2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
24 #include <linux/kvm_host.h>
25 #include <linux/module.h>
26 #include <linux/kernel.h>
28 #include <linux/highmem.h>
29 #include <linux/sched.h>
30 #include <linux/moduleparam.h>
31 #include <linux/mod_devicetable.h>
32 #include <linux/trace_events.h>
33 #include <linux/slab.h>
34 #include <linux/tboot.h>
35 #include <linux/hrtimer.h>
36 #include "kvm_cache_regs.h"
43 #include <asm/virtext.h>
45 #include <asm/fpu/internal.h>
46 #include <asm/perf_event.h>
47 #include <asm/debugreg.h>
48 #include <asm/kexec.h>
50 #include <asm/irq_remapping.h>
55 #define __ex(x) __kvm_handle_fault_on_reboot(x)
56 #define __ex_clear(x, reg) \
57 ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
59 MODULE_AUTHOR("Qumranet");
60 MODULE_LICENSE("GPL");
62 static const struct x86_cpu_id vmx_cpu_id[] = {
63 X86_FEATURE_MATCH(X86_FEATURE_VMX),
66 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
68 static bool __read_mostly enable_vpid = 1;
69 module_param_named(vpid, enable_vpid, bool, 0444);
71 static bool __read_mostly flexpriority_enabled = 1;
72 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
74 static bool __read_mostly enable_ept = 1;
75 module_param_named(ept, enable_ept, bool, S_IRUGO);
77 static bool __read_mostly enable_unrestricted_guest = 1;
78 module_param_named(unrestricted_guest,
79 enable_unrestricted_guest, bool, S_IRUGO);
81 static bool __read_mostly enable_ept_ad_bits = 1;
82 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
84 static bool __read_mostly emulate_invalid_guest_state = true;
85 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
87 static bool __read_mostly fasteoi = 1;
88 module_param(fasteoi, bool, S_IRUGO);
90 static bool __read_mostly enable_apicv = 1;
91 module_param(enable_apicv, bool, S_IRUGO);
93 static bool __read_mostly enable_shadow_vmcs = 1;
94 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
96 * If nested=1, nested virtualization is supported, i.e., guests may use
97 * VMX and be a hypervisor for its own guests. If nested=0, guests may not
98 * use VMX instructions.
100 static bool __read_mostly nested = 0;
101 module_param(nested, bool, S_IRUGO);
103 static u64 __read_mostly host_xss;
105 static bool __read_mostly enable_pml = 1;
106 module_param_named(pml, enable_pml, bool, S_IRUGO);
108 #define KVM_VMX_TSC_MULTIPLIER_MAX 0xffffffffffffffffULL
110 /* Guest_tsc -> host_tsc conversion requires 64-bit division. */
111 static int __read_mostly cpu_preemption_timer_multi;
112 static bool __read_mostly enable_preemption_timer = 1;
114 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
117 #define KVM_GUEST_CR0_MASK (X86_CR0_NW | X86_CR0_CD)
118 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST (X86_CR0_WP | X86_CR0_NE)
119 #define KVM_VM_CR0_ALWAYS_ON \
120 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
121 #define KVM_CR4_GUEST_OWNED_BITS \
122 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
123 | X86_CR4_OSXMMEXCPT | X86_CR4_TSD)
125 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
126 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
128 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
130 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
133 * Hyper-V requires all of these, so mark them as supported even though
134 * they are just treated the same as all-context.
136 #define VMX_VPID_EXTENT_SUPPORTED_MASK \
137 (VMX_VPID_EXTENT_INDIVIDUAL_ADDR_BIT | \
138 VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT | \
139 VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT | \
140 VMX_VPID_EXTENT_SINGLE_NON_GLOBAL_BIT)
143 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
144 * ple_gap: upper bound on the amount of time between two successive
145 * executions of PAUSE in a loop. Also indicate if ple enabled.
146 * According to test, this time is usually smaller than 128 cycles.
147 * ple_window: upper bound on the amount of time a guest is allowed to execute
148 * in a PAUSE loop. Tests indicate that most spinlocks are held for
149 * less than 2^12 cycles
150 * Time is measured based on a counter that runs at the same rate as the TSC,
151 * refer SDM volume 3b section 21.6.13 & 22.1.3.
153 #define KVM_VMX_DEFAULT_PLE_GAP 128
154 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
155 #define KVM_VMX_DEFAULT_PLE_WINDOW_GROW 2
156 #define KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK 0
157 #define KVM_VMX_DEFAULT_PLE_WINDOW_MAX \
158 INT_MAX / KVM_VMX_DEFAULT_PLE_WINDOW_GROW
160 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
161 module_param(ple_gap, int, S_IRUGO);
163 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
164 module_param(ple_window, int, S_IRUGO);
166 /* Default doubles per-vcpu window every exit. */
167 static int ple_window_grow = KVM_VMX_DEFAULT_PLE_WINDOW_GROW;
168 module_param(ple_window_grow, int, S_IRUGO);
170 /* Default resets per-vcpu window every exit to ple_window. */
171 static int ple_window_shrink = KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK;
172 module_param(ple_window_shrink, int, S_IRUGO);
174 /* Default is to compute the maximum so we can never overflow. */
175 static int ple_window_actual_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
176 static int ple_window_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
177 module_param(ple_window_max, int, S_IRUGO);
179 extern const ulong vmx_return;
181 #define NR_AUTOLOAD_MSRS 8
182 #define VMCS02_POOL_SIZE 1
191 * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
192 * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
193 * loaded on this CPU (so we can clear them if the CPU goes down).
197 struct vmcs *shadow_vmcs;
200 struct list_head loaded_vmcss_on_cpu_link;
203 struct shared_msr_entry {
210 * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
211 * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
212 * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
213 * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
214 * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
215 * More than one of these structures may exist, if L1 runs multiple L2 guests.
216 * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
217 * underlying hardware which will be used to run L2.
218 * This structure is packed to ensure that its layout is identical across
219 * machines (necessary for live migration).
220 * If there are changes in this struct, VMCS12_REVISION must be changed.
222 typedef u64 natural_width;
223 struct __packed vmcs12 {
224 /* According to the Intel spec, a VMCS region must start with the
225 * following two fields. Then follow implementation-specific data.
230 u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
231 u32 padding[7]; /* room for future expansion */
236 u64 vm_exit_msr_store_addr;
237 u64 vm_exit_msr_load_addr;
238 u64 vm_entry_msr_load_addr;
240 u64 virtual_apic_page_addr;
241 u64 apic_access_addr;
242 u64 posted_intr_desc_addr;
244 u64 eoi_exit_bitmap0;
245 u64 eoi_exit_bitmap1;
246 u64 eoi_exit_bitmap2;
247 u64 eoi_exit_bitmap3;
249 u64 guest_physical_address;
250 u64 vmcs_link_pointer;
251 u64 guest_ia32_debugctl;
254 u64 guest_ia32_perf_global_ctrl;
262 u64 host_ia32_perf_global_ctrl;
263 u64 padding64[8]; /* room for future expansion */
265 * To allow migration of L1 (complete with its L2 guests) between
266 * machines of different natural widths (32 or 64 bit), we cannot have
267 * unsigned long fields with no explict size. We use u64 (aliased
268 * natural_width) instead. Luckily, x86 is little-endian.
270 natural_width cr0_guest_host_mask;
271 natural_width cr4_guest_host_mask;
272 natural_width cr0_read_shadow;
273 natural_width cr4_read_shadow;
274 natural_width cr3_target_value0;
275 natural_width cr3_target_value1;
276 natural_width cr3_target_value2;
277 natural_width cr3_target_value3;
278 natural_width exit_qualification;
279 natural_width guest_linear_address;
280 natural_width guest_cr0;
281 natural_width guest_cr3;
282 natural_width guest_cr4;
283 natural_width guest_es_base;
284 natural_width guest_cs_base;
285 natural_width guest_ss_base;
286 natural_width guest_ds_base;
287 natural_width guest_fs_base;
288 natural_width guest_gs_base;
289 natural_width guest_ldtr_base;
290 natural_width guest_tr_base;
291 natural_width guest_gdtr_base;
292 natural_width guest_idtr_base;
293 natural_width guest_dr7;
294 natural_width guest_rsp;
295 natural_width guest_rip;
296 natural_width guest_rflags;
297 natural_width guest_pending_dbg_exceptions;
298 natural_width guest_sysenter_esp;
299 natural_width guest_sysenter_eip;
300 natural_width host_cr0;
301 natural_width host_cr3;
302 natural_width host_cr4;
303 natural_width host_fs_base;
304 natural_width host_gs_base;
305 natural_width host_tr_base;
306 natural_width host_gdtr_base;
307 natural_width host_idtr_base;
308 natural_width host_ia32_sysenter_esp;
309 natural_width host_ia32_sysenter_eip;
310 natural_width host_rsp;
311 natural_width host_rip;
312 natural_width paddingl[8]; /* room for future expansion */
313 u32 pin_based_vm_exec_control;
314 u32 cpu_based_vm_exec_control;
315 u32 exception_bitmap;
316 u32 page_fault_error_code_mask;
317 u32 page_fault_error_code_match;
318 u32 cr3_target_count;
319 u32 vm_exit_controls;
320 u32 vm_exit_msr_store_count;
321 u32 vm_exit_msr_load_count;
322 u32 vm_entry_controls;
323 u32 vm_entry_msr_load_count;
324 u32 vm_entry_intr_info_field;
325 u32 vm_entry_exception_error_code;
326 u32 vm_entry_instruction_len;
328 u32 secondary_vm_exec_control;
329 u32 vm_instruction_error;
331 u32 vm_exit_intr_info;
332 u32 vm_exit_intr_error_code;
333 u32 idt_vectoring_info_field;
334 u32 idt_vectoring_error_code;
335 u32 vm_exit_instruction_len;
336 u32 vmx_instruction_info;
343 u32 guest_ldtr_limit;
345 u32 guest_gdtr_limit;
346 u32 guest_idtr_limit;
347 u32 guest_es_ar_bytes;
348 u32 guest_cs_ar_bytes;
349 u32 guest_ss_ar_bytes;
350 u32 guest_ds_ar_bytes;
351 u32 guest_fs_ar_bytes;
352 u32 guest_gs_ar_bytes;
353 u32 guest_ldtr_ar_bytes;
354 u32 guest_tr_ar_bytes;
355 u32 guest_interruptibility_info;
356 u32 guest_activity_state;
357 u32 guest_sysenter_cs;
358 u32 host_ia32_sysenter_cs;
359 u32 vmx_preemption_timer_value;
360 u32 padding32[7]; /* room for future expansion */
361 u16 virtual_processor_id;
363 u16 guest_es_selector;
364 u16 guest_cs_selector;
365 u16 guest_ss_selector;
366 u16 guest_ds_selector;
367 u16 guest_fs_selector;
368 u16 guest_gs_selector;
369 u16 guest_ldtr_selector;
370 u16 guest_tr_selector;
371 u16 guest_intr_status;
372 u16 host_es_selector;
373 u16 host_cs_selector;
374 u16 host_ss_selector;
375 u16 host_ds_selector;
376 u16 host_fs_selector;
377 u16 host_gs_selector;
378 u16 host_tr_selector;
382 * VMCS12_REVISION is an arbitrary id that should be changed if the content or
383 * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
384 * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
386 #define VMCS12_REVISION 0x11e57ed0
389 * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
390 * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
391 * current implementation, 4K are reserved to avoid future complications.
393 #define VMCS12_SIZE 0x1000
395 /* Used to remember the last vmcs02 used for some recently used vmcs12s */
397 struct list_head list;
399 struct loaded_vmcs vmcs02;
403 * The nested_vmx structure is part of vcpu_vmx, and holds information we need
404 * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
407 /* Has the level1 guest done vmxon? */
411 /* The guest-physical address of the current VMCS L1 keeps for L2 */
413 /* The host-usable pointer to the above */
414 struct page *current_vmcs12_page;
415 struct vmcs12 *current_vmcs12;
417 * Cache of the guest's VMCS, existing outside of guest memory.
418 * Loaded from guest memory during VMPTRLD. Flushed to guest
419 * memory during VMXOFF, VMCLEAR, VMPTRLD.
421 struct vmcs12 *cached_vmcs12;
423 * Indicates if the shadow vmcs must be updated with the
424 * data hold by vmcs12
426 bool sync_shadow_vmcs;
428 /* vmcs02_list cache of VMCSs recently used to run L2 guests */
429 struct list_head vmcs02_pool;
431 bool change_vmcs01_virtual_x2apic_mode;
432 /* L2 must run next, and mustn't decide to exit to L1. */
433 bool nested_run_pending;
435 * Guest pages referred to in vmcs02 with host-physical pointers, so
436 * we must keep them pinned while L2 runs.
438 struct page *apic_access_page;
439 struct page *virtual_apic_page;
440 struct page *pi_desc_page;
441 struct pi_desc *pi_desc;
445 unsigned long *msr_bitmap;
447 struct hrtimer preemption_timer;
448 bool preemption_timer_expired;
450 /* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
457 * We only store the "true" versions of the VMX capability MSRs. We
458 * generate the "non-true" versions by setting the must-be-1 bits
459 * according to the SDM.
461 u32 nested_vmx_procbased_ctls_low;
462 u32 nested_vmx_procbased_ctls_high;
463 u32 nested_vmx_secondary_ctls_low;
464 u32 nested_vmx_secondary_ctls_high;
465 u32 nested_vmx_pinbased_ctls_low;
466 u32 nested_vmx_pinbased_ctls_high;
467 u32 nested_vmx_exit_ctls_low;
468 u32 nested_vmx_exit_ctls_high;
469 u32 nested_vmx_entry_ctls_low;
470 u32 nested_vmx_entry_ctls_high;
471 u32 nested_vmx_misc_low;
472 u32 nested_vmx_misc_high;
473 u32 nested_vmx_ept_caps;
474 u32 nested_vmx_vpid_caps;
475 u64 nested_vmx_basic;
476 u64 nested_vmx_cr0_fixed0;
477 u64 nested_vmx_cr0_fixed1;
478 u64 nested_vmx_cr4_fixed0;
479 u64 nested_vmx_cr4_fixed1;
480 u64 nested_vmx_vmcs_enum;
483 #define POSTED_INTR_ON 0
484 #define POSTED_INTR_SN 1
486 /* Posted-Interrupt Descriptor */
488 u32 pir[8]; /* Posted interrupt requested */
491 /* bit 256 - Outstanding Notification */
493 /* bit 257 - Suppress Notification */
495 /* bit 271:258 - Reserved */
497 /* bit 279:272 - Notification Vector */
499 /* bit 287:280 - Reserved */
501 /* bit 319:288 - Notification Destination */
509 static bool pi_test_and_set_on(struct pi_desc *pi_desc)
511 return test_and_set_bit(POSTED_INTR_ON,
512 (unsigned long *)&pi_desc->control);
515 static bool pi_test_and_clear_on(struct pi_desc *pi_desc)
517 return test_and_clear_bit(POSTED_INTR_ON,
518 (unsigned long *)&pi_desc->control);
521 static int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
523 return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
526 static inline void pi_clear_sn(struct pi_desc *pi_desc)
528 return clear_bit(POSTED_INTR_SN,
529 (unsigned long *)&pi_desc->control);
532 static inline void pi_set_sn(struct pi_desc *pi_desc)
534 return set_bit(POSTED_INTR_SN,
535 (unsigned long *)&pi_desc->control);
538 static inline void pi_clear_on(struct pi_desc *pi_desc)
540 clear_bit(POSTED_INTR_ON,
541 (unsigned long *)&pi_desc->control);
544 static inline int pi_test_on(struct pi_desc *pi_desc)
546 return test_bit(POSTED_INTR_ON,
547 (unsigned long *)&pi_desc->control);
550 static inline int pi_test_sn(struct pi_desc *pi_desc)
552 return test_bit(POSTED_INTR_SN,
553 (unsigned long *)&pi_desc->control);
557 struct kvm_vcpu vcpu;
558 unsigned long host_rsp;
560 bool nmi_known_unmasked;
562 u32 idt_vectoring_info;
564 struct shared_msr_entry *guest_msrs;
567 unsigned long host_idt_base;
569 u64 msr_host_kernel_gs_base;
570 u64 msr_guest_kernel_gs_base;
572 u32 vm_entry_controls_shadow;
573 u32 vm_exit_controls_shadow;
575 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
576 * non-nested (L1) guest, it always points to vmcs01. For a nested
577 * guest (L2), it points to a different VMCS.
579 struct loaded_vmcs vmcs01;
580 struct loaded_vmcs *loaded_vmcs;
581 bool __launched; /* temporary, used in vmx_vcpu_run */
582 struct msr_autoload {
584 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
585 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
589 u16 fs_sel, gs_sel, ldt_sel;
593 int gs_ldt_reload_needed;
594 int fs_reload_needed;
595 u64 msr_host_bndcfgs;
596 unsigned long vmcs_host_cr4; /* May not match real cr4 */
601 struct kvm_segment segs[8];
604 u32 bitmask; /* 4 bits per segment (1 bit per field) */
605 struct kvm_save_segment {
613 bool emulation_required;
617 /* Posted interrupt descriptor */
618 struct pi_desc pi_desc;
620 /* Support for a guest hypervisor (nested VMX) */
621 struct nested_vmx nested;
623 /* Dynamic PLE window. */
625 bool ple_window_dirty;
627 /* Support for PML */
628 #define PML_ENTITY_NUM 512
631 /* apic deadline value in host tsc */
634 u64 current_tsc_ratio;
636 bool guest_pkru_valid;
641 * Only bits masked by msr_ia32_feature_control_valid_bits can be set in
642 * msr_ia32_feature_control. FEATURE_CONTROL_LOCKED is always included
643 * in msr_ia32_feature_control_valid_bits.
645 u64 msr_ia32_feature_control;
646 u64 msr_ia32_feature_control_valid_bits;
649 enum segment_cache_field {
658 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
660 return container_of(vcpu, struct vcpu_vmx, vcpu);
663 static struct pi_desc *vcpu_to_pi_desc(struct kvm_vcpu *vcpu)
665 return &(to_vmx(vcpu)->pi_desc);
668 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
669 #define FIELD(number, name) [number] = VMCS12_OFFSET(name)
670 #define FIELD64(number, name) [number] = VMCS12_OFFSET(name), \
671 [number##_HIGH] = VMCS12_OFFSET(name)+4
674 static unsigned long shadow_read_only_fields[] = {
676 * We do NOT shadow fields that are modified when L0
677 * traps and emulates any vmx instruction (e.g. VMPTRLD,
678 * VMXON...) executed by L1.
679 * For example, VM_INSTRUCTION_ERROR is read
680 * by L1 if a vmx instruction fails (part of the error path).
681 * Note the code assumes this logic. If for some reason
682 * we start shadowing these fields then we need to
683 * force a shadow sync when L0 emulates vmx instructions
684 * (e.g. force a sync if VM_INSTRUCTION_ERROR is modified
685 * by nested_vmx_failValid)
689 VM_EXIT_INSTRUCTION_LEN,
690 IDT_VECTORING_INFO_FIELD,
691 IDT_VECTORING_ERROR_CODE,
692 VM_EXIT_INTR_ERROR_CODE,
694 GUEST_LINEAR_ADDRESS,
695 GUEST_PHYSICAL_ADDRESS
697 static int max_shadow_read_only_fields =
698 ARRAY_SIZE(shadow_read_only_fields);
700 static unsigned long shadow_read_write_fields[] = {
707 GUEST_INTERRUPTIBILITY_INFO,
720 CPU_BASED_VM_EXEC_CONTROL,
721 VM_ENTRY_EXCEPTION_ERROR_CODE,
722 VM_ENTRY_INTR_INFO_FIELD,
723 VM_ENTRY_INSTRUCTION_LEN,
724 VM_ENTRY_EXCEPTION_ERROR_CODE,
730 static int max_shadow_read_write_fields =
731 ARRAY_SIZE(shadow_read_write_fields);
733 static const unsigned short vmcs_field_to_offset_table[] = {
734 FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
735 FIELD(POSTED_INTR_NV, posted_intr_nv),
736 FIELD(GUEST_ES_SELECTOR, guest_es_selector),
737 FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
738 FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
739 FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
740 FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
741 FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
742 FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
743 FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
744 FIELD(GUEST_INTR_STATUS, guest_intr_status),
745 FIELD(HOST_ES_SELECTOR, host_es_selector),
746 FIELD(HOST_CS_SELECTOR, host_cs_selector),
747 FIELD(HOST_SS_SELECTOR, host_ss_selector),
748 FIELD(HOST_DS_SELECTOR, host_ds_selector),
749 FIELD(HOST_FS_SELECTOR, host_fs_selector),
750 FIELD(HOST_GS_SELECTOR, host_gs_selector),
751 FIELD(HOST_TR_SELECTOR, host_tr_selector),
752 FIELD64(IO_BITMAP_A, io_bitmap_a),
753 FIELD64(IO_BITMAP_B, io_bitmap_b),
754 FIELD64(MSR_BITMAP, msr_bitmap),
755 FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
756 FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
757 FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
758 FIELD64(TSC_OFFSET, tsc_offset),
759 FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
760 FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
761 FIELD64(POSTED_INTR_DESC_ADDR, posted_intr_desc_addr),
762 FIELD64(EPT_POINTER, ept_pointer),
763 FIELD64(EOI_EXIT_BITMAP0, eoi_exit_bitmap0),
764 FIELD64(EOI_EXIT_BITMAP1, eoi_exit_bitmap1),
765 FIELD64(EOI_EXIT_BITMAP2, eoi_exit_bitmap2),
766 FIELD64(EOI_EXIT_BITMAP3, eoi_exit_bitmap3),
767 FIELD64(XSS_EXIT_BITMAP, xss_exit_bitmap),
768 FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
769 FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
770 FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
771 FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
772 FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
773 FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
774 FIELD64(GUEST_PDPTR0, guest_pdptr0),
775 FIELD64(GUEST_PDPTR1, guest_pdptr1),
776 FIELD64(GUEST_PDPTR2, guest_pdptr2),
777 FIELD64(GUEST_PDPTR3, guest_pdptr3),
778 FIELD64(GUEST_BNDCFGS, guest_bndcfgs),
779 FIELD64(HOST_IA32_PAT, host_ia32_pat),
780 FIELD64(HOST_IA32_EFER, host_ia32_efer),
781 FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
782 FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
783 FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
784 FIELD(EXCEPTION_BITMAP, exception_bitmap),
785 FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
786 FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
787 FIELD(CR3_TARGET_COUNT, cr3_target_count),
788 FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
789 FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
790 FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
791 FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
792 FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
793 FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
794 FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
795 FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
796 FIELD(TPR_THRESHOLD, tpr_threshold),
797 FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
798 FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
799 FIELD(VM_EXIT_REASON, vm_exit_reason),
800 FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
801 FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
802 FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
803 FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
804 FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
805 FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
806 FIELD(GUEST_ES_LIMIT, guest_es_limit),
807 FIELD(GUEST_CS_LIMIT, guest_cs_limit),
808 FIELD(GUEST_SS_LIMIT, guest_ss_limit),
809 FIELD(GUEST_DS_LIMIT, guest_ds_limit),
810 FIELD(GUEST_FS_LIMIT, guest_fs_limit),
811 FIELD(GUEST_GS_LIMIT, guest_gs_limit),
812 FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
813 FIELD(GUEST_TR_LIMIT, guest_tr_limit),
814 FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
815 FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
816 FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
817 FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
818 FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
819 FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
820 FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
821 FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
822 FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
823 FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
824 FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
825 FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
826 FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
827 FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
828 FIELD(VMX_PREEMPTION_TIMER_VALUE, vmx_preemption_timer_value),
829 FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
830 FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
831 FIELD(CR0_READ_SHADOW, cr0_read_shadow),
832 FIELD(CR4_READ_SHADOW, cr4_read_shadow),
833 FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
834 FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
835 FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
836 FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
837 FIELD(EXIT_QUALIFICATION, exit_qualification),
838 FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
839 FIELD(GUEST_CR0, guest_cr0),
840 FIELD(GUEST_CR3, guest_cr3),
841 FIELD(GUEST_CR4, guest_cr4),
842 FIELD(GUEST_ES_BASE, guest_es_base),
843 FIELD(GUEST_CS_BASE, guest_cs_base),
844 FIELD(GUEST_SS_BASE, guest_ss_base),
845 FIELD(GUEST_DS_BASE, guest_ds_base),
846 FIELD(GUEST_FS_BASE, guest_fs_base),
847 FIELD(GUEST_GS_BASE, guest_gs_base),
848 FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
849 FIELD(GUEST_TR_BASE, guest_tr_base),
850 FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
851 FIELD(GUEST_IDTR_BASE, guest_idtr_base),
852 FIELD(GUEST_DR7, guest_dr7),
853 FIELD(GUEST_RSP, guest_rsp),
854 FIELD(GUEST_RIP, guest_rip),
855 FIELD(GUEST_RFLAGS, guest_rflags),
856 FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
857 FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
858 FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
859 FIELD(HOST_CR0, host_cr0),
860 FIELD(HOST_CR3, host_cr3),
861 FIELD(HOST_CR4, host_cr4),
862 FIELD(HOST_FS_BASE, host_fs_base),
863 FIELD(HOST_GS_BASE, host_gs_base),
864 FIELD(HOST_TR_BASE, host_tr_base),
865 FIELD(HOST_GDTR_BASE, host_gdtr_base),
866 FIELD(HOST_IDTR_BASE, host_idtr_base),
867 FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
868 FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
869 FIELD(HOST_RSP, host_rsp),
870 FIELD(HOST_RIP, host_rip),
873 static inline short vmcs_field_to_offset(unsigned long field)
875 BUILD_BUG_ON(ARRAY_SIZE(vmcs_field_to_offset_table) > SHRT_MAX);
877 if (field >= ARRAY_SIZE(vmcs_field_to_offset_table) ||
878 vmcs_field_to_offset_table[field] == 0)
881 return vmcs_field_to_offset_table[field];
884 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
886 return to_vmx(vcpu)->nested.cached_vmcs12;
889 static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
891 struct page *page = kvm_vcpu_gfn_to_page(vcpu, addr >> PAGE_SHIFT);
892 if (is_error_page(page))
898 static void nested_release_page(struct page *page)
900 kvm_release_page_dirty(page);
903 static void nested_release_page_clean(struct page *page)
905 kvm_release_page_clean(page);
908 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu);
909 static u64 construct_eptp(unsigned long root_hpa);
910 static bool vmx_xsaves_supported(void);
911 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
912 static void vmx_set_segment(struct kvm_vcpu *vcpu,
913 struct kvm_segment *var, int seg);
914 static void vmx_get_segment(struct kvm_vcpu *vcpu,
915 struct kvm_segment *var, int seg);
916 static bool guest_state_valid(struct kvm_vcpu *vcpu);
917 static u32 vmx_segment_access_rights(struct kvm_segment *var);
918 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx);
919 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx);
920 static int alloc_identity_pagetable(struct kvm *kvm);
922 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
923 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
925 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
926 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
928 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
929 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
932 * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
933 * can find which vCPU should be waken up.
935 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
936 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
941 VMX_MSR_BITMAP_LEGACY,
942 VMX_MSR_BITMAP_LONGMODE,
943 VMX_MSR_BITMAP_LEGACY_X2APIC_APICV,
944 VMX_MSR_BITMAP_LONGMODE_X2APIC_APICV,
945 VMX_MSR_BITMAP_LEGACY_X2APIC,
946 VMX_MSR_BITMAP_LONGMODE_X2APIC,
952 static unsigned long *vmx_bitmap[VMX_BITMAP_NR];
954 #define vmx_io_bitmap_a (vmx_bitmap[VMX_IO_BITMAP_A])
955 #define vmx_io_bitmap_b (vmx_bitmap[VMX_IO_BITMAP_B])
956 #define vmx_msr_bitmap_legacy (vmx_bitmap[VMX_MSR_BITMAP_LEGACY])
957 #define vmx_msr_bitmap_longmode (vmx_bitmap[VMX_MSR_BITMAP_LONGMODE])
958 #define vmx_msr_bitmap_legacy_x2apic_apicv (vmx_bitmap[VMX_MSR_BITMAP_LEGACY_X2APIC_APICV])
959 #define vmx_msr_bitmap_longmode_x2apic_apicv (vmx_bitmap[VMX_MSR_BITMAP_LONGMODE_X2APIC_APICV])
960 #define vmx_msr_bitmap_legacy_x2apic (vmx_bitmap[VMX_MSR_BITMAP_LEGACY_X2APIC])
961 #define vmx_msr_bitmap_longmode_x2apic (vmx_bitmap[VMX_MSR_BITMAP_LONGMODE_X2APIC])
962 #define vmx_vmread_bitmap (vmx_bitmap[VMX_VMREAD_BITMAP])
963 #define vmx_vmwrite_bitmap (vmx_bitmap[VMX_VMWRITE_BITMAP])
965 static bool cpu_has_load_ia32_efer;
966 static bool cpu_has_load_perf_global_ctrl;
968 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
969 static DEFINE_SPINLOCK(vmx_vpid_lock);
971 static struct vmcs_config {
976 u32 pin_based_exec_ctrl;
977 u32 cpu_based_exec_ctrl;
978 u32 cpu_based_2nd_exec_ctrl;
983 static struct vmx_capability {
988 #define VMX_SEGMENT_FIELD(seg) \
989 [VCPU_SREG_##seg] = { \
990 .selector = GUEST_##seg##_SELECTOR, \
991 .base = GUEST_##seg##_BASE, \
992 .limit = GUEST_##seg##_LIMIT, \
993 .ar_bytes = GUEST_##seg##_AR_BYTES, \
996 static const struct kvm_vmx_segment_field {
1001 } kvm_vmx_segment_fields[] = {
1002 VMX_SEGMENT_FIELD(CS),
1003 VMX_SEGMENT_FIELD(DS),
1004 VMX_SEGMENT_FIELD(ES),
1005 VMX_SEGMENT_FIELD(FS),
1006 VMX_SEGMENT_FIELD(GS),
1007 VMX_SEGMENT_FIELD(SS),
1008 VMX_SEGMENT_FIELD(TR),
1009 VMX_SEGMENT_FIELD(LDTR),
1012 static u64 host_efer;
1014 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
1017 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
1018 * away by decrementing the array size.
1020 static const u32 vmx_msr_index[] = {
1021 #ifdef CONFIG_X86_64
1022 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
1024 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
1027 static inline bool is_exception_n(u32 intr_info, u8 vector)
1029 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
1030 INTR_INFO_VALID_MASK)) ==
1031 (INTR_TYPE_HARD_EXCEPTION | vector | INTR_INFO_VALID_MASK);
1034 static inline bool is_debug(u32 intr_info)
1036 return is_exception_n(intr_info, DB_VECTOR);
1039 static inline bool is_breakpoint(u32 intr_info)
1041 return is_exception_n(intr_info, BP_VECTOR);
1044 static inline bool is_page_fault(u32 intr_info)
1046 return is_exception_n(intr_info, PF_VECTOR);
1049 static inline bool is_no_device(u32 intr_info)
1051 return is_exception_n(intr_info, NM_VECTOR);
1054 static inline bool is_invalid_opcode(u32 intr_info)
1056 return is_exception_n(intr_info, UD_VECTOR);
1059 static inline bool is_external_interrupt(u32 intr_info)
1061 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1062 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
1065 static inline bool is_machine_check(u32 intr_info)
1067 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
1068 INTR_INFO_VALID_MASK)) ==
1069 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
1072 static inline bool cpu_has_vmx_msr_bitmap(void)
1074 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
1077 static inline bool cpu_has_vmx_tpr_shadow(void)
1079 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
1082 static inline bool cpu_need_tpr_shadow(struct kvm_vcpu *vcpu)
1084 return cpu_has_vmx_tpr_shadow() && lapic_in_kernel(vcpu);
1087 static inline bool cpu_has_secondary_exec_ctrls(void)
1089 return vmcs_config.cpu_based_exec_ctrl &
1090 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1093 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
1095 return vmcs_config.cpu_based_2nd_exec_ctrl &
1096 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1099 static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
1101 return vmcs_config.cpu_based_2nd_exec_ctrl &
1102 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
1105 static inline bool cpu_has_vmx_apic_register_virt(void)
1107 return vmcs_config.cpu_based_2nd_exec_ctrl &
1108 SECONDARY_EXEC_APIC_REGISTER_VIRT;
1111 static inline bool cpu_has_vmx_virtual_intr_delivery(void)
1113 return vmcs_config.cpu_based_2nd_exec_ctrl &
1114 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
1118 * Comment's format: document - errata name - stepping - processor name.
1120 * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
1122 static u32 vmx_preemption_cpu_tfms[] = {
1123 /* 323344.pdf - BA86 - D0 - Xeon 7500 Series */
1125 /* 323056.pdf - AAX65 - C2 - Xeon L3406 */
1126 /* 322814.pdf - AAT59 - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
1127 /* 322911.pdf - AAU65 - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
1129 /* 322911.pdf - AAU65 - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
1131 /* 322373.pdf - AAO95 - B1 - Xeon 3400 Series */
1132 /* 322166.pdf - AAN92 - B1 - i7-800 and i5-700 Desktop */
1134 * 320767.pdf - AAP86 - B1 -
1135 * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
1138 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
1140 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
1142 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
1144 /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
1145 /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
1146 /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
1150 static inline bool cpu_has_broken_vmx_preemption_timer(void)
1152 u32 eax = cpuid_eax(0x00000001), i;
1154 /* Clear the reserved bits */
1155 eax &= ~(0x3U << 14 | 0xfU << 28);
1156 for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
1157 if (eax == vmx_preemption_cpu_tfms[i])
1163 static inline bool cpu_has_vmx_preemption_timer(void)
1165 return vmcs_config.pin_based_exec_ctrl &
1166 PIN_BASED_VMX_PREEMPTION_TIMER;
1169 static inline bool cpu_has_vmx_posted_intr(void)
1171 return IS_ENABLED(CONFIG_X86_LOCAL_APIC) &&
1172 vmcs_config.pin_based_exec_ctrl & PIN_BASED_POSTED_INTR;
1175 static inline bool cpu_has_vmx_apicv(void)
1177 return cpu_has_vmx_apic_register_virt() &&
1178 cpu_has_vmx_virtual_intr_delivery() &&
1179 cpu_has_vmx_posted_intr();
1182 static inline bool cpu_has_vmx_flexpriority(void)
1184 return cpu_has_vmx_tpr_shadow() &&
1185 cpu_has_vmx_virtualize_apic_accesses();
1188 static inline bool cpu_has_vmx_ept_execute_only(void)
1190 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
1193 static inline bool cpu_has_vmx_ept_2m_page(void)
1195 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
1198 static inline bool cpu_has_vmx_ept_1g_page(void)
1200 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
1203 static inline bool cpu_has_vmx_ept_4levels(void)
1205 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
1208 static inline bool cpu_has_vmx_ept_ad_bits(void)
1210 return vmx_capability.ept & VMX_EPT_AD_BIT;
1213 static inline bool cpu_has_vmx_invept_context(void)
1215 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
1218 static inline bool cpu_has_vmx_invept_global(void)
1220 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
1223 static inline bool cpu_has_vmx_invvpid_single(void)
1225 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
1228 static inline bool cpu_has_vmx_invvpid_global(void)
1230 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
1233 static inline bool cpu_has_vmx_invvpid(void)
1235 return vmx_capability.vpid & VMX_VPID_INVVPID_BIT;
1238 static inline bool cpu_has_vmx_ept(void)
1240 return vmcs_config.cpu_based_2nd_exec_ctrl &
1241 SECONDARY_EXEC_ENABLE_EPT;
1244 static inline bool cpu_has_vmx_unrestricted_guest(void)
1246 return vmcs_config.cpu_based_2nd_exec_ctrl &
1247 SECONDARY_EXEC_UNRESTRICTED_GUEST;
1250 static inline bool cpu_has_vmx_ple(void)
1252 return vmcs_config.cpu_based_2nd_exec_ctrl &
1253 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
1256 static inline bool cpu_has_vmx_basic_inout(void)
1258 return (((u64)vmcs_config.basic_cap << 32) & VMX_BASIC_INOUT);
1261 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
1263 return flexpriority_enabled && lapic_in_kernel(vcpu);
1266 static inline bool cpu_has_vmx_vpid(void)
1268 return vmcs_config.cpu_based_2nd_exec_ctrl &
1269 SECONDARY_EXEC_ENABLE_VPID;
1272 static inline bool cpu_has_vmx_rdtscp(void)
1274 return vmcs_config.cpu_based_2nd_exec_ctrl &
1275 SECONDARY_EXEC_RDTSCP;
1278 static inline bool cpu_has_vmx_invpcid(void)
1280 return vmcs_config.cpu_based_2nd_exec_ctrl &
1281 SECONDARY_EXEC_ENABLE_INVPCID;
1284 static inline bool cpu_has_vmx_wbinvd_exit(void)
1286 return vmcs_config.cpu_based_2nd_exec_ctrl &
1287 SECONDARY_EXEC_WBINVD_EXITING;
1290 static inline bool cpu_has_vmx_shadow_vmcs(void)
1293 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
1294 /* check if the cpu supports writing r/o exit information fields */
1295 if (!(vmx_msr & MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS))
1298 return vmcs_config.cpu_based_2nd_exec_ctrl &
1299 SECONDARY_EXEC_SHADOW_VMCS;
1302 static inline bool cpu_has_vmx_pml(void)
1304 return vmcs_config.cpu_based_2nd_exec_ctrl & SECONDARY_EXEC_ENABLE_PML;
1307 static inline bool cpu_has_vmx_tsc_scaling(void)
1309 return vmcs_config.cpu_based_2nd_exec_ctrl &
1310 SECONDARY_EXEC_TSC_SCALING;
1313 static inline bool report_flexpriority(void)
1315 return flexpriority_enabled;
1318 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
1320 return vmcs12->cpu_based_vm_exec_control & bit;
1323 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
1325 return (vmcs12->cpu_based_vm_exec_control &
1326 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
1327 (vmcs12->secondary_vm_exec_control & bit);
1330 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12)
1332 return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
1335 static inline bool nested_cpu_has_preemption_timer(struct vmcs12 *vmcs12)
1337 return vmcs12->pin_based_vm_exec_control &
1338 PIN_BASED_VMX_PREEMPTION_TIMER;
1341 static inline int nested_cpu_has_ept(struct vmcs12 *vmcs12)
1343 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_EPT);
1346 static inline bool nested_cpu_has_xsaves(struct vmcs12 *vmcs12)
1348 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES) &&
1349 vmx_xsaves_supported();
1352 static inline bool nested_cpu_has_virt_x2apic_mode(struct vmcs12 *vmcs12)
1354 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
1357 static inline bool nested_cpu_has_vpid(struct vmcs12 *vmcs12)
1359 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_VPID);
1362 static inline bool nested_cpu_has_apic_reg_virt(struct vmcs12 *vmcs12)
1364 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_APIC_REGISTER_VIRT);
1367 static inline bool nested_cpu_has_vid(struct vmcs12 *vmcs12)
1369 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
1372 static inline bool nested_cpu_has_posted_intr(struct vmcs12 *vmcs12)
1374 return vmcs12->pin_based_vm_exec_control & PIN_BASED_POSTED_INTR;
1377 static inline bool is_nmi(u32 intr_info)
1379 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1380 == (INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK);
1383 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
1385 unsigned long exit_qualification);
1386 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
1387 struct vmcs12 *vmcs12,
1388 u32 reason, unsigned long qualification);
1390 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
1394 for (i = 0; i < vmx->nmsrs; ++i)
1395 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
1400 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
1406 } operand = { vpid, 0, gva };
1408 asm volatile (__ex(ASM_VMX_INVVPID)
1409 /* CF==1 or ZF==1 --> rc = -1 */
1410 "; ja 1f ; ud2 ; 1:"
1411 : : "a"(&operand), "c"(ext) : "cc", "memory");
1414 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
1418 } operand = {eptp, gpa};
1420 asm volatile (__ex(ASM_VMX_INVEPT)
1421 /* CF==1 or ZF==1 --> rc = -1 */
1422 "; ja 1f ; ud2 ; 1:\n"
1423 : : "a" (&operand), "c" (ext) : "cc", "memory");
1426 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
1430 i = __find_msr_index(vmx, msr);
1432 return &vmx->guest_msrs[i];
1436 static void vmcs_clear(struct vmcs *vmcs)
1438 u64 phys_addr = __pa(vmcs);
1441 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
1442 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1445 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
1449 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
1451 vmcs_clear(loaded_vmcs->vmcs);
1452 if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
1453 vmcs_clear(loaded_vmcs->shadow_vmcs);
1454 loaded_vmcs->cpu = -1;
1455 loaded_vmcs->launched = 0;
1458 static void vmcs_load(struct vmcs *vmcs)
1460 u64 phys_addr = __pa(vmcs);
1463 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
1464 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1467 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
1471 #ifdef CONFIG_KEXEC_CORE
1473 * This bitmap is used to indicate whether the vmclear
1474 * operation is enabled on all cpus. All disabled by
1477 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
1479 static inline void crash_enable_local_vmclear(int cpu)
1481 cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
1484 static inline void crash_disable_local_vmclear(int cpu)
1486 cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
1489 static inline int crash_local_vmclear_enabled(int cpu)
1491 return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
1494 static void crash_vmclear_local_loaded_vmcss(void)
1496 int cpu = raw_smp_processor_id();
1497 struct loaded_vmcs *v;
1499 if (!crash_local_vmclear_enabled(cpu))
1502 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
1503 loaded_vmcss_on_cpu_link)
1504 vmcs_clear(v->vmcs);
1507 static inline void crash_enable_local_vmclear(int cpu) { }
1508 static inline void crash_disable_local_vmclear(int cpu) { }
1509 #endif /* CONFIG_KEXEC_CORE */
1511 static void __loaded_vmcs_clear(void *arg)
1513 struct loaded_vmcs *loaded_vmcs = arg;
1514 int cpu = raw_smp_processor_id();
1516 if (loaded_vmcs->cpu != cpu)
1517 return; /* vcpu migration can race with cpu offline */
1518 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1519 per_cpu(current_vmcs, cpu) = NULL;
1520 crash_disable_local_vmclear(cpu);
1521 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1524 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1525 * is before setting loaded_vmcs->vcpu to -1 which is done in
1526 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1527 * then adds the vmcs into percpu list before it is deleted.
1531 loaded_vmcs_init(loaded_vmcs);
1532 crash_enable_local_vmclear(cpu);
1535 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1537 int cpu = loaded_vmcs->cpu;
1540 smp_call_function_single(cpu,
1541 __loaded_vmcs_clear, loaded_vmcs, 1);
1544 static inline void vpid_sync_vcpu_single(int vpid)
1549 if (cpu_has_vmx_invvpid_single())
1550 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vpid, 0);
1553 static inline void vpid_sync_vcpu_global(void)
1555 if (cpu_has_vmx_invvpid_global())
1556 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1559 static inline void vpid_sync_context(int vpid)
1561 if (cpu_has_vmx_invvpid_single())
1562 vpid_sync_vcpu_single(vpid);
1564 vpid_sync_vcpu_global();
1567 static inline void ept_sync_global(void)
1569 if (cpu_has_vmx_invept_global())
1570 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1573 static inline void ept_sync_context(u64 eptp)
1576 if (cpu_has_vmx_invept_context())
1577 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1583 static __always_inline void vmcs_check16(unsigned long field)
1585 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1586 "16-bit accessor invalid for 64-bit field");
1587 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1588 "16-bit accessor invalid for 64-bit high field");
1589 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1590 "16-bit accessor invalid for 32-bit high field");
1591 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1592 "16-bit accessor invalid for natural width field");
1595 static __always_inline void vmcs_check32(unsigned long field)
1597 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1598 "32-bit accessor invalid for 16-bit field");
1599 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1600 "32-bit accessor invalid for natural width field");
1603 static __always_inline void vmcs_check64(unsigned long field)
1605 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1606 "64-bit accessor invalid for 16-bit field");
1607 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1608 "64-bit accessor invalid for 64-bit high field");
1609 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1610 "64-bit accessor invalid for 32-bit field");
1611 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1612 "64-bit accessor invalid for natural width field");
1615 static __always_inline void vmcs_checkl(unsigned long field)
1617 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1618 "Natural width accessor invalid for 16-bit field");
1619 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1620 "Natural width accessor invalid for 64-bit field");
1621 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1622 "Natural width accessor invalid for 64-bit high field");
1623 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1624 "Natural width accessor invalid for 32-bit field");
1627 static __always_inline unsigned long __vmcs_readl(unsigned long field)
1629 unsigned long value;
1631 asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1632 : "=a"(value) : "d"(field) : "cc");
1636 static __always_inline u16 vmcs_read16(unsigned long field)
1638 vmcs_check16(field);
1639 return __vmcs_readl(field);
1642 static __always_inline u32 vmcs_read32(unsigned long field)
1644 vmcs_check32(field);
1645 return __vmcs_readl(field);
1648 static __always_inline u64 vmcs_read64(unsigned long field)
1650 vmcs_check64(field);
1651 #ifdef CONFIG_X86_64
1652 return __vmcs_readl(field);
1654 return __vmcs_readl(field) | ((u64)__vmcs_readl(field+1) << 32);
1658 static __always_inline unsigned long vmcs_readl(unsigned long field)
1661 return __vmcs_readl(field);
1664 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1666 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1667 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1671 static __always_inline void __vmcs_writel(unsigned long field, unsigned long value)
1675 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1676 : "=q"(error) : "a"(value), "d"(field) : "cc");
1677 if (unlikely(error))
1678 vmwrite_error(field, value);
1681 static __always_inline void vmcs_write16(unsigned long field, u16 value)
1683 vmcs_check16(field);
1684 __vmcs_writel(field, value);
1687 static __always_inline void vmcs_write32(unsigned long field, u32 value)
1689 vmcs_check32(field);
1690 __vmcs_writel(field, value);
1693 static __always_inline void vmcs_write64(unsigned long field, u64 value)
1695 vmcs_check64(field);
1696 __vmcs_writel(field, value);
1697 #ifndef CONFIG_X86_64
1699 __vmcs_writel(field+1, value >> 32);
1703 static __always_inline void vmcs_writel(unsigned long field, unsigned long value)
1706 __vmcs_writel(field, value);
1709 static __always_inline void vmcs_clear_bits(unsigned long field, u32 mask)
1711 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1712 "vmcs_clear_bits does not support 64-bit fields");
1713 __vmcs_writel(field, __vmcs_readl(field) & ~mask);
1716 static __always_inline void vmcs_set_bits(unsigned long field, u32 mask)
1718 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1719 "vmcs_set_bits does not support 64-bit fields");
1720 __vmcs_writel(field, __vmcs_readl(field) | mask);
1723 static inline void vm_entry_controls_reset_shadow(struct vcpu_vmx *vmx)
1725 vmx->vm_entry_controls_shadow = vmcs_read32(VM_ENTRY_CONTROLS);
1728 static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val)
1730 vmcs_write32(VM_ENTRY_CONTROLS, val);
1731 vmx->vm_entry_controls_shadow = val;
1734 static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val)
1736 if (vmx->vm_entry_controls_shadow != val)
1737 vm_entry_controls_init(vmx, val);
1740 static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx)
1742 return vmx->vm_entry_controls_shadow;
1746 static inline void vm_entry_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1748 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) | val);
1751 static inline void vm_entry_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1753 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) & ~val);
1756 static inline void vm_exit_controls_reset_shadow(struct vcpu_vmx *vmx)
1758 vmx->vm_exit_controls_shadow = vmcs_read32(VM_EXIT_CONTROLS);
1761 static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val)
1763 vmcs_write32(VM_EXIT_CONTROLS, val);
1764 vmx->vm_exit_controls_shadow = val;
1767 static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val)
1769 if (vmx->vm_exit_controls_shadow != val)
1770 vm_exit_controls_init(vmx, val);
1773 static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx)
1775 return vmx->vm_exit_controls_shadow;
1779 static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1781 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val);
1784 static inline void vm_exit_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1786 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) & ~val);
1789 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1791 vmx->segment_cache.bitmask = 0;
1794 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1798 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1800 if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1801 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1802 vmx->segment_cache.bitmask = 0;
1804 ret = vmx->segment_cache.bitmask & mask;
1805 vmx->segment_cache.bitmask |= mask;
1809 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1811 u16 *p = &vmx->segment_cache.seg[seg].selector;
1813 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1814 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1818 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1820 ulong *p = &vmx->segment_cache.seg[seg].base;
1822 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1823 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1827 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1829 u32 *p = &vmx->segment_cache.seg[seg].limit;
1831 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1832 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1836 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1838 u32 *p = &vmx->segment_cache.seg[seg].ar;
1840 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1841 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1845 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1849 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1850 (1u << DB_VECTOR) | (1u << AC_VECTOR);
1851 if ((vcpu->guest_debug &
1852 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1853 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1854 eb |= 1u << BP_VECTOR;
1855 if (to_vmx(vcpu)->rmode.vm86_active)
1858 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1860 /* When we are running a nested L2 guest and L1 specified for it a
1861 * certain exception bitmap, we must trap the same exceptions and pass
1862 * them to L1. When running L2, we will only handle the exceptions
1863 * specified above if L1 did not want them.
1865 if (is_guest_mode(vcpu))
1866 eb |= get_vmcs12(vcpu)->exception_bitmap;
1868 vmcs_write32(EXCEPTION_BITMAP, eb);
1871 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1872 unsigned long entry, unsigned long exit)
1874 vm_entry_controls_clearbit(vmx, entry);
1875 vm_exit_controls_clearbit(vmx, exit);
1878 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1881 struct msr_autoload *m = &vmx->msr_autoload;
1885 if (cpu_has_load_ia32_efer) {
1886 clear_atomic_switch_msr_special(vmx,
1887 VM_ENTRY_LOAD_IA32_EFER,
1888 VM_EXIT_LOAD_IA32_EFER);
1892 case MSR_CORE_PERF_GLOBAL_CTRL:
1893 if (cpu_has_load_perf_global_ctrl) {
1894 clear_atomic_switch_msr_special(vmx,
1895 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1896 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1902 for (i = 0; i < m->nr; ++i)
1903 if (m->guest[i].index == msr)
1909 m->guest[i] = m->guest[m->nr];
1910 m->host[i] = m->host[m->nr];
1911 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1912 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1915 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1916 unsigned long entry, unsigned long exit,
1917 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
1918 u64 guest_val, u64 host_val)
1920 vmcs_write64(guest_val_vmcs, guest_val);
1921 vmcs_write64(host_val_vmcs, host_val);
1922 vm_entry_controls_setbit(vmx, entry);
1923 vm_exit_controls_setbit(vmx, exit);
1926 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1927 u64 guest_val, u64 host_val)
1930 struct msr_autoload *m = &vmx->msr_autoload;
1934 if (cpu_has_load_ia32_efer) {
1935 add_atomic_switch_msr_special(vmx,
1936 VM_ENTRY_LOAD_IA32_EFER,
1937 VM_EXIT_LOAD_IA32_EFER,
1940 guest_val, host_val);
1944 case MSR_CORE_PERF_GLOBAL_CTRL:
1945 if (cpu_has_load_perf_global_ctrl) {
1946 add_atomic_switch_msr_special(vmx,
1947 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1948 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1949 GUEST_IA32_PERF_GLOBAL_CTRL,
1950 HOST_IA32_PERF_GLOBAL_CTRL,
1951 guest_val, host_val);
1955 case MSR_IA32_PEBS_ENABLE:
1956 /* PEBS needs a quiescent period after being disabled (to write
1957 * a record). Disabling PEBS through VMX MSR swapping doesn't
1958 * provide that period, so a CPU could write host's record into
1961 wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
1964 for (i = 0; i < m->nr; ++i)
1965 if (m->guest[i].index == msr)
1968 if (i == NR_AUTOLOAD_MSRS) {
1969 printk_once(KERN_WARNING "Not enough msr switch entries. "
1970 "Can't add msr %x\n", msr);
1972 } else if (i == m->nr) {
1974 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1975 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1978 m->guest[i].index = msr;
1979 m->guest[i].value = guest_val;
1980 m->host[i].index = msr;
1981 m->host[i].value = host_val;
1984 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
1986 u64 guest_efer = vmx->vcpu.arch.efer;
1987 u64 ignore_bits = 0;
1991 * NX is needed to handle CR0.WP=1, CR4.SMEP=1. Testing
1992 * host CPUID is more efficient than testing guest CPUID
1993 * or CR4. Host SMEP is anyway a requirement for guest SMEP.
1995 if (boot_cpu_has(X86_FEATURE_SMEP))
1996 guest_efer |= EFER_NX;
1997 else if (!(guest_efer & EFER_NX))
1998 ignore_bits |= EFER_NX;
2002 * LMA and LME handled by hardware; SCE meaningless outside long mode.
2004 ignore_bits |= EFER_SCE;
2005 #ifdef CONFIG_X86_64
2006 ignore_bits |= EFER_LMA | EFER_LME;
2007 /* SCE is meaningful only in long mode on Intel */
2008 if (guest_efer & EFER_LMA)
2009 ignore_bits &= ~(u64)EFER_SCE;
2012 clear_atomic_switch_msr(vmx, MSR_EFER);
2015 * On EPT, we can't emulate NX, so we must switch EFER atomically.
2016 * On CPUs that support "load IA32_EFER", always switch EFER
2017 * atomically, since it's faster than switching it manually.
2019 if (cpu_has_load_ia32_efer ||
2020 (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
2021 if (!(guest_efer & EFER_LMA))
2022 guest_efer &= ~EFER_LME;
2023 if (guest_efer != host_efer)
2024 add_atomic_switch_msr(vmx, MSR_EFER,
2025 guest_efer, host_efer);
2028 guest_efer &= ~ignore_bits;
2029 guest_efer |= host_efer & ignore_bits;
2031 vmx->guest_msrs[efer_offset].data = guest_efer;
2032 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
2038 #ifdef CONFIG_X86_32
2040 * On 32-bit kernels, VM exits still load the FS and GS bases from the
2041 * VMCS rather than the segment table. KVM uses this helper to figure
2042 * out the current bases to poke them into the VMCS before entry.
2044 static unsigned long segment_base(u16 selector)
2046 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
2047 struct desc_struct *table;
2050 if (!(selector & ~SEGMENT_RPL_MASK))
2053 table = (struct desc_struct *)gdt->address;
2055 if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
2056 u16 ldt_selector = kvm_read_ldt();
2058 if (!(ldt_selector & ~SEGMENT_RPL_MASK))
2061 table = (struct desc_struct *)segment_base(ldt_selector);
2063 v = get_desc_base(&table[selector >> 3]);
2068 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
2070 struct vcpu_vmx *vmx = to_vmx(vcpu);
2073 if (vmx->host_state.loaded)
2076 vmx->host_state.loaded = 1;
2078 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
2079 * allow segment selectors with cpl > 0 or ti == 1.
2081 vmx->host_state.ldt_sel = kvm_read_ldt();
2082 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
2083 savesegment(fs, vmx->host_state.fs_sel);
2084 if (!(vmx->host_state.fs_sel & 7)) {
2085 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
2086 vmx->host_state.fs_reload_needed = 0;
2088 vmcs_write16(HOST_FS_SELECTOR, 0);
2089 vmx->host_state.fs_reload_needed = 1;
2091 savesegment(gs, vmx->host_state.gs_sel);
2092 if (!(vmx->host_state.gs_sel & 7))
2093 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
2095 vmcs_write16(HOST_GS_SELECTOR, 0);
2096 vmx->host_state.gs_ldt_reload_needed = 1;
2099 #ifdef CONFIG_X86_64
2100 savesegment(ds, vmx->host_state.ds_sel);
2101 savesegment(es, vmx->host_state.es_sel);
2104 #ifdef CONFIG_X86_64
2105 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
2106 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
2108 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
2109 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
2112 #ifdef CONFIG_X86_64
2113 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2114 if (is_long_mode(&vmx->vcpu))
2115 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2117 if (boot_cpu_has(X86_FEATURE_MPX))
2118 rdmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
2119 for (i = 0; i < vmx->save_nmsrs; ++i)
2120 kvm_set_shared_msr(vmx->guest_msrs[i].index,
2121 vmx->guest_msrs[i].data,
2122 vmx->guest_msrs[i].mask);
2125 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
2127 if (!vmx->host_state.loaded)
2130 ++vmx->vcpu.stat.host_state_reload;
2131 vmx->host_state.loaded = 0;
2132 #ifdef CONFIG_X86_64
2133 if (is_long_mode(&vmx->vcpu))
2134 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2136 if (vmx->host_state.gs_ldt_reload_needed) {
2137 kvm_load_ldt(vmx->host_state.ldt_sel);
2138 #ifdef CONFIG_X86_64
2139 load_gs_index(vmx->host_state.gs_sel);
2141 loadsegment(gs, vmx->host_state.gs_sel);
2144 if (vmx->host_state.fs_reload_needed)
2145 loadsegment(fs, vmx->host_state.fs_sel);
2146 #ifdef CONFIG_X86_64
2147 if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
2148 loadsegment(ds, vmx->host_state.ds_sel);
2149 loadsegment(es, vmx->host_state.es_sel);
2152 invalidate_tss_limit();
2153 #ifdef CONFIG_X86_64
2154 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2156 if (vmx->host_state.msr_host_bndcfgs)
2157 wrmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
2158 load_gdt(this_cpu_ptr(&host_gdt));
2161 static void vmx_load_host_state(struct vcpu_vmx *vmx)
2164 __vmx_load_host_state(vmx);
2168 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
2170 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2171 struct pi_desc old, new;
2174 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
2175 !irq_remapping_cap(IRQ_POSTING_CAP) ||
2176 !kvm_vcpu_apicv_active(vcpu))
2180 old.control = new.control = pi_desc->control;
2183 * If 'nv' field is POSTED_INTR_WAKEUP_VECTOR, there
2184 * are two possible cases:
2185 * 1. After running 'pre_block', context switch
2186 * happened. For this case, 'sn' was set in
2187 * vmx_vcpu_put(), so we need to clear it here.
2188 * 2. After running 'pre_block', we were blocked,
2189 * and woken up by some other guy. For this case,
2190 * we don't need to do anything, 'pi_post_block'
2191 * will do everything for us. However, we cannot
2192 * check whether it is case #1 or case #2 here
2193 * (maybe, not needed), so we also clear sn here,
2194 * I think it is not a big deal.
2196 if (pi_desc->nv != POSTED_INTR_WAKEUP_VECTOR) {
2197 if (vcpu->cpu != cpu) {
2198 dest = cpu_physical_id(cpu);
2200 if (x2apic_enabled())
2203 new.ndst = (dest << 8) & 0xFF00;
2206 /* set 'NV' to 'notification vector' */
2207 new.nv = POSTED_INTR_VECTOR;
2210 /* Allow posting non-urgent interrupts */
2212 } while (cmpxchg(&pi_desc->control, old.control,
2213 new.control) != old.control);
2216 static void decache_tsc_multiplier(struct vcpu_vmx *vmx)
2218 vmx->current_tsc_ratio = vmx->vcpu.arch.tsc_scaling_ratio;
2219 vmcs_write64(TSC_MULTIPLIER, vmx->current_tsc_ratio);
2223 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
2224 * vcpu mutex is already taken.
2226 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2228 struct vcpu_vmx *vmx = to_vmx(vcpu);
2229 bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
2231 if (!already_loaded) {
2232 loaded_vmcs_clear(vmx->loaded_vmcs);
2233 local_irq_disable();
2234 crash_disable_local_vmclear(cpu);
2237 * Read loaded_vmcs->cpu should be before fetching
2238 * loaded_vmcs->loaded_vmcss_on_cpu_link.
2239 * See the comments in __loaded_vmcs_clear().
2243 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
2244 &per_cpu(loaded_vmcss_on_cpu, cpu));
2245 crash_enable_local_vmclear(cpu);
2249 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
2250 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
2251 vmcs_load(vmx->loaded_vmcs->vmcs);
2254 if (!already_loaded) {
2255 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
2256 unsigned long sysenter_esp;
2258 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2261 * Linux uses per-cpu TSS and GDT, so set these when switching
2262 * processors. See 22.2.4.
2264 vmcs_writel(HOST_TR_BASE,
2265 (unsigned long)this_cpu_ptr(&cpu_tss));
2266 vmcs_writel(HOST_GDTR_BASE, gdt->address);
2269 * VM exits change the host TR limit to 0x67 after a VM
2270 * exit. This is okay, since 0x67 covers everything except
2271 * the IO bitmap and have have code to handle the IO bitmap
2272 * being lost after a VM exit.
2274 BUILD_BUG_ON(IO_BITMAP_OFFSET - 1 != 0x67);
2276 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
2277 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
2279 vmx->loaded_vmcs->cpu = cpu;
2282 /* Setup TSC multiplier */
2283 if (kvm_has_tsc_control &&
2284 vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
2285 decache_tsc_multiplier(vmx);
2287 vmx_vcpu_pi_load(vcpu, cpu);
2288 vmx->host_pkru = read_pkru();
2291 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
2293 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2295 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
2296 !irq_remapping_cap(IRQ_POSTING_CAP) ||
2297 !kvm_vcpu_apicv_active(vcpu))
2300 /* Set SN when the vCPU is preempted */
2301 if (vcpu->preempted)
2305 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
2307 vmx_vcpu_pi_put(vcpu);
2309 __vmx_load_host_state(to_vmx(vcpu));
2312 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
2315 * Return the cr0 value that a nested guest would read. This is a combination
2316 * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
2317 * its hypervisor (cr0_read_shadow).
2319 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
2321 return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
2322 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
2324 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
2326 return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
2327 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
2330 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
2332 unsigned long rflags, save_rflags;
2334 if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
2335 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2336 rflags = vmcs_readl(GUEST_RFLAGS);
2337 if (to_vmx(vcpu)->rmode.vm86_active) {
2338 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2339 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
2340 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2342 to_vmx(vcpu)->rflags = rflags;
2344 return to_vmx(vcpu)->rflags;
2347 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
2349 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2350 to_vmx(vcpu)->rflags = rflags;
2351 if (to_vmx(vcpu)->rmode.vm86_active) {
2352 to_vmx(vcpu)->rmode.save_rflags = rflags;
2353 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2355 vmcs_writel(GUEST_RFLAGS, rflags);
2358 static u32 vmx_get_pkru(struct kvm_vcpu *vcpu)
2360 return to_vmx(vcpu)->guest_pkru;
2363 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
2365 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2368 if (interruptibility & GUEST_INTR_STATE_STI)
2369 ret |= KVM_X86_SHADOW_INT_STI;
2370 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
2371 ret |= KVM_X86_SHADOW_INT_MOV_SS;
2376 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
2378 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2379 u32 interruptibility = interruptibility_old;
2381 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
2383 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
2384 interruptibility |= GUEST_INTR_STATE_MOV_SS;
2385 else if (mask & KVM_X86_SHADOW_INT_STI)
2386 interruptibility |= GUEST_INTR_STATE_STI;
2388 if ((interruptibility != interruptibility_old))
2389 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
2392 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
2396 rip = kvm_rip_read(vcpu);
2397 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
2398 kvm_rip_write(vcpu, rip);
2400 /* skipping an emulated instruction also counts */
2401 vmx_set_interrupt_shadow(vcpu, 0);
2405 * KVM wants to inject page-faults which it got to the guest. This function
2406 * checks whether in a nested guest, we need to inject them to L1 or L2.
2408 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned nr)
2410 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2412 if (!(vmcs12->exception_bitmap & (1u << nr)))
2415 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
2416 vmcs_read32(VM_EXIT_INTR_INFO),
2417 vmcs_readl(EXIT_QUALIFICATION));
2421 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
2422 bool has_error_code, u32 error_code,
2425 struct vcpu_vmx *vmx = to_vmx(vcpu);
2426 u32 intr_info = nr | INTR_INFO_VALID_MASK;
2428 if (!reinject && is_guest_mode(vcpu) &&
2429 nested_vmx_check_exception(vcpu, nr))
2432 if (has_error_code) {
2433 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
2434 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2437 if (vmx->rmode.vm86_active) {
2439 if (kvm_exception_is_soft(nr))
2440 inc_eip = vcpu->arch.event_exit_inst_len;
2441 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
2442 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2446 if (kvm_exception_is_soft(nr)) {
2447 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2448 vmx->vcpu.arch.event_exit_inst_len);
2449 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2451 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2453 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
2456 static bool vmx_rdtscp_supported(void)
2458 return cpu_has_vmx_rdtscp();
2461 static bool vmx_invpcid_supported(void)
2463 return cpu_has_vmx_invpcid() && enable_ept;
2467 * Swap MSR entry in host/guest MSR entry array.
2469 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
2471 struct shared_msr_entry tmp;
2473 tmp = vmx->guest_msrs[to];
2474 vmx->guest_msrs[to] = vmx->guest_msrs[from];
2475 vmx->guest_msrs[from] = tmp;
2478 static void vmx_set_msr_bitmap(struct kvm_vcpu *vcpu)
2480 unsigned long *msr_bitmap;
2482 if (is_guest_mode(vcpu))
2483 msr_bitmap = to_vmx(vcpu)->nested.msr_bitmap;
2484 else if (cpu_has_secondary_exec_ctrls() &&
2485 (vmcs_read32(SECONDARY_VM_EXEC_CONTROL) &
2486 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
2487 if (enable_apicv && kvm_vcpu_apicv_active(vcpu)) {
2488 if (is_long_mode(vcpu))
2489 msr_bitmap = vmx_msr_bitmap_longmode_x2apic_apicv;
2491 msr_bitmap = vmx_msr_bitmap_legacy_x2apic_apicv;
2493 if (is_long_mode(vcpu))
2494 msr_bitmap = vmx_msr_bitmap_longmode_x2apic;
2496 msr_bitmap = vmx_msr_bitmap_legacy_x2apic;
2499 if (is_long_mode(vcpu))
2500 msr_bitmap = vmx_msr_bitmap_longmode;
2502 msr_bitmap = vmx_msr_bitmap_legacy;
2505 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
2509 * Set up the vmcs to automatically save and restore system
2510 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
2511 * mode, as fiddling with msrs is very expensive.
2513 static void setup_msrs(struct vcpu_vmx *vmx)
2515 int save_nmsrs, index;
2518 #ifdef CONFIG_X86_64
2519 if (is_long_mode(&vmx->vcpu)) {
2520 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
2522 move_msr_up(vmx, index, save_nmsrs++);
2523 index = __find_msr_index(vmx, MSR_LSTAR);
2525 move_msr_up(vmx, index, save_nmsrs++);
2526 index = __find_msr_index(vmx, MSR_CSTAR);
2528 move_msr_up(vmx, index, save_nmsrs++);
2529 index = __find_msr_index(vmx, MSR_TSC_AUX);
2530 if (index >= 0 && guest_cpuid_has_rdtscp(&vmx->vcpu))
2531 move_msr_up(vmx, index, save_nmsrs++);
2533 * MSR_STAR is only needed on long mode guests, and only
2534 * if efer.sce is enabled.
2536 index = __find_msr_index(vmx, MSR_STAR);
2537 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
2538 move_msr_up(vmx, index, save_nmsrs++);
2541 index = __find_msr_index(vmx, MSR_EFER);
2542 if (index >= 0 && update_transition_efer(vmx, index))
2543 move_msr_up(vmx, index, save_nmsrs++);
2545 vmx->save_nmsrs = save_nmsrs;
2547 if (cpu_has_vmx_msr_bitmap())
2548 vmx_set_msr_bitmap(&vmx->vcpu);
2552 * reads and returns guest's timestamp counter "register"
2553 * guest_tsc = (host_tsc * tsc multiplier) >> 48 + tsc_offset
2554 * -- Intel TSC Scaling for Virtualization White Paper, sec 1.3
2556 static u64 guest_read_tsc(struct kvm_vcpu *vcpu)
2558 u64 host_tsc, tsc_offset;
2561 tsc_offset = vmcs_read64(TSC_OFFSET);
2562 return kvm_scale_tsc(vcpu, host_tsc) + tsc_offset;
2566 * writes 'offset' into guest's timestamp counter offset register
2568 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2570 if (is_guest_mode(vcpu)) {
2572 * We're here if L1 chose not to trap WRMSR to TSC. According
2573 * to the spec, this should set L1's TSC; The offset that L1
2574 * set for L2 remains unchanged, and still needs to be added
2575 * to the newly set TSC to get L2's TSC.
2577 struct vmcs12 *vmcs12;
2578 /* recalculate vmcs02.TSC_OFFSET: */
2579 vmcs12 = get_vmcs12(vcpu);
2580 vmcs_write64(TSC_OFFSET, offset +
2581 (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
2582 vmcs12->tsc_offset : 0));
2584 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2585 vmcs_read64(TSC_OFFSET), offset);
2586 vmcs_write64(TSC_OFFSET, offset);
2590 static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
2592 struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
2593 return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
2597 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
2598 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
2599 * all guests if the "nested" module option is off, and can also be disabled
2600 * for a single guest by disabling its VMX cpuid bit.
2602 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
2604 return nested && guest_cpuid_has_vmx(vcpu);
2608 * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
2609 * returned for the various VMX controls MSRs when nested VMX is enabled.
2610 * The same values should also be used to verify that vmcs12 control fields are
2611 * valid during nested entry from L1 to L2.
2612 * Each of these control msrs has a low and high 32-bit half: A low bit is on
2613 * if the corresponding bit in the (32-bit) control field *must* be on, and a
2614 * bit in the high half is on if the corresponding bit in the control field
2615 * may be on. See also vmx_control_verify().
2617 static void nested_vmx_setup_ctls_msrs(struct vcpu_vmx *vmx)
2620 * Note that as a general rule, the high half of the MSRs (bits in
2621 * the control fields which may be 1) should be initialized by the
2622 * intersection of the underlying hardware's MSR (i.e., features which
2623 * can be supported) and the list of features we want to expose -
2624 * because they are known to be properly supported in our code.
2625 * Also, usually, the low half of the MSRs (bits which must be 1) can
2626 * be set to 0, meaning that L1 may turn off any of these bits. The
2627 * reason is that if one of these bits is necessary, it will appear
2628 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
2629 * fields of vmcs01 and vmcs02, will turn these bits off - and
2630 * nested_vmx_exit_handled() will not pass related exits to L1.
2631 * These rules have exceptions below.
2634 /* pin-based controls */
2635 rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
2636 vmx->nested.nested_vmx_pinbased_ctls_low,
2637 vmx->nested.nested_vmx_pinbased_ctls_high);
2638 vmx->nested.nested_vmx_pinbased_ctls_low |=
2639 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2640 vmx->nested.nested_vmx_pinbased_ctls_high &=
2641 PIN_BASED_EXT_INTR_MASK |
2642 PIN_BASED_NMI_EXITING |
2643 PIN_BASED_VIRTUAL_NMIS;
2644 vmx->nested.nested_vmx_pinbased_ctls_high |=
2645 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2646 PIN_BASED_VMX_PREEMPTION_TIMER;
2647 if (kvm_vcpu_apicv_active(&vmx->vcpu))
2648 vmx->nested.nested_vmx_pinbased_ctls_high |=
2649 PIN_BASED_POSTED_INTR;
2652 rdmsr(MSR_IA32_VMX_EXIT_CTLS,
2653 vmx->nested.nested_vmx_exit_ctls_low,
2654 vmx->nested.nested_vmx_exit_ctls_high);
2655 vmx->nested.nested_vmx_exit_ctls_low =
2656 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
2658 vmx->nested.nested_vmx_exit_ctls_high &=
2659 #ifdef CONFIG_X86_64
2660 VM_EXIT_HOST_ADDR_SPACE_SIZE |
2662 VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
2663 vmx->nested.nested_vmx_exit_ctls_high |=
2664 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
2665 VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
2666 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
2668 if (kvm_mpx_supported())
2669 vmx->nested.nested_vmx_exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
2671 /* We support free control of debug control saving. */
2672 vmx->nested.nested_vmx_exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
2674 /* entry controls */
2675 rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
2676 vmx->nested.nested_vmx_entry_ctls_low,
2677 vmx->nested.nested_vmx_entry_ctls_high);
2678 vmx->nested.nested_vmx_entry_ctls_low =
2679 VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
2680 vmx->nested.nested_vmx_entry_ctls_high &=
2681 #ifdef CONFIG_X86_64
2682 VM_ENTRY_IA32E_MODE |
2684 VM_ENTRY_LOAD_IA32_PAT;
2685 vmx->nested.nested_vmx_entry_ctls_high |=
2686 (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
2687 if (kvm_mpx_supported())
2688 vmx->nested.nested_vmx_entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
2690 /* We support free control of debug control loading. */
2691 vmx->nested.nested_vmx_entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
2693 /* cpu-based controls */
2694 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
2695 vmx->nested.nested_vmx_procbased_ctls_low,
2696 vmx->nested.nested_vmx_procbased_ctls_high);
2697 vmx->nested.nested_vmx_procbased_ctls_low =
2698 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2699 vmx->nested.nested_vmx_procbased_ctls_high &=
2700 CPU_BASED_VIRTUAL_INTR_PENDING |
2701 CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
2702 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
2703 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
2704 CPU_BASED_CR3_STORE_EXITING |
2705 #ifdef CONFIG_X86_64
2706 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
2708 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
2709 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
2710 CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
2711 CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
2712 CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2714 * We can allow some features even when not supported by the
2715 * hardware. For example, L1 can specify an MSR bitmap - and we
2716 * can use it to avoid exits to L1 - even when L0 runs L2
2717 * without MSR bitmaps.
2719 vmx->nested.nested_vmx_procbased_ctls_high |=
2720 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2721 CPU_BASED_USE_MSR_BITMAPS;
2723 /* We support free control of CR3 access interception. */
2724 vmx->nested.nested_vmx_procbased_ctls_low &=
2725 ~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
2727 /* secondary cpu-based controls */
2728 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2729 vmx->nested.nested_vmx_secondary_ctls_low,
2730 vmx->nested.nested_vmx_secondary_ctls_high);
2731 vmx->nested.nested_vmx_secondary_ctls_low = 0;
2732 vmx->nested.nested_vmx_secondary_ctls_high &=
2733 SECONDARY_EXEC_RDRAND | SECONDARY_EXEC_RDSEED |
2734 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2735 SECONDARY_EXEC_RDTSCP |
2736 SECONDARY_EXEC_DESC |
2737 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2738 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2739 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2740 SECONDARY_EXEC_WBINVD_EXITING |
2741 SECONDARY_EXEC_XSAVES;
2744 /* nested EPT: emulate EPT also to L1 */
2745 vmx->nested.nested_vmx_secondary_ctls_high |=
2746 SECONDARY_EXEC_ENABLE_EPT;
2747 vmx->nested.nested_vmx_ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
2748 VMX_EPTP_WB_BIT | VMX_EPT_INVEPT_BIT;
2749 if (cpu_has_vmx_ept_execute_only())
2750 vmx->nested.nested_vmx_ept_caps |=
2751 VMX_EPT_EXECUTE_ONLY_BIT;
2752 vmx->nested.nested_vmx_ept_caps &= vmx_capability.ept;
2753 vmx->nested.nested_vmx_ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
2754 VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
2755 VMX_EPT_1GB_PAGE_BIT;
2756 if (enable_ept_ad_bits)
2757 vmx->nested.nested_vmx_ept_caps |= VMX_EPT_AD_BIT;
2759 vmx->nested.nested_vmx_ept_caps = 0;
2762 * Old versions of KVM use the single-context version without
2763 * checking for support, so declare that it is supported even
2764 * though it is treated as global context. The alternative is
2765 * not failing the single-context invvpid, and it is worse.
2768 vmx->nested.nested_vmx_secondary_ctls_high |=
2769 SECONDARY_EXEC_ENABLE_VPID;
2770 vmx->nested.nested_vmx_vpid_caps = VMX_VPID_INVVPID_BIT |
2771 VMX_VPID_EXTENT_SUPPORTED_MASK;
2773 vmx->nested.nested_vmx_vpid_caps = 0;
2775 if (enable_unrestricted_guest)
2776 vmx->nested.nested_vmx_secondary_ctls_high |=
2777 SECONDARY_EXEC_UNRESTRICTED_GUEST;
2779 /* miscellaneous data */
2780 rdmsr(MSR_IA32_VMX_MISC,
2781 vmx->nested.nested_vmx_misc_low,
2782 vmx->nested.nested_vmx_misc_high);
2783 vmx->nested.nested_vmx_misc_low &= VMX_MISC_SAVE_EFER_LMA;
2784 vmx->nested.nested_vmx_misc_low |=
2785 VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
2786 VMX_MISC_ACTIVITY_HLT;
2787 vmx->nested.nested_vmx_misc_high = 0;
2790 * This MSR reports some information about VMX support. We
2791 * should return information about the VMX we emulate for the
2792 * guest, and the VMCS structure we give it - not about the
2793 * VMX support of the underlying hardware.
2795 vmx->nested.nested_vmx_basic =
2797 VMX_BASIC_TRUE_CTLS |
2798 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2799 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2801 if (cpu_has_vmx_basic_inout())
2802 vmx->nested.nested_vmx_basic |= VMX_BASIC_INOUT;
2805 * These MSRs specify bits which the guest must keep fixed on
2806 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2807 * We picked the standard core2 setting.
2809 #define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2810 #define VMXON_CR4_ALWAYSON X86_CR4_VMXE
2811 vmx->nested.nested_vmx_cr0_fixed0 = VMXON_CR0_ALWAYSON;
2812 vmx->nested.nested_vmx_cr4_fixed0 = VMXON_CR4_ALWAYSON;
2814 /* These MSRs specify bits which the guest must keep fixed off. */
2815 rdmsrl(MSR_IA32_VMX_CR0_FIXED1, vmx->nested.nested_vmx_cr0_fixed1);
2816 rdmsrl(MSR_IA32_VMX_CR4_FIXED1, vmx->nested.nested_vmx_cr4_fixed1);
2818 /* highest index: VMX_PREEMPTION_TIMER_VALUE */
2819 vmx->nested.nested_vmx_vmcs_enum = 0x2e;
2823 * if fixed0[i] == 1: val[i] must be 1
2824 * if fixed1[i] == 0: val[i] must be 0
2826 static inline bool fixed_bits_valid(u64 val, u64 fixed0, u64 fixed1)
2828 return ((val & fixed1) | fixed0) == val;
2831 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2833 return fixed_bits_valid(control, low, high);
2836 static inline u64 vmx_control_msr(u32 low, u32 high)
2838 return low | ((u64)high << 32);
2841 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
2846 return (superset | subset) == superset;
2849 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
2851 const u64 feature_and_reserved =
2852 /* feature (except bit 48; see below) */
2853 BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
2855 BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
2856 u64 vmx_basic = vmx->nested.nested_vmx_basic;
2858 if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
2862 * KVM does not emulate a version of VMX that constrains physical
2863 * addresses of VMX structures (e.g. VMCS) to 32-bits.
2865 if (data & BIT_ULL(48))
2868 if (vmx_basic_vmcs_revision_id(vmx_basic) !=
2869 vmx_basic_vmcs_revision_id(data))
2872 if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
2875 vmx->nested.nested_vmx_basic = data;
2880 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
2885 switch (msr_index) {
2886 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2887 lowp = &vmx->nested.nested_vmx_pinbased_ctls_low;
2888 highp = &vmx->nested.nested_vmx_pinbased_ctls_high;
2890 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2891 lowp = &vmx->nested.nested_vmx_procbased_ctls_low;
2892 highp = &vmx->nested.nested_vmx_procbased_ctls_high;
2894 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2895 lowp = &vmx->nested.nested_vmx_exit_ctls_low;
2896 highp = &vmx->nested.nested_vmx_exit_ctls_high;
2898 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2899 lowp = &vmx->nested.nested_vmx_entry_ctls_low;
2900 highp = &vmx->nested.nested_vmx_entry_ctls_high;
2902 case MSR_IA32_VMX_PROCBASED_CTLS2:
2903 lowp = &vmx->nested.nested_vmx_secondary_ctls_low;
2904 highp = &vmx->nested.nested_vmx_secondary_ctls_high;
2910 supported = vmx_control_msr(*lowp, *highp);
2912 /* Check must-be-1 bits are still 1. */
2913 if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
2916 /* Check must-be-0 bits are still 0. */
2917 if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
2921 *highp = data >> 32;
2925 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
2927 const u64 feature_and_reserved_bits =
2929 BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
2930 BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
2932 GENMASK_ULL(13, 9) | BIT_ULL(31);
2935 vmx_misc = vmx_control_msr(vmx->nested.nested_vmx_misc_low,
2936 vmx->nested.nested_vmx_misc_high);
2938 if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
2941 if ((vmx->nested.nested_vmx_pinbased_ctls_high &
2942 PIN_BASED_VMX_PREEMPTION_TIMER) &&
2943 vmx_misc_preemption_timer_rate(data) !=
2944 vmx_misc_preemption_timer_rate(vmx_misc))
2947 if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
2950 if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
2953 if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
2956 vmx->nested.nested_vmx_misc_low = data;
2957 vmx->nested.nested_vmx_misc_high = data >> 32;
2961 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
2963 u64 vmx_ept_vpid_cap;
2965 vmx_ept_vpid_cap = vmx_control_msr(vmx->nested.nested_vmx_ept_caps,
2966 vmx->nested.nested_vmx_vpid_caps);
2968 /* Every bit is either reserved or a feature bit. */
2969 if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
2972 vmx->nested.nested_vmx_ept_caps = data;
2973 vmx->nested.nested_vmx_vpid_caps = data >> 32;
2977 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
2981 switch (msr_index) {
2982 case MSR_IA32_VMX_CR0_FIXED0:
2983 msr = &vmx->nested.nested_vmx_cr0_fixed0;
2985 case MSR_IA32_VMX_CR4_FIXED0:
2986 msr = &vmx->nested.nested_vmx_cr4_fixed0;
2993 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
2994 * must be 1 in the restored value.
2996 if (!is_bitwise_subset(data, *msr, -1ULL))
3004 * Called when userspace is restoring VMX MSRs.
3006 * Returns 0 on success, non-0 otherwise.
3008 static int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
3010 struct vcpu_vmx *vmx = to_vmx(vcpu);
3012 switch (msr_index) {
3013 case MSR_IA32_VMX_BASIC:
3014 return vmx_restore_vmx_basic(vmx, data);
3015 case MSR_IA32_VMX_PINBASED_CTLS:
3016 case MSR_IA32_VMX_PROCBASED_CTLS:
3017 case MSR_IA32_VMX_EXIT_CTLS:
3018 case MSR_IA32_VMX_ENTRY_CTLS:
3020 * The "non-true" VMX capability MSRs are generated from the
3021 * "true" MSRs, so we do not support restoring them directly.
3023 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
3024 * should restore the "true" MSRs with the must-be-1 bits
3025 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
3026 * DEFAULT SETTINGS".
3029 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
3030 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
3031 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
3032 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
3033 case MSR_IA32_VMX_PROCBASED_CTLS2:
3034 return vmx_restore_control_msr(vmx, msr_index, data);
3035 case MSR_IA32_VMX_MISC:
3036 return vmx_restore_vmx_misc(vmx, data);
3037 case MSR_IA32_VMX_CR0_FIXED0:
3038 case MSR_IA32_VMX_CR4_FIXED0:
3039 return vmx_restore_fixed0_msr(vmx, msr_index, data);
3040 case MSR_IA32_VMX_CR0_FIXED1:
3041 case MSR_IA32_VMX_CR4_FIXED1:
3043 * These MSRs are generated based on the vCPU's CPUID, so we
3044 * do not support restoring them directly.
3047 case MSR_IA32_VMX_EPT_VPID_CAP:
3048 return vmx_restore_vmx_ept_vpid_cap(vmx, data);
3049 case MSR_IA32_VMX_VMCS_ENUM:
3050 vmx->nested.nested_vmx_vmcs_enum = data;
3054 * The rest of the VMX capability MSRs do not support restore.
3060 /* Returns 0 on success, non-0 otherwise. */
3061 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
3063 struct vcpu_vmx *vmx = to_vmx(vcpu);
3065 switch (msr_index) {
3066 case MSR_IA32_VMX_BASIC:
3067 *pdata = vmx->nested.nested_vmx_basic;
3069 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
3070 case MSR_IA32_VMX_PINBASED_CTLS:
3071 *pdata = vmx_control_msr(
3072 vmx->nested.nested_vmx_pinbased_ctls_low,
3073 vmx->nested.nested_vmx_pinbased_ctls_high);
3074 if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
3075 *pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
3077 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
3078 case MSR_IA32_VMX_PROCBASED_CTLS:
3079 *pdata = vmx_control_msr(
3080 vmx->nested.nested_vmx_procbased_ctls_low,
3081 vmx->nested.nested_vmx_procbased_ctls_high);
3082 if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
3083 *pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
3085 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
3086 case MSR_IA32_VMX_EXIT_CTLS:
3087 *pdata = vmx_control_msr(
3088 vmx->nested.nested_vmx_exit_ctls_low,
3089 vmx->nested.nested_vmx_exit_ctls_high);
3090 if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
3091 *pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
3093 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
3094 case MSR_IA32_VMX_ENTRY_CTLS:
3095 *pdata = vmx_control_msr(
3096 vmx->nested.nested_vmx_entry_ctls_low,
3097 vmx->nested.nested_vmx_entry_ctls_high);
3098 if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
3099 *pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
3101 case MSR_IA32_VMX_MISC:
3102 *pdata = vmx_control_msr(
3103 vmx->nested.nested_vmx_misc_low,
3104 vmx->nested.nested_vmx_misc_high);
3106 case MSR_IA32_VMX_CR0_FIXED0:
3107 *pdata = vmx->nested.nested_vmx_cr0_fixed0;
3109 case MSR_IA32_VMX_CR0_FIXED1:
3110 *pdata = vmx->nested.nested_vmx_cr0_fixed1;
3112 case MSR_IA32_VMX_CR4_FIXED0:
3113 *pdata = vmx->nested.nested_vmx_cr4_fixed0;
3115 case MSR_IA32_VMX_CR4_FIXED1:
3116 *pdata = vmx->nested.nested_vmx_cr4_fixed1;
3118 case MSR_IA32_VMX_VMCS_ENUM:
3119 *pdata = vmx->nested.nested_vmx_vmcs_enum;
3121 case MSR_IA32_VMX_PROCBASED_CTLS2:
3122 *pdata = vmx_control_msr(
3123 vmx->nested.nested_vmx_secondary_ctls_low,
3124 vmx->nested.nested_vmx_secondary_ctls_high);
3126 case MSR_IA32_VMX_EPT_VPID_CAP:
3127 *pdata = vmx->nested.nested_vmx_ept_caps |
3128 ((u64)vmx->nested.nested_vmx_vpid_caps << 32);
3137 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
3140 uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
3142 return !(val & ~valid_bits);
3146 * Reads an msr value (of 'msr_index') into 'pdata'.
3147 * Returns 0 on success, non-0 otherwise.
3148 * Assumes vcpu_load() was already called.
3150 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3152 struct shared_msr_entry *msr;
3154 switch (msr_info->index) {
3155 #ifdef CONFIG_X86_64
3157 msr_info->data = vmcs_readl(GUEST_FS_BASE);
3160 msr_info->data = vmcs_readl(GUEST_GS_BASE);
3162 case MSR_KERNEL_GS_BASE:
3163 vmx_load_host_state(to_vmx(vcpu));
3164 msr_info->data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
3168 return kvm_get_msr_common(vcpu, msr_info);
3170 msr_info->data = guest_read_tsc(vcpu);
3172 case MSR_IA32_SYSENTER_CS:
3173 msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
3175 case MSR_IA32_SYSENTER_EIP:
3176 msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
3178 case MSR_IA32_SYSENTER_ESP:
3179 msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
3181 case MSR_IA32_BNDCFGS:
3182 if (!kvm_mpx_supported())
3184 msr_info->data = vmcs_read64(GUEST_BNDCFGS);
3186 case MSR_IA32_MCG_EXT_CTL:
3187 if (!msr_info->host_initiated &&
3188 !(to_vmx(vcpu)->msr_ia32_feature_control &
3189 FEATURE_CONTROL_LMCE))
3191 msr_info->data = vcpu->arch.mcg_ext_ctl;
3193 case MSR_IA32_FEATURE_CONTROL:
3194 msr_info->data = to_vmx(vcpu)->msr_ia32_feature_control;
3196 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3197 if (!nested_vmx_allowed(vcpu))
3199 return vmx_get_vmx_msr(vcpu, msr_info->index, &msr_info->data);
3201 if (!vmx_xsaves_supported())
3203 msr_info->data = vcpu->arch.ia32_xss;
3206 if (!guest_cpuid_has_rdtscp(vcpu) && !msr_info->host_initiated)
3208 /* Otherwise falls through */
3210 msr = find_msr_entry(to_vmx(vcpu), msr_info->index);
3212 msr_info->data = msr->data;
3215 return kvm_get_msr_common(vcpu, msr_info);
3221 static void vmx_leave_nested(struct kvm_vcpu *vcpu);
3224 * Writes msr value into into the appropriate "register".
3225 * Returns 0 on success, non-0 otherwise.
3226 * Assumes vcpu_load() was already called.
3228 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3230 struct vcpu_vmx *vmx = to_vmx(vcpu);
3231 struct shared_msr_entry *msr;
3233 u32 msr_index = msr_info->index;
3234 u64 data = msr_info->data;
3236 switch (msr_index) {
3238 ret = kvm_set_msr_common(vcpu, msr_info);
3240 #ifdef CONFIG_X86_64
3242 vmx_segment_cache_clear(vmx);
3243 vmcs_writel(GUEST_FS_BASE, data);
3246 vmx_segment_cache_clear(vmx);
3247 vmcs_writel(GUEST_GS_BASE, data);
3249 case MSR_KERNEL_GS_BASE:
3250 vmx_load_host_state(vmx);
3251 vmx->msr_guest_kernel_gs_base = data;
3254 case MSR_IA32_SYSENTER_CS:
3255 vmcs_write32(GUEST_SYSENTER_CS, data);
3257 case MSR_IA32_SYSENTER_EIP:
3258 vmcs_writel(GUEST_SYSENTER_EIP, data);
3260 case MSR_IA32_SYSENTER_ESP:
3261 vmcs_writel(GUEST_SYSENTER_ESP, data);
3263 case MSR_IA32_BNDCFGS:
3264 if (!kvm_mpx_supported())
3266 vmcs_write64(GUEST_BNDCFGS, data);
3269 kvm_write_tsc(vcpu, msr_info);
3271 case MSR_IA32_CR_PAT:
3272 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
3273 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
3275 vmcs_write64(GUEST_IA32_PAT, data);
3276 vcpu->arch.pat = data;
3279 ret = kvm_set_msr_common(vcpu, msr_info);
3281 case MSR_IA32_TSC_ADJUST:
3282 ret = kvm_set_msr_common(vcpu, msr_info);
3284 case MSR_IA32_MCG_EXT_CTL:
3285 if ((!msr_info->host_initiated &&
3286 !(to_vmx(vcpu)->msr_ia32_feature_control &
3287 FEATURE_CONTROL_LMCE)) ||
3288 (data & ~MCG_EXT_CTL_LMCE_EN))
3290 vcpu->arch.mcg_ext_ctl = data;
3292 case MSR_IA32_FEATURE_CONTROL:
3293 if (!vmx_feature_control_msr_valid(vcpu, data) ||
3294 (to_vmx(vcpu)->msr_ia32_feature_control &
3295 FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
3297 vmx->msr_ia32_feature_control = data;
3298 if (msr_info->host_initiated && data == 0)
3299 vmx_leave_nested(vcpu);
3301 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3302 if (!msr_info->host_initiated)
3303 return 1; /* they are read-only */
3304 if (!nested_vmx_allowed(vcpu))
3306 return vmx_set_vmx_msr(vcpu, msr_index, data);
3308 if (!vmx_xsaves_supported())
3311 * The only supported bit as of Skylake is bit 8, but
3312 * it is not supported on KVM.
3316 vcpu->arch.ia32_xss = data;
3317 if (vcpu->arch.ia32_xss != host_xss)
3318 add_atomic_switch_msr(vmx, MSR_IA32_XSS,
3319 vcpu->arch.ia32_xss, host_xss);
3321 clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
3324 if (!guest_cpuid_has_rdtscp(vcpu) && !msr_info->host_initiated)
3326 /* Check reserved bit, higher 32 bits should be zero */
3327 if ((data >> 32) != 0)
3329 /* Otherwise falls through */
3331 msr = find_msr_entry(vmx, msr_index);
3333 u64 old_msr_data = msr->data;
3335 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
3337 ret = kvm_set_shared_msr(msr->index, msr->data,
3341 msr->data = old_msr_data;
3345 ret = kvm_set_msr_common(vcpu, msr_info);
3351 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
3353 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
3356 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
3359 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
3361 case VCPU_EXREG_PDPTR:
3363 ept_save_pdptrs(vcpu);
3370 static __init int cpu_has_kvm_support(void)
3372 return cpu_has_vmx();
3375 static __init int vmx_disabled_by_bios(void)
3379 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
3380 if (msr & FEATURE_CONTROL_LOCKED) {
3381 /* launched w/ TXT and VMX disabled */
3382 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3385 /* launched w/o TXT and VMX only enabled w/ TXT */
3386 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3387 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3388 && !tboot_enabled()) {
3389 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
3390 "activate TXT before enabling KVM\n");
3393 /* launched w/o TXT and VMX disabled */
3394 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3395 && !tboot_enabled())
3402 static void kvm_cpu_vmxon(u64 addr)
3404 cr4_set_bits(X86_CR4_VMXE);
3405 intel_pt_handle_vmx(1);
3407 asm volatile (ASM_VMX_VMXON_RAX
3408 : : "a"(&addr), "m"(addr)
3412 static int hardware_enable(void)
3414 int cpu = raw_smp_processor_id();
3415 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
3418 if (cr4_read_shadow() & X86_CR4_VMXE)
3421 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
3422 INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
3423 spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
3426 * Now we can enable the vmclear operation in kdump
3427 * since the loaded_vmcss_on_cpu list on this cpu
3428 * has been initialized.
3430 * Though the cpu is not in VMX operation now, there
3431 * is no problem to enable the vmclear operation
3432 * for the loaded_vmcss_on_cpu list is empty!
3434 crash_enable_local_vmclear(cpu);
3436 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
3438 test_bits = FEATURE_CONTROL_LOCKED;
3439 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
3440 if (tboot_enabled())
3441 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
3443 if ((old & test_bits) != test_bits) {
3444 /* enable and lock */
3445 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
3447 kvm_cpu_vmxon(phys_addr);
3450 native_store_gdt(this_cpu_ptr(&host_gdt));
3455 static void vmclear_local_loaded_vmcss(void)
3457 int cpu = raw_smp_processor_id();
3458 struct loaded_vmcs *v, *n;
3460 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
3461 loaded_vmcss_on_cpu_link)
3462 __loaded_vmcs_clear(v);
3466 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
3469 static void kvm_cpu_vmxoff(void)
3471 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
3473 intel_pt_handle_vmx(0);
3474 cr4_clear_bits(X86_CR4_VMXE);
3477 static void hardware_disable(void)
3479 vmclear_local_loaded_vmcss();
3483 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
3484 u32 msr, u32 *result)
3486 u32 vmx_msr_low, vmx_msr_high;
3487 u32 ctl = ctl_min | ctl_opt;
3489 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3491 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
3492 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
3494 /* Ensure minimum (required) set of control bits are supported. */
3502 static __init bool allow_1_setting(u32 msr, u32 ctl)
3504 u32 vmx_msr_low, vmx_msr_high;
3506 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3507 return vmx_msr_high & ctl;
3510 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
3512 u32 vmx_msr_low, vmx_msr_high;
3513 u32 min, opt, min2, opt2;
3514 u32 _pin_based_exec_control = 0;
3515 u32 _cpu_based_exec_control = 0;
3516 u32 _cpu_based_2nd_exec_control = 0;
3517 u32 _vmexit_control = 0;
3518 u32 _vmentry_control = 0;
3520 min = CPU_BASED_HLT_EXITING |
3521 #ifdef CONFIG_X86_64
3522 CPU_BASED_CR8_LOAD_EXITING |
3523 CPU_BASED_CR8_STORE_EXITING |
3525 CPU_BASED_CR3_LOAD_EXITING |
3526 CPU_BASED_CR3_STORE_EXITING |
3527 CPU_BASED_USE_IO_BITMAPS |
3528 CPU_BASED_MOV_DR_EXITING |
3529 CPU_BASED_USE_TSC_OFFSETING |
3530 CPU_BASED_MWAIT_EXITING |
3531 CPU_BASED_MONITOR_EXITING |
3532 CPU_BASED_INVLPG_EXITING |
3533 CPU_BASED_RDPMC_EXITING;
3535 opt = CPU_BASED_TPR_SHADOW |
3536 CPU_BASED_USE_MSR_BITMAPS |
3537 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
3538 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
3539 &_cpu_based_exec_control) < 0)
3541 #ifdef CONFIG_X86_64
3542 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3543 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
3544 ~CPU_BASED_CR8_STORE_EXITING;
3546 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
3548 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
3549 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3550 SECONDARY_EXEC_WBINVD_EXITING |
3551 SECONDARY_EXEC_ENABLE_VPID |
3552 SECONDARY_EXEC_ENABLE_EPT |
3553 SECONDARY_EXEC_UNRESTRICTED_GUEST |
3554 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
3555 SECONDARY_EXEC_RDTSCP |
3556 SECONDARY_EXEC_ENABLE_INVPCID |
3557 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3558 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
3559 SECONDARY_EXEC_SHADOW_VMCS |
3560 SECONDARY_EXEC_XSAVES |
3561 SECONDARY_EXEC_ENABLE_PML |
3562 SECONDARY_EXEC_TSC_SCALING;
3563 if (adjust_vmx_controls(min2, opt2,
3564 MSR_IA32_VMX_PROCBASED_CTLS2,
3565 &_cpu_based_2nd_exec_control) < 0)
3568 #ifndef CONFIG_X86_64
3569 if (!(_cpu_based_2nd_exec_control &
3570 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
3571 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
3574 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3575 _cpu_based_2nd_exec_control &= ~(
3576 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3577 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3578 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3580 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
3581 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
3583 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
3584 CPU_BASED_CR3_STORE_EXITING |
3585 CPU_BASED_INVLPG_EXITING);
3586 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
3587 vmx_capability.ept, vmx_capability.vpid);
3590 min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
3591 #ifdef CONFIG_X86_64
3592 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
3594 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT |
3595 VM_EXIT_CLEAR_BNDCFGS;
3596 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
3597 &_vmexit_control) < 0)
3600 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING |
3601 PIN_BASED_VIRTUAL_NMIS;
3602 opt = PIN_BASED_POSTED_INTR | PIN_BASED_VMX_PREEMPTION_TIMER;
3603 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
3604 &_pin_based_exec_control) < 0)
3607 if (cpu_has_broken_vmx_preemption_timer())
3608 _pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3609 if (!(_cpu_based_2nd_exec_control &
3610 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
3611 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
3613 min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
3614 opt = VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS;
3615 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
3616 &_vmentry_control) < 0)
3619 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
3621 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
3622 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
3625 #ifdef CONFIG_X86_64
3626 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
3627 if (vmx_msr_high & (1u<<16))
3631 /* Require Write-Back (WB) memory type for VMCS accesses. */
3632 if (((vmx_msr_high >> 18) & 15) != 6)
3635 vmcs_conf->size = vmx_msr_high & 0x1fff;
3636 vmcs_conf->order = get_order(vmcs_conf->size);
3637 vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
3638 vmcs_conf->revision_id = vmx_msr_low;
3640 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
3641 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
3642 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
3643 vmcs_conf->vmexit_ctrl = _vmexit_control;
3644 vmcs_conf->vmentry_ctrl = _vmentry_control;
3646 cpu_has_load_ia32_efer =
3647 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3648 VM_ENTRY_LOAD_IA32_EFER)
3649 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3650 VM_EXIT_LOAD_IA32_EFER);
3652 cpu_has_load_perf_global_ctrl =
3653 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3654 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
3655 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3656 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
3659 * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
3660 * but due to errata below it can't be used. Workaround is to use
3661 * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
3663 * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
3668 * BC86,AAY89,BD102 (model 44)
3672 if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
3673 switch (boot_cpu_data.x86_model) {
3679 cpu_has_load_perf_global_ctrl = false;
3680 printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
3681 "does not work properly. Using workaround\n");
3688 if (boot_cpu_has(X86_FEATURE_XSAVES))
3689 rdmsrl(MSR_IA32_XSS, host_xss);
3694 static struct vmcs *alloc_vmcs_cpu(int cpu)
3696 int node = cpu_to_node(cpu);
3700 pages = __alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
3703 vmcs = page_address(pages);
3704 memset(vmcs, 0, vmcs_config.size);
3705 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
3709 static struct vmcs *alloc_vmcs(void)
3711 return alloc_vmcs_cpu(raw_smp_processor_id());
3714 static void free_vmcs(struct vmcs *vmcs)
3716 free_pages((unsigned long)vmcs, vmcs_config.order);
3720 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
3722 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
3724 if (!loaded_vmcs->vmcs)
3726 loaded_vmcs_clear(loaded_vmcs);
3727 free_vmcs(loaded_vmcs->vmcs);
3728 loaded_vmcs->vmcs = NULL;
3729 WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
3732 static void free_kvm_area(void)
3736 for_each_possible_cpu(cpu) {
3737 free_vmcs(per_cpu(vmxarea, cpu));
3738 per_cpu(vmxarea, cpu) = NULL;
3742 static void init_vmcs_shadow_fields(void)
3746 /* No checks for read only fields yet */
3748 for (i = j = 0; i < max_shadow_read_write_fields; i++) {
3749 switch (shadow_read_write_fields[i]) {
3751 if (!kvm_mpx_supported())
3759 shadow_read_write_fields[j] =
3760 shadow_read_write_fields[i];
3763 max_shadow_read_write_fields = j;
3765 /* shadowed fields guest access without vmexit */
3766 for (i = 0; i < max_shadow_read_write_fields; i++) {
3767 clear_bit(shadow_read_write_fields[i],
3768 vmx_vmwrite_bitmap);
3769 clear_bit(shadow_read_write_fields[i],
3772 for (i = 0; i < max_shadow_read_only_fields; i++)
3773 clear_bit(shadow_read_only_fields[i],
3777 static __init int alloc_kvm_area(void)
3781 for_each_possible_cpu(cpu) {
3784 vmcs = alloc_vmcs_cpu(cpu);
3790 per_cpu(vmxarea, cpu) = vmcs;
3795 static bool emulation_required(struct kvm_vcpu *vcpu)
3797 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
3800 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
3801 struct kvm_segment *save)
3803 if (!emulate_invalid_guest_state) {
3805 * CS and SS RPL should be equal during guest entry according
3806 * to VMX spec, but in reality it is not always so. Since vcpu
3807 * is in the middle of the transition from real mode to
3808 * protected mode it is safe to assume that RPL 0 is a good
3811 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
3812 save->selector &= ~SEGMENT_RPL_MASK;
3813 save->dpl = save->selector & SEGMENT_RPL_MASK;
3816 vmx_set_segment(vcpu, save, seg);
3819 static void enter_pmode(struct kvm_vcpu *vcpu)
3821 unsigned long flags;
3822 struct vcpu_vmx *vmx = to_vmx(vcpu);
3825 * Update real mode segment cache. It may be not up-to-date if sement
3826 * register was written while vcpu was in a guest mode.
3828 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3829 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3830 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3831 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3832 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3833 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3835 vmx->rmode.vm86_active = 0;
3837 vmx_segment_cache_clear(vmx);
3839 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3841 flags = vmcs_readl(GUEST_RFLAGS);
3842 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
3843 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
3844 vmcs_writel(GUEST_RFLAGS, flags);
3846 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
3847 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
3849 update_exception_bitmap(vcpu);
3851 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3852 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3853 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3854 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3855 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3856 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3859 static void fix_rmode_seg(int seg, struct kvm_segment *save)
3861 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3862 struct kvm_segment var = *save;
3865 if (seg == VCPU_SREG_CS)
3868 if (!emulate_invalid_guest_state) {
3869 var.selector = var.base >> 4;
3870 var.base = var.base & 0xffff0;
3880 if (save->base & 0xf)
3881 printk_once(KERN_WARNING "kvm: segment base is not "
3882 "paragraph aligned when entering "
3883 "protected mode (seg=%d)", seg);
3886 vmcs_write16(sf->selector, var.selector);
3887 vmcs_writel(sf->base, var.base);
3888 vmcs_write32(sf->limit, var.limit);
3889 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
3892 static void enter_rmode(struct kvm_vcpu *vcpu)
3894 unsigned long flags;
3895 struct vcpu_vmx *vmx = to_vmx(vcpu);
3897 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3898 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3899 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3900 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3901 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3902 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3903 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3905 vmx->rmode.vm86_active = 1;
3908 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
3909 * vcpu. Warn the user that an update is overdue.
3911 if (!vcpu->kvm->arch.tss_addr)
3912 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
3913 "called before entering vcpu\n");
3915 vmx_segment_cache_clear(vmx);
3917 vmcs_writel(GUEST_TR_BASE, vcpu->kvm->arch.tss_addr);
3918 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
3919 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3921 flags = vmcs_readl(GUEST_RFLAGS);
3922 vmx->rmode.save_rflags = flags;
3924 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
3926 vmcs_writel(GUEST_RFLAGS, flags);
3927 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
3928 update_exception_bitmap(vcpu);
3930 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3931 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3932 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3933 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3934 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3935 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3937 kvm_mmu_reset_context(vcpu);
3940 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
3942 struct vcpu_vmx *vmx = to_vmx(vcpu);
3943 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
3949 * Force kernel_gs_base reloading before EFER changes, as control
3950 * of this msr depends on is_long_mode().
3952 vmx_load_host_state(to_vmx(vcpu));
3953 vcpu->arch.efer = efer;
3954 if (efer & EFER_LMA) {
3955 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3958 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3960 msr->data = efer & ~EFER_LME;
3965 #ifdef CONFIG_X86_64
3967 static void enter_lmode(struct kvm_vcpu *vcpu)
3971 vmx_segment_cache_clear(to_vmx(vcpu));
3973 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
3974 if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
3975 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
3977 vmcs_write32(GUEST_TR_AR_BYTES,
3978 (guest_tr_ar & ~VMX_AR_TYPE_MASK)
3979 | VMX_AR_TYPE_BUSY_64_TSS);
3981 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
3984 static void exit_lmode(struct kvm_vcpu *vcpu)
3986 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3987 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
3992 static inline void __vmx_flush_tlb(struct kvm_vcpu *vcpu, int vpid)
3995 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
3997 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
3999 vpid_sync_context(vpid);
4003 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
4005 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->vpid);
4008 static void vmx_flush_tlb_ept_only(struct kvm_vcpu *vcpu)
4011 vmx_flush_tlb(vcpu);
4014 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
4016 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
4018 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
4019 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
4022 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
4024 if (enable_ept && is_paging(vcpu))
4025 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
4026 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
4029 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
4031 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
4033 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
4034 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
4037 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
4039 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4041 if (!test_bit(VCPU_EXREG_PDPTR,
4042 (unsigned long *)&vcpu->arch.regs_dirty))
4045 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
4046 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
4047 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
4048 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
4049 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
4053 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
4055 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4057 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
4058 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
4059 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
4060 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
4061 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
4064 __set_bit(VCPU_EXREG_PDPTR,
4065 (unsigned long *)&vcpu->arch.regs_avail);
4066 __set_bit(VCPU_EXREG_PDPTR,
4067 (unsigned long *)&vcpu->arch.regs_dirty);
4070 static bool nested_guest_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
4072 u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed0;
4073 u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed1;
4074 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4076 if (to_vmx(vcpu)->nested.nested_vmx_secondary_ctls_high &
4077 SECONDARY_EXEC_UNRESTRICTED_GUEST &&
4078 nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
4079 fixed0 &= ~(X86_CR0_PE | X86_CR0_PG);
4081 return fixed_bits_valid(val, fixed0, fixed1);
4084 static bool nested_host_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
4086 u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed0;
4087 u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed1;
4089 return fixed_bits_valid(val, fixed0, fixed1);
4092 static bool nested_cr4_valid(struct kvm_vcpu *vcpu, unsigned long val)
4094 u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr4_fixed0;
4095 u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr4_fixed1;
4097 return fixed_bits_valid(val, fixed0, fixed1);
4100 /* No difference in the restrictions on guest and host CR4 in VMX operation. */
4101 #define nested_guest_cr4_valid nested_cr4_valid
4102 #define nested_host_cr4_valid nested_cr4_valid
4104 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
4106 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
4108 struct kvm_vcpu *vcpu)
4110 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
4111 vmx_decache_cr3(vcpu);
4112 if (!(cr0 & X86_CR0_PG)) {
4113 /* From paging/starting to nonpaging */
4114 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
4115 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
4116 (CPU_BASED_CR3_LOAD_EXITING |
4117 CPU_BASED_CR3_STORE_EXITING));
4118 vcpu->arch.cr0 = cr0;
4119 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
4120 } else if (!is_paging(vcpu)) {
4121 /* From nonpaging to paging */
4122 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
4123 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
4124 ~(CPU_BASED_CR3_LOAD_EXITING |
4125 CPU_BASED_CR3_STORE_EXITING));
4126 vcpu->arch.cr0 = cr0;
4127 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
4130 if (!(cr0 & X86_CR0_WP))
4131 *hw_cr0 &= ~X86_CR0_WP;
4134 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
4136 struct vcpu_vmx *vmx = to_vmx(vcpu);
4137 unsigned long hw_cr0;
4139 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK);
4140 if (enable_unrestricted_guest)
4141 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
4143 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
4145 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
4148 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
4152 #ifdef CONFIG_X86_64
4153 if (vcpu->arch.efer & EFER_LME) {
4154 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
4156 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
4162 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
4164 vmcs_writel(CR0_READ_SHADOW, cr0);
4165 vmcs_writel(GUEST_CR0, hw_cr0);
4166 vcpu->arch.cr0 = cr0;
4168 /* depends on vcpu->arch.cr0 to be set to a new value */
4169 vmx->emulation_required = emulation_required(vcpu);
4172 static u64 construct_eptp(unsigned long root_hpa)
4176 /* TODO write the value reading from MSR */
4177 eptp = VMX_EPT_DEFAULT_MT |
4178 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
4179 if (enable_ept_ad_bits)
4180 eptp |= VMX_EPT_AD_ENABLE_BIT;
4181 eptp |= (root_hpa & PAGE_MASK);
4186 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
4188 unsigned long guest_cr3;
4193 eptp = construct_eptp(cr3);
4194 vmcs_write64(EPT_POINTER, eptp);
4195 if (is_paging(vcpu) || is_guest_mode(vcpu))
4196 guest_cr3 = kvm_read_cr3(vcpu);
4198 guest_cr3 = vcpu->kvm->arch.ept_identity_map_addr;
4199 ept_load_pdptrs(vcpu);
4202 vmx_flush_tlb(vcpu);
4203 vmcs_writel(GUEST_CR3, guest_cr3);
4206 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
4209 * Pass through host's Machine Check Enable value to hw_cr4, which
4210 * is in force while we are in guest mode. Do not let guests control
4211 * this bit, even if host CR4.MCE == 0.
4213 unsigned long hw_cr4 =
4214 (cr4_read_shadow() & X86_CR4_MCE) |
4215 (cr4 & ~X86_CR4_MCE) |
4216 (to_vmx(vcpu)->rmode.vm86_active ?
4217 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
4219 if (cr4 & X86_CR4_VMXE) {
4221 * To use VMXON (and later other VMX instructions), a guest
4222 * must first be able to turn on cr4.VMXE (see handle_vmon()).
4223 * So basically the check on whether to allow nested VMX
4226 if (!nested_vmx_allowed(vcpu))
4230 if (to_vmx(vcpu)->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
4233 vcpu->arch.cr4 = cr4;
4235 if (!is_paging(vcpu)) {
4236 hw_cr4 &= ~X86_CR4_PAE;
4237 hw_cr4 |= X86_CR4_PSE;
4238 } else if (!(cr4 & X86_CR4_PAE)) {
4239 hw_cr4 &= ~X86_CR4_PAE;
4243 if (!enable_unrestricted_guest && !is_paging(vcpu))
4245 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
4246 * hardware. To emulate this behavior, SMEP/SMAP/PKU needs
4247 * to be manually disabled when guest switches to non-paging
4250 * If !enable_unrestricted_guest, the CPU is always running
4251 * with CR0.PG=1 and CR4 needs to be modified.
4252 * If enable_unrestricted_guest, the CPU automatically
4253 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
4255 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
4257 vmcs_writel(CR4_READ_SHADOW, cr4);
4258 vmcs_writel(GUEST_CR4, hw_cr4);
4262 static void vmx_get_segment(struct kvm_vcpu *vcpu,
4263 struct kvm_segment *var, int seg)
4265 struct vcpu_vmx *vmx = to_vmx(vcpu);
4268 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
4269 *var = vmx->rmode.segs[seg];
4270 if (seg == VCPU_SREG_TR
4271 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
4273 var->base = vmx_read_guest_seg_base(vmx, seg);
4274 var->selector = vmx_read_guest_seg_selector(vmx, seg);
4277 var->base = vmx_read_guest_seg_base(vmx, seg);
4278 var->limit = vmx_read_guest_seg_limit(vmx, seg);
4279 var->selector = vmx_read_guest_seg_selector(vmx, seg);
4280 ar = vmx_read_guest_seg_ar(vmx, seg);
4281 var->unusable = (ar >> 16) & 1;
4282 var->type = ar & 15;
4283 var->s = (ar >> 4) & 1;
4284 var->dpl = (ar >> 5) & 3;
4286 * Some userspaces do not preserve unusable property. Since usable
4287 * segment has to be present according to VMX spec we can use present
4288 * property to amend userspace bug by making unusable segment always
4289 * nonpresent. vmx_segment_access_rights() already marks nonpresent
4290 * segment as unusable.
4292 var->present = !var->unusable;
4293 var->avl = (ar >> 12) & 1;
4294 var->l = (ar >> 13) & 1;
4295 var->db = (ar >> 14) & 1;
4296 var->g = (ar >> 15) & 1;
4299 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
4301 struct kvm_segment s;
4303 if (to_vmx(vcpu)->rmode.vm86_active) {
4304 vmx_get_segment(vcpu, &s, seg);
4307 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
4310 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
4312 struct vcpu_vmx *vmx = to_vmx(vcpu);
4314 if (unlikely(vmx->rmode.vm86_active))
4317 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
4318 return VMX_AR_DPL(ar);
4322 static u32 vmx_segment_access_rights(struct kvm_segment *var)
4326 if (var->unusable || !var->present)
4329 ar = var->type & 15;
4330 ar |= (var->s & 1) << 4;
4331 ar |= (var->dpl & 3) << 5;
4332 ar |= (var->present & 1) << 7;
4333 ar |= (var->avl & 1) << 12;
4334 ar |= (var->l & 1) << 13;
4335 ar |= (var->db & 1) << 14;
4336 ar |= (var->g & 1) << 15;
4342 static void vmx_set_segment(struct kvm_vcpu *vcpu,
4343 struct kvm_segment *var, int seg)
4345 struct vcpu_vmx *vmx = to_vmx(vcpu);
4346 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4348 vmx_segment_cache_clear(vmx);
4350 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
4351 vmx->rmode.segs[seg] = *var;
4352 if (seg == VCPU_SREG_TR)
4353 vmcs_write16(sf->selector, var->selector);
4355 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
4359 vmcs_writel(sf->base, var->base);
4360 vmcs_write32(sf->limit, var->limit);
4361 vmcs_write16(sf->selector, var->selector);
4364 * Fix the "Accessed" bit in AR field of segment registers for older
4366 * IA32 arch specifies that at the time of processor reset the
4367 * "Accessed" bit in the AR field of segment registers is 1. And qemu
4368 * is setting it to 0 in the userland code. This causes invalid guest
4369 * state vmexit when "unrestricted guest" mode is turned on.
4370 * Fix for this setup issue in cpu_reset is being pushed in the qemu
4371 * tree. Newer qemu binaries with that qemu fix would not need this
4374 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
4375 var->type |= 0x1; /* Accessed */
4377 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
4380 vmx->emulation_required = emulation_required(vcpu);
4383 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4385 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
4387 *db = (ar >> 14) & 1;
4388 *l = (ar >> 13) & 1;
4391 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4393 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
4394 dt->address = vmcs_readl(GUEST_IDTR_BASE);
4397 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4399 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
4400 vmcs_writel(GUEST_IDTR_BASE, dt->address);
4403 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4405 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
4406 dt->address = vmcs_readl(GUEST_GDTR_BASE);
4409 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4411 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
4412 vmcs_writel(GUEST_GDTR_BASE, dt->address);
4415 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
4417 struct kvm_segment var;
4420 vmx_get_segment(vcpu, &var, seg);
4422 if (seg == VCPU_SREG_CS)
4424 ar = vmx_segment_access_rights(&var);
4426 if (var.base != (var.selector << 4))
4428 if (var.limit != 0xffff)
4436 static bool code_segment_valid(struct kvm_vcpu *vcpu)
4438 struct kvm_segment cs;
4439 unsigned int cs_rpl;
4441 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4442 cs_rpl = cs.selector & SEGMENT_RPL_MASK;
4446 if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
4450 if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
4451 if (cs.dpl > cs_rpl)
4454 if (cs.dpl != cs_rpl)
4460 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
4464 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
4466 struct kvm_segment ss;
4467 unsigned int ss_rpl;
4469 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4470 ss_rpl = ss.selector & SEGMENT_RPL_MASK;
4474 if (ss.type != 3 && ss.type != 7)
4478 if (ss.dpl != ss_rpl) /* DPL != RPL */
4486 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
4488 struct kvm_segment var;
4491 vmx_get_segment(vcpu, &var, seg);
4492 rpl = var.selector & SEGMENT_RPL_MASK;
4500 if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
4501 if (var.dpl < rpl) /* DPL < RPL */
4505 /* TODO: Add other members to kvm_segment_field to allow checking for other access
4511 static bool tr_valid(struct kvm_vcpu *vcpu)
4513 struct kvm_segment tr;
4515 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
4519 if (tr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4521 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
4529 static bool ldtr_valid(struct kvm_vcpu *vcpu)
4531 struct kvm_segment ldtr;
4533 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
4537 if (ldtr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4547 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
4549 struct kvm_segment cs, ss;
4551 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4552 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4554 return ((cs.selector & SEGMENT_RPL_MASK) ==
4555 (ss.selector & SEGMENT_RPL_MASK));
4559 * Check if guest state is valid. Returns true if valid, false if
4561 * We assume that registers are always usable
4563 static bool guest_state_valid(struct kvm_vcpu *vcpu)
4565 if (enable_unrestricted_guest)
4568 /* real mode guest state checks */
4569 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
4570 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
4572 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
4574 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
4576 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
4578 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
4580 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
4583 /* protected mode guest state checks */
4584 if (!cs_ss_rpl_check(vcpu))
4586 if (!code_segment_valid(vcpu))
4588 if (!stack_segment_valid(vcpu))
4590 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
4592 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
4594 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
4596 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
4598 if (!tr_valid(vcpu))
4600 if (!ldtr_valid(vcpu))
4604 * - Add checks on RIP
4605 * - Add checks on RFLAGS
4611 static int init_rmode_tss(struct kvm *kvm)
4617 idx = srcu_read_lock(&kvm->srcu);
4618 fn = kvm->arch.tss_addr >> PAGE_SHIFT;
4619 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4622 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
4623 r = kvm_write_guest_page(kvm, fn++, &data,
4624 TSS_IOPB_BASE_OFFSET, sizeof(u16));
4627 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
4630 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4634 r = kvm_write_guest_page(kvm, fn, &data,
4635 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
4638 srcu_read_unlock(&kvm->srcu, idx);
4642 static int init_rmode_identity_map(struct kvm *kvm)
4645 kvm_pfn_t identity_map_pfn;
4651 /* Protect kvm->arch.ept_identity_pagetable_done. */
4652 mutex_lock(&kvm->slots_lock);
4654 if (likely(kvm->arch.ept_identity_pagetable_done))
4657 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
4659 r = alloc_identity_pagetable(kvm);
4663 idx = srcu_read_lock(&kvm->srcu);
4664 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
4667 /* Set up identity-mapping pagetable for EPT in real mode */
4668 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
4669 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
4670 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
4671 r = kvm_write_guest_page(kvm, identity_map_pfn,
4672 &tmp, i * sizeof(tmp), sizeof(tmp));
4676 kvm->arch.ept_identity_pagetable_done = true;
4679 srcu_read_unlock(&kvm->srcu, idx);
4682 mutex_unlock(&kvm->slots_lock);
4686 static void seg_setup(int seg)
4688 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4691 vmcs_write16(sf->selector, 0);
4692 vmcs_writel(sf->base, 0);
4693 vmcs_write32(sf->limit, 0xffff);
4695 if (seg == VCPU_SREG_CS)
4696 ar |= 0x08; /* code segment */
4698 vmcs_write32(sf->ar_bytes, ar);
4701 static int alloc_apic_access_page(struct kvm *kvm)
4706 mutex_lock(&kvm->slots_lock);
4707 if (kvm->arch.apic_access_page_done)
4709 r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
4710 APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
4714 page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
4715 if (is_error_page(page)) {
4721 * Do not pin the page in memory, so that memory hot-unplug
4722 * is able to migrate it.
4725 kvm->arch.apic_access_page_done = true;
4727 mutex_unlock(&kvm->slots_lock);
4731 static int alloc_identity_pagetable(struct kvm *kvm)
4733 /* Called with kvm->slots_lock held. */
4737 BUG_ON(kvm->arch.ept_identity_pagetable_done);
4739 r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
4740 kvm->arch.ept_identity_map_addr, PAGE_SIZE);
4745 static int allocate_vpid(void)
4751 spin_lock(&vmx_vpid_lock);
4752 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
4753 if (vpid < VMX_NR_VPIDS)
4754 __set_bit(vpid, vmx_vpid_bitmap);
4757 spin_unlock(&vmx_vpid_lock);
4761 static void free_vpid(int vpid)
4763 if (!enable_vpid || vpid == 0)
4765 spin_lock(&vmx_vpid_lock);
4766 __clear_bit(vpid, vmx_vpid_bitmap);
4767 spin_unlock(&vmx_vpid_lock);
4770 #define MSR_TYPE_R 1
4771 #define MSR_TYPE_W 2
4772 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
4775 int f = sizeof(unsigned long);
4777 if (!cpu_has_vmx_msr_bitmap())
4781 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4782 * have the write-low and read-high bitmap offsets the wrong way round.
4783 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4785 if (msr <= 0x1fff) {
4786 if (type & MSR_TYPE_R)
4788 __clear_bit(msr, msr_bitmap + 0x000 / f);
4790 if (type & MSR_TYPE_W)
4792 __clear_bit(msr, msr_bitmap + 0x800 / f);
4794 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4796 if (type & MSR_TYPE_R)
4798 __clear_bit(msr, msr_bitmap + 0x400 / f);
4800 if (type & MSR_TYPE_W)
4802 __clear_bit(msr, msr_bitmap + 0xc00 / f);
4808 * If a msr is allowed by L0, we should check whether it is allowed by L1.
4809 * The corresponding bit will be cleared unless both of L0 and L1 allow it.
4811 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
4812 unsigned long *msr_bitmap_nested,
4815 int f = sizeof(unsigned long);
4817 if (!cpu_has_vmx_msr_bitmap()) {
4823 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4824 * have the write-low and read-high bitmap offsets the wrong way round.
4825 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4827 if (msr <= 0x1fff) {
4828 if (type & MSR_TYPE_R &&
4829 !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
4831 __clear_bit(msr, msr_bitmap_nested + 0x000 / f);
4833 if (type & MSR_TYPE_W &&
4834 !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
4836 __clear_bit(msr, msr_bitmap_nested + 0x800 / f);
4838 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4840 if (type & MSR_TYPE_R &&
4841 !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
4843 __clear_bit(msr, msr_bitmap_nested + 0x400 / f);
4845 if (type & MSR_TYPE_W &&
4846 !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
4848 __clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
4853 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
4856 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy,
4857 msr, MSR_TYPE_R | MSR_TYPE_W);
4858 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode,
4859 msr, MSR_TYPE_R | MSR_TYPE_W);
4862 static void vmx_disable_intercept_msr_x2apic(u32 msr, int type, bool apicv_active)
4865 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic_apicv,
4867 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic_apicv,
4870 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4872 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4877 static bool vmx_get_enable_apicv(void)
4879 return enable_apicv;
4882 static void vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
4884 struct vcpu_vmx *vmx = to_vmx(vcpu);
4889 if (vmx->nested.pi_desc &&
4890 vmx->nested.pi_pending) {
4891 vmx->nested.pi_pending = false;
4892 if (!pi_test_and_clear_on(vmx->nested.pi_desc))
4895 max_irr = find_last_bit(
4896 (unsigned long *)vmx->nested.pi_desc->pir, 256);
4901 vapic_page = kmap(vmx->nested.virtual_apic_page);
4902 __kvm_apic_update_irr(vmx->nested.pi_desc->pir, vapic_page);
4903 kunmap(vmx->nested.virtual_apic_page);
4905 status = vmcs_read16(GUEST_INTR_STATUS);
4906 if ((u8)max_irr > ((u8)status & 0xff)) {
4908 status |= (u8)max_irr;
4909 vmcs_write16(GUEST_INTR_STATUS, status);
4914 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu)
4917 if (vcpu->mode == IN_GUEST_MODE) {
4918 struct vcpu_vmx *vmx = to_vmx(vcpu);
4921 * Currently, we don't support urgent interrupt,
4922 * all interrupts are recognized as non-urgent
4923 * interrupt, so we cannot post interrupts when
4926 * If the vcpu is in guest mode, it means it is
4927 * running instead of being scheduled out and
4928 * waiting in the run queue, and that's the only
4929 * case when 'SN' is set currently, warning if
4932 WARN_ON_ONCE(pi_test_sn(&vmx->pi_desc));
4934 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu),
4935 POSTED_INTR_VECTOR);
4942 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
4945 struct vcpu_vmx *vmx = to_vmx(vcpu);
4947 if (is_guest_mode(vcpu) &&
4948 vector == vmx->nested.posted_intr_nv) {
4949 /* the PIR and ON have been set by L1. */
4950 kvm_vcpu_trigger_posted_interrupt(vcpu);
4952 * If a posted intr is not recognized by hardware,
4953 * we will accomplish it in the next vmentry.
4955 vmx->nested.pi_pending = true;
4956 kvm_make_request(KVM_REQ_EVENT, vcpu);
4962 * Send interrupt to vcpu via posted interrupt way.
4963 * 1. If target vcpu is running(non-root mode), send posted interrupt
4964 * notification to vcpu and hardware will sync PIR to vIRR atomically.
4965 * 2. If target vcpu isn't running(root mode), kick it to pick up the
4966 * interrupt from PIR in next vmentry.
4968 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
4970 struct vcpu_vmx *vmx = to_vmx(vcpu);
4973 r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
4977 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
4980 /* If a previous notification has sent the IPI, nothing to do. */
4981 if (pi_test_and_set_on(&vmx->pi_desc))
4984 if (!kvm_vcpu_trigger_posted_interrupt(vcpu))
4985 kvm_vcpu_kick(vcpu);
4989 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
4990 * will not change in the lifetime of the guest.
4991 * Note that host-state that does change is set elsewhere. E.g., host-state
4992 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
4994 static void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
4999 unsigned long cr0, cr4;
5002 WARN_ON(cr0 & X86_CR0_TS);
5003 vmcs_writel(HOST_CR0, cr0); /* 22.2.3 */
5004 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
5006 /* Save the most likely value for this task's CR4 in the VMCS. */
5007 cr4 = cr4_read_shadow();
5008 vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
5009 vmx->host_state.vmcs_host_cr4 = cr4;
5011 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
5012 #ifdef CONFIG_X86_64
5014 * Load null selectors, so we can avoid reloading them in
5015 * __vmx_load_host_state(), in case userspace uses the null selectors
5016 * too (the expected case).
5018 vmcs_write16(HOST_DS_SELECTOR, 0);
5019 vmcs_write16(HOST_ES_SELECTOR, 0);
5021 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
5022 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
5024 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
5025 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
5027 native_store_idt(&dt);
5028 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
5029 vmx->host_idt_base = dt.address;
5031 vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
5033 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
5034 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
5035 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
5036 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
5038 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
5039 rdmsr(MSR_IA32_CR_PAT, low32, high32);
5040 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
5044 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
5046 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
5048 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
5049 if (is_guest_mode(&vmx->vcpu))
5050 vmx->vcpu.arch.cr4_guest_owned_bits &=
5051 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
5052 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
5055 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
5057 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
5059 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
5060 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
5061 /* Enable the preemption timer dynamically */
5062 pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
5063 return pin_based_exec_ctrl;
5066 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
5068 struct vcpu_vmx *vmx = to_vmx(vcpu);
5070 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
5071 if (cpu_has_secondary_exec_ctrls()) {
5072 if (kvm_vcpu_apicv_active(vcpu))
5073 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
5074 SECONDARY_EXEC_APIC_REGISTER_VIRT |
5075 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5077 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
5078 SECONDARY_EXEC_APIC_REGISTER_VIRT |
5079 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5082 if (cpu_has_vmx_msr_bitmap())
5083 vmx_set_msr_bitmap(vcpu);
5086 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
5088 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
5090 if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
5091 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
5093 if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
5094 exec_control &= ~CPU_BASED_TPR_SHADOW;
5095 #ifdef CONFIG_X86_64
5096 exec_control |= CPU_BASED_CR8_STORE_EXITING |
5097 CPU_BASED_CR8_LOAD_EXITING;
5101 exec_control |= CPU_BASED_CR3_STORE_EXITING |
5102 CPU_BASED_CR3_LOAD_EXITING |
5103 CPU_BASED_INVLPG_EXITING;
5104 return exec_control;
5107 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
5109 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
5110 if (!cpu_need_virtualize_apic_accesses(&vmx->vcpu))
5111 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
5113 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
5115 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
5116 enable_unrestricted_guest = 0;
5117 /* Enable INVPCID for non-ept guests may cause performance regression. */
5118 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
5120 if (!enable_unrestricted_guest)
5121 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
5123 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
5124 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
5125 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
5126 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5127 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
5128 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
5130 We can NOT enable shadow_vmcs here because we don't have yet
5133 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
5136 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
5138 return exec_control;
5141 static void ept_set_mmio_spte_mask(void)
5144 * EPT Misconfigurations can be generated if the value of bits 2:0
5145 * of an EPT paging-structure entry is 110b (write/execute).
5147 kvm_mmu_set_mmio_spte_mask(VMX_EPT_MISCONFIG_WX_VALUE);
5150 #define VMX_XSS_EXIT_BITMAP 0
5152 * Sets up the vmcs for emulated real mode.
5154 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
5156 #ifdef CONFIG_X86_64
5162 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
5163 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
5165 if (enable_shadow_vmcs) {
5166 vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
5167 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
5169 if (cpu_has_vmx_msr_bitmap())
5170 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
5172 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
5175 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
5176 vmx->hv_deadline_tsc = -1;
5178 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
5180 if (cpu_has_secondary_exec_ctrls()) {
5181 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
5182 vmx_secondary_exec_control(vmx));
5185 if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
5186 vmcs_write64(EOI_EXIT_BITMAP0, 0);
5187 vmcs_write64(EOI_EXIT_BITMAP1, 0);
5188 vmcs_write64(EOI_EXIT_BITMAP2, 0);
5189 vmcs_write64(EOI_EXIT_BITMAP3, 0);
5191 vmcs_write16(GUEST_INTR_STATUS, 0);
5193 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
5194 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
5198 vmcs_write32(PLE_GAP, ple_gap);
5199 vmx->ple_window = ple_window;
5200 vmx->ple_window_dirty = true;
5203 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
5204 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
5205 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
5207 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
5208 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
5209 vmx_set_constant_host_state(vmx);
5210 #ifdef CONFIG_X86_64
5211 rdmsrl(MSR_FS_BASE, a);
5212 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
5213 rdmsrl(MSR_GS_BASE, a);
5214 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
5216 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
5217 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
5220 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
5221 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
5222 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
5223 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
5224 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
5226 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
5227 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
5229 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
5230 u32 index = vmx_msr_index[i];
5231 u32 data_low, data_high;
5234 if (rdmsr_safe(index, &data_low, &data_high) < 0)
5236 if (wrmsr_safe(index, data_low, data_high) < 0)
5238 vmx->guest_msrs[j].index = i;
5239 vmx->guest_msrs[j].data = 0;
5240 vmx->guest_msrs[j].mask = -1ull;
5245 vm_exit_controls_init(vmx, vmcs_config.vmexit_ctrl);
5247 /* 22.2.1, 20.8.1 */
5248 vm_entry_controls_init(vmx, vmcs_config.vmentry_ctrl);
5250 vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
5251 vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
5253 set_cr4_guest_host_mask(vmx);
5255 if (vmx_xsaves_supported())
5256 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
5259 ASSERT(vmx->pml_pg);
5260 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
5261 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5267 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
5269 struct vcpu_vmx *vmx = to_vmx(vcpu);
5270 struct msr_data apic_base_msr;
5273 vmx->rmode.vm86_active = 0;
5275 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
5276 kvm_set_cr8(vcpu, 0);
5279 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
5280 MSR_IA32_APICBASE_ENABLE;
5281 if (kvm_vcpu_is_reset_bsp(vcpu))
5282 apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
5283 apic_base_msr.host_initiated = true;
5284 kvm_set_apic_base(vcpu, &apic_base_msr);
5287 vmx_segment_cache_clear(vmx);
5289 seg_setup(VCPU_SREG_CS);
5290 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
5291 vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
5293 seg_setup(VCPU_SREG_DS);
5294 seg_setup(VCPU_SREG_ES);
5295 seg_setup(VCPU_SREG_FS);
5296 seg_setup(VCPU_SREG_GS);
5297 seg_setup(VCPU_SREG_SS);
5299 vmcs_write16(GUEST_TR_SELECTOR, 0);
5300 vmcs_writel(GUEST_TR_BASE, 0);
5301 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
5302 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
5304 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
5305 vmcs_writel(GUEST_LDTR_BASE, 0);
5306 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
5307 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
5310 vmcs_write32(GUEST_SYSENTER_CS, 0);
5311 vmcs_writel(GUEST_SYSENTER_ESP, 0);
5312 vmcs_writel(GUEST_SYSENTER_EIP, 0);
5313 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
5316 vmcs_writel(GUEST_RFLAGS, 0x02);
5317 kvm_rip_write(vcpu, 0xfff0);
5319 vmcs_writel(GUEST_GDTR_BASE, 0);
5320 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
5322 vmcs_writel(GUEST_IDTR_BASE, 0);
5323 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
5325 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
5326 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
5327 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
5331 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
5333 if (cpu_has_vmx_tpr_shadow() && !init_event) {
5334 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
5335 if (cpu_need_tpr_shadow(vcpu))
5336 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
5337 __pa(vcpu->arch.apic->regs));
5338 vmcs_write32(TPR_THRESHOLD, 0);
5341 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
5343 if (kvm_vcpu_apicv_active(vcpu))
5344 memset(&vmx->pi_desc, 0, sizeof(struct pi_desc));
5347 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
5349 cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
5350 vmx->vcpu.arch.cr0 = cr0;
5351 vmx_set_cr0(vcpu, cr0); /* enter rmode */
5352 vmx_set_cr4(vcpu, 0);
5353 vmx_set_efer(vcpu, 0);
5355 update_exception_bitmap(vcpu);
5357 vpid_sync_context(vmx->vpid);
5361 * In nested virtualization, check if L1 asked to exit on external interrupts.
5362 * For most existing hypervisors, this will always return true.
5364 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
5366 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5367 PIN_BASED_EXT_INTR_MASK;
5371 * In nested virtualization, check if L1 has set
5372 * VM_EXIT_ACK_INTR_ON_EXIT
5374 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
5376 return get_vmcs12(vcpu)->vm_exit_controls &
5377 VM_EXIT_ACK_INTR_ON_EXIT;
5380 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
5382 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5383 PIN_BASED_NMI_EXITING;
5386 static void enable_irq_window(struct kvm_vcpu *vcpu)
5388 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
5389 CPU_BASED_VIRTUAL_INTR_PENDING);
5392 static void enable_nmi_window(struct kvm_vcpu *vcpu)
5394 if (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
5395 enable_irq_window(vcpu);
5399 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
5400 CPU_BASED_VIRTUAL_NMI_PENDING);
5403 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
5405 struct vcpu_vmx *vmx = to_vmx(vcpu);
5407 int irq = vcpu->arch.interrupt.nr;
5409 trace_kvm_inj_virq(irq);
5411 ++vcpu->stat.irq_injections;
5412 if (vmx->rmode.vm86_active) {
5414 if (vcpu->arch.interrupt.soft)
5415 inc_eip = vcpu->arch.event_exit_inst_len;
5416 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
5417 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5420 intr = irq | INTR_INFO_VALID_MASK;
5421 if (vcpu->arch.interrupt.soft) {
5422 intr |= INTR_TYPE_SOFT_INTR;
5423 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
5424 vmx->vcpu.arch.event_exit_inst_len);
5426 intr |= INTR_TYPE_EXT_INTR;
5427 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
5430 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
5432 struct vcpu_vmx *vmx = to_vmx(vcpu);
5434 if (!is_guest_mode(vcpu)) {
5435 ++vcpu->stat.nmi_injections;
5436 vmx->nmi_known_unmasked = false;
5439 if (vmx->rmode.vm86_active) {
5440 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
5441 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5445 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
5446 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
5449 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
5451 if (to_vmx(vcpu)->nmi_known_unmasked)
5453 return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
5456 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
5458 struct vcpu_vmx *vmx = to_vmx(vcpu);
5460 vmx->nmi_known_unmasked = !masked;
5462 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5463 GUEST_INTR_STATE_NMI);
5465 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
5466 GUEST_INTR_STATE_NMI);
5469 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
5471 if (to_vmx(vcpu)->nested.nested_run_pending)
5474 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5475 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
5476 | GUEST_INTR_STATE_NMI));
5479 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
5481 return (!to_vmx(vcpu)->nested.nested_run_pending &&
5482 vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
5483 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5484 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
5487 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
5491 ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
5495 kvm->arch.tss_addr = addr;
5496 return init_rmode_tss(kvm);
5499 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
5504 * Update instruction length as we may reinject the exception
5505 * from user space while in guest debugging mode.
5507 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
5508 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5509 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5513 if (vcpu->guest_debug &
5514 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5531 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
5532 int vec, u32 err_code)
5535 * Instruction with address size override prefix opcode 0x67
5536 * Cause the #SS fault with 0 error code in VM86 mode.
5538 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
5539 if (emulate_instruction(vcpu, 0) == EMULATE_DONE) {
5540 if (vcpu->arch.halt_request) {
5541 vcpu->arch.halt_request = 0;
5542 return kvm_vcpu_halt(vcpu);
5550 * Forward all other exceptions that are valid in real mode.
5551 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
5552 * the required debugging infrastructure rework.
5554 kvm_queue_exception(vcpu, vec);
5559 * Trigger machine check on the host. We assume all the MSRs are already set up
5560 * by the CPU and that we still run on the same CPU as the MCE occurred on.
5561 * We pass a fake environment to the machine check handler because we want
5562 * the guest to be always treated like user space, no matter what context
5563 * it used internally.
5565 static void kvm_machine_check(void)
5567 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
5568 struct pt_regs regs = {
5569 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
5570 .flags = X86_EFLAGS_IF,
5573 do_machine_check(®s, 0);
5577 static int handle_machine_check(struct kvm_vcpu *vcpu)
5579 /* already handled by vcpu_run */
5583 static int handle_exception(struct kvm_vcpu *vcpu)
5585 struct vcpu_vmx *vmx = to_vmx(vcpu);
5586 struct kvm_run *kvm_run = vcpu->run;
5587 u32 intr_info, ex_no, error_code;
5588 unsigned long cr2, rip, dr6;
5590 enum emulation_result er;
5592 vect_info = vmx->idt_vectoring_info;
5593 intr_info = vmx->exit_intr_info;
5595 if (is_machine_check(intr_info))
5596 return handle_machine_check(vcpu);
5598 if (is_nmi(intr_info))
5599 return 1; /* already handled by vmx_vcpu_run() */
5601 if (is_invalid_opcode(intr_info)) {
5602 if (is_guest_mode(vcpu)) {
5603 kvm_queue_exception(vcpu, UD_VECTOR);
5606 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
5607 if (er != EMULATE_DONE)
5608 kvm_queue_exception(vcpu, UD_VECTOR);
5613 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
5614 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
5617 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
5618 * MMIO, it is better to report an internal error.
5619 * See the comments in vmx_handle_exit.
5621 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
5622 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
5623 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5624 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
5625 vcpu->run->internal.ndata = 3;
5626 vcpu->run->internal.data[0] = vect_info;
5627 vcpu->run->internal.data[1] = intr_info;
5628 vcpu->run->internal.data[2] = error_code;
5632 if (is_page_fault(intr_info)) {
5633 /* EPT won't cause page fault directly */
5635 cr2 = vmcs_readl(EXIT_QUALIFICATION);
5636 trace_kvm_page_fault(cr2, error_code);
5638 if (kvm_event_needs_reinjection(vcpu))
5639 kvm_mmu_unprotect_page_virt(vcpu, cr2);
5640 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
5643 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
5645 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
5646 return handle_rmode_exception(vcpu, ex_no, error_code);
5650 kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
5653 dr6 = vmcs_readl(EXIT_QUALIFICATION);
5654 if (!(vcpu->guest_debug &
5655 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
5656 vcpu->arch.dr6 &= ~15;
5657 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5658 if (!(dr6 & ~DR6_RESERVED)) /* icebp */
5659 skip_emulated_instruction(vcpu);
5661 kvm_queue_exception(vcpu, DB_VECTOR);
5664 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
5665 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
5669 * Update instruction length as we may reinject #BP from
5670 * user space while in guest debugging mode. Reading it for
5671 * #DB as well causes no harm, it is not used in that case.
5673 vmx->vcpu.arch.event_exit_inst_len =
5674 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5675 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5676 rip = kvm_rip_read(vcpu);
5677 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
5678 kvm_run->debug.arch.exception = ex_no;
5681 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
5682 kvm_run->ex.exception = ex_no;
5683 kvm_run->ex.error_code = error_code;
5689 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
5691 ++vcpu->stat.irq_exits;
5695 static int handle_triple_fault(struct kvm_vcpu *vcpu)
5697 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5701 static int handle_io(struct kvm_vcpu *vcpu)
5703 unsigned long exit_qualification;
5704 int size, in, string, ret;
5707 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5708 string = (exit_qualification & 16) != 0;
5709 in = (exit_qualification & 8) != 0;
5711 ++vcpu->stat.io_exits;
5714 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5716 port = exit_qualification >> 16;
5717 size = (exit_qualification & 7) + 1;
5719 ret = kvm_skip_emulated_instruction(vcpu);
5722 * TODO: we might be squashing a KVM_GUESTDBG_SINGLESTEP-triggered
5723 * KVM_EXIT_DEBUG here.
5725 return kvm_fast_pio_out(vcpu, size, port) && ret;
5729 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
5732 * Patch in the VMCALL instruction:
5734 hypercall[0] = 0x0f;
5735 hypercall[1] = 0x01;
5736 hypercall[2] = 0xc1;
5739 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
5740 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
5742 if (is_guest_mode(vcpu)) {
5743 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5744 unsigned long orig_val = val;
5747 * We get here when L2 changed cr0 in a way that did not change
5748 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
5749 * but did change L0 shadowed bits. So we first calculate the
5750 * effective cr0 value that L1 would like to write into the
5751 * hardware. It consists of the L2-owned bits from the new
5752 * value combined with the L1-owned bits from L1's guest_cr0.
5754 val = (val & ~vmcs12->cr0_guest_host_mask) |
5755 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
5757 if (!nested_guest_cr0_valid(vcpu, val))
5760 if (kvm_set_cr0(vcpu, val))
5762 vmcs_writel(CR0_READ_SHADOW, orig_val);
5765 if (to_vmx(vcpu)->nested.vmxon &&
5766 !nested_host_cr0_valid(vcpu, val))
5769 return kvm_set_cr0(vcpu, val);
5773 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
5775 if (is_guest_mode(vcpu)) {
5776 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5777 unsigned long orig_val = val;
5779 /* analogously to handle_set_cr0 */
5780 val = (val & ~vmcs12->cr4_guest_host_mask) |
5781 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
5782 if (kvm_set_cr4(vcpu, val))
5784 vmcs_writel(CR4_READ_SHADOW, orig_val);
5787 return kvm_set_cr4(vcpu, val);
5790 static int handle_cr(struct kvm_vcpu *vcpu)
5792 unsigned long exit_qualification, val;
5798 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5799 cr = exit_qualification & 15;
5800 reg = (exit_qualification >> 8) & 15;
5801 switch ((exit_qualification >> 4) & 3) {
5802 case 0: /* mov to cr */
5803 val = kvm_register_readl(vcpu, reg);
5804 trace_kvm_cr_write(cr, val);
5807 err = handle_set_cr0(vcpu, val);
5808 return kvm_complete_insn_gp(vcpu, err);
5810 err = kvm_set_cr3(vcpu, val);
5811 return kvm_complete_insn_gp(vcpu, err);
5813 err = handle_set_cr4(vcpu, val);
5814 return kvm_complete_insn_gp(vcpu, err);
5816 u8 cr8_prev = kvm_get_cr8(vcpu);
5818 err = kvm_set_cr8(vcpu, cr8);
5819 ret = kvm_complete_insn_gp(vcpu, err);
5820 if (lapic_in_kernel(vcpu))
5822 if (cr8_prev <= cr8)
5825 * TODO: we might be squashing a
5826 * KVM_GUESTDBG_SINGLESTEP-triggered
5827 * KVM_EXIT_DEBUG here.
5829 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
5835 WARN_ONCE(1, "Guest should always own CR0.TS");
5836 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
5837 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
5838 return kvm_skip_emulated_instruction(vcpu);
5839 case 1: /*mov from cr*/
5842 val = kvm_read_cr3(vcpu);
5843 kvm_register_write(vcpu, reg, val);
5844 trace_kvm_cr_read(cr, val);
5845 return kvm_skip_emulated_instruction(vcpu);
5847 val = kvm_get_cr8(vcpu);
5848 kvm_register_write(vcpu, reg, val);
5849 trace_kvm_cr_read(cr, val);
5850 return kvm_skip_emulated_instruction(vcpu);
5854 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5855 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
5856 kvm_lmsw(vcpu, val);
5858 return kvm_skip_emulated_instruction(vcpu);
5862 vcpu->run->exit_reason = 0;
5863 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
5864 (int)(exit_qualification >> 4) & 3, cr);
5868 static int handle_dr(struct kvm_vcpu *vcpu)
5870 unsigned long exit_qualification;
5873 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5874 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
5876 /* First, if DR does not exist, trigger UD */
5877 if (!kvm_require_dr(vcpu, dr))
5880 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
5881 if (!kvm_require_cpl(vcpu, 0))
5883 dr7 = vmcs_readl(GUEST_DR7);
5886 * As the vm-exit takes precedence over the debug trap, we
5887 * need to emulate the latter, either for the host or the
5888 * guest debugging itself.
5890 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5891 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
5892 vcpu->run->debug.arch.dr7 = dr7;
5893 vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
5894 vcpu->run->debug.arch.exception = DB_VECTOR;
5895 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
5898 vcpu->arch.dr6 &= ~15;
5899 vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
5900 kvm_queue_exception(vcpu, DB_VECTOR);
5905 if (vcpu->guest_debug == 0) {
5906 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
5907 CPU_BASED_MOV_DR_EXITING);
5910 * No more DR vmexits; force a reload of the debug registers
5911 * and reenter on this instruction. The next vmexit will
5912 * retrieve the full state of the debug registers.
5914 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
5918 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
5919 if (exit_qualification & TYPE_MOV_FROM_DR) {
5922 if (kvm_get_dr(vcpu, dr, &val))
5924 kvm_register_write(vcpu, reg, val);
5926 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
5929 return kvm_skip_emulated_instruction(vcpu);
5932 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
5934 return vcpu->arch.dr6;
5937 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
5941 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
5943 get_debugreg(vcpu->arch.db[0], 0);
5944 get_debugreg(vcpu->arch.db[1], 1);
5945 get_debugreg(vcpu->arch.db[2], 2);
5946 get_debugreg(vcpu->arch.db[3], 3);
5947 get_debugreg(vcpu->arch.dr6, 6);
5948 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
5950 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
5951 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL, CPU_BASED_MOV_DR_EXITING);
5954 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
5956 vmcs_writel(GUEST_DR7, val);
5959 static int handle_cpuid(struct kvm_vcpu *vcpu)
5961 return kvm_emulate_cpuid(vcpu);
5964 static int handle_rdmsr(struct kvm_vcpu *vcpu)
5966 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5967 struct msr_data msr_info;
5969 msr_info.index = ecx;
5970 msr_info.host_initiated = false;
5971 if (vmx_get_msr(vcpu, &msr_info)) {
5972 trace_kvm_msr_read_ex(ecx);
5973 kvm_inject_gp(vcpu, 0);
5977 trace_kvm_msr_read(ecx, msr_info.data);
5979 /* FIXME: handling of bits 32:63 of rax, rdx */
5980 vcpu->arch.regs[VCPU_REGS_RAX] = msr_info.data & -1u;
5981 vcpu->arch.regs[VCPU_REGS_RDX] = (msr_info.data >> 32) & -1u;
5982 return kvm_skip_emulated_instruction(vcpu);
5985 static int handle_wrmsr(struct kvm_vcpu *vcpu)
5987 struct msr_data msr;
5988 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5989 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
5990 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
5994 msr.host_initiated = false;
5995 if (kvm_set_msr(vcpu, &msr) != 0) {
5996 trace_kvm_msr_write_ex(ecx, data);
5997 kvm_inject_gp(vcpu, 0);
6001 trace_kvm_msr_write(ecx, data);
6002 return kvm_skip_emulated_instruction(vcpu);
6005 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
6007 kvm_apic_update_ppr(vcpu);
6011 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
6013 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6014 CPU_BASED_VIRTUAL_INTR_PENDING);
6016 kvm_make_request(KVM_REQ_EVENT, vcpu);
6018 ++vcpu->stat.irq_window_exits;
6022 static int handle_halt(struct kvm_vcpu *vcpu)
6024 return kvm_emulate_halt(vcpu);
6027 static int handle_vmcall(struct kvm_vcpu *vcpu)
6029 return kvm_emulate_hypercall(vcpu);
6032 static int handle_invd(struct kvm_vcpu *vcpu)
6034 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6037 static int handle_invlpg(struct kvm_vcpu *vcpu)
6039 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6041 kvm_mmu_invlpg(vcpu, exit_qualification);
6042 return kvm_skip_emulated_instruction(vcpu);
6045 static int handle_rdpmc(struct kvm_vcpu *vcpu)
6049 err = kvm_rdpmc(vcpu);
6050 return kvm_complete_insn_gp(vcpu, err);
6053 static int handle_wbinvd(struct kvm_vcpu *vcpu)
6055 return kvm_emulate_wbinvd(vcpu);
6058 static int handle_xsetbv(struct kvm_vcpu *vcpu)
6060 u64 new_bv = kvm_read_edx_eax(vcpu);
6061 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
6063 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
6064 return kvm_skip_emulated_instruction(vcpu);
6068 static int handle_xsaves(struct kvm_vcpu *vcpu)
6070 kvm_skip_emulated_instruction(vcpu);
6071 WARN(1, "this should never happen\n");
6075 static int handle_xrstors(struct kvm_vcpu *vcpu)
6077 kvm_skip_emulated_instruction(vcpu);
6078 WARN(1, "this should never happen\n");
6082 static int handle_apic_access(struct kvm_vcpu *vcpu)
6084 if (likely(fasteoi)) {
6085 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6086 int access_type, offset;
6088 access_type = exit_qualification & APIC_ACCESS_TYPE;
6089 offset = exit_qualification & APIC_ACCESS_OFFSET;
6091 * Sane guest uses MOV to write EOI, with written value
6092 * not cared. So make a short-circuit here by avoiding
6093 * heavy instruction emulation.
6095 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
6096 (offset == APIC_EOI)) {
6097 kvm_lapic_set_eoi(vcpu);
6098 return kvm_skip_emulated_instruction(vcpu);
6101 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6104 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
6106 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6107 int vector = exit_qualification & 0xff;
6109 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
6110 kvm_apic_set_eoi_accelerated(vcpu, vector);
6114 static int handle_apic_write(struct kvm_vcpu *vcpu)
6116 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6117 u32 offset = exit_qualification & 0xfff;
6119 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
6120 kvm_apic_write_nodecode(vcpu, offset);
6124 static int handle_task_switch(struct kvm_vcpu *vcpu)
6126 struct vcpu_vmx *vmx = to_vmx(vcpu);
6127 unsigned long exit_qualification;
6128 bool has_error_code = false;
6131 int reason, type, idt_v, idt_index;
6133 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
6134 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
6135 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
6137 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6139 reason = (u32)exit_qualification >> 30;
6140 if (reason == TASK_SWITCH_GATE && idt_v) {
6142 case INTR_TYPE_NMI_INTR:
6143 vcpu->arch.nmi_injected = false;
6144 vmx_set_nmi_mask(vcpu, true);
6146 case INTR_TYPE_EXT_INTR:
6147 case INTR_TYPE_SOFT_INTR:
6148 kvm_clear_interrupt_queue(vcpu);
6150 case INTR_TYPE_HARD_EXCEPTION:
6151 if (vmx->idt_vectoring_info &
6152 VECTORING_INFO_DELIVER_CODE_MASK) {
6153 has_error_code = true;
6155 vmcs_read32(IDT_VECTORING_ERROR_CODE);
6158 case INTR_TYPE_SOFT_EXCEPTION:
6159 kvm_clear_exception_queue(vcpu);
6165 tss_selector = exit_qualification;
6167 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
6168 type != INTR_TYPE_EXT_INTR &&
6169 type != INTR_TYPE_NMI_INTR))
6170 skip_emulated_instruction(vcpu);
6172 if (kvm_task_switch(vcpu, tss_selector,
6173 type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
6174 has_error_code, error_code) == EMULATE_FAIL) {
6175 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6176 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6177 vcpu->run->internal.ndata = 0;
6182 * TODO: What about debug traps on tss switch?
6183 * Are we supposed to inject them and update dr6?
6189 static int handle_ept_violation(struct kvm_vcpu *vcpu)
6191 unsigned long exit_qualification;
6195 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6197 if (is_guest_mode(vcpu)
6198 && !(exit_qualification & EPT_VIOLATION_GVA_TRANSLATED)) {
6200 * Fix up exit_qualification according to whether guest
6201 * page table accesses are reads or writes.
6203 u64 eptp = nested_ept_get_cr3(vcpu);
6204 if (!(eptp & VMX_EPT_AD_ENABLE_BIT))
6205 exit_qualification &= ~EPT_VIOLATION_ACC_WRITE;
6209 * EPT violation happened while executing iret from NMI,
6210 * "blocked by NMI" bit has to be set before next VM entry.
6211 * There are errata that may cause this bit to not be set:
6214 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
6215 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
6216 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
6218 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6219 trace_kvm_page_fault(gpa, exit_qualification);
6221 /* Is it a read fault? */
6222 error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
6223 ? PFERR_USER_MASK : 0;
6224 /* Is it a write fault? */
6225 error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
6226 ? PFERR_WRITE_MASK : 0;
6227 /* Is it a fetch fault? */
6228 error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
6229 ? PFERR_FETCH_MASK : 0;
6230 /* ept page table entry is present? */
6231 error_code |= (exit_qualification &
6232 (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
6233 EPT_VIOLATION_EXECUTABLE))
6234 ? PFERR_PRESENT_MASK : 0;
6236 vcpu->arch.gpa_available = true;
6237 vcpu->arch.exit_qualification = exit_qualification;
6239 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
6242 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
6247 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6248 if (!kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
6249 trace_kvm_fast_mmio(gpa);
6250 return kvm_skip_emulated_instruction(vcpu);
6253 ret = handle_mmio_page_fault(vcpu, gpa, true);
6254 vcpu->arch.gpa_available = true;
6255 if (likely(ret == RET_MMIO_PF_EMULATE))
6256 return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) ==
6259 if (unlikely(ret == RET_MMIO_PF_INVALID))
6260 return kvm_mmu_page_fault(vcpu, gpa, 0, NULL, 0);
6262 if (unlikely(ret == RET_MMIO_PF_RETRY))
6265 /* It is the real ept misconfig */
6268 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
6269 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
6274 static int handle_nmi_window(struct kvm_vcpu *vcpu)
6276 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6277 CPU_BASED_VIRTUAL_NMI_PENDING);
6278 ++vcpu->stat.nmi_window_exits;
6279 kvm_make_request(KVM_REQ_EVENT, vcpu);
6284 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
6286 struct vcpu_vmx *vmx = to_vmx(vcpu);
6287 enum emulation_result err = EMULATE_DONE;
6290 bool intr_window_requested;
6291 unsigned count = 130;
6293 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
6294 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
6296 while (vmx->emulation_required && count-- != 0) {
6297 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
6298 return handle_interrupt_window(&vmx->vcpu);
6300 if (test_bit(KVM_REQ_EVENT, &vcpu->requests))
6303 err = emulate_instruction(vcpu, EMULTYPE_NO_REEXECUTE);
6305 if (err == EMULATE_USER_EXIT) {
6306 ++vcpu->stat.mmio_exits;
6311 if (err != EMULATE_DONE) {
6312 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6313 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6314 vcpu->run->internal.ndata = 0;
6318 if (vcpu->arch.halt_request) {
6319 vcpu->arch.halt_request = 0;
6320 ret = kvm_vcpu_halt(vcpu);
6324 if (signal_pending(current))
6334 static int __grow_ple_window(int val)
6336 if (ple_window_grow < 1)
6339 val = min(val, ple_window_actual_max);
6341 if (ple_window_grow < ple_window)
6342 val *= ple_window_grow;
6344 val += ple_window_grow;
6349 static int __shrink_ple_window(int val, int modifier, int minimum)
6354 if (modifier < ple_window)
6359 return max(val, minimum);
6362 static void grow_ple_window(struct kvm_vcpu *vcpu)
6364 struct vcpu_vmx *vmx = to_vmx(vcpu);
6365 int old = vmx->ple_window;
6367 vmx->ple_window = __grow_ple_window(old);
6369 if (vmx->ple_window != old)
6370 vmx->ple_window_dirty = true;
6372 trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old);
6375 static void shrink_ple_window(struct kvm_vcpu *vcpu)
6377 struct vcpu_vmx *vmx = to_vmx(vcpu);
6378 int old = vmx->ple_window;
6380 vmx->ple_window = __shrink_ple_window(old,
6381 ple_window_shrink, ple_window);
6383 if (vmx->ple_window != old)
6384 vmx->ple_window_dirty = true;
6386 trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old);
6390 * ple_window_actual_max is computed to be one grow_ple_window() below
6391 * ple_window_max. (See __grow_ple_window for the reason.)
6392 * This prevents overflows, because ple_window_max is int.
6393 * ple_window_max effectively rounded down to a multiple of ple_window_grow in
6395 * ple_window_max is also prevented from setting vmx->ple_window < ple_window.
6397 static void update_ple_window_actual_max(void)
6399 ple_window_actual_max =
6400 __shrink_ple_window(max(ple_window_max, ple_window),
6401 ple_window_grow, INT_MIN);
6405 * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
6407 static void wakeup_handler(void)
6409 struct kvm_vcpu *vcpu;
6410 int cpu = smp_processor_id();
6412 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6413 list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
6414 blocked_vcpu_list) {
6415 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6417 if (pi_test_on(pi_desc) == 1)
6418 kvm_vcpu_kick(vcpu);
6420 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6423 void vmx_enable_tdp(void)
6425 kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
6426 enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
6427 enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
6428 0ull, VMX_EPT_EXECUTABLE_MASK,
6429 cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
6430 enable_ept_ad_bits ? 0ull : VMX_EPT_RWX_MASK);
6432 ept_set_mmio_spte_mask();
6436 static __init int hardware_setup(void)
6438 int r = -ENOMEM, i, msr;
6440 rdmsrl_safe(MSR_EFER, &host_efer);
6442 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
6443 kvm_define_shared_msr(i, vmx_msr_index[i]);
6445 for (i = 0; i < VMX_BITMAP_NR; i++) {
6446 vmx_bitmap[i] = (unsigned long *)__get_free_page(GFP_KERNEL);
6451 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
6452 memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
6453 memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
6456 * Allow direct access to the PC debug port (it is often used for I/O
6457 * delays, but the vmexits simply slow things down).
6459 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
6460 clear_bit(0x80, vmx_io_bitmap_a);
6462 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
6464 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
6465 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
6467 if (setup_vmcs_config(&vmcs_config) < 0) {
6472 if (boot_cpu_has(X86_FEATURE_NX))
6473 kvm_enable_efer_bits(EFER_NX);
6475 if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
6476 !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
6479 if (!cpu_has_vmx_shadow_vmcs())
6480 enable_shadow_vmcs = 0;
6481 if (enable_shadow_vmcs)
6482 init_vmcs_shadow_fields();
6484 if (!cpu_has_vmx_ept() ||
6485 !cpu_has_vmx_ept_4levels()) {
6487 enable_unrestricted_guest = 0;
6488 enable_ept_ad_bits = 0;
6491 if (!cpu_has_vmx_ept_ad_bits())
6492 enable_ept_ad_bits = 0;
6494 if (!cpu_has_vmx_unrestricted_guest())
6495 enable_unrestricted_guest = 0;
6497 if (!cpu_has_vmx_flexpriority())
6498 flexpriority_enabled = 0;
6501 * set_apic_access_page_addr() is used to reload apic access
6502 * page upon invalidation. No need to do anything if not
6503 * using the APIC_ACCESS_ADDR VMCS field.
6505 if (!flexpriority_enabled)
6506 kvm_x86_ops->set_apic_access_page_addr = NULL;
6508 if (!cpu_has_vmx_tpr_shadow())
6509 kvm_x86_ops->update_cr8_intercept = NULL;
6511 if (enable_ept && !cpu_has_vmx_ept_2m_page())
6512 kvm_disable_largepages();
6514 if (!cpu_has_vmx_ple())
6517 if (!cpu_has_vmx_apicv()) {
6519 kvm_x86_ops->sync_pir_to_irr = NULL;
6522 if (cpu_has_vmx_tsc_scaling()) {
6523 kvm_has_tsc_control = true;
6524 kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
6525 kvm_tsc_scaling_ratio_frac_bits = 48;
6528 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
6529 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
6530 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
6531 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
6532 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
6533 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
6534 vmx_disable_intercept_for_msr(MSR_IA32_BNDCFGS, true);
6536 memcpy(vmx_msr_bitmap_legacy_x2apic_apicv,
6537 vmx_msr_bitmap_legacy, PAGE_SIZE);
6538 memcpy(vmx_msr_bitmap_longmode_x2apic_apicv,
6539 vmx_msr_bitmap_longmode, PAGE_SIZE);
6540 memcpy(vmx_msr_bitmap_legacy_x2apic,
6541 vmx_msr_bitmap_legacy, PAGE_SIZE);
6542 memcpy(vmx_msr_bitmap_longmode_x2apic,
6543 vmx_msr_bitmap_longmode, PAGE_SIZE);
6545 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
6547 for (msr = 0x800; msr <= 0x8ff; msr++) {
6548 if (msr == 0x839 /* TMCCT */)
6550 vmx_disable_intercept_msr_x2apic(msr, MSR_TYPE_R, true);
6554 * TPR reads and writes can be virtualized even if virtual interrupt
6555 * delivery is not in use.
6557 vmx_disable_intercept_msr_x2apic(0x808, MSR_TYPE_W, true);
6558 vmx_disable_intercept_msr_x2apic(0x808, MSR_TYPE_R | MSR_TYPE_W, false);
6561 vmx_disable_intercept_msr_x2apic(0x80b, MSR_TYPE_W, true);
6563 vmx_disable_intercept_msr_x2apic(0x83f, MSR_TYPE_W, true);
6570 update_ple_window_actual_max();
6573 * Only enable PML when hardware supports PML feature, and both EPT
6574 * and EPT A/D bit features are enabled -- PML depends on them to work.
6576 if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
6580 kvm_x86_ops->slot_enable_log_dirty = NULL;
6581 kvm_x86_ops->slot_disable_log_dirty = NULL;
6582 kvm_x86_ops->flush_log_dirty = NULL;
6583 kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
6586 if (cpu_has_vmx_preemption_timer() && enable_preemption_timer) {
6589 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
6590 cpu_preemption_timer_multi =
6591 vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
6593 kvm_x86_ops->set_hv_timer = NULL;
6594 kvm_x86_ops->cancel_hv_timer = NULL;
6597 kvm_set_posted_intr_wakeup_handler(wakeup_handler);
6599 kvm_mce_cap_supported |= MCG_LMCE_P;
6601 return alloc_kvm_area();
6604 for (i = 0; i < VMX_BITMAP_NR; i++)
6605 free_page((unsigned long)vmx_bitmap[i]);
6610 static __exit void hardware_unsetup(void)
6614 for (i = 0; i < VMX_BITMAP_NR; i++)
6615 free_page((unsigned long)vmx_bitmap[i]);
6621 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
6622 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
6624 static int handle_pause(struct kvm_vcpu *vcpu)
6627 grow_ple_window(vcpu);
6629 kvm_vcpu_on_spin(vcpu);
6630 return kvm_skip_emulated_instruction(vcpu);
6633 static int handle_nop(struct kvm_vcpu *vcpu)
6635 return kvm_skip_emulated_instruction(vcpu);
6638 static int handle_mwait(struct kvm_vcpu *vcpu)
6640 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
6641 return handle_nop(vcpu);
6644 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
6649 static int handle_monitor(struct kvm_vcpu *vcpu)
6651 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
6652 return handle_nop(vcpu);
6656 * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
6657 * We could reuse a single VMCS for all the L2 guests, but we also want the
6658 * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
6659 * allows keeping them loaded on the processor, and in the future will allow
6660 * optimizations where prepare_vmcs02 doesn't need to set all the fields on
6661 * every entry if they never change.
6662 * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
6663 * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
6665 * The following functions allocate and free a vmcs02 in this pool.
6668 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
6669 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
6671 struct vmcs02_list *item;
6672 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6673 if (item->vmptr == vmx->nested.current_vmptr) {
6674 list_move(&item->list, &vmx->nested.vmcs02_pool);
6675 return &item->vmcs02;
6678 if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
6679 /* Recycle the least recently used VMCS. */
6680 item = list_last_entry(&vmx->nested.vmcs02_pool,
6681 struct vmcs02_list, list);
6682 item->vmptr = vmx->nested.current_vmptr;
6683 list_move(&item->list, &vmx->nested.vmcs02_pool);
6684 return &item->vmcs02;
6687 /* Create a new VMCS */
6688 item = kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
6691 item->vmcs02.vmcs = alloc_vmcs();
6692 item->vmcs02.shadow_vmcs = NULL;
6693 if (!item->vmcs02.vmcs) {
6697 loaded_vmcs_init(&item->vmcs02);
6698 item->vmptr = vmx->nested.current_vmptr;
6699 list_add(&(item->list), &(vmx->nested.vmcs02_pool));
6700 vmx->nested.vmcs02_num++;
6701 return &item->vmcs02;
6704 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
6705 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
6707 struct vmcs02_list *item;
6708 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6709 if (item->vmptr == vmptr) {
6710 free_loaded_vmcs(&item->vmcs02);
6711 list_del(&item->list);
6713 vmx->nested.vmcs02_num--;
6719 * Free all VMCSs saved for this vcpu, except the one pointed by
6720 * vmx->loaded_vmcs. We must be running L1, so vmx->loaded_vmcs
6721 * must be &vmx->vmcs01.
6723 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
6725 struct vmcs02_list *item, *n;
6727 WARN_ON(vmx->loaded_vmcs != &vmx->vmcs01);
6728 list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
6730 * Something will leak if the above WARN triggers. Better than
6733 if (vmx->loaded_vmcs == &item->vmcs02)
6736 free_loaded_vmcs(&item->vmcs02);
6737 list_del(&item->list);
6739 vmx->nested.vmcs02_num--;
6744 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
6745 * set the success or error code of an emulated VMX instruction, as specified
6746 * by Vol 2B, VMX Instruction Reference, "Conventions".
6748 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
6750 vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
6751 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6752 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
6755 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
6757 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6758 & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
6759 X86_EFLAGS_SF | X86_EFLAGS_OF))
6763 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
6764 u32 vm_instruction_error)
6766 if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
6768 * failValid writes the error number to the current VMCS, which
6769 * can't be done there isn't a current VMCS.
6771 nested_vmx_failInvalid(vcpu);
6774 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6775 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6776 X86_EFLAGS_SF | X86_EFLAGS_OF))
6778 get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
6780 * We don't need to force a shadow sync because
6781 * VM_INSTRUCTION_ERROR is not shadowed
6785 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
6787 /* TODO: not to reset guest simply here. */
6788 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6789 pr_debug_ratelimited("kvm: nested vmx abort, indicator %d\n", indicator);
6792 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
6794 struct vcpu_vmx *vmx =
6795 container_of(timer, struct vcpu_vmx, nested.preemption_timer);
6797 vmx->nested.preemption_timer_expired = true;
6798 kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
6799 kvm_vcpu_kick(&vmx->vcpu);
6801 return HRTIMER_NORESTART;
6805 * Decode the memory-address operand of a vmx instruction, as recorded on an
6806 * exit caused by such an instruction (run by a guest hypervisor).
6807 * On success, returns 0. When the operand is invalid, returns 1 and throws
6810 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
6811 unsigned long exit_qualification,
6812 u32 vmx_instruction_info, bool wr, gva_t *ret)
6816 struct kvm_segment s;
6819 * According to Vol. 3B, "Information for VM Exits Due to Instruction
6820 * Execution", on an exit, vmx_instruction_info holds most of the
6821 * addressing components of the operand. Only the displacement part
6822 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
6823 * For how an actual address is calculated from all these components,
6824 * refer to Vol. 1, "Operand Addressing".
6826 int scaling = vmx_instruction_info & 3;
6827 int addr_size = (vmx_instruction_info >> 7) & 7;
6828 bool is_reg = vmx_instruction_info & (1u << 10);
6829 int seg_reg = (vmx_instruction_info >> 15) & 7;
6830 int index_reg = (vmx_instruction_info >> 18) & 0xf;
6831 bool index_is_valid = !(vmx_instruction_info & (1u << 22));
6832 int base_reg = (vmx_instruction_info >> 23) & 0xf;
6833 bool base_is_valid = !(vmx_instruction_info & (1u << 27));
6836 kvm_queue_exception(vcpu, UD_VECTOR);
6840 /* Addr = segment_base + offset */
6841 /* offset = base + [index * scale] + displacement */
6842 off = exit_qualification; /* holds the displacement */
6844 off += kvm_register_read(vcpu, base_reg);
6846 off += kvm_register_read(vcpu, index_reg)<<scaling;
6847 vmx_get_segment(vcpu, &s, seg_reg);
6848 *ret = s.base + off;
6850 if (addr_size == 1) /* 32 bit */
6853 /* Checks for #GP/#SS exceptions. */
6855 if (is_long_mode(vcpu)) {
6856 /* Long mode: #GP(0)/#SS(0) if the memory address is in a
6857 * non-canonical form. This is the only check on the memory
6858 * destination for long mode!
6860 exn = is_noncanonical_address(*ret);
6861 } else if (is_protmode(vcpu)) {
6862 /* Protected mode: apply checks for segment validity in the
6864 * - segment type check (#GP(0) may be thrown)
6865 * - usability check (#GP(0)/#SS(0))
6866 * - limit check (#GP(0)/#SS(0))
6869 /* #GP(0) if the destination operand is located in a
6870 * read-only data segment or any code segment.
6872 exn = ((s.type & 0xa) == 0 || (s.type & 8));
6874 /* #GP(0) if the source operand is located in an
6875 * execute-only code segment
6877 exn = ((s.type & 0xa) == 8);
6879 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6882 /* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
6884 exn = (s.unusable != 0);
6885 /* Protected mode: #GP(0)/#SS(0) if the memory
6886 * operand is outside the segment limit.
6888 exn = exn || (off + sizeof(u64) > s.limit);
6891 kvm_queue_exception_e(vcpu,
6892 seg_reg == VCPU_SREG_SS ?
6893 SS_VECTOR : GP_VECTOR,
6902 * This function performs the various checks including
6903 * - if it's 4KB aligned
6904 * - No bits beyond the physical address width are set
6905 * - Returns 0 on success or else 1
6906 * (Intel SDM Section 30.3)
6908 static int nested_vmx_check_vmptr(struct kvm_vcpu *vcpu, int exit_reason,
6913 struct x86_exception e;
6915 struct vcpu_vmx *vmx = to_vmx(vcpu);
6916 int maxphyaddr = cpuid_maxphyaddr(vcpu);
6918 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
6919 vmcs_read32(VMX_INSTRUCTION_INFO), false, &gva))
6922 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
6923 sizeof(vmptr), &e)) {
6924 kvm_inject_page_fault(vcpu, &e);
6928 switch (exit_reason) {
6929 case EXIT_REASON_VMON:
6932 * The first 4 bytes of VMXON region contain the supported
6933 * VMCS revision identifier
6935 * Note - IA32_VMX_BASIC[48] will never be 1
6936 * for the nested case;
6937 * which replaces physical address width with 32
6940 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6941 nested_vmx_failInvalid(vcpu);
6942 return kvm_skip_emulated_instruction(vcpu);
6945 page = nested_get_page(vcpu, vmptr);
6947 nested_vmx_failInvalid(vcpu);
6948 return kvm_skip_emulated_instruction(vcpu);
6950 if (*(u32 *)kmap(page) != VMCS12_REVISION) {
6952 nested_release_page_clean(page);
6953 nested_vmx_failInvalid(vcpu);
6954 return kvm_skip_emulated_instruction(vcpu);
6957 nested_release_page_clean(page);
6958 vmx->nested.vmxon_ptr = vmptr;
6960 case EXIT_REASON_VMCLEAR:
6961 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6962 nested_vmx_failValid(vcpu,
6963 VMXERR_VMCLEAR_INVALID_ADDRESS);
6964 return kvm_skip_emulated_instruction(vcpu);
6967 if (vmptr == vmx->nested.vmxon_ptr) {
6968 nested_vmx_failValid(vcpu,
6969 VMXERR_VMCLEAR_VMXON_POINTER);
6970 return kvm_skip_emulated_instruction(vcpu);
6973 case EXIT_REASON_VMPTRLD:
6974 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6975 nested_vmx_failValid(vcpu,
6976 VMXERR_VMPTRLD_INVALID_ADDRESS);
6977 return kvm_skip_emulated_instruction(vcpu);
6980 if (vmptr == vmx->nested.vmxon_ptr) {
6981 nested_vmx_failValid(vcpu,
6982 VMXERR_VMPTRLD_VMXON_POINTER);
6983 return kvm_skip_emulated_instruction(vcpu);
6987 return 1; /* shouldn't happen */
6995 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
6997 struct vcpu_vmx *vmx = to_vmx(vcpu);
6998 struct vmcs *shadow_vmcs;
7000 if (cpu_has_vmx_msr_bitmap()) {
7001 vmx->nested.msr_bitmap =
7002 (unsigned long *)__get_free_page(GFP_KERNEL);
7003 if (!vmx->nested.msr_bitmap)
7004 goto out_msr_bitmap;
7007 vmx->nested.cached_vmcs12 = kmalloc(VMCS12_SIZE, GFP_KERNEL);
7008 if (!vmx->nested.cached_vmcs12)
7009 goto out_cached_vmcs12;
7011 if (enable_shadow_vmcs) {
7012 shadow_vmcs = alloc_vmcs();
7014 goto out_shadow_vmcs;
7015 /* mark vmcs as shadow */
7016 shadow_vmcs->revision_id |= (1u << 31);
7017 /* init shadow vmcs */
7018 vmcs_clear(shadow_vmcs);
7019 vmx->vmcs01.shadow_vmcs = shadow_vmcs;
7022 INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
7023 vmx->nested.vmcs02_num = 0;
7025 hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
7026 HRTIMER_MODE_REL_PINNED);
7027 vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
7029 vmx->nested.vmxon = true;
7033 kfree(vmx->nested.cached_vmcs12);
7036 free_page((unsigned long)vmx->nested.msr_bitmap);
7043 * Emulate the VMXON instruction.
7044 * Currently, we just remember that VMX is active, and do not save or even
7045 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
7046 * do not currently need to store anything in that guest-allocated memory
7047 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
7048 * argument is different from the VMXON pointer (which the spec says they do).
7050 static int handle_vmon(struct kvm_vcpu *vcpu)
7053 struct kvm_segment cs;
7054 struct vcpu_vmx *vmx = to_vmx(vcpu);
7055 const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
7056 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
7058 /* The Intel VMX Instruction Reference lists a bunch of bits that
7059 * are prerequisite to running VMXON, most notably cr4.VMXE must be
7060 * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
7061 * Otherwise, we should fail with #UD. We test these now:
7063 if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
7064 !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
7065 (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
7066 kvm_queue_exception(vcpu, UD_VECTOR);
7070 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
7071 if (is_long_mode(vcpu) && !cs.l) {
7072 kvm_queue_exception(vcpu, UD_VECTOR);
7076 if (vmx_get_cpl(vcpu)) {
7077 kvm_inject_gp(vcpu, 0);
7081 if (vmx->nested.vmxon) {
7082 nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
7083 return kvm_skip_emulated_instruction(vcpu);
7086 if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
7087 != VMXON_NEEDED_FEATURES) {
7088 kvm_inject_gp(vcpu, 0);
7092 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMON, NULL))
7095 ret = enter_vmx_operation(vcpu);
7099 nested_vmx_succeed(vcpu);
7100 return kvm_skip_emulated_instruction(vcpu);
7104 * Intel's VMX Instruction Reference specifies a common set of prerequisites
7105 * for running VMX instructions (except VMXON, whose prerequisites are
7106 * slightly different). It also specifies what exception to inject otherwise.
7108 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
7110 struct kvm_segment cs;
7111 struct vcpu_vmx *vmx = to_vmx(vcpu);
7113 if (!vmx->nested.vmxon) {
7114 kvm_queue_exception(vcpu, UD_VECTOR);
7118 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
7119 if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
7120 (is_long_mode(vcpu) && !cs.l)) {
7121 kvm_queue_exception(vcpu, UD_VECTOR);
7125 if (vmx_get_cpl(vcpu)) {
7126 kvm_inject_gp(vcpu, 0);
7133 static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
7135 if (vmx->nested.current_vmptr == -1ull)
7138 /* current_vmptr and current_vmcs12 are always set/reset together */
7139 if (WARN_ON(vmx->nested.current_vmcs12 == NULL))
7142 if (enable_shadow_vmcs) {
7143 /* copy to memory all shadowed fields in case
7144 they were modified */
7145 copy_shadow_to_vmcs12(vmx);
7146 vmx->nested.sync_shadow_vmcs = false;
7147 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
7148 SECONDARY_EXEC_SHADOW_VMCS);
7149 vmcs_write64(VMCS_LINK_POINTER, -1ull);
7151 vmx->nested.posted_intr_nv = -1;
7153 /* Flush VMCS12 to guest memory */
7154 memcpy(vmx->nested.current_vmcs12, vmx->nested.cached_vmcs12,
7157 kunmap(vmx->nested.current_vmcs12_page);
7158 nested_release_page(vmx->nested.current_vmcs12_page);
7159 vmx->nested.current_vmptr = -1ull;
7160 vmx->nested.current_vmcs12 = NULL;
7164 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
7165 * just stops using VMX.
7167 static void free_nested(struct vcpu_vmx *vmx)
7169 if (!vmx->nested.vmxon)
7172 vmx->nested.vmxon = false;
7173 free_vpid(vmx->nested.vpid02);
7174 nested_release_vmcs12(vmx);
7175 if (vmx->nested.msr_bitmap) {
7176 free_page((unsigned long)vmx->nested.msr_bitmap);
7177 vmx->nested.msr_bitmap = NULL;
7179 if (enable_shadow_vmcs) {
7180 vmcs_clear(vmx->vmcs01.shadow_vmcs);
7181 free_vmcs(vmx->vmcs01.shadow_vmcs);
7182 vmx->vmcs01.shadow_vmcs = NULL;
7184 kfree(vmx->nested.cached_vmcs12);
7185 /* Unpin physical memory we referred to in current vmcs02 */
7186 if (vmx->nested.apic_access_page) {
7187 nested_release_page(vmx->nested.apic_access_page);
7188 vmx->nested.apic_access_page = NULL;
7190 if (vmx->nested.virtual_apic_page) {
7191 nested_release_page(vmx->nested.virtual_apic_page);
7192 vmx->nested.virtual_apic_page = NULL;
7194 if (vmx->nested.pi_desc_page) {
7195 kunmap(vmx->nested.pi_desc_page);
7196 nested_release_page(vmx->nested.pi_desc_page);
7197 vmx->nested.pi_desc_page = NULL;
7198 vmx->nested.pi_desc = NULL;
7201 nested_free_all_saved_vmcss(vmx);
7204 /* Emulate the VMXOFF instruction */
7205 static int handle_vmoff(struct kvm_vcpu *vcpu)
7207 if (!nested_vmx_check_permission(vcpu))
7209 free_nested(to_vmx(vcpu));
7210 nested_vmx_succeed(vcpu);
7211 return kvm_skip_emulated_instruction(vcpu);
7214 /* Emulate the VMCLEAR instruction */
7215 static int handle_vmclear(struct kvm_vcpu *vcpu)
7217 struct vcpu_vmx *vmx = to_vmx(vcpu);
7221 if (!nested_vmx_check_permission(vcpu))
7224 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMCLEAR, &vmptr))
7227 if (vmptr == vmx->nested.current_vmptr)
7228 nested_release_vmcs12(vmx);
7230 kvm_vcpu_write_guest(vcpu,
7231 vmptr + offsetof(struct vmcs12, launch_state),
7232 &zero, sizeof(zero));
7234 nested_free_vmcs02(vmx, vmptr);
7236 nested_vmx_succeed(vcpu);
7237 return kvm_skip_emulated_instruction(vcpu);
7240 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
7242 /* Emulate the VMLAUNCH instruction */
7243 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
7245 return nested_vmx_run(vcpu, true);
7248 /* Emulate the VMRESUME instruction */
7249 static int handle_vmresume(struct kvm_vcpu *vcpu)
7252 return nested_vmx_run(vcpu, false);
7255 enum vmcs_field_type {
7256 VMCS_FIELD_TYPE_U16 = 0,
7257 VMCS_FIELD_TYPE_U64 = 1,
7258 VMCS_FIELD_TYPE_U32 = 2,
7259 VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
7262 static inline int vmcs_field_type(unsigned long field)
7264 if (0x1 & field) /* the *_HIGH fields are all 32 bit */
7265 return VMCS_FIELD_TYPE_U32;
7266 return (field >> 13) & 0x3 ;
7269 static inline int vmcs_field_readonly(unsigned long field)
7271 return (((field >> 10) & 0x3) == 1);
7275 * Read a vmcs12 field. Since these can have varying lengths and we return
7276 * one type, we chose the biggest type (u64) and zero-extend the return value
7277 * to that size. Note that the caller, handle_vmread, might need to use only
7278 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
7279 * 64-bit fields are to be returned).
7281 static inline int vmcs12_read_any(struct kvm_vcpu *vcpu,
7282 unsigned long field, u64 *ret)
7284 short offset = vmcs_field_to_offset(field);
7290 p = ((char *)(get_vmcs12(vcpu))) + offset;
7292 switch (vmcs_field_type(field)) {
7293 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7294 *ret = *((natural_width *)p);
7296 case VMCS_FIELD_TYPE_U16:
7299 case VMCS_FIELD_TYPE_U32:
7302 case VMCS_FIELD_TYPE_U64:
7312 static inline int vmcs12_write_any(struct kvm_vcpu *vcpu,
7313 unsigned long field, u64 field_value){
7314 short offset = vmcs_field_to_offset(field);
7315 char *p = ((char *) get_vmcs12(vcpu)) + offset;
7319 switch (vmcs_field_type(field)) {
7320 case VMCS_FIELD_TYPE_U16:
7321 *(u16 *)p = field_value;
7323 case VMCS_FIELD_TYPE_U32:
7324 *(u32 *)p = field_value;
7326 case VMCS_FIELD_TYPE_U64:
7327 *(u64 *)p = field_value;
7329 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7330 *(natural_width *)p = field_value;
7339 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
7342 unsigned long field;
7344 struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
7345 const unsigned long *fields = shadow_read_write_fields;
7346 const int num_fields = max_shadow_read_write_fields;
7350 vmcs_load(shadow_vmcs);
7352 for (i = 0; i < num_fields; i++) {
7354 switch (vmcs_field_type(field)) {
7355 case VMCS_FIELD_TYPE_U16:
7356 field_value = vmcs_read16(field);
7358 case VMCS_FIELD_TYPE_U32:
7359 field_value = vmcs_read32(field);
7361 case VMCS_FIELD_TYPE_U64:
7362 field_value = vmcs_read64(field);
7364 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7365 field_value = vmcs_readl(field);
7371 vmcs12_write_any(&vmx->vcpu, field, field_value);
7374 vmcs_clear(shadow_vmcs);
7375 vmcs_load(vmx->loaded_vmcs->vmcs);
7380 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
7382 const unsigned long *fields[] = {
7383 shadow_read_write_fields,
7384 shadow_read_only_fields
7386 const int max_fields[] = {
7387 max_shadow_read_write_fields,
7388 max_shadow_read_only_fields
7391 unsigned long field;
7392 u64 field_value = 0;
7393 struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
7395 vmcs_load(shadow_vmcs);
7397 for (q = 0; q < ARRAY_SIZE(fields); q++) {
7398 for (i = 0; i < max_fields[q]; i++) {
7399 field = fields[q][i];
7400 vmcs12_read_any(&vmx->vcpu, field, &field_value);
7402 switch (vmcs_field_type(field)) {
7403 case VMCS_FIELD_TYPE_U16:
7404 vmcs_write16(field, (u16)field_value);
7406 case VMCS_FIELD_TYPE_U32:
7407 vmcs_write32(field, (u32)field_value);
7409 case VMCS_FIELD_TYPE_U64:
7410 vmcs_write64(field, (u64)field_value);
7412 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7413 vmcs_writel(field, (long)field_value);
7422 vmcs_clear(shadow_vmcs);
7423 vmcs_load(vmx->loaded_vmcs->vmcs);
7427 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
7428 * used before) all generate the same failure when it is missing.
7430 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
7432 struct vcpu_vmx *vmx = to_vmx(vcpu);
7433 if (vmx->nested.current_vmptr == -1ull) {
7434 nested_vmx_failInvalid(vcpu);
7440 static int handle_vmread(struct kvm_vcpu *vcpu)
7442 unsigned long field;
7444 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7445 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7448 if (!nested_vmx_check_permission(vcpu))
7451 if (!nested_vmx_check_vmcs12(vcpu))
7452 return kvm_skip_emulated_instruction(vcpu);
7454 /* Decode instruction info and find the field to read */
7455 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7456 /* Read the field, zero-extended to a u64 field_value */
7457 if (vmcs12_read_any(vcpu, field, &field_value) < 0) {
7458 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7459 return kvm_skip_emulated_instruction(vcpu);
7462 * Now copy part of this value to register or memory, as requested.
7463 * Note that the number of bits actually copied is 32 or 64 depending
7464 * on the guest's mode (32 or 64 bit), not on the given field's length.
7466 if (vmx_instruction_info & (1u << 10)) {
7467 kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
7470 if (get_vmx_mem_address(vcpu, exit_qualification,
7471 vmx_instruction_info, true, &gva))
7473 /* _system ok, as nested_vmx_check_permission verified cpl=0 */
7474 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
7475 &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
7478 nested_vmx_succeed(vcpu);
7479 return kvm_skip_emulated_instruction(vcpu);
7483 static int handle_vmwrite(struct kvm_vcpu *vcpu)
7485 unsigned long field;
7487 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7488 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7489 /* The value to write might be 32 or 64 bits, depending on L1's long
7490 * mode, and eventually we need to write that into a field of several
7491 * possible lengths. The code below first zero-extends the value to 64
7492 * bit (field_value), and then copies only the appropriate number of
7493 * bits into the vmcs12 field.
7495 u64 field_value = 0;
7496 struct x86_exception e;
7498 if (!nested_vmx_check_permission(vcpu))
7501 if (!nested_vmx_check_vmcs12(vcpu))
7502 return kvm_skip_emulated_instruction(vcpu);
7504 if (vmx_instruction_info & (1u << 10))
7505 field_value = kvm_register_readl(vcpu,
7506 (((vmx_instruction_info) >> 3) & 0xf));
7508 if (get_vmx_mem_address(vcpu, exit_qualification,
7509 vmx_instruction_info, false, &gva))
7511 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
7512 &field_value, (is_64_bit_mode(vcpu) ? 8 : 4), &e)) {
7513 kvm_inject_page_fault(vcpu, &e);
7519 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7520 if (vmcs_field_readonly(field)) {
7521 nested_vmx_failValid(vcpu,
7522 VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
7523 return kvm_skip_emulated_instruction(vcpu);
7526 if (vmcs12_write_any(vcpu, field, field_value) < 0) {
7527 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7528 return kvm_skip_emulated_instruction(vcpu);
7531 nested_vmx_succeed(vcpu);
7532 return kvm_skip_emulated_instruction(vcpu);
7535 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
7537 vmx->nested.current_vmptr = vmptr;
7538 if (enable_shadow_vmcs) {
7539 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
7540 SECONDARY_EXEC_SHADOW_VMCS);
7541 vmcs_write64(VMCS_LINK_POINTER,
7542 __pa(vmx->vmcs01.shadow_vmcs));
7543 vmx->nested.sync_shadow_vmcs = true;
7547 /* Emulate the VMPTRLD instruction */
7548 static int handle_vmptrld(struct kvm_vcpu *vcpu)
7550 struct vcpu_vmx *vmx = to_vmx(vcpu);
7553 if (!nested_vmx_check_permission(vcpu))
7556 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMPTRLD, &vmptr))
7559 if (vmx->nested.current_vmptr != vmptr) {
7560 struct vmcs12 *new_vmcs12;
7562 page = nested_get_page(vcpu, vmptr);
7564 nested_vmx_failInvalid(vcpu);
7565 return kvm_skip_emulated_instruction(vcpu);
7567 new_vmcs12 = kmap(page);
7568 if (new_vmcs12->revision_id != VMCS12_REVISION) {
7570 nested_release_page_clean(page);
7571 nested_vmx_failValid(vcpu,
7572 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
7573 return kvm_skip_emulated_instruction(vcpu);
7576 nested_release_vmcs12(vmx);
7577 vmx->nested.current_vmcs12 = new_vmcs12;
7578 vmx->nested.current_vmcs12_page = page;
7580 * Load VMCS12 from guest memory since it is not already
7583 memcpy(vmx->nested.cached_vmcs12,
7584 vmx->nested.current_vmcs12, VMCS12_SIZE);
7585 set_current_vmptr(vmx, vmptr);
7588 nested_vmx_succeed(vcpu);
7589 return kvm_skip_emulated_instruction(vcpu);
7592 /* Emulate the VMPTRST instruction */
7593 static int handle_vmptrst(struct kvm_vcpu *vcpu)
7595 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7596 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7598 struct x86_exception e;
7600 if (!nested_vmx_check_permission(vcpu))
7603 if (get_vmx_mem_address(vcpu, exit_qualification,
7604 vmx_instruction_info, true, &vmcs_gva))
7606 /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
7607 if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
7608 (void *)&to_vmx(vcpu)->nested.current_vmptr,
7610 kvm_inject_page_fault(vcpu, &e);
7613 nested_vmx_succeed(vcpu);
7614 return kvm_skip_emulated_instruction(vcpu);
7617 /* Emulate the INVEPT instruction */
7618 static int handle_invept(struct kvm_vcpu *vcpu)
7620 struct vcpu_vmx *vmx = to_vmx(vcpu);
7621 u32 vmx_instruction_info, types;
7624 struct x86_exception e;
7629 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7630 SECONDARY_EXEC_ENABLE_EPT) ||
7631 !(vmx->nested.nested_vmx_ept_caps & VMX_EPT_INVEPT_BIT)) {
7632 kvm_queue_exception(vcpu, UD_VECTOR);
7636 if (!nested_vmx_check_permission(vcpu))
7639 if (!kvm_read_cr0_bits(vcpu, X86_CR0_PE)) {
7640 kvm_queue_exception(vcpu, UD_VECTOR);
7644 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7645 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7647 types = (vmx->nested.nested_vmx_ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
7649 if (type >= 32 || !(types & (1 << type))) {
7650 nested_vmx_failValid(vcpu,
7651 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7652 return kvm_skip_emulated_instruction(vcpu);
7655 /* According to the Intel VMX instruction reference, the memory
7656 * operand is read even if it isn't needed (e.g., for type==global)
7658 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7659 vmx_instruction_info, false, &gva))
7661 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
7662 sizeof(operand), &e)) {
7663 kvm_inject_page_fault(vcpu, &e);
7668 case VMX_EPT_EXTENT_GLOBAL:
7670 * TODO: track mappings and invalidate
7671 * single context requests appropriately
7673 case VMX_EPT_EXTENT_CONTEXT:
7674 kvm_mmu_sync_roots(vcpu);
7675 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
7676 nested_vmx_succeed(vcpu);
7683 return kvm_skip_emulated_instruction(vcpu);
7686 static int handle_invvpid(struct kvm_vcpu *vcpu)
7688 struct vcpu_vmx *vmx = to_vmx(vcpu);
7689 u32 vmx_instruction_info;
7690 unsigned long type, types;
7692 struct x86_exception e;
7695 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7696 SECONDARY_EXEC_ENABLE_VPID) ||
7697 !(vmx->nested.nested_vmx_vpid_caps & VMX_VPID_INVVPID_BIT)) {
7698 kvm_queue_exception(vcpu, UD_VECTOR);
7702 if (!nested_vmx_check_permission(vcpu))
7705 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7706 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7708 types = (vmx->nested.nested_vmx_vpid_caps &
7709 VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
7711 if (type >= 32 || !(types & (1 << type))) {
7712 nested_vmx_failValid(vcpu,
7713 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7714 return kvm_skip_emulated_instruction(vcpu);
7717 /* according to the intel vmx instruction reference, the memory
7718 * operand is read even if it isn't needed (e.g., for type==global)
7720 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7721 vmx_instruction_info, false, &gva))
7723 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vpid,
7725 kvm_inject_page_fault(vcpu, &e);
7730 case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
7731 case VMX_VPID_EXTENT_SINGLE_CONTEXT:
7732 case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
7734 nested_vmx_failValid(vcpu,
7735 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7736 return kvm_skip_emulated_instruction(vcpu);
7739 case VMX_VPID_EXTENT_ALL_CONTEXT:
7743 return kvm_skip_emulated_instruction(vcpu);
7746 __vmx_flush_tlb(vcpu, vmx->nested.vpid02);
7747 nested_vmx_succeed(vcpu);
7749 return kvm_skip_emulated_instruction(vcpu);
7752 static int handle_pml_full(struct kvm_vcpu *vcpu)
7754 unsigned long exit_qualification;
7756 trace_kvm_pml_full(vcpu->vcpu_id);
7758 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7761 * PML buffer FULL happened while executing iret from NMI,
7762 * "blocked by NMI" bit has to be set before next VM entry.
7764 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
7765 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
7766 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
7767 GUEST_INTR_STATE_NMI);
7770 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
7771 * here.., and there's no userspace involvement needed for PML.
7776 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
7778 kvm_lapic_expired_hv_timer(vcpu);
7783 * The exit handlers return 1 if the exit was handled fully and guest execution
7784 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
7785 * to be done to userspace and return 0.
7787 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
7788 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
7789 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
7790 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
7791 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
7792 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
7793 [EXIT_REASON_CR_ACCESS] = handle_cr,
7794 [EXIT_REASON_DR_ACCESS] = handle_dr,
7795 [EXIT_REASON_CPUID] = handle_cpuid,
7796 [EXIT_REASON_MSR_READ] = handle_rdmsr,
7797 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
7798 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
7799 [EXIT_REASON_HLT] = handle_halt,
7800 [EXIT_REASON_INVD] = handle_invd,
7801 [EXIT_REASON_INVLPG] = handle_invlpg,
7802 [EXIT_REASON_RDPMC] = handle_rdpmc,
7803 [EXIT_REASON_VMCALL] = handle_vmcall,
7804 [EXIT_REASON_VMCLEAR] = handle_vmclear,
7805 [EXIT_REASON_VMLAUNCH] = handle_vmlaunch,
7806 [EXIT_REASON_VMPTRLD] = handle_vmptrld,
7807 [EXIT_REASON_VMPTRST] = handle_vmptrst,
7808 [EXIT_REASON_VMREAD] = handle_vmread,
7809 [EXIT_REASON_VMRESUME] = handle_vmresume,
7810 [EXIT_REASON_VMWRITE] = handle_vmwrite,
7811 [EXIT_REASON_VMOFF] = handle_vmoff,
7812 [EXIT_REASON_VMON] = handle_vmon,
7813 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
7814 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
7815 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
7816 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
7817 [EXIT_REASON_WBINVD] = handle_wbinvd,
7818 [EXIT_REASON_XSETBV] = handle_xsetbv,
7819 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
7820 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
7821 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
7822 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
7823 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
7824 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait,
7825 [EXIT_REASON_MONITOR_TRAP_FLAG] = handle_monitor_trap,
7826 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor,
7827 [EXIT_REASON_INVEPT] = handle_invept,
7828 [EXIT_REASON_INVVPID] = handle_invvpid,
7829 [EXIT_REASON_XSAVES] = handle_xsaves,
7830 [EXIT_REASON_XRSTORS] = handle_xrstors,
7831 [EXIT_REASON_PML_FULL] = handle_pml_full,
7832 [EXIT_REASON_PREEMPTION_TIMER] = handle_preemption_timer,
7835 static const int kvm_vmx_max_exit_handlers =
7836 ARRAY_SIZE(kvm_vmx_exit_handlers);
7838 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
7839 struct vmcs12 *vmcs12)
7841 unsigned long exit_qualification;
7842 gpa_t bitmap, last_bitmap;
7847 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7848 return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
7850 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7852 port = exit_qualification >> 16;
7853 size = (exit_qualification & 7) + 1;
7855 last_bitmap = (gpa_t)-1;
7860 bitmap = vmcs12->io_bitmap_a;
7861 else if (port < 0x10000)
7862 bitmap = vmcs12->io_bitmap_b;
7865 bitmap += (port & 0x7fff) / 8;
7867 if (last_bitmap != bitmap)
7868 if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
7870 if (b & (1 << (port & 7)))
7875 last_bitmap = bitmap;
7882 * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
7883 * rather than handle it ourselves in L0. I.e., check whether L1 expressed
7884 * disinterest in the current event (read or write a specific MSR) by using an
7885 * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
7887 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
7888 struct vmcs12 *vmcs12, u32 exit_reason)
7890 u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
7893 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
7897 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
7898 * for the four combinations of read/write and low/high MSR numbers.
7899 * First we need to figure out which of the four to use:
7901 bitmap = vmcs12->msr_bitmap;
7902 if (exit_reason == EXIT_REASON_MSR_WRITE)
7904 if (msr_index >= 0xc0000000) {
7905 msr_index -= 0xc0000000;
7909 /* Then read the msr_index'th bit from this bitmap: */
7910 if (msr_index < 1024*8) {
7912 if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
7914 return 1 & (b >> (msr_index & 7));
7916 return true; /* let L1 handle the wrong parameter */
7920 * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
7921 * rather than handle it ourselves in L0. I.e., check if L1 wanted to
7922 * intercept (via guest_host_mask etc.) the current event.
7924 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
7925 struct vmcs12 *vmcs12)
7927 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7928 int cr = exit_qualification & 15;
7929 int reg = (exit_qualification >> 8) & 15;
7930 unsigned long val = kvm_register_readl(vcpu, reg);
7932 switch ((exit_qualification >> 4) & 3) {
7933 case 0: /* mov to cr */
7936 if (vmcs12->cr0_guest_host_mask &
7937 (val ^ vmcs12->cr0_read_shadow))
7941 if ((vmcs12->cr3_target_count >= 1 &&
7942 vmcs12->cr3_target_value0 == val) ||
7943 (vmcs12->cr3_target_count >= 2 &&
7944 vmcs12->cr3_target_value1 == val) ||
7945 (vmcs12->cr3_target_count >= 3 &&
7946 vmcs12->cr3_target_value2 == val) ||
7947 (vmcs12->cr3_target_count >= 4 &&
7948 vmcs12->cr3_target_value3 == val))
7950 if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
7954 if (vmcs12->cr4_guest_host_mask &
7955 (vmcs12->cr4_read_shadow ^ val))
7959 if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
7965 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
7966 (vmcs12->cr0_read_shadow & X86_CR0_TS))
7969 case 1: /* mov from cr */
7972 if (vmcs12->cpu_based_vm_exec_control &
7973 CPU_BASED_CR3_STORE_EXITING)
7977 if (vmcs12->cpu_based_vm_exec_control &
7978 CPU_BASED_CR8_STORE_EXITING)
7985 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
7986 * cr0. Other attempted changes are ignored, with no exit.
7988 if (vmcs12->cr0_guest_host_mask & 0xe &
7989 (val ^ vmcs12->cr0_read_shadow))
7991 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
7992 !(vmcs12->cr0_read_shadow & 0x1) &&
8001 * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
8002 * should handle it ourselves in L0 (and then continue L2). Only call this
8003 * when in is_guest_mode (L2).
8005 static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu)
8007 u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8008 struct vcpu_vmx *vmx = to_vmx(vcpu);
8009 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8010 u32 exit_reason = vmx->exit_reason;
8012 trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
8013 vmcs_readl(EXIT_QUALIFICATION),
8014 vmx->idt_vectoring_info,
8016 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8019 if (vmx->nested.nested_run_pending)
8022 if (unlikely(vmx->fail)) {
8023 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
8024 vmcs_read32(VM_INSTRUCTION_ERROR));
8028 switch (exit_reason) {
8029 case EXIT_REASON_EXCEPTION_NMI:
8030 if (is_nmi(intr_info))
8032 else if (is_page_fault(intr_info))
8034 else if (is_no_device(intr_info) &&
8035 !(vmcs12->guest_cr0 & X86_CR0_TS))
8037 else if (is_debug(intr_info) &&
8039 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
8041 else if (is_breakpoint(intr_info) &&
8042 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
8044 return vmcs12->exception_bitmap &
8045 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
8046 case EXIT_REASON_EXTERNAL_INTERRUPT:
8048 case EXIT_REASON_TRIPLE_FAULT:
8050 case EXIT_REASON_PENDING_INTERRUPT:
8051 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
8052 case EXIT_REASON_NMI_WINDOW:
8053 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
8054 case EXIT_REASON_TASK_SWITCH:
8056 case EXIT_REASON_CPUID:
8058 case EXIT_REASON_HLT:
8059 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
8060 case EXIT_REASON_INVD:
8062 case EXIT_REASON_INVLPG:
8063 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
8064 case EXIT_REASON_RDPMC:
8065 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
8066 case EXIT_REASON_RDRAND:
8067 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND);
8068 case EXIT_REASON_RDSEED:
8069 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED);
8070 case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
8071 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
8072 case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
8073 case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
8074 case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
8075 case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
8076 case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
8077 case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
8079 * VMX instructions trap unconditionally. This allows L1 to
8080 * emulate them for its L2 guest, i.e., allows 3-level nesting!
8083 case EXIT_REASON_CR_ACCESS:
8084 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
8085 case EXIT_REASON_DR_ACCESS:
8086 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
8087 case EXIT_REASON_IO_INSTRUCTION:
8088 return nested_vmx_exit_handled_io(vcpu, vmcs12);
8089 case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
8090 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
8091 case EXIT_REASON_MSR_READ:
8092 case EXIT_REASON_MSR_WRITE:
8093 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
8094 case EXIT_REASON_INVALID_STATE:
8096 case EXIT_REASON_MWAIT_INSTRUCTION:
8097 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
8098 case EXIT_REASON_MONITOR_TRAP_FLAG:
8099 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_TRAP_FLAG);
8100 case EXIT_REASON_MONITOR_INSTRUCTION:
8101 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
8102 case EXIT_REASON_PAUSE_INSTRUCTION:
8103 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
8104 nested_cpu_has2(vmcs12,
8105 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
8106 case EXIT_REASON_MCE_DURING_VMENTRY:
8108 case EXIT_REASON_TPR_BELOW_THRESHOLD:
8109 return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
8110 case EXIT_REASON_APIC_ACCESS:
8111 return nested_cpu_has2(vmcs12,
8112 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
8113 case EXIT_REASON_APIC_WRITE:
8114 case EXIT_REASON_EOI_INDUCED:
8115 /* apic_write and eoi_induced should exit unconditionally. */
8117 case EXIT_REASON_EPT_VIOLATION:
8119 * L0 always deals with the EPT violation. If nested EPT is
8120 * used, and the nested mmu code discovers that the address is
8121 * missing in the guest EPT table (EPT12), the EPT violation
8122 * will be injected with nested_ept_inject_page_fault()
8125 case EXIT_REASON_EPT_MISCONFIG:
8127 * L2 never uses directly L1's EPT, but rather L0's own EPT
8128 * table (shadow on EPT) or a merged EPT table that L0 built
8129 * (EPT on EPT). So any problems with the structure of the
8130 * table is L0's fault.
8133 case EXIT_REASON_WBINVD:
8134 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
8135 case EXIT_REASON_XSETBV:
8137 case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
8139 * This should never happen, since it is not possible to
8140 * set XSS to a non-zero value---neither in L1 nor in L2.
8141 * If if it were, XSS would have to be checked against
8142 * the XSS exit bitmap in vmcs12.
8144 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
8145 case EXIT_REASON_PREEMPTION_TIMER:
8152 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
8154 *info1 = vmcs_readl(EXIT_QUALIFICATION);
8155 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
8158 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
8161 __free_page(vmx->pml_pg);
8166 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
8168 struct vcpu_vmx *vmx = to_vmx(vcpu);
8172 pml_idx = vmcs_read16(GUEST_PML_INDEX);
8174 /* Do nothing if PML buffer is empty */
8175 if (pml_idx == (PML_ENTITY_NUM - 1))
8178 /* PML index always points to next available PML buffer entity */
8179 if (pml_idx >= PML_ENTITY_NUM)
8184 pml_buf = page_address(vmx->pml_pg);
8185 for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
8188 gpa = pml_buf[pml_idx];
8189 WARN_ON(gpa & (PAGE_SIZE - 1));
8190 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
8193 /* reset PML index */
8194 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
8198 * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
8199 * Called before reporting dirty_bitmap to userspace.
8201 static void kvm_flush_pml_buffers(struct kvm *kvm)
8204 struct kvm_vcpu *vcpu;
8206 * We only need to kick vcpu out of guest mode here, as PML buffer
8207 * is flushed at beginning of all VMEXITs, and it's obvious that only
8208 * vcpus running in guest are possible to have unflushed GPAs in PML
8211 kvm_for_each_vcpu(i, vcpu, kvm)
8212 kvm_vcpu_kick(vcpu);
8215 static void vmx_dump_sel(char *name, uint32_t sel)
8217 pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
8218 name, vmcs_read16(sel),
8219 vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
8220 vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
8221 vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
8224 static void vmx_dump_dtsel(char *name, uint32_t limit)
8226 pr_err("%s limit=0x%08x, base=0x%016lx\n",
8227 name, vmcs_read32(limit),
8228 vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
8231 static void dump_vmcs(void)
8233 u32 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
8234 u32 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
8235 u32 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
8236 u32 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
8237 u32 secondary_exec_control = 0;
8238 unsigned long cr4 = vmcs_readl(GUEST_CR4);
8239 u64 efer = vmcs_read64(GUEST_IA32_EFER);
8242 if (cpu_has_secondary_exec_ctrls())
8243 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8245 pr_err("*** Guest State ***\n");
8246 pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8247 vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
8248 vmcs_readl(CR0_GUEST_HOST_MASK));
8249 pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8250 cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
8251 pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
8252 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
8253 (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
8255 pr_err("PDPTR0 = 0x%016llx PDPTR1 = 0x%016llx\n",
8256 vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
8257 pr_err("PDPTR2 = 0x%016llx PDPTR3 = 0x%016llx\n",
8258 vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
8260 pr_err("RSP = 0x%016lx RIP = 0x%016lx\n",
8261 vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
8262 pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n",
8263 vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
8264 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8265 vmcs_readl(GUEST_SYSENTER_ESP),
8266 vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
8267 vmx_dump_sel("CS: ", GUEST_CS_SELECTOR);
8268 vmx_dump_sel("DS: ", GUEST_DS_SELECTOR);
8269 vmx_dump_sel("SS: ", GUEST_SS_SELECTOR);
8270 vmx_dump_sel("ES: ", GUEST_ES_SELECTOR);
8271 vmx_dump_sel("FS: ", GUEST_FS_SELECTOR);
8272 vmx_dump_sel("GS: ", GUEST_GS_SELECTOR);
8273 vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
8274 vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
8275 vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
8276 vmx_dump_sel("TR: ", GUEST_TR_SELECTOR);
8277 if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
8278 (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
8279 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
8280 efer, vmcs_read64(GUEST_IA32_PAT));
8281 pr_err("DebugCtl = 0x%016llx DebugExceptions = 0x%016lx\n",
8282 vmcs_read64(GUEST_IA32_DEBUGCTL),
8283 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
8284 if (vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
8285 pr_err("PerfGlobCtl = 0x%016llx\n",
8286 vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
8287 if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
8288 pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
8289 pr_err("Interruptibility = %08x ActivityState = %08x\n",
8290 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
8291 vmcs_read32(GUEST_ACTIVITY_STATE));
8292 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
8293 pr_err("InterruptStatus = %04x\n",
8294 vmcs_read16(GUEST_INTR_STATUS));
8296 pr_err("*** Host State ***\n");
8297 pr_err("RIP = 0x%016lx RSP = 0x%016lx\n",
8298 vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
8299 pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
8300 vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
8301 vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
8302 vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
8303 vmcs_read16(HOST_TR_SELECTOR));
8304 pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
8305 vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
8306 vmcs_readl(HOST_TR_BASE));
8307 pr_err("GDTBase=%016lx IDTBase=%016lx\n",
8308 vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
8309 pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
8310 vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
8311 vmcs_readl(HOST_CR4));
8312 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8313 vmcs_readl(HOST_IA32_SYSENTER_ESP),
8314 vmcs_read32(HOST_IA32_SYSENTER_CS),
8315 vmcs_readl(HOST_IA32_SYSENTER_EIP));
8316 if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
8317 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
8318 vmcs_read64(HOST_IA32_EFER),
8319 vmcs_read64(HOST_IA32_PAT));
8320 if (vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
8321 pr_err("PerfGlobCtl = 0x%016llx\n",
8322 vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
8324 pr_err("*** Control State ***\n");
8325 pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
8326 pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
8327 pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
8328 pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
8329 vmcs_read32(EXCEPTION_BITMAP),
8330 vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
8331 vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
8332 pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
8333 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8334 vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
8335 vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
8336 pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
8337 vmcs_read32(VM_EXIT_INTR_INFO),
8338 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8339 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
8340 pr_err(" reason=%08x qualification=%016lx\n",
8341 vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
8342 pr_err("IDTVectoring: info=%08x errcode=%08x\n",
8343 vmcs_read32(IDT_VECTORING_INFO_FIELD),
8344 vmcs_read32(IDT_VECTORING_ERROR_CODE));
8345 pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
8346 if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
8347 pr_err("TSC Multiplier = 0x%016llx\n",
8348 vmcs_read64(TSC_MULTIPLIER));
8349 if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW)
8350 pr_err("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
8351 if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
8352 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
8353 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
8354 pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
8355 n = vmcs_read32(CR3_TARGET_COUNT);
8356 for (i = 0; i + 1 < n; i += 4)
8357 pr_err("CR3 target%u=%016lx target%u=%016lx\n",
8358 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
8359 i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
8361 pr_err("CR3 target%u=%016lx\n",
8362 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
8363 if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
8364 pr_err("PLE Gap=%08x Window=%08x\n",
8365 vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
8366 if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
8367 pr_err("Virtual processor ID = 0x%04x\n",
8368 vmcs_read16(VIRTUAL_PROCESSOR_ID));
8372 * The guest has exited. See if we can fix it or if we need userspace
8375 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
8377 struct vcpu_vmx *vmx = to_vmx(vcpu);
8378 u32 exit_reason = vmx->exit_reason;
8379 u32 vectoring_info = vmx->idt_vectoring_info;
8381 trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
8382 vcpu->arch.gpa_available = false;
8385 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
8386 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
8387 * querying dirty_bitmap, we only need to kick all vcpus out of guest
8388 * mode as if vcpus is in root mode, the PML buffer must has been
8392 vmx_flush_pml_buffer(vcpu);
8394 /* If guest state is invalid, start emulating */
8395 if (vmx->emulation_required)
8396 return handle_invalid_guest_state(vcpu);
8398 if (is_guest_mode(vcpu) && nested_vmx_exit_handled(vcpu)) {
8399 nested_vmx_vmexit(vcpu, exit_reason,
8400 vmcs_read32(VM_EXIT_INTR_INFO),
8401 vmcs_readl(EXIT_QUALIFICATION));
8405 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
8407 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8408 vcpu->run->fail_entry.hardware_entry_failure_reason
8413 if (unlikely(vmx->fail)) {
8414 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8415 vcpu->run->fail_entry.hardware_entry_failure_reason
8416 = vmcs_read32(VM_INSTRUCTION_ERROR);
8422 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
8423 * delivery event since it indicates guest is accessing MMIO.
8424 * The vm-exit can be triggered again after return to guest that
8425 * will cause infinite loop.
8427 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
8428 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
8429 exit_reason != EXIT_REASON_EPT_VIOLATION &&
8430 exit_reason != EXIT_REASON_PML_FULL &&
8431 exit_reason != EXIT_REASON_TASK_SWITCH)) {
8432 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8433 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
8434 vcpu->run->internal.ndata = 2;
8435 vcpu->run->internal.data[0] = vectoring_info;
8436 vcpu->run->internal.data[1] = exit_reason;
8440 if (exit_reason < kvm_vmx_max_exit_handlers
8441 && kvm_vmx_exit_handlers[exit_reason])
8442 return kvm_vmx_exit_handlers[exit_reason](vcpu);
8444 vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n",
8446 kvm_queue_exception(vcpu, UD_VECTOR);
8451 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
8453 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8455 if (is_guest_mode(vcpu) &&
8456 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
8459 if (irr == -1 || tpr < irr) {
8460 vmcs_write32(TPR_THRESHOLD, 0);
8464 vmcs_write32(TPR_THRESHOLD, irr);
8467 static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
8469 u32 sec_exec_control;
8471 /* Postpone execution until vmcs01 is the current VMCS. */
8472 if (is_guest_mode(vcpu)) {
8473 to_vmx(vcpu)->nested.change_vmcs01_virtual_x2apic_mode = true;
8477 if (!cpu_has_vmx_virtualize_x2apic_mode())
8480 if (!cpu_need_tpr_shadow(vcpu))
8483 sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8486 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8487 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8489 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8490 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8491 vmx_flush_tlb_ept_only(vcpu);
8493 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
8495 vmx_set_msr_bitmap(vcpu);
8498 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
8500 struct vcpu_vmx *vmx = to_vmx(vcpu);
8503 * Currently we do not handle the nested case where L2 has an
8504 * APIC access page of its own; that page is still pinned.
8505 * Hence, we skip the case where the VCPU is in guest mode _and_
8506 * L1 prepared an APIC access page for L2.
8508 * For the case where L1 and L2 share the same APIC access page
8509 * (flexpriority=Y but SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES clear
8510 * in the vmcs12), this function will only update either the vmcs01
8511 * or the vmcs02. If the former, the vmcs02 will be updated by
8512 * prepare_vmcs02. If the latter, the vmcs01 will be updated in
8513 * the next L2->L1 exit.
8515 if (!is_guest_mode(vcpu) ||
8516 !nested_cpu_has2(get_vmcs12(&vmx->vcpu),
8517 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
8518 vmcs_write64(APIC_ACCESS_ADDR, hpa);
8519 vmx_flush_tlb_ept_only(vcpu);
8523 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
8531 status = vmcs_read16(GUEST_INTR_STATUS);
8533 if (max_isr != old) {
8535 status |= max_isr << 8;
8536 vmcs_write16(GUEST_INTR_STATUS, status);
8540 static void vmx_set_rvi(int vector)
8548 status = vmcs_read16(GUEST_INTR_STATUS);
8549 old = (u8)status & 0xff;
8550 if ((u8)vector != old) {
8552 status |= (u8)vector;
8553 vmcs_write16(GUEST_INTR_STATUS, status);
8557 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
8559 if (!is_guest_mode(vcpu)) {
8560 vmx_set_rvi(max_irr);
8568 * In guest mode. If a vmexit is needed, vmx_check_nested_events
8571 if (nested_exit_on_intr(vcpu))
8575 * Else, fall back to pre-APICv interrupt injection since L2
8576 * is run without virtual interrupt delivery.
8578 if (!kvm_event_needs_reinjection(vcpu) &&
8579 vmx_interrupt_allowed(vcpu)) {
8580 kvm_queue_interrupt(vcpu, max_irr, false);
8581 vmx_inject_irq(vcpu);
8585 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
8587 struct vcpu_vmx *vmx = to_vmx(vcpu);
8590 WARN_ON(!vcpu->arch.apicv_active);
8591 if (pi_test_on(&vmx->pi_desc)) {
8592 pi_clear_on(&vmx->pi_desc);
8594 * IOMMU can write to PIR.ON, so the barrier matters even on UP.
8595 * But on x86 this is just a compiler barrier anyway.
8597 smp_mb__after_atomic();
8598 max_irr = kvm_apic_update_irr(vcpu, vmx->pi_desc.pir);
8600 max_irr = kvm_lapic_find_highest_irr(vcpu);
8602 vmx_hwapic_irr_update(vcpu, max_irr);
8606 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
8608 if (!kvm_vcpu_apicv_active(vcpu))
8611 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
8612 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
8613 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
8614 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
8617 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
8619 struct vcpu_vmx *vmx = to_vmx(vcpu);
8621 pi_clear_on(&vmx->pi_desc);
8622 memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
8625 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
8629 if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
8630 || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
8633 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8634 exit_intr_info = vmx->exit_intr_info;
8636 /* Handle machine checks before interrupts are enabled */
8637 if (is_machine_check(exit_intr_info))
8638 kvm_machine_check();
8640 /* We need to handle NMIs before interrupts are enabled */
8641 if (is_nmi(exit_intr_info)) {
8642 kvm_before_handle_nmi(&vmx->vcpu);
8644 kvm_after_handle_nmi(&vmx->vcpu);
8648 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
8650 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8651 register void *__sp asm(_ASM_SP);
8653 if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
8654 == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
8655 unsigned int vector;
8656 unsigned long entry;
8658 struct vcpu_vmx *vmx = to_vmx(vcpu);
8659 #ifdef CONFIG_X86_64
8663 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
8664 desc = (gate_desc *)vmx->host_idt_base + vector;
8665 entry = gate_offset(*desc);
8667 #ifdef CONFIG_X86_64
8668 "mov %%" _ASM_SP ", %[sp]\n\t"
8669 "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
8674 __ASM_SIZE(push) " $%c[cs]\n\t"
8675 "call *%[entry]\n\t"
8677 #ifdef CONFIG_X86_64
8683 [ss]"i"(__KERNEL_DS),
8684 [cs]"i"(__KERNEL_CS)
8689 static bool vmx_has_high_real_mode_segbase(void)
8691 return enable_unrestricted_guest || emulate_invalid_guest_state;
8694 static bool vmx_mpx_supported(void)
8696 return (vmcs_config.vmexit_ctrl & VM_EXIT_CLEAR_BNDCFGS) &&
8697 (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_BNDCFGS);
8700 static bool vmx_xsaves_supported(void)
8702 return vmcs_config.cpu_based_2nd_exec_ctrl &
8703 SECONDARY_EXEC_XSAVES;
8706 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
8711 bool idtv_info_valid;
8713 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8715 if (vmx->nmi_known_unmasked)
8718 * Can't use vmx->exit_intr_info since we're not sure what
8719 * the exit reason is.
8721 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8722 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
8723 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
8725 * SDM 3: 27.7.1.2 (September 2008)
8726 * Re-set bit "block by NMI" before VM entry if vmexit caused by
8727 * a guest IRET fault.
8728 * SDM 3: 23.2.2 (September 2008)
8729 * Bit 12 is undefined in any of the following cases:
8730 * If the VM exit sets the valid bit in the IDT-vectoring
8731 * information field.
8732 * If the VM exit is due to a double fault.
8734 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
8735 vector != DF_VECTOR && !idtv_info_valid)
8736 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
8737 GUEST_INTR_STATE_NMI);
8739 vmx->nmi_known_unmasked =
8740 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
8741 & GUEST_INTR_STATE_NMI);
8744 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
8745 u32 idt_vectoring_info,
8746 int instr_len_field,
8747 int error_code_field)
8751 bool idtv_info_valid;
8753 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8755 vcpu->arch.nmi_injected = false;
8756 kvm_clear_exception_queue(vcpu);
8757 kvm_clear_interrupt_queue(vcpu);
8759 if (!idtv_info_valid)
8762 kvm_make_request(KVM_REQ_EVENT, vcpu);
8764 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
8765 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
8768 case INTR_TYPE_NMI_INTR:
8769 vcpu->arch.nmi_injected = true;
8771 * SDM 3: 27.7.1.2 (September 2008)
8772 * Clear bit "block by NMI" before VM entry if a NMI
8775 vmx_set_nmi_mask(vcpu, false);
8777 case INTR_TYPE_SOFT_EXCEPTION:
8778 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8780 case INTR_TYPE_HARD_EXCEPTION:
8781 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
8782 u32 err = vmcs_read32(error_code_field);
8783 kvm_requeue_exception_e(vcpu, vector, err);
8785 kvm_requeue_exception(vcpu, vector);
8787 case INTR_TYPE_SOFT_INTR:
8788 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8790 case INTR_TYPE_EXT_INTR:
8791 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
8798 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
8800 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
8801 VM_EXIT_INSTRUCTION_LEN,
8802 IDT_VECTORING_ERROR_CODE);
8805 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
8807 __vmx_complete_interrupts(vcpu,
8808 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8809 VM_ENTRY_INSTRUCTION_LEN,
8810 VM_ENTRY_EXCEPTION_ERROR_CODE);
8812 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
8815 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
8818 struct perf_guest_switch_msr *msrs;
8820 msrs = perf_guest_get_msrs(&nr_msrs);
8825 for (i = 0; i < nr_msrs; i++)
8826 if (msrs[i].host == msrs[i].guest)
8827 clear_atomic_switch_msr(vmx, msrs[i].msr);
8829 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
8833 static void vmx_arm_hv_timer(struct kvm_vcpu *vcpu)
8835 struct vcpu_vmx *vmx = to_vmx(vcpu);
8839 if (vmx->hv_deadline_tsc == -1)
8843 if (vmx->hv_deadline_tsc > tscl)
8844 /* sure to be 32 bit only because checked on set_hv_timer */
8845 delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
8846 cpu_preemption_timer_multi);
8850 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
8853 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
8855 struct vcpu_vmx *vmx = to_vmx(vcpu);
8856 unsigned long debugctlmsr, cr4;
8858 /* Don't enter VMX if guest state is invalid, let the exit handler
8859 start emulation until we arrive back to a valid state */
8860 if (vmx->emulation_required)
8863 if (vmx->ple_window_dirty) {
8864 vmx->ple_window_dirty = false;
8865 vmcs_write32(PLE_WINDOW, vmx->ple_window);
8868 if (vmx->nested.sync_shadow_vmcs) {
8869 copy_vmcs12_to_shadow(vmx);
8870 vmx->nested.sync_shadow_vmcs = false;
8873 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
8874 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
8875 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
8876 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
8878 cr4 = cr4_read_shadow();
8879 if (unlikely(cr4 != vmx->host_state.vmcs_host_cr4)) {
8880 vmcs_writel(HOST_CR4, cr4);
8881 vmx->host_state.vmcs_host_cr4 = cr4;
8884 /* When single-stepping over STI and MOV SS, we must clear the
8885 * corresponding interruptibility bits in the guest state. Otherwise
8886 * vmentry fails as it then expects bit 14 (BS) in pending debug
8887 * exceptions being set, but that's not correct for the guest debugging
8889 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8890 vmx_set_interrupt_shadow(vcpu, 0);
8892 if (vmx->guest_pkru_valid)
8893 __write_pkru(vmx->guest_pkru);
8895 atomic_switch_perf_msrs(vmx);
8896 debugctlmsr = get_debugctlmsr();
8898 vmx_arm_hv_timer(vcpu);
8900 vmx->__launched = vmx->loaded_vmcs->launched;
8902 /* Store host registers */
8903 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
8904 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
8905 "push %%" _ASM_CX " \n\t"
8906 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
8908 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
8909 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
8911 /* Reload cr2 if changed */
8912 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
8913 "mov %%cr2, %%" _ASM_DX " \n\t"
8914 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
8916 "mov %%" _ASM_AX", %%cr2 \n\t"
8918 /* Check if vmlaunch of vmresume is needed */
8919 "cmpl $0, %c[launched](%0) \n\t"
8920 /* Load guest registers. Don't clobber flags. */
8921 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
8922 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
8923 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
8924 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
8925 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
8926 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
8927 #ifdef CONFIG_X86_64
8928 "mov %c[r8](%0), %%r8 \n\t"
8929 "mov %c[r9](%0), %%r9 \n\t"
8930 "mov %c[r10](%0), %%r10 \n\t"
8931 "mov %c[r11](%0), %%r11 \n\t"
8932 "mov %c[r12](%0), %%r12 \n\t"
8933 "mov %c[r13](%0), %%r13 \n\t"
8934 "mov %c[r14](%0), %%r14 \n\t"
8935 "mov %c[r15](%0), %%r15 \n\t"
8937 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
8939 /* Enter guest mode */
8941 __ex(ASM_VMX_VMLAUNCH) "\n\t"
8943 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
8945 /* Save guest registers, load host registers, keep flags */
8946 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
8948 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
8949 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
8950 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
8951 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
8952 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
8953 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
8954 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
8955 #ifdef CONFIG_X86_64
8956 "mov %%r8, %c[r8](%0) \n\t"
8957 "mov %%r9, %c[r9](%0) \n\t"
8958 "mov %%r10, %c[r10](%0) \n\t"
8959 "mov %%r11, %c[r11](%0) \n\t"
8960 "mov %%r12, %c[r12](%0) \n\t"
8961 "mov %%r13, %c[r13](%0) \n\t"
8962 "mov %%r14, %c[r14](%0) \n\t"
8963 "mov %%r15, %c[r15](%0) \n\t"
8965 "mov %%cr2, %%" _ASM_AX " \n\t"
8966 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
8968 "pop %%" _ASM_BP "; pop %%" _ASM_DX " \n\t"
8969 "setbe %c[fail](%0) \n\t"
8970 ".pushsection .rodata \n\t"
8971 ".global vmx_return \n\t"
8972 "vmx_return: " _ASM_PTR " 2b \n\t"
8974 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
8975 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
8976 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
8977 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
8978 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
8979 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
8980 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
8981 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
8982 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
8983 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
8984 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
8985 #ifdef CONFIG_X86_64
8986 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
8987 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
8988 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
8989 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
8990 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
8991 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
8992 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
8993 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
8995 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
8996 [wordsize]"i"(sizeof(ulong))
8998 #ifdef CONFIG_X86_64
8999 , "rax", "rbx", "rdi", "rsi"
9000 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
9002 , "eax", "ebx", "edi", "esi"
9006 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
9008 update_debugctlmsr(debugctlmsr);
9010 #ifndef CONFIG_X86_64
9012 * The sysexit path does not restore ds/es, so we must set them to
9013 * a reasonable value ourselves.
9015 * We can't defer this to vmx_load_host_state() since that function
9016 * may be executed in interrupt context, which saves and restore segments
9017 * around it, nullifying its effect.
9019 loadsegment(ds, __USER_DS);
9020 loadsegment(es, __USER_DS);
9023 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
9024 | (1 << VCPU_EXREG_RFLAGS)
9025 | (1 << VCPU_EXREG_PDPTR)
9026 | (1 << VCPU_EXREG_SEGMENTS)
9027 | (1 << VCPU_EXREG_CR3));
9028 vcpu->arch.regs_dirty = 0;
9030 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
9032 vmx->loaded_vmcs->launched = 1;
9034 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
9037 * eager fpu is enabled if PKEY is supported and CR4 is switched
9038 * back on host, so it is safe to read guest PKRU from current
9041 if (boot_cpu_has(X86_FEATURE_OSPKE)) {
9042 vmx->guest_pkru = __read_pkru();
9043 if (vmx->guest_pkru != vmx->host_pkru) {
9044 vmx->guest_pkru_valid = true;
9045 __write_pkru(vmx->host_pkru);
9047 vmx->guest_pkru_valid = false;
9051 * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
9052 * we did not inject a still-pending event to L1 now because of
9053 * nested_run_pending, we need to re-enable this bit.
9055 if (vmx->nested.nested_run_pending)
9056 kvm_make_request(KVM_REQ_EVENT, vcpu);
9058 vmx->nested.nested_run_pending = 0;
9060 vmx_complete_atomic_exit(vmx);
9061 vmx_recover_nmi_blocking(vmx);
9062 vmx_complete_interrupts(vmx);
9065 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
9067 struct vcpu_vmx *vmx = to_vmx(vcpu);
9070 if (vmx->loaded_vmcs == vmcs)
9074 vmx->loaded_vmcs = vmcs;
9076 vmx_vcpu_load(vcpu, cpu);
9082 * Ensure that the current vmcs of the logical processor is the
9083 * vmcs01 of the vcpu before calling free_nested().
9085 static void vmx_free_vcpu_nested(struct kvm_vcpu *vcpu)
9087 struct vcpu_vmx *vmx = to_vmx(vcpu);
9090 r = vcpu_load(vcpu);
9092 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
9097 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
9099 struct vcpu_vmx *vmx = to_vmx(vcpu);
9102 vmx_destroy_pml_buffer(vmx);
9103 free_vpid(vmx->vpid);
9104 leave_guest_mode(vcpu);
9105 vmx_free_vcpu_nested(vcpu);
9106 free_loaded_vmcs(vmx->loaded_vmcs);
9107 kfree(vmx->guest_msrs);
9108 kvm_vcpu_uninit(vcpu);
9109 kmem_cache_free(kvm_vcpu_cache, vmx);
9112 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
9115 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
9119 return ERR_PTR(-ENOMEM);
9121 vmx->vpid = allocate_vpid();
9123 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
9130 * If PML is turned on, failure on enabling PML just results in failure
9131 * of creating the vcpu, therefore we can simplify PML logic (by
9132 * avoiding dealing with cases, such as enabling PML partially on vcpus
9133 * for the guest, etc.
9136 vmx->pml_pg = alloc_page(GFP_KERNEL | __GFP_ZERO);
9141 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
9142 BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
9145 if (!vmx->guest_msrs)
9148 vmx->loaded_vmcs = &vmx->vmcs01;
9149 vmx->loaded_vmcs->vmcs = alloc_vmcs();
9150 vmx->loaded_vmcs->shadow_vmcs = NULL;
9151 if (!vmx->loaded_vmcs->vmcs)
9153 loaded_vmcs_init(vmx->loaded_vmcs);
9156 vmx_vcpu_load(&vmx->vcpu, cpu);
9157 vmx->vcpu.cpu = cpu;
9158 err = vmx_vcpu_setup(vmx);
9159 vmx_vcpu_put(&vmx->vcpu);
9163 if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9164 err = alloc_apic_access_page(kvm);
9170 if (!kvm->arch.ept_identity_map_addr)
9171 kvm->arch.ept_identity_map_addr =
9172 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
9173 err = init_rmode_identity_map(kvm);
9179 nested_vmx_setup_ctls_msrs(vmx);
9180 vmx->nested.vpid02 = allocate_vpid();
9183 vmx->nested.posted_intr_nv = -1;
9184 vmx->nested.current_vmptr = -1ull;
9185 vmx->nested.current_vmcs12 = NULL;
9187 vmx->msr_ia32_feature_control_valid_bits = FEATURE_CONTROL_LOCKED;
9192 free_vpid(vmx->nested.vpid02);
9193 free_loaded_vmcs(vmx->loaded_vmcs);
9195 kfree(vmx->guest_msrs);
9197 vmx_destroy_pml_buffer(vmx);
9199 kvm_vcpu_uninit(&vmx->vcpu);
9201 free_vpid(vmx->vpid);
9202 kmem_cache_free(kvm_vcpu_cache, vmx);
9203 return ERR_PTR(err);
9206 static void __init vmx_check_processor_compat(void *rtn)
9208 struct vmcs_config vmcs_conf;
9211 if (setup_vmcs_config(&vmcs_conf) < 0)
9213 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
9214 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
9215 smp_processor_id());
9220 static int get_ept_level(void)
9222 return VMX_EPT_DEFAULT_GAW + 1;
9225 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
9230 /* For VT-d and EPT combination
9231 * 1. MMIO: always map as UC
9233 * a. VT-d without snooping control feature: can't guarantee the
9234 * result, try to trust guest.
9235 * b. VT-d with snooping control feature: snooping control feature of
9236 * VT-d engine can guarantee the cache correctness. Just set it
9237 * to WB to keep consistent with host. So the same as item 3.
9238 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
9239 * consistent with host MTRR
9242 cache = MTRR_TYPE_UNCACHABLE;
9246 if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
9247 ipat = VMX_EPT_IPAT_BIT;
9248 cache = MTRR_TYPE_WRBACK;
9252 if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
9253 ipat = VMX_EPT_IPAT_BIT;
9254 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
9255 cache = MTRR_TYPE_WRBACK;
9257 cache = MTRR_TYPE_UNCACHABLE;
9261 cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
9264 return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
9267 static int vmx_get_lpage_level(void)
9269 if (enable_ept && !cpu_has_vmx_ept_1g_page())
9270 return PT_DIRECTORY_LEVEL;
9272 /* For shadow and EPT supported 1GB page */
9273 return PT_PDPE_LEVEL;
9276 static void vmcs_set_secondary_exec_control(u32 new_ctl)
9279 * These bits in the secondary execution controls field
9280 * are dynamic, the others are mostly based on the hypervisor
9281 * architecture and the guest's CPUID. Do not touch the
9285 SECONDARY_EXEC_SHADOW_VMCS |
9286 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
9287 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9289 u32 cur_ctl = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
9291 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
9292 (new_ctl & ~mask) | (cur_ctl & mask));
9296 * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
9297 * (indicating "allowed-1") if they are supported in the guest's CPUID.
9299 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
9301 struct vcpu_vmx *vmx = to_vmx(vcpu);
9302 struct kvm_cpuid_entry2 *entry;
9304 vmx->nested.nested_vmx_cr0_fixed1 = 0xffffffff;
9305 vmx->nested.nested_vmx_cr4_fixed1 = X86_CR4_PCE;
9307 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do { \
9308 if (entry && (entry->_reg & (_cpuid_mask))) \
9309 vmx->nested.nested_vmx_cr4_fixed1 |= (_cr4_mask); \
9312 entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
9313 cr4_fixed1_update(X86_CR4_VME, edx, bit(X86_FEATURE_VME));
9314 cr4_fixed1_update(X86_CR4_PVI, edx, bit(X86_FEATURE_VME));
9315 cr4_fixed1_update(X86_CR4_TSD, edx, bit(X86_FEATURE_TSC));
9316 cr4_fixed1_update(X86_CR4_DE, edx, bit(X86_FEATURE_DE));
9317 cr4_fixed1_update(X86_CR4_PSE, edx, bit(X86_FEATURE_PSE));
9318 cr4_fixed1_update(X86_CR4_PAE, edx, bit(X86_FEATURE_PAE));
9319 cr4_fixed1_update(X86_CR4_MCE, edx, bit(X86_FEATURE_MCE));
9320 cr4_fixed1_update(X86_CR4_PGE, edx, bit(X86_FEATURE_PGE));
9321 cr4_fixed1_update(X86_CR4_OSFXSR, edx, bit(X86_FEATURE_FXSR));
9322 cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, bit(X86_FEATURE_XMM));
9323 cr4_fixed1_update(X86_CR4_VMXE, ecx, bit(X86_FEATURE_VMX));
9324 cr4_fixed1_update(X86_CR4_SMXE, ecx, bit(X86_FEATURE_SMX));
9325 cr4_fixed1_update(X86_CR4_PCIDE, ecx, bit(X86_FEATURE_PCID));
9326 cr4_fixed1_update(X86_CR4_OSXSAVE, ecx, bit(X86_FEATURE_XSAVE));
9328 entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
9329 cr4_fixed1_update(X86_CR4_FSGSBASE, ebx, bit(X86_FEATURE_FSGSBASE));
9330 cr4_fixed1_update(X86_CR4_SMEP, ebx, bit(X86_FEATURE_SMEP));
9331 cr4_fixed1_update(X86_CR4_SMAP, ebx, bit(X86_FEATURE_SMAP));
9332 cr4_fixed1_update(X86_CR4_PKE, ecx, bit(X86_FEATURE_PKU));
9333 /* TODO: Use X86_CR4_UMIP and X86_FEATURE_UMIP macros */
9334 cr4_fixed1_update(bit(11), ecx, bit(2));
9336 #undef cr4_fixed1_update
9339 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
9341 struct kvm_cpuid_entry2 *best;
9342 struct vcpu_vmx *vmx = to_vmx(vcpu);
9343 u32 secondary_exec_ctl = vmx_secondary_exec_control(vmx);
9345 if (vmx_rdtscp_supported()) {
9346 bool rdtscp_enabled = guest_cpuid_has_rdtscp(vcpu);
9347 if (!rdtscp_enabled)
9348 secondary_exec_ctl &= ~SECONDARY_EXEC_RDTSCP;
9352 vmx->nested.nested_vmx_secondary_ctls_high |=
9353 SECONDARY_EXEC_RDTSCP;
9355 vmx->nested.nested_vmx_secondary_ctls_high &=
9356 ~SECONDARY_EXEC_RDTSCP;
9360 /* Exposing INVPCID only when PCID is exposed */
9361 best = kvm_find_cpuid_entry(vcpu, 0x7, 0);
9362 if (vmx_invpcid_supported() &&
9363 (!best || !(best->ebx & bit(X86_FEATURE_INVPCID)) ||
9364 !guest_cpuid_has_pcid(vcpu))) {
9365 secondary_exec_ctl &= ~SECONDARY_EXEC_ENABLE_INVPCID;
9368 best->ebx &= ~bit(X86_FEATURE_INVPCID);
9371 if (cpu_has_secondary_exec_ctrls())
9372 vmcs_set_secondary_exec_control(secondary_exec_ctl);
9374 if (nested_vmx_allowed(vcpu))
9375 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
9376 FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
9378 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
9379 ~FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
9381 if (nested_vmx_allowed(vcpu))
9382 nested_vmx_cr_fixed1_bits_update(vcpu);
9385 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
9387 if (func == 1 && nested)
9388 entry->ecx |= bit(X86_FEATURE_VMX);
9391 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
9392 struct x86_exception *fault)
9394 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9397 if (fault->error_code & PFERR_RSVD_MASK)
9398 exit_reason = EXIT_REASON_EPT_MISCONFIG;
9400 exit_reason = EXIT_REASON_EPT_VIOLATION;
9401 nested_vmx_vmexit(vcpu, exit_reason, 0, vcpu->arch.exit_qualification);
9402 vmcs12->guest_physical_address = fault->address;
9405 /* Callbacks for nested_ept_init_mmu_context: */
9407 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
9409 /* return the page table to be shadowed - in our case, EPT12 */
9410 return get_vmcs12(vcpu)->ept_pointer;
9413 static int nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
9417 WARN_ON(mmu_is_nested(vcpu));
9418 eptp = nested_ept_get_cr3(vcpu);
9419 if ((eptp & VMX_EPT_AD_ENABLE_BIT) && !enable_ept_ad_bits)
9422 kvm_mmu_unload(vcpu);
9423 kvm_init_shadow_ept_mmu(vcpu,
9424 to_vmx(vcpu)->nested.nested_vmx_ept_caps &
9425 VMX_EPT_EXECUTE_ONLY_BIT,
9426 eptp & VMX_EPT_AD_ENABLE_BIT);
9427 vcpu->arch.mmu.set_cr3 = vmx_set_cr3;
9428 vcpu->arch.mmu.get_cr3 = nested_ept_get_cr3;
9429 vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
9431 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
9435 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
9437 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
9440 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
9443 bool inequality, bit;
9445 bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
9447 (error_code & vmcs12->page_fault_error_code_mask) !=
9448 vmcs12->page_fault_error_code_match;
9449 return inequality ^ bit;
9452 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
9453 struct x86_exception *fault)
9455 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9457 WARN_ON(!is_guest_mode(vcpu));
9459 if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code))
9460 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
9461 vmcs_read32(VM_EXIT_INTR_INFO),
9462 vmcs_readl(EXIT_QUALIFICATION));
9464 kvm_inject_page_fault(vcpu, fault);
9467 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
9468 struct vmcs12 *vmcs12);
9470 static void nested_get_vmcs12_pages(struct kvm_vcpu *vcpu,
9471 struct vmcs12 *vmcs12)
9473 struct vcpu_vmx *vmx = to_vmx(vcpu);
9476 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
9478 * Translate L1 physical address to host physical
9479 * address for vmcs02. Keep the page pinned, so this
9480 * physical address remains valid. We keep a reference
9481 * to it so we can release it later.
9483 if (vmx->nested.apic_access_page) /* shouldn't happen */
9484 nested_release_page(vmx->nested.apic_access_page);
9485 vmx->nested.apic_access_page =
9486 nested_get_page(vcpu, vmcs12->apic_access_addr);
9488 * If translation failed, no matter: This feature asks
9489 * to exit when accessing the given address, and if it
9490 * can never be accessed, this feature won't do
9493 if (vmx->nested.apic_access_page) {
9494 hpa = page_to_phys(vmx->nested.apic_access_page);
9495 vmcs_write64(APIC_ACCESS_ADDR, hpa);
9497 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
9498 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
9500 } else if (!(nested_cpu_has_virt_x2apic_mode(vmcs12)) &&
9501 cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9502 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
9503 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
9504 kvm_vcpu_reload_apic_access_page(vcpu);
9507 if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
9508 if (vmx->nested.virtual_apic_page) /* shouldn't happen */
9509 nested_release_page(vmx->nested.virtual_apic_page);
9510 vmx->nested.virtual_apic_page =
9511 nested_get_page(vcpu, vmcs12->virtual_apic_page_addr);
9514 * If translation failed, VM entry will fail because
9515 * prepare_vmcs02 set VIRTUAL_APIC_PAGE_ADDR to -1ull.
9516 * Failing the vm entry is _not_ what the processor
9517 * does but it's basically the only possibility we
9518 * have. We could still enter the guest if CR8 load
9519 * exits are enabled, CR8 store exits are enabled, and
9520 * virtualize APIC access is disabled; in this case
9521 * the processor would never use the TPR shadow and we
9522 * could simply clear the bit from the execution
9523 * control. But such a configuration is useless, so
9524 * let's keep the code simple.
9526 if (vmx->nested.virtual_apic_page) {
9527 hpa = page_to_phys(vmx->nested.virtual_apic_page);
9528 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, hpa);
9532 if (nested_cpu_has_posted_intr(vmcs12)) {
9533 if (vmx->nested.pi_desc_page) { /* shouldn't happen */
9534 kunmap(vmx->nested.pi_desc_page);
9535 nested_release_page(vmx->nested.pi_desc_page);
9537 vmx->nested.pi_desc_page =
9538 nested_get_page(vcpu, vmcs12->posted_intr_desc_addr);
9539 vmx->nested.pi_desc =
9540 (struct pi_desc *)kmap(vmx->nested.pi_desc_page);
9541 if (!vmx->nested.pi_desc) {
9542 nested_release_page_clean(vmx->nested.pi_desc_page);
9545 vmx->nested.pi_desc =
9546 (struct pi_desc *)((void *)vmx->nested.pi_desc +
9547 (unsigned long)(vmcs12->posted_intr_desc_addr &
9549 vmcs_write64(POSTED_INTR_DESC_ADDR,
9550 page_to_phys(vmx->nested.pi_desc_page) +
9551 (unsigned long)(vmcs12->posted_intr_desc_addr &
9554 if (cpu_has_vmx_msr_bitmap() &&
9555 nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS) &&
9556 nested_vmx_merge_msr_bitmap(vcpu, vmcs12))
9559 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
9560 CPU_BASED_USE_MSR_BITMAPS);
9563 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
9565 u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
9566 struct vcpu_vmx *vmx = to_vmx(vcpu);
9568 if (vcpu->arch.virtual_tsc_khz == 0)
9571 /* Make sure short timeouts reliably trigger an immediate vmexit.
9572 * hrtimer_start does not guarantee this. */
9573 if (preemption_timeout <= 1) {
9574 vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
9578 preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
9579 preemption_timeout *= 1000000;
9580 do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
9581 hrtimer_start(&vmx->nested.preemption_timer,
9582 ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
9585 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
9586 struct vmcs12 *vmcs12)
9591 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
9594 if (vmcs12_read_any(vcpu, MSR_BITMAP, &addr)) {
9598 maxphyaddr = cpuid_maxphyaddr(vcpu);
9600 if (!PAGE_ALIGNED(vmcs12->msr_bitmap) ||
9601 ((addr + PAGE_SIZE) >> maxphyaddr))
9608 * Merge L0's and L1's MSR bitmap, return false to indicate that
9609 * we do not use the hardware.
9611 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
9612 struct vmcs12 *vmcs12)
9616 unsigned long *msr_bitmap_l1;
9617 unsigned long *msr_bitmap_l0 = to_vmx(vcpu)->nested.msr_bitmap;
9619 /* This shortcut is ok because we support only x2APIC MSRs so far. */
9620 if (!nested_cpu_has_virt_x2apic_mode(vmcs12))
9623 page = nested_get_page(vcpu, vmcs12->msr_bitmap);
9626 msr_bitmap_l1 = (unsigned long *)kmap(page);
9628 memset(msr_bitmap_l0, 0xff, PAGE_SIZE);
9630 if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
9631 if (nested_cpu_has_apic_reg_virt(vmcs12))
9632 for (msr = 0x800; msr <= 0x8ff; msr++)
9633 nested_vmx_disable_intercept_for_msr(
9634 msr_bitmap_l1, msr_bitmap_l0,
9637 nested_vmx_disable_intercept_for_msr(
9638 msr_bitmap_l1, msr_bitmap_l0,
9639 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
9640 MSR_TYPE_R | MSR_TYPE_W);
9642 if (nested_cpu_has_vid(vmcs12)) {
9643 nested_vmx_disable_intercept_for_msr(
9644 msr_bitmap_l1, msr_bitmap_l0,
9645 APIC_BASE_MSR + (APIC_EOI >> 4),
9647 nested_vmx_disable_intercept_for_msr(
9648 msr_bitmap_l1, msr_bitmap_l0,
9649 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
9654 nested_release_page_clean(page);
9659 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
9660 struct vmcs12 *vmcs12)
9662 if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9663 !nested_cpu_has_apic_reg_virt(vmcs12) &&
9664 !nested_cpu_has_vid(vmcs12) &&
9665 !nested_cpu_has_posted_intr(vmcs12))
9669 * If virtualize x2apic mode is enabled,
9670 * virtualize apic access must be disabled.
9672 if (nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9673 nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
9677 * If virtual interrupt delivery is enabled,
9678 * we must exit on external interrupts.
9680 if (nested_cpu_has_vid(vmcs12) &&
9681 !nested_exit_on_intr(vcpu))
9685 * bits 15:8 should be zero in posted_intr_nv,
9686 * the descriptor address has been already checked
9687 * in nested_get_vmcs12_pages.
9689 if (nested_cpu_has_posted_intr(vmcs12) &&
9690 (!nested_cpu_has_vid(vmcs12) ||
9691 !nested_exit_intr_ack_set(vcpu) ||
9692 vmcs12->posted_intr_nv & 0xff00))
9695 /* tpr shadow is needed by all apicv features. */
9696 if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
9702 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
9703 unsigned long count_field,
9704 unsigned long addr_field)
9709 if (vmcs12_read_any(vcpu, count_field, &count) ||
9710 vmcs12_read_any(vcpu, addr_field, &addr)) {
9716 maxphyaddr = cpuid_maxphyaddr(vcpu);
9717 if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
9718 (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr) {
9719 pr_debug_ratelimited(
9720 "nVMX: invalid MSR switch (0x%lx, %d, %llu, 0x%08llx)",
9721 addr_field, maxphyaddr, count, addr);
9727 static int nested_vmx_check_msr_switch_controls(struct kvm_vcpu *vcpu,
9728 struct vmcs12 *vmcs12)
9730 if (vmcs12->vm_exit_msr_load_count == 0 &&
9731 vmcs12->vm_exit_msr_store_count == 0 &&
9732 vmcs12->vm_entry_msr_load_count == 0)
9733 return 0; /* Fast path */
9734 if (nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_LOAD_COUNT,
9735 VM_EXIT_MSR_LOAD_ADDR) ||
9736 nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_STORE_COUNT,
9737 VM_EXIT_MSR_STORE_ADDR) ||
9738 nested_vmx_check_msr_switch(vcpu, VM_ENTRY_MSR_LOAD_COUNT,
9739 VM_ENTRY_MSR_LOAD_ADDR))
9744 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
9745 struct vmx_msr_entry *e)
9747 /* x2APIC MSR accesses are not allowed */
9748 if (vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8)
9750 if (e->index == MSR_IA32_UCODE_WRITE || /* SDM Table 35-2 */
9751 e->index == MSR_IA32_UCODE_REV)
9753 if (e->reserved != 0)
9758 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
9759 struct vmx_msr_entry *e)
9761 if (e->index == MSR_FS_BASE ||
9762 e->index == MSR_GS_BASE ||
9763 e->index == MSR_IA32_SMM_MONITOR_CTL || /* SMM is not supported */
9764 nested_vmx_msr_check_common(vcpu, e))
9769 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
9770 struct vmx_msr_entry *e)
9772 if (e->index == MSR_IA32_SMBASE || /* SMM is not supported */
9773 nested_vmx_msr_check_common(vcpu, e))
9779 * Load guest's/host's msr at nested entry/exit.
9780 * return 0 for success, entry index for failure.
9782 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9785 struct vmx_msr_entry e;
9786 struct msr_data msr;
9788 msr.host_initiated = false;
9789 for (i = 0; i < count; i++) {
9790 if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
9792 pr_debug_ratelimited(
9793 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9794 __func__, i, gpa + i * sizeof(e));
9797 if (nested_vmx_load_msr_check(vcpu, &e)) {
9798 pr_debug_ratelimited(
9799 "%s check failed (%u, 0x%x, 0x%x)\n",
9800 __func__, i, e.index, e.reserved);
9803 msr.index = e.index;
9805 if (kvm_set_msr(vcpu, &msr)) {
9806 pr_debug_ratelimited(
9807 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9808 __func__, i, e.index, e.value);
9817 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9820 struct vmx_msr_entry e;
9822 for (i = 0; i < count; i++) {
9823 struct msr_data msr_info;
9824 if (kvm_vcpu_read_guest(vcpu,
9825 gpa + i * sizeof(e),
9826 &e, 2 * sizeof(u32))) {
9827 pr_debug_ratelimited(
9828 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9829 __func__, i, gpa + i * sizeof(e));
9832 if (nested_vmx_store_msr_check(vcpu, &e)) {
9833 pr_debug_ratelimited(
9834 "%s check failed (%u, 0x%x, 0x%x)\n",
9835 __func__, i, e.index, e.reserved);
9838 msr_info.host_initiated = false;
9839 msr_info.index = e.index;
9840 if (kvm_get_msr(vcpu, &msr_info)) {
9841 pr_debug_ratelimited(
9842 "%s cannot read MSR (%u, 0x%x)\n",
9843 __func__, i, e.index);
9846 if (kvm_vcpu_write_guest(vcpu,
9847 gpa + i * sizeof(e) +
9848 offsetof(struct vmx_msr_entry, value),
9849 &msr_info.data, sizeof(msr_info.data))) {
9850 pr_debug_ratelimited(
9851 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9852 __func__, i, e.index, msr_info.data);
9859 static bool nested_cr3_valid(struct kvm_vcpu *vcpu, unsigned long val)
9861 unsigned long invalid_mask;
9863 invalid_mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
9864 return (val & invalid_mask) == 0;
9868 * Load guest's/host's cr3 at nested entry/exit. nested_ept is true if we are
9869 * emulating VM entry into a guest with EPT enabled.
9870 * Returns 0 on success, 1 on failure. Invalid state exit qualification code
9871 * is assigned to entry_failure_code on failure.
9873 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3, bool nested_ept,
9874 u32 *entry_failure_code)
9876 if (cr3 != kvm_read_cr3(vcpu) || (!nested_ept && pdptrs_changed(vcpu))) {
9877 if (!nested_cr3_valid(vcpu, cr3)) {
9878 *entry_failure_code = ENTRY_FAIL_DEFAULT;
9883 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
9884 * must not be dereferenced.
9886 if (!is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu) &&
9888 if (!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)) {
9889 *entry_failure_code = ENTRY_FAIL_PDPTE;
9894 vcpu->arch.cr3 = cr3;
9895 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
9898 kvm_mmu_reset_context(vcpu);
9903 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
9904 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
9905 * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
9906 * guest in a way that will both be appropriate to L1's requests, and our
9907 * needs. In addition to modifying the active vmcs (which is vmcs02), this
9908 * function also has additional necessary side-effects, like setting various
9909 * vcpu->arch fields.
9910 * Returns 0 on success, 1 on failure. Invalid state exit qualification code
9911 * is assigned to entry_failure_code on failure.
9913 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
9914 bool from_vmentry, u32 *entry_failure_code)
9916 struct vcpu_vmx *vmx = to_vmx(vcpu);
9919 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
9920 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
9921 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
9922 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
9923 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
9924 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
9925 vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
9926 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
9927 vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
9928 vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
9929 vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
9930 vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
9931 vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
9932 vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
9933 vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
9934 vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
9935 vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
9936 vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
9937 vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
9938 vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
9939 vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
9940 vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
9941 vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
9942 vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
9943 vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
9944 vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
9945 vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
9946 vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
9947 vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
9948 vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
9949 vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
9950 vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
9951 vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
9952 vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
9953 vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
9954 vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
9957 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
9958 kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
9959 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
9961 kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
9962 vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
9965 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
9966 vmcs12->vm_entry_intr_info_field);
9967 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
9968 vmcs12->vm_entry_exception_error_code);
9969 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
9970 vmcs12->vm_entry_instruction_len);
9971 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
9972 vmcs12->guest_interruptibility_info);
9974 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
9976 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
9977 vmx_set_rflags(vcpu, vmcs12->guest_rflags);
9978 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
9979 vmcs12->guest_pending_dbg_exceptions);
9980 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
9981 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
9983 if (nested_cpu_has_xsaves(vmcs12))
9984 vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
9985 vmcs_write64(VMCS_LINK_POINTER, -1ull);
9987 exec_control = vmcs12->pin_based_vm_exec_control;
9989 /* Preemption timer setting is only taken from vmcs01. */
9990 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
9991 exec_control |= vmcs_config.pin_based_exec_ctrl;
9992 if (vmx->hv_deadline_tsc == -1)
9993 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
9995 /* Posted interrupts setting is only taken from vmcs12. */
9996 if (nested_cpu_has_posted_intr(vmcs12)) {
9998 * Note that we use L0's vector here and in
9999 * vmx_deliver_nested_posted_interrupt.
10001 vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
10002 vmx->nested.pi_pending = false;
10003 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
10005 exec_control &= ~PIN_BASED_POSTED_INTR;
10008 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
10010 vmx->nested.preemption_timer_expired = false;
10011 if (nested_cpu_has_preemption_timer(vmcs12))
10012 vmx_start_preemption_timer(vcpu);
10015 * Whether page-faults are trapped is determined by a combination of
10016 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
10017 * If enable_ept, L0 doesn't care about page faults and we should
10018 * set all of these to L1's desires. However, if !enable_ept, L0 does
10019 * care about (at least some) page faults, and because it is not easy
10020 * (if at all possible?) to merge L0 and L1's desires, we simply ask
10021 * to exit on each and every L2 page fault. This is done by setting
10022 * MASK=MATCH=0 and (see below) EB.PF=1.
10023 * Note that below we don't need special code to set EB.PF beyond the
10024 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
10025 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
10026 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
10028 * A problem with this approach (when !enable_ept) is that L1 may be
10029 * injected with more page faults than it asked for. This could have
10030 * caused problems, but in practice existing hypervisors don't care.
10031 * To fix this, we will need to emulate the PFEC checking (on the L1
10032 * page tables), using walk_addr(), when injecting PFs to L1.
10034 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
10035 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
10036 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
10037 enable_ept ? vmcs12->page_fault_error_code_match : 0);
10039 if (cpu_has_secondary_exec_ctrls()) {
10040 exec_control = vmx_secondary_exec_control(vmx);
10042 /* Take the following fields only from vmcs12 */
10043 exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
10044 SECONDARY_EXEC_RDTSCP |
10045 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
10046 SECONDARY_EXEC_APIC_REGISTER_VIRT);
10047 if (nested_cpu_has(vmcs12,
10048 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
10049 exec_control |= vmcs12->secondary_vm_exec_control;
10051 if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
10052 vmcs_write64(EOI_EXIT_BITMAP0,
10053 vmcs12->eoi_exit_bitmap0);
10054 vmcs_write64(EOI_EXIT_BITMAP1,
10055 vmcs12->eoi_exit_bitmap1);
10056 vmcs_write64(EOI_EXIT_BITMAP2,
10057 vmcs12->eoi_exit_bitmap2);
10058 vmcs_write64(EOI_EXIT_BITMAP3,
10059 vmcs12->eoi_exit_bitmap3);
10060 vmcs_write16(GUEST_INTR_STATUS,
10061 vmcs12->guest_intr_status);
10065 * Write an illegal value to APIC_ACCESS_ADDR. Later,
10066 * nested_get_vmcs12_pages will either fix it up or
10067 * remove the VM execution control.
10069 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
10070 vmcs_write64(APIC_ACCESS_ADDR, -1ull);
10072 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
10077 * Set host-state according to L0's settings (vmcs12 is irrelevant here)
10078 * Some constant fields are set here by vmx_set_constant_host_state().
10079 * Other fields are different per CPU, and will be set later when
10080 * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
10082 vmx_set_constant_host_state(vmx);
10085 * Set the MSR load/store lists to match L0's settings.
10087 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
10088 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
10089 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
10090 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
10091 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
10094 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
10095 * entry, but only if the current (host) sp changed from the value
10096 * we wrote last (vmx->host_rsp). This cache is no longer relevant
10097 * if we switch vmcs, and rather than hold a separate cache per vmcs,
10098 * here we just force the write to happen on entry.
10102 exec_control = vmx_exec_control(vmx); /* L0's desires */
10103 exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
10104 exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
10105 exec_control &= ~CPU_BASED_TPR_SHADOW;
10106 exec_control |= vmcs12->cpu_based_vm_exec_control;
10109 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR. Later, if
10110 * nested_get_vmcs12_pages can't fix it up, the illegal value
10111 * will result in a VM entry failure.
10113 if (exec_control & CPU_BASED_TPR_SHADOW) {
10114 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, -1ull);
10115 vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
10119 * Merging of IO bitmap not currently supported.
10120 * Rather, exit every time.
10122 exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
10123 exec_control |= CPU_BASED_UNCOND_IO_EXITING;
10125 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
10127 /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
10128 * bitwise-or of what L1 wants to trap for L2, and what we want to
10129 * trap. Note that CR0.TS also needs updating - we do this later.
10131 update_exception_bitmap(vcpu);
10132 vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
10133 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
10135 /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
10136 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
10137 * bits are further modified by vmx_set_efer() below.
10139 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
10141 /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
10142 * emulated by vmx_set_efer(), below.
10144 vm_entry_controls_init(vmx,
10145 (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER &
10146 ~VM_ENTRY_IA32E_MODE) |
10147 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
10149 if (from_vmentry &&
10150 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
10151 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
10152 vcpu->arch.pat = vmcs12->guest_ia32_pat;
10153 } else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
10154 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
10157 set_cr4_guest_host_mask(vmx);
10159 if (from_vmentry &&
10160 vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)
10161 vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
10163 if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
10164 vmcs_write64(TSC_OFFSET,
10165 vcpu->arch.tsc_offset + vmcs12->tsc_offset);
10167 vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
10168 if (kvm_has_tsc_control)
10169 decache_tsc_multiplier(vmx);
10173 * There is no direct mapping between vpid02 and vpid12, the
10174 * vpid02 is per-vCPU for L0 and reused while the value of
10175 * vpid12 is changed w/ one invvpid during nested vmentry.
10176 * The vpid12 is allocated by L1 for L2, so it will not
10177 * influence global bitmap(for vpid01 and vpid02 allocation)
10178 * even if spawn a lot of nested vCPUs.
10180 if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02) {
10181 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
10182 if (vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
10183 vmx->nested.last_vpid = vmcs12->virtual_processor_id;
10184 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->nested.vpid02);
10187 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
10188 vmx_flush_tlb(vcpu);
10193 if (nested_cpu_has_ept(vmcs12)) {
10194 if (nested_ept_init_mmu_context(vcpu)) {
10195 *entry_failure_code = ENTRY_FAIL_DEFAULT;
10198 } else if (nested_cpu_has2(vmcs12,
10199 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
10200 vmx_flush_tlb_ept_only(vcpu);
10204 * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
10205 * bits which we consider mandatory enabled.
10206 * The CR0_READ_SHADOW is what L2 should have expected to read given
10207 * the specifications by L1; It's not enough to take
10208 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
10209 * have more bits than L1 expected.
10211 vmx_set_cr0(vcpu, vmcs12->guest_cr0);
10212 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
10214 vmx_set_cr4(vcpu, vmcs12->guest_cr4);
10215 vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
10217 if (from_vmentry &&
10218 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
10219 vcpu->arch.efer = vmcs12->guest_ia32_efer;
10220 else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
10221 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10223 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10224 /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
10225 vmx_set_efer(vcpu, vcpu->arch.efer);
10227 /* Shadow page tables on either EPT or shadow page tables. */
10228 if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
10229 entry_failure_code))
10233 vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
10236 * L1 may access the L2's PDPTR, so save them to construct vmcs12
10239 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
10240 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
10241 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
10242 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
10245 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
10246 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
10250 static int check_vmentry_prereqs(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10252 struct vcpu_vmx *vmx = to_vmx(vcpu);
10254 if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
10255 vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT)
10256 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10258 if (nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12))
10259 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10261 if (nested_vmx_check_apicv_controls(vcpu, vmcs12))
10262 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10264 if (nested_vmx_check_msr_switch_controls(vcpu, vmcs12))
10265 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10267 if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
10268 vmx->nested.nested_vmx_procbased_ctls_low,
10269 vmx->nested.nested_vmx_procbased_ctls_high) ||
10270 !vmx_control_verify(vmcs12->secondary_vm_exec_control,
10271 vmx->nested.nested_vmx_secondary_ctls_low,
10272 vmx->nested.nested_vmx_secondary_ctls_high) ||
10273 !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
10274 vmx->nested.nested_vmx_pinbased_ctls_low,
10275 vmx->nested.nested_vmx_pinbased_ctls_high) ||
10276 !vmx_control_verify(vmcs12->vm_exit_controls,
10277 vmx->nested.nested_vmx_exit_ctls_low,
10278 vmx->nested.nested_vmx_exit_ctls_high) ||
10279 !vmx_control_verify(vmcs12->vm_entry_controls,
10280 vmx->nested.nested_vmx_entry_ctls_low,
10281 vmx->nested.nested_vmx_entry_ctls_high))
10282 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10284 if (!nested_host_cr0_valid(vcpu, vmcs12->host_cr0) ||
10285 !nested_host_cr4_valid(vcpu, vmcs12->host_cr4) ||
10286 !nested_cr3_valid(vcpu, vmcs12->host_cr3))
10287 return VMXERR_ENTRY_INVALID_HOST_STATE_FIELD;
10292 static int check_vmentry_postreqs(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10297 *exit_qual = ENTRY_FAIL_DEFAULT;
10299 if (!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0) ||
10300 !nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4))
10303 if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_SHADOW_VMCS) &&
10304 vmcs12->vmcs_link_pointer != -1ull) {
10305 *exit_qual = ENTRY_FAIL_VMCS_LINK_PTR;
10310 * If the load IA32_EFER VM-entry control is 1, the following checks
10311 * are performed on the field for the IA32_EFER MSR:
10312 * - Bits reserved in the IA32_EFER MSR must be 0.
10313 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
10314 * the IA-32e mode guest VM-exit control. It must also be identical
10315 * to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
10318 if (to_vmx(vcpu)->nested.nested_run_pending &&
10319 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
10320 ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
10321 if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
10322 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
10323 ((vmcs12->guest_cr0 & X86_CR0_PG) &&
10324 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME)))
10329 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
10330 * IA32_EFER MSR must be 0 in the field for that register. In addition,
10331 * the values of the LMA and LME bits in the field must each be that of
10332 * the host address-space size VM-exit control.
10334 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
10335 ia32e = (vmcs12->vm_exit_controls &
10336 VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
10337 if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
10338 ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
10339 ia32e != !!(vmcs12->host_ia32_efer & EFER_LME))
10346 static int enter_vmx_non_root_mode(struct kvm_vcpu *vcpu, bool from_vmentry)
10348 struct vcpu_vmx *vmx = to_vmx(vcpu);
10349 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
10350 struct loaded_vmcs *vmcs02;
10354 vmcs02 = nested_get_current_vmcs02(vmx);
10358 enter_guest_mode(vcpu);
10360 if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
10361 vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
10363 vmx_switch_vmcs(vcpu, vmcs02);
10364 vmx_segment_cache_clear(vmx);
10366 if (prepare_vmcs02(vcpu, vmcs12, from_vmentry, &exit_qual)) {
10367 leave_guest_mode(vcpu);
10368 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
10369 nested_vmx_entry_failure(vcpu, vmcs12,
10370 EXIT_REASON_INVALID_STATE, exit_qual);
10374 nested_get_vmcs12_pages(vcpu, vmcs12);
10376 msr_entry_idx = nested_vmx_load_msr(vcpu,
10377 vmcs12->vm_entry_msr_load_addr,
10378 vmcs12->vm_entry_msr_load_count);
10379 if (msr_entry_idx) {
10380 leave_guest_mode(vcpu);
10381 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
10382 nested_vmx_entry_failure(vcpu, vmcs12,
10383 EXIT_REASON_MSR_LOAD_FAIL, msr_entry_idx);
10387 vmcs12->launch_state = 1;
10390 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
10391 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
10392 * returned as far as L1 is concerned. It will only return (and set
10393 * the success flag) when L2 exits (see nested_vmx_vmexit()).
10399 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
10400 * for running an L2 nested guest.
10402 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
10404 struct vmcs12 *vmcs12;
10405 struct vcpu_vmx *vmx = to_vmx(vcpu);
10409 if (!nested_vmx_check_permission(vcpu))
10412 if (!nested_vmx_check_vmcs12(vcpu))
10415 vmcs12 = get_vmcs12(vcpu);
10417 if (enable_shadow_vmcs)
10418 copy_shadow_to_vmcs12(vmx);
10421 * The nested entry process starts with enforcing various prerequisites
10422 * on vmcs12 as required by the Intel SDM, and act appropriately when
10423 * they fail: As the SDM explains, some conditions should cause the
10424 * instruction to fail, while others will cause the instruction to seem
10425 * to succeed, but return an EXIT_REASON_INVALID_STATE.
10426 * To speed up the normal (success) code path, we should avoid checking
10427 * for misconfigurations which will anyway be caught by the processor
10428 * when using the merged vmcs02.
10430 if (vmcs12->launch_state == launch) {
10431 nested_vmx_failValid(vcpu,
10432 launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
10433 : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
10437 ret = check_vmentry_prereqs(vcpu, vmcs12);
10439 nested_vmx_failValid(vcpu, ret);
10444 * After this point, the trap flag no longer triggers a singlestep trap
10445 * on the vm entry instructions; don't call kvm_skip_emulated_instruction.
10446 * This is not 100% correct; for performance reasons, we delegate most
10447 * of the checks on host state to the processor. If those fail,
10448 * the singlestep trap is missed.
10450 skip_emulated_instruction(vcpu);
10452 ret = check_vmentry_postreqs(vcpu, vmcs12, &exit_qual);
10454 nested_vmx_entry_failure(vcpu, vmcs12,
10455 EXIT_REASON_INVALID_STATE, exit_qual);
10460 * We're finally done with prerequisite checking, and can start with
10461 * the nested entry.
10464 ret = enter_vmx_non_root_mode(vcpu, true);
10468 if (vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT)
10469 return kvm_vcpu_halt(vcpu);
10471 vmx->nested.nested_run_pending = 1;
10476 return kvm_skip_emulated_instruction(vcpu);
10480 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
10481 * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
10482 * This function returns the new value we should put in vmcs12.guest_cr0.
10483 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
10484 * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
10485 * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
10486 * didn't trap the bit, because if L1 did, so would L0).
10487 * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have
10488 * been modified by L2, and L1 knows it. So just leave the old value of
10489 * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
10490 * isn't relevant, because if L0 traps this bit it can set it to anything.
10491 * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
10492 * changed these bits, and therefore they need to be updated, but L0
10493 * didn't necessarily allow them to be changed in GUEST_CR0 - and rather
10494 * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
10496 static inline unsigned long
10497 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10500 /*1*/ (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
10501 /*2*/ (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
10502 /*3*/ (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
10503 vcpu->arch.cr0_guest_owned_bits));
10506 static inline unsigned long
10507 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10510 /*1*/ (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
10511 /*2*/ (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
10512 /*3*/ (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
10513 vcpu->arch.cr4_guest_owned_bits));
10516 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
10517 struct vmcs12 *vmcs12)
10522 if (vcpu->arch.exception.pending && vcpu->arch.exception.reinject) {
10523 nr = vcpu->arch.exception.nr;
10524 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
10526 if (kvm_exception_is_soft(nr)) {
10527 vmcs12->vm_exit_instruction_len =
10528 vcpu->arch.event_exit_inst_len;
10529 idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
10531 idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
10533 if (vcpu->arch.exception.has_error_code) {
10534 idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
10535 vmcs12->idt_vectoring_error_code =
10536 vcpu->arch.exception.error_code;
10539 vmcs12->idt_vectoring_info_field = idt_vectoring;
10540 } else if (vcpu->arch.nmi_injected) {
10541 vmcs12->idt_vectoring_info_field =
10542 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
10543 } else if (vcpu->arch.interrupt.pending) {
10544 nr = vcpu->arch.interrupt.nr;
10545 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
10547 if (vcpu->arch.interrupt.soft) {
10548 idt_vectoring |= INTR_TYPE_SOFT_INTR;
10549 vmcs12->vm_entry_instruction_len =
10550 vcpu->arch.event_exit_inst_len;
10552 idt_vectoring |= INTR_TYPE_EXT_INTR;
10554 vmcs12->idt_vectoring_info_field = idt_vectoring;
10558 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
10560 struct vcpu_vmx *vmx = to_vmx(vcpu);
10562 if (vcpu->arch.exception.pending ||
10563 vcpu->arch.nmi_injected ||
10564 vcpu->arch.interrupt.pending)
10567 if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
10568 vmx->nested.preemption_timer_expired) {
10569 if (vmx->nested.nested_run_pending)
10571 nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
10575 if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
10576 if (vmx->nested.nested_run_pending)
10578 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
10579 NMI_VECTOR | INTR_TYPE_NMI_INTR |
10580 INTR_INFO_VALID_MASK, 0);
10582 * The NMI-triggered VM exit counts as injection:
10583 * clear this one and block further NMIs.
10585 vcpu->arch.nmi_pending = 0;
10586 vmx_set_nmi_mask(vcpu, true);
10590 if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
10591 nested_exit_on_intr(vcpu)) {
10592 if (vmx->nested.nested_run_pending)
10594 nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
10598 vmx_complete_nested_posted_interrupt(vcpu);
10602 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
10604 ktime_t remaining =
10605 hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
10608 if (ktime_to_ns(remaining) <= 0)
10611 value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
10612 do_div(value, 1000000);
10613 return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
10617 * Update the guest state fields of vmcs12 to reflect changes that
10618 * occurred while L2 was running. (The "IA-32e mode guest" bit of the
10619 * VM-entry controls is also updated, since this is really a guest
10622 static void sync_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10624 vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
10625 vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
10627 vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
10628 vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
10629 vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
10631 vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
10632 vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
10633 vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
10634 vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
10635 vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
10636 vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
10637 vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
10638 vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
10639 vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
10640 vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
10641 vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
10642 vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
10643 vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
10644 vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
10645 vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
10646 vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
10647 vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
10648 vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
10649 vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
10650 vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
10651 vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
10652 vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
10653 vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
10654 vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
10655 vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
10656 vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
10657 vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
10658 vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
10659 vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
10660 vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
10661 vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
10662 vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
10663 vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
10664 vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
10665 vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
10666 vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
10668 vmcs12->guest_interruptibility_info =
10669 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
10670 vmcs12->guest_pending_dbg_exceptions =
10671 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
10672 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
10673 vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
10675 vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
10677 if (nested_cpu_has_preemption_timer(vmcs12)) {
10678 if (vmcs12->vm_exit_controls &
10679 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
10680 vmcs12->vmx_preemption_timer_value =
10681 vmx_get_preemption_timer_value(vcpu);
10682 hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
10686 * In some cases (usually, nested EPT), L2 is allowed to change its
10687 * own CR3 without exiting. If it has changed it, we must keep it.
10688 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
10689 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
10691 * Additionally, restore L2's PDPTR to vmcs12.
10694 vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
10695 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
10696 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
10697 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
10698 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
10701 if (nested_cpu_has_ept(vmcs12))
10702 vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
10704 if (nested_cpu_has_vid(vmcs12))
10705 vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
10707 vmcs12->vm_entry_controls =
10708 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
10709 (vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
10711 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) {
10712 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
10713 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
10716 /* TODO: These cannot have changed unless we have MSR bitmaps and
10717 * the relevant bit asks not to trap the change */
10718 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
10719 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
10720 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
10721 vmcs12->guest_ia32_efer = vcpu->arch.efer;
10722 vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
10723 vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
10724 vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
10725 if (kvm_mpx_supported())
10726 vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
10727 if (nested_cpu_has_xsaves(vmcs12))
10728 vmcs12->xss_exit_bitmap = vmcs_read64(XSS_EXIT_BITMAP);
10732 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
10733 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
10734 * and this function updates it to reflect the changes to the guest state while
10735 * L2 was running (and perhaps made some exits which were handled directly by L0
10736 * without going back to L1), and to reflect the exit reason.
10737 * Note that we do not have to copy here all VMCS fields, just those that
10738 * could have changed by the L2 guest or the exit - i.e., the guest-state and
10739 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
10740 * which already writes to vmcs12 directly.
10742 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10743 u32 exit_reason, u32 exit_intr_info,
10744 unsigned long exit_qualification)
10746 /* update guest state fields: */
10747 sync_vmcs12(vcpu, vmcs12);
10749 /* update exit information fields: */
10751 vmcs12->vm_exit_reason = exit_reason;
10752 vmcs12->exit_qualification = exit_qualification;
10754 vmcs12->vm_exit_intr_info = exit_intr_info;
10755 if ((vmcs12->vm_exit_intr_info &
10756 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
10757 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK))
10758 vmcs12->vm_exit_intr_error_code =
10759 vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
10760 vmcs12->idt_vectoring_info_field = 0;
10761 vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
10762 vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
10764 if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
10765 /* vm_entry_intr_info_field is cleared on exit. Emulate this
10766 * instead of reading the real value. */
10767 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
10770 * Transfer the event that L0 or L1 may wanted to inject into
10771 * L2 to IDT_VECTORING_INFO_FIELD.
10773 vmcs12_save_pending_event(vcpu, vmcs12);
10777 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
10778 * preserved above and would only end up incorrectly in L1.
10780 vcpu->arch.nmi_injected = false;
10781 kvm_clear_exception_queue(vcpu);
10782 kvm_clear_interrupt_queue(vcpu);
10786 * A part of what we need to when the nested L2 guest exits and we want to
10787 * run its L1 parent, is to reset L1's guest state to the host state specified
10789 * This function is to be called not only on normal nested exit, but also on
10790 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
10791 * Failures During or After Loading Guest State").
10792 * This function should be called when the active VMCS is L1's (vmcs01).
10794 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
10795 struct vmcs12 *vmcs12)
10797 struct kvm_segment seg;
10798 u32 entry_failure_code;
10800 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
10801 vcpu->arch.efer = vmcs12->host_ia32_efer;
10802 else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
10803 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10805 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10806 vmx_set_efer(vcpu, vcpu->arch.efer);
10808 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
10809 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
10810 vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
10812 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
10813 * actually changed, because vmx_set_cr0 refers to efer set above.
10815 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
10816 * (KVM doesn't change it);
10818 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
10819 vmx_set_cr0(vcpu, vmcs12->host_cr0);
10821 /* Same as above - no reason to call set_cr4_guest_host_mask(). */
10822 vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
10823 kvm_set_cr4(vcpu, vmcs12->host_cr4);
10825 nested_ept_uninit_mmu_context(vcpu);
10828 * Only PDPTE load can fail as the value of cr3 was checked on entry and
10829 * couldn't have changed.
10831 if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, &entry_failure_code))
10832 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
10835 vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
10839 * Trivially support vpid by letting L2s share their parent
10840 * L1's vpid. TODO: move to a more elaborate solution, giving
10841 * each L2 its own vpid and exposing the vpid feature to L1.
10843 vmx_flush_tlb(vcpu);
10847 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
10848 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
10849 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
10850 vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
10851 vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
10853 /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1. */
10854 if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
10855 vmcs_write64(GUEST_BNDCFGS, 0);
10857 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
10858 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
10859 vcpu->arch.pat = vmcs12->host_ia32_pat;
10861 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
10862 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
10863 vmcs12->host_ia32_perf_global_ctrl);
10865 /* Set L1 segment info according to Intel SDM
10866 27.5.2 Loading Host Segment and Descriptor-Table Registers */
10867 seg = (struct kvm_segment) {
10869 .limit = 0xFFFFFFFF,
10870 .selector = vmcs12->host_cs_selector,
10876 if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
10880 vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
10881 seg = (struct kvm_segment) {
10883 .limit = 0xFFFFFFFF,
10890 seg.selector = vmcs12->host_ds_selector;
10891 vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
10892 seg.selector = vmcs12->host_es_selector;
10893 vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
10894 seg.selector = vmcs12->host_ss_selector;
10895 vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
10896 seg.selector = vmcs12->host_fs_selector;
10897 seg.base = vmcs12->host_fs_base;
10898 vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
10899 seg.selector = vmcs12->host_gs_selector;
10900 seg.base = vmcs12->host_gs_base;
10901 vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
10902 seg = (struct kvm_segment) {
10903 .base = vmcs12->host_tr_base,
10905 .selector = vmcs12->host_tr_selector,
10909 vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
10911 kvm_set_dr(vcpu, 7, 0x400);
10912 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
10914 if (cpu_has_vmx_msr_bitmap())
10915 vmx_set_msr_bitmap(vcpu);
10917 if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
10918 vmcs12->vm_exit_msr_load_count))
10919 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
10923 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
10924 * and modify vmcs12 to make it see what it would expect to see there if
10925 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
10927 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
10928 u32 exit_intr_info,
10929 unsigned long exit_qualification)
10931 struct vcpu_vmx *vmx = to_vmx(vcpu);
10932 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
10933 u32 vm_inst_error = 0;
10935 /* trying to cancel vmlaunch/vmresume is a bug */
10936 WARN_ON_ONCE(vmx->nested.nested_run_pending);
10938 leave_guest_mode(vcpu);
10939 prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
10940 exit_qualification);
10942 if (nested_vmx_store_msr(vcpu, vmcs12->vm_exit_msr_store_addr,
10943 vmcs12->vm_exit_msr_store_count))
10944 nested_vmx_abort(vcpu, VMX_ABORT_SAVE_GUEST_MSR_FAIL);
10946 if (unlikely(vmx->fail))
10947 vm_inst_error = vmcs_read32(VM_INSTRUCTION_ERROR);
10949 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
10951 if ((exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
10952 && nested_exit_intr_ack_set(vcpu)) {
10953 int irq = kvm_cpu_get_interrupt(vcpu);
10955 vmcs12->vm_exit_intr_info = irq |
10956 INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
10959 trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
10960 vmcs12->exit_qualification,
10961 vmcs12->idt_vectoring_info_field,
10962 vmcs12->vm_exit_intr_info,
10963 vmcs12->vm_exit_intr_error_code,
10966 vm_entry_controls_reset_shadow(vmx);
10967 vm_exit_controls_reset_shadow(vmx);
10968 vmx_segment_cache_clear(vmx);
10970 /* if no vmcs02 cache requested, remove the one we used */
10971 if (VMCS02_POOL_SIZE == 0)
10972 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
10974 load_vmcs12_host_state(vcpu, vmcs12);
10976 /* Update any VMCS fields that might have changed while L2 ran */
10977 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
10978 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
10979 vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
10980 if (vmx->hv_deadline_tsc == -1)
10981 vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
10982 PIN_BASED_VMX_PREEMPTION_TIMER);
10984 vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
10985 PIN_BASED_VMX_PREEMPTION_TIMER);
10986 if (kvm_has_tsc_control)
10987 decache_tsc_multiplier(vmx);
10989 if (vmx->nested.change_vmcs01_virtual_x2apic_mode) {
10990 vmx->nested.change_vmcs01_virtual_x2apic_mode = false;
10991 vmx_set_virtual_x2apic_mode(vcpu,
10992 vcpu->arch.apic_base & X2APIC_ENABLE);
10993 } else if (!nested_cpu_has_ept(vmcs12) &&
10994 nested_cpu_has2(vmcs12,
10995 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
10996 vmx_flush_tlb_ept_only(vcpu);
10999 /* This is needed for same reason as it was needed in prepare_vmcs02 */
11002 /* Unpin physical memory we referred to in vmcs02 */
11003 if (vmx->nested.apic_access_page) {
11004 nested_release_page(vmx->nested.apic_access_page);
11005 vmx->nested.apic_access_page = NULL;
11007 if (vmx->nested.virtual_apic_page) {
11008 nested_release_page(vmx->nested.virtual_apic_page);
11009 vmx->nested.virtual_apic_page = NULL;
11011 if (vmx->nested.pi_desc_page) {
11012 kunmap(vmx->nested.pi_desc_page);
11013 nested_release_page(vmx->nested.pi_desc_page);
11014 vmx->nested.pi_desc_page = NULL;
11015 vmx->nested.pi_desc = NULL;
11019 * We are now running in L2, mmu_notifier will force to reload the
11020 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
11022 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
11025 * Exiting from L2 to L1, we're now back to L1 which thinks it just
11026 * finished a VMLAUNCH or VMRESUME instruction, so we need to set the
11027 * success or failure flag accordingly.
11029 if (unlikely(vmx->fail)) {
11031 nested_vmx_failValid(vcpu, vm_inst_error);
11033 nested_vmx_succeed(vcpu);
11034 if (enable_shadow_vmcs)
11035 vmx->nested.sync_shadow_vmcs = true;
11037 /* in case we halted in L2 */
11038 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11042 * Forcibly leave nested mode in order to be able to reset the VCPU later on.
11044 static void vmx_leave_nested(struct kvm_vcpu *vcpu)
11046 if (is_guest_mode(vcpu)) {
11047 to_vmx(vcpu)->nested.nested_run_pending = 0;
11048 nested_vmx_vmexit(vcpu, -1, 0, 0);
11050 free_nested(to_vmx(vcpu));
11054 * L1's failure to enter L2 is a subset of a normal exit, as explained in
11055 * 23.7 "VM-entry failures during or after loading guest state" (this also
11056 * lists the acceptable exit-reason and exit-qualification parameters).
11057 * It should only be called before L2 actually succeeded to run, and when
11058 * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
11060 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
11061 struct vmcs12 *vmcs12,
11062 u32 reason, unsigned long qualification)
11064 load_vmcs12_host_state(vcpu, vmcs12);
11065 vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
11066 vmcs12->exit_qualification = qualification;
11067 nested_vmx_succeed(vcpu);
11068 if (enable_shadow_vmcs)
11069 to_vmx(vcpu)->nested.sync_shadow_vmcs = true;
11072 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
11073 struct x86_instruction_info *info,
11074 enum x86_intercept_stage stage)
11076 return X86EMUL_CONTINUE;
11079 #ifdef CONFIG_X86_64
11080 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
11081 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
11082 u64 divisor, u64 *result)
11084 u64 low = a << shift, high = a >> (64 - shift);
11086 /* To avoid the overflow on divq */
11087 if (high >= divisor)
11090 /* Low hold the result, high hold rem which is discarded */
11091 asm("divq %2\n\t" : "=a" (low), "=d" (high) :
11092 "rm" (divisor), "0" (low), "1" (high));
11098 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc)
11100 struct vcpu_vmx *vmx = to_vmx(vcpu);
11101 u64 tscl = rdtsc();
11102 u64 guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
11103 u64 delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
11105 /* Convert to host delta tsc if tsc scaling is enabled */
11106 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
11107 u64_shl_div_u64(delta_tsc,
11108 kvm_tsc_scaling_ratio_frac_bits,
11109 vcpu->arch.tsc_scaling_ratio,
11114 * If the delta tsc can't fit in the 32 bit after the multi shift,
11115 * we can't use the preemption timer.
11116 * It's possible that it fits on later vmentries, but checking
11117 * on every vmentry is costly so we just use an hrtimer.
11119 if (delta_tsc >> (cpu_preemption_timer_multi + 32))
11122 vmx->hv_deadline_tsc = tscl + delta_tsc;
11123 vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
11124 PIN_BASED_VMX_PREEMPTION_TIMER);
11128 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
11130 struct vcpu_vmx *vmx = to_vmx(vcpu);
11131 vmx->hv_deadline_tsc = -1;
11132 vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
11133 PIN_BASED_VMX_PREEMPTION_TIMER);
11137 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
11140 shrink_ple_window(vcpu);
11143 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
11144 struct kvm_memory_slot *slot)
11146 kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
11147 kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
11150 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
11151 struct kvm_memory_slot *slot)
11153 kvm_mmu_slot_set_dirty(kvm, slot);
11156 static void vmx_flush_log_dirty(struct kvm *kvm)
11158 kvm_flush_pml_buffers(kvm);
11161 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
11162 struct kvm_memory_slot *memslot,
11163 gfn_t offset, unsigned long mask)
11165 kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
11169 * This routine does the following things for vCPU which is going
11170 * to be blocked if VT-d PI is enabled.
11171 * - Store the vCPU to the wakeup list, so when interrupts happen
11172 * we can find the right vCPU to wake up.
11173 * - Change the Posted-interrupt descriptor as below:
11174 * 'NDST' <-- vcpu->pre_pcpu
11175 * 'NV' <-- POSTED_INTR_WAKEUP_VECTOR
11176 * - If 'ON' is set during this process, which means at least one
11177 * interrupt is posted for this vCPU, we cannot block it, in
11178 * this case, return 1, otherwise, return 0.
11181 static int pi_pre_block(struct kvm_vcpu *vcpu)
11183 unsigned long flags;
11185 struct pi_desc old, new;
11186 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
11188 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
11189 !irq_remapping_cap(IRQ_POSTING_CAP) ||
11190 !kvm_vcpu_apicv_active(vcpu))
11193 vcpu->pre_pcpu = vcpu->cpu;
11194 spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
11195 vcpu->pre_pcpu), flags);
11196 list_add_tail(&vcpu->blocked_vcpu_list,
11197 &per_cpu(blocked_vcpu_on_cpu,
11199 spin_unlock_irqrestore(&per_cpu(blocked_vcpu_on_cpu_lock,
11200 vcpu->pre_pcpu), flags);
11203 old.control = new.control = pi_desc->control;
11206 * We should not block the vCPU if
11207 * an interrupt is posted for it.
11209 if (pi_test_on(pi_desc) == 1) {
11210 spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
11211 vcpu->pre_pcpu), flags);
11212 list_del(&vcpu->blocked_vcpu_list);
11213 spin_unlock_irqrestore(
11214 &per_cpu(blocked_vcpu_on_cpu_lock,
11215 vcpu->pre_pcpu), flags);
11216 vcpu->pre_pcpu = -1;
11221 WARN((pi_desc->sn == 1),
11222 "Warning: SN field of posted-interrupts "
11223 "is set before blocking\n");
11226 * Since vCPU can be preempted during this process,
11227 * vcpu->cpu could be different with pre_pcpu, we
11228 * need to set pre_pcpu as the destination of wakeup
11229 * notification event, then we can find the right vCPU
11230 * to wakeup in wakeup handler if interrupts happen
11231 * when the vCPU is in blocked state.
11233 dest = cpu_physical_id(vcpu->pre_pcpu);
11235 if (x2apic_enabled())
11238 new.ndst = (dest << 8) & 0xFF00;
11240 /* set 'NV' to 'wakeup vector' */
11241 new.nv = POSTED_INTR_WAKEUP_VECTOR;
11242 } while (cmpxchg(&pi_desc->control, old.control,
11243 new.control) != old.control);
11248 static int vmx_pre_block(struct kvm_vcpu *vcpu)
11250 if (pi_pre_block(vcpu))
11253 if (kvm_lapic_hv_timer_in_use(vcpu))
11254 kvm_lapic_switch_to_sw_timer(vcpu);
11259 static void pi_post_block(struct kvm_vcpu *vcpu)
11261 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
11262 struct pi_desc old, new;
11264 unsigned long flags;
11266 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
11267 !irq_remapping_cap(IRQ_POSTING_CAP) ||
11268 !kvm_vcpu_apicv_active(vcpu))
11272 old.control = new.control = pi_desc->control;
11274 dest = cpu_physical_id(vcpu->cpu);
11276 if (x2apic_enabled())
11279 new.ndst = (dest << 8) & 0xFF00;
11281 /* Allow posting non-urgent interrupts */
11284 /* set 'NV' to 'notification vector' */
11285 new.nv = POSTED_INTR_VECTOR;
11286 } while (cmpxchg(&pi_desc->control, old.control,
11287 new.control) != old.control);
11289 if(vcpu->pre_pcpu != -1) {
11291 &per_cpu(blocked_vcpu_on_cpu_lock,
11292 vcpu->pre_pcpu), flags);
11293 list_del(&vcpu->blocked_vcpu_list);
11294 spin_unlock_irqrestore(
11295 &per_cpu(blocked_vcpu_on_cpu_lock,
11296 vcpu->pre_pcpu), flags);
11297 vcpu->pre_pcpu = -1;
11301 static void vmx_post_block(struct kvm_vcpu *vcpu)
11303 if (kvm_x86_ops->set_hv_timer)
11304 kvm_lapic_switch_to_hv_timer(vcpu);
11306 pi_post_block(vcpu);
11310 * vmx_update_pi_irte - set IRTE for Posted-Interrupts
11313 * @host_irq: host irq of the interrupt
11314 * @guest_irq: gsi of the interrupt
11315 * @set: set or unset PI
11316 * returns 0 on success, < 0 on failure
11318 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
11319 uint32_t guest_irq, bool set)
11321 struct kvm_kernel_irq_routing_entry *e;
11322 struct kvm_irq_routing_table *irq_rt;
11323 struct kvm_lapic_irq irq;
11324 struct kvm_vcpu *vcpu;
11325 struct vcpu_data vcpu_info;
11326 int idx, ret = -EINVAL;
11328 if (!kvm_arch_has_assigned_device(kvm) ||
11329 !irq_remapping_cap(IRQ_POSTING_CAP) ||
11330 !kvm_vcpu_apicv_active(kvm->vcpus[0]))
11333 idx = srcu_read_lock(&kvm->irq_srcu);
11334 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
11335 BUG_ON(guest_irq >= irq_rt->nr_rt_entries);
11337 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
11338 if (e->type != KVM_IRQ_ROUTING_MSI)
11341 * VT-d PI cannot support posting multicast/broadcast
11342 * interrupts to a vCPU, we still use interrupt remapping
11343 * for these kind of interrupts.
11345 * For lowest-priority interrupts, we only support
11346 * those with single CPU as the destination, e.g. user
11347 * configures the interrupts via /proc/irq or uses
11348 * irqbalance to make the interrupts single-CPU.
11350 * We will support full lowest-priority interrupt later.
11353 kvm_set_msi_irq(kvm, e, &irq);
11354 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
11356 * Make sure the IRTE is in remapped mode if
11357 * we don't handle it in posted mode.
11359 ret = irq_set_vcpu_affinity(host_irq, NULL);
11362 "failed to back to remapped mode, irq: %u\n",
11370 vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
11371 vcpu_info.vector = irq.vector;
11373 trace_kvm_pi_irte_update(vcpu->vcpu_id, host_irq, e->gsi,
11374 vcpu_info.vector, vcpu_info.pi_desc_addr, set);
11377 ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
11379 /* suppress notification event before unposting */
11380 pi_set_sn(vcpu_to_pi_desc(vcpu));
11381 ret = irq_set_vcpu_affinity(host_irq, NULL);
11382 pi_clear_sn(vcpu_to_pi_desc(vcpu));
11386 printk(KERN_INFO "%s: failed to update PI IRTE\n",
11394 srcu_read_unlock(&kvm->irq_srcu, idx);
11398 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
11400 if (vcpu->arch.mcg_cap & MCG_LMCE_P)
11401 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
11402 FEATURE_CONTROL_LMCE;
11404 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
11405 ~FEATURE_CONTROL_LMCE;
11408 static struct kvm_x86_ops vmx_x86_ops __ro_after_init = {
11409 .cpu_has_kvm_support = cpu_has_kvm_support,
11410 .disabled_by_bios = vmx_disabled_by_bios,
11411 .hardware_setup = hardware_setup,
11412 .hardware_unsetup = hardware_unsetup,
11413 .check_processor_compatibility = vmx_check_processor_compat,
11414 .hardware_enable = hardware_enable,
11415 .hardware_disable = hardware_disable,
11416 .cpu_has_accelerated_tpr = report_flexpriority,
11417 .cpu_has_high_real_mode_segbase = vmx_has_high_real_mode_segbase,
11419 .vcpu_create = vmx_create_vcpu,
11420 .vcpu_free = vmx_free_vcpu,
11421 .vcpu_reset = vmx_vcpu_reset,
11423 .prepare_guest_switch = vmx_save_host_state,
11424 .vcpu_load = vmx_vcpu_load,
11425 .vcpu_put = vmx_vcpu_put,
11427 .update_bp_intercept = update_exception_bitmap,
11428 .get_msr = vmx_get_msr,
11429 .set_msr = vmx_set_msr,
11430 .get_segment_base = vmx_get_segment_base,
11431 .get_segment = vmx_get_segment,
11432 .set_segment = vmx_set_segment,
11433 .get_cpl = vmx_get_cpl,
11434 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
11435 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
11436 .decache_cr3 = vmx_decache_cr3,
11437 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
11438 .set_cr0 = vmx_set_cr0,
11439 .set_cr3 = vmx_set_cr3,
11440 .set_cr4 = vmx_set_cr4,
11441 .set_efer = vmx_set_efer,
11442 .get_idt = vmx_get_idt,
11443 .set_idt = vmx_set_idt,
11444 .get_gdt = vmx_get_gdt,
11445 .set_gdt = vmx_set_gdt,
11446 .get_dr6 = vmx_get_dr6,
11447 .set_dr6 = vmx_set_dr6,
11448 .set_dr7 = vmx_set_dr7,
11449 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
11450 .cache_reg = vmx_cache_reg,
11451 .get_rflags = vmx_get_rflags,
11452 .set_rflags = vmx_set_rflags,
11454 .get_pkru = vmx_get_pkru,
11456 .tlb_flush = vmx_flush_tlb,
11458 .run = vmx_vcpu_run,
11459 .handle_exit = vmx_handle_exit,
11460 .skip_emulated_instruction = skip_emulated_instruction,
11461 .set_interrupt_shadow = vmx_set_interrupt_shadow,
11462 .get_interrupt_shadow = vmx_get_interrupt_shadow,
11463 .patch_hypercall = vmx_patch_hypercall,
11464 .set_irq = vmx_inject_irq,
11465 .set_nmi = vmx_inject_nmi,
11466 .queue_exception = vmx_queue_exception,
11467 .cancel_injection = vmx_cancel_injection,
11468 .interrupt_allowed = vmx_interrupt_allowed,
11469 .nmi_allowed = vmx_nmi_allowed,
11470 .get_nmi_mask = vmx_get_nmi_mask,
11471 .set_nmi_mask = vmx_set_nmi_mask,
11472 .enable_nmi_window = enable_nmi_window,
11473 .enable_irq_window = enable_irq_window,
11474 .update_cr8_intercept = update_cr8_intercept,
11475 .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode,
11476 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
11477 .get_enable_apicv = vmx_get_enable_apicv,
11478 .refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
11479 .load_eoi_exitmap = vmx_load_eoi_exitmap,
11480 .apicv_post_state_restore = vmx_apicv_post_state_restore,
11481 .hwapic_irr_update = vmx_hwapic_irr_update,
11482 .hwapic_isr_update = vmx_hwapic_isr_update,
11483 .sync_pir_to_irr = vmx_sync_pir_to_irr,
11484 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
11486 .set_tss_addr = vmx_set_tss_addr,
11487 .get_tdp_level = get_ept_level,
11488 .get_mt_mask = vmx_get_mt_mask,
11490 .get_exit_info = vmx_get_exit_info,
11492 .get_lpage_level = vmx_get_lpage_level,
11494 .cpuid_update = vmx_cpuid_update,
11496 .rdtscp_supported = vmx_rdtscp_supported,
11497 .invpcid_supported = vmx_invpcid_supported,
11499 .set_supported_cpuid = vmx_set_supported_cpuid,
11501 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
11503 .write_tsc_offset = vmx_write_tsc_offset,
11505 .set_tdp_cr3 = vmx_set_cr3,
11507 .check_intercept = vmx_check_intercept,
11508 .handle_external_intr = vmx_handle_external_intr,
11509 .mpx_supported = vmx_mpx_supported,
11510 .xsaves_supported = vmx_xsaves_supported,
11512 .check_nested_events = vmx_check_nested_events,
11514 .sched_in = vmx_sched_in,
11516 .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
11517 .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
11518 .flush_log_dirty = vmx_flush_log_dirty,
11519 .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
11521 .pre_block = vmx_pre_block,
11522 .post_block = vmx_post_block,
11524 .pmu_ops = &intel_pmu_ops,
11526 .update_pi_irte = vmx_update_pi_irte,
11528 #ifdef CONFIG_X86_64
11529 .set_hv_timer = vmx_set_hv_timer,
11530 .cancel_hv_timer = vmx_cancel_hv_timer,
11533 .setup_mce = vmx_setup_mce,
11536 static int __init vmx_init(void)
11538 int r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
11539 __alignof__(struct vcpu_vmx), THIS_MODULE);
11543 #ifdef CONFIG_KEXEC_CORE
11544 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
11545 crash_vmclear_local_loaded_vmcss);
11551 static void __exit vmx_exit(void)
11553 #ifdef CONFIG_KEXEC_CORE
11554 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
11561 module_init(vmx_init)
11562 module_exit(vmx_exit)