {
1670 kvm_load_ldt(vmx->host_state.ldt_sel);
1671 #ifdef CONFIG_X86_64
1672 load_gs_index(vmx->host_state.gs_sel);
1674 loadsegment(gs, vmx->host_state.gs_sel);
1677 if (vmx->host_state.fs_reload_needed)
1678 loadsegment(fs, vmx->host_state.fs_sel);
1679 #ifdef CONFIG_X86_64
1680 if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
1681 loadsegment(ds, vmx->host_state.ds_sel);
1682 loadsegment(es, vmx->host_state.es_sel);
1686 #ifdef CONFIG_X86_64
1687 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1690 * If the FPU is not active (through the host task or
1691 * the guest vcpu), then restore the cr0.TS bit.
1693 if (!user_has_fpu() && !vmx->vcpu.guest_fpu_loaded)
1695 load_gdt(&__get_cpu_var(host_gdt));
1698 static void vmx_load_host_state(struct vcpu_vmx *vmx)
1701 __vmx_load_host_state(vmx);
1706 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1707 * vcpu mutex is already taken.
1709 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1711 struct vcpu_vmx *vmx = to_vmx(vcpu);
1712 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1715 kvm_cpu_vmxon(phys_addr);
1716 else if (vmx->loaded_vmcs->cpu != cpu)
1717 loaded_vmcs_clear(vmx->loaded_vmcs);
1719 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1720 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1721 vmcs_load(vmx->loaded_vmcs->vmcs);
1724 if (vmx->loaded_vmcs->cpu != cpu) {
1725 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1726 unsigned long sysenter_esp;
1728 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1729 local_irq_disable();
1730 crash_disable_local_vmclear(cpu);
1733 * Read loaded_vmcs->cpu should be before fetching
1734 * loaded_vmcs->loaded_vmcss_on_cpu_link.
1735 * See the comments in __loaded_vmcs_clear().
1739 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1740 &per_cpu(loaded_vmcss_on_cpu, cpu));
1741 crash_enable_local_vmclear(cpu);
1745 * Linux uses per-cpu TSS and GDT, so set these when switching
1748 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
1749 vmcs_writel(HOST_GDTR_BASE, gdt->address); /* 22.2.4 */
1751 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1752 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1753 vmx->loaded_vmcs->cpu = cpu;
1757 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1759 __vmx_load_host_state(to_vmx(vcpu));
1760 if (!vmm_exclusive) {
1761 __loaded_vmcs_clear(to_vmx(vcpu)->loaded_vmcs);
1767 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
1771 if (vcpu->fpu_active)
1773 vcpu->fpu_active = 1;
1774 cr0 = vmcs_readl(GUEST_CR0);
1775 cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
1776 cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
1777 vmcs_writel(GUEST_CR0, cr0);
1778 update_exception_bitmap(vcpu);
1779 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
1780 if (is_guest_mode(vcpu))
1781 vcpu->arch.cr0_guest_owned_bits &=
1782 ~get_vmcs12(vcpu)->cr0_guest_host_mask;
1783 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1786 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
1789 * Return the cr0 value that a nested guest would read. This is a combination
1790 * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
1791 * its hypervisor (cr0_read_shadow).
1793 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
1795 return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
1796 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
1798 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
1800 return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
1801 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
1804 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
1806 /* Note that there is no vcpu->fpu_active = 0 here. The caller must
1807 * set this *before* calling this function.
1809 vmx_decache_cr0_guest_bits(vcpu);
1810 vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
1811 update_exception_bitmap(vcpu);
1812 vcpu->arch.cr0_guest_owned_bits = 0;
1813 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1814 if (is_guest_mode(vcpu)) {
1816 * L1's specified read shadow might not contain the TS bit,
1817 * so now that we turned on shadowing of this bit, we need to
1818 * set this bit of the shadow. Like in nested_vmx_run we need
1819 * nested_read_cr0(vmcs12), but vmcs12->guest_cr0 is not yet
1820 * up-to-date here because we just decached cr0.TS (and we'll
1821 * only update vmcs12->guest_cr0 on nested exit).
1823 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1824 vmcs12->guest_cr0 = (vmcs12->guest_cr0 & ~X86_CR0_TS) |
1825 (vcpu->arch.cr0 & X86_CR0_TS);
1826 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
1828 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
1831 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1833 unsigned long rflags, save_rflags;
1835 if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
1836 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1837 rflags = vmcs_readl(GUEST_RFLAGS);
1838 if (to_vmx(vcpu)->rmode.vm86_active) {
1839 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1840 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
1841 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1843 to_vmx(vcpu)->rflags = rflags;
1845 return to_vmx(vcpu)->rflags;
1848 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1850 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1851 to_vmx(vcpu)->rflags = rflags;
1852 if (to_vmx(vcpu)->rmode.vm86_active) {
1853 to_vmx(vcpu)->rmode.save_rflags = rflags;
1854 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1856 vmcs_writel(GUEST_RFLAGS, rflags);
1859 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1861 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1864 if (interruptibility & GUEST_INTR_STATE_STI)
1865 ret |= KVM_X86_SHADOW_INT_STI;
1866 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1867 ret |= KVM_X86_SHADOW_INT_MOV_SS;
1872 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1874 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1875 u32 interruptibility = interruptibility_old;
1877 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1879 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1880 interruptibility |= GUEST_INTR_STATE_MOV_SS;
1881 else if (mask & KVM_X86_SHADOW_INT_STI)
1882 interruptibility |= GUEST_INTR_STATE_STI;
1884 if ((interruptibility != interruptibility_old))
1885 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1888 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
1892 rip = kvm_rip_read(vcpu);
1893 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1894 kvm_rip_write(vcpu, rip);
1896 /* skipping an emulated instruction also counts */
1897 vmx_set_interrupt_shadow(vcpu, 0);
1901 * KVM wants to inject page-faults which it got to the guest. This function
1902 * checks whether in a nested guest, we need to inject them to L1 or L2.
1903 * This function assumes it is called with the exit reason in vmcs02 being
1904 * a #PF exception (this is the only case in which KVM injects a #PF when L2
1907 static int nested_pf_handled(struct kvm_vcpu *vcpu)
1909 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1911 /* TODO: also check PFEC_MATCH/MASK, not just EB.PF. */
1912 if (!(vmcs12->exception_bitmap & (1u << PF_VECTOR)))
1915 nested_vmx_vmexit(vcpu);
1919 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
1920 bool has_error_code, u32 error_code,
1923 struct vcpu_vmx *vmx = to_vmx(vcpu);
1924 u32 intr_info = nr | INTR_INFO_VALID_MASK;
1926 if (nr == PF_VECTOR && is_guest_mode(vcpu) &&
1927 !vmx->nested.nested_run_pending && nested_pf_handled(vcpu))
1930 if (has_error_code) {
1931 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1932 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1935 if (vmx->rmode.vm86_active) {
1937 if (kvm_exception_is_soft(nr))
1938 inc_eip = vcpu->arch.event_exit_inst_len;
1939 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
1940 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
1944 if (kvm_exception_is_soft(nr)) {
1945 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1946 vmx->vcpu.arch.event_exit_inst_len);
1947 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1949 intr_info |= INTR_TYPE_HARD_EXCEPTION;
1951 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1954 static bool vmx_rdtscp_supported(void)
1956 return cpu_has_vmx_rdtscp();
1959 static bool vmx_invpcid_supported(void)
1961 return cpu_has_vmx_invpcid() && enable_ept;
1965 * Swap MSR entry in host/guest MSR entry array.
1967 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1969 struct shared_msr_entry tmp;
1971 tmp = vmx->guest_msrs[to];
1972 vmx->guest_msrs[to] = vmx->guest_msrs[from];
1973 vmx->guest_msrs[from] = tmp;
1976 static void vmx_set_msr_bitmap(struct kvm_vcpu *vcpu)
1978 unsigned long *msr_bitmap;
1980 if (irqchip_in_kernel(vcpu->kvm) && apic_x2apic_mode(vcpu->arch.apic)) {
1981 if (is_long_mode(vcpu))
1982 msr_bitmap = vmx_msr_bitmap_longmode_x2apic;
1984 msr_bitmap = vmx_msr_bitmap_legacy_x2apic;
1986 if (is_long_mode(vcpu))
1987 msr_bitmap = vmx_msr_bitmap_longmode;
1989 msr_bitmap = vmx_msr_bitmap_legacy;
1992 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
1996 * Set up the vmcs to automatically save and restore system
1997 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
1998 * mode, as fiddling with msrs is very expensive.
2000 static void setup_msrs(struct vcpu_vmx *vmx)
2002 int save_nmsrs, index;
2005 #ifdef CONFIG_X86_64
2006 if (is_long_mode(&vmx->vcpu)) {
2007 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
2009 move_msr_up(vmx, index, save_nmsrs++);
2010 index = __find_msr_index(vmx, MSR_LSTAR);
2012 move_msr_up(vmx, index, save_nmsrs++);
2013 index = __find_msr_index(vmx, MSR_CSTAR);
2015 move_msr_up(vmx, index, save_nmsrs++);
2016 index = __find_msr_index(vmx, MSR_TSC_AUX);
2017 if (index >= 0 && vmx->rdtscp_enabled)
2018 move_msr_up(vmx, index, save_nmsrs++);
2020 * MSR_STAR is only needed on long mode guests, and only
2021 * if efer.sce is enabled.
2023 index = __find_msr_index(vmx, MSR_STAR);
2024 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
2025 move_msr_up(vmx, index, save_nmsrs++);
2028 index = __find_msr_index(vmx, MSR_EFER);
2029 if (index >= 0 && update_transition_efer(vmx, index))
2030 move_msr_up(vmx, index, save_nmsrs++);
2032 vmx->save_nmsrs = save_nmsrs;
2034 if (cpu_has_vmx_msr_bitmap())
2035 vmx_set_msr_bitmap(&vmx->vcpu);
2039 * reads and returns guest's timestamp counter "register"
2040 * guest_tsc = host_tsc + tsc_offset -- 21.3
2042 static u64 guest_read_tsc(void)
2044 u64 host_tsc, tsc_offset;
2047 tsc_offset = vmcs_read64(TSC_OFFSET);
2048 return host_tsc + tsc_offset;
2052 * Like guest_read_tsc, but always returns L1's notion of the timestamp
2053 * counter, even if a nested guest (L2) is currently running.
2055 u64 vmx_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2059 tsc_offset = is_guest_mode(vcpu) ?
2060 to_vmx(vcpu)->nested.vmcs01_tsc_offset :
2061 vmcs_read64(TSC_OFFSET);
2062 return host_tsc + tsc_offset;
2066 * Engage any workarounds for mis-matched TSC rates. Currently limited to
2067 * software catchup for faster rates on slower CPUs.
2069 static void vmx_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2074 if (user_tsc_khz > tsc_khz) {
2075 vcpu->arch.tsc_catchup = 1;
2076 vcpu->arch.tsc_always_catchup = 1;
2078 WARN(1, "user requested TSC rate below hardware speed\n");
2081 static u64 vmx_read_tsc_offset(struct kvm_vcpu *vcpu)
2083 return vmcs_read64(TSC_OFFSET);
2087 * writes 'offset' into guest's timestamp counter offset register
2089 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2091 if (is_guest_mode(vcpu)) {
2093 * We're here if L1 chose not to trap WRMSR to TSC. According
2094 * to the spec, this should set L1's TSC; The offset that L1
2095 * set for L2 remains unchanged, and still needs to be added
2096 * to the newly set TSC to get L2's TSC.
2098 struct vmcs12 *vmcs12;
2099 to_vmx(vcpu)->nested.vmcs01_tsc_offset = offset;
2100 /* recalculate vmcs02.TSC_OFFSET: */
2101 vmcs12 = get_vmcs12(vcpu);
2102 vmcs_write64(TSC_OFFSET, offset +
2103 (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
2104 vmcs12->tsc_offset : 0));
2106 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2107 vmcs_read64(TSC_OFFSET), offset);
2108 vmcs_write64(TSC_OFFSET, offset);
2112 static void vmx_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment, bool host)
2114 u64 offset = vmcs_read64(TSC_OFFSET);
2116 vmcs_write64(TSC_OFFSET, offset + adjustment);
2117 if (is_guest_mode(vcpu)) {
2118 /* Even when running L2, the adjustment needs to apply to L1 */
2119 to_vmx(vcpu)->nested.vmcs01_tsc_offset += adjustment;
2121 trace_kvm_write_tsc_offset(vcpu->vcpu_id, offset,
2122 offset + adjustment);
2125 static u64 vmx_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2127 return target_tsc - native_read_tsc();
2130 static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
2132 struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
2133 return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
2137 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
2138 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
2139 * all guests if the "nested" module option is off, and can also be disabled
2140 * for a single guest by disabling its VMX cpuid bit.
2142 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
2144 return nested && guest_cpuid_has_vmx(vcpu);
2148 * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
2149 * returned for the various VMX controls MSRs when nested VMX is enabled.
2150 * The same values should also be used to verify that vmcs12 control fields are
2151 * valid during nested entry from L1 to L2.
2152 * Each of these control msrs has a low and high 32-bit half: A low bit is on
2153 * if the corresponding bit in the (32-bit) control field *must* be on, and a
2154 * bit in the high half is on if the corresponding bit in the control field
2155 * may be on. See also vmx_control_verify().
2156 * TODO: allow these variables to be modified (downgraded) by module options
2159 static u32 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high;
2160 static u32 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high;
2161 static u32 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high;
2162 static u32 nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high;
2163 static u32 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high;
2164 static u32 nested_vmx_misc_low, nested_vmx_misc_high;
2165 static u32 nested_vmx_ept_caps;
2166 static __init void nested_vmx_setup_ctls_msrs(void)
2169 * Note that as a general rule, the high half of the MSRs (bits in
2170 * the control fields which may be 1) should be initialized by the
2171 * intersection of the underlying hardware's MSR (i.e., features which
2172 * can be supported) and the list of features we want to expose -
2173 * because they are known to be properly supported in our code.
2174 * Also, usually, the low half of the MSRs (bits which must be 1) can
2175 * be set to 0, meaning that L1 may turn off any of these bits. The
2176 * reason is that if one of these bits is necessary, it will appear
2177 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
2178 * fields of vmcs01 and vmcs02, will turn these bits off - and
2179 * nested_vmx_exit_handled() will not pass related exits to L1.
2180 * These rules have exceptions below.
2183 /* pin-based controls */
2184 rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
2185 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high);
2187 * According to the Intel spec, if bit 55 of VMX_BASIC is off (as it is
2188 * in our case), bits 1, 2 and 4 (i.e., 0x16) must be 1 in this MSR.
2190 nested_vmx_pinbased_ctls_low |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2191 nested_vmx_pinbased_ctls_high &= PIN_BASED_EXT_INTR_MASK |
2192 PIN_BASED_NMI_EXITING | PIN_BASED_VIRTUAL_NMIS |
2193 PIN_BASED_VMX_PREEMPTION_TIMER;
2194 nested_vmx_pinbased_ctls_high |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2198 * If bit 55 of VMX_BASIC is off, bits 0-8 and 10, 11, 13, 14, 16 and
2201 nested_vmx_exit_ctls_low = VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
2202 /* Note that guest use of VM_EXIT_ACK_INTR_ON_EXIT is not supported. */
2203 #ifdef CONFIG_X86_64
2204 nested_vmx_exit_ctls_high = VM_EXIT_HOST_ADDR_SPACE_SIZE;
2206 nested_vmx_exit_ctls_high = 0;
2208 nested_vmx_exit_ctls_high |= (VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
2209 VM_EXIT_LOAD_IA32_EFER);
2211 /* entry controls */
2212 rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
2213 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high);
2214 /* If bit 55 of VMX_BASIC is off, bits 0-8 and 12 must be 1. */
2215 nested_vmx_entry_ctls_low = VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
2216 nested_vmx_entry_ctls_high &=
2217 VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_IA32E_MODE;
2218 nested_vmx_entry_ctls_high |= (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR |
2219 VM_ENTRY_LOAD_IA32_EFER);
2220 /* cpu-based controls */
2221 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
2222 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high);
2223 nested_vmx_procbased_ctls_low = 0;
2224 nested_vmx_procbased_ctls_high &=
2225 CPU_BASED_VIRTUAL_INTR_PENDING | CPU_BASED_USE_TSC_OFFSETING |
2226 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
2227 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
2228 CPU_BASED_CR3_STORE_EXITING |
2229 #ifdef CONFIG_X86_64
2230 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
2232 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
2233 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_EXITING |
2234 CPU_BASED_RDPMC_EXITING | CPU_BASED_RDTSC_EXITING |
2235 CPU_BASED_PAUSE_EXITING |
2236 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2238 * We can allow some features even when not supported by the
2239 * hardware. For example, L1 can specify an MSR bitmap - and we
2240 * can use it to avoid exits to L1 - even when L0 runs L2
2241 * without MSR bitmaps.
2243 nested_vmx_procbased_ctls_high |= CPU_BASED_USE_MSR_BITMAPS;
2245 /* secondary cpu-based controls */
2246 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2247 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high);
2248 nested_vmx_secondary_ctls_low = 0;
2249 nested_vmx_secondary_ctls_high &=
2250 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2251 SECONDARY_EXEC_WBINVD_EXITING;
2253 /* miscellaneous data */
2254 rdmsr(MSR_IA32_VMX_MISC, nested_vmx_misc_low, nested_vmx_misc_high);
2255 nested_vmx_misc_low &= VMX_MISC_PREEMPTION_TIMER_RATE_MASK |
2256 VMX_MISC_SAVE_EFER_LMA;
2257 nested_vmx_misc_high = 0;
2260 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2263 * Bits 0 in high must be 0, and bits 1 in low must be 1.
2265 return ((control & high) | low) == control;
2268 static inline u64 vmx_control_msr(u32 low, u32 high)
2270 return low | ((u64)high << 32);
2274 * If we allow our guest to use VMX instructions (i.e., nested VMX), we should
2275 * also let it use VMX-specific MSRs.
2276 * vmx_get_vmx_msr() and vmx_set_vmx_msr() return 1 when we handled a
2277 * VMX-specific MSR, or 0 when we haven't (and the caller should handle it
2278 * like all other MSRs).
2280 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2282 if (!nested_vmx_allowed(vcpu) && msr_index >= MSR_IA32_VMX_BASIC &&
2283 msr_index <= MSR_IA32_VMX_TRUE_ENTRY_CTLS) {
2285 * According to the spec, processors which do not support VMX
2286 * should throw a #GP(0) when VMX capability MSRs are read.
2288 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
2292 switch (msr_index) {
2293 case MSR_IA32_FEATURE_CONTROL:
2294 if (nested_vmx_allowed(vcpu)) {
2295 *pdata = to_vmx(vcpu)->nested.msr_ia32_feature_control;
2299 case MSR_IA32_VMX_BASIC:
2301 * This MSR reports some information about VMX support. We
2302 * should return information about the VMX we emulate for the
2303 * guest, and the VMCS structure we give it - not about the
2304 * VMX support of the underlying hardware.
2306 *pdata = VMCS12_REVISION |
2307 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2308 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2310 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2311 case MSR_IA32_VMX_PINBASED_CTLS:
2312 *pdata = vmx_control_msr(nested_vmx_pinbased_ctls_low,
2313 nested_vmx_pinbased_ctls_high);
2315 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2316 case MSR_IA32_VMX_PROCBASED_CTLS:
2317 *pdata = vmx_control_msr(nested_vmx_procbased_ctls_low,
2318 nested_vmx_procbased_ctls_high);
2320 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2321 case MSR_IA32_VMX_EXIT_CTLS:
2322 *pdata = vmx_control_msr(nested_vmx_exit_ctls_low,
2323 nested_vmx_exit_ctls_high);
2325 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2326 case MSR_IA32_VMX_ENTRY_CTLS:
2327 *pdata = vmx_control_msr(nested_vmx_entry_ctls_low,
2328 nested_vmx_entry_ctls_high);
2330 case MSR_IA32_VMX_MISC:
2331 *pdata = vmx_control_msr(nested_vmx_misc_low,
2332 nested_vmx_misc_high);
2335 * These MSRs specify bits which the guest must keep fixed (on or off)
2336 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2337 * We picked the standard core2 setting.
2339 #define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2340 #define VMXON_CR4_ALWAYSON X86_CR4_VMXE
2341 case MSR_IA32_VMX_CR0_FIXED0:
2342 *pdata = VMXON_CR0_ALWAYSON;
2344 case MSR_IA32_VMX_CR0_FIXED1:
2347 case MSR_IA32_VMX_CR4_FIXED0:
2348 *pdata = VMXON_CR4_ALWAYSON;
2350 case MSR_IA32_VMX_CR4_FIXED1:
2353 case MSR_IA32_VMX_VMCS_ENUM:
2356 case MSR_IA32_VMX_PROCBASED_CTLS2:
2357 *pdata = vmx_control_msr(nested_vmx_secondary_ctls_low,
2358 nested_vmx_secondary_ctls_high);
2360 case MSR_IA32_VMX_EPT_VPID_CAP:
2361 /* Currently, no nested ept or nested vpid */
2371 static int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2373 u32 msr_index = msr_info->index;
2374 u64 data = msr_info->data;
2375 bool host_initialized = msr_info->host_initiated;
2377 if (!nested_vmx_allowed(vcpu))
2380 if (msr_index == MSR_IA32_FEATURE_CONTROL) {
2381 if (!host_initialized &&
2382 to_vmx(vcpu)->nested.msr_ia32_feature_control
2383 & FEATURE_CONTROL_LOCKED)
2385 to_vmx(vcpu)->nested.msr_ia32_feature_control = data;
2390 * No need to treat VMX capability MSRs specially: If we don't handle
2391 * them, handle_wrmsr will #GP(0), which is correct (they are readonly)
2397 * Reads an msr value (of 'msr_index') into 'pdata'.
2398 * Returns 0 on success, non-0 otherwise.
2399 * Assumes vcpu_load() was already called.
2401 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2404 struct shared_msr_entry *msr;
2407 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
2411 switch (msr_index) {
2412 #ifdef CONFIG_X86_64
2414 data = vmcs_readl(GUEST_FS_BASE);
2417 data = vmcs_readl(GUEST_GS_BASE);
2419 case MSR_KERNEL_GS_BASE:
2420 vmx_load_host_state(to_vmx(vcpu));
2421 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
2425 return kvm_get_msr_common(vcpu, msr_index, pdata);
2427 data = guest_read_tsc();
2429 case MSR_IA32_SYSENTER_CS:
2430 data = vmcs_read32(GUEST_SYSENTER_CS);
2432 case MSR_IA32_SYSENTER_EIP:
2433 data = vmcs_readl(GUEST_SYSENTER_EIP);
2435 case MSR_IA32_SYSENTER_ESP:
2436 data = vmcs_readl(GUEST_SYSENTER_ESP);
2439 if (!to_vmx(vcpu)->rdtscp_enabled)
2441 /* Otherwise falls through */
2443 if (vmx_get_vmx_msr(vcpu, msr_index, pdata))
2445 msr = find_msr_entry(to_vmx(vcpu), msr_index);
2450 return kvm_get_msr_common(vcpu, msr_index, pdata);
2458 * Writes msr value into into the appropriate "register".
2459 * Returns 0 on success, non-0 otherwise.
2460 * Assumes vcpu_load() was already called.
2462 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2464 struct vcpu_vmx *vmx = to_vmx(vcpu);
2465 struct shared_msr_entry *msr;
2467 u32 msr_index = msr_info->index;
2468 u64 data = msr_info->data;
2470 switch (msr_index) {
2472 ret = kvm_set_msr_common(vcpu, msr_info);
2474 #ifdef CONFIG_X86_64
2476 vmx_segment_cache_clear(vmx);
2477 vmcs_writel(GUEST_FS_BASE, data);
2480 vmx_segment_cache_clear(vmx);
2481 vmcs_writel(GUEST_GS_BASE, data);
2483 case MSR_KERNEL_GS_BASE:
2484 vmx_load_host_state(vmx);
2485 vmx->msr_guest_kernel_gs_base = data;
2488 case MSR_IA32_SYSENTER_CS:
2489 vmcs_write32(GUEST_SYSENTER_CS, data);
2491 case MSR_IA32_SYSENTER_EIP:
2492 vmcs_writel(GUEST_SYSENTER_EIP, data);
2494 case MSR_IA32_SYSENTER_ESP:
2495 vmcs_writel(GUEST_SYSENTER_ESP, data);
2498 kvm_write_tsc(vcpu, msr_info);
2500 case MSR_IA32_CR_PAT:
2501 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2502 vmcs_write64(GUEST_IA32_PAT, data);
2503 vcpu->arch.pat = data;
2506 ret = kvm_set_msr_common(vcpu, msr_info);
2508 case MSR_IA32_TSC_ADJUST:
2509 ret = kvm_set_msr_common(vcpu, msr_info);
2512 if (!vmx->rdtscp_enabled)
2514 /* Check reserved bit, higher 32 bits should be zero */
2515 if ((data >> 32) != 0)
2517 /* Otherwise falls through */
2519 if (vmx_set_vmx_msr(vcpu, msr_info))
2521 msr = find_msr_entry(vmx, msr_index);
2524 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
2526 kvm_set_shared_msr(msr->index, msr->data,
2532 ret = kvm_set_msr_common(vcpu, msr_info);
2538 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2540 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
2543 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2546 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2548 case VCPU_EXREG_PDPTR:
2550 ept_save_pdptrs(vcpu);
2557 static __init int cpu_has_kvm_support(void)
2559 return cpu_has_vmx();
2562 static __init int vmx_disabled_by_bios(void)
2566 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
2567 if (msr & FEATURE_CONTROL_LOCKED) {
2568 /* launched w/ TXT and VMX disabled */
2569 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2572 /* launched w/o TXT and VMX only enabled w/ TXT */
2573 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2574 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2575 && !tboot_enabled()) {
2576 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
2577 "activate TXT before enabling KVM\n");
2580 /* launched w/o TXT and VMX disabled */
2581 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2582 && !tboot_enabled())
2589 static void kvm_cpu_vmxon(u64 addr)
2591 asm volatile (ASM_VMX_VMXON_RAX
2592 : : "a"(&addr), "m"(addr)
2596 static int hardware_enable(void *garbage)
2598 int cpu = raw_smp_processor_id();
2599 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2602 if (read_cr4() & X86_CR4_VMXE)
2605 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2608 * Now we can enable the vmclear operation in kdump
2609 * since the loaded_vmcss_on_cpu list on this cpu
2610 * has been initialized.
2612 * Though the cpu is not in VMX operation now, there
2613 * is no problem to enable the vmclear operation
2614 * for the loaded_vmcss_on_cpu list is empty!
2616 crash_enable_local_vmclear(cpu);
2618 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
2620 test_bits = FEATURE_CONTROL_LOCKED;
2621 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
2622 if (tboot_enabled())
2623 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
2625 if ((old & test_bits) != test_bits) {
2626 /* enable and lock */
2627 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
2629 write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
2631 if (vmm_exclusive) {
2632 kvm_cpu_vmxon(phys_addr);
2636 native_store_gdt(&__get_cpu_var(host_gdt));
2641 static void vmclear_local_loaded_vmcss(void)
2643 int cpu = raw_smp_processor_id();
2644 struct loaded_vmcs *v, *n;
2646 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2647 loaded_vmcss_on_cpu_link)
2648 __loaded_vmcs_clear(v);
2652 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2655 static void kvm_cpu_vmxoff(void)
2657 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
2660 static void hardware_disable(void *garbage)
2662 if (vmm_exclusive) {
2663 vmclear_local_loaded_vmcss();
2666 write_cr4(read_cr4() & ~X86_CR4_VMXE);
2669 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2670 u32 msr, u32 *result)
2672 u32 vmx_msr_low, vmx_msr_high;
2673 u32 ctl = ctl_min | ctl_opt;
2675 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2677 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2678 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
2680 /* Ensure minimum (required) set of control bits are supported. */
2688 static __init bool allow_1_setting(u32 msr, u32 ctl)
2690 u32 vmx_msr_low, vmx_msr_high;
2692 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2693 return vmx_msr_high & ctl;
2696 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
2698 u32 vmx_msr_low, vmx_msr_high;
2699 u32 min, opt, min2, opt2;
2700 u32 _pin_based_exec_control = 0;
2701 u32 _cpu_based_exec_control = 0;
2702 u32 _cpu_based_2nd_exec_control = 0;
2703 u32 _vmexit_control = 0;
2704 u32 _vmentry_control = 0;
2706 min = CPU_BASED_HLT_EXITING |
2707 #ifdef CONFIG_X86_64
2708 CPU_BASED_CR8_LOAD_EXITING |
2709 CPU_BASED_CR8_STORE_EXITING |
2711 CPU_BASED_CR3_LOAD_EXITING |
2712 CPU_BASED_CR3_STORE_EXITING |
2713 CPU_BASED_USE_IO_BITMAPS |
2714 CPU_BASED_MOV_DR_EXITING |
2715 CPU_BASED_USE_TSC_OFFSETING |
2716 CPU_BASED_MWAIT_EXITING |
2717 CPU_BASED_MONITOR_EXITING |
2718 CPU_BASED_INVLPG_EXITING |
2719 CPU_BASED_RDPMC_EXITING;
2721 opt = CPU_BASED_TPR_SHADOW |
2722 CPU_BASED_USE_MSR_BITMAPS |
2723 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2724 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2725 &_cpu_based_exec_control) < 0)
2727 #ifdef CONFIG_X86_64
2728 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2729 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2730 ~CPU_BASED_CR8_STORE_EXITING;
2732 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2734 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2735 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2736 SECONDARY_EXEC_WBINVD_EXITING |
2737 SECONDARY_EXEC_ENABLE_VPID |
2738 SECONDARY_EXEC_ENABLE_EPT |
2739 SECONDARY_EXEC_UNRESTRICTED_GUEST |
2740 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2741 SECONDARY_EXEC_RDTSCP |
2742 SECONDARY_EXEC_ENABLE_INVPCID |
2743 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2744 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2745 SECONDARY_EXEC_SHADOW_VMCS;
2746 if (adjust_vmx_controls(min2, opt2,
2747 MSR_IA32_VMX_PROCBASED_CTLS2,
2748 &_cpu_based_2nd_exec_control) < 0)
2751 #ifndef CONFIG_X86_64
2752 if (!(_cpu_based_2nd_exec_control &
2753 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2754 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2757 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2758 _cpu_based_2nd_exec_control &= ~(
2759 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2760 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2761 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2763 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2764 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2766 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2767 CPU_BASED_CR3_STORE_EXITING |
2768 CPU_BASED_INVLPG_EXITING);
2769 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
2770 vmx_capability.ept, vmx_capability.vpid);
2774 #ifdef CONFIG_X86_64
2775 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2777 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT |
2778 VM_EXIT_ACK_INTR_ON_EXIT;
2779 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2780 &_vmexit_control) < 0)
2783 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2784 opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR;
2785 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2786 &_pin_based_exec_control) < 0)
2789 if (!(_cpu_based_2nd_exec_control &
2790 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) ||
2791 !(_vmexit_control & VM_EXIT_ACK_INTR_ON_EXIT))
2792 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2795 opt = VM_ENTRY_LOAD_IA32_PAT;
2796 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2797 &_vmentry_control) < 0)
2800 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2802 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2803 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2806 #ifdef CONFIG_X86_64
2807 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2808 if (vmx_msr_high & (1u<<16))
2812 /* Require Write-Back (WB) memory type for VMCS accesses. */
2813 if (((vmx_msr_high >> 18) & 15) != 6)
2816 vmcs_conf->size = vmx_msr_high & 0x1fff;
2817 vmcs_conf->order = get_order(vmcs_config.size);
2818 vmcs_conf->revision_id = vmx_msr_low;
2820 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2821 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2822 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2823 vmcs_conf->vmexit_ctrl = _vmexit_control;
2824 vmcs_conf->vmentry_ctrl = _vmentry_control;
2826 cpu_has_load_ia32_efer =
2827 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2828 VM_ENTRY_LOAD_IA32_EFER)
2829 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2830 VM_EXIT_LOAD_IA32_EFER);
2832 cpu_has_load_perf_global_ctrl =
2833 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2834 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
2835 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2836 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
2839 * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
2840 * but due to arrata below it can't be used. Workaround is to use
2841 * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2843 * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
2848 * BC86,AAY89,BD102 (model 44)
2852 if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
2853 switch (boot_cpu_data.x86_model) {
2859 cpu_has_load_perf_global_ctrl = false;
2860 printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2861 "does not work properly. Using workaround\n");
2871 static struct vmcs *alloc_vmcs_cpu(int cpu)
2873 int node = cpu_to_node(cpu);
2877 pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
2880 vmcs = page_address(pages);
2881 memset(vmcs, 0, vmcs_config.size);
2882 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
2886 static struct vmcs *alloc_vmcs(void)
2888 return alloc_vmcs_cpu(raw_smp_processor_id());
2891 static void free_vmcs(struct vmcs *vmcs)
2893 free_pages((unsigned long)vmcs, vmcs_config.order);
2897 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2899 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2901 if (!loaded_vmcs->vmcs)
2903 loaded_vmcs_clear(loaded_vmcs);
2904 free_vmcs(loaded_vmcs->vmcs);
2905 loaded_vmcs->vmcs = NULL;
2908 static void free_kvm_area(void)
2912 for_each_possible_cpu(cpu) {
2913 free_vmcs(per_cpu(vmxarea, cpu));
2914 per_cpu(vmxarea, cpu) = NULL;
2918 static __init int alloc_kvm_area(void)
2922 for_each_possible_cpu(cpu) {
2925 vmcs = alloc_vmcs_cpu(cpu);
2931 per_cpu(vmxarea, cpu) = vmcs;
2936 static __init int hardware_setup(void)
2938 if (setup_vmcs_config(&vmcs_config) < 0)
2941 if (boot_cpu_has(X86_FEATURE_NX))
2942 kvm_enable_efer_bits(EFER_NX);
2944 if (!cpu_has_vmx_vpid())
2946 if (!cpu_has_vmx_shadow_vmcs())
2947 enable_shadow_vmcs = 0;
2949 if (!cpu_has_vmx_ept() ||
2950 !cpu_has_vmx_ept_4levels()) {
2952 enable_unrestricted_guest = 0;
2953 enable_ept_ad_bits = 0;
2956 if (!cpu_has_vmx_ept_ad_bits())
2957 enable_ept_ad_bits = 0;
2959 if (!cpu_has_vmx_unrestricted_guest())
2960 enable_unrestricted_guest = 0;
2962 if (!cpu_has_vmx_flexpriority())
2963 flexpriority_enabled = 0;
2965 if (!cpu_has_vmx_tpr_shadow())
2966 kvm_x86_ops->update_cr8_intercept = NULL;
2968 if (enable_ept && !cpu_has_vmx_ept_2m_page())
2969 kvm_disable_largepages();
2971 if (!cpu_has_vmx_ple())
2974 if (!cpu_has_vmx_apicv())
2978 kvm_x86_ops->update_cr8_intercept = NULL;
2980 kvm_x86_ops->hwapic_irr_update = NULL;
2981 kvm_x86_ops->deliver_posted_interrupt = NULL;
2982 kvm_x86_ops->sync_pir_to_irr = vmx_sync_pir_to_irr_dummy;
2986 nested_vmx_setup_ctls_msrs();
2988 return alloc_kvm_area();
2991 static __exit void hardware_unsetup(void)
2996 static bool emulation_required(struct kvm_vcpu *vcpu)
2998 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
3001 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
3002 struct kvm_segment *save)
3004 if (!emulate_invalid_guest_state) {
3006 * CS and SS RPL should be equal during guest entry according
3007 * to VMX spec, but in reality it is not always so. Since vcpu
3008 * is in the middle of the transition from real mode to
3009 * protected mode it is safe to assume that RPL 0 is a good
3012 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
3013 save->selector &= ~SELECTOR_RPL_MASK;
3014 save->dpl = save->selector & SELECTOR_RPL_MASK;
3017 vmx_set_segment(vcpu, save, seg);
3020 static void enter_pmode(struct kvm_vcpu *vcpu)
3022 unsigned long flags;
3023 struct vcpu_vmx *vmx = to_vmx(vcpu);
3026 * Update real mode segment cache. It may be not up-to-date if sement
3027 * register was written while vcpu was in a guest mode.
3029 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3030 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3031 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3032 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3033 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3034 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3036 vmx->rmode.vm86_active = 0;
3038 vmx_segment_cache_clear(vmx);
3040 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3042 flags = vmcs_readl(GUEST_RFLAGS);
3043 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
3044 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
3045 vmcs_writel(GUEST_RFLAGS, flags);
3047 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
3048 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
3050 update_exception_bitmap(vcpu);
3052 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3053 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3054 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3055 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3056 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3057 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3059 /* CPL is always 0 when CPU enters protected mode */
3060 __set_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
3064 static void fix_rmode_seg(int seg, struct kvm_segment *save)
3066 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3067 struct kvm_segment var = *save;
3070 if (seg == VCPU_SREG_CS)
3073 if (!emulate_invalid_guest_state) {
3074 var.selector = var.base >> 4;
3075 var.base = var.base & 0xffff0;
3085 if (save->base & 0xf)
3086 printk_once(KERN_WARNING "kvm: segment base is not "
3087 "paragraph aligned when entering "
3088 "protected mode (seg=%d)", seg);
3091 vmcs_write16(sf->selector, var.selector);
3092 vmcs_write32(sf->base, var.base);
3093 vmcs_write32(sf->limit, var.limit);
3094 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
3097 static void enter_rmode(struct kvm_vcpu *vcpu)
3099 unsigned long flags;
3100 struct vcpu_vmx *vmx = to_vmx(vcpu);
3102 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3103 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3104 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3105 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3106 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3107 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3108 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3110 vmx->rmode.vm86_active = 1;
3113 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
3114 * vcpu. Warn the user that an update is overdue.
3116 if (!vcpu->kvm->arch.tss_addr)
3117 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
3118 "called before entering vcpu\n");
3120 vmx_segment_cache_clear(vmx);
3122 vmcs_writel(GUEST_TR_BASE, vcpu->kvm->arch.tss_addr);
3123 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
3124 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3126 flags = vmcs_readl(GUEST_RFLAGS);
3127 vmx->rmode.save_rflags = flags;
3129 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
3131 vmcs_writel(GUEST_RFLAGS, flags);
3132 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
3133 update_exception_bitmap(vcpu);
3135 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3136 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3137 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3138 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3139 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3140 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3142 kvm_mmu_reset_context(vcpu);
3145 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
3147 struct vcpu_vmx *vmx = to_vmx(vcpu);
3148 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
3154 * Force kernel_gs_base reloading before EFER changes, as control
3155 * of this msr depends on is_long_mode().
3157 vmx_load_host_state(to_vmx(vcpu));
3158 vcpu->arch.efer = efer;
3159 if (efer & EFER_LMA) {
3160 vmcs_write32(VM_ENTRY_CONTROLS,
3161 vmcs_read32(VM_ENTRY_CONTROLS) |
3162 VM_ENTRY_IA32E_MODE);
3165 vmcs_write32(VM_ENTRY_CONTROLS,
3166 vmcs_read32(VM_ENTRY_CONTROLS) &
3167 ~VM_ENTRY_IA32E_MODE);
3169 msr->data = efer & ~EFER_LME;
3174 #ifdef CONFIG_X86_64
3176 static void enter_lmode(struct kvm_vcpu *vcpu)
3180 vmx_segment_cache_clear(to_vmx(vcpu));
3182 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
3183 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
3184 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
3186 vmcs_write32(GUEST_TR_AR_BYTES,
3187 (guest_tr_ar & ~AR_TYPE_MASK)
3188 | AR_TYPE_BUSY_64_TSS);
3190 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
3193 static void exit_lmode(struct kvm_vcpu *vcpu)
3195 vmcs_write32(VM_ENTRY_CONTROLS,
3196 vmcs_read32(VM_ENTRY_CONTROLS)
3197 & ~VM_ENTRY_IA32E_MODE);
3198 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
3203 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
3205 vpid_sync_context(to_vmx(vcpu));
3207 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
3209 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
3213 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
3215 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
3217 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
3218 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
3221 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
3223 if (enable_ept && is_paging(vcpu))
3224 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3225 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
3228 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
3230 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
3232 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
3233 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
3236 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
3238 if (!test_bit(VCPU_EXREG_PDPTR,
3239 (unsigned long *)&vcpu->arch.regs_dirty))
3242 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3243 vmcs_write64(GUEST_PDPTR0, vcpu->arch.mmu.pdptrs[0]);
3244 vmcs_write64(GUEST_PDPTR1, vcpu->arch.mmu.pdptrs[1]);
3245 vmcs_write64(GUEST_PDPTR2, vcpu->arch.mmu.pdptrs[2]);
3246 vmcs_write64(GUEST_PDPTR3, vcpu->arch.mmu.pdptrs[3]);
3250 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
3252 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3253 vcpu->arch.mmu.pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
3254 vcpu->arch.mmu.pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
3255 vcpu->arch.mmu.pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
3256 vcpu->arch.mmu.pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
3259 __set_bit(VCPU_EXREG_PDPTR,
3260 (unsigned long *)&vcpu->arch.regs_avail);
3261 __set_bit(VCPU_EXREG_PDPTR,
3262 (unsigned long *)&vcpu->arch.regs_dirty);
3265 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
3267 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
3269 struct kvm_vcpu *vcpu)
3271 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3272 vmx_decache_cr3(vcpu);
3273 if (!(cr0 & X86_CR0_PG)) {
3274 /* From paging/starting to nonpaging */
3275 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3276 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
3277 (CPU_BASED_CR3_LOAD_EXITING |
3278 CPU_BASED_CR3_STORE_EXITING));
3279 vcpu->arch.cr0 = cr0;
3280 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3281 } else if (!is_paging(vcpu)) {
3282 /* From nonpaging to paging */
3283 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3284 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
3285 ~(CPU_BASED_CR3_LOAD_EXITING |
3286 CPU_BASED_CR3_STORE_EXITING));
3287 vcpu->arch.cr0 = cr0;
3288 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3291 if (!(cr0 & X86_CR0_WP))
3292 *hw_cr0 &= ~X86_CR0_WP;
3295 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
3297 struct vcpu_vmx *vmx = to_vmx(vcpu);
3298 unsigned long hw_cr0;
3300 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK);
3301 if (enable_unrestricted_guest)
3302 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
3304 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
3306 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
3309 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
3313 #ifdef CONFIG_X86_64
3314 if (vcpu->arch.efer & EFER_LME) {
3315 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
3317 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
3323 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
3325 if (!vcpu->fpu_active)
3326 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
3328 vmcs_writel(CR0_READ_SHADOW, cr0);
3329 vmcs_writel(GUEST_CR0, hw_cr0);
3330 vcpu->arch.cr0 = cr0;
3332 /* depends on vcpu->arch.cr0 to be set to a new value */
3333 vmx->emulation_required = emulation_required(vcpu);
3336 static u64 construct_eptp(unsigned long root_hpa)
3340 /* TODO write the value reading from MSR */
3341 eptp = VMX_EPT_DEFAULT_MT |
3342 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
3343 if (enable_ept_ad_bits)
3344 eptp |= VMX_EPT_AD_ENABLE_BIT;
3345 eptp |= (root_hpa & PAGE_MASK);
3350 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
3352 unsigned long guest_cr3;
3357 eptp = construct_eptp(cr3);
3358 vmcs_write64(EPT_POINTER, eptp);
3359 guest_cr3 = is_paging(vcpu) ? kvm_read_cr3(vcpu) :
3360 vcpu->kvm->arch.ept_identity_map_addr;
3361 ept_load_pdptrs(vcpu);
3364 vmx_flush_tlb(vcpu);
3365 vmcs_writel(GUEST_CR3, guest_cr3);
3368 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3370 unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
3371 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
3373 if (cr4 & X86_CR4_VMXE) {
3375 * To use VMXON (and later other VMX instructions), a guest
3376 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3377 * So basically the check on whether to allow nested VMX
3380 if (!nested_vmx_allowed(vcpu))
3383 if (to_vmx(vcpu)->nested.vmxon &&
3384 ((cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON))
3387 vcpu->arch.cr4 = cr4;
3389 if (!is_paging(vcpu)) {
3390 hw_cr4 &= ~X86_CR4_PAE;
3391 hw_cr4 |= X86_CR4_PSE;
3393 * SMEP is disabled if CPU is in non-paging mode in
3394 * hardware. However KVM always uses paging mode to
3395 * emulate guest non-paging mode with TDP.
3396 * To emulate this behavior, SMEP needs to be manually
3397 * disabled when guest switches to non-paging mode.
3399 hw_cr4 &= ~X86_CR4_SMEP;
3400 } else if (!(cr4 & X86_CR4_PAE)) {
3401 hw_cr4 &= ~X86_CR4_PAE;
3405 vmcs_writel(CR4_READ_SHADOW, cr4);
3406 vmcs_writel(GUEST_CR4, hw_cr4);
3410 static void vmx_get_segment(struct kvm_vcpu *vcpu,
3411 struct kvm_segment *var, int seg)
3413 struct vcpu_vmx *vmx = to_vmx(vcpu);
3416 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3417 *var = vmx->rmode.segs[seg];
3418 if (seg == VCPU_SREG_TR
3419 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3421 var->base = vmx_read_guest_seg_base(vmx, seg);
3422 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3425 var->base = vmx_read_guest_seg_base(vmx, seg);
3426 var->limit = vmx_read_guest_seg_limit(vmx, seg);
3427 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3428 ar = vmx_read_guest_seg_ar(vmx, seg);
3429 var->unusable = (ar >> 16) & 1;
3430 var->type = ar & 15;
3431 var->s = (ar >> 4) & 1;
3432 var->dpl = (ar >> 5) & 3;
3434 * Some userspaces do not preserve unusable property. Since usable
3435 * segment has to be present according to VMX spec we can use present
3436 * property to amend userspace bug by making unusable segment always
3437 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3438 * segment as unusable.
3440 var->present = !var->unusable;
3441 var->avl = (ar >> 12) & 1;
3442 var->l = (ar >> 13) & 1;
3443 var->db = (ar >> 14) & 1;
3444 var->g = (ar >> 15) & 1;
3447 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3449 struct kvm_segment s;
3451 if (to_vmx(vcpu)->rmode.vm86_active) {
3452 vmx_get_segment(vcpu, &s, seg);
3455 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3458 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
3460 struct vcpu_vmx *vmx = to_vmx(vcpu);
3462 if (!is_protmode(vcpu))
3465 if (!is_long_mode(vcpu)
3466 && (kvm_get_rflags(vcpu) & X86_EFLAGS_VM)) /* if virtual 8086 */
3469 if (!test_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail)) {
3470 __set_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
3471 vmx->cpl = vmx_read_guest_seg_selector(vmx, VCPU_SREG_CS) & 3;
3478 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3482 if (var->unusable || !var->present)
3485 ar = var->type & 15;
3486 ar |= (var->s & 1) << 4;
3487 ar |= (var->dpl & 3) << 5;
3488 ar |= (var->present & 1) << 7;
3489 ar |= (var->avl & 1) << 12;
3490 ar |= (var->l & 1) << 13;
3491 ar |= (var->db & 1) << 14;
3492 ar |= (var->g & 1) << 15;
3498 static void vmx_set_segment(struct kvm_vcpu *vcpu,
3499 struct kvm_segment *var, int seg)
3501 struct vcpu_vmx *vmx = to_vmx(vcpu);
3502 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3504 vmx_segment_cache_clear(vmx);
3505 if (seg == VCPU_SREG_CS)
3506 __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
3508 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3509 vmx->rmode.segs[seg] = *var;
3510 if (seg == VCPU_SREG_TR)
3511 vmcs_write16(sf->selector, var->selector);
3513 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3517 vmcs_writel(sf->base, var->base);
3518 vmcs_write32(sf->limit, var->limit);
3519 vmcs_write16(sf->selector, var->selector);
3522 * Fix the "Accessed" bit in AR field of segment registers for older
3524 * IA32 arch specifies that at the time of processor reset the
3525 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3526 * is setting it to 0 in the userland code. This causes invalid guest
3527 * state vmexit when "unrestricted guest" mode is turned on.
3528 * Fix for this setup issue in cpu_reset is being pushed in the qemu
3529 * tree. Newer qemu binaries with that qemu fix would not need this
3532 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3533 var->type |= 0x1; /* Accessed */
3535 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3538 vmx->emulation_required |= emulation_required(vcpu);
3541 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3543 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3545 *db = (ar >> 14) & 1;
3546 *l = (ar >> 13) & 1;
3549 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3551 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3552 dt->address = vmcs_readl(GUEST_IDTR_BASE);
3555 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3557 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3558 vmcs_writel(GUEST_IDTR_BASE, dt->address);
3561 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3563 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3564 dt->address = vmcs_readl(GUEST_GDTR_BASE);
3567 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3569 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3570 vmcs_writel(GUEST_GDTR_BASE, dt->address);
3573 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3575 struct kvm_segment var;
3578 vmx_get_segment(vcpu, &var, seg);
3580 if (seg == VCPU_SREG_CS)
3582 ar = vmx_segment_access_rights(&var);
3584 if (var.base != (var.selector << 4))
3586 if (var.limit != 0xffff)
3594 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3596 struct kvm_segment cs;
3597 unsigned int cs_rpl;
3599 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3600 cs_rpl = cs.selector & SELECTOR_RPL_MASK;
3604 if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
3608 if (cs.type & AR_TYPE_WRITEABLE_MASK) {
3609 if (cs.dpl > cs_rpl)
3612 if (cs.dpl != cs_rpl)
3618 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3622 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3624 struct kvm_segment ss;
3625 unsigned int ss_rpl;
3627 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3628 ss_rpl = ss.selector & SELECTOR_RPL_MASK;
3632 if (ss.type != 3 && ss.type != 7)
3636 if (ss.dpl != ss_rpl) /* DPL != RPL */
3644 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3646 struct kvm_segment var;
3649 vmx_get_segment(vcpu, &var, seg);
3650 rpl = var.selector & SELECTOR_RPL_MASK;
3658 if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
3659 if (var.dpl < rpl) /* DPL < RPL */
3663 /* TODO: Add other members to kvm_segment_field to allow checking for other access
3669 static bool tr_valid(struct kvm_vcpu *vcpu)
3671 struct kvm_segment tr;
3673 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3677 if (tr.selector & SELECTOR_TI_MASK) /* TI = 1 */
3679 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3687 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3689 struct kvm_segment ldtr;
3691 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3695 if (ldtr.selector & SELECTOR_TI_MASK) /* TI = 1 */
3705 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3707 struct kvm_segment cs, ss;
3709 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3710 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3712 return ((cs.selector & SELECTOR_RPL_MASK) ==
3713 (ss.selector & SELECTOR_RPL_MASK));
3717 * Check if guest state is valid. Returns true if valid, false if
3719 * We assume that registers are always usable
3721 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3723 if (enable_unrestricted_guest)
3726 /* real mode guest state checks */
3727 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3728 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3730 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3732 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3734 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3736 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3738 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3741 /* protected mode guest state checks */
3742 if (!cs_ss_rpl_check(vcpu))
3744 if (!code_segment_valid(vcpu))
3746 if (!stack_segment_valid(vcpu))
3748 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3750 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3752 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3754 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3756 if (!tr_valid(vcpu))
3758 if (!ldtr_valid(vcpu))
3762 * - Add checks on RIP
3763 * - Add checks on RFLAGS
3769 static int init_rmode_tss(struct kvm *kvm)
3773 int r, idx, ret = 0;
3775 idx = srcu_read_lock(&kvm->srcu);
3776 fn = kvm->arch.tss_addr >> PAGE_SHIFT;
3777 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3780 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3781 r = kvm_write_guest_page(kvm, fn++, &data,
3782 TSS_IOPB_BASE_OFFSET, sizeof(u16));
3785 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3788 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3792 r = kvm_write_guest_page(kvm, fn, &data,
3793 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3800 srcu_read_unlock(&kvm->srcu, idx);
3804 static int init_rmode_identity_map(struct kvm *kvm)
3807 pfn_t identity_map_pfn;
3812 if (unlikely(!kvm->arch.ept_identity_pagetable)) {
3813 printk(KERN_ERR "EPT: identity-mapping pagetable "
3814 "haven't been allocated!\n");
3817 if (likely(kvm->arch.ept_identity_pagetable_done))
3820 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
3821 idx = srcu_read_lock(&kvm->srcu);
3822 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3825 /* Set up identity-mapping pagetable for EPT in real mode */
3826 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3827 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3828 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3829 r = kvm_write_guest_page(kvm, identity_map_pfn,
3830 &tmp, i * sizeof(tmp), sizeof(tmp));
3834 kvm->arch.ept_identity_pagetable_done = true;
3837 srcu_read_unlock(&kvm->srcu, idx);
3841 static void seg_setup(int seg)
3843 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3846 vmcs_write16(sf->selector, 0);
3847 vmcs_writel(sf->base, 0);
3848 vmcs_write32(sf->limit, 0xffff);
3850 if (seg == VCPU_SREG_CS)
3851 ar |= 0x08; /* code segment */
3853 vmcs_write32(sf->ar_bytes, ar);
3856 static int alloc_apic_access_page(struct kvm *kvm)
3859 struct kvm_userspace_memory_region kvm_userspace_mem;
3862 mutex_lock(&kvm->slots_lock);
3863 if (kvm->arch.apic_access_page)
3865 kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
3866 kvm_userspace_mem.flags = 0;
3867 kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
3868 kvm_userspace_mem.memory_size = PAGE_SIZE;
3869 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem);
3873 page = gfn_to_page(kvm, 0xfee00);
3874 if (is_error_page(page)) {
3879 kvm->arch.apic_access_page = page;
3881 mutex_unlock(&kvm->slots_lock);
3885 static int alloc_identity_pagetable(struct kvm *kvm)
3888 struct kvm_userspace_memory_region kvm_userspace_mem;
3891 mutex_lock(&kvm->slots_lock);
3892 if (kvm->arch.ept_identity_pagetable)
3894 kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
3895 kvm_userspace_mem.flags = 0;
3896 kvm_userspace_mem.guest_phys_addr =
3897 kvm->arch.ept_identity_map_addr;
3898 kvm_userspace_mem.memory_size = PAGE_SIZE;
3899 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem);
3903 page = gfn_to_page(kvm, kvm->arch.ept_identity_map_addr >> PAGE_SHIFT);
3904 if (is_error_page(page)) {
3909 kvm->arch.ept_identity_pagetable = page;
3911 mutex_unlock(&kvm->slots_lock);
3915 static void allocate_vpid(struct vcpu_vmx *vmx)
3922 spin_lock(&vmx_vpid_lock);
3923 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3924 if (vpid < VMX_NR_VPIDS) {
3926 __set_bit(vpid, vmx_vpid_bitmap);
3928 spin_unlock(&vmx_vpid_lock);
3931 static void free_vpid(struct vcpu_vmx *vmx)
3935 spin_lock(&vmx_vpid_lock);
3937 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
3938 spin_unlock(&vmx_vpid_lock);
3941 #define MSR_TYPE_R 1
3942 #define MSR_TYPE_W 2
3943 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
3946 int f = sizeof(unsigned long);
3948 if (!cpu_has_vmx_msr_bitmap())
3952 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3953 * have the write-low and read-high bitmap offsets the wrong way round.
3954 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3956 if (msr <= 0x1fff) {
3957 if (type & MSR_TYPE_R)
3959 __clear_bit(msr, msr_bitmap + 0x000 / f);
3961 if (type & MSR_TYPE_W)
3963 __clear_bit(msr, msr_bitmap + 0x800 / f);
3965 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3967 if (type & MSR_TYPE_R)
3969 __clear_bit(msr, msr_bitmap + 0x400 / f);
3971 if (type & MSR_TYPE_W)
3973 __clear_bit(msr, msr_bitmap + 0xc00 / f);
3978 static void __vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
3981 int f = sizeof(unsigned long);
3983 if (!cpu_has_vmx_msr_bitmap())
3987 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3988 * have the write-low and read-high bitmap offsets the wrong way round.
3989 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3991 if (msr <= 0x1fff) {
3992 if (type & MSR_TYPE_R)
3994 __set_bit(msr, msr_bitmap + 0x000 / f);
3996 if (type & MSR_TYPE_W)
3998 __set_bit(msr, msr_bitmap + 0x800 / f);
4000 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4002 if (type & MSR_TYPE_R)
4004 __set_bit(msr, msr_bitmap + 0x400 / f);
4006 if (type & MSR_TYPE_W)
4008 __set_bit(msr, msr_bitmap + 0xc00 / f);
4013 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
4016 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy,
4017 msr, MSR_TYPE_R | MSR_TYPE_W);
4018 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode,
4019 msr, MSR_TYPE_R | MSR_TYPE_W);
4022 static void vmx_enable_intercept_msr_read_x2apic(u32 msr)
4024 __vmx_enable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4026 __vmx_enable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4030 static void vmx_disable_intercept_msr_read_x2apic(u32 msr)
4032 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4034 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4038 static void vmx_disable_intercept_msr_write_x2apic(u32 msr)
4040 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4042 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4046 static int vmx_vm_has_apicv(struct kvm *kvm)
4048 return enable_apicv && irqchip_in_kernel(kvm);
4052 * Send interrupt to vcpu via posted interrupt way.
4053 * 1. If target vcpu is running(non-root mode), send posted interrupt
4054 * notification to vcpu and hardware will sync PIR to vIRR atomically.
4055 * 2. If target vcpu isn't running(root mode), kick it to pick up the
4056 * interrupt from PIR in next vmentry.
4058 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
4060 struct vcpu_vmx *vmx = to_vmx(vcpu);
4063 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
4066 r = pi_test_and_set_on(&vmx->pi_desc);
4067 kvm_make_request(KVM_REQ_EVENT, vcpu);
4069 if (!r && (vcpu->mode == IN_GUEST_MODE))
4070 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu),
4071 POSTED_INTR_VECTOR);
4074 kvm_vcpu_kick(vcpu);
4077 static void vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
4079 struct vcpu_vmx *vmx = to_vmx(vcpu);
4081 if (!pi_test_and_clear_on(&vmx->pi_desc))
4084 kvm_apic_update_irr(vcpu, vmx->pi_desc.pir);
4087 static void vmx_sync_pir_to_irr_dummy(struct kvm_vcpu *vcpu)
4093 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
4094 * will not change in the lifetime of the guest.
4095 * Note that host-state that does change is set elsewhere. E.g., host-state
4096 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
4098 static void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
4104 vmcs_writel(HOST_CR0, read_cr0() & ~X86_CR0_TS); /* 22.2.3 */
4105 vmcs_writel(HOST_CR4, read_cr4()); /* 22.2.3, 22.2.5 */
4106 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
4108 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
4109 #ifdef CONFIG_X86_64
4111 * Load null selectors, so we can avoid reloading them in
4112 * __vmx_load_host_state(), in case userspace uses the null selectors
4113 * too (the expected case).
4115 vmcs_write16(HOST_DS_SELECTOR, 0);
4116 vmcs_write16(HOST_ES_SELECTOR, 0);
4118 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4119 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4121 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4122 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
4124 native_store_idt(&dt);
4125 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
4126 vmx->host_idt_base = dt.address;
4128 vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
4130 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
4131 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
4132 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
4133 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
4135 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
4136 rdmsr(MSR_IA32_CR_PAT, low32, high32);
4137 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
4141 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
4143 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
4145 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
4146 if (is_guest_mode(&vmx->vcpu))
4147 vmx->vcpu.arch.cr4_guest_owned_bits &=
4148 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
4149 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
4152 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
4154 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
4156 if (!vmx_vm_has_apicv(vmx->vcpu.kvm))
4157 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
4158 return pin_based_exec_ctrl;
4161 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
4163 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
4164 if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
4165 exec_control &= ~CPU_BASED_TPR_SHADOW;
4166 #ifdef CONFIG_X86_64
4167 exec_control |= CPU_BASED_CR8_STORE_EXITING |
4168 CPU_BASED_CR8_LOAD_EXITING;
4172 exec_control |= CPU_BASED_CR3_STORE_EXITING |
4173 CPU_BASED_CR3_LOAD_EXITING |
4174 CPU_BASED_INVLPG_EXITING;
4175 return exec_control;
4178 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
4180 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4181 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
4182 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4184 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4186 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4187 enable_unrestricted_guest = 0;
4188 /* Enable INVPCID for non-ept guests may cause performance regression. */
4189 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4191 if (!enable_unrestricted_guest)
4192 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4194 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4195 if (!vmx_vm_has_apicv(vmx->vcpu.kvm))
4196 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4197 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4198 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4199 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4201 We can NOT enable shadow_vmcs here because we don't have yet
4204 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4205 return exec_control;
4208 static void ept_set_mmio_spte_mask(void)
4211 * EPT Misconfigurations can be generated if the value of bits 2:0
4212 * of an EPT paging-structure entry is 110b (write/execute).
4213 * Also, magic bits (0x3ull << 62) is set to quickly identify mmio
4216 kvm_mmu_set_mmio_spte_mask((0x3ull << 62) | 0x6ull);
4220 * Sets up the vmcs for emulated real mode.
4222 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
4224 #ifdef CONFIG_X86_64
4230 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
4231 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
4233 if (enable_shadow_vmcs) {
4234 vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
4235 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
4237 if (cpu_has_vmx_msr_bitmap())
4238 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
4240 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4243 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
4245 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
4247 if (cpu_has_secondary_exec_ctrls()) {
4248 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
4249 vmx_secondary_exec_control(vmx));
4252 if (vmx_vm_has_apicv(vmx->vcpu.kvm)) {
4253 vmcs_write64(EOI_EXIT_BITMAP0, 0);
4254 vmcs_write64(EOI_EXIT_BITMAP1, 0);
4255 vmcs_write64(EOI_EXIT_BITMAP2, 0);
4256 vmcs_write64(EOI_EXIT_BITMAP3, 0);
4258 vmcs_write16(GUEST_INTR_STATUS, 0);
4260 vmcs_write64(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4261 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4265 vmcs_write32(PLE_GAP, ple_gap);
4266 vmcs_write32(PLE_WINDOW, ple_window);
4269 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4270 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4271 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
4273 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
4274 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
4275 vmx_set_constant_host_state(vmx);
4276 #ifdef CONFIG_X86_64
4277 rdmsrl(MSR_FS_BASE, a);
4278 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
4279 rdmsrl(MSR_GS_BASE, a);
4280 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
4282 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4283 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4286 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4287 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4288 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
4289 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4290 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
4292 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
4293 u32 msr_low, msr_high;
4295 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
4296 host_pat = msr_low | ((u64) msr_high << 32);
4297 /* Write the default value follow host pat */
4298 vmcs_write64(GUEST_IA32_PAT, host_pat);
4299 /* Keep arch.pat sync with GUEST_IA32_PAT */
4300 vmx->vcpu.arch.pat = host_pat;
4303 for (i = 0; i < NR_VMX_MSR; ++i) {
4304 u32 index = vmx_msr_index[i];
4305 u32 data_low, data_high;
4308 if (rdmsr_safe(index, &data_low, &data_high) < 0)
4310 if (wrmsr_safe(index, data_low, data_high) < 0)
4312 vmx->guest_msrs[j].index = i;
4313 vmx->guest_msrs[j].data = 0;
4314 vmx->guest_msrs[j].mask = -1ull;
4318 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
4320 /* 22.2.1, 20.8.1 */
4321 vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
4323 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
4324 set_cr4_guest_host_mask(vmx);
4329 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu)
4331 struct vcpu_vmx *vmx = to_vmx(vcpu);
4334 vmx->rmode.vm86_active = 0;
4336 vmx->soft_vnmi_blocked = 0;
4338 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4339 kvm_set_cr8(&vmx->vcpu, 0);
4340 msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
4341 if (kvm_vcpu_is_bsp(&vmx->vcpu))
4342 msr |= MSR_IA32_APICBASE_BSP;
4343 kvm_set_apic_base(&vmx->vcpu, msr);
4345 vmx_segment_cache_clear(vmx);
4347 seg_setup(VCPU_SREG_CS);
4348 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4349 vmcs_write32(GUEST_CS_BASE, 0xffff0000);
4351 seg_setup(VCPU_SREG_DS);
4352 seg_setup(VCPU_SREG_ES);
4353 seg_setup(VCPU_SREG_FS);
4354 seg_setup(VCPU_SREG_GS);
4355 seg_setup(VCPU_SREG_SS);
4357 vmcs_write16(GUEST_TR_SELECTOR, 0);
4358 vmcs_writel(GUEST_TR_BASE, 0);
4359 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4360 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4362 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4363 vmcs_writel(GUEST_LDTR_BASE, 0);
4364 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4365 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4367 vmcs_write32(GUEST_SYSENTER_CS, 0);
4368 vmcs_writel(GUEST_SYSENTER_ESP, 0);
4369 vmcs_writel(GUEST_SYSENTER_EIP, 0);
4371 vmcs_writel(GUEST_RFLAGS, 0x02);
4372 kvm_rip_write(vcpu, 0xfff0);
4374 vmcs_writel(GUEST_GDTR_BASE, 0);
4375 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4377 vmcs_writel(GUEST_IDTR_BASE, 0);
4378 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4380 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4381 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4382 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4384 /* Special registers */
4385 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4389 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
4391 if (cpu_has_vmx_tpr_shadow()) {
4392 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4393 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
4394 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4395 __pa(vmx->vcpu.arch.apic->regs));
4396 vmcs_write32(TPR_THRESHOLD, 0);
4399 if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
4400 vmcs_write64(APIC_ACCESS_ADDR,
4401 page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
4403 if (vmx_vm_has_apicv(vcpu->kvm))
4404 memset(&vmx->pi_desc, 0, sizeof(struct pi_desc));
4407 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4409 vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4410 vmx_set_cr0(&vmx->vcpu, kvm_read_cr0(vcpu)); /* enter rmode */
4411 vmx_set_cr4(&vmx->vcpu, 0);
4412 vmx_set_efer(&vmx->vcpu, 0);
4413 vmx_fpu_activate(&vmx->vcpu);
4414 update_exception_bitmap(&vmx->vcpu);
4416 vpid_sync_context(vmx);
4420 * In nested virtualization, check if L1 asked to exit on external interrupts.
4421 * For most existing hypervisors, this will always return true.
4423 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
4425 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
4426 PIN_BASED_EXT_INTR_MASK;
4429 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
4431 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
4432 PIN_BASED_NMI_EXITING;
4435 static int enable_irq_window(struct kvm_vcpu *vcpu)
4437 u32 cpu_based_vm_exec_control;
4439 if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu))
4441 * We get here if vmx_interrupt_allowed() said we can't
4442 * inject to L1 now because L2 must run. The caller will have
4443 * to make L2 exit right after entry, so we can inject to L1
4448 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4449 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
4450 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4454 static int enable_nmi_window(struct kvm_vcpu *vcpu)
4456 u32 cpu_based_vm_exec_control;
4458 if (!cpu_has_virtual_nmis())
4459 return enable_irq_window(vcpu);
4461 if (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI)
4462 return enable_irq_window(vcpu);
4464 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4465 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
4466 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4470 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4472 struct vcpu_vmx *vmx = to_vmx(vcpu);
4474 int irq = vcpu->arch.interrupt.nr;
4476 trace_kvm_inj_virq(irq);
4478 ++vcpu->stat.irq_injections;
4479 if (vmx->rmode.vm86_active) {
4481 if (vcpu->arch.interrupt.soft)
4482 inc_eip = vcpu->arch.event_exit_inst_len;
4483 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
4484 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4487 intr = irq | INTR_INFO_VALID_MASK;
4488 if (vcpu->arch.interrupt.soft) {
4489 intr |= INTR_TYPE_SOFT_INTR;
4490 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4491 vmx->vcpu.arch.event_exit_inst_len);
4493 intr |= INTR_TYPE_EXT_INTR;
4494 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4497 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4499 struct vcpu_vmx *vmx = to_vmx(vcpu);
4501 if (is_guest_mode(vcpu))
4504 if (!cpu_has_virtual_nmis()) {
4506 * Tracking the NMI-blocked state in software is built upon
4507 * finding the next open IRQ window. This, in turn, depends on
4508 * well-behaving guests: They have to keep IRQs disabled at
4509 * least as long as the NMI handler runs. Otherwise we may
4510 * cause NMI nesting, maybe breaking the guest. But as this is
4511 * highly unlikely, we can live with the residual risk.
4513 vmx->soft_vnmi_blocked = 1;
4514 vmx->vnmi_blocked_time = 0;
4517 ++vcpu->stat.nmi_injections;
4518 vmx->nmi_known_unmasked = false;
4519 if (vmx->rmode.vm86_active) {
4520 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
4521 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4524 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4525 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4528 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4530 if (!cpu_has_virtual_nmis())
4531 return to_vmx(vcpu)->soft_vnmi_blocked;
4532 if (to_vmx(vcpu)->nmi_known_unmasked)
4534 return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4537 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4539 struct vcpu_vmx *vmx = to_vmx(vcpu);
4541 if (!cpu_has_virtual_nmis()) {
4542 if (vmx->soft_vnmi_blocked != masked) {
4543 vmx->soft_vnmi_blocked = masked;
4544 vmx->vnmi_blocked_time = 0;
4547 vmx->nmi_known_unmasked = !masked;
4549 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4550 GUEST_INTR_STATE_NMI);
4552 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4553 GUEST_INTR_STATE_NMI);
4557 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4559 if (is_guest_mode(vcpu)) {
4560 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4562 if (to_vmx(vcpu)->nested.nested_run_pending)
4564 if (nested_exit_on_nmi(vcpu)) {
4565 nested_vmx_vmexit(vcpu);
4566 vmcs12->vm_exit_reason = EXIT_REASON_EXCEPTION_NMI;
4567 vmcs12->vm_exit_intr_info = NMI_VECTOR |
4568 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK;
4570 * The NMI-triggered VM exit counts as injection:
4571 * clear this one and block further NMIs.
4573 vcpu->arch.nmi_pending = 0;
4574 vmx_set_nmi_mask(vcpu, true);
4579 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
4582 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4583 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4584 | GUEST_INTR_STATE_NMI));
4587 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4589 if (is_guest_mode(vcpu)) {
4590 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4592 if (to_vmx(vcpu)->nested.nested_run_pending)
4594 if (nested_exit_on_intr(vcpu)) {
4595 nested_vmx_vmexit(vcpu);
4596 vmcs12->vm_exit_reason =
4597 EXIT_REASON_EXTERNAL_INTERRUPT;
4598 vmcs12->vm_exit_intr_info = 0;
4600 * fall through to normal code, but now in L1, not L2
4605 return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4606 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4607 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4610 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4613 struct kvm_userspace_memory_region tss_mem = {
4614 .slot = TSS_PRIVATE_MEMSLOT,
4615 .guest_phys_addr = addr,
4616 .memory_size = PAGE_SIZE * 3,
4620 ret = kvm_set_memory_region(kvm, &tss_mem);
4623 kvm->arch.tss_addr = addr;
4624 if (!init_rmode_tss(kvm))
4630 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4635 * Update instruction length as we may reinject the exception
4636 * from user space while in guest debugging mode.
4638 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4639 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4640 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4644 if (vcpu->guest_debug &
4645 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4662 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4663 int vec, u32 err_code)
4666 * Instruction with address size override prefix opcode 0x67
4667 * Cause the #SS fault with 0 error code in VM86 mode.
4669 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4670 if (emulate_instruction(vcpu, 0) == EMULATE_DONE) {
4671 if (vcpu->arch.halt_request) {
4672 vcpu->arch.halt_request = 0;
4673 return kvm_emulate_halt(vcpu);
4681 * Forward all other exceptions that are valid in real mode.
4682 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4683 * the required debugging infrastructure rework.
4685 kvm_queue_exception(vcpu, vec);
4690 * Trigger machine check on the host. We assume all the MSRs are already set up
4691 * by the CPU and that we still run on the same CPU as the MCE occurred on.
4692 * We pass a fake environment to the machine check handler because we want
4693 * the guest to be always treated like user space, no matter what context
4694 * it used internally.
4696 static void kvm_machine_check(void)
4698 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4699 struct pt_regs regs = {
4700 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
4701 .flags = X86_EFLAGS_IF,
4704 do_machine_check(®s, 0);
4708 static int handle_machine_check(struct kvm_vcpu *vcpu)
4710 /* already handled by vcpu_run */
4714 static int handle_exception(struct kvm_vcpu *vcpu)
4716 struct vcpu_vmx *vmx = to_vmx(vcpu);
4717 struct kvm_run *kvm_run = vcpu->run;
4718 u32 intr_info, ex_no, error_code;
4719 unsigned long cr2, rip, dr6;
4721 enum emulation_result er;
4723 vect_info = vmx->idt_vectoring_info;
4724 intr_info = vmx->exit_intr_info;
4726 if (is_machine_check(intr_info))
4727 return handle_machine_check(vcpu);
4729 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
4730 return 1; /* already handled by vmx_vcpu_run() */
4732 if (is_no_device(intr_info)) {
4733 vmx_fpu_activate(vcpu);
4737 if (is_invalid_opcode(intr_info)) {
4738 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
4739 if (er != EMULATE_DONE)
4740 kvm_queue_exception(vcpu, UD_VECTOR);
4745 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4746 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4749 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4750 * MMIO, it is better to report an internal error.
4751 * See the comments in vmx_handle_exit.
4753 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4754 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4755 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4756 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4757 vcpu->run->internal.ndata = 2;
4758 vcpu->run->internal.data[0] = vect_info;
4759 vcpu->run->internal.data[1] = intr_info;
4763 if (is_page_fault(intr_info)) {
4764 /* EPT won't cause page fault directly */
4766 cr2 = vmcs_readl(EXIT_QUALIFICATION);
4767 trace_kvm_page_fault(cr2, error_code);
4769 if (kvm_event_needs_reinjection(vcpu))
4770 kvm_mmu_unprotect_page_virt(vcpu, cr2);
4771 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
4774 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4776 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4777 return handle_rmode_exception(vcpu, ex_no, error_code);
4781 dr6 = vmcs_readl(EXIT_QUALIFICATION);
4782 if (!(vcpu->guest_debug &
4783 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4784 vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
4785 kvm_queue_exception(vcpu, DB_VECTOR);
4788 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4789 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4793 * Update instruction length as we may reinject #BP from
4794 * user space while in guest debugging mode. Reading it for
4795 * #DB as well causes no harm, it is not used in that case.
4797 vmx->vcpu.arch.event_exit_inst_len =
4798 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4799 kvm_run->exit_reason = KVM_EXIT_DEBUG;
4800 rip = kvm_rip_read(vcpu);
4801 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4802 kvm_run->debug.arch.exception = ex_no;
4805 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4806 kvm_run->ex.exception = ex_no;
4807 kvm_run->ex.error_code = error_code;
4813 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
4815 ++vcpu->stat.irq_exits;
4819 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4821 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4825 static int handle_io(struct kvm_vcpu *vcpu)
4827 unsigned long exit_qualification;
4828 int size, in, string;
4831 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4832 string = (exit_qualification & 16) != 0;
4833 in = (exit_qualification & 8) != 0;
4835 ++vcpu->stat.io_exits;
4838 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4840 port = exit_qualification >> 16;
4841 size = (exit_qualification & 7) + 1;
4842 skip_emulated_instruction(vcpu);
4844 return kvm_fast_pio_out(vcpu, size, port);
4848 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4851 * Patch in the VMCALL instruction:
4853 hypercall[0] = 0x0f;
4854 hypercall[1] = 0x01;
4855 hypercall[2] = 0xc1;
4858 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4859 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4861 if (is_guest_mode(vcpu)) {
4862 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4863 unsigned long orig_val = val;
4866 * We get here when L2 changed cr0 in a way that did not change
4867 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4868 * but did change L0 shadowed bits. So we first calculate the
4869 * effective cr0 value that L1 would like to write into the
4870 * hardware. It consists of the L2-owned bits from the new
4871 * value combined with the L1-owned bits from L1's guest_cr0.
4873 val = (val & ~vmcs12->cr0_guest_host_mask) |
4874 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
4876 /* TODO: will have to take unrestricted guest mode into
4878 if ((val & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON)
4881 if (kvm_set_cr0(vcpu, val))
4883 vmcs_writel(CR0_READ_SHADOW, orig_val);
4886 if (to_vmx(vcpu)->nested.vmxon &&
4887 ((val & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON))
4889 return kvm_set_cr0(vcpu, val);
4893 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4895 if (is_guest_mode(vcpu)) {
4896 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4897 unsigned long orig_val = val;
4899 /* analogously to handle_set_cr0 */
4900 val = (val & ~vmcs12->cr4_guest_host_mask) |
4901 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
4902 if (kvm_set_cr4(vcpu, val))
4904 vmcs_writel(CR4_READ_SHADOW, orig_val);
4907 return kvm_set_cr4(vcpu, val);
4910 /* called to set cr0 as approriate for clts instruction exit. */
4911 static void handle_clts(struct kvm_vcpu *vcpu)
4913 if (is_guest_mode(vcpu)) {
4915 * We get here when L2 did CLTS, and L1 didn't shadow CR0.TS
4916 * but we did (!fpu_active). We need to keep GUEST_CR0.TS on,
4917 * just pretend it's off (also in arch.cr0 for fpu_activate).
4919 vmcs_writel(CR0_READ_SHADOW,
4920 vmcs_readl(CR0_READ_SHADOW) & ~X86_CR0_TS);
4921 vcpu->arch.cr0 &= ~X86_CR0_TS;
4923 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4926 static int handle_cr(struct kvm_vcpu *vcpu)
4928 unsigned long exit_qualification, val;
4933 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4934 cr = exit_qualification & 15;
4935 reg = (exit_qualification >> 8) & 15;
4936 switch ((exit_qualification >> 4) & 3) {
4937 case 0: /* mov to cr */
4938 val = kvm_register_read(vcpu, reg);
4939 trace_kvm_cr_write(cr, val);
4942 err = handle_set_cr0(vcpu, val);
4943 kvm_complete_insn_gp(vcpu, err);
4946 err = kvm_set_cr3(vcpu, val);
4947 kvm_complete_insn_gp(vcpu, err);
4950 err = handle_set_cr4(vcpu, val);
4951 kvm_complete_insn_gp(vcpu, err);
4954 u8 cr8_prev = kvm_get_cr8(vcpu);
4955 u8 cr8 = kvm_register_read(vcpu, reg);
4956 err = kvm_set_cr8(vcpu, cr8);
4957 kvm_complete_insn_gp(vcpu, err);
4958 if (irqchip_in_kernel(vcpu->kvm))
4960 if (cr8_prev <= cr8)
4962 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4969 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4970 skip_emulated_instruction(vcpu);
4971 vmx_fpu_activate(vcpu);
4973 case 1: /*mov from cr*/
4976 val = kvm_read_cr3(vcpu);
4977 kvm_register_write(vcpu, reg, val);
4978 trace_kvm_cr_read(cr, val);
4979 skip_emulated_instruction(vcpu);
4982 val = kvm_get_cr8(vcpu);
4983 kvm_register_write(vcpu, reg, val);
4984 trace_kvm_cr_read(cr, val);
4985 skip_emulated_instruction(vcpu);
4990 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4991 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
4992 kvm_lmsw(vcpu, val);
4994 skip_emulated_instruction(vcpu);
4999 vcpu->run->exit_reason = 0;
5000 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
5001 (int)(exit_qualification >> 4) & 3, cr);
5005 static int handle_dr(struct kvm_vcpu *vcpu)
5007 unsigned long exit_qualification;
5010 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
5011 if (!kvm_require_cpl(vcpu, 0))
5013 dr = vmcs_readl(GUEST_DR7);
5016 * As the vm-exit takes precedence over the debug trap, we
5017 * need to emulate the latter, either for the host or the
5018 * guest debugging itself.
5020 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5021 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
5022 vcpu->run->debug.arch.dr7 = dr;
5023 vcpu->run->debug.arch.pc =
5024 vmcs_readl(GUEST_CS_BASE) +
5025 vmcs_readl(GUEST_RIP);
5026 vcpu->run->debug.arch.exception = DB_VECTOR;
5027 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
5030 vcpu->arch.dr7 &= ~DR7_GD;
5031 vcpu->arch.dr6 |= DR6_BD;
5032 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
5033 kvm_queue_exception(vcpu, DB_VECTOR);
5038 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5039 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
5040 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
5041 if (exit_qualification & TYPE_MOV_FROM_DR) {
5043 if (!kvm_get_dr(vcpu, dr, &val))
5044 kvm_register_write(vcpu, reg, val);
5046 kvm_set_dr(vcpu, dr, vcpu->arch.regs[reg]);
5047 skip_emulated_instruction(vcpu);
5051 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
5053 vmcs_writel(GUEST_DR7, val);
5056 static int handle_cpuid(struct kvm_vcpu *vcpu)
5058 kvm_emulate_cpuid(vcpu);
5062 static int handle_rdmsr(struct kvm_vcpu *vcpu)
5064 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5067 if (vmx_get_msr(vcpu, ecx, &data)) {
5068 trace_kvm_msr_read_ex(ecx);
5069 kvm_inject_gp(vcpu, 0);
5073 trace_kvm_msr_read(ecx, data);
5075 /* FIXME: handling of bits 32:63 of rax, rdx */
5076 vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
5077 vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
5078 skip_emulated_instruction(vcpu);
5082 static int handle_wrmsr(struct kvm_vcpu *vcpu)
5084 struct msr_data msr;
5085 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5086 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
5087 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
5091 msr.host_initiated = false;
5092 if (vmx_set_msr(vcpu, &msr) != 0) {
5093 trace_kvm_msr_write_ex(ecx, data);
5094 kvm_inject_gp(vcpu, 0);
5098 trace_kvm_msr_write(ecx, data);
5099 skip_emulated_instruction(vcpu);
5103 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
5105 kvm_make_request(KVM_REQ_EVENT, vcpu);
5109 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
5111 u32 cpu_based_vm_exec_control;
5113 /* clear pending irq */
5114 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5115 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
5116 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5118 kvm_make_request(KVM_REQ_EVENT, vcpu);
5120 ++vcpu->stat.irq_window_exits;
5123 * If the user space waits to inject interrupts, exit as soon as
5126 if (!irqchip_in_kernel(vcpu->kvm) &&
5127 vcpu->run->request_interrupt_window &&
5128 !kvm_cpu_has_interrupt(vcpu)) {
5129 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
5135 static int handle_halt(struct kvm_vcpu *vcpu)
5137 skip_emulated_instruction(vcpu);
5138 return kvm_emulate_halt(vcpu);
5141 static int handle_vmcall(struct kvm_vcpu *vcpu)
5143 skip_emulated_instruction(vcpu);
5144 kvm_emulate_hypercall(vcpu);
5148 static int handle_invd(struct kvm_vcpu *vcpu)
5150 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5153 static int handle_invlpg(struct kvm_vcpu *vcpu)
5155 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5157 kvm_mmu_invlpg(vcpu, exit_qualification);
5158 skip_emulated_instruction(vcpu);
5162 static int handle_rdpmc(struct kvm_vcpu *vcpu)
5166 err = kvm_rdpmc(vcpu);
5167 kvm_complete_insn_gp(vcpu, err);
5172 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5174 skip_emulated_instruction(vcpu);
5175 kvm_emulate_wbinvd(vcpu);
5179 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5181 u64 new_bv = kvm_read_edx_eax(vcpu);
5182 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5184 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5185 skip_emulated_instruction(vcpu);
5189 static int handle_apic_access(struct kvm_vcpu *vcpu)
5191 if (likely(fasteoi)) {
5192 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5193 int access_type, offset;
5195 access_type = exit_qualification & APIC_ACCESS_TYPE;
5196 offset = exit_qualification & APIC_ACCESS_OFFSET;
5198 * Sane guest uses MOV to write EOI, with written value
5199 * not cared. So make a short-circuit here by avoiding
5200 * heavy instruction emulation.
5202 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5203 (offset == APIC_EOI)) {
5204 kvm_lapic_set_eoi(vcpu);
5205 skip_emulated_instruction(vcpu);
5209 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5212 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5214 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5215 int vector = exit_qualification & 0xff;
5217 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5218 kvm_apic_set_eoi_accelerated(vcpu, vector);
5222 static int handle_apic_write(struct kvm_vcpu *vcpu)
5224 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5225 u32 offset = exit_qualification & 0xfff;
5227 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
5228 kvm_apic_write_nodecode(vcpu, offset);
5232 static int handle_task_switch(struct kvm_vcpu *vcpu)
5234 struct vcpu_vmx *vmx = to_vmx(vcpu);
5235 unsigned long exit_qualification;
5236 bool has_error_code = false;
5239 int reason, type, idt_v, idt_index;
5241 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5242 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5243 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5245 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5247 reason = (u32)exit_qualification >> 30;
5248 if (reason == TASK_SWITCH_GATE && idt_v) {
5250 case INTR_TYPE_NMI_INTR:
5251 vcpu->arch.nmi_injected = false;
5252 vmx_set_nmi_mask(vcpu, true);
5254 case INTR_TYPE_EXT_INTR:
5255 case INTR_TYPE_SOFT_INTR:
5256 kvm_clear_interrupt_queue(vcpu);
5258 case INTR_TYPE_HARD_EXCEPTION:
5259 if (vmx->idt_vectoring_info &
5260 VECTORING_INFO_DELIVER_CODE_MASK) {
5261 has_error_code = true;
5263 vmcs_read32(IDT_VECTORING_ERROR_CODE);
5266 case INTR_TYPE_SOFT_EXCEPTION:
5267 kvm_clear_exception_queue(vcpu);
5273 tss_selector = exit_qualification;
5275 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5276 type != INTR_TYPE_EXT_INTR &&
5277 type != INTR_TYPE_NMI_INTR))
5278 skip_emulated_instruction(vcpu);
5280 if (kvm_task_switch(vcpu, tss_selector,
5281 type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
5282 has_error_code, error_code) == EMULATE_FAIL) {
5283 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5284 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5285 vcpu->run->internal.ndata = 0;
5289 /* clear all local breakpoint enable flags */
5290 vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
5293 * TODO: What about debug traps on tss switch?
5294 * Are we supposed to inject them and update dr6?
5300 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5302 unsigned long exit_qualification;
5307 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5309 gla_validity = (exit_qualification >> 7) & 0x3;
5310 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
5311 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
5312 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
5313 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
5314 vmcs_readl(GUEST_LINEAR_ADDRESS));
5315 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
5316 (long unsigned int)exit_qualification);
5317 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5318 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
5322 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5323 trace_kvm_page_fault(gpa, exit_qualification);
5325 /* It is a write fault? */
5326 error_code = exit_qualification & (1U << 1);
5327 /* It is a fetch fault? */
5328 error_code |= (exit_qualification & (1U << 2)) << 2;
5329 /* ept page table is present? */
5330 error_code |= (exit_qualification >> 3) & 0x1;
5332 vcpu->arch.exit_qualification = exit_qualification;
5334 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5337 static u64 ept_rsvd_mask(u64 spte, int level)
5342 for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
5343 mask |= (1ULL << i);
5346 /* bits 7:3 reserved */
5348 else if (level == 2) {
5349 if (spte & (1ULL << 7))
5350 /* 2MB ref, bits 20:12 reserved */
5353 /* bits 6:3 reserved */
5360 static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
5363 printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
5365 /* 010b (write-only) */
5366 WARN_ON((spte & 0x7) == 0x2);
5368 /* 110b (write/execute) */
5369 WARN_ON((spte & 0x7) == 0x6);
5371 /* 100b (execute-only) and value not supported by logical processor */
5372 if (!cpu_has_vmx_ept_execute_only())
5373 WARN_ON((spte & 0x7) == 0x4);
5377 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
5379 if (rsvd_bits != 0) {
5380 printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
5381 __func__, rsvd_bits);
5385 if (level == 1 || (level == 2 && (spte & (1ULL << 7)))) {
5386 u64 ept_mem_type = (spte & 0x38) >> 3;
5388 if (ept_mem_type == 2 || ept_mem_type == 3 ||
5389 ept_mem_type == 7) {
5390 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
5391 __func__, ept_mem_type);
5398 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5401 int nr_sptes, i, ret;
5404 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5406 ret = handle_mmio_page_fault_common(vcpu, gpa, true);
5407 if (likely(ret == RET_MMIO_PF_EMULATE))
5408 return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) ==
5411 if (unlikely(ret == RET_MMIO_PF_INVALID))
5412 return kvm_mmu_page_fault(vcpu, gpa, 0, NULL, 0);
5414 if (unlikely(ret == RET_MMIO_PF_RETRY))
5417 /* It is the real ept misconfig */
5418 printk(KERN_ERR "EPT: Misconfiguration.\n");
5419 printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
5421 nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
5423 for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
5424 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
5426 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5427 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
5432 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5434 u32 cpu_based_vm_exec_control;
5436 /* clear pending NMI */
5437 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5438 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
5439 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5440 ++vcpu->stat.nmi_window_exits;
5441 kvm_make_request(KVM_REQ_EVENT, vcpu);
5446 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5448 struct vcpu_vmx *vmx = to_vmx(vcpu);
5449 enum emulation_result err = EMULATE_DONE;
5452 bool intr_window_requested;
5453 unsigned count = 130;
5455 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5456 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
5458 while (!guest_state_valid(vcpu) && count-- != 0) {
5459 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5460 return handle_interrupt_window(&vmx->vcpu);
5462 if (test_bit(KVM_REQ_EVENT, &vcpu->requests))
5465 err = emulate_instruction(vcpu, EMULTYPE_NO_REEXECUTE);
5467 if (err == EMULATE_USER_EXIT) {
5472 if (err != EMULATE_DONE) {
5473 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5474 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5475 vcpu->run->internal.ndata = 0;
5479 if (vcpu->arch.halt_request) {
5480 vcpu->arch.halt_request = 0;
5481 ret = kvm_emulate_halt(vcpu);
5485 if (signal_pending(current))
5491 vmx->emulation_required = emulation_required(vcpu);
5497 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5498 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5500 static int handle_pause(struct kvm_vcpu *vcpu)
5502 skip_emulated_instruction(vcpu);
5503 kvm_vcpu_on_spin(vcpu);
5508 static int handle_invalid_op(struct kvm_vcpu *vcpu)
5510 kvm_queue_exception(vcpu, UD_VECTOR);
5515 * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
5516 * We could reuse a single VMCS for all the L2 guests, but we also want the
5517 * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
5518 * allows keeping them loaded on the processor, and in the future will allow
5519 * optimizations where prepare_vmcs02 doesn't need to set all the fields on
5520 * every entry if they never change.
5521 * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
5522 * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
5524 * The following functions allocate and free a vmcs02 in this pool.
5527 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
5528 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
5530 struct vmcs02_list *item;
5531 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
5532 if (item->vmptr == vmx->nested.current_vmptr) {
5533 list_move(&item->list, &vmx->nested.vmcs02_pool);
5534 return &item->vmcs02;
5537 if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
5538 /* Recycle the least recently used VMCS. */
5539 item = list_entry(vmx->nested.vmcs02_pool.prev,
5540 struct vmcs02_list, list);
5541 item->vmptr = vmx->nested.current_vmptr;
5542 list_move(&item->list, &vmx->nested.vmcs02_pool);
5543 return &item->vmcs02;
5546 /* Create a new VMCS */
5547 item = kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
5550 item->vmcs02.vmcs = alloc_vmcs();
5551 if (!item->vmcs02.vmcs) {
5555 loaded_vmcs_init(&item->vmcs02);
5556 item->vmptr = vmx->nested.current_vmptr;
5557 list_add(&(item->list), &(vmx->nested.vmcs02_pool));
5558 vmx->nested.vmcs02_num++;
5559 return &item->vmcs02;
5562 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
5563 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
5565 struct vmcs02_list *item;
5566 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
5567 if (item->vmptr == vmptr) {
5568 free_loaded_vmcs(&item->vmcs02);
5569 list_del(&item->list);
5571 vmx->nested.vmcs02_num--;
5577 * Free all VMCSs saved for this vcpu, except the one pointed by
5578 * vmx->loaded_vmcs. These include the VMCSs in vmcs02_pool (except the one
5579 * currently used, if running L2), and vmcs01 when running L2.
5581 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
5583 struct vmcs02_list *item, *n;
5584 list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
5585 if (vmx->loaded_vmcs != &item->vmcs02)
5586 free_loaded_vmcs(&item->vmcs02);
5587 list_del(&item->list);
5590 vmx->nested.vmcs02_num = 0;
5592 if (vmx->loaded_vmcs != &vmx->vmcs01)
5593 free_loaded_vmcs(&vmx->vmcs01);
5597 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
5598 * set the success or error code of an emulated VMX instruction, as specified
5599 * by Vol 2B, VMX Instruction Reference, "Conventions".
5601 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
5603 vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
5604 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
5605 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
5608 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
5610 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
5611 & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
5612 X86_EFLAGS_SF | X86_EFLAGS_OF))
5616 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
5617 u32 vm_instruction_error)
5619 if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
5621 * failValid writes the error number to the current VMCS, which
5622 * can't be done there isn't a current VMCS.
5624 nested_vmx_failInvalid(vcpu);
5627 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
5628 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
5629 X86_EFLAGS_SF | X86_EFLAGS_OF))
5631 get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
5633 * We don't need to force a shadow sync because
5634 * VM_INSTRUCTION_ERROR is not shadowed
5639 * Emulate the VMXON instruction.
5640 * Currently, we just remember that VMX is active, and do not save or even
5641 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
5642 * do not currently need to store anything in that guest-allocated memory
5643 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
5644 * argument is different from the VMXON pointer (which the spec says they do).
5646 static int handle_vmon(struct kvm_vcpu *vcpu)
5648 struct kvm_segment cs;
5649 struct vcpu_vmx *vmx = to_vmx(vcpu);
5650 struct vmcs *shadow_vmcs;
5651 const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
5652 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
5654 /* The Intel VMX Instruction Reference lists a bunch of bits that
5655 * are prerequisite to running VMXON, most notably cr4.VMXE must be
5656 * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
5657 * Otherwise, we should fail with #UD. We test these now:
5659 if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
5660 !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
5661 (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
5662 kvm_queue_exception(vcpu, UD_VECTOR);
5666 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
5667 if (is_long_mode(vcpu) && !cs.l) {
5668 kvm_queue_exception(vcpu, UD_VECTOR);
5672 if (vmx_get_cpl(vcpu)) {
5673 kvm_inject_gp(vcpu, 0);
5676 if (vmx->nested.vmxon) {
5677 nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
5678 skip_emulated_instruction(vcpu);
5682 if ((vmx->nested.msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
5683 != VMXON_NEEDED_FEATURES) {
5684 kvm_inject_gp(vcpu, 0);
5688 if (enable_shadow_vmcs) {
5689 shadow_vmcs = alloc_vmcs();
5692 /* mark vmcs as shadow */
5693 shadow_vmcs->revision_id |= (1u << 31);
5694 /* init shadow vmcs */
5695 vmcs_clear(shadow_vmcs);
5696 vmx->nested.current_shadow_vmcs = shadow_vmcs;
5699 INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
5700 vmx->nested.vmcs02_num = 0;
5702 vmx->nested.vmxon = true;
5704 skip_emulated_instruction(vcpu);
5705 nested_vmx_succeed(vcpu);
5710 * Intel's VMX Instruction Reference specifies a common set of prerequisites
5711 * for running VMX instructions (except VMXON, whose prerequisites are
5712 * slightly different). It also specifies what exception to inject otherwise.
5714 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
5716 struct kvm_segment cs;
5717 struct vcpu_vmx *vmx = to_vmx(vcpu);
5719 if (!vmx->nested.vmxon) {
5720 kvm_queue_exception(vcpu, UD_VECTOR);
5724 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
5725 if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
5726 (is_long_mode(vcpu) && !cs.l)) {
5727 kvm_queue_exception(vcpu, UD_VECTOR);
5731 if (vmx_get_cpl(vcpu)) {
5732 kvm_inject_gp(vcpu, 0);
5739 static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
5742 if (enable_shadow_vmcs) {
5743 if (vmx->nested.current_vmcs12 != NULL) {
5744 /* copy to memory all shadowed fields in case
5745 they were modified */
5746 copy_shadow_to_vmcs12(vmx);
5747 vmx->nested.sync_shadow_vmcs = false;
5748 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5749 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
5750 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
5751 vmcs_write64(VMCS_LINK_POINTER, -1ull);
5754 kunmap(vmx->nested.current_vmcs12_page);
5755 nested_release_page(vmx->nested.current_vmcs12_page);
5759 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
5760 * just stops using VMX.
5762 static void free_nested(struct vcpu_vmx *vmx)
5764 if (!vmx->nested.vmxon)
5766 vmx->nested.vmxon = false;
5767 if (vmx->nested.current_vmptr != -1ull) {
5768 nested_release_vmcs12(vmx);
5769 vmx->nested.current_vmptr = -1ull;
5770 vmx->nested.current_vmcs12 = NULL;
5772 if (enable_shadow_vmcs)
5773 free_vmcs(vmx->nested.current_shadow_vmcs);
5774 /* Unpin physical memory we referred to in current vmcs02 */
5775 if (vmx->nested.apic_access_page) {
5776 nested_release_page(vmx->nested.apic_access_page);
5777 vmx->nested.apic_access_page = 0;
5780 nested_free_all_saved_vmcss(vmx);
5783 /* Emulate the VMXOFF instruction */
5784 static int handle_vmoff(struct kvm_vcpu *vcpu)
5786 if (!nested_vmx_check_permission(vcpu))
5788 free_nested(to_vmx(vcpu));
5789 skip_emulated_instruction(vcpu);
5790 nested_vmx_succeed(vcpu);
5795 * Decode the memory-address operand of a vmx instruction, as recorded on an
5796 * exit caused by such an instruction (run by a guest hypervisor).
5797 * On success, returns 0. When the operand is invalid, returns 1 and throws
5800 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
5801 unsigned long exit_qualification,
5802 u32 vmx_instruction_info, gva_t *ret)
5805 * According to Vol. 3B, "Information for VM Exits Due to Instruction
5806 * Execution", on an exit, vmx_instruction_info holds most of the
5807 * addressing components of the operand. Only the displacement part
5808 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
5809 * For how an actual address is calculated from all these components,
5810 * refer to Vol. 1, "Operand Addressing".
5812 int scaling = vmx_instruction_info & 3;
5813 int addr_size = (vmx_instruction_info >> 7) & 7;
5814 bool is_reg = vmx_instruction_info & (1u << 10);
5815 int seg_reg = (vmx_instruction_info >> 15) & 7;
5816 int index_reg = (vmx_instruction_info >> 18) & 0xf;
5817 bool index_is_valid = !(vmx_instruction_info & (1u << 22));
5818 int base_reg = (vmx_instruction_info >> 23) & 0xf;
5819 bool base_is_valid = !(vmx_instruction_info & (1u << 27));
5822 kvm_queue_exception(vcpu, UD_VECTOR);
5826 /* Addr = segment_base + offset */
5827 /* offset = base + [index * scale] + displacement */
5828 *ret = vmx_get_segment_base(vcpu, seg_reg);
5830 *ret += kvm_register_read(vcpu, base_reg);
5832 *ret += kvm_register_read(vcpu, index_reg)<<scaling;
5833 *ret += exit_qualification; /* holds the displacement */
5835 if (addr_size == 1) /* 32 bit */
5839 * TODO: throw #GP (and return 1) in various cases that the VM*
5840 * instructions require it - e.g., offset beyond segment limit,
5841 * unusable or unreadable/unwritable segment, non-canonical 64-bit
5842 * address, and so on. Currently these are not checked.
5847 /* Emulate the VMCLEAR instruction */
5848 static int handle_vmclear(struct kvm_vcpu *vcpu)
5850 struct vcpu_vmx *vmx = to_vmx(vcpu);
5853 struct vmcs12 *vmcs12;
5855 struct x86_exception e;
5857 if (!nested_vmx_check_permission(vcpu))
5860 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5861 vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
5864 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
5865 sizeof(vmptr), &e)) {
5866 kvm_inject_page_fault(vcpu, &e);
5870 if (!IS_ALIGNED(vmptr, PAGE_SIZE)) {
5871 nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
5872 skip_emulated_instruction(vcpu);
5876 if (vmptr == vmx->nested.current_vmptr) {
5877 nested_release_vmcs12(vmx);
5878 vmx->nested.current_vmptr = -1ull;
5879 vmx->nested.current_vmcs12 = NULL;
5882 page = nested_get_page(vcpu, vmptr);
5885 * For accurate processor emulation, VMCLEAR beyond available
5886 * physical memory should do nothing at all. However, it is
5887 * possible that a nested vmx bug, not a guest hypervisor bug,
5888 * resulted in this case, so let's shut down before doing any
5891 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5894 vmcs12 = kmap(page);
5895 vmcs12->launch_state = 0;
5897 nested_release_page(page);
5899 nested_free_vmcs02(vmx, vmptr);
5901 skip_emulated_instruction(vcpu);
5902 nested_vmx_succeed(vcpu);
5906 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
5908 /* Emulate the VMLAUNCH instruction */
5909 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
5911 return nested_vmx_run(vcpu, true);
5914 /* Emulate the VMRESUME instruction */
5915 static int handle_vmresume(struct kvm_vcpu *vcpu)
5918 return nested_vmx_run(vcpu, false);
5921 enum vmcs_field_type {
5922 VMCS_FIELD_TYPE_U16 = 0,
5923 VMCS_FIELD_TYPE_U64 = 1,
5924 VMCS_FIELD_TYPE_U32 = 2,
5925 VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
5928 static inline int vmcs_field_type(unsigned long field)
5930 if (0x1 & field) /* the *_HIGH fields are all 32 bit */
5931 return VMCS_FIELD_TYPE_U32;
5932 return (field >> 13) & 0x3 ;
5935 static inline int vmcs_field_readonly(unsigned long field)
5937 return (((field >> 10) & 0x3) == 1);
5941 * Read a vmcs12 field. Since these can have varying lengths and we return
5942 * one type, we chose the biggest type (u64) and zero-extend the return value
5943 * to that size. Note that the caller, handle_vmread, might need to use only
5944 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
5945 * 64-bit fields are to be returned).
5947 static inline bool vmcs12_read_any(struct kvm_vcpu *vcpu,
5948 unsigned long field, u64 *ret)
5950 short offset = vmcs_field_to_offset(field);
5956 p = ((char *)(get_vmcs12(vcpu))) + offset;
5958 switch (vmcs_field_type(field)) {
5959 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
5960 *ret = *((natural_width *)p);
5962 case VMCS_FIELD_TYPE_U16:
5965 case VMCS_FIELD_TYPE_U32:
5968 case VMCS_FIELD_TYPE_U64:
5972 return 0; /* can never happen. */
5977 static inline bool vmcs12_write_any(struct kvm_vcpu *vcpu,
5978 unsigned long field, u64 field_value){
5979 short offset = vmcs_field_to_offset(field);
5980 char *p = ((char *) get_vmcs12(vcpu)) + offset;
5984 switch (vmcs_field_type(field)) {
5985 case VMCS_FIELD_TYPE_U16:
5986 *(u16 *)p = field_value;
5988 case VMCS_FIELD_TYPE_U32:
5989 *(u32 *)p = field_value;
5991 case VMCS_FIELD_TYPE_U64:
5992 *(u64 *)p = field_value;
5994 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
5995 *(natural_width *)p = field_value;
5998 return false; /* can never happen. */
6003 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
6006 unsigned long field;
6008 struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
6009 const unsigned long *fields = shadow_read_write_fields;
6010 const int num_fields = max_shadow_read_write_fields;
6012 vmcs_load(shadow_vmcs);
6014 for (i = 0; i < num_fields; i++) {
6016 switch (vmcs_field_type(field)) {
6017 case VMCS_FIELD_TYPE_U16:
6018 field_value = vmcs_read16(field);
6020 case VMCS_FIELD_TYPE_U32:
6021 field_value = vmcs_read32(field);
6023 case VMCS_FIELD_TYPE_U64:
6024 field_value = vmcs_read64(field);
6026 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6027 field_value = vmcs_readl(field);
6030 vmcs12_write_any(&vmx->vcpu, field, field_value);
6033 vmcs_clear(shadow_vmcs);
6034 vmcs_load(vmx->loaded_vmcs->vmcs);
6037 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
6039 const unsigned long *fields[] = {
6040 shadow_read_write_fields,
6041 shadow_read_only_fields
6043 const int max_fields[] = {
6044 max_shadow_read_write_fields,
6045 max_shadow_read_only_fields
6048 unsigned long field;
6049 u64 field_value = 0;
6050 struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
6052 vmcs_load(shadow_vmcs);
6054 for (q = 0; q < ARRAY_SIZE(fields); q++) {
6055 for (i = 0; i < max_fields[q]; i++) {
6056 field = fields[q][i];
6057 vmcs12_read_any(&vmx->vcpu, field, &field_value);
6059 switch (vmcs_field_type(field)) {
6060 case VMCS_FIELD_TYPE_U16:
6061 vmcs_write16(field, (u16)field_value);
6063 case VMCS_FIELD_TYPE_U32:
6064 vmcs_write32(field, (u32)field_value);
6066 case VMCS_FIELD_TYPE_U64:
6067 vmcs_write64(field, (u64)field_value);
6069 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6070 vmcs_writel(field, (long)field_value);
6076 vmcs_clear(shadow_vmcs);
6077 vmcs_load(vmx->loaded_vmcs->vmcs);
6081 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
6082 * used before) all generate the same failure when it is missing.
6084 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
6086 struct vcpu_vmx *vmx = to_vmx(vcpu);
6087 if (vmx->nested.current_vmptr == -1ull) {
6088 nested_vmx_failInvalid(vcpu);
6089 skip_emulated_instruction(vcpu);
6095 static int handle_vmread(struct kvm_vcpu *vcpu)
6097 unsigned long field;
6099 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6100 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6103 if (!nested_vmx_check_permission(vcpu) ||
6104 !nested_vmx_check_vmcs12(vcpu))
6107 /* Decode instruction info and find the field to read */
6108 field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
6109 /* Read the field, zero-extended to a u64 field_value */
6110 if (!vmcs12_read_any(vcpu, field, &field_value)) {
6111 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
6112 skip_emulated_instruction(vcpu);
6116 * Now copy part of this value to register or memory, as requested.
6117 * Note that the number of bits actually copied is 32 or 64 depending
6118 * on the guest's mode (32 or 64 bit), not on the given field's length.
6120 if (vmx_instruction_info & (1u << 10)) {
6121 kvm_register_write(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
6124 if (get_vmx_mem_address(vcpu, exit_qualification,
6125 vmx_instruction_info, &gva))
6127 /* _system ok, as nested_vmx_check_permission verified cpl=0 */
6128 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
6129 &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
6132 nested_vmx_succeed(vcpu);
6133 skip_emulated_instruction(vcpu);
6138 static int handle_vmwrite(struct kvm_vcpu *vcpu)
6140 unsigned long field;
6142 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6143 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6144 /* The value to write might be 32 or 64 bits, depending on L1's long
6145 * mode, and eventually we need to write that into a field of several
6146 * possible lengths. The code below first zero-extends the value to 64
6147 * bit (field_value), and then copies only the approriate number of
6148 * bits into the vmcs12 field.
6150 u64 field_value = 0;
6151 struct x86_exception e;
6153 if (!nested_vmx_check_permission(vcpu) ||
6154 !nested_vmx_check_vmcs12(vcpu))
6157 if (vmx_instruction_info & (1u << 10))
6158 field_value = kvm_register_read(vcpu,
6159 (((vmx_instruction_info) >> 3) & 0xf));
6161 if (get_vmx_mem_address(vcpu, exit_qualification,
6162 vmx_instruction_info, &gva))
6164 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
6165 &field_value, (is_long_mode(vcpu) ? 8 : 4), &e)) {
6166 kvm_inject_page_fault(vcpu, &e);
6172 field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
6173 if (vmcs_field_readonly(field)) {
6174 nested_vmx_failValid(vcpu,
6175 VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
6176 skip_emulated_instruction(vcpu);
6180 if (!vmcs12_write_any(vcpu, field, field_value)) {
6181 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
6182 skip_emulated_instruction(vcpu);
6186 nested_vmx_succeed(vcpu);
6187 skip_emulated_instruction(vcpu);
6191 /* Emulate the VMPTRLD instruction */
6192 static int handle_vmptrld(struct kvm_vcpu *vcpu)
6194 struct vcpu_vmx *vmx = to_vmx(vcpu);
6197 struct x86_exception e;
6200 if (!nested_vmx_check_permission(vcpu))
6203 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
6204 vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
6207 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
6208 sizeof(vmptr), &e)) {
6209 kvm_inject_page_fault(vcpu, &e);
6213 if (!IS_ALIGNED(vmptr, PAGE_SIZE)) {
6214 nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
6215 skip_emulated_instruction(vcpu);
6219 if (vmx->nested.current_vmptr != vmptr) {
6220 struct vmcs12 *new_vmcs12;
6222 page = nested_get_page(vcpu, vmptr);
6224 nested_vmx_failInvalid(vcpu);
6225 skip_emulated_instruction(vcpu);
6228 new_vmcs12 = kmap(page);
6229 if (new_vmcs12->revision_id != VMCS12_REVISION) {
6231 nested_release_page_clean(page);
6232 nested_vmx_failValid(vcpu,
6233 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
6234 skip_emulated_instruction(vcpu);
6237 if (vmx->nested.current_vmptr != -1ull)
6238 nested_release_vmcs12(vmx);
6240 vmx->nested.current_vmptr = vmptr;
6241 vmx->nested.current_vmcs12 = new_vmcs12;
6242 vmx->nested.current_vmcs12_page = page;
6243 if (enable_shadow_vmcs) {
6244 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6245 exec_control |= SECONDARY_EXEC_SHADOW_VMCS;
6246 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
6247 vmcs_write64(VMCS_LINK_POINTER,
6248 __pa(vmx->nested.current_shadow_vmcs));
6249 vmx->nested.sync_shadow_vmcs = true;
6253 nested_vmx_succeed(vcpu);
6254 skip_emulated_instruction(vcpu);
6258 /* Emulate the VMPTRST instruction */
6259 static int handle_vmptrst(struct kvm_vcpu *vcpu)
6261 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6262 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6264 struct x86_exception e;
6266 if (!nested_vmx_check_permission(vcpu))
6269 if (get_vmx_mem_address(vcpu, exit_qualification,
6270 vmx_instruction_info, &vmcs_gva))
6272 /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
6273 if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
6274 (void *)&to_vmx(vcpu)->nested.current_vmptr,
6276 kvm_inject_page_fault(vcpu, &e);
6279 nested_vmx_succeed(vcpu);
6280 skip_emulated_instruction(vcpu);
6284 /* Emulate the INVEPT instruction */
6285 static int handle_invept(struct kvm_vcpu *vcpu)
6287 u32 vmx_instruction_info, types;
6290 struct x86_exception e;
6294 u64 eptp_mask = ((1ull << 51) - 1) & PAGE_MASK;
6296 if (!(nested_vmx_secondary_ctls_high & SECONDARY_EXEC_ENABLE_EPT) ||
6297 !(nested_vmx_ept_caps & VMX_EPT_INVEPT_BIT)) {
6298 kvm_queue_exception(vcpu, UD_VECTOR);
6302 if (!nested_vmx_check_permission(vcpu))
6305 if (!kvm_read_cr0_bits(vcpu, X86_CR0_PE)) {
6306 kvm_queue_exception(vcpu, UD_VECTOR);
6310 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6311 type = kvm_register_read(vcpu, (vmx_instruction_info >> 28) & 0xf);
6313 types = (nested_vmx_ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
6315 if (!(types & (1UL << type))) {
6316 nested_vmx_failValid(vcpu,
6317 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
6321 /* According to the Intel VMX instruction reference, the memory
6322 * operand is read even if it isn't needed (e.g., for type==global)
6324 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
6325 vmx_instruction_info, &gva))
6327 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
6328 sizeof(operand), &e)) {
6329 kvm_inject_page_fault(vcpu, &e);
6334 case VMX_EPT_EXTENT_CONTEXT:
6335 if ((operand.eptp & eptp_mask) !=
6336 (nested_ept_get_cr3(vcpu) & eptp_mask))
6338 case VMX_EPT_EXTENT_GLOBAL:
6339 kvm_mmu_sync_roots(vcpu);
6340 kvm_mmu_flush_tlb(vcpu);
6341 nested_vmx_succeed(vcpu);
6348 skip_emulated_instruction(vcpu);
6353 * The exit handlers return 1 if the exit was handled fully and guest execution
6354 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
6355 * to be done to userspace and return 0.
6357 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
6358 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
6359 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
6360 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
6361 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
6362 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
6363 [EXIT_REASON_CR_ACCESS] = handle_cr,
6364 [EXIT_REASON_DR_ACCESS] = handle_dr,
6365 [EXIT_REASON_CPUID] = handle_cpuid,
6366 [EXIT_REASON_MSR_READ] = handle_rdmsr,
6367 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
6368 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
6369 [EXIT_REASON_HLT] = handle_halt,
6370 [EXIT_REASON_INVD] = handle_invd,
6371 [EXIT_REASON_INVLPG] = handle_invlpg,
6372 [EXIT_REASON_RDPMC] = handle_rdpmc,
6373 [EXIT_REASON_VMCALL] = handle_vmcall,
6374 [EXIT_REASON_VMCLEAR] = handle_vmclear,
6375 [EXIT_REASON_VMLAUNCH] = handle_vmlaunch,
6376 [EXIT_REASON_VMPTRLD] = handle_vmptrld,
6377 [EXIT_REASON_VMPTRST] = handle_vmptrst,
6378 [EXIT_REASON_VMREAD] = handle_vmread,
6379 [EXIT_REASON_VMRESUME] = handle_vmresume,
6380 [EXIT_REASON_VMWRITE] = handle_vmwrite,
6381 [EXIT_REASON_VMOFF] = handle_vmoff,
6382 [EXIT_REASON_VMON] = handle_vmon,
6383 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
6384 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
6385 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
6386 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
6387 [EXIT_REASON_WBINVD] = handle_wbinvd,
6388 [EXIT_REASON_XSETBV] = handle_xsetbv,
6389 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
6390 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
6391 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
6392 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
6393 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
6394 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_invalid_op,
6395 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_invalid_op,
6396 [EXIT_REASON_INVEPT] = handle_invept,
6399 static const int kvm_vmx_max_exit_handlers =
6400 ARRAY_SIZE(kvm_vmx_exit_handlers);
6402 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
6403 struct vmcs12 *vmcs12)
6405 unsigned long exit_qualification;
6406 gpa_t bitmap, last_bitmap;
6411 if (nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING))
6414 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
6417 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6419 port = exit_qualification >> 16;
6420 size = (exit_qualification & 7) + 1;
6422 last_bitmap = (gpa_t)-1;
6427 bitmap = vmcs12->io_bitmap_a;
6428 else if (port < 0x10000)
6429 bitmap = vmcs12->io_bitmap_b;
6432 bitmap += (port & 0x7fff) / 8;
6434 if (last_bitmap != bitmap)
6435 if (kvm_read_guest(vcpu->kvm, bitmap, &b, 1))
6437 if (b & (1 << (port & 7)))
6442 last_bitmap = bitmap;
6449 * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
6450 * rather than handle it ourselves in L0. I.e., check whether L1 expressed
6451 * disinterest in the current event (read or write a specific MSR) by using an
6452 * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
6454 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
6455 struct vmcs12 *vmcs12, u32 exit_reason)
6457 u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
6460 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
6464 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
6465 * for the four combinations of read/write and low/high MSR numbers.
6466 * First we need to figure out which of the four to use:
6468 bitmap = vmcs12->msr_bitmap;
6469 if (exit_reason == EXIT_REASON_MSR_WRITE)
6471 if (msr_index >= 0xc0000000) {
6472 msr_index -= 0xc0000000;
6476 /* Then read the msr_index'th bit from this bitmap: */
6477 if (msr_index < 1024*8) {
6479 if (kvm_read_guest(vcpu->kvm, bitmap + msr_index/8, &b, 1))
6481 return 1 & (b >> (msr_index & 7));
6483 return 1; /* let L1 handle the wrong parameter */
6487 * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
6488 * rather than handle it ourselves in L0. I.e., check if L1 wanted to
6489 * intercept (via guest_host_mask etc.) the current event.
6491 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
6492 struct vmcs12 *vmcs12)
6494 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6495 int cr = exit_qualification & 15;
6496 int reg = (exit_qualification >> 8) & 15;
6497 unsigned long val = kvm_register_read(vcpu, reg);
6499 switch ((exit_qualification >> 4) & 3) {
6500 case 0: /* mov to cr */
6503 if (vmcs12->cr0_guest_host_mask &
6504 (val ^ vmcs12->cr0_read_shadow))
6508 if ((vmcs12->cr3_target_count >= 1 &&
6509 vmcs12->cr3_target_value0 == val) ||
6510 (vmcs12->cr3_target_count >= 2 &&
6511 vmcs12->cr3_target_value1 == val) ||
6512 (vmcs12->cr3_target_count >= 3 &&
6513 vmcs12->cr3_target_value2 == val) ||
6514 (vmcs12->cr3_target_count >= 4 &&
6515 vmcs12->cr3_target_value3 == val))
6517 if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
6521 if (vmcs12->cr4_guest_host_mask &
6522 (vmcs12->cr4_read_shadow ^ val))
6526 if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
6532 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
6533 (vmcs12->cr0_read_shadow & X86_CR0_TS))
6536 case 1: /* mov from cr */
6539 if (vmcs12->cpu_based_vm_exec_control &
6540 CPU_BASED_CR3_STORE_EXITING)
6544 if (vmcs12->cpu_based_vm_exec_control &
6545 CPU_BASED_CR8_STORE_EXITING)
6552 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
6553 * cr0. Other attempted changes are ignored, with no exit.
6555 if (vmcs12->cr0_guest_host_mask & 0xe &
6556 (val ^ vmcs12->cr0_read_shadow))
6558 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
6559 !(vmcs12->cr0_read_shadow & 0x1) &&
6568 * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
6569 * should handle it ourselves in L0 (and then continue L2). Only call this
6570 * when in is_guest_mode (L2).
6572 static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu)
6574 u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6575 struct vcpu_vmx *vmx = to_vmx(vcpu);
6576 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6577 u32 exit_reason = vmx->exit_reason;
6579 if (vmx->nested.nested_run_pending)
6582 if (unlikely(vmx->fail)) {
6583 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
6584 vmcs_read32(VM_INSTRUCTION_ERROR));
6588 switch (exit_reason) {
6589 case EXIT_REASON_EXCEPTION_NMI:
6590 if (!is_exception(intr_info))
6592 else if (is_page_fault(intr_info))
6594 return vmcs12->exception_bitmap &
6595 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
6596 case EXIT_REASON_EXTERNAL_INTERRUPT:
6598 case EXIT_REASON_TRIPLE_FAULT:
6600 case EXIT_REASON_PENDING_INTERRUPT:
6601 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
6602 case EXIT_REASON_NMI_WINDOW:
6603 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
6604 case EXIT_REASON_TASK_SWITCH:
6606 case EXIT_REASON_CPUID:
6608 case EXIT_REASON_HLT:
6609 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
6610 case EXIT_REASON_INVD:
6612 case EXIT_REASON_INVLPG:
6613 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
6614 case EXIT_REASON_RDPMC:
6615 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
6616 case EXIT_REASON_RDTSC:
6617 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
6618 case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
6619 case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
6620 case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
6621 case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
6622 case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
6623 case EXIT_REASON_INVEPT:
6625 * VMX instructions trap unconditionally. This allows L1 to
6626 * emulate them for its L2 guest, i.e., allows 3-level nesting!
6629 case EXIT_REASON_CR_ACCESS:
6630 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
6631 case EXIT_REASON_DR_ACCESS:
6632 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
6633 case EXIT_REASON_IO_INSTRUCTION:
6634 return nested_vmx_exit_handled_io(vcpu, vmcs12);
6635 case EXIT_REASON_MSR_READ:
6636 case EXIT_REASON_MSR_WRITE:
6637 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
6638 case EXIT_REASON_INVALID_STATE:
6640 case EXIT_REASON_MWAIT_INSTRUCTION:
6641 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
6642 case EXIT_REASON_MONITOR_INSTRUCTION:
6643 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
6644 case EXIT_REASON_PAUSE_INSTRUCTION:
6645 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
6646 nested_cpu_has2(vmcs12,
6647 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
6648 case EXIT_REASON_MCE_DURING_VMENTRY:
6650 case EXIT_REASON_TPR_BELOW_THRESHOLD:
6652 case EXIT_REASON_APIC_ACCESS:
6653 return nested_cpu_has2(vmcs12,
6654 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
6655 case EXIT_REASON_EPT_VIOLATION:
6656 case EXIT_REASON_EPT_MISCONFIG:
6658 case EXIT_REASON_PREEMPTION_TIMER:
6659 return vmcs12->pin_based_vm_exec_control &
6660 PIN_BASED_VMX_PREEMPTION_TIMER;
6661 case EXIT_REASON_WBINVD:
6662 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
6663 case EXIT_REASON_XSETBV:
6670 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
6672 *info1 = vmcs_readl(EXIT_QUALIFICATION);
6673 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
6677 * The guest has exited. See if we can fix it or if we need userspace
6680 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
6682 struct vcpu_vmx *vmx = to_vmx(vcpu);
6683 u32 exit_reason = vmx->exit_reason;
6684 u32 vectoring_info = vmx->idt_vectoring_info;
6686 /* If guest state is invalid, start emulating */
6687 if (vmx->emulation_required)
6688 return handle_invalid_guest_state(vcpu);
6691 * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
6692 * we did not inject a still-pending event to L1 now because of
6693 * nested_run_pending, we need to re-enable this bit.
6695 if (vmx->nested.nested_run_pending)
6696 kvm_make_request(KVM_REQ_EVENT, vcpu);
6698 if (!is_guest_mode(vcpu) && (exit_reason == EXIT_REASON_VMLAUNCH ||
6699 exit_reason == EXIT_REASON_VMRESUME))
6700 vmx->nested.nested_run_pending = 1;
6702 vmx->nested.nested_run_pending = 0;
6704 if (is_guest_mode(vcpu) && nested_vmx_exit_handled(vcpu)) {
6705 nested_vmx_vmexit(vcpu);
6709 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
6710 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
6711 vcpu->run->fail_entry.hardware_entry_failure_reason
6716 if (unlikely(vmx->fail)) {
6717 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
6718 vcpu->run->fail_entry.hardware_entry_failure_reason
6719 = vmcs_read32(VM_INSTRUCTION_ERROR);
6725 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
6726 * delivery event since it indicates guest is accessing MMIO.
6727 * The vm-exit can be triggered again after return to guest that
6728 * will cause infinite loop.
6730 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
6731 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
6732 exit_reason != EXIT_REASON_EPT_VIOLATION &&
6733 exit_reason != EXIT_REASON_TASK_SWITCH)) {
6734 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6735 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
6736 vcpu->run->internal.ndata = 2;
6737 vcpu->run->internal.data[0] = vectoring_info;
6738 vcpu->run->internal.data[1] = exit_reason;
6742 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked &&
6743 !(is_guest_mode(vcpu) && nested_cpu_has_virtual_nmis(
6744 get_vmcs12(vcpu), vcpu)))) {
6745 if (vmx_interrupt_allowed(vcpu)) {
6746 vmx->soft_vnmi_blocked = 0;
6747 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
6748 vcpu->arch.nmi_pending) {
6750 * This CPU don't support us in finding the end of an
6751 * NMI-blocked window if the guest runs with IRQs
6752 * disabled. So we pull the trigger after 1 s of
6753 * futile waiting, but inform the user about this.
6755 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
6756 "state on VCPU %d after 1 s timeout\n",
6757 __func__, vcpu->vcpu_id);
6758 vmx->soft_vnmi_blocked = 0;
6762 if (exit_reason < kvm_vmx_max_exit_handlers
6763 && kvm_vmx_exit_handlers[exit_reason])
6764 return kvm_vmx_exit_handlers[exit_reason](vcpu);
6766 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
6767 vcpu->run->hw.hardware_exit_reason = exit_reason;
6772 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
6774 if (irr == -1 || tpr < irr) {
6775 vmcs_write32(TPR_THRESHOLD, 0);
6779 vmcs_write32(TPR_THRESHOLD, irr);
6782 static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
6784 u32 sec_exec_control;
6787 * There is not point to enable virtualize x2apic without enable
6790 if (!cpu_has_vmx_virtualize_x2apic_mode() ||
6791 !vmx_vm_has_apicv(vcpu->kvm))
6794 if (!vm_need_tpr_shadow(vcpu->kvm))
6797 sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6800 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6801 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6803 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6804 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6806 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
6808 vmx_set_msr_bitmap(vcpu);
6811 static void vmx_hwapic_isr_update(struct kvm *kvm, int isr)
6816 if (!vmx_vm_has_apicv(kvm))
6822 status = vmcs_read16(GUEST_INTR_STATUS);
6827 vmcs_write16(GUEST_INTR_STATUS, status);
6831 static void vmx_set_rvi(int vector)
6836 status = vmcs_read16(GUEST_INTR_STATUS);
6837 old = (u8)status & 0xff;
6838 if ((u8)vector != old) {
6840 status |= (u8)vector;
6841 vmcs_write16(GUEST_INTR_STATUS, status);
6845 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
6850 vmx_set_rvi(max_irr);
6853 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
6855 if (!vmx_vm_has_apicv(vcpu->kvm))
6858 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
6859 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
6860 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
6861 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
6864 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
6868 if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
6869 || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
6872 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6873 exit_intr_info = vmx->exit_intr_info;
6875 /* Handle machine checks before interrupts are enabled */
6876 if (is_machine_check(exit_intr_info))
6877 kvm_machine_check();
6879 /* We need to handle NMIs before interrupts are enabled */
6880 if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
6881 (exit_intr_info & INTR_INFO_VALID_MASK)) {
6882 kvm_before_handle_nmi(&vmx->vcpu);
6884 kvm_after_handle_nmi(&vmx->vcpu);
6888 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
6890 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6893 * If external interrupt exists, IF bit is set in rflags/eflags on the
6894 * interrupt stack frame, and interrupt will be enabled on a return
6895 * from interrupt handler.
6897 if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
6898 == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
6899 unsigned int vector;
6900 unsigned long entry;
6902 struct vcpu_vmx *vmx = to_vmx(vcpu);
6903 #ifdef CONFIG_X86_64
6907 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6908 desc = (gate_desc *)vmx->host_idt_base + vector;
6909 entry = gate_offset(*desc);
6911 #ifdef CONFIG_X86_64
6912 "mov %%" _ASM_SP ", %[sp]\n\t"
6913 "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
6918 "orl $0x200, (%%" _ASM_SP ")\n\t"
6919 __ASM_SIZE(push) " $%c[cs]\n\t"
6920 "call *%[entry]\n\t"
6922 #ifdef CONFIG_X86_64
6927 [ss]"i"(__KERNEL_DS),
6928 [cs]"i"(__KERNEL_CS)
6934 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6939 bool idtv_info_valid;
6941 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6943 if (cpu_has_virtual_nmis()) {
6944 if (vmx->nmi_known_unmasked)
6947 * Can't use vmx->exit_intr_info since we're not sure what
6948 * the exit reason is.
6950 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6951 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6952 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6954 * SDM 3: 27.7.1.2 (September 2008)
6955 * Re-set bit "block by NMI" before VM entry if vmexit caused by
6956 * a guest IRET fault.
6957 * SDM 3: 23.2.2 (September 2008)
6958 * Bit 12 is undefined in any of the following cases:
6959 * If the VM exit sets the valid bit in the IDT-vectoring
6960 * information field.
6961 * If the VM exit is due to a double fault.
6963 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
6964 vector != DF_VECTOR && !idtv_info_valid)
6965 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6966 GUEST_INTR_STATE_NMI);
6968 vmx->nmi_known_unmasked =
6969 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
6970 & GUEST_INTR_STATE_NMI);
6971 } else if (unlikely(vmx->soft_vnmi_blocked))
6972 vmx->vnmi_blocked_time +=
6973 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
6976 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
6977 u32 idt_vectoring_info,
6978 int instr_len_field,
6979 int error_code_field)
6983 bool idtv_info_valid;
6985 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6987 vcpu->arch.nmi_injected = false;
6988 kvm_clear_exception_queue(vcpu);
6989 kvm_clear_interrupt_queue(vcpu);
6991 if (!idtv_info_valid)
6994 kvm_make_request(KVM_REQ_EVENT, vcpu);
6996 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6997 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
7000 case INTR_TYPE_NMI_INTR:
7001 vcpu->arch.nmi_injected = true;
7003 * SDM 3: 27.7.1.2 (September 2008)
7004 * Clear bit "block by NMI" before VM entry if a NMI
7007 vmx_set_nmi_mask(vcpu, false);
7009 case INTR_TYPE_SOFT_EXCEPTION:
7010 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
7012 case INTR_TYPE_HARD_EXCEPTION:
7013 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
7014 u32 err = vmcs_read32(error_code_field);
7015 kvm_queue_exception_e(vcpu, vector, err);
7017 kvm_queue_exception(vcpu, vector);
7019 case INTR_TYPE_SOFT_INTR:
7020 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
7022 case INTR_TYPE_EXT_INTR:
7023 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
7030 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
7032 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
7033 VM_EXIT_INSTRUCTION_LEN,
7034 IDT_VECTORING_ERROR_CODE);
7037 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
7039 __vmx_complete_interrupts(vcpu,
7040 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
7041 VM_ENTRY_INSTRUCTION_LEN,
7042 VM_ENTRY_EXCEPTION_ERROR_CODE);
7044 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
7047 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
7050 struct perf_guest_switch_msr *msrs;
7052 msrs = perf_guest_get_msrs(&nr_msrs);
7057 for (i = 0; i < nr_msrs; i++)
7058 if (msrs[i].host == msrs[i].guest)
7059 clear_atomic_switch_msr(vmx, msrs[i].msr);
7061 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
7065 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
7067 struct vcpu_vmx *vmx = to_vmx(vcpu);
7068 unsigned long debugctlmsr;
7070 /* Record the guest's net vcpu time for enforced NMI injections. */
7071 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
7072 vmx->entry_time = ktime_get();
7074 /* Don't enter VMX if guest state is invalid, let the exit handler
7075 start emulation until we arrive back to a valid state */
7076 if (vmx->emulation_required)
7079 if (vmx->nested.sync_shadow_vmcs) {
7080 copy_vmcs12_to_shadow(vmx);
7081 vmx->nested.sync_shadow_vmcs = false;
7084 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
7085 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
7086 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
7087 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
7089 /* When single-stepping over STI and MOV SS, we must clear the
7090 * corresponding interruptibility bits in the guest state. Otherwise
7091 * vmentry fails as it then expects bit 14 (BS) in pending debug
7092 * exceptions being set, but that's not correct for the guest debugging
7094 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7095 vmx_set_interrupt_shadow(vcpu, 0);
7097 atomic_switch_perf_msrs(vmx);
7098 debugctlmsr = get_debugctlmsr();
7100 vmx->__launched = vmx->loaded_vmcs->launched;
7102 /* Store host registers */
7103 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
7104 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
7105 "push %%" _ASM_CX " \n\t"
7106 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
7108 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
7109 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
7111 /* Reload cr2 if changed */
7112 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
7113 "mov %%cr2, %%" _ASM_DX " \n\t"
7114 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
7116 "mov %%" _ASM_AX", %%cr2 \n\t"
7118 /* Check if vmlaunch of vmresume is needed */
7119 "cmpl $0, %c[launched](%0) \n\t"
7120 /* Load guest registers. Don't clobber flags. */
7121 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
7122 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
7123 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
7124 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
7125 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
7126 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
7127 #ifdef CONFIG_X86_64
7128 "mov %c[r8](%0), %%r8 \n\t"
7129 "mov %c[r9](%0), %%r9 \n\t"
7130 "mov %c[r10](%0), %%r10 \n\t"
7131 "mov %c[r11](%0), %%r11 \n\t"
7132 "mov %c[r12](%0), %%r12 \n\t"
7133 "mov %c[r13](%0), %%r13 \n\t"
7134 "mov %c[r14](%0), %%r14 \n\t"
7135 "mov %c[r15](%0), %%r15 \n\t"
7137 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
7139 /* Enter guest mode */
7141 __ex(ASM_VMX_VMLAUNCH) "\n\t"
7143 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
7145 /* Save guest registers, load host registers, keep flags */
7146 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
7148 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
7149 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
7150 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
7151 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
7152 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
7153 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
7154 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
7155 #ifdef CONFIG_X86_64
7156 "mov %%r8, %c[r8](%0) \n\t"
7157 "mov %%r9, %c[r9](%0) \n\t"
7158 "mov %%r10, %c[r10](%0) \n\t"
7159 "mov %%r11, %c[r11](%0) \n\t"
7160 "mov %%r12, %c[r12](%0) \n\t"
7161 "mov %%r13, %c[r13](%0) \n\t"
7162 "mov %%r14, %c[r14](%0) \n\t"
7163 "mov %%r15, %c[r15](%0) \n\t"
7165 "mov %%cr2, %%" _ASM_AX " \n\t"
7166 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
7168 "pop %%" _ASM_BP "; pop %%" _ASM_DX " \n\t"
7169 "setbe %c[fail](%0) \n\t"
7170 ".pushsection .rodata \n\t"
7171 ".global vmx_return \n\t"
7172 "vmx_return: " _ASM_PTR " 2b \n\t"
7174 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
7175 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
7176 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
7177 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
7178 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
7179 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
7180 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
7181 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
7182 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
7183 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
7184 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
7185 #ifdef CONFIG_X86_64
7186 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
7187 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
7188 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
7189 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
7190 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
7191 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
7192 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
7193 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
7195 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
7196 [wordsize]"i"(sizeof(ulong))
7198 #ifdef CONFIG_X86_64
7199 , "rax", "rbx", "rdi", "rsi"
7200 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
7202 , "eax", "ebx", "edi", "esi"
7206 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
7208 update_debugctlmsr(debugctlmsr);
7210 #ifndef CONFIG_X86_64
7212 * The sysexit path does not restore ds/es, so we must set them to
7213 * a reasonable value ourselves.
7215 * We can't defer this to vmx_load_host_state() since that function
7216 * may be executed in interrupt context, which saves and restore segments
7217 * around it, nullifying its effect.
7219 loadsegment(ds, __USER_DS);
7220 loadsegment(es, __USER_DS);
7223 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
7224 | (1 << VCPU_EXREG_RFLAGS)
7225 | (1 << VCPU_EXREG_CPL)
7226 | (1 << VCPU_EXREG_PDPTR)
7227 | (1 << VCPU_EXREG_SEGMENTS)
7228 | (1 << VCPU_EXREG_CR3));
7229 vcpu->arch.regs_dirty = 0;
7231 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
7233 vmx->loaded_vmcs->launched = 1;
7235 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
7236 trace_kvm_exit(vmx->exit_reason, vcpu, KVM_ISA_VMX);
7238 vmx_complete_atomic_exit(vmx);
7239 vmx_recover_nmi_blocking(vmx);
7240 vmx_complete_interrupts(vmx);
7243 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
7245 struct vcpu_vmx *vmx = to_vmx(vcpu);
7249 free_loaded_vmcs(vmx->loaded_vmcs);
7250 kfree(vmx->guest_msrs);
7251 kvm_vcpu_uninit(vcpu);
7252 kmem_cache_free(kvm_vcpu_cache, vmx);
7255 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
7258 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
7262 return ERR_PTR(-ENOMEM);
7266 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
7270 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
7272 if (!vmx->guest_msrs) {
7276 vmx->loaded_vmcs = &vmx->vmcs01;
7277 vmx->loaded_vmcs->vmcs = alloc_vmcs();
7278 if (!vmx->loaded_vmcs->vmcs)
7281 kvm_cpu_vmxon(__pa(per_cpu(vmxarea, raw_smp_processor_id())));
7282 loaded_vmcs_init(vmx->loaded_vmcs);
7287 vmx_vcpu_load(&vmx->vcpu, cpu);
7288 vmx->vcpu.cpu = cpu;
7289 err = vmx_vcpu_setup(vmx);
7290 vmx_vcpu_put(&vmx->vcpu);
7294 if (vm_need_virtualize_apic_accesses(kvm)) {
7295 err = alloc_apic_access_page(kvm);
7301 if (!kvm->arch.ept_identity_map_addr)
7302 kvm->arch.ept_identity_map_addr =
7303 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
7305 if (alloc_identity_pagetable(kvm) != 0)
7307 if (!init_rmode_identity_map(kvm))
7311 vmx->nested.current_vmptr = -1ull;
7312 vmx->nested.current_vmcs12 = NULL;
7317 free_loaded_vmcs(vmx->loaded_vmcs);
7319 kfree(vmx->guest_msrs);
7321 kvm_vcpu_uninit(&vmx->vcpu);
7324 kmem_cache_free(kvm_vcpu_cache, vmx);
7325 return ERR_PTR(err);
7328 static void __init vmx_check_processor_compat(void *rtn)
7330 struct vmcs_config vmcs_conf;
7333 if (setup_vmcs_config(&vmcs_conf) < 0)
7335 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
7336 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
7337 smp_processor_id());
7342 static int get_ept_level(void)
7344 return VMX_EPT_DEFAULT_GAW + 1;
7347 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
7351 /* For VT-d and EPT combination
7352 * 1. MMIO: always map as UC
7354 * a. VT-d without snooping control feature: can't guarantee the
7355 * result, try to trust guest.
7356 * b. VT-d with snooping control feature: snooping control feature of
7357 * VT-d engine can guarantee the cache correctness. Just set it
7358 * to WB to keep consistent with host. So the same as item 3.
7359 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
7360 * consistent with host MTRR
7363 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
7364 else if (vcpu->kvm->arch.iommu_domain &&
7365 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY))
7366 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
7367 VMX_EPT_MT_EPTE_SHIFT;
7369 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
7375 static int vmx_get_lpage_level(void)
7377 if (enable_ept && !cpu_has_vmx_ept_1g_page())
7378 return PT_DIRECTORY_LEVEL;
7380 /* For shadow and EPT supported 1GB page */
7381 return PT_PDPE_LEVEL;
7384 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
7386 struct kvm_cpuid_entry2 *best;
7387 struct vcpu_vmx *vmx = to_vmx(vcpu);
7390 vmx->rdtscp_enabled = false;
7391 if (vmx_rdtscp_supported()) {
7392 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7393 if (exec_control & SECONDARY_EXEC_RDTSCP) {
7394 best = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
7395 if (best && (best->edx & bit(X86_FEATURE_RDTSCP)))
7396 vmx->rdtscp_enabled = true;
7398 exec_control &= ~SECONDARY_EXEC_RDTSCP;
7399 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
7405 /* Exposing INVPCID only when PCID is exposed */
7406 best = kvm_find_cpuid_entry(vcpu, 0x7, 0);
7407 if (vmx_invpcid_supported() &&
7408 best && (best->ebx & bit(X86_FEATURE_INVPCID)) &&
7409 guest_cpuid_has_pcid(vcpu)) {
7410 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7411 exec_control |= SECONDARY_EXEC_ENABLE_INVPCID;
7412 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
7415 if (cpu_has_secondary_exec_ctrls()) {
7416 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7417 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
7418 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
7422 best->ebx &= ~bit(X86_FEATURE_INVPCID);
7426 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
7428 if (func == 1 && nested)
7429 entry->ecx |= bit(X86_FEATURE_VMX);
7432 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
7433 struct x86_exception *fault)
7435 struct vmcs12 *vmcs12;
7436 nested_vmx_vmexit(vcpu);
7437 vmcs12 = get_vmcs12(vcpu);
7439 if (fault->error_code & PFERR_RSVD_MASK)
7440 vmcs12->vm_exit_reason = EXIT_REASON_EPT_MISCONFIG;
7442 vmcs12->vm_exit_reason = EXIT_REASON_EPT_VIOLATION;
7443 vmcs12->exit_qualification = vcpu->arch.exit_qualification;
7444 vmcs12->guest_physical_address = fault->address;
7447 /* Callbacks for nested_ept_init_mmu_context: */
7449 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
7451 /* return the page table to be shadowed - in our case, EPT12 */
7452 return get_vmcs12(vcpu)->ept_pointer;
7455 static int nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
7457 int r = kvm_init_shadow_ept_mmu(vcpu, &vcpu->arch.mmu,
7458 nested_vmx_ept_caps & VMX_EPT_EXECUTE_ONLY_BIT);
7460 vcpu->arch.mmu.set_cr3 = vmx_set_cr3;
7461 vcpu->arch.mmu.get_cr3 = nested_ept_get_cr3;
7462 vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
7464 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
7469 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
7471 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
7475 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
7476 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
7477 * with L0's requirements for its guest (a.k.a. vmsc01), so we can run the L2
7478 * guest in a way that will both be appropriate to L1's requests, and our
7479 * needs. In addition to modifying the active vmcs (which is vmcs02), this
7480 * function also has additional necessary side-effects, like setting various
7481 * vcpu->arch fields.
7483 static void prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
7485 struct vcpu_vmx *vmx = to_vmx(vcpu);
7488 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
7489 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
7490 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
7491 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
7492 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
7493 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
7494 vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
7495 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
7496 vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
7497 vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
7498 vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
7499 vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
7500 vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
7501 vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
7502 vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
7503 vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
7504 vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
7505 vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
7506 vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
7507 vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
7508 vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
7509 vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
7510 vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
7511 vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
7512 vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
7513 vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
7514 vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
7515 vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
7516 vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
7517 vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
7518 vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
7519 vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
7520 vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
7521 vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
7522 vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
7523 vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
7525 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
7526 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
7527 vmcs12->vm_entry_intr_info_field);
7528 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
7529 vmcs12->vm_entry_exception_error_code);
7530 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
7531 vmcs12->vm_entry_instruction_len);
7532 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
7533 vmcs12->guest_interruptibility_info);
7534 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
7535 kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
7536 vmx_set_rflags(vcpu, vmcs12->guest_rflags);
7537 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
7538 vmcs12->guest_pending_dbg_exceptions);
7539 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
7540 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
7542 vmcs_write64(VMCS_LINK_POINTER, -1ull);
7544 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
7545 (vmcs_config.pin_based_exec_ctrl |
7546 vmcs12->pin_based_vm_exec_control));
7548 if (vmcs12->pin_based_vm_exec_control & PIN_BASED_VMX_PREEMPTION_TIMER)
7549 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE,
7550 vmcs12->vmx_preemption_timer_value);
7553 * Whether page-faults are trapped is determined by a combination of
7554 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
7555 * If enable_ept, L0 doesn't care about page faults and we should
7556 * set all of these to L1's desires. However, if !enable_ept, L0 does
7557 * care about (at least some) page faults, and because it is not easy
7558 * (if at all possible?) to merge L0 and L1's desires, we simply ask
7559 * to exit on each and every L2 page fault. This is done by setting
7560 * MASK=MATCH=0 and (see below) EB.PF=1.
7561 * Note that below we don't need special code to set EB.PF beyond the
7562 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
7563 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
7564 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
7566 * A problem with this approach (when !enable_ept) is that L1 may be
7567 * injected with more page faults than it asked for. This could have
7568 * caused problems, but in practice existing hypervisors don't care.
7569 * To fix this, we will need to emulate the PFEC checking (on the L1
7570 * page tables), using walk_addr(), when injecting PFs to L1.
7572 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
7573 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
7574 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
7575 enable_ept ? vmcs12->page_fault_error_code_match : 0);
7577 if (cpu_has_secondary_exec_ctrls()) {
7578 u32 exec_control = vmx_secondary_exec_control(vmx);
7579 if (!vmx->rdtscp_enabled)
7580 exec_control &= ~SECONDARY_EXEC_RDTSCP;
7581 /* Take the following fields only from vmcs12 */
7582 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
7583 if (nested_cpu_has(vmcs12,
7584 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
7585 exec_control |= vmcs12->secondary_vm_exec_control;
7587 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) {
7589 * Translate L1 physical address to host physical
7590 * address for vmcs02. Keep the page pinned, so this
7591 * physical address remains valid. We keep a reference
7592 * to it so we can release it later.
7594 if (vmx->nested.apic_access_page) /* shouldn't happen */
7595 nested_release_page(vmx->nested.apic_access_page);
7596 vmx->nested.apic_access_page =
7597 nested_get_page(vcpu, vmcs12->apic_access_addr);
7599 * If translation failed, no matter: This feature asks
7600 * to exit when accessing the given address, and if it
7601 * can never be accessed, this feature won't do
7604 if (!vmx->nested.apic_access_page)
7606 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
7608 vmcs_write64(APIC_ACCESS_ADDR,
7609 page_to_phys(vmx->nested.apic_access_page));
7612 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
7617 * Set host-state according to L0's settings (vmcs12 is irrelevant here)
7618 * Some constant fields are set here by vmx_set_constant_host_state().
7619 * Other fields are different per CPU, and will be set later when
7620 * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
7622 vmx_set_constant_host_state(vmx);
7625 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
7626 * entry, but only if the current (host) sp changed from the value
7627 * we wrote last (vmx->host_rsp). This cache is no longer relevant
7628 * if we switch vmcs, and rather than hold a separate cache per vmcs,
7629 * here we just force the write to happen on entry.
7633 exec_control = vmx_exec_control(vmx); /* L0's desires */
7634 exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
7635 exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
7636 exec_control &= ~CPU_BASED_TPR_SHADOW;
7637 exec_control |= vmcs12->cpu_based_vm_exec_control;
7639 * Merging of IO and MSR bitmaps not currently supported.
7640 * Rather, exit every time.
7642 exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
7643 exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
7644 exec_control |= CPU_BASED_UNCOND_IO_EXITING;
7646 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
7648 /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
7649 * bitwise-or of what L1 wants to trap for L2, and what we want to
7650 * trap. Note that CR0.TS also needs updating - we do this later.
7652 update_exception_bitmap(vcpu);
7653 vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
7654 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
7656 /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
7657 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
7658 * bits are further modified by vmx_set_efer() below.
7660 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
7662 /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
7663 * emulated by vmx_set_efer(), below.
7665 vmcs_write32(VM_ENTRY_CONTROLS,
7666 (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER &
7667 ~VM_ENTRY_IA32E_MODE) |
7668 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
7670 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)
7671 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
7672 else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
7673 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
7676 set_cr4_guest_host_mask(vmx);
7678 if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
7679 vmcs_write64(TSC_OFFSET,
7680 vmx->nested.vmcs01_tsc_offset + vmcs12->tsc_offset);
7682 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
7686 * Trivially support vpid by letting L2s share their parent
7687 * L1's vpid. TODO: move to a more elaborate solution, giving
7688 * each L2 its own vpid and exposing the vpid feature to L1.
7690 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
7691 vmx_flush_tlb(vcpu);
7694 if (nested_cpu_has_ept(vmcs12)) {
7695 kvm_mmu_unload(vcpu);
7696 nested_ept_init_mmu_context(vcpu);
7699 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)
7700 vcpu->arch.efer = vmcs12->guest_ia32_efer;
7701 else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
7702 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
7704 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
7705 /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
7706 vmx_set_efer(vcpu, vcpu->arch.efer);
7709 * This sets GUEST_CR0 to vmcs12->guest_cr0, with possibly a modified
7710 * TS bit (for lazy fpu) and bits which we consider mandatory enabled.
7711 * The CR0_READ_SHADOW is what L2 should have expected to read given
7712 * the specifications by L1; It's not enough to take
7713 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
7714 * have more bits than L1 expected.
7716 vmx_set_cr0(vcpu, vmcs12->guest_cr0);
7717 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
7719 vmx_set_cr4(vcpu, vmcs12->guest_cr4);
7720 vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
7722 /* shadow page tables on either EPT or shadow page tables */
7723 kvm_set_cr3(vcpu, vmcs12->guest_cr3);
7724 kvm_mmu_reset_context(vcpu);
7727 * L1 may access the L2's PDPTR, so save them to construct vmcs12
7730 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
7731 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
7732 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
7733 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
7736 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
7737 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
7741 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
7742 * for running an L2 nested guest.
7744 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
7746 struct vmcs12 *vmcs12;
7747 struct vcpu_vmx *vmx = to_vmx(vcpu);
7749 struct loaded_vmcs *vmcs02;
7752 if (!nested_vmx_check_permission(vcpu) ||
7753 !nested_vmx_check_vmcs12(vcpu))
7756 skip_emulated_instruction(vcpu);
7757 vmcs12 = get_vmcs12(vcpu);
7759 if (enable_shadow_vmcs)
7760 copy_shadow_to_vmcs12(vmx);
7763 * The nested entry process starts with enforcing various prerequisites
7764 * on vmcs12 as required by the Intel SDM, and act appropriately when
7765 * they fail: As the SDM explains, some conditions should cause the
7766 * instruction to fail, while others will cause the instruction to seem
7767 * to succeed, but return an EXIT_REASON_INVALID_STATE.
7768 * To speed up the normal (success) code path, we should avoid checking
7769 * for misconfigurations which will anyway be caught by the processor
7770 * when using the merged vmcs02.
7772 if (vmcs12->launch_state == launch) {
7773 nested_vmx_failValid(vcpu,
7774 launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
7775 : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
7779 if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE) {
7780 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
7784 if ((vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_MSR_BITMAPS) &&
7785 !IS_ALIGNED(vmcs12->msr_bitmap, PAGE_SIZE)) {
7786 /*TODO: Also verify bits beyond physical address width are 0*/
7787 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
7791 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
7792 !IS_ALIGNED(vmcs12->apic_access_addr, PAGE_SIZE)) {
7793 /*TODO: Also verify bits beyond physical address width are 0*/
7794 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
7798 if (vmcs12->vm_entry_msr_load_count > 0 ||
7799 vmcs12->vm_exit_msr_load_count > 0 ||
7800 vmcs12->vm_exit_msr_store_count > 0) {
7801 pr_warn_ratelimited("%s: VMCS MSR_{LOAD,STORE} unsupported\n",
7803 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
7807 if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
7808 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high) ||
7809 !vmx_control_verify(vmcs12->secondary_vm_exec_control,
7810 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high) ||
7811 !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
7812 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high) ||
7813 !vmx_control_verify(vmcs12->vm_exit_controls,
7814 nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high) ||
7815 !vmx_control_verify(vmcs12->vm_entry_controls,
7816 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high))
7818 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
7822 if (((vmcs12->host_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
7823 ((vmcs12->host_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
7824 nested_vmx_failValid(vcpu,
7825 VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7829 if (((vmcs12->guest_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
7830 ((vmcs12->guest_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
7831 nested_vmx_entry_failure(vcpu, vmcs12,
7832 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
7835 if (vmcs12->vmcs_link_pointer != -1ull) {
7836 nested_vmx_entry_failure(vcpu, vmcs12,
7837 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_VMCS_LINK_PTR);
7842 * If the load IA32_EFER VM-entry control is 1, the following checks
7843 * are performed on the field for the IA32_EFER MSR:
7844 * - Bits reserved in the IA32_EFER MSR must be 0.
7845 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
7846 * the IA-32e mode guest VM-exit control. It must also be identical
7847 * to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
7850 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER) {
7851 ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
7852 if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
7853 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
7854 ((vmcs12->guest_cr0 & X86_CR0_PG) &&
7855 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))) {
7856 nested_vmx_entry_failure(vcpu, vmcs12,
7857 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
7863 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
7864 * IA32_EFER MSR must be 0 in the field for that register. In addition,
7865 * the values of the LMA and LME bits in the field must each be that of
7866 * the host address-space size VM-exit control.
7868 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
7869 ia32e = (vmcs12->vm_exit_controls &
7870 VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
7871 if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
7872 ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
7873 ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)) {
7874 nested_vmx_entry_failure(vcpu, vmcs12,
7875 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
7881 * We're finally done with prerequisite checking, and can start with
7885 vmcs02 = nested_get_current_vmcs02(vmx);
7889 enter_guest_mode(vcpu);
7891 vmx->nested.vmcs01_tsc_offset = vmcs_read64(TSC_OFFSET);
7894 vmx->loaded_vmcs = vmcs02;
7896 vmx_vcpu_load(vcpu, cpu);
7900 vmx_segment_cache_clear(vmx);
7902 vmcs12->launch_state = 1;
7904 prepare_vmcs02(vcpu, vmcs12);
7907 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
7908 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
7909 * returned as far as L1 is concerned. It will only return (and set
7910 * the success flag) when L2 exits (see nested_vmx_vmexit()).
7916 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
7917 * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
7918 * This function returns the new value we should put in vmcs12.guest_cr0.
7919 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
7920 * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
7921 * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
7922 * didn't trap the bit, because if L1 did, so would L0).
7923 * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have
7924 * been modified by L2, and L1 knows it. So just leave the old value of
7925 * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
7926 * isn't relevant, because if L0 traps this bit it can set it to anything.
7927 * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
7928 * changed these bits, and therefore they need to be updated, but L0
7929 * didn't necessarily allow them to be changed in GUEST_CR0 - and rather
7930 * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
7932 static inline unsigned long
7933 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
7936 /*1*/ (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
7937 /*2*/ (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
7938 /*3*/ (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
7939 vcpu->arch.cr0_guest_owned_bits));
7942 static inline unsigned long
7943 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
7946 /*1*/ (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
7947 /*2*/ (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
7948 /*3*/ (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
7949 vcpu->arch.cr4_guest_owned_bits));
7952 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
7953 struct vmcs12 *vmcs12)
7958 if (vcpu->arch.exception.pending) {
7959 nr = vcpu->arch.exception.nr;
7960 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
7962 if (kvm_exception_is_soft(nr)) {
7963 vmcs12->vm_exit_instruction_len =
7964 vcpu->arch.event_exit_inst_len;
7965 idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
7967 idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
7969 if (vcpu->arch.exception.has_error_code) {
7970 idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
7971 vmcs12->idt_vectoring_error_code =
7972 vcpu->arch.exception.error_code;
7975 vmcs12->idt_vectoring_info_field = idt_vectoring;
7976 } else if (vcpu->arch.nmi_pending) {
7977 vmcs12->idt_vectoring_info_field =
7978 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
7979 } else if (vcpu->arch.interrupt.pending) {
7980 nr = vcpu->arch.interrupt.nr;
7981 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
7983 if (vcpu->arch.interrupt.soft) {
7984 idt_vectoring |= INTR_TYPE_SOFT_INTR;
7985 vmcs12->vm_entry_instruction_len =
7986 vcpu->arch.event_exit_inst_len;
7988 idt_vectoring |= INTR_TYPE_EXT_INTR;
7990 vmcs12->idt_vectoring_info_field = idt_vectoring;
7995 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
7996 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
7997 * and this function updates it to reflect the changes to the guest state while
7998 * L2 was running (and perhaps made some exits which were handled directly by L0
7999 * without going back to L1), and to reflect the exit reason.
8000 * Note that we do not have to copy here all VMCS fields, just those that
8001 * could have changed by the L2 guest or the exit - i.e., the guest-state and
8002 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
8003 * which already writes to vmcs12 directly.
8005 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
8007 /* update guest state fields: */
8008 vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
8009 vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
8011 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
8012 vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
8013 vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
8014 vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
8016 vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
8017 vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
8018 vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
8019 vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
8020 vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
8021 vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
8022 vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
8023 vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
8024 vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
8025 vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
8026 vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
8027 vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
8028 vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
8029 vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
8030 vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
8031 vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
8032 vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
8033 vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
8034 vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
8035 vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
8036 vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
8037 vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
8038 vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
8039 vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
8040 vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
8041 vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
8042 vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
8043 vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
8044 vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
8045 vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
8046 vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
8047 vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
8048 vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
8049 vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
8050 vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
8051 vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
8053 vmcs12->guest_interruptibility_info =
8054 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
8055 vmcs12->guest_pending_dbg_exceptions =
8056 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
8059 * In some cases (usually, nested EPT), L2 is allowed to change its
8060 * own CR3 without exiting. If it has changed it, we must keep it.
8061 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
8062 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
8064 * Additionally, restore L2's PDPTR to vmcs12.
8067 vmcs12->guest_cr3 = vmcs_read64(GUEST_CR3);
8068 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
8069 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
8070 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
8071 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
8074 vmcs12->vm_entry_controls =
8075 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
8076 (vmcs_read32(VM_ENTRY_CONTROLS) & VM_ENTRY_IA32E_MODE);
8078 /* TODO: These cannot have changed unless we have MSR bitmaps and
8079 * the relevant bit asks not to trap the change */
8080 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
8081 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
8082 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
8083 vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
8084 vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
8085 vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
8087 /* update exit information fields: */
8089 vmcs12->vm_exit_reason = to_vmx(vcpu)->exit_reason;
8090 vmcs12->exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
8092 vmcs12->vm_exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8093 if ((vmcs12->vm_exit_intr_info &
8094 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
8095 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK))
8096 vmcs12->vm_exit_intr_error_code =
8097 vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
8098 vmcs12->idt_vectoring_info_field = 0;
8099 vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
8100 vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
8102 if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
8103 /* vm_entry_intr_info_field is cleared on exit. Emulate this
8104 * instead of reading the real value. */
8105 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
8108 * Transfer the event that L0 or L1 may wanted to inject into
8109 * L2 to IDT_VECTORING_INFO_FIELD.
8111 vmcs12_save_pending_event(vcpu, vmcs12);
8115 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
8116 * preserved above and would only end up incorrectly in L1.
8118 vcpu->arch.nmi_injected = false;
8119 kvm_clear_exception_queue(vcpu);
8120 kvm_clear_interrupt_queue(vcpu);
8124 * A part of what we need to when the nested L2 guest exits and we want to
8125 * run its L1 parent, is to reset L1's guest state to the host state specified
8127 * This function is to be called not only on normal nested exit, but also on
8128 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
8129 * Failures During or After Loading Guest State").
8130 * This function should be called when the active VMCS is L1's (vmcs01).
8132 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
8133 struct vmcs12 *vmcs12)
8135 struct kvm_segment seg;
8137 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
8138 vcpu->arch.efer = vmcs12->host_ia32_efer;
8139 else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
8140 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
8142 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
8143 vmx_set_efer(vcpu, vcpu->arch.efer);
8145 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
8146 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
8147 vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
8149 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
8150 * actually changed, because it depends on the current state of
8151 * fpu_active (which may have changed).
8152 * Note that vmx_set_cr0 refers to efer set above.
8154 kvm_set_cr0(vcpu, vmcs12->host_cr0);
8156 * If we did fpu_activate()/fpu_deactivate() during L2's run, we need
8157 * to apply the same changes to L1's vmcs. We just set cr0 correctly,
8158 * but we also need to update cr0_guest_host_mask and exception_bitmap.
8160 update_exception_bitmap(vcpu);
8161 vcpu->arch.cr0_guest_owned_bits = (vcpu->fpu_active ? X86_CR0_TS : 0);
8162 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
8165 * Note that CR4_GUEST_HOST_MASK is already set in the original vmcs01
8166 * (KVM doesn't change it)- no reason to call set_cr4_guest_host_mask();
8168 vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
8169 kvm_set_cr4(vcpu, vmcs12->host_cr4);
8171 if (nested_cpu_has_ept(vmcs12))
8172 nested_ept_uninit_mmu_context(vcpu);
8174 kvm_set_cr3(vcpu, vmcs12->host_cr3);
8175 kvm_mmu_reset_context(vcpu);
8179 * Trivially support vpid by letting L2s share their parent
8180 * L1's vpid. TODO: move to a more elaborate solution, giving
8181 * each L2 its own vpid and exposing the vpid feature to L1.
8183 vmx_flush_tlb(vcpu);
8187 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
8188 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
8189 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
8190 vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
8191 vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
8193 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT)
8194 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
8195 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
8196 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
8197 vmcs12->host_ia32_perf_global_ctrl);
8199 /* Set L1 segment info according to Intel SDM
8200 27.5.2 Loading Host Segment and Descriptor-Table Registers */
8201 seg = (struct kvm_segment) {
8203 .limit = 0xFFFFFFFF,
8204 .selector = vmcs12->host_cs_selector,
8210 if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
8214 vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
8215 seg = (struct kvm_segment) {
8217 .limit = 0xFFFFFFFF,
8224 seg.selector = vmcs12->host_ds_selector;
8225 vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
8226 seg.selector = vmcs12->host_es_selector;
8227 vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
8228 seg.selector = vmcs12->host_ss_selector;
8229 vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
8230 seg.selector = vmcs12->host_fs_selector;
8231 seg.base = vmcs12->host_fs_base;
8232 vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
8233 seg.selector = vmcs12->host_gs_selector;
8234 seg.base = vmcs12->host_gs_base;
8235 vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
8236 seg = (struct kvm_segment) {
8237 .base = vmcs12->host_tr_base,
8239 .selector = vmcs12->host_tr_selector,
8243 vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
8245 kvm_set_dr(vcpu, 7, 0x400);
8246 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
8250 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
8251 * and modify vmcs12 to make it see what it would expect to see there if
8252 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
8254 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu)
8256 struct vcpu_vmx *vmx = to_vmx(vcpu);
8258 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8260 /* trying to cancel vmlaunch/vmresume is a bug */
8261 WARN_ON_ONCE(vmx->nested.nested_run_pending);
8263 leave_guest_mode(vcpu);
8264 prepare_vmcs12(vcpu, vmcs12);
8267 vmx->loaded_vmcs = &vmx->vmcs01;
8269 vmx_vcpu_load(vcpu, cpu);
8273 vmx_segment_cache_clear(vmx);
8275 /* if no vmcs02 cache requested, remove the one we used */
8276 if (VMCS02_POOL_SIZE == 0)
8277 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
8279 load_vmcs12_host_state(vcpu, vmcs12);
8281 /* Update TSC_OFFSET if TSC was changed while L2 ran */
8282 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
8284 /* This is needed for same reason as it was needed in prepare_vmcs02 */
8287 /* Unpin physical memory we referred to in vmcs02 */
8288 if (vmx->nested.apic_access_page) {
8289 nested_release_page(vmx->nested.apic_access_page);
8290 vmx->nested.apic_access_page = 0;
8294 * Exiting from L2 to L1, we're now back to L1 which thinks it just
8295 * finished a VMLAUNCH or VMRESUME instruction, so we need to set the
8296 * success or failure flag accordingly.
8298 if (unlikely(vmx->fail)) {
8300 nested_vmx_failValid(vcpu, vmcs_read32(VM_INSTRUCTION_ERROR));
8302 nested_vmx_succeed(vcpu);
8303 if (enable_shadow_vmcs)
8304 vmx->nested.sync_shadow_vmcs = true;
8308 * L1's failure to enter L2 is a subset of a normal exit, as explained in
8309 * 23.7 "VM-entry failures during or after loading guest state" (this also
8310 * lists the acceptable exit-reason and exit-qualification parameters).
8311 * It should only be called before L2 actually succeeded to run, and when
8312 * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
8314 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
8315 struct vmcs12 *vmcs12,
8316 u32 reason, unsigned long qualification)
8318 load_vmcs12_host_state(vcpu, vmcs12);
8319 vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
8320 vmcs12->exit_qualification = qualification;
8321 nested_vmx_succeed(vcpu);
8322 if (enable_shadow_vmcs)
8323 to_vmx(vcpu)->nested.sync_shadow_vmcs = true;
8326 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
8327 struct x86_instruction_info *info,
8328 enum x86_intercept_stage stage)
8330 return X86EMUL_CONTINUE;
8333 static struct kvm_x86_ops vmx_x86_ops = {
8334 .cpu_has_kvm_support = cpu_has_kvm_support,
8335 .disabled_by_bios = vmx_disabled_by_bios,
8336 .hardware_setup = hardware_setup,
8337 .hardware_unsetup = hardware_unsetup,
8338 .check_processor_compatibility = vmx_check_processor_compat,
8339 .hardware_enable = hardware_enable,
8340 .hardware_disable = hardware_disable,
8341 .cpu_has_accelerated_tpr = report_flexpriority,
8343 .vcpu_create = vmx_create_vcpu,
8344 .vcpu_free = vmx_free_vcpu,
8345 .vcpu_reset = vmx_vcpu_reset,
8347 .prepare_guest_switch = vmx_save_host_state,
8348 .vcpu_load = vmx_vcpu_load,
8349 .vcpu_put = vmx_vcpu_put,
8351 .update_db_bp_intercept = update_exception_bitmap,
8352 .get_msr = vmx_get_msr,
8353 .set_msr = vmx_set_msr,
8354 .get_segment_base = vmx_get_segment_base,
8355 .get_segment = vmx_get_segment,
8356 .set_segment = vmx_set_segment,
8357 .get_cpl = vmx_get_cpl,
8358 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
8359 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
8360 .decache_cr3 = vmx_decache_cr3,
8361 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
8362 .set_cr0 = vmx_set_cr0,
8363 .set_cr3 = vmx_set_cr3,
8364 .set_cr4 = vmx_set_cr4,
8365 .set_efer = vmx_set_efer,
8366 .get_idt = vmx_get_idt,
8367 .set_idt = vmx_set_idt,
8368 .get_gdt = vmx_get_gdt,
8369 .set_gdt = vmx_set_gdt,
8370 .set_dr7 = vmx_set_dr7,
8371 .cache_reg = vmx_cache_reg,
8372 .get_rflags = vmx_get_rflags,
8373 .set_rflags = vmx_set_rflags,
8374 .fpu_activate = vmx_fpu_activate,
8375 .fpu_deactivate = vmx_fpu_deactivate,
8377 .tlb_flush = vmx_flush_tlb,
8379 .run = vmx_vcpu_run,
8380 .handle_exit = vmx_handle_exit,
8381 .skip_emulated_instruction = skip_emulated_instruction,
8382 .set_interrupt_shadow = vmx_set_interrupt_shadow,
8383 .get_interrupt_shadow = vmx_get_interrupt_shadow,
8384 .patch_hypercall = vmx_patch_hypercall,
8385 .set_irq = vmx_inject_irq,
8386 .set_nmi = vmx_inject_nmi,
8387 .queue_exception = vmx_queue_exception,
8388 .cancel_injection = vmx_cancel_injection,
8389 .interrupt_allowed = vmx_interrupt_allowed,
8390 .nmi_allowed = vmx_nmi_allowed,
8391 .get_nmi_mask = vmx_get_nmi_mask,
8392 .set_nmi_mask = vmx_set_nmi_mask,
8393 .enable_nmi_window = enable_nmi_window,
8394 .enable_irq_window = enable_irq_window,
8395 .update_cr8_intercept = update_cr8_intercept,
8396 .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode,
8397 .vm_has_apicv = vmx_vm_has_apicv,
8398 .load_eoi_exitmap = vmx_load_eoi_exitmap,
8399 .hwapic_irr_update = vmx_hwapic_irr_update,
8400 .hwapic_isr_update = vmx_hwapic_isr_update,
8401 .sync_pir_to_irr = vmx_sync_pir_to_irr,
8402 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
8404 .set_tss_addr = vmx_set_tss_addr,
8405 .get_tdp_level = get_ept_level,
8406 .get_mt_mask = vmx_get_mt_mask,
8408 .get_exit_info = vmx_get_exit_info,
8410 .get_lpage_level = vmx_get_lpage_level,
8412 .cpuid_update = vmx_cpuid_update,
8414 .rdtscp_supported = vmx_rdtscp_supported,
8415 .invpcid_supported = vmx_invpcid_supported,
8417 .set_supported_cpuid = vmx_set_supported_cpuid,
8419 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
8421 .set_tsc_khz = vmx_set_tsc_khz,
8422 .read_tsc_offset = vmx_read_tsc_offset,
8423 .write_tsc_offset = vmx_write_tsc_offset,
8424 .adjust_tsc_offset = vmx_adjust_tsc_offset,
8425 .compute_tsc_offset = vmx_compute_tsc_offset,
8426 .read_l1_tsc = vmx_read_l1_tsc,
8428 .set_tdp_cr3 = vmx_set_cr3,
8430 .check_intercept = vmx_check_intercept,
8431 .handle_external_intr = vmx_handle_external_intr,
8434 static int __init vmx_init(void)
8438 rdmsrl_safe(MSR_EFER, &host_efer);
8440 for (i = 0; i < NR_VMX_MSR; ++i)
8441 kvm_define_shared_msr(i, vmx_msr_index[i]);
8443 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
8444 if (!vmx_io_bitmap_a)
8449 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
8450 if (!vmx_io_bitmap_b)
8453 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
8454 if (!vmx_msr_bitmap_legacy)
8457 vmx_msr_bitmap_legacy_x2apic =
8458 (unsigned long *)__get_free_page(GFP_KERNEL);
8459 if (!vmx_msr_bitmap_legacy_x2apic)
8462 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
8463 if (!vmx_msr_bitmap_longmode)
8466 vmx_msr_bitmap_longmode_x2apic =
8467 (unsigned long *)__get_free_page(GFP_KERNEL);
8468 if (!vmx_msr_bitmap_longmode_x2apic)
8470 vmx_vmread_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
8471 if (!vmx_vmread_bitmap)
8474 vmx_vmwrite_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
8475 if (!vmx_vmwrite_bitmap)
8478 memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
8479 memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
8480 /* shadowed read/write fields */
8481 for (i = 0; i < max_shadow_read_write_fields; i++) {
8482 clear_bit(shadow_read_write_fields[i], vmx_vmwrite_bitmap);
8483 clear_bit(shadow_read_write_fields[i], vmx_vmread_bitmap);
8485 /* shadowed read only fields */
8486 for (i = 0; i < max_shadow_read_only_fields; i++)
8487 clear_bit(shadow_read_only_fields[i], vmx_vmread_bitmap);
8490 * Allow direct access to the PC debug port (it is often used for I/O
8491 * delays, but the vmexits simply slow things down).
8493 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
8494 clear_bit(0x80, vmx_io_bitmap_a);
8496 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
8498 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
8499 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
8501 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
8503 r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
8504 __alignof__(struct vcpu_vmx), THIS_MODULE);
8509 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
8510 crash_vmclear_local_loaded_vmcss);
8513 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
8514 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
8515 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
8516 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
8517 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
8518 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
8519 memcpy(vmx_msr_bitmap_legacy_x2apic,
8520 vmx_msr_bitmap_legacy, PAGE_SIZE);
8521 memcpy(vmx_msr_bitmap_longmode_x2apic,
8522 vmx_msr_bitmap_longmode, PAGE_SIZE);
8525 for (msr = 0x800; msr <= 0x8ff; msr++)
8526 vmx_disable_intercept_msr_read_x2apic(msr);
8528 /* According SDM, in x2apic mode, the whole id reg is used.
8529 * But in KVM, it only use the highest eight bits. Need to
8531 vmx_enable_intercept_msr_read_x2apic(0x802);
8533 vmx_enable_intercept_msr_read_x2apic(0x839);
8535 vmx_disable_intercept_msr_write_x2apic(0x808);
8537 vmx_disable_intercept_msr_write_x2apic(0x80b);
8539 vmx_disable_intercept_msr_write_x2apic(0x83f);
8543 kvm_mmu_set_mask_ptes(0ull,
8544 (enable_ept_ad_bits) ? VMX_EPT_ACCESS_BIT : 0ull,
8545 (enable_ept_ad_bits) ? VMX_EPT_DIRTY_BIT : 0ull,
8546 0ull, VMX_EPT_EXECUTABLE_MASK);
8547 ept_set_mmio_spte_mask();
8555 free_page((unsigned long)vmx_vmwrite_bitmap);
8557 free_page((unsigned long)vmx_vmread_bitmap);
8559 free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
8561 free_page((unsigned long)vmx_msr_bitmap_longmode);
8563 free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
8565 free_page((unsigned long)vmx_msr_bitmap_legacy);
8567 free_page((unsigned long)vmx_io_bitmap_b);
8569 free_page((unsigned long)vmx_io_bitmap_a);
8573 static void __exit vmx_exit(void)
8575 free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
8576 free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
8577 free_page((unsigned long)vmx_msr_bitmap_legacy);
8578 free_page((unsigned long)vmx_msr_bitmap_longmode);
8579 free_page((unsigned long)vmx_io_bitmap_b);
8580 free_page((unsigned long)vmx_io_bitmap_a);
8581 free_page((unsigned long)vmx_vmwrite_bitmap);
8582 free_page((unsigned long)vmx_vmread_bitmap);
8585 rcu_assign_pointer(crash_vmclear_loaded_vmcss, NULL);
8592 module_init(vmx_init)
8593 module_exit(vmx_exit)