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
601 enum segment_cache_field {
610 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
612 return container_of(vcpu, struct vcpu_vmx, vcpu);
615 static struct pi_desc *vcpu_to_pi_desc(struct kvm_vcpu *vcpu)
617 return &(to_vmx(vcpu)->pi_desc);
620 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
621 #define FIELD(number, name) [number] = VMCS12_OFFSET(name)
622 #define FIELD64(number, name) [number] = VMCS12_OFFSET(name), \
623 [number##_HIGH] = VMCS12_OFFSET(name)+4
626 static unsigned long shadow_read_only_fields[] = {
628 * We do NOT shadow fields that are modified when L0
629 * traps and emulates any vmx instruction (e.g. VMPTRLD,
630 * VMXON...) executed by L1.
631 * For example, VM_INSTRUCTION_ERROR is read
632 * by L1 if a vmx instruction fails (part of the error path).
633 * Note the code assumes this logic. If for some reason
634 * we start shadowing these fields then we need to
635 * force a shadow sync when L0 emulates vmx instructions
636 * (e.g. force a sync if VM_INSTRUCTION_ERROR is modified
637 * by nested_vmx_failValid)
641 VM_EXIT_INSTRUCTION_LEN,
642 IDT_VECTORING_INFO_FIELD,
643 IDT_VECTORING_ERROR_CODE,
644 VM_EXIT_INTR_ERROR_CODE,
646 GUEST_LINEAR_ADDRESS,
647 GUEST_PHYSICAL_ADDRESS
649 static int max_shadow_read_only_fields =
650 ARRAY_SIZE(shadow_read_only_fields);
652 static unsigned long shadow_read_write_fields[] = {
659 GUEST_INTERRUPTIBILITY_INFO,
672 CPU_BASED_VM_EXEC_CONTROL,
673 VM_ENTRY_EXCEPTION_ERROR_CODE,
674 VM_ENTRY_INTR_INFO_FIELD,
675 VM_ENTRY_INSTRUCTION_LEN,
676 VM_ENTRY_EXCEPTION_ERROR_CODE,
682 static int max_shadow_read_write_fields =
683 ARRAY_SIZE(shadow_read_write_fields);
685 static const unsigned short vmcs_field_to_offset_table[] = {
686 FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
687 FIELD(POSTED_INTR_NV, posted_intr_nv),
688 FIELD(GUEST_ES_SELECTOR, guest_es_selector),
689 FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
690 FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
691 FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
692 FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
693 FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
694 FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
695 FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
696 FIELD(GUEST_INTR_STATUS, guest_intr_status),
697 FIELD(HOST_ES_SELECTOR, host_es_selector),
698 FIELD(HOST_CS_SELECTOR, host_cs_selector),
699 FIELD(HOST_SS_SELECTOR, host_ss_selector),
700 FIELD(HOST_DS_SELECTOR, host_ds_selector),
701 FIELD(HOST_FS_SELECTOR, host_fs_selector),
702 FIELD(HOST_GS_SELECTOR, host_gs_selector),
703 FIELD(HOST_TR_SELECTOR, host_tr_selector),
704 FIELD64(IO_BITMAP_A, io_bitmap_a),
705 FIELD64(IO_BITMAP_B, io_bitmap_b),
706 FIELD64(MSR_BITMAP, msr_bitmap),
707 FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
708 FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
709 FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
710 FIELD64(TSC_OFFSET, tsc_offset),
711 FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
712 FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
713 FIELD64(POSTED_INTR_DESC_ADDR, posted_intr_desc_addr),
714 FIELD64(EPT_POINTER, ept_pointer),
715 FIELD64(EOI_EXIT_BITMAP0, eoi_exit_bitmap0),
716 FIELD64(EOI_EXIT_BITMAP1, eoi_exit_bitmap1),
717 FIELD64(EOI_EXIT_BITMAP2, eoi_exit_bitmap2),
718 FIELD64(EOI_EXIT_BITMAP3, eoi_exit_bitmap3),
719 FIELD64(XSS_EXIT_BITMAP, xss_exit_bitmap),
720 FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
721 FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
722 FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
723 FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
724 FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
725 FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
726 FIELD64(GUEST_PDPTR0, guest_pdptr0),
727 FIELD64(GUEST_PDPTR1, guest_pdptr1),
728 FIELD64(GUEST_PDPTR2, guest_pdptr2),
729 FIELD64(GUEST_PDPTR3, guest_pdptr3),
730 FIELD64(GUEST_BNDCFGS, guest_bndcfgs),
731 FIELD64(HOST_IA32_PAT, host_ia32_pat),
732 FIELD64(HOST_IA32_EFER, host_ia32_efer),
733 FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
734 FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
735 FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
736 FIELD(EXCEPTION_BITMAP, exception_bitmap),
737 FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
738 FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
739 FIELD(CR3_TARGET_COUNT, cr3_target_count),
740 FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
741 FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
742 FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
743 FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
744 FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
745 FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
746 FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
747 FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
748 FIELD(TPR_THRESHOLD, tpr_threshold),
749 FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
750 FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
751 FIELD(VM_EXIT_REASON, vm_exit_reason),
752 FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
753 FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
754 FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
755 FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
756 FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
757 FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
758 FIELD(GUEST_ES_LIMIT, guest_es_limit),
759 FIELD(GUEST_CS_LIMIT, guest_cs_limit),
760 FIELD(GUEST_SS_LIMIT, guest_ss_limit),
761 FIELD(GUEST_DS_LIMIT, guest_ds_limit),
762 FIELD(GUEST_FS_LIMIT, guest_fs_limit),
763 FIELD(GUEST_GS_LIMIT, guest_gs_limit),
764 FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
765 FIELD(GUEST_TR_LIMIT, guest_tr_limit),
766 FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
767 FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
768 FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
769 FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
770 FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
771 FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
772 FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
773 FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
774 FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
775 FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
776 FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
777 FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
778 FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
779 FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
780 FIELD(VMX_PREEMPTION_TIMER_VALUE, vmx_preemption_timer_value),
781 FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
782 FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
783 FIELD(CR0_READ_SHADOW, cr0_read_shadow),
784 FIELD(CR4_READ_SHADOW, cr4_read_shadow),
785 FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
786 FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
787 FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
788 FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
789 FIELD(EXIT_QUALIFICATION, exit_qualification),
790 FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
791 FIELD(GUEST_CR0, guest_cr0),
792 FIELD(GUEST_CR3, guest_cr3),
793 FIELD(GUEST_CR4, guest_cr4),
794 FIELD(GUEST_ES_BASE, guest_es_base),
795 FIELD(GUEST_CS_BASE, guest_cs_base),
796 FIELD(GUEST_SS_BASE, guest_ss_base),
797 FIELD(GUEST_DS_BASE, guest_ds_base),
798 FIELD(GUEST_FS_BASE, guest_fs_base),
799 FIELD(GUEST_GS_BASE, guest_gs_base),
800 FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
801 FIELD(GUEST_TR_BASE, guest_tr_base),
802 FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
803 FIELD(GUEST_IDTR_BASE, guest_idtr_base),
804 FIELD(GUEST_DR7, guest_dr7),
805 FIELD(GUEST_RSP, guest_rsp),
806 FIELD(GUEST_RIP, guest_rip),
807 FIELD(GUEST_RFLAGS, guest_rflags),
808 FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
809 FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
810 FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
811 FIELD(HOST_CR0, host_cr0),
812 FIELD(HOST_CR3, host_cr3),
813 FIELD(HOST_CR4, host_cr4),
814 FIELD(HOST_FS_BASE, host_fs_base),
815 FIELD(HOST_GS_BASE, host_gs_base),
816 FIELD(HOST_TR_BASE, host_tr_base),
817 FIELD(HOST_GDTR_BASE, host_gdtr_base),
818 FIELD(HOST_IDTR_BASE, host_idtr_base),
819 FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
820 FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
821 FIELD(HOST_RSP, host_rsp),
822 FIELD(HOST_RIP, host_rip),
825 static inline short vmcs_field_to_offset(unsigned long field)
827 BUILD_BUG_ON(ARRAY_SIZE(vmcs_field_to_offset_table) > SHRT_MAX);
829 if (field >= ARRAY_SIZE(vmcs_field_to_offset_table) ||
830 vmcs_field_to_offset_table[field] == 0)
833 return vmcs_field_to_offset_table[field];
836 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
838 return to_vmx(vcpu)->nested.current_vmcs12;
841 static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
843 struct page *page = kvm_vcpu_gfn_to_page(vcpu, addr >> PAGE_SHIFT);
844 if (is_error_page(page))
850 static void nested_release_page(struct page *page)
852 kvm_release_page_dirty(page);
855 static void nested_release_page_clean(struct page *page)
857 kvm_release_page_clean(page);
860 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu);
861 static u64 construct_eptp(unsigned long root_hpa);
862 static void kvm_cpu_vmxon(u64 addr);
863 static void kvm_cpu_vmxoff(void);
864 static bool vmx_mpx_supported(void);
865 static bool vmx_xsaves_supported(void);
866 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
867 static void vmx_set_segment(struct kvm_vcpu *vcpu,
868 struct kvm_segment *var, int seg);
869 static void vmx_get_segment(struct kvm_vcpu *vcpu,
870 struct kvm_segment *var, int seg);
871 static bool guest_state_valid(struct kvm_vcpu *vcpu);
872 static u32 vmx_segment_access_rights(struct kvm_segment *var);
873 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx);
874 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx);
875 static int alloc_identity_pagetable(struct kvm *kvm);
877 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
878 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
880 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
881 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
883 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
884 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
887 * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
888 * can find which vCPU should be waken up.
890 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
891 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
893 static unsigned long *vmx_io_bitmap_a;
894 static unsigned long *vmx_io_bitmap_b;
895 static unsigned long *vmx_msr_bitmap_legacy;
896 static unsigned long *vmx_msr_bitmap_longmode;
897 static unsigned long *vmx_msr_bitmap_legacy_x2apic;
898 static unsigned long *vmx_msr_bitmap_longmode_x2apic;
899 static unsigned long *vmx_msr_bitmap_nested;
900 static unsigned long *vmx_vmread_bitmap;
901 static unsigned long *vmx_vmwrite_bitmap;
903 static bool cpu_has_load_ia32_efer;
904 static bool cpu_has_load_perf_global_ctrl;
906 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
907 static DEFINE_SPINLOCK(vmx_vpid_lock);
909 static struct vmcs_config {
913 u32 pin_based_exec_ctrl;
914 u32 cpu_based_exec_ctrl;
915 u32 cpu_based_2nd_exec_ctrl;
920 static struct vmx_capability {
925 #define VMX_SEGMENT_FIELD(seg) \
926 [VCPU_SREG_##seg] = { \
927 .selector = GUEST_##seg##_SELECTOR, \
928 .base = GUEST_##seg##_BASE, \
929 .limit = GUEST_##seg##_LIMIT, \
930 .ar_bytes = GUEST_##seg##_AR_BYTES, \
933 static const struct kvm_vmx_segment_field {
938 } kvm_vmx_segment_fields[] = {
939 VMX_SEGMENT_FIELD(CS),
940 VMX_SEGMENT_FIELD(DS),
941 VMX_SEGMENT_FIELD(ES),
942 VMX_SEGMENT_FIELD(FS),
943 VMX_SEGMENT_FIELD(GS),
944 VMX_SEGMENT_FIELD(SS),
945 VMX_SEGMENT_FIELD(TR),
946 VMX_SEGMENT_FIELD(LDTR),
949 static u64 host_efer;
951 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
954 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
955 * away by decrementing the array size.
957 static const u32 vmx_msr_index[] = {
959 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
961 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
964 static inline bool is_exception_n(u32 intr_info, u8 vector)
966 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
967 INTR_INFO_VALID_MASK)) ==
968 (INTR_TYPE_HARD_EXCEPTION | vector | INTR_INFO_VALID_MASK);
971 static inline bool is_debug(u32 intr_info)
973 return is_exception_n(intr_info, DB_VECTOR);
976 static inline bool is_breakpoint(u32 intr_info)
978 return is_exception_n(intr_info, BP_VECTOR);
981 static inline bool is_page_fault(u32 intr_info)
983 return is_exception_n(intr_info, PF_VECTOR);
986 static inline bool is_no_device(u32 intr_info)
988 return is_exception_n(intr_info, NM_VECTOR);
991 static inline bool is_invalid_opcode(u32 intr_info)
993 return is_exception_n(intr_info, UD_VECTOR);
996 static inline bool is_external_interrupt(u32 intr_info)
998 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
999 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
1002 static inline bool is_machine_check(u32 intr_info)
1004 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
1005 INTR_INFO_VALID_MASK)) ==
1006 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
1009 static inline bool cpu_has_vmx_msr_bitmap(void)
1011 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
1014 static inline bool cpu_has_vmx_tpr_shadow(void)
1016 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
1019 static inline bool cpu_need_tpr_shadow(struct kvm_vcpu *vcpu)
1021 return cpu_has_vmx_tpr_shadow() && lapic_in_kernel(vcpu);
1024 static inline bool cpu_has_secondary_exec_ctrls(void)
1026 return vmcs_config.cpu_based_exec_ctrl &
1027 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1030 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
1032 return vmcs_config.cpu_based_2nd_exec_ctrl &
1033 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1036 static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
1038 return vmcs_config.cpu_based_2nd_exec_ctrl &
1039 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
1042 static inline bool cpu_has_vmx_apic_register_virt(void)
1044 return vmcs_config.cpu_based_2nd_exec_ctrl &
1045 SECONDARY_EXEC_APIC_REGISTER_VIRT;
1048 static inline bool cpu_has_vmx_virtual_intr_delivery(void)
1050 return vmcs_config.cpu_based_2nd_exec_ctrl &
1051 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
1054 static inline bool cpu_has_vmx_posted_intr(void)
1056 return IS_ENABLED(CONFIG_X86_LOCAL_APIC) &&
1057 vmcs_config.pin_based_exec_ctrl & PIN_BASED_POSTED_INTR;
1060 static inline bool cpu_has_vmx_apicv(void)
1062 return cpu_has_vmx_apic_register_virt() &&
1063 cpu_has_vmx_virtual_intr_delivery() &&
1064 cpu_has_vmx_posted_intr();
1067 static inline bool cpu_has_vmx_flexpriority(void)
1069 return cpu_has_vmx_tpr_shadow() &&
1070 cpu_has_vmx_virtualize_apic_accesses();
1073 static inline bool cpu_has_vmx_ept_execute_only(void)
1075 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
1078 static inline bool cpu_has_vmx_ept_2m_page(void)
1080 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
1083 static inline bool cpu_has_vmx_ept_1g_page(void)
1085 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
1088 static inline bool cpu_has_vmx_ept_4levels(void)
1090 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
1093 static inline bool cpu_has_vmx_ept_ad_bits(void)
1095 return vmx_capability.ept & VMX_EPT_AD_BIT;
1098 static inline bool cpu_has_vmx_invept_context(void)
1100 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
1103 static inline bool cpu_has_vmx_invept_global(void)
1105 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
1108 static inline bool cpu_has_vmx_invvpid_single(void)
1110 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
1113 static inline bool cpu_has_vmx_invvpid_global(void)
1115 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
1118 static inline bool cpu_has_vmx_ept(void)
1120 return vmcs_config.cpu_based_2nd_exec_ctrl &
1121 SECONDARY_EXEC_ENABLE_EPT;
1124 static inline bool cpu_has_vmx_unrestricted_guest(void)
1126 return vmcs_config.cpu_based_2nd_exec_ctrl &
1127 SECONDARY_EXEC_UNRESTRICTED_GUEST;
1130 static inline bool cpu_has_vmx_ple(void)
1132 return vmcs_config.cpu_based_2nd_exec_ctrl &
1133 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
1136 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
1138 return flexpriority_enabled && lapic_in_kernel(vcpu);
1141 static inline bool cpu_has_vmx_vpid(void)
1143 return vmcs_config.cpu_based_2nd_exec_ctrl &
1144 SECONDARY_EXEC_ENABLE_VPID;
1147 static inline bool cpu_has_vmx_rdtscp(void)
1149 return vmcs_config.cpu_based_2nd_exec_ctrl &
1150 SECONDARY_EXEC_RDTSCP;
1153 static inline bool cpu_has_vmx_invpcid(void)
1155 return vmcs_config.cpu_based_2nd_exec_ctrl &
1156 SECONDARY_EXEC_ENABLE_INVPCID;
1159 static inline bool cpu_has_virtual_nmis(void)
1161 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
1164 static inline bool cpu_has_vmx_wbinvd_exit(void)
1166 return vmcs_config.cpu_based_2nd_exec_ctrl &
1167 SECONDARY_EXEC_WBINVD_EXITING;
1170 static inline bool cpu_has_vmx_shadow_vmcs(void)
1173 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
1174 /* check if the cpu supports writing r/o exit information fields */
1175 if (!(vmx_msr & MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS))
1178 return vmcs_config.cpu_based_2nd_exec_ctrl &
1179 SECONDARY_EXEC_SHADOW_VMCS;
1182 static inline bool cpu_has_vmx_pml(void)
1184 return vmcs_config.cpu_based_2nd_exec_ctrl & SECONDARY_EXEC_ENABLE_PML;
1187 static inline bool cpu_has_vmx_tsc_scaling(void)
1189 return vmcs_config.cpu_based_2nd_exec_ctrl &
1190 SECONDARY_EXEC_TSC_SCALING;
1193 static inline bool report_flexpriority(void)
1195 return flexpriority_enabled;
1198 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
1200 return vmcs12->cpu_based_vm_exec_control & bit;
1203 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
1205 return (vmcs12->cpu_based_vm_exec_control &
1206 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
1207 (vmcs12->secondary_vm_exec_control & bit);
1210 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12)
1212 return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
1215 static inline bool nested_cpu_has_preemption_timer(struct vmcs12 *vmcs12)
1217 return vmcs12->pin_based_vm_exec_control &
1218 PIN_BASED_VMX_PREEMPTION_TIMER;
1221 static inline int nested_cpu_has_ept(struct vmcs12 *vmcs12)
1223 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_EPT);
1226 static inline bool nested_cpu_has_xsaves(struct vmcs12 *vmcs12)
1228 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES) &&
1229 vmx_xsaves_supported();
1232 static inline bool nested_cpu_has_virt_x2apic_mode(struct vmcs12 *vmcs12)
1234 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
1237 static inline bool nested_cpu_has_vpid(struct vmcs12 *vmcs12)
1239 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_VPID);
1242 static inline bool nested_cpu_has_apic_reg_virt(struct vmcs12 *vmcs12)
1244 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_APIC_REGISTER_VIRT);
1247 static inline bool nested_cpu_has_vid(struct vmcs12 *vmcs12)
1249 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
1252 static inline bool nested_cpu_has_posted_intr(struct vmcs12 *vmcs12)
1254 return vmcs12->pin_based_vm_exec_control & PIN_BASED_POSTED_INTR;
1257 static inline bool is_exception(u32 intr_info)
1259 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1260 == (INTR_TYPE_HARD_EXCEPTION | INTR_INFO_VALID_MASK);
1263 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
1265 unsigned long exit_qualification);
1266 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
1267 struct vmcs12 *vmcs12,
1268 u32 reason, unsigned long qualification);
1270 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
1274 for (i = 0; i < vmx->nmsrs; ++i)
1275 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
1280 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
1286 } operand = { vpid, 0, gva };
1288 asm volatile (__ex(ASM_VMX_INVVPID)
1289 /* CF==1 or ZF==1 --> rc = -1 */
1290 "; ja 1f ; ud2 ; 1:"
1291 : : "a"(&operand), "c"(ext) : "cc", "memory");
1294 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
1298 } operand = {eptp, gpa};
1300 asm volatile (__ex(ASM_VMX_INVEPT)
1301 /* CF==1 or ZF==1 --> rc = -1 */
1302 "; ja 1f ; ud2 ; 1:\n"
1303 : : "a" (&operand), "c" (ext) : "cc", "memory");
1306 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
1310 i = __find_msr_index(vmx, msr);
1312 return &vmx->guest_msrs[i];
1316 static void vmcs_clear(struct vmcs *vmcs)
1318 u64 phys_addr = __pa(vmcs);
1321 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
1322 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1325 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
1329 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
1331 vmcs_clear(loaded_vmcs->vmcs);
1332 loaded_vmcs->cpu = -1;
1333 loaded_vmcs->launched = 0;
1336 static void vmcs_load(struct vmcs *vmcs)
1338 u64 phys_addr = __pa(vmcs);
1341 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
1342 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1345 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
1349 #ifdef CONFIG_KEXEC_CORE
1351 * This bitmap is used to indicate whether the vmclear
1352 * operation is enabled on all cpus. All disabled by
1355 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
1357 static inline void crash_enable_local_vmclear(int cpu)
1359 cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
1362 static inline void crash_disable_local_vmclear(int cpu)
1364 cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
1367 static inline int crash_local_vmclear_enabled(int cpu)
1369 return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
1372 static void crash_vmclear_local_loaded_vmcss(void)
1374 int cpu = raw_smp_processor_id();
1375 struct loaded_vmcs *v;
1377 if (!crash_local_vmclear_enabled(cpu))
1380 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
1381 loaded_vmcss_on_cpu_link)
1382 vmcs_clear(v->vmcs);
1385 static inline void crash_enable_local_vmclear(int cpu) { }
1386 static inline void crash_disable_local_vmclear(int cpu) { }
1387 #endif /* CONFIG_KEXEC_CORE */
1389 static void __loaded_vmcs_clear(void *arg)
1391 struct loaded_vmcs *loaded_vmcs = arg;
1392 int cpu = raw_smp_processor_id();
1394 if (loaded_vmcs->cpu != cpu)
1395 return; /* vcpu migration can race with cpu offline */
1396 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1397 per_cpu(current_vmcs, cpu) = NULL;
1398 crash_disable_local_vmclear(cpu);
1399 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1402 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1403 * is before setting loaded_vmcs->vcpu to -1 which is done in
1404 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1405 * then adds the vmcs into percpu list before it is deleted.
1409 loaded_vmcs_init(loaded_vmcs);
1410 crash_enable_local_vmclear(cpu);
1413 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1415 int cpu = loaded_vmcs->cpu;
1418 smp_call_function_single(cpu,
1419 __loaded_vmcs_clear, loaded_vmcs, 1);
1422 static inline void vpid_sync_vcpu_single(int vpid)
1427 if (cpu_has_vmx_invvpid_single())
1428 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vpid, 0);
1431 static inline void vpid_sync_vcpu_global(void)
1433 if (cpu_has_vmx_invvpid_global())
1434 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1437 static inline void vpid_sync_context(int vpid)
1439 if (cpu_has_vmx_invvpid_single())
1440 vpid_sync_vcpu_single(vpid);
1442 vpid_sync_vcpu_global();
1445 static inline void ept_sync_global(void)
1447 if (cpu_has_vmx_invept_global())
1448 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1451 static inline void ept_sync_context(u64 eptp)
1454 if (cpu_has_vmx_invept_context())
1455 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1461 static __always_inline void vmcs_check16(unsigned long field)
1463 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1464 "16-bit accessor invalid for 64-bit field");
1465 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1466 "16-bit accessor invalid for 64-bit high field");
1467 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1468 "16-bit accessor invalid for 32-bit high field");
1469 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1470 "16-bit accessor invalid for natural width field");
1473 static __always_inline void vmcs_check32(unsigned long field)
1475 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1476 "32-bit accessor invalid for 16-bit field");
1477 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1478 "32-bit accessor invalid for natural width field");
1481 static __always_inline void vmcs_check64(unsigned long field)
1483 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1484 "64-bit accessor invalid for 16-bit field");
1485 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1486 "64-bit accessor invalid for 64-bit high field");
1487 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1488 "64-bit accessor invalid for 32-bit field");
1489 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1490 "64-bit accessor invalid for natural width field");
1493 static __always_inline void vmcs_checkl(unsigned long field)
1495 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1496 "Natural width accessor invalid for 16-bit field");
1497 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1498 "Natural width accessor invalid for 64-bit field");
1499 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1500 "Natural width accessor invalid for 64-bit high field");
1501 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1502 "Natural width accessor invalid for 32-bit field");
1505 static __always_inline unsigned long __vmcs_readl(unsigned long field)
1507 unsigned long value;
1509 asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1510 : "=a"(value) : "d"(field) : "cc");
1514 static __always_inline u16 vmcs_read16(unsigned long field)
1516 vmcs_check16(field);
1517 return __vmcs_readl(field);
1520 static __always_inline u32 vmcs_read32(unsigned long field)
1522 vmcs_check32(field);
1523 return __vmcs_readl(field);
1526 static __always_inline u64 vmcs_read64(unsigned long field)
1528 vmcs_check64(field);
1529 #ifdef CONFIG_X86_64
1530 return __vmcs_readl(field);
1532 return __vmcs_readl(field) | ((u64)__vmcs_readl(field+1) << 32);
1536 static __always_inline unsigned long vmcs_readl(unsigned long field)
1539 return __vmcs_readl(field);
1542 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1544 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1545 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1549 static __always_inline void __vmcs_writel(unsigned long field, unsigned long value)
1553 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1554 : "=q"(error) : "a"(value), "d"(field) : "cc");
1555 if (unlikely(error))
1556 vmwrite_error(field, value);
1559 static __always_inline void vmcs_write16(unsigned long field, u16 value)
1561 vmcs_check16(field);
1562 __vmcs_writel(field, value);
1565 static __always_inline void vmcs_write32(unsigned long field, u32 value)
1567 vmcs_check32(field);
1568 __vmcs_writel(field, value);
1571 static __always_inline void vmcs_write64(unsigned long field, u64 value)
1573 vmcs_check64(field);
1574 __vmcs_writel(field, value);
1575 #ifndef CONFIG_X86_64
1577 __vmcs_writel(field+1, value >> 32);
1581 static __always_inline void vmcs_writel(unsigned long field, unsigned long value)
1584 __vmcs_writel(field, value);
1587 static __always_inline void vmcs_clear_bits(unsigned long field, u32 mask)
1589 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1590 "vmcs_clear_bits does not support 64-bit fields");
1591 __vmcs_writel(field, __vmcs_readl(field) & ~mask);
1594 static __always_inline void vmcs_set_bits(unsigned long field, u32 mask)
1596 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1597 "vmcs_set_bits does not support 64-bit fields");
1598 __vmcs_writel(field, __vmcs_readl(field) | mask);
1601 static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val)
1603 vmcs_write32(VM_ENTRY_CONTROLS, val);
1604 vmx->vm_entry_controls_shadow = val;
1607 static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val)
1609 if (vmx->vm_entry_controls_shadow != val)
1610 vm_entry_controls_init(vmx, val);
1613 static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx)
1615 return vmx->vm_entry_controls_shadow;
1619 static inline void vm_entry_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1621 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) | val);
1624 static inline void vm_entry_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1626 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) & ~val);
1629 static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val)
1631 vmcs_write32(VM_EXIT_CONTROLS, val);
1632 vmx->vm_exit_controls_shadow = val;
1635 static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val)
1637 if (vmx->vm_exit_controls_shadow != val)
1638 vm_exit_controls_init(vmx, val);
1641 static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx)
1643 return vmx->vm_exit_controls_shadow;
1647 static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1649 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val);
1652 static inline void vm_exit_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1654 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) & ~val);
1657 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1659 vmx->segment_cache.bitmask = 0;
1662 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1666 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1668 if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1669 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1670 vmx->segment_cache.bitmask = 0;
1672 ret = vmx->segment_cache.bitmask & mask;
1673 vmx->segment_cache.bitmask |= mask;
1677 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1679 u16 *p = &vmx->segment_cache.seg[seg].selector;
1681 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1682 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1686 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1688 ulong *p = &vmx->segment_cache.seg[seg].base;
1690 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1691 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1695 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1697 u32 *p = &vmx->segment_cache.seg[seg].limit;
1699 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1700 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1704 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1706 u32 *p = &vmx->segment_cache.seg[seg].ar;
1708 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1709 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1713 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1717 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1718 (1u << NM_VECTOR) | (1u << DB_VECTOR) | (1u << AC_VECTOR);
1719 if ((vcpu->guest_debug &
1720 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1721 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1722 eb |= 1u << BP_VECTOR;
1723 if (to_vmx(vcpu)->rmode.vm86_active)
1726 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1727 if (vcpu->fpu_active)
1728 eb &= ~(1u << NM_VECTOR);
1730 /* When we are running a nested L2 guest and L1 specified for it a
1731 * certain exception bitmap, we must trap the same exceptions and pass
1732 * them to L1. When running L2, we will only handle the exceptions
1733 * specified above if L1 did not want them.
1735 if (is_guest_mode(vcpu))
1736 eb |= get_vmcs12(vcpu)->exception_bitmap;
1738 vmcs_write32(EXCEPTION_BITMAP, eb);
1741 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1742 unsigned long entry, unsigned long exit)
1744 vm_entry_controls_clearbit(vmx, entry);
1745 vm_exit_controls_clearbit(vmx, exit);
1748 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1751 struct msr_autoload *m = &vmx->msr_autoload;
1755 if (cpu_has_load_ia32_efer) {
1756 clear_atomic_switch_msr_special(vmx,
1757 VM_ENTRY_LOAD_IA32_EFER,
1758 VM_EXIT_LOAD_IA32_EFER);
1762 case MSR_CORE_PERF_GLOBAL_CTRL:
1763 if (cpu_has_load_perf_global_ctrl) {
1764 clear_atomic_switch_msr_special(vmx,
1765 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1766 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1772 for (i = 0; i < m->nr; ++i)
1773 if (m->guest[i].index == msr)
1779 m->guest[i] = m->guest[m->nr];
1780 m->host[i] = m->host[m->nr];
1781 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1782 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1785 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1786 unsigned long entry, unsigned long exit,
1787 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
1788 u64 guest_val, u64 host_val)
1790 vmcs_write64(guest_val_vmcs, guest_val);
1791 vmcs_write64(host_val_vmcs, host_val);
1792 vm_entry_controls_setbit(vmx, entry);
1793 vm_exit_controls_setbit(vmx, exit);
1796 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1797 u64 guest_val, u64 host_val)
1800 struct msr_autoload *m = &vmx->msr_autoload;
1804 if (cpu_has_load_ia32_efer) {
1805 add_atomic_switch_msr_special(vmx,
1806 VM_ENTRY_LOAD_IA32_EFER,
1807 VM_EXIT_LOAD_IA32_EFER,
1810 guest_val, host_val);
1814 case MSR_CORE_PERF_GLOBAL_CTRL:
1815 if (cpu_has_load_perf_global_ctrl) {
1816 add_atomic_switch_msr_special(vmx,
1817 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1818 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1819 GUEST_IA32_PERF_GLOBAL_CTRL,
1820 HOST_IA32_PERF_GLOBAL_CTRL,
1821 guest_val, host_val);
1827 for (i = 0; i < m->nr; ++i)
1828 if (m->guest[i].index == msr)
1831 if (i == NR_AUTOLOAD_MSRS) {
1832 printk_once(KERN_WARNING "Not enough msr switch entries. "
1833 "Can't add msr %x\n", msr);
1835 } else if (i == m->nr) {
1837 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1838 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1841 m->guest[i].index = msr;
1842 m->guest[i].value = guest_val;
1843 m->host[i].index = msr;
1844 m->host[i].value = host_val;
1847 static void reload_tss(void)
1850 * VT restores TR but not its size. Useless.
1852 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
1853 struct desc_struct *descs;
1855 descs = (void *)gdt->address;
1856 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
1860 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
1865 guest_efer = vmx->vcpu.arch.efer;
1868 * NX is emulated; LMA and LME handled by hardware; SCE meaningless
1871 ignore_bits = EFER_NX | EFER_SCE;
1872 #ifdef CONFIG_X86_64
1873 ignore_bits |= EFER_LMA | EFER_LME;
1874 /* SCE is meaningful only in long mode on Intel */
1875 if (guest_efer & EFER_LMA)
1876 ignore_bits &= ~(u64)EFER_SCE;
1878 guest_efer &= ~ignore_bits;
1879 guest_efer |= host_efer & ignore_bits;
1880 vmx->guest_msrs[efer_offset].data = guest_efer;
1881 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1883 clear_atomic_switch_msr(vmx, MSR_EFER);
1886 * On EPT, we can't emulate NX, so we must switch EFER atomically.
1887 * On CPUs that support "load IA32_EFER", always switch EFER
1888 * atomically, since it's faster than switching it manually.
1890 if (cpu_has_load_ia32_efer ||
1891 (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
1892 guest_efer = vmx->vcpu.arch.efer;
1893 if (!(guest_efer & EFER_LMA))
1894 guest_efer &= ~EFER_LME;
1895 if (guest_efer != host_efer)
1896 add_atomic_switch_msr(vmx, MSR_EFER,
1897 guest_efer, host_efer);
1904 static unsigned long segment_base(u16 selector)
1906 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
1907 struct desc_struct *d;
1908 unsigned long table_base;
1911 if (!(selector & ~3))
1914 table_base = gdt->address;
1916 if (selector & 4) { /* from ldt */
1917 u16 ldt_selector = kvm_read_ldt();
1919 if (!(ldt_selector & ~3))
1922 table_base = segment_base(ldt_selector);
1924 d = (struct desc_struct *)(table_base + (selector & ~7));
1925 v = get_desc_base(d);
1926 #ifdef CONFIG_X86_64
1927 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
1928 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
1933 static inline unsigned long kvm_read_tr_base(void)
1936 asm("str %0" : "=g"(tr));
1937 return segment_base(tr);
1940 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
1942 struct vcpu_vmx *vmx = to_vmx(vcpu);
1945 if (vmx->host_state.loaded)
1948 vmx->host_state.loaded = 1;
1950 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
1951 * allow segment selectors with cpl > 0 or ti == 1.
1953 vmx->host_state.ldt_sel = kvm_read_ldt();
1954 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
1955 savesegment(fs, vmx->host_state.fs_sel);
1956 if (!(vmx->host_state.fs_sel & 7)) {
1957 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
1958 vmx->host_state.fs_reload_needed = 0;
1960 vmcs_write16(HOST_FS_SELECTOR, 0);
1961 vmx->host_state.fs_reload_needed = 1;
1963 savesegment(gs, vmx->host_state.gs_sel);
1964 if (!(vmx->host_state.gs_sel & 7))
1965 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
1967 vmcs_write16(HOST_GS_SELECTOR, 0);
1968 vmx->host_state.gs_ldt_reload_needed = 1;
1971 #ifdef CONFIG_X86_64
1972 savesegment(ds, vmx->host_state.ds_sel);
1973 savesegment(es, vmx->host_state.es_sel);
1976 #ifdef CONFIG_X86_64
1977 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
1978 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
1980 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
1981 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
1984 #ifdef CONFIG_X86_64
1985 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1986 if (is_long_mode(&vmx->vcpu))
1987 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1989 if (boot_cpu_has(X86_FEATURE_MPX))
1990 rdmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
1991 for (i = 0; i < vmx->save_nmsrs; ++i)
1992 kvm_set_shared_msr(vmx->guest_msrs[i].index,
1993 vmx->guest_msrs[i].data,
1994 vmx->guest_msrs[i].mask);
1997 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
1999 if (!vmx->host_state.loaded)
2002 ++vmx->vcpu.stat.host_state_reload;
2003 vmx->host_state.loaded = 0;
2004 #ifdef CONFIG_X86_64
2005 if (is_long_mode(&vmx->vcpu))
2006 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2008 if (vmx->host_state.gs_ldt_reload_needed) {
2009 kvm_load_ldt(vmx->host_state.ldt_sel);
2010 #ifdef CONFIG_X86_64
2011 load_gs_index(vmx->host_state.gs_sel);
2013 loadsegment(gs, vmx->host_state.gs_sel);
2016 if (vmx->host_state.fs_reload_needed)
2017 loadsegment(fs, vmx->host_state.fs_sel);
2018 #ifdef CONFIG_X86_64
2019 if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
2020 loadsegment(ds, vmx->host_state.ds_sel);
2021 loadsegment(es, vmx->host_state.es_sel);
2025 #ifdef CONFIG_X86_64
2026 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2028 if (vmx->host_state.msr_host_bndcfgs)
2029 wrmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
2031 * If the FPU is not active (through the host task or
2032 * the guest vcpu), then restore the cr0.TS bit.
2034 if (!fpregs_active() && !vmx->vcpu.guest_fpu_loaded)
2036 load_gdt(this_cpu_ptr(&host_gdt));
2039 static void vmx_load_host_state(struct vcpu_vmx *vmx)
2042 __vmx_load_host_state(vmx);
2046 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
2048 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2049 struct pi_desc old, new;
2052 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
2053 !irq_remapping_cap(IRQ_POSTING_CAP))
2057 old.control = new.control = pi_desc->control;
2060 * If 'nv' field is POSTED_INTR_WAKEUP_VECTOR, there
2061 * are two possible cases:
2062 * 1. After running 'pre_block', context switch
2063 * happened. For this case, 'sn' was set in
2064 * vmx_vcpu_put(), so we need to clear it here.
2065 * 2. After running 'pre_block', we were blocked,
2066 * and woken up by some other guy. For this case,
2067 * we don't need to do anything, 'pi_post_block'
2068 * will do everything for us. However, we cannot
2069 * check whether it is case #1 or case #2 here
2070 * (maybe, not needed), so we also clear sn here,
2071 * I think it is not a big deal.
2073 if (pi_desc->nv != POSTED_INTR_WAKEUP_VECTOR) {
2074 if (vcpu->cpu != cpu) {
2075 dest = cpu_physical_id(cpu);
2077 if (x2apic_enabled())
2080 new.ndst = (dest << 8) & 0xFF00;
2083 /* set 'NV' to 'notification vector' */
2084 new.nv = POSTED_INTR_VECTOR;
2087 /* Allow posting non-urgent interrupts */
2089 } while (cmpxchg(&pi_desc->control, old.control,
2090 new.control) != old.control);
2093 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
2094 * vcpu mutex is already taken.
2096 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2098 struct vcpu_vmx *vmx = to_vmx(vcpu);
2099 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2102 kvm_cpu_vmxon(phys_addr);
2103 else if (vmx->loaded_vmcs->cpu != cpu)
2104 loaded_vmcs_clear(vmx->loaded_vmcs);
2106 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
2107 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
2108 vmcs_load(vmx->loaded_vmcs->vmcs);
2111 if (vmx->loaded_vmcs->cpu != cpu) {
2112 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
2113 unsigned long sysenter_esp;
2115 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2116 local_irq_disable();
2117 crash_disable_local_vmclear(cpu);
2120 * Read loaded_vmcs->cpu should be before fetching
2121 * loaded_vmcs->loaded_vmcss_on_cpu_link.
2122 * See the comments in __loaded_vmcs_clear().
2126 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
2127 &per_cpu(loaded_vmcss_on_cpu, cpu));
2128 crash_enable_local_vmclear(cpu);
2132 * Linux uses per-cpu TSS and GDT, so set these when switching
2135 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
2136 vmcs_writel(HOST_GDTR_BASE, gdt->address); /* 22.2.4 */
2138 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
2139 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
2141 /* Setup TSC multiplier */
2142 if (cpu_has_vmx_tsc_scaling())
2143 vmcs_write64(TSC_MULTIPLIER,
2144 vcpu->arch.tsc_scaling_ratio);
2146 vmx->loaded_vmcs->cpu = cpu;
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 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
5623 CPU_BASED_MOV_DR_EXITING);
5626 * No more DR vmexits; force a reload of the debug registers
5627 * and reenter on this instruction. The next vmexit will
5628 * retrieve the full state of the debug registers.
5630 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
5634 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
5635 if (exit_qualification & TYPE_MOV_FROM_DR) {
5638 if (kvm_get_dr(vcpu, dr, &val))
5640 kvm_register_write(vcpu, reg, val);
5642 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
5645 skip_emulated_instruction(vcpu);
5649 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
5651 return vcpu->arch.dr6;
5654 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
5658 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
5660 get_debugreg(vcpu->arch.db[0], 0);
5661 get_debugreg(vcpu->arch.db[1], 1);
5662 get_debugreg(vcpu->arch.db[2], 2);
5663 get_debugreg(vcpu->arch.db[3], 3);
5664 get_debugreg(vcpu->arch.dr6, 6);
5665 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
5667 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
5668 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL, CPU_BASED_MOV_DR_EXITING);
5671 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
5673 vmcs_writel(GUEST_DR7, val);
5676 static int handle_cpuid(struct kvm_vcpu *vcpu)
5678 kvm_emulate_cpuid(vcpu);
5682 static int handle_rdmsr(struct kvm_vcpu *vcpu)
5684 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5685 struct msr_data msr_info;
5687 msr_info.index = ecx;
5688 msr_info.host_initiated = false;
5689 if (vmx_get_msr(vcpu, &msr_info)) {
5690 trace_kvm_msr_read_ex(ecx);
5691 kvm_inject_gp(vcpu, 0);
5695 trace_kvm_msr_read(ecx, msr_info.data);
5697 /* FIXME: handling of bits 32:63 of rax, rdx */
5698 vcpu->arch.regs[VCPU_REGS_RAX] = msr_info.data & -1u;
5699 vcpu->arch.regs[VCPU_REGS_RDX] = (msr_info.data >> 32) & -1u;
5700 skip_emulated_instruction(vcpu);
5704 static int handle_wrmsr(struct kvm_vcpu *vcpu)
5706 struct msr_data msr;
5707 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5708 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
5709 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
5713 msr.host_initiated = false;
5714 if (kvm_set_msr(vcpu, &msr) != 0) {
5715 trace_kvm_msr_write_ex(ecx, data);
5716 kvm_inject_gp(vcpu, 0);
5720 trace_kvm_msr_write(ecx, data);
5721 skip_emulated_instruction(vcpu);
5725 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
5727 kvm_make_request(KVM_REQ_EVENT, vcpu);
5731 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
5733 u32 cpu_based_vm_exec_control;
5735 /* clear pending irq */
5736 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5737 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
5738 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5740 kvm_make_request(KVM_REQ_EVENT, vcpu);
5742 ++vcpu->stat.irq_window_exits;
5746 static int handle_halt(struct kvm_vcpu *vcpu)
5748 return kvm_emulate_halt(vcpu);
5751 static int handle_vmcall(struct kvm_vcpu *vcpu)
5753 return kvm_emulate_hypercall(vcpu);
5756 static int handle_invd(struct kvm_vcpu *vcpu)
5758 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5761 static int handle_invlpg(struct kvm_vcpu *vcpu)
5763 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5765 kvm_mmu_invlpg(vcpu, exit_qualification);
5766 skip_emulated_instruction(vcpu);
5770 static int handle_rdpmc(struct kvm_vcpu *vcpu)
5774 err = kvm_rdpmc(vcpu);
5775 kvm_complete_insn_gp(vcpu, err);
5780 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5782 kvm_emulate_wbinvd(vcpu);
5786 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5788 u64 new_bv = kvm_read_edx_eax(vcpu);
5789 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5791 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5792 skip_emulated_instruction(vcpu);
5796 static int handle_xsaves(struct kvm_vcpu *vcpu)
5798 skip_emulated_instruction(vcpu);
5799 WARN(1, "this should never happen\n");
5803 static int handle_xrstors(struct kvm_vcpu *vcpu)
5805 skip_emulated_instruction(vcpu);
5806 WARN(1, "this should never happen\n");
5810 static int handle_apic_access(struct kvm_vcpu *vcpu)
5812 if (likely(fasteoi)) {
5813 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5814 int access_type, offset;
5816 access_type = exit_qualification & APIC_ACCESS_TYPE;
5817 offset = exit_qualification & APIC_ACCESS_OFFSET;
5819 * Sane guest uses MOV to write EOI, with written value
5820 * not cared. So make a short-circuit here by avoiding
5821 * heavy instruction emulation.
5823 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5824 (offset == APIC_EOI)) {
5825 kvm_lapic_set_eoi(vcpu);
5826 skip_emulated_instruction(vcpu);
5830 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5833 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5835 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5836 int vector = exit_qualification & 0xff;
5838 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5839 kvm_apic_set_eoi_accelerated(vcpu, vector);
5843 static int handle_apic_write(struct kvm_vcpu *vcpu)
5845 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5846 u32 offset = exit_qualification & 0xfff;
5848 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
5849 kvm_apic_write_nodecode(vcpu, offset);
5853 static int handle_task_switch(struct kvm_vcpu *vcpu)
5855 struct vcpu_vmx *vmx = to_vmx(vcpu);
5856 unsigned long exit_qualification;
5857 bool has_error_code = false;
5860 int reason, type, idt_v, idt_index;
5862 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5863 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5864 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5866 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5868 reason = (u32)exit_qualification >> 30;
5869 if (reason == TASK_SWITCH_GATE && idt_v) {
5871 case INTR_TYPE_NMI_INTR:
5872 vcpu->arch.nmi_injected = false;
5873 vmx_set_nmi_mask(vcpu, true);
5875 case INTR_TYPE_EXT_INTR:
5876 case INTR_TYPE_SOFT_INTR:
5877 kvm_clear_interrupt_queue(vcpu);
5879 case INTR_TYPE_HARD_EXCEPTION:
5880 if (vmx->idt_vectoring_info &
5881 VECTORING_INFO_DELIVER_CODE_MASK) {
5882 has_error_code = true;
5884 vmcs_read32(IDT_VECTORING_ERROR_CODE);
5887 case INTR_TYPE_SOFT_EXCEPTION:
5888 kvm_clear_exception_queue(vcpu);
5894 tss_selector = exit_qualification;
5896 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5897 type != INTR_TYPE_EXT_INTR &&
5898 type != INTR_TYPE_NMI_INTR))
5899 skip_emulated_instruction(vcpu);
5901 if (kvm_task_switch(vcpu, tss_selector,
5902 type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
5903 has_error_code, error_code) == EMULATE_FAIL) {
5904 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5905 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5906 vcpu->run->internal.ndata = 0;
5911 * TODO: What about debug traps on tss switch?
5912 * Are we supposed to inject them and update dr6?
5918 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5920 unsigned long exit_qualification;
5925 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5927 gla_validity = (exit_qualification >> 7) & 0x3;
5928 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
5929 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
5930 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
5931 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
5932 vmcs_readl(GUEST_LINEAR_ADDRESS));
5933 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
5934 (long unsigned int)exit_qualification);
5935 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5936 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
5941 * EPT violation happened while executing iret from NMI,
5942 * "blocked by NMI" bit has to be set before next VM entry.
5943 * There are errata that may cause this bit to not be set:
5946 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5947 cpu_has_virtual_nmis() &&
5948 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
5949 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5951 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5952 trace_kvm_page_fault(gpa, exit_qualification);
5954 /* It is a write fault? */
5955 error_code = exit_qualification & PFERR_WRITE_MASK;
5956 /* It is a fetch fault? */
5957 error_code |= (exit_qualification << 2) & PFERR_FETCH_MASK;
5958 /* ept page table is present? */
5959 error_code |= (exit_qualification >> 3) & PFERR_PRESENT_MASK;
5961 vcpu->arch.exit_qualification = exit_qualification;
5963 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5966 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5971 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5972 if (!kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5973 skip_emulated_instruction(vcpu);
5974 trace_kvm_fast_mmio(gpa);
5978 ret = handle_mmio_page_fault(vcpu, gpa, true);
5979 if (likely(ret == RET_MMIO_PF_EMULATE))
5980 return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) ==
5983 if (unlikely(ret == RET_MMIO_PF_INVALID))
5984 return kvm_mmu_page_fault(vcpu, gpa, 0, NULL, 0);
5986 if (unlikely(ret == RET_MMIO_PF_RETRY))
5989 /* It is the real ept misconfig */
5992 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5993 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
5998 static int handle_nmi_window(struct kvm_vcpu *vcpu)
6000 u32 cpu_based_vm_exec_control;
6002 /* clear pending NMI */
6003 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
6004 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
6005 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
6006 ++vcpu->stat.nmi_window_exits;
6007 kvm_make_request(KVM_REQ_EVENT, vcpu);
6012 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
6014 struct vcpu_vmx *vmx = to_vmx(vcpu);
6015 enum emulation_result err = EMULATE_DONE;
6018 bool intr_window_requested;
6019 unsigned count = 130;
6021 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
6022 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
6024 while (vmx->emulation_required && count-- != 0) {
6025 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
6026 return handle_interrupt_window(&vmx->vcpu);
6028 if (test_bit(KVM_REQ_EVENT, &vcpu->requests))
6031 err = emulate_instruction(vcpu, EMULTYPE_NO_REEXECUTE);
6033 if (err == EMULATE_USER_EXIT) {
6034 ++vcpu->stat.mmio_exits;
6039 if (err != EMULATE_DONE) {
6040 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6041 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6042 vcpu->run->internal.ndata = 0;
6046 if (vcpu->arch.halt_request) {
6047 vcpu->arch.halt_request = 0;
6048 ret = kvm_vcpu_halt(vcpu);
6052 if (signal_pending(current))
6062 static int __grow_ple_window(int val)
6064 if (ple_window_grow < 1)
6067 val = min(val, ple_window_actual_max);
6069 if (ple_window_grow < ple_window)
6070 val *= ple_window_grow;
6072 val += ple_window_grow;
6077 static int __shrink_ple_window(int val, int modifier, int minimum)
6082 if (modifier < ple_window)
6087 return max(val, minimum);
6090 static void grow_ple_window(struct kvm_vcpu *vcpu)
6092 struct vcpu_vmx *vmx = to_vmx(vcpu);
6093 int old = vmx->ple_window;
6095 vmx->ple_window = __grow_ple_window(old);
6097 if (vmx->ple_window != old)
6098 vmx->ple_window_dirty = true;
6100 trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old);
6103 static void shrink_ple_window(struct kvm_vcpu *vcpu)
6105 struct vcpu_vmx *vmx = to_vmx(vcpu);
6106 int old = vmx->ple_window;
6108 vmx->ple_window = __shrink_ple_window(old,
6109 ple_window_shrink, ple_window);
6111 if (vmx->ple_window != old)
6112 vmx->ple_window_dirty = true;
6114 trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old);
6118 * ple_window_actual_max is computed to be one grow_ple_window() below
6119 * ple_window_max. (See __grow_ple_window for the reason.)
6120 * This prevents overflows, because ple_window_max is int.
6121 * ple_window_max effectively rounded down to a multiple of ple_window_grow in
6123 * ple_window_max is also prevented from setting vmx->ple_window < ple_window.
6125 static void update_ple_window_actual_max(void)
6127 ple_window_actual_max =
6128 __shrink_ple_window(max(ple_window_max, ple_window),
6129 ple_window_grow, INT_MIN);
6133 * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
6135 static void wakeup_handler(void)
6137 struct kvm_vcpu *vcpu;
6138 int cpu = smp_processor_id();
6140 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6141 list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
6142 blocked_vcpu_list) {
6143 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6145 if (pi_test_on(pi_desc) == 1)
6146 kvm_vcpu_kick(vcpu);
6148 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6151 static __init int hardware_setup(void)
6153 int r = -ENOMEM, i, msr;
6155 rdmsrl_safe(MSR_EFER, &host_efer);
6157 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
6158 kvm_define_shared_msr(i, vmx_msr_index[i]);
6160 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
6161 if (!vmx_io_bitmap_a)
6164 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
6165 if (!vmx_io_bitmap_b)
6168 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
6169 if (!vmx_msr_bitmap_legacy)
6172 vmx_msr_bitmap_legacy_x2apic =
6173 (unsigned long *)__get_free_page(GFP_KERNEL);
6174 if (!vmx_msr_bitmap_legacy_x2apic)
6177 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
6178 if (!vmx_msr_bitmap_longmode)
6181 vmx_msr_bitmap_longmode_x2apic =
6182 (unsigned long *)__get_free_page(GFP_KERNEL);
6183 if (!vmx_msr_bitmap_longmode_x2apic)
6187 vmx_msr_bitmap_nested =
6188 (unsigned long *)__get_free_page(GFP_KERNEL);
6189 if (!vmx_msr_bitmap_nested)
6193 vmx_vmread_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
6194 if (!vmx_vmread_bitmap)
6197 vmx_vmwrite_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
6198 if (!vmx_vmwrite_bitmap)
6201 memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
6202 memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
6205 * Allow direct access to the PC debug port (it is often used for I/O
6206 * delays, but the vmexits simply slow things down).
6208 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
6209 clear_bit(0x80, vmx_io_bitmap_a);
6211 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
6213 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
6214 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
6216 memset(vmx_msr_bitmap_nested, 0xff, PAGE_SIZE);
6218 if (setup_vmcs_config(&vmcs_config) < 0) {
6223 if (boot_cpu_has(X86_FEATURE_NX))
6224 kvm_enable_efer_bits(EFER_NX);
6226 if (!cpu_has_vmx_vpid())
6228 if (!cpu_has_vmx_shadow_vmcs())
6229 enable_shadow_vmcs = 0;
6230 if (enable_shadow_vmcs)
6231 init_vmcs_shadow_fields();
6233 if (!cpu_has_vmx_ept() ||
6234 !cpu_has_vmx_ept_4levels()) {
6236 enable_unrestricted_guest = 0;
6237 enable_ept_ad_bits = 0;
6240 if (!cpu_has_vmx_ept_ad_bits())
6241 enable_ept_ad_bits = 0;
6243 if (!cpu_has_vmx_unrestricted_guest())
6244 enable_unrestricted_guest = 0;
6246 if (!cpu_has_vmx_flexpriority())
6247 flexpriority_enabled = 0;
6250 * set_apic_access_page_addr() is used to reload apic access
6251 * page upon invalidation. No need to do anything if not
6252 * using the APIC_ACCESS_ADDR VMCS field.
6254 if (!flexpriority_enabled)
6255 kvm_x86_ops->set_apic_access_page_addr = NULL;
6257 if (!cpu_has_vmx_tpr_shadow())
6258 kvm_x86_ops->update_cr8_intercept = NULL;
6260 if (enable_ept && !cpu_has_vmx_ept_2m_page())
6261 kvm_disable_largepages();
6263 if (!cpu_has_vmx_ple())
6266 if (!cpu_has_vmx_apicv())
6269 if (cpu_has_vmx_tsc_scaling()) {
6270 kvm_has_tsc_control = true;
6271 kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
6272 kvm_tsc_scaling_ratio_frac_bits = 48;
6275 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
6276 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
6277 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
6278 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
6279 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
6280 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
6281 vmx_disable_intercept_for_msr(MSR_IA32_BNDCFGS, true);
6283 memcpy(vmx_msr_bitmap_legacy_x2apic,
6284 vmx_msr_bitmap_legacy, PAGE_SIZE);
6285 memcpy(vmx_msr_bitmap_longmode_x2apic,
6286 vmx_msr_bitmap_longmode, PAGE_SIZE);
6288 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
6291 for (msr = 0x800; msr <= 0x8ff; msr++)
6292 vmx_disable_intercept_msr_read_x2apic(msr);
6294 /* According SDM, in x2apic mode, the whole id reg is used.
6295 * But in KVM, it only use the highest eight bits. Need to
6297 vmx_enable_intercept_msr_read_x2apic(0x802);
6299 vmx_enable_intercept_msr_read_x2apic(0x839);
6301 vmx_disable_intercept_msr_write_x2apic(0x808);
6303 vmx_disable_intercept_msr_write_x2apic(0x80b);
6305 vmx_disable_intercept_msr_write_x2apic(0x83f);
6309 kvm_mmu_set_mask_ptes(0ull,
6310 (enable_ept_ad_bits) ? VMX_EPT_ACCESS_BIT : 0ull,
6311 (enable_ept_ad_bits) ? VMX_EPT_DIRTY_BIT : 0ull,
6312 0ull, VMX_EPT_EXECUTABLE_MASK);
6313 ept_set_mmio_spte_mask();
6318 update_ple_window_actual_max();
6321 * Only enable PML when hardware supports PML feature, and both EPT
6322 * and EPT A/D bit features are enabled -- PML depends on them to work.
6324 if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
6328 kvm_x86_ops->slot_enable_log_dirty = NULL;
6329 kvm_x86_ops->slot_disable_log_dirty = NULL;
6330 kvm_x86_ops->flush_log_dirty = NULL;
6331 kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
6334 kvm_set_posted_intr_wakeup_handler(wakeup_handler);
6336 return alloc_kvm_area();
6339 free_page((unsigned long)vmx_vmwrite_bitmap);
6341 free_page((unsigned long)vmx_vmread_bitmap);
6344 free_page((unsigned long)vmx_msr_bitmap_nested);
6346 free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
6348 free_page((unsigned long)vmx_msr_bitmap_longmode);
6350 free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
6352 free_page((unsigned long)vmx_msr_bitmap_legacy);
6354 free_page((unsigned long)vmx_io_bitmap_b);
6356 free_page((unsigned long)vmx_io_bitmap_a);
6361 static __exit void hardware_unsetup(void)
6363 free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
6364 free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
6365 free_page((unsigned long)vmx_msr_bitmap_legacy);
6366 free_page((unsigned long)vmx_msr_bitmap_longmode);
6367 free_page((unsigned long)vmx_io_bitmap_b);
6368 free_page((unsigned long)vmx_io_bitmap_a);
6369 free_page((unsigned long)vmx_vmwrite_bitmap);
6370 free_page((unsigned long)vmx_vmread_bitmap);
6372 free_page((unsigned long)vmx_msr_bitmap_nested);
6378 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
6379 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
6381 static int handle_pause(struct kvm_vcpu *vcpu)
6384 grow_ple_window(vcpu);
6386 skip_emulated_instruction(vcpu);
6387 kvm_vcpu_on_spin(vcpu);
6392 static int handle_nop(struct kvm_vcpu *vcpu)
6394 skip_emulated_instruction(vcpu);
6398 static int handle_mwait(struct kvm_vcpu *vcpu)
6400 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
6401 return handle_nop(vcpu);
6404 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
6409 static int handle_monitor(struct kvm_vcpu *vcpu)
6411 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
6412 return handle_nop(vcpu);
6416 * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
6417 * We could reuse a single VMCS for all the L2 guests, but we also want the
6418 * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
6419 * allows keeping them loaded on the processor, and in the future will allow
6420 * optimizations where prepare_vmcs02 doesn't need to set all the fields on
6421 * every entry if they never change.
6422 * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
6423 * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
6425 * The following functions allocate and free a vmcs02 in this pool.
6428 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
6429 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
6431 struct vmcs02_list *item;
6432 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6433 if (item->vmptr == vmx->nested.current_vmptr) {
6434 list_move(&item->list, &vmx->nested.vmcs02_pool);
6435 return &item->vmcs02;
6438 if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
6439 /* Recycle the least recently used VMCS. */
6440 item = list_last_entry(&vmx->nested.vmcs02_pool,
6441 struct vmcs02_list, list);
6442 item->vmptr = vmx->nested.current_vmptr;
6443 list_move(&item->list, &vmx->nested.vmcs02_pool);
6444 return &item->vmcs02;
6447 /* Create a new VMCS */
6448 item = kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
6451 item->vmcs02.vmcs = alloc_vmcs();
6452 if (!item->vmcs02.vmcs) {
6456 loaded_vmcs_init(&item->vmcs02);
6457 item->vmptr = vmx->nested.current_vmptr;
6458 list_add(&(item->list), &(vmx->nested.vmcs02_pool));
6459 vmx->nested.vmcs02_num++;
6460 return &item->vmcs02;
6463 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
6464 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
6466 struct vmcs02_list *item;
6467 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6468 if (item->vmptr == vmptr) {
6469 free_loaded_vmcs(&item->vmcs02);
6470 list_del(&item->list);
6472 vmx->nested.vmcs02_num--;
6478 * Free all VMCSs saved for this vcpu, except the one pointed by
6479 * vmx->loaded_vmcs. We must be running L1, so vmx->loaded_vmcs
6480 * must be &vmx->vmcs01.
6482 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
6484 struct vmcs02_list *item, *n;
6486 WARN_ON(vmx->loaded_vmcs != &vmx->vmcs01);
6487 list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
6489 * Something will leak if the above WARN triggers. Better than
6492 if (vmx->loaded_vmcs == &item->vmcs02)
6495 free_loaded_vmcs(&item->vmcs02);
6496 list_del(&item->list);
6498 vmx->nested.vmcs02_num--;
6503 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
6504 * set the success or error code of an emulated VMX instruction, as specified
6505 * by Vol 2B, VMX Instruction Reference, "Conventions".
6507 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
6509 vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
6510 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6511 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
6514 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
6516 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6517 & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
6518 X86_EFLAGS_SF | X86_EFLAGS_OF))
6522 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
6523 u32 vm_instruction_error)
6525 if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
6527 * failValid writes the error number to the current VMCS, which
6528 * can't be done there isn't a current VMCS.
6530 nested_vmx_failInvalid(vcpu);
6533 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6534 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6535 X86_EFLAGS_SF | X86_EFLAGS_OF))
6537 get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
6539 * We don't need to force a shadow sync because
6540 * VM_INSTRUCTION_ERROR is not shadowed
6544 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
6546 /* TODO: not to reset guest simply here. */
6547 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6548 pr_warn("kvm: nested vmx abort, indicator %d\n", indicator);
6551 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
6553 struct vcpu_vmx *vmx =
6554 container_of(timer, struct vcpu_vmx, nested.preemption_timer);
6556 vmx->nested.preemption_timer_expired = true;
6557 kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
6558 kvm_vcpu_kick(&vmx->vcpu);
6560 return HRTIMER_NORESTART;
6564 * Decode the memory-address operand of a vmx instruction, as recorded on an
6565 * exit caused by such an instruction (run by a guest hypervisor).
6566 * On success, returns 0. When the operand is invalid, returns 1 and throws
6569 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
6570 unsigned long exit_qualification,
6571 u32 vmx_instruction_info, bool wr, gva_t *ret)
6575 struct kvm_segment s;
6578 * According to Vol. 3B, "Information for VM Exits Due to Instruction
6579 * Execution", on an exit, vmx_instruction_info holds most of the
6580 * addressing components of the operand. Only the displacement part
6581 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
6582 * For how an actual address is calculated from all these components,
6583 * refer to Vol. 1, "Operand Addressing".
6585 int scaling = vmx_instruction_info & 3;
6586 int addr_size = (vmx_instruction_info >> 7) & 7;
6587 bool is_reg = vmx_instruction_info & (1u << 10);
6588 int seg_reg = (vmx_instruction_info >> 15) & 7;
6589 int index_reg = (vmx_instruction_info >> 18) & 0xf;
6590 bool index_is_valid = !(vmx_instruction_info & (1u << 22));
6591 int base_reg = (vmx_instruction_info >> 23) & 0xf;
6592 bool base_is_valid = !(vmx_instruction_info & (1u << 27));
6595 kvm_queue_exception(vcpu, UD_VECTOR);
6599 /* Addr = segment_base + offset */
6600 /* offset = base + [index * scale] + displacement */
6601 off = exit_qualification; /* holds the displacement */
6603 off += kvm_register_read(vcpu, base_reg);
6605 off += kvm_register_read(vcpu, index_reg)<<scaling;
6606 vmx_get_segment(vcpu, &s, seg_reg);
6607 *ret = s.base + off;
6609 if (addr_size == 1) /* 32 bit */
6612 /* Checks for #GP/#SS exceptions. */
6614 if (is_protmode(vcpu)) {
6615 /* Protected mode: apply checks for segment validity in the
6617 * - segment type check (#GP(0) may be thrown)
6618 * - usability check (#GP(0)/#SS(0))
6619 * - limit check (#GP(0)/#SS(0))
6622 /* #GP(0) if the destination operand is located in a
6623 * read-only data segment or any code segment.
6625 exn = ((s.type & 0xa) == 0 || (s.type & 8));
6627 /* #GP(0) if the source operand is located in an
6628 * execute-only code segment
6630 exn = ((s.type & 0xa) == 8);
6633 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6636 if (is_long_mode(vcpu)) {
6637 /* Long mode: #GP(0)/#SS(0) if the memory address is in a
6638 * non-canonical form. This is an only check for long mode.
6640 exn = is_noncanonical_address(*ret);
6641 } else if (is_protmode(vcpu)) {
6642 /* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
6644 exn = (s.unusable != 0);
6645 /* Protected mode: #GP(0)/#SS(0) if the memory
6646 * operand is outside the segment limit.
6648 exn = exn || (off + sizeof(u64) > s.limit);
6651 kvm_queue_exception_e(vcpu,
6652 seg_reg == VCPU_SREG_SS ?
6653 SS_VECTOR : GP_VECTOR,
6662 * This function performs the various checks including
6663 * - if it's 4KB aligned
6664 * - No bits beyond the physical address width are set
6665 * - Returns 0 on success or else 1
6666 * (Intel SDM Section 30.3)
6668 static int nested_vmx_check_vmptr(struct kvm_vcpu *vcpu, int exit_reason,
6673 struct x86_exception e;
6675 struct vcpu_vmx *vmx = to_vmx(vcpu);
6676 int maxphyaddr = cpuid_maxphyaddr(vcpu);
6678 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
6679 vmcs_read32(VMX_INSTRUCTION_INFO), false, &gva))
6682 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
6683 sizeof(vmptr), &e)) {
6684 kvm_inject_page_fault(vcpu, &e);
6688 switch (exit_reason) {
6689 case EXIT_REASON_VMON:
6692 * The first 4 bytes of VMXON region contain the supported
6693 * VMCS revision identifier
6695 * Note - IA32_VMX_BASIC[48] will never be 1
6696 * for the nested case;
6697 * which replaces physical address width with 32
6700 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6701 nested_vmx_failInvalid(vcpu);
6702 skip_emulated_instruction(vcpu);
6706 page = nested_get_page(vcpu, vmptr);
6708 *(u32 *)kmap(page) != VMCS12_REVISION) {
6709 nested_vmx_failInvalid(vcpu);
6711 skip_emulated_instruction(vcpu);
6715 vmx->nested.vmxon_ptr = vmptr;
6717 case EXIT_REASON_VMCLEAR:
6718 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6719 nested_vmx_failValid(vcpu,
6720 VMXERR_VMCLEAR_INVALID_ADDRESS);
6721 skip_emulated_instruction(vcpu);
6725 if (vmptr == vmx->nested.vmxon_ptr) {
6726 nested_vmx_failValid(vcpu,
6727 VMXERR_VMCLEAR_VMXON_POINTER);
6728 skip_emulated_instruction(vcpu);
6732 case EXIT_REASON_VMPTRLD:
6733 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6734 nested_vmx_failValid(vcpu,
6735 VMXERR_VMPTRLD_INVALID_ADDRESS);
6736 skip_emulated_instruction(vcpu);
6740 if (vmptr == vmx->nested.vmxon_ptr) {
6741 nested_vmx_failValid(vcpu,
6742 VMXERR_VMCLEAR_VMXON_POINTER);
6743 skip_emulated_instruction(vcpu);
6748 return 1; /* shouldn't happen */
6757 * Emulate the VMXON instruction.
6758 * Currently, we just remember that VMX is active, and do not save or even
6759 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
6760 * do not currently need to store anything in that guest-allocated memory
6761 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
6762 * argument is different from the VMXON pointer (which the spec says they do).
6764 static int handle_vmon(struct kvm_vcpu *vcpu)
6766 struct kvm_segment cs;
6767 struct vcpu_vmx *vmx = to_vmx(vcpu);
6768 struct vmcs *shadow_vmcs;
6769 const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
6770 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
6772 /* The Intel VMX Instruction Reference lists a bunch of bits that
6773 * are prerequisite to running VMXON, most notably cr4.VMXE must be
6774 * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
6775 * Otherwise, we should fail with #UD. We test these now:
6777 if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
6778 !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
6779 (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
6780 kvm_queue_exception(vcpu, UD_VECTOR);
6784 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
6785 if (is_long_mode(vcpu) && !cs.l) {
6786 kvm_queue_exception(vcpu, UD_VECTOR);
6790 if (vmx_get_cpl(vcpu)) {
6791 kvm_inject_gp(vcpu, 0);
6795 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMON, NULL))
6798 if (vmx->nested.vmxon) {
6799 nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
6800 skip_emulated_instruction(vcpu);
6804 if ((vmx->nested.msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
6805 != VMXON_NEEDED_FEATURES) {
6806 kvm_inject_gp(vcpu, 0);
6810 if (enable_shadow_vmcs) {
6811 shadow_vmcs = alloc_vmcs();
6814 /* mark vmcs as shadow */
6815 shadow_vmcs->revision_id |= (1u << 31);
6816 /* init shadow vmcs */
6817 vmcs_clear(shadow_vmcs);
6818 vmx->nested.current_shadow_vmcs = shadow_vmcs;
6821 INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
6822 vmx->nested.vmcs02_num = 0;
6824 hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
6826 vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
6828 vmx->nested.vmxon = true;
6830 skip_emulated_instruction(vcpu);
6831 nested_vmx_succeed(vcpu);
6836 * Intel's VMX Instruction Reference specifies a common set of prerequisites
6837 * for running VMX instructions (except VMXON, whose prerequisites are
6838 * slightly different). It also specifies what exception to inject otherwise.
6840 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
6842 struct kvm_segment cs;
6843 struct vcpu_vmx *vmx = to_vmx(vcpu);
6845 if (!vmx->nested.vmxon) {
6846 kvm_queue_exception(vcpu, UD_VECTOR);
6850 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
6851 if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
6852 (is_long_mode(vcpu) && !cs.l)) {
6853 kvm_queue_exception(vcpu, UD_VECTOR);
6857 if (vmx_get_cpl(vcpu)) {
6858 kvm_inject_gp(vcpu, 0);
6865 static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
6867 if (vmx->nested.current_vmptr == -1ull)
6870 /* current_vmptr and current_vmcs12 are always set/reset together */
6871 if (WARN_ON(vmx->nested.current_vmcs12 == NULL))
6874 if (enable_shadow_vmcs) {
6875 /* copy to memory all shadowed fields in case
6876 they were modified */
6877 copy_shadow_to_vmcs12(vmx);
6878 vmx->nested.sync_shadow_vmcs = false;
6879 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
6880 SECONDARY_EXEC_SHADOW_VMCS);
6881 vmcs_write64(VMCS_LINK_POINTER, -1ull);
6883 vmx->nested.posted_intr_nv = -1;
6884 kunmap(vmx->nested.current_vmcs12_page);
6885 nested_release_page(vmx->nested.current_vmcs12_page);
6886 vmx->nested.current_vmptr = -1ull;
6887 vmx->nested.current_vmcs12 = NULL;
6891 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
6892 * just stops using VMX.
6894 static void free_nested(struct vcpu_vmx *vmx)
6896 if (!vmx->nested.vmxon)
6899 vmx->nested.vmxon = false;
6900 free_vpid(vmx->nested.vpid02);
6901 nested_release_vmcs12(vmx);
6902 if (enable_shadow_vmcs)
6903 free_vmcs(vmx->nested.current_shadow_vmcs);
6904 /* Unpin physical memory we referred to in current vmcs02 */
6905 if (vmx->nested.apic_access_page) {
6906 nested_release_page(vmx->nested.apic_access_page);
6907 vmx->nested.apic_access_page = NULL;
6909 if (vmx->nested.virtual_apic_page) {
6910 nested_release_page(vmx->nested.virtual_apic_page);
6911 vmx->nested.virtual_apic_page = NULL;
6913 if (vmx->nested.pi_desc_page) {
6914 kunmap(vmx->nested.pi_desc_page);
6915 nested_release_page(vmx->nested.pi_desc_page);
6916 vmx->nested.pi_desc_page = NULL;
6917 vmx->nested.pi_desc = NULL;
6920 nested_free_all_saved_vmcss(vmx);
6923 /* Emulate the VMXOFF instruction */
6924 static int handle_vmoff(struct kvm_vcpu *vcpu)
6926 if (!nested_vmx_check_permission(vcpu))
6928 free_nested(to_vmx(vcpu));
6929 skip_emulated_instruction(vcpu);
6930 nested_vmx_succeed(vcpu);
6934 /* Emulate the VMCLEAR instruction */
6935 static int handle_vmclear(struct kvm_vcpu *vcpu)
6937 struct vcpu_vmx *vmx = to_vmx(vcpu);
6939 struct vmcs12 *vmcs12;
6942 if (!nested_vmx_check_permission(vcpu))
6945 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMCLEAR, &vmptr))
6948 if (vmptr == vmx->nested.current_vmptr)
6949 nested_release_vmcs12(vmx);
6951 page = nested_get_page(vcpu, vmptr);
6954 * For accurate processor emulation, VMCLEAR beyond available
6955 * physical memory should do nothing at all. However, it is
6956 * possible that a nested vmx bug, not a guest hypervisor bug,
6957 * resulted in this case, so let's shut down before doing any
6960 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6963 vmcs12 = kmap(page);
6964 vmcs12->launch_state = 0;
6966 nested_release_page(page);
6968 nested_free_vmcs02(vmx, vmptr);
6970 skip_emulated_instruction(vcpu);
6971 nested_vmx_succeed(vcpu);
6975 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
6977 /* Emulate the VMLAUNCH instruction */
6978 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
6980 return nested_vmx_run(vcpu, true);
6983 /* Emulate the VMRESUME instruction */
6984 static int handle_vmresume(struct kvm_vcpu *vcpu)
6987 return nested_vmx_run(vcpu, false);
6990 enum vmcs_field_type {
6991 VMCS_FIELD_TYPE_U16 = 0,
6992 VMCS_FIELD_TYPE_U64 = 1,
6993 VMCS_FIELD_TYPE_U32 = 2,
6994 VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
6997 static inline int vmcs_field_type(unsigned long field)
6999 if (0x1 & field) /* the *_HIGH fields are all 32 bit */
7000 return VMCS_FIELD_TYPE_U32;
7001 return (field >> 13) & 0x3 ;
7004 static inline int vmcs_field_readonly(unsigned long field)
7006 return (((field >> 10) & 0x3) == 1);
7010 * Read a vmcs12 field. Since these can have varying lengths and we return
7011 * one type, we chose the biggest type (u64) and zero-extend the return value
7012 * to that size. Note that the caller, handle_vmread, might need to use only
7013 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
7014 * 64-bit fields are to be returned).
7016 static inline int vmcs12_read_any(struct kvm_vcpu *vcpu,
7017 unsigned long field, u64 *ret)
7019 short offset = vmcs_field_to_offset(field);
7025 p = ((char *)(get_vmcs12(vcpu))) + offset;
7027 switch (vmcs_field_type(field)) {
7028 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7029 *ret = *((natural_width *)p);
7031 case VMCS_FIELD_TYPE_U16:
7034 case VMCS_FIELD_TYPE_U32:
7037 case VMCS_FIELD_TYPE_U64:
7047 static inline int vmcs12_write_any(struct kvm_vcpu *vcpu,
7048 unsigned long field, u64 field_value){
7049 short offset = vmcs_field_to_offset(field);
7050 char *p = ((char *) get_vmcs12(vcpu)) + offset;
7054 switch (vmcs_field_type(field)) {
7055 case VMCS_FIELD_TYPE_U16:
7056 *(u16 *)p = field_value;
7058 case VMCS_FIELD_TYPE_U32:
7059 *(u32 *)p = field_value;
7061 case VMCS_FIELD_TYPE_U64:
7062 *(u64 *)p = field_value;
7064 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7065 *(natural_width *)p = field_value;
7074 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
7077 unsigned long field;
7079 struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
7080 const unsigned long *fields = shadow_read_write_fields;
7081 const int num_fields = max_shadow_read_write_fields;
7085 vmcs_load(shadow_vmcs);
7087 for (i = 0; i < num_fields; i++) {
7089 switch (vmcs_field_type(field)) {
7090 case VMCS_FIELD_TYPE_U16:
7091 field_value = vmcs_read16(field);
7093 case VMCS_FIELD_TYPE_U32:
7094 field_value = vmcs_read32(field);
7096 case VMCS_FIELD_TYPE_U64:
7097 field_value = vmcs_read64(field);
7099 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7100 field_value = vmcs_readl(field);
7106 vmcs12_write_any(&vmx->vcpu, field, field_value);
7109 vmcs_clear(shadow_vmcs);
7110 vmcs_load(vmx->loaded_vmcs->vmcs);
7115 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
7117 const unsigned long *fields[] = {
7118 shadow_read_write_fields,
7119 shadow_read_only_fields
7121 const int max_fields[] = {
7122 max_shadow_read_write_fields,
7123 max_shadow_read_only_fields
7126 unsigned long field;
7127 u64 field_value = 0;
7128 struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
7130 vmcs_load(shadow_vmcs);
7132 for (q = 0; q < ARRAY_SIZE(fields); q++) {
7133 for (i = 0; i < max_fields[q]; i++) {
7134 field = fields[q][i];
7135 vmcs12_read_any(&vmx->vcpu, field, &field_value);
7137 switch (vmcs_field_type(field)) {
7138 case VMCS_FIELD_TYPE_U16:
7139 vmcs_write16(field, (u16)field_value);
7141 case VMCS_FIELD_TYPE_U32:
7142 vmcs_write32(field, (u32)field_value);
7144 case VMCS_FIELD_TYPE_U64:
7145 vmcs_write64(field, (u64)field_value);
7147 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7148 vmcs_writel(field, (long)field_value);
7157 vmcs_clear(shadow_vmcs);
7158 vmcs_load(vmx->loaded_vmcs->vmcs);
7162 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
7163 * used before) all generate the same failure when it is missing.
7165 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
7167 struct vcpu_vmx *vmx = to_vmx(vcpu);
7168 if (vmx->nested.current_vmptr == -1ull) {
7169 nested_vmx_failInvalid(vcpu);
7170 skip_emulated_instruction(vcpu);
7176 static int handle_vmread(struct kvm_vcpu *vcpu)
7178 unsigned long field;
7180 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7181 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7184 if (!nested_vmx_check_permission(vcpu) ||
7185 !nested_vmx_check_vmcs12(vcpu))
7188 /* Decode instruction info and find the field to read */
7189 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7190 /* Read the field, zero-extended to a u64 field_value */
7191 if (vmcs12_read_any(vcpu, field, &field_value) < 0) {
7192 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7193 skip_emulated_instruction(vcpu);
7197 * Now copy part of this value to register or memory, as requested.
7198 * Note that the number of bits actually copied is 32 or 64 depending
7199 * on the guest's mode (32 or 64 bit), not on the given field's length.
7201 if (vmx_instruction_info & (1u << 10)) {
7202 kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
7205 if (get_vmx_mem_address(vcpu, exit_qualification,
7206 vmx_instruction_info, true, &gva))
7208 /* _system ok, as nested_vmx_check_permission verified cpl=0 */
7209 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
7210 &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
7213 nested_vmx_succeed(vcpu);
7214 skip_emulated_instruction(vcpu);
7219 static int handle_vmwrite(struct kvm_vcpu *vcpu)
7221 unsigned long field;
7223 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7224 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7225 /* The value to write might be 32 or 64 bits, depending on L1's long
7226 * mode, and eventually we need to write that into a field of several
7227 * possible lengths. The code below first zero-extends the value to 64
7228 * bit (field_value), and then copies only the approriate number of
7229 * bits into the vmcs12 field.
7231 u64 field_value = 0;
7232 struct x86_exception e;
7234 if (!nested_vmx_check_permission(vcpu) ||
7235 !nested_vmx_check_vmcs12(vcpu))
7238 if (vmx_instruction_info & (1u << 10))
7239 field_value = kvm_register_readl(vcpu,
7240 (((vmx_instruction_info) >> 3) & 0xf));
7242 if (get_vmx_mem_address(vcpu, exit_qualification,
7243 vmx_instruction_info, false, &gva))
7245 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
7246 &field_value, (is_64_bit_mode(vcpu) ? 8 : 4), &e)) {
7247 kvm_inject_page_fault(vcpu, &e);
7253 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7254 if (vmcs_field_readonly(field)) {
7255 nested_vmx_failValid(vcpu,
7256 VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
7257 skip_emulated_instruction(vcpu);
7261 if (vmcs12_write_any(vcpu, field, field_value) < 0) {
7262 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7263 skip_emulated_instruction(vcpu);
7267 nested_vmx_succeed(vcpu);
7268 skip_emulated_instruction(vcpu);
7272 /* Emulate the VMPTRLD instruction */
7273 static int handle_vmptrld(struct kvm_vcpu *vcpu)
7275 struct vcpu_vmx *vmx = to_vmx(vcpu);
7278 if (!nested_vmx_check_permission(vcpu))
7281 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMPTRLD, &vmptr))
7284 if (vmx->nested.current_vmptr != vmptr) {
7285 struct vmcs12 *new_vmcs12;
7287 page = nested_get_page(vcpu, vmptr);
7289 nested_vmx_failInvalid(vcpu);
7290 skip_emulated_instruction(vcpu);
7293 new_vmcs12 = kmap(page);
7294 if (new_vmcs12->revision_id != VMCS12_REVISION) {
7296 nested_release_page_clean(page);
7297 nested_vmx_failValid(vcpu,
7298 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
7299 skip_emulated_instruction(vcpu);
7303 nested_release_vmcs12(vmx);
7304 vmx->nested.current_vmptr = vmptr;
7305 vmx->nested.current_vmcs12 = new_vmcs12;
7306 vmx->nested.current_vmcs12_page = page;
7307 if (enable_shadow_vmcs) {
7308 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
7309 SECONDARY_EXEC_SHADOW_VMCS);
7310 vmcs_write64(VMCS_LINK_POINTER,
7311 __pa(vmx->nested.current_shadow_vmcs));
7312 vmx->nested.sync_shadow_vmcs = true;
7316 nested_vmx_succeed(vcpu);
7317 skip_emulated_instruction(vcpu);
7321 /* Emulate the VMPTRST instruction */
7322 static int handle_vmptrst(struct kvm_vcpu *vcpu)
7324 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7325 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7327 struct x86_exception e;
7329 if (!nested_vmx_check_permission(vcpu))
7332 if (get_vmx_mem_address(vcpu, exit_qualification,
7333 vmx_instruction_info, true, &vmcs_gva))
7335 /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
7336 if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
7337 (void *)&to_vmx(vcpu)->nested.current_vmptr,
7339 kvm_inject_page_fault(vcpu, &e);
7342 nested_vmx_succeed(vcpu);
7343 skip_emulated_instruction(vcpu);
7347 /* Emulate the INVEPT instruction */
7348 static int handle_invept(struct kvm_vcpu *vcpu)
7350 struct vcpu_vmx *vmx = to_vmx(vcpu);
7351 u32 vmx_instruction_info, types;
7354 struct x86_exception e;
7359 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7360 SECONDARY_EXEC_ENABLE_EPT) ||
7361 !(vmx->nested.nested_vmx_ept_caps & VMX_EPT_INVEPT_BIT)) {
7362 kvm_queue_exception(vcpu, UD_VECTOR);
7366 if (!nested_vmx_check_permission(vcpu))
7369 if (!kvm_read_cr0_bits(vcpu, X86_CR0_PE)) {
7370 kvm_queue_exception(vcpu, UD_VECTOR);
7374 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7375 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7377 types = (vmx->nested.nested_vmx_ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
7379 if (!(types & (1UL << type))) {
7380 nested_vmx_failValid(vcpu,
7381 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7385 /* According to the Intel VMX instruction reference, the memory
7386 * operand is read even if it isn't needed (e.g., for type==global)
7388 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7389 vmx_instruction_info, false, &gva))
7391 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
7392 sizeof(operand), &e)) {
7393 kvm_inject_page_fault(vcpu, &e);
7398 case VMX_EPT_EXTENT_GLOBAL:
7399 kvm_mmu_sync_roots(vcpu);
7400 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
7401 nested_vmx_succeed(vcpu);
7404 /* Trap single context invalidation invept calls */
7409 skip_emulated_instruction(vcpu);
7413 static int handle_invvpid(struct kvm_vcpu *vcpu)
7415 struct vcpu_vmx *vmx = to_vmx(vcpu);
7416 u32 vmx_instruction_info;
7417 unsigned long type, types;
7419 struct x86_exception e;
7422 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7423 SECONDARY_EXEC_ENABLE_VPID) ||
7424 !(vmx->nested.nested_vmx_vpid_caps & VMX_VPID_INVVPID_BIT)) {
7425 kvm_queue_exception(vcpu, UD_VECTOR);
7429 if (!nested_vmx_check_permission(vcpu))
7432 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7433 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7435 types = (vmx->nested.nested_vmx_vpid_caps >> 8) & 0x7;
7437 if (!(types & (1UL << type))) {
7438 nested_vmx_failValid(vcpu,
7439 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7443 /* according to the intel vmx instruction reference, the memory
7444 * operand is read even if it isn't needed (e.g., for type==global)
7446 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7447 vmx_instruction_info, false, &gva))
7449 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vpid,
7451 kvm_inject_page_fault(vcpu, &e);
7456 case VMX_VPID_EXTENT_ALL_CONTEXT:
7457 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->nested.vpid02);
7458 nested_vmx_succeed(vcpu);
7461 /* Trap single context invalidation invvpid calls */
7466 skip_emulated_instruction(vcpu);
7470 static int handle_pml_full(struct kvm_vcpu *vcpu)
7472 unsigned long exit_qualification;
7474 trace_kvm_pml_full(vcpu->vcpu_id);
7476 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7479 * PML buffer FULL happened while executing iret from NMI,
7480 * "blocked by NMI" bit has to be set before next VM entry.
7482 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
7483 cpu_has_virtual_nmis() &&
7484 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
7485 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
7486 GUEST_INTR_STATE_NMI);
7489 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
7490 * here.., and there's no userspace involvement needed for PML.
7495 static int handle_pcommit(struct kvm_vcpu *vcpu)
7497 /* we never catch pcommit instruct for L1 guest. */
7503 * The exit handlers return 1 if the exit was handled fully and guest execution
7504 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
7505 * to be done to userspace and return 0.
7507 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
7508 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
7509 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
7510 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
7511 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
7512 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
7513 [EXIT_REASON_CR_ACCESS] = handle_cr,
7514 [EXIT_REASON_DR_ACCESS] = handle_dr,
7515 [EXIT_REASON_CPUID] = handle_cpuid,
7516 [EXIT_REASON_MSR_READ] = handle_rdmsr,
7517 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
7518 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
7519 [EXIT_REASON_HLT] = handle_halt,
7520 [EXIT_REASON_INVD] = handle_invd,
7521 [EXIT_REASON_INVLPG] = handle_invlpg,
7522 [EXIT_REASON_RDPMC] = handle_rdpmc,
7523 [EXIT_REASON_VMCALL] = handle_vmcall,
7524 [EXIT_REASON_VMCLEAR] = handle_vmclear,
7525 [EXIT_REASON_VMLAUNCH] = handle_vmlaunch,
7526 [EXIT_REASON_VMPTRLD] = handle_vmptrld,
7527 [EXIT_REASON_VMPTRST] = handle_vmptrst,
7528 [EXIT_REASON_VMREAD] = handle_vmread,
7529 [EXIT_REASON_VMRESUME] = handle_vmresume,
7530 [EXIT_REASON_VMWRITE] = handle_vmwrite,
7531 [EXIT_REASON_VMOFF] = handle_vmoff,
7532 [EXIT_REASON_VMON] = handle_vmon,
7533 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
7534 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
7535 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
7536 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
7537 [EXIT_REASON_WBINVD] = handle_wbinvd,
7538 [EXIT_REASON_XSETBV] = handle_xsetbv,
7539 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
7540 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
7541 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
7542 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
7543 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
7544 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait,
7545 [EXIT_REASON_MONITOR_TRAP_FLAG] = handle_monitor_trap,
7546 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor,
7547 [EXIT_REASON_INVEPT] = handle_invept,
7548 [EXIT_REASON_INVVPID] = handle_invvpid,
7549 [EXIT_REASON_XSAVES] = handle_xsaves,
7550 [EXIT_REASON_XRSTORS] = handle_xrstors,
7551 [EXIT_REASON_PML_FULL] = handle_pml_full,
7552 [EXIT_REASON_PCOMMIT] = handle_pcommit,
7555 static const int kvm_vmx_max_exit_handlers =
7556 ARRAY_SIZE(kvm_vmx_exit_handlers);
7558 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
7559 struct vmcs12 *vmcs12)
7561 unsigned long exit_qualification;
7562 gpa_t bitmap, last_bitmap;
7567 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7568 return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
7570 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7572 port = exit_qualification >> 16;
7573 size = (exit_qualification & 7) + 1;
7575 last_bitmap = (gpa_t)-1;
7580 bitmap = vmcs12->io_bitmap_a;
7581 else if (port < 0x10000)
7582 bitmap = vmcs12->io_bitmap_b;
7585 bitmap += (port & 0x7fff) / 8;
7587 if (last_bitmap != bitmap)
7588 if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
7590 if (b & (1 << (port & 7)))
7595 last_bitmap = bitmap;
7602 * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
7603 * rather than handle it ourselves in L0. I.e., check whether L1 expressed
7604 * disinterest in the current event (read or write a specific MSR) by using an
7605 * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
7607 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
7608 struct vmcs12 *vmcs12, u32 exit_reason)
7610 u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
7613 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
7617 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
7618 * for the four combinations of read/write and low/high MSR numbers.
7619 * First we need to figure out which of the four to use:
7621 bitmap = vmcs12->msr_bitmap;
7622 if (exit_reason == EXIT_REASON_MSR_WRITE)
7624 if (msr_index >= 0xc0000000) {
7625 msr_index -= 0xc0000000;
7629 /* Then read the msr_index'th bit from this bitmap: */
7630 if (msr_index < 1024*8) {
7632 if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
7634 return 1 & (b >> (msr_index & 7));
7636 return true; /* let L1 handle the wrong parameter */
7640 * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
7641 * rather than handle it ourselves in L0. I.e., check if L1 wanted to
7642 * intercept (via guest_host_mask etc.) the current event.
7644 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
7645 struct vmcs12 *vmcs12)
7647 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7648 int cr = exit_qualification & 15;
7649 int reg = (exit_qualification >> 8) & 15;
7650 unsigned long val = kvm_register_readl(vcpu, reg);
7652 switch ((exit_qualification >> 4) & 3) {
7653 case 0: /* mov to cr */
7656 if (vmcs12->cr0_guest_host_mask &
7657 (val ^ vmcs12->cr0_read_shadow))
7661 if ((vmcs12->cr3_target_count >= 1 &&
7662 vmcs12->cr3_target_value0 == val) ||
7663 (vmcs12->cr3_target_count >= 2 &&
7664 vmcs12->cr3_target_value1 == val) ||
7665 (vmcs12->cr3_target_count >= 3 &&
7666 vmcs12->cr3_target_value2 == val) ||
7667 (vmcs12->cr3_target_count >= 4 &&
7668 vmcs12->cr3_target_value3 == val))
7670 if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
7674 if (vmcs12->cr4_guest_host_mask &
7675 (vmcs12->cr4_read_shadow ^ val))
7679 if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
7685 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
7686 (vmcs12->cr0_read_shadow & X86_CR0_TS))
7689 case 1: /* mov from cr */
7692 if (vmcs12->cpu_based_vm_exec_control &
7693 CPU_BASED_CR3_STORE_EXITING)
7697 if (vmcs12->cpu_based_vm_exec_control &
7698 CPU_BASED_CR8_STORE_EXITING)
7705 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
7706 * cr0. Other attempted changes are ignored, with no exit.
7708 if (vmcs12->cr0_guest_host_mask & 0xe &
7709 (val ^ vmcs12->cr0_read_shadow))
7711 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
7712 !(vmcs12->cr0_read_shadow & 0x1) &&
7721 * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
7722 * should handle it ourselves in L0 (and then continue L2). Only call this
7723 * when in is_guest_mode (L2).
7725 static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu)
7727 u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
7728 struct vcpu_vmx *vmx = to_vmx(vcpu);
7729 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7730 u32 exit_reason = vmx->exit_reason;
7732 trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
7733 vmcs_readl(EXIT_QUALIFICATION),
7734 vmx->idt_vectoring_info,
7736 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
7739 if (vmx->nested.nested_run_pending)
7742 if (unlikely(vmx->fail)) {
7743 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
7744 vmcs_read32(VM_INSTRUCTION_ERROR));
7748 switch (exit_reason) {
7749 case EXIT_REASON_EXCEPTION_NMI:
7750 if (!is_exception(intr_info))
7752 else if (is_page_fault(intr_info))
7754 else if (is_no_device(intr_info) &&
7755 !(vmcs12->guest_cr0 & X86_CR0_TS))
7757 else if (is_debug(intr_info) &&
7759 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
7761 else if (is_breakpoint(intr_info) &&
7762 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
7764 return vmcs12->exception_bitmap &
7765 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
7766 case EXIT_REASON_EXTERNAL_INTERRUPT:
7768 case EXIT_REASON_TRIPLE_FAULT:
7770 case EXIT_REASON_PENDING_INTERRUPT:
7771 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
7772 case EXIT_REASON_NMI_WINDOW:
7773 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
7774 case EXIT_REASON_TASK_SWITCH:
7776 case EXIT_REASON_CPUID:
7777 if (kvm_register_read(vcpu, VCPU_REGS_RAX) == 0xa)
7780 case EXIT_REASON_HLT:
7781 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
7782 case EXIT_REASON_INVD:
7784 case EXIT_REASON_INVLPG:
7785 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
7786 case EXIT_REASON_RDPMC:
7787 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
7788 case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
7789 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
7790 case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
7791 case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
7792 case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
7793 case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
7794 case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
7795 case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
7797 * VMX instructions trap unconditionally. This allows L1 to
7798 * emulate them for its L2 guest, i.e., allows 3-level nesting!
7801 case EXIT_REASON_CR_ACCESS:
7802 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
7803 case EXIT_REASON_DR_ACCESS:
7804 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
7805 case EXIT_REASON_IO_INSTRUCTION:
7806 return nested_vmx_exit_handled_io(vcpu, vmcs12);
7807 case EXIT_REASON_MSR_READ:
7808 case EXIT_REASON_MSR_WRITE:
7809 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
7810 case EXIT_REASON_INVALID_STATE:
7812 case EXIT_REASON_MWAIT_INSTRUCTION:
7813 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
7814 case EXIT_REASON_MONITOR_TRAP_FLAG:
7815 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_TRAP_FLAG);
7816 case EXIT_REASON_MONITOR_INSTRUCTION:
7817 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
7818 case EXIT_REASON_PAUSE_INSTRUCTION:
7819 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
7820 nested_cpu_has2(vmcs12,
7821 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
7822 case EXIT_REASON_MCE_DURING_VMENTRY:
7824 case EXIT_REASON_TPR_BELOW_THRESHOLD:
7825 return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
7826 case EXIT_REASON_APIC_ACCESS:
7827 return nested_cpu_has2(vmcs12,
7828 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
7829 case EXIT_REASON_APIC_WRITE:
7830 case EXIT_REASON_EOI_INDUCED:
7831 /* apic_write and eoi_induced should exit unconditionally. */
7833 case EXIT_REASON_EPT_VIOLATION:
7835 * L0 always deals with the EPT violation. If nested EPT is
7836 * used, and the nested mmu code discovers that the address is
7837 * missing in the guest EPT table (EPT12), the EPT violation
7838 * will be injected with nested_ept_inject_page_fault()
7841 case EXIT_REASON_EPT_MISCONFIG:
7843 * L2 never uses directly L1's EPT, but rather L0's own EPT
7844 * table (shadow on EPT) or a merged EPT table that L0 built
7845 * (EPT on EPT). So any problems with the structure of the
7846 * table is L0's fault.
7849 case EXIT_REASON_WBINVD:
7850 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
7851 case EXIT_REASON_XSETBV:
7853 case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
7855 * This should never happen, since it is not possible to
7856 * set XSS to a non-zero value---neither in L1 nor in L2.
7857 * If if it were, XSS would have to be checked against
7858 * the XSS exit bitmap in vmcs12.
7860 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
7861 case EXIT_REASON_PCOMMIT:
7862 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_PCOMMIT);
7868 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
7870 *info1 = vmcs_readl(EXIT_QUALIFICATION);
7871 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
7874 static int vmx_create_pml_buffer(struct vcpu_vmx *vmx)
7876 struct page *pml_pg;
7878 pml_pg = alloc_page(GFP_KERNEL | __GFP_ZERO);
7882 vmx->pml_pg = pml_pg;
7884 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
7885 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
7890 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
7893 __free_page(vmx->pml_pg);
7898 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
7900 struct vcpu_vmx *vmx = to_vmx(vcpu);
7904 pml_idx = vmcs_read16(GUEST_PML_INDEX);
7906 /* Do nothing if PML buffer is empty */
7907 if (pml_idx == (PML_ENTITY_NUM - 1))
7910 /* PML index always points to next available PML buffer entity */
7911 if (pml_idx >= PML_ENTITY_NUM)
7916 pml_buf = page_address(vmx->pml_pg);
7917 for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
7920 gpa = pml_buf[pml_idx];
7921 WARN_ON(gpa & (PAGE_SIZE - 1));
7922 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
7925 /* reset PML index */
7926 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
7930 * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
7931 * Called before reporting dirty_bitmap to userspace.
7933 static void kvm_flush_pml_buffers(struct kvm *kvm)
7936 struct kvm_vcpu *vcpu;
7938 * We only need to kick vcpu out of guest mode here, as PML buffer
7939 * is flushed at beginning of all VMEXITs, and it's obvious that only
7940 * vcpus running in guest are possible to have unflushed GPAs in PML
7943 kvm_for_each_vcpu(i, vcpu, kvm)
7944 kvm_vcpu_kick(vcpu);
7947 static void vmx_dump_sel(char *name, uint32_t sel)
7949 pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
7950 name, vmcs_read32(sel),
7951 vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
7952 vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
7953 vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
7956 static void vmx_dump_dtsel(char *name, uint32_t limit)
7958 pr_err("%s limit=0x%08x, base=0x%016lx\n",
7959 name, vmcs_read32(limit),
7960 vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
7963 static void dump_vmcs(void)
7965 u32 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
7966 u32 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
7967 u32 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
7968 u32 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
7969 u32 secondary_exec_control = 0;
7970 unsigned long cr4 = vmcs_readl(GUEST_CR4);
7971 u64 efer = vmcs_read64(GUEST_IA32_EFER);
7974 if (cpu_has_secondary_exec_ctrls())
7975 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7977 pr_err("*** Guest State ***\n");
7978 pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
7979 vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
7980 vmcs_readl(CR0_GUEST_HOST_MASK));
7981 pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
7982 cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
7983 pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
7984 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
7985 (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
7987 pr_err("PDPTR0 = 0x%016llx PDPTR1 = 0x%016llx\n",
7988 vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
7989 pr_err("PDPTR2 = 0x%016llx PDPTR3 = 0x%016llx\n",
7990 vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
7992 pr_err("RSP = 0x%016lx RIP = 0x%016lx\n",
7993 vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
7994 pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n",
7995 vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
7996 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
7997 vmcs_readl(GUEST_SYSENTER_ESP),
7998 vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
7999 vmx_dump_sel("CS: ", GUEST_CS_SELECTOR);
8000 vmx_dump_sel("DS: ", GUEST_DS_SELECTOR);
8001 vmx_dump_sel("SS: ", GUEST_SS_SELECTOR);
8002 vmx_dump_sel("ES: ", GUEST_ES_SELECTOR);
8003 vmx_dump_sel("FS: ", GUEST_FS_SELECTOR);
8004 vmx_dump_sel("GS: ", GUEST_GS_SELECTOR);
8005 vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
8006 vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
8007 vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
8008 vmx_dump_sel("TR: ", GUEST_TR_SELECTOR);
8009 if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
8010 (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
8011 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
8012 efer, vmcs_read64(GUEST_IA32_PAT));
8013 pr_err("DebugCtl = 0x%016llx DebugExceptions = 0x%016lx\n",
8014 vmcs_read64(GUEST_IA32_DEBUGCTL),
8015 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
8016 if (vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
8017 pr_err("PerfGlobCtl = 0x%016llx\n",
8018 vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
8019 if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
8020 pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
8021 pr_err("Interruptibility = %08x ActivityState = %08x\n",
8022 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
8023 vmcs_read32(GUEST_ACTIVITY_STATE));
8024 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
8025 pr_err("InterruptStatus = %04x\n",
8026 vmcs_read16(GUEST_INTR_STATUS));
8028 pr_err("*** Host State ***\n");
8029 pr_err("RIP = 0x%016lx RSP = 0x%016lx\n",
8030 vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
8031 pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
8032 vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
8033 vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
8034 vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
8035 vmcs_read16(HOST_TR_SELECTOR));
8036 pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
8037 vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
8038 vmcs_readl(HOST_TR_BASE));
8039 pr_err("GDTBase=%016lx IDTBase=%016lx\n",
8040 vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
8041 pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
8042 vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
8043 vmcs_readl(HOST_CR4));
8044 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8045 vmcs_readl(HOST_IA32_SYSENTER_ESP),
8046 vmcs_read32(HOST_IA32_SYSENTER_CS),
8047 vmcs_readl(HOST_IA32_SYSENTER_EIP));
8048 if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
8049 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
8050 vmcs_read64(HOST_IA32_EFER),
8051 vmcs_read64(HOST_IA32_PAT));
8052 if (vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
8053 pr_err("PerfGlobCtl = 0x%016llx\n",
8054 vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
8056 pr_err("*** Control State ***\n");
8057 pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
8058 pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
8059 pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
8060 pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
8061 vmcs_read32(EXCEPTION_BITMAP),
8062 vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
8063 vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
8064 pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
8065 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8066 vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
8067 vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
8068 pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
8069 vmcs_read32(VM_EXIT_INTR_INFO),
8070 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8071 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
8072 pr_err(" reason=%08x qualification=%016lx\n",
8073 vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
8074 pr_err("IDTVectoring: info=%08x errcode=%08x\n",
8075 vmcs_read32(IDT_VECTORING_INFO_FIELD),
8076 vmcs_read32(IDT_VECTORING_ERROR_CODE));
8077 pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
8078 if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
8079 pr_err("TSC Multiplier = 0x%016llx\n",
8080 vmcs_read64(TSC_MULTIPLIER));
8081 if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW)
8082 pr_err("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
8083 if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
8084 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
8085 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
8086 pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
8087 n = vmcs_read32(CR3_TARGET_COUNT);
8088 for (i = 0; i + 1 < n; i += 4)
8089 pr_err("CR3 target%u=%016lx target%u=%016lx\n",
8090 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
8091 i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
8093 pr_err("CR3 target%u=%016lx\n",
8094 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
8095 if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
8096 pr_err("PLE Gap=%08x Window=%08x\n",
8097 vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
8098 if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
8099 pr_err("Virtual processor ID = 0x%04x\n",
8100 vmcs_read16(VIRTUAL_PROCESSOR_ID));
8104 * The guest has exited. See if we can fix it or if we need userspace
8107 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
8109 struct vcpu_vmx *vmx = to_vmx(vcpu);
8110 u32 exit_reason = vmx->exit_reason;
8111 u32 vectoring_info = vmx->idt_vectoring_info;
8113 trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
8116 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
8117 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
8118 * querying dirty_bitmap, we only need to kick all vcpus out of guest
8119 * mode as if vcpus is in root mode, the PML buffer must has been
8123 vmx_flush_pml_buffer(vcpu);
8125 /* If guest state is invalid, start emulating */
8126 if (vmx->emulation_required)
8127 return handle_invalid_guest_state(vcpu);
8129 if (is_guest_mode(vcpu) && nested_vmx_exit_handled(vcpu)) {
8130 nested_vmx_vmexit(vcpu, exit_reason,
8131 vmcs_read32(VM_EXIT_INTR_INFO),
8132 vmcs_readl(EXIT_QUALIFICATION));
8136 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
8138 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8139 vcpu->run->fail_entry.hardware_entry_failure_reason
8144 if (unlikely(vmx->fail)) {
8145 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8146 vcpu->run->fail_entry.hardware_entry_failure_reason
8147 = vmcs_read32(VM_INSTRUCTION_ERROR);
8153 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
8154 * delivery event since it indicates guest is accessing MMIO.
8155 * The vm-exit can be triggered again after return to guest that
8156 * will cause infinite loop.
8158 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
8159 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
8160 exit_reason != EXIT_REASON_EPT_VIOLATION &&
8161 exit_reason != EXIT_REASON_TASK_SWITCH)) {
8162 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8163 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
8164 vcpu->run->internal.ndata = 2;
8165 vcpu->run->internal.data[0] = vectoring_info;
8166 vcpu->run->internal.data[1] = exit_reason;
8170 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked &&
8171 !(is_guest_mode(vcpu) && nested_cpu_has_virtual_nmis(
8172 get_vmcs12(vcpu))))) {
8173 if (vmx_interrupt_allowed(vcpu)) {
8174 vmx->soft_vnmi_blocked = 0;
8175 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
8176 vcpu->arch.nmi_pending) {
8178 * This CPU don't support us in finding the end of an
8179 * NMI-blocked window if the guest runs with IRQs
8180 * disabled. So we pull the trigger after 1 s of
8181 * futile waiting, but inform the user about this.
8183 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
8184 "state on VCPU %d after 1 s timeout\n",
8185 __func__, vcpu->vcpu_id);
8186 vmx->soft_vnmi_blocked = 0;
8190 if (exit_reason < kvm_vmx_max_exit_handlers
8191 && kvm_vmx_exit_handlers[exit_reason])
8192 return kvm_vmx_exit_handlers[exit_reason](vcpu);
8194 WARN_ONCE(1, "vmx: unexpected exit reason 0x%x\n", exit_reason);
8195 kvm_queue_exception(vcpu, UD_VECTOR);
8200 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
8202 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8204 if (is_guest_mode(vcpu) &&
8205 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
8208 if (irr == -1 || tpr < irr) {
8209 vmcs_write32(TPR_THRESHOLD, 0);
8213 vmcs_write32(TPR_THRESHOLD, irr);
8216 static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
8218 u32 sec_exec_control;
8221 * There is not point to enable virtualize x2apic without enable
8224 if (!cpu_has_vmx_virtualize_x2apic_mode() ||
8225 !kvm_vcpu_apicv_active(vcpu))
8228 if (!cpu_need_tpr_shadow(vcpu))
8231 sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8234 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8235 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8237 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8238 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8240 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
8242 vmx_set_msr_bitmap(vcpu);
8245 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
8247 struct vcpu_vmx *vmx = to_vmx(vcpu);
8250 * Currently we do not handle the nested case where L2 has an
8251 * APIC access page of its own; that page is still pinned.
8252 * Hence, we skip the case where the VCPU is in guest mode _and_
8253 * L1 prepared an APIC access page for L2.
8255 * For the case where L1 and L2 share the same APIC access page
8256 * (flexpriority=Y but SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES clear
8257 * in the vmcs12), this function will only update either the vmcs01
8258 * or the vmcs02. If the former, the vmcs02 will be updated by
8259 * prepare_vmcs02. If the latter, the vmcs01 will be updated in
8260 * the next L2->L1 exit.
8262 if (!is_guest_mode(vcpu) ||
8263 !nested_cpu_has2(vmx->nested.current_vmcs12,
8264 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
8265 vmcs_write64(APIC_ACCESS_ADDR, hpa);
8268 static void vmx_hwapic_isr_update(struct kvm *kvm, int isr)
8276 status = vmcs_read16(GUEST_INTR_STATUS);
8281 vmcs_write16(GUEST_INTR_STATUS, status);
8285 static void vmx_set_rvi(int vector)
8293 status = vmcs_read16(GUEST_INTR_STATUS);
8294 old = (u8)status & 0xff;
8295 if ((u8)vector != old) {
8297 status |= (u8)vector;
8298 vmcs_write16(GUEST_INTR_STATUS, status);
8302 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
8304 if (!is_guest_mode(vcpu)) {
8305 vmx_set_rvi(max_irr);
8313 * In guest mode. If a vmexit is needed, vmx_check_nested_events
8316 if (nested_exit_on_intr(vcpu))
8320 * Else, fall back to pre-APICv interrupt injection since L2
8321 * is run without virtual interrupt delivery.
8323 if (!kvm_event_needs_reinjection(vcpu) &&
8324 vmx_interrupt_allowed(vcpu)) {
8325 kvm_queue_interrupt(vcpu, max_irr, false);
8326 vmx_inject_irq(vcpu);
8330 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
8332 if (!kvm_vcpu_apicv_active(vcpu))
8335 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
8336 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
8337 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
8338 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
8341 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
8345 if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
8346 || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
8349 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8350 exit_intr_info = vmx->exit_intr_info;
8352 /* Handle machine checks before interrupts are enabled */
8353 if (is_machine_check(exit_intr_info))
8354 kvm_machine_check();
8356 /* We need to handle NMIs before interrupts are enabled */
8357 if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
8358 (exit_intr_info & INTR_INFO_VALID_MASK)) {
8359 kvm_before_handle_nmi(&vmx->vcpu);
8361 kvm_after_handle_nmi(&vmx->vcpu);
8365 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
8367 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8370 * If external interrupt exists, IF bit is set in rflags/eflags on the
8371 * interrupt stack frame, and interrupt will be enabled on a return
8372 * from interrupt handler.
8374 if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
8375 == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
8376 unsigned int vector;
8377 unsigned long entry;
8379 struct vcpu_vmx *vmx = to_vmx(vcpu);
8380 #ifdef CONFIG_X86_64
8384 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
8385 desc = (gate_desc *)vmx->host_idt_base + vector;
8386 entry = gate_offset(*desc);
8388 #ifdef CONFIG_X86_64
8389 "mov %%" _ASM_SP ", %[sp]\n\t"
8390 "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
8395 "orl $0x200, (%%" _ASM_SP ")\n\t"
8396 __ASM_SIZE(push) " $%c[cs]\n\t"
8397 "call *%[entry]\n\t"
8399 #ifdef CONFIG_X86_64
8404 [ss]"i"(__KERNEL_DS),
8405 [cs]"i"(__KERNEL_CS)
8411 static bool vmx_has_high_real_mode_segbase(void)
8413 return enable_unrestricted_guest || emulate_invalid_guest_state;
8416 static bool vmx_mpx_supported(void)
8418 return (vmcs_config.vmexit_ctrl & VM_EXIT_CLEAR_BNDCFGS) &&
8419 (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_BNDCFGS);
8422 static bool vmx_xsaves_supported(void)
8424 return vmcs_config.cpu_based_2nd_exec_ctrl &
8425 SECONDARY_EXEC_XSAVES;
8428 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
8433 bool idtv_info_valid;
8435 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8437 if (cpu_has_virtual_nmis()) {
8438 if (vmx->nmi_known_unmasked)
8441 * Can't use vmx->exit_intr_info since we're not sure what
8442 * the exit reason is.
8444 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8445 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
8446 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
8448 * SDM 3: 27.7.1.2 (September 2008)
8449 * Re-set bit "block by NMI" before VM entry if vmexit caused by
8450 * a guest IRET fault.
8451 * SDM 3: 23.2.2 (September 2008)
8452 * Bit 12 is undefined in any of the following cases:
8453 * If the VM exit sets the valid bit in the IDT-vectoring
8454 * information field.
8455 * If the VM exit is due to a double fault.
8457 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
8458 vector != DF_VECTOR && !idtv_info_valid)
8459 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
8460 GUEST_INTR_STATE_NMI);
8462 vmx->nmi_known_unmasked =
8463 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
8464 & GUEST_INTR_STATE_NMI);
8465 } else if (unlikely(vmx->soft_vnmi_blocked))
8466 vmx->vnmi_blocked_time +=
8467 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
8470 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
8471 u32 idt_vectoring_info,
8472 int instr_len_field,
8473 int error_code_field)
8477 bool idtv_info_valid;
8479 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8481 vcpu->arch.nmi_injected = false;
8482 kvm_clear_exception_queue(vcpu);
8483 kvm_clear_interrupt_queue(vcpu);
8485 if (!idtv_info_valid)
8488 kvm_make_request(KVM_REQ_EVENT, vcpu);
8490 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
8491 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
8494 case INTR_TYPE_NMI_INTR:
8495 vcpu->arch.nmi_injected = true;
8497 * SDM 3: 27.7.1.2 (September 2008)
8498 * Clear bit "block by NMI" before VM entry if a NMI
8501 vmx_set_nmi_mask(vcpu, false);
8503 case INTR_TYPE_SOFT_EXCEPTION:
8504 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8506 case INTR_TYPE_HARD_EXCEPTION:
8507 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
8508 u32 err = vmcs_read32(error_code_field);
8509 kvm_requeue_exception_e(vcpu, vector, err);
8511 kvm_requeue_exception(vcpu, vector);
8513 case INTR_TYPE_SOFT_INTR:
8514 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8516 case INTR_TYPE_EXT_INTR:
8517 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
8524 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
8526 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
8527 VM_EXIT_INSTRUCTION_LEN,
8528 IDT_VECTORING_ERROR_CODE);
8531 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
8533 __vmx_complete_interrupts(vcpu,
8534 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8535 VM_ENTRY_INSTRUCTION_LEN,
8536 VM_ENTRY_EXCEPTION_ERROR_CODE);
8538 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
8541 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
8544 struct perf_guest_switch_msr *msrs;
8546 msrs = perf_guest_get_msrs(&nr_msrs);
8551 for (i = 0; i < nr_msrs; i++)
8552 if (msrs[i].host == msrs[i].guest)
8553 clear_atomic_switch_msr(vmx, msrs[i].msr);
8555 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
8559 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
8561 struct vcpu_vmx *vmx = to_vmx(vcpu);
8562 unsigned long debugctlmsr, cr4;
8564 /* Record the guest's net vcpu time for enforced NMI injections. */
8565 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
8566 vmx->entry_time = ktime_get();
8568 /* Don't enter VMX if guest state is invalid, let the exit handler
8569 start emulation until we arrive back to a valid state */
8570 if (vmx->emulation_required)
8573 if (vmx->ple_window_dirty) {
8574 vmx->ple_window_dirty = false;
8575 vmcs_write32(PLE_WINDOW, vmx->ple_window);
8578 if (vmx->nested.sync_shadow_vmcs) {
8579 copy_vmcs12_to_shadow(vmx);
8580 vmx->nested.sync_shadow_vmcs = false;
8583 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
8584 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
8585 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
8586 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
8588 cr4 = cr4_read_shadow();
8589 if (unlikely(cr4 != vmx->host_state.vmcs_host_cr4)) {
8590 vmcs_writel(HOST_CR4, cr4);
8591 vmx->host_state.vmcs_host_cr4 = cr4;
8594 /* When single-stepping over STI and MOV SS, we must clear the
8595 * corresponding interruptibility bits in the guest state. Otherwise
8596 * vmentry fails as it then expects bit 14 (BS) in pending debug
8597 * exceptions being set, but that's not correct for the guest debugging
8599 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8600 vmx_set_interrupt_shadow(vcpu, 0);
8602 atomic_switch_perf_msrs(vmx);
8603 debugctlmsr = get_debugctlmsr();
8605 vmx->__launched = vmx->loaded_vmcs->launched;
8607 /* Store host registers */
8608 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
8609 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
8610 "push %%" _ASM_CX " \n\t"
8611 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
8613 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
8614 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
8616 /* Reload cr2 if changed */
8617 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
8618 "mov %%cr2, %%" _ASM_DX " \n\t"
8619 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
8621 "mov %%" _ASM_AX", %%cr2 \n\t"
8623 /* Check if vmlaunch of vmresume is needed */
8624 "cmpl $0, %c[launched](%0) \n\t"
8625 /* Load guest registers. Don't clobber flags. */
8626 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
8627 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
8628 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
8629 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
8630 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
8631 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
8632 #ifdef CONFIG_X86_64
8633 "mov %c[r8](%0), %%r8 \n\t"
8634 "mov %c[r9](%0), %%r9 \n\t"
8635 "mov %c[r10](%0), %%r10 \n\t"
8636 "mov %c[r11](%0), %%r11 \n\t"
8637 "mov %c[r12](%0), %%r12 \n\t"
8638 "mov %c[r13](%0), %%r13 \n\t"
8639 "mov %c[r14](%0), %%r14 \n\t"
8640 "mov %c[r15](%0), %%r15 \n\t"
8642 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
8644 /* Enter guest mode */
8646 __ex(ASM_VMX_VMLAUNCH) "\n\t"
8648 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
8650 /* Save guest registers, load host registers, keep flags */
8651 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
8653 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
8654 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
8655 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
8656 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
8657 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
8658 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
8659 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
8660 #ifdef CONFIG_X86_64
8661 "mov %%r8, %c[r8](%0) \n\t"
8662 "mov %%r9, %c[r9](%0) \n\t"
8663 "mov %%r10, %c[r10](%0) \n\t"
8664 "mov %%r11, %c[r11](%0) \n\t"
8665 "mov %%r12, %c[r12](%0) \n\t"
8666 "mov %%r13, %c[r13](%0) \n\t"
8667 "mov %%r14, %c[r14](%0) \n\t"
8668 "mov %%r15, %c[r15](%0) \n\t"
8670 "mov %%cr2, %%" _ASM_AX " \n\t"
8671 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
8673 "pop %%" _ASM_BP "; pop %%" _ASM_DX " \n\t"
8674 "setbe %c[fail](%0) \n\t"
8675 ".pushsection .rodata \n\t"
8676 ".global vmx_return \n\t"
8677 "vmx_return: " _ASM_PTR " 2b \n\t"
8679 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
8680 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
8681 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
8682 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
8683 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
8684 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
8685 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
8686 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
8687 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
8688 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
8689 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
8690 #ifdef CONFIG_X86_64
8691 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
8692 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
8693 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
8694 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
8695 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
8696 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
8697 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
8698 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
8700 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
8701 [wordsize]"i"(sizeof(ulong))
8703 #ifdef CONFIG_X86_64
8704 , "rax", "rbx", "rdi", "rsi"
8705 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
8707 , "eax", "ebx", "edi", "esi"
8711 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
8713 update_debugctlmsr(debugctlmsr);
8715 #ifndef CONFIG_X86_64
8717 * The sysexit path does not restore ds/es, so we must set them to
8718 * a reasonable value ourselves.
8720 * We can't defer this to vmx_load_host_state() since that function
8721 * may be executed in interrupt context, which saves and restore segments
8722 * around it, nullifying its effect.
8724 loadsegment(ds, __USER_DS);
8725 loadsegment(es, __USER_DS);
8728 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
8729 | (1 << VCPU_EXREG_RFLAGS)
8730 | (1 << VCPU_EXREG_PDPTR)
8731 | (1 << VCPU_EXREG_SEGMENTS)
8732 | (1 << VCPU_EXREG_CR3));
8733 vcpu->arch.regs_dirty = 0;
8735 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
8737 vmx->loaded_vmcs->launched = 1;
8739 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
8742 * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
8743 * we did not inject a still-pending event to L1 now because of
8744 * nested_run_pending, we need to re-enable this bit.
8746 if (vmx->nested.nested_run_pending)
8747 kvm_make_request(KVM_REQ_EVENT, vcpu);
8749 vmx->nested.nested_run_pending = 0;
8751 vmx_complete_atomic_exit(vmx);
8752 vmx_recover_nmi_blocking(vmx);
8753 vmx_complete_interrupts(vmx);
8756 static void vmx_load_vmcs01(struct kvm_vcpu *vcpu)
8758 struct vcpu_vmx *vmx = to_vmx(vcpu);
8761 if (vmx->loaded_vmcs == &vmx->vmcs01)
8765 vmx->loaded_vmcs = &vmx->vmcs01;
8767 vmx_vcpu_load(vcpu, cpu);
8772 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
8774 struct vcpu_vmx *vmx = to_vmx(vcpu);
8777 vmx_destroy_pml_buffer(vmx);
8778 free_vpid(vmx->vpid);
8779 leave_guest_mode(vcpu);
8780 vmx_load_vmcs01(vcpu);
8782 free_loaded_vmcs(vmx->loaded_vmcs);
8783 kfree(vmx->guest_msrs);
8784 kvm_vcpu_uninit(vcpu);
8785 kmem_cache_free(kvm_vcpu_cache, vmx);
8788 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
8791 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
8795 return ERR_PTR(-ENOMEM);
8797 vmx->vpid = allocate_vpid();
8799 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
8803 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
8804 BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
8808 if (!vmx->guest_msrs) {
8812 vmx->loaded_vmcs = &vmx->vmcs01;
8813 vmx->loaded_vmcs->vmcs = alloc_vmcs();
8814 if (!vmx->loaded_vmcs->vmcs)
8817 kvm_cpu_vmxon(__pa(per_cpu(vmxarea, raw_smp_processor_id())));
8818 loaded_vmcs_init(vmx->loaded_vmcs);
8823 vmx_vcpu_load(&vmx->vcpu, cpu);
8824 vmx->vcpu.cpu = cpu;
8825 err = vmx_vcpu_setup(vmx);
8826 vmx_vcpu_put(&vmx->vcpu);
8830 if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
8831 err = alloc_apic_access_page(kvm);
8837 if (!kvm->arch.ept_identity_map_addr)
8838 kvm->arch.ept_identity_map_addr =
8839 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
8840 err = init_rmode_identity_map(kvm);
8846 nested_vmx_setup_ctls_msrs(vmx);
8847 vmx->nested.vpid02 = allocate_vpid();
8850 vmx->nested.posted_intr_nv = -1;
8851 vmx->nested.current_vmptr = -1ull;
8852 vmx->nested.current_vmcs12 = NULL;
8855 * If PML is turned on, failure on enabling PML just results in failure
8856 * of creating the vcpu, therefore we can simplify PML logic (by
8857 * avoiding dealing with cases, such as enabling PML partially on vcpus
8858 * for the guest, etc.
8861 err = vmx_create_pml_buffer(vmx);
8869 free_vpid(vmx->nested.vpid02);
8870 free_loaded_vmcs(vmx->loaded_vmcs);
8872 kfree(vmx->guest_msrs);
8874 kvm_vcpu_uninit(&vmx->vcpu);
8876 free_vpid(vmx->vpid);
8877 kmem_cache_free(kvm_vcpu_cache, vmx);
8878 return ERR_PTR(err);
8881 static void __init vmx_check_processor_compat(void *rtn)
8883 struct vmcs_config vmcs_conf;
8886 if (setup_vmcs_config(&vmcs_conf) < 0)
8888 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
8889 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
8890 smp_processor_id());
8895 static int get_ept_level(void)
8897 return VMX_EPT_DEFAULT_GAW + 1;
8900 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
8905 /* For VT-d and EPT combination
8906 * 1. MMIO: always map as UC
8908 * a. VT-d without snooping control feature: can't guarantee the
8909 * result, try to trust guest.
8910 * b. VT-d with snooping control feature: snooping control feature of
8911 * VT-d engine can guarantee the cache correctness. Just set it
8912 * to WB to keep consistent with host. So the same as item 3.
8913 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
8914 * consistent with host MTRR
8917 cache = MTRR_TYPE_UNCACHABLE;
8921 if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
8922 ipat = VMX_EPT_IPAT_BIT;
8923 cache = MTRR_TYPE_WRBACK;
8927 if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
8928 ipat = VMX_EPT_IPAT_BIT;
8929 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
8930 cache = MTRR_TYPE_WRBACK;
8932 cache = MTRR_TYPE_UNCACHABLE;
8936 cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
8939 return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
8942 static int vmx_get_lpage_level(void)
8944 if (enable_ept && !cpu_has_vmx_ept_1g_page())
8945 return PT_DIRECTORY_LEVEL;
8947 /* For shadow and EPT supported 1GB page */
8948 return PT_PDPE_LEVEL;
8951 static void vmcs_set_secondary_exec_control(u32 new_ctl)
8954 * These bits in the secondary execution controls field
8955 * are dynamic, the others are mostly based on the hypervisor
8956 * architecture and the guest's CPUID. Do not touch the
8960 SECONDARY_EXEC_SHADOW_VMCS |
8961 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
8962 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8964 u32 cur_ctl = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8966 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
8967 (new_ctl & ~mask) | (cur_ctl & mask));
8970 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
8972 struct kvm_cpuid_entry2 *best;
8973 struct vcpu_vmx *vmx = to_vmx(vcpu);
8974 u32 secondary_exec_ctl = vmx_secondary_exec_control(vmx);
8976 if (vmx_rdtscp_supported()) {
8977 bool rdtscp_enabled = guest_cpuid_has_rdtscp(vcpu);
8978 if (!rdtscp_enabled)
8979 secondary_exec_ctl &= ~SECONDARY_EXEC_RDTSCP;
8983 vmx->nested.nested_vmx_secondary_ctls_high |=
8984 SECONDARY_EXEC_RDTSCP;
8986 vmx->nested.nested_vmx_secondary_ctls_high &=
8987 ~SECONDARY_EXEC_RDTSCP;
8991 /* Exposing INVPCID only when PCID is exposed */
8992 best = kvm_find_cpuid_entry(vcpu, 0x7, 0);
8993 if (vmx_invpcid_supported() &&
8994 (!best || !(best->ebx & bit(X86_FEATURE_INVPCID)) ||
8995 !guest_cpuid_has_pcid(vcpu))) {
8996 secondary_exec_ctl &= ~SECONDARY_EXEC_ENABLE_INVPCID;
8999 best->ebx &= ~bit(X86_FEATURE_INVPCID);
9002 if (cpu_has_secondary_exec_ctrls())
9003 vmcs_set_secondary_exec_control(secondary_exec_ctl);
9005 if (static_cpu_has(X86_FEATURE_PCOMMIT) && nested) {
9006 if (guest_cpuid_has_pcommit(vcpu))
9007 vmx->nested.nested_vmx_secondary_ctls_high |=
9008 SECONDARY_EXEC_PCOMMIT;
9010 vmx->nested.nested_vmx_secondary_ctls_high &=
9011 ~SECONDARY_EXEC_PCOMMIT;
9015 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
9017 if (func == 1 && nested)
9018 entry->ecx |= bit(X86_FEATURE_VMX);
9021 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
9022 struct x86_exception *fault)
9024 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9027 if (fault->error_code & PFERR_RSVD_MASK)
9028 exit_reason = EXIT_REASON_EPT_MISCONFIG;
9030 exit_reason = EXIT_REASON_EPT_VIOLATION;
9031 nested_vmx_vmexit(vcpu, exit_reason, 0, vcpu->arch.exit_qualification);
9032 vmcs12->guest_physical_address = fault->address;
9035 /* Callbacks for nested_ept_init_mmu_context: */
9037 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
9039 /* return the page table to be shadowed - in our case, EPT12 */
9040 return get_vmcs12(vcpu)->ept_pointer;
9043 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
9045 WARN_ON(mmu_is_nested(vcpu));
9046 kvm_init_shadow_ept_mmu(vcpu,
9047 to_vmx(vcpu)->nested.nested_vmx_ept_caps &
9048 VMX_EPT_EXECUTE_ONLY_BIT);
9049 vcpu->arch.mmu.set_cr3 = vmx_set_cr3;
9050 vcpu->arch.mmu.get_cr3 = nested_ept_get_cr3;
9051 vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
9053 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
9056 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
9058 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
9061 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
9064 bool inequality, bit;
9066 bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
9068 (error_code & vmcs12->page_fault_error_code_mask) !=
9069 vmcs12->page_fault_error_code_match;
9070 return inequality ^ bit;
9073 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
9074 struct x86_exception *fault)
9076 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9078 WARN_ON(!is_guest_mode(vcpu));
9080 if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code))
9081 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
9082 vmcs_read32(VM_EXIT_INTR_INFO),
9083 vmcs_readl(EXIT_QUALIFICATION));
9085 kvm_inject_page_fault(vcpu, fault);
9088 static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu,
9089 struct vmcs12 *vmcs12)
9091 struct vcpu_vmx *vmx = to_vmx(vcpu);
9092 int maxphyaddr = cpuid_maxphyaddr(vcpu);
9094 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
9095 if (!PAGE_ALIGNED(vmcs12->apic_access_addr) ||
9096 vmcs12->apic_access_addr >> maxphyaddr)
9100 * Translate L1 physical address to host physical
9101 * address for vmcs02. Keep the page pinned, so this
9102 * physical address remains valid. We keep a reference
9103 * to it so we can release it later.
9105 if (vmx->nested.apic_access_page) /* shouldn't happen */
9106 nested_release_page(vmx->nested.apic_access_page);
9107 vmx->nested.apic_access_page =
9108 nested_get_page(vcpu, vmcs12->apic_access_addr);
9111 if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
9112 if (!PAGE_ALIGNED(vmcs12->virtual_apic_page_addr) ||
9113 vmcs12->virtual_apic_page_addr >> maxphyaddr)
9116 if (vmx->nested.virtual_apic_page) /* shouldn't happen */
9117 nested_release_page(vmx->nested.virtual_apic_page);
9118 vmx->nested.virtual_apic_page =
9119 nested_get_page(vcpu, vmcs12->virtual_apic_page_addr);
9122 * Failing the vm entry is _not_ what the processor does
9123 * but it's basically the only possibility we have.
9124 * We could still enter the guest if CR8 load exits are
9125 * enabled, CR8 store exits are enabled, and virtualize APIC
9126 * access is disabled; in this case the processor would never
9127 * use the TPR shadow and we could simply clear the bit from
9128 * the execution control. But such a configuration is useless,
9129 * so let's keep the code simple.
9131 if (!vmx->nested.virtual_apic_page)
9135 if (nested_cpu_has_posted_intr(vmcs12)) {
9136 if (!IS_ALIGNED(vmcs12->posted_intr_desc_addr, 64) ||
9137 vmcs12->posted_intr_desc_addr >> maxphyaddr)
9140 if (vmx->nested.pi_desc_page) { /* shouldn't happen */
9141 kunmap(vmx->nested.pi_desc_page);
9142 nested_release_page(vmx->nested.pi_desc_page);
9144 vmx->nested.pi_desc_page =
9145 nested_get_page(vcpu, vmcs12->posted_intr_desc_addr);
9146 if (!vmx->nested.pi_desc_page)
9149 vmx->nested.pi_desc =
9150 (struct pi_desc *)kmap(vmx->nested.pi_desc_page);
9151 if (!vmx->nested.pi_desc) {
9152 nested_release_page_clean(vmx->nested.pi_desc_page);
9155 vmx->nested.pi_desc =
9156 (struct pi_desc *)((void *)vmx->nested.pi_desc +
9157 (unsigned long)(vmcs12->posted_intr_desc_addr &
9164 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
9166 u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
9167 struct vcpu_vmx *vmx = to_vmx(vcpu);
9169 if (vcpu->arch.virtual_tsc_khz == 0)
9172 /* Make sure short timeouts reliably trigger an immediate vmexit.
9173 * hrtimer_start does not guarantee this. */
9174 if (preemption_timeout <= 1) {
9175 vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
9179 preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
9180 preemption_timeout *= 1000000;
9181 do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
9182 hrtimer_start(&vmx->nested.preemption_timer,
9183 ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
9186 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
9187 struct vmcs12 *vmcs12)
9192 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
9195 if (vmcs12_read_any(vcpu, MSR_BITMAP, &addr)) {
9199 maxphyaddr = cpuid_maxphyaddr(vcpu);
9201 if (!PAGE_ALIGNED(vmcs12->msr_bitmap) ||
9202 ((addr + PAGE_SIZE) >> maxphyaddr))
9209 * Merge L0's and L1's MSR bitmap, return false to indicate that
9210 * we do not use the hardware.
9212 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
9213 struct vmcs12 *vmcs12)
9217 unsigned long *msr_bitmap;
9219 if (!nested_cpu_has_virt_x2apic_mode(vmcs12))
9222 page = nested_get_page(vcpu, vmcs12->msr_bitmap);
9227 msr_bitmap = (unsigned long *)kmap(page);
9229 nested_release_page_clean(page);
9234 if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
9235 if (nested_cpu_has_apic_reg_virt(vmcs12))
9236 for (msr = 0x800; msr <= 0x8ff; msr++)
9237 nested_vmx_disable_intercept_for_msr(
9239 vmx_msr_bitmap_nested,
9241 /* TPR is allowed */
9242 nested_vmx_disable_intercept_for_msr(msr_bitmap,
9243 vmx_msr_bitmap_nested,
9244 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
9245 MSR_TYPE_R | MSR_TYPE_W);
9246 if (nested_cpu_has_vid(vmcs12)) {
9247 /* EOI and self-IPI are allowed */
9248 nested_vmx_disable_intercept_for_msr(
9250 vmx_msr_bitmap_nested,
9251 APIC_BASE_MSR + (APIC_EOI >> 4),
9253 nested_vmx_disable_intercept_for_msr(
9255 vmx_msr_bitmap_nested,
9256 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
9261 * Enable reading intercept of all the x2apic
9262 * MSRs. We should not rely on vmcs12 to do any
9263 * optimizations here, it may have been modified
9266 for (msr = 0x800; msr <= 0x8ff; msr++)
9267 __vmx_enable_intercept_for_msr(
9268 vmx_msr_bitmap_nested,
9272 __vmx_enable_intercept_for_msr(
9273 vmx_msr_bitmap_nested,
9274 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
9276 __vmx_enable_intercept_for_msr(
9277 vmx_msr_bitmap_nested,
9278 APIC_BASE_MSR + (APIC_EOI >> 4),
9280 __vmx_enable_intercept_for_msr(
9281 vmx_msr_bitmap_nested,
9282 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
9286 nested_release_page_clean(page);
9291 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
9292 struct vmcs12 *vmcs12)
9294 if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9295 !nested_cpu_has_apic_reg_virt(vmcs12) &&
9296 !nested_cpu_has_vid(vmcs12) &&
9297 !nested_cpu_has_posted_intr(vmcs12))
9301 * If virtualize x2apic mode is enabled,
9302 * virtualize apic access must be disabled.
9304 if (nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9305 nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
9309 * If virtual interrupt delivery is enabled,
9310 * we must exit on external interrupts.
9312 if (nested_cpu_has_vid(vmcs12) &&
9313 !nested_exit_on_intr(vcpu))
9317 * bits 15:8 should be zero in posted_intr_nv,
9318 * the descriptor address has been already checked
9319 * in nested_get_vmcs12_pages.
9321 if (nested_cpu_has_posted_intr(vmcs12) &&
9322 (!nested_cpu_has_vid(vmcs12) ||
9323 !nested_exit_intr_ack_set(vcpu) ||
9324 vmcs12->posted_intr_nv & 0xff00))
9327 /* tpr shadow is needed by all apicv features. */
9328 if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
9334 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
9335 unsigned long count_field,
9336 unsigned long addr_field)
9341 if (vmcs12_read_any(vcpu, count_field, &count) ||
9342 vmcs12_read_any(vcpu, addr_field, &addr)) {
9348 maxphyaddr = cpuid_maxphyaddr(vcpu);
9349 if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
9350 (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr) {
9351 pr_warn_ratelimited(
9352 "nVMX: invalid MSR switch (0x%lx, %d, %llu, 0x%08llx)",
9353 addr_field, maxphyaddr, count, addr);
9359 static int nested_vmx_check_msr_switch_controls(struct kvm_vcpu *vcpu,
9360 struct vmcs12 *vmcs12)
9362 if (vmcs12->vm_exit_msr_load_count == 0 &&
9363 vmcs12->vm_exit_msr_store_count == 0 &&
9364 vmcs12->vm_entry_msr_load_count == 0)
9365 return 0; /* Fast path */
9366 if (nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_LOAD_COUNT,
9367 VM_EXIT_MSR_LOAD_ADDR) ||
9368 nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_STORE_COUNT,
9369 VM_EXIT_MSR_STORE_ADDR) ||
9370 nested_vmx_check_msr_switch(vcpu, VM_ENTRY_MSR_LOAD_COUNT,
9371 VM_ENTRY_MSR_LOAD_ADDR))
9376 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
9377 struct vmx_msr_entry *e)
9379 /* x2APIC MSR accesses are not allowed */
9380 if (vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8)
9382 if (e->index == MSR_IA32_UCODE_WRITE || /* SDM Table 35-2 */
9383 e->index == MSR_IA32_UCODE_REV)
9385 if (e->reserved != 0)
9390 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
9391 struct vmx_msr_entry *e)
9393 if (e->index == MSR_FS_BASE ||
9394 e->index == MSR_GS_BASE ||
9395 e->index == MSR_IA32_SMM_MONITOR_CTL || /* SMM is not supported */
9396 nested_vmx_msr_check_common(vcpu, e))
9401 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
9402 struct vmx_msr_entry *e)
9404 if (e->index == MSR_IA32_SMBASE || /* SMM is not supported */
9405 nested_vmx_msr_check_common(vcpu, e))
9411 * Load guest's/host's msr at nested entry/exit.
9412 * return 0 for success, entry index for failure.
9414 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9417 struct vmx_msr_entry e;
9418 struct msr_data msr;
9420 msr.host_initiated = false;
9421 for (i = 0; i < count; i++) {
9422 if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
9424 pr_warn_ratelimited(
9425 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9426 __func__, i, gpa + i * sizeof(e));
9429 if (nested_vmx_load_msr_check(vcpu, &e)) {
9430 pr_warn_ratelimited(
9431 "%s check failed (%u, 0x%x, 0x%x)\n",
9432 __func__, i, e.index, e.reserved);
9435 msr.index = e.index;
9437 if (kvm_set_msr(vcpu, &msr)) {
9438 pr_warn_ratelimited(
9439 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9440 __func__, i, e.index, e.value);
9449 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9452 struct vmx_msr_entry e;
9454 for (i = 0; i < count; i++) {
9455 struct msr_data msr_info;
9456 if (kvm_vcpu_read_guest(vcpu,
9457 gpa + i * sizeof(e),
9458 &e, 2 * sizeof(u32))) {
9459 pr_warn_ratelimited(
9460 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9461 __func__, i, gpa + i * sizeof(e));
9464 if (nested_vmx_store_msr_check(vcpu, &e)) {
9465 pr_warn_ratelimited(
9466 "%s check failed (%u, 0x%x, 0x%x)\n",
9467 __func__, i, e.index, e.reserved);
9470 msr_info.host_initiated = false;
9471 msr_info.index = e.index;
9472 if (kvm_get_msr(vcpu, &msr_info)) {
9473 pr_warn_ratelimited(
9474 "%s cannot read MSR (%u, 0x%x)\n",
9475 __func__, i, e.index);
9478 if (kvm_vcpu_write_guest(vcpu,
9479 gpa + i * sizeof(e) +
9480 offsetof(struct vmx_msr_entry, value),
9481 &msr_info.data, sizeof(msr_info.data))) {
9482 pr_warn_ratelimited(
9483 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9484 __func__, i, e.index, msr_info.data);
9492 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
9493 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
9494 * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
9495 * guest in a way that will both be appropriate to L1's requests, and our
9496 * needs. In addition to modifying the active vmcs (which is vmcs02), this
9497 * function also has additional necessary side-effects, like setting various
9498 * vcpu->arch fields.
9500 static void prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
9502 struct vcpu_vmx *vmx = to_vmx(vcpu);
9505 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
9506 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
9507 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
9508 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
9509 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
9510 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
9511 vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
9512 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
9513 vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
9514 vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
9515 vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
9516 vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
9517 vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
9518 vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
9519 vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
9520 vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
9521 vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
9522 vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
9523 vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
9524 vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
9525 vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
9526 vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
9527 vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
9528 vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
9529 vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
9530 vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
9531 vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
9532 vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
9533 vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
9534 vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
9535 vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
9536 vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
9537 vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
9538 vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
9539 vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
9540 vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
9542 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
9543 kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
9544 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
9546 kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
9547 vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
9549 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
9550 vmcs12->vm_entry_intr_info_field);
9551 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
9552 vmcs12->vm_entry_exception_error_code);
9553 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
9554 vmcs12->vm_entry_instruction_len);
9555 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
9556 vmcs12->guest_interruptibility_info);
9557 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
9558 vmx_set_rflags(vcpu, vmcs12->guest_rflags);
9559 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
9560 vmcs12->guest_pending_dbg_exceptions);
9561 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
9562 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
9564 if (nested_cpu_has_xsaves(vmcs12))
9565 vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
9566 vmcs_write64(VMCS_LINK_POINTER, -1ull);
9568 exec_control = vmcs12->pin_based_vm_exec_control;
9569 exec_control |= vmcs_config.pin_based_exec_ctrl;
9570 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
9572 if (nested_cpu_has_posted_intr(vmcs12)) {
9574 * Note that we use L0's vector here and in
9575 * vmx_deliver_nested_posted_interrupt.
9577 vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
9578 vmx->nested.pi_pending = false;
9579 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
9580 vmcs_write64(POSTED_INTR_DESC_ADDR,
9581 page_to_phys(vmx->nested.pi_desc_page) +
9582 (unsigned long)(vmcs12->posted_intr_desc_addr &
9585 exec_control &= ~PIN_BASED_POSTED_INTR;
9587 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
9589 vmx->nested.preemption_timer_expired = false;
9590 if (nested_cpu_has_preemption_timer(vmcs12))
9591 vmx_start_preemption_timer(vcpu);
9594 * Whether page-faults are trapped is determined by a combination of
9595 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
9596 * If enable_ept, L0 doesn't care about page faults and we should
9597 * set all of these to L1's desires. However, if !enable_ept, L0 does
9598 * care about (at least some) page faults, and because it is not easy
9599 * (if at all possible?) to merge L0 and L1's desires, we simply ask
9600 * to exit on each and every L2 page fault. This is done by setting
9601 * MASK=MATCH=0 and (see below) EB.PF=1.
9602 * Note that below we don't need special code to set EB.PF beyond the
9603 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
9604 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
9605 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
9607 * A problem with this approach (when !enable_ept) is that L1 may be
9608 * injected with more page faults than it asked for. This could have
9609 * caused problems, but in practice existing hypervisors don't care.
9610 * To fix this, we will need to emulate the PFEC checking (on the L1
9611 * page tables), using walk_addr(), when injecting PFs to L1.
9613 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
9614 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
9615 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
9616 enable_ept ? vmcs12->page_fault_error_code_match : 0);
9618 if (cpu_has_secondary_exec_ctrls()) {
9619 exec_control = vmx_secondary_exec_control(vmx);
9621 /* Take the following fields only from vmcs12 */
9622 exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
9623 SECONDARY_EXEC_RDTSCP |
9624 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
9625 SECONDARY_EXEC_APIC_REGISTER_VIRT |
9626 SECONDARY_EXEC_PCOMMIT);
9627 if (nested_cpu_has(vmcs12,
9628 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
9629 exec_control |= vmcs12->secondary_vm_exec_control;
9631 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) {
9633 * If translation failed, no matter: This feature asks
9634 * to exit when accessing the given address, and if it
9635 * can never be accessed, this feature won't do
9638 if (!vmx->nested.apic_access_page)
9640 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9642 vmcs_write64(APIC_ACCESS_ADDR,
9643 page_to_phys(vmx->nested.apic_access_page));
9644 } else if (!(nested_cpu_has_virt_x2apic_mode(vmcs12)) &&
9645 cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9647 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9648 kvm_vcpu_reload_apic_access_page(vcpu);
9651 if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
9652 vmcs_write64(EOI_EXIT_BITMAP0,
9653 vmcs12->eoi_exit_bitmap0);
9654 vmcs_write64(EOI_EXIT_BITMAP1,
9655 vmcs12->eoi_exit_bitmap1);
9656 vmcs_write64(EOI_EXIT_BITMAP2,
9657 vmcs12->eoi_exit_bitmap2);
9658 vmcs_write64(EOI_EXIT_BITMAP3,
9659 vmcs12->eoi_exit_bitmap3);
9660 vmcs_write16(GUEST_INTR_STATUS,
9661 vmcs12->guest_intr_status);
9664 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
9669 * Set host-state according to L0's settings (vmcs12 is irrelevant here)
9670 * Some constant fields are set here by vmx_set_constant_host_state().
9671 * Other fields are different per CPU, and will be set later when
9672 * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
9674 vmx_set_constant_host_state(vmx);
9677 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
9678 * entry, but only if the current (host) sp changed from the value
9679 * we wrote last (vmx->host_rsp). This cache is no longer relevant
9680 * if we switch vmcs, and rather than hold a separate cache per vmcs,
9681 * here we just force the write to happen on entry.
9685 exec_control = vmx_exec_control(vmx); /* L0's desires */
9686 exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
9687 exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
9688 exec_control &= ~CPU_BASED_TPR_SHADOW;
9689 exec_control |= vmcs12->cpu_based_vm_exec_control;
9691 if (exec_control & CPU_BASED_TPR_SHADOW) {
9692 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
9693 page_to_phys(vmx->nested.virtual_apic_page));
9694 vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
9697 if (cpu_has_vmx_msr_bitmap() &&
9698 exec_control & CPU_BASED_USE_MSR_BITMAPS) {
9699 nested_vmx_merge_msr_bitmap(vcpu, vmcs12);
9700 /* MSR_BITMAP will be set by following vmx_set_efer. */
9702 exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
9705 * Merging of IO bitmap not currently supported.
9706 * Rather, exit every time.
9708 exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
9709 exec_control |= CPU_BASED_UNCOND_IO_EXITING;
9711 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
9713 /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
9714 * bitwise-or of what L1 wants to trap for L2, and what we want to
9715 * trap. Note that CR0.TS also needs updating - we do this later.
9717 update_exception_bitmap(vcpu);
9718 vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
9719 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
9721 /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
9722 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
9723 * bits are further modified by vmx_set_efer() below.
9725 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
9727 /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
9728 * emulated by vmx_set_efer(), below.
9730 vm_entry_controls_init(vmx,
9731 (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER &
9732 ~VM_ENTRY_IA32E_MODE) |
9733 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
9735 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) {
9736 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
9737 vcpu->arch.pat = vmcs12->guest_ia32_pat;
9738 } else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
9739 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
9742 set_cr4_guest_host_mask(vmx);
9744 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)
9745 vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
9747 if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
9748 vmcs_write64(TSC_OFFSET,
9749 vmx->nested.vmcs01_tsc_offset + vmcs12->tsc_offset);
9751 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
9755 * There is no direct mapping between vpid02 and vpid12, the
9756 * vpid02 is per-vCPU for L0 and reused while the value of
9757 * vpid12 is changed w/ one invvpid during nested vmentry.
9758 * The vpid12 is allocated by L1 for L2, so it will not
9759 * influence global bitmap(for vpid01 and vpid02 allocation)
9760 * even if spawn a lot of nested vCPUs.
9762 if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02) {
9763 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
9764 if (vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
9765 vmx->nested.last_vpid = vmcs12->virtual_processor_id;
9766 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->nested.vpid02);
9769 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
9770 vmx_flush_tlb(vcpu);
9775 if (nested_cpu_has_ept(vmcs12)) {
9776 kvm_mmu_unload(vcpu);
9777 nested_ept_init_mmu_context(vcpu);
9780 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)
9781 vcpu->arch.efer = vmcs12->guest_ia32_efer;
9782 else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
9783 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
9785 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
9786 /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
9787 vmx_set_efer(vcpu, vcpu->arch.efer);
9790 * This sets GUEST_CR0 to vmcs12->guest_cr0, with possibly a modified
9791 * TS bit (for lazy fpu) and bits which we consider mandatory enabled.
9792 * The CR0_READ_SHADOW is what L2 should have expected to read given
9793 * the specifications by L1; It's not enough to take
9794 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
9795 * have more bits than L1 expected.
9797 vmx_set_cr0(vcpu, vmcs12->guest_cr0);
9798 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
9800 vmx_set_cr4(vcpu, vmcs12->guest_cr4);
9801 vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
9803 /* shadow page tables on either EPT or shadow page tables */
9804 kvm_set_cr3(vcpu, vmcs12->guest_cr3);
9805 kvm_mmu_reset_context(vcpu);
9808 vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
9811 * L1 may access the L2's PDPTR, so save them to construct vmcs12
9814 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
9815 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
9816 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
9817 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
9820 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
9821 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
9825 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
9826 * for running an L2 nested guest.
9828 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
9830 struct vmcs12 *vmcs12;
9831 struct vcpu_vmx *vmx = to_vmx(vcpu);
9833 struct loaded_vmcs *vmcs02;
9837 if (!nested_vmx_check_permission(vcpu) ||
9838 !nested_vmx_check_vmcs12(vcpu))
9841 skip_emulated_instruction(vcpu);
9842 vmcs12 = get_vmcs12(vcpu);
9844 if (enable_shadow_vmcs)
9845 copy_shadow_to_vmcs12(vmx);
9848 * The nested entry process starts with enforcing various prerequisites
9849 * on vmcs12 as required by the Intel SDM, and act appropriately when
9850 * they fail: As the SDM explains, some conditions should cause the
9851 * instruction to fail, while others will cause the instruction to seem
9852 * to succeed, but return an EXIT_REASON_INVALID_STATE.
9853 * To speed up the normal (success) code path, we should avoid checking
9854 * for misconfigurations which will anyway be caught by the processor
9855 * when using the merged vmcs02.
9857 if (vmcs12->launch_state == launch) {
9858 nested_vmx_failValid(vcpu,
9859 launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
9860 : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
9864 if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
9865 vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT) {
9866 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9870 if (!nested_get_vmcs12_pages(vcpu, vmcs12)) {
9871 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9875 if (nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12)) {
9876 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9880 if (nested_vmx_check_apicv_controls(vcpu, vmcs12)) {
9881 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9885 if (nested_vmx_check_msr_switch_controls(vcpu, vmcs12)) {
9886 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9890 if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
9891 vmx->nested.nested_vmx_true_procbased_ctls_low,
9892 vmx->nested.nested_vmx_procbased_ctls_high) ||
9893 !vmx_control_verify(vmcs12->secondary_vm_exec_control,
9894 vmx->nested.nested_vmx_secondary_ctls_low,
9895 vmx->nested.nested_vmx_secondary_ctls_high) ||
9896 !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
9897 vmx->nested.nested_vmx_pinbased_ctls_low,
9898 vmx->nested.nested_vmx_pinbased_ctls_high) ||
9899 !vmx_control_verify(vmcs12->vm_exit_controls,
9900 vmx->nested.nested_vmx_true_exit_ctls_low,
9901 vmx->nested.nested_vmx_exit_ctls_high) ||
9902 !vmx_control_verify(vmcs12->vm_entry_controls,
9903 vmx->nested.nested_vmx_true_entry_ctls_low,
9904 vmx->nested.nested_vmx_entry_ctls_high))
9906 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9910 if (((vmcs12->host_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
9911 ((vmcs12->host_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
9912 nested_vmx_failValid(vcpu,
9913 VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
9917 if (!nested_cr0_valid(vcpu, vmcs12->guest_cr0) ||
9918 ((vmcs12->guest_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
9919 nested_vmx_entry_failure(vcpu, vmcs12,
9920 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
9923 if (vmcs12->vmcs_link_pointer != -1ull) {
9924 nested_vmx_entry_failure(vcpu, vmcs12,
9925 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_VMCS_LINK_PTR);
9930 * If the load IA32_EFER VM-entry control is 1, the following checks
9931 * are performed on the field for the IA32_EFER MSR:
9932 * - Bits reserved in the IA32_EFER MSR must be 0.
9933 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
9934 * the IA-32e mode guest VM-exit control. It must also be identical
9935 * to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
9938 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER) {
9939 ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
9940 if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
9941 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
9942 ((vmcs12->guest_cr0 & X86_CR0_PG) &&
9943 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))) {
9944 nested_vmx_entry_failure(vcpu, vmcs12,
9945 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
9951 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
9952 * IA32_EFER MSR must be 0 in the field for that register. In addition,
9953 * the values of the LMA and LME bits in the field must each be that of
9954 * the host address-space size VM-exit control.
9956 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
9957 ia32e = (vmcs12->vm_exit_controls &
9958 VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
9959 if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
9960 ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
9961 ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)) {
9962 nested_vmx_entry_failure(vcpu, vmcs12,
9963 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
9969 * We're finally done with prerequisite checking, and can start with
9973 vmcs02 = nested_get_current_vmcs02(vmx);
9977 enter_guest_mode(vcpu);
9979 vmx->nested.vmcs01_tsc_offset = vmcs_read64(TSC_OFFSET);
9981 if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
9982 vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
9985 vmx->loaded_vmcs = vmcs02;
9987 vmx_vcpu_load(vcpu, cpu);
9991 vmx_segment_cache_clear(vmx);
9993 prepare_vmcs02(vcpu, vmcs12);
9995 msr_entry_idx = nested_vmx_load_msr(vcpu,
9996 vmcs12->vm_entry_msr_load_addr,
9997 vmcs12->vm_entry_msr_load_count);
9998 if (msr_entry_idx) {
9999 leave_guest_mode(vcpu);
10000 vmx_load_vmcs01(vcpu);
10001 nested_vmx_entry_failure(vcpu, vmcs12,
10002 EXIT_REASON_MSR_LOAD_FAIL, msr_entry_idx);
10006 vmcs12->launch_state = 1;
10008 if (vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT)
10009 return kvm_vcpu_halt(vcpu);
10011 vmx->nested.nested_run_pending = 1;
10014 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
10015 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
10016 * returned as far as L1 is concerned. It will only return (and set
10017 * the success flag) when L2 exits (see nested_vmx_vmexit()).
10023 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
10024 * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
10025 * This function returns the new value we should put in vmcs12.guest_cr0.
10026 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
10027 * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
10028 * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
10029 * didn't trap the bit, because if L1 did, so would L0).
10030 * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have
10031 * been modified by L2, and L1 knows it. So just leave the old value of
10032 * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
10033 * isn't relevant, because if L0 traps this bit it can set it to anything.
10034 * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
10035 * changed these bits, and therefore they need to be updated, but L0
10036 * didn't necessarily allow them to be changed in GUEST_CR0 - and rather
10037 * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
10039 static inline unsigned long
10040 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10043 /*1*/ (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
10044 /*2*/ (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
10045 /*3*/ (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
10046 vcpu->arch.cr0_guest_owned_bits));
10049 static inline unsigned long
10050 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10053 /*1*/ (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
10054 /*2*/ (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
10055 /*3*/ (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
10056 vcpu->arch.cr4_guest_owned_bits));
10059 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
10060 struct vmcs12 *vmcs12)
10065 if (vcpu->arch.exception.pending && vcpu->arch.exception.reinject) {
10066 nr = vcpu->arch.exception.nr;
10067 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
10069 if (kvm_exception_is_soft(nr)) {
10070 vmcs12->vm_exit_instruction_len =
10071 vcpu->arch.event_exit_inst_len;
10072 idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
10074 idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
10076 if (vcpu->arch.exception.has_error_code) {
10077 idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
10078 vmcs12->idt_vectoring_error_code =
10079 vcpu->arch.exception.error_code;
10082 vmcs12->idt_vectoring_info_field = idt_vectoring;
10083 } else if (vcpu->arch.nmi_injected) {
10084 vmcs12->idt_vectoring_info_field =
10085 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
10086 } else if (vcpu->arch.interrupt.pending) {
10087 nr = vcpu->arch.interrupt.nr;
10088 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
10090 if (vcpu->arch.interrupt.soft) {
10091 idt_vectoring |= INTR_TYPE_SOFT_INTR;
10092 vmcs12->vm_entry_instruction_len =
10093 vcpu->arch.event_exit_inst_len;
10095 idt_vectoring |= INTR_TYPE_EXT_INTR;
10097 vmcs12->idt_vectoring_info_field = idt_vectoring;
10101 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
10103 struct vcpu_vmx *vmx = to_vmx(vcpu);
10105 if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
10106 vmx->nested.preemption_timer_expired) {
10107 if (vmx->nested.nested_run_pending)
10109 nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
10113 if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
10114 if (vmx->nested.nested_run_pending ||
10115 vcpu->arch.interrupt.pending)
10117 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
10118 NMI_VECTOR | INTR_TYPE_NMI_INTR |
10119 INTR_INFO_VALID_MASK, 0);
10121 * The NMI-triggered VM exit counts as injection:
10122 * clear this one and block further NMIs.
10124 vcpu->arch.nmi_pending = 0;
10125 vmx_set_nmi_mask(vcpu, true);
10129 if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
10130 nested_exit_on_intr(vcpu)) {
10131 if (vmx->nested.nested_run_pending)
10133 nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
10137 return vmx_complete_nested_posted_interrupt(vcpu);
10140 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
10142 ktime_t remaining =
10143 hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
10146 if (ktime_to_ns(remaining) <= 0)
10149 value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
10150 do_div(value, 1000000);
10151 return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
10155 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
10156 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
10157 * and this function updates it to reflect the changes to the guest state while
10158 * L2 was running (and perhaps made some exits which were handled directly by L0
10159 * without going back to L1), and to reflect the exit reason.
10160 * Note that we do not have to copy here all VMCS fields, just those that
10161 * could have changed by the L2 guest or the exit - i.e., the guest-state and
10162 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
10163 * which already writes to vmcs12 directly.
10165 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10166 u32 exit_reason, u32 exit_intr_info,
10167 unsigned long exit_qualification)
10169 /* update guest state fields: */
10170 vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
10171 vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
10173 vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
10174 vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
10175 vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
10177 vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
10178 vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
10179 vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
10180 vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
10181 vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
10182 vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
10183 vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
10184 vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
10185 vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
10186 vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
10187 vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
10188 vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
10189 vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
10190 vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
10191 vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
10192 vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
10193 vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
10194 vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
10195 vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
10196 vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
10197 vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
10198 vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
10199 vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
10200 vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
10201 vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
10202 vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
10203 vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
10204 vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
10205 vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
10206 vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
10207 vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
10208 vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
10209 vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
10210 vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
10211 vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
10212 vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
10214 vmcs12->guest_interruptibility_info =
10215 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
10216 vmcs12->guest_pending_dbg_exceptions =
10217 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
10218 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
10219 vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
10221 vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
10223 if (nested_cpu_has_preemption_timer(vmcs12)) {
10224 if (vmcs12->vm_exit_controls &
10225 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
10226 vmcs12->vmx_preemption_timer_value =
10227 vmx_get_preemption_timer_value(vcpu);
10228 hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
10232 * In some cases (usually, nested EPT), L2 is allowed to change its
10233 * own CR3 without exiting. If it has changed it, we must keep it.
10234 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
10235 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
10237 * Additionally, restore L2's PDPTR to vmcs12.
10240 vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
10241 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
10242 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
10243 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
10244 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
10247 if (nested_cpu_has_vid(vmcs12))
10248 vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
10250 vmcs12->vm_entry_controls =
10251 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
10252 (vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
10254 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) {
10255 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
10256 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
10259 /* TODO: These cannot have changed unless we have MSR bitmaps and
10260 * the relevant bit asks not to trap the change */
10261 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
10262 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
10263 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
10264 vmcs12->guest_ia32_efer = vcpu->arch.efer;
10265 vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
10266 vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
10267 vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
10268 if (vmx_mpx_supported())
10269 vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
10270 if (nested_cpu_has_xsaves(vmcs12))
10271 vmcs12->xss_exit_bitmap = vmcs_read64(XSS_EXIT_BITMAP);
10273 /* update exit information fields: */
10275 vmcs12->vm_exit_reason = exit_reason;
10276 vmcs12->exit_qualification = exit_qualification;
10278 vmcs12->vm_exit_intr_info = exit_intr_info;
10279 if ((vmcs12->vm_exit_intr_info &
10280 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
10281 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK))
10282 vmcs12->vm_exit_intr_error_code =
10283 vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
10284 vmcs12->idt_vectoring_info_field = 0;
10285 vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
10286 vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
10288 if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
10289 /* vm_entry_intr_info_field is cleared on exit. Emulate this
10290 * instead of reading the real value. */
10291 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
10294 * Transfer the event that L0 or L1 may wanted to inject into
10295 * L2 to IDT_VECTORING_INFO_FIELD.
10297 vmcs12_save_pending_event(vcpu, vmcs12);
10301 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
10302 * preserved above and would only end up incorrectly in L1.
10304 vcpu->arch.nmi_injected = false;
10305 kvm_clear_exception_queue(vcpu);
10306 kvm_clear_interrupt_queue(vcpu);
10310 * A part of what we need to when the nested L2 guest exits and we want to
10311 * run its L1 parent, is to reset L1's guest state to the host state specified
10313 * This function is to be called not only on normal nested exit, but also on
10314 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
10315 * Failures During or After Loading Guest State").
10316 * This function should be called when the active VMCS is L1's (vmcs01).
10318 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
10319 struct vmcs12 *vmcs12)
10321 struct kvm_segment seg;
10323 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
10324 vcpu->arch.efer = vmcs12->host_ia32_efer;
10325 else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
10326 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10328 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10329 vmx_set_efer(vcpu, vcpu->arch.efer);
10331 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
10332 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
10333 vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
10335 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
10336 * actually changed, because it depends on the current state of
10337 * fpu_active (which may have changed).
10338 * Note that vmx_set_cr0 refers to efer set above.
10340 vmx_set_cr0(vcpu, vmcs12->host_cr0);
10342 * If we did fpu_activate()/fpu_deactivate() during L2's run, we need
10343 * to apply the same changes to L1's vmcs. We just set cr0 correctly,
10344 * but we also need to update cr0_guest_host_mask and exception_bitmap.
10346 update_exception_bitmap(vcpu);
10347 vcpu->arch.cr0_guest_owned_bits = (vcpu->fpu_active ? X86_CR0_TS : 0);
10348 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
10351 * Note that CR4_GUEST_HOST_MASK is already set in the original vmcs01
10352 * (KVM doesn't change it)- no reason to call set_cr4_guest_host_mask();
10354 vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
10355 kvm_set_cr4(vcpu, vmcs12->host_cr4);
10357 nested_ept_uninit_mmu_context(vcpu);
10359 kvm_set_cr3(vcpu, vmcs12->host_cr3);
10360 kvm_mmu_reset_context(vcpu);
10363 vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
10367 * Trivially support vpid by letting L2s share their parent
10368 * L1's vpid. TODO: move to a more elaborate solution, giving
10369 * each L2 its own vpid and exposing the vpid feature to L1.
10371 vmx_flush_tlb(vcpu);
10375 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
10376 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
10377 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
10378 vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
10379 vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
10381 /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1. */
10382 if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
10383 vmcs_write64(GUEST_BNDCFGS, 0);
10385 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
10386 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
10387 vcpu->arch.pat = vmcs12->host_ia32_pat;
10389 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
10390 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
10391 vmcs12->host_ia32_perf_global_ctrl);
10393 /* Set L1 segment info according to Intel SDM
10394 27.5.2 Loading Host Segment and Descriptor-Table Registers */
10395 seg = (struct kvm_segment) {
10397 .limit = 0xFFFFFFFF,
10398 .selector = vmcs12->host_cs_selector,
10404 if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
10408 vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
10409 seg = (struct kvm_segment) {
10411 .limit = 0xFFFFFFFF,
10418 seg.selector = vmcs12->host_ds_selector;
10419 vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
10420 seg.selector = vmcs12->host_es_selector;
10421 vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
10422 seg.selector = vmcs12->host_ss_selector;
10423 vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
10424 seg.selector = vmcs12->host_fs_selector;
10425 seg.base = vmcs12->host_fs_base;
10426 vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
10427 seg.selector = vmcs12->host_gs_selector;
10428 seg.base = vmcs12->host_gs_base;
10429 vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
10430 seg = (struct kvm_segment) {
10431 .base = vmcs12->host_tr_base,
10433 .selector = vmcs12->host_tr_selector,
10437 vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
10439 kvm_set_dr(vcpu, 7, 0x400);
10440 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
10442 if (cpu_has_vmx_msr_bitmap())
10443 vmx_set_msr_bitmap(vcpu);
10445 if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
10446 vmcs12->vm_exit_msr_load_count))
10447 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
10451 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
10452 * and modify vmcs12 to make it see what it would expect to see there if
10453 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
10455 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
10456 u32 exit_intr_info,
10457 unsigned long exit_qualification)
10459 struct vcpu_vmx *vmx = to_vmx(vcpu);
10460 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
10462 /* trying to cancel vmlaunch/vmresume is a bug */
10463 WARN_ON_ONCE(vmx->nested.nested_run_pending);
10465 leave_guest_mode(vcpu);
10466 prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
10467 exit_qualification);
10469 if (nested_vmx_store_msr(vcpu, vmcs12->vm_exit_msr_store_addr,
10470 vmcs12->vm_exit_msr_store_count))
10471 nested_vmx_abort(vcpu, VMX_ABORT_SAVE_GUEST_MSR_FAIL);
10473 vmx_load_vmcs01(vcpu);
10475 if ((exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
10476 && nested_exit_intr_ack_set(vcpu)) {
10477 int irq = kvm_cpu_get_interrupt(vcpu);
10479 vmcs12->vm_exit_intr_info = irq |
10480 INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
10483 trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
10484 vmcs12->exit_qualification,
10485 vmcs12->idt_vectoring_info_field,
10486 vmcs12->vm_exit_intr_info,
10487 vmcs12->vm_exit_intr_error_code,
10490 vm_entry_controls_init(vmx, vmcs_read32(VM_ENTRY_CONTROLS));
10491 vm_exit_controls_init(vmx, vmcs_read32(VM_EXIT_CONTROLS));
10492 vmx_segment_cache_clear(vmx);
10494 /* if no vmcs02 cache requested, remove the one we used */
10495 if (VMCS02_POOL_SIZE == 0)
10496 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
10498 load_vmcs12_host_state(vcpu, vmcs12);
10500 /* Update TSC_OFFSET if TSC was changed while L2 ran */
10501 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
10503 /* This is needed for same reason as it was needed in prepare_vmcs02 */
10506 /* Unpin physical memory we referred to in vmcs02 */
10507 if (vmx->nested.apic_access_page) {
10508 nested_release_page(vmx->nested.apic_access_page);
10509 vmx->nested.apic_access_page = NULL;
10511 if (vmx->nested.virtual_apic_page) {
10512 nested_release_page(vmx->nested.virtual_apic_page);
10513 vmx->nested.virtual_apic_page = NULL;
10515 if (vmx->nested.pi_desc_page) {
10516 kunmap(vmx->nested.pi_desc_page);
10517 nested_release_page(vmx->nested.pi_desc_page);
10518 vmx->nested.pi_desc_page = NULL;
10519 vmx->nested.pi_desc = NULL;
10523 * We are now running in L2, mmu_notifier will force to reload the
10524 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
10526 kvm_vcpu_reload_apic_access_page(vcpu);
10529 * Exiting from L2 to L1, we're now back to L1 which thinks it just
10530 * finished a VMLAUNCH or VMRESUME instruction, so we need to set the
10531 * success or failure flag accordingly.
10533 if (unlikely(vmx->fail)) {
10535 nested_vmx_failValid(vcpu, vmcs_read32(VM_INSTRUCTION_ERROR));
10537 nested_vmx_succeed(vcpu);
10538 if (enable_shadow_vmcs)
10539 vmx->nested.sync_shadow_vmcs = true;
10541 /* in case we halted in L2 */
10542 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10546 * Forcibly leave nested mode in order to be able to reset the VCPU later on.
10548 static void vmx_leave_nested(struct kvm_vcpu *vcpu)
10550 if (is_guest_mode(vcpu))
10551 nested_vmx_vmexit(vcpu, -1, 0, 0);
10552 free_nested(to_vmx(vcpu));
10556 * L1's failure to enter L2 is a subset of a normal exit, as explained in
10557 * 23.7 "VM-entry failures during or after loading guest state" (this also
10558 * lists the acceptable exit-reason and exit-qualification parameters).
10559 * It should only be called before L2 actually succeeded to run, and when
10560 * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
10562 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
10563 struct vmcs12 *vmcs12,
10564 u32 reason, unsigned long qualification)
10566 load_vmcs12_host_state(vcpu, vmcs12);
10567 vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
10568 vmcs12->exit_qualification = qualification;
10569 nested_vmx_succeed(vcpu);
10570 if (enable_shadow_vmcs)
10571 to_vmx(vcpu)->nested.sync_shadow_vmcs = true;
10574 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
10575 struct x86_instruction_info *info,
10576 enum x86_intercept_stage stage)
10578 return X86EMUL_CONTINUE;
10581 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
10584 shrink_ple_window(vcpu);
10587 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
10588 struct kvm_memory_slot *slot)
10590 kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
10591 kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
10594 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
10595 struct kvm_memory_slot *slot)
10597 kvm_mmu_slot_set_dirty(kvm, slot);
10600 static void vmx_flush_log_dirty(struct kvm *kvm)
10602 kvm_flush_pml_buffers(kvm);
10605 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
10606 struct kvm_memory_slot *memslot,
10607 gfn_t offset, unsigned long mask)
10609 kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
10613 * This routine does the following things for vCPU which is going
10614 * to be blocked if VT-d PI is enabled.
10615 * - Store the vCPU to the wakeup list, so when interrupts happen
10616 * we can find the right vCPU to wake up.
10617 * - Change the Posted-interrupt descriptor as below:
10618 * 'NDST' <-- vcpu->pre_pcpu
10619 * 'NV' <-- POSTED_INTR_WAKEUP_VECTOR
10620 * - If 'ON' is set during this process, which means at least one
10621 * interrupt is posted for this vCPU, we cannot block it, in
10622 * this case, return 1, otherwise, return 0.
10625 static int vmx_pre_block(struct kvm_vcpu *vcpu)
10627 unsigned long flags;
10629 struct pi_desc old, new;
10630 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
10632 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
10633 !irq_remapping_cap(IRQ_POSTING_CAP))
10636 vcpu->pre_pcpu = vcpu->cpu;
10637 spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
10638 vcpu->pre_pcpu), flags);
10639 list_add_tail(&vcpu->blocked_vcpu_list,
10640 &per_cpu(blocked_vcpu_on_cpu,
10642 spin_unlock_irqrestore(&per_cpu(blocked_vcpu_on_cpu_lock,
10643 vcpu->pre_pcpu), flags);
10646 old.control = new.control = pi_desc->control;
10649 * We should not block the vCPU if
10650 * an interrupt is posted for it.
10652 if (pi_test_on(pi_desc) == 1) {
10653 spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
10654 vcpu->pre_pcpu), flags);
10655 list_del(&vcpu->blocked_vcpu_list);
10656 spin_unlock_irqrestore(
10657 &per_cpu(blocked_vcpu_on_cpu_lock,
10658 vcpu->pre_pcpu), flags);
10659 vcpu->pre_pcpu = -1;
10664 WARN((pi_desc->sn == 1),
10665 "Warning: SN field of posted-interrupts "
10666 "is set before blocking\n");
10669 * Since vCPU can be preempted during this process,
10670 * vcpu->cpu could be different with pre_pcpu, we
10671 * need to set pre_pcpu as the destination of wakeup
10672 * notification event, then we can find the right vCPU
10673 * to wakeup in wakeup handler if interrupts happen
10674 * when the vCPU is in blocked state.
10676 dest = cpu_physical_id(vcpu->pre_pcpu);
10678 if (x2apic_enabled())
10681 new.ndst = (dest << 8) & 0xFF00;
10683 /* set 'NV' to 'wakeup vector' */
10684 new.nv = POSTED_INTR_WAKEUP_VECTOR;
10685 } while (cmpxchg(&pi_desc->control, old.control,
10686 new.control) != old.control);
10691 static void vmx_post_block(struct kvm_vcpu *vcpu)
10693 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
10694 struct pi_desc old, new;
10696 unsigned long flags;
10698 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
10699 !irq_remapping_cap(IRQ_POSTING_CAP))
10703 old.control = new.control = pi_desc->control;
10705 dest = cpu_physical_id(vcpu->cpu);
10707 if (x2apic_enabled())
10710 new.ndst = (dest << 8) & 0xFF00;
10712 /* Allow posting non-urgent interrupts */
10715 /* set 'NV' to 'notification vector' */
10716 new.nv = POSTED_INTR_VECTOR;
10717 } while (cmpxchg(&pi_desc->control, old.control,
10718 new.control) != old.control);
10720 if(vcpu->pre_pcpu != -1) {
10722 &per_cpu(blocked_vcpu_on_cpu_lock,
10723 vcpu->pre_pcpu), flags);
10724 list_del(&vcpu->blocked_vcpu_list);
10725 spin_unlock_irqrestore(
10726 &per_cpu(blocked_vcpu_on_cpu_lock,
10727 vcpu->pre_pcpu), flags);
10728 vcpu->pre_pcpu = -1;
10733 * vmx_update_pi_irte - set IRTE for Posted-Interrupts
10736 * @host_irq: host irq of the interrupt
10737 * @guest_irq: gsi of the interrupt
10738 * @set: set or unset PI
10739 * returns 0 on success, < 0 on failure
10741 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
10742 uint32_t guest_irq, bool set)
10744 struct kvm_kernel_irq_routing_entry *e;
10745 struct kvm_irq_routing_table *irq_rt;
10746 struct kvm_lapic_irq irq;
10747 struct kvm_vcpu *vcpu;
10748 struct vcpu_data vcpu_info;
10749 int idx, ret = -EINVAL;
10751 if (!kvm_arch_has_assigned_device(kvm) ||
10752 !irq_remapping_cap(IRQ_POSTING_CAP))
10755 idx = srcu_read_lock(&kvm->irq_srcu);
10756 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
10757 BUG_ON(guest_irq >= irq_rt->nr_rt_entries);
10759 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
10760 if (e->type != KVM_IRQ_ROUTING_MSI)
10763 * VT-d PI cannot support posting multicast/broadcast
10764 * interrupts to a vCPU, we still use interrupt remapping
10765 * for these kind of interrupts.
10767 * For lowest-priority interrupts, we only support
10768 * those with single CPU as the destination, e.g. user
10769 * configures the interrupts via /proc/irq or uses
10770 * irqbalance to make the interrupts single-CPU.
10772 * We will support full lowest-priority interrupt later.
10775 kvm_set_msi_irq(e, &irq);
10776 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
10778 * Make sure the IRTE is in remapped mode if
10779 * we don't handle it in posted mode.
10781 ret = irq_set_vcpu_affinity(host_irq, NULL);
10784 "failed to back to remapped mode, irq: %u\n",
10792 vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
10793 vcpu_info.vector = irq.vector;
10795 trace_kvm_pi_irte_update(vcpu->vcpu_id, host_irq, e->gsi,
10796 vcpu_info.vector, vcpu_info.pi_desc_addr, set);
10799 ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
10801 /* suppress notification event before unposting */
10802 pi_set_sn(vcpu_to_pi_desc(vcpu));
10803 ret = irq_set_vcpu_affinity(host_irq, NULL);
10804 pi_clear_sn(vcpu_to_pi_desc(vcpu));
10808 printk(KERN_INFO "%s: failed to update PI IRTE\n",
10816 srcu_read_unlock(&kvm->irq_srcu, idx);
10820 static struct kvm_x86_ops vmx_x86_ops = {
10821 .cpu_has_kvm_support = cpu_has_kvm_support,
10822 .disabled_by_bios = vmx_disabled_by_bios,
10823 .hardware_setup = hardware_setup,
10824 .hardware_unsetup = hardware_unsetup,
10825 .check_processor_compatibility = vmx_check_processor_compat,
10826 .hardware_enable = hardware_enable,
10827 .hardware_disable = hardware_disable,
10828 .cpu_has_accelerated_tpr = report_flexpriority,
10829 .cpu_has_high_real_mode_segbase = vmx_has_high_real_mode_segbase,
10831 .vcpu_create = vmx_create_vcpu,
10832 .vcpu_free = vmx_free_vcpu,
10833 .vcpu_reset = vmx_vcpu_reset,
10835 .prepare_guest_switch = vmx_save_host_state,
10836 .vcpu_load = vmx_vcpu_load,
10837 .vcpu_put = vmx_vcpu_put,
10839 .update_bp_intercept = update_exception_bitmap,
10840 .get_msr = vmx_get_msr,
10841 .set_msr = vmx_set_msr,
10842 .get_segment_base = vmx_get_segment_base,
10843 .get_segment = vmx_get_segment,
10844 .set_segment = vmx_set_segment,
10845 .get_cpl = vmx_get_cpl,
10846 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
10847 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
10848 .decache_cr3 = vmx_decache_cr3,
10849 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
10850 .set_cr0 = vmx_set_cr0,
10851 .set_cr3 = vmx_set_cr3,
10852 .set_cr4 = vmx_set_cr4,
10853 .set_efer = vmx_set_efer,
10854 .get_idt = vmx_get_idt,
10855 .set_idt = vmx_set_idt,
10856 .get_gdt = vmx_get_gdt,
10857 .set_gdt = vmx_set_gdt,
10858 .get_dr6 = vmx_get_dr6,
10859 .set_dr6 = vmx_set_dr6,
10860 .set_dr7 = vmx_set_dr7,
10861 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
10862 .cache_reg = vmx_cache_reg,
10863 .get_rflags = vmx_get_rflags,
10864 .set_rflags = vmx_set_rflags,
10865 .fpu_activate = vmx_fpu_activate,
10866 .fpu_deactivate = vmx_fpu_deactivate,
10868 .tlb_flush = vmx_flush_tlb,
10870 .run = vmx_vcpu_run,
10871 .handle_exit = vmx_handle_exit,
10872 .skip_emulated_instruction = skip_emulated_instruction,
10873 .set_interrupt_shadow = vmx_set_interrupt_shadow,
10874 .get_interrupt_shadow = vmx_get_interrupt_shadow,
10875 .patch_hypercall = vmx_patch_hypercall,
10876 .set_irq = vmx_inject_irq,
10877 .set_nmi = vmx_inject_nmi,
10878 .queue_exception = vmx_queue_exception,
10879 .cancel_injection = vmx_cancel_injection,
10880 .interrupt_allowed = vmx_interrupt_allowed,
10881 .nmi_allowed = vmx_nmi_allowed,
10882 .get_nmi_mask = vmx_get_nmi_mask,
10883 .set_nmi_mask = vmx_set_nmi_mask,
10884 .enable_nmi_window = enable_nmi_window,
10885 .enable_irq_window = enable_irq_window,
10886 .update_cr8_intercept = update_cr8_intercept,
10887 .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode,
10888 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
10889 .get_enable_apicv = vmx_get_enable_apicv,
10890 .refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
10891 .load_eoi_exitmap = vmx_load_eoi_exitmap,
10892 .hwapic_irr_update = vmx_hwapic_irr_update,
10893 .hwapic_isr_update = vmx_hwapic_isr_update,
10894 .sync_pir_to_irr = vmx_sync_pir_to_irr,
10895 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
10897 .set_tss_addr = vmx_set_tss_addr,
10898 .get_tdp_level = get_ept_level,
10899 .get_mt_mask = vmx_get_mt_mask,
10901 .get_exit_info = vmx_get_exit_info,
10903 .get_lpage_level = vmx_get_lpage_level,
10905 .cpuid_update = vmx_cpuid_update,
10907 .rdtscp_supported = vmx_rdtscp_supported,
10908 .invpcid_supported = vmx_invpcid_supported,
10910 .set_supported_cpuid = vmx_set_supported_cpuid,
10912 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
10914 .read_tsc_offset = vmx_read_tsc_offset,
10915 .write_tsc_offset = vmx_write_tsc_offset,
10916 .adjust_tsc_offset_guest = vmx_adjust_tsc_offset_guest,
10917 .read_l1_tsc = vmx_read_l1_tsc,
10919 .set_tdp_cr3 = vmx_set_cr3,
10921 .check_intercept = vmx_check_intercept,
10922 .handle_external_intr = vmx_handle_external_intr,
10923 .mpx_supported = vmx_mpx_supported,
10924 .xsaves_supported = vmx_xsaves_supported,
10926 .check_nested_events = vmx_check_nested_events,
10928 .sched_in = vmx_sched_in,
10930 .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
10931 .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
10932 .flush_log_dirty = vmx_flush_log_dirty,
10933 .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
10935 .pre_block = vmx_pre_block,
10936 .post_block = vmx_post_block,
10938 .pmu_ops = &intel_pmu_ops,
10940 .update_pi_irte = vmx_update_pi_irte,
10943 static int __init vmx_init(void)
10945 int r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
10946 __alignof__(struct vcpu_vmx), THIS_MODULE);
10950 #ifdef CONFIG_KEXEC_CORE
10951 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
10952 crash_vmclear_local_loaded_vmcss);
10958 static void __exit vmx_exit(void)
10960 #ifdef CONFIG_KEXEC_CORE
10961 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
10968 module_init(vmx_init)
10969 module_exit(vmx_exit)