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),
724 static inline short vmcs_field_to_offset(unsigned long field)
726 BUILD_BUG_ON(ARRAY_SIZE(vmcs_field_to_offset_table) > SHRT_MAX);
728 if (field >= ARRAY_SIZE(vmcs_field_to_offset_table) ||
729 vmcs_field_to_offset_table[field] == 0)
732 return vmcs_field_to_offset_table[field];
735 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
737 return to_vmx(vcpu)->nested.current_vmcs12;
740 static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
742 struct page *page = gfn_to_page(vcpu->kvm, addr >> PAGE_SHIFT);
743 if (is_error_page(page))
749 static void nested_release_page(struct page *page)
751 kvm_release_page_dirty(page);
754 static void nested_release_page_clean(struct page *page)
756 kvm_release_page_clean(page);
759 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu);
760 static u64 construct_eptp(unsigned long root_hpa);
761 static void kvm_cpu_vmxon(u64 addr);
762 static void kvm_cpu_vmxoff(void);
763 static bool vmx_mpx_supported(void);
764 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
765 static void vmx_set_segment(struct kvm_vcpu *vcpu,
766 struct kvm_segment *var, int seg);
767 static void vmx_get_segment(struct kvm_vcpu *vcpu,
768 struct kvm_segment *var, int seg);
769 static bool guest_state_valid(struct kvm_vcpu *vcpu);
770 static u32 vmx_segment_access_rights(struct kvm_segment *var);
771 static void vmx_sync_pir_to_irr_dummy(struct kvm_vcpu *vcpu);
772 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx);
773 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx);
774 static int alloc_identity_pagetable(struct kvm *kvm);
776 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
777 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
779 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
780 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
782 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
783 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
785 static unsigned long *vmx_io_bitmap_a;
786 static unsigned long *vmx_io_bitmap_b;
787 static unsigned long *vmx_msr_bitmap_legacy;
788 static unsigned long *vmx_msr_bitmap_longmode;
789 static unsigned long *vmx_msr_bitmap_legacy_x2apic;
790 static unsigned long *vmx_msr_bitmap_longmode_x2apic;
791 static unsigned long *vmx_vmread_bitmap;
792 static unsigned long *vmx_vmwrite_bitmap;
794 static bool cpu_has_load_ia32_efer;
795 static bool cpu_has_load_perf_global_ctrl;
797 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
798 static DEFINE_SPINLOCK(vmx_vpid_lock);
800 static struct vmcs_config {
804 u32 pin_based_exec_ctrl;
805 u32 cpu_based_exec_ctrl;
806 u32 cpu_based_2nd_exec_ctrl;
811 static struct vmx_capability {
816 #define VMX_SEGMENT_FIELD(seg) \
817 [VCPU_SREG_##seg] = { \
818 .selector = GUEST_##seg##_SELECTOR, \
819 .base = GUEST_##seg##_BASE, \
820 .limit = GUEST_##seg##_LIMIT, \
821 .ar_bytes = GUEST_##seg##_AR_BYTES, \
824 static const struct kvm_vmx_segment_field {
829 } kvm_vmx_segment_fields[] = {
830 VMX_SEGMENT_FIELD(CS),
831 VMX_SEGMENT_FIELD(DS),
832 VMX_SEGMENT_FIELD(ES),
833 VMX_SEGMENT_FIELD(FS),
834 VMX_SEGMENT_FIELD(GS),
835 VMX_SEGMENT_FIELD(SS),
836 VMX_SEGMENT_FIELD(TR),
837 VMX_SEGMENT_FIELD(LDTR),
840 static u64 host_efer;
842 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
845 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
846 * away by decrementing the array size.
848 static const u32 vmx_msr_index[] = {
850 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
852 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
855 static inline bool is_page_fault(u32 intr_info)
857 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
858 INTR_INFO_VALID_MASK)) ==
859 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
862 static inline bool is_no_device(u32 intr_info)
864 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
865 INTR_INFO_VALID_MASK)) ==
866 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
869 static inline bool is_invalid_opcode(u32 intr_info)
871 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
872 INTR_INFO_VALID_MASK)) ==
873 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
876 static inline bool is_external_interrupt(u32 intr_info)
878 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
879 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
882 static inline bool is_machine_check(u32 intr_info)
884 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
885 INTR_INFO_VALID_MASK)) ==
886 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
889 static inline bool cpu_has_vmx_msr_bitmap(void)
891 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
894 static inline bool cpu_has_vmx_tpr_shadow(void)
896 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
899 static inline bool vm_need_tpr_shadow(struct kvm *kvm)
901 return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
904 static inline bool cpu_has_secondary_exec_ctrls(void)
906 return vmcs_config.cpu_based_exec_ctrl &
907 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
910 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
912 return vmcs_config.cpu_based_2nd_exec_ctrl &
913 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
916 static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
918 return vmcs_config.cpu_based_2nd_exec_ctrl &
919 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
922 static inline bool cpu_has_vmx_apic_register_virt(void)
924 return vmcs_config.cpu_based_2nd_exec_ctrl &
925 SECONDARY_EXEC_APIC_REGISTER_VIRT;
928 static inline bool cpu_has_vmx_virtual_intr_delivery(void)
930 return vmcs_config.cpu_based_2nd_exec_ctrl &
931 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
934 static inline bool cpu_has_vmx_posted_intr(void)
936 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_POSTED_INTR;
939 static inline bool cpu_has_vmx_apicv(void)
941 return cpu_has_vmx_apic_register_virt() &&
942 cpu_has_vmx_virtual_intr_delivery() &&
943 cpu_has_vmx_posted_intr();
946 static inline bool cpu_has_vmx_flexpriority(void)
948 return cpu_has_vmx_tpr_shadow() &&
949 cpu_has_vmx_virtualize_apic_accesses();
952 static inline bool cpu_has_vmx_ept_execute_only(void)
954 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
957 static inline bool cpu_has_vmx_eptp_uncacheable(void)
959 return vmx_capability.ept & VMX_EPTP_UC_BIT;
962 static inline bool cpu_has_vmx_eptp_writeback(void)
964 return vmx_capability.ept & VMX_EPTP_WB_BIT;
967 static inline bool cpu_has_vmx_ept_2m_page(void)
969 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
972 static inline bool cpu_has_vmx_ept_1g_page(void)
974 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
977 static inline bool cpu_has_vmx_ept_4levels(void)
979 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
982 static inline bool cpu_has_vmx_ept_ad_bits(void)
984 return vmx_capability.ept & VMX_EPT_AD_BIT;
987 static inline bool cpu_has_vmx_invept_context(void)
989 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
992 static inline bool cpu_has_vmx_invept_global(void)
994 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
997 static inline bool cpu_has_vmx_invvpid_single(void)
999 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
1002 static inline bool cpu_has_vmx_invvpid_global(void)
1004 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
1007 static inline bool cpu_has_vmx_ept(void)
1009 return vmcs_config.cpu_based_2nd_exec_ctrl &
1010 SECONDARY_EXEC_ENABLE_EPT;
1013 static inline bool cpu_has_vmx_unrestricted_guest(void)
1015 return vmcs_config.cpu_based_2nd_exec_ctrl &
1016 SECONDARY_EXEC_UNRESTRICTED_GUEST;
1019 static inline bool cpu_has_vmx_ple(void)
1021 return vmcs_config.cpu_based_2nd_exec_ctrl &
1022 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
1025 static inline bool vm_need_virtualize_apic_accesses(struct kvm *kvm)
1027 return flexpriority_enabled && irqchip_in_kernel(kvm);
1030 static inline bool cpu_has_vmx_vpid(void)
1032 return vmcs_config.cpu_based_2nd_exec_ctrl &
1033 SECONDARY_EXEC_ENABLE_VPID;
1036 static inline bool cpu_has_vmx_rdtscp(void)
1038 return vmcs_config.cpu_based_2nd_exec_ctrl &
1039 SECONDARY_EXEC_RDTSCP;
1042 static inline bool cpu_has_vmx_invpcid(void)
1044 return vmcs_config.cpu_based_2nd_exec_ctrl &
1045 SECONDARY_EXEC_ENABLE_INVPCID;
1048 static inline bool cpu_has_virtual_nmis(void)
1050 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
1053 static inline bool cpu_has_vmx_wbinvd_exit(void)
1055 return vmcs_config.cpu_based_2nd_exec_ctrl &
1056 SECONDARY_EXEC_WBINVD_EXITING;
1059 static inline bool cpu_has_vmx_shadow_vmcs(void)
1062 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
1063 /* check if the cpu supports writing r/o exit information fields */
1064 if (!(vmx_msr & MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS))
1067 return vmcs_config.cpu_based_2nd_exec_ctrl &
1068 SECONDARY_EXEC_SHADOW_VMCS;
1071 static inline bool report_flexpriority(void)
1073 return flexpriority_enabled;
1076 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
1078 return vmcs12->cpu_based_vm_exec_control & bit;
1081 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
1083 return (vmcs12->cpu_based_vm_exec_control &
1084 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
1085 (vmcs12->secondary_vm_exec_control & bit);
1088 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12)
1090 return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
1093 static inline bool nested_cpu_has_preemption_timer(struct vmcs12 *vmcs12)
1095 return vmcs12->pin_based_vm_exec_control &
1096 PIN_BASED_VMX_PREEMPTION_TIMER;
1099 static inline int nested_cpu_has_ept(struct vmcs12 *vmcs12)
1101 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_EPT);
1104 static inline bool is_exception(u32 intr_info)
1106 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1107 == (INTR_TYPE_HARD_EXCEPTION | INTR_INFO_VALID_MASK);
1110 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
1112 unsigned long exit_qualification);
1113 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
1114 struct vmcs12 *vmcs12,
1115 u32 reason, unsigned long qualification);
1117 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
1121 for (i = 0; i < vmx->nmsrs; ++i)
1122 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
1127 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
1133 } operand = { vpid, 0, gva };
1135 asm volatile (__ex(ASM_VMX_INVVPID)
1136 /* CF==1 or ZF==1 --> rc = -1 */
1137 "; ja 1f ; ud2 ; 1:"
1138 : : "a"(&operand), "c"(ext) : "cc", "memory");
1141 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
1145 } operand = {eptp, gpa};
1147 asm volatile (__ex(ASM_VMX_INVEPT)
1148 /* CF==1 or ZF==1 --> rc = -1 */
1149 "; ja 1f ; ud2 ; 1:\n"
1150 : : "a" (&operand), "c" (ext) : "cc", "memory");
1153 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
1157 i = __find_msr_index(vmx, msr);
1159 return &vmx->guest_msrs[i];
1163 static void vmcs_clear(struct vmcs *vmcs)
1165 u64 phys_addr = __pa(vmcs);
1168 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
1169 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1172 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
1176 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
1178 vmcs_clear(loaded_vmcs->vmcs);
1179 loaded_vmcs->cpu = -1;
1180 loaded_vmcs->launched = 0;
1183 static void vmcs_load(struct vmcs *vmcs)
1185 u64 phys_addr = __pa(vmcs);
1188 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
1189 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1192 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
1198 * This bitmap is used to indicate whether the vmclear
1199 * operation is enabled on all cpus. All disabled by
1202 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
1204 static inline void crash_enable_local_vmclear(int cpu)
1206 cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
1209 static inline void crash_disable_local_vmclear(int cpu)
1211 cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
1214 static inline int crash_local_vmclear_enabled(int cpu)
1216 return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
1219 static void crash_vmclear_local_loaded_vmcss(void)
1221 int cpu = raw_smp_processor_id();
1222 struct loaded_vmcs *v;
1224 if (!crash_local_vmclear_enabled(cpu))
1227 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
1228 loaded_vmcss_on_cpu_link)
1229 vmcs_clear(v->vmcs);
1232 static inline void crash_enable_local_vmclear(int cpu) { }
1233 static inline void crash_disable_local_vmclear(int cpu) { }
1234 #endif /* CONFIG_KEXEC */
1236 static void __loaded_vmcs_clear(void *arg)
1238 struct loaded_vmcs *loaded_vmcs = arg;
1239 int cpu = raw_smp_processor_id();
1241 if (loaded_vmcs->cpu != cpu)
1242 return; /* vcpu migration can race with cpu offline */
1243 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1244 per_cpu(current_vmcs, cpu) = NULL;
1245 crash_disable_local_vmclear(cpu);
1246 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1249 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1250 * is before setting loaded_vmcs->vcpu to -1 which is done in
1251 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1252 * then adds the vmcs into percpu list before it is deleted.
1256 loaded_vmcs_init(loaded_vmcs);
1257 crash_enable_local_vmclear(cpu);
1260 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1262 int cpu = loaded_vmcs->cpu;
1265 smp_call_function_single(cpu,
1266 __loaded_vmcs_clear, loaded_vmcs, 1);
1269 static inline void vpid_sync_vcpu_single(struct vcpu_vmx *vmx)
1274 if (cpu_has_vmx_invvpid_single())
1275 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
1278 static inline void vpid_sync_vcpu_global(void)
1280 if (cpu_has_vmx_invvpid_global())
1281 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1284 static inline void vpid_sync_context(struct vcpu_vmx *vmx)
1286 if (cpu_has_vmx_invvpid_single())
1287 vpid_sync_vcpu_single(vmx);
1289 vpid_sync_vcpu_global();
1292 static inline void ept_sync_global(void)
1294 if (cpu_has_vmx_invept_global())
1295 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1298 static inline void ept_sync_context(u64 eptp)
1301 if (cpu_has_vmx_invept_context())
1302 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1308 static __always_inline unsigned long vmcs_readl(unsigned long field)
1310 unsigned long value;
1312 asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1313 : "=a"(value) : "d"(field) : "cc");
1317 static __always_inline u16 vmcs_read16(unsigned long field)
1319 return vmcs_readl(field);
1322 static __always_inline u32 vmcs_read32(unsigned long field)
1324 return vmcs_readl(field);
1327 static __always_inline u64 vmcs_read64(unsigned long field)
1329 #ifdef CONFIG_X86_64
1330 return vmcs_readl(field);
1332 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
1336 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1338 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1339 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1343 static void vmcs_writel(unsigned long field, unsigned long value)
1347 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1348 : "=q"(error) : "a"(value), "d"(field) : "cc");
1349 if (unlikely(error))
1350 vmwrite_error(field, value);
1353 static void vmcs_write16(unsigned long field, u16 value)
1355 vmcs_writel(field, value);
1358 static void vmcs_write32(unsigned long field, u32 value)
1360 vmcs_writel(field, value);
1363 static void vmcs_write64(unsigned long field, u64 value)
1365 vmcs_writel(field, value);
1366 #ifndef CONFIG_X86_64
1368 vmcs_writel(field+1, value >> 32);
1372 static void vmcs_clear_bits(unsigned long field, u32 mask)
1374 vmcs_writel(field, vmcs_readl(field) & ~mask);
1377 static void vmcs_set_bits(unsigned long field, u32 mask)
1379 vmcs_writel(field, vmcs_readl(field) | mask);
1382 static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val)
1384 vmcs_write32(VM_ENTRY_CONTROLS, val);
1385 vmx->vm_entry_controls_shadow = val;
1388 static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val)
1390 if (vmx->vm_entry_controls_shadow != val)
1391 vm_entry_controls_init(vmx, val);
1394 static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx)
1396 return vmx->vm_entry_controls_shadow;
1400 static inline void vm_entry_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1402 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) | val);
1405 static inline void vm_entry_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1407 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) & ~val);
1410 static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val)
1412 vmcs_write32(VM_EXIT_CONTROLS, val);
1413 vmx->vm_exit_controls_shadow = val;
1416 static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val)
1418 if (vmx->vm_exit_controls_shadow != val)
1419 vm_exit_controls_init(vmx, val);
1422 static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx)
1424 return vmx->vm_exit_controls_shadow;
1428 static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1430 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val);
1433 static inline void vm_exit_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1435 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) & ~val);
1438 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1440 vmx->segment_cache.bitmask = 0;
1443 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1447 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1449 if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1450 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1451 vmx->segment_cache.bitmask = 0;
1453 ret = vmx->segment_cache.bitmask & mask;
1454 vmx->segment_cache.bitmask |= mask;
1458 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1460 u16 *p = &vmx->segment_cache.seg[seg].selector;
1462 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1463 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1467 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1469 ulong *p = &vmx->segment_cache.seg[seg].base;
1471 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1472 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1476 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1478 u32 *p = &vmx->segment_cache.seg[seg].limit;
1480 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1481 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1485 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1487 u32 *p = &vmx->segment_cache.seg[seg].ar;
1489 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1490 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1494 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1498 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1499 (1u << NM_VECTOR) | (1u << DB_VECTOR);
1500 if ((vcpu->guest_debug &
1501 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1502 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1503 eb |= 1u << BP_VECTOR;
1504 if (to_vmx(vcpu)->rmode.vm86_active)
1507 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1508 if (vcpu->fpu_active)
1509 eb &= ~(1u << NM_VECTOR);
1511 /* When we are running a nested L2 guest and L1 specified for it a
1512 * certain exception bitmap, we must trap the same exceptions and pass
1513 * them to L1. When running L2, we will only handle the exceptions
1514 * specified above if L1 did not want them.
1516 if (is_guest_mode(vcpu))
1517 eb |= get_vmcs12(vcpu)->exception_bitmap;
1519 vmcs_write32(EXCEPTION_BITMAP, eb);
1522 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1523 unsigned long entry, unsigned long exit)
1525 vm_entry_controls_clearbit(vmx, entry);
1526 vm_exit_controls_clearbit(vmx, exit);
1529 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1532 struct msr_autoload *m = &vmx->msr_autoload;
1536 if (cpu_has_load_ia32_efer) {
1537 clear_atomic_switch_msr_special(vmx,
1538 VM_ENTRY_LOAD_IA32_EFER,
1539 VM_EXIT_LOAD_IA32_EFER);
1543 case MSR_CORE_PERF_GLOBAL_CTRL:
1544 if (cpu_has_load_perf_global_ctrl) {
1545 clear_atomic_switch_msr_special(vmx,
1546 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1547 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1553 for (i = 0; i < m->nr; ++i)
1554 if (m->guest[i].index == msr)
1560 m->guest[i] = m->guest[m->nr];
1561 m->host[i] = m->host[m->nr];
1562 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1563 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1566 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1567 unsigned long entry, unsigned long exit,
1568 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
1569 u64 guest_val, u64 host_val)
1571 vmcs_write64(guest_val_vmcs, guest_val);
1572 vmcs_write64(host_val_vmcs, host_val);
1573 vm_entry_controls_setbit(vmx, entry);
1574 vm_exit_controls_setbit(vmx, exit);
1577 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1578 u64 guest_val, u64 host_val)
1581 struct msr_autoload *m = &vmx->msr_autoload;
1585 if (cpu_has_load_ia32_efer) {
1586 add_atomic_switch_msr_special(vmx,
1587 VM_ENTRY_LOAD_IA32_EFER,
1588 VM_EXIT_LOAD_IA32_EFER,
1591 guest_val, host_val);
1595 case MSR_CORE_PERF_GLOBAL_CTRL:
1596 if (cpu_has_load_perf_global_ctrl) {
1597 add_atomic_switch_msr_special(vmx,
1598 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1599 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1600 GUEST_IA32_PERF_GLOBAL_CTRL,
1601 HOST_IA32_PERF_GLOBAL_CTRL,
1602 guest_val, host_val);
1608 for (i = 0; i < m->nr; ++i)
1609 if (m->guest[i].index == msr)
1612 if (i == NR_AUTOLOAD_MSRS) {
1613 printk_once(KERN_WARNING "Not enough msr switch entries. "
1614 "Can't add msr %x\n", msr);
1616 } else if (i == m->nr) {
1618 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1619 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1622 m->guest[i].index = msr;
1623 m->guest[i].value = guest_val;
1624 m->host[i].index = msr;
1625 m->host[i].value = host_val;
1628 static void reload_tss(void)
1631 * VT restores TR but not its size. Useless.
1633 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
1634 struct desc_struct *descs;
1636 descs = (void *)gdt->address;
1637 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
1641 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
1646 guest_efer = vmx->vcpu.arch.efer;
1649 * NX is emulated; LMA and LME handled by hardware; SCE meaningless
1652 ignore_bits = EFER_NX | EFER_SCE;
1653 #ifdef CONFIG_X86_64
1654 ignore_bits |= EFER_LMA | EFER_LME;
1655 /* SCE is meaningful only in long mode on Intel */
1656 if (guest_efer & EFER_LMA)
1657 ignore_bits &= ~(u64)EFER_SCE;
1659 guest_efer &= ~ignore_bits;
1660 guest_efer |= host_efer & ignore_bits;
1661 vmx->guest_msrs[efer_offset].data = guest_efer;
1662 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1664 clear_atomic_switch_msr(vmx, MSR_EFER);
1667 * On EPT, we can't emulate NX, so we must switch EFER atomically.
1668 * On CPUs that support "load IA32_EFER", always switch EFER
1669 * atomically, since it's faster than switching it manually.
1671 if (cpu_has_load_ia32_efer ||
1672 (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
1673 guest_efer = vmx->vcpu.arch.efer;
1674 if (!(guest_efer & EFER_LMA))
1675 guest_efer &= ~EFER_LME;
1676 if (guest_efer != host_efer)
1677 add_atomic_switch_msr(vmx, MSR_EFER,
1678 guest_efer, host_efer);
1685 static unsigned long segment_base(u16 selector)
1687 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
1688 struct desc_struct *d;
1689 unsigned long table_base;
1692 if (!(selector & ~3))
1695 table_base = gdt->address;
1697 if (selector & 4) { /* from ldt */
1698 u16 ldt_selector = kvm_read_ldt();
1700 if (!(ldt_selector & ~3))
1703 table_base = segment_base(ldt_selector);
1705 d = (struct desc_struct *)(table_base + (selector & ~7));
1706 v = get_desc_base(d);
1707 #ifdef CONFIG_X86_64
1708 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
1709 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
1714 static inline unsigned long kvm_read_tr_base(void)
1717 asm("str %0" : "=g"(tr));
1718 return segment_base(tr);
1721 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
1723 struct vcpu_vmx *vmx = to_vmx(vcpu);
1726 if (vmx->host_state.loaded)
1729 vmx->host_state.loaded = 1;
1731 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
1732 * allow segment selectors with cpl > 0 or ti == 1.
1734 vmx->host_state.ldt_sel = kvm_read_ldt();
1735 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
1736 savesegment(fs, vmx->host_state.fs_sel);
1737 if (!(vmx->host_state.fs_sel & 7)) {
1738 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
1739 vmx->host_state.fs_reload_needed = 0;
1741 vmcs_write16(HOST_FS_SELECTOR, 0);
1742 vmx->host_state.fs_reload_needed = 1;
1744 savesegment(gs, vmx->host_state.gs_sel);
1745 if (!(vmx->host_state.gs_sel & 7))
1746 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
1748 vmcs_write16(HOST_GS_SELECTOR, 0);
1749 vmx->host_state.gs_ldt_reload_needed = 1;
1752 #ifdef CONFIG_X86_64
1753 savesegment(ds, vmx->host_state.ds_sel);
1754 savesegment(es, vmx->host_state.es_sel);
1757 #ifdef CONFIG_X86_64
1758 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
1759 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
1761 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
1762 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
1765 #ifdef CONFIG_X86_64
1766 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1767 if (is_long_mode(&vmx->vcpu))
1768 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1770 if (boot_cpu_has(X86_FEATURE_MPX))
1771 rdmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
1772 for (i = 0; i < vmx->save_nmsrs; ++i)
1773 kvm_set_shared_msr(vmx->guest_msrs[i].index,
1774 vmx->guest_msrs[i].data,
1775 vmx->guest_msrs[i].mask);
1778 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
1780 if (!vmx->host_state.loaded)
1783 ++vmx->vcpu.stat.host_state_reload;
1784 vmx->host_state.loaded = 0;
1785 #ifdef CONFIG_X86_64
1786 if (is_long_mode(&vmx->vcpu))
1787 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1789 if (vmx->host_state.gs_ldt_reload_needed) {
1790 kvm_load_ldt(vmx->host_state.ldt_sel);
1791 #ifdef CONFIG_X86_64
1792 load_gs_index(vmx->host_state.gs_sel);
1794 loadsegment(gs, vmx->host_state.gs_sel);
1797 if (vmx->host_state.fs_reload_needed)
1798 loadsegment(fs, vmx->host_state.fs_sel);
1799 #ifdef CONFIG_X86_64
1800 if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
1801 loadsegment(ds, vmx->host_state.ds_sel);
1802 loadsegment(es, vmx->host_state.es_sel);
1806 #ifdef CONFIG_X86_64
1807 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1809 if (vmx->host_state.msr_host_bndcfgs)
1810 wrmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
1812 * If the FPU is not active (through the host task or
1813 * the guest vcpu), then restore the cr0.TS bit.
1815 if (!user_has_fpu() && !vmx->vcpu.guest_fpu_loaded)
1817 load_gdt(this_cpu_ptr(&host_gdt));
1820 static void vmx_load_host_state(struct vcpu_vmx *vmx)
1823 __vmx_load_host_state(vmx);
1828 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1829 * vcpu mutex is already taken.
1831 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1833 struct vcpu_vmx *vmx = to_vmx(vcpu);
1834 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1837 kvm_cpu_vmxon(phys_addr);
1838 else if (vmx->loaded_vmcs->cpu != cpu)
1839 loaded_vmcs_clear(vmx->loaded_vmcs);
1841 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1842 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1843 vmcs_load(vmx->loaded_vmcs->vmcs);
1846 if (vmx->loaded_vmcs->cpu != cpu) {
1847 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
1848 unsigned long sysenter_esp;
1850 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1851 local_irq_disable();
1852 crash_disable_local_vmclear(cpu);
1855 * Read loaded_vmcs->cpu should be before fetching
1856 * loaded_vmcs->loaded_vmcss_on_cpu_link.
1857 * See the comments in __loaded_vmcs_clear().
1861 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1862 &per_cpu(loaded_vmcss_on_cpu, cpu));
1863 crash_enable_local_vmclear(cpu);
1867 * Linux uses per-cpu TSS and GDT, so set these when switching
1870 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
1871 vmcs_writel(HOST_GDTR_BASE, gdt->address); /* 22.2.4 */
1873 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1874 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1875 vmx->loaded_vmcs->cpu = cpu;
1879 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1881 __vmx_load_host_state(to_vmx(vcpu));
1882 if (!vmm_exclusive) {
1883 __loaded_vmcs_clear(to_vmx(vcpu)->loaded_vmcs);
1889 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
1893 if (vcpu->fpu_active)
1895 vcpu->fpu_active = 1;
1896 cr0 = vmcs_readl(GUEST_CR0);
1897 cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
1898 cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
1899 vmcs_writel(GUEST_CR0, cr0);
1900 update_exception_bitmap(vcpu);
1901 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
1902 if (is_guest_mode(vcpu))
1903 vcpu->arch.cr0_guest_owned_bits &=
1904 ~get_vmcs12(vcpu)->cr0_guest_host_mask;
1905 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1908 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
1911 * Return the cr0 value that a nested guest would read. This is a combination
1912 * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
1913 * its hypervisor (cr0_read_shadow).
1915 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
1917 return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
1918 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
1920 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
1922 return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
1923 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
1926 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
1928 /* Note that there is no vcpu->fpu_active = 0 here. The caller must
1929 * set this *before* calling this function.
1931 vmx_decache_cr0_guest_bits(vcpu);
1932 vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
1933 update_exception_bitmap(vcpu);
1934 vcpu->arch.cr0_guest_owned_bits = 0;
1935 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1936 if (is_guest_mode(vcpu)) {
1938 * L1's specified read shadow might not contain the TS bit,
1939 * so now that we turned on shadowing of this bit, we need to
1940 * set this bit of the shadow. Like in nested_vmx_run we need
1941 * nested_read_cr0(vmcs12), but vmcs12->guest_cr0 is not yet
1942 * up-to-date here because we just decached cr0.TS (and we'll
1943 * only update vmcs12->guest_cr0 on nested exit).
1945 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1946 vmcs12->guest_cr0 = (vmcs12->guest_cr0 & ~X86_CR0_TS) |
1947 (vcpu->arch.cr0 & X86_CR0_TS);
1948 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
1950 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
1953 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1955 unsigned long rflags, save_rflags;
1957 if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
1958 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1959 rflags = vmcs_readl(GUEST_RFLAGS);
1960 if (to_vmx(vcpu)->rmode.vm86_active) {
1961 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1962 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
1963 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1965 to_vmx(vcpu)->rflags = rflags;
1967 return to_vmx(vcpu)->rflags;
1970 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1972 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1973 to_vmx(vcpu)->rflags = rflags;
1974 if (to_vmx(vcpu)->rmode.vm86_active) {
1975 to_vmx(vcpu)->rmode.save_rflags = rflags;
1976 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1978 vmcs_writel(GUEST_RFLAGS, rflags);
1981 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
1983 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1986 if (interruptibility & GUEST_INTR_STATE_STI)
1987 ret |= KVM_X86_SHADOW_INT_STI;
1988 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1989 ret |= KVM_X86_SHADOW_INT_MOV_SS;
1994 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1996 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1997 u32 interruptibility = interruptibility_old;
1999 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
2001 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
2002 interruptibility |= GUEST_INTR_STATE_MOV_SS;
2003 else if (mask & KVM_X86_SHADOW_INT_STI)
2004 interruptibility |= GUEST_INTR_STATE_STI;
2006 if ((interruptibility != interruptibility_old))
2007 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
2010 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
2014 rip = kvm_rip_read(vcpu);
2015 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
2016 kvm_rip_write(vcpu, rip);
2018 /* skipping an emulated instruction also counts */
2019 vmx_set_interrupt_shadow(vcpu, 0);
2023 * KVM wants to inject page-faults which it got to the guest. This function
2024 * checks whether in a nested guest, we need to inject them to L1 or L2.
2026 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned nr)
2028 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2030 if (!(vmcs12->exception_bitmap & (1u << nr)))
2033 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
2034 vmcs_read32(VM_EXIT_INTR_INFO),
2035 vmcs_readl(EXIT_QUALIFICATION));
2039 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
2040 bool has_error_code, u32 error_code,
2043 struct vcpu_vmx *vmx = to_vmx(vcpu);
2044 u32 intr_info = nr | INTR_INFO_VALID_MASK;
2046 if (!reinject && is_guest_mode(vcpu) &&
2047 nested_vmx_check_exception(vcpu, nr))
2050 if (has_error_code) {
2051 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
2052 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2055 if (vmx->rmode.vm86_active) {
2057 if (kvm_exception_is_soft(nr))
2058 inc_eip = vcpu->arch.event_exit_inst_len;
2059 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
2060 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2064 if (kvm_exception_is_soft(nr)) {
2065 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2066 vmx->vcpu.arch.event_exit_inst_len);
2067 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2069 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2071 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
2074 static bool vmx_rdtscp_supported(void)
2076 return cpu_has_vmx_rdtscp();
2079 static bool vmx_invpcid_supported(void)
2081 return cpu_has_vmx_invpcid() && enable_ept;
2085 * Swap MSR entry in host/guest MSR entry array.
2087 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
2089 struct shared_msr_entry tmp;
2091 tmp = vmx->guest_msrs[to];
2092 vmx->guest_msrs[to] = vmx->guest_msrs[from];
2093 vmx->guest_msrs[from] = tmp;
2096 static void vmx_set_msr_bitmap(struct kvm_vcpu *vcpu)
2098 unsigned long *msr_bitmap;
2100 if (irqchip_in_kernel(vcpu->kvm) && apic_x2apic_mode(vcpu->arch.apic)) {
2101 if (is_long_mode(vcpu))
2102 msr_bitmap = vmx_msr_bitmap_longmode_x2apic;
2104 msr_bitmap = vmx_msr_bitmap_legacy_x2apic;
2106 if (is_long_mode(vcpu))
2107 msr_bitmap = vmx_msr_bitmap_longmode;
2109 msr_bitmap = vmx_msr_bitmap_legacy;
2112 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
2116 * Set up the vmcs to automatically save and restore system
2117 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
2118 * mode, as fiddling with msrs is very expensive.
2120 static void setup_msrs(struct vcpu_vmx *vmx)
2122 int save_nmsrs, index;
2125 #ifdef CONFIG_X86_64
2126 if (is_long_mode(&vmx->vcpu)) {
2127 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
2129 move_msr_up(vmx, index, save_nmsrs++);
2130 index = __find_msr_index(vmx, MSR_LSTAR);
2132 move_msr_up(vmx, index, save_nmsrs++);
2133 index = __find_msr_index(vmx, MSR_CSTAR);
2135 move_msr_up(vmx, index, save_nmsrs++);
2136 index = __find_msr_index(vmx, MSR_TSC_AUX);
2137 if (index >= 0 && vmx->rdtscp_enabled)
2138 move_msr_up(vmx, index, save_nmsrs++);
2140 * MSR_STAR is only needed on long mode guests, and only
2141 * if efer.sce is enabled.
2143 index = __find_msr_index(vmx, MSR_STAR);
2144 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
2145 move_msr_up(vmx, index, save_nmsrs++);
2148 index = __find_msr_index(vmx, MSR_EFER);
2149 if (index >= 0 && update_transition_efer(vmx, index))
2150 move_msr_up(vmx, index, save_nmsrs++);
2152 vmx->save_nmsrs = save_nmsrs;
2154 if (cpu_has_vmx_msr_bitmap())
2155 vmx_set_msr_bitmap(&vmx->vcpu);
2159 * reads and returns guest's timestamp counter "register"
2160 * guest_tsc = host_tsc + tsc_offset -- 21.3
2162 static u64 guest_read_tsc(void)
2164 u64 host_tsc, tsc_offset;
2167 tsc_offset = vmcs_read64(TSC_OFFSET);
2168 return host_tsc + tsc_offset;
2172 * Like guest_read_tsc, but always returns L1's notion of the timestamp
2173 * counter, even if a nested guest (L2) is currently running.
2175 static u64 vmx_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2179 tsc_offset = is_guest_mode(vcpu) ?
2180 to_vmx(vcpu)->nested.vmcs01_tsc_offset :
2181 vmcs_read64(TSC_OFFSET);
2182 return host_tsc + tsc_offset;
2186 * Engage any workarounds for mis-matched TSC rates. Currently limited to
2187 * software catchup for faster rates on slower CPUs.
2189 static void vmx_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2194 if (user_tsc_khz > tsc_khz) {
2195 vcpu->arch.tsc_catchup = 1;
2196 vcpu->arch.tsc_always_catchup = 1;
2198 WARN(1, "user requested TSC rate below hardware speed\n");
2201 static u64 vmx_read_tsc_offset(struct kvm_vcpu *vcpu)
2203 return vmcs_read64(TSC_OFFSET);
2207 * writes 'offset' into guest's timestamp counter offset register
2209 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2211 if (is_guest_mode(vcpu)) {
2213 * We're here if L1 chose not to trap WRMSR to TSC. According
2214 * to the spec, this should set L1's TSC; The offset that L1
2215 * set for L2 remains unchanged, and still needs to be added
2216 * to the newly set TSC to get L2's TSC.
2218 struct vmcs12 *vmcs12;
2219 to_vmx(vcpu)->nested.vmcs01_tsc_offset = offset;
2220 /* recalculate vmcs02.TSC_OFFSET: */
2221 vmcs12 = get_vmcs12(vcpu);
2222 vmcs_write64(TSC_OFFSET, offset +
2223 (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
2224 vmcs12->tsc_offset : 0));
2226 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2227 vmcs_read64(TSC_OFFSET), offset);
2228 vmcs_write64(TSC_OFFSET, offset);
2232 static void vmx_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment, bool host)
2234 u64 offset = vmcs_read64(TSC_OFFSET);
2236 vmcs_write64(TSC_OFFSET, offset + adjustment);
2237 if (is_guest_mode(vcpu)) {
2238 /* Even when running L2, the adjustment needs to apply to L1 */
2239 to_vmx(vcpu)->nested.vmcs01_tsc_offset += adjustment;
2241 trace_kvm_write_tsc_offset(vcpu->vcpu_id, offset,
2242 offset + adjustment);
2245 static u64 vmx_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2247 return target_tsc - native_read_tsc();
2250 static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
2252 struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
2253 return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
2257 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
2258 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
2259 * all guests if the "nested" module option is off, and can also be disabled
2260 * for a single guest by disabling its VMX cpuid bit.
2262 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
2264 return nested && guest_cpuid_has_vmx(vcpu);
2268 * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
2269 * returned for the various VMX controls MSRs when nested VMX is enabled.
2270 * The same values should also be used to verify that vmcs12 control fields are
2271 * valid during nested entry from L1 to L2.
2272 * Each of these control msrs has a low and high 32-bit half: A low bit is on
2273 * if the corresponding bit in the (32-bit) control field *must* be on, and a
2274 * bit in the high half is on if the corresponding bit in the control field
2275 * may be on. See also vmx_control_verify().
2276 * TODO: allow these variables to be modified (downgraded) by module options
2279 static u32 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high;
2280 static u32 nested_vmx_true_procbased_ctls_low;
2281 static u32 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high;
2282 static u32 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high;
2283 static u32 nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high;
2284 static u32 nested_vmx_true_exit_ctls_low;
2285 static u32 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high;
2286 static u32 nested_vmx_true_entry_ctls_low;
2287 static u32 nested_vmx_misc_low, nested_vmx_misc_high;
2288 static u32 nested_vmx_ept_caps;
2289 static __init void nested_vmx_setup_ctls_msrs(void)
2292 * Note that as a general rule, the high half of the MSRs (bits in
2293 * the control fields which may be 1) should be initialized by the
2294 * intersection of the underlying hardware's MSR (i.e., features which
2295 * can be supported) and the list of features we want to expose -
2296 * because they are known to be properly supported in our code.
2297 * Also, usually, the low half of the MSRs (bits which must be 1) can
2298 * be set to 0, meaning that L1 may turn off any of these bits. The
2299 * reason is that if one of these bits is necessary, it will appear
2300 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
2301 * fields of vmcs01 and vmcs02, will turn these bits off - and
2302 * nested_vmx_exit_handled() will not pass related exits to L1.
2303 * These rules have exceptions below.
2306 /* pin-based controls */
2307 rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
2308 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high);
2309 nested_vmx_pinbased_ctls_low |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2310 nested_vmx_pinbased_ctls_high &= PIN_BASED_EXT_INTR_MASK |
2311 PIN_BASED_NMI_EXITING | PIN_BASED_VIRTUAL_NMIS;
2312 nested_vmx_pinbased_ctls_high |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2313 PIN_BASED_VMX_PREEMPTION_TIMER;
2316 rdmsr(MSR_IA32_VMX_EXIT_CTLS,
2317 nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high);
2318 nested_vmx_exit_ctls_low = VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
2320 nested_vmx_exit_ctls_high &=
2321 #ifdef CONFIG_X86_64
2322 VM_EXIT_HOST_ADDR_SPACE_SIZE |
2324 VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
2325 nested_vmx_exit_ctls_high |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
2326 VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
2327 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
2329 if (vmx_mpx_supported())
2330 nested_vmx_exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
2332 /* We support free control of debug control saving. */
2333 nested_vmx_true_exit_ctls_low = nested_vmx_exit_ctls_low &
2334 ~VM_EXIT_SAVE_DEBUG_CONTROLS;
2336 /* entry controls */
2337 rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
2338 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high);
2339 nested_vmx_entry_ctls_low = VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
2340 nested_vmx_entry_ctls_high &=
2341 #ifdef CONFIG_X86_64
2342 VM_ENTRY_IA32E_MODE |
2344 VM_ENTRY_LOAD_IA32_PAT;
2345 nested_vmx_entry_ctls_high |= (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR |
2346 VM_ENTRY_LOAD_IA32_EFER);
2347 if (vmx_mpx_supported())
2348 nested_vmx_entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
2350 /* We support free control of debug control loading. */
2351 nested_vmx_true_entry_ctls_low = nested_vmx_entry_ctls_low &
2352 ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
2354 /* cpu-based controls */
2355 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
2356 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high);
2357 nested_vmx_procbased_ctls_low = CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2358 nested_vmx_procbased_ctls_high &=
2359 CPU_BASED_VIRTUAL_INTR_PENDING |
2360 CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
2361 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
2362 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
2363 CPU_BASED_CR3_STORE_EXITING |
2364 #ifdef CONFIG_X86_64
2365 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
2367 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
2368 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_EXITING |
2369 CPU_BASED_RDPMC_EXITING | CPU_BASED_RDTSC_EXITING |
2370 CPU_BASED_PAUSE_EXITING | CPU_BASED_TPR_SHADOW |
2371 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2373 * We can allow some features even when not supported by the
2374 * hardware. For example, L1 can specify an MSR bitmap - and we
2375 * can use it to avoid exits to L1 - even when L0 runs L2
2376 * without MSR bitmaps.
2378 nested_vmx_procbased_ctls_high |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2379 CPU_BASED_USE_MSR_BITMAPS;
2381 /* We support free control of CR3 access interception. */
2382 nested_vmx_true_procbased_ctls_low = nested_vmx_procbased_ctls_low &
2383 ~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
2385 /* secondary cpu-based controls */
2386 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2387 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high);
2388 nested_vmx_secondary_ctls_low = 0;
2389 nested_vmx_secondary_ctls_high &=
2390 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2391 SECONDARY_EXEC_UNRESTRICTED_GUEST |
2392 SECONDARY_EXEC_WBINVD_EXITING;
2395 /* nested EPT: emulate EPT also to L1 */
2396 nested_vmx_secondary_ctls_high |= SECONDARY_EXEC_ENABLE_EPT;
2397 nested_vmx_ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
2398 VMX_EPTP_WB_BIT | VMX_EPT_2MB_PAGE_BIT |
2400 nested_vmx_ept_caps &= vmx_capability.ept;
2402 * For nested guests, we don't do anything specific
2403 * for single context invalidation. Hence, only advertise
2404 * support for global context invalidation.
2406 nested_vmx_ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT;
2408 nested_vmx_ept_caps = 0;
2410 /* miscellaneous data */
2411 rdmsr(MSR_IA32_VMX_MISC, nested_vmx_misc_low, nested_vmx_misc_high);
2412 nested_vmx_misc_low &= VMX_MISC_SAVE_EFER_LMA;
2413 nested_vmx_misc_low |= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
2414 VMX_MISC_ACTIVITY_HLT;
2415 nested_vmx_misc_high = 0;
2418 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2421 * Bits 0 in high must be 0, and bits 1 in low must be 1.
2423 return ((control & high) | low) == control;
2426 static inline u64 vmx_control_msr(u32 low, u32 high)
2428 return low | ((u64)high << 32);
2431 /* Returns 0 on success, non-0 otherwise. */
2432 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2434 switch (msr_index) {
2435 case MSR_IA32_VMX_BASIC:
2437 * This MSR reports some information about VMX support. We
2438 * should return information about the VMX we emulate for the
2439 * guest, and the VMCS structure we give it - not about the
2440 * VMX support of the underlying hardware.
2442 *pdata = VMCS12_REVISION | VMX_BASIC_TRUE_CTLS |
2443 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2444 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2446 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2447 case MSR_IA32_VMX_PINBASED_CTLS:
2448 *pdata = vmx_control_msr(nested_vmx_pinbased_ctls_low,
2449 nested_vmx_pinbased_ctls_high);
2451 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2452 *pdata = vmx_control_msr(nested_vmx_true_procbased_ctls_low,
2453 nested_vmx_procbased_ctls_high);
2455 case MSR_IA32_VMX_PROCBASED_CTLS:
2456 *pdata = vmx_control_msr(nested_vmx_procbased_ctls_low,
2457 nested_vmx_procbased_ctls_high);
2459 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2460 *pdata = vmx_control_msr(nested_vmx_true_exit_ctls_low,
2461 nested_vmx_exit_ctls_high);
2463 case MSR_IA32_VMX_EXIT_CTLS:
2464 *pdata = vmx_control_msr(nested_vmx_exit_ctls_low,
2465 nested_vmx_exit_ctls_high);
2467 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2468 *pdata = vmx_control_msr(nested_vmx_true_entry_ctls_low,
2469 nested_vmx_entry_ctls_high);
2471 case MSR_IA32_VMX_ENTRY_CTLS:
2472 *pdata = vmx_control_msr(nested_vmx_entry_ctls_low,
2473 nested_vmx_entry_ctls_high);
2475 case MSR_IA32_VMX_MISC:
2476 *pdata = vmx_control_msr(nested_vmx_misc_low,
2477 nested_vmx_misc_high);
2480 * These MSRs specify bits which the guest must keep fixed (on or off)
2481 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2482 * We picked the standard core2 setting.
2484 #define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2485 #define VMXON_CR4_ALWAYSON X86_CR4_VMXE
2486 case MSR_IA32_VMX_CR0_FIXED0:
2487 *pdata = VMXON_CR0_ALWAYSON;
2489 case MSR_IA32_VMX_CR0_FIXED1:
2492 case MSR_IA32_VMX_CR4_FIXED0:
2493 *pdata = VMXON_CR4_ALWAYSON;
2495 case MSR_IA32_VMX_CR4_FIXED1:
2498 case MSR_IA32_VMX_VMCS_ENUM:
2499 *pdata = 0x2e; /* highest index: VMX_PREEMPTION_TIMER_VALUE */
2501 case MSR_IA32_VMX_PROCBASED_CTLS2:
2502 *pdata = vmx_control_msr(nested_vmx_secondary_ctls_low,
2503 nested_vmx_secondary_ctls_high);
2505 case MSR_IA32_VMX_EPT_VPID_CAP:
2506 /* Currently, no nested vpid support */
2507 *pdata = nested_vmx_ept_caps;
2517 * Reads an msr value (of 'msr_index') into 'pdata'.
2518 * Returns 0 on success, non-0 otherwise.
2519 * Assumes vcpu_load() was already called.
2521 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2524 struct shared_msr_entry *msr;
2527 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
2531 switch (msr_index) {
2532 #ifdef CONFIG_X86_64
2534 data = vmcs_readl(GUEST_FS_BASE);
2537 data = vmcs_readl(GUEST_GS_BASE);
2539 case MSR_KERNEL_GS_BASE:
2540 vmx_load_host_state(to_vmx(vcpu));
2541 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
2545 return kvm_get_msr_common(vcpu, msr_index, pdata);
2547 data = guest_read_tsc();
2549 case MSR_IA32_SYSENTER_CS:
2550 data = vmcs_read32(GUEST_SYSENTER_CS);
2552 case MSR_IA32_SYSENTER_EIP:
2553 data = vmcs_readl(GUEST_SYSENTER_EIP);
2555 case MSR_IA32_SYSENTER_ESP:
2556 data = vmcs_readl(GUEST_SYSENTER_ESP);
2558 case MSR_IA32_BNDCFGS:
2559 if (!vmx_mpx_supported())
2561 data = vmcs_read64(GUEST_BNDCFGS);
2563 case MSR_IA32_FEATURE_CONTROL:
2564 if (!nested_vmx_allowed(vcpu))
2566 data = to_vmx(vcpu)->nested.msr_ia32_feature_control;
2568 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2569 if (!nested_vmx_allowed(vcpu))
2571 return vmx_get_vmx_msr(vcpu, msr_index, pdata);
2573 if (!to_vmx(vcpu)->rdtscp_enabled)
2575 /* Otherwise falls through */
2577 msr = find_msr_entry(to_vmx(vcpu), msr_index);
2582 return kvm_get_msr_common(vcpu, msr_index, pdata);
2589 static void vmx_leave_nested(struct kvm_vcpu *vcpu);
2592 * Writes msr value into into the appropriate "register".
2593 * Returns 0 on success, non-0 otherwise.
2594 * Assumes vcpu_load() was already called.
2596 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2598 struct vcpu_vmx *vmx = to_vmx(vcpu);
2599 struct shared_msr_entry *msr;
2601 u32 msr_index = msr_info->index;
2602 u64 data = msr_info->data;
2604 switch (msr_index) {
2606 ret = kvm_set_msr_common(vcpu, msr_info);
2608 #ifdef CONFIG_X86_64
2610 vmx_segment_cache_clear(vmx);
2611 vmcs_writel(GUEST_FS_BASE, data);
2614 vmx_segment_cache_clear(vmx);
2615 vmcs_writel(GUEST_GS_BASE, data);
2617 case MSR_KERNEL_GS_BASE:
2618 vmx_load_host_state(vmx);
2619 vmx->msr_guest_kernel_gs_base = data;
2622 case MSR_IA32_SYSENTER_CS:
2623 vmcs_write32(GUEST_SYSENTER_CS, data);
2625 case MSR_IA32_SYSENTER_EIP:
2626 vmcs_writel(GUEST_SYSENTER_EIP, data);
2628 case MSR_IA32_SYSENTER_ESP:
2629 vmcs_writel(GUEST_SYSENTER_ESP, data);
2631 case MSR_IA32_BNDCFGS:
2632 if (!vmx_mpx_supported())
2634 vmcs_write64(GUEST_BNDCFGS, data);
2637 kvm_write_tsc(vcpu, msr_info);
2639 case MSR_IA32_CR_PAT:
2640 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2641 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
2643 vmcs_write64(GUEST_IA32_PAT, data);
2644 vcpu->arch.pat = data;
2647 ret = kvm_set_msr_common(vcpu, msr_info);
2649 case MSR_IA32_TSC_ADJUST:
2650 ret = kvm_set_msr_common(vcpu, msr_info);
2652 case MSR_IA32_FEATURE_CONTROL:
2653 if (!nested_vmx_allowed(vcpu) ||
2654 (to_vmx(vcpu)->nested.msr_ia32_feature_control &
2655 FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
2657 vmx->nested.msr_ia32_feature_control = data;
2658 if (msr_info->host_initiated && data == 0)
2659 vmx_leave_nested(vcpu);
2661 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2662 return 1; /* they are read-only */
2664 if (!vmx->rdtscp_enabled)
2666 /* Check reserved bit, higher 32 bits should be zero */
2667 if ((data >> 32) != 0)
2669 /* Otherwise falls through */
2671 msr = find_msr_entry(vmx, msr_index);
2673 u64 old_msr_data = msr->data;
2675 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
2677 ret = kvm_set_shared_msr(msr->index, msr->data,
2681 msr->data = old_msr_data;
2685 ret = kvm_set_msr_common(vcpu, msr_info);
2691 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2693 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
2696 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2699 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2701 case VCPU_EXREG_PDPTR:
2703 ept_save_pdptrs(vcpu);
2710 static __init int cpu_has_kvm_support(void)
2712 return cpu_has_vmx();
2715 static __init int vmx_disabled_by_bios(void)
2719 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
2720 if (msr & FEATURE_CONTROL_LOCKED) {
2721 /* launched w/ TXT and VMX disabled */
2722 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2725 /* launched w/o TXT and VMX only enabled w/ TXT */
2726 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2727 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2728 && !tboot_enabled()) {
2729 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
2730 "activate TXT before enabling KVM\n");
2733 /* launched w/o TXT and VMX disabled */
2734 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2735 && !tboot_enabled())
2742 static void kvm_cpu_vmxon(u64 addr)
2744 asm volatile (ASM_VMX_VMXON_RAX
2745 : : "a"(&addr), "m"(addr)
2749 static int hardware_enable(void)
2751 int cpu = raw_smp_processor_id();
2752 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2755 if (read_cr4() & X86_CR4_VMXE)
2758 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2761 * Now we can enable the vmclear operation in kdump
2762 * since the loaded_vmcss_on_cpu list on this cpu
2763 * has been initialized.
2765 * Though the cpu is not in VMX operation now, there
2766 * is no problem to enable the vmclear operation
2767 * for the loaded_vmcss_on_cpu list is empty!
2769 crash_enable_local_vmclear(cpu);
2771 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
2773 test_bits = FEATURE_CONTROL_LOCKED;
2774 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
2775 if (tboot_enabled())
2776 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
2778 if ((old & test_bits) != test_bits) {
2779 /* enable and lock */
2780 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
2782 write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
2784 if (vmm_exclusive) {
2785 kvm_cpu_vmxon(phys_addr);
2789 native_store_gdt(this_cpu_ptr(&host_gdt));
2794 static void vmclear_local_loaded_vmcss(void)
2796 int cpu = raw_smp_processor_id();
2797 struct loaded_vmcs *v, *n;
2799 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2800 loaded_vmcss_on_cpu_link)
2801 __loaded_vmcs_clear(v);
2805 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2808 static void kvm_cpu_vmxoff(void)
2810 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
2813 static void hardware_disable(void)
2815 if (vmm_exclusive) {
2816 vmclear_local_loaded_vmcss();
2819 write_cr4(read_cr4() & ~X86_CR4_VMXE);
2822 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2823 u32 msr, u32 *result)
2825 u32 vmx_msr_low, vmx_msr_high;
2826 u32 ctl = ctl_min | ctl_opt;
2828 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2830 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2831 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
2833 /* Ensure minimum (required) set of control bits are supported. */
2841 static __init bool allow_1_setting(u32 msr, u32 ctl)
2843 u32 vmx_msr_low, vmx_msr_high;
2845 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2846 return vmx_msr_high & ctl;
2849 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
2851 u32 vmx_msr_low, vmx_msr_high;
2852 u32 min, opt, min2, opt2;
2853 u32 _pin_based_exec_control = 0;
2854 u32 _cpu_based_exec_control = 0;
2855 u32 _cpu_based_2nd_exec_control = 0;
2856 u32 _vmexit_control = 0;
2857 u32 _vmentry_control = 0;
2859 min = CPU_BASED_HLT_EXITING |
2860 #ifdef CONFIG_X86_64
2861 CPU_BASED_CR8_LOAD_EXITING |
2862 CPU_BASED_CR8_STORE_EXITING |
2864 CPU_BASED_CR3_LOAD_EXITING |
2865 CPU_BASED_CR3_STORE_EXITING |
2866 CPU_BASED_USE_IO_BITMAPS |
2867 CPU_BASED_MOV_DR_EXITING |
2868 CPU_BASED_USE_TSC_OFFSETING |
2869 CPU_BASED_MWAIT_EXITING |
2870 CPU_BASED_MONITOR_EXITING |
2871 CPU_BASED_INVLPG_EXITING |
2872 CPU_BASED_RDPMC_EXITING;
2874 opt = CPU_BASED_TPR_SHADOW |
2875 CPU_BASED_USE_MSR_BITMAPS |
2876 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2877 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2878 &_cpu_based_exec_control) < 0)
2880 #ifdef CONFIG_X86_64
2881 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2882 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2883 ~CPU_BASED_CR8_STORE_EXITING;
2885 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2887 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2888 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2889 SECONDARY_EXEC_WBINVD_EXITING |
2890 SECONDARY_EXEC_ENABLE_VPID |
2891 SECONDARY_EXEC_ENABLE_EPT |
2892 SECONDARY_EXEC_UNRESTRICTED_GUEST |
2893 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2894 SECONDARY_EXEC_RDTSCP |
2895 SECONDARY_EXEC_ENABLE_INVPCID |
2896 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2897 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2898 SECONDARY_EXEC_SHADOW_VMCS;
2899 if (adjust_vmx_controls(min2, opt2,
2900 MSR_IA32_VMX_PROCBASED_CTLS2,
2901 &_cpu_based_2nd_exec_control) < 0)
2904 #ifndef CONFIG_X86_64
2905 if (!(_cpu_based_2nd_exec_control &
2906 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2907 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2910 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2911 _cpu_based_2nd_exec_control &= ~(
2912 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2913 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2914 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2916 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2917 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2919 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2920 CPU_BASED_CR3_STORE_EXITING |
2921 CPU_BASED_INVLPG_EXITING);
2922 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
2923 vmx_capability.ept, vmx_capability.vpid);
2926 min = VM_EXIT_SAVE_DEBUG_CONTROLS;
2927 #ifdef CONFIG_X86_64
2928 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2930 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT |
2931 VM_EXIT_ACK_INTR_ON_EXIT | VM_EXIT_CLEAR_BNDCFGS;
2932 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2933 &_vmexit_control) < 0)
2936 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2937 opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR;
2938 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2939 &_pin_based_exec_control) < 0)
2942 if (!(_cpu_based_2nd_exec_control &
2943 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) ||
2944 !(_vmexit_control & VM_EXIT_ACK_INTR_ON_EXIT))
2945 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2947 min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
2948 opt = VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS;
2949 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2950 &_vmentry_control) < 0)
2953 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2955 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2956 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2959 #ifdef CONFIG_X86_64
2960 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2961 if (vmx_msr_high & (1u<<16))
2965 /* Require Write-Back (WB) memory type for VMCS accesses. */
2966 if (((vmx_msr_high >> 18) & 15) != 6)
2969 vmcs_conf->size = vmx_msr_high & 0x1fff;
2970 vmcs_conf->order = get_order(vmcs_config.size);
2971 vmcs_conf->revision_id = vmx_msr_low;
2973 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2974 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2975 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2976 vmcs_conf->vmexit_ctrl = _vmexit_control;
2977 vmcs_conf->vmentry_ctrl = _vmentry_control;
2979 cpu_has_load_ia32_efer =
2980 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2981 VM_ENTRY_LOAD_IA32_EFER)
2982 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2983 VM_EXIT_LOAD_IA32_EFER);
2985 cpu_has_load_perf_global_ctrl =
2986 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2987 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
2988 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2989 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
2992 * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
2993 * but due to arrata below it can't be used. Workaround is to use
2994 * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2996 * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
3001 * BC86,AAY89,BD102 (model 44)
3005 if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
3006 switch (boot_cpu_data.x86_model) {
3012 cpu_has_load_perf_global_ctrl = false;
3013 printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
3014 "does not work properly. Using workaround\n");
3024 static struct vmcs *alloc_vmcs_cpu(int cpu)
3026 int node = cpu_to_node(cpu);
3030 pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
3033 vmcs = page_address(pages);
3034 memset(vmcs, 0, vmcs_config.size);
3035 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
3039 static struct vmcs *alloc_vmcs(void)
3041 return alloc_vmcs_cpu(raw_smp_processor_id());
3044 static void free_vmcs(struct vmcs *vmcs)
3046 free_pages((unsigned long)vmcs, vmcs_config.order);
3050 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
3052 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
3054 if (!loaded_vmcs->vmcs)
3056 loaded_vmcs_clear(loaded_vmcs);
3057 free_vmcs(loaded_vmcs->vmcs);
3058 loaded_vmcs->vmcs = NULL;
3061 static void free_kvm_area(void)
3065 for_each_possible_cpu(cpu) {
3066 free_vmcs(per_cpu(vmxarea, cpu));
3067 per_cpu(vmxarea, cpu) = NULL;
3071 static void init_vmcs_shadow_fields(void)
3075 /* No checks for read only fields yet */
3077 for (i = j = 0; i < max_shadow_read_write_fields; i++) {
3078 switch (shadow_read_write_fields[i]) {
3080 if (!vmx_mpx_supported())
3088 shadow_read_write_fields[j] =
3089 shadow_read_write_fields[i];
3092 max_shadow_read_write_fields = j;
3094 /* shadowed fields guest access without vmexit */
3095 for (i = 0; i < max_shadow_read_write_fields; i++) {
3096 clear_bit(shadow_read_write_fields[i],
3097 vmx_vmwrite_bitmap);
3098 clear_bit(shadow_read_write_fields[i],
3101 for (i = 0; i < max_shadow_read_only_fields; i++)
3102 clear_bit(shadow_read_only_fields[i],
3106 static __init int alloc_kvm_area(void)
3110 for_each_possible_cpu(cpu) {
3113 vmcs = alloc_vmcs_cpu(cpu);
3119 per_cpu(vmxarea, cpu) = vmcs;
3124 static bool emulation_required(struct kvm_vcpu *vcpu)
3126 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
3129 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
3130 struct kvm_segment *save)
3132 if (!emulate_invalid_guest_state) {
3134 * CS and SS RPL should be equal during guest entry according
3135 * to VMX spec, but in reality it is not always so. Since vcpu
3136 * is in the middle of the transition from real mode to
3137 * protected mode it is safe to assume that RPL 0 is a good
3140 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
3141 save->selector &= ~SELECTOR_RPL_MASK;
3142 save->dpl = save->selector & SELECTOR_RPL_MASK;
3145 vmx_set_segment(vcpu, save, seg);
3148 static void enter_pmode(struct kvm_vcpu *vcpu)
3150 unsigned long flags;
3151 struct vcpu_vmx *vmx = to_vmx(vcpu);
3154 * Update real mode segment cache. It may be not up-to-date if sement
3155 * register was written while vcpu was in a guest mode.
3157 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3158 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3159 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3160 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3161 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3162 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3164 vmx->rmode.vm86_active = 0;
3166 vmx_segment_cache_clear(vmx);
3168 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3170 flags = vmcs_readl(GUEST_RFLAGS);
3171 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
3172 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
3173 vmcs_writel(GUEST_RFLAGS, flags);
3175 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
3176 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
3178 update_exception_bitmap(vcpu);
3180 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3181 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3182 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3183 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3184 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3185 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3188 static void fix_rmode_seg(int seg, struct kvm_segment *save)
3190 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3191 struct kvm_segment var = *save;
3194 if (seg == VCPU_SREG_CS)
3197 if (!emulate_invalid_guest_state) {
3198 var.selector = var.base >> 4;
3199 var.base = var.base & 0xffff0;
3209 if (save->base & 0xf)
3210 printk_once(KERN_WARNING "kvm: segment base is not "
3211 "paragraph aligned when entering "
3212 "protected mode (seg=%d)", seg);
3215 vmcs_write16(sf->selector, var.selector);
3216 vmcs_write32(sf->base, var.base);
3217 vmcs_write32(sf->limit, var.limit);
3218 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
3221 static void enter_rmode(struct kvm_vcpu *vcpu)
3223 unsigned long flags;
3224 struct vcpu_vmx *vmx = to_vmx(vcpu);
3226 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3227 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3228 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3229 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3230 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3231 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3232 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3234 vmx->rmode.vm86_active = 1;
3237 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
3238 * vcpu. Warn the user that an update is overdue.
3240 if (!vcpu->kvm->arch.tss_addr)
3241 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
3242 "called before entering vcpu\n");
3244 vmx_segment_cache_clear(vmx);
3246 vmcs_writel(GUEST_TR_BASE, vcpu->kvm->arch.tss_addr);
3247 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
3248 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3250 flags = vmcs_readl(GUEST_RFLAGS);
3251 vmx->rmode.save_rflags = flags;
3253 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
3255 vmcs_writel(GUEST_RFLAGS, flags);
3256 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
3257 update_exception_bitmap(vcpu);
3259 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3260 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3261 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3262 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3263 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3264 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3266 kvm_mmu_reset_context(vcpu);
3269 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
3271 struct vcpu_vmx *vmx = to_vmx(vcpu);
3272 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
3278 * Force kernel_gs_base reloading before EFER changes, as control
3279 * of this msr depends on is_long_mode().
3281 vmx_load_host_state(to_vmx(vcpu));
3282 vcpu->arch.efer = efer;
3283 if (efer & EFER_LMA) {
3284 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3287 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3289 msr->data = efer & ~EFER_LME;
3294 #ifdef CONFIG_X86_64
3296 static void enter_lmode(struct kvm_vcpu *vcpu)
3300 vmx_segment_cache_clear(to_vmx(vcpu));
3302 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
3303 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
3304 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
3306 vmcs_write32(GUEST_TR_AR_BYTES,
3307 (guest_tr_ar & ~AR_TYPE_MASK)
3308 | AR_TYPE_BUSY_64_TSS);
3310 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
3313 static void exit_lmode(struct kvm_vcpu *vcpu)
3315 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3316 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
3321 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
3323 vpid_sync_context(to_vmx(vcpu));
3325 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
3327 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
3331 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
3333 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
3335 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
3336 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
3339 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
3341 if (enable_ept && is_paging(vcpu))
3342 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3343 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
3346 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
3348 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
3350 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
3351 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
3354 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
3356 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3358 if (!test_bit(VCPU_EXREG_PDPTR,
3359 (unsigned long *)&vcpu->arch.regs_dirty))
3362 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3363 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
3364 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
3365 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
3366 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
3370 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
3372 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3374 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3375 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
3376 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
3377 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
3378 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
3381 __set_bit(VCPU_EXREG_PDPTR,
3382 (unsigned long *)&vcpu->arch.regs_avail);
3383 __set_bit(VCPU_EXREG_PDPTR,
3384 (unsigned long *)&vcpu->arch.regs_dirty);
3387 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
3389 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
3391 struct kvm_vcpu *vcpu)
3393 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3394 vmx_decache_cr3(vcpu);
3395 if (!(cr0 & X86_CR0_PG)) {
3396 /* From paging/starting to nonpaging */
3397 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3398 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
3399 (CPU_BASED_CR3_LOAD_EXITING |
3400 CPU_BASED_CR3_STORE_EXITING));
3401 vcpu->arch.cr0 = cr0;
3402 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3403 } else if (!is_paging(vcpu)) {
3404 /* From nonpaging to paging */
3405 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3406 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
3407 ~(CPU_BASED_CR3_LOAD_EXITING |
3408 CPU_BASED_CR3_STORE_EXITING));
3409 vcpu->arch.cr0 = cr0;
3410 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3413 if (!(cr0 & X86_CR0_WP))
3414 *hw_cr0 &= ~X86_CR0_WP;
3417 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
3419 struct vcpu_vmx *vmx = to_vmx(vcpu);
3420 unsigned long hw_cr0;
3422 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK);
3423 if (enable_unrestricted_guest)
3424 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
3426 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
3428 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
3431 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
3435 #ifdef CONFIG_X86_64
3436 if (vcpu->arch.efer & EFER_LME) {
3437 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
3439 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
3445 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
3447 if (!vcpu->fpu_active)
3448 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
3450 vmcs_writel(CR0_READ_SHADOW, cr0);
3451 vmcs_writel(GUEST_CR0, hw_cr0);
3452 vcpu->arch.cr0 = cr0;
3454 /* depends on vcpu->arch.cr0 to be set to a new value */
3455 vmx->emulation_required = emulation_required(vcpu);
3458 static u64 construct_eptp(unsigned long root_hpa)
3462 /* TODO write the value reading from MSR */
3463 eptp = VMX_EPT_DEFAULT_MT |
3464 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
3465 if (enable_ept_ad_bits)
3466 eptp |= VMX_EPT_AD_ENABLE_BIT;
3467 eptp |= (root_hpa & PAGE_MASK);
3472 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
3474 unsigned long guest_cr3;
3479 eptp = construct_eptp(cr3);
3480 vmcs_write64(EPT_POINTER, eptp);
3481 if (is_paging(vcpu) || is_guest_mode(vcpu))
3482 guest_cr3 = kvm_read_cr3(vcpu);
3484 guest_cr3 = vcpu->kvm->arch.ept_identity_map_addr;
3485 ept_load_pdptrs(vcpu);
3488 vmx_flush_tlb(vcpu);
3489 vmcs_writel(GUEST_CR3, guest_cr3);
3492 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3494 unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
3495 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
3497 if (cr4 & X86_CR4_VMXE) {
3499 * To use VMXON (and later other VMX instructions), a guest
3500 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3501 * So basically the check on whether to allow nested VMX
3504 if (!nested_vmx_allowed(vcpu))
3507 if (to_vmx(vcpu)->nested.vmxon &&
3508 ((cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON))
3511 vcpu->arch.cr4 = cr4;
3513 if (!is_paging(vcpu)) {
3514 hw_cr4 &= ~X86_CR4_PAE;
3515 hw_cr4 |= X86_CR4_PSE;
3517 * SMEP/SMAP is disabled if CPU is in non-paging mode
3518 * in hardware. However KVM always uses paging mode to
3519 * emulate guest non-paging mode with TDP.
3520 * To emulate this behavior, SMEP/SMAP needs to be
3521 * manually disabled when guest switches to non-paging
3524 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP);
3525 } else if (!(cr4 & X86_CR4_PAE)) {
3526 hw_cr4 &= ~X86_CR4_PAE;
3530 vmcs_writel(CR4_READ_SHADOW, cr4);
3531 vmcs_writel(GUEST_CR4, hw_cr4);
3535 static void vmx_get_segment(struct kvm_vcpu *vcpu,
3536 struct kvm_segment *var, int seg)
3538 struct vcpu_vmx *vmx = to_vmx(vcpu);
3541 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3542 *var = vmx->rmode.segs[seg];
3543 if (seg == VCPU_SREG_TR
3544 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3546 var->base = vmx_read_guest_seg_base(vmx, seg);
3547 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3550 var->base = vmx_read_guest_seg_base(vmx, seg);
3551 var->limit = vmx_read_guest_seg_limit(vmx, seg);
3552 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3553 ar = vmx_read_guest_seg_ar(vmx, seg);
3554 var->unusable = (ar >> 16) & 1;
3555 var->type = ar & 15;
3556 var->s = (ar >> 4) & 1;
3557 var->dpl = (ar >> 5) & 3;
3559 * Some userspaces do not preserve unusable property. Since usable
3560 * segment has to be present according to VMX spec we can use present
3561 * property to amend userspace bug by making unusable segment always
3562 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3563 * segment as unusable.
3565 var->present = !var->unusable;
3566 var->avl = (ar >> 12) & 1;
3567 var->l = (ar >> 13) & 1;
3568 var->db = (ar >> 14) & 1;
3569 var->g = (ar >> 15) & 1;
3572 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3574 struct kvm_segment s;
3576 if (to_vmx(vcpu)->rmode.vm86_active) {
3577 vmx_get_segment(vcpu, &s, seg);
3580 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3583 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
3585 struct vcpu_vmx *vmx = to_vmx(vcpu);
3587 if (unlikely(vmx->rmode.vm86_active))
3590 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3595 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3599 if (var->unusable || !var->present)
3602 ar = var->type & 15;
3603 ar |= (var->s & 1) << 4;
3604 ar |= (var->dpl & 3) << 5;
3605 ar |= (var->present & 1) << 7;
3606 ar |= (var->avl & 1) << 12;
3607 ar |= (var->l & 1) << 13;
3608 ar |= (var->db & 1) << 14;
3609 ar |= (var->g & 1) << 15;
3615 static void vmx_set_segment(struct kvm_vcpu *vcpu,
3616 struct kvm_segment *var, int seg)
3618 struct vcpu_vmx *vmx = to_vmx(vcpu);
3619 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3621 vmx_segment_cache_clear(vmx);
3623 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3624 vmx->rmode.segs[seg] = *var;
3625 if (seg == VCPU_SREG_TR)
3626 vmcs_write16(sf->selector, var->selector);
3628 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3632 vmcs_writel(sf->base, var->base);
3633 vmcs_write32(sf->limit, var->limit);
3634 vmcs_write16(sf->selector, var->selector);
3637 * Fix the "Accessed" bit in AR field of segment registers for older
3639 * IA32 arch specifies that at the time of processor reset the
3640 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3641 * is setting it to 0 in the userland code. This causes invalid guest
3642 * state vmexit when "unrestricted guest" mode is turned on.
3643 * Fix for this setup issue in cpu_reset is being pushed in the qemu
3644 * tree. Newer qemu binaries with that qemu fix would not need this
3647 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3648 var->type |= 0x1; /* Accessed */
3650 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3653 vmx->emulation_required = emulation_required(vcpu);
3656 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3658 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3660 *db = (ar >> 14) & 1;
3661 *l = (ar >> 13) & 1;
3664 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3666 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3667 dt->address = vmcs_readl(GUEST_IDTR_BASE);
3670 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3672 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3673 vmcs_writel(GUEST_IDTR_BASE, dt->address);
3676 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3678 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3679 dt->address = vmcs_readl(GUEST_GDTR_BASE);
3682 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3684 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3685 vmcs_writel(GUEST_GDTR_BASE, dt->address);
3688 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3690 struct kvm_segment var;
3693 vmx_get_segment(vcpu, &var, seg);
3695 if (seg == VCPU_SREG_CS)
3697 ar = vmx_segment_access_rights(&var);
3699 if (var.base != (var.selector << 4))
3701 if (var.limit != 0xffff)
3709 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3711 struct kvm_segment cs;
3712 unsigned int cs_rpl;
3714 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3715 cs_rpl = cs.selector & SELECTOR_RPL_MASK;
3719 if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
3723 if (cs.type & AR_TYPE_WRITEABLE_MASK) {
3724 if (cs.dpl > cs_rpl)
3727 if (cs.dpl != cs_rpl)
3733 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3737 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3739 struct kvm_segment ss;
3740 unsigned int ss_rpl;
3742 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3743 ss_rpl = ss.selector & SELECTOR_RPL_MASK;
3747 if (ss.type != 3 && ss.type != 7)
3751 if (ss.dpl != ss_rpl) /* DPL != RPL */
3759 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3761 struct kvm_segment var;
3764 vmx_get_segment(vcpu, &var, seg);
3765 rpl = var.selector & SELECTOR_RPL_MASK;
3773 if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
3774 if (var.dpl < rpl) /* DPL < RPL */
3778 /* TODO: Add other members to kvm_segment_field to allow checking for other access
3784 static bool tr_valid(struct kvm_vcpu *vcpu)
3786 struct kvm_segment tr;
3788 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3792 if (tr.selector & SELECTOR_TI_MASK) /* TI = 1 */
3794 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3802 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3804 struct kvm_segment ldtr;
3806 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3810 if (ldtr.selector & SELECTOR_TI_MASK) /* TI = 1 */
3820 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3822 struct kvm_segment cs, ss;
3824 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3825 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3827 return ((cs.selector & SELECTOR_RPL_MASK) ==
3828 (ss.selector & SELECTOR_RPL_MASK));
3832 * Check if guest state is valid. Returns true if valid, false if
3834 * We assume that registers are always usable
3836 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3838 if (enable_unrestricted_guest)
3841 /* real mode guest state checks */
3842 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3843 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3845 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3847 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3849 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3851 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3853 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3856 /* protected mode guest state checks */
3857 if (!cs_ss_rpl_check(vcpu))
3859 if (!code_segment_valid(vcpu))
3861 if (!stack_segment_valid(vcpu))
3863 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3865 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3867 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3869 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3871 if (!tr_valid(vcpu))
3873 if (!ldtr_valid(vcpu))
3877 * - Add checks on RIP
3878 * - Add checks on RFLAGS
3884 static int init_rmode_tss(struct kvm *kvm)
3890 idx = srcu_read_lock(&kvm->srcu);
3891 fn = kvm->arch.tss_addr >> PAGE_SHIFT;
3892 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3895 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3896 r = kvm_write_guest_page(kvm, fn++, &data,
3897 TSS_IOPB_BASE_OFFSET, sizeof(u16));
3900 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3903 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3907 r = kvm_write_guest_page(kvm, fn, &data,
3908 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3911 srcu_read_unlock(&kvm->srcu, idx);
3915 static int init_rmode_identity_map(struct kvm *kvm)
3918 pfn_t identity_map_pfn;
3924 /* Protect kvm->arch.ept_identity_pagetable_done. */
3925 mutex_lock(&kvm->slots_lock);
3927 if (likely(kvm->arch.ept_identity_pagetable_done))
3930 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
3932 r = alloc_identity_pagetable(kvm);
3936 idx = srcu_read_lock(&kvm->srcu);
3937 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3940 /* Set up identity-mapping pagetable for EPT in real mode */
3941 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3942 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3943 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3944 r = kvm_write_guest_page(kvm, identity_map_pfn,
3945 &tmp, i * sizeof(tmp), sizeof(tmp));
3949 kvm->arch.ept_identity_pagetable_done = true;
3952 srcu_read_unlock(&kvm->srcu, idx);
3955 mutex_unlock(&kvm->slots_lock);
3959 static void seg_setup(int seg)
3961 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3964 vmcs_write16(sf->selector, 0);
3965 vmcs_writel(sf->base, 0);
3966 vmcs_write32(sf->limit, 0xffff);
3968 if (seg == VCPU_SREG_CS)
3969 ar |= 0x08; /* code segment */
3971 vmcs_write32(sf->ar_bytes, ar);
3974 static int alloc_apic_access_page(struct kvm *kvm)
3977 struct kvm_userspace_memory_region kvm_userspace_mem;
3980 mutex_lock(&kvm->slots_lock);
3981 if (kvm->arch.apic_access_page_done)
3983 kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
3984 kvm_userspace_mem.flags = 0;
3985 kvm_userspace_mem.guest_phys_addr = APIC_DEFAULT_PHYS_BASE;
3986 kvm_userspace_mem.memory_size = PAGE_SIZE;
3987 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem);
3991 page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
3992 if (is_error_page(page)) {
3998 * Do not pin the page in memory, so that memory hot-unplug
3999 * is able to migrate it.
4002 kvm->arch.apic_access_page_done = true;
4004 mutex_unlock(&kvm->slots_lock);
4008 static int alloc_identity_pagetable(struct kvm *kvm)
4010 /* Called with kvm->slots_lock held. */
4012 struct kvm_userspace_memory_region kvm_userspace_mem;
4015 BUG_ON(kvm->arch.ept_identity_pagetable_done);
4017 kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
4018 kvm_userspace_mem.flags = 0;
4019 kvm_userspace_mem.guest_phys_addr =
4020 kvm->arch.ept_identity_map_addr;
4021 kvm_userspace_mem.memory_size = PAGE_SIZE;
4022 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem);
4027 static void allocate_vpid(struct vcpu_vmx *vmx)
4034 spin_lock(&vmx_vpid_lock);
4035 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
4036 if (vpid < VMX_NR_VPIDS) {
4038 __set_bit(vpid, vmx_vpid_bitmap);
4040 spin_unlock(&vmx_vpid_lock);
4043 static void free_vpid(struct vcpu_vmx *vmx)
4047 spin_lock(&vmx_vpid_lock);
4049 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
4050 spin_unlock(&vmx_vpid_lock);
4053 #define MSR_TYPE_R 1
4054 #define MSR_TYPE_W 2
4055 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
4058 int f = sizeof(unsigned long);
4060 if (!cpu_has_vmx_msr_bitmap())
4064 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4065 * have the write-low and read-high bitmap offsets the wrong way round.
4066 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4068 if (msr <= 0x1fff) {
4069 if (type & MSR_TYPE_R)
4071 __clear_bit(msr, msr_bitmap + 0x000 / f);
4073 if (type & MSR_TYPE_W)
4075 __clear_bit(msr, msr_bitmap + 0x800 / f);
4077 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4079 if (type & MSR_TYPE_R)
4081 __clear_bit(msr, msr_bitmap + 0x400 / f);
4083 if (type & MSR_TYPE_W)
4085 __clear_bit(msr, msr_bitmap + 0xc00 / f);
4090 static void __vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
4093 int f = sizeof(unsigned long);
4095 if (!cpu_has_vmx_msr_bitmap())
4099 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4100 * have the write-low and read-high bitmap offsets the wrong way round.
4101 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4103 if (msr <= 0x1fff) {
4104 if (type & MSR_TYPE_R)
4106 __set_bit(msr, msr_bitmap + 0x000 / f);
4108 if (type & MSR_TYPE_W)
4110 __set_bit(msr, msr_bitmap + 0x800 / f);
4112 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4114 if (type & MSR_TYPE_R)
4116 __set_bit(msr, msr_bitmap + 0x400 / f);
4118 if (type & MSR_TYPE_W)
4120 __set_bit(msr, msr_bitmap + 0xc00 / f);
4125 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
4128 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy,
4129 msr, MSR_TYPE_R | MSR_TYPE_W);
4130 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode,
4131 msr, MSR_TYPE_R | MSR_TYPE_W);
4134 static void vmx_enable_intercept_msr_read_x2apic(u32 msr)
4136 __vmx_enable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4138 __vmx_enable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4142 static void vmx_disable_intercept_msr_read_x2apic(u32 msr)
4144 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4146 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4150 static void vmx_disable_intercept_msr_write_x2apic(u32 msr)
4152 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4154 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4158 static int vmx_vm_has_apicv(struct kvm *kvm)
4160 return enable_apicv && irqchip_in_kernel(kvm);
4164 * Send interrupt to vcpu via posted interrupt way.
4165 * 1. If target vcpu is running(non-root mode), send posted interrupt
4166 * notification to vcpu and hardware will sync PIR to vIRR atomically.
4167 * 2. If target vcpu isn't running(root mode), kick it to pick up the
4168 * interrupt from PIR in next vmentry.
4170 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
4172 struct vcpu_vmx *vmx = to_vmx(vcpu);
4175 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
4178 r = pi_test_and_set_on(&vmx->pi_desc);
4179 kvm_make_request(KVM_REQ_EVENT, vcpu);
4181 if (!r && (vcpu->mode == IN_GUEST_MODE))
4182 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu),
4183 POSTED_INTR_VECTOR);
4186 kvm_vcpu_kick(vcpu);
4189 static void vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
4191 struct vcpu_vmx *vmx = to_vmx(vcpu);
4193 if (!pi_test_and_clear_on(&vmx->pi_desc))
4196 kvm_apic_update_irr(vcpu, vmx->pi_desc.pir);
4199 static void vmx_sync_pir_to_irr_dummy(struct kvm_vcpu *vcpu)
4205 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
4206 * will not change in the lifetime of the guest.
4207 * Note that host-state that does change is set elsewhere. E.g., host-state
4208 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
4210 static void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
4217 vmcs_writel(HOST_CR0, read_cr0() & ~X86_CR0_TS); /* 22.2.3 */
4218 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
4220 /* Save the most likely value for this task's CR4 in the VMCS. */
4222 vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
4223 vmx->host_state.vmcs_host_cr4 = cr4;
4225 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
4226 #ifdef CONFIG_X86_64
4228 * Load null selectors, so we can avoid reloading them in
4229 * __vmx_load_host_state(), in case userspace uses the null selectors
4230 * too (the expected case).
4232 vmcs_write16(HOST_DS_SELECTOR, 0);
4233 vmcs_write16(HOST_ES_SELECTOR, 0);
4235 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4236 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4238 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4239 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
4241 native_store_idt(&dt);
4242 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
4243 vmx->host_idt_base = dt.address;
4245 vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
4247 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
4248 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
4249 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
4250 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
4252 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
4253 rdmsr(MSR_IA32_CR_PAT, low32, high32);
4254 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
4258 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
4260 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
4262 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
4263 if (is_guest_mode(&vmx->vcpu))
4264 vmx->vcpu.arch.cr4_guest_owned_bits &=
4265 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
4266 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
4269 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
4271 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
4273 if (!vmx_vm_has_apicv(vmx->vcpu.kvm))
4274 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
4275 return pin_based_exec_ctrl;
4278 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
4280 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
4282 if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
4283 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
4285 if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
4286 exec_control &= ~CPU_BASED_TPR_SHADOW;
4287 #ifdef CONFIG_X86_64
4288 exec_control |= CPU_BASED_CR8_STORE_EXITING |
4289 CPU_BASED_CR8_LOAD_EXITING;
4293 exec_control |= CPU_BASED_CR3_STORE_EXITING |
4294 CPU_BASED_CR3_LOAD_EXITING |
4295 CPU_BASED_INVLPG_EXITING;
4296 return exec_control;
4299 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
4301 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4302 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
4303 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4305 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4307 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4308 enable_unrestricted_guest = 0;
4309 /* Enable INVPCID for non-ept guests may cause performance regression. */
4310 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4312 if (!enable_unrestricted_guest)
4313 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4315 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4316 if (!vmx_vm_has_apicv(vmx->vcpu.kvm))
4317 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4318 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4319 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4320 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4322 We can NOT enable shadow_vmcs here because we don't have yet
4325 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4326 return exec_control;
4329 static void ept_set_mmio_spte_mask(void)
4332 * EPT Misconfigurations can be generated if the value of bits 2:0
4333 * of an EPT paging-structure entry is 110b (write/execute).
4334 * Also, magic bits (0x3ull << 62) is set to quickly identify mmio
4337 kvm_mmu_set_mmio_spte_mask((0x3ull << 62) | 0x6ull);
4341 * Sets up the vmcs for emulated real mode.
4343 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
4345 #ifdef CONFIG_X86_64
4351 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
4352 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
4354 if (enable_shadow_vmcs) {
4355 vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
4356 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
4358 if (cpu_has_vmx_msr_bitmap())
4359 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
4361 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4364 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
4366 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
4368 if (cpu_has_secondary_exec_ctrls()) {
4369 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
4370 vmx_secondary_exec_control(vmx));
4373 if (vmx_vm_has_apicv(vmx->vcpu.kvm)) {
4374 vmcs_write64(EOI_EXIT_BITMAP0, 0);
4375 vmcs_write64(EOI_EXIT_BITMAP1, 0);
4376 vmcs_write64(EOI_EXIT_BITMAP2, 0);
4377 vmcs_write64(EOI_EXIT_BITMAP3, 0);
4379 vmcs_write16(GUEST_INTR_STATUS, 0);
4381 vmcs_write64(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4382 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4386 vmcs_write32(PLE_GAP, ple_gap);
4387 vmx->ple_window = ple_window;
4388 vmx->ple_window_dirty = true;
4391 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4392 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4393 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
4395 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
4396 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
4397 vmx_set_constant_host_state(vmx);
4398 #ifdef CONFIG_X86_64
4399 rdmsrl(MSR_FS_BASE, a);
4400 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
4401 rdmsrl(MSR_GS_BASE, a);
4402 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
4404 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4405 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4408 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4409 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4410 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
4411 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4412 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
4414 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
4415 u32 msr_low, msr_high;
4417 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
4418 host_pat = msr_low | ((u64) msr_high << 32);
4419 /* Write the default value follow host pat */
4420 vmcs_write64(GUEST_IA32_PAT, host_pat);
4421 /* Keep arch.pat sync with GUEST_IA32_PAT */
4422 vmx->vcpu.arch.pat = host_pat;
4425 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
4426 u32 index = vmx_msr_index[i];
4427 u32 data_low, data_high;
4430 if (rdmsr_safe(index, &data_low, &data_high) < 0)
4432 if (wrmsr_safe(index, data_low, data_high) < 0)
4434 vmx->guest_msrs[j].index = i;
4435 vmx->guest_msrs[j].data = 0;
4436 vmx->guest_msrs[j].mask = -1ull;
4441 vm_exit_controls_init(vmx, vmcs_config.vmexit_ctrl);
4443 /* 22.2.1, 20.8.1 */
4444 vm_entry_controls_init(vmx, vmcs_config.vmentry_ctrl);
4446 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
4447 set_cr4_guest_host_mask(vmx);
4452 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu)
4454 struct vcpu_vmx *vmx = to_vmx(vcpu);
4455 struct msr_data apic_base_msr;
4457 vmx->rmode.vm86_active = 0;
4459 vmx->soft_vnmi_blocked = 0;
4461 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4462 kvm_set_cr8(&vmx->vcpu, 0);
4463 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE | MSR_IA32_APICBASE_ENABLE;
4464 if (kvm_vcpu_is_bsp(&vmx->vcpu))
4465 apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4466 apic_base_msr.host_initiated = true;
4467 kvm_set_apic_base(&vmx->vcpu, &apic_base_msr);
4469 vmx_segment_cache_clear(vmx);
4471 seg_setup(VCPU_SREG_CS);
4472 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4473 vmcs_write32(GUEST_CS_BASE, 0xffff0000);
4475 seg_setup(VCPU_SREG_DS);
4476 seg_setup(VCPU_SREG_ES);
4477 seg_setup(VCPU_SREG_FS);
4478 seg_setup(VCPU_SREG_GS);
4479 seg_setup(VCPU_SREG_SS);
4481 vmcs_write16(GUEST_TR_SELECTOR, 0);
4482 vmcs_writel(GUEST_TR_BASE, 0);
4483 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4484 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4486 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4487 vmcs_writel(GUEST_LDTR_BASE, 0);
4488 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4489 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4491 vmcs_write32(GUEST_SYSENTER_CS, 0);
4492 vmcs_writel(GUEST_SYSENTER_ESP, 0);
4493 vmcs_writel(GUEST_SYSENTER_EIP, 0);
4495 vmcs_writel(GUEST_RFLAGS, 0x02);
4496 kvm_rip_write(vcpu, 0xfff0);
4498 vmcs_writel(GUEST_GDTR_BASE, 0);
4499 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4501 vmcs_writel(GUEST_IDTR_BASE, 0);
4502 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4504 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4505 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4506 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4508 /* Special registers */
4509 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4513 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
4515 if (cpu_has_vmx_tpr_shadow()) {
4516 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4517 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
4518 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4519 __pa(vmx->vcpu.arch.apic->regs));
4520 vmcs_write32(TPR_THRESHOLD, 0);
4523 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4525 if (vmx_vm_has_apicv(vcpu->kvm))
4526 memset(&vmx->pi_desc, 0, sizeof(struct pi_desc));
4529 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4531 vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4532 vmx_set_cr0(&vmx->vcpu, kvm_read_cr0(vcpu)); /* enter rmode */
4533 vmx_set_cr4(&vmx->vcpu, 0);
4534 vmx_set_efer(&vmx->vcpu, 0);
4535 vmx_fpu_activate(&vmx->vcpu);
4536 update_exception_bitmap(&vmx->vcpu);
4538 vpid_sync_context(vmx);
4542 * In nested virtualization, check if L1 asked to exit on external interrupts.
4543 * For most existing hypervisors, this will always return true.
4545 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
4547 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
4548 PIN_BASED_EXT_INTR_MASK;
4552 * In nested virtualization, check if L1 has set
4553 * VM_EXIT_ACK_INTR_ON_EXIT
4555 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
4557 return get_vmcs12(vcpu)->vm_exit_controls &
4558 VM_EXIT_ACK_INTR_ON_EXIT;
4561 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
4563 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
4564 PIN_BASED_NMI_EXITING;
4567 static void enable_irq_window(struct kvm_vcpu *vcpu)
4569 u32 cpu_based_vm_exec_control;
4571 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4572 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
4573 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4576 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4578 u32 cpu_based_vm_exec_control;
4580 if (!cpu_has_virtual_nmis() ||
4581 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4582 enable_irq_window(vcpu);
4586 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4587 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
4588 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4591 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4593 struct vcpu_vmx *vmx = to_vmx(vcpu);
4595 int irq = vcpu->arch.interrupt.nr;
4597 trace_kvm_inj_virq(irq);
4599 ++vcpu->stat.irq_injections;
4600 if (vmx->rmode.vm86_active) {
4602 if (vcpu->arch.interrupt.soft)
4603 inc_eip = vcpu->arch.event_exit_inst_len;
4604 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
4605 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4608 intr = irq | INTR_INFO_VALID_MASK;
4609 if (vcpu->arch.interrupt.soft) {
4610 intr |= INTR_TYPE_SOFT_INTR;
4611 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4612 vmx->vcpu.arch.event_exit_inst_len);
4614 intr |= INTR_TYPE_EXT_INTR;
4615 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4618 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4620 struct vcpu_vmx *vmx = to_vmx(vcpu);
4622 if (is_guest_mode(vcpu))
4625 if (!cpu_has_virtual_nmis()) {
4627 * Tracking the NMI-blocked state in software is built upon
4628 * finding the next open IRQ window. This, in turn, depends on
4629 * well-behaving guests: They have to keep IRQs disabled at
4630 * least as long as the NMI handler runs. Otherwise we may
4631 * cause NMI nesting, maybe breaking the guest. But as this is
4632 * highly unlikely, we can live with the residual risk.
4634 vmx->soft_vnmi_blocked = 1;
4635 vmx->vnmi_blocked_time = 0;
4638 ++vcpu->stat.nmi_injections;
4639 vmx->nmi_known_unmasked = false;
4640 if (vmx->rmode.vm86_active) {
4641 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
4642 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4645 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4646 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4649 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4651 if (!cpu_has_virtual_nmis())
4652 return to_vmx(vcpu)->soft_vnmi_blocked;
4653 if (to_vmx(vcpu)->nmi_known_unmasked)
4655 return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4658 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4660 struct vcpu_vmx *vmx = to_vmx(vcpu);
4662 if (!cpu_has_virtual_nmis()) {
4663 if (vmx->soft_vnmi_blocked != masked) {
4664 vmx->soft_vnmi_blocked = masked;
4665 vmx->vnmi_blocked_time = 0;
4668 vmx->nmi_known_unmasked = !masked;
4670 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4671 GUEST_INTR_STATE_NMI);
4673 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4674 GUEST_INTR_STATE_NMI);
4678 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4680 if (to_vmx(vcpu)->nested.nested_run_pending)
4683 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
4686 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4687 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4688 | GUEST_INTR_STATE_NMI));
4691 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4693 return (!to_vmx(vcpu)->nested.nested_run_pending &&
4694 vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4695 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4696 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4699 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4702 struct kvm_userspace_memory_region tss_mem = {
4703 .slot = TSS_PRIVATE_MEMSLOT,
4704 .guest_phys_addr = addr,
4705 .memory_size = PAGE_SIZE * 3,
4709 ret = kvm_set_memory_region(kvm, &tss_mem);
4712 kvm->arch.tss_addr = addr;
4713 return init_rmode_tss(kvm);
4716 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4721 * Update instruction length as we may reinject the exception
4722 * from user space while in guest debugging mode.
4724 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4725 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4726 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4730 if (vcpu->guest_debug &
4731 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4748 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4749 int vec, u32 err_code)
4752 * Instruction with address size override prefix opcode 0x67
4753 * Cause the #SS fault with 0 error code in VM86 mode.
4755 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4756 if (emulate_instruction(vcpu, 0) == EMULATE_DONE) {
4757 if (vcpu->arch.halt_request) {
4758 vcpu->arch.halt_request = 0;
4759 return kvm_emulate_halt(vcpu);
4767 * Forward all other exceptions that are valid in real mode.
4768 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4769 * the required debugging infrastructure rework.
4771 kvm_queue_exception(vcpu, vec);
4776 * Trigger machine check on the host. We assume all the MSRs are already set up
4777 * by the CPU and that we still run on the same CPU as the MCE occurred on.
4778 * We pass a fake environment to the machine check handler because we want
4779 * the guest to be always treated like user space, no matter what context
4780 * it used internally.
4782 static void kvm_machine_check(void)
4784 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4785 struct pt_regs regs = {
4786 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
4787 .flags = X86_EFLAGS_IF,
4790 do_machine_check(®s, 0);
4794 static int handle_machine_check(struct kvm_vcpu *vcpu)
4796 /* already handled by vcpu_run */
4800 static int handle_exception(struct kvm_vcpu *vcpu)
4802 struct vcpu_vmx *vmx = to_vmx(vcpu);
4803 struct kvm_run *kvm_run = vcpu->run;
4804 u32 intr_info, ex_no, error_code;
4805 unsigned long cr2, rip, dr6;
4807 enum emulation_result er;
4809 vect_info = vmx->idt_vectoring_info;
4810 intr_info = vmx->exit_intr_info;
4812 if (is_machine_check(intr_info))
4813 return handle_machine_check(vcpu);
4815 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
4816 return 1; /* already handled by vmx_vcpu_run() */
4818 if (is_no_device(intr_info)) {
4819 vmx_fpu_activate(vcpu);
4823 if (is_invalid_opcode(intr_info)) {
4824 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
4825 if (er != EMULATE_DONE)
4826 kvm_queue_exception(vcpu, UD_VECTOR);
4831 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4832 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4835 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4836 * MMIO, it is better to report an internal error.
4837 * See the comments in vmx_handle_exit.
4839 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4840 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4841 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4842 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4843 vcpu->run->internal.ndata = 2;
4844 vcpu->run->internal.data[0] = vect_info;
4845 vcpu->run->internal.data[1] = intr_info;
4849 if (is_page_fault(intr_info)) {
4850 /* EPT won't cause page fault directly */
4852 cr2 = vmcs_readl(EXIT_QUALIFICATION);
4853 trace_kvm_page_fault(cr2, error_code);
4855 if (kvm_event_needs_reinjection(vcpu))
4856 kvm_mmu_unprotect_page_virt(vcpu, cr2);
4857 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
4860 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4862 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4863 return handle_rmode_exception(vcpu, ex_no, error_code);
4867 dr6 = vmcs_readl(EXIT_QUALIFICATION);
4868 if (!(vcpu->guest_debug &
4869 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4870 vcpu->arch.dr6 &= ~15;
4871 vcpu->arch.dr6 |= dr6 | DR6_RTM;
4872 if (!(dr6 & ~DR6_RESERVED)) /* icebp */
4873 skip_emulated_instruction(vcpu);
4875 kvm_queue_exception(vcpu, DB_VECTOR);
4878 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4879 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4883 * Update instruction length as we may reinject #BP from
4884 * user space while in guest debugging mode. Reading it for
4885 * #DB as well causes no harm, it is not used in that case.
4887 vmx->vcpu.arch.event_exit_inst_len =
4888 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4889 kvm_run->exit_reason = KVM_EXIT_DEBUG;
4890 rip = kvm_rip_read(vcpu);
4891 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4892 kvm_run->debug.arch.exception = ex_no;
4895 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4896 kvm_run->ex.exception = ex_no;
4897 kvm_run->ex.error_code = error_code;
4903 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
4905 ++vcpu->stat.irq_exits;
4909 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4911 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4915 static int handle_io(struct kvm_vcpu *vcpu)
4917 unsigned long exit_qualification;
4918 int size, in, string;
4921 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4922 string = (exit_qualification & 16) != 0;
4923 in = (exit_qualification & 8) != 0;
4925 ++vcpu->stat.io_exits;
4928 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4930 port = exit_qualification >> 16;
4931 size = (exit_qualification & 7) + 1;
4932 skip_emulated_instruction(vcpu);
4934 return kvm_fast_pio_out(vcpu, size, port);
4938 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4941 * Patch in the VMCALL instruction:
4943 hypercall[0] = 0x0f;
4944 hypercall[1] = 0x01;
4945 hypercall[2] = 0xc1;
4948 static bool nested_cr0_valid(struct vmcs12 *vmcs12, unsigned long val)
4950 unsigned long always_on = VMXON_CR0_ALWAYSON;
4952 if (nested_vmx_secondary_ctls_high &
4953 SECONDARY_EXEC_UNRESTRICTED_GUEST &&
4954 nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
4955 always_on &= ~(X86_CR0_PE | X86_CR0_PG);
4956 return (val & always_on) == always_on;
4959 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4960 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4962 if (is_guest_mode(vcpu)) {
4963 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4964 unsigned long orig_val = val;
4967 * We get here when L2 changed cr0 in a way that did not change
4968 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4969 * but did change L0 shadowed bits. So we first calculate the
4970 * effective cr0 value that L1 would like to write into the
4971 * hardware. It consists of the L2-owned bits from the new
4972 * value combined with the L1-owned bits from L1's guest_cr0.
4974 val = (val & ~vmcs12->cr0_guest_host_mask) |
4975 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
4977 if (!nested_cr0_valid(vmcs12, val))
4980 if (kvm_set_cr0(vcpu, val))
4982 vmcs_writel(CR0_READ_SHADOW, orig_val);
4985 if (to_vmx(vcpu)->nested.vmxon &&
4986 ((val & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON))
4988 return kvm_set_cr0(vcpu, val);
4992 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4994 if (is_guest_mode(vcpu)) {
4995 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4996 unsigned long orig_val = val;
4998 /* analogously to handle_set_cr0 */
4999 val = (val & ~vmcs12->cr4_guest_host_mask) |
5000 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
5001 if (kvm_set_cr4(vcpu, val))
5003 vmcs_writel(CR4_READ_SHADOW, orig_val);
5006 return kvm_set_cr4(vcpu, val);
5009 /* called to set cr0 as approriate for clts instruction exit. */
5010 static void handle_clts(struct kvm_vcpu *vcpu)
5012 if (is_guest_mode(vcpu)) {
5014 * We get here when L2 did CLTS, and L1 didn't shadow CR0.TS
5015 * but we did (!fpu_active). We need to keep GUEST_CR0.TS on,
5016 * just pretend it's off (also in arch.cr0 for fpu_activate).
5018 vmcs_writel(CR0_READ_SHADOW,
5019 vmcs_readl(CR0_READ_SHADOW) & ~X86_CR0_TS);
5020 vcpu->arch.cr0 &= ~X86_CR0_TS;
5022 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
5025 static int handle_cr(struct kvm_vcpu *vcpu)
5027 unsigned long exit_qualification, val;
5032 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5033 cr = exit_qualification & 15;
5034 reg = (exit_qualification >> 8) & 15;
5035 switch ((exit_qualification >> 4) & 3) {
5036 case 0: /* mov to cr */
5037 val = kvm_register_readl(vcpu, reg);
5038 trace_kvm_cr_write(cr, val);
5041 err = handle_set_cr0(vcpu, val);
5042 kvm_complete_insn_gp(vcpu, err);
5045 err = kvm_set_cr3(vcpu, val);
5046 kvm_complete_insn_gp(vcpu, err);
5049 err = handle_set_cr4(vcpu, val);
5050 kvm_complete_insn_gp(vcpu, err);
5053 u8 cr8_prev = kvm_get_cr8(vcpu);
5055 err = kvm_set_cr8(vcpu, cr8);
5056 kvm_complete_insn_gp(vcpu, err);
5057 if (irqchip_in_kernel(vcpu->kvm))
5059 if (cr8_prev <= cr8)
5061 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
5068 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
5069 skip_emulated_instruction(vcpu);
5070 vmx_fpu_activate(vcpu);
5072 case 1: /*mov from cr*/
5075 val = kvm_read_cr3(vcpu);
5076 kvm_register_write(vcpu, reg, val);
5077 trace_kvm_cr_read(cr, val);
5078 skip_emulated_instruction(vcpu);
5081 val = kvm_get_cr8(vcpu);
5082 kvm_register_write(vcpu, reg, val);
5083 trace_kvm_cr_read(cr, val);
5084 skip_emulated_instruction(vcpu);
5089 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5090 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
5091 kvm_lmsw(vcpu, val);
5093 skip_emulated_instruction(vcpu);
5098 vcpu->run->exit_reason = 0;
5099 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
5100 (int)(exit_qualification >> 4) & 3, cr);
5104 static int handle_dr(struct kvm_vcpu *vcpu)
5106 unsigned long exit_qualification;
5109 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5110 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
5112 /* First, if DR does not exist, trigger UD */
5113 if (!kvm_require_dr(vcpu, dr))
5116 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
5117 if (!kvm_require_cpl(vcpu, 0))
5119 dr7 = vmcs_readl(GUEST_DR7);
5122 * As the vm-exit takes precedence over the debug trap, we
5123 * need to emulate the latter, either for the host or the
5124 * guest debugging itself.
5126 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5127 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
5128 vcpu->run->debug.arch.dr7 = dr7;
5129 vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
5130 vcpu->run->debug.arch.exception = DB_VECTOR;
5131 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
5134 vcpu->arch.dr6 &= ~15;
5135 vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
5136 kvm_queue_exception(vcpu, DB_VECTOR);
5141 if (vcpu->guest_debug == 0) {
5142 u32 cpu_based_vm_exec_control;
5144 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5145 cpu_based_vm_exec_control &= ~CPU_BASED_MOV_DR_EXITING;
5146 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5149 * No more DR vmexits; force a reload of the debug registers
5150 * and reenter on this instruction. The next vmexit will
5151 * retrieve the full state of the debug registers.
5153 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
5157 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
5158 if (exit_qualification & TYPE_MOV_FROM_DR) {
5161 if (kvm_get_dr(vcpu, dr, &val))
5163 kvm_register_write(vcpu, reg, val);
5165 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
5168 skip_emulated_instruction(vcpu);
5172 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
5174 return vcpu->arch.dr6;
5177 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
5181 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
5183 u32 cpu_based_vm_exec_control;
5185 get_debugreg(vcpu->arch.db[0], 0);
5186 get_debugreg(vcpu->arch.db[1], 1);
5187 get_debugreg(vcpu->arch.db[2], 2);
5188 get_debugreg(vcpu->arch.db[3], 3);
5189 get_debugreg(vcpu->arch.dr6, 6);
5190 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
5192 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
5194 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5195 cpu_based_vm_exec_control |= CPU_BASED_MOV_DR_EXITING;
5196 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5199 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
5201 vmcs_writel(GUEST_DR7, val);
5204 static int handle_cpuid(struct kvm_vcpu *vcpu)
5206 kvm_emulate_cpuid(vcpu);
5210 static int handle_rdmsr(struct kvm_vcpu *vcpu)
5212 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5215 if (vmx_get_msr(vcpu, ecx, &data)) {
5216 trace_kvm_msr_read_ex(ecx);
5217 kvm_inject_gp(vcpu, 0);
5221 trace_kvm_msr_read(ecx, data);
5223 /* FIXME: handling of bits 32:63 of rax, rdx */
5224 vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
5225 vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
5226 skip_emulated_instruction(vcpu);
5230 static int handle_wrmsr(struct kvm_vcpu *vcpu)
5232 struct msr_data msr;
5233 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5234 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
5235 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
5239 msr.host_initiated = false;
5240 if (kvm_set_msr(vcpu, &msr) != 0) {
5241 trace_kvm_msr_write_ex(ecx, data);
5242 kvm_inject_gp(vcpu, 0);
5246 trace_kvm_msr_write(ecx, data);
5247 skip_emulated_instruction(vcpu);
5251 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
5253 kvm_make_request(KVM_REQ_EVENT, vcpu);
5257 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
5259 u32 cpu_based_vm_exec_control;
5261 /* clear pending irq */
5262 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5263 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
5264 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5266 kvm_make_request(KVM_REQ_EVENT, vcpu);
5268 ++vcpu->stat.irq_window_exits;
5271 * If the user space waits to inject interrupts, exit as soon as
5274 if (!irqchip_in_kernel(vcpu->kvm) &&
5275 vcpu->run->request_interrupt_window &&
5276 !kvm_cpu_has_interrupt(vcpu)) {
5277 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
5283 static int handle_halt(struct kvm_vcpu *vcpu)
5285 skip_emulated_instruction(vcpu);
5286 return kvm_emulate_halt(vcpu);
5289 static int handle_vmcall(struct kvm_vcpu *vcpu)
5291 skip_emulated_instruction(vcpu);
5292 kvm_emulate_hypercall(vcpu);
5296 static int handle_invd(struct kvm_vcpu *vcpu)
5298 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5301 static int handle_invlpg(struct kvm_vcpu *vcpu)
5303 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5305 kvm_mmu_invlpg(vcpu, exit_qualification);
5306 skip_emulated_instruction(vcpu);
5310 static int handle_rdpmc(struct kvm_vcpu *vcpu)
5314 err = kvm_rdpmc(vcpu);
5315 kvm_complete_insn_gp(vcpu, err);
5320 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5322 skip_emulated_instruction(vcpu);
5323 kvm_emulate_wbinvd(vcpu);
5327 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5329 u64 new_bv = kvm_read_edx_eax(vcpu);
5330 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5332 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5333 skip_emulated_instruction(vcpu);
5337 static int handle_apic_access(struct kvm_vcpu *vcpu)
5339 if (likely(fasteoi)) {
5340 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5341 int access_type, offset;
5343 access_type = exit_qualification & APIC_ACCESS_TYPE;
5344 offset = exit_qualification & APIC_ACCESS_OFFSET;
5346 * Sane guest uses MOV to write EOI, with written value
5347 * not cared. So make a short-circuit here by avoiding
5348 * heavy instruction emulation.
5350 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5351 (offset == APIC_EOI)) {
5352 kvm_lapic_set_eoi(vcpu);
5353 skip_emulated_instruction(vcpu);
5357 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5360 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5362 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5363 int vector = exit_qualification & 0xff;
5365 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5366 kvm_apic_set_eoi_accelerated(vcpu, vector);
5370 static int handle_apic_write(struct kvm_vcpu *vcpu)
5372 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5373 u32 offset = exit_qualification & 0xfff;
5375 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
5376 kvm_apic_write_nodecode(vcpu, offset);
5380 static int handle_task_switch(struct kvm_vcpu *vcpu)
5382 struct vcpu_vmx *vmx = to_vmx(vcpu);
5383 unsigned long exit_qualification;
5384 bool has_error_code = false;
5387 int reason, type, idt_v, idt_index;
5389 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5390 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5391 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5393 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5395 reason = (u32)exit_qualification >> 30;
5396 if (reason == TASK_SWITCH_GATE && idt_v) {
5398 case INTR_TYPE_NMI_INTR:
5399 vcpu->arch.nmi_injected = false;
5400 vmx_set_nmi_mask(vcpu, true);
5402 case INTR_TYPE_EXT_INTR:
5403 case INTR_TYPE_SOFT_INTR:
5404 kvm_clear_interrupt_queue(vcpu);
5406 case INTR_TYPE_HARD_EXCEPTION:
5407 if (vmx->idt_vectoring_info &
5408 VECTORING_INFO_DELIVER_CODE_MASK) {
5409 has_error_code = true;
5411 vmcs_read32(IDT_VECTORING_ERROR_CODE);
5414 case INTR_TYPE_SOFT_EXCEPTION:
5415 kvm_clear_exception_queue(vcpu);
5421 tss_selector = exit_qualification;
5423 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5424 type != INTR_TYPE_EXT_INTR &&
5425 type != INTR_TYPE_NMI_INTR))
5426 skip_emulated_instruction(vcpu);
5428 if (kvm_task_switch(vcpu, tss_selector,
5429 type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
5430 has_error_code, error_code) == EMULATE_FAIL) {
5431 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5432 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5433 vcpu->run->internal.ndata = 0;
5437 /* clear all local breakpoint enable flags */
5438 vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~0x155);
5441 * TODO: What about debug traps on tss switch?
5442 * Are we supposed to inject them and update dr6?
5448 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5450 unsigned long exit_qualification;
5455 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5457 gla_validity = (exit_qualification >> 7) & 0x3;
5458 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
5459 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
5460 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
5461 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
5462 vmcs_readl(GUEST_LINEAR_ADDRESS));
5463 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
5464 (long unsigned int)exit_qualification);
5465 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5466 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
5471 * EPT violation happened while executing iret from NMI,
5472 * "blocked by NMI" bit has to be set before next VM entry.
5473 * There are errata that may cause this bit to not be set:
5476 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5477 cpu_has_virtual_nmis() &&
5478 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
5479 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5481 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5482 trace_kvm_page_fault(gpa, exit_qualification);
5484 /* It is a write fault? */
5485 error_code = exit_qualification & PFERR_WRITE_MASK;
5486 /* It is a fetch fault? */
5487 error_code |= (exit_qualification << 2) & PFERR_FETCH_MASK;
5488 /* ept page table is present? */
5489 error_code |= (exit_qualification >> 3) & PFERR_PRESENT_MASK;
5491 vcpu->arch.exit_qualification = exit_qualification;
5493 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5496 static u64 ept_rsvd_mask(u64 spte, int level)
5501 for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
5502 mask |= (1ULL << i);
5505 /* bits 7:3 reserved */
5507 else if (spte & (1ULL << 7))
5509 * 1GB/2MB page, bits 29:12 or 20:12 reserved respectively,
5510 * level == 1 if the hypervisor is using the ignored bit 7.
5512 mask |= (PAGE_SIZE << ((level - 1) * 9)) - PAGE_SIZE;
5514 /* bits 6:3 reserved */
5520 static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
5523 printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
5525 /* 010b (write-only) */
5526 WARN_ON((spte & 0x7) == 0x2);
5528 /* 110b (write/execute) */
5529 WARN_ON((spte & 0x7) == 0x6);
5531 /* 100b (execute-only) and value not supported by logical processor */
5532 if (!cpu_has_vmx_ept_execute_only())
5533 WARN_ON((spte & 0x7) == 0x4);
5537 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
5539 if (rsvd_bits != 0) {
5540 printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
5541 __func__, rsvd_bits);
5545 /* bits 5:3 are _not_ reserved for large page or leaf page */
5546 if ((rsvd_bits & 0x38) == 0) {
5547 u64 ept_mem_type = (spte & 0x38) >> 3;
5549 if (ept_mem_type == 2 || ept_mem_type == 3 ||
5550 ept_mem_type == 7) {
5551 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
5552 __func__, ept_mem_type);
5559 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5562 int nr_sptes, i, ret;
5565 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5566 if (!kvm_io_bus_write(vcpu->kvm, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5567 skip_emulated_instruction(vcpu);
5571 ret = handle_mmio_page_fault_common(vcpu, gpa, true);
5572 if (likely(ret == RET_MMIO_PF_EMULATE))
5573 return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) ==
5576 if (unlikely(ret == RET_MMIO_PF_INVALID))
5577 return kvm_mmu_page_fault(vcpu, gpa, 0, NULL, 0);
5579 if (unlikely(ret == RET_MMIO_PF_RETRY))
5582 /* It is the real ept misconfig */
5583 printk(KERN_ERR "EPT: Misconfiguration.\n");
5584 printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
5586 nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
5588 for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
5589 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
5591 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5592 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
5597 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5599 u32 cpu_based_vm_exec_control;
5601 /* clear pending NMI */
5602 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5603 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
5604 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5605 ++vcpu->stat.nmi_window_exits;
5606 kvm_make_request(KVM_REQ_EVENT, vcpu);
5611 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5613 struct vcpu_vmx *vmx = to_vmx(vcpu);
5614 enum emulation_result err = EMULATE_DONE;
5617 bool intr_window_requested;
5618 unsigned count = 130;
5620 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5621 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
5623 while (vmx->emulation_required && count-- != 0) {
5624 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5625 return handle_interrupt_window(&vmx->vcpu);
5627 if (test_bit(KVM_REQ_EVENT, &vcpu->requests))
5630 err = emulate_instruction(vcpu, EMULTYPE_NO_REEXECUTE);
5632 if (err == EMULATE_USER_EXIT) {
5633 ++vcpu->stat.mmio_exits;
5638 if (err != EMULATE_DONE) {
5639 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5640 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5641 vcpu->run->internal.ndata = 0;
5645 if (vcpu->arch.halt_request) {
5646 vcpu->arch.halt_request = 0;
5647 ret = kvm_emulate_halt(vcpu);
5651 if (signal_pending(current))
5661 static int __grow_ple_window(int val)
5663 if (ple_window_grow < 1)
5666 val = min(val, ple_window_actual_max);
5668 if (ple_window_grow < ple_window)
5669 val *= ple_window_grow;
5671 val += ple_window_grow;
5676 static int __shrink_ple_window(int val, int modifier, int minimum)
5681 if (modifier < ple_window)
5686 return max(val, minimum);
5689 static void grow_ple_window(struct kvm_vcpu *vcpu)
5691 struct vcpu_vmx *vmx = to_vmx(vcpu);
5692 int old = vmx->ple_window;
5694 vmx->ple_window = __grow_ple_window(old);
5696 if (vmx->ple_window != old)
5697 vmx->ple_window_dirty = true;
5699 trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old);
5702 static void shrink_ple_window(struct kvm_vcpu *vcpu)
5704 struct vcpu_vmx *vmx = to_vmx(vcpu);
5705 int old = vmx->ple_window;
5707 vmx->ple_window = __shrink_ple_window(old,
5708 ple_window_shrink, ple_window);
5710 if (vmx->ple_window != old)
5711 vmx->ple_window_dirty = true;
5713 trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old);
5717 * ple_window_actual_max is computed to be one grow_ple_window() below
5718 * ple_window_max. (See __grow_ple_window for the reason.)
5719 * This prevents overflows, because ple_window_max is int.
5720 * ple_window_max effectively rounded down to a multiple of ple_window_grow in
5722 * ple_window_max is also prevented from setting vmx->ple_window < ple_window.
5724 static void update_ple_window_actual_max(void)
5726 ple_window_actual_max =
5727 __shrink_ple_window(max(ple_window_max, ple_window),
5728 ple_window_grow, INT_MIN);
5731 static __init int hardware_setup(void)
5733 int r = -ENOMEM, i, msr;
5735 rdmsrl_safe(MSR_EFER, &host_efer);
5737 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
5738 kvm_define_shared_msr(i, vmx_msr_index[i]);
5740 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
5741 if (!vmx_io_bitmap_a)
5744 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
5745 if (!vmx_io_bitmap_b)
5748 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
5749 if (!vmx_msr_bitmap_legacy)
5752 vmx_msr_bitmap_legacy_x2apic =
5753 (unsigned long *)__get_free_page(GFP_KERNEL);
5754 if (!vmx_msr_bitmap_legacy_x2apic)
5757 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
5758 if (!vmx_msr_bitmap_longmode)
5761 vmx_msr_bitmap_longmode_x2apic =
5762 (unsigned long *)__get_free_page(GFP_KERNEL);
5763 if (!vmx_msr_bitmap_longmode_x2apic)
5765 vmx_vmread_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
5766 if (!vmx_vmread_bitmap)
5769 vmx_vmwrite_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
5770 if (!vmx_vmwrite_bitmap)
5773 memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
5774 memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
5777 * Allow direct access to the PC debug port (it is often used for I/O
5778 * delays, but the vmexits simply slow things down).
5780 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
5781 clear_bit(0x80, vmx_io_bitmap_a);
5783 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
5785 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
5786 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
5788 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
5789 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
5790 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
5791 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
5792 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
5793 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
5794 vmx_disable_intercept_for_msr(MSR_IA32_BNDCFGS, true);
5796 memcpy(vmx_msr_bitmap_legacy_x2apic,
5797 vmx_msr_bitmap_legacy, PAGE_SIZE);
5798 memcpy(vmx_msr_bitmap_longmode_x2apic,
5799 vmx_msr_bitmap_longmode, PAGE_SIZE);
5802 for (msr = 0x800; msr <= 0x8ff; msr++)
5803 vmx_disable_intercept_msr_read_x2apic(msr);
5805 /* According SDM, in x2apic mode, the whole id reg is used.
5806 * But in KVM, it only use the highest eight bits. Need to
5808 vmx_enable_intercept_msr_read_x2apic(0x802);
5810 vmx_enable_intercept_msr_read_x2apic(0x839);
5812 vmx_disable_intercept_msr_write_x2apic(0x808);
5814 vmx_disable_intercept_msr_write_x2apic(0x80b);
5816 vmx_disable_intercept_msr_write_x2apic(0x83f);
5820 kvm_mmu_set_mask_ptes(0ull,
5821 (enable_ept_ad_bits) ? VMX_EPT_ACCESS_BIT : 0ull,
5822 (enable_ept_ad_bits) ? VMX_EPT_DIRTY_BIT : 0ull,
5823 0ull, VMX_EPT_EXECUTABLE_MASK);
5824 ept_set_mmio_spte_mask();
5829 update_ple_window_actual_max();
5831 if (setup_vmcs_config(&vmcs_config) < 0) {
5836 if (boot_cpu_has(X86_FEATURE_NX))
5837 kvm_enable_efer_bits(EFER_NX);
5839 if (!cpu_has_vmx_vpid())
5841 if (!cpu_has_vmx_shadow_vmcs())
5842 enable_shadow_vmcs = 0;
5843 if (enable_shadow_vmcs)
5844 init_vmcs_shadow_fields();
5846 if (!cpu_has_vmx_ept() ||
5847 !cpu_has_vmx_ept_4levels()) {
5849 enable_unrestricted_guest = 0;
5850 enable_ept_ad_bits = 0;
5853 if (!cpu_has_vmx_ept_ad_bits())
5854 enable_ept_ad_bits = 0;
5856 if (!cpu_has_vmx_unrestricted_guest())
5857 enable_unrestricted_guest = 0;
5859 if (!cpu_has_vmx_flexpriority()) {
5860 flexpriority_enabled = 0;
5863 * set_apic_access_page_addr() is used to reload apic access
5864 * page upon invalidation. No need to do anything if the
5865 * processor does not have the APIC_ACCESS_ADDR VMCS field.
5867 kvm_x86_ops->set_apic_access_page_addr = NULL;
5870 if (!cpu_has_vmx_tpr_shadow())
5871 kvm_x86_ops->update_cr8_intercept = NULL;
5873 if (enable_ept && !cpu_has_vmx_ept_2m_page())
5874 kvm_disable_largepages();
5876 if (!cpu_has_vmx_ple())
5879 if (!cpu_has_vmx_apicv())
5883 kvm_x86_ops->update_cr8_intercept = NULL;
5885 kvm_x86_ops->hwapic_irr_update = NULL;
5886 kvm_x86_ops->deliver_posted_interrupt = NULL;
5887 kvm_x86_ops->sync_pir_to_irr = vmx_sync_pir_to_irr_dummy;
5891 nested_vmx_setup_ctls_msrs();
5893 return alloc_kvm_area();
5896 free_page((unsigned long)vmx_vmwrite_bitmap);
5898 free_page((unsigned long)vmx_vmread_bitmap);
5900 free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
5902 free_page((unsigned long)vmx_msr_bitmap_longmode);
5904 free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
5906 free_page((unsigned long)vmx_msr_bitmap_legacy);
5908 free_page((unsigned long)vmx_io_bitmap_b);
5910 free_page((unsigned long)vmx_io_bitmap_a);
5915 static __exit void hardware_unsetup(void)
5917 free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
5918 free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
5919 free_page((unsigned long)vmx_msr_bitmap_legacy);
5920 free_page((unsigned long)vmx_msr_bitmap_longmode);
5921 free_page((unsigned long)vmx_io_bitmap_b);
5922 free_page((unsigned long)vmx_io_bitmap_a);
5923 free_page((unsigned long)vmx_vmwrite_bitmap);
5924 free_page((unsigned long)vmx_vmread_bitmap);
5930 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5931 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5933 static int handle_pause(struct kvm_vcpu *vcpu)
5936 grow_ple_window(vcpu);
5938 skip_emulated_instruction(vcpu);
5939 kvm_vcpu_on_spin(vcpu);
5944 static int handle_nop(struct kvm_vcpu *vcpu)
5946 skip_emulated_instruction(vcpu);
5950 static int handle_mwait(struct kvm_vcpu *vcpu)
5952 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
5953 return handle_nop(vcpu);
5956 static int handle_monitor(struct kvm_vcpu *vcpu)
5958 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
5959 return handle_nop(vcpu);
5963 * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
5964 * We could reuse a single VMCS for all the L2 guests, but we also want the
5965 * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
5966 * allows keeping them loaded on the processor, and in the future will allow
5967 * optimizations where prepare_vmcs02 doesn't need to set all the fields on
5968 * every entry if they never change.
5969 * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
5970 * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
5972 * The following functions allocate and free a vmcs02 in this pool.
5975 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
5976 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
5978 struct vmcs02_list *item;
5979 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
5980 if (item->vmptr == vmx->nested.current_vmptr) {
5981 list_move(&item->list, &vmx->nested.vmcs02_pool);
5982 return &item->vmcs02;
5985 if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
5986 /* Recycle the least recently used VMCS. */
5987 item = list_entry(vmx->nested.vmcs02_pool.prev,
5988 struct vmcs02_list, list);
5989 item->vmptr = vmx->nested.current_vmptr;
5990 list_move(&item->list, &vmx->nested.vmcs02_pool);
5991 return &item->vmcs02;
5994 /* Create a new VMCS */
5995 item = kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
5998 item->vmcs02.vmcs = alloc_vmcs();
5999 if (!item->vmcs02.vmcs) {
6003 loaded_vmcs_init(&item->vmcs02);
6004 item->vmptr = vmx->nested.current_vmptr;
6005 list_add(&(item->list), &(vmx->nested.vmcs02_pool));
6006 vmx->nested.vmcs02_num++;
6007 return &item->vmcs02;
6010 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
6011 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
6013 struct vmcs02_list *item;
6014 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6015 if (item->vmptr == vmptr) {
6016 free_loaded_vmcs(&item->vmcs02);
6017 list_del(&item->list);
6019 vmx->nested.vmcs02_num--;
6025 * Free all VMCSs saved for this vcpu, except the one pointed by
6026 * vmx->loaded_vmcs. We must be running L1, so vmx->loaded_vmcs
6027 * must be &vmx->vmcs01.
6029 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
6031 struct vmcs02_list *item, *n;
6033 WARN_ON(vmx->loaded_vmcs != &vmx->vmcs01);
6034 list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
6036 * Something will leak if the above WARN triggers. Better than
6039 if (vmx->loaded_vmcs == &item->vmcs02)
6042 free_loaded_vmcs(&item->vmcs02);
6043 list_del(&item->list);
6045 vmx->nested.vmcs02_num--;
6050 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
6051 * set the success or error code of an emulated VMX instruction, as specified
6052 * by Vol 2B, VMX Instruction Reference, "Conventions".
6054 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
6056 vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
6057 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6058 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
6061 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
6063 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6064 & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
6065 X86_EFLAGS_SF | X86_EFLAGS_OF))
6069 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
6070 u32 vm_instruction_error)
6072 if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
6074 * failValid writes the error number to the current VMCS, which
6075 * can't be done there isn't a current VMCS.
6077 nested_vmx_failInvalid(vcpu);
6080 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6081 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6082 X86_EFLAGS_SF | X86_EFLAGS_OF))
6084 get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
6086 * We don't need to force a shadow sync because
6087 * VM_INSTRUCTION_ERROR is not shadowed
6091 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
6093 struct vcpu_vmx *vmx =
6094 container_of(timer, struct vcpu_vmx, nested.preemption_timer);
6096 vmx->nested.preemption_timer_expired = true;
6097 kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
6098 kvm_vcpu_kick(&vmx->vcpu);
6100 return HRTIMER_NORESTART;
6104 * Decode the memory-address operand of a vmx instruction, as recorded on an
6105 * exit caused by such an instruction (run by a guest hypervisor).
6106 * On success, returns 0. When the operand is invalid, returns 1 and throws
6109 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
6110 unsigned long exit_qualification,
6111 u32 vmx_instruction_info, gva_t *ret)
6114 * According to Vol. 3B, "Information for VM Exits Due to Instruction
6115 * Execution", on an exit, vmx_instruction_info holds most of the
6116 * addressing components of the operand. Only the displacement part
6117 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
6118 * For how an actual address is calculated from all these components,
6119 * refer to Vol. 1, "Operand Addressing".
6121 int scaling = vmx_instruction_info & 3;
6122 int addr_size = (vmx_instruction_info >> 7) & 7;
6123 bool is_reg = vmx_instruction_info & (1u << 10);
6124 int seg_reg = (vmx_instruction_info >> 15) & 7;
6125 int index_reg = (vmx_instruction_info >> 18) & 0xf;
6126 bool index_is_valid = !(vmx_instruction_info & (1u << 22));
6127 int base_reg = (vmx_instruction_info >> 23) & 0xf;
6128 bool base_is_valid = !(vmx_instruction_info & (1u << 27));
6131 kvm_queue_exception(vcpu, UD_VECTOR);
6135 /* Addr = segment_base + offset */
6136 /* offset = base + [index * scale] + displacement */
6137 *ret = vmx_get_segment_base(vcpu, seg_reg);
6139 *ret += kvm_register_read(vcpu, base_reg);
6141 *ret += kvm_register_read(vcpu, index_reg)<<scaling;
6142 *ret += exit_qualification; /* holds the displacement */
6144 if (addr_size == 1) /* 32 bit */
6148 * TODO: throw #GP (and return 1) in various cases that the VM*
6149 * instructions require it - e.g., offset beyond segment limit,
6150 * unusable or unreadable/unwritable segment, non-canonical 64-bit
6151 * address, and so on. Currently these are not checked.
6157 * This function performs the various checks including
6158 * - if it's 4KB aligned
6159 * - No bits beyond the physical address width are set
6160 * - Returns 0 on success or else 1
6161 * (Intel SDM Section 30.3)
6163 static int nested_vmx_check_vmptr(struct kvm_vcpu *vcpu, int exit_reason,
6168 struct x86_exception e;
6170 struct vcpu_vmx *vmx = to_vmx(vcpu);
6171 int maxphyaddr = cpuid_maxphyaddr(vcpu);
6173 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
6174 vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
6177 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
6178 sizeof(vmptr), &e)) {
6179 kvm_inject_page_fault(vcpu, &e);
6183 switch (exit_reason) {
6184 case EXIT_REASON_VMON:
6187 * The first 4 bytes of VMXON region contain the supported
6188 * VMCS revision identifier
6190 * Note - IA32_VMX_BASIC[48] will never be 1
6191 * for the nested case;
6192 * which replaces physical address width with 32
6195 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6196 nested_vmx_failInvalid(vcpu);
6197 skip_emulated_instruction(vcpu);
6201 page = nested_get_page(vcpu, vmptr);
6203 *(u32 *)kmap(page) != VMCS12_REVISION) {
6204 nested_vmx_failInvalid(vcpu);
6206 skip_emulated_instruction(vcpu);
6210 vmx->nested.vmxon_ptr = vmptr;
6212 case EXIT_REASON_VMCLEAR:
6213 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6214 nested_vmx_failValid(vcpu,
6215 VMXERR_VMCLEAR_INVALID_ADDRESS);
6216 skip_emulated_instruction(vcpu);
6220 if (vmptr == vmx->nested.vmxon_ptr) {
6221 nested_vmx_failValid(vcpu,
6222 VMXERR_VMCLEAR_VMXON_POINTER);
6223 skip_emulated_instruction(vcpu);
6227 case EXIT_REASON_VMPTRLD:
6228 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6229 nested_vmx_failValid(vcpu,
6230 VMXERR_VMPTRLD_INVALID_ADDRESS);
6231 skip_emulated_instruction(vcpu);
6235 if (vmptr == vmx->nested.vmxon_ptr) {
6236 nested_vmx_failValid(vcpu,
6237 VMXERR_VMCLEAR_VMXON_POINTER);
6238 skip_emulated_instruction(vcpu);
6243 return 1; /* shouldn't happen */
6252 * Emulate the VMXON instruction.
6253 * Currently, we just remember that VMX is active, and do not save or even
6254 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
6255 * do not currently need to store anything in that guest-allocated memory
6256 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
6257 * argument is different from the VMXON pointer (which the spec says they do).
6259 static int handle_vmon(struct kvm_vcpu *vcpu)
6261 struct kvm_segment cs;
6262 struct vcpu_vmx *vmx = to_vmx(vcpu);
6263 struct vmcs *shadow_vmcs;
6264 const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
6265 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
6267 /* The Intel VMX Instruction Reference lists a bunch of bits that
6268 * are prerequisite to running VMXON, most notably cr4.VMXE must be
6269 * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
6270 * Otherwise, we should fail with #UD. We test these now:
6272 if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
6273 !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
6274 (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
6275 kvm_queue_exception(vcpu, UD_VECTOR);
6279 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
6280 if (is_long_mode(vcpu) && !cs.l) {
6281 kvm_queue_exception(vcpu, UD_VECTOR);
6285 if (vmx_get_cpl(vcpu)) {
6286 kvm_inject_gp(vcpu, 0);
6290 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMON, NULL))
6293 if (vmx->nested.vmxon) {
6294 nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
6295 skip_emulated_instruction(vcpu);
6299 if ((vmx->nested.msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
6300 != VMXON_NEEDED_FEATURES) {
6301 kvm_inject_gp(vcpu, 0);
6305 if (enable_shadow_vmcs) {
6306 shadow_vmcs = alloc_vmcs();
6309 /* mark vmcs as shadow */
6310 shadow_vmcs->revision_id |= (1u << 31);
6311 /* init shadow vmcs */
6312 vmcs_clear(shadow_vmcs);
6313 vmx->nested.current_shadow_vmcs = shadow_vmcs;
6316 INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
6317 vmx->nested.vmcs02_num = 0;
6319 hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
6321 vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
6323 vmx->nested.vmxon = true;
6325 skip_emulated_instruction(vcpu);
6326 nested_vmx_succeed(vcpu);
6331 * Intel's VMX Instruction Reference specifies a common set of prerequisites
6332 * for running VMX instructions (except VMXON, whose prerequisites are
6333 * slightly different). It also specifies what exception to inject otherwise.
6335 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
6337 struct kvm_segment cs;
6338 struct vcpu_vmx *vmx = to_vmx(vcpu);
6340 if (!vmx->nested.vmxon) {
6341 kvm_queue_exception(vcpu, UD_VECTOR);
6345 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
6346 if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
6347 (is_long_mode(vcpu) && !cs.l)) {
6348 kvm_queue_exception(vcpu, UD_VECTOR);
6352 if (vmx_get_cpl(vcpu)) {
6353 kvm_inject_gp(vcpu, 0);
6360 static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
6363 if (vmx->nested.current_vmptr == -1ull)
6366 /* current_vmptr and current_vmcs12 are always set/reset together */
6367 if (WARN_ON(vmx->nested.current_vmcs12 == NULL))
6370 if (enable_shadow_vmcs) {
6371 /* copy to memory all shadowed fields in case
6372 they were modified */
6373 copy_shadow_to_vmcs12(vmx);
6374 vmx->nested.sync_shadow_vmcs = false;
6375 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6376 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
6377 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
6378 vmcs_write64(VMCS_LINK_POINTER, -1ull);
6380 kunmap(vmx->nested.current_vmcs12_page);
6381 nested_release_page(vmx->nested.current_vmcs12_page);
6382 vmx->nested.current_vmptr = -1ull;
6383 vmx->nested.current_vmcs12 = NULL;
6387 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
6388 * just stops using VMX.
6390 static void free_nested(struct vcpu_vmx *vmx)
6392 if (!vmx->nested.vmxon)
6395 vmx->nested.vmxon = false;
6396 nested_release_vmcs12(vmx);
6397 if (enable_shadow_vmcs)
6398 free_vmcs(vmx->nested.current_shadow_vmcs);
6399 /* Unpin physical memory we referred to in current vmcs02 */
6400 if (vmx->nested.apic_access_page) {
6401 nested_release_page(vmx->nested.apic_access_page);
6402 vmx->nested.apic_access_page = NULL;
6404 if (vmx->nested.virtual_apic_page) {
6405 nested_release_page(vmx->nested.virtual_apic_page);
6406 vmx->nested.virtual_apic_page = NULL;
6409 nested_free_all_saved_vmcss(vmx);
6412 /* Emulate the VMXOFF instruction */
6413 static int handle_vmoff(struct kvm_vcpu *vcpu)
6415 if (!nested_vmx_check_permission(vcpu))
6417 free_nested(to_vmx(vcpu));
6418 skip_emulated_instruction(vcpu);
6419 nested_vmx_succeed(vcpu);
6423 /* Emulate the VMCLEAR instruction */
6424 static int handle_vmclear(struct kvm_vcpu *vcpu)
6426 struct vcpu_vmx *vmx = to_vmx(vcpu);
6428 struct vmcs12 *vmcs12;
6431 if (!nested_vmx_check_permission(vcpu))
6434 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMCLEAR, &vmptr))
6437 if (vmptr == vmx->nested.current_vmptr)
6438 nested_release_vmcs12(vmx);
6440 page = nested_get_page(vcpu, vmptr);
6443 * For accurate processor emulation, VMCLEAR beyond available
6444 * physical memory should do nothing at all. However, it is
6445 * possible that a nested vmx bug, not a guest hypervisor bug,
6446 * resulted in this case, so let's shut down before doing any
6449 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6452 vmcs12 = kmap(page);
6453 vmcs12->launch_state = 0;
6455 nested_release_page(page);
6457 nested_free_vmcs02(vmx, vmptr);
6459 skip_emulated_instruction(vcpu);
6460 nested_vmx_succeed(vcpu);
6464 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
6466 /* Emulate the VMLAUNCH instruction */
6467 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
6469 return nested_vmx_run(vcpu, true);
6472 /* Emulate the VMRESUME instruction */
6473 static int handle_vmresume(struct kvm_vcpu *vcpu)
6476 return nested_vmx_run(vcpu, false);
6479 enum vmcs_field_type {
6480 VMCS_FIELD_TYPE_U16 = 0,
6481 VMCS_FIELD_TYPE_U64 = 1,
6482 VMCS_FIELD_TYPE_U32 = 2,
6483 VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
6486 static inline int vmcs_field_type(unsigned long field)
6488 if (0x1 & field) /* the *_HIGH fields are all 32 bit */
6489 return VMCS_FIELD_TYPE_U32;
6490 return (field >> 13) & 0x3 ;
6493 static inline int vmcs_field_readonly(unsigned long field)
6495 return (((field >> 10) & 0x3) == 1);
6499 * Read a vmcs12 field. Since these can have varying lengths and we return
6500 * one type, we chose the biggest type (u64) and zero-extend the return value
6501 * to that size. Note that the caller, handle_vmread, might need to use only
6502 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
6503 * 64-bit fields are to be returned).
6505 static inline int vmcs12_read_any(struct kvm_vcpu *vcpu,
6506 unsigned long field, u64 *ret)
6508 short offset = vmcs_field_to_offset(field);
6514 p = ((char *)(get_vmcs12(vcpu))) + offset;
6516 switch (vmcs_field_type(field)) {
6517 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6518 *ret = *((natural_width *)p);
6520 case VMCS_FIELD_TYPE_U16:
6523 case VMCS_FIELD_TYPE_U32:
6526 case VMCS_FIELD_TYPE_U64:
6536 static inline int vmcs12_write_any(struct kvm_vcpu *vcpu,
6537 unsigned long field, u64 field_value){
6538 short offset = vmcs_field_to_offset(field);
6539 char *p = ((char *) get_vmcs12(vcpu)) + offset;
6543 switch (vmcs_field_type(field)) {
6544 case VMCS_FIELD_TYPE_U16:
6545 *(u16 *)p = field_value;
6547 case VMCS_FIELD_TYPE_U32:
6548 *(u32 *)p = field_value;
6550 case VMCS_FIELD_TYPE_U64:
6551 *(u64 *)p = field_value;
6553 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6554 *(natural_width *)p = field_value;
6563 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
6566 unsigned long field;
6568 struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
6569 const unsigned long *fields = shadow_read_write_fields;
6570 const int num_fields = max_shadow_read_write_fields;
6574 vmcs_load(shadow_vmcs);
6576 for (i = 0; i < num_fields; i++) {
6578 switch (vmcs_field_type(field)) {
6579 case VMCS_FIELD_TYPE_U16:
6580 field_value = vmcs_read16(field);
6582 case VMCS_FIELD_TYPE_U32:
6583 field_value = vmcs_read32(field);
6585 case VMCS_FIELD_TYPE_U64:
6586 field_value = vmcs_read64(field);
6588 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6589 field_value = vmcs_readl(field);
6595 vmcs12_write_any(&vmx->vcpu, field, field_value);
6598 vmcs_clear(shadow_vmcs);
6599 vmcs_load(vmx->loaded_vmcs->vmcs);
6604 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
6606 const unsigned long *fields[] = {
6607 shadow_read_write_fields,
6608 shadow_read_only_fields
6610 const int max_fields[] = {
6611 max_shadow_read_write_fields,
6612 max_shadow_read_only_fields
6615 unsigned long field;
6616 u64 field_value = 0;
6617 struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
6619 vmcs_load(shadow_vmcs);
6621 for (q = 0; q < ARRAY_SIZE(fields); q++) {
6622 for (i = 0; i < max_fields[q]; i++) {
6623 field = fields[q][i];
6624 vmcs12_read_any(&vmx->vcpu, field, &field_value);
6626 switch (vmcs_field_type(field)) {
6627 case VMCS_FIELD_TYPE_U16:
6628 vmcs_write16(field, (u16)field_value);
6630 case VMCS_FIELD_TYPE_U32:
6631 vmcs_write32(field, (u32)field_value);
6633 case VMCS_FIELD_TYPE_U64:
6634 vmcs_write64(field, (u64)field_value);
6636 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6637 vmcs_writel(field, (long)field_value);
6646 vmcs_clear(shadow_vmcs);
6647 vmcs_load(vmx->loaded_vmcs->vmcs);
6651 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
6652 * used before) all generate the same failure when it is missing.
6654 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
6656 struct vcpu_vmx *vmx = to_vmx(vcpu);
6657 if (vmx->nested.current_vmptr == -1ull) {
6658 nested_vmx_failInvalid(vcpu);
6659 skip_emulated_instruction(vcpu);
6665 static int handle_vmread(struct kvm_vcpu *vcpu)
6667 unsigned long field;
6669 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6670 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6673 if (!nested_vmx_check_permission(vcpu) ||
6674 !nested_vmx_check_vmcs12(vcpu))
6677 /* Decode instruction info and find the field to read */
6678 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
6679 /* Read the field, zero-extended to a u64 field_value */
6680 if (vmcs12_read_any(vcpu, field, &field_value) < 0) {
6681 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
6682 skip_emulated_instruction(vcpu);
6686 * Now copy part of this value to register or memory, as requested.
6687 * Note that the number of bits actually copied is 32 or 64 depending
6688 * on the guest's mode (32 or 64 bit), not on the given field's length.
6690 if (vmx_instruction_info & (1u << 10)) {
6691 kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
6694 if (get_vmx_mem_address(vcpu, exit_qualification,
6695 vmx_instruction_info, &gva))
6697 /* _system ok, as nested_vmx_check_permission verified cpl=0 */
6698 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
6699 &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
6702 nested_vmx_succeed(vcpu);
6703 skip_emulated_instruction(vcpu);
6708 static int handle_vmwrite(struct kvm_vcpu *vcpu)
6710 unsigned long field;
6712 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6713 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6714 /* The value to write might be 32 or 64 bits, depending on L1's long
6715 * mode, and eventually we need to write that into a field of several
6716 * possible lengths. The code below first zero-extends the value to 64
6717 * bit (field_value), and then copies only the approriate number of
6718 * bits into the vmcs12 field.
6720 u64 field_value = 0;
6721 struct x86_exception e;
6723 if (!nested_vmx_check_permission(vcpu) ||
6724 !nested_vmx_check_vmcs12(vcpu))
6727 if (vmx_instruction_info & (1u << 10))
6728 field_value = kvm_register_readl(vcpu,
6729 (((vmx_instruction_info) >> 3) & 0xf));
6731 if (get_vmx_mem_address(vcpu, exit_qualification,
6732 vmx_instruction_info, &gva))
6734 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
6735 &field_value, (is_64_bit_mode(vcpu) ? 8 : 4), &e)) {
6736 kvm_inject_page_fault(vcpu, &e);
6742 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
6743 if (vmcs_field_readonly(field)) {
6744 nested_vmx_failValid(vcpu,
6745 VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
6746 skip_emulated_instruction(vcpu);
6750 if (vmcs12_write_any(vcpu, field, field_value) < 0) {
6751 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
6752 skip_emulated_instruction(vcpu);
6756 nested_vmx_succeed(vcpu);
6757 skip_emulated_instruction(vcpu);
6761 /* Emulate the VMPTRLD instruction */
6762 static int handle_vmptrld(struct kvm_vcpu *vcpu)
6764 struct vcpu_vmx *vmx = to_vmx(vcpu);
6768 if (!nested_vmx_check_permission(vcpu))
6771 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMPTRLD, &vmptr))
6774 if (vmx->nested.current_vmptr != vmptr) {
6775 struct vmcs12 *new_vmcs12;
6777 page = nested_get_page(vcpu, vmptr);
6779 nested_vmx_failInvalid(vcpu);
6780 skip_emulated_instruction(vcpu);
6783 new_vmcs12 = kmap(page);
6784 if (new_vmcs12->revision_id != VMCS12_REVISION) {
6786 nested_release_page_clean(page);
6787 nested_vmx_failValid(vcpu,
6788 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
6789 skip_emulated_instruction(vcpu);
6793 nested_release_vmcs12(vmx);
6794 vmx->nested.current_vmptr = vmptr;
6795 vmx->nested.current_vmcs12 = new_vmcs12;
6796 vmx->nested.current_vmcs12_page = page;
6797 if (enable_shadow_vmcs) {
6798 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6799 exec_control |= SECONDARY_EXEC_SHADOW_VMCS;
6800 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
6801 vmcs_write64(VMCS_LINK_POINTER,
6802 __pa(vmx->nested.current_shadow_vmcs));
6803 vmx->nested.sync_shadow_vmcs = true;
6807 nested_vmx_succeed(vcpu);
6808 skip_emulated_instruction(vcpu);
6812 /* Emulate the VMPTRST instruction */
6813 static int handle_vmptrst(struct kvm_vcpu *vcpu)
6815 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6816 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6818 struct x86_exception e;
6820 if (!nested_vmx_check_permission(vcpu))
6823 if (get_vmx_mem_address(vcpu, exit_qualification,
6824 vmx_instruction_info, &vmcs_gva))
6826 /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
6827 if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
6828 (void *)&to_vmx(vcpu)->nested.current_vmptr,
6830 kvm_inject_page_fault(vcpu, &e);
6833 nested_vmx_succeed(vcpu);
6834 skip_emulated_instruction(vcpu);
6838 /* Emulate the INVEPT instruction */
6839 static int handle_invept(struct kvm_vcpu *vcpu)
6841 u32 vmx_instruction_info, types;
6844 struct x86_exception e;
6849 if (!(nested_vmx_secondary_ctls_high & SECONDARY_EXEC_ENABLE_EPT) ||
6850 !(nested_vmx_ept_caps & VMX_EPT_INVEPT_BIT)) {
6851 kvm_queue_exception(vcpu, UD_VECTOR);
6855 if (!nested_vmx_check_permission(vcpu))
6858 if (!kvm_read_cr0_bits(vcpu, X86_CR0_PE)) {
6859 kvm_queue_exception(vcpu, UD_VECTOR);
6863 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6864 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
6866 types = (nested_vmx_ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
6868 if (!(types & (1UL << type))) {
6869 nested_vmx_failValid(vcpu,
6870 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
6874 /* According to the Intel VMX instruction reference, the memory
6875 * operand is read even if it isn't needed (e.g., for type==global)
6877 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
6878 vmx_instruction_info, &gva))
6880 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
6881 sizeof(operand), &e)) {
6882 kvm_inject_page_fault(vcpu, &e);
6887 case VMX_EPT_EXTENT_GLOBAL:
6888 kvm_mmu_sync_roots(vcpu);
6889 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
6890 nested_vmx_succeed(vcpu);
6893 /* Trap single context invalidation invept calls */
6898 skip_emulated_instruction(vcpu);
6902 static int handle_invvpid(struct kvm_vcpu *vcpu)
6904 kvm_queue_exception(vcpu, UD_VECTOR);
6909 * The exit handlers return 1 if the exit was handled fully and guest execution
6910 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
6911 * to be done to userspace and return 0.
6913 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
6914 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
6915 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
6916 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
6917 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
6918 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
6919 [EXIT_REASON_CR_ACCESS] = handle_cr,
6920 [EXIT_REASON_DR_ACCESS] = handle_dr,
6921 [EXIT_REASON_CPUID] = handle_cpuid,
6922 [EXIT_REASON_MSR_READ] = handle_rdmsr,
6923 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
6924 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
6925 [EXIT_REASON_HLT] = handle_halt,
6926 [EXIT_REASON_INVD] = handle_invd,
6927 [EXIT_REASON_INVLPG] = handle_invlpg,
6928 [EXIT_REASON_RDPMC] = handle_rdpmc,
6929 [EXIT_REASON_VMCALL] = handle_vmcall,
6930 [EXIT_REASON_VMCLEAR] = handle_vmclear,
6931 [EXIT_REASON_VMLAUNCH] = handle_vmlaunch,
6932 [EXIT_REASON_VMPTRLD] = handle_vmptrld,
6933 [EXIT_REASON_VMPTRST] = handle_vmptrst,
6934 [EXIT_REASON_VMREAD] = handle_vmread,
6935 [EXIT_REASON_VMRESUME] = handle_vmresume,
6936 [EXIT_REASON_VMWRITE] = handle_vmwrite,
6937 [EXIT_REASON_VMOFF] = handle_vmoff,
6938 [EXIT_REASON_VMON] = handle_vmon,
6939 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
6940 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
6941 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
6942 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
6943 [EXIT_REASON_WBINVD] = handle_wbinvd,
6944 [EXIT_REASON_XSETBV] = handle_xsetbv,
6945 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
6946 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
6947 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
6948 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
6949 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
6950 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait,
6951 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor,
6952 [EXIT_REASON_INVEPT] = handle_invept,
6953 [EXIT_REASON_INVVPID] = handle_invvpid,
6956 static const int kvm_vmx_max_exit_handlers =
6957 ARRAY_SIZE(kvm_vmx_exit_handlers);
6959 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
6960 struct vmcs12 *vmcs12)
6962 unsigned long exit_qualification;
6963 gpa_t bitmap, last_bitmap;
6968 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
6969 return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
6971 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6973 port = exit_qualification >> 16;
6974 size = (exit_qualification & 7) + 1;
6976 last_bitmap = (gpa_t)-1;
6981 bitmap = vmcs12->io_bitmap_a;
6982 else if (port < 0x10000)
6983 bitmap = vmcs12->io_bitmap_b;
6986 bitmap += (port & 0x7fff) / 8;
6988 if (last_bitmap != bitmap)
6989 if (kvm_read_guest(vcpu->kvm, bitmap, &b, 1))
6991 if (b & (1 << (port & 7)))
6996 last_bitmap = bitmap;
7003 * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
7004 * rather than handle it ourselves in L0. I.e., check whether L1 expressed
7005 * disinterest in the current event (read or write a specific MSR) by using an
7006 * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
7008 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
7009 struct vmcs12 *vmcs12, u32 exit_reason)
7011 u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
7014 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
7018 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
7019 * for the four combinations of read/write and low/high MSR numbers.
7020 * First we need to figure out which of the four to use:
7022 bitmap = vmcs12->msr_bitmap;
7023 if (exit_reason == EXIT_REASON_MSR_WRITE)
7025 if (msr_index >= 0xc0000000) {
7026 msr_index -= 0xc0000000;
7030 /* Then read the msr_index'th bit from this bitmap: */
7031 if (msr_index < 1024*8) {
7033 if (kvm_read_guest(vcpu->kvm, bitmap + msr_index/8, &b, 1))
7035 return 1 & (b >> (msr_index & 7));
7037 return 1; /* let L1 handle the wrong parameter */
7041 * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
7042 * rather than handle it ourselves in L0. I.e., check if L1 wanted to
7043 * intercept (via guest_host_mask etc.) the current event.
7045 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
7046 struct vmcs12 *vmcs12)
7048 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7049 int cr = exit_qualification & 15;
7050 int reg = (exit_qualification >> 8) & 15;
7051 unsigned long val = kvm_register_readl(vcpu, reg);
7053 switch ((exit_qualification >> 4) & 3) {
7054 case 0: /* mov to cr */
7057 if (vmcs12->cr0_guest_host_mask &
7058 (val ^ vmcs12->cr0_read_shadow))
7062 if ((vmcs12->cr3_target_count >= 1 &&
7063 vmcs12->cr3_target_value0 == val) ||
7064 (vmcs12->cr3_target_count >= 2 &&
7065 vmcs12->cr3_target_value1 == val) ||
7066 (vmcs12->cr3_target_count >= 3 &&
7067 vmcs12->cr3_target_value2 == val) ||
7068 (vmcs12->cr3_target_count >= 4 &&
7069 vmcs12->cr3_target_value3 == val))
7071 if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
7075 if (vmcs12->cr4_guest_host_mask &
7076 (vmcs12->cr4_read_shadow ^ val))
7080 if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
7086 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
7087 (vmcs12->cr0_read_shadow & X86_CR0_TS))
7090 case 1: /* mov from cr */
7093 if (vmcs12->cpu_based_vm_exec_control &
7094 CPU_BASED_CR3_STORE_EXITING)
7098 if (vmcs12->cpu_based_vm_exec_control &
7099 CPU_BASED_CR8_STORE_EXITING)
7106 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
7107 * cr0. Other attempted changes are ignored, with no exit.
7109 if (vmcs12->cr0_guest_host_mask & 0xe &
7110 (val ^ vmcs12->cr0_read_shadow))
7112 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
7113 !(vmcs12->cr0_read_shadow & 0x1) &&
7122 * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
7123 * should handle it ourselves in L0 (and then continue L2). Only call this
7124 * when in is_guest_mode (L2).
7126 static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu)
7128 u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
7129 struct vcpu_vmx *vmx = to_vmx(vcpu);
7130 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7131 u32 exit_reason = vmx->exit_reason;
7133 trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
7134 vmcs_readl(EXIT_QUALIFICATION),
7135 vmx->idt_vectoring_info,
7137 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
7140 if (vmx->nested.nested_run_pending)
7143 if (unlikely(vmx->fail)) {
7144 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
7145 vmcs_read32(VM_INSTRUCTION_ERROR));
7149 switch (exit_reason) {
7150 case EXIT_REASON_EXCEPTION_NMI:
7151 if (!is_exception(intr_info))
7153 else if (is_page_fault(intr_info))
7155 else if (is_no_device(intr_info) &&
7156 !(vmcs12->guest_cr0 & X86_CR0_TS))
7158 return vmcs12->exception_bitmap &
7159 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
7160 case EXIT_REASON_EXTERNAL_INTERRUPT:
7162 case EXIT_REASON_TRIPLE_FAULT:
7164 case EXIT_REASON_PENDING_INTERRUPT:
7165 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
7166 case EXIT_REASON_NMI_WINDOW:
7167 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
7168 case EXIT_REASON_TASK_SWITCH:
7170 case EXIT_REASON_CPUID:
7171 if (kvm_register_read(vcpu, VCPU_REGS_RAX) == 0xa)
7174 case EXIT_REASON_HLT:
7175 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
7176 case EXIT_REASON_INVD:
7178 case EXIT_REASON_INVLPG:
7179 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
7180 case EXIT_REASON_RDPMC:
7181 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
7182 case EXIT_REASON_RDTSC:
7183 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
7184 case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
7185 case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
7186 case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
7187 case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
7188 case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
7189 case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
7191 * VMX instructions trap unconditionally. This allows L1 to
7192 * emulate them for its L2 guest, i.e., allows 3-level nesting!
7195 case EXIT_REASON_CR_ACCESS:
7196 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
7197 case EXIT_REASON_DR_ACCESS:
7198 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
7199 case EXIT_REASON_IO_INSTRUCTION:
7200 return nested_vmx_exit_handled_io(vcpu, vmcs12);
7201 case EXIT_REASON_MSR_READ:
7202 case EXIT_REASON_MSR_WRITE:
7203 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
7204 case EXIT_REASON_INVALID_STATE:
7206 case EXIT_REASON_MWAIT_INSTRUCTION:
7207 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
7208 case EXIT_REASON_MONITOR_INSTRUCTION:
7209 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
7210 case EXIT_REASON_PAUSE_INSTRUCTION:
7211 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
7212 nested_cpu_has2(vmcs12,
7213 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
7214 case EXIT_REASON_MCE_DURING_VMENTRY:
7216 case EXIT_REASON_TPR_BELOW_THRESHOLD:
7217 return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
7218 case EXIT_REASON_APIC_ACCESS:
7219 return nested_cpu_has2(vmcs12,
7220 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
7221 case EXIT_REASON_EPT_VIOLATION:
7223 * L0 always deals with the EPT violation. If nested EPT is
7224 * used, and the nested mmu code discovers that the address is
7225 * missing in the guest EPT table (EPT12), the EPT violation
7226 * will be injected with nested_ept_inject_page_fault()
7229 case EXIT_REASON_EPT_MISCONFIG:
7231 * L2 never uses directly L1's EPT, but rather L0's own EPT
7232 * table (shadow on EPT) or a merged EPT table that L0 built
7233 * (EPT on EPT). So any problems with the structure of the
7234 * table is L0's fault.
7237 case EXIT_REASON_WBINVD:
7238 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
7239 case EXIT_REASON_XSETBV:
7246 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
7248 *info1 = vmcs_readl(EXIT_QUALIFICATION);
7249 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
7253 * The guest has exited. See if we can fix it or if we need userspace
7256 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
7258 struct vcpu_vmx *vmx = to_vmx(vcpu);
7259 u32 exit_reason = vmx->exit_reason;
7260 u32 vectoring_info = vmx->idt_vectoring_info;
7262 /* If guest state is invalid, start emulating */
7263 if (vmx->emulation_required)
7264 return handle_invalid_guest_state(vcpu);
7266 if (is_guest_mode(vcpu) && nested_vmx_exit_handled(vcpu)) {
7267 nested_vmx_vmexit(vcpu, exit_reason,
7268 vmcs_read32(VM_EXIT_INTR_INFO),
7269 vmcs_readl(EXIT_QUALIFICATION));
7273 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
7274 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
7275 vcpu->run->fail_entry.hardware_entry_failure_reason
7280 if (unlikely(vmx->fail)) {
7281 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
7282 vcpu->run->fail_entry.hardware_entry_failure_reason
7283 = vmcs_read32(VM_INSTRUCTION_ERROR);
7289 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
7290 * delivery event since it indicates guest is accessing MMIO.
7291 * The vm-exit can be triggered again after return to guest that
7292 * will cause infinite loop.
7294 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
7295 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
7296 exit_reason != EXIT_REASON_EPT_VIOLATION &&
7297 exit_reason != EXIT_REASON_TASK_SWITCH)) {
7298 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7299 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
7300 vcpu->run->internal.ndata = 2;
7301 vcpu->run->internal.data[0] = vectoring_info;
7302 vcpu->run->internal.data[1] = exit_reason;
7306 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked &&
7307 !(is_guest_mode(vcpu) && nested_cpu_has_virtual_nmis(
7308 get_vmcs12(vcpu))))) {
7309 if (vmx_interrupt_allowed(vcpu)) {
7310 vmx->soft_vnmi_blocked = 0;
7311 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
7312 vcpu->arch.nmi_pending) {
7314 * This CPU don't support us in finding the end of an
7315 * NMI-blocked window if the guest runs with IRQs
7316 * disabled. So we pull the trigger after 1 s of
7317 * futile waiting, but inform the user about this.
7319 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
7320 "state on VCPU %d after 1 s timeout\n",
7321 __func__, vcpu->vcpu_id);
7322 vmx->soft_vnmi_blocked = 0;
7326 if (exit_reason < kvm_vmx_max_exit_handlers
7327 && kvm_vmx_exit_handlers[exit_reason])
7328 return kvm_vmx_exit_handlers[exit_reason](vcpu);
7330 WARN_ONCE(1, "vmx: unexpected exit reason 0x%x\n", exit_reason);
7331 kvm_queue_exception(vcpu, UD_VECTOR);
7336 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
7338 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7340 if (is_guest_mode(vcpu) &&
7341 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
7344 if (irr == -1 || tpr < irr) {
7345 vmcs_write32(TPR_THRESHOLD, 0);
7349 vmcs_write32(TPR_THRESHOLD, irr);
7352 static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
7354 u32 sec_exec_control;
7357 * There is not point to enable virtualize x2apic without enable
7360 if (!cpu_has_vmx_virtualize_x2apic_mode() ||
7361 !vmx_vm_has_apicv(vcpu->kvm))
7364 if (!vm_need_tpr_shadow(vcpu->kvm))
7367 sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7370 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
7371 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
7373 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
7374 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
7376 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
7378 vmx_set_msr_bitmap(vcpu);
7381 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
7383 struct vcpu_vmx *vmx = to_vmx(vcpu);
7386 * Currently we do not handle the nested case where L2 has an
7387 * APIC access page of its own; that page is still pinned.
7388 * Hence, we skip the case where the VCPU is in guest mode _and_
7389 * L1 prepared an APIC access page for L2.
7391 * For the case where L1 and L2 share the same APIC access page
7392 * (flexpriority=Y but SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES clear
7393 * in the vmcs12), this function will only update either the vmcs01
7394 * or the vmcs02. If the former, the vmcs02 will be updated by
7395 * prepare_vmcs02. If the latter, the vmcs01 will be updated in
7396 * the next L2->L1 exit.
7398 if (!is_guest_mode(vcpu) ||
7399 !nested_cpu_has2(vmx->nested.current_vmcs12,
7400 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
7401 vmcs_write64(APIC_ACCESS_ADDR, hpa);
7404 static void vmx_hwapic_isr_update(struct kvm *kvm, int isr)
7409 if (!vmx_vm_has_apicv(kvm))
7415 status = vmcs_read16(GUEST_INTR_STATUS);
7420 vmcs_write16(GUEST_INTR_STATUS, status);
7424 static void vmx_set_rvi(int vector)
7432 status = vmcs_read16(GUEST_INTR_STATUS);
7433 old = (u8)status & 0xff;
7434 if ((u8)vector != old) {
7436 status |= (u8)vector;
7437 vmcs_write16(GUEST_INTR_STATUS, status);
7441 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
7443 if (!is_guest_mode(vcpu)) {
7444 vmx_set_rvi(max_irr);
7452 * In guest mode. If a vmexit is needed, vmx_check_nested_events
7455 if (nested_exit_on_intr(vcpu))
7459 * Else, fall back to pre-APICv interrupt injection since L2
7460 * is run without virtual interrupt delivery.
7462 if (!kvm_event_needs_reinjection(vcpu) &&
7463 vmx_interrupt_allowed(vcpu)) {
7464 kvm_queue_interrupt(vcpu, max_irr, false);
7465 vmx_inject_irq(vcpu);
7469 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
7471 if (!vmx_vm_has_apicv(vcpu->kvm))
7474 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
7475 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
7476 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
7477 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
7480 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
7484 if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
7485 || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
7488 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
7489 exit_intr_info = vmx->exit_intr_info;
7491 /* Handle machine checks before interrupts are enabled */
7492 if (is_machine_check(exit_intr_info))
7493 kvm_machine_check();
7495 /* We need to handle NMIs before interrupts are enabled */
7496 if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
7497 (exit_intr_info & INTR_INFO_VALID_MASK)) {
7498 kvm_before_handle_nmi(&vmx->vcpu);
7500 kvm_after_handle_nmi(&vmx->vcpu);
7504 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
7506 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
7509 * If external interrupt exists, IF bit is set in rflags/eflags on the
7510 * interrupt stack frame, and interrupt will be enabled on a return
7511 * from interrupt handler.
7513 if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
7514 == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
7515 unsigned int vector;
7516 unsigned long entry;
7518 struct vcpu_vmx *vmx = to_vmx(vcpu);
7519 #ifdef CONFIG_X86_64
7523 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
7524 desc = (gate_desc *)vmx->host_idt_base + vector;
7525 entry = gate_offset(*desc);
7527 #ifdef CONFIG_X86_64
7528 "mov %%" _ASM_SP ", %[sp]\n\t"
7529 "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
7534 "orl $0x200, (%%" _ASM_SP ")\n\t"
7535 __ASM_SIZE(push) " $%c[cs]\n\t"
7536 "call *%[entry]\n\t"
7538 #ifdef CONFIG_X86_64
7543 [ss]"i"(__KERNEL_DS),
7544 [cs]"i"(__KERNEL_CS)
7550 static bool vmx_mpx_supported(void)
7552 return (vmcs_config.vmexit_ctrl & VM_EXIT_CLEAR_BNDCFGS) &&
7553 (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_BNDCFGS);
7556 static bool vmx_xsaves_supported(void)
7558 return vmcs_config.cpu_based_2nd_exec_ctrl &
7559 SECONDARY_EXEC_XSAVES;
7562 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
7567 bool idtv_info_valid;
7569 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
7571 if (cpu_has_virtual_nmis()) {
7572 if (vmx->nmi_known_unmasked)
7575 * Can't use vmx->exit_intr_info since we're not sure what
7576 * the exit reason is.
7578 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
7579 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
7580 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
7582 * SDM 3: 27.7.1.2 (September 2008)
7583 * Re-set bit "block by NMI" before VM entry if vmexit caused by
7584 * a guest IRET fault.
7585 * SDM 3: 23.2.2 (September 2008)
7586 * Bit 12 is undefined in any of the following cases:
7587 * If the VM exit sets the valid bit in the IDT-vectoring
7588 * information field.
7589 * If the VM exit is due to a double fault.
7591 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
7592 vector != DF_VECTOR && !idtv_info_valid)
7593 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
7594 GUEST_INTR_STATE_NMI);
7596 vmx->nmi_known_unmasked =
7597 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
7598 & GUEST_INTR_STATE_NMI);
7599 } else if (unlikely(vmx->soft_vnmi_blocked))
7600 vmx->vnmi_blocked_time +=
7601 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
7604 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
7605 u32 idt_vectoring_info,
7606 int instr_len_field,
7607 int error_code_field)
7611 bool idtv_info_valid;
7613 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
7615 vcpu->arch.nmi_injected = false;
7616 kvm_clear_exception_queue(vcpu);
7617 kvm_clear_interrupt_queue(vcpu);
7619 if (!idtv_info_valid)
7622 kvm_make_request(KVM_REQ_EVENT, vcpu);
7624 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
7625 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
7628 case INTR_TYPE_NMI_INTR:
7629 vcpu->arch.nmi_injected = true;
7631 * SDM 3: 27.7.1.2 (September 2008)
7632 * Clear bit "block by NMI" before VM entry if a NMI
7635 vmx_set_nmi_mask(vcpu, false);
7637 case INTR_TYPE_SOFT_EXCEPTION:
7638 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
7640 case INTR_TYPE_HARD_EXCEPTION:
7641 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
7642 u32 err = vmcs_read32(error_code_field);
7643 kvm_requeue_exception_e(vcpu, vector, err);
7645 kvm_requeue_exception(vcpu, vector);
7647 case INTR_TYPE_SOFT_INTR:
7648 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
7650 case INTR_TYPE_EXT_INTR:
7651 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
7658 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
7660 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
7661 VM_EXIT_INSTRUCTION_LEN,
7662 IDT_VECTORING_ERROR_CODE);
7665 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
7667 __vmx_complete_interrupts(vcpu,
7668 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
7669 VM_ENTRY_INSTRUCTION_LEN,
7670 VM_ENTRY_EXCEPTION_ERROR_CODE);
7672 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
7675 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
7678 struct perf_guest_switch_msr *msrs;
7680 msrs = perf_guest_get_msrs(&nr_msrs);
7685 for (i = 0; i < nr_msrs; i++)
7686 if (msrs[i].host == msrs[i].guest)
7687 clear_atomic_switch_msr(vmx, msrs[i].msr);
7689 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
7693 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
7695 struct vcpu_vmx *vmx = to_vmx(vcpu);
7696 unsigned long debugctlmsr, cr4;
7698 /* Record the guest's net vcpu time for enforced NMI injections. */
7699 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
7700 vmx->entry_time = ktime_get();
7702 /* Don't enter VMX if guest state is invalid, let the exit handler
7703 start emulation until we arrive back to a valid state */
7704 if (vmx->emulation_required)
7707 if (vmx->ple_window_dirty) {
7708 vmx->ple_window_dirty = false;
7709 vmcs_write32(PLE_WINDOW, vmx->ple_window);
7712 if (vmx->nested.sync_shadow_vmcs) {
7713 copy_vmcs12_to_shadow(vmx);
7714 vmx->nested.sync_shadow_vmcs = false;
7717 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
7718 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
7719 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
7720 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
7723 if (unlikely(cr4 != vmx->host_state.vmcs_host_cr4)) {
7724 vmcs_writel(HOST_CR4, cr4);
7725 vmx->host_state.vmcs_host_cr4 = cr4;
7728 /* When single-stepping over STI and MOV SS, we must clear the
7729 * corresponding interruptibility bits in the guest state. Otherwise
7730 * vmentry fails as it then expects bit 14 (BS) in pending debug
7731 * exceptions being set, but that's not correct for the guest debugging
7733 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7734 vmx_set_interrupt_shadow(vcpu, 0);
7736 atomic_switch_perf_msrs(vmx);
7737 debugctlmsr = get_debugctlmsr();
7739 vmx->__launched = vmx->loaded_vmcs->launched;
7741 /* Store host registers */
7742 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
7743 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
7744 "push %%" _ASM_CX " \n\t"
7745 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
7747 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
7748 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
7750 /* Reload cr2 if changed */
7751 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
7752 "mov %%cr2, %%" _ASM_DX " \n\t"
7753 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
7755 "mov %%" _ASM_AX", %%cr2 \n\t"
7757 /* Check if vmlaunch of vmresume is needed */
7758 "cmpl $0, %c[launched](%0) \n\t"
7759 /* Load guest registers. Don't clobber flags. */
7760 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
7761 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
7762 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
7763 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
7764 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
7765 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
7766 #ifdef CONFIG_X86_64
7767 "mov %c[r8](%0), %%r8 \n\t"
7768 "mov %c[r9](%0), %%r9 \n\t"
7769 "mov %c[r10](%0), %%r10 \n\t"
7770 "mov %c[r11](%0), %%r11 \n\t"
7771 "mov %c[r12](%0), %%r12 \n\t"
7772 "mov %c[r13](%0), %%r13 \n\t"
7773 "mov %c[r14](%0), %%r14 \n\t"
7774 "mov %c[r15](%0), %%r15 \n\t"
7776 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
7778 /* Enter guest mode */
7780 __ex(ASM_VMX_VMLAUNCH) "\n\t"
7782 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
7784 /* Save guest registers, load host registers, keep flags */
7785 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
7787 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
7788 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
7789 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
7790 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
7791 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
7792 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
7793 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
7794 #ifdef CONFIG_X86_64
7795 "mov %%r8, %c[r8](%0) \n\t"
7796 "mov %%r9, %c[r9](%0) \n\t"
7797 "mov %%r10, %c[r10](%0) \n\t"
7798 "mov %%r11, %c[r11](%0) \n\t"
7799 "mov %%r12, %c[r12](%0) \n\t"
7800 "mov %%r13, %c[r13](%0) \n\t"
7801 "mov %%r14, %c[r14](%0) \n\t"
7802 "mov %%r15, %c[r15](%0) \n\t"
7804 "mov %%cr2, %%" _ASM_AX " \n\t"
7805 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
7807 "pop %%" _ASM_BP "; pop %%" _ASM_DX " \n\t"
7808 "setbe %c[fail](%0) \n\t"
7809 ".pushsection .rodata \n\t"
7810 ".global vmx_return \n\t"
7811 "vmx_return: " _ASM_PTR " 2b \n\t"
7813 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
7814 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
7815 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
7816 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
7817 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
7818 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
7819 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
7820 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
7821 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
7822 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
7823 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
7824 #ifdef CONFIG_X86_64
7825 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
7826 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
7827 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
7828 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
7829 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
7830 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
7831 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
7832 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
7834 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
7835 [wordsize]"i"(sizeof(ulong))
7837 #ifdef CONFIG_X86_64
7838 , "rax", "rbx", "rdi", "rsi"
7839 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
7841 , "eax", "ebx", "edi", "esi"
7845 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
7847 update_debugctlmsr(debugctlmsr);
7849 #ifndef CONFIG_X86_64
7851 * The sysexit path does not restore ds/es, so we must set them to
7852 * a reasonable value ourselves.
7854 * We can't defer this to vmx_load_host_state() since that function
7855 * may be executed in interrupt context, which saves and restore segments
7856 * around it, nullifying its effect.
7858 loadsegment(ds, __USER_DS);
7859 loadsegment(es, __USER_DS);
7862 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
7863 | (1 << VCPU_EXREG_RFLAGS)
7864 | (1 << VCPU_EXREG_PDPTR)
7865 | (1 << VCPU_EXREG_SEGMENTS)
7866 | (1 << VCPU_EXREG_CR3));
7867 vcpu->arch.regs_dirty = 0;
7869 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
7871 vmx->loaded_vmcs->launched = 1;
7873 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
7874 trace_kvm_exit(vmx->exit_reason, vcpu, KVM_ISA_VMX);
7877 * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
7878 * we did not inject a still-pending event to L1 now because of
7879 * nested_run_pending, we need to re-enable this bit.
7881 if (vmx->nested.nested_run_pending)
7882 kvm_make_request(KVM_REQ_EVENT, vcpu);
7884 vmx->nested.nested_run_pending = 0;
7886 vmx_complete_atomic_exit(vmx);
7887 vmx_recover_nmi_blocking(vmx);
7888 vmx_complete_interrupts(vmx);
7891 static void vmx_load_vmcs01(struct kvm_vcpu *vcpu)
7893 struct vcpu_vmx *vmx = to_vmx(vcpu);
7896 if (vmx->loaded_vmcs == &vmx->vmcs01)
7900 vmx->loaded_vmcs = &vmx->vmcs01;
7902 vmx_vcpu_load(vcpu, cpu);
7907 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
7909 struct vcpu_vmx *vmx = to_vmx(vcpu);
7912 leave_guest_mode(vcpu);
7913 vmx_load_vmcs01(vcpu);
7915 free_loaded_vmcs(vmx->loaded_vmcs);
7916 kfree(vmx->guest_msrs);
7917 kvm_vcpu_uninit(vcpu);
7918 kmem_cache_free(kvm_vcpu_cache, vmx);
7921 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
7924 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
7928 return ERR_PTR(-ENOMEM);
7932 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
7936 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
7937 BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
7941 if (!vmx->guest_msrs) {
7945 vmx->loaded_vmcs = &vmx->vmcs01;
7946 vmx->loaded_vmcs->vmcs = alloc_vmcs();
7947 if (!vmx->loaded_vmcs->vmcs)
7950 kvm_cpu_vmxon(__pa(per_cpu(vmxarea, raw_smp_processor_id())));
7951 loaded_vmcs_init(vmx->loaded_vmcs);
7956 vmx_vcpu_load(&vmx->vcpu, cpu);
7957 vmx->vcpu.cpu = cpu;
7958 err = vmx_vcpu_setup(vmx);
7959 vmx_vcpu_put(&vmx->vcpu);
7963 if (vm_need_virtualize_apic_accesses(kvm)) {
7964 err = alloc_apic_access_page(kvm);
7970 if (!kvm->arch.ept_identity_map_addr)
7971 kvm->arch.ept_identity_map_addr =
7972 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
7973 err = init_rmode_identity_map(kvm);
7978 vmx->nested.current_vmptr = -1ull;
7979 vmx->nested.current_vmcs12 = NULL;
7984 free_loaded_vmcs(vmx->loaded_vmcs);
7986 kfree(vmx->guest_msrs);
7988 kvm_vcpu_uninit(&vmx->vcpu);
7991 kmem_cache_free(kvm_vcpu_cache, vmx);
7992 return ERR_PTR(err);
7995 static void __init vmx_check_processor_compat(void *rtn)
7997 struct vmcs_config vmcs_conf;
8000 if (setup_vmcs_config(&vmcs_conf) < 0)
8002 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
8003 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
8004 smp_processor_id());
8009 static int get_ept_level(void)
8011 return VMX_EPT_DEFAULT_GAW + 1;
8014 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
8018 /* For VT-d and EPT combination
8019 * 1. MMIO: always map as UC
8021 * a. VT-d without snooping control feature: can't guarantee the
8022 * result, try to trust guest.
8023 * b. VT-d with snooping control feature: snooping control feature of
8024 * VT-d engine can guarantee the cache correctness. Just set it
8025 * to WB to keep consistent with host. So the same as item 3.
8026 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
8027 * consistent with host MTRR
8030 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
8031 else if (kvm_arch_has_noncoherent_dma(vcpu->kvm))
8032 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
8033 VMX_EPT_MT_EPTE_SHIFT;
8035 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
8041 static int vmx_get_lpage_level(void)
8043 if (enable_ept && !cpu_has_vmx_ept_1g_page())
8044 return PT_DIRECTORY_LEVEL;
8046 /* For shadow and EPT supported 1GB page */
8047 return PT_PDPE_LEVEL;
8050 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
8052 struct kvm_cpuid_entry2 *best;
8053 struct vcpu_vmx *vmx = to_vmx(vcpu);
8056 vmx->rdtscp_enabled = false;
8057 if (vmx_rdtscp_supported()) {
8058 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8059 if (exec_control & SECONDARY_EXEC_RDTSCP) {
8060 best = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
8061 if (best && (best->edx & bit(X86_FEATURE_RDTSCP)))
8062 vmx->rdtscp_enabled = true;
8064 exec_control &= ~SECONDARY_EXEC_RDTSCP;
8065 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
8071 /* Exposing INVPCID only when PCID is exposed */
8072 best = kvm_find_cpuid_entry(vcpu, 0x7, 0);
8073 if (vmx_invpcid_supported() &&
8074 best && (best->ebx & bit(X86_FEATURE_INVPCID)) &&
8075 guest_cpuid_has_pcid(vcpu)) {
8076 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8077 exec_control |= SECONDARY_EXEC_ENABLE_INVPCID;
8078 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
8081 if (cpu_has_secondary_exec_ctrls()) {
8082 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8083 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
8084 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
8088 best->ebx &= ~bit(X86_FEATURE_INVPCID);
8092 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
8094 if (func == 1 && nested)
8095 entry->ecx |= bit(X86_FEATURE_VMX);
8098 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
8099 struct x86_exception *fault)
8101 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8104 if (fault->error_code & PFERR_RSVD_MASK)
8105 exit_reason = EXIT_REASON_EPT_MISCONFIG;
8107 exit_reason = EXIT_REASON_EPT_VIOLATION;
8108 nested_vmx_vmexit(vcpu, exit_reason, 0, vcpu->arch.exit_qualification);
8109 vmcs12->guest_physical_address = fault->address;
8112 /* Callbacks for nested_ept_init_mmu_context: */
8114 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
8116 /* return the page table to be shadowed - in our case, EPT12 */
8117 return get_vmcs12(vcpu)->ept_pointer;
8120 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
8122 kvm_init_shadow_ept_mmu(vcpu, &vcpu->arch.mmu,
8123 nested_vmx_ept_caps & VMX_EPT_EXECUTE_ONLY_BIT);
8125 vcpu->arch.mmu.set_cr3 = vmx_set_cr3;
8126 vcpu->arch.mmu.get_cr3 = nested_ept_get_cr3;
8127 vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
8129 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
8132 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
8134 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
8137 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
8138 struct x86_exception *fault)
8140 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8142 WARN_ON(!is_guest_mode(vcpu));
8144 /* TODO: also check PFEC_MATCH/MASK, not just EB.PF. */
8145 if (vmcs12->exception_bitmap & (1u << PF_VECTOR))
8146 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
8147 vmcs_read32(VM_EXIT_INTR_INFO),
8148 vmcs_readl(EXIT_QUALIFICATION));
8150 kvm_inject_page_fault(vcpu, fault);
8153 static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu,
8154 struct vmcs12 *vmcs12)
8156 struct vcpu_vmx *vmx = to_vmx(vcpu);
8158 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
8159 /* TODO: Also verify bits beyond physical address width are 0 */
8160 if (!PAGE_ALIGNED(vmcs12->apic_access_addr))
8164 * Translate L1 physical address to host physical
8165 * address for vmcs02. Keep the page pinned, so this
8166 * physical address remains valid. We keep a reference
8167 * to it so we can release it later.
8169 if (vmx->nested.apic_access_page) /* shouldn't happen */
8170 nested_release_page(vmx->nested.apic_access_page);
8171 vmx->nested.apic_access_page =
8172 nested_get_page(vcpu, vmcs12->apic_access_addr);
8175 if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
8176 /* TODO: Also verify bits beyond physical address width are 0 */
8177 if (!PAGE_ALIGNED(vmcs12->virtual_apic_page_addr))
8180 if (vmx->nested.virtual_apic_page) /* shouldn't happen */
8181 nested_release_page(vmx->nested.virtual_apic_page);
8182 vmx->nested.virtual_apic_page =
8183 nested_get_page(vcpu, vmcs12->virtual_apic_page_addr);
8186 * Failing the vm entry is _not_ what the processor does
8187 * but it's basically the only possibility we have.
8188 * We could still enter the guest if CR8 load exits are
8189 * enabled, CR8 store exits are enabled, and virtualize APIC
8190 * access is disabled; in this case the processor would never
8191 * use the TPR shadow and we could simply clear the bit from
8192 * the execution control. But such a configuration is useless,
8193 * so let's keep the code simple.
8195 if (!vmx->nested.virtual_apic_page)
8202 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
8204 u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
8205 struct vcpu_vmx *vmx = to_vmx(vcpu);
8207 if (vcpu->arch.virtual_tsc_khz == 0)
8210 /* Make sure short timeouts reliably trigger an immediate vmexit.
8211 * hrtimer_start does not guarantee this. */
8212 if (preemption_timeout <= 1) {
8213 vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
8217 preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
8218 preemption_timeout *= 1000000;
8219 do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
8220 hrtimer_start(&vmx->nested.preemption_timer,
8221 ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
8225 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
8226 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
8227 * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
8228 * guest in a way that will both be appropriate to L1's requests, and our
8229 * needs. In addition to modifying the active vmcs (which is vmcs02), this
8230 * function also has additional necessary side-effects, like setting various
8231 * vcpu->arch fields.
8233 static void prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
8235 struct vcpu_vmx *vmx = to_vmx(vcpu);
8238 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
8239 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
8240 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
8241 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
8242 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
8243 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
8244 vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
8245 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
8246 vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
8247 vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
8248 vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
8249 vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
8250 vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
8251 vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
8252 vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
8253 vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
8254 vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
8255 vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
8256 vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
8257 vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
8258 vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
8259 vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
8260 vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
8261 vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
8262 vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
8263 vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
8264 vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
8265 vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
8266 vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
8267 vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
8268 vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
8269 vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
8270 vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
8271 vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
8272 vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
8273 vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
8275 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
8276 kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
8277 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
8279 kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
8280 vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
8282 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
8283 vmcs12->vm_entry_intr_info_field);
8284 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
8285 vmcs12->vm_entry_exception_error_code);
8286 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
8287 vmcs12->vm_entry_instruction_len);
8288 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
8289 vmcs12->guest_interruptibility_info);
8290 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
8291 vmx_set_rflags(vcpu, vmcs12->guest_rflags);
8292 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
8293 vmcs12->guest_pending_dbg_exceptions);
8294 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
8295 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
8297 vmcs_write64(VMCS_LINK_POINTER, -1ull);
8299 exec_control = vmcs12->pin_based_vm_exec_control;
8300 exec_control |= vmcs_config.pin_based_exec_ctrl;
8301 exec_control &= ~(PIN_BASED_VMX_PREEMPTION_TIMER |
8302 PIN_BASED_POSTED_INTR);
8303 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
8305 vmx->nested.preemption_timer_expired = false;
8306 if (nested_cpu_has_preemption_timer(vmcs12))
8307 vmx_start_preemption_timer(vcpu);
8310 * Whether page-faults are trapped is determined by a combination of
8311 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
8312 * If enable_ept, L0 doesn't care about page faults and we should
8313 * set all of these to L1's desires. However, if !enable_ept, L0 does
8314 * care about (at least some) page faults, and because it is not easy
8315 * (if at all possible?) to merge L0 and L1's desires, we simply ask
8316 * to exit on each and every L2 page fault. This is done by setting
8317 * MASK=MATCH=0 and (see below) EB.PF=1.
8318 * Note that below we don't need special code to set EB.PF beyond the
8319 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
8320 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
8321 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
8323 * A problem with this approach (when !enable_ept) is that L1 may be
8324 * injected with more page faults than it asked for. This could have
8325 * caused problems, but in practice existing hypervisors don't care.
8326 * To fix this, we will need to emulate the PFEC checking (on the L1
8327 * page tables), using walk_addr(), when injecting PFs to L1.
8329 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
8330 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
8331 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
8332 enable_ept ? vmcs12->page_fault_error_code_match : 0);
8334 if (cpu_has_secondary_exec_ctrls()) {
8335 exec_control = vmx_secondary_exec_control(vmx);
8336 if (!vmx->rdtscp_enabled)
8337 exec_control &= ~SECONDARY_EXEC_RDTSCP;
8338 /* Take the following fields only from vmcs12 */
8339 exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
8340 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
8341 SECONDARY_EXEC_APIC_REGISTER_VIRT);
8342 if (nested_cpu_has(vmcs12,
8343 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
8344 exec_control |= vmcs12->secondary_vm_exec_control;
8346 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) {
8348 * If translation failed, no matter: This feature asks
8349 * to exit when accessing the given address, and if it
8350 * can never be accessed, this feature won't do
8353 if (!vmx->nested.apic_access_page)
8355 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8357 vmcs_write64(APIC_ACCESS_ADDR,
8358 page_to_phys(vmx->nested.apic_access_page));
8359 } else if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm)) {
8361 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8362 kvm_vcpu_reload_apic_access_page(vcpu);
8365 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
8370 * Set host-state according to L0's settings (vmcs12 is irrelevant here)
8371 * Some constant fields are set here by vmx_set_constant_host_state().
8372 * Other fields are different per CPU, and will be set later when
8373 * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
8375 vmx_set_constant_host_state(vmx);
8378 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
8379 * entry, but only if the current (host) sp changed from the value
8380 * we wrote last (vmx->host_rsp). This cache is no longer relevant
8381 * if we switch vmcs, and rather than hold a separate cache per vmcs,
8382 * here we just force the write to happen on entry.
8386 exec_control = vmx_exec_control(vmx); /* L0's desires */
8387 exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
8388 exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
8389 exec_control &= ~CPU_BASED_TPR_SHADOW;
8390 exec_control |= vmcs12->cpu_based_vm_exec_control;
8392 if (exec_control & CPU_BASED_TPR_SHADOW) {
8393 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
8394 page_to_phys(vmx->nested.virtual_apic_page));
8395 vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
8399 * Merging of IO and MSR bitmaps not currently supported.
8400 * Rather, exit every time.
8402 exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
8403 exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
8404 exec_control |= CPU_BASED_UNCOND_IO_EXITING;
8406 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
8408 /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
8409 * bitwise-or of what L1 wants to trap for L2, and what we want to
8410 * trap. Note that CR0.TS also needs updating - we do this later.
8412 update_exception_bitmap(vcpu);
8413 vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
8414 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
8416 /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
8417 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
8418 * bits are further modified by vmx_set_efer() below.
8420 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
8422 /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
8423 * emulated by vmx_set_efer(), below.
8425 vm_entry_controls_init(vmx,
8426 (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER &
8427 ~VM_ENTRY_IA32E_MODE) |
8428 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
8430 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) {
8431 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
8432 vcpu->arch.pat = vmcs12->guest_ia32_pat;
8433 } else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
8434 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
8437 set_cr4_guest_host_mask(vmx);
8439 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)
8440 vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
8442 if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
8443 vmcs_write64(TSC_OFFSET,
8444 vmx->nested.vmcs01_tsc_offset + vmcs12->tsc_offset);
8446 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
8450 * Trivially support vpid by letting L2s share their parent
8451 * L1's vpid. TODO: move to a more elaborate solution, giving
8452 * each L2 its own vpid and exposing the vpid feature to L1.
8454 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
8455 vmx_flush_tlb(vcpu);
8458 if (nested_cpu_has_ept(vmcs12)) {
8459 kvm_mmu_unload(vcpu);
8460 nested_ept_init_mmu_context(vcpu);
8463 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)
8464 vcpu->arch.efer = vmcs12->guest_ia32_efer;
8465 else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
8466 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
8468 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
8469 /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
8470 vmx_set_efer(vcpu, vcpu->arch.efer);
8473 * This sets GUEST_CR0 to vmcs12->guest_cr0, with possibly a modified
8474 * TS bit (for lazy fpu) and bits which we consider mandatory enabled.
8475 * The CR0_READ_SHADOW is what L2 should have expected to read given
8476 * the specifications by L1; It's not enough to take
8477 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
8478 * have more bits than L1 expected.
8480 vmx_set_cr0(vcpu, vmcs12->guest_cr0);
8481 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
8483 vmx_set_cr4(vcpu, vmcs12->guest_cr4);
8484 vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
8486 /* shadow page tables on either EPT or shadow page tables */
8487 kvm_set_cr3(vcpu, vmcs12->guest_cr3);
8488 kvm_mmu_reset_context(vcpu);
8491 vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
8494 * L1 may access the L2's PDPTR, so save them to construct vmcs12
8497 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
8498 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
8499 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
8500 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
8503 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
8504 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
8508 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
8509 * for running an L2 nested guest.
8511 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
8513 struct vmcs12 *vmcs12;
8514 struct vcpu_vmx *vmx = to_vmx(vcpu);
8516 struct loaded_vmcs *vmcs02;
8519 if (!nested_vmx_check_permission(vcpu) ||
8520 !nested_vmx_check_vmcs12(vcpu))
8523 skip_emulated_instruction(vcpu);
8524 vmcs12 = get_vmcs12(vcpu);
8526 if (enable_shadow_vmcs)
8527 copy_shadow_to_vmcs12(vmx);
8530 * The nested entry process starts with enforcing various prerequisites
8531 * on vmcs12 as required by the Intel SDM, and act appropriately when
8532 * they fail: As the SDM explains, some conditions should cause the
8533 * instruction to fail, while others will cause the instruction to seem
8534 * to succeed, but return an EXIT_REASON_INVALID_STATE.
8535 * To speed up the normal (success) code path, we should avoid checking
8536 * for misconfigurations which will anyway be caught by the processor
8537 * when using the merged vmcs02.
8539 if (vmcs12->launch_state == launch) {
8540 nested_vmx_failValid(vcpu,
8541 launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
8542 : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
8546 if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
8547 vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT) {
8548 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
8552 if ((vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_MSR_BITMAPS) &&
8553 !PAGE_ALIGNED(vmcs12->msr_bitmap)) {
8554 /*TODO: Also verify bits beyond physical address width are 0*/
8555 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
8559 if (!nested_get_vmcs12_pages(vcpu, vmcs12)) {
8560 /*TODO: Also verify bits beyond physical address width are 0*/
8561 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
8565 if (vmcs12->vm_entry_msr_load_count > 0 ||
8566 vmcs12->vm_exit_msr_load_count > 0 ||
8567 vmcs12->vm_exit_msr_store_count > 0) {
8568 pr_warn_ratelimited("%s: VMCS MSR_{LOAD,STORE} unsupported\n",
8570 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
8574 if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
8575 nested_vmx_true_procbased_ctls_low,
8576 nested_vmx_procbased_ctls_high) ||
8577 !vmx_control_verify(vmcs12->secondary_vm_exec_control,
8578 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high) ||
8579 !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
8580 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high) ||
8581 !vmx_control_verify(vmcs12->vm_exit_controls,
8582 nested_vmx_true_exit_ctls_low,
8583 nested_vmx_exit_ctls_high) ||
8584 !vmx_control_verify(vmcs12->vm_entry_controls,
8585 nested_vmx_true_entry_ctls_low,
8586 nested_vmx_entry_ctls_high))
8588 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
8592 if (((vmcs12->host_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
8593 ((vmcs12->host_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
8594 nested_vmx_failValid(vcpu,
8595 VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
8599 if (!nested_cr0_valid(vmcs12, vmcs12->guest_cr0) ||
8600 ((vmcs12->guest_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
8601 nested_vmx_entry_failure(vcpu, vmcs12,
8602 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
8605 if (vmcs12->vmcs_link_pointer != -1ull) {
8606 nested_vmx_entry_failure(vcpu, vmcs12,
8607 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_VMCS_LINK_PTR);
8612 * If the load IA32_EFER VM-entry control is 1, the following checks
8613 * are performed on the field for the IA32_EFER MSR:
8614 * - Bits reserved in the IA32_EFER MSR must be 0.
8615 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
8616 * the IA-32e mode guest VM-exit control. It must also be identical
8617 * to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
8620 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER) {
8621 ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
8622 if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
8623 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
8624 ((vmcs12->guest_cr0 & X86_CR0_PG) &&
8625 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))) {
8626 nested_vmx_entry_failure(vcpu, vmcs12,
8627 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
8633 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
8634 * IA32_EFER MSR must be 0 in the field for that register. In addition,
8635 * the values of the LMA and LME bits in the field must each be that of
8636 * the host address-space size VM-exit control.
8638 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
8639 ia32e = (vmcs12->vm_exit_controls &
8640 VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
8641 if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
8642 ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
8643 ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)) {
8644 nested_vmx_entry_failure(vcpu, vmcs12,
8645 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
8651 * We're finally done with prerequisite checking, and can start with
8655 vmcs02 = nested_get_current_vmcs02(vmx);
8659 enter_guest_mode(vcpu);
8661 vmx->nested.vmcs01_tsc_offset = vmcs_read64(TSC_OFFSET);
8663 if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
8664 vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
8667 vmx->loaded_vmcs = vmcs02;
8669 vmx_vcpu_load(vcpu, cpu);
8673 vmx_segment_cache_clear(vmx);
8675 vmcs12->launch_state = 1;
8677 prepare_vmcs02(vcpu, vmcs12);
8679 if (vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT)
8680 return kvm_emulate_halt(vcpu);
8682 vmx->nested.nested_run_pending = 1;
8685 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
8686 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
8687 * returned as far as L1 is concerned. It will only return (and set
8688 * the success flag) when L2 exits (see nested_vmx_vmexit()).
8694 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
8695 * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
8696 * This function returns the new value we should put in vmcs12.guest_cr0.
8697 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
8698 * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
8699 * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
8700 * didn't trap the bit, because if L1 did, so would L0).
8701 * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have
8702 * been modified by L2, and L1 knows it. So just leave the old value of
8703 * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
8704 * isn't relevant, because if L0 traps this bit it can set it to anything.
8705 * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
8706 * changed these bits, and therefore they need to be updated, but L0
8707 * didn't necessarily allow them to be changed in GUEST_CR0 - and rather
8708 * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
8710 static inline unsigned long
8711 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
8714 /*1*/ (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
8715 /*2*/ (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
8716 /*3*/ (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
8717 vcpu->arch.cr0_guest_owned_bits));
8720 static inline unsigned long
8721 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
8724 /*1*/ (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
8725 /*2*/ (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
8726 /*3*/ (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
8727 vcpu->arch.cr4_guest_owned_bits));
8730 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
8731 struct vmcs12 *vmcs12)
8736 if (vcpu->arch.exception.pending && vcpu->arch.exception.reinject) {
8737 nr = vcpu->arch.exception.nr;
8738 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
8740 if (kvm_exception_is_soft(nr)) {
8741 vmcs12->vm_exit_instruction_len =
8742 vcpu->arch.event_exit_inst_len;
8743 idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
8745 idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
8747 if (vcpu->arch.exception.has_error_code) {
8748 idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
8749 vmcs12->idt_vectoring_error_code =
8750 vcpu->arch.exception.error_code;
8753 vmcs12->idt_vectoring_info_field = idt_vectoring;
8754 } else if (vcpu->arch.nmi_injected) {
8755 vmcs12->idt_vectoring_info_field =
8756 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
8757 } else if (vcpu->arch.interrupt.pending) {
8758 nr = vcpu->arch.interrupt.nr;
8759 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
8761 if (vcpu->arch.interrupt.soft) {
8762 idt_vectoring |= INTR_TYPE_SOFT_INTR;
8763 vmcs12->vm_entry_instruction_len =
8764 vcpu->arch.event_exit_inst_len;
8766 idt_vectoring |= INTR_TYPE_EXT_INTR;
8768 vmcs12->idt_vectoring_info_field = idt_vectoring;
8772 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
8774 struct vcpu_vmx *vmx = to_vmx(vcpu);
8776 if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
8777 vmx->nested.preemption_timer_expired) {
8778 if (vmx->nested.nested_run_pending)
8780 nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
8784 if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
8785 if (vmx->nested.nested_run_pending ||
8786 vcpu->arch.interrupt.pending)
8788 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
8789 NMI_VECTOR | INTR_TYPE_NMI_INTR |
8790 INTR_INFO_VALID_MASK, 0);
8792 * The NMI-triggered VM exit counts as injection:
8793 * clear this one and block further NMIs.
8795 vcpu->arch.nmi_pending = 0;
8796 vmx_set_nmi_mask(vcpu, true);
8800 if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
8801 nested_exit_on_intr(vcpu)) {
8802 if (vmx->nested.nested_run_pending)
8804 nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
8810 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
8813 hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
8816 if (ktime_to_ns(remaining) <= 0)
8819 value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
8820 do_div(value, 1000000);
8821 return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
8825 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
8826 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
8827 * and this function updates it to reflect the changes to the guest state while
8828 * L2 was running (and perhaps made some exits which were handled directly by L0
8829 * without going back to L1), and to reflect the exit reason.
8830 * Note that we do not have to copy here all VMCS fields, just those that
8831 * could have changed by the L2 guest or the exit - i.e., the guest-state and
8832 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
8833 * which already writes to vmcs12 directly.
8835 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
8836 u32 exit_reason, u32 exit_intr_info,
8837 unsigned long exit_qualification)
8839 /* update guest state fields: */
8840 vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
8841 vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
8843 vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
8844 vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
8845 vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
8847 vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
8848 vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
8849 vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
8850 vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
8851 vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
8852 vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
8853 vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
8854 vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
8855 vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
8856 vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
8857 vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
8858 vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
8859 vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
8860 vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
8861 vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
8862 vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
8863 vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
8864 vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
8865 vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
8866 vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
8867 vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
8868 vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
8869 vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
8870 vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
8871 vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
8872 vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
8873 vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
8874 vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
8875 vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
8876 vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
8877 vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
8878 vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
8879 vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
8880 vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
8881 vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
8882 vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
8884 vmcs12->guest_interruptibility_info =
8885 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
8886 vmcs12->guest_pending_dbg_exceptions =
8887 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
8888 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
8889 vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
8891 vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
8893 if (nested_cpu_has_preemption_timer(vmcs12)) {
8894 if (vmcs12->vm_exit_controls &
8895 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
8896 vmcs12->vmx_preemption_timer_value =
8897 vmx_get_preemption_timer_value(vcpu);
8898 hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
8902 * In some cases (usually, nested EPT), L2 is allowed to change its
8903 * own CR3 without exiting. If it has changed it, we must keep it.
8904 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
8905 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
8907 * Additionally, restore L2's PDPTR to vmcs12.
8910 vmcs12->guest_cr3 = vmcs_read64(GUEST_CR3);
8911 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
8912 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
8913 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
8914 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
8917 vmcs12->vm_entry_controls =
8918 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
8919 (vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
8921 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) {
8922 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
8923 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
8926 /* TODO: These cannot have changed unless we have MSR bitmaps and
8927 * the relevant bit asks not to trap the change */
8928 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
8929 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
8930 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
8931 vmcs12->guest_ia32_efer = vcpu->arch.efer;
8932 vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
8933 vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
8934 vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
8935 if (vmx_mpx_supported())
8936 vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
8938 /* update exit information fields: */
8940 vmcs12->vm_exit_reason = exit_reason;
8941 vmcs12->exit_qualification = exit_qualification;
8943 vmcs12->vm_exit_intr_info = exit_intr_info;
8944 if ((vmcs12->vm_exit_intr_info &
8945 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
8946 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK))
8947 vmcs12->vm_exit_intr_error_code =
8948 vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
8949 vmcs12->idt_vectoring_info_field = 0;
8950 vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
8951 vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
8953 if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
8954 /* vm_entry_intr_info_field is cleared on exit. Emulate this
8955 * instead of reading the real value. */
8956 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
8959 * Transfer the event that L0 or L1 may wanted to inject into
8960 * L2 to IDT_VECTORING_INFO_FIELD.
8962 vmcs12_save_pending_event(vcpu, vmcs12);
8966 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
8967 * preserved above and would only end up incorrectly in L1.
8969 vcpu->arch.nmi_injected = false;
8970 kvm_clear_exception_queue(vcpu);
8971 kvm_clear_interrupt_queue(vcpu);
8975 * A part of what we need to when the nested L2 guest exits and we want to
8976 * run its L1 parent, is to reset L1's guest state to the host state specified
8978 * This function is to be called not only on normal nested exit, but also on
8979 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
8980 * Failures During or After Loading Guest State").
8981 * This function should be called when the active VMCS is L1's (vmcs01).
8983 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
8984 struct vmcs12 *vmcs12)
8986 struct kvm_segment seg;
8988 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
8989 vcpu->arch.efer = vmcs12->host_ia32_efer;
8990 else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
8991 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
8993 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
8994 vmx_set_efer(vcpu, vcpu->arch.efer);
8996 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
8997 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
8998 vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
9000 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
9001 * actually changed, because it depends on the current state of
9002 * fpu_active (which may have changed).
9003 * Note that vmx_set_cr0 refers to efer set above.
9005 vmx_set_cr0(vcpu, vmcs12->host_cr0);
9007 * If we did fpu_activate()/fpu_deactivate() during L2's run, we need
9008 * to apply the same changes to L1's vmcs. We just set cr0 correctly,
9009 * but we also need to update cr0_guest_host_mask and exception_bitmap.
9011 update_exception_bitmap(vcpu);
9012 vcpu->arch.cr0_guest_owned_bits = (vcpu->fpu_active ? X86_CR0_TS : 0);
9013 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
9016 * Note that CR4_GUEST_HOST_MASK is already set in the original vmcs01
9017 * (KVM doesn't change it)- no reason to call set_cr4_guest_host_mask();
9019 vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
9020 kvm_set_cr4(vcpu, vmcs12->host_cr4);
9022 nested_ept_uninit_mmu_context(vcpu);
9024 kvm_set_cr3(vcpu, vmcs12->host_cr3);
9025 kvm_mmu_reset_context(vcpu);
9028 vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
9032 * Trivially support vpid by letting L2s share their parent
9033 * L1's vpid. TODO: move to a more elaborate solution, giving
9034 * each L2 its own vpid and exposing the vpid feature to L1.
9036 vmx_flush_tlb(vcpu);
9040 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
9041 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
9042 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
9043 vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
9044 vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
9046 /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1. */
9047 if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
9048 vmcs_write64(GUEST_BNDCFGS, 0);
9050 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
9051 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
9052 vcpu->arch.pat = vmcs12->host_ia32_pat;
9054 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
9055 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
9056 vmcs12->host_ia32_perf_global_ctrl);
9058 /* Set L1 segment info according to Intel SDM
9059 27.5.2 Loading Host Segment and Descriptor-Table Registers */
9060 seg = (struct kvm_segment) {
9062 .limit = 0xFFFFFFFF,
9063 .selector = vmcs12->host_cs_selector,
9069 if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
9073 vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
9074 seg = (struct kvm_segment) {
9076 .limit = 0xFFFFFFFF,
9083 seg.selector = vmcs12->host_ds_selector;
9084 vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
9085 seg.selector = vmcs12->host_es_selector;
9086 vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
9087 seg.selector = vmcs12->host_ss_selector;
9088 vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
9089 seg.selector = vmcs12->host_fs_selector;
9090 seg.base = vmcs12->host_fs_base;
9091 vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
9092 seg.selector = vmcs12->host_gs_selector;
9093 seg.base = vmcs12->host_gs_base;
9094 vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
9095 seg = (struct kvm_segment) {
9096 .base = vmcs12->host_tr_base,
9098 .selector = vmcs12->host_tr_selector,
9102 vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
9104 kvm_set_dr(vcpu, 7, 0x400);
9105 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
9109 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
9110 * and modify vmcs12 to make it see what it would expect to see there if
9111 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
9113 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
9115 unsigned long exit_qualification)
9117 struct vcpu_vmx *vmx = to_vmx(vcpu);
9118 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9120 /* trying to cancel vmlaunch/vmresume is a bug */
9121 WARN_ON_ONCE(vmx->nested.nested_run_pending);
9123 leave_guest_mode(vcpu);
9124 prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
9125 exit_qualification);
9127 vmx_load_vmcs01(vcpu);
9129 if ((exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
9130 && nested_exit_intr_ack_set(vcpu)) {
9131 int irq = kvm_cpu_get_interrupt(vcpu);
9133 vmcs12->vm_exit_intr_info = irq |
9134 INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
9137 trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
9138 vmcs12->exit_qualification,
9139 vmcs12->idt_vectoring_info_field,
9140 vmcs12->vm_exit_intr_info,
9141 vmcs12->vm_exit_intr_error_code,
9144 vm_entry_controls_init(vmx, vmcs_read32(VM_ENTRY_CONTROLS));
9145 vm_exit_controls_init(vmx, vmcs_read32(VM_EXIT_CONTROLS));
9146 vmx_segment_cache_clear(vmx);
9148 /* if no vmcs02 cache requested, remove the one we used */
9149 if (VMCS02_POOL_SIZE == 0)
9150 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
9152 load_vmcs12_host_state(vcpu, vmcs12);
9154 /* Update TSC_OFFSET if TSC was changed while L2 ran */
9155 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
9157 /* This is needed for same reason as it was needed in prepare_vmcs02 */
9160 /* Unpin physical memory we referred to in vmcs02 */
9161 if (vmx->nested.apic_access_page) {
9162 nested_release_page(vmx->nested.apic_access_page);
9163 vmx->nested.apic_access_page = NULL;
9165 if (vmx->nested.virtual_apic_page) {
9166 nested_release_page(vmx->nested.virtual_apic_page);
9167 vmx->nested.virtual_apic_page = NULL;
9171 * We are now running in L2, mmu_notifier will force to reload the
9172 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
9174 kvm_vcpu_reload_apic_access_page(vcpu);
9177 * Exiting from L2 to L1, we're now back to L1 which thinks it just
9178 * finished a VMLAUNCH or VMRESUME instruction, so we need to set the
9179 * success or failure flag accordingly.
9181 if (unlikely(vmx->fail)) {
9183 nested_vmx_failValid(vcpu, vmcs_read32(VM_INSTRUCTION_ERROR));
9185 nested_vmx_succeed(vcpu);
9186 if (enable_shadow_vmcs)
9187 vmx->nested.sync_shadow_vmcs = true;
9189 /* in case we halted in L2 */
9190 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9194 * Forcibly leave nested mode in order to be able to reset the VCPU later on.
9196 static void vmx_leave_nested(struct kvm_vcpu *vcpu)
9198 if (is_guest_mode(vcpu))
9199 nested_vmx_vmexit(vcpu, -1, 0, 0);
9200 free_nested(to_vmx(vcpu));
9204 * L1's failure to enter L2 is a subset of a normal exit, as explained in
9205 * 23.7 "VM-entry failures during or after loading guest state" (this also
9206 * lists the acceptable exit-reason and exit-qualification parameters).
9207 * It should only be called before L2 actually succeeded to run, and when
9208 * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
9210 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
9211 struct vmcs12 *vmcs12,
9212 u32 reason, unsigned long qualification)
9214 load_vmcs12_host_state(vcpu, vmcs12);
9215 vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
9216 vmcs12->exit_qualification = qualification;
9217 nested_vmx_succeed(vcpu);
9218 if (enable_shadow_vmcs)
9219 to_vmx(vcpu)->nested.sync_shadow_vmcs = true;
9222 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
9223 struct x86_instruction_info *info,
9224 enum x86_intercept_stage stage)
9226 return X86EMUL_CONTINUE;
9229 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
9232 shrink_ple_window(vcpu);
9235 static struct kvm_x86_ops vmx_x86_ops = {
9236 .cpu_has_kvm_support = cpu_has_kvm_support,
9237 .disabled_by_bios = vmx_disabled_by_bios,
9238 .hardware_setup = hardware_setup,
9239 .hardware_unsetup = hardware_unsetup,
9240 .check_processor_compatibility = vmx_check_processor_compat,
9241 .hardware_enable = hardware_enable,
9242 .hardware_disable = hardware_disable,
9243 .cpu_has_accelerated_tpr = report_flexpriority,
9245 .vcpu_create = vmx_create_vcpu,
9246 .vcpu_free = vmx_free_vcpu,
9247 .vcpu_reset = vmx_vcpu_reset,
9249 .prepare_guest_switch = vmx_save_host_state,
9250 .vcpu_load = vmx_vcpu_load,
9251 .vcpu_put = vmx_vcpu_put,
9253 .update_db_bp_intercept = update_exception_bitmap,
9254 .get_msr = vmx_get_msr,
9255 .set_msr = vmx_set_msr,
9256 .get_segment_base = vmx_get_segment_base,
9257 .get_segment = vmx_get_segment,
9258 .set_segment = vmx_set_segment,
9259 .get_cpl = vmx_get_cpl,
9260 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
9261 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
9262 .decache_cr3 = vmx_decache_cr3,
9263 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
9264 .set_cr0 = vmx_set_cr0,
9265 .set_cr3 = vmx_set_cr3,
9266 .set_cr4 = vmx_set_cr4,
9267 .set_efer = vmx_set_efer,
9268 .get_idt = vmx_get_idt,
9269 .set_idt = vmx_set_idt,
9270 .get_gdt = vmx_get_gdt,
9271 .set_gdt = vmx_set_gdt,
9272 .get_dr6 = vmx_get_dr6,
9273 .set_dr6 = vmx_set_dr6,
9274 .set_dr7 = vmx_set_dr7,
9275 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
9276 .cache_reg = vmx_cache_reg,
9277 .get_rflags = vmx_get_rflags,
9278 .set_rflags = vmx_set_rflags,
9279 .fpu_deactivate = vmx_fpu_deactivate,
9281 .tlb_flush = vmx_flush_tlb,
9283 .run = vmx_vcpu_run,
9284 .handle_exit = vmx_handle_exit,
9285 .skip_emulated_instruction = skip_emulated_instruction,
9286 .set_interrupt_shadow = vmx_set_interrupt_shadow,
9287 .get_interrupt_shadow = vmx_get_interrupt_shadow,
9288 .patch_hypercall = vmx_patch_hypercall,
9289 .set_irq = vmx_inject_irq,
9290 .set_nmi = vmx_inject_nmi,
9291 .queue_exception = vmx_queue_exception,
9292 .cancel_injection = vmx_cancel_injection,
9293 .interrupt_allowed = vmx_interrupt_allowed,
9294 .nmi_allowed = vmx_nmi_allowed,
9295 .get_nmi_mask = vmx_get_nmi_mask,
9296 .set_nmi_mask = vmx_set_nmi_mask,
9297 .enable_nmi_window = enable_nmi_window,
9298 .enable_irq_window = enable_irq_window,
9299 .update_cr8_intercept = update_cr8_intercept,
9300 .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode,
9301 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
9302 .vm_has_apicv = vmx_vm_has_apicv,
9303 .load_eoi_exitmap = vmx_load_eoi_exitmap,
9304 .hwapic_irr_update = vmx_hwapic_irr_update,
9305 .hwapic_isr_update = vmx_hwapic_isr_update,
9306 .sync_pir_to_irr = vmx_sync_pir_to_irr,
9307 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
9309 .set_tss_addr = vmx_set_tss_addr,
9310 .get_tdp_level = get_ept_level,
9311 .get_mt_mask = vmx_get_mt_mask,
9313 .get_exit_info = vmx_get_exit_info,
9315 .get_lpage_level = vmx_get_lpage_level,
9317 .cpuid_update = vmx_cpuid_update,
9319 .rdtscp_supported = vmx_rdtscp_supported,
9320 .invpcid_supported = vmx_invpcid_supported,
9322 .set_supported_cpuid = vmx_set_supported_cpuid,
9324 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
9326 .set_tsc_khz = vmx_set_tsc_khz,
9327 .read_tsc_offset = vmx_read_tsc_offset,
9328 .write_tsc_offset = vmx_write_tsc_offset,
9329 .adjust_tsc_offset = vmx_adjust_tsc_offset,
9330 .compute_tsc_offset = vmx_compute_tsc_offset,
9331 .read_l1_tsc = vmx_read_l1_tsc,
9333 .set_tdp_cr3 = vmx_set_cr3,
9335 .check_intercept = vmx_check_intercept,
9336 .handle_external_intr = vmx_handle_external_intr,
9337 .mpx_supported = vmx_mpx_supported,
9338 .xsaves_supported = vmx_xsaves_supported,
9340 .check_nested_events = vmx_check_nested_events,
9342 .sched_in = vmx_sched_in,
9345 static int __init vmx_init(void)
9347 int r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
9348 __alignof__(struct vcpu_vmx), THIS_MODULE);
9353 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
9354 crash_vmclear_local_loaded_vmcss);
9360 static void __exit vmx_exit(void)
9363 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
9370 module_init(vmx_init)
9371 module_exit(vmx_exit)