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 <linux/frame.h>
37 #include "kvm_cache_regs.h"
44 #include <asm/virtext.h>
46 #include <asm/fpu/internal.h>
47 #include <asm/perf_event.h>
48 #include <asm/debugreg.h>
49 #include <asm/kexec.h>
51 #include <asm/irq_remapping.h>
52 #include <asm/mmu_context.h>
57 #define __ex(x) __kvm_handle_fault_on_reboot(x)
58 #define __ex_clear(x, reg) \
59 ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
61 MODULE_AUTHOR("Qumranet");
62 MODULE_LICENSE("GPL");
64 static const struct x86_cpu_id vmx_cpu_id
[] = {
65 X86_FEATURE_MATCH(X86_FEATURE_VMX
),
68 MODULE_DEVICE_TABLE(x86cpu
, vmx_cpu_id
);
70 static bool __read_mostly enable_vpid
= 1;
71 module_param_named(vpid
, enable_vpid
, bool, 0444);
73 static bool __read_mostly flexpriority_enabled
= 1;
74 module_param_named(flexpriority
, flexpriority_enabled
, bool, S_IRUGO
);
76 static bool __read_mostly enable_ept
= 1;
77 module_param_named(ept
, enable_ept
, bool, S_IRUGO
);
79 static bool __read_mostly enable_unrestricted_guest
= 1;
80 module_param_named(unrestricted_guest
,
81 enable_unrestricted_guest
, bool, S_IRUGO
);
83 static bool __read_mostly enable_ept_ad_bits
= 1;
84 module_param_named(eptad
, enable_ept_ad_bits
, bool, S_IRUGO
);
86 static bool __read_mostly emulate_invalid_guest_state
= true;
87 module_param(emulate_invalid_guest_state
, bool, S_IRUGO
);
89 static bool __read_mostly fasteoi
= 1;
90 module_param(fasteoi
, bool, S_IRUGO
);
92 static bool __read_mostly enable_apicv
= 1;
93 module_param(enable_apicv
, bool, S_IRUGO
);
95 static bool __read_mostly enable_shadow_vmcs
= 1;
96 module_param_named(enable_shadow_vmcs
, enable_shadow_vmcs
, bool, S_IRUGO
);
98 * If nested=1, nested virtualization is supported, i.e., guests may use
99 * VMX and be a hypervisor for its own guests. If nested=0, guests may not
100 * use VMX instructions.
102 static bool __read_mostly nested
= 0;
103 module_param(nested
, bool, S_IRUGO
);
105 static u64 __read_mostly host_xss
;
107 static bool __read_mostly enable_pml
= 1;
108 module_param_named(pml
, enable_pml
, bool, S_IRUGO
);
110 #define KVM_VMX_TSC_MULTIPLIER_MAX 0xffffffffffffffffULL
112 /* Guest_tsc -> host_tsc conversion requires 64-bit division. */
113 static int __read_mostly cpu_preemption_timer_multi
;
114 static bool __read_mostly enable_preemption_timer
= 1;
116 module_param_named(preemption_timer
, enable_preemption_timer
, bool, S_IRUGO
);
119 #define KVM_GUEST_CR0_MASK (X86_CR0_NW | X86_CR0_CD)
120 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST (X86_CR0_WP | X86_CR0_NE)
121 #define KVM_VM_CR0_ALWAYS_ON \
122 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
123 #define KVM_CR4_GUEST_OWNED_BITS \
124 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
125 | X86_CR4_OSXMMEXCPT | X86_CR4_TSD)
127 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
128 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
130 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
132 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
135 * Hyper-V requires all of these, so mark them as supported even though
136 * they are just treated the same as all-context.
138 #define VMX_VPID_EXTENT_SUPPORTED_MASK \
139 (VMX_VPID_EXTENT_INDIVIDUAL_ADDR_BIT | \
140 VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT | \
141 VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT | \
142 VMX_VPID_EXTENT_SINGLE_NON_GLOBAL_BIT)
145 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
146 * ple_gap: upper bound on the amount of time between two successive
147 * executions of PAUSE in a loop. Also indicate if ple enabled.
148 * According to test, this time is usually smaller than 128 cycles.
149 * ple_window: upper bound on the amount of time a guest is allowed to execute
150 * in a PAUSE loop. Tests indicate that most spinlocks are held for
151 * less than 2^12 cycles
152 * Time is measured based on a counter that runs at the same rate as the TSC,
153 * refer SDM volume 3b section 21.6.13 & 22.1.3.
155 #define KVM_VMX_DEFAULT_PLE_GAP 128
156 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
157 #define KVM_VMX_DEFAULT_PLE_WINDOW_GROW 2
158 #define KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK 0
159 #define KVM_VMX_DEFAULT_PLE_WINDOW_MAX \
160 INT_MAX / KVM_VMX_DEFAULT_PLE_WINDOW_GROW
162 static int ple_gap
= KVM_VMX_DEFAULT_PLE_GAP
;
163 module_param(ple_gap
, int, S_IRUGO
);
165 static int ple_window
= KVM_VMX_DEFAULT_PLE_WINDOW
;
166 module_param(ple_window
, int, S_IRUGO
);
168 /* Default doubles per-vcpu window every exit. */
169 static int ple_window_grow
= KVM_VMX_DEFAULT_PLE_WINDOW_GROW
;
170 module_param(ple_window_grow
, int, S_IRUGO
);
172 /* Default resets per-vcpu window every exit to ple_window. */
173 static int ple_window_shrink
= KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK
;
174 module_param(ple_window_shrink
, int, S_IRUGO
);
176 /* Default is to compute the maximum so we can never overflow. */
177 static int ple_window_actual_max
= KVM_VMX_DEFAULT_PLE_WINDOW_MAX
;
178 static int ple_window_max
= KVM_VMX_DEFAULT_PLE_WINDOW_MAX
;
179 module_param(ple_window_max
, int, S_IRUGO
);
181 extern const ulong vmx_return
;
183 #define NR_AUTOLOAD_MSRS 8
184 #define VMCS02_POOL_SIZE 1
193 * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
194 * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
195 * loaded on this CPU (so we can clear them if the CPU goes down).
199 struct vmcs
*shadow_vmcs
;
202 struct list_head loaded_vmcss_on_cpu_link
;
205 struct shared_msr_entry
{
212 * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
213 * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
214 * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
215 * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
216 * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
217 * More than one of these structures may exist, if L1 runs multiple L2 guests.
218 * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
219 * underlying hardware which will be used to run L2.
220 * This structure is packed to ensure that its layout is identical across
221 * machines (necessary for live migration).
222 * If there are changes in this struct, VMCS12_REVISION must be changed.
224 typedef u64 natural_width
;
225 struct __packed vmcs12
{
226 /* According to the Intel spec, a VMCS region must start with the
227 * following two fields. Then follow implementation-specific data.
232 u32 launch_state
; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
233 u32 padding
[7]; /* room for future expansion */
238 u64 vm_exit_msr_store_addr
;
239 u64 vm_exit_msr_load_addr
;
240 u64 vm_entry_msr_load_addr
;
242 u64 virtual_apic_page_addr
;
243 u64 apic_access_addr
;
244 u64 posted_intr_desc_addr
;
246 u64 eoi_exit_bitmap0
;
247 u64 eoi_exit_bitmap1
;
248 u64 eoi_exit_bitmap2
;
249 u64 eoi_exit_bitmap3
;
251 u64 guest_physical_address
;
252 u64 vmcs_link_pointer
;
254 u64 guest_ia32_debugctl
;
257 u64 guest_ia32_perf_global_ctrl
;
265 u64 host_ia32_perf_global_ctrl
;
266 u64 padding64
[8]; /* room for future expansion */
268 * To allow migration of L1 (complete with its L2 guests) between
269 * machines of different natural widths (32 or 64 bit), we cannot have
270 * unsigned long fields with no explict size. We use u64 (aliased
271 * natural_width) instead. Luckily, x86 is little-endian.
273 natural_width cr0_guest_host_mask
;
274 natural_width cr4_guest_host_mask
;
275 natural_width cr0_read_shadow
;
276 natural_width cr4_read_shadow
;
277 natural_width cr3_target_value0
;
278 natural_width cr3_target_value1
;
279 natural_width cr3_target_value2
;
280 natural_width cr3_target_value3
;
281 natural_width exit_qualification
;
282 natural_width guest_linear_address
;
283 natural_width guest_cr0
;
284 natural_width guest_cr3
;
285 natural_width guest_cr4
;
286 natural_width guest_es_base
;
287 natural_width guest_cs_base
;
288 natural_width guest_ss_base
;
289 natural_width guest_ds_base
;
290 natural_width guest_fs_base
;
291 natural_width guest_gs_base
;
292 natural_width guest_ldtr_base
;
293 natural_width guest_tr_base
;
294 natural_width guest_gdtr_base
;
295 natural_width guest_idtr_base
;
296 natural_width guest_dr7
;
297 natural_width guest_rsp
;
298 natural_width guest_rip
;
299 natural_width guest_rflags
;
300 natural_width guest_pending_dbg_exceptions
;
301 natural_width guest_sysenter_esp
;
302 natural_width guest_sysenter_eip
;
303 natural_width host_cr0
;
304 natural_width host_cr3
;
305 natural_width host_cr4
;
306 natural_width host_fs_base
;
307 natural_width host_gs_base
;
308 natural_width host_tr_base
;
309 natural_width host_gdtr_base
;
310 natural_width host_idtr_base
;
311 natural_width host_ia32_sysenter_esp
;
312 natural_width host_ia32_sysenter_eip
;
313 natural_width host_rsp
;
314 natural_width host_rip
;
315 natural_width paddingl
[8]; /* room for future expansion */
316 u32 pin_based_vm_exec_control
;
317 u32 cpu_based_vm_exec_control
;
318 u32 exception_bitmap
;
319 u32 page_fault_error_code_mask
;
320 u32 page_fault_error_code_match
;
321 u32 cr3_target_count
;
322 u32 vm_exit_controls
;
323 u32 vm_exit_msr_store_count
;
324 u32 vm_exit_msr_load_count
;
325 u32 vm_entry_controls
;
326 u32 vm_entry_msr_load_count
;
327 u32 vm_entry_intr_info_field
;
328 u32 vm_entry_exception_error_code
;
329 u32 vm_entry_instruction_len
;
331 u32 secondary_vm_exec_control
;
332 u32 vm_instruction_error
;
334 u32 vm_exit_intr_info
;
335 u32 vm_exit_intr_error_code
;
336 u32 idt_vectoring_info_field
;
337 u32 idt_vectoring_error_code
;
338 u32 vm_exit_instruction_len
;
339 u32 vmx_instruction_info
;
346 u32 guest_ldtr_limit
;
348 u32 guest_gdtr_limit
;
349 u32 guest_idtr_limit
;
350 u32 guest_es_ar_bytes
;
351 u32 guest_cs_ar_bytes
;
352 u32 guest_ss_ar_bytes
;
353 u32 guest_ds_ar_bytes
;
354 u32 guest_fs_ar_bytes
;
355 u32 guest_gs_ar_bytes
;
356 u32 guest_ldtr_ar_bytes
;
357 u32 guest_tr_ar_bytes
;
358 u32 guest_interruptibility_info
;
359 u32 guest_activity_state
;
360 u32 guest_sysenter_cs
;
361 u32 host_ia32_sysenter_cs
;
362 u32 vmx_preemption_timer_value
;
363 u32 padding32
[7]; /* room for future expansion */
364 u16 virtual_processor_id
;
366 u16 guest_es_selector
;
367 u16 guest_cs_selector
;
368 u16 guest_ss_selector
;
369 u16 guest_ds_selector
;
370 u16 guest_fs_selector
;
371 u16 guest_gs_selector
;
372 u16 guest_ldtr_selector
;
373 u16 guest_tr_selector
;
374 u16 guest_intr_status
;
376 u16 host_es_selector
;
377 u16 host_cs_selector
;
378 u16 host_ss_selector
;
379 u16 host_ds_selector
;
380 u16 host_fs_selector
;
381 u16 host_gs_selector
;
382 u16 host_tr_selector
;
386 * VMCS12_REVISION is an arbitrary id that should be changed if the content or
387 * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
388 * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
390 #define VMCS12_REVISION 0x11e57ed0
393 * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
394 * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
395 * current implementation, 4K are reserved to avoid future complications.
397 #define VMCS12_SIZE 0x1000
399 /* Used to remember the last vmcs02 used for some recently used vmcs12s */
401 struct list_head list
;
403 struct loaded_vmcs vmcs02
;
407 * The nested_vmx structure is part of vcpu_vmx, and holds information we need
408 * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
411 /* Has the level1 guest done vmxon? */
416 /* The guest-physical address of the current VMCS L1 keeps for L2 */
418 /* The host-usable pointer to the above */
419 struct page
*current_vmcs12_page
;
420 struct vmcs12
*current_vmcs12
;
422 * Cache of the guest's VMCS, existing outside of guest memory.
423 * Loaded from guest memory during VMPTRLD. Flushed to guest
424 * memory during VMXOFF, VMCLEAR, VMPTRLD.
426 struct vmcs12
*cached_vmcs12
;
428 * Indicates if the shadow vmcs must be updated with the
429 * data hold by vmcs12
431 bool sync_shadow_vmcs
;
433 /* vmcs02_list cache of VMCSs recently used to run L2 guests */
434 struct list_head vmcs02_pool
;
436 bool change_vmcs01_virtual_x2apic_mode
;
437 /* L2 must run next, and mustn't decide to exit to L1. */
438 bool nested_run_pending
;
440 * Guest pages referred to in vmcs02 with host-physical pointers, so
441 * we must keep them pinned while L2 runs.
443 struct page
*apic_access_page
;
444 struct page
*virtual_apic_page
;
445 struct page
*pi_desc_page
;
446 struct pi_desc
*pi_desc
;
450 unsigned long *msr_bitmap
;
452 struct hrtimer preemption_timer
;
453 bool preemption_timer_expired
;
455 /* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
462 * We only store the "true" versions of the VMX capability MSRs. We
463 * generate the "non-true" versions by setting the must-be-1 bits
464 * according to the SDM.
466 u32 nested_vmx_procbased_ctls_low
;
467 u32 nested_vmx_procbased_ctls_high
;
468 u32 nested_vmx_secondary_ctls_low
;
469 u32 nested_vmx_secondary_ctls_high
;
470 u32 nested_vmx_pinbased_ctls_low
;
471 u32 nested_vmx_pinbased_ctls_high
;
472 u32 nested_vmx_exit_ctls_low
;
473 u32 nested_vmx_exit_ctls_high
;
474 u32 nested_vmx_entry_ctls_low
;
475 u32 nested_vmx_entry_ctls_high
;
476 u32 nested_vmx_misc_low
;
477 u32 nested_vmx_misc_high
;
478 u32 nested_vmx_ept_caps
;
479 u32 nested_vmx_vpid_caps
;
480 u64 nested_vmx_basic
;
481 u64 nested_vmx_cr0_fixed0
;
482 u64 nested_vmx_cr0_fixed1
;
483 u64 nested_vmx_cr4_fixed0
;
484 u64 nested_vmx_cr4_fixed1
;
485 u64 nested_vmx_vmcs_enum
;
488 #define POSTED_INTR_ON 0
489 #define POSTED_INTR_SN 1
491 /* Posted-Interrupt Descriptor */
493 u32 pir
[8]; /* Posted interrupt requested */
496 /* bit 256 - Outstanding Notification */
498 /* bit 257 - Suppress Notification */
500 /* bit 271:258 - Reserved */
502 /* bit 279:272 - Notification Vector */
504 /* bit 287:280 - Reserved */
506 /* bit 319:288 - Notification Destination */
514 static bool pi_test_and_set_on(struct pi_desc
*pi_desc
)
516 return test_and_set_bit(POSTED_INTR_ON
,
517 (unsigned long *)&pi_desc
->control
);
520 static bool pi_test_and_clear_on(struct pi_desc
*pi_desc
)
522 return test_and_clear_bit(POSTED_INTR_ON
,
523 (unsigned long *)&pi_desc
->control
);
526 static int pi_test_and_set_pir(int vector
, struct pi_desc
*pi_desc
)
528 return test_and_set_bit(vector
, (unsigned long *)pi_desc
->pir
);
531 static inline void pi_clear_sn(struct pi_desc
*pi_desc
)
533 return clear_bit(POSTED_INTR_SN
,
534 (unsigned long *)&pi_desc
->control
);
537 static inline void pi_set_sn(struct pi_desc
*pi_desc
)
539 return set_bit(POSTED_INTR_SN
,
540 (unsigned long *)&pi_desc
->control
);
543 static inline void pi_clear_on(struct pi_desc
*pi_desc
)
545 clear_bit(POSTED_INTR_ON
,
546 (unsigned long *)&pi_desc
->control
);
549 static inline int pi_test_on(struct pi_desc
*pi_desc
)
551 return test_bit(POSTED_INTR_ON
,
552 (unsigned long *)&pi_desc
->control
);
555 static inline int pi_test_sn(struct pi_desc
*pi_desc
)
557 return test_bit(POSTED_INTR_SN
,
558 (unsigned long *)&pi_desc
->control
);
562 struct kvm_vcpu vcpu
;
563 unsigned long host_rsp
;
565 bool nmi_known_unmasked
;
567 u32 idt_vectoring_info
;
569 struct shared_msr_entry
*guest_msrs
;
572 unsigned long host_idt_base
;
574 u64 msr_host_kernel_gs_base
;
575 u64 msr_guest_kernel_gs_base
;
577 u32 vm_entry_controls_shadow
;
578 u32 vm_exit_controls_shadow
;
580 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
581 * non-nested (L1) guest, it always points to vmcs01. For a nested
582 * guest (L2), it points to a different VMCS.
584 struct loaded_vmcs vmcs01
;
585 struct loaded_vmcs
*loaded_vmcs
;
586 bool __launched
; /* temporary, used in vmx_vcpu_run */
587 struct msr_autoload
{
589 struct vmx_msr_entry guest
[NR_AUTOLOAD_MSRS
];
590 struct vmx_msr_entry host
[NR_AUTOLOAD_MSRS
];
594 u16 fs_sel
, gs_sel
, ldt_sel
;
598 int gs_ldt_reload_needed
;
599 int fs_reload_needed
;
600 u64 msr_host_bndcfgs
;
601 unsigned long vmcs_host_cr3
; /* May not match real cr3 */
602 unsigned long vmcs_host_cr4
; /* May not match real cr4 */
607 struct kvm_segment segs
[8];
610 u32 bitmask
; /* 4 bits per segment (1 bit per field) */
611 struct kvm_save_segment
{
619 bool emulation_required
;
623 /* Posted interrupt descriptor */
624 struct pi_desc pi_desc
;
626 /* Support for a guest hypervisor (nested VMX) */
627 struct nested_vmx nested
;
629 /* Dynamic PLE window. */
631 bool ple_window_dirty
;
633 /* Support for PML */
634 #define PML_ENTITY_NUM 512
637 /* apic deadline value in host tsc */
640 u64 current_tsc_ratio
;
642 bool guest_pkru_valid
;
647 * Only bits masked by msr_ia32_feature_control_valid_bits can be set in
648 * msr_ia32_feature_control. FEATURE_CONTROL_LOCKED is always included
649 * in msr_ia32_feature_control_valid_bits.
651 u64 msr_ia32_feature_control
;
652 u64 msr_ia32_feature_control_valid_bits
;
655 enum segment_cache_field
{
664 static inline struct vcpu_vmx
*to_vmx(struct kvm_vcpu
*vcpu
)
666 return container_of(vcpu
, struct vcpu_vmx
, vcpu
);
669 static struct pi_desc
*vcpu_to_pi_desc(struct kvm_vcpu
*vcpu
)
671 return &(to_vmx(vcpu
)->pi_desc
);
674 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
675 #define FIELD(number, name) [number] = VMCS12_OFFSET(name)
676 #define FIELD64(number, name) [number] = VMCS12_OFFSET(name), \
677 [number##_HIGH] = VMCS12_OFFSET(name)+4
680 static unsigned long shadow_read_only_fields
[] = {
682 * We do NOT shadow fields that are modified when L0
683 * traps and emulates any vmx instruction (e.g. VMPTRLD,
684 * VMXON...) executed by L1.
685 * For example, VM_INSTRUCTION_ERROR is read
686 * by L1 if a vmx instruction fails (part of the error path).
687 * Note the code assumes this logic. If for some reason
688 * we start shadowing these fields then we need to
689 * force a shadow sync when L0 emulates vmx instructions
690 * (e.g. force a sync if VM_INSTRUCTION_ERROR is modified
691 * by nested_vmx_failValid)
695 VM_EXIT_INSTRUCTION_LEN
,
696 IDT_VECTORING_INFO_FIELD
,
697 IDT_VECTORING_ERROR_CODE
,
698 VM_EXIT_INTR_ERROR_CODE
,
700 GUEST_LINEAR_ADDRESS
,
701 GUEST_PHYSICAL_ADDRESS
703 static int max_shadow_read_only_fields
=
704 ARRAY_SIZE(shadow_read_only_fields
);
706 static unsigned long shadow_read_write_fields
[] = {
713 GUEST_INTERRUPTIBILITY_INFO
,
726 CPU_BASED_VM_EXEC_CONTROL
,
727 VM_ENTRY_EXCEPTION_ERROR_CODE
,
728 VM_ENTRY_INTR_INFO_FIELD
,
729 VM_ENTRY_INSTRUCTION_LEN
,
730 VM_ENTRY_EXCEPTION_ERROR_CODE
,
736 static int max_shadow_read_write_fields
=
737 ARRAY_SIZE(shadow_read_write_fields
);
739 static const unsigned short vmcs_field_to_offset_table
[] = {
740 FIELD(VIRTUAL_PROCESSOR_ID
, virtual_processor_id
),
741 FIELD(POSTED_INTR_NV
, posted_intr_nv
),
742 FIELD(GUEST_ES_SELECTOR
, guest_es_selector
),
743 FIELD(GUEST_CS_SELECTOR
, guest_cs_selector
),
744 FIELD(GUEST_SS_SELECTOR
, guest_ss_selector
),
745 FIELD(GUEST_DS_SELECTOR
, guest_ds_selector
),
746 FIELD(GUEST_FS_SELECTOR
, guest_fs_selector
),
747 FIELD(GUEST_GS_SELECTOR
, guest_gs_selector
),
748 FIELD(GUEST_LDTR_SELECTOR
, guest_ldtr_selector
),
749 FIELD(GUEST_TR_SELECTOR
, guest_tr_selector
),
750 FIELD(GUEST_INTR_STATUS
, guest_intr_status
),
751 FIELD(GUEST_PML_INDEX
, guest_pml_index
),
752 FIELD(HOST_ES_SELECTOR
, host_es_selector
),
753 FIELD(HOST_CS_SELECTOR
, host_cs_selector
),
754 FIELD(HOST_SS_SELECTOR
, host_ss_selector
),
755 FIELD(HOST_DS_SELECTOR
, host_ds_selector
),
756 FIELD(HOST_FS_SELECTOR
, host_fs_selector
),
757 FIELD(HOST_GS_SELECTOR
, host_gs_selector
),
758 FIELD(HOST_TR_SELECTOR
, host_tr_selector
),
759 FIELD64(IO_BITMAP_A
, io_bitmap_a
),
760 FIELD64(IO_BITMAP_B
, io_bitmap_b
),
761 FIELD64(MSR_BITMAP
, msr_bitmap
),
762 FIELD64(VM_EXIT_MSR_STORE_ADDR
, vm_exit_msr_store_addr
),
763 FIELD64(VM_EXIT_MSR_LOAD_ADDR
, vm_exit_msr_load_addr
),
764 FIELD64(VM_ENTRY_MSR_LOAD_ADDR
, vm_entry_msr_load_addr
),
765 FIELD64(TSC_OFFSET
, tsc_offset
),
766 FIELD64(VIRTUAL_APIC_PAGE_ADDR
, virtual_apic_page_addr
),
767 FIELD64(APIC_ACCESS_ADDR
, apic_access_addr
),
768 FIELD64(POSTED_INTR_DESC_ADDR
, posted_intr_desc_addr
),
769 FIELD64(EPT_POINTER
, ept_pointer
),
770 FIELD64(EOI_EXIT_BITMAP0
, eoi_exit_bitmap0
),
771 FIELD64(EOI_EXIT_BITMAP1
, eoi_exit_bitmap1
),
772 FIELD64(EOI_EXIT_BITMAP2
, eoi_exit_bitmap2
),
773 FIELD64(EOI_EXIT_BITMAP3
, eoi_exit_bitmap3
),
774 FIELD64(XSS_EXIT_BITMAP
, xss_exit_bitmap
),
775 FIELD64(GUEST_PHYSICAL_ADDRESS
, guest_physical_address
),
776 FIELD64(VMCS_LINK_POINTER
, vmcs_link_pointer
),
777 FIELD64(PML_ADDRESS
, pml_address
),
778 FIELD64(GUEST_IA32_DEBUGCTL
, guest_ia32_debugctl
),
779 FIELD64(GUEST_IA32_PAT
, guest_ia32_pat
),
780 FIELD64(GUEST_IA32_EFER
, guest_ia32_efer
),
781 FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL
, guest_ia32_perf_global_ctrl
),
782 FIELD64(GUEST_PDPTR0
, guest_pdptr0
),
783 FIELD64(GUEST_PDPTR1
, guest_pdptr1
),
784 FIELD64(GUEST_PDPTR2
, guest_pdptr2
),
785 FIELD64(GUEST_PDPTR3
, guest_pdptr3
),
786 FIELD64(GUEST_BNDCFGS
, guest_bndcfgs
),
787 FIELD64(HOST_IA32_PAT
, host_ia32_pat
),
788 FIELD64(HOST_IA32_EFER
, host_ia32_efer
),
789 FIELD64(HOST_IA32_PERF_GLOBAL_CTRL
, host_ia32_perf_global_ctrl
),
790 FIELD(PIN_BASED_VM_EXEC_CONTROL
, pin_based_vm_exec_control
),
791 FIELD(CPU_BASED_VM_EXEC_CONTROL
, cpu_based_vm_exec_control
),
792 FIELD(EXCEPTION_BITMAP
, exception_bitmap
),
793 FIELD(PAGE_FAULT_ERROR_CODE_MASK
, page_fault_error_code_mask
),
794 FIELD(PAGE_FAULT_ERROR_CODE_MATCH
, page_fault_error_code_match
),
795 FIELD(CR3_TARGET_COUNT
, cr3_target_count
),
796 FIELD(VM_EXIT_CONTROLS
, vm_exit_controls
),
797 FIELD(VM_EXIT_MSR_STORE_COUNT
, vm_exit_msr_store_count
),
798 FIELD(VM_EXIT_MSR_LOAD_COUNT
, vm_exit_msr_load_count
),
799 FIELD(VM_ENTRY_CONTROLS
, vm_entry_controls
),
800 FIELD(VM_ENTRY_MSR_LOAD_COUNT
, vm_entry_msr_load_count
),
801 FIELD(VM_ENTRY_INTR_INFO_FIELD
, vm_entry_intr_info_field
),
802 FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE
, vm_entry_exception_error_code
),
803 FIELD(VM_ENTRY_INSTRUCTION_LEN
, vm_entry_instruction_len
),
804 FIELD(TPR_THRESHOLD
, tpr_threshold
),
805 FIELD(SECONDARY_VM_EXEC_CONTROL
, secondary_vm_exec_control
),
806 FIELD(VM_INSTRUCTION_ERROR
, vm_instruction_error
),
807 FIELD(VM_EXIT_REASON
, vm_exit_reason
),
808 FIELD(VM_EXIT_INTR_INFO
, vm_exit_intr_info
),
809 FIELD(VM_EXIT_INTR_ERROR_CODE
, vm_exit_intr_error_code
),
810 FIELD(IDT_VECTORING_INFO_FIELD
, idt_vectoring_info_field
),
811 FIELD(IDT_VECTORING_ERROR_CODE
, idt_vectoring_error_code
),
812 FIELD(VM_EXIT_INSTRUCTION_LEN
, vm_exit_instruction_len
),
813 FIELD(VMX_INSTRUCTION_INFO
, vmx_instruction_info
),
814 FIELD(GUEST_ES_LIMIT
, guest_es_limit
),
815 FIELD(GUEST_CS_LIMIT
, guest_cs_limit
),
816 FIELD(GUEST_SS_LIMIT
, guest_ss_limit
),
817 FIELD(GUEST_DS_LIMIT
, guest_ds_limit
),
818 FIELD(GUEST_FS_LIMIT
, guest_fs_limit
),
819 FIELD(GUEST_GS_LIMIT
, guest_gs_limit
),
820 FIELD(GUEST_LDTR_LIMIT
, guest_ldtr_limit
),
821 FIELD(GUEST_TR_LIMIT
, guest_tr_limit
),
822 FIELD(GUEST_GDTR_LIMIT
, guest_gdtr_limit
),
823 FIELD(GUEST_IDTR_LIMIT
, guest_idtr_limit
),
824 FIELD(GUEST_ES_AR_BYTES
, guest_es_ar_bytes
),
825 FIELD(GUEST_CS_AR_BYTES
, guest_cs_ar_bytes
),
826 FIELD(GUEST_SS_AR_BYTES
, guest_ss_ar_bytes
),
827 FIELD(GUEST_DS_AR_BYTES
, guest_ds_ar_bytes
),
828 FIELD(GUEST_FS_AR_BYTES
, guest_fs_ar_bytes
),
829 FIELD(GUEST_GS_AR_BYTES
, guest_gs_ar_bytes
),
830 FIELD(GUEST_LDTR_AR_BYTES
, guest_ldtr_ar_bytes
),
831 FIELD(GUEST_TR_AR_BYTES
, guest_tr_ar_bytes
),
832 FIELD(GUEST_INTERRUPTIBILITY_INFO
, guest_interruptibility_info
),
833 FIELD(GUEST_ACTIVITY_STATE
, guest_activity_state
),
834 FIELD(GUEST_SYSENTER_CS
, guest_sysenter_cs
),
835 FIELD(HOST_IA32_SYSENTER_CS
, host_ia32_sysenter_cs
),
836 FIELD(VMX_PREEMPTION_TIMER_VALUE
, vmx_preemption_timer_value
),
837 FIELD(CR0_GUEST_HOST_MASK
, cr0_guest_host_mask
),
838 FIELD(CR4_GUEST_HOST_MASK
, cr4_guest_host_mask
),
839 FIELD(CR0_READ_SHADOW
, cr0_read_shadow
),
840 FIELD(CR4_READ_SHADOW
, cr4_read_shadow
),
841 FIELD(CR3_TARGET_VALUE0
, cr3_target_value0
),
842 FIELD(CR3_TARGET_VALUE1
, cr3_target_value1
),
843 FIELD(CR3_TARGET_VALUE2
, cr3_target_value2
),
844 FIELD(CR3_TARGET_VALUE3
, cr3_target_value3
),
845 FIELD(EXIT_QUALIFICATION
, exit_qualification
),
846 FIELD(GUEST_LINEAR_ADDRESS
, guest_linear_address
),
847 FIELD(GUEST_CR0
, guest_cr0
),
848 FIELD(GUEST_CR3
, guest_cr3
),
849 FIELD(GUEST_CR4
, guest_cr4
),
850 FIELD(GUEST_ES_BASE
, guest_es_base
),
851 FIELD(GUEST_CS_BASE
, guest_cs_base
),
852 FIELD(GUEST_SS_BASE
, guest_ss_base
),
853 FIELD(GUEST_DS_BASE
, guest_ds_base
),
854 FIELD(GUEST_FS_BASE
, guest_fs_base
),
855 FIELD(GUEST_GS_BASE
, guest_gs_base
),
856 FIELD(GUEST_LDTR_BASE
, guest_ldtr_base
),
857 FIELD(GUEST_TR_BASE
, guest_tr_base
),
858 FIELD(GUEST_GDTR_BASE
, guest_gdtr_base
),
859 FIELD(GUEST_IDTR_BASE
, guest_idtr_base
),
860 FIELD(GUEST_DR7
, guest_dr7
),
861 FIELD(GUEST_RSP
, guest_rsp
),
862 FIELD(GUEST_RIP
, guest_rip
),
863 FIELD(GUEST_RFLAGS
, guest_rflags
),
864 FIELD(GUEST_PENDING_DBG_EXCEPTIONS
, guest_pending_dbg_exceptions
),
865 FIELD(GUEST_SYSENTER_ESP
, guest_sysenter_esp
),
866 FIELD(GUEST_SYSENTER_EIP
, guest_sysenter_eip
),
867 FIELD(HOST_CR0
, host_cr0
),
868 FIELD(HOST_CR3
, host_cr3
),
869 FIELD(HOST_CR4
, host_cr4
),
870 FIELD(HOST_FS_BASE
, host_fs_base
),
871 FIELD(HOST_GS_BASE
, host_gs_base
),
872 FIELD(HOST_TR_BASE
, host_tr_base
),
873 FIELD(HOST_GDTR_BASE
, host_gdtr_base
),
874 FIELD(HOST_IDTR_BASE
, host_idtr_base
),
875 FIELD(HOST_IA32_SYSENTER_ESP
, host_ia32_sysenter_esp
),
876 FIELD(HOST_IA32_SYSENTER_EIP
, host_ia32_sysenter_eip
),
877 FIELD(HOST_RSP
, host_rsp
),
878 FIELD(HOST_RIP
, host_rip
),
881 static inline short vmcs_field_to_offset(unsigned long field
)
883 BUILD_BUG_ON(ARRAY_SIZE(vmcs_field_to_offset_table
) > SHRT_MAX
);
885 if (field
>= ARRAY_SIZE(vmcs_field_to_offset_table
) ||
886 vmcs_field_to_offset_table
[field
] == 0)
889 return vmcs_field_to_offset_table
[field
];
892 static inline struct vmcs12
*get_vmcs12(struct kvm_vcpu
*vcpu
)
894 return to_vmx(vcpu
)->nested
.cached_vmcs12
;
897 static struct page
*nested_get_page(struct kvm_vcpu
*vcpu
, gpa_t addr
)
899 struct page
*page
= kvm_vcpu_gfn_to_page(vcpu
, addr
>> PAGE_SHIFT
);
900 if (is_error_page(page
))
906 static void nested_release_page(struct page
*page
)
908 kvm_release_page_dirty(page
);
911 static void nested_release_page_clean(struct page
*page
)
913 kvm_release_page_clean(page
);
916 static bool nested_ept_ad_enabled(struct kvm_vcpu
*vcpu
);
917 static unsigned long nested_ept_get_cr3(struct kvm_vcpu
*vcpu
);
918 static u64
construct_eptp(struct kvm_vcpu
*vcpu
, unsigned long root_hpa
);
919 static bool vmx_xsaves_supported(void);
920 static int vmx_set_tss_addr(struct kvm
*kvm
, unsigned int addr
);
921 static void vmx_set_segment(struct kvm_vcpu
*vcpu
,
922 struct kvm_segment
*var
, int seg
);
923 static void vmx_get_segment(struct kvm_vcpu
*vcpu
,
924 struct kvm_segment
*var
, int seg
);
925 static bool guest_state_valid(struct kvm_vcpu
*vcpu
);
926 static u32
vmx_segment_access_rights(struct kvm_segment
*var
);
927 static void copy_vmcs12_to_shadow(struct vcpu_vmx
*vmx
);
928 static void copy_shadow_to_vmcs12(struct vcpu_vmx
*vmx
);
929 static int alloc_identity_pagetable(struct kvm
*kvm
);
931 static DEFINE_PER_CPU(struct vmcs
*, vmxarea
);
932 static DEFINE_PER_CPU(struct vmcs
*, current_vmcs
);
934 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
935 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
937 static DEFINE_PER_CPU(struct list_head
, loaded_vmcss_on_cpu
);
940 * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
941 * can find which vCPU should be waken up.
943 static DEFINE_PER_CPU(struct list_head
, blocked_vcpu_on_cpu
);
944 static DEFINE_PER_CPU(spinlock_t
, blocked_vcpu_on_cpu_lock
);
949 VMX_MSR_BITMAP_LEGACY
,
950 VMX_MSR_BITMAP_LONGMODE
,
951 VMX_MSR_BITMAP_LEGACY_X2APIC_APICV
,
952 VMX_MSR_BITMAP_LONGMODE_X2APIC_APICV
,
953 VMX_MSR_BITMAP_LEGACY_X2APIC
,
954 VMX_MSR_BITMAP_LONGMODE_X2APIC
,
960 static unsigned long *vmx_bitmap
[VMX_BITMAP_NR
];
962 #define vmx_io_bitmap_a (vmx_bitmap[VMX_IO_BITMAP_A])
963 #define vmx_io_bitmap_b (vmx_bitmap[VMX_IO_BITMAP_B])
964 #define vmx_msr_bitmap_legacy (vmx_bitmap[VMX_MSR_BITMAP_LEGACY])
965 #define vmx_msr_bitmap_longmode (vmx_bitmap[VMX_MSR_BITMAP_LONGMODE])
966 #define vmx_msr_bitmap_legacy_x2apic_apicv (vmx_bitmap[VMX_MSR_BITMAP_LEGACY_X2APIC_APICV])
967 #define vmx_msr_bitmap_longmode_x2apic_apicv (vmx_bitmap[VMX_MSR_BITMAP_LONGMODE_X2APIC_APICV])
968 #define vmx_msr_bitmap_legacy_x2apic (vmx_bitmap[VMX_MSR_BITMAP_LEGACY_X2APIC])
969 #define vmx_msr_bitmap_longmode_x2apic (vmx_bitmap[VMX_MSR_BITMAP_LONGMODE_X2APIC])
970 #define vmx_vmread_bitmap (vmx_bitmap[VMX_VMREAD_BITMAP])
971 #define vmx_vmwrite_bitmap (vmx_bitmap[VMX_VMWRITE_BITMAP])
973 static bool cpu_has_load_ia32_efer
;
974 static bool cpu_has_load_perf_global_ctrl
;
976 static DECLARE_BITMAP(vmx_vpid_bitmap
, VMX_NR_VPIDS
);
977 static DEFINE_SPINLOCK(vmx_vpid_lock
);
979 static struct vmcs_config
{
984 u32 pin_based_exec_ctrl
;
985 u32 cpu_based_exec_ctrl
;
986 u32 cpu_based_2nd_exec_ctrl
;
991 static struct vmx_capability
{
996 #define VMX_SEGMENT_FIELD(seg) \
997 [VCPU_SREG_##seg] = { \
998 .selector = GUEST_##seg##_SELECTOR, \
999 .base = GUEST_##seg##_BASE, \
1000 .limit = GUEST_##seg##_LIMIT, \
1001 .ar_bytes = GUEST_##seg##_AR_BYTES, \
1004 static const struct kvm_vmx_segment_field
{
1009 } kvm_vmx_segment_fields
[] = {
1010 VMX_SEGMENT_FIELD(CS
),
1011 VMX_SEGMENT_FIELD(DS
),
1012 VMX_SEGMENT_FIELD(ES
),
1013 VMX_SEGMENT_FIELD(FS
),
1014 VMX_SEGMENT_FIELD(GS
),
1015 VMX_SEGMENT_FIELD(SS
),
1016 VMX_SEGMENT_FIELD(TR
),
1017 VMX_SEGMENT_FIELD(LDTR
),
1020 static u64 host_efer
;
1022 static void ept_save_pdptrs(struct kvm_vcpu
*vcpu
);
1025 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
1026 * away by decrementing the array size.
1028 static const u32 vmx_msr_index
[] = {
1029 #ifdef CONFIG_X86_64
1030 MSR_SYSCALL_MASK
, MSR_LSTAR
, MSR_CSTAR
,
1032 MSR_EFER
, MSR_TSC_AUX
, MSR_STAR
,
1035 static inline bool is_exception_n(u32 intr_info
, u8 vector
)
1037 return (intr_info
& (INTR_INFO_INTR_TYPE_MASK
| INTR_INFO_VECTOR_MASK
|
1038 INTR_INFO_VALID_MASK
)) ==
1039 (INTR_TYPE_HARD_EXCEPTION
| vector
| INTR_INFO_VALID_MASK
);
1042 static inline bool is_debug(u32 intr_info
)
1044 return is_exception_n(intr_info
, DB_VECTOR
);
1047 static inline bool is_breakpoint(u32 intr_info
)
1049 return is_exception_n(intr_info
, BP_VECTOR
);
1052 static inline bool is_page_fault(u32 intr_info
)
1054 return is_exception_n(intr_info
, PF_VECTOR
);
1057 static inline bool is_no_device(u32 intr_info
)
1059 return is_exception_n(intr_info
, NM_VECTOR
);
1062 static inline bool is_invalid_opcode(u32 intr_info
)
1064 return is_exception_n(intr_info
, UD_VECTOR
);
1067 static inline bool is_external_interrupt(u32 intr_info
)
1069 return (intr_info
& (INTR_INFO_INTR_TYPE_MASK
| INTR_INFO_VALID_MASK
))
1070 == (INTR_TYPE_EXT_INTR
| INTR_INFO_VALID_MASK
);
1073 static inline bool is_machine_check(u32 intr_info
)
1075 return (intr_info
& (INTR_INFO_INTR_TYPE_MASK
| INTR_INFO_VECTOR_MASK
|
1076 INTR_INFO_VALID_MASK
)) ==
1077 (INTR_TYPE_HARD_EXCEPTION
| MC_VECTOR
| INTR_INFO_VALID_MASK
);
1080 static inline bool cpu_has_vmx_msr_bitmap(void)
1082 return vmcs_config
.cpu_based_exec_ctrl
& CPU_BASED_USE_MSR_BITMAPS
;
1085 static inline bool cpu_has_vmx_tpr_shadow(void)
1087 return vmcs_config
.cpu_based_exec_ctrl
& CPU_BASED_TPR_SHADOW
;
1090 static inline bool cpu_need_tpr_shadow(struct kvm_vcpu
*vcpu
)
1092 return cpu_has_vmx_tpr_shadow() && lapic_in_kernel(vcpu
);
1095 static inline bool cpu_has_secondary_exec_ctrls(void)
1097 return vmcs_config
.cpu_based_exec_ctrl
&
1098 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS
;
1101 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
1103 return vmcs_config
.cpu_based_2nd_exec_ctrl
&
1104 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES
;
1107 static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
1109 return vmcs_config
.cpu_based_2nd_exec_ctrl
&
1110 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE
;
1113 static inline bool cpu_has_vmx_apic_register_virt(void)
1115 return vmcs_config
.cpu_based_2nd_exec_ctrl
&
1116 SECONDARY_EXEC_APIC_REGISTER_VIRT
;
1119 static inline bool cpu_has_vmx_virtual_intr_delivery(void)
1121 return vmcs_config
.cpu_based_2nd_exec_ctrl
&
1122 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY
;
1126 * Comment's format: document - errata name - stepping - processor name.
1128 * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
1130 static u32 vmx_preemption_cpu_tfms
[] = {
1131 /* 323344.pdf - BA86 - D0 - Xeon 7500 Series */
1133 /* 323056.pdf - AAX65 - C2 - Xeon L3406 */
1134 /* 322814.pdf - AAT59 - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
1135 /* 322911.pdf - AAU65 - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
1137 /* 322911.pdf - AAU65 - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
1139 /* 322373.pdf - AAO95 - B1 - Xeon 3400 Series */
1140 /* 322166.pdf - AAN92 - B1 - i7-800 and i5-700 Desktop */
1142 * 320767.pdf - AAP86 - B1 -
1143 * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
1146 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
1148 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
1150 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
1152 /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
1153 /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
1154 /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
1158 static inline bool cpu_has_broken_vmx_preemption_timer(void)
1160 u32 eax
= cpuid_eax(0x00000001), i
;
1162 /* Clear the reserved bits */
1163 eax
&= ~(0x3U
<< 14 | 0xfU
<< 28);
1164 for (i
= 0; i
< ARRAY_SIZE(vmx_preemption_cpu_tfms
); i
++)
1165 if (eax
== vmx_preemption_cpu_tfms
[i
])
1171 static inline bool cpu_has_vmx_preemption_timer(void)
1173 return vmcs_config
.pin_based_exec_ctrl
&
1174 PIN_BASED_VMX_PREEMPTION_TIMER
;
1177 static inline bool cpu_has_vmx_posted_intr(void)
1179 return IS_ENABLED(CONFIG_X86_LOCAL_APIC
) &&
1180 vmcs_config
.pin_based_exec_ctrl
& PIN_BASED_POSTED_INTR
;
1183 static inline bool cpu_has_vmx_apicv(void)
1185 return cpu_has_vmx_apic_register_virt() &&
1186 cpu_has_vmx_virtual_intr_delivery() &&
1187 cpu_has_vmx_posted_intr();
1190 static inline bool cpu_has_vmx_flexpriority(void)
1192 return cpu_has_vmx_tpr_shadow() &&
1193 cpu_has_vmx_virtualize_apic_accesses();
1196 static inline bool cpu_has_vmx_ept_execute_only(void)
1198 return vmx_capability
.ept
& VMX_EPT_EXECUTE_ONLY_BIT
;
1201 static inline bool cpu_has_vmx_ept_2m_page(void)
1203 return vmx_capability
.ept
& VMX_EPT_2MB_PAGE_BIT
;
1206 static inline bool cpu_has_vmx_ept_1g_page(void)
1208 return vmx_capability
.ept
& VMX_EPT_1GB_PAGE_BIT
;
1211 static inline bool cpu_has_vmx_ept_4levels(void)
1213 return vmx_capability
.ept
& VMX_EPT_PAGE_WALK_4_BIT
;
1216 static inline bool cpu_has_vmx_ept_ad_bits(void)
1218 return vmx_capability
.ept
& VMX_EPT_AD_BIT
;
1221 static inline bool cpu_has_vmx_invept_context(void)
1223 return vmx_capability
.ept
& VMX_EPT_EXTENT_CONTEXT_BIT
;
1226 static inline bool cpu_has_vmx_invept_global(void)
1228 return vmx_capability
.ept
& VMX_EPT_EXTENT_GLOBAL_BIT
;
1231 static inline bool cpu_has_vmx_invvpid_single(void)
1233 return vmx_capability
.vpid
& VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT
;
1236 static inline bool cpu_has_vmx_invvpid_global(void)
1238 return vmx_capability
.vpid
& VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT
;
1241 static inline bool cpu_has_vmx_invvpid(void)
1243 return vmx_capability
.vpid
& VMX_VPID_INVVPID_BIT
;
1246 static inline bool cpu_has_vmx_ept(void)
1248 return vmcs_config
.cpu_based_2nd_exec_ctrl
&
1249 SECONDARY_EXEC_ENABLE_EPT
;
1252 static inline bool cpu_has_vmx_unrestricted_guest(void)
1254 return vmcs_config
.cpu_based_2nd_exec_ctrl
&
1255 SECONDARY_EXEC_UNRESTRICTED_GUEST
;
1258 static inline bool cpu_has_vmx_ple(void)
1260 return vmcs_config
.cpu_based_2nd_exec_ctrl
&
1261 SECONDARY_EXEC_PAUSE_LOOP_EXITING
;
1264 static inline bool cpu_has_vmx_basic_inout(void)
1266 return (((u64
)vmcs_config
.basic_cap
<< 32) & VMX_BASIC_INOUT
);
1269 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu
*vcpu
)
1271 return flexpriority_enabled
&& lapic_in_kernel(vcpu
);
1274 static inline bool cpu_has_vmx_vpid(void)
1276 return vmcs_config
.cpu_based_2nd_exec_ctrl
&
1277 SECONDARY_EXEC_ENABLE_VPID
;
1280 static inline bool cpu_has_vmx_rdtscp(void)
1282 return vmcs_config
.cpu_based_2nd_exec_ctrl
&
1283 SECONDARY_EXEC_RDTSCP
;
1286 static inline bool cpu_has_vmx_invpcid(void)
1288 return vmcs_config
.cpu_based_2nd_exec_ctrl
&
1289 SECONDARY_EXEC_ENABLE_INVPCID
;
1292 static inline bool cpu_has_vmx_wbinvd_exit(void)
1294 return vmcs_config
.cpu_based_2nd_exec_ctrl
&
1295 SECONDARY_EXEC_WBINVD_EXITING
;
1298 static inline bool cpu_has_vmx_shadow_vmcs(void)
1301 rdmsrl(MSR_IA32_VMX_MISC
, vmx_msr
);
1302 /* check if the cpu supports writing r/o exit information fields */
1303 if (!(vmx_msr
& MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS
))
1306 return vmcs_config
.cpu_based_2nd_exec_ctrl
&
1307 SECONDARY_EXEC_SHADOW_VMCS
;
1310 static inline bool cpu_has_vmx_pml(void)
1312 return vmcs_config
.cpu_based_2nd_exec_ctrl
& SECONDARY_EXEC_ENABLE_PML
;
1315 static inline bool cpu_has_vmx_tsc_scaling(void)
1317 return vmcs_config
.cpu_based_2nd_exec_ctrl
&
1318 SECONDARY_EXEC_TSC_SCALING
;
1321 static inline bool report_flexpriority(void)
1323 return flexpriority_enabled
;
1326 static inline unsigned nested_cpu_vmx_misc_cr3_count(struct kvm_vcpu
*vcpu
)
1328 return vmx_misc_cr3_count(to_vmx(vcpu
)->nested
.nested_vmx_misc_low
);
1331 static inline bool nested_cpu_has(struct vmcs12
*vmcs12
, u32 bit
)
1333 return vmcs12
->cpu_based_vm_exec_control
& bit
;
1336 static inline bool nested_cpu_has2(struct vmcs12
*vmcs12
, u32 bit
)
1338 return (vmcs12
->cpu_based_vm_exec_control
&
1339 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS
) &&
1340 (vmcs12
->secondary_vm_exec_control
& bit
);
1343 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12
*vmcs12
)
1345 return vmcs12
->pin_based_vm_exec_control
& PIN_BASED_VIRTUAL_NMIS
;
1348 static inline bool nested_cpu_has_preemption_timer(struct vmcs12
*vmcs12
)
1350 return vmcs12
->pin_based_vm_exec_control
&
1351 PIN_BASED_VMX_PREEMPTION_TIMER
;
1354 static inline int nested_cpu_has_ept(struct vmcs12
*vmcs12
)
1356 return nested_cpu_has2(vmcs12
, SECONDARY_EXEC_ENABLE_EPT
);
1359 static inline bool nested_cpu_has_xsaves(struct vmcs12
*vmcs12
)
1361 return nested_cpu_has2(vmcs12
, SECONDARY_EXEC_XSAVES
) &&
1362 vmx_xsaves_supported();
1365 static inline bool nested_cpu_has_pml(struct vmcs12
*vmcs12
)
1367 return nested_cpu_has2(vmcs12
, SECONDARY_EXEC_ENABLE_PML
);
1370 static inline bool nested_cpu_has_virt_x2apic_mode(struct vmcs12
*vmcs12
)
1372 return nested_cpu_has2(vmcs12
, SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE
);
1375 static inline bool nested_cpu_has_vpid(struct vmcs12
*vmcs12
)
1377 return nested_cpu_has2(vmcs12
, SECONDARY_EXEC_ENABLE_VPID
);
1380 static inline bool nested_cpu_has_apic_reg_virt(struct vmcs12
*vmcs12
)
1382 return nested_cpu_has2(vmcs12
, SECONDARY_EXEC_APIC_REGISTER_VIRT
);
1385 static inline bool nested_cpu_has_vid(struct vmcs12
*vmcs12
)
1387 return nested_cpu_has2(vmcs12
, SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY
);
1390 static inline bool nested_cpu_has_posted_intr(struct vmcs12
*vmcs12
)
1392 return vmcs12
->pin_based_vm_exec_control
& PIN_BASED_POSTED_INTR
;
1395 static inline bool is_nmi(u32 intr_info
)
1397 return (intr_info
& (INTR_INFO_INTR_TYPE_MASK
| INTR_INFO_VALID_MASK
))
1398 == (INTR_TYPE_NMI_INTR
| INTR_INFO_VALID_MASK
);
1401 static void nested_vmx_vmexit(struct kvm_vcpu
*vcpu
, u32 exit_reason
,
1403 unsigned long exit_qualification
);
1404 static void nested_vmx_entry_failure(struct kvm_vcpu
*vcpu
,
1405 struct vmcs12
*vmcs12
,
1406 u32 reason
, unsigned long qualification
);
1408 static int __find_msr_index(struct vcpu_vmx
*vmx
, u32 msr
)
1412 for (i
= 0; i
< vmx
->nmsrs
; ++i
)
1413 if (vmx_msr_index
[vmx
->guest_msrs
[i
].index
] == msr
)
1418 static inline void __invvpid(int ext
, u16 vpid
, gva_t gva
)
1424 } operand
= { vpid
, 0, gva
};
1426 asm volatile (__ex(ASM_VMX_INVVPID
)
1427 /* CF==1 or ZF==1 --> rc = -1 */
1428 "; ja 1f ; ud2 ; 1:"
1429 : : "a"(&operand
), "c"(ext
) : "cc", "memory");
1432 static inline void __invept(int ext
, u64 eptp
, gpa_t gpa
)
1436 } operand
= {eptp
, gpa
};
1438 asm volatile (__ex(ASM_VMX_INVEPT
)
1439 /* CF==1 or ZF==1 --> rc = -1 */
1440 "; ja 1f ; ud2 ; 1:\n"
1441 : : "a" (&operand
), "c" (ext
) : "cc", "memory");
1444 static struct shared_msr_entry
*find_msr_entry(struct vcpu_vmx
*vmx
, u32 msr
)
1448 i
= __find_msr_index(vmx
, msr
);
1450 return &vmx
->guest_msrs
[i
];
1454 static void vmcs_clear(struct vmcs
*vmcs
)
1456 u64 phys_addr
= __pa(vmcs
);
1459 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX
) "; setna %0"
1460 : "=qm"(error
) : "a"(&phys_addr
), "m"(phys_addr
)
1463 printk(KERN_ERR
"kvm: vmclear fail: %p/%llx\n",
1467 static inline void loaded_vmcs_init(struct loaded_vmcs
*loaded_vmcs
)
1469 vmcs_clear(loaded_vmcs
->vmcs
);
1470 if (loaded_vmcs
->shadow_vmcs
&& loaded_vmcs
->launched
)
1471 vmcs_clear(loaded_vmcs
->shadow_vmcs
);
1472 loaded_vmcs
->cpu
= -1;
1473 loaded_vmcs
->launched
= 0;
1476 static void vmcs_load(struct vmcs
*vmcs
)
1478 u64 phys_addr
= __pa(vmcs
);
1481 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX
) "; setna %0"
1482 : "=qm"(error
) : "a"(&phys_addr
), "m"(phys_addr
)
1485 printk(KERN_ERR
"kvm: vmptrld %p/%llx failed\n",
1489 #ifdef CONFIG_KEXEC_CORE
1491 * This bitmap is used to indicate whether the vmclear
1492 * operation is enabled on all cpus. All disabled by
1495 static cpumask_t crash_vmclear_enabled_bitmap
= CPU_MASK_NONE
;
1497 static inline void crash_enable_local_vmclear(int cpu
)
1499 cpumask_set_cpu(cpu
, &crash_vmclear_enabled_bitmap
);
1502 static inline void crash_disable_local_vmclear(int cpu
)
1504 cpumask_clear_cpu(cpu
, &crash_vmclear_enabled_bitmap
);
1507 static inline int crash_local_vmclear_enabled(int cpu
)
1509 return cpumask_test_cpu(cpu
, &crash_vmclear_enabled_bitmap
);
1512 static void crash_vmclear_local_loaded_vmcss(void)
1514 int cpu
= raw_smp_processor_id();
1515 struct loaded_vmcs
*v
;
1517 if (!crash_local_vmclear_enabled(cpu
))
1520 list_for_each_entry(v
, &per_cpu(loaded_vmcss_on_cpu
, cpu
),
1521 loaded_vmcss_on_cpu_link
)
1522 vmcs_clear(v
->vmcs
);
1525 static inline void crash_enable_local_vmclear(int cpu
) { }
1526 static inline void crash_disable_local_vmclear(int cpu
) { }
1527 #endif /* CONFIG_KEXEC_CORE */
1529 static void __loaded_vmcs_clear(void *arg
)
1531 struct loaded_vmcs
*loaded_vmcs
= arg
;
1532 int cpu
= raw_smp_processor_id();
1534 if (loaded_vmcs
->cpu
!= cpu
)
1535 return; /* vcpu migration can race with cpu offline */
1536 if (per_cpu(current_vmcs
, cpu
) == loaded_vmcs
->vmcs
)
1537 per_cpu(current_vmcs
, cpu
) = NULL
;
1538 crash_disable_local_vmclear(cpu
);
1539 list_del(&loaded_vmcs
->loaded_vmcss_on_cpu_link
);
1542 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1543 * is before setting loaded_vmcs->vcpu to -1 which is done in
1544 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1545 * then adds the vmcs into percpu list before it is deleted.
1549 loaded_vmcs_init(loaded_vmcs
);
1550 crash_enable_local_vmclear(cpu
);
1553 static void loaded_vmcs_clear(struct loaded_vmcs
*loaded_vmcs
)
1555 int cpu
= loaded_vmcs
->cpu
;
1558 smp_call_function_single(cpu
,
1559 __loaded_vmcs_clear
, loaded_vmcs
, 1);
1562 static inline void vpid_sync_vcpu_single(int vpid
)
1567 if (cpu_has_vmx_invvpid_single())
1568 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT
, vpid
, 0);
1571 static inline void vpid_sync_vcpu_global(void)
1573 if (cpu_has_vmx_invvpid_global())
1574 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT
, 0, 0);
1577 static inline void vpid_sync_context(int vpid
)
1579 if (cpu_has_vmx_invvpid_single())
1580 vpid_sync_vcpu_single(vpid
);
1582 vpid_sync_vcpu_global();
1585 static inline void ept_sync_global(void)
1587 if (cpu_has_vmx_invept_global())
1588 __invept(VMX_EPT_EXTENT_GLOBAL
, 0, 0);
1591 static inline void ept_sync_context(u64 eptp
)
1594 if (cpu_has_vmx_invept_context())
1595 __invept(VMX_EPT_EXTENT_CONTEXT
, eptp
, 0);
1601 static __always_inline
void vmcs_check16(unsigned long field
)
1603 BUILD_BUG_ON_MSG(__builtin_constant_p(field
) && ((field
) & 0x6001) == 0x2000,
1604 "16-bit accessor invalid for 64-bit field");
1605 BUILD_BUG_ON_MSG(__builtin_constant_p(field
) && ((field
) & 0x6001) == 0x2001,
1606 "16-bit accessor invalid for 64-bit high field");
1607 BUILD_BUG_ON_MSG(__builtin_constant_p(field
) && ((field
) & 0x6000) == 0x4000,
1608 "16-bit accessor invalid for 32-bit high field");
1609 BUILD_BUG_ON_MSG(__builtin_constant_p(field
) && ((field
) & 0x6000) == 0x6000,
1610 "16-bit accessor invalid for natural width field");
1613 static __always_inline
void vmcs_check32(unsigned long field
)
1615 BUILD_BUG_ON_MSG(__builtin_constant_p(field
) && ((field
) & 0x6000) == 0,
1616 "32-bit accessor invalid for 16-bit field");
1617 BUILD_BUG_ON_MSG(__builtin_constant_p(field
) && ((field
) & 0x6000) == 0x6000,
1618 "32-bit accessor invalid for natural width field");
1621 static __always_inline
void vmcs_check64(unsigned long field
)
1623 BUILD_BUG_ON_MSG(__builtin_constant_p(field
) && ((field
) & 0x6000) == 0,
1624 "64-bit accessor invalid for 16-bit field");
1625 BUILD_BUG_ON_MSG(__builtin_constant_p(field
) && ((field
) & 0x6001) == 0x2001,
1626 "64-bit accessor invalid for 64-bit high field");
1627 BUILD_BUG_ON_MSG(__builtin_constant_p(field
) && ((field
) & 0x6000) == 0x4000,
1628 "64-bit accessor invalid for 32-bit field");
1629 BUILD_BUG_ON_MSG(__builtin_constant_p(field
) && ((field
) & 0x6000) == 0x6000,
1630 "64-bit accessor invalid for natural width field");
1633 static __always_inline
void vmcs_checkl(unsigned long field
)
1635 BUILD_BUG_ON_MSG(__builtin_constant_p(field
) && ((field
) & 0x6000) == 0,
1636 "Natural width accessor invalid for 16-bit field");
1637 BUILD_BUG_ON_MSG(__builtin_constant_p(field
) && ((field
) & 0x6001) == 0x2000,
1638 "Natural width accessor invalid for 64-bit field");
1639 BUILD_BUG_ON_MSG(__builtin_constant_p(field
) && ((field
) & 0x6001) == 0x2001,
1640 "Natural width accessor invalid for 64-bit high field");
1641 BUILD_BUG_ON_MSG(__builtin_constant_p(field
) && ((field
) & 0x6000) == 0x4000,
1642 "Natural width accessor invalid for 32-bit field");
1645 static __always_inline
unsigned long __vmcs_readl(unsigned long field
)
1647 unsigned long value
;
1649 asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX
, "%0")
1650 : "=a"(value
) : "d"(field
) : "cc");
1654 static __always_inline u16
vmcs_read16(unsigned long field
)
1656 vmcs_check16(field
);
1657 return __vmcs_readl(field
);
1660 static __always_inline u32
vmcs_read32(unsigned long field
)
1662 vmcs_check32(field
);
1663 return __vmcs_readl(field
);
1666 static __always_inline u64
vmcs_read64(unsigned long field
)
1668 vmcs_check64(field
);
1669 #ifdef CONFIG_X86_64
1670 return __vmcs_readl(field
);
1672 return __vmcs_readl(field
) | ((u64
)__vmcs_readl(field
+1) << 32);
1676 static __always_inline
unsigned long vmcs_readl(unsigned long field
)
1679 return __vmcs_readl(field
);
1682 static noinline
void vmwrite_error(unsigned long field
, unsigned long value
)
1684 printk(KERN_ERR
"vmwrite error: reg %lx value %lx (err %d)\n",
1685 field
, value
, vmcs_read32(VM_INSTRUCTION_ERROR
));
1689 static __always_inline
void __vmcs_writel(unsigned long field
, unsigned long value
)
1693 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX
) "; setna %0"
1694 : "=q"(error
) : "a"(value
), "d"(field
) : "cc");
1695 if (unlikely(error
))
1696 vmwrite_error(field
, value
);
1699 static __always_inline
void vmcs_write16(unsigned long field
, u16 value
)
1701 vmcs_check16(field
);
1702 __vmcs_writel(field
, value
);
1705 static __always_inline
void vmcs_write32(unsigned long field
, u32 value
)
1707 vmcs_check32(field
);
1708 __vmcs_writel(field
, value
);
1711 static __always_inline
void vmcs_write64(unsigned long field
, u64 value
)
1713 vmcs_check64(field
);
1714 __vmcs_writel(field
, value
);
1715 #ifndef CONFIG_X86_64
1717 __vmcs_writel(field
+1, value
>> 32);
1721 static __always_inline
void vmcs_writel(unsigned long field
, unsigned long value
)
1724 __vmcs_writel(field
, value
);
1727 static __always_inline
void vmcs_clear_bits(unsigned long field
, u32 mask
)
1729 BUILD_BUG_ON_MSG(__builtin_constant_p(field
) && ((field
) & 0x6000) == 0x2000,
1730 "vmcs_clear_bits does not support 64-bit fields");
1731 __vmcs_writel(field
, __vmcs_readl(field
) & ~mask
);
1734 static __always_inline
void vmcs_set_bits(unsigned long field
, u32 mask
)
1736 BUILD_BUG_ON_MSG(__builtin_constant_p(field
) && ((field
) & 0x6000) == 0x2000,
1737 "vmcs_set_bits does not support 64-bit fields");
1738 __vmcs_writel(field
, __vmcs_readl(field
) | mask
);
1741 static inline void vm_entry_controls_reset_shadow(struct vcpu_vmx
*vmx
)
1743 vmx
->vm_entry_controls_shadow
= vmcs_read32(VM_ENTRY_CONTROLS
);
1746 static inline void vm_entry_controls_init(struct vcpu_vmx
*vmx
, u32 val
)
1748 vmcs_write32(VM_ENTRY_CONTROLS
, val
);
1749 vmx
->vm_entry_controls_shadow
= val
;
1752 static inline void vm_entry_controls_set(struct vcpu_vmx
*vmx
, u32 val
)
1754 if (vmx
->vm_entry_controls_shadow
!= val
)
1755 vm_entry_controls_init(vmx
, val
);
1758 static inline u32
vm_entry_controls_get(struct vcpu_vmx
*vmx
)
1760 return vmx
->vm_entry_controls_shadow
;
1764 static inline void vm_entry_controls_setbit(struct vcpu_vmx
*vmx
, u32 val
)
1766 vm_entry_controls_set(vmx
, vm_entry_controls_get(vmx
) | val
);
1769 static inline void vm_entry_controls_clearbit(struct vcpu_vmx
*vmx
, u32 val
)
1771 vm_entry_controls_set(vmx
, vm_entry_controls_get(vmx
) & ~val
);
1774 static inline void vm_exit_controls_reset_shadow(struct vcpu_vmx
*vmx
)
1776 vmx
->vm_exit_controls_shadow
= vmcs_read32(VM_EXIT_CONTROLS
);
1779 static inline void vm_exit_controls_init(struct vcpu_vmx
*vmx
, u32 val
)
1781 vmcs_write32(VM_EXIT_CONTROLS
, val
);
1782 vmx
->vm_exit_controls_shadow
= val
;
1785 static inline void vm_exit_controls_set(struct vcpu_vmx
*vmx
, u32 val
)
1787 if (vmx
->vm_exit_controls_shadow
!= val
)
1788 vm_exit_controls_init(vmx
, val
);
1791 static inline u32
vm_exit_controls_get(struct vcpu_vmx
*vmx
)
1793 return vmx
->vm_exit_controls_shadow
;
1797 static inline void vm_exit_controls_setbit(struct vcpu_vmx
*vmx
, u32 val
)
1799 vm_exit_controls_set(vmx
, vm_exit_controls_get(vmx
) | val
);
1802 static inline void vm_exit_controls_clearbit(struct vcpu_vmx
*vmx
, u32 val
)
1804 vm_exit_controls_set(vmx
, vm_exit_controls_get(vmx
) & ~val
);
1807 static void vmx_segment_cache_clear(struct vcpu_vmx
*vmx
)
1809 vmx
->segment_cache
.bitmask
= 0;
1812 static bool vmx_segment_cache_test_set(struct vcpu_vmx
*vmx
, unsigned seg
,
1816 u32 mask
= 1 << (seg
* SEG_FIELD_NR
+ field
);
1818 if (!(vmx
->vcpu
.arch
.regs_avail
& (1 << VCPU_EXREG_SEGMENTS
))) {
1819 vmx
->vcpu
.arch
.regs_avail
|= (1 << VCPU_EXREG_SEGMENTS
);
1820 vmx
->segment_cache
.bitmask
= 0;
1822 ret
= vmx
->segment_cache
.bitmask
& mask
;
1823 vmx
->segment_cache
.bitmask
|= mask
;
1827 static u16
vmx_read_guest_seg_selector(struct vcpu_vmx
*vmx
, unsigned seg
)
1829 u16
*p
= &vmx
->segment_cache
.seg
[seg
].selector
;
1831 if (!vmx_segment_cache_test_set(vmx
, seg
, SEG_FIELD_SEL
))
1832 *p
= vmcs_read16(kvm_vmx_segment_fields
[seg
].selector
);
1836 static ulong
vmx_read_guest_seg_base(struct vcpu_vmx
*vmx
, unsigned seg
)
1838 ulong
*p
= &vmx
->segment_cache
.seg
[seg
].base
;
1840 if (!vmx_segment_cache_test_set(vmx
, seg
, SEG_FIELD_BASE
))
1841 *p
= vmcs_readl(kvm_vmx_segment_fields
[seg
].base
);
1845 static u32
vmx_read_guest_seg_limit(struct vcpu_vmx
*vmx
, unsigned seg
)
1847 u32
*p
= &vmx
->segment_cache
.seg
[seg
].limit
;
1849 if (!vmx_segment_cache_test_set(vmx
, seg
, SEG_FIELD_LIMIT
))
1850 *p
= vmcs_read32(kvm_vmx_segment_fields
[seg
].limit
);
1854 static u32
vmx_read_guest_seg_ar(struct vcpu_vmx
*vmx
, unsigned seg
)
1856 u32
*p
= &vmx
->segment_cache
.seg
[seg
].ar
;
1858 if (!vmx_segment_cache_test_set(vmx
, seg
, SEG_FIELD_AR
))
1859 *p
= vmcs_read32(kvm_vmx_segment_fields
[seg
].ar_bytes
);
1863 static void update_exception_bitmap(struct kvm_vcpu
*vcpu
)
1867 eb
= (1u << PF_VECTOR
) | (1u << UD_VECTOR
) | (1u << MC_VECTOR
) |
1868 (1u << DB_VECTOR
) | (1u << AC_VECTOR
);
1869 if ((vcpu
->guest_debug
&
1870 (KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_USE_SW_BP
)) ==
1871 (KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_USE_SW_BP
))
1872 eb
|= 1u << BP_VECTOR
;
1873 if (to_vmx(vcpu
)->rmode
.vm86_active
)
1876 eb
&= ~(1u << PF_VECTOR
); /* bypass_guest_pf = 0 */
1878 /* When we are running a nested L2 guest and L1 specified for it a
1879 * certain exception bitmap, we must trap the same exceptions and pass
1880 * them to L1. When running L2, we will only handle the exceptions
1881 * specified above if L1 did not want them.
1883 if (is_guest_mode(vcpu
))
1884 eb
|= get_vmcs12(vcpu
)->exception_bitmap
;
1886 vmcs_write32(EXCEPTION_BITMAP
, eb
);
1889 static void clear_atomic_switch_msr_special(struct vcpu_vmx
*vmx
,
1890 unsigned long entry
, unsigned long exit
)
1892 vm_entry_controls_clearbit(vmx
, entry
);
1893 vm_exit_controls_clearbit(vmx
, exit
);
1896 static void clear_atomic_switch_msr(struct vcpu_vmx
*vmx
, unsigned msr
)
1899 struct msr_autoload
*m
= &vmx
->msr_autoload
;
1903 if (cpu_has_load_ia32_efer
) {
1904 clear_atomic_switch_msr_special(vmx
,
1905 VM_ENTRY_LOAD_IA32_EFER
,
1906 VM_EXIT_LOAD_IA32_EFER
);
1910 case MSR_CORE_PERF_GLOBAL_CTRL
:
1911 if (cpu_has_load_perf_global_ctrl
) {
1912 clear_atomic_switch_msr_special(vmx
,
1913 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL
,
1914 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL
);
1920 for (i
= 0; i
< m
->nr
; ++i
)
1921 if (m
->guest
[i
].index
== msr
)
1927 m
->guest
[i
] = m
->guest
[m
->nr
];
1928 m
->host
[i
] = m
->host
[m
->nr
];
1929 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT
, m
->nr
);
1930 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT
, m
->nr
);
1933 static void add_atomic_switch_msr_special(struct vcpu_vmx
*vmx
,
1934 unsigned long entry
, unsigned long exit
,
1935 unsigned long guest_val_vmcs
, unsigned long host_val_vmcs
,
1936 u64 guest_val
, u64 host_val
)
1938 vmcs_write64(guest_val_vmcs
, guest_val
);
1939 vmcs_write64(host_val_vmcs
, host_val
);
1940 vm_entry_controls_setbit(vmx
, entry
);
1941 vm_exit_controls_setbit(vmx
, exit
);
1944 static void add_atomic_switch_msr(struct vcpu_vmx
*vmx
, unsigned msr
,
1945 u64 guest_val
, u64 host_val
)
1948 struct msr_autoload
*m
= &vmx
->msr_autoload
;
1952 if (cpu_has_load_ia32_efer
) {
1953 add_atomic_switch_msr_special(vmx
,
1954 VM_ENTRY_LOAD_IA32_EFER
,
1955 VM_EXIT_LOAD_IA32_EFER
,
1958 guest_val
, host_val
);
1962 case MSR_CORE_PERF_GLOBAL_CTRL
:
1963 if (cpu_has_load_perf_global_ctrl
) {
1964 add_atomic_switch_msr_special(vmx
,
1965 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL
,
1966 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL
,
1967 GUEST_IA32_PERF_GLOBAL_CTRL
,
1968 HOST_IA32_PERF_GLOBAL_CTRL
,
1969 guest_val
, host_val
);
1973 case MSR_IA32_PEBS_ENABLE
:
1974 /* PEBS needs a quiescent period after being disabled (to write
1975 * a record). Disabling PEBS through VMX MSR swapping doesn't
1976 * provide that period, so a CPU could write host's record into
1979 wrmsrl(MSR_IA32_PEBS_ENABLE
, 0);
1982 for (i
= 0; i
< m
->nr
; ++i
)
1983 if (m
->guest
[i
].index
== msr
)
1986 if (i
== NR_AUTOLOAD_MSRS
) {
1987 printk_once(KERN_WARNING
"Not enough msr switch entries. "
1988 "Can't add msr %x\n", msr
);
1990 } else if (i
== m
->nr
) {
1992 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT
, m
->nr
);
1993 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT
, m
->nr
);
1996 m
->guest
[i
].index
= msr
;
1997 m
->guest
[i
].value
= guest_val
;
1998 m
->host
[i
].index
= msr
;
1999 m
->host
[i
].value
= host_val
;
2002 static bool update_transition_efer(struct vcpu_vmx
*vmx
, int efer_offset
)
2004 u64 guest_efer
= vmx
->vcpu
.arch
.efer
;
2005 u64 ignore_bits
= 0;
2009 * NX is needed to handle CR0.WP=1, CR4.SMEP=1. Testing
2010 * host CPUID is more efficient than testing guest CPUID
2011 * or CR4. Host SMEP is anyway a requirement for guest SMEP.
2013 if (boot_cpu_has(X86_FEATURE_SMEP
))
2014 guest_efer
|= EFER_NX
;
2015 else if (!(guest_efer
& EFER_NX
))
2016 ignore_bits
|= EFER_NX
;
2020 * LMA and LME handled by hardware; SCE meaningless outside long mode.
2022 ignore_bits
|= EFER_SCE
;
2023 #ifdef CONFIG_X86_64
2024 ignore_bits
|= EFER_LMA
| EFER_LME
;
2025 /* SCE is meaningful only in long mode on Intel */
2026 if (guest_efer
& EFER_LMA
)
2027 ignore_bits
&= ~(u64
)EFER_SCE
;
2030 clear_atomic_switch_msr(vmx
, MSR_EFER
);
2033 * On EPT, we can't emulate NX, so we must switch EFER atomically.
2034 * On CPUs that support "load IA32_EFER", always switch EFER
2035 * atomically, since it's faster than switching it manually.
2037 if (cpu_has_load_ia32_efer
||
2038 (enable_ept
&& ((vmx
->vcpu
.arch
.efer
^ host_efer
) & EFER_NX
))) {
2039 if (!(guest_efer
& EFER_LMA
))
2040 guest_efer
&= ~EFER_LME
;
2041 if (guest_efer
!= host_efer
)
2042 add_atomic_switch_msr(vmx
, MSR_EFER
,
2043 guest_efer
, host_efer
);
2046 guest_efer
&= ~ignore_bits
;
2047 guest_efer
|= host_efer
& ignore_bits
;
2049 vmx
->guest_msrs
[efer_offset
].data
= guest_efer
;
2050 vmx
->guest_msrs
[efer_offset
].mask
= ~ignore_bits
;
2056 #ifdef CONFIG_X86_32
2058 * On 32-bit kernels, VM exits still load the FS and GS bases from the
2059 * VMCS rather than the segment table. KVM uses this helper to figure
2060 * out the current bases to poke them into the VMCS before entry.
2062 static unsigned long segment_base(u16 selector
)
2064 struct desc_struct
*table
;
2067 if (!(selector
& ~SEGMENT_RPL_MASK
))
2070 table
= get_current_gdt_ro();
2072 if ((selector
& SEGMENT_TI_MASK
) == SEGMENT_LDT
) {
2073 u16 ldt_selector
= kvm_read_ldt();
2075 if (!(ldt_selector
& ~SEGMENT_RPL_MASK
))
2078 table
= (struct desc_struct
*)segment_base(ldt_selector
);
2080 v
= get_desc_base(&table
[selector
>> 3]);
2085 static void vmx_save_host_state(struct kvm_vcpu
*vcpu
)
2087 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
2090 if (vmx
->host_state
.loaded
)
2093 vmx
->host_state
.loaded
= 1;
2095 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
2096 * allow segment selectors with cpl > 0 or ti == 1.
2098 vmx
->host_state
.ldt_sel
= kvm_read_ldt();
2099 vmx
->host_state
.gs_ldt_reload_needed
= vmx
->host_state
.ldt_sel
;
2100 savesegment(fs
, vmx
->host_state
.fs_sel
);
2101 if (!(vmx
->host_state
.fs_sel
& 7)) {
2102 vmcs_write16(HOST_FS_SELECTOR
, vmx
->host_state
.fs_sel
);
2103 vmx
->host_state
.fs_reload_needed
= 0;
2105 vmcs_write16(HOST_FS_SELECTOR
, 0);
2106 vmx
->host_state
.fs_reload_needed
= 1;
2108 savesegment(gs
, vmx
->host_state
.gs_sel
);
2109 if (!(vmx
->host_state
.gs_sel
& 7))
2110 vmcs_write16(HOST_GS_SELECTOR
, vmx
->host_state
.gs_sel
);
2112 vmcs_write16(HOST_GS_SELECTOR
, 0);
2113 vmx
->host_state
.gs_ldt_reload_needed
= 1;
2116 #ifdef CONFIG_X86_64
2117 savesegment(ds
, vmx
->host_state
.ds_sel
);
2118 savesegment(es
, vmx
->host_state
.es_sel
);
2121 #ifdef CONFIG_X86_64
2122 vmcs_writel(HOST_FS_BASE
, read_msr(MSR_FS_BASE
));
2123 vmcs_writel(HOST_GS_BASE
, read_msr(MSR_GS_BASE
));
2125 vmcs_writel(HOST_FS_BASE
, segment_base(vmx
->host_state
.fs_sel
));
2126 vmcs_writel(HOST_GS_BASE
, segment_base(vmx
->host_state
.gs_sel
));
2129 #ifdef CONFIG_X86_64
2130 rdmsrl(MSR_KERNEL_GS_BASE
, vmx
->msr_host_kernel_gs_base
);
2131 if (is_long_mode(&vmx
->vcpu
))
2132 wrmsrl(MSR_KERNEL_GS_BASE
, vmx
->msr_guest_kernel_gs_base
);
2134 if (boot_cpu_has(X86_FEATURE_MPX
))
2135 rdmsrl(MSR_IA32_BNDCFGS
, vmx
->host_state
.msr_host_bndcfgs
);
2136 for (i
= 0; i
< vmx
->save_nmsrs
; ++i
)
2137 kvm_set_shared_msr(vmx
->guest_msrs
[i
].index
,
2138 vmx
->guest_msrs
[i
].data
,
2139 vmx
->guest_msrs
[i
].mask
);
2142 static void __vmx_load_host_state(struct vcpu_vmx
*vmx
)
2144 if (!vmx
->host_state
.loaded
)
2147 ++vmx
->vcpu
.stat
.host_state_reload
;
2148 vmx
->host_state
.loaded
= 0;
2149 #ifdef CONFIG_X86_64
2150 if (is_long_mode(&vmx
->vcpu
))
2151 rdmsrl(MSR_KERNEL_GS_BASE
, vmx
->msr_guest_kernel_gs_base
);
2153 if (vmx
->host_state
.gs_ldt_reload_needed
) {
2154 kvm_load_ldt(vmx
->host_state
.ldt_sel
);
2155 #ifdef CONFIG_X86_64
2156 load_gs_index(vmx
->host_state
.gs_sel
);
2158 loadsegment(gs
, vmx
->host_state
.gs_sel
);
2161 if (vmx
->host_state
.fs_reload_needed
)
2162 loadsegment(fs
, vmx
->host_state
.fs_sel
);
2163 #ifdef CONFIG_X86_64
2164 if (unlikely(vmx
->host_state
.ds_sel
| vmx
->host_state
.es_sel
)) {
2165 loadsegment(ds
, vmx
->host_state
.ds_sel
);
2166 loadsegment(es
, vmx
->host_state
.es_sel
);
2169 invalidate_tss_limit();
2170 #ifdef CONFIG_X86_64
2171 wrmsrl(MSR_KERNEL_GS_BASE
, vmx
->msr_host_kernel_gs_base
);
2173 if (vmx
->host_state
.msr_host_bndcfgs
)
2174 wrmsrl(MSR_IA32_BNDCFGS
, vmx
->host_state
.msr_host_bndcfgs
);
2175 load_fixmap_gdt(raw_smp_processor_id());
2178 static void vmx_load_host_state(struct vcpu_vmx
*vmx
)
2181 __vmx_load_host_state(vmx
);
2185 static void vmx_vcpu_pi_load(struct kvm_vcpu
*vcpu
, int cpu
)
2187 struct pi_desc
*pi_desc
= vcpu_to_pi_desc(vcpu
);
2188 struct pi_desc old
, new;
2191 if (!kvm_arch_has_assigned_device(vcpu
->kvm
) ||
2192 !irq_remapping_cap(IRQ_POSTING_CAP
) ||
2193 !kvm_vcpu_apicv_active(vcpu
))
2197 old
.control
= new.control
= pi_desc
->control
;
2200 * If 'nv' field is POSTED_INTR_WAKEUP_VECTOR, there
2201 * are two possible cases:
2202 * 1. After running 'pre_block', context switch
2203 * happened. For this case, 'sn' was set in
2204 * vmx_vcpu_put(), so we need to clear it here.
2205 * 2. After running 'pre_block', we were blocked,
2206 * and woken up by some other guy. For this case,
2207 * we don't need to do anything, 'pi_post_block'
2208 * will do everything for us. However, we cannot
2209 * check whether it is case #1 or case #2 here
2210 * (maybe, not needed), so we also clear sn here,
2211 * I think it is not a big deal.
2213 if (pi_desc
->nv
!= POSTED_INTR_WAKEUP_VECTOR
) {
2214 if (vcpu
->cpu
!= cpu
) {
2215 dest
= cpu_physical_id(cpu
);
2217 if (x2apic_enabled())
2220 new.ndst
= (dest
<< 8) & 0xFF00;
2223 /* set 'NV' to 'notification vector' */
2224 new.nv
= POSTED_INTR_VECTOR
;
2227 /* Allow posting non-urgent interrupts */
2229 } while (cmpxchg(&pi_desc
->control
, old
.control
,
2230 new.control
) != old
.control
);
2233 static void decache_tsc_multiplier(struct vcpu_vmx
*vmx
)
2235 vmx
->current_tsc_ratio
= vmx
->vcpu
.arch
.tsc_scaling_ratio
;
2236 vmcs_write64(TSC_MULTIPLIER
, vmx
->current_tsc_ratio
);
2240 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
2241 * vcpu mutex is already taken.
2243 static void vmx_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
2245 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
2246 bool already_loaded
= vmx
->loaded_vmcs
->cpu
== cpu
;
2248 if (!already_loaded
) {
2249 loaded_vmcs_clear(vmx
->loaded_vmcs
);
2250 local_irq_disable();
2251 crash_disable_local_vmclear(cpu
);
2254 * Read loaded_vmcs->cpu should be before fetching
2255 * loaded_vmcs->loaded_vmcss_on_cpu_link.
2256 * See the comments in __loaded_vmcs_clear().
2260 list_add(&vmx
->loaded_vmcs
->loaded_vmcss_on_cpu_link
,
2261 &per_cpu(loaded_vmcss_on_cpu
, cpu
));
2262 crash_enable_local_vmclear(cpu
);
2266 if (per_cpu(current_vmcs
, cpu
) != vmx
->loaded_vmcs
->vmcs
) {
2267 per_cpu(current_vmcs
, cpu
) = vmx
->loaded_vmcs
->vmcs
;
2268 vmcs_load(vmx
->loaded_vmcs
->vmcs
);
2271 if (!already_loaded
) {
2272 void *gdt
= get_current_gdt_ro();
2273 unsigned long sysenter_esp
;
2275 kvm_make_request(KVM_REQ_TLB_FLUSH
, vcpu
);
2278 * Linux uses per-cpu TSS and GDT, so set these when switching
2279 * processors. See 22.2.4.
2281 vmcs_writel(HOST_TR_BASE
,
2282 (unsigned long)this_cpu_ptr(&cpu_tss
));
2283 vmcs_writel(HOST_GDTR_BASE
, (unsigned long)gdt
); /* 22.2.4 */
2286 * VM exits change the host TR limit to 0x67 after a VM
2287 * exit. This is okay, since 0x67 covers everything except
2288 * the IO bitmap and have have code to handle the IO bitmap
2289 * being lost after a VM exit.
2291 BUILD_BUG_ON(IO_BITMAP_OFFSET
- 1 != 0x67);
2293 rdmsrl(MSR_IA32_SYSENTER_ESP
, sysenter_esp
);
2294 vmcs_writel(HOST_IA32_SYSENTER_ESP
, sysenter_esp
); /* 22.2.3 */
2296 vmx
->loaded_vmcs
->cpu
= cpu
;
2299 /* Setup TSC multiplier */
2300 if (kvm_has_tsc_control
&&
2301 vmx
->current_tsc_ratio
!= vcpu
->arch
.tsc_scaling_ratio
)
2302 decache_tsc_multiplier(vmx
);
2304 vmx_vcpu_pi_load(vcpu
, cpu
);
2305 vmx
->host_pkru
= read_pkru();
2308 static void vmx_vcpu_pi_put(struct kvm_vcpu
*vcpu
)
2310 struct pi_desc
*pi_desc
= vcpu_to_pi_desc(vcpu
);
2312 if (!kvm_arch_has_assigned_device(vcpu
->kvm
) ||
2313 !irq_remapping_cap(IRQ_POSTING_CAP
) ||
2314 !kvm_vcpu_apicv_active(vcpu
))
2317 /* Set SN when the vCPU is preempted */
2318 if (vcpu
->preempted
)
2322 static void vmx_vcpu_put(struct kvm_vcpu
*vcpu
)
2324 vmx_vcpu_pi_put(vcpu
);
2326 __vmx_load_host_state(to_vmx(vcpu
));
2329 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu
*vcpu
);
2332 * Return the cr0 value that a nested guest would read. This is a combination
2333 * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
2334 * its hypervisor (cr0_read_shadow).
2336 static inline unsigned long nested_read_cr0(struct vmcs12
*fields
)
2338 return (fields
->guest_cr0
& ~fields
->cr0_guest_host_mask
) |
2339 (fields
->cr0_read_shadow
& fields
->cr0_guest_host_mask
);
2341 static inline unsigned long nested_read_cr4(struct vmcs12
*fields
)
2343 return (fields
->guest_cr4
& ~fields
->cr4_guest_host_mask
) |
2344 (fields
->cr4_read_shadow
& fields
->cr4_guest_host_mask
);
2347 static unsigned long vmx_get_rflags(struct kvm_vcpu
*vcpu
)
2349 unsigned long rflags
, save_rflags
;
2351 if (!test_bit(VCPU_EXREG_RFLAGS
, (ulong
*)&vcpu
->arch
.regs_avail
)) {
2352 __set_bit(VCPU_EXREG_RFLAGS
, (ulong
*)&vcpu
->arch
.regs_avail
);
2353 rflags
= vmcs_readl(GUEST_RFLAGS
);
2354 if (to_vmx(vcpu
)->rmode
.vm86_active
) {
2355 rflags
&= RMODE_GUEST_OWNED_EFLAGS_BITS
;
2356 save_rflags
= to_vmx(vcpu
)->rmode
.save_rflags
;
2357 rflags
|= save_rflags
& ~RMODE_GUEST_OWNED_EFLAGS_BITS
;
2359 to_vmx(vcpu
)->rflags
= rflags
;
2361 return to_vmx(vcpu
)->rflags
;
2364 static void vmx_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
2366 __set_bit(VCPU_EXREG_RFLAGS
, (ulong
*)&vcpu
->arch
.regs_avail
);
2367 to_vmx(vcpu
)->rflags
= rflags
;
2368 if (to_vmx(vcpu
)->rmode
.vm86_active
) {
2369 to_vmx(vcpu
)->rmode
.save_rflags
= rflags
;
2370 rflags
|= X86_EFLAGS_IOPL
| X86_EFLAGS_VM
;
2372 vmcs_writel(GUEST_RFLAGS
, rflags
);
2375 static u32
vmx_get_pkru(struct kvm_vcpu
*vcpu
)
2377 return to_vmx(vcpu
)->guest_pkru
;
2380 static u32
vmx_get_interrupt_shadow(struct kvm_vcpu
*vcpu
)
2382 u32 interruptibility
= vmcs_read32(GUEST_INTERRUPTIBILITY_INFO
);
2385 if (interruptibility
& GUEST_INTR_STATE_STI
)
2386 ret
|= KVM_X86_SHADOW_INT_STI
;
2387 if (interruptibility
& GUEST_INTR_STATE_MOV_SS
)
2388 ret
|= KVM_X86_SHADOW_INT_MOV_SS
;
2393 static void vmx_set_interrupt_shadow(struct kvm_vcpu
*vcpu
, int mask
)
2395 u32 interruptibility_old
= vmcs_read32(GUEST_INTERRUPTIBILITY_INFO
);
2396 u32 interruptibility
= interruptibility_old
;
2398 interruptibility
&= ~(GUEST_INTR_STATE_STI
| GUEST_INTR_STATE_MOV_SS
);
2400 if (mask
& KVM_X86_SHADOW_INT_MOV_SS
)
2401 interruptibility
|= GUEST_INTR_STATE_MOV_SS
;
2402 else if (mask
& KVM_X86_SHADOW_INT_STI
)
2403 interruptibility
|= GUEST_INTR_STATE_STI
;
2405 if ((interruptibility
!= interruptibility_old
))
2406 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO
, interruptibility
);
2409 static void skip_emulated_instruction(struct kvm_vcpu
*vcpu
)
2413 rip
= kvm_rip_read(vcpu
);
2414 rip
+= vmcs_read32(VM_EXIT_INSTRUCTION_LEN
);
2415 kvm_rip_write(vcpu
, rip
);
2417 /* skipping an emulated instruction also counts */
2418 vmx_set_interrupt_shadow(vcpu
, 0);
2422 * KVM wants to inject page-faults which it got to the guest. This function
2423 * checks whether in a nested guest, we need to inject them to L1 or L2.
2425 static int nested_vmx_check_exception(struct kvm_vcpu
*vcpu
, unsigned nr
)
2427 struct vmcs12
*vmcs12
= get_vmcs12(vcpu
);
2429 if (!(vmcs12
->exception_bitmap
& (1u << nr
)))
2432 nested_vmx_vmexit(vcpu
, EXIT_REASON_EXCEPTION_NMI
,
2433 vmcs_read32(VM_EXIT_INTR_INFO
),
2434 vmcs_readl(EXIT_QUALIFICATION
));
2438 static void vmx_queue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
,
2439 bool has_error_code
, u32 error_code
,
2442 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
2443 u32 intr_info
= nr
| INTR_INFO_VALID_MASK
;
2445 if (!reinject
&& is_guest_mode(vcpu
) &&
2446 nested_vmx_check_exception(vcpu
, nr
))
2449 if (has_error_code
) {
2450 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE
, error_code
);
2451 intr_info
|= INTR_INFO_DELIVER_CODE_MASK
;
2454 if (vmx
->rmode
.vm86_active
) {
2456 if (kvm_exception_is_soft(nr
))
2457 inc_eip
= vcpu
->arch
.event_exit_inst_len
;
2458 if (kvm_inject_realmode_interrupt(vcpu
, nr
, inc_eip
) != EMULATE_DONE
)
2459 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, vcpu
);
2463 if (kvm_exception_is_soft(nr
)) {
2464 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN
,
2465 vmx
->vcpu
.arch
.event_exit_inst_len
);
2466 intr_info
|= INTR_TYPE_SOFT_EXCEPTION
;
2468 intr_info
|= INTR_TYPE_HARD_EXCEPTION
;
2470 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD
, intr_info
);
2473 static bool vmx_rdtscp_supported(void)
2475 return cpu_has_vmx_rdtscp();
2478 static bool vmx_invpcid_supported(void)
2480 return cpu_has_vmx_invpcid() && enable_ept
;
2484 * Swap MSR entry in host/guest MSR entry array.
2486 static void move_msr_up(struct vcpu_vmx
*vmx
, int from
, int to
)
2488 struct shared_msr_entry tmp
;
2490 tmp
= vmx
->guest_msrs
[to
];
2491 vmx
->guest_msrs
[to
] = vmx
->guest_msrs
[from
];
2492 vmx
->guest_msrs
[from
] = tmp
;
2495 static void vmx_set_msr_bitmap(struct kvm_vcpu
*vcpu
)
2497 unsigned long *msr_bitmap
;
2499 if (is_guest_mode(vcpu
))
2500 msr_bitmap
= to_vmx(vcpu
)->nested
.msr_bitmap
;
2501 else if (cpu_has_secondary_exec_ctrls() &&
2502 (vmcs_read32(SECONDARY_VM_EXEC_CONTROL
) &
2503 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE
)) {
2504 if (enable_apicv
&& kvm_vcpu_apicv_active(vcpu
)) {
2505 if (is_long_mode(vcpu
))
2506 msr_bitmap
= vmx_msr_bitmap_longmode_x2apic_apicv
;
2508 msr_bitmap
= vmx_msr_bitmap_legacy_x2apic_apicv
;
2510 if (is_long_mode(vcpu
))
2511 msr_bitmap
= vmx_msr_bitmap_longmode_x2apic
;
2513 msr_bitmap
= vmx_msr_bitmap_legacy_x2apic
;
2516 if (is_long_mode(vcpu
))
2517 msr_bitmap
= vmx_msr_bitmap_longmode
;
2519 msr_bitmap
= vmx_msr_bitmap_legacy
;
2522 vmcs_write64(MSR_BITMAP
, __pa(msr_bitmap
));
2526 * Set up the vmcs to automatically save and restore system
2527 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
2528 * mode, as fiddling with msrs is very expensive.
2530 static void setup_msrs(struct vcpu_vmx
*vmx
)
2532 int save_nmsrs
, index
;
2535 #ifdef CONFIG_X86_64
2536 if (is_long_mode(&vmx
->vcpu
)) {
2537 index
= __find_msr_index(vmx
, MSR_SYSCALL_MASK
);
2539 move_msr_up(vmx
, index
, save_nmsrs
++);
2540 index
= __find_msr_index(vmx
, MSR_LSTAR
);
2542 move_msr_up(vmx
, index
, save_nmsrs
++);
2543 index
= __find_msr_index(vmx
, MSR_CSTAR
);
2545 move_msr_up(vmx
, index
, save_nmsrs
++);
2546 index
= __find_msr_index(vmx
, MSR_TSC_AUX
);
2547 if (index
>= 0 && guest_cpuid_has_rdtscp(&vmx
->vcpu
))
2548 move_msr_up(vmx
, index
, save_nmsrs
++);
2550 * MSR_STAR is only needed on long mode guests, and only
2551 * if efer.sce is enabled.
2553 index
= __find_msr_index(vmx
, MSR_STAR
);
2554 if ((index
>= 0) && (vmx
->vcpu
.arch
.efer
& EFER_SCE
))
2555 move_msr_up(vmx
, index
, save_nmsrs
++);
2558 index
= __find_msr_index(vmx
, MSR_EFER
);
2559 if (index
>= 0 && update_transition_efer(vmx
, index
))
2560 move_msr_up(vmx
, index
, save_nmsrs
++);
2562 vmx
->save_nmsrs
= save_nmsrs
;
2564 if (cpu_has_vmx_msr_bitmap())
2565 vmx_set_msr_bitmap(&vmx
->vcpu
);
2569 * reads and returns guest's timestamp counter "register"
2570 * guest_tsc = (host_tsc * tsc multiplier) >> 48 + tsc_offset
2571 * -- Intel TSC Scaling for Virtualization White Paper, sec 1.3
2573 static u64
guest_read_tsc(struct kvm_vcpu
*vcpu
)
2575 u64 host_tsc
, tsc_offset
;
2578 tsc_offset
= vmcs_read64(TSC_OFFSET
);
2579 return kvm_scale_tsc(vcpu
, host_tsc
) + tsc_offset
;
2583 * writes 'offset' into guest's timestamp counter offset register
2585 static void vmx_write_tsc_offset(struct kvm_vcpu
*vcpu
, u64 offset
)
2587 if (is_guest_mode(vcpu
)) {
2589 * We're here if L1 chose not to trap WRMSR to TSC. According
2590 * to the spec, this should set L1's TSC; The offset that L1
2591 * set for L2 remains unchanged, and still needs to be added
2592 * to the newly set TSC to get L2's TSC.
2594 struct vmcs12
*vmcs12
;
2595 /* recalculate vmcs02.TSC_OFFSET: */
2596 vmcs12
= get_vmcs12(vcpu
);
2597 vmcs_write64(TSC_OFFSET
, offset
+
2598 (nested_cpu_has(vmcs12
, CPU_BASED_USE_TSC_OFFSETING
) ?
2599 vmcs12
->tsc_offset
: 0));
2601 trace_kvm_write_tsc_offset(vcpu
->vcpu_id
,
2602 vmcs_read64(TSC_OFFSET
), offset
);
2603 vmcs_write64(TSC_OFFSET
, offset
);
2607 static bool guest_cpuid_has_vmx(struct kvm_vcpu
*vcpu
)
2609 struct kvm_cpuid_entry2
*best
= kvm_find_cpuid_entry(vcpu
, 1, 0);
2610 return best
&& (best
->ecx
& (1 << (X86_FEATURE_VMX
& 31)));
2614 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
2615 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
2616 * all guests if the "nested" module option is off, and can also be disabled
2617 * for a single guest by disabling its VMX cpuid bit.
2619 static inline bool nested_vmx_allowed(struct kvm_vcpu
*vcpu
)
2621 return nested
&& guest_cpuid_has_vmx(vcpu
);
2625 * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
2626 * returned for the various VMX controls MSRs when nested VMX is enabled.
2627 * The same values should also be used to verify that vmcs12 control fields are
2628 * valid during nested entry from L1 to L2.
2629 * Each of these control msrs has a low and high 32-bit half: A low bit is on
2630 * if the corresponding bit in the (32-bit) control field *must* be on, and a
2631 * bit in the high half is on if the corresponding bit in the control field
2632 * may be on. See also vmx_control_verify().
2634 static void nested_vmx_setup_ctls_msrs(struct vcpu_vmx
*vmx
)
2637 * Note that as a general rule, the high half of the MSRs (bits in
2638 * the control fields which may be 1) should be initialized by the
2639 * intersection of the underlying hardware's MSR (i.e., features which
2640 * can be supported) and the list of features we want to expose -
2641 * because they are known to be properly supported in our code.
2642 * Also, usually, the low half of the MSRs (bits which must be 1) can
2643 * be set to 0, meaning that L1 may turn off any of these bits. The
2644 * reason is that if one of these bits is necessary, it will appear
2645 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
2646 * fields of vmcs01 and vmcs02, will turn these bits off - and
2647 * nested_vmx_exit_handled() will not pass related exits to L1.
2648 * These rules have exceptions below.
2651 /* pin-based controls */
2652 rdmsr(MSR_IA32_VMX_PINBASED_CTLS
,
2653 vmx
->nested
.nested_vmx_pinbased_ctls_low
,
2654 vmx
->nested
.nested_vmx_pinbased_ctls_high
);
2655 vmx
->nested
.nested_vmx_pinbased_ctls_low
|=
2656 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR
;
2657 vmx
->nested
.nested_vmx_pinbased_ctls_high
&=
2658 PIN_BASED_EXT_INTR_MASK
|
2659 PIN_BASED_NMI_EXITING
|
2660 PIN_BASED_VIRTUAL_NMIS
;
2661 vmx
->nested
.nested_vmx_pinbased_ctls_high
|=
2662 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR
|
2663 PIN_BASED_VMX_PREEMPTION_TIMER
;
2664 if (kvm_vcpu_apicv_active(&vmx
->vcpu
))
2665 vmx
->nested
.nested_vmx_pinbased_ctls_high
|=
2666 PIN_BASED_POSTED_INTR
;
2669 rdmsr(MSR_IA32_VMX_EXIT_CTLS
,
2670 vmx
->nested
.nested_vmx_exit_ctls_low
,
2671 vmx
->nested
.nested_vmx_exit_ctls_high
);
2672 vmx
->nested
.nested_vmx_exit_ctls_low
=
2673 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR
;
2675 vmx
->nested
.nested_vmx_exit_ctls_high
&=
2676 #ifdef CONFIG_X86_64
2677 VM_EXIT_HOST_ADDR_SPACE_SIZE
|
2679 VM_EXIT_LOAD_IA32_PAT
| VM_EXIT_SAVE_IA32_PAT
;
2680 vmx
->nested
.nested_vmx_exit_ctls_high
|=
2681 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR
|
2682 VM_EXIT_LOAD_IA32_EFER
| VM_EXIT_SAVE_IA32_EFER
|
2683 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER
| VM_EXIT_ACK_INTR_ON_EXIT
;
2685 if (kvm_mpx_supported())
2686 vmx
->nested
.nested_vmx_exit_ctls_high
|= VM_EXIT_CLEAR_BNDCFGS
;
2688 /* We support free control of debug control saving. */
2689 vmx
->nested
.nested_vmx_exit_ctls_low
&= ~VM_EXIT_SAVE_DEBUG_CONTROLS
;
2691 /* entry controls */
2692 rdmsr(MSR_IA32_VMX_ENTRY_CTLS
,
2693 vmx
->nested
.nested_vmx_entry_ctls_low
,
2694 vmx
->nested
.nested_vmx_entry_ctls_high
);
2695 vmx
->nested
.nested_vmx_entry_ctls_low
=
2696 VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR
;
2697 vmx
->nested
.nested_vmx_entry_ctls_high
&=
2698 #ifdef CONFIG_X86_64
2699 VM_ENTRY_IA32E_MODE
|
2701 VM_ENTRY_LOAD_IA32_PAT
;
2702 vmx
->nested
.nested_vmx_entry_ctls_high
|=
2703 (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR
| VM_ENTRY_LOAD_IA32_EFER
);
2704 if (kvm_mpx_supported())
2705 vmx
->nested
.nested_vmx_entry_ctls_high
|= VM_ENTRY_LOAD_BNDCFGS
;
2707 /* We support free control of debug control loading. */
2708 vmx
->nested
.nested_vmx_entry_ctls_low
&= ~VM_ENTRY_LOAD_DEBUG_CONTROLS
;
2710 /* cpu-based controls */
2711 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS
,
2712 vmx
->nested
.nested_vmx_procbased_ctls_low
,
2713 vmx
->nested
.nested_vmx_procbased_ctls_high
);
2714 vmx
->nested
.nested_vmx_procbased_ctls_low
=
2715 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR
;
2716 vmx
->nested
.nested_vmx_procbased_ctls_high
&=
2717 CPU_BASED_VIRTUAL_INTR_PENDING
|
2718 CPU_BASED_VIRTUAL_NMI_PENDING
| CPU_BASED_USE_TSC_OFFSETING
|
2719 CPU_BASED_HLT_EXITING
| CPU_BASED_INVLPG_EXITING
|
2720 CPU_BASED_MWAIT_EXITING
| CPU_BASED_CR3_LOAD_EXITING
|
2721 CPU_BASED_CR3_STORE_EXITING
|
2722 #ifdef CONFIG_X86_64
2723 CPU_BASED_CR8_LOAD_EXITING
| CPU_BASED_CR8_STORE_EXITING
|
2725 CPU_BASED_MOV_DR_EXITING
| CPU_BASED_UNCOND_IO_EXITING
|
2726 CPU_BASED_USE_IO_BITMAPS
| CPU_BASED_MONITOR_TRAP_FLAG
|
2727 CPU_BASED_MONITOR_EXITING
| CPU_BASED_RDPMC_EXITING
|
2728 CPU_BASED_RDTSC_EXITING
| CPU_BASED_PAUSE_EXITING
|
2729 CPU_BASED_TPR_SHADOW
| CPU_BASED_ACTIVATE_SECONDARY_CONTROLS
;
2731 * We can allow some features even when not supported by the
2732 * hardware. For example, L1 can specify an MSR bitmap - and we
2733 * can use it to avoid exits to L1 - even when L0 runs L2
2734 * without MSR bitmaps.
2736 vmx
->nested
.nested_vmx_procbased_ctls_high
|=
2737 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR
|
2738 CPU_BASED_USE_MSR_BITMAPS
;
2740 /* We support free control of CR3 access interception. */
2741 vmx
->nested
.nested_vmx_procbased_ctls_low
&=
2742 ~(CPU_BASED_CR3_LOAD_EXITING
| CPU_BASED_CR3_STORE_EXITING
);
2744 /* secondary cpu-based controls */
2745 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2
,
2746 vmx
->nested
.nested_vmx_secondary_ctls_low
,
2747 vmx
->nested
.nested_vmx_secondary_ctls_high
);
2748 vmx
->nested
.nested_vmx_secondary_ctls_low
= 0;
2749 vmx
->nested
.nested_vmx_secondary_ctls_high
&=
2750 SECONDARY_EXEC_RDRAND
| SECONDARY_EXEC_RDSEED
|
2751 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES
|
2752 SECONDARY_EXEC_RDTSCP
|
2753 SECONDARY_EXEC_DESC
|
2754 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE
|
2755 SECONDARY_EXEC_APIC_REGISTER_VIRT
|
2756 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY
|
2757 SECONDARY_EXEC_WBINVD_EXITING
|
2758 SECONDARY_EXEC_XSAVES
;
2761 /* nested EPT: emulate EPT also to L1 */
2762 vmx
->nested
.nested_vmx_secondary_ctls_high
|=
2763 SECONDARY_EXEC_ENABLE_EPT
;
2764 vmx
->nested
.nested_vmx_ept_caps
= VMX_EPT_PAGE_WALK_4_BIT
|
2765 VMX_EPTP_WB_BIT
| VMX_EPT_INVEPT_BIT
;
2766 if (cpu_has_vmx_ept_execute_only())
2767 vmx
->nested
.nested_vmx_ept_caps
|=
2768 VMX_EPT_EXECUTE_ONLY_BIT
;
2769 vmx
->nested
.nested_vmx_ept_caps
&= vmx_capability
.ept
;
2770 vmx
->nested
.nested_vmx_ept_caps
|= VMX_EPT_EXTENT_GLOBAL_BIT
|
2771 VMX_EPT_EXTENT_CONTEXT_BIT
| VMX_EPT_2MB_PAGE_BIT
|
2772 VMX_EPT_1GB_PAGE_BIT
;
2773 if (enable_ept_ad_bits
) {
2774 vmx
->nested
.nested_vmx_secondary_ctls_high
|=
2775 SECONDARY_EXEC_ENABLE_PML
;
2776 vmx
->nested
.nested_vmx_ept_caps
|= VMX_EPT_AD_BIT
;
2779 vmx
->nested
.nested_vmx_ept_caps
= 0;
2782 * Old versions of KVM use the single-context version without
2783 * checking for support, so declare that it is supported even
2784 * though it is treated as global context. The alternative is
2785 * not failing the single-context invvpid, and it is worse.
2788 vmx
->nested
.nested_vmx_secondary_ctls_high
|=
2789 SECONDARY_EXEC_ENABLE_VPID
;
2790 vmx
->nested
.nested_vmx_vpid_caps
= VMX_VPID_INVVPID_BIT
|
2791 VMX_VPID_EXTENT_SUPPORTED_MASK
;
2793 vmx
->nested
.nested_vmx_vpid_caps
= 0;
2795 if (enable_unrestricted_guest
)
2796 vmx
->nested
.nested_vmx_secondary_ctls_high
|=
2797 SECONDARY_EXEC_UNRESTRICTED_GUEST
;
2799 /* miscellaneous data */
2800 rdmsr(MSR_IA32_VMX_MISC
,
2801 vmx
->nested
.nested_vmx_misc_low
,
2802 vmx
->nested
.nested_vmx_misc_high
);
2803 vmx
->nested
.nested_vmx_misc_low
&= VMX_MISC_SAVE_EFER_LMA
;
2804 vmx
->nested
.nested_vmx_misc_low
|=
2805 VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE
|
2806 VMX_MISC_ACTIVITY_HLT
;
2807 vmx
->nested
.nested_vmx_misc_high
= 0;
2810 * This MSR reports some information about VMX support. We
2811 * should return information about the VMX we emulate for the
2812 * guest, and the VMCS structure we give it - not about the
2813 * VMX support of the underlying hardware.
2815 vmx
->nested
.nested_vmx_basic
=
2817 VMX_BASIC_TRUE_CTLS
|
2818 ((u64
)VMCS12_SIZE
<< VMX_BASIC_VMCS_SIZE_SHIFT
) |
2819 (VMX_BASIC_MEM_TYPE_WB
<< VMX_BASIC_MEM_TYPE_SHIFT
);
2821 if (cpu_has_vmx_basic_inout())
2822 vmx
->nested
.nested_vmx_basic
|= VMX_BASIC_INOUT
;
2825 * These MSRs specify bits which the guest must keep fixed on
2826 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2827 * We picked the standard core2 setting.
2829 #define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2830 #define VMXON_CR4_ALWAYSON X86_CR4_VMXE
2831 vmx
->nested
.nested_vmx_cr0_fixed0
= VMXON_CR0_ALWAYSON
;
2832 vmx
->nested
.nested_vmx_cr4_fixed0
= VMXON_CR4_ALWAYSON
;
2834 /* These MSRs specify bits which the guest must keep fixed off. */
2835 rdmsrl(MSR_IA32_VMX_CR0_FIXED1
, vmx
->nested
.nested_vmx_cr0_fixed1
);
2836 rdmsrl(MSR_IA32_VMX_CR4_FIXED1
, vmx
->nested
.nested_vmx_cr4_fixed1
);
2838 /* highest index: VMX_PREEMPTION_TIMER_VALUE */
2839 vmx
->nested
.nested_vmx_vmcs_enum
= 0x2e;
2843 * if fixed0[i] == 1: val[i] must be 1
2844 * if fixed1[i] == 0: val[i] must be 0
2846 static inline bool fixed_bits_valid(u64 val
, u64 fixed0
, u64 fixed1
)
2848 return ((val
& fixed1
) | fixed0
) == val
;
2851 static inline bool vmx_control_verify(u32 control
, u32 low
, u32 high
)
2853 return fixed_bits_valid(control
, low
, high
);
2856 static inline u64
vmx_control_msr(u32 low
, u32 high
)
2858 return low
| ((u64
)high
<< 32);
2861 static bool is_bitwise_subset(u64 superset
, u64 subset
, u64 mask
)
2866 return (superset
| subset
) == superset
;
2869 static int vmx_restore_vmx_basic(struct vcpu_vmx
*vmx
, u64 data
)
2871 const u64 feature_and_reserved
=
2872 /* feature (except bit 48; see below) */
2873 BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
2875 BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
2876 u64 vmx_basic
= vmx
->nested
.nested_vmx_basic
;
2878 if (!is_bitwise_subset(vmx_basic
, data
, feature_and_reserved
))
2882 * KVM does not emulate a version of VMX that constrains physical
2883 * addresses of VMX structures (e.g. VMCS) to 32-bits.
2885 if (data
& BIT_ULL(48))
2888 if (vmx_basic_vmcs_revision_id(vmx_basic
) !=
2889 vmx_basic_vmcs_revision_id(data
))
2892 if (vmx_basic_vmcs_size(vmx_basic
) > vmx_basic_vmcs_size(data
))
2895 vmx
->nested
.nested_vmx_basic
= data
;
2900 vmx_restore_control_msr(struct vcpu_vmx
*vmx
, u32 msr_index
, u64 data
)
2905 switch (msr_index
) {
2906 case MSR_IA32_VMX_TRUE_PINBASED_CTLS
:
2907 lowp
= &vmx
->nested
.nested_vmx_pinbased_ctls_low
;
2908 highp
= &vmx
->nested
.nested_vmx_pinbased_ctls_high
;
2910 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS
:
2911 lowp
= &vmx
->nested
.nested_vmx_procbased_ctls_low
;
2912 highp
= &vmx
->nested
.nested_vmx_procbased_ctls_high
;
2914 case MSR_IA32_VMX_TRUE_EXIT_CTLS
:
2915 lowp
= &vmx
->nested
.nested_vmx_exit_ctls_low
;
2916 highp
= &vmx
->nested
.nested_vmx_exit_ctls_high
;
2918 case MSR_IA32_VMX_TRUE_ENTRY_CTLS
:
2919 lowp
= &vmx
->nested
.nested_vmx_entry_ctls_low
;
2920 highp
= &vmx
->nested
.nested_vmx_entry_ctls_high
;
2922 case MSR_IA32_VMX_PROCBASED_CTLS2
:
2923 lowp
= &vmx
->nested
.nested_vmx_secondary_ctls_low
;
2924 highp
= &vmx
->nested
.nested_vmx_secondary_ctls_high
;
2930 supported
= vmx_control_msr(*lowp
, *highp
);
2932 /* Check must-be-1 bits are still 1. */
2933 if (!is_bitwise_subset(data
, supported
, GENMASK_ULL(31, 0)))
2936 /* Check must-be-0 bits are still 0. */
2937 if (!is_bitwise_subset(supported
, data
, GENMASK_ULL(63, 32)))
2941 *highp
= data
>> 32;
2945 static int vmx_restore_vmx_misc(struct vcpu_vmx
*vmx
, u64 data
)
2947 const u64 feature_and_reserved_bits
=
2949 BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
2950 BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
2952 GENMASK_ULL(13, 9) | BIT_ULL(31);
2955 vmx_misc
= vmx_control_msr(vmx
->nested
.nested_vmx_misc_low
,
2956 vmx
->nested
.nested_vmx_misc_high
);
2958 if (!is_bitwise_subset(vmx_misc
, data
, feature_and_reserved_bits
))
2961 if ((vmx
->nested
.nested_vmx_pinbased_ctls_high
&
2962 PIN_BASED_VMX_PREEMPTION_TIMER
) &&
2963 vmx_misc_preemption_timer_rate(data
) !=
2964 vmx_misc_preemption_timer_rate(vmx_misc
))
2967 if (vmx_misc_cr3_count(data
) > vmx_misc_cr3_count(vmx_misc
))
2970 if (vmx_misc_max_msr(data
) > vmx_misc_max_msr(vmx_misc
))
2973 if (vmx_misc_mseg_revid(data
) != vmx_misc_mseg_revid(vmx_misc
))
2976 vmx
->nested
.nested_vmx_misc_low
= data
;
2977 vmx
->nested
.nested_vmx_misc_high
= data
>> 32;
2981 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx
*vmx
, u64 data
)
2983 u64 vmx_ept_vpid_cap
;
2985 vmx_ept_vpid_cap
= vmx_control_msr(vmx
->nested
.nested_vmx_ept_caps
,
2986 vmx
->nested
.nested_vmx_vpid_caps
);
2988 /* Every bit is either reserved or a feature bit. */
2989 if (!is_bitwise_subset(vmx_ept_vpid_cap
, data
, -1ULL))
2992 vmx
->nested
.nested_vmx_ept_caps
= data
;
2993 vmx
->nested
.nested_vmx_vpid_caps
= data
>> 32;
2997 static int vmx_restore_fixed0_msr(struct vcpu_vmx
*vmx
, u32 msr_index
, u64 data
)
3001 switch (msr_index
) {
3002 case MSR_IA32_VMX_CR0_FIXED0
:
3003 msr
= &vmx
->nested
.nested_vmx_cr0_fixed0
;
3005 case MSR_IA32_VMX_CR4_FIXED0
:
3006 msr
= &vmx
->nested
.nested_vmx_cr4_fixed0
;
3013 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
3014 * must be 1 in the restored value.
3016 if (!is_bitwise_subset(data
, *msr
, -1ULL))
3024 * Called when userspace is restoring VMX MSRs.
3026 * Returns 0 on success, non-0 otherwise.
3028 static int vmx_set_vmx_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64 data
)
3030 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
3032 switch (msr_index
) {
3033 case MSR_IA32_VMX_BASIC
:
3034 return vmx_restore_vmx_basic(vmx
, data
);
3035 case MSR_IA32_VMX_PINBASED_CTLS
:
3036 case MSR_IA32_VMX_PROCBASED_CTLS
:
3037 case MSR_IA32_VMX_EXIT_CTLS
:
3038 case MSR_IA32_VMX_ENTRY_CTLS
:
3040 * The "non-true" VMX capability MSRs are generated from the
3041 * "true" MSRs, so we do not support restoring them directly.
3043 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
3044 * should restore the "true" MSRs with the must-be-1 bits
3045 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
3046 * DEFAULT SETTINGS".
3049 case MSR_IA32_VMX_TRUE_PINBASED_CTLS
:
3050 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS
:
3051 case MSR_IA32_VMX_TRUE_EXIT_CTLS
:
3052 case MSR_IA32_VMX_TRUE_ENTRY_CTLS
:
3053 case MSR_IA32_VMX_PROCBASED_CTLS2
:
3054 return vmx_restore_control_msr(vmx
, msr_index
, data
);
3055 case MSR_IA32_VMX_MISC
:
3056 return vmx_restore_vmx_misc(vmx
, data
);
3057 case MSR_IA32_VMX_CR0_FIXED0
:
3058 case MSR_IA32_VMX_CR4_FIXED0
:
3059 return vmx_restore_fixed0_msr(vmx
, msr_index
, data
);
3060 case MSR_IA32_VMX_CR0_FIXED1
:
3061 case MSR_IA32_VMX_CR4_FIXED1
:
3063 * These MSRs are generated based on the vCPU's CPUID, so we
3064 * do not support restoring them directly.
3067 case MSR_IA32_VMX_EPT_VPID_CAP
:
3068 return vmx_restore_vmx_ept_vpid_cap(vmx
, data
);
3069 case MSR_IA32_VMX_VMCS_ENUM
:
3070 vmx
->nested
.nested_vmx_vmcs_enum
= data
;
3074 * The rest of the VMX capability MSRs do not support restore.
3080 /* Returns 0 on success, non-0 otherwise. */
3081 static int vmx_get_vmx_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64
*pdata
)
3083 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
3085 switch (msr_index
) {
3086 case MSR_IA32_VMX_BASIC
:
3087 *pdata
= vmx
->nested
.nested_vmx_basic
;
3089 case MSR_IA32_VMX_TRUE_PINBASED_CTLS
:
3090 case MSR_IA32_VMX_PINBASED_CTLS
:
3091 *pdata
= vmx_control_msr(
3092 vmx
->nested
.nested_vmx_pinbased_ctls_low
,
3093 vmx
->nested
.nested_vmx_pinbased_ctls_high
);
3094 if (msr_index
== MSR_IA32_VMX_PINBASED_CTLS
)
3095 *pdata
|= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR
;
3097 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS
:
3098 case MSR_IA32_VMX_PROCBASED_CTLS
:
3099 *pdata
= vmx_control_msr(
3100 vmx
->nested
.nested_vmx_procbased_ctls_low
,
3101 vmx
->nested
.nested_vmx_procbased_ctls_high
);
3102 if (msr_index
== MSR_IA32_VMX_PROCBASED_CTLS
)
3103 *pdata
|= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR
;
3105 case MSR_IA32_VMX_TRUE_EXIT_CTLS
:
3106 case MSR_IA32_VMX_EXIT_CTLS
:
3107 *pdata
= vmx_control_msr(
3108 vmx
->nested
.nested_vmx_exit_ctls_low
,
3109 vmx
->nested
.nested_vmx_exit_ctls_high
);
3110 if (msr_index
== MSR_IA32_VMX_EXIT_CTLS
)
3111 *pdata
|= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR
;
3113 case MSR_IA32_VMX_TRUE_ENTRY_CTLS
:
3114 case MSR_IA32_VMX_ENTRY_CTLS
:
3115 *pdata
= vmx_control_msr(
3116 vmx
->nested
.nested_vmx_entry_ctls_low
,
3117 vmx
->nested
.nested_vmx_entry_ctls_high
);
3118 if (msr_index
== MSR_IA32_VMX_ENTRY_CTLS
)
3119 *pdata
|= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR
;
3121 case MSR_IA32_VMX_MISC
:
3122 *pdata
= vmx_control_msr(
3123 vmx
->nested
.nested_vmx_misc_low
,
3124 vmx
->nested
.nested_vmx_misc_high
);
3126 case MSR_IA32_VMX_CR0_FIXED0
:
3127 *pdata
= vmx
->nested
.nested_vmx_cr0_fixed0
;
3129 case MSR_IA32_VMX_CR0_FIXED1
:
3130 *pdata
= vmx
->nested
.nested_vmx_cr0_fixed1
;
3132 case MSR_IA32_VMX_CR4_FIXED0
:
3133 *pdata
= vmx
->nested
.nested_vmx_cr4_fixed0
;
3135 case MSR_IA32_VMX_CR4_FIXED1
:
3136 *pdata
= vmx
->nested
.nested_vmx_cr4_fixed1
;
3138 case MSR_IA32_VMX_VMCS_ENUM
:
3139 *pdata
= vmx
->nested
.nested_vmx_vmcs_enum
;
3141 case MSR_IA32_VMX_PROCBASED_CTLS2
:
3142 *pdata
= vmx_control_msr(
3143 vmx
->nested
.nested_vmx_secondary_ctls_low
,
3144 vmx
->nested
.nested_vmx_secondary_ctls_high
);
3146 case MSR_IA32_VMX_EPT_VPID_CAP
:
3147 *pdata
= vmx
->nested
.nested_vmx_ept_caps
|
3148 ((u64
)vmx
->nested
.nested_vmx_vpid_caps
<< 32);
3157 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu
*vcpu
,
3160 uint64_t valid_bits
= to_vmx(vcpu
)->msr_ia32_feature_control_valid_bits
;
3162 return !(val
& ~valid_bits
);
3166 * Reads an msr value (of 'msr_index') into 'pdata'.
3167 * Returns 0 on success, non-0 otherwise.
3168 * Assumes vcpu_load() was already called.
3170 static int vmx_get_msr(struct kvm_vcpu
*vcpu
, struct msr_data
*msr_info
)
3172 struct shared_msr_entry
*msr
;
3174 switch (msr_info
->index
) {
3175 #ifdef CONFIG_X86_64
3177 msr_info
->data
= vmcs_readl(GUEST_FS_BASE
);
3180 msr_info
->data
= vmcs_readl(GUEST_GS_BASE
);
3182 case MSR_KERNEL_GS_BASE
:
3183 vmx_load_host_state(to_vmx(vcpu
));
3184 msr_info
->data
= to_vmx(vcpu
)->msr_guest_kernel_gs_base
;
3188 return kvm_get_msr_common(vcpu
, msr_info
);
3190 msr_info
->data
= guest_read_tsc(vcpu
);
3192 case MSR_IA32_SYSENTER_CS
:
3193 msr_info
->data
= vmcs_read32(GUEST_SYSENTER_CS
);
3195 case MSR_IA32_SYSENTER_EIP
:
3196 msr_info
->data
= vmcs_readl(GUEST_SYSENTER_EIP
);
3198 case MSR_IA32_SYSENTER_ESP
:
3199 msr_info
->data
= vmcs_readl(GUEST_SYSENTER_ESP
);
3201 case MSR_IA32_BNDCFGS
:
3202 if (!kvm_mpx_supported() ||
3203 (!msr_info
->host_initiated
&& !guest_cpuid_has_mpx(vcpu
)))
3205 msr_info
->data
= vmcs_read64(GUEST_BNDCFGS
);
3207 case MSR_IA32_MCG_EXT_CTL
:
3208 if (!msr_info
->host_initiated
&&
3209 !(to_vmx(vcpu
)->msr_ia32_feature_control
&
3210 FEATURE_CONTROL_LMCE
))
3212 msr_info
->data
= vcpu
->arch
.mcg_ext_ctl
;
3214 case MSR_IA32_FEATURE_CONTROL
:
3215 msr_info
->data
= to_vmx(vcpu
)->msr_ia32_feature_control
;
3217 case MSR_IA32_VMX_BASIC
... MSR_IA32_VMX_VMFUNC
:
3218 if (!nested_vmx_allowed(vcpu
))
3220 return vmx_get_vmx_msr(vcpu
, msr_info
->index
, &msr_info
->data
);
3222 if (!vmx_xsaves_supported())
3224 msr_info
->data
= vcpu
->arch
.ia32_xss
;
3227 if (!guest_cpuid_has_rdtscp(vcpu
) && !msr_info
->host_initiated
)
3229 /* Otherwise falls through */
3231 msr
= find_msr_entry(to_vmx(vcpu
), msr_info
->index
);
3233 msr_info
->data
= msr
->data
;
3236 return kvm_get_msr_common(vcpu
, msr_info
);
3242 static void vmx_leave_nested(struct kvm_vcpu
*vcpu
);
3245 * Writes msr value into into the appropriate "register".
3246 * Returns 0 on success, non-0 otherwise.
3247 * Assumes vcpu_load() was already called.
3249 static int vmx_set_msr(struct kvm_vcpu
*vcpu
, struct msr_data
*msr_info
)
3251 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
3252 struct shared_msr_entry
*msr
;
3254 u32 msr_index
= msr_info
->index
;
3255 u64 data
= msr_info
->data
;
3257 switch (msr_index
) {
3259 ret
= kvm_set_msr_common(vcpu
, msr_info
);
3261 #ifdef CONFIG_X86_64
3263 vmx_segment_cache_clear(vmx
);
3264 vmcs_writel(GUEST_FS_BASE
, data
);
3267 vmx_segment_cache_clear(vmx
);
3268 vmcs_writel(GUEST_GS_BASE
, data
);
3270 case MSR_KERNEL_GS_BASE
:
3271 vmx_load_host_state(vmx
);
3272 vmx
->msr_guest_kernel_gs_base
= data
;
3275 case MSR_IA32_SYSENTER_CS
:
3276 vmcs_write32(GUEST_SYSENTER_CS
, data
);
3278 case MSR_IA32_SYSENTER_EIP
:
3279 vmcs_writel(GUEST_SYSENTER_EIP
, data
);
3281 case MSR_IA32_SYSENTER_ESP
:
3282 vmcs_writel(GUEST_SYSENTER_ESP
, data
);
3284 case MSR_IA32_BNDCFGS
:
3285 if (!kvm_mpx_supported() ||
3286 (!msr_info
->host_initiated
&& !guest_cpuid_has_mpx(vcpu
)))
3288 if (is_noncanonical_address(data
& PAGE_MASK
) ||
3289 (data
& MSR_IA32_BNDCFGS_RSVD
))
3291 vmcs_write64(GUEST_BNDCFGS
, data
);
3294 kvm_write_tsc(vcpu
, msr_info
);
3296 case MSR_IA32_CR_PAT
:
3297 if (vmcs_config
.vmentry_ctrl
& VM_ENTRY_LOAD_IA32_PAT
) {
3298 if (!kvm_mtrr_valid(vcpu
, MSR_IA32_CR_PAT
, data
))
3300 vmcs_write64(GUEST_IA32_PAT
, data
);
3301 vcpu
->arch
.pat
= data
;
3304 ret
= kvm_set_msr_common(vcpu
, msr_info
);
3306 case MSR_IA32_TSC_ADJUST
:
3307 ret
= kvm_set_msr_common(vcpu
, msr_info
);
3309 case MSR_IA32_MCG_EXT_CTL
:
3310 if ((!msr_info
->host_initiated
&&
3311 !(to_vmx(vcpu
)->msr_ia32_feature_control
&
3312 FEATURE_CONTROL_LMCE
)) ||
3313 (data
& ~MCG_EXT_CTL_LMCE_EN
))
3315 vcpu
->arch
.mcg_ext_ctl
= data
;
3317 case MSR_IA32_FEATURE_CONTROL
:
3318 if (!vmx_feature_control_msr_valid(vcpu
, data
) ||
3319 (to_vmx(vcpu
)->msr_ia32_feature_control
&
3320 FEATURE_CONTROL_LOCKED
&& !msr_info
->host_initiated
))
3322 vmx
->msr_ia32_feature_control
= data
;
3323 if (msr_info
->host_initiated
&& data
== 0)
3324 vmx_leave_nested(vcpu
);
3326 case MSR_IA32_VMX_BASIC
... MSR_IA32_VMX_VMFUNC
:
3327 if (!msr_info
->host_initiated
)
3328 return 1; /* they are read-only */
3329 if (!nested_vmx_allowed(vcpu
))
3331 return vmx_set_vmx_msr(vcpu
, msr_index
, data
);
3333 if (!vmx_xsaves_supported())
3336 * The only supported bit as of Skylake is bit 8, but
3337 * it is not supported on KVM.
3341 vcpu
->arch
.ia32_xss
= data
;
3342 if (vcpu
->arch
.ia32_xss
!= host_xss
)
3343 add_atomic_switch_msr(vmx
, MSR_IA32_XSS
,
3344 vcpu
->arch
.ia32_xss
, host_xss
);
3346 clear_atomic_switch_msr(vmx
, MSR_IA32_XSS
);
3349 if (!guest_cpuid_has_rdtscp(vcpu
) && !msr_info
->host_initiated
)
3351 /* Check reserved bit, higher 32 bits should be zero */
3352 if ((data
>> 32) != 0)
3354 /* Otherwise falls through */
3356 msr
= find_msr_entry(vmx
, msr_index
);
3358 u64 old_msr_data
= msr
->data
;
3360 if (msr
- vmx
->guest_msrs
< vmx
->save_nmsrs
) {
3362 ret
= kvm_set_shared_msr(msr
->index
, msr
->data
,
3366 msr
->data
= old_msr_data
;
3370 ret
= kvm_set_msr_common(vcpu
, msr_info
);
3376 static void vmx_cache_reg(struct kvm_vcpu
*vcpu
, enum kvm_reg reg
)
3378 __set_bit(reg
, (unsigned long *)&vcpu
->arch
.regs_avail
);
3381 vcpu
->arch
.regs
[VCPU_REGS_RSP
] = vmcs_readl(GUEST_RSP
);
3384 vcpu
->arch
.regs
[VCPU_REGS_RIP
] = vmcs_readl(GUEST_RIP
);
3386 case VCPU_EXREG_PDPTR
:
3388 ept_save_pdptrs(vcpu
);
3395 static __init
int cpu_has_kvm_support(void)
3397 return cpu_has_vmx();
3400 static __init
int vmx_disabled_by_bios(void)
3404 rdmsrl(MSR_IA32_FEATURE_CONTROL
, msr
);
3405 if (msr
& FEATURE_CONTROL_LOCKED
) {
3406 /* launched w/ TXT and VMX disabled */
3407 if (!(msr
& FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX
)
3410 /* launched w/o TXT and VMX only enabled w/ TXT */
3411 if (!(msr
& FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX
)
3412 && (msr
& FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX
)
3413 && !tboot_enabled()) {
3414 printk(KERN_WARNING
"kvm: disable TXT in the BIOS or "
3415 "activate TXT before enabling KVM\n");
3418 /* launched w/o TXT and VMX disabled */
3419 if (!(msr
& FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX
)
3420 && !tboot_enabled())
3427 static void kvm_cpu_vmxon(u64 addr
)
3429 cr4_set_bits(X86_CR4_VMXE
);
3430 intel_pt_handle_vmx(1);
3432 asm volatile (ASM_VMX_VMXON_RAX
3433 : : "a"(&addr
), "m"(addr
)
3437 static int hardware_enable(void)
3439 int cpu
= raw_smp_processor_id();
3440 u64 phys_addr
= __pa(per_cpu(vmxarea
, cpu
));
3443 if (cr4_read_shadow() & X86_CR4_VMXE
)
3446 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu
, cpu
));
3447 INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu
, cpu
));
3448 spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock
, cpu
));
3451 * Now we can enable the vmclear operation in kdump
3452 * since the loaded_vmcss_on_cpu list on this cpu
3453 * has been initialized.
3455 * Though the cpu is not in VMX operation now, there
3456 * is no problem to enable the vmclear operation
3457 * for the loaded_vmcss_on_cpu list is empty!
3459 crash_enable_local_vmclear(cpu
);
3461 rdmsrl(MSR_IA32_FEATURE_CONTROL
, old
);
3463 test_bits
= FEATURE_CONTROL_LOCKED
;
3464 test_bits
|= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX
;
3465 if (tboot_enabled())
3466 test_bits
|= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX
;
3468 if ((old
& test_bits
) != test_bits
) {
3469 /* enable and lock */
3470 wrmsrl(MSR_IA32_FEATURE_CONTROL
, old
| test_bits
);
3472 kvm_cpu_vmxon(phys_addr
);
3478 static void vmclear_local_loaded_vmcss(void)
3480 int cpu
= raw_smp_processor_id();
3481 struct loaded_vmcs
*v
, *n
;
3483 list_for_each_entry_safe(v
, n
, &per_cpu(loaded_vmcss_on_cpu
, cpu
),
3484 loaded_vmcss_on_cpu_link
)
3485 __loaded_vmcs_clear(v
);
3489 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
3492 static void kvm_cpu_vmxoff(void)
3494 asm volatile (__ex(ASM_VMX_VMXOFF
) : : : "cc");
3496 intel_pt_handle_vmx(0);
3497 cr4_clear_bits(X86_CR4_VMXE
);
3500 static void hardware_disable(void)
3502 vmclear_local_loaded_vmcss();
3506 static __init
int adjust_vmx_controls(u32 ctl_min
, u32 ctl_opt
,
3507 u32 msr
, u32
*result
)
3509 u32 vmx_msr_low
, vmx_msr_high
;
3510 u32 ctl
= ctl_min
| ctl_opt
;
3512 rdmsr(msr
, vmx_msr_low
, vmx_msr_high
);
3514 ctl
&= vmx_msr_high
; /* bit == 0 in high word ==> must be zero */
3515 ctl
|= vmx_msr_low
; /* bit == 1 in low word ==> must be one */
3517 /* Ensure minimum (required) set of control bits are supported. */
3525 static __init
bool allow_1_setting(u32 msr
, u32 ctl
)
3527 u32 vmx_msr_low
, vmx_msr_high
;
3529 rdmsr(msr
, vmx_msr_low
, vmx_msr_high
);
3530 return vmx_msr_high
& ctl
;
3533 static __init
int setup_vmcs_config(struct vmcs_config
*vmcs_conf
)
3535 u32 vmx_msr_low
, vmx_msr_high
;
3536 u32 min
, opt
, min2
, opt2
;
3537 u32 _pin_based_exec_control
= 0;
3538 u32 _cpu_based_exec_control
= 0;
3539 u32 _cpu_based_2nd_exec_control
= 0;
3540 u32 _vmexit_control
= 0;
3541 u32 _vmentry_control
= 0;
3543 min
= CPU_BASED_HLT_EXITING
|
3544 #ifdef CONFIG_X86_64
3545 CPU_BASED_CR8_LOAD_EXITING
|
3546 CPU_BASED_CR8_STORE_EXITING
|
3548 CPU_BASED_CR3_LOAD_EXITING
|
3549 CPU_BASED_CR3_STORE_EXITING
|
3550 CPU_BASED_USE_IO_BITMAPS
|
3551 CPU_BASED_MOV_DR_EXITING
|
3552 CPU_BASED_USE_TSC_OFFSETING
|
3553 CPU_BASED_INVLPG_EXITING
|
3554 CPU_BASED_RDPMC_EXITING
;
3556 if (!kvm_mwait_in_guest())
3557 min
|= CPU_BASED_MWAIT_EXITING
|
3558 CPU_BASED_MONITOR_EXITING
;
3560 opt
= CPU_BASED_TPR_SHADOW
|
3561 CPU_BASED_USE_MSR_BITMAPS
|
3562 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS
;
3563 if (adjust_vmx_controls(min
, opt
, MSR_IA32_VMX_PROCBASED_CTLS
,
3564 &_cpu_based_exec_control
) < 0)
3566 #ifdef CONFIG_X86_64
3567 if ((_cpu_based_exec_control
& CPU_BASED_TPR_SHADOW
))
3568 _cpu_based_exec_control
&= ~CPU_BASED_CR8_LOAD_EXITING
&
3569 ~CPU_BASED_CR8_STORE_EXITING
;
3571 if (_cpu_based_exec_control
& CPU_BASED_ACTIVATE_SECONDARY_CONTROLS
) {
3573 opt2
= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES
|
3574 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE
|
3575 SECONDARY_EXEC_WBINVD_EXITING
|
3576 SECONDARY_EXEC_ENABLE_VPID
|
3577 SECONDARY_EXEC_ENABLE_EPT
|
3578 SECONDARY_EXEC_UNRESTRICTED_GUEST
|
3579 SECONDARY_EXEC_PAUSE_LOOP_EXITING
|
3580 SECONDARY_EXEC_RDTSCP
|
3581 SECONDARY_EXEC_ENABLE_INVPCID
|
3582 SECONDARY_EXEC_APIC_REGISTER_VIRT
|
3583 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY
|
3584 SECONDARY_EXEC_SHADOW_VMCS
|
3585 SECONDARY_EXEC_XSAVES
|
3586 SECONDARY_EXEC_ENABLE_PML
|
3587 SECONDARY_EXEC_TSC_SCALING
;
3588 if (adjust_vmx_controls(min2
, opt2
,
3589 MSR_IA32_VMX_PROCBASED_CTLS2
,
3590 &_cpu_based_2nd_exec_control
) < 0)
3593 #ifndef CONFIG_X86_64
3594 if (!(_cpu_based_2nd_exec_control
&
3595 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES
))
3596 _cpu_based_exec_control
&= ~CPU_BASED_TPR_SHADOW
;
3599 if (!(_cpu_based_exec_control
& CPU_BASED_TPR_SHADOW
))
3600 _cpu_based_2nd_exec_control
&= ~(
3601 SECONDARY_EXEC_APIC_REGISTER_VIRT
|
3602 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE
|
3603 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY
);
3605 if (_cpu_based_2nd_exec_control
& SECONDARY_EXEC_ENABLE_EPT
) {
3606 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
3608 _cpu_based_exec_control
&= ~(CPU_BASED_CR3_LOAD_EXITING
|
3609 CPU_BASED_CR3_STORE_EXITING
|
3610 CPU_BASED_INVLPG_EXITING
);
3611 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP
,
3612 vmx_capability
.ept
, vmx_capability
.vpid
);
3615 min
= VM_EXIT_SAVE_DEBUG_CONTROLS
| VM_EXIT_ACK_INTR_ON_EXIT
;
3616 #ifdef CONFIG_X86_64
3617 min
|= VM_EXIT_HOST_ADDR_SPACE_SIZE
;
3619 opt
= VM_EXIT_SAVE_IA32_PAT
| VM_EXIT_LOAD_IA32_PAT
|
3620 VM_EXIT_CLEAR_BNDCFGS
;
3621 if (adjust_vmx_controls(min
, opt
, MSR_IA32_VMX_EXIT_CTLS
,
3622 &_vmexit_control
) < 0)
3625 min
= PIN_BASED_EXT_INTR_MASK
| PIN_BASED_NMI_EXITING
|
3626 PIN_BASED_VIRTUAL_NMIS
;
3627 opt
= PIN_BASED_POSTED_INTR
| PIN_BASED_VMX_PREEMPTION_TIMER
;
3628 if (adjust_vmx_controls(min
, opt
, MSR_IA32_VMX_PINBASED_CTLS
,
3629 &_pin_based_exec_control
) < 0)
3632 if (cpu_has_broken_vmx_preemption_timer())
3633 _pin_based_exec_control
&= ~PIN_BASED_VMX_PREEMPTION_TIMER
;
3634 if (!(_cpu_based_2nd_exec_control
&
3635 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY
))
3636 _pin_based_exec_control
&= ~PIN_BASED_POSTED_INTR
;
3638 min
= VM_ENTRY_LOAD_DEBUG_CONTROLS
;
3639 opt
= VM_ENTRY_LOAD_IA32_PAT
| VM_ENTRY_LOAD_BNDCFGS
;
3640 if (adjust_vmx_controls(min
, opt
, MSR_IA32_VMX_ENTRY_CTLS
,
3641 &_vmentry_control
) < 0)
3644 rdmsr(MSR_IA32_VMX_BASIC
, vmx_msr_low
, vmx_msr_high
);
3646 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
3647 if ((vmx_msr_high
& 0x1fff) > PAGE_SIZE
)
3650 #ifdef CONFIG_X86_64
3651 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
3652 if (vmx_msr_high
& (1u<<16))
3656 /* Require Write-Back (WB) memory type for VMCS accesses. */
3657 if (((vmx_msr_high
>> 18) & 15) != 6)
3660 vmcs_conf
->size
= vmx_msr_high
& 0x1fff;
3661 vmcs_conf
->order
= get_order(vmcs_conf
->size
);
3662 vmcs_conf
->basic_cap
= vmx_msr_high
& ~0x1fff;
3663 vmcs_conf
->revision_id
= vmx_msr_low
;
3665 vmcs_conf
->pin_based_exec_ctrl
= _pin_based_exec_control
;
3666 vmcs_conf
->cpu_based_exec_ctrl
= _cpu_based_exec_control
;
3667 vmcs_conf
->cpu_based_2nd_exec_ctrl
= _cpu_based_2nd_exec_control
;
3668 vmcs_conf
->vmexit_ctrl
= _vmexit_control
;
3669 vmcs_conf
->vmentry_ctrl
= _vmentry_control
;
3671 cpu_has_load_ia32_efer
=
3672 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS
,
3673 VM_ENTRY_LOAD_IA32_EFER
)
3674 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS
,
3675 VM_EXIT_LOAD_IA32_EFER
);
3677 cpu_has_load_perf_global_ctrl
=
3678 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS
,
3679 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL
)
3680 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS
,
3681 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL
);
3684 * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
3685 * but due to errata below it can't be used. Workaround is to use
3686 * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
3688 * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
3693 * BC86,AAY89,BD102 (model 44)
3697 if (cpu_has_load_perf_global_ctrl
&& boot_cpu_data
.x86
== 0x6) {
3698 switch (boot_cpu_data
.x86_model
) {
3704 cpu_has_load_perf_global_ctrl
= false;
3705 printk_once(KERN_WARNING
"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
3706 "does not work properly. Using workaround\n");
3713 if (boot_cpu_has(X86_FEATURE_XSAVES
))
3714 rdmsrl(MSR_IA32_XSS
, host_xss
);
3719 static struct vmcs
*alloc_vmcs_cpu(int cpu
)
3721 int node
= cpu_to_node(cpu
);
3725 pages
= __alloc_pages_node(node
, GFP_KERNEL
, vmcs_config
.order
);
3728 vmcs
= page_address(pages
);
3729 memset(vmcs
, 0, vmcs_config
.size
);
3730 vmcs
->revision_id
= vmcs_config
.revision_id
; /* vmcs revision id */
3734 static struct vmcs
*alloc_vmcs(void)
3736 return alloc_vmcs_cpu(raw_smp_processor_id());
3739 static void free_vmcs(struct vmcs
*vmcs
)
3741 free_pages((unsigned long)vmcs
, vmcs_config
.order
);
3745 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
3747 static void free_loaded_vmcs(struct loaded_vmcs
*loaded_vmcs
)
3749 if (!loaded_vmcs
->vmcs
)
3751 loaded_vmcs_clear(loaded_vmcs
);
3752 free_vmcs(loaded_vmcs
->vmcs
);
3753 loaded_vmcs
->vmcs
= NULL
;
3754 WARN_ON(loaded_vmcs
->shadow_vmcs
!= NULL
);
3757 static void free_kvm_area(void)
3761 for_each_possible_cpu(cpu
) {
3762 free_vmcs(per_cpu(vmxarea
, cpu
));
3763 per_cpu(vmxarea
, cpu
) = NULL
;
3767 enum vmcs_field_type
{
3768 VMCS_FIELD_TYPE_U16
= 0,
3769 VMCS_FIELD_TYPE_U64
= 1,
3770 VMCS_FIELD_TYPE_U32
= 2,
3771 VMCS_FIELD_TYPE_NATURAL_WIDTH
= 3
3774 static inline int vmcs_field_type(unsigned long field
)
3776 if (0x1 & field
) /* the *_HIGH fields are all 32 bit */
3777 return VMCS_FIELD_TYPE_U32
;
3778 return (field
>> 13) & 0x3 ;
3781 static inline int vmcs_field_readonly(unsigned long field
)
3783 return (((field
>> 10) & 0x3) == 1);
3786 static void init_vmcs_shadow_fields(void)
3790 /* No checks for read only fields yet */
3792 for (i
= j
= 0; i
< max_shadow_read_write_fields
; i
++) {
3793 switch (shadow_read_write_fields
[i
]) {
3795 if (!kvm_mpx_supported())
3803 shadow_read_write_fields
[j
] =
3804 shadow_read_write_fields
[i
];
3807 max_shadow_read_write_fields
= j
;
3809 /* shadowed fields guest access without vmexit */
3810 for (i
= 0; i
< max_shadow_read_write_fields
; i
++) {
3811 unsigned long field
= shadow_read_write_fields
[i
];
3813 clear_bit(field
, vmx_vmwrite_bitmap
);
3814 clear_bit(field
, vmx_vmread_bitmap
);
3815 if (vmcs_field_type(field
) == VMCS_FIELD_TYPE_U64
) {
3816 clear_bit(field
+ 1, vmx_vmwrite_bitmap
);
3817 clear_bit(field
+ 1, vmx_vmread_bitmap
);
3820 for (i
= 0; i
< max_shadow_read_only_fields
; i
++) {
3821 unsigned long field
= shadow_read_only_fields
[i
];
3823 clear_bit(field
, vmx_vmread_bitmap
);
3824 if (vmcs_field_type(field
) == VMCS_FIELD_TYPE_U64
)
3825 clear_bit(field
+ 1, vmx_vmread_bitmap
);
3829 static __init
int alloc_kvm_area(void)
3833 for_each_possible_cpu(cpu
) {
3836 vmcs
= alloc_vmcs_cpu(cpu
);
3842 per_cpu(vmxarea
, cpu
) = vmcs
;
3847 static bool emulation_required(struct kvm_vcpu
*vcpu
)
3849 return emulate_invalid_guest_state
&& !guest_state_valid(vcpu
);
3852 static void fix_pmode_seg(struct kvm_vcpu
*vcpu
, int seg
,
3853 struct kvm_segment
*save
)
3855 if (!emulate_invalid_guest_state
) {
3857 * CS and SS RPL should be equal during guest entry according
3858 * to VMX spec, but in reality it is not always so. Since vcpu
3859 * is in the middle of the transition from real mode to
3860 * protected mode it is safe to assume that RPL 0 is a good
3863 if (seg
== VCPU_SREG_CS
|| seg
== VCPU_SREG_SS
)
3864 save
->selector
&= ~SEGMENT_RPL_MASK
;
3865 save
->dpl
= save
->selector
& SEGMENT_RPL_MASK
;
3868 vmx_set_segment(vcpu
, save
, seg
);
3871 static void enter_pmode(struct kvm_vcpu
*vcpu
)
3873 unsigned long flags
;
3874 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
3877 * Update real mode segment cache. It may be not up-to-date if sement
3878 * register was written while vcpu was in a guest mode.
3880 vmx_get_segment(vcpu
, &vmx
->rmode
.segs
[VCPU_SREG_ES
], VCPU_SREG_ES
);
3881 vmx_get_segment(vcpu
, &vmx
->rmode
.segs
[VCPU_SREG_DS
], VCPU_SREG_DS
);
3882 vmx_get_segment(vcpu
, &vmx
->rmode
.segs
[VCPU_SREG_FS
], VCPU_SREG_FS
);
3883 vmx_get_segment(vcpu
, &vmx
->rmode
.segs
[VCPU_SREG_GS
], VCPU_SREG_GS
);
3884 vmx_get_segment(vcpu
, &vmx
->rmode
.segs
[VCPU_SREG_SS
], VCPU_SREG_SS
);
3885 vmx_get_segment(vcpu
, &vmx
->rmode
.segs
[VCPU_SREG_CS
], VCPU_SREG_CS
);
3887 vmx
->rmode
.vm86_active
= 0;
3889 vmx_segment_cache_clear(vmx
);
3891 vmx_set_segment(vcpu
, &vmx
->rmode
.segs
[VCPU_SREG_TR
], VCPU_SREG_TR
);
3893 flags
= vmcs_readl(GUEST_RFLAGS
);
3894 flags
&= RMODE_GUEST_OWNED_EFLAGS_BITS
;
3895 flags
|= vmx
->rmode
.save_rflags
& ~RMODE_GUEST_OWNED_EFLAGS_BITS
;
3896 vmcs_writel(GUEST_RFLAGS
, flags
);
3898 vmcs_writel(GUEST_CR4
, (vmcs_readl(GUEST_CR4
) & ~X86_CR4_VME
) |
3899 (vmcs_readl(CR4_READ_SHADOW
) & X86_CR4_VME
));
3901 update_exception_bitmap(vcpu
);
3903 fix_pmode_seg(vcpu
, VCPU_SREG_CS
, &vmx
->rmode
.segs
[VCPU_SREG_CS
]);
3904 fix_pmode_seg(vcpu
, VCPU_SREG_SS
, &vmx
->rmode
.segs
[VCPU_SREG_SS
]);
3905 fix_pmode_seg(vcpu
, VCPU_SREG_ES
, &vmx
->rmode
.segs
[VCPU_SREG_ES
]);
3906 fix_pmode_seg(vcpu
, VCPU_SREG_DS
, &vmx
->rmode
.segs
[VCPU_SREG_DS
]);
3907 fix_pmode_seg(vcpu
, VCPU_SREG_FS
, &vmx
->rmode
.segs
[VCPU_SREG_FS
]);
3908 fix_pmode_seg(vcpu
, VCPU_SREG_GS
, &vmx
->rmode
.segs
[VCPU_SREG_GS
]);
3911 static void fix_rmode_seg(int seg
, struct kvm_segment
*save
)
3913 const struct kvm_vmx_segment_field
*sf
= &kvm_vmx_segment_fields
[seg
];
3914 struct kvm_segment var
= *save
;
3917 if (seg
== VCPU_SREG_CS
)
3920 if (!emulate_invalid_guest_state
) {
3921 var
.selector
= var
.base
>> 4;
3922 var
.base
= var
.base
& 0xffff0;
3932 if (save
->base
& 0xf)
3933 printk_once(KERN_WARNING
"kvm: segment base is not "
3934 "paragraph aligned when entering "
3935 "protected mode (seg=%d)", seg
);
3938 vmcs_write16(sf
->selector
, var
.selector
);
3939 vmcs_writel(sf
->base
, var
.base
);
3940 vmcs_write32(sf
->limit
, var
.limit
);
3941 vmcs_write32(sf
->ar_bytes
, vmx_segment_access_rights(&var
));
3944 static void enter_rmode(struct kvm_vcpu
*vcpu
)
3946 unsigned long flags
;
3947 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
3949 vmx_get_segment(vcpu
, &vmx
->rmode
.segs
[VCPU_SREG_TR
], VCPU_SREG_TR
);
3950 vmx_get_segment(vcpu
, &vmx
->rmode
.segs
[VCPU_SREG_ES
], VCPU_SREG_ES
);
3951 vmx_get_segment(vcpu
, &vmx
->rmode
.segs
[VCPU_SREG_DS
], VCPU_SREG_DS
);
3952 vmx_get_segment(vcpu
, &vmx
->rmode
.segs
[VCPU_SREG_FS
], VCPU_SREG_FS
);
3953 vmx_get_segment(vcpu
, &vmx
->rmode
.segs
[VCPU_SREG_GS
], VCPU_SREG_GS
);
3954 vmx_get_segment(vcpu
, &vmx
->rmode
.segs
[VCPU_SREG_SS
], VCPU_SREG_SS
);
3955 vmx_get_segment(vcpu
, &vmx
->rmode
.segs
[VCPU_SREG_CS
], VCPU_SREG_CS
);
3957 vmx
->rmode
.vm86_active
= 1;
3960 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
3961 * vcpu. Warn the user that an update is overdue.
3963 if (!vcpu
->kvm
->arch
.tss_addr
)
3964 printk_once(KERN_WARNING
"kvm: KVM_SET_TSS_ADDR need to be "
3965 "called before entering vcpu\n");
3967 vmx_segment_cache_clear(vmx
);
3969 vmcs_writel(GUEST_TR_BASE
, vcpu
->kvm
->arch
.tss_addr
);
3970 vmcs_write32(GUEST_TR_LIMIT
, RMODE_TSS_SIZE
- 1);
3971 vmcs_write32(GUEST_TR_AR_BYTES
, 0x008b);
3973 flags
= vmcs_readl(GUEST_RFLAGS
);
3974 vmx
->rmode
.save_rflags
= flags
;
3976 flags
|= X86_EFLAGS_IOPL
| X86_EFLAGS_VM
;
3978 vmcs_writel(GUEST_RFLAGS
, flags
);
3979 vmcs_writel(GUEST_CR4
, vmcs_readl(GUEST_CR4
) | X86_CR4_VME
);
3980 update_exception_bitmap(vcpu
);
3982 fix_rmode_seg(VCPU_SREG_SS
, &vmx
->rmode
.segs
[VCPU_SREG_SS
]);
3983 fix_rmode_seg(VCPU_SREG_CS
, &vmx
->rmode
.segs
[VCPU_SREG_CS
]);
3984 fix_rmode_seg(VCPU_SREG_ES
, &vmx
->rmode
.segs
[VCPU_SREG_ES
]);
3985 fix_rmode_seg(VCPU_SREG_DS
, &vmx
->rmode
.segs
[VCPU_SREG_DS
]);
3986 fix_rmode_seg(VCPU_SREG_GS
, &vmx
->rmode
.segs
[VCPU_SREG_GS
]);
3987 fix_rmode_seg(VCPU_SREG_FS
, &vmx
->rmode
.segs
[VCPU_SREG_FS
]);
3989 kvm_mmu_reset_context(vcpu
);
3992 static void vmx_set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
3994 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
3995 struct shared_msr_entry
*msr
= find_msr_entry(vmx
, MSR_EFER
);
4001 * Force kernel_gs_base reloading before EFER changes, as control
4002 * of this msr depends on is_long_mode().
4004 vmx_load_host_state(to_vmx(vcpu
));
4005 vcpu
->arch
.efer
= efer
;
4006 if (efer
& EFER_LMA
) {
4007 vm_entry_controls_setbit(to_vmx(vcpu
), VM_ENTRY_IA32E_MODE
);
4010 vm_entry_controls_clearbit(to_vmx(vcpu
), VM_ENTRY_IA32E_MODE
);
4012 msr
->data
= efer
& ~EFER_LME
;
4017 #ifdef CONFIG_X86_64
4019 static void enter_lmode(struct kvm_vcpu
*vcpu
)
4023 vmx_segment_cache_clear(to_vmx(vcpu
));
4025 guest_tr_ar
= vmcs_read32(GUEST_TR_AR_BYTES
);
4026 if ((guest_tr_ar
& VMX_AR_TYPE_MASK
) != VMX_AR_TYPE_BUSY_64_TSS
) {
4027 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
4029 vmcs_write32(GUEST_TR_AR_BYTES
,
4030 (guest_tr_ar
& ~VMX_AR_TYPE_MASK
)
4031 | VMX_AR_TYPE_BUSY_64_TSS
);
4033 vmx_set_efer(vcpu
, vcpu
->arch
.efer
| EFER_LMA
);
4036 static void exit_lmode(struct kvm_vcpu
*vcpu
)
4038 vm_entry_controls_clearbit(to_vmx(vcpu
), VM_ENTRY_IA32E_MODE
);
4039 vmx_set_efer(vcpu
, vcpu
->arch
.efer
& ~EFER_LMA
);
4044 static inline void __vmx_flush_tlb(struct kvm_vcpu
*vcpu
, int vpid
)
4047 if (!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
))
4049 ept_sync_context(construct_eptp(vcpu
, vcpu
->arch
.mmu
.root_hpa
));
4051 vpid_sync_context(vpid
);
4055 static void vmx_flush_tlb(struct kvm_vcpu
*vcpu
)
4057 __vmx_flush_tlb(vcpu
, to_vmx(vcpu
)->vpid
);
4060 static void vmx_flush_tlb_ept_only(struct kvm_vcpu
*vcpu
)
4063 vmx_flush_tlb(vcpu
);
4066 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu
*vcpu
)
4068 ulong cr0_guest_owned_bits
= vcpu
->arch
.cr0_guest_owned_bits
;
4070 vcpu
->arch
.cr0
&= ~cr0_guest_owned_bits
;
4071 vcpu
->arch
.cr0
|= vmcs_readl(GUEST_CR0
) & cr0_guest_owned_bits
;
4074 static void vmx_decache_cr3(struct kvm_vcpu
*vcpu
)
4076 if (enable_ept
&& is_paging(vcpu
))
4077 vcpu
->arch
.cr3
= vmcs_readl(GUEST_CR3
);
4078 __set_bit(VCPU_EXREG_CR3
, (ulong
*)&vcpu
->arch
.regs_avail
);
4081 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu
*vcpu
)
4083 ulong cr4_guest_owned_bits
= vcpu
->arch
.cr4_guest_owned_bits
;
4085 vcpu
->arch
.cr4
&= ~cr4_guest_owned_bits
;
4086 vcpu
->arch
.cr4
|= vmcs_readl(GUEST_CR4
) & cr4_guest_owned_bits
;
4089 static void ept_load_pdptrs(struct kvm_vcpu
*vcpu
)
4091 struct kvm_mmu
*mmu
= vcpu
->arch
.walk_mmu
;
4093 if (!test_bit(VCPU_EXREG_PDPTR
,
4094 (unsigned long *)&vcpu
->arch
.regs_dirty
))
4097 if (is_paging(vcpu
) && is_pae(vcpu
) && !is_long_mode(vcpu
)) {
4098 vmcs_write64(GUEST_PDPTR0
, mmu
->pdptrs
[0]);
4099 vmcs_write64(GUEST_PDPTR1
, mmu
->pdptrs
[1]);
4100 vmcs_write64(GUEST_PDPTR2
, mmu
->pdptrs
[2]);
4101 vmcs_write64(GUEST_PDPTR3
, mmu
->pdptrs
[3]);
4105 static void ept_save_pdptrs(struct kvm_vcpu
*vcpu
)
4107 struct kvm_mmu
*mmu
= vcpu
->arch
.walk_mmu
;
4109 if (is_paging(vcpu
) && is_pae(vcpu
) && !is_long_mode(vcpu
)) {
4110 mmu
->pdptrs
[0] = vmcs_read64(GUEST_PDPTR0
);
4111 mmu
->pdptrs
[1] = vmcs_read64(GUEST_PDPTR1
);
4112 mmu
->pdptrs
[2] = vmcs_read64(GUEST_PDPTR2
);
4113 mmu
->pdptrs
[3] = vmcs_read64(GUEST_PDPTR3
);
4116 __set_bit(VCPU_EXREG_PDPTR
,
4117 (unsigned long *)&vcpu
->arch
.regs_avail
);
4118 __set_bit(VCPU_EXREG_PDPTR
,
4119 (unsigned long *)&vcpu
->arch
.regs_dirty
);
4122 static bool nested_guest_cr0_valid(struct kvm_vcpu
*vcpu
, unsigned long val
)
4124 u64 fixed0
= to_vmx(vcpu
)->nested
.nested_vmx_cr0_fixed0
;
4125 u64 fixed1
= to_vmx(vcpu
)->nested
.nested_vmx_cr0_fixed1
;
4126 struct vmcs12
*vmcs12
= get_vmcs12(vcpu
);
4128 if (to_vmx(vcpu
)->nested
.nested_vmx_secondary_ctls_high
&
4129 SECONDARY_EXEC_UNRESTRICTED_GUEST
&&
4130 nested_cpu_has2(vmcs12
, SECONDARY_EXEC_UNRESTRICTED_GUEST
))
4131 fixed0
&= ~(X86_CR0_PE
| X86_CR0_PG
);
4133 return fixed_bits_valid(val
, fixed0
, fixed1
);
4136 static bool nested_host_cr0_valid(struct kvm_vcpu
*vcpu
, unsigned long val
)
4138 u64 fixed0
= to_vmx(vcpu
)->nested
.nested_vmx_cr0_fixed0
;
4139 u64 fixed1
= to_vmx(vcpu
)->nested
.nested_vmx_cr0_fixed1
;
4141 return fixed_bits_valid(val
, fixed0
, fixed1
);
4144 static bool nested_cr4_valid(struct kvm_vcpu
*vcpu
, unsigned long val
)
4146 u64 fixed0
= to_vmx(vcpu
)->nested
.nested_vmx_cr4_fixed0
;
4147 u64 fixed1
= to_vmx(vcpu
)->nested
.nested_vmx_cr4_fixed1
;
4149 return fixed_bits_valid(val
, fixed0
, fixed1
);
4152 /* No difference in the restrictions on guest and host CR4 in VMX operation. */
4153 #define nested_guest_cr4_valid nested_cr4_valid
4154 #define nested_host_cr4_valid nested_cr4_valid
4156 static int vmx_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
);
4158 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0
,
4160 struct kvm_vcpu
*vcpu
)
4162 if (!test_bit(VCPU_EXREG_CR3
, (ulong
*)&vcpu
->arch
.regs_avail
))
4163 vmx_decache_cr3(vcpu
);
4164 if (!(cr0
& X86_CR0_PG
)) {
4165 /* From paging/starting to nonpaging */
4166 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL
,
4167 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL
) |
4168 (CPU_BASED_CR3_LOAD_EXITING
|
4169 CPU_BASED_CR3_STORE_EXITING
));
4170 vcpu
->arch
.cr0
= cr0
;
4171 vmx_set_cr4(vcpu
, kvm_read_cr4(vcpu
));
4172 } else if (!is_paging(vcpu
)) {
4173 /* From nonpaging to paging */
4174 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL
,
4175 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL
) &
4176 ~(CPU_BASED_CR3_LOAD_EXITING
|
4177 CPU_BASED_CR3_STORE_EXITING
));
4178 vcpu
->arch
.cr0
= cr0
;
4179 vmx_set_cr4(vcpu
, kvm_read_cr4(vcpu
));
4182 if (!(cr0
& X86_CR0_WP
))
4183 *hw_cr0
&= ~X86_CR0_WP
;
4186 static void vmx_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
4188 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
4189 unsigned long hw_cr0
;
4191 hw_cr0
= (cr0
& ~KVM_GUEST_CR0_MASK
);
4192 if (enable_unrestricted_guest
)
4193 hw_cr0
|= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST
;
4195 hw_cr0
|= KVM_VM_CR0_ALWAYS_ON
;
4197 if (vmx
->rmode
.vm86_active
&& (cr0
& X86_CR0_PE
))
4200 if (!vmx
->rmode
.vm86_active
&& !(cr0
& X86_CR0_PE
))
4204 #ifdef CONFIG_X86_64
4205 if (vcpu
->arch
.efer
& EFER_LME
) {
4206 if (!is_paging(vcpu
) && (cr0
& X86_CR0_PG
))
4208 if (is_paging(vcpu
) && !(cr0
& X86_CR0_PG
))
4214 ept_update_paging_mode_cr0(&hw_cr0
, cr0
, vcpu
);
4216 vmcs_writel(CR0_READ_SHADOW
, cr0
);
4217 vmcs_writel(GUEST_CR0
, hw_cr0
);
4218 vcpu
->arch
.cr0
= cr0
;
4220 /* depends on vcpu->arch.cr0 to be set to a new value */
4221 vmx
->emulation_required
= emulation_required(vcpu
);
4224 static u64
construct_eptp(struct kvm_vcpu
*vcpu
, unsigned long root_hpa
)
4228 /* TODO write the value reading from MSR */
4229 eptp
= VMX_EPT_DEFAULT_MT
|
4230 VMX_EPT_DEFAULT_GAW
<< VMX_EPT_GAW_EPTP_SHIFT
;
4231 if (enable_ept_ad_bits
&&
4232 (!is_guest_mode(vcpu
) || nested_ept_ad_enabled(vcpu
)))
4233 eptp
|= VMX_EPT_AD_ENABLE_BIT
;
4234 eptp
|= (root_hpa
& PAGE_MASK
);
4239 static void vmx_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
4241 unsigned long guest_cr3
;
4246 eptp
= construct_eptp(vcpu
, cr3
);
4247 vmcs_write64(EPT_POINTER
, eptp
);
4248 if (is_paging(vcpu
) || is_guest_mode(vcpu
))
4249 guest_cr3
= kvm_read_cr3(vcpu
);
4251 guest_cr3
= vcpu
->kvm
->arch
.ept_identity_map_addr
;
4252 ept_load_pdptrs(vcpu
);
4255 vmx_flush_tlb(vcpu
);
4256 vmcs_writel(GUEST_CR3
, guest_cr3
);
4259 static int vmx_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
4262 * Pass through host's Machine Check Enable value to hw_cr4, which
4263 * is in force while we are in guest mode. Do not let guests control
4264 * this bit, even if host CR4.MCE == 0.
4266 unsigned long hw_cr4
=
4267 (cr4_read_shadow() & X86_CR4_MCE
) |
4268 (cr4
& ~X86_CR4_MCE
) |
4269 (to_vmx(vcpu
)->rmode
.vm86_active
?
4270 KVM_RMODE_VM_CR4_ALWAYS_ON
: KVM_PMODE_VM_CR4_ALWAYS_ON
);
4272 if (cr4
& X86_CR4_VMXE
) {
4274 * To use VMXON (and later other VMX instructions), a guest
4275 * must first be able to turn on cr4.VMXE (see handle_vmon()).
4276 * So basically the check on whether to allow nested VMX
4279 if (!nested_vmx_allowed(vcpu
))
4283 if (to_vmx(vcpu
)->nested
.vmxon
&& !nested_cr4_valid(vcpu
, cr4
))
4286 vcpu
->arch
.cr4
= cr4
;
4288 if (!is_paging(vcpu
)) {
4289 hw_cr4
&= ~X86_CR4_PAE
;
4290 hw_cr4
|= X86_CR4_PSE
;
4291 } else if (!(cr4
& X86_CR4_PAE
)) {
4292 hw_cr4
&= ~X86_CR4_PAE
;
4296 if (!enable_unrestricted_guest
&& !is_paging(vcpu
))
4298 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
4299 * hardware. To emulate this behavior, SMEP/SMAP/PKU needs
4300 * to be manually disabled when guest switches to non-paging
4303 * If !enable_unrestricted_guest, the CPU is always running
4304 * with CR0.PG=1 and CR4 needs to be modified.
4305 * If enable_unrestricted_guest, the CPU automatically
4306 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
4308 hw_cr4
&= ~(X86_CR4_SMEP
| X86_CR4_SMAP
| X86_CR4_PKE
);
4310 vmcs_writel(CR4_READ_SHADOW
, cr4
);
4311 vmcs_writel(GUEST_CR4
, hw_cr4
);
4315 static void vmx_get_segment(struct kvm_vcpu
*vcpu
,
4316 struct kvm_segment
*var
, int seg
)
4318 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
4321 if (vmx
->rmode
.vm86_active
&& seg
!= VCPU_SREG_LDTR
) {
4322 *var
= vmx
->rmode
.segs
[seg
];
4323 if (seg
== VCPU_SREG_TR
4324 || var
->selector
== vmx_read_guest_seg_selector(vmx
, seg
))
4326 var
->base
= vmx_read_guest_seg_base(vmx
, seg
);
4327 var
->selector
= vmx_read_guest_seg_selector(vmx
, seg
);
4330 var
->base
= vmx_read_guest_seg_base(vmx
, seg
);
4331 var
->limit
= vmx_read_guest_seg_limit(vmx
, seg
);
4332 var
->selector
= vmx_read_guest_seg_selector(vmx
, seg
);
4333 ar
= vmx_read_guest_seg_ar(vmx
, seg
);
4334 var
->unusable
= (ar
>> 16) & 1;
4335 var
->type
= ar
& 15;
4336 var
->s
= (ar
>> 4) & 1;
4337 var
->dpl
= (ar
>> 5) & 3;
4339 * Some userspaces do not preserve unusable property. Since usable
4340 * segment has to be present according to VMX spec we can use present
4341 * property to amend userspace bug by making unusable segment always
4342 * nonpresent. vmx_segment_access_rights() already marks nonpresent
4343 * segment as unusable.
4345 var
->present
= !var
->unusable
;
4346 var
->avl
= (ar
>> 12) & 1;
4347 var
->l
= (ar
>> 13) & 1;
4348 var
->db
= (ar
>> 14) & 1;
4349 var
->g
= (ar
>> 15) & 1;
4352 static u64
vmx_get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
4354 struct kvm_segment s
;
4356 if (to_vmx(vcpu
)->rmode
.vm86_active
) {
4357 vmx_get_segment(vcpu
, &s
, seg
);
4360 return vmx_read_guest_seg_base(to_vmx(vcpu
), seg
);
4363 static int vmx_get_cpl(struct kvm_vcpu
*vcpu
)
4365 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
4367 if (unlikely(vmx
->rmode
.vm86_active
))
4370 int ar
= vmx_read_guest_seg_ar(vmx
, VCPU_SREG_SS
);
4371 return VMX_AR_DPL(ar
);
4375 static u32
vmx_segment_access_rights(struct kvm_segment
*var
)
4379 if (var
->unusable
|| !var
->present
)
4382 ar
= var
->type
& 15;
4383 ar
|= (var
->s
& 1) << 4;
4384 ar
|= (var
->dpl
& 3) << 5;
4385 ar
|= (var
->present
& 1) << 7;
4386 ar
|= (var
->avl
& 1) << 12;
4387 ar
|= (var
->l
& 1) << 13;
4388 ar
|= (var
->db
& 1) << 14;
4389 ar
|= (var
->g
& 1) << 15;
4395 static void vmx_set_segment(struct kvm_vcpu
*vcpu
,
4396 struct kvm_segment
*var
, int seg
)
4398 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
4399 const struct kvm_vmx_segment_field
*sf
= &kvm_vmx_segment_fields
[seg
];
4401 vmx_segment_cache_clear(vmx
);
4403 if (vmx
->rmode
.vm86_active
&& seg
!= VCPU_SREG_LDTR
) {
4404 vmx
->rmode
.segs
[seg
] = *var
;
4405 if (seg
== VCPU_SREG_TR
)
4406 vmcs_write16(sf
->selector
, var
->selector
);
4408 fix_rmode_seg(seg
, &vmx
->rmode
.segs
[seg
]);
4412 vmcs_writel(sf
->base
, var
->base
);
4413 vmcs_write32(sf
->limit
, var
->limit
);
4414 vmcs_write16(sf
->selector
, var
->selector
);
4417 * Fix the "Accessed" bit in AR field of segment registers for older
4419 * IA32 arch specifies that at the time of processor reset the
4420 * "Accessed" bit in the AR field of segment registers is 1. And qemu
4421 * is setting it to 0 in the userland code. This causes invalid guest
4422 * state vmexit when "unrestricted guest" mode is turned on.
4423 * Fix for this setup issue in cpu_reset is being pushed in the qemu
4424 * tree. Newer qemu binaries with that qemu fix would not need this
4427 if (enable_unrestricted_guest
&& (seg
!= VCPU_SREG_LDTR
))
4428 var
->type
|= 0x1; /* Accessed */
4430 vmcs_write32(sf
->ar_bytes
, vmx_segment_access_rights(var
));
4433 vmx
->emulation_required
= emulation_required(vcpu
);
4436 static void vmx_get_cs_db_l_bits(struct kvm_vcpu
*vcpu
, int *db
, int *l
)
4438 u32 ar
= vmx_read_guest_seg_ar(to_vmx(vcpu
), VCPU_SREG_CS
);
4440 *db
= (ar
>> 14) & 1;
4441 *l
= (ar
>> 13) & 1;
4444 static void vmx_get_idt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
4446 dt
->size
= vmcs_read32(GUEST_IDTR_LIMIT
);
4447 dt
->address
= vmcs_readl(GUEST_IDTR_BASE
);
4450 static void vmx_set_idt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
4452 vmcs_write32(GUEST_IDTR_LIMIT
, dt
->size
);
4453 vmcs_writel(GUEST_IDTR_BASE
, dt
->address
);
4456 static void vmx_get_gdt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
4458 dt
->size
= vmcs_read32(GUEST_GDTR_LIMIT
);
4459 dt
->address
= vmcs_readl(GUEST_GDTR_BASE
);
4462 static void vmx_set_gdt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
4464 vmcs_write32(GUEST_GDTR_LIMIT
, dt
->size
);
4465 vmcs_writel(GUEST_GDTR_BASE
, dt
->address
);
4468 static bool rmode_segment_valid(struct kvm_vcpu
*vcpu
, int seg
)
4470 struct kvm_segment var
;
4473 vmx_get_segment(vcpu
, &var
, seg
);
4475 if (seg
== VCPU_SREG_CS
)
4477 ar
= vmx_segment_access_rights(&var
);
4479 if (var
.base
!= (var
.selector
<< 4))
4481 if (var
.limit
!= 0xffff)
4489 static bool code_segment_valid(struct kvm_vcpu
*vcpu
)
4491 struct kvm_segment cs
;
4492 unsigned int cs_rpl
;
4494 vmx_get_segment(vcpu
, &cs
, VCPU_SREG_CS
);
4495 cs_rpl
= cs
.selector
& SEGMENT_RPL_MASK
;
4499 if (~cs
.type
& (VMX_AR_TYPE_CODE_MASK
|VMX_AR_TYPE_ACCESSES_MASK
))
4503 if (cs
.type
& VMX_AR_TYPE_WRITEABLE_MASK
) {
4504 if (cs
.dpl
> cs_rpl
)
4507 if (cs
.dpl
!= cs_rpl
)
4513 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
4517 static bool stack_segment_valid(struct kvm_vcpu
*vcpu
)
4519 struct kvm_segment ss
;
4520 unsigned int ss_rpl
;
4522 vmx_get_segment(vcpu
, &ss
, VCPU_SREG_SS
);
4523 ss_rpl
= ss
.selector
& SEGMENT_RPL_MASK
;
4527 if (ss
.type
!= 3 && ss
.type
!= 7)
4531 if (ss
.dpl
!= ss_rpl
) /* DPL != RPL */
4539 static bool data_segment_valid(struct kvm_vcpu
*vcpu
, int seg
)
4541 struct kvm_segment var
;
4544 vmx_get_segment(vcpu
, &var
, seg
);
4545 rpl
= var
.selector
& SEGMENT_RPL_MASK
;
4553 if (~var
.type
& (VMX_AR_TYPE_CODE_MASK
|VMX_AR_TYPE_WRITEABLE_MASK
)) {
4554 if (var
.dpl
< rpl
) /* DPL < RPL */
4558 /* TODO: Add other members to kvm_segment_field to allow checking for other access
4564 static bool tr_valid(struct kvm_vcpu
*vcpu
)
4566 struct kvm_segment tr
;
4568 vmx_get_segment(vcpu
, &tr
, VCPU_SREG_TR
);
4572 if (tr
.selector
& SEGMENT_TI_MASK
) /* TI = 1 */
4574 if (tr
.type
!= 3 && tr
.type
!= 11) /* TODO: Check if guest is in IA32e mode */
4582 static bool ldtr_valid(struct kvm_vcpu
*vcpu
)
4584 struct kvm_segment ldtr
;
4586 vmx_get_segment(vcpu
, &ldtr
, VCPU_SREG_LDTR
);
4590 if (ldtr
.selector
& SEGMENT_TI_MASK
) /* TI = 1 */
4600 static bool cs_ss_rpl_check(struct kvm_vcpu
*vcpu
)
4602 struct kvm_segment cs
, ss
;
4604 vmx_get_segment(vcpu
, &cs
, VCPU_SREG_CS
);
4605 vmx_get_segment(vcpu
, &ss
, VCPU_SREG_SS
);
4607 return ((cs
.selector
& SEGMENT_RPL_MASK
) ==
4608 (ss
.selector
& SEGMENT_RPL_MASK
));
4612 * Check if guest state is valid. Returns true if valid, false if
4614 * We assume that registers are always usable
4616 static bool guest_state_valid(struct kvm_vcpu
*vcpu
)
4618 if (enable_unrestricted_guest
)
4621 /* real mode guest state checks */
4622 if (!is_protmode(vcpu
) || (vmx_get_rflags(vcpu
) & X86_EFLAGS_VM
)) {
4623 if (!rmode_segment_valid(vcpu
, VCPU_SREG_CS
))
4625 if (!rmode_segment_valid(vcpu
, VCPU_SREG_SS
))
4627 if (!rmode_segment_valid(vcpu
, VCPU_SREG_DS
))
4629 if (!rmode_segment_valid(vcpu
, VCPU_SREG_ES
))
4631 if (!rmode_segment_valid(vcpu
, VCPU_SREG_FS
))
4633 if (!rmode_segment_valid(vcpu
, VCPU_SREG_GS
))
4636 /* protected mode guest state checks */
4637 if (!cs_ss_rpl_check(vcpu
))
4639 if (!code_segment_valid(vcpu
))
4641 if (!stack_segment_valid(vcpu
))
4643 if (!data_segment_valid(vcpu
, VCPU_SREG_DS
))
4645 if (!data_segment_valid(vcpu
, VCPU_SREG_ES
))
4647 if (!data_segment_valid(vcpu
, VCPU_SREG_FS
))
4649 if (!data_segment_valid(vcpu
, VCPU_SREG_GS
))
4651 if (!tr_valid(vcpu
))
4653 if (!ldtr_valid(vcpu
))
4657 * - Add checks on RIP
4658 * - Add checks on RFLAGS
4664 static bool page_address_valid(struct kvm_vcpu
*vcpu
, gpa_t gpa
)
4666 return PAGE_ALIGNED(gpa
) && !(gpa
>> cpuid_maxphyaddr(vcpu
));
4669 static int init_rmode_tss(struct kvm
*kvm
)
4675 idx
= srcu_read_lock(&kvm
->srcu
);
4676 fn
= kvm
->arch
.tss_addr
>> PAGE_SHIFT
;
4677 r
= kvm_clear_guest_page(kvm
, fn
, 0, PAGE_SIZE
);
4680 data
= TSS_BASE_SIZE
+ TSS_REDIRECTION_SIZE
;
4681 r
= kvm_write_guest_page(kvm
, fn
++, &data
,
4682 TSS_IOPB_BASE_OFFSET
, sizeof(u16
));
4685 r
= kvm_clear_guest_page(kvm
, fn
++, 0, PAGE_SIZE
);
4688 r
= kvm_clear_guest_page(kvm
, fn
, 0, PAGE_SIZE
);
4692 r
= kvm_write_guest_page(kvm
, fn
, &data
,
4693 RMODE_TSS_SIZE
- 2 * PAGE_SIZE
- 1,
4696 srcu_read_unlock(&kvm
->srcu
, idx
);
4700 static int init_rmode_identity_map(struct kvm
*kvm
)
4703 kvm_pfn_t identity_map_pfn
;
4709 /* Protect kvm->arch.ept_identity_pagetable_done. */
4710 mutex_lock(&kvm
->slots_lock
);
4712 if (likely(kvm
->arch
.ept_identity_pagetable_done
))
4715 identity_map_pfn
= kvm
->arch
.ept_identity_map_addr
>> PAGE_SHIFT
;
4717 r
= alloc_identity_pagetable(kvm
);
4721 idx
= srcu_read_lock(&kvm
->srcu
);
4722 r
= kvm_clear_guest_page(kvm
, identity_map_pfn
, 0, PAGE_SIZE
);
4725 /* Set up identity-mapping pagetable for EPT in real mode */
4726 for (i
= 0; i
< PT32_ENT_PER_PAGE
; i
++) {
4727 tmp
= (i
<< 22) + (_PAGE_PRESENT
| _PAGE_RW
| _PAGE_USER
|
4728 _PAGE_ACCESSED
| _PAGE_DIRTY
| _PAGE_PSE
);
4729 r
= kvm_write_guest_page(kvm
, identity_map_pfn
,
4730 &tmp
, i
* sizeof(tmp
), sizeof(tmp
));
4734 kvm
->arch
.ept_identity_pagetable_done
= true;
4737 srcu_read_unlock(&kvm
->srcu
, idx
);
4740 mutex_unlock(&kvm
->slots_lock
);
4744 static void seg_setup(int seg
)
4746 const struct kvm_vmx_segment_field
*sf
= &kvm_vmx_segment_fields
[seg
];
4749 vmcs_write16(sf
->selector
, 0);
4750 vmcs_writel(sf
->base
, 0);
4751 vmcs_write32(sf
->limit
, 0xffff);
4753 if (seg
== VCPU_SREG_CS
)
4754 ar
|= 0x08; /* code segment */
4756 vmcs_write32(sf
->ar_bytes
, ar
);
4759 static int alloc_apic_access_page(struct kvm
*kvm
)
4764 mutex_lock(&kvm
->slots_lock
);
4765 if (kvm
->arch
.apic_access_page_done
)
4767 r
= __x86_set_memory_region(kvm
, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT
,
4768 APIC_DEFAULT_PHYS_BASE
, PAGE_SIZE
);
4772 page
= gfn_to_page(kvm
, APIC_DEFAULT_PHYS_BASE
>> PAGE_SHIFT
);
4773 if (is_error_page(page
)) {
4779 * Do not pin the page in memory, so that memory hot-unplug
4780 * is able to migrate it.
4783 kvm
->arch
.apic_access_page_done
= true;
4785 mutex_unlock(&kvm
->slots_lock
);
4789 static int alloc_identity_pagetable(struct kvm
*kvm
)
4791 /* Called with kvm->slots_lock held. */
4795 BUG_ON(kvm
->arch
.ept_identity_pagetable_done
);
4797 r
= __x86_set_memory_region(kvm
, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT
,
4798 kvm
->arch
.ept_identity_map_addr
, PAGE_SIZE
);
4803 static int allocate_vpid(void)
4809 spin_lock(&vmx_vpid_lock
);
4810 vpid
= find_first_zero_bit(vmx_vpid_bitmap
, VMX_NR_VPIDS
);
4811 if (vpid
< VMX_NR_VPIDS
)
4812 __set_bit(vpid
, vmx_vpid_bitmap
);
4815 spin_unlock(&vmx_vpid_lock
);
4819 static void free_vpid(int vpid
)
4821 if (!enable_vpid
|| vpid
== 0)
4823 spin_lock(&vmx_vpid_lock
);
4824 __clear_bit(vpid
, vmx_vpid_bitmap
);
4825 spin_unlock(&vmx_vpid_lock
);
4828 #define MSR_TYPE_R 1
4829 #define MSR_TYPE_W 2
4830 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap
,
4833 int f
= sizeof(unsigned long);
4835 if (!cpu_has_vmx_msr_bitmap())
4839 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4840 * have the write-low and read-high bitmap offsets the wrong way round.
4841 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4843 if (msr
<= 0x1fff) {
4844 if (type
& MSR_TYPE_R
)
4846 __clear_bit(msr
, msr_bitmap
+ 0x000 / f
);
4848 if (type
& MSR_TYPE_W
)
4850 __clear_bit(msr
, msr_bitmap
+ 0x800 / f
);
4852 } else if ((msr
>= 0xc0000000) && (msr
<= 0xc0001fff)) {
4854 if (type
& MSR_TYPE_R
)
4856 __clear_bit(msr
, msr_bitmap
+ 0x400 / f
);
4858 if (type
& MSR_TYPE_W
)
4860 __clear_bit(msr
, msr_bitmap
+ 0xc00 / f
);
4866 * If a msr is allowed by L0, we should check whether it is allowed by L1.
4867 * The corresponding bit will be cleared unless both of L0 and L1 allow it.
4869 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1
,
4870 unsigned long *msr_bitmap_nested
,
4873 int f
= sizeof(unsigned long);
4875 if (!cpu_has_vmx_msr_bitmap()) {
4881 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4882 * have the write-low and read-high bitmap offsets the wrong way round.
4883 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4885 if (msr
<= 0x1fff) {
4886 if (type
& MSR_TYPE_R
&&
4887 !test_bit(msr
, msr_bitmap_l1
+ 0x000 / f
))
4889 __clear_bit(msr
, msr_bitmap_nested
+ 0x000 / f
);
4891 if (type
& MSR_TYPE_W
&&
4892 !test_bit(msr
, msr_bitmap_l1
+ 0x800 / f
))
4894 __clear_bit(msr
, msr_bitmap_nested
+ 0x800 / f
);
4896 } else if ((msr
>= 0xc0000000) && (msr
<= 0xc0001fff)) {
4898 if (type
& MSR_TYPE_R
&&
4899 !test_bit(msr
, msr_bitmap_l1
+ 0x400 / f
))
4901 __clear_bit(msr
, msr_bitmap_nested
+ 0x400 / f
);
4903 if (type
& MSR_TYPE_W
&&
4904 !test_bit(msr
, msr_bitmap_l1
+ 0xc00 / f
))
4906 __clear_bit(msr
, msr_bitmap_nested
+ 0xc00 / f
);
4911 static void vmx_disable_intercept_for_msr(u32 msr
, bool longmode_only
)
4914 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy
,
4915 msr
, MSR_TYPE_R
| MSR_TYPE_W
);
4916 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode
,
4917 msr
, MSR_TYPE_R
| MSR_TYPE_W
);
4920 static void vmx_disable_intercept_msr_x2apic(u32 msr
, int type
, bool apicv_active
)
4923 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic_apicv
,
4925 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic_apicv
,
4928 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic
,
4930 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic
,
4935 static bool vmx_get_enable_apicv(void)
4937 return enable_apicv
;
4940 static void vmx_complete_nested_posted_interrupt(struct kvm_vcpu
*vcpu
)
4942 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
4947 if (vmx
->nested
.pi_desc
&&
4948 vmx
->nested
.pi_pending
) {
4949 vmx
->nested
.pi_pending
= false;
4950 if (!pi_test_and_clear_on(vmx
->nested
.pi_desc
))
4953 max_irr
= find_last_bit(
4954 (unsigned long *)vmx
->nested
.pi_desc
->pir
, 256);
4959 vapic_page
= kmap(vmx
->nested
.virtual_apic_page
);
4960 __kvm_apic_update_irr(vmx
->nested
.pi_desc
->pir
, vapic_page
);
4961 kunmap(vmx
->nested
.virtual_apic_page
);
4963 status
= vmcs_read16(GUEST_INTR_STATUS
);
4964 if ((u8
)max_irr
> ((u8
)status
& 0xff)) {
4966 status
|= (u8
)max_irr
;
4967 vmcs_write16(GUEST_INTR_STATUS
, status
);
4972 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu
*vcpu
)
4975 if (vcpu
->mode
== IN_GUEST_MODE
) {
4976 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
4979 * Currently, we don't support urgent interrupt,
4980 * all interrupts are recognized as non-urgent
4981 * interrupt, so we cannot post interrupts when
4984 * If the vcpu is in guest mode, it means it is
4985 * running instead of being scheduled out and
4986 * waiting in the run queue, and that's the only
4987 * case when 'SN' is set currently, warning if
4990 WARN_ON_ONCE(pi_test_sn(&vmx
->pi_desc
));
4992 apic
->send_IPI_mask(get_cpu_mask(vcpu
->cpu
),
4993 POSTED_INTR_VECTOR
);
5000 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu
*vcpu
,
5003 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
5005 if (is_guest_mode(vcpu
) &&
5006 vector
== vmx
->nested
.posted_intr_nv
) {
5007 /* the PIR and ON have been set by L1. */
5008 kvm_vcpu_trigger_posted_interrupt(vcpu
);
5010 * If a posted intr is not recognized by hardware,
5011 * we will accomplish it in the next vmentry.
5013 vmx
->nested
.pi_pending
= true;
5014 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
5020 * Send interrupt to vcpu via posted interrupt way.
5021 * 1. If target vcpu is running(non-root mode), send posted interrupt
5022 * notification to vcpu and hardware will sync PIR to vIRR atomically.
5023 * 2. If target vcpu isn't running(root mode), kick it to pick up the
5024 * interrupt from PIR in next vmentry.
5026 static void vmx_deliver_posted_interrupt(struct kvm_vcpu
*vcpu
, int vector
)
5028 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
5031 r
= vmx_deliver_nested_posted_interrupt(vcpu
, vector
);
5035 if (pi_test_and_set_pir(vector
, &vmx
->pi_desc
))
5038 /* If a previous notification has sent the IPI, nothing to do. */
5039 if (pi_test_and_set_on(&vmx
->pi_desc
))
5042 if (!kvm_vcpu_trigger_posted_interrupt(vcpu
))
5043 kvm_vcpu_kick(vcpu
);
5047 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
5048 * will not change in the lifetime of the guest.
5049 * Note that host-state that does change is set elsewhere. E.g., host-state
5050 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
5052 static void vmx_set_constant_host_state(struct vcpu_vmx
*vmx
)
5057 unsigned long cr0
, cr3
, cr4
;
5060 WARN_ON(cr0
& X86_CR0_TS
);
5061 vmcs_writel(HOST_CR0
, cr0
); /* 22.2.3 */
5064 * Save the most likely value for this task's CR3 in the VMCS.
5065 * We can't use __get_current_cr3_fast() because we're not atomic.
5068 vmcs_writel(HOST_CR3
, cr3
); /* 22.2.3 FIXME: shadow tables */
5069 vmx
->host_state
.vmcs_host_cr3
= cr3
;
5071 /* Save the most likely value for this task's CR4 in the VMCS. */
5072 cr4
= cr4_read_shadow();
5073 vmcs_writel(HOST_CR4
, cr4
); /* 22.2.3, 22.2.5 */
5074 vmx
->host_state
.vmcs_host_cr4
= cr4
;
5076 vmcs_write16(HOST_CS_SELECTOR
, __KERNEL_CS
); /* 22.2.4 */
5077 #ifdef CONFIG_X86_64
5079 * Load null selectors, so we can avoid reloading them in
5080 * __vmx_load_host_state(), in case userspace uses the null selectors
5081 * too (the expected case).
5083 vmcs_write16(HOST_DS_SELECTOR
, 0);
5084 vmcs_write16(HOST_ES_SELECTOR
, 0);
5086 vmcs_write16(HOST_DS_SELECTOR
, __KERNEL_DS
); /* 22.2.4 */
5087 vmcs_write16(HOST_ES_SELECTOR
, __KERNEL_DS
); /* 22.2.4 */
5089 vmcs_write16(HOST_SS_SELECTOR
, __KERNEL_DS
); /* 22.2.4 */
5090 vmcs_write16(HOST_TR_SELECTOR
, GDT_ENTRY_TSS
*8); /* 22.2.4 */
5092 native_store_idt(&dt
);
5093 vmcs_writel(HOST_IDTR_BASE
, dt
.address
); /* 22.2.4 */
5094 vmx
->host_idt_base
= dt
.address
;
5096 vmcs_writel(HOST_RIP
, vmx_return
); /* 22.2.5 */
5098 rdmsr(MSR_IA32_SYSENTER_CS
, low32
, high32
);
5099 vmcs_write32(HOST_IA32_SYSENTER_CS
, low32
);
5100 rdmsrl(MSR_IA32_SYSENTER_EIP
, tmpl
);
5101 vmcs_writel(HOST_IA32_SYSENTER_EIP
, tmpl
); /* 22.2.3 */
5103 if (vmcs_config
.vmexit_ctrl
& VM_EXIT_LOAD_IA32_PAT
) {
5104 rdmsr(MSR_IA32_CR_PAT
, low32
, high32
);
5105 vmcs_write64(HOST_IA32_PAT
, low32
| ((u64
) high32
<< 32));
5109 static void set_cr4_guest_host_mask(struct vcpu_vmx
*vmx
)
5111 vmx
->vcpu
.arch
.cr4_guest_owned_bits
= KVM_CR4_GUEST_OWNED_BITS
;
5113 vmx
->vcpu
.arch
.cr4_guest_owned_bits
|= X86_CR4_PGE
;
5114 if (is_guest_mode(&vmx
->vcpu
))
5115 vmx
->vcpu
.arch
.cr4_guest_owned_bits
&=
5116 ~get_vmcs12(&vmx
->vcpu
)->cr4_guest_host_mask
;
5117 vmcs_writel(CR4_GUEST_HOST_MASK
, ~vmx
->vcpu
.arch
.cr4_guest_owned_bits
);
5120 static u32
vmx_pin_based_exec_ctrl(struct vcpu_vmx
*vmx
)
5122 u32 pin_based_exec_ctrl
= vmcs_config
.pin_based_exec_ctrl
;
5124 if (!kvm_vcpu_apicv_active(&vmx
->vcpu
))
5125 pin_based_exec_ctrl
&= ~PIN_BASED_POSTED_INTR
;
5126 /* Enable the preemption timer dynamically */
5127 pin_based_exec_ctrl
&= ~PIN_BASED_VMX_PREEMPTION_TIMER
;
5128 return pin_based_exec_ctrl
;
5131 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu
*vcpu
)
5133 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
5135 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL
, vmx_pin_based_exec_ctrl(vmx
));
5136 if (cpu_has_secondary_exec_ctrls()) {
5137 if (kvm_vcpu_apicv_active(vcpu
))
5138 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL
,
5139 SECONDARY_EXEC_APIC_REGISTER_VIRT
|
5140 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY
);
5142 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL
,
5143 SECONDARY_EXEC_APIC_REGISTER_VIRT
|
5144 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY
);
5147 if (cpu_has_vmx_msr_bitmap())
5148 vmx_set_msr_bitmap(vcpu
);
5151 static u32
vmx_exec_control(struct vcpu_vmx
*vmx
)
5153 u32 exec_control
= vmcs_config
.cpu_based_exec_ctrl
;
5155 if (vmx
->vcpu
.arch
.switch_db_regs
& KVM_DEBUGREG_WONT_EXIT
)
5156 exec_control
&= ~CPU_BASED_MOV_DR_EXITING
;
5158 if (!cpu_need_tpr_shadow(&vmx
->vcpu
)) {
5159 exec_control
&= ~CPU_BASED_TPR_SHADOW
;
5160 #ifdef CONFIG_X86_64
5161 exec_control
|= CPU_BASED_CR8_STORE_EXITING
|
5162 CPU_BASED_CR8_LOAD_EXITING
;
5166 exec_control
|= CPU_BASED_CR3_STORE_EXITING
|
5167 CPU_BASED_CR3_LOAD_EXITING
|
5168 CPU_BASED_INVLPG_EXITING
;
5169 return exec_control
;
5172 static u32
vmx_secondary_exec_control(struct vcpu_vmx
*vmx
)
5174 u32 exec_control
= vmcs_config
.cpu_based_2nd_exec_ctrl
;
5175 if (!cpu_need_virtualize_apic_accesses(&vmx
->vcpu
))
5176 exec_control
&= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES
;
5178 exec_control
&= ~SECONDARY_EXEC_ENABLE_VPID
;
5180 exec_control
&= ~SECONDARY_EXEC_ENABLE_EPT
;
5181 enable_unrestricted_guest
= 0;
5182 /* Enable INVPCID for non-ept guests may cause performance regression. */
5183 exec_control
&= ~SECONDARY_EXEC_ENABLE_INVPCID
;
5185 if (!enable_unrestricted_guest
)
5186 exec_control
&= ~SECONDARY_EXEC_UNRESTRICTED_GUEST
;
5188 exec_control
&= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING
;
5189 if (!kvm_vcpu_apicv_active(&vmx
->vcpu
))
5190 exec_control
&= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT
|
5191 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY
);
5192 exec_control
&= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE
;
5193 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
5195 We can NOT enable shadow_vmcs here because we don't have yet
5198 exec_control
&= ~SECONDARY_EXEC_SHADOW_VMCS
;
5201 exec_control
&= ~SECONDARY_EXEC_ENABLE_PML
;
5203 return exec_control
;
5206 static void ept_set_mmio_spte_mask(void)
5209 * EPT Misconfigurations can be generated if the value of bits 2:0
5210 * of an EPT paging-structure entry is 110b (write/execute).
5212 kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK
,
5213 VMX_EPT_MISCONFIG_WX_VALUE
);
5216 #define VMX_XSS_EXIT_BITMAP 0
5218 * Sets up the vmcs for emulated real mode.
5220 static int vmx_vcpu_setup(struct vcpu_vmx
*vmx
)
5222 #ifdef CONFIG_X86_64
5228 vmcs_write64(IO_BITMAP_A
, __pa(vmx_io_bitmap_a
));
5229 vmcs_write64(IO_BITMAP_B
, __pa(vmx_io_bitmap_b
));
5231 if (enable_shadow_vmcs
) {
5232 vmcs_write64(VMREAD_BITMAP
, __pa(vmx_vmread_bitmap
));
5233 vmcs_write64(VMWRITE_BITMAP
, __pa(vmx_vmwrite_bitmap
));
5235 if (cpu_has_vmx_msr_bitmap())
5236 vmcs_write64(MSR_BITMAP
, __pa(vmx_msr_bitmap_legacy
));
5238 vmcs_write64(VMCS_LINK_POINTER
, -1ull); /* 22.3.1.5 */
5241 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL
, vmx_pin_based_exec_ctrl(vmx
));
5242 vmx
->hv_deadline_tsc
= -1;
5244 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL
, vmx_exec_control(vmx
));
5246 if (cpu_has_secondary_exec_ctrls()) {
5247 vmcs_write32(SECONDARY_VM_EXEC_CONTROL
,
5248 vmx_secondary_exec_control(vmx
));
5251 if (kvm_vcpu_apicv_active(&vmx
->vcpu
)) {
5252 vmcs_write64(EOI_EXIT_BITMAP0
, 0);
5253 vmcs_write64(EOI_EXIT_BITMAP1
, 0);
5254 vmcs_write64(EOI_EXIT_BITMAP2
, 0);
5255 vmcs_write64(EOI_EXIT_BITMAP3
, 0);
5257 vmcs_write16(GUEST_INTR_STATUS
, 0);
5259 vmcs_write16(POSTED_INTR_NV
, POSTED_INTR_VECTOR
);
5260 vmcs_write64(POSTED_INTR_DESC_ADDR
, __pa((&vmx
->pi_desc
)));
5264 vmcs_write32(PLE_GAP
, ple_gap
);
5265 vmx
->ple_window
= ple_window
;
5266 vmx
->ple_window_dirty
= true;
5269 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK
, 0);
5270 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH
, 0);
5271 vmcs_write32(CR3_TARGET_COUNT
, 0); /* 22.2.1 */
5273 vmcs_write16(HOST_FS_SELECTOR
, 0); /* 22.2.4 */
5274 vmcs_write16(HOST_GS_SELECTOR
, 0); /* 22.2.4 */
5275 vmx_set_constant_host_state(vmx
);
5276 #ifdef CONFIG_X86_64
5277 rdmsrl(MSR_FS_BASE
, a
);
5278 vmcs_writel(HOST_FS_BASE
, a
); /* 22.2.4 */
5279 rdmsrl(MSR_GS_BASE
, a
);
5280 vmcs_writel(HOST_GS_BASE
, a
); /* 22.2.4 */
5282 vmcs_writel(HOST_FS_BASE
, 0); /* 22.2.4 */
5283 vmcs_writel(HOST_GS_BASE
, 0); /* 22.2.4 */
5286 vmcs_write32(VM_EXIT_MSR_STORE_COUNT
, 0);
5287 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT
, 0);
5288 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR
, __pa(vmx
->msr_autoload
.host
));
5289 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT
, 0);
5290 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR
, __pa(vmx
->msr_autoload
.guest
));
5292 if (vmcs_config
.vmentry_ctrl
& VM_ENTRY_LOAD_IA32_PAT
)
5293 vmcs_write64(GUEST_IA32_PAT
, vmx
->vcpu
.arch
.pat
);
5295 for (i
= 0; i
< ARRAY_SIZE(vmx_msr_index
); ++i
) {
5296 u32 index
= vmx_msr_index
[i
];
5297 u32 data_low
, data_high
;
5300 if (rdmsr_safe(index
, &data_low
, &data_high
) < 0)
5302 if (wrmsr_safe(index
, data_low
, data_high
) < 0)
5304 vmx
->guest_msrs
[j
].index
= i
;
5305 vmx
->guest_msrs
[j
].data
= 0;
5306 vmx
->guest_msrs
[j
].mask
= -1ull;
5311 vm_exit_controls_init(vmx
, vmcs_config
.vmexit_ctrl
);
5313 /* 22.2.1, 20.8.1 */
5314 vm_entry_controls_init(vmx
, vmcs_config
.vmentry_ctrl
);
5316 vmx
->vcpu
.arch
.cr0_guest_owned_bits
= X86_CR0_TS
;
5317 vmcs_writel(CR0_GUEST_HOST_MASK
, ~X86_CR0_TS
);
5319 set_cr4_guest_host_mask(vmx
);
5321 if (vmx_xsaves_supported())
5322 vmcs_write64(XSS_EXIT_BITMAP
, VMX_XSS_EXIT_BITMAP
);
5325 ASSERT(vmx
->pml_pg
);
5326 vmcs_write64(PML_ADDRESS
, page_to_phys(vmx
->pml_pg
));
5327 vmcs_write16(GUEST_PML_INDEX
, PML_ENTITY_NUM
- 1);
5333 static void vmx_vcpu_reset(struct kvm_vcpu
*vcpu
, bool init_event
)
5335 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
5336 struct msr_data apic_base_msr
;
5339 vmx
->rmode
.vm86_active
= 0;
5341 vmx
->vcpu
.arch
.regs
[VCPU_REGS_RDX
] = get_rdx_init_val();
5342 kvm_set_cr8(vcpu
, 0);
5345 apic_base_msr
.data
= APIC_DEFAULT_PHYS_BASE
|
5346 MSR_IA32_APICBASE_ENABLE
;
5347 if (kvm_vcpu_is_reset_bsp(vcpu
))
5348 apic_base_msr
.data
|= MSR_IA32_APICBASE_BSP
;
5349 apic_base_msr
.host_initiated
= true;
5350 kvm_set_apic_base(vcpu
, &apic_base_msr
);
5353 vmx_segment_cache_clear(vmx
);
5355 seg_setup(VCPU_SREG_CS
);
5356 vmcs_write16(GUEST_CS_SELECTOR
, 0xf000);
5357 vmcs_writel(GUEST_CS_BASE
, 0xffff0000ul
);
5359 seg_setup(VCPU_SREG_DS
);
5360 seg_setup(VCPU_SREG_ES
);
5361 seg_setup(VCPU_SREG_FS
);
5362 seg_setup(VCPU_SREG_GS
);
5363 seg_setup(VCPU_SREG_SS
);
5365 vmcs_write16(GUEST_TR_SELECTOR
, 0);
5366 vmcs_writel(GUEST_TR_BASE
, 0);
5367 vmcs_write32(GUEST_TR_LIMIT
, 0xffff);
5368 vmcs_write32(GUEST_TR_AR_BYTES
, 0x008b);
5370 vmcs_write16(GUEST_LDTR_SELECTOR
, 0);
5371 vmcs_writel(GUEST_LDTR_BASE
, 0);
5372 vmcs_write32(GUEST_LDTR_LIMIT
, 0xffff);
5373 vmcs_write32(GUEST_LDTR_AR_BYTES
, 0x00082);
5376 vmcs_write32(GUEST_SYSENTER_CS
, 0);
5377 vmcs_writel(GUEST_SYSENTER_ESP
, 0);
5378 vmcs_writel(GUEST_SYSENTER_EIP
, 0);
5379 vmcs_write64(GUEST_IA32_DEBUGCTL
, 0);
5382 vmcs_writel(GUEST_RFLAGS
, 0x02);
5383 kvm_rip_write(vcpu
, 0xfff0);
5385 vmcs_writel(GUEST_GDTR_BASE
, 0);
5386 vmcs_write32(GUEST_GDTR_LIMIT
, 0xffff);
5388 vmcs_writel(GUEST_IDTR_BASE
, 0);
5389 vmcs_write32(GUEST_IDTR_LIMIT
, 0xffff);
5391 vmcs_write32(GUEST_ACTIVITY_STATE
, GUEST_ACTIVITY_ACTIVE
);
5392 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO
, 0);
5393 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS
, 0);
5397 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD
, 0); /* 22.2.1 */
5399 if (cpu_has_vmx_tpr_shadow() && !init_event
) {
5400 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR
, 0);
5401 if (cpu_need_tpr_shadow(vcpu
))
5402 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR
,
5403 __pa(vcpu
->arch
.apic
->regs
));
5404 vmcs_write32(TPR_THRESHOLD
, 0);
5407 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD
, vcpu
);
5409 if (kvm_vcpu_apicv_active(vcpu
))
5410 memset(&vmx
->pi_desc
, 0, sizeof(struct pi_desc
));
5413 vmcs_write16(VIRTUAL_PROCESSOR_ID
, vmx
->vpid
);
5415 cr0
= X86_CR0_NW
| X86_CR0_CD
| X86_CR0_ET
;
5416 vmx
->vcpu
.arch
.cr0
= cr0
;
5417 vmx_set_cr0(vcpu
, cr0
); /* enter rmode */
5418 vmx_set_cr4(vcpu
, 0);
5419 vmx_set_efer(vcpu
, 0);
5421 update_exception_bitmap(vcpu
);
5423 vpid_sync_context(vmx
->vpid
);
5427 * In nested virtualization, check if L1 asked to exit on external interrupts.
5428 * For most existing hypervisors, this will always return true.
5430 static bool nested_exit_on_intr(struct kvm_vcpu
*vcpu
)
5432 return get_vmcs12(vcpu
)->pin_based_vm_exec_control
&
5433 PIN_BASED_EXT_INTR_MASK
;
5437 * In nested virtualization, check if L1 has set
5438 * VM_EXIT_ACK_INTR_ON_EXIT
5440 static bool nested_exit_intr_ack_set(struct kvm_vcpu
*vcpu
)
5442 return get_vmcs12(vcpu
)->vm_exit_controls
&
5443 VM_EXIT_ACK_INTR_ON_EXIT
;
5446 static bool nested_exit_on_nmi(struct kvm_vcpu
*vcpu
)
5448 return get_vmcs12(vcpu
)->pin_based_vm_exec_control
&
5449 PIN_BASED_NMI_EXITING
;
5452 static void enable_irq_window(struct kvm_vcpu
*vcpu
)
5454 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL
,
5455 CPU_BASED_VIRTUAL_INTR_PENDING
);
5458 static void enable_nmi_window(struct kvm_vcpu
*vcpu
)
5460 if (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO
) & GUEST_INTR_STATE_STI
) {
5461 enable_irq_window(vcpu
);
5465 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL
,
5466 CPU_BASED_VIRTUAL_NMI_PENDING
);
5469 static void vmx_inject_irq(struct kvm_vcpu
*vcpu
)
5471 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
5473 int irq
= vcpu
->arch
.interrupt
.nr
;
5475 trace_kvm_inj_virq(irq
);
5477 ++vcpu
->stat
.irq_injections
;
5478 if (vmx
->rmode
.vm86_active
) {
5480 if (vcpu
->arch
.interrupt
.soft
)
5481 inc_eip
= vcpu
->arch
.event_exit_inst_len
;
5482 if (kvm_inject_realmode_interrupt(vcpu
, irq
, inc_eip
) != EMULATE_DONE
)
5483 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, vcpu
);
5486 intr
= irq
| INTR_INFO_VALID_MASK
;
5487 if (vcpu
->arch
.interrupt
.soft
) {
5488 intr
|= INTR_TYPE_SOFT_INTR
;
5489 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN
,
5490 vmx
->vcpu
.arch
.event_exit_inst_len
);
5492 intr
|= INTR_TYPE_EXT_INTR
;
5493 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD
, intr
);
5496 static void vmx_inject_nmi(struct kvm_vcpu
*vcpu
)
5498 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
5500 if (!is_guest_mode(vcpu
)) {
5501 ++vcpu
->stat
.nmi_injections
;
5502 vmx
->nmi_known_unmasked
= false;
5505 if (vmx
->rmode
.vm86_active
) {
5506 if (kvm_inject_realmode_interrupt(vcpu
, NMI_VECTOR
, 0) != EMULATE_DONE
)
5507 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, vcpu
);
5511 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD
,
5512 INTR_TYPE_NMI_INTR
| INTR_INFO_VALID_MASK
| NMI_VECTOR
);
5515 static bool vmx_get_nmi_mask(struct kvm_vcpu
*vcpu
)
5517 if (to_vmx(vcpu
)->nmi_known_unmasked
)
5519 return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO
) & GUEST_INTR_STATE_NMI
;
5522 static void vmx_set_nmi_mask(struct kvm_vcpu
*vcpu
, bool masked
)
5524 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
5526 vmx
->nmi_known_unmasked
= !masked
;
5528 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO
,
5529 GUEST_INTR_STATE_NMI
);
5531 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO
,
5532 GUEST_INTR_STATE_NMI
);
5535 static int vmx_nmi_allowed(struct kvm_vcpu
*vcpu
)
5537 if (to_vmx(vcpu
)->nested
.nested_run_pending
)
5540 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO
) &
5541 (GUEST_INTR_STATE_MOV_SS
| GUEST_INTR_STATE_STI
5542 | GUEST_INTR_STATE_NMI
));
5545 static int vmx_interrupt_allowed(struct kvm_vcpu
*vcpu
)
5547 return (!to_vmx(vcpu
)->nested
.nested_run_pending
&&
5548 vmcs_readl(GUEST_RFLAGS
) & X86_EFLAGS_IF
) &&
5549 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO
) &
5550 (GUEST_INTR_STATE_STI
| GUEST_INTR_STATE_MOV_SS
));
5553 static int vmx_set_tss_addr(struct kvm
*kvm
, unsigned int addr
)
5557 ret
= x86_set_memory_region(kvm
, TSS_PRIVATE_MEMSLOT
, addr
,
5561 kvm
->arch
.tss_addr
= addr
;
5562 return init_rmode_tss(kvm
);
5565 static bool rmode_exception(struct kvm_vcpu
*vcpu
, int vec
)
5570 * Update instruction length as we may reinject the exception
5571 * from user space while in guest debugging mode.
5573 to_vmx(vcpu
)->vcpu
.arch
.event_exit_inst_len
=
5574 vmcs_read32(VM_EXIT_INSTRUCTION_LEN
);
5575 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_SW_BP
)
5579 if (vcpu
->guest_debug
&
5580 (KVM_GUESTDBG_SINGLESTEP
| KVM_GUESTDBG_USE_HW_BP
))
5597 static int handle_rmode_exception(struct kvm_vcpu
*vcpu
,
5598 int vec
, u32 err_code
)
5601 * Instruction with address size override prefix opcode 0x67
5602 * Cause the #SS fault with 0 error code in VM86 mode.
5604 if (((vec
== GP_VECTOR
) || (vec
== SS_VECTOR
)) && err_code
== 0) {
5605 if (emulate_instruction(vcpu
, 0) == EMULATE_DONE
) {
5606 if (vcpu
->arch
.halt_request
) {
5607 vcpu
->arch
.halt_request
= 0;
5608 return kvm_vcpu_halt(vcpu
);
5616 * Forward all other exceptions that are valid in real mode.
5617 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
5618 * the required debugging infrastructure rework.
5620 kvm_queue_exception(vcpu
, vec
);
5625 * Trigger machine check on the host. We assume all the MSRs are already set up
5626 * by the CPU and that we still run on the same CPU as the MCE occurred on.
5627 * We pass a fake environment to the machine check handler because we want
5628 * the guest to be always treated like user space, no matter what context
5629 * it used internally.
5631 static void kvm_machine_check(void)
5633 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
5634 struct pt_regs regs
= {
5635 .cs
= 3, /* Fake ring 3 no matter what the guest ran on */
5636 .flags
= X86_EFLAGS_IF
,
5639 do_machine_check(®s
, 0);
5643 static int handle_machine_check(struct kvm_vcpu
*vcpu
)
5645 /* already handled by vcpu_run */
5649 static int handle_exception(struct kvm_vcpu
*vcpu
)
5651 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
5652 struct kvm_run
*kvm_run
= vcpu
->run
;
5653 u32 intr_info
, ex_no
, error_code
;
5654 unsigned long cr2
, rip
, dr6
;
5656 enum emulation_result er
;
5658 vect_info
= vmx
->idt_vectoring_info
;
5659 intr_info
= vmx
->exit_intr_info
;
5661 if (is_machine_check(intr_info
))
5662 return handle_machine_check(vcpu
);
5664 if (is_nmi(intr_info
))
5665 return 1; /* already handled by vmx_vcpu_run() */
5667 if (is_invalid_opcode(intr_info
)) {
5668 if (is_guest_mode(vcpu
)) {
5669 kvm_queue_exception(vcpu
, UD_VECTOR
);
5672 er
= emulate_instruction(vcpu
, EMULTYPE_TRAP_UD
);
5673 if (er
!= EMULATE_DONE
)
5674 kvm_queue_exception(vcpu
, UD_VECTOR
);
5679 if (intr_info
& INTR_INFO_DELIVER_CODE_MASK
)
5680 error_code
= vmcs_read32(VM_EXIT_INTR_ERROR_CODE
);
5683 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
5684 * MMIO, it is better to report an internal error.
5685 * See the comments in vmx_handle_exit.
5687 if ((vect_info
& VECTORING_INFO_VALID_MASK
) &&
5688 !(is_page_fault(intr_info
) && !(error_code
& PFERR_RSVD_MASK
))) {
5689 vcpu
->run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
5690 vcpu
->run
->internal
.suberror
= KVM_INTERNAL_ERROR_SIMUL_EX
;
5691 vcpu
->run
->internal
.ndata
= 3;
5692 vcpu
->run
->internal
.data
[0] = vect_info
;
5693 vcpu
->run
->internal
.data
[1] = intr_info
;
5694 vcpu
->run
->internal
.data
[2] = error_code
;
5698 if (is_page_fault(intr_info
)) {
5699 /* EPT won't cause page fault directly */
5701 cr2
= vmcs_readl(EXIT_QUALIFICATION
);
5702 trace_kvm_page_fault(cr2
, error_code
);
5704 if (kvm_event_needs_reinjection(vcpu
))
5705 kvm_mmu_unprotect_page_virt(vcpu
, cr2
);
5706 return kvm_mmu_page_fault(vcpu
, cr2
, error_code
, NULL
, 0);
5709 ex_no
= intr_info
& INTR_INFO_VECTOR_MASK
;
5711 if (vmx
->rmode
.vm86_active
&& rmode_exception(vcpu
, ex_no
))
5712 return handle_rmode_exception(vcpu
, ex_no
, error_code
);
5716 kvm_queue_exception_e(vcpu
, AC_VECTOR
, error_code
);
5719 dr6
= vmcs_readl(EXIT_QUALIFICATION
);
5720 if (!(vcpu
->guest_debug
&
5721 (KVM_GUESTDBG_SINGLESTEP
| KVM_GUESTDBG_USE_HW_BP
))) {
5722 vcpu
->arch
.dr6
&= ~15;
5723 vcpu
->arch
.dr6
|= dr6
| DR6_RTM
;
5724 if (!(dr6
& ~DR6_RESERVED
)) /* icebp */
5725 skip_emulated_instruction(vcpu
);
5727 kvm_queue_exception(vcpu
, DB_VECTOR
);
5730 kvm_run
->debug
.arch
.dr6
= dr6
| DR6_FIXED_1
;
5731 kvm_run
->debug
.arch
.dr7
= vmcs_readl(GUEST_DR7
);
5735 * Update instruction length as we may reinject #BP from
5736 * user space while in guest debugging mode. Reading it for
5737 * #DB as well causes no harm, it is not used in that case.
5739 vmx
->vcpu
.arch
.event_exit_inst_len
=
5740 vmcs_read32(VM_EXIT_INSTRUCTION_LEN
);
5741 kvm_run
->exit_reason
= KVM_EXIT_DEBUG
;
5742 rip
= kvm_rip_read(vcpu
);
5743 kvm_run
->debug
.arch
.pc
= vmcs_readl(GUEST_CS_BASE
) + rip
;
5744 kvm_run
->debug
.arch
.exception
= ex_no
;
5747 kvm_run
->exit_reason
= KVM_EXIT_EXCEPTION
;
5748 kvm_run
->ex
.exception
= ex_no
;
5749 kvm_run
->ex
.error_code
= error_code
;
5755 static int handle_external_interrupt(struct kvm_vcpu
*vcpu
)
5757 ++vcpu
->stat
.irq_exits
;
5761 static int handle_triple_fault(struct kvm_vcpu
*vcpu
)
5763 vcpu
->run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
5767 static int handle_io(struct kvm_vcpu
*vcpu
)
5769 unsigned long exit_qualification
;
5770 int size
, in
, string
, ret
;
5773 exit_qualification
= vmcs_readl(EXIT_QUALIFICATION
);
5774 string
= (exit_qualification
& 16) != 0;
5775 in
= (exit_qualification
& 8) != 0;
5777 ++vcpu
->stat
.io_exits
;
5780 return emulate_instruction(vcpu
, 0) == EMULATE_DONE
;
5782 port
= exit_qualification
>> 16;
5783 size
= (exit_qualification
& 7) + 1;
5785 ret
= kvm_skip_emulated_instruction(vcpu
);
5788 * TODO: we might be squashing a KVM_GUESTDBG_SINGLESTEP-triggered
5789 * KVM_EXIT_DEBUG here.
5791 return kvm_fast_pio_out(vcpu
, size
, port
) && ret
;
5795 vmx_patch_hypercall(struct kvm_vcpu
*vcpu
, unsigned char *hypercall
)
5798 * Patch in the VMCALL instruction:
5800 hypercall
[0] = 0x0f;
5801 hypercall
[1] = 0x01;
5802 hypercall
[2] = 0xc1;
5805 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
5806 static int handle_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long val
)
5808 if (is_guest_mode(vcpu
)) {
5809 struct vmcs12
*vmcs12
= get_vmcs12(vcpu
);
5810 unsigned long orig_val
= val
;
5813 * We get here when L2 changed cr0 in a way that did not change
5814 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
5815 * but did change L0 shadowed bits. So we first calculate the
5816 * effective cr0 value that L1 would like to write into the
5817 * hardware. It consists of the L2-owned bits from the new
5818 * value combined with the L1-owned bits from L1's guest_cr0.
5820 val
= (val
& ~vmcs12
->cr0_guest_host_mask
) |
5821 (vmcs12
->guest_cr0
& vmcs12
->cr0_guest_host_mask
);
5823 if (!nested_guest_cr0_valid(vcpu
, val
))
5826 if (kvm_set_cr0(vcpu
, val
))
5828 vmcs_writel(CR0_READ_SHADOW
, orig_val
);
5831 if (to_vmx(vcpu
)->nested
.vmxon
&&
5832 !nested_host_cr0_valid(vcpu
, val
))
5835 return kvm_set_cr0(vcpu
, val
);
5839 static int handle_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long val
)
5841 if (is_guest_mode(vcpu
)) {
5842 struct vmcs12
*vmcs12
= get_vmcs12(vcpu
);
5843 unsigned long orig_val
= val
;
5845 /* analogously to handle_set_cr0 */
5846 val
= (val
& ~vmcs12
->cr4_guest_host_mask
) |
5847 (vmcs12
->guest_cr4
& vmcs12
->cr4_guest_host_mask
);
5848 if (kvm_set_cr4(vcpu
, val
))
5850 vmcs_writel(CR4_READ_SHADOW
, orig_val
);
5853 return kvm_set_cr4(vcpu
, val
);
5856 static int handle_cr(struct kvm_vcpu
*vcpu
)
5858 unsigned long exit_qualification
, val
;
5864 exit_qualification
= vmcs_readl(EXIT_QUALIFICATION
);
5865 cr
= exit_qualification
& 15;
5866 reg
= (exit_qualification
>> 8) & 15;
5867 switch ((exit_qualification
>> 4) & 3) {
5868 case 0: /* mov to cr */
5869 val
= kvm_register_readl(vcpu
, reg
);
5870 trace_kvm_cr_write(cr
, val
);
5873 err
= handle_set_cr0(vcpu
, val
);
5874 return kvm_complete_insn_gp(vcpu
, err
);
5876 err
= kvm_set_cr3(vcpu
, val
);
5877 return kvm_complete_insn_gp(vcpu
, err
);
5879 err
= handle_set_cr4(vcpu
, val
);
5880 return kvm_complete_insn_gp(vcpu
, err
);
5882 u8 cr8_prev
= kvm_get_cr8(vcpu
);
5884 err
= kvm_set_cr8(vcpu
, cr8
);
5885 ret
= kvm_complete_insn_gp(vcpu
, err
);
5886 if (lapic_in_kernel(vcpu
))
5888 if (cr8_prev
<= cr8
)
5891 * TODO: we might be squashing a
5892 * KVM_GUESTDBG_SINGLESTEP-triggered
5893 * KVM_EXIT_DEBUG here.
5895 vcpu
->run
->exit_reason
= KVM_EXIT_SET_TPR
;
5901 WARN_ONCE(1, "Guest should always own CR0.TS");
5902 vmx_set_cr0(vcpu
, kvm_read_cr0_bits(vcpu
, ~X86_CR0_TS
));
5903 trace_kvm_cr_write(0, kvm_read_cr0(vcpu
));
5904 return kvm_skip_emulated_instruction(vcpu
);
5905 case 1: /*mov from cr*/
5908 val
= kvm_read_cr3(vcpu
);
5909 kvm_register_write(vcpu
, reg
, val
);
5910 trace_kvm_cr_read(cr
, val
);
5911 return kvm_skip_emulated_instruction(vcpu
);
5913 val
= kvm_get_cr8(vcpu
);
5914 kvm_register_write(vcpu
, reg
, val
);
5915 trace_kvm_cr_read(cr
, val
);
5916 return kvm_skip_emulated_instruction(vcpu
);
5920 val
= (exit_qualification
>> LMSW_SOURCE_DATA_SHIFT
) & 0x0f;
5921 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu
) & ~0xful
) | val
);
5922 kvm_lmsw(vcpu
, val
);
5924 return kvm_skip_emulated_instruction(vcpu
);
5928 vcpu
->run
->exit_reason
= 0;
5929 vcpu_unimpl(vcpu
, "unhandled control register: op %d cr %d\n",
5930 (int)(exit_qualification
>> 4) & 3, cr
);
5934 static int handle_dr(struct kvm_vcpu
*vcpu
)
5936 unsigned long exit_qualification
;
5939 exit_qualification
= vmcs_readl(EXIT_QUALIFICATION
);
5940 dr
= exit_qualification
& DEBUG_REG_ACCESS_NUM
;
5942 /* First, if DR does not exist, trigger UD */
5943 if (!kvm_require_dr(vcpu
, dr
))
5946 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
5947 if (!kvm_require_cpl(vcpu
, 0))
5949 dr7
= vmcs_readl(GUEST_DR7
);
5952 * As the vm-exit takes precedence over the debug trap, we
5953 * need to emulate the latter, either for the host or the
5954 * guest debugging itself.
5956 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
) {
5957 vcpu
->run
->debug
.arch
.dr6
= vcpu
->arch
.dr6
;
5958 vcpu
->run
->debug
.arch
.dr7
= dr7
;
5959 vcpu
->run
->debug
.arch
.pc
= kvm_get_linear_rip(vcpu
);
5960 vcpu
->run
->debug
.arch
.exception
= DB_VECTOR
;
5961 vcpu
->run
->exit_reason
= KVM_EXIT_DEBUG
;
5964 vcpu
->arch
.dr6
&= ~15;
5965 vcpu
->arch
.dr6
|= DR6_BD
| DR6_RTM
;
5966 kvm_queue_exception(vcpu
, DB_VECTOR
);
5971 if (vcpu
->guest_debug
== 0) {
5972 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL
,
5973 CPU_BASED_MOV_DR_EXITING
);
5976 * No more DR vmexits; force a reload of the debug registers
5977 * and reenter on this instruction. The next vmexit will
5978 * retrieve the full state of the debug registers.
5980 vcpu
->arch
.switch_db_regs
|= KVM_DEBUGREG_WONT_EXIT
;
5984 reg
= DEBUG_REG_ACCESS_REG(exit_qualification
);
5985 if (exit_qualification
& TYPE_MOV_FROM_DR
) {
5988 if (kvm_get_dr(vcpu
, dr
, &val
))
5990 kvm_register_write(vcpu
, reg
, val
);
5992 if (kvm_set_dr(vcpu
, dr
, kvm_register_readl(vcpu
, reg
)))
5995 return kvm_skip_emulated_instruction(vcpu
);
5998 static u64
vmx_get_dr6(struct kvm_vcpu
*vcpu
)
6000 return vcpu
->arch
.dr6
;
6003 static void vmx_set_dr6(struct kvm_vcpu
*vcpu
, unsigned long val
)
6007 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu
*vcpu
)
6009 get_debugreg(vcpu
->arch
.db
[0], 0);
6010 get_debugreg(vcpu
->arch
.db
[1], 1);
6011 get_debugreg(vcpu
->arch
.db
[2], 2);
6012 get_debugreg(vcpu
->arch
.db
[3], 3);
6013 get_debugreg(vcpu
->arch
.dr6
, 6);
6014 vcpu
->arch
.dr7
= vmcs_readl(GUEST_DR7
);
6016 vcpu
->arch
.switch_db_regs
&= ~KVM_DEBUGREG_WONT_EXIT
;
6017 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL
, CPU_BASED_MOV_DR_EXITING
);
6020 static void vmx_set_dr7(struct kvm_vcpu
*vcpu
, unsigned long val
)
6022 vmcs_writel(GUEST_DR7
, val
);
6025 static int handle_cpuid(struct kvm_vcpu
*vcpu
)
6027 return kvm_emulate_cpuid(vcpu
);
6030 static int handle_rdmsr(struct kvm_vcpu
*vcpu
)
6032 u32 ecx
= vcpu
->arch
.regs
[VCPU_REGS_RCX
];
6033 struct msr_data msr_info
;
6035 msr_info
.index
= ecx
;
6036 msr_info
.host_initiated
= false;
6037 if (vmx_get_msr(vcpu
, &msr_info
)) {
6038 trace_kvm_msr_read_ex(ecx
);
6039 kvm_inject_gp(vcpu
, 0);
6043 trace_kvm_msr_read(ecx
, msr_info
.data
);
6045 /* FIXME: handling of bits 32:63 of rax, rdx */
6046 vcpu
->arch
.regs
[VCPU_REGS_RAX
] = msr_info
.data
& -1u;
6047 vcpu
->arch
.regs
[VCPU_REGS_RDX
] = (msr_info
.data
>> 32) & -1u;
6048 return kvm_skip_emulated_instruction(vcpu
);
6051 static int handle_wrmsr(struct kvm_vcpu
*vcpu
)
6053 struct msr_data msr
;
6054 u32 ecx
= vcpu
->arch
.regs
[VCPU_REGS_RCX
];
6055 u64 data
= (vcpu
->arch
.regs
[VCPU_REGS_RAX
] & -1u)
6056 | ((u64
)(vcpu
->arch
.regs
[VCPU_REGS_RDX
] & -1u) << 32);
6060 msr
.host_initiated
= false;
6061 if (kvm_set_msr(vcpu
, &msr
) != 0) {
6062 trace_kvm_msr_write_ex(ecx
, data
);
6063 kvm_inject_gp(vcpu
, 0);
6067 trace_kvm_msr_write(ecx
, data
);
6068 return kvm_skip_emulated_instruction(vcpu
);
6071 static int handle_tpr_below_threshold(struct kvm_vcpu
*vcpu
)
6073 kvm_apic_update_ppr(vcpu
);
6077 static int handle_interrupt_window(struct kvm_vcpu
*vcpu
)
6079 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL
,
6080 CPU_BASED_VIRTUAL_INTR_PENDING
);
6082 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6084 ++vcpu
->stat
.irq_window_exits
;
6088 static int handle_halt(struct kvm_vcpu
*vcpu
)
6090 return kvm_emulate_halt(vcpu
);
6093 static int handle_vmcall(struct kvm_vcpu
*vcpu
)
6095 return kvm_emulate_hypercall(vcpu
);
6098 static int handle_invd(struct kvm_vcpu
*vcpu
)
6100 return emulate_instruction(vcpu
, 0) == EMULATE_DONE
;
6103 static int handle_invlpg(struct kvm_vcpu
*vcpu
)
6105 unsigned long exit_qualification
= vmcs_readl(EXIT_QUALIFICATION
);
6107 kvm_mmu_invlpg(vcpu
, exit_qualification
);
6108 return kvm_skip_emulated_instruction(vcpu
);
6111 static int handle_rdpmc(struct kvm_vcpu
*vcpu
)
6115 err
= kvm_rdpmc(vcpu
);
6116 return kvm_complete_insn_gp(vcpu
, err
);
6119 static int handle_wbinvd(struct kvm_vcpu
*vcpu
)
6121 return kvm_emulate_wbinvd(vcpu
);
6124 static int handle_xsetbv(struct kvm_vcpu
*vcpu
)
6126 u64 new_bv
= kvm_read_edx_eax(vcpu
);
6127 u32 index
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
6129 if (kvm_set_xcr(vcpu
, index
, new_bv
) == 0)
6130 return kvm_skip_emulated_instruction(vcpu
);
6134 static int handle_xsaves(struct kvm_vcpu
*vcpu
)
6136 kvm_skip_emulated_instruction(vcpu
);
6137 WARN(1, "this should never happen\n");
6141 static int handle_xrstors(struct kvm_vcpu
*vcpu
)
6143 kvm_skip_emulated_instruction(vcpu
);
6144 WARN(1, "this should never happen\n");
6148 static int handle_apic_access(struct kvm_vcpu
*vcpu
)
6150 if (likely(fasteoi
)) {
6151 unsigned long exit_qualification
= vmcs_readl(EXIT_QUALIFICATION
);
6152 int access_type
, offset
;
6154 access_type
= exit_qualification
& APIC_ACCESS_TYPE
;
6155 offset
= exit_qualification
& APIC_ACCESS_OFFSET
;
6157 * Sane guest uses MOV to write EOI, with written value
6158 * not cared. So make a short-circuit here by avoiding
6159 * heavy instruction emulation.
6161 if ((access_type
== TYPE_LINEAR_APIC_INST_WRITE
) &&
6162 (offset
== APIC_EOI
)) {
6163 kvm_lapic_set_eoi(vcpu
);
6164 return kvm_skip_emulated_instruction(vcpu
);
6167 return emulate_instruction(vcpu
, 0) == EMULATE_DONE
;
6170 static int handle_apic_eoi_induced(struct kvm_vcpu
*vcpu
)
6172 unsigned long exit_qualification
= vmcs_readl(EXIT_QUALIFICATION
);
6173 int vector
= exit_qualification
& 0xff;
6175 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
6176 kvm_apic_set_eoi_accelerated(vcpu
, vector
);
6180 static int handle_apic_write(struct kvm_vcpu
*vcpu
)
6182 unsigned long exit_qualification
= vmcs_readl(EXIT_QUALIFICATION
);
6183 u32 offset
= exit_qualification
& 0xfff;
6185 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
6186 kvm_apic_write_nodecode(vcpu
, offset
);
6190 static int handle_task_switch(struct kvm_vcpu
*vcpu
)
6192 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
6193 unsigned long exit_qualification
;
6194 bool has_error_code
= false;
6197 int reason
, type
, idt_v
, idt_index
;
6199 idt_v
= (vmx
->idt_vectoring_info
& VECTORING_INFO_VALID_MASK
);
6200 idt_index
= (vmx
->idt_vectoring_info
& VECTORING_INFO_VECTOR_MASK
);
6201 type
= (vmx
->idt_vectoring_info
& VECTORING_INFO_TYPE_MASK
);
6203 exit_qualification
= vmcs_readl(EXIT_QUALIFICATION
);
6205 reason
= (u32
)exit_qualification
>> 30;
6206 if (reason
== TASK_SWITCH_GATE
&& idt_v
) {
6208 case INTR_TYPE_NMI_INTR
:
6209 vcpu
->arch
.nmi_injected
= false;
6210 vmx_set_nmi_mask(vcpu
, true);
6212 case INTR_TYPE_EXT_INTR
:
6213 case INTR_TYPE_SOFT_INTR
:
6214 kvm_clear_interrupt_queue(vcpu
);
6216 case INTR_TYPE_HARD_EXCEPTION
:
6217 if (vmx
->idt_vectoring_info
&
6218 VECTORING_INFO_DELIVER_CODE_MASK
) {
6219 has_error_code
= true;
6221 vmcs_read32(IDT_VECTORING_ERROR_CODE
);
6224 case INTR_TYPE_SOFT_EXCEPTION
:
6225 kvm_clear_exception_queue(vcpu
);
6231 tss_selector
= exit_qualification
;
6233 if (!idt_v
|| (type
!= INTR_TYPE_HARD_EXCEPTION
&&
6234 type
!= INTR_TYPE_EXT_INTR
&&
6235 type
!= INTR_TYPE_NMI_INTR
))
6236 skip_emulated_instruction(vcpu
);
6238 if (kvm_task_switch(vcpu
, tss_selector
,
6239 type
== INTR_TYPE_SOFT_INTR
? idt_index
: -1, reason
,
6240 has_error_code
, error_code
) == EMULATE_FAIL
) {
6241 vcpu
->run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
6242 vcpu
->run
->internal
.suberror
= KVM_INTERNAL_ERROR_EMULATION
;
6243 vcpu
->run
->internal
.ndata
= 0;
6248 * TODO: What about debug traps on tss switch?
6249 * Are we supposed to inject them and update dr6?
6255 static int handle_ept_violation(struct kvm_vcpu
*vcpu
)
6257 unsigned long exit_qualification
;
6261 exit_qualification
= vmcs_readl(EXIT_QUALIFICATION
);
6264 * EPT violation happened while executing iret from NMI,
6265 * "blocked by NMI" bit has to be set before next VM entry.
6266 * There are errata that may cause this bit to not be set:
6269 if (!(to_vmx(vcpu
)->idt_vectoring_info
& VECTORING_INFO_VALID_MASK
) &&
6270 (exit_qualification
& INTR_INFO_UNBLOCK_NMI
))
6271 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO
, GUEST_INTR_STATE_NMI
);
6273 gpa
= vmcs_read64(GUEST_PHYSICAL_ADDRESS
);
6274 trace_kvm_page_fault(gpa
, exit_qualification
);
6276 /* Is it a read fault? */
6277 error_code
= (exit_qualification
& EPT_VIOLATION_ACC_READ
)
6278 ? PFERR_USER_MASK
: 0;
6279 /* Is it a write fault? */
6280 error_code
|= (exit_qualification
& EPT_VIOLATION_ACC_WRITE
)
6281 ? PFERR_WRITE_MASK
: 0;
6282 /* Is it a fetch fault? */
6283 error_code
|= (exit_qualification
& EPT_VIOLATION_ACC_INSTR
)
6284 ? PFERR_FETCH_MASK
: 0;
6285 /* ept page table entry is present? */
6286 error_code
|= (exit_qualification
&
6287 (EPT_VIOLATION_READABLE
| EPT_VIOLATION_WRITABLE
|
6288 EPT_VIOLATION_EXECUTABLE
))
6289 ? PFERR_PRESENT_MASK
: 0;
6291 vcpu
->arch
.gpa_available
= true;
6292 vcpu
->arch
.exit_qualification
= exit_qualification
;
6294 return kvm_mmu_page_fault(vcpu
, gpa
, error_code
, NULL
, 0);
6297 static int handle_ept_misconfig(struct kvm_vcpu
*vcpu
)
6302 gpa
= vmcs_read64(GUEST_PHYSICAL_ADDRESS
);
6303 if (!kvm_io_bus_write(vcpu
, KVM_FAST_MMIO_BUS
, gpa
, 0, NULL
)) {
6304 trace_kvm_fast_mmio(gpa
);
6305 return kvm_skip_emulated_instruction(vcpu
);
6308 ret
= handle_mmio_page_fault(vcpu
, gpa
, true);
6309 vcpu
->arch
.gpa_available
= true;
6310 if (likely(ret
== RET_MMIO_PF_EMULATE
))
6311 return x86_emulate_instruction(vcpu
, gpa
, 0, NULL
, 0) ==
6314 if (unlikely(ret
== RET_MMIO_PF_INVALID
))
6315 return kvm_mmu_page_fault(vcpu
, gpa
, 0, NULL
, 0);
6317 if (unlikely(ret
== RET_MMIO_PF_RETRY
))
6320 /* It is the real ept misconfig */
6323 vcpu
->run
->exit_reason
= KVM_EXIT_UNKNOWN
;
6324 vcpu
->run
->hw
.hardware_exit_reason
= EXIT_REASON_EPT_MISCONFIG
;
6329 static int handle_nmi_window(struct kvm_vcpu
*vcpu
)
6331 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL
,
6332 CPU_BASED_VIRTUAL_NMI_PENDING
);
6333 ++vcpu
->stat
.nmi_window_exits
;
6334 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
6339 static int handle_invalid_guest_state(struct kvm_vcpu
*vcpu
)
6341 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
6342 enum emulation_result err
= EMULATE_DONE
;
6345 bool intr_window_requested
;
6346 unsigned count
= 130;
6348 cpu_exec_ctrl
= vmcs_read32(CPU_BASED_VM_EXEC_CONTROL
);
6349 intr_window_requested
= cpu_exec_ctrl
& CPU_BASED_VIRTUAL_INTR_PENDING
;
6351 while (vmx
->emulation_required
&& count
-- != 0) {
6352 if (intr_window_requested
&& vmx_interrupt_allowed(vcpu
))
6353 return handle_interrupt_window(&vmx
->vcpu
);
6355 if (kvm_test_request(KVM_REQ_EVENT
, vcpu
))
6358 err
= emulate_instruction(vcpu
, EMULTYPE_NO_REEXECUTE
);
6360 if (err
== EMULATE_USER_EXIT
) {
6361 ++vcpu
->stat
.mmio_exits
;
6366 if (err
!= EMULATE_DONE
) {
6367 vcpu
->run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
6368 vcpu
->run
->internal
.suberror
= KVM_INTERNAL_ERROR_EMULATION
;
6369 vcpu
->run
->internal
.ndata
= 0;
6373 if (vcpu
->arch
.halt_request
) {
6374 vcpu
->arch
.halt_request
= 0;
6375 ret
= kvm_vcpu_halt(vcpu
);
6379 if (signal_pending(current
))
6389 static int __grow_ple_window(int val
)
6391 if (ple_window_grow
< 1)
6394 val
= min(val
, ple_window_actual_max
);
6396 if (ple_window_grow
< ple_window
)
6397 val
*= ple_window_grow
;
6399 val
+= ple_window_grow
;
6404 static int __shrink_ple_window(int val
, int modifier
, int minimum
)
6409 if (modifier
< ple_window
)
6414 return max(val
, minimum
);
6417 static void grow_ple_window(struct kvm_vcpu
*vcpu
)
6419 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
6420 int old
= vmx
->ple_window
;
6422 vmx
->ple_window
= __grow_ple_window(old
);
6424 if (vmx
->ple_window
!= old
)
6425 vmx
->ple_window_dirty
= true;
6427 trace_kvm_ple_window_grow(vcpu
->vcpu_id
, vmx
->ple_window
, old
);
6430 static void shrink_ple_window(struct kvm_vcpu
*vcpu
)
6432 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
6433 int old
= vmx
->ple_window
;
6435 vmx
->ple_window
= __shrink_ple_window(old
,
6436 ple_window_shrink
, ple_window
);
6438 if (vmx
->ple_window
!= old
)
6439 vmx
->ple_window_dirty
= true;
6441 trace_kvm_ple_window_shrink(vcpu
->vcpu_id
, vmx
->ple_window
, old
);
6445 * ple_window_actual_max is computed to be one grow_ple_window() below
6446 * ple_window_max. (See __grow_ple_window for the reason.)
6447 * This prevents overflows, because ple_window_max is int.
6448 * ple_window_max effectively rounded down to a multiple of ple_window_grow in
6450 * ple_window_max is also prevented from setting vmx->ple_window < ple_window.
6452 static void update_ple_window_actual_max(void)
6454 ple_window_actual_max
=
6455 __shrink_ple_window(max(ple_window_max
, ple_window
),
6456 ple_window_grow
, INT_MIN
);
6460 * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
6462 static void wakeup_handler(void)
6464 struct kvm_vcpu
*vcpu
;
6465 int cpu
= smp_processor_id();
6467 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock
, cpu
));
6468 list_for_each_entry(vcpu
, &per_cpu(blocked_vcpu_on_cpu
, cpu
),
6469 blocked_vcpu_list
) {
6470 struct pi_desc
*pi_desc
= vcpu_to_pi_desc(vcpu
);
6472 if (pi_test_on(pi_desc
) == 1)
6473 kvm_vcpu_kick(vcpu
);
6475 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock
, cpu
));
6478 void vmx_enable_tdp(void)
6480 kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK
,
6481 enable_ept_ad_bits
? VMX_EPT_ACCESS_BIT
: 0ull,
6482 enable_ept_ad_bits
? VMX_EPT_DIRTY_BIT
: 0ull,
6483 0ull, VMX_EPT_EXECUTABLE_MASK
,
6484 cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK
,
6487 ept_set_mmio_spte_mask();
6491 static __init
int hardware_setup(void)
6493 int r
= -ENOMEM
, i
, msr
;
6495 rdmsrl_safe(MSR_EFER
, &host_efer
);
6497 for (i
= 0; i
< ARRAY_SIZE(vmx_msr_index
); ++i
)
6498 kvm_define_shared_msr(i
, vmx_msr_index
[i
]);
6500 for (i
= 0; i
< VMX_BITMAP_NR
; i
++) {
6501 vmx_bitmap
[i
] = (unsigned long *)__get_free_page(GFP_KERNEL
);
6506 vmx_io_bitmap_b
= (unsigned long *)__get_free_page(GFP_KERNEL
);
6507 memset(vmx_vmread_bitmap
, 0xff, PAGE_SIZE
);
6508 memset(vmx_vmwrite_bitmap
, 0xff, PAGE_SIZE
);
6511 * Allow direct access to the PC debug port (it is often used for I/O
6512 * delays, but the vmexits simply slow things down).
6514 memset(vmx_io_bitmap_a
, 0xff, PAGE_SIZE
);
6515 clear_bit(0x80, vmx_io_bitmap_a
);
6517 memset(vmx_io_bitmap_b
, 0xff, PAGE_SIZE
);
6519 memset(vmx_msr_bitmap_legacy
, 0xff, PAGE_SIZE
);
6520 memset(vmx_msr_bitmap_longmode
, 0xff, PAGE_SIZE
);
6522 if (setup_vmcs_config(&vmcs_config
) < 0) {
6527 if (boot_cpu_has(X86_FEATURE_NX
))
6528 kvm_enable_efer_bits(EFER_NX
);
6530 if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
6531 !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
6534 if (!cpu_has_vmx_shadow_vmcs())
6535 enable_shadow_vmcs
= 0;
6536 if (enable_shadow_vmcs
)
6537 init_vmcs_shadow_fields();
6539 if (!cpu_has_vmx_ept() ||
6540 !cpu_has_vmx_ept_4levels()) {
6542 enable_unrestricted_guest
= 0;
6543 enable_ept_ad_bits
= 0;
6546 if (!cpu_has_vmx_ept_ad_bits() || !enable_ept
)
6547 enable_ept_ad_bits
= 0;
6549 if (!cpu_has_vmx_unrestricted_guest())
6550 enable_unrestricted_guest
= 0;
6552 if (!cpu_has_vmx_flexpriority())
6553 flexpriority_enabled
= 0;
6556 * set_apic_access_page_addr() is used to reload apic access
6557 * page upon invalidation. No need to do anything if not
6558 * using the APIC_ACCESS_ADDR VMCS field.
6560 if (!flexpriority_enabled
)
6561 kvm_x86_ops
->set_apic_access_page_addr
= NULL
;
6563 if (!cpu_has_vmx_tpr_shadow())
6564 kvm_x86_ops
->update_cr8_intercept
= NULL
;
6566 if (enable_ept
&& !cpu_has_vmx_ept_2m_page())
6567 kvm_disable_largepages();
6569 if (!cpu_has_vmx_ple())
6572 if (!cpu_has_vmx_apicv()) {
6574 kvm_x86_ops
->sync_pir_to_irr
= NULL
;
6577 if (cpu_has_vmx_tsc_scaling()) {
6578 kvm_has_tsc_control
= true;
6579 kvm_max_tsc_scaling_ratio
= KVM_VMX_TSC_MULTIPLIER_MAX
;
6580 kvm_tsc_scaling_ratio_frac_bits
= 48;
6583 vmx_disable_intercept_for_msr(MSR_FS_BASE
, false);
6584 vmx_disable_intercept_for_msr(MSR_GS_BASE
, false);
6585 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE
, true);
6586 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS
, false);
6587 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP
, false);
6588 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP
, false);
6590 memcpy(vmx_msr_bitmap_legacy_x2apic_apicv
,
6591 vmx_msr_bitmap_legacy
, PAGE_SIZE
);
6592 memcpy(vmx_msr_bitmap_longmode_x2apic_apicv
,
6593 vmx_msr_bitmap_longmode
, PAGE_SIZE
);
6594 memcpy(vmx_msr_bitmap_legacy_x2apic
,
6595 vmx_msr_bitmap_legacy
, PAGE_SIZE
);
6596 memcpy(vmx_msr_bitmap_longmode_x2apic
,
6597 vmx_msr_bitmap_longmode
, PAGE_SIZE
);
6599 set_bit(0, vmx_vpid_bitmap
); /* 0 is reserved for host */
6601 for (msr
= 0x800; msr
<= 0x8ff; msr
++) {
6602 if (msr
== 0x839 /* TMCCT */)
6604 vmx_disable_intercept_msr_x2apic(msr
, MSR_TYPE_R
, true);
6608 * TPR reads and writes can be virtualized even if virtual interrupt
6609 * delivery is not in use.
6611 vmx_disable_intercept_msr_x2apic(0x808, MSR_TYPE_W
, true);
6612 vmx_disable_intercept_msr_x2apic(0x808, MSR_TYPE_R
| MSR_TYPE_W
, false);
6615 vmx_disable_intercept_msr_x2apic(0x80b, MSR_TYPE_W
, true);
6617 vmx_disable_intercept_msr_x2apic(0x83f, MSR_TYPE_W
, true);
6624 update_ple_window_actual_max();
6627 * Only enable PML when hardware supports PML feature, and both EPT
6628 * and EPT A/D bit features are enabled -- PML depends on them to work.
6630 if (!enable_ept
|| !enable_ept_ad_bits
|| !cpu_has_vmx_pml())
6634 kvm_x86_ops
->slot_enable_log_dirty
= NULL
;
6635 kvm_x86_ops
->slot_disable_log_dirty
= NULL
;
6636 kvm_x86_ops
->flush_log_dirty
= NULL
;
6637 kvm_x86_ops
->enable_log_dirty_pt_masked
= NULL
;
6640 if (cpu_has_vmx_preemption_timer() && enable_preemption_timer
) {
6643 rdmsrl(MSR_IA32_VMX_MISC
, vmx_msr
);
6644 cpu_preemption_timer_multi
=
6645 vmx_msr
& VMX_MISC_PREEMPTION_TIMER_RATE_MASK
;
6647 kvm_x86_ops
->set_hv_timer
= NULL
;
6648 kvm_x86_ops
->cancel_hv_timer
= NULL
;
6651 kvm_set_posted_intr_wakeup_handler(wakeup_handler
);
6653 kvm_mce_cap_supported
|= MCG_LMCE_P
;
6655 return alloc_kvm_area();
6658 for (i
= 0; i
< VMX_BITMAP_NR
; i
++)
6659 free_page((unsigned long)vmx_bitmap
[i
]);
6664 static __exit
void hardware_unsetup(void)
6668 for (i
= 0; i
< VMX_BITMAP_NR
; i
++)
6669 free_page((unsigned long)vmx_bitmap
[i
]);
6675 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
6676 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
6678 static int handle_pause(struct kvm_vcpu
*vcpu
)
6681 grow_ple_window(vcpu
);
6683 kvm_vcpu_on_spin(vcpu
);
6684 return kvm_skip_emulated_instruction(vcpu
);
6687 static int handle_nop(struct kvm_vcpu
*vcpu
)
6689 return kvm_skip_emulated_instruction(vcpu
);
6692 static int handle_mwait(struct kvm_vcpu
*vcpu
)
6694 printk_once(KERN_WARNING
"kvm: MWAIT instruction emulated as NOP!\n");
6695 return handle_nop(vcpu
);
6698 static int handle_monitor_trap(struct kvm_vcpu
*vcpu
)
6703 static int handle_monitor(struct kvm_vcpu
*vcpu
)
6705 printk_once(KERN_WARNING
"kvm: MONITOR instruction emulated as NOP!\n");
6706 return handle_nop(vcpu
);
6710 * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
6711 * We could reuse a single VMCS for all the L2 guests, but we also want the
6712 * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
6713 * allows keeping them loaded on the processor, and in the future will allow
6714 * optimizations where prepare_vmcs02 doesn't need to set all the fields on
6715 * every entry if they never change.
6716 * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
6717 * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
6719 * The following functions allocate and free a vmcs02 in this pool.
6722 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
6723 static struct loaded_vmcs
*nested_get_current_vmcs02(struct vcpu_vmx
*vmx
)
6725 struct vmcs02_list
*item
;
6726 list_for_each_entry(item
, &vmx
->nested
.vmcs02_pool
, list
)
6727 if (item
->vmptr
== vmx
->nested
.current_vmptr
) {
6728 list_move(&item
->list
, &vmx
->nested
.vmcs02_pool
);
6729 return &item
->vmcs02
;
6732 if (vmx
->nested
.vmcs02_num
>= max(VMCS02_POOL_SIZE
, 1)) {
6733 /* Recycle the least recently used VMCS. */
6734 item
= list_last_entry(&vmx
->nested
.vmcs02_pool
,
6735 struct vmcs02_list
, list
);
6736 item
->vmptr
= vmx
->nested
.current_vmptr
;
6737 list_move(&item
->list
, &vmx
->nested
.vmcs02_pool
);
6738 return &item
->vmcs02
;
6741 /* Create a new VMCS */
6742 item
= kmalloc(sizeof(struct vmcs02_list
), GFP_KERNEL
);
6745 item
->vmcs02
.vmcs
= alloc_vmcs();
6746 item
->vmcs02
.shadow_vmcs
= NULL
;
6747 if (!item
->vmcs02
.vmcs
) {
6751 loaded_vmcs_init(&item
->vmcs02
);
6752 item
->vmptr
= vmx
->nested
.current_vmptr
;
6753 list_add(&(item
->list
), &(vmx
->nested
.vmcs02_pool
));
6754 vmx
->nested
.vmcs02_num
++;
6755 return &item
->vmcs02
;
6758 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
6759 static void nested_free_vmcs02(struct vcpu_vmx
*vmx
, gpa_t vmptr
)
6761 struct vmcs02_list
*item
;
6762 list_for_each_entry(item
, &vmx
->nested
.vmcs02_pool
, list
)
6763 if (item
->vmptr
== vmptr
) {
6764 free_loaded_vmcs(&item
->vmcs02
);
6765 list_del(&item
->list
);
6767 vmx
->nested
.vmcs02_num
--;
6773 * Free all VMCSs saved for this vcpu, except the one pointed by
6774 * vmx->loaded_vmcs. We must be running L1, so vmx->loaded_vmcs
6775 * must be &vmx->vmcs01.
6777 static void nested_free_all_saved_vmcss(struct vcpu_vmx
*vmx
)
6779 struct vmcs02_list
*item
, *n
;
6781 WARN_ON(vmx
->loaded_vmcs
!= &vmx
->vmcs01
);
6782 list_for_each_entry_safe(item
, n
, &vmx
->nested
.vmcs02_pool
, list
) {
6784 * Something will leak if the above WARN triggers. Better than
6787 if (vmx
->loaded_vmcs
== &item
->vmcs02
)
6790 free_loaded_vmcs(&item
->vmcs02
);
6791 list_del(&item
->list
);
6793 vmx
->nested
.vmcs02_num
--;
6798 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
6799 * set the success or error code of an emulated VMX instruction, as specified
6800 * by Vol 2B, VMX Instruction Reference, "Conventions".
6802 static void nested_vmx_succeed(struct kvm_vcpu
*vcpu
)
6804 vmx_set_rflags(vcpu
, vmx_get_rflags(vcpu
)
6805 & ~(X86_EFLAGS_CF
| X86_EFLAGS_PF
| X86_EFLAGS_AF
|
6806 X86_EFLAGS_ZF
| X86_EFLAGS_SF
| X86_EFLAGS_OF
));
6809 static void nested_vmx_failInvalid(struct kvm_vcpu
*vcpu
)
6811 vmx_set_rflags(vcpu
, (vmx_get_rflags(vcpu
)
6812 & ~(X86_EFLAGS_PF
| X86_EFLAGS_AF
| X86_EFLAGS_ZF
|
6813 X86_EFLAGS_SF
| X86_EFLAGS_OF
))
6817 static void nested_vmx_failValid(struct kvm_vcpu
*vcpu
,
6818 u32 vm_instruction_error
)
6820 if (to_vmx(vcpu
)->nested
.current_vmptr
== -1ull) {
6822 * failValid writes the error number to the current VMCS, which
6823 * can't be done there isn't a current VMCS.
6825 nested_vmx_failInvalid(vcpu
);
6828 vmx_set_rflags(vcpu
, (vmx_get_rflags(vcpu
)
6829 & ~(X86_EFLAGS_CF
| X86_EFLAGS_PF
| X86_EFLAGS_AF
|
6830 X86_EFLAGS_SF
| X86_EFLAGS_OF
))
6832 get_vmcs12(vcpu
)->vm_instruction_error
= vm_instruction_error
;
6834 * We don't need to force a shadow sync because
6835 * VM_INSTRUCTION_ERROR is not shadowed
6839 static void nested_vmx_abort(struct kvm_vcpu
*vcpu
, u32 indicator
)
6841 /* TODO: not to reset guest simply here. */
6842 kvm_make_request(KVM_REQ_TRIPLE_FAULT
, vcpu
);
6843 pr_debug_ratelimited("kvm: nested vmx abort, indicator %d\n", indicator
);
6846 static enum hrtimer_restart
vmx_preemption_timer_fn(struct hrtimer
*timer
)
6848 struct vcpu_vmx
*vmx
=
6849 container_of(timer
, struct vcpu_vmx
, nested
.preemption_timer
);
6851 vmx
->nested
.preemption_timer_expired
= true;
6852 kvm_make_request(KVM_REQ_EVENT
, &vmx
->vcpu
);
6853 kvm_vcpu_kick(&vmx
->vcpu
);
6855 return HRTIMER_NORESTART
;
6859 * Decode the memory-address operand of a vmx instruction, as recorded on an
6860 * exit caused by such an instruction (run by a guest hypervisor).
6861 * On success, returns 0. When the operand is invalid, returns 1 and throws
6864 static int get_vmx_mem_address(struct kvm_vcpu
*vcpu
,
6865 unsigned long exit_qualification
,
6866 u32 vmx_instruction_info
, bool wr
, gva_t
*ret
)
6870 struct kvm_segment s
;
6873 * According to Vol. 3B, "Information for VM Exits Due to Instruction
6874 * Execution", on an exit, vmx_instruction_info holds most of the
6875 * addressing components of the operand. Only the displacement part
6876 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
6877 * For how an actual address is calculated from all these components,
6878 * refer to Vol. 1, "Operand Addressing".
6880 int scaling
= vmx_instruction_info
& 3;
6881 int addr_size
= (vmx_instruction_info
>> 7) & 7;
6882 bool is_reg
= vmx_instruction_info
& (1u << 10);
6883 int seg_reg
= (vmx_instruction_info
>> 15) & 7;
6884 int index_reg
= (vmx_instruction_info
>> 18) & 0xf;
6885 bool index_is_valid
= !(vmx_instruction_info
& (1u << 22));
6886 int base_reg
= (vmx_instruction_info
>> 23) & 0xf;
6887 bool base_is_valid
= !(vmx_instruction_info
& (1u << 27));
6890 kvm_queue_exception(vcpu
, UD_VECTOR
);
6894 /* Addr = segment_base + offset */
6895 /* offset = base + [index * scale] + displacement */
6896 off
= exit_qualification
; /* holds the displacement */
6898 off
+= kvm_register_read(vcpu
, base_reg
);
6900 off
+= kvm_register_read(vcpu
, index_reg
)<<scaling
;
6901 vmx_get_segment(vcpu
, &s
, seg_reg
);
6902 *ret
= s
.base
+ off
;
6904 if (addr_size
== 1) /* 32 bit */
6907 /* Checks for #GP/#SS exceptions. */
6909 if (is_long_mode(vcpu
)) {
6910 /* Long mode: #GP(0)/#SS(0) if the memory address is in a
6911 * non-canonical form. This is the only check on the memory
6912 * destination for long mode!
6914 exn
= is_noncanonical_address(*ret
);
6915 } else if (is_protmode(vcpu
)) {
6916 /* Protected mode: apply checks for segment validity in the
6918 * - segment type check (#GP(0) may be thrown)
6919 * - usability check (#GP(0)/#SS(0))
6920 * - limit check (#GP(0)/#SS(0))
6923 /* #GP(0) if the destination operand is located in a
6924 * read-only data segment or any code segment.
6926 exn
= ((s
.type
& 0xa) == 0 || (s
.type
& 8));
6928 /* #GP(0) if the source operand is located in an
6929 * execute-only code segment
6931 exn
= ((s
.type
& 0xa) == 8);
6933 kvm_queue_exception_e(vcpu
, GP_VECTOR
, 0);
6936 /* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
6938 exn
= (s
.unusable
!= 0);
6939 /* Protected mode: #GP(0)/#SS(0) if the memory
6940 * operand is outside the segment limit.
6942 exn
= exn
|| (off
+ sizeof(u64
) > s
.limit
);
6945 kvm_queue_exception_e(vcpu
,
6946 seg_reg
== VCPU_SREG_SS
?
6947 SS_VECTOR
: GP_VECTOR
,
6955 static int nested_vmx_get_vmptr(struct kvm_vcpu
*vcpu
, gpa_t
*vmpointer
)
6958 struct x86_exception e
;
6960 if (get_vmx_mem_address(vcpu
, vmcs_readl(EXIT_QUALIFICATION
),
6961 vmcs_read32(VMX_INSTRUCTION_INFO
), false, &gva
))
6964 if (kvm_read_guest_virt(&vcpu
->arch
.emulate_ctxt
, gva
, vmpointer
,
6965 sizeof(*vmpointer
), &e
)) {
6966 kvm_inject_page_fault(vcpu
, &e
);
6973 static int enter_vmx_operation(struct kvm_vcpu
*vcpu
)
6975 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
6976 struct vmcs
*shadow_vmcs
;
6978 if (cpu_has_vmx_msr_bitmap()) {
6979 vmx
->nested
.msr_bitmap
=
6980 (unsigned long *)__get_free_page(GFP_KERNEL
);
6981 if (!vmx
->nested
.msr_bitmap
)
6982 goto out_msr_bitmap
;
6985 vmx
->nested
.cached_vmcs12
= kmalloc(VMCS12_SIZE
, GFP_KERNEL
);
6986 if (!vmx
->nested
.cached_vmcs12
)
6987 goto out_cached_vmcs12
;
6989 if (enable_shadow_vmcs
) {
6990 shadow_vmcs
= alloc_vmcs();
6992 goto out_shadow_vmcs
;
6993 /* mark vmcs as shadow */
6994 shadow_vmcs
->revision_id
|= (1u << 31);
6995 /* init shadow vmcs */
6996 vmcs_clear(shadow_vmcs
);
6997 vmx
->vmcs01
.shadow_vmcs
= shadow_vmcs
;
7000 INIT_LIST_HEAD(&(vmx
->nested
.vmcs02_pool
));
7001 vmx
->nested
.vmcs02_num
= 0;
7003 hrtimer_init(&vmx
->nested
.preemption_timer
, CLOCK_MONOTONIC
,
7004 HRTIMER_MODE_REL_PINNED
);
7005 vmx
->nested
.preemption_timer
.function
= vmx_preemption_timer_fn
;
7007 vmx
->nested
.vmxon
= true;
7011 kfree(vmx
->nested
.cached_vmcs12
);
7014 free_page((unsigned long)vmx
->nested
.msr_bitmap
);
7021 * Emulate the VMXON instruction.
7022 * Currently, we just remember that VMX is active, and do not save or even
7023 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
7024 * do not currently need to store anything in that guest-allocated memory
7025 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
7026 * argument is different from the VMXON pointer (which the spec says they do).
7028 static int handle_vmon(struct kvm_vcpu
*vcpu
)
7033 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
7034 const u64 VMXON_NEEDED_FEATURES
= FEATURE_CONTROL_LOCKED
7035 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX
;
7038 * The Intel VMX Instruction Reference lists a bunch of bits that are
7039 * prerequisite to running VMXON, most notably cr4.VMXE must be set to
7040 * 1 (see vmx_set_cr4() for when we allow the guest to set this).
7041 * Otherwise, we should fail with #UD. But most faulting conditions
7042 * have already been checked by hardware, prior to the VM-exit for
7043 * VMXON. We do test guest cr4.VMXE because processor CR4 always has
7044 * that bit set to 1 in non-root mode.
7046 if (!kvm_read_cr4_bits(vcpu
, X86_CR4_VMXE
)) {
7047 kvm_queue_exception(vcpu
, UD_VECTOR
);
7051 if (vmx
->nested
.vmxon
) {
7052 nested_vmx_failValid(vcpu
, VMXERR_VMXON_IN_VMX_ROOT_OPERATION
);
7053 return kvm_skip_emulated_instruction(vcpu
);
7056 if ((vmx
->msr_ia32_feature_control
& VMXON_NEEDED_FEATURES
)
7057 != VMXON_NEEDED_FEATURES
) {
7058 kvm_inject_gp(vcpu
, 0);
7062 if (nested_vmx_get_vmptr(vcpu
, &vmptr
))
7067 * The first 4 bytes of VMXON region contain the supported
7068 * VMCS revision identifier
7070 * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
7071 * which replaces physical address width with 32
7073 if (!PAGE_ALIGNED(vmptr
) || (vmptr
>> cpuid_maxphyaddr(vcpu
))) {
7074 nested_vmx_failInvalid(vcpu
);
7075 return kvm_skip_emulated_instruction(vcpu
);
7078 page
= nested_get_page(vcpu
, vmptr
);
7080 nested_vmx_failInvalid(vcpu
);
7081 return kvm_skip_emulated_instruction(vcpu
);
7083 if (*(u32
*)kmap(page
) != VMCS12_REVISION
) {
7085 nested_release_page_clean(page
);
7086 nested_vmx_failInvalid(vcpu
);
7087 return kvm_skip_emulated_instruction(vcpu
);
7090 nested_release_page_clean(page
);
7092 vmx
->nested
.vmxon_ptr
= vmptr
;
7093 ret
= enter_vmx_operation(vcpu
);
7097 nested_vmx_succeed(vcpu
);
7098 return kvm_skip_emulated_instruction(vcpu
);
7102 * Intel's VMX Instruction Reference specifies a common set of prerequisites
7103 * for running VMX instructions (except VMXON, whose prerequisites are
7104 * slightly different). It also specifies what exception to inject otherwise.
7105 * Note that many of these exceptions have priority over VM exits, so they
7106 * don't have to be checked again here.
7108 static int nested_vmx_check_permission(struct kvm_vcpu
*vcpu
)
7110 if (!to_vmx(vcpu
)->nested
.vmxon
) {
7111 kvm_queue_exception(vcpu
, UD_VECTOR
);
7117 static inline void nested_release_vmcs12(struct vcpu_vmx
*vmx
)
7119 if (vmx
->nested
.current_vmptr
== -1ull)
7122 /* current_vmptr and current_vmcs12 are always set/reset together */
7123 if (WARN_ON(vmx
->nested
.current_vmcs12
== NULL
))
7126 if (enable_shadow_vmcs
) {
7127 /* copy to memory all shadowed fields in case
7128 they were modified */
7129 copy_shadow_to_vmcs12(vmx
);
7130 vmx
->nested
.sync_shadow_vmcs
= false;
7131 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL
,
7132 SECONDARY_EXEC_SHADOW_VMCS
);
7133 vmcs_write64(VMCS_LINK_POINTER
, -1ull);
7135 vmx
->nested
.posted_intr_nv
= -1;
7137 /* Flush VMCS12 to guest memory */
7138 memcpy(vmx
->nested
.current_vmcs12
, vmx
->nested
.cached_vmcs12
,
7141 kunmap(vmx
->nested
.current_vmcs12_page
);
7142 nested_release_page(vmx
->nested
.current_vmcs12_page
);
7143 vmx
->nested
.current_vmptr
= -1ull;
7144 vmx
->nested
.current_vmcs12
= NULL
;
7148 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
7149 * just stops using VMX.
7151 static void free_nested(struct vcpu_vmx
*vmx
)
7153 if (!vmx
->nested
.vmxon
)
7156 vmx
->nested
.vmxon
= false;
7157 free_vpid(vmx
->nested
.vpid02
);
7158 nested_release_vmcs12(vmx
);
7159 if (vmx
->nested
.msr_bitmap
) {
7160 free_page((unsigned long)vmx
->nested
.msr_bitmap
);
7161 vmx
->nested
.msr_bitmap
= NULL
;
7163 if (enable_shadow_vmcs
) {
7164 vmcs_clear(vmx
->vmcs01
.shadow_vmcs
);
7165 free_vmcs(vmx
->vmcs01
.shadow_vmcs
);
7166 vmx
->vmcs01
.shadow_vmcs
= NULL
;
7168 kfree(vmx
->nested
.cached_vmcs12
);
7169 /* Unpin physical memory we referred to in current vmcs02 */
7170 if (vmx
->nested
.apic_access_page
) {
7171 nested_release_page(vmx
->nested
.apic_access_page
);
7172 vmx
->nested
.apic_access_page
= NULL
;
7174 if (vmx
->nested
.virtual_apic_page
) {
7175 nested_release_page(vmx
->nested
.virtual_apic_page
);
7176 vmx
->nested
.virtual_apic_page
= NULL
;
7178 if (vmx
->nested
.pi_desc_page
) {
7179 kunmap(vmx
->nested
.pi_desc_page
);
7180 nested_release_page(vmx
->nested
.pi_desc_page
);
7181 vmx
->nested
.pi_desc_page
= NULL
;
7182 vmx
->nested
.pi_desc
= NULL
;
7185 nested_free_all_saved_vmcss(vmx
);
7188 /* Emulate the VMXOFF instruction */
7189 static int handle_vmoff(struct kvm_vcpu
*vcpu
)
7191 if (!nested_vmx_check_permission(vcpu
))
7193 free_nested(to_vmx(vcpu
));
7194 nested_vmx_succeed(vcpu
);
7195 return kvm_skip_emulated_instruction(vcpu
);
7198 /* Emulate the VMCLEAR instruction */
7199 static int handle_vmclear(struct kvm_vcpu
*vcpu
)
7201 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
7205 if (!nested_vmx_check_permission(vcpu
))
7208 if (nested_vmx_get_vmptr(vcpu
, &vmptr
))
7211 if (!PAGE_ALIGNED(vmptr
) || (vmptr
>> cpuid_maxphyaddr(vcpu
))) {
7212 nested_vmx_failValid(vcpu
, VMXERR_VMCLEAR_INVALID_ADDRESS
);
7213 return kvm_skip_emulated_instruction(vcpu
);
7216 if (vmptr
== vmx
->nested
.vmxon_ptr
) {
7217 nested_vmx_failValid(vcpu
, VMXERR_VMCLEAR_VMXON_POINTER
);
7218 return kvm_skip_emulated_instruction(vcpu
);
7221 if (vmptr
== vmx
->nested
.current_vmptr
)
7222 nested_release_vmcs12(vmx
);
7224 kvm_vcpu_write_guest(vcpu
,
7225 vmptr
+ offsetof(struct vmcs12
, launch_state
),
7226 &zero
, sizeof(zero
));
7228 nested_free_vmcs02(vmx
, vmptr
);
7230 nested_vmx_succeed(vcpu
);
7231 return kvm_skip_emulated_instruction(vcpu
);
7234 static int nested_vmx_run(struct kvm_vcpu
*vcpu
, bool launch
);
7236 /* Emulate the VMLAUNCH instruction */
7237 static int handle_vmlaunch(struct kvm_vcpu
*vcpu
)
7239 return nested_vmx_run(vcpu
, true);
7242 /* Emulate the VMRESUME instruction */
7243 static int handle_vmresume(struct kvm_vcpu
*vcpu
)
7246 return nested_vmx_run(vcpu
, false);
7250 * Read a vmcs12 field. Since these can have varying lengths and we return
7251 * one type, we chose the biggest type (u64) and zero-extend the return value
7252 * to that size. Note that the caller, handle_vmread, might need to use only
7253 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
7254 * 64-bit fields are to be returned).
7256 static inline int vmcs12_read_any(struct kvm_vcpu
*vcpu
,
7257 unsigned long field
, u64
*ret
)
7259 short offset
= vmcs_field_to_offset(field
);
7265 p
= ((char *)(get_vmcs12(vcpu
))) + offset
;
7267 switch (vmcs_field_type(field
)) {
7268 case VMCS_FIELD_TYPE_NATURAL_WIDTH
:
7269 *ret
= *((natural_width
*)p
);
7271 case VMCS_FIELD_TYPE_U16
:
7274 case VMCS_FIELD_TYPE_U32
:
7277 case VMCS_FIELD_TYPE_U64
:
7287 static inline int vmcs12_write_any(struct kvm_vcpu
*vcpu
,
7288 unsigned long field
, u64 field_value
){
7289 short offset
= vmcs_field_to_offset(field
);
7290 char *p
= ((char *) get_vmcs12(vcpu
)) + offset
;
7294 switch (vmcs_field_type(field
)) {
7295 case VMCS_FIELD_TYPE_U16
:
7296 *(u16
*)p
= field_value
;
7298 case VMCS_FIELD_TYPE_U32
:
7299 *(u32
*)p
= field_value
;
7301 case VMCS_FIELD_TYPE_U64
:
7302 *(u64
*)p
= field_value
;
7304 case VMCS_FIELD_TYPE_NATURAL_WIDTH
:
7305 *(natural_width
*)p
= field_value
;
7314 static void copy_shadow_to_vmcs12(struct vcpu_vmx
*vmx
)
7317 unsigned long field
;
7319 struct vmcs
*shadow_vmcs
= vmx
->vmcs01
.shadow_vmcs
;
7320 const unsigned long *fields
= shadow_read_write_fields
;
7321 const int num_fields
= max_shadow_read_write_fields
;
7325 vmcs_load(shadow_vmcs
);
7327 for (i
= 0; i
< num_fields
; i
++) {
7329 switch (vmcs_field_type(field
)) {
7330 case VMCS_FIELD_TYPE_U16
:
7331 field_value
= vmcs_read16(field
);
7333 case VMCS_FIELD_TYPE_U32
:
7334 field_value
= vmcs_read32(field
);
7336 case VMCS_FIELD_TYPE_U64
:
7337 field_value
= vmcs_read64(field
);
7339 case VMCS_FIELD_TYPE_NATURAL_WIDTH
:
7340 field_value
= vmcs_readl(field
);
7346 vmcs12_write_any(&vmx
->vcpu
, field
, field_value
);
7349 vmcs_clear(shadow_vmcs
);
7350 vmcs_load(vmx
->loaded_vmcs
->vmcs
);
7355 static void copy_vmcs12_to_shadow(struct vcpu_vmx
*vmx
)
7357 const unsigned long *fields
[] = {
7358 shadow_read_write_fields
,
7359 shadow_read_only_fields
7361 const int max_fields
[] = {
7362 max_shadow_read_write_fields
,
7363 max_shadow_read_only_fields
7366 unsigned long field
;
7367 u64 field_value
= 0;
7368 struct vmcs
*shadow_vmcs
= vmx
->vmcs01
.shadow_vmcs
;
7370 vmcs_load(shadow_vmcs
);
7372 for (q
= 0; q
< ARRAY_SIZE(fields
); q
++) {
7373 for (i
= 0; i
< max_fields
[q
]; i
++) {
7374 field
= fields
[q
][i
];
7375 vmcs12_read_any(&vmx
->vcpu
, field
, &field_value
);
7377 switch (vmcs_field_type(field
)) {
7378 case VMCS_FIELD_TYPE_U16
:
7379 vmcs_write16(field
, (u16
)field_value
);
7381 case VMCS_FIELD_TYPE_U32
:
7382 vmcs_write32(field
, (u32
)field_value
);
7384 case VMCS_FIELD_TYPE_U64
:
7385 vmcs_write64(field
, (u64
)field_value
);
7387 case VMCS_FIELD_TYPE_NATURAL_WIDTH
:
7388 vmcs_writel(field
, (long)field_value
);
7397 vmcs_clear(shadow_vmcs
);
7398 vmcs_load(vmx
->loaded_vmcs
->vmcs
);
7402 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
7403 * used before) all generate the same failure when it is missing.
7405 static int nested_vmx_check_vmcs12(struct kvm_vcpu
*vcpu
)
7407 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
7408 if (vmx
->nested
.current_vmptr
== -1ull) {
7409 nested_vmx_failInvalid(vcpu
);
7415 static int handle_vmread(struct kvm_vcpu
*vcpu
)
7417 unsigned long field
;
7419 unsigned long exit_qualification
= vmcs_readl(EXIT_QUALIFICATION
);
7420 u32 vmx_instruction_info
= vmcs_read32(VMX_INSTRUCTION_INFO
);
7423 if (!nested_vmx_check_permission(vcpu
))
7426 if (!nested_vmx_check_vmcs12(vcpu
))
7427 return kvm_skip_emulated_instruction(vcpu
);
7429 /* Decode instruction info and find the field to read */
7430 field
= kvm_register_readl(vcpu
, (((vmx_instruction_info
) >> 28) & 0xf));
7431 /* Read the field, zero-extended to a u64 field_value */
7432 if (vmcs12_read_any(vcpu
, field
, &field_value
) < 0) {
7433 nested_vmx_failValid(vcpu
, VMXERR_UNSUPPORTED_VMCS_COMPONENT
);
7434 return kvm_skip_emulated_instruction(vcpu
);
7437 * Now copy part of this value to register or memory, as requested.
7438 * Note that the number of bits actually copied is 32 or 64 depending
7439 * on the guest's mode (32 or 64 bit), not on the given field's length.
7441 if (vmx_instruction_info
& (1u << 10)) {
7442 kvm_register_writel(vcpu
, (((vmx_instruction_info
) >> 3) & 0xf),
7445 if (get_vmx_mem_address(vcpu
, exit_qualification
,
7446 vmx_instruction_info
, true, &gva
))
7448 /* _system ok, as hardware has verified cpl=0 */
7449 kvm_write_guest_virt_system(&vcpu
->arch
.emulate_ctxt
, gva
,
7450 &field_value
, (is_long_mode(vcpu
) ? 8 : 4), NULL
);
7453 nested_vmx_succeed(vcpu
);
7454 return kvm_skip_emulated_instruction(vcpu
);
7458 static int handle_vmwrite(struct kvm_vcpu
*vcpu
)
7460 unsigned long field
;
7462 unsigned long exit_qualification
= vmcs_readl(EXIT_QUALIFICATION
);
7463 u32 vmx_instruction_info
= vmcs_read32(VMX_INSTRUCTION_INFO
);
7464 /* The value to write might be 32 or 64 bits, depending on L1's long
7465 * mode, and eventually we need to write that into a field of several
7466 * possible lengths. The code below first zero-extends the value to 64
7467 * bit (field_value), and then copies only the appropriate number of
7468 * bits into the vmcs12 field.
7470 u64 field_value
= 0;
7471 struct x86_exception e
;
7473 if (!nested_vmx_check_permission(vcpu
))
7476 if (!nested_vmx_check_vmcs12(vcpu
))
7477 return kvm_skip_emulated_instruction(vcpu
);
7479 if (vmx_instruction_info
& (1u << 10))
7480 field_value
= kvm_register_readl(vcpu
,
7481 (((vmx_instruction_info
) >> 3) & 0xf));
7483 if (get_vmx_mem_address(vcpu
, exit_qualification
,
7484 vmx_instruction_info
, false, &gva
))
7486 if (kvm_read_guest_virt(&vcpu
->arch
.emulate_ctxt
, gva
,
7487 &field_value
, (is_64_bit_mode(vcpu
) ? 8 : 4), &e
)) {
7488 kvm_inject_page_fault(vcpu
, &e
);
7494 field
= kvm_register_readl(vcpu
, (((vmx_instruction_info
) >> 28) & 0xf));
7495 if (vmcs_field_readonly(field
)) {
7496 nested_vmx_failValid(vcpu
,
7497 VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT
);
7498 return kvm_skip_emulated_instruction(vcpu
);
7501 if (vmcs12_write_any(vcpu
, field
, field_value
) < 0) {
7502 nested_vmx_failValid(vcpu
, VMXERR_UNSUPPORTED_VMCS_COMPONENT
);
7503 return kvm_skip_emulated_instruction(vcpu
);
7506 nested_vmx_succeed(vcpu
);
7507 return kvm_skip_emulated_instruction(vcpu
);
7510 static void set_current_vmptr(struct vcpu_vmx
*vmx
, gpa_t vmptr
)
7512 vmx
->nested
.current_vmptr
= vmptr
;
7513 if (enable_shadow_vmcs
) {
7514 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL
,
7515 SECONDARY_EXEC_SHADOW_VMCS
);
7516 vmcs_write64(VMCS_LINK_POINTER
,
7517 __pa(vmx
->vmcs01
.shadow_vmcs
));
7518 vmx
->nested
.sync_shadow_vmcs
= true;
7522 /* Emulate the VMPTRLD instruction */
7523 static int handle_vmptrld(struct kvm_vcpu
*vcpu
)
7525 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
7528 if (!nested_vmx_check_permission(vcpu
))
7531 if (nested_vmx_get_vmptr(vcpu
, &vmptr
))
7534 if (!PAGE_ALIGNED(vmptr
) || (vmptr
>> cpuid_maxphyaddr(vcpu
))) {
7535 nested_vmx_failValid(vcpu
, VMXERR_VMPTRLD_INVALID_ADDRESS
);
7536 return kvm_skip_emulated_instruction(vcpu
);
7539 if (vmptr
== vmx
->nested
.vmxon_ptr
) {
7540 nested_vmx_failValid(vcpu
, VMXERR_VMPTRLD_VMXON_POINTER
);
7541 return kvm_skip_emulated_instruction(vcpu
);
7544 if (vmx
->nested
.current_vmptr
!= vmptr
) {
7545 struct vmcs12
*new_vmcs12
;
7547 page
= nested_get_page(vcpu
, vmptr
);
7549 nested_vmx_failInvalid(vcpu
);
7550 return kvm_skip_emulated_instruction(vcpu
);
7552 new_vmcs12
= kmap(page
);
7553 if (new_vmcs12
->revision_id
!= VMCS12_REVISION
) {
7555 nested_release_page_clean(page
);
7556 nested_vmx_failValid(vcpu
,
7557 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID
);
7558 return kvm_skip_emulated_instruction(vcpu
);
7561 nested_release_vmcs12(vmx
);
7562 vmx
->nested
.current_vmcs12
= new_vmcs12
;
7563 vmx
->nested
.current_vmcs12_page
= page
;
7565 * Load VMCS12 from guest memory since it is not already
7568 memcpy(vmx
->nested
.cached_vmcs12
,
7569 vmx
->nested
.current_vmcs12
, VMCS12_SIZE
);
7570 set_current_vmptr(vmx
, vmptr
);
7573 nested_vmx_succeed(vcpu
);
7574 return kvm_skip_emulated_instruction(vcpu
);
7577 /* Emulate the VMPTRST instruction */
7578 static int handle_vmptrst(struct kvm_vcpu
*vcpu
)
7580 unsigned long exit_qualification
= vmcs_readl(EXIT_QUALIFICATION
);
7581 u32 vmx_instruction_info
= vmcs_read32(VMX_INSTRUCTION_INFO
);
7583 struct x86_exception e
;
7585 if (!nested_vmx_check_permission(vcpu
))
7588 if (get_vmx_mem_address(vcpu
, exit_qualification
,
7589 vmx_instruction_info
, true, &vmcs_gva
))
7591 /* ok to use *_system, as hardware has verified cpl=0 */
7592 if (kvm_write_guest_virt_system(&vcpu
->arch
.emulate_ctxt
, vmcs_gva
,
7593 (void *)&to_vmx(vcpu
)->nested
.current_vmptr
,
7595 kvm_inject_page_fault(vcpu
, &e
);
7598 nested_vmx_succeed(vcpu
);
7599 return kvm_skip_emulated_instruction(vcpu
);
7602 /* Emulate the INVEPT instruction */
7603 static int handle_invept(struct kvm_vcpu
*vcpu
)
7605 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
7606 u32 vmx_instruction_info
, types
;
7609 struct x86_exception e
;
7614 if (!(vmx
->nested
.nested_vmx_secondary_ctls_high
&
7615 SECONDARY_EXEC_ENABLE_EPT
) ||
7616 !(vmx
->nested
.nested_vmx_ept_caps
& VMX_EPT_INVEPT_BIT
)) {
7617 kvm_queue_exception(vcpu
, UD_VECTOR
);
7621 if (!nested_vmx_check_permission(vcpu
))
7624 vmx_instruction_info
= vmcs_read32(VMX_INSTRUCTION_INFO
);
7625 type
= kvm_register_readl(vcpu
, (vmx_instruction_info
>> 28) & 0xf);
7627 types
= (vmx
->nested
.nested_vmx_ept_caps
>> VMX_EPT_EXTENT_SHIFT
) & 6;
7629 if (type
>= 32 || !(types
& (1 << type
))) {
7630 nested_vmx_failValid(vcpu
,
7631 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID
);
7632 return kvm_skip_emulated_instruction(vcpu
);
7635 /* According to the Intel VMX instruction reference, the memory
7636 * operand is read even if it isn't needed (e.g., for type==global)
7638 if (get_vmx_mem_address(vcpu
, vmcs_readl(EXIT_QUALIFICATION
),
7639 vmx_instruction_info
, false, &gva
))
7641 if (kvm_read_guest_virt(&vcpu
->arch
.emulate_ctxt
, gva
, &operand
,
7642 sizeof(operand
), &e
)) {
7643 kvm_inject_page_fault(vcpu
, &e
);
7648 case VMX_EPT_EXTENT_GLOBAL
:
7650 * TODO: track mappings and invalidate
7651 * single context requests appropriately
7653 case VMX_EPT_EXTENT_CONTEXT
:
7654 kvm_mmu_sync_roots(vcpu
);
7655 kvm_make_request(KVM_REQ_TLB_FLUSH
, vcpu
);
7656 nested_vmx_succeed(vcpu
);
7663 return kvm_skip_emulated_instruction(vcpu
);
7666 static int handle_invvpid(struct kvm_vcpu
*vcpu
)
7668 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
7669 u32 vmx_instruction_info
;
7670 unsigned long type
, types
;
7672 struct x86_exception e
;
7678 if (!(vmx
->nested
.nested_vmx_secondary_ctls_high
&
7679 SECONDARY_EXEC_ENABLE_VPID
) ||
7680 !(vmx
->nested
.nested_vmx_vpid_caps
& VMX_VPID_INVVPID_BIT
)) {
7681 kvm_queue_exception(vcpu
, UD_VECTOR
);
7685 if (!nested_vmx_check_permission(vcpu
))
7688 vmx_instruction_info
= vmcs_read32(VMX_INSTRUCTION_INFO
);
7689 type
= kvm_register_readl(vcpu
, (vmx_instruction_info
>> 28) & 0xf);
7691 types
= (vmx
->nested
.nested_vmx_vpid_caps
&
7692 VMX_VPID_EXTENT_SUPPORTED_MASK
) >> 8;
7694 if (type
>= 32 || !(types
& (1 << type
))) {
7695 nested_vmx_failValid(vcpu
,
7696 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID
);
7697 return kvm_skip_emulated_instruction(vcpu
);
7700 /* according to the intel vmx instruction reference, the memory
7701 * operand is read even if it isn't needed (e.g., for type==global)
7703 if (get_vmx_mem_address(vcpu
, vmcs_readl(EXIT_QUALIFICATION
),
7704 vmx_instruction_info
, false, &gva
))
7706 if (kvm_read_guest_virt(&vcpu
->arch
.emulate_ctxt
, gva
, &operand
,
7707 sizeof(operand
), &e
)) {
7708 kvm_inject_page_fault(vcpu
, &e
);
7711 if (operand
.vpid
>> 16) {
7712 nested_vmx_failValid(vcpu
,
7713 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID
);
7714 return kvm_skip_emulated_instruction(vcpu
);
7718 case VMX_VPID_EXTENT_INDIVIDUAL_ADDR
:
7719 if (is_noncanonical_address(operand
.gla
)) {
7720 nested_vmx_failValid(vcpu
,
7721 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID
);
7722 return kvm_skip_emulated_instruction(vcpu
);
7725 case VMX_VPID_EXTENT_SINGLE_CONTEXT
:
7726 case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL
:
7727 if (!operand
.vpid
) {
7728 nested_vmx_failValid(vcpu
,
7729 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID
);
7730 return kvm_skip_emulated_instruction(vcpu
);
7733 case VMX_VPID_EXTENT_ALL_CONTEXT
:
7737 return kvm_skip_emulated_instruction(vcpu
);
7740 __vmx_flush_tlb(vcpu
, vmx
->nested
.vpid02
);
7741 nested_vmx_succeed(vcpu
);
7743 return kvm_skip_emulated_instruction(vcpu
);
7746 static int handle_pml_full(struct kvm_vcpu
*vcpu
)
7748 unsigned long exit_qualification
;
7750 trace_kvm_pml_full(vcpu
->vcpu_id
);
7752 exit_qualification
= vmcs_readl(EXIT_QUALIFICATION
);
7755 * PML buffer FULL happened while executing iret from NMI,
7756 * "blocked by NMI" bit has to be set before next VM entry.
7758 if (!(to_vmx(vcpu
)->idt_vectoring_info
& VECTORING_INFO_VALID_MASK
) &&
7759 (exit_qualification
& INTR_INFO_UNBLOCK_NMI
))
7760 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO
,
7761 GUEST_INTR_STATE_NMI
);
7764 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
7765 * here.., and there's no userspace involvement needed for PML.
7770 static int handle_preemption_timer(struct kvm_vcpu
*vcpu
)
7772 kvm_lapic_expired_hv_timer(vcpu
);
7777 * The exit handlers return 1 if the exit was handled fully and guest execution
7778 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
7779 * to be done to userspace and return 0.
7781 static int (*const kvm_vmx_exit_handlers
[])(struct kvm_vcpu
*vcpu
) = {
7782 [EXIT_REASON_EXCEPTION_NMI
] = handle_exception
,
7783 [EXIT_REASON_EXTERNAL_INTERRUPT
] = handle_external_interrupt
,
7784 [EXIT_REASON_TRIPLE_FAULT
] = handle_triple_fault
,
7785 [EXIT_REASON_NMI_WINDOW
] = handle_nmi_window
,
7786 [EXIT_REASON_IO_INSTRUCTION
] = handle_io
,
7787 [EXIT_REASON_CR_ACCESS
] = handle_cr
,
7788 [EXIT_REASON_DR_ACCESS
] = handle_dr
,
7789 [EXIT_REASON_CPUID
] = handle_cpuid
,
7790 [EXIT_REASON_MSR_READ
] = handle_rdmsr
,
7791 [EXIT_REASON_MSR_WRITE
] = handle_wrmsr
,
7792 [EXIT_REASON_PENDING_INTERRUPT
] = handle_interrupt_window
,
7793 [EXIT_REASON_HLT
] = handle_halt
,
7794 [EXIT_REASON_INVD
] = handle_invd
,
7795 [EXIT_REASON_INVLPG
] = handle_invlpg
,
7796 [EXIT_REASON_RDPMC
] = handle_rdpmc
,
7797 [EXIT_REASON_VMCALL
] = handle_vmcall
,
7798 [EXIT_REASON_VMCLEAR
] = handle_vmclear
,
7799 [EXIT_REASON_VMLAUNCH
] = handle_vmlaunch
,
7800 [EXIT_REASON_VMPTRLD
] = handle_vmptrld
,
7801 [EXIT_REASON_VMPTRST
] = handle_vmptrst
,
7802 [EXIT_REASON_VMREAD
] = handle_vmread
,
7803 [EXIT_REASON_VMRESUME
] = handle_vmresume
,
7804 [EXIT_REASON_VMWRITE
] = handle_vmwrite
,
7805 [EXIT_REASON_VMOFF
] = handle_vmoff
,
7806 [EXIT_REASON_VMON
] = handle_vmon
,
7807 [EXIT_REASON_TPR_BELOW_THRESHOLD
] = handle_tpr_below_threshold
,
7808 [EXIT_REASON_APIC_ACCESS
] = handle_apic_access
,
7809 [EXIT_REASON_APIC_WRITE
] = handle_apic_write
,
7810 [EXIT_REASON_EOI_INDUCED
] = handle_apic_eoi_induced
,
7811 [EXIT_REASON_WBINVD
] = handle_wbinvd
,
7812 [EXIT_REASON_XSETBV
] = handle_xsetbv
,
7813 [EXIT_REASON_TASK_SWITCH
] = handle_task_switch
,
7814 [EXIT_REASON_MCE_DURING_VMENTRY
] = handle_machine_check
,
7815 [EXIT_REASON_EPT_VIOLATION
] = handle_ept_violation
,
7816 [EXIT_REASON_EPT_MISCONFIG
] = handle_ept_misconfig
,
7817 [EXIT_REASON_PAUSE_INSTRUCTION
] = handle_pause
,
7818 [EXIT_REASON_MWAIT_INSTRUCTION
] = handle_mwait
,
7819 [EXIT_REASON_MONITOR_TRAP_FLAG
] = handle_monitor_trap
,
7820 [EXIT_REASON_MONITOR_INSTRUCTION
] = handle_monitor
,
7821 [EXIT_REASON_INVEPT
] = handle_invept
,
7822 [EXIT_REASON_INVVPID
] = handle_invvpid
,
7823 [EXIT_REASON_XSAVES
] = handle_xsaves
,
7824 [EXIT_REASON_XRSTORS
] = handle_xrstors
,
7825 [EXIT_REASON_PML_FULL
] = handle_pml_full
,
7826 [EXIT_REASON_PREEMPTION_TIMER
] = handle_preemption_timer
,
7829 static const int kvm_vmx_max_exit_handlers
=
7830 ARRAY_SIZE(kvm_vmx_exit_handlers
);
7832 static bool nested_vmx_exit_handled_io(struct kvm_vcpu
*vcpu
,
7833 struct vmcs12
*vmcs12
)
7835 unsigned long exit_qualification
;
7836 gpa_t bitmap
, last_bitmap
;
7841 if (!nested_cpu_has(vmcs12
, CPU_BASED_USE_IO_BITMAPS
))
7842 return nested_cpu_has(vmcs12
, CPU_BASED_UNCOND_IO_EXITING
);
7844 exit_qualification
= vmcs_readl(EXIT_QUALIFICATION
);
7846 port
= exit_qualification
>> 16;
7847 size
= (exit_qualification
& 7) + 1;
7849 last_bitmap
= (gpa_t
)-1;
7854 bitmap
= vmcs12
->io_bitmap_a
;
7855 else if (port
< 0x10000)
7856 bitmap
= vmcs12
->io_bitmap_b
;
7859 bitmap
+= (port
& 0x7fff) / 8;
7861 if (last_bitmap
!= bitmap
)
7862 if (kvm_vcpu_read_guest(vcpu
, bitmap
, &b
, 1))
7864 if (b
& (1 << (port
& 7)))
7869 last_bitmap
= bitmap
;
7876 * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
7877 * rather than handle it ourselves in L0. I.e., check whether L1 expressed
7878 * disinterest in the current event (read or write a specific MSR) by using an
7879 * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
7881 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu
*vcpu
,
7882 struct vmcs12
*vmcs12
, u32 exit_reason
)
7884 u32 msr_index
= vcpu
->arch
.regs
[VCPU_REGS_RCX
];
7887 if (!nested_cpu_has(vmcs12
, CPU_BASED_USE_MSR_BITMAPS
))
7891 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
7892 * for the four combinations of read/write and low/high MSR numbers.
7893 * First we need to figure out which of the four to use:
7895 bitmap
= vmcs12
->msr_bitmap
;
7896 if (exit_reason
== EXIT_REASON_MSR_WRITE
)
7898 if (msr_index
>= 0xc0000000) {
7899 msr_index
-= 0xc0000000;
7903 /* Then read the msr_index'th bit from this bitmap: */
7904 if (msr_index
< 1024*8) {
7906 if (kvm_vcpu_read_guest(vcpu
, bitmap
+ msr_index
/8, &b
, 1))
7908 return 1 & (b
>> (msr_index
& 7));
7910 return true; /* let L1 handle the wrong parameter */
7914 * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
7915 * rather than handle it ourselves in L0. I.e., check if L1 wanted to
7916 * intercept (via guest_host_mask etc.) the current event.
7918 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu
*vcpu
,
7919 struct vmcs12
*vmcs12
)
7921 unsigned long exit_qualification
= vmcs_readl(EXIT_QUALIFICATION
);
7922 int cr
= exit_qualification
& 15;
7926 switch ((exit_qualification
>> 4) & 3) {
7927 case 0: /* mov to cr */
7928 reg
= (exit_qualification
>> 8) & 15;
7929 val
= kvm_register_readl(vcpu
, reg
);
7932 if (vmcs12
->cr0_guest_host_mask
&
7933 (val
^ vmcs12
->cr0_read_shadow
))
7937 if ((vmcs12
->cr3_target_count
>= 1 &&
7938 vmcs12
->cr3_target_value0
== val
) ||
7939 (vmcs12
->cr3_target_count
>= 2 &&
7940 vmcs12
->cr3_target_value1
== val
) ||
7941 (vmcs12
->cr3_target_count
>= 3 &&
7942 vmcs12
->cr3_target_value2
== val
) ||
7943 (vmcs12
->cr3_target_count
>= 4 &&
7944 vmcs12
->cr3_target_value3
== val
))
7946 if (nested_cpu_has(vmcs12
, CPU_BASED_CR3_LOAD_EXITING
))
7950 if (vmcs12
->cr4_guest_host_mask
&
7951 (vmcs12
->cr4_read_shadow
^ val
))
7955 if (nested_cpu_has(vmcs12
, CPU_BASED_CR8_LOAD_EXITING
))
7961 if ((vmcs12
->cr0_guest_host_mask
& X86_CR0_TS
) &&
7962 (vmcs12
->cr0_read_shadow
& X86_CR0_TS
))
7965 case 1: /* mov from cr */
7968 if (vmcs12
->cpu_based_vm_exec_control
&
7969 CPU_BASED_CR3_STORE_EXITING
)
7973 if (vmcs12
->cpu_based_vm_exec_control
&
7974 CPU_BASED_CR8_STORE_EXITING
)
7981 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
7982 * cr0. Other attempted changes are ignored, with no exit.
7984 val
= (exit_qualification
>> LMSW_SOURCE_DATA_SHIFT
) & 0x0f;
7985 if (vmcs12
->cr0_guest_host_mask
& 0xe &
7986 (val
^ vmcs12
->cr0_read_shadow
))
7988 if ((vmcs12
->cr0_guest_host_mask
& 0x1) &&
7989 !(vmcs12
->cr0_read_shadow
& 0x1) &&
7998 * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
7999 * should handle it ourselves in L0 (and then continue L2). Only call this
8000 * when in is_guest_mode (L2).
8002 static bool nested_vmx_exit_handled(struct kvm_vcpu
*vcpu
)
8004 u32 intr_info
= vmcs_read32(VM_EXIT_INTR_INFO
);
8005 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
8006 struct vmcs12
*vmcs12
= get_vmcs12(vcpu
);
8007 u32 exit_reason
= vmx
->exit_reason
;
8009 trace_kvm_nested_vmexit(kvm_rip_read(vcpu
), exit_reason
,
8010 vmcs_readl(EXIT_QUALIFICATION
),
8011 vmx
->idt_vectoring_info
,
8013 vmcs_read32(VM_EXIT_INTR_ERROR_CODE
),
8016 if (vmx
->nested
.nested_run_pending
)
8019 if (unlikely(vmx
->fail
)) {
8020 pr_info_ratelimited("%s failed vm entry %x\n", __func__
,
8021 vmcs_read32(VM_INSTRUCTION_ERROR
));
8025 switch (exit_reason
) {
8026 case EXIT_REASON_EXCEPTION_NMI
:
8027 if (is_nmi(intr_info
))
8029 else if (is_page_fault(intr_info
))
8031 else if (is_no_device(intr_info
) &&
8032 !(vmcs12
->guest_cr0
& X86_CR0_TS
))
8034 else if (is_debug(intr_info
) &&
8036 (KVM_GUESTDBG_SINGLESTEP
| KVM_GUESTDBG_USE_HW_BP
))
8038 else if (is_breakpoint(intr_info
) &&
8039 vcpu
->guest_debug
& KVM_GUESTDBG_USE_SW_BP
)
8041 return vmcs12
->exception_bitmap
&
8042 (1u << (intr_info
& INTR_INFO_VECTOR_MASK
));
8043 case EXIT_REASON_EXTERNAL_INTERRUPT
:
8045 case EXIT_REASON_TRIPLE_FAULT
:
8047 case EXIT_REASON_PENDING_INTERRUPT
:
8048 return nested_cpu_has(vmcs12
, CPU_BASED_VIRTUAL_INTR_PENDING
);
8049 case EXIT_REASON_NMI_WINDOW
:
8050 return nested_cpu_has(vmcs12
, CPU_BASED_VIRTUAL_NMI_PENDING
);
8051 case EXIT_REASON_TASK_SWITCH
:
8053 case EXIT_REASON_CPUID
:
8055 case EXIT_REASON_HLT
:
8056 return nested_cpu_has(vmcs12
, CPU_BASED_HLT_EXITING
);
8057 case EXIT_REASON_INVD
:
8059 case EXIT_REASON_INVLPG
:
8060 return nested_cpu_has(vmcs12
, CPU_BASED_INVLPG_EXITING
);
8061 case EXIT_REASON_RDPMC
:
8062 return nested_cpu_has(vmcs12
, CPU_BASED_RDPMC_EXITING
);
8063 case EXIT_REASON_RDRAND
:
8064 return nested_cpu_has2(vmcs12
, SECONDARY_EXEC_RDRAND
);
8065 case EXIT_REASON_RDSEED
:
8066 return nested_cpu_has2(vmcs12
, SECONDARY_EXEC_RDSEED
);
8067 case EXIT_REASON_RDTSC
: case EXIT_REASON_RDTSCP
:
8068 return nested_cpu_has(vmcs12
, CPU_BASED_RDTSC_EXITING
);
8069 case EXIT_REASON_VMCALL
: case EXIT_REASON_VMCLEAR
:
8070 case EXIT_REASON_VMLAUNCH
: case EXIT_REASON_VMPTRLD
:
8071 case EXIT_REASON_VMPTRST
: case EXIT_REASON_VMREAD
:
8072 case EXIT_REASON_VMRESUME
: case EXIT_REASON_VMWRITE
:
8073 case EXIT_REASON_VMOFF
: case EXIT_REASON_VMON
:
8074 case EXIT_REASON_INVEPT
: case EXIT_REASON_INVVPID
:
8076 * VMX instructions trap unconditionally. This allows L1 to
8077 * emulate them for its L2 guest, i.e., allows 3-level nesting!
8080 case EXIT_REASON_CR_ACCESS
:
8081 return nested_vmx_exit_handled_cr(vcpu
, vmcs12
);
8082 case EXIT_REASON_DR_ACCESS
:
8083 return nested_cpu_has(vmcs12
, CPU_BASED_MOV_DR_EXITING
);
8084 case EXIT_REASON_IO_INSTRUCTION
:
8085 return nested_vmx_exit_handled_io(vcpu
, vmcs12
);
8086 case EXIT_REASON_GDTR_IDTR
: case EXIT_REASON_LDTR_TR
:
8087 return nested_cpu_has2(vmcs12
, SECONDARY_EXEC_DESC
);
8088 case EXIT_REASON_MSR_READ
:
8089 case EXIT_REASON_MSR_WRITE
:
8090 return nested_vmx_exit_handled_msr(vcpu
, vmcs12
, exit_reason
);
8091 case EXIT_REASON_INVALID_STATE
:
8093 case EXIT_REASON_MWAIT_INSTRUCTION
:
8094 return nested_cpu_has(vmcs12
, CPU_BASED_MWAIT_EXITING
);
8095 case EXIT_REASON_MONITOR_TRAP_FLAG
:
8096 return nested_cpu_has(vmcs12
, CPU_BASED_MONITOR_TRAP_FLAG
);
8097 case EXIT_REASON_MONITOR_INSTRUCTION
:
8098 return nested_cpu_has(vmcs12
, CPU_BASED_MONITOR_EXITING
);
8099 case EXIT_REASON_PAUSE_INSTRUCTION
:
8100 return nested_cpu_has(vmcs12
, CPU_BASED_PAUSE_EXITING
) ||
8101 nested_cpu_has2(vmcs12
,
8102 SECONDARY_EXEC_PAUSE_LOOP_EXITING
);
8103 case EXIT_REASON_MCE_DURING_VMENTRY
:
8105 case EXIT_REASON_TPR_BELOW_THRESHOLD
:
8106 return nested_cpu_has(vmcs12
, CPU_BASED_TPR_SHADOW
);
8107 case EXIT_REASON_APIC_ACCESS
:
8108 return nested_cpu_has2(vmcs12
,
8109 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES
);
8110 case EXIT_REASON_APIC_WRITE
:
8111 case EXIT_REASON_EOI_INDUCED
:
8112 /* apic_write and eoi_induced should exit unconditionally. */
8114 case EXIT_REASON_EPT_VIOLATION
:
8116 * L0 always deals with the EPT violation. If nested EPT is
8117 * used, and the nested mmu code discovers that the address is
8118 * missing in the guest EPT table (EPT12), the EPT violation
8119 * will be injected with nested_ept_inject_page_fault()
8122 case EXIT_REASON_EPT_MISCONFIG
:
8124 * L2 never uses directly L1's EPT, but rather L0's own EPT
8125 * table (shadow on EPT) or a merged EPT table that L0 built
8126 * (EPT on EPT). So any problems with the structure of the
8127 * table is L0's fault.
8130 case EXIT_REASON_WBINVD
:
8131 return nested_cpu_has2(vmcs12
, SECONDARY_EXEC_WBINVD_EXITING
);
8132 case EXIT_REASON_XSETBV
:
8134 case EXIT_REASON_XSAVES
: case EXIT_REASON_XRSTORS
:
8136 * This should never happen, since it is not possible to
8137 * set XSS to a non-zero value---neither in L1 nor in L2.
8138 * If if it were, XSS would have to be checked against
8139 * the XSS exit bitmap in vmcs12.
8141 return nested_cpu_has2(vmcs12
, SECONDARY_EXEC_XSAVES
);
8142 case EXIT_REASON_PREEMPTION_TIMER
:
8144 case EXIT_REASON_PML_FULL
:
8145 /* We emulate PML support to L1. */
8152 static void vmx_get_exit_info(struct kvm_vcpu
*vcpu
, u64
*info1
, u64
*info2
)
8154 *info1
= vmcs_readl(EXIT_QUALIFICATION
);
8155 *info2
= vmcs_read32(VM_EXIT_INTR_INFO
);
8158 static void vmx_destroy_pml_buffer(struct vcpu_vmx
*vmx
)
8161 __free_page(vmx
->pml_pg
);
8166 static void vmx_flush_pml_buffer(struct kvm_vcpu
*vcpu
)
8168 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
8172 pml_idx
= vmcs_read16(GUEST_PML_INDEX
);
8174 /* Do nothing if PML buffer is empty */
8175 if (pml_idx
== (PML_ENTITY_NUM
- 1))
8178 /* PML index always points to next available PML buffer entity */
8179 if (pml_idx
>= PML_ENTITY_NUM
)
8184 pml_buf
= page_address(vmx
->pml_pg
);
8185 for (; pml_idx
< PML_ENTITY_NUM
; pml_idx
++) {
8188 gpa
= pml_buf
[pml_idx
];
8189 WARN_ON(gpa
& (PAGE_SIZE
- 1));
8190 kvm_vcpu_mark_page_dirty(vcpu
, gpa
>> PAGE_SHIFT
);
8193 /* reset PML index */
8194 vmcs_write16(GUEST_PML_INDEX
, PML_ENTITY_NUM
- 1);
8198 * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
8199 * Called before reporting dirty_bitmap to userspace.
8201 static void kvm_flush_pml_buffers(struct kvm
*kvm
)
8204 struct kvm_vcpu
*vcpu
;
8206 * We only need to kick vcpu out of guest mode here, as PML buffer
8207 * is flushed at beginning of all VMEXITs, and it's obvious that only
8208 * vcpus running in guest are possible to have unflushed GPAs in PML
8211 kvm_for_each_vcpu(i
, vcpu
, kvm
)
8212 kvm_vcpu_kick(vcpu
);
8215 static void vmx_dump_sel(char *name
, uint32_t sel
)
8217 pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
8218 name
, vmcs_read16(sel
),
8219 vmcs_read32(sel
+ GUEST_ES_AR_BYTES
- GUEST_ES_SELECTOR
),
8220 vmcs_read32(sel
+ GUEST_ES_LIMIT
- GUEST_ES_SELECTOR
),
8221 vmcs_readl(sel
+ GUEST_ES_BASE
- GUEST_ES_SELECTOR
));
8224 static void vmx_dump_dtsel(char *name
, uint32_t limit
)
8226 pr_err("%s limit=0x%08x, base=0x%016lx\n",
8227 name
, vmcs_read32(limit
),
8228 vmcs_readl(limit
+ GUEST_GDTR_BASE
- GUEST_GDTR_LIMIT
));
8231 static void dump_vmcs(void)
8233 u32 vmentry_ctl
= vmcs_read32(VM_ENTRY_CONTROLS
);
8234 u32 vmexit_ctl
= vmcs_read32(VM_EXIT_CONTROLS
);
8235 u32 cpu_based_exec_ctrl
= vmcs_read32(CPU_BASED_VM_EXEC_CONTROL
);
8236 u32 pin_based_exec_ctrl
= vmcs_read32(PIN_BASED_VM_EXEC_CONTROL
);
8237 u32 secondary_exec_control
= 0;
8238 unsigned long cr4
= vmcs_readl(GUEST_CR4
);
8239 u64 efer
= vmcs_read64(GUEST_IA32_EFER
);
8242 if (cpu_has_secondary_exec_ctrls())
8243 secondary_exec_control
= vmcs_read32(SECONDARY_VM_EXEC_CONTROL
);
8245 pr_err("*** Guest State ***\n");
8246 pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8247 vmcs_readl(GUEST_CR0
), vmcs_readl(CR0_READ_SHADOW
),
8248 vmcs_readl(CR0_GUEST_HOST_MASK
));
8249 pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8250 cr4
, vmcs_readl(CR4_READ_SHADOW
), vmcs_readl(CR4_GUEST_HOST_MASK
));
8251 pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3
));
8252 if ((secondary_exec_control
& SECONDARY_EXEC_ENABLE_EPT
) &&
8253 (cr4
& X86_CR4_PAE
) && !(efer
& EFER_LMA
))
8255 pr_err("PDPTR0 = 0x%016llx PDPTR1 = 0x%016llx\n",
8256 vmcs_read64(GUEST_PDPTR0
), vmcs_read64(GUEST_PDPTR1
));
8257 pr_err("PDPTR2 = 0x%016llx PDPTR3 = 0x%016llx\n",
8258 vmcs_read64(GUEST_PDPTR2
), vmcs_read64(GUEST_PDPTR3
));
8260 pr_err("RSP = 0x%016lx RIP = 0x%016lx\n",
8261 vmcs_readl(GUEST_RSP
), vmcs_readl(GUEST_RIP
));
8262 pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n",
8263 vmcs_readl(GUEST_RFLAGS
), vmcs_readl(GUEST_DR7
));
8264 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8265 vmcs_readl(GUEST_SYSENTER_ESP
),
8266 vmcs_read32(GUEST_SYSENTER_CS
), vmcs_readl(GUEST_SYSENTER_EIP
));
8267 vmx_dump_sel("CS: ", GUEST_CS_SELECTOR
);
8268 vmx_dump_sel("DS: ", GUEST_DS_SELECTOR
);
8269 vmx_dump_sel("SS: ", GUEST_SS_SELECTOR
);
8270 vmx_dump_sel("ES: ", GUEST_ES_SELECTOR
);
8271 vmx_dump_sel("FS: ", GUEST_FS_SELECTOR
);
8272 vmx_dump_sel("GS: ", GUEST_GS_SELECTOR
);
8273 vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT
);
8274 vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR
);
8275 vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT
);
8276 vmx_dump_sel("TR: ", GUEST_TR_SELECTOR
);
8277 if ((vmexit_ctl
& (VM_EXIT_SAVE_IA32_PAT
| VM_EXIT_SAVE_IA32_EFER
)) ||
8278 (vmentry_ctl
& (VM_ENTRY_LOAD_IA32_PAT
| VM_ENTRY_LOAD_IA32_EFER
)))
8279 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
8280 efer
, vmcs_read64(GUEST_IA32_PAT
));
8281 pr_err("DebugCtl = 0x%016llx DebugExceptions = 0x%016lx\n",
8282 vmcs_read64(GUEST_IA32_DEBUGCTL
),
8283 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS
));
8284 if (vmentry_ctl
& VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL
)
8285 pr_err("PerfGlobCtl = 0x%016llx\n",
8286 vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL
));
8287 if (vmentry_ctl
& VM_ENTRY_LOAD_BNDCFGS
)
8288 pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS
));
8289 pr_err("Interruptibility = %08x ActivityState = %08x\n",
8290 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO
),
8291 vmcs_read32(GUEST_ACTIVITY_STATE
));
8292 if (secondary_exec_control
& SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY
)
8293 pr_err("InterruptStatus = %04x\n",
8294 vmcs_read16(GUEST_INTR_STATUS
));
8296 pr_err("*** Host State ***\n");
8297 pr_err("RIP = 0x%016lx RSP = 0x%016lx\n",
8298 vmcs_readl(HOST_RIP
), vmcs_readl(HOST_RSP
));
8299 pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
8300 vmcs_read16(HOST_CS_SELECTOR
), vmcs_read16(HOST_SS_SELECTOR
),
8301 vmcs_read16(HOST_DS_SELECTOR
), vmcs_read16(HOST_ES_SELECTOR
),
8302 vmcs_read16(HOST_FS_SELECTOR
), vmcs_read16(HOST_GS_SELECTOR
),
8303 vmcs_read16(HOST_TR_SELECTOR
));
8304 pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
8305 vmcs_readl(HOST_FS_BASE
), vmcs_readl(HOST_GS_BASE
),
8306 vmcs_readl(HOST_TR_BASE
));
8307 pr_err("GDTBase=%016lx IDTBase=%016lx\n",
8308 vmcs_readl(HOST_GDTR_BASE
), vmcs_readl(HOST_IDTR_BASE
));
8309 pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
8310 vmcs_readl(HOST_CR0
), vmcs_readl(HOST_CR3
),
8311 vmcs_readl(HOST_CR4
));
8312 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8313 vmcs_readl(HOST_IA32_SYSENTER_ESP
),
8314 vmcs_read32(HOST_IA32_SYSENTER_CS
),
8315 vmcs_readl(HOST_IA32_SYSENTER_EIP
));
8316 if (vmexit_ctl
& (VM_EXIT_LOAD_IA32_PAT
| VM_EXIT_LOAD_IA32_EFER
))
8317 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
8318 vmcs_read64(HOST_IA32_EFER
),
8319 vmcs_read64(HOST_IA32_PAT
));
8320 if (vmexit_ctl
& VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL
)
8321 pr_err("PerfGlobCtl = 0x%016llx\n",
8322 vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL
));
8324 pr_err("*** Control State ***\n");
8325 pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
8326 pin_based_exec_ctrl
, cpu_based_exec_ctrl
, secondary_exec_control
);
8327 pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl
, vmexit_ctl
);
8328 pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
8329 vmcs_read32(EXCEPTION_BITMAP
),
8330 vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK
),
8331 vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH
));
8332 pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
8333 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD
),
8334 vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE
),
8335 vmcs_read32(VM_ENTRY_INSTRUCTION_LEN
));
8336 pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
8337 vmcs_read32(VM_EXIT_INTR_INFO
),
8338 vmcs_read32(VM_EXIT_INTR_ERROR_CODE
),
8339 vmcs_read32(VM_EXIT_INSTRUCTION_LEN
));
8340 pr_err(" reason=%08x qualification=%016lx\n",
8341 vmcs_read32(VM_EXIT_REASON
), vmcs_readl(EXIT_QUALIFICATION
));
8342 pr_err("IDTVectoring: info=%08x errcode=%08x\n",
8343 vmcs_read32(IDT_VECTORING_INFO_FIELD
),
8344 vmcs_read32(IDT_VECTORING_ERROR_CODE
));
8345 pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET
));
8346 if (secondary_exec_control
& SECONDARY_EXEC_TSC_SCALING
)
8347 pr_err("TSC Multiplier = 0x%016llx\n",
8348 vmcs_read64(TSC_MULTIPLIER
));
8349 if (cpu_based_exec_ctrl
& CPU_BASED_TPR_SHADOW
)
8350 pr_err("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD
));
8351 if (pin_based_exec_ctrl
& PIN_BASED_POSTED_INTR
)
8352 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV
));
8353 if ((secondary_exec_control
& SECONDARY_EXEC_ENABLE_EPT
))
8354 pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER
));
8355 n
= vmcs_read32(CR3_TARGET_COUNT
);
8356 for (i
= 0; i
+ 1 < n
; i
+= 4)
8357 pr_err("CR3 target%u=%016lx target%u=%016lx\n",
8358 i
, vmcs_readl(CR3_TARGET_VALUE0
+ i
* 2),
8359 i
+ 1, vmcs_readl(CR3_TARGET_VALUE0
+ i
* 2 + 2));
8361 pr_err("CR3 target%u=%016lx\n",
8362 i
, vmcs_readl(CR3_TARGET_VALUE0
+ i
* 2));
8363 if (secondary_exec_control
& SECONDARY_EXEC_PAUSE_LOOP_EXITING
)
8364 pr_err("PLE Gap=%08x Window=%08x\n",
8365 vmcs_read32(PLE_GAP
), vmcs_read32(PLE_WINDOW
));
8366 if (secondary_exec_control
& SECONDARY_EXEC_ENABLE_VPID
)
8367 pr_err("Virtual processor ID = 0x%04x\n",
8368 vmcs_read16(VIRTUAL_PROCESSOR_ID
));
8372 * The guest has exited. See if we can fix it or if we need userspace
8375 static int vmx_handle_exit(struct kvm_vcpu
*vcpu
)
8377 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
8378 u32 exit_reason
= vmx
->exit_reason
;
8379 u32 vectoring_info
= vmx
->idt_vectoring_info
;
8381 trace_kvm_exit(exit_reason
, vcpu
, KVM_ISA_VMX
);
8382 vcpu
->arch
.gpa_available
= false;
8385 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
8386 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
8387 * querying dirty_bitmap, we only need to kick all vcpus out of guest
8388 * mode as if vcpus is in root mode, the PML buffer must has been
8392 vmx_flush_pml_buffer(vcpu
);
8394 /* If guest state is invalid, start emulating */
8395 if (vmx
->emulation_required
)
8396 return handle_invalid_guest_state(vcpu
);
8398 if (is_guest_mode(vcpu
) && nested_vmx_exit_handled(vcpu
)) {
8399 nested_vmx_vmexit(vcpu
, exit_reason
,
8400 vmcs_read32(VM_EXIT_INTR_INFO
),
8401 vmcs_readl(EXIT_QUALIFICATION
));
8405 if (exit_reason
& VMX_EXIT_REASONS_FAILED_VMENTRY
) {
8407 vcpu
->run
->exit_reason
= KVM_EXIT_FAIL_ENTRY
;
8408 vcpu
->run
->fail_entry
.hardware_entry_failure_reason
8413 if (unlikely(vmx
->fail
)) {
8414 vcpu
->run
->exit_reason
= KVM_EXIT_FAIL_ENTRY
;
8415 vcpu
->run
->fail_entry
.hardware_entry_failure_reason
8416 = vmcs_read32(VM_INSTRUCTION_ERROR
);
8422 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
8423 * delivery event since it indicates guest is accessing MMIO.
8424 * The vm-exit can be triggered again after return to guest that
8425 * will cause infinite loop.
8427 if ((vectoring_info
& VECTORING_INFO_VALID_MASK
) &&
8428 (exit_reason
!= EXIT_REASON_EXCEPTION_NMI
&&
8429 exit_reason
!= EXIT_REASON_EPT_VIOLATION
&&
8430 exit_reason
!= EXIT_REASON_PML_FULL
&&
8431 exit_reason
!= EXIT_REASON_TASK_SWITCH
)) {
8432 vcpu
->run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
8433 vcpu
->run
->internal
.suberror
= KVM_INTERNAL_ERROR_DELIVERY_EV
;
8434 vcpu
->run
->internal
.ndata
= 3;
8435 vcpu
->run
->internal
.data
[0] = vectoring_info
;
8436 vcpu
->run
->internal
.data
[1] = exit_reason
;
8437 vcpu
->run
->internal
.data
[2] = vcpu
->arch
.exit_qualification
;
8438 if (exit_reason
== EXIT_REASON_EPT_MISCONFIG
) {
8439 vcpu
->run
->internal
.ndata
++;
8440 vcpu
->run
->internal
.data
[3] =
8441 vmcs_read64(GUEST_PHYSICAL_ADDRESS
);
8446 if (exit_reason
< kvm_vmx_max_exit_handlers
8447 && kvm_vmx_exit_handlers
[exit_reason
])
8448 return kvm_vmx_exit_handlers
[exit_reason
](vcpu
);
8450 vcpu_unimpl(vcpu
, "vmx: unexpected exit reason 0x%x\n",
8452 kvm_queue_exception(vcpu
, UD_VECTOR
);
8457 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
, int tpr
, int irr
)
8459 struct vmcs12
*vmcs12
= get_vmcs12(vcpu
);
8461 if (is_guest_mode(vcpu
) &&
8462 nested_cpu_has(vmcs12
, CPU_BASED_TPR_SHADOW
))
8465 if (irr
== -1 || tpr
< irr
) {
8466 vmcs_write32(TPR_THRESHOLD
, 0);
8470 vmcs_write32(TPR_THRESHOLD
, irr
);
8473 static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu
*vcpu
, bool set
)
8475 u32 sec_exec_control
;
8477 /* Postpone execution until vmcs01 is the current VMCS. */
8478 if (is_guest_mode(vcpu
)) {
8479 to_vmx(vcpu
)->nested
.change_vmcs01_virtual_x2apic_mode
= true;
8483 if (!cpu_has_vmx_virtualize_x2apic_mode())
8486 if (!cpu_need_tpr_shadow(vcpu
))
8489 sec_exec_control
= vmcs_read32(SECONDARY_VM_EXEC_CONTROL
);
8492 sec_exec_control
&= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES
;
8493 sec_exec_control
|= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE
;
8495 sec_exec_control
&= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE
;
8496 sec_exec_control
|= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES
;
8497 vmx_flush_tlb_ept_only(vcpu
);
8499 vmcs_write32(SECONDARY_VM_EXEC_CONTROL
, sec_exec_control
);
8501 vmx_set_msr_bitmap(vcpu
);
8504 static void vmx_set_apic_access_page_addr(struct kvm_vcpu
*vcpu
, hpa_t hpa
)
8506 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
8509 * Currently we do not handle the nested case where L2 has an
8510 * APIC access page of its own; that page is still pinned.
8511 * Hence, we skip the case where the VCPU is in guest mode _and_
8512 * L1 prepared an APIC access page for L2.
8514 * For the case where L1 and L2 share the same APIC access page
8515 * (flexpriority=Y but SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES clear
8516 * in the vmcs12), this function will only update either the vmcs01
8517 * or the vmcs02. If the former, the vmcs02 will be updated by
8518 * prepare_vmcs02. If the latter, the vmcs01 will be updated in
8519 * the next L2->L1 exit.
8521 if (!is_guest_mode(vcpu
) ||
8522 !nested_cpu_has2(get_vmcs12(&vmx
->vcpu
),
8523 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES
)) {
8524 vmcs_write64(APIC_ACCESS_ADDR
, hpa
);
8525 vmx_flush_tlb_ept_only(vcpu
);
8529 static void vmx_hwapic_isr_update(struct kvm_vcpu
*vcpu
, int max_isr
)
8537 status
= vmcs_read16(GUEST_INTR_STATUS
);
8539 if (max_isr
!= old
) {
8541 status
|= max_isr
<< 8;
8542 vmcs_write16(GUEST_INTR_STATUS
, status
);
8546 static void vmx_set_rvi(int vector
)
8554 status
= vmcs_read16(GUEST_INTR_STATUS
);
8555 old
= (u8
)status
& 0xff;
8556 if ((u8
)vector
!= old
) {
8558 status
|= (u8
)vector
;
8559 vmcs_write16(GUEST_INTR_STATUS
, status
);
8563 static void vmx_hwapic_irr_update(struct kvm_vcpu
*vcpu
, int max_irr
)
8565 if (!is_guest_mode(vcpu
)) {
8566 vmx_set_rvi(max_irr
);
8574 * In guest mode. If a vmexit is needed, vmx_check_nested_events
8577 if (nested_exit_on_intr(vcpu
))
8581 * Else, fall back to pre-APICv interrupt injection since L2
8582 * is run without virtual interrupt delivery.
8584 if (!kvm_event_needs_reinjection(vcpu
) &&
8585 vmx_interrupt_allowed(vcpu
)) {
8586 kvm_queue_interrupt(vcpu
, max_irr
, false);
8587 vmx_inject_irq(vcpu
);
8591 static int vmx_sync_pir_to_irr(struct kvm_vcpu
*vcpu
)
8593 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
8596 WARN_ON(!vcpu
->arch
.apicv_active
);
8597 if (pi_test_on(&vmx
->pi_desc
)) {
8598 pi_clear_on(&vmx
->pi_desc
);
8600 * IOMMU can write to PIR.ON, so the barrier matters even on UP.
8601 * But on x86 this is just a compiler barrier anyway.
8603 smp_mb__after_atomic();
8604 max_irr
= kvm_apic_update_irr(vcpu
, vmx
->pi_desc
.pir
);
8606 max_irr
= kvm_lapic_find_highest_irr(vcpu
);
8608 vmx_hwapic_irr_update(vcpu
, max_irr
);
8612 static void vmx_load_eoi_exitmap(struct kvm_vcpu
*vcpu
, u64
*eoi_exit_bitmap
)
8614 if (!kvm_vcpu_apicv_active(vcpu
))
8617 vmcs_write64(EOI_EXIT_BITMAP0
, eoi_exit_bitmap
[0]);
8618 vmcs_write64(EOI_EXIT_BITMAP1
, eoi_exit_bitmap
[1]);
8619 vmcs_write64(EOI_EXIT_BITMAP2
, eoi_exit_bitmap
[2]);
8620 vmcs_write64(EOI_EXIT_BITMAP3
, eoi_exit_bitmap
[3]);
8623 static void vmx_apicv_post_state_restore(struct kvm_vcpu
*vcpu
)
8625 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
8627 pi_clear_on(&vmx
->pi_desc
);
8628 memset(vmx
->pi_desc
.pir
, 0, sizeof(vmx
->pi_desc
.pir
));
8631 static void vmx_complete_atomic_exit(struct vcpu_vmx
*vmx
)
8635 if (!(vmx
->exit_reason
== EXIT_REASON_MCE_DURING_VMENTRY
8636 || vmx
->exit_reason
== EXIT_REASON_EXCEPTION_NMI
))
8639 vmx
->exit_intr_info
= vmcs_read32(VM_EXIT_INTR_INFO
);
8640 exit_intr_info
= vmx
->exit_intr_info
;
8642 /* Handle machine checks before interrupts are enabled */
8643 if (is_machine_check(exit_intr_info
))
8644 kvm_machine_check();
8646 /* We need to handle NMIs before interrupts are enabled */
8647 if (is_nmi(exit_intr_info
)) {
8648 kvm_before_handle_nmi(&vmx
->vcpu
);
8650 kvm_after_handle_nmi(&vmx
->vcpu
);
8654 static void vmx_handle_external_intr(struct kvm_vcpu
*vcpu
)
8656 u32 exit_intr_info
= vmcs_read32(VM_EXIT_INTR_INFO
);
8657 register void *__sp
asm(_ASM_SP
);
8659 if ((exit_intr_info
& (INTR_INFO_VALID_MASK
| INTR_INFO_INTR_TYPE_MASK
))
8660 == (INTR_INFO_VALID_MASK
| INTR_TYPE_EXT_INTR
)) {
8661 unsigned int vector
;
8662 unsigned long entry
;
8664 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
8665 #ifdef CONFIG_X86_64
8669 vector
= exit_intr_info
& INTR_INFO_VECTOR_MASK
;
8670 desc
= (gate_desc
*)vmx
->host_idt_base
+ vector
;
8671 entry
= gate_offset(*desc
);
8673 #ifdef CONFIG_X86_64
8674 "mov %%" _ASM_SP
", %[sp]\n\t"
8675 "and $0xfffffffffffffff0, %%" _ASM_SP
"\n\t"
8680 __ASM_SIZE(push
) " $%c[cs]\n\t"
8681 "call *%[entry]\n\t"
8683 #ifdef CONFIG_X86_64
8689 [ss
]"i"(__KERNEL_DS
),
8690 [cs
]"i"(__KERNEL_CS
)
8694 STACK_FRAME_NON_STANDARD(vmx_handle_external_intr
);
8696 static bool vmx_has_high_real_mode_segbase(void)
8698 return enable_unrestricted_guest
|| emulate_invalid_guest_state
;
8701 static bool vmx_mpx_supported(void)
8703 return (vmcs_config
.vmexit_ctrl
& VM_EXIT_CLEAR_BNDCFGS
) &&
8704 (vmcs_config
.vmentry_ctrl
& VM_ENTRY_LOAD_BNDCFGS
);
8707 static bool vmx_xsaves_supported(void)
8709 return vmcs_config
.cpu_based_2nd_exec_ctrl
&
8710 SECONDARY_EXEC_XSAVES
;
8713 static void vmx_recover_nmi_blocking(struct vcpu_vmx
*vmx
)
8718 bool idtv_info_valid
;
8720 idtv_info_valid
= vmx
->idt_vectoring_info
& VECTORING_INFO_VALID_MASK
;
8722 if (vmx
->nmi_known_unmasked
)
8725 * Can't use vmx->exit_intr_info since we're not sure what
8726 * the exit reason is.
8728 exit_intr_info
= vmcs_read32(VM_EXIT_INTR_INFO
);
8729 unblock_nmi
= (exit_intr_info
& INTR_INFO_UNBLOCK_NMI
) != 0;
8730 vector
= exit_intr_info
& INTR_INFO_VECTOR_MASK
;
8732 * SDM 3: 27.7.1.2 (September 2008)
8733 * Re-set bit "block by NMI" before VM entry if vmexit caused by
8734 * a guest IRET fault.
8735 * SDM 3: 23.2.2 (September 2008)
8736 * Bit 12 is undefined in any of the following cases:
8737 * If the VM exit sets the valid bit in the IDT-vectoring
8738 * information field.
8739 * If the VM exit is due to a double fault.
8741 if ((exit_intr_info
& INTR_INFO_VALID_MASK
) && unblock_nmi
&&
8742 vector
!= DF_VECTOR
&& !idtv_info_valid
)
8743 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO
,
8744 GUEST_INTR_STATE_NMI
);
8746 vmx
->nmi_known_unmasked
=
8747 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO
)
8748 & GUEST_INTR_STATE_NMI
);
8751 static void __vmx_complete_interrupts(struct kvm_vcpu
*vcpu
,
8752 u32 idt_vectoring_info
,
8753 int instr_len_field
,
8754 int error_code_field
)
8758 bool idtv_info_valid
;
8760 idtv_info_valid
= idt_vectoring_info
& VECTORING_INFO_VALID_MASK
;
8762 vcpu
->arch
.nmi_injected
= false;
8763 kvm_clear_exception_queue(vcpu
);
8764 kvm_clear_interrupt_queue(vcpu
);
8766 if (!idtv_info_valid
)
8769 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
8771 vector
= idt_vectoring_info
& VECTORING_INFO_VECTOR_MASK
;
8772 type
= idt_vectoring_info
& VECTORING_INFO_TYPE_MASK
;
8775 case INTR_TYPE_NMI_INTR
:
8776 vcpu
->arch
.nmi_injected
= true;
8778 * SDM 3: 27.7.1.2 (September 2008)
8779 * Clear bit "block by NMI" before VM entry if a NMI
8782 vmx_set_nmi_mask(vcpu
, false);
8784 case INTR_TYPE_SOFT_EXCEPTION
:
8785 vcpu
->arch
.event_exit_inst_len
= vmcs_read32(instr_len_field
);
8787 case INTR_TYPE_HARD_EXCEPTION
:
8788 if (idt_vectoring_info
& VECTORING_INFO_DELIVER_CODE_MASK
) {
8789 u32 err
= vmcs_read32(error_code_field
);
8790 kvm_requeue_exception_e(vcpu
, vector
, err
);
8792 kvm_requeue_exception(vcpu
, vector
);
8794 case INTR_TYPE_SOFT_INTR
:
8795 vcpu
->arch
.event_exit_inst_len
= vmcs_read32(instr_len_field
);
8797 case INTR_TYPE_EXT_INTR
:
8798 kvm_queue_interrupt(vcpu
, vector
, type
== INTR_TYPE_SOFT_INTR
);
8805 static void vmx_complete_interrupts(struct vcpu_vmx
*vmx
)
8807 __vmx_complete_interrupts(&vmx
->vcpu
, vmx
->idt_vectoring_info
,
8808 VM_EXIT_INSTRUCTION_LEN
,
8809 IDT_VECTORING_ERROR_CODE
);
8812 static void vmx_cancel_injection(struct kvm_vcpu
*vcpu
)
8814 __vmx_complete_interrupts(vcpu
,
8815 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD
),
8816 VM_ENTRY_INSTRUCTION_LEN
,
8817 VM_ENTRY_EXCEPTION_ERROR_CODE
);
8819 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD
, 0);
8822 static void atomic_switch_perf_msrs(struct vcpu_vmx
*vmx
)
8825 struct perf_guest_switch_msr
*msrs
;
8827 msrs
= perf_guest_get_msrs(&nr_msrs
);
8832 for (i
= 0; i
< nr_msrs
; i
++)
8833 if (msrs
[i
].host
== msrs
[i
].guest
)
8834 clear_atomic_switch_msr(vmx
, msrs
[i
].msr
);
8836 add_atomic_switch_msr(vmx
, msrs
[i
].msr
, msrs
[i
].guest
,
8840 static void vmx_arm_hv_timer(struct kvm_vcpu
*vcpu
)
8842 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
8846 if (vmx
->hv_deadline_tsc
== -1)
8850 if (vmx
->hv_deadline_tsc
> tscl
)
8851 /* sure to be 32 bit only because checked on set_hv_timer */
8852 delta_tsc
= (u32
)((vmx
->hv_deadline_tsc
- tscl
) >>
8853 cpu_preemption_timer_multi
);
8857 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE
, delta_tsc
);
8860 static void __noclone
vmx_vcpu_run(struct kvm_vcpu
*vcpu
)
8862 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
8863 unsigned long debugctlmsr
, cr3
, cr4
;
8865 /* Don't enter VMX if guest state is invalid, let the exit handler
8866 start emulation until we arrive back to a valid state */
8867 if (vmx
->emulation_required
)
8870 if (vmx
->ple_window_dirty
) {
8871 vmx
->ple_window_dirty
= false;
8872 vmcs_write32(PLE_WINDOW
, vmx
->ple_window
);
8875 if (vmx
->nested
.sync_shadow_vmcs
) {
8876 copy_vmcs12_to_shadow(vmx
);
8877 vmx
->nested
.sync_shadow_vmcs
= false;
8880 if (test_bit(VCPU_REGS_RSP
, (unsigned long *)&vcpu
->arch
.regs_dirty
))
8881 vmcs_writel(GUEST_RSP
, vcpu
->arch
.regs
[VCPU_REGS_RSP
]);
8882 if (test_bit(VCPU_REGS_RIP
, (unsigned long *)&vcpu
->arch
.regs_dirty
))
8883 vmcs_writel(GUEST_RIP
, vcpu
->arch
.regs
[VCPU_REGS_RIP
]);
8885 cr3
= __get_current_cr3_fast();
8886 if (unlikely(cr3
!= vmx
->host_state
.vmcs_host_cr3
)) {
8887 vmcs_writel(HOST_CR3
, cr3
);
8888 vmx
->host_state
.vmcs_host_cr3
= cr3
;
8891 cr4
= cr4_read_shadow();
8892 if (unlikely(cr4
!= vmx
->host_state
.vmcs_host_cr4
)) {
8893 vmcs_writel(HOST_CR4
, cr4
);
8894 vmx
->host_state
.vmcs_host_cr4
= cr4
;
8897 /* When single-stepping over STI and MOV SS, we must clear the
8898 * corresponding interruptibility bits in the guest state. Otherwise
8899 * vmentry fails as it then expects bit 14 (BS) in pending debug
8900 * exceptions being set, but that's not correct for the guest debugging
8902 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)
8903 vmx_set_interrupt_shadow(vcpu
, 0);
8905 if (vmx
->guest_pkru_valid
)
8906 __write_pkru(vmx
->guest_pkru
);
8908 atomic_switch_perf_msrs(vmx
);
8909 debugctlmsr
= get_debugctlmsr();
8911 vmx_arm_hv_timer(vcpu
);
8913 vmx
->__launched
= vmx
->loaded_vmcs
->launched
;
8915 /* Store host registers */
8916 "push %%" _ASM_DX
"; push %%" _ASM_BP
";"
8917 "push %%" _ASM_CX
" \n\t" /* placeholder for guest rcx */
8918 "push %%" _ASM_CX
" \n\t"
8919 "cmp %%" _ASM_SP
", %c[host_rsp](%0) \n\t"
8921 "mov %%" _ASM_SP
", %c[host_rsp](%0) \n\t"
8922 __ex(ASM_VMX_VMWRITE_RSP_RDX
) "\n\t"
8924 /* Reload cr2 if changed */
8925 "mov %c[cr2](%0), %%" _ASM_AX
" \n\t"
8926 "mov %%cr2, %%" _ASM_DX
" \n\t"
8927 "cmp %%" _ASM_AX
", %%" _ASM_DX
" \n\t"
8929 "mov %%" _ASM_AX
", %%cr2 \n\t"
8931 /* Check if vmlaunch of vmresume is needed */
8932 "cmpl $0, %c[launched](%0) \n\t"
8933 /* Load guest registers. Don't clobber flags. */
8934 "mov %c[rax](%0), %%" _ASM_AX
" \n\t"
8935 "mov %c[rbx](%0), %%" _ASM_BX
" \n\t"
8936 "mov %c[rdx](%0), %%" _ASM_DX
" \n\t"
8937 "mov %c[rsi](%0), %%" _ASM_SI
" \n\t"
8938 "mov %c[rdi](%0), %%" _ASM_DI
" \n\t"
8939 "mov %c[rbp](%0), %%" _ASM_BP
" \n\t"
8940 #ifdef CONFIG_X86_64
8941 "mov %c[r8](%0), %%r8 \n\t"
8942 "mov %c[r9](%0), %%r9 \n\t"
8943 "mov %c[r10](%0), %%r10 \n\t"
8944 "mov %c[r11](%0), %%r11 \n\t"
8945 "mov %c[r12](%0), %%r12 \n\t"
8946 "mov %c[r13](%0), %%r13 \n\t"
8947 "mov %c[r14](%0), %%r14 \n\t"
8948 "mov %c[r15](%0), %%r15 \n\t"
8950 "mov %c[rcx](%0), %%" _ASM_CX
" \n\t" /* kills %0 (ecx) */
8952 /* Enter guest mode */
8954 __ex(ASM_VMX_VMLAUNCH
) "\n\t"
8956 "1: " __ex(ASM_VMX_VMRESUME
) "\n\t"
8958 /* Save guest registers, load host registers, keep flags */
8959 "mov %0, %c[wordsize](%%" _ASM_SP
") \n\t"
8961 "mov %%" _ASM_AX
", %c[rax](%0) \n\t"
8962 "mov %%" _ASM_BX
", %c[rbx](%0) \n\t"
8963 __ASM_SIZE(pop
) " %c[rcx](%0) \n\t"
8964 "mov %%" _ASM_DX
", %c[rdx](%0) \n\t"
8965 "mov %%" _ASM_SI
", %c[rsi](%0) \n\t"
8966 "mov %%" _ASM_DI
", %c[rdi](%0) \n\t"
8967 "mov %%" _ASM_BP
", %c[rbp](%0) \n\t"
8968 #ifdef CONFIG_X86_64
8969 "mov %%r8, %c[r8](%0) \n\t"
8970 "mov %%r9, %c[r9](%0) \n\t"
8971 "mov %%r10, %c[r10](%0) \n\t"
8972 "mov %%r11, %c[r11](%0) \n\t"
8973 "mov %%r12, %c[r12](%0) \n\t"
8974 "mov %%r13, %c[r13](%0) \n\t"
8975 "mov %%r14, %c[r14](%0) \n\t"
8976 "mov %%r15, %c[r15](%0) \n\t"
8978 "mov %%cr2, %%" _ASM_AX
" \n\t"
8979 "mov %%" _ASM_AX
", %c[cr2](%0) \n\t"
8981 "pop %%" _ASM_BP
"; pop %%" _ASM_DX
" \n\t"
8982 "setbe %c[fail](%0) \n\t"
8983 ".pushsection .rodata \n\t"
8984 ".global vmx_return \n\t"
8985 "vmx_return: " _ASM_PTR
" 2b \n\t"
8987 : : "c"(vmx
), "d"((unsigned long)HOST_RSP
),
8988 [launched
]"i"(offsetof(struct vcpu_vmx
, __launched
)),
8989 [fail
]"i"(offsetof(struct vcpu_vmx
, fail
)),
8990 [host_rsp
]"i"(offsetof(struct vcpu_vmx
, host_rsp
)),
8991 [rax
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.regs
[VCPU_REGS_RAX
])),
8992 [rbx
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.regs
[VCPU_REGS_RBX
])),
8993 [rcx
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.regs
[VCPU_REGS_RCX
])),
8994 [rdx
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.regs
[VCPU_REGS_RDX
])),
8995 [rsi
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.regs
[VCPU_REGS_RSI
])),
8996 [rdi
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.regs
[VCPU_REGS_RDI
])),
8997 [rbp
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.regs
[VCPU_REGS_RBP
])),
8998 #ifdef CONFIG_X86_64
8999 [r8
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.regs
[VCPU_REGS_R8
])),
9000 [r9
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.regs
[VCPU_REGS_R9
])),
9001 [r10
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.regs
[VCPU_REGS_R10
])),
9002 [r11
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.regs
[VCPU_REGS_R11
])),
9003 [r12
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.regs
[VCPU_REGS_R12
])),
9004 [r13
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.regs
[VCPU_REGS_R13
])),
9005 [r14
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.regs
[VCPU_REGS_R14
])),
9006 [r15
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.regs
[VCPU_REGS_R15
])),
9008 [cr2
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.cr2
)),
9009 [wordsize
]"i"(sizeof(ulong
))
9011 #ifdef CONFIG_X86_64
9012 , "rax", "rbx", "rdi", "rsi"
9013 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
9015 , "eax", "ebx", "edi", "esi"
9019 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
9021 update_debugctlmsr(debugctlmsr
);
9023 #ifndef CONFIG_X86_64
9025 * The sysexit path does not restore ds/es, so we must set them to
9026 * a reasonable value ourselves.
9028 * We can't defer this to vmx_load_host_state() since that function
9029 * may be executed in interrupt context, which saves and restore segments
9030 * around it, nullifying its effect.
9032 loadsegment(ds
, __USER_DS
);
9033 loadsegment(es
, __USER_DS
);
9036 vcpu
->arch
.regs_avail
= ~((1 << VCPU_REGS_RIP
) | (1 << VCPU_REGS_RSP
)
9037 | (1 << VCPU_EXREG_RFLAGS
)
9038 | (1 << VCPU_EXREG_PDPTR
)
9039 | (1 << VCPU_EXREG_SEGMENTS
)
9040 | (1 << VCPU_EXREG_CR3
));
9041 vcpu
->arch
.regs_dirty
= 0;
9043 vmx
->idt_vectoring_info
= vmcs_read32(IDT_VECTORING_INFO_FIELD
);
9045 vmx
->loaded_vmcs
->launched
= 1;
9047 vmx
->exit_reason
= vmcs_read32(VM_EXIT_REASON
);
9050 * eager fpu is enabled if PKEY is supported and CR4 is switched
9051 * back on host, so it is safe to read guest PKRU from current
9054 if (boot_cpu_has(X86_FEATURE_OSPKE
)) {
9055 vmx
->guest_pkru
= __read_pkru();
9056 if (vmx
->guest_pkru
!= vmx
->host_pkru
) {
9057 vmx
->guest_pkru_valid
= true;
9058 __write_pkru(vmx
->host_pkru
);
9060 vmx
->guest_pkru_valid
= false;
9064 * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
9065 * we did not inject a still-pending event to L1 now because of
9066 * nested_run_pending, we need to re-enable this bit.
9068 if (vmx
->nested
.nested_run_pending
)
9069 kvm_make_request(KVM_REQ_EVENT
, vcpu
);
9071 vmx
->nested
.nested_run_pending
= 0;
9073 vmx_complete_atomic_exit(vmx
);
9074 vmx_recover_nmi_blocking(vmx
);
9075 vmx_complete_interrupts(vmx
);
9077 STACK_FRAME_NON_STANDARD(vmx_vcpu_run
);
9079 static void vmx_switch_vmcs(struct kvm_vcpu
*vcpu
, struct loaded_vmcs
*vmcs
)
9081 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
9084 if (vmx
->loaded_vmcs
== vmcs
)
9088 vmx
->loaded_vmcs
= vmcs
;
9090 vmx_vcpu_load(vcpu
, cpu
);
9096 * Ensure that the current vmcs of the logical processor is the
9097 * vmcs01 of the vcpu before calling free_nested().
9099 static void vmx_free_vcpu_nested(struct kvm_vcpu
*vcpu
)
9101 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
9104 r
= vcpu_load(vcpu
);
9106 vmx_switch_vmcs(vcpu
, &vmx
->vmcs01
);
9111 static void vmx_free_vcpu(struct kvm_vcpu
*vcpu
)
9113 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
9116 vmx_destroy_pml_buffer(vmx
);
9117 free_vpid(vmx
->vpid
);
9118 leave_guest_mode(vcpu
);
9119 vmx_free_vcpu_nested(vcpu
);
9120 free_loaded_vmcs(vmx
->loaded_vmcs
);
9121 kfree(vmx
->guest_msrs
);
9122 kvm_vcpu_uninit(vcpu
);
9123 kmem_cache_free(kvm_vcpu_cache
, vmx
);
9126 static struct kvm_vcpu
*vmx_create_vcpu(struct kvm
*kvm
, unsigned int id
)
9129 struct vcpu_vmx
*vmx
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL
);
9133 return ERR_PTR(-ENOMEM
);
9135 vmx
->vpid
= allocate_vpid();
9137 err
= kvm_vcpu_init(&vmx
->vcpu
, kvm
, id
);
9144 * If PML is turned on, failure on enabling PML just results in failure
9145 * of creating the vcpu, therefore we can simplify PML logic (by
9146 * avoiding dealing with cases, such as enabling PML partially on vcpus
9147 * for the guest, etc.
9150 vmx
->pml_pg
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
9155 vmx
->guest_msrs
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
9156 BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index
) * sizeof(vmx
->guest_msrs
[0])
9159 if (!vmx
->guest_msrs
)
9162 vmx
->loaded_vmcs
= &vmx
->vmcs01
;
9163 vmx
->loaded_vmcs
->vmcs
= alloc_vmcs();
9164 vmx
->loaded_vmcs
->shadow_vmcs
= NULL
;
9165 if (!vmx
->loaded_vmcs
->vmcs
)
9167 loaded_vmcs_init(vmx
->loaded_vmcs
);
9170 vmx_vcpu_load(&vmx
->vcpu
, cpu
);
9171 vmx
->vcpu
.cpu
= cpu
;
9172 err
= vmx_vcpu_setup(vmx
);
9173 vmx_vcpu_put(&vmx
->vcpu
);
9177 if (cpu_need_virtualize_apic_accesses(&vmx
->vcpu
)) {
9178 err
= alloc_apic_access_page(kvm
);
9184 if (!kvm
->arch
.ept_identity_map_addr
)
9185 kvm
->arch
.ept_identity_map_addr
=
9186 VMX_EPT_IDENTITY_PAGETABLE_ADDR
;
9187 err
= init_rmode_identity_map(kvm
);
9193 nested_vmx_setup_ctls_msrs(vmx
);
9194 vmx
->nested
.vpid02
= allocate_vpid();
9197 vmx
->nested
.posted_intr_nv
= -1;
9198 vmx
->nested
.current_vmptr
= -1ull;
9199 vmx
->nested
.current_vmcs12
= NULL
;
9201 vmx
->msr_ia32_feature_control_valid_bits
= FEATURE_CONTROL_LOCKED
;
9206 free_vpid(vmx
->nested
.vpid02
);
9207 free_loaded_vmcs(vmx
->loaded_vmcs
);
9209 kfree(vmx
->guest_msrs
);
9211 vmx_destroy_pml_buffer(vmx
);
9213 kvm_vcpu_uninit(&vmx
->vcpu
);
9215 free_vpid(vmx
->vpid
);
9216 kmem_cache_free(kvm_vcpu_cache
, vmx
);
9217 return ERR_PTR(err
);
9220 static void __init
vmx_check_processor_compat(void *rtn
)
9222 struct vmcs_config vmcs_conf
;
9225 if (setup_vmcs_config(&vmcs_conf
) < 0)
9227 if (memcmp(&vmcs_config
, &vmcs_conf
, sizeof(struct vmcs_config
)) != 0) {
9228 printk(KERN_ERR
"kvm: CPU %d feature inconsistency!\n",
9229 smp_processor_id());
9234 static int get_ept_level(void)
9236 return VMX_EPT_DEFAULT_GAW
+ 1;
9239 static u64
vmx_get_mt_mask(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool is_mmio
)
9244 /* For VT-d and EPT combination
9245 * 1. MMIO: always map as UC
9247 * a. VT-d without snooping control feature: can't guarantee the
9248 * result, try to trust guest.
9249 * b. VT-d with snooping control feature: snooping control feature of
9250 * VT-d engine can guarantee the cache correctness. Just set it
9251 * to WB to keep consistent with host. So the same as item 3.
9252 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
9253 * consistent with host MTRR
9256 cache
= MTRR_TYPE_UNCACHABLE
;
9260 if (!kvm_arch_has_noncoherent_dma(vcpu
->kvm
)) {
9261 ipat
= VMX_EPT_IPAT_BIT
;
9262 cache
= MTRR_TYPE_WRBACK
;
9266 if (kvm_read_cr0(vcpu
) & X86_CR0_CD
) {
9267 ipat
= VMX_EPT_IPAT_BIT
;
9268 if (kvm_check_has_quirk(vcpu
->kvm
, KVM_X86_QUIRK_CD_NW_CLEARED
))
9269 cache
= MTRR_TYPE_WRBACK
;
9271 cache
= MTRR_TYPE_UNCACHABLE
;
9275 cache
= kvm_mtrr_get_guest_memory_type(vcpu
, gfn
);
9278 return (cache
<< VMX_EPT_MT_EPTE_SHIFT
) | ipat
;
9281 static int vmx_get_lpage_level(void)
9283 if (enable_ept
&& !cpu_has_vmx_ept_1g_page())
9284 return PT_DIRECTORY_LEVEL
;
9286 /* For shadow and EPT supported 1GB page */
9287 return PT_PDPE_LEVEL
;
9290 static void vmcs_set_secondary_exec_control(u32 new_ctl
)
9293 * These bits in the secondary execution controls field
9294 * are dynamic, the others are mostly based on the hypervisor
9295 * architecture and the guest's CPUID. Do not touch the
9299 SECONDARY_EXEC_SHADOW_VMCS
|
9300 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE
|
9301 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES
;
9303 u32 cur_ctl
= vmcs_read32(SECONDARY_VM_EXEC_CONTROL
);
9305 vmcs_write32(SECONDARY_VM_EXEC_CONTROL
,
9306 (new_ctl
& ~mask
) | (cur_ctl
& mask
));
9310 * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
9311 * (indicating "allowed-1") if they are supported in the guest's CPUID.
9313 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu
*vcpu
)
9315 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
9316 struct kvm_cpuid_entry2
*entry
;
9318 vmx
->nested
.nested_vmx_cr0_fixed1
= 0xffffffff;
9319 vmx
->nested
.nested_vmx_cr4_fixed1
= X86_CR4_PCE
;
9321 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do { \
9322 if (entry && (entry->_reg & (_cpuid_mask))) \
9323 vmx->nested.nested_vmx_cr4_fixed1 |= (_cr4_mask); \
9326 entry
= kvm_find_cpuid_entry(vcpu
, 0x1, 0);
9327 cr4_fixed1_update(X86_CR4_VME
, edx
, bit(X86_FEATURE_VME
));
9328 cr4_fixed1_update(X86_CR4_PVI
, edx
, bit(X86_FEATURE_VME
));
9329 cr4_fixed1_update(X86_CR4_TSD
, edx
, bit(X86_FEATURE_TSC
));
9330 cr4_fixed1_update(X86_CR4_DE
, edx
, bit(X86_FEATURE_DE
));
9331 cr4_fixed1_update(X86_CR4_PSE
, edx
, bit(X86_FEATURE_PSE
));
9332 cr4_fixed1_update(X86_CR4_PAE
, edx
, bit(X86_FEATURE_PAE
));
9333 cr4_fixed1_update(X86_CR4_MCE
, edx
, bit(X86_FEATURE_MCE
));
9334 cr4_fixed1_update(X86_CR4_PGE
, edx
, bit(X86_FEATURE_PGE
));
9335 cr4_fixed1_update(X86_CR4_OSFXSR
, edx
, bit(X86_FEATURE_FXSR
));
9336 cr4_fixed1_update(X86_CR4_OSXMMEXCPT
, edx
, bit(X86_FEATURE_XMM
));
9337 cr4_fixed1_update(X86_CR4_VMXE
, ecx
, bit(X86_FEATURE_VMX
));
9338 cr4_fixed1_update(X86_CR4_SMXE
, ecx
, bit(X86_FEATURE_SMX
));
9339 cr4_fixed1_update(X86_CR4_PCIDE
, ecx
, bit(X86_FEATURE_PCID
));
9340 cr4_fixed1_update(X86_CR4_OSXSAVE
, ecx
, bit(X86_FEATURE_XSAVE
));
9342 entry
= kvm_find_cpuid_entry(vcpu
, 0x7, 0);
9343 cr4_fixed1_update(X86_CR4_FSGSBASE
, ebx
, bit(X86_FEATURE_FSGSBASE
));
9344 cr4_fixed1_update(X86_CR4_SMEP
, ebx
, bit(X86_FEATURE_SMEP
));
9345 cr4_fixed1_update(X86_CR4_SMAP
, ebx
, bit(X86_FEATURE_SMAP
));
9346 cr4_fixed1_update(X86_CR4_PKE
, ecx
, bit(X86_FEATURE_PKU
));
9347 /* TODO: Use X86_CR4_UMIP and X86_FEATURE_UMIP macros */
9348 cr4_fixed1_update(bit(11), ecx
, bit(2));
9350 #undef cr4_fixed1_update
9353 static void vmx_cpuid_update(struct kvm_vcpu
*vcpu
)
9355 struct kvm_cpuid_entry2
*best
;
9356 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
9357 u32 secondary_exec_ctl
= vmx_secondary_exec_control(vmx
);
9359 if (vmx_rdtscp_supported()) {
9360 bool rdtscp_enabled
= guest_cpuid_has_rdtscp(vcpu
);
9361 if (!rdtscp_enabled
)
9362 secondary_exec_ctl
&= ~SECONDARY_EXEC_RDTSCP
;
9366 vmx
->nested
.nested_vmx_secondary_ctls_high
|=
9367 SECONDARY_EXEC_RDTSCP
;
9369 vmx
->nested
.nested_vmx_secondary_ctls_high
&=
9370 ~SECONDARY_EXEC_RDTSCP
;
9374 /* Exposing INVPCID only when PCID is exposed */
9375 best
= kvm_find_cpuid_entry(vcpu
, 0x7, 0);
9376 if (vmx_invpcid_supported() &&
9377 (!best
|| !(best
->ebx
& bit(X86_FEATURE_INVPCID
)) ||
9378 !guest_cpuid_has_pcid(vcpu
))) {
9379 secondary_exec_ctl
&= ~SECONDARY_EXEC_ENABLE_INVPCID
;
9382 best
->ebx
&= ~bit(X86_FEATURE_INVPCID
);
9385 if (cpu_has_secondary_exec_ctrls())
9386 vmcs_set_secondary_exec_control(secondary_exec_ctl
);
9388 if (nested_vmx_allowed(vcpu
))
9389 to_vmx(vcpu
)->msr_ia32_feature_control_valid_bits
|=
9390 FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX
;
9392 to_vmx(vcpu
)->msr_ia32_feature_control_valid_bits
&=
9393 ~FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX
;
9395 if (nested_vmx_allowed(vcpu
))
9396 nested_vmx_cr_fixed1_bits_update(vcpu
);
9399 static void vmx_set_supported_cpuid(u32 func
, struct kvm_cpuid_entry2
*entry
)
9401 if (func
== 1 && nested
)
9402 entry
->ecx
|= bit(X86_FEATURE_VMX
);
9405 static void nested_ept_inject_page_fault(struct kvm_vcpu
*vcpu
,
9406 struct x86_exception
*fault
)
9408 struct vmcs12
*vmcs12
= get_vmcs12(vcpu
);
9409 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
9411 unsigned long exit_qualification
= vcpu
->arch
.exit_qualification
;
9413 if (vmx
->nested
.pml_full
) {
9414 exit_reason
= EXIT_REASON_PML_FULL
;
9415 vmx
->nested
.pml_full
= false;
9416 exit_qualification
&= INTR_INFO_UNBLOCK_NMI
;
9417 } else if (fault
->error_code
& PFERR_RSVD_MASK
)
9418 exit_reason
= EXIT_REASON_EPT_MISCONFIG
;
9420 exit_reason
= EXIT_REASON_EPT_VIOLATION
;
9422 nested_vmx_vmexit(vcpu
, exit_reason
, 0, exit_qualification
);
9423 vmcs12
->guest_physical_address
= fault
->address
;
9426 static bool nested_ept_ad_enabled(struct kvm_vcpu
*vcpu
)
9428 return nested_ept_get_cr3(vcpu
) & VMX_EPT_AD_ENABLE_BIT
;
9431 /* Callbacks for nested_ept_init_mmu_context: */
9433 static unsigned long nested_ept_get_cr3(struct kvm_vcpu
*vcpu
)
9435 /* return the page table to be shadowed - in our case, EPT12 */
9436 return get_vmcs12(vcpu
)->ept_pointer
;
9439 static int nested_ept_init_mmu_context(struct kvm_vcpu
*vcpu
)
9443 WARN_ON(mmu_is_nested(vcpu
));
9444 wants_ad
= nested_ept_ad_enabled(vcpu
);
9445 if (wants_ad
&& !enable_ept_ad_bits
)
9448 kvm_mmu_unload(vcpu
);
9449 kvm_init_shadow_ept_mmu(vcpu
,
9450 to_vmx(vcpu
)->nested
.nested_vmx_ept_caps
&
9451 VMX_EPT_EXECUTE_ONLY_BIT
,
9453 vcpu
->arch
.mmu
.set_cr3
= vmx_set_cr3
;
9454 vcpu
->arch
.mmu
.get_cr3
= nested_ept_get_cr3
;
9455 vcpu
->arch
.mmu
.inject_page_fault
= nested_ept_inject_page_fault
;
9457 vcpu
->arch
.walk_mmu
= &vcpu
->arch
.nested_mmu
;
9461 static void nested_ept_uninit_mmu_context(struct kvm_vcpu
*vcpu
)
9463 vcpu
->arch
.walk_mmu
= &vcpu
->arch
.mmu
;
9466 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12
*vmcs12
,
9469 bool inequality
, bit
;
9471 bit
= (vmcs12
->exception_bitmap
& (1u << PF_VECTOR
)) != 0;
9473 (error_code
& vmcs12
->page_fault_error_code_mask
) !=
9474 vmcs12
->page_fault_error_code_match
;
9475 return inequality
^ bit
;
9478 static void vmx_inject_page_fault_nested(struct kvm_vcpu
*vcpu
,
9479 struct x86_exception
*fault
)
9481 struct vmcs12
*vmcs12
= get_vmcs12(vcpu
);
9483 WARN_ON(!is_guest_mode(vcpu
));
9485 if (nested_vmx_is_page_fault_vmexit(vmcs12
, fault
->error_code
))
9486 nested_vmx_vmexit(vcpu
, to_vmx(vcpu
)->exit_reason
,
9487 vmcs_read32(VM_EXIT_INTR_INFO
),
9488 vmcs_readl(EXIT_QUALIFICATION
));
9490 kvm_inject_page_fault(vcpu
, fault
);
9493 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu
*vcpu
,
9494 struct vmcs12
*vmcs12
);
9496 static void nested_get_vmcs12_pages(struct kvm_vcpu
*vcpu
,
9497 struct vmcs12
*vmcs12
)
9499 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
9502 if (nested_cpu_has2(vmcs12
, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES
)) {
9504 * Translate L1 physical address to host physical
9505 * address for vmcs02. Keep the page pinned, so this
9506 * physical address remains valid. We keep a reference
9507 * to it so we can release it later.
9509 if (vmx
->nested
.apic_access_page
) /* shouldn't happen */
9510 nested_release_page(vmx
->nested
.apic_access_page
);
9511 vmx
->nested
.apic_access_page
=
9512 nested_get_page(vcpu
, vmcs12
->apic_access_addr
);
9514 * If translation failed, no matter: This feature asks
9515 * to exit when accessing the given address, and if it
9516 * can never be accessed, this feature won't do
9519 if (vmx
->nested
.apic_access_page
) {
9520 hpa
= page_to_phys(vmx
->nested
.apic_access_page
);
9521 vmcs_write64(APIC_ACCESS_ADDR
, hpa
);
9523 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL
,
9524 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES
);
9526 } else if (!(nested_cpu_has_virt_x2apic_mode(vmcs12
)) &&
9527 cpu_need_virtualize_apic_accesses(&vmx
->vcpu
)) {
9528 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL
,
9529 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES
);
9530 kvm_vcpu_reload_apic_access_page(vcpu
);
9533 if (nested_cpu_has(vmcs12
, CPU_BASED_TPR_SHADOW
)) {
9534 if (vmx
->nested
.virtual_apic_page
) /* shouldn't happen */
9535 nested_release_page(vmx
->nested
.virtual_apic_page
);
9536 vmx
->nested
.virtual_apic_page
=
9537 nested_get_page(vcpu
, vmcs12
->virtual_apic_page_addr
);
9540 * If translation failed, VM entry will fail because
9541 * prepare_vmcs02 set VIRTUAL_APIC_PAGE_ADDR to -1ull.
9542 * Failing the vm entry is _not_ what the processor
9543 * does but it's basically the only possibility we
9544 * have. We could still enter the guest if CR8 load
9545 * exits are enabled, CR8 store exits are enabled, and
9546 * virtualize APIC access is disabled; in this case
9547 * the processor would never use the TPR shadow and we
9548 * could simply clear the bit from the execution
9549 * control. But such a configuration is useless, so
9550 * let's keep the code simple.
9552 if (vmx
->nested
.virtual_apic_page
) {
9553 hpa
= page_to_phys(vmx
->nested
.virtual_apic_page
);
9554 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR
, hpa
);
9558 if (nested_cpu_has_posted_intr(vmcs12
)) {
9559 if (vmx
->nested
.pi_desc_page
) { /* shouldn't happen */
9560 kunmap(vmx
->nested
.pi_desc_page
);
9561 nested_release_page(vmx
->nested
.pi_desc_page
);
9563 vmx
->nested
.pi_desc_page
=
9564 nested_get_page(vcpu
, vmcs12
->posted_intr_desc_addr
);
9565 vmx
->nested
.pi_desc
=
9566 (struct pi_desc
*)kmap(vmx
->nested
.pi_desc_page
);
9567 if (!vmx
->nested
.pi_desc
) {
9568 nested_release_page_clean(vmx
->nested
.pi_desc_page
);
9571 vmx
->nested
.pi_desc
=
9572 (struct pi_desc
*)((void *)vmx
->nested
.pi_desc
+
9573 (unsigned long)(vmcs12
->posted_intr_desc_addr
&
9575 vmcs_write64(POSTED_INTR_DESC_ADDR
,
9576 page_to_phys(vmx
->nested
.pi_desc_page
) +
9577 (unsigned long)(vmcs12
->posted_intr_desc_addr
&
9580 if (cpu_has_vmx_msr_bitmap() &&
9581 nested_cpu_has(vmcs12
, CPU_BASED_USE_MSR_BITMAPS
) &&
9582 nested_vmx_merge_msr_bitmap(vcpu
, vmcs12
))
9585 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL
,
9586 CPU_BASED_USE_MSR_BITMAPS
);
9589 static void vmx_start_preemption_timer(struct kvm_vcpu
*vcpu
)
9591 u64 preemption_timeout
= get_vmcs12(vcpu
)->vmx_preemption_timer_value
;
9592 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
9594 if (vcpu
->arch
.virtual_tsc_khz
== 0)
9597 /* Make sure short timeouts reliably trigger an immediate vmexit.
9598 * hrtimer_start does not guarantee this. */
9599 if (preemption_timeout
<= 1) {
9600 vmx_preemption_timer_fn(&vmx
->nested
.preemption_timer
);
9604 preemption_timeout
<<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE
;
9605 preemption_timeout
*= 1000000;
9606 do_div(preemption_timeout
, vcpu
->arch
.virtual_tsc_khz
);
9607 hrtimer_start(&vmx
->nested
.preemption_timer
,
9608 ns_to_ktime(preemption_timeout
), HRTIMER_MODE_REL
);
9611 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu
*vcpu
,
9612 struct vmcs12
*vmcs12
)
9614 if (!nested_cpu_has(vmcs12
, CPU_BASED_USE_IO_BITMAPS
))
9617 if (!page_address_valid(vcpu
, vmcs12
->io_bitmap_a
) ||
9618 !page_address_valid(vcpu
, vmcs12
->io_bitmap_b
))
9624 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu
*vcpu
,
9625 struct vmcs12
*vmcs12
)
9627 if (!nested_cpu_has(vmcs12
, CPU_BASED_USE_MSR_BITMAPS
))
9630 if (!page_address_valid(vcpu
, vmcs12
->msr_bitmap
))
9637 * Merge L0's and L1's MSR bitmap, return false to indicate that
9638 * we do not use the hardware.
9640 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu
*vcpu
,
9641 struct vmcs12
*vmcs12
)
9645 unsigned long *msr_bitmap_l1
;
9646 unsigned long *msr_bitmap_l0
= to_vmx(vcpu
)->nested
.msr_bitmap
;
9648 /* This shortcut is ok because we support only x2APIC MSRs so far. */
9649 if (!nested_cpu_has_virt_x2apic_mode(vmcs12
))
9652 page
= nested_get_page(vcpu
, vmcs12
->msr_bitmap
);
9655 msr_bitmap_l1
= (unsigned long *)kmap(page
);
9657 memset(msr_bitmap_l0
, 0xff, PAGE_SIZE
);
9659 if (nested_cpu_has_virt_x2apic_mode(vmcs12
)) {
9660 if (nested_cpu_has_apic_reg_virt(vmcs12
))
9661 for (msr
= 0x800; msr
<= 0x8ff; msr
++)
9662 nested_vmx_disable_intercept_for_msr(
9663 msr_bitmap_l1
, msr_bitmap_l0
,
9666 nested_vmx_disable_intercept_for_msr(
9667 msr_bitmap_l1
, msr_bitmap_l0
,
9668 APIC_BASE_MSR
+ (APIC_TASKPRI
>> 4),
9669 MSR_TYPE_R
| MSR_TYPE_W
);
9671 if (nested_cpu_has_vid(vmcs12
)) {
9672 nested_vmx_disable_intercept_for_msr(
9673 msr_bitmap_l1
, msr_bitmap_l0
,
9674 APIC_BASE_MSR
+ (APIC_EOI
>> 4),
9676 nested_vmx_disable_intercept_for_msr(
9677 msr_bitmap_l1
, msr_bitmap_l0
,
9678 APIC_BASE_MSR
+ (APIC_SELF_IPI
>> 4),
9683 nested_release_page_clean(page
);
9688 static int nested_vmx_check_apicv_controls(struct kvm_vcpu
*vcpu
,
9689 struct vmcs12
*vmcs12
)
9691 if (!nested_cpu_has_virt_x2apic_mode(vmcs12
) &&
9692 !nested_cpu_has_apic_reg_virt(vmcs12
) &&
9693 !nested_cpu_has_vid(vmcs12
) &&
9694 !nested_cpu_has_posted_intr(vmcs12
))
9698 * If virtualize x2apic mode is enabled,
9699 * virtualize apic access must be disabled.
9701 if (nested_cpu_has_virt_x2apic_mode(vmcs12
) &&
9702 nested_cpu_has2(vmcs12
, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES
))
9706 * If virtual interrupt delivery is enabled,
9707 * we must exit on external interrupts.
9709 if (nested_cpu_has_vid(vmcs12
) &&
9710 !nested_exit_on_intr(vcpu
))
9714 * bits 15:8 should be zero in posted_intr_nv,
9715 * the descriptor address has been already checked
9716 * in nested_get_vmcs12_pages.
9718 if (nested_cpu_has_posted_intr(vmcs12
) &&
9719 (!nested_cpu_has_vid(vmcs12
) ||
9720 !nested_exit_intr_ack_set(vcpu
) ||
9721 vmcs12
->posted_intr_nv
& 0xff00))
9724 /* tpr shadow is needed by all apicv features. */
9725 if (!nested_cpu_has(vmcs12
, CPU_BASED_TPR_SHADOW
))
9731 static int nested_vmx_check_msr_switch(struct kvm_vcpu
*vcpu
,
9732 unsigned long count_field
,
9733 unsigned long addr_field
)
9738 if (vmcs12_read_any(vcpu
, count_field
, &count
) ||
9739 vmcs12_read_any(vcpu
, addr_field
, &addr
)) {
9745 maxphyaddr
= cpuid_maxphyaddr(vcpu
);
9746 if (!IS_ALIGNED(addr
, 16) || addr
>> maxphyaddr
||
9747 (addr
+ count
* sizeof(struct vmx_msr_entry
) - 1) >> maxphyaddr
) {
9748 pr_debug_ratelimited(
9749 "nVMX: invalid MSR switch (0x%lx, %d, %llu, 0x%08llx)",
9750 addr_field
, maxphyaddr
, count
, addr
);
9756 static int nested_vmx_check_msr_switch_controls(struct kvm_vcpu
*vcpu
,
9757 struct vmcs12
*vmcs12
)
9759 if (vmcs12
->vm_exit_msr_load_count
== 0 &&
9760 vmcs12
->vm_exit_msr_store_count
== 0 &&
9761 vmcs12
->vm_entry_msr_load_count
== 0)
9762 return 0; /* Fast path */
9763 if (nested_vmx_check_msr_switch(vcpu
, VM_EXIT_MSR_LOAD_COUNT
,
9764 VM_EXIT_MSR_LOAD_ADDR
) ||
9765 nested_vmx_check_msr_switch(vcpu
, VM_EXIT_MSR_STORE_COUNT
,
9766 VM_EXIT_MSR_STORE_ADDR
) ||
9767 nested_vmx_check_msr_switch(vcpu
, VM_ENTRY_MSR_LOAD_COUNT
,
9768 VM_ENTRY_MSR_LOAD_ADDR
))
9773 static int nested_vmx_check_pml_controls(struct kvm_vcpu
*vcpu
,
9774 struct vmcs12
*vmcs12
)
9776 u64 address
= vmcs12
->pml_address
;
9777 int maxphyaddr
= cpuid_maxphyaddr(vcpu
);
9779 if (nested_cpu_has2(vmcs12
, SECONDARY_EXEC_ENABLE_PML
)) {
9780 if (!nested_cpu_has_ept(vmcs12
) ||
9781 !IS_ALIGNED(address
, 4096) ||
9782 address
>> maxphyaddr
)
9789 static int nested_vmx_msr_check_common(struct kvm_vcpu
*vcpu
,
9790 struct vmx_msr_entry
*e
)
9792 /* x2APIC MSR accesses are not allowed */
9793 if (vcpu
->arch
.apic_base
& X2APIC_ENABLE
&& e
->index
>> 8 == 0x8)
9795 if (e
->index
== MSR_IA32_UCODE_WRITE
|| /* SDM Table 35-2 */
9796 e
->index
== MSR_IA32_UCODE_REV
)
9798 if (e
->reserved
!= 0)
9803 static int nested_vmx_load_msr_check(struct kvm_vcpu
*vcpu
,
9804 struct vmx_msr_entry
*e
)
9806 if (e
->index
== MSR_FS_BASE
||
9807 e
->index
== MSR_GS_BASE
||
9808 e
->index
== MSR_IA32_SMM_MONITOR_CTL
|| /* SMM is not supported */
9809 nested_vmx_msr_check_common(vcpu
, e
))
9814 static int nested_vmx_store_msr_check(struct kvm_vcpu
*vcpu
,
9815 struct vmx_msr_entry
*e
)
9817 if (e
->index
== MSR_IA32_SMBASE
|| /* SMM is not supported */
9818 nested_vmx_msr_check_common(vcpu
, e
))
9824 * Load guest's/host's msr at nested entry/exit.
9825 * return 0 for success, entry index for failure.
9827 static u32
nested_vmx_load_msr(struct kvm_vcpu
*vcpu
, u64 gpa
, u32 count
)
9830 struct vmx_msr_entry e
;
9831 struct msr_data msr
;
9833 msr
.host_initiated
= false;
9834 for (i
= 0; i
< count
; i
++) {
9835 if (kvm_vcpu_read_guest(vcpu
, gpa
+ i
* sizeof(e
),
9837 pr_debug_ratelimited(
9838 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9839 __func__
, i
, gpa
+ i
* sizeof(e
));
9842 if (nested_vmx_load_msr_check(vcpu
, &e
)) {
9843 pr_debug_ratelimited(
9844 "%s check failed (%u, 0x%x, 0x%x)\n",
9845 __func__
, i
, e
.index
, e
.reserved
);
9848 msr
.index
= e
.index
;
9850 if (kvm_set_msr(vcpu
, &msr
)) {
9851 pr_debug_ratelimited(
9852 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9853 __func__
, i
, e
.index
, e
.value
);
9862 static int nested_vmx_store_msr(struct kvm_vcpu
*vcpu
, u64 gpa
, u32 count
)
9865 struct vmx_msr_entry e
;
9867 for (i
= 0; i
< count
; i
++) {
9868 struct msr_data msr_info
;
9869 if (kvm_vcpu_read_guest(vcpu
,
9870 gpa
+ i
* sizeof(e
),
9871 &e
, 2 * sizeof(u32
))) {
9872 pr_debug_ratelimited(
9873 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9874 __func__
, i
, gpa
+ i
* sizeof(e
));
9877 if (nested_vmx_store_msr_check(vcpu
, &e
)) {
9878 pr_debug_ratelimited(
9879 "%s check failed (%u, 0x%x, 0x%x)\n",
9880 __func__
, i
, e
.index
, e
.reserved
);
9883 msr_info
.host_initiated
= false;
9884 msr_info
.index
= e
.index
;
9885 if (kvm_get_msr(vcpu
, &msr_info
)) {
9886 pr_debug_ratelimited(
9887 "%s cannot read MSR (%u, 0x%x)\n",
9888 __func__
, i
, e
.index
);
9891 if (kvm_vcpu_write_guest(vcpu
,
9892 gpa
+ i
* sizeof(e
) +
9893 offsetof(struct vmx_msr_entry
, value
),
9894 &msr_info
.data
, sizeof(msr_info
.data
))) {
9895 pr_debug_ratelimited(
9896 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9897 __func__
, i
, e
.index
, msr_info
.data
);
9904 static bool nested_cr3_valid(struct kvm_vcpu
*vcpu
, unsigned long val
)
9906 unsigned long invalid_mask
;
9908 invalid_mask
= (~0ULL) << cpuid_maxphyaddr(vcpu
);
9909 return (val
& invalid_mask
) == 0;
9913 * Load guest's/host's cr3 at nested entry/exit. nested_ept is true if we are
9914 * emulating VM entry into a guest with EPT enabled.
9915 * Returns 0 on success, 1 on failure. Invalid state exit qualification code
9916 * is assigned to entry_failure_code on failure.
9918 static int nested_vmx_load_cr3(struct kvm_vcpu
*vcpu
, unsigned long cr3
, bool nested_ept
,
9919 u32
*entry_failure_code
)
9921 if (cr3
!= kvm_read_cr3(vcpu
) || (!nested_ept
&& pdptrs_changed(vcpu
))) {
9922 if (!nested_cr3_valid(vcpu
, cr3
)) {
9923 *entry_failure_code
= ENTRY_FAIL_DEFAULT
;
9928 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
9929 * must not be dereferenced.
9931 if (!is_long_mode(vcpu
) && is_pae(vcpu
) && is_paging(vcpu
) &&
9933 if (!load_pdptrs(vcpu
, vcpu
->arch
.walk_mmu
, cr3
)) {
9934 *entry_failure_code
= ENTRY_FAIL_PDPTE
;
9939 vcpu
->arch
.cr3
= cr3
;
9940 __set_bit(VCPU_EXREG_CR3
, (ulong
*)&vcpu
->arch
.regs_avail
);
9943 kvm_mmu_reset_context(vcpu
);
9948 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
9949 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
9950 * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
9951 * guest in a way that will both be appropriate to L1's requests, and our
9952 * needs. In addition to modifying the active vmcs (which is vmcs02), this
9953 * function also has additional necessary side-effects, like setting various
9954 * vcpu->arch fields.
9955 * Returns 0 on success, 1 on failure. Invalid state exit qualification code
9956 * is assigned to entry_failure_code on failure.
9958 static int prepare_vmcs02(struct kvm_vcpu
*vcpu
, struct vmcs12
*vmcs12
,
9959 bool from_vmentry
, u32
*entry_failure_code
)
9961 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
9962 u32 exec_control
, vmcs12_exec_ctrl
;
9964 vmcs_write16(GUEST_ES_SELECTOR
, vmcs12
->guest_es_selector
);
9965 vmcs_write16(GUEST_CS_SELECTOR
, vmcs12
->guest_cs_selector
);
9966 vmcs_write16(GUEST_SS_SELECTOR
, vmcs12
->guest_ss_selector
);
9967 vmcs_write16(GUEST_DS_SELECTOR
, vmcs12
->guest_ds_selector
);
9968 vmcs_write16(GUEST_FS_SELECTOR
, vmcs12
->guest_fs_selector
);
9969 vmcs_write16(GUEST_GS_SELECTOR
, vmcs12
->guest_gs_selector
);
9970 vmcs_write16(GUEST_LDTR_SELECTOR
, vmcs12
->guest_ldtr_selector
);
9971 vmcs_write16(GUEST_TR_SELECTOR
, vmcs12
->guest_tr_selector
);
9972 vmcs_write32(GUEST_ES_LIMIT
, vmcs12
->guest_es_limit
);
9973 vmcs_write32(GUEST_CS_LIMIT
, vmcs12
->guest_cs_limit
);
9974 vmcs_write32(GUEST_SS_LIMIT
, vmcs12
->guest_ss_limit
);
9975 vmcs_write32(GUEST_DS_LIMIT
, vmcs12
->guest_ds_limit
);
9976 vmcs_write32(GUEST_FS_LIMIT
, vmcs12
->guest_fs_limit
);
9977 vmcs_write32(GUEST_GS_LIMIT
, vmcs12
->guest_gs_limit
);
9978 vmcs_write32(GUEST_LDTR_LIMIT
, vmcs12
->guest_ldtr_limit
);
9979 vmcs_write32(GUEST_TR_LIMIT
, vmcs12
->guest_tr_limit
);
9980 vmcs_write32(GUEST_GDTR_LIMIT
, vmcs12
->guest_gdtr_limit
);
9981 vmcs_write32(GUEST_IDTR_LIMIT
, vmcs12
->guest_idtr_limit
);
9982 vmcs_write32(GUEST_ES_AR_BYTES
, vmcs12
->guest_es_ar_bytes
);
9983 vmcs_write32(GUEST_CS_AR_BYTES
, vmcs12
->guest_cs_ar_bytes
);
9984 vmcs_write32(GUEST_SS_AR_BYTES
, vmcs12
->guest_ss_ar_bytes
);
9985 vmcs_write32(GUEST_DS_AR_BYTES
, vmcs12
->guest_ds_ar_bytes
);
9986 vmcs_write32(GUEST_FS_AR_BYTES
, vmcs12
->guest_fs_ar_bytes
);
9987 vmcs_write32(GUEST_GS_AR_BYTES
, vmcs12
->guest_gs_ar_bytes
);
9988 vmcs_write32(GUEST_LDTR_AR_BYTES
, vmcs12
->guest_ldtr_ar_bytes
);
9989 vmcs_write32(GUEST_TR_AR_BYTES
, vmcs12
->guest_tr_ar_bytes
);
9990 vmcs_writel(GUEST_ES_BASE
, vmcs12
->guest_es_base
);
9991 vmcs_writel(GUEST_CS_BASE
, vmcs12
->guest_cs_base
);
9992 vmcs_writel(GUEST_SS_BASE
, vmcs12
->guest_ss_base
);
9993 vmcs_writel(GUEST_DS_BASE
, vmcs12
->guest_ds_base
);
9994 vmcs_writel(GUEST_FS_BASE
, vmcs12
->guest_fs_base
);
9995 vmcs_writel(GUEST_GS_BASE
, vmcs12
->guest_gs_base
);
9996 vmcs_writel(GUEST_LDTR_BASE
, vmcs12
->guest_ldtr_base
);
9997 vmcs_writel(GUEST_TR_BASE
, vmcs12
->guest_tr_base
);
9998 vmcs_writel(GUEST_GDTR_BASE
, vmcs12
->guest_gdtr_base
);
9999 vmcs_writel(GUEST_IDTR_BASE
, vmcs12
->guest_idtr_base
);
10001 if (from_vmentry
&&
10002 (vmcs12
->vm_entry_controls
& VM_ENTRY_LOAD_DEBUG_CONTROLS
)) {
10003 kvm_set_dr(vcpu
, 7, vmcs12
->guest_dr7
);
10004 vmcs_write64(GUEST_IA32_DEBUGCTL
, vmcs12
->guest_ia32_debugctl
);
10006 kvm_set_dr(vcpu
, 7, vcpu
->arch
.dr7
);
10007 vmcs_write64(GUEST_IA32_DEBUGCTL
, vmx
->nested
.vmcs01_debugctl
);
10009 if (from_vmentry
) {
10010 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD
,
10011 vmcs12
->vm_entry_intr_info_field
);
10012 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE
,
10013 vmcs12
->vm_entry_exception_error_code
);
10014 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN
,
10015 vmcs12
->vm_entry_instruction_len
);
10016 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO
,
10017 vmcs12
->guest_interruptibility_info
);
10019 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD
, 0);
10021 vmcs_write32(GUEST_SYSENTER_CS
, vmcs12
->guest_sysenter_cs
);
10022 vmx_set_rflags(vcpu
, vmcs12
->guest_rflags
);
10023 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS
,
10024 vmcs12
->guest_pending_dbg_exceptions
);
10025 vmcs_writel(GUEST_SYSENTER_ESP
, vmcs12
->guest_sysenter_esp
);
10026 vmcs_writel(GUEST_SYSENTER_EIP
, vmcs12
->guest_sysenter_eip
);
10028 if (nested_cpu_has_xsaves(vmcs12
))
10029 vmcs_write64(XSS_EXIT_BITMAP
, vmcs12
->xss_exit_bitmap
);
10030 vmcs_write64(VMCS_LINK_POINTER
, -1ull);
10032 exec_control
= vmcs12
->pin_based_vm_exec_control
;
10034 /* Preemption timer setting is only taken from vmcs01. */
10035 exec_control
&= ~PIN_BASED_VMX_PREEMPTION_TIMER
;
10036 exec_control
|= vmcs_config
.pin_based_exec_ctrl
;
10037 if (vmx
->hv_deadline_tsc
== -1)
10038 exec_control
&= ~PIN_BASED_VMX_PREEMPTION_TIMER
;
10040 /* Posted interrupts setting is only taken from vmcs12. */
10041 if (nested_cpu_has_posted_intr(vmcs12
)) {
10043 * Note that we use L0's vector here and in
10044 * vmx_deliver_nested_posted_interrupt.
10046 vmx
->nested
.posted_intr_nv
= vmcs12
->posted_intr_nv
;
10047 vmx
->nested
.pi_pending
= false;
10048 vmcs_write16(POSTED_INTR_NV
, POSTED_INTR_VECTOR
);
10050 exec_control
&= ~PIN_BASED_POSTED_INTR
;
10053 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL
, exec_control
);
10055 vmx
->nested
.preemption_timer_expired
= false;
10056 if (nested_cpu_has_preemption_timer(vmcs12
))
10057 vmx_start_preemption_timer(vcpu
);
10060 * Whether page-faults are trapped is determined by a combination of
10061 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
10062 * If enable_ept, L0 doesn't care about page faults and we should
10063 * set all of these to L1's desires. However, if !enable_ept, L0 does
10064 * care about (at least some) page faults, and because it is not easy
10065 * (if at all possible?) to merge L0 and L1's desires, we simply ask
10066 * to exit on each and every L2 page fault. This is done by setting
10067 * MASK=MATCH=0 and (see below) EB.PF=1.
10068 * Note that below we don't need special code to set EB.PF beyond the
10069 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
10070 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
10071 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
10073 * A problem with this approach (when !enable_ept) is that L1 may be
10074 * injected with more page faults than it asked for. This could have
10075 * caused problems, but in practice existing hypervisors don't care.
10076 * To fix this, we will need to emulate the PFEC checking (on the L1
10077 * page tables), using walk_addr(), when injecting PFs to L1.
10079 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK
,
10080 enable_ept
? vmcs12
->page_fault_error_code_mask
: 0);
10081 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH
,
10082 enable_ept
? vmcs12
->page_fault_error_code_match
: 0);
10084 if (cpu_has_secondary_exec_ctrls()) {
10085 exec_control
= vmx_secondary_exec_control(vmx
);
10087 /* Take the following fields only from vmcs12 */
10088 exec_control
&= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES
|
10089 SECONDARY_EXEC_RDTSCP
|
10090 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY
|
10091 SECONDARY_EXEC_APIC_REGISTER_VIRT
);
10092 if (nested_cpu_has(vmcs12
,
10093 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS
)) {
10094 vmcs12_exec_ctrl
= vmcs12
->secondary_vm_exec_control
&
10095 ~SECONDARY_EXEC_ENABLE_PML
;
10096 exec_control
|= vmcs12_exec_ctrl
;
10099 if (exec_control
& SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY
) {
10100 vmcs_write64(EOI_EXIT_BITMAP0
,
10101 vmcs12
->eoi_exit_bitmap0
);
10102 vmcs_write64(EOI_EXIT_BITMAP1
,
10103 vmcs12
->eoi_exit_bitmap1
);
10104 vmcs_write64(EOI_EXIT_BITMAP2
,
10105 vmcs12
->eoi_exit_bitmap2
);
10106 vmcs_write64(EOI_EXIT_BITMAP3
,
10107 vmcs12
->eoi_exit_bitmap3
);
10108 vmcs_write16(GUEST_INTR_STATUS
,
10109 vmcs12
->guest_intr_status
);
10113 * Write an illegal value to APIC_ACCESS_ADDR. Later,
10114 * nested_get_vmcs12_pages will either fix it up or
10115 * remove the VM execution control.
10117 if (exec_control
& SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES
)
10118 vmcs_write64(APIC_ACCESS_ADDR
, -1ull);
10120 vmcs_write32(SECONDARY_VM_EXEC_CONTROL
, exec_control
);
10125 * Set host-state according to L0's settings (vmcs12 is irrelevant here)
10126 * Some constant fields are set here by vmx_set_constant_host_state().
10127 * Other fields are different per CPU, and will be set later when
10128 * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
10130 vmx_set_constant_host_state(vmx
);
10133 * Set the MSR load/store lists to match L0's settings.
10135 vmcs_write32(VM_EXIT_MSR_STORE_COUNT
, 0);
10136 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT
, vmx
->msr_autoload
.nr
);
10137 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR
, __pa(vmx
->msr_autoload
.host
));
10138 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT
, vmx
->msr_autoload
.nr
);
10139 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR
, __pa(vmx
->msr_autoload
.guest
));
10142 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
10143 * entry, but only if the current (host) sp changed from the value
10144 * we wrote last (vmx->host_rsp). This cache is no longer relevant
10145 * if we switch vmcs, and rather than hold a separate cache per vmcs,
10146 * here we just force the write to happen on entry.
10150 exec_control
= vmx_exec_control(vmx
); /* L0's desires */
10151 exec_control
&= ~CPU_BASED_VIRTUAL_INTR_PENDING
;
10152 exec_control
&= ~CPU_BASED_VIRTUAL_NMI_PENDING
;
10153 exec_control
&= ~CPU_BASED_TPR_SHADOW
;
10154 exec_control
|= vmcs12
->cpu_based_vm_exec_control
;
10157 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR. Later, if
10158 * nested_get_vmcs12_pages can't fix it up, the illegal value
10159 * will result in a VM entry failure.
10161 if (exec_control
& CPU_BASED_TPR_SHADOW
) {
10162 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR
, -1ull);
10163 vmcs_write32(TPR_THRESHOLD
, vmcs12
->tpr_threshold
);
10167 * Merging of IO bitmap not currently supported.
10168 * Rather, exit every time.
10170 exec_control
&= ~CPU_BASED_USE_IO_BITMAPS
;
10171 exec_control
|= CPU_BASED_UNCOND_IO_EXITING
;
10173 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL
, exec_control
);
10175 /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
10176 * bitwise-or of what L1 wants to trap for L2, and what we want to
10177 * trap. Note that CR0.TS also needs updating - we do this later.
10179 update_exception_bitmap(vcpu
);
10180 vcpu
->arch
.cr0_guest_owned_bits
&= ~vmcs12
->cr0_guest_host_mask
;
10181 vmcs_writel(CR0_GUEST_HOST_MASK
, ~vcpu
->arch
.cr0_guest_owned_bits
);
10183 /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
10184 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
10185 * bits are further modified by vmx_set_efer() below.
10187 vmcs_write32(VM_EXIT_CONTROLS
, vmcs_config
.vmexit_ctrl
);
10189 /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
10190 * emulated by vmx_set_efer(), below.
10192 vm_entry_controls_init(vmx
,
10193 (vmcs12
->vm_entry_controls
& ~VM_ENTRY_LOAD_IA32_EFER
&
10194 ~VM_ENTRY_IA32E_MODE
) |
10195 (vmcs_config
.vmentry_ctrl
& ~VM_ENTRY_IA32E_MODE
));
10197 if (from_vmentry
&&
10198 (vmcs12
->vm_entry_controls
& VM_ENTRY_LOAD_IA32_PAT
)) {
10199 vmcs_write64(GUEST_IA32_PAT
, vmcs12
->guest_ia32_pat
);
10200 vcpu
->arch
.pat
= vmcs12
->guest_ia32_pat
;
10201 } else if (vmcs_config
.vmentry_ctrl
& VM_ENTRY_LOAD_IA32_PAT
) {
10202 vmcs_write64(GUEST_IA32_PAT
, vmx
->vcpu
.arch
.pat
);
10205 set_cr4_guest_host_mask(vmx
);
10207 if (from_vmentry
&&
10208 vmcs12
->vm_entry_controls
& VM_ENTRY_LOAD_BNDCFGS
)
10209 vmcs_write64(GUEST_BNDCFGS
, vmcs12
->guest_bndcfgs
);
10211 if (vmcs12
->cpu_based_vm_exec_control
& CPU_BASED_USE_TSC_OFFSETING
)
10212 vmcs_write64(TSC_OFFSET
,
10213 vcpu
->arch
.tsc_offset
+ vmcs12
->tsc_offset
);
10215 vmcs_write64(TSC_OFFSET
, vcpu
->arch
.tsc_offset
);
10216 if (kvm_has_tsc_control
)
10217 decache_tsc_multiplier(vmx
);
10221 * There is no direct mapping between vpid02 and vpid12, the
10222 * vpid02 is per-vCPU for L0 and reused while the value of
10223 * vpid12 is changed w/ one invvpid during nested vmentry.
10224 * The vpid12 is allocated by L1 for L2, so it will not
10225 * influence global bitmap(for vpid01 and vpid02 allocation)
10226 * even if spawn a lot of nested vCPUs.
10228 if (nested_cpu_has_vpid(vmcs12
) && vmx
->nested
.vpid02
) {
10229 vmcs_write16(VIRTUAL_PROCESSOR_ID
, vmx
->nested
.vpid02
);
10230 if (vmcs12
->virtual_processor_id
!= vmx
->nested
.last_vpid
) {
10231 vmx
->nested
.last_vpid
= vmcs12
->virtual_processor_id
;
10232 __vmx_flush_tlb(vcpu
, to_vmx(vcpu
)->nested
.vpid02
);
10235 vmcs_write16(VIRTUAL_PROCESSOR_ID
, vmx
->vpid
);
10236 vmx_flush_tlb(vcpu
);
10243 * Conceptually we want to copy the PML address and index from
10244 * vmcs01 here, and then back to vmcs01 on nested vmexit. But,
10245 * since we always flush the log on each vmexit, this happens
10246 * to be equivalent to simply resetting the fields in vmcs02.
10248 ASSERT(vmx
->pml_pg
);
10249 vmcs_write64(PML_ADDRESS
, page_to_phys(vmx
->pml_pg
));
10250 vmcs_write16(GUEST_PML_INDEX
, PML_ENTITY_NUM
- 1);
10253 if (nested_cpu_has_ept(vmcs12
)) {
10254 if (nested_ept_init_mmu_context(vcpu
)) {
10255 *entry_failure_code
= ENTRY_FAIL_DEFAULT
;
10258 } else if (nested_cpu_has2(vmcs12
,
10259 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES
)) {
10260 vmx_flush_tlb_ept_only(vcpu
);
10264 * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
10265 * bits which we consider mandatory enabled.
10266 * The CR0_READ_SHADOW is what L2 should have expected to read given
10267 * the specifications by L1; It's not enough to take
10268 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
10269 * have more bits than L1 expected.
10271 vmx_set_cr0(vcpu
, vmcs12
->guest_cr0
);
10272 vmcs_writel(CR0_READ_SHADOW
, nested_read_cr0(vmcs12
));
10274 vmx_set_cr4(vcpu
, vmcs12
->guest_cr4
);
10275 vmcs_writel(CR4_READ_SHADOW
, nested_read_cr4(vmcs12
));
10277 if (from_vmentry
&&
10278 (vmcs12
->vm_entry_controls
& VM_ENTRY_LOAD_IA32_EFER
))
10279 vcpu
->arch
.efer
= vmcs12
->guest_ia32_efer
;
10280 else if (vmcs12
->vm_entry_controls
& VM_ENTRY_IA32E_MODE
)
10281 vcpu
->arch
.efer
|= (EFER_LMA
| EFER_LME
);
10283 vcpu
->arch
.efer
&= ~(EFER_LMA
| EFER_LME
);
10284 /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
10285 vmx_set_efer(vcpu
, vcpu
->arch
.efer
);
10287 /* Shadow page tables on either EPT or shadow page tables. */
10288 if (nested_vmx_load_cr3(vcpu
, vmcs12
->guest_cr3
, nested_cpu_has_ept(vmcs12
),
10289 entry_failure_code
))
10293 vcpu
->arch
.walk_mmu
->inject_page_fault
= vmx_inject_page_fault_nested
;
10296 * L1 may access the L2's PDPTR, so save them to construct vmcs12
10299 vmcs_write64(GUEST_PDPTR0
, vmcs12
->guest_pdptr0
);
10300 vmcs_write64(GUEST_PDPTR1
, vmcs12
->guest_pdptr1
);
10301 vmcs_write64(GUEST_PDPTR2
, vmcs12
->guest_pdptr2
);
10302 vmcs_write64(GUEST_PDPTR3
, vmcs12
->guest_pdptr3
);
10305 kvm_register_write(vcpu
, VCPU_REGS_RSP
, vmcs12
->guest_rsp
);
10306 kvm_register_write(vcpu
, VCPU_REGS_RIP
, vmcs12
->guest_rip
);
10310 static int check_vmentry_prereqs(struct kvm_vcpu
*vcpu
, struct vmcs12
*vmcs12
)
10312 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
10314 if (vmcs12
->guest_activity_state
!= GUEST_ACTIVITY_ACTIVE
&&
10315 vmcs12
->guest_activity_state
!= GUEST_ACTIVITY_HLT
)
10316 return VMXERR_ENTRY_INVALID_CONTROL_FIELD
;
10318 if (nested_vmx_check_io_bitmap_controls(vcpu
, vmcs12
))
10319 return VMXERR_ENTRY_INVALID_CONTROL_FIELD
;
10321 if (nested_vmx_check_msr_bitmap_controls(vcpu
, vmcs12
))
10322 return VMXERR_ENTRY_INVALID_CONTROL_FIELD
;
10324 if (nested_vmx_check_apicv_controls(vcpu
, vmcs12
))
10325 return VMXERR_ENTRY_INVALID_CONTROL_FIELD
;
10327 if (nested_vmx_check_msr_switch_controls(vcpu
, vmcs12
))
10328 return VMXERR_ENTRY_INVALID_CONTROL_FIELD
;
10330 if (nested_vmx_check_pml_controls(vcpu
, vmcs12
))
10331 return VMXERR_ENTRY_INVALID_CONTROL_FIELD
;
10333 if (!vmx_control_verify(vmcs12
->cpu_based_vm_exec_control
,
10334 vmx
->nested
.nested_vmx_procbased_ctls_low
,
10335 vmx
->nested
.nested_vmx_procbased_ctls_high
) ||
10336 (nested_cpu_has(vmcs12
, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS
) &&
10337 !vmx_control_verify(vmcs12
->secondary_vm_exec_control
,
10338 vmx
->nested
.nested_vmx_secondary_ctls_low
,
10339 vmx
->nested
.nested_vmx_secondary_ctls_high
)) ||
10340 !vmx_control_verify(vmcs12
->pin_based_vm_exec_control
,
10341 vmx
->nested
.nested_vmx_pinbased_ctls_low
,
10342 vmx
->nested
.nested_vmx_pinbased_ctls_high
) ||
10343 !vmx_control_verify(vmcs12
->vm_exit_controls
,
10344 vmx
->nested
.nested_vmx_exit_ctls_low
,
10345 vmx
->nested
.nested_vmx_exit_ctls_high
) ||
10346 !vmx_control_verify(vmcs12
->vm_entry_controls
,
10347 vmx
->nested
.nested_vmx_entry_ctls_low
,
10348 vmx
->nested
.nested_vmx_entry_ctls_high
))
10349 return VMXERR_ENTRY_INVALID_CONTROL_FIELD
;
10351 if (vmcs12
->cr3_target_count
> nested_cpu_vmx_misc_cr3_count(vcpu
))
10352 return VMXERR_ENTRY_INVALID_CONTROL_FIELD
;
10354 if (!nested_host_cr0_valid(vcpu
, vmcs12
->host_cr0
) ||
10355 !nested_host_cr4_valid(vcpu
, vmcs12
->host_cr4
) ||
10356 !nested_cr3_valid(vcpu
, vmcs12
->host_cr3
))
10357 return VMXERR_ENTRY_INVALID_HOST_STATE_FIELD
;
10362 static int check_vmentry_postreqs(struct kvm_vcpu
*vcpu
, struct vmcs12
*vmcs12
,
10367 *exit_qual
= ENTRY_FAIL_DEFAULT
;
10369 if (!nested_guest_cr0_valid(vcpu
, vmcs12
->guest_cr0
) ||
10370 !nested_guest_cr4_valid(vcpu
, vmcs12
->guest_cr4
))
10373 if (!nested_cpu_has2(vmcs12
, SECONDARY_EXEC_SHADOW_VMCS
) &&
10374 vmcs12
->vmcs_link_pointer
!= -1ull) {
10375 *exit_qual
= ENTRY_FAIL_VMCS_LINK_PTR
;
10380 * If the load IA32_EFER VM-entry control is 1, the following checks
10381 * are performed on the field for the IA32_EFER MSR:
10382 * - Bits reserved in the IA32_EFER MSR must be 0.
10383 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
10384 * the IA-32e mode guest VM-exit control. It must also be identical
10385 * to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
10388 if (to_vmx(vcpu
)->nested
.nested_run_pending
&&
10389 (vmcs12
->vm_entry_controls
& VM_ENTRY_LOAD_IA32_EFER
)) {
10390 ia32e
= (vmcs12
->vm_entry_controls
& VM_ENTRY_IA32E_MODE
) != 0;
10391 if (!kvm_valid_efer(vcpu
, vmcs12
->guest_ia32_efer
) ||
10392 ia32e
!= !!(vmcs12
->guest_ia32_efer
& EFER_LMA
) ||
10393 ((vmcs12
->guest_cr0
& X86_CR0_PG
) &&
10394 ia32e
!= !!(vmcs12
->guest_ia32_efer
& EFER_LME
)))
10399 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
10400 * IA32_EFER MSR must be 0 in the field for that register. In addition,
10401 * the values of the LMA and LME bits in the field must each be that of
10402 * the host address-space size VM-exit control.
10404 if (vmcs12
->vm_exit_controls
& VM_EXIT_LOAD_IA32_EFER
) {
10405 ia32e
= (vmcs12
->vm_exit_controls
&
10406 VM_EXIT_HOST_ADDR_SPACE_SIZE
) != 0;
10407 if (!kvm_valid_efer(vcpu
, vmcs12
->host_ia32_efer
) ||
10408 ia32e
!= !!(vmcs12
->host_ia32_efer
& EFER_LMA
) ||
10409 ia32e
!= !!(vmcs12
->host_ia32_efer
& EFER_LME
))
10416 static int enter_vmx_non_root_mode(struct kvm_vcpu
*vcpu
, bool from_vmentry
)
10418 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
10419 struct vmcs12
*vmcs12
= get_vmcs12(vcpu
);
10420 struct loaded_vmcs
*vmcs02
;
10424 vmcs02
= nested_get_current_vmcs02(vmx
);
10428 enter_guest_mode(vcpu
);
10430 if (!(vmcs12
->vm_entry_controls
& VM_ENTRY_LOAD_DEBUG_CONTROLS
))
10431 vmx
->nested
.vmcs01_debugctl
= vmcs_read64(GUEST_IA32_DEBUGCTL
);
10433 vmx_switch_vmcs(vcpu
, vmcs02
);
10434 vmx_segment_cache_clear(vmx
);
10436 if (prepare_vmcs02(vcpu
, vmcs12
, from_vmentry
, &exit_qual
)) {
10437 leave_guest_mode(vcpu
);
10438 vmx_switch_vmcs(vcpu
, &vmx
->vmcs01
);
10439 nested_vmx_entry_failure(vcpu
, vmcs12
,
10440 EXIT_REASON_INVALID_STATE
, exit_qual
);
10444 nested_get_vmcs12_pages(vcpu
, vmcs12
);
10446 msr_entry_idx
= nested_vmx_load_msr(vcpu
,
10447 vmcs12
->vm_entry_msr_load_addr
,
10448 vmcs12
->vm_entry_msr_load_count
);
10449 if (msr_entry_idx
) {
10450 leave_guest_mode(vcpu
);
10451 vmx_switch_vmcs(vcpu
, &vmx
->vmcs01
);
10452 nested_vmx_entry_failure(vcpu
, vmcs12
,
10453 EXIT_REASON_MSR_LOAD_FAIL
, msr_entry_idx
);
10458 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
10459 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
10460 * returned as far as L1 is concerned. It will only return (and set
10461 * the success flag) when L2 exits (see nested_vmx_vmexit()).
10467 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
10468 * for running an L2 nested guest.
10470 static int nested_vmx_run(struct kvm_vcpu
*vcpu
, bool launch
)
10472 struct vmcs12
*vmcs12
;
10473 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
10477 if (!nested_vmx_check_permission(vcpu
))
10480 if (!nested_vmx_check_vmcs12(vcpu
))
10483 vmcs12
= get_vmcs12(vcpu
);
10485 if (enable_shadow_vmcs
)
10486 copy_shadow_to_vmcs12(vmx
);
10489 * The nested entry process starts with enforcing various prerequisites
10490 * on vmcs12 as required by the Intel SDM, and act appropriately when
10491 * they fail: As the SDM explains, some conditions should cause the
10492 * instruction to fail, while others will cause the instruction to seem
10493 * to succeed, but return an EXIT_REASON_INVALID_STATE.
10494 * To speed up the normal (success) code path, we should avoid checking
10495 * for misconfigurations which will anyway be caught by the processor
10496 * when using the merged vmcs02.
10498 if (vmcs12
->launch_state
== launch
) {
10499 nested_vmx_failValid(vcpu
,
10500 launch
? VMXERR_VMLAUNCH_NONCLEAR_VMCS
10501 : VMXERR_VMRESUME_NONLAUNCHED_VMCS
);
10505 ret
= check_vmentry_prereqs(vcpu
, vmcs12
);
10507 nested_vmx_failValid(vcpu
, ret
);
10512 * After this point, the trap flag no longer triggers a singlestep trap
10513 * on the vm entry instructions; don't call kvm_skip_emulated_instruction.
10514 * This is not 100% correct; for performance reasons, we delegate most
10515 * of the checks on host state to the processor. If those fail,
10516 * the singlestep trap is missed.
10518 skip_emulated_instruction(vcpu
);
10520 ret
= check_vmentry_postreqs(vcpu
, vmcs12
, &exit_qual
);
10522 nested_vmx_entry_failure(vcpu
, vmcs12
,
10523 EXIT_REASON_INVALID_STATE
, exit_qual
);
10528 * We're finally done with prerequisite checking, and can start with
10529 * the nested entry.
10532 ret
= enter_vmx_non_root_mode(vcpu
, true);
10536 if (vmcs12
->guest_activity_state
== GUEST_ACTIVITY_HLT
)
10537 return kvm_vcpu_halt(vcpu
);
10539 vmx
->nested
.nested_run_pending
= 1;
10544 return kvm_skip_emulated_instruction(vcpu
);
10548 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
10549 * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
10550 * This function returns the new value we should put in vmcs12.guest_cr0.
10551 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
10552 * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
10553 * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
10554 * didn't trap the bit, because if L1 did, so would L0).
10555 * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have
10556 * been modified by L2, and L1 knows it. So just leave the old value of
10557 * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
10558 * isn't relevant, because if L0 traps this bit it can set it to anything.
10559 * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
10560 * changed these bits, and therefore they need to be updated, but L0
10561 * didn't necessarily allow them to be changed in GUEST_CR0 - and rather
10562 * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
10564 static inline unsigned long
10565 vmcs12_guest_cr0(struct kvm_vcpu
*vcpu
, struct vmcs12
*vmcs12
)
10568 /*1*/ (vmcs_readl(GUEST_CR0
) & vcpu
->arch
.cr0_guest_owned_bits
) |
10569 /*2*/ (vmcs12
->guest_cr0
& vmcs12
->cr0_guest_host_mask
) |
10570 /*3*/ (vmcs_readl(CR0_READ_SHADOW
) & ~(vmcs12
->cr0_guest_host_mask
|
10571 vcpu
->arch
.cr0_guest_owned_bits
));
10574 static inline unsigned long
10575 vmcs12_guest_cr4(struct kvm_vcpu
*vcpu
, struct vmcs12
*vmcs12
)
10578 /*1*/ (vmcs_readl(GUEST_CR4
) & vcpu
->arch
.cr4_guest_owned_bits
) |
10579 /*2*/ (vmcs12
->guest_cr4
& vmcs12
->cr4_guest_host_mask
) |
10580 /*3*/ (vmcs_readl(CR4_READ_SHADOW
) & ~(vmcs12
->cr4_guest_host_mask
|
10581 vcpu
->arch
.cr4_guest_owned_bits
));
10584 static void vmcs12_save_pending_event(struct kvm_vcpu
*vcpu
,
10585 struct vmcs12
*vmcs12
)
10590 if (vcpu
->arch
.exception
.pending
&& vcpu
->arch
.exception
.reinject
) {
10591 nr
= vcpu
->arch
.exception
.nr
;
10592 idt_vectoring
= nr
| VECTORING_INFO_VALID_MASK
;
10594 if (kvm_exception_is_soft(nr
)) {
10595 vmcs12
->vm_exit_instruction_len
=
10596 vcpu
->arch
.event_exit_inst_len
;
10597 idt_vectoring
|= INTR_TYPE_SOFT_EXCEPTION
;
10599 idt_vectoring
|= INTR_TYPE_HARD_EXCEPTION
;
10601 if (vcpu
->arch
.exception
.has_error_code
) {
10602 idt_vectoring
|= VECTORING_INFO_DELIVER_CODE_MASK
;
10603 vmcs12
->idt_vectoring_error_code
=
10604 vcpu
->arch
.exception
.error_code
;
10607 vmcs12
->idt_vectoring_info_field
= idt_vectoring
;
10608 } else if (vcpu
->arch
.nmi_injected
) {
10609 vmcs12
->idt_vectoring_info_field
=
10610 INTR_TYPE_NMI_INTR
| INTR_INFO_VALID_MASK
| NMI_VECTOR
;
10611 } else if (vcpu
->arch
.interrupt
.pending
) {
10612 nr
= vcpu
->arch
.interrupt
.nr
;
10613 idt_vectoring
= nr
| VECTORING_INFO_VALID_MASK
;
10615 if (vcpu
->arch
.interrupt
.soft
) {
10616 idt_vectoring
|= INTR_TYPE_SOFT_INTR
;
10617 vmcs12
->vm_entry_instruction_len
=
10618 vcpu
->arch
.event_exit_inst_len
;
10620 idt_vectoring
|= INTR_TYPE_EXT_INTR
;
10622 vmcs12
->idt_vectoring_info_field
= idt_vectoring
;
10626 static int vmx_check_nested_events(struct kvm_vcpu
*vcpu
, bool external_intr
)
10628 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
10630 if (vcpu
->arch
.exception
.pending
||
10631 vcpu
->arch
.nmi_injected
||
10632 vcpu
->arch
.interrupt
.pending
)
10635 if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu
)) &&
10636 vmx
->nested
.preemption_timer_expired
) {
10637 if (vmx
->nested
.nested_run_pending
)
10639 nested_vmx_vmexit(vcpu
, EXIT_REASON_PREEMPTION_TIMER
, 0, 0);
10643 if (vcpu
->arch
.nmi_pending
&& nested_exit_on_nmi(vcpu
)) {
10644 if (vmx
->nested
.nested_run_pending
)
10646 nested_vmx_vmexit(vcpu
, EXIT_REASON_EXCEPTION_NMI
,
10647 NMI_VECTOR
| INTR_TYPE_NMI_INTR
|
10648 INTR_INFO_VALID_MASK
, 0);
10650 * The NMI-triggered VM exit counts as injection:
10651 * clear this one and block further NMIs.
10653 vcpu
->arch
.nmi_pending
= 0;
10654 vmx_set_nmi_mask(vcpu
, true);
10658 if ((kvm_cpu_has_interrupt(vcpu
) || external_intr
) &&
10659 nested_exit_on_intr(vcpu
)) {
10660 if (vmx
->nested
.nested_run_pending
)
10662 nested_vmx_vmexit(vcpu
, EXIT_REASON_EXTERNAL_INTERRUPT
, 0, 0);
10666 vmx_complete_nested_posted_interrupt(vcpu
);
10670 static u32
vmx_get_preemption_timer_value(struct kvm_vcpu
*vcpu
)
10672 ktime_t remaining
=
10673 hrtimer_get_remaining(&to_vmx(vcpu
)->nested
.preemption_timer
);
10676 if (ktime_to_ns(remaining
) <= 0)
10679 value
= ktime_to_ns(remaining
) * vcpu
->arch
.virtual_tsc_khz
;
10680 do_div(value
, 1000000);
10681 return value
>> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE
;
10685 * Update the guest state fields of vmcs12 to reflect changes that
10686 * occurred while L2 was running. (The "IA-32e mode guest" bit of the
10687 * VM-entry controls is also updated, since this is really a guest
10690 static void sync_vmcs12(struct kvm_vcpu
*vcpu
, struct vmcs12
*vmcs12
)
10692 vmcs12
->guest_cr0
= vmcs12_guest_cr0(vcpu
, vmcs12
);
10693 vmcs12
->guest_cr4
= vmcs12_guest_cr4(vcpu
, vmcs12
);
10695 vmcs12
->guest_rsp
= kvm_register_read(vcpu
, VCPU_REGS_RSP
);
10696 vmcs12
->guest_rip
= kvm_register_read(vcpu
, VCPU_REGS_RIP
);
10697 vmcs12
->guest_rflags
= vmcs_readl(GUEST_RFLAGS
);
10699 vmcs12
->guest_es_selector
= vmcs_read16(GUEST_ES_SELECTOR
);
10700 vmcs12
->guest_cs_selector
= vmcs_read16(GUEST_CS_SELECTOR
);
10701 vmcs12
->guest_ss_selector
= vmcs_read16(GUEST_SS_SELECTOR
);
10702 vmcs12
->guest_ds_selector
= vmcs_read16(GUEST_DS_SELECTOR
);
10703 vmcs12
->guest_fs_selector
= vmcs_read16(GUEST_FS_SELECTOR
);
10704 vmcs12
->guest_gs_selector
= vmcs_read16(GUEST_GS_SELECTOR
);
10705 vmcs12
->guest_ldtr_selector
= vmcs_read16(GUEST_LDTR_SELECTOR
);
10706 vmcs12
->guest_tr_selector
= vmcs_read16(GUEST_TR_SELECTOR
);
10707 vmcs12
->guest_es_limit
= vmcs_read32(GUEST_ES_LIMIT
);
10708 vmcs12
->guest_cs_limit
= vmcs_read32(GUEST_CS_LIMIT
);
10709 vmcs12
->guest_ss_limit
= vmcs_read32(GUEST_SS_LIMIT
);
10710 vmcs12
->guest_ds_limit
= vmcs_read32(GUEST_DS_LIMIT
);
10711 vmcs12
->guest_fs_limit
= vmcs_read32(GUEST_FS_LIMIT
);
10712 vmcs12
->guest_gs_limit
= vmcs_read32(GUEST_GS_LIMIT
);
10713 vmcs12
->guest_ldtr_limit
= vmcs_read32(GUEST_LDTR_LIMIT
);
10714 vmcs12
->guest_tr_limit
= vmcs_read32(GUEST_TR_LIMIT
);
10715 vmcs12
->guest_gdtr_limit
= vmcs_read32(GUEST_GDTR_LIMIT
);
10716 vmcs12
->guest_idtr_limit
= vmcs_read32(GUEST_IDTR_LIMIT
);
10717 vmcs12
->guest_es_ar_bytes
= vmcs_read32(GUEST_ES_AR_BYTES
);
10718 vmcs12
->guest_cs_ar_bytes
= vmcs_read32(GUEST_CS_AR_BYTES
);
10719 vmcs12
->guest_ss_ar_bytes
= vmcs_read32(GUEST_SS_AR_BYTES
);
10720 vmcs12
->guest_ds_ar_bytes
= vmcs_read32(GUEST_DS_AR_BYTES
);
10721 vmcs12
->guest_fs_ar_bytes
= vmcs_read32(GUEST_FS_AR_BYTES
);
10722 vmcs12
->guest_gs_ar_bytes
= vmcs_read32(GUEST_GS_AR_BYTES
);
10723 vmcs12
->guest_ldtr_ar_bytes
= vmcs_read32(GUEST_LDTR_AR_BYTES
);
10724 vmcs12
->guest_tr_ar_bytes
= vmcs_read32(GUEST_TR_AR_BYTES
);
10725 vmcs12
->guest_es_base
= vmcs_readl(GUEST_ES_BASE
);
10726 vmcs12
->guest_cs_base
= vmcs_readl(GUEST_CS_BASE
);
10727 vmcs12
->guest_ss_base
= vmcs_readl(GUEST_SS_BASE
);
10728 vmcs12
->guest_ds_base
= vmcs_readl(GUEST_DS_BASE
);
10729 vmcs12
->guest_fs_base
= vmcs_readl(GUEST_FS_BASE
);
10730 vmcs12
->guest_gs_base
= vmcs_readl(GUEST_GS_BASE
);
10731 vmcs12
->guest_ldtr_base
= vmcs_readl(GUEST_LDTR_BASE
);
10732 vmcs12
->guest_tr_base
= vmcs_readl(GUEST_TR_BASE
);
10733 vmcs12
->guest_gdtr_base
= vmcs_readl(GUEST_GDTR_BASE
);
10734 vmcs12
->guest_idtr_base
= vmcs_readl(GUEST_IDTR_BASE
);
10736 vmcs12
->guest_interruptibility_info
=
10737 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO
);
10738 vmcs12
->guest_pending_dbg_exceptions
=
10739 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS
);
10740 if (vcpu
->arch
.mp_state
== KVM_MP_STATE_HALTED
)
10741 vmcs12
->guest_activity_state
= GUEST_ACTIVITY_HLT
;
10743 vmcs12
->guest_activity_state
= GUEST_ACTIVITY_ACTIVE
;
10745 if (nested_cpu_has_preemption_timer(vmcs12
)) {
10746 if (vmcs12
->vm_exit_controls
&
10747 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER
)
10748 vmcs12
->vmx_preemption_timer_value
=
10749 vmx_get_preemption_timer_value(vcpu
);
10750 hrtimer_cancel(&to_vmx(vcpu
)->nested
.preemption_timer
);
10754 * In some cases (usually, nested EPT), L2 is allowed to change its
10755 * own CR3 without exiting. If it has changed it, we must keep it.
10756 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
10757 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
10759 * Additionally, restore L2's PDPTR to vmcs12.
10762 vmcs12
->guest_cr3
= vmcs_readl(GUEST_CR3
);
10763 vmcs12
->guest_pdptr0
= vmcs_read64(GUEST_PDPTR0
);
10764 vmcs12
->guest_pdptr1
= vmcs_read64(GUEST_PDPTR1
);
10765 vmcs12
->guest_pdptr2
= vmcs_read64(GUEST_PDPTR2
);
10766 vmcs12
->guest_pdptr3
= vmcs_read64(GUEST_PDPTR3
);
10769 vmcs12
->guest_linear_address
= vmcs_readl(GUEST_LINEAR_ADDRESS
);
10771 if (nested_cpu_has_vid(vmcs12
))
10772 vmcs12
->guest_intr_status
= vmcs_read16(GUEST_INTR_STATUS
);
10774 vmcs12
->vm_entry_controls
=
10775 (vmcs12
->vm_entry_controls
& ~VM_ENTRY_IA32E_MODE
) |
10776 (vm_entry_controls_get(to_vmx(vcpu
)) & VM_ENTRY_IA32E_MODE
);
10778 if (vmcs12
->vm_exit_controls
& VM_EXIT_SAVE_DEBUG_CONTROLS
) {
10779 kvm_get_dr(vcpu
, 7, (unsigned long *)&vmcs12
->guest_dr7
);
10780 vmcs12
->guest_ia32_debugctl
= vmcs_read64(GUEST_IA32_DEBUGCTL
);
10783 /* TODO: These cannot have changed unless we have MSR bitmaps and
10784 * the relevant bit asks not to trap the change */
10785 if (vmcs12
->vm_exit_controls
& VM_EXIT_SAVE_IA32_PAT
)
10786 vmcs12
->guest_ia32_pat
= vmcs_read64(GUEST_IA32_PAT
);
10787 if (vmcs12
->vm_exit_controls
& VM_EXIT_SAVE_IA32_EFER
)
10788 vmcs12
->guest_ia32_efer
= vcpu
->arch
.efer
;
10789 vmcs12
->guest_sysenter_cs
= vmcs_read32(GUEST_SYSENTER_CS
);
10790 vmcs12
->guest_sysenter_esp
= vmcs_readl(GUEST_SYSENTER_ESP
);
10791 vmcs12
->guest_sysenter_eip
= vmcs_readl(GUEST_SYSENTER_EIP
);
10792 if (kvm_mpx_supported())
10793 vmcs12
->guest_bndcfgs
= vmcs_read64(GUEST_BNDCFGS
);
10797 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
10798 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
10799 * and this function updates it to reflect the changes to the guest state while
10800 * L2 was running (and perhaps made some exits which were handled directly by L0
10801 * without going back to L1), and to reflect the exit reason.
10802 * Note that we do not have to copy here all VMCS fields, just those that
10803 * could have changed by the L2 guest or the exit - i.e., the guest-state and
10804 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
10805 * which already writes to vmcs12 directly.
10807 static void prepare_vmcs12(struct kvm_vcpu
*vcpu
, struct vmcs12
*vmcs12
,
10808 u32 exit_reason
, u32 exit_intr_info
,
10809 unsigned long exit_qualification
)
10811 /* update guest state fields: */
10812 sync_vmcs12(vcpu
, vmcs12
);
10814 /* update exit information fields: */
10816 vmcs12
->vm_exit_reason
= exit_reason
;
10817 vmcs12
->exit_qualification
= exit_qualification
;
10819 vmcs12
->vm_exit_intr_info
= exit_intr_info
;
10820 if ((vmcs12
->vm_exit_intr_info
&
10821 (INTR_INFO_VALID_MASK
| INTR_INFO_DELIVER_CODE_MASK
)) ==
10822 (INTR_INFO_VALID_MASK
| INTR_INFO_DELIVER_CODE_MASK
))
10823 vmcs12
->vm_exit_intr_error_code
=
10824 vmcs_read32(VM_EXIT_INTR_ERROR_CODE
);
10825 vmcs12
->idt_vectoring_info_field
= 0;
10826 vmcs12
->vm_exit_instruction_len
= vmcs_read32(VM_EXIT_INSTRUCTION_LEN
);
10827 vmcs12
->vmx_instruction_info
= vmcs_read32(VMX_INSTRUCTION_INFO
);
10829 if (!(vmcs12
->vm_exit_reason
& VMX_EXIT_REASONS_FAILED_VMENTRY
)) {
10830 vmcs12
->launch_state
= 1;
10832 /* vm_entry_intr_info_field is cleared on exit. Emulate this
10833 * instead of reading the real value. */
10834 vmcs12
->vm_entry_intr_info_field
&= ~INTR_INFO_VALID_MASK
;
10837 * Transfer the event that L0 or L1 may wanted to inject into
10838 * L2 to IDT_VECTORING_INFO_FIELD.
10840 vmcs12_save_pending_event(vcpu
, vmcs12
);
10844 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
10845 * preserved above and would only end up incorrectly in L1.
10847 vcpu
->arch
.nmi_injected
= false;
10848 kvm_clear_exception_queue(vcpu
);
10849 kvm_clear_interrupt_queue(vcpu
);
10853 * A part of what we need to when the nested L2 guest exits and we want to
10854 * run its L1 parent, is to reset L1's guest state to the host state specified
10856 * This function is to be called not only on normal nested exit, but also on
10857 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
10858 * Failures During or After Loading Guest State").
10859 * This function should be called when the active VMCS is L1's (vmcs01).
10861 static void load_vmcs12_host_state(struct kvm_vcpu
*vcpu
,
10862 struct vmcs12
*vmcs12
)
10864 struct kvm_segment seg
;
10865 u32 entry_failure_code
;
10867 if (vmcs12
->vm_exit_controls
& VM_EXIT_LOAD_IA32_EFER
)
10868 vcpu
->arch
.efer
= vmcs12
->host_ia32_efer
;
10869 else if (vmcs12
->vm_exit_controls
& VM_EXIT_HOST_ADDR_SPACE_SIZE
)
10870 vcpu
->arch
.efer
|= (EFER_LMA
| EFER_LME
);
10872 vcpu
->arch
.efer
&= ~(EFER_LMA
| EFER_LME
);
10873 vmx_set_efer(vcpu
, vcpu
->arch
.efer
);
10875 kvm_register_write(vcpu
, VCPU_REGS_RSP
, vmcs12
->host_rsp
);
10876 kvm_register_write(vcpu
, VCPU_REGS_RIP
, vmcs12
->host_rip
);
10877 vmx_set_rflags(vcpu
, X86_EFLAGS_FIXED
);
10879 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
10880 * actually changed, because vmx_set_cr0 refers to efer set above.
10882 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
10883 * (KVM doesn't change it);
10885 vcpu
->arch
.cr0_guest_owned_bits
= X86_CR0_TS
;
10886 vmx_set_cr0(vcpu
, vmcs12
->host_cr0
);
10888 /* Same as above - no reason to call set_cr4_guest_host_mask(). */
10889 vcpu
->arch
.cr4_guest_owned_bits
= ~vmcs_readl(CR4_GUEST_HOST_MASK
);
10890 kvm_set_cr4(vcpu
, vmcs12
->host_cr4
);
10892 nested_ept_uninit_mmu_context(vcpu
);
10895 * Only PDPTE load can fail as the value of cr3 was checked on entry and
10896 * couldn't have changed.
10898 if (nested_vmx_load_cr3(vcpu
, vmcs12
->host_cr3
, false, &entry_failure_code
))
10899 nested_vmx_abort(vcpu
, VMX_ABORT_LOAD_HOST_PDPTE_FAIL
);
10902 vcpu
->arch
.walk_mmu
->inject_page_fault
= kvm_inject_page_fault
;
10906 * Trivially support vpid by letting L2s share their parent
10907 * L1's vpid. TODO: move to a more elaborate solution, giving
10908 * each L2 its own vpid and exposing the vpid feature to L1.
10910 vmx_flush_tlb(vcpu
);
10914 vmcs_write32(GUEST_SYSENTER_CS
, vmcs12
->host_ia32_sysenter_cs
);
10915 vmcs_writel(GUEST_SYSENTER_ESP
, vmcs12
->host_ia32_sysenter_esp
);
10916 vmcs_writel(GUEST_SYSENTER_EIP
, vmcs12
->host_ia32_sysenter_eip
);
10917 vmcs_writel(GUEST_IDTR_BASE
, vmcs12
->host_idtr_base
);
10918 vmcs_writel(GUEST_GDTR_BASE
, vmcs12
->host_gdtr_base
);
10920 /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1. */
10921 if (vmcs12
->vm_exit_controls
& VM_EXIT_CLEAR_BNDCFGS
)
10922 vmcs_write64(GUEST_BNDCFGS
, 0);
10924 if (vmcs12
->vm_exit_controls
& VM_EXIT_LOAD_IA32_PAT
) {
10925 vmcs_write64(GUEST_IA32_PAT
, vmcs12
->host_ia32_pat
);
10926 vcpu
->arch
.pat
= vmcs12
->host_ia32_pat
;
10928 if (vmcs12
->vm_exit_controls
& VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL
)
10929 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL
,
10930 vmcs12
->host_ia32_perf_global_ctrl
);
10932 /* Set L1 segment info according to Intel SDM
10933 27.5.2 Loading Host Segment and Descriptor-Table Registers */
10934 seg
= (struct kvm_segment
) {
10936 .limit
= 0xFFFFFFFF,
10937 .selector
= vmcs12
->host_cs_selector
,
10943 if (vmcs12
->vm_exit_controls
& VM_EXIT_HOST_ADDR_SPACE_SIZE
)
10947 vmx_set_segment(vcpu
, &seg
, VCPU_SREG_CS
);
10948 seg
= (struct kvm_segment
) {
10950 .limit
= 0xFFFFFFFF,
10957 seg
.selector
= vmcs12
->host_ds_selector
;
10958 vmx_set_segment(vcpu
, &seg
, VCPU_SREG_DS
);
10959 seg
.selector
= vmcs12
->host_es_selector
;
10960 vmx_set_segment(vcpu
, &seg
, VCPU_SREG_ES
);
10961 seg
.selector
= vmcs12
->host_ss_selector
;
10962 vmx_set_segment(vcpu
, &seg
, VCPU_SREG_SS
);
10963 seg
.selector
= vmcs12
->host_fs_selector
;
10964 seg
.base
= vmcs12
->host_fs_base
;
10965 vmx_set_segment(vcpu
, &seg
, VCPU_SREG_FS
);
10966 seg
.selector
= vmcs12
->host_gs_selector
;
10967 seg
.base
= vmcs12
->host_gs_base
;
10968 vmx_set_segment(vcpu
, &seg
, VCPU_SREG_GS
);
10969 seg
= (struct kvm_segment
) {
10970 .base
= vmcs12
->host_tr_base
,
10972 .selector
= vmcs12
->host_tr_selector
,
10976 vmx_set_segment(vcpu
, &seg
, VCPU_SREG_TR
);
10978 kvm_set_dr(vcpu
, 7, 0x400);
10979 vmcs_write64(GUEST_IA32_DEBUGCTL
, 0);
10981 if (cpu_has_vmx_msr_bitmap())
10982 vmx_set_msr_bitmap(vcpu
);
10984 if (nested_vmx_load_msr(vcpu
, vmcs12
->vm_exit_msr_load_addr
,
10985 vmcs12
->vm_exit_msr_load_count
))
10986 nested_vmx_abort(vcpu
, VMX_ABORT_LOAD_HOST_MSR_FAIL
);
10990 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
10991 * and modify vmcs12 to make it see what it would expect to see there if
10992 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
10994 static void nested_vmx_vmexit(struct kvm_vcpu
*vcpu
, u32 exit_reason
,
10995 u32 exit_intr_info
,
10996 unsigned long exit_qualification
)
10998 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
10999 struct vmcs12
*vmcs12
= get_vmcs12(vcpu
);
11000 u32 vm_inst_error
= 0;
11002 /* trying to cancel vmlaunch/vmresume is a bug */
11003 WARN_ON_ONCE(vmx
->nested
.nested_run_pending
);
11005 leave_guest_mode(vcpu
);
11006 prepare_vmcs12(vcpu
, vmcs12
, exit_reason
, exit_intr_info
,
11007 exit_qualification
);
11009 if (nested_vmx_store_msr(vcpu
, vmcs12
->vm_exit_msr_store_addr
,
11010 vmcs12
->vm_exit_msr_store_count
))
11011 nested_vmx_abort(vcpu
, VMX_ABORT_SAVE_GUEST_MSR_FAIL
);
11013 if (unlikely(vmx
->fail
))
11014 vm_inst_error
= vmcs_read32(VM_INSTRUCTION_ERROR
);
11016 vmx_switch_vmcs(vcpu
, &vmx
->vmcs01
);
11018 if ((exit_reason
== EXIT_REASON_EXTERNAL_INTERRUPT
)
11019 && nested_exit_intr_ack_set(vcpu
)) {
11020 int irq
= kvm_cpu_get_interrupt(vcpu
);
11022 vmcs12
->vm_exit_intr_info
= irq
|
11023 INTR_INFO_VALID_MASK
| INTR_TYPE_EXT_INTR
;
11026 trace_kvm_nested_vmexit_inject(vmcs12
->vm_exit_reason
,
11027 vmcs12
->exit_qualification
,
11028 vmcs12
->idt_vectoring_info_field
,
11029 vmcs12
->vm_exit_intr_info
,
11030 vmcs12
->vm_exit_intr_error_code
,
11033 vm_entry_controls_reset_shadow(vmx
);
11034 vm_exit_controls_reset_shadow(vmx
);
11035 vmx_segment_cache_clear(vmx
);
11037 /* if no vmcs02 cache requested, remove the one we used */
11038 if (VMCS02_POOL_SIZE
== 0)
11039 nested_free_vmcs02(vmx
, vmx
->nested
.current_vmptr
);
11041 load_vmcs12_host_state(vcpu
, vmcs12
);
11043 /* Update any VMCS fields that might have changed while L2 ran */
11044 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT
, vmx
->msr_autoload
.nr
);
11045 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT
, vmx
->msr_autoload
.nr
);
11046 vmcs_write64(TSC_OFFSET
, vcpu
->arch
.tsc_offset
);
11047 if (vmx
->hv_deadline_tsc
== -1)
11048 vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL
,
11049 PIN_BASED_VMX_PREEMPTION_TIMER
);
11051 vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL
,
11052 PIN_BASED_VMX_PREEMPTION_TIMER
);
11053 if (kvm_has_tsc_control
)
11054 decache_tsc_multiplier(vmx
);
11056 if (vmx
->nested
.change_vmcs01_virtual_x2apic_mode
) {
11057 vmx
->nested
.change_vmcs01_virtual_x2apic_mode
= false;
11058 vmx_set_virtual_x2apic_mode(vcpu
,
11059 vcpu
->arch
.apic_base
& X2APIC_ENABLE
);
11060 } else if (!nested_cpu_has_ept(vmcs12
) &&
11061 nested_cpu_has2(vmcs12
,
11062 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES
)) {
11063 vmx_flush_tlb_ept_only(vcpu
);
11066 /* This is needed for same reason as it was needed in prepare_vmcs02 */
11069 /* Unpin physical memory we referred to in vmcs02 */
11070 if (vmx
->nested
.apic_access_page
) {
11071 nested_release_page(vmx
->nested
.apic_access_page
);
11072 vmx
->nested
.apic_access_page
= NULL
;
11074 if (vmx
->nested
.virtual_apic_page
) {
11075 nested_release_page(vmx
->nested
.virtual_apic_page
);
11076 vmx
->nested
.virtual_apic_page
= NULL
;
11078 if (vmx
->nested
.pi_desc_page
) {
11079 kunmap(vmx
->nested
.pi_desc_page
);
11080 nested_release_page(vmx
->nested
.pi_desc_page
);
11081 vmx
->nested
.pi_desc_page
= NULL
;
11082 vmx
->nested
.pi_desc
= NULL
;
11086 * We are now running in L2, mmu_notifier will force to reload the
11087 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
11089 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD
, vcpu
);
11092 * Exiting from L2 to L1, we're now back to L1 which thinks it just
11093 * finished a VMLAUNCH or VMRESUME instruction, so we need to set the
11094 * success or failure flag accordingly.
11096 if (unlikely(vmx
->fail
)) {
11098 nested_vmx_failValid(vcpu
, vm_inst_error
);
11100 nested_vmx_succeed(vcpu
);
11101 if (enable_shadow_vmcs
)
11102 vmx
->nested
.sync_shadow_vmcs
= true;
11104 /* in case we halted in L2 */
11105 vcpu
->arch
.mp_state
= KVM_MP_STATE_RUNNABLE
;
11109 * Forcibly leave nested mode in order to be able to reset the VCPU later on.
11111 static void vmx_leave_nested(struct kvm_vcpu
*vcpu
)
11113 if (is_guest_mode(vcpu
)) {
11114 to_vmx(vcpu
)->nested
.nested_run_pending
= 0;
11115 nested_vmx_vmexit(vcpu
, -1, 0, 0);
11117 free_nested(to_vmx(vcpu
));
11121 * L1's failure to enter L2 is a subset of a normal exit, as explained in
11122 * 23.7 "VM-entry failures during or after loading guest state" (this also
11123 * lists the acceptable exit-reason and exit-qualification parameters).
11124 * It should only be called before L2 actually succeeded to run, and when
11125 * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
11127 static void nested_vmx_entry_failure(struct kvm_vcpu
*vcpu
,
11128 struct vmcs12
*vmcs12
,
11129 u32 reason
, unsigned long qualification
)
11131 load_vmcs12_host_state(vcpu
, vmcs12
);
11132 vmcs12
->vm_exit_reason
= reason
| VMX_EXIT_REASONS_FAILED_VMENTRY
;
11133 vmcs12
->exit_qualification
= qualification
;
11134 nested_vmx_succeed(vcpu
);
11135 if (enable_shadow_vmcs
)
11136 to_vmx(vcpu
)->nested
.sync_shadow_vmcs
= true;
11139 static int vmx_check_intercept(struct kvm_vcpu
*vcpu
,
11140 struct x86_instruction_info
*info
,
11141 enum x86_intercept_stage stage
)
11143 return X86EMUL_CONTINUE
;
11146 #ifdef CONFIG_X86_64
11147 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
11148 static inline int u64_shl_div_u64(u64 a
, unsigned int shift
,
11149 u64 divisor
, u64
*result
)
11151 u64 low
= a
<< shift
, high
= a
>> (64 - shift
);
11153 /* To avoid the overflow on divq */
11154 if (high
>= divisor
)
11157 /* Low hold the result, high hold rem which is discarded */
11158 asm("divq %2\n\t" : "=a" (low
), "=d" (high
) :
11159 "rm" (divisor
), "0" (low
), "1" (high
));
11165 static int vmx_set_hv_timer(struct kvm_vcpu
*vcpu
, u64 guest_deadline_tsc
)
11167 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
11168 u64 tscl
= rdtsc();
11169 u64 guest_tscl
= kvm_read_l1_tsc(vcpu
, tscl
);
11170 u64 delta_tsc
= max(guest_deadline_tsc
, guest_tscl
) - guest_tscl
;
11172 /* Convert to host delta tsc if tsc scaling is enabled */
11173 if (vcpu
->arch
.tsc_scaling_ratio
!= kvm_default_tsc_scaling_ratio
&&
11174 u64_shl_div_u64(delta_tsc
,
11175 kvm_tsc_scaling_ratio_frac_bits
,
11176 vcpu
->arch
.tsc_scaling_ratio
,
11181 * If the delta tsc can't fit in the 32 bit after the multi shift,
11182 * we can't use the preemption timer.
11183 * It's possible that it fits on later vmentries, but checking
11184 * on every vmentry is costly so we just use an hrtimer.
11186 if (delta_tsc
>> (cpu_preemption_timer_multi
+ 32))
11189 vmx
->hv_deadline_tsc
= tscl
+ delta_tsc
;
11190 vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL
,
11191 PIN_BASED_VMX_PREEMPTION_TIMER
);
11193 return delta_tsc
== 0;
11196 static void vmx_cancel_hv_timer(struct kvm_vcpu
*vcpu
)
11198 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
11199 vmx
->hv_deadline_tsc
= -1;
11200 vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL
,
11201 PIN_BASED_VMX_PREEMPTION_TIMER
);
11205 static void vmx_sched_in(struct kvm_vcpu
*vcpu
, int cpu
)
11208 shrink_ple_window(vcpu
);
11211 static void vmx_slot_enable_log_dirty(struct kvm
*kvm
,
11212 struct kvm_memory_slot
*slot
)
11214 kvm_mmu_slot_leaf_clear_dirty(kvm
, slot
);
11215 kvm_mmu_slot_largepage_remove_write_access(kvm
, slot
);
11218 static void vmx_slot_disable_log_dirty(struct kvm
*kvm
,
11219 struct kvm_memory_slot
*slot
)
11221 kvm_mmu_slot_set_dirty(kvm
, slot
);
11224 static void vmx_flush_log_dirty(struct kvm
*kvm
)
11226 kvm_flush_pml_buffers(kvm
);
11229 static int vmx_write_pml_buffer(struct kvm_vcpu
*vcpu
)
11231 struct vmcs12
*vmcs12
;
11232 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
11234 struct page
*page
= NULL
;
11237 if (is_guest_mode(vcpu
)) {
11238 WARN_ON_ONCE(vmx
->nested
.pml_full
);
11241 * Check if PML is enabled for the nested guest.
11242 * Whether eptp bit 6 is set is already checked
11243 * as part of A/D emulation.
11245 vmcs12
= get_vmcs12(vcpu
);
11246 if (!nested_cpu_has_pml(vmcs12
))
11249 if (vmcs12
->guest_pml_index
>= PML_ENTITY_NUM
) {
11250 vmx
->nested
.pml_full
= true;
11254 gpa
= vmcs_read64(GUEST_PHYSICAL_ADDRESS
) & ~0xFFFull
;
11256 page
= nested_get_page(vcpu
, vmcs12
->pml_address
);
11260 pml_address
= kmap(page
);
11261 pml_address
[vmcs12
->guest_pml_index
--] = gpa
;
11263 nested_release_page_clean(page
);
11269 static void vmx_enable_log_dirty_pt_masked(struct kvm
*kvm
,
11270 struct kvm_memory_slot
*memslot
,
11271 gfn_t offset
, unsigned long mask
)
11273 kvm_mmu_clear_dirty_pt_masked(kvm
, memslot
, offset
, mask
);
11277 * This routine does the following things for vCPU which is going
11278 * to be blocked if VT-d PI is enabled.
11279 * - Store the vCPU to the wakeup list, so when interrupts happen
11280 * we can find the right vCPU to wake up.
11281 * - Change the Posted-interrupt descriptor as below:
11282 * 'NDST' <-- vcpu->pre_pcpu
11283 * 'NV' <-- POSTED_INTR_WAKEUP_VECTOR
11284 * - If 'ON' is set during this process, which means at least one
11285 * interrupt is posted for this vCPU, we cannot block it, in
11286 * this case, return 1, otherwise, return 0.
11289 static int pi_pre_block(struct kvm_vcpu
*vcpu
)
11291 unsigned long flags
;
11293 struct pi_desc old
, new;
11294 struct pi_desc
*pi_desc
= vcpu_to_pi_desc(vcpu
);
11296 if (!kvm_arch_has_assigned_device(vcpu
->kvm
) ||
11297 !irq_remapping_cap(IRQ_POSTING_CAP
) ||
11298 !kvm_vcpu_apicv_active(vcpu
))
11301 vcpu
->pre_pcpu
= vcpu
->cpu
;
11302 spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock
,
11303 vcpu
->pre_pcpu
), flags
);
11304 list_add_tail(&vcpu
->blocked_vcpu_list
,
11305 &per_cpu(blocked_vcpu_on_cpu
,
11307 spin_unlock_irqrestore(&per_cpu(blocked_vcpu_on_cpu_lock
,
11308 vcpu
->pre_pcpu
), flags
);
11311 old
.control
= new.control
= pi_desc
->control
;
11314 * We should not block the vCPU if
11315 * an interrupt is posted for it.
11317 if (pi_test_on(pi_desc
) == 1) {
11318 spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock
,
11319 vcpu
->pre_pcpu
), flags
);
11320 list_del(&vcpu
->blocked_vcpu_list
);
11321 spin_unlock_irqrestore(
11322 &per_cpu(blocked_vcpu_on_cpu_lock
,
11323 vcpu
->pre_pcpu
), flags
);
11324 vcpu
->pre_pcpu
= -1;
11329 WARN((pi_desc
->sn
== 1),
11330 "Warning: SN field of posted-interrupts "
11331 "is set before blocking\n");
11334 * Since vCPU can be preempted during this process,
11335 * vcpu->cpu could be different with pre_pcpu, we
11336 * need to set pre_pcpu as the destination of wakeup
11337 * notification event, then we can find the right vCPU
11338 * to wakeup in wakeup handler if interrupts happen
11339 * when the vCPU is in blocked state.
11341 dest
= cpu_physical_id(vcpu
->pre_pcpu
);
11343 if (x2apic_enabled())
11346 new.ndst
= (dest
<< 8) & 0xFF00;
11348 /* set 'NV' to 'wakeup vector' */
11349 new.nv
= POSTED_INTR_WAKEUP_VECTOR
;
11350 } while (cmpxchg(&pi_desc
->control
, old
.control
,
11351 new.control
) != old
.control
);
11356 static int vmx_pre_block(struct kvm_vcpu
*vcpu
)
11358 if (pi_pre_block(vcpu
))
11361 if (kvm_lapic_hv_timer_in_use(vcpu
))
11362 kvm_lapic_switch_to_sw_timer(vcpu
);
11367 static void pi_post_block(struct kvm_vcpu
*vcpu
)
11369 struct pi_desc
*pi_desc
= vcpu_to_pi_desc(vcpu
);
11370 struct pi_desc old
, new;
11372 unsigned long flags
;
11374 if (!kvm_arch_has_assigned_device(vcpu
->kvm
) ||
11375 !irq_remapping_cap(IRQ_POSTING_CAP
) ||
11376 !kvm_vcpu_apicv_active(vcpu
))
11380 old
.control
= new.control
= pi_desc
->control
;
11382 dest
= cpu_physical_id(vcpu
->cpu
);
11384 if (x2apic_enabled())
11387 new.ndst
= (dest
<< 8) & 0xFF00;
11389 /* Allow posting non-urgent interrupts */
11392 /* set 'NV' to 'notification vector' */
11393 new.nv
= POSTED_INTR_VECTOR
;
11394 } while (cmpxchg(&pi_desc
->control
, old
.control
,
11395 new.control
) != old
.control
);
11397 if(vcpu
->pre_pcpu
!= -1) {
11399 &per_cpu(blocked_vcpu_on_cpu_lock
,
11400 vcpu
->pre_pcpu
), flags
);
11401 list_del(&vcpu
->blocked_vcpu_list
);
11402 spin_unlock_irqrestore(
11403 &per_cpu(blocked_vcpu_on_cpu_lock
,
11404 vcpu
->pre_pcpu
), flags
);
11405 vcpu
->pre_pcpu
= -1;
11409 static void vmx_post_block(struct kvm_vcpu
*vcpu
)
11411 if (kvm_x86_ops
->set_hv_timer
)
11412 kvm_lapic_switch_to_hv_timer(vcpu
);
11414 pi_post_block(vcpu
);
11418 * vmx_update_pi_irte - set IRTE for Posted-Interrupts
11421 * @host_irq: host irq of the interrupt
11422 * @guest_irq: gsi of the interrupt
11423 * @set: set or unset PI
11424 * returns 0 on success, < 0 on failure
11426 static int vmx_update_pi_irte(struct kvm
*kvm
, unsigned int host_irq
,
11427 uint32_t guest_irq
, bool set
)
11429 struct kvm_kernel_irq_routing_entry
*e
;
11430 struct kvm_irq_routing_table
*irq_rt
;
11431 struct kvm_lapic_irq irq
;
11432 struct kvm_vcpu
*vcpu
;
11433 struct vcpu_data vcpu_info
;
11434 int idx
, ret
= -EINVAL
;
11436 if (!kvm_arch_has_assigned_device(kvm
) ||
11437 !irq_remapping_cap(IRQ_POSTING_CAP
) ||
11438 !kvm_vcpu_apicv_active(kvm
->vcpus
[0]))
11441 idx
= srcu_read_lock(&kvm
->irq_srcu
);
11442 irq_rt
= srcu_dereference(kvm
->irq_routing
, &kvm
->irq_srcu
);
11443 BUG_ON(guest_irq
>= irq_rt
->nr_rt_entries
);
11445 hlist_for_each_entry(e
, &irq_rt
->map
[guest_irq
], link
) {
11446 if (e
->type
!= KVM_IRQ_ROUTING_MSI
)
11449 * VT-d PI cannot support posting multicast/broadcast
11450 * interrupts to a vCPU, we still use interrupt remapping
11451 * for these kind of interrupts.
11453 * For lowest-priority interrupts, we only support
11454 * those with single CPU as the destination, e.g. user
11455 * configures the interrupts via /proc/irq or uses
11456 * irqbalance to make the interrupts single-CPU.
11458 * We will support full lowest-priority interrupt later.
11461 kvm_set_msi_irq(kvm
, e
, &irq
);
11462 if (!kvm_intr_is_single_vcpu(kvm
, &irq
, &vcpu
)) {
11464 * Make sure the IRTE is in remapped mode if
11465 * we don't handle it in posted mode.
11467 ret
= irq_set_vcpu_affinity(host_irq
, NULL
);
11470 "failed to back to remapped mode, irq: %u\n",
11478 vcpu_info
.pi_desc_addr
= __pa(vcpu_to_pi_desc(vcpu
));
11479 vcpu_info
.vector
= irq
.vector
;
11481 trace_kvm_pi_irte_update(vcpu
->vcpu_id
, host_irq
, e
->gsi
,
11482 vcpu_info
.vector
, vcpu_info
.pi_desc_addr
, set
);
11485 ret
= irq_set_vcpu_affinity(host_irq
, &vcpu_info
);
11487 /* suppress notification event before unposting */
11488 pi_set_sn(vcpu_to_pi_desc(vcpu
));
11489 ret
= irq_set_vcpu_affinity(host_irq
, NULL
);
11490 pi_clear_sn(vcpu_to_pi_desc(vcpu
));
11494 printk(KERN_INFO
"%s: failed to update PI IRTE\n",
11502 srcu_read_unlock(&kvm
->irq_srcu
, idx
);
11506 static void vmx_setup_mce(struct kvm_vcpu
*vcpu
)
11508 if (vcpu
->arch
.mcg_cap
& MCG_LMCE_P
)
11509 to_vmx(vcpu
)->msr_ia32_feature_control_valid_bits
|=
11510 FEATURE_CONTROL_LMCE
;
11512 to_vmx(vcpu
)->msr_ia32_feature_control_valid_bits
&=
11513 ~FEATURE_CONTROL_LMCE
;
11516 static struct kvm_x86_ops vmx_x86_ops __ro_after_init
= {
11517 .cpu_has_kvm_support
= cpu_has_kvm_support
,
11518 .disabled_by_bios
= vmx_disabled_by_bios
,
11519 .hardware_setup
= hardware_setup
,
11520 .hardware_unsetup
= hardware_unsetup
,
11521 .check_processor_compatibility
= vmx_check_processor_compat
,
11522 .hardware_enable
= hardware_enable
,
11523 .hardware_disable
= hardware_disable
,
11524 .cpu_has_accelerated_tpr
= report_flexpriority
,
11525 .cpu_has_high_real_mode_segbase
= vmx_has_high_real_mode_segbase
,
11527 .vcpu_create
= vmx_create_vcpu
,
11528 .vcpu_free
= vmx_free_vcpu
,
11529 .vcpu_reset
= vmx_vcpu_reset
,
11531 .prepare_guest_switch
= vmx_save_host_state
,
11532 .vcpu_load
= vmx_vcpu_load
,
11533 .vcpu_put
= vmx_vcpu_put
,
11535 .update_bp_intercept
= update_exception_bitmap
,
11536 .get_msr
= vmx_get_msr
,
11537 .set_msr
= vmx_set_msr
,
11538 .get_segment_base
= vmx_get_segment_base
,
11539 .get_segment
= vmx_get_segment
,
11540 .set_segment
= vmx_set_segment
,
11541 .get_cpl
= vmx_get_cpl
,
11542 .get_cs_db_l_bits
= vmx_get_cs_db_l_bits
,
11543 .decache_cr0_guest_bits
= vmx_decache_cr0_guest_bits
,
11544 .decache_cr3
= vmx_decache_cr3
,
11545 .decache_cr4_guest_bits
= vmx_decache_cr4_guest_bits
,
11546 .set_cr0
= vmx_set_cr0
,
11547 .set_cr3
= vmx_set_cr3
,
11548 .set_cr4
= vmx_set_cr4
,
11549 .set_efer
= vmx_set_efer
,
11550 .get_idt
= vmx_get_idt
,
11551 .set_idt
= vmx_set_idt
,
11552 .get_gdt
= vmx_get_gdt
,
11553 .set_gdt
= vmx_set_gdt
,
11554 .get_dr6
= vmx_get_dr6
,
11555 .set_dr6
= vmx_set_dr6
,
11556 .set_dr7
= vmx_set_dr7
,
11557 .sync_dirty_debug_regs
= vmx_sync_dirty_debug_regs
,
11558 .cache_reg
= vmx_cache_reg
,
11559 .get_rflags
= vmx_get_rflags
,
11560 .set_rflags
= vmx_set_rflags
,
11562 .get_pkru
= vmx_get_pkru
,
11564 .tlb_flush
= vmx_flush_tlb
,
11566 .run
= vmx_vcpu_run
,
11567 .handle_exit
= vmx_handle_exit
,
11568 .skip_emulated_instruction
= skip_emulated_instruction
,
11569 .set_interrupt_shadow
= vmx_set_interrupt_shadow
,
11570 .get_interrupt_shadow
= vmx_get_interrupt_shadow
,
11571 .patch_hypercall
= vmx_patch_hypercall
,
11572 .set_irq
= vmx_inject_irq
,
11573 .set_nmi
= vmx_inject_nmi
,
11574 .queue_exception
= vmx_queue_exception
,
11575 .cancel_injection
= vmx_cancel_injection
,
11576 .interrupt_allowed
= vmx_interrupt_allowed
,
11577 .nmi_allowed
= vmx_nmi_allowed
,
11578 .get_nmi_mask
= vmx_get_nmi_mask
,
11579 .set_nmi_mask
= vmx_set_nmi_mask
,
11580 .enable_nmi_window
= enable_nmi_window
,
11581 .enable_irq_window
= enable_irq_window
,
11582 .update_cr8_intercept
= update_cr8_intercept
,
11583 .set_virtual_x2apic_mode
= vmx_set_virtual_x2apic_mode
,
11584 .set_apic_access_page_addr
= vmx_set_apic_access_page_addr
,
11585 .get_enable_apicv
= vmx_get_enable_apicv
,
11586 .refresh_apicv_exec_ctrl
= vmx_refresh_apicv_exec_ctrl
,
11587 .load_eoi_exitmap
= vmx_load_eoi_exitmap
,
11588 .apicv_post_state_restore
= vmx_apicv_post_state_restore
,
11589 .hwapic_irr_update
= vmx_hwapic_irr_update
,
11590 .hwapic_isr_update
= vmx_hwapic_isr_update
,
11591 .sync_pir_to_irr
= vmx_sync_pir_to_irr
,
11592 .deliver_posted_interrupt
= vmx_deliver_posted_interrupt
,
11594 .set_tss_addr
= vmx_set_tss_addr
,
11595 .get_tdp_level
= get_ept_level
,
11596 .get_mt_mask
= vmx_get_mt_mask
,
11598 .get_exit_info
= vmx_get_exit_info
,
11600 .get_lpage_level
= vmx_get_lpage_level
,
11602 .cpuid_update
= vmx_cpuid_update
,
11604 .rdtscp_supported
= vmx_rdtscp_supported
,
11605 .invpcid_supported
= vmx_invpcid_supported
,
11607 .set_supported_cpuid
= vmx_set_supported_cpuid
,
11609 .has_wbinvd_exit
= cpu_has_vmx_wbinvd_exit
,
11611 .write_tsc_offset
= vmx_write_tsc_offset
,
11613 .set_tdp_cr3
= vmx_set_cr3
,
11615 .check_intercept
= vmx_check_intercept
,
11616 .handle_external_intr
= vmx_handle_external_intr
,
11617 .mpx_supported
= vmx_mpx_supported
,
11618 .xsaves_supported
= vmx_xsaves_supported
,
11620 .check_nested_events
= vmx_check_nested_events
,
11622 .sched_in
= vmx_sched_in
,
11624 .slot_enable_log_dirty
= vmx_slot_enable_log_dirty
,
11625 .slot_disable_log_dirty
= vmx_slot_disable_log_dirty
,
11626 .flush_log_dirty
= vmx_flush_log_dirty
,
11627 .enable_log_dirty_pt_masked
= vmx_enable_log_dirty_pt_masked
,
11628 .write_log_dirty
= vmx_write_pml_buffer
,
11630 .pre_block
= vmx_pre_block
,
11631 .post_block
= vmx_post_block
,
11633 .pmu_ops
= &intel_pmu_ops
,
11635 .update_pi_irte
= vmx_update_pi_irte
,
11637 #ifdef CONFIG_X86_64
11638 .set_hv_timer
= vmx_set_hv_timer
,
11639 .cancel_hv_timer
= vmx_cancel_hv_timer
,
11642 .setup_mce
= vmx_setup_mce
,
11645 static int __init
vmx_init(void)
11647 int r
= kvm_init(&vmx_x86_ops
, sizeof(struct vcpu_vmx
),
11648 __alignof__(struct vcpu_vmx
), THIS_MODULE
);
11652 #ifdef CONFIG_KEXEC_CORE
11653 rcu_assign_pointer(crash_vmclear_loaded_vmcss
,
11654 crash_vmclear_local_loaded_vmcss
);
11660 static void __exit
vmx_exit(void)
11662 #ifdef CONFIG_KEXEC_CORE
11663 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss
, NULL
);
11670 module_init(vmx_init
)
11671 module_exit(vmx_exit
)