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.
24 #include <linux/kvm_host.h>
25 #include <linux/module.h>
26 #include <linux/kernel.h>
28 #include <linux/highmem.h>
29 #include <linux/sched.h>
30 #include <linux/moduleparam.h>
31 #include <linux/mod_devicetable.h>
32 #include <linux/trace_events.h>
33 #include <linux/slab.h>
34 #include <linux/tboot.h>
35 #include <linux/hrtimer.h>
36 #include "kvm_cache_regs.h"
43 #include <asm/virtext.h>
45 #include <asm/fpu/internal.h>
46 #include <asm/perf_event.h>
47 #include <asm/debugreg.h>
48 #include <asm/kexec.h>
50 #include <asm/irq_remapping.h>
55 #define __ex(x) __kvm_handle_fault_on_reboot(x)
56 #define __ex_clear(x, reg) \
57 ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
59 MODULE_AUTHOR("Qumranet");
60 MODULE_LICENSE("GPL");
62 static const struct x86_cpu_id vmx_cpu_id[] = {
63 X86_FEATURE_MATCH(X86_FEATURE_VMX),
66 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
68 static bool __read_mostly enable_vpid = 1;
69 module_param_named(vpid, enable_vpid, bool, 0444);
71 static bool __read_mostly flexpriority_enabled = 1;
72 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
74 static bool __read_mostly enable_ept = 1;
75 module_param_named(ept, enable_ept, bool, S_IRUGO);
77 static bool __read_mostly enable_unrestricted_guest = 1;
78 module_param_named(unrestricted_guest,
79 enable_unrestricted_guest, bool, S_IRUGO);
81 static bool __read_mostly enable_ept_ad_bits = 1;
82 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
84 static bool __read_mostly emulate_invalid_guest_state = true;
85 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
87 static bool __read_mostly vmm_exclusive = 1;
88 module_param(vmm_exclusive, bool, S_IRUGO);
90 static bool __read_mostly fasteoi = 1;
91 module_param(fasteoi, bool, S_IRUGO);
93 static bool __read_mostly enable_apicv = 1;
94 module_param(enable_apicv, bool, S_IRUGO);
96 static bool __read_mostly enable_shadow_vmcs = 1;
97 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
99 * If nested=1, nested virtualization is supported, i.e., guests may use
100 * VMX and be a hypervisor for its own guests. If nested=0, guests may not
101 * use VMX instructions.
103 static bool __read_mostly nested = 0;
104 module_param(nested, bool, S_IRUGO);
106 static u64 __read_mostly host_xss;
108 static bool __read_mostly enable_pml = 1;
109 module_param_named(pml, enable_pml, bool, S_IRUGO);
111 #define KVM_VMX_TSC_MULTIPLIER_MAX 0xffffffffffffffffULL
113 /* Guest_tsc -> host_tsc conversion requires 64-bit division. */
114 static int __read_mostly cpu_preemption_timer_multi;
115 static bool __read_mostly enable_preemption_timer = 1;
117 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
120 #define KVM_GUEST_CR0_MASK (X86_CR0_NW | X86_CR0_CD)
121 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST (X86_CR0_WP | X86_CR0_NE)
122 #define KVM_VM_CR0_ALWAYS_ON \
123 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
124 #define KVM_CR4_GUEST_OWNED_BITS \
125 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
126 | X86_CR4_OSXMMEXCPT | X86_CR4_TSD)
128 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
129 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
131 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
133 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
136 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
137 * ple_gap: upper bound on the amount of time between two successive
138 * executions of PAUSE in a loop. Also indicate if ple enabled.
139 * According to test, this time is usually smaller than 128 cycles.
140 * ple_window: upper bound on the amount of time a guest is allowed to execute
141 * in a PAUSE loop. Tests indicate that most spinlocks are held for
142 * less than 2^12 cycles
143 * Time is measured based on a counter that runs at the same rate as the TSC,
144 * refer SDM volume 3b section 21.6.13 & 22.1.3.
146 #define KVM_VMX_DEFAULT_PLE_GAP 128
147 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
148 #define KVM_VMX_DEFAULT_PLE_WINDOW_GROW 2
149 #define KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK 0
150 #define KVM_VMX_DEFAULT_PLE_WINDOW_MAX \
151 INT_MAX / KVM_VMX_DEFAULT_PLE_WINDOW_GROW
153 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
154 module_param(ple_gap, int, S_IRUGO);
156 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
157 module_param(ple_window, int, S_IRUGO);
159 /* Default doubles per-vcpu window every exit. */
160 static int ple_window_grow = KVM_VMX_DEFAULT_PLE_WINDOW_GROW;
161 module_param(ple_window_grow, int, S_IRUGO);
163 /* Default resets per-vcpu window every exit to ple_window. */
164 static int ple_window_shrink = KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK;
165 module_param(ple_window_shrink, int, S_IRUGO);
167 /* Default is to compute the maximum so we can never overflow. */
168 static int ple_window_actual_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
169 static int ple_window_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
170 module_param(ple_window_max, int, S_IRUGO);
172 extern const ulong vmx_return;
174 #define NR_AUTOLOAD_MSRS 8
175 #define VMCS02_POOL_SIZE 1
184 * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
185 * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
186 * loaded on this CPU (so we can clear them if the CPU goes down).
192 struct list_head loaded_vmcss_on_cpu_link;
195 struct shared_msr_entry {
202 * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
203 * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
204 * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
205 * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
206 * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
207 * More than one of these structures may exist, if L1 runs multiple L2 guests.
208 * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
209 * underlying hardware which will be used to run L2.
210 * This structure is packed to ensure that its layout is identical across
211 * machines (necessary for live migration).
212 * If there are changes in this struct, VMCS12_REVISION must be changed.
214 typedef u64 natural_width;
215 struct __packed vmcs12 {
216 /* According to the Intel spec, a VMCS region must start with the
217 * following two fields. Then follow implementation-specific data.
222 u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
223 u32 padding[7]; /* room for future expansion */
228 u64 vm_exit_msr_store_addr;
229 u64 vm_exit_msr_load_addr;
230 u64 vm_entry_msr_load_addr;
232 u64 virtual_apic_page_addr;
233 u64 apic_access_addr;
234 u64 posted_intr_desc_addr;
236 u64 eoi_exit_bitmap0;
237 u64 eoi_exit_bitmap1;
238 u64 eoi_exit_bitmap2;
239 u64 eoi_exit_bitmap3;
241 u64 guest_physical_address;
242 u64 vmcs_link_pointer;
243 u64 guest_ia32_debugctl;
246 u64 guest_ia32_perf_global_ctrl;
254 u64 host_ia32_perf_global_ctrl;
255 u64 padding64[8]; /* room for future expansion */
257 * To allow migration of L1 (complete with its L2 guests) between
258 * machines of different natural widths (32 or 64 bit), we cannot have
259 * unsigned long fields with no explict size. We use u64 (aliased
260 * natural_width) instead. Luckily, x86 is little-endian.
262 natural_width cr0_guest_host_mask;
263 natural_width cr4_guest_host_mask;
264 natural_width cr0_read_shadow;
265 natural_width cr4_read_shadow;
266 natural_width cr3_target_value0;
267 natural_width cr3_target_value1;
268 natural_width cr3_target_value2;
269 natural_width cr3_target_value3;
270 natural_width exit_qualification;
271 natural_width guest_linear_address;
272 natural_width guest_cr0;
273 natural_width guest_cr3;
274 natural_width guest_cr4;
275 natural_width guest_es_base;
276 natural_width guest_cs_base;
277 natural_width guest_ss_base;
278 natural_width guest_ds_base;
279 natural_width guest_fs_base;
280 natural_width guest_gs_base;
281 natural_width guest_ldtr_base;
282 natural_width guest_tr_base;
283 natural_width guest_gdtr_base;
284 natural_width guest_idtr_base;
285 natural_width guest_dr7;
286 natural_width guest_rsp;
287 natural_width guest_rip;
288 natural_width guest_rflags;
289 natural_width guest_pending_dbg_exceptions;
290 natural_width guest_sysenter_esp;
291 natural_width guest_sysenter_eip;
292 natural_width host_cr0;
293 natural_width host_cr3;
294 natural_width host_cr4;
295 natural_width host_fs_base;
296 natural_width host_gs_base;
297 natural_width host_tr_base;
298 natural_width host_gdtr_base;
299 natural_width host_idtr_base;
300 natural_width host_ia32_sysenter_esp;
301 natural_width host_ia32_sysenter_eip;
302 natural_width host_rsp;
303 natural_width host_rip;
304 natural_width paddingl[8]; /* room for future expansion */
305 u32 pin_based_vm_exec_control;
306 u32 cpu_based_vm_exec_control;
307 u32 exception_bitmap;
308 u32 page_fault_error_code_mask;
309 u32 page_fault_error_code_match;
310 u32 cr3_target_count;
311 u32 vm_exit_controls;
312 u32 vm_exit_msr_store_count;
313 u32 vm_exit_msr_load_count;
314 u32 vm_entry_controls;
315 u32 vm_entry_msr_load_count;
316 u32 vm_entry_intr_info_field;
317 u32 vm_entry_exception_error_code;
318 u32 vm_entry_instruction_len;
320 u32 secondary_vm_exec_control;
321 u32 vm_instruction_error;
323 u32 vm_exit_intr_info;
324 u32 vm_exit_intr_error_code;
325 u32 idt_vectoring_info_field;
326 u32 idt_vectoring_error_code;
327 u32 vm_exit_instruction_len;
328 u32 vmx_instruction_info;
335 u32 guest_ldtr_limit;
337 u32 guest_gdtr_limit;
338 u32 guest_idtr_limit;
339 u32 guest_es_ar_bytes;
340 u32 guest_cs_ar_bytes;
341 u32 guest_ss_ar_bytes;
342 u32 guest_ds_ar_bytes;
343 u32 guest_fs_ar_bytes;
344 u32 guest_gs_ar_bytes;
345 u32 guest_ldtr_ar_bytes;
346 u32 guest_tr_ar_bytes;
347 u32 guest_interruptibility_info;
348 u32 guest_activity_state;
349 u32 guest_sysenter_cs;
350 u32 host_ia32_sysenter_cs;
351 u32 vmx_preemption_timer_value;
352 u32 padding32[7]; /* room for future expansion */
353 u16 virtual_processor_id;
355 u16 guest_es_selector;
356 u16 guest_cs_selector;
357 u16 guest_ss_selector;
358 u16 guest_ds_selector;
359 u16 guest_fs_selector;
360 u16 guest_gs_selector;
361 u16 guest_ldtr_selector;
362 u16 guest_tr_selector;
363 u16 guest_intr_status;
364 u16 host_es_selector;
365 u16 host_cs_selector;
366 u16 host_ss_selector;
367 u16 host_ds_selector;
368 u16 host_fs_selector;
369 u16 host_gs_selector;
370 u16 host_tr_selector;
374 * VMCS12_REVISION is an arbitrary id that should be changed if the content or
375 * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
376 * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
378 #define VMCS12_REVISION 0x11e57ed0
381 * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
382 * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
383 * current implementation, 4K are reserved to avoid future complications.
385 #define VMCS12_SIZE 0x1000
387 /* Used to remember the last vmcs02 used for some recently used vmcs12s */
389 struct list_head list;
391 struct loaded_vmcs vmcs02;
395 * The nested_vmx structure is part of vcpu_vmx, and holds information we need
396 * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
399 /* Has the level1 guest done vmxon? */
403 /* The guest-physical address of the current VMCS L1 keeps for L2 */
405 /* The host-usable pointer to the above */
406 struct page *current_vmcs12_page;
407 struct vmcs12 *current_vmcs12;
408 struct vmcs *current_shadow_vmcs;
410 * Indicates if the shadow vmcs must be updated with the
411 * data hold by vmcs12
413 bool sync_shadow_vmcs;
415 /* vmcs02_list cache of VMCSs recently used to run L2 guests */
416 struct list_head vmcs02_pool;
418 u64 vmcs01_tsc_offset;
419 /* L2 must run next, and mustn't decide to exit to L1. */
420 bool nested_run_pending;
422 * Guest pages referred to in vmcs02 with host-physical pointers, so
423 * we must keep them pinned while L2 runs.
425 struct page *apic_access_page;
426 struct page *virtual_apic_page;
427 struct page *pi_desc_page;
428 struct pi_desc *pi_desc;
432 struct hrtimer preemption_timer;
433 bool preemption_timer_expired;
435 /* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
441 u32 nested_vmx_procbased_ctls_low;
442 u32 nested_vmx_procbased_ctls_high;
443 u32 nested_vmx_true_procbased_ctls_low;
444 u32 nested_vmx_secondary_ctls_low;
445 u32 nested_vmx_secondary_ctls_high;
446 u32 nested_vmx_pinbased_ctls_low;
447 u32 nested_vmx_pinbased_ctls_high;
448 u32 nested_vmx_exit_ctls_low;
449 u32 nested_vmx_exit_ctls_high;
450 u32 nested_vmx_true_exit_ctls_low;
451 u32 nested_vmx_entry_ctls_low;
452 u32 nested_vmx_entry_ctls_high;
453 u32 nested_vmx_true_entry_ctls_low;
454 u32 nested_vmx_misc_low;
455 u32 nested_vmx_misc_high;
456 u32 nested_vmx_ept_caps;
457 u32 nested_vmx_vpid_caps;
460 #define POSTED_INTR_ON 0
461 #define POSTED_INTR_SN 1
463 /* Posted-Interrupt Descriptor */
465 u32 pir[8]; /* Posted interrupt requested */
468 /* bit 256 - Outstanding Notification */
470 /* bit 257 - Suppress Notification */
472 /* bit 271:258 - Reserved */
474 /* bit 279:272 - Notification Vector */
476 /* bit 287:280 - Reserved */
478 /* bit 319:288 - Notification Destination */
486 static bool pi_test_and_set_on(struct pi_desc *pi_desc)
488 return test_and_set_bit(POSTED_INTR_ON,
489 (unsigned long *)&pi_desc->control);
492 static bool pi_test_and_clear_on(struct pi_desc *pi_desc)
494 return test_and_clear_bit(POSTED_INTR_ON,
495 (unsigned long *)&pi_desc->control);
498 static int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
500 return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
503 static inline void pi_clear_sn(struct pi_desc *pi_desc)
505 return clear_bit(POSTED_INTR_SN,
506 (unsigned long *)&pi_desc->control);
509 static inline void pi_set_sn(struct pi_desc *pi_desc)
511 return set_bit(POSTED_INTR_SN,
512 (unsigned long *)&pi_desc->control);
515 static inline int pi_test_on(struct pi_desc *pi_desc)
517 return test_bit(POSTED_INTR_ON,
518 (unsigned long *)&pi_desc->control);
521 static inline int pi_test_sn(struct pi_desc *pi_desc)
523 return test_bit(POSTED_INTR_SN,
524 (unsigned long *)&pi_desc->control);
528 struct kvm_vcpu vcpu;
529 unsigned long host_rsp;
531 bool nmi_known_unmasked;
533 u32 idt_vectoring_info;
535 struct shared_msr_entry *guest_msrs;
538 unsigned long host_idt_base;
540 u64 msr_host_kernel_gs_base;
541 u64 msr_guest_kernel_gs_base;
543 u32 vm_entry_controls_shadow;
544 u32 vm_exit_controls_shadow;
546 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
547 * non-nested (L1) guest, it always points to vmcs01. For a nested
548 * guest (L2), it points to a different VMCS.
550 struct loaded_vmcs vmcs01;
551 struct loaded_vmcs *loaded_vmcs;
552 bool __launched; /* temporary, used in vmx_vcpu_run */
553 struct msr_autoload {
555 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
556 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
560 u16 fs_sel, gs_sel, ldt_sel;
564 int gs_ldt_reload_needed;
565 int fs_reload_needed;
566 u64 msr_host_bndcfgs;
567 unsigned long vmcs_host_cr4; /* May not match real cr4 */
572 struct kvm_segment segs[8];
575 u32 bitmask; /* 4 bits per segment (1 bit per field) */
576 struct kvm_save_segment {
584 bool emulation_required;
586 /* Support for vnmi-less CPUs */
587 int soft_vnmi_blocked;
589 s64 vnmi_blocked_time;
592 /* Posted interrupt descriptor */
593 struct pi_desc pi_desc;
595 /* Support for a guest hypervisor (nested VMX) */
596 struct nested_vmx nested;
598 /* Dynamic PLE window. */
600 bool ple_window_dirty;
602 /* Support for PML */
603 #define PML_ENTITY_NUM 512
606 /* apic deadline value in host tsc */
609 u64 current_tsc_ratio;
611 bool guest_pkru_valid;
616 * Only bits masked by msr_ia32_feature_control_valid_bits can be set in
617 * msr_ia32_feature_control. FEATURE_CONTROL_LOCKED is always included
618 * in msr_ia32_feature_control_valid_bits.
620 u64 msr_ia32_feature_control;
621 u64 msr_ia32_feature_control_valid_bits;
624 enum segment_cache_field {
633 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
635 return container_of(vcpu, struct vcpu_vmx, vcpu);
638 static struct pi_desc *vcpu_to_pi_desc(struct kvm_vcpu *vcpu)
640 return &(to_vmx(vcpu)->pi_desc);
643 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
644 #define FIELD(number, name) [number] = VMCS12_OFFSET(name)
645 #define FIELD64(number, name) [number] = VMCS12_OFFSET(name), \
646 [number##_HIGH] = VMCS12_OFFSET(name)+4
649 static unsigned long shadow_read_only_fields[] = {
651 * We do NOT shadow fields that are modified when L0
652 * traps and emulates any vmx instruction (e.g. VMPTRLD,
653 * VMXON...) executed by L1.
654 * For example, VM_INSTRUCTION_ERROR is read
655 * by L1 if a vmx instruction fails (part of the error path).
656 * Note the code assumes this logic. If for some reason
657 * we start shadowing these fields then we need to
658 * force a shadow sync when L0 emulates vmx instructions
659 * (e.g. force a sync if VM_INSTRUCTION_ERROR is modified
660 * by nested_vmx_failValid)
664 VM_EXIT_INSTRUCTION_LEN,
665 IDT_VECTORING_INFO_FIELD,
666 IDT_VECTORING_ERROR_CODE,
667 VM_EXIT_INTR_ERROR_CODE,
669 GUEST_LINEAR_ADDRESS,
670 GUEST_PHYSICAL_ADDRESS
672 static int max_shadow_read_only_fields =
673 ARRAY_SIZE(shadow_read_only_fields);
675 static unsigned long shadow_read_write_fields[] = {
682 GUEST_INTERRUPTIBILITY_INFO,
695 CPU_BASED_VM_EXEC_CONTROL,
696 VM_ENTRY_EXCEPTION_ERROR_CODE,
697 VM_ENTRY_INTR_INFO_FIELD,
698 VM_ENTRY_INSTRUCTION_LEN,
699 VM_ENTRY_EXCEPTION_ERROR_CODE,
705 static int max_shadow_read_write_fields =
706 ARRAY_SIZE(shadow_read_write_fields);
708 static const unsigned short vmcs_field_to_offset_table[] = {
709 FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
710 FIELD(POSTED_INTR_NV, posted_intr_nv),
711 FIELD(GUEST_ES_SELECTOR, guest_es_selector),
712 FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
713 FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
714 FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
715 FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
716 FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
717 FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
718 FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
719 FIELD(GUEST_INTR_STATUS, guest_intr_status),
720 FIELD(HOST_ES_SELECTOR, host_es_selector),
721 FIELD(HOST_CS_SELECTOR, host_cs_selector),
722 FIELD(HOST_SS_SELECTOR, host_ss_selector),
723 FIELD(HOST_DS_SELECTOR, host_ds_selector),
724 FIELD(HOST_FS_SELECTOR, host_fs_selector),
725 FIELD(HOST_GS_SELECTOR, host_gs_selector),
726 FIELD(HOST_TR_SELECTOR, host_tr_selector),
727 FIELD64(IO_BITMAP_A, io_bitmap_a),
728 FIELD64(IO_BITMAP_B, io_bitmap_b),
729 FIELD64(MSR_BITMAP, msr_bitmap),
730 FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
731 FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
732 FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
733 FIELD64(TSC_OFFSET, tsc_offset),
734 FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
735 FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
736 FIELD64(POSTED_INTR_DESC_ADDR, posted_intr_desc_addr),
737 FIELD64(EPT_POINTER, ept_pointer),
738 FIELD64(EOI_EXIT_BITMAP0, eoi_exit_bitmap0),
739 FIELD64(EOI_EXIT_BITMAP1, eoi_exit_bitmap1),
740 FIELD64(EOI_EXIT_BITMAP2, eoi_exit_bitmap2),
741 FIELD64(EOI_EXIT_BITMAP3, eoi_exit_bitmap3),
742 FIELD64(XSS_EXIT_BITMAP, xss_exit_bitmap),
743 FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
744 FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
745 FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
746 FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
747 FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
748 FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
749 FIELD64(GUEST_PDPTR0, guest_pdptr0),
750 FIELD64(GUEST_PDPTR1, guest_pdptr1),
751 FIELD64(GUEST_PDPTR2, guest_pdptr2),
752 FIELD64(GUEST_PDPTR3, guest_pdptr3),
753 FIELD64(GUEST_BNDCFGS, guest_bndcfgs),
754 FIELD64(HOST_IA32_PAT, host_ia32_pat),
755 FIELD64(HOST_IA32_EFER, host_ia32_efer),
756 FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
757 FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
758 FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
759 FIELD(EXCEPTION_BITMAP, exception_bitmap),
760 FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
761 FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
762 FIELD(CR3_TARGET_COUNT, cr3_target_count),
763 FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
764 FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
765 FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
766 FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
767 FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
768 FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
769 FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
770 FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
771 FIELD(TPR_THRESHOLD, tpr_threshold),
772 FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
773 FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
774 FIELD(VM_EXIT_REASON, vm_exit_reason),
775 FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
776 FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
777 FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
778 FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
779 FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
780 FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
781 FIELD(GUEST_ES_LIMIT, guest_es_limit),
782 FIELD(GUEST_CS_LIMIT, guest_cs_limit),
783 FIELD(GUEST_SS_LIMIT, guest_ss_limit),
784 FIELD(GUEST_DS_LIMIT, guest_ds_limit),
785 FIELD(GUEST_FS_LIMIT, guest_fs_limit),
786 FIELD(GUEST_GS_LIMIT, guest_gs_limit),
787 FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
788 FIELD(GUEST_TR_LIMIT, guest_tr_limit),
789 FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
790 FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
791 FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
792 FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
793 FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
794 FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
795 FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
796 FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
797 FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
798 FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
799 FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
800 FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
801 FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
802 FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
803 FIELD(VMX_PREEMPTION_TIMER_VALUE, vmx_preemption_timer_value),
804 FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
805 FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
806 FIELD(CR0_READ_SHADOW, cr0_read_shadow),
807 FIELD(CR4_READ_SHADOW, cr4_read_shadow),
808 FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
809 FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
810 FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
811 FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
812 FIELD(EXIT_QUALIFICATION, exit_qualification),
813 FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
814 FIELD(GUEST_CR0, guest_cr0),
815 FIELD(GUEST_CR3, guest_cr3),
816 FIELD(GUEST_CR4, guest_cr4),
817 FIELD(GUEST_ES_BASE, guest_es_base),
818 FIELD(GUEST_CS_BASE, guest_cs_base),
819 FIELD(GUEST_SS_BASE, guest_ss_base),
820 FIELD(GUEST_DS_BASE, guest_ds_base),
821 FIELD(GUEST_FS_BASE, guest_fs_base),
822 FIELD(GUEST_GS_BASE, guest_gs_base),
823 FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
824 FIELD(GUEST_TR_BASE, guest_tr_base),
825 FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
826 FIELD(GUEST_IDTR_BASE, guest_idtr_base),
827 FIELD(GUEST_DR7, guest_dr7),
828 FIELD(GUEST_RSP, guest_rsp),
829 FIELD(GUEST_RIP, guest_rip),
830 FIELD(GUEST_RFLAGS, guest_rflags),
831 FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
832 FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
833 FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
834 FIELD(HOST_CR0, host_cr0),
835 FIELD(HOST_CR3, host_cr3),
836 FIELD(HOST_CR4, host_cr4),
837 FIELD(HOST_FS_BASE, host_fs_base),
838 FIELD(HOST_GS_BASE, host_gs_base),
839 FIELD(HOST_TR_BASE, host_tr_base),
840 FIELD(HOST_GDTR_BASE, host_gdtr_base),
841 FIELD(HOST_IDTR_BASE, host_idtr_base),
842 FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
843 FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
844 FIELD(HOST_RSP, host_rsp),
845 FIELD(HOST_RIP, host_rip),
848 static inline short vmcs_field_to_offset(unsigned long field)
850 BUILD_BUG_ON(ARRAY_SIZE(vmcs_field_to_offset_table) > SHRT_MAX);
852 if (field >= ARRAY_SIZE(vmcs_field_to_offset_table) ||
853 vmcs_field_to_offset_table[field] == 0)
856 return vmcs_field_to_offset_table[field];
859 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
861 return to_vmx(vcpu)->nested.current_vmcs12;
864 static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
866 struct page *page = kvm_vcpu_gfn_to_page(vcpu, addr >> PAGE_SHIFT);
867 if (is_error_page(page))
873 static void nested_release_page(struct page *page)
875 kvm_release_page_dirty(page);
878 static void nested_release_page_clean(struct page *page)
880 kvm_release_page_clean(page);
883 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu);
884 static u64 construct_eptp(unsigned long root_hpa);
885 static void kvm_cpu_vmxon(u64 addr);
886 static void kvm_cpu_vmxoff(void);
887 static bool vmx_xsaves_supported(void);
888 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
889 static void vmx_set_segment(struct kvm_vcpu *vcpu,
890 struct kvm_segment *var, int seg);
891 static void vmx_get_segment(struct kvm_vcpu *vcpu,
892 struct kvm_segment *var, int seg);
893 static bool guest_state_valid(struct kvm_vcpu *vcpu);
894 static u32 vmx_segment_access_rights(struct kvm_segment *var);
895 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx);
896 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx);
897 static int alloc_identity_pagetable(struct kvm *kvm);
899 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
900 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
902 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
903 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
905 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
906 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
909 * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
910 * can find which vCPU should be waken up.
912 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
913 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
915 static unsigned long *vmx_io_bitmap_a;
916 static unsigned long *vmx_io_bitmap_b;
917 static unsigned long *vmx_msr_bitmap_legacy;
918 static unsigned long *vmx_msr_bitmap_longmode;
919 static unsigned long *vmx_msr_bitmap_legacy_x2apic;
920 static unsigned long *vmx_msr_bitmap_longmode_x2apic;
921 static unsigned long *vmx_msr_bitmap_nested;
922 static unsigned long *vmx_vmread_bitmap;
923 static unsigned long *vmx_vmwrite_bitmap;
925 static bool cpu_has_load_ia32_efer;
926 static bool cpu_has_load_perf_global_ctrl;
928 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
929 static DEFINE_SPINLOCK(vmx_vpid_lock);
931 static struct vmcs_config {
935 u32 pin_based_exec_ctrl;
936 u32 cpu_based_exec_ctrl;
937 u32 cpu_based_2nd_exec_ctrl;
942 static struct vmx_capability {
947 #define VMX_SEGMENT_FIELD(seg) \
948 [VCPU_SREG_##seg] = { \
949 .selector = GUEST_##seg##_SELECTOR, \
950 .base = GUEST_##seg##_BASE, \
951 .limit = GUEST_##seg##_LIMIT, \
952 .ar_bytes = GUEST_##seg##_AR_BYTES, \
955 static const struct kvm_vmx_segment_field {
960 } kvm_vmx_segment_fields[] = {
961 VMX_SEGMENT_FIELD(CS),
962 VMX_SEGMENT_FIELD(DS),
963 VMX_SEGMENT_FIELD(ES),
964 VMX_SEGMENT_FIELD(FS),
965 VMX_SEGMENT_FIELD(GS),
966 VMX_SEGMENT_FIELD(SS),
967 VMX_SEGMENT_FIELD(TR),
968 VMX_SEGMENT_FIELD(LDTR),
971 static u64 host_efer;
973 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
976 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
977 * away by decrementing the array size.
979 static const u32 vmx_msr_index[] = {
981 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
983 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
986 static inline bool is_exception_n(u32 intr_info, u8 vector)
988 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
989 INTR_INFO_VALID_MASK)) ==
990 (INTR_TYPE_HARD_EXCEPTION | vector | INTR_INFO_VALID_MASK);
993 static inline bool is_debug(u32 intr_info)
995 return is_exception_n(intr_info, DB_VECTOR);
998 static inline bool is_breakpoint(u32 intr_info)
1000 return is_exception_n(intr_info, BP_VECTOR);
1003 static inline bool is_page_fault(u32 intr_info)
1005 return is_exception_n(intr_info, PF_VECTOR);
1008 static inline bool is_no_device(u32 intr_info)
1010 return is_exception_n(intr_info, NM_VECTOR);
1013 static inline bool is_invalid_opcode(u32 intr_info)
1015 return is_exception_n(intr_info, UD_VECTOR);
1018 static inline bool is_external_interrupt(u32 intr_info)
1020 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1021 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
1024 static inline bool is_machine_check(u32 intr_info)
1026 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
1027 INTR_INFO_VALID_MASK)) ==
1028 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
1031 static inline bool cpu_has_vmx_msr_bitmap(void)
1033 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
1036 static inline bool cpu_has_vmx_tpr_shadow(void)
1038 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
1041 static inline bool cpu_need_tpr_shadow(struct kvm_vcpu *vcpu)
1043 return cpu_has_vmx_tpr_shadow() && lapic_in_kernel(vcpu);
1046 static inline bool cpu_has_secondary_exec_ctrls(void)
1048 return vmcs_config.cpu_based_exec_ctrl &
1049 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1052 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
1054 return vmcs_config.cpu_based_2nd_exec_ctrl &
1055 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1058 static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
1060 return vmcs_config.cpu_based_2nd_exec_ctrl &
1061 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
1064 static inline bool cpu_has_vmx_apic_register_virt(void)
1066 return vmcs_config.cpu_based_2nd_exec_ctrl &
1067 SECONDARY_EXEC_APIC_REGISTER_VIRT;
1070 static inline bool cpu_has_vmx_virtual_intr_delivery(void)
1072 return vmcs_config.cpu_based_2nd_exec_ctrl &
1073 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
1077 * Comment's format: document - errata name - stepping - processor name.
1079 * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
1081 static u32 vmx_preemption_cpu_tfms[] = {
1082 /* 323344.pdf - BA86 - D0 - Xeon 7500 Series */
1084 /* 323056.pdf - AAX65 - C2 - Xeon L3406 */
1085 /* 322814.pdf - AAT59 - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
1086 /* 322911.pdf - AAU65 - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
1088 /* 322911.pdf - AAU65 - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
1090 /* 322373.pdf - AAO95 - B1 - Xeon 3400 Series */
1091 /* 322166.pdf - AAN92 - B1 - i7-800 and i5-700 Desktop */
1093 * 320767.pdf - AAP86 - B1 -
1094 * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
1097 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
1099 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
1101 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
1103 /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
1104 /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
1105 /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
1109 static inline bool cpu_has_broken_vmx_preemption_timer(void)
1111 u32 eax = cpuid_eax(0x00000001), i;
1113 /* Clear the reserved bits */
1114 eax &= ~(0x3U << 14 | 0xfU << 28);
1115 for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
1116 if (eax == vmx_preemption_cpu_tfms[i])
1122 static inline bool cpu_has_vmx_preemption_timer(void)
1124 return vmcs_config.pin_based_exec_ctrl &
1125 PIN_BASED_VMX_PREEMPTION_TIMER;
1128 static inline bool cpu_has_vmx_posted_intr(void)
1130 return IS_ENABLED(CONFIG_X86_LOCAL_APIC) &&
1131 vmcs_config.pin_based_exec_ctrl & PIN_BASED_POSTED_INTR;
1134 static inline bool cpu_has_vmx_apicv(void)
1136 return cpu_has_vmx_apic_register_virt() &&
1137 cpu_has_vmx_virtual_intr_delivery() &&
1138 cpu_has_vmx_posted_intr();
1141 static inline bool cpu_has_vmx_flexpriority(void)
1143 return cpu_has_vmx_tpr_shadow() &&
1144 cpu_has_vmx_virtualize_apic_accesses();
1147 static inline bool cpu_has_vmx_ept_execute_only(void)
1149 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
1152 static inline bool cpu_has_vmx_ept_2m_page(void)
1154 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
1157 static inline bool cpu_has_vmx_ept_1g_page(void)
1159 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
1162 static inline bool cpu_has_vmx_ept_4levels(void)
1164 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
1167 static inline bool cpu_has_vmx_ept_ad_bits(void)
1169 return vmx_capability.ept & VMX_EPT_AD_BIT;
1172 static inline bool cpu_has_vmx_invept_context(void)
1174 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
1177 static inline bool cpu_has_vmx_invept_global(void)
1179 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
1182 static inline bool cpu_has_vmx_invvpid_single(void)
1184 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
1187 static inline bool cpu_has_vmx_invvpid_global(void)
1189 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
1192 static inline bool cpu_has_vmx_ept(void)
1194 return vmcs_config.cpu_based_2nd_exec_ctrl &
1195 SECONDARY_EXEC_ENABLE_EPT;
1198 static inline bool cpu_has_vmx_unrestricted_guest(void)
1200 return vmcs_config.cpu_based_2nd_exec_ctrl &
1201 SECONDARY_EXEC_UNRESTRICTED_GUEST;
1204 static inline bool cpu_has_vmx_ple(void)
1206 return vmcs_config.cpu_based_2nd_exec_ctrl &
1207 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
1210 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
1212 return flexpriority_enabled && lapic_in_kernel(vcpu);
1215 static inline bool cpu_has_vmx_vpid(void)
1217 return vmcs_config.cpu_based_2nd_exec_ctrl &
1218 SECONDARY_EXEC_ENABLE_VPID;
1221 static inline bool cpu_has_vmx_rdtscp(void)
1223 return vmcs_config.cpu_based_2nd_exec_ctrl &
1224 SECONDARY_EXEC_RDTSCP;
1227 static inline bool cpu_has_vmx_invpcid(void)
1229 return vmcs_config.cpu_based_2nd_exec_ctrl &
1230 SECONDARY_EXEC_ENABLE_INVPCID;
1233 static inline bool cpu_has_virtual_nmis(void)
1235 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
1238 static inline bool cpu_has_vmx_wbinvd_exit(void)
1240 return vmcs_config.cpu_based_2nd_exec_ctrl &
1241 SECONDARY_EXEC_WBINVD_EXITING;
1244 static inline bool cpu_has_vmx_shadow_vmcs(void)
1247 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
1248 /* check if the cpu supports writing r/o exit information fields */
1249 if (!(vmx_msr & MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS))
1252 return vmcs_config.cpu_based_2nd_exec_ctrl &
1253 SECONDARY_EXEC_SHADOW_VMCS;
1256 static inline bool cpu_has_vmx_pml(void)
1258 return vmcs_config.cpu_based_2nd_exec_ctrl & SECONDARY_EXEC_ENABLE_PML;
1261 static inline bool cpu_has_vmx_tsc_scaling(void)
1263 return vmcs_config.cpu_based_2nd_exec_ctrl &
1264 SECONDARY_EXEC_TSC_SCALING;
1267 static inline bool report_flexpriority(void)
1269 return flexpriority_enabled;
1272 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
1274 return vmcs12->cpu_based_vm_exec_control & bit;
1277 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
1279 return (vmcs12->cpu_based_vm_exec_control &
1280 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
1281 (vmcs12->secondary_vm_exec_control & bit);
1284 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12)
1286 return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
1289 static inline bool nested_cpu_has_preemption_timer(struct vmcs12 *vmcs12)
1291 return vmcs12->pin_based_vm_exec_control &
1292 PIN_BASED_VMX_PREEMPTION_TIMER;
1295 static inline int nested_cpu_has_ept(struct vmcs12 *vmcs12)
1297 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_EPT);
1300 static inline bool nested_cpu_has_xsaves(struct vmcs12 *vmcs12)
1302 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES) &&
1303 vmx_xsaves_supported();
1306 static inline bool nested_cpu_has_virt_x2apic_mode(struct vmcs12 *vmcs12)
1308 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
1311 static inline bool nested_cpu_has_vpid(struct vmcs12 *vmcs12)
1313 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_VPID);
1316 static inline bool nested_cpu_has_apic_reg_virt(struct vmcs12 *vmcs12)
1318 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_APIC_REGISTER_VIRT);
1321 static inline bool nested_cpu_has_vid(struct vmcs12 *vmcs12)
1323 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
1326 static inline bool nested_cpu_has_posted_intr(struct vmcs12 *vmcs12)
1328 return vmcs12->pin_based_vm_exec_control & PIN_BASED_POSTED_INTR;
1331 static inline bool is_exception(u32 intr_info)
1333 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1334 == (INTR_TYPE_HARD_EXCEPTION | INTR_INFO_VALID_MASK);
1337 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
1339 unsigned long exit_qualification);
1340 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
1341 struct vmcs12 *vmcs12,
1342 u32 reason, unsigned long qualification);
1344 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
1348 for (i = 0; i < vmx->nmsrs; ++i)
1349 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
1354 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
1360 } operand = { vpid, 0, gva };
1362 asm volatile (__ex(ASM_VMX_INVVPID)
1363 /* CF==1 or ZF==1 --> rc = -1 */
1364 "; ja 1f ; ud2 ; 1:"
1365 : : "a"(&operand), "c"(ext) : "cc", "memory");
1368 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
1372 } operand = {eptp, gpa};
1374 asm volatile (__ex(ASM_VMX_INVEPT)
1375 /* CF==1 or ZF==1 --> rc = -1 */
1376 "; ja 1f ; ud2 ; 1:\n"
1377 : : "a" (&operand), "c" (ext) : "cc", "memory");
1380 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
1384 i = __find_msr_index(vmx, msr);
1386 return &vmx->guest_msrs[i];
1390 static void vmcs_clear(struct vmcs *vmcs)
1392 u64 phys_addr = __pa(vmcs);
1395 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
1396 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1399 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
1403 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
1405 vmcs_clear(loaded_vmcs->vmcs);
1406 loaded_vmcs->cpu = -1;
1407 loaded_vmcs->launched = 0;
1410 static void vmcs_load(struct vmcs *vmcs)
1412 u64 phys_addr = __pa(vmcs);
1415 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
1416 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1419 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
1423 #ifdef CONFIG_KEXEC_CORE
1425 * This bitmap is used to indicate whether the vmclear
1426 * operation is enabled on all cpus. All disabled by
1429 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
1431 static inline void crash_enable_local_vmclear(int cpu)
1433 cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
1436 static inline void crash_disable_local_vmclear(int cpu)
1438 cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
1441 static inline int crash_local_vmclear_enabled(int cpu)
1443 return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
1446 static void crash_vmclear_local_loaded_vmcss(void)
1448 int cpu = raw_smp_processor_id();
1449 struct loaded_vmcs *v;
1451 if (!crash_local_vmclear_enabled(cpu))
1454 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
1455 loaded_vmcss_on_cpu_link)
1456 vmcs_clear(v->vmcs);
1459 static inline void crash_enable_local_vmclear(int cpu) { }
1460 static inline void crash_disable_local_vmclear(int cpu) { }
1461 #endif /* CONFIG_KEXEC_CORE */
1463 static void __loaded_vmcs_clear(void *arg)
1465 struct loaded_vmcs *loaded_vmcs = arg;
1466 int cpu = raw_smp_processor_id();
1468 if (loaded_vmcs->cpu != cpu)
1469 return; /* vcpu migration can race with cpu offline */
1470 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1471 per_cpu(current_vmcs, cpu) = NULL;
1472 crash_disable_local_vmclear(cpu);
1473 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1476 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1477 * is before setting loaded_vmcs->vcpu to -1 which is done in
1478 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1479 * then adds the vmcs into percpu list before it is deleted.
1483 loaded_vmcs_init(loaded_vmcs);
1484 crash_enable_local_vmclear(cpu);
1487 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1489 int cpu = loaded_vmcs->cpu;
1492 smp_call_function_single(cpu,
1493 __loaded_vmcs_clear, loaded_vmcs, 1);
1496 static inline void vpid_sync_vcpu_single(int vpid)
1501 if (cpu_has_vmx_invvpid_single())
1502 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vpid, 0);
1505 static inline void vpid_sync_vcpu_global(void)
1507 if (cpu_has_vmx_invvpid_global())
1508 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1511 static inline void vpid_sync_context(int vpid)
1513 if (cpu_has_vmx_invvpid_single())
1514 vpid_sync_vcpu_single(vpid);
1516 vpid_sync_vcpu_global();
1519 static inline void ept_sync_global(void)
1521 if (cpu_has_vmx_invept_global())
1522 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1525 static inline void ept_sync_context(u64 eptp)
1528 if (cpu_has_vmx_invept_context())
1529 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1535 static __always_inline void vmcs_check16(unsigned long field)
1537 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1538 "16-bit accessor invalid for 64-bit field");
1539 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1540 "16-bit accessor invalid for 64-bit high field");
1541 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1542 "16-bit accessor invalid for 32-bit high field");
1543 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1544 "16-bit accessor invalid for natural width field");
1547 static __always_inline void vmcs_check32(unsigned long field)
1549 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1550 "32-bit accessor invalid for 16-bit field");
1551 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1552 "32-bit accessor invalid for natural width field");
1555 static __always_inline void vmcs_check64(unsigned long field)
1557 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1558 "64-bit accessor invalid for 16-bit field");
1559 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1560 "64-bit accessor invalid for 64-bit high field");
1561 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1562 "64-bit accessor invalid for 32-bit field");
1563 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1564 "64-bit accessor invalid for natural width field");
1567 static __always_inline void vmcs_checkl(unsigned long field)
1569 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1570 "Natural width accessor invalid for 16-bit field");
1571 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1572 "Natural width accessor invalid for 64-bit field");
1573 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1574 "Natural width accessor invalid for 64-bit high field");
1575 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1576 "Natural width accessor invalid for 32-bit field");
1579 static __always_inline unsigned long __vmcs_readl(unsigned long field)
1581 unsigned long value;
1583 asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1584 : "=a"(value) : "d"(field) : "cc");
1588 static __always_inline u16 vmcs_read16(unsigned long field)
1590 vmcs_check16(field);
1591 return __vmcs_readl(field);
1594 static __always_inline u32 vmcs_read32(unsigned long field)
1596 vmcs_check32(field);
1597 return __vmcs_readl(field);
1600 static __always_inline u64 vmcs_read64(unsigned long field)
1602 vmcs_check64(field);
1603 #ifdef CONFIG_X86_64
1604 return __vmcs_readl(field);
1606 return __vmcs_readl(field) | ((u64)__vmcs_readl(field+1) << 32);
1610 static __always_inline unsigned long vmcs_readl(unsigned long field)
1613 return __vmcs_readl(field);
1616 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1618 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1619 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1623 static __always_inline void __vmcs_writel(unsigned long field, unsigned long value)
1627 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1628 : "=q"(error) : "a"(value), "d"(field) : "cc");
1629 if (unlikely(error))
1630 vmwrite_error(field, value);
1633 static __always_inline void vmcs_write16(unsigned long field, u16 value)
1635 vmcs_check16(field);
1636 __vmcs_writel(field, value);
1639 static __always_inline void vmcs_write32(unsigned long field, u32 value)
1641 vmcs_check32(field);
1642 __vmcs_writel(field, value);
1645 static __always_inline void vmcs_write64(unsigned long field, u64 value)
1647 vmcs_check64(field);
1648 __vmcs_writel(field, value);
1649 #ifndef CONFIG_X86_64
1651 __vmcs_writel(field+1, value >> 32);
1655 static __always_inline void vmcs_writel(unsigned long field, unsigned long value)
1658 __vmcs_writel(field, value);
1661 static __always_inline void vmcs_clear_bits(unsigned long field, u32 mask)
1663 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1664 "vmcs_clear_bits does not support 64-bit fields");
1665 __vmcs_writel(field, __vmcs_readl(field) & ~mask);
1668 static __always_inline void vmcs_set_bits(unsigned long field, u32 mask)
1670 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1671 "vmcs_set_bits does not support 64-bit fields");
1672 __vmcs_writel(field, __vmcs_readl(field) | mask);
1675 static inline void vm_entry_controls_reset_shadow(struct vcpu_vmx *vmx)
1677 vmx->vm_entry_controls_shadow = vmcs_read32(VM_ENTRY_CONTROLS);
1680 static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val)
1682 vmcs_write32(VM_ENTRY_CONTROLS, val);
1683 vmx->vm_entry_controls_shadow = val;
1686 static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val)
1688 if (vmx->vm_entry_controls_shadow != val)
1689 vm_entry_controls_init(vmx, val);
1692 static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx)
1694 return vmx->vm_entry_controls_shadow;
1698 static inline void vm_entry_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1700 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) | val);
1703 static inline void vm_entry_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1705 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) & ~val);
1708 static inline void vm_exit_controls_reset_shadow(struct vcpu_vmx *vmx)
1710 vmx->vm_exit_controls_shadow = vmcs_read32(VM_EXIT_CONTROLS);
1713 static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val)
1715 vmcs_write32(VM_EXIT_CONTROLS, val);
1716 vmx->vm_exit_controls_shadow = val;
1719 static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val)
1721 if (vmx->vm_exit_controls_shadow != val)
1722 vm_exit_controls_init(vmx, val);
1725 static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx)
1727 return vmx->vm_exit_controls_shadow;
1731 static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1733 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val);
1736 static inline void vm_exit_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1738 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) & ~val);
1741 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1743 vmx->segment_cache.bitmask = 0;
1746 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1750 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1752 if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1753 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1754 vmx->segment_cache.bitmask = 0;
1756 ret = vmx->segment_cache.bitmask & mask;
1757 vmx->segment_cache.bitmask |= mask;
1761 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1763 u16 *p = &vmx->segment_cache.seg[seg].selector;
1765 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1766 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1770 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1772 ulong *p = &vmx->segment_cache.seg[seg].base;
1774 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1775 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1779 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1781 u32 *p = &vmx->segment_cache.seg[seg].limit;
1783 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1784 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1788 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1790 u32 *p = &vmx->segment_cache.seg[seg].ar;
1792 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1793 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1797 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1801 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1802 (1u << NM_VECTOR) | (1u << DB_VECTOR) | (1u << AC_VECTOR);
1803 if ((vcpu->guest_debug &
1804 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1805 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1806 eb |= 1u << BP_VECTOR;
1807 if (to_vmx(vcpu)->rmode.vm86_active)
1810 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1811 if (vcpu->fpu_active)
1812 eb &= ~(1u << NM_VECTOR);
1814 /* When we are running a nested L2 guest and L1 specified for it a
1815 * certain exception bitmap, we must trap the same exceptions and pass
1816 * them to L1. When running L2, we will only handle the exceptions
1817 * specified above if L1 did not want them.
1819 if (is_guest_mode(vcpu))
1820 eb |= get_vmcs12(vcpu)->exception_bitmap;
1822 vmcs_write32(EXCEPTION_BITMAP, eb);
1825 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1826 unsigned long entry, unsigned long exit)
1828 vm_entry_controls_clearbit(vmx, entry);
1829 vm_exit_controls_clearbit(vmx, exit);
1832 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1835 struct msr_autoload *m = &vmx->msr_autoload;
1839 if (cpu_has_load_ia32_efer) {
1840 clear_atomic_switch_msr_special(vmx,
1841 VM_ENTRY_LOAD_IA32_EFER,
1842 VM_EXIT_LOAD_IA32_EFER);
1846 case MSR_CORE_PERF_GLOBAL_CTRL:
1847 if (cpu_has_load_perf_global_ctrl) {
1848 clear_atomic_switch_msr_special(vmx,
1849 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1850 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1856 for (i = 0; i < m->nr; ++i)
1857 if (m->guest[i].index == msr)
1863 m->guest[i] = m->guest[m->nr];
1864 m->host[i] = m->host[m->nr];
1865 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1866 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1869 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1870 unsigned long entry, unsigned long exit,
1871 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
1872 u64 guest_val, u64 host_val)
1874 vmcs_write64(guest_val_vmcs, guest_val);
1875 vmcs_write64(host_val_vmcs, host_val);
1876 vm_entry_controls_setbit(vmx, entry);
1877 vm_exit_controls_setbit(vmx, exit);
1880 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1881 u64 guest_val, u64 host_val)
1884 struct msr_autoload *m = &vmx->msr_autoload;
1888 if (cpu_has_load_ia32_efer) {
1889 add_atomic_switch_msr_special(vmx,
1890 VM_ENTRY_LOAD_IA32_EFER,
1891 VM_EXIT_LOAD_IA32_EFER,
1894 guest_val, host_val);
1898 case MSR_CORE_PERF_GLOBAL_CTRL:
1899 if (cpu_has_load_perf_global_ctrl) {
1900 add_atomic_switch_msr_special(vmx,
1901 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1902 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1903 GUEST_IA32_PERF_GLOBAL_CTRL,
1904 HOST_IA32_PERF_GLOBAL_CTRL,
1905 guest_val, host_val);
1909 case MSR_IA32_PEBS_ENABLE:
1910 /* PEBS needs a quiescent period after being disabled (to write
1911 * a record). Disabling PEBS through VMX MSR swapping doesn't
1912 * provide that period, so a CPU could write host's record into
1915 wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
1918 for (i = 0; i < m->nr; ++i)
1919 if (m->guest[i].index == msr)
1922 if (i == NR_AUTOLOAD_MSRS) {
1923 printk_once(KERN_WARNING "Not enough msr switch entries. "
1924 "Can't add msr %x\n", msr);
1926 } else if (i == m->nr) {
1928 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1929 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1932 m->guest[i].index = msr;
1933 m->guest[i].value = guest_val;
1934 m->host[i].index = msr;
1935 m->host[i].value = host_val;
1938 static void reload_tss(void)
1941 * VT restores TR but not its size. Useless.
1943 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
1944 struct desc_struct *descs;
1946 descs = (void *)gdt->address;
1947 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
1951 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
1953 u64 guest_efer = vmx->vcpu.arch.efer;
1954 u64 ignore_bits = 0;
1958 * NX is needed to handle CR0.WP=1, CR4.SMEP=1. Testing
1959 * host CPUID is more efficient than testing guest CPUID
1960 * or CR4. Host SMEP is anyway a requirement for guest SMEP.
1962 if (boot_cpu_has(X86_FEATURE_SMEP))
1963 guest_efer |= EFER_NX;
1964 else if (!(guest_efer & EFER_NX))
1965 ignore_bits |= EFER_NX;
1969 * LMA and LME handled by hardware; SCE meaningless outside long mode.
1971 ignore_bits |= EFER_SCE;
1972 #ifdef CONFIG_X86_64
1973 ignore_bits |= EFER_LMA | EFER_LME;
1974 /* SCE is meaningful only in long mode on Intel */
1975 if (guest_efer & EFER_LMA)
1976 ignore_bits &= ~(u64)EFER_SCE;
1979 clear_atomic_switch_msr(vmx, MSR_EFER);
1982 * On EPT, we can't emulate NX, so we must switch EFER atomically.
1983 * On CPUs that support "load IA32_EFER", always switch EFER
1984 * atomically, since it's faster than switching it manually.
1986 if (cpu_has_load_ia32_efer ||
1987 (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
1988 if (!(guest_efer & EFER_LMA))
1989 guest_efer &= ~EFER_LME;
1990 if (guest_efer != host_efer)
1991 add_atomic_switch_msr(vmx, MSR_EFER,
1992 guest_efer, host_efer);
1995 guest_efer &= ~ignore_bits;
1996 guest_efer |= host_efer & ignore_bits;
1998 vmx->guest_msrs[efer_offset].data = guest_efer;
1999 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
2005 static unsigned long segment_base(u16 selector)
2007 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
2008 struct desc_struct *d;
2009 unsigned long table_base;
2012 if (!(selector & ~3))
2015 table_base = gdt->address;
2017 if (selector & 4) { /* from ldt */
2018 u16 ldt_selector = kvm_read_ldt();
2020 if (!(ldt_selector & ~3))
2023 table_base = segment_base(ldt_selector);
2025 d = (struct desc_struct *)(table_base + (selector & ~7));
2026 v = get_desc_base(d);
2027 #ifdef CONFIG_X86_64
2028 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
2029 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
2034 static inline unsigned long kvm_read_tr_base(void)
2037 asm("str %0" : "=g"(tr));
2038 return segment_base(tr);
2041 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
2043 struct vcpu_vmx *vmx = to_vmx(vcpu);
2046 if (vmx->host_state.loaded)
2049 vmx->host_state.loaded = 1;
2051 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
2052 * allow segment selectors with cpl > 0 or ti == 1.
2054 vmx->host_state.ldt_sel = kvm_read_ldt();
2055 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
2056 savesegment(fs, vmx->host_state.fs_sel);
2057 if (!(vmx->host_state.fs_sel & 7)) {
2058 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
2059 vmx->host_state.fs_reload_needed = 0;
2061 vmcs_write16(HOST_FS_SELECTOR, 0);
2062 vmx->host_state.fs_reload_needed = 1;
2064 savesegment(gs, vmx->host_state.gs_sel);
2065 if (!(vmx->host_state.gs_sel & 7))
2066 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
2068 vmcs_write16(HOST_GS_SELECTOR, 0);
2069 vmx->host_state.gs_ldt_reload_needed = 1;
2072 #ifdef CONFIG_X86_64
2073 savesegment(ds, vmx->host_state.ds_sel);
2074 savesegment(es, vmx->host_state.es_sel);
2077 #ifdef CONFIG_X86_64
2078 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
2079 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
2081 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
2082 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
2085 #ifdef CONFIG_X86_64
2086 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2087 if (is_long_mode(&vmx->vcpu))
2088 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2090 if (boot_cpu_has(X86_FEATURE_MPX))
2091 rdmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
2092 for (i = 0; i < vmx->save_nmsrs; ++i)
2093 kvm_set_shared_msr(vmx->guest_msrs[i].index,
2094 vmx->guest_msrs[i].data,
2095 vmx->guest_msrs[i].mask);
2098 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
2100 if (!vmx->host_state.loaded)
2103 ++vmx->vcpu.stat.host_state_reload;
2104 vmx->host_state.loaded = 0;
2105 #ifdef CONFIG_X86_64
2106 if (is_long_mode(&vmx->vcpu))
2107 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2109 if (vmx->host_state.gs_ldt_reload_needed) {
2110 kvm_load_ldt(vmx->host_state.ldt_sel);
2111 #ifdef CONFIG_X86_64
2112 load_gs_index(vmx->host_state.gs_sel);
2114 loadsegment(gs, vmx->host_state.gs_sel);
2117 if (vmx->host_state.fs_reload_needed)
2118 loadsegment(fs, vmx->host_state.fs_sel);
2119 #ifdef CONFIG_X86_64
2120 if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
2121 loadsegment(ds, vmx->host_state.ds_sel);
2122 loadsegment(es, vmx->host_state.es_sel);
2126 #ifdef CONFIG_X86_64
2127 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2129 if (vmx->host_state.msr_host_bndcfgs)
2130 wrmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
2132 * If the FPU is not active (through the host task or
2133 * the guest vcpu), then restore the cr0.TS bit.
2135 if (!fpregs_active() && !vmx->vcpu.guest_fpu_loaded)
2137 load_gdt(this_cpu_ptr(&host_gdt));
2140 static void vmx_load_host_state(struct vcpu_vmx *vmx)
2143 __vmx_load_host_state(vmx);
2147 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
2149 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2150 struct pi_desc old, new;
2153 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
2154 !irq_remapping_cap(IRQ_POSTING_CAP))
2158 old.control = new.control = pi_desc->control;
2161 * If 'nv' field is POSTED_INTR_WAKEUP_VECTOR, there
2162 * are two possible cases:
2163 * 1. After running 'pre_block', context switch
2164 * happened. For this case, 'sn' was set in
2165 * vmx_vcpu_put(), so we need to clear it here.
2166 * 2. After running 'pre_block', we were blocked,
2167 * and woken up by some other guy. For this case,
2168 * we don't need to do anything, 'pi_post_block'
2169 * will do everything for us. However, we cannot
2170 * check whether it is case #1 or case #2 here
2171 * (maybe, not needed), so we also clear sn here,
2172 * I think it is not a big deal.
2174 if (pi_desc->nv != POSTED_INTR_WAKEUP_VECTOR) {
2175 if (vcpu->cpu != cpu) {
2176 dest = cpu_physical_id(cpu);
2178 if (x2apic_enabled())
2181 new.ndst = (dest << 8) & 0xFF00;
2184 /* set 'NV' to 'notification vector' */
2185 new.nv = POSTED_INTR_VECTOR;
2188 /* Allow posting non-urgent interrupts */
2190 } while (cmpxchg(&pi_desc->control, old.control,
2191 new.control) != old.control);
2195 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
2196 * vcpu mutex is already taken.
2198 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2200 struct vcpu_vmx *vmx = to_vmx(vcpu);
2201 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2204 kvm_cpu_vmxon(phys_addr);
2205 else if (vmx->loaded_vmcs->cpu != cpu)
2206 loaded_vmcs_clear(vmx->loaded_vmcs);
2208 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
2209 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
2210 vmcs_load(vmx->loaded_vmcs->vmcs);
2213 if (vmx->loaded_vmcs->cpu != cpu) {
2214 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
2215 unsigned long sysenter_esp;
2217 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2218 local_irq_disable();
2219 crash_disable_local_vmclear(cpu);
2222 * Read loaded_vmcs->cpu should be before fetching
2223 * loaded_vmcs->loaded_vmcss_on_cpu_link.
2224 * See the comments in __loaded_vmcs_clear().
2228 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
2229 &per_cpu(loaded_vmcss_on_cpu, cpu));
2230 crash_enable_local_vmclear(cpu);
2234 * Linux uses per-cpu TSS and GDT, so set these when switching
2237 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
2238 vmcs_writel(HOST_GDTR_BASE, gdt->address); /* 22.2.4 */
2240 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
2241 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
2243 vmx->loaded_vmcs->cpu = cpu;
2246 /* Setup TSC multiplier */
2247 if (kvm_has_tsc_control &&
2248 vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio) {
2249 vmx->current_tsc_ratio = vcpu->arch.tsc_scaling_ratio;
2250 vmcs_write64(TSC_MULTIPLIER, vmx->current_tsc_ratio);
2253 vmx_vcpu_pi_load(vcpu, cpu);
2254 vmx->host_pkru = read_pkru();
2257 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
2259 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2261 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
2262 !irq_remapping_cap(IRQ_POSTING_CAP))
2265 /* Set SN when the vCPU is preempted */
2266 if (vcpu->preempted)
2270 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
2272 vmx_vcpu_pi_put(vcpu);
2274 __vmx_load_host_state(to_vmx(vcpu));
2275 if (!vmm_exclusive) {
2276 __loaded_vmcs_clear(to_vmx(vcpu)->loaded_vmcs);
2282 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
2286 if (vcpu->fpu_active)
2288 vcpu->fpu_active = 1;
2289 cr0 = vmcs_readl(GUEST_CR0);
2290 cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
2291 cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
2292 vmcs_writel(GUEST_CR0, cr0);
2293 update_exception_bitmap(vcpu);
2294 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
2295 if (is_guest_mode(vcpu))
2296 vcpu->arch.cr0_guest_owned_bits &=
2297 ~get_vmcs12(vcpu)->cr0_guest_host_mask;
2298 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2301 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
2304 * Return the cr0 value that a nested guest would read. This is a combination
2305 * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
2306 * its hypervisor (cr0_read_shadow).
2308 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
2310 return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
2311 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
2313 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
2315 return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
2316 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
2319 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
2321 /* Note that there is no vcpu->fpu_active = 0 here. The caller must
2322 * set this *before* calling this function.
2324 vmx_decache_cr0_guest_bits(vcpu);
2325 vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
2326 update_exception_bitmap(vcpu);
2327 vcpu->arch.cr0_guest_owned_bits = 0;
2328 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2329 if (is_guest_mode(vcpu)) {
2331 * L1's specified read shadow might not contain the TS bit,
2332 * so now that we turned on shadowing of this bit, we need to
2333 * set this bit of the shadow. Like in nested_vmx_run we need
2334 * nested_read_cr0(vmcs12), but vmcs12->guest_cr0 is not yet
2335 * up-to-date here because we just decached cr0.TS (and we'll
2336 * only update vmcs12->guest_cr0 on nested exit).
2338 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2339 vmcs12->guest_cr0 = (vmcs12->guest_cr0 & ~X86_CR0_TS) |
2340 (vcpu->arch.cr0 & X86_CR0_TS);
2341 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2343 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
2346 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
2348 unsigned long rflags, save_rflags;
2350 if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
2351 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2352 rflags = vmcs_readl(GUEST_RFLAGS);
2353 if (to_vmx(vcpu)->rmode.vm86_active) {
2354 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2355 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
2356 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2358 to_vmx(vcpu)->rflags = rflags;
2360 return to_vmx(vcpu)->rflags;
2363 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
2365 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2366 to_vmx(vcpu)->rflags = rflags;
2367 if (to_vmx(vcpu)->rmode.vm86_active) {
2368 to_vmx(vcpu)->rmode.save_rflags = rflags;
2369 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2371 vmcs_writel(GUEST_RFLAGS, rflags);
2374 static u32 vmx_get_pkru(struct kvm_vcpu *vcpu)
2376 return to_vmx(vcpu)->guest_pkru;
2379 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
2381 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2384 if (interruptibility & GUEST_INTR_STATE_STI)
2385 ret |= KVM_X86_SHADOW_INT_STI;
2386 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
2387 ret |= KVM_X86_SHADOW_INT_MOV_SS;
2392 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
2394 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2395 u32 interruptibility = interruptibility_old;
2397 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
2399 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
2400 interruptibility |= GUEST_INTR_STATE_MOV_SS;
2401 else if (mask & KVM_X86_SHADOW_INT_STI)
2402 interruptibility |= GUEST_INTR_STATE_STI;
2404 if ((interruptibility != interruptibility_old))
2405 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
2408 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
2412 rip = kvm_rip_read(vcpu);
2413 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
2414 kvm_rip_write(vcpu, rip);
2416 /* skipping an emulated instruction also counts */
2417 vmx_set_interrupt_shadow(vcpu, 0);
2421 * KVM wants to inject page-faults which it got to the guest. This function
2422 * checks whether in a nested guest, we need to inject them to L1 or L2.
2424 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned nr)
2426 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2428 if (!(vmcs12->exception_bitmap & (1u << nr)))
2431 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
2432 vmcs_read32(VM_EXIT_INTR_INFO),
2433 vmcs_readl(EXIT_QUALIFICATION));
2437 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
2438 bool has_error_code, u32 error_code,
2441 struct vcpu_vmx *vmx = to_vmx(vcpu);
2442 u32 intr_info = nr | INTR_INFO_VALID_MASK;
2444 if (!reinject && is_guest_mode(vcpu) &&
2445 nested_vmx_check_exception(vcpu, nr))
2448 if (has_error_code) {
2449 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
2450 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2453 if (vmx->rmode.vm86_active) {
2455 if (kvm_exception_is_soft(nr))
2456 inc_eip = vcpu->arch.event_exit_inst_len;
2457 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
2458 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2462 if (kvm_exception_is_soft(nr)) {
2463 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2464 vmx->vcpu.arch.event_exit_inst_len);
2465 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2467 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2469 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
2472 static bool vmx_rdtscp_supported(void)
2474 return cpu_has_vmx_rdtscp();
2477 static bool vmx_invpcid_supported(void)
2479 return cpu_has_vmx_invpcid() && enable_ept;
2483 * Swap MSR entry in host/guest MSR entry array.
2485 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
2487 struct shared_msr_entry tmp;
2489 tmp = vmx->guest_msrs[to];
2490 vmx->guest_msrs[to] = vmx->guest_msrs[from];
2491 vmx->guest_msrs[from] = tmp;
2494 static void vmx_set_msr_bitmap(struct kvm_vcpu *vcpu)
2496 unsigned long *msr_bitmap;
2498 if (is_guest_mode(vcpu))
2499 msr_bitmap = vmx_msr_bitmap_nested;
2500 else if (cpu_has_secondary_exec_ctrls() &&
2501 (vmcs_read32(SECONDARY_VM_EXEC_CONTROL) &
2502 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
2503 if (is_long_mode(vcpu))
2504 msr_bitmap = vmx_msr_bitmap_longmode_x2apic;
2506 msr_bitmap = vmx_msr_bitmap_legacy_x2apic;
2508 if (is_long_mode(vcpu))
2509 msr_bitmap = vmx_msr_bitmap_longmode;
2511 msr_bitmap = vmx_msr_bitmap_legacy;
2514 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
2518 * Set up the vmcs to automatically save and restore system
2519 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
2520 * mode, as fiddling with msrs is very expensive.
2522 static void setup_msrs(struct vcpu_vmx *vmx)
2524 int save_nmsrs, index;
2527 #ifdef CONFIG_X86_64
2528 if (is_long_mode(&vmx->vcpu)) {
2529 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
2531 move_msr_up(vmx, index, save_nmsrs++);
2532 index = __find_msr_index(vmx, MSR_LSTAR);
2534 move_msr_up(vmx, index, save_nmsrs++);
2535 index = __find_msr_index(vmx, MSR_CSTAR);
2537 move_msr_up(vmx, index, save_nmsrs++);
2538 index = __find_msr_index(vmx, MSR_TSC_AUX);
2539 if (index >= 0 && guest_cpuid_has_rdtscp(&vmx->vcpu))
2540 move_msr_up(vmx, index, save_nmsrs++);
2542 * MSR_STAR is only needed on long mode guests, and only
2543 * if efer.sce is enabled.
2545 index = __find_msr_index(vmx, MSR_STAR);
2546 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
2547 move_msr_up(vmx, index, save_nmsrs++);
2550 index = __find_msr_index(vmx, MSR_EFER);
2551 if (index >= 0 && update_transition_efer(vmx, index))
2552 move_msr_up(vmx, index, save_nmsrs++);
2554 vmx->save_nmsrs = save_nmsrs;
2556 if (cpu_has_vmx_msr_bitmap())
2557 vmx_set_msr_bitmap(&vmx->vcpu);
2561 * reads and returns guest's timestamp counter "register"
2562 * guest_tsc = (host_tsc * tsc multiplier) >> 48 + tsc_offset
2563 * -- Intel TSC Scaling for Virtualization White Paper, sec 1.3
2565 static u64 guest_read_tsc(struct kvm_vcpu *vcpu)
2567 u64 host_tsc, tsc_offset;
2570 tsc_offset = vmcs_read64(TSC_OFFSET);
2571 return kvm_scale_tsc(vcpu, host_tsc) + tsc_offset;
2575 * Like guest_read_tsc, but always returns L1's notion of the timestamp
2576 * counter, even if a nested guest (L2) is currently running.
2578 static u64 vmx_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2582 tsc_offset = is_guest_mode(vcpu) ?
2583 to_vmx(vcpu)->nested.vmcs01_tsc_offset :
2584 vmcs_read64(TSC_OFFSET);
2585 return host_tsc + tsc_offset;
2588 static u64 vmx_read_tsc_offset(struct kvm_vcpu *vcpu)
2590 return vmcs_read64(TSC_OFFSET);
2594 * writes 'offset' into guest's timestamp counter offset register
2596 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2598 if (is_guest_mode(vcpu)) {
2600 * We're here if L1 chose not to trap WRMSR to TSC. According
2601 * to the spec, this should set L1's TSC; The offset that L1
2602 * set for L2 remains unchanged, and still needs to be added
2603 * to the newly set TSC to get L2's TSC.
2605 struct vmcs12 *vmcs12;
2606 to_vmx(vcpu)->nested.vmcs01_tsc_offset = offset;
2607 /* recalculate vmcs02.TSC_OFFSET: */
2608 vmcs12 = get_vmcs12(vcpu);
2609 vmcs_write64(TSC_OFFSET, offset +
2610 (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
2611 vmcs12->tsc_offset : 0));
2613 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2614 vmcs_read64(TSC_OFFSET), offset);
2615 vmcs_write64(TSC_OFFSET, offset);
2619 static void vmx_adjust_tsc_offset_guest(struct kvm_vcpu *vcpu, s64 adjustment)
2621 u64 offset = vmcs_read64(TSC_OFFSET);
2623 vmcs_write64(TSC_OFFSET, offset + adjustment);
2624 if (is_guest_mode(vcpu)) {
2625 /* Even when running L2, the adjustment needs to apply to L1 */
2626 to_vmx(vcpu)->nested.vmcs01_tsc_offset += adjustment;
2628 trace_kvm_write_tsc_offset(vcpu->vcpu_id, offset,
2629 offset + adjustment);
2632 static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
2634 struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
2635 return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
2639 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
2640 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
2641 * all guests if the "nested" module option is off, and can also be disabled
2642 * for a single guest by disabling its VMX cpuid bit.
2644 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
2646 return nested && guest_cpuid_has_vmx(vcpu);
2650 * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
2651 * returned for the various VMX controls MSRs when nested VMX is enabled.
2652 * The same values should also be used to verify that vmcs12 control fields are
2653 * valid during nested entry from L1 to L2.
2654 * Each of these control msrs has a low and high 32-bit half: A low bit is on
2655 * if the corresponding bit in the (32-bit) control field *must* be on, and a
2656 * bit in the high half is on if the corresponding bit in the control field
2657 * may be on. See also vmx_control_verify().
2659 static void nested_vmx_setup_ctls_msrs(struct vcpu_vmx *vmx)
2662 * Note that as a general rule, the high half of the MSRs (bits in
2663 * the control fields which may be 1) should be initialized by the
2664 * intersection of the underlying hardware's MSR (i.e., features which
2665 * can be supported) and the list of features we want to expose -
2666 * because they are known to be properly supported in our code.
2667 * Also, usually, the low half of the MSRs (bits which must be 1) can
2668 * be set to 0, meaning that L1 may turn off any of these bits. The
2669 * reason is that if one of these bits is necessary, it will appear
2670 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
2671 * fields of vmcs01 and vmcs02, will turn these bits off - and
2672 * nested_vmx_exit_handled() will not pass related exits to L1.
2673 * These rules have exceptions below.
2676 /* pin-based controls */
2677 rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
2678 vmx->nested.nested_vmx_pinbased_ctls_low,
2679 vmx->nested.nested_vmx_pinbased_ctls_high);
2680 vmx->nested.nested_vmx_pinbased_ctls_low |=
2681 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2682 vmx->nested.nested_vmx_pinbased_ctls_high &=
2683 PIN_BASED_EXT_INTR_MASK |
2684 PIN_BASED_NMI_EXITING |
2685 PIN_BASED_VIRTUAL_NMIS;
2686 vmx->nested.nested_vmx_pinbased_ctls_high |=
2687 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2688 PIN_BASED_VMX_PREEMPTION_TIMER;
2689 if (kvm_vcpu_apicv_active(&vmx->vcpu))
2690 vmx->nested.nested_vmx_pinbased_ctls_high |=
2691 PIN_BASED_POSTED_INTR;
2694 rdmsr(MSR_IA32_VMX_EXIT_CTLS,
2695 vmx->nested.nested_vmx_exit_ctls_low,
2696 vmx->nested.nested_vmx_exit_ctls_high);
2697 vmx->nested.nested_vmx_exit_ctls_low =
2698 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
2700 vmx->nested.nested_vmx_exit_ctls_high &=
2701 #ifdef CONFIG_X86_64
2702 VM_EXIT_HOST_ADDR_SPACE_SIZE |
2704 VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
2705 vmx->nested.nested_vmx_exit_ctls_high |=
2706 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
2707 VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
2708 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
2710 if (kvm_mpx_supported())
2711 vmx->nested.nested_vmx_exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
2713 /* We support free control of debug control saving. */
2714 vmx->nested.nested_vmx_true_exit_ctls_low =
2715 vmx->nested.nested_vmx_exit_ctls_low &
2716 ~VM_EXIT_SAVE_DEBUG_CONTROLS;
2718 /* entry controls */
2719 rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
2720 vmx->nested.nested_vmx_entry_ctls_low,
2721 vmx->nested.nested_vmx_entry_ctls_high);
2722 vmx->nested.nested_vmx_entry_ctls_low =
2723 VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
2724 vmx->nested.nested_vmx_entry_ctls_high &=
2725 #ifdef CONFIG_X86_64
2726 VM_ENTRY_IA32E_MODE |
2728 VM_ENTRY_LOAD_IA32_PAT;
2729 vmx->nested.nested_vmx_entry_ctls_high |=
2730 (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
2731 if (kvm_mpx_supported())
2732 vmx->nested.nested_vmx_entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
2734 /* We support free control of debug control loading. */
2735 vmx->nested.nested_vmx_true_entry_ctls_low =
2736 vmx->nested.nested_vmx_entry_ctls_low &
2737 ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
2739 /* cpu-based controls */
2740 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
2741 vmx->nested.nested_vmx_procbased_ctls_low,
2742 vmx->nested.nested_vmx_procbased_ctls_high);
2743 vmx->nested.nested_vmx_procbased_ctls_low =
2744 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2745 vmx->nested.nested_vmx_procbased_ctls_high &=
2746 CPU_BASED_VIRTUAL_INTR_PENDING |
2747 CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
2748 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
2749 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
2750 CPU_BASED_CR3_STORE_EXITING |
2751 #ifdef CONFIG_X86_64
2752 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
2754 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
2755 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
2756 CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
2757 CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
2758 CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2760 * We can allow some features even when not supported by the
2761 * hardware. For example, L1 can specify an MSR bitmap - and we
2762 * can use it to avoid exits to L1 - even when L0 runs L2
2763 * without MSR bitmaps.
2765 vmx->nested.nested_vmx_procbased_ctls_high |=
2766 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2767 CPU_BASED_USE_MSR_BITMAPS;
2769 /* We support free control of CR3 access interception. */
2770 vmx->nested.nested_vmx_true_procbased_ctls_low =
2771 vmx->nested.nested_vmx_procbased_ctls_low &
2772 ~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
2774 /* secondary cpu-based controls */
2775 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2776 vmx->nested.nested_vmx_secondary_ctls_low,
2777 vmx->nested.nested_vmx_secondary_ctls_high);
2778 vmx->nested.nested_vmx_secondary_ctls_low = 0;
2779 vmx->nested.nested_vmx_secondary_ctls_high &=
2780 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2781 SECONDARY_EXEC_RDTSCP |
2782 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2783 SECONDARY_EXEC_ENABLE_VPID |
2784 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2785 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2786 SECONDARY_EXEC_WBINVD_EXITING |
2787 SECONDARY_EXEC_XSAVES |
2788 SECONDARY_EXEC_PCOMMIT;
2791 /* nested EPT: emulate EPT also to L1 */
2792 vmx->nested.nested_vmx_secondary_ctls_high |=
2793 SECONDARY_EXEC_ENABLE_EPT;
2794 vmx->nested.nested_vmx_ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
2795 VMX_EPTP_WB_BIT | VMX_EPT_2MB_PAGE_BIT |
2797 vmx->nested.nested_vmx_ept_caps &= vmx_capability.ept;
2799 * For nested guests, we don't do anything specific
2800 * for single context invalidation. Hence, only advertise
2801 * support for global context invalidation.
2803 vmx->nested.nested_vmx_ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT;
2805 vmx->nested.nested_vmx_ept_caps = 0;
2808 * Old versions of KVM use the single-context version without
2809 * checking for support, so declare that it is supported even
2810 * though it is treated as global context. The alternative is
2811 * not failing the single-context invvpid, and it is worse.
2814 vmx->nested.nested_vmx_vpid_caps = VMX_VPID_INVVPID_BIT |
2815 VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT |
2816 VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
2818 vmx->nested.nested_vmx_vpid_caps = 0;
2820 if (enable_unrestricted_guest)
2821 vmx->nested.nested_vmx_secondary_ctls_high |=
2822 SECONDARY_EXEC_UNRESTRICTED_GUEST;
2824 /* miscellaneous data */
2825 rdmsr(MSR_IA32_VMX_MISC,
2826 vmx->nested.nested_vmx_misc_low,
2827 vmx->nested.nested_vmx_misc_high);
2828 vmx->nested.nested_vmx_misc_low &= VMX_MISC_SAVE_EFER_LMA;
2829 vmx->nested.nested_vmx_misc_low |=
2830 VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
2831 VMX_MISC_ACTIVITY_HLT;
2832 vmx->nested.nested_vmx_misc_high = 0;
2835 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2838 * Bits 0 in high must be 0, and bits 1 in low must be 1.
2840 return ((control & high) | low) == control;
2843 static inline u64 vmx_control_msr(u32 low, u32 high)
2845 return low | ((u64)high << 32);
2848 /* Returns 0 on success, non-0 otherwise. */
2849 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2851 struct vcpu_vmx *vmx = to_vmx(vcpu);
2853 switch (msr_index) {
2854 case MSR_IA32_VMX_BASIC:
2856 * This MSR reports some information about VMX support. We
2857 * should return information about the VMX we emulate for the
2858 * guest, and the VMCS structure we give it - not about the
2859 * VMX support of the underlying hardware.
2861 *pdata = VMCS12_REVISION | VMX_BASIC_TRUE_CTLS |
2862 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2863 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2865 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2866 case MSR_IA32_VMX_PINBASED_CTLS:
2867 *pdata = vmx_control_msr(
2868 vmx->nested.nested_vmx_pinbased_ctls_low,
2869 vmx->nested.nested_vmx_pinbased_ctls_high);
2871 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2872 *pdata = vmx_control_msr(
2873 vmx->nested.nested_vmx_true_procbased_ctls_low,
2874 vmx->nested.nested_vmx_procbased_ctls_high);
2876 case MSR_IA32_VMX_PROCBASED_CTLS:
2877 *pdata = vmx_control_msr(
2878 vmx->nested.nested_vmx_procbased_ctls_low,
2879 vmx->nested.nested_vmx_procbased_ctls_high);
2881 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2882 *pdata = vmx_control_msr(
2883 vmx->nested.nested_vmx_true_exit_ctls_low,
2884 vmx->nested.nested_vmx_exit_ctls_high);
2886 case MSR_IA32_VMX_EXIT_CTLS:
2887 *pdata = vmx_control_msr(
2888 vmx->nested.nested_vmx_exit_ctls_low,
2889 vmx->nested.nested_vmx_exit_ctls_high);
2891 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2892 *pdata = vmx_control_msr(
2893 vmx->nested.nested_vmx_true_entry_ctls_low,
2894 vmx->nested.nested_vmx_entry_ctls_high);
2896 case MSR_IA32_VMX_ENTRY_CTLS:
2897 *pdata = vmx_control_msr(
2898 vmx->nested.nested_vmx_entry_ctls_low,
2899 vmx->nested.nested_vmx_entry_ctls_high);
2901 case MSR_IA32_VMX_MISC:
2902 *pdata = vmx_control_msr(
2903 vmx->nested.nested_vmx_misc_low,
2904 vmx->nested.nested_vmx_misc_high);
2907 * These MSRs specify bits which the guest must keep fixed (on or off)
2908 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2909 * We picked the standard core2 setting.
2911 #define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2912 #define VMXON_CR4_ALWAYSON X86_CR4_VMXE
2913 case MSR_IA32_VMX_CR0_FIXED0:
2914 *pdata = VMXON_CR0_ALWAYSON;
2916 case MSR_IA32_VMX_CR0_FIXED1:
2919 case MSR_IA32_VMX_CR4_FIXED0:
2920 *pdata = VMXON_CR4_ALWAYSON;
2922 case MSR_IA32_VMX_CR4_FIXED1:
2925 case MSR_IA32_VMX_VMCS_ENUM:
2926 *pdata = 0x2e; /* highest index: VMX_PREEMPTION_TIMER_VALUE */
2928 case MSR_IA32_VMX_PROCBASED_CTLS2:
2929 *pdata = vmx_control_msr(
2930 vmx->nested.nested_vmx_secondary_ctls_low,
2931 vmx->nested.nested_vmx_secondary_ctls_high);
2933 case MSR_IA32_VMX_EPT_VPID_CAP:
2934 /* Currently, no nested vpid support */
2935 *pdata = vmx->nested.nested_vmx_ept_caps |
2936 ((u64)vmx->nested.nested_vmx_vpid_caps << 32);
2945 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
2948 uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
2950 return !(val & ~valid_bits);
2954 * Reads an msr value (of 'msr_index') into 'pdata'.
2955 * Returns 0 on success, non-0 otherwise.
2956 * Assumes vcpu_load() was already called.
2958 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2960 struct shared_msr_entry *msr;
2962 switch (msr_info->index) {
2963 #ifdef CONFIG_X86_64
2965 msr_info->data = vmcs_readl(GUEST_FS_BASE);
2968 msr_info->data = vmcs_readl(GUEST_GS_BASE);
2970 case MSR_KERNEL_GS_BASE:
2971 vmx_load_host_state(to_vmx(vcpu));
2972 msr_info->data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
2976 return kvm_get_msr_common(vcpu, msr_info);
2978 msr_info->data = guest_read_tsc(vcpu);
2980 case MSR_IA32_SYSENTER_CS:
2981 msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
2983 case MSR_IA32_SYSENTER_EIP:
2984 msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
2986 case MSR_IA32_SYSENTER_ESP:
2987 msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
2989 case MSR_IA32_BNDCFGS:
2990 if (!kvm_mpx_supported())
2992 msr_info->data = vmcs_read64(GUEST_BNDCFGS);
2994 case MSR_IA32_MCG_EXT_CTL:
2995 if (!msr_info->host_initiated &&
2996 !(to_vmx(vcpu)->msr_ia32_feature_control &
2997 FEATURE_CONTROL_LMCE))
2999 msr_info->data = vcpu->arch.mcg_ext_ctl;
3001 case MSR_IA32_FEATURE_CONTROL:
3002 msr_info->data = to_vmx(vcpu)->msr_ia32_feature_control;
3004 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3005 if (!nested_vmx_allowed(vcpu))
3007 return vmx_get_vmx_msr(vcpu, msr_info->index, &msr_info->data);
3009 if (!vmx_xsaves_supported())
3011 msr_info->data = vcpu->arch.ia32_xss;
3014 if (!guest_cpuid_has_rdtscp(vcpu) && !msr_info->host_initiated)
3016 /* Otherwise falls through */
3018 msr = find_msr_entry(to_vmx(vcpu), msr_info->index);
3020 msr_info->data = msr->data;
3023 return kvm_get_msr_common(vcpu, msr_info);
3029 static void vmx_leave_nested(struct kvm_vcpu *vcpu);
3032 * Writes msr value into into the appropriate "register".
3033 * Returns 0 on success, non-0 otherwise.
3034 * Assumes vcpu_load() was already called.
3036 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3038 struct vcpu_vmx *vmx = to_vmx(vcpu);
3039 struct shared_msr_entry *msr;
3041 u32 msr_index = msr_info->index;
3042 u64 data = msr_info->data;
3044 switch (msr_index) {
3046 ret = kvm_set_msr_common(vcpu, msr_info);
3048 #ifdef CONFIG_X86_64
3050 vmx_segment_cache_clear(vmx);
3051 vmcs_writel(GUEST_FS_BASE, data);
3054 vmx_segment_cache_clear(vmx);
3055 vmcs_writel(GUEST_GS_BASE, data);
3057 case MSR_KERNEL_GS_BASE:
3058 vmx_load_host_state(vmx);
3059 vmx->msr_guest_kernel_gs_base = data;
3062 case MSR_IA32_SYSENTER_CS:
3063 vmcs_write32(GUEST_SYSENTER_CS, data);
3065 case MSR_IA32_SYSENTER_EIP:
3066 vmcs_writel(GUEST_SYSENTER_EIP, data);
3068 case MSR_IA32_SYSENTER_ESP:
3069 vmcs_writel(GUEST_SYSENTER_ESP, data);
3071 case MSR_IA32_BNDCFGS:
3072 if (!kvm_mpx_supported())
3074 vmcs_write64(GUEST_BNDCFGS, data);
3077 kvm_write_tsc(vcpu, msr_info);
3079 case MSR_IA32_CR_PAT:
3080 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
3081 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
3083 vmcs_write64(GUEST_IA32_PAT, data);
3084 vcpu->arch.pat = data;
3087 ret = kvm_set_msr_common(vcpu, msr_info);
3089 case MSR_IA32_TSC_ADJUST:
3090 ret = kvm_set_msr_common(vcpu, msr_info);
3092 case MSR_IA32_MCG_EXT_CTL:
3093 if ((!msr_info->host_initiated &&
3094 !(to_vmx(vcpu)->msr_ia32_feature_control &
3095 FEATURE_CONTROL_LMCE)) ||
3096 (data & ~MCG_EXT_CTL_LMCE_EN))
3098 vcpu->arch.mcg_ext_ctl = data;
3100 case MSR_IA32_FEATURE_CONTROL:
3101 if (!vmx_feature_control_msr_valid(vcpu, data) ||
3102 (to_vmx(vcpu)->msr_ia32_feature_control &
3103 FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
3105 vmx->msr_ia32_feature_control = data;
3106 if (msr_info->host_initiated && data == 0)
3107 vmx_leave_nested(vcpu);
3109 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3110 return 1; /* they are read-only */
3112 if (!vmx_xsaves_supported())
3115 * The only supported bit as of Skylake is bit 8, but
3116 * it is not supported on KVM.
3120 vcpu->arch.ia32_xss = data;
3121 if (vcpu->arch.ia32_xss != host_xss)
3122 add_atomic_switch_msr(vmx, MSR_IA32_XSS,
3123 vcpu->arch.ia32_xss, host_xss);
3125 clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
3128 if (!guest_cpuid_has_rdtscp(vcpu) && !msr_info->host_initiated)
3130 /* Check reserved bit, higher 32 bits should be zero */
3131 if ((data >> 32) != 0)
3133 /* Otherwise falls through */
3135 msr = find_msr_entry(vmx, msr_index);
3137 u64 old_msr_data = msr->data;
3139 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
3141 ret = kvm_set_shared_msr(msr->index, msr->data,
3145 msr->data = old_msr_data;
3149 ret = kvm_set_msr_common(vcpu, msr_info);
3155 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
3157 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
3160 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
3163 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
3165 case VCPU_EXREG_PDPTR:
3167 ept_save_pdptrs(vcpu);
3174 static __init int cpu_has_kvm_support(void)
3176 return cpu_has_vmx();
3179 static __init int vmx_disabled_by_bios(void)
3183 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
3184 if (msr & FEATURE_CONTROL_LOCKED) {
3185 /* launched w/ TXT and VMX disabled */
3186 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3189 /* launched w/o TXT and VMX only enabled w/ TXT */
3190 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3191 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3192 && !tboot_enabled()) {
3193 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
3194 "activate TXT before enabling KVM\n");
3197 /* launched w/o TXT and VMX disabled */
3198 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3199 && !tboot_enabled())
3206 static void kvm_cpu_vmxon(u64 addr)
3208 intel_pt_handle_vmx(1);
3210 asm volatile (ASM_VMX_VMXON_RAX
3211 : : "a"(&addr), "m"(addr)
3215 static int hardware_enable(void)
3217 int cpu = raw_smp_processor_id();
3218 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
3221 if (cr4_read_shadow() & X86_CR4_VMXE)
3224 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
3225 INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
3226 spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
3229 * Now we can enable the vmclear operation in kdump
3230 * since the loaded_vmcss_on_cpu list on this cpu
3231 * has been initialized.
3233 * Though the cpu is not in VMX operation now, there
3234 * is no problem to enable the vmclear operation
3235 * for the loaded_vmcss_on_cpu list is empty!
3237 crash_enable_local_vmclear(cpu);
3239 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
3241 test_bits = FEATURE_CONTROL_LOCKED;
3242 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
3243 if (tboot_enabled())
3244 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
3246 if ((old & test_bits) != test_bits) {
3247 /* enable and lock */
3248 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
3250 cr4_set_bits(X86_CR4_VMXE);
3252 if (vmm_exclusive) {
3253 kvm_cpu_vmxon(phys_addr);
3257 native_store_gdt(this_cpu_ptr(&host_gdt));
3262 static void vmclear_local_loaded_vmcss(void)
3264 int cpu = raw_smp_processor_id();
3265 struct loaded_vmcs *v, *n;
3267 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
3268 loaded_vmcss_on_cpu_link)
3269 __loaded_vmcs_clear(v);
3273 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
3276 static void kvm_cpu_vmxoff(void)
3278 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
3280 intel_pt_handle_vmx(0);
3283 static void hardware_disable(void)
3285 if (vmm_exclusive) {
3286 vmclear_local_loaded_vmcss();
3289 cr4_clear_bits(X86_CR4_VMXE);
3292 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
3293 u32 msr, u32 *result)
3295 u32 vmx_msr_low, vmx_msr_high;
3296 u32 ctl = ctl_min | ctl_opt;
3298 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3300 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
3301 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
3303 /* Ensure minimum (required) set of control bits are supported. */
3311 static __init bool allow_1_setting(u32 msr, u32 ctl)
3313 u32 vmx_msr_low, vmx_msr_high;
3315 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3316 return vmx_msr_high & ctl;
3319 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
3321 u32 vmx_msr_low, vmx_msr_high;
3322 u32 min, opt, min2, opt2;
3323 u32 _pin_based_exec_control = 0;
3324 u32 _cpu_based_exec_control = 0;
3325 u32 _cpu_based_2nd_exec_control = 0;
3326 u32 _vmexit_control = 0;
3327 u32 _vmentry_control = 0;
3329 min = CPU_BASED_HLT_EXITING |
3330 #ifdef CONFIG_X86_64
3331 CPU_BASED_CR8_LOAD_EXITING |
3332 CPU_BASED_CR8_STORE_EXITING |
3334 CPU_BASED_CR3_LOAD_EXITING |
3335 CPU_BASED_CR3_STORE_EXITING |
3336 CPU_BASED_USE_IO_BITMAPS |
3337 CPU_BASED_MOV_DR_EXITING |
3338 CPU_BASED_USE_TSC_OFFSETING |
3339 CPU_BASED_MWAIT_EXITING |
3340 CPU_BASED_MONITOR_EXITING |
3341 CPU_BASED_INVLPG_EXITING |
3342 CPU_BASED_RDPMC_EXITING;
3344 opt = CPU_BASED_TPR_SHADOW |
3345 CPU_BASED_USE_MSR_BITMAPS |
3346 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
3347 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
3348 &_cpu_based_exec_control) < 0)
3350 #ifdef CONFIG_X86_64
3351 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3352 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
3353 ~CPU_BASED_CR8_STORE_EXITING;
3355 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
3357 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
3358 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3359 SECONDARY_EXEC_WBINVD_EXITING |
3360 SECONDARY_EXEC_ENABLE_VPID |
3361 SECONDARY_EXEC_ENABLE_EPT |
3362 SECONDARY_EXEC_UNRESTRICTED_GUEST |
3363 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
3364 SECONDARY_EXEC_RDTSCP |
3365 SECONDARY_EXEC_ENABLE_INVPCID |
3366 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3367 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
3368 SECONDARY_EXEC_SHADOW_VMCS |
3369 SECONDARY_EXEC_XSAVES |
3370 SECONDARY_EXEC_ENABLE_PML |
3371 SECONDARY_EXEC_PCOMMIT |
3372 SECONDARY_EXEC_TSC_SCALING;
3373 if (adjust_vmx_controls(min2, opt2,
3374 MSR_IA32_VMX_PROCBASED_CTLS2,
3375 &_cpu_based_2nd_exec_control) < 0)
3378 #ifndef CONFIG_X86_64
3379 if (!(_cpu_based_2nd_exec_control &
3380 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
3381 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
3384 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3385 _cpu_based_2nd_exec_control &= ~(
3386 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3387 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3388 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3390 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
3391 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
3393 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
3394 CPU_BASED_CR3_STORE_EXITING |
3395 CPU_BASED_INVLPG_EXITING);
3396 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
3397 vmx_capability.ept, vmx_capability.vpid);
3400 min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
3401 #ifdef CONFIG_X86_64
3402 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
3404 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT |
3405 VM_EXIT_CLEAR_BNDCFGS;
3406 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
3407 &_vmexit_control) < 0)
3410 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
3411 opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
3412 PIN_BASED_VMX_PREEMPTION_TIMER;
3413 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
3414 &_pin_based_exec_control) < 0)
3417 if (cpu_has_broken_vmx_preemption_timer())
3418 _pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3419 if (!(_cpu_based_2nd_exec_control &
3420 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
3421 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
3423 min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
3424 opt = VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS;
3425 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
3426 &_vmentry_control) < 0)
3429 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
3431 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
3432 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
3435 #ifdef CONFIG_X86_64
3436 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
3437 if (vmx_msr_high & (1u<<16))
3441 /* Require Write-Back (WB) memory type for VMCS accesses. */
3442 if (((vmx_msr_high >> 18) & 15) != 6)
3445 vmcs_conf->size = vmx_msr_high & 0x1fff;
3446 vmcs_conf->order = get_order(vmcs_config.size);
3447 vmcs_conf->revision_id = vmx_msr_low;
3449 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
3450 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
3451 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
3452 vmcs_conf->vmexit_ctrl = _vmexit_control;
3453 vmcs_conf->vmentry_ctrl = _vmentry_control;
3455 cpu_has_load_ia32_efer =
3456 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3457 VM_ENTRY_LOAD_IA32_EFER)
3458 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3459 VM_EXIT_LOAD_IA32_EFER);
3461 cpu_has_load_perf_global_ctrl =
3462 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3463 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
3464 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3465 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
3468 * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
3469 * but due to errata below it can't be used. Workaround is to use
3470 * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
3472 * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
3477 * BC86,AAY89,BD102 (model 44)
3481 if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
3482 switch (boot_cpu_data.x86_model) {
3488 cpu_has_load_perf_global_ctrl = false;
3489 printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
3490 "does not work properly. Using workaround\n");
3497 if (boot_cpu_has(X86_FEATURE_XSAVES))
3498 rdmsrl(MSR_IA32_XSS, host_xss);
3503 static struct vmcs *alloc_vmcs_cpu(int cpu)
3505 int node = cpu_to_node(cpu);
3509 pages = __alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
3512 vmcs = page_address(pages);
3513 memset(vmcs, 0, vmcs_config.size);
3514 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
3518 static struct vmcs *alloc_vmcs(void)
3520 return alloc_vmcs_cpu(raw_smp_processor_id());
3523 static void free_vmcs(struct vmcs *vmcs)
3525 free_pages((unsigned long)vmcs, vmcs_config.order);
3529 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
3531 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
3533 if (!loaded_vmcs->vmcs)
3535 loaded_vmcs_clear(loaded_vmcs);
3536 free_vmcs(loaded_vmcs->vmcs);
3537 loaded_vmcs->vmcs = NULL;
3540 static void free_kvm_area(void)
3544 for_each_possible_cpu(cpu) {
3545 free_vmcs(per_cpu(vmxarea, cpu));
3546 per_cpu(vmxarea, cpu) = NULL;
3550 static void init_vmcs_shadow_fields(void)
3554 /* No checks for read only fields yet */
3556 for (i = j = 0; i < max_shadow_read_write_fields; i++) {
3557 switch (shadow_read_write_fields[i]) {
3559 if (!kvm_mpx_supported())
3567 shadow_read_write_fields[j] =
3568 shadow_read_write_fields[i];
3571 max_shadow_read_write_fields = j;
3573 /* shadowed fields guest access without vmexit */
3574 for (i = 0; i < max_shadow_read_write_fields; i++) {
3575 clear_bit(shadow_read_write_fields[i],
3576 vmx_vmwrite_bitmap);
3577 clear_bit(shadow_read_write_fields[i],
3580 for (i = 0; i < max_shadow_read_only_fields; i++)
3581 clear_bit(shadow_read_only_fields[i],
3585 static __init int alloc_kvm_area(void)
3589 for_each_possible_cpu(cpu) {
3592 vmcs = alloc_vmcs_cpu(cpu);
3598 per_cpu(vmxarea, cpu) = vmcs;
3603 static bool emulation_required(struct kvm_vcpu *vcpu)
3605 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
3608 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
3609 struct kvm_segment *save)
3611 if (!emulate_invalid_guest_state) {
3613 * CS and SS RPL should be equal during guest entry according
3614 * to VMX spec, but in reality it is not always so. Since vcpu
3615 * is in the middle of the transition from real mode to
3616 * protected mode it is safe to assume that RPL 0 is a good
3619 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
3620 save->selector &= ~SEGMENT_RPL_MASK;
3621 save->dpl = save->selector & SEGMENT_RPL_MASK;
3624 vmx_set_segment(vcpu, save, seg);
3627 static void enter_pmode(struct kvm_vcpu *vcpu)
3629 unsigned long flags;
3630 struct vcpu_vmx *vmx = to_vmx(vcpu);
3633 * Update real mode segment cache. It may be not up-to-date if sement
3634 * register was written while vcpu was in a guest mode.
3636 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3637 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3638 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3639 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3640 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3641 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3643 vmx->rmode.vm86_active = 0;
3645 vmx_segment_cache_clear(vmx);
3647 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3649 flags = vmcs_readl(GUEST_RFLAGS);
3650 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
3651 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
3652 vmcs_writel(GUEST_RFLAGS, flags);
3654 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
3655 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
3657 update_exception_bitmap(vcpu);
3659 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3660 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3661 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3662 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3663 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3664 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3667 static void fix_rmode_seg(int seg, struct kvm_segment *save)
3669 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3670 struct kvm_segment var = *save;
3673 if (seg == VCPU_SREG_CS)
3676 if (!emulate_invalid_guest_state) {
3677 var.selector = var.base >> 4;
3678 var.base = var.base & 0xffff0;
3688 if (save->base & 0xf)
3689 printk_once(KERN_WARNING "kvm: segment base is not "
3690 "paragraph aligned when entering "
3691 "protected mode (seg=%d)", seg);
3694 vmcs_write16(sf->selector, var.selector);
3695 vmcs_write32(sf->base, var.base);
3696 vmcs_write32(sf->limit, var.limit);
3697 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
3700 static void enter_rmode(struct kvm_vcpu *vcpu)
3702 unsigned long flags;
3703 struct vcpu_vmx *vmx = to_vmx(vcpu);
3705 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3706 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3707 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3708 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3709 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3710 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3711 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3713 vmx->rmode.vm86_active = 1;
3716 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
3717 * vcpu. Warn the user that an update is overdue.
3719 if (!vcpu->kvm->arch.tss_addr)
3720 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
3721 "called before entering vcpu\n");
3723 vmx_segment_cache_clear(vmx);
3725 vmcs_writel(GUEST_TR_BASE, vcpu->kvm->arch.tss_addr);
3726 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
3727 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3729 flags = vmcs_readl(GUEST_RFLAGS);
3730 vmx->rmode.save_rflags = flags;
3732 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
3734 vmcs_writel(GUEST_RFLAGS, flags);
3735 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
3736 update_exception_bitmap(vcpu);
3738 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3739 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3740 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3741 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3742 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3743 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3745 kvm_mmu_reset_context(vcpu);
3748 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
3750 struct vcpu_vmx *vmx = to_vmx(vcpu);
3751 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
3757 * Force kernel_gs_base reloading before EFER changes, as control
3758 * of this msr depends on is_long_mode().
3760 vmx_load_host_state(to_vmx(vcpu));
3761 vcpu->arch.efer = efer;
3762 if (efer & EFER_LMA) {
3763 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3766 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3768 msr->data = efer & ~EFER_LME;
3773 #ifdef CONFIG_X86_64
3775 static void enter_lmode(struct kvm_vcpu *vcpu)
3779 vmx_segment_cache_clear(to_vmx(vcpu));
3781 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
3782 if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
3783 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
3785 vmcs_write32(GUEST_TR_AR_BYTES,
3786 (guest_tr_ar & ~VMX_AR_TYPE_MASK)
3787 | VMX_AR_TYPE_BUSY_64_TSS);
3789 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
3792 static void exit_lmode(struct kvm_vcpu *vcpu)
3794 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3795 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
3800 static inline void __vmx_flush_tlb(struct kvm_vcpu *vcpu, int vpid)
3802 vpid_sync_context(vpid);
3804 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
3806 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
3810 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
3812 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->vpid);
3815 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
3817 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
3819 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
3820 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
3823 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
3825 if (enable_ept && is_paging(vcpu))
3826 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3827 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
3830 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
3832 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
3834 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
3835 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
3838 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
3840 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3842 if (!test_bit(VCPU_EXREG_PDPTR,
3843 (unsigned long *)&vcpu->arch.regs_dirty))
3846 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3847 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
3848 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
3849 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
3850 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
3854 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
3856 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3858 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3859 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
3860 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
3861 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
3862 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
3865 __set_bit(VCPU_EXREG_PDPTR,
3866 (unsigned long *)&vcpu->arch.regs_avail);
3867 __set_bit(VCPU_EXREG_PDPTR,
3868 (unsigned long *)&vcpu->arch.regs_dirty);
3871 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
3873 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
3875 struct kvm_vcpu *vcpu)
3877 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3878 vmx_decache_cr3(vcpu);
3879 if (!(cr0 & X86_CR0_PG)) {
3880 /* From paging/starting to nonpaging */
3881 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3882 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
3883 (CPU_BASED_CR3_LOAD_EXITING |
3884 CPU_BASED_CR3_STORE_EXITING));
3885 vcpu->arch.cr0 = cr0;
3886 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3887 } else if (!is_paging(vcpu)) {
3888 /* From nonpaging to paging */
3889 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3890 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
3891 ~(CPU_BASED_CR3_LOAD_EXITING |
3892 CPU_BASED_CR3_STORE_EXITING));
3893 vcpu->arch.cr0 = cr0;
3894 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3897 if (!(cr0 & X86_CR0_WP))
3898 *hw_cr0 &= ~X86_CR0_WP;
3901 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
3903 struct vcpu_vmx *vmx = to_vmx(vcpu);
3904 unsigned long hw_cr0;
3906 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK);
3907 if (enable_unrestricted_guest)
3908 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
3910 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
3912 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
3915 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
3919 #ifdef CONFIG_X86_64
3920 if (vcpu->arch.efer & EFER_LME) {
3921 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
3923 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
3929 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
3931 if (!vcpu->fpu_active)
3932 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
3934 vmcs_writel(CR0_READ_SHADOW, cr0);
3935 vmcs_writel(GUEST_CR0, hw_cr0);
3936 vcpu->arch.cr0 = cr0;
3938 /* depends on vcpu->arch.cr0 to be set to a new value */
3939 vmx->emulation_required = emulation_required(vcpu);
3942 static u64 construct_eptp(unsigned long root_hpa)
3946 /* TODO write the value reading from MSR */
3947 eptp = VMX_EPT_DEFAULT_MT |
3948 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
3949 if (enable_ept_ad_bits)
3950 eptp |= VMX_EPT_AD_ENABLE_BIT;
3951 eptp |= (root_hpa & PAGE_MASK);
3956 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
3958 unsigned long guest_cr3;
3963 eptp = construct_eptp(cr3);
3964 vmcs_write64(EPT_POINTER, eptp);
3965 if (is_paging(vcpu) || is_guest_mode(vcpu))
3966 guest_cr3 = kvm_read_cr3(vcpu);
3968 guest_cr3 = vcpu->kvm->arch.ept_identity_map_addr;
3969 ept_load_pdptrs(vcpu);
3972 vmx_flush_tlb(vcpu);
3973 vmcs_writel(GUEST_CR3, guest_cr3);
3976 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3979 * Pass through host's Machine Check Enable value to hw_cr4, which
3980 * is in force while we are in guest mode. Do not let guests control
3981 * this bit, even if host CR4.MCE == 0.
3983 unsigned long hw_cr4 =
3984 (cr4_read_shadow() & X86_CR4_MCE) |
3985 (cr4 & ~X86_CR4_MCE) |
3986 (to_vmx(vcpu)->rmode.vm86_active ?
3987 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
3989 if (cr4 & X86_CR4_VMXE) {
3991 * To use VMXON (and later other VMX instructions), a guest
3992 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3993 * So basically the check on whether to allow nested VMX
3996 if (!nested_vmx_allowed(vcpu))
3999 if (to_vmx(vcpu)->nested.vmxon &&
4000 ((cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON))
4003 vcpu->arch.cr4 = cr4;
4005 if (!is_paging(vcpu)) {
4006 hw_cr4 &= ~X86_CR4_PAE;
4007 hw_cr4 |= X86_CR4_PSE;
4008 } else if (!(cr4 & X86_CR4_PAE)) {
4009 hw_cr4 &= ~X86_CR4_PAE;
4013 if (!enable_unrestricted_guest && !is_paging(vcpu))
4015 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
4016 * hardware. To emulate this behavior, SMEP/SMAP/PKU needs
4017 * to be manually disabled when guest switches to non-paging
4020 * If !enable_unrestricted_guest, the CPU is always running
4021 * with CR0.PG=1 and CR4 needs to be modified.
4022 * If enable_unrestricted_guest, the CPU automatically
4023 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
4025 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
4027 vmcs_writel(CR4_READ_SHADOW, cr4);
4028 vmcs_writel(GUEST_CR4, hw_cr4);
4032 static void vmx_get_segment(struct kvm_vcpu *vcpu,
4033 struct kvm_segment *var, int seg)
4035 struct vcpu_vmx *vmx = to_vmx(vcpu);
4038 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
4039 *var = vmx->rmode.segs[seg];
4040 if (seg == VCPU_SREG_TR
4041 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
4043 var->base = vmx_read_guest_seg_base(vmx, seg);
4044 var->selector = vmx_read_guest_seg_selector(vmx, seg);
4047 var->base = vmx_read_guest_seg_base(vmx, seg);
4048 var->limit = vmx_read_guest_seg_limit(vmx, seg);
4049 var->selector = vmx_read_guest_seg_selector(vmx, seg);
4050 ar = vmx_read_guest_seg_ar(vmx, seg);
4051 var->unusable = (ar >> 16) & 1;
4052 var->type = ar & 15;
4053 var->s = (ar >> 4) & 1;
4054 var->dpl = (ar >> 5) & 3;
4056 * Some userspaces do not preserve unusable property. Since usable
4057 * segment has to be present according to VMX spec we can use present
4058 * property to amend userspace bug by making unusable segment always
4059 * nonpresent. vmx_segment_access_rights() already marks nonpresent
4060 * segment as unusable.
4062 var->present = !var->unusable;
4063 var->avl = (ar >> 12) & 1;
4064 var->l = (ar >> 13) & 1;
4065 var->db = (ar >> 14) & 1;
4066 var->g = (ar >> 15) & 1;
4069 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
4071 struct kvm_segment s;
4073 if (to_vmx(vcpu)->rmode.vm86_active) {
4074 vmx_get_segment(vcpu, &s, seg);
4077 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
4080 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
4082 struct vcpu_vmx *vmx = to_vmx(vcpu);
4084 if (unlikely(vmx->rmode.vm86_active))
4087 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
4088 return VMX_AR_DPL(ar);
4092 static u32 vmx_segment_access_rights(struct kvm_segment *var)
4096 if (var->unusable || !var->present)
4099 ar = var->type & 15;
4100 ar |= (var->s & 1) << 4;
4101 ar |= (var->dpl & 3) << 5;
4102 ar |= (var->present & 1) << 7;
4103 ar |= (var->avl & 1) << 12;
4104 ar |= (var->l & 1) << 13;
4105 ar |= (var->db & 1) << 14;
4106 ar |= (var->g & 1) << 15;
4112 static void vmx_set_segment(struct kvm_vcpu *vcpu,
4113 struct kvm_segment *var, int seg)
4115 struct vcpu_vmx *vmx = to_vmx(vcpu);
4116 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4118 vmx_segment_cache_clear(vmx);
4120 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
4121 vmx->rmode.segs[seg] = *var;
4122 if (seg == VCPU_SREG_TR)
4123 vmcs_write16(sf->selector, var->selector);
4125 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
4129 vmcs_writel(sf->base, var->base);
4130 vmcs_write32(sf->limit, var->limit);
4131 vmcs_write16(sf->selector, var->selector);
4134 * Fix the "Accessed" bit in AR field of segment registers for older
4136 * IA32 arch specifies that at the time of processor reset the
4137 * "Accessed" bit in the AR field of segment registers is 1. And qemu
4138 * is setting it to 0 in the userland code. This causes invalid guest
4139 * state vmexit when "unrestricted guest" mode is turned on.
4140 * Fix for this setup issue in cpu_reset is being pushed in the qemu
4141 * tree. Newer qemu binaries with that qemu fix would not need this
4144 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
4145 var->type |= 0x1; /* Accessed */
4147 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
4150 vmx->emulation_required = emulation_required(vcpu);
4153 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4155 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
4157 *db = (ar >> 14) & 1;
4158 *l = (ar >> 13) & 1;
4161 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4163 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
4164 dt->address = vmcs_readl(GUEST_IDTR_BASE);
4167 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4169 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
4170 vmcs_writel(GUEST_IDTR_BASE, dt->address);
4173 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4175 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
4176 dt->address = vmcs_readl(GUEST_GDTR_BASE);
4179 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4181 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
4182 vmcs_writel(GUEST_GDTR_BASE, dt->address);
4185 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
4187 struct kvm_segment var;
4190 vmx_get_segment(vcpu, &var, seg);
4192 if (seg == VCPU_SREG_CS)
4194 ar = vmx_segment_access_rights(&var);
4196 if (var.base != (var.selector << 4))
4198 if (var.limit != 0xffff)
4206 static bool code_segment_valid(struct kvm_vcpu *vcpu)
4208 struct kvm_segment cs;
4209 unsigned int cs_rpl;
4211 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4212 cs_rpl = cs.selector & SEGMENT_RPL_MASK;
4216 if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
4220 if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
4221 if (cs.dpl > cs_rpl)
4224 if (cs.dpl != cs_rpl)
4230 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
4234 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
4236 struct kvm_segment ss;
4237 unsigned int ss_rpl;
4239 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4240 ss_rpl = ss.selector & SEGMENT_RPL_MASK;
4244 if (ss.type != 3 && ss.type != 7)
4248 if (ss.dpl != ss_rpl) /* DPL != RPL */
4256 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
4258 struct kvm_segment var;
4261 vmx_get_segment(vcpu, &var, seg);
4262 rpl = var.selector & SEGMENT_RPL_MASK;
4270 if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
4271 if (var.dpl < rpl) /* DPL < RPL */
4275 /* TODO: Add other members to kvm_segment_field to allow checking for other access
4281 static bool tr_valid(struct kvm_vcpu *vcpu)
4283 struct kvm_segment tr;
4285 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
4289 if (tr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4291 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
4299 static bool ldtr_valid(struct kvm_vcpu *vcpu)
4301 struct kvm_segment ldtr;
4303 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
4307 if (ldtr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4317 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
4319 struct kvm_segment cs, ss;
4321 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4322 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4324 return ((cs.selector & SEGMENT_RPL_MASK) ==
4325 (ss.selector & SEGMENT_RPL_MASK));
4329 * Check if guest state is valid. Returns true if valid, false if
4331 * We assume that registers are always usable
4333 static bool guest_state_valid(struct kvm_vcpu *vcpu)
4335 if (enable_unrestricted_guest)
4338 /* real mode guest state checks */
4339 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
4340 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
4342 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
4344 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
4346 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
4348 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
4350 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
4353 /* protected mode guest state checks */
4354 if (!cs_ss_rpl_check(vcpu))
4356 if (!code_segment_valid(vcpu))
4358 if (!stack_segment_valid(vcpu))
4360 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
4362 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
4364 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
4366 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
4368 if (!tr_valid(vcpu))
4370 if (!ldtr_valid(vcpu))
4374 * - Add checks on RIP
4375 * - Add checks on RFLAGS
4381 static int init_rmode_tss(struct kvm *kvm)
4387 idx = srcu_read_lock(&kvm->srcu);
4388 fn = kvm->arch.tss_addr >> PAGE_SHIFT;
4389 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4392 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
4393 r = kvm_write_guest_page(kvm, fn++, &data,
4394 TSS_IOPB_BASE_OFFSET, sizeof(u16));
4397 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
4400 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4404 r = kvm_write_guest_page(kvm, fn, &data,
4405 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
4408 srcu_read_unlock(&kvm->srcu, idx);
4412 static int init_rmode_identity_map(struct kvm *kvm)
4415 kvm_pfn_t identity_map_pfn;
4421 /* Protect kvm->arch.ept_identity_pagetable_done. */
4422 mutex_lock(&kvm->slots_lock);
4424 if (likely(kvm->arch.ept_identity_pagetable_done))
4427 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
4429 r = alloc_identity_pagetable(kvm);
4433 idx = srcu_read_lock(&kvm->srcu);
4434 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
4437 /* Set up identity-mapping pagetable for EPT in real mode */
4438 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
4439 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
4440 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
4441 r = kvm_write_guest_page(kvm, identity_map_pfn,
4442 &tmp, i * sizeof(tmp), sizeof(tmp));
4446 kvm->arch.ept_identity_pagetable_done = true;
4449 srcu_read_unlock(&kvm->srcu, idx);
4452 mutex_unlock(&kvm->slots_lock);
4456 static void seg_setup(int seg)
4458 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4461 vmcs_write16(sf->selector, 0);
4462 vmcs_writel(sf->base, 0);
4463 vmcs_write32(sf->limit, 0xffff);
4465 if (seg == VCPU_SREG_CS)
4466 ar |= 0x08; /* code segment */
4468 vmcs_write32(sf->ar_bytes, ar);
4471 static int alloc_apic_access_page(struct kvm *kvm)
4476 mutex_lock(&kvm->slots_lock);
4477 if (kvm->arch.apic_access_page_done)
4479 r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
4480 APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
4484 page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
4485 if (is_error_page(page)) {
4491 * Do not pin the page in memory, so that memory hot-unplug
4492 * is able to migrate it.
4495 kvm->arch.apic_access_page_done = true;
4497 mutex_unlock(&kvm->slots_lock);
4501 static int alloc_identity_pagetable(struct kvm *kvm)
4503 /* Called with kvm->slots_lock held. */
4507 BUG_ON(kvm->arch.ept_identity_pagetable_done);
4509 r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
4510 kvm->arch.ept_identity_map_addr, PAGE_SIZE);
4515 static int allocate_vpid(void)
4521 spin_lock(&vmx_vpid_lock);
4522 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
4523 if (vpid < VMX_NR_VPIDS)
4524 __set_bit(vpid, vmx_vpid_bitmap);
4527 spin_unlock(&vmx_vpid_lock);
4531 static void free_vpid(int vpid)
4533 if (!enable_vpid || vpid == 0)
4535 spin_lock(&vmx_vpid_lock);
4536 __clear_bit(vpid, vmx_vpid_bitmap);
4537 spin_unlock(&vmx_vpid_lock);
4540 #define MSR_TYPE_R 1
4541 #define MSR_TYPE_W 2
4542 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
4545 int f = sizeof(unsigned long);
4547 if (!cpu_has_vmx_msr_bitmap())
4551 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4552 * have the write-low and read-high bitmap offsets the wrong way round.
4553 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4555 if (msr <= 0x1fff) {
4556 if (type & MSR_TYPE_R)
4558 __clear_bit(msr, msr_bitmap + 0x000 / f);
4560 if (type & MSR_TYPE_W)
4562 __clear_bit(msr, msr_bitmap + 0x800 / f);
4564 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4566 if (type & MSR_TYPE_R)
4568 __clear_bit(msr, msr_bitmap + 0x400 / f);
4570 if (type & MSR_TYPE_W)
4572 __clear_bit(msr, msr_bitmap + 0xc00 / f);
4577 static void __vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
4580 int f = sizeof(unsigned long);
4582 if (!cpu_has_vmx_msr_bitmap())
4586 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4587 * have the write-low and read-high bitmap offsets the wrong way round.
4588 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4590 if (msr <= 0x1fff) {
4591 if (type & MSR_TYPE_R)
4593 __set_bit(msr, msr_bitmap + 0x000 / f);
4595 if (type & MSR_TYPE_W)
4597 __set_bit(msr, msr_bitmap + 0x800 / f);
4599 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4601 if (type & MSR_TYPE_R)
4603 __set_bit(msr, msr_bitmap + 0x400 / f);
4605 if (type & MSR_TYPE_W)
4607 __set_bit(msr, msr_bitmap + 0xc00 / f);
4613 * If a msr is allowed by L0, we should check whether it is allowed by L1.
4614 * The corresponding bit will be cleared unless both of L0 and L1 allow it.
4616 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
4617 unsigned long *msr_bitmap_nested,
4620 int f = sizeof(unsigned long);
4622 if (!cpu_has_vmx_msr_bitmap()) {
4628 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4629 * have the write-low and read-high bitmap offsets the wrong way round.
4630 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4632 if (msr <= 0x1fff) {
4633 if (type & MSR_TYPE_R &&
4634 !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
4636 __clear_bit(msr, msr_bitmap_nested + 0x000 / f);
4638 if (type & MSR_TYPE_W &&
4639 !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
4641 __clear_bit(msr, msr_bitmap_nested + 0x800 / f);
4643 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4645 if (type & MSR_TYPE_R &&
4646 !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
4648 __clear_bit(msr, msr_bitmap_nested + 0x400 / f);
4650 if (type & MSR_TYPE_W &&
4651 !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
4653 __clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
4658 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
4661 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy,
4662 msr, MSR_TYPE_R | MSR_TYPE_W);
4663 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode,
4664 msr, MSR_TYPE_R | MSR_TYPE_W);
4667 static void vmx_enable_intercept_msr_read_x2apic(u32 msr)
4669 __vmx_enable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4671 __vmx_enable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4675 static void vmx_disable_intercept_msr_read_x2apic(u32 msr)
4677 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4679 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4683 static void vmx_disable_intercept_msr_write_x2apic(u32 msr)
4685 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4687 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4691 static bool vmx_get_enable_apicv(void)
4693 return enable_apicv;
4696 static int vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
4698 struct vcpu_vmx *vmx = to_vmx(vcpu);
4703 if (vmx->nested.pi_desc &&
4704 vmx->nested.pi_pending) {
4705 vmx->nested.pi_pending = false;
4706 if (!pi_test_and_clear_on(vmx->nested.pi_desc))
4709 max_irr = find_last_bit(
4710 (unsigned long *)vmx->nested.pi_desc->pir, 256);
4715 vapic_page = kmap(vmx->nested.virtual_apic_page);
4720 __kvm_apic_update_irr(vmx->nested.pi_desc->pir, vapic_page);
4721 kunmap(vmx->nested.virtual_apic_page);
4723 status = vmcs_read16(GUEST_INTR_STATUS);
4724 if ((u8)max_irr > ((u8)status & 0xff)) {
4726 status |= (u8)max_irr;
4727 vmcs_write16(GUEST_INTR_STATUS, status);
4733 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu)
4736 if (vcpu->mode == IN_GUEST_MODE) {
4737 struct vcpu_vmx *vmx = to_vmx(vcpu);
4740 * Currently, we don't support urgent interrupt,
4741 * all interrupts are recognized as non-urgent
4742 * interrupt, so we cannot post interrupts when
4745 * If the vcpu is in guest mode, it means it is
4746 * running instead of being scheduled out and
4747 * waiting in the run queue, and that's the only
4748 * case when 'SN' is set currently, warning if
4751 WARN_ON_ONCE(pi_test_sn(&vmx->pi_desc));
4753 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu),
4754 POSTED_INTR_VECTOR);
4761 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
4764 struct vcpu_vmx *vmx = to_vmx(vcpu);
4766 if (is_guest_mode(vcpu) &&
4767 vector == vmx->nested.posted_intr_nv) {
4768 /* the PIR and ON have been set by L1. */
4769 kvm_vcpu_trigger_posted_interrupt(vcpu);
4771 * If a posted intr is not recognized by hardware,
4772 * we will accomplish it in the next vmentry.
4774 vmx->nested.pi_pending = true;
4775 kvm_make_request(KVM_REQ_EVENT, vcpu);
4781 * Send interrupt to vcpu via posted interrupt way.
4782 * 1. If target vcpu is running(non-root mode), send posted interrupt
4783 * notification to vcpu and hardware will sync PIR to vIRR atomically.
4784 * 2. If target vcpu isn't running(root mode), kick it to pick up the
4785 * interrupt from PIR in next vmentry.
4787 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
4789 struct vcpu_vmx *vmx = to_vmx(vcpu);
4792 r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
4796 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
4799 r = pi_test_and_set_on(&vmx->pi_desc);
4800 kvm_make_request(KVM_REQ_EVENT, vcpu);
4801 if (r || !kvm_vcpu_trigger_posted_interrupt(vcpu))
4802 kvm_vcpu_kick(vcpu);
4805 static void vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
4807 struct vcpu_vmx *vmx = to_vmx(vcpu);
4809 if (!pi_test_and_clear_on(&vmx->pi_desc))
4812 kvm_apic_update_irr(vcpu, vmx->pi_desc.pir);
4816 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
4817 * will not change in the lifetime of the guest.
4818 * Note that host-state that does change is set elsewhere. E.g., host-state
4819 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
4821 static void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
4828 vmcs_writel(HOST_CR0, read_cr0() & ~X86_CR0_TS); /* 22.2.3 */
4829 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
4831 /* Save the most likely value for this task's CR4 in the VMCS. */
4832 cr4 = cr4_read_shadow();
4833 vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
4834 vmx->host_state.vmcs_host_cr4 = cr4;
4836 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
4837 #ifdef CONFIG_X86_64
4839 * Load null selectors, so we can avoid reloading them in
4840 * __vmx_load_host_state(), in case userspace uses the null selectors
4841 * too (the expected case).
4843 vmcs_write16(HOST_DS_SELECTOR, 0);
4844 vmcs_write16(HOST_ES_SELECTOR, 0);
4846 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4847 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4849 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4850 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
4852 native_store_idt(&dt);
4853 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
4854 vmx->host_idt_base = dt.address;
4856 vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
4858 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
4859 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
4860 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
4861 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
4863 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
4864 rdmsr(MSR_IA32_CR_PAT, low32, high32);
4865 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
4869 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
4871 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
4873 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
4874 if (is_guest_mode(&vmx->vcpu))
4875 vmx->vcpu.arch.cr4_guest_owned_bits &=
4876 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
4877 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
4880 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
4882 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
4884 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
4885 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
4886 /* Enable the preemption timer dynamically */
4887 pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
4888 return pin_based_exec_ctrl;
4891 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
4893 struct vcpu_vmx *vmx = to_vmx(vcpu);
4895 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
4896 if (cpu_has_secondary_exec_ctrls()) {
4897 if (kvm_vcpu_apicv_active(vcpu))
4898 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
4899 SECONDARY_EXEC_APIC_REGISTER_VIRT |
4900 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4902 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
4903 SECONDARY_EXEC_APIC_REGISTER_VIRT |
4904 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4907 if (cpu_has_vmx_msr_bitmap())
4908 vmx_set_msr_bitmap(vcpu);
4911 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
4913 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
4915 if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
4916 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
4918 if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
4919 exec_control &= ~CPU_BASED_TPR_SHADOW;
4920 #ifdef CONFIG_X86_64
4921 exec_control |= CPU_BASED_CR8_STORE_EXITING |
4922 CPU_BASED_CR8_LOAD_EXITING;
4926 exec_control |= CPU_BASED_CR3_STORE_EXITING |
4927 CPU_BASED_CR3_LOAD_EXITING |
4928 CPU_BASED_INVLPG_EXITING;
4929 return exec_control;
4932 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
4934 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4935 if (!cpu_need_virtualize_apic_accesses(&vmx->vcpu))
4936 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4938 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4940 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4941 enable_unrestricted_guest = 0;
4942 /* Enable INVPCID for non-ept guests may cause performance regression. */
4943 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4945 if (!enable_unrestricted_guest)
4946 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4948 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4949 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
4950 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4951 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4952 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4953 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4955 We can NOT enable shadow_vmcs here because we don't have yet
4958 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4961 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4963 /* Currently, we allow L1 guest to directly run pcommit instruction. */
4964 exec_control &= ~SECONDARY_EXEC_PCOMMIT;
4966 return exec_control;
4969 static void ept_set_mmio_spte_mask(void)
4972 * EPT Misconfigurations can be generated if the value of bits 2:0
4973 * of an EPT paging-structure entry is 110b (write/execute).
4974 * Also, magic bits (0x3ull << 62) is set to quickly identify mmio
4977 kvm_mmu_set_mmio_spte_mask((0x3ull << 62) | 0x6ull);
4980 #define VMX_XSS_EXIT_BITMAP 0
4982 * Sets up the vmcs for emulated real mode.
4984 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
4986 #ifdef CONFIG_X86_64
4992 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
4993 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
4995 if (enable_shadow_vmcs) {
4996 vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
4997 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
4999 if (cpu_has_vmx_msr_bitmap())
5000 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
5002 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
5005 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
5006 vmx->hv_deadline_tsc = -1;
5008 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
5010 if (cpu_has_secondary_exec_ctrls())
5011 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
5012 vmx_secondary_exec_control(vmx));
5014 if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
5015 vmcs_write64(EOI_EXIT_BITMAP0, 0);
5016 vmcs_write64(EOI_EXIT_BITMAP1, 0);
5017 vmcs_write64(EOI_EXIT_BITMAP2, 0);
5018 vmcs_write64(EOI_EXIT_BITMAP3, 0);
5020 vmcs_write16(GUEST_INTR_STATUS, 0);
5022 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
5023 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
5027 vmcs_write32(PLE_GAP, ple_gap);
5028 vmx->ple_window = ple_window;
5029 vmx->ple_window_dirty = true;
5032 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
5033 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
5034 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
5036 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
5037 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
5038 vmx_set_constant_host_state(vmx);
5039 #ifdef CONFIG_X86_64
5040 rdmsrl(MSR_FS_BASE, a);
5041 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
5042 rdmsrl(MSR_GS_BASE, a);
5043 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
5045 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
5046 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
5049 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
5050 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
5051 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
5052 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
5053 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
5055 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
5056 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
5058 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
5059 u32 index = vmx_msr_index[i];
5060 u32 data_low, data_high;
5063 if (rdmsr_safe(index, &data_low, &data_high) < 0)
5065 if (wrmsr_safe(index, data_low, data_high) < 0)
5067 vmx->guest_msrs[j].index = i;
5068 vmx->guest_msrs[j].data = 0;
5069 vmx->guest_msrs[j].mask = -1ull;
5074 vm_exit_controls_init(vmx, vmcs_config.vmexit_ctrl);
5076 /* 22.2.1, 20.8.1 */
5077 vm_entry_controls_init(vmx, vmcs_config.vmentry_ctrl);
5079 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
5080 set_cr4_guest_host_mask(vmx);
5082 if (vmx_xsaves_supported())
5083 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
5088 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
5090 struct vcpu_vmx *vmx = to_vmx(vcpu);
5091 struct msr_data apic_base_msr;
5094 vmx->rmode.vm86_active = 0;
5096 vmx->soft_vnmi_blocked = 0;
5098 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
5099 kvm_set_cr8(vcpu, 0);
5102 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
5103 MSR_IA32_APICBASE_ENABLE;
5104 if (kvm_vcpu_is_reset_bsp(vcpu))
5105 apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
5106 apic_base_msr.host_initiated = true;
5107 kvm_set_apic_base(vcpu, &apic_base_msr);
5110 vmx_segment_cache_clear(vmx);
5112 seg_setup(VCPU_SREG_CS);
5113 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
5114 vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
5116 seg_setup(VCPU_SREG_DS);
5117 seg_setup(VCPU_SREG_ES);
5118 seg_setup(VCPU_SREG_FS);
5119 seg_setup(VCPU_SREG_GS);
5120 seg_setup(VCPU_SREG_SS);
5122 vmcs_write16(GUEST_TR_SELECTOR, 0);
5123 vmcs_writel(GUEST_TR_BASE, 0);
5124 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
5125 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
5127 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
5128 vmcs_writel(GUEST_LDTR_BASE, 0);
5129 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
5130 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
5133 vmcs_write32(GUEST_SYSENTER_CS, 0);
5134 vmcs_writel(GUEST_SYSENTER_ESP, 0);
5135 vmcs_writel(GUEST_SYSENTER_EIP, 0);
5136 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
5139 vmcs_writel(GUEST_RFLAGS, 0x02);
5140 kvm_rip_write(vcpu, 0xfff0);
5142 vmcs_writel(GUEST_GDTR_BASE, 0);
5143 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
5145 vmcs_writel(GUEST_IDTR_BASE, 0);
5146 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
5148 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
5149 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
5150 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
5154 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
5156 if (cpu_has_vmx_tpr_shadow() && !init_event) {
5157 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
5158 if (cpu_need_tpr_shadow(vcpu))
5159 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
5160 __pa(vcpu->arch.apic->regs));
5161 vmcs_write32(TPR_THRESHOLD, 0);
5164 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
5166 if (kvm_vcpu_apicv_active(vcpu))
5167 memset(&vmx->pi_desc, 0, sizeof(struct pi_desc));
5170 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
5172 cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
5173 vmx->vcpu.arch.cr0 = cr0;
5174 vmx_set_cr0(vcpu, cr0); /* enter rmode */
5175 vmx_set_cr4(vcpu, 0);
5176 vmx_set_efer(vcpu, 0);
5177 vmx_fpu_activate(vcpu);
5178 update_exception_bitmap(vcpu);
5180 vpid_sync_context(vmx->vpid);
5184 * In nested virtualization, check if L1 asked to exit on external interrupts.
5185 * For most existing hypervisors, this will always return true.
5187 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
5189 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5190 PIN_BASED_EXT_INTR_MASK;
5194 * In nested virtualization, check if L1 has set
5195 * VM_EXIT_ACK_INTR_ON_EXIT
5197 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
5199 return get_vmcs12(vcpu)->vm_exit_controls &
5200 VM_EXIT_ACK_INTR_ON_EXIT;
5203 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
5205 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5206 PIN_BASED_NMI_EXITING;
5209 static void enable_irq_window(struct kvm_vcpu *vcpu)
5211 u32 cpu_based_vm_exec_control;
5213 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5214 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
5215 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5218 static void enable_nmi_window(struct kvm_vcpu *vcpu)
5220 u32 cpu_based_vm_exec_control;
5222 if (!cpu_has_virtual_nmis() ||
5223 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
5224 enable_irq_window(vcpu);
5228 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5229 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
5230 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5233 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
5235 struct vcpu_vmx *vmx = to_vmx(vcpu);
5237 int irq = vcpu->arch.interrupt.nr;
5239 trace_kvm_inj_virq(irq);
5241 ++vcpu->stat.irq_injections;
5242 if (vmx->rmode.vm86_active) {
5244 if (vcpu->arch.interrupt.soft)
5245 inc_eip = vcpu->arch.event_exit_inst_len;
5246 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
5247 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5250 intr = irq | INTR_INFO_VALID_MASK;
5251 if (vcpu->arch.interrupt.soft) {
5252 intr |= INTR_TYPE_SOFT_INTR;
5253 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
5254 vmx->vcpu.arch.event_exit_inst_len);
5256 intr |= INTR_TYPE_EXT_INTR;
5257 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
5260 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
5262 struct vcpu_vmx *vmx = to_vmx(vcpu);
5264 if (is_guest_mode(vcpu))
5267 if (!cpu_has_virtual_nmis()) {
5269 * Tracking the NMI-blocked state in software is built upon
5270 * finding the next open IRQ window. This, in turn, depends on
5271 * well-behaving guests: They have to keep IRQs disabled at
5272 * least as long as the NMI handler runs. Otherwise we may
5273 * cause NMI nesting, maybe breaking the guest. But as this is
5274 * highly unlikely, we can live with the residual risk.
5276 vmx->soft_vnmi_blocked = 1;
5277 vmx->vnmi_blocked_time = 0;
5280 ++vcpu->stat.nmi_injections;
5281 vmx->nmi_known_unmasked = false;
5282 if (vmx->rmode.vm86_active) {
5283 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
5284 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5287 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
5288 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
5291 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
5293 if (!cpu_has_virtual_nmis())
5294 return to_vmx(vcpu)->soft_vnmi_blocked;
5295 if (to_vmx(vcpu)->nmi_known_unmasked)
5297 return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
5300 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
5302 struct vcpu_vmx *vmx = to_vmx(vcpu);
5304 if (!cpu_has_virtual_nmis()) {
5305 if (vmx->soft_vnmi_blocked != masked) {
5306 vmx->soft_vnmi_blocked = masked;
5307 vmx->vnmi_blocked_time = 0;
5310 vmx->nmi_known_unmasked = !masked;
5312 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5313 GUEST_INTR_STATE_NMI);
5315 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
5316 GUEST_INTR_STATE_NMI);
5320 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
5322 if (to_vmx(vcpu)->nested.nested_run_pending)
5325 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
5328 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5329 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
5330 | GUEST_INTR_STATE_NMI));
5333 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
5335 return (!to_vmx(vcpu)->nested.nested_run_pending &&
5336 vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
5337 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5338 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
5341 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
5345 ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
5349 kvm->arch.tss_addr = addr;
5350 return init_rmode_tss(kvm);
5353 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
5358 * Update instruction length as we may reinject the exception
5359 * from user space while in guest debugging mode.
5361 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
5362 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5363 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5367 if (vcpu->guest_debug &
5368 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5385 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
5386 int vec, u32 err_code)
5389 * Instruction with address size override prefix opcode 0x67
5390 * Cause the #SS fault with 0 error code in VM86 mode.
5392 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
5393 if (emulate_instruction(vcpu, 0) == EMULATE_DONE) {
5394 if (vcpu->arch.halt_request) {
5395 vcpu->arch.halt_request = 0;
5396 return kvm_vcpu_halt(vcpu);
5404 * Forward all other exceptions that are valid in real mode.
5405 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
5406 * the required debugging infrastructure rework.
5408 kvm_queue_exception(vcpu, vec);
5413 * Trigger machine check on the host. We assume all the MSRs are already set up
5414 * by the CPU and that we still run on the same CPU as the MCE occurred on.
5415 * We pass a fake environment to the machine check handler because we want
5416 * the guest to be always treated like user space, no matter what context
5417 * it used internally.
5419 static void kvm_machine_check(void)
5421 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
5422 struct pt_regs regs = {
5423 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
5424 .flags = X86_EFLAGS_IF,
5427 do_machine_check(®s, 0);
5431 static int handle_machine_check(struct kvm_vcpu *vcpu)
5433 /* already handled by vcpu_run */
5437 static int handle_exception(struct kvm_vcpu *vcpu)
5439 struct vcpu_vmx *vmx = to_vmx(vcpu);
5440 struct kvm_run *kvm_run = vcpu->run;
5441 u32 intr_info, ex_no, error_code;
5442 unsigned long cr2, rip, dr6;
5444 enum emulation_result er;
5446 vect_info = vmx->idt_vectoring_info;
5447 intr_info = vmx->exit_intr_info;
5449 if (is_machine_check(intr_info))
5450 return handle_machine_check(vcpu);
5452 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
5453 return 1; /* already handled by vmx_vcpu_run() */
5455 if (is_no_device(intr_info)) {
5456 vmx_fpu_activate(vcpu);
5460 if (is_invalid_opcode(intr_info)) {
5461 if (is_guest_mode(vcpu)) {
5462 kvm_queue_exception(vcpu, UD_VECTOR);
5465 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
5466 if (er != EMULATE_DONE)
5467 kvm_queue_exception(vcpu, UD_VECTOR);
5472 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
5473 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
5476 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
5477 * MMIO, it is better to report an internal error.
5478 * See the comments in vmx_handle_exit.
5480 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
5481 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
5482 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5483 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
5484 vcpu->run->internal.ndata = 3;
5485 vcpu->run->internal.data[0] = vect_info;
5486 vcpu->run->internal.data[1] = intr_info;
5487 vcpu->run->internal.data[2] = error_code;
5491 if (is_page_fault(intr_info)) {
5492 /* EPT won't cause page fault directly */
5494 cr2 = vmcs_readl(EXIT_QUALIFICATION);
5495 trace_kvm_page_fault(cr2, error_code);
5497 if (kvm_event_needs_reinjection(vcpu))
5498 kvm_mmu_unprotect_page_virt(vcpu, cr2);
5499 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
5502 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
5504 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
5505 return handle_rmode_exception(vcpu, ex_no, error_code);
5509 kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
5512 dr6 = vmcs_readl(EXIT_QUALIFICATION);
5513 if (!(vcpu->guest_debug &
5514 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
5515 vcpu->arch.dr6 &= ~15;
5516 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5517 if (!(dr6 & ~DR6_RESERVED)) /* icebp */
5518 skip_emulated_instruction(vcpu);
5520 kvm_queue_exception(vcpu, DB_VECTOR);
5523 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
5524 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
5528 * Update instruction length as we may reinject #BP from
5529 * user space while in guest debugging mode. Reading it for
5530 * #DB as well causes no harm, it is not used in that case.
5532 vmx->vcpu.arch.event_exit_inst_len =
5533 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5534 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5535 rip = kvm_rip_read(vcpu);
5536 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
5537 kvm_run->debug.arch.exception = ex_no;
5540 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
5541 kvm_run->ex.exception = ex_no;
5542 kvm_run->ex.error_code = error_code;
5548 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
5550 ++vcpu->stat.irq_exits;
5554 static int handle_triple_fault(struct kvm_vcpu *vcpu)
5556 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5560 static int handle_io(struct kvm_vcpu *vcpu)
5562 unsigned long exit_qualification;
5563 int size, in, string;
5566 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5567 string = (exit_qualification & 16) != 0;
5568 in = (exit_qualification & 8) != 0;
5570 ++vcpu->stat.io_exits;
5573 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5575 port = exit_qualification >> 16;
5576 size = (exit_qualification & 7) + 1;
5577 skip_emulated_instruction(vcpu);
5579 return kvm_fast_pio_out(vcpu, size, port);
5583 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
5586 * Patch in the VMCALL instruction:
5588 hypercall[0] = 0x0f;
5589 hypercall[1] = 0x01;
5590 hypercall[2] = 0xc1;
5593 static bool nested_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
5595 unsigned long always_on = VMXON_CR0_ALWAYSON;
5596 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5598 if (to_vmx(vcpu)->nested.nested_vmx_secondary_ctls_high &
5599 SECONDARY_EXEC_UNRESTRICTED_GUEST &&
5600 nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
5601 always_on &= ~(X86_CR0_PE | X86_CR0_PG);
5602 return (val & always_on) == always_on;
5605 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
5606 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
5608 if (is_guest_mode(vcpu)) {
5609 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5610 unsigned long orig_val = val;
5613 * We get here when L2 changed cr0 in a way that did not change
5614 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
5615 * but did change L0 shadowed bits. So we first calculate the
5616 * effective cr0 value that L1 would like to write into the
5617 * hardware. It consists of the L2-owned bits from the new
5618 * value combined with the L1-owned bits from L1's guest_cr0.
5620 val = (val & ~vmcs12->cr0_guest_host_mask) |
5621 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
5623 if (!nested_cr0_valid(vcpu, val))
5626 if (kvm_set_cr0(vcpu, val))
5628 vmcs_writel(CR0_READ_SHADOW, orig_val);
5631 if (to_vmx(vcpu)->nested.vmxon &&
5632 ((val & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON))
5634 return kvm_set_cr0(vcpu, val);
5638 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
5640 if (is_guest_mode(vcpu)) {
5641 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5642 unsigned long orig_val = val;
5644 /* analogously to handle_set_cr0 */
5645 val = (val & ~vmcs12->cr4_guest_host_mask) |
5646 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
5647 if (kvm_set_cr4(vcpu, val))
5649 vmcs_writel(CR4_READ_SHADOW, orig_val);
5652 return kvm_set_cr4(vcpu, val);
5655 /* called to set cr0 as appropriate for clts instruction exit. */
5656 static void handle_clts(struct kvm_vcpu *vcpu)
5658 if (is_guest_mode(vcpu)) {
5660 * We get here when L2 did CLTS, and L1 didn't shadow CR0.TS
5661 * but we did (!fpu_active). We need to keep GUEST_CR0.TS on,
5662 * just pretend it's off (also in arch.cr0 for fpu_activate).
5664 vmcs_writel(CR0_READ_SHADOW,
5665 vmcs_readl(CR0_READ_SHADOW) & ~X86_CR0_TS);
5666 vcpu->arch.cr0 &= ~X86_CR0_TS;
5668 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
5671 static int handle_cr(struct kvm_vcpu *vcpu)
5673 unsigned long exit_qualification, val;
5678 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5679 cr = exit_qualification & 15;
5680 reg = (exit_qualification >> 8) & 15;
5681 switch ((exit_qualification >> 4) & 3) {
5682 case 0: /* mov to cr */
5683 val = kvm_register_readl(vcpu, reg);
5684 trace_kvm_cr_write(cr, val);
5687 err = handle_set_cr0(vcpu, val);
5688 kvm_complete_insn_gp(vcpu, err);
5691 err = kvm_set_cr3(vcpu, val);
5692 kvm_complete_insn_gp(vcpu, err);
5695 err = handle_set_cr4(vcpu, val);
5696 kvm_complete_insn_gp(vcpu, err);
5699 u8 cr8_prev = kvm_get_cr8(vcpu);
5701 err = kvm_set_cr8(vcpu, cr8);
5702 kvm_complete_insn_gp(vcpu, err);
5703 if (lapic_in_kernel(vcpu))
5705 if (cr8_prev <= cr8)
5707 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
5714 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
5715 skip_emulated_instruction(vcpu);
5716 vmx_fpu_activate(vcpu);
5718 case 1: /*mov from cr*/
5721 val = kvm_read_cr3(vcpu);
5722 kvm_register_write(vcpu, reg, val);
5723 trace_kvm_cr_read(cr, val);
5724 skip_emulated_instruction(vcpu);
5727 val = kvm_get_cr8(vcpu);
5728 kvm_register_write(vcpu, reg, val);
5729 trace_kvm_cr_read(cr, val);
5730 skip_emulated_instruction(vcpu);
5735 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5736 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
5737 kvm_lmsw(vcpu, val);
5739 skip_emulated_instruction(vcpu);
5744 vcpu->run->exit_reason = 0;
5745 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
5746 (int)(exit_qualification >> 4) & 3, cr);
5750 static int handle_dr(struct kvm_vcpu *vcpu)
5752 unsigned long exit_qualification;
5755 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5756 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
5758 /* First, if DR does not exist, trigger UD */
5759 if (!kvm_require_dr(vcpu, dr))
5762 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
5763 if (!kvm_require_cpl(vcpu, 0))
5765 dr7 = vmcs_readl(GUEST_DR7);
5768 * As the vm-exit takes precedence over the debug trap, we
5769 * need to emulate the latter, either for the host or the
5770 * guest debugging itself.
5772 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5773 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
5774 vcpu->run->debug.arch.dr7 = dr7;
5775 vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
5776 vcpu->run->debug.arch.exception = DB_VECTOR;
5777 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
5780 vcpu->arch.dr6 &= ~15;
5781 vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
5782 kvm_queue_exception(vcpu, DB_VECTOR);
5787 if (vcpu->guest_debug == 0) {
5788 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
5789 CPU_BASED_MOV_DR_EXITING);
5792 * No more DR vmexits; force a reload of the debug registers
5793 * and reenter on this instruction. The next vmexit will
5794 * retrieve the full state of the debug registers.
5796 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
5800 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
5801 if (exit_qualification & TYPE_MOV_FROM_DR) {
5804 if (kvm_get_dr(vcpu, dr, &val))
5806 kvm_register_write(vcpu, reg, val);
5808 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
5811 skip_emulated_instruction(vcpu);
5815 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
5817 return vcpu->arch.dr6;
5820 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
5824 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
5826 get_debugreg(vcpu->arch.db[0], 0);
5827 get_debugreg(vcpu->arch.db[1], 1);
5828 get_debugreg(vcpu->arch.db[2], 2);
5829 get_debugreg(vcpu->arch.db[3], 3);
5830 get_debugreg(vcpu->arch.dr6, 6);
5831 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
5833 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
5834 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL, CPU_BASED_MOV_DR_EXITING);
5837 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
5839 vmcs_writel(GUEST_DR7, val);
5842 static int handle_cpuid(struct kvm_vcpu *vcpu)
5844 kvm_emulate_cpuid(vcpu);
5848 static int handle_rdmsr(struct kvm_vcpu *vcpu)
5850 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5851 struct msr_data msr_info;
5853 msr_info.index = ecx;
5854 msr_info.host_initiated = false;
5855 if (vmx_get_msr(vcpu, &msr_info)) {
5856 trace_kvm_msr_read_ex(ecx);
5857 kvm_inject_gp(vcpu, 0);
5861 trace_kvm_msr_read(ecx, msr_info.data);
5863 /* FIXME: handling of bits 32:63 of rax, rdx */
5864 vcpu->arch.regs[VCPU_REGS_RAX] = msr_info.data & -1u;
5865 vcpu->arch.regs[VCPU_REGS_RDX] = (msr_info.data >> 32) & -1u;
5866 skip_emulated_instruction(vcpu);
5870 static int handle_wrmsr(struct kvm_vcpu *vcpu)
5872 struct msr_data msr;
5873 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5874 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
5875 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
5879 msr.host_initiated = false;
5880 if (kvm_set_msr(vcpu, &msr) != 0) {
5881 trace_kvm_msr_write_ex(ecx, data);
5882 kvm_inject_gp(vcpu, 0);
5886 trace_kvm_msr_write(ecx, data);
5887 skip_emulated_instruction(vcpu);
5891 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
5893 kvm_make_request(KVM_REQ_EVENT, vcpu);
5897 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
5899 u32 cpu_based_vm_exec_control;
5901 /* clear pending irq */
5902 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5903 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
5904 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5906 kvm_make_request(KVM_REQ_EVENT, vcpu);
5908 ++vcpu->stat.irq_window_exits;
5912 static int handle_halt(struct kvm_vcpu *vcpu)
5914 return kvm_emulate_halt(vcpu);
5917 static int handle_vmcall(struct kvm_vcpu *vcpu)
5919 return kvm_emulate_hypercall(vcpu);
5922 static int handle_invd(struct kvm_vcpu *vcpu)
5924 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5927 static int handle_invlpg(struct kvm_vcpu *vcpu)
5929 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5931 kvm_mmu_invlpg(vcpu, exit_qualification);
5932 skip_emulated_instruction(vcpu);
5936 static int handle_rdpmc(struct kvm_vcpu *vcpu)
5940 err = kvm_rdpmc(vcpu);
5941 kvm_complete_insn_gp(vcpu, err);
5946 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5948 kvm_emulate_wbinvd(vcpu);
5952 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5954 u64 new_bv = kvm_read_edx_eax(vcpu);
5955 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5957 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5958 skip_emulated_instruction(vcpu);
5962 static int handle_xsaves(struct kvm_vcpu *vcpu)
5964 skip_emulated_instruction(vcpu);
5965 WARN(1, "this should never happen\n");
5969 static int handle_xrstors(struct kvm_vcpu *vcpu)
5971 skip_emulated_instruction(vcpu);
5972 WARN(1, "this should never happen\n");
5976 static int handle_apic_access(struct kvm_vcpu *vcpu)
5978 if (likely(fasteoi)) {
5979 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5980 int access_type, offset;
5982 access_type = exit_qualification & APIC_ACCESS_TYPE;
5983 offset = exit_qualification & APIC_ACCESS_OFFSET;
5985 * Sane guest uses MOV to write EOI, with written value
5986 * not cared. So make a short-circuit here by avoiding
5987 * heavy instruction emulation.
5989 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5990 (offset == APIC_EOI)) {
5991 kvm_lapic_set_eoi(vcpu);
5992 skip_emulated_instruction(vcpu);
5996 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5999 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
6001 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6002 int vector = exit_qualification & 0xff;
6004 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
6005 kvm_apic_set_eoi_accelerated(vcpu, vector);
6009 static int handle_apic_write(struct kvm_vcpu *vcpu)
6011 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6012 u32 offset = exit_qualification & 0xfff;
6014 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
6015 kvm_apic_write_nodecode(vcpu, offset);
6019 static int handle_task_switch(struct kvm_vcpu *vcpu)
6021 struct vcpu_vmx *vmx = to_vmx(vcpu);
6022 unsigned long exit_qualification;
6023 bool has_error_code = false;
6026 int reason, type, idt_v, idt_index;
6028 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
6029 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
6030 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
6032 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6034 reason = (u32)exit_qualification >> 30;
6035 if (reason == TASK_SWITCH_GATE && idt_v) {
6037 case INTR_TYPE_NMI_INTR:
6038 vcpu->arch.nmi_injected = false;
6039 vmx_set_nmi_mask(vcpu, true);
6041 case INTR_TYPE_EXT_INTR:
6042 case INTR_TYPE_SOFT_INTR:
6043 kvm_clear_interrupt_queue(vcpu);
6045 case INTR_TYPE_HARD_EXCEPTION:
6046 if (vmx->idt_vectoring_info &
6047 VECTORING_INFO_DELIVER_CODE_MASK) {
6048 has_error_code = true;
6050 vmcs_read32(IDT_VECTORING_ERROR_CODE);
6053 case INTR_TYPE_SOFT_EXCEPTION:
6054 kvm_clear_exception_queue(vcpu);
6060 tss_selector = exit_qualification;
6062 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
6063 type != INTR_TYPE_EXT_INTR &&
6064 type != INTR_TYPE_NMI_INTR))
6065 skip_emulated_instruction(vcpu);
6067 if (kvm_task_switch(vcpu, tss_selector,
6068 type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
6069 has_error_code, error_code) == EMULATE_FAIL) {
6070 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6071 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6072 vcpu->run->internal.ndata = 0;
6077 * TODO: What about debug traps on tss switch?
6078 * Are we supposed to inject them and update dr6?
6084 static int handle_ept_violation(struct kvm_vcpu *vcpu)
6086 unsigned long exit_qualification;
6091 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6093 gla_validity = (exit_qualification >> 7) & 0x3;
6094 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
6095 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
6096 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
6097 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
6098 vmcs_readl(GUEST_LINEAR_ADDRESS));
6099 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
6100 (long unsigned int)exit_qualification);
6101 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
6102 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
6107 * EPT violation happened while executing iret from NMI,
6108 * "blocked by NMI" bit has to be set before next VM entry.
6109 * There are errata that may cause this bit to not be set:
6112 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
6113 cpu_has_virtual_nmis() &&
6114 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
6115 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
6117 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6118 trace_kvm_page_fault(gpa, exit_qualification);
6120 /* It is a write fault? */
6121 error_code = exit_qualification & PFERR_WRITE_MASK;
6122 /* It is a fetch fault? */
6123 error_code |= (exit_qualification << 2) & PFERR_FETCH_MASK;
6124 /* ept page table is present? */
6125 error_code |= (exit_qualification >> 3) & PFERR_PRESENT_MASK;
6127 vcpu->arch.exit_qualification = exit_qualification;
6129 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
6132 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
6137 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6138 if (!kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
6139 skip_emulated_instruction(vcpu);
6140 trace_kvm_fast_mmio(gpa);
6144 ret = handle_mmio_page_fault(vcpu, gpa, true);
6145 if (likely(ret == RET_MMIO_PF_EMULATE))
6146 return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) ==
6149 if (unlikely(ret == RET_MMIO_PF_INVALID))
6150 return kvm_mmu_page_fault(vcpu, gpa, 0, NULL, 0);
6152 if (unlikely(ret == RET_MMIO_PF_RETRY))
6155 /* It is the real ept misconfig */
6158 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
6159 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
6164 static int handle_nmi_window(struct kvm_vcpu *vcpu)
6166 u32 cpu_based_vm_exec_control;
6168 /* clear pending NMI */
6169 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
6170 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
6171 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
6172 ++vcpu->stat.nmi_window_exits;
6173 kvm_make_request(KVM_REQ_EVENT, vcpu);
6178 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
6180 struct vcpu_vmx *vmx = to_vmx(vcpu);
6181 enum emulation_result err = EMULATE_DONE;
6184 bool intr_window_requested;
6185 unsigned count = 130;
6187 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
6188 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
6190 while (vmx->emulation_required && count-- != 0) {
6191 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
6192 return handle_interrupt_window(&vmx->vcpu);
6194 if (test_bit(KVM_REQ_EVENT, &vcpu->requests))
6197 err = emulate_instruction(vcpu, EMULTYPE_NO_REEXECUTE);
6199 if (err == EMULATE_USER_EXIT) {
6200 ++vcpu->stat.mmio_exits;
6205 if (err != EMULATE_DONE) {
6206 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6207 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6208 vcpu->run->internal.ndata = 0;
6212 if (vcpu->arch.halt_request) {
6213 vcpu->arch.halt_request = 0;
6214 ret = kvm_vcpu_halt(vcpu);
6218 if (signal_pending(current))
6228 static int __grow_ple_window(int val)
6230 if (ple_window_grow < 1)
6233 val = min(val, ple_window_actual_max);
6235 if (ple_window_grow < ple_window)
6236 val *= ple_window_grow;
6238 val += ple_window_grow;
6243 static int __shrink_ple_window(int val, int modifier, int minimum)
6248 if (modifier < ple_window)
6253 return max(val, minimum);
6256 static void grow_ple_window(struct kvm_vcpu *vcpu)
6258 struct vcpu_vmx *vmx = to_vmx(vcpu);
6259 int old = vmx->ple_window;
6261 vmx->ple_window = __grow_ple_window(old);
6263 if (vmx->ple_window != old)
6264 vmx->ple_window_dirty = true;
6266 trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old);
6269 static void shrink_ple_window(struct kvm_vcpu *vcpu)
6271 struct vcpu_vmx *vmx = to_vmx(vcpu);
6272 int old = vmx->ple_window;
6274 vmx->ple_window = __shrink_ple_window(old,
6275 ple_window_shrink, ple_window);
6277 if (vmx->ple_window != old)
6278 vmx->ple_window_dirty = true;
6280 trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old);
6284 * ple_window_actual_max is computed to be one grow_ple_window() below
6285 * ple_window_max. (See __grow_ple_window for the reason.)
6286 * This prevents overflows, because ple_window_max is int.
6287 * ple_window_max effectively rounded down to a multiple of ple_window_grow in
6289 * ple_window_max is also prevented from setting vmx->ple_window < ple_window.
6291 static void update_ple_window_actual_max(void)
6293 ple_window_actual_max =
6294 __shrink_ple_window(max(ple_window_max, ple_window),
6295 ple_window_grow, INT_MIN);
6299 * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
6301 static void wakeup_handler(void)
6303 struct kvm_vcpu *vcpu;
6304 int cpu = smp_processor_id();
6306 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6307 list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
6308 blocked_vcpu_list) {
6309 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6311 if (pi_test_on(pi_desc) == 1)
6312 kvm_vcpu_kick(vcpu);
6314 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6317 static __init int hardware_setup(void)
6319 int r = -ENOMEM, i, msr;
6321 rdmsrl_safe(MSR_EFER, &host_efer);
6323 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
6324 kvm_define_shared_msr(i, vmx_msr_index[i]);
6326 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
6327 if (!vmx_io_bitmap_a)
6330 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
6331 if (!vmx_io_bitmap_b)
6334 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
6335 if (!vmx_msr_bitmap_legacy)
6338 vmx_msr_bitmap_legacy_x2apic =
6339 (unsigned long *)__get_free_page(GFP_KERNEL);
6340 if (!vmx_msr_bitmap_legacy_x2apic)
6343 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
6344 if (!vmx_msr_bitmap_longmode)
6347 vmx_msr_bitmap_longmode_x2apic =
6348 (unsigned long *)__get_free_page(GFP_KERNEL);
6349 if (!vmx_msr_bitmap_longmode_x2apic)
6353 vmx_msr_bitmap_nested =
6354 (unsigned long *)__get_free_page(GFP_KERNEL);
6355 if (!vmx_msr_bitmap_nested)
6359 vmx_vmread_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
6360 if (!vmx_vmread_bitmap)
6363 vmx_vmwrite_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
6364 if (!vmx_vmwrite_bitmap)
6367 memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
6368 memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
6371 * Allow direct access to the PC debug port (it is often used for I/O
6372 * delays, but the vmexits simply slow things down).
6374 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
6375 clear_bit(0x80, vmx_io_bitmap_a);
6377 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
6379 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
6380 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
6382 memset(vmx_msr_bitmap_nested, 0xff, PAGE_SIZE);
6384 if (setup_vmcs_config(&vmcs_config) < 0) {
6389 if (boot_cpu_has(X86_FEATURE_NX))
6390 kvm_enable_efer_bits(EFER_NX);
6392 if (!cpu_has_vmx_vpid())
6394 if (!cpu_has_vmx_shadow_vmcs())
6395 enable_shadow_vmcs = 0;
6396 if (enable_shadow_vmcs)
6397 init_vmcs_shadow_fields();
6399 if (!cpu_has_vmx_ept() ||
6400 !cpu_has_vmx_ept_4levels()) {
6402 enable_unrestricted_guest = 0;
6403 enable_ept_ad_bits = 0;
6406 if (!cpu_has_vmx_ept_ad_bits())
6407 enable_ept_ad_bits = 0;
6409 if (!cpu_has_vmx_unrestricted_guest())
6410 enable_unrestricted_guest = 0;
6412 if (!cpu_has_vmx_flexpriority())
6413 flexpriority_enabled = 0;
6416 * set_apic_access_page_addr() is used to reload apic access
6417 * page upon invalidation. No need to do anything if not
6418 * using the APIC_ACCESS_ADDR VMCS field.
6420 if (!flexpriority_enabled)
6421 kvm_x86_ops->set_apic_access_page_addr = NULL;
6423 if (!cpu_has_vmx_tpr_shadow())
6424 kvm_x86_ops->update_cr8_intercept = NULL;
6426 if (enable_ept && !cpu_has_vmx_ept_2m_page())
6427 kvm_disable_largepages();
6429 if (!cpu_has_vmx_ple())
6432 if (!cpu_has_vmx_apicv())
6435 if (cpu_has_vmx_tsc_scaling()) {
6436 kvm_has_tsc_control = true;
6437 kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
6438 kvm_tsc_scaling_ratio_frac_bits = 48;
6441 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
6442 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
6443 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
6444 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
6445 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
6446 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
6447 vmx_disable_intercept_for_msr(MSR_IA32_BNDCFGS, true);
6449 memcpy(vmx_msr_bitmap_legacy_x2apic,
6450 vmx_msr_bitmap_legacy, PAGE_SIZE);
6451 memcpy(vmx_msr_bitmap_longmode_x2apic,
6452 vmx_msr_bitmap_longmode, PAGE_SIZE);
6454 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
6456 for (msr = 0x800; msr <= 0x8ff; msr++)
6457 vmx_disable_intercept_msr_read_x2apic(msr);
6459 /* According SDM, in x2apic mode, the whole id reg is used. But in
6460 * KVM, it only use the highest eight bits. Need to intercept it */
6461 vmx_enable_intercept_msr_read_x2apic(0x802);
6463 vmx_enable_intercept_msr_read_x2apic(0x839);
6465 vmx_disable_intercept_msr_write_x2apic(0x808);
6467 vmx_disable_intercept_msr_write_x2apic(0x80b);
6469 vmx_disable_intercept_msr_write_x2apic(0x83f);
6472 kvm_mmu_set_mask_ptes(0ull,
6473 (enable_ept_ad_bits) ? VMX_EPT_ACCESS_BIT : 0ull,
6474 (enable_ept_ad_bits) ? VMX_EPT_DIRTY_BIT : 0ull,
6475 0ull, VMX_EPT_EXECUTABLE_MASK);
6476 0ull, VMX_EPT_EXECUTABLE_MASK, VMX_EPT_READABLE_MASK);
6477 ept_set_mmio_spte_mask();
6482 update_ple_window_actual_max();
6485 * Only enable PML when hardware supports PML feature, and both EPT
6486 * and EPT A/D bit features are enabled -- PML depends on them to work.
6488 if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
6492 kvm_x86_ops->slot_enable_log_dirty = NULL;
6493 kvm_x86_ops->slot_disable_log_dirty = NULL;
6494 kvm_x86_ops->flush_log_dirty = NULL;
6495 kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
6498 if (cpu_has_vmx_preemption_timer() && enable_preemption_timer) {
6501 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
6502 cpu_preemption_timer_multi =
6503 vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
6505 kvm_x86_ops->set_hv_timer = NULL;
6506 kvm_x86_ops->cancel_hv_timer = NULL;
6509 kvm_set_posted_intr_wakeup_handler(wakeup_handler);
6511 kvm_mce_cap_supported |= MCG_LMCE_P;
6513 return alloc_kvm_area();
6516 free_page((unsigned long)vmx_vmwrite_bitmap);
6518 free_page((unsigned long)vmx_vmread_bitmap);
6521 free_page((unsigned long)vmx_msr_bitmap_nested);
6523 free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
6525 free_page((unsigned long)vmx_msr_bitmap_longmode);
6527 free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
6529 free_page((unsigned long)vmx_msr_bitmap_legacy);
6531 free_page((unsigned long)vmx_io_bitmap_b);
6533 free_page((unsigned long)vmx_io_bitmap_a);
6538 static __exit void hardware_unsetup(void)
6540 free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
6541 free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
6542 free_page((unsigned long)vmx_msr_bitmap_legacy);
6543 free_page((unsigned long)vmx_msr_bitmap_longmode);
6544 free_page((unsigned long)vmx_io_bitmap_b);
6545 free_page((unsigned long)vmx_io_bitmap_a);
6546 free_page((unsigned long)vmx_vmwrite_bitmap);
6547 free_page((unsigned long)vmx_vmread_bitmap);
6549 free_page((unsigned long)vmx_msr_bitmap_nested);
6555 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
6556 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
6558 static int handle_pause(struct kvm_vcpu *vcpu)
6561 grow_ple_window(vcpu);
6563 skip_emulated_instruction(vcpu);
6564 kvm_vcpu_on_spin(vcpu);
6569 static int handle_nop(struct kvm_vcpu *vcpu)
6571 skip_emulated_instruction(vcpu);
6575 static int handle_mwait(struct kvm_vcpu *vcpu)
6577 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
6578 return handle_nop(vcpu);
6581 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
6586 static int handle_monitor(struct kvm_vcpu *vcpu)
6588 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
6589 return handle_nop(vcpu);
6593 * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
6594 * We could reuse a single VMCS for all the L2 guests, but we also want the
6595 * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
6596 * allows keeping them loaded on the processor, and in the future will allow
6597 * optimizations where prepare_vmcs02 doesn't need to set all the fields on
6598 * every entry if they never change.
6599 * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
6600 * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
6602 * The following functions allocate and free a vmcs02 in this pool.
6605 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
6606 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
6608 struct vmcs02_list *item;
6609 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6610 if (item->vmptr == vmx->nested.current_vmptr) {
6611 list_move(&item->list, &vmx->nested.vmcs02_pool);
6612 return &item->vmcs02;
6615 if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
6616 /* Recycle the least recently used VMCS. */
6617 item = list_last_entry(&vmx->nested.vmcs02_pool,
6618 struct vmcs02_list, list);
6619 item->vmptr = vmx->nested.current_vmptr;
6620 list_move(&item->list, &vmx->nested.vmcs02_pool);
6621 return &item->vmcs02;
6624 /* Create a new VMCS */
6625 item = kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
6628 item->vmcs02.vmcs = alloc_vmcs();
6629 if (!item->vmcs02.vmcs) {
6633 loaded_vmcs_init(&item->vmcs02);
6634 item->vmptr = vmx->nested.current_vmptr;
6635 list_add(&(item->list), &(vmx->nested.vmcs02_pool));
6636 vmx->nested.vmcs02_num++;
6637 return &item->vmcs02;
6640 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
6641 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
6643 struct vmcs02_list *item;
6644 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6645 if (item->vmptr == vmptr) {
6646 free_loaded_vmcs(&item->vmcs02);
6647 list_del(&item->list);
6649 vmx->nested.vmcs02_num--;
6655 * Free all VMCSs saved for this vcpu, except the one pointed by
6656 * vmx->loaded_vmcs. We must be running L1, so vmx->loaded_vmcs
6657 * must be &vmx->vmcs01.
6659 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
6661 struct vmcs02_list *item, *n;
6663 WARN_ON(vmx->loaded_vmcs != &vmx->vmcs01);
6664 list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
6666 * Something will leak if the above WARN triggers. Better than
6669 if (vmx->loaded_vmcs == &item->vmcs02)
6672 free_loaded_vmcs(&item->vmcs02);
6673 list_del(&item->list);
6675 vmx->nested.vmcs02_num--;
6680 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
6681 * set the success or error code of an emulated VMX instruction, as specified
6682 * by Vol 2B, VMX Instruction Reference, "Conventions".
6684 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
6686 vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
6687 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6688 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
6691 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
6693 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6694 & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
6695 X86_EFLAGS_SF | X86_EFLAGS_OF))
6699 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
6700 u32 vm_instruction_error)
6702 if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
6704 * failValid writes the error number to the current VMCS, which
6705 * can't be done there isn't a current VMCS.
6707 nested_vmx_failInvalid(vcpu);
6710 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6711 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6712 X86_EFLAGS_SF | X86_EFLAGS_OF))
6714 get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
6716 * We don't need to force a shadow sync because
6717 * VM_INSTRUCTION_ERROR is not shadowed
6721 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
6723 /* TODO: not to reset guest simply here. */
6724 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6725 pr_warn("kvm: nested vmx abort, indicator %d\n", indicator);
6728 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
6730 struct vcpu_vmx *vmx =
6731 container_of(timer, struct vcpu_vmx, nested.preemption_timer);
6733 vmx->nested.preemption_timer_expired = true;
6734 kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
6735 kvm_vcpu_kick(&vmx->vcpu);
6737 return HRTIMER_NORESTART;
6741 * Decode the memory-address operand of a vmx instruction, as recorded on an
6742 * exit caused by such an instruction (run by a guest hypervisor).
6743 * On success, returns 0. When the operand is invalid, returns 1 and throws
6746 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
6747 unsigned long exit_qualification,
6748 u32 vmx_instruction_info, bool wr, gva_t *ret)
6752 struct kvm_segment s;
6755 * According to Vol. 3B, "Information for VM Exits Due to Instruction
6756 * Execution", on an exit, vmx_instruction_info holds most of the
6757 * addressing components of the operand. Only the displacement part
6758 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
6759 * For how an actual address is calculated from all these components,
6760 * refer to Vol. 1, "Operand Addressing".
6762 int scaling = vmx_instruction_info & 3;
6763 int addr_size = (vmx_instruction_info >> 7) & 7;
6764 bool is_reg = vmx_instruction_info & (1u << 10);
6765 int seg_reg = (vmx_instruction_info >> 15) & 7;
6766 int index_reg = (vmx_instruction_info >> 18) & 0xf;
6767 bool index_is_valid = !(vmx_instruction_info & (1u << 22));
6768 int base_reg = (vmx_instruction_info >> 23) & 0xf;
6769 bool base_is_valid = !(vmx_instruction_info & (1u << 27));
6772 kvm_queue_exception(vcpu, UD_VECTOR);
6776 /* Addr = segment_base + offset */
6777 /* offset = base + [index * scale] + displacement */
6778 off = exit_qualification; /* holds the displacement */
6780 off += kvm_register_read(vcpu, base_reg);
6782 off += kvm_register_read(vcpu, index_reg)<<scaling;
6783 vmx_get_segment(vcpu, &s, seg_reg);
6784 *ret = s.base + off;
6786 if (addr_size == 1) /* 32 bit */
6789 /* Checks for #GP/#SS exceptions. */
6791 if (is_protmode(vcpu)) {
6792 /* Protected mode: apply checks for segment validity in the
6794 * - segment type check (#GP(0) may be thrown)
6795 * - usability check (#GP(0)/#SS(0))
6796 * - limit check (#GP(0)/#SS(0))
6799 /* #GP(0) if the destination operand is located in a
6800 * read-only data segment or any code segment.
6802 exn = ((s.type & 0xa) == 0 || (s.type & 8));
6804 /* #GP(0) if the source operand is located in an
6805 * execute-only code segment
6807 exn = ((s.type & 0xa) == 8);
6810 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6813 if (is_long_mode(vcpu)) {
6814 /* Long mode: #GP(0)/#SS(0) if the memory address is in a
6815 * non-canonical form. This is an only check for long mode.
6817 exn = is_noncanonical_address(*ret);
6818 } else if (is_protmode(vcpu)) {
6819 /* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
6821 exn = (s.unusable != 0);
6822 /* Protected mode: #GP(0)/#SS(0) if the memory
6823 * operand is outside the segment limit.
6825 exn = exn || (off + sizeof(u64) > s.limit);
6828 kvm_queue_exception_e(vcpu,
6829 seg_reg == VCPU_SREG_SS ?
6830 SS_VECTOR : GP_VECTOR,
6839 * This function performs the various checks including
6840 * - if it's 4KB aligned
6841 * - No bits beyond the physical address width are set
6842 * - Returns 0 on success or else 1
6843 * (Intel SDM Section 30.3)
6845 static int nested_vmx_check_vmptr(struct kvm_vcpu *vcpu, int exit_reason,
6850 struct x86_exception e;
6852 struct vcpu_vmx *vmx = to_vmx(vcpu);
6853 int maxphyaddr = cpuid_maxphyaddr(vcpu);
6855 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
6856 vmcs_read32(VMX_INSTRUCTION_INFO), false, &gva))
6859 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
6860 sizeof(vmptr), &e)) {
6861 kvm_inject_page_fault(vcpu, &e);
6865 switch (exit_reason) {
6866 case EXIT_REASON_VMON:
6869 * The first 4 bytes of VMXON region contain the supported
6870 * VMCS revision identifier
6872 * Note - IA32_VMX_BASIC[48] will never be 1
6873 * for the nested case;
6874 * which replaces physical address width with 32
6877 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6878 nested_vmx_failInvalid(vcpu);
6879 skip_emulated_instruction(vcpu);
6883 page = nested_get_page(vcpu, vmptr);
6885 *(u32 *)kmap(page) != VMCS12_REVISION) {
6886 nested_vmx_failInvalid(vcpu);
6888 skip_emulated_instruction(vcpu);
6892 vmx->nested.vmxon_ptr = vmptr;
6894 case EXIT_REASON_VMCLEAR:
6895 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6896 nested_vmx_failValid(vcpu,
6897 VMXERR_VMCLEAR_INVALID_ADDRESS);
6898 skip_emulated_instruction(vcpu);
6902 if (vmptr == vmx->nested.vmxon_ptr) {
6903 nested_vmx_failValid(vcpu,
6904 VMXERR_VMCLEAR_VMXON_POINTER);
6905 skip_emulated_instruction(vcpu);
6909 case EXIT_REASON_VMPTRLD:
6910 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6911 nested_vmx_failValid(vcpu,
6912 VMXERR_VMPTRLD_INVALID_ADDRESS);
6913 skip_emulated_instruction(vcpu);
6917 if (vmptr == vmx->nested.vmxon_ptr) {
6918 nested_vmx_failValid(vcpu,
6919 VMXERR_VMCLEAR_VMXON_POINTER);
6920 skip_emulated_instruction(vcpu);
6925 return 1; /* shouldn't happen */
6934 * Emulate the VMXON instruction.
6935 * Currently, we just remember that VMX is active, and do not save or even
6936 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
6937 * do not currently need to store anything in that guest-allocated memory
6938 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
6939 * argument is different from the VMXON pointer (which the spec says they do).
6941 static int handle_vmon(struct kvm_vcpu *vcpu)
6943 struct kvm_segment cs;
6944 struct vcpu_vmx *vmx = to_vmx(vcpu);
6945 struct vmcs *shadow_vmcs;
6946 const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
6947 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
6949 /* The Intel VMX Instruction Reference lists a bunch of bits that
6950 * are prerequisite to running VMXON, most notably cr4.VMXE must be
6951 * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
6952 * Otherwise, we should fail with #UD. We test these now:
6954 if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
6955 !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
6956 (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
6957 kvm_queue_exception(vcpu, UD_VECTOR);
6961 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
6962 if (is_long_mode(vcpu) && !cs.l) {
6963 kvm_queue_exception(vcpu, UD_VECTOR);
6967 if (vmx_get_cpl(vcpu)) {
6968 kvm_inject_gp(vcpu, 0);
6972 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMON, NULL))
6975 if (vmx->nested.vmxon) {
6976 nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
6977 skip_emulated_instruction(vcpu);
6981 if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
6982 != VMXON_NEEDED_FEATURES) {
6983 kvm_inject_gp(vcpu, 0);
6987 if (enable_shadow_vmcs) {
6988 shadow_vmcs = alloc_vmcs();
6991 /* mark vmcs as shadow */
6992 shadow_vmcs->revision_id |= (1u << 31);
6993 /* init shadow vmcs */
6994 vmcs_clear(shadow_vmcs);
6995 vmx->nested.current_shadow_vmcs = shadow_vmcs;
6998 INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
6999 vmx->nested.vmcs02_num = 0;
7001 hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
7003 vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
7005 vmx->nested.vmxon = true;
7007 skip_emulated_instruction(vcpu);
7008 nested_vmx_succeed(vcpu);
7013 * Intel's VMX Instruction Reference specifies a common set of prerequisites
7014 * for running VMX instructions (except VMXON, whose prerequisites are
7015 * slightly different). It also specifies what exception to inject otherwise.
7017 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
7019 struct kvm_segment cs;
7020 struct vcpu_vmx *vmx = to_vmx(vcpu);
7022 if (!vmx->nested.vmxon) {
7023 kvm_queue_exception(vcpu, UD_VECTOR);
7027 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
7028 if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
7029 (is_long_mode(vcpu) && !cs.l)) {
7030 kvm_queue_exception(vcpu, UD_VECTOR);
7034 if (vmx_get_cpl(vcpu)) {
7035 kvm_inject_gp(vcpu, 0);
7042 static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
7044 if (vmx->nested.current_vmptr == -1ull)
7047 /* current_vmptr and current_vmcs12 are always set/reset together */
7048 if (WARN_ON(vmx->nested.current_vmcs12 == NULL))
7051 if (enable_shadow_vmcs) {
7052 /* copy to memory all shadowed fields in case
7053 they were modified */
7054 copy_shadow_to_vmcs12(vmx);
7055 vmx->nested.sync_shadow_vmcs = false;
7056 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
7057 SECONDARY_EXEC_SHADOW_VMCS);
7058 vmcs_write64(VMCS_LINK_POINTER, -1ull);
7060 vmx->nested.posted_intr_nv = -1;
7061 kunmap(vmx->nested.current_vmcs12_page);
7062 nested_release_page(vmx->nested.current_vmcs12_page);
7063 vmx->nested.current_vmptr = -1ull;
7064 vmx->nested.current_vmcs12 = NULL;
7068 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
7069 * just stops using VMX.
7071 static void free_nested(struct vcpu_vmx *vmx)
7073 if (!vmx->nested.vmxon)
7076 vmx->nested.vmxon = false;
7077 free_vpid(vmx->nested.vpid02);
7078 nested_release_vmcs12(vmx);
7079 if (enable_shadow_vmcs)
7080 free_vmcs(vmx->nested.current_shadow_vmcs);
7081 /* Unpin physical memory we referred to in current vmcs02 */
7082 if (vmx->nested.apic_access_page) {
7083 nested_release_page(vmx->nested.apic_access_page);
7084 vmx->nested.apic_access_page = NULL;
7086 if (vmx->nested.virtual_apic_page) {
7087 nested_release_page(vmx->nested.virtual_apic_page);
7088 vmx->nested.virtual_apic_page = NULL;
7090 if (vmx->nested.pi_desc_page) {
7091 kunmap(vmx->nested.pi_desc_page);
7092 nested_release_page(vmx->nested.pi_desc_page);
7093 vmx->nested.pi_desc_page = NULL;
7094 vmx->nested.pi_desc = NULL;
7097 nested_free_all_saved_vmcss(vmx);
7100 /* Emulate the VMXOFF instruction */
7101 static int handle_vmoff(struct kvm_vcpu *vcpu)
7103 if (!nested_vmx_check_permission(vcpu))
7105 free_nested(to_vmx(vcpu));
7106 skip_emulated_instruction(vcpu);
7107 nested_vmx_succeed(vcpu);
7111 /* Emulate the VMCLEAR instruction */
7112 static int handle_vmclear(struct kvm_vcpu *vcpu)
7114 struct vcpu_vmx *vmx = to_vmx(vcpu);
7116 struct vmcs12 *vmcs12;
7119 if (!nested_vmx_check_permission(vcpu))
7122 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMCLEAR, &vmptr))
7125 if (vmptr == vmx->nested.current_vmptr)
7126 nested_release_vmcs12(vmx);
7128 page = nested_get_page(vcpu, vmptr);
7131 * For accurate processor emulation, VMCLEAR beyond available
7132 * physical memory should do nothing at all. However, it is
7133 * possible that a nested vmx bug, not a guest hypervisor bug,
7134 * resulted in this case, so let's shut down before doing any
7137 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7140 vmcs12 = kmap(page);
7141 vmcs12->launch_state = 0;
7143 nested_release_page(page);
7145 nested_free_vmcs02(vmx, vmptr);
7147 skip_emulated_instruction(vcpu);
7148 nested_vmx_succeed(vcpu);
7152 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
7154 /* Emulate the VMLAUNCH instruction */
7155 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
7157 return nested_vmx_run(vcpu, true);
7160 /* Emulate the VMRESUME instruction */
7161 static int handle_vmresume(struct kvm_vcpu *vcpu)
7164 return nested_vmx_run(vcpu, false);
7167 enum vmcs_field_type {
7168 VMCS_FIELD_TYPE_U16 = 0,
7169 VMCS_FIELD_TYPE_U64 = 1,
7170 VMCS_FIELD_TYPE_U32 = 2,
7171 VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
7174 static inline int vmcs_field_type(unsigned long field)
7176 if (0x1 & field) /* the *_HIGH fields are all 32 bit */
7177 return VMCS_FIELD_TYPE_U32;
7178 return (field >> 13) & 0x3 ;
7181 static inline int vmcs_field_readonly(unsigned long field)
7183 return (((field >> 10) & 0x3) == 1);
7187 * Read a vmcs12 field. Since these can have varying lengths and we return
7188 * one type, we chose the biggest type (u64) and zero-extend the return value
7189 * to that size. Note that the caller, handle_vmread, might need to use only
7190 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
7191 * 64-bit fields are to be returned).
7193 static inline int vmcs12_read_any(struct kvm_vcpu *vcpu,
7194 unsigned long field, u64 *ret)
7196 short offset = vmcs_field_to_offset(field);
7202 p = ((char *)(get_vmcs12(vcpu))) + offset;
7204 switch (vmcs_field_type(field)) {
7205 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7206 *ret = *((natural_width *)p);
7208 case VMCS_FIELD_TYPE_U16:
7211 case VMCS_FIELD_TYPE_U32:
7214 case VMCS_FIELD_TYPE_U64:
7224 static inline int vmcs12_write_any(struct kvm_vcpu *vcpu,
7225 unsigned long field, u64 field_value){
7226 short offset = vmcs_field_to_offset(field);
7227 char *p = ((char *) get_vmcs12(vcpu)) + offset;
7231 switch (vmcs_field_type(field)) {
7232 case VMCS_FIELD_TYPE_U16:
7233 *(u16 *)p = field_value;
7235 case VMCS_FIELD_TYPE_U32:
7236 *(u32 *)p = field_value;
7238 case VMCS_FIELD_TYPE_U64:
7239 *(u64 *)p = field_value;
7241 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7242 *(natural_width *)p = field_value;
7251 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
7254 unsigned long field;
7256 struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
7257 const unsigned long *fields = shadow_read_write_fields;
7258 const int num_fields = max_shadow_read_write_fields;
7262 vmcs_load(shadow_vmcs);
7264 for (i = 0; i < num_fields; i++) {
7266 switch (vmcs_field_type(field)) {
7267 case VMCS_FIELD_TYPE_U16:
7268 field_value = vmcs_read16(field);
7270 case VMCS_FIELD_TYPE_U32:
7271 field_value = vmcs_read32(field);
7273 case VMCS_FIELD_TYPE_U64:
7274 field_value = vmcs_read64(field);
7276 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7277 field_value = vmcs_readl(field);
7283 vmcs12_write_any(&vmx->vcpu, field, field_value);
7286 vmcs_clear(shadow_vmcs);
7287 vmcs_load(vmx->loaded_vmcs->vmcs);
7292 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
7294 const unsigned long *fields[] = {
7295 shadow_read_write_fields,
7296 shadow_read_only_fields
7298 const int max_fields[] = {
7299 max_shadow_read_write_fields,
7300 max_shadow_read_only_fields
7303 unsigned long field;
7304 u64 field_value = 0;
7305 struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
7307 vmcs_load(shadow_vmcs);
7309 for (q = 0; q < ARRAY_SIZE(fields); q++) {
7310 for (i = 0; i < max_fields[q]; i++) {
7311 field = fields[q][i];
7312 vmcs12_read_any(&vmx->vcpu, field, &field_value);
7314 switch (vmcs_field_type(field)) {
7315 case VMCS_FIELD_TYPE_U16:
7316 vmcs_write16(field, (u16)field_value);
7318 case VMCS_FIELD_TYPE_U32:
7319 vmcs_write32(field, (u32)field_value);
7321 case VMCS_FIELD_TYPE_U64:
7322 vmcs_write64(field, (u64)field_value);
7324 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7325 vmcs_writel(field, (long)field_value);
7334 vmcs_clear(shadow_vmcs);
7335 vmcs_load(vmx->loaded_vmcs->vmcs);
7339 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
7340 * used before) all generate the same failure when it is missing.
7342 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
7344 struct vcpu_vmx *vmx = to_vmx(vcpu);
7345 if (vmx->nested.current_vmptr == -1ull) {
7346 nested_vmx_failInvalid(vcpu);
7347 skip_emulated_instruction(vcpu);
7353 static int handle_vmread(struct kvm_vcpu *vcpu)
7355 unsigned long field;
7357 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7358 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7361 if (!nested_vmx_check_permission(vcpu) ||
7362 !nested_vmx_check_vmcs12(vcpu))
7365 /* Decode instruction info and find the field to read */
7366 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7367 /* Read the field, zero-extended to a u64 field_value */
7368 if (vmcs12_read_any(vcpu, field, &field_value) < 0) {
7369 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7370 skip_emulated_instruction(vcpu);
7374 * Now copy part of this value to register or memory, as requested.
7375 * Note that the number of bits actually copied is 32 or 64 depending
7376 * on the guest's mode (32 or 64 bit), not on the given field's length.
7378 if (vmx_instruction_info & (1u << 10)) {
7379 kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
7382 if (get_vmx_mem_address(vcpu, exit_qualification,
7383 vmx_instruction_info, true, &gva))
7385 /* _system ok, as nested_vmx_check_permission verified cpl=0 */
7386 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
7387 &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
7390 nested_vmx_succeed(vcpu);
7391 skip_emulated_instruction(vcpu);
7396 static int handle_vmwrite(struct kvm_vcpu *vcpu)
7398 unsigned long field;
7400 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7401 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7402 /* The value to write might be 32 or 64 bits, depending on L1's long
7403 * mode, and eventually we need to write that into a field of several
7404 * possible lengths. The code below first zero-extends the value to 64
7405 * bit (field_value), and then copies only the appropriate number of
7406 * bits into the vmcs12 field.
7408 u64 field_value = 0;
7409 struct x86_exception e;
7411 if (!nested_vmx_check_permission(vcpu) ||
7412 !nested_vmx_check_vmcs12(vcpu))
7415 if (vmx_instruction_info & (1u << 10))
7416 field_value = kvm_register_readl(vcpu,
7417 (((vmx_instruction_info) >> 3) & 0xf));
7419 if (get_vmx_mem_address(vcpu, exit_qualification,
7420 vmx_instruction_info, false, &gva))
7422 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
7423 &field_value, (is_64_bit_mode(vcpu) ? 8 : 4), &e)) {
7424 kvm_inject_page_fault(vcpu, &e);
7430 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7431 if (vmcs_field_readonly(field)) {
7432 nested_vmx_failValid(vcpu,
7433 VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
7434 skip_emulated_instruction(vcpu);
7438 if (vmcs12_write_any(vcpu, field, field_value) < 0) {
7439 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7440 skip_emulated_instruction(vcpu);
7444 nested_vmx_succeed(vcpu);
7445 skip_emulated_instruction(vcpu);
7449 /* Emulate the VMPTRLD instruction */
7450 static int handle_vmptrld(struct kvm_vcpu *vcpu)
7452 struct vcpu_vmx *vmx = to_vmx(vcpu);
7455 if (!nested_vmx_check_permission(vcpu))
7458 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMPTRLD, &vmptr))
7461 if (vmx->nested.current_vmptr != vmptr) {
7462 struct vmcs12 *new_vmcs12;
7464 page = nested_get_page(vcpu, vmptr);
7466 nested_vmx_failInvalid(vcpu);
7467 skip_emulated_instruction(vcpu);
7470 new_vmcs12 = kmap(page);
7471 if (new_vmcs12->revision_id != VMCS12_REVISION) {
7473 nested_release_page_clean(page);
7474 nested_vmx_failValid(vcpu,
7475 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
7476 skip_emulated_instruction(vcpu);
7480 nested_release_vmcs12(vmx);
7481 vmx->nested.current_vmptr = vmptr;
7482 vmx->nested.current_vmcs12 = new_vmcs12;
7483 vmx->nested.current_vmcs12_page = page;
7484 if (enable_shadow_vmcs) {
7485 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
7486 SECONDARY_EXEC_SHADOW_VMCS);
7487 vmcs_write64(VMCS_LINK_POINTER,
7488 __pa(vmx->nested.current_shadow_vmcs));
7489 vmx->nested.sync_shadow_vmcs = true;
7493 nested_vmx_succeed(vcpu);
7494 skip_emulated_instruction(vcpu);
7498 /* Emulate the VMPTRST instruction */
7499 static int handle_vmptrst(struct kvm_vcpu *vcpu)
7501 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7502 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7504 struct x86_exception e;
7506 if (!nested_vmx_check_permission(vcpu))
7509 if (get_vmx_mem_address(vcpu, exit_qualification,
7510 vmx_instruction_info, true, &vmcs_gva))
7512 /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
7513 if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
7514 (void *)&to_vmx(vcpu)->nested.current_vmptr,
7516 kvm_inject_page_fault(vcpu, &e);
7519 nested_vmx_succeed(vcpu);
7520 skip_emulated_instruction(vcpu);
7524 /* Emulate the INVEPT instruction */
7525 static int handle_invept(struct kvm_vcpu *vcpu)
7527 struct vcpu_vmx *vmx = to_vmx(vcpu);
7528 u32 vmx_instruction_info, types;
7531 struct x86_exception e;
7536 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7537 SECONDARY_EXEC_ENABLE_EPT) ||
7538 !(vmx->nested.nested_vmx_ept_caps & VMX_EPT_INVEPT_BIT)) {
7539 kvm_queue_exception(vcpu, UD_VECTOR);
7543 if (!nested_vmx_check_permission(vcpu))
7546 if (!kvm_read_cr0_bits(vcpu, X86_CR0_PE)) {
7547 kvm_queue_exception(vcpu, UD_VECTOR);
7551 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7552 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7554 types = (vmx->nested.nested_vmx_ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
7556 if (!(types & (1UL << type))) {
7557 nested_vmx_failValid(vcpu,
7558 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7559 skip_emulated_instruction(vcpu);
7563 /* According to the Intel VMX instruction reference, the memory
7564 * operand is read even if it isn't needed (e.g., for type==global)
7566 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7567 vmx_instruction_info, false, &gva))
7569 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
7570 sizeof(operand), &e)) {
7571 kvm_inject_page_fault(vcpu, &e);
7576 case VMX_EPT_EXTENT_GLOBAL:
7577 kvm_mmu_sync_roots(vcpu);
7578 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
7579 nested_vmx_succeed(vcpu);
7582 /* Trap single context invalidation invept calls */
7587 skip_emulated_instruction(vcpu);
7591 static int handle_invvpid(struct kvm_vcpu *vcpu)
7593 struct vcpu_vmx *vmx = to_vmx(vcpu);
7594 u32 vmx_instruction_info;
7595 unsigned long type, types;
7597 struct x86_exception e;
7600 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7601 SECONDARY_EXEC_ENABLE_VPID) ||
7602 !(vmx->nested.nested_vmx_vpid_caps & VMX_VPID_INVVPID_BIT)) {
7603 kvm_queue_exception(vcpu, UD_VECTOR);
7607 if (!nested_vmx_check_permission(vcpu))
7610 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7611 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7613 types = (vmx->nested.nested_vmx_vpid_caps >> 8) & 0x7;
7615 if (!(types & (1UL << type))) {
7616 nested_vmx_failValid(vcpu,
7617 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7618 skip_emulated_instruction(vcpu);
7622 /* according to the intel vmx instruction reference, the memory
7623 * operand is read even if it isn't needed (e.g., for type==global)
7625 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7626 vmx_instruction_info, false, &gva))
7628 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vpid,
7630 kvm_inject_page_fault(vcpu, &e);
7635 case VMX_VPID_EXTENT_SINGLE_CONTEXT:
7637 * Old versions of KVM use the single-context version so we
7638 * have to support it; just treat it the same as all-context.
7640 case VMX_VPID_EXTENT_ALL_CONTEXT:
7641 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->nested.vpid02);
7642 nested_vmx_succeed(vcpu);
7645 /* Trap individual address invalidation invvpid calls */
7650 skip_emulated_instruction(vcpu);
7654 static int handle_pml_full(struct kvm_vcpu *vcpu)
7656 unsigned long exit_qualification;
7658 trace_kvm_pml_full(vcpu->vcpu_id);
7660 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7663 * PML buffer FULL happened while executing iret from NMI,
7664 * "blocked by NMI" bit has to be set before next VM entry.
7666 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
7667 cpu_has_virtual_nmis() &&
7668 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
7669 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
7670 GUEST_INTR_STATE_NMI);
7673 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
7674 * here.., and there's no userspace involvement needed for PML.
7679 static int handle_pcommit(struct kvm_vcpu *vcpu)
7681 /* we never catch pcommit instruct for L1 guest. */
7686 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
7688 kvm_lapic_expired_hv_timer(vcpu);
7693 * The exit handlers return 1 if the exit was handled fully and guest execution
7694 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
7695 * to be done to userspace and return 0.
7697 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
7698 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
7699 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
7700 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
7701 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
7702 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
7703 [EXIT_REASON_CR_ACCESS] = handle_cr,
7704 [EXIT_REASON_DR_ACCESS] = handle_dr,
7705 [EXIT_REASON_CPUID] = handle_cpuid,
7706 [EXIT_REASON_MSR_READ] = handle_rdmsr,
7707 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
7708 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
7709 [EXIT_REASON_HLT] = handle_halt,
7710 [EXIT_REASON_INVD] = handle_invd,
7711 [EXIT_REASON_INVLPG] = handle_invlpg,
7712 [EXIT_REASON_RDPMC] = handle_rdpmc,
7713 [EXIT_REASON_VMCALL] = handle_vmcall,
7714 [EXIT_REASON_VMCLEAR] = handle_vmclear,
7715 [EXIT_REASON_VMLAUNCH] = handle_vmlaunch,
7716 [EXIT_REASON_VMPTRLD] = handle_vmptrld,
7717 [EXIT_REASON_VMPTRST] = handle_vmptrst,
7718 [EXIT_REASON_VMREAD] = handle_vmread,
7719 [EXIT_REASON_VMRESUME] = handle_vmresume,
7720 [EXIT_REASON_VMWRITE] = handle_vmwrite,
7721 [EXIT_REASON_VMOFF] = handle_vmoff,
7722 [EXIT_REASON_VMON] = handle_vmon,
7723 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
7724 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
7725 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
7726 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
7727 [EXIT_REASON_WBINVD] = handle_wbinvd,
7728 [EXIT_REASON_XSETBV] = handle_xsetbv,
7729 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
7730 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
7731 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
7732 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
7733 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
7734 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait,
7735 [EXIT_REASON_MONITOR_TRAP_FLAG] = handle_monitor_trap,
7736 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor,
7737 [EXIT_REASON_INVEPT] = handle_invept,
7738 [EXIT_REASON_INVVPID] = handle_invvpid,
7739 [EXIT_REASON_XSAVES] = handle_xsaves,
7740 [EXIT_REASON_XRSTORS] = handle_xrstors,
7741 [EXIT_REASON_PML_FULL] = handle_pml_full,
7742 [EXIT_REASON_PCOMMIT] = handle_pcommit,
7743 [EXIT_REASON_PREEMPTION_TIMER] = handle_preemption_timer,
7746 static const int kvm_vmx_max_exit_handlers =
7747 ARRAY_SIZE(kvm_vmx_exit_handlers);
7749 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
7750 struct vmcs12 *vmcs12)
7752 unsigned long exit_qualification;
7753 gpa_t bitmap, last_bitmap;
7758 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7759 return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
7761 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7763 port = exit_qualification >> 16;
7764 size = (exit_qualification & 7) + 1;
7766 last_bitmap = (gpa_t)-1;
7771 bitmap = vmcs12->io_bitmap_a;
7772 else if (port < 0x10000)
7773 bitmap = vmcs12->io_bitmap_b;
7776 bitmap += (port & 0x7fff) / 8;
7778 if (last_bitmap != bitmap)
7779 if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
7781 if (b & (1 << (port & 7)))
7786 last_bitmap = bitmap;
7793 * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
7794 * rather than handle it ourselves in L0. I.e., check whether L1 expressed
7795 * disinterest in the current event (read or write a specific MSR) by using an
7796 * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
7798 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
7799 struct vmcs12 *vmcs12, u32 exit_reason)
7801 u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
7804 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
7808 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
7809 * for the four combinations of read/write and low/high MSR numbers.
7810 * First we need to figure out which of the four to use:
7812 bitmap = vmcs12->msr_bitmap;
7813 if (exit_reason == EXIT_REASON_MSR_WRITE)
7815 if (msr_index >= 0xc0000000) {
7816 msr_index -= 0xc0000000;
7820 /* Then read the msr_index'th bit from this bitmap: */
7821 if (msr_index < 1024*8) {
7823 if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
7825 return 1 & (b >> (msr_index & 7));
7827 return true; /* let L1 handle the wrong parameter */
7831 * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
7832 * rather than handle it ourselves in L0. I.e., check if L1 wanted to
7833 * intercept (via guest_host_mask etc.) the current event.
7835 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
7836 struct vmcs12 *vmcs12)
7838 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7839 int cr = exit_qualification & 15;
7840 int reg = (exit_qualification >> 8) & 15;
7841 unsigned long val = kvm_register_readl(vcpu, reg);
7843 switch ((exit_qualification >> 4) & 3) {
7844 case 0: /* mov to cr */
7847 if (vmcs12->cr0_guest_host_mask &
7848 (val ^ vmcs12->cr0_read_shadow))
7852 if ((vmcs12->cr3_target_count >= 1 &&
7853 vmcs12->cr3_target_value0 == val) ||
7854 (vmcs12->cr3_target_count >= 2 &&
7855 vmcs12->cr3_target_value1 == val) ||
7856 (vmcs12->cr3_target_count >= 3 &&
7857 vmcs12->cr3_target_value2 == val) ||
7858 (vmcs12->cr3_target_count >= 4 &&
7859 vmcs12->cr3_target_value3 == val))
7861 if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
7865 if (vmcs12->cr4_guest_host_mask &
7866 (vmcs12->cr4_read_shadow ^ val))
7870 if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
7876 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
7877 (vmcs12->cr0_read_shadow & X86_CR0_TS))
7880 case 1: /* mov from cr */
7883 if (vmcs12->cpu_based_vm_exec_control &
7884 CPU_BASED_CR3_STORE_EXITING)
7888 if (vmcs12->cpu_based_vm_exec_control &
7889 CPU_BASED_CR8_STORE_EXITING)
7896 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
7897 * cr0. Other attempted changes are ignored, with no exit.
7899 if (vmcs12->cr0_guest_host_mask & 0xe &
7900 (val ^ vmcs12->cr0_read_shadow))
7902 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
7903 !(vmcs12->cr0_read_shadow & 0x1) &&
7912 * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
7913 * should handle it ourselves in L0 (and then continue L2). Only call this
7914 * when in is_guest_mode (L2).
7916 static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu)
7918 u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
7919 struct vcpu_vmx *vmx = to_vmx(vcpu);
7920 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7921 u32 exit_reason = vmx->exit_reason;
7923 trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
7924 vmcs_readl(EXIT_QUALIFICATION),
7925 vmx->idt_vectoring_info,
7927 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
7930 if (vmx->nested.nested_run_pending)
7933 if (unlikely(vmx->fail)) {
7934 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
7935 vmcs_read32(VM_INSTRUCTION_ERROR));
7939 switch (exit_reason) {
7940 case EXIT_REASON_EXCEPTION_NMI:
7941 if (!is_exception(intr_info))
7943 else if (is_page_fault(intr_info))
7945 else if (is_no_device(intr_info) &&
7946 !(vmcs12->guest_cr0 & X86_CR0_TS))
7948 else if (is_debug(intr_info) &&
7950 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
7952 else if (is_breakpoint(intr_info) &&
7953 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
7955 return vmcs12->exception_bitmap &
7956 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
7957 case EXIT_REASON_EXTERNAL_INTERRUPT:
7959 case EXIT_REASON_TRIPLE_FAULT:
7961 case EXIT_REASON_PENDING_INTERRUPT:
7962 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
7963 case EXIT_REASON_NMI_WINDOW:
7964 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
7965 case EXIT_REASON_TASK_SWITCH:
7967 case EXIT_REASON_CPUID:
7968 if (kvm_register_read(vcpu, VCPU_REGS_RAX) == 0xa)
7971 case EXIT_REASON_HLT:
7972 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
7973 case EXIT_REASON_INVD:
7975 case EXIT_REASON_INVLPG:
7976 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
7977 case EXIT_REASON_RDPMC:
7978 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
7979 case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
7980 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
7981 case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
7982 case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
7983 case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
7984 case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
7985 case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
7986 case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
7988 * VMX instructions trap unconditionally. This allows L1 to
7989 * emulate them for its L2 guest, i.e., allows 3-level nesting!
7992 case EXIT_REASON_CR_ACCESS:
7993 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
7994 case EXIT_REASON_DR_ACCESS:
7995 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
7996 case EXIT_REASON_IO_INSTRUCTION:
7997 return nested_vmx_exit_handled_io(vcpu, vmcs12);
7998 case EXIT_REASON_MSR_READ:
7999 case EXIT_REASON_MSR_WRITE:
8000 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
8001 case EXIT_REASON_INVALID_STATE:
8003 case EXIT_REASON_MWAIT_INSTRUCTION:
8004 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
8005 case EXIT_REASON_MONITOR_TRAP_FLAG:
8006 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_TRAP_FLAG);
8007 case EXIT_REASON_MONITOR_INSTRUCTION:
8008 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
8009 case EXIT_REASON_PAUSE_INSTRUCTION:
8010 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
8011 nested_cpu_has2(vmcs12,
8012 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
8013 case EXIT_REASON_MCE_DURING_VMENTRY:
8015 case EXIT_REASON_TPR_BELOW_THRESHOLD:
8016 return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
8017 case EXIT_REASON_APIC_ACCESS:
8018 return nested_cpu_has2(vmcs12,
8019 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
8020 case EXIT_REASON_APIC_WRITE:
8021 case EXIT_REASON_EOI_INDUCED:
8022 /* apic_write and eoi_induced should exit unconditionally. */
8024 case EXIT_REASON_EPT_VIOLATION:
8026 * L0 always deals with the EPT violation. If nested EPT is
8027 * used, and the nested mmu code discovers that the address is
8028 * missing in the guest EPT table (EPT12), the EPT violation
8029 * will be injected with nested_ept_inject_page_fault()
8032 case EXIT_REASON_EPT_MISCONFIG:
8034 * L2 never uses directly L1's EPT, but rather L0's own EPT
8035 * table (shadow on EPT) or a merged EPT table that L0 built
8036 * (EPT on EPT). So any problems with the structure of the
8037 * table is L0's fault.
8040 case EXIT_REASON_WBINVD:
8041 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
8042 case EXIT_REASON_XSETBV:
8044 case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
8046 * This should never happen, since it is not possible to
8047 * set XSS to a non-zero value---neither in L1 nor in L2.
8048 * If if it were, XSS would have to be checked against
8049 * the XSS exit bitmap in vmcs12.
8051 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
8052 case EXIT_REASON_PCOMMIT:
8053 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_PCOMMIT);
8054 case EXIT_REASON_PREEMPTION_TIMER:
8061 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
8063 *info1 = vmcs_readl(EXIT_QUALIFICATION);
8064 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
8067 static int vmx_create_pml_buffer(struct vcpu_vmx *vmx)
8069 struct page *pml_pg;
8071 pml_pg = alloc_page(GFP_KERNEL | __GFP_ZERO);
8075 vmx->pml_pg = pml_pg;
8077 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
8078 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
8083 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
8086 __free_page(vmx->pml_pg);
8091 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
8093 struct vcpu_vmx *vmx = to_vmx(vcpu);
8097 pml_idx = vmcs_read16(GUEST_PML_INDEX);
8099 /* Do nothing if PML buffer is empty */
8100 if (pml_idx == (PML_ENTITY_NUM - 1))
8103 /* PML index always points to next available PML buffer entity */
8104 if (pml_idx >= PML_ENTITY_NUM)
8109 pml_buf = page_address(vmx->pml_pg);
8110 for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
8113 gpa = pml_buf[pml_idx];
8114 WARN_ON(gpa & (PAGE_SIZE - 1));
8115 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
8118 /* reset PML index */
8119 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
8123 * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
8124 * Called before reporting dirty_bitmap to userspace.
8126 static void kvm_flush_pml_buffers(struct kvm *kvm)
8129 struct kvm_vcpu *vcpu;
8131 * We only need to kick vcpu out of guest mode here, as PML buffer
8132 * is flushed at beginning of all VMEXITs, and it's obvious that only
8133 * vcpus running in guest are possible to have unflushed GPAs in PML
8136 kvm_for_each_vcpu(i, vcpu, kvm)
8137 kvm_vcpu_kick(vcpu);
8140 static void vmx_dump_sel(char *name, uint32_t sel)
8142 pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
8143 name, vmcs_read32(sel),
8144 vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
8145 vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
8146 vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
8149 static void vmx_dump_dtsel(char *name, uint32_t limit)
8151 pr_err("%s limit=0x%08x, base=0x%016lx\n",
8152 name, vmcs_read32(limit),
8153 vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
8156 static void dump_vmcs(void)
8158 u32 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
8159 u32 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
8160 u32 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
8161 u32 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
8162 u32 secondary_exec_control = 0;
8163 unsigned long cr4 = vmcs_readl(GUEST_CR4);
8164 u64 efer = vmcs_read64(GUEST_IA32_EFER);
8167 if (cpu_has_secondary_exec_ctrls())
8168 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8170 pr_err("*** Guest State ***\n");
8171 pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8172 vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
8173 vmcs_readl(CR0_GUEST_HOST_MASK));
8174 pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8175 cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
8176 pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
8177 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
8178 (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
8180 pr_err("PDPTR0 = 0x%016llx PDPTR1 = 0x%016llx\n",
8181 vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
8182 pr_err("PDPTR2 = 0x%016llx PDPTR3 = 0x%016llx\n",
8183 vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
8185 pr_err("RSP = 0x%016lx RIP = 0x%016lx\n",
8186 vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
8187 pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n",
8188 vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
8189 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8190 vmcs_readl(GUEST_SYSENTER_ESP),
8191 vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
8192 vmx_dump_sel("CS: ", GUEST_CS_SELECTOR);
8193 vmx_dump_sel("DS: ", GUEST_DS_SELECTOR);
8194 vmx_dump_sel("SS: ", GUEST_SS_SELECTOR);
8195 vmx_dump_sel("ES: ", GUEST_ES_SELECTOR);
8196 vmx_dump_sel("FS: ", GUEST_FS_SELECTOR);
8197 vmx_dump_sel("GS: ", GUEST_GS_SELECTOR);
8198 vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
8199 vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
8200 vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
8201 vmx_dump_sel("TR: ", GUEST_TR_SELECTOR);
8202 if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
8203 (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
8204 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
8205 efer, vmcs_read64(GUEST_IA32_PAT));
8206 pr_err("DebugCtl = 0x%016llx DebugExceptions = 0x%016lx\n",
8207 vmcs_read64(GUEST_IA32_DEBUGCTL),
8208 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
8209 if (vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
8210 pr_err("PerfGlobCtl = 0x%016llx\n",
8211 vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
8212 if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
8213 pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
8214 pr_err("Interruptibility = %08x ActivityState = %08x\n",
8215 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
8216 vmcs_read32(GUEST_ACTIVITY_STATE));
8217 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
8218 pr_err("InterruptStatus = %04x\n",
8219 vmcs_read16(GUEST_INTR_STATUS));
8221 pr_err("*** Host State ***\n");
8222 pr_err("RIP = 0x%016lx RSP = 0x%016lx\n",
8223 vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
8224 pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
8225 vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
8226 vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
8227 vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
8228 vmcs_read16(HOST_TR_SELECTOR));
8229 pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
8230 vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
8231 vmcs_readl(HOST_TR_BASE));
8232 pr_err("GDTBase=%016lx IDTBase=%016lx\n",
8233 vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
8234 pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
8235 vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
8236 vmcs_readl(HOST_CR4));
8237 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8238 vmcs_readl(HOST_IA32_SYSENTER_ESP),
8239 vmcs_read32(HOST_IA32_SYSENTER_CS),
8240 vmcs_readl(HOST_IA32_SYSENTER_EIP));
8241 if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
8242 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
8243 vmcs_read64(HOST_IA32_EFER),
8244 vmcs_read64(HOST_IA32_PAT));
8245 if (vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
8246 pr_err("PerfGlobCtl = 0x%016llx\n",
8247 vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
8249 pr_err("*** Control State ***\n");
8250 pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
8251 pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
8252 pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
8253 pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
8254 vmcs_read32(EXCEPTION_BITMAP),
8255 vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
8256 vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
8257 pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
8258 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8259 vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
8260 vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
8261 pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
8262 vmcs_read32(VM_EXIT_INTR_INFO),
8263 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8264 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
8265 pr_err(" reason=%08x qualification=%016lx\n",
8266 vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
8267 pr_err("IDTVectoring: info=%08x errcode=%08x\n",
8268 vmcs_read32(IDT_VECTORING_INFO_FIELD),
8269 vmcs_read32(IDT_VECTORING_ERROR_CODE));
8270 pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
8271 if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
8272 pr_err("TSC Multiplier = 0x%016llx\n",
8273 vmcs_read64(TSC_MULTIPLIER));
8274 if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW)
8275 pr_err("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
8276 if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
8277 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
8278 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
8279 pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
8280 n = vmcs_read32(CR3_TARGET_COUNT);
8281 for (i = 0; i + 1 < n; i += 4)
8282 pr_err("CR3 target%u=%016lx target%u=%016lx\n",
8283 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
8284 i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
8286 pr_err("CR3 target%u=%016lx\n",
8287 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
8288 if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
8289 pr_err("PLE Gap=%08x Window=%08x\n",
8290 vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
8291 if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
8292 pr_err("Virtual processor ID = 0x%04x\n",
8293 vmcs_read16(VIRTUAL_PROCESSOR_ID));
8297 * The guest has exited. See if we can fix it or if we need userspace
8300 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
8302 struct vcpu_vmx *vmx = to_vmx(vcpu);
8303 u32 exit_reason = vmx->exit_reason;
8304 u32 vectoring_info = vmx->idt_vectoring_info;
8306 trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
8309 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
8310 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
8311 * querying dirty_bitmap, we only need to kick all vcpus out of guest
8312 * mode as if vcpus is in root mode, the PML buffer must has been
8316 vmx_flush_pml_buffer(vcpu);
8318 /* If guest state is invalid, start emulating */
8319 if (vmx->emulation_required)
8320 return handle_invalid_guest_state(vcpu);
8322 if (is_guest_mode(vcpu) && nested_vmx_exit_handled(vcpu)) {
8323 nested_vmx_vmexit(vcpu, exit_reason,
8324 vmcs_read32(VM_EXIT_INTR_INFO),
8325 vmcs_readl(EXIT_QUALIFICATION));
8329 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
8331 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8332 vcpu->run->fail_entry.hardware_entry_failure_reason
8337 if (unlikely(vmx->fail)) {
8338 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8339 vcpu->run->fail_entry.hardware_entry_failure_reason
8340 = vmcs_read32(VM_INSTRUCTION_ERROR);
8346 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
8347 * delivery event since it indicates guest is accessing MMIO.
8348 * The vm-exit can be triggered again after return to guest that
8349 * will cause infinite loop.
8351 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
8352 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
8353 exit_reason != EXIT_REASON_EPT_VIOLATION &&
8354 exit_reason != EXIT_REASON_TASK_SWITCH)) {
8355 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8356 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
8357 vcpu->run->internal.ndata = 2;
8358 vcpu->run->internal.data[0] = vectoring_info;
8359 vcpu->run->internal.data[1] = exit_reason;
8363 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked &&
8364 !(is_guest_mode(vcpu) && nested_cpu_has_virtual_nmis(
8365 get_vmcs12(vcpu))))) {
8366 if (vmx_interrupt_allowed(vcpu)) {
8367 vmx->soft_vnmi_blocked = 0;
8368 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
8369 vcpu->arch.nmi_pending) {
8371 * This CPU don't support us in finding the end of an
8372 * NMI-blocked window if the guest runs with IRQs
8373 * disabled. So we pull the trigger after 1 s of
8374 * futile waiting, but inform the user about this.
8376 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
8377 "state on VCPU %d after 1 s timeout\n",
8378 __func__, vcpu->vcpu_id);
8379 vmx->soft_vnmi_blocked = 0;
8383 if (exit_reason < kvm_vmx_max_exit_handlers
8384 && kvm_vmx_exit_handlers[exit_reason])
8385 return kvm_vmx_exit_handlers[exit_reason](vcpu);
8387 WARN_ONCE(1, "vmx: unexpected exit reason 0x%x\n", exit_reason);
8388 kvm_queue_exception(vcpu, UD_VECTOR);
8393 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
8395 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8397 if (is_guest_mode(vcpu) &&
8398 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
8401 if (irr == -1 || tpr < irr) {
8402 vmcs_write32(TPR_THRESHOLD, 0);
8406 vmcs_write32(TPR_THRESHOLD, irr);
8409 static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
8411 u32 sec_exec_control;
8414 * There is not point to enable virtualize x2apic without enable
8417 if (!cpu_has_vmx_virtualize_x2apic_mode() ||
8418 !kvm_vcpu_apicv_active(vcpu))
8421 if (!cpu_need_tpr_shadow(vcpu))
8424 sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8427 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8428 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8430 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8431 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8433 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
8435 vmx_set_msr_bitmap(vcpu);
8438 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
8440 struct vcpu_vmx *vmx = to_vmx(vcpu);
8443 * Currently we do not handle the nested case where L2 has an
8444 * APIC access page of its own; that page is still pinned.
8445 * Hence, we skip the case where the VCPU is in guest mode _and_
8446 * L1 prepared an APIC access page for L2.
8448 * For the case where L1 and L2 share the same APIC access page
8449 * (flexpriority=Y but SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES clear
8450 * in the vmcs12), this function will only update either the vmcs01
8451 * or the vmcs02. If the former, the vmcs02 will be updated by
8452 * prepare_vmcs02. If the latter, the vmcs01 will be updated in
8453 * the next L2->L1 exit.
8455 if (!is_guest_mode(vcpu) ||
8456 !nested_cpu_has2(vmx->nested.current_vmcs12,
8457 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
8458 vmcs_write64(APIC_ACCESS_ADDR, hpa);
8461 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
8469 status = vmcs_read16(GUEST_INTR_STATUS);
8471 if (max_isr != old) {
8473 status |= max_isr << 8;
8474 vmcs_write16(GUEST_INTR_STATUS, status);
8478 static void vmx_set_rvi(int vector)
8486 status = vmcs_read16(GUEST_INTR_STATUS);
8487 old = (u8)status & 0xff;
8488 if ((u8)vector != old) {
8490 status |= (u8)vector;
8491 vmcs_write16(GUEST_INTR_STATUS, status);
8495 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
8497 if (!is_guest_mode(vcpu)) {
8498 vmx_set_rvi(max_irr);
8506 * In guest mode. If a vmexit is needed, vmx_check_nested_events
8509 if (nested_exit_on_intr(vcpu))
8513 * Else, fall back to pre-APICv interrupt injection since L2
8514 * is run without virtual interrupt delivery.
8516 if (!kvm_event_needs_reinjection(vcpu) &&
8517 vmx_interrupt_allowed(vcpu)) {
8518 kvm_queue_interrupt(vcpu, max_irr, false);
8519 vmx_inject_irq(vcpu);
8523 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
8525 if (!kvm_vcpu_apicv_active(vcpu))
8528 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
8529 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
8530 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
8531 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
8534 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
8538 if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
8539 || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
8542 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8543 exit_intr_info = vmx->exit_intr_info;
8545 /* Handle machine checks before interrupts are enabled */
8546 if (is_machine_check(exit_intr_info))
8547 kvm_machine_check();
8549 /* We need to handle NMIs before interrupts are enabled */
8550 if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
8551 (exit_intr_info & INTR_INFO_VALID_MASK)) {
8552 kvm_before_handle_nmi(&vmx->vcpu);
8554 kvm_after_handle_nmi(&vmx->vcpu);
8558 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
8560 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8561 register void *__sp asm(_ASM_SP);
8564 * If external interrupt exists, IF bit is set in rflags/eflags on the
8565 * interrupt stack frame, and interrupt will be enabled on a return
8566 * from interrupt handler.
8568 if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
8569 == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
8570 unsigned int vector;
8571 unsigned long entry;
8573 struct vcpu_vmx *vmx = to_vmx(vcpu);
8574 #ifdef CONFIG_X86_64
8578 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
8579 desc = (gate_desc *)vmx->host_idt_base + vector;
8580 entry = gate_offset(*desc);
8582 #ifdef CONFIG_X86_64
8583 "mov %%" _ASM_SP ", %[sp]\n\t"
8584 "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
8589 __ASM_SIZE(push) " $%c[cs]\n\t"
8590 "call *%[entry]\n\t"
8592 #ifdef CONFIG_X86_64
8598 [ss]"i"(__KERNEL_DS),
8599 [cs]"i"(__KERNEL_CS)
8604 static bool vmx_has_high_real_mode_segbase(void)
8606 return enable_unrestricted_guest || emulate_invalid_guest_state;
8609 static bool vmx_mpx_supported(void)
8611 return (vmcs_config.vmexit_ctrl & VM_EXIT_CLEAR_BNDCFGS) &&
8612 (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_BNDCFGS);
8615 static bool vmx_xsaves_supported(void)
8617 return vmcs_config.cpu_based_2nd_exec_ctrl &
8618 SECONDARY_EXEC_XSAVES;
8621 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
8626 bool idtv_info_valid;
8628 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8630 if (cpu_has_virtual_nmis()) {
8631 if (vmx->nmi_known_unmasked)
8634 * Can't use vmx->exit_intr_info since we're not sure what
8635 * the exit reason is.
8637 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8638 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
8639 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
8641 * SDM 3: 27.7.1.2 (September 2008)
8642 * Re-set bit "block by NMI" before VM entry if vmexit caused by
8643 * a guest IRET fault.
8644 * SDM 3: 23.2.2 (September 2008)
8645 * Bit 12 is undefined in any of the following cases:
8646 * If the VM exit sets the valid bit in the IDT-vectoring
8647 * information field.
8648 * If the VM exit is due to a double fault.
8650 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
8651 vector != DF_VECTOR && !idtv_info_valid)
8652 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
8653 GUEST_INTR_STATE_NMI);
8655 vmx->nmi_known_unmasked =
8656 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
8657 & GUEST_INTR_STATE_NMI);
8658 } else if (unlikely(vmx->soft_vnmi_blocked))
8659 vmx->vnmi_blocked_time +=
8660 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
8663 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
8664 u32 idt_vectoring_info,
8665 int instr_len_field,
8666 int error_code_field)
8670 bool idtv_info_valid;
8672 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8674 vcpu->arch.nmi_injected = false;
8675 kvm_clear_exception_queue(vcpu);
8676 kvm_clear_interrupt_queue(vcpu);
8678 if (!idtv_info_valid)
8681 kvm_make_request(KVM_REQ_EVENT, vcpu);
8683 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
8684 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
8687 case INTR_TYPE_NMI_INTR:
8688 vcpu->arch.nmi_injected = true;
8690 * SDM 3: 27.7.1.2 (September 2008)
8691 * Clear bit "block by NMI" before VM entry if a NMI
8694 vmx_set_nmi_mask(vcpu, false);
8696 case INTR_TYPE_SOFT_EXCEPTION:
8697 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8699 case INTR_TYPE_HARD_EXCEPTION:
8700 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
8701 u32 err = vmcs_read32(error_code_field);
8702 kvm_requeue_exception_e(vcpu, vector, err);
8704 kvm_requeue_exception(vcpu, vector);
8706 case INTR_TYPE_SOFT_INTR:
8707 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8709 case INTR_TYPE_EXT_INTR:
8710 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
8717 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
8719 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
8720 VM_EXIT_INSTRUCTION_LEN,
8721 IDT_VECTORING_ERROR_CODE);
8724 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
8726 __vmx_complete_interrupts(vcpu,
8727 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8728 VM_ENTRY_INSTRUCTION_LEN,
8729 VM_ENTRY_EXCEPTION_ERROR_CODE);
8731 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
8734 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
8737 struct perf_guest_switch_msr *msrs;
8739 msrs = perf_guest_get_msrs(&nr_msrs);
8744 for (i = 0; i < nr_msrs; i++)
8745 if (msrs[i].host == msrs[i].guest)
8746 clear_atomic_switch_msr(vmx, msrs[i].msr);
8748 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
8752 void vmx_arm_hv_timer(struct kvm_vcpu *vcpu)
8754 struct vcpu_vmx *vmx = to_vmx(vcpu);
8758 if (vmx->hv_deadline_tsc == -1)
8762 if (vmx->hv_deadline_tsc > tscl)
8763 /* sure to be 32 bit only because checked on set_hv_timer */
8764 delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
8765 cpu_preemption_timer_multi);
8769 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
8772 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
8774 struct vcpu_vmx *vmx = to_vmx(vcpu);
8775 unsigned long debugctlmsr, cr4;
8777 /* Record the guest's net vcpu time for enforced NMI injections. */
8778 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
8779 vmx->entry_time = ktime_get();
8781 /* Don't enter VMX if guest state is invalid, let the exit handler
8782 start emulation until we arrive back to a valid state */
8783 if (vmx->emulation_required)
8786 if (vmx->ple_window_dirty) {
8787 vmx->ple_window_dirty = false;
8788 vmcs_write32(PLE_WINDOW, vmx->ple_window);
8791 if (vmx->nested.sync_shadow_vmcs) {
8792 copy_vmcs12_to_shadow(vmx);
8793 vmx->nested.sync_shadow_vmcs = false;
8796 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
8797 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
8798 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
8799 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
8801 cr4 = cr4_read_shadow();
8802 if (unlikely(cr4 != vmx->host_state.vmcs_host_cr4)) {
8803 vmcs_writel(HOST_CR4, cr4);
8804 vmx->host_state.vmcs_host_cr4 = cr4;
8807 /* When single-stepping over STI and MOV SS, we must clear the
8808 * corresponding interruptibility bits in the guest state. Otherwise
8809 * vmentry fails as it then expects bit 14 (BS) in pending debug
8810 * exceptions being set, but that's not correct for the guest debugging
8812 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8813 vmx_set_interrupt_shadow(vcpu, 0);
8815 if (vmx->guest_pkru_valid)
8816 __write_pkru(vmx->guest_pkru);
8818 atomic_switch_perf_msrs(vmx);
8819 debugctlmsr = get_debugctlmsr();
8821 vmx_arm_hv_timer(vcpu);
8823 vmx->__launched = vmx->loaded_vmcs->launched;
8825 /* Store host registers */
8826 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
8827 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
8828 "push %%" _ASM_CX " \n\t"
8829 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
8831 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
8832 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
8834 /* Reload cr2 if changed */
8835 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
8836 "mov %%cr2, %%" _ASM_DX " \n\t"
8837 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
8839 "mov %%" _ASM_AX", %%cr2 \n\t"
8841 /* Check if vmlaunch of vmresume is needed */
8842 "cmpl $0, %c[launched](%0) \n\t"
8843 /* Load guest registers. Don't clobber flags. */
8844 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
8845 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
8846 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
8847 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
8848 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
8849 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
8850 #ifdef CONFIG_X86_64
8851 "mov %c[r8](%0), %%r8 \n\t"
8852 "mov %c[r9](%0), %%r9 \n\t"
8853 "mov %c[r10](%0), %%r10 \n\t"
8854 "mov %c[r11](%0), %%r11 \n\t"
8855 "mov %c[r12](%0), %%r12 \n\t"
8856 "mov %c[r13](%0), %%r13 \n\t"
8857 "mov %c[r14](%0), %%r14 \n\t"
8858 "mov %c[r15](%0), %%r15 \n\t"
8860 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
8862 /* Enter guest mode */
8864 __ex(ASM_VMX_VMLAUNCH) "\n\t"
8866 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
8868 /* Save guest registers, load host registers, keep flags */
8869 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
8871 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
8872 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
8873 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
8874 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
8875 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
8876 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
8877 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
8878 #ifdef CONFIG_X86_64
8879 "mov %%r8, %c[r8](%0) \n\t"
8880 "mov %%r9, %c[r9](%0) \n\t"
8881 "mov %%r10, %c[r10](%0) \n\t"
8882 "mov %%r11, %c[r11](%0) \n\t"
8883 "mov %%r12, %c[r12](%0) \n\t"
8884 "mov %%r13, %c[r13](%0) \n\t"
8885 "mov %%r14, %c[r14](%0) \n\t"
8886 "mov %%r15, %c[r15](%0) \n\t"
8888 "mov %%cr2, %%" _ASM_AX " \n\t"
8889 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
8891 "pop %%" _ASM_BP "; pop %%" _ASM_DX " \n\t"
8892 "setbe %c[fail](%0) \n\t"
8893 ".pushsection .rodata \n\t"
8894 ".global vmx_return \n\t"
8895 "vmx_return: " _ASM_PTR " 2b \n\t"
8897 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
8898 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
8899 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
8900 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
8901 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
8902 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
8903 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
8904 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
8905 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
8906 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
8907 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
8908 #ifdef CONFIG_X86_64
8909 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
8910 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
8911 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
8912 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
8913 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
8914 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
8915 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
8916 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
8918 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
8919 [wordsize]"i"(sizeof(ulong))
8921 #ifdef CONFIG_X86_64
8922 , "rax", "rbx", "rdi", "rsi"
8923 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
8925 , "eax", "ebx", "edi", "esi"
8929 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
8931 update_debugctlmsr(debugctlmsr);
8933 #ifndef CONFIG_X86_64
8935 * The sysexit path does not restore ds/es, so we must set them to
8936 * a reasonable value ourselves.
8938 * We can't defer this to vmx_load_host_state() since that function
8939 * may be executed in interrupt context, which saves and restore segments
8940 * around it, nullifying its effect.
8942 loadsegment(ds, __USER_DS);
8943 loadsegment(es, __USER_DS);
8946 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
8947 | (1 << VCPU_EXREG_RFLAGS)
8948 | (1 << VCPU_EXREG_PDPTR)
8949 | (1 << VCPU_EXREG_SEGMENTS)
8950 | (1 << VCPU_EXREG_CR3));
8951 vcpu->arch.regs_dirty = 0;
8953 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
8955 vmx->loaded_vmcs->launched = 1;
8957 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
8960 * eager fpu is enabled if PKEY is supported and CR4 is switched
8961 * back on host, so it is safe to read guest PKRU from current
8964 if (boot_cpu_has(X86_FEATURE_OSPKE)) {
8965 vmx->guest_pkru = __read_pkru();
8966 if (vmx->guest_pkru != vmx->host_pkru) {
8967 vmx->guest_pkru_valid = true;
8968 __write_pkru(vmx->host_pkru);
8970 vmx->guest_pkru_valid = false;
8974 * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
8975 * we did not inject a still-pending event to L1 now because of
8976 * nested_run_pending, we need to re-enable this bit.
8978 if (vmx->nested.nested_run_pending)
8979 kvm_make_request(KVM_REQ_EVENT, vcpu);
8981 vmx->nested.nested_run_pending = 0;
8983 vmx_complete_atomic_exit(vmx);
8984 vmx_recover_nmi_blocking(vmx);
8985 vmx_complete_interrupts(vmx);
8988 static void vmx_load_vmcs01(struct kvm_vcpu *vcpu)
8990 struct vcpu_vmx *vmx = to_vmx(vcpu);
8993 if (vmx->loaded_vmcs == &vmx->vmcs01)
8997 vmx->loaded_vmcs = &vmx->vmcs01;
8999 vmx_vcpu_load(vcpu, cpu);
9004 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
9006 struct vcpu_vmx *vmx = to_vmx(vcpu);
9009 vmx_destroy_pml_buffer(vmx);
9010 free_vpid(vmx->vpid);
9011 leave_guest_mode(vcpu);
9012 vmx_load_vmcs01(vcpu);
9014 free_loaded_vmcs(vmx->loaded_vmcs);
9015 kfree(vmx->guest_msrs);
9016 kvm_vcpu_uninit(vcpu);
9017 kmem_cache_free(kvm_vcpu_cache, vmx);
9020 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
9023 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
9027 return ERR_PTR(-ENOMEM);
9029 vmx->vpid = allocate_vpid();
9031 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
9035 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
9036 BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
9040 if (!vmx->guest_msrs) {
9044 vmx->loaded_vmcs = &vmx->vmcs01;
9045 vmx->loaded_vmcs->vmcs = alloc_vmcs();
9046 if (!vmx->loaded_vmcs->vmcs)
9049 kvm_cpu_vmxon(__pa(per_cpu(vmxarea, raw_smp_processor_id())));
9050 loaded_vmcs_init(vmx->loaded_vmcs);
9055 vmx_vcpu_load(&vmx->vcpu, cpu);
9056 vmx->vcpu.cpu = cpu;
9057 err = vmx_vcpu_setup(vmx);
9058 vmx_vcpu_put(&vmx->vcpu);
9062 if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9063 err = alloc_apic_access_page(kvm);
9069 if (!kvm->arch.ept_identity_map_addr)
9070 kvm->arch.ept_identity_map_addr =
9071 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
9072 err = init_rmode_identity_map(kvm);
9078 nested_vmx_setup_ctls_msrs(vmx);
9079 vmx->nested.vpid02 = allocate_vpid();
9082 vmx->nested.posted_intr_nv = -1;
9083 vmx->nested.current_vmptr = -1ull;
9084 vmx->nested.current_vmcs12 = NULL;
9087 * If PML is turned on, failure on enabling PML just results in failure
9088 * of creating the vcpu, therefore we can simplify PML logic (by
9089 * avoiding dealing with cases, such as enabling PML partially on vcpus
9090 * for the guest, etc.
9093 err = vmx_create_pml_buffer(vmx);
9098 vmx->msr_ia32_feature_control_valid_bits = FEATURE_CONTROL_LOCKED;
9103 free_vpid(vmx->nested.vpid02);
9104 free_loaded_vmcs(vmx->loaded_vmcs);
9106 kfree(vmx->guest_msrs);
9108 kvm_vcpu_uninit(&vmx->vcpu);
9110 free_vpid(vmx->vpid);
9111 kmem_cache_free(kvm_vcpu_cache, vmx);
9112 return ERR_PTR(err);
9115 static void __init vmx_check_processor_compat(void *rtn)
9117 struct vmcs_config vmcs_conf;
9120 if (setup_vmcs_config(&vmcs_conf) < 0)
9122 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
9123 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
9124 smp_processor_id());
9129 static int get_ept_level(void)
9131 return VMX_EPT_DEFAULT_GAW + 1;
9134 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
9139 /* For VT-d and EPT combination
9140 * 1. MMIO: always map as UC
9142 * a. VT-d without snooping control feature: can't guarantee the
9143 * result, try to trust guest.
9144 * b. VT-d with snooping control feature: snooping control feature of
9145 * VT-d engine can guarantee the cache correctness. Just set it
9146 * to WB to keep consistent with host. So the same as item 3.
9147 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
9148 * consistent with host MTRR
9151 cache = MTRR_TYPE_UNCACHABLE;
9155 if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
9156 ipat = VMX_EPT_IPAT_BIT;
9157 cache = MTRR_TYPE_WRBACK;
9161 if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
9162 ipat = VMX_EPT_IPAT_BIT;
9163 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
9164 cache = MTRR_TYPE_WRBACK;
9166 cache = MTRR_TYPE_UNCACHABLE;
9170 cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
9173 return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
9176 static int vmx_get_lpage_level(void)
9178 if (enable_ept && !cpu_has_vmx_ept_1g_page())
9179 return PT_DIRECTORY_LEVEL;
9181 /* For shadow and EPT supported 1GB page */
9182 return PT_PDPE_LEVEL;
9185 static void vmcs_set_secondary_exec_control(u32 new_ctl)
9188 * These bits in the secondary execution controls field
9189 * are dynamic, the others are mostly based on the hypervisor
9190 * architecture and the guest's CPUID. Do not touch the
9194 SECONDARY_EXEC_SHADOW_VMCS |
9195 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
9196 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9198 u32 cur_ctl = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
9200 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
9201 (new_ctl & ~mask) | (cur_ctl & mask));
9204 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
9206 struct kvm_cpuid_entry2 *best;
9207 struct vcpu_vmx *vmx = to_vmx(vcpu);
9208 u32 secondary_exec_ctl = vmx_secondary_exec_control(vmx);
9210 if (vmx_rdtscp_supported()) {
9211 bool rdtscp_enabled = guest_cpuid_has_rdtscp(vcpu);
9212 if (!rdtscp_enabled)
9213 secondary_exec_ctl &= ~SECONDARY_EXEC_RDTSCP;
9217 vmx->nested.nested_vmx_secondary_ctls_high |=
9218 SECONDARY_EXEC_RDTSCP;
9220 vmx->nested.nested_vmx_secondary_ctls_high &=
9221 ~SECONDARY_EXEC_RDTSCP;
9225 /* Exposing INVPCID only when PCID is exposed */
9226 best = kvm_find_cpuid_entry(vcpu, 0x7, 0);
9227 if (vmx_invpcid_supported() &&
9228 (!best || !(best->ebx & bit(X86_FEATURE_INVPCID)) ||
9229 !guest_cpuid_has_pcid(vcpu))) {
9230 secondary_exec_ctl &= ~SECONDARY_EXEC_ENABLE_INVPCID;
9233 best->ebx &= ~bit(X86_FEATURE_INVPCID);
9236 if (cpu_has_secondary_exec_ctrls())
9237 vmcs_set_secondary_exec_control(secondary_exec_ctl);
9239 if (static_cpu_has(X86_FEATURE_PCOMMIT) && nested) {
9240 if (guest_cpuid_has_pcommit(vcpu))
9241 vmx->nested.nested_vmx_secondary_ctls_high |=
9242 SECONDARY_EXEC_PCOMMIT;
9244 vmx->nested.nested_vmx_secondary_ctls_high &=
9245 ~SECONDARY_EXEC_PCOMMIT;
9248 if (nested_vmx_allowed(vcpu))
9249 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
9250 FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
9252 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
9253 ~FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
9256 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
9258 if (func == 1 && nested)
9259 entry->ecx |= bit(X86_FEATURE_VMX);
9262 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
9263 struct x86_exception *fault)
9265 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9268 if (fault->error_code & PFERR_RSVD_MASK)
9269 exit_reason = EXIT_REASON_EPT_MISCONFIG;
9271 exit_reason = EXIT_REASON_EPT_VIOLATION;
9272 nested_vmx_vmexit(vcpu, exit_reason, 0, vcpu->arch.exit_qualification);
9273 vmcs12->guest_physical_address = fault->address;
9276 /* Callbacks for nested_ept_init_mmu_context: */
9278 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
9280 /* return the page table to be shadowed - in our case, EPT12 */
9281 return get_vmcs12(vcpu)->ept_pointer;
9284 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
9286 WARN_ON(mmu_is_nested(vcpu));
9287 kvm_init_shadow_ept_mmu(vcpu,
9288 to_vmx(vcpu)->nested.nested_vmx_ept_caps &
9289 VMX_EPT_EXECUTE_ONLY_BIT);
9290 vcpu->arch.mmu.set_cr3 = vmx_set_cr3;
9291 vcpu->arch.mmu.get_cr3 = nested_ept_get_cr3;
9292 vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
9294 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
9297 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
9299 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
9302 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
9305 bool inequality, bit;
9307 bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
9309 (error_code & vmcs12->page_fault_error_code_mask) !=
9310 vmcs12->page_fault_error_code_match;
9311 return inequality ^ bit;
9314 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
9315 struct x86_exception *fault)
9317 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9319 WARN_ON(!is_guest_mode(vcpu));
9321 if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code))
9322 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
9323 vmcs_read32(VM_EXIT_INTR_INFO),
9324 vmcs_readl(EXIT_QUALIFICATION));
9326 kvm_inject_page_fault(vcpu, fault);
9329 static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu,
9330 struct vmcs12 *vmcs12)
9332 struct vcpu_vmx *vmx = to_vmx(vcpu);
9333 int maxphyaddr = cpuid_maxphyaddr(vcpu);
9335 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
9336 if (!PAGE_ALIGNED(vmcs12->apic_access_addr) ||
9337 vmcs12->apic_access_addr >> maxphyaddr)
9341 * Translate L1 physical address to host physical
9342 * address for vmcs02. Keep the page pinned, so this
9343 * physical address remains valid. We keep a reference
9344 * to it so we can release it later.
9346 if (vmx->nested.apic_access_page) /* shouldn't happen */
9347 nested_release_page(vmx->nested.apic_access_page);
9348 vmx->nested.apic_access_page =
9349 nested_get_page(vcpu, vmcs12->apic_access_addr);
9352 if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
9353 if (!PAGE_ALIGNED(vmcs12->virtual_apic_page_addr) ||
9354 vmcs12->virtual_apic_page_addr >> maxphyaddr)
9357 if (vmx->nested.virtual_apic_page) /* shouldn't happen */
9358 nested_release_page(vmx->nested.virtual_apic_page);
9359 vmx->nested.virtual_apic_page =
9360 nested_get_page(vcpu, vmcs12->virtual_apic_page_addr);
9363 * Failing the vm entry is _not_ what the processor does
9364 * but it's basically the only possibility we have.
9365 * We could still enter the guest if CR8 load exits are
9366 * enabled, CR8 store exits are enabled, and virtualize APIC
9367 * access is disabled; in this case the processor would never
9368 * use the TPR shadow and we could simply clear the bit from
9369 * the execution control. But such a configuration is useless,
9370 * so let's keep the code simple.
9372 if (!vmx->nested.virtual_apic_page)
9376 if (nested_cpu_has_posted_intr(vmcs12)) {
9377 if (!IS_ALIGNED(vmcs12->posted_intr_desc_addr, 64) ||
9378 vmcs12->posted_intr_desc_addr >> maxphyaddr)
9381 if (vmx->nested.pi_desc_page) { /* shouldn't happen */
9382 kunmap(vmx->nested.pi_desc_page);
9383 nested_release_page(vmx->nested.pi_desc_page);
9385 vmx->nested.pi_desc_page =
9386 nested_get_page(vcpu, vmcs12->posted_intr_desc_addr);
9387 if (!vmx->nested.pi_desc_page)
9390 vmx->nested.pi_desc =
9391 (struct pi_desc *)kmap(vmx->nested.pi_desc_page);
9392 if (!vmx->nested.pi_desc) {
9393 nested_release_page_clean(vmx->nested.pi_desc_page);
9396 vmx->nested.pi_desc =
9397 (struct pi_desc *)((void *)vmx->nested.pi_desc +
9398 (unsigned long)(vmcs12->posted_intr_desc_addr &
9405 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
9407 u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
9408 struct vcpu_vmx *vmx = to_vmx(vcpu);
9410 if (vcpu->arch.virtual_tsc_khz == 0)
9413 /* Make sure short timeouts reliably trigger an immediate vmexit.
9414 * hrtimer_start does not guarantee this. */
9415 if (preemption_timeout <= 1) {
9416 vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
9420 preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
9421 preemption_timeout *= 1000000;
9422 do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
9423 hrtimer_start(&vmx->nested.preemption_timer,
9424 ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
9427 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
9428 struct vmcs12 *vmcs12)
9433 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
9436 if (vmcs12_read_any(vcpu, MSR_BITMAP, &addr)) {
9440 maxphyaddr = cpuid_maxphyaddr(vcpu);
9442 if (!PAGE_ALIGNED(vmcs12->msr_bitmap) ||
9443 ((addr + PAGE_SIZE) >> maxphyaddr))
9450 * Merge L0's and L1's MSR bitmap, return false to indicate that
9451 * we do not use the hardware.
9453 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
9454 struct vmcs12 *vmcs12)
9458 unsigned long *msr_bitmap;
9460 if (!nested_cpu_has_virt_x2apic_mode(vmcs12))
9463 page = nested_get_page(vcpu, vmcs12->msr_bitmap);
9468 msr_bitmap = (unsigned long *)kmap(page);
9470 nested_release_page_clean(page);
9475 if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
9476 if (nested_cpu_has_apic_reg_virt(vmcs12))
9477 for (msr = 0x800; msr <= 0x8ff; msr++)
9478 nested_vmx_disable_intercept_for_msr(
9480 vmx_msr_bitmap_nested,
9482 /* TPR is allowed */
9483 nested_vmx_disable_intercept_for_msr(msr_bitmap,
9484 vmx_msr_bitmap_nested,
9485 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
9486 MSR_TYPE_R | MSR_TYPE_W);
9487 if (nested_cpu_has_vid(vmcs12)) {
9488 /* EOI and self-IPI are allowed */
9489 nested_vmx_disable_intercept_for_msr(
9491 vmx_msr_bitmap_nested,
9492 APIC_BASE_MSR + (APIC_EOI >> 4),
9494 nested_vmx_disable_intercept_for_msr(
9496 vmx_msr_bitmap_nested,
9497 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
9502 * Enable reading intercept of all the x2apic
9503 * MSRs. We should not rely on vmcs12 to do any
9504 * optimizations here, it may have been modified
9507 for (msr = 0x800; msr <= 0x8ff; msr++)
9508 __vmx_enable_intercept_for_msr(
9509 vmx_msr_bitmap_nested,
9513 __vmx_enable_intercept_for_msr(
9514 vmx_msr_bitmap_nested,
9515 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
9517 __vmx_enable_intercept_for_msr(
9518 vmx_msr_bitmap_nested,
9519 APIC_BASE_MSR + (APIC_EOI >> 4),
9521 __vmx_enable_intercept_for_msr(
9522 vmx_msr_bitmap_nested,
9523 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
9527 nested_release_page_clean(page);
9532 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
9533 struct vmcs12 *vmcs12)
9535 if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9536 !nested_cpu_has_apic_reg_virt(vmcs12) &&
9537 !nested_cpu_has_vid(vmcs12) &&
9538 !nested_cpu_has_posted_intr(vmcs12))
9542 * If virtualize x2apic mode is enabled,
9543 * virtualize apic access must be disabled.
9545 if (nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9546 nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
9550 * If virtual interrupt delivery is enabled,
9551 * we must exit on external interrupts.
9553 if (nested_cpu_has_vid(vmcs12) &&
9554 !nested_exit_on_intr(vcpu))
9558 * bits 15:8 should be zero in posted_intr_nv,
9559 * the descriptor address has been already checked
9560 * in nested_get_vmcs12_pages.
9562 if (nested_cpu_has_posted_intr(vmcs12) &&
9563 (!nested_cpu_has_vid(vmcs12) ||
9564 !nested_exit_intr_ack_set(vcpu) ||
9565 vmcs12->posted_intr_nv & 0xff00))
9568 /* tpr shadow is needed by all apicv features. */
9569 if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
9575 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
9576 unsigned long count_field,
9577 unsigned long addr_field)
9582 if (vmcs12_read_any(vcpu, count_field, &count) ||
9583 vmcs12_read_any(vcpu, addr_field, &addr)) {
9589 maxphyaddr = cpuid_maxphyaddr(vcpu);
9590 if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
9591 (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr) {
9592 pr_warn_ratelimited(
9593 "nVMX: invalid MSR switch (0x%lx, %d, %llu, 0x%08llx)",
9594 addr_field, maxphyaddr, count, addr);
9600 static int nested_vmx_check_msr_switch_controls(struct kvm_vcpu *vcpu,
9601 struct vmcs12 *vmcs12)
9603 if (vmcs12->vm_exit_msr_load_count == 0 &&
9604 vmcs12->vm_exit_msr_store_count == 0 &&
9605 vmcs12->vm_entry_msr_load_count == 0)
9606 return 0; /* Fast path */
9607 if (nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_LOAD_COUNT,
9608 VM_EXIT_MSR_LOAD_ADDR) ||
9609 nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_STORE_COUNT,
9610 VM_EXIT_MSR_STORE_ADDR) ||
9611 nested_vmx_check_msr_switch(vcpu, VM_ENTRY_MSR_LOAD_COUNT,
9612 VM_ENTRY_MSR_LOAD_ADDR))
9617 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
9618 struct vmx_msr_entry *e)
9620 /* x2APIC MSR accesses are not allowed */
9621 if (vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8)
9623 if (e->index == MSR_IA32_UCODE_WRITE || /* SDM Table 35-2 */
9624 e->index == MSR_IA32_UCODE_REV)
9626 if (e->reserved != 0)
9631 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
9632 struct vmx_msr_entry *e)
9634 if (e->index == MSR_FS_BASE ||
9635 e->index == MSR_GS_BASE ||
9636 e->index == MSR_IA32_SMM_MONITOR_CTL || /* SMM is not supported */
9637 nested_vmx_msr_check_common(vcpu, e))
9642 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
9643 struct vmx_msr_entry *e)
9645 if (e->index == MSR_IA32_SMBASE || /* SMM is not supported */
9646 nested_vmx_msr_check_common(vcpu, e))
9652 * Load guest's/host's msr at nested entry/exit.
9653 * return 0 for success, entry index for failure.
9655 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9658 struct vmx_msr_entry e;
9659 struct msr_data msr;
9661 msr.host_initiated = false;
9662 for (i = 0; i < count; i++) {
9663 if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
9665 pr_warn_ratelimited(
9666 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9667 __func__, i, gpa + i * sizeof(e));
9670 if (nested_vmx_load_msr_check(vcpu, &e)) {
9671 pr_warn_ratelimited(
9672 "%s check failed (%u, 0x%x, 0x%x)\n",
9673 __func__, i, e.index, e.reserved);
9676 msr.index = e.index;
9678 if (kvm_set_msr(vcpu, &msr)) {
9679 pr_warn_ratelimited(
9680 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9681 __func__, i, e.index, e.value);
9690 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9693 struct vmx_msr_entry e;
9695 for (i = 0; i < count; i++) {
9696 struct msr_data msr_info;
9697 if (kvm_vcpu_read_guest(vcpu,
9698 gpa + i * sizeof(e),
9699 &e, 2 * sizeof(u32))) {
9700 pr_warn_ratelimited(
9701 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9702 __func__, i, gpa + i * sizeof(e));
9705 if (nested_vmx_store_msr_check(vcpu, &e)) {
9706 pr_warn_ratelimited(
9707 "%s check failed (%u, 0x%x, 0x%x)\n",
9708 __func__, i, e.index, e.reserved);
9711 msr_info.host_initiated = false;
9712 msr_info.index = e.index;
9713 if (kvm_get_msr(vcpu, &msr_info)) {
9714 pr_warn_ratelimited(
9715 "%s cannot read MSR (%u, 0x%x)\n",
9716 __func__, i, e.index);
9719 if (kvm_vcpu_write_guest(vcpu,
9720 gpa + i * sizeof(e) +
9721 offsetof(struct vmx_msr_entry, value),
9722 &msr_info.data, sizeof(msr_info.data))) {
9723 pr_warn_ratelimited(
9724 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9725 __func__, i, e.index, msr_info.data);
9733 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
9734 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
9735 * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
9736 * guest in a way that will both be appropriate to L1's requests, and our
9737 * needs. In addition to modifying the active vmcs (which is vmcs02), this
9738 * function also has additional necessary side-effects, like setting various
9739 * vcpu->arch fields.
9741 static void prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
9743 struct vcpu_vmx *vmx = to_vmx(vcpu);
9746 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
9747 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
9748 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
9749 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
9750 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
9751 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
9752 vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
9753 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
9754 vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
9755 vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
9756 vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
9757 vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
9758 vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
9759 vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
9760 vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
9761 vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
9762 vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
9763 vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
9764 vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
9765 vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
9766 vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
9767 vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
9768 vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
9769 vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
9770 vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
9771 vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
9772 vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
9773 vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
9774 vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
9775 vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
9776 vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
9777 vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
9778 vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
9779 vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
9780 vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
9781 vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
9783 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
9784 kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
9785 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
9787 kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
9788 vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
9790 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
9791 vmcs12->vm_entry_intr_info_field);
9792 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
9793 vmcs12->vm_entry_exception_error_code);
9794 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
9795 vmcs12->vm_entry_instruction_len);
9796 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
9797 vmcs12->guest_interruptibility_info);
9798 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
9799 vmx_set_rflags(vcpu, vmcs12->guest_rflags);
9800 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
9801 vmcs12->guest_pending_dbg_exceptions);
9802 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
9803 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
9805 if (nested_cpu_has_xsaves(vmcs12))
9806 vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
9807 vmcs_write64(VMCS_LINK_POINTER, -1ull);
9809 exec_control = vmcs12->pin_based_vm_exec_control;
9811 /* Preemption timer setting is only taken from vmcs01. */
9812 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
9813 exec_control |= vmcs_config.pin_based_exec_ctrl;
9814 if (vmx->hv_deadline_tsc == -1)
9815 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
9817 /* Posted interrupts setting is only taken from vmcs12. */
9818 if (nested_cpu_has_posted_intr(vmcs12)) {
9820 * Note that we use L0's vector here and in
9821 * vmx_deliver_nested_posted_interrupt.
9823 vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
9824 vmx->nested.pi_pending = false;
9825 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
9826 vmcs_write64(POSTED_INTR_DESC_ADDR,
9827 page_to_phys(vmx->nested.pi_desc_page) +
9828 (unsigned long)(vmcs12->posted_intr_desc_addr &
9831 exec_control &= ~PIN_BASED_POSTED_INTR;
9833 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
9835 vmx->nested.preemption_timer_expired = false;
9836 if (nested_cpu_has_preemption_timer(vmcs12))
9837 vmx_start_preemption_timer(vcpu);
9840 * Whether page-faults are trapped is determined by a combination of
9841 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
9842 * If enable_ept, L0 doesn't care about page faults and we should
9843 * set all of these to L1's desires. However, if !enable_ept, L0 does
9844 * care about (at least some) page faults, and because it is not easy
9845 * (if at all possible?) to merge L0 and L1's desires, we simply ask
9846 * to exit on each and every L2 page fault. This is done by setting
9847 * MASK=MATCH=0 and (see below) EB.PF=1.
9848 * Note that below we don't need special code to set EB.PF beyond the
9849 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
9850 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
9851 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
9853 * A problem with this approach (when !enable_ept) is that L1 may be
9854 * injected with more page faults than it asked for. This could have
9855 * caused problems, but in practice existing hypervisors don't care.
9856 * To fix this, we will need to emulate the PFEC checking (on the L1
9857 * page tables), using walk_addr(), when injecting PFs to L1.
9859 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
9860 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
9861 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
9862 enable_ept ? vmcs12->page_fault_error_code_match : 0);
9864 if (cpu_has_secondary_exec_ctrls()) {
9865 exec_control = vmx_secondary_exec_control(vmx);
9867 /* Take the following fields only from vmcs12 */
9868 exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
9869 SECONDARY_EXEC_RDTSCP |
9870 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
9871 SECONDARY_EXEC_APIC_REGISTER_VIRT |
9872 SECONDARY_EXEC_PCOMMIT);
9873 if (nested_cpu_has(vmcs12,
9874 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
9875 exec_control |= vmcs12->secondary_vm_exec_control;
9877 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) {
9879 * If translation failed, no matter: This feature asks
9880 * to exit when accessing the given address, and if it
9881 * can never be accessed, this feature won't do
9884 if (!vmx->nested.apic_access_page)
9886 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9888 vmcs_write64(APIC_ACCESS_ADDR,
9889 page_to_phys(vmx->nested.apic_access_page));
9890 } else if (!(nested_cpu_has_virt_x2apic_mode(vmcs12)) &&
9891 cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9893 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9894 kvm_vcpu_reload_apic_access_page(vcpu);
9897 if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
9898 vmcs_write64(EOI_EXIT_BITMAP0,
9899 vmcs12->eoi_exit_bitmap0);
9900 vmcs_write64(EOI_EXIT_BITMAP1,
9901 vmcs12->eoi_exit_bitmap1);
9902 vmcs_write64(EOI_EXIT_BITMAP2,
9903 vmcs12->eoi_exit_bitmap2);
9904 vmcs_write64(EOI_EXIT_BITMAP3,
9905 vmcs12->eoi_exit_bitmap3);
9906 vmcs_write16(GUEST_INTR_STATUS,
9907 vmcs12->guest_intr_status);
9910 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
9915 * Set host-state according to L0's settings (vmcs12 is irrelevant here)
9916 * Some constant fields are set here by vmx_set_constant_host_state().
9917 * Other fields are different per CPU, and will be set later when
9918 * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
9920 vmx_set_constant_host_state(vmx);
9923 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
9924 * entry, but only if the current (host) sp changed from the value
9925 * we wrote last (vmx->host_rsp). This cache is no longer relevant
9926 * if we switch vmcs, and rather than hold a separate cache per vmcs,
9927 * here we just force the write to happen on entry.
9931 exec_control = vmx_exec_control(vmx); /* L0's desires */
9932 exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
9933 exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
9934 exec_control &= ~CPU_BASED_TPR_SHADOW;
9935 exec_control |= vmcs12->cpu_based_vm_exec_control;
9937 if (exec_control & CPU_BASED_TPR_SHADOW) {
9938 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
9939 page_to_phys(vmx->nested.virtual_apic_page));
9940 vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
9943 if (cpu_has_vmx_msr_bitmap() &&
9944 exec_control & CPU_BASED_USE_MSR_BITMAPS) {
9945 nested_vmx_merge_msr_bitmap(vcpu, vmcs12);
9946 /* MSR_BITMAP will be set by following vmx_set_efer. */
9948 exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
9951 * Merging of IO bitmap not currently supported.
9952 * Rather, exit every time.
9954 exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
9955 exec_control |= CPU_BASED_UNCOND_IO_EXITING;
9957 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
9959 /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
9960 * bitwise-or of what L1 wants to trap for L2, and what we want to
9961 * trap. Note that CR0.TS also needs updating - we do this later.
9963 update_exception_bitmap(vcpu);
9964 vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
9965 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
9967 /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
9968 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
9969 * bits are further modified by vmx_set_efer() below.
9971 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
9973 /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
9974 * emulated by vmx_set_efer(), below.
9976 vm_entry_controls_init(vmx,
9977 (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER &
9978 ~VM_ENTRY_IA32E_MODE) |
9979 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
9981 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) {
9982 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
9983 vcpu->arch.pat = vmcs12->guest_ia32_pat;
9984 } else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
9985 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
9988 set_cr4_guest_host_mask(vmx);
9990 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)
9991 vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
9993 if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
9994 vmcs_write64(TSC_OFFSET,
9995 vmx->nested.vmcs01_tsc_offset + vmcs12->tsc_offset);
9997 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
10001 * There is no direct mapping between vpid02 and vpid12, the
10002 * vpid02 is per-vCPU for L0 and reused while the value of
10003 * vpid12 is changed w/ one invvpid during nested vmentry.
10004 * The vpid12 is allocated by L1 for L2, so it will not
10005 * influence global bitmap(for vpid01 and vpid02 allocation)
10006 * even if spawn a lot of nested vCPUs.
10008 if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02) {
10009 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
10010 if (vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
10011 vmx->nested.last_vpid = vmcs12->virtual_processor_id;
10012 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->nested.vpid02);
10015 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
10016 vmx_flush_tlb(vcpu);
10021 if (nested_cpu_has_ept(vmcs12)) {
10022 kvm_mmu_unload(vcpu);
10023 nested_ept_init_mmu_context(vcpu);
10026 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)
10027 vcpu->arch.efer = vmcs12->guest_ia32_efer;
10028 else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
10029 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10031 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10032 /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
10033 vmx_set_efer(vcpu, vcpu->arch.efer);
10036 * This sets GUEST_CR0 to vmcs12->guest_cr0, with possibly a modified
10037 * TS bit (for lazy fpu) and bits which we consider mandatory enabled.
10038 * The CR0_READ_SHADOW is what L2 should have expected to read given
10039 * the specifications by L1; It's not enough to take
10040 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
10041 * have more bits than L1 expected.
10043 vmx_set_cr0(vcpu, vmcs12->guest_cr0);
10044 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
10046 vmx_set_cr4(vcpu, vmcs12->guest_cr4);
10047 vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
10049 /* shadow page tables on either EPT or shadow page tables */
10050 kvm_set_cr3(vcpu, vmcs12->guest_cr3);
10051 kvm_mmu_reset_context(vcpu);
10054 vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
10057 * L1 may access the L2's PDPTR, so save them to construct vmcs12
10060 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
10061 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
10062 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
10063 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
10066 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
10067 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
10071 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
10072 * for running an L2 nested guest.
10074 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
10076 struct vmcs12 *vmcs12;
10077 struct vcpu_vmx *vmx = to_vmx(vcpu);
10079 struct loaded_vmcs *vmcs02;
10083 if (!nested_vmx_check_permission(vcpu) ||
10084 !nested_vmx_check_vmcs12(vcpu))
10087 skip_emulated_instruction(vcpu);
10088 vmcs12 = get_vmcs12(vcpu);
10090 if (enable_shadow_vmcs)
10091 copy_shadow_to_vmcs12(vmx);
10094 * The nested entry process starts with enforcing various prerequisites
10095 * on vmcs12 as required by the Intel SDM, and act appropriately when
10096 * they fail: As the SDM explains, some conditions should cause the
10097 * instruction to fail, while others will cause the instruction to seem
10098 * to succeed, but return an EXIT_REASON_INVALID_STATE.
10099 * To speed up the normal (success) code path, we should avoid checking
10100 * for misconfigurations which will anyway be caught by the processor
10101 * when using the merged vmcs02.
10103 if (vmcs12->launch_state == launch) {
10104 nested_vmx_failValid(vcpu,
10105 launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
10106 : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
10110 if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
10111 vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT) {
10112 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
10116 if (!nested_get_vmcs12_pages(vcpu, vmcs12)) {
10117 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
10121 if (nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12)) {
10122 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
10126 if (nested_vmx_check_apicv_controls(vcpu, vmcs12)) {
10127 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
10131 if (nested_vmx_check_msr_switch_controls(vcpu, vmcs12)) {
10132 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
10136 if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
10137 vmx->nested.nested_vmx_true_procbased_ctls_low,
10138 vmx->nested.nested_vmx_procbased_ctls_high) ||
10139 !vmx_control_verify(vmcs12->secondary_vm_exec_control,
10140 vmx->nested.nested_vmx_secondary_ctls_low,
10141 vmx->nested.nested_vmx_secondary_ctls_high) ||
10142 !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
10143 vmx->nested.nested_vmx_pinbased_ctls_low,
10144 vmx->nested.nested_vmx_pinbased_ctls_high) ||
10145 !vmx_control_verify(vmcs12->vm_exit_controls,
10146 vmx->nested.nested_vmx_true_exit_ctls_low,
10147 vmx->nested.nested_vmx_exit_ctls_high) ||
10148 !vmx_control_verify(vmcs12->vm_entry_controls,
10149 vmx->nested.nested_vmx_true_entry_ctls_low,
10150 vmx->nested.nested_vmx_entry_ctls_high))
10152 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
10156 if (((vmcs12->host_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
10157 ((vmcs12->host_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
10158 nested_vmx_failValid(vcpu,
10159 VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
10163 if (!nested_cr0_valid(vcpu, vmcs12->guest_cr0) ||
10164 ((vmcs12->guest_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
10165 nested_vmx_entry_failure(vcpu, vmcs12,
10166 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
10169 if (vmcs12->vmcs_link_pointer != -1ull) {
10170 nested_vmx_entry_failure(vcpu, vmcs12,
10171 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_VMCS_LINK_PTR);
10176 * If the load IA32_EFER VM-entry control is 1, the following checks
10177 * are performed on the field for the IA32_EFER MSR:
10178 * - Bits reserved in the IA32_EFER MSR must be 0.
10179 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
10180 * the IA-32e mode guest VM-exit control. It must also be identical
10181 * to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
10184 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER) {
10185 ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
10186 if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
10187 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
10188 ((vmcs12->guest_cr0 & X86_CR0_PG) &&
10189 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))) {
10190 nested_vmx_entry_failure(vcpu, vmcs12,
10191 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
10197 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
10198 * IA32_EFER MSR must be 0 in the field for that register. In addition,
10199 * the values of the LMA and LME bits in the field must each be that of
10200 * the host address-space size VM-exit control.
10202 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
10203 ia32e = (vmcs12->vm_exit_controls &
10204 VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
10205 if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
10206 ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
10207 ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)) {
10208 nested_vmx_entry_failure(vcpu, vmcs12,
10209 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
10215 * We're finally done with prerequisite checking, and can start with
10216 * the nested entry.
10219 vmcs02 = nested_get_current_vmcs02(vmx);
10223 enter_guest_mode(vcpu);
10225 vmx->nested.vmcs01_tsc_offset = vmcs_read64(TSC_OFFSET);
10227 if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
10228 vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
10231 vmx->loaded_vmcs = vmcs02;
10232 vmx_vcpu_put(vcpu);
10233 vmx_vcpu_load(vcpu, cpu);
10237 vmx_segment_cache_clear(vmx);
10239 prepare_vmcs02(vcpu, vmcs12);
10241 msr_entry_idx = nested_vmx_load_msr(vcpu,
10242 vmcs12->vm_entry_msr_load_addr,
10243 vmcs12->vm_entry_msr_load_count);
10244 if (msr_entry_idx) {
10245 leave_guest_mode(vcpu);
10246 vmx_load_vmcs01(vcpu);
10247 nested_vmx_entry_failure(vcpu, vmcs12,
10248 EXIT_REASON_MSR_LOAD_FAIL, msr_entry_idx);
10252 vmcs12->launch_state = 1;
10254 if (vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT)
10255 return kvm_vcpu_halt(vcpu);
10257 vmx->nested.nested_run_pending = 1;
10260 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
10261 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
10262 * returned as far as L1 is concerned. It will only return (and set
10263 * the success flag) when L2 exits (see nested_vmx_vmexit()).
10269 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
10270 * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
10271 * This function returns the new value we should put in vmcs12.guest_cr0.
10272 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
10273 * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
10274 * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
10275 * didn't trap the bit, because if L1 did, so would L0).
10276 * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have
10277 * been modified by L2, and L1 knows it. So just leave the old value of
10278 * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
10279 * isn't relevant, because if L0 traps this bit it can set it to anything.
10280 * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
10281 * changed these bits, and therefore they need to be updated, but L0
10282 * didn't necessarily allow them to be changed in GUEST_CR0 - and rather
10283 * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
10285 static inline unsigned long
10286 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10289 /*1*/ (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
10290 /*2*/ (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
10291 /*3*/ (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
10292 vcpu->arch.cr0_guest_owned_bits));
10295 static inline unsigned long
10296 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10299 /*1*/ (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
10300 /*2*/ (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
10301 /*3*/ (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
10302 vcpu->arch.cr4_guest_owned_bits));
10305 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
10306 struct vmcs12 *vmcs12)
10311 if (vcpu->arch.exception.pending && vcpu->arch.exception.reinject) {
10312 nr = vcpu->arch.exception.nr;
10313 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
10315 if (kvm_exception_is_soft(nr)) {
10316 vmcs12->vm_exit_instruction_len =
10317 vcpu->arch.event_exit_inst_len;
10318 idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
10320 idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
10322 if (vcpu->arch.exception.has_error_code) {
10323 idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
10324 vmcs12->idt_vectoring_error_code =
10325 vcpu->arch.exception.error_code;
10328 vmcs12->idt_vectoring_info_field = idt_vectoring;
10329 } else if (vcpu->arch.nmi_injected) {
10330 vmcs12->idt_vectoring_info_field =
10331 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
10332 } else if (vcpu->arch.interrupt.pending) {
10333 nr = vcpu->arch.interrupt.nr;
10334 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
10336 if (vcpu->arch.interrupt.soft) {
10337 idt_vectoring |= INTR_TYPE_SOFT_INTR;
10338 vmcs12->vm_entry_instruction_len =
10339 vcpu->arch.event_exit_inst_len;
10341 idt_vectoring |= INTR_TYPE_EXT_INTR;
10343 vmcs12->idt_vectoring_info_field = idt_vectoring;
10347 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
10349 struct vcpu_vmx *vmx = to_vmx(vcpu);
10351 if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
10352 vmx->nested.preemption_timer_expired) {
10353 if (vmx->nested.nested_run_pending)
10355 nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
10359 if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
10360 if (vmx->nested.nested_run_pending ||
10361 vcpu->arch.interrupt.pending)
10363 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
10364 NMI_VECTOR | INTR_TYPE_NMI_INTR |
10365 INTR_INFO_VALID_MASK, 0);
10367 * The NMI-triggered VM exit counts as injection:
10368 * clear this one and block further NMIs.
10370 vcpu->arch.nmi_pending = 0;
10371 vmx_set_nmi_mask(vcpu, true);
10375 if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
10376 nested_exit_on_intr(vcpu)) {
10377 if (vmx->nested.nested_run_pending)
10379 nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
10383 return vmx_complete_nested_posted_interrupt(vcpu);
10386 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
10388 ktime_t remaining =
10389 hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
10392 if (ktime_to_ns(remaining) <= 0)
10395 value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
10396 do_div(value, 1000000);
10397 return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
10401 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
10402 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
10403 * and this function updates it to reflect the changes to the guest state while
10404 * L2 was running (and perhaps made some exits which were handled directly by L0
10405 * without going back to L1), and to reflect the exit reason.
10406 * Note that we do not have to copy here all VMCS fields, just those that
10407 * could have changed by the L2 guest or the exit - i.e., the guest-state and
10408 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
10409 * which already writes to vmcs12 directly.
10411 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10412 u32 exit_reason, u32 exit_intr_info,
10413 unsigned long exit_qualification)
10415 /* update guest state fields: */
10416 vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
10417 vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
10419 vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
10420 vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
10421 vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
10423 vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
10424 vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
10425 vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
10426 vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
10427 vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
10428 vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
10429 vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
10430 vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
10431 vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
10432 vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
10433 vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
10434 vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
10435 vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
10436 vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
10437 vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
10438 vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
10439 vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
10440 vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
10441 vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
10442 vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
10443 vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
10444 vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
10445 vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
10446 vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
10447 vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
10448 vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
10449 vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
10450 vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
10451 vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
10452 vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
10453 vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
10454 vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
10455 vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
10456 vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
10457 vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
10458 vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
10460 vmcs12->guest_interruptibility_info =
10461 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
10462 vmcs12->guest_pending_dbg_exceptions =
10463 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
10464 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
10465 vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
10467 vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
10469 if (nested_cpu_has_preemption_timer(vmcs12)) {
10470 if (vmcs12->vm_exit_controls &
10471 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
10472 vmcs12->vmx_preemption_timer_value =
10473 vmx_get_preemption_timer_value(vcpu);
10474 hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
10478 * In some cases (usually, nested EPT), L2 is allowed to change its
10479 * own CR3 without exiting. If it has changed it, we must keep it.
10480 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
10481 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
10483 * Additionally, restore L2's PDPTR to vmcs12.
10486 vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
10487 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
10488 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
10489 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
10490 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
10493 if (nested_cpu_has_vid(vmcs12))
10494 vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
10496 vmcs12->vm_entry_controls =
10497 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
10498 (vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
10500 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) {
10501 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
10502 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
10505 /* TODO: These cannot have changed unless we have MSR bitmaps and
10506 * the relevant bit asks not to trap the change */
10507 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
10508 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
10509 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
10510 vmcs12->guest_ia32_efer = vcpu->arch.efer;
10511 vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
10512 vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
10513 vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
10514 if (kvm_mpx_supported())
10515 vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
10516 if (nested_cpu_has_xsaves(vmcs12))
10517 vmcs12->xss_exit_bitmap = vmcs_read64(XSS_EXIT_BITMAP);
10519 /* update exit information fields: */
10521 vmcs12->vm_exit_reason = exit_reason;
10522 vmcs12->exit_qualification = exit_qualification;
10524 vmcs12->vm_exit_intr_info = exit_intr_info;
10525 if ((vmcs12->vm_exit_intr_info &
10526 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
10527 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK))
10528 vmcs12->vm_exit_intr_error_code =
10529 vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
10530 vmcs12->idt_vectoring_info_field = 0;
10531 vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
10532 vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
10534 if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
10535 /* vm_entry_intr_info_field is cleared on exit. Emulate this
10536 * instead of reading the real value. */
10537 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
10540 * Transfer the event that L0 or L1 may wanted to inject into
10541 * L2 to IDT_VECTORING_INFO_FIELD.
10543 vmcs12_save_pending_event(vcpu, vmcs12);
10547 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
10548 * preserved above and would only end up incorrectly in L1.
10550 vcpu->arch.nmi_injected = false;
10551 kvm_clear_exception_queue(vcpu);
10552 kvm_clear_interrupt_queue(vcpu);
10556 * A part of what we need to when the nested L2 guest exits and we want to
10557 * run its L1 parent, is to reset L1's guest state to the host state specified
10559 * This function is to be called not only on normal nested exit, but also on
10560 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
10561 * Failures During or After Loading Guest State").
10562 * This function should be called when the active VMCS is L1's (vmcs01).
10564 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
10565 struct vmcs12 *vmcs12)
10567 struct kvm_segment seg;
10569 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
10570 vcpu->arch.efer = vmcs12->host_ia32_efer;
10571 else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
10572 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10574 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10575 vmx_set_efer(vcpu, vcpu->arch.efer);
10577 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
10578 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
10579 vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
10581 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
10582 * actually changed, because it depends on the current state of
10583 * fpu_active (which may have changed).
10584 * Note that vmx_set_cr0 refers to efer set above.
10586 vmx_set_cr0(vcpu, vmcs12->host_cr0);
10588 * If we did fpu_activate()/fpu_deactivate() during L2's run, we need
10589 * to apply the same changes to L1's vmcs. We just set cr0 correctly,
10590 * but we also need to update cr0_guest_host_mask and exception_bitmap.
10592 update_exception_bitmap(vcpu);
10593 vcpu->arch.cr0_guest_owned_bits = (vcpu->fpu_active ? X86_CR0_TS : 0);
10594 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
10597 * Note that CR4_GUEST_HOST_MASK is already set in the original vmcs01
10598 * (KVM doesn't change it)- no reason to call set_cr4_guest_host_mask();
10600 vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
10601 kvm_set_cr4(vcpu, vmcs12->host_cr4);
10603 nested_ept_uninit_mmu_context(vcpu);
10605 kvm_set_cr3(vcpu, vmcs12->host_cr3);
10606 kvm_mmu_reset_context(vcpu);
10609 vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
10613 * Trivially support vpid by letting L2s share their parent
10614 * L1's vpid. TODO: move to a more elaborate solution, giving
10615 * each L2 its own vpid and exposing the vpid feature to L1.
10617 vmx_flush_tlb(vcpu);
10621 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
10622 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
10623 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
10624 vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
10625 vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
10627 /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1. */
10628 if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
10629 vmcs_write64(GUEST_BNDCFGS, 0);
10631 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
10632 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
10633 vcpu->arch.pat = vmcs12->host_ia32_pat;
10635 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
10636 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
10637 vmcs12->host_ia32_perf_global_ctrl);
10639 /* Set L1 segment info according to Intel SDM
10640 27.5.2 Loading Host Segment and Descriptor-Table Registers */
10641 seg = (struct kvm_segment) {
10643 .limit = 0xFFFFFFFF,
10644 .selector = vmcs12->host_cs_selector,
10650 if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
10654 vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
10655 seg = (struct kvm_segment) {
10657 .limit = 0xFFFFFFFF,
10664 seg.selector = vmcs12->host_ds_selector;
10665 vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
10666 seg.selector = vmcs12->host_es_selector;
10667 vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
10668 seg.selector = vmcs12->host_ss_selector;
10669 vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
10670 seg.selector = vmcs12->host_fs_selector;
10671 seg.base = vmcs12->host_fs_base;
10672 vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
10673 seg.selector = vmcs12->host_gs_selector;
10674 seg.base = vmcs12->host_gs_base;
10675 vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
10676 seg = (struct kvm_segment) {
10677 .base = vmcs12->host_tr_base,
10679 .selector = vmcs12->host_tr_selector,
10683 vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
10685 kvm_set_dr(vcpu, 7, 0x400);
10686 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
10688 if (cpu_has_vmx_msr_bitmap())
10689 vmx_set_msr_bitmap(vcpu);
10691 if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
10692 vmcs12->vm_exit_msr_load_count))
10693 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
10697 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
10698 * and modify vmcs12 to make it see what it would expect to see there if
10699 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
10701 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
10702 u32 exit_intr_info,
10703 unsigned long exit_qualification)
10705 struct vcpu_vmx *vmx = to_vmx(vcpu);
10706 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
10708 /* trying to cancel vmlaunch/vmresume is a bug */
10709 WARN_ON_ONCE(vmx->nested.nested_run_pending);
10711 leave_guest_mode(vcpu);
10712 prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
10713 exit_qualification);
10715 if (nested_vmx_store_msr(vcpu, vmcs12->vm_exit_msr_store_addr,
10716 vmcs12->vm_exit_msr_store_count))
10717 nested_vmx_abort(vcpu, VMX_ABORT_SAVE_GUEST_MSR_FAIL);
10719 vmx_load_vmcs01(vcpu);
10721 if ((exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
10722 && nested_exit_intr_ack_set(vcpu)) {
10723 int irq = kvm_cpu_get_interrupt(vcpu);
10725 vmcs12->vm_exit_intr_info = irq |
10726 INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
10729 trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
10730 vmcs12->exit_qualification,
10731 vmcs12->idt_vectoring_info_field,
10732 vmcs12->vm_exit_intr_info,
10733 vmcs12->vm_exit_intr_error_code,
10736 vm_entry_controls_reset_shadow(vmx);
10737 vm_exit_controls_reset_shadow(vmx);
10738 vmx_segment_cache_clear(vmx);
10740 /* if no vmcs02 cache requested, remove the one we used */
10741 if (VMCS02_POOL_SIZE == 0)
10742 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
10744 load_vmcs12_host_state(vcpu, vmcs12);
10746 /* Update any VMCS fields that might have changed while L2 ran */
10747 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
10748 if (vmx->hv_deadline_tsc == -1)
10749 vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
10750 PIN_BASED_VMX_PREEMPTION_TIMER);
10752 vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
10753 PIN_BASED_VMX_PREEMPTION_TIMER);
10755 /* This is needed for same reason as it was needed in prepare_vmcs02 */
10758 /* Unpin physical memory we referred to in vmcs02 */
10759 if (vmx->nested.apic_access_page) {
10760 nested_release_page(vmx->nested.apic_access_page);
10761 vmx->nested.apic_access_page = NULL;
10763 if (vmx->nested.virtual_apic_page) {
10764 nested_release_page(vmx->nested.virtual_apic_page);
10765 vmx->nested.virtual_apic_page = NULL;
10767 if (vmx->nested.pi_desc_page) {
10768 kunmap(vmx->nested.pi_desc_page);
10769 nested_release_page(vmx->nested.pi_desc_page);
10770 vmx->nested.pi_desc_page = NULL;
10771 vmx->nested.pi_desc = NULL;
10775 * We are now running in L2, mmu_notifier will force to reload the
10776 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
10778 kvm_vcpu_reload_apic_access_page(vcpu);
10781 * Exiting from L2 to L1, we're now back to L1 which thinks it just
10782 * finished a VMLAUNCH or VMRESUME instruction, so we need to set the
10783 * success or failure flag accordingly.
10785 if (unlikely(vmx->fail)) {
10787 nested_vmx_failValid(vcpu, vmcs_read32(VM_INSTRUCTION_ERROR));
10789 nested_vmx_succeed(vcpu);
10790 if (enable_shadow_vmcs)
10791 vmx->nested.sync_shadow_vmcs = true;
10793 /* in case we halted in L2 */
10794 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10798 * Forcibly leave nested mode in order to be able to reset the VCPU later on.
10800 static void vmx_leave_nested(struct kvm_vcpu *vcpu)
10802 if (is_guest_mode(vcpu))
10803 nested_vmx_vmexit(vcpu, -1, 0, 0);
10804 free_nested(to_vmx(vcpu));
10808 * L1's failure to enter L2 is a subset of a normal exit, as explained in
10809 * 23.7 "VM-entry failures during or after loading guest state" (this also
10810 * lists the acceptable exit-reason and exit-qualification parameters).
10811 * It should only be called before L2 actually succeeded to run, and when
10812 * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
10814 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
10815 struct vmcs12 *vmcs12,
10816 u32 reason, unsigned long qualification)
10818 load_vmcs12_host_state(vcpu, vmcs12);
10819 vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
10820 vmcs12->exit_qualification = qualification;
10821 nested_vmx_succeed(vcpu);
10822 if (enable_shadow_vmcs)
10823 to_vmx(vcpu)->nested.sync_shadow_vmcs = true;
10826 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
10827 struct x86_instruction_info *info,
10828 enum x86_intercept_stage stage)
10830 return X86EMUL_CONTINUE;
10833 #ifdef CONFIG_X86_64
10834 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
10835 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
10836 u64 divisor, u64 *result)
10838 u64 low = a << shift, high = a >> (64 - shift);
10840 /* To avoid the overflow on divq */
10841 if (high >= divisor)
10844 /* Low hold the result, high hold rem which is discarded */
10845 asm("divq %2\n\t" : "=a" (low), "=d" (high) :
10846 "rm" (divisor), "0" (low), "1" (high));
10852 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc)
10854 struct vcpu_vmx *vmx = to_vmx(vcpu);
10855 u64 tscl = rdtsc();
10856 u64 guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
10857 u64 delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
10859 /* Convert to host delta tsc if tsc scaling is enabled */
10860 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
10861 u64_shl_div_u64(delta_tsc,
10862 kvm_tsc_scaling_ratio_frac_bits,
10863 vcpu->arch.tsc_scaling_ratio,
10868 * If the delta tsc can't fit in the 32 bit after the multi shift,
10869 * we can't use the preemption timer.
10870 * It's possible that it fits on later vmentries, but checking
10871 * on every vmentry is costly so we just use an hrtimer.
10873 if (delta_tsc >> (cpu_preemption_timer_multi + 32))
10876 vmx->hv_deadline_tsc = tscl + delta_tsc;
10877 vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
10878 PIN_BASED_VMX_PREEMPTION_TIMER);
10882 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
10884 struct vcpu_vmx *vmx = to_vmx(vcpu);
10885 vmx->hv_deadline_tsc = -1;
10886 vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
10887 PIN_BASED_VMX_PREEMPTION_TIMER);
10891 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
10894 shrink_ple_window(vcpu);
10897 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
10898 struct kvm_memory_slot *slot)
10900 kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
10901 kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
10904 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
10905 struct kvm_memory_slot *slot)
10907 kvm_mmu_slot_set_dirty(kvm, slot);
10910 static void vmx_flush_log_dirty(struct kvm *kvm)
10912 kvm_flush_pml_buffers(kvm);
10915 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
10916 struct kvm_memory_slot *memslot,
10917 gfn_t offset, unsigned long mask)
10919 kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
10923 * This routine does the following things for vCPU which is going
10924 * to be blocked if VT-d PI is enabled.
10925 * - Store the vCPU to the wakeup list, so when interrupts happen
10926 * we can find the right vCPU to wake up.
10927 * - Change the Posted-interrupt descriptor as below:
10928 * 'NDST' <-- vcpu->pre_pcpu
10929 * 'NV' <-- POSTED_INTR_WAKEUP_VECTOR
10930 * - If 'ON' is set during this process, which means at least one
10931 * interrupt is posted for this vCPU, we cannot block it, in
10932 * this case, return 1, otherwise, return 0.
10935 static int pi_pre_block(struct kvm_vcpu *vcpu)
10937 unsigned long flags;
10939 struct pi_desc old, new;
10940 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
10942 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
10943 !irq_remapping_cap(IRQ_POSTING_CAP))
10946 vcpu->pre_pcpu = vcpu->cpu;
10947 spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
10948 vcpu->pre_pcpu), flags);
10949 list_add_tail(&vcpu->blocked_vcpu_list,
10950 &per_cpu(blocked_vcpu_on_cpu,
10952 spin_unlock_irqrestore(&per_cpu(blocked_vcpu_on_cpu_lock,
10953 vcpu->pre_pcpu), flags);
10956 old.control = new.control = pi_desc->control;
10959 * We should not block the vCPU if
10960 * an interrupt is posted for it.
10962 if (pi_test_on(pi_desc) == 1) {
10963 spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
10964 vcpu->pre_pcpu), flags);
10965 list_del(&vcpu->blocked_vcpu_list);
10966 spin_unlock_irqrestore(
10967 &per_cpu(blocked_vcpu_on_cpu_lock,
10968 vcpu->pre_pcpu), flags);
10969 vcpu->pre_pcpu = -1;
10974 WARN((pi_desc->sn == 1),
10975 "Warning: SN field of posted-interrupts "
10976 "is set before blocking\n");
10979 * Since vCPU can be preempted during this process,
10980 * vcpu->cpu could be different with pre_pcpu, we
10981 * need to set pre_pcpu as the destination of wakeup
10982 * notification event, then we can find the right vCPU
10983 * to wakeup in wakeup handler if interrupts happen
10984 * when the vCPU is in blocked state.
10986 dest = cpu_physical_id(vcpu->pre_pcpu);
10988 if (x2apic_enabled())
10991 new.ndst = (dest << 8) & 0xFF00;
10993 /* set 'NV' to 'wakeup vector' */
10994 new.nv = POSTED_INTR_WAKEUP_VECTOR;
10995 } while (cmpxchg(&pi_desc->control, old.control,
10996 new.control) != old.control);
11001 static int vmx_pre_block(struct kvm_vcpu *vcpu)
11003 if (pi_pre_block(vcpu))
11006 if (kvm_lapic_hv_timer_in_use(vcpu))
11007 kvm_lapic_switch_to_sw_timer(vcpu);
11012 static void pi_post_block(struct kvm_vcpu *vcpu)
11014 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
11015 struct pi_desc old, new;
11017 unsigned long flags;
11019 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
11020 !irq_remapping_cap(IRQ_POSTING_CAP))
11024 old.control = new.control = pi_desc->control;
11026 dest = cpu_physical_id(vcpu->cpu);
11028 if (x2apic_enabled())
11031 new.ndst = (dest << 8) & 0xFF00;
11033 /* Allow posting non-urgent interrupts */
11036 /* set 'NV' to 'notification vector' */
11037 new.nv = POSTED_INTR_VECTOR;
11038 } while (cmpxchg(&pi_desc->control, old.control,
11039 new.control) != old.control);
11041 if(vcpu->pre_pcpu != -1) {
11043 &per_cpu(blocked_vcpu_on_cpu_lock,
11044 vcpu->pre_pcpu), flags);
11045 list_del(&vcpu->blocked_vcpu_list);
11046 spin_unlock_irqrestore(
11047 &per_cpu(blocked_vcpu_on_cpu_lock,
11048 vcpu->pre_pcpu), flags);
11049 vcpu->pre_pcpu = -1;
11053 static void vmx_post_block(struct kvm_vcpu *vcpu)
11055 if (kvm_x86_ops->set_hv_timer)
11056 kvm_lapic_switch_to_hv_timer(vcpu);
11058 pi_post_block(vcpu);
11062 * vmx_update_pi_irte - set IRTE for Posted-Interrupts
11065 * @host_irq: host irq of the interrupt
11066 * @guest_irq: gsi of the interrupt
11067 * @set: set or unset PI
11068 * returns 0 on success, < 0 on failure
11070 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
11071 uint32_t guest_irq, bool set)
11073 struct kvm_kernel_irq_routing_entry *e;
11074 struct kvm_irq_routing_table *irq_rt;
11075 struct kvm_lapic_irq irq;
11076 struct kvm_vcpu *vcpu;
11077 struct vcpu_data vcpu_info;
11078 int idx, ret = -EINVAL;
11080 if (!kvm_arch_has_assigned_device(kvm) ||
11081 !irq_remapping_cap(IRQ_POSTING_CAP))
11084 idx = srcu_read_lock(&kvm->irq_srcu);
11085 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
11086 BUG_ON(guest_irq >= irq_rt->nr_rt_entries);
11088 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
11089 if (e->type != KVM_IRQ_ROUTING_MSI)
11092 * VT-d PI cannot support posting multicast/broadcast
11093 * interrupts to a vCPU, we still use interrupt remapping
11094 * for these kind of interrupts.
11096 * For lowest-priority interrupts, we only support
11097 * those with single CPU as the destination, e.g. user
11098 * configures the interrupts via /proc/irq or uses
11099 * irqbalance to make the interrupts single-CPU.
11101 * We will support full lowest-priority interrupt later.
11104 kvm_set_msi_irq(e, &irq);
11105 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
11107 * Make sure the IRTE is in remapped mode if
11108 * we don't handle it in posted mode.
11110 ret = irq_set_vcpu_affinity(host_irq, NULL);
11113 "failed to back to remapped mode, irq: %u\n",
11121 vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
11122 vcpu_info.vector = irq.vector;
11124 trace_kvm_pi_irte_update(vcpu->vcpu_id, host_irq, e->gsi,
11125 vcpu_info.vector, vcpu_info.pi_desc_addr, set);
11128 ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
11130 /* suppress notification event before unposting */
11131 pi_set_sn(vcpu_to_pi_desc(vcpu));
11132 ret = irq_set_vcpu_affinity(host_irq, NULL);
11133 pi_clear_sn(vcpu_to_pi_desc(vcpu));
11137 printk(KERN_INFO "%s: failed to update PI IRTE\n",
11145 srcu_read_unlock(&kvm->irq_srcu, idx);
11149 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
11151 if (vcpu->arch.mcg_cap & MCG_LMCE_P)
11152 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
11153 FEATURE_CONTROL_LMCE;
11155 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
11156 ~FEATURE_CONTROL_LMCE;
11159 static struct kvm_x86_ops vmx_x86_ops = {
11160 .cpu_has_kvm_support = cpu_has_kvm_support,
11161 .disabled_by_bios = vmx_disabled_by_bios,
11162 .hardware_setup = hardware_setup,
11163 .hardware_unsetup = hardware_unsetup,
11164 .check_processor_compatibility = vmx_check_processor_compat,
11165 .hardware_enable = hardware_enable,
11166 .hardware_disable = hardware_disable,
11167 .cpu_has_accelerated_tpr = report_flexpriority,
11168 .cpu_has_high_real_mode_segbase = vmx_has_high_real_mode_segbase,
11170 .vcpu_create = vmx_create_vcpu,
11171 .vcpu_free = vmx_free_vcpu,
11172 .vcpu_reset = vmx_vcpu_reset,
11174 .prepare_guest_switch = vmx_save_host_state,
11175 .vcpu_load = vmx_vcpu_load,
11176 .vcpu_put = vmx_vcpu_put,
11178 .update_bp_intercept = update_exception_bitmap,
11179 .get_msr = vmx_get_msr,
11180 .set_msr = vmx_set_msr,
11181 .get_segment_base = vmx_get_segment_base,
11182 .get_segment = vmx_get_segment,
11183 .set_segment = vmx_set_segment,
11184 .get_cpl = vmx_get_cpl,
11185 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
11186 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
11187 .decache_cr3 = vmx_decache_cr3,
11188 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
11189 .set_cr0 = vmx_set_cr0,
11190 .set_cr3 = vmx_set_cr3,
11191 .set_cr4 = vmx_set_cr4,
11192 .set_efer = vmx_set_efer,
11193 .get_idt = vmx_get_idt,
11194 .set_idt = vmx_set_idt,
11195 .get_gdt = vmx_get_gdt,
11196 .set_gdt = vmx_set_gdt,
11197 .get_dr6 = vmx_get_dr6,
11198 .set_dr6 = vmx_set_dr6,
11199 .set_dr7 = vmx_set_dr7,
11200 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
11201 .cache_reg = vmx_cache_reg,
11202 .get_rflags = vmx_get_rflags,
11203 .set_rflags = vmx_set_rflags,
11205 .get_pkru = vmx_get_pkru,
11207 .fpu_activate = vmx_fpu_activate,
11208 .fpu_deactivate = vmx_fpu_deactivate,
11210 .tlb_flush = vmx_flush_tlb,
11212 .run = vmx_vcpu_run,
11213 .handle_exit = vmx_handle_exit,
11214 .skip_emulated_instruction = skip_emulated_instruction,
11215 .set_interrupt_shadow = vmx_set_interrupt_shadow,
11216 .get_interrupt_shadow = vmx_get_interrupt_shadow,
11217 .patch_hypercall = vmx_patch_hypercall,
11218 .set_irq = vmx_inject_irq,
11219 .set_nmi = vmx_inject_nmi,
11220 .queue_exception = vmx_queue_exception,
11221 .cancel_injection = vmx_cancel_injection,
11222 .interrupt_allowed = vmx_interrupt_allowed,
11223 .nmi_allowed = vmx_nmi_allowed,
11224 .get_nmi_mask = vmx_get_nmi_mask,
11225 .set_nmi_mask = vmx_set_nmi_mask,
11226 .enable_nmi_window = enable_nmi_window,
11227 .enable_irq_window = enable_irq_window,
11228 .update_cr8_intercept = update_cr8_intercept,
11229 .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode,
11230 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
11231 .get_enable_apicv = vmx_get_enable_apicv,
11232 .refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
11233 .load_eoi_exitmap = vmx_load_eoi_exitmap,
11234 .hwapic_irr_update = vmx_hwapic_irr_update,
11235 .hwapic_isr_update = vmx_hwapic_isr_update,
11236 .sync_pir_to_irr = vmx_sync_pir_to_irr,
11237 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
11239 .set_tss_addr = vmx_set_tss_addr,
11240 .get_tdp_level = get_ept_level,
11241 .get_mt_mask = vmx_get_mt_mask,
11243 .get_exit_info = vmx_get_exit_info,
11245 .get_lpage_level = vmx_get_lpage_level,
11247 .cpuid_update = vmx_cpuid_update,
11249 .rdtscp_supported = vmx_rdtscp_supported,
11250 .invpcid_supported = vmx_invpcid_supported,
11252 .set_supported_cpuid = vmx_set_supported_cpuid,
11254 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
11256 .read_tsc_offset = vmx_read_tsc_offset,
11257 .write_tsc_offset = vmx_write_tsc_offset,
11258 .adjust_tsc_offset_guest = vmx_adjust_tsc_offset_guest,
11259 .read_l1_tsc = vmx_read_l1_tsc,
11261 .set_tdp_cr3 = vmx_set_cr3,
11263 .check_intercept = vmx_check_intercept,
11264 .handle_external_intr = vmx_handle_external_intr,
11265 .mpx_supported = vmx_mpx_supported,
11266 .xsaves_supported = vmx_xsaves_supported,
11268 .check_nested_events = vmx_check_nested_events,
11270 .sched_in = vmx_sched_in,
11272 .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
11273 .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
11274 .flush_log_dirty = vmx_flush_log_dirty,
11275 .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
11277 .pre_block = vmx_pre_block,
11278 .post_block = vmx_post_block,
11280 .pmu_ops = &intel_pmu_ops,
11282 .update_pi_irte = vmx_update_pi_irte,
11284 #ifdef CONFIG_X86_64
11285 .set_hv_timer = vmx_set_hv_timer,
11286 .cancel_hv_timer = vmx_cancel_hv_timer,
11289 .setup_mce = vmx_setup_mce,
11292 static int __init vmx_init(void)
11294 int r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
11295 __alignof__(struct vcpu_vmx), THIS_MODULE);
11299 #ifdef CONFIG_KEXEC_CORE
11300 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
11301 crash_vmclear_local_loaded_vmcss);
11307 static void __exit vmx_exit(void)
11309 #ifdef CONFIG_KEXEC_CORE
11310 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
11317 module_init(vmx_init)
11318 module_exit(vmx_exit)