2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
24 #include <linux/kvm_host.h>
25 #include <linux/module.h>
26 #include <linux/kernel.h>
28 #include <linux/highmem.h>
29 #include <linux/sched.h>
30 #include <linux/moduleparam.h>
31 #include <linux/mod_devicetable.h>
32 #include <linux/trace_events.h>
33 #include <linux/slab.h>
34 #include <linux/tboot.h>
35 #include <linux/hrtimer.h>
36 #include "kvm_cache_regs.h"
43 #include <asm/virtext.h>
45 #include <asm/fpu/internal.h>
46 #include <asm/perf_event.h>
47 #include <asm/debugreg.h>
48 #include <asm/kexec.h>
50 #include <asm/irq_remapping.h>
55 #define __ex(x) __kvm_handle_fault_on_reboot(x)
56 #define __ex_clear(x, reg) \
57 ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
59 MODULE_AUTHOR("Qumranet");
60 MODULE_LICENSE("GPL");
62 static const struct x86_cpu_id vmx_cpu_id[] = {
63 X86_FEATURE_MATCH(X86_FEATURE_VMX),
66 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
68 static bool __read_mostly enable_vpid = 1;
69 module_param_named(vpid, enable_vpid, bool, 0444);
71 static bool __read_mostly flexpriority_enabled = 1;
72 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
74 static bool __read_mostly enable_ept = 1;
75 module_param_named(ept, enable_ept, bool, S_IRUGO);
77 static bool __read_mostly enable_unrestricted_guest = 1;
78 module_param_named(unrestricted_guest,
79 enable_unrestricted_guest, bool, S_IRUGO);
81 static bool __read_mostly enable_ept_ad_bits = 1;
82 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
84 static bool __read_mostly emulate_invalid_guest_state = true;
85 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
87 static bool __read_mostly vmm_exclusive = 1;
88 module_param(vmm_exclusive, bool, S_IRUGO);
90 static bool __read_mostly fasteoi = 1;
91 module_param(fasteoi, bool, S_IRUGO);
93 static bool __read_mostly enable_apicv = 1;
94 module_param(enable_apicv, bool, S_IRUGO);
96 static bool __read_mostly enable_shadow_vmcs = 1;
97 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
99 * If nested=1, nested virtualization is supported, i.e., guests may use
100 * VMX and be a hypervisor for its own guests. If nested=0, guests may not
101 * use VMX instructions.
103 static bool __read_mostly nested = 0;
104 module_param(nested, bool, S_IRUGO);
106 static u64 __read_mostly host_xss;
108 static bool __read_mostly enable_pml = 1;
109 module_param_named(pml, enable_pml, bool, S_IRUGO);
111 #define KVM_VMX_TSC_MULTIPLIER_MAX 0xffffffffffffffffULL
113 #define KVM_GUEST_CR0_MASK (X86_CR0_NW | X86_CR0_CD)
114 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST (X86_CR0_WP | X86_CR0_NE)
115 #define KVM_VM_CR0_ALWAYS_ON \
116 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
117 #define KVM_CR4_GUEST_OWNED_BITS \
118 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
119 | X86_CR4_OSXMMEXCPT | X86_CR4_TSD)
121 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
122 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
124 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
126 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
129 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
130 * ple_gap: upper bound on the amount of time between two successive
131 * executions of PAUSE in a loop. Also indicate if ple enabled.
132 * According to test, this time is usually smaller than 128 cycles.
133 * ple_window: upper bound on the amount of time a guest is allowed to execute
134 * in a PAUSE loop. Tests indicate that most spinlocks are held for
135 * less than 2^12 cycles
136 * Time is measured based on a counter that runs at the same rate as the TSC,
137 * refer SDM volume 3b section 21.6.13 & 22.1.3.
139 #define KVM_VMX_DEFAULT_PLE_GAP 128
140 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
141 #define KVM_VMX_DEFAULT_PLE_WINDOW_GROW 2
142 #define KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK 0
143 #define KVM_VMX_DEFAULT_PLE_WINDOW_MAX \
144 INT_MAX / KVM_VMX_DEFAULT_PLE_WINDOW_GROW
146 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
147 module_param(ple_gap, int, S_IRUGO);
149 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
150 module_param(ple_window, int, S_IRUGO);
152 /* Default doubles per-vcpu window every exit. */
153 static int ple_window_grow = KVM_VMX_DEFAULT_PLE_WINDOW_GROW;
154 module_param(ple_window_grow, int, S_IRUGO);
156 /* Default resets per-vcpu window every exit to ple_window. */
157 static int ple_window_shrink = KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK;
158 module_param(ple_window_shrink, int, S_IRUGO);
160 /* Default is to compute the maximum so we can never overflow. */
161 static int ple_window_actual_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
162 static int ple_window_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
163 module_param(ple_window_max, int, S_IRUGO);
165 extern const ulong vmx_return;
167 #define NR_AUTOLOAD_MSRS 8
168 #define VMCS02_POOL_SIZE 1
177 * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
178 * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
179 * loaded on this CPU (so we can clear them if the CPU goes down).
185 struct list_head loaded_vmcss_on_cpu_link;
188 struct shared_msr_entry {
195 * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
196 * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
197 * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
198 * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
199 * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
200 * More than one of these structures may exist, if L1 runs multiple L2 guests.
201 * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
202 * underlying hardware which will be used to run L2.
203 * This structure is packed to ensure that its layout is identical across
204 * machines (necessary for live migration).
205 * If there are changes in this struct, VMCS12_REVISION must be changed.
207 typedef u64 natural_width;
208 struct __packed vmcs12 {
209 /* According to the Intel spec, a VMCS region must start with the
210 * following two fields. Then follow implementation-specific data.
215 u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
216 u32 padding[7]; /* room for future expansion */
221 u64 vm_exit_msr_store_addr;
222 u64 vm_exit_msr_load_addr;
223 u64 vm_entry_msr_load_addr;
225 u64 virtual_apic_page_addr;
226 u64 apic_access_addr;
227 u64 posted_intr_desc_addr;
229 u64 eoi_exit_bitmap0;
230 u64 eoi_exit_bitmap1;
231 u64 eoi_exit_bitmap2;
232 u64 eoi_exit_bitmap3;
234 u64 guest_physical_address;
235 u64 vmcs_link_pointer;
236 u64 guest_ia32_debugctl;
239 u64 guest_ia32_perf_global_ctrl;
247 u64 host_ia32_perf_global_ctrl;
248 u64 padding64[8]; /* room for future expansion */
250 * To allow migration of L1 (complete with its L2 guests) between
251 * machines of different natural widths (32 or 64 bit), we cannot have
252 * unsigned long fields with no explict size. We use u64 (aliased
253 * natural_width) instead. Luckily, x86 is little-endian.
255 natural_width cr0_guest_host_mask;
256 natural_width cr4_guest_host_mask;
257 natural_width cr0_read_shadow;
258 natural_width cr4_read_shadow;
259 natural_width cr3_target_value0;
260 natural_width cr3_target_value1;
261 natural_width cr3_target_value2;
262 natural_width cr3_target_value3;
263 natural_width exit_qualification;
264 natural_width guest_linear_address;
265 natural_width guest_cr0;
266 natural_width guest_cr3;
267 natural_width guest_cr4;
268 natural_width guest_es_base;
269 natural_width guest_cs_base;
270 natural_width guest_ss_base;
271 natural_width guest_ds_base;
272 natural_width guest_fs_base;
273 natural_width guest_gs_base;
274 natural_width guest_ldtr_base;
275 natural_width guest_tr_base;
276 natural_width guest_gdtr_base;
277 natural_width guest_idtr_base;
278 natural_width guest_dr7;
279 natural_width guest_rsp;
280 natural_width guest_rip;
281 natural_width guest_rflags;
282 natural_width guest_pending_dbg_exceptions;
283 natural_width guest_sysenter_esp;
284 natural_width guest_sysenter_eip;
285 natural_width host_cr0;
286 natural_width host_cr3;
287 natural_width host_cr4;
288 natural_width host_fs_base;
289 natural_width host_gs_base;
290 natural_width host_tr_base;
291 natural_width host_gdtr_base;
292 natural_width host_idtr_base;
293 natural_width host_ia32_sysenter_esp;
294 natural_width host_ia32_sysenter_eip;
295 natural_width host_rsp;
296 natural_width host_rip;
297 natural_width paddingl[8]; /* room for future expansion */
298 u32 pin_based_vm_exec_control;
299 u32 cpu_based_vm_exec_control;
300 u32 exception_bitmap;
301 u32 page_fault_error_code_mask;
302 u32 page_fault_error_code_match;
303 u32 cr3_target_count;
304 u32 vm_exit_controls;
305 u32 vm_exit_msr_store_count;
306 u32 vm_exit_msr_load_count;
307 u32 vm_entry_controls;
308 u32 vm_entry_msr_load_count;
309 u32 vm_entry_intr_info_field;
310 u32 vm_entry_exception_error_code;
311 u32 vm_entry_instruction_len;
313 u32 secondary_vm_exec_control;
314 u32 vm_instruction_error;
316 u32 vm_exit_intr_info;
317 u32 vm_exit_intr_error_code;
318 u32 idt_vectoring_info_field;
319 u32 idt_vectoring_error_code;
320 u32 vm_exit_instruction_len;
321 u32 vmx_instruction_info;
328 u32 guest_ldtr_limit;
330 u32 guest_gdtr_limit;
331 u32 guest_idtr_limit;
332 u32 guest_es_ar_bytes;
333 u32 guest_cs_ar_bytes;
334 u32 guest_ss_ar_bytes;
335 u32 guest_ds_ar_bytes;
336 u32 guest_fs_ar_bytes;
337 u32 guest_gs_ar_bytes;
338 u32 guest_ldtr_ar_bytes;
339 u32 guest_tr_ar_bytes;
340 u32 guest_interruptibility_info;
341 u32 guest_activity_state;
342 u32 guest_sysenter_cs;
343 u32 host_ia32_sysenter_cs;
344 u32 vmx_preemption_timer_value;
345 u32 padding32[7]; /* room for future expansion */
346 u16 virtual_processor_id;
348 u16 guest_es_selector;
349 u16 guest_cs_selector;
350 u16 guest_ss_selector;
351 u16 guest_ds_selector;
352 u16 guest_fs_selector;
353 u16 guest_gs_selector;
354 u16 guest_ldtr_selector;
355 u16 guest_tr_selector;
356 u16 guest_intr_status;
357 u16 host_es_selector;
358 u16 host_cs_selector;
359 u16 host_ss_selector;
360 u16 host_ds_selector;
361 u16 host_fs_selector;
362 u16 host_gs_selector;
363 u16 host_tr_selector;
367 * VMCS12_REVISION is an arbitrary id that should be changed if the content or
368 * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
369 * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
371 #define VMCS12_REVISION 0x11e57ed0
374 * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
375 * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
376 * current implementation, 4K are reserved to avoid future complications.
378 #define VMCS12_SIZE 0x1000
380 /* Used to remember the last vmcs02 used for some recently used vmcs12s */
382 struct list_head list;
384 struct loaded_vmcs vmcs02;
388 * The nested_vmx structure is part of vcpu_vmx, and holds information we need
389 * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
392 /* Has the level1 guest done vmxon? */
396 /* The guest-physical address of the current VMCS L1 keeps for L2 */
398 /* The host-usable pointer to the above */
399 struct page *current_vmcs12_page;
400 struct vmcs12 *current_vmcs12;
401 struct vmcs *current_shadow_vmcs;
403 * Indicates if the shadow vmcs must be updated with the
404 * data hold by vmcs12
406 bool sync_shadow_vmcs;
408 /* vmcs02_list cache of VMCSs recently used to run L2 guests */
409 struct list_head vmcs02_pool;
411 u64 vmcs01_tsc_offset;
412 /* L2 must run next, and mustn't decide to exit to L1. */
413 bool nested_run_pending;
415 * Guest pages referred to in vmcs02 with host-physical pointers, so
416 * we must keep them pinned while L2 runs.
418 struct page *apic_access_page;
419 struct page *virtual_apic_page;
420 struct page *pi_desc_page;
421 struct pi_desc *pi_desc;
424 u64 msr_ia32_feature_control;
426 struct hrtimer preemption_timer;
427 bool preemption_timer_expired;
429 /* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
435 u32 nested_vmx_procbased_ctls_low;
436 u32 nested_vmx_procbased_ctls_high;
437 u32 nested_vmx_true_procbased_ctls_low;
438 u32 nested_vmx_secondary_ctls_low;
439 u32 nested_vmx_secondary_ctls_high;
440 u32 nested_vmx_pinbased_ctls_low;
441 u32 nested_vmx_pinbased_ctls_high;
442 u32 nested_vmx_exit_ctls_low;
443 u32 nested_vmx_exit_ctls_high;
444 u32 nested_vmx_true_exit_ctls_low;
445 u32 nested_vmx_entry_ctls_low;
446 u32 nested_vmx_entry_ctls_high;
447 u32 nested_vmx_true_entry_ctls_low;
448 u32 nested_vmx_misc_low;
449 u32 nested_vmx_misc_high;
450 u32 nested_vmx_ept_caps;
451 u32 nested_vmx_vpid_caps;
454 #define POSTED_INTR_ON 0
455 #define POSTED_INTR_SN 1
457 /* Posted-Interrupt Descriptor */
459 u32 pir[8]; /* Posted interrupt requested */
462 /* bit 256 - Outstanding Notification */
464 /* bit 257 - Suppress Notification */
466 /* bit 271:258 - Reserved */
468 /* bit 279:272 - Notification Vector */
470 /* bit 287:280 - Reserved */
472 /* bit 319:288 - Notification Destination */
480 static bool pi_test_and_set_on(struct pi_desc *pi_desc)
482 return test_and_set_bit(POSTED_INTR_ON,
483 (unsigned long *)&pi_desc->control);
486 static bool pi_test_and_clear_on(struct pi_desc *pi_desc)
488 return test_and_clear_bit(POSTED_INTR_ON,
489 (unsigned long *)&pi_desc->control);
492 static int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
494 return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
497 static inline void pi_clear_sn(struct pi_desc *pi_desc)
499 return clear_bit(POSTED_INTR_SN,
500 (unsigned long *)&pi_desc->control);
503 static inline void pi_set_sn(struct pi_desc *pi_desc)
505 return set_bit(POSTED_INTR_SN,
506 (unsigned long *)&pi_desc->control);
509 static inline int pi_test_on(struct pi_desc *pi_desc)
511 return test_bit(POSTED_INTR_ON,
512 (unsigned long *)&pi_desc->control);
515 static inline int pi_test_sn(struct pi_desc *pi_desc)
517 return test_bit(POSTED_INTR_SN,
518 (unsigned long *)&pi_desc->control);
522 struct kvm_vcpu vcpu;
523 unsigned long host_rsp;
525 bool nmi_known_unmasked;
527 u32 idt_vectoring_info;
529 struct shared_msr_entry *guest_msrs;
532 unsigned long host_idt_base;
534 u64 msr_host_kernel_gs_base;
535 u64 msr_guest_kernel_gs_base;
537 u32 vm_entry_controls_shadow;
538 u32 vm_exit_controls_shadow;
540 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
541 * non-nested (L1) guest, it always points to vmcs01. For a nested
542 * guest (L2), it points to a different VMCS.
544 struct loaded_vmcs vmcs01;
545 struct loaded_vmcs *loaded_vmcs;
546 bool __launched; /* temporary, used in vmx_vcpu_run */
547 struct msr_autoload {
549 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
550 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
554 u16 fs_sel, gs_sel, ldt_sel;
558 int gs_ldt_reload_needed;
559 int fs_reload_needed;
560 u64 msr_host_bndcfgs;
561 unsigned long vmcs_host_cr4; /* May not match real cr4 */
566 struct kvm_segment segs[8];
569 u32 bitmask; /* 4 bits per segment (1 bit per field) */
570 struct kvm_save_segment {
578 bool emulation_required;
580 /* Support for vnmi-less CPUs */
581 int soft_vnmi_blocked;
583 s64 vnmi_blocked_time;
586 /* Posted interrupt descriptor */
587 struct pi_desc pi_desc;
589 /* Support for a guest hypervisor (nested VMX) */
590 struct nested_vmx nested;
592 /* Dynamic PLE window. */
594 bool ple_window_dirty;
596 /* Support for PML */
597 #define PML_ENTITY_NUM 512
600 u64 current_tsc_ratio;
603 enum segment_cache_field {
612 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
614 return container_of(vcpu, struct vcpu_vmx, vcpu);
617 static struct pi_desc *vcpu_to_pi_desc(struct kvm_vcpu *vcpu)
619 return &(to_vmx(vcpu)->pi_desc);
622 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
623 #define FIELD(number, name) [number] = VMCS12_OFFSET(name)
624 #define FIELD64(number, name) [number] = VMCS12_OFFSET(name), \
625 [number##_HIGH] = VMCS12_OFFSET(name)+4
628 static unsigned long shadow_read_only_fields[] = {
630 * We do NOT shadow fields that are modified when L0
631 * traps and emulates any vmx instruction (e.g. VMPTRLD,
632 * VMXON...) executed by L1.
633 * For example, VM_INSTRUCTION_ERROR is read
634 * by L1 if a vmx instruction fails (part of the error path).
635 * Note the code assumes this logic. If for some reason
636 * we start shadowing these fields then we need to
637 * force a shadow sync when L0 emulates vmx instructions
638 * (e.g. force a sync if VM_INSTRUCTION_ERROR is modified
639 * by nested_vmx_failValid)
643 VM_EXIT_INSTRUCTION_LEN,
644 IDT_VECTORING_INFO_FIELD,
645 IDT_VECTORING_ERROR_CODE,
646 VM_EXIT_INTR_ERROR_CODE,
648 GUEST_LINEAR_ADDRESS,
649 GUEST_PHYSICAL_ADDRESS
651 static int max_shadow_read_only_fields =
652 ARRAY_SIZE(shadow_read_only_fields);
654 static unsigned long shadow_read_write_fields[] = {
661 GUEST_INTERRUPTIBILITY_INFO,
674 CPU_BASED_VM_EXEC_CONTROL,
675 VM_ENTRY_EXCEPTION_ERROR_CODE,
676 VM_ENTRY_INTR_INFO_FIELD,
677 VM_ENTRY_INSTRUCTION_LEN,
678 VM_ENTRY_EXCEPTION_ERROR_CODE,
684 static int max_shadow_read_write_fields =
685 ARRAY_SIZE(shadow_read_write_fields);
687 static const unsigned short vmcs_field_to_offset_table[] = {
688 FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
689 FIELD(POSTED_INTR_NV, posted_intr_nv),
690 FIELD(GUEST_ES_SELECTOR, guest_es_selector),
691 FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
692 FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
693 FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
694 FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
695 FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
696 FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
697 FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
698 FIELD(GUEST_INTR_STATUS, guest_intr_status),
699 FIELD(HOST_ES_SELECTOR, host_es_selector),
700 FIELD(HOST_CS_SELECTOR, host_cs_selector),
701 FIELD(HOST_SS_SELECTOR, host_ss_selector),
702 FIELD(HOST_DS_SELECTOR, host_ds_selector),
703 FIELD(HOST_FS_SELECTOR, host_fs_selector),
704 FIELD(HOST_GS_SELECTOR, host_gs_selector),
705 FIELD(HOST_TR_SELECTOR, host_tr_selector),
706 FIELD64(IO_BITMAP_A, io_bitmap_a),
707 FIELD64(IO_BITMAP_B, io_bitmap_b),
708 FIELD64(MSR_BITMAP, msr_bitmap),
709 FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
710 FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
711 FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
712 FIELD64(TSC_OFFSET, tsc_offset),
713 FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
714 FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
715 FIELD64(POSTED_INTR_DESC_ADDR, posted_intr_desc_addr),
716 FIELD64(EPT_POINTER, ept_pointer),
717 FIELD64(EOI_EXIT_BITMAP0, eoi_exit_bitmap0),
718 FIELD64(EOI_EXIT_BITMAP1, eoi_exit_bitmap1),
719 FIELD64(EOI_EXIT_BITMAP2, eoi_exit_bitmap2),
720 FIELD64(EOI_EXIT_BITMAP3, eoi_exit_bitmap3),
721 FIELD64(XSS_EXIT_BITMAP, xss_exit_bitmap),
722 FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
723 FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
724 FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
725 FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
726 FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
727 FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
728 FIELD64(GUEST_PDPTR0, guest_pdptr0),
729 FIELD64(GUEST_PDPTR1, guest_pdptr1),
730 FIELD64(GUEST_PDPTR2, guest_pdptr2),
731 FIELD64(GUEST_PDPTR3, guest_pdptr3),
732 FIELD64(GUEST_BNDCFGS, guest_bndcfgs),
733 FIELD64(HOST_IA32_PAT, host_ia32_pat),
734 FIELD64(HOST_IA32_EFER, host_ia32_efer),
735 FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
736 FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
737 FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
738 FIELD(EXCEPTION_BITMAP, exception_bitmap),
739 FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
740 FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
741 FIELD(CR3_TARGET_COUNT, cr3_target_count),
742 FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
743 FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
744 FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
745 FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
746 FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
747 FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
748 FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
749 FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
750 FIELD(TPR_THRESHOLD, tpr_threshold),
751 FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
752 FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
753 FIELD(VM_EXIT_REASON, vm_exit_reason),
754 FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
755 FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
756 FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
757 FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
758 FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
759 FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
760 FIELD(GUEST_ES_LIMIT, guest_es_limit),
761 FIELD(GUEST_CS_LIMIT, guest_cs_limit),
762 FIELD(GUEST_SS_LIMIT, guest_ss_limit),
763 FIELD(GUEST_DS_LIMIT, guest_ds_limit),
764 FIELD(GUEST_FS_LIMIT, guest_fs_limit),
765 FIELD(GUEST_GS_LIMIT, guest_gs_limit),
766 FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
767 FIELD(GUEST_TR_LIMIT, guest_tr_limit),
768 FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
769 FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
770 FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
771 FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
772 FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
773 FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
774 FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
775 FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
776 FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
777 FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
778 FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
779 FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
780 FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
781 FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
782 FIELD(VMX_PREEMPTION_TIMER_VALUE, vmx_preemption_timer_value),
783 FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
784 FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
785 FIELD(CR0_READ_SHADOW, cr0_read_shadow),
786 FIELD(CR4_READ_SHADOW, cr4_read_shadow),
787 FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
788 FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
789 FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
790 FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
791 FIELD(EXIT_QUALIFICATION, exit_qualification),
792 FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
793 FIELD(GUEST_CR0, guest_cr0),
794 FIELD(GUEST_CR3, guest_cr3),
795 FIELD(GUEST_CR4, guest_cr4),
796 FIELD(GUEST_ES_BASE, guest_es_base),
797 FIELD(GUEST_CS_BASE, guest_cs_base),
798 FIELD(GUEST_SS_BASE, guest_ss_base),
799 FIELD(GUEST_DS_BASE, guest_ds_base),
800 FIELD(GUEST_FS_BASE, guest_fs_base),
801 FIELD(GUEST_GS_BASE, guest_gs_base),
802 FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
803 FIELD(GUEST_TR_BASE, guest_tr_base),
804 FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
805 FIELD(GUEST_IDTR_BASE, guest_idtr_base),
806 FIELD(GUEST_DR7, guest_dr7),
807 FIELD(GUEST_RSP, guest_rsp),
808 FIELD(GUEST_RIP, guest_rip),
809 FIELD(GUEST_RFLAGS, guest_rflags),
810 FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
811 FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
812 FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
813 FIELD(HOST_CR0, host_cr0),
814 FIELD(HOST_CR3, host_cr3),
815 FIELD(HOST_CR4, host_cr4),
816 FIELD(HOST_FS_BASE, host_fs_base),
817 FIELD(HOST_GS_BASE, host_gs_base),
818 FIELD(HOST_TR_BASE, host_tr_base),
819 FIELD(HOST_GDTR_BASE, host_gdtr_base),
820 FIELD(HOST_IDTR_BASE, host_idtr_base),
821 FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
822 FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
823 FIELD(HOST_RSP, host_rsp),
824 FIELD(HOST_RIP, host_rip),
827 static inline short vmcs_field_to_offset(unsigned long field)
829 BUILD_BUG_ON(ARRAY_SIZE(vmcs_field_to_offset_table) > SHRT_MAX);
831 if (field >= ARRAY_SIZE(vmcs_field_to_offset_table) ||
832 vmcs_field_to_offset_table[field] == 0)
835 return vmcs_field_to_offset_table[field];
838 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
840 return to_vmx(vcpu)->nested.current_vmcs12;
843 static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
845 struct page *page = kvm_vcpu_gfn_to_page(vcpu, addr >> PAGE_SHIFT);
846 if (is_error_page(page))
852 static void nested_release_page(struct page *page)
854 kvm_release_page_dirty(page);
857 static void nested_release_page_clean(struct page *page)
859 kvm_release_page_clean(page);
862 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu);
863 static u64 construct_eptp(unsigned long root_hpa);
864 static void kvm_cpu_vmxon(u64 addr);
865 static void kvm_cpu_vmxoff(void);
866 static bool vmx_mpx_supported(void);
867 static bool vmx_xsaves_supported(void);
868 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
869 static void vmx_set_segment(struct kvm_vcpu *vcpu,
870 struct kvm_segment *var, int seg);
871 static void vmx_get_segment(struct kvm_vcpu *vcpu,
872 struct kvm_segment *var, int seg);
873 static bool guest_state_valid(struct kvm_vcpu *vcpu);
874 static u32 vmx_segment_access_rights(struct kvm_segment *var);
875 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx);
876 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx);
877 static int alloc_identity_pagetable(struct kvm *kvm);
879 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
880 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
882 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
883 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
885 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
886 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
889 * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
890 * can find which vCPU should be waken up.
892 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
893 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
895 static unsigned long *vmx_io_bitmap_a;
896 static unsigned long *vmx_io_bitmap_b;
897 static unsigned long *vmx_msr_bitmap_legacy;
898 static unsigned long *vmx_msr_bitmap_longmode;
899 static unsigned long *vmx_msr_bitmap_legacy_x2apic;
900 static unsigned long *vmx_msr_bitmap_longmode_x2apic;
901 static unsigned long *vmx_msr_bitmap_nested;
902 static unsigned long *vmx_vmread_bitmap;
903 static unsigned long *vmx_vmwrite_bitmap;
905 static bool cpu_has_load_ia32_efer;
906 static bool cpu_has_load_perf_global_ctrl;
908 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
909 static DEFINE_SPINLOCK(vmx_vpid_lock);
911 static struct vmcs_config {
915 u32 pin_based_exec_ctrl;
916 u32 cpu_based_exec_ctrl;
917 u32 cpu_based_2nd_exec_ctrl;
922 static struct vmx_capability {
927 #define VMX_SEGMENT_FIELD(seg) \
928 [VCPU_SREG_##seg] = { \
929 .selector = GUEST_##seg##_SELECTOR, \
930 .base = GUEST_##seg##_BASE, \
931 .limit = GUEST_##seg##_LIMIT, \
932 .ar_bytes = GUEST_##seg##_AR_BYTES, \
935 static const struct kvm_vmx_segment_field {
940 } kvm_vmx_segment_fields[] = {
941 VMX_SEGMENT_FIELD(CS),
942 VMX_SEGMENT_FIELD(DS),
943 VMX_SEGMENT_FIELD(ES),
944 VMX_SEGMENT_FIELD(FS),
945 VMX_SEGMENT_FIELD(GS),
946 VMX_SEGMENT_FIELD(SS),
947 VMX_SEGMENT_FIELD(TR),
948 VMX_SEGMENT_FIELD(LDTR),
951 static u64 host_efer;
953 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
956 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
957 * away by decrementing the array size.
959 static const u32 vmx_msr_index[] = {
961 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
963 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
966 static inline bool is_page_fault(u32 intr_info)
968 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
969 INTR_INFO_VALID_MASK)) ==
970 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
973 static inline bool is_no_device(u32 intr_info)
975 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
976 INTR_INFO_VALID_MASK)) ==
977 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
980 static inline bool is_invalid_opcode(u32 intr_info)
982 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
983 INTR_INFO_VALID_MASK)) ==
984 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
987 static inline bool is_external_interrupt(u32 intr_info)
989 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
990 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
993 static inline bool is_machine_check(u32 intr_info)
995 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
996 INTR_INFO_VALID_MASK)) ==
997 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
1000 static inline bool cpu_has_vmx_msr_bitmap(void)
1002 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
1005 static inline bool cpu_has_vmx_tpr_shadow(void)
1007 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
1010 static inline bool cpu_need_tpr_shadow(struct kvm_vcpu *vcpu)
1012 return cpu_has_vmx_tpr_shadow() && lapic_in_kernel(vcpu);
1015 static inline bool cpu_has_secondary_exec_ctrls(void)
1017 return vmcs_config.cpu_based_exec_ctrl &
1018 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1021 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
1023 return vmcs_config.cpu_based_2nd_exec_ctrl &
1024 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1027 static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
1029 return vmcs_config.cpu_based_2nd_exec_ctrl &
1030 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
1033 static inline bool cpu_has_vmx_apic_register_virt(void)
1035 return vmcs_config.cpu_based_2nd_exec_ctrl &
1036 SECONDARY_EXEC_APIC_REGISTER_VIRT;
1039 static inline bool cpu_has_vmx_virtual_intr_delivery(void)
1041 return vmcs_config.cpu_based_2nd_exec_ctrl &
1042 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
1045 static inline bool cpu_has_vmx_posted_intr(void)
1047 return IS_ENABLED(CONFIG_X86_LOCAL_APIC) &&
1048 vmcs_config.pin_based_exec_ctrl & PIN_BASED_POSTED_INTR;
1051 static inline bool cpu_has_vmx_apicv(void)
1053 return cpu_has_vmx_apic_register_virt() &&
1054 cpu_has_vmx_virtual_intr_delivery() &&
1055 cpu_has_vmx_posted_intr();
1058 static inline bool cpu_has_vmx_flexpriority(void)
1060 return cpu_has_vmx_tpr_shadow() &&
1061 cpu_has_vmx_virtualize_apic_accesses();
1064 static inline bool cpu_has_vmx_ept_execute_only(void)
1066 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
1069 static inline bool cpu_has_vmx_ept_2m_page(void)
1071 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
1074 static inline bool cpu_has_vmx_ept_1g_page(void)
1076 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
1079 static inline bool cpu_has_vmx_ept_4levels(void)
1081 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
1084 static inline bool cpu_has_vmx_ept_ad_bits(void)
1086 return vmx_capability.ept & VMX_EPT_AD_BIT;
1089 static inline bool cpu_has_vmx_invept_context(void)
1091 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
1094 static inline bool cpu_has_vmx_invept_global(void)
1096 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
1099 static inline bool cpu_has_vmx_invvpid_single(void)
1101 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
1104 static inline bool cpu_has_vmx_invvpid_global(void)
1106 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
1109 static inline bool cpu_has_vmx_ept(void)
1111 return vmcs_config.cpu_based_2nd_exec_ctrl &
1112 SECONDARY_EXEC_ENABLE_EPT;
1115 static inline bool cpu_has_vmx_unrestricted_guest(void)
1117 return vmcs_config.cpu_based_2nd_exec_ctrl &
1118 SECONDARY_EXEC_UNRESTRICTED_GUEST;
1121 static inline bool cpu_has_vmx_ple(void)
1123 return vmcs_config.cpu_based_2nd_exec_ctrl &
1124 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
1127 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
1129 return flexpriority_enabled && lapic_in_kernel(vcpu);
1132 static inline bool cpu_has_vmx_vpid(void)
1134 return vmcs_config.cpu_based_2nd_exec_ctrl &
1135 SECONDARY_EXEC_ENABLE_VPID;
1138 static inline bool cpu_has_vmx_rdtscp(void)
1140 return vmcs_config.cpu_based_2nd_exec_ctrl &
1141 SECONDARY_EXEC_RDTSCP;
1144 static inline bool cpu_has_vmx_invpcid(void)
1146 return vmcs_config.cpu_based_2nd_exec_ctrl &
1147 SECONDARY_EXEC_ENABLE_INVPCID;
1150 static inline bool cpu_has_virtual_nmis(void)
1152 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
1155 static inline bool cpu_has_vmx_wbinvd_exit(void)
1157 return vmcs_config.cpu_based_2nd_exec_ctrl &
1158 SECONDARY_EXEC_WBINVD_EXITING;
1161 static inline bool cpu_has_vmx_shadow_vmcs(void)
1164 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
1165 /* check if the cpu supports writing r/o exit information fields */
1166 if (!(vmx_msr & MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS))
1169 return vmcs_config.cpu_based_2nd_exec_ctrl &
1170 SECONDARY_EXEC_SHADOW_VMCS;
1173 static inline bool cpu_has_vmx_pml(void)
1175 return vmcs_config.cpu_based_2nd_exec_ctrl & SECONDARY_EXEC_ENABLE_PML;
1178 static inline bool cpu_has_vmx_tsc_scaling(void)
1180 return vmcs_config.cpu_based_2nd_exec_ctrl &
1181 SECONDARY_EXEC_TSC_SCALING;
1184 static inline bool report_flexpriority(void)
1186 return flexpriority_enabled;
1189 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
1191 return vmcs12->cpu_based_vm_exec_control & bit;
1194 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
1196 return (vmcs12->cpu_based_vm_exec_control &
1197 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
1198 (vmcs12->secondary_vm_exec_control & bit);
1201 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12)
1203 return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
1206 static inline bool nested_cpu_has_preemption_timer(struct vmcs12 *vmcs12)
1208 return vmcs12->pin_based_vm_exec_control &
1209 PIN_BASED_VMX_PREEMPTION_TIMER;
1212 static inline int nested_cpu_has_ept(struct vmcs12 *vmcs12)
1214 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_EPT);
1217 static inline bool nested_cpu_has_xsaves(struct vmcs12 *vmcs12)
1219 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES) &&
1220 vmx_xsaves_supported();
1223 static inline bool nested_cpu_has_virt_x2apic_mode(struct vmcs12 *vmcs12)
1225 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
1228 static inline bool nested_cpu_has_vpid(struct vmcs12 *vmcs12)
1230 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_VPID);
1233 static inline bool nested_cpu_has_apic_reg_virt(struct vmcs12 *vmcs12)
1235 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_APIC_REGISTER_VIRT);
1238 static inline bool nested_cpu_has_vid(struct vmcs12 *vmcs12)
1240 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
1243 static inline bool nested_cpu_has_posted_intr(struct vmcs12 *vmcs12)
1245 return vmcs12->pin_based_vm_exec_control & PIN_BASED_POSTED_INTR;
1248 static inline bool is_exception(u32 intr_info)
1250 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1251 == (INTR_TYPE_HARD_EXCEPTION | INTR_INFO_VALID_MASK);
1254 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
1256 unsigned long exit_qualification);
1257 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
1258 struct vmcs12 *vmcs12,
1259 u32 reason, unsigned long qualification);
1261 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
1265 for (i = 0; i < vmx->nmsrs; ++i)
1266 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
1271 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
1277 } operand = { vpid, 0, gva };
1279 asm volatile (__ex(ASM_VMX_INVVPID)
1280 /* CF==1 or ZF==1 --> rc = -1 */
1281 "; ja 1f ; ud2 ; 1:"
1282 : : "a"(&operand), "c"(ext) : "cc", "memory");
1285 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
1289 } operand = {eptp, gpa};
1291 asm volatile (__ex(ASM_VMX_INVEPT)
1292 /* CF==1 or ZF==1 --> rc = -1 */
1293 "; ja 1f ; ud2 ; 1:\n"
1294 : : "a" (&operand), "c" (ext) : "cc", "memory");
1297 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
1301 i = __find_msr_index(vmx, msr);
1303 return &vmx->guest_msrs[i];
1307 static void vmcs_clear(struct vmcs *vmcs)
1309 u64 phys_addr = __pa(vmcs);
1312 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
1313 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1316 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
1320 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
1322 vmcs_clear(loaded_vmcs->vmcs);
1323 loaded_vmcs->cpu = -1;
1324 loaded_vmcs->launched = 0;
1327 static void vmcs_load(struct vmcs *vmcs)
1329 u64 phys_addr = __pa(vmcs);
1332 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
1333 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1336 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
1340 #ifdef CONFIG_KEXEC_CORE
1342 * This bitmap is used to indicate whether the vmclear
1343 * operation is enabled on all cpus. All disabled by
1346 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
1348 static inline void crash_enable_local_vmclear(int cpu)
1350 cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
1353 static inline void crash_disable_local_vmclear(int cpu)
1355 cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
1358 static inline int crash_local_vmclear_enabled(int cpu)
1360 return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
1363 static void crash_vmclear_local_loaded_vmcss(void)
1365 int cpu = raw_smp_processor_id();
1366 struct loaded_vmcs *v;
1368 if (!crash_local_vmclear_enabled(cpu))
1371 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
1372 loaded_vmcss_on_cpu_link)
1373 vmcs_clear(v->vmcs);
1376 static inline void crash_enable_local_vmclear(int cpu) { }
1377 static inline void crash_disable_local_vmclear(int cpu) { }
1378 #endif /* CONFIG_KEXEC_CORE */
1380 static void __loaded_vmcs_clear(void *arg)
1382 struct loaded_vmcs *loaded_vmcs = arg;
1383 int cpu = raw_smp_processor_id();
1385 if (loaded_vmcs->cpu != cpu)
1386 return; /* vcpu migration can race with cpu offline */
1387 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1388 per_cpu(current_vmcs, cpu) = NULL;
1389 crash_disable_local_vmclear(cpu);
1390 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1393 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1394 * is before setting loaded_vmcs->vcpu to -1 which is done in
1395 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1396 * then adds the vmcs into percpu list before it is deleted.
1400 loaded_vmcs_init(loaded_vmcs);
1401 crash_enable_local_vmclear(cpu);
1404 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1406 int cpu = loaded_vmcs->cpu;
1409 smp_call_function_single(cpu,
1410 __loaded_vmcs_clear, loaded_vmcs, 1);
1413 static inline void vpid_sync_vcpu_single(int vpid)
1418 if (cpu_has_vmx_invvpid_single())
1419 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vpid, 0);
1422 static inline void vpid_sync_vcpu_global(void)
1424 if (cpu_has_vmx_invvpid_global())
1425 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1428 static inline void vpid_sync_context(int vpid)
1430 if (cpu_has_vmx_invvpid_single())
1431 vpid_sync_vcpu_single(vpid);
1433 vpid_sync_vcpu_global();
1436 static inline void ept_sync_global(void)
1438 if (cpu_has_vmx_invept_global())
1439 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1442 static inline void ept_sync_context(u64 eptp)
1445 if (cpu_has_vmx_invept_context())
1446 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1452 static __always_inline void vmcs_check16(unsigned long field)
1454 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1455 "16-bit accessor invalid for 64-bit field");
1456 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1457 "16-bit accessor invalid for 64-bit high field");
1458 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1459 "16-bit accessor invalid for 32-bit high field");
1460 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1461 "16-bit accessor invalid for natural width field");
1464 static __always_inline void vmcs_check32(unsigned long field)
1466 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1467 "32-bit accessor invalid for 16-bit field");
1468 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1469 "32-bit accessor invalid for natural width field");
1472 static __always_inline void vmcs_check64(unsigned long field)
1474 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1475 "64-bit accessor invalid for 16-bit field");
1476 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1477 "64-bit accessor invalid for 64-bit high field");
1478 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1479 "64-bit accessor invalid for 32-bit field");
1480 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1481 "64-bit accessor invalid for natural width field");
1484 static __always_inline void vmcs_checkl(unsigned long field)
1486 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1487 "Natural width accessor invalid for 16-bit field");
1488 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1489 "Natural width accessor invalid for 64-bit field");
1490 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1491 "Natural width accessor invalid for 64-bit high field");
1492 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1493 "Natural width accessor invalid for 32-bit field");
1496 static __always_inline unsigned long __vmcs_readl(unsigned long field)
1498 unsigned long value;
1500 asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1501 : "=a"(value) : "d"(field) : "cc");
1505 static __always_inline u16 vmcs_read16(unsigned long field)
1507 vmcs_check16(field);
1508 return __vmcs_readl(field);
1511 static __always_inline u32 vmcs_read32(unsigned long field)
1513 vmcs_check32(field);
1514 return __vmcs_readl(field);
1517 static __always_inline u64 vmcs_read64(unsigned long field)
1519 vmcs_check64(field);
1520 #ifdef CONFIG_X86_64
1521 return __vmcs_readl(field);
1523 return __vmcs_readl(field) | ((u64)__vmcs_readl(field+1) << 32);
1527 static __always_inline unsigned long vmcs_readl(unsigned long field)
1530 return __vmcs_readl(field);
1533 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1535 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1536 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1540 static __always_inline void __vmcs_writel(unsigned long field, unsigned long value)
1544 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1545 : "=q"(error) : "a"(value), "d"(field) : "cc");
1546 if (unlikely(error))
1547 vmwrite_error(field, value);
1550 static __always_inline void vmcs_write16(unsigned long field, u16 value)
1552 vmcs_check16(field);
1553 __vmcs_writel(field, value);
1556 static __always_inline void vmcs_write32(unsigned long field, u32 value)
1558 vmcs_check32(field);
1559 __vmcs_writel(field, value);
1562 static __always_inline void vmcs_write64(unsigned long field, u64 value)
1564 vmcs_check64(field);
1565 __vmcs_writel(field, value);
1566 #ifndef CONFIG_X86_64
1568 __vmcs_writel(field+1, value >> 32);
1572 static __always_inline void vmcs_writel(unsigned long field, unsigned long value)
1575 __vmcs_writel(field, value);
1578 static __always_inline void vmcs_clear_bits(unsigned long field, u32 mask)
1580 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1581 "vmcs_clear_bits does not support 64-bit fields");
1582 __vmcs_writel(field, __vmcs_readl(field) & ~mask);
1585 static __always_inline void vmcs_set_bits(unsigned long field, u32 mask)
1587 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1588 "vmcs_set_bits does not support 64-bit fields");
1589 __vmcs_writel(field, __vmcs_readl(field) | mask);
1592 static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val)
1594 vmcs_write32(VM_ENTRY_CONTROLS, val);
1595 vmx->vm_entry_controls_shadow = val;
1598 static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val)
1600 if (vmx->vm_entry_controls_shadow != val)
1601 vm_entry_controls_init(vmx, val);
1604 static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx)
1606 return vmx->vm_entry_controls_shadow;
1610 static inline void vm_entry_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1612 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) | val);
1615 static inline void vm_entry_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1617 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) & ~val);
1620 static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val)
1622 vmcs_write32(VM_EXIT_CONTROLS, val);
1623 vmx->vm_exit_controls_shadow = val;
1626 static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val)
1628 if (vmx->vm_exit_controls_shadow != val)
1629 vm_exit_controls_init(vmx, val);
1632 static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx)
1634 return vmx->vm_exit_controls_shadow;
1638 static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1640 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val);
1643 static inline void vm_exit_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1645 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) & ~val);
1648 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1650 vmx->segment_cache.bitmask = 0;
1653 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1657 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1659 if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1660 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1661 vmx->segment_cache.bitmask = 0;
1663 ret = vmx->segment_cache.bitmask & mask;
1664 vmx->segment_cache.bitmask |= mask;
1668 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1670 u16 *p = &vmx->segment_cache.seg[seg].selector;
1672 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1673 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1677 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1679 ulong *p = &vmx->segment_cache.seg[seg].base;
1681 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1682 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1686 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1688 u32 *p = &vmx->segment_cache.seg[seg].limit;
1690 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1691 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1695 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1697 u32 *p = &vmx->segment_cache.seg[seg].ar;
1699 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1700 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1704 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1708 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1709 (1u << NM_VECTOR) | (1u << DB_VECTOR) | (1u << AC_VECTOR);
1710 if ((vcpu->guest_debug &
1711 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1712 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1713 eb |= 1u << BP_VECTOR;
1714 if (to_vmx(vcpu)->rmode.vm86_active)
1717 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1718 if (vcpu->fpu_active)
1719 eb &= ~(1u << NM_VECTOR);
1721 /* When we are running a nested L2 guest and L1 specified for it a
1722 * certain exception bitmap, we must trap the same exceptions and pass
1723 * them to L1. When running L2, we will only handle the exceptions
1724 * specified above if L1 did not want them.
1726 if (is_guest_mode(vcpu))
1727 eb |= get_vmcs12(vcpu)->exception_bitmap;
1729 vmcs_write32(EXCEPTION_BITMAP, eb);
1732 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1733 unsigned long entry, unsigned long exit)
1735 vm_entry_controls_clearbit(vmx, entry);
1736 vm_exit_controls_clearbit(vmx, exit);
1739 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1742 struct msr_autoload *m = &vmx->msr_autoload;
1746 if (cpu_has_load_ia32_efer) {
1747 clear_atomic_switch_msr_special(vmx,
1748 VM_ENTRY_LOAD_IA32_EFER,
1749 VM_EXIT_LOAD_IA32_EFER);
1753 case MSR_CORE_PERF_GLOBAL_CTRL:
1754 if (cpu_has_load_perf_global_ctrl) {
1755 clear_atomic_switch_msr_special(vmx,
1756 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1757 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1763 for (i = 0; i < m->nr; ++i)
1764 if (m->guest[i].index == msr)
1770 m->guest[i] = m->guest[m->nr];
1771 m->host[i] = m->host[m->nr];
1772 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1773 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1776 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1777 unsigned long entry, unsigned long exit,
1778 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
1779 u64 guest_val, u64 host_val)
1781 vmcs_write64(guest_val_vmcs, guest_val);
1782 vmcs_write64(host_val_vmcs, host_val);
1783 vm_entry_controls_setbit(vmx, entry);
1784 vm_exit_controls_setbit(vmx, exit);
1787 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1788 u64 guest_val, u64 host_val)
1791 struct msr_autoload *m = &vmx->msr_autoload;
1795 if (cpu_has_load_ia32_efer) {
1796 add_atomic_switch_msr_special(vmx,
1797 VM_ENTRY_LOAD_IA32_EFER,
1798 VM_EXIT_LOAD_IA32_EFER,
1801 guest_val, host_val);
1805 case MSR_CORE_PERF_GLOBAL_CTRL:
1806 if (cpu_has_load_perf_global_ctrl) {
1807 add_atomic_switch_msr_special(vmx,
1808 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1809 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1810 GUEST_IA32_PERF_GLOBAL_CTRL,
1811 HOST_IA32_PERF_GLOBAL_CTRL,
1812 guest_val, host_val);
1816 case MSR_IA32_PEBS_ENABLE:
1817 /* PEBS needs a quiescent period after being disabled (to write
1818 * a record). Disabling PEBS through VMX MSR swapping doesn't
1819 * provide that period, so a CPU could write host's record into
1822 wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
1825 for (i = 0; i < m->nr; ++i)
1826 if (m->guest[i].index == msr)
1829 if (i == NR_AUTOLOAD_MSRS) {
1830 printk_once(KERN_WARNING "Not enough msr switch entries. "
1831 "Can't add msr %x\n", msr);
1833 } else if (i == m->nr) {
1835 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1836 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1839 m->guest[i].index = msr;
1840 m->guest[i].value = guest_val;
1841 m->host[i].index = msr;
1842 m->host[i].value = host_val;
1845 static void reload_tss(void)
1848 * VT restores TR but not its size. Useless.
1850 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
1851 struct desc_struct *descs;
1853 descs = (void *)gdt->address;
1854 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
1858 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
1863 guest_efer = vmx->vcpu.arch.efer;
1866 * NX is emulated; LMA and LME handled by hardware; SCE meaningless
1869 ignore_bits = EFER_NX | EFER_SCE;
1870 #ifdef CONFIG_X86_64
1871 ignore_bits |= EFER_LMA | EFER_LME;
1872 /* SCE is meaningful only in long mode on Intel */
1873 if (guest_efer & EFER_LMA)
1874 ignore_bits &= ~(u64)EFER_SCE;
1876 guest_efer &= ~ignore_bits;
1877 guest_efer |= host_efer & ignore_bits;
1878 vmx->guest_msrs[efer_offset].data = guest_efer;
1879 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1881 clear_atomic_switch_msr(vmx, MSR_EFER);
1884 * On EPT, we can't emulate NX, so we must switch EFER atomically.
1885 * On CPUs that support "load IA32_EFER", always switch EFER
1886 * atomically, since it's faster than switching it manually.
1888 if (cpu_has_load_ia32_efer ||
1889 (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
1890 guest_efer = vmx->vcpu.arch.efer;
1891 if (!(guest_efer & EFER_LMA))
1892 guest_efer &= ~EFER_LME;
1893 if (guest_efer != host_efer)
1894 add_atomic_switch_msr(vmx, MSR_EFER,
1895 guest_efer, host_efer);
1902 static unsigned long segment_base(u16 selector)
1904 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
1905 struct desc_struct *d;
1906 unsigned long table_base;
1909 if (!(selector & ~3))
1912 table_base = gdt->address;
1914 if (selector & 4) { /* from ldt */
1915 u16 ldt_selector = kvm_read_ldt();
1917 if (!(ldt_selector & ~3))
1920 table_base = segment_base(ldt_selector);
1922 d = (struct desc_struct *)(table_base + (selector & ~7));
1923 v = get_desc_base(d);
1924 #ifdef CONFIG_X86_64
1925 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
1926 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
1931 static inline unsigned long kvm_read_tr_base(void)
1934 asm("str %0" : "=g"(tr));
1935 return segment_base(tr);
1938 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
1940 struct vcpu_vmx *vmx = to_vmx(vcpu);
1943 if (vmx->host_state.loaded)
1946 vmx->host_state.loaded = 1;
1948 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
1949 * allow segment selectors with cpl > 0 or ti == 1.
1951 vmx->host_state.ldt_sel = kvm_read_ldt();
1952 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
1953 savesegment(fs, vmx->host_state.fs_sel);
1954 if (!(vmx->host_state.fs_sel & 7)) {
1955 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
1956 vmx->host_state.fs_reload_needed = 0;
1958 vmcs_write16(HOST_FS_SELECTOR, 0);
1959 vmx->host_state.fs_reload_needed = 1;
1961 savesegment(gs, vmx->host_state.gs_sel);
1962 if (!(vmx->host_state.gs_sel & 7))
1963 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
1965 vmcs_write16(HOST_GS_SELECTOR, 0);
1966 vmx->host_state.gs_ldt_reload_needed = 1;
1969 #ifdef CONFIG_X86_64
1970 savesegment(ds, vmx->host_state.ds_sel);
1971 savesegment(es, vmx->host_state.es_sel);
1974 #ifdef CONFIG_X86_64
1975 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
1976 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
1978 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
1979 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
1982 #ifdef CONFIG_X86_64
1983 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1984 if (is_long_mode(&vmx->vcpu))
1985 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1987 if (boot_cpu_has(X86_FEATURE_MPX))
1988 rdmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
1989 for (i = 0; i < vmx->save_nmsrs; ++i)
1990 kvm_set_shared_msr(vmx->guest_msrs[i].index,
1991 vmx->guest_msrs[i].data,
1992 vmx->guest_msrs[i].mask);
1995 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
1997 if (!vmx->host_state.loaded)
2000 ++vmx->vcpu.stat.host_state_reload;
2001 vmx->host_state.loaded = 0;
2002 #ifdef CONFIG_X86_64
2003 if (is_long_mode(&vmx->vcpu))
2004 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2006 if (vmx->host_state.gs_ldt_reload_needed) {
2007 kvm_load_ldt(vmx->host_state.ldt_sel);
2008 #ifdef CONFIG_X86_64
2009 load_gs_index(vmx->host_state.gs_sel);
2011 loadsegment(gs, vmx->host_state.gs_sel);
2014 if (vmx->host_state.fs_reload_needed)
2015 loadsegment(fs, vmx->host_state.fs_sel);
2016 #ifdef CONFIG_X86_64
2017 if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
2018 loadsegment(ds, vmx->host_state.ds_sel);
2019 loadsegment(es, vmx->host_state.es_sel);
2023 #ifdef CONFIG_X86_64
2024 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2026 if (vmx->host_state.msr_host_bndcfgs)
2027 wrmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
2029 * If the FPU is not active (through the host task or
2030 * the guest vcpu), then restore the cr0.TS bit.
2032 if (!fpregs_active() && !vmx->vcpu.guest_fpu_loaded)
2034 load_gdt(this_cpu_ptr(&host_gdt));
2037 static void vmx_load_host_state(struct vcpu_vmx *vmx)
2040 __vmx_load_host_state(vmx);
2044 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
2046 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2047 struct pi_desc old, new;
2050 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
2051 !irq_remapping_cap(IRQ_POSTING_CAP))
2055 old.control = new.control = pi_desc->control;
2058 * If 'nv' field is POSTED_INTR_WAKEUP_VECTOR, there
2059 * are two possible cases:
2060 * 1. After running 'pre_block', context switch
2061 * happened. For this case, 'sn' was set in
2062 * vmx_vcpu_put(), so we need to clear it here.
2063 * 2. After running 'pre_block', we were blocked,
2064 * and woken up by some other guy. For this case,
2065 * we don't need to do anything, 'pi_post_block'
2066 * will do everything for us. However, we cannot
2067 * check whether it is case #1 or case #2 here
2068 * (maybe, not needed), so we also clear sn here,
2069 * I think it is not a big deal.
2071 if (pi_desc->nv != POSTED_INTR_WAKEUP_VECTOR) {
2072 if (vcpu->cpu != cpu) {
2073 dest = cpu_physical_id(cpu);
2075 if (x2apic_enabled())
2078 new.ndst = (dest << 8) & 0xFF00;
2081 /* set 'NV' to 'notification vector' */
2082 new.nv = POSTED_INTR_VECTOR;
2085 /* Allow posting non-urgent interrupts */
2087 } while (cmpxchg(&pi_desc->control, old.control,
2088 new.control) != old.control);
2091 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
2092 * vcpu mutex is already taken.
2094 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2096 struct vcpu_vmx *vmx = to_vmx(vcpu);
2097 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2100 kvm_cpu_vmxon(phys_addr);
2101 else if (vmx->loaded_vmcs->cpu != cpu)
2102 loaded_vmcs_clear(vmx->loaded_vmcs);
2104 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
2105 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
2106 vmcs_load(vmx->loaded_vmcs->vmcs);
2109 if (vmx->loaded_vmcs->cpu != cpu) {
2110 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
2111 unsigned long sysenter_esp;
2113 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2114 local_irq_disable();
2115 crash_disable_local_vmclear(cpu);
2118 * Read loaded_vmcs->cpu should be before fetching
2119 * loaded_vmcs->loaded_vmcss_on_cpu_link.
2120 * See the comments in __loaded_vmcs_clear().
2124 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
2125 &per_cpu(loaded_vmcss_on_cpu, cpu));
2126 crash_enable_local_vmclear(cpu);
2130 * Linux uses per-cpu TSS and GDT, so set these when switching
2133 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
2134 vmcs_writel(HOST_GDTR_BASE, gdt->address); /* 22.2.4 */
2136 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
2137 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
2139 vmx->loaded_vmcs->cpu = cpu;
2142 /* Setup TSC multiplier */
2143 if (kvm_has_tsc_control &&
2144 vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio) {
2145 vmx->current_tsc_ratio = vcpu->arch.tsc_scaling_ratio;
2146 vmcs_write64(TSC_MULTIPLIER, vmx->current_tsc_ratio);
2149 vmx_vcpu_pi_load(vcpu, cpu);
2152 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
2154 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2156 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
2157 !irq_remapping_cap(IRQ_POSTING_CAP))
2160 /* Set SN when the vCPU is preempted */
2161 if (vcpu->preempted)
2165 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
2167 vmx_vcpu_pi_put(vcpu);
2169 __vmx_load_host_state(to_vmx(vcpu));
2170 if (!vmm_exclusive) {
2171 __loaded_vmcs_clear(to_vmx(vcpu)->loaded_vmcs);
2177 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
2181 if (vcpu->fpu_active)
2183 vcpu->fpu_active = 1;
2184 cr0 = vmcs_readl(GUEST_CR0);
2185 cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
2186 cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
2187 vmcs_writel(GUEST_CR0, cr0);
2188 update_exception_bitmap(vcpu);
2189 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
2190 if (is_guest_mode(vcpu))
2191 vcpu->arch.cr0_guest_owned_bits &=
2192 ~get_vmcs12(vcpu)->cr0_guest_host_mask;
2193 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2196 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
2199 * Return the cr0 value that a nested guest would read. This is a combination
2200 * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
2201 * its hypervisor (cr0_read_shadow).
2203 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
2205 return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
2206 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
2208 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
2210 return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
2211 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
2214 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
2216 /* Note that there is no vcpu->fpu_active = 0 here. The caller must
2217 * set this *before* calling this function.
2219 vmx_decache_cr0_guest_bits(vcpu);
2220 vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
2221 update_exception_bitmap(vcpu);
2222 vcpu->arch.cr0_guest_owned_bits = 0;
2223 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2224 if (is_guest_mode(vcpu)) {
2226 * L1's specified read shadow might not contain the TS bit,
2227 * so now that we turned on shadowing of this bit, we need to
2228 * set this bit of the shadow. Like in nested_vmx_run we need
2229 * nested_read_cr0(vmcs12), but vmcs12->guest_cr0 is not yet
2230 * up-to-date here because we just decached cr0.TS (and we'll
2231 * only update vmcs12->guest_cr0 on nested exit).
2233 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2234 vmcs12->guest_cr0 = (vmcs12->guest_cr0 & ~X86_CR0_TS) |
2235 (vcpu->arch.cr0 & X86_CR0_TS);
2236 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2238 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
2241 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
2243 unsigned long rflags, save_rflags;
2245 if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
2246 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2247 rflags = vmcs_readl(GUEST_RFLAGS);
2248 if (to_vmx(vcpu)->rmode.vm86_active) {
2249 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2250 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
2251 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2253 to_vmx(vcpu)->rflags = rflags;
2255 return to_vmx(vcpu)->rflags;
2258 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
2260 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2261 to_vmx(vcpu)->rflags = rflags;
2262 if (to_vmx(vcpu)->rmode.vm86_active) {
2263 to_vmx(vcpu)->rmode.save_rflags = rflags;
2264 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2266 vmcs_writel(GUEST_RFLAGS, rflags);
2269 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
2271 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2274 if (interruptibility & GUEST_INTR_STATE_STI)
2275 ret |= KVM_X86_SHADOW_INT_STI;
2276 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
2277 ret |= KVM_X86_SHADOW_INT_MOV_SS;
2282 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
2284 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2285 u32 interruptibility = interruptibility_old;
2287 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
2289 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
2290 interruptibility |= GUEST_INTR_STATE_MOV_SS;
2291 else if (mask & KVM_X86_SHADOW_INT_STI)
2292 interruptibility |= GUEST_INTR_STATE_STI;
2294 if ((interruptibility != interruptibility_old))
2295 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
2298 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
2302 rip = kvm_rip_read(vcpu);
2303 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
2304 kvm_rip_write(vcpu, rip);
2306 /* skipping an emulated instruction also counts */
2307 vmx_set_interrupt_shadow(vcpu, 0);
2311 * KVM wants to inject page-faults which it got to the guest. This function
2312 * checks whether in a nested guest, we need to inject them to L1 or L2.
2314 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned nr)
2316 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2318 if (!(vmcs12->exception_bitmap & (1u << nr)))
2321 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
2322 vmcs_read32(VM_EXIT_INTR_INFO),
2323 vmcs_readl(EXIT_QUALIFICATION));
2327 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
2328 bool has_error_code, u32 error_code,
2331 struct vcpu_vmx *vmx = to_vmx(vcpu);
2332 u32 intr_info = nr | INTR_INFO_VALID_MASK;
2334 if (!reinject && is_guest_mode(vcpu) &&
2335 nested_vmx_check_exception(vcpu, nr))
2338 if (has_error_code) {
2339 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
2340 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2343 if (vmx->rmode.vm86_active) {
2345 if (kvm_exception_is_soft(nr))
2346 inc_eip = vcpu->arch.event_exit_inst_len;
2347 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
2348 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2352 if (kvm_exception_is_soft(nr)) {
2353 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2354 vmx->vcpu.arch.event_exit_inst_len);
2355 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2357 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2359 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
2362 static bool vmx_rdtscp_supported(void)
2364 return cpu_has_vmx_rdtscp();
2367 static bool vmx_invpcid_supported(void)
2369 return cpu_has_vmx_invpcid() && enable_ept;
2373 * Swap MSR entry in host/guest MSR entry array.
2375 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
2377 struct shared_msr_entry tmp;
2379 tmp = vmx->guest_msrs[to];
2380 vmx->guest_msrs[to] = vmx->guest_msrs[from];
2381 vmx->guest_msrs[from] = tmp;
2384 static void vmx_set_msr_bitmap(struct kvm_vcpu *vcpu)
2386 unsigned long *msr_bitmap;
2388 if (is_guest_mode(vcpu))
2389 msr_bitmap = vmx_msr_bitmap_nested;
2390 else if (vcpu->arch.apic_base & X2APIC_ENABLE) {
2391 if (is_long_mode(vcpu))
2392 msr_bitmap = vmx_msr_bitmap_longmode_x2apic;
2394 msr_bitmap = vmx_msr_bitmap_legacy_x2apic;
2396 if (is_long_mode(vcpu))
2397 msr_bitmap = vmx_msr_bitmap_longmode;
2399 msr_bitmap = vmx_msr_bitmap_legacy;
2402 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
2406 * Set up the vmcs to automatically save and restore system
2407 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
2408 * mode, as fiddling with msrs is very expensive.
2410 static void setup_msrs(struct vcpu_vmx *vmx)
2412 int save_nmsrs, index;
2415 #ifdef CONFIG_X86_64
2416 if (is_long_mode(&vmx->vcpu)) {
2417 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
2419 move_msr_up(vmx, index, save_nmsrs++);
2420 index = __find_msr_index(vmx, MSR_LSTAR);
2422 move_msr_up(vmx, index, save_nmsrs++);
2423 index = __find_msr_index(vmx, MSR_CSTAR);
2425 move_msr_up(vmx, index, save_nmsrs++);
2426 index = __find_msr_index(vmx, MSR_TSC_AUX);
2427 if (index >= 0 && guest_cpuid_has_rdtscp(&vmx->vcpu))
2428 move_msr_up(vmx, index, save_nmsrs++);
2430 * MSR_STAR is only needed on long mode guests, and only
2431 * if efer.sce is enabled.
2433 index = __find_msr_index(vmx, MSR_STAR);
2434 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
2435 move_msr_up(vmx, index, save_nmsrs++);
2438 index = __find_msr_index(vmx, MSR_EFER);
2439 if (index >= 0 && update_transition_efer(vmx, index))
2440 move_msr_up(vmx, index, save_nmsrs++);
2442 vmx->save_nmsrs = save_nmsrs;
2444 if (cpu_has_vmx_msr_bitmap())
2445 vmx_set_msr_bitmap(&vmx->vcpu);
2449 * reads and returns guest's timestamp counter "register"
2450 * guest_tsc = (host_tsc * tsc multiplier) >> 48 + tsc_offset
2451 * -- Intel TSC Scaling for Virtualization White Paper, sec 1.3
2453 static u64 guest_read_tsc(struct kvm_vcpu *vcpu)
2455 u64 host_tsc, tsc_offset;
2458 tsc_offset = vmcs_read64(TSC_OFFSET);
2459 return kvm_scale_tsc(vcpu, host_tsc) + tsc_offset;
2463 * Like guest_read_tsc, but always returns L1's notion of the timestamp
2464 * counter, even if a nested guest (L2) is currently running.
2466 static u64 vmx_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2470 tsc_offset = is_guest_mode(vcpu) ?
2471 to_vmx(vcpu)->nested.vmcs01_tsc_offset :
2472 vmcs_read64(TSC_OFFSET);
2473 return host_tsc + tsc_offset;
2476 static u64 vmx_read_tsc_offset(struct kvm_vcpu *vcpu)
2478 return vmcs_read64(TSC_OFFSET);
2482 * writes 'offset' into guest's timestamp counter offset register
2484 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2486 if (is_guest_mode(vcpu)) {
2488 * We're here if L1 chose not to trap WRMSR to TSC. According
2489 * to the spec, this should set L1's TSC; The offset that L1
2490 * set for L2 remains unchanged, and still needs to be added
2491 * to the newly set TSC to get L2's TSC.
2493 struct vmcs12 *vmcs12;
2494 to_vmx(vcpu)->nested.vmcs01_tsc_offset = offset;
2495 /* recalculate vmcs02.TSC_OFFSET: */
2496 vmcs12 = get_vmcs12(vcpu);
2497 vmcs_write64(TSC_OFFSET, offset +
2498 (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
2499 vmcs12->tsc_offset : 0));
2501 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2502 vmcs_read64(TSC_OFFSET), offset);
2503 vmcs_write64(TSC_OFFSET, offset);
2507 static void vmx_adjust_tsc_offset_guest(struct kvm_vcpu *vcpu, s64 adjustment)
2509 u64 offset = vmcs_read64(TSC_OFFSET);
2511 vmcs_write64(TSC_OFFSET, offset + adjustment);
2512 if (is_guest_mode(vcpu)) {
2513 /* Even when running L2, the adjustment needs to apply to L1 */
2514 to_vmx(vcpu)->nested.vmcs01_tsc_offset += adjustment;
2516 trace_kvm_write_tsc_offset(vcpu->vcpu_id, offset,
2517 offset + adjustment);
2520 static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
2522 struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
2523 return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
2527 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
2528 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
2529 * all guests if the "nested" module option is off, and can also be disabled
2530 * for a single guest by disabling its VMX cpuid bit.
2532 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
2534 return nested && guest_cpuid_has_vmx(vcpu);
2538 * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
2539 * returned for the various VMX controls MSRs when nested VMX is enabled.
2540 * The same values should also be used to verify that vmcs12 control fields are
2541 * valid during nested entry from L1 to L2.
2542 * Each of these control msrs has a low and high 32-bit half: A low bit is on
2543 * if the corresponding bit in the (32-bit) control field *must* be on, and a
2544 * bit in the high half is on if the corresponding bit in the control field
2545 * may be on. See also vmx_control_verify().
2547 static void nested_vmx_setup_ctls_msrs(struct vcpu_vmx *vmx)
2550 * Note that as a general rule, the high half of the MSRs (bits in
2551 * the control fields which may be 1) should be initialized by the
2552 * intersection of the underlying hardware's MSR (i.e., features which
2553 * can be supported) and the list of features we want to expose -
2554 * because they are known to be properly supported in our code.
2555 * Also, usually, the low half of the MSRs (bits which must be 1) can
2556 * be set to 0, meaning that L1 may turn off any of these bits. The
2557 * reason is that if one of these bits is necessary, it will appear
2558 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
2559 * fields of vmcs01 and vmcs02, will turn these bits off - and
2560 * nested_vmx_exit_handled() will not pass related exits to L1.
2561 * These rules have exceptions below.
2564 /* pin-based controls */
2565 rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
2566 vmx->nested.nested_vmx_pinbased_ctls_low,
2567 vmx->nested.nested_vmx_pinbased_ctls_high);
2568 vmx->nested.nested_vmx_pinbased_ctls_low |=
2569 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2570 vmx->nested.nested_vmx_pinbased_ctls_high &=
2571 PIN_BASED_EXT_INTR_MASK |
2572 PIN_BASED_NMI_EXITING |
2573 PIN_BASED_VIRTUAL_NMIS;
2574 vmx->nested.nested_vmx_pinbased_ctls_high |=
2575 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2576 PIN_BASED_VMX_PREEMPTION_TIMER;
2577 if (kvm_vcpu_apicv_active(&vmx->vcpu))
2578 vmx->nested.nested_vmx_pinbased_ctls_high |=
2579 PIN_BASED_POSTED_INTR;
2582 rdmsr(MSR_IA32_VMX_EXIT_CTLS,
2583 vmx->nested.nested_vmx_exit_ctls_low,
2584 vmx->nested.nested_vmx_exit_ctls_high);
2585 vmx->nested.nested_vmx_exit_ctls_low =
2586 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
2588 vmx->nested.nested_vmx_exit_ctls_high &=
2589 #ifdef CONFIG_X86_64
2590 VM_EXIT_HOST_ADDR_SPACE_SIZE |
2592 VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
2593 vmx->nested.nested_vmx_exit_ctls_high |=
2594 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
2595 VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
2596 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
2598 if (vmx_mpx_supported())
2599 vmx->nested.nested_vmx_exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
2601 /* We support free control of debug control saving. */
2602 vmx->nested.nested_vmx_true_exit_ctls_low =
2603 vmx->nested.nested_vmx_exit_ctls_low &
2604 ~VM_EXIT_SAVE_DEBUG_CONTROLS;
2606 /* entry controls */
2607 rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
2608 vmx->nested.nested_vmx_entry_ctls_low,
2609 vmx->nested.nested_vmx_entry_ctls_high);
2610 vmx->nested.nested_vmx_entry_ctls_low =
2611 VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
2612 vmx->nested.nested_vmx_entry_ctls_high &=
2613 #ifdef CONFIG_X86_64
2614 VM_ENTRY_IA32E_MODE |
2616 VM_ENTRY_LOAD_IA32_PAT;
2617 vmx->nested.nested_vmx_entry_ctls_high |=
2618 (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
2619 if (vmx_mpx_supported())
2620 vmx->nested.nested_vmx_entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
2622 /* We support free control of debug control loading. */
2623 vmx->nested.nested_vmx_true_entry_ctls_low =
2624 vmx->nested.nested_vmx_entry_ctls_low &
2625 ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
2627 /* cpu-based controls */
2628 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
2629 vmx->nested.nested_vmx_procbased_ctls_low,
2630 vmx->nested.nested_vmx_procbased_ctls_high);
2631 vmx->nested.nested_vmx_procbased_ctls_low =
2632 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2633 vmx->nested.nested_vmx_procbased_ctls_high &=
2634 CPU_BASED_VIRTUAL_INTR_PENDING |
2635 CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
2636 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
2637 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
2638 CPU_BASED_CR3_STORE_EXITING |
2639 #ifdef CONFIG_X86_64
2640 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
2642 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
2643 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
2644 CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
2645 CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
2646 CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2648 * We can allow some features even when not supported by the
2649 * hardware. For example, L1 can specify an MSR bitmap - and we
2650 * can use it to avoid exits to L1 - even when L0 runs L2
2651 * without MSR bitmaps.
2653 vmx->nested.nested_vmx_procbased_ctls_high |=
2654 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2655 CPU_BASED_USE_MSR_BITMAPS;
2657 /* We support free control of CR3 access interception. */
2658 vmx->nested.nested_vmx_true_procbased_ctls_low =
2659 vmx->nested.nested_vmx_procbased_ctls_low &
2660 ~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
2662 /* secondary cpu-based controls */
2663 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2664 vmx->nested.nested_vmx_secondary_ctls_low,
2665 vmx->nested.nested_vmx_secondary_ctls_high);
2666 vmx->nested.nested_vmx_secondary_ctls_low = 0;
2667 vmx->nested.nested_vmx_secondary_ctls_high &=
2668 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2669 SECONDARY_EXEC_RDTSCP |
2670 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2671 SECONDARY_EXEC_ENABLE_VPID |
2672 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2673 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2674 SECONDARY_EXEC_WBINVD_EXITING |
2675 SECONDARY_EXEC_XSAVES |
2676 SECONDARY_EXEC_PCOMMIT;
2679 /* nested EPT: emulate EPT also to L1 */
2680 vmx->nested.nested_vmx_secondary_ctls_high |=
2681 SECONDARY_EXEC_ENABLE_EPT;
2682 vmx->nested.nested_vmx_ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
2683 VMX_EPTP_WB_BIT | VMX_EPT_2MB_PAGE_BIT |
2685 vmx->nested.nested_vmx_ept_caps &= vmx_capability.ept;
2687 * For nested guests, we don't do anything specific
2688 * for single context invalidation. Hence, only advertise
2689 * support for global context invalidation.
2691 vmx->nested.nested_vmx_ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT;
2693 vmx->nested.nested_vmx_ept_caps = 0;
2696 vmx->nested.nested_vmx_vpid_caps = VMX_VPID_INVVPID_BIT |
2697 VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
2699 vmx->nested.nested_vmx_vpid_caps = 0;
2701 if (enable_unrestricted_guest)
2702 vmx->nested.nested_vmx_secondary_ctls_high |=
2703 SECONDARY_EXEC_UNRESTRICTED_GUEST;
2705 /* miscellaneous data */
2706 rdmsr(MSR_IA32_VMX_MISC,
2707 vmx->nested.nested_vmx_misc_low,
2708 vmx->nested.nested_vmx_misc_high);
2709 vmx->nested.nested_vmx_misc_low &= VMX_MISC_SAVE_EFER_LMA;
2710 vmx->nested.nested_vmx_misc_low |=
2711 VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
2712 VMX_MISC_ACTIVITY_HLT;
2713 vmx->nested.nested_vmx_misc_high = 0;
2716 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2719 * Bits 0 in high must be 0, and bits 1 in low must be 1.
2721 return ((control & high) | low) == control;
2724 static inline u64 vmx_control_msr(u32 low, u32 high)
2726 return low | ((u64)high << 32);
2729 /* Returns 0 on success, non-0 otherwise. */
2730 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2732 struct vcpu_vmx *vmx = to_vmx(vcpu);
2734 switch (msr_index) {
2735 case MSR_IA32_VMX_BASIC:
2737 * This MSR reports some information about VMX support. We
2738 * should return information about the VMX we emulate for the
2739 * guest, and the VMCS structure we give it - not about the
2740 * VMX support of the underlying hardware.
2742 *pdata = VMCS12_REVISION | VMX_BASIC_TRUE_CTLS |
2743 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2744 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2746 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2747 case MSR_IA32_VMX_PINBASED_CTLS:
2748 *pdata = vmx_control_msr(
2749 vmx->nested.nested_vmx_pinbased_ctls_low,
2750 vmx->nested.nested_vmx_pinbased_ctls_high);
2752 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2753 *pdata = vmx_control_msr(
2754 vmx->nested.nested_vmx_true_procbased_ctls_low,
2755 vmx->nested.nested_vmx_procbased_ctls_high);
2757 case MSR_IA32_VMX_PROCBASED_CTLS:
2758 *pdata = vmx_control_msr(
2759 vmx->nested.nested_vmx_procbased_ctls_low,
2760 vmx->nested.nested_vmx_procbased_ctls_high);
2762 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2763 *pdata = vmx_control_msr(
2764 vmx->nested.nested_vmx_true_exit_ctls_low,
2765 vmx->nested.nested_vmx_exit_ctls_high);
2767 case MSR_IA32_VMX_EXIT_CTLS:
2768 *pdata = vmx_control_msr(
2769 vmx->nested.nested_vmx_exit_ctls_low,
2770 vmx->nested.nested_vmx_exit_ctls_high);
2772 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2773 *pdata = vmx_control_msr(
2774 vmx->nested.nested_vmx_true_entry_ctls_low,
2775 vmx->nested.nested_vmx_entry_ctls_high);
2777 case MSR_IA32_VMX_ENTRY_CTLS:
2778 *pdata = vmx_control_msr(
2779 vmx->nested.nested_vmx_entry_ctls_low,
2780 vmx->nested.nested_vmx_entry_ctls_high);
2782 case MSR_IA32_VMX_MISC:
2783 *pdata = vmx_control_msr(
2784 vmx->nested.nested_vmx_misc_low,
2785 vmx->nested.nested_vmx_misc_high);
2788 * These MSRs specify bits which the guest must keep fixed (on or off)
2789 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2790 * We picked the standard core2 setting.
2792 #define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2793 #define VMXON_CR4_ALWAYSON X86_CR4_VMXE
2794 case MSR_IA32_VMX_CR0_FIXED0:
2795 *pdata = VMXON_CR0_ALWAYSON;
2797 case MSR_IA32_VMX_CR0_FIXED1:
2800 case MSR_IA32_VMX_CR4_FIXED0:
2801 *pdata = VMXON_CR4_ALWAYSON;
2803 case MSR_IA32_VMX_CR4_FIXED1:
2806 case MSR_IA32_VMX_VMCS_ENUM:
2807 *pdata = 0x2e; /* highest index: VMX_PREEMPTION_TIMER_VALUE */
2809 case MSR_IA32_VMX_PROCBASED_CTLS2:
2810 *pdata = vmx_control_msr(
2811 vmx->nested.nested_vmx_secondary_ctls_low,
2812 vmx->nested.nested_vmx_secondary_ctls_high);
2814 case MSR_IA32_VMX_EPT_VPID_CAP:
2815 /* Currently, no nested vpid support */
2816 *pdata = vmx->nested.nested_vmx_ept_caps |
2817 ((u64)vmx->nested.nested_vmx_vpid_caps << 32);
2827 * Reads an msr value (of 'msr_index') into 'pdata'.
2828 * Returns 0 on success, non-0 otherwise.
2829 * Assumes vcpu_load() was already called.
2831 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2833 struct shared_msr_entry *msr;
2835 switch (msr_info->index) {
2836 #ifdef CONFIG_X86_64
2838 msr_info->data = vmcs_readl(GUEST_FS_BASE);
2841 msr_info->data = vmcs_readl(GUEST_GS_BASE);
2843 case MSR_KERNEL_GS_BASE:
2844 vmx_load_host_state(to_vmx(vcpu));
2845 msr_info->data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
2849 return kvm_get_msr_common(vcpu, msr_info);
2851 msr_info->data = guest_read_tsc(vcpu);
2853 case MSR_IA32_SYSENTER_CS:
2854 msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
2856 case MSR_IA32_SYSENTER_EIP:
2857 msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
2859 case MSR_IA32_SYSENTER_ESP:
2860 msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
2862 case MSR_IA32_BNDCFGS:
2863 if (!vmx_mpx_supported())
2865 msr_info->data = vmcs_read64(GUEST_BNDCFGS);
2867 case MSR_IA32_FEATURE_CONTROL:
2868 if (!nested_vmx_allowed(vcpu))
2870 msr_info->data = to_vmx(vcpu)->nested.msr_ia32_feature_control;
2872 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2873 if (!nested_vmx_allowed(vcpu))
2875 return vmx_get_vmx_msr(vcpu, msr_info->index, &msr_info->data);
2877 if (!vmx_xsaves_supported())
2879 msr_info->data = vcpu->arch.ia32_xss;
2882 if (!guest_cpuid_has_rdtscp(vcpu) && !msr_info->host_initiated)
2884 /* Otherwise falls through */
2886 msr = find_msr_entry(to_vmx(vcpu), msr_info->index);
2888 msr_info->data = msr->data;
2891 return kvm_get_msr_common(vcpu, msr_info);
2897 static void vmx_leave_nested(struct kvm_vcpu *vcpu);
2900 * Writes msr value into into the appropriate "register".
2901 * Returns 0 on success, non-0 otherwise.
2902 * Assumes vcpu_load() was already called.
2904 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2906 struct vcpu_vmx *vmx = to_vmx(vcpu);
2907 struct shared_msr_entry *msr;
2909 u32 msr_index = msr_info->index;
2910 u64 data = msr_info->data;
2912 switch (msr_index) {
2914 ret = kvm_set_msr_common(vcpu, msr_info);
2916 #ifdef CONFIG_X86_64
2918 vmx_segment_cache_clear(vmx);
2919 vmcs_writel(GUEST_FS_BASE, data);
2922 vmx_segment_cache_clear(vmx);
2923 vmcs_writel(GUEST_GS_BASE, data);
2925 case MSR_KERNEL_GS_BASE:
2926 vmx_load_host_state(vmx);
2927 vmx->msr_guest_kernel_gs_base = data;
2930 case MSR_IA32_SYSENTER_CS:
2931 vmcs_write32(GUEST_SYSENTER_CS, data);
2933 case MSR_IA32_SYSENTER_EIP:
2934 vmcs_writel(GUEST_SYSENTER_EIP, data);
2936 case MSR_IA32_SYSENTER_ESP:
2937 vmcs_writel(GUEST_SYSENTER_ESP, data);
2939 case MSR_IA32_BNDCFGS:
2940 if (!vmx_mpx_supported())
2942 vmcs_write64(GUEST_BNDCFGS, data);
2945 kvm_write_tsc(vcpu, msr_info);
2947 case MSR_IA32_CR_PAT:
2948 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2949 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
2951 vmcs_write64(GUEST_IA32_PAT, data);
2952 vcpu->arch.pat = data;
2955 ret = kvm_set_msr_common(vcpu, msr_info);
2957 case MSR_IA32_TSC_ADJUST:
2958 ret = kvm_set_msr_common(vcpu, msr_info);
2960 case MSR_IA32_FEATURE_CONTROL:
2961 if (!nested_vmx_allowed(vcpu) ||
2962 (to_vmx(vcpu)->nested.msr_ia32_feature_control &
2963 FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
2965 vmx->nested.msr_ia32_feature_control = data;
2966 if (msr_info->host_initiated && data == 0)
2967 vmx_leave_nested(vcpu);
2969 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2970 return 1; /* they are read-only */
2972 if (!vmx_xsaves_supported())
2975 * The only supported bit as of Skylake is bit 8, but
2976 * it is not supported on KVM.
2980 vcpu->arch.ia32_xss = data;
2981 if (vcpu->arch.ia32_xss != host_xss)
2982 add_atomic_switch_msr(vmx, MSR_IA32_XSS,
2983 vcpu->arch.ia32_xss, host_xss);
2985 clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
2988 if (!guest_cpuid_has_rdtscp(vcpu) && !msr_info->host_initiated)
2990 /* Check reserved bit, higher 32 bits should be zero */
2991 if ((data >> 32) != 0)
2993 /* Otherwise falls through */
2995 msr = find_msr_entry(vmx, msr_index);
2997 u64 old_msr_data = msr->data;
2999 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
3001 ret = kvm_set_shared_msr(msr->index, msr->data,
3005 msr->data = old_msr_data;
3009 ret = kvm_set_msr_common(vcpu, msr_info);
3015 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
3017 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
3020 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
3023 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
3025 case VCPU_EXREG_PDPTR:
3027 ept_save_pdptrs(vcpu);
3034 static __init int cpu_has_kvm_support(void)
3036 return cpu_has_vmx();
3039 static __init int vmx_disabled_by_bios(void)
3043 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
3044 if (msr & FEATURE_CONTROL_LOCKED) {
3045 /* launched w/ TXT and VMX disabled */
3046 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3049 /* launched w/o TXT and VMX only enabled w/ TXT */
3050 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3051 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3052 && !tboot_enabled()) {
3053 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
3054 "activate TXT before enabling KVM\n");
3057 /* launched w/o TXT and VMX disabled */
3058 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3059 && !tboot_enabled())
3066 static void kvm_cpu_vmxon(u64 addr)
3068 asm volatile (ASM_VMX_VMXON_RAX
3069 : : "a"(&addr), "m"(addr)
3073 static int hardware_enable(void)
3075 int cpu = raw_smp_processor_id();
3076 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
3079 if (cr4_read_shadow() & X86_CR4_VMXE)
3082 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
3083 INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
3084 spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
3087 * Now we can enable the vmclear operation in kdump
3088 * since the loaded_vmcss_on_cpu list on this cpu
3089 * has been initialized.
3091 * Though the cpu is not in VMX operation now, there
3092 * is no problem to enable the vmclear operation
3093 * for the loaded_vmcss_on_cpu list is empty!
3095 crash_enable_local_vmclear(cpu);
3097 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
3099 test_bits = FEATURE_CONTROL_LOCKED;
3100 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
3101 if (tboot_enabled())
3102 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
3104 if ((old & test_bits) != test_bits) {
3105 /* enable and lock */
3106 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
3108 cr4_set_bits(X86_CR4_VMXE);
3110 if (vmm_exclusive) {
3111 kvm_cpu_vmxon(phys_addr);
3115 native_store_gdt(this_cpu_ptr(&host_gdt));
3120 static void vmclear_local_loaded_vmcss(void)
3122 int cpu = raw_smp_processor_id();
3123 struct loaded_vmcs *v, *n;
3125 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
3126 loaded_vmcss_on_cpu_link)
3127 __loaded_vmcs_clear(v);
3131 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
3134 static void kvm_cpu_vmxoff(void)
3136 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
3139 static void hardware_disable(void)
3141 if (vmm_exclusive) {
3142 vmclear_local_loaded_vmcss();
3145 cr4_clear_bits(X86_CR4_VMXE);
3148 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
3149 u32 msr, u32 *result)
3151 u32 vmx_msr_low, vmx_msr_high;
3152 u32 ctl = ctl_min | ctl_opt;
3154 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3156 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
3157 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
3159 /* Ensure minimum (required) set of control bits are supported. */
3167 static __init bool allow_1_setting(u32 msr, u32 ctl)
3169 u32 vmx_msr_low, vmx_msr_high;
3171 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3172 return vmx_msr_high & ctl;
3175 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
3177 u32 vmx_msr_low, vmx_msr_high;
3178 u32 min, opt, min2, opt2;
3179 u32 _pin_based_exec_control = 0;
3180 u32 _cpu_based_exec_control = 0;
3181 u32 _cpu_based_2nd_exec_control = 0;
3182 u32 _vmexit_control = 0;
3183 u32 _vmentry_control = 0;
3185 min = CPU_BASED_HLT_EXITING |
3186 #ifdef CONFIG_X86_64
3187 CPU_BASED_CR8_LOAD_EXITING |
3188 CPU_BASED_CR8_STORE_EXITING |
3190 CPU_BASED_CR3_LOAD_EXITING |
3191 CPU_BASED_CR3_STORE_EXITING |
3192 CPU_BASED_USE_IO_BITMAPS |
3193 CPU_BASED_MOV_DR_EXITING |
3194 CPU_BASED_USE_TSC_OFFSETING |
3195 CPU_BASED_MWAIT_EXITING |
3196 CPU_BASED_MONITOR_EXITING |
3197 CPU_BASED_INVLPG_EXITING |
3198 CPU_BASED_RDPMC_EXITING;
3200 opt = CPU_BASED_TPR_SHADOW |
3201 CPU_BASED_USE_MSR_BITMAPS |
3202 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
3203 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
3204 &_cpu_based_exec_control) < 0)
3206 #ifdef CONFIG_X86_64
3207 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3208 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
3209 ~CPU_BASED_CR8_STORE_EXITING;
3211 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
3213 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
3214 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3215 SECONDARY_EXEC_WBINVD_EXITING |
3216 SECONDARY_EXEC_ENABLE_VPID |
3217 SECONDARY_EXEC_ENABLE_EPT |
3218 SECONDARY_EXEC_UNRESTRICTED_GUEST |
3219 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
3220 SECONDARY_EXEC_RDTSCP |
3221 SECONDARY_EXEC_ENABLE_INVPCID |
3222 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3223 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
3224 SECONDARY_EXEC_SHADOW_VMCS |
3225 SECONDARY_EXEC_XSAVES |
3226 SECONDARY_EXEC_ENABLE_PML |
3227 SECONDARY_EXEC_PCOMMIT |
3228 SECONDARY_EXEC_TSC_SCALING;
3229 if (adjust_vmx_controls(min2, opt2,
3230 MSR_IA32_VMX_PROCBASED_CTLS2,
3231 &_cpu_based_2nd_exec_control) < 0)
3234 #ifndef CONFIG_X86_64
3235 if (!(_cpu_based_2nd_exec_control &
3236 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
3237 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
3240 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3241 _cpu_based_2nd_exec_control &= ~(
3242 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3243 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3244 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3246 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
3247 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
3249 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
3250 CPU_BASED_CR3_STORE_EXITING |
3251 CPU_BASED_INVLPG_EXITING);
3252 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
3253 vmx_capability.ept, vmx_capability.vpid);
3256 min = VM_EXIT_SAVE_DEBUG_CONTROLS;
3257 #ifdef CONFIG_X86_64
3258 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
3260 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT |
3261 VM_EXIT_ACK_INTR_ON_EXIT | VM_EXIT_CLEAR_BNDCFGS;
3262 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
3263 &_vmexit_control) < 0)
3266 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
3267 opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR;
3268 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
3269 &_pin_based_exec_control) < 0)
3272 if (!(_cpu_based_2nd_exec_control &
3273 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) ||
3274 !(_vmexit_control & VM_EXIT_ACK_INTR_ON_EXIT))
3275 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
3277 min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
3278 opt = VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS;
3279 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
3280 &_vmentry_control) < 0)
3283 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
3285 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
3286 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
3289 #ifdef CONFIG_X86_64
3290 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
3291 if (vmx_msr_high & (1u<<16))
3295 /* Require Write-Back (WB) memory type for VMCS accesses. */
3296 if (((vmx_msr_high >> 18) & 15) != 6)
3299 vmcs_conf->size = vmx_msr_high & 0x1fff;
3300 vmcs_conf->order = get_order(vmcs_config.size);
3301 vmcs_conf->revision_id = vmx_msr_low;
3303 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
3304 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
3305 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
3306 vmcs_conf->vmexit_ctrl = _vmexit_control;
3307 vmcs_conf->vmentry_ctrl = _vmentry_control;
3309 cpu_has_load_ia32_efer =
3310 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3311 VM_ENTRY_LOAD_IA32_EFER)
3312 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3313 VM_EXIT_LOAD_IA32_EFER);
3315 cpu_has_load_perf_global_ctrl =
3316 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3317 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
3318 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3319 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
3322 * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
3323 * but due to arrata below it can't be used. Workaround is to use
3324 * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
3326 * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
3331 * BC86,AAY89,BD102 (model 44)
3335 if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
3336 switch (boot_cpu_data.x86_model) {
3342 cpu_has_load_perf_global_ctrl = false;
3343 printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
3344 "does not work properly. Using workaround\n");
3352 rdmsrl(MSR_IA32_XSS, host_xss);
3357 static struct vmcs *alloc_vmcs_cpu(int cpu)
3359 int node = cpu_to_node(cpu);
3363 pages = __alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
3366 vmcs = page_address(pages);
3367 memset(vmcs, 0, vmcs_config.size);
3368 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
3372 static struct vmcs *alloc_vmcs(void)
3374 return alloc_vmcs_cpu(raw_smp_processor_id());
3377 static void free_vmcs(struct vmcs *vmcs)
3379 free_pages((unsigned long)vmcs, vmcs_config.order);
3383 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
3385 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
3387 if (!loaded_vmcs->vmcs)
3389 loaded_vmcs_clear(loaded_vmcs);
3390 free_vmcs(loaded_vmcs->vmcs);
3391 loaded_vmcs->vmcs = NULL;
3394 static void free_kvm_area(void)
3398 for_each_possible_cpu(cpu) {
3399 free_vmcs(per_cpu(vmxarea, cpu));
3400 per_cpu(vmxarea, cpu) = NULL;
3404 static void init_vmcs_shadow_fields(void)
3408 /* No checks for read only fields yet */
3410 for (i = j = 0; i < max_shadow_read_write_fields; i++) {
3411 switch (shadow_read_write_fields[i]) {
3413 if (!vmx_mpx_supported())
3421 shadow_read_write_fields[j] =
3422 shadow_read_write_fields[i];
3425 max_shadow_read_write_fields = j;
3427 /* shadowed fields guest access without vmexit */
3428 for (i = 0; i < max_shadow_read_write_fields; i++) {
3429 clear_bit(shadow_read_write_fields[i],
3430 vmx_vmwrite_bitmap);
3431 clear_bit(shadow_read_write_fields[i],
3434 for (i = 0; i < max_shadow_read_only_fields; i++)
3435 clear_bit(shadow_read_only_fields[i],
3439 static __init int alloc_kvm_area(void)
3443 for_each_possible_cpu(cpu) {
3446 vmcs = alloc_vmcs_cpu(cpu);
3452 per_cpu(vmxarea, cpu) = vmcs;
3457 static bool emulation_required(struct kvm_vcpu *vcpu)
3459 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
3462 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
3463 struct kvm_segment *save)
3465 if (!emulate_invalid_guest_state) {
3467 * CS and SS RPL should be equal during guest entry according
3468 * to VMX spec, but in reality it is not always so. Since vcpu
3469 * is in the middle of the transition from real mode to
3470 * protected mode it is safe to assume that RPL 0 is a good
3473 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
3474 save->selector &= ~SEGMENT_RPL_MASK;
3475 save->dpl = save->selector & SEGMENT_RPL_MASK;
3478 vmx_set_segment(vcpu, save, seg);
3481 static void enter_pmode(struct kvm_vcpu *vcpu)
3483 unsigned long flags;
3484 struct vcpu_vmx *vmx = to_vmx(vcpu);
3487 * Update real mode segment cache. It may be not up-to-date if sement
3488 * register was written while vcpu was in a guest mode.
3490 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3491 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3492 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3493 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3494 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3495 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3497 vmx->rmode.vm86_active = 0;
3499 vmx_segment_cache_clear(vmx);
3501 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3503 flags = vmcs_readl(GUEST_RFLAGS);
3504 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
3505 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
3506 vmcs_writel(GUEST_RFLAGS, flags);
3508 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
3509 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
3511 update_exception_bitmap(vcpu);
3513 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3514 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3515 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3516 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3517 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3518 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3521 static void fix_rmode_seg(int seg, struct kvm_segment *save)
3523 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3524 struct kvm_segment var = *save;
3527 if (seg == VCPU_SREG_CS)
3530 if (!emulate_invalid_guest_state) {
3531 var.selector = var.base >> 4;
3532 var.base = var.base & 0xffff0;
3542 if (save->base & 0xf)
3543 printk_once(KERN_WARNING "kvm: segment base is not "
3544 "paragraph aligned when entering "
3545 "protected mode (seg=%d)", seg);
3548 vmcs_write16(sf->selector, var.selector);
3549 vmcs_write32(sf->base, var.base);
3550 vmcs_write32(sf->limit, var.limit);
3551 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
3554 static void enter_rmode(struct kvm_vcpu *vcpu)
3556 unsigned long flags;
3557 struct vcpu_vmx *vmx = to_vmx(vcpu);
3559 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3560 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3561 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3562 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3563 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3564 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3565 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3567 vmx->rmode.vm86_active = 1;
3570 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
3571 * vcpu. Warn the user that an update is overdue.
3573 if (!vcpu->kvm->arch.tss_addr)
3574 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
3575 "called before entering vcpu\n");
3577 vmx_segment_cache_clear(vmx);
3579 vmcs_writel(GUEST_TR_BASE, vcpu->kvm->arch.tss_addr);
3580 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
3581 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3583 flags = vmcs_readl(GUEST_RFLAGS);
3584 vmx->rmode.save_rflags = flags;
3586 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
3588 vmcs_writel(GUEST_RFLAGS, flags);
3589 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
3590 update_exception_bitmap(vcpu);
3592 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3593 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3594 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3595 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3596 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3597 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3599 kvm_mmu_reset_context(vcpu);
3602 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
3604 struct vcpu_vmx *vmx = to_vmx(vcpu);
3605 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
3611 * Force kernel_gs_base reloading before EFER changes, as control
3612 * of this msr depends on is_long_mode().
3614 vmx_load_host_state(to_vmx(vcpu));
3615 vcpu->arch.efer = efer;
3616 if (efer & EFER_LMA) {
3617 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3620 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3622 msr->data = efer & ~EFER_LME;
3627 #ifdef CONFIG_X86_64
3629 static void enter_lmode(struct kvm_vcpu *vcpu)
3633 vmx_segment_cache_clear(to_vmx(vcpu));
3635 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
3636 if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
3637 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
3639 vmcs_write32(GUEST_TR_AR_BYTES,
3640 (guest_tr_ar & ~VMX_AR_TYPE_MASK)
3641 | VMX_AR_TYPE_BUSY_64_TSS);
3643 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
3646 static void exit_lmode(struct kvm_vcpu *vcpu)
3648 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3649 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
3654 static inline void __vmx_flush_tlb(struct kvm_vcpu *vcpu, int vpid)
3656 vpid_sync_context(vpid);
3658 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
3660 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
3664 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
3666 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->vpid);
3669 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
3671 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
3673 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
3674 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
3677 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
3679 if (enable_ept && is_paging(vcpu))
3680 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3681 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
3684 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
3686 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
3688 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
3689 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
3692 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
3694 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3696 if (!test_bit(VCPU_EXREG_PDPTR,
3697 (unsigned long *)&vcpu->arch.regs_dirty))
3700 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3701 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
3702 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
3703 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
3704 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
3708 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
3710 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3712 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3713 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
3714 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
3715 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
3716 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
3719 __set_bit(VCPU_EXREG_PDPTR,
3720 (unsigned long *)&vcpu->arch.regs_avail);
3721 __set_bit(VCPU_EXREG_PDPTR,
3722 (unsigned long *)&vcpu->arch.regs_dirty);
3725 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
3727 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
3729 struct kvm_vcpu *vcpu)
3731 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3732 vmx_decache_cr3(vcpu);
3733 if (!(cr0 & X86_CR0_PG)) {
3734 /* From paging/starting to nonpaging */
3735 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3736 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
3737 (CPU_BASED_CR3_LOAD_EXITING |
3738 CPU_BASED_CR3_STORE_EXITING));
3739 vcpu->arch.cr0 = cr0;
3740 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3741 } else if (!is_paging(vcpu)) {
3742 /* From nonpaging to paging */
3743 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3744 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
3745 ~(CPU_BASED_CR3_LOAD_EXITING |
3746 CPU_BASED_CR3_STORE_EXITING));
3747 vcpu->arch.cr0 = cr0;
3748 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3751 if (!(cr0 & X86_CR0_WP))
3752 *hw_cr0 &= ~X86_CR0_WP;
3755 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
3757 struct vcpu_vmx *vmx = to_vmx(vcpu);
3758 unsigned long hw_cr0;
3760 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK);
3761 if (enable_unrestricted_guest)
3762 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
3764 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
3766 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
3769 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
3773 #ifdef CONFIG_X86_64
3774 if (vcpu->arch.efer & EFER_LME) {
3775 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
3777 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
3783 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
3785 if (!vcpu->fpu_active)
3786 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
3788 vmcs_writel(CR0_READ_SHADOW, cr0);
3789 vmcs_writel(GUEST_CR0, hw_cr0);
3790 vcpu->arch.cr0 = cr0;
3792 /* depends on vcpu->arch.cr0 to be set to a new value */
3793 vmx->emulation_required = emulation_required(vcpu);
3796 static u64 construct_eptp(unsigned long root_hpa)
3800 /* TODO write the value reading from MSR */
3801 eptp = VMX_EPT_DEFAULT_MT |
3802 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
3803 if (enable_ept_ad_bits)
3804 eptp |= VMX_EPT_AD_ENABLE_BIT;
3805 eptp |= (root_hpa & PAGE_MASK);
3810 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
3812 unsigned long guest_cr3;
3817 eptp = construct_eptp(cr3);
3818 vmcs_write64(EPT_POINTER, eptp);
3819 if (is_paging(vcpu) || is_guest_mode(vcpu))
3820 guest_cr3 = kvm_read_cr3(vcpu);
3822 guest_cr3 = vcpu->kvm->arch.ept_identity_map_addr;
3823 ept_load_pdptrs(vcpu);
3826 vmx_flush_tlb(vcpu);
3827 vmcs_writel(GUEST_CR3, guest_cr3);
3830 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3833 * Pass through host's Machine Check Enable value to hw_cr4, which
3834 * is in force while we are in guest mode. Do not let guests control
3835 * this bit, even if host CR4.MCE == 0.
3837 unsigned long hw_cr4 =
3838 (cr4_read_shadow() & X86_CR4_MCE) |
3839 (cr4 & ~X86_CR4_MCE) |
3840 (to_vmx(vcpu)->rmode.vm86_active ?
3841 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
3843 if (cr4 & X86_CR4_VMXE) {
3845 * To use VMXON (and later other VMX instructions), a guest
3846 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3847 * So basically the check on whether to allow nested VMX
3850 if (!nested_vmx_allowed(vcpu))
3853 if (to_vmx(vcpu)->nested.vmxon &&
3854 ((cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON))
3857 vcpu->arch.cr4 = cr4;
3859 if (!is_paging(vcpu)) {
3860 hw_cr4 &= ~X86_CR4_PAE;
3861 hw_cr4 |= X86_CR4_PSE;
3862 } else if (!(cr4 & X86_CR4_PAE)) {
3863 hw_cr4 &= ~X86_CR4_PAE;
3867 if (!enable_unrestricted_guest && !is_paging(vcpu))
3869 * SMEP/SMAP is disabled if CPU is in non-paging mode in
3870 * hardware. However KVM always uses paging mode without
3871 * unrestricted guest.
3872 * To emulate this behavior, SMEP/SMAP needs to be manually
3873 * disabled when guest switches to non-paging mode.
3875 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP);
3877 vmcs_writel(CR4_READ_SHADOW, cr4);
3878 vmcs_writel(GUEST_CR4, hw_cr4);
3882 static void vmx_get_segment(struct kvm_vcpu *vcpu,
3883 struct kvm_segment *var, int seg)
3885 struct vcpu_vmx *vmx = to_vmx(vcpu);
3888 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3889 *var = vmx->rmode.segs[seg];
3890 if (seg == VCPU_SREG_TR
3891 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3893 var->base = vmx_read_guest_seg_base(vmx, seg);
3894 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3897 var->base = vmx_read_guest_seg_base(vmx, seg);
3898 var->limit = vmx_read_guest_seg_limit(vmx, seg);
3899 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3900 ar = vmx_read_guest_seg_ar(vmx, seg);
3901 var->unusable = (ar >> 16) & 1;
3902 var->type = ar & 15;
3903 var->s = (ar >> 4) & 1;
3904 var->dpl = (ar >> 5) & 3;
3906 * Some userspaces do not preserve unusable property. Since usable
3907 * segment has to be present according to VMX spec we can use present
3908 * property to amend userspace bug by making unusable segment always
3909 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3910 * segment as unusable.
3912 var->present = !var->unusable;
3913 var->avl = (ar >> 12) & 1;
3914 var->l = (ar >> 13) & 1;
3915 var->db = (ar >> 14) & 1;
3916 var->g = (ar >> 15) & 1;
3919 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3921 struct kvm_segment s;
3923 if (to_vmx(vcpu)->rmode.vm86_active) {
3924 vmx_get_segment(vcpu, &s, seg);
3927 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3930 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
3932 struct vcpu_vmx *vmx = to_vmx(vcpu);
3934 if (unlikely(vmx->rmode.vm86_active))
3937 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3938 return VMX_AR_DPL(ar);
3942 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3946 if (var->unusable || !var->present)
3949 ar = var->type & 15;
3950 ar |= (var->s & 1) << 4;
3951 ar |= (var->dpl & 3) << 5;
3952 ar |= (var->present & 1) << 7;
3953 ar |= (var->avl & 1) << 12;
3954 ar |= (var->l & 1) << 13;
3955 ar |= (var->db & 1) << 14;
3956 ar |= (var->g & 1) << 15;
3962 static void vmx_set_segment(struct kvm_vcpu *vcpu,
3963 struct kvm_segment *var, int seg)
3965 struct vcpu_vmx *vmx = to_vmx(vcpu);
3966 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3968 vmx_segment_cache_clear(vmx);
3970 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3971 vmx->rmode.segs[seg] = *var;
3972 if (seg == VCPU_SREG_TR)
3973 vmcs_write16(sf->selector, var->selector);
3975 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3979 vmcs_writel(sf->base, var->base);
3980 vmcs_write32(sf->limit, var->limit);
3981 vmcs_write16(sf->selector, var->selector);
3984 * Fix the "Accessed" bit in AR field of segment registers for older
3986 * IA32 arch specifies that at the time of processor reset the
3987 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3988 * is setting it to 0 in the userland code. This causes invalid guest
3989 * state vmexit when "unrestricted guest" mode is turned on.
3990 * Fix for this setup issue in cpu_reset is being pushed in the qemu
3991 * tree. Newer qemu binaries with that qemu fix would not need this
3994 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3995 var->type |= 0x1; /* Accessed */
3997 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
4000 vmx->emulation_required = emulation_required(vcpu);
4003 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4005 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
4007 *db = (ar >> 14) & 1;
4008 *l = (ar >> 13) & 1;
4011 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4013 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
4014 dt->address = vmcs_readl(GUEST_IDTR_BASE);
4017 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4019 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
4020 vmcs_writel(GUEST_IDTR_BASE, dt->address);
4023 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4025 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
4026 dt->address = vmcs_readl(GUEST_GDTR_BASE);
4029 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4031 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
4032 vmcs_writel(GUEST_GDTR_BASE, dt->address);
4035 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
4037 struct kvm_segment var;
4040 vmx_get_segment(vcpu, &var, seg);
4042 if (seg == VCPU_SREG_CS)
4044 ar = vmx_segment_access_rights(&var);
4046 if (var.base != (var.selector << 4))
4048 if (var.limit != 0xffff)
4056 static bool code_segment_valid(struct kvm_vcpu *vcpu)
4058 struct kvm_segment cs;
4059 unsigned int cs_rpl;
4061 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4062 cs_rpl = cs.selector & SEGMENT_RPL_MASK;
4066 if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
4070 if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
4071 if (cs.dpl > cs_rpl)
4074 if (cs.dpl != cs_rpl)
4080 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
4084 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
4086 struct kvm_segment ss;
4087 unsigned int ss_rpl;
4089 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4090 ss_rpl = ss.selector & SEGMENT_RPL_MASK;
4094 if (ss.type != 3 && ss.type != 7)
4098 if (ss.dpl != ss_rpl) /* DPL != RPL */
4106 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
4108 struct kvm_segment var;
4111 vmx_get_segment(vcpu, &var, seg);
4112 rpl = var.selector & SEGMENT_RPL_MASK;
4120 if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
4121 if (var.dpl < rpl) /* DPL < RPL */
4125 /* TODO: Add other members to kvm_segment_field to allow checking for other access
4131 static bool tr_valid(struct kvm_vcpu *vcpu)
4133 struct kvm_segment tr;
4135 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
4139 if (tr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4141 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
4149 static bool ldtr_valid(struct kvm_vcpu *vcpu)
4151 struct kvm_segment ldtr;
4153 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
4157 if (ldtr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4167 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
4169 struct kvm_segment cs, ss;
4171 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4172 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4174 return ((cs.selector & SEGMENT_RPL_MASK) ==
4175 (ss.selector & SEGMENT_RPL_MASK));
4179 * Check if guest state is valid. Returns true if valid, false if
4181 * We assume that registers are always usable
4183 static bool guest_state_valid(struct kvm_vcpu *vcpu)
4185 if (enable_unrestricted_guest)
4188 /* real mode guest state checks */
4189 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
4190 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
4192 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
4194 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
4196 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
4198 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
4200 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
4203 /* protected mode guest state checks */
4204 if (!cs_ss_rpl_check(vcpu))
4206 if (!code_segment_valid(vcpu))
4208 if (!stack_segment_valid(vcpu))
4210 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
4212 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
4214 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
4216 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
4218 if (!tr_valid(vcpu))
4220 if (!ldtr_valid(vcpu))
4224 * - Add checks on RIP
4225 * - Add checks on RFLAGS
4231 static int init_rmode_tss(struct kvm *kvm)
4237 idx = srcu_read_lock(&kvm->srcu);
4238 fn = kvm->arch.tss_addr >> PAGE_SHIFT;
4239 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4242 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
4243 r = kvm_write_guest_page(kvm, fn++, &data,
4244 TSS_IOPB_BASE_OFFSET, sizeof(u16));
4247 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
4250 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4254 r = kvm_write_guest_page(kvm, fn, &data,
4255 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
4258 srcu_read_unlock(&kvm->srcu, idx);
4262 static int init_rmode_identity_map(struct kvm *kvm)
4265 kvm_pfn_t identity_map_pfn;
4271 /* Protect kvm->arch.ept_identity_pagetable_done. */
4272 mutex_lock(&kvm->slots_lock);
4274 if (likely(kvm->arch.ept_identity_pagetable_done))
4277 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
4279 r = alloc_identity_pagetable(kvm);
4283 idx = srcu_read_lock(&kvm->srcu);
4284 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
4287 /* Set up identity-mapping pagetable for EPT in real mode */
4288 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
4289 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
4290 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
4291 r = kvm_write_guest_page(kvm, identity_map_pfn,
4292 &tmp, i * sizeof(tmp), sizeof(tmp));
4296 kvm->arch.ept_identity_pagetable_done = true;
4299 srcu_read_unlock(&kvm->srcu, idx);
4302 mutex_unlock(&kvm->slots_lock);
4306 static void seg_setup(int seg)
4308 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4311 vmcs_write16(sf->selector, 0);
4312 vmcs_writel(sf->base, 0);
4313 vmcs_write32(sf->limit, 0xffff);
4315 if (seg == VCPU_SREG_CS)
4316 ar |= 0x08; /* code segment */
4318 vmcs_write32(sf->ar_bytes, ar);
4321 static int alloc_apic_access_page(struct kvm *kvm)
4326 mutex_lock(&kvm->slots_lock);
4327 if (kvm->arch.apic_access_page_done)
4329 r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
4330 APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
4334 page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
4335 if (is_error_page(page)) {
4341 * Do not pin the page in memory, so that memory hot-unplug
4342 * is able to migrate it.
4345 kvm->arch.apic_access_page_done = true;
4347 mutex_unlock(&kvm->slots_lock);
4351 static int alloc_identity_pagetable(struct kvm *kvm)
4353 /* Called with kvm->slots_lock held. */
4357 BUG_ON(kvm->arch.ept_identity_pagetable_done);
4359 r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
4360 kvm->arch.ept_identity_map_addr, PAGE_SIZE);
4365 static int allocate_vpid(void)
4371 spin_lock(&vmx_vpid_lock);
4372 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
4373 if (vpid < VMX_NR_VPIDS)
4374 __set_bit(vpid, vmx_vpid_bitmap);
4377 spin_unlock(&vmx_vpid_lock);
4381 static void free_vpid(int vpid)
4383 if (!enable_vpid || vpid == 0)
4385 spin_lock(&vmx_vpid_lock);
4386 __clear_bit(vpid, vmx_vpid_bitmap);
4387 spin_unlock(&vmx_vpid_lock);
4390 #define MSR_TYPE_R 1
4391 #define MSR_TYPE_W 2
4392 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
4395 int f = sizeof(unsigned long);
4397 if (!cpu_has_vmx_msr_bitmap())
4401 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4402 * have the write-low and read-high bitmap offsets the wrong way round.
4403 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4405 if (msr <= 0x1fff) {
4406 if (type & MSR_TYPE_R)
4408 __clear_bit(msr, msr_bitmap + 0x000 / f);
4410 if (type & MSR_TYPE_W)
4412 __clear_bit(msr, msr_bitmap + 0x800 / f);
4414 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4416 if (type & MSR_TYPE_R)
4418 __clear_bit(msr, msr_bitmap + 0x400 / f);
4420 if (type & MSR_TYPE_W)
4422 __clear_bit(msr, msr_bitmap + 0xc00 / f);
4427 static void __vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
4430 int f = sizeof(unsigned long);
4432 if (!cpu_has_vmx_msr_bitmap())
4436 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4437 * have the write-low and read-high bitmap offsets the wrong way round.
4438 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4440 if (msr <= 0x1fff) {
4441 if (type & MSR_TYPE_R)
4443 __set_bit(msr, msr_bitmap + 0x000 / f);
4445 if (type & MSR_TYPE_W)
4447 __set_bit(msr, msr_bitmap + 0x800 / f);
4449 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4451 if (type & MSR_TYPE_R)
4453 __set_bit(msr, msr_bitmap + 0x400 / f);
4455 if (type & MSR_TYPE_W)
4457 __set_bit(msr, msr_bitmap + 0xc00 / f);
4463 * If a msr is allowed by L0, we should check whether it is allowed by L1.
4464 * The corresponding bit will be cleared unless both of L0 and L1 allow it.
4466 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
4467 unsigned long *msr_bitmap_nested,
4470 int f = sizeof(unsigned long);
4472 if (!cpu_has_vmx_msr_bitmap()) {
4478 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4479 * have the write-low and read-high bitmap offsets the wrong way round.
4480 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4482 if (msr <= 0x1fff) {
4483 if (type & MSR_TYPE_R &&
4484 !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
4486 __clear_bit(msr, msr_bitmap_nested + 0x000 / f);
4488 if (type & MSR_TYPE_W &&
4489 !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
4491 __clear_bit(msr, msr_bitmap_nested + 0x800 / f);
4493 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4495 if (type & MSR_TYPE_R &&
4496 !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
4498 __clear_bit(msr, msr_bitmap_nested + 0x400 / f);
4500 if (type & MSR_TYPE_W &&
4501 !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
4503 __clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
4508 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
4511 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy,
4512 msr, MSR_TYPE_R | MSR_TYPE_W);
4513 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode,
4514 msr, MSR_TYPE_R | MSR_TYPE_W);
4517 static void vmx_enable_intercept_msr_read_x2apic(u32 msr)
4519 __vmx_enable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4521 __vmx_enable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4525 static void vmx_disable_intercept_msr_read_x2apic(u32 msr)
4527 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4529 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4533 static void vmx_disable_intercept_msr_write_x2apic(u32 msr)
4535 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4537 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4541 static bool vmx_get_enable_apicv(void)
4543 return enable_apicv;
4546 static int vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
4548 struct vcpu_vmx *vmx = to_vmx(vcpu);
4553 if (vmx->nested.pi_desc &&
4554 vmx->nested.pi_pending) {
4555 vmx->nested.pi_pending = false;
4556 if (!pi_test_and_clear_on(vmx->nested.pi_desc))
4559 max_irr = find_last_bit(
4560 (unsigned long *)vmx->nested.pi_desc->pir, 256);
4565 vapic_page = kmap(vmx->nested.virtual_apic_page);
4570 __kvm_apic_update_irr(vmx->nested.pi_desc->pir, vapic_page);
4571 kunmap(vmx->nested.virtual_apic_page);
4573 status = vmcs_read16(GUEST_INTR_STATUS);
4574 if ((u8)max_irr > ((u8)status & 0xff)) {
4576 status |= (u8)max_irr;
4577 vmcs_write16(GUEST_INTR_STATUS, status);
4583 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu)
4586 if (vcpu->mode == IN_GUEST_MODE) {
4587 struct vcpu_vmx *vmx = to_vmx(vcpu);
4590 * Currently, we don't support urgent interrupt,
4591 * all interrupts are recognized as non-urgent
4592 * interrupt, so we cannot post interrupts when
4595 * If the vcpu is in guest mode, it means it is
4596 * running instead of being scheduled out and
4597 * waiting in the run queue, and that's the only
4598 * case when 'SN' is set currently, warning if
4601 WARN_ON_ONCE(pi_test_sn(&vmx->pi_desc));
4603 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu),
4604 POSTED_INTR_VECTOR);
4611 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
4614 struct vcpu_vmx *vmx = to_vmx(vcpu);
4616 if (is_guest_mode(vcpu) &&
4617 vector == vmx->nested.posted_intr_nv) {
4618 /* the PIR and ON have been set by L1. */
4619 kvm_vcpu_trigger_posted_interrupt(vcpu);
4621 * If a posted intr is not recognized by hardware,
4622 * we will accomplish it in the next vmentry.
4624 vmx->nested.pi_pending = true;
4625 kvm_make_request(KVM_REQ_EVENT, vcpu);
4631 * Send interrupt to vcpu via posted interrupt way.
4632 * 1. If target vcpu is running(non-root mode), send posted interrupt
4633 * notification to vcpu and hardware will sync PIR to vIRR atomically.
4634 * 2. If target vcpu isn't running(root mode), kick it to pick up the
4635 * interrupt from PIR in next vmentry.
4637 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
4639 struct vcpu_vmx *vmx = to_vmx(vcpu);
4642 r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
4646 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
4649 r = pi_test_and_set_on(&vmx->pi_desc);
4650 kvm_make_request(KVM_REQ_EVENT, vcpu);
4651 if (r || !kvm_vcpu_trigger_posted_interrupt(vcpu))
4652 kvm_vcpu_kick(vcpu);
4655 static void vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
4657 struct vcpu_vmx *vmx = to_vmx(vcpu);
4659 if (!pi_test_and_clear_on(&vmx->pi_desc))
4662 kvm_apic_update_irr(vcpu, vmx->pi_desc.pir);
4666 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
4667 * will not change in the lifetime of the guest.
4668 * Note that host-state that does change is set elsewhere. E.g., host-state
4669 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
4671 static void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
4678 vmcs_writel(HOST_CR0, read_cr0() & ~X86_CR0_TS); /* 22.2.3 */
4679 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
4681 /* Save the most likely value for this task's CR4 in the VMCS. */
4682 cr4 = cr4_read_shadow();
4683 vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
4684 vmx->host_state.vmcs_host_cr4 = cr4;
4686 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
4687 #ifdef CONFIG_X86_64
4689 * Load null selectors, so we can avoid reloading them in
4690 * __vmx_load_host_state(), in case userspace uses the null selectors
4691 * too (the expected case).
4693 vmcs_write16(HOST_DS_SELECTOR, 0);
4694 vmcs_write16(HOST_ES_SELECTOR, 0);
4696 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4697 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4699 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4700 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
4702 native_store_idt(&dt);
4703 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
4704 vmx->host_idt_base = dt.address;
4706 vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
4708 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
4709 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
4710 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
4711 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
4713 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
4714 rdmsr(MSR_IA32_CR_PAT, low32, high32);
4715 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
4719 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
4721 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
4723 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
4724 if (is_guest_mode(&vmx->vcpu))
4725 vmx->vcpu.arch.cr4_guest_owned_bits &=
4726 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
4727 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
4730 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
4732 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
4734 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
4735 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
4736 return pin_based_exec_ctrl;
4739 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
4741 struct vcpu_vmx *vmx = to_vmx(vcpu);
4743 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
4746 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
4748 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
4750 if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
4751 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
4753 if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
4754 exec_control &= ~CPU_BASED_TPR_SHADOW;
4755 #ifdef CONFIG_X86_64
4756 exec_control |= CPU_BASED_CR8_STORE_EXITING |
4757 CPU_BASED_CR8_LOAD_EXITING;
4761 exec_control |= CPU_BASED_CR3_STORE_EXITING |
4762 CPU_BASED_CR3_LOAD_EXITING |
4763 CPU_BASED_INVLPG_EXITING;
4764 return exec_control;
4767 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
4769 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4770 if (!cpu_need_virtualize_apic_accesses(&vmx->vcpu))
4771 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4773 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4775 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4776 enable_unrestricted_guest = 0;
4777 /* Enable INVPCID for non-ept guests may cause performance regression. */
4778 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4780 if (!enable_unrestricted_guest)
4781 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4783 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4784 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
4785 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4786 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4787 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4788 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4790 We can NOT enable shadow_vmcs here because we don't have yet
4793 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4796 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4798 /* Currently, we allow L1 guest to directly run pcommit instruction. */
4799 exec_control &= ~SECONDARY_EXEC_PCOMMIT;
4801 return exec_control;
4804 static void ept_set_mmio_spte_mask(void)
4807 * EPT Misconfigurations can be generated if the value of bits 2:0
4808 * of an EPT paging-structure entry is 110b (write/execute).
4809 * Also, magic bits (0x3ull << 62) is set to quickly identify mmio
4812 kvm_mmu_set_mmio_spte_mask((0x3ull << 62) | 0x6ull);
4815 #define VMX_XSS_EXIT_BITMAP 0
4817 * Sets up the vmcs for emulated real mode.
4819 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
4821 #ifdef CONFIG_X86_64
4827 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
4828 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
4830 if (enable_shadow_vmcs) {
4831 vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
4832 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
4834 if (cpu_has_vmx_msr_bitmap())
4835 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
4837 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4840 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
4842 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
4844 if (cpu_has_secondary_exec_ctrls())
4845 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
4846 vmx_secondary_exec_control(vmx));
4848 if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
4849 vmcs_write64(EOI_EXIT_BITMAP0, 0);
4850 vmcs_write64(EOI_EXIT_BITMAP1, 0);
4851 vmcs_write64(EOI_EXIT_BITMAP2, 0);
4852 vmcs_write64(EOI_EXIT_BITMAP3, 0);
4854 vmcs_write16(GUEST_INTR_STATUS, 0);
4856 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4857 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4861 vmcs_write32(PLE_GAP, ple_gap);
4862 vmx->ple_window = ple_window;
4863 vmx->ple_window_dirty = true;
4866 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4867 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4868 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
4870 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
4871 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
4872 vmx_set_constant_host_state(vmx);
4873 #ifdef CONFIG_X86_64
4874 rdmsrl(MSR_FS_BASE, a);
4875 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
4876 rdmsrl(MSR_GS_BASE, a);
4877 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
4879 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4880 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4883 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4884 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4885 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
4886 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4887 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
4889 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4890 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4892 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
4893 u32 index = vmx_msr_index[i];
4894 u32 data_low, data_high;
4897 if (rdmsr_safe(index, &data_low, &data_high) < 0)
4899 if (wrmsr_safe(index, data_low, data_high) < 0)
4901 vmx->guest_msrs[j].index = i;
4902 vmx->guest_msrs[j].data = 0;
4903 vmx->guest_msrs[j].mask = -1ull;
4908 vm_exit_controls_init(vmx, vmcs_config.vmexit_ctrl);
4910 /* 22.2.1, 20.8.1 */
4911 vm_entry_controls_init(vmx, vmcs_config.vmentry_ctrl);
4913 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
4914 set_cr4_guest_host_mask(vmx);
4916 if (vmx_xsaves_supported())
4917 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4922 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4924 struct vcpu_vmx *vmx = to_vmx(vcpu);
4925 struct msr_data apic_base_msr;
4928 vmx->rmode.vm86_active = 0;
4930 vmx->soft_vnmi_blocked = 0;
4932 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4933 kvm_set_cr8(vcpu, 0);
4936 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
4937 MSR_IA32_APICBASE_ENABLE;
4938 if (kvm_vcpu_is_reset_bsp(vcpu))
4939 apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4940 apic_base_msr.host_initiated = true;
4941 kvm_set_apic_base(vcpu, &apic_base_msr);
4944 vmx_segment_cache_clear(vmx);
4946 seg_setup(VCPU_SREG_CS);
4947 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4948 vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
4950 seg_setup(VCPU_SREG_DS);
4951 seg_setup(VCPU_SREG_ES);
4952 seg_setup(VCPU_SREG_FS);
4953 seg_setup(VCPU_SREG_GS);
4954 seg_setup(VCPU_SREG_SS);
4956 vmcs_write16(GUEST_TR_SELECTOR, 0);
4957 vmcs_writel(GUEST_TR_BASE, 0);
4958 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4959 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4961 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4962 vmcs_writel(GUEST_LDTR_BASE, 0);
4963 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4964 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4967 vmcs_write32(GUEST_SYSENTER_CS, 0);
4968 vmcs_writel(GUEST_SYSENTER_ESP, 0);
4969 vmcs_writel(GUEST_SYSENTER_EIP, 0);
4970 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4973 vmcs_writel(GUEST_RFLAGS, 0x02);
4974 kvm_rip_write(vcpu, 0xfff0);
4976 vmcs_writel(GUEST_GDTR_BASE, 0);
4977 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4979 vmcs_writel(GUEST_IDTR_BASE, 0);
4980 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4982 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4983 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4984 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4988 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
4990 if (cpu_has_vmx_tpr_shadow() && !init_event) {
4991 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4992 if (cpu_need_tpr_shadow(vcpu))
4993 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4994 __pa(vcpu->arch.apic->regs));
4995 vmcs_write32(TPR_THRESHOLD, 0);
4998 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
5000 if (kvm_vcpu_apicv_active(vcpu))
5001 memset(&vmx->pi_desc, 0, sizeof(struct pi_desc));
5004 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
5006 cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
5007 vmx_set_cr0(vcpu, cr0); /* enter rmode */
5008 vmx->vcpu.arch.cr0 = cr0;
5009 vmx_set_cr4(vcpu, 0);
5010 vmx_set_efer(vcpu, 0);
5011 vmx_fpu_activate(vcpu);
5012 update_exception_bitmap(vcpu);
5014 vpid_sync_context(vmx->vpid);
5018 * In nested virtualization, check if L1 asked to exit on external interrupts.
5019 * For most existing hypervisors, this will always return true.
5021 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
5023 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5024 PIN_BASED_EXT_INTR_MASK;
5028 * In nested virtualization, check if L1 has set
5029 * VM_EXIT_ACK_INTR_ON_EXIT
5031 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
5033 return get_vmcs12(vcpu)->vm_exit_controls &
5034 VM_EXIT_ACK_INTR_ON_EXIT;
5037 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
5039 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5040 PIN_BASED_NMI_EXITING;
5043 static void enable_irq_window(struct kvm_vcpu *vcpu)
5045 u32 cpu_based_vm_exec_control;
5047 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5048 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
5049 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5052 static void enable_nmi_window(struct kvm_vcpu *vcpu)
5054 u32 cpu_based_vm_exec_control;
5056 if (!cpu_has_virtual_nmis() ||
5057 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
5058 enable_irq_window(vcpu);
5062 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5063 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
5064 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5067 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
5069 struct vcpu_vmx *vmx = to_vmx(vcpu);
5071 int irq = vcpu->arch.interrupt.nr;
5073 trace_kvm_inj_virq(irq);
5075 ++vcpu->stat.irq_injections;
5076 if (vmx->rmode.vm86_active) {
5078 if (vcpu->arch.interrupt.soft)
5079 inc_eip = vcpu->arch.event_exit_inst_len;
5080 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
5081 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5084 intr = irq | INTR_INFO_VALID_MASK;
5085 if (vcpu->arch.interrupt.soft) {
5086 intr |= INTR_TYPE_SOFT_INTR;
5087 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
5088 vmx->vcpu.arch.event_exit_inst_len);
5090 intr |= INTR_TYPE_EXT_INTR;
5091 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
5094 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
5096 struct vcpu_vmx *vmx = to_vmx(vcpu);
5098 if (is_guest_mode(vcpu))
5101 if (!cpu_has_virtual_nmis()) {
5103 * Tracking the NMI-blocked state in software is built upon
5104 * finding the next open IRQ window. This, in turn, depends on
5105 * well-behaving guests: They have to keep IRQs disabled at
5106 * least as long as the NMI handler runs. Otherwise we may
5107 * cause NMI nesting, maybe breaking the guest. But as this is
5108 * highly unlikely, we can live with the residual risk.
5110 vmx->soft_vnmi_blocked = 1;
5111 vmx->vnmi_blocked_time = 0;
5114 ++vcpu->stat.nmi_injections;
5115 vmx->nmi_known_unmasked = false;
5116 if (vmx->rmode.vm86_active) {
5117 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
5118 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5121 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
5122 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
5125 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
5127 if (!cpu_has_virtual_nmis())
5128 return to_vmx(vcpu)->soft_vnmi_blocked;
5129 if (to_vmx(vcpu)->nmi_known_unmasked)
5131 return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
5134 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
5136 struct vcpu_vmx *vmx = to_vmx(vcpu);
5138 if (!cpu_has_virtual_nmis()) {
5139 if (vmx->soft_vnmi_blocked != masked) {
5140 vmx->soft_vnmi_blocked = masked;
5141 vmx->vnmi_blocked_time = 0;
5144 vmx->nmi_known_unmasked = !masked;
5146 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5147 GUEST_INTR_STATE_NMI);
5149 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
5150 GUEST_INTR_STATE_NMI);
5154 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
5156 if (to_vmx(vcpu)->nested.nested_run_pending)
5159 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
5162 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5163 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
5164 | GUEST_INTR_STATE_NMI));
5167 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
5169 return (!to_vmx(vcpu)->nested.nested_run_pending &&
5170 vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
5171 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5172 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
5175 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
5179 ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
5183 kvm->arch.tss_addr = addr;
5184 return init_rmode_tss(kvm);
5187 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
5192 * Update instruction length as we may reinject the exception
5193 * from user space while in guest debugging mode.
5195 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
5196 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5197 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5201 if (vcpu->guest_debug &
5202 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5219 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
5220 int vec, u32 err_code)
5223 * Instruction with address size override prefix opcode 0x67
5224 * Cause the #SS fault with 0 error code in VM86 mode.
5226 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
5227 if (emulate_instruction(vcpu, 0) == EMULATE_DONE) {
5228 if (vcpu->arch.halt_request) {
5229 vcpu->arch.halt_request = 0;
5230 return kvm_vcpu_halt(vcpu);
5238 * Forward all other exceptions that are valid in real mode.
5239 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
5240 * the required debugging infrastructure rework.
5242 kvm_queue_exception(vcpu, vec);
5247 * Trigger machine check on the host. We assume all the MSRs are already set up
5248 * by the CPU and that we still run on the same CPU as the MCE occurred on.
5249 * We pass a fake environment to the machine check handler because we want
5250 * the guest to be always treated like user space, no matter what context
5251 * it used internally.
5253 static void kvm_machine_check(void)
5255 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
5256 struct pt_regs regs = {
5257 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
5258 .flags = X86_EFLAGS_IF,
5261 do_machine_check(®s, 0);
5265 static int handle_machine_check(struct kvm_vcpu *vcpu)
5267 /* already handled by vcpu_run */
5271 static int handle_exception(struct kvm_vcpu *vcpu)
5273 struct vcpu_vmx *vmx = to_vmx(vcpu);
5274 struct kvm_run *kvm_run = vcpu->run;
5275 u32 intr_info, ex_no, error_code;
5276 unsigned long cr2, rip, dr6;
5278 enum emulation_result er;
5280 vect_info = vmx->idt_vectoring_info;
5281 intr_info = vmx->exit_intr_info;
5283 if (is_machine_check(intr_info))
5284 return handle_machine_check(vcpu);
5286 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
5287 return 1; /* already handled by vmx_vcpu_run() */
5289 if (is_no_device(intr_info)) {
5290 vmx_fpu_activate(vcpu);
5294 if (is_invalid_opcode(intr_info)) {
5295 if (is_guest_mode(vcpu)) {
5296 kvm_queue_exception(vcpu, UD_VECTOR);
5299 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
5300 if (er != EMULATE_DONE)
5301 kvm_queue_exception(vcpu, UD_VECTOR);
5306 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
5307 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
5310 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
5311 * MMIO, it is better to report an internal error.
5312 * See the comments in vmx_handle_exit.
5314 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
5315 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
5316 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5317 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
5318 vcpu->run->internal.ndata = 3;
5319 vcpu->run->internal.data[0] = vect_info;
5320 vcpu->run->internal.data[1] = intr_info;
5321 vcpu->run->internal.data[2] = error_code;
5325 if (is_page_fault(intr_info)) {
5326 /* EPT won't cause page fault directly */
5328 cr2 = vmcs_readl(EXIT_QUALIFICATION);
5329 trace_kvm_page_fault(cr2, error_code);
5331 if (kvm_event_needs_reinjection(vcpu))
5332 kvm_mmu_unprotect_page_virt(vcpu, cr2);
5333 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
5336 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
5338 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
5339 return handle_rmode_exception(vcpu, ex_no, error_code);
5343 kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
5346 dr6 = vmcs_readl(EXIT_QUALIFICATION);
5347 if (!(vcpu->guest_debug &
5348 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
5349 vcpu->arch.dr6 &= ~15;
5350 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5351 if (!(dr6 & ~DR6_RESERVED)) /* icebp */
5352 skip_emulated_instruction(vcpu);
5354 kvm_queue_exception(vcpu, DB_VECTOR);
5357 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
5358 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
5362 * Update instruction length as we may reinject #BP from
5363 * user space while in guest debugging mode. Reading it for
5364 * #DB as well causes no harm, it is not used in that case.
5366 vmx->vcpu.arch.event_exit_inst_len =
5367 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5368 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5369 rip = kvm_rip_read(vcpu);
5370 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
5371 kvm_run->debug.arch.exception = ex_no;
5374 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
5375 kvm_run->ex.exception = ex_no;
5376 kvm_run->ex.error_code = error_code;
5382 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
5384 ++vcpu->stat.irq_exits;
5388 static int handle_triple_fault(struct kvm_vcpu *vcpu)
5390 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5394 static int handle_io(struct kvm_vcpu *vcpu)
5396 unsigned long exit_qualification;
5397 int size, in, string;
5400 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5401 string = (exit_qualification & 16) != 0;
5402 in = (exit_qualification & 8) != 0;
5404 ++vcpu->stat.io_exits;
5407 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5409 port = exit_qualification >> 16;
5410 size = (exit_qualification & 7) + 1;
5411 skip_emulated_instruction(vcpu);
5413 return kvm_fast_pio_out(vcpu, size, port);
5417 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
5420 * Patch in the VMCALL instruction:
5422 hypercall[0] = 0x0f;
5423 hypercall[1] = 0x01;
5424 hypercall[2] = 0xc1;
5427 static bool nested_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
5429 unsigned long always_on = VMXON_CR0_ALWAYSON;
5430 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5432 if (to_vmx(vcpu)->nested.nested_vmx_secondary_ctls_high &
5433 SECONDARY_EXEC_UNRESTRICTED_GUEST &&
5434 nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
5435 always_on &= ~(X86_CR0_PE | X86_CR0_PG);
5436 return (val & always_on) == always_on;
5439 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
5440 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
5442 if (is_guest_mode(vcpu)) {
5443 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5444 unsigned long orig_val = val;
5447 * We get here when L2 changed cr0 in a way that did not change
5448 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
5449 * but did change L0 shadowed bits. So we first calculate the
5450 * effective cr0 value that L1 would like to write into the
5451 * hardware. It consists of the L2-owned bits from the new
5452 * value combined with the L1-owned bits from L1's guest_cr0.
5454 val = (val & ~vmcs12->cr0_guest_host_mask) |
5455 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
5457 if (!nested_cr0_valid(vcpu, val))
5460 if (kvm_set_cr0(vcpu, val))
5462 vmcs_writel(CR0_READ_SHADOW, orig_val);
5465 if (to_vmx(vcpu)->nested.vmxon &&
5466 ((val & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON))
5468 return kvm_set_cr0(vcpu, val);
5472 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
5474 if (is_guest_mode(vcpu)) {
5475 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5476 unsigned long orig_val = val;
5478 /* analogously to handle_set_cr0 */
5479 val = (val & ~vmcs12->cr4_guest_host_mask) |
5480 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
5481 if (kvm_set_cr4(vcpu, val))
5483 vmcs_writel(CR4_READ_SHADOW, orig_val);
5486 return kvm_set_cr4(vcpu, val);
5489 /* called to set cr0 as approriate for clts instruction exit. */
5490 static void handle_clts(struct kvm_vcpu *vcpu)
5492 if (is_guest_mode(vcpu)) {
5494 * We get here when L2 did CLTS, and L1 didn't shadow CR0.TS
5495 * but we did (!fpu_active). We need to keep GUEST_CR0.TS on,
5496 * just pretend it's off (also in arch.cr0 for fpu_activate).
5498 vmcs_writel(CR0_READ_SHADOW,
5499 vmcs_readl(CR0_READ_SHADOW) & ~X86_CR0_TS);
5500 vcpu->arch.cr0 &= ~X86_CR0_TS;
5502 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
5505 static int handle_cr(struct kvm_vcpu *vcpu)
5507 unsigned long exit_qualification, val;
5512 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5513 cr = exit_qualification & 15;
5514 reg = (exit_qualification >> 8) & 15;
5515 switch ((exit_qualification >> 4) & 3) {
5516 case 0: /* mov to cr */
5517 val = kvm_register_readl(vcpu, reg);
5518 trace_kvm_cr_write(cr, val);
5521 err = handle_set_cr0(vcpu, val);
5522 kvm_complete_insn_gp(vcpu, err);
5525 err = kvm_set_cr3(vcpu, val);
5526 kvm_complete_insn_gp(vcpu, err);
5529 err = handle_set_cr4(vcpu, val);
5530 kvm_complete_insn_gp(vcpu, err);
5533 u8 cr8_prev = kvm_get_cr8(vcpu);
5535 err = kvm_set_cr8(vcpu, cr8);
5536 kvm_complete_insn_gp(vcpu, err);
5537 if (lapic_in_kernel(vcpu))
5539 if (cr8_prev <= cr8)
5541 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
5548 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
5549 skip_emulated_instruction(vcpu);
5550 vmx_fpu_activate(vcpu);
5552 case 1: /*mov from cr*/
5555 val = kvm_read_cr3(vcpu);
5556 kvm_register_write(vcpu, reg, val);
5557 trace_kvm_cr_read(cr, val);
5558 skip_emulated_instruction(vcpu);
5561 val = kvm_get_cr8(vcpu);
5562 kvm_register_write(vcpu, reg, val);
5563 trace_kvm_cr_read(cr, val);
5564 skip_emulated_instruction(vcpu);
5569 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5570 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
5571 kvm_lmsw(vcpu, val);
5573 skip_emulated_instruction(vcpu);
5578 vcpu->run->exit_reason = 0;
5579 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
5580 (int)(exit_qualification >> 4) & 3, cr);
5584 static int handle_dr(struct kvm_vcpu *vcpu)
5586 unsigned long exit_qualification;
5589 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5590 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
5592 /* First, if DR does not exist, trigger UD */
5593 if (!kvm_require_dr(vcpu, dr))
5596 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
5597 if (!kvm_require_cpl(vcpu, 0))
5599 dr7 = vmcs_readl(GUEST_DR7);
5602 * As the vm-exit takes precedence over the debug trap, we
5603 * need to emulate the latter, either for the host or the
5604 * guest debugging itself.
5606 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5607 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
5608 vcpu->run->debug.arch.dr7 = dr7;
5609 vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
5610 vcpu->run->debug.arch.exception = DB_VECTOR;
5611 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
5614 vcpu->arch.dr6 &= ~15;
5615 vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
5616 kvm_queue_exception(vcpu, DB_VECTOR);
5621 if (vcpu->guest_debug == 0) {
5622 u32 cpu_based_vm_exec_control;
5624 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5625 cpu_based_vm_exec_control &= ~CPU_BASED_MOV_DR_EXITING;
5626 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5629 * No more DR vmexits; force a reload of the debug registers
5630 * and reenter on this instruction. The next vmexit will
5631 * retrieve the full state of the debug registers.
5633 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
5637 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
5638 if (exit_qualification & TYPE_MOV_FROM_DR) {
5641 if (kvm_get_dr(vcpu, dr, &val))
5643 kvm_register_write(vcpu, reg, val);
5645 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
5648 skip_emulated_instruction(vcpu);
5652 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
5654 return vcpu->arch.dr6;
5657 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
5661 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
5663 u32 cpu_based_vm_exec_control;
5665 get_debugreg(vcpu->arch.db[0], 0);
5666 get_debugreg(vcpu->arch.db[1], 1);
5667 get_debugreg(vcpu->arch.db[2], 2);
5668 get_debugreg(vcpu->arch.db[3], 3);
5669 get_debugreg(vcpu->arch.dr6, 6);
5670 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
5672 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
5674 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5675 cpu_based_vm_exec_control |= CPU_BASED_MOV_DR_EXITING;
5676 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5679 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
5681 vmcs_writel(GUEST_DR7, val);
5684 static int handle_cpuid(struct kvm_vcpu *vcpu)
5686 kvm_emulate_cpuid(vcpu);
5690 static int handle_rdmsr(struct kvm_vcpu *vcpu)
5692 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5693 struct msr_data msr_info;
5695 msr_info.index = ecx;
5696 msr_info.host_initiated = false;
5697 if (vmx_get_msr(vcpu, &msr_info)) {
5698 trace_kvm_msr_read_ex(ecx);
5699 kvm_inject_gp(vcpu, 0);
5703 trace_kvm_msr_read(ecx, msr_info.data);
5705 /* FIXME: handling of bits 32:63 of rax, rdx */
5706 vcpu->arch.regs[VCPU_REGS_RAX] = msr_info.data & -1u;
5707 vcpu->arch.regs[VCPU_REGS_RDX] = (msr_info.data >> 32) & -1u;
5708 skip_emulated_instruction(vcpu);
5712 static int handle_wrmsr(struct kvm_vcpu *vcpu)
5714 struct msr_data msr;
5715 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5716 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
5717 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
5721 msr.host_initiated = false;
5722 if (kvm_set_msr(vcpu, &msr) != 0) {
5723 trace_kvm_msr_write_ex(ecx, data);
5724 kvm_inject_gp(vcpu, 0);
5728 trace_kvm_msr_write(ecx, data);
5729 skip_emulated_instruction(vcpu);
5733 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
5735 kvm_make_request(KVM_REQ_EVENT, vcpu);
5739 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
5741 u32 cpu_based_vm_exec_control;
5743 /* clear pending irq */
5744 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5745 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
5746 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5748 kvm_make_request(KVM_REQ_EVENT, vcpu);
5750 ++vcpu->stat.irq_window_exits;
5754 static int handle_halt(struct kvm_vcpu *vcpu)
5756 return kvm_emulate_halt(vcpu);
5759 static int handle_vmcall(struct kvm_vcpu *vcpu)
5761 kvm_emulate_hypercall(vcpu);
5765 static int handle_invd(struct kvm_vcpu *vcpu)
5767 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5770 static int handle_invlpg(struct kvm_vcpu *vcpu)
5772 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5774 kvm_mmu_invlpg(vcpu, exit_qualification);
5775 skip_emulated_instruction(vcpu);
5779 static int handle_rdpmc(struct kvm_vcpu *vcpu)
5783 err = kvm_rdpmc(vcpu);
5784 kvm_complete_insn_gp(vcpu, err);
5789 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5791 kvm_emulate_wbinvd(vcpu);
5795 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5797 u64 new_bv = kvm_read_edx_eax(vcpu);
5798 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5800 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5801 skip_emulated_instruction(vcpu);
5805 static int handle_xsaves(struct kvm_vcpu *vcpu)
5807 skip_emulated_instruction(vcpu);
5808 WARN(1, "this should never happen\n");
5812 static int handle_xrstors(struct kvm_vcpu *vcpu)
5814 skip_emulated_instruction(vcpu);
5815 WARN(1, "this should never happen\n");
5819 static int handle_apic_access(struct kvm_vcpu *vcpu)
5821 if (likely(fasteoi)) {
5822 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5823 int access_type, offset;
5825 access_type = exit_qualification & APIC_ACCESS_TYPE;
5826 offset = exit_qualification & APIC_ACCESS_OFFSET;
5828 * Sane guest uses MOV to write EOI, with written value
5829 * not cared. So make a short-circuit here by avoiding
5830 * heavy instruction emulation.
5832 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5833 (offset == APIC_EOI)) {
5834 kvm_lapic_set_eoi(vcpu);
5835 skip_emulated_instruction(vcpu);
5839 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5842 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5844 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5845 int vector = exit_qualification & 0xff;
5847 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5848 kvm_apic_set_eoi_accelerated(vcpu, vector);
5852 static int handle_apic_write(struct kvm_vcpu *vcpu)
5854 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5855 u32 offset = exit_qualification & 0xfff;
5857 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
5858 kvm_apic_write_nodecode(vcpu, offset);
5862 static int handle_task_switch(struct kvm_vcpu *vcpu)
5864 struct vcpu_vmx *vmx = to_vmx(vcpu);
5865 unsigned long exit_qualification;
5866 bool has_error_code = false;
5869 int reason, type, idt_v, idt_index;
5871 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5872 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5873 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5875 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5877 reason = (u32)exit_qualification >> 30;
5878 if (reason == TASK_SWITCH_GATE && idt_v) {
5880 case INTR_TYPE_NMI_INTR:
5881 vcpu->arch.nmi_injected = false;
5882 vmx_set_nmi_mask(vcpu, true);
5884 case INTR_TYPE_EXT_INTR:
5885 case INTR_TYPE_SOFT_INTR:
5886 kvm_clear_interrupt_queue(vcpu);
5888 case INTR_TYPE_HARD_EXCEPTION:
5889 if (vmx->idt_vectoring_info &
5890 VECTORING_INFO_DELIVER_CODE_MASK) {
5891 has_error_code = true;
5893 vmcs_read32(IDT_VECTORING_ERROR_CODE);
5896 case INTR_TYPE_SOFT_EXCEPTION:
5897 kvm_clear_exception_queue(vcpu);
5903 tss_selector = exit_qualification;
5905 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5906 type != INTR_TYPE_EXT_INTR &&
5907 type != INTR_TYPE_NMI_INTR))
5908 skip_emulated_instruction(vcpu);
5910 if (kvm_task_switch(vcpu, tss_selector,
5911 type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
5912 has_error_code, error_code) == EMULATE_FAIL) {
5913 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5914 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5915 vcpu->run->internal.ndata = 0;
5920 * TODO: What about debug traps on tss switch?
5921 * Are we supposed to inject them and update dr6?
5927 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5929 unsigned long exit_qualification;
5934 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5936 gla_validity = (exit_qualification >> 7) & 0x3;
5937 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
5938 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
5939 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
5940 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
5941 vmcs_readl(GUEST_LINEAR_ADDRESS));
5942 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
5943 (long unsigned int)exit_qualification);
5944 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5945 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
5950 * EPT violation happened while executing iret from NMI,
5951 * "blocked by NMI" bit has to be set before next VM entry.
5952 * There are errata that may cause this bit to not be set:
5955 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5956 cpu_has_virtual_nmis() &&
5957 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
5958 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5960 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5961 trace_kvm_page_fault(gpa, exit_qualification);
5963 /* It is a write fault? */
5964 error_code = exit_qualification & PFERR_WRITE_MASK;
5965 /* It is a fetch fault? */
5966 error_code |= (exit_qualification << 2) & PFERR_FETCH_MASK;
5967 /* ept page table is present? */
5968 error_code |= (exit_qualification >> 3) & PFERR_PRESENT_MASK;
5970 vcpu->arch.exit_qualification = exit_qualification;
5972 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5975 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5980 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5981 if (!kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5982 skip_emulated_instruction(vcpu);
5983 trace_kvm_fast_mmio(gpa);
5987 ret = handle_mmio_page_fault(vcpu, gpa, true);
5988 if (likely(ret == RET_MMIO_PF_EMULATE))
5989 return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) ==
5992 if (unlikely(ret == RET_MMIO_PF_INVALID))
5993 return kvm_mmu_page_fault(vcpu, gpa, 0, NULL, 0);
5995 if (unlikely(ret == RET_MMIO_PF_RETRY))
5998 /* It is the real ept misconfig */
6001 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
6002 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
6007 static int handle_nmi_window(struct kvm_vcpu *vcpu)
6009 u32 cpu_based_vm_exec_control;
6011 /* clear pending NMI */
6012 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
6013 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
6014 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
6015 ++vcpu->stat.nmi_window_exits;
6016 kvm_make_request(KVM_REQ_EVENT, vcpu);
6021 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
6023 struct vcpu_vmx *vmx = to_vmx(vcpu);
6024 enum emulation_result err = EMULATE_DONE;
6027 bool intr_window_requested;
6028 unsigned count = 130;
6030 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
6031 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
6033 while (vmx->emulation_required && count-- != 0) {
6034 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
6035 return handle_interrupt_window(&vmx->vcpu);
6037 if (test_bit(KVM_REQ_EVENT, &vcpu->requests))
6040 err = emulate_instruction(vcpu, EMULTYPE_NO_REEXECUTE);
6042 if (err == EMULATE_USER_EXIT) {
6043 ++vcpu->stat.mmio_exits;
6048 if (err != EMULATE_DONE) {
6049 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6050 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6051 vcpu->run->internal.ndata = 0;
6055 if (vcpu->arch.halt_request) {
6056 vcpu->arch.halt_request = 0;
6057 ret = kvm_vcpu_halt(vcpu);
6061 if (signal_pending(current))
6071 static int __grow_ple_window(int val)
6073 if (ple_window_grow < 1)
6076 val = min(val, ple_window_actual_max);
6078 if (ple_window_grow < ple_window)
6079 val *= ple_window_grow;
6081 val += ple_window_grow;
6086 static int __shrink_ple_window(int val, int modifier, int minimum)
6091 if (modifier < ple_window)
6096 return max(val, minimum);
6099 static void grow_ple_window(struct kvm_vcpu *vcpu)
6101 struct vcpu_vmx *vmx = to_vmx(vcpu);
6102 int old = vmx->ple_window;
6104 vmx->ple_window = __grow_ple_window(old);
6106 if (vmx->ple_window != old)
6107 vmx->ple_window_dirty = true;
6109 trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old);
6112 static void shrink_ple_window(struct kvm_vcpu *vcpu)
6114 struct vcpu_vmx *vmx = to_vmx(vcpu);
6115 int old = vmx->ple_window;
6117 vmx->ple_window = __shrink_ple_window(old,
6118 ple_window_shrink, ple_window);
6120 if (vmx->ple_window != old)
6121 vmx->ple_window_dirty = true;
6123 trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old);
6127 * ple_window_actual_max is computed to be one grow_ple_window() below
6128 * ple_window_max. (See __grow_ple_window for the reason.)
6129 * This prevents overflows, because ple_window_max is int.
6130 * ple_window_max effectively rounded down to a multiple of ple_window_grow in
6132 * ple_window_max is also prevented from setting vmx->ple_window < ple_window.
6134 static void update_ple_window_actual_max(void)
6136 ple_window_actual_max =
6137 __shrink_ple_window(max(ple_window_max, ple_window),
6138 ple_window_grow, INT_MIN);
6142 * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
6144 static void wakeup_handler(void)
6146 struct kvm_vcpu *vcpu;
6147 int cpu = smp_processor_id();
6149 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6150 list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
6151 blocked_vcpu_list) {
6152 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6154 if (pi_test_on(pi_desc) == 1)
6155 kvm_vcpu_kick(vcpu);
6157 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6160 static __init int hardware_setup(void)
6162 int r = -ENOMEM, i, msr;
6164 rdmsrl_safe(MSR_EFER, &host_efer);
6166 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
6167 kvm_define_shared_msr(i, vmx_msr_index[i]);
6169 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
6170 if (!vmx_io_bitmap_a)
6173 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
6174 if (!vmx_io_bitmap_b)
6177 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
6178 if (!vmx_msr_bitmap_legacy)
6181 vmx_msr_bitmap_legacy_x2apic =
6182 (unsigned long *)__get_free_page(GFP_KERNEL);
6183 if (!vmx_msr_bitmap_legacy_x2apic)
6186 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
6187 if (!vmx_msr_bitmap_longmode)
6190 vmx_msr_bitmap_longmode_x2apic =
6191 (unsigned long *)__get_free_page(GFP_KERNEL);
6192 if (!vmx_msr_bitmap_longmode_x2apic)
6196 vmx_msr_bitmap_nested =
6197 (unsigned long *)__get_free_page(GFP_KERNEL);
6198 if (!vmx_msr_bitmap_nested)
6202 vmx_vmread_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
6203 if (!vmx_vmread_bitmap)
6206 vmx_vmwrite_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
6207 if (!vmx_vmwrite_bitmap)
6210 memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
6211 memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
6214 * Allow direct access to the PC debug port (it is often used for I/O
6215 * delays, but the vmexits simply slow things down).
6217 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
6218 clear_bit(0x80, vmx_io_bitmap_a);
6220 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
6222 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
6223 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
6225 memset(vmx_msr_bitmap_nested, 0xff, PAGE_SIZE);
6227 if (setup_vmcs_config(&vmcs_config) < 0) {
6232 if (boot_cpu_has(X86_FEATURE_NX))
6233 kvm_enable_efer_bits(EFER_NX);
6235 if (!cpu_has_vmx_vpid())
6237 if (!cpu_has_vmx_shadow_vmcs())
6238 enable_shadow_vmcs = 0;
6239 if (enable_shadow_vmcs)
6240 init_vmcs_shadow_fields();
6242 if (!cpu_has_vmx_ept() ||
6243 !cpu_has_vmx_ept_4levels()) {
6245 enable_unrestricted_guest = 0;
6246 enable_ept_ad_bits = 0;
6249 if (!cpu_has_vmx_ept_ad_bits())
6250 enable_ept_ad_bits = 0;
6252 if (!cpu_has_vmx_unrestricted_guest())
6253 enable_unrestricted_guest = 0;
6255 if (!cpu_has_vmx_flexpriority())
6256 flexpriority_enabled = 0;
6259 * set_apic_access_page_addr() is used to reload apic access
6260 * page upon invalidation. No need to do anything if not
6261 * using the APIC_ACCESS_ADDR VMCS field.
6263 if (!flexpriority_enabled)
6264 kvm_x86_ops->set_apic_access_page_addr = NULL;
6266 if (!cpu_has_vmx_tpr_shadow())
6267 kvm_x86_ops->update_cr8_intercept = NULL;
6269 if (enable_ept && !cpu_has_vmx_ept_2m_page())
6270 kvm_disable_largepages();
6272 if (!cpu_has_vmx_ple())
6275 if (!cpu_has_vmx_apicv())
6278 if (cpu_has_vmx_tsc_scaling()) {
6279 kvm_has_tsc_control = true;
6280 kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
6281 kvm_tsc_scaling_ratio_frac_bits = 48;
6284 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
6285 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
6286 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
6287 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
6288 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
6289 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
6290 vmx_disable_intercept_for_msr(MSR_IA32_BNDCFGS, true);
6292 memcpy(vmx_msr_bitmap_legacy_x2apic,
6293 vmx_msr_bitmap_legacy, PAGE_SIZE);
6294 memcpy(vmx_msr_bitmap_longmode_x2apic,
6295 vmx_msr_bitmap_longmode, PAGE_SIZE);
6297 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
6300 for (msr = 0x800; msr <= 0x8ff; msr++)
6301 vmx_disable_intercept_msr_read_x2apic(msr);
6303 /* According SDM, in x2apic mode, the whole id reg is used.
6304 * But in KVM, it only use the highest eight bits. Need to
6306 vmx_enable_intercept_msr_read_x2apic(0x802);
6308 vmx_enable_intercept_msr_read_x2apic(0x839);
6310 vmx_disable_intercept_msr_write_x2apic(0x808);
6312 vmx_disable_intercept_msr_write_x2apic(0x80b);
6314 vmx_disable_intercept_msr_write_x2apic(0x83f);
6318 kvm_mmu_set_mask_ptes(0ull,
6319 (enable_ept_ad_bits) ? VMX_EPT_ACCESS_BIT : 0ull,
6320 (enable_ept_ad_bits) ? VMX_EPT_DIRTY_BIT : 0ull,
6321 0ull, VMX_EPT_EXECUTABLE_MASK);
6322 ept_set_mmio_spte_mask();
6327 update_ple_window_actual_max();
6330 * Only enable PML when hardware supports PML feature, and both EPT
6331 * and EPT A/D bit features are enabled -- PML depends on them to work.
6333 if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
6337 kvm_x86_ops->slot_enable_log_dirty = NULL;
6338 kvm_x86_ops->slot_disable_log_dirty = NULL;
6339 kvm_x86_ops->flush_log_dirty = NULL;
6340 kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
6343 kvm_set_posted_intr_wakeup_handler(wakeup_handler);
6345 return alloc_kvm_area();
6348 free_page((unsigned long)vmx_vmwrite_bitmap);
6350 free_page((unsigned long)vmx_vmread_bitmap);
6353 free_page((unsigned long)vmx_msr_bitmap_nested);
6355 free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
6357 free_page((unsigned long)vmx_msr_bitmap_longmode);
6359 free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
6361 free_page((unsigned long)vmx_msr_bitmap_legacy);
6363 free_page((unsigned long)vmx_io_bitmap_b);
6365 free_page((unsigned long)vmx_io_bitmap_a);
6370 static __exit void hardware_unsetup(void)
6372 free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
6373 free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
6374 free_page((unsigned long)vmx_msr_bitmap_legacy);
6375 free_page((unsigned long)vmx_msr_bitmap_longmode);
6376 free_page((unsigned long)vmx_io_bitmap_b);
6377 free_page((unsigned long)vmx_io_bitmap_a);
6378 free_page((unsigned long)vmx_vmwrite_bitmap);
6379 free_page((unsigned long)vmx_vmread_bitmap);
6381 free_page((unsigned long)vmx_msr_bitmap_nested);
6387 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
6388 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
6390 static int handle_pause(struct kvm_vcpu *vcpu)
6393 grow_ple_window(vcpu);
6395 skip_emulated_instruction(vcpu);
6396 kvm_vcpu_on_spin(vcpu);
6401 static int handle_nop(struct kvm_vcpu *vcpu)
6403 skip_emulated_instruction(vcpu);
6407 static int handle_mwait(struct kvm_vcpu *vcpu)
6409 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
6410 return handle_nop(vcpu);
6413 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
6418 static int handle_monitor(struct kvm_vcpu *vcpu)
6420 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
6421 return handle_nop(vcpu);
6425 * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
6426 * We could reuse a single VMCS for all the L2 guests, but we also want the
6427 * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
6428 * allows keeping them loaded on the processor, and in the future will allow
6429 * optimizations where prepare_vmcs02 doesn't need to set all the fields on
6430 * every entry if they never change.
6431 * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
6432 * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
6434 * The following functions allocate and free a vmcs02 in this pool.
6437 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
6438 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
6440 struct vmcs02_list *item;
6441 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6442 if (item->vmptr == vmx->nested.current_vmptr) {
6443 list_move(&item->list, &vmx->nested.vmcs02_pool);
6444 return &item->vmcs02;
6447 if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
6448 /* Recycle the least recently used VMCS. */
6449 item = list_entry(vmx->nested.vmcs02_pool.prev,
6450 struct vmcs02_list, list);
6451 item->vmptr = vmx->nested.current_vmptr;
6452 list_move(&item->list, &vmx->nested.vmcs02_pool);
6453 return &item->vmcs02;
6456 /* Create a new VMCS */
6457 item = kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
6460 item->vmcs02.vmcs = alloc_vmcs();
6461 if (!item->vmcs02.vmcs) {
6465 loaded_vmcs_init(&item->vmcs02);
6466 item->vmptr = vmx->nested.current_vmptr;
6467 list_add(&(item->list), &(vmx->nested.vmcs02_pool));
6468 vmx->nested.vmcs02_num++;
6469 return &item->vmcs02;
6472 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
6473 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
6475 struct vmcs02_list *item;
6476 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6477 if (item->vmptr == vmptr) {
6478 free_loaded_vmcs(&item->vmcs02);
6479 list_del(&item->list);
6481 vmx->nested.vmcs02_num--;
6487 * Free all VMCSs saved for this vcpu, except the one pointed by
6488 * vmx->loaded_vmcs. We must be running L1, so vmx->loaded_vmcs
6489 * must be &vmx->vmcs01.
6491 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
6493 struct vmcs02_list *item, *n;
6495 WARN_ON(vmx->loaded_vmcs != &vmx->vmcs01);
6496 list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
6498 * Something will leak if the above WARN triggers. Better than
6501 if (vmx->loaded_vmcs == &item->vmcs02)
6504 free_loaded_vmcs(&item->vmcs02);
6505 list_del(&item->list);
6507 vmx->nested.vmcs02_num--;
6512 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
6513 * set the success or error code of an emulated VMX instruction, as specified
6514 * by Vol 2B, VMX Instruction Reference, "Conventions".
6516 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
6518 vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
6519 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6520 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
6523 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
6525 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6526 & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
6527 X86_EFLAGS_SF | X86_EFLAGS_OF))
6531 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
6532 u32 vm_instruction_error)
6534 if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
6536 * failValid writes the error number to the current VMCS, which
6537 * can't be done there isn't a current VMCS.
6539 nested_vmx_failInvalid(vcpu);
6542 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6543 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6544 X86_EFLAGS_SF | X86_EFLAGS_OF))
6546 get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
6548 * We don't need to force a shadow sync because
6549 * VM_INSTRUCTION_ERROR is not shadowed
6553 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
6555 /* TODO: not to reset guest simply here. */
6556 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6557 pr_warn("kvm: nested vmx abort, indicator %d\n", indicator);
6560 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
6562 struct vcpu_vmx *vmx =
6563 container_of(timer, struct vcpu_vmx, nested.preemption_timer);
6565 vmx->nested.preemption_timer_expired = true;
6566 kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
6567 kvm_vcpu_kick(&vmx->vcpu);
6569 return HRTIMER_NORESTART;
6573 * Decode the memory-address operand of a vmx instruction, as recorded on an
6574 * exit caused by such an instruction (run by a guest hypervisor).
6575 * On success, returns 0. When the operand is invalid, returns 1 and throws
6578 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
6579 unsigned long exit_qualification,
6580 u32 vmx_instruction_info, bool wr, gva_t *ret)
6584 struct kvm_segment s;
6587 * According to Vol. 3B, "Information for VM Exits Due to Instruction
6588 * Execution", on an exit, vmx_instruction_info holds most of the
6589 * addressing components of the operand. Only the displacement part
6590 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
6591 * For how an actual address is calculated from all these components,
6592 * refer to Vol. 1, "Operand Addressing".
6594 int scaling = vmx_instruction_info & 3;
6595 int addr_size = (vmx_instruction_info >> 7) & 7;
6596 bool is_reg = vmx_instruction_info & (1u << 10);
6597 int seg_reg = (vmx_instruction_info >> 15) & 7;
6598 int index_reg = (vmx_instruction_info >> 18) & 0xf;
6599 bool index_is_valid = !(vmx_instruction_info & (1u << 22));
6600 int base_reg = (vmx_instruction_info >> 23) & 0xf;
6601 bool base_is_valid = !(vmx_instruction_info & (1u << 27));
6604 kvm_queue_exception(vcpu, UD_VECTOR);
6608 /* Addr = segment_base + offset */
6609 /* offset = base + [index * scale] + displacement */
6610 off = exit_qualification; /* holds the displacement */
6612 off += kvm_register_read(vcpu, base_reg);
6614 off += kvm_register_read(vcpu, index_reg)<<scaling;
6615 vmx_get_segment(vcpu, &s, seg_reg);
6616 *ret = s.base + off;
6618 if (addr_size == 1) /* 32 bit */
6621 /* Checks for #GP/#SS exceptions. */
6623 if (is_protmode(vcpu)) {
6624 /* Protected mode: apply checks for segment validity in the
6626 * - segment type check (#GP(0) may be thrown)
6627 * - usability check (#GP(0)/#SS(0))
6628 * - limit check (#GP(0)/#SS(0))
6631 /* #GP(0) if the destination operand is located in a
6632 * read-only data segment or any code segment.
6634 exn = ((s.type & 0xa) == 0 || (s.type & 8));
6636 /* #GP(0) if the source operand is located in an
6637 * execute-only code segment
6639 exn = ((s.type & 0xa) == 8);
6642 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6645 if (is_long_mode(vcpu)) {
6646 /* Long mode: #GP(0)/#SS(0) if the memory address is in a
6647 * non-canonical form. This is an only check for long mode.
6649 exn = is_noncanonical_address(*ret);
6650 } else if (is_protmode(vcpu)) {
6651 /* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
6653 exn = (s.unusable != 0);
6654 /* Protected mode: #GP(0)/#SS(0) if the memory
6655 * operand is outside the segment limit.
6657 exn = exn || (off + sizeof(u64) > s.limit);
6660 kvm_queue_exception_e(vcpu,
6661 seg_reg == VCPU_SREG_SS ?
6662 SS_VECTOR : GP_VECTOR,
6671 * This function performs the various checks including
6672 * - if it's 4KB aligned
6673 * - No bits beyond the physical address width are set
6674 * - Returns 0 on success or else 1
6675 * (Intel SDM Section 30.3)
6677 static int nested_vmx_check_vmptr(struct kvm_vcpu *vcpu, int exit_reason,
6682 struct x86_exception e;
6684 struct vcpu_vmx *vmx = to_vmx(vcpu);
6685 int maxphyaddr = cpuid_maxphyaddr(vcpu);
6687 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
6688 vmcs_read32(VMX_INSTRUCTION_INFO), false, &gva))
6691 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
6692 sizeof(vmptr), &e)) {
6693 kvm_inject_page_fault(vcpu, &e);
6697 switch (exit_reason) {
6698 case EXIT_REASON_VMON:
6701 * The first 4 bytes of VMXON region contain the supported
6702 * VMCS revision identifier
6704 * Note - IA32_VMX_BASIC[48] will never be 1
6705 * for the nested case;
6706 * which replaces physical address width with 32
6709 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6710 nested_vmx_failInvalid(vcpu);
6711 skip_emulated_instruction(vcpu);
6715 page = nested_get_page(vcpu, vmptr);
6717 *(u32 *)kmap(page) != VMCS12_REVISION) {
6718 nested_vmx_failInvalid(vcpu);
6720 skip_emulated_instruction(vcpu);
6724 vmx->nested.vmxon_ptr = vmptr;
6726 case EXIT_REASON_VMCLEAR:
6727 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6728 nested_vmx_failValid(vcpu,
6729 VMXERR_VMCLEAR_INVALID_ADDRESS);
6730 skip_emulated_instruction(vcpu);
6734 if (vmptr == vmx->nested.vmxon_ptr) {
6735 nested_vmx_failValid(vcpu,
6736 VMXERR_VMCLEAR_VMXON_POINTER);
6737 skip_emulated_instruction(vcpu);
6741 case EXIT_REASON_VMPTRLD:
6742 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6743 nested_vmx_failValid(vcpu,
6744 VMXERR_VMPTRLD_INVALID_ADDRESS);
6745 skip_emulated_instruction(vcpu);
6749 if (vmptr == vmx->nested.vmxon_ptr) {
6750 nested_vmx_failValid(vcpu,
6751 VMXERR_VMCLEAR_VMXON_POINTER);
6752 skip_emulated_instruction(vcpu);
6757 return 1; /* shouldn't happen */
6766 * Emulate the VMXON instruction.
6767 * Currently, we just remember that VMX is active, and do not save or even
6768 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
6769 * do not currently need to store anything in that guest-allocated memory
6770 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
6771 * argument is different from the VMXON pointer (which the spec says they do).
6773 static int handle_vmon(struct kvm_vcpu *vcpu)
6775 struct kvm_segment cs;
6776 struct vcpu_vmx *vmx = to_vmx(vcpu);
6777 struct vmcs *shadow_vmcs;
6778 const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
6779 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
6781 /* The Intel VMX Instruction Reference lists a bunch of bits that
6782 * are prerequisite to running VMXON, most notably cr4.VMXE must be
6783 * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
6784 * Otherwise, we should fail with #UD. We test these now:
6786 if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
6787 !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
6788 (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
6789 kvm_queue_exception(vcpu, UD_VECTOR);
6793 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
6794 if (is_long_mode(vcpu) && !cs.l) {
6795 kvm_queue_exception(vcpu, UD_VECTOR);
6799 if (vmx_get_cpl(vcpu)) {
6800 kvm_inject_gp(vcpu, 0);
6804 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMON, NULL))
6807 if (vmx->nested.vmxon) {
6808 nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
6809 skip_emulated_instruction(vcpu);
6813 if ((vmx->nested.msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
6814 != VMXON_NEEDED_FEATURES) {
6815 kvm_inject_gp(vcpu, 0);
6819 if (enable_shadow_vmcs) {
6820 shadow_vmcs = alloc_vmcs();
6823 /* mark vmcs as shadow */
6824 shadow_vmcs->revision_id |= (1u << 31);
6825 /* init shadow vmcs */
6826 vmcs_clear(shadow_vmcs);
6827 vmx->nested.current_shadow_vmcs = shadow_vmcs;
6830 INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
6831 vmx->nested.vmcs02_num = 0;
6833 hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
6835 vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
6837 vmx->nested.vmxon = true;
6839 skip_emulated_instruction(vcpu);
6840 nested_vmx_succeed(vcpu);
6845 * Intel's VMX Instruction Reference specifies a common set of prerequisites
6846 * for running VMX instructions (except VMXON, whose prerequisites are
6847 * slightly different). It also specifies what exception to inject otherwise.
6849 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
6851 struct kvm_segment cs;
6852 struct vcpu_vmx *vmx = to_vmx(vcpu);
6854 if (!vmx->nested.vmxon) {
6855 kvm_queue_exception(vcpu, UD_VECTOR);
6859 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
6860 if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
6861 (is_long_mode(vcpu) && !cs.l)) {
6862 kvm_queue_exception(vcpu, UD_VECTOR);
6866 if (vmx_get_cpl(vcpu)) {
6867 kvm_inject_gp(vcpu, 0);
6874 static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
6876 if (vmx->nested.current_vmptr == -1ull)
6879 /* current_vmptr and current_vmcs12 are always set/reset together */
6880 if (WARN_ON(vmx->nested.current_vmcs12 == NULL))
6883 if (enable_shadow_vmcs) {
6884 /* copy to memory all shadowed fields in case
6885 they were modified */
6886 copy_shadow_to_vmcs12(vmx);
6887 vmx->nested.sync_shadow_vmcs = false;
6888 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
6889 SECONDARY_EXEC_SHADOW_VMCS);
6890 vmcs_write64(VMCS_LINK_POINTER, -1ull);
6892 vmx->nested.posted_intr_nv = -1;
6893 kunmap(vmx->nested.current_vmcs12_page);
6894 nested_release_page(vmx->nested.current_vmcs12_page);
6895 vmx->nested.current_vmptr = -1ull;
6896 vmx->nested.current_vmcs12 = NULL;
6900 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
6901 * just stops using VMX.
6903 static void free_nested(struct vcpu_vmx *vmx)
6905 if (!vmx->nested.vmxon)
6908 vmx->nested.vmxon = false;
6909 free_vpid(vmx->nested.vpid02);
6910 nested_release_vmcs12(vmx);
6911 if (enable_shadow_vmcs)
6912 free_vmcs(vmx->nested.current_shadow_vmcs);
6913 /* Unpin physical memory we referred to in current vmcs02 */
6914 if (vmx->nested.apic_access_page) {
6915 nested_release_page(vmx->nested.apic_access_page);
6916 vmx->nested.apic_access_page = NULL;
6918 if (vmx->nested.virtual_apic_page) {
6919 nested_release_page(vmx->nested.virtual_apic_page);
6920 vmx->nested.virtual_apic_page = NULL;
6922 if (vmx->nested.pi_desc_page) {
6923 kunmap(vmx->nested.pi_desc_page);
6924 nested_release_page(vmx->nested.pi_desc_page);
6925 vmx->nested.pi_desc_page = NULL;
6926 vmx->nested.pi_desc = NULL;
6929 nested_free_all_saved_vmcss(vmx);
6932 /* Emulate the VMXOFF instruction */
6933 static int handle_vmoff(struct kvm_vcpu *vcpu)
6935 if (!nested_vmx_check_permission(vcpu))
6937 free_nested(to_vmx(vcpu));
6938 skip_emulated_instruction(vcpu);
6939 nested_vmx_succeed(vcpu);
6943 /* Emulate the VMCLEAR instruction */
6944 static int handle_vmclear(struct kvm_vcpu *vcpu)
6946 struct vcpu_vmx *vmx = to_vmx(vcpu);
6948 struct vmcs12 *vmcs12;
6951 if (!nested_vmx_check_permission(vcpu))
6954 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMCLEAR, &vmptr))
6957 if (vmptr == vmx->nested.current_vmptr)
6958 nested_release_vmcs12(vmx);
6960 page = nested_get_page(vcpu, vmptr);
6963 * For accurate processor emulation, VMCLEAR beyond available
6964 * physical memory should do nothing at all. However, it is
6965 * possible that a nested vmx bug, not a guest hypervisor bug,
6966 * resulted in this case, so let's shut down before doing any
6969 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6972 vmcs12 = kmap(page);
6973 vmcs12->launch_state = 0;
6975 nested_release_page(page);
6977 nested_free_vmcs02(vmx, vmptr);
6979 skip_emulated_instruction(vcpu);
6980 nested_vmx_succeed(vcpu);
6984 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
6986 /* Emulate the VMLAUNCH instruction */
6987 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
6989 return nested_vmx_run(vcpu, true);
6992 /* Emulate the VMRESUME instruction */
6993 static int handle_vmresume(struct kvm_vcpu *vcpu)
6996 return nested_vmx_run(vcpu, false);
6999 enum vmcs_field_type {
7000 VMCS_FIELD_TYPE_U16 = 0,
7001 VMCS_FIELD_TYPE_U64 = 1,
7002 VMCS_FIELD_TYPE_U32 = 2,
7003 VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
7006 static inline int vmcs_field_type(unsigned long field)
7008 if (0x1 & field) /* the *_HIGH fields are all 32 bit */
7009 return VMCS_FIELD_TYPE_U32;
7010 return (field >> 13) & 0x3 ;
7013 static inline int vmcs_field_readonly(unsigned long field)
7015 return (((field >> 10) & 0x3) == 1);
7019 * Read a vmcs12 field. Since these can have varying lengths and we return
7020 * one type, we chose the biggest type (u64) and zero-extend the return value
7021 * to that size. Note that the caller, handle_vmread, might need to use only
7022 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
7023 * 64-bit fields are to be returned).
7025 static inline int vmcs12_read_any(struct kvm_vcpu *vcpu,
7026 unsigned long field, u64 *ret)
7028 short offset = vmcs_field_to_offset(field);
7034 p = ((char *)(get_vmcs12(vcpu))) + offset;
7036 switch (vmcs_field_type(field)) {
7037 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7038 *ret = *((natural_width *)p);
7040 case VMCS_FIELD_TYPE_U16:
7043 case VMCS_FIELD_TYPE_U32:
7046 case VMCS_FIELD_TYPE_U64:
7056 static inline int vmcs12_write_any(struct kvm_vcpu *vcpu,
7057 unsigned long field, u64 field_value){
7058 short offset = vmcs_field_to_offset(field);
7059 char *p = ((char *) get_vmcs12(vcpu)) + offset;
7063 switch (vmcs_field_type(field)) {
7064 case VMCS_FIELD_TYPE_U16:
7065 *(u16 *)p = field_value;
7067 case VMCS_FIELD_TYPE_U32:
7068 *(u32 *)p = field_value;
7070 case VMCS_FIELD_TYPE_U64:
7071 *(u64 *)p = field_value;
7073 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7074 *(natural_width *)p = field_value;
7083 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
7086 unsigned long field;
7088 struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
7089 const unsigned long *fields = shadow_read_write_fields;
7090 const int num_fields = max_shadow_read_write_fields;
7094 vmcs_load(shadow_vmcs);
7096 for (i = 0; i < num_fields; i++) {
7098 switch (vmcs_field_type(field)) {
7099 case VMCS_FIELD_TYPE_U16:
7100 field_value = vmcs_read16(field);
7102 case VMCS_FIELD_TYPE_U32:
7103 field_value = vmcs_read32(field);
7105 case VMCS_FIELD_TYPE_U64:
7106 field_value = vmcs_read64(field);
7108 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7109 field_value = vmcs_readl(field);
7115 vmcs12_write_any(&vmx->vcpu, field, field_value);
7118 vmcs_clear(shadow_vmcs);
7119 vmcs_load(vmx->loaded_vmcs->vmcs);
7124 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
7126 const unsigned long *fields[] = {
7127 shadow_read_write_fields,
7128 shadow_read_only_fields
7130 const int max_fields[] = {
7131 max_shadow_read_write_fields,
7132 max_shadow_read_only_fields
7135 unsigned long field;
7136 u64 field_value = 0;
7137 struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
7139 vmcs_load(shadow_vmcs);
7141 for (q = 0; q < ARRAY_SIZE(fields); q++) {
7142 for (i = 0; i < max_fields[q]; i++) {
7143 field = fields[q][i];
7144 vmcs12_read_any(&vmx->vcpu, field, &field_value);
7146 switch (vmcs_field_type(field)) {
7147 case VMCS_FIELD_TYPE_U16:
7148 vmcs_write16(field, (u16)field_value);
7150 case VMCS_FIELD_TYPE_U32:
7151 vmcs_write32(field, (u32)field_value);
7153 case VMCS_FIELD_TYPE_U64:
7154 vmcs_write64(field, (u64)field_value);
7156 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7157 vmcs_writel(field, (long)field_value);
7166 vmcs_clear(shadow_vmcs);
7167 vmcs_load(vmx->loaded_vmcs->vmcs);
7171 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
7172 * used before) all generate the same failure when it is missing.
7174 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
7176 struct vcpu_vmx *vmx = to_vmx(vcpu);
7177 if (vmx->nested.current_vmptr == -1ull) {
7178 nested_vmx_failInvalid(vcpu);
7179 skip_emulated_instruction(vcpu);
7185 static int handle_vmread(struct kvm_vcpu *vcpu)
7187 unsigned long field;
7189 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7190 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7193 if (!nested_vmx_check_permission(vcpu) ||
7194 !nested_vmx_check_vmcs12(vcpu))
7197 /* Decode instruction info and find the field to read */
7198 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7199 /* Read the field, zero-extended to a u64 field_value */
7200 if (vmcs12_read_any(vcpu, field, &field_value) < 0) {
7201 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7202 skip_emulated_instruction(vcpu);
7206 * Now copy part of this value to register or memory, as requested.
7207 * Note that the number of bits actually copied is 32 or 64 depending
7208 * on the guest's mode (32 or 64 bit), not on the given field's length.
7210 if (vmx_instruction_info & (1u << 10)) {
7211 kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
7214 if (get_vmx_mem_address(vcpu, exit_qualification,
7215 vmx_instruction_info, true, &gva))
7217 /* _system ok, as nested_vmx_check_permission verified cpl=0 */
7218 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
7219 &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
7222 nested_vmx_succeed(vcpu);
7223 skip_emulated_instruction(vcpu);
7228 static int handle_vmwrite(struct kvm_vcpu *vcpu)
7230 unsigned long field;
7232 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7233 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7234 /* The value to write might be 32 or 64 bits, depending on L1's long
7235 * mode, and eventually we need to write that into a field of several
7236 * possible lengths. The code below first zero-extends the value to 64
7237 * bit (field_value), and then copies only the approriate number of
7238 * bits into the vmcs12 field.
7240 u64 field_value = 0;
7241 struct x86_exception e;
7243 if (!nested_vmx_check_permission(vcpu) ||
7244 !nested_vmx_check_vmcs12(vcpu))
7247 if (vmx_instruction_info & (1u << 10))
7248 field_value = kvm_register_readl(vcpu,
7249 (((vmx_instruction_info) >> 3) & 0xf));
7251 if (get_vmx_mem_address(vcpu, exit_qualification,
7252 vmx_instruction_info, false, &gva))
7254 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
7255 &field_value, (is_64_bit_mode(vcpu) ? 8 : 4), &e)) {
7256 kvm_inject_page_fault(vcpu, &e);
7262 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7263 if (vmcs_field_readonly(field)) {
7264 nested_vmx_failValid(vcpu,
7265 VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
7266 skip_emulated_instruction(vcpu);
7270 if (vmcs12_write_any(vcpu, field, field_value) < 0) {
7271 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7272 skip_emulated_instruction(vcpu);
7276 nested_vmx_succeed(vcpu);
7277 skip_emulated_instruction(vcpu);
7281 /* Emulate the VMPTRLD instruction */
7282 static int handle_vmptrld(struct kvm_vcpu *vcpu)
7284 struct vcpu_vmx *vmx = to_vmx(vcpu);
7287 if (!nested_vmx_check_permission(vcpu))
7290 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMPTRLD, &vmptr))
7293 if (vmx->nested.current_vmptr != vmptr) {
7294 struct vmcs12 *new_vmcs12;
7296 page = nested_get_page(vcpu, vmptr);
7298 nested_vmx_failInvalid(vcpu);
7299 skip_emulated_instruction(vcpu);
7302 new_vmcs12 = kmap(page);
7303 if (new_vmcs12->revision_id != VMCS12_REVISION) {
7305 nested_release_page_clean(page);
7306 nested_vmx_failValid(vcpu,
7307 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
7308 skip_emulated_instruction(vcpu);
7312 nested_release_vmcs12(vmx);
7313 vmx->nested.current_vmptr = vmptr;
7314 vmx->nested.current_vmcs12 = new_vmcs12;
7315 vmx->nested.current_vmcs12_page = page;
7316 if (enable_shadow_vmcs) {
7317 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
7318 SECONDARY_EXEC_SHADOW_VMCS);
7319 vmcs_write64(VMCS_LINK_POINTER,
7320 __pa(vmx->nested.current_shadow_vmcs));
7321 vmx->nested.sync_shadow_vmcs = true;
7325 nested_vmx_succeed(vcpu);
7326 skip_emulated_instruction(vcpu);
7330 /* Emulate the VMPTRST instruction */
7331 static int handle_vmptrst(struct kvm_vcpu *vcpu)
7333 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7334 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7336 struct x86_exception e;
7338 if (!nested_vmx_check_permission(vcpu))
7341 if (get_vmx_mem_address(vcpu, exit_qualification,
7342 vmx_instruction_info, true, &vmcs_gva))
7344 /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
7345 if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
7346 (void *)&to_vmx(vcpu)->nested.current_vmptr,
7348 kvm_inject_page_fault(vcpu, &e);
7351 nested_vmx_succeed(vcpu);
7352 skip_emulated_instruction(vcpu);
7356 /* Emulate the INVEPT instruction */
7357 static int handle_invept(struct kvm_vcpu *vcpu)
7359 struct vcpu_vmx *vmx = to_vmx(vcpu);
7360 u32 vmx_instruction_info, types;
7363 struct x86_exception e;
7368 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7369 SECONDARY_EXEC_ENABLE_EPT) ||
7370 !(vmx->nested.nested_vmx_ept_caps & VMX_EPT_INVEPT_BIT)) {
7371 kvm_queue_exception(vcpu, UD_VECTOR);
7375 if (!nested_vmx_check_permission(vcpu))
7378 if (!kvm_read_cr0_bits(vcpu, X86_CR0_PE)) {
7379 kvm_queue_exception(vcpu, UD_VECTOR);
7383 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7384 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7386 types = (vmx->nested.nested_vmx_ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
7388 if (!(types & (1UL << type))) {
7389 nested_vmx_failValid(vcpu,
7390 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7394 /* According to the Intel VMX instruction reference, the memory
7395 * operand is read even if it isn't needed (e.g., for type==global)
7397 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7398 vmx_instruction_info, false, &gva))
7400 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
7401 sizeof(operand), &e)) {
7402 kvm_inject_page_fault(vcpu, &e);
7407 case VMX_EPT_EXTENT_GLOBAL:
7408 kvm_mmu_sync_roots(vcpu);
7409 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
7410 nested_vmx_succeed(vcpu);
7413 /* Trap single context invalidation invept calls */
7418 skip_emulated_instruction(vcpu);
7422 static int handle_invvpid(struct kvm_vcpu *vcpu)
7424 struct vcpu_vmx *vmx = to_vmx(vcpu);
7425 u32 vmx_instruction_info;
7426 unsigned long type, types;
7428 struct x86_exception e;
7431 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7432 SECONDARY_EXEC_ENABLE_VPID) ||
7433 !(vmx->nested.nested_vmx_vpid_caps & VMX_VPID_INVVPID_BIT)) {
7434 kvm_queue_exception(vcpu, UD_VECTOR);
7438 if (!nested_vmx_check_permission(vcpu))
7441 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7442 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7444 types = (vmx->nested.nested_vmx_vpid_caps >> 8) & 0x7;
7446 if (!(types & (1UL << type))) {
7447 nested_vmx_failValid(vcpu,
7448 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7452 /* according to the intel vmx instruction reference, the memory
7453 * operand is read even if it isn't needed (e.g., for type==global)
7455 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7456 vmx_instruction_info, false, &gva))
7458 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vpid,
7460 kvm_inject_page_fault(vcpu, &e);
7465 case VMX_VPID_EXTENT_ALL_CONTEXT:
7466 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->nested.vpid02);
7467 nested_vmx_succeed(vcpu);
7470 /* Trap single context invalidation invvpid calls */
7475 skip_emulated_instruction(vcpu);
7479 static int handle_pml_full(struct kvm_vcpu *vcpu)
7481 unsigned long exit_qualification;
7483 trace_kvm_pml_full(vcpu->vcpu_id);
7485 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7488 * PML buffer FULL happened while executing iret from NMI,
7489 * "blocked by NMI" bit has to be set before next VM entry.
7491 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
7492 cpu_has_virtual_nmis() &&
7493 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
7494 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
7495 GUEST_INTR_STATE_NMI);
7498 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
7499 * here.., and there's no userspace involvement needed for PML.
7504 static int handle_pcommit(struct kvm_vcpu *vcpu)
7506 /* we never catch pcommit instruct for L1 guest. */
7512 * The exit handlers return 1 if the exit was handled fully and guest execution
7513 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
7514 * to be done to userspace and return 0.
7516 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
7517 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
7518 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
7519 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
7520 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
7521 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
7522 [EXIT_REASON_CR_ACCESS] = handle_cr,
7523 [EXIT_REASON_DR_ACCESS] = handle_dr,
7524 [EXIT_REASON_CPUID] = handle_cpuid,
7525 [EXIT_REASON_MSR_READ] = handle_rdmsr,
7526 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
7527 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
7528 [EXIT_REASON_HLT] = handle_halt,
7529 [EXIT_REASON_INVD] = handle_invd,
7530 [EXIT_REASON_INVLPG] = handle_invlpg,
7531 [EXIT_REASON_RDPMC] = handle_rdpmc,
7532 [EXIT_REASON_VMCALL] = handle_vmcall,
7533 [EXIT_REASON_VMCLEAR] = handle_vmclear,
7534 [EXIT_REASON_VMLAUNCH] = handle_vmlaunch,
7535 [EXIT_REASON_VMPTRLD] = handle_vmptrld,
7536 [EXIT_REASON_VMPTRST] = handle_vmptrst,
7537 [EXIT_REASON_VMREAD] = handle_vmread,
7538 [EXIT_REASON_VMRESUME] = handle_vmresume,
7539 [EXIT_REASON_VMWRITE] = handle_vmwrite,
7540 [EXIT_REASON_VMOFF] = handle_vmoff,
7541 [EXIT_REASON_VMON] = handle_vmon,
7542 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
7543 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
7544 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
7545 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
7546 [EXIT_REASON_WBINVD] = handle_wbinvd,
7547 [EXIT_REASON_XSETBV] = handle_xsetbv,
7548 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
7549 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
7550 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
7551 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
7552 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
7553 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait,
7554 [EXIT_REASON_MONITOR_TRAP_FLAG] = handle_monitor_trap,
7555 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor,
7556 [EXIT_REASON_INVEPT] = handle_invept,
7557 [EXIT_REASON_INVVPID] = handle_invvpid,
7558 [EXIT_REASON_XSAVES] = handle_xsaves,
7559 [EXIT_REASON_XRSTORS] = handle_xrstors,
7560 [EXIT_REASON_PML_FULL] = handle_pml_full,
7561 [EXIT_REASON_PCOMMIT] = handle_pcommit,
7564 static const int kvm_vmx_max_exit_handlers =
7565 ARRAY_SIZE(kvm_vmx_exit_handlers);
7567 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
7568 struct vmcs12 *vmcs12)
7570 unsigned long exit_qualification;
7571 gpa_t bitmap, last_bitmap;
7576 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7577 return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
7579 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7581 port = exit_qualification >> 16;
7582 size = (exit_qualification & 7) + 1;
7584 last_bitmap = (gpa_t)-1;
7589 bitmap = vmcs12->io_bitmap_a;
7590 else if (port < 0x10000)
7591 bitmap = vmcs12->io_bitmap_b;
7594 bitmap += (port & 0x7fff) / 8;
7596 if (last_bitmap != bitmap)
7597 if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
7599 if (b & (1 << (port & 7)))
7604 last_bitmap = bitmap;
7611 * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
7612 * rather than handle it ourselves in L0. I.e., check whether L1 expressed
7613 * disinterest in the current event (read or write a specific MSR) by using an
7614 * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
7616 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
7617 struct vmcs12 *vmcs12, u32 exit_reason)
7619 u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
7622 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
7626 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
7627 * for the four combinations of read/write and low/high MSR numbers.
7628 * First we need to figure out which of the four to use:
7630 bitmap = vmcs12->msr_bitmap;
7631 if (exit_reason == EXIT_REASON_MSR_WRITE)
7633 if (msr_index >= 0xc0000000) {
7634 msr_index -= 0xc0000000;
7638 /* Then read the msr_index'th bit from this bitmap: */
7639 if (msr_index < 1024*8) {
7641 if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
7643 return 1 & (b >> (msr_index & 7));
7645 return true; /* let L1 handle the wrong parameter */
7649 * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
7650 * rather than handle it ourselves in L0. I.e., check if L1 wanted to
7651 * intercept (via guest_host_mask etc.) the current event.
7653 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
7654 struct vmcs12 *vmcs12)
7656 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7657 int cr = exit_qualification & 15;
7658 int reg = (exit_qualification >> 8) & 15;
7659 unsigned long val = kvm_register_readl(vcpu, reg);
7661 switch ((exit_qualification >> 4) & 3) {
7662 case 0: /* mov to cr */
7665 if (vmcs12->cr0_guest_host_mask &
7666 (val ^ vmcs12->cr0_read_shadow))
7670 if ((vmcs12->cr3_target_count >= 1 &&
7671 vmcs12->cr3_target_value0 == val) ||
7672 (vmcs12->cr3_target_count >= 2 &&
7673 vmcs12->cr3_target_value1 == val) ||
7674 (vmcs12->cr3_target_count >= 3 &&
7675 vmcs12->cr3_target_value2 == val) ||
7676 (vmcs12->cr3_target_count >= 4 &&
7677 vmcs12->cr3_target_value3 == val))
7679 if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
7683 if (vmcs12->cr4_guest_host_mask &
7684 (vmcs12->cr4_read_shadow ^ val))
7688 if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
7694 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
7695 (vmcs12->cr0_read_shadow & X86_CR0_TS))
7698 case 1: /* mov from cr */
7701 if (vmcs12->cpu_based_vm_exec_control &
7702 CPU_BASED_CR3_STORE_EXITING)
7706 if (vmcs12->cpu_based_vm_exec_control &
7707 CPU_BASED_CR8_STORE_EXITING)
7714 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
7715 * cr0. Other attempted changes are ignored, with no exit.
7717 if (vmcs12->cr0_guest_host_mask & 0xe &
7718 (val ^ vmcs12->cr0_read_shadow))
7720 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
7721 !(vmcs12->cr0_read_shadow & 0x1) &&
7730 * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
7731 * should handle it ourselves in L0 (and then continue L2). Only call this
7732 * when in is_guest_mode (L2).
7734 static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu)
7736 u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
7737 struct vcpu_vmx *vmx = to_vmx(vcpu);
7738 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7739 u32 exit_reason = vmx->exit_reason;
7741 trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
7742 vmcs_readl(EXIT_QUALIFICATION),
7743 vmx->idt_vectoring_info,
7745 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
7748 if (vmx->nested.nested_run_pending)
7751 if (unlikely(vmx->fail)) {
7752 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
7753 vmcs_read32(VM_INSTRUCTION_ERROR));
7757 switch (exit_reason) {
7758 case EXIT_REASON_EXCEPTION_NMI:
7759 if (!is_exception(intr_info))
7761 else if (is_page_fault(intr_info))
7763 else if (is_no_device(intr_info) &&
7764 !(vmcs12->guest_cr0 & X86_CR0_TS))
7766 return vmcs12->exception_bitmap &
7767 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
7768 case EXIT_REASON_EXTERNAL_INTERRUPT:
7770 case EXIT_REASON_TRIPLE_FAULT:
7772 case EXIT_REASON_PENDING_INTERRUPT:
7773 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
7774 case EXIT_REASON_NMI_WINDOW:
7775 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
7776 case EXIT_REASON_TASK_SWITCH:
7778 case EXIT_REASON_CPUID:
7779 if (kvm_register_read(vcpu, VCPU_REGS_RAX) == 0xa)
7782 case EXIT_REASON_HLT:
7783 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
7784 case EXIT_REASON_INVD:
7786 case EXIT_REASON_INVLPG:
7787 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
7788 case EXIT_REASON_RDPMC:
7789 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
7790 case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
7791 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
7792 case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
7793 case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
7794 case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
7795 case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
7796 case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
7797 case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
7799 * VMX instructions trap unconditionally. This allows L1 to
7800 * emulate them for its L2 guest, i.e., allows 3-level nesting!
7803 case EXIT_REASON_CR_ACCESS:
7804 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
7805 case EXIT_REASON_DR_ACCESS:
7806 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
7807 case EXIT_REASON_IO_INSTRUCTION:
7808 return nested_vmx_exit_handled_io(vcpu, vmcs12);
7809 case EXIT_REASON_MSR_READ:
7810 case EXIT_REASON_MSR_WRITE:
7811 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
7812 case EXIT_REASON_INVALID_STATE:
7814 case EXIT_REASON_MWAIT_INSTRUCTION:
7815 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
7816 case EXIT_REASON_MONITOR_TRAP_FLAG:
7817 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_TRAP_FLAG);
7818 case EXIT_REASON_MONITOR_INSTRUCTION:
7819 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
7820 case EXIT_REASON_PAUSE_INSTRUCTION:
7821 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
7822 nested_cpu_has2(vmcs12,
7823 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
7824 case EXIT_REASON_MCE_DURING_VMENTRY:
7826 case EXIT_REASON_TPR_BELOW_THRESHOLD:
7827 return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
7828 case EXIT_REASON_APIC_ACCESS:
7829 return nested_cpu_has2(vmcs12,
7830 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
7831 case EXIT_REASON_APIC_WRITE:
7832 case EXIT_REASON_EOI_INDUCED:
7833 /* apic_write and eoi_induced should exit unconditionally. */
7835 case EXIT_REASON_EPT_VIOLATION:
7837 * L0 always deals with the EPT violation. If nested EPT is
7838 * used, and the nested mmu code discovers that the address is
7839 * missing in the guest EPT table (EPT12), the EPT violation
7840 * will be injected with nested_ept_inject_page_fault()
7843 case EXIT_REASON_EPT_MISCONFIG:
7845 * L2 never uses directly L1's EPT, but rather L0's own EPT
7846 * table (shadow on EPT) or a merged EPT table that L0 built
7847 * (EPT on EPT). So any problems with the structure of the
7848 * table is L0's fault.
7851 case EXIT_REASON_WBINVD:
7852 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
7853 case EXIT_REASON_XSETBV:
7855 case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
7857 * This should never happen, since it is not possible to
7858 * set XSS to a non-zero value---neither in L1 nor in L2.
7859 * If if it were, XSS would have to be checked against
7860 * the XSS exit bitmap in vmcs12.
7862 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
7863 case EXIT_REASON_PCOMMIT:
7864 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_PCOMMIT);
7870 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
7872 *info1 = vmcs_readl(EXIT_QUALIFICATION);
7873 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
7876 static int vmx_create_pml_buffer(struct vcpu_vmx *vmx)
7878 struct page *pml_pg;
7880 pml_pg = alloc_page(GFP_KERNEL | __GFP_ZERO);
7884 vmx->pml_pg = pml_pg;
7886 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
7887 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
7892 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
7895 __free_page(vmx->pml_pg);
7900 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
7902 struct vcpu_vmx *vmx = to_vmx(vcpu);
7906 pml_idx = vmcs_read16(GUEST_PML_INDEX);
7908 /* Do nothing if PML buffer is empty */
7909 if (pml_idx == (PML_ENTITY_NUM - 1))
7912 /* PML index always points to next available PML buffer entity */
7913 if (pml_idx >= PML_ENTITY_NUM)
7918 pml_buf = page_address(vmx->pml_pg);
7919 for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
7922 gpa = pml_buf[pml_idx];
7923 WARN_ON(gpa & (PAGE_SIZE - 1));
7924 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
7927 /* reset PML index */
7928 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
7932 * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
7933 * Called before reporting dirty_bitmap to userspace.
7935 static void kvm_flush_pml_buffers(struct kvm *kvm)
7938 struct kvm_vcpu *vcpu;
7940 * We only need to kick vcpu out of guest mode here, as PML buffer
7941 * is flushed at beginning of all VMEXITs, and it's obvious that only
7942 * vcpus running in guest are possible to have unflushed GPAs in PML
7945 kvm_for_each_vcpu(i, vcpu, kvm)
7946 kvm_vcpu_kick(vcpu);
7949 static void vmx_dump_sel(char *name, uint32_t sel)
7951 pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
7952 name, vmcs_read32(sel),
7953 vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
7954 vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
7955 vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
7958 static void vmx_dump_dtsel(char *name, uint32_t limit)
7960 pr_err("%s limit=0x%08x, base=0x%016lx\n",
7961 name, vmcs_read32(limit),
7962 vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
7965 static void dump_vmcs(void)
7967 u32 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
7968 u32 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
7969 u32 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
7970 u32 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
7971 u32 secondary_exec_control = 0;
7972 unsigned long cr4 = vmcs_readl(GUEST_CR4);
7973 u64 efer = vmcs_read64(GUEST_IA32_EFER);
7976 if (cpu_has_secondary_exec_ctrls())
7977 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7979 pr_err("*** Guest State ***\n");
7980 pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
7981 vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
7982 vmcs_readl(CR0_GUEST_HOST_MASK));
7983 pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
7984 cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
7985 pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
7986 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
7987 (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
7989 pr_err("PDPTR0 = 0x%016llx PDPTR1 = 0x%016llx\n",
7990 vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
7991 pr_err("PDPTR2 = 0x%016llx PDPTR3 = 0x%016llx\n",
7992 vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
7994 pr_err("RSP = 0x%016lx RIP = 0x%016lx\n",
7995 vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
7996 pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n",
7997 vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
7998 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
7999 vmcs_readl(GUEST_SYSENTER_ESP),
8000 vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
8001 vmx_dump_sel("CS: ", GUEST_CS_SELECTOR);
8002 vmx_dump_sel("DS: ", GUEST_DS_SELECTOR);
8003 vmx_dump_sel("SS: ", GUEST_SS_SELECTOR);
8004 vmx_dump_sel("ES: ", GUEST_ES_SELECTOR);
8005 vmx_dump_sel("FS: ", GUEST_FS_SELECTOR);
8006 vmx_dump_sel("GS: ", GUEST_GS_SELECTOR);
8007 vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
8008 vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
8009 vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
8010 vmx_dump_sel("TR: ", GUEST_TR_SELECTOR);
8011 if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
8012 (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
8013 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
8014 efer, vmcs_read64(GUEST_IA32_PAT));
8015 pr_err("DebugCtl = 0x%016llx DebugExceptions = 0x%016lx\n",
8016 vmcs_read64(GUEST_IA32_DEBUGCTL),
8017 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
8018 if (vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
8019 pr_err("PerfGlobCtl = 0x%016llx\n",
8020 vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
8021 if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
8022 pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
8023 pr_err("Interruptibility = %08x ActivityState = %08x\n",
8024 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
8025 vmcs_read32(GUEST_ACTIVITY_STATE));
8026 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
8027 pr_err("InterruptStatus = %04x\n",
8028 vmcs_read16(GUEST_INTR_STATUS));
8030 pr_err("*** Host State ***\n");
8031 pr_err("RIP = 0x%016lx RSP = 0x%016lx\n",
8032 vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
8033 pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
8034 vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
8035 vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
8036 vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
8037 vmcs_read16(HOST_TR_SELECTOR));
8038 pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
8039 vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
8040 vmcs_readl(HOST_TR_BASE));
8041 pr_err("GDTBase=%016lx IDTBase=%016lx\n",
8042 vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
8043 pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
8044 vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
8045 vmcs_readl(HOST_CR4));
8046 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8047 vmcs_readl(HOST_IA32_SYSENTER_ESP),
8048 vmcs_read32(HOST_IA32_SYSENTER_CS),
8049 vmcs_readl(HOST_IA32_SYSENTER_EIP));
8050 if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
8051 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
8052 vmcs_read64(HOST_IA32_EFER),
8053 vmcs_read64(HOST_IA32_PAT));
8054 if (vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
8055 pr_err("PerfGlobCtl = 0x%016llx\n",
8056 vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
8058 pr_err("*** Control State ***\n");
8059 pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
8060 pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
8061 pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
8062 pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
8063 vmcs_read32(EXCEPTION_BITMAP),
8064 vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
8065 vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
8066 pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
8067 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8068 vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
8069 vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
8070 pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
8071 vmcs_read32(VM_EXIT_INTR_INFO),
8072 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8073 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
8074 pr_err(" reason=%08x qualification=%016lx\n",
8075 vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
8076 pr_err("IDTVectoring: info=%08x errcode=%08x\n",
8077 vmcs_read32(IDT_VECTORING_INFO_FIELD),
8078 vmcs_read32(IDT_VECTORING_ERROR_CODE));
8079 pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
8080 if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
8081 pr_err("TSC Multiplier = 0x%016llx\n",
8082 vmcs_read64(TSC_MULTIPLIER));
8083 if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW)
8084 pr_err("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
8085 if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
8086 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
8087 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
8088 pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
8089 n = vmcs_read32(CR3_TARGET_COUNT);
8090 for (i = 0; i + 1 < n; i += 4)
8091 pr_err("CR3 target%u=%016lx target%u=%016lx\n",
8092 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
8093 i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
8095 pr_err("CR3 target%u=%016lx\n",
8096 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
8097 if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
8098 pr_err("PLE Gap=%08x Window=%08x\n",
8099 vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
8100 if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
8101 pr_err("Virtual processor ID = 0x%04x\n",
8102 vmcs_read16(VIRTUAL_PROCESSOR_ID));
8106 * The guest has exited. See if we can fix it or if we need userspace
8109 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
8111 struct vcpu_vmx *vmx = to_vmx(vcpu);
8112 u32 exit_reason = vmx->exit_reason;
8113 u32 vectoring_info = vmx->idt_vectoring_info;
8115 trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
8118 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
8119 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
8120 * querying dirty_bitmap, we only need to kick all vcpus out of guest
8121 * mode as if vcpus is in root mode, the PML buffer must has been
8125 vmx_flush_pml_buffer(vcpu);
8127 /* If guest state is invalid, start emulating */
8128 if (vmx->emulation_required)
8129 return handle_invalid_guest_state(vcpu);
8131 if (is_guest_mode(vcpu) && nested_vmx_exit_handled(vcpu)) {
8132 nested_vmx_vmexit(vcpu, exit_reason,
8133 vmcs_read32(VM_EXIT_INTR_INFO),
8134 vmcs_readl(EXIT_QUALIFICATION));
8138 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
8140 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8141 vcpu->run->fail_entry.hardware_entry_failure_reason
8146 if (unlikely(vmx->fail)) {
8147 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8148 vcpu->run->fail_entry.hardware_entry_failure_reason
8149 = vmcs_read32(VM_INSTRUCTION_ERROR);
8155 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
8156 * delivery event since it indicates guest is accessing MMIO.
8157 * The vm-exit can be triggered again after return to guest that
8158 * will cause infinite loop.
8160 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
8161 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
8162 exit_reason != EXIT_REASON_EPT_VIOLATION &&
8163 exit_reason != EXIT_REASON_TASK_SWITCH)) {
8164 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8165 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
8166 vcpu->run->internal.ndata = 2;
8167 vcpu->run->internal.data[0] = vectoring_info;
8168 vcpu->run->internal.data[1] = exit_reason;
8172 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked &&
8173 !(is_guest_mode(vcpu) && nested_cpu_has_virtual_nmis(
8174 get_vmcs12(vcpu))))) {
8175 if (vmx_interrupt_allowed(vcpu)) {
8176 vmx->soft_vnmi_blocked = 0;
8177 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
8178 vcpu->arch.nmi_pending) {
8180 * This CPU don't support us in finding the end of an
8181 * NMI-blocked window if the guest runs with IRQs
8182 * disabled. So we pull the trigger after 1 s of
8183 * futile waiting, but inform the user about this.
8185 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
8186 "state on VCPU %d after 1 s timeout\n",
8187 __func__, vcpu->vcpu_id);
8188 vmx->soft_vnmi_blocked = 0;
8192 if (exit_reason < kvm_vmx_max_exit_handlers
8193 && kvm_vmx_exit_handlers[exit_reason])
8194 return kvm_vmx_exit_handlers[exit_reason](vcpu);
8196 WARN_ONCE(1, "vmx: unexpected exit reason 0x%x\n", exit_reason);
8197 kvm_queue_exception(vcpu, UD_VECTOR);
8202 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
8204 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8206 if (is_guest_mode(vcpu) &&
8207 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
8210 if (irr == -1 || tpr < irr) {
8211 vmcs_write32(TPR_THRESHOLD, 0);
8215 vmcs_write32(TPR_THRESHOLD, irr);
8218 static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
8220 u32 sec_exec_control;
8223 * There is not point to enable virtualize x2apic without enable
8226 if (!cpu_has_vmx_virtualize_x2apic_mode() ||
8227 !kvm_vcpu_apicv_active(vcpu))
8230 if (!cpu_need_tpr_shadow(vcpu))
8233 sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8236 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8237 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8239 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8240 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8242 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
8244 vmx_set_msr_bitmap(vcpu);
8247 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
8249 struct vcpu_vmx *vmx = to_vmx(vcpu);
8252 * Currently we do not handle the nested case where L2 has an
8253 * APIC access page of its own; that page is still pinned.
8254 * Hence, we skip the case where the VCPU is in guest mode _and_
8255 * L1 prepared an APIC access page for L2.
8257 * For the case where L1 and L2 share the same APIC access page
8258 * (flexpriority=Y but SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES clear
8259 * in the vmcs12), this function will only update either the vmcs01
8260 * or the vmcs02. If the former, the vmcs02 will be updated by
8261 * prepare_vmcs02. If the latter, the vmcs01 will be updated in
8262 * the next L2->L1 exit.
8264 if (!is_guest_mode(vcpu) ||
8265 !nested_cpu_has2(vmx->nested.current_vmcs12,
8266 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
8267 vmcs_write64(APIC_ACCESS_ADDR, hpa);
8270 static void vmx_hwapic_isr_update(struct kvm *kvm, int isr)
8278 status = vmcs_read16(GUEST_INTR_STATUS);
8283 vmcs_write16(GUEST_INTR_STATUS, status);
8287 static void vmx_set_rvi(int vector)
8295 status = vmcs_read16(GUEST_INTR_STATUS);
8296 old = (u8)status & 0xff;
8297 if ((u8)vector != old) {
8299 status |= (u8)vector;
8300 vmcs_write16(GUEST_INTR_STATUS, status);
8304 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
8306 if (!is_guest_mode(vcpu)) {
8307 vmx_set_rvi(max_irr);
8315 * In guest mode. If a vmexit is needed, vmx_check_nested_events
8318 if (nested_exit_on_intr(vcpu))
8322 * Else, fall back to pre-APICv interrupt injection since L2
8323 * is run without virtual interrupt delivery.
8325 if (!kvm_event_needs_reinjection(vcpu) &&
8326 vmx_interrupt_allowed(vcpu)) {
8327 kvm_queue_interrupt(vcpu, max_irr, false);
8328 vmx_inject_irq(vcpu);
8332 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
8334 if (!kvm_vcpu_apicv_active(vcpu))
8337 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
8338 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
8339 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
8340 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
8343 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
8347 if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
8348 || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
8351 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8352 exit_intr_info = vmx->exit_intr_info;
8354 /* Handle machine checks before interrupts are enabled */
8355 if (is_machine_check(exit_intr_info))
8356 kvm_machine_check();
8358 /* We need to handle NMIs before interrupts are enabled */
8359 if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
8360 (exit_intr_info & INTR_INFO_VALID_MASK)) {
8361 kvm_before_handle_nmi(&vmx->vcpu);
8363 kvm_after_handle_nmi(&vmx->vcpu);
8367 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
8369 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8372 * If external interrupt exists, IF bit is set in rflags/eflags on the
8373 * interrupt stack frame, and interrupt will be enabled on a return
8374 * from interrupt handler.
8376 if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
8377 == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
8378 unsigned int vector;
8379 unsigned long entry;
8381 struct vcpu_vmx *vmx = to_vmx(vcpu);
8382 #ifdef CONFIG_X86_64
8386 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
8387 desc = (gate_desc *)vmx->host_idt_base + vector;
8388 entry = gate_offset(*desc);
8390 #ifdef CONFIG_X86_64
8391 "mov %%" _ASM_SP ", %[sp]\n\t"
8392 "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
8397 "orl $0x200, (%%" _ASM_SP ")\n\t"
8398 __ASM_SIZE(push) " $%c[cs]\n\t"
8399 "call *%[entry]\n\t"
8401 #ifdef CONFIG_X86_64
8406 [ss]"i"(__KERNEL_DS),
8407 [cs]"i"(__KERNEL_CS)
8413 static bool vmx_has_high_real_mode_segbase(void)
8415 return enable_unrestricted_guest || emulate_invalid_guest_state;
8418 static bool vmx_mpx_supported(void)
8420 return (vmcs_config.vmexit_ctrl & VM_EXIT_CLEAR_BNDCFGS) &&
8421 (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_BNDCFGS);
8424 static bool vmx_xsaves_supported(void)
8426 return vmcs_config.cpu_based_2nd_exec_ctrl &
8427 SECONDARY_EXEC_XSAVES;
8430 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
8435 bool idtv_info_valid;
8437 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8439 if (cpu_has_virtual_nmis()) {
8440 if (vmx->nmi_known_unmasked)
8443 * Can't use vmx->exit_intr_info since we're not sure what
8444 * the exit reason is.
8446 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8447 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
8448 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
8450 * SDM 3: 27.7.1.2 (September 2008)
8451 * Re-set bit "block by NMI" before VM entry if vmexit caused by
8452 * a guest IRET fault.
8453 * SDM 3: 23.2.2 (September 2008)
8454 * Bit 12 is undefined in any of the following cases:
8455 * If the VM exit sets the valid bit in the IDT-vectoring
8456 * information field.
8457 * If the VM exit is due to a double fault.
8459 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
8460 vector != DF_VECTOR && !idtv_info_valid)
8461 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
8462 GUEST_INTR_STATE_NMI);
8464 vmx->nmi_known_unmasked =
8465 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
8466 & GUEST_INTR_STATE_NMI);
8467 } else if (unlikely(vmx->soft_vnmi_blocked))
8468 vmx->vnmi_blocked_time +=
8469 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
8472 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
8473 u32 idt_vectoring_info,
8474 int instr_len_field,
8475 int error_code_field)
8479 bool idtv_info_valid;
8481 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8483 vcpu->arch.nmi_injected = false;
8484 kvm_clear_exception_queue(vcpu);
8485 kvm_clear_interrupt_queue(vcpu);
8487 if (!idtv_info_valid)
8490 kvm_make_request(KVM_REQ_EVENT, vcpu);
8492 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
8493 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
8496 case INTR_TYPE_NMI_INTR:
8497 vcpu->arch.nmi_injected = true;
8499 * SDM 3: 27.7.1.2 (September 2008)
8500 * Clear bit "block by NMI" before VM entry if a NMI
8503 vmx_set_nmi_mask(vcpu, false);
8505 case INTR_TYPE_SOFT_EXCEPTION:
8506 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8508 case INTR_TYPE_HARD_EXCEPTION:
8509 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
8510 u32 err = vmcs_read32(error_code_field);
8511 kvm_requeue_exception_e(vcpu, vector, err);
8513 kvm_requeue_exception(vcpu, vector);
8515 case INTR_TYPE_SOFT_INTR:
8516 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8518 case INTR_TYPE_EXT_INTR:
8519 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
8526 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
8528 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
8529 VM_EXIT_INSTRUCTION_LEN,
8530 IDT_VECTORING_ERROR_CODE);
8533 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
8535 __vmx_complete_interrupts(vcpu,
8536 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8537 VM_ENTRY_INSTRUCTION_LEN,
8538 VM_ENTRY_EXCEPTION_ERROR_CODE);
8540 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
8543 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
8546 struct perf_guest_switch_msr *msrs;
8548 msrs = perf_guest_get_msrs(&nr_msrs);
8553 for (i = 0; i < nr_msrs; i++)
8554 if (msrs[i].host == msrs[i].guest)
8555 clear_atomic_switch_msr(vmx, msrs[i].msr);
8557 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
8561 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
8563 struct vcpu_vmx *vmx = to_vmx(vcpu);
8564 unsigned long debugctlmsr, cr4;
8566 /* Record the guest's net vcpu time for enforced NMI injections. */
8567 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
8568 vmx->entry_time = ktime_get();
8570 /* Don't enter VMX if guest state is invalid, let the exit handler
8571 start emulation until we arrive back to a valid state */
8572 if (vmx->emulation_required)
8575 if (vmx->ple_window_dirty) {
8576 vmx->ple_window_dirty = false;
8577 vmcs_write32(PLE_WINDOW, vmx->ple_window);
8580 if (vmx->nested.sync_shadow_vmcs) {
8581 copy_vmcs12_to_shadow(vmx);
8582 vmx->nested.sync_shadow_vmcs = false;
8585 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
8586 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
8587 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
8588 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
8590 cr4 = cr4_read_shadow();
8591 if (unlikely(cr4 != vmx->host_state.vmcs_host_cr4)) {
8592 vmcs_writel(HOST_CR4, cr4);
8593 vmx->host_state.vmcs_host_cr4 = cr4;
8596 /* When single-stepping over STI and MOV SS, we must clear the
8597 * corresponding interruptibility bits in the guest state. Otherwise
8598 * vmentry fails as it then expects bit 14 (BS) in pending debug
8599 * exceptions being set, but that's not correct for the guest debugging
8601 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8602 vmx_set_interrupt_shadow(vcpu, 0);
8604 atomic_switch_perf_msrs(vmx);
8605 debugctlmsr = get_debugctlmsr();
8607 vmx->__launched = vmx->loaded_vmcs->launched;
8609 /* Store host registers */
8610 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
8611 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
8612 "push %%" _ASM_CX " \n\t"
8613 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
8615 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
8616 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
8618 /* Reload cr2 if changed */
8619 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
8620 "mov %%cr2, %%" _ASM_DX " \n\t"
8621 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
8623 "mov %%" _ASM_AX", %%cr2 \n\t"
8625 /* Check if vmlaunch of vmresume is needed */
8626 "cmpl $0, %c[launched](%0) \n\t"
8627 /* Load guest registers. Don't clobber flags. */
8628 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
8629 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
8630 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
8631 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
8632 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
8633 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
8634 #ifdef CONFIG_X86_64
8635 "mov %c[r8](%0), %%r8 \n\t"
8636 "mov %c[r9](%0), %%r9 \n\t"
8637 "mov %c[r10](%0), %%r10 \n\t"
8638 "mov %c[r11](%0), %%r11 \n\t"
8639 "mov %c[r12](%0), %%r12 \n\t"
8640 "mov %c[r13](%0), %%r13 \n\t"
8641 "mov %c[r14](%0), %%r14 \n\t"
8642 "mov %c[r15](%0), %%r15 \n\t"
8644 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
8646 /* Enter guest mode */
8648 __ex(ASM_VMX_VMLAUNCH) "\n\t"
8650 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
8652 /* Save guest registers, load host registers, keep flags */
8653 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
8655 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
8656 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
8657 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
8658 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
8659 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
8660 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
8661 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
8662 #ifdef CONFIG_X86_64
8663 "mov %%r8, %c[r8](%0) \n\t"
8664 "mov %%r9, %c[r9](%0) \n\t"
8665 "mov %%r10, %c[r10](%0) \n\t"
8666 "mov %%r11, %c[r11](%0) \n\t"
8667 "mov %%r12, %c[r12](%0) \n\t"
8668 "mov %%r13, %c[r13](%0) \n\t"
8669 "mov %%r14, %c[r14](%0) \n\t"
8670 "mov %%r15, %c[r15](%0) \n\t"
8672 "mov %%cr2, %%" _ASM_AX " \n\t"
8673 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
8675 "pop %%" _ASM_BP "; pop %%" _ASM_DX " \n\t"
8676 "setbe %c[fail](%0) \n\t"
8677 ".pushsection .rodata \n\t"
8678 ".global vmx_return \n\t"
8679 "vmx_return: " _ASM_PTR " 2b \n\t"
8681 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
8682 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
8683 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
8684 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
8685 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
8686 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
8687 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
8688 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
8689 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
8690 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
8691 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
8692 #ifdef CONFIG_X86_64
8693 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
8694 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
8695 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
8696 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
8697 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
8698 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
8699 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
8700 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
8702 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
8703 [wordsize]"i"(sizeof(ulong))
8705 #ifdef CONFIG_X86_64
8706 , "rax", "rbx", "rdi", "rsi"
8707 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
8709 , "eax", "ebx", "edi", "esi"
8713 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
8715 update_debugctlmsr(debugctlmsr);
8717 #ifndef CONFIG_X86_64
8719 * The sysexit path does not restore ds/es, so we must set them to
8720 * a reasonable value ourselves.
8722 * We can't defer this to vmx_load_host_state() since that function
8723 * may be executed in interrupt context, which saves and restore segments
8724 * around it, nullifying its effect.
8726 loadsegment(ds, __USER_DS);
8727 loadsegment(es, __USER_DS);
8730 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
8731 | (1 << VCPU_EXREG_RFLAGS)
8732 | (1 << VCPU_EXREG_PDPTR)
8733 | (1 << VCPU_EXREG_SEGMENTS)
8734 | (1 << VCPU_EXREG_CR3));
8735 vcpu->arch.regs_dirty = 0;
8737 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
8739 vmx->loaded_vmcs->launched = 1;
8741 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
8744 * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
8745 * we did not inject a still-pending event to L1 now because of
8746 * nested_run_pending, we need to re-enable this bit.
8748 if (vmx->nested.nested_run_pending)
8749 kvm_make_request(KVM_REQ_EVENT, vcpu);
8751 vmx->nested.nested_run_pending = 0;
8753 vmx_complete_atomic_exit(vmx);
8754 vmx_recover_nmi_blocking(vmx);
8755 vmx_complete_interrupts(vmx);
8758 static void vmx_load_vmcs01(struct kvm_vcpu *vcpu)
8760 struct vcpu_vmx *vmx = to_vmx(vcpu);
8763 if (vmx->loaded_vmcs == &vmx->vmcs01)
8767 vmx->loaded_vmcs = &vmx->vmcs01;
8769 vmx_vcpu_load(vcpu, cpu);
8774 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
8776 struct vcpu_vmx *vmx = to_vmx(vcpu);
8779 vmx_destroy_pml_buffer(vmx);
8780 free_vpid(vmx->vpid);
8781 leave_guest_mode(vcpu);
8782 vmx_load_vmcs01(vcpu);
8784 free_loaded_vmcs(vmx->loaded_vmcs);
8785 kfree(vmx->guest_msrs);
8786 kvm_vcpu_uninit(vcpu);
8787 kmem_cache_free(kvm_vcpu_cache, vmx);
8790 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
8793 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
8797 return ERR_PTR(-ENOMEM);
8799 vmx->vpid = allocate_vpid();
8801 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
8805 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
8806 BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
8810 if (!vmx->guest_msrs) {
8814 vmx->loaded_vmcs = &vmx->vmcs01;
8815 vmx->loaded_vmcs->vmcs = alloc_vmcs();
8816 if (!vmx->loaded_vmcs->vmcs)
8819 kvm_cpu_vmxon(__pa(per_cpu(vmxarea, raw_smp_processor_id())));
8820 loaded_vmcs_init(vmx->loaded_vmcs);
8825 vmx_vcpu_load(&vmx->vcpu, cpu);
8826 vmx->vcpu.cpu = cpu;
8827 err = vmx_vcpu_setup(vmx);
8828 vmx_vcpu_put(&vmx->vcpu);
8832 if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
8833 err = alloc_apic_access_page(kvm);
8839 if (!kvm->arch.ept_identity_map_addr)
8840 kvm->arch.ept_identity_map_addr =
8841 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
8842 err = init_rmode_identity_map(kvm);
8848 nested_vmx_setup_ctls_msrs(vmx);
8849 vmx->nested.vpid02 = allocate_vpid();
8852 vmx->nested.posted_intr_nv = -1;
8853 vmx->nested.current_vmptr = -1ull;
8854 vmx->nested.current_vmcs12 = NULL;
8857 * If PML is turned on, failure on enabling PML just results in failure
8858 * of creating the vcpu, therefore we can simplify PML logic (by
8859 * avoiding dealing with cases, such as enabling PML partially on vcpus
8860 * for the guest, etc.
8863 err = vmx_create_pml_buffer(vmx);
8871 free_vpid(vmx->nested.vpid02);
8872 free_loaded_vmcs(vmx->loaded_vmcs);
8874 kfree(vmx->guest_msrs);
8876 kvm_vcpu_uninit(&vmx->vcpu);
8878 free_vpid(vmx->vpid);
8879 kmem_cache_free(kvm_vcpu_cache, vmx);
8880 return ERR_PTR(err);
8883 static void __init vmx_check_processor_compat(void *rtn)
8885 struct vmcs_config vmcs_conf;
8888 if (setup_vmcs_config(&vmcs_conf) < 0)
8890 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
8891 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
8892 smp_processor_id());
8897 static int get_ept_level(void)
8899 return VMX_EPT_DEFAULT_GAW + 1;
8902 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
8907 /* For VT-d and EPT combination
8908 * 1. MMIO: always map as UC
8910 * a. VT-d without snooping control feature: can't guarantee the
8911 * result, try to trust guest.
8912 * b. VT-d with snooping control feature: snooping control feature of
8913 * VT-d engine can guarantee the cache correctness. Just set it
8914 * to WB to keep consistent with host. So the same as item 3.
8915 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
8916 * consistent with host MTRR
8919 cache = MTRR_TYPE_UNCACHABLE;
8923 if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
8924 ipat = VMX_EPT_IPAT_BIT;
8925 cache = MTRR_TYPE_WRBACK;
8929 if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
8930 ipat = VMX_EPT_IPAT_BIT;
8931 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
8932 cache = MTRR_TYPE_WRBACK;
8934 cache = MTRR_TYPE_UNCACHABLE;
8938 cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
8941 return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
8944 static int vmx_get_lpage_level(void)
8946 if (enable_ept && !cpu_has_vmx_ept_1g_page())
8947 return PT_DIRECTORY_LEVEL;
8949 /* For shadow and EPT supported 1GB page */
8950 return PT_PDPE_LEVEL;
8953 static void vmcs_set_secondary_exec_control(u32 new_ctl)
8956 * These bits in the secondary execution controls field
8957 * are dynamic, the others are mostly based on the hypervisor
8958 * architecture and the guest's CPUID. Do not touch the
8962 SECONDARY_EXEC_SHADOW_VMCS |
8963 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
8964 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8966 u32 cur_ctl = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8968 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
8969 (new_ctl & ~mask) | (cur_ctl & mask));
8972 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
8974 struct kvm_cpuid_entry2 *best;
8975 struct vcpu_vmx *vmx = to_vmx(vcpu);
8976 u32 secondary_exec_ctl = vmx_secondary_exec_control(vmx);
8978 if (vmx_rdtscp_supported()) {
8979 bool rdtscp_enabled = guest_cpuid_has_rdtscp(vcpu);
8980 if (!rdtscp_enabled)
8981 secondary_exec_ctl &= ~SECONDARY_EXEC_RDTSCP;
8985 vmx->nested.nested_vmx_secondary_ctls_high |=
8986 SECONDARY_EXEC_RDTSCP;
8988 vmx->nested.nested_vmx_secondary_ctls_high &=
8989 ~SECONDARY_EXEC_RDTSCP;
8993 /* Exposing INVPCID only when PCID is exposed */
8994 best = kvm_find_cpuid_entry(vcpu, 0x7, 0);
8995 if (vmx_invpcid_supported() &&
8996 (!best || !(best->ebx & bit(X86_FEATURE_INVPCID)) ||
8997 !guest_cpuid_has_pcid(vcpu))) {
8998 secondary_exec_ctl &= ~SECONDARY_EXEC_ENABLE_INVPCID;
9001 best->ebx &= ~bit(X86_FEATURE_INVPCID);
9004 if (cpu_has_secondary_exec_ctrls())
9005 vmcs_set_secondary_exec_control(secondary_exec_ctl);
9007 if (static_cpu_has(X86_FEATURE_PCOMMIT) && nested) {
9008 if (guest_cpuid_has_pcommit(vcpu))
9009 vmx->nested.nested_vmx_secondary_ctls_high |=
9010 SECONDARY_EXEC_PCOMMIT;
9012 vmx->nested.nested_vmx_secondary_ctls_high &=
9013 ~SECONDARY_EXEC_PCOMMIT;
9017 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
9019 if (func == 1 && nested)
9020 entry->ecx |= bit(X86_FEATURE_VMX);
9023 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
9024 struct x86_exception *fault)
9026 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9029 if (fault->error_code & PFERR_RSVD_MASK)
9030 exit_reason = EXIT_REASON_EPT_MISCONFIG;
9032 exit_reason = EXIT_REASON_EPT_VIOLATION;
9033 nested_vmx_vmexit(vcpu, exit_reason, 0, vcpu->arch.exit_qualification);
9034 vmcs12->guest_physical_address = fault->address;
9037 /* Callbacks for nested_ept_init_mmu_context: */
9039 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
9041 /* return the page table to be shadowed - in our case, EPT12 */
9042 return get_vmcs12(vcpu)->ept_pointer;
9045 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
9047 WARN_ON(mmu_is_nested(vcpu));
9048 kvm_init_shadow_ept_mmu(vcpu,
9049 to_vmx(vcpu)->nested.nested_vmx_ept_caps &
9050 VMX_EPT_EXECUTE_ONLY_BIT);
9051 vcpu->arch.mmu.set_cr3 = vmx_set_cr3;
9052 vcpu->arch.mmu.get_cr3 = nested_ept_get_cr3;
9053 vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
9055 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
9058 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
9060 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
9063 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
9066 bool inequality, bit;
9068 bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
9070 (error_code & vmcs12->page_fault_error_code_mask) !=
9071 vmcs12->page_fault_error_code_match;
9072 return inequality ^ bit;
9075 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
9076 struct x86_exception *fault)
9078 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9080 WARN_ON(!is_guest_mode(vcpu));
9082 if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code))
9083 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
9084 vmcs_read32(VM_EXIT_INTR_INFO),
9085 vmcs_readl(EXIT_QUALIFICATION));
9087 kvm_inject_page_fault(vcpu, fault);
9090 static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu,
9091 struct vmcs12 *vmcs12)
9093 struct vcpu_vmx *vmx = to_vmx(vcpu);
9094 int maxphyaddr = cpuid_maxphyaddr(vcpu);
9096 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
9097 if (!PAGE_ALIGNED(vmcs12->apic_access_addr) ||
9098 vmcs12->apic_access_addr >> maxphyaddr)
9102 * Translate L1 physical address to host physical
9103 * address for vmcs02. Keep the page pinned, so this
9104 * physical address remains valid. We keep a reference
9105 * to it so we can release it later.
9107 if (vmx->nested.apic_access_page) /* shouldn't happen */
9108 nested_release_page(vmx->nested.apic_access_page);
9109 vmx->nested.apic_access_page =
9110 nested_get_page(vcpu, vmcs12->apic_access_addr);
9113 if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
9114 if (!PAGE_ALIGNED(vmcs12->virtual_apic_page_addr) ||
9115 vmcs12->virtual_apic_page_addr >> maxphyaddr)
9118 if (vmx->nested.virtual_apic_page) /* shouldn't happen */
9119 nested_release_page(vmx->nested.virtual_apic_page);
9120 vmx->nested.virtual_apic_page =
9121 nested_get_page(vcpu, vmcs12->virtual_apic_page_addr);
9124 * Failing the vm entry is _not_ what the processor does
9125 * but it's basically the only possibility we have.
9126 * We could still enter the guest if CR8 load exits are
9127 * enabled, CR8 store exits are enabled, and virtualize APIC
9128 * access is disabled; in this case the processor would never
9129 * use the TPR shadow and we could simply clear the bit from
9130 * the execution control. But such a configuration is useless,
9131 * so let's keep the code simple.
9133 if (!vmx->nested.virtual_apic_page)
9137 if (nested_cpu_has_posted_intr(vmcs12)) {
9138 if (!IS_ALIGNED(vmcs12->posted_intr_desc_addr, 64) ||
9139 vmcs12->posted_intr_desc_addr >> maxphyaddr)
9142 if (vmx->nested.pi_desc_page) { /* shouldn't happen */
9143 kunmap(vmx->nested.pi_desc_page);
9144 nested_release_page(vmx->nested.pi_desc_page);
9146 vmx->nested.pi_desc_page =
9147 nested_get_page(vcpu, vmcs12->posted_intr_desc_addr);
9148 if (!vmx->nested.pi_desc_page)
9151 vmx->nested.pi_desc =
9152 (struct pi_desc *)kmap(vmx->nested.pi_desc_page);
9153 if (!vmx->nested.pi_desc) {
9154 nested_release_page_clean(vmx->nested.pi_desc_page);
9157 vmx->nested.pi_desc =
9158 (struct pi_desc *)((void *)vmx->nested.pi_desc +
9159 (unsigned long)(vmcs12->posted_intr_desc_addr &
9166 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
9168 u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
9169 struct vcpu_vmx *vmx = to_vmx(vcpu);
9171 if (vcpu->arch.virtual_tsc_khz == 0)
9174 /* Make sure short timeouts reliably trigger an immediate vmexit.
9175 * hrtimer_start does not guarantee this. */
9176 if (preemption_timeout <= 1) {
9177 vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
9181 preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
9182 preemption_timeout *= 1000000;
9183 do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
9184 hrtimer_start(&vmx->nested.preemption_timer,
9185 ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
9188 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
9189 struct vmcs12 *vmcs12)
9194 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
9197 if (vmcs12_read_any(vcpu, MSR_BITMAP, &addr)) {
9201 maxphyaddr = cpuid_maxphyaddr(vcpu);
9203 if (!PAGE_ALIGNED(vmcs12->msr_bitmap) ||
9204 ((addr + PAGE_SIZE) >> maxphyaddr))
9211 * Merge L0's and L1's MSR bitmap, return false to indicate that
9212 * we do not use the hardware.
9214 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
9215 struct vmcs12 *vmcs12)
9219 unsigned long *msr_bitmap;
9221 if (!nested_cpu_has_virt_x2apic_mode(vmcs12))
9224 page = nested_get_page(vcpu, vmcs12->msr_bitmap);
9229 msr_bitmap = (unsigned long *)kmap(page);
9231 nested_release_page_clean(page);
9236 if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
9237 if (nested_cpu_has_apic_reg_virt(vmcs12))
9238 for (msr = 0x800; msr <= 0x8ff; msr++)
9239 nested_vmx_disable_intercept_for_msr(
9241 vmx_msr_bitmap_nested,
9243 /* TPR is allowed */
9244 nested_vmx_disable_intercept_for_msr(msr_bitmap,
9245 vmx_msr_bitmap_nested,
9246 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
9247 MSR_TYPE_R | MSR_TYPE_W);
9248 if (nested_cpu_has_vid(vmcs12)) {
9249 /* EOI and self-IPI are allowed */
9250 nested_vmx_disable_intercept_for_msr(
9252 vmx_msr_bitmap_nested,
9253 APIC_BASE_MSR + (APIC_EOI >> 4),
9255 nested_vmx_disable_intercept_for_msr(
9257 vmx_msr_bitmap_nested,
9258 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
9263 * Enable reading intercept of all the x2apic
9264 * MSRs. We should not rely on vmcs12 to do any
9265 * optimizations here, it may have been modified
9268 for (msr = 0x800; msr <= 0x8ff; msr++)
9269 __vmx_enable_intercept_for_msr(
9270 vmx_msr_bitmap_nested,
9274 __vmx_enable_intercept_for_msr(
9275 vmx_msr_bitmap_nested,
9276 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
9278 __vmx_enable_intercept_for_msr(
9279 vmx_msr_bitmap_nested,
9280 APIC_BASE_MSR + (APIC_EOI >> 4),
9282 __vmx_enable_intercept_for_msr(
9283 vmx_msr_bitmap_nested,
9284 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
9288 nested_release_page_clean(page);
9293 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
9294 struct vmcs12 *vmcs12)
9296 if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9297 !nested_cpu_has_apic_reg_virt(vmcs12) &&
9298 !nested_cpu_has_vid(vmcs12) &&
9299 !nested_cpu_has_posted_intr(vmcs12))
9303 * If virtualize x2apic mode is enabled,
9304 * virtualize apic access must be disabled.
9306 if (nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9307 nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
9311 * If virtual interrupt delivery is enabled,
9312 * we must exit on external interrupts.
9314 if (nested_cpu_has_vid(vmcs12) &&
9315 !nested_exit_on_intr(vcpu))
9319 * bits 15:8 should be zero in posted_intr_nv,
9320 * the descriptor address has been already checked
9321 * in nested_get_vmcs12_pages.
9323 if (nested_cpu_has_posted_intr(vmcs12) &&
9324 (!nested_cpu_has_vid(vmcs12) ||
9325 !nested_exit_intr_ack_set(vcpu) ||
9326 vmcs12->posted_intr_nv & 0xff00))
9329 /* tpr shadow is needed by all apicv features. */
9330 if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
9336 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
9337 unsigned long count_field,
9338 unsigned long addr_field)
9343 if (vmcs12_read_any(vcpu, count_field, &count) ||
9344 vmcs12_read_any(vcpu, addr_field, &addr)) {
9350 maxphyaddr = cpuid_maxphyaddr(vcpu);
9351 if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
9352 (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr) {
9353 pr_warn_ratelimited(
9354 "nVMX: invalid MSR switch (0x%lx, %d, %llu, 0x%08llx)",
9355 addr_field, maxphyaddr, count, addr);
9361 static int nested_vmx_check_msr_switch_controls(struct kvm_vcpu *vcpu,
9362 struct vmcs12 *vmcs12)
9364 if (vmcs12->vm_exit_msr_load_count == 0 &&
9365 vmcs12->vm_exit_msr_store_count == 0 &&
9366 vmcs12->vm_entry_msr_load_count == 0)
9367 return 0; /* Fast path */
9368 if (nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_LOAD_COUNT,
9369 VM_EXIT_MSR_LOAD_ADDR) ||
9370 nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_STORE_COUNT,
9371 VM_EXIT_MSR_STORE_ADDR) ||
9372 nested_vmx_check_msr_switch(vcpu, VM_ENTRY_MSR_LOAD_COUNT,
9373 VM_ENTRY_MSR_LOAD_ADDR))
9378 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
9379 struct vmx_msr_entry *e)
9381 /* x2APIC MSR accesses are not allowed */
9382 if (vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8)
9384 if (e->index == MSR_IA32_UCODE_WRITE || /* SDM Table 35-2 */
9385 e->index == MSR_IA32_UCODE_REV)
9387 if (e->reserved != 0)
9392 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
9393 struct vmx_msr_entry *e)
9395 if (e->index == MSR_FS_BASE ||
9396 e->index == MSR_GS_BASE ||
9397 e->index == MSR_IA32_SMM_MONITOR_CTL || /* SMM is not supported */
9398 nested_vmx_msr_check_common(vcpu, e))
9403 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
9404 struct vmx_msr_entry *e)
9406 if (e->index == MSR_IA32_SMBASE || /* SMM is not supported */
9407 nested_vmx_msr_check_common(vcpu, e))
9413 * Load guest's/host's msr at nested entry/exit.
9414 * return 0 for success, entry index for failure.
9416 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9419 struct vmx_msr_entry e;
9420 struct msr_data msr;
9422 msr.host_initiated = false;
9423 for (i = 0; i < count; i++) {
9424 if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
9426 pr_warn_ratelimited(
9427 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9428 __func__, i, gpa + i * sizeof(e));
9431 if (nested_vmx_load_msr_check(vcpu, &e)) {
9432 pr_warn_ratelimited(
9433 "%s check failed (%u, 0x%x, 0x%x)\n",
9434 __func__, i, e.index, e.reserved);
9437 msr.index = e.index;
9439 if (kvm_set_msr(vcpu, &msr)) {
9440 pr_warn_ratelimited(
9441 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9442 __func__, i, e.index, e.value);
9451 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9454 struct vmx_msr_entry e;
9456 for (i = 0; i < count; i++) {
9457 struct msr_data msr_info;
9458 if (kvm_vcpu_read_guest(vcpu,
9459 gpa + i * sizeof(e),
9460 &e, 2 * sizeof(u32))) {
9461 pr_warn_ratelimited(
9462 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9463 __func__, i, gpa + i * sizeof(e));
9466 if (nested_vmx_store_msr_check(vcpu, &e)) {
9467 pr_warn_ratelimited(
9468 "%s check failed (%u, 0x%x, 0x%x)\n",
9469 __func__, i, e.index, e.reserved);
9472 msr_info.host_initiated = false;
9473 msr_info.index = e.index;
9474 if (kvm_get_msr(vcpu, &msr_info)) {
9475 pr_warn_ratelimited(
9476 "%s cannot read MSR (%u, 0x%x)\n",
9477 __func__, i, e.index);
9480 if (kvm_vcpu_write_guest(vcpu,
9481 gpa + i * sizeof(e) +
9482 offsetof(struct vmx_msr_entry, value),
9483 &msr_info.data, sizeof(msr_info.data))) {
9484 pr_warn_ratelimited(
9485 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9486 __func__, i, e.index, msr_info.data);
9494 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
9495 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
9496 * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
9497 * guest in a way that will both be appropriate to L1's requests, and our
9498 * needs. In addition to modifying the active vmcs (which is vmcs02), this
9499 * function also has additional necessary side-effects, like setting various
9500 * vcpu->arch fields.
9502 static void prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
9504 struct vcpu_vmx *vmx = to_vmx(vcpu);
9507 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
9508 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
9509 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
9510 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
9511 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
9512 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
9513 vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
9514 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
9515 vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
9516 vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
9517 vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
9518 vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
9519 vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
9520 vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
9521 vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
9522 vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
9523 vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
9524 vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
9525 vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
9526 vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
9527 vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
9528 vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
9529 vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
9530 vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
9531 vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
9532 vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
9533 vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
9534 vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
9535 vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
9536 vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
9537 vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
9538 vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
9539 vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
9540 vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
9541 vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
9542 vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
9544 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
9545 kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
9546 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
9548 kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
9549 vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
9551 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
9552 vmcs12->vm_entry_intr_info_field);
9553 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
9554 vmcs12->vm_entry_exception_error_code);
9555 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
9556 vmcs12->vm_entry_instruction_len);
9557 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
9558 vmcs12->guest_interruptibility_info);
9559 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
9560 vmx_set_rflags(vcpu, vmcs12->guest_rflags);
9561 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
9562 vmcs12->guest_pending_dbg_exceptions);
9563 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
9564 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
9566 if (nested_cpu_has_xsaves(vmcs12))
9567 vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
9568 vmcs_write64(VMCS_LINK_POINTER, -1ull);
9570 exec_control = vmcs12->pin_based_vm_exec_control;
9571 exec_control |= vmcs_config.pin_based_exec_ctrl;
9572 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
9574 if (nested_cpu_has_posted_intr(vmcs12)) {
9576 * Note that we use L0's vector here and in
9577 * vmx_deliver_nested_posted_interrupt.
9579 vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
9580 vmx->nested.pi_pending = false;
9581 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
9582 vmcs_write64(POSTED_INTR_DESC_ADDR,
9583 page_to_phys(vmx->nested.pi_desc_page) +
9584 (unsigned long)(vmcs12->posted_intr_desc_addr &
9587 exec_control &= ~PIN_BASED_POSTED_INTR;
9589 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
9591 vmx->nested.preemption_timer_expired = false;
9592 if (nested_cpu_has_preemption_timer(vmcs12))
9593 vmx_start_preemption_timer(vcpu);
9596 * Whether page-faults are trapped is determined by a combination of
9597 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
9598 * If enable_ept, L0 doesn't care about page faults and we should
9599 * set all of these to L1's desires. However, if !enable_ept, L0 does
9600 * care about (at least some) page faults, and because it is not easy
9601 * (if at all possible?) to merge L0 and L1's desires, we simply ask
9602 * to exit on each and every L2 page fault. This is done by setting
9603 * MASK=MATCH=0 and (see below) EB.PF=1.
9604 * Note that below we don't need special code to set EB.PF beyond the
9605 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
9606 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
9607 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
9609 * A problem with this approach (when !enable_ept) is that L1 may be
9610 * injected with more page faults than it asked for. This could have
9611 * caused problems, but in practice existing hypervisors don't care.
9612 * To fix this, we will need to emulate the PFEC checking (on the L1
9613 * page tables), using walk_addr(), when injecting PFs to L1.
9615 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
9616 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
9617 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
9618 enable_ept ? vmcs12->page_fault_error_code_match : 0);
9620 if (cpu_has_secondary_exec_ctrls()) {
9621 exec_control = vmx_secondary_exec_control(vmx);
9623 /* Take the following fields only from vmcs12 */
9624 exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
9625 SECONDARY_EXEC_RDTSCP |
9626 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
9627 SECONDARY_EXEC_APIC_REGISTER_VIRT |
9628 SECONDARY_EXEC_PCOMMIT);
9629 if (nested_cpu_has(vmcs12,
9630 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
9631 exec_control |= vmcs12->secondary_vm_exec_control;
9633 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) {
9635 * If translation failed, no matter: This feature asks
9636 * to exit when accessing the given address, and if it
9637 * can never be accessed, this feature won't do
9640 if (!vmx->nested.apic_access_page)
9642 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9644 vmcs_write64(APIC_ACCESS_ADDR,
9645 page_to_phys(vmx->nested.apic_access_page));
9646 } else if (!(nested_cpu_has_virt_x2apic_mode(vmcs12)) &&
9647 cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9649 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9650 kvm_vcpu_reload_apic_access_page(vcpu);
9653 if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
9654 vmcs_write64(EOI_EXIT_BITMAP0,
9655 vmcs12->eoi_exit_bitmap0);
9656 vmcs_write64(EOI_EXIT_BITMAP1,
9657 vmcs12->eoi_exit_bitmap1);
9658 vmcs_write64(EOI_EXIT_BITMAP2,
9659 vmcs12->eoi_exit_bitmap2);
9660 vmcs_write64(EOI_EXIT_BITMAP3,
9661 vmcs12->eoi_exit_bitmap3);
9662 vmcs_write16(GUEST_INTR_STATUS,
9663 vmcs12->guest_intr_status);
9666 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
9671 * Set host-state according to L0's settings (vmcs12 is irrelevant here)
9672 * Some constant fields are set here by vmx_set_constant_host_state().
9673 * Other fields are different per CPU, and will be set later when
9674 * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
9676 vmx_set_constant_host_state(vmx);
9679 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
9680 * entry, but only if the current (host) sp changed from the value
9681 * we wrote last (vmx->host_rsp). This cache is no longer relevant
9682 * if we switch vmcs, and rather than hold a separate cache per vmcs,
9683 * here we just force the write to happen on entry.
9687 exec_control = vmx_exec_control(vmx); /* L0's desires */
9688 exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
9689 exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
9690 exec_control &= ~CPU_BASED_TPR_SHADOW;
9691 exec_control |= vmcs12->cpu_based_vm_exec_control;
9693 if (exec_control & CPU_BASED_TPR_SHADOW) {
9694 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
9695 page_to_phys(vmx->nested.virtual_apic_page));
9696 vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
9699 if (cpu_has_vmx_msr_bitmap() &&
9700 exec_control & CPU_BASED_USE_MSR_BITMAPS) {
9701 nested_vmx_merge_msr_bitmap(vcpu, vmcs12);
9702 /* MSR_BITMAP will be set by following vmx_set_efer. */
9704 exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
9707 * Merging of IO bitmap not currently supported.
9708 * Rather, exit every time.
9710 exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
9711 exec_control |= CPU_BASED_UNCOND_IO_EXITING;
9713 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
9715 /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
9716 * bitwise-or of what L1 wants to trap for L2, and what we want to
9717 * trap. Note that CR0.TS also needs updating - we do this later.
9719 update_exception_bitmap(vcpu);
9720 vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
9721 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
9723 /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
9724 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
9725 * bits are further modified by vmx_set_efer() below.
9727 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
9729 /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
9730 * emulated by vmx_set_efer(), below.
9732 vm_entry_controls_init(vmx,
9733 (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER &
9734 ~VM_ENTRY_IA32E_MODE) |
9735 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
9737 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) {
9738 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
9739 vcpu->arch.pat = vmcs12->guest_ia32_pat;
9740 } else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
9741 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
9744 set_cr4_guest_host_mask(vmx);
9746 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)
9747 vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
9749 if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
9750 vmcs_write64(TSC_OFFSET,
9751 vmx->nested.vmcs01_tsc_offset + vmcs12->tsc_offset);
9753 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
9757 * There is no direct mapping between vpid02 and vpid12, the
9758 * vpid02 is per-vCPU for L0 and reused while the value of
9759 * vpid12 is changed w/ one invvpid during nested vmentry.
9760 * The vpid12 is allocated by L1 for L2, so it will not
9761 * influence global bitmap(for vpid01 and vpid02 allocation)
9762 * even if spawn a lot of nested vCPUs.
9764 if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02) {
9765 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
9766 if (vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
9767 vmx->nested.last_vpid = vmcs12->virtual_processor_id;
9768 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->nested.vpid02);
9771 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
9772 vmx_flush_tlb(vcpu);
9777 if (nested_cpu_has_ept(vmcs12)) {
9778 kvm_mmu_unload(vcpu);
9779 nested_ept_init_mmu_context(vcpu);
9782 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)
9783 vcpu->arch.efer = vmcs12->guest_ia32_efer;
9784 else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
9785 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
9787 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
9788 /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
9789 vmx_set_efer(vcpu, vcpu->arch.efer);
9792 * This sets GUEST_CR0 to vmcs12->guest_cr0, with possibly a modified
9793 * TS bit (for lazy fpu) and bits which we consider mandatory enabled.
9794 * The CR0_READ_SHADOW is what L2 should have expected to read given
9795 * the specifications by L1; It's not enough to take
9796 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
9797 * have more bits than L1 expected.
9799 vmx_set_cr0(vcpu, vmcs12->guest_cr0);
9800 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
9802 vmx_set_cr4(vcpu, vmcs12->guest_cr4);
9803 vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
9805 /* shadow page tables on either EPT or shadow page tables */
9806 kvm_set_cr3(vcpu, vmcs12->guest_cr3);
9807 kvm_mmu_reset_context(vcpu);
9810 vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
9813 * L1 may access the L2's PDPTR, so save them to construct vmcs12
9816 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
9817 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
9818 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
9819 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
9822 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
9823 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
9827 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
9828 * for running an L2 nested guest.
9830 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
9832 struct vmcs12 *vmcs12;
9833 struct vcpu_vmx *vmx = to_vmx(vcpu);
9835 struct loaded_vmcs *vmcs02;
9839 if (!nested_vmx_check_permission(vcpu) ||
9840 !nested_vmx_check_vmcs12(vcpu))
9843 skip_emulated_instruction(vcpu);
9844 vmcs12 = get_vmcs12(vcpu);
9846 if (enable_shadow_vmcs)
9847 copy_shadow_to_vmcs12(vmx);
9850 * The nested entry process starts with enforcing various prerequisites
9851 * on vmcs12 as required by the Intel SDM, and act appropriately when
9852 * they fail: As the SDM explains, some conditions should cause the
9853 * instruction to fail, while others will cause the instruction to seem
9854 * to succeed, but return an EXIT_REASON_INVALID_STATE.
9855 * To speed up the normal (success) code path, we should avoid checking
9856 * for misconfigurations which will anyway be caught by the processor
9857 * when using the merged vmcs02.
9859 if (vmcs12->launch_state == launch) {
9860 nested_vmx_failValid(vcpu,
9861 launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
9862 : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
9866 if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
9867 vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT) {
9868 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9872 if (!nested_get_vmcs12_pages(vcpu, vmcs12)) {
9873 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9877 if (nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12)) {
9878 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9882 if (nested_vmx_check_apicv_controls(vcpu, vmcs12)) {
9883 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9887 if (nested_vmx_check_msr_switch_controls(vcpu, vmcs12)) {
9888 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9892 if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
9893 vmx->nested.nested_vmx_true_procbased_ctls_low,
9894 vmx->nested.nested_vmx_procbased_ctls_high) ||
9895 !vmx_control_verify(vmcs12->secondary_vm_exec_control,
9896 vmx->nested.nested_vmx_secondary_ctls_low,
9897 vmx->nested.nested_vmx_secondary_ctls_high) ||
9898 !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
9899 vmx->nested.nested_vmx_pinbased_ctls_low,
9900 vmx->nested.nested_vmx_pinbased_ctls_high) ||
9901 !vmx_control_verify(vmcs12->vm_exit_controls,
9902 vmx->nested.nested_vmx_true_exit_ctls_low,
9903 vmx->nested.nested_vmx_exit_ctls_high) ||
9904 !vmx_control_verify(vmcs12->vm_entry_controls,
9905 vmx->nested.nested_vmx_true_entry_ctls_low,
9906 vmx->nested.nested_vmx_entry_ctls_high))
9908 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9912 if (((vmcs12->host_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
9913 ((vmcs12->host_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
9914 nested_vmx_failValid(vcpu,
9915 VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
9919 if (!nested_cr0_valid(vcpu, vmcs12->guest_cr0) ||
9920 ((vmcs12->guest_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
9921 nested_vmx_entry_failure(vcpu, vmcs12,
9922 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
9925 if (vmcs12->vmcs_link_pointer != -1ull) {
9926 nested_vmx_entry_failure(vcpu, vmcs12,
9927 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_VMCS_LINK_PTR);
9932 * If the load IA32_EFER VM-entry control is 1, the following checks
9933 * are performed on the field for the IA32_EFER MSR:
9934 * - Bits reserved in the IA32_EFER MSR must be 0.
9935 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
9936 * the IA-32e mode guest VM-exit control. It must also be identical
9937 * to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
9940 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER) {
9941 ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
9942 if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
9943 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
9944 ((vmcs12->guest_cr0 & X86_CR0_PG) &&
9945 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))) {
9946 nested_vmx_entry_failure(vcpu, vmcs12,
9947 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
9953 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
9954 * IA32_EFER MSR must be 0 in the field for that register. In addition,
9955 * the values of the LMA and LME bits in the field must each be that of
9956 * the host address-space size VM-exit control.
9958 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
9959 ia32e = (vmcs12->vm_exit_controls &
9960 VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
9961 if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
9962 ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
9963 ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)) {
9964 nested_vmx_entry_failure(vcpu, vmcs12,
9965 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
9971 * We're finally done with prerequisite checking, and can start with
9975 vmcs02 = nested_get_current_vmcs02(vmx);
9979 enter_guest_mode(vcpu);
9981 vmx->nested.vmcs01_tsc_offset = vmcs_read64(TSC_OFFSET);
9983 if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
9984 vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
9987 vmx->loaded_vmcs = vmcs02;
9989 vmx_vcpu_load(vcpu, cpu);
9993 vmx_segment_cache_clear(vmx);
9995 prepare_vmcs02(vcpu, vmcs12);
9997 msr_entry_idx = nested_vmx_load_msr(vcpu,
9998 vmcs12->vm_entry_msr_load_addr,
9999 vmcs12->vm_entry_msr_load_count);
10000 if (msr_entry_idx) {
10001 leave_guest_mode(vcpu);
10002 vmx_load_vmcs01(vcpu);
10003 nested_vmx_entry_failure(vcpu, vmcs12,
10004 EXIT_REASON_MSR_LOAD_FAIL, msr_entry_idx);
10008 vmcs12->launch_state = 1;
10010 if (vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT)
10011 return kvm_vcpu_halt(vcpu);
10013 vmx->nested.nested_run_pending = 1;
10016 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
10017 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
10018 * returned as far as L1 is concerned. It will only return (and set
10019 * the success flag) when L2 exits (see nested_vmx_vmexit()).
10025 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
10026 * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
10027 * This function returns the new value we should put in vmcs12.guest_cr0.
10028 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
10029 * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
10030 * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
10031 * didn't trap the bit, because if L1 did, so would L0).
10032 * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have
10033 * been modified by L2, and L1 knows it. So just leave the old value of
10034 * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
10035 * isn't relevant, because if L0 traps this bit it can set it to anything.
10036 * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
10037 * changed these bits, and therefore they need to be updated, but L0
10038 * didn't necessarily allow them to be changed in GUEST_CR0 - and rather
10039 * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
10041 static inline unsigned long
10042 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10045 /*1*/ (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
10046 /*2*/ (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
10047 /*3*/ (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
10048 vcpu->arch.cr0_guest_owned_bits));
10051 static inline unsigned long
10052 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10055 /*1*/ (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
10056 /*2*/ (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
10057 /*3*/ (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
10058 vcpu->arch.cr4_guest_owned_bits));
10061 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
10062 struct vmcs12 *vmcs12)
10067 if (vcpu->arch.exception.pending && vcpu->arch.exception.reinject) {
10068 nr = vcpu->arch.exception.nr;
10069 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
10071 if (kvm_exception_is_soft(nr)) {
10072 vmcs12->vm_exit_instruction_len =
10073 vcpu->arch.event_exit_inst_len;
10074 idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
10076 idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
10078 if (vcpu->arch.exception.has_error_code) {
10079 idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
10080 vmcs12->idt_vectoring_error_code =
10081 vcpu->arch.exception.error_code;
10084 vmcs12->idt_vectoring_info_field = idt_vectoring;
10085 } else if (vcpu->arch.nmi_injected) {
10086 vmcs12->idt_vectoring_info_field =
10087 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
10088 } else if (vcpu->arch.interrupt.pending) {
10089 nr = vcpu->arch.interrupt.nr;
10090 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
10092 if (vcpu->arch.interrupt.soft) {
10093 idt_vectoring |= INTR_TYPE_SOFT_INTR;
10094 vmcs12->vm_entry_instruction_len =
10095 vcpu->arch.event_exit_inst_len;
10097 idt_vectoring |= INTR_TYPE_EXT_INTR;
10099 vmcs12->idt_vectoring_info_field = idt_vectoring;
10103 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
10105 struct vcpu_vmx *vmx = to_vmx(vcpu);
10107 if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
10108 vmx->nested.preemption_timer_expired) {
10109 if (vmx->nested.nested_run_pending)
10111 nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
10115 if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
10116 if (vmx->nested.nested_run_pending ||
10117 vcpu->arch.interrupt.pending)
10119 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
10120 NMI_VECTOR | INTR_TYPE_NMI_INTR |
10121 INTR_INFO_VALID_MASK, 0);
10123 * The NMI-triggered VM exit counts as injection:
10124 * clear this one and block further NMIs.
10126 vcpu->arch.nmi_pending = 0;
10127 vmx_set_nmi_mask(vcpu, true);
10131 if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
10132 nested_exit_on_intr(vcpu)) {
10133 if (vmx->nested.nested_run_pending)
10135 nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
10139 return vmx_complete_nested_posted_interrupt(vcpu);
10142 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
10144 ktime_t remaining =
10145 hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
10148 if (ktime_to_ns(remaining) <= 0)
10151 value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
10152 do_div(value, 1000000);
10153 return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
10157 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
10158 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
10159 * and this function updates it to reflect the changes to the guest state while
10160 * L2 was running (and perhaps made some exits which were handled directly by L0
10161 * without going back to L1), and to reflect the exit reason.
10162 * Note that we do not have to copy here all VMCS fields, just those that
10163 * could have changed by the L2 guest or the exit - i.e., the guest-state and
10164 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
10165 * which already writes to vmcs12 directly.
10167 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10168 u32 exit_reason, u32 exit_intr_info,
10169 unsigned long exit_qualification)
10171 /* update guest state fields: */
10172 vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
10173 vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
10175 vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
10176 vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
10177 vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
10179 vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
10180 vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
10181 vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
10182 vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
10183 vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
10184 vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
10185 vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
10186 vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
10187 vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
10188 vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
10189 vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
10190 vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
10191 vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
10192 vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
10193 vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
10194 vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
10195 vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
10196 vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
10197 vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
10198 vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
10199 vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
10200 vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
10201 vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
10202 vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
10203 vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
10204 vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
10205 vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
10206 vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
10207 vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
10208 vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
10209 vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
10210 vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
10211 vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
10212 vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
10213 vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
10214 vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
10216 vmcs12->guest_interruptibility_info =
10217 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
10218 vmcs12->guest_pending_dbg_exceptions =
10219 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
10220 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
10221 vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
10223 vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
10225 if (nested_cpu_has_preemption_timer(vmcs12)) {
10226 if (vmcs12->vm_exit_controls &
10227 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
10228 vmcs12->vmx_preemption_timer_value =
10229 vmx_get_preemption_timer_value(vcpu);
10230 hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
10234 * In some cases (usually, nested EPT), L2 is allowed to change its
10235 * own CR3 without exiting. If it has changed it, we must keep it.
10236 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
10237 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
10239 * Additionally, restore L2's PDPTR to vmcs12.
10242 vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
10243 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
10244 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
10245 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
10246 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
10249 if (nested_cpu_has_vid(vmcs12))
10250 vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
10252 vmcs12->vm_entry_controls =
10253 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
10254 (vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
10256 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) {
10257 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
10258 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
10261 /* TODO: These cannot have changed unless we have MSR bitmaps and
10262 * the relevant bit asks not to trap the change */
10263 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
10264 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
10265 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
10266 vmcs12->guest_ia32_efer = vcpu->arch.efer;
10267 vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
10268 vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
10269 vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
10270 if (vmx_mpx_supported())
10271 vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
10272 if (nested_cpu_has_xsaves(vmcs12))
10273 vmcs12->xss_exit_bitmap = vmcs_read64(XSS_EXIT_BITMAP);
10275 /* update exit information fields: */
10277 vmcs12->vm_exit_reason = exit_reason;
10278 vmcs12->exit_qualification = exit_qualification;
10280 vmcs12->vm_exit_intr_info = exit_intr_info;
10281 if ((vmcs12->vm_exit_intr_info &
10282 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
10283 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK))
10284 vmcs12->vm_exit_intr_error_code =
10285 vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
10286 vmcs12->idt_vectoring_info_field = 0;
10287 vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
10288 vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
10290 if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
10291 /* vm_entry_intr_info_field is cleared on exit. Emulate this
10292 * instead of reading the real value. */
10293 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
10296 * Transfer the event that L0 or L1 may wanted to inject into
10297 * L2 to IDT_VECTORING_INFO_FIELD.
10299 vmcs12_save_pending_event(vcpu, vmcs12);
10303 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
10304 * preserved above and would only end up incorrectly in L1.
10306 vcpu->arch.nmi_injected = false;
10307 kvm_clear_exception_queue(vcpu);
10308 kvm_clear_interrupt_queue(vcpu);
10312 * A part of what we need to when the nested L2 guest exits and we want to
10313 * run its L1 parent, is to reset L1's guest state to the host state specified
10315 * This function is to be called not only on normal nested exit, but also on
10316 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
10317 * Failures During or After Loading Guest State").
10318 * This function should be called when the active VMCS is L1's (vmcs01).
10320 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
10321 struct vmcs12 *vmcs12)
10323 struct kvm_segment seg;
10325 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
10326 vcpu->arch.efer = vmcs12->host_ia32_efer;
10327 else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
10328 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10330 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10331 vmx_set_efer(vcpu, vcpu->arch.efer);
10333 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
10334 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
10335 vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
10337 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
10338 * actually changed, because it depends on the current state of
10339 * fpu_active (which may have changed).
10340 * Note that vmx_set_cr0 refers to efer set above.
10342 vmx_set_cr0(vcpu, vmcs12->host_cr0);
10344 * If we did fpu_activate()/fpu_deactivate() during L2's run, we need
10345 * to apply the same changes to L1's vmcs. We just set cr0 correctly,
10346 * but we also need to update cr0_guest_host_mask and exception_bitmap.
10348 update_exception_bitmap(vcpu);
10349 vcpu->arch.cr0_guest_owned_bits = (vcpu->fpu_active ? X86_CR0_TS : 0);
10350 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
10353 * Note that CR4_GUEST_HOST_MASK is already set in the original vmcs01
10354 * (KVM doesn't change it)- no reason to call set_cr4_guest_host_mask();
10356 vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
10357 kvm_set_cr4(vcpu, vmcs12->host_cr4);
10359 nested_ept_uninit_mmu_context(vcpu);
10361 kvm_set_cr3(vcpu, vmcs12->host_cr3);
10362 kvm_mmu_reset_context(vcpu);
10365 vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
10369 * Trivially support vpid by letting L2s share their parent
10370 * L1's vpid. TODO: move to a more elaborate solution, giving
10371 * each L2 its own vpid and exposing the vpid feature to L1.
10373 vmx_flush_tlb(vcpu);
10377 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
10378 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
10379 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
10380 vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
10381 vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
10383 /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1. */
10384 if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
10385 vmcs_write64(GUEST_BNDCFGS, 0);
10387 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
10388 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
10389 vcpu->arch.pat = vmcs12->host_ia32_pat;
10391 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
10392 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
10393 vmcs12->host_ia32_perf_global_ctrl);
10395 /* Set L1 segment info according to Intel SDM
10396 27.5.2 Loading Host Segment and Descriptor-Table Registers */
10397 seg = (struct kvm_segment) {
10399 .limit = 0xFFFFFFFF,
10400 .selector = vmcs12->host_cs_selector,
10406 if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
10410 vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
10411 seg = (struct kvm_segment) {
10413 .limit = 0xFFFFFFFF,
10420 seg.selector = vmcs12->host_ds_selector;
10421 vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
10422 seg.selector = vmcs12->host_es_selector;
10423 vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
10424 seg.selector = vmcs12->host_ss_selector;
10425 vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
10426 seg.selector = vmcs12->host_fs_selector;
10427 seg.base = vmcs12->host_fs_base;
10428 vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
10429 seg.selector = vmcs12->host_gs_selector;
10430 seg.base = vmcs12->host_gs_base;
10431 vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
10432 seg = (struct kvm_segment) {
10433 .base = vmcs12->host_tr_base,
10435 .selector = vmcs12->host_tr_selector,
10439 vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
10441 kvm_set_dr(vcpu, 7, 0x400);
10442 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
10444 if (cpu_has_vmx_msr_bitmap())
10445 vmx_set_msr_bitmap(vcpu);
10447 if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
10448 vmcs12->vm_exit_msr_load_count))
10449 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
10453 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
10454 * and modify vmcs12 to make it see what it would expect to see there if
10455 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
10457 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
10458 u32 exit_intr_info,
10459 unsigned long exit_qualification)
10461 struct vcpu_vmx *vmx = to_vmx(vcpu);
10462 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
10464 /* trying to cancel vmlaunch/vmresume is a bug */
10465 WARN_ON_ONCE(vmx->nested.nested_run_pending);
10467 leave_guest_mode(vcpu);
10468 prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
10469 exit_qualification);
10471 if (nested_vmx_store_msr(vcpu, vmcs12->vm_exit_msr_store_addr,
10472 vmcs12->vm_exit_msr_store_count))
10473 nested_vmx_abort(vcpu, VMX_ABORT_SAVE_GUEST_MSR_FAIL);
10475 vmx_load_vmcs01(vcpu);
10477 if ((exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
10478 && nested_exit_intr_ack_set(vcpu)) {
10479 int irq = kvm_cpu_get_interrupt(vcpu);
10481 vmcs12->vm_exit_intr_info = irq |
10482 INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
10485 trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
10486 vmcs12->exit_qualification,
10487 vmcs12->idt_vectoring_info_field,
10488 vmcs12->vm_exit_intr_info,
10489 vmcs12->vm_exit_intr_error_code,
10492 vm_entry_controls_init(vmx, vmcs_read32(VM_ENTRY_CONTROLS));
10493 vm_exit_controls_init(vmx, vmcs_read32(VM_EXIT_CONTROLS));
10494 vmx_segment_cache_clear(vmx);
10496 /* if no vmcs02 cache requested, remove the one we used */
10497 if (VMCS02_POOL_SIZE == 0)
10498 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
10500 load_vmcs12_host_state(vcpu, vmcs12);
10502 /* Update TSC_OFFSET if TSC was changed while L2 ran */
10503 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
10505 /* This is needed for same reason as it was needed in prepare_vmcs02 */
10508 /* Unpin physical memory we referred to in vmcs02 */
10509 if (vmx->nested.apic_access_page) {
10510 nested_release_page(vmx->nested.apic_access_page);
10511 vmx->nested.apic_access_page = NULL;
10513 if (vmx->nested.virtual_apic_page) {
10514 nested_release_page(vmx->nested.virtual_apic_page);
10515 vmx->nested.virtual_apic_page = NULL;
10517 if (vmx->nested.pi_desc_page) {
10518 kunmap(vmx->nested.pi_desc_page);
10519 nested_release_page(vmx->nested.pi_desc_page);
10520 vmx->nested.pi_desc_page = NULL;
10521 vmx->nested.pi_desc = NULL;
10525 * We are now running in L2, mmu_notifier will force to reload the
10526 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
10528 kvm_vcpu_reload_apic_access_page(vcpu);
10531 * Exiting from L2 to L1, we're now back to L1 which thinks it just
10532 * finished a VMLAUNCH or VMRESUME instruction, so we need to set the
10533 * success or failure flag accordingly.
10535 if (unlikely(vmx->fail)) {
10537 nested_vmx_failValid(vcpu, vmcs_read32(VM_INSTRUCTION_ERROR));
10539 nested_vmx_succeed(vcpu);
10540 if (enable_shadow_vmcs)
10541 vmx->nested.sync_shadow_vmcs = true;
10543 /* in case we halted in L2 */
10544 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10548 * Forcibly leave nested mode in order to be able to reset the VCPU later on.
10550 static void vmx_leave_nested(struct kvm_vcpu *vcpu)
10552 if (is_guest_mode(vcpu))
10553 nested_vmx_vmexit(vcpu, -1, 0, 0);
10554 free_nested(to_vmx(vcpu));
10558 * L1's failure to enter L2 is a subset of a normal exit, as explained in
10559 * 23.7 "VM-entry failures during or after loading guest state" (this also
10560 * lists the acceptable exit-reason and exit-qualification parameters).
10561 * It should only be called before L2 actually succeeded to run, and when
10562 * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
10564 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
10565 struct vmcs12 *vmcs12,
10566 u32 reason, unsigned long qualification)
10568 load_vmcs12_host_state(vcpu, vmcs12);
10569 vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
10570 vmcs12->exit_qualification = qualification;
10571 nested_vmx_succeed(vcpu);
10572 if (enable_shadow_vmcs)
10573 to_vmx(vcpu)->nested.sync_shadow_vmcs = true;
10576 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
10577 struct x86_instruction_info *info,
10578 enum x86_intercept_stage stage)
10580 return X86EMUL_CONTINUE;
10583 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
10586 shrink_ple_window(vcpu);
10589 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
10590 struct kvm_memory_slot *slot)
10592 kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
10593 kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
10596 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
10597 struct kvm_memory_slot *slot)
10599 kvm_mmu_slot_set_dirty(kvm, slot);
10602 static void vmx_flush_log_dirty(struct kvm *kvm)
10604 kvm_flush_pml_buffers(kvm);
10607 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
10608 struct kvm_memory_slot *memslot,
10609 gfn_t offset, unsigned long mask)
10611 kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
10615 * This routine does the following things for vCPU which is going
10616 * to be blocked if VT-d PI is enabled.
10617 * - Store the vCPU to the wakeup list, so when interrupts happen
10618 * we can find the right vCPU to wake up.
10619 * - Change the Posted-interrupt descriptor as below:
10620 * 'NDST' <-- vcpu->pre_pcpu
10621 * 'NV' <-- POSTED_INTR_WAKEUP_VECTOR
10622 * - If 'ON' is set during this process, which means at least one
10623 * interrupt is posted for this vCPU, we cannot block it, in
10624 * this case, return 1, otherwise, return 0.
10627 static int vmx_pre_block(struct kvm_vcpu *vcpu)
10629 unsigned long flags;
10631 struct pi_desc old, new;
10632 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
10634 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
10635 !irq_remapping_cap(IRQ_POSTING_CAP))
10638 vcpu->pre_pcpu = vcpu->cpu;
10639 spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
10640 vcpu->pre_pcpu), flags);
10641 list_add_tail(&vcpu->blocked_vcpu_list,
10642 &per_cpu(blocked_vcpu_on_cpu,
10644 spin_unlock_irqrestore(&per_cpu(blocked_vcpu_on_cpu_lock,
10645 vcpu->pre_pcpu), flags);
10648 old.control = new.control = pi_desc->control;
10651 * We should not block the vCPU if
10652 * an interrupt is posted for it.
10654 if (pi_test_on(pi_desc) == 1) {
10655 spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
10656 vcpu->pre_pcpu), flags);
10657 list_del(&vcpu->blocked_vcpu_list);
10658 spin_unlock_irqrestore(
10659 &per_cpu(blocked_vcpu_on_cpu_lock,
10660 vcpu->pre_pcpu), flags);
10661 vcpu->pre_pcpu = -1;
10666 WARN((pi_desc->sn == 1),
10667 "Warning: SN field of posted-interrupts "
10668 "is set before blocking\n");
10671 * Since vCPU can be preempted during this process,
10672 * vcpu->cpu could be different with pre_pcpu, we
10673 * need to set pre_pcpu as the destination of wakeup
10674 * notification event, then we can find the right vCPU
10675 * to wakeup in wakeup handler if interrupts happen
10676 * when the vCPU is in blocked state.
10678 dest = cpu_physical_id(vcpu->pre_pcpu);
10680 if (x2apic_enabled())
10683 new.ndst = (dest << 8) & 0xFF00;
10685 /* set 'NV' to 'wakeup vector' */
10686 new.nv = POSTED_INTR_WAKEUP_VECTOR;
10687 } while (cmpxchg(&pi_desc->control, old.control,
10688 new.control) != old.control);
10693 static void vmx_post_block(struct kvm_vcpu *vcpu)
10695 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
10696 struct pi_desc old, new;
10698 unsigned long flags;
10700 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
10701 !irq_remapping_cap(IRQ_POSTING_CAP))
10705 old.control = new.control = pi_desc->control;
10707 dest = cpu_physical_id(vcpu->cpu);
10709 if (x2apic_enabled())
10712 new.ndst = (dest << 8) & 0xFF00;
10714 /* Allow posting non-urgent interrupts */
10717 /* set 'NV' to 'notification vector' */
10718 new.nv = POSTED_INTR_VECTOR;
10719 } while (cmpxchg(&pi_desc->control, old.control,
10720 new.control) != old.control);
10722 if(vcpu->pre_pcpu != -1) {
10724 &per_cpu(blocked_vcpu_on_cpu_lock,
10725 vcpu->pre_pcpu), flags);
10726 list_del(&vcpu->blocked_vcpu_list);
10727 spin_unlock_irqrestore(
10728 &per_cpu(blocked_vcpu_on_cpu_lock,
10729 vcpu->pre_pcpu), flags);
10730 vcpu->pre_pcpu = -1;
10735 * vmx_update_pi_irte - set IRTE for Posted-Interrupts
10738 * @host_irq: host irq of the interrupt
10739 * @guest_irq: gsi of the interrupt
10740 * @set: set or unset PI
10741 * returns 0 on success, < 0 on failure
10743 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
10744 uint32_t guest_irq, bool set)
10746 struct kvm_kernel_irq_routing_entry *e;
10747 struct kvm_irq_routing_table *irq_rt;
10748 struct kvm_lapic_irq irq;
10749 struct kvm_vcpu *vcpu;
10750 struct vcpu_data vcpu_info;
10751 int idx, ret = -EINVAL;
10753 if (!kvm_arch_has_assigned_device(kvm) ||
10754 !irq_remapping_cap(IRQ_POSTING_CAP))
10757 idx = srcu_read_lock(&kvm->irq_srcu);
10758 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
10759 BUG_ON(guest_irq >= irq_rt->nr_rt_entries);
10761 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
10762 if (e->type != KVM_IRQ_ROUTING_MSI)
10765 * VT-d PI cannot support posting multicast/broadcast
10766 * interrupts to a vCPU, we still use interrupt remapping
10767 * for these kind of interrupts.
10769 * For lowest-priority interrupts, we only support
10770 * those with single CPU as the destination, e.g. user
10771 * configures the interrupts via /proc/irq or uses
10772 * irqbalance to make the interrupts single-CPU.
10774 * We will support full lowest-priority interrupt later.
10777 kvm_set_msi_irq(e, &irq);
10778 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu))
10781 vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
10782 vcpu_info.vector = irq.vector;
10784 trace_kvm_pi_irte_update(vcpu->vcpu_id, e->gsi,
10785 vcpu_info.vector, vcpu_info.pi_desc_addr, set);
10788 ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
10790 /* suppress notification event before unposting */
10791 pi_set_sn(vcpu_to_pi_desc(vcpu));
10792 ret = irq_set_vcpu_affinity(host_irq, NULL);
10793 pi_clear_sn(vcpu_to_pi_desc(vcpu));
10797 printk(KERN_INFO "%s: failed to update PI IRTE\n",
10805 srcu_read_unlock(&kvm->irq_srcu, idx);
10809 static struct kvm_x86_ops vmx_x86_ops = {
10810 .cpu_has_kvm_support = cpu_has_kvm_support,
10811 .disabled_by_bios = vmx_disabled_by_bios,
10812 .hardware_setup = hardware_setup,
10813 .hardware_unsetup = hardware_unsetup,
10814 .check_processor_compatibility = vmx_check_processor_compat,
10815 .hardware_enable = hardware_enable,
10816 .hardware_disable = hardware_disable,
10817 .cpu_has_accelerated_tpr = report_flexpriority,
10818 .cpu_has_high_real_mode_segbase = vmx_has_high_real_mode_segbase,
10820 .vcpu_create = vmx_create_vcpu,
10821 .vcpu_free = vmx_free_vcpu,
10822 .vcpu_reset = vmx_vcpu_reset,
10824 .prepare_guest_switch = vmx_save_host_state,
10825 .vcpu_load = vmx_vcpu_load,
10826 .vcpu_put = vmx_vcpu_put,
10828 .update_bp_intercept = update_exception_bitmap,
10829 .get_msr = vmx_get_msr,
10830 .set_msr = vmx_set_msr,
10831 .get_segment_base = vmx_get_segment_base,
10832 .get_segment = vmx_get_segment,
10833 .set_segment = vmx_set_segment,
10834 .get_cpl = vmx_get_cpl,
10835 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
10836 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
10837 .decache_cr3 = vmx_decache_cr3,
10838 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
10839 .set_cr0 = vmx_set_cr0,
10840 .set_cr3 = vmx_set_cr3,
10841 .set_cr4 = vmx_set_cr4,
10842 .set_efer = vmx_set_efer,
10843 .get_idt = vmx_get_idt,
10844 .set_idt = vmx_set_idt,
10845 .get_gdt = vmx_get_gdt,
10846 .set_gdt = vmx_set_gdt,
10847 .get_dr6 = vmx_get_dr6,
10848 .set_dr6 = vmx_set_dr6,
10849 .set_dr7 = vmx_set_dr7,
10850 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
10851 .cache_reg = vmx_cache_reg,
10852 .get_rflags = vmx_get_rflags,
10853 .set_rflags = vmx_set_rflags,
10854 .fpu_activate = vmx_fpu_activate,
10855 .fpu_deactivate = vmx_fpu_deactivate,
10857 .tlb_flush = vmx_flush_tlb,
10859 .run = vmx_vcpu_run,
10860 .handle_exit = vmx_handle_exit,
10861 .skip_emulated_instruction = skip_emulated_instruction,
10862 .set_interrupt_shadow = vmx_set_interrupt_shadow,
10863 .get_interrupt_shadow = vmx_get_interrupt_shadow,
10864 .patch_hypercall = vmx_patch_hypercall,
10865 .set_irq = vmx_inject_irq,
10866 .set_nmi = vmx_inject_nmi,
10867 .queue_exception = vmx_queue_exception,
10868 .cancel_injection = vmx_cancel_injection,
10869 .interrupt_allowed = vmx_interrupt_allowed,
10870 .nmi_allowed = vmx_nmi_allowed,
10871 .get_nmi_mask = vmx_get_nmi_mask,
10872 .set_nmi_mask = vmx_set_nmi_mask,
10873 .enable_nmi_window = enable_nmi_window,
10874 .enable_irq_window = enable_irq_window,
10875 .update_cr8_intercept = update_cr8_intercept,
10876 .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode,
10877 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
10878 .get_enable_apicv = vmx_get_enable_apicv,
10879 .refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
10880 .load_eoi_exitmap = vmx_load_eoi_exitmap,
10881 .hwapic_irr_update = vmx_hwapic_irr_update,
10882 .hwapic_isr_update = vmx_hwapic_isr_update,
10883 .sync_pir_to_irr = vmx_sync_pir_to_irr,
10884 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
10886 .set_tss_addr = vmx_set_tss_addr,
10887 .get_tdp_level = get_ept_level,
10888 .get_mt_mask = vmx_get_mt_mask,
10890 .get_exit_info = vmx_get_exit_info,
10892 .get_lpage_level = vmx_get_lpage_level,
10894 .cpuid_update = vmx_cpuid_update,
10896 .rdtscp_supported = vmx_rdtscp_supported,
10897 .invpcid_supported = vmx_invpcid_supported,
10899 .set_supported_cpuid = vmx_set_supported_cpuid,
10901 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
10903 .read_tsc_offset = vmx_read_tsc_offset,
10904 .write_tsc_offset = vmx_write_tsc_offset,
10905 .adjust_tsc_offset_guest = vmx_adjust_tsc_offset_guest,
10906 .read_l1_tsc = vmx_read_l1_tsc,
10908 .set_tdp_cr3 = vmx_set_cr3,
10910 .check_intercept = vmx_check_intercept,
10911 .handle_external_intr = vmx_handle_external_intr,
10912 .mpx_supported = vmx_mpx_supported,
10913 .xsaves_supported = vmx_xsaves_supported,
10915 .check_nested_events = vmx_check_nested_events,
10917 .sched_in = vmx_sched_in,
10919 .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
10920 .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
10921 .flush_log_dirty = vmx_flush_log_dirty,
10922 .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
10924 .pre_block = vmx_pre_block,
10925 .post_block = vmx_post_block,
10927 .pmu_ops = &intel_pmu_ops,
10929 .update_pi_irte = vmx_update_pi_irte,
10932 static int __init vmx_init(void)
10934 int r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
10935 __alignof__(struct vcpu_vmx), THIS_MODULE);
10939 #ifdef CONFIG_KEXEC_CORE
10940 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
10941 crash_vmclear_local_loaded_vmcss);
10947 static void __exit vmx_exit(void)
10949 #ifdef CONFIG_KEXEC_CORE
10950 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
10957 module_init(vmx_init)
10958 module_exit(vmx_exit)