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/nospec-branch.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 enable_vnmi = 1;
77 module_param_named(vnmi, enable_vnmi, bool, S_IRUGO);
79 static bool __read_mostly flexpriority_enabled = 1;
80 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
82 static bool __read_mostly enable_ept = 1;
83 module_param_named(ept, enable_ept, bool, S_IRUGO);
85 static bool __read_mostly enable_unrestricted_guest = 1;
86 module_param_named(unrestricted_guest,
87 enable_unrestricted_guest, bool, S_IRUGO);
89 static bool __read_mostly enable_ept_ad_bits = 1;
90 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
92 static bool __read_mostly emulate_invalid_guest_state = true;
93 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
95 static bool __read_mostly fasteoi = 1;
96 module_param(fasteoi, bool, S_IRUGO);
98 static bool __read_mostly enable_apicv = 1;
99 module_param(enable_apicv, bool, S_IRUGO);
101 static bool __read_mostly enable_shadow_vmcs = 1;
102 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
104 * If nested=1, nested virtualization is supported, i.e., guests may use
105 * VMX and be a hypervisor for its own guests. If nested=0, guests may not
106 * use VMX instructions.
108 static bool __read_mostly nested = 0;
109 module_param(nested, bool, S_IRUGO);
111 static u64 __read_mostly host_xss;
113 static bool __read_mostly enable_pml = 1;
114 module_param_named(pml, enable_pml, bool, S_IRUGO);
118 #define MSR_TYPE_RW 3
120 #define MSR_BITMAP_MODE_X2APIC 1
121 #define MSR_BITMAP_MODE_X2APIC_APICV 2
122 #define MSR_BITMAP_MODE_LM 4
124 #define KVM_VMX_TSC_MULTIPLIER_MAX 0xffffffffffffffffULL
126 /* Guest_tsc -> host_tsc conversion requires 64-bit division. */
127 static int __read_mostly cpu_preemption_timer_multi;
128 static bool __read_mostly enable_preemption_timer = 1;
130 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
133 #define KVM_GUEST_CR0_MASK (X86_CR0_NW | X86_CR0_CD)
134 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST (X86_CR0_WP | X86_CR0_NE)
135 #define KVM_VM_CR0_ALWAYS_ON \
136 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
137 #define KVM_CR4_GUEST_OWNED_BITS \
138 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
139 | X86_CR4_OSXMMEXCPT | X86_CR4_LA57 | X86_CR4_TSD)
141 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
142 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
144 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
146 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
149 * Hyper-V requires all of these, so mark them as supported even though
150 * they are just treated the same as all-context.
152 #define VMX_VPID_EXTENT_SUPPORTED_MASK \
153 (VMX_VPID_EXTENT_INDIVIDUAL_ADDR_BIT | \
154 VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT | \
155 VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT | \
156 VMX_VPID_EXTENT_SINGLE_NON_GLOBAL_BIT)
159 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
160 * ple_gap: upper bound on the amount of time between two successive
161 * executions of PAUSE in a loop. Also indicate if ple enabled.
162 * According to test, this time is usually smaller than 128 cycles.
163 * ple_window: upper bound on the amount of time a guest is allowed to execute
164 * in a PAUSE loop. Tests indicate that most spinlocks are held for
165 * less than 2^12 cycles
166 * Time is measured based on a counter that runs at the same rate as the TSC,
167 * refer SDM volume 3b section 21.6.13 & 22.1.3.
169 #define KVM_VMX_DEFAULT_PLE_GAP 128
170 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
171 #define KVM_VMX_DEFAULT_PLE_WINDOW_GROW 2
172 #define KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK 0
173 #define KVM_VMX_DEFAULT_PLE_WINDOW_MAX \
174 INT_MAX / KVM_VMX_DEFAULT_PLE_WINDOW_GROW
176 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
177 module_param(ple_gap, int, S_IRUGO);
179 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
180 module_param(ple_window, int, S_IRUGO);
182 /* Default doubles per-vcpu window every exit. */
183 static int ple_window_grow = KVM_VMX_DEFAULT_PLE_WINDOW_GROW;
184 module_param(ple_window_grow, int, S_IRUGO);
186 /* Default resets per-vcpu window every exit to ple_window. */
187 static int ple_window_shrink = KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK;
188 module_param(ple_window_shrink, int, S_IRUGO);
190 /* Default is to compute the maximum so we can never overflow. */
191 static int ple_window_actual_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
192 static int ple_window_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
193 module_param(ple_window_max, int, S_IRUGO);
195 extern const ulong vmx_return;
197 #define NR_AUTOLOAD_MSRS 8
206 * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
207 * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
208 * loaded on this CPU (so we can clear them if the CPU goes down).
212 struct vmcs *shadow_vmcs;
215 bool nmi_known_unmasked;
216 unsigned long vmcs_host_cr3; /* May not match real cr3 */
217 unsigned long vmcs_host_cr4; /* May not match real cr4 */
218 /* Support for vnmi-less CPUs */
219 int soft_vnmi_blocked;
221 s64 vnmi_blocked_time;
222 unsigned long *msr_bitmap;
223 struct list_head loaded_vmcss_on_cpu_link;
226 struct shared_msr_entry {
233 * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
234 * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
235 * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
236 * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
237 * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
238 * More than one of these structures may exist, if L1 runs multiple L2 guests.
239 * nested_vmx_run() will use the data here to build the vmcs02: a VMCS for the
240 * underlying hardware which will be used to run L2.
241 * This structure is packed to ensure that its layout is identical across
242 * machines (necessary for live migration).
243 * If there are changes in this struct, VMCS12_REVISION must be changed.
245 typedef u64 natural_width;
246 struct __packed vmcs12 {
247 /* According to the Intel spec, a VMCS region must start with the
248 * following two fields. Then follow implementation-specific data.
253 u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
254 u32 padding[7]; /* room for future expansion */
259 u64 vm_exit_msr_store_addr;
260 u64 vm_exit_msr_load_addr;
261 u64 vm_entry_msr_load_addr;
263 u64 virtual_apic_page_addr;
264 u64 apic_access_addr;
265 u64 posted_intr_desc_addr;
266 u64 vm_function_control;
268 u64 eoi_exit_bitmap0;
269 u64 eoi_exit_bitmap1;
270 u64 eoi_exit_bitmap2;
271 u64 eoi_exit_bitmap3;
272 u64 eptp_list_address;
274 u64 guest_physical_address;
275 u64 vmcs_link_pointer;
277 u64 guest_ia32_debugctl;
280 u64 guest_ia32_perf_global_ctrl;
288 u64 host_ia32_perf_global_ctrl;
289 u64 padding64[8]; /* room for future expansion */
291 * To allow migration of L1 (complete with its L2 guests) between
292 * machines of different natural widths (32 or 64 bit), we cannot have
293 * unsigned long fields with no explict size. We use u64 (aliased
294 * natural_width) instead. Luckily, x86 is little-endian.
296 natural_width cr0_guest_host_mask;
297 natural_width cr4_guest_host_mask;
298 natural_width cr0_read_shadow;
299 natural_width cr4_read_shadow;
300 natural_width cr3_target_value0;
301 natural_width cr3_target_value1;
302 natural_width cr3_target_value2;
303 natural_width cr3_target_value3;
304 natural_width exit_qualification;
305 natural_width guest_linear_address;
306 natural_width guest_cr0;
307 natural_width guest_cr3;
308 natural_width guest_cr4;
309 natural_width guest_es_base;
310 natural_width guest_cs_base;
311 natural_width guest_ss_base;
312 natural_width guest_ds_base;
313 natural_width guest_fs_base;
314 natural_width guest_gs_base;
315 natural_width guest_ldtr_base;
316 natural_width guest_tr_base;
317 natural_width guest_gdtr_base;
318 natural_width guest_idtr_base;
319 natural_width guest_dr7;
320 natural_width guest_rsp;
321 natural_width guest_rip;
322 natural_width guest_rflags;
323 natural_width guest_pending_dbg_exceptions;
324 natural_width guest_sysenter_esp;
325 natural_width guest_sysenter_eip;
326 natural_width host_cr0;
327 natural_width host_cr3;
328 natural_width host_cr4;
329 natural_width host_fs_base;
330 natural_width host_gs_base;
331 natural_width host_tr_base;
332 natural_width host_gdtr_base;
333 natural_width host_idtr_base;
334 natural_width host_ia32_sysenter_esp;
335 natural_width host_ia32_sysenter_eip;
336 natural_width host_rsp;
337 natural_width host_rip;
338 natural_width paddingl[8]; /* room for future expansion */
339 u32 pin_based_vm_exec_control;
340 u32 cpu_based_vm_exec_control;
341 u32 exception_bitmap;
342 u32 page_fault_error_code_mask;
343 u32 page_fault_error_code_match;
344 u32 cr3_target_count;
345 u32 vm_exit_controls;
346 u32 vm_exit_msr_store_count;
347 u32 vm_exit_msr_load_count;
348 u32 vm_entry_controls;
349 u32 vm_entry_msr_load_count;
350 u32 vm_entry_intr_info_field;
351 u32 vm_entry_exception_error_code;
352 u32 vm_entry_instruction_len;
354 u32 secondary_vm_exec_control;
355 u32 vm_instruction_error;
357 u32 vm_exit_intr_info;
358 u32 vm_exit_intr_error_code;
359 u32 idt_vectoring_info_field;
360 u32 idt_vectoring_error_code;
361 u32 vm_exit_instruction_len;
362 u32 vmx_instruction_info;
369 u32 guest_ldtr_limit;
371 u32 guest_gdtr_limit;
372 u32 guest_idtr_limit;
373 u32 guest_es_ar_bytes;
374 u32 guest_cs_ar_bytes;
375 u32 guest_ss_ar_bytes;
376 u32 guest_ds_ar_bytes;
377 u32 guest_fs_ar_bytes;
378 u32 guest_gs_ar_bytes;
379 u32 guest_ldtr_ar_bytes;
380 u32 guest_tr_ar_bytes;
381 u32 guest_interruptibility_info;
382 u32 guest_activity_state;
383 u32 guest_sysenter_cs;
384 u32 host_ia32_sysenter_cs;
385 u32 vmx_preemption_timer_value;
386 u32 padding32[7]; /* room for future expansion */
387 u16 virtual_processor_id;
389 u16 guest_es_selector;
390 u16 guest_cs_selector;
391 u16 guest_ss_selector;
392 u16 guest_ds_selector;
393 u16 guest_fs_selector;
394 u16 guest_gs_selector;
395 u16 guest_ldtr_selector;
396 u16 guest_tr_selector;
397 u16 guest_intr_status;
399 u16 host_es_selector;
400 u16 host_cs_selector;
401 u16 host_ss_selector;
402 u16 host_ds_selector;
403 u16 host_fs_selector;
404 u16 host_gs_selector;
405 u16 host_tr_selector;
409 * VMCS12_REVISION is an arbitrary id that should be changed if the content or
410 * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
411 * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
413 #define VMCS12_REVISION 0x11e57ed0
416 * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
417 * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
418 * current implementation, 4K are reserved to avoid future complications.
420 #define VMCS12_SIZE 0x1000
423 * VMCS12_MAX_FIELD_INDEX is the highest index value used in any
424 * supported VMCS12 field encoding.
426 #define VMCS12_MAX_FIELD_INDEX 0x17
429 * The nested_vmx structure is part of vcpu_vmx, and holds information we need
430 * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
433 /* Has the level1 guest done vmxon? */
438 /* The guest-physical address of the current VMCS L1 keeps for L2 */
441 * Cache of the guest's VMCS, existing outside of guest memory.
442 * Loaded from guest memory during VMPTRLD. Flushed to guest
443 * memory during VMCLEAR and VMPTRLD.
445 struct vmcs12 *cached_vmcs12;
447 * Indicates if the shadow vmcs must be updated with the
448 * data hold by vmcs12
450 bool sync_shadow_vmcs;
453 bool change_vmcs01_virtual_x2apic_mode;
454 /* L2 must run next, and mustn't decide to exit to L1. */
455 bool nested_run_pending;
457 struct loaded_vmcs vmcs02;
460 * Guest pages referred to in the vmcs02 with host-physical
461 * pointers, so we must keep them pinned while L2 runs.
463 struct page *apic_access_page;
464 struct page *virtual_apic_page;
465 struct page *pi_desc_page;
466 struct pi_desc *pi_desc;
470 struct hrtimer preemption_timer;
471 bool preemption_timer_expired;
473 /* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
480 * We only store the "true" versions of the VMX capability MSRs. We
481 * generate the "non-true" versions by setting the must-be-1 bits
482 * according to the SDM.
484 u32 nested_vmx_procbased_ctls_low;
485 u32 nested_vmx_procbased_ctls_high;
486 u32 nested_vmx_secondary_ctls_low;
487 u32 nested_vmx_secondary_ctls_high;
488 u32 nested_vmx_pinbased_ctls_low;
489 u32 nested_vmx_pinbased_ctls_high;
490 u32 nested_vmx_exit_ctls_low;
491 u32 nested_vmx_exit_ctls_high;
492 u32 nested_vmx_entry_ctls_low;
493 u32 nested_vmx_entry_ctls_high;
494 u32 nested_vmx_misc_low;
495 u32 nested_vmx_misc_high;
496 u32 nested_vmx_ept_caps;
497 u32 nested_vmx_vpid_caps;
498 u64 nested_vmx_basic;
499 u64 nested_vmx_cr0_fixed0;
500 u64 nested_vmx_cr0_fixed1;
501 u64 nested_vmx_cr4_fixed0;
502 u64 nested_vmx_cr4_fixed1;
503 u64 nested_vmx_vmcs_enum;
504 u64 nested_vmx_vmfunc_controls;
506 /* SMM related state */
508 /* in VMX operation on SMM entry? */
510 /* in guest mode on SMM entry? */
515 #define POSTED_INTR_ON 0
516 #define POSTED_INTR_SN 1
518 /* Posted-Interrupt Descriptor */
520 u32 pir[8]; /* Posted interrupt requested */
523 /* bit 256 - Outstanding Notification */
525 /* bit 257 - Suppress Notification */
527 /* bit 271:258 - Reserved */
529 /* bit 279:272 - Notification Vector */
531 /* bit 287:280 - Reserved */
533 /* bit 319:288 - Notification Destination */
541 static bool pi_test_and_set_on(struct pi_desc *pi_desc)
543 return test_and_set_bit(POSTED_INTR_ON,
544 (unsigned long *)&pi_desc->control);
547 static bool pi_test_and_clear_on(struct pi_desc *pi_desc)
549 return test_and_clear_bit(POSTED_INTR_ON,
550 (unsigned long *)&pi_desc->control);
553 static int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
555 return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
558 static inline void pi_clear_sn(struct pi_desc *pi_desc)
560 return clear_bit(POSTED_INTR_SN,
561 (unsigned long *)&pi_desc->control);
564 static inline void pi_set_sn(struct pi_desc *pi_desc)
566 return set_bit(POSTED_INTR_SN,
567 (unsigned long *)&pi_desc->control);
570 static inline void pi_clear_on(struct pi_desc *pi_desc)
572 clear_bit(POSTED_INTR_ON,
573 (unsigned long *)&pi_desc->control);
576 static inline int pi_test_on(struct pi_desc *pi_desc)
578 return test_bit(POSTED_INTR_ON,
579 (unsigned long *)&pi_desc->control);
582 static inline int pi_test_sn(struct pi_desc *pi_desc)
584 return test_bit(POSTED_INTR_SN,
585 (unsigned long *)&pi_desc->control);
589 struct kvm_vcpu vcpu;
590 unsigned long host_rsp;
594 u32 idt_vectoring_info;
596 struct shared_msr_entry *guest_msrs;
599 unsigned long host_idt_base;
601 u64 msr_host_kernel_gs_base;
602 u64 msr_guest_kernel_gs_base;
605 u64 arch_capabilities;
608 u32 vm_entry_controls_shadow;
609 u32 vm_exit_controls_shadow;
610 u32 secondary_exec_control;
613 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
614 * non-nested (L1) guest, it always points to vmcs01. For a nested
615 * guest (L2), it points to a different VMCS.
617 struct loaded_vmcs vmcs01;
618 struct loaded_vmcs *loaded_vmcs;
619 bool __launched; /* temporary, used in vmx_vcpu_run */
620 struct msr_autoload {
622 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
623 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
627 u16 fs_sel, gs_sel, ldt_sel;
631 int gs_ldt_reload_needed;
632 int fs_reload_needed;
633 u64 msr_host_bndcfgs;
638 struct kvm_segment segs[8];
641 u32 bitmask; /* 4 bits per segment (1 bit per field) */
642 struct kvm_save_segment {
650 bool emulation_required;
654 /* Posted interrupt descriptor */
655 struct pi_desc pi_desc;
657 /* Support for a guest hypervisor (nested VMX) */
658 struct nested_vmx nested;
660 /* Dynamic PLE window. */
662 bool ple_window_dirty;
664 /* Support for PML */
665 #define PML_ENTITY_NUM 512
668 /* apic deadline value in host tsc */
671 u64 current_tsc_ratio;
675 unsigned long host_debugctlmsr;
678 * Only bits masked by msr_ia32_feature_control_valid_bits can be set in
679 * msr_ia32_feature_control. FEATURE_CONTROL_LOCKED is always included
680 * in msr_ia32_feature_control_valid_bits.
682 u64 msr_ia32_feature_control;
683 u64 msr_ia32_feature_control_valid_bits;
686 enum segment_cache_field {
695 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
697 return container_of(vcpu, struct vcpu_vmx, vcpu);
700 static struct pi_desc *vcpu_to_pi_desc(struct kvm_vcpu *vcpu)
702 return &(to_vmx(vcpu)->pi_desc);
705 #define ROL16(val, n) ((u16)(((u16)(val) << (n)) | ((u16)(val) >> (16 - (n)))))
706 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
707 #define FIELD(number, name) [ROL16(number, 6)] = VMCS12_OFFSET(name)
708 #define FIELD64(number, name) \
709 FIELD(number, name), \
710 [ROL16(number##_HIGH, 6)] = VMCS12_OFFSET(name) + sizeof(u32)
713 static u16 shadow_read_only_fields[] = {
714 #define SHADOW_FIELD_RO(x) x,
715 #include "vmx_shadow_fields.h"
717 static int max_shadow_read_only_fields =
718 ARRAY_SIZE(shadow_read_only_fields);
720 static u16 shadow_read_write_fields[] = {
721 #define SHADOW_FIELD_RW(x) x,
722 #include "vmx_shadow_fields.h"
724 static int max_shadow_read_write_fields =
725 ARRAY_SIZE(shadow_read_write_fields);
727 static const unsigned short vmcs_field_to_offset_table[] = {
728 FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
729 FIELD(POSTED_INTR_NV, posted_intr_nv),
730 FIELD(GUEST_ES_SELECTOR, guest_es_selector),
731 FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
732 FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
733 FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
734 FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
735 FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
736 FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
737 FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
738 FIELD(GUEST_INTR_STATUS, guest_intr_status),
739 FIELD(GUEST_PML_INDEX, guest_pml_index),
740 FIELD(HOST_ES_SELECTOR, host_es_selector),
741 FIELD(HOST_CS_SELECTOR, host_cs_selector),
742 FIELD(HOST_SS_SELECTOR, host_ss_selector),
743 FIELD(HOST_DS_SELECTOR, host_ds_selector),
744 FIELD(HOST_FS_SELECTOR, host_fs_selector),
745 FIELD(HOST_GS_SELECTOR, host_gs_selector),
746 FIELD(HOST_TR_SELECTOR, host_tr_selector),
747 FIELD64(IO_BITMAP_A, io_bitmap_a),
748 FIELD64(IO_BITMAP_B, io_bitmap_b),
749 FIELD64(MSR_BITMAP, msr_bitmap),
750 FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
751 FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
752 FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
753 FIELD64(TSC_OFFSET, tsc_offset),
754 FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
755 FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
756 FIELD64(POSTED_INTR_DESC_ADDR, posted_intr_desc_addr),
757 FIELD64(VM_FUNCTION_CONTROL, vm_function_control),
758 FIELD64(EPT_POINTER, ept_pointer),
759 FIELD64(EOI_EXIT_BITMAP0, eoi_exit_bitmap0),
760 FIELD64(EOI_EXIT_BITMAP1, eoi_exit_bitmap1),
761 FIELD64(EOI_EXIT_BITMAP2, eoi_exit_bitmap2),
762 FIELD64(EOI_EXIT_BITMAP3, eoi_exit_bitmap3),
763 FIELD64(EPTP_LIST_ADDRESS, eptp_list_address),
764 FIELD64(XSS_EXIT_BITMAP, xss_exit_bitmap),
765 FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
766 FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
767 FIELD64(PML_ADDRESS, pml_address),
768 FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
769 FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
770 FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
771 FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
772 FIELD64(GUEST_PDPTR0, guest_pdptr0),
773 FIELD64(GUEST_PDPTR1, guest_pdptr1),
774 FIELD64(GUEST_PDPTR2, guest_pdptr2),
775 FIELD64(GUEST_PDPTR3, guest_pdptr3),
776 FIELD64(GUEST_BNDCFGS, guest_bndcfgs),
777 FIELD64(HOST_IA32_PAT, host_ia32_pat),
778 FIELD64(HOST_IA32_EFER, host_ia32_efer),
779 FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
780 FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
781 FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
782 FIELD(EXCEPTION_BITMAP, exception_bitmap),
783 FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
784 FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
785 FIELD(CR3_TARGET_COUNT, cr3_target_count),
786 FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
787 FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
788 FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
789 FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
790 FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
791 FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
792 FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
793 FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
794 FIELD(TPR_THRESHOLD, tpr_threshold),
795 FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
796 FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
797 FIELD(VM_EXIT_REASON, vm_exit_reason),
798 FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
799 FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
800 FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
801 FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
802 FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
803 FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
804 FIELD(GUEST_ES_LIMIT, guest_es_limit),
805 FIELD(GUEST_CS_LIMIT, guest_cs_limit),
806 FIELD(GUEST_SS_LIMIT, guest_ss_limit),
807 FIELD(GUEST_DS_LIMIT, guest_ds_limit),
808 FIELD(GUEST_FS_LIMIT, guest_fs_limit),
809 FIELD(GUEST_GS_LIMIT, guest_gs_limit),
810 FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
811 FIELD(GUEST_TR_LIMIT, guest_tr_limit),
812 FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
813 FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
814 FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
815 FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
816 FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
817 FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
818 FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
819 FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
820 FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
821 FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
822 FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
823 FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
824 FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
825 FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
826 FIELD(VMX_PREEMPTION_TIMER_VALUE, vmx_preemption_timer_value),
827 FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
828 FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
829 FIELD(CR0_READ_SHADOW, cr0_read_shadow),
830 FIELD(CR4_READ_SHADOW, cr4_read_shadow),
831 FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
832 FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
833 FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
834 FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
835 FIELD(EXIT_QUALIFICATION, exit_qualification),
836 FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
837 FIELD(GUEST_CR0, guest_cr0),
838 FIELD(GUEST_CR3, guest_cr3),
839 FIELD(GUEST_CR4, guest_cr4),
840 FIELD(GUEST_ES_BASE, guest_es_base),
841 FIELD(GUEST_CS_BASE, guest_cs_base),
842 FIELD(GUEST_SS_BASE, guest_ss_base),
843 FIELD(GUEST_DS_BASE, guest_ds_base),
844 FIELD(GUEST_FS_BASE, guest_fs_base),
845 FIELD(GUEST_GS_BASE, guest_gs_base),
846 FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
847 FIELD(GUEST_TR_BASE, guest_tr_base),
848 FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
849 FIELD(GUEST_IDTR_BASE, guest_idtr_base),
850 FIELD(GUEST_DR7, guest_dr7),
851 FIELD(GUEST_RSP, guest_rsp),
852 FIELD(GUEST_RIP, guest_rip),
853 FIELD(GUEST_RFLAGS, guest_rflags),
854 FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
855 FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
856 FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
857 FIELD(HOST_CR0, host_cr0),
858 FIELD(HOST_CR3, host_cr3),
859 FIELD(HOST_CR4, host_cr4),
860 FIELD(HOST_FS_BASE, host_fs_base),
861 FIELD(HOST_GS_BASE, host_gs_base),
862 FIELD(HOST_TR_BASE, host_tr_base),
863 FIELD(HOST_GDTR_BASE, host_gdtr_base),
864 FIELD(HOST_IDTR_BASE, host_idtr_base),
865 FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
866 FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
867 FIELD(HOST_RSP, host_rsp),
868 FIELD(HOST_RIP, host_rip),
871 static inline short vmcs_field_to_offset(unsigned long field)
873 const size_t size = ARRAY_SIZE(vmcs_field_to_offset_table);
874 unsigned short offset;
880 index = ROL16(field, 6);
884 index = array_index_nospec(index, size);
885 offset = vmcs_field_to_offset_table[index];
891 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
893 return to_vmx(vcpu)->nested.cached_vmcs12;
896 static bool nested_ept_ad_enabled(struct kvm_vcpu *vcpu);
897 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu);
898 static u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa);
899 static bool vmx_xsaves_supported(void);
900 static void vmx_set_segment(struct kvm_vcpu *vcpu,
901 struct kvm_segment *var, int seg);
902 static void vmx_get_segment(struct kvm_vcpu *vcpu,
903 struct kvm_segment *var, int seg);
904 static bool guest_state_valid(struct kvm_vcpu *vcpu);
905 static u32 vmx_segment_access_rights(struct kvm_segment *var);
906 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx);
907 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu);
908 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked);
909 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
911 static void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu);
912 static void __always_inline vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
915 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
916 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
918 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
919 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
921 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
924 * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
925 * can find which vCPU should be waken up.
927 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
928 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
936 static unsigned long *vmx_bitmap[VMX_BITMAP_NR];
938 #define vmx_vmread_bitmap (vmx_bitmap[VMX_VMREAD_BITMAP])
939 #define vmx_vmwrite_bitmap (vmx_bitmap[VMX_VMWRITE_BITMAP])
941 static bool cpu_has_load_ia32_efer;
942 static bool cpu_has_load_perf_global_ctrl;
944 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
945 static DEFINE_SPINLOCK(vmx_vpid_lock);
947 static struct vmcs_config {
952 u32 pin_based_exec_ctrl;
953 u32 cpu_based_exec_ctrl;
954 u32 cpu_based_2nd_exec_ctrl;
959 static struct vmx_capability {
964 #define VMX_SEGMENT_FIELD(seg) \
965 [VCPU_SREG_##seg] = { \
966 .selector = GUEST_##seg##_SELECTOR, \
967 .base = GUEST_##seg##_BASE, \
968 .limit = GUEST_##seg##_LIMIT, \
969 .ar_bytes = GUEST_##seg##_AR_BYTES, \
972 static const struct kvm_vmx_segment_field {
977 } kvm_vmx_segment_fields[] = {
978 VMX_SEGMENT_FIELD(CS),
979 VMX_SEGMENT_FIELD(DS),
980 VMX_SEGMENT_FIELD(ES),
981 VMX_SEGMENT_FIELD(FS),
982 VMX_SEGMENT_FIELD(GS),
983 VMX_SEGMENT_FIELD(SS),
984 VMX_SEGMENT_FIELD(TR),
985 VMX_SEGMENT_FIELD(LDTR),
988 static u64 host_efer;
990 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
993 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
994 * away by decrementing the array size.
996 static const u32 vmx_msr_index[] = {
998 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
1000 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
1003 static inline bool is_exception_n(u32 intr_info, u8 vector)
1005 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
1006 INTR_INFO_VALID_MASK)) ==
1007 (INTR_TYPE_HARD_EXCEPTION | vector | INTR_INFO_VALID_MASK);
1010 static inline bool is_debug(u32 intr_info)
1012 return is_exception_n(intr_info, DB_VECTOR);
1015 static inline bool is_breakpoint(u32 intr_info)
1017 return is_exception_n(intr_info, BP_VECTOR);
1020 static inline bool is_page_fault(u32 intr_info)
1022 return is_exception_n(intr_info, PF_VECTOR);
1025 static inline bool is_no_device(u32 intr_info)
1027 return is_exception_n(intr_info, NM_VECTOR);
1030 static inline bool is_invalid_opcode(u32 intr_info)
1032 return is_exception_n(intr_info, UD_VECTOR);
1035 static inline bool is_external_interrupt(u32 intr_info)
1037 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1038 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
1041 static inline bool is_machine_check(u32 intr_info)
1043 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
1044 INTR_INFO_VALID_MASK)) ==
1045 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
1048 static inline bool cpu_has_vmx_msr_bitmap(void)
1050 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
1053 static inline bool cpu_has_vmx_tpr_shadow(void)
1055 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
1058 static inline bool cpu_need_tpr_shadow(struct kvm_vcpu *vcpu)
1060 return cpu_has_vmx_tpr_shadow() && lapic_in_kernel(vcpu);
1063 static inline bool cpu_has_secondary_exec_ctrls(void)
1065 return vmcs_config.cpu_based_exec_ctrl &
1066 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1069 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
1071 return vmcs_config.cpu_based_2nd_exec_ctrl &
1072 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1075 static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
1077 return vmcs_config.cpu_based_2nd_exec_ctrl &
1078 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
1081 static inline bool cpu_has_vmx_apic_register_virt(void)
1083 return vmcs_config.cpu_based_2nd_exec_ctrl &
1084 SECONDARY_EXEC_APIC_REGISTER_VIRT;
1087 static inline bool cpu_has_vmx_virtual_intr_delivery(void)
1089 return vmcs_config.cpu_based_2nd_exec_ctrl &
1090 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
1094 * Comment's format: document - errata name - stepping - processor name.
1096 * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
1098 static u32 vmx_preemption_cpu_tfms[] = {
1099 /* 323344.pdf - BA86 - D0 - Xeon 7500 Series */
1101 /* 323056.pdf - AAX65 - C2 - Xeon L3406 */
1102 /* 322814.pdf - AAT59 - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
1103 /* 322911.pdf - AAU65 - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
1105 /* 322911.pdf - AAU65 - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
1107 /* 322373.pdf - AAO95 - B1 - Xeon 3400 Series */
1108 /* 322166.pdf - AAN92 - B1 - i7-800 and i5-700 Desktop */
1110 * 320767.pdf - AAP86 - B1 -
1111 * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
1114 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
1116 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
1118 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
1120 /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
1121 /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
1122 /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
1126 static inline bool cpu_has_broken_vmx_preemption_timer(void)
1128 u32 eax = cpuid_eax(0x00000001), i;
1130 /* Clear the reserved bits */
1131 eax &= ~(0x3U << 14 | 0xfU << 28);
1132 for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
1133 if (eax == vmx_preemption_cpu_tfms[i])
1139 static inline bool cpu_has_vmx_preemption_timer(void)
1141 return vmcs_config.pin_based_exec_ctrl &
1142 PIN_BASED_VMX_PREEMPTION_TIMER;
1145 static inline bool cpu_has_vmx_posted_intr(void)
1147 return IS_ENABLED(CONFIG_X86_LOCAL_APIC) &&
1148 vmcs_config.pin_based_exec_ctrl & PIN_BASED_POSTED_INTR;
1151 static inline bool cpu_has_vmx_apicv(void)
1153 return cpu_has_vmx_apic_register_virt() &&
1154 cpu_has_vmx_virtual_intr_delivery() &&
1155 cpu_has_vmx_posted_intr();
1158 static inline bool cpu_has_vmx_flexpriority(void)
1160 return cpu_has_vmx_tpr_shadow() &&
1161 cpu_has_vmx_virtualize_apic_accesses();
1164 static inline bool cpu_has_vmx_ept_execute_only(void)
1166 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
1169 static inline bool cpu_has_vmx_ept_2m_page(void)
1171 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
1174 static inline bool cpu_has_vmx_ept_1g_page(void)
1176 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
1179 static inline bool cpu_has_vmx_ept_4levels(void)
1181 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
1184 static inline bool cpu_has_vmx_ept_mt_wb(void)
1186 return vmx_capability.ept & VMX_EPTP_WB_BIT;
1189 static inline bool cpu_has_vmx_ept_5levels(void)
1191 return vmx_capability.ept & VMX_EPT_PAGE_WALK_5_BIT;
1194 static inline bool cpu_has_vmx_ept_ad_bits(void)
1196 return vmx_capability.ept & VMX_EPT_AD_BIT;
1199 static inline bool cpu_has_vmx_invept_context(void)
1201 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
1204 static inline bool cpu_has_vmx_invept_global(void)
1206 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
1209 static inline bool cpu_has_vmx_invvpid_single(void)
1211 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
1214 static inline bool cpu_has_vmx_invvpid_global(void)
1216 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
1219 static inline bool cpu_has_vmx_invvpid(void)
1221 return vmx_capability.vpid & VMX_VPID_INVVPID_BIT;
1224 static inline bool cpu_has_vmx_ept(void)
1226 return vmcs_config.cpu_based_2nd_exec_ctrl &
1227 SECONDARY_EXEC_ENABLE_EPT;
1230 static inline bool cpu_has_vmx_unrestricted_guest(void)
1232 return vmcs_config.cpu_based_2nd_exec_ctrl &
1233 SECONDARY_EXEC_UNRESTRICTED_GUEST;
1236 static inline bool cpu_has_vmx_ple(void)
1238 return vmcs_config.cpu_based_2nd_exec_ctrl &
1239 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
1242 static inline bool cpu_has_vmx_basic_inout(void)
1244 return (((u64)vmcs_config.basic_cap << 32) & VMX_BASIC_INOUT);
1247 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
1249 return flexpriority_enabled && lapic_in_kernel(vcpu);
1252 static inline bool cpu_has_vmx_vpid(void)
1254 return vmcs_config.cpu_based_2nd_exec_ctrl &
1255 SECONDARY_EXEC_ENABLE_VPID;
1258 static inline bool cpu_has_vmx_rdtscp(void)
1260 return vmcs_config.cpu_based_2nd_exec_ctrl &
1261 SECONDARY_EXEC_RDTSCP;
1264 static inline bool cpu_has_vmx_invpcid(void)
1266 return vmcs_config.cpu_based_2nd_exec_ctrl &
1267 SECONDARY_EXEC_ENABLE_INVPCID;
1270 static inline bool cpu_has_virtual_nmis(void)
1272 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
1275 static inline bool cpu_has_vmx_wbinvd_exit(void)
1277 return vmcs_config.cpu_based_2nd_exec_ctrl &
1278 SECONDARY_EXEC_WBINVD_EXITING;
1281 static inline bool cpu_has_vmx_shadow_vmcs(void)
1284 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
1285 /* check if the cpu supports writing r/o exit information fields */
1286 if (!(vmx_msr & MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS))
1289 return vmcs_config.cpu_based_2nd_exec_ctrl &
1290 SECONDARY_EXEC_SHADOW_VMCS;
1293 static inline bool cpu_has_vmx_pml(void)
1295 return vmcs_config.cpu_based_2nd_exec_ctrl & SECONDARY_EXEC_ENABLE_PML;
1298 static inline bool cpu_has_vmx_tsc_scaling(void)
1300 return vmcs_config.cpu_based_2nd_exec_ctrl &
1301 SECONDARY_EXEC_TSC_SCALING;
1304 static inline bool cpu_has_vmx_vmfunc(void)
1306 return vmcs_config.cpu_based_2nd_exec_ctrl &
1307 SECONDARY_EXEC_ENABLE_VMFUNC;
1310 static inline bool report_flexpriority(void)
1312 return flexpriority_enabled;
1315 static inline unsigned nested_cpu_vmx_misc_cr3_count(struct kvm_vcpu *vcpu)
1317 return vmx_misc_cr3_count(to_vmx(vcpu)->nested.nested_vmx_misc_low);
1320 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
1322 return vmcs12->cpu_based_vm_exec_control & bit;
1325 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
1327 return (vmcs12->cpu_based_vm_exec_control &
1328 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
1329 (vmcs12->secondary_vm_exec_control & bit);
1332 static inline bool nested_cpu_has_preemption_timer(struct vmcs12 *vmcs12)
1334 return vmcs12->pin_based_vm_exec_control &
1335 PIN_BASED_VMX_PREEMPTION_TIMER;
1338 static inline int nested_cpu_has_ept(struct vmcs12 *vmcs12)
1340 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_EPT);
1343 static inline bool nested_cpu_has_xsaves(struct vmcs12 *vmcs12)
1345 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
1348 static inline bool nested_cpu_has_pml(struct vmcs12 *vmcs12)
1350 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_PML);
1353 static inline bool nested_cpu_has_virt_x2apic_mode(struct vmcs12 *vmcs12)
1355 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
1358 static inline bool nested_cpu_has_vpid(struct vmcs12 *vmcs12)
1360 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_VPID);
1363 static inline bool nested_cpu_has_apic_reg_virt(struct vmcs12 *vmcs12)
1365 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_APIC_REGISTER_VIRT);
1368 static inline bool nested_cpu_has_vid(struct vmcs12 *vmcs12)
1370 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
1373 static inline bool nested_cpu_has_posted_intr(struct vmcs12 *vmcs12)
1375 return vmcs12->pin_based_vm_exec_control & PIN_BASED_POSTED_INTR;
1378 static inline bool nested_cpu_has_vmfunc(struct vmcs12 *vmcs12)
1380 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_VMFUNC);
1383 static inline bool nested_cpu_has_eptp_switching(struct vmcs12 *vmcs12)
1385 return nested_cpu_has_vmfunc(vmcs12) &&
1386 (vmcs12->vm_function_control &
1387 VMX_VMFUNC_EPTP_SWITCHING);
1390 static inline bool is_nmi(u32 intr_info)
1392 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1393 == (INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK);
1396 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
1398 unsigned long exit_qualification);
1399 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
1400 struct vmcs12 *vmcs12,
1401 u32 reason, unsigned long qualification);
1403 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
1407 for (i = 0; i < vmx->nmsrs; ++i)
1408 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
1413 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
1419 } operand = { vpid, 0, gva };
1421 asm volatile (__ex(ASM_VMX_INVVPID)
1422 /* CF==1 or ZF==1 --> rc = -1 */
1423 "; ja 1f ; ud2 ; 1:"
1424 : : "a"(&operand), "c"(ext) : "cc", "memory");
1427 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
1431 } operand = {eptp, gpa};
1433 asm volatile (__ex(ASM_VMX_INVEPT)
1434 /* CF==1 or ZF==1 --> rc = -1 */
1435 "; ja 1f ; ud2 ; 1:\n"
1436 : : "a" (&operand), "c" (ext) : "cc", "memory");
1439 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
1443 i = __find_msr_index(vmx, msr);
1445 return &vmx->guest_msrs[i];
1449 static void vmcs_clear(struct vmcs *vmcs)
1451 u64 phys_addr = __pa(vmcs);
1454 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
1455 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1458 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
1462 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
1464 vmcs_clear(loaded_vmcs->vmcs);
1465 if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
1466 vmcs_clear(loaded_vmcs->shadow_vmcs);
1467 loaded_vmcs->cpu = -1;
1468 loaded_vmcs->launched = 0;
1471 static void vmcs_load(struct vmcs *vmcs)
1473 u64 phys_addr = __pa(vmcs);
1476 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
1477 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1480 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
1484 #ifdef CONFIG_KEXEC_CORE
1486 * This bitmap is used to indicate whether the vmclear
1487 * operation is enabled on all cpus. All disabled by
1490 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
1492 static inline void crash_enable_local_vmclear(int cpu)
1494 cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
1497 static inline void crash_disable_local_vmclear(int cpu)
1499 cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
1502 static inline int crash_local_vmclear_enabled(int cpu)
1504 return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
1507 static void crash_vmclear_local_loaded_vmcss(void)
1509 int cpu = raw_smp_processor_id();
1510 struct loaded_vmcs *v;
1512 if (!crash_local_vmclear_enabled(cpu))
1515 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
1516 loaded_vmcss_on_cpu_link)
1517 vmcs_clear(v->vmcs);
1520 static inline void crash_enable_local_vmclear(int cpu) { }
1521 static inline void crash_disable_local_vmclear(int cpu) { }
1522 #endif /* CONFIG_KEXEC_CORE */
1524 static void __loaded_vmcs_clear(void *arg)
1526 struct loaded_vmcs *loaded_vmcs = arg;
1527 int cpu = raw_smp_processor_id();
1529 if (loaded_vmcs->cpu != cpu)
1530 return; /* vcpu migration can race with cpu offline */
1531 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1532 per_cpu(current_vmcs, cpu) = NULL;
1533 crash_disable_local_vmclear(cpu);
1534 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1537 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1538 * is before setting loaded_vmcs->vcpu to -1 which is done in
1539 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1540 * then adds the vmcs into percpu list before it is deleted.
1544 loaded_vmcs_init(loaded_vmcs);
1545 crash_enable_local_vmclear(cpu);
1548 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1550 int cpu = loaded_vmcs->cpu;
1553 smp_call_function_single(cpu,
1554 __loaded_vmcs_clear, loaded_vmcs, 1);
1557 static inline void vpid_sync_vcpu_single(int vpid)
1562 if (cpu_has_vmx_invvpid_single())
1563 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vpid, 0);
1566 static inline void vpid_sync_vcpu_global(void)
1568 if (cpu_has_vmx_invvpid_global())
1569 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1572 static inline void vpid_sync_context(int vpid)
1574 if (cpu_has_vmx_invvpid_single())
1575 vpid_sync_vcpu_single(vpid);
1577 vpid_sync_vcpu_global();
1580 static inline void ept_sync_global(void)
1582 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1585 static inline void ept_sync_context(u64 eptp)
1587 if (cpu_has_vmx_invept_context())
1588 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1593 static __always_inline void vmcs_check16(unsigned long field)
1595 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1596 "16-bit accessor invalid for 64-bit field");
1597 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1598 "16-bit accessor invalid for 64-bit high field");
1599 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1600 "16-bit accessor invalid for 32-bit high field");
1601 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1602 "16-bit accessor invalid for natural width field");
1605 static __always_inline void vmcs_check32(unsigned long field)
1607 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1608 "32-bit accessor invalid for 16-bit field");
1609 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1610 "32-bit accessor invalid for natural width field");
1613 static __always_inline void vmcs_check64(unsigned long field)
1615 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1616 "64-bit accessor invalid for 16-bit field");
1617 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1618 "64-bit accessor invalid for 64-bit high field");
1619 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1620 "64-bit accessor invalid for 32-bit field");
1621 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1622 "64-bit accessor invalid for natural width field");
1625 static __always_inline void vmcs_checkl(unsigned long field)
1627 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1628 "Natural width accessor invalid for 16-bit field");
1629 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1630 "Natural width accessor invalid for 64-bit field");
1631 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1632 "Natural width accessor invalid for 64-bit high field");
1633 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1634 "Natural width accessor invalid for 32-bit field");
1637 static __always_inline unsigned long __vmcs_readl(unsigned long field)
1639 unsigned long value;
1641 asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1642 : "=a"(value) : "d"(field) : "cc");
1646 static __always_inline u16 vmcs_read16(unsigned long field)
1648 vmcs_check16(field);
1649 return __vmcs_readl(field);
1652 static __always_inline u32 vmcs_read32(unsigned long field)
1654 vmcs_check32(field);
1655 return __vmcs_readl(field);
1658 static __always_inline u64 vmcs_read64(unsigned long field)
1660 vmcs_check64(field);
1661 #ifdef CONFIG_X86_64
1662 return __vmcs_readl(field);
1664 return __vmcs_readl(field) | ((u64)__vmcs_readl(field+1) << 32);
1668 static __always_inline unsigned long vmcs_readl(unsigned long field)
1671 return __vmcs_readl(field);
1674 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1676 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1677 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1681 static __always_inline void __vmcs_writel(unsigned long field, unsigned long value)
1685 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1686 : "=q"(error) : "a"(value), "d"(field) : "cc");
1687 if (unlikely(error))
1688 vmwrite_error(field, value);
1691 static __always_inline void vmcs_write16(unsigned long field, u16 value)
1693 vmcs_check16(field);
1694 __vmcs_writel(field, value);
1697 static __always_inline void vmcs_write32(unsigned long field, u32 value)
1699 vmcs_check32(field);
1700 __vmcs_writel(field, value);
1703 static __always_inline void vmcs_write64(unsigned long field, u64 value)
1705 vmcs_check64(field);
1706 __vmcs_writel(field, value);
1707 #ifndef CONFIG_X86_64
1709 __vmcs_writel(field+1, value >> 32);
1713 static __always_inline void vmcs_writel(unsigned long field, unsigned long value)
1716 __vmcs_writel(field, value);
1719 static __always_inline void vmcs_clear_bits(unsigned long field, u32 mask)
1721 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1722 "vmcs_clear_bits does not support 64-bit fields");
1723 __vmcs_writel(field, __vmcs_readl(field) & ~mask);
1726 static __always_inline void vmcs_set_bits(unsigned long field, u32 mask)
1728 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1729 "vmcs_set_bits does not support 64-bit fields");
1730 __vmcs_writel(field, __vmcs_readl(field) | mask);
1733 static inline void vm_entry_controls_reset_shadow(struct vcpu_vmx *vmx)
1735 vmx->vm_entry_controls_shadow = vmcs_read32(VM_ENTRY_CONTROLS);
1738 static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val)
1740 vmcs_write32(VM_ENTRY_CONTROLS, val);
1741 vmx->vm_entry_controls_shadow = val;
1744 static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val)
1746 if (vmx->vm_entry_controls_shadow != val)
1747 vm_entry_controls_init(vmx, val);
1750 static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx)
1752 return vmx->vm_entry_controls_shadow;
1756 static inline void vm_entry_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1758 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) | val);
1761 static inline void vm_entry_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1763 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) & ~val);
1766 static inline void vm_exit_controls_reset_shadow(struct vcpu_vmx *vmx)
1768 vmx->vm_exit_controls_shadow = vmcs_read32(VM_EXIT_CONTROLS);
1771 static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val)
1773 vmcs_write32(VM_EXIT_CONTROLS, val);
1774 vmx->vm_exit_controls_shadow = val;
1777 static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val)
1779 if (vmx->vm_exit_controls_shadow != val)
1780 vm_exit_controls_init(vmx, val);
1783 static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx)
1785 return vmx->vm_exit_controls_shadow;
1789 static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1791 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val);
1794 static inline void vm_exit_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1796 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) & ~val);
1799 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1801 vmx->segment_cache.bitmask = 0;
1804 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1808 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1810 if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1811 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1812 vmx->segment_cache.bitmask = 0;
1814 ret = vmx->segment_cache.bitmask & mask;
1815 vmx->segment_cache.bitmask |= mask;
1819 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1821 u16 *p = &vmx->segment_cache.seg[seg].selector;
1823 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1824 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1828 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1830 ulong *p = &vmx->segment_cache.seg[seg].base;
1832 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1833 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1837 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1839 u32 *p = &vmx->segment_cache.seg[seg].limit;
1841 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1842 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1846 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1848 u32 *p = &vmx->segment_cache.seg[seg].ar;
1850 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1851 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1855 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1859 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1860 (1u << DB_VECTOR) | (1u << AC_VECTOR);
1861 if ((vcpu->guest_debug &
1862 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1863 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1864 eb |= 1u << BP_VECTOR;
1865 if (to_vmx(vcpu)->rmode.vm86_active)
1868 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1870 /* When we are running a nested L2 guest and L1 specified for it a
1871 * certain exception bitmap, we must trap the same exceptions and pass
1872 * them to L1. When running L2, we will only handle the exceptions
1873 * specified above if L1 did not want them.
1875 if (is_guest_mode(vcpu))
1876 eb |= get_vmcs12(vcpu)->exception_bitmap;
1878 vmcs_write32(EXCEPTION_BITMAP, eb);
1882 * Check if MSR is intercepted for currently loaded MSR bitmap.
1884 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr)
1886 unsigned long *msr_bitmap;
1887 int f = sizeof(unsigned long);
1889 if (!cpu_has_vmx_msr_bitmap())
1892 msr_bitmap = to_vmx(vcpu)->loaded_vmcs->msr_bitmap;
1894 if (msr <= 0x1fff) {
1895 return !!test_bit(msr, msr_bitmap + 0x800 / f);
1896 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
1898 return !!test_bit(msr, msr_bitmap + 0xc00 / f);
1905 * Check if MSR is intercepted for L01 MSR bitmap.
1907 static bool msr_write_intercepted_l01(struct kvm_vcpu *vcpu, u32 msr)
1909 unsigned long *msr_bitmap;
1910 int f = sizeof(unsigned long);
1912 if (!cpu_has_vmx_msr_bitmap())
1915 msr_bitmap = to_vmx(vcpu)->vmcs01.msr_bitmap;
1917 if (msr <= 0x1fff) {
1918 return !!test_bit(msr, msr_bitmap + 0x800 / f);
1919 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
1921 return !!test_bit(msr, msr_bitmap + 0xc00 / f);
1927 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1928 unsigned long entry, unsigned long exit)
1930 vm_entry_controls_clearbit(vmx, entry);
1931 vm_exit_controls_clearbit(vmx, exit);
1934 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1937 struct msr_autoload *m = &vmx->msr_autoload;
1941 if (cpu_has_load_ia32_efer) {
1942 clear_atomic_switch_msr_special(vmx,
1943 VM_ENTRY_LOAD_IA32_EFER,
1944 VM_EXIT_LOAD_IA32_EFER);
1948 case MSR_CORE_PERF_GLOBAL_CTRL:
1949 if (cpu_has_load_perf_global_ctrl) {
1950 clear_atomic_switch_msr_special(vmx,
1951 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1952 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1958 for (i = 0; i < m->nr; ++i)
1959 if (m->guest[i].index == msr)
1965 m->guest[i] = m->guest[m->nr];
1966 m->host[i] = m->host[m->nr];
1967 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1968 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1971 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1972 unsigned long entry, unsigned long exit,
1973 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
1974 u64 guest_val, u64 host_val)
1976 vmcs_write64(guest_val_vmcs, guest_val);
1977 vmcs_write64(host_val_vmcs, host_val);
1978 vm_entry_controls_setbit(vmx, entry);
1979 vm_exit_controls_setbit(vmx, exit);
1982 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1983 u64 guest_val, u64 host_val)
1986 struct msr_autoload *m = &vmx->msr_autoload;
1990 if (cpu_has_load_ia32_efer) {
1991 add_atomic_switch_msr_special(vmx,
1992 VM_ENTRY_LOAD_IA32_EFER,
1993 VM_EXIT_LOAD_IA32_EFER,
1996 guest_val, host_val);
2000 case MSR_CORE_PERF_GLOBAL_CTRL:
2001 if (cpu_has_load_perf_global_ctrl) {
2002 add_atomic_switch_msr_special(vmx,
2003 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
2004 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
2005 GUEST_IA32_PERF_GLOBAL_CTRL,
2006 HOST_IA32_PERF_GLOBAL_CTRL,
2007 guest_val, host_val);
2011 case MSR_IA32_PEBS_ENABLE:
2012 /* PEBS needs a quiescent period after being disabled (to write
2013 * a record). Disabling PEBS through VMX MSR swapping doesn't
2014 * provide that period, so a CPU could write host's record into
2017 wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
2020 for (i = 0; i < m->nr; ++i)
2021 if (m->guest[i].index == msr)
2024 if (i == NR_AUTOLOAD_MSRS) {
2025 printk_once(KERN_WARNING "Not enough msr switch entries. "
2026 "Can't add msr %x\n", msr);
2028 } else if (i == m->nr) {
2030 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
2031 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
2034 m->guest[i].index = msr;
2035 m->guest[i].value = guest_val;
2036 m->host[i].index = msr;
2037 m->host[i].value = host_val;
2040 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
2042 u64 guest_efer = vmx->vcpu.arch.efer;
2043 u64 ignore_bits = 0;
2047 * NX is needed to handle CR0.WP=1, CR4.SMEP=1. Testing
2048 * host CPUID is more efficient than testing guest CPUID
2049 * or CR4. Host SMEP is anyway a requirement for guest SMEP.
2051 if (boot_cpu_has(X86_FEATURE_SMEP))
2052 guest_efer |= EFER_NX;
2053 else if (!(guest_efer & EFER_NX))
2054 ignore_bits |= EFER_NX;
2058 * LMA and LME handled by hardware; SCE meaningless outside long mode.
2060 ignore_bits |= EFER_SCE;
2061 #ifdef CONFIG_X86_64
2062 ignore_bits |= EFER_LMA | EFER_LME;
2063 /* SCE is meaningful only in long mode on Intel */
2064 if (guest_efer & EFER_LMA)
2065 ignore_bits &= ~(u64)EFER_SCE;
2068 clear_atomic_switch_msr(vmx, MSR_EFER);
2071 * On EPT, we can't emulate NX, so we must switch EFER atomically.
2072 * On CPUs that support "load IA32_EFER", always switch EFER
2073 * atomically, since it's faster than switching it manually.
2075 if (cpu_has_load_ia32_efer ||
2076 (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
2077 if (!(guest_efer & EFER_LMA))
2078 guest_efer &= ~EFER_LME;
2079 if (guest_efer != host_efer)
2080 add_atomic_switch_msr(vmx, MSR_EFER,
2081 guest_efer, host_efer);
2084 guest_efer &= ~ignore_bits;
2085 guest_efer |= host_efer & ignore_bits;
2087 vmx->guest_msrs[efer_offset].data = guest_efer;
2088 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
2094 #ifdef CONFIG_X86_32
2096 * On 32-bit kernels, VM exits still load the FS and GS bases from the
2097 * VMCS rather than the segment table. KVM uses this helper to figure
2098 * out the current bases to poke them into the VMCS before entry.
2100 static unsigned long segment_base(u16 selector)
2102 struct desc_struct *table;
2105 if (!(selector & ~SEGMENT_RPL_MASK))
2108 table = get_current_gdt_ro();
2110 if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
2111 u16 ldt_selector = kvm_read_ldt();
2113 if (!(ldt_selector & ~SEGMENT_RPL_MASK))
2116 table = (struct desc_struct *)segment_base(ldt_selector);
2118 v = get_desc_base(&table[selector >> 3]);
2123 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
2125 struct vcpu_vmx *vmx = to_vmx(vcpu);
2128 if (vmx->host_state.loaded)
2131 vmx->host_state.loaded = 1;
2133 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
2134 * allow segment selectors with cpl > 0 or ti == 1.
2136 vmx->host_state.ldt_sel = kvm_read_ldt();
2137 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
2138 savesegment(fs, vmx->host_state.fs_sel);
2139 if (!(vmx->host_state.fs_sel & 7)) {
2140 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
2141 vmx->host_state.fs_reload_needed = 0;
2143 vmcs_write16(HOST_FS_SELECTOR, 0);
2144 vmx->host_state.fs_reload_needed = 1;
2146 savesegment(gs, vmx->host_state.gs_sel);
2147 if (!(vmx->host_state.gs_sel & 7))
2148 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
2150 vmcs_write16(HOST_GS_SELECTOR, 0);
2151 vmx->host_state.gs_ldt_reload_needed = 1;
2154 #ifdef CONFIG_X86_64
2155 savesegment(ds, vmx->host_state.ds_sel);
2156 savesegment(es, vmx->host_state.es_sel);
2159 #ifdef CONFIG_X86_64
2160 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
2161 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
2163 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
2164 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
2167 #ifdef CONFIG_X86_64
2168 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2169 if (is_long_mode(&vmx->vcpu))
2170 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2172 if (boot_cpu_has(X86_FEATURE_MPX))
2173 rdmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
2174 for (i = 0; i < vmx->save_nmsrs; ++i)
2175 kvm_set_shared_msr(vmx->guest_msrs[i].index,
2176 vmx->guest_msrs[i].data,
2177 vmx->guest_msrs[i].mask);
2180 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
2182 if (!vmx->host_state.loaded)
2185 ++vmx->vcpu.stat.host_state_reload;
2186 vmx->host_state.loaded = 0;
2187 #ifdef CONFIG_X86_64
2188 if (is_long_mode(&vmx->vcpu))
2189 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2191 if (vmx->host_state.gs_ldt_reload_needed) {
2192 kvm_load_ldt(vmx->host_state.ldt_sel);
2193 #ifdef CONFIG_X86_64
2194 load_gs_index(vmx->host_state.gs_sel);
2196 loadsegment(gs, vmx->host_state.gs_sel);
2199 if (vmx->host_state.fs_reload_needed)
2200 loadsegment(fs, vmx->host_state.fs_sel);
2201 #ifdef CONFIG_X86_64
2202 if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
2203 loadsegment(ds, vmx->host_state.ds_sel);
2204 loadsegment(es, vmx->host_state.es_sel);
2207 invalidate_tss_limit();
2208 #ifdef CONFIG_X86_64
2209 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2211 if (vmx->host_state.msr_host_bndcfgs)
2212 wrmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
2213 load_fixmap_gdt(raw_smp_processor_id());
2216 static void vmx_load_host_state(struct vcpu_vmx *vmx)
2219 __vmx_load_host_state(vmx);
2223 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
2225 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2226 struct pi_desc old, new;
2230 * In case of hot-plug or hot-unplug, we may have to undo
2231 * vmx_vcpu_pi_put even if there is no assigned device. And we
2232 * always keep PI.NDST up to date for simplicity: it makes the
2233 * code easier, and CPU migration is not a fast path.
2235 if (!pi_test_sn(pi_desc) && vcpu->cpu == cpu)
2239 * First handle the simple case where no cmpxchg is necessary; just
2240 * allow posting non-urgent interrupts.
2242 * If the 'nv' field is POSTED_INTR_WAKEUP_VECTOR, do not change
2243 * PI.NDST: pi_post_block will do it for us and the wakeup_handler
2244 * expects the VCPU to be on the blocked_vcpu_list that matches
2247 if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR ||
2249 pi_clear_sn(pi_desc);
2253 /* The full case. */
2255 old.control = new.control = pi_desc->control;
2257 dest = cpu_physical_id(cpu);
2259 if (x2apic_enabled())
2262 new.ndst = (dest << 8) & 0xFF00;
2265 } while (cmpxchg64(&pi_desc->control, old.control,
2266 new.control) != old.control);
2269 static void decache_tsc_multiplier(struct vcpu_vmx *vmx)
2271 vmx->current_tsc_ratio = vmx->vcpu.arch.tsc_scaling_ratio;
2272 vmcs_write64(TSC_MULTIPLIER, vmx->current_tsc_ratio);
2276 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
2277 * vcpu mutex is already taken.
2279 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2281 struct vcpu_vmx *vmx = to_vmx(vcpu);
2282 bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
2284 if (!already_loaded) {
2285 loaded_vmcs_clear(vmx->loaded_vmcs);
2286 local_irq_disable();
2287 crash_disable_local_vmclear(cpu);
2290 * Read loaded_vmcs->cpu should be before fetching
2291 * loaded_vmcs->loaded_vmcss_on_cpu_link.
2292 * See the comments in __loaded_vmcs_clear().
2296 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
2297 &per_cpu(loaded_vmcss_on_cpu, cpu));
2298 crash_enable_local_vmclear(cpu);
2302 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
2303 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
2304 vmcs_load(vmx->loaded_vmcs->vmcs);
2305 indirect_branch_prediction_barrier();
2308 if (!already_loaded) {
2309 void *gdt = get_current_gdt_ro();
2310 unsigned long sysenter_esp;
2312 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2315 * Linux uses per-cpu TSS and GDT, so set these when switching
2316 * processors. See 22.2.4.
2318 vmcs_writel(HOST_TR_BASE,
2319 (unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss);
2320 vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt); /* 22.2.4 */
2323 * VM exits change the host TR limit to 0x67 after a VM
2324 * exit. This is okay, since 0x67 covers everything except
2325 * the IO bitmap and have have code to handle the IO bitmap
2326 * being lost after a VM exit.
2328 BUILD_BUG_ON(IO_BITMAP_OFFSET - 1 != 0x67);
2330 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
2331 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
2333 vmx->loaded_vmcs->cpu = cpu;
2336 /* Setup TSC multiplier */
2337 if (kvm_has_tsc_control &&
2338 vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
2339 decache_tsc_multiplier(vmx);
2341 vmx_vcpu_pi_load(vcpu, cpu);
2342 vmx->host_pkru = read_pkru();
2343 vmx->host_debugctlmsr = get_debugctlmsr();
2346 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
2348 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2350 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
2351 !irq_remapping_cap(IRQ_POSTING_CAP) ||
2352 !kvm_vcpu_apicv_active(vcpu))
2355 /* Set SN when the vCPU is preempted */
2356 if (vcpu->preempted)
2360 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
2362 vmx_vcpu_pi_put(vcpu);
2364 __vmx_load_host_state(to_vmx(vcpu));
2367 static bool emulation_required(struct kvm_vcpu *vcpu)
2369 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
2372 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
2375 * Return the cr0 value that a nested guest would read. This is a combination
2376 * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
2377 * its hypervisor (cr0_read_shadow).
2379 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
2381 return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
2382 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
2384 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
2386 return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
2387 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
2390 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
2392 unsigned long rflags, save_rflags;
2394 if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
2395 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2396 rflags = vmcs_readl(GUEST_RFLAGS);
2397 if (to_vmx(vcpu)->rmode.vm86_active) {
2398 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2399 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
2400 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2402 to_vmx(vcpu)->rflags = rflags;
2404 return to_vmx(vcpu)->rflags;
2407 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
2409 unsigned long old_rflags = vmx_get_rflags(vcpu);
2411 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2412 to_vmx(vcpu)->rflags = rflags;
2413 if (to_vmx(vcpu)->rmode.vm86_active) {
2414 to_vmx(vcpu)->rmode.save_rflags = rflags;
2415 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2417 vmcs_writel(GUEST_RFLAGS, rflags);
2419 if ((old_rflags ^ to_vmx(vcpu)->rflags) & X86_EFLAGS_VM)
2420 to_vmx(vcpu)->emulation_required = emulation_required(vcpu);
2423 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
2425 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2428 if (interruptibility & GUEST_INTR_STATE_STI)
2429 ret |= KVM_X86_SHADOW_INT_STI;
2430 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
2431 ret |= KVM_X86_SHADOW_INT_MOV_SS;
2436 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
2438 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2439 u32 interruptibility = interruptibility_old;
2441 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
2443 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
2444 interruptibility |= GUEST_INTR_STATE_MOV_SS;
2445 else if (mask & KVM_X86_SHADOW_INT_STI)
2446 interruptibility |= GUEST_INTR_STATE_STI;
2448 if ((interruptibility != interruptibility_old))
2449 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
2452 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
2456 rip = kvm_rip_read(vcpu);
2457 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
2458 kvm_rip_write(vcpu, rip);
2460 /* skipping an emulated instruction also counts */
2461 vmx_set_interrupt_shadow(vcpu, 0);
2464 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu,
2465 unsigned long exit_qual)
2467 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2468 unsigned int nr = vcpu->arch.exception.nr;
2469 u32 intr_info = nr | INTR_INFO_VALID_MASK;
2471 if (vcpu->arch.exception.has_error_code) {
2472 vmcs12->vm_exit_intr_error_code = vcpu->arch.exception.error_code;
2473 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2476 if (kvm_exception_is_soft(nr))
2477 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2479 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2481 if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
2482 vmx_get_nmi_mask(vcpu))
2483 intr_info |= INTR_INFO_UNBLOCK_NMI;
2485 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
2489 * KVM wants to inject page-faults which it got to the guest. This function
2490 * checks whether in a nested guest, we need to inject them to L1 or L2.
2492 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned long *exit_qual)
2494 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2495 unsigned int nr = vcpu->arch.exception.nr;
2497 if (nr == PF_VECTOR) {
2498 if (vcpu->arch.exception.nested_apf) {
2499 *exit_qual = vcpu->arch.apf.nested_apf_token;
2503 * FIXME: we must not write CR2 when L1 intercepts an L2 #PF exception.
2504 * The fix is to add the ancillary datum (CR2 or DR6) to structs
2505 * kvm_queued_exception and kvm_vcpu_events, so that CR2 and DR6
2506 * can be written only when inject_pending_event runs. This should be
2507 * conditional on a new capability---if the capability is disabled,
2508 * kvm_multiple_exception would write the ancillary information to
2509 * CR2 or DR6, for backwards ABI-compatibility.
2511 if (nested_vmx_is_page_fault_vmexit(vmcs12,
2512 vcpu->arch.exception.error_code)) {
2513 *exit_qual = vcpu->arch.cr2;
2517 if (vmcs12->exception_bitmap & (1u << nr)) {
2518 if (nr == DB_VECTOR)
2519 *exit_qual = vcpu->arch.dr6;
2529 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
2531 struct vcpu_vmx *vmx = to_vmx(vcpu);
2532 unsigned nr = vcpu->arch.exception.nr;
2533 bool has_error_code = vcpu->arch.exception.has_error_code;
2534 u32 error_code = vcpu->arch.exception.error_code;
2535 u32 intr_info = nr | INTR_INFO_VALID_MASK;
2537 if (has_error_code) {
2538 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
2539 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2542 if (vmx->rmode.vm86_active) {
2544 if (kvm_exception_is_soft(nr))
2545 inc_eip = vcpu->arch.event_exit_inst_len;
2546 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
2547 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2551 if (kvm_exception_is_soft(nr)) {
2552 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2553 vmx->vcpu.arch.event_exit_inst_len);
2554 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2556 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2558 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
2561 static bool vmx_rdtscp_supported(void)
2563 return cpu_has_vmx_rdtscp();
2566 static bool vmx_invpcid_supported(void)
2568 return cpu_has_vmx_invpcid() && enable_ept;
2572 * Swap MSR entry in host/guest MSR entry array.
2574 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
2576 struct shared_msr_entry tmp;
2578 tmp = vmx->guest_msrs[to];
2579 vmx->guest_msrs[to] = vmx->guest_msrs[from];
2580 vmx->guest_msrs[from] = tmp;
2584 * Set up the vmcs to automatically save and restore system
2585 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
2586 * mode, as fiddling with msrs is very expensive.
2588 static void setup_msrs(struct vcpu_vmx *vmx)
2590 int save_nmsrs, index;
2593 #ifdef CONFIG_X86_64
2594 if (is_long_mode(&vmx->vcpu)) {
2595 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
2597 move_msr_up(vmx, index, save_nmsrs++);
2598 index = __find_msr_index(vmx, MSR_LSTAR);
2600 move_msr_up(vmx, index, save_nmsrs++);
2601 index = __find_msr_index(vmx, MSR_CSTAR);
2603 move_msr_up(vmx, index, save_nmsrs++);
2604 index = __find_msr_index(vmx, MSR_TSC_AUX);
2605 if (index >= 0 && guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP))
2606 move_msr_up(vmx, index, save_nmsrs++);
2608 * MSR_STAR is only needed on long mode guests, and only
2609 * if efer.sce is enabled.
2611 index = __find_msr_index(vmx, MSR_STAR);
2612 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
2613 move_msr_up(vmx, index, save_nmsrs++);
2616 index = __find_msr_index(vmx, MSR_EFER);
2617 if (index >= 0 && update_transition_efer(vmx, index))
2618 move_msr_up(vmx, index, save_nmsrs++);
2620 vmx->save_nmsrs = save_nmsrs;
2622 if (cpu_has_vmx_msr_bitmap())
2623 vmx_update_msr_bitmap(&vmx->vcpu);
2627 * reads and returns guest's timestamp counter "register"
2628 * guest_tsc = (host_tsc * tsc multiplier) >> 48 + tsc_offset
2629 * -- Intel TSC Scaling for Virtualization White Paper, sec 1.3
2631 static u64 guest_read_tsc(struct kvm_vcpu *vcpu)
2633 u64 host_tsc, tsc_offset;
2636 tsc_offset = vmcs_read64(TSC_OFFSET);
2637 return kvm_scale_tsc(vcpu, host_tsc) + tsc_offset;
2641 * writes 'offset' into guest's timestamp counter offset register
2643 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2645 if (is_guest_mode(vcpu)) {
2647 * We're here if L1 chose not to trap WRMSR to TSC. According
2648 * to the spec, this should set L1's TSC; The offset that L1
2649 * set for L2 remains unchanged, and still needs to be added
2650 * to the newly set TSC to get L2's TSC.
2652 struct vmcs12 *vmcs12;
2653 /* recalculate vmcs02.TSC_OFFSET: */
2654 vmcs12 = get_vmcs12(vcpu);
2655 vmcs_write64(TSC_OFFSET, offset +
2656 (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
2657 vmcs12->tsc_offset : 0));
2659 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2660 vmcs_read64(TSC_OFFSET), offset);
2661 vmcs_write64(TSC_OFFSET, offset);
2666 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
2667 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
2668 * all guests if the "nested" module option is off, and can also be disabled
2669 * for a single guest by disabling its VMX cpuid bit.
2671 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
2673 return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
2677 * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
2678 * returned for the various VMX controls MSRs when nested VMX is enabled.
2679 * The same values should also be used to verify that vmcs12 control fields are
2680 * valid during nested entry from L1 to L2.
2681 * Each of these control msrs has a low and high 32-bit half: A low bit is on
2682 * if the corresponding bit in the (32-bit) control field *must* be on, and a
2683 * bit in the high half is on if the corresponding bit in the control field
2684 * may be on. See also vmx_control_verify().
2686 static void nested_vmx_setup_ctls_msrs(struct vcpu_vmx *vmx)
2689 * Note that as a general rule, the high half of the MSRs (bits in
2690 * the control fields which may be 1) should be initialized by the
2691 * intersection of the underlying hardware's MSR (i.e., features which
2692 * can be supported) and the list of features we want to expose -
2693 * because they are known to be properly supported in our code.
2694 * Also, usually, the low half of the MSRs (bits which must be 1) can
2695 * be set to 0, meaning that L1 may turn off any of these bits. The
2696 * reason is that if one of these bits is necessary, it will appear
2697 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
2698 * fields of vmcs01 and vmcs02, will turn these bits off - and
2699 * nested_vmx_exit_reflected() will not pass related exits to L1.
2700 * These rules have exceptions below.
2703 /* pin-based controls */
2704 rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
2705 vmx->nested.nested_vmx_pinbased_ctls_low,
2706 vmx->nested.nested_vmx_pinbased_ctls_high);
2707 vmx->nested.nested_vmx_pinbased_ctls_low |=
2708 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2709 vmx->nested.nested_vmx_pinbased_ctls_high &=
2710 PIN_BASED_EXT_INTR_MASK |
2711 PIN_BASED_NMI_EXITING |
2712 PIN_BASED_VIRTUAL_NMIS;
2713 vmx->nested.nested_vmx_pinbased_ctls_high |=
2714 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2715 PIN_BASED_VMX_PREEMPTION_TIMER;
2716 if (kvm_vcpu_apicv_active(&vmx->vcpu))
2717 vmx->nested.nested_vmx_pinbased_ctls_high |=
2718 PIN_BASED_POSTED_INTR;
2721 rdmsr(MSR_IA32_VMX_EXIT_CTLS,
2722 vmx->nested.nested_vmx_exit_ctls_low,
2723 vmx->nested.nested_vmx_exit_ctls_high);
2724 vmx->nested.nested_vmx_exit_ctls_low =
2725 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
2727 vmx->nested.nested_vmx_exit_ctls_high &=
2728 #ifdef CONFIG_X86_64
2729 VM_EXIT_HOST_ADDR_SPACE_SIZE |
2731 VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
2732 vmx->nested.nested_vmx_exit_ctls_high |=
2733 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
2734 VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
2735 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
2737 if (kvm_mpx_supported())
2738 vmx->nested.nested_vmx_exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
2740 /* We support free control of debug control saving. */
2741 vmx->nested.nested_vmx_exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
2743 /* entry controls */
2744 rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
2745 vmx->nested.nested_vmx_entry_ctls_low,
2746 vmx->nested.nested_vmx_entry_ctls_high);
2747 vmx->nested.nested_vmx_entry_ctls_low =
2748 VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
2749 vmx->nested.nested_vmx_entry_ctls_high &=
2750 #ifdef CONFIG_X86_64
2751 VM_ENTRY_IA32E_MODE |
2753 VM_ENTRY_LOAD_IA32_PAT;
2754 vmx->nested.nested_vmx_entry_ctls_high |=
2755 (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
2756 if (kvm_mpx_supported())
2757 vmx->nested.nested_vmx_entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
2759 /* We support free control of debug control loading. */
2760 vmx->nested.nested_vmx_entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
2762 /* cpu-based controls */
2763 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
2764 vmx->nested.nested_vmx_procbased_ctls_low,
2765 vmx->nested.nested_vmx_procbased_ctls_high);
2766 vmx->nested.nested_vmx_procbased_ctls_low =
2767 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2768 vmx->nested.nested_vmx_procbased_ctls_high &=
2769 CPU_BASED_VIRTUAL_INTR_PENDING |
2770 CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
2771 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
2772 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
2773 CPU_BASED_CR3_STORE_EXITING |
2774 #ifdef CONFIG_X86_64
2775 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
2777 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
2778 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
2779 CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
2780 CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
2781 CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2783 * We can allow some features even when not supported by the
2784 * hardware. For example, L1 can specify an MSR bitmap - and we
2785 * can use it to avoid exits to L1 - even when L0 runs L2
2786 * without MSR bitmaps.
2788 vmx->nested.nested_vmx_procbased_ctls_high |=
2789 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2790 CPU_BASED_USE_MSR_BITMAPS;
2792 /* We support free control of CR3 access interception. */
2793 vmx->nested.nested_vmx_procbased_ctls_low &=
2794 ~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
2797 * secondary cpu-based controls. Do not include those that
2798 * depend on CPUID bits, they are added later by vmx_cpuid_update.
2800 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2801 vmx->nested.nested_vmx_secondary_ctls_low,
2802 vmx->nested.nested_vmx_secondary_ctls_high);
2803 vmx->nested.nested_vmx_secondary_ctls_low = 0;
2804 vmx->nested.nested_vmx_secondary_ctls_high &=
2805 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2806 SECONDARY_EXEC_DESC |
2807 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2808 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2809 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2810 SECONDARY_EXEC_WBINVD_EXITING;
2813 /* nested EPT: emulate EPT also to L1 */
2814 vmx->nested.nested_vmx_secondary_ctls_high |=
2815 SECONDARY_EXEC_ENABLE_EPT;
2816 vmx->nested.nested_vmx_ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
2817 VMX_EPTP_WB_BIT | VMX_EPT_INVEPT_BIT;
2818 if (cpu_has_vmx_ept_execute_only())
2819 vmx->nested.nested_vmx_ept_caps |=
2820 VMX_EPT_EXECUTE_ONLY_BIT;
2821 vmx->nested.nested_vmx_ept_caps &= vmx_capability.ept;
2822 vmx->nested.nested_vmx_ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
2823 VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
2824 VMX_EPT_1GB_PAGE_BIT;
2825 if (enable_ept_ad_bits) {
2826 vmx->nested.nested_vmx_secondary_ctls_high |=
2827 SECONDARY_EXEC_ENABLE_PML;
2828 vmx->nested.nested_vmx_ept_caps |= VMX_EPT_AD_BIT;
2832 if (cpu_has_vmx_vmfunc()) {
2833 vmx->nested.nested_vmx_secondary_ctls_high |=
2834 SECONDARY_EXEC_ENABLE_VMFUNC;
2836 * Advertise EPTP switching unconditionally
2837 * since we emulate it
2840 vmx->nested.nested_vmx_vmfunc_controls =
2841 VMX_VMFUNC_EPTP_SWITCHING;
2845 * Old versions of KVM use the single-context version without
2846 * checking for support, so declare that it is supported even
2847 * though it is treated as global context. The alternative is
2848 * not failing the single-context invvpid, and it is worse.
2851 vmx->nested.nested_vmx_secondary_ctls_high |=
2852 SECONDARY_EXEC_ENABLE_VPID;
2853 vmx->nested.nested_vmx_vpid_caps = VMX_VPID_INVVPID_BIT |
2854 VMX_VPID_EXTENT_SUPPORTED_MASK;
2857 if (enable_unrestricted_guest)
2858 vmx->nested.nested_vmx_secondary_ctls_high |=
2859 SECONDARY_EXEC_UNRESTRICTED_GUEST;
2861 /* miscellaneous data */
2862 rdmsr(MSR_IA32_VMX_MISC,
2863 vmx->nested.nested_vmx_misc_low,
2864 vmx->nested.nested_vmx_misc_high);
2865 vmx->nested.nested_vmx_misc_low &= VMX_MISC_SAVE_EFER_LMA;
2866 vmx->nested.nested_vmx_misc_low |=
2867 VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
2868 VMX_MISC_ACTIVITY_HLT;
2869 vmx->nested.nested_vmx_misc_high = 0;
2872 * This MSR reports some information about VMX support. We
2873 * should return information about the VMX we emulate for the
2874 * guest, and the VMCS structure we give it - not about the
2875 * VMX support of the underlying hardware.
2877 vmx->nested.nested_vmx_basic =
2879 VMX_BASIC_TRUE_CTLS |
2880 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2881 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2883 if (cpu_has_vmx_basic_inout())
2884 vmx->nested.nested_vmx_basic |= VMX_BASIC_INOUT;
2887 * These MSRs specify bits which the guest must keep fixed on
2888 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2889 * We picked the standard core2 setting.
2891 #define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2892 #define VMXON_CR4_ALWAYSON X86_CR4_VMXE
2893 vmx->nested.nested_vmx_cr0_fixed0 = VMXON_CR0_ALWAYSON;
2894 vmx->nested.nested_vmx_cr4_fixed0 = VMXON_CR4_ALWAYSON;
2896 /* These MSRs specify bits which the guest must keep fixed off. */
2897 rdmsrl(MSR_IA32_VMX_CR0_FIXED1, vmx->nested.nested_vmx_cr0_fixed1);
2898 rdmsrl(MSR_IA32_VMX_CR4_FIXED1, vmx->nested.nested_vmx_cr4_fixed1);
2900 /* highest index: VMX_PREEMPTION_TIMER_VALUE */
2901 vmx->nested.nested_vmx_vmcs_enum = VMCS12_MAX_FIELD_INDEX << 1;
2905 * if fixed0[i] == 1: val[i] must be 1
2906 * if fixed1[i] == 0: val[i] must be 0
2908 static inline bool fixed_bits_valid(u64 val, u64 fixed0, u64 fixed1)
2910 return ((val & fixed1) | fixed0) == val;
2913 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2915 return fixed_bits_valid(control, low, high);
2918 static inline u64 vmx_control_msr(u32 low, u32 high)
2920 return low | ((u64)high << 32);
2923 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
2928 return (superset | subset) == superset;
2931 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
2933 const u64 feature_and_reserved =
2934 /* feature (except bit 48; see below) */
2935 BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
2937 BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
2938 u64 vmx_basic = vmx->nested.nested_vmx_basic;
2940 if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
2944 * KVM does not emulate a version of VMX that constrains physical
2945 * addresses of VMX structures (e.g. VMCS) to 32-bits.
2947 if (data & BIT_ULL(48))
2950 if (vmx_basic_vmcs_revision_id(vmx_basic) !=
2951 vmx_basic_vmcs_revision_id(data))
2954 if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
2957 vmx->nested.nested_vmx_basic = data;
2962 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
2967 switch (msr_index) {
2968 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2969 lowp = &vmx->nested.nested_vmx_pinbased_ctls_low;
2970 highp = &vmx->nested.nested_vmx_pinbased_ctls_high;
2972 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2973 lowp = &vmx->nested.nested_vmx_procbased_ctls_low;
2974 highp = &vmx->nested.nested_vmx_procbased_ctls_high;
2976 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2977 lowp = &vmx->nested.nested_vmx_exit_ctls_low;
2978 highp = &vmx->nested.nested_vmx_exit_ctls_high;
2980 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2981 lowp = &vmx->nested.nested_vmx_entry_ctls_low;
2982 highp = &vmx->nested.nested_vmx_entry_ctls_high;
2984 case MSR_IA32_VMX_PROCBASED_CTLS2:
2985 lowp = &vmx->nested.nested_vmx_secondary_ctls_low;
2986 highp = &vmx->nested.nested_vmx_secondary_ctls_high;
2992 supported = vmx_control_msr(*lowp, *highp);
2994 /* Check must-be-1 bits are still 1. */
2995 if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
2998 /* Check must-be-0 bits are still 0. */
2999 if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
3003 *highp = data >> 32;
3007 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
3009 const u64 feature_and_reserved_bits =
3011 BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
3012 BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
3014 GENMASK_ULL(13, 9) | BIT_ULL(31);
3017 vmx_misc = vmx_control_msr(vmx->nested.nested_vmx_misc_low,
3018 vmx->nested.nested_vmx_misc_high);
3020 if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
3023 if ((vmx->nested.nested_vmx_pinbased_ctls_high &
3024 PIN_BASED_VMX_PREEMPTION_TIMER) &&
3025 vmx_misc_preemption_timer_rate(data) !=
3026 vmx_misc_preemption_timer_rate(vmx_misc))
3029 if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
3032 if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
3035 if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
3038 vmx->nested.nested_vmx_misc_low = data;
3039 vmx->nested.nested_vmx_misc_high = data >> 32;
3043 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
3045 u64 vmx_ept_vpid_cap;
3047 vmx_ept_vpid_cap = vmx_control_msr(vmx->nested.nested_vmx_ept_caps,
3048 vmx->nested.nested_vmx_vpid_caps);
3050 /* Every bit is either reserved or a feature bit. */
3051 if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
3054 vmx->nested.nested_vmx_ept_caps = data;
3055 vmx->nested.nested_vmx_vpid_caps = data >> 32;
3059 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
3063 switch (msr_index) {
3064 case MSR_IA32_VMX_CR0_FIXED0:
3065 msr = &vmx->nested.nested_vmx_cr0_fixed0;
3067 case MSR_IA32_VMX_CR4_FIXED0:
3068 msr = &vmx->nested.nested_vmx_cr4_fixed0;
3075 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
3076 * must be 1 in the restored value.
3078 if (!is_bitwise_subset(data, *msr, -1ULL))
3086 * Called when userspace is restoring VMX MSRs.
3088 * Returns 0 on success, non-0 otherwise.
3090 static int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
3092 struct vcpu_vmx *vmx = to_vmx(vcpu);
3094 switch (msr_index) {
3095 case MSR_IA32_VMX_BASIC:
3096 return vmx_restore_vmx_basic(vmx, data);
3097 case MSR_IA32_VMX_PINBASED_CTLS:
3098 case MSR_IA32_VMX_PROCBASED_CTLS:
3099 case MSR_IA32_VMX_EXIT_CTLS:
3100 case MSR_IA32_VMX_ENTRY_CTLS:
3102 * The "non-true" VMX capability MSRs are generated from the
3103 * "true" MSRs, so we do not support restoring them directly.
3105 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
3106 * should restore the "true" MSRs with the must-be-1 bits
3107 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
3108 * DEFAULT SETTINGS".
3111 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
3112 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
3113 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
3114 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
3115 case MSR_IA32_VMX_PROCBASED_CTLS2:
3116 return vmx_restore_control_msr(vmx, msr_index, data);
3117 case MSR_IA32_VMX_MISC:
3118 return vmx_restore_vmx_misc(vmx, data);
3119 case MSR_IA32_VMX_CR0_FIXED0:
3120 case MSR_IA32_VMX_CR4_FIXED0:
3121 return vmx_restore_fixed0_msr(vmx, msr_index, data);
3122 case MSR_IA32_VMX_CR0_FIXED1:
3123 case MSR_IA32_VMX_CR4_FIXED1:
3125 * These MSRs are generated based on the vCPU's CPUID, so we
3126 * do not support restoring them directly.
3129 case MSR_IA32_VMX_EPT_VPID_CAP:
3130 return vmx_restore_vmx_ept_vpid_cap(vmx, data);
3131 case MSR_IA32_VMX_VMCS_ENUM:
3132 vmx->nested.nested_vmx_vmcs_enum = data;
3136 * The rest of the VMX capability MSRs do not support restore.
3142 /* Returns 0 on success, non-0 otherwise. */
3143 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
3145 struct vcpu_vmx *vmx = to_vmx(vcpu);
3147 switch (msr_index) {
3148 case MSR_IA32_VMX_BASIC:
3149 *pdata = vmx->nested.nested_vmx_basic;
3151 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
3152 case MSR_IA32_VMX_PINBASED_CTLS:
3153 *pdata = vmx_control_msr(
3154 vmx->nested.nested_vmx_pinbased_ctls_low,
3155 vmx->nested.nested_vmx_pinbased_ctls_high);
3156 if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
3157 *pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
3159 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
3160 case MSR_IA32_VMX_PROCBASED_CTLS:
3161 *pdata = vmx_control_msr(
3162 vmx->nested.nested_vmx_procbased_ctls_low,
3163 vmx->nested.nested_vmx_procbased_ctls_high);
3164 if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
3165 *pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
3167 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
3168 case MSR_IA32_VMX_EXIT_CTLS:
3169 *pdata = vmx_control_msr(
3170 vmx->nested.nested_vmx_exit_ctls_low,
3171 vmx->nested.nested_vmx_exit_ctls_high);
3172 if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
3173 *pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
3175 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
3176 case MSR_IA32_VMX_ENTRY_CTLS:
3177 *pdata = vmx_control_msr(
3178 vmx->nested.nested_vmx_entry_ctls_low,
3179 vmx->nested.nested_vmx_entry_ctls_high);
3180 if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
3181 *pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
3183 case MSR_IA32_VMX_MISC:
3184 *pdata = vmx_control_msr(
3185 vmx->nested.nested_vmx_misc_low,
3186 vmx->nested.nested_vmx_misc_high);
3188 case MSR_IA32_VMX_CR0_FIXED0:
3189 *pdata = vmx->nested.nested_vmx_cr0_fixed0;
3191 case MSR_IA32_VMX_CR0_FIXED1:
3192 *pdata = vmx->nested.nested_vmx_cr0_fixed1;
3194 case MSR_IA32_VMX_CR4_FIXED0:
3195 *pdata = vmx->nested.nested_vmx_cr4_fixed0;
3197 case MSR_IA32_VMX_CR4_FIXED1:
3198 *pdata = vmx->nested.nested_vmx_cr4_fixed1;
3200 case MSR_IA32_VMX_VMCS_ENUM:
3201 *pdata = vmx->nested.nested_vmx_vmcs_enum;
3203 case MSR_IA32_VMX_PROCBASED_CTLS2:
3204 *pdata = vmx_control_msr(
3205 vmx->nested.nested_vmx_secondary_ctls_low,
3206 vmx->nested.nested_vmx_secondary_ctls_high);
3208 case MSR_IA32_VMX_EPT_VPID_CAP:
3209 *pdata = vmx->nested.nested_vmx_ept_caps |
3210 ((u64)vmx->nested.nested_vmx_vpid_caps << 32);
3212 case MSR_IA32_VMX_VMFUNC:
3213 *pdata = vmx->nested.nested_vmx_vmfunc_controls;
3222 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
3225 uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
3227 return !(val & ~valid_bits);
3231 * Reads an msr value (of 'msr_index') into 'pdata'.
3232 * Returns 0 on success, non-0 otherwise.
3233 * Assumes vcpu_load() was already called.
3235 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3237 struct vcpu_vmx *vmx = to_vmx(vcpu);
3238 struct shared_msr_entry *msr;
3240 switch (msr_info->index) {
3241 #ifdef CONFIG_X86_64
3243 msr_info->data = vmcs_readl(GUEST_FS_BASE);
3246 msr_info->data = vmcs_readl(GUEST_GS_BASE);
3248 case MSR_KERNEL_GS_BASE:
3249 vmx_load_host_state(vmx);
3250 msr_info->data = vmx->msr_guest_kernel_gs_base;
3254 return kvm_get_msr_common(vcpu, msr_info);
3256 msr_info->data = guest_read_tsc(vcpu);
3258 case MSR_IA32_SPEC_CTRL:
3259 if (!msr_info->host_initiated &&
3260 !guest_cpuid_has(vcpu, X86_FEATURE_IBRS) &&
3261 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
3264 msr_info->data = to_vmx(vcpu)->spec_ctrl;
3266 case MSR_IA32_ARCH_CAPABILITIES:
3267 if (!msr_info->host_initiated &&
3268 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3270 msr_info->data = to_vmx(vcpu)->arch_capabilities;
3272 case MSR_IA32_SYSENTER_CS:
3273 msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
3275 case MSR_IA32_SYSENTER_EIP:
3276 msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
3278 case MSR_IA32_SYSENTER_ESP:
3279 msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
3281 case MSR_IA32_BNDCFGS:
3282 if (!kvm_mpx_supported() ||
3283 (!msr_info->host_initiated &&
3284 !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
3286 msr_info->data = vmcs_read64(GUEST_BNDCFGS);
3288 case MSR_IA32_MCG_EXT_CTL:
3289 if (!msr_info->host_initiated &&
3290 !(vmx->msr_ia32_feature_control &
3291 FEATURE_CONTROL_LMCE))
3293 msr_info->data = vcpu->arch.mcg_ext_ctl;
3295 case MSR_IA32_FEATURE_CONTROL:
3296 msr_info->data = vmx->msr_ia32_feature_control;
3298 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3299 if (!nested_vmx_allowed(vcpu))
3301 return vmx_get_vmx_msr(vcpu, msr_info->index, &msr_info->data);
3303 if (!vmx_xsaves_supported())
3305 msr_info->data = vcpu->arch.ia32_xss;
3308 if (!msr_info->host_initiated &&
3309 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
3311 /* Otherwise falls through */
3313 msr = find_msr_entry(vmx, msr_info->index);
3315 msr_info->data = msr->data;
3318 return kvm_get_msr_common(vcpu, msr_info);
3324 static void vmx_leave_nested(struct kvm_vcpu *vcpu);
3327 * Writes msr value into into the appropriate "register".
3328 * Returns 0 on success, non-0 otherwise.
3329 * Assumes vcpu_load() was already called.
3331 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3333 struct vcpu_vmx *vmx = to_vmx(vcpu);
3334 struct shared_msr_entry *msr;
3336 u32 msr_index = msr_info->index;
3337 u64 data = msr_info->data;
3339 switch (msr_index) {
3341 ret = kvm_set_msr_common(vcpu, msr_info);
3343 #ifdef CONFIG_X86_64
3345 vmx_segment_cache_clear(vmx);
3346 vmcs_writel(GUEST_FS_BASE, data);
3349 vmx_segment_cache_clear(vmx);
3350 vmcs_writel(GUEST_GS_BASE, data);
3352 case MSR_KERNEL_GS_BASE:
3353 vmx_load_host_state(vmx);
3354 vmx->msr_guest_kernel_gs_base = data;
3357 case MSR_IA32_SYSENTER_CS:
3358 vmcs_write32(GUEST_SYSENTER_CS, data);
3360 case MSR_IA32_SYSENTER_EIP:
3361 vmcs_writel(GUEST_SYSENTER_EIP, data);
3363 case MSR_IA32_SYSENTER_ESP:
3364 vmcs_writel(GUEST_SYSENTER_ESP, data);
3366 case MSR_IA32_BNDCFGS:
3367 if (!kvm_mpx_supported() ||
3368 (!msr_info->host_initiated &&
3369 !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
3371 if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
3372 (data & MSR_IA32_BNDCFGS_RSVD))
3374 vmcs_write64(GUEST_BNDCFGS, data);
3377 kvm_write_tsc(vcpu, msr_info);
3379 case MSR_IA32_SPEC_CTRL:
3380 if (!msr_info->host_initiated &&
3381 !guest_cpuid_has(vcpu, X86_FEATURE_IBRS) &&
3382 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
3385 /* The STIBP bit doesn't fault even if it's not advertised */
3386 if (data & ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP))
3389 vmx->spec_ctrl = data;
3396 * When it's written (to non-zero) for the first time, pass
3400 * The handling of the MSR bitmap for L2 guests is done in
3401 * nested_vmx_merge_msr_bitmap. We should not touch the
3402 * vmcs02.msr_bitmap here since it gets completely overwritten
3403 * in the merging. We update the vmcs01 here for L1 as well
3404 * since it will end up touching the MSR anyway now.
3406 vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap,
3410 case MSR_IA32_PRED_CMD:
3411 if (!msr_info->host_initiated &&
3412 !guest_cpuid_has(vcpu, X86_FEATURE_IBPB) &&
3413 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL))
3416 if (data & ~PRED_CMD_IBPB)
3422 wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
3426 * When it's written (to non-zero) for the first time, pass
3430 * The handling of the MSR bitmap for L2 guests is done in
3431 * nested_vmx_merge_msr_bitmap. We should not touch the
3432 * vmcs02.msr_bitmap here since it gets completely overwritten
3435 vmx_disable_intercept_for_msr(vmx->vmcs01.msr_bitmap, MSR_IA32_PRED_CMD,
3438 case MSR_IA32_ARCH_CAPABILITIES:
3439 if (!msr_info->host_initiated)
3441 vmx->arch_capabilities = data;
3443 case MSR_IA32_CR_PAT:
3444 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
3445 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
3447 vmcs_write64(GUEST_IA32_PAT, data);
3448 vcpu->arch.pat = data;
3451 ret = kvm_set_msr_common(vcpu, msr_info);
3453 case MSR_IA32_TSC_ADJUST:
3454 ret = kvm_set_msr_common(vcpu, msr_info);
3456 case MSR_IA32_MCG_EXT_CTL:
3457 if ((!msr_info->host_initiated &&
3458 !(to_vmx(vcpu)->msr_ia32_feature_control &
3459 FEATURE_CONTROL_LMCE)) ||
3460 (data & ~MCG_EXT_CTL_LMCE_EN))
3462 vcpu->arch.mcg_ext_ctl = data;
3464 case MSR_IA32_FEATURE_CONTROL:
3465 if (!vmx_feature_control_msr_valid(vcpu, data) ||
3466 (to_vmx(vcpu)->msr_ia32_feature_control &
3467 FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
3469 vmx->msr_ia32_feature_control = data;
3470 if (msr_info->host_initiated && data == 0)
3471 vmx_leave_nested(vcpu);
3473 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3474 if (!msr_info->host_initiated)
3475 return 1; /* they are read-only */
3476 if (!nested_vmx_allowed(vcpu))
3478 return vmx_set_vmx_msr(vcpu, msr_index, data);
3480 if (!vmx_xsaves_supported())
3483 * The only supported bit as of Skylake is bit 8, but
3484 * it is not supported on KVM.
3488 vcpu->arch.ia32_xss = data;
3489 if (vcpu->arch.ia32_xss != host_xss)
3490 add_atomic_switch_msr(vmx, MSR_IA32_XSS,
3491 vcpu->arch.ia32_xss, host_xss);
3493 clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
3496 if (!msr_info->host_initiated &&
3497 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
3499 /* Check reserved bit, higher 32 bits should be zero */
3500 if ((data >> 32) != 0)
3502 /* Otherwise falls through */
3504 msr = find_msr_entry(vmx, msr_index);
3506 u64 old_msr_data = msr->data;
3508 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
3510 ret = kvm_set_shared_msr(msr->index, msr->data,
3514 msr->data = old_msr_data;
3518 ret = kvm_set_msr_common(vcpu, msr_info);
3524 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
3526 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
3529 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
3532 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
3534 case VCPU_EXREG_PDPTR:
3536 ept_save_pdptrs(vcpu);
3543 static __init int cpu_has_kvm_support(void)
3545 return cpu_has_vmx();
3548 static __init int vmx_disabled_by_bios(void)
3552 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
3553 if (msr & FEATURE_CONTROL_LOCKED) {
3554 /* launched w/ TXT and VMX disabled */
3555 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3558 /* launched w/o TXT and VMX only enabled w/ TXT */
3559 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3560 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3561 && !tboot_enabled()) {
3562 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
3563 "activate TXT before enabling KVM\n");
3566 /* launched w/o TXT and VMX disabled */
3567 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3568 && !tboot_enabled())
3575 static void kvm_cpu_vmxon(u64 addr)
3577 cr4_set_bits(X86_CR4_VMXE);
3578 intel_pt_handle_vmx(1);
3580 asm volatile (ASM_VMX_VMXON_RAX
3581 : : "a"(&addr), "m"(addr)
3585 static int hardware_enable(void)
3587 int cpu = raw_smp_processor_id();
3588 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
3591 if (cr4_read_shadow() & X86_CR4_VMXE)
3594 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
3595 INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
3596 spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
3599 * Now we can enable the vmclear operation in kdump
3600 * since the loaded_vmcss_on_cpu list on this cpu
3601 * has been initialized.
3603 * Though the cpu is not in VMX operation now, there
3604 * is no problem to enable the vmclear operation
3605 * for the loaded_vmcss_on_cpu list is empty!
3607 crash_enable_local_vmclear(cpu);
3609 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
3611 test_bits = FEATURE_CONTROL_LOCKED;
3612 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
3613 if (tboot_enabled())
3614 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
3616 if ((old & test_bits) != test_bits) {
3617 /* enable and lock */
3618 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
3620 kvm_cpu_vmxon(phys_addr);
3627 static void vmclear_local_loaded_vmcss(void)
3629 int cpu = raw_smp_processor_id();
3630 struct loaded_vmcs *v, *n;
3632 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
3633 loaded_vmcss_on_cpu_link)
3634 __loaded_vmcs_clear(v);
3638 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
3641 static void kvm_cpu_vmxoff(void)
3643 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
3645 intel_pt_handle_vmx(0);
3646 cr4_clear_bits(X86_CR4_VMXE);
3649 static void hardware_disable(void)
3651 vmclear_local_loaded_vmcss();
3655 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
3656 u32 msr, u32 *result)
3658 u32 vmx_msr_low, vmx_msr_high;
3659 u32 ctl = ctl_min | ctl_opt;
3661 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3663 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
3664 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
3666 /* Ensure minimum (required) set of control bits are supported. */
3674 static __init bool allow_1_setting(u32 msr, u32 ctl)
3676 u32 vmx_msr_low, vmx_msr_high;
3678 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3679 return vmx_msr_high & ctl;
3682 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
3684 u32 vmx_msr_low, vmx_msr_high;
3685 u32 min, opt, min2, opt2;
3686 u32 _pin_based_exec_control = 0;
3687 u32 _cpu_based_exec_control = 0;
3688 u32 _cpu_based_2nd_exec_control = 0;
3689 u32 _vmexit_control = 0;
3690 u32 _vmentry_control = 0;
3692 min = CPU_BASED_HLT_EXITING |
3693 #ifdef CONFIG_X86_64
3694 CPU_BASED_CR8_LOAD_EXITING |
3695 CPU_BASED_CR8_STORE_EXITING |
3697 CPU_BASED_CR3_LOAD_EXITING |
3698 CPU_BASED_CR3_STORE_EXITING |
3699 CPU_BASED_UNCOND_IO_EXITING |
3700 CPU_BASED_MOV_DR_EXITING |
3701 CPU_BASED_USE_TSC_OFFSETING |
3702 CPU_BASED_INVLPG_EXITING |
3703 CPU_BASED_RDPMC_EXITING;
3705 if (!kvm_mwait_in_guest())
3706 min |= CPU_BASED_MWAIT_EXITING |
3707 CPU_BASED_MONITOR_EXITING;
3709 opt = CPU_BASED_TPR_SHADOW |
3710 CPU_BASED_USE_MSR_BITMAPS |
3711 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
3712 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
3713 &_cpu_based_exec_control) < 0)
3715 #ifdef CONFIG_X86_64
3716 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3717 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
3718 ~CPU_BASED_CR8_STORE_EXITING;
3720 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
3722 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
3723 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3724 SECONDARY_EXEC_WBINVD_EXITING |
3725 SECONDARY_EXEC_ENABLE_VPID |
3726 SECONDARY_EXEC_ENABLE_EPT |
3727 SECONDARY_EXEC_UNRESTRICTED_GUEST |
3728 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
3729 SECONDARY_EXEC_DESC |
3730 SECONDARY_EXEC_RDTSCP |
3731 SECONDARY_EXEC_ENABLE_INVPCID |
3732 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3733 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
3734 SECONDARY_EXEC_SHADOW_VMCS |
3735 SECONDARY_EXEC_XSAVES |
3736 SECONDARY_EXEC_RDSEED_EXITING |
3737 SECONDARY_EXEC_RDRAND_EXITING |
3738 SECONDARY_EXEC_ENABLE_PML |
3739 SECONDARY_EXEC_TSC_SCALING |
3740 SECONDARY_EXEC_ENABLE_VMFUNC;
3741 if (adjust_vmx_controls(min2, opt2,
3742 MSR_IA32_VMX_PROCBASED_CTLS2,
3743 &_cpu_based_2nd_exec_control) < 0)
3746 #ifndef CONFIG_X86_64
3747 if (!(_cpu_based_2nd_exec_control &
3748 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
3749 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
3752 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3753 _cpu_based_2nd_exec_control &= ~(
3754 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3755 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3756 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3758 rdmsr_safe(MSR_IA32_VMX_EPT_VPID_CAP,
3759 &vmx_capability.ept, &vmx_capability.vpid);
3761 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
3762 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
3764 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
3765 CPU_BASED_CR3_STORE_EXITING |
3766 CPU_BASED_INVLPG_EXITING);
3767 } else if (vmx_capability.ept) {
3768 vmx_capability.ept = 0;
3769 pr_warn_once("EPT CAP should not exist if not support "
3770 "1-setting enable EPT VM-execution control\n");
3772 if (!(_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_VPID) &&
3773 vmx_capability.vpid) {
3774 vmx_capability.vpid = 0;
3775 pr_warn_once("VPID CAP should not exist if not support "
3776 "1-setting enable VPID VM-execution control\n");
3779 min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
3780 #ifdef CONFIG_X86_64
3781 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
3783 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT |
3784 VM_EXIT_CLEAR_BNDCFGS;
3785 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
3786 &_vmexit_control) < 0)
3789 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
3790 opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
3791 PIN_BASED_VMX_PREEMPTION_TIMER;
3792 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
3793 &_pin_based_exec_control) < 0)
3796 if (cpu_has_broken_vmx_preemption_timer())
3797 _pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3798 if (!(_cpu_based_2nd_exec_control &
3799 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
3800 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
3802 min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
3803 opt = VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS;
3804 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
3805 &_vmentry_control) < 0)
3808 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
3810 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
3811 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
3814 #ifdef CONFIG_X86_64
3815 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
3816 if (vmx_msr_high & (1u<<16))
3820 /* Require Write-Back (WB) memory type for VMCS accesses. */
3821 if (((vmx_msr_high >> 18) & 15) != 6)
3824 vmcs_conf->size = vmx_msr_high & 0x1fff;
3825 vmcs_conf->order = get_order(vmcs_conf->size);
3826 vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
3827 vmcs_conf->revision_id = vmx_msr_low;
3829 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
3830 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
3831 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
3832 vmcs_conf->vmexit_ctrl = _vmexit_control;
3833 vmcs_conf->vmentry_ctrl = _vmentry_control;
3835 cpu_has_load_ia32_efer =
3836 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3837 VM_ENTRY_LOAD_IA32_EFER)
3838 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3839 VM_EXIT_LOAD_IA32_EFER);
3841 cpu_has_load_perf_global_ctrl =
3842 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3843 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
3844 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3845 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
3848 * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
3849 * but due to errata below it can't be used. Workaround is to use
3850 * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
3852 * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
3857 * BC86,AAY89,BD102 (model 44)
3861 if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
3862 switch (boot_cpu_data.x86_model) {
3868 cpu_has_load_perf_global_ctrl = false;
3869 printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
3870 "does not work properly. Using workaround\n");
3877 if (boot_cpu_has(X86_FEATURE_XSAVES))
3878 rdmsrl(MSR_IA32_XSS, host_xss);
3883 static struct vmcs *alloc_vmcs_cpu(int cpu)
3885 int node = cpu_to_node(cpu);
3889 pages = __alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
3892 vmcs = page_address(pages);
3893 memset(vmcs, 0, vmcs_config.size);
3894 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
3898 static void free_vmcs(struct vmcs *vmcs)
3900 free_pages((unsigned long)vmcs, vmcs_config.order);
3904 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
3906 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
3908 if (!loaded_vmcs->vmcs)
3910 loaded_vmcs_clear(loaded_vmcs);
3911 free_vmcs(loaded_vmcs->vmcs);
3912 loaded_vmcs->vmcs = NULL;
3913 if (loaded_vmcs->msr_bitmap)
3914 free_page((unsigned long)loaded_vmcs->msr_bitmap);
3915 WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
3918 static struct vmcs *alloc_vmcs(void)
3920 return alloc_vmcs_cpu(raw_smp_processor_id());
3923 static int alloc_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
3925 loaded_vmcs->vmcs = alloc_vmcs();
3926 if (!loaded_vmcs->vmcs)
3929 loaded_vmcs->shadow_vmcs = NULL;
3930 loaded_vmcs_init(loaded_vmcs);
3932 if (cpu_has_vmx_msr_bitmap()) {
3933 loaded_vmcs->msr_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
3934 if (!loaded_vmcs->msr_bitmap)
3936 memset(loaded_vmcs->msr_bitmap, 0xff, PAGE_SIZE);
3941 free_loaded_vmcs(loaded_vmcs);
3945 static void free_kvm_area(void)
3949 for_each_possible_cpu(cpu) {
3950 free_vmcs(per_cpu(vmxarea, cpu));
3951 per_cpu(vmxarea, cpu) = NULL;
3955 enum vmcs_field_width {
3956 VMCS_FIELD_WIDTH_U16 = 0,
3957 VMCS_FIELD_WIDTH_U64 = 1,
3958 VMCS_FIELD_WIDTH_U32 = 2,
3959 VMCS_FIELD_WIDTH_NATURAL_WIDTH = 3
3962 static inline int vmcs_field_width(unsigned long field)
3964 if (0x1 & field) /* the *_HIGH fields are all 32 bit */
3965 return VMCS_FIELD_WIDTH_U32;
3966 return (field >> 13) & 0x3 ;
3969 static inline int vmcs_field_readonly(unsigned long field)
3971 return (((field >> 10) & 0x3) == 1);
3974 static void init_vmcs_shadow_fields(void)
3978 for (i = j = 0; i < max_shadow_read_only_fields; i++) {
3979 u16 field = shadow_read_only_fields[i];
3980 if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
3981 (i + 1 == max_shadow_read_only_fields ||
3982 shadow_read_only_fields[i + 1] != field + 1))
3983 pr_err("Missing field from shadow_read_only_field %x\n",
3986 clear_bit(field, vmx_vmread_bitmap);
3987 #ifdef CONFIG_X86_64
3992 shadow_read_only_fields[j] = field;
3995 max_shadow_read_only_fields = j;
3997 for (i = j = 0; i < max_shadow_read_write_fields; i++) {
3998 u16 field = shadow_read_write_fields[i];
3999 if (vmcs_field_width(field) == VMCS_FIELD_WIDTH_U64 &&
4000 (i + 1 == max_shadow_read_write_fields ||
4001 shadow_read_write_fields[i + 1] != field + 1))
4002 pr_err("Missing field from shadow_read_write_field %x\n",
4006 * PML and the preemption timer can be emulated, but the
4007 * processor cannot vmwrite to fields that don't exist
4011 case GUEST_PML_INDEX:
4012 if (!cpu_has_vmx_pml())
4015 case VMX_PREEMPTION_TIMER_VALUE:
4016 if (!cpu_has_vmx_preemption_timer())
4019 case GUEST_INTR_STATUS:
4020 if (!cpu_has_vmx_apicv())
4027 clear_bit(field, vmx_vmwrite_bitmap);
4028 clear_bit(field, vmx_vmread_bitmap);
4029 #ifdef CONFIG_X86_64
4034 shadow_read_write_fields[j] = field;
4037 max_shadow_read_write_fields = j;
4040 static __init int alloc_kvm_area(void)
4044 for_each_possible_cpu(cpu) {
4047 vmcs = alloc_vmcs_cpu(cpu);
4053 per_cpu(vmxarea, cpu) = vmcs;
4058 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
4059 struct kvm_segment *save)
4061 if (!emulate_invalid_guest_state) {
4063 * CS and SS RPL should be equal during guest entry according
4064 * to VMX spec, but in reality it is not always so. Since vcpu
4065 * is in the middle of the transition from real mode to
4066 * protected mode it is safe to assume that RPL 0 is a good
4069 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
4070 save->selector &= ~SEGMENT_RPL_MASK;
4071 save->dpl = save->selector & SEGMENT_RPL_MASK;
4074 vmx_set_segment(vcpu, save, seg);
4077 static void enter_pmode(struct kvm_vcpu *vcpu)
4079 unsigned long flags;
4080 struct vcpu_vmx *vmx = to_vmx(vcpu);
4083 * Update real mode segment cache. It may be not up-to-date if sement
4084 * register was written while vcpu was in a guest mode.
4086 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
4087 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
4088 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
4089 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
4090 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
4091 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
4093 vmx->rmode.vm86_active = 0;
4095 vmx_segment_cache_clear(vmx);
4097 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
4099 flags = vmcs_readl(GUEST_RFLAGS);
4100 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
4101 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
4102 vmcs_writel(GUEST_RFLAGS, flags);
4104 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
4105 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
4107 update_exception_bitmap(vcpu);
4109 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
4110 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
4111 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
4112 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
4113 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
4114 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
4117 static void fix_rmode_seg(int seg, struct kvm_segment *save)
4119 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4120 struct kvm_segment var = *save;
4123 if (seg == VCPU_SREG_CS)
4126 if (!emulate_invalid_guest_state) {
4127 var.selector = var.base >> 4;
4128 var.base = var.base & 0xffff0;
4138 if (save->base & 0xf)
4139 printk_once(KERN_WARNING "kvm: segment base is not "
4140 "paragraph aligned when entering "
4141 "protected mode (seg=%d)", seg);
4144 vmcs_write16(sf->selector, var.selector);
4145 vmcs_writel(sf->base, var.base);
4146 vmcs_write32(sf->limit, var.limit);
4147 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
4150 static void enter_rmode(struct kvm_vcpu *vcpu)
4152 unsigned long flags;
4153 struct vcpu_vmx *vmx = to_vmx(vcpu);
4155 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
4156 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
4157 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
4158 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
4159 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
4160 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
4161 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
4163 vmx->rmode.vm86_active = 1;
4166 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
4167 * vcpu. Warn the user that an update is overdue.
4169 if (!vcpu->kvm->arch.tss_addr)
4170 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
4171 "called before entering vcpu\n");
4173 vmx_segment_cache_clear(vmx);
4175 vmcs_writel(GUEST_TR_BASE, vcpu->kvm->arch.tss_addr);
4176 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
4177 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4179 flags = vmcs_readl(GUEST_RFLAGS);
4180 vmx->rmode.save_rflags = flags;
4182 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
4184 vmcs_writel(GUEST_RFLAGS, flags);
4185 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
4186 update_exception_bitmap(vcpu);
4188 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
4189 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
4190 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
4191 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
4192 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
4193 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
4195 kvm_mmu_reset_context(vcpu);
4198 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
4200 struct vcpu_vmx *vmx = to_vmx(vcpu);
4201 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
4207 * Force kernel_gs_base reloading before EFER changes, as control
4208 * of this msr depends on is_long_mode().
4210 vmx_load_host_state(to_vmx(vcpu));
4211 vcpu->arch.efer = efer;
4212 if (efer & EFER_LMA) {
4213 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
4216 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
4218 msr->data = efer & ~EFER_LME;
4223 #ifdef CONFIG_X86_64
4225 static void enter_lmode(struct kvm_vcpu *vcpu)
4229 vmx_segment_cache_clear(to_vmx(vcpu));
4231 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
4232 if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
4233 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
4235 vmcs_write32(GUEST_TR_AR_BYTES,
4236 (guest_tr_ar & ~VMX_AR_TYPE_MASK)
4237 | VMX_AR_TYPE_BUSY_64_TSS);
4239 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
4242 static void exit_lmode(struct kvm_vcpu *vcpu)
4244 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
4245 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
4250 static inline void __vmx_flush_tlb(struct kvm_vcpu *vcpu, int vpid,
4251 bool invalidate_gpa)
4253 if (enable_ept && (invalidate_gpa || !enable_vpid)) {
4254 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
4256 ept_sync_context(construct_eptp(vcpu, vcpu->arch.mmu.root_hpa));
4258 vpid_sync_context(vpid);
4262 static void vmx_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
4264 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->vpid, invalidate_gpa);
4267 static void vmx_flush_tlb_ept_only(struct kvm_vcpu *vcpu)
4270 vmx_flush_tlb(vcpu, true);
4273 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
4275 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
4277 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
4278 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
4281 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
4283 if (enable_ept && is_paging(vcpu))
4284 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
4285 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
4288 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
4290 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
4292 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
4293 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
4296 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
4298 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4300 if (!test_bit(VCPU_EXREG_PDPTR,
4301 (unsigned long *)&vcpu->arch.regs_dirty))
4304 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
4305 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
4306 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
4307 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
4308 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
4312 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
4314 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4316 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
4317 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
4318 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
4319 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
4320 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
4323 __set_bit(VCPU_EXREG_PDPTR,
4324 (unsigned long *)&vcpu->arch.regs_avail);
4325 __set_bit(VCPU_EXREG_PDPTR,
4326 (unsigned long *)&vcpu->arch.regs_dirty);
4329 static bool nested_guest_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
4331 u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed0;
4332 u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed1;
4333 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4335 if (to_vmx(vcpu)->nested.nested_vmx_secondary_ctls_high &
4336 SECONDARY_EXEC_UNRESTRICTED_GUEST &&
4337 nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
4338 fixed0 &= ~(X86_CR0_PE | X86_CR0_PG);
4340 return fixed_bits_valid(val, fixed0, fixed1);
4343 static bool nested_host_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
4345 u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed0;
4346 u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed1;
4348 return fixed_bits_valid(val, fixed0, fixed1);
4351 static bool nested_cr4_valid(struct kvm_vcpu *vcpu, unsigned long val)
4353 u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr4_fixed0;
4354 u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr4_fixed1;
4356 return fixed_bits_valid(val, fixed0, fixed1);
4359 /* No difference in the restrictions on guest and host CR4 in VMX operation. */
4360 #define nested_guest_cr4_valid nested_cr4_valid
4361 #define nested_host_cr4_valid nested_cr4_valid
4363 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
4365 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
4367 struct kvm_vcpu *vcpu)
4369 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
4370 vmx_decache_cr3(vcpu);
4371 if (!(cr0 & X86_CR0_PG)) {
4372 /* From paging/starting to nonpaging */
4373 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
4374 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
4375 (CPU_BASED_CR3_LOAD_EXITING |
4376 CPU_BASED_CR3_STORE_EXITING));
4377 vcpu->arch.cr0 = cr0;
4378 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
4379 } else if (!is_paging(vcpu)) {
4380 /* From nonpaging to paging */
4381 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
4382 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
4383 ~(CPU_BASED_CR3_LOAD_EXITING |
4384 CPU_BASED_CR3_STORE_EXITING));
4385 vcpu->arch.cr0 = cr0;
4386 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
4389 if (!(cr0 & X86_CR0_WP))
4390 *hw_cr0 &= ~X86_CR0_WP;
4393 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
4395 struct vcpu_vmx *vmx = to_vmx(vcpu);
4396 unsigned long hw_cr0;
4398 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK);
4399 if (enable_unrestricted_guest)
4400 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
4402 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
4404 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
4407 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
4411 #ifdef CONFIG_X86_64
4412 if (vcpu->arch.efer & EFER_LME) {
4413 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
4415 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
4421 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
4423 vmcs_writel(CR0_READ_SHADOW, cr0);
4424 vmcs_writel(GUEST_CR0, hw_cr0);
4425 vcpu->arch.cr0 = cr0;
4427 /* depends on vcpu->arch.cr0 to be set to a new value */
4428 vmx->emulation_required = emulation_required(vcpu);
4431 static int get_ept_level(struct kvm_vcpu *vcpu)
4433 if (cpu_has_vmx_ept_5levels() && (cpuid_maxphyaddr(vcpu) > 48))
4438 static u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa)
4440 u64 eptp = VMX_EPTP_MT_WB;
4442 eptp |= (get_ept_level(vcpu) == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
4444 if (enable_ept_ad_bits &&
4445 (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
4446 eptp |= VMX_EPTP_AD_ENABLE_BIT;
4447 eptp |= (root_hpa & PAGE_MASK);
4452 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
4454 unsigned long guest_cr3;
4459 eptp = construct_eptp(vcpu, cr3);
4460 vmcs_write64(EPT_POINTER, eptp);
4461 if (is_paging(vcpu) || is_guest_mode(vcpu))
4462 guest_cr3 = kvm_read_cr3(vcpu);
4464 guest_cr3 = vcpu->kvm->arch.ept_identity_map_addr;
4465 ept_load_pdptrs(vcpu);
4468 vmx_flush_tlb(vcpu, true);
4469 vmcs_writel(GUEST_CR3, guest_cr3);
4472 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
4475 * Pass through host's Machine Check Enable value to hw_cr4, which
4476 * is in force while we are in guest mode. Do not let guests control
4477 * this bit, even if host CR4.MCE == 0.
4479 unsigned long hw_cr4 =
4480 (cr4_read_shadow() & X86_CR4_MCE) |
4481 (cr4 & ~X86_CR4_MCE) |
4482 (to_vmx(vcpu)->rmode.vm86_active ?
4483 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
4485 if ((cr4 & X86_CR4_UMIP) && !boot_cpu_has(X86_FEATURE_UMIP)) {
4486 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
4487 SECONDARY_EXEC_DESC);
4488 hw_cr4 &= ~X86_CR4_UMIP;
4490 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
4491 SECONDARY_EXEC_DESC);
4493 if (cr4 & X86_CR4_VMXE) {
4495 * To use VMXON (and later other VMX instructions), a guest
4496 * must first be able to turn on cr4.VMXE (see handle_vmon()).
4497 * So basically the check on whether to allow nested VMX
4500 if (!nested_vmx_allowed(vcpu))
4504 if (to_vmx(vcpu)->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
4507 vcpu->arch.cr4 = cr4;
4509 if (!is_paging(vcpu)) {
4510 hw_cr4 &= ~X86_CR4_PAE;
4511 hw_cr4 |= X86_CR4_PSE;
4512 } else if (!(cr4 & X86_CR4_PAE)) {
4513 hw_cr4 &= ~X86_CR4_PAE;
4517 if (!enable_unrestricted_guest && !is_paging(vcpu))
4519 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
4520 * hardware. To emulate this behavior, SMEP/SMAP/PKU needs
4521 * to be manually disabled when guest switches to non-paging
4524 * If !enable_unrestricted_guest, the CPU is always running
4525 * with CR0.PG=1 and CR4 needs to be modified.
4526 * If enable_unrestricted_guest, the CPU automatically
4527 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
4529 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
4531 vmcs_writel(CR4_READ_SHADOW, cr4);
4532 vmcs_writel(GUEST_CR4, hw_cr4);
4536 static void vmx_get_segment(struct kvm_vcpu *vcpu,
4537 struct kvm_segment *var, int seg)
4539 struct vcpu_vmx *vmx = to_vmx(vcpu);
4542 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
4543 *var = vmx->rmode.segs[seg];
4544 if (seg == VCPU_SREG_TR
4545 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
4547 var->base = vmx_read_guest_seg_base(vmx, seg);
4548 var->selector = vmx_read_guest_seg_selector(vmx, seg);
4551 var->base = vmx_read_guest_seg_base(vmx, seg);
4552 var->limit = vmx_read_guest_seg_limit(vmx, seg);
4553 var->selector = vmx_read_guest_seg_selector(vmx, seg);
4554 ar = vmx_read_guest_seg_ar(vmx, seg);
4555 var->unusable = (ar >> 16) & 1;
4556 var->type = ar & 15;
4557 var->s = (ar >> 4) & 1;
4558 var->dpl = (ar >> 5) & 3;
4560 * Some userspaces do not preserve unusable property. Since usable
4561 * segment has to be present according to VMX spec we can use present
4562 * property to amend userspace bug by making unusable segment always
4563 * nonpresent. vmx_segment_access_rights() already marks nonpresent
4564 * segment as unusable.
4566 var->present = !var->unusable;
4567 var->avl = (ar >> 12) & 1;
4568 var->l = (ar >> 13) & 1;
4569 var->db = (ar >> 14) & 1;
4570 var->g = (ar >> 15) & 1;
4573 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
4575 struct kvm_segment s;
4577 if (to_vmx(vcpu)->rmode.vm86_active) {
4578 vmx_get_segment(vcpu, &s, seg);
4581 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
4584 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
4586 struct vcpu_vmx *vmx = to_vmx(vcpu);
4588 if (unlikely(vmx->rmode.vm86_active))
4591 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
4592 return VMX_AR_DPL(ar);
4596 static u32 vmx_segment_access_rights(struct kvm_segment *var)
4600 if (var->unusable || !var->present)
4603 ar = var->type & 15;
4604 ar |= (var->s & 1) << 4;
4605 ar |= (var->dpl & 3) << 5;
4606 ar |= (var->present & 1) << 7;
4607 ar |= (var->avl & 1) << 12;
4608 ar |= (var->l & 1) << 13;
4609 ar |= (var->db & 1) << 14;
4610 ar |= (var->g & 1) << 15;
4616 static void vmx_set_segment(struct kvm_vcpu *vcpu,
4617 struct kvm_segment *var, int seg)
4619 struct vcpu_vmx *vmx = to_vmx(vcpu);
4620 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4622 vmx_segment_cache_clear(vmx);
4624 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
4625 vmx->rmode.segs[seg] = *var;
4626 if (seg == VCPU_SREG_TR)
4627 vmcs_write16(sf->selector, var->selector);
4629 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
4633 vmcs_writel(sf->base, var->base);
4634 vmcs_write32(sf->limit, var->limit);
4635 vmcs_write16(sf->selector, var->selector);
4638 * Fix the "Accessed" bit in AR field of segment registers for older
4640 * IA32 arch specifies that at the time of processor reset the
4641 * "Accessed" bit in the AR field of segment registers is 1. And qemu
4642 * is setting it to 0 in the userland code. This causes invalid guest
4643 * state vmexit when "unrestricted guest" mode is turned on.
4644 * Fix for this setup issue in cpu_reset is being pushed in the qemu
4645 * tree. Newer qemu binaries with that qemu fix would not need this
4648 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
4649 var->type |= 0x1; /* Accessed */
4651 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
4654 vmx->emulation_required = emulation_required(vcpu);
4657 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4659 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
4661 *db = (ar >> 14) & 1;
4662 *l = (ar >> 13) & 1;
4665 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4667 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
4668 dt->address = vmcs_readl(GUEST_IDTR_BASE);
4671 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4673 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
4674 vmcs_writel(GUEST_IDTR_BASE, dt->address);
4677 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4679 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
4680 dt->address = vmcs_readl(GUEST_GDTR_BASE);
4683 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4685 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
4686 vmcs_writel(GUEST_GDTR_BASE, dt->address);
4689 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
4691 struct kvm_segment var;
4694 vmx_get_segment(vcpu, &var, seg);
4696 if (seg == VCPU_SREG_CS)
4698 ar = vmx_segment_access_rights(&var);
4700 if (var.base != (var.selector << 4))
4702 if (var.limit != 0xffff)
4710 static bool code_segment_valid(struct kvm_vcpu *vcpu)
4712 struct kvm_segment cs;
4713 unsigned int cs_rpl;
4715 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4716 cs_rpl = cs.selector & SEGMENT_RPL_MASK;
4720 if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
4724 if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
4725 if (cs.dpl > cs_rpl)
4728 if (cs.dpl != cs_rpl)
4734 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
4738 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
4740 struct kvm_segment ss;
4741 unsigned int ss_rpl;
4743 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4744 ss_rpl = ss.selector & SEGMENT_RPL_MASK;
4748 if (ss.type != 3 && ss.type != 7)
4752 if (ss.dpl != ss_rpl) /* DPL != RPL */
4760 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
4762 struct kvm_segment var;
4765 vmx_get_segment(vcpu, &var, seg);
4766 rpl = var.selector & SEGMENT_RPL_MASK;
4774 if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
4775 if (var.dpl < rpl) /* DPL < RPL */
4779 /* TODO: Add other members to kvm_segment_field to allow checking for other access
4785 static bool tr_valid(struct kvm_vcpu *vcpu)
4787 struct kvm_segment tr;
4789 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
4793 if (tr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4795 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
4803 static bool ldtr_valid(struct kvm_vcpu *vcpu)
4805 struct kvm_segment ldtr;
4807 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
4811 if (ldtr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4821 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
4823 struct kvm_segment cs, ss;
4825 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4826 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4828 return ((cs.selector & SEGMENT_RPL_MASK) ==
4829 (ss.selector & SEGMENT_RPL_MASK));
4833 * Check if guest state is valid. Returns true if valid, false if
4835 * We assume that registers are always usable
4837 static bool guest_state_valid(struct kvm_vcpu *vcpu)
4839 if (enable_unrestricted_guest)
4842 /* real mode guest state checks */
4843 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
4844 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
4846 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
4848 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
4850 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
4852 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
4854 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
4857 /* protected mode guest state checks */
4858 if (!cs_ss_rpl_check(vcpu))
4860 if (!code_segment_valid(vcpu))
4862 if (!stack_segment_valid(vcpu))
4864 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
4866 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
4868 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
4870 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
4872 if (!tr_valid(vcpu))
4874 if (!ldtr_valid(vcpu))
4878 * - Add checks on RIP
4879 * - Add checks on RFLAGS
4885 static bool page_address_valid(struct kvm_vcpu *vcpu, gpa_t gpa)
4887 return PAGE_ALIGNED(gpa) && !(gpa >> cpuid_maxphyaddr(vcpu));
4890 static int init_rmode_tss(struct kvm *kvm)
4896 idx = srcu_read_lock(&kvm->srcu);
4897 fn = kvm->arch.tss_addr >> PAGE_SHIFT;
4898 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4901 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
4902 r = kvm_write_guest_page(kvm, fn++, &data,
4903 TSS_IOPB_BASE_OFFSET, sizeof(u16));
4906 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
4909 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4913 r = kvm_write_guest_page(kvm, fn, &data,
4914 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
4917 srcu_read_unlock(&kvm->srcu, idx);
4921 static int init_rmode_identity_map(struct kvm *kvm)
4924 kvm_pfn_t identity_map_pfn;
4927 /* Protect kvm->arch.ept_identity_pagetable_done. */
4928 mutex_lock(&kvm->slots_lock);
4930 if (likely(kvm->arch.ept_identity_pagetable_done))
4933 if (!kvm->arch.ept_identity_map_addr)
4934 kvm->arch.ept_identity_map_addr = VMX_EPT_IDENTITY_PAGETABLE_ADDR;
4935 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
4937 r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
4938 kvm->arch.ept_identity_map_addr, PAGE_SIZE);
4942 idx = srcu_read_lock(&kvm->srcu);
4943 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
4946 /* Set up identity-mapping pagetable for EPT in real mode */
4947 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
4948 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
4949 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
4950 r = kvm_write_guest_page(kvm, identity_map_pfn,
4951 &tmp, i * sizeof(tmp), sizeof(tmp));
4955 kvm->arch.ept_identity_pagetable_done = true;
4958 srcu_read_unlock(&kvm->srcu, idx);
4961 mutex_unlock(&kvm->slots_lock);
4965 static void seg_setup(int seg)
4967 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4970 vmcs_write16(sf->selector, 0);
4971 vmcs_writel(sf->base, 0);
4972 vmcs_write32(sf->limit, 0xffff);
4974 if (seg == VCPU_SREG_CS)
4975 ar |= 0x08; /* code segment */
4977 vmcs_write32(sf->ar_bytes, ar);
4980 static int alloc_apic_access_page(struct kvm *kvm)
4985 mutex_lock(&kvm->slots_lock);
4986 if (kvm->arch.apic_access_page_done)
4988 r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
4989 APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
4993 page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
4994 if (is_error_page(page)) {
5000 * Do not pin the page in memory, so that memory hot-unplug
5001 * is able to migrate it.
5004 kvm->arch.apic_access_page_done = true;
5006 mutex_unlock(&kvm->slots_lock);
5010 static int allocate_vpid(void)
5016 spin_lock(&vmx_vpid_lock);
5017 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
5018 if (vpid < VMX_NR_VPIDS)
5019 __set_bit(vpid, vmx_vpid_bitmap);
5022 spin_unlock(&vmx_vpid_lock);
5026 static void free_vpid(int vpid)
5028 if (!enable_vpid || vpid == 0)
5030 spin_lock(&vmx_vpid_lock);
5031 __clear_bit(vpid, vmx_vpid_bitmap);
5032 spin_unlock(&vmx_vpid_lock);
5035 static void __always_inline vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
5038 int f = sizeof(unsigned long);
5040 if (!cpu_has_vmx_msr_bitmap())
5044 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
5045 * have the write-low and read-high bitmap offsets the wrong way round.
5046 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
5048 if (msr <= 0x1fff) {
5049 if (type & MSR_TYPE_R)
5051 __clear_bit(msr, msr_bitmap + 0x000 / f);
5053 if (type & MSR_TYPE_W)
5055 __clear_bit(msr, msr_bitmap + 0x800 / f);
5057 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
5059 if (type & MSR_TYPE_R)
5061 __clear_bit(msr, msr_bitmap + 0x400 / f);
5063 if (type & MSR_TYPE_W)
5065 __clear_bit(msr, msr_bitmap + 0xc00 / f);
5070 static void __always_inline vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
5073 int f = sizeof(unsigned long);
5075 if (!cpu_has_vmx_msr_bitmap())
5079 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
5080 * have the write-low and read-high bitmap offsets the wrong way round.
5081 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
5083 if (msr <= 0x1fff) {
5084 if (type & MSR_TYPE_R)
5086 __set_bit(msr, msr_bitmap + 0x000 / f);
5088 if (type & MSR_TYPE_W)
5090 __set_bit(msr, msr_bitmap + 0x800 / f);
5092 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
5094 if (type & MSR_TYPE_R)
5096 __set_bit(msr, msr_bitmap + 0x400 / f);
5098 if (type & MSR_TYPE_W)
5100 __set_bit(msr, msr_bitmap + 0xc00 / f);
5105 static void __always_inline vmx_set_intercept_for_msr(unsigned long *msr_bitmap,
5106 u32 msr, int type, bool value)
5109 vmx_enable_intercept_for_msr(msr_bitmap, msr, type);
5111 vmx_disable_intercept_for_msr(msr_bitmap, msr, type);
5115 * If a msr is allowed by L0, we should check whether it is allowed by L1.
5116 * The corresponding bit will be cleared unless both of L0 and L1 allow it.
5118 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
5119 unsigned long *msr_bitmap_nested,
5122 int f = sizeof(unsigned long);
5125 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
5126 * have the write-low and read-high bitmap offsets the wrong way round.
5127 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
5129 if (msr <= 0x1fff) {
5130 if (type & MSR_TYPE_R &&
5131 !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
5133 __clear_bit(msr, msr_bitmap_nested + 0x000 / f);
5135 if (type & MSR_TYPE_W &&
5136 !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
5138 __clear_bit(msr, msr_bitmap_nested + 0x800 / f);
5140 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
5142 if (type & MSR_TYPE_R &&
5143 !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
5145 __clear_bit(msr, msr_bitmap_nested + 0x400 / f);
5147 if (type & MSR_TYPE_W &&
5148 !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
5150 __clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
5155 static u8 vmx_msr_bitmap_mode(struct kvm_vcpu *vcpu)
5159 if (cpu_has_secondary_exec_ctrls() &&
5160 (vmcs_read32(SECONDARY_VM_EXEC_CONTROL) &
5161 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
5162 mode |= MSR_BITMAP_MODE_X2APIC;
5163 if (enable_apicv && kvm_vcpu_apicv_active(vcpu))
5164 mode |= MSR_BITMAP_MODE_X2APIC_APICV;
5167 if (is_long_mode(vcpu))
5168 mode |= MSR_BITMAP_MODE_LM;
5173 #define X2APIC_MSR(r) (APIC_BASE_MSR + ((r) >> 4))
5175 static void vmx_update_msr_bitmap_x2apic(unsigned long *msr_bitmap,
5180 for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
5181 unsigned word = msr / BITS_PER_LONG;
5182 msr_bitmap[word] = (mode & MSR_BITMAP_MODE_X2APIC_APICV) ? 0 : ~0;
5183 msr_bitmap[word + (0x800 / sizeof(long))] = ~0;
5186 if (mode & MSR_BITMAP_MODE_X2APIC) {
5188 * TPR reads and writes can be virtualized even if virtual interrupt
5189 * delivery is not in use.
5191 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TASKPRI), MSR_TYPE_RW);
5192 if (mode & MSR_BITMAP_MODE_X2APIC_APICV) {
5193 vmx_enable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_TMCCT), MSR_TYPE_R);
5194 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_EOI), MSR_TYPE_W);
5195 vmx_disable_intercept_for_msr(msr_bitmap, X2APIC_MSR(APIC_SELF_IPI), MSR_TYPE_W);
5200 static void vmx_update_msr_bitmap(struct kvm_vcpu *vcpu)
5202 struct vcpu_vmx *vmx = to_vmx(vcpu);
5203 unsigned long *msr_bitmap = vmx->vmcs01.msr_bitmap;
5204 u8 mode = vmx_msr_bitmap_mode(vcpu);
5205 u8 changed = mode ^ vmx->msr_bitmap_mode;
5210 vmx_set_intercept_for_msr(msr_bitmap, MSR_KERNEL_GS_BASE, MSR_TYPE_RW,
5211 !(mode & MSR_BITMAP_MODE_LM));
5213 if (changed & (MSR_BITMAP_MODE_X2APIC | MSR_BITMAP_MODE_X2APIC_APICV))
5214 vmx_update_msr_bitmap_x2apic(msr_bitmap, mode);
5216 vmx->msr_bitmap_mode = mode;
5219 static bool vmx_get_enable_apicv(struct kvm_vcpu *vcpu)
5221 return enable_apicv;
5224 static void nested_mark_vmcs12_pages_dirty(struct kvm_vcpu *vcpu)
5226 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5230 * Don't need to mark the APIC access page dirty; it is never
5231 * written to by the CPU during APIC virtualization.
5234 if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
5235 gfn = vmcs12->virtual_apic_page_addr >> PAGE_SHIFT;
5236 kvm_vcpu_mark_page_dirty(vcpu, gfn);
5239 if (nested_cpu_has_posted_intr(vmcs12)) {
5240 gfn = vmcs12->posted_intr_desc_addr >> PAGE_SHIFT;
5241 kvm_vcpu_mark_page_dirty(vcpu, gfn);
5246 static void vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
5248 struct vcpu_vmx *vmx = to_vmx(vcpu);
5253 if (!vmx->nested.pi_desc || !vmx->nested.pi_pending)
5256 vmx->nested.pi_pending = false;
5257 if (!pi_test_and_clear_on(vmx->nested.pi_desc))
5260 max_irr = find_last_bit((unsigned long *)vmx->nested.pi_desc->pir, 256);
5261 if (max_irr != 256) {
5262 vapic_page = kmap(vmx->nested.virtual_apic_page);
5263 __kvm_apic_update_irr(vmx->nested.pi_desc->pir,
5264 vapic_page, &max_irr);
5265 kunmap(vmx->nested.virtual_apic_page);
5267 status = vmcs_read16(GUEST_INTR_STATUS);
5268 if ((u8)max_irr > ((u8)status & 0xff)) {
5270 status |= (u8)max_irr;
5271 vmcs_write16(GUEST_INTR_STATUS, status);
5275 nested_mark_vmcs12_pages_dirty(vcpu);
5278 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
5282 int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
5284 if (vcpu->mode == IN_GUEST_MODE) {
5286 * The vector of interrupt to be delivered to vcpu had
5287 * been set in PIR before this function.
5289 * Following cases will be reached in this block, and
5290 * we always send a notification event in all cases as
5293 * Case 1: vcpu keeps in non-root mode. Sending a
5294 * notification event posts the interrupt to vcpu.
5296 * Case 2: vcpu exits to root mode and is still
5297 * runnable. PIR will be synced to vIRR before the
5298 * next vcpu entry. Sending a notification event in
5299 * this case has no effect, as vcpu is not in root
5302 * Case 3: vcpu exits to root mode and is blocked.
5303 * vcpu_block() has already synced PIR to vIRR and
5304 * never blocks vcpu if vIRR is not cleared. Therefore,
5305 * a blocked vcpu here does not wait for any requested
5306 * interrupts in PIR, and sending a notification event
5307 * which has no effect is safe here.
5310 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
5317 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
5320 struct vcpu_vmx *vmx = to_vmx(vcpu);
5322 if (is_guest_mode(vcpu) &&
5323 vector == vmx->nested.posted_intr_nv) {
5325 * If a posted intr is not recognized by hardware,
5326 * we will accomplish it in the next vmentry.
5328 vmx->nested.pi_pending = true;
5329 kvm_make_request(KVM_REQ_EVENT, vcpu);
5330 /* the PIR and ON have been set by L1. */
5331 if (!kvm_vcpu_trigger_posted_interrupt(vcpu, true))
5332 kvm_vcpu_kick(vcpu);
5338 * Send interrupt to vcpu via posted interrupt way.
5339 * 1. If target vcpu is running(non-root mode), send posted interrupt
5340 * notification to vcpu and hardware will sync PIR to vIRR atomically.
5341 * 2. If target vcpu isn't running(root mode), kick it to pick up the
5342 * interrupt from PIR in next vmentry.
5344 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
5346 struct vcpu_vmx *vmx = to_vmx(vcpu);
5349 r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
5353 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
5356 /* If a previous notification has sent the IPI, nothing to do. */
5357 if (pi_test_and_set_on(&vmx->pi_desc))
5360 if (!kvm_vcpu_trigger_posted_interrupt(vcpu, false))
5361 kvm_vcpu_kick(vcpu);
5365 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
5366 * will not change in the lifetime of the guest.
5367 * Note that host-state that does change is set elsewhere. E.g., host-state
5368 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
5370 static void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
5375 unsigned long cr0, cr3, cr4;
5378 WARN_ON(cr0 & X86_CR0_TS);
5379 vmcs_writel(HOST_CR0, cr0); /* 22.2.3 */
5382 * Save the most likely value for this task's CR3 in the VMCS.
5383 * We can't use __get_current_cr3_fast() because we're not atomic.
5386 vmcs_writel(HOST_CR3, cr3); /* 22.2.3 FIXME: shadow tables */
5387 vmx->loaded_vmcs->vmcs_host_cr3 = cr3;
5389 /* Save the most likely value for this task's CR4 in the VMCS. */
5390 cr4 = cr4_read_shadow();
5391 vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
5392 vmx->loaded_vmcs->vmcs_host_cr4 = cr4;
5394 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
5395 #ifdef CONFIG_X86_64
5397 * Load null selectors, so we can avoid reloading them in
5398 * __vmx_load_host_state(), in case userspace uses the null selectors
5399 * too (the expected case).
5401 vmcs_write16(HOST_DS_SELECTOR, 0);
5402 vmcs_write16(HOST_ES_SELECTOR, 0);
5404 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
5405 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
5407 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
5408 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
5411 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
5412 vmx->host_idt_base = dt.address;
5414 vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
5416 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
5417 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
5418 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
5419 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
5421 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
5422 rdmsr(MSR_IA32_CR_PAT, low32, high32);
5423 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
5427 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
5429 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
5431 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
5432 if (is_guest_mode(&vmx->vcpu))
5433 vmx->vcpu.arch.cr4_guest_owned_bits &=
5434 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
5435 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
5438 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
5440 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
5442 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
5443 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
5446 pin_based_exec_ctrl &= ~PIN_BASED_VIRTUAL_NMIS;
5448 /* Enable the preemption timer dynamically */
5449 pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
5450 return pin_based_exec_ctrl;
5453 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
5455 struct vcpu_vmx *vmx = to_vmx(vcpu);
5457 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
5458 if (cpu_has_secondary_exec_ctrls()) {
5459 if (kvm_vcpu_apicv_active(vcpu))
5460 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
5461 SECONDARY_EXEC_APIC_REGISTER_VIRT |
5462 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5464 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
5465 SECONDARY_EXEC_APIC_REGISTER_VIRT |
5466 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5469 if (cpu_has_vmx_msr_bitmap())
5470 vmx_update_msr_bitmap(vcpu);
5473 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
5475 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
5477 if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
5478 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
5480 if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
5481 exec_control &= ~CPU_BASED_TPR_SHADOW;
5482 #ifdef CONFIG_X86_64
5483 exec_control |= CPU_BASED_CR8_STORE_EXITING |
5484 CPU_BASED_CR8_LOAD_EXITING;
5488 exec_control |= CPU_BASED_CR3_STORE_EXITING |
5489 CPU_BASED_CR3_LOAD_EXITING |
5490 CPU_BASED_INVLPG_EXITING;
5491 return exec_control;
5494 static bool vmx_rdrand_supported(void)
5496 return vmcs_config.cpu_based_2nd_exec_ctrl &
5497 SECONDARY_EXEC_RDRAND_EXITING;
5500 static bool vmx_rdseed_supported(void)
5502 return vmcs_config.cpu_based_2nd_exec_ctrl &
5503 SECONDARY_EXEC_RDSEED_EXITING;
5506 static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
5508 struct kvm_vcpu *vcpu = &vmx->vcpu;
5510 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
5512 if (!cpu_need_virtualize_apic_accesses(vcpu))
5513 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
5515 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
5517 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
5518 enable_unrestricted_guest = 0;
5519 /* Enable INVPCID for non-ept guests may cause performance regression. */
5520 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
5522 if (!enable_unrestricted_guest)
5523 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
5525 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
5526 if (!kvm_vcpu_apicv_active(vcpu))
5527 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
5528 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5529 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
5531 /* SECONDARY_EXEC_DESC is enabled/disabled on writes to CR4.UMIP,
5532 * in vmx_set_cr4. */
5533 exec_control &= ~SECONDARY_EXEC_DESC;
5535 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
5537 We can NOT enable shadow_vmcs here because we don't have yet
5540 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
5543 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
5545 if (vmx_xsaves_supported()) {
5546 /* Exposing XSAVES only when XSAVE is exposed */
5547 bool xsaves_enabled =
5548 guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
5549 guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
5551 if (!xsaves_enabled)
5552 exec_control &= ~SECONDARY_EXEC_XSAVES;
5556 vmx->nested.nested_vmx_secondary_ctls_high |=
5557 SECONDARY_EXEC_XSAVES;
5559 vmx->nested.nested_vmx_secondary_ctls_high &=
5560 ~SECONDARY_EXEC_XSAVES;
5564 if (vmx_rdtscp_supported()) {
5565 bool rdtscp_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP);
5566 if (!rdtscp_enabled)
5567 exec_control &= ~SECONDARY_EXEC_RDTSCP;
5571 vmx->nested.nested_vmx_secondary_ctls_high |=
5572 SECONDARY_EXEC_RDTSCP;
5574 vmx->nested.nested_vmx_secondary_ctls_high &=
5575 ~SECONDARY_EXEC_RDTSCP;
5579 if (vmx_invpcid_supported()) {
5580 /* Exposing INVPCID only when PCID is exposed */
5581 bool invpcid_enabled =
5582 guest_cpuid_has(vcpu, X86_FEATURE_INVPCID) &&
5583 guest_cpuid_has(vcpu, X86_FEATURE_PCID);
5585 if (!invpcid_enabled) {
5586 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
5587 guest_cpuid_clear(vcpu, X86_FEATURE_INVPCID);
5591 if (invpcid_enabled)
5592 vmx->nested.nested_vmx_secondary_ctls_high |=
5593 SECONDARY_EXEC_ENABLE_INVPCID;
5595 vmx->nested.nested_vmx_secondary_ctls_high &=
5596 ~SECONDARY_EXEC_ENABLE_INVPCID;
5600 if (vmx_rdrand_supported()) {
5601 bool rdrand_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDRAND);
5603 exec_control &= ~SECONDARY_EXEC_RDRAND_EXITING;
5607 vmx->nested.nested_vmx_secondary_ctls_high |=
5608 SECONDARY_EXEC_RDRAND_EXITING;
5610 vmx->nested.nested_vmx_secondary_ctls_high &=
5611 ~SECONDARY_EXEC_RDRAND_EXITING;
5615 if (vmx_rdseed_supported()) {
5616 bool rdseed_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDSEED);
5618 exec_control &= ~SECONDARY_EXEC_RDSEED_EXITING;
5622 vmx->nested.nested_vmx_secondary_ctls_high |=
5623 SECONDARY_EXEC_RDSEED_EXITING;
5625 vmx->nested.nested_vmx_secondary_ctls_high &=
5626 ~SECONDARY_EXEC_RDSEED_EXITING;
5630 vmx->secondary_exec_control = exec_control;
5633 static void ept_set_mmio_spte_mask(void)
5636 * EPT Misconfigurations can be generated if the value of bits 2:0
5637 * of an EPT paging-structure entry is 110b (write/execute).
5639 kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK,
5640 VMX_EPT_MISCONFIG_WX_VALUE);
5643 #define VMX_XSS_EXIT_BITMAP 0
5645 * Sets up the vmcs for emulated real mode.
5647 static void vmx_vcpu_setup(struct vcpu_vmx *vmx)
5649 #ifdef CONFIG_X86_64
5654 if (enable_shadow_vmcs) {
5655 vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
5656 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
5658 if (cpu_has_vmx_msr_bitmap())
5659 vmcs_write64(MSR_BITMAP, __pa(vmx->vmcs01.msr_bitmap));
5661 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
5664 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
5665 vmx->hv_deadline_tsc = -1;
5667 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
5669 if (cpu_has_secondary_exec_ctrls()) {
5670 vmx_compute_secondary_exec_control(vmx);
5671 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
5672 vmx->secondary_exec_control);
5675 if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
5676 vmcs_write64(EOI_EXIT_BITMAP0, 0);
5677 vmcs_write64(EOI_EXIT_BITMAP1, 0);
5678 vmcs_write64(EOI_EXIT_BITMAP2, 0);
5679 vmcs_write64(EOI_EXIT_BITMAP3, 0);
5681 vmcs_write16(GUEST_INTR_STATUS, 0);
5683 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
5684 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
5688 vmcs_write32(PLE_GAP, ple_gap);
5689 vmx->ple_window = ple_window;
5690 vmx->ple_window_dirty = true;
5693 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
5694 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
5695 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
5697 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
5698 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
5699 vmx_set_constant_host_state(vmx);
5700 #ifdef CONFIG_X86_64
5701 rdmsrl(MSR_FS_BASE, a);
5702 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
5703 rdmsrl(MSR_GS_BASE, a);
5704 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
5706 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
5707 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
5710 if (cpu_has_vmx_vmfunc())
5711 vmcs_write64(VM_FUNCTION_CONTROL, 0);
5713 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
5714 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
5715 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
5716 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
5717 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
5719 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
5720 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
5722 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
5723 u32 index = vmx_msr_index[i];
5724 u32 data_low, data_high;
5727 if (rdmsr_safe(index, &data_low, &data_high) < 0)
5729 if (wrmsr_safe(index, data_low, data_high) < 0)
5731 vmx->guest_msrs[j].index = i;
5732 vmx->guest_msrs[j].data = 0;
5733 vmx->guest_msrs[j].mask = -1ull;
5737 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
5738 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, vmx->arch_capabilities);
5740 vm_exit_controls_init(vmx, vmcs_config.vmexit_ctrl);
5742 /* 22.2.1, 20.8.1 */
5743 vm_entry_controls_init(vmx, vmcs_config.vmentry_ctrl);
5745 vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
5746 vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
5748 set_cr4_guest_host_mask(vmx);
5750 if (vmx_xsaves_supported())
5751 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
5754 ASSERT(vmx->pml_pg);
5755 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
5756 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5760 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
5762 struct vcpu_vmx *vmx = to_vmx(vcpu);
5763 struct msr_data apic_base_msr;
5766 vmx->rmode.vm86_active = 0;
5769 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
5770 kvm_set_cr8(vcpu, 0);
5773 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
5774 MSR_IA32_APICBASE_ENABLE;
5775 if (kvm_vcpu_is_reset_bsp(vcpu))
5776 apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
5777 apic_base_msr.host_initiated = true;
5778 kvm_set_apic_base(vcpu, &apic_base_msr);
5781 vmx_segment_cache_clear(vmx);
5783 seg_setup(VCPU_SREG_CS);
5784 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
5785 vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
5787 seg_setup(VCPU_SREG_DS);
5788 seg_setup(VCPU_SREG_ES);
5789 seg_setup(VCPU_SREG_FS);
5790 seg_setup(VCPU_SREG_GS);
5791 seg_setup(VCPU_SREG_SS);
5793 vmcs_write16(GUEST_TR_SELECTOR, 0);
5794 vmcs_writel(GUEST_TR_BASE, 0);
5795 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
5796 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
5798 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
5799 vmcs_writel(GUEST_LDTR_BASE, 0);
5800 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
5801 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
5804 vmcs_write32(GUEST_SYSENTER_CS, 0);
5805 vmcs_writel(GUEST_SYSENTER_ESP, 0);
5806 vmcs_writel(GUEST_SYSENTER_EIP, 0);
5807 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
5810 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
5811 kvm_rip_write(vcpu, 0xfff0);
5813 vmcs_writel(GUEST_GDTR_BASE, 0);
5814 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
5816 vmcs_writel(GUEST_IDTR_BASE, 0);
5817 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
5819 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
5820 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
5821 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
5822 if (kvm_mpx_supported())
5823 vmcs_write64(GUEST_BNDCFGS, 0);
5827 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
5829 if (cpu_has_vmx_tpr_shadow() && !init_event) {
5830 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
5831 if (cpu_need_tpr_shadow(vcpu))
5832 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
5833 __pa(vcpu->arch.apic->regs));
5834 vmcs_write32(TPR_THRESHOLD, 0);
5837 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
5840 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
5842 cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
5843 vmx->vcpu.arch.cr0 = cr0;
5844 vmx_set_cr0(vcpu, cr0); /* enter rmode */
5845 vmx_set_cr4(vcpu, 0);
5846 vmx_set_efer(vcpu, 0);
5848 update_exception_bitmap(vcpu);
5850 vpid_sync_context(vmx->vpid);
5854 * In nested virtualization, check if L1 asked to exit on external interrupts.
5855 * For most existing hypervisors, this will always return true.
5857 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
5859 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5860 PIN_BASED_EXT_INTR_MASK;
5864 * In nested virtualization, check if L1 has set
5865 * VM_EXIT_ACK_INTR_ON_EXIT
5867 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
5869 return get_vmcs12(vcpu)->vm_exit_controls &
5870 VM_EXIT_ACK_INTR_ON_EXIT;
5873 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
5875 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5876 PIN_BASED_NMI_EXITING;
5879 static void enable_irq_window(struct kvm_vcpu *vcpu)
5881 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
5882 CPU_BASED_VIRTUAL_INTR_PENDING);
5885 static void enable_nmi_window(struct kvm_vcpu *vcpu)
5888 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
5889 enable_irq_window(vcpu);
5893 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
5894 CPU_BASED_VIRTUAL_NMI_PENDING);
5897 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
5899 struct vcpu_vmx *vmx = to_vmx(vcpu);
5901 int irq = vcpu->arch.interrupt.nr;
5903 trace_kvm_inj_virq(irq);
5905 ++vcpu->stat.irq_injections;
5906 if (vmx->rmode.vm86_active) {
5908 if (vcpu->arch.interrupt.soft)
5909 inc_eip = vcpu->arch.event_exit_inst_len;
5910 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
5911 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5914 intr = irq | INTR_INFO_VALID_MASK;
5915 if (vcpu->arch.interrupt.soft) {
5916 intr |= INTR_TYPE_SOFT_INTR;
5917 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
5918 vmx->vcpu.arch.event_exit_inst_len);
5920 intr |= INTR_TYPE_EXT_INTR;
5921 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
5924 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
5926 struct vcpu_vmx *vmx = to_vmx(vcpu);
5930 * Tracking the NMI-blocked state in software is built upon
5931 * finding the next open IRQ window. This, in turn, depends on
5932 * well-behaving guests: They have to keep IRQs disabled at
5933 * least as long as the NMI handler runs. Otherwise we may
5934 * cause NMI nesting, maybe breaking the guest. But as this is
5935 * highly unlikely, we can live with the residual risk.
5937 vmx->loaded_vmcs->soft_vnmi_blocked = 1;
5938 vmx->loaded_vmcs->vnmi_blocked_time = 0;
5941 ++vcpu->stat.nmi_injections;
5942 vmx->loaded_vmcs->nmi_known_unmasked = false;
5944 if (vmx->rmode.vm86_active) {
5945 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
5946 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5950 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
5951 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
5954 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
5956 struct vcpu_vmx *vmx = to_vmx(vcpu);
5960 return vmx->loaded_vmcs->soft_vnmi_blocked;
5961 if (vmx->loaded_vmcs->nmi_known_unmasked)
5963 masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
5964 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
5968 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
5970 struct vcpu_vmx *vmx = to_vmx(vcpu);
5973 if (vmx->loaded_vmcs->soft_vnmi_blocked != masked) {
5974 vmx->loaded_vmcs->soft_vnmi_blocked = masked;
5975 vmx->loaded_vmcs->vnmi_blocked_time = 0;
5978 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
5980 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5981 GUEST_INTR_STATE_NMI);
5983 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
5984 GUEST_INTR_STATE_NMI);
5988 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
5990 if (to_vmx(vcpu)->nested.nested_run_pending)
5994 to_vmx(vcpu)->loaded_vmcs->soft_vnmi_blocked)
5997 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5998 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
5999 | GUEST_INTR_STATE_NMI));
6002 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
6004 return (!to_vmx(vcpu)->nested.nested_run_pending &&
6005 vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
6006 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
6007 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
6010 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
6014 ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
6018 kvm->arch.tss_addr = addr;
6019 return init_rmode_tss(kvm);
6022 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
6027 * Update instruction length as we may reinject the exception
6028 * from user space while in guest debugging mode.
6030 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
6031 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
6032 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
6036 if (vcpu->guest_debug &
6037 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
6054 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
6055 int vec, u32 err_code)
6058 * Instruction with address size override prefix opcode 0x67
6059 * Cause the #SS fault with 0 error code in VM86 mode.
6061 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
6062 if (emulate_instruction(vcpu, 0) == EMULATE_DONE) {
6063 if (vcpu->arch.halt_request) {
6064 vcpu->arch.halt_request = 0;
6065 return kvm_vcpu_halt(vcpu);
6073 * Forward all other exceptions that are valid in real mode.
6074 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
6075 * the required debugging infrastructure rework.
6077 kvm_queue_exception(vcpu, vec);
6082 * Trigger machine check on the host. We assume all the MSRs are already set up
6083 * by the CPU and that we still run on the same CPU as the MCE occurred on.
6084 * We pass a fake environment to the machine check handler because we want
6085 * the guest to be always treated like user space, no matter what context
6086 * it used internally.
6088 static void kvm_machine_check(void)
6090 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
6091 struct pt_regs regs = {
6092 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
6093 .flags = X86_EFLAGS_IF,
6096 do_machine_check(®s, 0);
6100 static int handle_machine_check(struct kvm_vcpu *vcpu)
6102 /* already handled by vcpu_run */
6106 static int handle_exception(struct kvm_vcpu *vcpu)
6108 struct vcpu_vmx *vmx = to_vmx(vcpu);
6109 struct kvm_run *kvm_run = vcpu->run;
6110 u32 intr_info, ex_no, error_code;
6111 unsigned long cr2, rip, dr6;
6113 enum emulation_result er;
6115 vect_info = vmx->idt_vectoring_info;
6116 intr_info = vmx->exit_intr_info;
6118 if (is_machine_check(intr_info))
6119 return handle_machine_check(vcpu);
6121 if (is_nmi(intr_info))
6122 return 1; /* already handled by vmx_vcpu_run() */
6124 if (is_invalid_opcode(intr_info)) {
6125 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
6126 if (er == EMULATE_USER_EXIT)
6128 if (er != EMULATE_DONE)
6129 kvm_queue_exception(vcpu, UD_VECTOR);
6134 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
6135 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
6138 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
6139 * MMIO, it is better to report an internal error.
6140 * See the comments in vmx_handle_exit.
6142 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
6143 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
6144 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6145 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
6146 vcpu->run->internal.ndata = 3;
6147 vcpu->run->internal.data[0] = vect_info;
6148 vcpu->run->internal.data[1] = intr_info;
6149 vcpu->run->internal.data[2] = error_code;
6153 if (is_page_fault(intr_info)) {
6154 cr2 = vmcs_readl(EXIT_QUALIFICATION);
6155 /* EPT won't cause page fault directly */
6156 WARN_ON_ONCE(!vcpu->arch.apf.host_apf_reason && enable_ept);
6157 return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0);
6160 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
6162 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
6163 return handle_rmode_exception(vcpu, ex_no, error_code);
6167 kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
6170 dr6 = vmcs_readl(EXIT_QUALIFICATION);
6171 if (!(vcpu->guest_debug &
6172 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
6173 vcpu->arch.dr6 &= ~15;
6174 vcpu->arch.dr6 |= dr6 | DR6_RTM;
6175 if (!(dr6 & ~DR6_RESERVED)) /* icebp */
6176 skip_emulated_instruction(vcpu);
6178 kvm_queue_exception(vcpu, DB_VECTOR);
6181 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
6182 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
6186 * Update instruction length as we may reinject #BP from
6187 * user space while in guest debugging mode. Reading it for
6188 * #DB as well causes no harm, it is not used in that case.
6190 vmx->vcpu.arch.event_exit_inst_len =
6191 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
6192 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6193 rip = kvm_rip_read(vcpu);
6194 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
6195 kvm_run->debug.arch.exception = ex_no;
6198 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
6199 kvm_run->ex.exception = ex_no;
6200 kvm_run->ex.error_code = error_code;
6206 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
6208 ++vcpu->stat.irq_exits;
6212 static int handle_triple_fault(struct kvm_vcpu *vcpu)
6214 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
6215 vcpu->mmio_needed = 0;
6219 static int handle_io(struct kvm_vcpu *vcpu)
6221 unsigned long exit_qualification;
6222 int size, in, string, ret;
6225 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6226 string = (exit_qualification & 16) != 0;
6227 in = (exit_qualification & 8) != 0;
6229 ++vcpu->stat.io_exits;
6232 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6234 port = exit_qualification >> 16;
6235 size = (exit_qualification & 7) + 1;
6237 ret = kvm_skip_emulated_instruction(vcpu);
6240 * TODO: we might be squashing a KVM_GUESTDBG_SINGLESTEP-triggered
6241 * KVM_EXIT_DEBUG here.
6243 return kvm_fast_pio_out(vcpu, size, port) && ret;
6247 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
6250 * Patch in the VMCALL instruction:
6252 hypercall[0] = 0x0f;
6253 hypercall[1] = 0x01;
6254 hypercall[2] = 0xc1;
6257 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
6258 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
6260 if (is_guest_mode(vcpu)) {
6261 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6262 unsigned long orig_val = val;
6265 * We get here when L2 changed cr0 in a way that did not change
6266 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
6267 * but did change L0 shadowed bits. So we first calculate the
6268 * effective cr0 value that L1 would like to write into the
6269 * hardware. It consists of the L2-owned bits from the new
6270 * value combined with the L1-owned bits from L1's guest_cr0.
6272 val = (val & ~vmcs12->cr0_guest_host_mask) |
6273 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
6275 if (!nested_guest_cr0_valid(vcpu, val))
6278 if (kvm_set_cr0(vcpu, val))
6280 vmcs_writel(CR0_READ_SHADOW, orig_val);
6283 if (to_vmx(vcpu)->nested.vmxon &&
6284 !nested_host_cr0_valid(vcpu, val))
6287 return kvm_set_cr0(vcpu, val);
6291 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
6293 if (is_guest_mode(vcpu)) {
6294 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6295 unsigned long orig_val = val;
6297 /* analogously to handle_set_cr0 */
6298 val = (val & ~vmcs12->cr4_guest_host_mask) |
6299 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
6300 if (kvm_set_cr4(vcpu, val))
6302 vmcs_writel(CR4_READ_SHADOW, orig_val);
6305 return kvm_set_cr4(vcpu, val);
6308 static int handle_desc(struct kvm_vcpu *vcpu)
6310 WARN_ON(!(vcpu->arch.cr4 & X86_CR4_UMIP));
6311 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6314 static int handle_cr(struct kvm_vcpu *vcpu)
6316 unsigned long exit_qualification, val;
6322 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6323 cr = exit_qualification & 15;
6324 reg = (exit_qualification >> 8) & 15;
6325 switch ((exit_qualification >> 4) & 3) {
6326 case 0: /* mov to cr */
6327 val = kvm_register_readl(vcpu, reg);
6328 trace_kvm_cr_write(cr, val);
6331 err = handle_set_cr0(vcpu, val);
6332 return kvm_complete_insn_gp(vcpu, err);
6334 err = kvm_set_cr3(vcpu, val);
6335 return kvm_complete_insn_gp(vcpu, err);
6337 err = handle_set_cr4(vcpu, val);
6338 return kvm_complete_insn_gp(vcpu, err);
6340 u8 cr8_prev = kvm_get_cr8(vcpu);
6342 err = kvm_set_cr8(vcpu, cr8);
6343 ret = kvm_complete_insn_gp(vcpu, err);
6344 if (lapic_in_kernel(vcpu))
6346 if (cr8_prev <= cr8)
6349 * TODO: we might be squashing a
6350 * KVM_GUESTDBG_SINGLESTEP-triggered
6351 * KVM_EXIT_DEBUG here.
6353 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
6359 WARN_ONCE(1, "Guest should always own CR0.TS");
6360 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
6361 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
6362 return kvm_skip_emulated_instruction(vcpu);
6363 case 1: /*mov from cr*/
6366 val = kvm_read_cr3(vcpu);
6367 kvm_register_write(vcpu, reg, val);
6368 trace_kvm_cr_read(cr, val);
6369 return kvm_skip_emulated_instruction(vcpu);
6371 val = kvm_get_cr8(vcpu);
6372 kvm_register_write(vcpu, reg, val);
6373 trace_kvm_cr_read(cr, val);
6374 return kvm_skip_emulated_instruction(vcpu);
6378 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
6379 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
6380 kvm_lmsw(vcpu, val);
6382 return kvm_skip_emulated_instruction(vcpu);
6386 vcpu->run->exit_reason = 0;
6387 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
6388 (int)(exit_qualification >> 4) & 3, cr);
6392 static int handle_dr(struct kvm_vcpu *vcpu)
6394 unsigned long exit_qualification;
6397 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6398 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
6400 /* First, if DR does not exist, trigger UD */
6401 if (!kvm_require_dr(vcpu, dr))
6404 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
6405 if (!kvm_require_cpl(vcpu, 0))
6407 dr7 = vmcs_readl(GUEST_DR7);
6410 * As the vm-exit takes precedence over the debug trap, we
6411 * need to emulate the latter, either for the host or the
6412 * guest debugging itself.
6414 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6415 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
6416 vcpu->run->debug.arch.dr7 = dr7;
6417 vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
6418 vcpu->run->debug.arch.exception = DB_VECTOR;
6419 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
6422 vcpu->arch.dr6 &= ~15;
6423 vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
6424 kvm_queue_exception(vcpu, DB_VECTOR);
6429 if (vcpu->guest_debug == 0) {
6430 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6431 CPU_BASED_MOV_DR_EXITING);
6434 * No more DR vmexits; force a reload of the debug registers
6435 * and reenter on this instruction. The next vmexit will
6436 * retrieve the full state of the debug registers.
6438 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
6442 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
6443 if (exit_qualification & TYPE_MOV_FROM_DR) {
6446 if (kvm_get_dr(vcpu, dr, &val))
6448 kvm_register_write(vcpu, reg, val);
6450 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
6453 return kvm_skip_emulated_instruction(vcpu);
6456 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
6458 return vcpu->arch.dr6;
6461 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
6465 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
6467 get_debugreg(vcpu->arch.db[0], 0);
6468 get_debugreg(vcpu->arch.db[1], 1);
6469 get_debugreg(vcpu->arch.db[2], 2);
6470 get_debugreg(vcpu->arch.db[3], 3);
6471 get_debugreg(vcpu->arch.dr6, 6);
6472 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
6474 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
6475 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL, CPU_BASED_MOV_DR_EXITING);
6478 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
6480 vmcs_writel(GUEST_DR7, val);
6483 static int handle_cpuid(struct kvm_vcpu *vcpu)
6485 return kvm_emulate_cpuid(vcpu);
6488 static int handle_rdmsr(struct kvm_vcpu *vcpu)
6490 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
6491 struct msr_data msr_info;
6493 msr_info.index = ecx;
6494 msr_info.host_initiated = false;
6495 if (vmx_get_msr(vcpu, &msr_info)) {
6496 trace_kvm_msr_read_ex(ecx);
6497 kvm_inject_gp(vcpu, 0);
6501 trace_kvm_msr_read(ecx, msr_info.data);
6503 /* FIXME: handling of bits 32:63 of rax, rdx */
6504 vcpu->arch.regs[VCPU_REGS_RAX] = msr_info.data & -1u;
6505 vcpu->arch.regs[VCPU_REGS_RDX] = (msr_info.data >> 32) & -1u;
6506 return kvm_skip_emulated_instruction(vcpu);
6509 static int handle_wrmsr(struct kvm_vcpu *vcpu)
6511 struct msr_data msr;
6512 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
6513 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
6514 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
6518 msr.host_initiated = false;
6519 if (kvm_set_msr(vcpu, &msr) != 0) {
6520 trace_kvm_msr_write_ex(ecx, data);
6521 kvm_inject_gp(vcpu, 0);
6525 trace_kvm_msr_write(ecx, data);
6526 return kvm_skip_emulated_instruction(vcpu);
6529 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
6531 kvm_apic_update_ppr(vcpu);
6535 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
6537 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6538 CPU_BASED_VIRTUAL_INTR_PENDING);
6540 kvm_make_request(KVM_REQ_EVENT, vcpu);
6542 ++vcpu->stat.irq_window_exits;
6546 static int handle_halt(struct kvm_vcpu *vcpu)
6548 return kvm_emulate_halt(vcpu);
6551 static int handle_vmcall(struct kvm_vcpu *vcpu)
6553 return kvm_emulate_hypercall(vcpu);
6556 static int handle_invd(struct kvm_vcpu *vcpu)
6558 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6561 static int handle_invlpg(struct kvm_vcpu *vcpu)
6563 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6565 kvm_mmu_invlpg(vcpu, exit_qualification);
6566 return kvm_skip_emulated_instruction(vcpu);
6569 static int handle_rdpmc(struct kvm_vcpu *vcpu)
6573 err = kvm_rdpmc(vcpu);
6574 return kvm_complete_insn_gp(vcpu, err);
6577 static int handle_wbinvd(struct kvm_vcpu *vcpu)
6579 return kvm_emulate_wbinvd(vcpu);
6582 static int handle_xsetbv(struct kvm_vcpu *vcpu)
6584 u64 new_bv = kvm_read_edx_eax(vcpu);
6585 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
6587 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
6588 return kvm_skip_emulated_instruction(vcpu);
6592 static int handle_xsaves(struct kvm_vcpu *vcpu)
6594 kvm_skip_emulated_instruction(vcpu);
6595 WARN(1, "this should never happen\n");
6599 static int handle_xrstors(struct kvm_vcpu *vcpu)
6601 kvm_skip_emulated_instruction(vcpu);
6602 WARN(1, "this should never happen\n");
6606 static int handle_apic_access(struct kvm_vcpu *vcpu)
6608 if (likely(fasteoi)) {
6609 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6610 int access_type, offset;
6612 access_type = exit_qualification & APIC_ACCESS_TYPE;
6613 offset = exit_qualification & APIC_ACCESS_OFFSET;
6615 * Sane guest uses MOV to write EOI, with written value
6616 * not cared. So make a short-circuit here by avoiding
6617 * heavy instruction emulation.
6619 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
6620 (offset == APIC_EOI)) {
6621 kvm_lapic_set_eoi(vcpu);
6622 return kvm_skip_emulated_instruction(vcpu);
6625 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6628 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
6630 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6631 int vector = exit_qualification & 0xff;
6633 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
6634 kvm_apic_set_eoi_accelerated(vcpu, vector);
6638 static int handle_apic_write(struct kvm_vcpu *vcpu)
6640 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6641 u32 offset = exit_qualification & 0xfff;
6643 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
6644 kvm_apic_write_nodecode(vcpu, offset);
6648 static int handle_task_switch(struct kvm_vcpu *vcpu)
6650 struct vcpu_vmx *vmx = to_vmx(vcpu);
6651 unsigned long exit_qualification;
6652 bool has_error_code = false;
6655 int reason, type, idt_v, idt_index;
6657 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
6658 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
6659 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
6661 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6663 reason = (u32)exit_qualification >> 30;
6664 if (reason == TASK_SWITCH_GATE && idt_v) {
6666 case INTR_TYPE_NMI_INTR:
6667 vcpu->arch.nmi_injected = false;
6668 vmx_set_nmi_mask(vcpu, true);
6670 case INTR_TYPE_EXT_INTR:
6671 case INTR_TYPE_SOFT_INTR:
6672 kvm_clear_interrupt_queue(vcpu);
6674 case INTR_TYPE_HARD_EXCEPTION:
6675 if (vmx->idt_vectoring_info &
6676 VECTORING_INFO_DELIVER_CODE_MASK) {
6677 has_error_code = true;
6679 vmcs_read32(IDT_VECTORING_ERROR_CODE);
6682 case INTR_TYPE_SOFT_EXCEPTION:
6683 kvm_clear_exception_queue(vcpu);
6689 tss_selector = exit_qualification;
6691 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
6692 type != INTR_TYPE_EXT_INTR &&
6693 type != INTR_TYPE_NMI_INTR))
6694 skip_emulated_instruction(vcpu);
6696 if (kvm_task_switch(vcpu, tss_selector,
6697 type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
6698 has_error_code, error_code) == EMULATE_FAIL) {
6699 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6700 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6701 vcpu->run->internal.ndata = 0;
6706 * TODO: What about debug traps on tss switch?
6707 * Are we supposed to inject them and update dr6?
6713 static int handle_ept_violation(struct kvm_vcpu *vcpu)
6715 unsigned long exit_qualification;
6719 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6722 * EPT violation happened while executing iret from NMI,
6723 * "blocked by NMI" bit has to be set before next VM entry.
6724 * There are errata that may cause this bit to not be set:
6727 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
6729 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
6730 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
6732 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6733 trace_kvm_page_fault(gpa, exit_qualification);
6735 /* Is it a read fault? */
6736 error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
6737 ? PFERR_USER_MASK : 0;
6738 /* Is it a write fault? */
6739 error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
6740 ? PFERR_WRITE_MASK : 0;
6741 /* Is it a fetch fault? */
6742 error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
6743 ? PFERR_FETCH_MASK : 0;
6744 /* ept page table entry is present? */
6745 error_code |= (exit_qualification &
6746 (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
6747 EPT_VIOLATION_EXECUTABLE))
6748 ? PFERR_PRESENT_MASK : 0;
6750 error_code |= (exit_qualification & 0x100) != 0 ?
6751 PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
6753 vcpu->arch.exit_qualification = exit_qualification;
6754 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
6757 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
6763 * A nested guest cannot optimize MMIO vmexits, because we have an
6764 * nGPA here instead of the required GPA.
6766 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6767 if (!is_guest_mode(vcpu) &&
6768 !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
6769 trace_kvm_fast_mmio(gpa);
6771 * Doing kvm_skip_emulated_instruction() depends on undefined
6772 * behavior: Intel's manual doesn't mandate
6773 * VM_EXIT_INSTRUCTION_LEN to be set in VMCS when EPT MISCONFIG
6774 * occurs and while on real hardware it was observed to be set,
6775 * other hypervisors (namely Hyper-V) don't set it, we end up
6776 * advancing IP with some random value. Disable fast mmio when
6777 * running nested and keep it for real hardware in hope that
6778 * VM_EXIT_INSTRUCTION_LEN will always be set correctly.
6780 if (!static_cpu_has(X86_FEATURE_HYPERVISOR))
6781 return kvm_skip_emulated_instruction(vcpu);
6783 return x86_emulate_instruction(vcpu, gpa, EMULTYPE_SKIP,
6784 NULL, 0) == EMULATE_DONE;
6787 ret = kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
6791 /* It is the real ept misconfig */
6794 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
6795 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
6800 static int handle_nmi_window(struct kvm_vcpu *vcpu)
6802 WARN_ON_ONCE(!enable_vnmi);
6803 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6804 CPU_BASED_VIRTUAL_NMI_PENDING);
6805 ++vcpu->stat.nmi_window_exits;
6806 kvm_make_request(KVM_REQ_EVENT, vcpu);
6811 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
6813 struct vcpu_vmx *vmx = to_vmx(vcpu);
6814 enum emulation_result err = EMULATE_DONE;
6817 bool intr_window_requested;
6818 unsigned count = 130;
6820 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
6821 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
6823 while (vmx->emulation_required && count-- != 0) {
6824 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
6825 return handle_interrupt_window(&vmx->vcpu);
6827 if (kvm_test_request(KVM_REQ_EVENT, vcpu))
6830 err = emulate_instruction(vcpu, 0);
6832 if (err == EMULATE_USER_EXIT) {
6833 ++vcpu->stat.mmio_exits;
6838 if (err != EMULATE_DONE) {
6839 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6840 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6841 vcpu->run->internal.ndata = 0;
6845 if (vcpu->arch.halt_request) {
6846 vcpu->arch.halt_request = 0;
6847 ret = kvm_vcpu_halt(vcpu);
6851 if (signal_pending(current))
6861 static int __grow_ple_window(int val)
6863 if (ple_window_grow < 1)
6866 val = min(val, ple_window_actual_max);
6868 if (ple_window_grow < ple_window)
6869 val *= ple_window_grow;
6871 val += ple_window_grow;
6876 static int __shrink_ple_window(int val, int modifier, int minimum)
6881 if (modifier < ple_window)
6886 return max(val, minimum);
6889 static void grow_ple_window(struct kvm_vcpu *vcpu)
6891 struct vcpu_vmx *vmx = to_vmx(vcpu);
6892 int old = vmx->ple_window;
6894 vmx->ple_window = __grow_ple_window(old);
6896 if (vmx->ple_window != old)
6897 vmx->ple_window_dirty = true;
6899 trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old);
6902 static void shrink_ple_window(struct kvm_vcpu *vcpu)
6904 struct vcpu_vmx *vmx = to_vmx(vcpu);
6905 int old = vmx->ple_window;
6907 vmx->ple_window = __shrink_ple_window(old,
6908 ple_window_shrink, ple_window);
6910 if (vmx->ple_window != old)
6911 vmx->ple_window_dirty = true;
6913 trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old);
6917 * ple_window_actual_max is computed to be one grow_ple_window() below
6918 * ple_window_max. (See __grow_ple_window for the reason.)
6919 * This prevents overflows, because ple_window_max is int.
6920 * ple_window_max effectively rounded down to a multiple of ple_window_grow in
6922 * ple_window_max is also prevented from setting vmx->ple_window < ple_window.
6924 static void update_ple_window_actual_max(void)
6926 ple_window_actual_max =
6927 __shrink_ple_window(max(ple_window_max, ple_window),
6928 ple_window_grow, INT_MIN);
6932 * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
6934 static void wakeup_handler(void)
6936 struct kvm_vcpu *vcpu;
6937 int cpu = smp_processor_id();
6939 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6940 list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
6941 blocked_vcpu_list) {
6942 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6944 if (pi_test_on(pi_desc) == 1)
6945 kvm_vcpu_kick(vcpu);
6947 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6950 void vmx_enable_tdp(void)
6952 kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
6953 enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
6954 enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
6955 0ull, VMX_EPT_EXECUTABLE_MASK,
6956 cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
6957 VMX_EPT_RWX_MASK, 0ull);
6959 ept_set_mmio_spte_mask();
6963 static __init int hardware_setup(void)
6967 rdmsrl_safe(MSR_EFER, &host_efer);
6969 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
6970 kvm_define_shared_msr(i, vmx_msr_index[i]);
6972 for (i = 0; i < VMX_BITMAP_NR; i++) {
6973 vmx_bitmap[i] = (unsigned long *)__get_free_page(GFP_KERNEL);
6978 memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
6979 memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
6981 if (setup_vmcs_config(&vmcs_config) < 0) {
6986 if (boot_cpu_has(X86_FEATURE_NX))
6987 kvm_enable_efer_bits(EFER_NX);
6989 if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
6990 !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
6993 if (!cpu_has_vmx_ept() ||
6994 !cpu_has_vmx_ept_4levels() ||
6995 !cpu_has_vmx_ept_mt_wb() ||
6996 !cpu_has_vmx_invept_global())
6999 if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
7000 enable_ept_ad_bits = 0;
7002 if (!cpu_has_vmx_unrestricted_guest() || !enable_ept)
7003 enable_unrestricted_guest = 0;
7005 if (!cpu_has_vmx_flexpriority())
7006 flexpriority_enabled = 0;
7008 if (!cpu_has_virtual_nmis())
7012 * set_apic_access_page_addr() is used to reload apic access
7013 * page upon invalidation. No need to do anything if not
7014 * using the APIC_ACCESS_ADDR VMCS field.
7016 if (!flexpriority_enabled)
7017 kvm_x86_ops->set_apic_access_page_addr = NULL;
7019 if (!cpu_has_vmx_tpr_shadow())
7020 kvm_x86_ops->update_cr8_intercept = NULL;
7022 if (enable_ept && !cpu_has_vmx_ept_2m_page())
7023 kvm_disable_largepages();
7025 if (!cpu_has_vmx_ple()) {
7028 ple_window_grow = 0;
7030 ple_window_shrink = 0;
7033 if (!cpu_has_vmx_apicv()) {
7035 kvm_x86_ops->sync_pir_to_irr = NULL;
7038 if (cpu_has_vmx_tsc_scaling()) {
7039 kvm_has_tsc_control = true;
7040 kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
7041 kvm_tsc_scaling_ratio_frac_bits = 48;
7044 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
7051 update_ple_window_actual_max();
7054 * Only enable PML when hardware supports PML feature, and both EPT
7055 * and EPT A/D bit features are enabled -- PML depends on them to work.
7057 if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
7061 kvm_x86_ops->slot_enable_log_dirty = NULL;
7062 kvm_x86_ops->slot_disable_log_dirty = NULL;
7063 kvm_x86_ops->flush_log_dirty = NULL;
7064 kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
7067 if (cpu_has_vmx_preemption_timer() && enable_preemption_timer) {
7070 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
7071 cpu_preemption_timer_multi =
7072 vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
7074 kvm_x86_ops->set_hv_timer = NULL;
7075 kvm_x86_ops->cancel_hv_timer = NULL;
7078 if (!cpu_has_vmx_shadow_vmcs())
7079 enable_shadow_vmcs = 0;
7080 if (enable_shadow_vmcs)
7081 init_vmcs_shadow_fields();
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->arch.emulate_ctxt, gva, vmpointer,
7321 sizeof(*vmpointer), &e)) {
7322 kvm_inject_page_fault(vcpu, &e);
7329 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
7331 struct vcpu_vmx *vmx = to_vmx(vcpu);
7332 struct vmcs *shadow_vmcs;
7335 r = alloc_loaded_vmcs(&vmx->nested.vmcs02);
7339 vmx->nested.cached_vmcs12 = kmalloc(VMCS12_SIZE, GFP_KERNEL);
7340 if (!vmx->nested.cached_vmcs12)
7341 goto out_cached_vmcs12;
7343 if (enable_shadow_vmcs) {
7344 shadow_vmcs = alloc_vmcs();
7346 goto out_shadow_vmcs;
7347 /* mark vmcs as shadow */
7348 shadow_vmcs->revision_id |= (1u << 31);
7349 /* init shadow vmcs */
7350 vmcs_clear(shadow_vmcs);
7351 vmx->vmcs01.shadow_vmcs = shadow_vmcs;
7354 hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
7355 HRTIMER_MODE_REL_PINNED);
7356 vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
7358 vmx->nested.vmxon = true;
7362 kfree(vmx->nested.cached_vmcs12);
7365 free_loaded_vmcs(&vmx->nested.vmcs02);
7372 * Emulate the VMXON instruction.
7373 * Currently, we just remember that VMX is active, and do not save or even
7374 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
7375 * do not currently need to store anything in that guest-allocated memory
7376 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
7377 * argument is different from the VMXON pointer (which the spec says they do).
7379 static int handle_vmon(struct kvm_vcpu *vcpu)
7384 struct vcpu_vmx *vmx = to_vmx(vcpu);
7385 const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
7386 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
7389 * The Intel VMX Instruction Reference lists a bunch of bits that are
7390 * prerequisite to running VMXON, most notably cr4.VMXE must be set to
7391 * 1 (see vmx_set_cr4() for when we allow the guest to set this).
7392 * Otherwise, we should fail with #UD. But most faulting conditions
7393 * have already been checked by hardware, prior to the VM-exit for
7394 * VMXON. We do test guest cr4.VMXE because processor CR4 always has
7395 * that bit set to 1 in non-root mode.
7397 if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE)) {
7398 kvm_queue_exception(vcpu, UD_VECTOR);
7402 if (vmx->nested.vmxon) {
7403 nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
7404 return kvm_skip_emulated_instruction(vcpu);
7407 if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
7408 != VMXON_NEEDED_FEATURES) {
7409 kvm_inject_gp(vcpu, 0);
7413 if (nested_vmx_get_vmptr(vcpu, &vmptr))
7418 * The first 4 bytes of VMXON region contain the supported
7419 * VMCS revision identifier
7421 * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
7422 * which replaces physical address width with 32
7424 if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu))) {
7425 nested_vmx_failInvalid(vcpu);
7426 return kvm_skip_emulated_instruction(vcpu);
7429 page = kvm_vcpu_gpa_to_page(vcpu, vmptr);
7430 if (is_error_page(page)) {
7431 nested_vmx_failInvalid(vcpu);
7432 return kvm_skip_emulated_instruction(vcpu);
7434 if (*(u32 *)kmap(page) != VMCS12_REVISION) {
7436 kvm_release_page_clean(page);
7437 nested_vmx_failInvalid(vcpu);
7438 return kvm_skip_emulated_instruction(vcpu);
7441 kvm_release_page_clean(page);
7443 vmx->nested.vmxon_ptr = vmptr;
7444 ret = enter_vmx_operation(vcpu);
7448 nested_vmx_succeed(vcpu);
7449 return kvm_skip_emulated_instruction(vcpu);
7453 * Intel's VMX Instruction Reference specifies a common set of prerequisites
7454 * for running VMX instructions (except VMXON, whose prerequisites are
7455 * slightly different). It also specifies what exception to inject otherwise.
7456 * Note that many of these exceptions have priority over VM exits, so they
7457 * don't have to be checked again here.
7459 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
7461 if (!to_vmx(vcpu)->nested.vmxon) {
7462 kvm_queue_exception(vcpu, UD_VECTOR);
7468 static void vmx_disable_shadow_vmcs(struct vcpu_vmx *vmx)
7470 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL, SECONDARY_EXEC_SHADOW_VMCS);
7471 vmcs_write64(VMCS_LINK_POINTER, -1ull);
7474 static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
7476 if (vmx->nested.current_vmptr == -1ull)
7479 if (enable_shadow_vmcs) {
7480 /* copy to memory all shadowed fields in case
7481 they were modified */
7482 copy_shadow_to_vmcs12(vmx);
7483 vmx->nested.sync_shadow_vmcs = false;
7484 vmx_disable_shadow_vmcs(vmx);
7486 vmx->nested.posted_intr_nv = -1;
7488 /* Flush VMCS12 to guest memory */
7489 kvm_vcpu_write_guest_page(&vmx->vcpu,
7490 vmx->nested.current_vmptr >> PAGE_SHIFT,
7491 vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
7493 vmx->nested.current_vmptr = -1ull;
7497 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
7498 * just stops using VMX.
7500 static void free_nested(struct vcpu_vmx *vmx)
7502 if (!vmx->nested.vmxon && !vmx->nested.smm.vmxon)
7505 vmx->nested.vmxon = false;
7506 vmx->nested.smm.vmxon = false;
7507 free_vpid(vmx->nested.vpid02);
7508 vmx->nested.posted_intr_nv = -1;
7509 vmx->nested.current_vmptr = -1ull;
7510 if (enable_shadow_vmcs) {
7511 vmx_disable_shadow_vmcs(vmx);
7512 vmcs_clear(vmx->vmcs01.shadow_vmcs);
7513 free_vmcs(vmx->vmcs01.shadow_vmcs);
7514 vmx->vmcs01.shadow_vmcs = NULL;
7516 kfree(vmx->nested.cached_vmcs12);
7517 /* Unpin physical memory we referred to in the vmcs02 */
7518 if (vmx->nested.apic_access_page) {
7519 kvm_release_page_dirty(vmx->nested.apic_access_page);
7520 vmx->nested.apic_access_page = NULL;
7522 if (vmx->nested.virtual_apic_page) {
7523 kvm_release_page_dirty(vmx->nested.virtual_apic_page);
7524 vmx->nested.virtual_apic_page = NULL;
7526 if (vmx->nested.pi_desc_page) {
7527 kunmap(vmx->nested.pi_desc_page);
7528 kvm_release_page_dirty(vmx->nested.pi_desc_page);
7529 vmx->nested.pi_desc_page = NULL;
7530 vmx->nested.pi_desc = NULL;
7533 free_loaded_vmcs(&vmx->nested.vmcs02);
7536 /* Emulate the VMXOFF instruction */
7537 static int handle_vmoff(struct kvm_vcpu *vcpu)
7539 if (!nested_vmx_check_permission(vcpu))
7541 free_nested(to_vmx(vcpu));
7542 nested_vmx_succeed(vcpu);
7543 return kvm_skip_emulated_instruction(vcpu);
7546 /* Emulate the VMCLEAR instruction */
7547 static int handle_vmclear(struct kvm_vcpu *vcpu)
7549 struct vcpu_vmx *vmx = to_vmx(vcpu);
7553 if (!nested_vmx_check_permission(vcpu))
7556 if (nested_vmx_get_vmptr(vcpu, &vmptr))
7559 if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu))) {
7560 nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
7561 return kvm_skip_emulated_instruction(vcpu);
7564 if (vmptr == vmx->nested.vmxon_ptr) {
7565 nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_VMXON_POINTER);
7566 return kvm_skip_emulated_instruction(vcpu);
7569 if (vmptr == vmx->nested.current_vmptr)
7570 nested_release_vmcs12(vmx);
7572 kvm_vcpu_write_guest(vcpu,
7573 vmptr + offsetof(struct vmcs12, launch_state),
7574 &zero, sizeof(zero));
7576 nested_vmx_succeed(vcpu);
7577 return kvm_skip_emulated_instruction(vcpu);
7580 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
7582 /* Emulate the VMLAUNCH instruction */
7583 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
7585 return nested_vmx_run(vcpu, true);
7588 /* Emulate the VMRESUME instruction */
7589 static int handle_vmresume(struct kvm_vcpu *vcpu)
7592 return nested_vmx_run(vcpu, false);
7596 * Read a vmcs12 field. Since these can have varying lengths and we return
7597 * one type, we chose the biggest type (u64) and zero-extend the return value
7598 * to that size. Note that the caller, handle_vmread, might need to use only
7599 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
7600 * 64-bit fields are to be returned).
7602 static inline int vmcs12_read_any(struct kvm_vcpu *vcpu,
7603 unsigned long field, u64 *ret)
7605 short offset = vmcs_field_to_offset(field);
7611 p = ((char *)(get_vmcs12(vcpu))) + offset;
7613 switch (vmcs_field_width(field)) {
7614 case VMCS_FIELD_WIDTH_NATURAL_WIDTH:
7615 *ret = *((natural_width *)p);
7617 case VMCS_FIELD_WIDTH_U16:
7620 case VMCS_FIELD_WIDTH_U32:
7623 case VMCS_FIELD_WIDTH_U64:
7633 static inline int vmcs12_write_any(struct kvm_vcpu *vcpu,
7634 unsigned long field, u64 field_value){
7635 short offset = vmcs_field_to_offset(field);
7636 char *p = ((char *) get_vmcs12(vcpu)) + offset;
7640 switch (vmcs_field_width(field)) {
7641 case VMCS_FIELD_WIDTH_U16:
7642 *(u16 *)p = field_value;
7644 case VMCS_FIELD_WIDTH_U32:
7645 *(u32 *)p = field_value;
7647 case VMCS_FIELD_WIDTH_U64:
7648 *(u64 *)p = field_value;
7650 case VMCS_FIELD_WIDTH_NATURAL_WIDTH:
7651 *(natural_width *)p = field_value;
7660 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
7663 unsigned long field;
7665 struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
7666 const u16 *fields = shadow_read_write_fields;
7667 const int num_fields = max_shadow_read_write_fields;
7671 vmcs_load(shadow_vmcs);
7673 for (i = 0; i < num_fields; i++) {
7675 field_value = __vmcs_readl(field);
7676 vmcs12_write_any(&vmx->vcpu, field, field_value);
7679 vmcs_clear(shadow_vmcs);
7680 vmcs_load(vmx->loaded_vmcs->vmcs);
7685 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
7687 const u16 *fields[] = {
7688 shadow_read_write_fields,
7689 shadow_read_only_fields
7691 const int max_fields[] = {
7692 max_shadow_read_write_fields,
7693 max_shadow_read_only_fields
7696 unsigned long field;
7697 u64 field_value = 0;
7698 struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
7700 vmcs_load(shadow_vmcs);
7702 for (q = 0; q < ARRAY_SIZE(fields); q++) {
7703 for (i = 0; i < max_fields[q]; i++) {
7704 field = fields[q][i];
7705 vmcs12_read_any(&vmx->vcpu, field, &field_value);
7706 __vmcs_writel(field, field_value);
7710 vmcs_clear(shadow_vmcs);
7711 vmcs_load(vmx->loaded_vmcs->vmcs);
7715 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
7716 * used before) all generate the same failure when it is missing.
7718 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
7720 struct vcpu_vmx *vmx = to_vmx(vcpu);
7721 if (vmx->nested.current_vmptr == -1ull) {
7722 nested_vmx_failInvalid(vcpu);
7728 static int handle_vmread(struct kvm_vcpu *vcpu)
7730 unsigned long field;
7732 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7733 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7736 if (!nested_vmx_check_permission(vcpu))
7739 if (!nested_vmx_check_vmcs12(vcpu))
7740 return kvm_skip_emulated_instruction(vcpu);
7742 /* Decode instruction info and find the field to read */
7743 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7744 /* Read the field, zero-extended to a u64 field_value */
7745 if (vmcs12_read_any(vcpu, field, &field_value) < 0) {
7746 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7747 return kvm_skip_emulated_instruction(vcpu);
7750 * Now copy part of this value to register or memory, as requested.
7751 * Note that the number of bits actually copied is 32 or 64 depending
7752 * on the guest's mode (32 or 64 bit), not on the given field's length.
7754 if (vmx_instruction_info & (1u << 10)) {
7755 kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
7758 if (get_vmx_mem_address(vcpu, exit_qualification,
7759 vmx_instruction_info, true, &gva))
7761 /* _system ok, as hardware has verified cpl=0 */
7762 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
7763 &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
7766 nested_vmx_succeed(vcpu);
7767 return kvm_skip_emulated_instruction(vcpu);
7771 static int handle_vmwrite(struct kvm_vcpu *vcpu)
7773 unsigned long field;
7775 struct vcpu_vmx *vmx = to_vmx(vcpu);
7776 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7777 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7779 /* The value to write might be 32 or 64 bits, depending on L1's long
7780 * mode, and eventually we need to write that into a field of several
7781 * possible lengths. The code below first zero-extends the value to 64
7782 * bit (field_value), and then copies only the appropriate number of
7783 * bits into the vmcs12 field.
7785 u64 field_value = 0;
7786 struct x86_exception e;
7788 if (!nested_vmx_check_permission(vcpu))
7791 if (!nested_vmx_check_vmcs12(vcpu))
7792 return kvm_skip_emulated_instruction(vcpu);
7794 if (vmx_instruction_info & (1u << 10))
7795 field_value = kvm_register_readl(vcpu,
7796 (((vmx_instruction_info) >> 3) & 0xf));
7798 if (get_vmx_mem_address(vcpu, exit_qualification,
7799 vmx_instruction_info, false, &gva))
7801 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
7802 &field_value, (is_64_bit_mode(vcpu) ? 8 : 4), &e)) {
7803 kvm_inject_page_fault(vcpu, &e);
7809 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7810 if (vmcs_field_readonly(field)) {
7811 nested_vmx_failValid(vcpu,
7812 VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
7813 return kvm_skip_emulated_instruction(vcpu);
7816 if (vmcs12_write_any(vcpu, field, field_value) < 0) {
7817 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7818 return kvm_skip_emulated_instruction(vcpu);
7822 #define SHADOW_FIELD_RW(x) case x:
7823 #include "vmx_shadow_fields.h"
7825 * The fields that can be updated by L1 without a vmexit are
7826 * always updated in the vmcs02, the others go down the slow
7827 * path of prepare_vmcs02.
7831 vmx->nested.dirty_vmcs12 = true;
7835 nested_vmx_succeed(vcpu);
7836 return kvm_skip_emulated_instruction(vcpu);
7839 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
7841 vmx->nested.current_vmptr = vmptr;
7842 if (enable_shadow_vmcs) {
7843 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
7844 SECONDARY_EXEC_SHADOW_VMCS);
7845 vmcs_write64(VMCS_LINK_POINTER,
7846 __pa(vmx->vmcs01.shadow_vmcs));
7847 vmx->nested.sync_shadow_vmcs = true;
7849 vmx->nested.dirty_vmcs12 = true;
7852 /* Emulate the VMPTRLD instruction */
7853 static int handle_vmptrld(struct kvm_vcpu *vcpu)
7855 struct vcpu_vmx *vmx = to_vmx(vcpu);
7858 if (!nested_vmx_check_permission(vcpu))
7861 if (nested_vmx_get_vmptr(vcpu, &vmptr))
7864 if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu))) {
7865 nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
7866 return kvm_skip_emulated_instruction(vcpu);
7869 if (vmptr == vmx->nested.vmxon_ptr) {
7870 nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_VMXON_POINTER);
7871 return kvm_skip_emulated_instruction(vcpu);
7874 if (vmx->nested.current_vmptr != vmptr) {
7875 struct vmcs12 *new_vmcs12;
7877 page = kvm_vcpu_gpa_to_page(vcpu, vmptr);
7878 if (is_error_page(page)) {
7879 nested_vmx_failInvalid(vcpu);
7880 return kvm_skip_emulated_instruction(vcpu);
7882 new_vmcs12 = kmap(page);
7883 if (new_vmcs12->revision_id != VMCS12_REVISION) {
7885 kvm_release_page_clean(page);
7886 nested_vmx_failValid(vcpu,
7887 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
7888 return kvm_skip_emulated_instruction(vcpu);
7891 nested_release_vmcs12(vmx);
7893 * Load VMCS12 from guest memory since it is not already
7896 memcpy(vmx->nested.cached_vmcs12, new_vmcs12, VMCS12_SIZE);
7898 kvm_release_page_clean(page);
7900 set_current_vmptr(vmx, vmptr);
7903 nested_vmx_succeed(vcpu);
7904 return kvm_skip_emulated_instruction(vcpu);
7907 /* Emulate the VMPTRST instruction */
7908 static int handle_vmptrst(struct kvm_vcpu *vcpu)
7910 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7911 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7913 struct x86_exception e;
7915 if (!nested_vmx_check_permission(vcpu))
7918 if (get_vmx_mem_address(vcpu, exit_qualification,
7919 vmx_instruction_info, true, &vmcs_gva))
7921 /* ok to use *_system, as hardware has verified cpl=0 */
7922 if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
7923 (void *)&to_vmx(vcpu)->nested.current_vmptr,
7925 kvm_inject_page_fault(vcpu, &e);
7928 nested_vmx_succeed(vcpu);
7929 return kvm_skip_emulated_instruction(vcpu);
7932 /* Emulate the INVEPT instruction */
7933 static int handle_invept(struct kvm_vcpu *vcpu)
7935 struct vcpu_vmx *vmx = to_vmx(vcpu);
7936 u32 vmx_instruction_info, types;
7939 struct x86_exception e;
7944 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7945 SECONDARY_EXEC_ENABLE_EPT) ||
7946 !(vmx->nested.nested_vmx_ept_caps & VMX_EPT_INVEPT_BIT)) {
7947 kvm_queue_exception(vcpu, UD_VECTOR);
7951 if (!nested_vmx_check_permission(vcpu))
7954 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7955 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7957 types = (vmx->nested.nested_vmx_ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
7959 if (type >= 32 || !(types & (1 << type))) {
7960 nested_vmx_failValid(vcpu,
7961 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7962 return kvm_skip_emulated_instruction(vcpu);
7965 /* According to the Intel VMX instruction reference, the memory
7966 * operand is read even if it isn't needed (e.g., for type==global)
7968 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7969 vmx_instruction_info, false, &gva))
7971 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
7972 sizeof(operand), &e)) {
7973 kvm_inject_page_fault(vcpu, &e);
7978 case VMX_EPT_EXTENT_GLOBAL:
7980 * TODO: track mappings and invalidate
7981 * single context requests appropriately
7983 case VMX_EPT_EXTENT_CONTEXT:
7984 kvm_mmu_sync_roots(vcpu);
7985 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
7986 nested_vmx_succeed(vcpu);
7993 return kvm_skip_emulated_instruction(vcpu);
7996 static int handle_invvpid(struct kvm_vcpu *vcpu)
7998 struct vcpu_vmx *vmx = to_vmx(vcpu);
7999 u32 vmx_instruction_info;
8000 unsigned long type, types;
8002 struct x86_exception e;
8008 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
8009 SECONDARY_EXEC_ENABLE_VPID) ||
8010 !(vmx->nested.nested_vmx_vpid_caps & VMX_VPID_INVVPID_BIT)) {
8011 kvm_queue_exception(vcpu, UD_VECTOR);
8015 if (!nested_vmx_check_permission(vcpu))
8018 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
8019 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
8021 types = (vmx->nested.nested_vmx_vpid_caps &
8022 VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
8024 if (type >= 32 || !(types & (1 << type))) {
8025 nested_vmx_failValid(vcpu,
8026 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
8027 return kvm_skip_emulated_instruction(vcpu);
8030 /* according to the intel vmx instruction reference, the memory
8031 * operand is read even if it isn't needed (e.g., for type==global)
8033 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
8034 vmx_instruction_info, false, &gva))
8036 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
8037 sizeof(operand), &e)) {
8038 kvm_inject_page_fault(vcpu, &e);
8041 if (operand.vpid >> 16) {
8042 nested_vmx_failValid(vcpu,
8043 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
8044 return kvm_skip_emulated_instruction(vcpu);
8048 case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
8049 if (is_noncanonical_address(operand.gla, vcpu)) {
8050 nested_vmx_failValid(vcpu,
8051 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
8052 return kvm_skip_emulated_instruction(vcpu);
8055 case VMX_VPID_EXTENT_SINGLE_CONTEXT:
8056 case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
8057 if (!operand.vpid) {
8058 nested_vmx_failValid(vcpu,
8059 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
8060 return kvm_skip_emulated_instruction(vcpu);
8063 case VMX_VPID_EXTENT_ALL_CONTEXT:
8067 return kvm_skip_emulated_instruction(vcpu);
8070 __vmx_flush_tlb(vcpu, vmx->nested.vpid02, true);
8071 nested_vmx_succeed(vcpu);
8073 return kvm_skip_emulated_instruction(vcpu);
8076 static int handle_pml_full(struct kvm_vcpu *vcpu)
8078 unsigned long exit_qualification;
8080 trace_kvm_pml_full(vcpu->vcpu_id);
8082 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
8085 * PML buffer FULL happened while executing iret from NMI,
8086 * "blocked by NMI" bit has to be set before next VM entry.
8088 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
8090 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
8091 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
8092 GUEST_INTR_STATE_NMI);
8095 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
8096 * here.., and there's no userspace involvement needed for PML.
8101 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
8103 kvm_lapic_expired_hv_timer(vcpu);
8107 static bool valid_ept_address(struct kvm_vcpu *vcpu, u64 address)
8109 struct vcpu_vmx *vmx = to_vmx(vcpu);
8110 int maxphyaddr = cpuid_maxphyaddr(vcpu);
8112 /* Check for memory type validity */
8113 switch (address & VMX_EPTP_MT_MASK) {
8114 case VMX_EPTP_MT_UC:
8115 if (!(vmx->nested.nested_vmx_ept_caps & VMX_EPTP_UC_BIT))
8118 case VMX_EPTP_MT_WB:
8119 if (!(vmx->nested.nested_vmx_ept_caps & VMX_EPTP_WB_BIT))
8126 /* only 4 levels page-walk length are valid */
8127 if ((address & VMX_EPTP_PWL_MASK) != VMX_EPTP_PWL_4)
8130 /* Reserved bits should not be set */
8131 if (address >> maxphyaddr || ((address >> 7) & 0x1f))
8134 /* AD, if set, should be supported */
8135 if (address & VMX_EPTP_AD_ENABLE_BIT) {
8136 if (!(vmx->nested.nested_vmx_ept_caps & VMX_EPT_AD_BIT))
8143 static int nested_vmx_eptp_switching(struct kvm_vcpu *vcpu,
8144 struct vmcs12 *vmcs12)
8146 u32 index = vcpu->arch.regs[VCPU_REGS_RCX];
8148 bool accessed_dirty;
8149 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
8151 if (!nested_cpu_has_eptp_switching(vmcs12) ||
8152 !nested_cpu_has_ept(vmcs12))
8155 if (index >= VMFUNC_EPTP_ENTRIES)
8159 if (kvm_vcpu_read_guest_page(vcpu, vmcs12->eptp_list_address >> PAGE_SHIFT,
8160 &address, index * 8, 8))
8163 accessed_dirty = !!(address & VMX_EPTP_AD_ENABLE_BIT);
8166 * If the (L2) guest does a vmfunc to the currently
8167 * active ept pointer, we don't have to do anything else
8169 if (vmcs12->ept_pointer != address) {
8170 if (!valid_ept_address(vcpu, address))
8173 kvm_mmu_unload(vcpu);
8174 mmu->ept_ad = accessed_dirty;
8175 mmu->base_role.ad_disabled = !accessed_dirty;
8176 vmcs12->ept_pointer = address;
8178 * TODO: Check what's the correct approach in case
8179 * mmu reload fails. Currently, we just let the next
8180 * reload potentially fail
8182 kvm_mmu_reload(vcpu);
8188 static int handle_vmfunc(struct kvm_vcpu *vcpu)
8190 struct vcpu_vmx *vmx = to_vmx(vcpu);
8191 struct vmcs12 *vmcs12;
8192 u32 function = vcpu->arch.regs[VCPU_REGS_RAX];
8195 * VMFUNC is only supported for nested guests, but we always enable the
8196 * secondary control for simplicity; for non-nested mode, fake that we
8197 * didn't by injecting #UD.
8199 if (!is_guest_mode(vcpu)) {
8200 kvm_queue_exception(vcpu, UD_VECTOR);
8204 vmcs12 = get_vmcs12(vcpu);
8205 if ((vmcs12->vm_function_control & (1 << function)) == 0)
8210 if (nested_vmx_eptp_switching(vcpu, vmcs12))
8216 return kvm_skip_emulated_instruction(vcpu);
8219 nested_vmx_vmexit(vcpu, vmx->exit_reason,
8220 vmcs_read32(VM_EXIT_INTR_INFO),
8221 vmcs_readl(EXIT_QUALIFICATION));
8226 * The exit handlers return 1 if the exit was handled fully and guest execution
8227 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
8228 * to be done to userspace and return 0.
8230 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
8231 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
8232 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
8233 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
8234 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
8235 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
8236 [EXIT_REASON_CR_ACCESS] = handle_cr,
8237 [EXIT_REASON_DR_ACCESS] = handle_dr,
8238 [EXIT_REASON_CPUID] = handle_cpuid,
8239 [EXIT_REASON_MSR_READ] = handle_rdmsr,
8240 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
8241 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
8242 [EXIT_REASON_HLT] = handle_halt,
8243 [EXIT_REASON_INVD] = handle_invd,
8244 [EXIT_REASON_INVLPG] = handle_invlpg,
8245 [EXIT_REASON_RDPMC] = handle_rdpmc,
8246 [EXIT_REASON_VMCALL] = handle_vmcall,
8247 [EXIT_REASON_VMCLEAR] = handle_vmclear,
8248 [EXIT_REASON_VMLAUNCH] = handle_vmlaunch,
8249 [EXIT_REASON_VMPTRLD] = handle_vmptrld,
8250 [EXIT_REASON_VMPTRST] = handle_vmptrst,
8251 [EXIT_REASON_VMREAD] = handle_vmread,
8252 [EXIT_REASON_VMRESUME] = handle_vmresume,
8253 [EXIT_REASON_VMWRITE] = handle_vmwrite,
8254 [EXIT_REASON_VMOFF] = handle_vmoff,
8255 [EXIT_REASON_VMON] = handle_vmon,
8256 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
8257 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
8258 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
8259 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
8260 [EXIT_REASON_WBINVD] = handle_wbinvd,
8261 [EXIT_REASON_XSETBV] = handle_xsetbv,
8262 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
8263 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
8264 [EXIT_REASON_GDTR_IDTR] = handle_desc,
8265 [EXIT_REASON_LDTR_TR] = handle_desc,
8266 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
8267 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
8268 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
8269 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait,
8270 [EXIT_REASON_MONITOR_TRAP_FLAG] = handle_monitor_trap,
8271 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor,
8272 [EXIT_REASON_INVEPT] = handle_invept,
8273 [EXIT_REASON_INVVPID] = handle_invvpid,
8274 [EXIT_REASON_RDRAND] = handle_invalid_op,
8275 [EXIT_REASON_RDSEED] = handle_invalid_op,
8276 [EXIT_REASON_XSAVES] = handle_xsaves,
8277 [EXIT_REASON_XRSTORS] = handle_xrstors,
8278 [EXIT_REASON_PML_FULL] = handle_pml_full,
8279 [EXIT_REASON_VMFUNC] = handle_vmfunc,
8280 [EXIT_REASON_PREEMPTION_TIMER] = handle_preemption_timer,
8283 static const int kvm_vmx_max_exit_handlers =
8284 ARRAY_SIZE(kvm_vmx_exit_handlers);
8286 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
8287 struct vmcs12 *vmcs12)
8289 unsigned long exit_qualification;
8290 gpa_t bitmap, last_bitmap;
8295 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
8296 return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
8298 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
8300 port = exit_qualification >> 16;
8301 size = (exit_qualification & 7) + 1;
8303 last_bitmap = (gpa_t)-1;
8308 bitmap = vmcs12->io_bitmap_a;
8309 else if (port < 0x10000)
8310 bitmap = vmcs12->io_bitmap_b;
8313 bitmap += (port & 0x7fff) / 8;
8315 if (last_bitmap != bitmap)
8316 if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
8318 if (b & (1 << (port & 7)))
8323 last_bitmap = bitmap;
8330 * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
8331 * rather than handle it ourselves in L0. I.e., check whether L1 expressed
8332 * disinterest in the current event (read or write a specific MSR) by using an
8333 * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
8335 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
8336 struct vmcs12 *vmcs12, u32 exit_reason)
8338 u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
8341 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
8345 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
8346 * for the four combinations of read/write and low/high MSR numbers.
8347 * First we need to figure out which of the four to use:
8349 bitmap = vmcs12->msr_bitmap;
8350 if (exit_reason == EXIT_REASON_MSR_WRITE)
8352 if (msr_index >= 0xc0000000) {
8353 msr_index -= 0xc0000000;
8357 /* Then read the msr_index'th bit from this bitmap: */
8358 if (msr_index < 1024*8) {
8360 if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
8362 return 1 & (b >> (msr_index & 7));
8364 return true; /* let L1 handle the wrong parameter */
8368 * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
8369 * rather than handle it ourselves in L0. I.e., check if L1 wanted to
8370 * intercept (via guest_host_mask etc.) the current event.
8372 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
8373 struct vmcs12 *vmcs12)
8375 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
8376 int cr = exit_qualification & 15;
8380 switch ((exit_qualification >> 4) & 3) {
8381 case 0: /* mov to cr */
8382 reg = (exit_qualification >> 8) & 15;
8383 val = kvm_register_readl(vcpu, reg);
8386 if (vmcs12->cr0_guest_host_mask &
8387 (val ^ vmcs12->cr0_read_shadow))
8391 if ((vmcs12->cr3_target_count >= 1 &&
8392 vmcs12->cr3_target_value0 == val) ||
8393 (vmcs12->cr3_target_count >= 2 &&
8394 vmcs12->cr3_target_value1 == val) ||
8395 (vmcs12->cr3_target_count >= 3 &&
8396 vmcs12->cr3_target_value2 == val) ||
8397 (vmcs12->cr3_target_count >= 4 &&
8398 vmcs12->cr3_target_value3 == val))
8400 if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
8404 if (vmcs12->cr4_guest_host_mask &
8405 (vmcs12->cr4_read_shadow ^ val))
8409 if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
8415 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
8416 (vmcs12->cr0_read_shadow & X86_CR0_TS))
8419 case 1: /* mov from cr */
8422 if (vmcs12->cpu_based_vm_exec_control &
8423 CPU_BASED_CR3_STORE_EXITING)
8427 if (vmcs12->cpu_based_vm_exec_control &
8428 CPU_BASED_CR8_STORE_EXITING)
8435 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
8436 * cr0. Other attempted changes are ignored, with no exit.
8438 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
8439 if (vmcs12->cr0_guest_host_mask & 0xe &
8440 (val ^ vmcs12->cr0_read_shadow))
8442 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
8443 !(vmcs12->cr0_read_shadow & 0x1) &&
8452 * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
8453 * should handle it ourselves in L0 (and then continue L2). Only call this
8454 * when in is_guest_mode (L2).
8456 static bool nested_vmx_exit_reflected(struct kvm_vcpu *vcpu, u32 exit_reason)
8458 u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8459 struct vcpu_vmx *vmx = to_vmx(vcpu);
8460 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8462 if (vmx->nested.nested_run_pending)
8465 if (unlikely(vmx->fail)) {
8466 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
8467 vmcs_read32(VM_INSTRUCTION_ERROR));
8472 * The host physical addresses of some pages of guest memory
8473 * are loaded into the vmcs02 (e.g. vmcs12's Virtual APIC
8474 * Page). The CPU may write to these pages via their host
8475 * physical address while L2 is running, bypassing any
8476 * address-translation-based dirty tracking (e.g. EPT write
8479 * Mark them dirty on every exit from L2 to prevent them from
8480 * getting out of sync with dirty tracking.
8482 nested_mark_vmcs12_pages_dirty(vcpu);
8484 trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
8485 vmcs_readl(EXIT_QUALIFICATION),
8486 vmx->idt_vectoring_info,
8488 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8491 switch (exit_reason) {
8492 case EXIT_REASON_EXCEPTION_NMI:
8493 if (is_nmi(intr_info))
8495 else if (is_page_fault(intr_info))
8496 return !vmx->vcpu.arch.apf.host_apf_reason && enable_ept;
8497 else if (is_no_device(intr_info) &&
8498 !(vmcs12->guest_cr0 & X86_CR0_TS))
8500 else if (is_debug(intr_info) &&
8502 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
8504 else if (is_breakpoint(intr_info) &&
8505 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
8507 return vmcs12->exception_bitmap &
8508 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
8509 case EXIT_REASON_EXTERNAL_INTERRUPT:
8511 case EXIT_REASON_TRIPLE_FAULT:
8513 case EXIT_REASON_PENDING_INTERRUPT:
8514 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
8515 case EXIT_REASON_NMI_WINDOW:
8516 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
8517 case EXIT_REASON_TASK_SWITCH:
8519 case EXIT_REASON_CPUID:
8521 case EXIT_REASON_HLT:
8522 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
8523 case EXIT_REASON_INVD:
8525 case EXIT_REASON_INVLPG:
8526 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
8527 case EXIT_REASON_RDPMC:
8528 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
8529 case EXIT_REASON_RDRAND:
8530 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND_EXITING);
8531 case EXIT_REASON_RDSEED:
8532 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED_EXITING);
8533 case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
8534 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
8535 case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
8536 case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
8537 case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
8538 case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
8539 case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
8540 case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
8542 * VMX instructions trap unconditionally. This allows L1 to
8543 * emulate them for its L2 guest, i.e., allows 3-level nesting!
8546 case EXIT_REASON_CR_ACCESS:
8547 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
8548 case EXIT_REASON_DR_ACCESS:
8549 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
8550 case EXIT_REASON_IO_INSTRUCTION:
8551 return nested_vmx_exit_handled_io(vcpu, vmcs12);
8552 case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
8553 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
8554 case EXIT_REASON_MSR_READ:
8555 case EXIT_REASON_MSR_WRITE:
8556 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
8557 case EXIT_REASON_INVALID_STATE:
8559 case EXIT_REASON_MWAIT_INSTRUCTION:
8560 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
8561 case EXIT_REASON_MONITOR_TRAP_FLAG:
8562 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_TRAP_FLAG);
8563 case EXIT_REASON_MONITOR_INSTRUCTION:
8564 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
8565 case EXIT_REASON_PAUSE_INSTRUCTION:
8566 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
8567 nested_cpu_has2(vmcs12,
8568 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
8569 case EXIT_REASON_MCE_DURING_VMENTRY:
8571 case EXIT_REASON_TPR_BELOW_THRESHOLD:
8572 return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
8573 case EXIT_REASON_APIC_ACCESS:
8574 return nested_cpu_has2(vmcs12,
8575 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
8576 case EXIT_REASON_APIC_WRITE:
8577 case EXIT_REASON_EOI_INDUCED:
8578 /* apic_write and eoi_induced should exit unconditionally. */
8580 case EXIT_REASON_EPT_VIOLATION:
8582 * L0 always deals with the EPT violation. If nested EPT is
8583 * used, and the nested mmu code discovers that the address is
8584 * missing in the guest EPT table (EPT12), the EPT violation
8585 * will be injected with nested_ept_inject_page_fault()
8588 case EXIT_REASON_EPT_MISCONFIG:
8590 * L2 never uses directly L1's EPT, but rather L0's own EPT
8591 * table (shadow on EPT) or a merged EPT table that L0 built
8592 * (EPT on EPT). So any problems with the structure of the
8593 * table is L0's fault.
8596 case EXIT_REASON_INVPCID:
8598 nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_INVPCID) &&
8599 nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
8600 case EXIT_REASON_WBINVD:
8601 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
8602 case EXIT_REASON_XSETBV:
8604 case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
8606 * This should never happen, since it is not possible to
8607 * set XSS to a non-zero value---neither in L1 nor in L2.
8608 * If if it were, XSS would have to be checked against
8609 * the XSS exit bitmap in vmcs12.
8611 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
8612 case EXIT_REASON_PREEMPTION_TIMER:
8614 case EXIT_REASON_PML_FULL:
8615 /* We emulate PML support to L1. */
8617 case EXIT_REASON_VMFUNC:
8618 /* VM functions are emulated through L2->L0 vmexits. */
8625 static int nested_vmx_reflect_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason)
8627 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8630 * At this point, the exit interruption info in exit_intr_info
8631 * is only valid for EXCEPTION_NMI exits. For EXTERNAL_INTERRUPT
8632 * we need to query the in-kernel LAPIC.
8634 WARN_ON(exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT);
8635 if ((exit_intr_info &
8636 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
8637 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) {
8638 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8639 vmcs12->vm_exit_intr_error_code =
8640 vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
8643 nested_vmx_vmexit(vcpu, exit_reason, exit_intr_info,
8644 vmcs_readl(EXIT_QUALIFICATION));
8648 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
8650 *info1 = vmcs_readl(EXIT_QUALIFICATION);
8651 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
8654 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
8657 __free_page(vmx->pml_pg);
8662 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
8664 struct vcpu_vmx *vmx = to_vmx(vcpu);
8668 pml_idx = vmcs_read16(GUEST_PML_INDEX);
8670 /* Do nothing if PML buffer is empty */
8671 if (pml_idx == (PML_ENTITY_NUM - 1))
8674 /* PML index always points to next available PML buffer entity */
8675 if (pml_idx >= PML_ENTITY_NUM)
8680 pml_buf = page_address(vmx->pml_pg);
8681 for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
8684 gpa = pml_buf[pml_idx];
8685 WARN_ON(gpa & (PAGE_SIZE - 1));
8686 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
8689 /* reset PML index */
8690 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
8694 * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
8695 * Called before reporting dirty_bitmap to userspace.
8697 static void kvm_flush_pml_buffers(struct kvm *kvm)
8700 struct kvm_vcpu *vcpu;
8702 * We only need to kick vcpu out of guest mode here, as PML buffer
8703 * is flushed at beginning of all VMEXITs, and it's obvious that only
8704 * vcpus running in guest are possible to have unflushed GPAs in PML
8707 kvm_for_each_vcpu(i, vcpu, kvm)
8708 kvm_vcpu_kick(vcpu);
8711 static void vmx_dump_sel(char *name, uint32_t sel)
8713 pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
8714 name, vmcs_read16(sel),
8715 vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
8716 vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
8717 vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
8720 static void vmx_dump_dtsel(char *name, uint32_t limit)
8722 pr_err("%s limit=0x%08x, base=0x%016lx\n",
8723 name, vmcs_read32(limit),
8724 vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
8727 static void dump_vmcs(void)
8729 u32 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
8730 u32 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
8731 u32 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
8732 u32 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
8733 u32 secondary_exec_control = 0;
8734 unsigned long cr4 = vmcs_readl(GUEST_CR4);
8735 u64 efer = vmcs_read64(GUEST_IA32_EFER);
8738 if (cpu_has_secondary_exec_ctrls())
8739 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8741 pr_err("*** Guest State ***\n");
8742 pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8743 vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
8744 vmcs_readl(CR0_GUEST_HOST_MASK));
8745 pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8746 cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
8747 pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
8748 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
8749 (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
8751 pr_err("PDPTR0 = 0x%016llx PDPTR1 = 0x%016llx\n",
8752 vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
8753 pr_err("PDPTR2 = 0x%016llx PDPTR3 = 0x%016llx\n",
8754 vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
8756 pr_err("RSP = 0x%016lx RIP = 0x%016lx\n",
8757 vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
8758 pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n",
8759 vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
8760 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8761 vmcs_readl(GUEST_SYSENTER_ESP),
8762 vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
8763 vmx_dump_sel("CS: ", GUEST_CS_SELECTOR);
8764 vmx_dump_sel("DS: ", GUEST_DS_SELECTOR);
8765 vmx_dump_sel("SS: ", GUEST_SS_SELECTOR);
8766 vmx_dump_sel("ES: ", GUEST_ES_SELECTOR);
8767 vmx_dump_sel("FS: ", GUEST_FS_SELECTOR);
8768 vmx_dump_sel("GS: ", GUEST_GS_SELECTOR);
8769 vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
8770 vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
8771 vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
8772 vmx_dump_sel("TR: ", GUEST_TR_SELECTOR);
8773 if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
8774 (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
8775 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
8776 efer, vmcs_read64(GUEST_IA32_PAT));
8777 pr_err("DebugCtl = 0x%016llx DebugExceptions = 0x%016lx\n",
8778 vmcs_read64(GUEST_IA32_DEBUGCTL),
8779 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
8780 if (vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
8781 pr_err("PerfGlobCtl = 0x%016llx\n",
8782 vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
8783 if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
8784 pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
8785 pr_err("Interruptibility = %08x ActivityState = %08x\n",
8786 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
8787 vmcs_read32(GUEST_ACTIVITY_STATE));
8788 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
8789 pr_err("InterruptStatus = %04x\n",
8790 vmcs_read16(GUEST_INTR_STATUS));
8792 pr_err("*** Host State ***\n");
8793 pr_err("RIP = 0x%016lx RSP = 0x%016lx\n",
8794 vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
8795 pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
8796 vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
8797 vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
8798 vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
8799 vmcs_read16(HOST_TR_SELECTOR));
8800 pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
8801 vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
8802 vmcs_readl(HOST_TR_BASE));
8803 pr_err("GDTBase=%016lx IDTBase=%016lx\n",
8804 vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
8805 pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
8806 vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
8807 vmcs_readl(HOST_CR4));
8808 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8809 vmcs_readl(HOST_IA32_SYSENTER_ESP),
8810 vmcs_read32(HOST_IA32_SYSENTER_CS),
8811 vmcs_readl(HOST_IA32_SYSENTER_EIP));
8812 if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
8813 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
8814 vmcs_read64(HOST_IA32_EFER),
8815 vmcs_read64(HOST_IA32_PAT));
8816 if (vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
8817 pr_err("PerfGlobCtl = 0x%016llx\n",
8818 vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
8820 pr_err("*** Control State ***\n");
8821 pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
8822 pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
8823 pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
8824 pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
8825 vmcs_read32(EXCEPTION_BITMAP),
8826 vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
8827 vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
8828 pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
8829 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8830 vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
8831 vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
8832 pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
8833 vmcs_read32(VM_EXIT_INTR_INFO),
8834 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8835 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
8836 pr_err(" reason=%08x qualification=%016lx\n",
8837 vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
8838 pr_err("IDTVectoring: info=%08x errcode=%08x\n",
8839 vmcs_read32(IDT_VECTORING_INFO_FIELD),
8840 vmcs_read32(IDT_VECTORING_ERROR_CODE));
8841 pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
8842 if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
8843 pr_err("TSC Multiplier = 0x%016llx\n",
8844 vmcs_read64(TSC_MULTIPLIER));
8845 if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW)
8846 pr_err("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
8847 if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
8848 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
8849 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
8850 pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
8851 n = vmcs_read32(CR3_TARGET_COUNT);
8852 for (i = 0; i + 1 < n; i += 4)
8853 pr_err("CR3 target%u=%016lx target%u=%016lx\n",
8854 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
8855 i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
8857 pr_err("CR3 target%u=%016lx\n",
8858 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
8859 if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
8860 pr_err("PLE Gap=%08x Window=%08x\n",
8861 vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
8862 if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
8863 pr_err("Virtual processor ID = 0x%04x\n",
8864 vmcs_read16(VIRTUAL_PROCESSOR_ID));
8868 * The guest has exited. See if we can fix it or if we need userspace
8871 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
8873 struct vcpu_vmx *vmx = to_vmx(vcpu);
8874 u32 exit_reason = vmx->exit_reason;
8875 u32 vectoring_info = vmx->idt_vectoring_info;
8877 trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
8880 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
8881 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
8882 * querying dirty_bitmap, we only need to kick all vcpus out of guest
8883 * mode as if vcpus is in root mode, the PML buffer must has been
8887 vmx_flush_pml_buffer(vcpu);
8889 /* If guest state is invalid, start emulating */
8890 if (vmx->emulation_required)
8891 return handle_invalid_guest_state(vcpu);
8893 if (is_guest_mode(vcpu) && nested_vmx_exit_reflected(vcpu, exit_reason))
8894 return nested_vmx_reflect_vmexit(vcpu, exit_reason);
8896 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
8898 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8899 vcpu->run->fail_entry.hardware_entry_failure_reason
8904 if (unlikely(vmx->fail)) {
8905 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8906 vcpu->run->fail_entry.hardware_entry_failure_reason
8907 = vmcs_read32(VM_INSTRUCTION_ERROR);
8913 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
8914 * delivery event since it indicates guest is accessing MMIO.
8915 * The vm-exit can be triggered again after return to guest that
8916 * will cause infinite loop.
8918 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
8919 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
8920 exit_reason != EXIT_REASON_EPT_VIOLATION &&
8921 exit_reason != EXIT_REASON_PML_FULL &&
8922 exit_reason != EXIT_REASON_TASK_SWITCH)) {
8923 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8924 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
8925 vcpu->run->internal.ndata = 3;
8926 vcpu->run->internal.data[0] = vectoring_info;
8927 vcpu->run->internal.data[1] = exit_reason;
8928 vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
8929 if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
8930 vcpu->run->internal.ndata++;
8931 vcpu->run->internal.data[3] =
8932 vmcs_read64(GUEST_PHYSICAL_ADDRESS);
8937 if (unlikely(!enable_vnmi &&
8938 vmx->loaded_vmcs->soft_vnmi_blocked)) {
8939 if (vmx_interrupt_allowed(vcpu)) {
8940 vmx->loaded_vmcs->soft_vnmi_blocked = 0;
8941 } else if (vmx->loaded_vmcs->vnmi_blocked_time > 1000000000LL &&
8942 vcpu->arch.nmi_pending) {
8944 * This CPU don't support us in finding the end of an
8945 * NMI-blocked window if the guest runs with IRQs
8946 * disabled. So we pull the trigger after 1 s of
8947 * futile waiting, but inform the user about this.
8949 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
8950 "state on VCPU %d after 1 s timeout\n",
8951 __func__, vcpu->vcpu_id);
8952 vmx->loaded_vmcs->soft_vnmi_blocked = 0;
8956 if (exit_reason < kvm_vmx_max_exit_handlers
8957 && kvm_vmx_exit_handlers[exit_reason])
8958 return kvm_vmx_exit_handlers[exit_reason](vcpu);
8960 vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n",
8962 kvm_queue_exception(vcpu, UD_VECTOR);
8967 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
8969 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8971 if (is_guest_mode(vcpu) &&
8972 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
8975 if (irr == -1 || tpr < irr) {
8976 vmcs_write32(TPR_THRESHOLD, 0);
8980 vmcs_write32(TPR_THRESHOLD, irr);
8983 static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
8985 u32 sec_exec_control;
8987 /* Postpone execution until vmcs01 is the current VMCS. */
8988 if (is_guest_mode(vcpu)) {
8989 to_vmx(vcpu)->nested.change_vmcs01_virtual_x2apic_mode = true;
8993 if (!cpu_has_vmx_virtualize_x2apic_mode())
8996 if (!cpu_need_tpr_shadow(vcpu))
8999 sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
9002 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9003 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
9005 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
9006 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9007 vmx_flush_tlb_ept_only(vcpu);
9009 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
9011 vmx_update_msr_bitmap(vcpu);
9014 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
9016 struct vcpu_vmx *vmx = to_vmx(vcpu);
9019 * Currently we do not handle the nested case where L2 has an
9020 * APIC access page of its own; that page is still pinned.
9021 * Hence, we skip the case where the VCPU is in guest mode _and_
9022 * L1 prepared an APIC access page for L2.
9024 * For the case where L1 and L2 share the same APIC access page
9025 * (flexpriority=Y but SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES clear
9026 * in the vmcs12), this function will only update either the vmcs01
9027 * or the vmcs02. If the former, the vmcs02 will be updated by
9028 * prepare_vmcs02. If the latter, the vmcs01 will be updated in
9029 * the next L2->L1 exit.
9031 if (!is_guest_mode(vcpu) ||
9032 !nested_cpu_has2(get_vmcs12(&vmx->vcpu),
9033 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
9034 vmcs_write64(APIC_ACCESS_ADDR, hpa);
9035 vmx_flush_tlb_ept_only(vcpu);
9039 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
9047 status = vmcs_read16(GUEST_INTR_STATUS);
9049 if (max_isr != old) {
9051 status |= max_isr << 8;
9052 vmcs_write16(GUEST_INTR_STATUS, status);
9056 static void vmx_set_rvi(int vector)
9064 status = vmcs_read16(GUEST_INTR_STATUS);
9065 old = (u8)status & 0xff;
9066 if ((u8)vector != old) {
9068 status |= (u8)vector;
9069 vmcs_write16(GUEST_INTR_STATUS, status);
9073 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
9076 * When running L2, updating RVI is only relevant when
9077 * vmcs12 virtual-interrupt-delivery enabled.
9078 * However, it can be enabled only when L1 also
9079 * intercepts external-interrupts and in that case
9080 * we should not update vmcs02 RVI but instead intercept
9081 * interrupt. Therefore, do nothing when running L2.
9083 if (!is_guest_mode(vcpu))
9084 vmx_set_rvi(max_irr);
9087 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
9089 struct vcpu_vmx *vmx = to_vmx(vcpu);
9091 bool max_irr_updated;
9093 WARN_ON(!vcpu->arch.apicv_active);
9094 if (pi_test_on(&vmx->pi_desc)) {
9095 pi_clear_on(&vmx->pi_desc);
9097 * IOMMU can write to PIR.ON, so the barrier matters even on UP.
9098 * But on x86 this is just a compiler barrier anyway.
9100 smp_mb__after_atomic();
9102 kvm_apic_update_irr(vcpu, vmx->pi_desc.pir, &max_irr);
9105 * If we are running L2 and L1 has a new pending interrupt
9106 * which can be injected, we should re-evaluate
9107 * what should be done with this new L1 interrupt.
9108 * If L1 intercepts external-interrupts, we should
9109 * exit from L2 to L1. Otherwise, interrupt should be
9110 * delivered directly to L2.
9112 if (is_guest_mode(vcpu) && max_irr_updated) {
9113 if (nested_exit_on_intr(vcpu))
9114 kvm_vcpu_exiting_guest_mode(vcpu);
9116 kvm_make_request(KVM_REQ_EVENT, vcpu);
9119 max_irr = kvm_lapic_find_highest_irr(vcpu);
9121 vmx_hwapic_irr_update(vcpu, max_irr);
9125 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
9127 if (!kvm_vcpu_apicv_active(vcpu))
9130 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
9131 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
9132 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
9133 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
9136 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
9138 struct vcpu_vmx *vmx = to_vmx(vcpu);
9140 pi_clear_on(&vmx->pi_desc);
9141 memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
9144 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
9146 u32 exit_intr_info = 0;
9147 u16 basic_exit_reason = (u16)vmx->exit_reason;
9149 if (!(basic_exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
9150 || basic_exit_reason == EXIT_REASON_EXCEPTION_NMI))
9153 if (!(vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
9154 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
9155 vmx->exit_intr_info = exit_intr_info;
9157 /* if exit due to PF check for async PF */
9158 if (is_page_fault(exit_intr_info))
9159 vmx->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
9161 /* Handle machine checks before interrupts are enabled */
9162 if (basic_exit_reason == EXIT_REASON_MCE_DURING_VMENTRY ||
9163 is_machine_check(exit_intr_info))
9164 kvm_machine_check();
9166 /* We need to handle NMIs before interrupts are enabled */
9167 if (is_nmi(exit_intr_info)) {
9168 kvm_before_handle_nmi(&vmx->vcpu);
9170 kvm_after_handle_nmi(&vmx->vcpu);
9174 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
9176 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
9178 if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
9179 == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
9180 unsigned int vector;
9181 unsigned long entry;
9183 struct vcpu_vmx *vmx = to_vmx(vcpu);
9184 #ifdef CONFIG_X86_64
9188 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
9189 desc = (gate_desc *)vmx->host_idt_base + vector;
9190 entry = gate_offset(desc);
9192 #ifdef CONFIG_X86_64
9193 "mov %%" _ASM_SP ", %[sp]\n\t"
9194 "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
9199 __ASM_SIZE(push) " $%c[cs]\n\t"
9202 #ifdef CONFIG_X86_64
9207 THUNK_TARGET(entry),
9208 [ss]"i"(__KERNEL_DS),
9209 [cs]"i"(__KERNEL_CS)
9213 STACK_FRAME_NON_STANDARD(vmx_handle_external_intr);
9215 static bool vmx_has_high_real_mode_segbase(void)
9217 return enable_unrestricted_guest || emulate_invalid_guest_state;
9220 static bool vmx_mpx_supported(void)
9222 return (vmcs_config.vmexit_ctrl & VM_EXIT_CLEAR_BNDCFGS) &&
9223 (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_BNDCFGS);
9226 static bool vmx_xsaves_supported(void)
9228 return vmcs_config.cpu_based_2nd_exec_ctrl &
9229 SECONDARY_EXEC_XSAVES;
9232 static bool vmx_umip_emulated(void)
9234 return vmcs_config.cpu_based_2nd_exec_ctrl &
9235 SECONDARY_EXEC_DESC;
9238 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
9243 bool idtv_info_valid;
9245 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
9248 if (vmx->loaded_vmcs->nmi_known_unmasked)
9251 * Can't use vmx->exit_intr_info since we're not sure what
9252 * the exit reason is.
9254 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
9255 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
9256 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
9258 * SDM 3: 27.7.1.2 (September 2008)
9259 * Re-set bit "block by NMI" before VM entry if vmexit caused by
9260 * a guest IRET fault.
9261 * SDM 3: 23.2.2 (September 2008)
9262 * Bit 12 is undefined in any of the following cases:
9263 * If the VM exit sets the valid bit in the IDT-vectoring
9264 * information field.
9265 * If the VM exit is due to a double fault.
9267 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
9268 vector != DF_VECTOR && !idtv_info_valid)
9269 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
9270 GUEST_INTR_STATE_NMI);
9272 vmx->loaded_vmcs->nmi_known_unmasked =
9273 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
9274 & GUEST_INTR_STATE_NMI);
9275 } else if (unlikely(vmx->loaded_vmcs->soft_vnmi_blocked))
9276 vmx->loaded_vmcs->vnmi_blocked_time +=
9277 ktime_to_ns(ktime_sub(ktime_get(),
9278 vmx->loaded_vmcs->entry_time));
9281 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
9282 u32 idt_vectoring_info,
9283 int instr_len_field,
9284 int error_code_field)
9288 bool idtv_info_valid;
9290 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
9292 vcpu->arch.nmi_injected = false;
9293 kvm_clear_exception_queue(vcpu);
9294 kvm_clear_interrupt_queue(vcpu);
9296 if (!idtv_info_valid)
9299 kvm_make_request(KVM_REQ_EVENT, vcpu);
9301 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
9302 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
9305 case INTR_TYPE_NMI_INTR:
9306 vcpu->arch.nmi_injected = true;
9308 * SDM 3: 27.7.1.2 (September 2008)
9309 * Clear bit "block by NMI" before VM entry if a NMI
9312 vmx_set_nmi_mask(vcpu, false);
9314 case INTR_TYPE_SOFT_EXCEPTION:
9315 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
9317 case INTR_TYPE_HARD_EXCEPTION:
9318 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
9319 u32 err = vmcs_read32(error_code_field);
9320 kvm_requeue_exception_e(vcpu, vector, err);
9322 kvm_requeue_exception(vcpu, vector);
9324 case INTR_TYPE_SOFT_INTR:
9325 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
9327 case INTR_TYPE_EXT_INTR:
9328 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
9335 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
9337 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
9338 VM_EXIT_INSTRUCTION_LEN,
9339 IDT_VECTORING_ERROR_CODE);
9342 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
9344 __vmx_complete_interrupts(vcpu,
9345 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
9346 VM_ENTRY_INSTRUCTION_LEN,
9347 VM_ENTRY_EXCEPTION_ERROR_CODE);
9349 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
9352 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
9355 struct perf_guest_switch_msr *msrs;
9357 msrs = perf_guest_get_msrs(&nr_msrs);
9362 for (i = 0; i < nr_msrs; i++)
9363 if (msrs[i].host == msrs[i].guest)
9364 clear_atomic_switch_msr(vmx, msrs[i].msr);
9366 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
9370 static void vmx_arm_hv_timer(struct kvm_vcpu *vcpu)
9372 struct vcpu_vmx *vmx = to_vmx(vcpu);
9376 if (vmx->hv_deadline_tsc == -1)
9380 if (vmx->hv_deadline_tsc > tscl)
9381 /* sure to be 32 bit only because checked on set_hv_timer */
9382 delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
9383 cpu_preemption_timer_multi);
9387 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
9390 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
9392 struct vcpu_vmx *vmx = to_vmx(vcpu);
9393 unsigned long cr3, cr4;
9395 /* Record the guest's net vcpu time for enforced NMI injections. */
9396 if (unlikely(!enable_vnmi &&
9397 vmx->loaded_vmcs->soft_vnmi_blocked))
9398 vmx->loaded_vmcs->entry_time = ktime_get();
9400 /* Don't enter VMX if guest state is invalid, let the exit handler
9401 start emulation until we arrive back to a valid state */
9402 if (vmx->emulation_required)
9405 if (vmx->ple_window_dirty) {
9406 vmx->ple_window_dirty = false;
9407 vmcs_write32(PLE_WINDOW, vmx->ple_window);
9410 if (vmx->nested.sync_shadow_vmcs) {
9411 copy_vmcs12_to_shadow(vmx);
9412 vmx->nested.sync_shadow_vmcs = false;
9415 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
9416 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
9417 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
9418 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
9420 cr3 = __get_current_cr3_fast();
9421 if (unlikely(cr3 != vmx->loaded_vmcs->vmcs_host_cr3)) {
9422 vmcs_writel(HOST_CR3, cr3);
9423 vmx->loaded_vmcs->vmcs_host_cr3 = cr3;
9426 cr4 = cr4_read_shadow();
9427 if (unlikely(cr4 != vmx->loaded_vmcs->vmcs_host_cr4)) {
9428 vmcs_writel(HOST_CR4, cr4);
9429 vmx->loaded_vmcs->vmcs_host_cr4 = cr4;
9432 /* When single-stepping over STI and MOV SS, we must clear the
9433 * corresponding interruptibility bits in the guest state. Otherwise
9434 * vmentry fails as it then expects bit 14 (BS) in pending debug
9435 * exceptions being set, but that's not correct for the guest debugging
9437 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9438 vmx_set_interrupt_shadow(vcpu, 0);
9440 if (static_cpu_has(X86_FEATURE_PKU) &&
9441 kvm_read_cr4_bits(vcpu, X86_CR4_PKE) &&
9442 vcpu->arch.pkru != vmx->host_pkru)
9443 __write_pkru(vcpu->arch.pkru);
9445 atomic_switch_perf_msrs(vmx);
9447 vmx_arm_hv_timer(vcpu);
9450 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
9451 * it's non-zero. Since vmentry is serialising on affected CPUs, there
9452 * is no need to worry about the conditional branch over the wrmsr
9453 * being speculatively taken.
9456 native_wrmsrl(MSR_IA32_SPEC_CTRL, vmx->spec_ctrl);
9458 vmx->__launched = vmx->loaded_vmcs->launched;
9460 /* Store host registers */
9461 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
9462 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
9463 "push %%" _ASM_CX " \n\t"
9464 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
9466 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
9467 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
9469 /* Reload cr2 if changed */
9470 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
9471 "mov %%cr2, %%" _ASM_DX " \n\t"
9472 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
9474 "mov %%" _ASM_AX", %%cr2 \n\t"
9476 /* Check if vmlaunch of vmresume is needed */
9477 "cmpl $0, %c[launched](%0) \n\t"
9478 /* Load guest registers. Don't clobber flags. */
9479 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
9480 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
9481 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
9482 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
9483 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
9484 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
9485 #ifdef CONFIG_X86_64
9486 "mov %c[r8](%0), %%r8 \n\t"
9487 "mov %c[r9](%0), %%r9 \n\t"
9488 "mov %c[r10](%0), %%r10 \n\t"
9489 "mov %c[r11](%0), %%r11 \n\t"
9490 "mov %c[r12](%0), %%r12 \n\t"
9491 "mov %c[r13](%0), %%r13 \n\t"
9492 "mov %c[r14](%0), %%r14 \n\t"
9493 "mov %c[r15](%0), %%r15 \n\t"
9495 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
9497 /* Enter guest mode */
9499 __ex(ASM_VMX_VMLAUNCH) "\n\t"
9501 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
9503 /* Save guest registers, load host registers, keep flags */
9504 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
9506 "setbe %c[fail](%0)\n\t"
9507 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
9508 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
9509 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
9510 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
9511 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
9512 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
9513 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
9514 #ifdef CONFIG_X86_64
9515 "mov %%r8, %c[r8](%0) \n\t"
9516 "mov %%r9, %c[r9](%0) \n\t"
9517 "mov %%r10, %c[r10](%0) \n\t"
9518 "mov %%r11, %c[r11](%0) \n\t"
9519 "mov %%r12, %c[r12](%0) \n\t"
9520 "mov %%r13, %c[r13](%0) \n\t"
9521 "mov %%r14, %c[r14](%0) \n\t"
9522 "mov %%r15, %c[r15](%0) \n\t"
9523 "xor %%r8d, %%r8d \n\t"
9524 "xor %%r9d, %%r9d \n\t"
9525 "xor %%r10d, %%r10d \n\t"
9526 "xor %%r11d, %%r11d \n\t"
9527 "xor %%r12d, %%r12d \n\t"
9528 "xor %%r13d, %%r13d \n\t"
9529 "xor %%r14d, %%r14d \n\t"
9530 "xor %%r15d, %%r15d \n\t"
9532 "mov %%cr2, %%" _ASM_AX " \n\t"
9533 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
9535 "xor %%eax, %%eax \n\t"
9536 "xor %%ebx, %%ebx \n\t"
9537 "xor %%esi, %%esi \n\t"
9538 "xor %%edi, %%edi \n\t"
9539 "pop %%" _ASM_BP "; pop %%" _ASM_DX " \n\t"
9540 ".pushsection .rodata \n\t"
9541 ".global vmx_return \n\t"
9542 "vmx_return: " _ASM_PTR " 2b \n\t"
9544 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
9545 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
9546 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
9547 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
9548 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
9549 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
9550 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
9551 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
9552 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
9553 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
9554 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
9555 #ifdef CONFIG_X86_64
9556 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
9557 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
9558 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
9559 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
9560 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
9561 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
9562 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
9563 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
9565 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
9566 [wordsize]"i"(sizeof(ulong))
9568 #ifdef CONFIG_X86_64
9569 , "rax", "rbx", "rdi", "rsi"
9570 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
9572 , "eax", "ebx", "edi", "esi"
9577 * We do not use IBRS in the kernel. If this vCPU has used the
9578 * SPEC_CTRL MSR it may have left it on; save the value and
9579 * turn it off. This is much more efficient than blindly adding
9580 * it to the atomic save/restore list. Especially as the former
9581 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
9583 * For non-nested case:
9584 * If the L01 MSR bitmap does not intercept the MSR, then we need to
9588 * If the L02 MSR bitmap does not intercept the MSR, then we need to
9591 if (!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL))
9592 vmx->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
9595 native_wrmsrl(MSR_IA32_SPEC_CTRL, 0);
9597 /* Eliminate branch target predictions from guest mode */
9600 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
9601 if (vmx->host_debugctlmsr)
9602 update_debugctlmsr(vmx->host_debugctlmsr);
9604 #ifndef CONFIG_X86_64
9606 * The sysexit path does not restore ds/es, so we must set them to
9607 * a reasonable value ourselves.
9609 * We can't defer this to vmx_load_host_state() since that function
9610 * may be executed in interrupt context, which saves and restore segments
9611 * around it, nullifying its effect.
9613 loadsegment(ds, __USER_DS);
9614 loadsegment(es, __USER_DS);
9617 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
9618 | (1 << VCPU_EXREG_RFLAGS)
9619 | (1 << VCPU_EXREG_PDPTR)
9620 | (1 << VCPU_EXREG_SEGMENTS)
9621 | (1 << VCPU_EXREG_CR3));
9622 vcpu->arch.regs_dirty = 0;
9625 * eager fpu is enabled if PKEY is supported and CR4 is switched
9626 * back on host, so it is safe to read guest PKRU from current
9629 if (static_cpu_has(X86_FEATURE_PKU) &&
9630 kvm_read_cr4_bits(vcpu, X86_CR4_PKE)) {
9631 vcpu->arch.pkru = __read_pkru();
9632 if (vcpu->arch.pkru != vmx->host_pkru)
9633 __write_pkru(vmx->host_pkru);
9637 * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
9638 * we did not inject a still-pending event to L1 now because of
9639 * nested_run_pending, we need to re-enable this bit.
9641 if (vmx->nested.nested_run_pending)
9642 kvm_make_request(KVM_REQ_EVENT, vcpu);
9644 vmx->nested.nested_run_pending = 0;
9645 vmx->idt_vectoring_info = 0;
9647 vmx->exit_reason = vmx->fail ? 0xdead : vmcs_read32(VM_EXIT_REASON);
9648 if (vmx->fail || (vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
9651 vmx->loaded_vmcs->launched = 1;
9652 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
9654 vmx_complete_atomic_exit(vmx);
9655 vmx_recover_nmi_blocking(vmx);
9656 vmx_complete_interrupts(vmx);
9658 STACK_FRAME_NON_STANDARD(vmx_vcpu_run);
9660 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
9662 struct vcpu_vmx *vmx = to_vmx(vcpu);
9665 if (vmx->loaded_vmcs == vmcs)
9669 vmx->loaded_vmcs = vmcs;
9671 vmx_vcpu_load(vcpu, cpu);
9676 * Ensure that the current vmcs of the logical processor is the
9677 * vmcs01 of the vcpu before calling free_nested().
9679 static void vmx_free_vcpu_nested(struct kvm_vcpu *vcpu)
9681 struct vcpu_vmx *vmx = to_vmx(vcpu);
9684 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
9689 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
9691 struct vcpu_vmx *vmx = to_vmx(vcpu);
9694 vmx_destroy_pml_buffer(vmx);
9695 free_vpid(vmx->vpid);
9696 leave_guest_mode(vcpu);
9697 vmx_free_vcpu_nested(vcpu);
9698 free_loaded_vmcs(vmx->loaded_vmcs);
9699 kfree(vmx->guest_msrs);
9700 kvm_vcpu_uninit(vcpu);
9701 kmem_cache_free(kvm_vcpu_cache, vmx);
9704 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
9707 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
9708 unsigned long *msr_bitmap;
9712 return ERR_PTR(-ENOMEM);
9714 vmx->vpid = allocate_vpid();
9716 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
9723 * If PML is turned on, failure on enabling PML just results in failure
9724 * of creating the vcpu, therefore we can simplify PML logic (by
9725 * avoiding dealing with cases, such as enabling PML partially on vcpus
9726 * for the guest, etc.
9729 vmx->pml_pg = alloc_page(GFP_KERNEL | __GFP_ZERO);
9734 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
9735 BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
9738 if (!vmx->guest_msrs)
9741 err = alloc_loaded_vmcs(&vmx->vmcs01);
9745 msr_bitmap = vmx->vmcs01.msr_bitmap;
9746 vmx_disable_intercept_for_msr(msr_bitmap, MSR_FS_BASE, MSR_TYPE_RW);
9747 vmx_disable_intercept_for_msr(msr_bitmap, MSR_GS_BASE, MSR_TYPE_RW);
9748 vmx_disable_intercept_for_msr(msr_bitmap, MSR_KERNEL_GS_BASE, MSR_TYPE_RW);
9749 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_CS, MSR_TYPE_RW);
9750 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_ESP, MSR_TYPE_RW);
9751 vmx_disable_intercept_for_msr(msr_bitmap, MSR_IA32_SYSENTER_EIP, MSR_TYPE_RW);
9752 vmx->msr_bitmap_mode = 0;
9754 vmx->loaded_vmcs = &vmx->vmcs01;
9756 vmx_vcpu_load(&vmx->vcpu, cpu);
9757 vmx->vcpu.cpu = cpu;
9758 vmx_vcpu_setup(vmx);
9759 vmx_vcpu_put(&vmx->vcpu);
9761 if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9762 err = alloc_apic_access_page(kvm);
9768 err = init_rmode_identity_map(kvm);
9774 nested_vmx_setup_ctls_msrs(vmx);
9775 vmx->nested.vpid02 = allocate_vpid();
9778 vmx->nested.posted_intr_nv = -1;
9779 vmx->nested.current_vmptr = -1ull;
9781 vmx->msr_ia32_feature_control_valid_bits = FEATURE_CONTROL_LOCKED;
9784 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
9785 * or POSTED_INTR_WAKEUP_VECTOR.
9787 vmx->pi_desc.nv = POSTED_INTR_VECTOR;
9788 vmx->pi_desc.sn = 1;
9793 free_vpid(vmx->nested.vpid02);
9794 free_loaded_vmcs(vmx->loaded_vmcs);
9796 kfree(vmx->guest_msrs);
9798 vmx_destroy_pml_buffer(vmx);
9800 kvm_vcpu_uninit(&vmx->vcpu);
9802 free_vpid(vmx->vpid);
9803 kmem_cache_free(kvm_vcpu_cache, vmx);
9804 return ERR_PTR(err);
9807 static void __init vmx_check_processor_compat(void *rtn)
9809 struct vmcs_config vmcs_conf;
9812 if (setup_vmcs_config(&vmcs_conf) < 0)
9814 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
9815 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
9816 smp_processor_id());
9821 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
9826 /* For VT-d and EPT combination
9827 * 1. MMIO: always map as UC
9829 * a. VT-d without snooping control feature: can't guarantee the
9830 * result, try to trust guest.
9831 * b. VT-d with snooping control feature: snooping control feature of
9832 * VT-d engine can guarantee the cache correctness. Just set it
9833 * to WB to keep consistent with host. So the same as item 3.
9834 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
9835 * consistent with host MTRR
9838 cache = MTRR_TYPE_UNCACHABLE;
9842 if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
9843 ipat = VMX_EPT_IPAT_BIT;
9844 cache = MTRR_TYPE_WRBACK;
9848 if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
9849 ipat = VMX_EPT_IPAT_BIT;
9850 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
9851 cache = MTRR_TYPE_WRBACK;
9853 cache = MTRR_TYPE_UNCACHABLE;
9857 cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
9860 return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
9863 static int vmx_get_lpage_level(void)
9865 if (enable_ept && !cpu_has_vmx_ept_1g_page())
9866 return PT_DIRECTORY_LEVEL;
9868 /* For shadow and EPT supported 1GB page */
9869 return PT_PDPE_LEVEL;
9872 static void vmcs_set_secondary_exec_control(u32 new_ctl)
9875 * These bits in the secondary execution controls field
9876 * are dynamic, the others are mostly based on the hypervisor
9877 * architecture and the guest's CPUID. Do not touch the
9881 SECONDARY_EXEC_SHADOW_VMCS |
9882 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
9883 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
9884 SECONDARY_EXEC_DESC;
9886 u32 cur_ctl = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
9888 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
9889 (new_ctl & ~mask) | (cur_ctl & mask));
9893 * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
9894 * (indicating "allowed-1") if they are supported in the guest's CPUID.
9896 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
9898 struct vcpu_vmx *vmx = to_vmx(vcpu);
9899 struct kvm_cpuid_entry2 *entry;
9901 vmx->nested.nested_vmx_cr0_fixed1 = 0xffffffff;
9902 vmx->nested.nested_vmx_cr4_fixed1 = X86_CR4_PCE;
9904 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do { \
9905 if (entry && (entry->_reg & (_cpuid_mask))) \
9906 vmx->nested.nested_vmx_cr4_fixed1 |= (_cr4_mask); \
9909 entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
9910 cr4_fixed1_update(X86_CR4_VME, edx, bit(X86_FEATURE_VME));
9911 cr4_fixed1_update(X86_CR4_PVI, edx, bit(X86_FEATURE_VME));
9912 cr4_fixed1_update(X86_CR4_TSD, edx, bit(X86_FEATURE_TSC));
9913 cr4_fixed1_update(X86_CR4_DE, edx, bit(X86_FEATURE_DE));
9914 cr4_fixed1_update(X86_CR4_PSE, edx, bit(X86_FEATURE_PSE));
9915 cr4_fixed1_update(X86_CR4_PAE, edx, bit(X86_FEATURE_PAE));
9916 cr4_fixed1_update(X86_CR4_MCE, edx, bit(X86_FEATURE_MCE));
9917 cr4_fixed1_update(X86_CR4_PGE, edx, bit(X86_FEATURE_PGE));
9918 cr4_fixed1_update(X86_CR4_OSFXSR, edx, bit(X86_FEATURE_FXSR));
9919 cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, bit(X86_FEATURE_XMM));
9920 cr4_fixed1_update(X86_CR4_VMXE, ecx, bit(X86_FEATURE_VMX));
9921 cr4_fixed1_update(X86_CR4_SMXE, ecx, bit(X86_FEATURE_SMX));
9922 cr4_fixed1_update(X86_CR4_PCIDE, ecx, bit(X86_FEATURE_PCID));
9923 cr4_fixed1_update(X86_CR4_OSXSAVE, ecx, bit(X86_FEATURE_XSAVE));
9925 entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
9926 cr4_fixed1_update(X86_CR4_FSGSBASE, ebx, bit(X86_FEATURE_FSGSBASE));
9927 cr4_fixed1_update(X86_CR4_SMEP, ebx, bit(X86_FEATURE_SMEP));
9928 cr4_fixed1_update(X86_CR4_SMAP, ebx, bit(X86_FEATURE_SMAP));
9929 cr4_fixed1_update(X86_CR4_PKE, ecx, bit(X86_FEATURE_PKU));
9930 cr4_fixed1_update(X86_CR4_UMIP, ecx, bit(X86_FEATURE_UMIP));
9932 #undef cr4_fixed1_update
9935 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
9937 struct vcpu_vmx *vmx = to_vmx(vcpu);
9939 if (cpu_has_secondary_exec_ctrls()) {
9940 vmx_compute_secondary_exec_control(vmx);
9941 vmcs_set_secondary_exec_control(vmx->secondary_exec_control);
9944 if (nested_vmx_allowed(vcpu))
9945 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
9946 FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
9948 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
9949 ~FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
9951 if (nested_vmx_allowed(vcpu))
9952 nested_vmx_cr_fixed1_bits_update(vcpu);
9955 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
9957 if (func == 1 && nested)
9958 entry->ecx |= bit(X86_FEATURE_VMX);
9961 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
9962 struct x86_exception *fault)
9964 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9965 struct vcpu_vmx *vmx = to_vmx(vcpu);
9967 unsigned long exit_qualification = vcpu->arch.exit_qualification;
9969 if (vmx->nested.pml_full) {
9970 exit_reason = EXIT_REASON_PML_FULL;
9971 vmx->nested.pml_full = false;
9972 exit_qualification &= INTR_INFO_UNBLOCK_NMI;
9973 } else if (fault->error_code & PFERR_RSVD_MASK)
9974 exit_reason = EXIT_REASON_EPT_MISCONFIG;
9976 exit_reason = EXIT_REASON_EPT_VIOLATION;
9978 nested_vmx_vmexit(vcpu, exit_reason, 0, exit_qualification);
9979 vmcs12->guest_physical_address = fault->address;
9982 static bool nested_ept_ad_enabled(struct kvm_vcpu *vcpu)
9984 return nested_ept_get_cr3(vcpu) & VMX_EPTP_AD_ENABLE_BIT;
9987 /* Callbacks for nested_ept_init_mmu_context: */
9989 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
9991 /* return the page table to be shadowed - in our case, EPT12 */
9992 return get_vmcs12(vcpu)->ept_pointer;
9995 static int nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
9997 WARN_ON(mmu_is_nested(vcpu));
9998 if (!valid_ept_address(vcpu, nested_ept_get_cr3(vcpu)))
10001 kvm_mmu_unload(vcpu);
10002 kvm_init_shadow_ept_mmu(vcpu,
10003 to_vmx(vcpu)->nested.nested_vmx_ept_caps &
10004 VMX_EPT_EXECUTE_ONLY_BIT,
10005 nested_ept_ad_enabled(vcpu));
10006 vcpu->arch.mmu.set_cr3 = vmx_set_cr3;
10007 vcpu->arch.mmu.get_cr3 = nested_ept_get_cr3;
10008 vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
10010 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
10014 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
10016 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
10019 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
10022 bool inequality, bit;
10024 bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
10026 (error_code & vmcs12->page_fault_error_code_mask) !=
10027 vmcs12->page_fault_error_code_match;
10028 return inequality ^ bit;
10031 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
10032 struct x86_exception *fault)
10034 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
10036 WARN_ON(!is_guest_mode(vcpu));
10038 if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code) &&
10039 !to_vmx(vcpu)->nested.nested_run_pending) {
10040 vmcs12->vm_exit_intr_error_code = fault->error_code;
10041 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
10042 PF_VECTOR | INTR_TYPE_HARD_EXCEPTION |
10043 INTR_INFO_DELIVER_CODE_MASK | INTR_INFO_VALID_MASK,
10046 kvm_inject_page_fault(vcpu, fault);
10050 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
10051 struct vmcs12 *vmcs12);
10053 static void nested_get_vmcs12_pages(struct kvm_vcpu *vcpu,
10054 struct vmcs12 *vmcs12)
10056 struct vcpu_vmx *vmx = to_vmx(vcpu);
10060 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
10062 * Translate L1 physical address to host physical
10063 * address for vmcs02. Keep the page pinned, so this
10064 * physical address remains valid. We keep a reference
10065 * to it so we can release it later.
10067 if (vmx->nested.apic_access_page) { /* shouldn't happen */
10068 kvm_release_page_dirty(vmx->nested.apic_access_page);
10069 vmx->nested.apic_access_page = NULL;
10071 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->apic_access_addr);
10073 * If translation failed, no matter: This feature asks
10074 * to exit when accessing the given address, and if it
10075 * can never be accessed, this feature won't do
10078 if (!is_error_page(page)) {
10079 vmx->nested.apic_access_page = page;
10080 hpa = page_to_phys(vmx->nested.apic_access_page);
10081 vmcs_write64(APIC_ACCESS_ADDR, hpa);
10083 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
10084 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
10086 } else if (!(nested_cpu_has_virt_x2apic_mode(vmcs12)) &&
10087 cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
10088 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
10089 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
10090 kvm_vcpu_reload_apic_access_page(vcpu);
10093 if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
10094 if (vmx->nested.virtual_apic_page) { /* shouldn't happen */
10095 kvm_release_page_dirty(vmx->nested.virtual_apic_page);
10096 vmx->nested.virtual_apic_page = NULL;
10098 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->virtual_apic_page_addr);
10101 * If translation failed, VM entry will fail because
10102 * prepare_vmcs02 set VIRTUAL_APIC_PAGE_ADDR to -1ull.
10103 * Failing the vm entry is _not_ what the processor
10104 * does but it's basically the only possibility we
10105 * have. We could still enter the guest if CR8 load
10106 * exits are enabled, CR8 store exits are enabled, and
10107 * virtualize APIC access is disabled; in this case
10108 * the processor would never use the TPR shadow and we
10109 * could simply clear the bit from the execution
10110 * control. But such a configuration is useless, so
10111 * let's keep the code simple.
10113 if (!is_error_page(page)) {
10114 vmx->nested.virtual_apic_page = page;
10115 hpa = page_to_phys(vmx->nested.virtual_apic_page);
10116 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, hpa);
10120 if (nested_cpu_has_posted_intr(vmcs12)) {
10121 if (vmx->nested.pi_desc_page) { /* shouldn't happen */
10122 kunmap(vmx->nested.pi_desc_page);
10123 kvm_release_page_dirty(vmx->nested.pi_desc_page);
10124 vmx->nested.pi_desc_page = NULL;
10126 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->posted_intr_desc_addr);
10127 if (is_error_page(page))
10129 vmx->nested.pi_desc_page = page;
10130 vmx->nested.pi_desc = kmap(vmx->nested.pi_desc_page);
10131 vmx->nested.pi_desc =
10132 (struct pi_desc *)((void *)vmx->nested.pi_desc +
10133 (unsigned long)(vmcs12->posted_intr_desc_addr &
10135 vmcs_write64(POSTED_INTR_DESC_ADDR,
10136 page_to_phys(vmx->nested.pi_desc_page) +
10137 (unsigned long)(vmcs12->posted_intr_desc_addr &
10140 if (nested_vmx_prepare_msr_bitmap(vcpu, vmcs12))
10141 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
10142 CPU_BASED_USE_MSR_BITMAPS);
10144 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
10145 CPU_BASED_USE_MSR_BITMAPS);
10148 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
10150 u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
10151 struct vcpu_vmx *vmx = to_vmx(vcpu);
10153 if (vcpu->arch.virtual_tsc_khz == 0)
10156 /* Make sure short timeouts reliably trigger an immediate vmexit.
10157 * hrtimer_start does not guarantee this. */
10158 if (preemption_timeout <= 1) {
10159 vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
10163 preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
10164 preemption_timeout *= 1000000;
10165 do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
10166 hrtimer_start(&vmx->nested.preemption_timer,
10167 ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
10170 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu *vcpu,
10171 struct vmcs12 *vmcs12)
10173 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
10176 if (!page_address_valid(vcpu, vmcs12->io_bitmap_a) ||
10177 !page_address_valid(vcpu, vmcs12->io_bitmap_b))
10183 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
10184 struct vmcs12 *vmcs12)
10186 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
10189 if (!page_address_valid(vcpu, vmcs12->msr_bitmap))
10195 static int nested_vmx_check_tpr_shadow_controls(struct kvm_vcpu *vcpu,
10196 struct vmcs12 *vmcs12)
10198 if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
10201 if (!page_address_valid(vcpu, vmcs12->virtual_apic_page_addr))
10208 * Merge L0's and L1's MSR bitmap, return false to indicate that
10209 * we do not use the hardware.
10211 static inline bool nested_vmx_prepare_msr_bitmap(struct kvm_vcpu *vcpu,
10212 struct vmcs12 *vmcs12)
10216 unsigned long *msr_bitmap_l1;
10217 unsigned long *msr_bitmap_l0 = to_vmx(vcpu)->nested.vmcs02.msr_bitmap;
10219 * pred_cmd & spec_ctrl are trying to verify two things:
10221 * 1. L0 gave a permission to L1 to actually passthrough the MSR. This
10222 * ensures that we do not accidentally generate an L02 MSR bitmap
10223 * from the L12 MSR bitmap that is too permissive.
10224 * 2. That L1 or L2s have actually used the MSR. This avoids
10225 * unnecessarily merging of the bitmap if the MSR is unused. This
10226 * works properly because we only update the L01 MSR bitmap lazily.
10227 * So even if L0 should pass L1 these MSRs, the L01 bitmap is only
10228 * updated to reflect this when L1 (or its L2s) actually write to
10231 bool pred_cmd = !msr_write_intercepted_l01(vcpu, MSR_IA32_PRED_CMD);
10232 bool spec_ctrl = !msr_write_intercepted_l01(vcpu, MSR_IA32_SPEC_CTRL);
10234 /* Nothing to do if the MSR bitmap is not in use. */
10235 if (!cpu_has_vmx_msr_bitmap() ||
10236 !nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
10239 if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
10240 !pred_cmd && !spec_ctrl)
10243 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->msr_bitmap);
10244 if (is_error_page(page))
10247 msr_bitmap_l1 = (unsigned long *)kmap(page);
10248 if (nested_cpu_has_apic_reg_virt(vmcs12)) {
10250 * L0 need not intercept reads for MSRs between 0x800 and 0x8ff, it
10251 * just lets the processor take the value from the virtual-APIC page;
10252 * take those 256 bits directly from the L1 bitmap.
10254 for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
10255 unsigned word = msr / BITS_PER_LONG;
10256 msr_bitmap_l0[word] = msr_bitmap_l1[word];
10257 msr_bitmap_l0[word + (0x800 / sizeof(long))] = ~0;
10260 for (msr = 0x800; msr <= 0x8ff; msr += BITS_PER_LONG) {
10261 unsigned word = msr / BITS_PER_LONG;
10262 msr_bitmap_l0[word] = ~0;
10263 msr_bitmap_l0[word + (0x800 / sizeof(long))] = ~0;
10267 nested_vmx_disable_intercept_for_msr(
10268 msr_bitmap_l1, msr_bitmap_l0,
10269 X2APIC_MSR(APIC_TASKPRI),
10272 if (nested_cpu_has_vid(vmcs12)) {
10273 nested_vmx_disable_intercept_for_msr(
10274 msr_bitmap_l1, msr_bitmap_l0,
10275 X2APIC_MSR(APIC_EOI),
10277 nested_vmx_disable_intercept_for_msr(
10278 msr_bitmap_l1, msr_bitmap_l0,
10279 X2APIC_MSR(APIC_SELF_IPI),
10284 nested_vmx_disable_intercept_for_msr(
10285 msr_bitmap_l1, msr_bitmap_l0,
10286 MSR_IA32_SPEC_CTRL,
10287 MSR_TYPE_R | MSR_TYPE_W);
10290 nested_vmx_disable_intercept_for_msr(
10291 msr_bitmap_l1, msr_bitmap_l0,
10296 kvm_release_page_clean(page);
10301 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
10302 struct vmcs12 *vmcs12)
10304 if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
10305 !nested_cpu_has_apic_reg_virt(vmcs12) &&
10306 !nested_cpu_has_vid(vmcs12) &&
10307 !nested_cpu_has_posted_intr(vmcs12))
10311 * If virtualize x2apic mode is enabled,
10312 * virtualize apic access must be disabled.
10314 if (nested_cpu_has_virt_x2apic_mode(vmcs12) &&
10315 nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
10319 * If virtual interrupt delivery is enabled,
10320 * we must exit on external interrupts.
10322 if (nested_cpu_has_vid(vmcs12) &&
10323 !nested_exit_on_intr(vcpu))
10327 * bits 15:8 should be zero in posted_intr_nv,
10328 * the descriptor address has been already checked
10329 * in nested_get_vmcs12_pages.
10331 if (nested_cpu_has_posted_intr(vmcs12) &&
10332 (!nested_cpu_has_vid(vmcs12) ||
10333 !nested_exit_intr_ack_set(vcpu) ||
10334 vmcs12->posted_intr_nv & 0xff00))
10337 /* tpr shadow is needed by all apicv features. */
10338 if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
10344 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
10345 unsigned long count_field,
10346 unsigned long addr_field)
10351 if (vmcs12_read_any(vcpu, count_field, &count) ||
10352 vmcs12_read_any(vcpu, addr_field, &addr)) {
10358 maxphyaddr = cpuid_maxphyaddr(vcpu);
10359 if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
10360 (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr) {
10361 pr_debug_ratelimited(
10362 "nVMX: invalid MSR switch (0x%lx, %d, %llu, 0x%08llx)",
10363 addr_field, maxphyaddr, count, addr);
10369 static int nested_vmx_check_msr_switch_controls(struct kvm_vcpu *vcpu,
10370 struct vmcs12 *vmcs12)
10372 if (vmcs12->vm_exit_msr_load_count == 0 &&
10373 vmcs12->vm_exit_msr_store_count == 0 &&
10374 vmcs12->vm_entry_msr_load_count == 0)
10375 return 0; /* Fast path */
10376 if (nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_LOAD_COUNT,
10377 VM_EXIT_MSR_LOAD_ADDR) ||
10378 nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_STORE_COUNT,
10379 VM_EXIT_MSR_STORE_ADDR) ||
10380 nested_vmx_check_msr_switch(vcpu, VM_ENTRY_MSR_LOAD_COUNT,
10381 VM_ENTRY_MSR_LOAD_ADDR))
10386 static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
10387 struct vmcs12 *vmcs12)
10389 u64 address = vmcs12->pml_address;
10390 int maxphyaddr = cpuid_maxphyaddr(vcpu);
10392 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_PML)) {
10393 if (!nested_cpu_has_ept(vmcs12) ||
10394 !IS_ALIGNED(address, 4096) ||
10395 address >> maxphyaddr)
10402 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
10403 struct vmx_msr_entry *e)
10405 /* x2APIC MSR accesses are not allowed */
10406 if (vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8)
10408 if (e->index == MSR_IA32_UCODE_WRITE || /* SDM Table 35-2 */
10409 e->index == MSR_IA32_UCODE_REV)
10411 if (e->reserved != 0)
10416 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
10417 struct vmx_msr_entry *e)
10419 if (e->index == MSR_FS_BASE ||
10420 e->index == MSR_GS_BASE ||
10421 e->index == MSR_IA32_SMM_MONITOR_CTL || /* SMM is not supported */
10422 nested_vmx_msr_check_common(vcpu, e))
10427 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
10428 struct vmx_msr_entry *e)
10430 if (e->index == MSR_IA32_SMBASE || /* SMM is not supported */
10431 nested_vmx_msr_check_common(vcpu, e))
10437 * Load guest's/host's msr at nested entry/exit.
10438 * return 0 for success, entry index for failure.
10440 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
10443 struct vmx_msr_entry e;
10444 struct msr_data msr;
10446 msr.host_initiated = false;
10447 for (i = 0; i < count; i++) {
10448 if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
10450 pr_debug_ratelimited(
10451 "%s cannot read MSR entry (%u, 0x%08llx)\n",
10452 __func__, i, gpa + i * sizeof(e));
10455 if (nested_vmx_load_msr_check(vcpu, &e)) {
10456 pr_debug_ratelimited(
10457 "%s check failed (%u, 0x%x, 0x%x)\n",
10458 __func__, i, e.index, e.reserved);
10461 msr.index = e.index;
10462 msr.data = e.value;
10463 if (kvm_set_msr(vcpu, &msr)) {
10464 pr_debug_ratelimited(
10465 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
10466 __func__, i, e.index, e.value);
10475 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
10478 struct vmx_msr_entry e;
10480 for (i = 0; i < count; i++) {
10481 struct msr_data msr_info;
10482 if (kvm_vcpu_read_guest(vcpu,
10483 gpa + i * sizeof(e),
10484 &e, 2 * sizeof(u32))) {
10485 pr_debug_ratelimited(
10486 "%s cannot read MSR entry (%u, 0x%08llx)\n",
10487 __func__, i, gpa + i * sizeof(e));
10490 if (nested_vmx_store_msr_check(vcpu, &e)) {
10491 pr_debug_ratelimited(
10492 "%s check failed (%u, 0x%x, 0x%x)\n",
10493 __func__, i, e.index, e.reserved);
10496 msr_info.host_initiated = false;
10497 msr_info.index = e.index;
10498 if (kvm_get_msr(vcpu, &msr_info)) {
10499 pr_debug_ratelimited(
10500 "%s cannot read MSR (%u, 0x%x)\n",
10501 __func__, i, e.index);
10504 if (kvm_vcpu_write_guest(vcpu,
10505 gpa + i * sizeof(e) +
10506 offsetof(struct vmx_msr_entry, value),
10507 &msr_info.data, sizeof(msr_info.data))) {
10508 pr_debug_ratelimited(
10509 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
10510 __func__, i, e.index, msr_info.data);
10517 static bool nested_cr3_valid(struct kvm_vcpu *vcpu, unsigned long val)
10519 unsigned long invalid_mask;
10521 invalid_mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
10522 return (val & invalid_mask) == 0;
10526 * Load guest's/host's cr3 at nested entry/exit. nested_ept is true if we are
10527 * emulating VM entry into a guest with EPT enabled.
10528 * Returns 0 on success, 1 on failure. Invalid state exit qualification code
10529 * is assigned to entry_failure_code on failure.
10531 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3, bool nested_ept,
10532 u32 *entry_failure_code)
10534 if (cr3 != kvm_read_cr3(vcpu) || (!nested_ept && pdptrs_changed(vcpu))) {
10535 if (!nested_cr3_valid(vcpu, cr3)) {
10536 *entry_failure_code = ENTRY_FAIL_DEFAULT;
10541 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
10542 * must not be dereferenced.
10544 if (!is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu) &&
10546 if (!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)) {
10547 *entry_failure_code = ENTRY_FAIL_PDPTE;
10552 vcpu->arch.cr3 = cr3;
10553 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
10556 kvm_mmu_reset_context(vcpu);
10560 static void prepare_vmcs02_full(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10563 struct vcpu_vmx *vmx = to_vmx(vcpu);
10565 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
10566 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
10567 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
10568 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
10569 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
10570 vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
10571 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
10572 vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
10573 vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
10574 vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
10575 vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
10576 vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
10577 vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
10578 vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
10579 vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
10580 vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
10581 vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
10582 vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
10583 vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
10584 vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
10585 vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
10586 vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
10587 vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
10588 vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
10589 vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
10590 vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
10591 vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
10592 vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
10593 vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
10594 vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
10595 vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
10597 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
10598 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
10599 vmcs12->guest_pending_dbg_exceptions);
10600 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
10601 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
10603 if (nested_cpu_has_xsaves(vmcs12))
10604 vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
10605 vmcs_write64(VMCS_LINK_POINTER, -1ull);
10607 if (cpu_has_vmx_posted_intr())
10608 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_NESTED_VECTOR);
10611 * Whether page-faults are trapped is determined by a combination of
10612 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
10613 * If enable_ept, L0 doesn't care about page faults and we should
10614 * set all of these to L1's desires. However, if !enable_ept, L0 does
10615 * care about (at least some) page faults, and because it is not easy
10616 * (if at all possible?) to merge L0 and L1's desires, we simply ask
10617 * to exit on each and every L2 page fault. This is done by setting
10618 * MASK=MATCH=0 and (see below) EB.PF=1.
10619 * Note that below we don't need special code to set EB.PF beyond the
10620 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
10621 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
10622 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
10624 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
10625 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
10626 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
10627 enable_ept ? vmcs12->page_fault_error_code_match : 0);
10629 /* All VMFUNCs are currently emulated through L0 vmexits. */
10630 if (cpu_has_vmx_vmfunc())
10631 vmcs_write64(VM_FUNCTION_CONTROL, 0);
10633 if (cpu_has_vmx_apicv()) {
10634 vmcs_write64(EOI_EXIT_BITMAP0, vmcs12->eoi_exit_bitmap0);
10635 vmcs_write64(EOI_EXIT_BITMAP1, vmcs12->eoi_exit_bitmap1);
10636 vmcs_write64(EOI_EXIT_BITMAP2, vmcs12->eoi_exit_bitmap2);
10637 vmcs_write64(EOI_EXIT_BITMAP3, vmcs12->eoi_exit_bitmap3);
10641 * Set host-state according to L0's settings (vmcs12 is irrelevant here)
10642 * Some constant fields are set here by vmx_set_constant_host_state().
10643 * Other fields are different per CPU, and will be set later when
10644 * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
10646 vmx_set_constant_host_state(vmx);
10649 * Set the MSR load/store lists to match L0's settings.
10651 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
10652 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
10653 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
10654 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
10655 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
10657 set_cr4_guest_host_mask(vmx);
10659 if (vmx_mpx_supported())
10660 vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
10663 if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02)
10664 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
10666 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
10670 * L1 may access the L2's PDPTR, so save them to construct vmcs12
10673 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
10674 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
10675 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
10676 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
10679 if (cpu_has_vmx_msr_bitmap())
10680 vmcs_write64(MSR_BITMAP, __pa(vmx->nested.vmcs02.msr_bitmap));
10684 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
10685 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
10686 * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
10687 * guest in a way that will both be appropriate to L1's requests, and our
10688 * needs. In addition to modifying the active vmcs (which is vmcs02), this
10689 * function also has additional necessary side-effects, like setting various
10690 * vcpu->arch fields.
10691 * Returns 0 on success, 1 on failure. Invalid state exit qualification code
10692 * is assigned to entry_failure_code on failure.
10694 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10695 bool from_vmentry, u32 *entry_failure_code)
10697 struct vcpu_vmx *vmx = to_vmx(vcpu);
10698 u32 exec_control, vmcs12_exec_ctrl;
10701 * First, the fields that are shadowed. This must be kept in sync
10702 * with vmx_shadow_fields.h.
10705 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
10706 vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
10707 vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
10708 vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
10709 vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
10712 * Not in vmcs02: GUEST_PML_INDEX, HOST_FS_SELECTOR, HOST_GS_SELECTOR,
10713 * HOST_FS_BASE, HOST_GS_BASE.
10716 if (from_vmentry &&
10717 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
10718 kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
10719 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
10721 kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
10722 vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
10724 if (from_vmentry) {
10725 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
10726 vmcs12->vm_entry_intr_info_field);
10727 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
10728 vmcs12->vm_entry_exception_error_code);
10729 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
10730 vmcs12->vm_entry_instruction_len);
10731 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
10732 vmcs12->guest_interruptibility_info);
10733 vmx->loaded_vmcs->nmi_known_unmasked =
10734 !(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
10736 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
10738 vmx_set_rflags(vcpu, vmcs12->guest_rflags);
10740 exec_control = vmcs12->pin_based_vm_exec_control;
10742 /* Preemption timer setting is only taken from vmcs01. */
10743 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
10744 exec_control |= vmcs_config.pin_based_exec_ctrl;
10745 if (vmx->hv_deadline_tsc == -1)
10746 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
10748 /* Posted interrupts setting is only taken from vmcs12. */
10749 if (nested_cpu_has_posted_intr(vmcs12)) {
10750 vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
10751 vmx->nested.pi_pending = false;
10753 exec_control &= ~PIN_BASED_POSTED_INTR;
10756 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
10758 vmx->nested.preemption_timer_expired = false;
10759 if (nested_cpu_has_preemption_timer(vmcs12))
10760 vmx_start_preemption_timer(vcpu);
10762 if (cpu_has_secondary_exec_ctrls()) {
10763 exec_control = vmx->secondary_exec_control;
10765 /* Take the following fields only from vmcs12 */
10766 exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
10767 SECONDARY_EXEC_ENABLE_INVPCID |
10768 SECONDARY_EXEC_RDTSCP |
10769 SECONDARY_EXEC_XSAVES |
10770 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
10771 SECONDARY_EXEC_APIC_REGISTER_VIRT |
10772 SECONDARY_EXEC_ENABLE_VMFUNC);
10773 if (nested_cpu_has(vmcs12,
10774 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)) {
10775 vmcs12_exec_ctrl = vmcs12->secondary_vm_exec_control &
10776 ~SECONDARY_EXEC_ENABLE_PML;
10777 exec_control |= vmcs12_exec_ctrl;
10780 if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
10781 vmcs_write16(GUEST_INTR_STATUS,
10782 vmcs12->guest_intr_status);
10785 * Write an illegal value to APIC_ACCESS_ADDR. Later,
10786 * nested_get_vmcs12_pages will either fix it up or
10787 * remove the VM execution control.
10789 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
10790 vmcs_write64(APIC_ACCESS_ADDR, -1ull);
10792 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
10796 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
10797 * entry, but only if the current (host) sp changed from the value
10798 * we wrote last (vmx->host_rsp). This cache is no longer relevant
10799 * if we switch vmcs, and rather than hold a separate cache per vmcs,
10800 * here we just force the write to happen on entry.
10804 exec_control = vmx_exec_control(vmx); /* L0's desires */
10805 exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
10806 exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
10807 exec_control &= ~CPU_BASED_TPR_SHADOW;
10808 exec_control |= vmcs12->cpu_based_vm_exec_control;
10811 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR. Later, if
10812 * nested_get_vmcs12_pages can't fix it up, the illegal value
10813 * will result in a VM entry failure.
10815 if (exec_control & CPU_BASED_TPR_SHADOW) {
10816 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, -1ull);
10817 vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
10819 #ifdef CONFIG_X86_64
10820 exec_control |= CPU_BASED_CR8_LOAD_EXITING |
10821 CPU_BASED_CR8_STORE_EXITING;
10826 * A vmexit (to either L1 hypervisor or L0 userspace) is always needed
10827 * for I/O port accesses.
10829 exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
10830 exec_control |= CPU_BASED_UNCOND_IO_EXITING;
10832 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
10834 /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
10835 * bitwise-or of what L1 wants to trap for L2, and what we want to
10836 * trap. Note that CR0.TS also needs updating - we do this later.
10838 update_exception_bitmap(vcpu);
10839 vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
10840 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
10842 /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
10843 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
10844 * bits are further modified by vmx_set_efer() below.
10846 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
10848 /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
10849 * emulated by vmx_set_efer(), below.
10851 vm_entry_controls_init(vmx,
10852 (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER &
10853 ~VM_ENTRY_IA32E_MODE) |
10854 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
10856 if (from_vmentry &&
10857 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
10858 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
10859 vcpu->arch.pat = vmcs12->guest_ia32_pat;
10860 } else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
10861 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
10864 if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
10865 vmcs_write64(TSC_OFFSET,
10866 vcpu->arch.tsc_offset + vmcs12->tsc_offset);
10868 vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
10869 if (kvm_has_tsc_control)
10870 decache_tsc_multiplier(vmx);
10874 * There is no direct mapping between vpid02 and vpid12, the
10875 * vpid02 is per-vCPU for L0 and reused while the value of
10876 * vpid12 is changed w/ one invvpid during nested vmentry.
10877 * The vpid12 is allocated by L1 for L2, so it will not
10878 * influence global bitmap(for vpid01 and vpid02 allocation)
10879 * even if spawn a lot of nested vCPUs.
10881 if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02) {
10882 if (vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
10883 vmx->nested.last_vpid = vmcs12->virtual_processor_id;
10884 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->nested.vpid02, true);
10887 vmx_flush_tlb(vcpu, true);
10893 * Conceptually we want to copy the PML address and index from
10894 * vmcs01 here, and then back to vmcs01 on nested vmexit. But,
10895 * since we always flush the log on each vmexit, this happens
10896 * to be equivalent to simply resetting the fields in vmcs02.
10898 ASSERT(vmx->pml_pg);
10899 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
10900 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
10903 if (nested_cpu_has_ept(vmcs12)) {
10904 if (nested_ept_init_mmu_context(vcpu)) {
10905 *entry_failure_code = ENTRY_FAIL_DEFAULT;
10908 } else if (nested_cpu_has2(vmcs12,
10909 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
10910 vmx_flush_tlb_ept_only(vcpu);
10914 * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
10915 * bits which we consider mandatory enabled.
10916 * The CR0_READ_SHADOW is what L2 should have expected to read given
10917 * the specifications by L1; It's not enough to take
10918 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
10919 * have more bits than L1 expected.
10921 vmx_set_cr0(vcpu, vmcs12->guest_cr0);
10922 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
10924 vmx_set_cr4(vcpu, vmcs12->guest_cr4);
10925 vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
10927 if (from_vmentry &&
10928 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
10929 vcpu->arch.efer = vmcs12->guest_ia32_efer;
10930 else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
10931 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10933 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10934 /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
10935 vmx_set_efer(vcpu, vcpu->arch.efer);
10937 if (vmx->nested.dirty_vmcs12) {
10938 prepare_vmcs02_full(vcpu, vmcs12, from_vmentry);
10939 vmx->nested.dirty_vmcs12 = false;
10942 /* Shadow page tables on either EPT or shadow page tables. */
10943 if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
10944 entry_failure_code))
10948 vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
10950 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
10951 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
10955 static int check_vmentry_prereqs(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10957 struct vcpu_vmx *vmx = to_vmx(vcpu);
10959 if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
10960 vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT)
10961 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10963 if (nested_vmx_check_io_bitmap_controls(vcpu, vmcs12))
10964 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10966 if (nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12))
10967 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10969 if (nested_vmx_check_tpr_shadow_controls(vcpu, vmcs12))
10970 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10972 if (nested_vmx_check_apicv_controls(vcpu, vmcs12))
10973 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10975 if (nested_vmx_check_msr_switch_controls(vcpu, vmcs12))
10976 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10978 if (nested_vmx_check_pml_controls(vcpu, vmcs12))
10979 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10981 if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
10982 vmx->nested.nested_vmx_procbased_ctls_low,
10983 vmx->nested.nested_vmx_procbased_ctls_high) ||
10984 (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
10985 !vmx_control_verify(vmcs12->secondary_vm_exec_control,
10986 vmx->nested.nested_vmx_secondary_ctls_low,
10987 vmx->nested.nested_vmx_secondary_ctls_high)) ||
10988 !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
10989 vmx->nested.nested_vmx_pinbased_ctls_low,
10990 vmx->nested.nested_vmx_pinbased_ctls_high) ||
10991 !vmx_control_verify(vmcs12->vm_exit_controls,
10992 vmx->nested.nested_vmx_exit_ctls_low,
10993 vmx->nested.nested_vmx_exit_ctls_high) ||
10994 !vmx_control_verify(vmcs12->vm_entry_controls,
10995 vmx->nested.nested_vmx_entry_ctls_low,
10996 vmx->nested.nested_vmx_entry_ctls_high))
10997 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10999 if (nested_cpu_has_vmfunc(vmcs12)) {
11000 if (vmcs12->vm_function_control &
11001 ~vmx->nested.nested_vmx_vmfunc_controls)
11002 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
11004 if (nested_cpu_has_eptp_switching(vmcs12)) {
11005 if (!nested_cpu_has_ept(vmcs12) ||
11006 !page_address_valid(vcpu, vmcs12->eptp_list_address))
11007 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
11011 if (vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu))
11012 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
11014 if (!nested_host_cr0_valid(vcpu, vmcs12->host_cr0) ||
11015 !nested_host_cr4_valid(vcpu, vmcs12->host_cr4) ||
11016 !nested_cr3_valid(vcpu, vmcs12->host_cr3))
11017 return VMXERR_ENTRY_INVALID_HOST_STATE_FIELD;
11022 static int check_vmentry_postreqs(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
11027 *exit_qual = ENTRY_FAIL_DEFAULT;
11029 if (!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0) ||
11030 !nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4))
11033 if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_SHADOW_VMCS) &&
11034 vmcs12->vmcs_link_pointer != -1ull) {
11035 *exit_qual = ENTRY_FAIL_VMCS_LINK_PTR;
11040 * If the load IA32_EFER VM-entry control is 1, the following checks
11041 * are performed on the field for the IA32_EFER MSR:
11042 * - Bits reserved in the IA32_EFER MSR must be 0.
11043 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
11044 * the IA-32e mode guest VM-exit control. It must also be identical
11045 * to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
11048 if (to_vmx(vcpu)->nested.nested_run_pending &&
11049 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
11050 ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
11051 if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
11052 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
11053 ((vmcs12->guest_cr0 & X86_CR0_PG) &&
11054 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME)))
11059 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
11060 * IA32_EFER MSR must be 0 in the field for that register. In addition,
11061 * the values of the LMA and LME bits in the field must each be that of
11062 * the host address-space size VM-exit control.
11064 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
11065 ia32e = (vmcs12->vm_exit_controls &
11066 VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
11067 if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
11068 ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
11069 ia32e != !!(vmcs12->host_ia32_efer & EFER_LME))
11073 if ((vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS) &&
11074 (is_noncanonical_address(vmcs12->guest_bndcfgs & PAGE_MASK, vcpu) ||
11075 (vmcs12->guest_bndcfgs & MSR_IA32_BNDCFGS_RSVD)))
11081 static int enter_vmx_non_root_mode(struct kvm_vcpu *vcpu, bool from_vmentry)
11083 struct vcpu_vmx *vmx = to_vmx(vcpu);
11084 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
11088 enter_guest_mode(vcpu);
11090 if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
11091 vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
11093 vmx_switch_vmcs(vcpu, &vmx->nested.vmcs02);
11094 vmx_segment_cache_clear(vmx);
11096 if (prepare_vmcs02(vcpu, vmcs12, from_vmentry, &exit_qual)) {
11097 leave_guest_mode(vcpu);
11098 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
11099 nested_vmx_entry_failure(vcpu, vmcs12,
11100 EXIT_REASON_INVALID_STATE, exit_qual);
11104 nested_get_vmcs12_pages(vcpu, vmcs12);
11106 msr_entry_idx = nested_vmx_load_msr(vcpu,
11107 vmcs12->vm_entry_msr_load_addr,
11108 vmcs12->vm_entry_msr_load_count);
11109 if (msr_entry_idx) {
11110 leave_guest_mode(vcpu);
11111 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
11112 nested_vmx_entry_failure(vcpu, vmcs12,
11113 EXIT_REASON_MSR_LOAD_FAIL, msr_entry_idx);
11118 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
11119 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
11120 * returned as far as L1 is concerned. It will only return (and set
11121 * the success flag) when L2 exits (see nested_vmx_vmexit()).
11127 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
11128 * for running an L2 nested guest.
11130 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
11132 struct vmcs12 *vmcs12;
11133 struct vcpu_vmx *vmx = to_vmx(vcpu);
11134 u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
11138 if (!nested_vmx_check_permission(vcpu))
11141 if (!nested_vmx_check_vmcs12(vcpu))
11144 vmcs12 = get_vmcs12(vcpu);
11146 if (enable_shadow_vmcs)
11147 copy_shadow_to_vmcs12(vmx);
11150 * The nested entry process starts with enforcing various prerequisites
11151 * on vmcs12 as required by the Intel SDM, and act appropriately when
11152 * they fail: As the SDM explains, some conditions should cause the
11153 * instruction to fail, while others will cause the instruction to seem
11154 * to succeed, but return an EXIT_REASON_INVALID_STATE.
11155 * To speed up the normal (success) code path, we should avoid checking
11156 * for misconfigurations which will anyway be caught by the processor
11157 * when using the merged vmcs02.
11159 if (interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS) {
11160 nested_vmx_failValid(vcpu,
11161 VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
11165 if (vmcs12->launch_state == launch) {
11166 nested_vmx_failValid(vcpu,
11167 launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
11168 : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
11172 ret = check_vmentry_prereqs(vcpu, vmcs12);
11174 nested_vmx_failValid(vcpu, ret);
11179 * After this point, the trap flag no longer triggers a singlestep trap
11180 * on the vm entry instructions; don't call kvm_skip_emulated_instruction.
11181 * This is not 100% correct; for performance reasons, we delegate most
11182 * of the checks on host state to the processor. If those fail,
11183 * the singlestep trap is missed.
11185 skip_emulated_instruction(vcpu);
11187 ret = check_vmentry_postreqs(vcpu, vmcs12, &exit_qual);
11189 nested_vmx_entry_failure(vcpu, vmcs12,
11190 EXIT_REASON_INVALID_STATE, exit_qual);
11195 * We're finally done with prerequisite checking, and can start with
11196 * the nested entry.
11199 ret = enter_vmx_non_root_mode(vcpu, true);
11203 if (vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT)
11204 return kvm_vcpu_halt(vcpu);
11206 vmx->nested.nested_run_pending = 1;
11211 return kvm_skip_emulated_instruction(vcpu);
11215 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
11216 * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
11217 * This function returns the new value we should put in vmcs12.guest_cr0.
11218 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
11219 * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
11220 * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
11221 * didn't trap the bit, because if L1 did, so would L0).
11222 * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have
11223 * been modified by L2, and L1 knows it. So just leave the old value of
11224 * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
11225 * isn't relevant, because if L0 traps this bit it can set it to anything.
11226 * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
11227 * changed these bits, and therefore they need to be updated, but L0
11228 * didn't necessarily allow them to be changed in GUEST_CR0 - and rather
11229 * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
11231 static inline unsigned long
11232 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
11235 /*1*/ (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
11236 /*2*/ (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
11237 /*3*/ (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
11238 vcpu->arch.cr0_guest_owned_bits));
11241 static inline unsigned long
11242 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
11245 /*1*/ (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
11246 /*2*/ (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
11247 /*3*/ (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
11248 vcpu->arch.cr4_guest_owned_bits));
11251 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
11252 struct vmcs12 *vmcs12)
11257 if (vcpu->arch.exception.injected) {
11258 nr = vcpu->arch.exception.nr;
11259 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
11261 if (kvm_exception_is_soft(nr)) {
11262 vmcs12->vm_exit_instruction_len =
11263 vcpu->arch.event_exit_inst_len;
11264 idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
11266 idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
11268 if (vcpu->arch.exception.has_error_code) {
11269 idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
11270 vmcs12->idt_vectoring_error_code =
11271 vcpu->arch.exception.error_code;
11274 vmcs12->idt_vectoring_info_field = idt_vectoring;
11275 } else if (vcpu->arch.nmi_injected) {
11276 vmcs12->idt_vectoring_info_field =
11277 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
11278 } else if (vcpu->arch.interrupt.pending) {
11279 nr = vcpu->arch.interrupt.nr;
11280 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
11282 if (vcpu->arch.interrupt.soft) {
11283 idt_vectoring |= INTR_TYPE_SOFT_INTR;
11284 vmcs12->vm_entry_instruction_len =
11285 vcpu->arch.event_exit_inst_len;
11287 idt_vectoring |= INTR_TYPE_EXT_INTR;
11289 vmcs12->idt_vectoring_info_field = idt_vectoring;
11293 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
11295 struct vcpu_vmx *vmx = to_vmx(vcpu);
11296 unsigned long exit_qual;
11297 bool block_nested_events =
11298 vmx->nested.nested_run_pending || kvm_event_needs_reinjection(vcpu);
11300 if (vcpu->arch.exception.pending &&
11301 nested_vmx_check_exception(vcpu, &exit_qual)) {
11302 if (block_nested_events)
11304 nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
11308 if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
11309 vmx->nested.preemption_timer_expired) {
11310 if (block_nested_events)
11312 nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
11316 if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
11317 if (block_nested_events)
11319 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
11320 NMI_VECTOR | INTR_TYPE_NMI_INTR |
11321 INTR_INFO_VALID_MASK, 0);
11323 * The NMI-triggered VM exit counts as injection:
11324 * clear this one and block further NMIs.
11326 vcpu->arch.nmi_pending = 0;
11327 vmx_set_nmi_mask(vcpu, true);
11331 if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
11332 nested_exit_on_intr(vcpu)) {
11333 if (block_nested_events)
11335 nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
11339 vmx_complete_nested_posted_interrupt(vcpu);
11343 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
11345 ktime_t remaining =
11346 hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
11349 if (ktime_to_ns(remaining) <= 0)
11352 value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
11353 do_div(value, 1000000);
11354 return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
11358 * Update the guest state fields of vmcs12 to reflect changes that
11359 * occurred while L2 was running. (The "IA-32e mode guest" bit of the
11360 * VM-entry controls is also updated, since this is really a guest
11363 static void sync_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
11365 vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
11366 vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
11368 vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
11369 vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
11370 vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
11372 vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
11373 vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
11374 vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
11375 vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
11376 vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
11377 vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
11378 vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
11379 vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
11380 vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
11381 vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
11382 vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
11383 vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
11384 vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
11385 vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
11386 vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
11387 vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
11388 vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
11389 vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
11390 vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
11391 vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
11392 vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
11393 vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
11394 vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
11395 vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
11396 vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
11397 vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
11398 vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
11399 vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
11400 vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
11401 vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
11402 vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
11403 vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
11404 vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
11405 vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
11406 vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
11407 vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
11409 vmcs12->guest_interruptibility_info =
11410 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
11411 vmcs12->guest_pending_dbg_exceptions =
11412 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
11413 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
11414 vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
11416 vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
11418 if (nested_cpu_has_preemption_timer(vmcs12)) {
11419 if (vmcs12->vm_exit_controls &
11420 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
11421 vmcs12->vmx_preemption_timer_value =
11422 vmx_get_preemption_timer_value(vcpu);
11423 hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
11427 * In some cases (usually, nested EPT), L2 is allowed to change its
11428 * own CR3 without exiting. If it has changed it, we must keep it.
11429 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
11430 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
11432 * Additionally, restore L2's PDPTR to vmcs12.
11435 vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
11436 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
11437 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
11438 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
11439 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
11442 vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
11444 if (nested_cpu_has_vid(vmcs12))
11445 vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
11447 vmcs12->vm_entry_controls =
11448 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
11449 (vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
11451 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) {
11452 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
11453 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
11456 /* TODO: These cannot have changed unless we have MSR bitmaps and
11457 * the relevant bit asks not to trap the change */
11458 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
11459 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
11460 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
11461 vmcs12->guest_ia32_efer = vcpu->arch.efer;
11462 vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
11463 vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
11464 vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
11465 if (kvm_mpx_supported())
11466 vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
11470 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
11471 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
11472 * and this function updates it to reflect the changes to the guest state while
11473 * L2 was running (and perhaps made some exits which were handled directly by L0
11474 * without going back to L1), and to reflect the exit reason.
11475 * Note that we do not have to copy here all VMCS fields, just those that
11476 * could have changed by the L2 guest or the exit - i.e., the guest-state and
11477 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
11478 * which already writes to vmcs12 directly.
11480 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
11481 u32 exit_reason, u32 exit_intr_info,
11482 unsigned long exit_qualification)
11484 /* update guest state fields: */
11485 sync_vmcs12(vcpu, vmcs12);
11487 /* update exit information fields: */
11489 vmcs12->vm_exit_reason = exit_reason;
11490 vmcs12->exit_qualification = exit_qualification;
11491 vmcs12->vm_exit_intr_info = exit_intr_info;
11493 vmcs12->idt_vectoring_info_field = 0;
11494 vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
11495 vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
11497 if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
11498 vmcs12->launch_state = 1;
11500 /* vm_entry_intr_info_field is cleared on exit. Emulate this
11501 * instead of reading the real value. */
11502 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
11505 * Transfer the event that L0 or L1 may wanted to inject into
11506 * L2 to IDT_VECTORING_INFO_FIELD.
11508 vmcs12_save_pending_event(vcpu, vmcs12);
11512 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
11513 * preserved above and would only end up incorrectly in L1.
11515 vcpu->arch.nmi_injected = false;
11516 kvm_clear_exception_queue(vcpu);
11517 kvm_clear_interrupt_queue(vcpu);
11520 static void load_vmcs12_mmu_host_state(struct kvm_vcpu *vcpu,
11521 struct vmcs12 *vmcs12)
11523 u32 entry_failure_code;
11525 nested_ept_uninit_mmu_context(vcpu);
11528 * Only PDPTE load can fail as the value of cr3 was checked on entry and
11529 * couldn't have changed.
11531 if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, &entry_failure_code))
11532 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
11535 vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
11539 * A part of what we need to when the nested L2 guest exits and we want to
11540 * run its L1 parent, is to reset L1's guest state to the host state specified
11542 * This function is to be called not only on normal nested exit, but also on
11543 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
11544 * Failures During or After Loading Guest State").
11545 * This function should be called when the active VMCS is L1's (vmcs01).
11547 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
11548 struct vmcs12 *vmcs12)
11550 struct kvm_segment seg;
11552 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
11553 vcpu->arch.efer = vmcs12->host_ia32_efer;
11554 else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
11555 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
11557 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
11558 vmx_set_efer(vcpu, vcpu->arch.efer);
11560 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
11561 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
11562 vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
11564 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
11565 * actually changed, because vmx_set_cr0 refers to efer set above.
11567 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
11568 * (KVM doesn't change it);
11570 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
11571 vmx_set_cr0(vcpu, vmcs12->host_cr0);
11573 /* Same as above - no reason to call set_cr4_guest_host_mask(). */
11574 vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
11575 vmx_set_cr4(vcpu, vmcs12->host_cr4);
11577 load_vmcs12_mmu_host_state(vcpu, vmcs12);
11581 * Trivially support vpid by letting L2s share their parent
11582 * L1's vpid. TODO: move to a more elaborate solution, giving
11583 * each L2 its own vpid and exposing the vpid feature to L1.
11585 vmx_flush_tlb(vcpu, true);
11588 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
11589 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
11590 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
11591 vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
11592 vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
11593 vmcs_write32(GUEST_IDTR_LIMIT, 0xFFFF);
11594 vmcs_write32(GUEST_GDTR_LIMIT, 0xFFFF);
11596 /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1. */
11597 if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
11598 vmcs_write64(GUEST_BNDCFGS, 0);
11600 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
11601 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
11602 vcpu->arch.pat = vmcs12->host_ia32_pat;
11604 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
11605 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
11606 vmcs12->host_ia32_perf_global_ctrl);
11608 /* Set L1 segment info according to Intel SDM
11609 27.5.2 Loading Host Segment and Descriptor-Table Registers */
11610 seg = (struct kvm_segment) {
11612 .limit = 0xFFFFFFFF,
11613 .selector = vmcs12->host_cs_selector,
11619 if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
11623 vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
11624 seg = (struct kvm_segment) {
11626 .limit = 0xFFFFFFFF,
11633 seg.selector = vmcs12->host_ds_selector;
11634 vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
11635 seg.selector = vmcs12->host_es_selector;
11636 vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
11637 seg.selector = vmcs12->host_ss_selector;
11638 vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
11639 seg.selector = vmcs12->host_fs_selector;
11640 seg.base = vmcs12->host_fs_base;
11641 vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
11642 seg.selector = vmcs12->host_gs_selector;
11643 seg.base = vmcs12->host_gs_base;
11644 vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
11645 seg = (struct kvm_segment) {
11646 .base = vmcs12->host_tr_base,
11648 .selector = vmcs12->host_tr_selector,
11652 vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
11654 kvm_set_dr(vcpu, 7, 0x400);
11655 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
11657 if (cpu_has_vmx_msr_bitmap())
11658 vmx_update_msr_bitmap(vcpu);
11660 if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
11661 vmcs12->vm_exit_msr_load_count))
11662 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
11666 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
11667 * and modify vmcs12 to make it see what it would expect to see there if
11668 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
11670 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
11671 u32 exit_intr_info,
11672 unsigned long exit_qualification)
11674 struct vcpu_vmx *vmx = to_vmx(vcpu);
11675 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
11677 /* trying to cancel vmlaunch/vmresume is a bug */
11678 WARN_ON_ONCE(vmx->nested.nested_run_pending);
11681 * The only expected VM-instruction error is "VM entry with
11682 * invalid control field(s)." Anything else indicates a
11685 WARN_ON_ONCE(vmx->fail && (vmcs_read32(VM_INSTRUCTION_ERROR) !=
11686 VMXERR_ENTRY_INVALID_CONTROL_FIELD));
11688 leave_guest_mode(vcpu);
11690 if (likely(!vmx->fail)) {
11691 if (exit_reason == -1)
11692 sync_vmcs12(vcpu, vmcs12);
11694 prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
11695 exit_qualification);
11697 if (nested_vmx_store_msr(vcpu, vmcs12->vm_exit_msr_store_addr,
11698 vmcs12->vm_exit_msr_store_count))
11699 nested_vmx_abort(vcpu, VMX_ABORT_SAVE_GUEST_MSR_FAIL);
11702 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
11703 vm_entry_controls_reset_shadow(vmx);
11704 vm_exit_controls_reset_shadow(vmx);
11705 vmx_segment_cache_clear(vmx);
11707 /* Update any VMCS fields that might have changed while L2 ran */
11708 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
11709 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
11710 vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
11711 if (vmx->hv_deadline_tsc == -1)
11712 vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
11713 PIN_BASED_VMX_PREEMPTION_TIMER);
11715 vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
11716 PIN_BASED_VMX_PREEMPTION_TIMER);
11717 if (kvm_has_tsc_control)
11718 decache_tsc_multiplier(vmx);
11720 if (vmx->nested.change_vmcs01_virtual_x2apic_mode) {
11721 vmx->nested.change_vmcs01_virtual_x2apic_mode = false;
11722 vmx_set_virtual_x2apic_mode(vcpu,
11723 vcpu->arch.apic_base & X2APIC_ENABLE);
11724 } else if (!nested_cpu_has_ept(vmcs12) &&
11725 nested_cpu_has2(vmcs12,
11726 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
11727 vmx_flush_tlb_ept_only(vcpu);
11730 /* This is needed for same reason as it was needed in prepare_vmcs02 */
11733 /* Unpin physical memory we referred to in vmcs02 */
11734 if (vmx->nested.apic_access_page) {
11735 kvm_release_page_dirty(vmx->nested.apic_access_page);
11736 vmx->nested.apic_access_page = NULL;
11738 if (vmx->nested.virtual_apic_page) {
11739 kvm_release_page_dirty(vmx->nested.virtual_apic_page);
11740 vmx->nested.virtual_apic_page = NULL;
11742 if (vmx->nested.pi_desc_page) {
11743 kunmap(vmx->nested.pi_desc_page);
11744 kvm_release_page_dirty(vmx->nested.pi_desc_page);
11745 vmx->nested.pi_desc_page = NULL;
11746 vmx->nested.pi_desc = NULL;
11750 * We are now running in L2, mmu_notifier will force to reload the
11751 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
11753 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
11755 if (enable_shadow_vmcs && exit_reason != -1)
11756 vmx->nested.sync_shadow_vmcs = true;
11758 /* in case we halted in L2 */
11759 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11761 if (likely(!vmx->fail)) {
11763 * TODO: SDM says that with acknowledge interrupt on
11764 * exit, bit 31 of the VM-exit interrupt information
11765 * (valid interrupt) is always set to 1 on
11766 * EXIT_REASON_EXTERNAL_INTERRUPT, so we shouldn't
11767 * need kvm_cpu_has_interrupt(). See the commit
11768 * message for details.
11770 if (nested_exit_intr_ack_set(vcpu) &&
11771 exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT &&
11772 kvm_cpu_has_interrupt(vcpu)) {
11773 int irq = kvm_cpu_get_interrupt(vcpu);
11775 vmcs12->vm_exit_intr_info = irq |
11776 INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
11779 if (exit_reason != -1)
11780 trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
11781 vmcs12->exit_qualification,
11782 vmcs12->idt_vectoring_info_field,
11783 vmcs12->vm_exit_intr_info,
11784 vmcs12->vm_exit_intr_error_code,
11787 load_vmcs12_host_state(vcpu, vmcs12);
11793 * After an early L2 VM-entry failure, we're now back
11794 * in L1 which thinks it just finished a VMLAUNCH or
11795 * VMRESUME instruction, so we need to set the failure
11796 * flag and the VM-instruction error field of the VMCS
11799 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
11801 load_vmcs12_mmu_host_state(vcpu, vmcs12);
11804 * The emulated instruction was already skipped in
11805 * nested_vmx_run, but the updated RIP was never
11806 * written back to the vmcs01.
11808 skip_emulated_instruction(vcpu);
11813 * Forcibly leave nested mode in order to be able to reset the VCPU later on.
11815 static void vmx_leave_nested(struct kvm_vcpu *vcpu)
11817 if (is_guest_mode(vcpu)) {
11818 to_vmx(vcpu)->nested.nested_run_pending = 0;
11819 nested_vmx_vmexit(vcpu, -1, 0, 0);
11821 free_nested(to_vmx(vcpu));
11825 * L1's failure to enter L2 is a subset of a normal exit, as explained in
11826 * 23.7 "VM-entry failures during or after loading guest state" (this also
11827 * lists the acceptable exit-reason and exit-qualification parameters).
11828 * It should only be called before L2 actually succeeded to run, and when
11829 * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
11831 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
11832 struct vmcs12 *vmcs12,
11833 u32 reason, unsigned long qualification)
11835 load_vmcs12_host_state(vcpu, vmcs12);
11836 vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
11837 vmcs12->exit_qualification = qualification;
11838 nested_vmx_succeed(vcpu);
11839 if (enable_shadow_vmcs)
11840 to_vmx(vcpu)->nested.sync_shadow_vmcs = true;
11843 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
11844 struct x86_instruction_info *info,
11845 enum x86_intercept_stage stage)
11847 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
11848 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
11851 * RDPID causes #UD if disabled through secondary execution controls.
11852 * Because it is marked as EmulateOnUD, we need to intercept it here.
11854 if (info->intercept == x86_intercept_rdtscp &&
11855 !nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDTSCP)) {
11856 ctxt->exception.vector = UD_VECTOR;
11857 ctxt->exception.error_code_valid = false;
11858 return X86EMUL_PROPAGATE_FAULT;
11861 /* TODO: check more intercepts... */
11862 return X86EMUL_CONTINUE;
11865 #ifdef CONFIG_X86_64
11866 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
11867 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
11868 u64 divisor, u64 *result)
11870 u64 low = a << shift, high = a >> (64 - shift);
11872 /* To avoid the overflow on divq */
11873 if (high >= divisor)
11876 /* Low hold the result, high hold rem which is discarded */
11877 asm("divq %2\n\t" : "=a" (low), "=d" (high) :
11878 "rm" (divisor), "0" (low), "1" (high));
11884 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc)
11886 struct vcpu_vmx *vmx = to_vmx(vcpu);
11887 u64 tscl = rdtsc();
11888 u64 guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
11889 u64 delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
11891 /* Convert to host delta tsc if tsc scaling is enabled */
11892 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
11893 u64_shl_div_u64(delta_tsc,
11894 kvm_tsc_scaling_ratio_frac_bits,
11895 vcpu->arch.tsc_scaling_ratio,
11900 * If the delta tsc can't fit in the 32 bit after the multi shift,
11901 * we can't use the preemption timer.
11902 * It's possible that it fits on later vmentries, but checking
11903 * on every vmentry is costly so we just use an hrtimer.
11905 if (delta_tsc >> (cpu_preemption_timer_multi + 32))
11908 vmx->hv_deadline_tsc = tscl + delta_tsc;
11909 vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
11910 PIN_BASED_VMX_PREEMPTION_TIMER);
11912 return delta_tsc == 0;
11915 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
11917 struct vcpu_vmx *vmx = to_vmx(vcpu);
11918 vmx->hv_deadline_tsc = -1;
11919 vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
11920 PIN_BASED_VMX_PREEMPTION_TIMER);
11924 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
11927 shrink_ple_window(vcpu);
11930 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
11931 struct kvm_memory_slot *slot)
11933 kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
11934 kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
11937 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
11938 struct kvm_memory_slot *slot)
11940 kvm_mmu_slot_set_dirty(kvm, slot);
11943 static void vmx_flush_log_dirty(struct kvm *kvm)
11945 kvm_flush_pml_buffers(kvm);
11948 static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu)
11950 struct vmcs12 *vmcs12;
11951 struct vcpu_vmx *vmx = to_vmx(vcpu);
11953 struct page *page = NULL;
11956 if (is_guest_mode(vcpu)) {
11957 WARN_ON_ONCE(vmx->nested.pml_full);
11960 * Check if PML is enabled for the nested guest.
11961 * Whether eptp bit 6 is set is already checked
11962 * as part of A/D emulation.
11964 vmcs12 = get_vmcs12(vcpu);
11965 if (!nested_cpu_has_pml(vmcs12))
11968 if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
11969 vmx->nested.pml_full = true;
11973 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull;
11975 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->pml_address);
11976 if (is_error_page(page))
11979 pml_address = kmap(page);
11980 pml_address[vmcs12->guest_pml_index--] = gpa;
11982 kvm_release_page_clean(page);
11988 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
11989 struct kvm_memory_slot *memslot,
11990 gfn_t offset, unsigned long mask)
11992 kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
11995 static void __pi_post_block(struct kvm_vcpu *vcpu)
11997 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
11998 struct pi_desc old, new;
12002 old.control = new.control = pi_desc->control;
12003 WARN(old.nv != POSTED_INTR_WAKEUP_VECTOR,
12004 "Wakeup handler not enabled while the VCPU is blocked\n");
12006 dest = cpu_physical_id(vcpu->cpu);
12008 if (x2apic_enabled())
12011 new.ndst = (dest << 8) & 0xFF00;
12013 /* set 'NV' to 'notification vector' */
12014 new.nv = POSTED_INTR_VECTOR;
12015 } while (cmpxchg64(&pi_desc->control, old.control,
12016 new.control) != old.control);
12018 if (!WARN_ON_ONCE(vcpu->pre_pcpu == -1)) {
12019 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
12020 list_del(&vcpu->blocked_vcpu_list);
12021 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
12022 vcpu->pre_pcpu = -1;
12027 * This routine does the following things for vCPU which is going
12028 * to be blocked if VT-d PI is enabled.
12029 * - Store the vCPU to the wakeup list, so when interrupts happen
12030 * we can find the right vCPU to wake up.
12031 * - Change the Posted-interrupt descriptor as below:
12032 * 'NDST' <-- vcpu->pre_pcpu
12033 * 'NV' <-- POSTED_INTR_WAKEUP_VECTOR
12034 * - If 'ON' is set during this process, which means at least one
12035 * interrupt is posted for this vCPU, we cannot block it, in
12036 * this case, return 1, otherwise, return 0.
12039 static int pi_pre_block(struct kvm_vcpu *vcpu)
12042 struct pi_desc old, new;
12043 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
12045 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
12046 !irq_remapping_cap(IRQ_POSTING_CAP) ||
12047 !kvm_vcpu_apicv_active(vcpu))
12050 WARN_ON(irqs_disabled());
12051 local_irq_disable();
12052 if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) {
12053 vcpu->pre_pcpu = vcpu->cpu;
12054 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
12055 list_add_tail(&vcpu->blocked_vcpu_list,
12056 &per_cpu(blocked_vcpu_on_cpu,
12058 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
12062 old.control = new.control = pi_desc->control;
12064 WARN((pi_desc->sn == 1),
12065 "Warning: SN field of posted-interrupts "
12066 "is set before blocking\n");
12069 * Since vCPU can be preempted during this process,
12070 * vcpu->cpu could be different with pre_pcpu, we
12071 * need to set pre_pcpu as the destination of wakeup
12072 * notification event, then we can find the right vCPU
12073 * to wakeup in wakeup handler if interrupts happen
12074 * when the vCPU is in blocked state.
12076 dest = cpu_physical_id(vcpu->pre_pcpu);
12078 if (x2apic_enabled())
12081 new.ndst = (dest << 8) & 0xFF00;
12083 /* set 'NV' to 'wakeup vector' */
12084 new.nv = POSTED_INTR_WAKEUP_VECTOR;
12085 } while (cmpxchg64(&pi_desc->control, old.control,
12086 new.control) != old.control);
12088 /* We should not block the vCPU if an interrupt is posted for it. */
12089 if (pi_test_on(pi_desc) == 1)
12090 __pi_post_block(vcpu);
12092 local_irq_enable();
12093 return (vcpu->pre_pcpu == -1);
12096 static int vmx_pre_block(struct kvm_vcpu *vcpu)
12098 if (pi_pre_block(vcpu))
12101 if (kvm_lapic_hv_timer_in_use(vcpu))
12102 kvm_lapic_switch_to_sw_timer(vcpu);
12107 static void pi_post_block(struct kvm_vcpu *vcpu)
12109 if (vcpu->pre_pcpu == -1)
12112 WARN_ON(irqs_disabled());
12113 local_irq_disable();
12114 __pi_post_block(vcpu);
12115 local_irq_enable();
12118 static void vmx_post_block(struct kvm_vcpu *vcpu)
12120 if (kvm_x86_ops->set_hv_timer)
12121 kvm_lapic_switch_to_hv_timer(vcpu);
12123 pi_post_block(vcpu);
12127 * vmx_update_pi_irte - set IRTE for Posted-Interrupts
12130 * @host_irq: host irq of the interrupt
12131 * @guest_irq: gsi of the interrupt
12132 * @set: set or unset PI
12133 * returns 0 on success, < 0 on failure
12135 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
12136 uint32_t guest_irq, bool set)
12138 struct kvm_kernel_irq_routing_entry *e;
12139 struct kvm_irq_routing_table *irq_rt;
12140 struct kvm_lapic_irq irq;
12141 struct kvm_vcpu *vcpu;
12142 struct vcpu_data vcpu_info;
12145 if (!kvm_arch_has_assigned_device(kvm) ||
12146 !irq_remapping_cap(IRQ_POSTING_CAP) ||
12147 !kvm_vcpu_apicv_active(kvm->vcpus[0]))
12150 idx = srcu_read_lock(&kvm->irq_srcu);
12151 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
12152 if (guest_irq >= irq_rt->nr_rt_entries ||
12153 hlist_empty(&irq_rt->map[guest_irq])) {
12154 pr_warn_once("no route for guest_irq %u/%u (broken user space?)\n",
12155 guest_irq, irq_rt->nr_rt_entries);
12159 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
12160 if (e->type != KVM_IRQ_ROUTING_MSI)
12163 * VT-d PI cannot support posting multicast/broadcast
12164 * interrupts to a vCPU, we still use interrupt remapping
12165 * for these kind of interrupts.
12167 * For lowest-priority interrupts, we only support
12168 * those with single CPU as the destination, e.g. user
12169 * configures the interrupts via /proc/irq or uses
12170 * irqbalance to make the interrupts single-CPU.
12172 * We will support full lowest-priority interrupt later.
12175 kvm_set_msi_irq(kvm, e, &irq);
12176 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
12178 * Make sure the IRTE is in remapped mode if
12179 * we don't handle it in posted mode.
12181 ret = irq_set_vcpu_affinity(host_irq, NULL);
12184 "failed to back to remapped mode, irq: %u\n",
12192 vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
12193 vcpu_info.vector = irq.vector;
12195 trace_kvm_pi_irte_update(vcpu->vcpu_id, host_irq, e->gsi,
12196 vcpu_info.vector, vcpu_info.pi_desc_addr, set);
12199 ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
12201 ret = irq_set_vcpu_affinity(host_irq, NULL);
12204 printk(KERN_INFO "%s: failed to update PI IRTE\n",
12212 srcu_read_unlock(&kvm->irq_srcu, idx);
12216 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
12218 if (vcpu->arch.mcg_cap & MCG_LMCE_P)
12219 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
12220 FEATURE_CONTROL_LMCE;
12222 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
12223 ~FEATURE_CONTROL_LMCE;
12226 static int vmx_smi_allowed(struct kvm_vcpu *vcpu)
12228 /* we need a nested vmexit to enter SMM, postpone if run is pending */
12229 if (to_vmx(vcpu)->nested.nested_run_pending)
12234 static int vmx_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
12236 struct vcpu_vmx *vmx = to_vmx(vcpu);
12238 vmx->nested.smm.guest_mode = is_guest_mode(vcpu);
12239 if (vmx->nested.smm.guest_mode)
12240 nested_vmx_vmexit(vcpu, -1, 0, 0);
12242 vmx->nested.smm.vmxon = vmx->nested.vmxon;
12243 vmx->nested.vmxon = false;
12247 static int vmx_pre_leave_smm(struct kvm_vcpu *vcpu, u64 smbase)
12249 struct vcpu_vmx *vmx = to_vmx(vcpu);
12252 if (vmx->nested.smm.vmxon) {
12253 vmx->nested.vmxon = true;
12254 vmx->nested.smm.vmxon = false;
12257 if (vmx->nested.smm.guest_mode) {
12258 vcpu->arch.hflags &= ~HF_SMM_MASK;
12259 ret = enter_vmx_non_root_mode(vcpu, false);
12260 vcpu->arch.hflags |= HF_SMM_MASK;
12264 vmx->nested.smm.guest_mode = false;
12269 static int enable_smi_window(struct kvm_vcpu *vcpu)
12274 static struct kvm_x86_ops vmx_x86_ops __ro_after_init = {
12275 .cpu_has_kvm_support = cpu_has_kvm_support,
12276 .disabled_by_bios = vmx_disabled_by_bios,
12277 .hardware_setup = hardware_setup,
12278 .hardware_unsetup = hardware_unsetup,
12279 .check_processor_compatibility = vmx_check_processor_compat,
12280 .hardware_enable = hardware_enable,
12281 .hardware_disable = hardware_disable,
12282 .cpu_has_accelerated_tpr = report_flexpriority,
12283 .cpu_has_high_real_mode_segbase = vmx_has_high_real_mode_segbase,
12285 .vcpu_create = vmx_create_vcpu,
12286 .vcpu_free = vmx_free_vcpu,
12287 .vcpu_reset = vmx_vcpu_reset,
12289 .prepare_guest_switch = vmx_save_host_state,
12290 .vcpu_load = vmx_vcpu_load,
12291 .vcpu_put = vmx_vcpu_put,
12293 .update_bp_intercept = update_exception_bitmap,
12294 .get_msr = vmx_get_msr,
12295 .set_msr = vmx_set_msr,
12296 .get_segment_base = vmx_get_segment_base,
12297 .get_segment = vmx_get_segment,
12298 .set_segment = vmx_set_segment,
12299 .get_cpl = vmx_get_cpl,
12300 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
12301 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
12302 .decache_cr3 = vmx_decache_cr3,
12303 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
12304 .set_cr0 = vmx_set_cr0,
12305 .set_cr3 = vmx_set_cr3,
12306 .set_cr4 = vmx_set_cr4,
12307 .set_efer = vmx_set_efer,
12308 .get_idt = vmx_get_idt,
12309 .set_idt = vmx_set_idt,
12310 .get_gdt = vmx_get_gdt,
12311 .set_gdt = vmx_set_gdt,
12312 .get_dr6 = vmx_get_dr6,
12313 .set_dr6 = vmx_set_dr6,
12314 .set_dr7 = vmx_set_dr7,
12315 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
12316 .cache_reg = vmx_cache_reg,
12317 .get_rflags = vmx_get_rflags,
12318 .set_rflags = vmx_set_rflags,
12320 .tlb_flush = vmx_flush_tlb,
12322 .run = vmx_vcpu_run,
12323 .handle_exit = vmx_handle_exit,
12324 .skip_emulated_instruction = skip_emulated_instruction,
12325 .set_interrupt_shadow = vmx_set_interrupt_shadow,
12326 .get_interrupt_shadow = vmx_get_interrupt_shadow,
12327 .patch_hypercall = vmx_patch_hypercall,
12328 .set_irq = vmx_inject_irq,
12329 .set_nmi = vmx_inject_nmi,
12330 .queue_exception = vmx_queue_exception,
12331 .cancel_injection = vmx_cancel_injection,
12332 .interrupt_allowed = vmx_interrupt_allowed,
12333 .nmi_allowed = vmx_nmi_allowed,
12334 .get_nmi_mask = vmx_get_nmi_mask,
12335 .set_nmi_mask = vmx_set_nmi_mask,
12336 .enable_nmi_window = enable_nmi_window,
12337 .enable_irq_window = enable_irq_window,
12338 .update_cr8_intercept = update_cr8_intercept,
12339 .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode,
12340 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
12341 .get_enable_apicv = vmx_get_enable_apicv,
12342 .refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
12343 .load_eoi_exitmap = vmx_load_eoi_exitmap,
12344 .apicv_post_state_restore = vmx_apicv_post_state_restore,
12345 .hwapic_irr_update = vmx_hwapic_irr_update,
12346 .hwapic_isr_update = vmx_hwapic_isr_update,
12347 .sync_pir_to_irr = vmx_sync_pir_to_irr,
12348 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
12350 .set_tss_addr = vmx_set_tss_addr,
12351 .get_tdp_level = get_ept_level,
12352 .get_mt_mask = vmx_get_mt_mask,
12354 .get_exit_info = vmx_get_exit_info,
12356 .get_lpage_level = vmx_get_lpage_level,
12358 .cpuid_update = vmx_cpuid_update,
12360 .rdtscp_supported = vmx_rdtscp_supported,
12361 .invpcid_supported = vmx_invpcid_supported,
12363 .set_supported_cpuid = vmx_set_supported_cpuid,
12365 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
12367 .write_tsc_offset = vmx_write_tsc_offset,
12369 .set_tdp_cr3 = vmx_set_cr3,
12371 .check_intercept = vmx_check_intercept,
12372 .handle_external_intr = vmx_handle_external_intr,
12373 .mpx_supported = vmx_mpx_supported,
12374 .xsaves_supported = vmx_xsaves_supported,
12375 .umip_emulated = vmx_umip_emulated,
12377 .check_nested_events = vmx_check_nested_events,
12379 .sched_in = vmx_sched_in,
12381 .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
12382 .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
12383 .flush_log_dirty = vmx_flush_log_dirty,
12384 .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
12385 .write_log_dirty = vmx_write_pml_buffer,
12387 .pre_block = vmx_pre_block,
12388 .post_block = vmx_post_block,
12390 .pmu_ops = &intel_pmu_ops,
12392 .update_pi_irte = vmx_update_pi_irte,
12394 #ifdef CONFIG_X86_64
12395 .set_hv_timer = vmx_set_hv_timer,
12396 .cancel_hv_timer = vmx_cancel_hv_timer,
12399 .setup_mce = vmx_setup_mce,
12401 .smi_allowed = vmx_smi_allowed,
12402 .pre_enter_smm = vmx_pre_enter_smm,
12403 .pre_leave_smm = vmx_pre_leave_smm,
12404 .enable_smi_window = enable_smi_window,
12407 static int __init vmx_init(void)
12409 int r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
12410 __alignof__(struct vcpu_vmx), THIS_MODULE);
12414 #ifdef CONFIG_KEXEC_CORE
12415 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
12416 crash_vmclear_local_loaded_vmcss);
12422 static void __exit vmx_exit(void)
12424 #ifdef CONFIG_KEXEC_CORE
12425 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
12432 module_init(vmx_init)
12433 module_exit(vmx_exit)