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.
23 #include <linux/kvm_host.h>
24 #include <linux/module.h>
25 #include <linux/kernel.h>
27 #include <linux/highmem.h>
28 #include <linux/sched.h>
29 #include <linux/moduleparam.h>
30 #include <linux/mod_devicetable.h>
31 #include <linux/ftrace_event.h>
32 #include <linux/slab.h>
33 #include <linux/tboot.h>
34 #include <linux/hrtimer.h>
35 #include "kvm_cache_regs.h"
41 #include <asm/virtext.h>
45 #include <asm/perf_event.h>
46 #include <asm/debugreg.h>
47 #include <asm/kexec.h>
51 #define __ex(x) __kvm_handle_fault_on_reboot(x)
52 #define __ex_clear(x, reg) \
53 ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
55 MODULE_AUTHOR("Qumranet");
56 MODULE_LICENSE("GPL");
58 static const struct x86_cpu_id vmx_cpu_id[] = {
59 X86_FEATURE_MATCH(X86_FEATURE_VMX),
62 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
64 static bool __read_mostly enable_vpid = 1;
65 module_param_named(vpid, enable_vpid, bool, 0444);
67 static bool __read_mostly flexpriority_enabled = 1;
68 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
70 static bool __read_mostly enable_ept = 1;
71 module_param_named(ept, enable_ept, bool, S_IRUGO);
73 static bool __read_mostly enable_unrestricted_guest = 1;
74 module_param_named(unrestricted_guest,
75 enable_unrestricted_guest, bool, S_IRUGO);
77 static bool __read_mostly enable_ept_ad_bits = 1;
78 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
80 static bool __read_mostly emulate_invalid_guest_state = true;
81 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
83 static bool __read_mostly vmm_exclusive = 1;
84 module_param(vmm_exclusive, bool, S_IRUGO);
86 static bool __read_mostly fasteoi = 1;
87 module_param(fasteoi, bool, S_IRUGO);
89 static bool __read_mostly enable_apicv = 1;
90 module_param(enable_apicv, bool, S_IRUGO);
92 static bool __read_mostly enable_shadow_vmcs = 1;
93 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
95 * If nested=1, nested virtualization is supported, i.e., guests may use
96 * VMX and be a hypervisor for its own guests. If nested=0, guests may not
97 * use VMX instructions.
99 static bool __read_mostly nested = 0;
100 module_param(nested, bool, S_IRUGO);
102 #define KVM_GUEST_CR0_MASK (X86_CR0_NW | X86_CR0_CD)
103 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST (X86_CR0_WP | X86_CR0_NE)
104 #define KVM_VM_CR0_ALWAYS_ON \
105 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
106 #define KVM_CR4_GUEST_OWNED_BITS \
107 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
108 | X86_CR4_OSXMMEXCPT | X86_CR4_TSD)
110 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
111 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
113 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
115 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
118 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
119 * ple_gap: upper bound on the amount of time between two successive
120 * executions of PAUSE in a loop. Also indicate if ple enabled.
121 * According to test, this time is usually smaller than 128 cycles.
122 * ple_window: upper bound on the amount of time a guest is allowed to execute
123 * in a PAUSE loop. Tests indicate that most spinlocks are held for
124 * less than 2^12 cycles
125 * Time is measured based on a counter that runs at the same rate as the TSC,
126 * refer SDM volume 3b section 21.6.13 & 22.1.3.
128 #define KVM_VMX_DEFAULT_PLE_GAP 128
129 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
130 #define KVM_VMX_DEFAULT_PLE_WINDOW_GROW 2
131 #define KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK 0
132 #define KVM_VMX_DEFAULT_PLE_WINDOW_MAX \
133 INT_MAX / KVM_VMX_DEFAULT_PLE_WINDOW_GROW
135 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
136 module_param(ple_gap, int, S_IRUGO);
138 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
139 module_param(ple_window, int, S_IRUGO);
141 /* Default doubles per-vcpu window every exit. */
142 static int ple_window_grow = KVM_VMX_DEFAULT_PLE_WINDOW_GROW;
143 module_param(ple_window_grow, int, S_IRUGO);
145 /* Default resets per-vcpu window every exit to ple_window. */
146 static int ple_window_shrink = KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK;
147 module_param(ple_window_shrink, int, S_IRUGO);
149 /* Default is to compute the maximum so we can never overflow. */
150 static int ple_window_actual_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
151 static int ple_window_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
152 module_param(ple_window_max, int, S_IRUGO);
154 extern const ulong vmx_return;
156 #define NR_AUTOLOAD_MSRS 8
157 #define VMCS02_POOL_SIZE 1
166 * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
167 * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
168 * loaded on this CPU (so we can clear them if the CPU goes down).
174 struct list_head loaded_vmcss_on_cpu_link;
177 struct shared_msr_entry {
184 * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
185 * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
186 * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
187 * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
188 * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
189 * More than one of these structures may exist, if L1 runs multiple L2 guests.
190 * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
191 * underlying hardware which will be used to run L2.
192 * This structure is packed to ensure that its layout is identical across
193 * machines (necessary for live migration).
194 * If there are changes in this struct, VMCS12_REVISION must be changed.
196 typedef u64 natural_width;
197 struct __packed vmcs12 {
198 /* According to the Intel spec, a VMCS region must start with the
199 * following two fields. Then follow implementation-specific data.
204 u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
205 u32 padding[7]; /* room for future expansion */
210 u64 vm_exit_msr_store_addr;
211 u64 vm_exit_msr_load_addr;
212 u64 vm_entry_msr_load_addr;
214 u64 virtual_apic_page_addr;
215 u64 apic_access_addr;
217 u64 guest_physical_address;
218 u64 vmcs_link_pointer;
219 u64 guest_ia32_debugctl;
222 u64 guest_ia32_perf_global_ctrl;
230 u64 host_ia32_perf_global_ctrl;
231 u64 padding64[8]; /* room for future expansion */
233 * To allow migration of L1 (complete with its L2 guests) between
234 * machines of different natural widths (32 or 64 bit), we cannot have
235 * unsigned long fields with no explict size. We use u64 (aliased
236 * natural_width) instead. Luckily, x86 is little-endian.
238 natural_width cr0_guest_host_mask;
239 natural_width cr4_guest_host_mask;
240 natural_width cr0_read_shadow;
241 natural_width cr4_read_shadow;
242 natural_width cr3_target_value0;
243 natural_width cr3_target_value1;
244 natural_width cr3_target_value2;
245 natural_width cr3_target_value3;
246 natural_width exit_qualification;
247 natural_width guest_linear_address;
248 natural_width guest_cr0;
249 natural_width guest_cr3;
250 natural_width guest_cr4;
251 natural_width guest_es_base;
252 natural_width guest_cs_base;
253 natural_width guest_ss_base;
254 natural_width guest_ds_base;
255 natural_width guest_fs_base;
256 natural_width guest_gs_base;
257 natural_width guest_ldtr_base;
258 natural_width guest_tr_base;
259 natural_width guest_gdtr_base;
260 natural_width guest_idtr_base;
261 natural_width guest_dr7;
262 natural_width guest_rsp;
263 natural_width guest_rip;
264 natural_width guest_rflags;
265 natural_width guest_pending_dbg_exceptions;
266 natural_width guest_sysenter_esp;
267 natural_width guest_sysenter_eip;
268 natural_width host_cr0;
269 natural_width host_cr3;
270 natural_width host_cr4;
271 natural_width host_fs_base;
272 natural_width host_gs_base;
273 natural_width host_tr_base;
274 natural_width host_gdtr_base;
275 natural_width host_idtr_base;
276 natural_width host_ia32_sysenter_esp;
277 natural_width host_ia32_sysenter_eip;
278 natural_width host_rsp;
279 natural_width host_rip;
280 natural_width paddingl[8]; /* room for future expansion */
281 u32 pin_based_vm_exec_control;
282 u32 cpu_based_vm_exec_control;
283 u32 exception_bitmap;
284 u32 page_fault_error_code_mask;
285 u32 page_fault_error_code_match;
286 u32 cr3_target_count;
287 u32 vm_exit_controls;
288 u32 vm_exit_msr_store_count;
289 u32 vm_exit_msr_load_count;
290 u32 vm_entry_controls;
291 u32 vm_entry_msr_load_count;
292 u32 vm_entry_intr_info_field;
293 u32 vm_entry_exception_error_code;
294 u32 vm_entry_instruction_len;
296 u32 secondary_vm_exec_control;
297 u32 vm_instruction_error;
299 u32 vm_exit_intr_info;
300 u32 vm_exit_intr_error_code;
301 u32 idt_vectoring_info_field;
302 u32 idt_vectoring_error_code;
303 u32 vm_exit_instruction_len;
304 u32 vmx_instruction_info;
311 u32 guest_ldtr_limit;
313 u32 guest_gdtr_limit;
314 u32 guest_idtr_limit;
315 u32 guest_es_ar_bytes;
316 u32 guest_cs_ar_bytes;
317 u32 guest_ss_ar_bytes;
318 u32 guest_ds_ar_bytes;
319 u32 guest_fs_ar_bytes;
320 u32 guest_gs_ar_bytes;
321 u32 guest_ldtr_ar_bytes;
322 u32 guest_tr_ar_bytes;
323 u32 guest_interruptibility_info;
324 u32 guest_activity_state;
325 u32 guest_sysenter_cs;
326 u32 host_ia32_sysenter_cs;
327 u32 vmx_preemption_timer_value;
328 u32 padding32[7]; /* room for future expansion */
329 u16 virtual_processor_id;
330 u16 guest_es_selector;
331 u16 guest_cs_selector;
332 u16 guest_ss_selector;
333 u16 guest_ds_selector;
334 u16 guest_fs_selector;
335 u16 guest_gs_selector;
336 u16 guest_ldtr_selector;
337 u16 guest_tr_selector;
338 u16 host_es_selector;
339 u16 host_cs_selector;
340 u16 host_ss_selector;
341 u16 host_ds_selector;
342 u16 host_fs_selector;
343 u16 host_gs_selector;
344 u16 host_tr_selector;
348 * VMCS12_REVISION is an arbitrary id that should be changed if the content or
349 * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
350 * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
352 #define VMCS12_REVISION 0x11e57ed0
355 * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
356 * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
357 * current implementation, 4K are reserved to avoid future complications.
359 #define VMCS12_SIZE 0x1000
361 /* Used to remember the last vmcs02 used for some recently used vmcs12s */
363 struct list_head list;
365 struct loaded_vmcs vmcs02;
369 * The nested_vmx structure is part of vcpu_vmx, and holds information we need
370 * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
373 /* Has the level1 guest done vmxon? */
377 /* The guest-physical address of the current VMCS L1 keeps for L2 */
379 /* The host-usable pointer to the above */
380 struct page *current_vmcs12_page;
381 struct vmcs12 *current_vmcs12;
382 struct vmcs *current_shadow_vmcs;
384 * Indicates if the shadow vmcs must be updated with the
385 * data hold by vmcs12
387 bool sync_shadow_vmcs;
389 /* vmcs02_list cache of VMCSs recently used to run L2 guests */
390 struct list_head vmcs02_pool;
392 u64 vmcs01_tsc_offset;
393 /* L2 must run next, and mustn't decide to exit to L1. */
394 bool nested_run_pending;
396 * Guest pages referred to in vmcs02 with host-physical pointers, so
397 * we must keep them pinned while L2 runs.
399 struct page *apic_access_page;
400 struct page *virtual_apic_page;
401 u64 msr_ia32_feature_control;
403 struct hrtimer preemption_timer;
404 bool preemption_timer_expired;
406 /* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
410 #define POSTED_INTR_ON 0
411 /* Posted-Interrupt Descriptor */
413 u32 pir[8]; /* Posted interrupt requested */
414 u32 control; /* bit 0 of control is outstanding notification bit */
418 static bool pi_test_and_set_on(struct pi_desc *pi_desc)
420 return test_and_set_bit(POSTED_INTR_ON,
421 (unsigned long *)&pi_desc->control);
424 static bool pi_test_and_clear_on(struct pi_desc *pi_desc)
426 return test_and_clear_bit(POSTED_INTR_ON,
427 (unsigned long *)&pi_desc->control);
430 static int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
432 return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
436 struct kvm_vcpu vcpu;
437 unsigned long host_rsp;
439 bool nmi_known_unmasked;
441 u32 idt_vectoring_info;
443 struct shared_msr_entry *guest_msrs;
446 unsigned long host_idt_base;
448 u64 msr_host_kernel_gs_base;
449 u64 msr_guest_kernel_gs_base;
451 u32 vm_entry_controls_shadow;
452 u32 vm_exit_controls_shadow;
454 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
455 * non-nested (L1) guest, it always points to vmcs01. For a nested
456 * guest (L2), it points to a different VMCS.
458 struct loaded_vmcs vmcs01;
459 struct loaded_vmcs *loaded_vmcs;
460 bool __launched; /* temporary, used in vmx_vcpu_run */
461 struct msr_autoload {
463 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
464 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
468 u16 fs_sel, gs_sel, ldt_sel;
472 int gs_ldt_reload_needed;
473 int fs_reload_needed;
474 u64 msr_host_bndcfgs;
475 unsigned long vmcs_host_cr4; /* May not match real cr4 */
480 struct kvm_segment segs[8];
483 u32 bitmask; /* 4 bits per segment (1 bit per field) */
484 struct kvm_save_segment {
492 bool emulation_required;
494 /* Support for vnmi-less CPUs */
495 int soft_vnmi_blocked;
497 s64 vnmi_blocked_time;
502 /* Posted interrupt descriptor */
503 struct pi_desc pi_desc;
505 /* Support for a guest hypervisor (nested VMX) */
506 struct nested_vmx nested;
508 /* Dynamic PLE window. */
510 bool ple_window_dirty;
513 enum segment_cache_field {
522 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
524 return container_of(vcpu, struct vcpu_vmx, vcpu);
527 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
528 #define FIELD(number, name) [number] = VMCS12_OFFSET(name)
529 #define FIELD64(number, name) [number] = VMCS12_OFFSET(name), \
530 [number##_HIGH] = VMCS12_OFFSET(name)+4
533 static unsigned long shadow_read_only_fields[] = {
535 * We do NOT shadow fields that are modified when L0
536 * traps and emulates any vmx instruction (e.g. VMPTRLD,
537 * VMXON...) executed by L1.
538 * For example, VM_INSTRUCTION_ERROR is read
539 * by L1 if a vmx instruction fails (part of the error path).
540 * Note the code assumes this logic. If for some reason
541 * we start shadowing these fields then we need to
542 * force a shadow sync when L0 emulates vmx instructions
543 * (e.g. force a sync if VM_INSTRUCTION_ERROR is modified
544 * by nested_vmx_failValid)
548 VM_EXIT_INSTRUCTION_LEN,
549 IDT_VECTORING_INFO_FIELD,
550 IDT_VECTORING_ERROR_CODE,
551 VM_EXIT_INTR_ERROR_CODE,
553 GUEST_LINEAR_ADDRESS,
554 GUEST_PHYSICAL_ADDRESS
556 static int max_shadow_read_only_fields =
557 ARRAY_SIZE(shadow_read_only_fields);
559 static unsigned long shadow_read_write_fields[] = {
566 GUEST_INTERRUPTIBILITY_INFO,
579 CPU_BASED_VM_EXEC_CONTROL,
580 VM_ENTRY_EXCEPTION_ERROR_CODE,
581 VM_ENTRY_INTR_INFO_FIELD,
582 VM_ENTRY_INSTRUCTION_LEN,
583 VM_ENTRY_EXCEPTION_ERROR_CODE,
589 static int max_shadow_read_write_fields =
590 ARRAY_SIZE(shadow_read_write_fields);
592 static const unsigned short vmcs_field_to_offset_table[] = {
593 FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
594 FIELD(GUEST_ES_SELECTOR, guest_es_selector),
595 FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
596 FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
597 FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
598 FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
599 FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
600 FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
601 FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
602 FIELD(HOST_ES_SELECTOR, host_es_selector),
603 FIELD(HOST_CS_SELECTOR, host_cs_selector),
604 FIELD(HOST_SS_SELECTOR, host_ss_selector),
605 FIELD(HOST_DS_SELECTOR, host_ds_selector),
606 FIELD(HOST_FS_SELECTOR, host_fs_selector),
607 FIELD(HOST_GS_SELECTOR, host_gs_selector),
608 FIELD(HOST_TR_SELECTOR, host_tr_selector),
609 FIELD64(IO_BITMAP_A, io_bitmap_a),
610 FIELD64(IO_BITMAP_B, io_bitmap_b),
611 FIELD64(MSR_BITMAP, msr_bitmap),
612 FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
613 FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
614 FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
615 FIELD64(TSC_OFFSET, tsc_offset),
616 FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
617 FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
618 FIELD64(EPT_POINTER, ept_pointer),
619 FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
620 FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
621 FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
622 FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
623 FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
624 FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
625 FIELD64(GUEST_PDPTR0, guest_pdptr0),
626 FIELD64(GUEST_PDPTR1, guest_pdptr1),
627 FIELD64(GUEST_PDPTR2, guest_pdptr2),
628 FIELD64(GUEST_PDPTR3, guest_pdptr3),
629 FIELD64(GUEST_BNDCFGS, guest_bndcfgs),
630 FIELD64(HOST_IA32_PAT, host_ia32_pat),
631 FIELD64(HOST_IA32_EFER, host_ia32_efer),
632 FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
633 FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
634 FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
635 FIELD(EXCEPTION_BITMAP, exception_bitmap),
636 FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
637 FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
638 FIELD(CR3_TARGET_COUNT, cr3_target_count),
639 FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
640 FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
641 FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
642 FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
643 FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
644 FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
645 FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
646 FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
647 FIELD(TPR_THRESHOLD, tpr_threshold),
648 FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
649 FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
650 FIELD(VM_EXIT_REASON, vm_exit_reason),
651 FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
652 FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
653 FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
654 FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
655 FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
656 FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
657 FIELD(GUEST_ES_LIMIT, guest_es_limit),
658 FIELD(GUEST_CS_LIMIT, guest_cs_limit),
659 FIELD(GUEST_SS_LIMIT, guest_ss_limit),
660 FIELD(GUEST_DS_LIMIT, guest_ds_limit),
661 FIELD(GUEST_FS_LIMIT, guest_fs_limit),
662 FIELD(GUEST_GS_LIMIT, guest_gs_limit),
663 FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
664 FIELD(GUEST_TR_LIMIT, guest_tr_limit),
665 FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
666 FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
667 FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
668 FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
669 FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
670 FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
671 FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
672 FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
673 FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
674 FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
675 FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
676 FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
677 FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
678 FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
679 FIELD(VMX_PREEMPTION_TIMER_VALUE, vmx_preemption_timer_value),
680 FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
681 FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
682 FIELD(CR0_READ_SHADOW, cr0_read_shadow),
683 FIELD(CR4_READ_SHADOW, cr4_read_shadow),
684 FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
685 FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
686 FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
687 FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
688 FIELD(EXIT_QUALIFICATION, exit_qualification),
689 FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
690 FIELD(GUEST_CR0, guest_cr0),
691 FIELD(GUEST_CR3, guest_cr3),
692 FIELD(GUEST_CR4, guest_cr4),
693 FIELD(GUEST_ES_BASE, guest_es_base),
694 FIELD(GUEST_CS_BASE, guest_cs_base),
695 FIELD(GUEST_SS_BASE, guest_ss_base),
696 FIELD(GUEST_DS_BASE, guest_ds_base),
697 FIELD(GUEST_FS_BASE, guest_fs_base),
698 FIELD(GUEST_GS_BASE, guest_gs_base),
699 FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
700 FIELD(GUEST_TR_BASE, guest_tr_base),
701 FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
702 FIELD(GUEST_IDTR_BASE, guest_idtr_base),
703 FIELD(GUEST_DR7, guest_dr7),
704 FIELD(GUEST_RSP, guest_rsp),
705 FIELD(GUEST_RIP, guest_rip),
706 FIELD(GUEST_RFLAGS, guest_rflags),
707 FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
708 FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
709 FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
710 FIELD(HOST_CR0, host_cr0),
711 FIELD(HOST_CR3, host_cr3),
712 FIELD(HOST_CR4, host_cr4),
713 FIELD(HOST_FS_BASE, host_fs_base),
714 FIELD(HOST_GS_BASE, host_gs_base),
715 FIELD(HOST_TR_BASE, host_tr_base),
716 FIELD(HOST_GDTR_BASE, host_gdtr_base),
717 FIELD(HOST_IDTR_BASE, host_idtr_base),
718 FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
719 FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
720 FIELD(HOST_RSP, host_rsp),
721 FIELD(HOST_RIP, host_rip),
723 static const int max_vmcs_field = ARRAY_SIZE(vmcs_field_to_offset_table);
725 static inline short vmcs_field_to_offset(unsigned long field)
727 if (field >= max_vmcs_field || vmcs_field_to_offset_table[field] == 0)
729 return vmcs_field_to_offset_table[field];
732 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
734 return to_vmx(vcpu)->nested.current_vmcs12;
737 static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
739 struct page *page = gfn_to_page(vcpu->kvm, addr >> PAGE_SHIFT);
740 if (is_error_page(page))
746 static void nested_release_page(struct page *page)
748 kvm_release_page_dirty(page);
751 static void nested_release_page_clean(struct page *page)
753 kvm_release_page_clean(page);
756 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu);
757 static u64 construct_eptp(unsigned long root_hpa);
758 static void kvm_cpu_vmxon(u64 addr);
759 static void kvm_cpu_vmxoff(void);
760 static bool vmx_mpx_supported(void);
761 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
762 static void vmx_set_segment(struct kvm_vcpu *vcpu,
763 struct kvm_segment *var, int seg);
764 static void vmx_get_segment(struct kvm_vcpu *vcpu,
765 struct kvm_segment *var, int seg);
766 static bool guest_state_valid(struct kvm_vcpu *vcpu);
767 static u32 vmx_segment_access_rights(struct kvm_segment *var);
768 static void vmx_sync_pir_to_irr_dummy(struct kvm_vcpu *vcpu);
769 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx);
770 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx);
771 static int alloc_identity_pagetable(struct kvm *kvm);
773 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
774 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
776 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
777 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
779 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
780 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
782 static unsigned long *vmx_io_bitmap_a;
783 static unsigned long *vmx_io_bitmap_b;
784 static unsigned long *vmx_msr_bitmap_legacy;
785 static unsigned long *vmx_msr_bitmap_longmode;
786 static unsigned long *vmx_msr_bitmap_legacy_x2apic;
787 static unsigned long *vmx_msr_bitmap_longmode_x2apic;
788 static unsigned long *vmx_vmread_bitmap;
789 static unsigned long *vmx_vmwrite_bitmap;
791 static bool cpu_has_load_ia32_efer;
792 static bool cpu_has_load_perf_global_ctrl;
794 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
795 static DEFINE_SPINLOCK(vmx_vpid_lock);
797 static struct vmcs_config {
801 u32 pin_based_exec_ctrl;
802 u32 cpu_based_exec_ctrl;
803 u32 cpu_based_2nd_exec_ctrl;
808 static struct vmx_capability {
813 #define VMX_SEGMENT_FIELD(seg) \
814 [VCPU_SREG_##seg] = { \
815 .selector = GUEST_##seg##_SELECTOR, \
816 .base = GUEST_##seg##_BASE, \
817 .limit = GUEST_##seg##_LIMIT, \
818 .ar_bytes = GUEST_##seg##_AR_BYTES, \
821 static const struct kvm_vmx_segment_field {
826 } kvm_vmx_segment_fields[] = {
827 VMX_SEGMENT_FIELD(CS),
828 VMX_SEGMENT_FIELD(DS),
829 VMX_SEGMENT_FIELD(ES),
830 VMX_SEGMENT_FIELD(FS),
831 VMX_SEGMENT_FIELD(GS),
832 VMX_SEGMENT_FIELD(SS),
833 VMX_SEGMENT_FIELD(TR),
834 VMX_SEGMENT_FIELD(LDTR),
837 static u64 host_efer;
839 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
842 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
843 * away by decrementing the array size.
845 static const u32 vmx_msr_index[] = {
847 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
849 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
852 static inline bool is_page_fault(u32 intr_info)
854 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
855 INTR_INFO_VALID_MASK)) ==
856 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
859 static inline bool is_no_device(u32 intr_info)
861 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
862 INTR_INFO_VALID_MASK)) ==
863 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
866 static inline bool is_invalid_opcode(u32 intr_info)
868 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
869 INTR_INFO_VALID_MASK)) ==
870 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
873 static inline bool is_external_interrupt(u32 intr_info)
875 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
876 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
879 static inline bool is_machine_check(u32 intr_info)
881 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
882 INTR_INFO_VALID_MASK)) ==
883 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
886 static inline bool cpu_has_vmx_msr_bitmap(void)
888 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
891 static inline bool cpu_has_vmx_tpr_shadow(void)
893 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
896 static inline bool vm_need_tpr_shadow(struct kvm *kvm)
898 return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
901 static inline bool cpu_has_secondary_exec_ctrls(void)
903 return vmcs_config.cpu_based_exec_ctrl &
904 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
907 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
909 return vmcs_config.cpu_based_2nd_exec_ctrl &
910 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
913 static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
915 return vmcs_config.cpu_based_2nd_exec_ctrl &
916 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
919 static inline bool cpu_has_vmx_apic_register_virt(void)
921 return vmcs_config.cpu_based_2nd_exec_ctrl &
922 SECONDARY_EXEC_APIC_REGISTER_VIRT;
925 static inline bool cpu_has_vmx_virtual_intr_delivery(void)
927 return vmcs_config.cpu_based_2nd_exec_ctrl &
928 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
931 static inline bool cpu_has_vmx_posted_intr(void)
933 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_POSTED_INTR;
936 static inline bool cpu_has_vmx_apicv(void)
938 return cpu_has_vmx_apic_register_virt() &&
939 cpu_has_vmx_virtual_intr_delivery() &&
940 cpu_has_vmx_posted_intr();
943 static inline bool cpu_has_vmx_flexpriority(void)
945 return cpu_has_vmx_tpr_shadow() &&
946 cpu_has_vmx_virtualize_apic_accesses();
949 static inline bool cpu_has_vmx_ept_execute_only(void)
951 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
954 static inline bool cpu_has_vmx_eptp_uncacheable(void)
956 return vmx_capability.ept & VMX_EPTP_UC_BIT;
959 static inline bool cpu_has_vmx_eptp_writeback(void)
961 return vmx_capability.ept & VMX_EPTP_WB_BIT;
964 static inline bool cpu_has_vmx_ept_2m_page(void)
966 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
969 static inline bool cpu_has_vmx_ept_1g_page(void)
971 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
974 static inline bool cpu_has_vmx_ept_4levels(void)
976 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
979 static inline bool cpu_has_vmx_ept_ad_bits(void)
981 return vmx_capability.ept & VMX_EPT_AD_BIT;
984 static inline bool cpu_has_vmx_invept_context(void)
986 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
989 static inline bool cpu_has_vmx_invept_global(void)
991 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
994 static inline bool cpu_has_vmx_invvpid_single(void)
996 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
999 static inline bool cpu_has_vmx_invvpid_global(void)
1001 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
1004 static inline bool cpu_has_vmx_ept(void)
1006 return vmcs_config.cpu_based_2nd_exec_ctrl &
1007 SECONDARY_EXEC_ENABLE_EPT;
1010 static inline bool cpu_has_vmx_unrestricted_guest(void)
1012 return vmcs_config.cpu_based_2nd_exec_ctrl &
1013 SECONDARY_EXEC_UNRESTRICTED_GUEST;
1016 static inline bool cpu_has_vmx_ple(void)
1018 return vmcs_config.cpu_based_2nd_exec_ctrl &
1019 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
1022 static inline bool vm_need_virtualize_apic_accesses(struct kvm *kvm)
1024 return flexpriority_enabled && irqchip_in_kernel(kvm);
1027 static inline bool cpu_has_vmx_vpid(void)
1029 return vmcs_config.cpu_based_2nd_exec_ctrl &
1030 SECONDARY_EXEC_ENABLE_VPID;
1033 static inline bool cpu_has_vmx_rdtscp(void)
1035 return vmcs_config.cpu_based_2nd_exec_ctrl &
1036 SECONDARY_EXEC_RDTSCP;
1039 static inline bool cpu_has_vmx_invpcid(void)
1041 return vmcs_config.cpu_based_2nd_exec_ctrl &
1042 SECONDARY_EXEC_ENABLE_INVPCID;
1045 static inline bool cpu_has_virtual_nmis(void)
1047 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
1050 static inline bool cpu_has_vmx_wbinvd_exit(void)
1052 return vmcs_config.cpu_based_2nd_exec_ctrl &
1053 SECONDARY_EXEC_WBINVD_EXITING;
1056 static inline bool cpu_has_vmx_shadow_vmcs(void)
1059 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
1060 /* check if the cpu supports writing r/o exit information fields */
1061 if (!(vmx_msr & MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS))
1064 return vmcs_config.cpu_based_2nd_exec_ctrl &
1065 SECONDARY_EXEC_SHADOW_VMCS;
1068 static inline bool report_flexpriority(void)
1070 return flexpriority_enabled;
1073 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
1075 return vmcs12->cpu_based_vm_exec_control & bit;
1078 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
1080 return (vmcs12->cpu_based_vm_exec_control &
1081 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
1082 (vmcs12->secondary_vm_exec_control & bit);
1085 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12)
1087 return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
1090 static inline bool nested_cpu_has_preemption_timer(struct vmcs12 *vmcs12)
1092 return vmcs12->pin_based_vm_exec_control &
1093 PIN_BASED_VMX_PREEMPTION_TIMER;
1096 static inline int nested_cpu_has_ept(struct vmcs12 *vmcs12)
1098 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_EPT);
1101 static inline bool is_exception(u32 intr_info)
1103 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1104 == (INTR_TYPE_HARD_EXCEPTION | INTR_INFO_VALID_MASK);
1107 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
1109 unsigned long exit_qualification);
1110 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
1111 struct vmcs12 *vmcs12,
1112 u32 reason, unsigned long qualification);
1114 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
1118 for (i = 0; i < vmx->nmsrs; ++i)
1119 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
1124 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
1130 } operand = { vpid, 0, gva };
1132 asm volatile (__ex(ASM_VMX_INVVPID)
1133 /* CF==1 or ZF==1 --> rc = -1 */
1134 "; ja 1f ; ud2 ; 1:"
1135 : : "a"(&operand), "c"(ext) : "cc", "memory");
1138 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
1142 } operand = {eptp, gpa};
1144 asm volatile (__ex(ASM_VMX_INVEPT)
1145 /* CF==1 or ZF==1 --> rc = -1 */
1146 "; ja 1f ; ud2 ; 1:\n"
1147 : : "a" (&operand), "c" (ext) : "cc", "memory");
1150 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
1154 i = __find_msr_index(vmx, msr);
1156 return &vmx->guest_msrs[i];
1160 static void vmcs_clear(struct vmcs *vmcs)
1162 u64 phys_addr = __pa(vmcs);
1165 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
1166 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1169 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
1173 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
1175 vmcs_clear(loaded_vmcs->vmcs);
1176 loaded_vmcs->cpu = -1;
1177 loaded_vmcs->launched = 0;
1180 static void vmcs_load(struct vmcs *vmcs)
1182 u64 phys_addr = __pa(vmcs);
1185 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
1186 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1189 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
1195 * This bitmap is used to indicate whether the vmclear
1196 * operation is enabled on all cpus. All disabled by
1199 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
1201 static inline void crash_enable_local_vmclear(int cpu)
1203 cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
1206 static inline void crash_disable_local_vmclear(int cpu)
1208 cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
1211 static inline int crash_local_vmclear_enabled(int cpu)
1213 return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
1216 static void crash_vmclear_local_loaded_vmcss(void)
1218 int cpu = raw_smp_processor_id();
1219 struct loaded_vmcs *v;
1221 if (!crash_local_vmclear_enabled(cpu))
1224 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
1225 loaded_vmcss_on_cpu_link)
1226 vmcs_clear(v->vmcs);
1229 static inline void crash_enable_local_vmclear(int cpu) { }
1230 static inline void crash_disable_local_vmclear(int cpu) { }
1231 #endif /* CONFIG_KEXEC */
1233 static void __loaded_vmcs_clear(void *arg)
1235 struct loaded_vmcs *loaded_vmcs = arg;
1236 int cpu = raw_smp_processor_id();
1238 if (loaded_vmcs->cpu != cpu)
1239 return; /* vcpu migration can race with cpu offline */
1240 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1241 per_cpu(current_vmcs, cpu) = NULL;
1242 crash_disable_local_vmclear(cpu);
1243 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1246 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1247 * is before setting loaded_vmcs->vcpu to -1 which is done in
1248 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1249 * then adds the vmcs into percpu list before it is deleted.
1253 loaded_vmcs_init(loaded_vmcs);
1254 crash_enable_local_vmclear(cpu);
1257 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1259 int cpu = loaded_vmcs->cpu;
1262 smp_call_function_single(cpu,
1263 __loaded_vmcs_clear, loaded_vmcs, 1);
1266 static inline void vpid_sync_vcpu_single(struct vcpu_vmx *vmx)
1271 if (cpu_has_vmx_invvpid_single())
1272 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
1275 static inline void vpid_sync_vcpu_global(void)
1277 if (cpu_has_vmx_invvpid_global())
1278 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1281 static inline void vpid_sync_context(struct vcpu_vmx *vmx)
1283 if (cpu_has_vmx_invvpid_single())
1284 vpid_sync_vcpu_single(vmx);
1286 vpid_sync_vcpu_global();
1289 static inline void ept_sync_global(void)
1291 if (cpu_has_vmx_invept_global())
1292 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1295 static inline void ept_sync_context(u64 eptp)
1298 if (cpu_has_vmx_invept_context())
1299 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1305 static __always_inline unsigned long vmcs_readl(unsigned long field)
1307 unsigned long value;
1309 asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1310 : "=a"(value) : "d"(field) : "cc");
1314 static __always_inline u16 vmcs_read16(unsigned long field)
1316 return vmcs_readl(field);
1319 static __always_inline u32 vmcs_read32(unsigned long field)
1321 return vmcs_readl(field);
1324 static __always_inline u64 vmcs_read64(unsigned long field)
1326 #ifdef CONFIG_X86_64
1327 return vmcs_readl(field);
1329 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
1333 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1335 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1336 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1340 static void vmcs_writel(unsigned long field, unsigned long value)
1344 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1345 : "=q"(error) : "a"(value), "d"(field) : "cc");
1346 if (unlikely(error))
1347 vmwrite_error(field, value);
1350 static void vmcs_write16(unsigned long field, u16 value)
1352 vmcs_writel(field, value);
1355 static void vmcs_write32(unsigned long field, u32 value)
1357 vmcs_writel(field, value);
1360 static void vmcs_write64(unsigned long field, u64 value)
1362 vmcs_writel(field, value);
1363 #ifndef CONFIG_X86_64
1365 vmcs_writel(field+1, value >> 32);
1369 static void vmcs_clear_bits(unsigned long field, u32 mask)
1371 vmcs_writel(field, vmcs_readl(field) & ~mask);
1374 static void vmcs_set_bits(unsigned long field, u32 mask)
1376 vmcs_writel(field, vmcs_readl(field) | mask);
1379 static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val)
1381 vmcs_write32(VM_ENTRY_CONTROLS, val);
1382 vmx->vm_entry_controls_shadow = val;
1385 static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val)
1387 if (vmx->vm_entry_controls_shadow != val)
1388 vm_entry_controls_init(vmx, val);
1391 static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx)
1393 return vmx->vm_entry_controls_shadow;
1397 static inline void vm_entry_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1399 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) | val);
1402 static inline void vm_entry_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1404 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) & ~val);
1407 static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val)
1409 vmcs_write32(VM_EXIT_CONTROLS, val);
1410 vmx->vm_exit_controls_shadow = val;
1413 static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val)
1415 if (vmx->vm_exit_controls_shadow != val)
1416 vm_exit_controls_init(vmx, val);
1419 static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx)
1421 return vmx->vm_exit_controls_shadow;
1425 static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1427 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val);
1430 static inline void vm_exit_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1432 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) & ~val);
1435 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1437 vmx->segment_cache.bitmask = 0;
1440 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1444 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1446 if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1447 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1448 vmx->segment_cache.bitmask = 0;
1450 ret = vmx->segment_cache.bitmask & mask;
1451 vmx->segment_cache.bitmask |= mask;
1455 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1457 u16 *p = &vmx->segment_cache.seg[seg].selector;
1459 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1460 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1464 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1466 ulong *p = &vmx->segment_cache.seg[seg].base;
1468 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1469 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1473 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1475 u32 *p = &vmx->segment_cache.seg[seg].limit;
1477 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1478 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1482 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1484 u32 *p = &vmx->segment_cache.seg[seg].ar;
1486 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1487 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1491 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1495 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1496 (1u << NM_VECTOR) | (1u << DB_VECTOR);
1497 if ((vcpu->guest_debug &
1498 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1499 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1500 eb |= 1u << BP_VECTOR;
1501 if (to_vmx(vcpu)->rmode.vm86_active)
1504 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1505 if (vcpu->fpu_active)
1506 eb &= ~(1u << NM_VECTOR);
1508 /* When we are running a nested L2 guest and L1 specified for it a
1509 * certain exception bitmap, we must trap the same exceptions and pass
1510 * them to L1. When running L2, we will only handle the exceptions
1511 * specified above if L1 did not want them.
1513 if (is_guest_mode(vcpu))
1514 eb |= get_vmcs12(vcpu)->exception_bitmap;
1516 vmcs_write32(EXCEPTION_BITMAP, eb);
1519 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1520 unsigned long entry, unsigned long exit)
1522 vm_entry_controls_clearbit(vmx, entry);
1523 vm_exit_controls_clearbit(vmx, exit);
1526 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1529 struct msr_autoload *m = &vmx->msr_autoload;
1533 if (cpu_has_load_ia32_efer) {
1534 clear_atomic_switch_msr_special(vmx,
1535 VM_ENTRY_LOAD_IA32_EFER,
1536 VM_EXIT_LOAD_IA32_EFER);
1540 case MSR_CORE_PERF_GLOBAL_CTRL:
1541 if (cpu_has_load_perf_global_ctrl) {
1542 clear_atomic_switch_msr_special(vmx,
1543 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1544 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1550 for (i = 0; i < m->nr; ++i)
1551 if (m->guest[i].index == msr)
1557 m->guest[i] = m->guest[m->nr];
1558 m->host[i] = m->host[m->nr];
1559 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1560 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1563 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1564 unsigned long entry, unsigned long exit,
1565 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
1566 u64 guest_val, u64 host_val)
1568 vmcs_write64(guest_val_vmcs, guest_val);
1569 vmcs_write64(host_val_vmcs, host_val);
1570 vm_entry_controls_setbit(vmx, entry);
1571 vm_exit_controls_setbit(vmx, exit);
1574 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1575 u64 guest_val, u64 host_val)
1578 struct msr_autoload *m = &vmx->msr_autoload;
1582 if (cpu_has_load_ia32_efer) {
1583 add_atomic_switch_msr_special(vmx,
1584 VM_ENTRY_LOAD_IA32_EFER,
1585 VM_EXIT_LOAD_IA32_EFER,
1588 guest_val, host_val);
1592 case MSR_CORE_PERF_GLOBAL_CTRL:
1593 if (cpu_has_load_perf_global_ctrl) {
1594 add_atomic_switch_msr_special(vmx,
1595 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1596 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1597 GUEST_IA32_PERF_GLOBAL_CTRL,
1598 HOST_IA32_PERF_GLOBAL_CTRL,
1599 guest_val, host_val);
1605 for (i = 0; i < m->nr; ++i)
1606 if (m->guest[i].index == msr)
1609 if (i == NR_AUTOLOAD_MSRS) {
1610 printk_once(KERN_WARNING "Not enough msr switch entries. "
1611 "Can't add msr %x\n", msr);
1613 } else if (i == m->nr) {
1615 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1616 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1619 m->guest[i].index = msr;
1620 m->guest[i].value = guest_val;
1621 m->host[i].index = msr;
1622 m->host[i].value = host_val;
1625 static void reload_tss(void)
1628 * VT restores TR but not its size. Useless.
1630 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
1631 struct desc_struct *descs;
1633 descs = (void *)gdt->address;
1634 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
1638 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
1643 guest_efer = vmx->vcpu.arch.efer;
1646 * NX is emulated; LMA and LME handled by hardware; SCE meaningless
1649 ignore_bits = EFER_NX | EFER_SCE;
1650 #ifdef CONFIG_X86_64
1651 ignore_bits |= EFER_LMA | EFER_LME;
1652 /* SCE is meaningful only in long mode on Intel */
1653 if (guest_efer & EFER_LMA)
1654 ignore_bits &= ~(u64)EFER_SCE;
1656 guest_efer &= ~ignore_bits;
1657 guest_efer |= host_efer & ignore_bits;
1658 vmx->guest_msrs[efer_offset].data = guest_efer;
1659 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1661 clear_atomic_switch_msr(vmx, MSR_EFER);
1662 /* On ept, can't emulate nx, and must switch nx atomically */
1663 if (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX)) {
1664 guest_efer = vmx->vcpu.arch.efer;
1665 if (!(guest_efer & EFER_LMA))
1666 guest_efer &= ~EFER_LME;
1667 add_atomic_switch_msr(vmx, MSR_EFER, guest_efer, host_efer);
1674 static unsigned long segment_base(u16 selector)
1676 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
1677 struct desc_struct *d;
1678 unsigned long table_base;
1681 if (!(selector & ~3))
1684 table_base = gdt->address;
1686 if (selector & 4) { /* from ldt */
1687 u16 ldt_selector = kvm_read_ldt();
1689 if (!(ldt_selector & ~3))
1692 table_base = segment_base(ldt_selector);
1694 d = (struct desc_struct *)(table_base + (selector & ~7));
1695 v = get_desc_base(d);
1696 #ifdef CONFIG_X86_64
1697 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
1698 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
1703 static inline unsigned long kvm_read_tr_base(void)
1706 asm("str %0" : "=g"(tr));
1707 return segment_base(tr);
1710 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
1712 struct vcpu_vmx *vmx = to_vmx(vcpu);
1715 if (vmx->host_state.loaded)
1718 vmx->host_state.loaded = 1;
1720 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
1721 * allow segment selectors with cpl > 0 or ti == 1.
1723 vmx->host_state.ldt_sel = kvm_read_ldt();
1724 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
1725 savesegment(fs, vmx->host_state.fs_sel);
1726 if (!(vmx->host_state.fs_sel & 7)) {
1727 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
1728 vmx->host_state.fs_reload_needed = 0;
1730 vmcs_write16(HOST_FS_SELECTOR, 0);
1731 vmx->host_state.fs_reload_needed = 1;
1733 savesegment(gs, vmx->host_state.gs_sel);
1734 if (!(vmx->host_state.gs_sel & 7))
1735 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
1737 vmcs_write16(HOST_GS_SELECTOR, 0);
1738 vmx->host_state.gs_ldt_reload_needed = 1;
1741 #ifdef CONFIG_X86_64
1742 savesegment(ds, vmx->host_state.ds_sel);
1743 savesegment(es, vmx->host_state.es_sel);
1746 #ifdef CONFIG_X86_64
1747 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
1748 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
1750 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
1751 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
1754 #ifdef CONFIG_X86_64
1755 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1756 if (is_long_mode(&vmx->vcpu))
1757 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1759 if (boot_cpu_has(X86_FEATURE_MPX))
1760 rdmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
1761 for (i = 0; i < vmx->save_nmsrs; ++i)
1762 kvm_set_shared_msr(vmx->guest_msrs[i].index,
1763 vmx->guest_msrs[i].data,
1764 vmx->guest_msrs[i].mask);
1767 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
1769 if (!vmx->host_state.loaded)
1772 ++vmx->vcpu.stat.host_state_reload;
1773 vmx->host_state.loaded = 0;
1774 #ifdef CONFIG_X86_64
1775 if (is_long_mode(&vmx->vcpu))
1776 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1778 if (vmx->host_state.gs_ldt_reload_needed) {
1779 kvm_load_ldt(vmx->host_state.ldt_sel);
1780 #ifdef CONFIG_X86_64
1781 load_gs_index(vmx->host_state.gs_sel);
1783 loadsegment(gs, vmx->host_state.gs_sel);
1786 if (vmx->host_state.fs_reload_needed)
1787 loadsegment(fs, vmx->host_state.fs_sel);
1788 #ifdef CONFIG_X86_64
1789 if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
1790 loadsegment(ds, vmx->host_state.ds_sel);
1791 loadsegment(es, vmx->host_state.es_sel);
1795 #ifdef CONFIG_X86_64
1796 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1798 if (vmx->host_state.msr_host_bndcfgs)
1799 wrmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
1801 * If the FPU is not active (through the host task or
1802 * the guest vcpu), then restore the cr0.TS bit.
1804 if (!user_has_fpu() && !vmx->vcpu.guest_fpu_loaded)
1806 load_gdt(this_cpu_ptr(&host_gdt));
1809 static void vmx_load_host_state(struct vcpu_vmx *vmx)
1812 __vmx_load_host_state(vmx);
1817 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1818 * vcpu mutex is already taken.
1820 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1822 struct vcpu_vmx *vmx = to_vmx(vcpu);
1823 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1826 kvm_cpu_vmxon(phys_addr);
1827 else if (vmx->loaded_vmcs->cpu != cpu)
1828 loaded_vmcs_clear(vmx->loaded_vmcs);
1830 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1831 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1832 vmcs_load(vmx->loaded_vmcs->vmcs);
1835 if (vmx->loaded_vmcs->cpu != cpu) {
1836 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
1837 unsigned long sysenter_esp;
1839 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1840 local_irq_disable();
1841 crash_disable_local_vmclear(cpu);
1844 * Read loaded_vmcs->cpu should be before fetching
1845 * loaded_vmcs->loaded_vmcss_on_cpu_link.
1846 * See the comments in __loaded_vmcs_clear().
1850 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1851 &per_cpu(loaded_vmcss_on_cpu, cpu));
1852 crash_enable_local_vmclear(cpu);
1856 * Linux uses per-cpu TSS and GDT, so set these when switching
1859 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
1860 vmcs_writel(HOST_GDTR_BASE, gdt->address); /* 22.2.4 */
1862 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1863 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1864 vmx->loaded_vmcs->cpu = cpu;
1868 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1870 __vmx_load_host_state(to_vmx(vcpu));
1871 if (!vmm_exclusive) {
1872 __loaded_vmcs_clear(to_vmx(vcpu)->loaded_vmcs);
1878 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
1882 if (vcpu->fpu_active)
1884 vcpu->fpu_active = 1;
1885 cr0 = vmcs_readl(GUEST_CR0);
1886 cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
1887 cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
1888 vmcs_writel(GUEST_CR0, cr0);
1889 update_exception_bitmap(vcpu);
1890 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
1891 if (is_guest_mode(vcpu))
1892 vcpu->arch.cr0_guest_owned_bits &=
1893 ~get_vmcs12(vcpu)->cr0_guest_host_mask;
1894 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1897 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
1900 * Return the cr0 value that a nested guest would read. This is a combination
1901 * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
1902 * its hypervisor (cr0_read_shadow).
1904 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
1906 return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
1907 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
1909 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
1911 return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
1912 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
1915 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
1917 /* Note that there is no vcpu->fpu_active = 0 here. The caller must
1918 * set this *before* calling this function.
1920 vmx_decache_cr0_guest_bits(vcpu);
1921 vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
1922 update_exception_bitmap(vcpu);
1923 vcpu->arch.cr0_guest_owned_bits = 0;
1924 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1925 if (is_guest_mode(vcpu)) {
1927 * L1's specified read shadow might not contain the TS bit,
1928 * so now that we turned on shadowing of this bit, we need to
1929 * set this bit of the shadow. Like in nested_vmx_run we need
1930 * nested_read_cr0(vmcs12), but vmcs12->guest_cr0 is not yet
1931 * up-to-date here because we just decached cr0.TS (and we'll
1932 * only update vmcs12->guest_cr0 on nested exit).
1934 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1935 vmcs12->guest_cr0 = (vmcs12->guest_cr0 & ~X86_CR0_TS) |
1936 (vcpu->arch.cr0 & X86_CR0_TS);
1937 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
1939 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
1942 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1944 unsigned long rflags, save_rflags;
1946 if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
1947 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1948 rflags = vmcs_readl(GUEST_RFLAGS);
1949 if (to_vmx(vcpu)->rmode.vm86_active) {
1950 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1951 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
1952 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1954 to_vmx(vcpu)->rflags = rflags;
1956 return to_vmx(vcpu)->rflags;
1959 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1961 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1962 to_vmx(vcpu)->rflags = rflags;
1963 if (to_vmx(vcpu)->rmode.vm86_active) {
1964 to_vmx(vcpu)->rmode.save_rflags = rflags;
1965 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1967 vmcs_writel(GUEST_RFLAGS, rflags);
1970 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
1972 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1975 if (interruptibility & GUEST_INTR_STATE_STI)
1976 ret |= KVM_X86_SHADOW_INT_STI;
1977 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1978 ret |= KVM_X86_SHADOW_INT_MOV_SS;
1983 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1985 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1986 u32 interruptibility = interruptibility_old;
1988 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1990 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1991 interruptibility |= GUEST_INTR_STATE_MOV_SS;
1992 else if (mask & KVM_X86_SHADOW_INT_STI)
1993 interruptibility |= GUEST_INTR_STATE_STI;
1995 if ((interruptibility != interruptibility_old))
1996 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1999 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
2003 rip = kvm_rip_read(vcpu);
2004 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
2005 kvm_rip_write(vcpu, rip);
2007 /* skipping an emulated instruction also counts */
2008 vmx_set_interrupt_shadow(vcpu, 0);
2012 * KVM wants to inject page-faults which it got to the guest. This function
2013 * checks whether in a nested guest, we need to inject them to L1 or L2.
2015 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned nr)
2017 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2019 if (!(vmcs12->exception_bitmap & (1u << nr)))
2022 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
2023 vmcs_read32(VM_EXIT_INTR_INFO),
2024 vmcs_readl(EXIT_QUALIFICATION));
2028 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
2029 bool has_error_code, u32 error_code,
2032 struct vcpu_vmx *vmx = to_vmx(vcpu);
2033 u32 intr_info = nr | INTR_INFO_VALID_MASK;
2035 if (!reinject && is_guest_mode(vcpu) &&
2036 nested_vmx_check_exception(vcpu, nr))
2039 if (has_error_code) {
2040 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
2041 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2044 if (vmx->rmode.vm86_active) {
2046 if (kvm_exception_is_soft(nr))
2047 inc_eip = vcpu->arch.event_exit_inst_len;
2048 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
2049 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2053 if (kvm_exception_is_soft(nr)) {
2054 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2055 vmx->vcpu.arch.event_exit_inst_len);
2056 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2058 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2060 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
2063 static bool vmx_rdtscp_supported(void)
2065 return cpu_has_vmx_rdtscp();
2068 static bool vmx_invpcid_supported(void)
2070 return cpu_has_vmx_invpcid() && enable_ept;
2074 * Swap MSR entry in host/guest MSR entry array.
2076 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
2078 struct shared_msr_entry tmp;
2080 tmp = vmx->guest_msrs[to];
2081 vmx->guest_msrs[to] = vmx->guest_msrs[from];
2082 vmx->guest_msrs[from] = tmp;
2085 static void vmx_set_msr_bitmap(struct kvm_vcpu *vcpu)
2087 unsigned long *msr_bitmap;
2089 if (irqchip_in_kernel(vcpu->kvm) && apic_x2apic_mode(vcpu->arch.apic)) {
2090 if (is_long_mode(vcpu))
2091 msr_bitmap = vmx_msr_bitmap_longmode_x2apic;
2093 msr_bitmap = vmx_msr_bitmap_legacy_x2apic;
2095 if (is_long_mode(vcpu))
2096 msr_bitmap = vmx_msr_bitmap_longmode;
2098 msr_bitmap = vmx_msr_bitmap_legacy;
2101 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
2105 * Set up the vmcs to automatically save and restore system
2106 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
2107 * mode, as fiddling with msrs is very expensive.
2109 static void setup_msrs(struct vcpu_vmx *vmx)
2111 int save_nmsrs, index;
2114 #ifdef CONFIG_X86_64
2115 if (is_long_mode(&vmx->vcpu)) {
2116 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
2118 move_msr_up(vmx, index, save_nmsrs++);
2119 index = __find_msr_index(vmx, MSR_LSTAR);
2121 move_msr_up(vmx, index, save_nmsrs++);
2122 index = __find_msr_index(vmx, MSR_CSTAR);
2124 move_msr_up(vmx, index, save_nmsrs++);
2125 index = __find_msr_index(vmx, MSR_TSC_AUX);
2126 if (index >= 0 && vmx->rdtscp_enabled)
2127 move_msr_up(vmx, index, save_nmsrs++);
2129 * MSR_STAR is only needed on long mode guests, and only
2130 * if efer.sce is enabled.
2132 index = __find_msr_index(vmx, MSR_STAR);
2133 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
2134 move_msr_up(vmx, index, save_nmsrs++);
2137 index = __find_msr_index(vmx, MSR_EFER);
2138 if (index >= 0 && update_transition_efer(vmx, index))
2139 move_msr_up(vmx, index, save_nmsrs++);
2141 vmx->save_nmsrs = save_nmsrs;
2143 if (cpu_has_vmx_msr_bitmap())
2144 vmx_set_msr_bitmap(&vmx->vcpu);
2148 * reads and returns guest's timestamp counter "register"
2149 * guest_tsc = host_tsc + tsc_offset -- 21.3
2151 static u64 guest_read_tsc(void)
2153 u64 host_tsc, tsc_offset;
2156 tsc_offset = vmcs_read64(TSC_OFFSET);
2157 return host_tsc + tsc_offset;
2161 * Like guest_read_tsc, but always returns L1's notion of the timestamp
2162 * counter, even if a nested guest (L2) is currently running.
2164 static u64 vmx_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2168 tsc_offset = is_guest_mode(vcpu) ?
2169 to_vmx(vcpu)->nested.vmcs01_tsc_offset :
2170 vmcs_read64(TSC_OFFSET);
2171 return host_tsc + tsc_offset;
2175 * Engage any workarounds for mis-matched TSC rates. Currently limited to
2176 * software catchup for faster rates on slower CPUs.
2178 static void vmx_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2183 if (user_tsc_khz > tsc_khz) {
2184 vcpu->arch.tsc_catchup = 1;
2185 vcpu->arch.tsc_always_catchup = 1;
2187 WARN(1, "user requested TSC rate below hardware speed\n");
2190 static u64 vmx_read_tsc_offset(struct kvm_vcpu *vcpu)
2192 return vmcs_read64(TSC_OFFSET);
2196 * writes 'offset' into guest's timestamp counter offset register
2198 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2200 if (is_guest_mode(vcpu)) {
2202 * We're here if L1 chose not to trap WRMSR to TSC. According
2203 * to the spec, this should set L1's TSC; The offset that L1
2204 * set for L2 remains unchanged, and still needs to be added
2205 * to the newly set TSC to get L2's TSC.
2207 struct vmcs12 *vmcs12;
2208 to_vmx(vcpu)->nested.vmcs01_tsc_offset = offset;
2209 /* recalculate vmcs02.TSC_OFFSET: */
2210 vmcs12 = get_vmcs12(vcpu);
2211 vmcs_write64(TSC_OFFSET, offset +
2212 (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
2213 vmcs12->tsc_offset : 0));
2215 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2216 vmcs_read64(TSC_OFFSET), offset);
2217 vmcs_write64(TSC_OFFSET, offset);
2221 static void vmx_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment, bool host)
2223 u64 offset = vmcs_read64(TSC_OFFSET);
2225 vmcs_write64(TSC_OFFSET, offset + adjustment);
2226 if (is_guest_mode(vcpu)) {
2227 /* Even when running L2, the adjustment needs to apply to L1 */
2228 to_vmx(vcpu)->nested.vmcs01_tsc_offset += adjustment;
2230 trace_kvm_write_tsc_offset(vcpu->vcpu_id, offset,
2231 offset + adjustment);
2234 static u64 vmx_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2236 return target_tsc - native_read_tsc();
2239 static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
2241 struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
2242 return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
2246 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
2247 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
2248 * all guests if the "nested" module option is off, and can also be disabled
2249 * for a single guest by disabling its VMX cpuid bit.
2251 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
2253 return nested && guest_cpuid_has_vmx(vcpu);
2257 * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
2258 * returned for the various VMX controls MSRs when nested VMX is enabled.
2259 * The same values should also be used to verify that vmcs12 control fields are
2260 * valid during nested entry from L1 to L2.
2261 * Each of these control msrs has a low and high 32-bit half: A low bit is on
2262 * if the corresponding bit in the (32-bit) control field *must* be on, and a
2263 * bit in the high half is on if the corresponding bit in the control field
2264 * may be on. See also vmx_control_verify().
2265 * TODO: allow these variables to be modified (downgraded) by module options
2268 static u32 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high;
2269 static u32 nested_vmx_true_procbased_ctls_low;
2270 static u32 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high;
2271 static u32 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high;
2272 static u32 nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high;
2273 static u32 nested_vmx_true_exit_ctls_low;
2274 static u32 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high;
2275 static u32 nested_vmx_true_entry_ctls_low;
2276 static u32 nested_vmx_misc_low, nested_vmx_misc_high;
2277 static u32 nested_vmx_ept_caps;
2278 static __init void nested_vmx_setup_ctls_msrs(void)
2281 * Note that as a general rule, the high half of the MSRs (bits in
2282 * the control fields which may be 1) should be initialized by the
2283 * intersection of the underlying hardware's MSR (i.e., features which
2284 * can be supported) and the list of features we want to expose -
2285 * because they are known to be properly supported in our code.
2286 * Also, usually, the low half of the MSRs (bits which must be 1) can
2287 * be set to 0, meaning that L1 may turn off any of these bits. The
2288 * reason is that if one of these bits is necessary, it will appear
2289 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
2290 * fields of vmcs01 and vmcs02, will turn these bits off - and
2291 * nested_vmx_exit_handled() will not pass related exits to L1.
2292 * These rules have exceptions below.
2295 /* pin-based controls */
2296 rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
2297 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high);
2298 nested_vmx_pinbased_ctls_low |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2299 nested_vmx_pinbased_ctls_high &= PIN_BASED_EXT_INTR_MASK |
2300 PIN_BASED_NMI_EXITING | PIN_BASED_VIRTUAL_NMIS;
2301 nested_vmx_pinbased_ctls_high |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2302 PIN_BASED_VMX_PREEMPTION_TIMER;
2305 rdmsr(MSR_IA32_VMX_EXIT_CTLS,
2306 nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high);
2307 nested_vmx_exit_ctls_low = VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
2309 nested_vmx_exit_ctls_high &=
2310 #ifdef CONFIG_X86_64
2311 VM_EXIT_HOST_ADDR_SPACE_SIZE |
2313 VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
2314 nested_vmx_exit_ctls_high |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
2315 VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
2316 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
2318 if (vmx_mpx_supported())
2319 nested_vmx_exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
2321 /* We support free control of debug control saving. */
2322 nested_vmx_true_exit_ctls_low = nested_vmx_exit_ctls_low &
2323 ~VM_EXIT_SAVE_DEBUG_CONTROLS;
2325 /* entry controls */
2326 rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
2327 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high);
2328 nested_vmx_entry_ctls_low = VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
2329 nested_vmx_entry_ctls_high &=
2330 #ifdef CONFIG_X86_64
2331 VM_ENTRY_IA32E_MODE |
2333 VM_ENTRY_LOAD_IA32_PAT;
2334 nested_vmx_entry_ctls_high |= (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR |
2335 VM_ENTRY_LOAD_IA32_EFER);
2336 if (vmx_mpx_supported())
2337 nested_vmx_entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
2339 /* We support free control of debug control loading. */
2340 nested_vmx_true_entry_ctls_low = nested_vmx_entry_ctls_low &
2341 ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
2343 /* cpu-based controls */
2344 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
2345 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high);
2346 nested_vmx_procbased_ctls_low = CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2347 nested_vmx_procbased_ctls_high &=
2348 CPU_BASED_VIRTUAL_INTR_PENDING |
2349 CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
2350 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
2351 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
2352 CPU_BASED_CR3_STORE_EXITING |
2353 #ifdef CONFIG_X86_64
2354 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
2356 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
2357 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_EXITING |
2358 CPU_BASED_RDPMC_EXITING | CPU_BASED_RDTSC_EXITING |
2359 CPU_BASED_PAUSE_EXITING | CPU_BASED_TPR_SHADOW |
2360 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2362 * We can allow some features even when not supported by the
2363 * hardware. For example, L1 can specify an MSR bitmap - and we
2364 * can use it to avoid exits to L1 - even when L0 runs L2
2365 * without MSR bitmaps.
2367 nested_vmx_procbased_ctls_high |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2368 CPU_BASED_USE_MSR_BITMAPS;
2370 /* We support free control of CR3 access interception. */
2371 nested_vmx_true_procbased_ctls_low = nested_vmx_procbased_ctls_low &
2372 ~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
2374 /* secondary cpu-based controls */
2375 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2376 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high);
2377 nested_vmx_secondary_ctls_low = 0;
2378 nested_vmx_secondary_ctls_high &=
2379 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2380 SECONDARY_EXEC_UNRESTRICTED_GUEST |
2381 SECONDARY_EXEC_WBINVD_EXITING;
2384 /* nested EPT: emulate EPT also to L1 */
2385 nested_vmx_secondary_ctls_high |= SECONDARY_EXEC_ENABLE_EPT;
2386 nested_vmx_ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
2387 VMX_EPTP_WB_BIT | VMX_EPT_2MB_PAGE_BIT |
2389 nested_vmx_ept_caps &= vmx_capability.ept;
2391 * For nested guests, we don't do anything specific
2392 * for single context invalidation. Hence, only advertise
2393 * support for global context invalidation.
2395 nested_vmx_ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT;
2397 nested_vmx_ept_caps = 0;
2399 /* miscellaneous data */
2400 rdmsr(MSR_IA32_VMX_MISC, nested_vmx_misc_low, nested_vmx_misc_high);
2401 nested_vmx_misc_low &= VMX_MISC_SAVE_EFER_LMA;
2402 nested_vmx_misc_low |= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
2403 VMX_MISC_ACTIVITY_HLT;
2404 nested_vmx_misc_high = 0;
2407 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2410 * Bits 0 in high must be 0, and bits 1 in low must be 1.
2412 return ((control & high) | low) == control;
2415 static inline u64 vmx_control_msr(u32 low, u32 high)
2417 return low | ((u64)high << 32);
2420 /* Returns 0 on success, non-0 otherwise. */
2421 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2423 switch (msr_index) {
2424 case MSR_IA32_VMX_BASIC:
2426 * This MSR reports some information about VMX support. We
2427 * should return information about the VMX we emulate for the
2428 * guest, and the VMCS structure we give it - not about the
2429 * VMX support of the underlying hardware.
2431 *pdata = VMCS12_REVISION | VMX_BASIC_TRUE_CTLS |
2432 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2433 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2435 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2436 case MSR_IA32_VMX_PINBASED_CTLS:
2437 *pdata = vmx_control_msr(nested_vmx_pinbased_ctls_low,
2438 nested_vmx_pinbased_ctls_high);
2440 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2441 *pdata = vmx_control_msr(nested_vmx_true_procbased_ctls_low,
2442 nested_vmx_procbased_ctls_high);
2444 case MSR_IA32_VMX_PROCBASED_CTLS:
2445 *pdata = vmx_control_msr(nested_vmx_procbased_ctls_low,
2446 nested_vmx_procbased_ctls_high);
2448 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2449 *pdata = vmx_control_msr(nested_vmx_true_exit_ctls_low,
2450 nested_vmx_exit_ctls_high);
2452 case MSR_IA32_VMX_EXIT_CTLS:
2453 *pdata = vmx_control_msr(nested_vmx_exit_ctls_low,
2454 nested_vmx_exit_ctls_high);
2456 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2457 *pdata = vmx_control_msr(nested_vmx_true_entry_ctls_low,
2458 nested_vmx_entry_ctls_high);
2460 case MSR_IA32_VMX_ENTRY_CTLS:
2461 *pdata = vmx_control_msr(nested_vmx_entry_ctls_low,
2462 nested_vmx_entry_ctls_high);
2464 case MSR_IA32_VMX_MISC:
2465 *pdata = vmx_control_msr(nested_vmx_misc_low,
2466 nested_vmx_misc_high);
2469 * These MSRs specify bits which the guest must keep fixed (on or off)
2470 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2471 * We picked the standard core2 setting.
2473 #define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2474 #define VMXON_CR4_ALWAYSON X86_CR4_VMXE
2475 case MSR_IA32_VMX_CR0_FIXED0:
2476 *pdata = VMXON_CR0_ALWAYSON;
2478 case MSR_IA32_VMX_CR0_FIXED1:
2481 case MSR_IA32_VMX_CR4_FIXED0:
2482 *pdata = VMXON_CR4_ALWAYSON;
2484 case MSR_IA32_VMX_CR4_FIXED1:
2487 case MSR_IA32_VMX_VMCS_ENUM:
2488 *pdata = 0x2e; /* highest index: VMX_PREEMPTION_TIMER_VALUE */
2490 case MSR_IA32_VMX_PROCBASED_CTLS2:
2491 *pdata = vmx_control_msr(nested_vmx_secondary_ctls_low,
2492 nested_vmx_secondary_ctls_high);
2494 case MSR_IA32_VMX_EPT_VPID_CAP:
2495 /* Currently, no nested vpid support */
2496 *pdata = nested_vmx_ept_caps;
2506 * Reads an msr value (of 'msr_index') into 'pdata'.
2507 * Returns 0 on success, non-0 otherwise.
2508 * Assumes vcpu_load() was already called.
2510 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2513 struct shared_msr_entry *msr;
2516 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
2520 switch (msr_index) {
2521 #ifdef CONFIG_X86_64
2523 data = vmcs_readl(GUEST_FS_BASE);
2526 data = vmcs_readl(GUEST_GS_BASE);
2528 case MSR_KERNEL_GS_BASE:
2529 vmx_load_host_state(to_vmx(vcpu));
2530 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
2534 return kvm_get_msr_common(vcpu, msr_index, pdata);
2536 data = guest_read_tsc();
2538 case MSR_IA32_SYSENTER_CS:
2539 data = vmcs_read32(GUEST_SYSENTER_CS);
2541 case MSR_IA32_SYSENTER_EIP:
2542 data = vmcs_readl(GUEST_SYSENTER_EIP);
2544 case MSR_IA32_SYSENTER_ESP:
2545 data = vmcs_readl(GUEST_SYSENTER_ESP);
2547 case MSR_IA32_BNDCFGS:
2548 if (!vmx_mpx_supported())
2550 data = vmcs_read64(GUEST_BNDCFGS);
2552 case MSR_IA32_FEATURE_CONTROL:
2553 if (!nested_vmx_allowed(vcpu))
2555 data = to_vmx(vcpu)->nested.msr_ia32_feature_control;
2557 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2558 if (!nested_vmx_allowed(vcpu))
2560 return vmx_get_vmx_msr(vcpu, msr_index, pdata);
2562 if (!to_vmx(vcpu)->rdtscp_enabled)
2564 /* Otherwise falls through */
2566 msr = find_msr_entry(to_vmx(vcpu), msr_index);
2571 return kvm_get_msr_common(vcpu, msr_index, pdata);
2578 static void vmx_leave_nested(struct kvm_vcpu *vcpu);
2581 * Writes msr value into into the appropriate "register".
2582 * Returns 0 on success, non-0 otherwise.
2583 * Assumes vcpu_load() was already called.
2585 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2587 struct vcpu_vmx *vmx = to_vmx(vcpu);
2588 struct shared_msr_entry *msr;
2590 u32 msr_index = msr_info->index;
2591 u64 data = msr_info->data;
2593 switch (msr_index) {
2595 ret = kvm_set_msr_common(vcpu, msr_info);
2597 #ifdef CONFIG_X86_64
2599 vmx_segment_cache_clear(vmx);
2600 vmcs_writel(GUEST_FS_BASE, data);
2603 vmx_segment_cache_clear(vmx);
2604 vmcs_writel(GUEST_GS_BASE, data);
2606 case MSR_KERNEL_GS_BASE:
2607 vmx_load_host_state(vmx);
2608 vmx->msr_guest_kernel_gs_base = data;
2611 case MSR_IA32_SYSENTER_CS:
2612 vmcs_write32(GUEST_SYSENTER_CS, data);
2614 case MSR_IA32_SYSENTER_EIP:
2615 vmcs_writel(GUEST_SYSENTER_EIP, data);
2617 case MSR_IA32_SYSENTER_ESP:
2618 vmcs_writel(GUEST_SYSENTER_ESP, data);
2620 case MSR_IA32_BNDCFGS:
2621 if (!vmx_mpx_supported())
2623 vmcs_write64(GUEST_BNDCFGS, data);
2626 kvm_write_tsc(vcpu, msr_info);
2628 case MSR_IA32_CR_PAT:
2629 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2630 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
2632 vmcs_write64(GUEST_IA32_PAT, data);
2633 vcpu->arch.pat = data;
2636 ret = kvm_set_msr_common(vcpu, msr_info);
2638 case MSR_IA32_TSC_ADJUST:
2639 ret = kvm_set_msr_common(vcpu, msr_info);
2641 case MSR_IA32_FEATURE_CONTROL:
2642 if (!nested_vmx_allowed(vcpu) ||
2643 (to_vmx(vcpu)->nested.msr_ia32_feature_control &
2644 FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
2646 vmx->nested.msr_ia32_feature_control = data;
2647 if (msr_info->host_initiated && data == 0)
2648 vmx_leave_nested(vcpu);
2650 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2651 return 1; /* they are read-only */
2653 if (!vmx->rdtscp_enabled)
2655 /* Check reserved bit, higher 32 bits should be zero */
2656 if ((data >> 32) != 0)
2658 /* Otherwise falls through */
2660 msr = find_msr_entry(vmx, msr_index);
2662 u64 old_msr_data = msr->data;
2664 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
2666 ret = kvm_set_shared_msr(msr->index, msr->data,
2670 msr->data = old_msr_data;
2674 ret = kvm_set_msr_common(vcpu, msr_info);
2680 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2682 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
2685 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2688 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2690 case VCPU_EXREG_PDPTR:
2692 ept_save_pdptrs(vcpu);
2699 static __init int cpu_has_kvm_support(void)
2701 return cpu_has_vmx();
2704 static __init int vmx_disabled_by_bios(void)
2708 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
2709 if (msr & FEATURE_CONTROL_LOCKED) {
2710 /* launched w/ TXT and VMX disabled */
2711 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2714 /* launched w/o TXT and VMX only enabled w/ TXT */
2715 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2716 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2717 && !tboot_enabled()) {
2718 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
2719 "activate TXT before enabling KVM\n");
2722 /* launched w/o TXT and VMX disabled */
2723 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2724 && !tboot_enabled())
2731 static void kvm_cpu_vmxon(u64 addr)
2733 asm volatile (ASM_VMX_VMXON_RAX
2734 : : "a"(&addr), "m"(addr)
2738 static int hardware_enable(void)
2740 int cpu = raw_smp_processor_id();
2741 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2744 if (read_cr4() & X86_CR4_VMXE)
2747 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2750 * Now we can enable the vmclear operation in kdump
2751 * since the loaded_vmcss_on_cpu list on this cpu
2752 * has been initialized.
2754 * Though the cpu is not in VMX operation now, there
2755 * is no problem to enable the vmclear operation
2756 * for the loaded_vmcss_on_cpu list is empty!
2758 crash_enable_local_vmclear(cpu);
2760 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
2762 test_bits = FEATURE_CONTROL_LOCKED;
2763 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
2764 if (tboot_enabled())
2765 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
2767 if ((old & test_bits) != test_bits) {
2768 /* enable and lock */
2769 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
2771 write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
2773 if (vmm_exclusive) {
2774 kvm_cpu_vmxon(phys_addr);
2778 native_store_gdt(this_cpu_ptr(&host_gdt));
2783 static void vmclear_local_loaded_vmcss(void)
2785 int cpu = raw_smp_processor_id();
2786 struct loaded_vmcs *v, *n;
2788 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2789 loaded_vmcss_on_cpu_link)
2790 __loaded_vmcs_clear(v);
2794 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2797 static void kvm_cpu_vmxoff(void)
2799 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
2802 static void hardware_disable(void)
2804 if (vmm_exclusive) {
2805 vmclear_local_loaded_vmcss();
2808 write_cr4(read_cr4() & ~X86_CR4_VMXE);
2811 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2812 u32 msr, u32 *result)
2814 u32 vmx_msr_low, vmx_msr_high;
2815 u32 ctl = ctl_min | ctl_opt;
2817 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2819 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2820 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
2822 /* Ensure minimum (required) set of control bits are supported. */
2830 static __init bool allow_1_setting(u32 msr, u32 ctl)
2832 u32 vmx_msr_low, vmx_msr_high;
2834 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2835 return vmx_msr_high & ctl;
2838 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
2840 u32 vmx_msr_low, vmx_msr_high;
2841 u32 min, opt, min2, opt2;
2842 u32 _pin_based_exec_control = 0;
2843 u32 _cpu_based_exec_control = 0;
2844 u32 _cpu_based_2nd_exec_control = 0;
2845 u32 _vmexit_control = 0;
2846 u32 _vmentry_control = 0;
2848 min = CPU_BASED_HLT_EXITING |
2849 #ifdef CONFIG_X86_64
2850 CPU_BASED_CR8_LOAD_EXITING |
2851 CPU_BASED_CR8_STORE_EXITING |
2853 CPU_BASED_CR3_LOAD_EXITING |
2854 CPU_BASED_CR3_STORE_EXITING |
2855 CPU_BASED_USE_IO_BITMAPS |
2856 CPU_BASED_MOV_DR_EXITING |
2857 CPU_BASED_USE_TSC_OFFSETING |
2858 CPU_BASED_MWAIT_EXITING |
2859 CPU_BASED_MONITOR_EXITING |
2860 CPU_BASED_INVLPG_EXITING |
2861 CPU_BASED_RDPMC_EXITING;
2863 opt = CPU_BASED_TPR_SHADOW |
2864 CPU_BASED_USE_MSR_BITMAPS |
2865 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2866 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2867 &_cpu_based_exec_control) < 0)
2869 #ifdef CONFIG_X86_64
2870 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2871 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2872 ~CPU_BASED_CR8_STORE_EXITING;
2874 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2876 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2877 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2878 SECONDARY_EXEC_WBINVD_EXITING |
2879 SECONDARY_EXEC_ENABLE_VPID |
2880 SECONDARY_EXEC_ENABLE_EPT |
2881 SECONDARY_EXEC_UNRESTRICTED_GUEST |
2882 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2883 SECONDARY_EXEC_RDTSCP |
2884 SECONDARY_EXEC_ENABLE_INVPCID |
2885 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2886 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2887 SECONDARY_EXEC_SHADOW_VMCS;
2888 if (adjust_vmx_controls(min2, opt2,
2889 MSR_IA32_VMX_PROCBASED_CTLS2,
2890 &_cpu_based_2nd_exec_control) < 0)
2893 #ifndef CONFIG_X86_64
2894 if (!(_cpu_based_2nd_exec_control &
2895 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2896 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2899 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2900 _cpu_based_2nd_exec_control &= ~(
2901 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2902 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2903 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2905 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2906 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2908 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2909 CPU_BASED_CR3_STORE_EXITING |
2910 CPU_BASED_INVLPG_EXITING);
2911 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
2912 vmx_capability.ept, vmx_capability.vpid);
2915 min = VM_EXIT_SAVE_DEBUG_CONTROLS;
2916 #ifdef CONFIG_X86_64
2917 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2919 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT |
2920 VM_EXIT_ACK_INTR_ON_EXIT | VM_EXIT_CLEAR_BNDCFGS;
2921 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2922 &_vmexit_control) < 0)
2925 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2926 opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR;
2927 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2928 &_pin_based_exec_control) < 0)
2931 if (!(_cpu_based_2nd_exec_control &
2932 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) ||
2933 !(_vmexit_control & VM_EXIT_ACK_INTR_ON_EXIT))
2934 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2936 min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
2937 opt = VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS;
2938 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2939 &_vmentry_control) < 0)
2942 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2944 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2945 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2948 #ifdef CONFIG_X86_64
2949 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2950 if (vmx_msr_high & (1u<<16))
2954 /* Require Write-Back (WB) memory type for VMCS accesses. */
2955 if (((vmx_msr_high >> 18) & 15) != 6)
2958 vmcs_conf->size = vmx_msr_high & 0x1fff;
2959 vmcs_conf->order = get_order(vmcs_config.size);
2960 vmcs_conf->revision_id = vmx_msr_low;
2962 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2963 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2964 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2965 vmcs_conf->vmexit_ctrl = _vmexit_control;
2966 vmcs_conf->vmentry_ctrl = _vmentry_control;
2968 cpu_has_load_ia32_efer =
2969 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2970 VM_ENTRY_LOAD_IA32_EFER)
2971 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2972 VM_EXIT_LOAD_IA32_EFER);
2974 cpu_has_load_perf_global_ctrl =
2975 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2976 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
2977 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2978 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
2981 * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
2982 * but due to arrata below it can't be used. Workaround is to use
2983 * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2985 * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
2990 * BC86,AAY89,BD102 (model 44)
2994 if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
2995 switch (boot_cpu_data.x86_model) {
3001 cpu_has_load_perf_global_ctrl = false;
3002 printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
3003 "does not work properly. Using workaround\n");
3013 static struct vmcs *alloc_vmcs_cpu(int cpu)
3015 int node = cpu_to_node(cpu);
3019 pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
3022 vmcs = page_address(pages);
3023 memset(vmcs, 0, vmcs_config.size);
3024 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
3028 static struct vmcs *alloc_vmcs(void)
3030 return alloc_vmcs_cpu(raw_smp_processor_id());
3033 static void free_vmcs(struct vmcs *vmcs)
3035 free_pages((unsigned long)vmcs, vmcs_config.order);
3039 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
3041 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
3043 if (!loaded_vmcs->vmcs)
3045 loaded_vmcs_clear(loaded_vmcs);
3046 free_vmcs(loaded_vmcs->vmcs);
3047 loaded_vmcs->vmcs = NULL;
3050 static void free_kvm_area(void)
3054 for_each_possible_cpu(cpu) {
3055 free_vmcs(per_cpu(vmxarea, cpu));
3056 per_cpu(vmxarea, cpu) = NULL;
3060 static void init_vmcs_shadow_fields(void)
3064 /* No checks for read only fields yet */
3066 for (i = j = 0; i < max_shadow_read_write_fields; i++) {
3067 switch (shadow_read_write_fields[i]) {
3069 if (!vmx_mpx_supported())
3077 shadow_read_write_fields[j] =
3078 shadow_read_write_fields[i];
3081 max_shadow_read_write_fields = j;
3083 /* shadowed fields guest access without vmexit */
3084 for (i = 0; i < max_shadow_read_write_fields; i++) {
3085 clear_bit(shadow_read_write_fields[i],
3086 vmx_vmwrite_bitmap);
3087 clear_bit(shadow_read_write_fields[i],
3090 for (i = 0; i < max_shadow_read_only_fields; i++)
3091 clear_bit(shadow_read_only_fields[i],
3095 static __init int alloc_kvm_area(void)
3099 for_each_possible_cpu(cpu) {
3102 vmcs = alloc_vmcs_cpu(cpu);
3108 per_cpu(vmxarea, cpu) = vmcs;
3113 static bool emulation_required(struct kvm_vcpu *vcpu)
3115 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
3118 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
3119 struct kvm_segment *save)
3121 if (!emulate_invalid_guest_state) {
3123 * CS and SS RPL should be equal during guest entry according
3124 * to VMX spec, but in reality it is not always so. Since vcpu
3125 * is in the middle of the transition from real mode to
3126 * protected mode it is safe to assume that RPL 0 is a good
3129 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
3130 save->selector &= ~SELECTOR_RPL_MASK;
3131 save->dpl = save->selector & SELECTOR_RPL_MASK;
3134 vmx_set_segment(vcpu, save, seg);
3137 static void enter_pmode(struct kvm_vcpu *vcpu)
3139 unsigned long flags;
3140 struct vcpu_vmx *vmx = to_vmx(vcpu);
3143 * Update real mode segment cache. It may be not up-to-date if sement
3144 * register was written while vcpu was in a guest mode.
3146 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3147 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3148 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3149 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3150 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3151 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3153 vmx->rmode.vm86_active = 0;
3155 vmx_segment_cache_clear(vmx);
3157 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3159 flags = vmcs_readl(GUEST_RFLAGS);
3160 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
3161 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
3162 vmcs_writel(GUEST_RFLAGS, flags);
3164 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
3165 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
3167 update_exception_bitmap(vcpu);
3169 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3170 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3171 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3172 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3173 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3174 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3177 static void fix_rmode_seg(int seg, struct kvm_segment *save)
3179 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3180 struct kvm_segment var = *save;
3183 if (seg == VCPU_SREG_CS)
3186 if (!emulate_invalid_guest_state) {
3187 var.selector = var.base >> 4;
3188 var.base = var.base & 0xffff0;
3198 if (save->base & 0xf)
3199 printk_once(KERN_WARNING "kvm: segment base is not "
3200 "paragraph aligned when entering "
3201 "protected mode (seg=%d)", seg);
3204 vmcs_write16(sf->selector, var.selector);
3205 vmcs_write32(sf->base, var.base);
3206 vmcs_write32(sf->limit, var.limit);
3207 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
3210 static void enter_rmode(struct kvm_vcpu *vcpu)
3212 unsigned long flags;
3213 struct vcpu_vmx *vmx = to_vmx(vcpu);
3215 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3216 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3217 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3218 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3219 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3220 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3221 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3223 vmx->rmode.vm86_active = 1;
3226 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
3227 * vcpu. Warn the user that an update is overdue.
3229 if (!vcpu->kvm->arch.tss_addr)
3230 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
3231 "called before entering vcpu\n");
3233 vmx_segment_cache_clear(vmx);
3235 vmcs_writel(GUEST_TR_BASE, vcpu->kvm->arch.tss_addr);
3236 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
3237 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3239 flags = vmcs_readl(GUEST_RFLAGS);
3240 vmx->rmode.save_rflags = flags;
3242 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
3244 vmcs_writel(GUEST_RFLAGS, flags);
3245 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
3246 update_exception_bitmap(vcpu);
3248 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3249 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3250 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3251 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3252 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3253 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3255 kvm_mmu_reset_context(vcpu);
3258 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
3260 struct vcpu_vmx *vmx = to_vmx(vcpu);
3261 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
3267 * Force kernel_gs_base reloading before EFER changes, as control
3268 * of this msr depends on is_long_mode().
3270 vmx_load_host_state(to_vmx(vcpu));
3271 vcpu->arch.efer = efer;
3272 if (efer & EFER_LMA) {
3273 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3276 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3278 msr->data = efer & ~EFER_LME;
3283 #ifdef CONFIG_X86_64
3285 static void enter_lmode(struct kvm_vcpu *vcpu)
3289 vmx_segment_cache_clear(to_vmx(vcpu));
3291 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
3292 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
3293 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
3295 vmcs_write32(GUEST_TR_AR_BYTES,
3296 (guest_tr_ar & ~AR_TYPE_MASK)
3297 | AR_TYPE_BUSY_64_TSS);
3299 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
3302 static void exit_lmode(struct kvm_vcpu *vcpu)
3304 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3305 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
3310 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
3312 vpid_sync_context(to_vmx(vcpu));
3314 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
3316 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
3320 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
3322 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
3324 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
3325 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
3328 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
3330 if (enable_ept && is_paging(vcpu))
3331 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3332 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
3335 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
3337 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
3339 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
3340 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
3343 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
3345 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3347 if (!test_bit(VCPU_EXREG_PDPTR,
3348 (unsigned long *)&vcpu->arch.regs_dirty))
3351 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3352 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
3353 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
3354 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
3355 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
3359 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
3361 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3363 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3364 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
3365 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
3366 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
3367 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
3370 __set_bit(VCPU_EXREG_PDPTR,
3371 (unsigned long *)&vcpu->arch.regs_avail);
3372 __set_bit(VCPU_EXREG_PDPTR,
3373 (unsigned long *)&vcpu->arch.regs_dirty);
3376 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
3378 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
3380 struct kvm_vcpu *vcpu)
3382 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3383 vmx_decache_cr3(vcpu);
3384 if (!(cr0 & X86_CR0_PG)) {
3385 /* From paging/starting to nonpaging */
3386 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3387 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
3388 (CPU_BASED_CR3_LOAD_EXITING |
3389 CPU_BASED_CR3_STORE_EXITING));
3390 vcpu->arch.cr0 = cr0;
3391 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3392 } else if (!is_paging(vcpu)) {
3393 /* From nonpaging to paging */
3394 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3395 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
3396 ~(CPU_BASED_CR3_LOAD_EXITING |
3397 CPU_BASED_CR3_STORE_EXITING));
3398 vcpu->arch.cr0 = cr0;
3399 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3402 if (!(cr0 & X86_CR0_WP))
3403 *hw_cr0 &= ~X86_CR0_WP;
3406 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
3408 struct vcpu_vmx *vmx = to_vmx(vcpu);
3409 unsigned long hw_cr0;
3411 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK);
3412 if (enable_unrestricted_guest)
3413 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
3415 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
3417 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
3420 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
3424 #ifdef CONFIG_X86_64
3425 if (vcpu->arch.efer & EFER_LME) {
3426 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
3428 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
3434 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
3436 if (!vcpu->fpu_active)
3437 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
3439 vmcs_writel(CR0_READ_SHADOW, cr0);
3440 vmcs_writel(GUEST_CR0, hw_cr0);
3441 vcpu->arch.cr0 = cr0;
3443 /* depends on vcpu->arch.cr0 to be set to a new value */
3444 vmx->emulation_required = emulation_required(vcpu);
3447 static u64 construct_eptp(unsigned long root_hpa)
3451 /* TODO write the value reading from MSR */
3452 eptp = VMX_EPT_DEFAULT_MT |
3453 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
3454 if (enable_ept_ad_bits)
3455 eptp |= VMX_EPT_AD_ENABLE_BIT;
3456 eptp |= (root_hpa & PAGE_MASK);
3461 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
3463 unsigned long guest_cr3;
3468 eptp = construct_eptp(cr3);
3469 vmcs_write64(EPT_POINTER, eptp);
3470 if (is_paging(vcpu) || is_guest_mode(vcpu))
3471 guest_cr3 = kvm_read_cr3(vcpu);
3473 guest_cr3 = vcpu->kvm->arch.ept_identity_map_addr;
3474 ept_load_pdptrs(vcpu);
3477 vmx_flush_tlb(vcpu);
3478 vmcs_writel(GUEST_CR3, guest_cr3);
3481 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3483 unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
3484 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
3486 if (cr4 & X86_CR4_VMXE) {
3488 * To use VMXON (and later other VMX instructions), a guest
3489 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3490 * So basically the check on whether to allow nested VMX
3493 if (!nested_vmx_allowed(vcpu))
3496 if (to_vmx(vcpu)->nested.vmxon &&
3497 ((cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON))
3500 vcpu->arch.cr4 = cr4;
3502 if (!is_paging(vcpu)) {
3503 hw_cr4 &= ~X86_CR4_PAE;
3504 hw_cr4 |= X86_CR4_PSE;
3506 * SMEP/SMAP is disabled if CPU is in non-paging mode
3507 * in hardware. However KVM always uses paging mode to
3508 * emulate guest non-paging mode with TDP.
3509 * To emulate this behavior, SMEP/SMAP needs to be
3510 * manually disabled when guest switches to non-paging
3513 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP);
3514 } else if (!(cr4 & X86_CR4_PAE)) {
3515 hw_cr4 &= ~X86_CR4_PAE;
3519 vmcs_writel(CR4_READ_SHADOW, cr4);
3520 vmcs_writel(GUEST_CR4, hw_cr4);
3524 static void vmx_get_segment(struct kvm_vcpu *vcpu,
3525 struct kvm_segment *var, int seg)
3527 struct vcpu_vmx *vmx = to_vmx(vcpu);
3530 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3531 *var = vmx->rmode.segs[seg];
3532 if (seg == VCPU_SREG_TR
3533 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3535 var->base = vmx_read_guest_seg_base(vmx, seg);
3536 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3539 var->base = vmx_read_guest_seg_base(vmx, seg);
3540 var->limit = vmx_read_guest_seg_limit(vmx, seg);
3541 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3542 ar = vmx_read_guest_seg_ar(vmx, seg);
3543 var->unusable = (ar >> 16) & 1;
3544 var->type = ar & 15;
3545 var->s = (ar >> 4) & 1;
3546 var->dpl = (ar >> 5) & 3;
3548 * Some userspaces do not preserve unusable property. Since usable
3549 * segment has to be present according to VMX spec we can use present
3550 * property to amend userspace bug by making unusable segment always
3551 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3552 * segment as unusable.
3554 var->present = !var->unusable;
3555 var->avl = (ar >> 12) & 1;
3556 var->l = (ar >> 13) & 1;
3557 var->db = (ar >> 14) & 1;
3558 var->g = (ar >> 15) & 1;
3561 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3563 struct kvm_segment s;
3565 if (to_vmx(vcpu)->rmode.vm86_active) {
3566 vmx_get_segment(vcpu, &s, seg);
3569 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3572 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
3574 struct vcpu_vmx *vmx = to_vmx(vcpu);
3576 if (unlikely(vmx->rmode.vm86_active))
3579 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3584 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3588 if (var->unusable || !var->present)
3591 ar = var->type & 15;
3592 ar |= (var->s & 1) << 4;
3593 ar |= (var->dpl & 3) << 5;
3594 ar |= (var->present & 1) << 7;
3595 ar |= (var->avl & 1) << 12;
3596 ar |= (var->l & 1) << 13;
3597 ar |= (var->db & 1) << 14;
3598 ar |= (var->g & 1) << 15;
3604 static void vmx_set_segment(struct kvm_vcpu *vcpu,
3605 struct kvm_segment *var, int seg)
3607 struct vcpu_vmx *vmx = to_vmx(vcpu);
3608 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3610 vmx_segment_cache_clear(vmx);
3612 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3613 vmx->rmode.segs[seg] = *var;
3614 if (seg == VCPU_SREG_TR)
3615 vmcs_write16(sf->selector, var->selector);
3617 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3621 vmcs_writel(sf->base, var->base);
3622 vmcs_write32(sf->limit, var->limit);
3623 vmcs_write16(sf->selector, var->selector);
3626 * Fix the "Accessed" bit in AR field of segment registers for older
3628 * IA32 arch specifies that at the time of processor reset the
3629 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3630 * is setting it to 0 in the userland code. This causes invalid guest
3631 * state vmexit when "unrestricted guest" mode is turned on.
3632 * Fix for this setup issue in cpu_reset is being pushed in the qemu
3633 * tree. Newer qemu binaries with that qemu fix would not need this
3636 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3637 var->type |= 0x1; /* Accessed */
3639 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3642 vmx->emulation_required = emulation_required(vcpu);
3645 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3647 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3649 *db = (ar >> 14) & 1;
3650 *l = (ar >> 13) & 1;
3653 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3655 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3656 dt->address = vmcs_readl(GUEST_IDTR_BASE);
3659 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3661 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3662 vmcs_writel(GUEST_IDTR_BASE, dt->address);
3665 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3667 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3668 dt->address = vmcs_readl(GUEST_GDTR_BASE);
3671 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3673 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3674 vmcs_writel(GUEST_GDTR_BASE, dt->address);
3677 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3679 struct kvm_segment var;
3682 vmx_get_segment(vcpu, &var, seg);
3684 if (seg == VCPU_SREG_CS)
3686 ar = vmx_segment_access_rights(&var);
3688 if (var.base != (var.selector << 4))
3690 if (var.limit != 0xffff)
3698 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3700 struct kvm_segment cs;
3701 unsigned int cs_rpl;
3703 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3704 cs_rpl = cs.selector & SELECTOR_RPL_MASK;
3708 if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
3712 if (cs.type & AR_TYPE_WRITEABLE_MASK) {
3713 if (cs.dpl > cs_rpl)
3716 if (cs.dpl != cs_rpl)
3722 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3726 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3728 struct kvm_segment ss;
3729 unsigned int ss_rpl;
3731 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3732 ss_rpl = ss.selector & SELECTOR_RPL_MASK;
3736 if (ss.type != 3 && ss.type != 7)
3740 if (ss.dpl != ss_rpl) /* DPL != RPL */
3748 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3750 struct kvm_segment var;
3753 vmx_get_segment(vcpu, &var, seg);
3754 rpl = var.selector & SELECTOR_RPL_MASK;
3762 if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
3763 if (var.dpl < rpl) /* DPL < RPL */
3767 /* TODO: Add other members to kvm_segment_field to allow checking for other access
3773 static bool tr_valid(struct kvm_vcpu *vcpu)
3775 struct kvm_segment tr;
3777 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3781 if (tr.selector & SELECTOR_TI_MASK) /* TI = 1 */
3783 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3791 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3793 struct kvm_segment ldtr;
3795 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3799 if (ldtr.selector & SELECTOR_TI_MASK) /* TI = 1 */
3809 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3811 struct kvm_segment cs, ss;
3813 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3814 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3816 return ((cs.selector & SELECTOR_RPL_MASK) ==
3817 (ss.selector & SELECTOR_RPL_MASK));
3821 * Check if guest state is valid. Returns true if valid, false if
3823 * We assume that registers are always usable
3825 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3827 if (enable_unrestricted_guest)
3830 /* real mode guest state checks */
3831 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3832 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3834 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3836 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3838 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3840 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3842 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3845 /* protected mode guest state checks */
3846 if (!cs_ss_rpl_check(vcpu))
3848 if (!code_segment_valid(vcpu))
3850 if (!stack_segment_valid(vcpu))
3852 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3854 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3856 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3858 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3860 if (!tr_valid(vcpu))
3862 if (!ldtr_valid(vcpu))
3866 * - Add checks on RIP
3867 * - Add checks on RFLAGS
3873 static int init_rmode_tss(struct kvm *kvm)
3879 idx = srcu_read_lock(&kvm->srcu);
3880 fn = kvm->arch.tss_addr >> PAGE_SHIFT;
3881 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3884 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3885 r = kvm_write_guest_page(kvm, fn++, &data,
3886 TSS_IOPB_BASE_OFFSET, sizeof(u16));
3889 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3892 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3896 r = kvm_write_guest_page(kvm, fn, &data,
3897 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3900 srcu_read_unlock(&kvm->srcu, idx);
3904 static int init_rmode_identity_map(struct kvm *kvm)
3907 pfn_t identity_map_pfn;
3913 /* Protect kvm->arch.ept_identity_pagetable_done. */
3914 mutex_lock(&kvm->slots_lock);
3916 if (likely(kvm->arch.ept_identity_pagetable_done))
3919 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
3921 r = alloc_identity_pagetable(kvm);
3925 idx = srcu_read_lock(&kvm->srcu);
3926 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3929 /* Set up identity-mapping pagetable for EPT in real mode */
3930 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3931 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3932 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3933 r = kvm_write_guest_page(kvm, identity_map_pfn,
3934 &tmp, i * sizeof(tmp), sizeof(tmp));
3938 kvm->arch.ept_identity_pagetable_done = true;
3941 srcu_read_unlock(&kvm->srcu, idx);
3944 mutex_unlock(&kvm->slots_lock);
3948 static void seg_setup(int seg)
3950 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3953 vmcs_write16(sf->selector, 0);
3954 vmcs_writel(sf->base, 0);
3955 vmcs_write32(sf->limit, 0xffff);
3957 if (seg == VCPU_SREG_CS)
3958 ar |= 0x08; /* code segment */
3960 vmcs_write32(sf->ar_bytes, ar);
3963 static int alloc_apic_access_page(struct kvm *kvm)
3966 struct kvm_userspace_memory_region kvm_userspace_mem;
3969 mutex_lock(&kvm->slots_lock);
3970 if (kvm->arch.apic_access_page_done)
3972 kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
3973 kvm_userspace_mem.flags = 0;
3974 kvm_userspace_mem.guest_phys_addr = APIC_DEFAULT_PHYS_BASE;
3975 kvm_userspace_mem.memory_size = PAGE_SIZE;
3976 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem);
3980 page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
3981 if (is_error_page(page)) {
3987 * Do not pin the page in memory, so that memory hot-unplug
3988 * is able to migrate it.
3991 kvm->arch.apic_access_page_done = true;
3993 mutex_unlock(&kvm->slots_lock);
3997 static int alloc_identity_pagetable(struct kvm *kvm)
3999 /* Called with kvm->slots_lock held. */
4001 struct kvm_userspace_memory_region kvm_userspace_mem;
4004 BUG_ON(kvm->arch.ept_identity_pagetable_done);
4006 kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
4007 kvm_userspace_mem.flags = 0;
4008 kvm_userspace_mem.guest_phys_addr =
4009 kvm->arch.ept_identity_map_addr;
4010 kvm_userspace_mem.memory_size = PAGE_SIZE;
4011 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem);
4016 static void allocate_vpid(struct vcpu_vmx *vmx)
4023 spin_lock(&vmx_vpid_lock);
4024 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
4025 if (vpid < VMX_NR_VPIDS) {
4027 __set_bit(vpid, vmx_vpid_bitmap);
4029 spin_unlock(&vmx_vpid_lock);
4032 static void free_vpid(struct vcpu_vmx *vmx)
4036 spin_lock(&vmx_vpid_lock);
4038 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
4039 spin_unlock(&vmx_vpid_lock);
4042 #define MSR_TYPE_R 1
4043 #define MSR_TYPE_W 2
4044 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
4047 int f = sizeof(unsigned long);
4049 if (!cpu_has_vmx_msr_bitmap())
4053 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4054 * have the write-low and read-high bitmap offsets the wrong way round.
4055 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4057 if (msr <= 0x1fff) {
4058 if (type & MSR_TYPE_R)
4060 __clear_bit(msr, msr_bitmap + 0x000 / f);
4062 if (type & MSR_TYPE_W)
4064 __clear_bit(msr, msr_bitmap + 0x800 / f);
4066 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4068 if (type & MSR_TYPE_R)
4070 __clear_bit(msr, msr_bitmap + 0x400 / f);
4072 if (type & MSR_TYPE_W)
4074 __clear_bit(msr, msr_bitmap + 0xc00 / f);
4079 static void __vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
4082 int f = sizeof(unsigned long);
4084 if (!cpu_has_vmx_msr_bitmap())
4088 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4089 * have the write-low and read-high bitmap offsets the wrong way round.
4090 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4092 if (msr <= 0x1fff) {
4093 if (type & MSR_TYPE_R)
4095 __set_bit(msr, msr_bitmap + 0x000 / f);
4097 if (type & MSR_TYPE_W)
4099 __set_bit(msr, msr_bitmap + 0x800 / f);
4101 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4103 if (type & MSR_TYPE_R)
4105 __set_bit(msr, msr_bitmap + 0x400 / f);
4107 if (type & MSR_TYPE_W)
4109 __set_bit(msr, msr_bitmap + 0xc00 / f);
4114 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
4117 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy,
4118 msr, MSR_TYPE_R | MSR_TYPE_W);
4119 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode,
4120 msr, MSR_TYPE_R | MSR_TYPE_W);
4123 static void vmx_enable_intercept_msr_read_x2apic(u32 msr)
4125 __vmx_enable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4127 __vmx_enable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4131 static void vmx_disable_intercept_msr_read_x2apic(u32 msr)
4133 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4135 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4139 static void vmx_disable_intercept_msr_write_x2apic(u32 msr)
4141 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4143 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4147 static int vmx_vm_has_apicv(struct kvm *kvm)
4149 return enable_apicv && irqchip_in_kernel(kvm);
4153 * Send interrupt to vcpu via posted interrupt way.
4154 * 1. If target vcpu is running(non-root mode), send posted interrupt
4155 * notification to vcpu and hardware will sync PIR to vIRR atomically.
4156 * 2. If target vcpu isn't running(root mode), kick it to pick up the
4157 * interrupt from PIR in next vmentry.
4159 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
4161 struct vcpu_vmx *vmx = to_vmx(vcpu);
4164 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
4167 r = pi_test_and_set_on(&vmx->pi_desc);
4168 kvm_make_request(KVM_REQ_EVENT, vcpu);
4170 if (!r && (vcpu->mode == IN_GUEST_MODE))
4171 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu),
4172 POSTED_INTR_VECTOR);
4175 kvm_vcpu_kick(vcpu);
4178 static void vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
4180 struct vcpu_vmx *vmx = to_vmx(vcpu);
4182 if (!pi_test_and_clear_on(&vmx->pi_desc))
4185 kvm_apic_update_irr(vcpu, vmx->pi_desc.pir);
4188 static void vmx_sync_pir_to_irr_dummy(struct kvm_vcpu *vcpu)
4194 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
4195 * will not change in the lifetime of the guest.
4196 * Note that host-state that does change is set elsewhere. E.g., host-state
4197 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
4199 static void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
4206 vmcs_writel(HOST_CR0, read_cr0() & ~X86_CR0_TS); /* 22.2.3 */
4207 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
4209 /* Save the most likely value for this task's CR4 in the VMCS. */
4211 vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
4212 vmx->host_state.vmcs_host_cr4 = cr4;
4214 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
4215 #ifdef CONFIG_X86_64
4217 * Load null selectors, so we can avoid reloading them in
4218 * __vmx_load_host_state(), in case userspace uses the null selectors
4219 * too (the expected case).
4221 vmcs_write16(HOST_DS_SELECTOR, 0);
4222 vmcs_write16(HOST_ES_SELECTOR, 0);
4224 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4225 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4227 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4228 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
4230 native_store_idt(&dt);
4231 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
4232 vmx->host_idt_base = dt.address;
4234 vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
4236 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
4237 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
4238 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
4239 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
4241 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
4242 rdmsr(MSR_IA32_CR_PAT, low32, high32);
4243 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
4247 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
4249 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
4251 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
4252 if (is_guest_mode(&vmx->vcpu))
4253 vmx->vcpu.arch.cr4_guest_owned_bits &=
4254 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
4255 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
4258 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
4260 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
4262 if (!vmx_vm_has_apicv(vmx->vcpu.kvm))
4263 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
4264 return pin_based_exec_ctrl;
4267 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
4269 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
4271 if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
4272 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
4274 if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
4275 exec_control &= ~CPU_BASED_TPR_SHADOW;
4276 #ifdef CONFIG_X86_64
4277 exec_control |= CPU_BASED_CR8_STORE_EXITING |
4278 CPU_BASED_CR8_LOAD_EXITING;
4282 exec_control |= CPU_BASED_CR3_STORE_EXITING |
4283 CPU_BASED_CR3_LOAD_EXITING |
4284 CPU_BASED_INVLPG_EXITING;
4285 return exec_control;
4288 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
4290 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4291 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
4292 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4294 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4296 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4297 enable_unrestricted_guest = 0;
4298 /* Enable INVPCID for non-ept guests may cause performance regression. */
4299 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4301 if (!enable_unrestricted_guest)
4302 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4304 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4305 if (!vmx_vm_has_apicv(vmx->vcpu.kvm))
4306 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4307 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4308 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4309 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4311 We can NOT enable shadow_vmcs here because we don't have yet
4314 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4315 return exec_control;
4318 static void ept_set_mmio_spte_mask(void)
4321 * EPT Misconfigurations can be generated if the value of bits 2:0
4322 * of an EPT paging-structure entry is 110b (write/execute).
4323 * Also, magic bits (0x3ull << 62) is set to quickly identify mmio
4326 kvm_mmu_set_mmio_spte_mask((0x3ull << 62) | 0x6ull);
4330 * Sets up the vmcs for emulated real mode.
4332 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
4334 #ifdef CONFIG_X86_64
4340 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
4341 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
4343 if (enable_shadow_vmcs) {
4344 vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
4345 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
4347 if (cpu_has_vmx_msr_bitmap())
4348 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
4350 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4353 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
4355 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
4357 if (cpu_has_secondary_exec_ctrls()) {
4358 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
4359 vmx_secondary_exec_control(vmx));
4362 if (vmx_vm_has_apicv(vmx->vcpu.kvm)) {
4363 vmcs_write64(EOI_EXIT_BITMAP0, 0);
4364 vmcs_write64(EOI_EXIT_BITMAP1, 0);
4365 vmcs_write64(EOI_EXIT_BITMAP2, 0);
4366 vmcs_write64(EOI_EXIT_BITMAP3, 0);
4368 vmcs_write16(GUEST_INTR_STATUS, 0);
4370 vmcs_write64(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4371 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4375 vmcs_write32(PLE_GAP, ple_gap);
4376 vmx->ple_window = ple_window;
4377 vmx->ple_window_dirty = true;
4380 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4381 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4382 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
4384 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
4385 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
4386 vmx_set_constant_host_state(vmx);
4387 #ifdef CONFIG_X86_64
4388 rdmsrl(MSR_FS_BASE, a);
4389 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
4390 rdmsrl(MSR_GS_BASE, a);
4391 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
4393 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4394 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4397 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4398 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4399 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
4400 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4401 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
4403 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
4404 u32 msr_low, msr_high;
4406 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
4407 host_pat = msr_low | ((u64) msr_high << 32);
4408 /* Write the default value follow host pat */
4409 vmcs_write64(GUEST_IA32_PAT, host_pat);
4410 /* Keep arch.pat sync with GUEST_IA32_PAT */
4411 vmx->vcpu.arch.pat = host_pat;
4414 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
4415 u32 index = vmx_msr_index[i];
4416 u32 data_low, data_high;
4419 if (rdmsr_safe(index, &data_low, &data_high) < 0)
4421 if (wrmsr_safe(index, data_low, data_high) < 0)
4423 vmx->guest_msrs[j].index = i;
4424 vmx->guest_msrs[j].data = 0;
4425 vmx->guest_msrs[j].mask = -1ull;
4430 vm_exit_controls_init(vmx, vmcs_config.vmexit_ctrl);
4432 /* 22.2.1, 20.8.1 */
4433 vm_entry_controls_init(vmx, vmcs_config.vmentry_ctrl);
4435 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
4436 set_cr4_guest_host_mask(vmx);
4441 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu)
4443 struct vcpu_vmx *vmx = to_vmx(vcpu);
4444 struct msr_data apic_base_msr;
4446 vmx->rmode.vm86_active = 0;
4448 vmx->soft_vnmi_blocked = 0;
4450 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4451 kvm_set_cr8(&vmx->vcpu, 0);
4452 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE | MSR_IA32_APICBASE_ENABLE;
4453 if (kvm_vcpu_is_bsp(&vmx->vcpu))
4454 apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4455 apic_base_msr.host_initiated = true;
4456 kvm_set_apic_base(&vmx->vcpu, &apic_base_msr);
4458 vmx_segment_cache_clear(vmx);
4460 seg_setup(VCPU_SREG_CS);
4461 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4462 vmcs_write32(GUEST_CS_BASE, 0xffff0000);
4464 seg_setup(VCPU_SREG_DS);
4465 seg_setup(VCPU_SREG_ES);
4466 seg_setup(VCPU_SREG_FS);
4467 seg_setup(VCPU_SREG_GS);
4468 seg_setup(VCPU_SREG_SS);
4470 vmcs_write16(GUEST_TR_SELECTOR, 0);
4471 vmcs_writel(GUEST_TR_BASE, 0);
4472 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4473 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4475 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4476 vmcs_writel(GUEST_LDTR_BASE, 0);
4477 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4478 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4480 vmcs_write32(GUEST_SYSENTER_CS, 0);
4481 vmcs_writel(GUEST_SYSENTER_ESP, 0);
4482 vmcs_writel(GUEST_SYSENTER_EIP, 0);
4484 vmcs_writel(GUEST_RFLAGS, 0x02);
4485 kvm_rip_write(vcpu, 0xfff0);
4487 vmcs_writel(GUEST_GDTR_BASE, 0);
4488 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4490 vmcs_writel(GUEST_IDTR_BASE, 0);
4491 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4493 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4494 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4495 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4497 /* Special registers */
4498 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4502 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
4504 if (cpu_has_vmx_tpr_shadow()) {
4505 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4506 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
4507 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4508 __pa(vmx->vcpu.arch.apic->regs));
4509 vmcs_write32(TPR_THRESHOLD, 0);
4512 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4514 if (vmx_vm_has_apicv(vcpu->kvm))
4515 memset(&vmx->pi_desc, 0, sizeof(struct pi_desc));
4518 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4520 vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4521 vmx_set_cr0(&vmx->vcpu, kvm_read_cr0(vcpu)); /* enter rmode */
4522 vmx_set_cr4(&vmx->vcpu, 0);
4523 vmx_set_efer(&vmx->vcpu, 0);
4524 vmx_fpu_activate(&vmx->vcpu);
4525 update_exception_bitmap(&vmx->vcpu);
4527 vpid_sync_context(vmx);
4531 * In nested virtualization, check if L1 asked to exit on external interrupts.
4532 * For most existing hypervisors, this will always return true.
4534 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
4536 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
4537 PIN_BASED_EXT_INTR_MASK;
4541 * In nested virtualization, check if L1 has set
4542 * VM_EXIT_ACK_INTR_ON_EXIT
4544 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
4546 return get_vmcs12(vcpu)->vm_exit_controls &
4547 VM_EXIT_ACK_INTR_ON_EXIT;
4550 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
4552 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
4553 PIN_BASED_NMI_EXITING;
4556 static void enable_irq_window(struct kvm_vcpu *vcpu)
4558 u32 cpu_based_vm_exec_control;
4560 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4561 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
4562 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4565 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4567 u32 cpu_based_vm_exec_control;
4569 if (!cpu_has_virtual_nmis() ||
4570 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4571 enable_irq_window(vcpu);
4575 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4576 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
4577 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4580 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4582 struct vcpu_vmx *vmx = to_vmx(vcpu);
4584 int irq = vcpu->arch.interrupt.nr;
4586 trace_kvm_inj_virq(irq);
4588 ++vcpu->stat.irq_injections;
4589 if (vmx->rmode.vm86_active) {
4591 if (vcpu->arch.interrupt.soft)
4592 inc_eip = vcpu->arch.event_exit_inst_len;
4593 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
4594 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4597 intr = irq | INTR_INFO_VALID_MASK;
4598 if (vcpu->arch.interrupt.soft) {
4599 intr |= INTR_TYPE_SOFT_INTR;
4600 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4601 vmx->vcpu.arch.event_exit_inst_len);
4603 intr |= INTR_TYPE_EXT_INTR;
4604 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4607 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4609 struct vcpu_vmx *vmx = to_vmx(vcpu);
4611 if (is_guest_mode(vcpu))
4614 if (!cpu_has_virtual_nmis()) {
4616 * Tracking the NMI-blocked state in software is built upon
4617 * finding the next open IRQ window. This, in turn, depends on
4618 * well-behaving guests: They have to keep IRQs disabled at
4619 * least as long as the NMI handler runs. Otherwise we may
4620 * cause NMI nesting, maybe breaking the guest. But as this is
4621 * highly unlikely, we can live with the residual risk.
4623 vmx->soft_vnmi_blocked = 1;
4624 vmx->vnmi_blocked_time = 0;
4627 ++vcpu->stat.nmi_injections;
4628 vmx->nmi_known_unmasked = false;
4629 if (vmx->rmode.vm86_active) {
4630 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
4631 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4634 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4635 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4638 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4640 if (!cpu_has_virtual_nmis())
4641 return to_vmx(vcpu)->soft_vnmi_blocked;
4642 if (to_vmx(vcpu)->nmi_known_unmasked)
4644 return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4647 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4649 struct vcpu_vmx *vmx = to_vmx(vcpu);
4651 if (!cpu_has_virtual_nmis()) {
4652 if (vmx->soft_vnmi_blocked != masked) {
4653 vmx->soft_vnmi_blocked = masked;
4654 vmx->vnmi_blocked_time = 0;
4657 vmx->nmi_known_unmasked = !masked;
4659 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4660 GUEST_INTR_STATE_NMI);
4662 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4663 GUEST_INTR_STATE_NMI);
4667 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4669 if (to_vmx(vcpu)->nested.nested_run_pending)
4672 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
4675 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4676 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4677 | GUEST_INTR_STATE_NMI));
4680 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4682 return (!to_vmx(vcpu)->nested.nested_run_pending &&
4683 vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4684 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4685 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4688 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4691 struct kvm_userspace_memory_region tss_mem = {
4692 .slot = TSS_PRIVATE_MEMSLOT,
4693 .guest_phys_addr = addr,
4694 .memory_size = PAGE_SIZE * 3,
4698 ret = kvm_set_memory_region(kvm, &tss_mem);
4701 kvm->arch.tss_addr = addr;
4702 return init_rmode_tss(kvm);
4705 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4710 * Update instruction length as we may reinject the exception
4711 * from user space while in guest debugging mode.
4713 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4714 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4715 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4719 if (vcpu->guest_debug &
4720 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4737 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4738 int vec, u32 err_code)
4741 * Instruction with address size override prefix opcode 0x67
4742 * Cause the #SS fault with 0 error code in VM86 mode.
4744 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4745 if (emulate_instruction(vcpu, 0) == EMULATE_DONE) {
4746 if (vcpu->arch.halt_request) {
4747 vcpu->arch.halt_request = 0;
4748 return kvm_emulate_halt(vcpu);
4756 * Forward all other exceptions that are valid in real mode.
4757 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4758 * the required debugging infrastructure rework.
4760 kvm_queue_exception(vcpu, vec);
4765 * Trigger machine check on the host. We assume all the MSRs are already set up
4766 * by the CPU and that we still run on the same CPU as the MCE occurred on.
4767 * We pass a fake environment to the machine check handler because we want
4768 * the guest to be always treated like user space, no matter what context
4769 * it used internally.
4771 static void kvm_machine_check(void)
4773 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4774 struct pt_regs regs = {
4775 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
4776 .flags = X86_EFLAGS_IF,
4779 do_machine_check(®s, 0);
4783 static int handle_machine_check(struct kvm_vcpu *vcpu)
4785 /* already handled by vcpu_run */
4789 static int handle_exception(struct kvm_vcpu *vcpu)
4791 struct vcpu_vmx *vmx = to_vmx(vcpu);
4792 struct kvm_run *kvm_run = vcpu->run;
4793 u32 intr_info, ex_no, error_code;
4794 unsigned long cr2, rip, dr6;
4796 enum emulation_result er;
4798 vect_info = vmx->idt_vectoring_info;
4799 intr_info = vmx->exit_intr_info;
4801 if (is_machine_check(intr_info))
4802 return handle_machine_check(vcpu);
4804 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
4805 return 1; /* already handled by vmx_vcpu_run() */
4807 if (is_no_device(intr_info)) {
4808 vmx_fpu_activate(vcpu);
4812 if (is_invalid_opcode(intr_info)) {
4813 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
4814 if (er != EMULATE_DONE)
4815 kvm_queue_exception(vcpu, UD_VECTOR);
4820 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4821 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4824 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4825 * MMIO, it is better to report an internal error.
4826 * See the comments in vmx_handle_exit.
4828 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4829 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4830 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4831 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4832 vcpu->run->internal.ndata = 2;
4833 vcpu->run->internal.data[0] = vect_info;
4834 vcpu->run->internal.data[1] = intr_info;
4838 if (is_page_fault(intr_info)) {
4839 /* EPT won't cause page fault directly */
4841 cr2 = vmcs_readl(EXIT_QUALIFICATION);
4842 trace_kvm_page_fault(cr2, error_code);
4844 if (kvm_event_needs_reinjection(vcpu))
4845 kvm_mmu_unprotect_page_virt(vcpu, cr2);
4846 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
4849 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4851 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4852 return handle_rmode_exception(vcpu, ex_no, error_code);
4856 dr6 = vmcs_readl(EXIT_QUALIFICATION);
4857 if (!(vcpu->guest_debug &
4858 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4859 vcpu->arch.dr6 &= ~15;
4860 vcpu->arch.dr6 |= dr6 | DR6_RTM;
4861 if (!(dr6 & ~DR6_RESERVED)) /* icebp */
4862 skip_emulated_instruction(vcpu);
4864 kvm_queue_exception(vcpu, DB_VECTOR);
4867 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4868 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4872 * Update instruction length as we may reinject #BP from
4873 * user space while in guest debugging mode. Reading it for
4874 * #DB as well causes no harm, it is not used in that case.
4876 vmx->vcpu.arch.event_exit_inst_len =
4877 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4878 kvm_run->exit_reason = KVM_EXIT_DEBUG;
4879 rip = kvm_rip_read(vcpu);
4880 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4881 kvm_run->debug.arch.exception = ex_no;
4884 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4885 kvm_run->ex.exception = ex_no;
4886 kvm_run->ex.error_code = error_code;
4892 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
4894 ++vcpu->stat.irq_exits;
4898 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4900 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4904 static int handle_io(struct kvm_vcpu *vcpu)
4906 unsigned long exit_qualification;
4907 int size, in, string;
4910 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4911 string = (exit_qualification & 16) != 0;
4912 in = (exit_qualification & 8) != 0;
4914 ++vcpu->stat.io_exits;
4917 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4919 port = exit_qualification >> 16;
4920 size = (exit_qualification & 7) + 1;
4921 skip_emulated_instruction(vcpu);
4923 return kvm_fast_pio_out(vcpu, size, port);
4927 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4930 * Patch in the VMCALL instruction:
4932 hypercall[0] = 0x0f;
4933 hypercall[1] = 0x01;
4934 hypercall[2] = 0xc1;
4937 static bool nested_cr0_valid(struct vmcs12 *vmcs12, unsigned long val)
4939 unsigned long always_on = VMXON_CR0_ALWAYSON;
4941 if (nested_vmx_secondary_ctls_high &
4942 SECONDARY_EXEC_UNRESTRICTED_GUEST &&
4943 nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
4944 always_on &= ~(X86_CR0_PE | X86_CR0_PG);
4945 return (val & always_on) == always_on;
4948 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4949 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4951 if (is_guest_mode(vcpu)) {
4952 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4953 unsigned long orig_val = val;
4956 * We get here when L2 changed cr0 in a way that did not change
4957 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4958 * but did change L0 shadowed bits. So we first calculate the
4959 * effective cr0 value that L1 would like to write into the
4960 * hardware. It consists of the L2-owned bits from the new
4961 * value combined with the L1-owned bits from L1's guest_cr0.
4963 val = (val & ~vmcs12->cr0_guest_host_mask) |
4964 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
4966 if (!nested_cr0_valid(vmcs12, val))
4969 if (kvm_set_cr0(vcpu, val))
4971 vmcs_writel(CR0_READ_SHADOW, orig_val);
4974 if (to_vmx(vcpu)->nested.vmxon &&
4975 ((val & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON))
4977 return kvm_set_cr0(vcpu, val);
4981 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4983 if (is_guest_mode(vcpu)) {
4984 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4985 unsigned long orig_val = val;
4987 /* analogously to handle_set_cr0 */
4988 val = (val & ~vmcs12->cr4_guest_host_mask) |
4989 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
4990 if (kvm_set_cr4(vcpu, val))
4992 vmcs_writel(CR4_READ_SHADOW, orig_val);
4995 return kvm_set_cr4(vcpu, val);
4998 /* called to set cr0 as approriate for clts instruction exit. */
4999 static void handle_clts(struct kvm_vcpu *vcpu)
5001 if (is_guest_mode(vcpu)) {
5003 * We get here when L2 did CLTS, and L1 didn't shadow CR0.TS
5004 * but we did (!fpu_active). We need to keep GUEST_CR0.TS on,
5005 * just pretend it's off (also in arch.cr0 for fpu_activate).
5007 vmcs_writel(CR0_READ_SHADOW,
5008 vmcs_readl(CR0_READ_SHADOW) & ~X86_CR0_TS);
5009 vcpu->arch.cr0 &= ~X86_CR0_TS;
5011 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
5014 static int handle_cr(struct kvm_vcpu *vcpu)
5016 unsigned long exit_qualification, val;
5021 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5022 cr = exit_qualification & 15;
5023 reg = (exit_qualification >> 8) & 15;
5024 switch ((exit_qualification >> 4) & 3) {
5025 case 0: /* mov to cr */
5026 val = kvm_register_readl(vcpu, reg);
5027 trace_kvm_cr_write(cr, val);
5030 err = handle_set_cr0(vcpu, val);
5031 kvm_complete_insn_gp(vcpu, err);
5034 err = kvm_set_cr3(vcpu, val);
5035 kvm_complete_insn_gp(vcpu, err);
5038 err = handle_set_cr4(vcpu, val);
5039 kvm_complete_insn_gp(vcpu, err);
5042 u8 cr8_prev = kvm_get_cr8(vcpu);
5044 err = kvm_set_cr8(vcpu, cr8);
5045 kvm_complete_insn_gp(vcpu, err);
5046 if (irqchip_in_kernel(vcpu->kvm))
5048 if (cr8_prev <= cr8)
5050 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
5057 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
5058 skip_emulated_instruction(vcpu);
5059 vmx_fpu_activate(vcpu);
5061 case 1: /*mov from cr*/
5064 val = kvm_read_cr3(vcpu);
5065 kvm_register_write(vcpu, reg, val);
5066 trace_kvm_cr_read(cr, val);
5067 skip_emulated_instruction(vcpu);
5070 val = kvm_get_cr8(vcpu);
5071 kvm_register_write(vcpu, reg, val);
5072 trace_kvm_cr_read(cr, val);
5073 skip_emulated_instruction(vcpu);
5078 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5079 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
5080 kvm_lmsw(vcpu, val);
5082 skip_emulated_instruction(vcpu);
5087 vcpu->run->exit_reason = 0;
5088 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
5089 (int)(exit_qualification >> 4) & 3, cr);
5093 static int handle_dr(struct kvm_vcpu *vcpu)
5095 unsigned long exit_qualification;
5098 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5099 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
5101 /* First, if DR does not exist, trigger UD */
5102 if (!kvm_require_dr(vcpu, dr))
5105 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
5106 if (!kvm_require_cpl(vcpu, 0))
5108 dr7 = vmcs_readl(GUEST_DR7);
5111 * As the vm-exit takes precedence over the debug trap, we
5112 * need to emulate the latter, either for the host or the
5113 * guest debugging itself.
5115 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5116 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
5117 vcpu->run->debug.arch.dr7 = dr7;
5118 vcpu->run->debug.arch.pc =
5119 vmcs_readl(GUEST_CS_BASE) +
5120 vmcs_readl(GUEST_RIP);
5121 vcpu->run->debug.arch.exception = DB_VECTOR;
5122 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
5125 vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
5126 kvm_queue_exception(vcpu, DB_VECTOR);
5131 if (vcpu->guest_debug == 0) {
5132 u32 cpu_based_vm_exec_control;
5134 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5135 cpu_based_vm_exec_control &= ~CPU_BASED_MOV_DR_EXITING;
5136 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5139 * No more DR vmexits; force a reload of the debug registers
5140 * and reenter on this instruction. The next vmexit will
5141 * retrieve the full state of the debug registers.
5143 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
5147 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
5148 if (exit_qualification & TYPE_MOV_FROM_DR) {
5151 if (kvm_get_dr(vcpu, dr, &val))
5153 kvm_register_write(vcpu, reg, val);
5155 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
5158 skip_emulated_instruction(vcpu);
5162 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
5164 return vcpu->arch.dr6;
5167 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
5171 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
5173 u32 cpu_based_vm_exec_control;
5175 get_debugreg(vcpu->arch.db[0], 0);
5176 get_debugreg(vcpu->arch.db[1], 1);
5177 get_debugreg(vcpu->arch.db[2], 2);
5178 get_debugreg(vcpu->arch.db[3], 3);
5179 get_debugreg(vcpu->arch.dr6, 6);
5180 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
5182 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
5184 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5185 cpu_based_vm_exec_control |= CPU_BASED_MOV_DR_EXITING;
5186 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5189 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
5191 vmcs_writel(GUEST_DR7, val);
5194 static int handle_cpuid(struct kvm_vcpu *vcpu)
5196 kvm_emulate_cpuid(vcpu);
5200 static int handle_rdmsr(struct kvm_vcpu *vcpu)
5202 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5205 if (vmx_get_msr(vcpu, ecx, &data)) {
5206 trace_kvm_msr_read_ex(ecx);
5207 kvm_inject_gp(vcpu, 0);
5211 trace_kvm_msr_read(ecx, data);
5213 /* FIXME: handling of bits 32:63 of rax, rdx */
5214 vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
5215 vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
5216 skip_emulated_instruction(vcpu);
5220 static int handle_wrmsr(struct kvm_vcpu *vcpu)
5222 struct msr_data msr;
5223 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5224 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
5225 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
5229 msr.host_initiated = false;
5230 if (kvm_set_msr(vcpu, &msr) != 0) {
5231 trace_kvm_msr_write_ex(ecx, data);
5232 kvm_inject_gp(vcpu, 0);
5236 trace_kvm_msr_write(ecx, data);
5237 skip_emulated_instruction(vcpu);
5241 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
5243 kvm_make_request(KVM_REQ_EVENT, vcpu);
5247 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
5249 u32 cpu_based_vm_exec_control;
5251 /* clear pending irq */
5252 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5253 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
5254 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5256 kvm_make_request(KVM_REQ_EVENT, vcpu);
5258 ++vcpu->stat.irq_window_exits;
5261 * If the user space waits to inject interrupts, exit as soon as
5264 if (!irqchip_in_kernel(vcpu->kvm) &&
5265 vcpu->run->request_interrupt_window &&
5266 !kvm_cpu_has_interrupt(vcpu)) {
5267 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
5273 static int handle_halt(struct kvm_vcpu *vcpu)
5275 skip_emulated_instruction(vcpu);
5276 return kvm_emulate_halt(vcpu);
5279 static int handle_vmcall(struct kvm_vcpu *vcpu)
5281 skip_emulated_instruction(vcpu);
5282 kvm_emulate_hypercall(vcpu);
5286 static int handle_invd(struct kvm_vcpu *vcpu)
5288 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5291 static int handle_invlpg(struct kvm_vcpu *vcpu)
5293 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5295 kvm_mmu_invlpg(vcpu, exit_qualification);
5296 skip_emulated_instruction(vcpu);
5300 static int handle_rdpmc(struct kvm_vcpu *vcpu)
5304 err = kvm_rdpmc(vcpu);
5305 kvm_complete_insn_gp(vcpu, err);
5310 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5312 skip_emulated_instruction(vcpu);
5313 kvm_emulate_wbinvd(vcpu);
5317 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5319 u64 new_bv = kvm_read_edx_eax(vcpu);
5320 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5322 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5323 skip_emulated_instruction(vcpu);
5327 static int handle_apic_access(struct kvm_vcpu *vcpu)
5329 if (likely(fasteoi)) {
5330 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5331 int access_type, offset;
5333 access_type = exit_qualification & APIC_ACCESS_TYPE;
5334 offset = exit_qualification & APIC_ACCESS_OFFSET;
5336 * Sane guest uses MOV to write EOI, with written value
5337 * not cared. So make a short-circuit here by avoiding
5338 * heavy instruction emulation.
5340 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5341 (offset == APIC_EOI)) {
5342 kvm_lapic_set_eoi(vcpu);
5343 skip_emulated_instruction(vcpu);
5347 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5350 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5352 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5353 int vector = exit_qualification & 0xff;
5355 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5356 kvm_apic_set_eoi_accelerated(vcpu, vector);
5360 static int handle_apic_write(struct kvm_vcpu *vcpu)
5362 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5363 u32 offset = exit_qualification & 0xfff;
5365 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
5366 kvm_apic_write_nodecode(vcpu, offset);
5370 static int handle_task_switch(struct kvm_vcpu *vcpu)
5372 struct vcpu_vmx *vmx = to_vmx(vcpu);
5373 unsigned long exit_qualification;
5374 bool has_error_code = false;
5377 int reason, type, idt_v, idt_index;
5379 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5380 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5381 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5383 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5385 reason = (u32)exit_qualification >> 30;
5386 if (reason == TASK_SWITCH_GATE && idt_v) {
5388 case INTR_TYPE_NMI_INTR:
5389 vcpu->arch.nmi_injected = false;
5390 vmx_set_nmi_mask(vcpu, true);
5392 case INTR_TYPE_EXT_INTR:
5393 case INTR_TYPE_SOFT_INTR:
5394 kvm_clear_interrupt_queue(vcpu);
5396 case INTR_TYPE_HARD_EXCEPTION:
5397 if (vmx->idt_vectoring_info &
5398 VECTORING_INFO_DELIVER_CODE_MASK) {
5399 has_error_code = true;
5401 vmcs_read32(IDT_VECTORING_ERROR_CODE);
5404 case INTR_TYPE_SOFT_EXCEPTION:
5405 kvm_clear_exception_queue(vcpu);
5411 tss_selector = exit_qualification;
5413 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5414 type != INTR_TYPE_EXT_INTR &&
5415 type != INTR_TYPE_NMI_INTR))
5416 skip_emulated_instruction(vcpu);
5418 if (kvm_task_switch(vcpu, tss_selector,
5419 type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
5420 has_error_code, error_code) == EMULATE_FAIL) {
5421 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5422 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5423 vcpu->run->internal.ndata = 0;
5427 /* clear all local breakpoint enable flags */
5428 vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~0x155);
5431 * TODO: What about debug traps on tss switch?
5432 * Are we supposed to inject them and update dr6?
5438 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5440 unsigned long exit_qualification;
5445 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5447 gla_validity = (exit_qualification >> 7) & 0x3;
5448 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
5449 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
5450 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
5451 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
5452 vmcs_readl(GUEST_LINEAR_ADDRESS));
5453 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
5454 (long unsigned int)exit_qualification);
5455 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5456 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
5461 * EPT violation happened while executing iret from NMI,
5462 * "blocked by NMI" bit has to be set before next VM entry.
5463 * There are errata that may cause this bit to not be set:
5466 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5467 cpu_has_virtual_nmis() &&
5468 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
5469 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5471 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5472 trace_kvm_page_fault(gpa, exit_qualification);
5474 /* It is a write fault? */
5475 error_code = exit_qualification & (1U << 1);
5476 /* It is a fetch fault? */
5477 error_code |= (exit_qualification & (1U << 2)) << 2;
5478 /* ept page table is present? */
5479 error_code |= (exit_qualification >> 3) & 0x1;
5481 vcpu->arch.exit_qualification = exit_qualification;
5483 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5486 static u64 ept_rsvd_mask(u64 spte, int level)
5491 for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
5492 mask |= (1ULL << i);
5495 /* bits 7:3 reserved */
5497 else if (spte & (1ULL << 7))
5499 * 1GB/2MB page, bits 29:12 or 20:12 reserved respectively,
5500 * level == 1 if the hypervisor is using the ignored bit 7.
5502 mask |= (PAGE_SIZE << ((level - 1) * 9)) - PAGE_SIZE;
5504 /* bits 6:3 reserved */
5510 static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
5513 printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
5515 /* 010b (write-only) */
5516 WARN_ON((spte & 0x7) == 0x2);
5518 /* 110b (write/execute) */
5519 WARN_ON((spte & 0x7) == 0x6);
5521 /* 100b (execute-only) and value not supported by logical processor */
5522 if (!cpu_has_vmx_ept_execute_only())
5523 WARN_ON((spte & 0x7) == 0x4);
5527 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
5529 if (rsvd_bits != 0) {
5530 printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
5531 __func__, rsvd_bits);
5535 /* bits 5:3 are _not_ reserved for large page or leaf page */
5536 if ((rsvd_bits & 0x38) == 0) {
5537 u64 ept_mem_type = (spte & 0x38) >> 3;
5539 if (ept_mem_type == 2 || ept_mem_type == 3 ||
5540 ept_mem_type == 7) {
5541 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
5542 __func__, ept_mem_type);
5549 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5552 int nr_sptes, i, ret;
5555 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5556 if (!kvm_io_bus_write(vcpu->kvm, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5557 skip_emulated_instruction(vcpu);
5561 ret = handle_mmio_page_fault_common(vcpu, gpa, true);
5562 if (likely(ret == RET_MMIO_PF_EMULATE))
5563 return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) ==
5566 if (unlikely(ret == RET_MMIO_PF_INVALID))
5567 return kvm_mmu_page_fault(vcpu, gpa, 0, NULL, 0);
5569 if (unlikely(ret == RET_MMIO_PF_RETRY))
5572 /* It is the real ept misconfig */
5573 printk(KERN_ERR "EPT: Misconfiguration.\n");
5574 printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
5576 nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
5578 for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
5579 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
5581 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5582 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
5587 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5589 u32 cpu_based_vm_exec_control;
5591 /* clear pending NMI */
5592 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5593 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
5594 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5595 ++vcpu->stat.nmi_window_exits;
5596 kvm_make_request(KVM_REQ_EVENT, vcpu);
5601 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5603 struct vcpu_vmx *vmx = to_vmx(vcpu);
5604 enum emulation_result err = EMULATE_DONE;
5607 bool intr_window_requested;
5608 unsigned count = 130;
5610 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5611 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
5613 while (vmx->emulation_required && count-- != 0) {
5614 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5615 return handle_interrupt_window(&vmx->vcpu);
5617 if (test_bit(KVM_REQ_EVENT, &vcpu->requests))
5620 err = emulate_instruction(vcpu, EMULTYPE_NO_REEXECUTE);
5622 if (err == EMULATE_USER_EXIT) {
5623 ++vcpu->stat.mmio_exits;
5628 if (err != EMULATE_DONE) {
5629 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5630 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5631 vcpu->run->internal.ndata = 0;
5635 if (vcpu->arch.halt_request) {
5636 vcpu->arch.halt_request = 0;
5637 ret = kvm_emulate_halt(vcpu);
5641 if (signal_pending(current))
5651 static int __grow_ple_window(int val)
5653 if (ple_window_grow < 1)
5656 val = min(val, ple_window_actual_max);
5658 if (ple_window_grow < ple_window)
5659 val *= ple_window_grow;
5661 val += ple_window_grow;
5666 static int __shrink_ple_window(int val, int modifier, int minimum)
5671 if (modifier < ple_window)
5676 return max(val, minimum);
5679 static void grow_ple_window(struct kvm_vcpu *vcpu)
5681 struct vcpu_vmx *vmx = to_vmx(vcpu);
5682 int old = vmx->ple_window;
5684 vmx->ple_window = __grow_ple_window(old);
5686 if (vmx->ple_window != old)
5687 vmx->ple_window_dirty = true;
5689 trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old);
5692 static void shrink_ple_window(struct kvm_vcpu *vcpu)
5694 struct vcpu_vmx *vmx = to_vmx(vcpu);
5695 int old = vmx->ple_window;
5697 vmx->ple_window = __shrink_ple_window(old,
5698 ple_window_shrink, ple_window);
5700 if (vmx->ple_window != old)
5701 vmx->ple_window_dirty = true;
5703 trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old);
5707 * ple_window_actual_max is computed to be one grow_ple_window() below
5708 * ple_window_max. (See __grow_ple_window for the reason.)
5709 * This prevents overflows, because ple_window_max is int.
5710 * ple_window_max effectively rounded down to a multiple of ple_window_grow in
5712 * ple_window_max is also prevented from setting vmx->ple_window < ple_window.
5714 static void update_ple_window_actual_max(void)
5716 ple_window_actual_max =
5717 __shrink_ple_window(max(ple_window_max, ple_window),
5718 ple_window_grow, INT_MIN);
5721 static __init int hardware_setup(void)
5723 if (setup_vmcs_config(&vmcs_config) < 0)
5726 if (boot_cpu_has(X86_FEATURE_NX))
5727 kvm_enable_efer_bits(EFER_NX);
5729 if (!cpu_has_vmx_vpid())
5731 if (!cpu_has_vmx_shadow_vmcs())
5732 enable_shadow_vmcs = 0;
5733 if (enable_shadow_vmcs)
5734 init_vmcs_shadow_fields();
5736 if (!cpu_has_vmx_ept() ||
5737 !cpu_has_vmx_ept_4levels()) {
5739 enable_unrestricted_guest = 0;
5740 enable_ept_ad_bits = 0;
5743 if (!cpu_has_vmx_ept_ad_bits())
5744 enable_ept_ad_bits = 0;
5746 if (!cpu_has_vmx_unrestricted_guest())
5747 enable_unrestricted_guest = 0;
5749 if (!cpu_has_vmx_flexpriority()) {
5750 flexpriority_enabled = 0;
5753 * set_apic_access_page_addr() is used to reload apic access
5754 * page upon invalidation. No need to do anything if the
5755 * processor does not have the APIC_ACCESS_ADDR VMCS field.
5757 kvm_x86_ops->set_apic_access_page_addr = NULL;
5760 if (!cpu_has_vmx_tpr_shadow())
5761 kvm_x86_ops->update_cr8_intercept = NULL;
5763 if (enable_ept && !cpu_has_vmx_ept_2m_page())
5764 kvm_disable_largepages();
5766 if (!cpu_has_vmx_ple())
5769 if (!cpu_has_vmx_apicv())
5773 kvm_x86_ops->update_cr8_intercept = NULL;
5775 kvm_x86_ops->hwapic_irr_update = NULL;
5776 kvm_x86_ops->deliver_posted_interrupt = NULL;
5777 kvm_x86_ops->sync_pir_to_irr = vmx_sync_pir_to_irr_dummy;
5781 nested_vmx_setup_ctls_msrs();
5783 return alloc_kvm_area();
5786 static __exit void hardware_unsetup(void)
5792 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5793 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5795 static int handle_pause(struct kvm_vcpu *vcpu)
5798 grow_ple_window(vcpu);
5800 skip_emulated_instruction(vcpu);
5801 kvm_vcpu_on_spin(vcpu);
5806 static int handle_nop(struct kvm_vcpu *vcpu)
5808 skip_emulated_instruction(vcpu);
5812 static int handle_mwait(struct kvm_vcpu *vcpu)
5814 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
5815 return handle_nop(vcpu);
5818 static int handle_monitor(struct kvm_vcpu *vcpu)
5820 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
5821 return handle_nop(vcpu);
5825 * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
5826 * We could reuse a single VMCS for all the L2 guests, but we also want the
5827 * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
5828 * allows keeping them loaded on the processor, and in the future will allow
5829 * optimizations where prepare_vmcs02 doesn't need to set all the fields on
5830 * every entry if they never change.
5831 * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
5832 * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
5834 * The following functions allocate and free a vmcs02 in this pool.
5837 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
5838 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
5840 struct vmcs02_list *item;
5841 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
5842 if (item->vmptr == vmx->nested.current_vmptr) {
5843 list_move(&item->list, &vmx->nested.vmcs02_pool);
5844 return &item->vmcs02;
5847 if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
5848 /* Recycle the least recently used VMCS. */
5849 item = list_entry(vmx->nested.vmcs02_pool.prev,
5850 struct vmcs02_list, list);
5851 item->vmptr = vmx->nested.current_vmptr;
5852 list_move(&item->list, &vmx->nested.vmcs02_pool);
5853 return &item->vmcs02;
5856 /* Create a new VMCS */
5857 item = kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
5860 item->vmcs02.vmcs = alloc_vmcs();
5861 if (!item->vmcs02.vmcs) {
5865 loaded_vmcs_init(&item->vmcs02);
5866 item->vmptr = vmx->nested.current_vmptr;
5867 list_add(&(item->list), &(vmx->nested.vmcs02_pool));
5868 vmx->nested.vmcs02_num++;
5869 return &item->vmcs02;
5872 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
5873 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
5875 struct vmcs02_list *item;
5876 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
5877 if (item->vmptr == vmptr) {
5878 free_loaded_vmcs(&item->vmcs02);
5879 list_del(&item->list);
5881 vmx->nested.vmcs02_num--;
5887 * Free all VMCSs saved for this vcpu, except the one pointed by
5888 * vmx->loaded_vmcs. We must be running L1, so vmx->loaded_vmcs
5889 * must be &vmx->vmcs01.
5891 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
5893 struct vmcs02_list *item, *n;
5895 WARN_ON(vmx->loaded_vmcs != &vmx->vmcs01);
5896 list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
5898 * Something will leak if the above WARN triggers. Better than
5901 if (vmx->loaded_vmcs == &item->vmcs02)
5904 free_loaded_vmcs(&item->vmcs02);
5905 list_del(&item->list);
5907 vmx->nested.vmcs02_num--;
5912 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
5913 * set the success or error code of an emulated VMX instruction, as specified
5914 * by Vol 2B, VMX Instruction Reference, "Conventions".
5916 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
5918 vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
5919 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
5920 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
5923 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
5925 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
5926 & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
5927 X86_EFLAGS_SF | X86_EFLAGS_OF))
5931 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
5932 u32 vm_instruction_error)
5934 if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
5936 * failValid writes the error number to the current VMCS, which
5937 * can't be done there isn't a current VMCS.
5939 nested_vmx_failInvalid(vcpu);
5942 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
5943 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
5944 X86_EFLAGS_SF | X86_EFLAGS_OF))
5946 get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
5948 * We don't need to force a shadow sync because
5949 * VM_INSTRUCTION_ERROR is not shadowed
5953 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
5955 struct vcpu_vmx *vmx =
5956 container_of(timer, struct vcpu_vmx, nested.preemption_timer);
5958 vmx->nested.preemption_timer_expired = true;
5959 kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
5960 kvm_vcpu_kick(&vmx->vcpu);
5962 return HRTIMER_NORESTART;
5966 * Decode the memory-address operand of a vmx instruction, as recorded on an
5967 * exit caused by such an instruction (run by a guest hypervisor).
5968 * On success, returns 0. When the operand is invalid, returns 1 and throws
5971 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
5972 unsigned long exit_qualification,
5973 u32 vmx_instruction_info, gva_t *ret)
5976 * According to Vol. 3B, "Information for VM Exits Due to Instruction
5977 * Execution", on an exit, vmx_instruction_info holds most of the
5978 * addressing components of the operand. Only the displacement part
5979 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
5980 * For how an actual address is calculated from all these components,
5981 * refer to Vol. 1, "Operand Addressing".
5983 int scaling = vmx_instruction_info & 3;
5984 int addr_size = (vmx_instruction_info >> 7) & 7;
5985 bool is_reg = vmx_instruction_info & (1u << 10);
5986 int seg_reg = (vmx_instruction_info >> 15) & 7;
5987 int index_reg = (vmx_instruction_info >> 18) & 0xf;
5988 bool index_is_valid = !(vmx_instruction_info & (1u << 22));
5989 int base_reg = (vmx_instruction_info >> 23) & 0xf;
5990 bool base_is_valid = !(vmx_instruction_info & (1u << 27));
5993 kvm_queue_exception(vcpu, UD_VECTOR);
5997 /* Addr = segment_base + offset */
5998 /* offset = base + [index * scale] + displacement */
5999 *ret = vmx_get_segment_base(vcpu, seg_reg);
6001 *ret += kvm_register_read(vcpu, base_reg);
6003 *ret += kvm_register_read(vcpu, index_reg)<<scaling;
6004 *ret += exit_qualification; /* holds the displacement */
6006 if (addr_size == 1) /* 32 bit */
6010 * TODO: throw #GP (and return 1) in various cases that the VM*
6011 * instructions require it - e.g., offset beyond segment limit,
6012 * unusable or unreadable/unwritable segment, non-canonical 64-bit
6013 * address, and so on. Currently these are not checked.
6019 * This function performs the various checks including
6020 * - if it's 4KB aligned
6021 * - No bits beyond the physical address width are set
6022 * - Returns 0 on success or else 1
6023 * (Intel SDM Section 30.3)
6025 static int nested_vmx_check_vmptr(struct kvm_vcpu *vcpu, int exit_reason,
6030 struct x86_exception e;
6032 struct vcpu_vmx *vmx = to_vmx(vcpu);
6033 int maxphyaddr = cpuid_maxphyaddr(vcpu);
6035 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
6036 vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
6039 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
6040 sizeof(vmptr), &e)) {
6041 kvm_inject_page_fault(vcpu, &e);
6045 switch (exit_reason) {
6046 case EXIT_REASON_VMON:
6049 * The first 4 bytes of VMXON region contain the supported
6050 * VMCS revision identifier
6052 * Note - IA32_VMX_BASIC[48] will never be 1
6053 * for the nested case;
6054 * which replaces physical address width with 32
6057 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6058 nested_vmx_failInvalid(vcpu);
6059 skip_emulated_instruction(vcpu);
6063 page = nested_get_page(vcpu, vmptr);
6065 *(u32 *)kmap(page) != VMCS12_REVISION) {
6066 nested_vmx_failInvalid(vcpu);
6068 skip_emulated_instruction(vcpu);
6072 vmx->nested.vmxon_ptr = vmptr;
6074 case EXIT_REASON_VMCLEAR:
6075 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6076 nested_vmx_failValid(vcpu,
6077 VMXERR_VMCLEAR_INVALID_ADDRESS);
6078 skip_emulated_instruction(vcpu);
6082 if (vmptr == vmx->nested.vmxon_ptr) {
6083 nested_vmx_failValid(vcpu,
6084 VMXERR_VMCLEAR_VMXON_POINTER);
6085 skip_emulated_instruction(vcpu);
6089 case EXIT_REASON_VMPTRLD:
6090 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6091 nested_vmx_failValid(vcpu,
6092 VMXERR_VMPTRLD_INVALID_ADDRESS);
6093 skip_emulated_instruction(vcpu);
6097 if (vmptr == vmx->nested.vmxon_ptr) {
6098 nested_vmx_failValid(vcpu,
6099 VMXERR_VMCLEAR_VMXON_POINTER);
6100 skip_emulated_instruction(vcpu);
6105 return 1; /* shouldn't happen */
6114 * Emulate the VMXON instruction.
6115 * Currently, we just remember that VMX is active, and do not save or even
6116 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
6117 * do not currently need to store anything in that guest-allocated memory
6118 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
6119 * argument is different from the VMXON pointer (which the spec says they do).
6121 static int handle_vmon(struct kvm_vcpu *vcpu)
6123 struct kvm_segment cs;
6124 struct vcpu_vmx *vmx = to_vmx(vcpu);
6125 struct vmcs *shadow_vmcs;
6126 const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
6127 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
6129 /* The Intel VMX Instruction Reference lists a bunch of bits that
6130 * are prerequisite to running VMXON, most notably cr4.VMXE must be
6131 * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
6132 * Otherwise, we should fail with #UD. We test these now:
6134 if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
6135 !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
6136 (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
6137 kvm_queue_exception(vcpu, UD_VECTOR);
6141 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
6142 if (is_long_mode(vcpu) && !cs.l) {
6143 kvm_queue_exception(vcpu, UD_VECTOR);
6147 if (vmx_get_cpl(vcpu)) {
6148 kvm_inject_gp(vcpu, 0);
6152 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMON, NULL))
6155 if (vmx->nested.vmxon) {
6156 nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
6157 skip_emulated_instruction(vcpu);
6161 if ((vmx->nested.msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
6162 != VMXON_NEEDED_FEATURES) {
6163 kvm_inject_gp(vcpu, 0);
6167 if (enable_shadow_vmcs) {
6168 shadow_vmcs = alloc_vmcs();
6171 /* mark vmcs as shadow */
6172 shadow_vmcs->revision_id |= (1u << 31);
6173 /* init shadow vmcs */
6174 vmcs_clear(shadow_vmcs);
6175 vmx->nested.current_shadow_vmcs = shadow_vmcs;
6178 INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
6179 vmx->nested.vmcs02_num = 0;
6181 hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
6183 vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
6185 vmx->nested.vmxon = true;
6187 skip_emulated_instruction(vcpu);
6188 nested_vmx_succeed(vcpu);
6193 * Intel's VMX Instruction Reference specifies a common set of prerequisites
6194 * for running VMX instructions (except VMXON, whose prerequisites are
6195 * slightly different). It also specifies what exception to inject otherwise.
6197 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
6199 struct kvm_segment cs;
6200 struct vcpu_vmx *vmx = to_vmx(vcpu);
6202 if (!vmx->nested.vmxon) {
6203 kvm_queue_exception(vcpu, UD_VECTOR);
6207 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
6208 if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
6209 (is_long_mode(vcpu) && !cs.l)) {
6210 kvm_queue_exception(vcpu, UD_VECTOR);
6214 if (vmx_get_cpl(vcpu)) {
6215 kvm_inject_gp(vcpu, 0);
6222 static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
6225 if (vmx->nested.current_vmptr == -1ull)
6228 /* current_vmptr and current_vmcs12 are always set/reset together */
6229 if (WARN_ON(vmx->nested.current_vmcs12 == NULL))
6232 if (enable_shadow_vmcs) {
6233 /* copy to memory all shadowed fields in case
6234 they were modified */
6235 copy_shadow_to_vmcs12(vmx);
6236 vmx->nested.sync_shadow_vmcs = false;
6237 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6238 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
6239 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
6240 vmcs_write64(VMCS_LINK_POINTER, -1ull);
6242 kunmap(vmx->nested.current_vmcs12_page);
6243 nested_release_page(vmx->nested.current_vmcs12_page);
6244 vmx->nested.current_vmptr = -1ull;
6245 vmx->nested.current_vmcs12 = NULL;
6249 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
6250 * just stops using VMX.
6252 static void free_nested(struct vcpu_vmx *vmx)
6254 if (!vmx->nested.vmxon)
6257 vmx->nested.vmxon = false;
6258 nested_release_vmcs12(vmx);
6259 if (enable_shadow_vmcs)
6260 free_vmcs(vmx->nested.current_shadow_vmcs);
6261 /* Unpin physical memory we referred to in current vmcs02 */
6262 if (vmx->nested.apic_access_page) {
6263 nested_release_page(vmx->nested.apic_access_page);
6264 vmx->nested.apic_access_page = NULL;
6266 if (vmx->nested.virtual_apic_page) {
6267 nested_release_page(vmx->nested.virtual_apic_page);
6268 vmx->nested.virtual_apic_page = NULL;
6271 nested_free_all_saved_vmcss(vmx);
6274 /* Emulate the VMXOFF instruction */
6275 static int handle_vmoff(struct kvm_vcpu *vcpu)
6277 if (!nested_vmx_check_permission(vcpu))
6279 free_nested(to_vmx(vcpu));
6280 skip_emulated_instruction(vcpu);
6281 nested_vmx_succeed(vcpu);
6285 /* Emulate the VMCLEAR instruction */
6286 static int handle_vmclear(struct kvm_vcpu *vcpu)
6288 struct vcpu_vmx *vmx = to_vmx(vcpu);
6290 struct vmcs12 *vmcs12;
6293 if (!nested_vmx_check_permission(vcpu))
6296 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMCLEAR, &vmptr))
6299 if (vmptr == vmx->nested.current_vmptr)
6300 nested_release_vmcs12(vmx);
6302 page = nested_get_page(vcpu, vmptr);
6305 * For accurate processor emulation, VMCLEAR beyond available
6306 * physical memory should do nothing at all. However, it is
6307 * possible that a nested vmx bug, not a guest hypervisor bug,
6308 * resulted in this case, so let's shut down before doing any
6311 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6314 vmcs12 = kmap(page);
6315 vmcs12->launch_state = 0;
6317 nested_release_page(page);
6319 nested_free_vmcs02(vmx, vmptr);
6321 skip_emulated_instruction(vcpu);
6322 nested_vmx_succeed(vcpu);
6326 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
6328 /* Emulate the VMLAUNCH instruction */
6329 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
6331 return nested_vmx_run(vcpu, true);
6334 /* Emulate the VMRESUME instruction */
6335 static int handle_vmresume(struct kvm_vcpu *vcpu)
6338 return nested_vmx_run(vcpu, false);
6341 enum vmcs_field_type {
6342 VMCS_FIELD_TYPE_U16 = 0,
6343 VMCS_FIELD_TYPE_U64 = 1,
6344 VMCS_FIELD_TYPE_U32 = 2,
6345 VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
6348 static inline int vmcs_field_type(unsigned long field)
6350 if (0x1 & field) /* the *_HIGH fields are all 32 bit */
6351 return VMCS_FIELD_TYPE_U32;
6352 return (field >> 13) & 0x3 ;
6355 static inline int vmcs_field_readonly(unsigned long field)
6357 return (((field >> 10) & 0x3) == 1);
6361 * Read a vmcs12 field. Since these can have varying lengths and we return
6362 * one type, we chose the biggest type (u64) and zero-extend the return value
6363 * to that size. Note that the caller, handle_vmread, might need to use only
6364 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
6365 * 64-bit fields are to be returned).
6367 static inline bool vmcs12_read_any(struct kvm_vcpu *vcpu,
6368 unsigned long field, u64 *ret)
6370 short offset = vmcs_field_to_offset(field);
6376 p = ((char *)(get_vmcs12(vcpu))) + offset;
6378 switch (vmcs_field_type(field)) {
6379 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6380 *ret = *((natural_width *)p);
6382 case VMCS_FIELD_TYPE_U16:
6385 case VMCS_FIELD_TYPE_U32:
6388 case VMCS_FIELD_TYPE_U64:
6392 return 0; /* can never happen. */
6397 static inline bool vmcs12_write_any(struct kvm_vcpu *vcpu,
6398 unsigned long field, u64 field_value){
6399 short offset = vmcs_field_to_offset(field);
6400 char *p = ((char *) get_vmcs12(vcpu)) + offset;
6404 switch (vmcs_field_type(field)) {
6405 case VMCS_FIELD_TYPE_U16:
6406 *(u16 *)p = field_value;
6408 case VMCS_FIELD_TYPE_U32:
6409 *(u32 *)p = field_value;
6411 case VMCS_FIELD_TYPE_U64:
6412 *(u64 *)p = field_value;
6414 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6415 *(natural_width *)p = field_value;
6418 return false; /* can never happen. */
6423 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
6426 unsigned long field;
6428 struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
6429 const unsigned long *fields = shadow_read_write_fields;
6430 const int num_fields = max_shadow_read_write_fields;
6434 vmcs_load(shadow_vmcs);
6436 for (i = 0; i < num_fields; i++) {
6438 switch (vmcs_field_type(field)) {
6439 case VMCS_FIELD_TYPE_U16:
6440 field_value = vmcs_read16(field);
6442 case VMCS_FIELD_TYPE_U32:
6443 field_value = vmcs_read32(field);
6445 case VMCS_FIELD_TYPE_U64:
6446 field_value = vmcs_read64(field);
6448 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6449 field_value = vmcs_readl(field);
6452 vmcs12_write_any(&vmx->vcpu, field, field_value);
6455 vmcs_clear(shadow_vmcs);
6456 vmcs_load(vmx->loaded_vmcs->vmcs);
6461 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
6463 const unsigned long *fields[] = {
6464 shadow_read_write_fields,
6465 shadow_read_only_fields
6467 const int max_fields[] = {
6468 max_shadow_read_write_fields,
6469 max_shadow_read_only_fields
6472 unsigned long field;
6473 u64 field_value = 0;
6474 struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
6476 vmcs_load(shadow_vmcs);
6478 for (q = 0; q < ARRAY_SIZE(fields); q++) {
6479 for (i = 0; i < max_fields[q]; i++) {
6480 field = fields[q][i];
6481 vmcs12_read_any(&vmx->vcpu, field, &field_value);
6483 switch (vmcs_field_type(field)) {
6484 case VMCS_FIELD_TYPE_U16:
6485 vmcs_write16(field, (u16)field_value);
6487 case VMCS_FIELD_TYPE_U32:
6488 vmcs_write32(field, (u32)field_value);
6490 case VMCS_FIELD_TYPE_U64:
6491 vmcs_write64(field, (u64)field_value);
6493 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6494 vmcs_writel(field, (long)field_value);
6500 vmcs_clear(shadow_vmcs);
6501 vmcs_load(vmx->loaded_vmcs->vmcs);
6505 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
6506 * used before) all generate the same failure when it is missing.
6508 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
6510 struct vcpu_vmx *vmx = to_vmx(vcpu);
6511 if (vmx->nested.current_vmptr == -1ull) {
6512 nested_vmx_failInvalid(vcpu);
6513 skip_emulated_instruction(vcpu);
6519 static int handle_vmread(struct kvm_vcpu *vcpu)
6521 unsigned long field;
6523 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6524 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6527 if (!nested_vmx_check_permission(vcpu) ||
6528 !nested_vmx_check_vmcs12(vcpu))
6531 /* Decode instruction info and find the field to read */
6532 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
6533 /* Read the field, zero-extended to a u64 field_value */
6534 if (!vmcs12_read_any(vcpu, field, &field_value)) {
6535 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
6536 skip_emulated_instruction(vcpu);
6540 * Now copy part of this value to register or memory, as requested.
6541 * Note that the number of bits actually copied is 32 or 64 depending
6542 * on the guest's mode (32 or 64 bit), not on the given field's length.
6544 if (vmx_instruction_info & (1u << 10)) {
6545 kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
6548 if (get_vmx_mem_address(vcpu, exit_qualification,
6549 vmx_instruction_info, &gva))
6551 /* _system ok, as nested_vmx_check_permission verified cpl=0 */
6552 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
6553 &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
6556 nested_vmx_succeed(vcpu);
6557 skip_emulated_instruction(vcpu);
6562 static int handle_vmwrite(struct kvm_vcpu *vcpu)
6564 unsigned long field;
6566 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6567 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6568 /* The value to write might be 32 or 64 bits, depending on L1's long
6569 * mode, and eventually we need to write that into a field of several
6570 * possible lengths. The code below first zero-extends the value to 64
6571 * bit (field_value), and then copies only the approriate number of
6572 * bits into the vmcs12 field.
6574 u64 field_value = 0;
6575 struct x86_exception e;
6577 if (!nested_vmx_check_permission(vcpu) ||
6578 !nested_vmx_check_vmcs12(vcpu))
6581 if (vmx_instruction_info & (1u << 10))
6582 field_value = kvm_register_readl(vcpu,
6583 (((vmx_instruction_info) >> 3) & 0xf));
6585 if (get_vmx_mem_address(vcpu, exit_qualification,
6586 vmx_instruction_info, &gva))
6588 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
6589 &field_value, (is_64_bit_mode(vcpu) ? 8 : 4), &e)) {
6590 kvm_inject_page_fault(vcpu, &e);
6596 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
6597 if (vmcs_field_readonly(field)) {
6598 nested_vmx_failValid(vcpu,
6599 VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
6600 skip_emulated_instruction(vcpu);
6604 if (!vmcs12_write_any(vcpu, field, field_value)) {
6605 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
6606 skip_emulated_instruction(vcpu);
6610 nested_vmx_succeed(vcpu);
6611 skip_emulated_instruction(vcpu);
6615 /* Emulate the VMPTRLD instruction */
6616 static int handle_vmptrld(struct kvm_vcpu *vcpu)
6618 struct vcpu_vmx *vmx = to_vmx(vcpu);
6622 if (!nested_vmx_check_permission(vcpu))
6625 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMPTRLD, &vmptr))
6628 if (vmx->nested.current_vmptr != vmptr) {
6629 struct vmcs12 *new_vmcs12;
6631 page = nested_get_page(vcpu, vmptr);
6633 nested_vmx_failInvalid(vcpu);
6634 skip_emulated_instruction(vcpu);
6637 new_vmcs12 = kmap(page);
6638 if (new_vmcs12->revision_id != VMCS12_REVISION) {
6640 nested_release_page_clean(page);
6641 nested_vmx_failValid(vcpu,
6642 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
6643 skip_emulated_instruction(vcpu);
6647 nested_release_vmcs12(vmx);
6648 vmx->nested.current_vmptr = vmptr;
6649 vmx->nested.current_vmcs12 = new_vmcs12;
6650 vmx->nested.current_vmcs12_page = page;
6651 if (enable_shadow_vmcs) {
6652 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6653 exec_control |= SECONDARY_EXEC_SHADOW_VMCS;
6654 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
6655 vmcs_write64(VMCS_LINK_POINTER,
6656 __pa(vmx->nested.current_shadow_vmcs));
6657 vmx->nested.sync_shadow_vmcs = true;
6661 nested_vmx_succeed(vcpu);
6662 skip_emulated_instruction(vcpu);
6666 /* Emulate the VMPTRST instruction */
6667 static int handle_vmptrst(struct kvm_vcpu *vcpu)
6669 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6670 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6672 struct x86_exception e;
6674 if (!nested_vmx_check_permission(vcpu))
6677 if (get_vmx_mem_address(vcpu, exit_qualification,
6678 vmx_instruction_info, &vmcs_gva))
6680 /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
6681 if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
6682 (void *)&to_vmx(vcpu)->nested.current_vmptr,
6684 kvm_inject_page_fault(vcpu, &e);
6687 nested_vmx_succeed(vcpu);
6688 skip_emulated_instruction(vcpu);
6692 /* Emulate the INVEPT instruction */
6693 static int handle_invept(struct kvm_vcpu *vcpu)
6695 u32 vmx_instruction_info, types;
6698 struct x86_exception e;
6703 if (!(nested_vmx_secondary_ctls_high & SECONDARY_EXEC_ENABLE_EPT) ||
6704 !(nested_vmx_ept_caps & VMX_EPT_INVEPT_BIT)) {
6705 kvm_queue_exception(vcpu, UD_VECTOR);
6709 if (!nested_vmx_check_permission(vcpu))
6712 if (!kvm_read_cr0_bits(vcpu, X86_CR0_PE)) {
6713 kvm_queue_exception(vcpu, UD_VECTOR);
6717 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6718 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
6720 types = (nested_vmx_ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
6722 if (!(types & (1UL << type))) {
6723 nested_vmx_failValid(vcpu,
6724 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
6728 /* According to the Intel VMX instruction reference, the memory
6729 * operand is read even if it isn't needed (e.g., for type==global)
6731 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
6732 vmx_instruction_info, &gva))
6734 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
6735 sizeof(operand), &e)) {
6736 kvm_inject_page_fault(vcpu, &e);
6741 case VMX_EPT_EXTENT_GLOBAL:
6742 kvm_mmu_sync_roots(vcpu);
6743 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
6744 nested_vmx_succeed(vcpu);
6747 /* Trap single context invalidation invept calls */
6752 skip_emulated_instruction(vcpu);
6756 static int handle_invvpid(struct kvm_vcpu *vcpu)
6758 kvm_queue_exception(vcpu, UD_VECTOR);
6763 * The exit handlers return 1 if the exit was handled fully and guest execution
6764 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
6765 * to be done to userspace and return 0.
6767 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
6768 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
6769 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
6770 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
6771 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
6772 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
6773 [EXIT_REASON_CR_ACCESS] = handle_cr,
6774 [EXIT_REASON_DR_ACCESS] = handle_dr,
6775 [EXIT_REASON_CPUID] = handle_cpuid,
6776 [EXIT_REASON_MSR_READ] = handle_rdmsr,
6777 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
6778 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
6779 [EXIT_REASON_HLT] = handle_halt,
6780 [EXIT_REASON_INVD] = handle_invd,
6781 [EXIT_REASON_INVLPG] = handle_invlpg,
6782 [EXIT_REASON_RDPMC] = handle_rdpmc,
6783 [EXIT_REASON_VMCALL] = handle_vmcall,
6784 [EXIT_REASON_VMCLEAR] = handle_vmclear,
6785 [EXIT_REASON_VMLAUNCH] = handle_vmlaunch,
6786 [EXIT_REASON_VMPTRLD] = handle_vmptrld,
6787 [EXIT_REASON_VMPTRST] = handle_vmptrst,
6788 [EXIT_REASON_VMREAD] = handle_vmread,
6789 [EXIT_REASON_VMRESUME] = handle_vmresume,
6790 [EXIT_REASON_VMWRITE] = handle_vmwrite,
6791 [EXIT_REASON_VMOFF] = handle_vmoff,
6792 [EXIT_REASON_VMON] = handle_vmon,
6793 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
6794 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
6795 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
6796 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
6797 [EXIT_REASON_WBINVD] = handle_wbinvd,
6798 [EXIT_REASON_XSETBV] = handle_xsetbv,
6799 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
6800 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
6801 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
6802 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
6803 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
6804 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait,
6805 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor,
6806 [EXIT_REASON_INVEPT] = handle_invept,
6807 [EXIT_REASON_INVVPID] = handle_invvpid,
6810 static const int kvm_vmx_max_exit_handlers =
6811 ARRAY_SIZE(kvm_vmx_exit_handlers);
6813 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
6814 struct vmcs12 *vmcs12)
6816 unsigned long exit_qualification;
6817 gpa_t bitmap, last_bitmap;
6822 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
6823 return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
6825 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6827 port = exit_qualification >> 16;
6828 size = (exit_qualification & 7) + 1;
6830 last_bitmap = (gpa_t)-1;
6835 bitmap = vmcs12->io_bitmap_a;
6836 else if (port < 0x10000)
6837 bitmap = vmcs12->io_bitmap_b;
6840 bitmap += (port & 0x7fff) / 8;
6842 if (last_bitmap != bitmap)
6843 if (kvm_read_guest(vcpu->kvm, bitmap, &b, 1))
6845 if (b & (1 << (port & 7)))
6850 last_bitmap = bitmap;
6857 * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
6858 * rather than handle it ourselves in L0. I.e., check whether L1 expressed
6859 * disinterest in the current event (read or write a specific MSR) by using an
6860 * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
6862 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
6863 struct vmcs12 *vmcs12, u32 exit_reason)
6865 u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
6868 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
6872 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
6873 * for the four combinations of read/write and low/high MSR numbers.
6874 * First we need to figure out which of the four to use:
6876 bitmap = vmcs12->msr_bitmap;
6877 if (exit_reason == EXIT_REASON_MSR_WRITE)
6879 if (msr_index >= 0xc0000000) {
6880 msr_index -= 0xc0000000;
6884 /* Then read the msr_index'th bit from this bitmap: */
6885 if (msr_index < 1024*8) {
6887 if (kvm_read_guest(vcpu->kvm, bitmap + msr_index/8, &b, 1))
6889 return 1 & (b >> (msr_index & 7));
6891 return 1; /* let L1 handle the wrong parameter */
6895 * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
6896 * rather than handle it ourselves in L0. I.e., check if L1 wanted to
6897 * intercept (via guest_host_mask etc.) the current event.
6899 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
6900 struct vmcs12 *vmcs12)
6902 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6903 int cr = exit_qualification & 15;
6904 int reg = (exit_qualification >> 8) & 15;
6905 unsigned long val = kvm_register_readl(vcpu, reg);
6907 switch ((exit_qualification >> 4) & 3) {
6908 case 0: /* mov to cr */
6911 if (vmcs12->cr0_guest_host_mask &
6912 (val ^ vmcs12->cr0_read_shadow))
6916 if ((vmcs12->cr3_target_count >= 1 &&
6917 vmcs12->cr3_target_value0 == val) ||
6918 (vmcs12->cr3_target_count >= 2 &&
6919 vmcs12->cr3_target_value1 == val) ||
6920 (vmcs12->cr3_target_count >= 3 &&
6921 vmcs12->cr3_target_value2 == val) ||
6922 (vmcs12->cr3_target_count >= 4 &&
6923 vmcs12->cr3_target_value3 == val))
6925 if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
6929 if (vmcs12->cr4_guest_host_mask &
6930 (vmcs12->cr4_read_shadow ^ val))
6934 if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
6940 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
6941 (vmcs12->cr0_read_shadow & X86_CR0_TS))
6944 case 1: /* mov from cr */
6947 if (vmcs12->cpu_based_vm_exec_control &
6948 CPU_BASED_CR3_STORE_EXITING)
6952 if (vmcs12->cpu_based_vm_exec_control &
6953 CPU_BASED_CR8_STORE_EXITING)
6960 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
6961 * cr0. Other attempted changes are ignored, with no exit.
6963 if (vmcs12->cr0_guest_host_mask & 0xe &
6964 (val ^ vmcs12->cr0_read_shadow))
6966 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
6967 !(vmcs12->cr0_read_shadow & 0x1) &&
6976 * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
6977 * should handle it ourselves in L0 (and then continue L2). Only call this
6978 * when in is_guest_mode (L2).
6980 static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu)
6982 u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6983 struct vcpu_vmx *vmx = to_vmx(vcpu);
6984 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6985 u32 exit_reason = vmx->exit_reason;
6987 trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
6988 vmcs_readl(EXIT_QUALIFICATION),
6989 vmx->idt_vectoring_info,
6991 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
6994 if (vmx->nested.nested_run_pending)
6997 if (unlikely(vmx->fail)) {
6998 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
6999 vmcs_read32(VM_INSTRUCTION_ERROR));
7003 switch (exit_reason) {
7004 case EXIT_REASON_EXCEPTION_NMI:
7005 if (!is_exception(intr_info))
7007 else if (is_page_fault(intr_info))
7009 else if (is_no_device(intr_info) &&
7010 !(vmcs12->guest_cr0 & X86_CR0_TS))
7012 return vmcs12->exception_bitmap &
7013 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
7014 case EXIT_REASON_EXTERNAL_INTERRUPT:
7016 case EXIT_REASON_TRIPLE_FAULT:
7018 case EXIT_REASON_PENDING_INTERRUPT:
7019 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
7020 case EXIT_REASON_NMI_WINDOW:
7021 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
7022 case EXIT_REASON_TASK_SWITCH:
7024 case EXIT_REASON_CPUID:
7025 if (kvm_register_read(vcpu, VCPU_REGS_RAX) == 0xa)
7028 case EXIT_REASON_HLT:
7029 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
7030 case EXIT_REASON_INVD:
7032 case EXIT_REASON_INVLPG:
7033 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
7034 case EXIT_REASON_RDPMC:
7035 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
7036 case EXIT_REASON_RDTSC:
7037 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
7038 case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
7039 case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
7040 case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
7041 case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
7042 case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
7043 case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
7045 * VMX instructions trap unconditionally. This allows L1 to
7046 * emulate them for its L2 guest, i.e., allows 3-level nesting!
7049 case EXIT_REASON_CR_ACCESS:
7050 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
7051 case EXIT_REASON_DR_ACCESS:
7052 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
7053 case EXIT_REASON_IO_INSTRUCTION:
7054 return nested_vmx_exit_handled_io(vcpu, vmcs12);
7055 case EXIT_REASON_MSR_READ:
7056 case EXIT_REASON_MSR_WRITE:
7057 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
7058 case EXIT_REASON_INVALID_STATE:
7060 case EXIT_REASON_MWAIT_INSTRUCTION:
7061 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
7062 case EXIT_REASON_MONITOR_INSTRUCTION:
7063 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
7064 case EXIT_REASON_PAUSE_INSTRUCTION:
7065 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
7066 nested_cpu_has2(vmcs12,
7067 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
7068 case EXIT_REASON_MCE_DURING_VMENTRY:
7070 case EXIT_REASON_TPR_BELOW_THRESHOLD:
7071 return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
7072 case EXIT_REASON_APIC_ACCESS:
7073 return nested_cpu_has2(vmcs12,
7074 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
7075 case EXIT_REASON_EPT_VIOLATION:
7077 * L0 always deals with the EPT violation. If nested EPT is
7078 * used, and the nested mmu code discovers that the address is
7079 * missing in the guest EPT table (EPT12), the EPT violation
7080 * will be injected with nested_ept_inject_page_fault()
7083 case EXIT_REASON_EPT_MISCONFIG:
7085 * L2 never uses directly L1's EPT, but rather L0's own EPT
7086 * table (shadow on EPT) or a merged EPT table that L0 built
7087 * (EPT on EPT). So any problems with the structure of the
7088 * table is L0's fault.
7091 case EXIT_REASON_WBINVD:
7092 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
7093 case EXIT_REASON_XSETBV:
7100 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
7102 *info1 = vmcs_readl(EXIT_QUALIFICATION);
7103 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
7107 * The guest has exited. See if we can fix it or if we need userspace
7110 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
7112 struct vcpu_vmx *vmx = to_vmx(vcpu);
7113 u32 exit_reason = vmx->exit_reason;
7114 u32 vectoring_info = vmx->idt_vectoring_info;
7116 /* If guest state is invalid, start emulating */
7117 if (vmx->emulation_required)
7118 return handle_invalid_guest_state(vcpu);
7120 if (is_guest_mode(vcpu) && nested_vmx_exit_handled(vcpu)) {
7121 nested_vmx_vmexit(vcpu, exit_reason,
7122 vmcs_read32(VM_EXIT_INTR_INFO),
7123 vmcs_readl(EXIT_QUALIFICATION));
7127 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
7128 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
7129 vcpu->run->fail_entry.hardware_entry_failure_reason
7134 if (unlikely(vmx->fail)) {
7135 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
7136 vcpu->run->fail_entry.hardware_entry_failure_reason
7137 = vmcs_read32(VM_INSTRUCTION_ERROR);
7143 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
7144 * delivery event since it indicates guest is accessing MMIO.
7145 * The vm-exit can be triggered again after return to guest that
7146 * will cause infinite loop.
7148 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
7149 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
7150 exit_reason != EXIT_REASON_EPT_VIOLATION &&
7151 exit_reason != EXIT_REASON_TASK_SWITCH)) {
7152 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7153 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
7154 vcpu->run->internal.ndata = 2;
7155 vcpu->run->internal.data[0] = vectoring_info;
7156 vcpu->run->internal.data[1] = exit_reason;
7160 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked &&
7161 !(is_guest_mode(vcpu) && nested_cpu_has_virtual_nmis(
7162 get_vmcs12(vcpu))))) {
7163 if (vmx_interrupt_allowed(vcpu)) {
7164 vmx->soft_vnmi_blocked = 0;
7165 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
7166 vcpu->arch.nmi_pending) {
7168 * This CPU don't support us in finding the end of an
7169 * NMI-blocked window if the guest runs with IRQs
7170 * disabled. So we pull the trigger after 1 s of
7171 * futile waiting, but inform the user about this.
7173 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
7174 "state on VCPU %d after 1 s timeout\n",
7175 __func__, vcpu->vcpu_id);
7176 vmx->soft_vnmi_blocked = 0;
7180 if (exit_reason < kvm_vmx_max_exit_handlers
7181 && kvm_vmx_exit_handlers[exit_reason])
7182 return kvm_vmx_exit_handlers[exit_reason](vcpu);
7184 WARN_ONCE(1, "vmx: unexpected exit reason 0x%x\n", exit_reason);
7185 kvm_queue_exception(vcpu, UD_VECTOR);
7190 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
7192 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7194 if (is_guest_mode(vcpu) &&
7195 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
7198 if (irr == -1 || tpr < irr) {
7199 vmcs_write32(TPR_THRESHOLD, 0);
7203 vmcs_write32(TPR_THRESHOLD, irr);
7206 static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
7208 u32 sec_exec_control;
7211 * There is not point to enable virtualize x2apic without enable
7214 if (!cpu_has_vmx_virtualize_x2apic_mode() ||
7215 !vmx_vm_has_apicv(vcpu->kvm))
7218 if (!vm_need_tpr_shadow(vcpu->kvm))
7221 sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7224 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
7225 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
7227 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
7228 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
7230 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
7232 vmx_set_msr_bitmap(vcpu);
7235 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
7237 struct vcpu_vmx *vmx = to_vmx(vcpu);
7240 * Currently we do not handle the nested case where L2 has an
7241 * APIC access page of its own; that page is still pinned.
7242 * Hence, we skip the case where the VCPU is in guest mode _and_
7243 * L1 prepared an APIC access page for L2.
7245 * For the case where L1 and L2 share the same APIC access page
7246 * (flexpriority=Y but SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES clear
7247 * in the vmcs12), this function will only update either the vmcs01
7248 * or the vmcs02. If the former, the vmcs02 will be updated by
7249 * prepare_vmcs02. If the latter, the vmcs01 will be updated in
7250 * the next L2->L1 exit.
7252 if (!is_guest_mode(vcpu) ||
7253 !nested_cpu_has2(vmx->nested.current_vmcs12,
7254 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
7255 vmcs_write64(APIC_ACCESS_ADDR, hpa);
7258 static void vmx_hwapic_isr_update(struct kvm *kvm, int isr)
7263 if (!vmx_vm_has_apicv(kvm))
7269 status = vmcs_read16(GUEST_INTR_STATUS);
7274 vmcs_write16(GUEST_INTR_STATUS, status);
7278 static void vmx_set_rvi(int vector)
7283 status = vmcs_read16(GUEST_INTR_STATUS);
7284 old = (u8)status & 0xff;
7285 if ((u8)vector != old) {
7287 status |= (u8)vector;
7288 vmcs_write16(GUEST_INTR_STATUS, status);
7292 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
7298 * If a vmexit is needed, vmx_check_nested_events handles it.
7300 if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu))
7303 if (!is_guest_mode(vcpu)) {
7304 vmx_set_rvi(max_irr);
7309 * Fall back to pre-APICv interrupt injection since L2
7310 * is run without virtual interrupt delivery.
7312 if (!kvm_event_needs_reinjection(vcpu) &&
7313 vmx_interrupt_allowed(vcpu)) {
7314 kvm_queue_interrupt(vcpu, max_irr, false);
7315 vmx_inject_irq(vcpu);
7319 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
7321 if (!vmx_vm_has_apicv(vcpu->kvm))
7324 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
7325 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
7326 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
7327 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
7330 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
7334 if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
7335 || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
7338 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
7339 exit_intr_info = vmx->exit_intr_info;
7341 /* Handle machine checks before interrupts are enabled */
7342 if (is_machine_check(exit_intr_info))
7343 kvm_machine_check();
7345 /* We need to handle NMIs before interrupts are enabled */
7346 if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
7347 (exit_intr_info & INTR_INFO_VALID_MASK)) {
7348 kvm_before_handle_nmi(&vmx->vcpu);
7350 kvm_after_handle_nmi(&vmx->vcpu);
7354 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
7356 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
7359 * If external interrupt exists, IF bit is set in rflags/eflags on the
7360 * interrupt stack frame, and interrupt will be enabled on a return
7361 * from interrupt handler.
7363 if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
7364 == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
7365 unsigned int vector;
7366 unsigned long entry;
7368 struct vcpu_vmx *vmx = to_vmx(vcpu);
7369 #ifdef CONFIG_X86_64
7373 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
7374 desc = (gate_desc *)vmx->host_idt_base + vector;
7375 entry = gate_offset(*desc);
7377 #ifdef CONFIG_X86_64
7378 "mov %%" _ASM_SP ", %[sp]\n\t"
7379 "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
7384 "orl $0x200, (%%" _ASM_SP ")\n\t"
7385 __ASM_SIZE(push) " $%c[cs]\n\t"
7386 "call *%[entry]\n\t"
7388 #ifdef CONFIG_X86_64
7393 [ss]"i"(__KERNEL_DS),
7394 [cs]"i"(__KERNEL_CS)
7400 static bool vmx_mpx_supported(void)
7402 return (vmcs_config.vmexit_ctrl & VM_EXIT_CLEAR_BNDCFGS) &&
7403 (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_BNDCFGS);
7406 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
7411 bool idtv_info_valid;
7413 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
7415 if (cpu_has_virtual_nmis()) {
7416 if (vmx->nmi_known_unmasked)
7419 * Can't use vmx->exit_intr_info since we're not sure what
7420 * the exit reason is.
7422 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
7423 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
7424 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
7426 * SDM 3: 27.7.1.2 (September 2008)
7427 * Re-set bit "block by NMI" before VM entry if vmexit caused by
7428 * a guest IRET fault.
7429 * SDM 3: 23.2.2 (September 2008)
7430 * Bit 12 is undefined in any of the following cases:
7431 * If the VM exit sets the valid bit in the IDT-vectoring
7432 * information field.
7433 * If the VM exit is due to a double fault.
7435 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
7436 vector != DF_VECTOR && !idtv_info_valid)
7437 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
7438 GUEST_INTR_STATE_NMI);
7440 vmx->nmi_known_unmasked =
7441 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
7442 & GUEST_INTR_STATE_NMI);
7443 } else if (unlikely(vmx->soft_vnmi_blocked))
7444 vmx->vnmi_blocked_time +=
7445 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
7448 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
7449 u32 idt_vectoring_info,
7450 int instr_len_field,
7451 int error_code_field)
7455 bool idtv_info_valid;
7457 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
7459 vcpu->arch.nmi_injected = false;
7460 kvm_clear_exception_queue(vcpu);
7461 kvm_clear_interrupt_queue(vcpu);
7463 if (!idtv_info_valid)
7466 kvm_make_request(KVM_REQ_EVENT, vcpu);
7468 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
7469 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
7472 case INTR_TYPE_NMI_INTR:
7473 vcpu->arch.nmi_injected = true;
7475 * SDM 3: 27.7.1.2 (September 2008)
7476 * Clear bit "block by NMI" before VM entry if a NMI
7479 vmx_set_nmi_mask(vcpu, false);
7481 case INTR_TYPE_SOFT_EXCEPTION:
7482 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
7484 case INTR_TYPE_HARD_EXCEPTION:
7485 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
7486 u32 err = vmcs_read32(error_code_field);
7487 kvm_requeue_exception_e(vcpu, vector, err);
7489 kvm_requeue_exception(vcpu, vector);
7491 case INTR_TYPE_SOFT_INTR:
7492 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
7494 case INTR_TYPE_EXT_INTR:
7495 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
7502 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
7504 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
7505 VM_EXIT_INSTRUCTION_LEN,
7506 IDT_VECTORING_ERROR_CODE);
7509 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
7511 __vmx_complete_interrupts(vcpu,
7512 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
7513 VM_ENTRY_INSTRUCTION_LEN,
7514 VM_ENTRY_EXCEPTION_ERROR_CODE);
7516 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
7519 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
7522 struct perf_guest_switch_msr *msrs;
7524 msrs = perf_guest_get_msrs(&nr_msrs);
7529 for (i = 0; i < nr_msrs; i++)
7530 if (msrs[i].host == msrs[i].guest)
7531 clear_atomic_switch_msr(vmx, msrs[i].msr);
7533 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
7537 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
7539 struct vcpu_vmx *vmx = to_vmx(vcpu);
7540 unsigned long debugctlmsr, cr4;
7542 /* Record the guest's net vcpu time for enforced NMI injections. */
7543 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
7544 vmx->entry_time = ktime_get();
7546 /* Don't enter VMX if guest state is invalid, let the exit handler
7547 start emulation until we arrive back to a valid state */
7548 if (vmx->emulation_required)
7551 if (vmx->ple_window_dirty) {
7552 vmx->ple_window_dirty = false;
7553 vmcs_write32(PLE_WINDOW, vmx->ple_window);
7556 if (vmx->nested.sync_shadow_vmcs) {
7557 copy_vmcs12_to_shadow(vmx);
7558 vmx->nested.sync_shadow_vmcs = false;
7561 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
7562 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
7563 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
7564 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
7567 if (unlikely(cr4 != vmx->host_state.vmcs_host_cr4)) {
7568 vmcs_writel(HOST_CR4, cr4);
7569 vmx->host_state.vmcs_host_cr4 = cr4;
7572 /* When single-stepping over STI and MOV SS, we must clear the
7573 * corresponding interruptibility bits in the guest state. Otherwise
7574 * vmentry fails as it then expects bit 14 (BS) in pending debug
7575 * exceptions being set, but that's not correct for the guest debugging
7577 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7578 vmx_set_interrupt_shadow(vcpu, 0);
7580 atomic_switch_perf_msrs(vmx);
7581 debugctlmsr = get_debugctlmsr();
7583 vmx->__launched = vmx->loaded_vmcs->launched;
7585 /* Store host registers */
7586 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
7587 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
7588 "push %%" _ASM_CX " \n\t"
7589 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
7591 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
7592 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
7594 /* Reload cr2 if changed */
7595 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
7596 "mov %%cr2, %%" _ASM_DX " \n\t"
7597 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
7599 "mov %%" _ASM_AX", %%cr2 \n\t"
7601 /* Check if vmlaunch of vmresume is needed */
7602 "cmpl $0, %c[launched](%0) \n\t"
7603 /* Load guest registers. Don't clobber flags. */
7604 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
7605 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
7606 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
7607 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
7608 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
7609 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
7610 #ifdef CONFIG_X86_64
7611 "mov %c[r8](%0), %%r8 \n\t"
7612 "mov %c[r9](%0), %%r9 \n\t"
7613 "mov %c[r10](%0), %%r10 \n\t"
7614 "mov %c[r11](%0), %%r11 \n\t"
7615 "mov %c[r12](%0), %%r12 \n\t"
7616 "mov %c[r13](%0), %%r13 \n\t"
7617 "mov %c[r14](%0), %%r14 \n\t"
7618 "mov %c[r15](%0), %%r15 \n\t"
7620 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
7622 /* Enter guest mode */
7624 __ex(ASM_VMX_VMLAUNCH) "\n\t"
7626 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
7628 /* Save guest registers, load host registers, keep flags */
7629 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
7631 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
7632 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
7633 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
7634 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
7635 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
7636 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
7637 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
7638 #ifdef CONFIG_X86_64
7639 "mov %%r8, %c[r8](%0) \n\t"
7640 "mov %%r9, %c[r9](%0) \n\t"
7641 "mov %%r10, %c[r10](%0) \n\t"
7642 "mov %%r11, %c[r11](%0) \n\t"
7643 "mov %%r12, %c[r12](%0) \n\t"
7644 "mov %%r13, %c[r13](%0) \n\t"
7645 "mov %%r14, %c[r14](%0) \n\t"
7646 "mov %%r15, %c[r15](%0) \n\t"
7648 "mov %%cr2, %%" _ASM_AX " \n\t"
7649 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
7651 "pop %%" _ASM_BP "; pop %%" _ASM_DX " \n\t"
7652 "setbe %c[fail](%0) \n\t"
7653 ".pushsection .rodata \n\t"
7654 ".global vmx_return \n\t"
7655 "vmx_return: " _ASM_PTR " 2b \n\t"
7657 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
7658 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
7659 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
7660 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
7661 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
7662 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
7663 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
7664 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
7665 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
7666 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
7667 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
7668 #ifdef CONFIG_X86_64
7669 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
7670 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
7671 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
7672 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
7673 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
7674 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
7675 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
7676 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
7678 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
7679 [wordsize]"i"(sizeof(ulong))
7681 #ifdef CONFIG_X86_64
7682 , "rax", "rbx", "rdi", "rsi"
7683 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
7685 , "eax", "ebx", "edi", "esi"
7689 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
7691 update_debugctlmsr(debugctlmsr);
7693 #ifndef CONFIG_X86_64
7695 * The sysexit path does not restore ds/es, so we must set them to
7696 * a reasonable value ourselves.
7698 * We can't defer this to vmx_load_host_state() since that function
7699 * may be executed in interrupt context, which saves and restore segments
7700 * around it, nullifying its effect.
7702 loadsegment(ds, __USER_DS);
7703 loadsegment(es, __USER_DS);
7706 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
7707 | (1 << VCPU_EXREG_RFLAGS)
7708 | (1 << VCPU_EXREG_PDPTR)
7709 | (1 << VCPU_EXREG_SEGMENTS)
7710 | (1 << VCPU_EXREG_CR3));
7711 vcpu->arch.regs_dirty = 0;
7713 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
7715 vmx->loaded_vmcs->launched = 1;
7717 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
7718 trace_kvm_exit(vmx->exit_reason, vcpu, KVM_ISA_VMX);
7721 * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
7722 * we did not inject a still-pending event to L1 now because of
7723 * nested_run_pending, we need to re-enable this bit.
7725 if (vmx->nested.nested_run_pending)
7726 kvm_make_request(KVM_REQ_EVENT, vcpu);
7728 vmx->nested.nested_run_pending = 0;
7730 vmx_complete_atomic_exit(vmx);
7731 vmx_recover_nmi_blocking(vmx);
7732 vmx_complete_interrupts(vmx);
7735 static void vmx_load_vmcs01(struct kvm_vcpu *vcpu)
7737 struct vcpu_vmx *vmx = to_vmx(vcpu);
7740 if (vmx->loaded_vmcs == &vmx->vmcs01)
7744 vmx->loaded_vmcs = &vmx->vmcs01;
7746 vmx_vcpu_load(vcpu, cpu);
7751 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
7753 struct vcpu_vmx *vmx = to_vmx(vcpu);
7756 leave_guest_mode(vcpu);
7757 vmx_load_vmcs01(vcpu);
7759 free_loaded_vmcs(vmx->loaded_vmcs);
7760 kfree(vmx->guest_msrs);
7761 kvm_vcpu_uninit(vcpu);
7762 kmem_cache_free(kvm_vcpu_cache, vmx);
7765 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
7768 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
7772 return ERR_PTR(-ENOMEM);
7776 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
7780 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
7781 BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
7785 if (!vmx->guest_msrs) {
7789 vmx->loaded_vmcs = &vmx->vmcs01;
7790 vmx->loaded_vmcs->vmcs = alloc_vmcs();
7791 if (!vmx->loaded_vmcs->vmcs)
7794 kvm_cpu_vmxon(__pa(per_cpu(vmxarea, raw_smp_processor_id())));
7795 loaded_vmcs_init(vmx->loaded_vmcs);
7800 vmx_vcpu_load(&vmx->vcpu, cpu);
7801 vmx->vcpu.cpu = cpu;
7802 err = vmx_vcpu_setup(vmx);
7803 vmx_vcpu_put(&vmx->vcpu);
7807 if (vm_need_virtualize_apic_accesses(kvm)) {
7808 err = alloc_apic_access_page(kvm);
7814 if (!kvm->arch.ept_identity_map_addr)
7815 kvm->arch.ept_identity_map_addr =
7816 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
7817 err = init_rmode_identity_map(kvm);
7822 vmx->nested.current_vmptr = -1ull;
7823 vmx->nested.current_vmcs12 = NULL;
7828 free_loaded_vmcs(vmx->loaded_vmcs);
7830 kfree(vmx->guest_msrs);
7832 kvm_vcpu_uninit(&vmx->vcpu);
7835 kmem_cache_free(kvm_vcpu_cache, vmx);
7836 return ERR_PTR(err);
7839 static void __init vmx_check_processor_compat(void *rtn)
7841 struct vmcs_config vmcs_conf;
7844 if (setup_vmcs_config(&vmcs_conf) < 0)
7846 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
7847 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
7848 smp_processor_id());
7853 static int get_ept_level(void)
7855 return VMX_EPT_DEFAULT_GAW + 1;
7858 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
7862 /* For VT-d and EPT combination
7863 * 1. MMIO: always map as UC
7865 * a. VT-d without snooping control feature: can't guarantee the
7866 * result, try to trust guest.
7867 * b. VT-d with snooping control feature: snooping control feature of
7868 * VT-d engine can guarantee the cache correctness. Just set it
7869 * to WB to keep consistent with host. So the same as item 3.
7870 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
7871 * consistent with host MTRR
7874 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
7875 else if (kvm_arch_has_noncoherent_dma(vcpu->kvm))
7876 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
7877 VMX_EPT_MT_EPTE_SHIFT;
7879 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
7885 static int vmx_get_lpage_level(void)
7887 if (enable_ept && !cpu_has_vmx_ept_1g_page())
7888 return PT_DIRECTORY_LEVEL;
7890 /* For shadow and EPT supported 1GB page */
7891 return PT_PDPE_LEVEL;
7894 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
7896 struct kvm_cpuid_entry2 *best;
7897 struct vcpu_vmx *vmx = to_vmx(vcpu);
7900 vmx->rdtscp_enabled = false;
7901 if (vmx_rdtscp_supported()) {
7902 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7903 if (exec_control & SECONDARY_EXEC_RDTSCP) {
7904 best = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
7905 if (best && (best->edx & bit(X86_FEATURE_RDTSCP)))
7906 vmx->rdtscp_enabled = true;
7908 exec_control &= ~SECONDARY_EXEC_RDTSCP;
7909 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
7915 /* Exposing INVPCID only when PCID is exposed */
7916 best = kvm_find_cpuid_entry(vcpu, 0x7, 0);
7917 if (vmx_invpcid_supported() &&
7918 best && (best->ebx & bit(X86_FEATURE_INVPCID)) &&
7919 guest_cpuid_has_pcid(vcpu)) {
7920 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7921 exec_control |= SECONDARY_EXEC_ENABLE_INVPCID;
7922 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
7925 if (cpu_has_secondary_exec_ctrls()) {
7926 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7927 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
7928 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
7932 best->ebx &= ~bit(X86_FEATURE_INVPCID);
7936 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
7938 if (func == 1 && nested)
7939 entry->ecx |= bit(X86_FEATURE_VMX);
7942 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
7943 struct x86_exception *fault)
7945 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7948 if (fault->error_code & PFERR_RSVD_MASK)
7949 exit_reason = EXIT_REASON_EPT_MISCONFIG;
7951 exit_reason = EXIT_REASON_EPT_VIOLATION;
7952 nested_vmx_vmexit(vcpu, exit_reason, 0, vcpu->arch.exit_qualification);
7953 vmcs12->guest_physical_address = fault->address;
7956 /* Callbacks for nested_ept_init_mmu_context: */
7958 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
7960 /* return the page table to be shadowed - in our case, EPT12 */
7961 return get_vmcs12(vcpu)->ept_pointer;
7964 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
7966 kvm_init_shadow_ept_mmu(vcpu, &vcpu->arch.mmu,
7967 nested_vmx_ept_caps & VMX_EPT_EXECUTE_ONLY_BIT);
7969 vcpu->arch.mmu.set_cr3 = vmx_set_cr3;
7970 vcpu->arch.mmu.get_cr3 = nested_ept_get_cr3;
7971 vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
7973 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
7976 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
7978 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
7981 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
7982 struct x86_exception *fault)
7984 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7986 WARN_ON(!is_guest_mode(vcpu));
7988 /* TODO: also check PFEC_MATCH/MASK, not just EB.PF. */
7989 if (vmcs12->exception_bitmap & (1u << PF_VECTOR))
7990 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
7991 vmcs_read32(VM_EXIT_INTR_INFO),
7992 vmcs_readl(EXIT_QUALIFICATION));
7994 kvm_inject_page_fault(vcpu, fault);
7997 static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu,
7998 struct vmcs12 *vmcs12)
8000 struct vcpu_vmx *vmx = to_vmx(vcpu);
8002 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
8003 /* TODO: Also verify bits beyond physical address width are 0 */
8004 if (!PAGE_ALIGNED(vmcs12->apic_access_addr))
8008 * Translate L1 physical address to host physical
8009 * address for vmcs02. Keep the page pinned, so this
8010 * physical address remains valid. We keep a reference
8011 * to it so we can release it later.
8013 if (vmx->nested.apic_access_page) /* shouldn't happen */
8014 nested_release_page(vmx->nested.apic_access_page);
8015 vmx->nested.apic_access_page =
8016 nested_get_page(vcpu, vmcs12->apic_access_addr);
8019 if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
8020 /* TODO: Also verify bits beyond physical address width are 0 */
8021 if (!PAGE_ALIGNED(vmcs12->virtual_apic_page_addr))
8024 if (vmx->nested.virtual_apic_page) /* shouldn't happen */
8025 nested_release_page(vmx->nested.virtual_apic_page);
8026 vmx->nested.virtual_apic_page =
8027 nested_get_page(vcpu, vmcs12->virtual_apic_page_addr);
8030 * Failing the vm entry is _not_ what the processor does
8031 * but it's basically the only possibility we have.
8032 * We could still enter the guest if CR8 load exits are
8033 * enabled, CR8 store exits are enabled, and virtualize APIC
8034 * access is disabled; in this case the processor would never
8035 * use the TPR shadow and we could simply clear the bit from
8036 * the execution control. But such a configuration is useless,
8037 * so let's keep the code simple.
8039 if (!vmx->nested.virtual_apic_page)
8046 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
8048 u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
8049 struct vcpu_vmx *vmx = to_vmx(vcpu);
8051 if (vcpu->arch.virtual_tsc_khz == 0)
8054 /* Make sure short timeouts reliably trigger an immediate vmexit.
8055 * hrtimer_start does not guarantee this. */
8056 if (preemption_timeout <= 1) {
8057 vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
8061 preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
8062 preemption_timeout *= 1000000;
8063 do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
8064 hrtimer_start(&vmx->nested.preemption_timer,
8065 ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
8069 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
8070 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
8071 * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
8072 * guest in a way that will both be appropriate to L1's requests, and our
8073 * needs. In addition to modifying the active vmcs (which is vmcs02), this
8074 * function also has additional necessary side-effects, like setting various
8075 * vcpu->arch fields.
8077 static void prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
8079 struct vcpu_vmx *vmx = to_vmx(vcpu);
8082 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
8083 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
8084 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
8085 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
8086 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
8087 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
8088 vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
8089 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
8090 vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
8091 vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
8092 vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
8093 vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
8094 vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
8095 vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
8096 vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
8097 vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
8098 vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
8099 vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
8100 vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
8101 vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
8102 vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
8103 vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
8104 vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
8105 vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
8106 vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
8107 vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
8108 vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
8109 vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
8110 vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
8111 vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
8112 vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
8113 vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
8114 vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
8115 vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
8116 vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
8117 vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
8119 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
8120 kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
8121 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
8123 kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
8124 vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
8126 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
8127 vmcs12->vm_entry_intr_info_field);
8128 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
8129 vmcs12->vm_entry_exception_error_code);
8130 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
8131 vmcs12->vm_entry_instruction_len);
8132 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
8133 vmcs12->guest_interruptibility_info);
8134 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
8135 vmx_set_rflags(vcpu, vmcs12->guest_rflags);
8136 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
8137 vmcs12->guest_pending_dbg_exceptions);
8138 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
8139 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
8141 vmcs_write64(VMCS_LINK_POINTER, -1ull);
8143 exec_control = vmcs12->pin_based_vm_exec_control;
8144 exec_control |= vmcs_config.pin_based_exec_ctrl;
8145 exec_control &= ~(PIN_BASED_VMX_PREEMPTION_TIMER |
8146 PIN_BASED_POSTED_INTR);
8147 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
8149 vmx->nested.preemption_timer_expired = false;
8150 if (nested_cpu_has_preemption_timer(vmcs12))
8151 vmx_start_preemption_timer(vcpu);
8154 * Whether page-faults are trapped is determined by a combination of
8155 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
8156 * If enable_ept, L0 doesn't care about page faults and we should
8157 * set all of these to L1's desires. However, if !enable_ept, L0 does
8158 * care about (at least some) page faults, and because it is not easy
8159 * (if at all possible?) to merge L0 and L1's desires, we simply ask
8160 * to exit on each and every L2 page fault. This is done by setting
8161 * MASK=MATCH=0 and (see below) EB.PF=1.
8162 * Note that below we don't need special code to set EB.PF beyond the
8163 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
8164 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
8165 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
8167 * A problem with this approach (when !enable_ept) is that L1 may be
8168 * injected with more page faults than it asked for. This could have
8169 * caused problems, but in practice existing hypervisors don't care.
8170 * To fix this, we will need to emulate the PFEC checking (on the L1
8171 * page tables), using walk_addr(), when injecting PFs to L1.
8173 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
8174 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
8175 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
8176 enable_ept ? vmcs12->page_fault_error_code_match : 0);
8178 if (cpu_has_secondary_exec_ctrls()) {
8179 exec_control = vmx_secondary_exec_control(vmx);
8180 if (!vmx->rdtscp_enabled)
8181 exec_control &= ~SECONDARY_EXEC_RDTSCP;
8182 /* Take the following fields only from vmcs12 */
8183 exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
8184 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
8185 SECONDARY_EXEC_APIC_REGISTER_VIRT);
8186 if (nested_cpu_has(vmcs12,
8187 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
8188 exec_control |= vmcs12->secondary_vm_exec_control;
8190 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) {
8192 * If translation failed, no matter: This feature asks
8193 * to exit when accessing the given address, and if it
8194 * can never be accessed, this feature won't do
8197 if (!vmx->nested.apic_access_page)
8199 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8201 vmcs_write64(APIC_ACCESS_ADDR,
8202 page_to_phys(vmx->nested.apic_access_page));
8203 } else if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm)) {
8205 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8206 kvm_vcpu_reload_apic_access_page(vcpu);
8209 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
8214 * Set host-state according to L0's settings (vmcs12 is irrelevant here)
8215 * Some constant fields are set here by vmx_set_constant_host_state().
8216 * Other fields are different per CPU, and will be set later when
8217 * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
8219 vmx_set_constant_host_state(vmx);
8222 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
8223 * entry, but only if the current (host) sp changed from the value
8224 * we wrote last (vmx->host_rsp). This cache is no longer relevant
8225 * if we switch vmcs, and rather than hold a separate cache per vmcs,
8226 * here we just force the write to happen on entry.
8230 exec_control = vmx_exec_control(vmx); /* L0's desires */
8231 exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
8232 exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
8233 exec_control &= ~CPU_BASED_TPR_SHADOW;
8234 exec_control |= vmcs12->cpu_based_vm_exec_control;
8236 if (exec_control & CPU_BASED_TPR_SHADOW) {
8237 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
8238 page_to_phys(vmx->nested.virtual_apic_page));
8239 vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
8243 * Merging of IO and MSR bitmaps not currently supported.
8244 * Rather, exit every time.
8246 exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
8247 exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
8248 exec_control |= CPU_BASED_UNCOND_IO_EXITING;
8250 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
8252 /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
8253 * bitwise-or of what L1 wants to trap for L2, and what we want to
8254 * trap. Note that CR0.TS also needs updating - we do this later.
8256 update_exception_bitmap(vcpu);
8257 vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
8258 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
8260 /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
8261 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
8262 * bits are further modified by vmx_set_efer() below.
8264 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
8266 /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
8267 * emulated by vmx_set_efer(), below.
8269 vm_entry_controls_init(vmx,
8270 (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER &
8271 ~VM_ENTRY_IA32E_MODE) |
8272 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
8274 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) {
8275 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
8276 vcpu->arch.pat = vmcs12->guest_ia32_pat;
8277 } else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
8278 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
8281 set_cr4_guest_host_mask(vmx);
8283 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)
8284 vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
8286 if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
8287 vmcs_write64(TSC_OFFSET,
8288 vmx->nested.vmcs01_tsc_offset + vmcs12->tsc_offset);
8290 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
8294 * Trivially support vpid by letting L2s share their parent
8295 * L1's vpid. TODO: move to a more elaborate solution, giving
8296 * each L2 its own vpid and exposing the vpid feature to L1.
8298 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
8299 vmx_flush_tlb(vcpu);
8302 if (nested_cpu_has_ept(vmcs12)) {
8303 kvm_mmu_unload(vcpu);
8304 nested_ept_init_mmu_context(vcpu);
8307 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)
8308 vcpu->arch.efer = vmcs12->guest_ia32_efer;
8309 else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
8310 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
8312 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
8313 /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
8314 vmx_set_efer(vcpu, vcpu->arch.efer);
8317 * This sets GUEST_CR0 to vmcs12->guest_cr0, with possibly a modified
8318 * TS bit (for lazy fpu) and bits which we consider mandatory enabled.
8319 * The CR0_READ_SHADOW is what L2 should have expected to read given
8320 * the specifications by L1; It's not enough to take
8321 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
8322 * have more bits than L1 expected.
8324 vmx_set_cr0(vcpu, vmcs12->guest_cr0);
8325 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
8327 vmx_set_cr4(vcpu, vmcs12->guest_cr4);
8328 vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
8330 /* shadow page tables on either EPT or shadow page tables */
8331 kvm_set_cr3(vcpu, vmcs12->guest_cr3);
8332 kvm_mmu_reset_context(vcpu);
8335 vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
8338 * L1 may access the L2's PDPTR, so save them to construct vmcs12
8341 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
8342 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
8343 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
8344 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
8347 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
8348 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
8352 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
8353 * for running an L2 nested guest.
8355 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
8357 struct vmcs12 *vmcs12;
8358 struct vcpu_vmx *vmx = to_vmx(vcpu);
8360 struct loaded_vmcs *vmcs02;
8363 if (!nested_vmx_check_permission(vcpu) ||
8364 !nested_vmx_check_vmcs12(vcpu))
8367 skip_emulated_instruction(vcpu);
8368 vmcs12 = get_vmcs12(vcpu);
8370 if (enable_shadow_vmcs)
8371 copy_shadow_to_vmcs12(vmx);
8374 * The nested entry process starts with enforcing various prerequisites
8375 * on vmcs12 as required by the Intel SDM, and act appropriately when
8376 * they fail: As the SDM explains, some conditions should cause the
8377 * instruction to fail, while others will cause the instruction to seem
8378 * to succeed, but return an EXIT_REASON_INVALID_STATE.
8379 * To speed up the normal (success) code path, we should avoid checking
8380 * for misconfigurations which will anyway be caught by the processor
8381 * when using the merged vmcs02.
8383 if (vmcs12->launch_state == launch) {
8384 nested_vmx_failValid(vcpu,
8385 launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
8386 : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
8390 if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
8391 vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT) {
8392 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
8396 if ((vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_MSR_BITMAPS) &&
8397 !PAGE_ALIGNED(vmcs12->msr_bitmap)) {
8398 /*TODO: Also verify bits beyond physical address width are 0*/
8399 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
8403 if (!nested_get_vmcs12_pages(vcpu, vmcs12)) {
8404 /*TODO: Also verify bits beyond physical address width are 0*/
8405 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
8409 if (vmcs12->vm_entry_msr_load_count > 0 ||
8410 vmcs12->vm_exit_msr_load_count > 0 ||
8411 vmcs12->vm_exit_msr_store_count > 0) {
8412 pr_warn_ratelimited("%s: VMCS MSR_{LOAD,STORE} unsupported\n",
8414 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
8418 if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
8419 nested_vmx_true_procbased_ctls_low,
8420 nested_vmx_procbased_ctls_high) ||
8421 !vmx_control_verify(vmcs12->secondary_vm_exec_control,
8422 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high) ||
8423 !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
8424 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high) ||
8425 !vmx_control_verify(vmcs12->vm_exit_controls,
8426 nested_vmx_true_exit_ctls_low,
8427 nested_vmx_exit_ctls_high) ||
8428 !vmx_control_verify(vmcs12->vm_entry_controls,
8429 nested_vmx_true_entry_ctls_low,
8430 nested_vmx_entry_ctls_high))
8432 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
8436 if (((vmcs12->host_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
8437 ((vmcs12->host_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
8438 nested_vmx_failValid(vcpu,
8439 VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
8443 if (!nested_cr0_valid(vmcs12, vmcs12->guest_cr0) ||
8444 ((vmcs12->guest_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
8445 nested_vmx_entry_failure(vcpu, vmcs12,
8446 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
8449 if (vmcs12->vmcs_link_pointer != -1ull) {
8450 nested_vmx_entry_failure(vcpu, vmcs12,
8451 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_VMCS_LINK_PTR);
8456 * If the load IA32_EFER VM-entry control is 1, the following checks
8457 * are performed on the field for the IA32_EFER MSR:
8458 * - Bits reserved in the IA32_EFER MSR must be 0.
8459 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
8460 * the IA-32e mode guest VM-exit control. It must also be identical
8461 * to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
8464 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER) {
8465 ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
8466 if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
8467 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
8468 ((vmcs12->guest_cr0 & X86_CR0_PG) &&
8469 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))) {
8470 nested_vmx_entry_failure(vcpu, vmcs12,
8471 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
8477 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
8478 * IA32_EFER MSR must be 0 in the field for that register. In addition,
8479 * the values of the LMA and LME bits in the field must each be that of
8480 * the host address-space size VM-exit control.
8482 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
8483 ia32e = (vmcs12->vm_exit_controls &
8484 VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
8485 if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
8486 ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
8487 ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)) {
8488 nested_vmx_entry_failure(vcpu, vmcs12,
8489 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
8495 * We're finally done with prerequisite checking, and can start with
8499 vmcs02 = nested_get_current_vmcs02(vmx);
8503 enter_guest_mode(vcpu);
8505 vmx->nested.vmcs01_tsc_offset = vmcs_read64(TSC_OFFSET);
8507 if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
8508 vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
8511 vmx->loaded_vmcs = vmcs02;
8513 vmx_vcpu_load(vcpu, cpu);
8517 vmx_segment_cache_clear(vmx);
8519 vmcs12->launch_state = 1;
8521 prepare_vmcs02(vcpu, vmcs12);
8523 if (vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT)
8524 return kvm_emulate_halt(vcpu);
8526 vmx->nested.nested_run_pending = 1;
8529 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
8530 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
8531 * returned as far as L1 is concerned. It will only return (and set
8532 * the success flag) when L2 exits (see nested_vmx_vmexit()).
8538 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
8539 * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
8540 * This function returns the new value we should put in vmcs12.guest_cr0.
8541 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
8542 * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
8543 * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
8544 * didn't trap the bit, because if L1 did, so would L0).
8545 * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have
8546 * been modified by L2, and L1 knows it. So just leave the old value of
8547 * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
8548 * isn't relevant, because if L0 traps this bit it can set it to anything.
8549 * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
8550 * changed these bits, and therefore they need to be updated, but L0
8551 * didn't necessarily allow them to be changed in GUEST_CR0 - and rather
8552 * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
8554 static inline unsigned long
8555 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
8558 /*1*/ (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
8559 /*2*/ (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
8560 /*3*/ (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
8561 vcpu->arch.cr0_guest_owned_bits));
8564 static inline unsigned long
8565 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
8568 /*1*/ (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
8569 /*2*/ (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
8570 /*3*/ (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
8571 vcpu->arch.cr4_guest_owned_bits));
8574 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
8575 struct vmcs12 *vmcs12)
8580 if (vcpu->arch.exception.pending && vcpu->arch.exception.reinject) {
8581 nr = vcpu->arch.exception.nr;
8582 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
8584 if (kvm_exception_is_soft(nr)) {
8585 vmcs12->vm_exit_instruction_len =
8586 vcpu->arch.event_exit_inst_len;
8587 idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
8589 idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
8591 if (vcpu->arch.exception.has_error_code) {
8592 idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
8593 vmcs12->idt_vectoring_error_code =
8594 vcpu->arch.exception.error_code;
8597 vmcs12->idt_vectoring_info_field = idt_vectoring;
8598 } else if (vcpu->arch.nmi_injected) {
8599 vmcs12->idt_vectoring_info_field =
8600 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
8601 } else if (vcpu->arch.interrupt.pending) {
8602 nr = vcpu->arch.interrupt.nr;
8603 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
8605 if (vcpu->arch.interrupt.soft) {
8606 idt_vectoring |= INTR_TYPE_SOFT_INTR;
8607 vmcs12->vm_entry_instruction_len =
8608 vcpu->arch.event_exit_inst_len;
8610 idt_vectoring |= INTR_TYPE_EXT_INTR;
8612 vmcs12->idt_vectoring_info_field = idt_vectoring;
8616 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
8618 struct vcpu_vmx *vmx = to_vmx(vcpu);
8620 if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
8621 vmx->nested.preemption_timer_expired) {
8622 if (vmx->nested.nested_run_pending)
8624 nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
8628 if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
8629 if (vmx->nested.nested_run_pending ||
8630 vcpu->arch.interrupt.pending)
8632 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
8633 NMI_VECTOR | INTR_TYPE_NMI_INTR |
8634 INTR_INFO_VALID_MASK, 0);
8636 * The NMI-triggered VM exit counts as injection:
8637 * clear this one and block further NMIs.
8639 vcpu->arch.nmi_pending = 0;
8640 vmx_set_nmi_mask(vcpu, true);
8644 if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
8645 nested_exit_on_intr(vcpu)) {
8646 if (vmx->nested.nested_run_pending)
8648 nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
8654 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
8657 hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
8660 if (ktime_to_ns(remaining) <= 0)
8663 value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
8664 do_div(value, 1000000);
8665 return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
8669 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
8670 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
8671 * and this function updates it to reflect the changes to the guest state while
8672 * L2 was running (and perhaps made some exits which were handled directly by L0
8673 * without going back to L1), and to reflect the exit reason.
8674 * Note that we do not have to copy here all VMCS fields, just those that
8675 * could have changed by the L2 guest or the exit - i.e., the guest-state and
8676 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
8677 * which already writes to vmcs12 directly.
8679 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
8680 u32 exit_reason, u32 exit_intr_info,
8681 unsigned long exit_qualification)
8683 /* update guest state fields: */
8684 vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
8685 vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
8687 vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
8688 vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
8689 vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
8691 vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
8692 vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
8693 vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
8694 vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
8695 vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
8696 vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
8697 vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
8698 vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
8699 vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
8700 vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
8701 vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
8702 vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
8703 vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
8704 vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
8705 vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
8706 vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
8707 vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
8708 vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
8709 vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
8710 vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
8711 vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
8712 vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
8713 vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
8714 vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
8715 vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
8716 vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
8717 vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
8718 vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
8719 vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
8720 vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
8721 vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
8722 vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
8723 vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
8724 vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
8725 vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
8726 vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
8728 vmcs12->guest_interruptibility_info =
8729 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
8730 vmcs12->guest_pending_dbg_exceptions =
8731 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
8732 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
8733 vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
8735 vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
8737 if (nested_cpu_has_preemption_timer(vmcs12)) {
8738 if (vmcs12->vm_exit_controls &
8739 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
8740 vmcs12->vmx_preemption_timer_value =
8741 vmx_get_preemption_timer_value(vcpu);
8742 hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
8746 * In some cases (usually, nested EPT), L2 is allowed to change its
8747 * own CR3 without exiting. If it has changed it, we must keep it.
8748 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
8749 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
8751 * Additionally, restore L2's PDPTR to vmcs12.
8754 vmcs12->guest_cr3 = vmcs_read64(GUEST_CR3);
8755 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
8756 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
8757 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
8758 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
8761 vmcs12->vm_entry_controls =
8762 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
8763 (vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
8765 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) {
8766 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
8767 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
8770 /* TODO: These cannot have changed unless we have MSR bitmaps and
8771 * the relevant bit asks not to trap the change */
8772 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
8773 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
8774 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
8775 vmcs12->guest_ia32_efer = vcpu->arch.efer;
8776 vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
8777 vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
8778 vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
8779 if (vmx_mpx_supported())
8780 vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
8782 /* update exit information fields: */
8784 vmcs12->vm_exit_reason = exit_reason;
8785 vmcs12->exit_qualification = exit_qualification;
8787 vmcs12->vm_exit_intr_info = exit_intr_info;
8788 if ((vmcs12->vm_exit_intr_info &
8789 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
8790 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK))
8791 vmcs12->vm_exit_intr_error_code =
8792 vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
8793 vmcs12->idt_vectoring_info_field = 0;
8794 vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
8795 vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
8797 if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
8798 /* vm_entry_intr_info_field is cleared on exit. Emulate this
8799 * instead of reading the real value. */
8800 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
8803 * Transfer the event that L0 or L1 may wanted to inject into
8804 * L2 to IDT_VECTORING_INFO_FIELD.
8806 vmcs12_save_pending_event(vcpu, vmcs12);
8810 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
8811 * preserved above and would only end up incorrectly in L1.
8813 vcpu->arch.nmi_injected = false;
8814 kvm_clear_exception_queue(vcpu);
8815 kvm_clear_interrupt_queue(vcpu);
8819 * A part of what we need to when the nested L2 guest exits and we want to
8820 * run its L1 parent, is to reset L1's guest state to the host state specified
8822 * This function is to be called not only on normal nested exit, but also on
8823 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
8824 * Failures During or After Loading Guest State").
8825 * This function should be called when the active VMCS is L1's (vmcs01).
8827 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
8828 struct vmcs12 *vmcs12)
8830 struct kvm_segment seg;
8832 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
8833 vcpu->arch.efer = vmcs12->host_ia32_efer;
8834 else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
8835 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
8837 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
8838 vmx_set_efer(vcpu, vcpu->arch.efer);
8840 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
8841 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
8842 vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
8844 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
8845 * actually changed, because it depends on the current state of
8846 * fpu_active (which may have changed).
8847 * Note that vmx_set_cr0 refers to efer set above.
8849 vmx_set_cr0(vcpu, vmcs12->host_cr0);
8851 * If we did fpu_activate()/fpu_deactivate() during L2's run, we need
8852 * to apply the same changes to L1's vmcs. We just set cr0 correctly,
8853 * but we also need to update cr0_guest_host_mask and exception_bitmap.
8855 update_exception_bitmap(vcpu);
8856 vcpu->arch.cr0_guest_owned_bits = (vcpu->fpu_active ? X86_CR0_TS : 0);
8857 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
8860 * Note that CR4_GUEST_HOST_MASK is already set in the original vmcs01
8861 * (KVM doesn't change it)- no reason to call set_cr4_guest_host_mask();
8863 vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
8864 kvm_set_cr4(vcpu, vmcs12->host_cr4);
8866 nested_ept_uninit_mmu_context(vcpu);
8868 kvm_set_cr3(vcpu, vmcs12->host_cr3);
8869 kvm_mmu_reset_context(vcpu);
8872 vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
8876 * Trivially support vpid by letting L2s share their parent
8877 * L1's vpid. TODO: move to a more elaborate solution, giving
8878 * each L2 its own vpid and exposing the vpid feature to L1.
8880 vmx_flush_tlb(vcpu);
8884 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
8885 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
8886 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
8887 vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
8888 vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
8890 /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1. */
8891 if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
8892 vmcs_write64(GUEST_BNDCFGS, 0);
8894 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
8895 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
8896 vcpu->arch.pat = vmcs12->host_ia32_pat;
8898 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
8899 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
8900 vmcs12->host_ia32_perf_global_ctrl);
8902 /* Set L1 segment info according to Intel SDM
8903 27.5.2 Loading Host Segment and Descriptor-Table Registers */
8904 seg = (struct kvm_segment) {
8906 .limit = 0xFFFFFFFF,
8907 .selector = vmcs12->host_cs_selector,
8913 if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
8917 vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
8918 seg = (struct kvm_segment) {
8920 .limit = 0xFFFFFFFF,
8927 seg.selector = vmcs12->host_ds_selector;
8928 vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
8929 seg.selector = vmcs12->host_es_selector;
8930 vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
8931 seg.selector = vmcs12->host_ss_selector;
8932 vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
8933 seg.selector = vmcs12->host_fs_selector;
8934 seg.base = vmcs12->host_fs_base;
8935 vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
8936 seg.selector = vmcs12->host_gs_selector;
8937 seg.base = vmcs12->host_gs_base;
8938 vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
8939 seg = (struct kvm_segment) {
8940 .base = vmcs12->host_tr_base,
8942 .selector = vmcs12->host_tr_selector,
8946 vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
8948 kvm_set_dr(vcpu, 7, 0x400);
8949 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
8953 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
8954 * and modify vmcs12 to make it see what it would expect to see there if
8955 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
8957 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
8959 unsigned long exit_qualification)
8961 struct vcpu_vmx *vmx = to_vmx(vcpu);
8962 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8964 /* trying to cancel vmlaunch/vmresume is a bug */
8965 WARN_ON_ONCE(vmx->nested.nested_run_pending);
8967 leave_guest_mode(vcpu);
8968 prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
8969 exit_qualification);
8971 vmx_load_vmcs01(vcpu);
8973 if ((exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
8974 && nested_exit_intr_ack_set(vcpu)) {
8975 int irq = kvm_cpu_get_interrupt(vcpu);
8977 vmcs12->vm_exit_intr_info = irq |
8978 INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
8981 trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
8982 vmcs12->exit_qualification,
8983 vmcs12->idt_vectoring_info_field,
8984 vmcs12->vm_exit_intr_info,
8985 vmcs12->vm_exit_intr_error_code,
8988 vm_entry_controls_init(vmx, vmcs_read32(VM_ENTRY_CONTROLS));
8989 vm_exit_controls_init(vmx, vmcs_read32(VM_EXIT_CONTROLS));
8990 vmx_segment_cache_clear(vmx);
8992 /* if no vmcs02 cache requested, remove the one we used */
8993 if (VMCS02_POOL_SIZE == 0)
8994 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
8996 load_vmcs12_host_state(vcpu, vmcs12);
8998 /* Update TSC_OFFSET if TSC was changed while L2 ran */
8999 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
9001 /* This is needed for same reason as it was needed in prepare_vmcs02 */
9004 /* Unpin physical memory we referred to in vmcs02 */
9005 if (vmx->nested.apic_access_page) {
9006 nested_release_page(vmx->nested.apic_access_page);
9007 vmx->nested.apic_access_page = NULL;
9009 if (vmx->nested.virtual_apic_page) {
9010 nested_release_page(vmx->nested.virtual_apic_page);
9011 vmx->nested.virtual_apic_page = NULL;
9015 * We are now running in L2, mmu_notifier will force to reload the
9016 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
9018 kvm_vcpu_reload_apic_access_page(vcpu);
9021 * Exiting from L2 to L1, we're now back to L1 which thinks it just
9022 * finished a VMLAUNCH or VMRESUME instruction, so we need to set the
9023 * success or failure flag accordingly.
9025 if (unlikely(vmx->fail)) {
9027 nested_vmx_failValid(vcpu, vmcs_read32(VM_INSTRUCTION_ERROR));
9029 nested_vmx_succeed(vcpu);
9030 if (enable_shadow_vmcs)
9031 vmx->nested.sync_shadow_vmcs = true;
9033 /* in case we halted in L2 */
9034 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9038 * Forcibly leave nested mode in order to be able to reset the VCPU later on.
9040 static void vmx_leave_nested(struct kvm_vcpu *vcpu)
9042 if (is_guest_mode(vcpu))
9043 nested_vmx_vmexit(vcpu, -1, 0, 0);
9044 free_nested(to_vmx(vcpu));
9048 * L1's failure to enter L2 is a subset of a normal exit, as explained in
9049 * 23.7 "VM-entry failures during or after loading guest state" (this also
9050 * lists the acceptable exit-reason and exit-qualification parameters).
9051 * It should only be called before L2 actually succeeded to run, and when
9052 * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
9054 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
9055 struct vmcs12 *vmcs12,
9056 u32 reason, unsigned long qualification)
9058 load_vmcs12_host_state(vcpu, vmcs12);
9059 vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
9060 vmcs12->exit_qualification = qualification;
9061 nested_vmx_succeed(vcpu);
9062 if (enable_shadow_vmcs)
9063 to_vmx(vcpu)->nested.sync_shadow_vmcs = true;
9066 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
9067 struct x86_instruction_info *info,
9068 enum x86_intercept_stage stage)
9070 return X86EMUL_CONTINUE;
9073 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
9076 shrink_ple_window(vcpu);
9079 static struct kvm_x86_ops vmx_x86_ops = {
9080 .cpu_has_kvm_support = cpu_has_kvm_support,
9081 .disabled_by_bios = vmx_disabled_by_bios,
9082 .hardware_setup = hardware_setup,
9083 .hardware_unsetup = hardware_unsetup,
9084 .check_processor_compatibility = vmx_check_processor_compat,
9085 .hardware_enable = hardware_enable,
9086 .hardware_disable = hardware_disable,
9087 .cpu_has_accelerated_tpr = report_flexpriority,
9089 .vcpu_create = vmx_create_vcpu,
9090 .vcpu_free = vmx_free_vcpu,
9091 .vcpu_reset = vmx_vcpu_reset,
9093 .prepare_guest_switch = vmx_save_host_state,
9094 .vcpu_load = vmx_vcpu_load,
9095 .vcpu_put = vmx_vcpu_put,
9097 .update_db_bp_intercept = update_exception_bitmap,
9098 .get_msr = vmx_get_msr,
9099 .set_msr = vmx_set_msr,
9100 .get_segment_base = vmx_get_segment_base,
9101 .get_segment = vmx_get_segment,
9102 .set_segment = vmx_set_segment,
9103 .get_cpl = vmx_get_cpl,
9104 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
9105 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
9106 .decache_cr3 = vmx_decache_cr3,
9107 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
9108 .set_cr0 = vmx_set_cr0,
9109 .set_cr3 = vmx_set_cr3,
9110 .set_cr4 = vmx_set_cr4,
9111 .set_efer = vmx_set_efer,
9112 .get_idt = vmx_get_idt,
9113 .set_idt = vmx_set_idt,
9114 .get_gdt = vmx_get_gdt,
9115 .set_gdt = vmx_set_gdt,
9116 .get_dr6 = vmx_get_dr6,
9117 .set_dr6 = vmx_set_dr6,
9118 .set_dr7 = vmx_set_dr7,
9119 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
9120 .cache_reg = vmx_cache_reg,
9121 .get_rflags = vmx_get_rflags,
9122 .set_rflags = vmx_set_rflags,
9123 .fpu_deactivate = vmx_fpu_deactivate,
9125 .tlb_flush = vmx_flush_tlb,
9127 .run = vmx_vcpu_run,
9128 .handle_exit = vmx_handle_exit,
9129 .skip_emulated_instruction = skip_emulated_instruction,
9130 .set_interrupt_shadow = vmx_set_interrupt_shadow,
9131 .get_interrupt_shadow = vmx_get_interrupt_shadow,
9132 .patch_hypercall = vmx_patch_hypercall,
9133 .set_irq = vmx_inject_irq,
9134 .set_nmi = vmx_inject_nmi,
9135 .queue_exception = vmx_queue_exception,
9136 .cancel_injection = vmx_cancel_injection,
9137 .interrupt_allowed = vmx_interrupt_allowed,
9138 .nmi_allowed = vmx_nmi_allowed,
9139 .get_nmi_mask = vmx_get_nmi_mask,
9140 .set_nmi_mask = vmx_set_nmi_mask,
9141 .enable_nmi_window = enable_nmi_window,
9142 .enable_irq_window = enable_irq_window,
9143 .update_cr8_intercept = update_cr8_intercept,
9144 .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode,
9145 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
9146 .vm_has_apicv = vmx_vm_has_apicv,
9147 .load_eoi_exitmap = vmx_load_eoi_exitmap,
9148 .hwapic_irr_update = vmx_hwapic_irr_update,
9149 .hwapic_isr_update = vmx_hwapic_isr_update,
9150 .sync_pir_to_irr = vmx_sync_pir_to_irr,
9151 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
9153 .set_tss_addr = vmx_set_tss_addr,
9154 .get_tdp_level = get_ept_level,
9155 .get_mt_mask = vmx_get_mt_mask,
9157 .get_exit_info = vmx_get_exit_info,
9159 .get_lpage_level = vmx_get_lpage_level,
9161 .cpuid_update = vmx_cpuid_update,
9163 .rdtscp_supported = vmx_rdtscp_supported,
9164 .invpcid_supported = vmx_invpcid_supported,
9166 .set_supported_cpuid = vmx_set_supported_cpuid,
9168 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
9170 .set_tsc_khz = vmx_set_tsc_khz,
9171 .read_tsc_offset = vmx_read_tsc_offset,
9172 .write_tsc_offset = vmx_write_tsc_offset,
9173 .adjust_tsc_offset = vmx_adjust_tsc_offset,
9174 .compute_tsc_offset = vmx_compute_tsc_offset,
9175 .read_l1_tsc = vmx_read_l1_tsc,
9177 .set_tdp_cr3 = vmx_set_cr3,
9179 .check_intercept = vmx_check_intercept,
9180 .handle_external_intr = vmx_handle_external_intr,
9181 .mpx_supported = vmx_mpx_supported,
9183 .check_nested_events = vmx_check_nested_events,
9185 .sched_in = vmx_sched_in,
9188 static int __init vmx_init(void)
9192 rdmsrl_safe(MSR_EFER, &host_efer);
9194 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
9195 kvm_define_shared_msr(i, vmx_msr_index[i]);
9197 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
9198 if (!vmx_io_bitmap_a)
9203 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
9204 if (!vmx_io_bitmap_b)
9207 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
9208 if (!vmx_msr_bitmap_legacy)
9211 vmx_msr_bitmap_legacy_x2apic =
9212 (unsigned long *)__get_free_page(GFP_KERNEL);
9213 if (!vmx_msr_bitmap_legacy_x2apic)
9216 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
9217 if (!vmx_msr_bitmap_longmode)
9220 vmx_msr_bitmap_longmode_x2apic =
9221 (unsigned long *)__get_free_page(GFP_KERNEL);
9222 if (!vmx_msr_bitmap_longmode_x2apic)
9224 vmx_vmread_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
9225 if (!vmx_vmread_bitmap)
9228 vmx_vmwrite_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
9229 if (!vmx_vmwrite_bitmap)
9232 memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
9233 memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
9236 * Allow direct access to the PC debug port (it is often used for I/O
9237 * delays, but the vmexits simply slow things down).
9239 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
9240 clear_bit(0x80, vmx_io_bitmap_a);
9242 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
9244 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
9245 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
9247 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
9249 r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
9250 __alignof__(struct vcpu_vmx), THIS_MODULE);
9255 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
9256 crash_vmclear_local_loaded_vmcss);
9259 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
9260 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
9261 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
9262 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
9263 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
9264 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
9265 vmx_disable_intercept_for_msr(MSR_IA32_BNDCFGS, true);
9267 memcpy(vmx_msr_bitmap_legacy_x2apic,
9268 vmx_msr_bitmap_legacy, PAGE_SIZE);
9269 memcpy(vmx_msr_bitmap_longmode_x2apic,
9270 vmx_msr_bitmap_longmode, PAGE_SIZE);
9273 for (msr = 0x800; msr <= 0x8ff; msr++)
9274 vmx_disable_intercept_msr_read_x2apic(msr);
9276 /* According SDM, in x2apic mode, the whole id reg is used.
9277 * But in KVM, it only use the highest eight bits. Need to
9279 vmx_enable_intercept_msr_read_x2apic(0x802);
9281 vmx_enable_intercept_msr_read_x2apic(0x839);
9283 vmx_disable_intercept_msr_write_x2apic(0x808);
9285 vmx_disable_intercept_msr_write_x2apic(0x80b);
9287 vmx_disable_intercept_msr_write_x2apic(0x83f);
9291 kvm_mmu_set_mask_ptes(0ull,
9292 (enable_ept_ad_bits) ? VMX_EPT_ACCESS_BIT : 0ull,
9293 (enable_ept_ad_bits) ? VMX_EPT_DIRTY_BIT : 0ull,
9294 0ull, VMX_EPT_EXECUTABLE_MASK);
9295 ept_set_mmio_spte_mask();
9300 update_ple_window_actual_max();
9305 free_page((unsigned long)vmx_vmwrite_bitmap);
9307 free_page((unsigned long)vmx_vmread_bitmap);
9309 free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
9311 free_page((unsigned long)vmx_msr_bitmap_longmode);
9313 free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
9315 free_page((unsigned long)vmx_msr_bitmap_legacy);
9317 free_page((unsigned long)vmx_io_bitmap_b);
9319 free_page((unsigned long)vmx_io_bitmap_a);
9323 static void __exit vmx_exit(void)
9325 free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
9326 free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
9327 free_page((unsigned long)vmx_msr_bitmap_legacy);
9328 free_page((unsigned long)vmx_msr_bitmap_longmode);
9329 free_page((unsigned long)vmx_io_bitmap_b);
9330 free_page((unsigned long)vmx_io_bitmap_a);
9331 free_page((unsigned long)vmx_vmwrite_bitmap);
9332 free_page((unsigned long)vmx_vmread_bitmap);
9335 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
9342 module_init(vmx_init)
9343 module_exit(vmx_exit)