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 <linux/nospec.h>
38 #include "kvm_cache_regs.h"
45 #include <asm/virtext.h>
47 #include <asm/fpu/internal.h>
48 #include <asm/perf_event.h>
49 #include <asm/debugreg.h>
50 #include <asm/kexec.h>
52 #include <asm/irq_remapping.h>
53 #include <asm/mmu_context.h>
54 #include <asm/microcode.h>
55 #include <asm/spec-ctrl.h>
60 #define __ex(x) __kvm_handle_fault_on_reboot(x)
61 #define __ex_clear(x, reg) \
62 ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
67 static const struct x86_cpu_id vmx_cpu_id[] = {
68 X86_FEATURE_MATCH(X86_FEATURE_VMX),
71 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
73 static bool __read_mostly enable_vpid = 1;
74 module_param_named(vpid, enable_vpid, bool, 0444);
76 static bool __read_mostly flexpriority_enabled = 1;
77 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
79 static bool __read_mostly enable_ept = 1;
80 module_param_named(ept, enable_ept, bool, S_IRUGO);
82 static bool __read_mostly enable_unrestricted_guest = 1;
83 module_param_named(unrestricted_guest,
84 enable_unrestricted_guest, bool, S_IRUGO);
86 static bool __read_mostly enable_ept_ad_bits = 1;
87 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
89 static bool __read_mostly emulate_invalid_guest_state = true;
90 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
92 static bool __read_mostly fasteoi = 1;
93 module_param(fasteoi, bool, S_IRUGO);
95 static bool __read_mostly enable_apicv = 1;
96 module_param(enable_apicv, bool, S_IRUGO);
98 static bool __read_mostly enable_shadow_vmcs = 1;
99 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
101 * If nested=1, nested virtualization is supported, i.e., guests may use
102 * VMX and be a hypervisor for its own guests. If nested=0, guests may not
103 * use VMX instructions.
105 static bool __read_mostly nested = 0;
106 module_param(nested, bool, S_IRUGO);
108 static u64 __read_mostly host_xss;
110 static bool __read_mostly enable_pml = 1;
111 module_param_named(pml, enable_pml, bool, S_IRUGO);
115 #define MSR_TYPE_RW 3
117 #define MSR_BITMAP_MODE_X2APIC 1
118 #define MSR_BITMAP_MODE_X2APIC_APICV 2
119 #define MSR_BITMAP_MODE_LM 4
121 #define KVM_VMX_TSC_MULTIPLIER_MAX 0xffffffffffffffffULL
123 /* Guest_tsc -> host_tsc conversion requires 64-bit division. */
124 static int __read_mostly cpu_preemption_timer_multi;
125 static bool __read_mostly enable_preemption_timer = 1;
127 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
130 #define KVM_GUEST_CR0_MASK (X86_CR0_NW | X86_CR0_CD)
131 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST (X86_CR0_WP | X86_CR0_NE)
132 #define KVM_VM_CR0_ALWAYS_ON \
133 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
134 #define KVM_CR4_GUEST_OWNED_BITS \
135 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
136 | X86_CR4_OSXMMEXCPT | X86_CR4_LA57 | X86_CR4_TSD)
138 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
139 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
141 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
143 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
146 * Hyper-V requires all of these, so mark them as supported even though
147 * they are just treated the same as all-context.
149 #define VMX_VPID_EXTENT_SUPPORTED_MASK \
150 (VMX_VPID_EXTENT_INDIVIDUAL_ADDR_BIT | \
151 VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT | \
152 VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT | \
153 VMX_VPID_EXTENT_SINGLE_NON_GLOBAL_BIT)
156 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
157 * ple_gap: upper bound on the amount of time between two successive
158 * executions of PAUSE in a loop. Also indicate if ple enabled.
159 * According to test, this time is usually smaller than 128 cycles.
160 * ple_window: upper bound on the amount of time a guest is allowed to execute
161 * in a PAUSE loop. Tests indicate that most spinlocks are held for
162 * less than 2^12 cycles
163 * Time is measured based on a counter that runs at the same rate as the TSC,
164 * refer SDM volume 3b section 21.6.13 & 22.1.3.
166 #define KVM_VMX_DEFAULT_PLE_GAP 128
167 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
168 #define KVM_VMX_DEFAULT_PLE_WINDOW_GROW 2
169 #define KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK 0
170 #define KVM_VMX_DEFAULT_PLE_WINDOW_MAX \
171 INT_MAX / KVM_VMX_DEFAULT_PLE_WINDOW_GROW
173 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
174 module_param(ple_gap, int, S_IRUGO);
176 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
177 module_param(ple_window, int, S_IRUGO);
179 /* Default doubles per-vcpu window every exit. */
180 static int ple_window_grow = KVM_VMX_DEFAULT_PLE_WINDOW_GROW;
181 module_param(ple_window_grow, int, S_IRUGO);
183 /* Default resets per-vcpu window every exit to ple_window. */
184 static int ple_window_shrink = KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK;
185 module_param(ple_window_shrink, int, S_IRUGO);
187 /* Default is to compute the maximum so we can never overflow. */
188 static int ple_window_actual_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
189 static int ple_window_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
190 module_param(ple_window_max, int, S_IRUGO);
192 extern const ulong vmx_return;
194 #define NR_AUTOLOAD_MSRS 8
203 * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
204 * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
205 * loaded on this CPU (so we can clear them if the CPU goes down).
209 struct vmcs *shadow_vmcs;
212 bool nmi_known_unmasked;
213 unsigned long vmcs_host_cr3; /* May not match real cr3 */
214 unsigned long vmcs_host_cr4; /* May not match real cr4 */
215 /* Support for vnmi-less CPUs */
216 int soft_vnmi_blocked;
218 s64 vnmi_blocked_time;
219 unsigned long *msr_bitmap;
220 struct list_head loaded_vmcss_on_cpu_link;
223 struct shared_msr_entry {
230 * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
231 * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
232 * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
233 * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
234 * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
235 * More than one of these structures may exist, if L1 runs multiple L2 guests.
236 * nested_vmx_run() will use the data here to build the vmcs02: a VMCS for the
237 * underlying hardware which will be used to run L2.
238 * This structure is packed to ensure that its layout is identical across
239 * machines (necessary for live migration).
240 * If there are changes in this struct, VMCS12_REVISION must be changed.
242 typedef u64 natural_width;
243 struct __packed vmcs12 {
244 /* According to the Intel spec, a VMCS region must start with the
245 * following two fields. Then follow implementation-specific data.
250 u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
251 u32 padding[7]; /* room for future expansion */
256 u64 vm_exit_msr_store_addr;
257 u64 vm_exit_msr_load_addr;
258 u64 vm_entry_msr_load_addr;
260 u64 virtual_apic_page_addr;
261 u64 apic_access_addr;
262 u64 posted_intr_desc_addr;
263 u64 vm_function_control;
265 u64 eoi_exit_bitmap0;
266 u64 eoi_exit_bitmap1;
267 u64 eoi_exit_bitmap2;
268 u64 eoi_exit_bitmap3;
269 u64 eptp_list_address;
271 u64 guest_physical_address;
272 u64 vmcs_link_pointer;
274 u64 guest_ia32_debugctl;
277 u64 guest_ia32_perf_global_ctrl;
285 u64 host_ia32_perf_global_ctrl;
286 u64 padding64[8]; /* room for future expansion */
288 * To allow migration of L1 (complete with its L2 guests) between
289 * machines of different natural widths (32 or 64 bit), we cannot have
290 * unsigned long fields with no explict size. We use u64 (aliased
291 * natural_width) instead. Luckily, x86 is little-endian.
293 natural_width cr0_guest_host_mask;
294 natural_width cr4_guest_host_mask;
295 natural_width cr0_read_shadow;
296 natural_width cr4_read_shadow;
297 natural_width cr3_target_value0;
298 natural_width cr3_target_value1;
299 natural_width cr3_target_value2;
300 natural_width cr3_target_value3;
301 natural_width exit_qualification;
302 natural_width guest_linear_address;
303 natural_width guest_cr0;
304 natural_width guest_cr3;
305 natural_width guest_cr4;
306 natural_width guest_es_base;
307 natural_width guest_cs_base;
308 natural_width guest_ss_base;
309 natural_width guest_ds_base;
310 natural_width guest_fs_base;
311 natural_width guest_gs_base;
312 natural_width guest_ldtr_base;
313 natural_width guest_tr_base;
314 natural_width guest_gdtr_base;
315 natural_width guest_idtr_base;
316 natural_width guest_dr7;
317 natural_width guest_rsp;
318 natural_width guest_rip;
319 natural_width guest_rflags;
320 natural_width guest_pending_dbg_exceptions;
321 natural_width guest_sysenter_esp;
322 natural_width guest_sysenter_eip;
323 natural_width host_cr0;
324 natural_width host_cr3;
325 natural_width host_cr4;
326 natural_width host_fs_base;
327 natural_width host_gs_base;
328 natural_width host_tr_base;
329 natural_width host_gdtr_base;
330 natural_width host_idtr_base;
331 natural_width host_ia32_sysenter_esp;
332 natural_width host_ia32_sysenter_eip;
333 natural_width host_rsp;
334 natural_width host_rip;
335 natural_width paddingl[8]; /* room for future expansion */
336 u32 pin_based_vm_exec_control;
337 u32 cpu_based_vm_exec_control;
338 u32 exception_bitmap;
339 u32 page_fault_error_code_mask;
340 u32 page_fault_error_code_match;
341 u32 cr3_target_count;
342 u32 vm_exit_controls;
343 u32 vm_exit_msr_store_count;
344 u32 vm_exit_msr_load_count;
345 u32 vm_entry_controls;
346 u32 vm_entry_msr_load_count;
347 u32 vm_entry_intr_info_field;
348 u32 vm_entry_exception_error_code;
349 u32 vm_entry_instruction_len;
351 u32 secondary_vm_exec_control;
352 u32 vm_instruction_error;
354 u32 vm_exit_intr_info;
355 u32 vm_exit_intr_error_code;
356 u32 idt_vectoring_info_field;
357 u32 idt_vectoring_error_code;
358 u32 vm_exit_instruction_len;
359 u32 vmx_instruction_info;
366 u32 guest_ldtr_limit;
368 u32 guest_gdtr_limit;
369 u32 guest_idtr_limit;
370 u32 guest_es_ar_bytes;
371 u32 guest_cs_ar_bytes;
372 u32 guest_ss_ar_bytes;
373 u32 guest_ds_ar_bytes;
374 u32 guest_fs_ar_bytes;
375 u32 guest_gs_ar_bytes;
376 u32 guest_ldtr_ar_bytes;
377 u32 guest_tr_ar_bytes;
378 u32 guest_interruptibility_info;
379 u32 guest_activity_state;
380 u32 guest_sysenter_cs;
381 u32 host_ia32_sysenter_cs;
382 u32 vmx_preemption_timer_value;
383 u32 padding32[7]; /* room for future expansion */
384 u16 virtual_processor_id;
386 u16 guest_es_selector;
387 u16 guest_cs_selector;
388 u16 guest_ss_selector;
389 u16 guest_ds_selector;
390 u16 guest_fs_selector;
391 u16 guest_gs_selector;
392 u16 guest_ldtr_selector;
393 u16 guest_tr_selector;
394 u16 guest_intr_status;
396 u16 host_es_selector;
397 u16 host_cs_selector;
398 u16 host_ss_selector;
399 u16 host_ds_selector;
400 u16 host_fs_selector;
401 u16 host_gs_selector;
402 u16 host_tr_selector;
406 * VMCS12_REVISION is an arbitrary id that should be changed if the content or
407 * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
408 * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
410 #define VMCS12_REVISION 0x11e57ed0
413 * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
414 * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
415 * current implementation, 4K are reserved to avoid future complications.
417 #define VMCS12_SIZE 0x1000
420 * The nested_vmx structure is part of vcpu_vmx, and holds information we need
421 * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
424 /* Has the level1 guest done vmxon? */
429 /* The guest-physical address of the current VMCS L1 keeps for L2 */
432 * Cache of the guest's VMCS, existing outside of guest memory.
433 * Loaded from guest memory during VMPTRLD. Flushed to guest
434 * memory during VMCLEAR and VMPTRLD.
436 struct vmcs12 *cached_vmcs12;
438 * Indicates if the shadow vmcs must be updated with the
439 * data hold by vmcs12
441 bool sync_shadow_vmcs;
443 bool change_vmcs01_virtual_x2apic_mode;
444 /* L2 must run next, and mustn't decide to exit to L1. */
445 bool nested_run_pending;
447 struct loaded_vmcs vmcs02;
450 * Guest pages referred to in the vmcs02 with host-physical
451 * pointers, so we must keep them pinned while L2 runs.
453 struct page *apic_access_page;
454 struct page *virtual_apic_page;
455 struct page *pi_desc_page;
456 struct pi_desc *pi_desc;
460 struct hrtimer preemption_timer;
461 bool preemption_timer_expired;
463 /* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
470 * We only store the "true" versions of the VMX capability MSRs. We
471 * generate the "non-true" versions by setting the must-be-1 bits
472 * according to the SDM.
474 u32 nested_vmx_procbased_ctls_low;
475 u32 nested_vmx_procbased_ctls_high;
476 u32 nested_vmx_secondary_ctls_low;
477 u32 nested_vmx_secondary_ctls_high;
478 u32 nested_vmx_pinbased_ctls_low;
479 u32 nested_vmx_pinbased_ctls_high;
480 u32 nested_vmx_exit_ctls_low;
481 u32 nested_vmx_exit_ctls_high;
482 u32 nested_vmx_entry_ctls_low;
483 u32 nested_vmx_entry_ctls_high;
484 u32 nested_vmx_misc_low;
485 u32 nested_vmx_misc_high;
486 u32 nested_vmx_ept_caps;
487 u32 nested_vmx_vpid_caps;
488 u64 nested_vmx_basic;
489 u64 nested_vmx_cr0_fixed0;
490 u64 nested_vmx_cr0_fixed1;
491 u64 nested_vmx_cr4_fixed0;
492 u64 nested_vmx_cr4_fixed1;
493 u64 nested_vmx_vmcs_enum;
494 u64 nested_vmx_vmfunc_controls;
497 #define POSTED_INTR_ON 0
498 #define POSTED_INTR_SN 1
500 /* Posted-Interrupt Descriptor */
502 u32 pir[8]; /* Posted interrupt requested */
505 /* bit 256 - Outstanding Notification */
507 /* bit 257 - Suppress Notification */
509 /* bit 271:258 - Reserved */
511 /* bit 279:272 - Notification Vector */
513 /* bit 287:280 - Reserved */
515 /* bit 319:288 - Notification Destination */
523 static bool pi_test_and_set_on(struct pi_desc *pi_desc)
525 return test_and_set_bit(POSTED_INTR_ON,
526 (unsigned long *)&pi_desc->control);
529 static bool pi_test_and_clear_on(struct pi_desc *pi_desc)
531 return test_and_clear_bit(POSTED_INTR_ON,
532 (unsigned long *)&pi_desc->control);
535 static int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
537 return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
540 static inline void pi_clear_sn(struct pi_desc *pi_desc)
542 return clear_bit(POSTED_INTR_SN,
543 (unsigned long *)&pi_desc->control);
546 static inline void pi_set_sn(struct pi_desc *pi_desc)
548 return set_bit(POSTED_INTR_SN,
549 (unsigned long *)&pi_desc->control);
552 static inline void pi_clear_on(struct pi_desc *pi_desc)
554 clear_bit(POSTED_INTR_ON,
555 (unsigned long *)&pi_desc->control);
558 static inline int pi_test_on(struct pi_desc *pi_desc)
560 return test_bit(POSTED_INTR_ON,
561 (unsigned long *)&pi_desc->control);
564 static inline int pi_test_sn(struct pi_desc *pi_desc)
566 return test_bit(POSTED_INTR_SN,
567 (unsigned long *)&pi_desc->control);
571 struct kvm_vcpu vcpu;
572 unsigned long host_rsp;
576 u32 idt_vectoring_info;
578 struct shared_msr_entry *guest_msrs;
581 unsigned long host_idt_base;
583 u64 msr_host_kernel_gs_base;
584 u64 msr_guest_kernel_gs_base;
587 u64 arch_capabilities;
590 u32 vm_entry_controls_shadow;
591 u32 vm_exit_controls_shadow;
592 u32 secondary_exec_control;
595 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
596 * non-nested (L1) guest, it always points to vmcs01. For a nested
597 * guest (L2), it points to a different VMCS.
599 struct loaded_vmcs vmcs01;
600 struct loaded_vmcs *loaded_vmcs;
601 bool __launched; /* temporary, used in vmx_vcpu_run */
602 struct msr_autoload {
604 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
605 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
609 u16 fs_sel, gs_sel, ldt_sel;
613 int gs_ldt_reload_needed;
614 int fs_reload_needed;
615 u64 msr_host_bndcfgs;
620 struct kvm_segment segs[8];
623 u32 bitmask; /* 4 bits per segment (1 bit per field) */
624 struct kvm_save_segment {
632 bool emulation_required;
636 /* Posted interrupt descriptor */
637 struct pi_desc pi_desc;
639 /* Support for a guest hypervisor (nested VMX) */
640 struct nested_vmx nested;
642 /* Dynamic PLE window. */
644 bool ple_window_dirty;
646 /* Support for PML */
647 #define PML_ENTITY_NUM 512
650 /* apic deadline value in host tsc */
653 u64 current_tsc_ratio;
658 * Only bits masked by msr_ia32_feature_control_valid_bits can be set in
659 * msr_ia32_feature_control. FEATURE_CONTROL_LOCKED is always included
660 * in msr_ia32_feature_control_valid_bits.
662 u64 msr_ia32_feature_control;
663 u64 msr_ia32_feature_control_valid_bits;
666 enum segment_cache_field {
675 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
677 return container_of(vcpu, struct vcpu_vmx, vcpu);
680 static struct pi_desc *vcpu_to_pi_desc(struct kvm_vcpu *vcpu)
682 return &(to_vmx(vcpu)->pi_desc);
685 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
686 #define FIELD(number, name) [number] = VMCS12_OFFSET(name)
687 #define FIELD64(number, name) [number] = VMCS12_OFFSET(name), \
688 [number##_HIGH] = VMCS12_OFFSET(name)+4
691 static unsigned long shadow_read_only_fields[] = {
693 * We do NOT shadow fields that are modified when L0
694 * traps and emulates any vmx instruction (e.g. VMPTRLD,
695 * VMXON...) executed by L1.
696 * For example, VM_INSTRUCTION_ERROR is read
697 * by L1 if a vmx instruction fails (part of the error path).
698 * Note the code assumes this logic. If for some reason
699 * we start shadowing these fields then we need to
700 * force a shadow sync when L0 emulates vmx instructions
701 * (e.g. force a sync if VM_INSTRUCTION_ERROR is modified
702 * by nested_vmx_failValid)
706 VM_EXIT_INSTRUCTION_LEN,
707 IDT_VECTORING_INFO_FIELD,
708 IDT_VECTORING_ERROR_CODE,
709 VM_EXIT_INTR_ERROR_CODE,
711 GUEST_LINEAR_ADDRESS,
712 GUEST_PHYSICAL_ADDRESS
714 static int max_shadow_read_only_fields =
715 ARRAY_SIZE(shadow_read_only_fields);
717 static unsigned long shadow_read_write_fields[] = {
724 GUEST_INTERRUPTIBILITY_INFO,
737 CPU_BASED_VM_EXEC_CONTROL,
738 VM_ENTRY_EXCEPTION_ERROR_CODE,
739 VM_ENTRY_INTR_INFO_FIELD,
740 VM_ENTRY_INSTRUCTION_LEN,
741 VM_ENTRY_EXCEPTION_ERROR_CODE,
747 static int max_shadow_read_write_fields =
748 ARRAY_SIZE(shadow_read_write_fields);
750 static const unsigned short vmcs_field_to_offset_table[] = {
751 FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
752 FIELD(POSTED_INTR_NV, posted_intr_nv),
753 FIELD(GUEST_ES_SELECTOR, guest_es_selector),
754 FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
755 FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
756 FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
757 FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
758 FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
759 FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
760 FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
761 FIELD(GUEST_INTR_STATUS, guest_intr_status),
762 FIELD(GUEST_PML_INDEX, guest_pml_index),
763 FIELD(HOST_ES_SELECTOR, host_es_selector),
764 FIELD(HOST_CS_SELECTOR, host_cs_selector),
765 FIELD(HOST_SS_SELECTOR, host_ss_selector),
766 FIELD(HOST_DS_SELECTOR, host_ds_selector),
767 FIELD(HOST_FS_SELECTOR, host_fs_selector),
768 FIELD(HOST_GS_SELECTOR, host_gs_selector),
769 FIELD(HOST_TR_SELECTOR, host_tr_selector),
770 FIELD64(IO_BITMAP_A, io_bitmap_a),
771 FIELD64(IO_BITMAP_B, io_bitmap_b),
772 FIELD64(MSR_BITMAP, msr_bitmap),
773 FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
774 FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
775 FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
776 FIELD64(TSC_OFFSET, tsc_offset),
777 FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
778 FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
779 FIELD64(POSTED_INTR_DESC_ADDR, posted_intr_desc_addr),
780 FIELD64(VM_FUNCTION_CONTROL, vm_function_control),
781 FIELD64(EPT_POINTER, ept_pointer),
782 FIELD64(EOI_EXIT_BITMAP0, eoi_exit_bitmap0),
783 FIELD64(EOI_EXIT_BITMAP1, eoi_exit_bitmap1),
784 FIELD64(EOI_EXIT_BITMAP2, eoi_exit_bitmap2),
785 FIELD64(EOI_EXIT_BITMAP3, eoi_exit_bitmap3),
786 FIELD64(EPTP_LIST_ADDRESS, eptp_list_address),
787 FIELD64(XSS_EXIT_BITMAP, xss_exit_bitmap),
788 FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
789 FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
790 FIELD64(PML_ADDRESS, pml_address),
791 FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
792 FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
793 FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
794 FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
795 FIELD64(GUEST_PDPTR0, guest_pdptr0),
796 FIELD64(GUEST_PDPTR1, guest_pdptr1),
797 FIELD64(GUEST_PDPTR2, guest_pdptr2),
798 FIELD64(GUEST_PDPTR3, guest_pdptr3),
799 FIELD64(GUEST_BNDCFGS, guest_bndcfgs),
800 FIELD64(HOST_IA32_PAT, host_ia32_pat),
801 FIELD64(HOST_IA32_EFER, host_ia32_efer),
802 FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
803 FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
804 FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
805 FIELD(EXCEPTION_BITMAP, exception_bitmap),
806 FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
807 FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
808 FIELD(CR3_TARGET_COUNT, cr3_target_count),
809 FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
810 FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
811 FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
812 FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
813 FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
814 FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
815 FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
816 FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
817 FIELD(TPR_THRESHOLD, tpr_threshold),
818 FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
819 FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
820 FIELD(VM_EXIT_REASON, vm_exit_reason),
821 FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
822 FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
823 FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
824 FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
825 FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
826 FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
827 FIELD(GUEST_ES_LIMIT, guest_es_limit),
828 FIELD(GUEST_CS_LIMIT, guest_cs_limit),
829 FIELD(GUEST_SS_LIMIT, guest_ss_limit),
830 FIELD(GUEST_DS_LIMIT, guest_ds_limit),
831 FIELD(GUEST_FS_LIMIT, guest_fs_limit),
832 FIELD(GUEST_GS_LIMIT, guest_gs_limit),
833 FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
834 FIELD(GUEST_TR_LIMIT, guest_tr_limit),
835 FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
836 FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
837 FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
838 FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
839 FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
840 FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
841 FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
842 FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
843 FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
844 FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
845 FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
846 FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
847 FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
848 FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
849 FIELD(VMX_PREEMPTION_TIMER_VALUE, vmx_preemption_timer_value),
850 FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
851 FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
852 FIELD(CR0_READ_SHADOW, cr0_read_shadow),
853 FIELD(CR4_READ_SHADOW, cr4_read_shadow),
854 FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
855 FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
856 FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
857 FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
858 FIELD(EXIT_QUALIFICATION, exit_qualification),
859 FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
860 FIELD(GUEST_CR0, guest_cr0),
861 FIELD(GUEST_CR3, guest_cr3),
862 FIELD(GUEST_CR4, guest_cr4),
863 FIELD(GUEST_ES_BASE, guest_es_base),
864 FIELD(GUEST_CS_BASE, guest_cs_base),
865 FIELD(GUEST_SS_BASE, guest_ss_base),
866 FIELD(GUEST_DS_BASE, guest_ds_base),
867 FIELD(GUEST_FS_BASE, guest_fs_base),
868 FIELD(GUEST_GS_BASE, guest_gs_base),
869 FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
870 FIELD(GUEST_TR_BASE, guest_tr_base),
871 FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
872 FIELD(GUEST_IDTR_BASE, guest_idtr_base),
873 FIELD(GUEST_DR7, guest_dr7),
874 FIELD(GUEST_RSP, guest_rsp),
875 FIELD(GUEST_RIP, guest_rip),
876 FIELD(GUEST_RFLAGS, guest_rflags),
877 FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
878 FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
879 FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
880 FIELD(HOST_CR0, host_cr0),
881 FIELD(HOST_CR3, host_cr3),
882 FIELD(HOST_CR4, host_cr4),
883 FIELD(HOST_FS_BASE, host_fs_base),
884 FIELD(HOST_GS_BASE, host_gs_base),
885 FIELD(HOST_TR_BASE, host_tr_base),
886 FIELD(HOST_GDTR_BASE, host_gdtr_base),
887 FIELD(HOST_IDTR_BASE, host_idtr_base),
888 FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
889 FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
890 FIELD(HOST_RSP, host_rsp),
891 FIELD(HOST_RIP, host_rip),
894 static inline short vmcs_field_to_offset(unsigned long field)
896 const size_t size = ARRAY_SIZE(vmcs_field_to_offset_table);
897 unsigned short offset;
899 BUILD_BUG_ON(size > SHRT_MAX);
903 field = array_index_nospec(field, size);
904 offset = vmcs_field_to_offset_table[field];
910 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
912 return to_vmx(vcpu)->nested.cached_vmcs12;
915 static bool nested_ept_ad_enabled(struct kvm_vcpu *vcpu);
916 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu);
917 static u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa);
918 static bool vmx_xsaves_supported(void);
919 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
920 static void vmx_set_segment(struct kvm_vcpu *vcpu,
921 struct kvm_segment *var, int seg);
922 static void vmx_get_segment(struct kvm_vcpu *vcpu,
923 struct kvm_segment *var, int seg);
924 static bool guest_state_valid(struct kvm_vcpu *vcpu);
925 static u32 vmx_segment_access_rights(struct kvm_segment *var);
926 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx);
927 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx);
928 static int alloc_identity_pagetable(struct kvm *kvm);
929 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu);
930 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked);
931 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
933 static void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu);
934 static void __always_inline vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
937 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
938 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
940 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
941 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
943 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
946 * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
947 * can find which vCPU should be waken up.
949 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
950 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
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_vmread_bitmap (vmx_bitmap[VMX_VMREAD_BITMAP])
965 #define vmx_vmwrite_bitmap (vmx_bitmap[VMX_VMWRITE_BITMAP])
967 static bool cpu_has_load_ia32_efer;
968 static bool cpu_has_load_perf_global_ctrl;
970 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
971 static DEFINE_SPINLOCK(vmx_vpid_lock);
973 static struct vmcs_config {
978 u32 pin_based_exec_ctrl;
979 u32 cpu_based_exec_ctrl;
980 u32 cpu_based_2nd_exec_ctrl;
985 static struct vmx_capability {
990 #define VMX_SEGMENT_FIELD(seg) \
991 [VCPU_SREG_##seg] = { \
992 .selector = GUEST_##seg##_SELECTOR, \
993 .base = GUEST_##seg##_BASE, \
994 .limit = GUEST_##seg##_LIMIT, \
995 .ar_bytes = GUEST_##seg##_AR_BYTES, \
998 static const struct kvm_vmx_segment_field {
1003 } kvm_vmx_segment_fields[] = {
1004 VMX_SEGMENT_FIELD(CS),
1005 VMX_SEGMENT_FIELD(DS),
1006 VMX_SEGMENT_FIELD(ES),
1007 VMX_SEGMENT_FIELD(FS),
1008 VMX_SEGMENT_FIELD(GS),
1009 VMX_SEGMENT_FIELD(SS),
1010 VMX_SEGMENT_FIELD(TR),
1011 VMX_SEGMENT_FIELD(LDTR),
1014 static u64 host_efer;
1016 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
1019 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
1020 * away by decrementing the array size.
1022 static const u32 vmx_msr_index[] = {
1023 #ifdef CONFIG_X86_64
1024 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
1026 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
1029 static inline bool is_exception_n(u32 intr_info, u8 vector)
1031 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
1032 INTR_INFO_VALID_MASK)) ==
1033 (INTR_TYPE_HARD_EXCEPTION | vector | INTR_INFO_VALID_MASK);
1036 static inline bool is_debug(u32 intr_info)
1038 return is_exception_n(intr_info, DB_VECTOR);
1041 static inline bool is_breakpoint(u32 intr_info)
1043 return is_exception_n(intr_info, BP_VECTOR);
1046 static inline bool is_page_fault(u32 intr_info)
1048 return is_exception_n(intr_info, PF_VECTOR);
1051 static inline bool is_no_device(u32 intr_info)
1053 return is_exception_n(intr_info, NM_VECTOR);
1056 static inline bool is_invalid_opcode(u32 intr_info)
1058 return is_exception_n(intr_info, UD_VECTOR);
1061 static inline bool is_external_interrupt(u32 intr_info)
1063 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1064 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
1067 static inline bool is_machine_check(u32 intr_info)
1069 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
1070 INTR_INFO_VALID_MASK)) ==
1071 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
1074 /* Undocumented: icebp/int1 */
1075 static inline bool is_icebp(u32 intr_info)
1077 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1078 == (INTR_TYPE_PRIV_SW_EXCEPTION | INTR_INFO_VALID_MASK);
1081 static inline bool cpu_has_vmx_msr_bitmap(void)
1083 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
1086 static inline bool cpu_has_vmx_tpr_shadow(void)
1088 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
1091 static inline bool cpu_need_tpr_shadow(struct kvm_vcpu *vcpu)
1093 return cpu_has_vmx_tpr_shadow() && lapic_in_kernel(vcpu);
1096 static inline bool cpu_has_secondary_exec_ctrls(void)
1098 return vmcs_config.cpu_based_exec_ctrl &
1099 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1102 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
1104 return vmcs_config.cpu_based_2nd_exec_ctrl &
1105 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1108 static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
1110 return vmcs_config.cpu_based_2nd_exec_ctrl &
1111 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
1114 static inline bool cpu_has_vmx_apic_register_virt(void)
1116 return vmcs_config.cpu_based_2nd_exec_ctrl &
1117 SECONDARY_EXEC_APIC_REGISTER_VIRT;
1120 static inline bool cpu_has_vmx_virtual_intr_delivery(void)
1122 return vmcs_config.cpu_based_2nd_exec_ctrl &
1123 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
1127 * Comment's format: document - errata name - stepping - processor name.
1129 * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
1131 static u32 vmx_preemption_cpu_tfms[] = {
1132 /* 323344.pdf - BA86 - D0 - Xeon 7500 Series */
1134 /* 323056.pdf - AAX65 - C2 - Xeon L3406 */
1135 /* 322814.pdf - AAT59 - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
1136 /* 322911.pdf - AAU65 - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
1138 /* 322911.pdf - AAU65 - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
1140 /* 322373.pdf - AAO95 - B1 - Xeon 3400 Series */
1141 /* 322166.pdf - AAN92 - B1 - i7-800 and i5-700 Desktop */
1143 * 320767.pdf - AAP86 - B1 -
1144 * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
1147 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
1149 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
1151 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
1153 /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
1154 /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
1155 /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
1159 static inline bool cpu_has_broken_vmx_preemption_timer(void)
1161 u32 eax = cpuid_eax(0x00000001), i;
1163 /* Clear the reserved bits */
1164 eax &= ~(0x3U << 14 | 0xfU << 28);
1165 for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
1166 if (eax == vmx_preemption_cpu_tfms[i])
1172 static inline bool cpu_has_vmx_preemption_timer(void)
1174 return vmcs_config.pin_based_exec_ctrl &
1175 PIN_BASED_VMX_PREEMPTION_TIMER;
1178 static inline bool cpu_has_vmx_posted_intr(void)
1180 return IS_ENABLED(CONFIG_X86_LOCAL_APIC) &&
1181 vmcs_config.pin_based_exec_ctrl & PIN_BASED_POSTED_INTR;
1184 static inline bool cpu_has_vmx_apicv(void)
1186 return cpu_has_vmx_apic_register_virt() &&
1187 cpu_has_vmx_virtual_intr_delivery() &&
1188 cpu_has_vmx_posted_intr();
1191 static inline bool cpu_has_vmx_flexpriority(void)
1193 return cpu_has_vmx_tpr_shadow() &&
1194 cpu_has_vmx_virtualize_apic_accesses();
1197 static inline bool cpu_has_vmx_ept_execute_only(void)
1199 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
1202 static inline bool cpu_has_vmx_ept_2m_page(void)
1204 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
1207 static inline bool cpu_has_vmx_ept_1g_page(void)
1209 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
1212 static inline bool cpu_has_vmx_ept_4levels(void)
1214 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
1217 static inline bool cpu_has_vmx_ept_mt_wb(void)
1219 return vmx_capability.ept & VMX_EPTP_WB_BIT;
1222 static inline bool cpu_has_vmx_ept_5levels(void)
1224 return vmx_capability.ept & VMX_EPT_PAGE_WALK_5_BIT;
1227 static inline bool cpu_has_vmx_ept_ad_bits(void)
1229 return vmx_capability.ept & VMX_EPT_AD_BIT;
1232 static inline bool cpu_has_vmx_invept_context(void)
1234 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
1237 static inline bool cpu_has_vmx_invept_global(void)
1239 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
1242 static inline bool cpu_has_vmx_invvpid_single(void)
1244 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
1247 static inline bool cpu_has_vmx_invvpid_global(void)
1249 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
1252 static inline bool cpu_has_vmx_invvpid(void)
1254 return vmx_capability.vpid & VMX_VPID_INVVPID_BIT;
1257 static inline bool cpu_has_vmx_ept(void)
1259 return vmcs_config.cpu_based_2nd_exec_ctrl &
1260 SECONDARY_EXEC_ENABLE_EPT;
1263 static inline bool cpu_has_vmx_unrestricted_guest(void)
1265 return vmcs_config.cpu_based_2nd_exec_ctrl &
1266 SECONDARY_EXEC_UNRESTRICTED_GUEST;
1269 static inline bool cpu_has_vmx_ple(void)
1271 return vmcs_config.cpu_based_2nd_exec_ctrl &
1272 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
1275 static inline bool cpu_has_vmx_basic_inout(void)
1277 return (((u64)vmcs_config.basic_cap << 32) & VMX_BASIC_INOUT);
1280 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
1282 return flexpriority_enabled && lapic_in_kernel(vcpu);
1285 static inline bool cpu_has_vmx_vpid(void)
1287 return vmcs_config.cpu_based_2nd_exec_ctrl &
1288 SECONDARY_EXEC_ENABLE_VPID;
1291 static inline bool cpu_has_vmx_rdtscp(void)
1293 return vmcs_config.cpu_based_2nd_exec_ctrl &
1294 SECONDARY_EXEC_RDTSCP;
1297 static inline bool cpu_has_vmx_invpcid(void)
1299 return vmcs_config.cpu_based_2nd_exec_ctrl &
1300 SECONDARY_EXEC_ENABLE_INVPCID;
1303 static inline bool cpu_has_virtual_nmis(void)
1305 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
1308 static inline bool cpu_has_vmx_wbinvd_exit(void)
1310 return vmcs_config.cpu_based_2nd_exec_ctrl &
1311 SECONDARY_EXEC_WBINVD_EXITING;
1314 static inline bool cpu_has_vmx_shadow_vmcs(void)
1317 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
1318 /* check if the cpu supports writing r/o exit information fields */
1319 if (!(vmx_msr & MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS))
1322 return vmcs_config.cpu_based_2nd_exec_ctrl &
1323 SECONDARY_EXEC_SHADOW_VMCS;
1326 static inline bool cpu_has_vmx_pml(void)
1328 return vmcs_config.cpu_based_2nd_exec_ctrl & SECONDARY_EXEC_ENABLE_PML;
1331 static inline bool cpu_has_vmx_tsc_scaling(void)
1333 return vmcs_config.cpu_based_2nd_exec_ctrl &
1334 SECONDARY_EXEC_TSC_SCALING;
1337 static inline bool cpu_has_vmx_vmfunc(void)
1339 return vmcs_config.cpu_based_2nd_exec_ctrl &
1340 SECONDARY_EXEC_ENABLE_VMFUNC;
1343 static inline bool report_flexpriority(void)
1345 return flexpriority_enabled;
1348 static inline unsigned nested_cpu_vmx_misc_cr3_count(struct kvm_vcpu *vcpu)
1350 return vmx_misc_cr3_count(to_vmx(vcpu)->nested.nested_vmx_misc_low);
1353 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
1355 return vmcs12->cpu_based_vm_exec_control & bit;
1358 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
1360 return (vmcs12->cpu_based_vm_exec_control &
1361 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
1362 (vmcs12->secondary_vm_exec_control & bit);
1365 static inline bool nested_cpu_has_preemption_timer(struct vmcs12 *vmcs12)
1367 return vmcs12->pin_based_vm_exec_control &
1368 PIN_BASED_VMX_PREEMPTION_TIMER;
1371 static inline int nested_cpu_has_ept(struct vmcs12 *vmcs12)
1373 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_EPT);
1376 static inline bool nested_cpu_has_xsaves(struct vmcs12 *vmcs12)
1378 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
1381 static inline bool nested_cpu_has_pml(struct vmcs12 *vmcs12)
1383 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_PML);
1386 static inline bool nested_cpu_has_virt_x2apic_mode(struct vmcs12 *vmcs12)
1388 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
1391 static inline bool nested_cpu_has_vpid(struct vmcs12 *vmcs12)
1393 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_VPID);
1396 static inline bool nested_cpu_has_apic_reg_virt(struct vmcs12 *vmcs12)
1398 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_APIC_REGISTER_VIRT);
1401 static inline bool nested_cpu_has_vid(struct vmcs12 *vmcs12)
1403 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
1406 static inline bool nested_cpu_has_posted_intr(struct vmcs12 *vmcs12)
1408 return vmcs12->pin_based_vm_exec_control & PIN_BASED_POSTED_INTR;
1411 static inline bool nested_cpu_has_vmfunc(struct vmcs12 *vmcs12)
1413 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_VMFUNC);
1416 static inline bool nested_cpu_has_eptp_switching(struct vmcs12 *vmcs12)
1418 return nested_cpu_has_vmfunc(vmcs12) &&
1419 (vmcs12->vm_function_control &
1420 VMX_VMFUNC_EPTP_SWITCHING);
1423 static inline bool is_nmi(u32 intr_info)
1425 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1426 == (INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK);
1429 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
1431 unsigned long exit_qualification);
1432 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
1433 struct vmcs12 *vmcs12,
1434 u32 reason, unsigned long qualification);
1436 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
1440 for (i = 0; i < vmx->nmsrs; ++i)
1441 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
1446 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
1452 } operand = { vpid, 0, gva };
1454 asm volatile (__ex(ASM_VMX_INVVPID)
1455 /* CF==1 or ZF==1 --> rc = -1 */
1456 "; ja 1f ; ud2 ; 1:"
1457 : : "a"(&operand), "c"(ext) : "cc", "memory");
1460 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
1464 } operand = {eptp, gpa};
1466 asm volatile (__ex(ASM_VMX_INVEPT)
1467 /* CF==1 or ZF==1 --> rc = -1 */
1468 "; ja 1f ; ud2 ; 1:\n"
1469 : : "a" (&operand), "c" (ext) : "cc", "memory");
1472 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
1476 i = __find_msr_index(vmx, msr);
1478 return &vmx->guest_msrs[i];
1482 static void vmcs_clear(struct vmcs *vmcs)
1484 u64 phys_addr = __pa(vmcs);
1487 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
1488 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1491 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
1495 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
1497 vmcs_clear(loaded_vmcs->vmcs);
1498 if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
1499 vmcs_clear(loaded_vmcs->shadow_vmcs);
1500 loaded_vmcs->cpu = -1;
1501 loaded_vmcs->launched = 0;
1504 static void vmcs_load(struct vmcs *vmcs)
1506 u64 phys_addr = __pa(vmcs);
1509 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
1510 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1513 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
1517 #ifdef CONFIG_KEXEC_CORE
1519 * This bitmap is used to indicate whether the vmclear
1520 * operation is enabled on all cpus. All disabled by
1523 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
1525 static inline void crash_enable_local_vmclear(int cpu)
1527 cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
1530 static inline void crash_disable_local_vmclear(int cpu)
1532 cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
1535 static inline int crash_local_vmclear_enabled(int cpu)
1537 return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
1540 static void crash_vmclear_local_loaded_vmcss(void)
1542 int cpu = raw_smp_processor_id();
1543 struct loaded_vmcs *v;
1545 if (!crash_local_vmclear_enabled(cpu))
1548 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
1549 loaded_vmcss_on_cpu_link)
1550 vmcs_clear(v->vmcs);
1553 static inline void crash_enable_local_vmclear(int cpu) { }
1554 static inline void crash_disable_local_vmclear(int cpu) { }
1555 #endif /* CONFIG_KEXEC_CORE */
1557 static void __loaded_vmcs_clear(void *arg)
1559 struct loaded_vmcs *loaded_vmcs = arg;
1560 int cpu = raw_smp_processor_id();
1562 if (loaded_vmcs->cpu != cpu)
1563 return; /* vcpu migration can race with cpu offline */
1564 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1565 per_cpu(current_vmcs, cpu) = NULL;
1566 crash_disable_local_vmclear(cpu);
1567 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1570 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1571 * is before setting loaded_vmcs->vcpu to -1 which is done in
1572 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1573 * then adds the vmcs into percpu list before it is deleted.
1577 loaded_vmcs_init(loaded_vmcs);
1578 crash_enable_local_vmclear(cpu);
1581 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1583 int cpu = loaded_vmcs->cpu;
1586 smp_call_function_single(cpu,
1587 __loaded_vmcs_clear, loaded_vmcs, 1);
1590 static inline void vpid_sync_vcpu_single(int vpid)
1595 if (cpu_has_vmx_invvpid_single())
1596 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vpid, 0);
1599 static inline void vpid_sync_vcpu_global(void)
1601 if (cpu_has_vmx_invvpid_global())
1602 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1605 static inline void vpid_sync_context(int vpid)
1607 if (cpu_has_vmx_invvpid_single())
1608 vpid_sync_vcpu_single(vpid);
1610 vpid_sync_vcpu_global();
1613 static inline void ept_sync_global(void)
1615 if (cpu_has_vmx_invept_global())
1616 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1619 static inline void ept_sync_context(u64 eptp)
1622 if (cpu_has_vmx_invept_context())
1623 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1629 static __always_inline void vmcs_check16(unsigned long field)
1631 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1632 "16-bit accessor invalid for 64-bit field");
1633 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1634 "16-bit accessor invalid for 64-bit high field");
1635 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1636 "16-bit accessor invalid for 32-bit high field");
1637 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1638 "16-bit accessor invalid for natural width field");
1641 static __always_inline void vmcs_check32(unsigned long field)
1643 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1644 "32-bit accessor invalid for 16-bit field");
1645 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1646 "32-bit accessor invalid for natural width field");
1649 static __always_inline void vmcs_check64(unsigned long field)
1651 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1652 "64-bit accessor invalid for 16-bit field");
1653 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1654 "64-bit accessor invalid for 64-bit high field");
1655 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1656 "64-bit accessor invalid for 32-bit field");
1657 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1658 "64-bit accessor invalid for natural width field");
1661 static __always_inline void vmcs_checkl(unsigned long field)
1663 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1664 "Natural width accessor invalid for 16-bit field");
1665 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1666 "Natural width accessor invalid for 64-bit field");
1667 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1668 "Natural width accessor invalid for 64-bit high field");
1669 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1670 "Natural width accessor invalid for 32-bit field");
1673 static __always_inline unsigned long __vmcs_readl(unsigned long field)
1675 unsigned long value;
1677 asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1678 : "=a"(value) : "d"(field) : "cc");
1682 static __always_inline u16 vmcs_read16(unsigned long field)
1684 vmcs_check16(field);
1685 return __vmcs_readl(field);
1688 static __always_inline u32 vmcs_read32(unsigned long field)
1690 vmcs_check32(field);
1691 return __vmcs_readl(field);
1694 static __always_inline u64 vmcs_read64(unsigned long field)
1696 vmcs_check64(field);
1697 #ifdef CONFIG_X86_64
1698 return __vmcs_readl(field);
1700 return __vmcs_readl(field) | ((u64)__vmcs_readl(field+1) << 32);
1704 static __always_inline unsigned long vmcs_readl(unsigned long field)
1707 return __vmcs_readl(field);
1710 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1712 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1713 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1717 static __always_inline void __vmcs_writel(unsigned long field, unsigned long value)
1721 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1722 : "=q"(error) : "a"(value), "d"(field) : "cc");
1723 if (unlikely(error))
1724 vmwrite_error(field, value);
1727 static __always_inline void vmcs_write16(unsigned long field, u16 value)
1729 vmcs_check16(field);
1730 __vmcs_writel(field, value);
1733 static __always_inline void vmcs_write32(unsigned long field, u32 value)
1735 vmcs_check32(field);
1736 __vmcs_writel(field, value);
1739 static __always_inline void vmcs_write64(unsigned long field, u64 value)
1741 vmcs_check64(field);
1742 __vmcs_writel(field, value);
1743 #ifndef CONFIG_X86_64
1745 __vmcs_writel(field+1, value >> 32);
1749 static __always_inline void vmcs_writel(unsigned long field, unsigned long value)
1752 __vmcs_writel(field, value);
1755 static __always_inline void vmcs_clear_bits(unsigned long field, u32 mask)
1757 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1758 "vmcs_clear_bits does not support 64-bit fields");
1759 __vmcs_writel(field, __vmcs_readl(field) & ~mask);
1762 static __always_inline void vmcs_set_bits(unsigned long field, u32 mask)
1764 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1765 "vmcs_set_bits does not support 64-bit fields");
1766 __vmcs_writel(field, __vmcs_readl(field) | mask);
1769 static inline void vm_entry_controls_reset_shadow(struct vcpu_vmx *vmx)
1771 vmx->vm_entry_controls_shadow = vmcs_read32(VM_ENTRY_CONTROLS);
1774 static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val)
1776 vmcs_write32(VM_ENTRY_CONTROLS, val);
1777 vmx->vm_entry_controls_shadow = val;
1780 static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val)
1782 if (vmx->vm_entry_controls_shadow != val)
1783 vm_entry_controls_init(vmx, val);
1786 static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx)
1788 return vmx->vm_entry_controls_shadow;
1792 static inline void vm_entry_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1794 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) | val);
1797 static inline void vm_entry_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1799 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) & ~val);
1802 static inline void vm_exit_controls_reset_shadow(struct vcpu_vmx *vmx)
1804 vmx->vm_exit_controls_shadow = vmcs_read32(VM_EXIT_CONTROLS);
1807 static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val)
1809 vmcs_write32(VM_EXIT_CONTROLS, val);
1810 vmx->vm_exit_controls_shadow = val;
1813 static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val)
1815 if (vmx->vm_exit_controls_shadow != val)
1816 vm_exit_controls_init(vmx, val);
1819 static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx)
1821 return vmx->vm_exit_controls_shadow;
1825 static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1827 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val);
1830 static inline void vm_exit_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1832 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) & ~val);
1835 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1837 vmx->segment_cache.bitmask = 0;
1840 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1844 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1846 if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1847 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1848 vmx->segment_cache.bitmask = 0;
1850 ret = vmx->segment_cache.bitmask & mask;
1851 vmx->segment_cache.bitmask |= mask;
1855 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1857 u16 *p = &vmx->segment_cache.seg[seg].selector;
1859 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1860 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1864 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1866 ulong *p = &vmx->segment_cache.seg[seg].base;
1868 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1869 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1873 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1875 u32 *p = &vmx->segment_cache.seg[seg].limit;
1877 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1878 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1882 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1884 u32 *p = &vmx->segment_cache.seg[seg].ar;
1886 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1887 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1891 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1895 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1896 (1u << DB_VECTOR) | (1u << AC_VECTOR);
1897 if ((vcpu->guest_debug &
1898 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1899 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1900 eb |= 1u << BP_VECTOR;
1901 if (to_vmx(vcpu)->rmode.vm86_active)
1904 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1906 /* When we are running a nested L2 guest and L1 specified for it a
1907 * certain exception bitmap, we must trap the same exceptions and pass
1908 * them to L1. When running L2, we will only handle the exceptions
1909 * specified above if L1 did not want them.
1911 if (is_guest_mode(vcpu))
1912 eb |= get_vmcs12(vcpu)->exception_bitmap;
1914 vmcs_write32(EXCEPTION_BITMAP, eb);
1918 * Check if MSR is intercepted for currently loaded MSR bitmap.
1920 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr)
1922 unsigned long *msr_bitmap;
1923 int f = sizeof(unsigned long);
1925 if (!cpu_has_vmx_msr_bitmap())
1928 msr_bitmap = to_vmx(vcpu)->loaded_vmcs->msr_bitmap;
1930 if (msr <= 0x1fff) {
1931 return !!test_bit(msr, msr_bitmap + 0x800 / f);
1932 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
1934 return !!test_bit(msr, msr_bitmap + 0xc00 / f);
1941 * Check if MSR is intercepted for L01 MSR bitmap.
1943 static bool msr_write_intercepted_l01(struct kvm_vcpu *vcpu, u32 msr)
1945 unsigned long *msr_bitmap;
1946 int f = sizeof(unsigned long);
1948 if (!cpu_has_vmx_msr_bitmap())
1951 msr_bitmap = to_vmx(vcpu)->vmcs01.msr_bitmap;
1953 if (msr <= 0x1fff) {
1954 return !!test_bit(msr, msr_bitmap + 0x800 / f);
1955 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
1957 return !!test_bit(msr, msr_bitmap + 0xc00 / f);
1963 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1964 unsigned long entry, unsigned long exit)
1966 vm_entry_controls_clearbit(vmx, entry);
1967 vm_exit_controls_clearbit(vmx, exit);
1970 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1973 struct msr_autoload *m = &vmx->msr_autoload;
1977 if (cpu_has_load_ia32_efer) {
1978 clear_atomic_switch_msr_special(vmx,
1979 VM_ENTRY_LOAD_IA32_EFER,
1980 VM_EXIT_LOAD_IA32_EFER);
1984 case MSR_CORE_PERF_GLOBAL_CTRL:
1985 if (cpu_has_load_perf_global_ctrl) {
1986 clear_atomic_switch_msr_special(vmx,
1987 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1988 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1994 for (i = 0; i < m->nr; ++i)
1995 if (m->guest[i].index == msr)
2001 m->guest[i] = m->guest[m->nr];
2002 m->host[i] = m->host[m->nr];
2003 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
2004 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
2007 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
2008 unsigned long entry, unsigned long exit,
2009 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
2010 u64 guest_val, u64 host_val)
2012 vmcs_write64(guest_val_vmcs, guest_val);
2013 vmcs_write64(host_val_vmcs, host_val);
2014 vm_entry_controls_setbit(vmx, entry);
2015 vm_exit_controls_setbit(vmx, exit);
2018 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
2019 u64 guest_val, u64 host_val)
2022 struct msr_autoload *m = &vmx->msr_autoload;
2026 if (cpu_has_load_ia32_efer) {
2027 add_atomic_switch_msr_special(vmx,
2028 VM_ENTRY_LOAD_IA32_EFER,
2029 VM_EXIT_LOAD_IA32_EFER,
2032 guest_val, host_val);
2036 case MSR_CORE_PERF_GLOBAL_CTRL:
2037 if (cpu_has_load_perf_global_ctrl) {
2038 add_atomic_switch_msr_special(vmx,
2039 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
2040 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
2041 GUEST_IA32_PERF_GLOBAL_CTRL,
2042 HOST_IA32_PERF_GLOBAL_CTRL,
2043 guest_val, host_val);
2047 case MSR_IA32_PEBS_ENABLE:
2048 /* PEBS needs a quiescent period after being disabled (to write
2049 * a record). Disabling PEBS through VMX MSR swapping doesn't
2050 * provide that period, so a CPU could write host's record into
2053 wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
2056 for (i = 0; i < m->nr; ++i)
2057 if (m->guest[i].index == msr)
2060 if (i == NR_AUTOLOAD_MSRS) {
2061 printk_once(KERN_WARNING "Not enough msr switch entries. "
2062 "Can't add msr %x\n", msr);
2064 } else if (i == m->nr) {
2066 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
2067 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
2070 m->guest[i].index = msr;
2071 m->guest[i].value = guest_val;
2072 m->host[i].index = msr;
2073 m->host[i].value = host_val;
2076 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
2078 u64 guest_efer = vmx->vcpu.arch.efer;
2079 u64 ignore_bits = 0;
2083 * NX is needed to handle CR0.WP=1, CR4.SMEP=1. Testing
2084 * host CPUID is more efficient than testing guest CPUID
2085 * or CR4. Host SMEP is anyway a requirement for guest SMEP.
2087 if (boot_cpu_has(X86_FEATURE_SMEP))
2088 guest_efer |= EFER_NX;
2089 else if (!(guest_efer & EFER_NX))
2090 ignore_bits |= EFER_NX;
2094 * LMA and LME handled by hardware; SCE meaningless outside long mode.
2096 ignore_bits |= EFER_SCE;
2097 #ifdef CONFIG_X86_64
2098 ignore_bits |= EFER_LMA | EFER_LME;
2099 /* SCE is meaningful only in long mode on Intel */
2100 if (guest_efer & EFER_LMA)
2101 ignore_bits &= ~(u64)EFER_SCE;
2104 clear_atomic_switch_msr(vmx, MSR_EFER);
2107 * On EPT, we can't emulate NX, so we must switch EFER atomically.
2108 * On CPUs that support "load IA32_EFER", always switch EFER
2109 * atomically, since it's faster than switching it manually.
2111 if (cpu_has_load_ia32_efer ||
2112 (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
2113 if (!(guest_efer & EFER_LMA))
2114 guest_efer &= ~EFER_LME;
2115 if (guest_efer != host_efer)
2116 add_atomic_switch_msr(vmx, MSR_EFER,
2117 guest_efer, host_efer);
2120 guest_efer &= ~ignore_bits;
2121 guest_efer |= host_efer & ignore_bits;
2123 vmx->guest_msrs[efer_offset].data = guest_efer;
2124 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
2130 #ifdef CONFIG_X86_32
2132 * On 32-bit kernels, VM exits still load the FS and GS bases from the
2133 * VMCS rather than the segment table. KVM uses this helper to figure
2134 * out the current bases to poke them into the VMCS before entry.
2136 static unsigned long segment_base(u16 selector)
2138 struct desc_struct *table;
2141 if (!(selector & ~SEGMENT_RPL_MASK))
2144 table = get_current_gdt_ro();
2146 if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
2147 u16 ldt_selector = kvm_read_ldt();
2149 if (!(ldt_selector & ~SEGMENT_RPL_MASK))
2152 table = (struct desc_struct *)segment_base(ldt_selector);
2154 v = get_desc_base(&table[selector >> 3]);
2159 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
2161 struct vcpu_vmx *vmx = to_vmx(vcpu);
2164 if (vmx->host_state.loaded)
2167 vmx->host_state.loaded = 1;
2169 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
2170 * allow segment selectors with cpl > 0 or ti == 1.
2172 vmx->host_state.ldt_sel = kvm_read_ldt();
2173 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
2174 savesegment(fs, vmx->host_state.fs_sel);
2175 if (!(vmx->host_state.fs_sel & 7)) {
2176 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
2177 vmx->host_state.fs_reload_needed = 0;
2179 vmcs_write16(HOST_FS_SELECTOR, 0);
2180 vmx->host_state.fs_reload_needed = 1;
2182 savesegment(gs, vmx->host_state.gs_sel);
2183 if (!(vmx->host_state.gs_sel & 7))
2184 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
2186 vmcs_write16(HOST_GS_SELECTOR, 0);
2187 vmx->host_state.gs_ldt_reload_needed = 1;
2190 #ifdef CONFIG_X86_64
2191 savesegment(ds, vmx->host_state.ds_sel);
2192 savesegment(es, vmx->host_state.es_sel);
2195 #ifdef CONFIG_X86_64
2196 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
2197 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
2199 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
2200 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
2203 #ifdef CONFIG_X86_64
2204 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2205 if (is_long_mode(&vmx->vcpu))
2206 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2208 if (boot_cpu_has(X86_FEATURE_MPX))
2209 rdmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
2210 for (i = 0; i < vmx->save_nmsrs; ++i)
2211 kvm_set_shared_msr(vmx->guest_msrs[i].index,
2212 vmx->guest_msrs[i].data,
2213 vmx->guest_msrs[i].mask);
2216 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
2218 if (!vmx->host_state.loaded)
2221 ++vmx->vcpu.stat.host_state_reload;
2222 vmx->host_state.loaded = 0;
2223 #ifdef CONFIG_X86_64
2224 if (is_long_mode(&vmx->vcpu))
2225 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2227 if (vmx->host_state.gs_ldt_reload_needed) {
2228 kvm_load_ldt(vmx->host_state.ldt_sel);
2229 #ifdef CONFIG_X86_64
2230 load_gs_index(vmx->host_state.gs_sel);
2232 loadsegment(gs, vmx->host_state.gs_sel);
2235 if (vmx->host_state.fs_reload_needed)
2236 loadsegment(fs, vmx->host_state.fs_sel);
2237 #ifdef CONFIG_X86_64
2238 if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
2239 loadsegment(ds, vmx->host_state.ds_sel);
2240 loadsegment(es, vmx->host_state.es_sel);
2243 invalidate_tss_limit();
2244 #ifdef CONFIG_X86_64
2245 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2247 if (vmx->host_state.msr_host_bndcfgs)
2248 wrmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
2249 load_fixmap_gdt(raw_smp_processor_id());
2252 static void vmx_load_host_state(struct vcpu_vmx *vmx)
2255 __vmx_load_host_state(vmx);
2259 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
2261 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2262 struct pi_desc old, new;
2266 * In case of hot-plug or hot-unplug, we may have to undo
2267 * vmx_vcpu_pi_put even if there is no assigned device. And we
2268 * always keep PI.NDST up to date for simplicity: it makes the
2269 * code easier, and CPU migration is not a fast path.
2271 if (!pi_test_sn(pi_desc) && vcpu->cpu == cpu)
2275 * First handle the simple case where no cmpxchg is necessary; just
2276 * allow posting non-urgent interrupts.
2278 * If the 'nv' field is POSTED_INTR_WAKEUP_VECTOR, do not change
2279 * PI.NDST: pi_post_block will do it for us and the wakeup_handler
2280 * expects the VCPU to be on the blocked_vcpu_list that matches
2283 if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR ||
2285 pi_clear_sn(pi_desc);
2289 /* The full case. */
2291 old.control = new.control = pi_desc->control;
2293 dest = cpu_physical_id(cpu);
2295 if (x2apic_enabled())
2298 new.ndst = (dest << 8) & 0xFF00;
2301 } while (cmpxchg64(&pi_desc->control, old.control,
2302 new.control) != old.control);
2305 static void decache_tsc_multiplier(struct vcpu_vmx *vmx)
2307 vmx->current_tsc_ratio = vmx->vcpu.arch.tsc_scaling_ratio;
2308 vmcs_write64(TSC_MULTIPLIER, vmx->current_tsc_ratio);
2312 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
2313 * vcpu mutex is already taken.
2315 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2317 struct vcpu_vmx *vmx = to_vmx(vcpu);
2318 bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
2320 if (!already_loaded) {
2321 loaded_vmcs_clear(vmx->loaded_vmcs);
2322 local_irq_disable();
2323 crash_disable_local_vmclear(cpu);
2326 * Read loaded_vmcs->cpu should be before fetching
2327 * loaded_vmcs->loaded_vmcss_on_cpu_link.
2328 * See the comments in __loaded_vmcs_clear().
2332 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
2333 &per_cpu(loaded_vmcss_on_cpu, cpu));
2334 crash_enable_local_vmclear(cpu);
2338 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
2339 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
2340 vmcs_load(vmx->loaded_vmcs->vmcs);
2341 indirect_branch_prediction_barrier();
2344 if (!already_loaded) {
2345 void *gdt = get_current_gdt_ro();
2346 unsigned long sysenter_esp;
2348 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2351 * Linux uses per-cpu TSS and GDT, so set these when switching
2352 * processors. See 22.2.4.
2354 vmcs_writel(HOST_TR_BASE,
2355 (unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss);
2356 vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt); /* 22.2.4 */
2359 * VM exits change the host TR limit to 0x67 after a VM
2360 * exit. This is okay, since 0x67 covers everything except
2361 * the IO bitmap and have have code to handle the IO bitmap
2362 * being lost after a VM exit.
2364 BUILD_BUG_ON(IO_BITMAP_OFFSET - 1 != 0x67);
2366 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
2367 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
2369 vmx->loaded_vmcs->cpu = cpu;
2372 /* Setup TSC multiplier */
2373 if (kvm_has_tsc_control &&
2374 vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
2375 decache_tsc_multiplier(vmx);
2377 vmx_vcpu_pi_load(vcpu, cpu);
2378 vmx->host_pkru = read_pkru();
2381 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
2383 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2385 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
2386 !irq_remapping_cap(IRQ_POSTING_CAP) ||
2387 !kvm_vcpu_apicv_active(vcpu))
2390 /* Set SN when the vCPU is preempted */
2391 if (vcpu->preempted)
2395 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
2397 vmx_vcpu_pi_put(vcpu);
2399 __vmx_load_host_state(to_vmx(vcpu));
2402 static bool emulation_required(struct kvm_vcpu *vcpu)
2404 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
2407 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
2410 * Return the cr0 value that a nested guest would read. This is a combination
2411 * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
2412 * its hypervisor (cr0_read_shadow).
2414 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
2416 return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
2417 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
2419 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
2421 return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
2422 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
2425 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
2427 unsigned long rflags, save_rflags;
2429 if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
2430 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2431 rflags = vmcs_readl(GUEST_RFLAGS);
2432 if (to_vmx(vcpu)->rmode.vm86_active) {
2433 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2434 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
2435 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2437 to_vmx(vcpu)->rflags = rflags;
2439 return to_vmx(vcpu)->rflags;
2442 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
2444 unsigned long old_rflags = vmx_get_rflags(vcpu);
2446 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2447 to_vmx(vcpu)->rflags = rflags;
2448 if (to_vmx(vcpu)->rmode.vm86_active) {
2449 to_vmx(vcpu)->rmode.save_rflags = rflags;
2450 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2452 vmcs_writel(GUEST_RFLAGS, rflags);
2454 if ((old_rflags ^ to_vmx(vcpu)->rflags) & X86_EFLAGS_VM)
2455 to_vmx(vcpu)->emulation_required = emulation_required(vcpu);
2458 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
2460 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2463 if (interruptibility & GUEST_INTR_STATE_STI)
2464 ret |= KVM_X86_SHADOW_INT_STI;
2465 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
2466 ret |= KVM_X86_SHADOW_INT_MOV_SS;
2471 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
2473 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2474 u32 interruptibility = interruptibility_old;
2476 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
2478 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
2479 interruptibility |= GUEST_INTR_STATE_MOV_SS;
2480 else if (mask & KVM_X86_SHADOW_INT_STI)
2481 interruptibility |= GUEST_INTR_STATE_STI;
2483 if ((interruptibility != interruptibility_old))
2484 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
2487 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
2491 rip = kvm_rip_read(vcpu);
2492 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
2493 kvm_rip_write(vcpu, rip);
2495 /* skipping an emulated instruction also counts */
2496 vmx_set_interrupt_shadow(vcpu, 0);
2499 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu,
2500 unsigned long exit_qual)
2502 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2503 unsigned int nr = vcpu->arch.exception.nr;
2504 u32 intr_info = nr | INTR_INFO_VALID_MASK;
2506 if (vcpu->arch.exception.has_error_code) {
2507 vmcs12->vm_exit_intr_error_code = vcpu->arch.exception.error_code;
2508 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2511 if (kvm_exception_is_soft(nr))
2512 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2514 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2516 if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
2517 vmx_get_nmi_mask(vcpu))
2518 intr_info |= INTR_INFO_UNBLOCK_NMI;
2520 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
2524 * KVM wants to inject page-faults which it got to the guest. This function
2525 * checks whether in a nested guest, we need to inject them to L1 or L2.
2527 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned long *exit_qual)
2529 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2530 unsigned int nr = vcpu->arch.exception.nr;
2532 if (nr == PF_VECTOR) {
2533 if (vcpu->arch.exception.nested_apf) {
2534 *exit_qual = vcpu->arch.apf.nested_apf_token;
2538 * FIXME: we must not write CR2 when L1 intercepts an L2 #PF exception.
2539 * The fix is to add the ancillary datum (CR2 or DR6) to structs
2540 * kvm_queued_exception and kvm_vcpu_events, so that CR2 and DR6
2541 * can be written only when inject_pending_event runs. This should be
2542 * conditional on a new capability---if the capability is disabled,
2543 * kvm_multiple_exception would write the ancillary information to
2544 * CR2 or DR6, for backwards ABI-compatibility.
2546 if (nested_vmx_is_page_fault_vmexit(vmcs12,
2547 vcpu->arch.exception.error_code)) {
2548 *exit_qual = vcpu->arch.cr2;
2552 if (vmcs12->exception_bitmap & (1u << nr)) {
2553 if (nr == DB_VECTOR)
2554 *exit_qual = vcpu->arch.dr6;
2564 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
2566 struct vcpu_vmx *vmx = to_vmx(vcpu);
2567 unsigned nr = vcpu->arch.exception.nr;
2568 bool has_error_code = vcpu->arch.exception.has_error_code;
2569 u32 error_code = vcpu->arch.exception.error_code;
2570 u32 intr_info = nr | INTR_INFO_VALID_MASK;
2572 if (has_error_code) {
2573 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
2574 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2577 if (vmx->rmode.vm86_active) {
2579 if (kvm_exception_is_soft(nr))
2580 inc_eip = vcpu->arch.event_exit_inst_len;
2581 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
2582 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2586 WARN_ON_ONCE(vmx->emulation_required);
2588 if (kvm_exception_is_soft(nr)) {
2589 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2590 vmx->vcpu.arch.event_exit_inst_len);
2591 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2593 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2595 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
2598 static bool vmx_rdtscp_supported(void)
2600 return cpu_has_vmx_rdtscp();
2603 static bool vmx_invpcid_supported(void)
2605 return cpu_has_vmx_invpcid() && enable_ept;
2609 * Swap MSR entry in host/guest MSR entry array.
2611 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
2613 struct shared_msr_entry tmp;
2615 tmp = vmx->guest_msrs[to];
2616 vmx->guest_msrs[to] = vmx->guest_msrs[from];
2617 vmx->guest_msrs[from] = tmp;
2621 * Set up the vmcs to automatically save and restore system
2622 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
2623 * mode, as fiddling with msrs is very expensive.
2625 static void setup_msrs(struct vcpu_vmx *vmx)
2627 int save_nmsrs, index;
2630 #ifdef CONFIG_X86_64
2631 if (is_long_mode(&vmx->vcpu)) {
2632 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
2634 move_msr_up(vmx, index, save_nmsrs++);
2635 index = __find_msr_index(vmx, MSR_LSTAR);
2637 move_msr_up(vmx, index, save_nmsrs++);
2638 index = __find_msr_index(vmx, MSR_CSTAR);
2640 move_msr_up(vmx, index, save_nmsrs++);
2641 index = __find_msr_index(vmx, MSR_TSC_AUX);
2642 if (index >= 0 && guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP))
2643 move_msr_up(vmx, index, save_nmsrs++);
2645 * MSR_STAR is only needed on long mode guests, and only
2646 * if efer.sce is enabled.
2648 index = __find_msr_index(vmx, MSR_STAR);
2649 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
2650 move_msr_up(vmx, index, save_nmsrs++);
2653 index = __find_msr_index(vmx, MSR_EFER);
2654 if (index >= 0 && update_transition_efer(vmx, index))
2655 move_msr_up(vmx, index, save_nmsrs++);
2657 vmx->save_nmsrs = save_nmsrs;
2659 if (cpu_has_vmx_msr_bitmap())
2660 vmx_update_msr_bitmap(&vmx->vcpu);
2664 * reads and returns guest's timestamp counter "register"
2665 * guest_tsc = (host_tsc * tsc multiplier) >> 48 + tsc_offset
2666 * -- Intel TSC Scaling for Virtualization White Paper, sec 1.3
2668 static u64 guest_read_tsc(struct kvm_vcpu *vcpu)
2670 u64 host_tsc, tsc_offset;
2673 tsc_offset = vmcs_read64(TSC_OFFSET);
2674 return kvm_scale_tsc(vcpu, host_tsc) + tsc_offset;
2678 * writes 'offset' into guest's timestamp counter offset register
2680 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2682 if (is_guest_mode(vcpu)) {
2684 * We're here if L1 chose not to trap WRMSR to TSC. According
2685 * to the spec, this should set L1's TSC; The offset that L1
2686 * set for L2 remains unchanged, and still needs to be added
2687 * to the newly set TSC to get L2's TSC.
2689 struct vmcs12 *vmcs12;
2690 /* recalculate vmcs02.TSC_OFFSET: */
2691 vmcs12 = get_vmcs12(vcpu);
2692 vmcs_write64(TSC_OFFSET, offset +
2693 (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
2694 vmcs12->tsc_offset : 0));
2696 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2697 vmcs_read64(TSC_OFFSET), offset);
2698 vmcs_write64(TSC_OFFSET, offset);
2703 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
2704 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
2705 * all guests if the "nested" module option is off, and can also be disabled
2706 * for a single guest by disabling its VMX cpuid bit.
2708 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
2710 return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
2714 * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
2715 * returned for the various VMX controls MSRs when nested VMX is enabled.
2716 * The same values should also be used to verify that vmcs12 control fields are
2717 * valid during nested entry from L1 to L2.
2718 * Each of these control msrs has a low and high 32-bit half: A low bit is on
2719 * if the corresponding bit in the (32-bit) control field *must* be on, and a
2720 * bit in the high half is on if the corresponding bit in the control field
2721 * may be on. See also vmx_control_verify().
2723 static void nested_vmx_setup_ctls_msrs(struct vcpu_vmx *vmx)
2726 * Note that as a general rule, the high half of the MSRs (bits in
2727 * the control fields which may be 1) should be initialized by the
2728 * intersection of the underlying hardware's MSR (i.e., features which
2729 * can be supported) and the list of features we want to expose -
2730 * because they are known to be properly supported in our code.
2731 * Also, usually, the low half of the MSRs (bits which must be 1) can
2732 * be set to 0, meaning that L1 may turn off any of these bits. The
2733 * reason is that if one of these bits is necessary, it will appear
2734 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
2735 * fields of vmcs01 and vmcs02, will turn these bits off - and
2736 * nested_vmx_exit_reflected() will not pass related exits to L1.
2737 * These rules have exceptions below.
2740 /* pin-based controls */
2741 rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
2742 vmx->nested.nested_vmx_pinbased_ctls_low,
2743 vmx->nested.nested_vmx_pinbased_ctls_high);
2744 vmx->nested.nested_vmx_pinbased_ctls_low |=
2745 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2746 vmx->nested.nested_vmx_pinbased_ctls_high &=
2747 PIN_BASED_EXT_INTR_MASK |
2748 PIN_BASED_NMI_EXITING |
2749 PIN_BASED_VIRTUAL_NMIS;
2750 vmx->nested.nested_vmx_pinbased_ctls_high |=
2751 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2752 PIN_BASED_VMX_PREEMPTION_TIMER;
2753 if (kvm_vcpu_apicv_active(&vmx->vcpu))
2754 vmx->nested.nested_vmx_pinbased_ctls_high |=
2755 PIN_BASED_POSTED_INTR;
2758 rdmsr(MSR_IA32_VMX_EXIT_CTLS,
2759 vmx->nested.nested_vmx_exit_ctls_low,
2760 vmx->nested.nested_vmx_exit_ctls_high);
2761 vmx->nested.nested_vmx_exit_ctls_low =
2762 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
2764 vmx->nested.nested_vmx_exit_ctls_high &=
2765 #ifdef CONFIG_X86_64
2766 VM_EXIT_HOST_ADDR_SPACE_SIZE |
2768 VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
2769 vmx->nested.nested_vmx_exit_ctls_high |=
2770 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
2771 VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
2772 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
2774 if (kvm_mpx_supported())
2775 vmx->nested.nested_vmx_exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
2777 /* We support free control of debug control saving. */
2778 vmx->nested.nested_vmx_exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
2780 /* entry controls */
2781 rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
2782 vmx->nested.nested_vmx_entry_ctls_low,
2783 vmx->nested.nested_vmx_entry_ctls_high);
2784 vmx->nested.nested_vmx_entry_ctls_low =
2785 VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
2786 vmx->nested.nested_vmx_entry_ctls_high &=
2787 #ifdef CONFIG_X86_64
2788 VM_ENTRY_IA32E_MODE |
2790 VM_ENTRY_LOAD_IA32_PAT;
2791 vmx->nested.nested_vmx_entry_ctls_high |=
2792 (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
2793 if (kvm_mpx_supported())
2794 vmx->nested.nested_vmx_entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
2796 /* We support free control of debug control loading. */
2797 vmx->nested.nested_vmx_entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
2799 /* cpu-based controls */
2800 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
2801 vmx->nested.nested_vmx_procbased_ctls_low,
2802 vmx->nested.nested_vmx_procbased_ctls_high);
2803 vmx->nested.nested_vmx_procbased_ctls_low =
2804 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2805 vmx->nested.nested_vmx_procbased_ctls_high &=
2806 CPU_BASED_VIRTUAL_INTR_PENDING |
2807 CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
2808 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
2809 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
2810 CPU_BASED_CR3_STORE_EXITING |
2811 #ifdef CONFIG_X86_64
2812 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
2814 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
2815 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
2816 CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
2817 CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
2818 CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2820 * We can allow some features even when not supported by the
2821 * hardware. For example, L1 can specify an MSR bitmap - and we
2822 * can use it to avoid exits to L1 - even when L0 runs L2
2823 * without MSR bitmaps.
2825 vmx->nested.nested_vmx_procbased_ctls_high |=
2826 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2827 CPU_BASED_USE_MSR_BITMAPS;
2829 /* We support free control of CR3 access interception. */
2830 vmx->nested.nested_vmx_procbased_ctls_low &=
2831 ~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
2834 * secondary cpu-based controls. Do not include those that
2835 * depend on CPUID bits, they are added later by vmx_cpuid_update.
2837 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2838 vmx->nested.nested_vmx_secondary_ctls_low,
2839 vmx->nested.nested_vmx_secondary_ctls_high);
2840 vmx->nested.nested_vmx_secondary_ctls_low = 0;
2841 vmx->nested.nested_vmx_secondary_ctls_high &=
2842 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2843 SECONDARY_EXEC_DESC |
2844 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2845 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2846 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2847 SECONDARY_EXEC_WBINVD_EXITING;
2850 /* nested EPT: emulate EPT also to L1 */
2851 vmx->nested.nested_vmx_secondary_ctls_high |=
2852 SECONDARY_EXEC_ENABLE_EPT;
2853 vmx->nested.nested_vmx_ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
2854 VMX_EPTP_WB_BIT | VMX_EPT_INVEPT_BIT;
2855 if (cpu_has_vmx_ept_execute_only())
2856 vmx->nested.nested_vmx_ept_caps |=
2857 VMX_EPT_EXECUTE_ONLY_BIT;
2858 vmx->nested.nested_vmx_ept_caps &= vmx_capability.ept;
2859 vmx->nested.nested_vmx_ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
2860 VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
2861 VMX_EPT_1GB_PAGE_BIT;
2862 if (enable_ept_ad_bits) {
2863 vmx->nested.nested_vmx_secondary_ctls_high |=
2864 SECONDARY_EXEC_ENABLE_PML;
2865 vmx->nested.nested_vmx_ept_caps |= VMX_EPT_AD_BIT;
2868 vmx->nested.nested_vmx_ept_caps = 0;
2870 if (cpu_has_vmx_vmfunc()) {
2871 vmx->nested.nested_vmx_secondary_ctls_high |=
2872 SECONDARY_EXEC_ENABLE_VMFUNC;
2874 * Advertise EPTP switching unconditionally
2875 * since we emulate it
2878 vmx->nested.nested_vmx_vmfunc_controls =
2879 VMX_VMFUNC_EPTP_SWITCHING;
2883 * Old versions of KVM use the single-context version without
2884 * checking for support, so declare that it is supported even
2885 * though it is treated as global context. The alternative is
2886 * not failing the single-context invvpid, and it is worse.
2889 vmx->nested.nested_vmx_secondary_ctls_high |=
2890 SECONDARY_EXEC_ENABLE_VPID;
2891 vmx->nested.nested_vmx_vpid_caps = VMX_VPID_INVVPID_BIT |
2892 VMX_VPID_EXTENT_SUPPORTED_MASK;
2894 vmx->nested.nested_vmx_vpid_caps = 0;
2896 if (enable_unrestricted_guest)
2897 vmx->nested.nested_vmx_secondary_ctls_high |=
2898 SECONDARY_EXEC_UNRESTRICTED_GUEST;
2900 /* miscellaneous data */
2901 rdmsr(MSR_IA32_VMX_MISC,
2902 vmx->nested.nested_vmx_misc_low,
2903 vmx->nested.nested_vmx_misc_high);
2904 vmx->nested.nested_vmx_misc_low &= VMX_MISC_SAVE_EFER_LMA;
2905 vmx->nested.nested_vmx_misc_low |=
2906 VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
2907 VMX_MISC_ACTIVITY_HLT;
2908 vmx->nested.nested_vmx_misc_high = 0;
2911 * This MSR reports some information about VMX support. We
2912 * should return information about the VMX we emulate for the
2913 * guest, and the VMCS structure we give it - not about the
2914 * VMX support of the underlying hardware.
2916 vmx->nested.nested_vmx_basic =
2918 VMX_BASIC_TRUE_CTLS |
2919 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2920 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2922 if (cpu_has_vmx_basic_inout())
2923 vmx->nested.nested_vmx_basic |= VMX_BASIC_INOUT;
2926 * These MSRs specify bits which the guest must keep fixed on
2927 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2928 * We picked the standard core2 setting.
2930 #define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2931 #define VMXON_CR4_ALWAYSON X86_CR4_VMXE
2932 vmx->nested.nested_vmx_cr0_fixed0 = VMXON_CR0_ALWAYSON;
2933 vmx->nested.nested_vmx_cr4_fixed0 = VMXON_CR4_ALWAYSON;
2935 /* These MSRs specify bits which the guest must keep fixed off. */
2936 rdmsrl(MSR_IA32_VMX_CR0_FIXED1, vmx->nested.nested_vmx_cr0_fixed1);
2937 rdmsrl(MSR_IA32_VMX_CR4_FIXED1, vmx->nested.nested_vmx_cr4_fixed1);
2939 /* highest index: VMX_PREEMPTION_TIMER_VALUE */
2940 vmx->nested.nested_vmx_vmcs_enum = 0x2e;
2944 * if fixed0[i] == 1: val[i] must be 1
2945 * if fixed1[i] == 0: val[i] must be 0
2947 static inline bool fixed_bits_valid(u64 val, u64 fixed0, u64 fixed1)
2949 return ((val & fixed1) | fixed0) == val;
2952 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2954 return fixed_bits_valid(control, low, high);
2957 static inline u64 vmx_control_msr(u32 low, u32 high)
2959 return low | ((u64)high << 32);
2962 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
2967 return (superset | subset) == superset;
2970 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
2972 const u64 feature_and_reserved =
2973 /* feature (except bit 48; see below) */
2974 BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
2976 BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
2977 u64 vmx_basic = vmx->nested.nested_vmx_basic;
2979 if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
2983 * KVM does not emulate a version of VMX that constrains physical
2984 * addresses of VMX structures (e.g. VMCS) to 32-bits.
2986 if (data & BIT_ULL(48))
2989 if (vmx_basic_vmcs_revision_id(vmx_basic) !=
2990 vmx_basic_vmcs_revision_id(data))
2993 if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
2996 vmx->nested.nested_vmx_basic = data;
3001 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
3006 switch (msr_index) {
3007 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
3008 lowp = &vmx->nested.nested_vmx_pinbased_ctls_low;
3009 highp = &vmx->nested.nested_vmx_pinbased_ctls_high;
3011 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
3012 lowp = &vmx->nested.nested_vmx_procbased_ctls_low;
3013 highp = &vmx->nested.nested_vmx_procbased_ctls_high;
3015 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
3016 lowp = &vmx->nested.nested_vmx_exit_ctls_low;
3017 highp = &vmx->nested.nested_vmx_exit_ctls_high;
3019 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
3020 lowp = &vmx->nested.nested_vmx_entry_ctls_low;
3021 highp = &vmx->nested.nested_vmx_entry_ctls_high;
3023 case MSR_IA32_VMX_PROCBASED_CTLS2:
3024 lowp = &vmx->nested.nested_vmx_secondary_ctls_low;
3025 highp = &vmx->nested.nested_vmx_secondary_ctls_high;
3031 supported = vmx_control_msr(*lowp, *highp);
3033 /* Check must-be-1 bits are still 1. */
3034 if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
3037 /* Check must-be-0 bits are still 0. */
3038 if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
3042 *highp = data >> 32;
3046 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
3048 const u64 feature_and_reserved_bits =
3050 BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
3051 BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
3053 GENMASK_ULL(13, 9) | BIT_ULL(31);
3056 vmx_misc = vmx_control_msr(vmx->nested.nested_vmx_misc_low,
3057 vmx->nested.nested_vmx_misc_high);
3059 if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
3062 if ((vmx->nested.nested_vmx_pinbased_ctls_high &
3063 PIN_BASED_VMX_PREEMPTION_TIMER) &&
3064 vmx_misc_preemption_timer_rate(data) !=
3065 vmx_misc_preemption_timer_rate(vmx_misc))
3068 if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
3071 if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
3074 if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
3077 vmx->nested.nested_vmx_misc_low = data;
3078 vmx->nested.nested_vmx_misc_high = data >> 32;
3082 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
3084 u64 vmx_ept_vpid_cap;
3086 vmx_ept_vpid_cap = vmx_control_msr(vmx->nested.nested_vmx_ept_caps,
3087 vmx->nested.nested_vmx_vpid_caps);
3089 /* Every bit is either reserved or a feature bit. */
3090 if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
3093 vmx->nested.nested_vmx_ept_caps = data;
3094 vmx->nested.nested_vmx_vpid_caps = data >> 32;
3098 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
3102 switch (msr_index) {
3103 case MSR_IA32_VMX_CR0_FIXED0:
3104 msr = &vmx->nested.nested_vmx_cr0_fixed0;
3106 case MSR_IA32_VMX_CR4_FIXED0:
3107 msr = &vmx->nested.nested_vmx_cr4_fixed0;
3114 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
3115 * must be 1 in the restored value.
3117 if (!is_bitwise_subset(data, *msr, -1ULL))
3125 * Called when userspace is restoring VMX MSRs.
3127 * Returns 0 on success, non-0 otherwise.
3129 static int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
3131 struct vcpu_vmx *vmx = to_vmx(vcpu);
3133 switch (msr_index) {
3134 case MSR_IA32_VMX_BASIC:
3135 return vmx_restore_vmx_basic(vmx, data);
3136 case MSR_IA32_VMX_PINBASED_CTLS:
3137 case MSR_IA32_VMX_PROCBASED_CTLS:
3138 case MSR_IA32_VMX_EXIT_CTLS:
3139 case MSR_IA32_VMX_ENTRY_CTLS:
3141 * The "non-true" VMX capability MSRs are generated from the
3142 * "true" MSRs, so we do not support restoring them directly.
3144 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
3145 * should restore the "true" MSRs with the must-be-1 bits
3146 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
3147 * DEFAULT SETTINGS".
3150 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
3151 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
3152 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
3153 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
3154 case MSR_IA32_VMX_PROCBASED_CTLS2:
3155 return vmx_restore_control_msr(vmx, msr_index, data);
3156 case MSR_IA32_VMX_MISC:
3157 return vmx_restore_vmx_misc(vmx, data);
3158 case MSR_IA32_VMX_CR0_FIXED0:
3159 case MSR_IA32_VMX_CR4_FIXED0:
3160 return vmx_restore_fixed0_msr(vmx, msr_index, data);
3161 case MSR_IA32_VMX_CR0_FIXED1:
3162 case MSR_IA32_VMX_CR4_FIXED1:
3164 * These MSRs are generated based on the vCPU's CPUID, so we
3165 * do not support restoring them directly.
3168 case MSR_IA32_VMX_EPT_VPID_CAP:
3169 return vmx_restore_vmx_ept_vpid_cap(vmx, data);
3170 case MSR_IA32_VMX_VMCS_ENUM:
3171 vmx->nested.nested_vmx_vmcs_enum = data;
3175 * The rest of the VMX capability MSRs do not support restore.
3181 /* Returns 0 on success, non-0 otherwise. */
3182 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
3184 struct vcpu_vmx *vmx = to_vmx(vcpu);
3186 switch (msr_index) {
3187 case MSR_IA32_VMX_BASIC:
3188 *pdata = vmx->nested.nested_vmx_basic;
3190 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
3191 case MSR_IA32_VMX_PINBASED_CTLS:
3192 *pdata = vmx_control_msr(
3193 vmx->nested.nested_vmx_pinbased_ctls_low,
3194 vmx->nested.nested_vmx_pinbased_ctls_high);
3195 if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
3196 *pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
3198 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
3199 case MSR_IA32_VMX_PROCBASED_CTLS:
3200 *pdata = vmx_control_msr(
3201 vmx->nested.nested_vmx_procbased_ctls_low,
3202 vmx->nested.nested_vmx_procbased_ctls_high);
3203 if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
3204 *pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
3206 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
3207 case MSR_IA32_VMX_EXIT_CTLS:
3208 *pdata = vmx_control_msr(
3209 vmx->nested.nested_vmx_exit_ctls_low,
3210 vmx->nested.nested_vmx_exit_ctls_high);
3211 if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
3212 *pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
3214 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
3215 case MSR_IA32_VMX_ENTRY_CTLS:
3216 *pdata = vmx_control_msr(
3217 vmx->nested.nested_vmx_entry_ctls_low,
3218 vmx->nested.nested_vmx_entry_ctls_high);
3219 if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
3220 *pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
3222 case MSR_IA32_VMX_MISC:
3223 *pdata = vmx_control_msr(
3224 vmx->nested.nested_vmx_misc_low,
3225 vmx->nested.nested_vmx_misc_high);
3227 case MSR_IA32_VMX_CR0_FIXED0:
3228 *pdata = vmx->nested.nested_vmx_cr0_fixed0;
3230 case MSR_IA32_VMX_CR0_FIXED1:
3231 *pdata = vmx->nested.nested_vmx_cr0_fixed1;
3233 case MSR_IA32_VMX_CR4_FIXED0:
3234 *pdata = vmx->nested.nested_vmx_cr4_fixed0;
3236 case MSR_IA32_VMX_CR4_FIXED1:
3237 *pdata = vmx->nested.nested_vmx_cr4_fixed1;
3239 case MSR_IA32_VMX_VMCS_ENUM:
3240 *pdata = vmx->nested.nested_vmx_vmcs_enum;
3242 case MSR_IA32_VMX_PROCBASED_CTLS2:
3243 *pdata = vmx_control_msr(
3244 vmx->nested.nested_vmx_secondary_ctls_low,
3245 vmx->nested.nested_vmx_secondary_ctls_high);
3247 case MSR_IA32_VMX_EPT_VPID_CAP:
3248 *pdata = vmx->nested.nested_vmx_ept_caps |
3249 ((u64)vmx->nested.nested_vmx_vpid_caps << 32);
3251 case MSR_IA32_VMX_VMFUNC:
3252 *pdata = vmx->nested.nested_vmx_vmfunc_controls;
3261 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
3264 uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
3266 return !(val & ~valid_bits);
3270 * Reads an msr value (of 'msr_index') into 'pdata'.
3271 * Returns 0 on success, non-0 otherwise.
3272 * Assumes vcpu_load() was already called.
3274 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3276 struct shared_msr_entry *msr;
3278 switch (msr_info->index) {
3279 #ifdef CONFIG_X86_64
3281 msr_info->data = vmcs_readl(GUEST_FS_BASE);
3284 msr_info->data = vmcs_readl(GUEST_GS_BASE);
3286 case MSR_KERNEL_GS_BASE:
3287 vmx_load_host_state(to_vmx(vcpu));
3288 msr_info->data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
3292 return kvm_get_msr_common(vcpu, msr_info);
3294 msr_info->data = guest_read_tsc(vcpu);
3296 case MSR_IA32_SPEC_CTRL:
3297 if (!msr_info->host_initiated &&
3298 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
3301 msr_info->data = to_vmx(vcpu)->spec_ctrl;
3303 case MSR_IA32_ARCH_CAPABILITIES:
3304 if (!msr_info->host_initiated &&
3305 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3307 msr_info->data = to_vmx(vcpu)->arch_capabilities;
3309 case MSR_IA32_SYSENTER_CS:
3310 msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
3312 case MSR_IA32_SYSENTER_EIP:
3313 msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
3315 case MSR_IA32_SYSENTER_ESP:
3316 msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
3318 case MSR_IA32_BNDCFGS:
3319 if (!kvm_mpx_supported() ||
3320 (!msr_info->host_initiated &&
3321 !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
3323 msr_info->data = vmcs_read64(GUEST_BNDCFGS);
3325 case MSR_IA32_MCG_EXT_CTL:
3326 if (!msr_info->host_initiated &&
3327 !(to_vmx(vcpu)->msr_ia32_feature_control &
3328 FEATURE_CONTROL_LMCE))
3330 msr_info->data = vcpu->arch.mcg_ext_ctl;
3332 case MSR_IA32_FEATURE_CONTROL:
3333 msr_info->data = to_vmx(vcpu)->msr_ia32_feature_control;
3335 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3336 if (!nested_vmx_allowed(vcpu))
3338 return vmx_get_vmx_msr(vcpu, msr_info->index, &msr_info->data);
3340 if (!vmx_xsaves_supported())
3342 msr_info->data = vcpu->arch.ia32_xss;
3345 if (!msr_info->host_initiated &&
3346 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
3348 /* Otherwise falls through */
3350 msr = find_msr_entry(to_vmx(vcpu), msr_info->index);
3352 msr_info->data = msr->data;
3355 return kvm_get_msr_common(vcpu, msr_info);
3361 static void vmx_leave_nested(struct kvm_vcpu *vcpu);
3364 * Writes msr value into into the appropriate "register".
3365 * Returns 0 on success, non-0 otherwise.
3366 * Assumes vcpu_load() was already called.
3368 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3370 struct vcpu_vmx *vmx = to_vmx(vcpu);
3371 struct shared_msr_entry *msr;
3373 u32 msr_index = msr_info->index;
3374 u64 data = msr_info->data;
3376 switch (msr_index) {
3378 ret = kvm_set_msr_common(vcpu, msr_info);
3380 #ifdef CONFIG_X86_64
3382 vmx_segment_cache_clear(vmx);
3383 vmcs_writel(GUEST_FS_BASE, data);
3386 vmx_segment_cache_clear(vmx);
3387 vmcs_writel(GUEST_GS_BASE, data);
3389 case MSR_KERNEL_GS_BASE:
3390 vmx_load_host_state(vmx);
3391 vmx->msr_guest_kernel_gs_base = data;
3394 case MSR_IA32_SYSENTER_CS:
3395 vmcs_write32(GUEST_SYSENTER_CS, data);
3397 case MSR_IA32_SYSENTER_EIP:
3398 vmcs_writel(GUEST_SYSENTER_EIP, data);
3400 case MSR_IA32_SYSENTER_ESP:
3401 vmcs_writel(GUEST_SYSENTER_ESP, data);
3403 case MSR_IA32_BNDCFGS:
3404 if (!kvm_mpx_supported() ||
3405 (!msr_info->host_initiated &&
3406 !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
3408 if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
3409 (data & MSR_IA32_BNDCFGS_RSVD))
3411 vmcs_write64(GUEST_BNDCFGS, data);
3414 kvm_write_tsc(vcpu, msr_info);
3416 case MSR_IA32_SPEC_CTRL:
3417 if (!msr_info->host_initiated &&
3418 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
3421 /* The STIBP bit doesn't fault even if it's not advertised */
3422 if (data & ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_SSBD))
3425 vmx->spec_ctrl = data;
3432 * When it's written (to non-zero) for the first time, pass
3436 * The handling of the MSR bitmap for L2 guests is done in
3437 * nested_vmx_merge_msr_bitmap. We should not touch the
3438 * vmcs02.msr_bitmap here since it gets completely overwritten
3439 * in the merging. We update the vmcs01 here for L1 as well
3440 * since it will end up touching the MSR anyway now.
3442 vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap,
3446 case MSR_IA32_PRED_CMD:
3447 if (!msr_info->host_initiated &&
3448 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
3451 if (data & ~PRED_CMD_IBPB)
3457 wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
3461 * When it's written (to non-zero) for the first time, pass
3465 * The handling of the MSR bitmap for L2 guests is done in
3466 * nested_vmx_merge_msr_bitmap. We should not touch the
3467 * vmcs02.msr_bitmap here since it gets completely overwritten
3470 vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap, MSR_IA32_PRED_CMD,
3473 case MSR_IA32_ARCH_CAPABILITIES:
3474 if (!msr_info->host_initiated)
3476 vmx->arch_capabilities = data;
3478 case MSR_IA32_CR_PAT:
3479 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
3480 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
3482 vmcs_write64(GUEST_IA32_PAT, data);
3483 vcpu->arch.pat = data;
3486 ret = kvm_set_msr_common(vcpu, msr_info);
3488 case MSR_IA32_TSC_ADJUST:
3489 ret = kvm_set_msr_common(vcpu, msr_info);
3491 case MSR_IA32_MCG_EXT_CTL:
3492 if ((!msr_info->host_initiated &&
3493 !(to_vmx(vcpu)->msr_ia32_feature_control &
3494 FEATURE_CONTROL_LMCE)) ||
3495 (data & ~MCG_EXT_CTL_LMCE_EN))
3497 vcpu->arch.mcg_ext_ctl = data;
3499 case MSR_IA32_FEATURE_CONTROL:
3500 if (!vmx_feature_control_msr_valid(vcpu, data) ||
3501 (to_vmx(vcpu)->msr_ia32_feature_control &
3502 FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
3504 vmx->msr_ia32_feature_control = data;
3505 if (msr_info->host_initiated && data == 0)
3506 vmx_leave_nested(vcpu);
3508 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3509 if (!msr_info->host_initiated)
3510 return 1; /* they are read-only */
3511 if (!nested_vmx_allowed(vcpu))
3513 return vmx_set_vmx_msr(vcpu, msr_index, data);
3515 if (!vmx_xsaves_supported())
3518 * The only supported bit as of Skylake is bit 8, but
3519 * it is not supported on KVM.
3523 vcpu->arch.ia32_xss = data;
3524 if (vcpu->arch.ia32_xss != host_xss)
3525 add_atomic_switch_msr(vmx, MSR_IA32_XSS,
3526 vcpu->arch.ia32_xss, host_xss);
3528 clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
3531 if (!msr_info->host_initiated &&
3532 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
3534 /* Check reserved bit, higher 32 bits should be zero */
3535 if ((data >> 32) != 0)
3537 /* Otherwise falls through */
3539 msr = find_msr_entry(vmx, msr_index);
3541 u64 old_msr_data = msr->data;
3543 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
3545 ret = kvm_set_shared_msr(msr->index, msr->data,
3549 msr->data = old_msr_data;
3553 ret = kvm_set_msr_common(vcpu, msr_info);
3559 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
3561 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
3564 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
3567 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
3569 case VCPU_EXREG_PDPTR:
3571 ept_save_pdptrs(vcpu);
3578 static __init int cpu_has_kvm_support(void)
3580 return cpu_has_vmx();
3583 static __init int vmx_disabled_by_bios(void)
3587 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
3588 if (msr & FEATURE_CONTROL_LOCKED) {
3589 /* launched w/ TXT and VMX disabled */
3590 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3593 /* launched w/o TXT and VMX only enabled w/ TXT */
3594 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3595 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3596 && !tboot_enabled()) {
3597 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
3598 "activate TXT before enabling KVM\n");
3601 /* launched w/o TXT and VMX disabled */
3602 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3603 && !tboot_enabled())
3610 static void kvm_cpu_vmxon(u64 addr)
3612 cr4_set_bits(X86_CR4_VMXE);
3613 intel_pt_handle_vmx(1);
3615 asm volatile (ASM_VMX_VMXON_RAX
3616 : : "a"(&addr), "m"(addr)
3620 static int hardware_enable(void)
3622 int cpu = raw_smp_processor_id();
3623 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
3626 if (cr4_read_shadow() & X86_CR4_VMXE)
3629 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
3630 INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
3631 spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
3634 * Now we can enable the vmclear operation in kdump
3635 * since the loaded_vmcss_on_cpu list on this cpu
3636 * has been initialized.
3638 * Though the cpu is not in VMX operation now, there
3639 * is no problem to enable the vmclear operation
3640 * for the loaded_vmcss_on_cpu list is empty!
3642 crash_enable_local_vmclear(cpu);
3644 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
3646 test_bits = FEATURE_CONTROL_LOCKED;
3647 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
3648 if (tboot_enabled())
3649 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
3651 if ((old & test_bits) != test_bits) {
3652 /* enable and lock */
3653 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
3655 kvm_cpu_vmxon(phys_addr);
3661 static void vmclear_local_loaded_vmcss(void)
3663 int cpu = raw_smp_processor_id();
3664 struct loaded_vmcs *v, *n;
3666 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
3667 loaded_vmcss_on_cpu_link)
3668 __loaded_vmcs_clear(v);
3672 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
3675 static void kvm_cpu_vmxoff(void)
3677 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
3679 intel_pt_handle_vmx(0);
3680 cr4_clear_bits(X86_CR4_VMXE);
3683 static void hardware_disable(void)
3685 vmclear_local_loaded_vmcss();
3689 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
3690 u32 msr, u32 *result)
3692 u32 vmx_msr_low, vmx_msr_high;
3693 u32 ctl = ctl_min | ctl_opt;
3695 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3697 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
3698 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
3700 /* Ensure minimum (required) set of control bits are supported. */
3708 static __init bool allow_1_setting(u32 msr, u32 ctl)
3710 u32 vmx_msr_low, vmx_msr_high;
3712 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3713 return vmx_msr_high & ctl;
3716 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
3718 u32 vmx_msr_low, vmx_msr_high;
3719 u32 min, opt, min2, opt2;
3720 u32 _pin_based_exec_control = 0;
3721 u32 _cpu_based_exec_control = 0;
3722 u32 _cpu_based_2nd_exec_control = 0;
3723 u32 _vmexit_control = 0;
3724 u32 _vmentry_control = 0;
3726 min = CPU_BASED_HLT_EXITING |
3727 #ifdef CONFIG_X86_64
3728 CPU_BASED_CR8_LOAD_EXITING |
3729 CPU_BASED_CR8_STORE_EXITING |
3731 CPU_BASED_CR3_LOAD_EXITING |
3732 CPU_BASED_CR3_STORE_EXITING |
3733 CPU_BASED_USE_IO_BITMAPS |
3734 CPU_BASED_MOV_DR_EXITING |
3735 CPU_BASED_USE_TSC_OFFSETING |
3736 CPU_BASED_INVLPG_EXITING |
3737 CPU_BASED_RDPMC_EXITING;
3739 if (!kvm_mwait_in_guest())
3740 min |= CPU_BASED_MWAIT_EXITING |
3741 CPU_BASED_MONITOR_EXITING;
3743 opt = CPU_BASED_TPR_SHADOW |
3744 CPU_BASED_USE_MSR_BITMAPS |
3745 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
3746 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
3747 &_cpu_based_exec_control) < 0)
3749 #ifdef CONFIG_X86_64
3750 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3751 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
3752 ~CPU_BASED_CR8_STORE_EXITING;
3754 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
3756 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
3757 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3758 SECONDARY_EXEC_WBINVD_EXITING |
3759 SECONDARY_EXEC_ENABLE_VPID |
3760 SECONDARY_EXEC_ENABLE_EPT |
3761 SECONDARY_EXEC_UNRESTRICTED_GUEST |
3762 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
3763 SECONDARY_EXEC_RDTSCP |
3764 SECONDARY_EXEC_ENABLE_INVPCID |
3765 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3766 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
3767 SECONDARY_EXEC_SHADOW_VMCS |
3768 SECONDARY_EXEC_XSAVES |
3769 SECONDARY_EXEC_RDSEED |
3770 SECONDARY_EXEC_RDRAND |
3771 SECONDARY_EXEC_ENABLE_PML |
3772 SECONDARY_EXEC_TSC_SCALING |
3773 SECONDARY_EXEC_ENABLE_VMFUNC;
3774 if (adjust_vmx_controls(min2, opt2,
3775 MSR_IA32_VMX_PROCBASED_CTLS2,
3776 &_cpu_based_2nd_exec_control) < 0)
3779 #ifndef CONFIG_X86_64
3780 if (!(_cpu_based_2nd_exec_control &
3781 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
3782 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
3785 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3786 _cpu_based_2nd_exec_control &= ~(
3787 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3788 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3789 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3791 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
3792 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
3794 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
3795 CPU_BASED_CR3_STORE_EXITING |
3796 CPU_BASED_INVLPG_EXITING);
3797 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
3798 vmx_capability.ept, vmx_capability.vpid);
3801 min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
3802 #ifdef CONFIG_X86_64
3803 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
3805 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT |
3806 VM_EXIT_CLEAR_BNDCFGS;
3807 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
3808 &_vmexit_control) < 0)
3811 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
3812 opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
3813 PIN_BASED_VMX_PREEMPTION_TIMER;
3814 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
3815 &_pin_based_exec_control) < 0)
3818 if (cpu_has_broken_vmx_preemption_timer())
3819 _pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3820 if (!(_cpu_based_2nd_exec_control &
3821 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
3822 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
3824 min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
3825 opt = VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS;
3826 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
3827 &_vmentry_control) < 0)
3830 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
3832 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
3833 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
3836 #ifdef CONFIG_X86_64
3837 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
3838 if (vmx_msr_high & (1u<<16))
3842 /* Require Write-Back (WB) memory type for VMCS accesses. */
3843 if (((vmx_msr_high >> 18) & 15) != 6)
3846 vmcs_conf->size = vmx_msr_high & 0x1fff;
3847 vmcs_conf->order = get_order(vmcs_conf->size);
3848 vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
3849 vmcs_conf->revision_id = vmx_msr_low;
3851 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
3852 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
3853 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
3854 vmcs_conf->vmexit_ctrl = _vmexit_control;
3855 vmcs_conf->vmentry_ctrl = _vmentry_control;
3857 cpu_has_load_ia32_efer =
3858 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3859 VM_ENTRY_LOAD_IA32_EFER)
3860 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3861 VM_EXIT_LOAD_IA32_EFER);
3863 cpu_has_load_perf_global_ctrl =
3864 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3865 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
3866 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3867 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
3870 * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
3871 * but due to errata below it can't be used. Workaround is to use
3872 * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
3874 * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
3879 * BC86,AAY89,BD102 (model 44)
3883 if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
3884 switch (boot_cpu_data.x86_model) {
3890 cpu_has_load_perf_global_ctrl = false;
3891 printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
3892 "does not work properly. Using workaround\n");
3899 if (boot_cpu_has(X86_FEATURE_XSAVES))
3900 rdmsrl(MSR_IA32_XSS, host_xss);
3905 static struct vmcs *alloc_vmcs_cpu(int cpu)
3907 int node = cpu_to_node(cpu);
3911 pages = __alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
3914 vmcs = page_address(pages);
3915 memset(vmcs, 0, vmcs_config.size);
3916 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
3920 static void free_vmcs(struct vmcs *vmcs)
3922 free_pages((unsigned long)vmcs, vmcs_config.order);
3926 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
3928 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
3930 if (!loaded_vmcs->vmcs)
3932 loaded_vmcs_clear(loaded_vmcs);
3933 free_vmcs(loaded_vmcs->vmcs);
3934 loaded_vmcs->vmcs = NULL;
3935 if (loaded_vmcs->msr_bitmap)
3936 free_page((unsigned long)loaded_vmcs->msr_bitmap);
3937 WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
3940 static struct vmcs *alloc_vmcs(void)
3942 return alloc_vmcs_cpu(raw_smp_processor_id());
3945 static int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
3947 loaded_vmcs->vmcs = alloc_vmcs();
3948 if (!loaded_vmcs->vmcs)
3951 loaded_vmcs->shadow_vmcs = NULL;
3952 loaded_vmcs_init(loaded_vmcs);
3954 if (cpu_has_vmx_msr_bitmap()) {
3955 loaded_vmcs->msr_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
3956 if (!loaded_vmcs->msr_bitmap)
3958 memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE);
3963 free_loaded_vmcs(loaded_vmcs);
3967 static void free_kvm_area(void)
3971 for_each_possible_cpu(cpu) {
3972 free_vmcs(per_cpu(vmxarea, cpu));
3973 per_cpu(vmxarea, cpu) = NULL;
3977 enum vmcs_field_type {
3978 VMCS_FIELD_TYPE_U16 = 0,
3979 VMCS_FIELD_TYPE_U64 = 1,
3980 VMCS_FIELD_TYPE_U32 = 2,
3981 VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
3984 static inline int vmcs_field_type(unsigned long field)
3986 if (0x1 & field) /* the *_HIGH fields are all 32 bit */
3987 return VMCS_FIELD_TYPE_U32;
3988 return (field >> 13) & 0x3 ;
3991 static inline int vmcs_field_readonly(unsigned long field)
3993 return (((field >> 10) & 0x3) == 1);
3996 static void init_vmcs_shadow_fields(void)
4000 /* No checks for read only fields yet */
4002 for (i = j = 0; i < max_shadow_read_write_fields; i++) {
4003 switch (shadow_read_write_fields[i]) {
4005 if (!kvm_mpx_supported())
4013 shadow_read_write_fields[j] =
4014 shadow_read_write_fields[i];
4017 max_shadow_read_write_fields = j;
4019 /* shadowed fields guest access without vmexit */
4020 for (i = 0; i < max_shadow_read_write_fields; i++) {
4021 unsigned long field = shadow_read_write_fields[i];
4023 clear_bit(field, vmx_vmwrite_bitmap);
4024 clear_bit(field, vmx_vmread_bitmap);
4025 if (vmcs_field_type(field) == VMCS_FIELD_TYPE_U64) {
4026 clear_bit(field + 1, vmx_vmwrite_bitmap);
4027 clear_bit(field + 1, vmx_vmread_bitmap);
4030 for (i = 0; i < max_shadow_read_only_fields; i++) {
4031 unsigned long field = shadow_read_only_fields[i];
4033 clear_bit(field, vmx_vmread_bitmap);
4034 if (vmcs_field_type(field) == VMCS_FIELD_TYPE_U64)
4035 clear_bit(field + 1, vmx_vmread_bitmap);
4039 static __init int alloc_kvm_area(void)
4043 for_each_possible_cpu(cpu) {
4046 vmcs = alloc_vmcs_cpu(cpu);
4052 per_cpu(vmxarea, cpu) = vmcs;
4057 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
4058 struct kvm_segment *save)
4060 if (!emulate_invalid_guest_state) {
4062 * CS and SS RPL should be equal during guest entry according
4063 * to VMX spec, but in reality it is not always so. Since vcpu
4064 * is in the middle of the transition from real mode to
4065 * protected mode it is safe to assume that RPL 0 is a good
4068 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
4069 save->selector &= ~SEGMENT_RPL_MASK;
4070 save->dpl = save->selector & SEGMENT_RPL_MASK;
4073 vmx_set_segment(vcpu, save, seg);
4076 static void enter_pmode(struct kvm_vcpu *vcpu)
4078 unsigned long flags;
4079 struct vcpu_vmx *vmx = to_vmx(vcpu);
4082 * Update real mode segment cache. It may be not up-to-date if sement
4083 * register was written while vcpu was in a guest mode.
4085 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
4086 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
4087 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
4088 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
4089 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
4090 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
4092 vmx->rmode.vm86_active = 0;
4094 vmx_segment_cache_clear(vmx);
4096 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
4098 flags = vmcs_readl(GUEST_RFLAGS);
4099 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
4100 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
4101 vmcs_writel(GUEST_RFLAGS, flags);
4103 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
4104 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
4106 update_exception_bitmap(vcpu);
4108 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
4109 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
4110 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
4111 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
4112 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
4113 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
4116 static void fix_rmode_seg(int seg, struct kvm_segment *save)
4118 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4119 struct kvm_segment var = *save;
4122 if (seg == VCPU_SREG_CS)
4125 if (!emulate_invalid_guest_state) {
4126 var.selector = var.base >> 4;
4127 var.base = var.base & 0xffff0;
4137 if (save->base & 0xf)
4138 printk_once(KERN_WARNING "kvm: segment base is not "
4139 "paragraph aligned when entering "
4140 "protected mode (seg=%d)", seg);
4143 vmcs_write16(sf->selector, var.selector);
4144 vmcs_writel(sf->base, var.base);
4145 vmcs_write32(sf->limit, var.limit);
4146 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
4149 static void enter_rmode(struct kvm_vcpu *vcpu)
4151 unsigned long flags;
4152 struct vcpu_vmx *vmx = to_vmx(vcpu);
4154 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
4155 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
4156 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
4157 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
4158 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
4159 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
4160 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
4162 vmx->rmode.vm86_active = 1;
4165 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
4166 * vcpu. Warn the user that an update is overdue.
4168 if (!vcpu->kvm->arch.tss_addr)
4169 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
4170 "called before entering vcpu\n");
4172 vmx_segment_cache_clear(vmx);
4174 vmcs_writel(GUEST_TR_BASE, vcpu->kvm->arch.tss_addr);
4175 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
4176 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4178 flags = vmcs_readl(GUEST_RFLAGS);
4179 vmx->rmode.save_rflags = flags;
4181 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
4183 vmcs_writel(GUEST_RFLAGS, flags);
4184 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
4185 update_exception_bitmap(vcpu);
4187 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
4188 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
4189 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
4190 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
4191 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
4192 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
4194 kvm_mmu_reset_context(vcpu);
4197 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
4199 struct vcpu_vmx *vmx = to_vmx(vcpu);
4200 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
4206 * Force kernel_gs_base reloading before EFER changes, as control
4207 * of this msr depends on is_long_mode().
4209 vmx_load_host_state(to_vmx(vcpu));
4210 vcpu->arch.efer = efer;
4211 if (efer & EFER_LMA) {
4212 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
4215 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
4217 msr->data = efer & ~EFER_LME;
4222 #ifdef CONFIG_X86_64
4224 static void enter_lmode(struct kvm_vcpu *vcpu)
4228 vmx_segment_cache_clear(to_vmx(vcpu));
4230 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
4231 if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
4232 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
4234 vmcs_write32(GUEST_TR_AR_BYTES,
4235 (guest_tr_ar & ~VMX_AR_TYPE_MASK)
4236 | VMX_AR_TYPE_BUSY_64_TSS);
4238 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
4241 static void exit_lmode(struct kvm_vcpu *vcpu)
4243 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
4244 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
4249 static inline void __vmx_flush_tlb(struct kvm_vcpu *vcpu, int vpid)
4252 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
4254 ept_sync_context(construct_eptp(vcpu, vcpu->arch.mmu.root_hpa));
4256 vpid_sync_context(vpid);
4260 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
4262 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->vpid);
4265 static void vmx_flush_tlb_ept_only(struct kvm_vcpu *vcpu)
4268 vmx_flush_tlb(vcpu);
4271 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
4273 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
4275 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
4276 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
4279 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
4281 if (enable_ept && is_paging(vcpu))
4282 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
4283 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
4286 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
4288 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
4290 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
4291 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
4294 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
4296 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4298 if (!test_bit(VCPU_EXREG_PDPTR,
4299 (unsigned long *)&vcpu->arch.regs_dirty))
4302 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
4303 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
4304 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
4305 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
4306 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
4310 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
4312 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4314 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
4315 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
4316 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
4317 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
4318 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
4321 __set_bit(VCPU_EXREG_PDPTR,
4322 (unsigned long *)&vcpu->arch.regs_avail);
4323 __set_bit(VCPU_EXREG_PDPTR,
4324 (unsigned long *)&vcpu->arch.regs_dirty);
4327 static bool nested_guest_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
4329 u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed0;
4330 u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed1;
4331 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4333 if (to_vmx(vcpu)->nested.nested_vmx_secondary_ctls_high &
4334 SECONDARY_EXEC_UNRESTRICTED_GUEST &&
4335 nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
4336 fixed0 &= ~(X86_CR0_PE | X86_CR0_PG);
4338 return fixed_bits_valid(val, fixed0, fixed1);
4341 static bool nested_host_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
4343 u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed0;
4344 u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed1;
4346 return fixed_bits_valid(val, fixed0, fixed1);
4349 static bool nested_cr4_valid(struct kvm_vcpu *vcpu, unsigned long val)
4351 u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr4_fixed0;
4352 u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr4_fixed1;
4354 return fixed_bits_valid(val, fixed0, fixed1);
4357 /* No difference in the restrictions on guest and host CR4 in VMX operation. */
4358 #define nested_guest_cr4_valid nested_cr4_valid
4359 #define nested_host_cr4_valid nested_cr4_valid
4361 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
4363 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
4365 struct kvm_vcpu *vcpu)
4367 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
4368 vmx_decache_cr3(vcpu);
4369 if (!(cr0 & X86_CR0_PG)) {
4370 /* From paging/starting to nonpaging */
4371 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
4372 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
4373 (CPU_BASED_CR3_LOAD_EXITING |
4374 CPU_BASED_CR3_STORE_EXITING));
4375 vcpu->arch.cr0 = cr0;
4376 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
4377 } else if (!is_paging(vcpu)) {
4378 /* From nonpaging to paging */
4379 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
4380 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
4381 ~(CPU_BASED_CR3_LOAD_EXITING |
4382 CPU_BASED_CR3_STORE_EXITING));
4383 vcpu->arch.cr0 = cr0;
4384 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
4387 if (!(cr0 & X86_CR0_WP))
4388 *hw_cr0 &= ~X86_CR0_WP;
4391 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
4393 struct vcpu_vmx *vmx = to_vmx(vcpu);
4394 unsigned long hw_cr0;
4396 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK);
4397 if (enable_unrestricted_guest)
4398 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
4400 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
4402 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
4405 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
4409 #ifdef CONFIG_X86_64
4410 if (vcpu->arch.efer & EFER_LME) {
4411 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
4413 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
4419 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
4421 vmcs_writel(CR0_READ_SHADOW, cr0);
4422 vmcs_writel(GUEST_CR0, hw_cr0);
4423 vcpu->arch.cr0 = cr0;
4425 /* depends on vcpu->arch.cr0 to be set to a new value */
4426 vmx->emulation_required = emulation_required(vcpu);
4429 static int get_ept_level(struct kvm_vcpu *vcpu)
4431 if (cpu_has_vmx_ept_5levels() && (cpuid_maxphyaddr(vcpu) > 48))
4436 static u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa)
4438 u64 eptp = VMX_EPTP_MT_WB;
4440 eptp |= (get_ept_level(vcpu) == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
4442 if (enable_ept_ad_bits &&
4443 (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
4444 eptp |= VMX_EPTP_AD_ENABLE_BIT;
4445 eptp |= (root_hpa & PAGE_MASK);
4450 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
4452 unsigned long guest_cr3;
4457 eptp = construct_eptp(vcpu, cr3);
4458 vmcs_write64(EPT_POINTER, eptp);
4459 if (is_paging(vcpu) || is_guest_mode(vcpu))
4460 guest_cr3 = kvm_read_cr3(vcpu);
4462 guest_cr3 = vcpu->kvm->arch.ept_identity_map_addr;
4463 ept_load_pdptrs(vcpu);
4466 vmx_flush_tlb(vcpu);
4467 vmcs_writel(GUEST_CR3, guest_cr3);
4470 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
4473 * Pass through host's Machine Check Enable value to hw_cr4, which
4474 * is in force while we are in guest mode. Do not let guests control
4475 * this bit, even if host CR4.MCE == 0.
4477 unsigned long hw_cr4 =
4478 (cr4_read_shadow() & X86_CR4_MCE) |
4479 (cr4 & ~X86_CR4_MCE) |
4480 (to_vmx(vcpu)->rmode.vm86_active ?
4481 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
4483 if (cr4 & X86_CR4_VMXE) {
4485 * To use VMXON (and later other VMX instructions), a guest
4486 * must first be able to turn on cr4.VMXE (see handle_vmon()).
4487 * So basically the check on whether to allow nested VMX
4490 if (!nested_vmx_allowed(vcpu))
4494 if (to_vmx(vcpu)->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
4497 vcpu->arch.cr4 = cr4;
4499 if (!is_paging(vcpu)) {
4500 hw_cr4 &= ~X86_CR4_PAE;
4501 hw_cr4 |= X86_CR4_PSE;
4502 } else if (!(cr4 & X86_CR4_PAE)) {
4503 hw_cr4 &= ~X86_CR4_PAE;
4507 if (!enable_unrestricted_guest && !is_paging(vcpu))
4509 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
4510 * hardware. To emulate this behavior, SMEP/SMAP/PKU needs
4511 * to be manually disabled when guest switches to non-paging
4514 * If !enable_unrestricted_guest, the CPU is always running
4515 * with CR0.PG=1 and CR4 needs to be modified.
4516 * If enable_unrestricted_guest, the CPU automatically
4517 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
4519 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
4521 vmcs_writel(CR4_READ_SHADOW, cr4);
4522 vmcs_writel(GUEST_CR4, hw_cr4);
4526 static void vmx_get_segment(struct kvm_vcpu *vcpu,
4527 struct kvm_segment *var, int seg)
4529 struct vcpu_vmx *vmx = to_vmx(vcpu);
4532 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
4533 *var = vmx->rmode.segs[seg];
4534 if (seg == VCPU_SREG_TR
4535 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
4537 var->base = vmx_read_guest_seg_base(vmx, seg);
4538 var->selector = vmx_read_guest_seg_selector(vmx, seg);
4541 var->base = vmx_read_guest_seg_base(vmx, seg);
4542 var->limit = vmx_read_guest_seg_limit(vmx, seg);
4543 var->selector = vmx_read_guest_seg_selector(vmx, seg);
4544 ar = vmx_read_guest_seg_ar(vmx, seg);
4545 var->unusable = (ar >> 16) & 1;
4546 var->type = ar & 15;
4547 var->s = (ar >> 4) & 1;
4548 var->dpl = (ar >> 5) & 3;
4550 * Some userspaces do not preserve unusable property. Since usable
4551 * segment has to be present according to VMX spec we can use present
4552 * property to amend userspace bug by making unusable segment always
4553 * nonpresent. vmx_segment_access_rights() already marks nonpresent
4554 * segment as unusable.
4556 var->present = !var->unusable;
4557 var->avl = (ar >> 12) & 1;
4558 var->l = (ar >> 13) & 1;
4559 var->db = (ar >> 14) & 1;
4560 var->g = (ar >> 15) & 1;
4563 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
4565 struct kvm_segment s;
4567 if (to_vmx(vcpu)->rmode.vm86_active) {
4568 vmx_get_segment(vcpu, &s, seg);
4571 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
4574 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
4576 struct vcpu_vmx *vmx = to_vmx(vcpu);
4578 if (unlikely(vmx->rmode.vm86_active))
4581 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
4582 return VMX_AR_DPL(ar);
4586 static u32 vmx_segment_access_rights(struct kvm_segment *var)
4590 if (var->unusable || !var->present)
4593 ar = var->type & 15;
4594 ar |= (var->s & 1) << 4;
4595 ar |= (var->dpl & 3) << 5;
4596 ar |= (var->present & 1) << 7;
4597 ar |= (var->avl & 1) << 12;
4598 ar |= (var->l & 1) << 13;
4599 ar |= (var->db & 1) << 14;
4600 ar |= (var->g & 1) << 15;
4606 static void vmx_set_segment(struct kvm_vcpu *vcpu,
4607 struct kvm_segment *var, int seg)
4609 struct vcpu_vmx *vmx = to_vmx(vcpu);
4610 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4612 vmx_segment_cache_clear(vmx);
4614 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
4615 vmx->rmode.segs[seg] = *var;
4616 if (seg == VCPU_SREG_TR)
4617 vmcs_write16(sf->selector, var->selector);
4619 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
4623 vmcs_writel(sf->base, var->base);
4624 vmcs_write32(sf->limit, var->limit);
4625 vmcs_write16(sf->selector, var->selector);
4628 * Fix the "Accessed" bit in AR field of segment registers for older
4630 * IA32 arch specifies that at the time of processor reset the
4631 * "Accessed" bit in the AR field of segment registers is 1. And qemu
4632 * is setting it to 0 in the userland code. This causes invalid guest
4633 * state vmexit when "unrestricted guest" mode is turned on.
4634 * Fix for this setup issue in cpu_reset is being pushed in the qemu
4635 * tree. Newer qemu binaries with that qemu fix would not need this
4638 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
4639 var->type |= 0x1; /* Accessed */
4641 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
4644 vmx->emulation_required = emulation_required(vcpu);
4647 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4649 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
4651 *db = (ar >> 14) & 1;
4652 *l = (ar >> 13) & 1;
4655 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4657 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
4658 dt->address = vmcs_readl(GUEST_IDTR_BASE);
4661 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4663 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
4664 vmcs_writel(GUEST_IDTR_BASE, dt->address);
4667 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4669 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
4670 dt->address = vmcs_readl(GUEST_GDTR_BASE);
4673 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4675 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
4676 vmcs_writel(GUEST_GDTR_BASE, dt->address);
4679 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
4681 struct kvm_segment var;
4684 vmx_get_segment(vcpu, &var, seg);
4686 if (seg == VCPU_SREG_CS)
4688 ar = vmx_segment_access_rights(&var);
4690 if (var.base != (var.selector << 4))
4692 if (var.limit != 0xffff)
4700 static bool code_segment_valid(struct kvm_vcpu *vcpu)
4702 struct kvm_segment cs;
4703 unsigned int cs_rpl;
4705 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4706 cs_rpl = cs.selector & SEGMENT_RPL_MASK;
4710 if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
4714 if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
4715 if (cs.dpl > cs_rpl)
4718 if (cs.dpl != cs_rpl)
4724 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
4728 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
4730 struct kvm_segment ss;
4731 unsigned int ss_rpl;
4733 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4734 ss_rpl = ss.selector & SEGMENT_RPL_MASK;
4738 if (ss.type != 3 && ss.type != 7)
4742 if (ss.dpl != ss_rpl) /* DPL != RPL */
4750 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
4752 struct kvm_segment var;
4755 vmx_get_segment(vcpu, &var, seg);
4756 rpl = var.selector & SEGMENT_RPL_MASK;
4764 if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
4765 if (var.dpl < rpl) /* DPL < RPL */
4769 /* TODO: Add other members to kvm_segment_field to allow checking for other access
4775 static bool tr_valid(struct kvm_vcpu *vcpu)
4777 struct kvm_segment tr;
4779 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
4783 if (tr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4785 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
4793 static bool ldtr_valid(struct kvm_vcpu *vcpu)
4795 struct kvm_segment ldtr;
4797 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
4801 if (ldtr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4811 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
4813 struct kvm_segment cs, ss;
4815 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4816 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4818 return ((cs.selector & SEGMENT_RPL_MASK) ==
4819 (ss.selector & SEGMENT_RPL_MASK));
4823 * Check if guest state is valid. Returns true if valid, false if
4825 * We assume that registers are always usable
4827 static bool guest_state_valid(struct kvm_vcpu *vcpu)
4829 if (enable_unrestricted_guest)
4832 /* real mode guest state checks */
4833 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
4834 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
4836 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
4838 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
4840 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
4842 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
4844 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
4847 /* protected mode guest state checks */
4848 if (!cs_ss_rpl_check(vcpu))
4850 if (!code_segment_valid(vcpu))
4852 if (!stack_segment_valid(vcpu))
4854 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
4856 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
4858 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
4860 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
4862 if (!tr_valid(vcpu))
4864 if (!ldtr_valid(vcpu))
4868 * - Add checks on RIP
4869 * - Add checks on RFLAGS
4875 static bool page_address_valid(struct kvm_vcpu *vcpu, gpa_t gpa)
4877 return PAGE_ALIGNED(gpa) && !(gpa >> cpuid_maxphyaddr(vcpu));
4880 static int init_rmode_tss(struct kvm *kvm)
4886 idx = srcu_read_lock(&kvm->srcu);
4887 fn = kvm->arch.tss_addr >> PAGE_SHIFT;
4888 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4891 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
4892 r = kvm_write_guest_page(kvm, fn++, &data,
4893 TSS_IOPB_BASE_OFFSET, sizeof(u16));
4896 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
4899 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4903 r = kvm_write_guest_page(kvm, fn, &data,
4904 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
4907 srcu_read_unlock(&kvm->srcu, idx);
4911 static int init_rmode_identity_map(struct kvm *kvm)
4914 kvm_pfn_t identity_map_pfn;
4920 /* Protect kvm->arch.ept_identity_pagetable_done. */
4921 mutex_lock(&kvm->slots_lock);
4923 if (likely(kvm->arch.ept_identity_pagetable_done))
4926 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
4928 r = alloc_identity_pagetable(kvm);
4932 idx = srcu_read_lock(&kvm->srcu);
4933 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
4936 /* Set up identity-mapping pagetable for EPT in real mode */
4937 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
4938 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
4939 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
4940 r = kvm_write_guest_page(kvm, identity_map_pfn,
4941 &tmp, i * sizeof(tmp), sizeof(tmp));
4945 kvm->arch.ept_identity_pagetable_done = true;
4948 srcu_read_unlock(&kvm->srcu, idx);
4951 mutex_unlock(&kvm->slots_lock);
4955 static void seg_setup(int seg)
4957 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4960 vmcs_write16(sf->selector, 0);
4961 vmcs_writel(sf->base, 0);
4962 vmcs_write32(sf->limit, 0xffff);
4964 if (seg == VCPU_SREG_CS)
4965 ar |= 0x08; /* code segment */
4967 vmcs_write32(sf->ar_bytes, ar);
4970 static int alloc_apic_access_page(struct kvm *kvm)
4975 mutex_lock(&kvm->slots_lock);
4976 if (kvm->arch.apic_access_page_done)
4978 r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
4979 APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
4983 page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
4984 if (is_error_page(page)) {
4990 * Do not pin the page in memory, so that memory hot-unplug
4991 * is able to migrate it.
4994 kvm->arch.apic_access_page_done = true;
4996 mutex_unlock(&kvm->slots_lock);
5000 static int alloc_identity_pagetable(struct kvm *kvm)
5002 /* Called with kvm->slots_lock held. */
5006 BUG_ON(kvm->arch.ept_identity_pagetable_done);
5008 r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
5009 kvm->arch.ept_identity_map_addr, PAGE_SIZE);
5014 static int allocate_vpid(void)
5020 spin_lock(&vmx_vpid_lock);
5021 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
5022 if (vpid < VMX_NR_VPIDS)
5023 __set_bit(vpid, vmx_vpid_bitmap);
5026 spin_unlock(&vmx_vpid_lock);
5030 static void free_vpid(int vpid)
5032 if (!enable_vpid || vpid == 0)
5034 spin_lock(&vmx_vpid_lock);
5035 __clear_bit(vpid, vmx_vpid_bitmap);
5036 spin_unlock(&vmx_vpid_lock);
5039 static void __always_inline vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
5042 int f = sizeof(unsigned long);
5044 if (!cpu_has_vmx_msr_bitmap())
5048 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
5049 * have the write-low and read-high bitmap offsets the wrong way round.
5050 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
5052 if (msr <= 0x1fff) {
5053 if (type & MSR_TYPE_R)
5055 __clear_bit(msr, msr_bitmap + 0x000 / f);
5057 if (type & MSR_TYPE_W)
5059 __clear_bit(msr, msr_bitmap + 0x800 / f);
5061 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
5063 if (type & MSR_TYPE_R)
5065 __clear_bit(msr, msr_bitmap + 0x400 / f);
5067 if (type & MSR_TYPE_W)
5069 __clear_bit(msr, msr_bitmap + 0xc00 / f);
5074 static void __always_inline vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
5077 int f = sizeof(unsigned long);
5079 if (!cpu_has_vmx_msr_bitmap())
5083 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
5084 * have the write-low and read-high bitmap offsets the wrong way round.
5085 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
5087 if (msr <= 0x1fff) {
5088 if (type & MSR_TYPE_R)
5090 __set_bit(msr, msr_bitmap + 0x000 / f);
5092 if (type & MSR_TYPE_W)
5094 __set_bit(msr, msr_bitmap + 0x800 / f);
5096 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
5098 if (type & MSR_TYPE_R)
5100 __set_bit(msr, msr_bitmap + 0x400 / f);
5102 if (type & MSR_TYPE_W)
5104 __set_bit(msr, msr_bitmap + 0xc00 / f);
5109 static void __always_inline vmx_set_intercept_for_msr(unsigned long *msr_bitmap,
5110 u32 msr, int type, bool value)
5113 vmx_enable_intercept_for_msr(msr_bitmap, msr, type);
5115 vmx_disable_intercept_for_msr(msr_bitmap, msr, type);
5119 * If a msr is allowed by L0, we should check whether it is allowed by L1.
5120 * The corresponding bit will be cleared unless both of L0 and L1 allow it.
5122 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
5123 unsigned long *msr_bitmap_nested,
5126 int f = sizeof(unsigned long);
5128 if (!cpu_has_vmx_msr_bitmap()) {
5134 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
5135 * have the write-low and read-high bitmap offsets the wrong way round.
5136 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
5138 if (msr <= 0x1fff) {
5139 if (type & MSR_TYPE_R &&
5140 !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
5142 __clear_bit(msr, msr_bitmap_nested + 0x000 / f);
5144 if (type & MSR_TYPE_W &&
5145 !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
5147 __clear_bit(msr, msr_bitmap_nested + 0x800 / f);
5149 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
5151 if (type & MSR_TYPE_R &&
5152 !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
5154 __clear_bit(msr, msr_bitmap_nested + 0x400 / f);
5156 if (type & MSR_TYPE_W &&
5157 !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
5159 __clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
5164 static u8 vmx_msr_bitmap_mode(struct kvm_vcpu *vcpu)
5168 if (cpu_has_secondary_exec_ctrls() &&
5169 (vmcs_read32(SECONDARY_VM_EXEC_CONTROL) &
5170 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
5171 mode |= MSR_BITMAP_MODE_X2APIC;
5172 if (enable_apicv && kvm_vcpu_apicv_active(vcpu))
5173 mode |= MSR_BITMAP_MODE_X2APIC_APICV;
5176 if (is_long_mode(vcpu))
5177 mode |= MSR_BITMAP_MODE_LM;
5182 #define X2APIC_MSR(r) (APIC_BASE_MSR + ((r) >> 4))
5184 static void vmx_update_msr_bitmap_x2apic(unsigned long *msr_bitmap,
5189 for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
5190 unsigned word = msr / BITS_PER_LONG;
5191 msr_bitmap[word] = (mode & MSR_BITMAP_MODE_X2APIC_APICV) ? 0 : ~0;
5192 msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
5195 if (mode & MSR_BITMAP_MODE_X2APIC) {
5197 * TPR reads and writes can be virtualized even if virtual interrupt
5198 * delivery is not in use.
5200 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW);
5201 if (mode & MSR_BITMAP_MODE_X2APIC_APICV) {
5202 vmx_enable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_R);
5203 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_EOI), MSR_TYPE_W);
5204 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W);
5209 static void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu)
5211 struct vcpu_vmx *vmx = to_vmx(vcpu);
5212 unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
5213 u8 mode = vmx_msr_bitmap_mode(vcpu);
5214 u8 changed = mode ^ vmx->msr_bitmap_mode;
5219 vmx_set_intercept_for_msr(msr_bitmap, MSR_KERNEL_GS_BASE, MSR_TYPE_RW,
5220 !(mode & MSR_BITMAP_MODE_LM));
5222 if (changed & (MSR_BITMAP_MODE_X2APIC | MSR_BITMAP_MODE_X2APIC_APICV))
5223 vmx_update_msr_bitmap_x2apic(msr_bitmap, mode);
5225 vmx->msr_bitmap_mode = mode;
5228 static bool vmx_get_enable_apicv(struct kvm_vcpu *vcpu)
5230 return enable_apicv;
5233 static void nested_mark_vmcs12_pages_dirty(struct kvm_vcpu *vcpu)
5235 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5239 * Don't need to mark the APIC access page dirty; it is never
5240 * written to by the CPU during APIC virtualization.
5243 if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
5244 gfn = vmcs12->virtual_apic_page_addr >> PAGE_SHIFT;
5245 kvm_vcpu_mark_page_dirty(vcpu, gfn);
5248 if (nested_cpu_has_posted_intr(vmcs12)) {
5249 gfn = vmcs12->posted_intr_desc_addr >> PAGE_SHIFT;
5250 kvm_vcpu_mark_page_dirty(vcpu, gfn);
5255 static void vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
5257 struct vcpu_vmx *vmx = to_vmx(vcpu);
5262 if (!vmx->nested.pi_desc || !vmx->nested.pi_pending)
5265 vmx->nested.pi_pending = false;
5266 if (!pi_test_and_clear_on(vmx->nested.pi_desc))
5269 max_irr = find_last_bit((unsigned long *)vmx->nested.pi_desc->pir, 256);
5270 if (max_irr != 256) {
5271 vapic_page = kmap(vmx->nested.virtual_apic_page);
5272 __kvm_apic_update_irr(vmx->nested.pi_desc->pir, vapic_page);
5273 kunmap(vmx->nested.virtual_apic_page);
5275 status = vmcs_read16(GUEST_INTR_STATUS);
5276 if ((u8)max_irr > ((u8)status & 0xff)) {
5278 status |= (u8)max_irr;
5279 vmcs_write16(GUEST_INTR_STATUS, status);
5283 nested_mark_vmcs12_pages_dirty(vcpu);
5286 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
5290 int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
5292 if (vcpu->mode == IN_GUEST_MODE) {
5294 * The vector of interrupt to be delivered to vcpu had
5295 * been set in PIR before this function.
5297 * Following cases will be reached in this block, and
5298 * we always send a notification event in all cases as
5301 * Case 1: vcpu keeps in non-root mode. Sending a
5302 * notification event posts the interrupt to vcpu.
5304 * Case 2: vcpu exits to root mode and is still
5305 * runnable. PIR will be synced to vIRR before the
5306 * next vcpu entry. Sending a notification event in
5307 * this case has no effect, as vcpu is not in root
5310 * Case 3: vcpu exits to root mode and is blocked.
5311 * vcpu_block() has already synced PIR to vIRR and
5312 * never blocks vcpu if vIRR is not cleared. Therefore,
5313 * a blocked vcpu here does not wait for any requested
5314 * interrupts in PIR, and sending a notification event
5315 * which has no effect is safe here.
5318 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
5325 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
5328 struct vcpu_vmx *vmx = to_vmx(vcpu);
5330 if (is_guest_mode(vcpu) &&
5331 vector == vmx->nested.posted_intr_nv) {
5333 * If a posted intr is not recognized by hardware,
5334 * we will accomplish it in the next vmentry.
5336 vmx->nested.pi_pending = true;
5337 kvm_make_request(KVM_REQ_EVENT, vcpu);
5338 /* the PIR and ON have been set by L1. */
5339 if (!kvm_vcpu_trigger_posted_interrupt(vcpu, true))
5340 kvm_vcpu_kick(vcpu);
5346 * Send interrupt to vcpu via posted interrupt way.
5347 * 1. If target vcpu is running(non-root mode), send posted interrupt
5348 * notification to vcpu and hardware will sync PIR to vIRR atomically.
5349 * 2. If target vcpu isn't running(root mode), kick it to pick up the
5350 * interrupt from PIR in next vmentry.
5352 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
5354 struct vcpu_vmx *vmx = to_vmx(vcpu);
5357 r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
5361 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
5364 /* If a previous notification has sent the IPI, nothing to do. */
5365 if (pi_test_and_set_on(&vmx->pi_desc))
5368 if (!kvm_vcpu_trigger_posted_interrupt(vcpu, false))
5369 kvm_vcpu_kick(vcpu);
5373 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
5374 * will not change in the lifetime of the guest.
5375 * Note that host-state that does change is set elsewhere. E.g., host-state
5376 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
5378 static void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
5383 unsigned long cr0, cr3, cr4;
5386 WARN_ON(cr0 & X86_CR0_TS);
5387 vmcs_writel(HOST_CR0, cr0); /* 22.2.3 */
5390 * Save the most likely value for this task's CR3 in the VMCS.
5391 * We can't use __get_current_cr3_fast() because we're not atomic.
5394 vmcs_writel(HOST_CR3, cr3); /* 22.2.3 FIXME: shadow tables */
5395 vmx->loaded_vmcs->vmcs_host_cr3 = cr3;
5397 /* Save the most likely value for this task's CR4 in the VMCS. */
5398 cr4 = cr4_read_shadow();
5399 vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
5400 vmx->loaded_vmcs->vmcs_host_cr4 = cr4;
5402 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
5403 #ifdef CONFIG_X86_64
5405 * Load null selectors, so we can avoid reloading them in
5406 * __vmx_load_host_state(), in case userspace uses the null selectors
5407 * too (the expected case).
5409 vmcs_write16(HOST_DS_SELECTOR, 0);
5410 vmcs_write16(HOST_ES_SELECTOR, 0);
5412 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
5413 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
5415 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
5416 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
5419 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
5420 vmx->host_idt_base = dt.address;
5422 vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
5424 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
5425 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
5426 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
5427 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
5429 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
5430 rdmsr(MSR_IA32_CR_PAT, low32, high32);
5431 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
5435 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
5437 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
5439 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
5440 if (is_guest_mode(&vmx->vcpu))
5441 vmx->vcpu.arch.cr4_guest_owned_bits &=
5442 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
5443 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
5446 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
5448 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
5450 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
5451 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
5452 /* Enable the preemption timer dynamically */
5453 pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
5454 return pin_based_exec_ctrl;
5457 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
5459 struct vcpu_vmx *vmx = to_vmx(vcpu);
5461 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
5462 if (cpu_has_secondary_exec_ctrls()) {
5463 if (kvm_vcpu_apicv_active(vcpu))
5464 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
5465 SECONDARY_EXEC_APIC_REGISTER_VIRT |
5466 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5468 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
5469 SECONDARY_EXEC_APIC_REGISTER_VIRT |
5470 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5473 if (cpu_has_vmx_msr_bitmap())
5474 vmx_update_msr_bitmap(vcpu);
5477 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
5479 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
5481 if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
5482 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
5484 if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
5485 exec_control &= ~CPU_BASED_TPR_SHADOW;
5486 #ifdef CONFIG_X86_64
5487 exec_control |= CPU_BASED_CR8_STORE_EXITING |
5488 CPU_BASED_CR8_LOAD_EXITING;
5492 exec_control |= CPU_BASED_CR3_STORE_EXITING |
5493 CPU_BASED_CR3_LOAD_EXITING |
5494 CPU_BASED_INVLPG_EXITING;
5495 return exec_control;
5498 static bool vmx_rdrand_supported(void)
5500 return vmcs_config.cpu_based_2nd_exec_ctrl &
5501 SECONDARY_EXEC_RDRAND;
5504 static bool vmx_rdseed_supported(void)
5506 return vmcs_config.cpu_based_2nd_exec_ctrl &
5507 SECONDARY_EXEC_RDSEED;
5510 static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
5512 struct kvm_vcpu *vcpu = &vmx->vcpu;
5514 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
5515 if (!cpu_need_virtualize_apic_accesses(vcpu))
5516 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
5518 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
5520 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
5521 enable_unrestricted_guest = 0;
5522 /* Enable INVPCID for non-ept guests may cause performance regression. */
5523 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
5525 if (!enable_unrestricted_guest)
5526 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
5528 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
5529 if (!kvm_vcpu_apicv_active(vcpu))
5530 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
5531 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5532 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
5533 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
5535 We can NOT enable shadow_vmcs here because we don't have yet
5538 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
5541 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
5543 if (vmx_xsaves_supported()) {
5544 /* Exposing XSAVES only when XSAVE is exposed */
5545 bool xsaves_enabled =
5546 guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
5547 guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
5549 if (!xsaves_enabled)
5550 exec_control &= ~SECONDARY_EXEC_XSAVES;
5554 vmx->nested.nested_vmx_secondary_ctls_high |=
5555 SECONDARY_EXEC_XSAVES;
5557 vmx->nested.nested_vmx_secondary_ctls_high &=
5558 ~SECONDARY_EXEC_XSAVES;
5562 if (vmx_rdtscp_supported()) {
5563 bool rdtscp_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP);
5564 if (!rdtscp_enabled)
5565 exec_control &= ~SECONDARY_EXEC_RDTSCP;
5569 vmx->nested.nested_vmx_secondary_ctls_high |=
5570 SECONDARY_EXEC_RDTSCP;
5572 vmx->nested.nested_vmx_secondary_ctls_high &=
5573 ~SECONDARY_EXEC_RDTSCP;
5577 if (vmx_invpcid_supported()) {
5578 /* Exposing INVPCID only when PCID is exposed */
5579 bool invpcid_enabled =
5580 guest_cpuid_has(vcpu, X86_FEATURE_INVPCID) &&
5581 guest_cpuid_has(vcpu, X86_FEATURE_PCID);
5583 if (!invpcid_enabled) {
5584 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
5585 guest_cpuid_clear(vcpu, X86_FEATURE_INVPCID);
5589 if (invpcid_enabled)
5590 vmx->nested.nested_vmx_secondary_ctls_high |=
5591 SECONDARY_EXEC_ENABLE_INVPCID;
5593 vmx->nested.nested_vmx_secondary_ctls_high &=
5594 ~SECONDARY_EXEC_ENABLE_INVPCID;
5598 if (vmx_rdrand_supported()) {
5599 bool rdrand_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDRAND);
5601 exec_control &= ~SECONDARY_EXEC_RDRAND;
5605 vmx->nested.nested_vmx_secondary_ctls_high |=
5606 SECONDARY_EXEC_RDRAND;
5608 vmx->nested.nested_vmx_secondary_ctls_high &=
5609 ~SECONDARY_EXEC_RDRAND;
5613 if (vmx_rdseed_supported()) {
5614 bool rdseed_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDSEED);
5616 exec_control &= ~SECONDARY_EXEC_RDSEED;
5620 vmx->nested.nested_vmx_secondary_ctls_high |=
5621 SECONDARY_EXEC_RDSEED;
5623 vmx->nested.nested_vmx_secondary_ctls_high &=
5624 ~SECONDARY_EXEC_RDSEED;
5628 vmx->secondary_exec_control = exec_control;
5631 static void ept_set_mmio_spte_mask(void)
5634 * EPT Misconfigurations can be generated if the value of bits 2:0
5635 * of an EPT paging-structure entry is 110b (write/execute).
5637 kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK,
5638 VMX_EPT_MISCONFIG_WX_VALUE);
5641 #define VMX_XSS_EXIT_BITMAP 0
5643 * Sets up the vmcs for emulated real mode.
5645 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
5647 #ifdef CONFIG_X86_64
5653 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
5654 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
5656 if (enable_shadow_vmcs) {
5657 vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
5658 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
5660 if (cpu_has_vmx_msr_bitmap())
5661 vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap));
5663 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
5666 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
5667 vmx->hv_deadline_tsc = -1;
5669 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
5671 if (cpu_has_secondary_exec_ctrls()) {
5672 vmx_compute_secondary_exec_control(vmx);
5673 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
5674 vmx->secondary_exec_control);
5677 if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
5678 vmcs_write64(EOI_EXIT_BITMAP0, 0);
5679 vmcs_write64(EOI_EXIT_BITMAP1, 0);
5680 vmcs_write64(EOI_EXIT_BITMAP2, 0);
5681 vmcs_write64(EOI_EXIT_BITMAP3, 0);
5683 vmcs_write16(GUEST_INTR_STATUS, 0);
5685 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
5686 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
5690 vmcs_write32(PLE_GAP, ple_gap);
5691 vmx->ple_window = ple_window;
5692 vmx->ple_window_dirty = true;
5695 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
5696 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
5697 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
5699 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
5700 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
5701 vmx_set_constant_host_state(vmx);
5702 #ifdef CONFIG_X86_64
5703 rdmsrl(MSR_FS_BASE, a);
5704 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
5705 rdmsrl(MSR_GS_BASE, a);
5706 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
5708 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
5709 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
5712 if (cpu_has_vmx_vmfunc())
5713 vmcs_write64(VM_FUNCTION_CONTROL, 0);
5715 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
5716 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
5717 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
5718 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
5719 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
5721 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
5722 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
5724 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
5725 u32 index = vmx_msr_index[i];
5726 u32 data_low, data_high;
5729 if (rdmsr_safe(index, &data_low, &data_high) < 0)
5731 if (wrmsr_safe(index, data_low, data_high) < 0)
5733 vmx->guest_msrs[j].index = i;
5734 vmx->guest_msrs[j].data = 0;
5735 vmx->guest_msrs[j].mask = -1ull;
5739 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
5740 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, vmx->arch_capabilities);
5742 vm_exit_controls_init(vmx, vmcs_config.vmexit_ctrl);
5744 /* 22.2.1, 20.8.1 */
5745 vm_entry_controls_init(vmx, vmcs_config.vmentry_ctrl);
5747 vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
5748 vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
5750 set_cr4_guest_host_mask(vmx);
5752 if (vmx_xsaves_supported())
5753 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
5756 ASSERT(vmx->pml_pg);
5757 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
5758 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5764 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
5766 struct vcpu_vmx *vmx = to_vmx(vcpu);
5767 struct msr_data apic_base_msr;
5770 vmx->rmode.vm86_active = 0;
5773 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
5774 kvm_set_cr8(vcpu, 0);
5777 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
5778 MSR_IA32_APICBASE_ENABLE;
5779 if (kvm_vcpu_is_reset_bsp(vcpu))
5780 apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
5781 apic_base_msr.host_initiated = true;
5782 kvm_set_apic_base(vcpu, &apic_base_msr);
5785 vmx_segment_cache_clear(vmx);
5787 seg_setup(VCPU_SREG_CS);
5788 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
5789 vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
5791 seg_setup(VCPU_SREG_DS);
5792 seg_setup(VCPU_SREG_ES);
5793 seg_setup(VCPU_SREG_FS);
5794 seg_setup(VCPU_SREG_GS);
5795 seg_setup(VCPU_SREG_SS);
5797 vmcs_write16(GUEST_TR_SELECTOR, 0);
5798 vmcs_writel(GUEST_TR_BASE, 0);
5799 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
5800 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
5802 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
5803 vmcs_writel(GUEST_LDTR_BASE, 0);
5804 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
5805 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
5808 vmcs_write32(GUEST_SYSENTER_CS, 0);
5809 vmcs_writel(GUEST_SYSENTER_ESP, 0);
5810 vmcs_writel(GUEST_SYSENTER_EIP, 0);
5811 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
5814 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
5815 kvm_rip_write(vcpu, 0xfff0);
5817 vmcs_writel(GUEST_GDTR_BASE, 0);
5818 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
5820 vmcs_writel(GUEST_IDTR_BASE, 0);
5821 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
5823 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
5824 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
5825 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
5829 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
5831 if (cpu_has_vmx_tpr_shadow() && !init_event) {
5832 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
5833 if (cpu_need_tpr_shadow(vcpu))
5834 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
5835 __pa(vcpu->arch.apic->regs));
5836 vmcs_write32(TPR_THRESHOLD, 0);
5839 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
5842 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
5844 cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
5845 vmx->vcpu.arch.cr0 = cr0;
5846 vmx_set_cr0(vcpu, cr0); /* enter rmode */
5847 vmx_set_cr4(vcpu, 0);
5848 vmx_set_efer(vcpu, 0);
5850 update_exception_bitmap(vcpu);
5852 vpid_sync_context(vmx->vpid);
5856 * In nested virtualization, check if L1 asked to exit on external interrupts.
5857 * For most existing hypervisors, this will always return true.
5859 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
5861 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5862 PIN_BASED_EXT_INTR_MASK;
5866 * In nested virtualization, check if L1 has set
5867 * VM_EXIT_ACK_INTR_ON_EXIT
5869 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
5871 return get_vmcs12(vcpu)->vm_exit_controls &
5872 VM_EXIT_ACK_INTR_ON_EXIT;
5875 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
5877 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5878 PIN_BASED_NMI_EXITING;
5881 static void enable_irq_window(struct kvm_vcpu *vcpu)
5883 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
5884 CPU_BASED_VIRTUAL_INTR_PENDING);
5887 static void enable_nmi_window(struct kvm_vcpu *vcpu)
5889 if (!cpu_has_virtual_nmis() ||
5890 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
5891 enable_irq_window(vcpu);
5895 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
5896 CPU_BASED_VIRTUAL_NMI_PENDING);
5899 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
5901 struct vcpu_vmx *vmx = to_vmx(vcpu);
5903 int irq = vcpu->arch.interrupt.nr;
5905 trace_kvm_inj_virq(irq);
5907 ++vcpu->stat.irq_injections;
5908 if (vmx->rmode.vm86_active) {
5910 if (vcpu->arch.interrupt.soft)
5911 inc_eip = vcpu->arch.event_exit_inst_len;
5912 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
5913 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5916 intr = irq | INTR_INFO_VALID_MASK;
5917 if (vcpu->arch.interrupt.soft) {
5918 intr |= INTR_TYPE_SOFT_INTR;
5919 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
5920 vmx->vcpu.arch.event_exit_inst_len);
5922 intr |= INTR_TYPE_EXT_INTR;
5923 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
5926 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
5928 struct vcpu_vmx *vmx = to_vmx(vcpu);
5930 if (!cpu_has_virtual_nmis()) {
5932 * Tracking the NMI-blocked state in software is built upon
5933 * finding the next open IRQ window. This, in turn, depends on
5934 * well-behaving guests: They have to keep IRQs disabled at
5935 * least as long as the NMI handler runs. Otherwise we may
5936 * cause NMI nesting, maybe breaking the guest. But as this is
5937 * highly unlikely, we can live with the residual risk.
5939 vmx->loaded_vmcs->soft_vnmi_blocked = 1;
5940 vmx->loaded_vmcs->vnmi_blocked_time = 0;
5943 ++vcpu->stat.nmi_injections;
5944 vmx->loaded_vmcs->nmi_known_unmasked = false;
5946 if (vmx->rmode.vm86_active) {
5947 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
5948 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5952 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
5953 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
5956 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
5958 struct vcpu_vmx *vmx = to_vmx(vcpu);
5961 if (!cpu_has_virtual_nmis())
5962 return vmx->loaded_vmcs->soft_vnmi_blocked;
5963 if (vmx->loaded_vmcs->nmi_known_unmasked)
5965 masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
5966 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
5970 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
5972 struct vcpu_vmx *vmx = to_vmx(vcpu);
5974 if (!cpu_has_virtual_nmis()) {
5975 if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
5976 vmx->loaded_vmcs->soft_vnmi_blocked = masked;
5977 vmx->loaded_vmcs->vnmi_blocked_time = 0;
5980 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
5982 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5983 GUEST_INTR_STATE_NMI);
5985 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
5986 GUEST_INTR_STATE_NMI);
5990 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
5992 if (to_vmx(vcpu)->nested.nested_run_pending)
5995 if (!cpu_has_virtual_nmis() &&
5996 to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
5999 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
6000 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
6001 | GUEST_INTR_STATE_NMI));
6004 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
6006 return (!to_vmx(vcpu)->nested.nested_run_pending &&
6007 vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
6008 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
6009 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
6012 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
6016 ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
6020 kvm->arch.tss_addr = addr;
6021 return init_rmode_tss(kvm);
6024 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
6029 * Update instruction length as we may reinject the exception
6030 * from user space while in guest debugging mode.
6032 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
6033 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
6034 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
6038 if (vcpu->guest_debug &
6039 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
6056 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
6057 int vec, u32 err_code)
6060 * Instruction with address size override prefix opcode 0x67
6061 * Cause the #SS fault with 0 error code in VM86 mode.
6063 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
6064 if (emulate_instruction(vcpu, 0) == EMULATE_DONE) {
6065 if (vcpu->arch.halt_request) {
6066 vcpu->arch.halt_request = 0;
6067 return kvm_vcpu_halt(vcpu);
6075 * Forward all other exceptions that are valid in real mode.
6076 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
6077 * the required debugging infrastructure rework.
6079 kvm_queue_exception(vcpu, vec);
6084 * Trigger machine check on the host. We assume all the MSRs are already set up
6085 * by the CPU and that we still run on the same CPU as the MCE occurred on.
6086 * We pass a fake environment to the machine check handler because we want
6087 * the guest to be always treated like user space, no matter what context
6088 * it used internally.
6090 static void kvm_machine_check(void)
6092 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
6093 struct pt_regs regs = {
6094 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
6095 .flags = X86_EFLAGS_IF,
6098 do_machine_check(®s, 0);
6102 static int handle_machine_check(struct kvm_vcpu *vcpu)
6104 /* already handled by vcpu_run */
6108 static int handle_exception(struct kvm_vcpu *vcpu)
6110 struct vcpu_vmx *vmx = to_vmx(vcpu);
6111 struct kvm_run *kvm_run = vcpu->run;
6112 u32 intr_info, ex_no, error_code;
6113 unsigned long cr2, rip, dr6;
6115 enum emulation_result er;
6117 vect_info = vmx->idt_vectoring_info;
6118 intr_info = vmx->exit_intr_info;
6120 if (is_machine_check(intr_info))
6121 return handle_machine_check(vcpu);
6123 if (is_nmi(intr_info))
6124 return 1; /* already handled by vmx_vcpu_run() */
6126 if (is_invalid_opcode(intr_info)) {
6127 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
6128 if (er == EMULATE_USER_EXIT)
6130 if (er != EMULATE_DONE)
6131 kvm_queue_exception(vcpu, UD_VECTOR);
6136 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
6137 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
6140 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
6141 * MMIO, it is better to report an internal error.
6142 * See the comments in vmx_handle_exit.
6144 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
6145 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
6146 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6147 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
6148 vcpu->run->internal.ndata = 3;
6149 vcpu->run->internal.data[0] = vect_info;
6150 vcpu->run->internal.data[1] = intr_info;
6151 vcpu->run->internal.data[2] = error_code;
6155 if (is_page_fault(intr_info)) {
6156 cr2 = vmcs_readl(EXIT_QUALIFICATION);
6157 /* EPT won't cause page fault directly */
6158 WARN_ON_ONCE(!vcpu->arch.apf.host_apf_reason && enable_ept);
6159 return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0,
6163 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
6165 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
6166 return handle_rmode_exception(vcpu, ex_no, error_code);
6170 kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
6173 dr6 = vmcs_readl(EXIT_QUALIFICATION);
6174 if (!(vcpu->guest_debug &
6175 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
6176 vcpu->arch.dr6 &= ~15;
6177 vcpu->arch.dr6 |= dr6 | DR6_RTM;
6178 if (is_icebp(intr_info))
6179 skip_emulated_instruction(vcpu);
6181 kvm_queue_exception(vcpu, DB_VECTOR);
6184 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
6185 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
6189 * Update instruction length as we may reinject #BP from
6190 * user space while in guest debugging mode. Reading it for
6191 * #DB as well causes no harm, it is not used in that case.
6193 vmx->vcpu.arch.event_exit_inst_len =
6194 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
6195 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6196 rip = kvm_rip_read(vcpu);
6197 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
6198 kvm_run->debug.arch.exception = ex_no;
6201 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
6202 kvm_run->ex.exception = ex_no;
6203 kvm_run->ex.error_code = error_code;
6209 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
6211 ++vcpu->stat.irq_exits;
6215 static int handle_triple_fault(struct kvm_vcpu *vcpu)
6217 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6218 vcpu->mmio_needed = 0;
6222 static int handle_io(struct kvm_vcpu *vcpu)
6224 unsigned long exit_qualification;
6225 int size, in, string, ret;
6228 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6229 string = (exit_qualification & 16) != 0;
6230 in = (exit_qualification & 8) != 0;
6232 ++vcpu->stat.io_exits;
6235 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6237 port = exit_qualification >> 16;
6238 size = (exit_qualification & 7) + 1;
6240 ret = kvm_skip_emulated_instruction(vcpu);
6243 * TODO: we might be squashing a KVM_GUESTDBG_SINGLESTEP-triggered
6244 * KVM_EXIT_DEBUG here.
6246 return kvm_fast_pio_out(vcpu, size, port) && ret;
6250 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
6253 * Patch in the VMCALL instruction:
6255 hypercall[0] = 0x0f;
6256 hypercall[1] = 0x01;
6257 hypercall[2] = 0xc1;
6260 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
6261 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
6263 if (is_guest_mode(vcpu)) {
6264 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6265 unsigned long orig_val = val;
6268 * We get here when L2 changed cr0 in a way that did not change
6269 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
6270 * but did change L0 shadowed bits. So we first calculate the
6271 * effective cr0 value that L1 would like to write into the
6272 * hardware. It consists of the L2-owned bits from the new
6273 * value combined with the L1-owned bits from L1's guest_cr0.
6275 val = (val & ~vmcs12->cr0_guest_host_mask) |
6276 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
6278 if (!nested_guest_cr0_valid(vcpu, val))
6281 if (kvm_set_cr0(vcpu, val))
6283 vmcs_writel(CR0_READ_SHADOW, orig_val);
6286 if (to_vmx(vcpu)->nested.vmxon &&
6287 !nested_host_cr0_valid(vcpu, val))
6290 return kvm_set_cr0(vcpu, val);
6294 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
6296 if (is_guest_mode(vcpu)) {
6297 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6298 unsigned long orig_val = val;
6300 /* analogously to handle_set_cr0 */
6301 val = (val & ~vmcs12->cr4_guest_host_mask) |
6302 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
6303 if (kvm_set_cr4(vcpu, val))
6305 vmcs_writel(CR4_READ_SHADOW, orig_val);
6308 return kvm_set_cr4(vcpu, val);
6311 static int handle_cr(struct kvm_vcpu *vcpu)
6313 unsigned long exit_qualification, val;
6319 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6320 cr = exit_qualification & 15;
6321 reg = (exit_qualification >> 8) & 15;
6322 switch ((exit_qualification >> 4) & 3) {
6323 case 0: /* mov to cr */
6324 val = kvm_register_readl(vcpu, reg);
6325 trace_kvm_cr_write(cr, val);
6328 err = handle_set_cr0(vcpu, val);
6329 return kvm_complete_insn_gp(vcpu, err);
6331 err = kvm_set_cr3(vcpu, val);
6332 return kvm_complete_insn_gp(vcpu, err);
6334 err = handle_set_cr4(vcpu, val);
6335 return kvm_complete_insn_gp(vcpu, err);
6337 u8 cr8_prev = kvm_get_cr8(vcpu);
6339 err = kvm_set_cr8(vcpu, cr8);
6340 ret = kvm_complete_insn_gp(vcpu, err);
6341 if (lapic_in_kernel(vcpu))
6343 if (cr8_prev <= cr8)
6346 * TODO: we might be squashing a
6347 * KVM_GUESTDBG_SINGLESTEP-triggered
6348 * KVM_EXIT_DEBUG here.
6350 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
6356 WARN_ONCE(1, "Guest should always own CR0.TS");
6357 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
6358 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
6359 return kvm_skip_emulated_instruction(vcpu);
6360 case 1: /*mov from cr*/
6363 val = kvm_read_cr3(vcpu);
6364 kvm_register_write(vcpu, reg, val);
6365 trace_kvm_cr_read(cr, val);
6366 return kvm_skip_emulated_instruction(vcpu);
6368 val = kvm_get_cr8(vcpu);
6369 kvm_register_write(vcpu, reg, val);
6370 trace_kvm_cr_read(cr, val);
6371 return kvm_skip_emulated_instruction(vcpu);
6375 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
6376 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
6377 kvm_lmsw(vcpu, val);
6379 return kvm_skip_emulated_instruction(vcpu);
6383 vcpu->run->exit_reason = 0;
6384 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
6385 (int)(exit_qualification >> 4) & 3, cr);
6389 static int handle_dr(struct kvm_vcpu *vcpu)
6391 unsigned long exit_qualification;
6394 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6395 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
6397 /* First, if DR does not exist, trigger UD */
6398 if (!kvm_require_dr(vcpu, dr))
6401 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
6402 if (!kvm_require_cpl(vcpu, 0))
6404 dr7 = vmcs_readl(GUEST_DR7);
6407 * As the vm-exit takes precedence over the debug trap, we
6408 * need to emulate the latter, either for the host or the
6409 * guest debugging itself.
6411 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6412 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
6413 vcpu->run->debug.arch.dr7 = dr7;
6414 vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
6415 vcpu->run->debug.arch.exception = DB_VECTOR;
6416 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
6419 vcpu->arch.dr6 &= ~15;
6420 vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
6421 kvm_queue_exception(vcpu, DB_VECTOR);
6426 if (vcpu->guest_debug == 0) {
6427 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6428 CPU_BASED_MOV_DR_EXITING);
6431 * No more DR vmexits; force a reload of the debug registers
6432 * and reenter on this instruction. The next vmexit will
6433 * retrieve the full state of the debug registers.
6435 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
6439 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
6440 if (exit_qualification & TYPE_MOV_FROM_DR) {
6443 if (kvm_get_dr(vcpu, dr, &val))
6445 kvm_register_write(vcpu, reg, val);
6447 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
6450 return kvm_skip_emulated_instruction(vcpu);
6453 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
6455 return vcpu->arch.dr6;
6458 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
6462 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
6464 get_debugreg(vcpu->arch.db[0], 0);
6465 get_debugreg(vcpu->arch.db[1], 1);
6466 get_debugreg(vcpu->arch.db[2], 2);
6467 get_debugreg(vcpu->arch.db[3], 3);
6468 get_debugreg(vcpu->arch.dr6, 6);
6469 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
6471 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
6472 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL, CPU_BASED_MOV_DR_EXITING);
6475 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
6477 vmcs_writel(GUEST_DR7, val);
6480 static int handle_cpuid(struct kvm_vcpu *vcpu)
6482 return kvm_emulate_cpuid(vcpu);
6485 static int handle_rdmsr(struct kvm_vcpu *vcpu)
6487 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
6488 struct msr_data msr_info;
6490 msr_info.index = ecx;
6491 msr_info.host_initiated = false;
6492 if (vmx_get_msr(vcpu, &msr_info)) {
6493 trace_kvm_msr_read_ex(ecx);
6494 kvm_inject_gp(vcpu, 0);
6498 trace_kvm_msr_read(ecx, msr_info.data);
6500 /* FIXME: handling of bits 32:63 of rax, rdx */
6501 vcpu->arch.regs[VCPU_REGS_RAX] = msr_info.data & -1u;
6502 vcpu->arch.regs[VCPU_REGS_RDX] = (msr_info.data >> 32) & -1u;
6503 return kvm_skip_emulated_instruction(vcpu);
6506 static int handle_wrmsr(struct kvm_vcpu *vcpu)
6508 struct msr_data msr;
6509 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
6510 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
6511 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
6515 msr.host_initiated = false;
6516 if (kvm_set_msr(vcpu, &msr) != 0) {
6517 trace_kvm_msr_write_ex(ecx, data);
6518 kvm_inject_gp(vcpu, 0);
6522 trace_kvm_msr_write(ecx, data);
6523 return kvm_skip_emulated_instruction(vcpu);
6526 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
6528 kvm_apic_update_ppr(vcpu);
6532 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
6534 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6535 CPU_BASED_VIRTUAL_INTR_PENDING);
6537 kvm_make_request(KVM_REQ_EVENT, vcpu);
6539 ++vcpu->stat.irq_window_exits;
6543 static int handle_halt(struct kvm_vcpu *vcpu)
6545 return kvm_emulate_halt(vcpu);
6548 static int handle_vmcall(struct kvm_vcpu *vcpu)
6550 return kvm_emulate_hypercall(vcpu);
6553 static int handle_invd(struct kvm_vcpu *vcpu)
6555 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6558 static int handle_invlpg(struct kvm_vcpu *vcpu)
6560 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6562 kvm_mmu_invlpg(vcpu, exit_qualification);
6563 return kvm_skip_emulated_instruction(vcpu);
6566 static int handle_rdpmc(struct kvm_vcpu *vcpu)
6570 err = kvm_rdpmc(vcpu);
6571 return kvm_complete_insn_gp(vcpu, err);
6574 static int handle_wbinvd(struct kvm_vcpu *vcpu)
6576 return kvm_emulate_wbinvd(vcpu);
6579 static int handle_xsetbv(struct kvm_vcpu *vcpu)
6581 u64 new_bv = kvm_read_edx_eax(vcpu);
6582 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
6584 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
6585 return kvm_skip_emulated_instruction(vcpu);
6589 static int handle_xsaves(struct kvm_vcpu *vcpu)
6591 kvm_skip_emulated_instruction(vcpu);
6592 WARN(1, "this should never happen\n");
6596 static int handle_xrstors(struct kvm_vcpu *vcpu)
6598 kvm_skip_emulated_instruction(vcpu);
6599 WARN(1, "this should never happen\n");
6603 static int handle_apic_access(struct kvm_vcpu *vcpu)
6605 if (likely(fasteoi)) {
6606 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6607 int access_type, offset;
6609 access_type = exit_qualification & APIC_ACCESS_TYPE;
6610 offset = exit_qualification & APIC_ACCESS_OFFSET;
6612 * Sane guest uses MOV to write EOI, with written value
6613 * not cared. So make a short-circuit here by avoiding
6614 * heavy instruction emulation.
6616 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
6617 (offset == APIC_EOI)) {
6618 kvm_lapic_set_eoi(vcpu);
6619 return kvm_skip_emulated_instruction(vcpu);
6622 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6625 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
6627 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6628 int vector = exit_qualification & 0xff;
6630 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
6631 kvm_apic_set_eoi_accelerated(vcpu, vector);
6635 static int handle_apic_write(struct kvm_vcpu *vcpu)
6637 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6638 u32 offset = exit_qualification & 0xfff;
6640 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
6641 kvm_apic_write_nodecode(vcpu, offset);
6645 static int handle_task_switch(struct kvm_vcpu *vcpu)
6647 struct vcpu_vmx *vmx = to_vmx(vcpu);
6648 unsigned long exit_qualification;
6649 bool has_error_code = false;
6652 int reason, type, idt_v, idt_index;
6654 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
6655 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
6656 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
6658 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6660 reason = (u32)exit_qualification >> 30;
6661 if (reason == TASK_SWITCH_GATE && idt_v) {
6663 case INTR_TYPE_NMI_INTR:
6664 vcpu->arch.nmi_injected = false;
6665 vmx_set_nmi_mask(vcpu, true);
6667 case INTR_TYPE_EXT_INTR:
6668 case INTR_TYPE_SOFT_INTR:
6669 kvm_clear_interrupt_queue(vcpu);
6671 case INTR_TYPE_HARD_EXCEPTION:
6672 if (vmx->idt_vectoring_info &
6673 VECTORING_INFO_DELIVER_CODE_MASK) {
6674 has_error_code = true;
6676 vmcs_read32(IDT_VECTORING_ERROR_CODE);
6679 case INTR_TYPE_SOFT_EXCEPTION:
6680 kvm_clear_exception_queue(vcpu);
6686 tss_selector = exit_qualification;
6688 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
6689 type != INTR_TYPE_EXT_INTR &&
6690 type != INTR_TYPE_NMI_INTR))
6691 skip_emulated_instruction(vcpu);
6693 if (kvm_task_switch(vcpu, tss_selector,
6694 type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
6695 has_error_code, error_code) == EMULATE_FAIL) {
6696 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6697 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6698 vcpu->run->internal.ndata = 0;
6703 * TODO: What about debug traps on tss switch?
6704 * Are we supposed to inject them and update dr6?
6710 static int handle_ept_violation(struct kvm_vcpu *vcpu)
6712 unsigned long exit_qualification;
6716 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6719 * EPT violation happened while executing iret from NMI,
6720 * "blocked by NMI" bit has to be set before next VM entry.
6721 * There are errata that may cause this bit to not be set:
6724 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
6725 cpu_has_virtual_nmis() &&
6726 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
6727 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
6729 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6730 trace_kvm_page_fault(gpa, exit_qualification);
6732 /* Is it a read fault? */
6733 error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
6734 ? PFERR_USER_MASK : 0;
6735 /* Is it a write fault? */
6736 error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
6737 ? PFERR_WRITE_MASK : 0;
6738 /* Is it a fetch fault? */
6739 error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
6740 ? PFERR_FETCH_MASK : 0;
6741 /* ept page table entry is present? */
6742 error_code |= (exit_qualification &
6743 (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
6744 EPT_VIOLATION_EXECUTABLE))
6745 ? PFERR_PRESENT_MASK : 0;
6747 error_code |= (exit_qualification & 0x100) != 0 ?
6748 PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
6750 vcpu->arch.exit_qualification = exit_qualification;
6751 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
6754 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
6760 * A nested guest cannot optimize MMIO vmexits, because we have an
6761 * nGPA here instead of the required GPA.
6763 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6764 if (!is_guest_mode(vcpu) &&
6765 !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
6766 trace_kvm_fast_mmio(gpa);
6768 * Doing kvm_skip_emulated_instruction() depends on undefined
6769 * behavior: Intel's manual doesn't mandate
6770 * VM_EXIT_INSTRUCTION_LEN to be set in VMCS when EPT MISCONFIG
6771 * occurs and while on real hardware it was observed to be set,
6772 * other hypervisors (namely Hyper-V) don't set it, we end up
6773 * advancing IP with some random value. Disable fast mmio when
6774 * running nested and keep it for real hardware in hope that
6775 * VM_EXIT_INSTRUCTION_LEN will always be set correctly.
6777 if (!static_cpu_has(X86_FEATURE_HYPERVISOR))
6778 return kvm_skip_emulated_instruction(vcpu);
6780 return x86_emulate_instruction(vcpu, gpa, EMULTYPE_SKIP,
6781 NULL, 0) == EMULATE_DONE;
6784 ret = kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
6788 /* It is the real ept misconfig */
6791 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
6792 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
6797 static int handle_nmi_window(struct kvm_vcpu *vcpu)
6799 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6800 CPU_BASED_VIRTUAL_NMI_PENDING);
6801 ++vcpu->stat.nmi_window_exits;
6802 kvm_make_request(KVM_REQ_EVENT, vcpu);
6807 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
6809 struct vcpu_vmx *vmx = to_vmx(vcpu);
6810 enum emulation_result err = EMULATE_DONE;
6813 bool intr_window_requested;
6814 unsigned count = 130;
6816 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
6817 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
6819 while (vmx->emulation_required && count-- != 0) {
6820 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
6821 return handle_interrupt_window(&vmx->vcpu);
6823 if (kvm_test_request(KVM_REQ_EVENT, vcpu))
6826 err = emulate_instruction(vcpu, 0);
6828 if (err == EMULATE_USER_EXIT) {
6829 ++vcpu->stat.mmio_exits;
6834 if (err != EMULATE_DONE)
6835 goto emulation_error;
6837 if (vmx->emulation_required && !vmx->rmode.vm86_active &&
6838 vcpu->arch.exception.pending)
6839 goto emulation_error;
6841 if (vcpu->arch.halt_request) {
6842 vcpu->arch.halt_request = 0;
6843 ret = kvm_vcpu_halt(vcpu);
6847 if (signal_pending(current))
6857 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6858 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6859 vcpu->run->internal.ndata = 0;
6863 static int __grow_ple_window(int val)
6865 if (ple_window_grow < 1)
6868 val = min(val, ple_window_actual_max);
6870 if (ple_window_grow < ple_window)
6871 val *= ple_window_grow;
6873 val += ple_window_grow;
6878 static int __shrink_ple_window(int val, int modifier, int minimum)
6883 if (modifier < ple_window)
6888 return max(val, minimum);
6891 static void grow_ple_window(struct kvm_vcpu *vcpu)
6893 struct vcpu_vmx *vmx = to_vmx(vcpu);
6894 int old = vmx->ple_window;
6896 vmx->ple_window = __grow_ple_window(old);
6898 if (vmx->ple_window != old)
6899 vmx->ple_window_dirty = true;
6901 trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old);
6904 static void shrink_ple_window(struct kvm_vcpu *vcpu)
6906 struct vcpu_vmx *vmx = to_vmx(vcpu);
6907 int old = vmx->ple_window;
6909 vmx->ple_window = __shrink_ple_window(old,
6910 ple_window_shrink, ple_window);
6912 if (vmx->ple_window != old)
6913 vmx->ple_window_dirty = true;
6915 trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old);
6919 * ple_window_actual_max is computed to be one grow_ple_window() below
6920 * ple_window_max. (See __grow_ple_window for the reason.)
6921 * This prevents overflows, because ple_window_max is int.
6922 * ple_window_max effectively rounded down to a multiple of ple_window_grow in
6924 * ple_window_max is also prevented from setting vmx->ple_window < ple_window.
6926 static void update_ple_window_actual_max(void)
6928 ple_window_actual_max =
6929 __shrink_ple_window(max(ple_window_max, ple_window),
6930 ple_window_grow, INT_MIN);
6934 * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
6936 static void wakeup_handler(void)
6938 struct kvm_vcpu *vcpu;
6939 int cpu = smp_processor_id();
6941 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6942 list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
6943 blocked_vcpu_list) {
6944 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6946 if (pi_test_on(pi_desc) == 1)
6947 kvm_vcpu_kick(vcpu);
6949 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6952 void vmx_enable_tdp(void)
6954 kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
6955 enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
6956 enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
6957 0ull, VMX_EPT_EXECUTABLE_MASK,
6958 cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
6959 VMX_EPT_RWX_MASK, 0ull);
6961 ept_set_mmio_spte_mask();
6965 static __init int hardware_setup(void)
6969 rdmsrl_safe(MSR_EFER, &host_efer);
6971 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
6972 kvm_define_shared_msr(i, vmx_msr_index[i]);
6974 for (i = 0; i < VMX_BITMAP_NR; i++) {
6975 vmx_bitmap[i] = (unsigned long *)__get_free_page(GFP_KERNEL);
6980 memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
6981 memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
6983 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
6985 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
6987 if (setup_vmcs_config(&vmcs_config) < 0) {
6992 if (boot_cpu_has(X86_FEATURE_NX))
6993 kvm_enable_efer_bits(EFER_NX);
6995 if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
6996 !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
6999 if (!cpu_has_vmx_shadow_vmcs())
7000 enable_shadow_vmcs = 0;
7001 if (enable_shadow_vmcs)
7002 init_vmcs_shadow_fields();
7004 if (!cpu_has_vmx_ept() ||
7005 !cpu_has_vmx_ept_4levels() ||
7006 !cpu_has_vmx_ept_mt_wb()) {
7008 enable_unrestricted_guest = 0;
7009 enable_ept_ad_bits = 0;
7012 if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
7013 enable_ept_ad_bits = 0;
7015 if (!cpu_has_vmx_unrestricted_guest())
7016 enable_unrestricted_guest = 0;
7018 if (!cpu_has_vmx_flexpriority())
7019 flexpriority_enabled = 0;
7022 * set_apic_access_page_addr() is used to reload apic access
7023 * page upon invalidation. No need to do anything if not
7024 * using the APIC_ACCESS_ADDR VMCS field.
7026 if (!flexpriority_enabled)
7027 kvm_x86_ops->set_apic_access_page_addr = NULL;
7029 if (!cpu_has_vmx_tpr_shadow())
7030 kvm_x86_ops->update_cr8_intercept = NULL;
7032 if (enable_ept && !cpu_has_vmx_ept_2m_page())
7033 kvm_disable_largepages();
7035 if (!cpu_has_vmx_ple())
7038 if (!cpu_has_vmx_apicv()) {
7040 kvm_x86_ops->sync_pir_to_irr = NULL;
7043 if (cpu_has_vmx_tsc_scaling()) {
7044 kvm_has_tsc_control = true;
7045 kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
7046 kvm_tsc_scaling_ratio_frac_bits = 48;
7049 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
7056 update_ple_window_actual_max();
7059 * Only enable PML when hardware supports PML feature, and both EPT
7060 * and EPT A/D bit features are enabled -- PML depends on them to work.
7062 if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
7066 kvm_x86_ops->slot_enable_log_dirty = NULL;
7067 kvm_x86_ops->slot_disable_log_dirty = NULL;
7068 kvm_x86_ops->flush_log_dirty = NULL;
7069 kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
7072 if (cpu_has_vmx_preemption_timer() && enable_preemption_timer) {
7075 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
7076 cpu_preemption_timer_multi =
7077 vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
7079 kvm_x86_ops->set_hv_timer = NULL;
7080 kvm_x86_ops->cancel_hv_timer = NULL;
7083 kvm_set_posted_intr_wakeup_handler(wakeup_handler);
7085 kvm_mce_cap_supported |= MCG_LMCE_P;
7087 return alloc_kvm_area();
7090 for (i = 0; i < VMX_BITMAP_NR; i++)
7091 free_page((unsigned long)vmx_bitmap[i]);
7096 static __exit void hardware_unsetup(void)
7100 for (i = 0; i < VMX_BITMAP_NR; i++)
7101 free_page((unsigned long)vmx_bitmap[i]);
7107 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
7108 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
7110 static int handle_pause(struct kvm_vcpu *vcpu)
7113 grow_ple_window(vcpu);
7116 * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
7117 * VM-execution control is ignored if CPL > 0. OTOH, KVM
7118 * never set PAUSE_EXITING and just set PLE if supported,
7119 * so the vcpu must be CPL=0 if it gets a PAUSE exit.
7121 kvm_vcpu_on_spin(vcpu, true);
7122 return kvm_skip_emulated_instruction(vcpu);
7125 static int handle_nop(struct kvm_vcpu *vcpu)
7127 return kvm_skip_emulated_instruction(vcpu);
7130 static int handle_mwait(struct kvm_vcpu *vcpu)
7132 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
7133 return handle_nop(vcpu);
7136 static int handle_invalid_op(struct kvm_vcpu *vcpu)
7138 kvm_queue_exception(vcpu, UD_VECTOR);
7142 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
7147 static int handle_monitor(struct kvm_vcpu *vcpu)
7149 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
7150 return handle_nop(vcpu);
7154 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
7155 * set the success or error code of an emulated VMX instruction, as specified
7156 * by Vol 2B, VMX Instruction Reference, "Conventions".
7158 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
7160 vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
7161 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
7162 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
7165 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
7167 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
7168 & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
7169 X86_EFLAGS_SF | X86_EFLAGS_OF))
7173 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
7174 u32 vm_instruction_error)
7176 if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
7178 * failValid writes the error number to the current VMCS, which
7179 * can't be done there isn't a current VMCS.
7181 nested_vmx_failInvalid(vcpu);
7184 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
7185 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
7186 X86_EFLAGS_SF | X86_EFLAGS_OF))
7188 get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
7190 * We don't need to force a shadow sync because
7191 * VM_INSTRUCTION_ERROR is not shadowed
7195 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
7197 /* TODO: not to reset guest simply here. */
7198 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7199 pr_debug_ratelimited("kvm: nested vmx abort, indicator %d\n", indicator);
7202 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
7204 struct vcpu_vmx *vmx =
7205 container_of(timer, struct vcpu_vmx, nested.preemption_timer);
7207 vmx->nested.preemption_timer_expired = true;
7208 kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
7209 kvm_vcpu_kick(&vmx->vcpu);
7211 return HRTIMER_NORESTART;
7215 * Decode the memory-address operand of a vmx instruction, as recorded on an
7216 * exit caused by such an instruction (run by a guest hypervisor).
7217 * On success, returns 0. When the operand is invalid, returns 1 and throws
7220 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
7221 unsigned long exit_qualification,
7222 u32 vmx_instruction_info, bool wr, gva_t *ret)
7226 struct kvm_segment s;
7229 * According to Vol. 3B, "Information for VM Exits Due to Instruction
7230 * Execution", on an exit, vmx_instruction_info holds most of the
7231 * addressing components of the operand. Only the displacement part
7232 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
7233 * For how an actual address is calculated from all these components,
7234 * refer to Vol. 1, "Operand Addressing".
7236 int scaling = vmx_instruction_info & 3;
7237 int addr_size = (vmx_instruction_info >> 7) & 7;
7238 bool is_reg = vmx_instruction_info & (1u << 10);
7239 int seg_reg = (vmx_instruction_info >> 15) & 7;
7240 int index_reg = (vmx_instruction_info >> 18) & 0xf;
7241 bool index_is_valid = !(vmx_instruction_info & (1u << 22));
7242 int base_reg = (vmx_instruction_info >> 23) & 0xf;
7243 bool base_is_valid = !(vmx_instruction_info & (1u << 27));
7246 kvm_queue_exception(vcpu, UD_VECTOR);
7250 /* Addr = segment_base + offset */
7251 /* offset = base + [index * scale] + displacement */
7252 off = exit_qualification; /* holds the displacement */
7254 off += kvm_register_read(vcpu, base_reg);
7256 off += kvm_register_read(vcpu, index_reg)<<scaling;
7257 vmx_get_segment(vcpu, &s, seg_reg);
7258 *ret = s.base + off;
7260 if (addr_size == 1) /* 32 bit */
7263 /* Checks for #GP/#SS exceptions. */
7265 if (is_long_mode(vcpu)) {
7266 /* Long mode: #GP(0)/#SS(0) if the memory address is in a
7267 * non-canonical form. This is the only check on the memory
7268 * destination for long mode!
7270 exn = is_noncanonical_address(*ret, vcpu);
7271 } else if (is_protmode(vcpu)) {
7272 /* Protected mode: apply checks for segment validity in the
7274 * - segment type check (#GP(0) may be thrown)
7275 * - usability check (#GP(0)/#SS(0))
7276 * - limit check (#GP(0)/#SS(0))
7279 /* #GP(0) if the destination operand is located in a
7280 * read-only data segment or any code segment.
7282 exn = ((s.type & 0xa) == 0 || (s.type & 8));
7284 /* #GP(0) if the source operand is located in an
7285 * execute-only code segment
7287 exn = ((s.type & 0xa) == 8);
7289 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7292 /* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
7294 exn = (s.unusable != 0);
7295 /* Protected mode: #GP(0)/#SS(0) if the memory
7296 * operand is outside the segment limit.
7298 exn = exn || (off + sizeof(u64) > s.limit);
7301 kvm_queue_exception_e(vcpu,
7302 seg_reg == VCPU_SREG_SS ?
7303 SS_VECTOR : GP_VECTOR,
7311 static int nested_vmx_get_vmptr(struct kvm_vcpu *vcpu, gpa_t *vmpointer)
7314 struct x86_exception e;
7316 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7317 vmcs_read32(VMX_INSTRUCTION_INFO), false, &gva))
7320 if (kvm_read_guest_virt(vcpu, gva, vmpointer, sizeof(*vmpointer), &e)) {
7321 kvm_inject_page_fault(vcpu, &e);
7328 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
7330 struct vcpu_vmx *vmx = to_vmx(vcpu);
7331 struct vmcs *shadow_vmcs;
7334 r = alloc_loaded_vmcs(&vmx->nested.vmcs02);
7338 vmx->nested.cached_vmcs12 = kmalloc(VMCS12_SIZE, GFP_KERNEL);
7339 if (!vmx->nested.cached_vmcs12)
7340 goto out_cached_vmcs12;
7342 if (enable_shadow_vmcs) {
7343 shadow_vmcs = alloc_vmcs();
7345 goto out_shadow_vmcs;
7346 /* mark vmcs as shadow */
7347 shadow_vmcs->revision_id |= (1u << 31);
7348 /* init shadow vmcs */
7349 vmcs_clear(shadow_vmcs);
7350 vmx->vmcs01.shadow_vmcs = shadow_vmcs;
7353 hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
7354 HRTIMER_MODE_REL_PINNED);
7355 vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
7357 vmx->nested.vpid02 = allocate_vpid();
7359 vmx->nested.vmxon = true;
7363 kfree(vmx->nested.cached_vmcs12);
7366 free_loaded_vmcs(&vmx->nested.vmcs02);
7373 * Emulate the VMXON instruction.
7374 * Currently, we just remember that VMX is active, and do not save or even
7375 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
7376 * do not currently need to store anything in that guest-allocated memory
7377 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
7378 * argument is different from the VMXON pointer (which the spec says they do).
7380 static int handle_vmon(struct kvm_vcpu *vcpu)
7385 struct vcpu_vmx *vmx = to_vmx(vcpu);
7386 const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
7387 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
7390 * The Intel VMX Instruction Reference lists a bunch of bits that are
7391 * prerequisite to running VMXON, most notably cr4.VMXE must be set to
7392 * 1 (see vmx_set_cr4() for when we allow the guest to set this).
7393 * Otherwise, we should fail with #UD. But most faulting conditions
7394 * have already been checked by hardware, prior to the VM-exit for
7395 * VMXON. We do test guest cr4.VMXE because processor CR4 always has
7396 * that bit set to 1 in non-root mode.
7398 if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE)) {
7399 kvm_queue_exception(vcpu, UD_VECTOR);
7403 /* CPL=0 must be checked manually. */
7404 if (vmx_get_cpl(vcpu)) {
7405 kvm_queue_exception(vcpu, UD_VECTOR);
7409 if (vmx->nested.vmxon) {
7410 nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
7411 return kvm_skip_emulated_instruction(vcpu);
7414 if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
7415 != VMXON_NEEDED_FEATURES) {
7416 kvm_inject_gp(vcpu, 0);
7420 if (nested_vmx_get_vmptr(vcpu, &vmptr))
7425 * The first 4 bytes of VMXON region contain the supported
7426 * VMCS revision identifier
7428 * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
7429 * which replaces physical address width with 32
7431 if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu))) {
7432 nested_vmx_failInvalid(vcpu);
7433 return kvm_skip_emulated_instruction(vcpu);
7436 page = kvm_vcpu_gpa_to_page(vcpu, vmptr);
7437 if (is_error_page(page)) {
7438 nested_vmx_failInvalid(vcpu);
7439 return kvm_skip_emulated_instruction(vcpu);
7441 if (*(u32 *)kmap(page) != VMCS12_REVISION) {
7443 kvm_release_page_clean(page);
7444 nested_vmx_failInvalid(vcpu);
7445 return kvm_skip_emulated_instruction(vcpu);
7448 kvm_release_page_clean(page);
7450 vmx->nested.vmxon_ptr = vmptr;
7451 ret = enter_vmx_operation(vcpu);
7455 nested_vmx_succeed(vcpu);
7456 return kvm_skip_emulated_instruction(vcpu);
7460 * Intel's VMX Instruction Reference specifies a common set of prerequisites
7461 * for running VMX instructions (except VMXON, whose prerequisites are
7462 * slightly different). It also specifies what exception to inject otherwise.
7463 * Note that many of these exceptions have priority over VM exits, so they
7464 * don't have to be checked again here.
7466 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
7468 if (vmx_get_cpl(vcpu)) {
7469 kvm_queue_exception(vcpu, UD_VECTOR);
7473 if (!to_vmx(vcpu)->nested.vmxon) {
7474 kvm_queue_exception(vcpu, UD_VECTOR);
7480 static void vmx_disable_shadow_vmcs(struct vcpu_vmx *vmx)
7482 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL, SECONDARY_EXEC_SHADOW_VMCS);
7483 vmcs_write64(VMCS_LINK_POINTER, -1ull);
7486 static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
7488 if (vmx->nested.current_vmptr == -1ull)
7491 if (enable_shadow_vmcs) {
7492 /* copy to memory all shadowed fields in case
7493 they were modified */
7494 copy_shadow_to_vmcs12(vmx);
7495 vmx->nested.sync_shadow_vmcs = false;
7496 vmx_disable_shadow_vmcs(vmx);
7498 vmx->nested.posted_intr_nv = -1;
7500 /* Flush VMCS12 to guest memory */
7501 kvm_vcpu_write_guest_page(&vmx->vcpu,
7502 vmx->nested.current_vmptr >> PAGE_SHIFT,
7503 vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
7505 vmx->nested.current_vmptr = -1ull;
7509 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
7510 * just stops using VMX.
7512 static void free_nested(struct vcpu_vmx *vmx)
7514 if (!vmx->nested.vmxon)
7517 vmx->nested.vmxon = false;
7518 free_vpid(vmx->nested.vpid02);
7519 vmx->nested.posted_intr_nv = -1;
7520 vmx->nested.current_vmptr = -1ull;
7521 if (enable_shadow_vmcs) {
7522 vmx_disable_shadow_vmcs(vmx);
7523 vmcs_clear(vmx->vmcs01.shadow_vmcs);
7524 free_vmcs(vmx->vmcs01.shadow_vmcs);
7525 vmx->vmcs01.shadow_vmcs = NULL;
7527 kfree(vmx->nested.cached_vmcs12);
7528 /* Unpin physical memory we referred to in the vmcs02 */
7529 if (vmx->nested.apic_access_page) {
7530 kvm_release_page_dirty(vmx->nested.apic_access_page);
7531 vmx->nested.apic_access_page = NULL;
7533 if (vmx->nested.virtual_apic_page) {
7534 kvm_release_page_dirty(vmx->nested.virtual_apic_page);
7535 vmx->nested.virtual_apic_page = NULL;
7537 if (vmx->nested.pi_desc_page) {
7538 kunmap(vmx->nested.pi_desc_page);
7539 kvm_release_page_dirty(vmx->nested.pi_desc_page);
7540 vmx->nested.pi_desc_page = NULL;
7541 vmx->nested.pi_desc = NULL;
7544 free_loaded_vmcs(&vmx->nested.vmcs02);
7547 /* Emulate the VMXOFF instruction */
7548 static int handle_vmoff(struct kvm_vcpu *vcpu)
7550 if (!nested_vmx_check_permission(vcpu))
7552 free_nested(to_vmx(vcpu));
7553 nested_vmx_succeed(vcpu);
7554 return kvm_skip_emulated_instruction(vcpu);
7557 /* Emulate the VMCLEAR instruction */
7558 static int handle_vmclear(struct kvm_vcpu *vcpu)
7560 struct vcpu_vmx *vmx = to_vmx(vcpu);
7564 if (!nested_vmx_check_permission(vcpu))
7567 if (nested_vmx_get_vmptr(vcpu, &vmptr))
7570 if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu))) {
7571 nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
7572 return kvm_skip_emulated_instruction(vcpu);
7575 if (vmptr == vmx->nested.vmxon_ptr) {
7576 nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_VMXON_POINTER);
7577 return kvm_skip_emulated_instruction(vcpu);
7580 if (vmptr == vmx->nested.current_vmptr)
7581 nested_release_vmcs12(vmx);
7583 kvm_vcpu_write_guest(vcpu,
7584 vmptr + offsetof(struct vmcs12, launch_state),
7585 &zero, sizeof(zero));
7587 nested_vmx_succeed(vcpu);
7588 return kvm_skip_emulated_instruction(vcpu);
7591 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
7593 /* Emulate the VMLAUNCH instruction */
7594 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
7596 return nested_vmx_run(vcpu, true);
7599 /* Emulate the VMRESUME instruction */
7600 static int handle_vmresume(struct kvm_vcpu *vcpu)
7603 return nested_vmx_run(vcpu, false);
7607 * Read a vmcs12 field. Since these can have varying lengths and we return
7608 * one type, we chose the biggest type (u64) and zero-extend the return value
7609 * to that size. Note that the caller, handle_vmread, might need to use only
7610 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
7611 * 64-bit fields are to be returned).
7613 static inline int vmcs12_read_any(struct kvm_vcpu *vcpu,
7614 unsigned long field, u64 *ret)
7616 short offset = vmcs_field_to_offset(field);
7622 p = ((char *)(get_vmcs12(vcpu))) + offset;
7624 switch (vmcs_field_type(field)) {
7625 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7626 *ret = *((natural_width *)p);
7628 case VMCS_FIELD_TYPE_U16:
7631 case VMCS_FIELD_TYPE_U32:
7634 case VMCS_FIELD_TYPE_U64:
7644 static inline int vmcs12_write_any(struct kvm_vcpu *vcpu,
7645 unsigned long field, u64 field_value){
7646 short offset = vmcs_field_to_offset(field);
7647 char *p = ((char *) get_vmcs12(vcpu)) + offset;
7651 switch (vmcs_field_type(field)) {
7652 case VMCS_FIELD_TYPE_U16:
7653 *(u16 *)p = field_value;
7655 case VMCS_FIELD_TYPE_U32:
7656 *(u32 *)p = field_value;
7658 case VMCS_FIELD_TYPE_U64:
7659 *(u64 *)p = field_value;
7661 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7662 *(natural_width *)p = field_value;
7671 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
7674 unsigned long field;
7676 struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
7677 const unsigned long *fields = shadow_read_write_fields;
7678 const int num_fields = max_shadow_read_write_fields;
7682 vmcs_load(shadow_vmcs);
7684 for (i = 0; i < num_fields; i++) {
7686 switch (vmcs_field_type(field)) {
7687 case VMCS_FIELD_TYPE_U16:
7688 field_value = vmcs_read16(field);
7690 case VMCS_FIELD_TYPE_U32:
7691 field_value = vmcs_read32(field);
7693 case VMCS_FIELD_TYPE_U64:
7694 field_value = vmcs_read64(field);
7696 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7697 field_value = vmcs_readl(field);
7703 vmcs12_write_any(&vmx->vcpu, field, field_value);
7706 vmcs_clear(shadow_vmcs);
7707 vmcs_load(vmx->loaded_vmcs->vmcs);
7712 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
7714 const unsigned long *fields[] = {
7715 shadow_read_write_fields,
7716 shadow_read_only_fields
7718 const int max_fields[] = {
7719 max_shadow_read_write_fields,
7720 max_shadow_read_only_fields
7723 unsigned long field;
7724 u64 field_value = 0;
7725 struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
7727 vmcs_load(shadow_vmcs);
7729 for (q = 0; q < ARRAY_SIZE(fields); q++) {
7730 for (i = 0; i < max_fields[q]; i++) {
7731 field = fields[q][i];
7732 vmcs12_read_any(&vmx->vcpu, field, &field_value);
7734 switch (vmcs_field_type(field)) {
7735 case VMCS_FIELD_TYPE_U16:
7736 vmcs_write16(field, (u16)field_value);
7738 case VMCS_FIELD_TYPE_U32:
7739 vmcs_write32(field, (u32)field_value);
7741 case VMCS_FIELD_TYPE_U64:
7742 vmcs_write64(field, (u64)field_value);
7744 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7745 vmcs_writel(field, (long)field_value);
7754 vmcs_clear(shadow_vmcs);
7755 vmcs_load(vmx->loaded_vmcs->vmcs);
7759 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
7760 * used before) all generate the same failure when it is missing.
7762 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
7764 struct vcpu_vmx *vmx = to_vmx(vcpu);
7765 if (vmx->nested.current_vmptr == -1ull) {
7766 nested_vmx_failInvalid(vcpu);
7772 static int handle_vmread(struct kvm_vcpu *vcpu)
7774 unsigned long field;
7776 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7777 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7780 if (!nested_vmx_check_permission(vcpu))
7783 if (!nested_vmx_check_vmcs12(vcpu))
7784 return kvm_skip_emulated_instruction(vcpu);
7786 /* Decode instruction info and find the field to read */
7787 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7788 /* Read the field, zero-extended to a u64 field_value */
7789 if (vmcs12_read_any(vcpu, field, &field_value) < 0) {
7790 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7791 return kvm_skip_emulated_instruction(vcpu);
7794 * Now copy part of this value to register or memory, as requested.
7795 * Note that the number of bits actually copied is 32 or 64 depending
7796 * on the guest's mode (32 or 64 bit), not on the given field's length.
7798 if (vmx_instruction_info & (1u << 10)) {
7799 kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
7802 if (get_vmx_mem_address(vcpu, exit_qualification,
7803 vmx_instruction_info, true, &gva))
7805 /* _system ok, nested_vmx_check_permission has verified cpl=0 */
7806 kvm_write_guest_virt_system(vcpu, gva, &field_value,
7807 (is_long_mode(vcpu) ? 8 : 4), NULL);
7810 nested_vmx_succeed(vcpu);
7811 return kvm_skip_emulated_instruction(vcpu);
7815 static int handle_vmwrite(struct kvm_vcpu *vcpu)
7817 unsigned long field;
7819 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7820 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7821 /* The value to write might be 32 or 64 bits, depending on L1's long
7822 * mode, and eventually we need to write that into a field of several
7823 * possible lengths. The code below first zero-extends the value to 64
7824 * bit (field_value), and then copies only the appropriate number of
7825 * bits into the vmcs12 field.
7827 u64 field_value = 0;
7828 struct x86_exception e;
7830 if (!nested_vmx_check_permission(vcpu))
7833 if (!nested_vmx_check_vmcs12(vcpu))
7834 return kvm_skip_emulated_instruction(vcpu);
7836 if (vmx_instruction_info & (1u << 10))
7837 field_value = kvm_register_readl(vcpu,
7838 (((vmx_instruction_info) >> 3) & 0xf));
7840 if (get_vmx_mem_address(vcpu, exit_qualification,
7841 vmx_instruction_info, false, &gva))
7843 if (kvm_read_guest_virt(vcpu, gva, &field_value,
7844 (is_64_bit_mode(vcpu) ? 8 : 4), &e)) {
7845 kvm_inject_page_fault(vcpu, &e);
7851 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7852 if (vmcs_field_readonly(field)) {
7853 nested_vmx_failValid(vcpu,
7854 VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
7855 return kvm_skip_emulated_instruction(vcpu);
7858 if (vmcs12_write_any(vcpu, field, field_value) < 0) {
7859 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7860 return kvm_skip_emulated_instruction(vcpu);
7863 nested_vmx_succeed(vcpu);
7864 return kvm_skip_emulated_instruction(vcpu);
7867 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
7869 vmx->nested.current_vmptr = vmptr;
7870 if (enable_shadow_vmcs) {
7871 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
7872 SECONDARY_EXEC_SHADOW_VMCS);
7873 vmcs_write64(VMCS_LINK_POINTER,
7874 __pa(vmx->vmcs01.shadow_vmcs));
7875 vmx->nested.sync_shadow_vmcs = true;
7879 /* Emulate the VMPTRLD instruction */
7880 static int handle_vmptrld(struct kvm_vcpu *vcpu)
7882 struct vcpu_vmx *vmx = to_vmx(vcpu);
7885 if (!nested_vmx_check_permission(vcpu))
7888 if (nested_vmx_get_vmptr(vcpu, &vmptr))
7891 if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu))) {
7892 nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
7893 return kvm_skip_emulated_instruction(vcpu);
7896 if (vmptr == vmx->nested.vmxon_ptr) {
7897 nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_VMXON_POINTER);
7898 return kvm_skip_emulated_instruction(vcpu);
7901 if (vmx->nested.current_vmptr != vmptr) {
7902 struct vmcs12 *new_vmcs12;
7904 page = kvm_vcpu_gpa_to_page(vcpu, vmptr);
7905 if (is_error_page(page)) {
7906 nested_vmx_failInvalid(vcpu);
7907 return kvm_skip_emulated_instruction(vcpu);
7909 new_vmcs12 = kmap(page);
7910 if (new_vmcs12->revision_id != VMCS12_REVISION) {
7912 kvm_release_page_clean(page);
7913 nested_vmx_failValid(vcpu,
7914 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
7915 return kvm_skip_emulated_instruction(vcpu);
7918 nested_release_vmcs12(vmx);
7920 * Load VMCS12 from guest memory since it is not already
7923 memcpy(vmx->nested.cached_vmcs12, new_vmcs12, VMCS12_SIZE);
7925 kvm_release_page_clean(page);
7927 set_current_vmptr(vmx, vmptr);
7930 nested_vmx_succeed(vcpu);
7931 return kvm_skip_emulated_instruction(vcpu);
7934 /* Emulate the VMPTRST instruction */
7935 static int handle_vmptrst(struct kvm_vcpu *vcpu)
7937 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7938 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7940 struct x86_exception e;
7942 if (!nested_vmx_check_permission(vcpu))
7945 if (get_vmx_mem_address(vcpu, exit_qualification,
7946 vmx_instruction_info, true, &vmcs_gva))
7948 /* *_system ok, nested_vmx_check_permission has verified cpl=0 */
7949 if (kvm_write_guest_virt_system(vcpu, vmcs_gva,
7950 (void *)&to_vmx(vcpu)->nested.current_vmptr,
7952 kvm_inject_page_fault(vcpu, &e);
7955 nested_vmx_succeed(vcpu);
7956 return kvm_skip_emulated_instruction(vcpu);
7959 /* Emulate the INVEPT instruction */
7960 static int handle_invept(struct kvm_vcpu *vcpu)
7962 struct vcpu_vmx *vmx = to_vmx(vcpu);
7963 u32 vmx_instruction_info, types;
7966 struct x86_exception e;
7971 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7972 SECONDARY_EXEC_ENABLE_EPT) ||
7973 !(vmx->nested.nested_vmx_ept_caps & VMX_EPT_INVEPT_BIT)) {
7974 kvm_queue_exception(vcpu, UD_VECTOR);
7978 if (!nested_vmx_check_permission(vcpu))
7981 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7982 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7984 types = (vmx->nested.nested_vmx_ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
7986 if (type >= 32 || !(types & (1 << type))) {
7987 nested_vmx_failValid(vcpu,
7988 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7989 return kvm_skip_emulated_instruction(vcpu);
7992 /* According to the Intel VMX instruction reference, the memory
7993 * operand is read even if it isn't needed (e.g., for type==global)
7995 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7996 vmx_instruction_info, false, &gva))
7998 if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
7999 kvm_inject_page_fault(vcpu, &e);
8004 case VMX_EPT_EXTENT_GLOBAL:
8006 * TODO: track mappings and invalidate
8007 * single context requests appropriately
8009 case VMX_EPT_EXTENT_CONTEXT:
8010 kvm_mmu_sync_roots(vcpu);
8011 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
8012 nested_vmx_succeed(vcpu);
8019 return kvm_skip_emulated_instruction(vcpu);
8022 static int handle_invvpid(struct kvm_vcpu *vcpu)
8024 struct vcpu_vmx *vmx = to_vmx(vcpu);
8025 u32 vmx_instruction_info;
8026 unsigned long type, types;
8028 struct x86_exception e;
8034 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
8035 SECONDARY_EXEC_ENABLE_VPID) ||
8036 !(vmx->nested.nested_vmx_vpid_caps & VMX_VPID_INVVPID_BIT)) {
8037 kvm_queue_exception(vcpu, UD_VECTOR);
8041 if (!nested_vmx_check_permission(vcpu))
8044 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
8045 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
8047 types = (vmx->nested.nested_vmx_vpid_caps &
8048 VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
8050 if (type >= 32 || !(types & (1 << type))) {
8051 nested_vmx_failValid(vcpu,
8052 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
8053 return kvm_skip_emulated_instruction(vcpu);
8056 /* according to the intel vmx instruction reference, the memory
8057 * operand is read even if it isn't needed (e.g., for type==global)
8059 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
8060 vmx_instruction_info, false, &gva))
8062 if (kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e)) {
8063 kvm_inject_page_fault(vcpu, &e);
8066 if (operand.vpid >> 16) {
8067 nested_vmx_failValid(vcpu,
8068 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
8069 return kvm_skip_emulated_instruction(vcpu);
8073 case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
8074 if (is_noncanonical_address(operand.gla, vcpu)) {
8075 nested_vmx_failValid(vcpu,
8076 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
8077 return kvm_skip_emulated_instruction(vcpu);
8080 case VMX_VPID_EXTENT_SINGLE_CONTEXT:
8081 case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
8082 if (!operand.vpid) {
8083 nested_vmx_failValid(vcpu,
8084 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
8085 return kvm_skip_emulated_instruction(vcpu);
8088 case VMX_VPID_EXTENT_ALL_CONTEXT:
8092 return kvm_skip_emulated_instruction(vcpu);
8095 __vmx_flush_tlb(vcpu, vmx->nested.vpid02);
8096 nested_vmx_succeed(vcpu);
8098 return kvm_skip_emulated_instruction(vcpu);
8101 static int handle_pml_full(struct kvm_vcpu *vcpu)
8103 unsigned long exit_qualification;
8105 trace_kvm_pml_full(vcpu->vcpu_id);
8107 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
8110 * PML buffer FULL happened while executing iret from NMI,
8111 * "blocked by NMI" bit has to be set before next VM entry.
8113 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
8114 cpu_has_virtual_nmis() &&
8115 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
8116 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
8117 GUEST_INTR_STATE_NMI);
8120 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
8121 * here.., and there's no userspace involvement needed for PML.
8126 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
8128 kvm_lapic_expired_hv_timer(vcpu);
8132 static bool valid_ept_address(struct kvm_vcpu *vcpu, u64 address)
8134 struct vcpu_vmx *vmx = to_vmx(vcpu);
8135 int maxphyaddr = cpuid_maxphyaddr(vcpu);
8137 /* Check for memory type validity */
8138 switch (address & VMX_EPTP_MT_MASK) {
8139 case VMX_EPTP_MT_UC:
8140 if (!(vmx->nested.nested_vmx_ept_caps & VMX_EPTP_UC_BIT))
8143 case VMX_EPTP_MT_WB:
8144 if (!(vmx->nested.nested_vmx_ept_caps & VMX_EPTP_WB_BIT))
8151 /* only 4 levels page-walk length are valid */
8152 if ((address & VMX_EPTP_PWL_MASK) != VMX_EPTP_PWL_4)
8155 /* Reserved bits should not be set */
8156 if (address >> maxphyaddr || ((address >> 7) & 0x1f))
8159 /* AD, if set, should be supported */
8160 if (address & VMX_EPTP_AD_ENABLE_BIT) {
8161 if (!(vmx->nested.nested_vmx_ept_caps & VMX_EPT_AD_BIT))
8168 static int nested_vmx_eptp_switching(struct kvm_vcpu *vcpu,
8169 struct vmcs12 *vmcs12)
8171 u32 index = vcpu->arch.regs[VCPU_REGS_RCX];
8173 bool accessed_dirty;
8174 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
8176 if (!nested_cpu_has_eptp_switching(vmcs12) ||
8177 !nested_cpu_has_ept(vmcs12))
8180 if (index >= VMFUNC_EPTP_ENTRIES)
8184 if (kvm_vcpu_read_guest_page(vcpu, vmcs12->eptp_list_address >> PAGE_SHIFT,
8185 &address, index * 8, 8))
8188 accessed_dirty = !!(address & VMX_EPTP_AD_ENABLE_BIT);
8191 * If the (L2) guest does a vmfunc to the currently
8192 * active ept pointer, we don't have to do anything else
8194 if (vmcs12->ept_pointer != address) {
8195 if (!valid_ept_address(vcpu, address))
8198 kvm_mmu_unload(vcpu);
8199 mmu->ept_ad = accessed_dirty;
8200 mmu->base_role.ad_disabled = !accessed_dirty;
8201 vmcs12->ept_pointer = address;
8203 * TODO: Check what's the correct approach in case
8204 * mmu reload fails. Currently, we just let the next
8205 * reload potentially fail
8207 kvm_mmu_reload(vcpu);
8213 static int handle_vmfunc(struct kvm_vcpu *vcpu)
8215 struct vcpu_vmx *vmx = to_vmx(vcpu);
8216 struct vmcs12 *vmcs12;
8217 u32 function = vcpu->arch.regs[VCPU_REGS_RAX];
8220 * VMFUNC is only supported for nested guests, but we always enable the
8221 * secondary control for simplicity; for non-nested mode, fake that we
8222 * didn't by injecting #UD.
8224 if (!is_guest_mode(vcpu)) {
8225 kvm_queue_exception(vcpu, UD_VECTOR);
8229 vmcs12 = get_vmcs12(vcpu);
8230 if ((vmcs12->vm_function_control & (1 << function)) == 0)
8235 if (nested_vmx_eptp_switching(vcpu, vmcs12))
8241 return kvm_skip_emulated_instruction(vcpu);
8244 nested_vmx_vmexit(vcpu, vmx->exit_reason,
8245 vmcs_read32(VM_EXIT_INTR_INFO),
8246 vmcs_readl(EXIT_QUALIFICATION));
8251 * The exit handlers return 1 if the exit was handled fully and guest execution
8252 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
8253 * to be done to userspace and return 0.
8255 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
8256 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
8257 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
8258 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
8259 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
8260 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
8261 [EXIT_REASON_CR_ACCESS] = handle_cr,
8262 [EXIT_REASON_DR_ACCESS] = handle_dr,
8263 [EXIT_REASON_CPUID] = handle_cpuid,
8264 [EXIT_REASON_MSR_READ] = handle_rdmsr,
8265 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
8266 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
8267 [EXIT_REASON_HLT] = handle_halt,
8268 [EXIT_REASON_INVD] = handle_invd,
8269 [EXIT_REASON_INVLPG] = handle_invlpg,
8270 [EXIT_REASON_RDPMC] = handle_rdpmc,
8271 [EXIT_REASON_VMCALL] = handle_vmcall,
8272 [EXIT_REASON_VMCLEAR] = handle_vmclear,
8273 [EXIT_REASON_VMLAUNCH] = handle_vmlaunch,
8274 [EXIT_REASON_VMPTRLD] = handle_vmptrld,
8275 [EXIT_REASON_VMPTRST] = handle_vmptrst,
8276 [EXIT_REASON_VMREAD] = handle_vmread,
8277 [EXIT_REASON_VMRESUME] = handle_vmresume,
8278 [EXIT_REASON_VMWRITE] = handle_vmwrite,
8279 [EXIT_REASON_VMOFF] = handle_vmoff,
8280 [EXIT_REASON_VMON] = handle_vmon,
8281 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
8282 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
8283 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
8284 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
8285 [EXIT_REASON_WBINVD] = handle_wbinvd,
8286 [EXIT_REASON_XSETBV] = handle_xsetbv,
8287 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
8288 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
8289 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
8290 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
8291 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
8292 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait,
8293 [EXIT_REASON_MONITOR_TRAP_FLAG] = handle_monitor_trap,
8294 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor,
8295 [EXIT_REASON_INVEPT] = handle_invept,
8296 [EXIT_REASON_INVVPID] = handle_invvpid,
8297 [EXIT_REASON_RDRAND] = handle_invalid_op,
8298 [EXIT_REASON_RDSEED] = handle_invalid_op,
8299 [EXIT_REASON_XSAVES] = handle_xsaves,
8300 [EXIT_REASON_XRSTORS] = handle_xrstors,
8301 [EXIT_REASON_PML_FULL] = handle_pml_full,
8302 [EXIT_REASON_VMFUNC] = handle_vmfunc,
8303 [EXIT_REASON_PREEMPTION_TIMER] = handle_preemption_timer,
8306 static const int kvm_vmx_max_exit_handlers =
8307 ARRAY_SIZE(kvm_vmx_exit_handlers);
8309 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
8310 struct vmcs12 *vmcs12)
8312 unsigned long exit_qualification;
8313 gpa_t bitmap, last_bitmap;
8318 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
8319 return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
8321 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
8323 port = exit_qualification >> 16;
8324 size = (exit_qualification & 7) + 1;
8326 last_bitmap = (gpa_t)-1;
8331 bitmap = vmcs12->io_bitmap_a;
8332 else if (port < 0x10000)
8333 bitmap = vmcs12->io_bitmap_b;
8336 bitmap += (port & 0x7fff) / 8;
8338 if (last_bitmap != bitmap)
8339 if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
8341 if (b & (1 << (port & 7)))
8346 last_bitmap = bitmap;
8353 * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
8354 * rather than handle it ourselves in L0. I.e., check whether L1 expressed
8355 * disinterest in the current event (read or write a specific MSR) by using an
8356 * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
8358 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
8359 struct vmcs12 *vmcs12, u32 exit_reason)
8361 u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
8364 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
8368 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
8369 * for the four combinations of read/write and low/high MSR numbers.
8370 * First we need to figure out which of the four to use:
8372 bitmap = vmcs12->msr_bitmap;
8373 if (exit_reason == EXIT_REASON_MSR_WRITE)
8375 if (msr_index >= 0xc0000000) {
8376 msr_index -= 0xc0000000;
8380 /* Then read the msr_index'th bit from this bitmap: */
8381 if (msr_index < 1024*8) {
8383 if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
8385 return 1 & (b >> (msr_index & 7));
8387 return true; /* let L1 handle the wrong parameter */
8391 * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
8392 * rather than handle it ourselves in L0. I.e., check if L1 wanted to
8393 * intercept (via guest_host_mask etc.) the current event.
8395 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
8396 struct vmcs12 *vmcs12)
8398 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
8399 int cr = exit_qualification & 15;
8403 switch ((exit_qualification >> 4) & 3) {
8404 case 0: /* mov to cr */
8405 reg = (exit_qualification >> 8) & 15;
8406 val = kvm_register_readl(vcpu, reg);
8409 if (vmcs12->cr0_guest_host_mask &
8410 (val ^ vmcs12->cr0_read_shadow))
8414 if ((vmcs12->cr3_target_count >= 1 &&
8415 vmcs12->cr3_target_value0 == val) ||
8416 (vmcs12->cr3_target_count >= 2 &&
8417 vmcs12->cr3_target_value1 == val) ||
8418 (vmcs12->cr3_target_count >= 3 &&
8419 vmcs12->cr3_target_value2 == val) ||
8420 (vmcs12->cr3_target_count >= 4 &&
8421 vmcs12->cr3_target_value3 == val))
8423 if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
8427 if (vmcs12->cr4_guest_host_mask &
8428 (vmcs12->cr4_read_shadow ^ val))
8432 if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
8438 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
8439 (vmcs12->cr0_read_shadow & X86_CR0_TS))
8442 case 1: /* mov from cr */
8445 if (vmcs12->cpu_based_vm_exec_control &
8446 CPU_BASED_CR3_STORE_EXITING)
8450 if (vmcs12->cpu_based_vm_exec_control &
8451 CPU_BASED_CR8_STORE_EXITING)
8458 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
8459 * cr0. Other attempted changes are ignored, with no exit.
8461 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
8462 if (vmcs12->cr0_guest_host_mask & 0xe &
8463 (val ^ vmcs12->cr0_read_shadow))
8465 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
8466 !(vmcs12->cr0_read_shadow & 0x1) &&
8475 * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
8476 * should handle it ourselves in L0 (and then continue L2). Only call this
8477 * when in is_guest_mode (L2).
8479 static bool nested_vmx_exit_reflected(struct kvm_vcpu *vcpu, u32 exit_reason)
8481 u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8482 struct vcpu_vmx *vmx = to_vmx(vcpu);
8483 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8485 if (vmx->nested.nested_run_pending)
8488 if (unlikely(vmx->fail)) {
8489 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
8490 vmcs_read32(VM_INSTRUCTION_ERROR));
8495 * The host physical addresses of some pages of guest memory
8496 * are loaded into the vmcs02 (e.g. vmcs12's Virtual APIC
8497 * Page). The CPU may write to these pages via their host
8498 * physical address while L2 is running, bypassing any
8499 * address-translation-based dirty tracking (e.g. EPT write
8502 * Mark them dirty on every exit from L2 to prevent them from
8503 * getting out of sync with dirty tracking.
8505 nested_mark_vmcs12_pages_dirty(vcpu);
8507 trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
8508 vmcs_readl(EXIT_QUALIFICATION),
8509 vmx->idt_vectoring_info,
8511 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8514 switch (exit_reason) {
8515 case EXIT_REASON_EXCEPTION_NMI:
8516 if (is_nmi(intr_info))
8518 else if (is_page_fault(intr_info))
8519 return !vmx->vcpu.arch.apf.host_apf_reason && enable_ept;
8520 else if (is_no_device(intr_info) &&
8521 !(vmcs12->guest_cr0 & X86_CR0_TS))
8523 else if (is_debug(intr_info) &&
8525 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
8527 else if (is_breakpoint(intr_info) &&
8528 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
8530 return vmcs12->exception_bitmap &
8531 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
8532 case EXIT_REASON_EXTERNAL_INTERRUPT:
8534 case EXIT_REASON_TRIPLE_FAULT:
8536 case EXIT_REASON_PENDING_INTERRUPT:
8537 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
8538 case EXIT_REASON_NMI_WINDOW:
8539 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
8540 case EXIT_REASON_TASK_SWITCH:
8542 case EXIT_REASON_CPUID:
8544 case EXIT_REASON_HLT:
8545 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
8546 case EXIT_REASON_INVD:
8548 case EXIT_REASON_INVLPG:
8549 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
8550 case EXIT_REASON_RDPMC:
8551 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
8552 case EXIT_REASON_RDRAND:
8553 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND);
8554 case EXIT_REASON_RDSEED:
8555 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED);
8556 case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
8557 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
8558 case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
8559 case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
8560 case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
8561 case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
8562 case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
8563 case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
8565 * VMX instructions trap unconditionally. This allows L1 to
8566 * emulate them for its L2 guest, i.e., allows 3-level nesting!
8569 case EXIT_REASON_CR_ACCESS:
8570 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
8571 case EXIT_REASON_DR_ACCESS:
8572 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
8573 case EXIT_REASON_IO_INSTRUCTION:
8574 return nested_vmx_exit_handled_io(vcpu, vmcs12);
8575 case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
8576 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
8577 case EXIT_REASON_MSR_READ:
8578 case EXIT_REASON_MSR_WRITE:
8579 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
8580 case EXIT_REASON_INVALID_STATE:
8582 case EXIT_REASON_MWAIT_INSTRUCTION:
8583 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
8584 case EXIT_REASON_MONITOR_TRAP_FLAG:
8585 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_TRAP_FLAG);
8586 case EXIT_REASON_MONITOR_INSTRUCTION:
8587 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
8588 case EXIT_REASON_PAUSE_INSTRUCTION:
8589 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
8590 nested_cpu_has2(vmcs12,
8591 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
8592 case EXIT_REASON_MCE_DURING_VMENTRY:
8594 case EXIT_REASON_TPR_BELOW_THRESHOLD:
8595 return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
8596 case EXIT_REASON_APIC_ACCESS:
8597 return nested_cpu_has2(vmcs12,
8598 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
8599 case EXIT_REASON_APIC_WRITE:
8600 case EXIT_REASON_EOI_INDUCED:
8601 /* apic_write and eoi_induced should exit unconditionally. */
8603 case EXIT_REASON_EPT_VIOLATION:
8605 * L0 always deals with the EPT violation. If nested EPT is
8606 * used, and the nested mmu code discovers that the address is
8607 * missing in the guest EPT table (EPT12), the EPT violation
8608 * will be injected with nested_ept_inject_page_fault()
8611 case EXIT_REASON_EPT_MISCONFIG:
8613 * L2 never uses directly L1's EPT, but rather L0's own EPT
8614 * table (shadow on EPT) or a merged EPT table that L0 built
8615 * (EPT on EPT). So any problems with the structure of the
8616 * table is L0's fault.
8619 case EXIT_REASON_INVPCID:
8621 nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_INVPCID) &&
8622 nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
8623 case EXIT_REASON_WBINVD:
8624 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
8625 case EXIT_REASON_XSETBV:
8627 case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
8629 * This should never happen, since it is not possible to
8630 * set XSS to a non-zero value---neither in L1 nor in L2.
8631 * If if it were, XSS would have to be checked against
8632 * the XSS exit bitmap in vmcs12.
8634 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
8635 case EXIT_REASON_PREEMPTION_TIMER:
8637 case EXIT_REASON_PML_FULL:
8638 /* We emulate PML support to L1. */
8640 case EXIT_REASON_VMFUNC:
8641 /* VM functions are emulated through L2->L0 vmexits. */
8648 static int nested_vmx_reflect_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason)
8650 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8653 * At this point, the exit interruption info in exit_intr_info
8654 * is only valid for EXCEPTION_NMI exits. For EXTERNAL_INTERRUPT
8655 * we need to query the in-kernel LAPIC.
8657 WARN_ON(exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT);
8658 if ((exit_intr_info &
8659 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
8660 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) {
8661 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8662 vmcs12->vm_exit_intr_error_code =
8663 vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
8666 nested_vmx_vmexit(vcpu, exit_reason, exit_intr_info,
8667 vmcs_readl(EXIT_QUALIFICATION));
8671 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
8673 *info1 = vmcs_readl(EXIT_QUALIFICATION);
8674 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
8677 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
8680 __free_page(vmx->pml_pg);
8685 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
8687 struct vcpu_vmx *vmx = to_vmx(vcpu);
8691 pml_idx = vmcs_read16(GUEST_PML_INDEX);
8693 /* Do nothing if PML buffer is empty */
8694 if (pml_idx == (PML_ENTITY_NUM - 1))
8697 /* PML index always points to next available PML buffer entity */
8698 if (pml_idx >= PML_ENTITY_NUM)
8703 pml_buf = page_address(vmx->pml_pg);
8704 for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
8707 gpa = pml_buf[pml_idx];
8708 WARN_ON(gpa & (PAGE_SIZE - 1));
8709 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
8712 /* reset PML index */
8713 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
8717 * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
8718 * Called before reporting dirty_bitmap to userspace.
8720 static void kvm_flush_pml_buffers(struct kvm *kvm)
8723 struct kvm_vcpu *vcpu;
8725 * We only need to kick vcpu out of guest mode here, as PML buffer
8726 * is flushed at beginning of all VMEXITs, and it's obvious that only
8727 * vcpus running in guest are possible to have unflushed GPAs in PML
8730 kvm_for_each_vcpu(i, vcpu, kvm)
8731 kvm_vcpu_kick(vcpu);
8734 static void vmx_dump_sel(char *name, uint32_t sel)
8736 pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
8737 name, vmcs_read16(sel),
8738 vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
8739 vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
8740 vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
8743 static void vmx_dump_dtsel(char *name, uint32_t limit)
8745 pr_err("%s limit=0x%08x, base=0x%016lx\n",
8746 name, vmcs_read32(limit),
8747 vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
8750 static void dump_vmcs(void)
8752 u32 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
8753 u32 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
8754 u32 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
8755 u32 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
8756 u32 secondary_exec_control = 0;
8757 unsigned long cr4 = vmcs_readl(GUEST_CR4);
8758 u64 efer = vmcs_read64(GUEST_IA32_EFER);
8761 if (cpu_has_secondary_exec_ctrls())
8762 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8764 pr_err("*** Guest State ***\n");
8765 pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8766 vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
8767 vmcs_readl(CR0_GUEST_HOST_MASK));
8768 pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8769 cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
8770 pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
8771 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
8772 (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
8774 pr_err("PDPTR0 = 0x%016llx PDPTR1 = 0x%016llx\n",
8775 vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
8776 pr_err("PDPTR2 = 0x%016llx PDPTR3 = 0x%016llx\n",
8777 vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
8779 pr_err("RSP = 0x%016lx RIP = 0x%016lx\n",
8780 vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
8781 pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n",
8782 vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
8783 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8784 vmcs_readl(GUEST_SYSENTER_ESP),
8785 vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
8786 vmx_dump_sel("CS: ", GUEST_CS_SELECTOR);
8787 vmx_dump_sel("DS: ", GUEST_DS_SELECTOR);
8788 vmx_dump_sel("SS: ", GUEST_SS_SELECTOR);
8789 vmx_dump_sel("ES: ", GUEST_ES_SELECTOR);
8790 vmx_dump_sel("FS: ", GUEST_FS_SELECTOR);
8791 vmx_dump_sel("GS: ", GUEST_GS_SELECTOR);
8792 vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
8793 vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
8794 vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
8795 vmx_dump_sel("TR: ", GUEST_TR_SELECTOR);
8796 if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
8797 (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
8798 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
8799 efer, vmcs_read64(GUEST_IA32_PAT));
8800 pr_err("DebugCtl = 0x%016llx DebugExceptions = 0x%016lx\n",
8801 vmcs_read64(GUEST_IA32_DEBUGCTL),
8802 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
8803 if (vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
8804 pr_err("PerfGlobCtl = 0x%016llx\n",
8805 vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
8806 if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
8807 pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
8808 pr_err("Interruptibility = %08x ActivityState = %08x\n",
8809 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
8810 vmcs_read32(GUEST_ACTIVITY_STATE));
8811 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
8812 pr_err("InterruptStatus = %04x\n",
8813 vmcs_read16(GUEST_INTR_STATUS));
8815 pr_err("*** Host State ***\n");
8816 pr_err("RIP = 0x%016lx RSP = 0x%016lx\n",
8817 vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
8818 pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
8819 vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
8820 vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
8821 vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
8822 vmcs_read16(HOST_TR_SELECTOR));
8823 pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
8824 vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
8825 vmcs_readl(HOST_TR_BASE));
8826 pr_err("GDTBase=%016lx IDTBase=%016lx\n",
8827 vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
8828 pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
8829 vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
8830 vmcs_readl(HOST_CR4));
8831 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8832 vmcs_readl(HOST_IA32_SYSENTER_ESP),
8833 vmcs_read32(HOST_IA32_SYSENTER_CS),
8834 vmcs_readl(HOST_IA32_SYSENTER_EIP));
8835 if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
8836 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
8837 vmcs_read64(HOST_IA32_EFER),
8838 vmcs_read64(HOST_IA32_PAT));
8839 if (vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
8840 pr_err("PerfGlobCtl = 0x%016llx\n",
8841 vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
8843 pr_err("*** Control State ***\n");
8844 pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
8845 pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
8846 pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
8847 pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
8848 vmcs_read32(EXCEPTION_BITMAP),
8849 vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
8850 vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
8851 pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
8852 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8853 vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
8854 vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
8855 pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
8856 vmcs_read32(VM_EXIT_INTR_INFO),
8857 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8858 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
8859 pr_err(" reason=%08x qualification=%016lx\n",
8860 vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
8861 pr_err("IDTVectoring: info=%08x errcode=%08x\n",
8862 vmcs_read32(IDT_VECTORING_INFO_FIELD),
8863 vmcs_read32(IDT_VECTORING_ERROR_CODE));
8864 pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
8865 if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
8866 pr_err("TSC Multiplier = 0x%016llx\n",
8867 vmcs_read64(TSC_MULTIPLIER));
8868 if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW)
8869 pr_err("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
8870 if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
8871 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
8872 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
8873 pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
8874 n = vmcs_read32(CR3_TARGET_COUNT);
8875 for (i = 0; i + 1 < n; i += 4)
8876 pr_err("CR3 target%u=%016lx target%u=%016lx\n",
8877 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
8878 i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
8880 pr_err("CR3 target%u=%016lx\n",
8881 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
8882 if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
8883 pr_err("PLE Gap=%08x Window=%08x\n",
8884 vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
8885 if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
8886 pr_err("Virtual processor ID = 0x%04x\n",
8887 vmcs_read16(VIRTUAL_PROCESSOR_ID));
8891 * The guest has exited. See if we can fix it or if we need userspace
8894 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
8896 struct vcpu_vmx *vmx = to_vmx(vcpu);
8897 u32 exit_reason = vmx->exit_reason;
8898 u32 vectoring_info = vmx->idt_vectoring_info;
8900 trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
8903 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
8904 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
8905 * querying dirty_bitmap, we only need to kick all vcpus out of guest
8906 * mode as if vcpus is in root mode, the PML buffer must has been
8910 vmx_flush_pml_buffer(vcpu);
8912 /* If guest state is invalid, start emulating */
8913 if (vmx->emulation_required)
8914 return handle_invalid_guest_state(vcpu);
8916 if (is_guest_mode(vcpu) && nested_vmx_exit_reflected(vcpu, exit_reason))
8917 return nested_vmx_reflect_vmexit(vcpu, exit_reason);
8919 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
8921 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8922 vcpu->run->fail_entry.hardware_entry_failure_reason
8927 if (unlikely(vmx->fail)) {
8928 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8929 vcpu->run->fail_entry.hardware_entry_failure_reason
8930 = vmcs_read32(VM_INSTRUCTION_ERROR);
8936 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
8937 * delivery event since it indicates guest is accessing MMIO.
8938 * The vm-exit can be triggered again after return to guest that
8939 * will cause infinite loop.
8941 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
8942 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
8943 exit_reason != EXIT_REASON_EPT_VIOLATION &&
8944 exit_reason != EXIT_REASON_PML_FULL &&
8945 exit_reason != EXIT_REASON_TASK_SWITCH)) {
8946 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8947 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
8948 vcpu->run->internal.ndata = 3;
8949 vcpu->run->internal.data[0] = vectoring_info;
8950 vcpu->run->internal.data[1] = exit_reason;
8951 vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
8952 if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
8953 vcpu->run->internal.ndata++;
8954 vcpu->run->internal.data[3] =
8955 vmcs_read64(GUEST_PHYSICAL_ADDRESS);
8960 if (unlikely(!cpu_has_virtual_nmis() &&
8961 vmx->loaded_vmcs->soft_vnmi_blocked)) {
8962 if (vmx_interrupt_allowed(vcpu)) {
8963 vmx->loaded_vmcs->soft_vnmi_blocked = 0;
8964 } else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
8965 vcpu->arch.nmi_pending) {
8967 * This CPU don't support us in finding the end of an
8968 * NMI-blocked window if the guest runs with IRQs
8969 * disabled. So we pull the trigger after 1 s of
8970 * futile waiting, but inform the user about this.
8972 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
8973 "state on VCPU %d after 1 s timeout\n",
8974 __func__, vcpu->vcpu_id);
8975 vmx->loaded_vmcs->soft_vnmi_blocked = 0;
8979 if (exit_reason < kvm_vmx_max_exit_handlers
8980 && kvm_vmx_exit_handlers[exit_reason])
8981 return kvm_vmx_exit_handlers[exit_reason](vcpu);
8983 vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n",
8985 kvm_queue_exception(vcpu, UD_VECTOR);
8990 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
8992 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8994 if (is_guest_mode(vcpu) &&
8995 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
8998 if (irr == -1 || tpr < irr) {
8999 vmcs_write32(TPR_THRESHOLD, 0);
9003 vmcs_write32(TPR_THRESHOLD, irr);
9006 static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
9008 u32 sec_exec_control;
9010 /* Postpone execution until vmcs01 is the current VMCS. */
9011 if (is_guest_mode(vcpu)) {
9012 to_vmx(vcpu)->nested.change_vmcs01_virtual_x2apic_mode = true;
9016 if (!cpu_has_vmx_virtualize_x2apic_mode())
9019 if (!cpu_need_tpr_shadow(vcpu))
9022 sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
9025 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9026 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
9028 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
9029 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9030 vmx_flush_tlb_ept_only(vcpu);
9032 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
9034 vmx_update_msr_bitmap(vcpu);
9037 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
9039 struct vcpu_vmx *vmx = to_vmx(vcpu);
9042 * Currently we do not handle the nested case where L2 has an
9043 * APIC access page of its own; that page is still pinned.
9044 * Hence, we skip the case where the VCPU is in guest mode _and_
9045 * L1 prepared an APIC access page for L2.
9047 * For the case where L1 and L2 share the same APIC access page
9048 * (flexpriority=Y but SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES clear
9049 * in the vmcs12), this function will only update either the vmcs01
9050 * or the vmcs02. If the former, the vmcs02 will be updated by
9051 * prepare_vmcs02. If the latter, the vmcs01 will be updated in
9052 * the next L2->L1 exit.
9054 if (!is_guest_mode(vcpu) ||
9055 !nested_cpu_has2(get_vmcs12(&vmx->vcpu),
9056 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
9057 vmcs_write64(APIC_ACCESS_ADDR, hpa);
9058 vmx_flush_tlb_ept_only(vcpu);
9062 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
9070 status = vmcs_read16(GUEST_INTR_STATUS);
9072 if (max_isr != old) {
9074 status |= max_isr << 8;
9075 vmcs_write16(GUEST_INTR_STATUS, status);
9079 static void vmx_set_rvi(int vector)
9087 status = vmcs_read16(GUEST_INTR_STATUS);
9088 old = (u8)status & 0xff;
9089 if ((u8)vector != old) {
9091 status |= (u8)vector;
9092 vmcs_write16(GUEST_INTR_STATUS, status);
9096 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
9098 if (!is_guest_mode(vcpu)) {
9099 vmx_set_rvi(max_irr);
9107 * In guest mode. If a vmexit is needed, vmx_check_nested_events
9110 if (nested_exit_on_intr(vcpu))
9114 * Else, fall back to pre-APICv interrupt injection since L2
9115 * is run without virtual interrupt delivery.
9117 if (!kvm_event_needs_reinjection(vcpu) &&
9118 vmx_interrupt_allowed(vcpu)) {
9119 kvm_queue_interrupt(vcpu, max_irr, false);
9120 vmx_inject_irq(vcpu);
9124 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
9126 struct vcpu_vmx *vmx = to_vmx(vcpu);
9129 WARN_ON(!vcpu->arch.apicv_active);
9130 if (pi_test_on(&vmx->pi_desc)) {
9131 pi_clear_on(&vmx->pi_desc);
9133 * IOMMU can write to PIR.ON, so the barrier matters even on UP.
9134 * But on x86 this is just a compiler barrier anyway.
9136 smp_mb__after_atomic();
9137 max_irr = kvm_apic_update_irr(vcpu, vmx->pi_desc.pir);
9139 max_irr = kvm_lapic_find_highest_irr(vcpu);
9141 vmx_hwapic_irr_update(vcpu, max_irr);
9145 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
9147 if (!kvm_vcpu_apicv_active(vcpu))
9150 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
9151 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
9152 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
9153 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
9156 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
9158 struct vcpu_vmx *vmx = to_vmx(vcpu);
9160 pi_clear_on(&vmx->pi_desc);
9161 memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
9164 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
9166 u32 exit_intr_info = 0;
9167 u16 basic_exit_reason = (u16)vmx->exit_reason;
9169 if (!(basic_exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
9170 || basic_exit_reason == EXIT_REASON_EXCEPTION_NMI))
9173 if (!(vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
9174 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
9175 vmx->exit_intr_info = exit_intr_info;
9177 /* if exit due to PF check for async PF */
9178 if (is_page_fault(exit_intr_info))
9179 vmx->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
9181 /* Handle machine checks before interrupts are enabled */
9182 if (basic_exit_reason == EXIT_REASON_MCE_DURING_VMENTRY ||
9183 is_machine_check(exit_intr_info))
9184 kvm_machine_check();
9186 /* We need to handle NMIs before interrupts are enabled */
9187 if (is_nmi(exit_intr_info)) {
9188 kvm_before_handle_nmi(&vmx->vcpu);
9190 kvm_after_handle_nmi(&vmx->vcpu);
9194 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
9196 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
9198 if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
9199 == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
9200 unsigned int vector;
9201 unsigned long entry;
9203 struct vcpu_vmx *vmx = to_vmx(vcpu);
9204 #ifdef CONFIG_X86_64
9208 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
9209 desc = (gate_desc *)vmx->host_idt_base + vector;
9210 entry = gate_offset(desc);
9212 #ifdef CONFIG_X86_64
9213 "mov %%" _ASM_SP ", %[sp]\n\t"
9214 "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
9219 __ASM_SIZE(push) " $%c[cs]\n\t"
9222 #ifdef CONFIG_X86_64
9227 THUNK_TARGET(entry),
9228 [ss]"i"(__KERNEL_DS),
9229 [cs]"i"(__KERNEL_CS)
9233 STACK_FRAME_NON_STANDARD(vmx_handle_external_intr);
9235 static bool vmx_has_emulated_msr(int index)
9238 case MSR_IA32_SMBASE:
9240 * We cannot do SMM unless we can run the guest in big
9243 return enable_unrestricted_guest || emulate_invalid_guest_state;
9244 case MSR_AMD64_VIRT_SPEC_CTRL:
9245 /* This is AMD only. */
9252 static bool vmx_mpx_supported(void)
9254 return (vmcs_config.vmexit_ctrl & VM_EXIT_CLEAR_BNDCFGS) &&
9255 (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_BNDCFGS);
9258 static bool vmx_xsaves_supported(void)
9260 return vmcs_config.cpu_based_2nd_exec_ctrl &
9261 SECONDARY_EXEC_XSAVES;
9264 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
9269 bool idtv_info_valid;
9271 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
9273 if (cpu_has_virtual_nmis()) {
9274 if (vmx->loaded_vmcs->nmi_known_unmasked)
9277 * Can't use vmx->exit_intr_info since we're not sure what
9278 * the exit reason is.
9280 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
9281 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
9282 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
9284 * SDM 3: 27.7.1.2 (September 2008)
9285 * Re-set bit "block by NMI" before VM entry if vmexit caused by
9286 * a guest IRET fault.
9287 * SDM 3: 23.2.2 (September 2008)
9288 * Bit 12 is undefined in any of the following cases:
9289 * If the VM exit sets the valid bit in the IDT-vectoring
9290 * information field.
9291 * If the VM exit is due to a double fault.
9293 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
9294 vector != DF_VECTOR && !idtv_info_valid)
9295 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
9296 GUEST_INTR_STATE_NMI);
9298 vmx->loaded_vmcs->nmi_known_unmasked =
9299 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
9300 & GUEST_INTR_STATE_NMI);
9301 } else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
9302 vmx->loaded_vmcs->vnmi_blocked_time +=
9303 ktime_to_ns(ktime_sub(ktime_get(),
9304 vmx->loaded_vmcs->entry_time));
9307 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
9308 u32 idt_vectoring_info,
9309 int instr_len_field,
9310 int error_code_field)
9314 bool idtv_info_valid;
9316 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
9318 vcpu->arch.nmi_injected = false;
9319 kvm_clear_exception_queue(vcpu);
9320 kvm_clear_interrupt_queue(vcpu);
9322 if (!idtv_info_valid)
9325 kvm_make_request(KVM_REQ_EVENT, vcpu);
9327 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
9328 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
9331 case INTR_TYPE_NMI_INTR:
9332 vcpu->arch.nmi_injected = true;
9334 * SDM 3: 27.7.1.2 (September 2008)
9335 * Clear bit "block by NMI" before VM entry if a NMI
9338 vmx_set_nmi_mask(vcpu, false);
9340 case INTR_TYPE_SOFT_EXCEPTION:
9341 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
9343 case INTR_TYPE_HARD_EXCEPTION:
9344 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
9345 u32 err = vmcs_read32(error_code_field);
9346 kvm_requeue_exception_e(vcpu, vector, err);
9348 kvm_requeue_exception(vcpu, vector);
9350 case INTR_TYPE_SOFT_INTR:
9351 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
9353 case INTR_TYPE_EXT_INTR:
9354 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
9361 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
9363 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
9364 VM_EXIT_INSTRUCTION_LEN,
9365 IDT_VECTORING_ERROR_CODE);
9368 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
9370 __vmx_complete_interrupts(vcpu,
9371 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
9372 VM_ENTRY_INSTRUCTION_LEN,
9373 VM_ENTRY_EXCEPTION_ERROR_CODE);
9375 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
9378 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
9381 struct perf_guest_switch_msr *msrs;
9383 msrs = perf_guest_get_msrs(&nr_msrs);
9388 for (i = 0; i < nr_msrs; i++)
9389 if (msrs[i].host == msrs[i].guest)
9390 clear_atomic_switch_msr(vmx, msrs[i].msr);
9392 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
9396 static void vmx_arm_hv_timer(struct kvm_vcpu *vcpu)
9398 struct vcpu_vmx *vmx = to_vmx(vcpu);
9402 if (vmx->hv_deadline_tsc == -1)
9406 if (vmx->hv_deadline_tsc > tscl)
9407 /* sure to be 32 bit only because checked on set_hv_timer */
9408 delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
9409 cpu_preemption_timer_multi);
9413 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
9416 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
9418 struct vcpu_vmx *vmx = to_vmx(vcpu);
9419 unsigned long debugctlmsr, cr3, cr4;
9421 /* Record the guest's net vcpu time for enforced NMI injections. */
9422 if (unlikely(!cpu_has_virtual_nmis() &&
9423 vmx->loaded_vmcs->soft_vnmi_blocked))
9424 vmx->loaded_vmcs->entry_time = ktime_get();
9426 /* Don't enter VMX if guest state is invalid, let the exit handler
9427 start emulation until we arrive back to a valid state */
9428 if (vmx->emulation_required)
9431 if (vmx->ple_window_dirty) {
9432 vmx->ple_window_dirty = false;
9433 vmcs_write32(PLE_WINDOW, vmx->ple_window);
9436 if (vmx->nested.sync_shadow_vmcs) {
9437 copy_vmcs12_to_shadow(vmx);
9438 vmx->nested.sync_shadow_vmcs = false;
9441 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
9442 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
9443 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
9444 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
9446 cr3 = __get_current_cr3_fast();
9447 if (unlikely(cr3 != vmx->loaded_vmcs->vmcs_host_cr3)) {
9448 vmcs_writel(HOST_CR3, cr3);
9449 vmx->loaded_vmcs->vmcs_host_cr3 = cr3;
9452 cr4 = cr4_read_shadow();
9453 if (unlikely(cr4 != vmx->loaded_vmcs->vmcs_host_cr4)) {
9454 vmcs_writel(HOST_CR4, cr4);
9455 vmx->loaded_vmcs->vmcs_host_cr4 = cr4;
9458 /* When single-stepping over STI and MOV SS, we must clear the
9459 * corresponding interruptibility bits in the guest state. Otherwise
9460 * vmentry fails as it then expects bit 14 (BS) in pending debug
9461 * exceptions being set, but that's not correct for the guest debugging
9463 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9464 vmx_set_interrupt_shadow(vcpu, 0);
9466 if (static_cpu_has(X86_FEATURE_PKU) &&
9467 kvm_read_cr4_bits(vcpu, X86_CR4_PKE) &&
9468 vcpu->arch.pkru != vmx->host_pkru)
9469 __write_pkru(vcpu->arch.pkru);
9471 atomic_switch_perf_msrs(vmx);
9472 debugctlmsr = get_debugctlmsr();
9474 vmx_arm_hv_timer(vcpu);
9477 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
9478 * it's non-zero. Since vmentry is serialising on affected CPUs, there
9479 * is no need to worry about the conditional branch over the wrmsr
9480 * being speculatively taken.
9482 x86_spec_ctrl_set_guest(vmx->spec_ctrl, 0);
9484 vmx->__launched = vmx->loaded_vmcs->launched;
9487 /* Store host registers */
9488 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
9489 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
9490 "push %%" _ASM_CX " \n\t"
9491 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
9493 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
9494 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
9496 /* Reload cr2 if changed */
9497 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
9498 "mov %%cr2, %%" _ASM_DX " \n\t"
9499 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
9501 "mov %%" _ASM_AX", %%cr2 \n\t"
9503 /* Check if vmlaunch of vmresume is needed */
9504 "cmpl $0, %c[launched](%0) \n\t"
9505 /* Load guest registers. Don't clobber flags. */
9506 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
9507 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
9508 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
9509 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
9510 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
9511 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
9512 #ifdef CONFIG_X86_64
9513 "mov %c[r8](%0), %%r8 \n\t"
9514 "mov %c[r9](%0), %%r9 \n\t"
9515 "mov %c[r10](%0), %%r10 \n\t"
9516 "mov %c[r11](%0), %%r11 \n\t"
9517 "mov %c[r12](%0), %%r12 \n\t"
9518 "mov %c[r13](%0), %%r13 \n\t"
9519 "mov %c[r14](%0), %%r14 \n\t"
9520 "mov %c[r15](%0), %%r15 \n\t"
9522 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
9524 /* Enter guest mode */
9526 __ex(ASM_VMX_VMLAUNCH) "\n\t"
9528 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
9530 /* Save guest registers, load host registers, keep flags */
9531 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
9533 "setbe %c[fail](%0)\n\t"
9534 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
9535 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
9536 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
9537 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
9538 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
9539 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
9540 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
9541 #ifdef CONFIG_X86_64
9542 "mov %%r8, %c[r8](%0) \n\t"
9543 "mov %%r9, %c[r9](%0) \n\t"
9544 "mov %%r10, %c[r10](%0) \n\t"
9545 "mov %%r11, %c[r11](%0) \n\t"
9546 "mov %%r12, %c[r12](%0) \n\t"
9547 "mov %%r13, %c[r13](%0) \n\t"
9548 "mov %%r14, %c[r14](%0) \n\t"
9549 "mov %%r15, %c[r15](%0) \n\t"
9550 "xor %%r8d, %%r8d \n\t"
9551 "xor %%r9d, %%r9d \n\t"
9552 "xor %%r10d, %%r10d \n\t"
9553 "xor %%r11d, %%r11d \n\t"
9554 "xor %%r12d, %%r12d \n\t"
9555 "xor %%r13d, %%r13d \n\t"
9556 "xor %%r14d, %%r14d \n\t"
9557 "xor %%r15d, %%r15d \n\t"
9559 "mov %%cr2, %%" _ASM_AX " \n\t"
9560 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
9562 "xor %%eax, %%eax \n\t"
9563 "xor %%ebx, %%ebx \n\t"
9564 "xor %%esi, %%esi \n\t"
9565 "xor %%edi, %%edi \n\t"
9566 "pop %%" _ASM_BP "; pop %%" _ASM_DX " \n\t"
9567 ".pushsection .rodata \n\t"
9568 ".global vmx_return \n\t"
9569 "vmx_return: " _ASM_PTR " 2b \n\t"
9571 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
9572 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
9573 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
9574 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
9575 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
9576 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
9577 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
9578 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
9579 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
9580 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
9581 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
9582 #ifdef CONFIG_X86_64
9583 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
9584 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
9585 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
9586 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
9587 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
9588 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
9589 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
9590 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
9592 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
9593 [wordsize]"i"(sizeof(ulong))
9595 #ifdef CONFIG_X86_64
9596 , "rax", "rbx", "rdi", "rsi"
9597 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
9599 , "eax", "ebx", "edi", "esi"
9604 * We do not use IBRS in the kernel. If this vCPU has used the
9605 * SPEC_CTRL MSR it may have left it on; save the value and
9606 * turn it off. This is much more efficient than blindly adding
9607 * it to the atomic save/restore list. Especially as the former
9608 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
9610 * For non-nested case:
9611 * If the L01 MSR bitmap does not intercept the MSR, then we need to
9615 * If the L02 MSR bitmap does not intercept the MSR, then we need to
9618 if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
9619 vmx->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
9621 x86_spec_ctrl_restore_host(vmx->spec_ctrl, 0);
9623 /* Eliminate branch target predictions from guest mode */
9626 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
9628 update_debugctlmsr(debugctlmsr);
9630 #ifndef CONFIG_X86_64
9632 * The sysexit path does not restore ds/es, so we must set them to
9633 * a reasonable value ourselves.
9635 * We can't defer this to vmx_load_host_state() since that function
9636 * may be executed in interrupt context, which saves and restore segments
9637 * around it, nullifying its effect.
9639 loadsegment(ds, __USER_DS);
9640 loadsegment(es, __USER_DS);
9643 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
9644 | (1 << VCPU_EXREG_RFLAGS)
9645 | (1 << VCPU_EXREG_PDPTR)
9646 | (1 << VCPU_EXREG_SEGMENTS)
9647 | (1 << VCPU_EXREG_CR3));
9648 vcpu->arch.regs_dirty = 0;
9651 * eager fpu is enabled if PKEY is supported and CR4 is switched
9652 * back on host, so it is safe to read guest PKRU from current
9655 if (static_cpu_has(X86_FEATURE_PKU) &&
9656 kvm_read_cr4_bits(vcpu, X86_CR4_PKE)) {
9657 vcpu->arch.pkru = __read_pkru();
9658 if (vcpu->arch.pkru != vmx->host_pkru)
9659 __write_pkru(vmx->host_pkru);
9663 * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
9664 * we did not inject a still-pending event to L1 now because of
9665 * nested_run_pending, we need to re-enable this bit.
9667 if (vmx->nested.nested_run_pending)
9668 kvm_make_request(KVM_REQ_EVENT, vcpu);
9670 vmx->nested.nested_run_pending = 0;
9671 vmx->idt_vectoring_info = 0;
9673 vmx->exit_reason = vmx->fail ? 0xdead : vmcs_read32(VM_EXIT_REASON);
9674 if (vmx->fail || (vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
9677 vmx->loaded_vmcs->launched = 1;
9678 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
9680 vmx_complete_atomic_exit(vmx);
9681 vmx_recover_nmi_blocking(vmx);
9682 vmx_complete_interrupts(vmx);
9684 STACK_FRAME_NON_STANDARD(vmx_vcpu_run);
9686 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
9688 struct vcpu_vmx *vmx = to_vmx(vcpu);
9691 if (vmx->loaded_vmcs == vmcs)
9695 vmx->loaded_vmcs = vmcs;
9697 vmx_vcpu_load(vcpu, cpu);
9703 * Ensure that the current vmcs of the logical processor is the
9704 * vmcs01 of the vcpu before calling free_nested().
9706 static void vmx_free_vcpu_nested(struct kvm_vcpu *vcpu)
9708 struct vcpu_vmx *vmx = to_vmx(vcpu);
9711 r = vcpu_load(vcpu);
9713 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
9718 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
9720 struct vcpu_vmx *vmx = to_vmx(vcpu);
9723 vmx_destroy_pml_buffer(vmx);
9724 free_vpid(vmx->vpid);
9725 leave_guest_mode(vcpu);
9726 vmx_free_vcpu_nested(vcpu);
9727 free_loaded_vmcs(vmx->loaded_vmcs);
9728 kfree(vmx->guest_msrs);
9729 kvm_vcpu_uninit(vcpu);
9730 kmem_cache_free(kvm_vcpu_cache, vmx);
9733 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
9736 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
9737 unsigned long *msr_bitmap;
9741 return ERR_PTR(-ENOMEM);
9743 vmx->vpid = allocate_vpid();
9745 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
9752 * If PML is turned on, failure on enabling PML just results in failure
9753 * of creating the vcpu, therefore we can simplify PML logic (by
9754 * avoiding dealing with cases, such as enabling PML partially on vcpus
9755 * for the guest, etc.
9758 vmx->pml_pg = alloc_page(GFP_KERNEL | __GFP_ZERO);
9763 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
9764 BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
9767 if (!vmx->guest_msrs)
9770 err = alloc_loaded_vmcs(&vmx->vmcs01);
9774 msr_bitmap = vmx->vmcs01.msr_bitmap;
9775 vmx_disable_intercept_for_msr(msr_bitmap, MSR_FS_BASE, MSR_TYPE_RW);
9776 vmx_disable_intercept_for_msr(msr_bitmap, MSR_GS_BASE, MSR_TYPE_RW);
9777 vmx_disable_intercept_for_msr(msr_bitmap, MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
9778 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW);
9779 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW);
9780 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW);
9781 vmx->msr_bitmap_mode = 0;
9783 vmx->loaded_vmcs = &vmx->vmcs01;
9785 vmx_vcpu_load(&vmx->vcpu, cpu);
9786 vmx->vcpu.cpu = cpu;
9787 err = vmx_vcpu_setup(vmx);
9788 vmx_vcpu_put(&vmx->vcpu);
9792 if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9793 err = alloc_apic_access_page(kvm);
9799 if (!kvm->arch.ept_identity_map_addr)
9800 kvm->arch.ept_identity_map_addr =
9801 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
9802 err = init_rmode_identity_map(kvm);
9808 nested_vmx_setup_ctls_msrs(vmx);
9810 vmx->nested.posted_intr_nv = -1;
9811 vmx->nested.current_vmptr = -1ull;
9813 vmx->msr_ia32_feature_control_valid_bits = FEATURE_CONTROL_LOCKED;
9816 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
9817 * or POSTED_INTR_WAKEUP_VECTOR.
9819 vmx->pi_desc.nv = POSTED_INTR_VECTOR;
9820 vmx->pi_desc.sn = 1;
9825 free_loaded_vmcs(vmx->loaded_vmcs);
9827 kfree(vmx->guest_msrs);
9829 vmx_destroy_pml_buffer(vmx);
9831 kvm_vcpu_uninit(&vmx->vcpu);
9833 free_vpid(vmx->vpid);
9834 kmem_cache_free(kvm_vcpu_cache, vmx);
9835 return ERR_PTR(err);
9838 static void __init vmx_check_processor_compat(void *rtn)
9840 struct vmcs_config vmcs_conf;
9843 if (setup_vmcs_config(&vmcs_conf) < 0)
9845 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
9846 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
9847 smp_processor_id());
9852 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
9857 /* For VT-d and EPT combination
9858 * 1. MMIO: always map as UC
9860 * a. VT-d without snooping control feature: can't guarantee the
9861 * result, try to trust guest.
9862 * b. VT-d with snooping control feature: snooping control feature of
9863 * VT-d engine can guarantee the cache correctness. Just set it
9864 * to WB to keep consistent with host. So the same as item 3.
9865 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
9866 * consistent with host MTRR
9869 cache = MTRR_TYPE_UNCACHABLE;
9873 if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
9874 ipat = VMX_EPT_IPAT_BIT;
9875 cache = MTRR_TYPE_WRBACK;
9879 if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
9880 ipat = VMX_EPT_IPAT_BIT;
9881 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
9882 cache = MTRR_TYPE_WRBACK;
9884 cache = MTRR_TYPE_UNCACHABLE;
9888 cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
9891 return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
9894 static int vmx_get_lpage_level(void)
9896 if (enable_ept && !cpu_has_vmx_ept_1g_page())
9897 return PT_DIRECTORY_LEVEL;
9899 /* For shadow and EPT supported 1GB page */
9900 return PT_PDPE_LEVEL;
9903 static void vmcs_set_secondary_exec_control(u32 new_ctl)
9906 * These bits in the secondary execution controls field
9907 * are dynamic, the others are mostly based on the hypervisor
9908 * architecture and the guest's CPUID. Do not touch the
9912 SECONDARY_EXEC_SHADOW_VMCS |
9913 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
9914 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9916 u32 cur_ctl = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
9918 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
9919 (new_ctl & ~mask) | (cur_ctl & mask));
9923 * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
9924 * (indicating "allowed-1") if they are supported in the guest's CPUID.
9926 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
9928 struct vcpu_vmx *vmx = to_vmx(vcpu);
9929 struct kvm_cpuid_entry2 *entry;
9931 vmx->nested.nested_vmx_cr0_fixed1 = 0xffffffff;
9932 vmx->nested.nested_vmx_cr4_fixed1 = X86_CR4_PCE;
9934 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do { \
9935 if (entry && (entry->_reg & (_cpuid_mask))) \
9936 vmx->nested.nested_vmx_cr4_fixed1 |= (_cr4_mask); \
9939 entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
9940 cr4_fixed1_update(X86_CR4_VME, edx, bit(X86_FEATURE_VME));
9941 cr4_fixed1_update(X86_CR4_PVI, edx, bit(X86_FEATURE_VME));
9942 cr4_fixed1_update(X86_CR4_TSD, edx, bit(X86_FEATURE_TSC));
9943 cr4_fixed1_update(X86_CR4_DE, edx, bit(X86_FEATURE_DE));
9944 cr4_fixed1_update(X86_CR4_PSE, edx, bit(X86_FEATURE_PSE));
9945 cr4_fixed1_update(X86_CR4_PAE, edx, bit(X86_FEATURE_PAE));
9946 cr4_fixed1_update(X86_CR4_MCE, edx, bit(X86_FEATURE_MCE));
9947 cr4_fixed1_update(X86_CR4_PGE, edx, bit(X86_FEATURE_PGE));
9948 cr4_fixed1_update(X86_CR4_OSFXSR, edx, bit(X86_FEATURE_FXSR));
9949 cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, bit(X86_FEATURE_XMM));
9950 cr4_fixed1_update(X86_CR4_VMXE, ecx, bit(X86_FEATURE_VMX));
9951 cr4_fixed1_update(X86_CR4_SMXE, ecx, bit(X86_FEATURE_SMX));
9952 cr4_fixed1_update(X86_CR4_PCIDE, ecx, bit(X86_FEATURE_PCID));
9953 cr4_fixed1_update(X86_CR4_OSXSAVE, ecx, bit(X86_FEATURE_XSAVE));
9955 entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
9956 cr4_fixed1_update(X86_CR4_FSGSBASE, ebx, bit(X86_FEATURE_FSGSBASE));
9957 cr4_fixed1_update(X86_CR4_SMEP, ebx, bit(X86_FEATURE_SMEP));
9958 cr4_fixed1_update(X86_CR4_SMAP, ebx, bit(X86_FEATURE_SMAP));
9959 cr4_fixed1_update(X86_CR4_PKE, ecx, bit(X86_FEATURE_PKU));
9960 /* TODO: Use X86_CR4_UMIP and X86_FEATURE_UMIP macros */
9961 cr4_fixed1_update(bit(11), ecx, bit(2));
9963 #undef cr4_fixed1_update
9966 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
9968 struct vcpu_vmx *vmx = to_vmx(vcpu);
9970 if (cpu_has_secondary_exec_ctrls()) {
9971 vmx_compute_secondary_exec_control(vmx);
9972 vmcs_set_secondary_exec_control(vmx->secondary_exec_control);
9975 if (nested_vmx_allowed(vcpu))
9976 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
9977 FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
9979 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
9980 ~FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
9982 if (nested_vmx_allowed(vcpu))
9983 nested_vmx_cr_fixed1_bits_update(vcpu);
9986 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
9988 if (func == 1 && nested)
9989 entry->ecx |= bit(X86_FEATURE_VMX);
9992 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
9993 struct x86_exception *fault)
9995 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9996 struct vcpu_vmx *vmx = to_vmx(vcpu);
9998 unsigned long exit_qualification = vcpu->arch.exit_qualification;
10000 if (vmx->nested.pml_full) {
10001 exit_reason = EXIT_REASON_PML_FULL;
10002 vmx->nested.pml_full = false;
10003 exit_qualification &= INTR_INFO_UNBLOCK_NMI;
10004 } else if (fault->error_code & PFERR_RSVD_MASK)
10005 exit_reason = EXIT_REASON_EPT_MISCONFIG;
10007 exit_reason = EXIT_REASON_EPT_VIOLATION;
10009 nested_vmx_vmexit(vcpu, exit_reason, 0, exit_qualification);
10010 vmcs12->guest_physical_address = fault->address;
10013 static bool nested_ept_ad_enabled(struct kvm_vcpu *vcpu)
10015 return nested_ept_get_cr3(vcpu) & VMX_EPTP_AD_ENABLE_BIT;
10018 /* Callbacks for nested_ept_init_mmu_context: */
10020 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
10022 /* return the page table to be shadowed - in our case, EPT12 */
10023 return get_vmcs12(vcpu)->ept_pointer;
10026 static int nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
10028 WARN_ON(mmu_is_nested(vcpu));
10029 if (!valid_ept_address(vcpu, nested_ept_get_cr3(vcpu)))
10032 kvm_mmu_unload(vcpu);
10033 kvm_init_shadow_ept_mmu(vcpu,
10034 to_vmx(vcpu)->nested.nested_vmx_ept_caps &
10035 VMX_EPT_EXECUTE_ONLY_BIT,
10036 nested_ept_ad_enabled(vcpu));
10037 vcpu->arch.mmu.set_cr3 = vmx_set_cr3;
10038 vcpu->arch.mmu.get_cr3 = nested_ept_get_cr3;
10039 vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
10041 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
10045 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
10047 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
10050 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
10053 bool inequality, bit;
10055 bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
10057 (error_code & vmcs12->page_fault_error_code_mask) !=
10058 vmcs12->page_fault_error_code_match;
10059 return inequality ^ bit;
10062 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
10063 struct x86_exception *fault)
10065 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
10067 WARN_ON(!is_guest_mode(vcpu));
10069 if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code) &&
10070 !to_vmx(vcpu)->nested.nested_run_pending) {
10071 vmcs12->vm_exit_intr_error_code = fault->error_code;
10072 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
10073 PF_VECTOR | INTR_TYPE_HARD_EXCEPTION |
10074 INTR_INFO_DELIVER_CODE_MASK | INTR_INFO_VALID_MASK,
10077 kvm_inject_page_fault(vcpu, fault);
10081 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
10082 struct vmcs12 *vmcs12);
10084 static void nested_get_vmcs12_pages(struct kvm_vcpu *vcpu,
10085 struct vmcs12 *vmcs12)
10087 struct vcpu_vmx *vmx = to_vmx(vcpu);
10091 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
10093 * Translate L1 physical address to host physical
10094 * address for vmcs02. Keep the page pinned, so this
10095 * physical address remains valid. We keep a reference
10096 * to it so we can release it later.
10098 if (vmx->nested.apic_access_page) { /* shouldn't happen */
10099 kvm_release_page_dirty(vmx->nested.apic_access_page);
10100 vmx->nested.apic_access_page = NULL;
10102 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->apic_access_addr);
10104 * If translation failed, no matter: This feature asks
10105 * to exit when accessing the given address, and if it
10106 * can never be accessed, this feature won't do
10109 if (!is_error_page(page)) {
10110 vmx->nested.apic_access_page = page;
10111 hpa = page_to_phys(vmx->nested.apic_access_page);
10112 vmcs_write64(APIC_ACCESS_ADDR, hpa);
10114 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
10115 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
10117 } else if (!(nested_cpu_has_virt_x2apic_mode(vmcs12)) &&
10118 cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
10119 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
10120 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
10121 kvm_vcpu_reload_apic_access_page(vcpu);
10124 if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
10125 if (vmx->nested.virtual_apic_page) { /* shouldn't happen */
10126 kvm_release_page_dirty(vmx->nested.virtual_apic_page);
10127 vmx->nested.virtual_apic_page = NULL;
10129 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->virtual_apic_page_addr);
10132 * If translation failed, VM entry will fail because
10133 * prepare_vmcs02 set VIRTUAL_APIC_PAGE_ADDR to -1ull.
10134 * Failing the vm entry is _not_ what the processor
10135 * does but it's basically the only possibility we
10136 * have. We could still enter the guest if CR8 load
10137 * exits are enabled, CR8 store exits are enabled, and
10138 * virtualize APIC access is disabled; in this case
10139 * the processor would never use the TPR shadow and we
10140 * could simply clear the bit from the execution
10141 * control. But such a configuration is useless, so
10142 * let's keep the code simple.
10144 if (!is_error_page(page)) {
10145 vmx->nested.virtual_apic_page = page;
10146 hpa = page_to_phys(vmx->nested.virtual_apic_page);
10147 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, hpa);
10151 if (nested_cpu_has_posted_intr(vmcs12)) {
10152 if (vmx->nested.pi_desc_page) { /* shouldn't happen */
10153 kunmap(vmx->nested.pi_desc_page);
10154 kvm_release_page_dirty(vmx->nested.pi_desc_page);
10155 vmx->nested.pi_desc_page = NULL;
10157 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->posted_intr_desc_addr);
10158 if (is_error_page(page))
10160 vmx->nested.pi_desc_page = page;
10161 vmx->nested.pi_desc = kmap(vmx->nested.pi_desc_page);
10162 vmx->nested.pi_desc =
10163 (struct pi_desc *)((void *)vmx->nested.pi_desc +
10164 (unsigned long)(vmcs12->posted_intr_desc_addr &
10166 vmcs_write64(POSTED_INTR_DESC_ADDR,
10167 page_to_phys(vmx->nested.pi_desc_page) +
10168 (unsigned long)(vmcs12->posted_intr_desc_addr &
10171 if (cpu_has_vmx_msr_bitmap() &&
10172 nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS) &&
10173 nested_vmx_merge_msr_bitmap(vcpu, vmcs12))
10174 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
10175 CPU_BASED_USE_MSR_BITMAPS);
10177 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
10178 CPU_BASED_USE_MSR_BITMAPS);
10181 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
10183 u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
10184 struct vcpu_vmx *vmx = to_vmx(vcpu);
10186 if (vcpu->arch.virtual_tsc_khz == 0)
10189 /* Make sure short timeouts reliably trigger an immediate vmexit.
10190 * hrtimer_start does not guarantee this. */
10191 if (preemption_timeout <= 1) {
10192 vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
10196 preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
10197 preemption_timeout *= 1000000;
10198 do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
10199 hrtimer_start(&vmx->nested.preemption_timer,
10200 ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
10203 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu *vcpu,
10204 struct vmcs12 *vmcs12)
10206 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
10209 if (!page_address_valid(vcpu, vmcs12->io_bitmap_a) ||
10210 !page_address_valid(vcpu, vmcs12->io_bitmap_b))
10216 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
10217 struct vmcs12 *vmcs12)
10219 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
10222 if (!page_address_valid(vcpu, vmcs12->msr_bitmap))
10228 static int nested_vmx_check_tpr_shadow_controls(struct kvm_vcpu *vcpu,
10229 struct vmcs12 *vmcs12)
10231 if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
10234 if (!page_address_valid(vcpu, vmcs12->virtual_apic_page_addr))
10241 * Merge L0's and L1's MSR bitmap, return false to indicate that
10242 * we do not use the hardware.
10244 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
10245 struct vmcs12 *vmcs12)
10249 unsigned long *msr_bitmap_l1;
10250 unsigned long *msr_bitmap_l0 = to_vmx(vcpu)->nested.vmcs02.msr_bitmap;
10252 * pred_cmd & spec_ctrl are trying to verify two things:
10254 * 1. L0 gave a permission to L1 to actually passthrough the MSR. This
10255 * ensures that we do not accidentally generate an L02 MSR bitmap
10256 * from the L12 MSR bitmap that is too permissive.
10257 * 2. That L1 or L2s have actually used the MSR. This avoids
10258 * unnecessarily merging of the bitmap if the MSR is unused. This
10259 * works properly because we only update the L01 MSR bitmap lazily.
10260 * So even if L0 should pass L1 these MSRs, the L01 bitmap is only
10261 * updated to reflect this when L1 (or its L2s) actually write to
10264 bool pred_cmd = !msr_write_intercepted_l01(vcpu, MSR_IA32_PRED_CMD);
10265 bool spec_ctrl = !msr_write_intercepted_l01(vcpu, MSR_IA32_SPEC_CTRL);
10267 if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
10268 !pred_cmd && !spec_ctrl)
10271 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->msr_bitmap);
10272 if (is_error_page(page))
10274 msr_bitmap_l1 = (unsigned long *)kmap(page);
10276 memset(msr_bitmap_l0, 0xff, PAGE_SIZE);
10278 if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
10279 if (nested_cpu_has_apic_reg_virt(vmcs12))
10280 for (msr = 0x800; msr <= 0x8ff; msr++)
10281 nested_vmx_disable_intercept_for_msr(
10282 msr_bitmap_l1, msr_bitmap_l0,
10285 nested_vmx_disable_intercept_for_msr(
10286 msr_bitmap_l1, msr_bitmap_l0,
10287 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
10288 MSR_TYPE_R | MSR_TYPE_W);
10290 if (nested_cpu_has_vid(vmcs12)) {
10291 nested_vmx_disable_intercept_for_msr(
10292 msr_bitmap_l1, msr_bitmap_l0,
10293 APIC_BASE_MSR + (APIC_EOI >> 4),
10295 nested_vmx_disable_intercept_for_msr(
10296 msr_bitmap_l1, msr_bitmap_l0,
10297 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
10303 nested_vmx_disable_intercept_for_msr(
10304 msr_bitmap_l1, msr_bitmap_l0,
10305 MSR_IA32_SPEC_CTRL,
10306 MSR_TYPE_R | MSR_TYPE_W);
10309 nested_vmx_disable_intercept_for_msr(
10310 msr_bitmap_l1, msr_bitmap_l0,
10315 kvm_release_page_clean(page);
10320 static int nested_vmx_check_apic_access_controls(struct kvm_vcpu *vcpu,
10321 struct vmcs12 *vmcs12)
10323 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
10324 !page_address_valid(vcpu, vmcs12->apic_access_addr))
10330 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
10331 struct vmcs12 *vmcs12)
10333 if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
10334 !nested_cpu_has_apic_reg_virt(vmcs12) &&
10335 !nested_cpu_has_vid(vmcs12) &&
10336 !nested_cpu_has_posted_intr(vmcs12))
10340 * If virtualize x2apic mode is enabled,
10341 * virtualize apic access must be disabled.
10343 if (nested_cpu_has_virt_x2apic_mode(vmcs12) &&
10344 nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
10348 * If virtual interrupt delivery is enabled,
10349 * we must exit on external interrupts.
10351 if (nested_cpu_has_vid(vmcs12) &&
10352 !nested_exit_on_intr(vcpu))
10356 * bits 15:8 should be zero in posted_intr_nv,
10357 * the descriptor address has been already checked
10358 * in nested_get_vmcs12_pages.
10360 if (nested_cpu_has_posted_intr(vmcs12) &&
10361 (!nested_cpu_has_vid(vmcs12) ||
10362 !nested_exit_intr_ack_set(vcpu) ||
10363 vmcs12->posted_intr_nv & 0xff00))
10366 /* tpr shadow is needed by all apicv features. */
10367 if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
10373 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
10374 unsigned long count_field,
10375 unsigned long addr_field)
10380 if (vmcs12_read_any(vcpu, count_field, &count) ||
10381 vmcs12_read_any(vcpu, addr_field, &addr)) {
10387 maxphyaddr = cpuid_maxphyaddr(vcpu);
10388 if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
10389 (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr) {
10390 pr_debug_ratelimited(
10391 "nVMX: invalid MSR switch (0x%lx, %d, %llu, 0x%08llx)",
10392 addr_field, maxphyaddr, count, addr);
10398 static int nested_vmx_check_msr_switch_controls(struct kvm_vcpu *vcpu,
10399 struct vmcs12 *vmcs12)
10401 if (vmcs12->vm_exit_msr_load_count == 0 &&
10402 vmcs12->vm_exit_msr_store_count == 0 &&
10403 vmcs12->vm_entry_msr_load_count == 0)
10404 return 0; /* Fast path */
10405 if (nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_LOAD_COUNT,
10406 VM_EXIT_MSR_LOAD_ADDR) ||
10407 nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_STORE_COUNT,
10408 VM_EXIT_MSR_STORE_ADDR) ||
10409 nested_vmx_check_msr_switch(vcpu, VM_ENTRY_MSR_LOAD_COUNT,
10410 VM_ENTRY_MSR_LOAD_ADDR))
10415 static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
10416 struct vmcs12 *vmcs12)
10418 u64 address = vmcs12->pml_address;
10419 int maxphyaddr = cpuid_maxphyaddr(vcpu);
10421 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_PML)) {
10422 if (!nested_cpu_has_ept(vmcs12) ||
10423 !IS_ALIGNED(address, 4096) ||
10424 address >> maxphyaddr)
10431 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
10432 struct vmx_msr_entry *e)
10434 /* x2APIC MSR accesses are not allowed */
10435 if (vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8)
10437 if (e->index == MSR_IA32_UCODE_WRITE || /* SDM Table 35-2 */
10438 e->index == MSR_IA32_UCODE_REV)
10440 if (e->reserved != 0)
10445 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
10446 struct vmx_msr_entry *e)
10448 if (e->index == MSR_FS_BASE ||
10449 e->index == MSR_GS_BASE ||
10450 e->index == MSR_IA32_SMM_MONITOR_CTL || /* SMM is not supported */
10451 nested_vmx_msr_check_common(vcpu, e))
10456 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
10457 struct vmx_msr_entry *e)
10459 if (e->index == MSR_IA32_SMBASE || /* SMM is not supported */
10460 nested_vmx_msr_check_common(vcpu, e))
10466 * Load guest's/host's msr at nested entry/exit.
10467 * return 0 for success, entry index for failure.
10469 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
10472 struct vmx_msr_entry e;
10473 struct msr_data msr;
10475 msr.host_initiated = false;
10476 for (i = 0; i < count; i++) {
10477 if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
10479 pr_debug_ratelimited(
10480 "%s cannot read MSR entry (%u, 0x%08llx)\n",
10481 __func__, i, gpa + i * sizeof(e));
10484 if (nested_vmx_load_msr_check(vcpu, &e)) {
10485 pr_debug_ratelimited(
10486 "%s check failed (%u, 0x%x, 0x%x)\n",
10487 __func__, i, e.index, e.reserved);
10490 msr.index = e.index;
10491 msr.data = e.value;
10492 if (kvm_set_msr(vcpu, &msr)) {
10493 pr_debug_ratelimited(
10494 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
10495 __func__, i, e.index, e.value);
10504 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
10507 struct vmx_msr_entry e;
10509 for (i = 0; i < count; i++) {
10510 struct msr_data msr_info;
10511 if (kvm_vcpu_read_guest(vcpu,
10512 gpa + i * sizeof(e),
10513 &e, 2 * sizeof(u32))) {
10514 pr_debug_ratelimited(
10515 "%s cannot read MSR entry (%u, 0x%08llx)\n",
10516 __func__, i, gpa + i * sizeof(e));
10519 if (nested_vmx_store_msr_check(vcpu, &e)) {
10520 pr_debug_ratelimited(
10521 "%s check failed (%u, 0x%x, 0x%x)\n",
10522 __func__, i, e.index, e.reserved);
10525 msr_info.host_initiated = false;
10526 msr_info.index = e.index;
10527 if (kvm_get_msr(vcpu, &msr_info)) {
10528 pr_debug_ratelimited(
10529 "%s cannot read MSR (%u, 0x%x)\n",
10530 __func__, i, e.index);
10533 if (kvm_vcpu_write_guest(vcpu,
10534 gpa + i * sizeof(e) +
10535 offsetof(struct vmx_msr_entry, value),
10536 &msr_info.data, sizeof(msr_info.data))) {
10537 pr_debug_ratelimited(
10538 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
10539 __func__, i, e.index, msr_info.data);
10546 static bool nested_cr3_valid(struct kvm_vcpu *vcpu, unsigned long val)
10548 unsigned long invalid_mask;
10550 invalid_mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
10551 return (val & invalid_mask) == 0;
10555 * Load guest's/host's cr3 at nested entry/exit. nested_ept is true if we are
10556 * emulating VM entry into a guest with EPT enabled.
10557 * Returns 0 on success, 1 on failure. Invalid state exit qualification code
10558 * is assigned to entry_failure_code on failure.
10560 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3, bool nested_ept,
10561 u32 *entry_failure_code)
10563 if (cr3 != kvm_read_cr3(vcpu) || (!nested_ept && pdptrs_changed(vcpu))) {
10564 if (!nested_cr3_valid(vcpu, cr3)) {
10565 *entry_failure_code = ENTRY_FAIL_DEFAULT;
10570 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
10571 * must not be dereferenced.
10573 if (!is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu) &&
10575 if (!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)) {
10576 *entry_failure_code = ENTRY_FAIL_PDPTE;
10581 vcpu->arch.cr3 = cr3;
10582 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
10585 kvm_mmu_reset_context(vcpu);
10590 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
10591 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
10592 * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
10593 * guest in a way that will both be appropriate to L1's requests, and our
10594 * needs. In addition to modifying the active vmcs (which is vmcs02), this
10595 * function also has additional necessary side-effects, like setting various
10596 * vcpu->arch fields.
10597 * Returns 0 on success, 1 on failure. Invalid state exit qualification code
10598 * is assigned to entry_failure_code on failure.
10600 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10601 bool from_vmentry, u32 *entry_failure_code)
10603 struct vcpu_vmx *vmx = to_vmx(vcpu);
10604 u32 exec_control, vmcs12_exec_ctrl;
10606 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
10607 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
10608 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
10609 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
10610 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
10611 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
10612 vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
10613 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
10614 vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
10615 vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
10616 vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
10617 vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
10618 vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
10619 vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
10620 vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
10621 vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
10622 vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
10623 vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
10624 vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
10625 vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
10626 vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
10627 vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
10628 vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
10629 vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
10630 vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
10631 vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
10632 vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
10633 vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
10634 vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
10635 vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
10636 vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
10637 vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
10638 vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
10639 vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
10640 vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
10641 vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
10643 if (from_vmentry &&
10644 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
10645 kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
10646 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
10648 kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
10649 vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
10651 if (from_vmentry) {
10652 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
10653 vmcs12->vm_entry_intr_info_field);
10654 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
10655 vmcs12->vm_entry_exception_error_code);
10656 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
10657 vmcs12->vm_entry_instruction_len);
10658 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
10659 vmcs12->guest_interruptibility_info);
10660 vmx->loaded_vmcs->nmi_known_unmasked =
10661 !(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
10663 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
10665 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
10666 vmx_set_rflags(vcpu, vmcs12->guest_rflags);
10667 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
10668 vmcs12->guest_pending_dbg_exceptions);
10669 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
10670 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
10672 if (nested_cpu_has_xsaves(vmcs12))
10673 vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
10674 vmcs_write64(VMCS_LINK_POINTER, -1ull);
10676 exec_control = vmcs12->pin_based_vm_exec_control;
10678 /* Preemption timer setting is only taken from vmcs01. */
10679 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
10680 exec_control |= vmcs_config.pin_based_exec_ctrl;
10681 if (vmx->hv_deadline_tsc == -1)
10682 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
10684 /* Posted interrupts setting is only taken from vmcs12. */
10685 if (nested_cpu_has_posted_intr(vmcs12)) {
10686 vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
10687 vmx->nested.pi_pending = false;
10688 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_NESTED_VECTOR);
10690 exec_control &= ~PIN_BASED_POSTED_INTR;
10693 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
10695 vmx->nested.preemption_timer_expired = false;
10696 if (nested_cpu_has_preemption_timer(vmcs12))
10697 vmx_start_preemption_timer(vcpu);
10700 * Whether page-faults are trapped is determined by a combination of
10701 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
10702 * If enable_ept, L0 doesn't care about page faults and we should
10703 * set all of these to L1's desires. However, if !enable_ept, L0 does
10704 * care about (at least some) page faults, and because it is not easy
10705 * (if at all possible?) to merge L0 and L1's desires, we simply ask
10706 * to exit on each and every L2 page fault. This is done by setting
10707 * MASK=MATCH=0 and (see below) EB.PF=1.
10708 * Note that below we don't need special code to set EB.PF beyond the
10709 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
10710 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
10711 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
10713 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
10714 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
10715 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
10716 enable_ept ? vmcs12->page_fault_error_code_match : 0);
10718 if (cpu_has_secondary_exec_ctrls()) {
10719 exec_control = vmx->secondary_exec_control;
10721 /* Take the following fields only from vmcs12 */
10722 exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
10723 SECONDARY_EXEC_ENABLE_INVPCID |
10724 SECONDARY_EXEC_RDTSCP |
10725 SECONDARY_EXEC_XSAVES |
10726 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
10727 SECONDARY_EXEC_APIC_REGISTER_VIRT |
10728 SECONDARY_EXEC_ENABLE_VMFUNC);
10729 if (nested_cpu_has(vmcs12,
10730 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)) {
10731 vmcs12_exec_ctrl = vmcs12->secondary_vm_exec_control &
10732 ~SECONDARY_EXEC_ENABLE_PML;
10733 exec_control |= vmcs12_exec_ctrl;
10736 /* All VMFUNCs are currently emulated through L0 vmexits. */
10737 if (exec_control & SECONDARY_EXEC_ENABLE_VMFUNC)
10738 vmcs_write64(VM_FUNCTION_CONTROL, 0);
10740 if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
10741 vmcs_write64(EOI_EXIT_BITMAP0,
10742 vmcs12->eoi_exit_bitmap0);
10743 vmcs_write64(EOI_EXIT_BITMAP1,
10744 vmcs12->eoi_exit_bitmap1);
10745 vmcs_write64(EOI_EXIT_BITMAP2,
10746 vmcs12->eoi_exit_bitmap2);
10747 vmcs_write64(EOI_EXIT_BITMAP3,
10748 vmcs12->eoi_exit_bitmap3);
10749 vmcs_write16(GUEST_INTR_STATUS,
10750 vmcs12->guest_intr_status);
10754 * Write an illegal value to APIC_ACCESS_ADDR. Later,
10755 * nested_get_vmcs12_pages will either fix it up or
10756 * remove the VM execution control.
10758 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
10759 vmcs_write64(APIC_ACCESS_ADDR, -1ull);
10761 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
10766 * Set host-state according to L0's settings (vmcs12 is irrelevant here)
10767 * Some constant fields are set here by vmx_set_constant_host_state().
10768 * Other fields are different per CPU, and will be set later when
10769 * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
10771 vmx_set_constant_host_state(vmx);
10774 * Set the MSR load/store lists to match L0's settings.
10776 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
10777 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
10778 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
10779 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
10780 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
10783 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
10784 * entry, but only if the current (host) sp changed from the value
10785 * we wrote last (vmx->host_rsp). This cache is no longer relevant
10786 * if we switch vmcs, and rather than hold a separate cache per vmcs,
10787 * here we just force the write to happen on entry.
10791 exec_control = vmx_exec_control(vmx); /* L0's desires */
10792 exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
10793 exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
10794 exec_control &= ~CPU_BASED_TPR_SHADOW;
10795 exec_control |= vmcs12->cpu_based_vm_exec_control;
10798 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR. Later, if
10799 * nested_get_vmcs12_pages can't fix it up, the illegal value
10800 * will result in a VM entry failure.
10802 if (exec_control & CPU_BASED_TPR_SHADOW) {
10803 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, -1ull);
10804 vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
10806 #ifdef CONFIG_X86_64
10807 exec_control |= CPU_BASED_CR8_LOAD_EXITING |
10808 CPU_BASED_CR8_STORE_EXITING;
10813 * Merging of IO bitmap not currently supported.
10814 * Rather, exit every time.
10816 exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
10817 exec_control |= CPU_BASED_UNCOND_IO_EXITING;
10819 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
10821 /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
10822 * bitwise-or of what L1 wants to trap for L2, and what we want to
10823 * trap. Note that CR0.TS also needs updating - we do this later.
10825 update_exception_bitmap(vcpu);
10826 vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
10827 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
10829 /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
10830 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
10831 * bits are further modified by vmx_set_efer() below.
10833 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
10835 /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
10836 * emulated by vmx_set_efer(), below.
10838 vm_entry_controls_init(vmx,
10839 (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER &
10840 ~VM_ENTRY_IA32E_MODE) |
10841 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
10843 if (from_vmentry &&
10844 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
10845 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
10846 vcpu->arch.pat = vmcs12->guest_ia32_pat;
10847 } else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
10848 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
10851 set_cr4_guest_host_mask(vmx);
10853 if (from_vmentry &&
10854 vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)
10855 vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
10857 if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
10858 vmcs_write64(TSC_OFFSET,
10859 vcpu->arch.tsc_offset + vmcs12->tsc_offset);
10861 vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
10862 if (kvm_has_tsc_control)
10863 decache_tsc_multiplier(vmx);
10865 if (cpu_has_vmx_msr_bitmap())
10866 vmcs_write64(MSR_BITMAP, __pa(vmx->nested.vmcs02.msr_bitmap));
10870 * There is no direct mapping between vpid02 and vpid12, the
10871 * vpid02 is per-vCPU for L0 and reused while the value of
10872 * vpid12 is changed w/ one invvpid during nested vmentry.
10873 * The vpid12 is allocated by L1 for L2, so it will not
10874 * influence global bitmap(for vpid01 and vpid02 allocation)
10875 * even if spawn a lot of nested vCPUs.
10877 if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02) {
10878 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
10879 if (vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
10880 vmx->nested.last_vpid = vmcs12->virtual_processor_id;
10881 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->nested.vpid02);
10884 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
10885 vmx_flush_tlb(vcpu);
10892 * Conceptually we want to copy the PML address and index from
10893 * vmcs01 here, and then back to vmcs01 on nested vmexit. But,
10894 * since we always flush the log on each vmexit, this happens
10895 * to be equivalent to simply resetting the fields in vmcs02.
10897 ASSERT(vmx->pml_pg);
10898 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
10899 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
10902 if (nested_cpu_has_ept(vmcs12)) {
10903 if (nested_ept_init_mmu_context(vcpu)) {
10904 *entry_failure_code = ENTRY_FAIL_DEFAULT;
10907 } else if (nested_cpu_has2(vmcs12,
10908 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
10909 vmx_flush_tlb_ept_only(vcpu);
10913 * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
10914 * bits which we consider mandatory enabled.
10915 * The CR0_READ_SHADOW is what L2 should have expected to read given
10916 * the specifications by L1; It's not enough to take
10917 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
10918 * have more bits than L1 expected.
10920 vmx_set_cr0(vcpu, vmcs12->guest_cr0);
10921 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
10923 vmx_set_cr4(vcpu, vmcs12->guest_cr4);
10924 vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
10926 if (from_vmentry &&
10927 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
10928 vcpu->arch.efer = vmcs12->guest_ia32_efer;
10929 else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
10930 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10932 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10933 /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
10934 vmx_set_efer(vcpu, vcpu->arch.efer);
10936 /* Shadow page tables on either EPT or shadow page tables. */
10937 if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
10938 entry_failure_code))
10942 vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
10945 * L1 may access the L2's PDPTR, so save them to construct vmcs12
10948 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
10949 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
10950 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
10951 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
10954 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
10955 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
10959 static int check_vmentry_prereqs(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10961 struct vcpu_vmx *vmx = to_vmx(vcpu);
10963 if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
10964 vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT)
10965 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10967 if (nested_vmx_check_io_bitmap_controls(vcpu, vmcs12))
10968 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10970 if (nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12))
10971 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10973 if (nested_vmx_check_apic_access_controls(vcpu, vmcs12))
10974 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10976 if (nested_vmx_check_tpr_shadow_controls(vcpu, vmcs12))
10977 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10979 if (nested_vmx_check_apicv_controls(vcpu, vmcs12))
10980 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10982 if (nested_vmx_check_msr_switch_controls(vcpu, vmcs12))
10983 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10985 if (nested_vmx_check_pml_controls(vcpu, vmcs12))
10986 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10988 if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
10989 vmx->nested.nested_vmx_procbased_ctls_low,
10990 vmx->nested.nested_vmx_procbased_ctls_high) ||
10991 (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
10992 !vmx_control_verify(vmcs12->secondary_vm_exec_control,
10993 vmx->nested.nested_vmx_secondary_ctls_low,
10994 vmx->nested.nested_vmx_secondary_ctls_high)) ||
10995 !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
10996 vmx->nested.nested_vmx_pinbased_ctls_low,
10997 vmx->nested.nested_vmx_pinbased_ctls_high) ||
10998 !vmx_control_verify(vmcs12->vm_exit_controls,
10999 vmx->nested.nested_vmx_exit_ctls_low,
11000 vmx->nested.nested_vmx_exit_ctls_high) ||
11001 !vmx_control_verify(vmcs12->vm_entry_controls,
11002 vmx->nested.nested_vmx_entry_ctls_low,
11003 vmx->nested.nested_vmx_entry_ctls_high))
11004 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
11006 if (nested_cpu_has_vmfunc(vmcs12)) {
11007 if (vmcs12->vm_function_control &
11008 ~vmx->nested.nested_vmx_vmfunc_controls)
11009 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
11011 if (nested_cpu_has_eptp_switching(vmcs12)) {
11012 if (!nested_cpu_has_ept(vmcs12) ||
11013 !page_address_valid(vcpu, vmcs12->eptp_list_address))
11014 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
11018 if (vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu))
11019 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
11021 if (!nested_host_cr0_valid(vcpu, vmcs12->host_cr0) ||
11022 !nested_host_cr4_valid(vcpu, vmcs12->host_cr4) ||
11023 !nested_cr3_valid(vcpu, vmcs12->host_cr3))
11024 return VMXERR_ENTRY_INVALID_HOST_STATE_FIELD;
11029 static int check_vmentry_postreqs(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
11034 *exit_qual = ENTRY_FAIL_DEFAULT;
11036 if (!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0) ||
11037 !nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4))
11040 if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_SHADOW_VMCS) &&
11041 vmcs12->vmcs_link_pointer != -1ull) {
11042 *exit_qual = ENTRY_FAIL_VMCS_LINK_PTR;
11047 * If the load IA32_EFER VM-entry control is 1, the following checks
11048 * are performed on the field for the IA32_EFER MSR:
11049 * - Bits reserved in the IA32_EFER MSR must be 0.
11050 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
11051 * the IA-32e mode guest VM-exit control. It must also be identical
11052 * to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
11055 if (to_vmx(vcpu)->nested.nested_run_pending &&
11056 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
11057 ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
11058 if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
11059 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
11060 ((vmcs12->guest_cr0 & X86_CR0_PG) &&
11061 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME)))
11066 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
11067 * IA32_EFER MSR must be 0 in the field for that register. In addition,
11068 * the values of the LMA and LME bits in the field must each be that of
11069 * the host address-space size VM-exit control.
11071 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
11072 ia32e = (vmcs12->vm_exit_controls &
11073 VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
11074 if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
11075 ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
11076 ia32e != !!(vmcs12->host_ia32_efer & EFER_LME))
11083 static int enter_vmx_non_root_mode(struct kvm_vcpu *vcpu, bool from_vmentry)
11085 struct vcpu_vmx *vmx = to_vmx(vcpu);
11086 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
11090 enter_guest_mode(vcpu);
11092 if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
11093 vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
11095 vmx_switch_vmcs(vcpu, &vmx->nested.vmcs02);
11096 vmx_segment_cache_clear(vmx);
11098 if (prepare_vmcs02(vcpu, vmcs12, from_vmentry, &exit_qual)) {
11099 leave_guest_mode(vcpu);
11100 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
11101 nested_vmx_entry_failure(vcpu, vmcs12,
11102 EXIT_REASON_INVALID_STATE, exit_qual);
11106 nested_get_vmcs12_pages(vcpu, vmcs12);
11108 msr_entry_idx = nested_vmx_load_msr(vcpu,
11109 vmcs12->vm_entry_msr_load_addr,
11110 vmcs12->vm_entry_msr_load_count);
11111 if (msr_entry_idx) {
11112 leave_guest_mode(vcpu);
11113 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
11114 nested_vmx_entry_failure(vcpu, vmcs12,
11115 EXIT_REASON_MSR_LOAD_FAIL, msr_entry_idx);
11120 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
11121 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
11122 * returned as far as L1 is concerned. It will only return (and set
11123 * the success flag) when L2 exits (see nested_vmx_vmexit()).
11129 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
11130 * for running an L2 nested guest.
11132 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
11134 struct vmcs12 *vmcs12;
11135 struct vcpu_vmx *vmx = to_vmx(vcpu);
11136 u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
11140 if (!nested_vmx_check_permission(vcpu))
11143 if (!nested_vmx_check_vmcs12(vcpu))
11146 vmcs12 = get_vmcs12(vcpu);
11148 if (enable_shadow_vmcs)
11149 copy_shadow_to_vmcs12(vmx);
11152 * The nested entry process starts with enforcing various prerequisites
11153 * on vmcs12 as required by the Intel SDM, and act appropriately when
11154 * they fail: As the SDM explains, some conditions should cause the
11155 * instruction to fail, while others will cause the instruction to seem
11156 * to succeed, but return an EXIT_REASON_INVALID_STATE.
11157 * To speed up the normal (success) code path, we should avoid checking
11158 * for misconfigurations which will anyway be caught by the processor
11159 * when using the merged vmcs02.
11161 if (interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS) {
11162 nested_vmx_failValid(vcpu,
11163 VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
11167 if (vmcs12->launch_state == launch) {
11168 nested_vmx_failValid(vcpu,
11169 launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
11170 : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
11174 ret = check_vmentry_prereqs(vcpu, vmcs12);
11176 nested_vmx_failValid(vcpu, ret);
11181 * After this point, the trap flag no longer triggers a singlestep trap
11182 * on the vm entry instructions; don't call kvm_skip_emulated_instruction.
11183 * This is not 100% correct; for performance reasons, we delegate most
11184 * of the checks on host state to the processor. If those fail,
11185 * the singlestep trap is missed.
11187 skip_emulated_instruction(vcpu);
11189 ret = check_vmentry_postreqs(vcpu, vmcs12, &exit_qual);
11191 nested_vmx_entry_failure(vcpu, vmcs12,
11192 EXIT_REASON_INVALID_STATE, exit_qual);
11197 * We're finally done with prerequisite checking, and can start with
11198 * the nested entry.
11201 ret = enter_vmx_non_root_mode(vcpu, true);
11206 * If we're entering a halted L2 vcpu and the L2 vcpu won't be woken
11207 * by event injection, halt vcpu.
11209 if ((vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT) &&
11210 !(vmcs12->vm_entry_intr_info_field & INTR_INFO_VALID_MASK))
11211 return kvm_vcpu_halt(vcpu);
11213 vmx->nested.nested_run_pending = 1;
11218 return kvm_skip_emulated_instruction(vcpu);
11222 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
11223 * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
11224 * This function returns the new value we should put in vmcs12.guest_cr0.
11225 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
11226 * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
11227 * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
11228 * didn't trap the bit, because if L1 did, so would L0).
11229 * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have
11230 * been modified by L2, and L1 knows it. So just leave the old value of
11231 * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
11232 * isn't relevant, because if L0 traps this bit it can set it to anything.
11233 * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
11234 * changed these bits, and therefore they need to be updated, but L0
11235 * didn't necessarily allow them to be changed in GUEST_CR0 - and rather
11236 * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
11238 static inline unsigned long
11239 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
11242 /*1*/ (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
11243 /*2*/ (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
11244 /*3*/ (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
11245 vcpu->arch.cr0_guest_owned_bits));
11248 static inline unsigned long
11249 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
11252 /*1*/ (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
11253 /*2*/ (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
11254 /*3*/ (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
11255 vcpu->arch.cr4_guest_owned_bits));
11258 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
11259 struct vmcs12 *vmcs12)
11264 if (vcpu->arch.exception.injected) {
11265 nr = vcpu->arch.exception.nr;
11266 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
11268 if (kvm_exception_is_soft(nr)) {
11269 vmcs12->vm_exit_instruction_len =
11270 vcpu->arch.event_exit_inst_len;
11271 idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
11273 idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
11275 if (vcpu->arch.exception.has_error_code) {
11276 idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
11277 vmcs12->idt_vectoring_error_code =
11278 vcpu->arch.exception.error_code;
11281 vmcs12->idt_vectoring_info_field = idt_vectoring;
11282 } else if (vcpu->arch.nmi_injected) {
11283 vmcs12->idt_vectoring_info_field =
11284 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
11285 } else if (vcpu->arch.interrupt.pending) {
11286 nr = vcpu->arch.interrupt.nr;
11287 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
11289 if (vcpu->arch.interrupt.soft) {
11290 idt_vectoring |= INTR_TYPE_SOFT_INTR;
11291 vmcs12->vm_entry_instruction_len =
11292 vcpu->arch.event_exit_inst_len;
11294 idt_vectoring |= INTR_TYPE_EXT_INTR;
11296 vmcs12->idt_vectoring_info_field = idt_vectoring;
11300 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
11302 struct vcpu_vmx *vmx = to_vmx(vcpu);
11303 unsigned long exit_qual;
11304 bool block_nested_events =
11305 vmx->nested.nested_run_pending || kvm_event_needs_reinjection(vcpu);
11307 if (vcpu->arch.exception.pending &&
11308 nested_vmx_check_exception(vcpu, &exit_qual)) {
11309 if (block_nested_events)
11311 nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
11315 if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
11316 vmx->nested.preemption_timer_expired) {
11317 if (block_nested_events)
11319 nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
11323 if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
11324 if (block_nested_events)
11326 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
11327 NMI_VECTOR | INTR_TYPE_NMI_INTR |
11328 INTR_INFO_VALID_MASK, 0);
11330 * The NMI-triggered VM exit counts as injection:
11331 * clear this one and block further NMIs.
11333 vcpu->arch.nmi_pending = 0;
11334 vmx_set_nmi_mask(vcpu, true);
11338 if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
11339 nested_exit_on_intr(vcpu)) {
11340 if (block_nested_events)
11342 nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
11346 vmx_complete_nested_posted_interrupt(vcpu);
11350 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
11352 ktime_t remaining =
11353 hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
11356 if (ktime_to_ns(remaining) <= 0)
11359 value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
11360 do_div(value, 1000000);
11361 return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
11365 * Update the guest state fields of vmcs12 to reflect changes that
11366 * occurred while L2 was running. (The "IA-32e mode guest" bit of the
11367 * VM-entry controls is also updated, since this is really a guest
11370 static void sync_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
11372 vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
11373 vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
11375 vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
11376 vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
11377 vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
11379 vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
11380 vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
11381 vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
11382 vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
11383 vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
11384 vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
11385 vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
11386 vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
11387 vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
11388 vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
11389 vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
11390 vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
11391 vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
11392 vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
11393 vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
11394 vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
11395 vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
11396 vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
11397 vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
11398 vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
11399 vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
11400 vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
11401 vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
11402 vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
11403 vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
11404 vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
11405 vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
11406 vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
11407 vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
11408 vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
11409 vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
11410 vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
11411 vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
11412 vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
11413 vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
11414 vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
11416 vmcs12->guest_interruptibility_info =
11417 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
11418 vmcs12->guest_pending_dbg_exceptions =
11419 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
11420 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
11421 vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
11423 vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
11425 if (nested_cpu_has_preemption_timer(vmcs12)) {
11426 if (vmcs12->vm_exit_controls &
11427 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
11428 vmcs12->vmx_preemption_timer_value =
11429 vmx_get_preemption_timer_value(vcpu);
11430 hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
11434 * In some cases (usually, nested EPT), L2 is allowed to change its
11435 * own CR3 without exiting. If it has changed it, we must keep it.
11436 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
11437 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
11439 * Additionally, restore L2's PDPTR to vmcs12.
11442 vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
11443 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
11444 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
11445 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
11446 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
11449 vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
11451 if (nested_cpu_has_vid(vmcs12))
11452 vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
11454 vmcs12->vm_entry_controls =
11455 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
11456 (vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
11458 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) {
11459 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
11460 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
11463 /* TODO: These cannot have changed unless we have MSR bitmaps and
11464 * the relevant bit asks not to trap the change */
11465 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
11466 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
11467 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
11468 vmcs12->guest_ia32_efer = vcpu->arch.efer;
11469 vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
11470 vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
11471 vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
11472 if (kvm_mpx_supported())
11473 vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
11477 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
11478 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
11479 * and this function updates it to reflect the changes to the guest state while
11480 * L2 was running (and perhaps made some exits which were handled directly by L0
11481 * without going back to L1), and to reflect the exit reason.
11482 * Note that we do not have to copy here all VMCS fields, just those that
11483 * could have changed by the L2 guest or the exit - i.e., the guest-state and
11484 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
11485 * which already writes to vmcs12 directly.
11487 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
11488 u32 exit_reason, u32 exit_intr_info,
11489 unsigned long exit_qualification)
11491 /* update guest state fields: */
11492 sync_vmcs12(vcpu, vmcs12);
11494 /* update exit information fields: */
11496 vmcs12->vm_exit_reason = exit_reason;
11497 vmcs12->exit_qualification = exit_qualification;
11498 vmcs12->vm_exit_intr_info = exit_intr_info;
11500 vmcs12->idt_vectoring_info_field = 0;
11501 vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
11502 vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
11504 if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
11505 vmcs12->launch_state = 1;
11507 /* vm_entry_intr_info_field is cleared on exit. Emulate this
11508 * instead of reading the real value. */
11509 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
11512 * Transfer the event that L0 or L1 may wanted to inject into
11513 * L2 to IDT_VECTORING_INFO_FIELD.
11515 vmcs12_save_pending_event(vcpu, vmcs12);
11519 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
11520 * preserved above and would only end up incorrectly in L1.
11522 vcpu->arch.nmi_injected = false;
11523 kvm_clear_exception_queue(vcpu);
11524 kvm_clear_interrupt_queue(vcpu);
11527 static void load_vmcs12_mmu_host_state(struct kvm_vcpu *vcpu,
11528 struct vmcs12 *vmcs12)
11530 u32 entry_failure_code;
11532 nested_ept_uninit_mmu_context(vcpu);
11535 * Only PDPTE load can fail as the value of cr3 was checked on entry and
11536 * couldn't have changed.
11538 if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, &entry_failure_code))
11539 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
11542 vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
11546 * A part of what we need to when the nested L2 guest exits and we want to
11547 * run its L1 parent, is to reset L1's guest state to the host state specified
11549 * This function is to be called not only on normal nested exit, but also on
11550 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
11551 * Failures During or After Loading Guest State").
11552 * This function should be called when the active VMCS is L1's (vmcs01).
11554 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
11555 struct vmcs12 *vmcs12)
11557 struct kvm_segment seg;
11559 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
11560 vcpu->arch.efer = vmcs12->host_ia32_efer;
11561 else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
11562 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
11564 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
11565 vmx_set_efer(vcpu, vcpu->arch.efer);
11567 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
11568 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
11569 vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
11571 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
11572 * actually changed, because vmx_set_cr0 refers to efer set above.
11574 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
11575 * (KVM doesn't change it);
11577 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
11578 vmx_set_cr0(vcpu, vmcs12->host_cr0);
11580 /* Same as above - no reason to call set_cr4_guest_host_mask(). */
11581 vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
11582 vmx_set_cr4(vcpu, vmcs12->host_cr4);
11584 load_vmcs12_mmu_host_state(vcpu, vmcs12);
11588 * Trivially support vpid by letting L2s share their parent
11589 * L1's vpid. TODO: move to a more elaborate solution, giving
11590 * each L2 its own vpid and exposing the vpid feature to L1.
11592 vmx_flush_tlb(vcpu);
11594 /* Restore posted intr vector. */
11595 if (nested_cpu_has_posted_intr(vmcs12))
11596 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
11598 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
11599 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
11600 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
11601 vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
11602 vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
11603 vmcs_write32(GUEST_IDTR_LIMIT, 0xFFFF);
11604 vmcs_write32(GUEST_GDTR_LIMIT, 0xFFFF);
11606 /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1. */
11607 if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
11608 vmcs_write64(GUEST_BNDCFGS, 0);
11610 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
11611 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
11612 vcpu->arch.pat = vmcs12->host_ia32_pat;
11614 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
11615 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
11616 vmcs12->host_ia32_perf_global_ctrl);
11618 /* Set L1 segment info according to Intel SDM
11619 27.5.2 Loading Host Segment and Descriptor-Table Registers */
11620 seg = (struct kvm_segment) {
11622 .limit = 0xFFFFFFFF,
11623 .selector = vmcs12->host_cs_selector,
11629 if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
11633 vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
11634 seg = (struct kvm_segment) {
11636 .limit = 0xFFFFFFFF,
11643 seg.selector = vmcs12->host_ds_selector;
11644 vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
11645 seg.selector = vmcs12->host_es_selector;
11646 vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
11647 seg.selector = vmcs12->host_ss_selector;
11648 vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
11649 seg.selector = vmcs12->host_fs_selector;
11650 seg.base = vmcs12->host_fs_base;
11651 vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
11652 seg.selector = vmcs12->host_gs_selector;
11653 seg.base = vmcs12->host_gs_base;
11654 vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
11655 seg = (struct kvm_segment) {
11656 .base = vmcs12->host_tr_base,
11658 .selector = vmcs12->host_tr_selector,
11662 vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
11664 kvm_set_dr(vcpu, 7, 0x400);
11665 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
11667 if (cpu_has_vmx_msr_bitmap())
11668 vmx_update_msr_bitmap(vcpu);
11670 if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
11671 vmcs12->vm_exit_msr_load_count))
11672 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
11676 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
11677 * and modify vmcs12 to make it see what it would expect to see there if
11678 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
11680 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
11681 u32 exit_intr_info,
11682 unsigned long exit_qualification)
11684 struct vcpu_vmx *vmx = to_vmx(vcpu);
11685 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
11687 /* trying to cancel vmlaunch/vmresume is a bug */
11688 WARN_ON_ONCE(vmx->nested.nested_run_pending);
11691 * The only expected VM-instruction error is "VM entry with
11692 * invalid control field(s)." Anything else indicates a
11695 WARN_ON_ONCE(vmx->fail && (vmcs_read32(VM_INSTRUCTION_ERROR) !=
11696 VMXERR_ENTRY_INVALID_CONTROL_FIELD));
11698 leave_guest_mode(vcpu);
11700 if (likely(!vmx->fail)) {
11701 prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
11702 exit_qualification);
11704 if (nested_vmx_store_msr(vcpu, vmcs12->vm_exit_msr_store_addr,
11705 vmcs12->vm_exit_msr_store_count))
11706 nested_vmx_abort(vcpu, VMX_ABORT_SAVE_GUEST_MSR_FAIL);
11709 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
11710 vm_entry_controls_reset_shadow(vmx);
11711 vm_exit_controls_reset_shadow(vmx);
11712 vmx_segment_cache_clear(vmx);
11714 /* Update any VMCS fields that might have changed while L2 ran */
11715 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
11716 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
11717 vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
11718 if (vmx->hv_deadline_tsc == -1)
11719 vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
11720 PIN_BASED_VMX_PREEMPTION_TIMER);
11722 vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
11723 PIN_BASED_VMX_PREEMPTION_TIMER);
11724 if (kvm_has_tsc_control)
11725 decache_tsc_multiplier(vmx);
11727 if (vmx->nested.change_vmcs01_virtual_x2apic_mode) {
11728 vmx->nested.change_vmcs01_virtual_x2apic_mode = false;
11729 vmx_set_virtual_x2apic_mode(vcpu,
11730 vcpu->arch.apic_base & X2APIC_ENABLE);
11731 } else if (!nested_cpu_has_ept(vmcs12) &&
11732 nested_cpu_has2(vmcs12,
11733 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
11734 vmx_flush_tlb_ept_only(vcpu);
11737 /* This is needed for same reason as it was needed in prepare_vmcs02 */
11740 /* Unpin physical memory we referred to in vmcs02 */
11741 if (vmx->nested.apic_access_page) {
11742 kvm_release_page_dirty(vmx->nested.apic_access_page);
11743 vmx->nested.apic_access_page = NULL;
11745 if (vmx->nested.virtual_apic_page) {
11746 kvm_release_page_dirty(vmx->nested.virtual_apic_page);
11747 vmx->nested.virtual_apic_page = NULL;
11749 if (vmx->nested.pi_desc_page) {
11750 kunmap(vmx->nested.pi_desc_page);
11751 kvm_release_page_dirty(vmx->nested.pi_desc_page);
11752 vmx->nested.pi_desc_page = NULL;
11753 vmx->nested.pi_desc = NULL;
11757 * We are now running in L2, mmu_notifier will force to reload the
11758 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
11760 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
11762 if (enable_shadow_vmcs)
11763 vmx->nested.sync_shadow_vmcs = true;
11765 /* in case we halted in L2 */
11766 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11768 if (likely(!vmx->fail)) {
11770 * TODO: SDM says that with acknowledge interrupt on
11771 * exit, bit 31 of the VM-exit interrupt information
11772 * (valid interrupt) is always set to 1 on
11773 * EXIT_REASON_EXTERNAL_INTERRUPT, so we shouldn't
11774 * need kvm_cpu_has_interrupt(). See the commit
11775 * message for details.
11777 if (nested_exit_intr_ack_set(vcpu) &&
11778 exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT &&
11779 kvm_cpu_has_interrupt(vcpu)) {
11780 int irq = kvm_cpu_get_interrupt(vcpu);
11782 vmcs12->vm_exit_intr_info = irq |
11783 INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
11786 trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
11787 vmcs12->exit_qualification,
11788 vmcs12->idt_vectoring_info_field,
11789 vmcs12->vm_exit_intr_info,
11790 vmcs12->vm_exit_intr_error_code,
11793 load_vmcs12_host_state(vcpu, vmcs12);
11799 * After an early L2 VM-entry failure, we're now back
11800 * in L1 which thinks it just finished a VMLAUNCH or
11801 * VMRESUME instruction, so we need to set the failure
11802 * flag and the VM-instruction error field of the VMCS
11805 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
11807 load_vmcs12_mmu_host_state(vcpu, vmcs12);
11810 * The emulated instruction was already skipped in
11811 * nested_vmx_run, but the updated RIP was never
11812 * written back to the vmcs01.
11814 skip_emulated_instruction(vcpu);
11819 * Forcibly leave nested mode in order to be able to reset the VCPU later on.
11821 static void vmx_leave_nested(struct kvm_vcpu *vcpu)
11823 if (is_guest_mode(vcpu)) {
11824 to_vmx(vcpu)->nested.nested_run_pending = 0;
11825 nested_vmx_vmexit(vcpu, -1, 0, 0);
11827 free_nested(to_vmx(vcpu));
11831 * L1's failure to enter L2 is a subset of a normal exit, as explained in
11832 * 23.7 "VM-entry failures during or after loading guest state" (this also
11833 * lists the acceptable exit-reason and exit-qualification parameters).
11834 * It should only be called before L2 actually succeeded to run, and when
11835 * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
11837 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
11838 struct vmcs12 *vmcs12,
11839 u32 reason, unsigned long qualification)
11841 load_vmcs12_host_state(vcpu, vmcs12);
11842 vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
11843 vmcs12->exit_qualification = qualification;
11844 nested_vmx_succeed(vcpu);
11845 if (enable_shadow_vmcs)
11846 to_vmx(vcpu)->nested.sync_shadow_vmcs = true;
11849 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
11850 struct x86_instruction_info *info,
11851 enum x86_intercept_stage stage)
11853 return X86EMUL_CONTINUE;
11856 #ifdef CONFIG_X86_64
11857 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
11858 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
11859 u64 divisor, u64 *result)
11861 u64 low = a << shift, high = a >> (64 - shift);
11863 /* To avoid the overflow on divq */
11864 if (high >= divisor)
11867 /* Low hold the result, high hold rem which is discarded */
11868 asm("divq %2\n\t" : "=a" (low), "=d" (high) :
11869 "rm" (divisor), "0" (low), "1" (high));
11875 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc)
11877 struct vcpu_vmx *vmx = to_vmx(vcpu);
11878 u64 tscl = rdtsc();
11879 u64 guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
11880 u64 delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
11882 /* Convert to host delta tsc if tsc scaling is enabled */
11883 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
11884 u64_shl_div_u64(delta_tsc,
11885 kvm_tsc_scaling_ratio_frac_bits,
11886 vcpu->arch.tsc_scaling_ratio,
11891 * If the delta tsc can't fit in the 32 bit after the multi shift,
11892 * we can't use the preemption timer.
11893 * It's possible that it fits on later vmentries, but checking
11894 * on every vmentry is costly so we just use an hrtimer.
11896 if (delta_tsc >> (cpu_preemption_timer_multi + 32))
11899 vmx->hv_deadline_tsc = tscl + delta_tsc;
11900 vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
11901 PIN_BASED_VMX_PREEMPTION_TIMER);
11903 return delta_tsc == 0;
11906 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
11908 struct vcpu_vmx *vmx = to_vmx(vcpu);
11909 vmx->hv_deadline_tsc = -1;
11910 vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
11911 PIN_BASED_VMX_PREEMPTION_TIMER);
11915 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
11918 shrink_ple_window(vcpu);
11921 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
11922 struct kvm_memory_slot *slot)
11924 kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
11925 kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
11928 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
11929 struct kvm_memory_slot *slot)
11931 kvm_mmu_slot_set_dirty(kvm, slot);
11934 static void vmx_flush_log_dirty(struct kvm *kvm)
11936 kvm_flush_pml_buffers(kvm);
11939 static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu)
11941 struct vmcs12 *vmcs12;
11942 struct vcpu_vmx *vmx = to_vmx(vcpu);
11944 struct page *page = NULL;
11947 if (is_guest_mode(vcpu)) {
11948 WARN_ON_ONCE(vmx->nested.pml_full);
11951 * Check if PML is enabled for the nested guest.
11952 * Whether eptp bit 6 is set is already checked
11953 * as part of A/D emulation.
11955 vmcs12 = get_vmcs12(vcpu);
11956 if (!nested_cpu_has_pml(vmcs12))
11959 if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
11960 vmx->nested.pml_full = true;
11964 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull;
11966 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->pml_address);
11967 if (is_error_page(page))
11970 pml_address = kmap(page);
11971 pml_address[vmcs12->guest_pml_index--] = gpa;
11973 kvm_release_page_clean(page);
11979 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
11980 struct kvm_memory_slot *memslot,
11981 gfn_t offset, unsigned long mask)
11983 kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
11986 static void __pi_post_block(struct kvm_vcpu *vcpu)
11988 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
11989 struct pi_desc old, new;
11993 old.control = new.control = pi_desc->control;
11994 WARN(old.nv != POSTED_INTR_WAKEUP_VECTOR,
11995 "Wakeup handler not enabled while the VCPU is blocked\n");
11997 dest = cpu_physical_id(vcpu->cpu);
11999 if (x2apic_enabled())
12002 new.ndst = (dest << 8) & 0xFF00;
12004 /* set 'NV' to 'notification vector' */
12005 new.nv = POSTED_INTR_VECTOR;
12006 } while (cmpxchg64(&pi_desc->control, old.control,
12007 new.control) != old.control);
12009 if (!WARN_ON_ONCE(vcpu->pre_pcpu == -1)) {
12010 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
12011 list_del(&vcpu->blocked_vcpu_list);
12012 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
12013 vcpu->pre_pcpu = -1;
12018 * This routine does the following things for vCPU which is going
12019 * to be blocked if VT-d PI is enabled.
12020 * - Store the vCPU to the wakeup list, so when interrupts happen
12021 * we can find the right vCPU to wake up.
12022 * - Change the Posted-interrupt descriptor as below:
12023 * 'NDST' <-- vcpu->pre_pcpu
12024 * 'NV' <-- POSTED_INTR_WAKEUP_VECTOR
12025 * - If 'ON' is set during this process, which means at least one
12026 * interrupt is posted for this vCPU, we cannot block it, in
12027 * this case, return 1, otherwise, return 0.
12030 static int pi_pre_block(struct kvm_vcpu *vcpu)
12033 struct pi_desc old, new;
12034 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
12036 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
12037 !irq_remapping_cap(IRQ_POSTING_CAP) ||
12038 !kvm_vcpu_apicv_active(vcpu))
12041 WARN_ON(irqs_disabled());
12042 local_irq_disable();
12043 if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) {
12044 vcpu->pre_pcpu = vcpu->cpu;
12045 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
12046 list_add_tail(&vcpu->blocked_vcpu_list,
12047 &per_cpu(blocked_vcpu_on_cpu,
12049 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
12053 old.control = new.control = pi_desc->control;
12055 WARN((pi_desc->sn == 1),
12056 "Warning: SN field of posted-interrupts "
12057 "is set before blocking\n");
12060 * Since vCPU can be preempted during this process,
12061 * vcpu->cpu could be different with pre_pcpu, we
12062 * need to set pre_pcpu as the destination of wakeup
12063 * notification event, then we can find the right vCPU
12064 * to wakeup in wakeup handler if interrupts happen
12065 * when the vCPU is in blocked state.
12067 dest = cpu_physical_id(vcpu->pre_pcpu);
12069 if (x2apic_enabled())
12072 new.ndst = (dest << 8) & 0xFF00;
12074 /* set 'NV' to 'wakeup vector' */
12075 new.nv = POSTED_INTR_WAKEUP_VECTOR;
12076 } while (cmpxchg64(&pi_desc->control, old.control,
12077 new.control) != old.control);
12079 /* We should not block the vCPU if an interrupt is posted for it. */
12080 if (pi_test_on(pi_desc) == 1)
12081 __pi_post_block(vcpu);
12083 local_irq_enable();
12084 return (vcpu->pre_pcpu == -1);
12087 static int vmx_pre_block(struct kvm_vcpu *vcpu)
12089 if (pi_pre_block(vcpu))
12092 if (kvm_lapic_hv_timer_in_use(vcpu))
12093 kvm_lapic_switch_to_sw_timer(vcpu);
12098 static void pi_post_block(struct kvm_vcpu *vcpu)
12100 if (vcpu->pre_pcpu == -1)
12103 WARN_ON(irqs_disabled());
12104 local_irq_disable();
12105 __pi_post_block(vcpu);
12106 local_irq_enable();
12109 static void vmx_post_block(struct kvm_vcpu *vcpu)
12111 if (kvm_x86_ops->set_hv_timer)
12112 kvm_lapic_switch_to_hv_timer(vcpu);
12114 pi_post_block(vcpu);
12118 * vmx_update_pi_irte - set IRTE for Posted-Interrupts
12121 * @host_irq: host irq of the interrupt
12122 * @guest_irq: gsi of the interrupt
12123 * @set: set or unset PI
12124 * returns 0 on success, < 0 on failure
12126 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
12127 uint32_t guest_irq, bool set)
12129 struct kvm_kernel_irq_routing_entry *e;
12130 struct kvm_irq_routing_table *irq_rt;
12131 struct kvm_lapic_irq irq;
12132 struct kvm_vcpu *vcpu;
12133 struct vcpu_data vcpu_info;
12136 if (!kvm_arch_has_assigned_device(kvm) ||
12137 !irq_remapping_cap(IRQ_POSTING_CAP) ||
12138 !kvm_vcpu_apicv_active(kvm->vcpus[0]))
12141 idx = srcu_read_lock(&kvm->irq_srcu);
12142 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
12143 if (guest_irq >= irq_rt->nr_rt_entries ||
12144 hlist_empty(&irq_rt->map[guest_irq])) {
12145 pr_warn_once("no route for guest_irq %u/%u (broken user space?)\n",
12146 guest_irq, irq_rt->nr_rt_entries);
12150 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
12151 if (e->type != KVM_IRQ_ROUTING_MSI)
12154 * VT-d PI cannot support posting multicast/broadcast
12155 * interrupts to a vCPU, we still use interrupt remapping
12156 * for these kind of interrupts.
12158 * For lowest-priority interrupts, we only support
12159 * those with single CPU as the destination, e.g. user
12160 * configures the interrupts via /proc/irq or uses
12161 * irqbalance to make the interrupts single-CPU.
12163 * We will support full lowest-priority interrupt later.
12166 kvm_set_msi_irq(kvm, e, &irq);
12167 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
12169 * Make sure the IRTE is in remapped mode if
12170 * we don't handle it in posted mode.
12172 ret = irq_set_vcpu_affinity(host_irq, NULL);
12175 "failed to back to remapped mode, irq: %u\n",
12183 vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
12184 vcpu_info.vector = irq.vector;
12186 trace_kvm_pi_irte_update(host_irq, vcpu->vcpu_id, e->gsi,
12187 vcpu_info.vector, vcpu_info.pi_desc_addr, set);
12190 ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
12192 ret = irq_set_vcpu_affinity(host_irq, NULL);
12195 printk(KERN_INFO "%s: failed to update PI IRTE\n",
12203 srcu_read_unlock(&kvm->irq_srcu, idx);
12207 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
12209 if (vcpu->arch.mcg_cap & MCG_LMCE_P)
12210 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
12211 FEATURE_CONTROL_LMCE;
12213 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
12214 ~FEATURE_CONTROL_LMCE;
12217 static struct kvm_x86_ops vmx_x86_ops __ro_after_init = {
12218 .cpu_has_kvm_support = cpu_has_kvm_support,
12219 .disabled_by_bios = vmx_disabled_by_bios,
12220 .hardware_setup = hardware_setup,
12221 .hardware_unsetup = hardware_unsetup,
12222 .check_processor_compatibility = vmx_check_processor_compat,
12223 .hardware_enable = hardware_enable,
12224 .hardware_disable = hardware_disable,
12225 .cpu_has_accelerated_tpr = report_flexpriority,
12226 .has_emulated_msr = vmx_has_emulated_msr,
12228 .vcpu_create = vmx_create_vcpu,
12229 .vcpu_free = vmx_free_vcpu,
12230 .vcpu_reset = vmx_vcpu_reset,
12232 .prepare_guest_switch = vmx_save_host_state,
12233 .vcpu_load = vmx_vcpu_load,
12234 .vcpu_put = vmx_vcpu_put,
12236 .update_bp_intercept = update_exception_bitmap,
12237 .get_msr = vmx_get_msr,
12238 .set_msr = vmx_set_msr,
12239 .get_segment_base = vmx_get_segment_base,
12240 .get_segment = vmx_get_segment,
12241 .set_segment = vmx_set_segment,
12242 .get_cpl = vmx_get_cpl,
12243 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
12244 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
12245 .decache_cr3 = vmx_decache_cr3,
12246 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
12247 .set_cr0 = vmx_set_cr0,
12248 .set_cr3 = vmx_set_cr3,
12249 .set_cr4 = vmx_set_cr4,
12250 .set_efer = vmx_set_efer,
12251 .get_idt = vmx_get_idt,
12252 .set_idt = vmx_set_idt,
12253 .get_gdt = vmx_get_gdt,
12254 .set_gdt = vmx_set_gdt,
12255 .get_dr6 = vmx_get_dr6,
12256 .set_dr6 = vmx_set_dr6,
12257 .set_dr7 = vmx_set_dr7,
12258 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
12259 .cache_reg = vmx_cache_reg,
12260 .get_rflags = vmx_get_rflags,
12261 .set_rflags = vmx_set_rflags,
12263 .tlb_flush = vmx_flush_tlb,
12265 .run = vmx_vcpu_run,
12266 .handle_exit = vmx_handle_exit,
12267 .skip_emulated_instruction = skip_emulated_instruction,
12268 .set_interrupt_shadow = vmx_set_interrupt_shadow,
12269 .get_interrupt_shadow = vmx_get_interrupt_shadow,
12270 .patch_hypercall = vmx_patch_hypercall,
12271 .set_irq = vmx_inject_irq,
12272 .set_nmi = vmx_inject_nmi,
12273 .queue_exception = vmx_queue_exception,
12274 .cancel_injection = vmx_cancel_injection,
12275 .interrupt_allowed = vmx_interrupt_allowed,
12276 .nmi_allowed = vmx_nmi_allowed,
12277 .get_nmi_mask = vmx_get_nmi_mask,
12278 .set_nmi_mask = vmx_set_nmi_mask,
12279 .enable_nmi_window = enable_nmi_window,
12280 .enable_irq_window = enable_irq_window,
12281 .update_cr8_intercept = update_cr8_intercept,
12282 .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode,
12283 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
12284 .get_enable_apicv = vmx_get_enable_apicv,
12285 .refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
12286 .load_eoi_exitmap = vmx_load_eoi_exitmap,
12287 .apicv_post_state_restore = vmx_apicv_post_state_restore,
12288 .hwapic_irr_update = vmx_hwapic_irr_update,
12289 .hwapic_isr_update = vmx_hwapic_isr_update,
12290 .sync_pir_to_irr = vmx_sync_pir_to_irr,
12291 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
12293 .set_tss_addr = vmx_set_tss_addr,
12294 .get_tdp_level = get_ept_level,
12295 .get_mt_mask = vmx_get_mt_mask,
12297 .get_exit_info = vmx_get_exit_info,
12299 .get_lpage_level = vmx_get_lpage_level,
12301 .cpuid_update = vmx_cpuid_update,
12303 .rdtscp_supported = vmx_rdtscp_supported,
12304 .invpcid_supported = vmx_invpcid_supported,
12306 .set_supported_cpuid = vmx_set_supported_cpuid,
12308 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
12310 .write_tsc_offset = vmx_write_tsc_offset,
12312 .set_tdp_cr3 = vmx_set_cr3,
12314 .check_intercept = vmx_check_intercept,
12315 .handle_external_intr = vmx_handle_external_intr,
12316 .mpx_supported = vmx_mpx_supported,
12317 .xsaves_supported = vmx_xsaves_supported,
12319 .check_nested_events = vmx_check_nested_events,
12321 .sched_in = vmx_sched_in,
12323 .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
12324 .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
12325 .flush_log_dirty = vmx_flush_log_dirty,
12326 .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
12327 .write_log_dirty = vmx_write_pml_buffer,
12329 .pre_block = vmx_pre_block,
12330 .post_block = vmx_post_block,
12332 .pmu_ops = &intel_pmu_ops,
12334 .update_pi_irte = vmx_update_pi_irte,
12336 #ifdef CONFIG_X86_64
12337 .set_hv_timer = vmx_set_hv_timer,
12338 .cancel_hv_timer = vmx_cancel_hv_timer,
12341 .setup_mce = vmx_setup_mce,
12344 static int __init vmx_init(void)
12346 int r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
12347 __alignof__(struct vcpu_vmx), THIS_MODULE);
12351 #ifdef CONFIG_KEXEC_CORE
12352 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
12353 crash_vmclear_local_loaded_vmcss);
12359 static void __exit vmx_exit(void)
12361 #ifdef CONFIG_KEXEC_CORE
12362 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
12369 module_init(vmx_init)
12370 module_exit(vmx_exit)