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_page_fault(u32 intr_info)
966 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
967 INTR_INFO_VALID_MASK)) ==
968 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
971 static inline bool is_no_device(u32 intr_info)
973 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
974 INTR_INFO_VALID_MASK)) ==
975 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
978 static inline bool is_invalid_opcode(u32 intr_info)
980 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
981 INTR_INFO_VALID_MASK)) ==
982 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
985 static inline bool is_external_interrupt(u32 intr_info)
987 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
988 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
991 static inline bool is_machine_check(u32 intr_info)
993 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
994 INTR_INFO_VALID_MASK)) ==
995 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
998 static inline bool cpu_has_vmx_msr_bitmap(void)
1000 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
1003 static inline bool cpu_has_vmx_tpr_shadow(void)
1005 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
1008 static inline bool cpu_need_tpr_shadow(struct kvm_vcpu *vcpu)
1010 return cpu_has_vmx_tpr_shadow() && lapic_in_kernel(vcpu);
1013 static inline bool cpu_has_secondary_exec_ctrls(void)
1015 return vmcs_config.cpu_based_exec_ctrl &
1016 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1019 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
1021 return vmcs_config.cpu_based_2nd_exec_ctrl &
1022 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1025 static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
1027 return vmcs_config.cpu_based_2nd_exec_ctrl &
1028 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
1031 static inline bool cpu_has_vmx_apic_register_virt(void)
1033 return vmcs_config.cpu_based_2nd_exec_ctrl &
1034 SECONDARY_EXEC_APIC_REGISTER_VIRT;
1037 static inline bool cpu_has_vmx_virtual_intr_delivery(void)
1039 return vmcs_config.cpu_based_2nd_exec_ctrl &
1040 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
1043 static inline bool cpu_has_vmx_posted_intr(void)
1045 return IS_ENABLED(CONFIG_X86_LOCAL_APIC) &&
1046 vmcs_config.pin_based_exec_ctrl & PIN_BASED_POSTED_INTR;
1049 static inline bool cpu_has_vmx_apicv(void)
1051 return cpu_has_vmx_apic_register_virt() &&
1052 cpu_has_vmx_virtual_intr_delivery() &&
1053 cpu_has_vmx_posted_intr();
1056 static inline bool cpu_has_vmx_flexpriority(void)
1058 return cpu_has_vmx_tpr_shadow() &&
1059 cpu_has_vmx_virtualize_apic_accesses();
1062 static inline bool cpu_has_vmx_ept_execute_only(void)
1064 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
1067 static inline bool cpu_has_vmx_ept_2m_page(void)
1069 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
1072 static inline bool cpu_has_vmx_ept_1g_page(void)
1074 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
1077 static inline bool cpu_has_vmx_ept_4levels(void)
1079 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
1082 static inline bool cpu_has_vmx_ept_ad_bits(void)
1084 return vmx_capability.ept & VMX_EPT_AD_BIT;
1087 static inline bool cpu_has_vmx_invept_context(void)
1089 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
1092 static inline bool cpu_has_vmx_invept_global(void)
1094 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
1097 static inline bool cpu_has_vmx_invvpid_single(void)
1099 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
1102 static inline bool cpu_has_vmx_invvpid_global(void)
1104 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
1107 static inline bool cpu_has_vmx_ept(void)
1109 return vmcs_config.cpu_based_2nd_exec_ctrl &
1110 SECONDARY_EXEC_ENABLE_EPT;
1113 static inline bool cpu_has_vmx_unrestricted_guest(void)
1115 return vmcs_config.cpu_based_2nd_exec_ctrl &
1116 SECONDARY_EXEC_UNRESTRICTED_GUEST;
1119 static inline bool cpu_has_vmx_ple(void)
1121 return vmcs_config.cpu_based_2nd_exec_ctrl &
1122 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
1125 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
1127 return flexpriority_enabled && lapic_in_kernel(vcpu);
1130 static inline bool cpu_has_vmx_vpid(void)
1132 return vmcs_config.cpu_based_2nd_exec_ctrl &
1133 SECONDARY_EXEC_ENABLE_VPID;
1136 static inline bool cpu_has_vmx_rdtscp(void)
1138 return vmcs_config.cpu_based_2nd_exec_ctrl &
1139 SECONDARY_EXEC_RDTSCP;
1142 static inline bool cpu_has_vmx_invpcid(void)
1144 return vmcs_config.cpu_based_2nd_exec_ctrl &
1145 SECONDARY_EXEC_ENABLE_INVPCID;
1148 static inline bool cpu_has_virtual_nmis(void)
1150 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
1153 static inline bool cpu_has_vmx_wbinvd_exit(void)
1155 return vmcs_config.cpu_based_2nd_exec_ctrl &
1156 SECONDARY_EXEC_WBINVD_EXITING;
1159 static inline bool cpu_has_vmx_shadow_vmcs(void)
1162 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
1163 /* check if the cpu supports writing r/o exit information fields */
1164 if (!(vmx_msr & MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS))
1167 return vmcs_config.cpu_based_2nd_exec_ctrl &
1168 SECONDARY_EXEC_SHADOW_VMCS;
1171 static inline bool cpu_has_vmx_pml(void)
1173 return vmcs_config.cpu_based_2nd_exec_ctrl & SECONDARY_EXEC_ENABLE_PML;
1176 static inline bool cpu_has_vmx_tsc_scaling(void)
1178 return vmcs_config.cpu_based_2nd_exec_ctrl &
1179 SECONDARY_EXEC_TSC_SCALING;
1182 static inline bool report_flexpriority(void)
1184 return flexpriority_enabled;
1187 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
1189 return vmcs12->cpu_based_vm_exec_control & bit;
1192 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
1194 return (vmcs12->cpu_based_vm_exec_control &
1195 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
1196 (vmcs12->secondary_vm_exec_control & bit);
1199 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12)
1201 return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
1204 static inline bool nested_cpu_has_preemption_timer(struct vmcs12 *vmcs12)
1206 return vmcs12->pin_based_vm_exec_control &
1207 PIN_BASED_VMX_PREEMPTION_TIMER;
1210 static inline int nested_cpu_has_ept(struct vmcs12 *vmcs12)
1212 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_EPT);
1215 static inline bool nested_cpu_has_xsaves(struct vmcs12 *vmcs12)
1217 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES) &&
1218 vmx_xsaves_supported();
1221 static inline bool nested_cpu_has_virt_x2apic_mode(struct vmcs12 *vmcs12)
1223 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
1226 static inline bool nested_cpu_has_vpid(struct vmcs12 *vmcs12)
1228 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_VPID);
1231 static inline bool nested_cpu_has_apic_reg_virt(struct vmcs12 *vmcs12)
1233 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_APIC_REGISTER_VIRT);
1236 static inline bool nested_cpu_has_vid(struct vmcs12 *vmcs12)
1238 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
1241 static inline bool nested_cpu_has_posted_intr(struct vmcs12 *vmcs12)
1243 return vmcs12->pin_based_vm_exec_control & PIN_BASED_POSTED_INTR;
1246 static inline bool is_exception(u32 intr_info)
1248 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1249 == (INTR_TYPE_HARD_EXCEPTION | INTR_INFO_VALID_MASK);
1252 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
1254 unsigned long exit_qualification);
1255 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
1256 struct vmcs12 *vmcs12,
1257 u32 reason, unsigned long qualification);
1259 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
1263 for (i = 0; i < vmx->nmsrs; ++i)
1264 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
1269 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
1275 } operand = { vpid, 0, gva };
1277 asm volatile (__ex(ASM_VMX_INVVPID)
1278 /* CF==1 or ZF==1 --> rc = -1 */
1279 "; ja 1f ; ud2 ; 1:"
1280 : : "a"(&operand), "c"(ext) : "cc", "memory");
1283 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
1287 } operand = {eptp, gpa};
1289 asm volatile (__ex(ASM_VMX_INVEPT)
1290 /* CF==1 or ZF==1 --> rc = -1 */
1291 "; ja 1f ; ud2 ; 1:\n"
1292 : : "a" (&operand), "c" (ext) : "cc", "memory");
1295 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
1299 i = __find_msr_index(vmx, msr);
1301 return &vmx->guest_msrs[i];
1305 static void vmcs_clear(struct vmcs *vmcs)
1307 u64 phys_addr = __pa(vmcs);
1310 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
1311 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1314 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
1318 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
1320 vmcs_clear(loaded_vmcs->vmcs);
1321 loaded_vmcs->cpu = -1;
1322 loaded_vmcs->launched = 0;
1325 static void vmcs_load(struct vmcs *vmcs)
1327 u64 phys_addr = __pa(vmcs);
1330 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
1331 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1334 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
1338 #ifdef CONFIG_KEXEC_CORE
1340 * This bitmap is used to indicate whether the vmclear
1341 * operation is enabled on all cpus. All disabled by
1344 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
1346 static inline void crash_enable_local_vmclear(int cpu)
1348 cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
1351 static inline void crash_disable_local_vmclear(int cpu)
1353 cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
1356 static inline int crash_local_vmclear_enabled(int cpu)
1358 return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
1361 static void crash_vmclear_local_loaded_vmcss(void)
1363 int cpu = raw_smp_processor_id();
1364 struct loaded_vmcs *v;
1366 if (!crash_local_vmclear_enabled(cpu))
1369 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
1370 loaded_vmcss_on_cpu_link)
1371 vmcs_clear(v->vmcs);
1374 static inline void crash_enable_local_vmclear(int cpu) { }
1375 static inline void crash_disable_local_vmclear(int cpu) { }
1376 #endif /* CONFIG_KEXEC_CORE */
1378 static void __loaded_vmcs_clear(void *arg)
1380 struct loaded_vmcs *loaded_vmcs = arg;
1381 int cpu = raw_smp_processor_id();
1383 if (loaded_vmcs->cpu != cpu)
1384 return; /* vcpu migration can race with cpu offline */
1385 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1386 per_cpu(current_vmcs, cpu) = NULL;
1387 crash_disable_local_vmclear(cpu);
1388 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1391 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1392 * is before setting loaded_vmcs->vcpu to -1 which is done in
1393 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1394 * then adds the vmcs into percpu list before it is deleted.
1398 loaded_vmcs_init(loaded_vmcs);
1399 crash_enable_local_vmclear(cpu);
1402 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1404 int cpu = loaded_vmcs->cpu;
1407 smp_call_function_single(cpu,
1408 __loaded_vmcs_clear, loaded_vmcs, 1);
1411 static inline void vpid_sync_vcpu_single(int vpid)
1416 if (cpu_has_vmx_invvpid_single())
1417 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vpid, 0);
1420 static inline void vpid_sync_vcpu_global(void)
1422 if (cpu_has_vmx_invvpid_global())
1423 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1426 static inline void vpid_sync_context(int vpid)
1428 if (cpu_has_vmx_invvpid_single())
1429 vpid_sync_vcpu_single(vpid);
1431 vpid_sync_vcpu_global();
1434 static inline void ept_sync_global(void)
1436 if (cpu_has_vmx_invept_global())
1437 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1440 static inline void ept_sync_context(u64 eptp)
1443 if (cpu_has_vmx_invept_context())
1444 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1450 static __always_inline unsigned long vmcs_readl(unsigned long field)
1452 unsigned long value;
1454 asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1455 : "=a"(value) : "d"(field) : "cc");
1459 static __always_inline u16 vmcs_read16(unsigned long field)
1461 return vmcs_readl(field);
1464 static __always_inline u32 vmcs_read32(unsigned long field)
1466 return vmcs_readl(field);
1469 static __always_inline u64 vmcs_read64(unsigned long field)
1471 #ifdef CONFIG_X86_64
1472 return vmcs_readl(field);
1474 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
1478 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1480 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1481 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1485 static void vmcs_writel(unsigned long field, unsigned long value)
1489 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1490 : "=q"(error) : "a"(value), "d"(field) : "cc");
1491 if (unlikely(error))
1492 vmwrite_error(field, value);
1495 static void vmcs_write16(unsigned long field, u16 value)
1497 vmcs_writel(field, value);
1500 static void vmcs_write32(unsigned long field, u32 value)
1502 vmcs_writel(field, value);
1505 static void vmcs_write64(unsigned long field, u64 value)
1507 vmcs_writel(field, value);
1508 #ifndef CONFIG_X86_64
1510 vmcs_writel(field+1, value >> 32);
1514 static void vmcs_clear_bits(unsigned long field, u32 mask)
1516 vmcs_writel(field, vmcs_readl(field) & ~mask);
1519 static void vmcs_set_bits(unsigned long field, u32 mask)
1521 vmcs_writel(field, vmcs_readl(field) | mask);
1524 static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val)
1526 vmcs_write32(VM_ENTRY_CONTROLS, val);
1527 vmx->vm_entry_controls_shadow = val;
1530 static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val)
1532 if (vmx->vm_entry_controls_shadow != val)
1533 vm_entry_controls_init(vmx, val);
1536 static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx)
1538 return vmx->vm_entry_controls_shadow;
1542 static inline void vm_entry_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1544 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) | val);
1547 static inline void vm_entry_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1549 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) & ~val);
1552 static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val)
1554 vmcs_write32(VM_EXIT_CONTROLS, val);
1555 vmx->vm_exit_controls_shadow = val;
1558 static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val)
1560 if (vmx->vm_exit_controls_shadow != val)
1561 vm_exit_controls_init(vmx, val);
1564 static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx)
1566 return vmx->vm_exit_controls_shadow;
1570 static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1572 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val);
1575 static inline void vm_exit_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1577 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) & ~val);
1580 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1582 vmx->segment_cache.bitmask = 0;
1585 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1589 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1591 if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1592 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1593 vmx->segment_cache.bitmask = 0;
1595 ret = vmx->segment_cache.bitmask & mask;
1596 vmx->segment_cache.bitmask |= mask;
1600 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1602 u16 *p = &vmx->segment_cache.seg[seg].selector;
1604 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1605 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1609 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1611 ulong *p = &vmx->segment_cache.seg[seg].base;
1613 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1614 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1618 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1620 u32 *p = &vmx->segment_cache.seg[seg].limit;
1622 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1623 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1627 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1629 u32 *p = &vmx->segment_cache.seg[seg].ar;
1631 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1632 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1636 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1640 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1641 (1u << NM_VECTOR) | (1u << DB_VECTOR) | (1u << AC_VECTOR);
1642 if ((vcpu->guest_debug &
1643 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1644 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1645 eb |= 1u << BP_VECTOR;
1646 if (to_vmx(vcpu)->rmode.vm86_active)
1649 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1650 if (vcpu->fpu_active)
1651 eb &= ~(1u << NM_VECTOR);
1653 /* When we are running a nested L2 guest and L1 specified for it a
1654 * certain exception bitmap, we must trap the same exceptions and pass
1655 * them to L1. When running L2, we will only handle the exceptions
1656 * specified above if L1 did not want them.
1658 if (is_guest_mode(vcpu))
1659 eb |= get_vmcs12(vcpu)->exception_bitmap;
1661 vmcs_write32(EXCEPTION_BITMAP, eb);
1664 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1665 unsigned long entry, unsigned long exit)
1667 vm_entry_controls_clearbit(vmx, entry);
1668 vm_exit_controls_clearbit(vmx, exit);
1671 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1674 struct msr_autoload *m = &vmx->msr_autoload;
1678 if (cpu_has_load_ia32_efer) {
1679 clear_atomic_switch_msr_special(vmx,
1680 VM_ENTRY_LOAD_IA32_EFER,
1681 VM_EXIT_LOAD_IA32_EFER);
1685 case MSR_CORE_PERF_GLOBAL_CTRL:
1686 if (cpu_has_load_perf_global_ctrl) {
1687 clear_atomic_switch_msr_special(vmx,
1688 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1689 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1695 for (i = 0; i < m->nr; ++i)
1696 if (m->guest[i].index == msr)
1702 m->guest[i] = m->guest[m->nr];
1703 m->host[i] = m->host[m->nr];
1704 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1705 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1708 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1709 unsigned long entry, unsigned long exit,
1710 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
1711 u64 guest_val, u64 host_val)
1713 vmcs_write64(guest_val_vmcs, guest_val);
1714 vmcs_write64(host_val_vmcs, host_val);
1715 vm_entry_controls_setbit(vmx, entry);
1716 vm_exit_controls_setbit(vmx, exit);
1719 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1720 u64 guest_val, u64 host_val)
1723 struct msr_autoload *m = &vmx->msr_autoload;
1727 if (cpu_has_load_ia32_efer) {
1728 add_atomic_switch_msr_special(vmx,
1729 VM_ENTRY_LOAD_IA32_EFER,
1730 VM_EXIT_LOAD_IA32_EFER,
1733 guest_val, host_val);
1737 case MSR_CORE_PERF_GLOBAL_CTRL:
1738 if (cpu_has_load_perf_global_ctrl) {
1739 add_atomic_switch_msr_special(vmx,
1740 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1741 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1742 GUEST_IA32_PERF_GLOBAL_CTRL,
1743 HOST_IA32_PERF_GLOBAL_CTRL,
1744 guest_val, host_val);
1750 for (i = 0; i < m->nr; ++i)
1751 if (m->guest[i].index == msr)
1754 if (i == NR_AUTOLOAD_MSRS) {
1755 printk_once(KERN_WARNING "Not enough msr switch entries. "
1756 "Can't add msr %x\n", msr);
1758 } else if (i == m->nr) {
1760 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1761 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1764 m->guest[i].index = msr;
1765 m->guest[i].value = guest_val;
1766 m->host[i].index = msr;
1767 m->host[i].value = host_val;
1770 static void reload_tss(void)
1773 * VT restores TR but not its size. Useless.
1775 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
1776 struct desc_struct *descs;
1778 descs = (void *)gdt->address;
1779 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
1783 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
1788 guest_efer = vmx->vcpu.arch.efer;
1791 * NX is emulated; LMA and LME handled by hardware; SCE meaningless
1794 ignore_bits = EFER_NX | EFER_SCE;
1795 #ifdef CONFIG_X86_64
1796 ignore_bits |= EFER_LMA | EFER_LME;
1797 /* SCE is meaningful only in long mode on Intel */
1798 if (guest_efer & EFER_LMA)
1799 ignore_bits &= ~(u64)EFER_SCE;
1801 guest_efer &= ~ignore_bits;
1802 guest_efer |= host_efer & ignore_bits;
1803 vmx->guest_msrs[efer_offset].data = guest_efer;
1804 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1806 clear_atomic_switch_msr(vmx, MSR_EFER);
1809 * On EPT, we can't emulate NX, so we must switch EFER atomically.
1810 * On CPUs that support "load IA32_EFER", always switch EFER
1811 * atomically, since it's faster than switching it manually.
1813 if (cpu_has_load_ia32_efer ||
1814 (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
1815 guest_efer = vmx->vcpu.arch.efer;
1816 if (!(guest_efer & EFER_LMA))
1817 guest_efer &= ~EFER_LME;
1818 if (guest_efer != host_efer)
1819 add_atomic_switch_msr(vmx, MSR_EFER,
1820 guest_efer, host_efer);
1827 static unsigned long segment_base(u16 selector)
1829 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
1830 struct desc_struct *d;
1831 unsigned long table_base;
1834 if (!(selector & ~3))
1837 table_base = gdt->address;
1839 if (selector & 4) { /* from ldt */
1840 u16 ldt_selector = kvm_read_ldt();
1842 if (!(ldt_selector & ~3))
1845 table_base = segment_base(ldt_selector);
1847 d = (struct desc_struct *)(table_base + (selector & ~7));
1848 v = get_desc_base(d);
1849 #ifdef CONFIG_X86_64
1850 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
1851 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
1856 static inline unsigned long kvm_read_tr_base(void)
1859 asm("str %0" : "=g"(tr));
1860 return segment_base(tr);
1863 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
1865 struct vcpu_vmx *vmx = to_vmx(vcpu);
1868 if (vmx->host_state.loaded)
1871 vmx->host_state.loaded = 1;
1873 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
1874 * allow segment selectors with cpl > 0 or ti == 1.
1876 vmx->host_state.ldt_sel = kvm_read_ldt();
1877 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
1878 savesegment(fs, vmx->host_state.fs_sel);
1879 if (!(vmx->host_state.fs_sel & 7)) {
1880 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
1881 vmx->host_state.fs_reload_needed = 0;
1883 vmcs_write16(HOST_FS_SELECTOR, 0);
1884 vmx->host_state.fs_reload_needed = 1;
1886 savesegment(gs, vmx->host_state.gs_sel);
1887 if (!(vmx->host_state.gs_sel & 7))
1888 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
1890 vmcs_write16(HOST_GS_SELECTOR, 0);
1891 vmx->host_state.gs_ldt_reload_needed = 1;
1894 #ifdef CONFIG_X86_64
1895 savesegment(ds, vmx->host_state.ds_sel);
1896 savesegment(es, vmx->host_state.es_sel);
1899 #ifdef CONFIG_X86_64
1900 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
1901 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
1903 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
1904 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
1907 #ifdef CONFIG_X86_64
1908 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1909 if (is_long_mode(&vmx->vcpu))
1910 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1912 if (boot_cpu_has(X86_FEATURE_MPX))
1913 rdmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
1914 for (i = 0; i < vmx->save_nmsrs; ++i)
1915 kvm_set_shared_msr(vmx->guest_msrs[i].index,
1916 vmx->guest_msrs[i].data,
1917 vmx->guest_msrs[i].mask);
1920 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
1922 if (!vmx->host_state.loaded)
1925 ++vmx->vcpu.stat.host_state_reload;
1926 vmx->host_state.loaded = 0;
1927 #ifdef CONFIG_X86_64
1928 if (is_long_mode(&vmx->vcpu))
1929 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1931 if (vmx->host_state.gs_ldt_reload_needed) {
1932 kvm_load_ldt(vmx->host_state.ldt_sel);
1933 #ifdef CONFIG_X86_64
1934 load_gs_index(vmx->host_state.gs_sel);
1936 loadsegment(gs, vmx->host_state.gs_sel);
1939 if (vmx->host_state.fs_reload_needed)
1940 loadsegment(fs, vmx->host_state.fs_sel);
1941 #ifdef CONFIG_X86_64
1942 if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
1943 loadsegment(ds, vmx->host_state.ds_sel);
1944 loadsegment(es, vmx->host_state.es_sel);
1948 #ifdef CONFIG_X86_64
1949 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1951 if (vmx->host_state.msr_host_bndcfgs)
1952 wrmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
1954 * If the FPU is not active (through the host task or
1955 * the guest vcpu), then restore the cr0.TS bit.
1957 if (!fpregs_active() && !vmx->vcpu.guest_fpu_loaded)
1959 load_gdt(this_cpu_ptr(&host_gdt));
1962 static void vmx_load_host_state(struct vcpu_vmx *vmx)
1965 __vmx_load_host_state(vmx);
1969 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
1971 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1972 struct pi_desc old, new;
1975 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
1976 !irq_remapping_cap(IRQ_POSTING_CAP))
1980 old.control = new.control = pi_desc->control;
1983 * If 'nv' field is POSTED_INTR_WAKEUP_VECTOR, there
1984 * are two possible cases:
1985 * 1. After running 'pre_block', context switch
1986 * happened. For this case, 'sn' was set in
1987 * vmx_vcpu_put(), so we need to clear it here.
1988 * 2. After running 'pre_block', we were blocked,
1989 * and woken up by some other guy. For this case,
1990 * we don't need to do anything, 'pi_post_block'
1991 * will do everything for us. However, we cannot
1992 * check whether it is case #1 or case #2 here
1993 * (maybe, not needed), so we also clear sn here,
1994 * I think it is not a big deal.
1996 if (pi_desc->nv != POSTED_INTR_WAKEUP_VECTOR) {
1997 if (vcpu->cpu != cpu) {
1998 dest = cpu_physical_id(cpu);
2000 if (x2apic_enabled())
2003 new.ndst = (dest << 8) & 0xFF00;
2006 /* set 'NV' to 'notification vector' */
2007 new.nv = POSTED_INTR_VECTOR;
2010 /* Allow posting non-urgent interrupts */
2012 } while (cmpxchg(&pi_desc->control, old.control,
2013 new.control) != old.control);
2016 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
2017 * vcpu mutex is already taken.
2019 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2021 struct vcpu_vmx *vmx = to_vmx(vcpu);
2022 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2025 kvm_cpu_vmxon(phys_addr);
2026 else if (vmx->loaded_vmcs->cpu != cpu)
2027 loaded_vmcs_clear(vmx->loaded_vmcs);
2029 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
2030 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
2031 vmcs_load(vmx->loaded_vmcs->vmcs);
2034 if (vmx->loaded_vmcs->cpu != cpu) {
2035 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
2036 unsigned long sysenter_esp;
2038 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2039 local_irq_disable();
2040 crash_disable_local_vmclear(cpu);
2043 * Read loaded_vmcs->cpu should be before fetching
2044 * loaded_vmcs->loaded_vmcss_on_cpu_link.
2045 * See the comments in __loaded_vmcs_clear().
2049 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
2050 &per_cpu(loaded_vmcss_on_cpu, cpu));
2051 crash_enable_local_vmclear(cpu);
2055 * Linux uses per-cpu TSS and GDT, so set these when switching
2058 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
2059 vmcs_writel(HOST_GDTR_BASE, gdt->address); /* 22.2.4 */
2061 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
2062 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
2064 /* Setup TSC multiplier */
2065 if (cpu_has_vmx_tsc_scaling())
2066 vmcs_write64(TSC_MULTIPLIER,
2067 vcpu->arch.tsc_scaling_ratio);
2069 vmx->loaded_vmcs->cpu = cpu;
2072 vmx_vcpu_pi_load(vcpu, cpu);
2075 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
2077 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2079 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
2080 !irq_remapping_cap(IRQ_POSTING_CAP))
2083 /* Set SN when the vCPU is preempted */
2084 if (vcpu->preempted)
2088 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
2090 vmx_vcpu_pi_put(vcpu);
2092 __vmx_load_host_state(to_vmx(vcpu));
2093 if (!vmm_exclusive) {
2094 __loaded_vmcs_clear(to_vmx(vcpu)->loaded_vmcs);
2100 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
2104 if (vcpu->fpu_active)
2106 vcpu->fpu_active = 1;
2107 cr0 = vmcs_readl(GUEST_CR0);
2108 cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
2109 cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
2110 vmcs_writel(GUEST_CR0, cr0);
2111 update_exception_bitmap(vcpu);
2112 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
2113 if (is_guest_mode(vcpu))
2114 vcpu->arch.cr0_guest_owned_bits &=
2115 ~get_vmcs12(vcpu)->cr0_guest_host_mask;
2116 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2119 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
2122 * Return the cr0 value that a nested guest would read. This is a combination
2123 * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
2124 * its hypervisor (cr0_read_shadow).
2126 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
2128 return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
2129 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
2131 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
2133 return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
2134 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
2137 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
2139 /* Note that there is no vcpu->fpu_active = 0 here. The caller must
2140 * set this *before* calling this function.
2142 vmx_decache_cr0_guest_bits(vcpu);
2143 vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
2144 update_exception_bitmap(vcpu);
2145 vcpu->arch.cr0_guest_owned_bits = 0;
2146 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2147 if (is_guest_mode(vcpu)) {
2149 * L1's specified read shadow might not contain the TS bit,
2150 * so now that we turned on shadowing of this bit, we need to
2151 * set this bit of the shadow. Like in nested_vmx_run we need
2152 * nested_read_cr0(vmcs12), but vmcs12->guest_cr0 is not yet
2153 * up-to-date here because we just decached cr0.TS (and we'll
2154 * only update vmcs12->guest_cr0 on nested exit).
2156 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2157 vmcs12->guest_cr0 = (vmcs12->guest_cr0 & ~X86_CR0_TS) |
2158 (vcpu->arch.cr0 & X86_CR0_TS);
2159 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2161 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
2164 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
2166 unsigned long rflags, save_rflags;
2168 if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
2169 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2170 rflags = vmcs_readl(GUEST_RFLAGS);
2171 if (to_vmx(vcpu)->rmode.vm86_active) {
2172 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2173 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
2174 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2176 to_vmx(vcpu)->rflags = rflags;
2178 return to_vmx(vcpu)->rflags;
2181 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
2183 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2184 to_vmx(vcpu)->rflags = rflags;
2185 if (to_vmx(vcpu)->rmode.vm86_active) {
2186 to_vmx(vcpu)->rmode.save_rflags = rflags;
2187 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2189 vmcs_writel(GUEST_RFLAGS, rflags);
2192 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
2194 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2197 if (interruptibility & GUEST_INTR_STATE_STI)
2198 ret |= KVM_X86_SHADOW_INT_STI;
2199 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
2200 ret |= KVM_X86_SHADOW_INT_MOV_SS;
2205 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
2207 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2208 u32 interruptibility = interruptibility_old;
2210 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
2212 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
2213 interruptibility |= GUEST_INTR_STATE_MOV_SS;
2214 else if (mask & KVM_X86_SHADOW_INT_STI)
2215 interruptibility |= GUEST_INTR_STATE_STI;
2217 if ((interruptibility != interruptibility_old))
2218 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
2221 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
2225 rip = kvm_rip_read(vcpu);
2226 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
2227 kvm_rip_write(vcpu, rip);
2229 /* skipping an emulated instruction also counts */
2230 vmx_set_interrupt_shadow(vcpu, 0);
2234 * KVM wants to inject page-faults which it got to the guest. This function
2235 * checks whether in a nested guest, we need to inject them to L1 or L2.
2237 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned nr)
2239 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2241 if (!(vmcs12->exception_bitmap & (1u << nr)))
2244 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
2245 vmcs_read32(VM_EXIT_INTR_INFO),
2246 vmcs_readl(EXIT_QUALIFICATION));
2250 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
2251 bool has_error_code, u32 error_code,
2254 struct vcpu_vmx *vmx = to_vmx(vcpu);
2255 u32 intr_info = nr | INTR_INFO_VALID_MASK;
2257 if (!reinject && is_guest_mode(vcpu) &&
2258 nested_vmx_check_exception(vcpu, nr))
2261 if (has_error_code) {
2262 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
2263 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2266 if (vmx->rmode.vm86_active) {
2268 if (kvm_exception_is_soft(nr))
2269 inc_eip = vcpu->arch.event_exit_inst_len;
2270 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
2271 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2275 if (kvm_exception_is_soft(nr)) {
2276 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2277 vmx->vcpu.arch.event_exit_inst_len);
2278 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2280 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2282 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
2285 static bool vmx_rdtscp_supported(void)
2287 return cpu_has_vmx_rdtscp();
2290 static bool vmx_invpcid_supported(void)
2292 return cpu_has_vmx_invpcid() && enable_ept;
2296 * Swap MSR entry in host/guest MSR entry array.
2298 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
2300 struct shared_msr_entry tmp;
2302 tmp = vmx->guest_msrs[to];
2303 vmx->guest_msrs[to] = vmx->guest_msrs[from];
2304 vmx->guest_msrs[from] = tmp;
2307 static void vmx_set_msr_bitmap(struct kvm_vcpu *vcpu)
2309 unsigned long *msr_bitmap;
2311 if (is_guest_mode(vcpu))
2312 msr_bitmap = vmx_msr_bitmap_nested;
2313 else if (vcpu->arch.apic_base & X2APIC_ENABLE) {
2314 if (is_long_mode(vcpu))
2315 msr_bitmap = vmx_msr_bitmap_longmode_x2apic;
2317 msr_bitmap = vmx_msr_bitmap_legacy_x2apic;
2319 if (is_long_mode(vcpu))
2320 msr_bitmap = vmx_msr_bitmap_longmode;
2322 msr_bitmap = vmx_msr_bitmap_legacy;
2325 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
2329 * Set up the vmcs to automatically save and restore system
2330 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
2331 * mode, as fiddling with msrs is very expensive.
2333 static void setup_msrs(struct vcpu_vmx *vmx)
2335 int save_nmsrs, index;
2338 #ifdef CONFIG_X86_64
2339 if (is_long_mode(&vmx->vcpu)) {
2340 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
2342 move_msr_up(vmx, index, save_nmsrs++);
2343 index = __find_msr_index(vmx, MSR_LSTAR);
2345 move_msr_up(vmx, index, save_nmsrs++);
2346 index = __find_msr_index(vmx, MSR_CSTAR);
2348 move_msr_up(vmx, index, save_nmsrs++);
2349 index = __find_msr_index(vmx, MSR_TSC_AUX);
2350 if (index >= 0 && guest_cpuid_has_rdtscp(&vmx->vcpu))
2351 move_msr_up(vmx, index, save_nmsrs++);
2353 * MSR_STAR is only needed on long mode guests, and only
2354 * if efer.sce is enabled.
2356 index = __find_msr_index(vmx, MSR_STAR);
2357 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
2358 move_msr_up(vmx, index, save_nmsrs++);
2361 index = __find_msr_index(vmx, MSR_EFER);
2362 if (index >= 0 && update_transition_efer(vmx, index))
2363 move_msr_up(vmx, index, save_nmsrs++);
2365 vmx->save_nmsrs = save_nmsrs;
2367 if (cpu_has_vmx_msr_bitmap())
2368 vmx_set_msr_bitmap(&vmx->vcpu);
2372 * reads and returns guest's timestamp counter "register"
2373 * guest_tsc = (host_tsc * tsc multiplier) >> 48 + tsc_offset
2374 * -- Intel TSC Scaling for Virtualization White Paper, sec 1.3
2376 static u64 guest_read_tsc(struct kvm_vcpu *vcpu)
2378 u64 host_tsc, tsc_offset;
2381 tsc_offset = vmcs_read64(TSC_OFFSET);
2382 return kvm_scale_tsc(vcpu, host_tsc) + tsc_offset;
2386 * Like guest_read_tsc, but always returns L1's notion of the timestamp
2387 * counter, even if a nested guest (L2) is currently running.
2389 static u64 vmx_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2393 tsc_offset = is_guest_mode(vcpu) ?
2394 to_vmx(vcpu)->nested.vmcs01_tsc_offset :
2395 vmcs_read64(TSC_OFFSET);
2396 return host_tsc + tsc_offset;
2399 static u64 vmx_read_tsc_offset(struct kvm_vcpu *vcpu)
2401 return vmcs_read64(TSC_OFFSET);
2405 * writes 'offset' into guest's timestamp counter offset register
2407 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2409 if (is_guest_mode(vcpu)) {
2411 * We're here if L1 chose not to trap WRMSR to TSC. According
2412 * to the spec, this should set L1's TSC; The offset that L1
2413 * set for L2 remains unchanged, and still needs to be added
2414 * to the newly set TSC to get L2's TSC.
2416 struct vmcs12 *vmcs12;
2417 to_vmx(vcpu)->nested.vmcs01_tsc_offset = offset;
2418 /* recalculate vmcs02.TSC_OFFSET: */
2419 vmcs12 = get_vmcs12(vcpu);
2420 vmcs_write64(TSC_OFFSET, offset +
2421 (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
2422 vmcs12->tsc_offset : 0));
2424 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2425 vmcs_read64(TSC_OFFSET), offset);
2426 vmcs_write64(TSC_OFFSET, offset);
2430 static void vmx_adjust_tsc_offset_guest(struct kvm_vcpu *vcpu, s64 adjustment)
2432 u64 offset = vmcs_read64(TSC_OFFSET);
2434 vmcs_write64(TSC_OFFSET, offset + adjustment);
2435 if (is_guest_mode(vcpu)) {
2436 /* Even when running L2, the adjustment needs to apply to L1 */
2437 to_vmx(vcpu)->nested.vmcs01_tsc_offset += adjustment;
2439 trace_kvm_write_tsc_offset(vcpu->vcpu_id, offset,
2440 offset + adjustment);
2443 static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
2445 struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
2446 return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
2450 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
2451 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
2452 * all guests if the "nested" module option is off, and can also be disabled
2453 * for a single guest by disabling its VMX cpuid bit.
2455 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
2457 return nested && guest_cpuid_has_vmx(vcpu);
2461 * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
2462 * returned for the various VMX controls MSRs when nested VMX is enabled.
2463 * The same values should also be used to verify that vmcs12 control fields are
2464 * valid during nested entry from L1 to L2.
2465 * Each of these control msrs has a low and high 32-bit half: A low bit is on
2466 * if the corresponding bit in the (32-bit) control field *must* be on, and a
2467 * bit in the high half is on if the corresponding bit in the control field
2468 * may be on. See also vmx_control_verify().
2470 static void nested_vmx_setup_ctls_msrs(struct vcpu_vmx *vmx)
2473 * Note that as a general rule, the high half of the MSRs (bits in
2474 * the control fields which may be 1) should be initialized by the
2475 * intersection of the underlying hardware's MSR (i.e., features which
2476 * can be supported) and the list of features we want to expose -
2477 * because they are known to be properly supported in our code.
2478 * Also, usually, the low half of the MSRs (bits which must be 1) can
2479 * be set to 0, meaning that L1 may turn off any of these bits. The
2480 * reason is that if one of these bits is necessary, it will appear
2481 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
2482 * fields of vmcs01 and vmcs02, will turn these bits off - and
2483 * nested_vmx_exit_handled() will not pass related exits to L1.
2484 * These rules have exceptions below.
2487 /* pin-based controls */
2488 rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
2489 vmx->nested.nested_vmx_pinbased_ctls_low,
2490 vmx->nested.nested_vmx_pinbased_ctls_high);
2491 vmx->nested.nested_vmx_pinbased_ctls_low |=
2492 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2493 vmx->nested.nested_vmx_pinbased_ctls_high &=
2494 PIN_BASED_EXT_INTR_MASK |
2495 PIN_BASED_NMI_EXITING |
2496 PIN_BASED_VIRTUAL_NMIS;
2497 vmx->nested.nested_vmx_pinbased_ctls_high |=
2498 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2499 PIN_BASED_VMX_PREEMPTION_TIMER;
2500 if (kvm_vcpu_apicv_active(&vmx->vcpu))
2501 vmx->nested.nested_vmx_pinbased_ctls_high |=
2502 PIN_BASED_POSTED_INTR;
2505 rdmsr(MSR_IA32_VMX_EXIT_CTLS,
2506 vmx->nested.nested_vmx_exit_ctls_low,
2507 vmx->nested.nested_vmx_exit_ctls_high);
2508 vmx->nested.nested_vmx_exit_ctls_low =
2509 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
2511 vmx->nested.nested_vmx_exit_ctls_high &=
2512 #ifdef CONFIG_X86_64
2513 VM_EXIT_HOST_ADDR_SPACE_SIZE |
2515 VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
2516 vmx->nested.nested_vmx_exit_ctls_high |=
2517 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
2518 VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
2519 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
2521 if (vmx_mpx_supported())
2522 vmx->nested.nested_vmx_exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
2524 /* We support free control of debug control saving. */
2525 vmx->nested.nested_vmx_true_exit_ctls_low =
2526 vmx->nested.nested_vmx_exit_ctls_low &
2527 ~VM_EXIT_SAVE_DEBUG_CONTROLS;
2529 /* entry controls */
2530 rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
2531 vmx->nested.nested_vmx_entry_ctls_low,
2532 vmx->nested.nested_vmx_entry_ctls_high);
2533 vmx->nested.nested_vmx_entry_ctls_low =
2534 VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
2535 vmx->nested.nested_vmx_entry_ctls_high &=
2536 #ifdef CONFIG_X86_64
2537 VM_ENTRY_IA32E_MODE |
2539 VM_ENTRY_LOAD_IA32_PAT;
2540 vmx->nested.nested_vmx_entry_ctls_high |=
2541 (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
2542 if (vmx_mpx_supported())
2543 vmx->nested.nested_vmx_entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
2545 /* We support free control of debug control loading. */
2546 vmx->nested.nested_vmx_true_entry_ctls_low =
2547 vmx->nested.nested_vmx_entry_ctls_low &
2548 ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
2550 /* cpu-based controls */
2551 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
2552 vmx->nested.nested_vmx_procbased_ctls_low,
2553 vmx->nested.nested_vmx_procbased_ctls_high);
2554 vmx->nested.nested_vmx_procbased_ctls_low =
2555 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2556 vmx->nested.nested_vmx_procbased_ctls_high &=
2557 CPU_BASED_VIRTUAL_INTR_PENDING |
2558 CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
2559 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
2560 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
2561 CPU_BASED_CR3_STORE_EXITING |
2562 #ifdef CONFIG_X86_64
2563 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
2565 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
2566 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
2567 CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
2568 CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
2569 CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2571 * We can allow some features even when not supported by the
2572 * hardware. For example, L1 can specify an MSR bitmap - and we
2573 * can use it to avoid exits to L1 - even when L0 runs L2
2574 * without MSR bitmaps.
2576 vmx->nested.nested_vmx_procbased_ctls_high |=
2577 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2578 CPU_BASED_USE_MSR_BITMAPS;
2580 /* We support free control of CR3 access interception. */
2581 vmx->nested.nested_vmx_true_procbased_ctls_low =
2582 vmx->nested.nested_vmx_procbased_ctls_low &
2583 ~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
2585 /* secondary cpu-based controls */
2586 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2587 vmx->nested.nested_vmx_secondary_ctls_low,
2588 vmx->nested.nested_vmx_secondary_ctls_high);
2589 vmx->nested.nested_vmx_secondary_ctls_low = 0;
2590 vmx->nested.nested_vmx_secondary_ctls_high &=
2591 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2592 SECONDARY_EXEC_RDTSCP |
2593 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2594 SECONDARY_EXEC_ENABLE_VPID |
2595 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2596 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2597 SECONDARY_EXEC_WBINVD_EXITING |
2598 SECONDARY_EXEC_XSAVES |
2599 SECONDARY_EXEC_PCOMMIT;
2602 /* nested EPT: emulate EPT also to L1 */
2603 vmx->nested.nested_vmx_secondary_ctls_high |=
2604 SECONDARY_EXEC_ENABLE_EPT;
2605 vmx->nested.nested_vmx_ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
2606 VMX_EPTP_WB_BIT | VMX_EPT_2MB_PAGE_BIT |
2608 vmx->nested.nested_vmx_ept_caps &= vmx_capability.ept;
2610 * For nested guests, we don't do anything specific
2611 * for single context invalidation. Hence, only advertise
2612 * support for global context invalidation.
2614 vmx->nested.nested_vmx_ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT;
2616 vmx->nested.nested_vmx_ept_caps = 0;
2619 vmx->nested.nested_vmx_vpid_caps = VMX_VPID_INVVPID_BIT |
2620 VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
2622 vmx->nested.nested_vmx_vpid_caps = 0;
2624 if (enable_unrestricted_guest)
2625 vmx->nested.nested_vmx_secondary_ctls_high |=
2626 SECONDARY_EXEC_UNRESTRICTED_GUEST;
2628 /* miscellaneous data */
2629 rdmsr(MSR_IA32_VMX_MISC,
2630 vmx->nested.nested_vmx_misc_low,
2631 vmx->nested.nested_vmx_misc_high);
2632 vmx->nested.nested_vmx_misc_low &= VMX_MISC_SAVE_EFER_LMA;
2633 vmx->nested.nested_vmx_misc_low |=
2634 VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
2635 VMX_MISC_ACTIVITY_HLT;
2636 vmx->nested.nested_vmx_misc_high = 0;
2639 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2642 * Bits 0 in high must be 0, and bits 1 in low must be 1.
2644 return ((control & high) | low) == control;
2647 static inline u64 vmx_control_msr(u32 low, u32 high)
2649 return low | ((u64)high << 32);
2652 /* Returns 0 on success, non-0 otherwise. */
2653 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2655 struct vcpu_vmx *vmx = to_vmx(vcpu);
2657 switch (msr_index) {
2658 case MSR_IA32_VMX_BASIC:
2660 * This MSR reports some information about VMX support. We
2661 * should return information about the VMX we emulate for the
2662 * guest, and the VMCS structure we give it - not about the
2663 * VMX support of the underlying hardware.
2665 *pdata = VMCS12_REVISION | VMX_BASIC_TRUE_CTLS |
2666 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2667 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2669 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2670 case MSR_IA32_VMX_PINBASED_CTLS:
2671 *pdata = vmx_control_msr(
2672 vmx->nested.nested_vmx_pinbased_ctls_low,
2673 vmx->nested.nested_vmx_pinbased_ctls_high);
2675 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2676 *pdata = vmx_control_msr(
2677 vmx->nested.nested_vmx_true_procbased_ctls_low,
2678 vmx->nested.nested_vmx_procbased_ctls_high);
2680 case MSR_IA32_VMX_PROCBASED_CTLS:
2681 *pdata = vmx_control_msr(
2682 vmx->nested.nested_vmx_procbased_ctls_low,
2683 vmx->nested.nested_vmx_procbased_ctls_high);
2685 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2686 *pdata = vmx_control_msr(
2687 vmx->nested.nested_vmx_true_exit_ctls_low,
2688 vmx->nested.nested_vmx_exit_ctls_high);
2690 case MSR_IA32_VMX_EXIT_CTLS:
2691 *pdata = vmx_control_msr(
2692 vmx->nested.nested_vmx_exit_ctls_low,
2693 vmx->nested.nested_vmx_exit_ctls_high);
2695 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2696 *pdata = vmx_control_msr(
2697 vmx->nested.nested_vmx_true_entry_ctls_low,
2698 vmx->nested.nested_vmx_entry_ctls_high);
2700 case MSR_IA32_VMX_ENTRY_CTLS:
2701 *pdata = vmx_control_msr(
2702 vmx->nested.nested_vmx_entry_ctls_low,
2703 vmx->nested.nested_vmx_entry_ctls_high);
2705 case MSR_IA32_VMX_MISC:
2706 *pdata = vmx_control_msr(
2707 vmx->nested.nested_vmx_misc_low,
2708 vmx->nested.nested_vmx_misc_high);
2711 * These MSRs specify bits which the guest must keep fixed (on or off)
2712 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2713 * We picked the standard core2 setting.
2715 #define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2716 #define VMXON_CR4_ALWAYSON X86_CR4_VMXE
2717 case MSR_IA32_VMX_CR0_FIXED0:
2718 *pdata = VMXON_CR0_ALWAYSON;
2720 case MSR_IA32_VMX_CR0_FIXED1:
2723 case MSR_IA32_VMX_CR4_FIXED0:
2724 *pdata = VMXON_CR4_ALWAYSON;
2726 case MSR_IA32_VMX_CR4_FIXED1:
2729 case MSR_IA32_VMX_VMCS_ENUM:
2730 *pdata = 0x2e; /* highest index: VMX_PREEMPTION_TIMER_VALUE */
2732 case MSR_IA32_VMX_PROCBASED_CTLS2:
2733 *pdata = vmx_control_msr(
2734 vmx->nested.nested_vmx_secondary_ctls_low,
2735 vmx->nested.nested_vmx_secondary_ctls_high);
2737 case MSR_IA32_VMX_EPT_VPID_CAP:
2738 /* Currently, no nested vpid support */
2739 *pdata = vmx->nested.nested_vmx_ept_caps |
2740 ((u64)vmx->nested.nested_vmx_vpid_caps << 32);
2750 * Reads an msr value (of 'msr_index') into 'pdata'.
2751 * Returns 0 on success, non-0 otherwise.
2752 * Assumes vcpu_load() was already called.
2754 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2756 struct shared_msr_entry *msr;
2758 switch (msr_info->index) {
2759 #ifdef CONFIG_X86_64
2761 msr_info->data = vmcs_readl(GUEST_FS_BASE);
2764 msr_info->data = vmcs_readl(GUEST_GS_BASE);
2766 case MSR_KERNEL_GS_BASE:
2767 vmx_load_host_state(to_vmx(vcpu));
2768 msr_info->data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
2772 return kvm_get_msr_common(vcpu, msr_info);
2774 msr_info->data = guest_read_tsc(vcpu);
2776 case MSR_IA32_SYSENTER_CS:
2777 msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
2779 case MSR_IA32_SYSENTER_EIP:
2780 msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
2782 case MSR_IA32_SYSENTER_ESP:
2783 msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
2785 case MSR_IA32_BNDCFGS:
2786 if (!vmx_mpx_supported())
2788 msr_info->data = vmcs_read64(GUEST_BNDCFGS);
2790 case MSR_IA32_FEATURE_CONTROL:
2791 if (!nested_vmx_allowed(vcpu))
2793 msr_info->data = to_vmx(vcpu)->nested.msr_ia32_feature_control;
2795 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2796 if (!nested_vmx_allowed(vcpu))
2798 return vmx_get_vmx_msr(vcpu, msr_info->index, &msr_info->data);
2800 if (!vmx_xsaves_supported())
2802 msr_info->data = vcpu->arch.ia32_xss;
2805 if (!guest_cpuid_has_rdtscp(vcpu))
2807 /* Otherwise falls through */
2809 msr = find_msr_entry(to_vmx(vcpu), msr_info->index);
2811 msr_info->data = msr->data;
2814 return kvm_get_msr_common(vcpu, msr_info);
2820 static void vmx_leave_nested(struct kvm_vcpu *vcpu);
2823 * Writes msr value into into the appropriate "register".
2824 * Returns 0 on success, non-0 otherwise.
2825 * Assumes vcpu_load() was already called.
2827 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2829 struct vcpu_vmx *vmx = to_vmx(vcpu);
2830 struct shared_msr_entry *msr;
2832 u32 msr_index = msr_info->index;
2833 u64 data = msr_info->data;
2835 switch (msr_index) {
2837 ret = kvm_set_msr_common(vcpu, msr_info);
2839 #ifdef CONFIG_X86_64
2841 vmx_segment_cache_clear(vmx);
2842 vmcs_writel(GUEST_FS_BASE, data);
2845 vmx_segment_cache_clear(vmx);
2846 vmcs_writel(GUEST_GS_BASE, data);
2848 case MSR_KERNEL_GS_BASE:
2849 vmx_load_host_state(vmx);
2850 vmx->msr_guest_kernel_gs_base = data;
2853 case MSR_IA32_SYSENTER_CS:
2854 vmcs_write32(GUEST_SYSENTER_CS, data);
2856 case MSR_IA32_SYSENTER_EIP:
2857 vmcs_writel(GUEST_SYSENTER_EIP, data);
2859 case MSR_IA32_SYSENTER_ESP:
2860 vmcs_writel(GUEST_SYSENTER_ESP, data);
2862 case MSR_IA32_BNDCFGS:
2863 if (!vmx_mpx_supported())
2865 vmcs_write64(GUEST_BNDCFGS, data);
2868 kvm_write_tsc(vcpu, msr_info);
2870 case MSR_IA32_CR_PAT:
2871 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2872 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
2874 vmcs_write64(GUEST_IA32_PAT, data);
2875 vcpu->arch.pat = data;
2878 ret = kvm_set_msr_common(vcpu, msr_info);
2880 case MSR_IA32_TSC_ADJUST:
2881 ret = kvm_set_msr_common(vcpu, msr_info);
2883 case MSR_IA32_FEATURE_CONTROL:
2884 if (!nested_vmx_allowed(vcpu) ||
2885 (to_vmx(vcpu)->nested.msr_ia32_feature_control &
2886 FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
2888 vmx->nested.msr_ia32_feature_control = data;
2889 if (msr_info->host_initiated && data == 0)
2890 vmx_leave_nested(vcpu);
2892 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2893 return 1; /* they are read-only */
2895 if (!vmx_xsaves_supported())
2898 * The only supported bit as of Skylake is bit 8, but
2899 * it is not supported on KVM.
2903 vcpu->arch.ia32_xss = data;
2904 if (vcpu->arch.ia32_xss != host_xss)
2905 add_atomic_switch_msr(vmx, MSR_IA32_XSS,
2906 vcpu->arch.ia32_xss, host_xss);
2908 clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
2911 if (!guest_cpuid_has_rdtscp(vcpu))
2913 /* Check reserved bit, higher 32 bits should be zero */
2914 if ((data >> 32) != 0)
2916 /* Otherwise falls through */
2918 msr = find_msr_entry(vmx, msr_index);
2920 u64 old_msr_data = msr->data;
2922 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
2924 ret = kvm_set_shared_msr(msr->index, msr->data,
2928 msr->data = old_msr_data;
2932 ret = kvm_set_msr_common(vcpu, msr_info);
2938 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2940 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
2943 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2946 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2948 case VCPU_EXREG_PDPTR:
2950 ept_save_pdptrs(vcpu);
2957 static __init int cpu_has_kvm_support(void)
2959 return cpu_has_vmx();
2962 static __init int vmx_disabled_by_bios(void)
2966 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
2967 if (msr & FEATURE_CONTROL_LOCKED) {
2968 /* launched w/ TXT and VMX disabled */
2969 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2972 /* launched w/o TXT and VMX only enabled w/ TXT */
2973 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2974 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2975 && !tboot_enabled()) {
2976 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
2977 "activate TXT before enabling KVM\n");
2980 /* launched w/o TXT and VMX disabled */
2981 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2982 && !tboot_enabled())
2989 static void kvm_cpu_vmxon(u64 addr)
2991 asm volatile (ASM_VMX_VMXON_RAX
2992 : : "a"(&addr), "m"(addr)
2996 static int hardware_enable(void)
2998 int cpu = raw_smp_processor_id();
2999 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
3002 if (cr4_read_shadow() & X86_CR4_VMXE)
3005 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
3006 INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
3007 spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
3010 * Now we can enable the vmclear operation in kdump
3011 * since the loaded_vmcss_on_cpu list on this cpu
3012 * has been initialized.
3014 * Though the cpu is not in VMX operation now, there
3015 * is no problem to enable the vmclear operation
3016 * for the loaded_vmcss_on_cpu list is empty!
3018 crash_enable_local_vmclear(cpu);
3020 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
3022 test_bits = FEATURE_CONTROL_LOCKED;
3023 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
3024 if (tboot_enabled())
3025 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
3027 if ((old & test_bits) != test_bits) {
3028 /* enable and lock */
3029 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
3031 cr4_set_bits(X86_CR4_VMXE);
3033 if (vmm_exclusive) {
3034 kvm_cpu_vmxon(phys_addr);
3038 native_store_gdt(this_cpu_ptr(&host_gdt));
3043 static void vmclear_local_loaded_vmcss(void)
3045 int cpu = raw_smp_processor_id();
3046 struct loaded_vmcs *v, *n;
3048 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
3049 loaded_vmcss_on_cpu_link)
3050 __loaded_vmcs_clear(v);
3054 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
3057 static void kvm_cpu_vmxoff(void)
3059 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
3062 static void hardware_disable(void)
3064 if (vmm_exclusive) {
3065 vmclear_local_loaded_vmcss();
3068 cr4_clear_bits(X86_CR4_VMXE);
3071 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
3072 u32 msr, u32 *result)
3074 u32 vmx_msr_low, vmx_msr_high;
3075 u32 ctl = ctl_min | ctl_opt;
3077 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3079 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
3080 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
3082 /* Ensure minimum (required) set of control bits are supported. */
3090 static __init bool allow_1_setting(u32 msr, u32 ctl)
3092 u32 vmx_msr_low, vmx_msr_high;
3094 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3095 return vmx_msr_high & ctl;
3098 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
3100 u32 vmx_msr_low, vmx_msr_high;
3101 u32 min, opt, min2, opt2;
3102 u32 _pin_based_exec_control = 0;
3103 u32 _cpu_based_exec_control = 0;
3104 u32 _cpu_based_2nd_exec_control = 0;
3105 u32 _vmexit_control = 0;
3106 u32 _vmentry_control = 0;
3108 min = CPU_BASED_HLT_EXITING |
3109 #ifdef CONFIG_X86_64
3110 CPU_BASED_CR8_LOAD_EXITING |
3111 CPU_BASED_CR8_STORE_EXITING |
3113 CPU_BASED_CR3_LOAD_EXITING |
3114 CPU_BASED_CR3_STORE_EXITING |
3115 CPU_BASED_USE_IO_BITMAPS |
3116 CPU_BASED_MOV_DR_EXITING |
3117 CPU_BASED_USE_TSC_OFFSETING |
3118 CPU_BASED_MWAIT_EXITING |
3119 CPU_BASED_MONITOR_EXITING |
3120 CPU_BASED_INVLPG_EXITING |
3121 CPU_BASED_RDPMC_EXITING;
3123 opt = CPU_BASED_TPR_SHADOW |
3124 CPU_BASED_USE_MSR_BITMAPS |
3125 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
3126 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
3127 &_cpu_based_exec_control) < 0)
3129 #ifdef CONFIG_X86_64
3130 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3131 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
3132 ~CPU_BASED_CR8_STORE_EXITING;
3134 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
3136 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
3137 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3138 SECONDARY_EXEC_WBINVD_EXITING |
3139 SECONDARY_EXEC_ENABLE_VPID |
3140 SECONDARY_EXEC_ENABLE_EPT |
3141 SECONDARY_EXEC_UNRESTRICTED_GUEST |
3142 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
3143 SECONDARY_EXEC_RDTSCP |
3144 SECONDARY_EXEC_ENABLE_INVPCID |
3145 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3146 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
3147 SECONDARY_EXEC_SHADOW_VMCS |
3148 SECONDARY_EXEC_XSAVES |
3149 SECONDARY_EXEC_ENABLE_PML |
3150 SECONDARY_EXEC_PCOMMIT |
3151 SECONDARY_EXEC_TSC_SCALING;
3152 if (adjust_vmx_controls(min2, opt2,
3153 MSR_IA32_VMX_PROCBASED_CTLS2,
3154 &_cpu_based_2nd_exec_control) < 0)
3157 #ifndef CONFIG_X86_64
3158 if (!(_cpu_based_2nd_exec_control &
3159 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
3160 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
3163 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3164 _cpu_based_2nd_exec_control &= ~(
3165 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3166 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3167 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3169 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
3170 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
3172 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
3173 CPU_BASED_CR3_STORE_EXITING |
3174 CPU_BASED_INVLPG_EXITING);
3175 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
3176 vmx_capability.ept, vmx_capability.vpid);
3179 min = VM_EXIT_SAVE_DEBUG_CONTROLS;
3180 #ifdef CONFIG_X86_64
3181 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
3183 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT |
3184 VM_EXIT_ACK_INTR_ON_EXIT | VM_EXIT_CLEAR_BNDCFGS;
3185 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
3186 &_vmexit_control) < 0)
3189 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
3190 opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR;
3191 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
3192 &_pin_based_exec_control) < 0)
3195 if (!(_cpu_based_2nd_exec_control &
3196 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) ||
3197 !(_vmexit_control & VM_EXIT_ACK_INTR_ON_EXIT))
3198 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
3200 min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
3201 opt = VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS;
3202 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
3203 &_vmentry_control) < 0)
3206 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
3208 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
3209 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
3212 #ifdef CONFIG_X86_64
3213 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
3214 if (vmx_msr_high & (1u<<16))
3218 /* Require Write-Back (WB) memory type for VMCS accesses. */
3219 if (((vmx_msr_high >> 18) & 15) != 6)
3222 vmcs_conf->size = vmx_msr_high & 0x1fff;
3223 vmcs_conf->order = get_order(vmcs_config.size);
3224 vmcs_conf->revision_id = vmx_msr_low;
3226 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
3227 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
3228 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
3229 vmcs_conf->vmexit_ctrl = _vmexit_control;
3230 vmcs_conf->vmentry_ctrl = _vmentry_control;
3232 cpu_has_load_ia32_efer =
3233 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3234 VM_ENTRY_LOAD_IA32_EFER)
3235 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3236 VM_EXIT_LOAD_IA32_EFER);
3238 cpu_has_load_perf_global_ctrl =
3239 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3240 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
3241 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3242 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
3245 * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
3246 * but due to arrata below it can't be used. Workaround is to use
3247 * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
3249 * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
3254 * BC86,AAY89,BD102 (model 44)
3258 if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
3259 switch (boot_cpu_data.x86_model) {
3265 cpu_has_load_perf_global_ctrl = false;
3266 printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
3267 "does not work properly. Using workaround\n");
3275 rdmsrl(MSR_IA32_XSS, host_xss);
3280 static struct vmcs *alloc_vmcs_cpu(int cpu)
3282 int node = cpu_to_node(cpu);
3286 pages = __alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
3289 vmcs = page_address(pages);
3290 memset(vmcs, 0, vmcs_config.size);
3291 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
3295 static struct vmcs *alloc_vmcs(void)
3297 return alloc_vmcs_cpu(raw_smp_processor_id());
3300 static void free_vmcs(struct vmcs *vmcs)
3302 free_pages((unsigned long)vmcs, vmcs_config.order);
3306 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
3308 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
3310 if (!loaded_vmcs->vmcs)
3312 loaded_vmcs_clear(loaded_vmcs);
3313 free_vmcs(loaded_vmcs->vmcs);
3314 loaded_vmcs->vmcs = NULL;
3317 static void free_kvm_area(void)
3321 for_each_possible_cpu(cpu) {
3322 free_vmcs(per_cpu(vmxarea, cpu));
3323 per_cpu(vmxarea, cpu) = NULL;
3327 static void init_vmcs_shadow_fields(void)
3331 /* No checks for read only fields yet */
3333 for (i = j = 0; i < max_shadow_read_write_fields; i++) {
3334 switch (shadow_read_write_fields[i]) {
3336 if (!vmx_mpx_supported())
3344 shadow_read_write_fields[j] =
3345 shadow_read_write_fields[i];
3348 max_shadow_read_write_fields = j;
3350 /* shadowed fields guest access without vmexit */
3351 for (i = 0; i < max_shadow_read_write_fields; i++) {
3352 clear_bit(shadow_read_write_fields[i],
3353 vmx_vmwrite_bitmap);
3354 clear_bit(shadow_read_write_fields[i],
3357 for (i = 0; i < max_shadow_read_only_fields; i++)
3358 clear_bit(shadow_read_only_fields[i],
3362 static __init int alloc_kvm_area(void)
3366 for_each_possible_cpu(cpu) {
3369 vmcs = alloc_vmcs_cpu(cpu);
3375 per_cpu(vmxarea, cpu) = vmcs;
3380 static bool emulation_required(struct kvm_vcpu *vcpu)
3382 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
3385 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
3386 struct kvm_segment *save)
3388 if (!emulate_invalid_guest_state) {
3390 * CS and SS RPL should be equal during guest entry according
3391 * to VMX spec, but in reality it is not always so. Since vcpu
3392 * is in the middle of the transition from real mode to
3393 * protected mode it is safe to assume that RPL 0 is a good
3396 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
3397 save->selector &= ~SEGMENT_RPL_MASK;
3398 save->dpl = save->selector & SEGMENT_RPL_MASK;
3401 vmx_set_segment(vcpu, save, seg);
3404 static void enter_pmode(struct kvm_vcpu *vcpu)
3406 unsigned long flags;
3407 struct vcpu_vmx *vmx = to_vmx(vcpu);
3410 * Update real mode segment cache. It may be not up-to-date if sement
3411 * register was written while vcpu was in a guest mode.
3413 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3414 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3415 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3416 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3417 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3418 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3420 vmx->rmode.vm86_active = 0;
3422 vmx_segment_cache_clear(vmx);
3424 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3426 flags = vmcs_readl(GUEST_RFLAGS);
3427 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
3428 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
3429 vmcs_writel(GUEST_RFLAGS, flags);
3431 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
3432 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
3434 update_exception_bitmap(vcpu);
3436 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3437 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3438 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3439 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3440 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3441 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3444 static void fix_rmode_seg(int seg, struct kvm_segment *save)
3446 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3447 struct kvm_segment var = *save;
3450 if (seg == VCPU_SREG_CS)
3453 if (!emulate_invalid_guest_state) {
3454 var.selector = var.base >> 4;
3455 var.base = var.base & 0xffff0;
3465 if (save->base & 0xf)
3466 printk_once(KERN_WARNING "kvm: segment base is not "
3467 "paragraph aligned when entering "
3468 "protected mode (seg=%d)", seg);
3471 vmcs_write16(sf->selector, var.selector);
3472 vmcs_write32(sf->base, var.base);
3473 vmcs_write32(sf->limit, var.limit);
3474 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
3477 static void enter_rmode(struct kvm_vcpu *vcpu)
3479 unsigned long flags;
3480 struct vcpu_vmx *vmx = to_vmx(vcpu);
3482 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3483 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3484 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3485 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3486 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3487 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3488 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3490 vmx->rmode.vm86_active = 1;
3493 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
3494 * vcpu. Warn the user that an update is overdue.
3496 if (!vcpu->kvm->arch.tss_addr)
3497 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
3498 "called before entering vcpu\n");
3500 vmx_segment_cache_clear(vmx);
3502 vmcs_writel(GUEST_TR_BASE, vcpu->kvm->arch.tss_addr);
3503 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
3504 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3506 flags = vmcs_readl(GUEST_RFLAGS);
3507 vmx->rmode.save_rflags = flags;
3509 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
3511 vmcs_writel(GUEST_RFLAGS, flags);
3512 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
3513 update_exception_bitmap(vcpu);
3515 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3516 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3517 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3518 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3519 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3520 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3522 kvm_mmu_reset_context(vcpu);
3525 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
3527 struct vcpu_vmx *vmx = to_vmx(vcpu);
3528 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
3534 * Force kernel_gs_base reloading before EFER changes, as control
3535 * of this msr depends on is_long_mode().
3537 vmx_load_host_state(to_vmx(vcpu));
3538 vcpu->arch.efer = efer;
3539 if (efer & EFER_LMA) {
3540 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3543 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3545 msr->data = efer & ~EFER_LME;
3550 #ifdef CONFIG_X86_64
3552 static void enter_lmode(struct kvm_vcpu *vcpu)
3556 vmx_segment_cache_clear(to_vmx(vcpu));
3558 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
3559 if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
3560 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
3562 vmcs_write32(GUEST_TR_AR_BYTES,
3563 (guest_tr_ar & ~VMX_AR_TYPE_MASK)
3564 | VMX_AR_TYPE_BUSY_64_TSS);
3566 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
3569 static void exit_lmode(struct kvm_vcpu *vcpu)
3571 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3572 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
3577 static inline void __vmx_flush_tlb(struct kvm_vcpu *vcpu, int vpid)
3579 vpid_sync_context(vpid);
3581 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
3583 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
3587 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
3589 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->vpid);
3592 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
3594 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
3596 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
3597 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
3600 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
3602 if (enable_ept && is_paging(vcpu))
3603 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3604 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
3607 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
3609 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
3611 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
3612 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
3615 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
3617 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3619 if (!test_bit(VCPU_EXREG_PDPTR,
3620 (unsigned long *)&vcpu->arch.regs_dirty))
3623 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3624 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
3625 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
3626 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
3627 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
3631 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
3633 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3635 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3636 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
3637 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
3638 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
3639 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
3642 __set_bit(VCPU_EXREG_PDPTR,
3643 (unsigned long *)&vcpu->arch.regs_avail);
3644 __set_bit(VCPU_EXREG_PDPTR,
3645 (unsigned long *)&vcpu->arch.regs_dirty);
3648 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
3650 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
3652 struct kvm_vcpu *vcpu)
3654 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3655 vmx_decache_cr3(vcpu);
3656 if (!(cr0 & X86_CR0_PG)) {
3657 /* From paging/starting to nonpaging */
3658 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3659 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
3660 (CPU_BASED_CR3_LOAD_EXITING |
3661 CPU_BASED_CR3_STORE_EXITING));
3662 vcpu->arch.cr0 = cr0;
3663 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3664 } else if (!is_paging(vcpu)) {
3665 /* From nonpaging to paging */
3666 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3667 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
3668 ~(CPU_BASED_CR3_LOAD_EXITING |
3669 CPU_BASED_CR3_STORE_EXITING));
3670 vcpu->arch.cr0 = cr0;
3671 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3674 if (!(cr0 & X86_CR0_WP))
3675 *hw_cr0 &= ~X86_CR0_WP;
3678 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
3680 struct vcpu_vmx *vmx = to_vmx(vcpu);
3681 unsigned long hw_cr0;
3683 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK);
3684 if (enable_unrestricted_guest)
3685 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
3687 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
3689 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
3692 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
3696 #ifdef CONFIG_X86_64
3697 if (vcpu->arch.efer & EFER_LME) {
3698 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
3700 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
3706 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
3708 if (!vcpu->fpu_active)
3709 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
3711 vmcs_writel(CR0_READ_SHADOW, cr0);
3712 vmcs_writel(GUEST_CR0, hw_cr0);
3713 vcpu->arch.cr0 = cr0;
3715 /* depends on vcpu->arch.cr0 to be set to a new value */
3716 vmx->emulation_required = emulation_required(vcpu);
3719 static u64 construct_eptp(unsigned long root_hpa)
3723 /* TODO write the value reading from MSR */
3724 eptp = VMX_EPT_DEFAULT_MT |
3725 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
3726 if (enable_ept_ad_bits)
3727 eptp |= VMX_EPT_AD_ENABLE_BIT;
3728 eptp |= (root_hpa & PAGE_MASK);
3733 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
3735 unsigned long guest_cr3;
3740 eptp = construct_eptp(cr3);
3741 vmcs_write64(EPT_POINTER, eptp);
3742 if (is_paging(vcpu) || is_guest_mode(vcpu))
3743 guest_cr3 = kvm_read_cr3(vcpu);
3745 guest_cr3 = vcpu->kvm->arch.ept_identity_map_addr;
3746 ept_load_pdptrs(vcpu);
3749 vmx_flush_tlb(vcpu);
3750 vmcs_writel(GUEST_CR3, guest_cr3);
3753 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3756 * Pass through host's Machine Check Enable value to hw_cr4, which
3757 * is in force while we are in guest mode. Do not let guests control
3758 * this bit, even if host CR4.MCE == 0.
3760 unsigned long hw_cr4 =
3761 (cr4_read_shadow() & X86_CR4_MCE) |
3762 (cr4 & ~X86_CR4_MCE) |
3763 (to_vmx(vcpu)->rmode.vm86_active ?
3764 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
3766 if (cr4 & X86_CR4_VMXE) {
3768 * To use VMXON (and later other VMX instructions), a guest
3769 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3770 * So basically the check on whether to allow nested VMX
3773 if (!nested_vmx_allowed(vcpu))
3776 if (to_vmx(vcpu)->nested.vmxon &&
3777 ((cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON))
3780 vcpu->arch.cr4 = cr4;
3782 if (!is_paging(vcpu)) {
3783 hw_cr4 &= ~X86_CR4_PAE;
3784 hw_cr4 |= X86_CR4_PSE;
3785 } else if (!(cr4 & X86_CR4_PAE)) {
3786 hw_cr4 &= ~X86_CR4_PAE;
3790 if (!enable_unrestricted_guest && !is_paging(vcpu))
3792 * SMEP/SMAP is disabled if CPU is in non-paging mode in
3793 * hardware. However KVM always uses paging mode without
3794 * unrestricted guest.
3795 * To emulate this behavior, SMEP/SMAP needs to be manually
3796 * disabled when guest switches to non-paging mode.
3798 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP);
3800 vmcs_writel(CR4_READ_SHADOW, cr4);
3801 vmcs_writel(GUEST_CR4, hw_cr4);
3805 static void vmx_get_segment(struct kvm_vcpu *vcpu,
3806 struct kvm_segment *var, int seg)
3808 struct vcpu_vmx *vmx = to_vmx(vcpu);
3811 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3812 *var = vmx->rmode.segs[seg];
3813 if (seg == VCPU_SREG_TR
3814 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3816 var->base = vmx_read_guest_seg_base(vmx, seg);
3817 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3820 var->base = vmx_read_guest_seg_base(vmx, seg);
3821 var->limit = vmx_read_guest_seg_limit(vmx, seg);
3822 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3823 ar = vmx_read_guest_seg_ar(vmx, seg);
3824 var->unusable = (ar >> 16) & 1;
3825 var->type = ar & 15;
3826 var->s = (ar >> 4) & 1;
3827 var->dpl = (ar >> 5) & 3;
3829 * Some userspaces do not preserve unusable property. Since usable
3830 * segment has to be present according to VMX spec we can use present
3831 * property to amend userspace bug by making unusable segment always
3832 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3833 * segment as unusable.
3835 var->present = !var->unusable;
3836 var->avl = (ar >> 12) & 1;
3837 var->l = (ar >> 13) & 1;
3838 var->db = (ar >> 14) & 1;
3839 var->g = (ar >> 15) & 1;
3842 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3844 struct kvm_segment s;
3846 if (to_vmx(vcpu)->rmode.vm86_active) {
3847 vmx_get_segment(vcpu, &s, seg);
3850 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3853 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
3855 struct vcpu_vmx *vmx = to_vmx(vcpu);
3857 if (unlikely(vmx->rmode.vm86_active))
3860 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3861 return VMX_AR_DPL(ar);
3865 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3869 if (var->unusable || !var->present)
3872 ar = var->type & 15;
3873 ar |= (var->s & 1) << 4;
3874 ar |= (var->dpl & 3) << 5;
3875 ar |= (var->present & 1) << 7;
3876 ar |= (var->avl & 1) << 12;
3877 ar |= (var->l & 1) << 13;
3878 ar |= (var->db & 1) << 14;
3879 ar |= (var->g & 1) << 15;
3885 static void vmx_set_segment(struct kvm_vcpu *vcpu,
3886 struct kvm_segment *var, int seg)
3888 struct vcpu_vmx *vmx = to_vmx(vcpu);
3889 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3891 vmx_segment_cache_clear(vmx);
3893 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3894 vmx->rmode.segs[seg] = *var;
3895 if (seg == VCPU_SREG_TR)
3896 vmcs_write16(sf->selector, var->selector);
3898 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3902 vmcs_writel(sf->base, var->base);
3903 vmcs_write32(sf->limit, var->limit);
3904 vmcs_write16(sf->selector, var->selector);
3907 * Fix the "Accessed" bit in AR field of segment registers for older
3909 * IA32 arch specifies that at the time of processor reset the
3910 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3911 * is setting it to 0 in the userland code. This causes invalid guest
3912 * state vmexit when "unrestricted guest" mode is turned on.
3913 * Fix for this setup issue in cpu_reset is being pushed in the qemu
3914 * tree. Newer qemu binaries with that qemu fix would not need this
3917 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3918 var->type |= 0x1; /* Accessed */
3920 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3923 vmx->emulation_required = emulation_required(vcpu);
3926 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3928 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3930 *db = (ar >> 14) & 1;
3931 *l = (ar >> 13) & 1;
3934 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3936 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3937 dt->address = vmcs_readl(GUEST_IDTR_BASE);
3940 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3942 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3943 vmcs_writel(GUEST_IDTR_BASE, dt->address);
3946 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3948 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3949 dt->address = vmcs_readl(GUEST_GDTR_BASE);
3952 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3954 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3955 vmcs_writel(GUEST_GDTR_BASE, dt->address);
3958 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3960 struct kvm_segment var;
3963 vmx_get_segment(vcpu, &var, seg);
3965 if (seg == VCPU_SREG_CS)
3967 ar = vmx_segment_access_rights(&var);
3969 if (var.base != (var.selector << 4))
3971 if (var.limit != 0xffff)
3979 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3981 struct kvm_segment cs;
3982 unsigned int cs_rpl;
3984 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3985 cs_rpl = cs.selector & SEGMENT_RPL_MASK;
3989 if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
3993 if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
3994 if (cs.dpl > cs_rpl)
3997 if (cs.dpl != cs_rpl)
4003 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
4007 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
4009 struct kvm_segment ss;
4010 unsigned int ss_rpl;
4012 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4013 ss_rpl = ss.selector & SEGMENT_RPL_MASK;
4017 if (ss.type != 3 && ss.type != 7)
4021 if (ss.dpl != ss_rpl) /* DPL != RPL */
4029 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
4031 struct kvm_segment var;
4034 vmx_get_segment(vcpu, &var, seg);
4035 rpl = var.selector & SEGMENT_RPL_MASK;
4043 if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
4044 if (var.dpl < rpl) /* DPL < RPL */
4048 /* TODO: Add other members to kvm_segment_field to allow checking for other access
4054 static bool tr_valid(struct kvm_vcpu *vcpu)
4056 struct kvm_segment tr;
4058 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
4062 if (tr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4064 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
4072 static bool ldtr_valid(struct kvm_vcpu *vcpu)
4074 struct kvm_segment ldtr;
4076 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
4080 if (ldtr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4090 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
4092 struct kvm_segment cs, ss;
4094 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4095 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4097 return ((cs.selector & SEGMENT_RPL_MASK) ==
4098 (ss.selector & SEGMENT_RPL_MASK));
4102 * Check if guest state is valid. Returns true if valid, false if
4104 * We assume that registers are always usable
4106 static bool guest_state_valid(struct kvm_vcpu *vcpu)
4108 if (enable_unrestricted_guest)
4111 /* real mode guest state checks */
4112 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
4113 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
4115 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
4117 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
4119 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
4121 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
4123 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
4126 /* protected mode guest state checks */
4127 if (!cs_ss_rpl_check(vcpu))
4129 if (!code_segment_valid(vcpu))
4131 if (!stack_segment_valid(vcpu))
4133 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
4135 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
4137 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
4139 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
4141 if (!tr_valid(vcpu))
4143 if (!ldtr_valid(vcpu))
4147 * - Add checks on RIP
4148 * - Add checks on RFLAGS
4154 static int init_rmode_tss(struct kvm *kvm)
4160 idx = srcu_read_lock(&kvm->srcu);
4161 fn = kvm->arch.tss_addr >> PAGE_SHIFT;
4162 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4165 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
4166 r = kvm_write_guest_page(kvm, fn++, &data,
4167 TSS_IOPB_BASE_OFFSET, sizeof(u16));
4170 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
4173 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4177 r = kvm_write_guest_page(kvm, fn, &data,
4178 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
4181 srcu_read_unlock(&kvm->srcu, idx);
4185 static int init_rmode_identity_map(struct kvm *kvm)
4188 pfn_t identity_map_pfn;
4194 /* Protect kvm->arch.ept_identity_pagetable_done. */
4195 mutex_lock(&kvm->slots_lock);
4197 if (likely(kvm->arch.ept_identity_pagetable_done))
4200 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
4202 r = alloc_identity_pagetable(kvm);
4206 idx = srcu_read_lock(&kvm->srcu);
4207 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
4210 /* Set up identity-mapping pagetable for EPT in real mode */
4211 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
4212 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
4213 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
4214 r = kvm_write_guest_page(kvm, identity_map_pfn,
4215 &tmp, i * sizeof(tmp), sizeof(tmp));
4219 kvm->arch.ept_identity_pagetable_done = true;
4222 srcu_read_unlock(&kvm->srcu, idx);
4225 mutex_unlock(&kvm->slots_lock);
4229 static void seg_setup(int seg)
4231 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4234 vmcs_write16(sf->selector, 0);
4235 vmcs_writel(sf->base, 0);
4236 vmcs_write32(sf->limit, 0xffff);
4238 if (seg == VCPU_SREG_CS)
4239 ar |= 0x08; /* code segment */
4241 vmcs_write32(sf->ar_bytes, ar);
4244 static int alloc_apic_access_page(struct kvm *kvm)
4249 mutex_lock(&kvm->slots_lock);
4250 if (kvm->arch.apic_access_page_done)
4252 r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
4253 APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
4257 page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
4258 if (is_error_page(page)) {
4264 * Do not pin the page in memory, so that memory hot-unplug
4265 * is able to migrate it.
4268 kvm->arch.apic_access_page_done = true;
4270 mutex_unlock(&kvm->slots_lock);
4274 static int alloc_identity_pagetable(struct kvm *kvm)
4276 /* Called with kvm->slots_lock held. */
4280 BUG_ON(kvm->arch.ept_identity_pagetable_done);
4282 r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
4283 kvm->arch.ept_identity_map_addr, PAGE_SIZE);
4288 static int allocate_vpid(void)
4294 spin_lock(&vmx_vpid_lock);
4295 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
4296 if (vpid < VMX_NR_VPIDS)
4297 __set_bit(vpid, vmx_vpid_bitmap);
4300 spin_unlock(&vmx_vpid_lock);
4304 static void free_vpid(int vpid)
4306 if (!enable_vpid || vpid == 0)
4308 spin_lock(&vmx_vpid_lock);
4309 __clear_bit(vpid, vmx_vpid_bitmap);
4310 spin_unlock(&vmx_vpid_lock);
4313 #define MSR_TYPE_R 1
4314 #define MSR_TYPE_W 2
4315 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
4318 int f = sizeof(unsigned long);
4320 if (!cpu_has_vmx_msr_bitmap())
4324 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4325 * have the write-low and read-high bitmap offsets the wrong way round.
4326 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4328 if (msr <= 0x1fff) {
4329 if (type & MSR_TYPE_R)
4331 __clear_bit(msr, msr_bitmap + 0x000 / f);
4333 if (type & MSR_TYPE_W)
4335 __clear_bit(msr, msr_bitmap + 0x800 / f);
4337 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4339 if (type & MSR_TYPE_R)
4341 __clear_bit(msr, msr_bitmap + 0x400 / f);
4343 if (type & MSR_TYPE_W)
4345 __clear_bit(msr, msr_bitmap + 0xc00 / f);
4350 static void __vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
4353 int f = sizeof(unsigned long);
4355 if (!cpu_has_vmx_msr_bitmap())
4359 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4360 * have the write-low and read-high bitmap offsets the wrong way round.
4361 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4363 if (msr <= 0x1fff) {
4364 if (type & MSR_TYPE_R)
4366 __set_bit(msr, msr_bitmap + 0x000 / f);
4368 if (type & MSR_TYPE_W)
4370 __set_bit(msr, msr_bitmap + 0x800 / f);
4372 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4374 if (type & MSR_TYPE_R)
4376 __set_bit(msr, msr_bitmap + 0x400 / f);
4378 if (type & MSR_TYPE_W)
4380 __set_bit(msr, msr_bitmap + 0xc00 / f);
4386 * If a msr is allowed by L0, we should check whether it is allowed by L1.
4387 * The corresponding bit will be cleared unless both of L0 and L1 allow it.
4389 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
4390 unsigned long *msr_bitmap_nested,
4393 int f = sizeof(unsigned long);
4395 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 &&
4407 !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
4409 __clear_bit(msr, msr_bitmap_nested + 0x000 / f);
4411 if (type & MSR_TYPE_W &&
4412 !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
4414 __clear_bit(msr, msr_bitmap_nested + 0x800 / f);
4416 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4418 if (type & MSR_TYPE_R &&
4419 !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
4421 __clear_bit(msr, msr_bitmap_nested + 0x400 / f);
4423 if (type & MSR_TYPE_W &&
4424 !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
4426 __clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
4431 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
4434 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy,
4435 msr, MSR_TYPE_R | MSR_TYPE_W);
4436 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode,
4437 msr, MSR_TYPE_R | MSR_TYPE_W);
4440 static void vmx_enable_intercept_msr_read_x2apic(u32 msr)
4442 __vmx_enable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4444 __vmx_enable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4448 static void vmx_disable_intercept_msr_read_x2apic(u32 msr)
4450 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4452 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4456 static void vmx_disable_intercept_msr_write_x2apic(u32 msr)
4458 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4460 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4464 static bool vmx_get_enable_apicv(void)
4466 return enable_apicv;
4469 static int vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
4471 struct vcpu_vmx *vmx = to_vmx(vcpu);
4476 if (vmx->nested.pi_desc &&
4477 vmx->nested.pi_pending) {
4478 vmx->nested.pi_pending = false;
4479 if (!pi_test_and_clear_on(vmx->nested.pi_desc))
4482 max_irr = find_last_bit(
4483 (unsigned long *)vmx->nested.pi_desc->pir, 256);
4488 vapic_page = kmap(vmx->nested.virtual_apic_page);
4493 __kvm_apic_update_irr(vmx->nested.pi_desc->pir, vapic_page);
4494 kunmap(vmx->nested.virtual_apic_page);
4496 status = vmcs_read16(GUEST_INTR_STATUS);
4497 if ((u8)max_irr > ((u8)status & 0xff)) {
4499 status |= (u8)max_irr;
4500 vmcs_write16(GUEST_INTR_STATUS, status);
4506 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu)
4509 if (vcpu->mode == IN_GUEST_MODE) {
4510 struct vcpu_vmx *vmx = to_vmx(vcpu);
4513 * Currently, we don't support urgent interrupt,
4514 * all interrupts are recognized as non-urgent
4515 * interrupt, so we cannot post interrupts when
4518 * If the vcpu is in guest mode, it means it is
4519 * running instead of being scheduled out and
4520 * waiting in the run queue, and that's the only
4521 * case when 'SN' is set currently, warning if
4524 WARN_ON_ONCE(pi_test_sn(&vmx->pi_desc));
4526 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu),
4527 POSTED_INTR_VECTOR);
4534 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
4537 struct vcpu_vmx *vmx = to_vmx(vcpu);
4539 if (is_guest_mode(vcpu) &&
4540 vector == vmx->nested.posted_intr_nv) {
4541 /* the PIR and ON have been set by L1. */
4542 kvm_vcpu_trigger_posted_interrupt(vcpu);
4544 * If a posted intr is not recognized by hardware,
4545 * we will accomplish it in the next vmentry.
4547 vmx->nested.pi_pending = true;
4548 kvm_make_request(KVM_REQ_EVENT, vcpu);
4554 * Send interrupt to vcpu via posted interrupt way.
4555 * 1. If target vcpu is running(non-root mode), send posted interrupt
4556 * notification to vcpu and hardware will sync PIR to vIRR atomically.
4557 * 2. If target vcpu isn't running(root mode), kick it to pick up the
4558 * interrupt from PIR in next vmentry.
4560 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
4562 struct vcpu_vmx *vmx = to_vmx(vcpu);
4565 r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
4569 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
4572 r = pi_test_and_set_on(&vmx->pi_desc);
4573 kvm_make_request(KVM_REQ_EVENT, vcpu);
4574 if (r || !kvm_vcpu_trigger_posted_interrupt(vcpu))
4575 kvm_vcpu_kick(vcpu);
4578 static void vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
4580 struct vcpu_vmx *vmx = to_vmx(vcpu);
4582 if (!pi_test_and_clear_on(&vmx->pi_desc))
4585 kvm_apic_update_irr(vcpu, vmx->pi_desc.pir);
4589 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
4590 * will not change in the lifetime of the guest.
4591 * Note that host-state that does change is set elsewhere. E.g., host-state
4592 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
4594 static void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
4601 vmcs_writel(HOST_CR0, read_cr0() & ~X86_CR0_TS); /* 22.2.3 */
4602 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
4604 /* Save the most likely value for this task's CR4 in the VMCS. */
4605 cr4 = cr4_read_shadow();
4606 vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
4607 vmx->host_state.vmcs_host_cr4 = cr4;
4609 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
4610 #ifdef CONFIG_X86_64
4612 * Load null selectors, so we can avoid reloading them in
4613 * __vmx_load_host_state(), in case userspace uses the null selectors
4614 * too (the expected case).
4616 vmcs_write16(HOST_DS_SELECTOR, 0);
4617 vmcs_write16(HOST_ES_SELECTOR, 0);
4619 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4620 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4622 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4623 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
4625 native_store_idt(&dt);
4626 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
4627 vmx->host_idt_base = dt.address;
4629 vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
4631 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
4632 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
4633 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
4634 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
4636 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
4637 rdmsr(MSR_IA32_CR_PAT, low32, high32);
4638 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
4642 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
4644 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
4646 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
4647 if (is_guest_mode(&vmx->vcpu))
4648 vmx->vcpu.arch.cr4_guest_owned_bits &=
4649 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
4650 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
4653 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
4655 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
4657 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
4658 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
4659 return pin_based_exec_ctrl;
4662 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
4664 struct vcpu_vmx *vmx = to_vmx(vcpu);
4666 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
4669 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
4671 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
4673 if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
4674 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
4676 if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
4677 exec_control &= ~CPU_BASED_TPR_SHADOW;
4678 #ifdef CONFIG_X86_64
4679 exec_control |= CPU_BASED_CR8_STORE_EXITING |
4680 CPU_BASED_CR8_LOAD_EXITING;
4684 exec_control |= CPU_BASED_CR3_STORE_EXITING |
4685 CPU_BASED_CR3_LOAD_EXITING |
4686 CPU_BASED_INVLPG_EXITING;
4687 return exec_control;
4690 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
4692 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4693 if (!cpu_need_virtualize_apic_accesses(&vmx->vcpu))
4694 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4696 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4698 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4699 enable_unrestricted_guest = 0;
4700 /* Enable INVPCID for non-ept guests may cause performance regression. */
4701 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4703 if (!enable_unrestricted_guest)
4704 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4706 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4707 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
4708 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4709 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4710 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4711 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4713 We can NOT enable shadow_vmcs here because we don't have yet
4716 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4719 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4721 /* Currently, we allow L1 guest to directly run pcommit instruction. */
4722 exec_control &= ~SECONDARY_EXEC_PCOMMIT;
4724 return exec_control;
4727 static void ept_set_mmio_spte_mask(void)
4730 * EPT Misconfigurations can be generated if the value of bits 2:0
4731 * of an EPT paging-structure entry is 110b (write/execute).
4732 * Also, magic bits (0x3ull << 62) is set to quickly identify mmio
4735 kvm_mmu_set_mmio_spte_mask((0x3ull << 62) | 0x6ull);
4738 #define VMX_XSS_EXIT_BITMAP 0
4740 * Sets up the vmcs for emulated real mode.
4742 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
4744 #ifdef CONFIG_X86_64
4750 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
4751 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
4753 if (enable_shadow_vmcs) {
4754 vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
4755 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
4757 if (cpu_has_vmx_msr_bitmap())
4758 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
4760 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4763 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
4765 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
4767 if (cpu_has_secondary_exec_ctrls())
4768 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
4769 vmx_secondary_exec_control(vmx));
4771 if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
4772 vmcs_write64(EOI_EXIT_BITMAP0, 0);
4773 vmcs_write64(EOI_EXIT_BITMAP1, 0);
4774 vmcs_write64(EOI_EXIT_BITMAP2, 0);
4775 vmcs_write64(EOI_EXIT_BITMAP3, 0);
4777 vmcs_write16(GUEST_INTR_STATUS, 0);
4779 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4780 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4784 vmcs_write32(PLE_GAP, ple_gap);
4785 vmx->ple_window = ple_window;
4786 vmx->ple_window_dirty = true;
4789 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4790 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4791 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
4793 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
4794 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
4795 vmx_set_constant_host_state(vmx);
4796 #ifdef CONFIG_X86_64
4797 rdmsrl(MSR_FS_BASE, a);
4798 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
4799 rdmsrl(MSR_GS_BASE, a);
4800 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
4802 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4803 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4806 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4807 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4808 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
4809 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4810 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
4812 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4813 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4815 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
4816 u32 index = vmx_msr_index[i];
4817 u32 data_low, data_high;
4820 if (rdmsr_safe(index, &data_low, &data_high) < 0)
4822 if (wrmsr_safe(index, data_low, data_high) < 0)
4824 vmx->guest_msrs[j].index = i;
4825 vmx->guest_msrs[j].data = 0;
4826 vmx->guest_msrs[j].mask = -1ull;
4831 vm_exit_controls_init(vmx, vmcs_config.vmexit_ctrl);
4833 /* 22.2.1, 20.8.1 */
4834 vm_entry_controls_init(vmx, vmcs_config.vmentry_ctrl);
4836 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
4837 set_cr4_guest_host_mask(vmx);
4839 if (vmx_xsaves_supported())
4840 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4845 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4847 struct vcpu_vmx *vmx = to_vmx(vcpu);
4848 struct msr_data apic_base_msr;
4851 vmx->rmode.vm86_active = 0;
4853 vmx->soft_vnmi_blocked = 0;
4855 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4856 kvm_set_cr8(vcpu, 0);
4859 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
4860 MSR_IA32_APICBASE_ENABLE;
4861 if (kvm_vcpu_is_reset_bsp(vcpu))
4862 apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4863 apic_base_msr.host_initiated = true;
4864 kvm_set_apic_base(vcpu, &apic_base_msr);
4867 vmx_segment_cache_clear(vmx);
4869 seg_setup(VCPU_SREG_CS);
4870 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4871 vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
4873 seg_setup(VCPU_SREG_DS);
4874 seg_setup(VCPU_SREG_ES);
4875 seg_setup(VCPU_SREG_FS);
4876 seg_setup(VCPU_SREG_GS);
4877 seg_setup(VCPU_SREG_SS);
4879 vmcs_write16(GUEST_TR_SELECTOR, 0);
4880 vmcs_writel(GUEST_TR_BASE, 0);
4881 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4882 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4884 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4885 vmcs_writel(GUEST_LDTR_BASE, 0);
4886 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4887 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4890 vmcs_write32(GUEST_SYSENTER_CS, 0);
4891 vmcs_writel(GUEST_SYSENTER_ESP, 0);
4892 vmcs_writel(GUEST_SYSENTER_EIP, 0);
4893 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4896 vmcs_writel(GUEST_RFLAGS, 0x02);
4897 kvm_rip_write(vcpu, 0xfff0);
4899 vmcs_writel(GUEST_GDTR_BASE, 0);
4900 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4902 vmcs_writel(GUEST_IDTR_BASE, 0);
4903 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4905 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4906 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4907 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4911 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
4913 if (cpu_has_vmx_tpr_shadow() && !init_event) {
4914 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4915 if (cpu_need_tpr_shadow(vcpu))
4916 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4917 __pa(vcpu->arch.apic->regs));
4918 vmcs_write32(TPR_THRESHOLD, 0);
4921 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4923 if (kvm_vcpu_apicv_active(vcpu))
4924 memset(&vmx->pi_desc, 0, sizeof(struct pi_desc));
4927 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4929 cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4930 vmx_set_cr0(vcpu, cr0); /* enter rmode */
4931 vmx->vcpu.arch.cr0 = cr0;
4932 vmx_set_cr4(vcpu, 0);
4933 vmx_set_efer(vcpu, 0);
4934 vmx_fpu_activate(vcpu);
4935 update_exception_bitmap(vcpu);
4937 vpid_sync_context(vmx->vpid);
4941 * In nested virtualization, check if L1 asked to exit on external interrupts.
4942 * For most existing hypervisors, this will always return true.
4944 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
4946 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
4947 PIN_BASED_EXT_INTR_MASK;
4951 * In nested virtualization, check if L1 has set
4952 * VM_EXIT_ACK_INTR_ON_EXIT
4954 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
4956 return get_vmcs12(vcpu)->vm_exit_controls &
4957 VM_EXIT_ACK_INTR_ON_EXIT;
4960 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
4962 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
4963 PIN_BASED_NMI_EXITING;
4966 static void enable_irq_window(struct kvm_vcpu *vcpu)
4968 u32 cpu_based_vm_exec_control;
4970 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4971 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
4972 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4975 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4977 u32 cpu_based_vm_exec_control;
4979 if (!cpu_has_virtual_nmis() ||
4980 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4981 enable_irq_window(vcpu);
4985 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4986 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
4987 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4990 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4992 struct vcpu_vmx *vmx = to_vmx(vcpu);
4994 int irq = vcpu->arch.interrupt.nr;
4996 trace_kvm_inj_virq(irq);
4998 ++vcpu->stat.irq_injections;
4999 if (vmx->rmode.vm86_active) {
5001 if (vcpu->arch.interrupt.soft)
5002 inc_eip = vcpu->arch.event_exit_inst_len;
5003 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
5004 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5007 intr = irq | INTR_INFO_VALID_MASK;
5008 if (vcpu->arch.interrupt.soft) {
5009 intr |= INTR_TYPE_SOFT_INTR;
5010 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
5011 vmx->vcpu.arch.event_exit_inst_len);
5013 intr |= INTR_TYPE_EXT_INTR;
5014 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
5017 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
5019 struct vcpu_vmx *vmx = to_vmx(vcpu);
5021 if (is_guest_mode(vcpu))
5024 if (!cpu_has_virtual_nmis()) {
5026 * Tracking the NMI-blocked state in software is built upon
5027 * finding the next open IRQ window. This, in turn, depends on
5028 * well-behaving guests: They have to keep IRQs disabled at
5029 * least as long as the NMI handler runs. Otherwise we may
5030 * cause NMI nesting, maybe breaking the guest. But as this is
5031 * highly unlikely, we can live with the residual risk.
5033 vmx->soft_vnmi_blocked = 1;
5034 vmx->vnmi_blocked_time = 0;
5037 ++vcpu->stat.nmi_injections;
5038 vmx->nmi_known_unmasked = false;
5039 if (vmx->rmode.vm86_active) {
5040 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
5041 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5044 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
5045 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
5048 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
5050 if (!cpu_has_virtual_nmis())
5051 return to_vmx(vcpu)->soft_vnmi_blocked;
5052 if (to_vmx(vcpu)->nmi_known_unmasked)
5054 return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
5057 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
5059 struct vcpu_vmx *vmx = to_vmx(vcpu);
5061 if (!cpu_has_virtual_nmis()) {
5062 if (vmx->soft_vnmi_blocked != masked) {
5063 vmx->soft_vnmi_blocked = masked;
5064 vmx->vnmi_blocked_time = 0;
5067 vmx->nmi_known_unmasked = !masked;
5069 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5070 GUEST_INTR_STATE_NMI);
5072 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
5073 GUEST_INTR_STATE_NMI);
5077 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
5079 if (to_vmx(vcpu)->nested.nested_run_pending)
5082 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
5085 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5086 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
5087 | GUEST_INTR_STATE_NMI));
5090 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
5092 return (!to_vmx(vcpu)->nested.nested_run_pending &&
5093 vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
5094 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5095 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
5098 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
5102 ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
5106 kvm->arch.tss_addr = addr;
5107 return init_rmode_tss(kvm);
5110 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
5115 * Update instruction length as we may reinject the exception
5116 * from user space while in guest debugging mode.
5118 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
5119 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5120 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5124 if (vcpu->guest_debug &
5125 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5142 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
5143 int vec, u32 err_code)
5146 * Instruction with address size override prefix opcode 0x67
5147 * Cause the #SS fault with 0 error code in VM86 mode.
5149 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
5150 if (emulate_instruction(vcpu, 0) == EMULATE_DONE) {
5151 if (vcpu->arch.halt_request) {
5152 vcpu->arch.halt_request = 0;
5153 return kvm_vcpu_halt(vcpu);
5161 * Forward all other exceptions that are valid in real mode.
5162 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
5163 * the required debugging infrastructure rework.
5165 kvm_queue_exception(vcpu, vec);
5170 * Trigger machine check on the host. We assume all the MSRs are already set up
5171 * by the CPU and that we still run on the same CPU as the MCE occurred on.
5172 * We pass a fake environment to the machine check handler because we want
5173 * the guest to be always treated like user space, no matter what context
5174 * it used internally.
5176 static void kvm_machine_check(void)
5178 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
5179 struct pt_regs regs = {
5180 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
5181 .flags = X86_EFLAGS_IF,
5184 do_machine_check(®s, 0);
5188 static int handle_machine_check(struct kvm_vcpu *vcpu)
5190 /* already handled by vcpu_run */
5194 static int handle_exception(struct kvm_vcpu *vcpu)
5196 struct vcpu_vmx *vmx = to_vmx(vcpu);
5197 struct kvm_run *kvm_run = vcpu->run;
5198 u32 intr_info, ex_no, error_code;
5199 unsigned long cr2, rip, dr6;
5201 enum emulation_result er;
5203 vect_info = vmx->idt_vectoring_info;
5204 intr_info = vmx->exit_intr_info;
5206 if (is_machine_check(intr_info))
5207 return handle_machine_check(vcpu);
5209 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
5210 return 1; /* already handled by vmx_vcpu_run() */
5212 if (is_no_device(intr_info)) {
5213 vmx_fpu_activate(vcpu);
5217 if (is_invalid_opcode(intr_info)) {
5218 if (is_guest_mode(vcpu)) {
5219 kvm_queue_exception(vcpu, UD_VECTOR);
5222 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
5223 if (er != EMULATE_DONE)
5224 kvm_queue_exception(vcpu, UD_VECTOR);
5229 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
5230 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
5233 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
5234 * MMIO, it is better to report an internal error.
5235 * See the comments in vmx_handle_exit.
5237 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
5238 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
5239 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5240 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
5241 vcpu->run->internal.ndata = 3;
5242 vcpu->run->internal.data[0] = vect_info;
5243 vcpu->run->internal.data[1] = intr_info;
5244 vcpu->run->internal.data[2] = error_code;
5248 if (is_page_fault(intr_info)) {
5249 /* EPT won't cause page fault directly */
5251 cr2 = vmcs_readl(EXIT_QUALIFICATION);
5252 trace_kvm_page_fault(cr2, error_code);
5254 if (kvm_event_needs_reinjection(vcpu))
5255 kvm_mmu_unprotect_page_virt(vcpu, cr2);
5256 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
5259 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
5261 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
5262 return handle_rmode_exception(vcpu, ex_no, error_code);
5266 kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
5269 dr6 = vmcs_readl(EXIT_QUALIFICATION);
5270 if (!(vcpu->guest_debug &
5271 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
5272 vcpu->arch.dr6 &= ~15;
5273 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5274 if (!(dr6 & ~DR6_RESERVED)) /* icebp */
5275 skip_emulated_instruction(vcpu);
5277 kvm_queue_exception(vcpu, DB_VECTOR);
5280 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
5281 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
5285 * Update instruction length as we may reinject #BP from
5286 * user space while in guest debugging mode. Reading it for
5287 * #DB as well causes no harm, it is not used in that case.
5289 vmx->vcpu.arch.event_exit_inst_len =
5290 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5291 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5292 rip = kvm_rip_read(vcpu);
5293 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
5294 kvm_run->debug.arch.exception = ex_no;
5297 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
5298 kvm_run->ex.exception = ex_no;
5299 kvm_run->ex.error_code = error_code;
5305 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
5307 ++vcpu->stat.irq_exits;
5311 static int handle_triple_fault(struct kvm_vcpu *vcpu)
5313 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5317 static int handle_io(struct kvm_vcpu *vcpu)
5319 unsigned long exit_qualification;
5320 int size, in, string;
5323 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5324 string = (exit_qualification & 16) != 0;
5325 in = (exit_qualification & 8) != 0;
5327 ++vcpu->stat.io_exits;
5330 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5332 port = exit_qualification >> 16;
5333 size = (exit_qualification & 7) + 1;
5334 skip_emulated_instruction(vcpu);
5336 return kvm_fast_pio_out(vcpu, size, port);
5340 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
5343 * Patch in the VMCALL instruction:
5345 hypercall[0] = 0x0f;
5346 hypercall[1] = 0x01;
5347 hypercall[2] = 0xc1;
5350 static bool nested_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
5352 unsigned long always_on = VMXON_CR0_ALWAYSON;
5353 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5355 if (to_vmx(vcpu)->nested.nested_vmx_secondary_ctls_high &
5356 SECONDARY_EXEC_UNRESTRICTED_GUEST &&
5357 nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
5358 always_on &= ~(X86_CR0_PE | X86_CR0_PG);
5359 return (val & always_on) == always_on;
5362 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
5363 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
5365 if (is_guest_mode(vcpu)) {
5366 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5367 unsigned long orig_val = val;
5370 * We get here when L2 changed cr0 in a way that did not change
5371 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
5372 * but did change L0 shadowed bits. So we first calculate the
5373 * effective cr0 value that L1 would like to write into the
5374 * hardware. It consists of the L2-owned bits from the new
5375 * value combined with the L1-owned bits from L1's guest_cr0.
5377 val = (val & ~vmcs12->cr0_guest_host_mask) |
5378 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
5380 if (!nested_cr0_valid(vcpu, val))
5383 if (kvm_set_cr0(vcpu, val))
5385 vmcs_writel(CR0_READ_SHADOW, orig_val);
5388 if (to_vmx(vcpu)->nested.vmxon &&
5389 ((val & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON))
5391 return kvm_set_cr0(vcpu, val);
5395 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
5397 if (is_guest_mode(vcpu)) {
5398 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5399 unsigned long orig_val = val;
5401 /* analogously to handle_set_cr0 */
5402 val = (val & ~vmcs12->cr4_guest_host_mask) |
5403 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
5404 if (kvm_set_cr4(vcpu, val))
5406 vmcs_writel(CR4_READ_SHADOW, orig_val);
5409 return kvm_set_cr4(vcpu, val);
5412 /* called to set cr0 as approriate for clts instruction exit. */
5413 static void handle_clts(struct kvm_vcpu *vcpu)
5415 if (is_guest_mode(vcpu)) {
5417 * We get here when L2 did CLTS, and L1 didn't shadow CR0.TS
5418 * but we did (!fpu_active). We need to keep GUEST_CR0.TS on,
5419 * just pretend it's off (also in arch.cr0 for fpu_activate).
5421 vmcs_writel(CR0_READ_SHADOW,
5422 vmcs_readl(CR0_READ_SHADOW) & ~X86_CR0_TS);
5423 vcpu->arch.cr0 &= ~X86_CR0_TS;
5425 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
5428 static int handle_cr(struct kvm_vcpu *vcpu)
5430 unsigned long exit_qualification, val;
5435 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5436 cr = exit_qualification & 15;
5437 reg = (exit_qualification >> 8) & 15;
5438 switch ((exit_qualification >> 4) & 3) {
5439 case 0: /* mov to cr */
5440 val = kvm_register_readl(vcpu, reg);
5441 trace_kvm_cr_write(cr, val);
5444 err = handle_set_cr0(vcpu, val);
5445 kvm_complete_insn_gp(vcpu, err);
5448 err = kvm_set_cr3(vcpu, val);
5449 kvm_complete_insn_gp(vcpu, err);
5452 err = handle_set_cr4(vcpu, val);
5453 kvm_complete_insn_gp(vcpu, err);
5456 u8 cr8_prev = kvm_get_cr8(vcpu);
5458 err = kvm_set_cr8(vcpu, cr8);
5459 kvm_complete_insn_gp(vcpu, err);
5460 if (lapic_in_kernel(vcpu))
5462 if (cr8_prev <= cr8)
5464 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
5471 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
5472 skip_emulated_instruction(vcpu);
5473 vmx_fpu_activate(vcpu);
5475 case 1: /*mov from cr*/
5478 val = kvm_read_cr3(vcpu);
5479 kvm_register_write(vcpu, reg, val);
5480 trace_kvm_cr_read(cr, val);
5481 skip_emulated_instruction(vcpu);
5484 val = kvm_get_cr8(vcpu);
5485 kvm_register_write(vcpu, reg, val);
5486 trace_kvm_cr_read(cr, val);
5487 skip_emulated_instruction(vcpu);
5492 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5493 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
5494 kvm_lmsw(vcpu, val);
5496 skip_emulated_instruction(vcpu);
5501 vcpu->run->exit_reason = 0;
5502 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
5503 (int)(exit_qualification >> 4) & 3, cr);
5507 static int handle_dr(struct kvm_vcpu *vcpu)
5509 unsigned long exit_qualification;
5512 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5513 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
5515 /* First, if DR does not exist, trigger UD */
5516 if (!kvm_require_dr(vcpu, dr))
5519 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
5520 if (!kvm_require_cpl(vcpu, 0))
5522 dr7 = vmcs_readl(GUEST_DR7);
5525 * As the vm-exit takes precedence over the debug trap, we
5526 * need to emulate the latter, either for the host or the
5527 * guest debugging itself.
5529 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5530 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
5531 vcpu->run->debug.arch.dr7 = dr7;
5532 vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
5533 vcpu->run->debug.arch.exception = DB_VECTOR;
5534 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
5537 vcpu->arch.dr6 &= ~15;
5538 vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
5539 kvm_queue_exception(vcpu, DB_VECTOR);
5544 if (vcpu->guest_debug == 0) {
5545 u32 cpu_based_vm_exec_control;
5547 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5548 cpu_based_vm_exec_control &= ~CPU_BASED_MOV_DR_EXITING;
5549 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5552 * No more DR vmexits; force a reload of the debug registers
5553 * and reenter on this instruction. The next vmexit will
5554 * retrieve the full state of the debug registers.
5556 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
5560 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
5561 if (exit_qualification & TYPE_MOV_FROM_DR) {
5564 if (kvm_get_dr(vcpu, dr, &val))
5566 kvm_register_write(vcpu, reg, val);
5568 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
5571 skip_emulated_instruction(vcpu);
5575 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
5577 return vcpu->arch.dr6;
5580 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
5584 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
5586 u32 cpu_based_vm_exec_control;
5588 get_debugreg(vcpu->arch.db[0], 0);
5589 get_debugreg(vcpu->arch.db[1], 1);
5590 get_debugreg(vcpu->arch.db[2], 2);
5591 get_debugreg(vcpu->arch.db[3], 3);
5592 get_debugreg(vcpu->arch.dr6, 6);
5593 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
5595 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
5597 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5598 cpu_based_vm_exec_control |= CPU_BASED_MOV_DR_EXITING;
5599 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5602 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
5604 vmcs_writel(GUEST_DR7, val);
5607 static int handle_cpuid(struct kvm_vcpu *vcpu)
5609 kvm_emulate_cpuid(vcpu);
5613 static int handle_rdmsr(struct kvm_vcpu *vcpu)
5615 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5616 struct msr_data msr_info;
5618 msr_info.index = ecx;
5619 msr_info.host_initiated = false;
5620 if (vmx_get_msr(vcpu, &msr_info)) {
5621 trace_kvm_msr_read_ex(ecx);
5622 kvm_inject_gp(vcpu, 0);
5626 trace_kvm_msr_read(ecx, msr_info.data);
5628 /* FIXME: handling of bits 32:63 of rax, rdx */
5629 vcpu->arch.regs[VCPU_REGS_RAX] = msr_info.data & -1u;
5630 vcpu->arch.regs[VCPU_REGS_RDX] = (msr_info.data >> 32) & -1u;
5631 skip_emulated_instruction(vcpu);
5635 static int handle_wrmsr(struct kvm_vcpu *vcpu)
5637 struct msr_data msr;
5638 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5639 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
5640 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
5644 msr.host_initiated = false;
5645 if (kvm_set_msr(vcpu, &msr) != 0) {
5646 trace_kvm_msr_write_ex(ecx, data);
5647 kvm_inject_gp(vcpu, 0);
5651 trace_kvm_msr_write(ecx, data);
5652 skip_emulated_instruction(vcpu);
5656 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
5658 kvm_make_request(KVM_REQ_EVENT, vcpu);
5662 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
5664 u32 cpu_based_vm_exec_control;
5666 /* clear pending irq */
5667 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5668 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
5669 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5671 kvm_make_request(KVM_REQ_EVENT, vcpu);
5673 ++vcpu->stat.irq_window_exits;
5677 static int handle_halt(struct kvm_vcpu *vcpu)
5679 return kvm_emulate_halt(vcpu);
5682 static int handle_vmcall(struct kvm_vcpu *vcpu)
5684 kvm_emulate_hypercall(vcpu);
5688 static int handle_invd(struct kvm_vcpu *vcpu)
5690 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5693 static int handle_invlpg(struct kvm_vcpu *vcpu)
5695 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5697 kvm_mmu_invlpg(vcpu, exit_qualification);
5698 skip_emulated_instruction(vcpu);
5702 static int handle_rdpmc(struct kvm_vcpu *vcpu)
5706 err = kvm_rdpmc(vcpu);
5707 kvm_complete_insn_gp(vcpu, err);
5712 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5714 kvm_emulate_wbinvd(vcpu);
5718 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5720 u64 new_bv = kvm_read_edx_eax(vcpu);
5721 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5723 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5724 skip_emulated_instruction(vcpu);
5728 static int handle_xsaves(struct kvm_vcpu *vcpu)
5730 skip_emulated_instruction(vcpu);
5731 WARN(1, "this should never happen\n");
5735 static int handle_xrstors(struct kvm_vcpu *vcpu)
5737 skip_emulated_instruction(vcpu);
5738 WARN(1, "this should never happen\n");
5742 static int handle_apic_access(struct kvm_vcpu *vcpu)
5744 if (likely(fasteoi)) {
5745 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5746 int access_type, offset;
5748 access_type = exit_qualification & APIC_ACCESS_TYPE;
5749 offset = exit_qualification & APIC_ACCESS_OFFSET;
5751 * Sane guest uses MOV to write EOI, with written value
5752 * not cared. So make a short-circuit here by avoiding
5753 * heavy instruction emulation.
5755 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5756 (offset == APIC_EOI)) {
5757 kvm_lapic_set_eoi(vcpu);
5758 skip_emulated_instruction(vcpu);
5762 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5765 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5767 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5768 int vector = exit_qualification & 0xff;
5770 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5771 kvm_apic_set_eoi_accelerated(vcpu, vector);
5775 static int handle_apic_write(struct kvm_vcpu *vcpu)
5777 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5778 u32 offset = exit_qualification & 0xfff;
5780 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
5781 kvm_apic_write_nodecode(vcpu, offset);
5785 static int handle_task_switch(struct kvm_vcpu *vcpu)
5787 struct vcpu_vmx *vmx = to_vmx(vcpu);
5788 unsigned long exit_qualification;
5789 bool has_error_code = false;
5792 int reason, type, idt_v, idt_index;
5794 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5795 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5796 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5798 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5800 reason = (u32)exit_qualification >> 30;
5801 if (reason == TASK_SWITCH_GATE && idt_v) {
5803 case INTR_TYPE_NMI_INTR:
5804 vcpu->arch.nmi_injected = false;
5805 vmx_set_nmi_mask(vcpu, true);
5807 case INTR_TYPE_EXT_INTR:
5808 case INTR_TYPE_SOFT_INTR:
5809 kvm_clear_interrupt_queue(vcpu);
5811 case INTR_TYPE_HARD_EXCEPTION:
5812 if (vmx->idt_vectoring_info &
5813 VECTORING_INFO_DELIVER_CODE_MASK) {
5814 has_error_code = true;
5816 vmcs_read32(IDT_VECTORING_ERROR_CODE);
5819 case INTR_TYPE_SOFT_EXCEPTION:
5820 kvm_clear_exception_queue(vcpu);
5826 tss_selector = exit_qualification;
5828 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5829 type != INTR_TYPE_EXT_INTR &&
5830 type != INTR_TYPE_NMI_INTR))
5831 skip_emulated_instruction(vcpu);
5833 if (kvm_task_switch(vcpu, tss_selector,
5834 type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
5835 has_error_code, error_code) == EMULATE_FAIL) {
5836 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5837 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5838 vcpu->run->internal.ndata = 0;
5843 * TODO: What about debug traps on tss switch?
5844 * Are we supposed to inject them and update dr6?
5850 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5852 unsigned long exit_qualification;
5857 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5859 gla_validity = (exit_qualification >> 7) & 0x3;
5860 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
5861 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
5862 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
5863 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
5864 vmcs_readl(GUEST_LINEAR_ADDRESS));
5865 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
5866 (long unsigned int)exit_qualification);
5867 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5868 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
5873 * EPT violation happened while executing iret from NMI,
5874 * "blocked by NMI" bit has to be set before next VM entry.
5875 * There are errata that may cause this bit to not be set:
5878 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5879 cpu_has_virtual_nmis() &&
5880 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
5881 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5883 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5884 trace_kvm_page_fault(gpa, exit_qualification);
5886 /* It is a write fault? */
5887 error_code = exit_qualification & PFERR_WRITE_MASK;
5888 /* It is a fetch fault? */
5889 error_code |= (exit_qualification << 2) & PFERR_FETCH_MASK;
5890 /* ept page table is present? */
5891 error_code |= (exit_qualification >> 3) & PFERR_PRESENT_MASK;
5893 vcpu->arch.exit_qualification = exit_qualification;
5895 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5898 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5903 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5904 if (!kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5905 skip_emulated_instruction(vcpu);
5906 trace_kvm_fast_mmio(gpa);
5910 ret = handle_mmio_page_fault(vcpu, gpa, true);
5911 if (likely(ret == RET_MMIO_PF_EMULATE))
5912 return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) ==
5915 if (unlikely(ret == RET_MMIO_PF_INVALID))
5916 return kvm_mmu_page_fault(vcpu, gpa, 0, NULL, 0);
5918 if (unlikely(ret == RET_MMIO_PF_RETRY))
5921 /* It is the real ept misconfig */
5924 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5925 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
5930 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5932 u32 cpu_based_vm_exec_control;
5934 /* clear pending NMI */
5935 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5936 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
5937 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5938 ++vcpu->stat.nmi_window_exits;
5939 kvm_make_request(KVM_REQ_EVENT, vcpu);
5944 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5946 struct vcpu_vmx *vmx = to_vmx(vcpu);
5947 enum emulation_result err = EMULATE_DONE;
5950 bool intr_window_requested;
5951 unsigned count = 130;
5953 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5954 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
5956 while (vmx->emulation_required && count-- != 0) {
5957 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5958 return handle_interrupt_window(&vmx->vcpu);
5960 if (test_bit(KVM_REQ_EVENT, &vcpu->requests))
5963 err = emulate_instruction(vcpu, EMULTYPE_NO_REEXECUTE);
5965 if (err == EMULATE_USER_EXIT) {
5966 ++vcpu->stat.mmio_exits;
5971 if (err != EMULATE_DONE) {
5972 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5973 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5974 vcpu->run->internal.ndata = 0;
5978 if (vcpu->arch.halt_request) {
5979 vcpu->arch.halt_request = 0;
5980 ret = kvm_vcpu_halt(vcpu);
5984 if (signal_pending(current))
5994 static int __grow_ple_window(int val)
5996 if (ple_window_grow < 1)
5999 val = min(val, ple_window_actual_max);
6001 if (ple_window_grow < ple_window)
6002 val *= ple_window_grow;
6004 val += ple_window_grow;
6009 static int __shrink_ple_window(int val, int modifier, int minimum)
6014 if (modifier < ple_window)
6019 return max(val, minimum);
6022 static void grow_ple_window(struct kvm_vcpu *vcpu)
6024 struct vcpu_vmx *vmx = to_vmx(vcpu);
6025 int old = vmx->ple_window;
6027 vmx->ple_window = __grow_ple_window(old);
6029 if (vmx->ple_window != old)
6030 vmx->ple_window_dirty = true;
6032 trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old);
6035 static void shrink_ple_window(struct kvm_vcpu *vcpu)
6037 struct vcpu_vmx *vmx = to_vmx(vcpu);
6038 int old = vmx->ple_window;
6040 vmx->ple_window = __shrink_ple_window(old,
6041 ple_window_shrink, ple_window);
6043 if (vmx->ple_window != old)
6044 vmx->ple_window_dirty = true;
6046 trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old);
6050 * ple_window_actual_max is computed to be one grow_ple_window() below
6051 * ple_window_max. (See __grow_ple_window for the reason.)
6052 * This prevents overflows, because ple_window_max is int.
6053 * ple_window_max effectively rounded down to a multiple of ple_window_grow in
6055 * ple_window_max is also prevented from setting vmx->ple_window < ple_window.
6057 static void update_ple_window_actual_max(void)
6059 ple_window_actual_max =
6060 __shrink_ple_window(max(ple_window_max, ple_window),
6061 ple_window_grow, INT_MIN);
6065 * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
6067 static void wakeup_handler(void)
6069 struct kvm_vcpu *vcpu;
6070 int cpu = smp_processor_id();
6072 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6073 list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
6074 blocked_vcpu_list) {
6075 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6077 if (pi_test_on(pi_desc) == 1)
6078 kvm_vcpu_kick(vcpu);
6080 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6083 static __init int hardware_setup(void)
6085 int r = -ENOMEM, i, msr;
6087 rdmsrl_safe(MSR_EFER, &host_efer);
6089 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
6090 kvm_define_shared_msr(i, vmx_msr_index[i]);
6092 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
6093 if (!vmx_io_bitmap_a)
6096 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
6097 if (!vmx_io_bitmap_b)
6100 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
6101 if (!vmx_msr_bitmap_legacy)
6104 vmx_msr_bitmap_legacy_x2apic =
6105 (unsigned long *)__get_free_page(GFP_KERNEL);
6106 if (!vmx_msr_bitmap_legacy_x2apic)
6109 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
6110 if (!vmx_msr_bitmap_longmode)
6113 vmx_msr_bitmap_longmode_x2apic =
6114 (unsigned long *)__get_free_page(GFP_KERNEL);
6115 if (!vmx_msr_bitmap_longmode_x2apic)
6119 vmx_msr_bitmap_nested =
6120 (unsigned long *)__get_free_page(GFP_KERNEL);
6121 if (!vmx_msr_bitmap_nested)
6125 vmx_vmread_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
6126 if (!vmx_vmread_bitmap)
6129 vmx_vmwrite_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
6130 if (!vmx_vmwrite_bitmap)
6133 memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
6134 memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
6137 * Allow direct access to the PC debug port (it is often used for I/O
6138 * delays, but the vmexits simply slow things down).
6140 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
6141 clear_bit(0x80, vmx_io_bitmap_a);
6143 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
6145 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
6146 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
6148 memset(vmx_msr_bitmap_nested, 0xff, PAGE_SIZE);
6150 if (setup_vmcs_config(&vmcs_config) < 0) {
6155 if (boot_cpu_has(X86_FEATURE_NX))
6156 kvm_enable_efer_bits(EFER_NX);
6158 if (!cpu_has_vmx_vpid())
6160 if (!cpu_has_vmx_shadow_vmcs())
6161 enable_shadow_vmcs = 0;
6162 if (enable_shadow_vmcs)
6163 init_vmcs_shadow_fields();
6165 if (!cpu_has_vmx_ept() ||
6166 !cpu_has_vmx_ept_4levels()) {
6168 enable_unrestricted_guest = 0;
6169 enable_ept_ad_bits = 0;
6172 if (!cpu_has_vmx_ept_ad_bits())
6173 enable_ept_ad_bits = 0;
6175 if (!cpu_has_vmx_unrestricted_guest())
6176 enable_unrestricted_guest = 0;
6178 if (!cpu_has_vmx_flexpriority())
6179 flexpriority_enabled = 0;
6182 * set_apic_access_page_addr() is used to reload apic access
6183 * page upon invalidation. No need to do anything if not
6184 * using the APIC_ACCESS_ADDR VMCS field.
6186 if (!flexpriority_enabled)
6187 kvm_x86_ops->set_apic_access_page_addr = NULL;
6189 if (!cpu_has_vmx_tpr_shadow())
6190 kvm_x86_ops->update_cr8_intercept = NULL;
6192 if (enable_ept && !cpu_has_vmx_ept_2m_page())
6193 kvm_disable_largepages();
6195 if (!cpu_has_vmx_ple())
6198 if (!cpu_has_vmx_apicv())
6201 if (cpu_has_vmx_tsc_scaling()) {
6202 kvm_has_tsc_control = true;
6203 kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
6204 kvm_tsc_scaling_ratio_frac_bits = 48;
6207 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
6208 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
6209 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
6210 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
6211 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
6212 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
6213 vmx_disable_intercept_for_msr(MSR_IA32_BNDCFGS, true);
6215 memcpy(vmx_msr_bitmap_legacy_x2apic,
6216 vmx_msr_bitmap_legacy, PAGE_SIZE);
6217 memcpy(vmx_msr_bitmap_longmode_x2apic,
6218 vmx_msr_bitmap_longmode, PAGE_SIZE);
6220 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
6223 for (msr = 0x800; msr <= 0x8ff; msr++)
6224 vmx_disable_intercept_msr_read_x2apic(msr);
6226 /* According SDM, in x2apic mode, the whole id reg is used.
6227 * But in KVM, it only use the highest eight bits. Need to
6229 vmx_enable_intercept_msr_read_x2apic(0x802);
6231 vmx_enable_intercept_msr_read_x2apic(0x839);
6233 vmx_disable_intercept_msr_write_x2apic(0x808);
6235 vmx_disable_intercept_msr_write_x2apic(0x80b);
6237 vmx_disable_intercept_msr_write_x2apic(0x83f);
6241 kvm_mmu_set_mask_ptes(0ull,
6242 (enable_ept_ad_bits) ? VMX_EPT_ACCESS_BIT : 0ull,
6243 (enable_ept_ad_bits) ? VMX_EPT_DIRTY_BIT : 0ull,
6244 0ull, VMX_EPT_EXECUTABLE_MASK);
6245 ept_set_mmio_spte_mask();
6250 update_ple_window_actual_max();
6253 * Only enable PML when hardware supports PML feature, and both EPT
6254 * and EPT A/D bit features are enabled -- PML depends on them to work.
6256 if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
6260 kvm_x86_ops->slot_enable_log_dirty = NULL;
6261 kvm_x86_ops->slot_disable_log_dirty = NULL;
6262 kvm_x86_ops->flush_log_dirty = NULL;
6263 kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
6266 kvm_set_posted_intr_wakeup_handler(wakeup_handler);
6268 return alloc_kvm_area();
6271 free_page((unsigned long)vmx_vmwrite_bitmap);
6273 free_page((unsigned long)vmx_vmread_bitmap);
6276 free_page((unsigned long)vmx_msr_bitmap_nested);
6278 free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
6280 free_page((unsigned long)vmx_msr_bitmap_longmode);
6282 free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
6284 free_page((unsigned long)vmx_msr_bitmap_legacy);
6286 free_page((unsigned long)vmx_io_bitmap_b);
6288 free_page((unsigned long)vmx_io_bitmap_a);
6293 static __exit void hardware_unsetup(void)
6295 free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
6296 free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
6297 free_page((unsigned long)vmx_msr_bitmap_legacy);
6298 free_page((unsigned long)vmx_msr_bitmap_longmode);
6299 free_page((unsigned long)vmx_io_bitmap_b);
6300 free_page((unsigned long)vmx_io_bitmap_a);
6301 free_page((unsigned long)vmx_vmwrite_bitmap);
6302 free_page((unsigned long)vmx_vmread_bitmap);
6304 free_page((unsigned long)vmx_msr_bitmap_nested);
6310 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
6311 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
6313 static int handle_pause(struct kvm_vcpu *vcpu)
6316 grow_ple_window(vcpu);
6318 skip_emulated_instruction(vcpu);
6319 kvm_vcpu_on_spin(vcpu);
6324 static int handle_nop(struct kvm_vcpu *vcpu)
6326 skip_emulated_instruction(vcpu);
6330 static int handle_mwait(struct kvm_vcpu *vcpu)
6332 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
6333 return handle_nop(vcpu);
6336 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
6341 static int handle_monitor(struct kvm_vcpu *vcpu)
6343 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
6344 return handle_nop(vcpu);
6348 * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
6349 * We could reuse a single VMCS for all the L2 guests, but we also want the
6350 * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
6351 * allows keeping them loaded on the processor, and in the future will allow
6352 * optimizations where prepare_vmcs02 doesn't need to set all the fields on
6353 * every entry if they never change.
6354 * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
6355 * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
6357 * The following functions allocate and free a vmcs02 in this pool.
6360 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
6361 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
6363 struct vmcs02_list *item;
6364 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6365 if (item->vmptr == vmx->nested.current_vmptr) {
6366 list_move(&item->list, &vmx->nested.vmcs02_pool);
6367 return &item->vmcs02;
6370 if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
6371 /* Recycle the least recently used VMCS. */
6372 item = list_entry(vmx->nested.vmcs02_pool.prev,
6373 struct vmcs02_list, list);
6374 item->vmptr = vmx->nested.current_vmptr;
6375 list_move(&item->list, &vmx->nested.vmcs02_pool);
6376 return &item->vmcs02;
6379 /* Create a new VMCS */
6380 item = kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
6383 item->vmcs02.vmcs = alloc_vmcs();
6384 if (!item->vmcs02.vmcs) {
6388 loaded_vmcs_init(&item->vmcs02);
6389 item->vmptr = vmx->nested.current_vmptr;
6390 list_add(&(item->list), &(vmx->nested.vmcs02_pool));
6391 vmx->nested.vmcs02_num++;
6392 return &item->vmcs02;
6395 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
6396 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
6398 struct vmcs02_list *item;
6399 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6400 if (item->vmptr == vmptr) {
6401 free_loaded_vmcs(&item->vmcs02);
6402 list_del(&item->list);
6404 vmx->nested.vmcs02_num--;
6410 * Free all VMCSs saved for this vcpu, except the one pointed by
6411 * vmx->loaded_vmcs. We must be running L1, so vmx->loaded_vmcs
6412 * must be &vmx->vmcs01.
6414 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
6416 struct vmcs02_list *item, *n;
6418 WARN_ON(vmx->loaded_vmcs != &vmx->vmcs01);
6419 list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
6421 * Something will leak if the above WARN triggers. Better than
6424 if (vmx->loaded_vmcs == &item->vmcs02)
6427 free_loaded_vmcs(&item->vmcs02);
6428 list_del(&item->list);
6430 vmx->nested.vmcs02_num--;
6435 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
6436 * set the success or error code of an emulated VMX instruction, as specified
6437 * by Vol 2B, VMX Instruction Reference, "Conventions".
6439 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
6441 vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
6442 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6443 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
6446 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
6448 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6449 & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
6450 X86_EFLAGS_SF | X86_EFLAGS_OF))
6454 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
6455 u32 vm_instruction_error)
6457 if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
6459 * failValid writes the error number to the current VMCS, which
6460 * can't be done there isn't a current VMCS.
6462 nested_vmx_failInvalid(vcpu);
6465 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6466 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6467 X86_EFLAGS_SF | X86_EFLAGS_OF))
6469 get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
6471 * We don't need to force a shadow sync because
6472 * VM_INSTRUCTION_ERROR is not shadowed
6476 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
6478 /* TODO: not to reset guest simply here. */
6479 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6480 pr_warn("kvm: nested vmx abort, indicator %d\n", indicator);
6483 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
6485 struct vcpu_vmx *vmx =
6486 container_of(timer, struct vcpu_vmx, nested.preemption_timer);
6488 vmx->nested.preemption_timer_expired = true;
6489 kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
6490 kvm_vcpu_kick(&vmx->vcpu);
6492 return HRTIMER_NORESTART;
6496 * Decode the memory-address operand of a vmx instruction, as recorded on an
6497 * exit caused by such an instruction (run by a guest hypervisor).
6498 * On success, returns 0. When the operand is invalid, returns 1 and throws
6501 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
6502 unsigned long exit_qualification,
6503 u32 vmx_instruction_info, bool wr, gva_t *ret)
6507 struct kvm_segment s;
6510 * According to Vol. 3B, "Information for VM Exits Due to Instruction
6511 * Execution", on an exit, vmx_instruction_info holds most of the
6512 * addressing components of the operand. Only the displacement part
6513 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
6514 * For how an actual address is calculated from all these components,
6515 * refer to Vol. 1, "Operand Addressing".
6517 int scaling = vmx_instruction_info & 3;
6518 int addr_size = (vmx_instruction_info >> 7) & 7;
6519 bool is_reg = vmx_instruction_info & (1u << 10);
6520 int seg_reg = (vmx_instruction_info >> 15) & 7;
6521 int index_reg = (vmx_instruction_info >> 18) & 0xf;
6522 bool index_is_valid = !(vmx_instruction_info & (1u << 22));
6523 int base_reg = (vmx_instruction_info >> 23) & 0xf;
6524 bool base_is_valid = !(vmx_instruction_info & (1u << 27));
6527 kvm_queue_exception(vcpu, UD_VECTOR);
6531 /* Addr = segment_base + offset */
6532 /* offset = base + [index * scale] + displacement */
6533 off = exit_qualification; /* holds the displacement */
6535 off += kvm_register_read(vcpu, base_reg);
6537 off += kvm_register_read(vcpu, index_reg)<<scaling;
6538 vmx_get_segment(vcpu, &s, seg_reg);
6539 *ret = s.base + off;
6541 if (addr_size == 1) /* 32 bit */
6544 /* Checks for #GP/#SS exceptions. */
6546 if (is_protmode(vcpu)) {
6547 /* Protected mode: apply checks for segment validity in the
6549 * - segment type check (#GP(0) may be thrown)
6550 * - usability check (#GP(0)/#SS(0))
6551 * - limit check (#GP(0)/#SS(0))
6554 /* #GP(0) if the destination operand is located in a
6555 * read-only data segment or any code segment.
6557 exn = ((s.type & 0xa) == 0 || (s.type & 8));
6559 /* #GP(0) if the source operand is located in an
6560 * execute-only code segment
6562 exn = ((s.type & 0xa) == 8);
6565 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6568 if (is_long_mode(vcpu)) {
6569 /* Long mode: #GP(0)/#SS(0) if the memory address is in a
6570 * non-canonical form. This is an only check for long mode.
6572 exn = is_noncanonical_address(*ret);
6573 } else if (is_protmode(vcpu)) {
6574 /* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
6576 exn = (s.unusable != 0);
6577 /* Protected mode: #GP(0)/#SS(0) if the memory
6578 * operand is outside the segment limit.
6580 exn = exn || (off + sizeof(u64) > s.limit);
6583 kvm_queue_exception_e(vcpu,
6584 seg_reg == VCPU_SREG_SS ?
6585 SS_VECTOR : GP_VECTOR,
6594 * This function performs the various checks including
6595 * - if it's 4KB aligned
6596 * - No bits beyond the physical address width are set
6597 * - Returns 0 on success or else 1
6598 * (Intel SDM Section 30.3)
6600 static int nested_vmx_check_vmptr(struct kvm_vcpu *vcpu, int exit_reason,
6605 struct x86_exception e;
6607 struct vcpu_vmx *vmx = to_vmx(vcpu);
6608 int maxphyaddr = cpuid_maxphyaddr(vcpu);
6610 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
6611 vmcs_read32(VMX_INSTRUCTION_INFO), false, &gva))
6614 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
6615 sizeof(vmptr), &e)) {
6616 kvm_inject_page_fault(vcpu, &e);
6620 switch (exit_reason) {
6621 case EXIT_REASON_VMON:
6624 * The first 4 bytes of VMXON region contain the supported
6625 * VMCS revision identifier
6627 * Note - IA32_VMX_BASIC[48] will never be 1
6628 * for the nested case;
6629 * which replaces physical address width with 32
6632 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6633 nested_vmx_failInvalid(vcpu);
6634 skip_emulated_instruction(vcpu);
6638 page = nested_get_page(vcpu, vmptr);
6640 *(u32 *)kmap(page) != VMCS12_REVISION) {
6641 nested_vmx_failInvalid(vcpu);
6643 skip_emulated_instruction(vcpu);
6647 vmx->nested.vmxon_ptr = vmptr;
6649 case EXIT_REASON_VMCLEAR:
6650 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6651 nested_vmx_failValid(vcpu,
6652 VMXERR_VMCLEAR_INVALID_ADDRESS);
6653 skip_emulated_instruction(vcpu);
6657 if (vmptr == vmx->nested.vmxon_ptr) {
6658 nested_vmx_failValid(vcpu,
6659 VMXERR_VMCLEAR_VMXON_POINTER);
6660 skip_emulated_instruction(vcpu);
6664 case EXIT_REASON_VMPTRLD:
6665 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6666 nested_vmx_failValid(vcpu,
6667 VMXERR_VMPTRLD_INVALID_ADDRESS);
6668 skip_emulated_instruction(vcpu);
6672 if (vmptr == vmx->nested.vmxon_ptr) {
6673 nested_vmx_failValid(vcpu,
6674 VMXERR_VMCLEAR_VMXON_POINTER);
6675 skip_emulated_instruction(vcpu);
6680 return 1; /* shouldn't happen */
6689 * Emulate the VMXON instruction.
6690 * Currently, we just remember that VMX is active, and do not save or even
6691 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
6692 * do not currently need to store anything in that guest-allocated memory
6693 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
6694 * argument is different from the VMXON pointer (which the spec says they do).
6696 static int handle_vmon(struct kvm_vcpu *vcpu)
6698 struct kvm_segment cs;
6699 struct vcpu_vmx *vmx = to_vmx(vcpu);
6700 struct vmcs *shadow_vmcs;
6701 const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
6702 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
6704 /* The Intel VMX Instruction Reference lists a bunch of bits that
6705 * are prerequisite to running VMXON, most notably cr4.VMXE must be
6706 * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
6707 * Otherwise, we should fail with #UD. We test these now:
6709 if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
6710 !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
6711 (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
6712 kvm_queue_exception(vcpu, UD_VECTOR);
6716 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
6717 if (is_long_mode(vcpu) && !cs.l) {
6718 kvm_queue_exception(vcpu, UD_VECTOR);
6722 if (vmx_get_cpl(vcpu)) {
6723 kvm_inject_gp(vcpu, 0);
6727 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMON, NULL))
6730 if (vmx->nested.vmxon) {
6731 nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
6732 skip_emulated_instruction(vcpu);
6736 if ((vmx->nested.msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
6737 != VMXON_NEEDED_FEATURES) {
6738 kvm_inject_gp(vcpu, 0);
6742 if (enable_shadow_vmcs) {
6743 shadow_vmcs = alloc_vmcs();
6746 /* mark vmcs as shadow */
6747 shadow_vmcs->revision_id |= (1u << 31);
6748 /* init shadow vmcs */
6749 vmcs_clear(shadow_vmcs);
6750 vmx->nested.current_shadow_vmcs = shadow_vmcs;
6753 INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
6754 vmx->nested.vmcs02_num = 0;
6756 hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
6758 vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
6760 vmx->nested.vmxon = true;
6762 skip_emulated_instruction(vcpu);
6763 nested_vmx_succeed(vcpu);
6768 * Intel's VMX Instruction Reference specifies a common set of prerequisites
6769 * for running VMX instructions (except VMXON, whose prerequisites are
6770 * slightly different). It also specifies what exception to inject otherwise.
6772 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
6774 struct kvm_segment cs;
6775 struct vcpu_vmx *vmx = to_vmx(vcpu);
6777 if (!vmx->nested.vmxon) {
6778 kvm_queue_exception(vcpu, UD_VECTOR);
6782 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
6783 if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
6784 (is_long_mode(vcpu) && !cs.l)) {
6785 kvm_queue_exception(vcpu, UD_VECTOR);
6789 if (vmx_get_cpl(vcpu)) {
6790 kvm_inject_gp(vcpu, 0);
6797 static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
6799 if (vmx->nested.current_vmptr == -1ull)
6802 /* current_vmptr and current_vmcs12 are always set/reset together */
6803 if (WARN_ON(vmx->nested.current_vmcs12 == NULL))
6806 if (enable_shadow_vmcs) {
6807 /* copy to memory all shadowed fields in case
6808 they were modified */
6809 copy_shadow_to_vmcs12(vmx);
6810 vmx->nested.sync_shadow_vmcs = false;
6811 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
6812 SECONDARY_EXEC_SHADOW_VMCS);
6813 vmcs_write64(VMCS_LINK_POINTER, -1ull);
6815 vmx->nested.posted_intr_nv = -1;
6816 kunmap(vmx->nested.current_vmcs12_page);
6817 nested_release_page(vmx->nested.current_vmcs12_page);
6818 vmx->nested.current_vmptr = -1ull;
6819 vmx->nested.current_vmcs12 = NULL;
6823 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
6824 * just stops using VMX.
6826 static void free_nested(struct vcpu_vmx *vmx)
6828 if (!vmx->nested.vmxon)
6831 vmx->nested.vmxon = false;
6832 free_vpid(vmx->nested.vpid02);
6833 nested_release_vmcs12(vmx);
6834 if (enable_shadow_vmcs)
6835 free_vmcs(vmx->nested.current_shadow_vmcs);
6836 /* Unpin physical memory we referred to in current vmcs02 */
6837 if (vmx->nested.apic_access_page) {
6838 nested_release_page(vmx->nested.apic_access_page);
6839 vmx->nested.apic_access_page = NULL;
6841 if (vmx->nested.virtual_apic_page) {
6842 nested_release_page(vmx->nested.virtual_apic_page);
6843 vmx->nested.virtual_apic_page = NULL;
6845 if (vmx->nested.pi_desc_page) {
6846 kunmap(vmx->nested.pi_desc_page);
6847 nested_release_page(vmx->nested.pi_desc_page);
6848 vmx->nested.pi_desc_page = NULL;
6849 vmx->nested.pi_desc = NULL;
6852 nested_free_all_saved_vmcss(vmx);
6855 /* Emulate the VMXOFF instruction */
6856 static int handle_vmoff(struct kvm_vcpu *vcpu)
6858 if (!nested_vmx_check_permission(vcpu))
6860 free_nested(to_vmx(vcpu));
6861 skip_emulated_instruction(vcpu);
6862 nested_vmx_succeed(vcpu);
6866 /* Emulate the VMCLEAR instruction */
6867 static int handle_vmclear(struct kvm_vcpu *vcpu)
6869 struct vcpu_vmx *vmx = to_vmx(vcpu);
6871 struct vmcs12 *vmcs12;
6874 if (!nested_vmx_check_permission(vcpu))
6877 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMCLEAR, &vmptr))
6880 if (vmptr == vmx->nested.current_vmptr)
6881 nested_release_vmcs12(vmx);
6883 page = nested_get_page(vcpu, vmptr);
6886 * For accurate processor emulation, VMCLEAR beyond available
6887 * physical memory should do nothing at all. However, it is
6888 * possible that a nested vmx bug, not a guest hypervisor bug,
6889 * resulted in this case, so let's shut down before doing any
6892 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6895 vmcs12 = kmap(page);
6896 vmcs12->launch_state = 0;
6898 nested_release_page(page);
6900 nested_free_vmcs02(vmx, vmptr);
6902 skip_emulated_instruction(vcpu);
6903 nested_vmx_succeed(vcpu);
6907 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
6909 /* Emulate the VMLAUNCH instruction */
6910 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
6912 return nested_vmx_run(vcpu, true);
6915 /* Emulate the VMRESUME instruction */
6916 static int handle_vmresume(struct kvm_vcpu *vcpu)
6919 return nested_vmx_run(vcpu, false);
6922 enum vmcs_field_type {
6923 VMCS_FIELD_TYPE_U16 = 0,
6924 VMCS_FIELD_TYPE_U64 = 1,
6925 VMCS_FIELD_TYPE_U32 = 2,
6926 VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
6929 static inline int vmcs_field_type(unsigned long field)
6931 if (0x1 & field) /* the *_HIGH fields are all 32 bit */
6932 return VMCS_FIELD_TYPE_U32;
6933 return (field >> 13) & 0x3 ;
6936 static inline int vmcs_field_readonly(unsigned long field)
6938 return (((field >> 10) & 0x3) == 1);
6942 * Read a vmcs12 field. Since these can have varying lengths and we return
6943 * one type, we chose the biggest type (u64) and zero-extend the return value
6944 * to that size. Note that the caller, handle_vmread, might need to use only
6945 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
6946 * 64-bit fields are to be returned).
6948 static inline int vmcs12_read_any(struct kvm_vcpu *vcpu,
6949 unsigned long field, u64 *ret)
6951 short offset = vmcs_field_to_offset(field);
6957 p = ((char *)(get_vmcs12(vcpu))) + offset;
6959 switch (vmcs_field_type(field)) {
6960 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6961 *ret = *((natural_width *)p);
6963 case VMCS_FIELD_TYPE_U16:
6966 case VMCS_FIELD_TYPE_U32:
6969 case VMCS_FIELD_TYPE_U64:
6979 static inline int vmcs12_write_any(struct kvm_vcpu *vcpu,
6980 unsigned long field, u64 field_value){
6981 short offset = vmcs_field_to_offset(field);
6982 char *p = ((char *) get_vmcs12(vcpu)) + offset;
6986 switch (vmcs_field_type(field)) {
6987 case VMCS_FIELD_TYPE_U16:
6988 *(u16 *)p = field_value;
6990 case VMCS_FIELD_TYPE_U32:
6991 *(u32 *)p = field_value;
6993 case VMCS_FIELD_TYPE_U64:
6994 *(u64 *)p = field_value;
6996 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6997 *(natural_width *)p = field_value;
7006 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
7009 unsigned long field;
7011 struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
7012 const unsigned long *fields = shadow_read_write_fields;
7013 const int num_fields = max_shadow_read_write_fields;
7017 vmcs_load(shadow_vmcs);
7019 for (i = 0; i < num_fields; i++) {
7021 switch (vmcs_field_type(field)) {
7022 case VMCS_FIELD_TYPE_U16:
7023 field_value = vmcs_read16(field);
7025 case VMCS_FIELD_TYPE_U32:
7026 field_value = vmcs_read32(field);
7028 case VMCS_FIELD_TYPE_U64:
7029 field_value = vmcs_read64(field);
7031 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7032 field_value = vmcs_readl(field);
7038 vmcs12_write_any(&vmx->vcpu, field, field_value);
7041 vmcs_clear(shadow_vmcs);
7042 vmcs_load(vmx->loaded_vmcs->vmcs);
7047 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
7049 const unsigned long *fields[] = {
7050 shadow_read_write_fields,
7051 shadow_read_only_fields
7053 const int max_fields[] = {
7054 max_shadow_read_write_fields,
7055 max_shadow_read_only_fields
7058 unsigned long field;
7059 u64 field_value = 0;
7060 struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
7062 vmcs_load(shadow_vmcs);
7064 for (q = 0; q < ARRAY_SIZE(fields); q++) {
7065 for (i = 0; i < max_fields[q]; i++) {
7066 field = fields[q][i];
7067 vmcs12_read_any(&vmx->vcpu, field, &field_value);
7069 switch (vmcs_field_type(field)) {
7070 case VMCS_FIELD_TYPE_U16:
7071 vmcs_write16(field, (u16)field_value);
7073 case VMCS_FIELD_TYPE_U32:
7074 vmcs_write32(field, (u32)field_value);
7076 case VMCS_FIELD_TYPE_U64:
7077 vmcs_write64(field, (u64)field_value);
7079 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7080 vmcs_writel(field, (long)field_value);
7089 vmcs_clear(shadow_vmcs);
7090 vmcs_load(vmx->loaded_vmcs->vmcs);
7094 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
7095 * used before) all generate the same failure when it is missing.
7097 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
7099 struct vcpu_vmx *vmx = to_vmx(vcpu);
7100 if (vmx->nested.current_vmptr == -1ull) {
7101 nested_vmx_failInvalid(vcpu);
7102 skip_emulated_instruction(vcpu);
7108 static int handle_vmread(struct kvm_vcpu *vcpu)
7110 unsigned long field;
7112 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7113 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7116 if (!nested_vmx_check_permission(vcpu) ||
7117 !nested_vmx_check_vmcs12(vcpu))
7120 /* Decode instruction info and find the field to read */
7121 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7122 /* Read the field, zero-extended to a u64 field_value */
7123 if (vmcs12_read_any(vcpu, field, &field_value) < 0) {
7124 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7125 skip_emulated_instruction(vcpu);
7129 * Now copy part of this value to register or memory, as requested.
7130 * Note that the number of bits actually copied is 32 or 64 depending
7131 * on the guest's mode (32 or 64 bit), not on the given field's length.
7133 if (vmx_instruction_info & (1u << 10)) {
7134 kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
7137 if (get_vmx_mem_address(vcpu, exit_qualification,
7138 vmx_instruction_info, true, &gva))
7140 /* _system ok, as nested_vmx_check_permission verified cpl=0 */
7141 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
7142 &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
7145 nested_vmx_succeed(vcpu);
7146 skip_emulated_instruction(vcpu);
7151 static int handle_vmwrite(struct kvm_vcpu *vcpu)
7153 unsigned long field;
7155 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7156 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7157 /* The value to write might be 32 or 64 bits, depending on L1's long
7158 * mode, and eventually we need to write that into a field of several
7159 * possible lengths. The code below first zero-extends the value to 64
7160 * bit (field_value), and then copies only the approriate number of
7161 * bits into the vmcs12 field.
7163 u64 field_value = 0;
7164 struct x86_exception e;
7166 if (!nested_vmx_check_permission(vcpu) ||
7167 !nested_vmx_check_vmcs12(vcpu))
7170 if (vmx_instruction_info & (1u << 10))
7171 field_value = kvm_register_readl(vcpu,
7172 (((vmx_instruction_info) >> 3) & 0xf));
7174 if (get_vmx_mem_address(vcpu, exit_qualification,
7175 vmx_instruction_info, false, &gva))
7177 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
7178 &field_value, (is_64_bit_mode(vcpu) ? 8 : 4), &e)) {
7179 kvm_inject_page_fault(vcpu, &e);
7185 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7186 if (vmcs_field_readonly(field)) {
7187 nested_vmx_failValid(vcpu,
7188 VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
7189 skip_emulated_instruction(vcpu);
7193 if (vmcs12_write_any(vcpu, field, field_value) < 0) {
7194 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7195 skip_emulated_instruction(vcpu);
7199 nested_vmx_succeed(vcpu);
7200 skip_emulated_instruction(vcpu);
7204 /* Emulate the VMPTRLD instruction */
7205 static int handle_vmptrld(struct kvm_vcpu *vcpu)
7207 struct vcpu_vmx *vmx = to_vmx(vcpu);
7210 if (!nested_vmx_check_permission(vcpu))
7213 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMPTRLD, &vmptr))
7216 if (vmx->nested.current_vmptr != vmptr) {
7217 struct vmcs12 *new_vmcs12;
7219 page = nested_get_page(vcpu, vmptr);
7221 nested_vmx_failInvalid(vcpu);
7222 skip_emulated_instruction(vcpu);
7225 new_vmcs12 = kmap(page);
7226 if (new_vmcs12->revision_id != VMCS12_REVISION) {
7228 nested_release_page_clean(page);
7229 nested_vmx_failValid(vcpu,
7230 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
7231 skip_emulated_instruction(vcpu);
7235 nested_release_vmcs12(vmx);
7236 vmx->nested.current_vmptr = vmptr;
7237 vmx->nested.current_vmcs12 = new_vmcs12;
7238 vmx->nested.current_vmcs12_page = page;
7239 if (enable_shadow_vmcs) {
7240 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
7241 SECONDARY_EXEC_SHADOW_VMCS);
7242 vmcs_write64(VMCS_LINK_POINTER,
7243 __pa(vmx->nested.current_shadow_vmcs));
7244 vmx->nested.sync_shadow_vmcs = true;
7248 nested_vmx_succeed(vcpu);
7249 skip_emulated_instruction(vcpu);
7253 /* Emulate the VMPTRST instruction */
7254 static int handle_vmptrst(struct kvm_vcpu *vcpu)
7256 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7257 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7259 struct x86_exception e;
7261 if (!nested_vmx_check_permission(vcpu))
7264 if (get_vmx_mem_address(vcpu, exit_qualification,
7265 vmx_instruction_info, true, &vmcs_gva))
7267 /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
7268 if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
7269 (void *)&to_vmx(vcpu)->nested.current_vmptr,
7271 kvm_inject_page_fault(vcpu, &e);
7274 nested_vmx_succeed(vcpu);
7275 skip_emulated_instruction(vcpu);
7279 /* Emulate the INVEPT instruction */
7280 static int handle_invept(struct kvm_vcpu *vcpu)
7282 struct vcpu_vmx *vmx = to_vmx(vcpu);
7283 u32 vmx_instruction_info, types;
7286 struct x86_exception e;
7291 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7292 SECONDARY_EXEC_ENABLE_EPT) ||
7293 !(vmx->nested.nested_vmx_ept_caps & VMX_EPT_INVEPT_BIT)) {
7294 kvm_queue_exception(vcpu, UD_VECTOR);
7298 if (!nested_vmx_check_permission(vcpu))
7301 if (!kvm_read_cr0_bits(vcpu, X86_CR0_PE)) {
7302 kvm_queue_exception(vcpu, UD_VECTOR);
7306 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7307 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7309 types = (vmx->nested.nested_vmx_ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
7311 if (!(types & (1UL << type))) {
7312 nested_vmx_failValid(vcpu,
7313 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7317 /* According to the Intel VMX instruction reference, the memory
7318 * operand is read even if it isn't needed (e.g., for type==global)
7320 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7321 vmx_instruction_info, false, &gva))
7323 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
7324 sizeof(operand), &e)) {
7325 kvm_inject_page_fault(vcpu, &e);
7330 case VMX_EPT_EXTENT_GLOBAL:
7331 kvm_mmu_sync_roots(vcpu);
7332 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
7333 nested_vmx_succeed(vcpu);
7336 /* Trap single context invalidation invept calls */
7341 skip_emulated_instruction(vcpu);
7345 static int handle_invvpid(struct kvm_vcpu *vcpu)
7347 struct vcpu_vmx *vmx = to_vmx(vcpu);
7348 u32 vmx_instruction_info;
7349 unsigned long type, types;
7351 struct x86_exception e;
7354 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7355 SECONDARY_EXEC_ENABLE_VPID) ||
7356 !(vmx->nested.nested_vmx_vpid_caps & VMX_VPID_INVVPID_BIT)) {
7357 kvm_queue_exception(vcpu, UD_VECTOR);
7361 if (!nested_vmx_check_permission(vcpu))
7364 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7365 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7367 types = (vmx->nested.nested_vmx_vpid_caps >> 8) & 0x7;
7369 if (!(types & (1UL << type))) {
7370 nested_vmx_failValid(vcpu,
7371 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7375 /* according to the intel vmx instruction reference, the memory
7376 * operand is read even if it isn't needed (e.g., for type==global)
7378 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7379 vmx_instruction_info, false, &gva))
7381 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vpid,
7383 kvm_inject_page_fault(vcpu, &e);
7388 case VMX_VPID_EXTENT_ALL_CONTEXT:
7389 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->nested.vpid02);
7390 nested_vmx_succeed(vcpu);
7393 /* Trap single context invalidation invvpid calls */
7398 skip_emulated_instruction(vcpu);
7402 static int handle_pml_full(struct kvm_vcpu *vcpu)
7404 unsigned long exit_qualification;
7406 trace_kvm_pml_full(vcpu->vcpu_id);
7408 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7411 * PML buffer FULL happened while executing iret from NMI,
7412 * "blocked by NMI" bit has to be set before next VM entry.
7414 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
7415 cpu_has_virtual_nmis() &&
7416 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
7417 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
7418 GUEST_INTR_STATE_NMI);
7421 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
7422 * here.., and there's no userspace involvement needed for PML.
7427 static int handle_pcommit(struct kvm_vcpu *vcpu)
7429 /* we never catch pcommit instruct for L1 guest. */
7435 * The exit handlers return 1 if the exit was handled fully and guest execution
7436 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
7437 * to be done to userspace and return 0.
7439 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
7440 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
7441 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
7442 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
7443 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
7444 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
7445 [EXIT_REASON_CR_ACCESS] = handle_cr,
7446 [EXIT_REASON_DR_ACCESS] = handle_dr,
7447 [EXIT_REASON_CPUID] = handle_cpuid,
7448 [EXIT_REASON_MSR_READ] = handle_rdmsr,
7449 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
7450 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
7451 [EXIT_REASON_HLT] = handle_halt,
7452 [EXIT_REASON_INVD] = handle_invd,
7453 [EXIT_REASON_INVLPG] = handle_invlpg,
7454 [EXIT_REASON_RDPMC] = handle_rdpmc,
7455 [EXIT_REASON_VMCALL] = handle_vmcall,
7456 [EXIT_REASON_VMCLEAR] = handle_vmclear,
7457 [EXIT_REASON_VMLAUNCH] = handle_vmlaunch,
7458 [EXIT_REASON_VMPTRLD] = handle_vmptrld,
7459 [EXIT_REASON_VMPTRST] = handle_vmptrst,
7460 [EXIT_REASON_VMREAD] = handle_vmread,
7461 [EXIT_REASON_VMRESUME] = handle_vmresume,
7462 [EXIT_REASON_VMWRITE] = handle_vmwrite,
7463 [EXIT_REASON_VMOFF] = handle_vmoff,
7464 [EXIT_REASON_VMON] = handle_vmon,
7465 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
7466 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
7467 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
7468 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
7469 [EXIT_REASON_WBINVD] = handle_wbinvd,
7470 [EXIT_REASON_XSETBV] = handle_xsetbv,
7471 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
7472 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
7473 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
7474 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
7475 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
7476 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait,
7477 [EXIT_REASON_MONITOR_TRAP_FLAG] = handle_monitor_trap,
7478 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor,
7479 [EXIT_REASON_INVEPT] = handle_invept,
7480 [EXIT_REASON_INVVPID] = handle_invvpid,
7481 [EXIT_REASON_XSAVES] = handle_xsaves,
7482 [EXIT_REASON_XRSTORS] = handle_xrstors,
7483 [EXIT_REASON_PML_FULL] = handle_pml_full,
7484 [EXIT_REASON_PCOMMIT] = handle_pcommit,
7487 static const int kvm_vmx_max_exit_handlers =
7488 ARRAY_SIZE(kvm_vmx_exit_handlers);
7490 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
7491 struct vmcs12 *vmcs12)
7493 unsigned long exit_qualification;
7494 gpa_t bitmap, last_bitmap;
7499 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7500 return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
7502 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7504 port = exit_qualification >> 16;
7505 size = (exit_qualification & 7) + 1;
7507 last_bitmap = (gpa_t)-1;
7512 bitmap = vmcs12->io_bitmap_a;
7513 else if (port < 0x10000)
7514 bitmap = vmcs12->io_bitmap_b;
7517 bitmap += (port & 0x7fff) / 8;
7519 if (last_bitmap != bitmap)
7520 if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
7522 if (b & (1 << (port & 7)))
7527 last_bitmap = bitmap;
7534 * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
7535 * rather than handle it ourselves in L0. I.e., check whether L1 expressed
7536 * disinterest in the current event (read or write a specific MSR) by using an
7537 * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
7539 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
7540 struct vmcs12 *vmcs12, u32 exit_reason)
7542 u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
7545 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
7549 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
7550 * for the four combinations of read/write and low/high MSR numbers.
7551 * First we need to figure out which of the four to use:
7553 bitmap = vmcs12->msr_bitmap;
7554 if (exit_reason == EXIT_REASON_MSR_WRITE)
7556 if (msr_index >= 0xc0000000) {
7557 msr_index -= 0xc0000000;
7561 /* Then read the msr_index'th bit from this bitmap: */
7562 if (msr_index < 1024*8) {
7564 if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
7566 return 1 & (b >> (msr_index & 7));
7568 return true; /* let L1 handle the wrong parameter */
7572 * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
7573 * rather than handle it ourselves in L0. I.e., check if L1 wanted to
7574 * intercept (via guest_host_mask etc.) the current event.
7576 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
7577 struct vmcs12 *vmcs12)
7579 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7580 int cr = exit_qualification & 15;
7581 int reg = (exit_qualification >> 8) & 15;
7582 unsigned long val = kvm_register_readl(vcpu, reg);
7584 switch ((exit_qualification >> 4) & 3) {
7585 case 0: /* mov to cr */
7588 if (vmcs12->cr0_guest_host_mask &
7589 (val ^ vmcs12->cr0_read_shadow))
7593 if ((vmcs12->cr3_target_count >= 1 &&
7594 vmcs12->cr3_target_value0 == val) ||
7595 (vmcs12->cr3_target_count >= 2 &&
7596 vmcs12->cr3_target_value1 == val) ||
7597 (vmcs12->cr3_target_count >= 3 &&
7598 vmcs12->cr3_target_value2 == val) ||
7599 (vmcs12->cr3_target_count >= 4 &&
7600 vmcs12->cr3_target_value3 == val))
7602 if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
7606 if (vmcs12->cr4_guest_host_mask &
7607 (vmcs12->cr4_read_shadow ^ val))
7611 if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
7617 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
7618 (vmcs12->cr0_read_shadow & X86_CR0_TS))
7621 case 1: /* mov from cr */
7624 if (vmcs12->cpu_based_vm_exec_control &
7625 CPU_BASED_CR3_STORE_EXITING)
7629 if (vmcs12->cpu_based_vm_exec_control &
7630 CPU_BASED_CR8_STORE_EXITING)
7637 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
7638 * cr0. Other attempted changes are ignored, with no exit.
7640 if (vmcs12->cr0_guest_host_mask & 0xe &
7641 (val ^ vmcs12->cr0_read_shadow))
7643 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
7644 !(vmcs12->cr0_read_shadow & 0x1) &&
7653 * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
7654 * should handle it ourselves in L0 (and then continue L2). Only call this
7655 * when in is_guest_mode (L2).
7657 static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu)
7659 u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
7660 struct vcpu_vmx *vmx = to_vmx(vcpu);
7661 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7662 u32 exit_reason = vmx->exit_reason;
7664 trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
7665 vmcs_readl(EXIT_QUALIFICATION),
7666 vmx->idt_vectoring_info,
7668 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
7671 if (vmx->nested.nested_run_pending)
7674 if (unlikely(vmx->fail)) {
7675 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
7676 vmcs_read32(VM_INSTRUCTION_ERROR));
7680 switch (exit_reason) {
7681 case EXIT_REASON_EXCEPTION_NMI:
7682 if (!is_exception(intr_info))
7684 else if (is_page_fault(intr_info))
7686 else if (is_no_device(intr_info) &&
7687 !(vmcs12->guest_cr0 & X86_CR0_TS))
7689 return vmcs12->exception_bitmap &
7690 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
7691 case EXIT_REASON_EXTERNAL_INTERRUPT:
7693 case EXIT_REASON_TRIPLE_FAULT:
7695 case EXIT_REASON_PENDING_INTERRUPT:
7696 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
7697 case EXIT_REASON_NMI_WINDOW:
7698 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
7699 case EXIT_REASON_TASK_SWITCH:
7701 case EXIT_REASON_CPUID:
7702 if (kvm_register_read(vcpu, VCPU_REGS_RAX) == 0xa)
7705 case EXIT_REASON_HLT:
7706 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
7707 case EXIT_REASON_INVD:
7709 case EXIT_REASON_INVLPG:
7710 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
7711 case EXIT_REASON_RDPMC:
7712 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
7713 case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
7714 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
7715 case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
7716 case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
7717 case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
7718 case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
7719 case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
7720 case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
7722 * VMX instructions trap unconditionally. This allows L1 to
7723 * emulate them for its L2 guest, i.e., allows 3-level nesting!
7726 case EXIT_REASON_CR_ACCESS:
7727 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
7728 case EXIT_REASON_DR_ACCESS:
7729 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
7730 case EXIT_REASON_IO_INSTRUCTION:
7731 return nested_vmx_exit_handled_io(vcpu, vmcs12);
7732 case EXIT_REASON_MSR_READ:
7733 case EXIT_REASON_MSR_WRITE:
7734 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
7735 case EXIT_REASON_INVALID_STATE:
7737 case EXIT_REASON_MWAIT_INSTRUCTION:
7738 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
7739 case EXIT_REASON_MONITOR_TRAP_FLAG:
7740 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_TRAP_FLAG);
7741 case EXIT_REASON_MONITOR_INSTRUCTION:
7742 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
7743 case EXIT_REASON_PAUSE_INSTRUCTION:
7744 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
7745 nested_cpu_has2(vmcs12,
7746 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
7747 case EXIT_REASON_MCE_DURING_VMENTRY:
7749 case EXIT_REASON_TPR_BELOW_THRESHOLD:
7750 return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
7751 case EXIT_REASON_APIC_ACCESS:
7752 return nested_cpu_has2(vmcs12,
7753 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
7754 case EXIT_REASON_APIC_WRITE:
7755 case EXIT_REASON_EOI_INDUCED:
7756 /* apic_write and eoi_induced should exit unconditionally. */
7758 case EXIT_REASON_EPT_VIOLATION:
7760 * L0 always deals with the EPT violation. If nested EPT is
7761 * used, and the nested mmu code discovers that the address is
7762 * missing in the guest EPT table (EPT12), the EPT violation
7763 * will be injected with nested_ept_inject_page_fault()
7766 case EXIT_REASON_EPT_MISCONFIG:
7768 * L2 never uses directly L1's EPT, but rather L0's own EPT
7769 * table (shadow on EPT) or a merged EPT table that L0 built
7770 * (EPT on EPT). So any problems with the structure of the
7771 * table is L0's fault.
7774 case EXIT_REASON_WBINVD:
7775 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
7776 case EXIT_REASON_XSETBV:
7778 case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
7780 * This should never happen, since it is not possible to
7781 * set XSS to a non-zero value---neither in L1 nor in L2.
7782 * If if it were, XSS would have to be checked against
7783 * the XSS exit bitmap in vmcs12.
7785 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
7786 case EXIT_REASON_PCOMMIT:
7787 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_PCOMMIT);
7793 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
7795 *info1 = vmcs_readl(EXIT_QUALIFICATION);
7796 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
7799 static int vmx_create_pml_buffer(struct vcpu_vmx *vmx)
7801 struct page *pml_pg;
7803 pml_pg = alloc_page(GFP_KERNEL | __GFP_ZERO);
7807 vmx->pml_pg = pml_pg;
7809 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
7810 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
7815 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
7818 __free_page(vmx->pml_pg);
7823 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
7825 struct vcpu_vmx *vmx = to_vmx(vcpu);
7829 pml_idx = vmcs_read16(GUEST_PML_INDEX);
7831 /* Do nothing if PML buffer is empty */
7832 if (pml_idx == (PML_ENTITY_NUM - 1))
7835 /* PML index always points to next available PML buffer entity */
7836 if (pml_idx >= PML_ENTITY_NUM)
7841 pml_buf = page_address(vmx->pml_pg);
7842 for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
7845 gpa = pml_buf[pml_idx];
7846 WARN_ON(gpa & (PAGE_SIZE - 1));
7847 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
7850 /* reset PML index */
7851 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
7855 * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
7856 * Called before reporting dirty_bitmap to userspace.
7858 static void kvm_flush_pml_buffers(struct kvm *kvm)
7861 struct kvm_vcpu *vcpu;
7863 * We only need to kick vcpu out of guest mode here, as PML buffer
7864 * is flushed at beginning of all VMEXITs, and it's obvious that only
7865 * vcpus running in guest are possible to have unflushed GPAs in PML
7868 kvm_for_each_vcpu(i, vcpu, kvm)
7869 kvm_vcpu_kick(vcpu);
7872 static void vmx_dump_sel(char *name, uint32_t sel)
7874 pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
7875 name, vmcs_read32(sel),
7876 vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
7877 vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
7878 vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
7881 static void vmx_dump_dtsel(char *name, uint32_t limit)
7883 pr_err("%s limit=0x%08x, base=0x%016lx\n",
7884 name, vmcs_read32(limit),
7885 vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
7888 static void dump_vmcs(void)
7890 u32 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
7891 u32 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
7892 u32 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
7893 u32 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
7894 u32 secondary_exec_control = 0;
7895 unsigned long cr4 = vmcs_readl(GUEST_CR4);
7896 u64 efer = vmcs_read64(GUEST_IA32_EFER);
7899 if (cpu_has_secondary_exec_ctrls())
7900 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7902 pr_err("*** Guest State ***\n");
7903 pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
7904 vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
7905 vmcs_readl(CR0_GUEST_HOST_MASK));
7906 pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
7907 cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
7908 pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
7909 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
7910 (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
7912 pr_err("PDPTR0 = 0x%016llx PDPTR1 = 0x%016llx\n",
7913 vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
7914 pr_err("PDPTR2 = 0x%016llx PDPTR3 = 0x%016llx\n",
7915 vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
7917 pr_err("RSP = 0x%016lx RIP = 0x%016lx\n",
7918 vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
7919 pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n",
7920 vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
7921 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
7922 vmcs_readl(GUEST_SYSENTER_ESP),
7923 vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
7924 vmx_dump_sel("CS: ", GUEST_CS_SELECTOR);
7925 vmx_dump_sel("DS: ", GUEST_DS_SELECTOR);
7926 vmx_dump_sel("SS: ", GUEST_SS_SELECTOR);
7927 vmx_dump_sel("ES: ", GUEST_ES_SELECTOR);
7928 vmx_dump_sel("FS: ", GUEST_FS_SELECTOR);
7929 vmx_dump_sel("GS: ", GUEST_GS_SELECTOR);
7930 vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
7931 vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
7932 vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
7933 vmx_dump_sel("TR: ", GUEST_TR_SELECTOR);
7934 if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
7935 (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
7936 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
7937 efer, vmcs_read64(GUEST_IA32_PAT));
7938 pr_err("DebugCtl = 0x%016llx DebugExceptions = 0x%016lx\n",
7939 vmcs_read64(GUEST_IA32_DEBUGCTL),
7940 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
7941 if (vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
7942 pr_err("PerfGlobCtl = 0x%016llx\n",
7943 vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
7944 if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
7945 pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
7946 pr_err("Interruptibility = %08x ActivityState = %08x\n",
7947 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
7948 vmcs_read32(GUEST_ACTIVITY_STATE));
7949 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
7950 pr_err("InterruptStatus = %04x\n",
7951 vmcs_read16(GUEST_INTR_STATUS));
7953 pr_err("*** Host State ***\n");
7954 pr_err("RIP = 0x%016lx RSP = 0x%016lx\n",
7955 vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
7956 pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
7957 vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
7958 vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
7959 vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
7960 vmcs_read16(HOST_TR_SELECTOR));
7961 pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
7962 vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
7963 vmcs_readl(HOST_TR_BASE));
7964 pr_err("GDTBase=%016lx IDTBase=%016lx\n",
7965 vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
7966 pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
7967 vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
7968 vmcs_readl(HOST_CR4));
7969 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
7970 vmcs_readl(HOST_IA32_SYSENTER_ESP),
7971 vmcs_read32(HOST_IA32_SYSENTER_CS),
7972 vmcs_readl(HOST_IA32_SYSENTER_EIP));
7973 if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
7974 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
7975 vmcs_read64(HOST_IA32_EFER),
7976 vmcs_read64(HOST_IA32_PAT));
7977 if (vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
7978 pr_err("PerfGlobCtl = 0x%016llx\n",
7979 vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
7981 pr_err("*** Control State ***\n");
7982 pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
7983 pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
7984 pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
7985 pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
7986 vmcs_read32(EXCEPTION_BITMAP),
7987 vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
7988 vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
7989 pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
7990 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
7991 vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
7992 vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
7993 pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
7994 vmcs_read32(VM_EXIT_INTR_INFO),
7995 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
7996 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
7997 pr_err(" reason=%08x qualification=%016lx\n",
7998 vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
7999 pr_err("IDTVectoring: info=%08x errcode=%08x\n",
8000 vmcs_read32(IDT_VECTORING_INFO_FIELD),
8001 vmcs_read32(IDT_VECTORING_ERROR_CODE));
8002 pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
8003 if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
8004 pr_err("TSC Multiplier = 0x%016llx\n",
8005 vmcs_read64(TSC_MULTIPLIER));
8006 if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW)
8007 pr_err("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
8008 if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
8009 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
8010 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
8011 pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
8012 n = vmcs_read32(CR3_TARGET_COUNT);
8013 for (i = 0; i + 1 < n; i += 4)
8014 pr_err("CR3 target%u=%016lx target%u=%016lx\n",
8015 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
8016 i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
8018 pr_err("CR3 target%u=%016lx\n",
8019 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
8020 if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
8021 pr_err("PLE Gap=%08x Window=%08x\n",
8022 vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
8023 if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
8024 pr_err("Virtual processor ID = 0x%04x\n",
8025 vmcs_read16(VIRTUAL_PROCESSOR_ID));
8029 * The guest has exited. See if we can fix it or if we need userspace
8032 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
8034 struct vcpu_vmx *vmx = to_vmx(vcpu);
8035 u32 exit_reason = vmx->exit_reason;
8036 u32 vectoring_info = vmx->idt_vectoring_info;
8039 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
8040 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
8041 * querying dirty_bitmap, we only need to kick all vcpus out of guest
8042 * mode as if vcpus is in root mode, the PML buffer must has been
8046 vmx_flush_pml_buffer(vcpu);
8048 /* If guest state is invalid, start emulating */
8049 if (vmx->emulation_required)
8050 return handle_invalid_guest_state(vcpu);
8052 if (is_guest_mode(vcpu) && nested_vmx_exit_handled(vcpu)) {
8053 nested_vmx_vmexit(vcpu, exit_reason,
8054 vmcs_read32(VM_EXIT_INTR_INFO),
8055 vmcs_readl(EXIT_QUALIFICATION));
8059 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
8061 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8062 vcpu->run->fail_entry.hardware_entry_failure_reason
8067 if (unlikely(vmx->fail)) {
8068 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8069 vcpu->run->fail_entry.hardware_entry_failure_reason
8070 = vmcs_read32(VM_INSTRUCTION_ERROR);
8076 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
8077 * delivery event since it indicates guest is accessing MMIO.
8078 * The vm-exit can be triggered again after return to guest that
8079 * will cause infinite loop.
8081 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
8082 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
8083 exit_reason != EXIT_REASON_EPT_VIOLATION &&
8084 exit_reason != EXIT_REASON_TASK_SWITCH)) {
8085 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8086 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
8087 vcpu->run->internal.ndata = 2;
8088 vcpu->run->internal.data[0] = vectoring_info;
8089 vcpu->run->internal.data[1] = exit_reason;
8093 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked &&
8094 !(is_guest_mode(vcpu) && nested_cpu_has_virtual_nmis(
8095 get_vmcs12(vcpu))))) {
8096 if (vmx_interrupt_allowed(vcpu)) {
8097 vmx->soft_vnmi_blocked = 0;
8098 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
8099 vcpu->arch.nmi_pending) {
8101 * This CPU don't support us in finding the end of an
8102 * NMI-blocked window if the guest runs with IRQs
8103 * disabled. So we pull the trigger after 1 s of
8104 * futile waiting, but inform the user about this.
8106 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
8107 "state on VCPU %d after 1 s timeout\n",
8108 __func__, vcpu->vcpu_id);
8109 vmx->soft_vnmi_blocked = 0;
8113 if (exit_reason < kvm_vmx_max_exit_handlers
8114 && kvm_vmx_exit_handlers[exit_reason])
8115 return kvm_vmx_exit_handlers[exit_reason](vcpu);
8117 WARN_ONCE(1, "vmx: unexpected exit reason 0x%x\n", exit_reason);
8118 kvm_queue_exception(vcpu, UD_VECTOR);
8123 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
8125 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8127 if (is_guest_mode(vcpu) &&
8128 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
8131 if (irr == -1 || tpr < irr) {
8132 vmcs_write32(TPR_THRESHOLD, 0);
8136 vmcs_write32(TPR_THRESHOLD, irr);
8139 static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
8141 u32 sec_exec_control;
8144 * There is not point to enable virtualize x2apic without enable
8147 if (!cpu_has_vmx_virtualize_x2apic_mode() ||
8148 !kvm_vcpu_apicv_active(vcpu))
8151 if (!cpu_need_tpr_shadow(vcpu))
8154 sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8157 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8158 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8160 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8161 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8163 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
8165 vmx_set_msr_bitmap(vcpu);
8168 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
8170 struct vcpu_vmx *vmx = to_vmx(vcpu);
8173 * Currently we do not handle the nested case where L2 has an
8174 * APIC access page of its own; that page is still pinned.
8175 * Hence, we skip the case where the VCPU is in guest mode _and_
8176 * L1 prepared an APIC access page for L2.
8178 * For the case where L1 and L2 share the same APIC access page
8179 * (flexpriority=Y but SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES clear
8180 * in the vmcs12), this function will only update either the vmcs01
8181 * or the vmcs02. If the former, the vmcs02 will be updated by
8182 * prepare_vmcs02. If the latter, the vmcs01 will be updated in
8183 * the next L2->L1 exit.
8185 if (!is_guest_mode(vcpu) ||
8186 !nested_cpu_has2(vmx->nested.current_vmcs12,
8187 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
8188 vmcs_write64(APIC_ACCESS_ADDR, hpa);
8191 static void vmx_hwapic_isr_update(struct kvm *kvm, int isr)
8199 status = vmcs_read16(GUEST_INTR_STATUS);
8204 vmcs_write16(GUEST_INTR_STATUS, status);
8208 static void vmx_set_rvi(int vector)
8216 status = vmcs_read16(GUEST_INTR_STATUS);
8217 old = (u8)status & 0xff;
8218 if ((u8)vector != old) {
8220 status |= (u8)vector;
8221 vmcs_write16(GUEST_INTR_STATUS, status);
8225 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
8227 if (!is_guest_mode(vcpu)) {
8228 vmx_set_rvi(max_irr);
8236 * In guest mode. If a vmexit is needed, vmx_check_nested_events
8239 if (nested_exit_on_intr(vcpu))
8243 * Else, fall back to pre-APICv interrupt injection since L2
8244 * is run without virtual interrupt delivery.
8246 if (!kvm_event_needs_reinjection(vcpu) &&
8247 vmx_interrupt_allowed(vcpu)) {
8248 kvm_queue_interrupt(vcpu, max_irr, false);
8249 vmx_inject_irq(vcpu);
8253 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
8255 if (!kvm_vcpu_apicv_active(vcpu))
8258 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
8259 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
8260 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
8261 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
8264 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
8268 if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
8269 || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
8272 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8273 exit_intr_info = vmx->exit_intr_info;
8275 /* Handle machine checks before interrupts are enabled */
8276 if (is_machine_check(exit_intr_info))
8277 kvm_machine_check();
8279 /* We need to handle NMIs before interrupts are enabled */
8280 if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
8281 (exit_intr_info & INTR_INFO_VALID_MASK)) {
8282 kvm_before_handle_nmi(&vmx->vcpu);
8284 kvm_after_handle_nmi(&vmx->vcpu);
8288 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
8290 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8293 * If external interrupt exists, IF bit is set in rflags/eflags on the
8294 * interrupt stack frame, and interrupt will be enabled on a return
8295 * from interrupt handler.
8297 if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
8298 == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
8299 unsigned int vector;
8300 unsigned long entry;
8302 struct vcpu_vmx *vmx = to_vmx(vcpu);
8303 #ifdef CONFIG_X86_64
8307 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
8308 desc = (gate_desc *)vmx->host_idt_base + vector;
8309 entry = gate_offset(*desc);
8311 #ifdef CONFIG_X86_64
8312 "mov %%" _ASM_SP ", %[sp]\n\t"
8313 "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
8318 "orl $0x200, (%%" _ASM_SP ")\n\t"
8319 __ASM_SIZE(push) " $%c[cs]\n\t"
8320 "call *%[entry]\n\t"
8322 #ifdef CONFIG_X86_64
8327 [ss]"i"(__KERNEL_DS),
8328 [cs]"i"(__KERNEL_CS)
8334 static bool vmx_has_high_real_mode_segbase(void)
8336 return enable_unrestricted_guest || emulate_invalid_guest_state;
8339 static bool vmx_mpx_supported(void)
8341 return (vmcs_config.vmexit_ctrl & VM_EXIT_CLEAR_BNDCFGS) &&
8342 (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_BNDCFGS);
8345 static bool vmx_xsaves_supported(void)
8347 return vmcs_config.cpu_based_2nd_exec_ctrl &
8348 SECONDARY_EXEC_XSAVES;
8351 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
8356 bool idtv_info_valid;
8358 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8360 if (cpu_has_virtual_nmis()) {
8361 if (vmx->nmi_known_unmasked)
8364 * Can't use vmx->exit_intr_info since we're not sure what
8365 * the exit reason is.
8367 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8368 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
8369 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
8371 * SDM 3: 27.7.1.2 (September 2008)
8372 * Re-set bit "block by NMI" before VM entry if vmexit caused by
8373 * a guest IRET fault.
8374 * SDM 3: 23.2.2 (September 2008)
8375 * Bit 12 is undefined in any of the following cases:
8376 * If the VM exit sets the valid bit in the IDT-vectoring
8377 * information field.
8378 * If the VM exit is due to a double fault.
8380 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
8381 vector != DF_VECTOR && !idtv_info_valid)
8382 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
8383 GUEST_INTR_STATE_NMI);
8385 vmx->nmi_known_unmasked =
8386 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
8387 & GUEST_INTR_STATE_NMI);
8388 } else if (unlikely(vmx->soft_vnmi_blocked))
8389 vmx->vnmi_blocked_time +=
8390 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
8393 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
8394 u32 idt_vectoring_info,
8395 int instr_len_field,
8396 int error_code_field)
8400 bool idtv_info_valid;
8402 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8404 vcpu->arch.nmi_injected = false;
8405 kvm_clear_exception_queue(vcpu);
8406 kvm_clear_interrupt_queue(vcpu);
8408 if (!idtv_info_valid)
8411 kvm_make_request(KVM_REQ_EVENT, vcpu);
8413 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
8414 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
8417 case INTR_TYPE_NMI_INTR:
8418 vcpu->arch.nmi_injected = true;
8420 * SDM 3: 27.7.1.2 (September 2008)
8421 * Clear bit "block by NMI" before VM entry if a NMI
8424 vmx_set_nmi_mask(vcpu, false);
8426 case INTR_TYPE_SOFT_EXCEPTION:
8427 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8429 case INTR_TYPE_HARD_EXCEPTION:
8430 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
8431 u32 err = vmcs_read32(error_code_field);
8432 kvm_requeue_exception_e(vcpu, vector, err);
8434 kvm_requeue_exception(vcpu, vector);
8436 case INTR_TYPE_SOFT_INTR:
8437 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8439 case INTR_TYPE_EXT_INTR:
8440 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
8447 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
8449 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
8450 VM_EXIT_INSTRUCTION_LEN,
8451 IDT_VECTORING_ERROR_CODE);
8454 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
8456 __vmx_complete_interrupts(vcpu,
8457 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8458 VM_ENTRY_INSTRUCTION_LEN,
8459 VM_ENTRY_EXCEPTION_ERROR_CODE);
8461 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
8464 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
8467 struct perf_guest_switch_msr *msrs;
8469 msrs = perf_guest_get_msrs(&nr_msrs);
8474 for (i = 0; i < nr_msrs; i++)
8475 if (msrs[i].host == msrs[i].guest)
8476 clear_atomic_switch_msr(vmx, msrs[i].msr);
8478 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
8482 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
8484 struct vcpu_vmx *vmx = to_vmx(vcpu);
8485 unsigned long debugctlmsr, cr4;
8487 /* Record the guest's net vcpu time for enforced NMI injections. */
8488 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
8489 vmx->entry_time = ktime_get();
8491 /* Don't enter VMX if guest state is invalid, let the exit handler
8492 start emulation until we arrive back to a valid state */
8493 if (vmx->emulation_required)
8496 if (vmx->ple_window_dirty) {
8497 vmx->ple_window_dirty = false;
8498 vmcs_write32(PLE_WINDOW, vmx->ple_window);
8501 if (vmx->nested.sync_shadow_vmcs) {
8502 copy_vmcs12_to_shadow(vmx);
8503 vmx->nested.sync_shadow_vmcs = false;
8506 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
8507 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
8508 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
8509 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
8511 cr4 = cr4_read_shadow();
8512 if (unlikely(cr4 != vmx->host_state.vmcs_host_cr4)) {
8513 vmcs_writel(HOST_CR4, cr4);
8514 vmx->host_state.vmcs_host_cr4 = cr4;
8517 /* When single-stepping over STI and MOV SS, we must clear the
8518 * corresponding interruptibility bits in the guest state. Otherwise
8519 * vmentry fails as it then expects bit 14 (BS) in pending debug
8520 * exceptions being set, but that's not correct for the guest debugging
8522 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8523 vmx_set_interrupt_shadow(vcpu, 0);
8525 atomic_switch_perf_msrs(vmx);
8526 debugctlmsr = get_debugctlmsr();
8528 vmx->__launched = vmx->loaded_vmcs->launched;
8530 /* Store host registers */
8531 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
8532 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
8533 "push %%" _ASM_CX " \n\t"
8534 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
8536 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
8537 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
8539 /* Reload cr2 if changed */
8540 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
8541 "mov %%cr2, %%" _ASM_DX " \n\t"
8542 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
8544 "mov %%" _ASM_AX", %%cr2 \n\t"
8546 /* Check if vmlaunch of vmresume is needed */
8547 "cmpl $0, %c[launched](%0) \n\t"
8548 /* Load guest registers. Don't clobber flags. */
8549 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
8550 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
8551 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
8552 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
8553 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
8554 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
8555 #ifdef CONFIG_X86_64
8556 "mov %c[r8](%0), %%r8 \n\t"
8557 "mov %c[r9](%0), %%r9 \n\t"
8558 "mov %c[r10](%0), %%r10 \n\t"
8559 "mov %c[r11](%0), %%r11 \n\t"
8560 "mov %c[r12](%0), %%r12 \n\t"
8561 "mov %c[r13](%0), %%r13 \n\t"
8562 "mov %c[r14](%0), %%r14 \n\t"
8563 "mov %c[r15](%0), %%r15 \n\t"
8565 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
8567 /* Enter guest mode */
8569 __ex(ASM_VMX_VMLAUNCH) "\n\t"
8571 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
8573 /* Save guest registers, load host registers, keep flags */
8574 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
8576 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
8577 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
8578 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
8579 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
8580 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
8581 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
8582 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
8583 #ifdef CONFIG_X86_64
8584 "mov %%r8, %c[r8](%0) \n\t"
8585 "mov %%r9, %c[r9](%0) \n\t"
8586 "mov %%r10, %c[r10](%0) \n\t"
8587 "mov %%r11, %c[r11](%0) \n\t"
8588 "mov %%r12, %c[r12](%0) \n\t"
8589 "mov %%r13, %c[r13](%0) \n\t"
8590 "mov %%r14, %c[r14](%0) \n\t"
8591 "mov %%r15, %c[r15](%0) \n\t"
8593 "mov %%cr2, %%" _ASM_AX " \n\t"
8594 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
8596 "pop %%" _ASM_BP "; pop %%" _ASM_DX " \n\t"
8597 "setbe %c[fail](%0) \n\t"
8598 ".pushsection .rodata \n\t"
8599 ".global vmx_return \n\t"
8600 "vmx_return: " _ASM_PTR " 2b \n\t"
8602 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
8603 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
8604 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
8605 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
8606 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
8607 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
8608 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
8609 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
8610 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
8611 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
8612 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
8613 #ifdef CONFIG_X86_64
8614 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
8615 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
8616 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
8617 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
8618 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
8619 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
8620 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
8621 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
8623 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
8624 [wordsize]"i"(sizeof(ulong))
8626 #ifdef CONFIG_X86_64
8627 , "rax", "rbx", "rdi", "rsi"
8628 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
8630 , "eax", "ebx", "edi", "esi"
8634 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
8636 update_debugctlmsr(debugctlmsr);
8638 #ifndef CONFIG_X86_64
8640 * The sysexit path does not restore ds/es, so we must set them to
8641 * a reasonable value ourselves.
8643 * We can't defer this to vmx_load_host_state() since that function
8644 * may be executed in interrupt context, which saves and restore segments
8645 * around it, nullifying its effect.
8647 loadsegment(ds, __USER_DS);
8648 loadsegment(es, __USER_DS);
8651 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
8652 | (1 << VCPU_EXREG_RFLAGS)
8653 | (1 << VCPU_EXREG_PDPTR)
8654 | (1 << VCPU_EXREG_SEGMENTS)
8655 | (1 << VCPU_EXREG_CR3));
8656 vcpu->arch.regs_dirty = 0;
8658 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
8660 vmx->loaded_vmcs->launched = 1;
8662 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
8663 trace_kvm_exit(vmx->exit_reason, vcpu, KVM_ISA_VMX);
8666 * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
8667 * we did not inject a still-pending event to L1 now because of
8668 * nested_run_pending, we need to re-enable this bit.
8670 if (vmx->nested.nested_run_pending)
8671 kvm_make_request(KVM_REQ_EVENT, vcpu);
8673 vmx->nested.nested_run_pending = 0;
8675 vmx_complete_atomic_exit(vmx);
8676 vmx_recover_nmi_blocking(vmx);
8677 vmx_complete_interrupts(vmx);
8680 static void vmx_load_vmcs01(struct kvm_vcpu *vcpu)
8682 struct vcpu_vmx *vmx = to_vmx(vcpu);
8685 if (vmx->loaded_vmcs == &vmx->vmcs01)
8689 vmx->loaded_vmcs = &vmx->vmcs01;
8691 vmx_vcpu_load(vcpu, cpu);
8696 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
8698 struct vcpu_vmx *vmx = to_vmx(vcpu);
8701 vmx_destroy_pml_buffer(vmx);
8702 free_vpid(vmx->vpid);
8703 leave_guest_mode(vcpu);
8704 vmx_load_vmcs01(vcpu);
8706 free_loaded_vmcs(vmx->loaded_vmcs);
8707 kfree(vmx->guest_msrs);
8708 kvm_vcpu_uninit(vcpu);
8709 kmem_cache_free(kvm_vcpu_cache, vmx);
8712 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
8715 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
8719 return ERR_PTR(-ENOMEM);
8721 vmx->vpid = allocate_vpid();
8723 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
8727 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
8728 BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
8732 if (!vmx->guest_msrs) {
8736 vmx->loaded_vmcs = &vmx->vmcs01;
8737 vmx->loaded_vmcs->vmcs = alloc_vmcs();
8738 if (!vmx->loaded_vmcs->vmcs)
8741 kvm_cpu_vmxon(__pa(per_cpu(vmxarea, raw_smp_processor_id())));
8742 loaded_vmcs_init(vmx->loaded_vmcs);
8747 vmx_vcpu_load(&vmx->vcpu, cpu);
8748 vmx->vcpu.cpu = cpu;
8749 err = vmx_vcpu_setup(vmx);
8750 vmx_vcpu_put(&vmx->vcpu);
8754 if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
8755 err = alloc_apic_access_page(kvm);
8761 if (!kvm->arch.ept_identity_map_addr)
8762 kvm->arch.ept_identity_map_addr =
8763 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
8764 err = init_rmode_identity_map(kvm);
8770 nested_vmx_setup_ctls_msrs(vmx);
8771 vmx->nested.vpid02 = allocate_vpid();
8774 vmx->nested.posted_intr_nv = -1;
8775 vmx->nested.current_vmptr = -1ull;
8776 vmx->nested.current_vmcs12 = NULL;
8779 * If PML is turned on, failure on enabling PML just results in failure
8780 * of creating the vcpu, therefore we can simplify PML logic (by
8781 * avoiding dealing with cases, such as enabling PML partially on vcpus
8782 * for the guest, etc.
8785 err = vmx_create_pml_buffer(vmx);
8793 free_vpid(vmx->nested.vpid02);
8794 free_loaded_vmcs(vmx->loaded_vmcs);
8796 kfree(vmx->guest_msrs);
8798 kvm_vcpu_uninit(&vmx->vcpu);
8800 free_vpid(vmx->vpid);
8801 kmem_cache_free(kvm_vcpu_cache, vmx);
8802 return ERR_PTR(err);
8805 static void __init vmx_check_processor_compat(void *rtn)
8807 struct vmcs_config vmcs_conf;
8810 if (setup_vmcs_config(&vmcs_conf) < 0)
8812 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
8813 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
8814 smp_processor_id());
8819 static int get_ept_level(void)
8821 return VMX_EPT_DEFAULT_GAW + 1;
8824 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
8829 /* For VT-d and EPT combination
8830 * 1. MMIO: always map as UC
8832 * a. VT-d without snooping control feature: can't guarantee the
8833 * result, try to trust guest.
8834 * b. VT-d with snooping control feature: snooping control feature of
8835 * VT-d engine can guarantee the cache correctness. Just set it
8836 * to WB to keep consistent with host. So the same as item 3.
8837 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
8838 * consistent with host MTRR
8841 cache = MTRR_TYPE_UNCACHABLE;
8845 if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
8846 ipat = VMX_EPT_IPAT_BIT;
8847 cache = MTRR_TYPE_WRBACK;
8851 if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
8852 ipat = VMX_EPT_IPAT_BIT;
8853 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
8854 cache = MTRR_TYPE_WRBACK;
8856 cache = MTRR_TYPE_UNCACHABLE;
8860 cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
8863 return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
8866 static int vmx_get_lpage_level(void)
8868 if (enable_ept && !cpu_has_vmx_ept_1g_page())
8869 return PT_DIRECTORY_LEVEL;
8871 /* For shadow and EPT supported 1GB page */
8872 return PT_PDPE_LEVEL;
8875 static void vmcs_set_secondary_exec_control(u32 new_ctl)
8878 * These bits in the secondary execution controls field
8879 * are dynamic, the others are mostly based on the hypervisor
8880 * architecture and the guest's CPUID. Do not touch the
8884 SECONDARY_EXEC_SHADOW_VMCS |
8885 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
8886 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8888 u32 cur_ctl = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8890 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
8891 (new_ctl & ~mask) | (cur_ctl & mask));
8894 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
8896 struct kvm_cpuid_entry2 *best;
8897 struct vcpu_vmx *vmx = to_vmx(vcpu);
8898 u32 secondary_exec_ctl = vmx_secondary_exec_control(vmx);
8900 if (vmx_rdtscp_supported()) {
8901 bool rdtscp_enabled = guest_cpuid_has_rdtscp(vcpu);
8902 if (!rdtscp_enabled)
8903 secondary_exec_ctl &= ~SECONDARY_EXEC_RDTSCP;
8907 vmx->nested.nested_vmx_secondary_ctls_high |=
8908 SECONDARY_EXEC_RDTSCP;
8910 vmx->nested.nested_vmx_secondary_ctls_high &=
8911 ~SECONDARY_EXEC_RDTSCP;
8915 /* Exposing INVPCID only when PCID is exposed */
8916 best = kvm_find_cpuid_entry(vcpu, 0x7, 0);
8917 if (vmx_invpcid_supported() &&
8918 (!best || !(best->ebx & bit(X86_FEATURE_INVPCID)) ||
8919 !guest_cpuid_has_pcid(vcpu))) {
8920 secondary_exec_ctl &= ~SECONDARY_EXEC_ENABLE_INVPCID;
8923 best->ebx &= ~bit(X86_FEATURE_INVPCID);
8926 vmcs_set_secondary_exec_control(secondary_exec_ctl);
8928 if (static_cpu_has(X86_FEATURE_PCOMMIT) && nested) {
8929 if (guest_cpuid_has_pcommit(vcpu))
8930 vmx->nested.nested_vmx_secondary_ctls_high |=
8931 SECONDARY_EXEC_PCOMMIT;
8933 vmx->nested.nested_vmx_secondary_ctls_high &=
8934 ~SECONDARY_EXEC_PCOMMIT;
8938 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
8940 if (func == 1 && nested)
8941 entry->ecx |= bit(X86_FEATURE_VMX);
8944 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
8945 struct x86_exception *fault)
8947 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8950 if (fault->error_code & PFERR_RSVD_MASK)
8951 exit_reason = EXIT_REASON_EPT_MISCONFIG;
8953 exit_reason = EXIT_REASON_EPT_VIOLATION;
8954 nested_vmx_vmexit(vcpu, exit_reason, 0, vcpu->arch.exit_qualification);
8955 vmcs12->guest_physical_address = fault->address;
8958 /* Callbacks for nested_ept_init_mmu_context: */
8960 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
8962 /* return the page table to be shadowed - in our case, EPT12 */
8963 return get_vmcs12(vcpu)->ept_pointer;
8966 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
8968 WARN_ON(mmu_is_nested(vcpu));
8969 kvm_init_shadow_ept_mmu(vcpu,
8970 to_vmx(vcpu)->nested.nested_vmx_ept_caps &
8971 VMX_EPT_EXECUTE_ONLY_BIT);
8972 vcpu->arch.mmu.set_cr3 = vmx_set_cr3;
8973 vcpu->arch.mmu.get_cr3 = nested_ept_get_cr3;
8974 vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
8976 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
8979 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
8981 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
8984 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
8987 bool inequality, bit;
8989 bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
8991 (error_code & vmcs12->page_fault_error_code_mask) !=
8992 vmcs12->page_fault_error_code_match;
8993 return inequality ^ bit;
8996 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
8997 struct x86_exception *fault)
8999 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9001 WARN_ON(!is_guest_mode(vcpu));
9003 if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code))
9004 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
9005 vmcs_read32(VM_EXIT_INTR_INFO),
9006 vmcs_readl(EXIT_QUALIFICATION));
9008 kvm_inject_page_fault(vcpu, fault);
9011 static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu,
9012 struct vmcs12 *vmcs12)
9014 struct vcpu_vmx *vmx = to_vmx(vcpu);
9015 int maxphyaddr = cpuid_maxphyaddr(vcpu);
9017 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
9018 if (!PAGE_ALIGNED(vmcs12->apic_access_addr) ||
9019 vmcs12->apic_access_addr >> maxphyaddr)
9023 * Translate L1 physical address to host physical
9024 * address for vmcs02. Keep the page pinned, so this
9025 * physical address remains valid. We keep a reference
9026 * to it so we can release it later.
9028 if (vmx->nested.apic_access_page) /* shouldn't happen */
9029 nested_release_page(vmx->nested.apic_access_page);
9030 vmx->nested.apic_access_page =
9031 nested_get_page(vcpu, vmcs12->apic_access_addr);
9034 if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
9035 if (!PAGE_ALIGNED(vmcs12->virtual_apic_page_addr) ||
9036 vmcs12->virtual_apic_page_addr >> maxphyaddr)
9039 if (vmx->nested.virtual_apic_page) /* shouldn't happen */
9040 nested_release_page(vmx->nested.virtual_apic_page);
9041 vmx->nested.virtual_apic_page =
9042 nested_get_page(vcpu, vmcs12->virtual_apic_page_addr);
9045 * Failing the vm entry is _not_ what the processor does
9046 * but it's basically the only possibility we have.
9047 * We could still enter the guest if CR8 load exits are
9048 * enabled, CR8 store exits are enabled, and virtualize APIC
9049 * access is disabled; in this case the processor would never
9050 * use the TPR shadow and we could simply clear the bit from
9051 * the execution control. But such a configuration is useless,
9052 * so let's keep the code simple.
9054 if (!vmx->nested.virtual_apic_page)
9058 if (nested_cpu_has_posted_intr(vmcs12)) {
9059 if (!IS_ALIGNED(vmcs12->posted_intr_desc_addr, 64) ||
9060 vmcs12->posted_intr_desc_addr >> maxphyaddr)
9063 if (vmx->nested.pi_desc_page) { /* shouldn't happen */
9064 kunmap(vmx->nested.pi_desc_page);
9065 nested_release_page(vmx->nested.pi_desc_page);
9067 vmx->nested.pi_desc_page =
9068 nested_get_page(vcpu, vmcs12->posted_intr_desc_addr);
9069 if (!vmx->nested.pi_desc_page)
9072 vmx->nested.pi_desc =
9073 (struct pi_desc *)kmap(vmx->nested.pi_desc_page);
9074 if (!vmx->nested.pi_desc) {
9075 nested_release_page_clean(vmx->nested.pi_desc_page);
9078 vmx->nested.pi_desc =
9079 (struct pi_desc *)((void *)vmx->nested.pi_desc +
9080 (unsigned long)(vmcs12->posted_intr_desc_addr &
9087 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
9089 u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
9090 struct vcpu_vmx *vmx = to_vmx(vcpu);
9092 if (vcpu->arch.virtual_tsc_khz == 0)
9095 /* Make sure short timeouts reliably trigger an immediate vmexit.
9096 * hrtimer_start does not guarantee this. */
9097 if (preemption_timeout <= 1) {
9098 vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
9102 preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
9103 preemption_timeout *= 1000000;
9104 do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
9105 hrtimer_start(&vmx->nested.preemption_timer,
9106 ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
9109 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
9110 struct vmcs12 *vmcs12)
9115 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
9118 if (vmcs12_read_any(vcpu, MSR_BITMAP, &addr)) {
9122 maxphyaddr = cpuid_maxphyaddr(vcpu);
9124 if (!PAGE_ALIGNED(vmcs12->msr_bitmap) ||
9125 ((addr + PAGE_SIZE) >> maxphyaddr))
9132 * Merge L0's and L1's MSR bitmap, return false to indicate that
9133 * we do not use the hardware.
9135 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
9136 struct vmcs12 *vmcs12)
9140 unsigned long *msr_bitmap;
9142 if (!nested_cpu_has_virt_x2apic_mode(vmcs12))
9145 page = nested_get_page(vcpu, vmcs12->msr_bitmap);
9150 msr_bitmap = (unsigned long *)kmap(page);
9152 nested_release_page_clean(page);
9157 if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
9158 if (nested_cpu_has_apic_reg_virt(vmcs12))
9159 for (msr = 0x800; msr <= 0x8ff; msr++)
9160 nested_vmx_disable_intercept_for_msr(
9162 vmx_msr_bitmap_nested,
9164 /* TPR is allowed */
9165 nested_vmx_disable_intercept_for_msr(msr_bitmap,
9166 vmx_msr_bitmap_nested,
9167 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
9168 MSR_TYPE_R | MSR_TYPE_W);
9169 if (nested_cpu_has_vid(vmcs12)) {
9170 /* EOI and self-IPI are allowed */
9171 nested_vmx_disable_intercept_for_msr(
9173 vmx_msr_bitmap_nested,
9174 APIC_BASE_MSR + (APIC_EOI >> 4),
9176 nested_vmx_disable_intercept_for_msr(
9178 vmx_msr_bitmap_nested,
9179 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
9184 * Enable reading intercept of all the x2apic
9185 * MSRs. We should not rely on vmcs12 to do any
9186 * optimizations here, it may have been modified
9189 for (msr = 0x800; msr <= 0x8ff; msr++)
9190 __vmx_enable_intercept_for_msr(
9191 vmx_msr_bitmap_nested,
9195 __vmx_enable_intercept_for_msr(
9196 vmx_msr_bitmap_nested,
9197 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
9199 __vmx_enable_intercept_for_msr(
9200 vmx_msr_bitmap_nested,
9201 APIC_BASE_MSR + (APIC_EOI >> 4),
9203 __vmx_enable_intercept_for_msr(
9204 vmx_msr_bitmap_nested,
9205 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
9209 nested_release_page_clean(page);
9214 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
9215 struct vmcs12 *vmcs12)
9217 if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9218 !nested_cpu_has_apic_reg_virt(vmcs12) &&
9219 !nested_cpu_has_vid(vmcs12) &&
9220 !nested_cpu_has_posted_intr(vmcs12))
9224 * If virtualize x2apic mode is enabled,
9225 * virtualize apic access must be disabled.
9227 if (nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9228 nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
9232 * If virtual interrupt delivery is enabled,
9233 * we must exit on external interrupts.
9235 if (nested_cpu_has_vid(vmcs12) &&
9236 !nested_exit_on_intr(vcpu))
9240 * bits 15:8 should be zero in posted_intr_nv,
9241 * the descriptor address has been already checked
9242 * in nested_get_vmcs12_pages.
9244 if (nested_cpu_has_posted_intr(vmcs12) &&
9245 (!nested_cpu_has_vid(vmcs12) ||
9246 !nested_exit_intr_ack_set(vcpu) ||
9247 vmcs12->posted_intr_nv & 0xff00))
9250 /* tpr shadow is needed by all apicv features. */
9251 if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
9257 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
9258 unsigned long count_field,
9259 unsigned long addr_field)
9264 if (vmcs12_read_any(vcpu, count_field, &count) ||
9265 vmcs12_read_any(vcpu, addr_field, &addr)) {
9271 maxphyaddr = cpuid_maxphyaddr(vcpu);
9272 if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
9273 (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr) {
9274 pr_warn_ratelimited(
9275 "nVMX: invalid MSR switch (0x%lx, %d, %llu, 0x%08llx)",
9276 addr_field, maxphyaddr, count, addr);
9282 static int nested_vmx_check_msr_switch_controls(struct kvm_vcpu *vcpu,
9283 struct vmcs12 *vmcs12)
9285 if (vmcs12->vm_exit_msr_load_count == 0 &&
9286 vmcs12->vm_exit_msr_store_count == 0 &&
9287 vmcs12->vm_entry_msr_load_count == 0)
9288 return 0; /* Fast path */
9289 if (nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_LOAD_COUNT,
9290 VM_EXIT_MSR_LOAD_ADDR) ||
9291 nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_STORE_COUNT,
9292 VM_EXIT_MSR_STORE_ADDR) ||
9293 nested_vmx_check_msr_switch(vcpu, VM_ENTRY_MSR_LOAD_COUNT,
9294 VM_ENTRY_MSR_LOAD_ADDR))
9299 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
9300 struct vmx_msr_entry *e)
9302 /* x2APIC MSR accesses are not allowed */
9303 if (vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8)
9305 if (e->index == MSR_IA32_UCODE_WRITE || /* SDM Table 35-2 */
9306 e->index == MSR_IA32_UCODE_REV)
9308 if (e->reserved != 0)
9313 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
9314 struct vmx_msr_entry *e)
9316 if (e->index == MSR_FS_BASE ||
9317 e->index == MSR_GS_BASE ||
9318 e->index == MSR_IA32_SMM_MONITOR_CTL || /* SMM is not supported */
9319 nested_vmx_msr_check_common(vcpu, e))
9324 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
9325 struct vmx_msr_entry *e)
9327 if (e->index == MSR_IA32_SMBASE || /* SMM is not supported */
9328 nested_vmx_msr_check_common(vcpu, e))
9334 * Load guest's/host's msr at nested entry/exit.
9335 * return 0 for success, entry index for failure.
9337 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9340 struct vmx_msr_entry e;
9341 struct msr_data msr;
9343 msr.host_initiated = false;
9344 for (i = 0; i < count; i++) {
9345 if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
9347 pr_warn_ratelimited(
9348 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9349 __func__, i, gpa + i * sizeof(e));
9352 if (nested_vmx_load_msr_check(vcpu, &e)) {
9353 pr_warn_ratelimited(
9354 "%s check failed (%u, 0x%x, 0x%x)\n",
9355 __func__, i, e.index, e.reserved);
9358 msr.index = e.index;
9360 if (kvm_set_msr(vcpu, &msr)) {
9361 pr_warn_ratelimited(
9362 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9363 __func__, i, e.index, e.value);
9372 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9375 struct vmx_msr_entry e;
9377 for (i = 0; i < count; i++) {
9378 struct msr_data msr_info;
9379 if (kvm_vcpu_read_guest(vcpu,
9380 gpa + i * sizeof(e),
9381 &e, 2 * sizeof(u32))) {
9382 pr_warn_ratelimited(
9383 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9384 __func__, i, gpa + i * sizeof(e));
9387 if (nested_vmx_store_msr_check(vcpu, &e)) {
9388 pr_warn_ratelimited(
9389 "%s check failed (%u, 0x%x, 0x%x)\n",
9390 __func__, i, e.index, e.reserved);
9393 msr_info.host_initiated = false;
9394 msr_info.index = e.index;
9395 if (kvm_get_msr(vcpu, &msr_info)) {
9396 pr_warn_ratelimited(
9397 "%s cannot read MSR (%u, 0x%x)\n",
9398 __func__, i, e.index);
9401 if (kvm_vcpu_write_guest(vcpu,
9402 gpa + i * sizeof(e) +
9403 offsetof(struct vmx_msr_entry, value),
9404 &msr_info.data, sizeof(msr_info.data))) {
9405 pr_warn_ratelimited(
9406 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9407 __func__, i, e.index, msr_info.data);
9415 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
9416 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
9417 * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
9418 * guest in a way that will both be appropriate to L1's requests, and our
9419 * needs. In addition to modifying the active vmcs (which is vmcs02), this
9420 * function also has additional necessary side-effects, like setting various
9421 * vcpu->arch fields.
9423 static void prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
9425 struct vcpu_vmx *vmx = to_vmx(vcpu);
9428 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
9429 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
9430 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
9431 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
9432 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
9433 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
9434 vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
9435 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
9436 vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
9437 vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
9438 vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
9439 vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
9440 vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
9441 vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
9442 vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
9443 vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
9444 vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
9445 vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
9446 vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
9447 vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
9448 vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
9449 vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
9450 vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
9451 vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
9452 vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
9453 vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
9454 vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
9455 vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
9456 vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
9457 vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
9458 vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
9459 vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
9460 vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
9461 vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
9462 vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
9463 vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
9465 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
9466 kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
9467 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
9469 kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
9470 vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
9472 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
9473 vmcs12->vm_entry_intr_info_field);
9474 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
9475 vmcs12->vm_entry_exception_error_code);
9476 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
9477 vmcs12->vm_entry_instruction_len);
9478 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
9479 vmcs12->guest_interruptibility_info);
9480 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
9481 vmx_set_rflags(vcpu, vmcs12->guest_rflags);
9482 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
9483 vmcs12->guest_pending_dbg_exceptions);
9484 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
9485 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
9487 if (nested_cpu_has_xsaves(vmcs12))
9488 vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
9489 vmcs_write64(VMCS_LINK_POINTER, -1ull);
9491 exec_control = vmcs12->pin_based_vm_exec_control;
9492 exec_control |= vmcs_config.pin_based_exec_ctrl;
9493 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
9495 if (nested_cpu_has_posted_intr(vmcs12)) {
9497 * Note that we use L0's vector here and in
9498 * vmx_deliver_nested_posted_interrupt.
9500 vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
9501 vmx->nested.pi_pending = false;
9502 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
9503 vmcs_write64(POSTED_INTR_DESC_ADDR,
9504 page_to_phys(vmx->nested.pi_desc_page) +
9505 (unsigned long)(vmcs12->posted_intr_desc_addr &
9508 exec_control &= ~PIN_BASED_POSTED_INTR;
9510 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
9512 vmx->nested.preemption_timer_expired = false;
9513 if (nested_cpu_has_preemption_timer(vmcs12))
9514 vmx_start_preemption_timer(vcpu);
9517 * Whether page-faults are trapped is determined by a combination of
9518 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
9519 * If enable_ept, L0 doesn't care about page faults and we should
9520 * set all of these to L1's desires. However, if !enable_ept, L0 does
9521 * care about (at least some) page faults, and because it is not easy
9522 * (if at all possible?) to merge L0 and L1's desires, we simply ask
9523 * to exit on each and every L2 page fault. This is done by setting
9524 * MASK=MATCH=0 and (see below) EB.PF=1.
9525 * Note that below we don't need special code to set EB.PF beyond the
9526 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
9527 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
9528 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
9530 * A problem with this approach (when !enable_ept) is that L1 may be
9531 * injected with more page faults than it asked for. This could have
9532 * caused problems, but in practice existing hypervisors don't care.
9533 * To fix this, we will need to emulate the PFEC checking (on the L1
9534 * page tables), using walk_addr(), when injecting PFs to L1.
9536 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
9537 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
9538 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
9539 enable_ept ? vmcs12->page_fault_error_code_match : 0);
9541 if (cpu_has_secondary_exec_ctrls()) {
9542 exec_control = vmx_secondary_exec_control(vmx);
9544 /* Take the following fields only from vmcs12 */
9545 exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
9546 SECONDARY_EXEC_RDTSCP |
9547 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
9548 SECONDARY_EXEC_APIC_REGISTER_VIRT |
9549 SECONDARY_EXEC_PCOMMIT);
9550 if (nested_cpu_has(vmcs12,
9551 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
9552 exec_control |= vmcs12->secondary_vm_exec_control;
9554 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) {
9556 * If translation failed, no matter: This feature asks
9557 * to exit when accessing the given address, and if it
9558 * can never be accessed, this feature won't do
9561 if (!vmx->nested.apic_access_page)
9563 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9565 vmcs_write64(APIC_ACCESS_ADDR,
9566 page_to_phys(vmx->nested.apic_access_page));
9567 } else if (!(nested_cpu_has_virt_x2apic_mode(vmcs12)) &&
9568 cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9570 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9571 kvm_vcpu_reload_apic_access_page(vcpu);
9574 if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
9575 vmcs_write64(EOI_EXIT_BITMAP0,
9576 vmcs12->eoi_exit_bitmap0);
9577 vmcs_write64(EOI_EXIT_BITMAP1,
9578 vmcs12->eoi_exit_bitmap1);
9579 vmcs_write64(EOI_EXIT_BITMAP2,
9580 vmcs12->eoi_exit_bitmap2);
9581 vmcs_write64(EOI_EXIT_BITMAP3,
9582 vmcs12->eoi_exit_bitmap3);
9583 vmcs_write16(GUEST_INTR_STATUS,
9584 vmcs12->guest_intr_status);
9587 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
9592 * Set host-state according to L0's settings (vmcs12 is irrelevant here)
9593 * Some constant fields are set here by vmx_set_constant_host_state().
9594 * Other fields are different per CPU, and will be set later when
9595 * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
9597 vmx_set_constant_host_state(vmx);
9600 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
9601 * entry, but only if the current (host) sp changed from the value
9602 * we wrote last (vmx->host_rsp). This cache is no longer relevant
9603 * if we switch vmcs, and rather than hold a separate cache per vmcs,
9604 * here we just force the write to happen on entry.
9608 exec_control = vmx_exec_control(vmx); /* L0's desires */
9609 exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
9610 exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
9611 exec_control &= ~CPU_BASED_TPR_SHADOW;
9612 exec_control |= vmcs12->cpu_based_vm_exec_control;
9614 if (exec_control & CPU_BASED_TPR_SHADOW) {
9615 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
9616 page_to_phys(vmx->nested.virtual_apic_page));
9617 vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
9620 if (cpu_has_vmx_msr_bitmap() &&
9621 exec_control & CPU_BASED_USE_MSR_BITMAPS) {
9622 nested_vmx_merge_msr_bitmap(vcpu, vmcs12);
9623 /* MSR_BITMAP will be set by following vmx_set_efer. */
9625 exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
9628 * Merging of IO bitmap not currently supported.
9629 * Rather, exit every time.
9631 exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
9632 exec_control |= CPU_BASED_UNCOND_IO_EXITING;
9634 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
9636 /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
9637 * bitwise-or of what L1 wants to trap for L2, and what we want to
9638 * trap. Note that CR0.TS also needs updating - we do this later.
9640 update_exception_bitmap(vcpu);
9641 vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
9642 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
9644 /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
9645 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
9646 * bits are further modified by vmx_set_efer() below.
9648 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
9650 /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
9651 * emulated by vmx_set_efer(), below.
9653 vm_entry_controls_init(vmx,
9654 (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER &
9655 ~VM_ENTRY_IA32E_MODE) |
9656 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
9658 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) {
9659 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
9660 vcpu->arch.pat = vmcs12->guest_ia32_pat;
9661 } else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
9662 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
9665 set_cr4_guest_host_mask(vmx);
9667 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)
9668 vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
9670 if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
9671 vmcs_write64(TSC_OFFSET,
9672 vmx->nested.vmcs01_tsc_offset + vmcs12->tsc_offset);
9674 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
9678 * There is no direct mapping between vpid02 and vpid12, the
9679 * vpid02 is per-vCPU for L0 and reused while the value of
9680 * vpid12 is changed w/ one invvpid during nested vmentry.
9681 * The vpid12 is allocated by L1 for L2, so it will not
9682 * influence global bitmap(for vpid01 and vpid02 allocation)
9683 * even if spawn a lot of nested vCPUs.
9685 if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02) {
9686 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
9687 if (vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
9688 vmx->nested.last_vpid = vmcs12->virtual_processor_id;
9689 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->nested.vpid02);
9692 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
9693 vmx_flush_tlb(vcpu);
9698 if (nested_cpu_has_ept(vmcs12)) {
9699 kvm_mmu_unload(vcpu);
9700 nested_ept_init_mmu_context(vcpu);
9703 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)
9704 vcpu->arch.efer = vmcs12->guest_ia32_efer;
9705 else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
9706 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
9708 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
9709 /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
9710 vmx_set_efer(vcpu, vcpu->arch.efer);
9713 * This sets GUEST_CR0 to vmcs12->guest_cr0, with possibly a modified
9714 * TS bit (for lazy fpu) and bits which we consider mandatory enabled.
9715 * The CR0_READ_SHADOW is what L2 should have expected to read given
9716 * the specifications by L1; It's not enough to take
9717 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
9718 * have more bits than L1 expected.
9720 vmx_set_cr0(vcpu, vmcs12->guest_cr0);
9721 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
9723 vmx_set_cr4(vcpu, vmcs12->guest_cr4);
9724 vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
9726 /* shadow page tables on either EPT or shadow page tables */
9727 kvm_set_cr3(vcpu, vmcs12->guest_cr3);
9728 kvm_mmu_reset_context(vcpu);
9731 vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
9734 * L1 may access the L2's PDPTR, so save them to construct vmcs12
9737 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
9738 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
9739 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
9740 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
9743 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
9744 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
9748 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
9749 * for running an L2 nested guest.
9751 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
9753 struct vmcs12 *vmcs12;
9754 struct vcpu_vmx *vmx = to_vmx(vcpu);
9756 struct loaded_vmcs *vmcs02;
9760 if (!nested_vmx_check_permission(vcpu) ||
9761 !nested_vmx_check_vmcs12(vcpu))
9764 skip_emulated_instruction(vcpu);
9765 vmcs12 = get_vmcs12(vcpu);
9767 if (enable_shadow_vmcs)
9768 copy_shadow_to_vmcs12(vmx);
9771 * The nested entry process starts with enforcing various prerequisites
9772 * on vmcs12 as required by the Intel SDM, and act appropriately when
9773 * they fail: As the SDM explains, some conditions should cause the
9774 * instruction to fail, while others will cause the instruction to seem
9775 * to succeed, but return an EXIT_REASON_INVALID_STATE.
9776 * To speed up the normal (success) code path, we should avoid checking
9777 * for misconfigurations which will anyway be caught by the processor
9778 * when using the merged vmcs02.
9780 if (vmcs12->launch_state == launch) {
9781 nested_vmx_failValid(vcpu,
9782 launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
9783 : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
9787 if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
9788 vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT) {
9789 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9793 if (!nested_get_vmcs12_pages(vcpu, vmcs12)) {
9794 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9798 if (nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12)) {
9799 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9803 if (nested_vmx_check_apicv_controls(vcpu, vmcs12)) {
9804 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9808 if (nested_vmx_check_msr_switch_controls(vcpu, vmcs12)) {
9809 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9813 if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
9814 vmx->nested.nested_vmx_true_procbased_ctls_low,
9815 vmx->nested.nested_vmx_procbased_ctls_high) ||
9816 !vmx_control_verify(vmcs12->secondary_vm_exec_control,
9817 vmx->nested.nested_vmx_secondary_ctls_low,
9818 vmx->nested.nested_vmx_secondary_ctls_high) ||
9819 !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
9820 vmx->nested.nested_vmx_pinbased_ctls_low,
9821 vmx->nested.nested_vmx_pinbased_ctls_high) ||
9822 !vmx_control_verify(vmcs12->vm_exit_controls,
9823 vmx->nested.nested_vmx_true_exit_ctls_low,
9824 vmx->nested.nested_vmx_exit_ctls_high) ||
9825 !vmx_control_verify(vmcs12->vm_entry_controls,
9826 vmx->nested.nested_vmx_true_entry_ctls_low,
9827 vmx->nested.nested_vmx_entry_ctls_high))
9829 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9833 if (((vmcs12->host_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
9834 ((vmcs12->host_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
9835 nested_vmx_failValid(vcpu,
9836 VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
9840 if (!nested_cr0_valid(vcpu, vmcs12->guest_cr0) ||
9841 ((vmcs12->guest_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
9842 nested_vmx_entry_failure(vcpu, vmcs12,
9843 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
9846 if (vmcs12->vmcs_link_pointer != -1ull) {
9847 nested_vmx_entry_failure(vcpu, vmcs12,
9848 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_VMCS_LINK_PTR);
9853 * If the load IA32_EFER VM-entry control is 1, the following checks
9854 * are performed on the field for the IA32_EFER MSR:
9855 * - Bits reserved in the IA32_EFER MSR must be 0.
9856 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
9857 * the IA-32e mode guest VM-exit control. It must also be identical
9858 * to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
9861 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER) {
9862 ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
9863 if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
9864 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
9865 ((vmcs12->guest_cr0 & X86_CR0_PG) &&
9866 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))) {
9867 nested_vmx_entry_failure(vcpu, vmcs12,
9868 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
9874 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
9875 * IA32_EFER MSR must be 0 in the field for that register. In addition,
9876 * the values of the LMA and LME bits in the field must each be that of
9877 * the host address-space size VM-exit control.
9879 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
9880 ia32e = (vmcs12->vm_exit_controls &
9881 VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
9882 if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
9883 ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
9884 ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)) {
9885 nested_vmx_entry_failure(vcpu, vmcs12,
9886 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
9892 * We're finally done with prerequisite checking, and can start with
9896 vmcs02 = nested_get_current_vmcs02(vmx);
9900 enter_guest_mode(vcpu);
9902 vmx->nested.vmcs01_tsc_offset = vmcs_read64(TSC_OFFSET);
9904 if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
9905 vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
9908 vmx->loaded_vmcs = vmcs02;
9910 vmx_vcpu_load(vcpu, cpu);
9914 vmx_segment_cache_clear(vmx);
9916 prepare_vmcs02(vcpu, vmcs12);
9918 msr_entry_idx = nested_vmx_load_msr(vcpu,
9919 vmcs12->vm_entry_msr_load_addr,
9920 vmcs12->vm_entry_msr_load_count);
9921 if (msr_entry_idx) {
9922 leave_guest_mode(vcpu);
9923 vmx_load_vmcs01(vcpu);
9924 nested_vmx_entry_failure(vcpu, vmcs12,
9925 EXIT_REASON_MSR_LOAD_FAIL, msr_entry_idx);
9929 vmcs12->launch_state = 1;
9931 if (vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT)
9932 return kvm_vcpu_halt(vcpu);
9934 vmx->nested.nested_run_pending = 1;
9937 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
9938 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
9939 * returned as far as L1 is concerned. It will only return (and set
9940 * the success flag) when L2 exits (see nested_vmx_vmexit()).
9946 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
9947 * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
9948 * This function returns the new value we should put in vmcs12.guest_cr0.
9949 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
9950 * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
9951 * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
9952 * didn't trap the bit, because if L1 did, so would L0).
9953 * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have
9954 * been modified by L2, and L1 knows it. So just leave the old value of
9955 * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
9956 * isn't relevant, because if L0 traps this bit it can set it to anything.
9957 * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
9958 * changed these bits, and therefore they need to be updated, but L0
9959 * didn't necessarily allow them to be changed in GUEST_CR0 - and rather
9960 * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
9962 static inline unsigned long
9963 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
9966 /*1*/ (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
9967 /*2*/ (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
9968 /*3*/ (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
9969 vcpu->arch.cr0_guest_owned_bits));
9972 static inline unsigned long
9973 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
9976 /*1*/ (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
9977 /*2*/ (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
9978 /*3*/ (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
9979 vcpu->arch.cr4_guest_owned_bits));
9982 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
9983 struct vmcs12 *vmcs12)
9988 if (vcpu->arch.exception.pending && vcpu->arch.exception.reinject) {
9989 nr = vcpu->arch.exception.nr;
9990 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
9992 if (kvm_exception_is_soft(nr)) {
9993 vmcs12->vm_exit_instruction_len =
9994 vcpu->arch.event_exit_inst_len;
9995 idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
9997 idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
9999 if (vcpu->arch.exception.has_error_code) {
10000 idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
10001 vmcs12->idt_vectoring_error_code =
10002 vcpu->arch.exception.error_code;
10005 vmcs12->idt_vectoring_info_field = idt_vectoring;
10006 } else if (vcpu->arch.nmi_injected) {
10007 vmcs12->idt_vectoring_info_field =
10008 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
10009 } else if (vcpu->arch.interrupt.pending) {
10010 nr = vcpu->arch.interrupt.nr;
10011 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
10013 if (vcpu->arch.interrupt.soft) {
10014 idt_vectoring |= INTR_TYPE_SOFT_INTR;
10015 vmcs12->vm_entry_instruction_len =
10016 vcpu->arch.event_exit_inst_len;
10018 idt_vectoring |= INTR_TYPE_EXT_INTR;
10020 vmcs12->idt_vectoring_info_field = idt_vectoring;
10024 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
10026 struct vcpu_vmx *vmx = to_vmx(vcpu);
10028 if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
10029 vmx->nested.preemption_timer_expired) {
10030 if (vmx->nested.nested_run_pending)
10032 nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
10036 if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
10037 if (vmx->nested.nested_run_pending ||
10038 vcpu->arch.interrupt.pending)
10040 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
10041 NMI_VECTOR | INTR_TYPE_NMI_INTR |
10042 INTR_INFO_VALID_MASK, 0);
10044 * The NMI-triggered VM exit counts as injection:
10045 * clear this one and block further NMIs.
10047 vcpu->arch.nmi_pending = 0;
10048 vmx_set_nmi_mask(vcpu, true);
10052 if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
10053 nested_exit_on_intr(vcpu)) {
10054 if (vmx->nested.nested_run_pending)
10056 nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
10060 return vmx_complete_nested_posted_interrupt(vcpu);
10063 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
10065 ktime_t remaining =
10066 hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
10069 if (ktime_to_ns(remaining) <= 0)
10072 value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
10073 do_div(value, 1000000);
10074 return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
10078 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
10079 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
10080 * and this function updates it to reflect the changes to the guest state while
10081 * L2 was running (and perhaps made some exits which were handled directly by L0
10082 * without going back to L1), and to reflect the exit reason.
10083 * Note that we do not have to copy here all VMCS fields, just those that
10084 * could have changed by the L2 guest or the exit - i.e., the guest-state and
10085 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
10086 * which already writes to vmcs12 directly.
10088 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10089 u32 exit_reason, u32 exit_intr_info,
10090 unsigned long exit_qualification)
10092 /* update guest state fields: */
10093 vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
10094 vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
10096 vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
10097 vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
10098 vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
10100 vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
10101 vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
10102 vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
10103 vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
10104 vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
10105 vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
10106 vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
10107 vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
10108 vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
10109 vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
10110 vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
10111 vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
10112 vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
10113 vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
10114 vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
10115 vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
10116 vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
10117 vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
10118 vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
10119 vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
10120 vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
10121 vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
10122 vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
10123 vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
10124 vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
10125 vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
10126 vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
10127 vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
10128 vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
10129 vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
10130 vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
10131 vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
10132 vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
10133 vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
10134 vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
10135 vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
10137 vmcs12->guest_interruptibility_info =
10138 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
10139 vmcs12->guest_pending_dbg_exceptions =
10140 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
10141 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
10142 vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
10144 vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
10146 if (nested_cpu_has_preemption_timer(vmcs12)) {
10147 if (vmcs12->vm_exit_controls &
10148 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
10149 vmcs12->vmx_preemption_timer_value =
10150 vmx_get_preemption_timer_value(vcpu);
10151 hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
10155 * In some cases (usually, nested EPT), L2 is allowed to change its
10156 * own CR3 without exiting. If it has changed it, we must keep it.
10157 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
10158 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
10160 * Additionally, restore L2's PDPTR to vmcs12.
10163 vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
10164 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
10165 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
10166 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
10167 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
10170 if (nested_cpu_has_vid(vmcs12))
10171 vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
10173 vmcs12->vm_entry_controls =
10174 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
10175 (vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
10177 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) {
10178 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
10179 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
10182 /* TODO: These cannot have changed unless we have MSR bitmaps and
10183 * the relevant bit asks not to trap the change */
10184 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
10185 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
10186 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
10187 vmcs12->guest_ia32_efer = vcpu->arch.efer;
10188 vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
10189 vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
10190 vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
10191 if (vmx_mpx_supported())
10192 vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
10193 if (nested_cpu_has_xsaves(vmcs12))
10194 vmcs12->xss_exit_bitmap = vmcs_read64(XSS_EXIT_BITMAP);
10196 /* update exit information fields: */
10198 vmcs12->vm_exit_reason = exit_reason;
10199 vmcs12->exit_qualification = exit_qualification;
10201 vmcs12->vm_exit_intr_info = exit_intr_info;
10202 if ((vmcs12->vm_exit_intr_info &
10203 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
10204 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK))
10205 vmcs12->vm_exit_intr_error_code =
10206 vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
10207 vmcs12->idt_vectoring_info_field = 0;
10208 vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
10209 vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
10211 if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
10212 /* vm_entry_intr_info_field is cleared on exit. Emulate this
10213 * instead of reading the real value. */
10214 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
10217 * Transfer the event that L0 or L1 may wanted to inject into
10218 * L2 to IDT_VECTORING_INFO_FIELD.
10220 vmcs12_save_pending_event(vcpu, vmcs12);
10224 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
10225 * preserved above and would only end up incorrectly in L1.
10227 vcpu->arch.nmi_injected = false;
10228 kvm_clear_exception_queue(vcpu);
10229 kvm_clear_interrupt_queue(vcpu);
10233 * A part of what we need to when the nested L2 guest exits and we want to
10234 * run its L1 parent, is to reset L1's guest state to the host state specified
10236 * This function is to be called not only on normal nested exit, but also on
10237 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
10238 * Failures During or After Loading Guest State").
10239 * This function should be called when the active VMCS is L1's (vmcs01).
10241 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
10242 struct vmcs12 *vmcs12)
10244 struct kvm_segment seg;
10246 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
10247 vcpu->arch.efer = vmcs12->host_ia32_efer;
10248 else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
10249 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10251 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10252 vmx_set_efer(vcpu, vcpu->arch.efer);
10254 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
10255 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
10256 vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
10258 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
10259 * actually changed, because it depends on the current state of
10260 * fpu_active (which may have changed).
10261 * Note that vmx_set_cr0 refers to efer set above.
10263 vmx_set_cr0(vcpu, vmcs12->host_cr0);
10265 * If we did fpu_activate()/fpu_deactivate() during L2's run, we need
10266 * to apply the same changes to L1's vmcs. We just set cr0 correctly,
10267 * but we also need to update cr0_guest_host_mask and exception_bitmap.
10269 update_exception_bitmap(vcpu);
10270 vcpu->arch.cr0_guest_owned_bits = (vcpu->fpu_active ? X86_CR0_TS : 0);
10271 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
10274 * Note that CR4_GUEST_HOST_MASK is already set in the original vmcs01
10275 * (KVM doesn't change it)- no reason to call set_cr4_guest_host_mask();
10277 vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
10278 kvm_set_cr4(vcpu, vmcs12->host_cr4);
10280 nested_ept_uninit_mmu_context(vcpu);
10282 kvm_set_cr3(vcpu, vmcs12->host_cr3);
10283 kvm_mmu_reset_context(vcpu);
10286 vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
10290 * Trivially support vpid by letting L2s share their parent
10291 * L1's vpid. TODO: move to a more elaborate solution, giving
10292 * each L2 its own vpid and exposing the vpid feature to L1.
10294 vmx_flush_tlb(vcpu);
10298 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
10299 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
10300 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
10301 vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
10302 vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
10304 /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1. */
10305 if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
10306 vmcs_write64(GUEST_BNDCFGS, 0);
10308 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
10309 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
10310 vcpu->arch.pat = vmcs12->host_ia32_pat;
10312 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
10313 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
10314 vmcs12->host_ia32_perf_global_ctrl);
10316 /* Set L1 segment info according to Intel SDM
10317 27.5.2 Loading Host Segment and Descriptor-Table Registers */
10318 seg = (struct kvm_segment) {
10320 .limit = 0xFFFFFFFF,
10321 .selector = vmcs12->host_cs_selector,
10327 if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
10331 vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
10332 seg = (struct kvm_segment) {
10334 .limit = 0xFFFFFFFF,
10341 seg.selector = vmcs12->host_ds_selector;
10342 vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
10343 seg.selector = vmcs12->host_es_selector;
10344 vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
10345 seg.selector = vmcs12->host_ss_selector;
10346 vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
10347 seg.selector = vmcs12->host_fs_selector;
10348 seg.base = vmcs12->host_fs_base;
10349 vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
10350 seg.selector = vmcs12->host_gs_selector;
10351 seg.base = vmcs12->host_gs_base;
10352 vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
10353 seg = (struct kvm_segment) {
10354 .base = vmcs12->host_tr_base,
10356 .selector = vmcs12->host_tr_selector,
10360 vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
10362 kvm_set_dr(vcpu, 7, 0x400);
10363 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
10365 if (cpu_has_vmx_msr_bitmap())
10366 vmx_set_msr_bitmap(vcpu);
10368 if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
10369 vmcs12->vm_exit_msr_load_count))
10370 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
10374 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
10375 * and modify vmcs12 to make it see what it would expect to see there if
10376 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
10378 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
10379 u32 exit_intr_info,
10380 unsigned long exit_qualification)
10382 struct vcpu_vmx *vmx = to_vmx(vcpu);
10383 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
10385 /* trying to cancel vmlaunch/vmresume is a bug */
10386 WARN_ON_ONCE(vmx->nested.nested_run_pending);
10388 leave_guest_mode(vcpu);
10389 prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
10390 exit_qualification);
10392 if (nested_vmx_store_msr(vcpu, vmcs12->vm_exit_msr_store_addr,
10393 vmcs12->vm_exit_msr_store_count))
10394 nested_vmx_abort(vcpu, VMX_ABORT_SAVE_GUEST_MSR_FAIL);
10396 vmx_load_vmcs01(vcpu);
10398 if ((exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
10399 && nested_exit_intr_ack_set(vcpu)) {
10400 int irq = kvm_cpu_get_interrupt(vcpu);
10402 vmcs12->vm_exit_intr_info = irq |
10403 INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
10406 trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
10407 vmcs12->exit_qualification,
10408 vmcs12->idt_vectoring_info_field,
10409 vmcs12->vm_exit_intr_info,
10410 vmcs12->vm_exit_intr_error_code,
10413 vm_entry_controls_init(vmx, vmcs_read32(VM_ENTRY_CONTROLS));
10414 vm_exit_controls_init(vmx, vmcs_read32(VM_EXIT_CONTROLS));
10415 vmx_segment_cache_clear(vmx);
10417 /* if no vmcs02 cache requested, remove the one we used */
10418 if (VMCS02_POOL_SIZE == 0)
10419 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
10421 load_vmcs12_host_state(vcpu, vmcs12);
10423 /* Update TSC_OFFSET if TSC was changed while L2 ran */
10424 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
10426 /* This is needed for same reason as it was needed in prepare_vmcs02 */
10429 /* Unpin physical memory we referred to in vmcs02 */
10430 if (vmx->nested.apic_access_page) {
10431 nested_release_page(vmx->nested.apic_access_page);
10432 vmx->nested.apic_access_page = NULL;
10434 if (vmx->nested.virtual_apic_page) {
10435 nested_release_page(vmx->nested.virtual_apic_page);
10436 vmx->nested.virtual_apic_page = NULL;
10438 if (vmx->nested.pi_desc_page) {
10439 kunmap(vmx->nested.pi_desc_page);
10440 nested_release_page(vmx->nested.pi_desc_page);
10441 vmx->nested.pi_desc_page = NULL;
10442 vmx->nested.pi_desc = NULL;
10446 * We are now running in L2, mmu_notifier will force to reload the
10447 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
10449 kvm_vcpu_reload_apic_access_page(vcpu);
10452 * Exiting from L2 to L1, we're now back to L1 which thinks it just
10453 * finished a VMLAUNCH or VMRESUME instruction, so we need to set the
10454 * success or failure flag accordingly.
10456 if (unlikely(vmx->fail)) {
10458 nested_vmx_failValid(vcpu, vmcs_read32(VM_INSTRUCTION_ERROR));
10460 nested_vmx_succeed(vcpu);
10461 if (enable_shadow_vmcs)
10462 vmx->nested.sync_shadow_vmcs = true;
10464 /* in case we halted in L2 */
10465 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10469 * Forcibly leave nested mode in order to be able to reset the VCPU later on.
10471 static void vmx_leave_nested(struct kvm_vcpu *vcpu)
10473 if (is_guest_mode(vcpu))
10474 nested_vmx_vmexit(vcpu, -1, 0, 0);
10475 free_nested(to_vmx(vcpu));
10479 * L1's failure to enter L2 is a subset of a normal exit, as explained in
10480 * 23.7 "VM-entry failures during or after loading guest state" (this also
10481 * lists the acceptable exit-reason and exit-qualification parameters).
10482 * It should only be called before L2 actually succeeded to run, and when
10483 * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
10485 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
10486 struct vmcs12 *vmcs12,
10487 u32 reason, unsigned long qualification)
10489 load_vmcs12_host_state(vcpu, vmcs12);
10490 vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
10491 vmcs12->exit_qualification = qualification;
10492 nested_vmx_succeed(vcpu);
10493 if (enable_shadow_vmcs)
10494 to_vmx(vcpu)->nested.sync_shadow_vmcs = true;
10497 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
10498 struct x86_instruction_info *info,
10499 enum x86_intercept_stage stage)
10501 return X86EMUL_CONTINUE;
10504 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
10507 shrink_ple_window(vcpu);
10510 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
10511 struct kvm_memory_slot *slot)
10513 kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
10514 kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
10517 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
10518 struct kvm_memory_slot *slot)
10520 kvm_mmu_slot_set_dirty(kvm, slot);
10523 static void vmx_flush_log_dirty(struct kvm *kvm)
10525 kvm_flush_pml_buffers(kvm);
10528 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
10529 struct kvm_memory_slot *memslot,
10530 gfn_t offset, unsigned long mask)
10532 kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
10536 * This routine does the following things for vCPU which is going
10537 * to be blocked if VT-d PI is enabled.
10538 * - Store the vCPU to the wakeup list, so when interrupts happen
10539 * we can find the right vCPU to wake up.
10540 * - Change the Posted-interrupt descriptor as below:
10541 * 'NDST' <-- vcpu->pre_pcpu
10542 * 'NV' <-- POSTED_INTR_WAKEUP_VECTOR
10543 * - If 'ON' is set during this process, which means at least one
10544 * interrupt is posted for this vCPU, we cannot block it, in
10545 * this case, return 1, otherwise, return 0.
10548 static int vmx_pre_block(struct kvm_vcpu *vcpu)
10550 unsigned long flags;
10552 struct pi_desc old, new;
10553 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
10555 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
10556 !irq_remapping_cap(IRQ_POSTING_CAP))
10559 vcpu->pre_pcpu = vcpu->cpu;
10560 spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
10561 vcpu->pre_pcpu), flags);
10562 list_add_tail(&vcpu->blocked_vcpu_list,
10563 &per_cpu(blocked_vcpu_on_cpu,
10565 spin_unlock_irqrestore(&per_cpu(blocked_vcpu_on_cpu_lock,
10566 vcpu->pre_pcpu), flags);
10569 old.control = new.control = pi_desc->control;
10572 * We should not block the vCPU if
10573 * an interrupt is posted for it.
10575 if (pi_test_on(pi_desc) == 1) {
10576 spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
10577 vcpu->pre_pcpu), flags);
10578 list_del(&vcpu->blocked_vcpu_list);
10579 spin_unlock_irqrestore(
10580 &per_cpu(blocked_vcpu_on_cpu_lock,
10581 vcpu->pre_pcpu), flags);
10582 vcpu->pre_pcpu = -1;
10587 WARN((pi_desc->sn == 1),
10588 "Warning: SN field of posted-interrupts "
10589 "is set before blocking\n");
10592 * Since vCPU can be preempted during this process,
10593 * vcpu->cpu could be different with pre_pcpu, we
10594 * need to set pre_pcpu as the destination of wakeup
10595 * notification event, then we can find the right vCPU
10596 * to wakeup in wakeup handler if interrupts happen
10597 * when the vCPU is in blocked state.
10599 dest = cpu_physical_id(vcpu->pre_pcpu);
10601 if (x2apic_enabled())
10604 new.ndst = (dest << 8) & 0xFF00;
10606 /* set 'NV' to 'wakeup vector' */
10607 new.nv = POSTED_INTR_WAKEUP_VECTOR;
10608 } while (cmpxchg(&pi_desc->control, old.control,
10609 new.control) != old.control);
10614 static void vmx_post_block(struct kvm_vcpu *vcpu)
10616 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
10617 struct pi_desc old, new;
10619 unsigned long flags;
10621 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
10622 !irq_remapping_cap(IRQ_POSTING_CAP))
10626 old.control = new.control = pi_desc->control;
10628 dest = cpu_physical_id(vcpu->cpu);
10630 if (x2apic_enabled())
10633 new.ndst = (dest << 8) & 0xFF00;
10635 /* Allow posting non-urgent interrupts */
10638 /* set 'NV' to 'notification vector' */
10639 new.nv = POSTED_INTR_VECTOR;
10640 } while (cmpxchg(&pi_desc->control, old.control,
10641 new.control) != old.control);
10643 if(vcpu->pre_pcpu != -1) {
10645 &per_cpu(blocked_vcpu_on_cpu_lock,
10646 vcpu->pre_pcpu), flags);
10647 list_del(&vcpu->blocked_vcpu_list);
10648 spin_unlock_irqrestore(
10649 &per_cpu(blocked_vcpu_on_cpu_lock,
10650 vcpu->pre_pcpu), flags);
10651 vcpu->pre_pcpu = -1;
10656 * vmx_update_pi_irte - set IRTE for Posted-Interrupts
10659 * @host_irq: host irq of the interrupt
10660 * @guest_irq: gsi of the interrupt
10661 * @set: set or unset PI
10662 * returns 0 on success, < 0 on failure
10664 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
10665 uint32_t guest_irq, bool set)
10667 struct kvm_kernel_irq_routing_entry *e;
10668 struct kvm_irq_routing_table *irq_rt;
10669 struct kvm_lapic_irq irq;
10670 struct kvm_vcpu *vcpu;
10671 struct vcpu_data vcpu_info;
10672 int idx, ret = -EINVAL;
10674 if (!kvm_arch_has_assigned_device(kvm) ||
10675 !irq_remapping_cap(IRQ_POSTING_CAP))
10678 idx = srcu_read_lock(&kvm->irq_srcu);
10679 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
10680 BUG_ON(guest_irq >= irq_rt->nr_rt_entries);
10682 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
10683 if (e->type != KVM_IRQ_ROUTING_MSI)
10686 * VT-d PI cannot support posting multicast/broadcast
10687 * interrupts to a vCPU, we still use interrupt remapping
10688 * for these kind of interrupts.
10690 * For lowest-priority interrupts, we only support
10691 * those with single CPU as the destination, e.g. user
10692 * configures the interrupts via /proc/irq or uses
10693 * irqbalance to make the interrupts single-CPU.
10695 * We will support full lowest-priority interrupt later.
10698 kvm_set_msi_irq(e, &irq);
10699 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu))
10702 vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
10703 vcpu_info.vector = irq.vector;
10705 trace_kvm_pi_irte_update(vcpu->vcpu_id, e->gsi,
10706 vcpu_info.vector, vcpu_info.pi_desc_addr, set);
10709 ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
10711 /* suppress notification event before unposting */
10712 pi_set_sn(vcpu_to_pi_desc(vcpu));
10713 ret = irq_set_vcpu_affinity(host_irq, NULL);
10714 pi_clear_sn(vcpu_to_pi_desc(vcpu));
10718 printk(KERN_INFO "%s: failed to update PI IRTE\n",
10726 srcu_read_unlock(&kvm->irq_srcu, idx);
10730 static struct kvm_x86_ops vmx_x86_ops = {
10731 .cpu_has_kvm_support = cpu_has_kvm_support,
10732 .disabled_by_bios = vmx_disabled_by_bios,
10733 .hardware_setup = hardware_setup,
10734 .hardware_unsetup = hardware_unsetup,
10735 .check_processor_compatibility = vmx_check_processor_compat,
10736 .hardware_enable = hardware_enable,
10737 .hardware_disable = hardware_disable,
10738 .cpu_has_accelerated_tpr = report_flexpriority,
10739 .cpu_has_high_real_mode_segbase = vmx_has_high_real_mode_segbase,
10741 .vcpu_create = vmx_create_vcpu,
10742 .vcpu_free = vmx_free_vcpu,
10743 .vcpu_reset = vmx_vcpu_reset,
10745 .prepare_guest_switch = vmx_save_host_state,
10746 .vcpu_load = vmx_vcpu_load,
10747 .vcpu_put = vmx_vcpu_put,
10749 .update_bp_intercept = update_exception_bitmap,
10750 .get_msr = vmx_get_msr,
10751 .set_msr = vmx_set_msr,
10752 .get_segment_base = vmx_get_segment_base,
10753 .get_segment = vmx_get_segment,
10754 .set_segment = vmx_set_segment,
10755 .get_cpl = vmx_get_cpl,
10756 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
10757 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
10758 .decache_cr3 = vmx_decache_cr3,
10759 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
10760 .set_cr0 = vmx_set_cr0,
10761 .set_cr3 = vmx_set_cr3,
10762 .set_cr4 = vmx_set_cr4,
10763 .set_efer = vmx_set_efer,
10764 .get_idt = vmx_get_idt,
10765 .set_idt = vmx_set_idt,
10766 .get_gdt = vmx_get_gdt,
10767 .set_gdt = vmx_set_gdt,
10768 .get_dr6 = vmx_get_dr6,
10769 .set_dr6 = vmx_set_dr6,
10770 .set_dr7 = vmx_set_dr7,
10771 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
10772 .cache_reg = vmx_cache_reg,
10773 .get_rflags = vmx_get_rflags,
10774 .set_rflags = vmx_set_rflags,
10775 .fpu_activate = vmx_fpu_activate,
10776 .fpu_deactivate = vmx_fpu_deactivate,
10778 .tlb_flush = vmx_flush_tlb,
10780 .run = vmx_vcpu_run,
10781 .handle_exit = vmx_handle_exit,
10782 .skip_emulated_instruction = skip_emulated_instruction,
10783 .set_interrupt_shadow = vmx_set_interrupt_shadow,
10784 .get_interrupt_shadow = vmx_get_interrupt_shadow,
10785 .patch_hypercall = vmx_patch_hypercall,
10786 .set_irq = vmx_inject_irq,
10787 .set_nmi = vmx_inject_nmi,
10788 .queue_exception = vmx_queue_exception,
10789 .cancel_injection = vmx_cancel_injection,
10790 .interrupt_allowed = vmx_interrupt_allowed,
10791 .nmi_allowed = vmx_nmi_allowed,
10792 .get_nmi_mask = vmx_get_nmi_mask,
10793 .set_nmi_mask = vmx_set_nmi_mask,
10794 .enable_nmi_window = enable_nmi_window,
10795 .enable_irq_window = enable_irq_window,
10796 .update_cr8_intercept = update_cr8_intercept,
10797 .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode,
10798 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
10799 .get_enable_apicv = vmx_get_enable_apicv,
10800 .refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
10801 .load_eoi_exitmap = vmx_load_eoi_exitmap,
10802 .hwapic_irr_update = vmx_hwapic_irr_update,
10803 .hwapic_isr_update = vmx_hwapic_isr_update,
10804 .sync_pir_to_irr = vmx_sync_pir_to_irr,
10805 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
10807 .set_tss_addr = vmx_set_tss_addr,
10808 .get_tdp_level = get_ept_level,
10809 .get_mt_mask = vmx_get_mt_mask,
10811 .get_exit_info = vmx_get_exit_info,
10813 .get_lpage_level = vmx_get_lpage_level,
10815 .cpuid_update = vmx_cpuid_update,
10817 .rdtscp_supported = vmx_rdtscp_supported,
10818 .invpcid_supported = vmx_invpcid_supported,
10820 .set_supported_cpuid = vmx_set_supported_cpuid,
10822 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
10824 .read_tsc_offset = vmx_read_tsc_offset,
10825 .write_tsc_offset = vmx_write_tsc_offset,
10826 .adjust_tsc_offset_guest = vmx_adjust_tsc_offset_guest,
10827 .read_l1_tsc = vmx_read_l1_tsc,
10829 .set_tdp_cr3 = vmx_set_cr3,
10831 .check_intercept = vmx_check_intercept,
10832 .handle_external_intr = vmx_handle_external_intr,
10833 .mpx_supported = vmx_mpx_supported,
10834 .xsaves_supported = vmx_xsaves_supported,
10836 .check_nested_events = vmx_check_nested_events,
10838 .sched_in = vmx_sched_in,
10840 .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
10841 .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
10842 .flush_log_dirty = vmx_flush_log_dirty,
10843 .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
10845 .pre_block = vmx_pre_block,
10846 .post_block = vmx_post_block,
10848 .pmu_ops = &intel_pmu_ops,
10850 .update_pi_irte = vmx_update_pi_irte,
10853 static int __init vmx_init(void)
10855 int r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
10856 __alignof__(struct vcpu_vmx), THIS_MODULE);
10860 #ifdef CONFIG_KEXEC_CORE
10861 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
10862 crash_vmclear_local_loaded_vmcss);
10868 static void __exit vmx_exit(void)
10870 #ifdef CONFIG_KEXEC_CORE
10871 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
10878 module_init(vmx_init)
10879 module_exit(vmx_exit)