2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
23 #include <linux/kvm_host.h>
24 #include <linux/module.h>
25 #include <linux/kernel.h>
27 #include <linux/highmem.h>
28 #include <linux/sched.h>
29 #include <linux/moduleparam.h>
30 #include <linux/mod_devicetable.h>
31 #include <linux/ftrace_event.h>
32 #include <linux/slab.h>
33 #include <linux/tboot.h>
34 #include <linux/hrtimer.h>
35 #include "kvm_cache_regs.h"
41 #include <asm/virtext.h>
45 #include <asm/perf_event.h>
46 #include <asm/debugreg.h>
47 #include <asm/kexec.h>
52 #define __ex(x) __kvm_handle_fault_on_reboot(x)
53 #define __ex_clear(x, reg) \
54 ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
56 MODULE_AUTHOR("Qumranet");
57 MODULE_LICENSE("GPL");
59 static const struct x86_cpu_id vmx_cpu_id[] = {
60 X86_FEATURE_MATCH(X86_FEATURE_VMX),
63 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
65 static bool __read_mostly enable_vpid = 1;
66 module_param_named(vpid, enable_vpid, bool, 0444);
68 static bool __read_mostly flexpriority_enabled = 1;
69 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
71 static bool __read_mostly enable_ept = 1;
72 module_param_named(ept, enable_ept, bool, S_IRUGO);
74 static bool __read_mostly enable_unrestricted_guest = 1;
75 module_param_named(unrestricted_guest,
76 enable_unrestricted_guest, bool, S_IRUGO);
78 static bool __read_mostly enable_ept_ad_bits = 1;
79 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
81 static bool __read_mostly emulate_invalid_guest_state = true;
82 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
84 static bool __read_mostly vmm_exclusive = 1;
85 module_param(vmm_exclusive, 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_GUEST_CR0_MASK (X86_CR0_NW | X86_CR0_CD)
109 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST (X86_CR0_WP | X86_CR0_NE)
110 #define KVM_VM_CR0_ALWAYS_ON \
111 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
112 #define KVM_CR4_GUEST_OWNED_BITS \
113 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
114 | X86_CR4_OSXMMEXCPT | X86_CR4_TSD)
116 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
117 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
119 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
121 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
124 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
125 * ple_gap: upper bound on the amount of time between two successive
126 * executions of PAUSE in a loop. Also indicate if ple enabled.
127 * According to test, this time is usually smaller than 128 cycles.
128 * ple_window: upper bound on the amount of time a guest is allowed to execute
129 * in a PAUSE loop. Tests indicate that most spinlocks are held for
130 * less than 2^12 cycles
131 * Time is measured based on a counter that runs at the same rate as the TSC,
132 * refer SDM volume 3b section 21.6.13 & 22.1.3.
134 #define KVM_VMX_DEFAULT_PLE_GAP 128
135 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
136 #define KVM_VMX_DEFAULT_PLE_WINDOW_GROW 2
137 #define KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK 0
138 #define KVM_VMX_DEFAULT_PLE_WINDOW_MAX \
139 INT_MAX / KVM_VMX_DEFAULT_PLE_WINDOW_GROW
141 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
142 module_param(ple_gap, int, S_IRUGO);
144 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
145 module_param(ple_window, int, S_IRUGO);
147 /* Default doubles per-vcpu window every exit. */
148 static int ple_window_grow = KVM_VMX_DEFAULT_PLE_WINDOW_GROW;
149 module_param(ple_window_grow, int, S_IRUGO);
151 /* Default resets per-vcpu window every exit to ple_window. */
152 static int ple_window_shrink = KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK;
153 module_param(ple_window_shrink, int, S_IRUGO);
155 /* Default is to compute the maximum so we can never overflow. */
156 static int ple_window_actual_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
157 static int ple_window_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
158 module_param(ple_window_max, int, S_IRUGO);
160 extern const ulong vmx_return;
162 #define NR_AUTOLOAD_MSRS 8
163 #define VMCS02_POOL_SIZE 1
172 * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
173 * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
174 * loaded on this CPU (so we can clear them if the CPU goes down).
180 struct list_head loaded_vmcss_on_cpu_link;
183 struct shared_msr_entry {
190 * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
191 * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
192 * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
193 * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
194 * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
195 * More than one of these structures may exist, if L1 runs multiple L2 guests.
196 * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
197 * underlying hardware which will be used to run L2.
198 * This structure is packed to ensure that its layout is identical across
199 * machines (necessary for live migration).
200 * If there are changes in this struct, VMCS12_REVISION must be changed.
202 typedef u64 natural_width;
203 struct __packed vmcs12 {
204 /* According to the Intel spec, a VMCS region must start with the
205 * following two fields. Then follow implementation-specific data.
210 u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
211 u32 padding[7]; /* room for future expansion */
216 u64 vm_exit_msr_store_addr;
217 u64 vm_exit_msr_load_addr;
218 u64 vm_entry_msr_load_addr;
220 u64 virtual_apic_page_addr;
221 u64 apic_access_addr;
222 u64 posted_intr_desc_addr;
224 u64 eoi_exit_bitmap0;
225 u64 eoi_exit_bitmap1;
226 u64 eoi_exit_bitmap2;
227 u64 eoi_exit_bitmap3;
229 u64 guest_physical_address;
230 u64 vmcs_link_pointer;
231 u64 guest_ia32_debugctl;
234 u64 guest_ia32_perf_global_ctrl;
242 u64 host_ia32_perf_global_ctrl;
243 u64 padding64[8]; /* room for future expansion */
245 * To allow migration of L1 (complete with its L2 guests) between
246 * machines of different natural widths (32 or 64 bit), we cannot have
247 * unsigned long fields with no explict size. We use u64 (aliased
248 * natural_width) instead. Luckily, x86 is little-endian.
250 natural_width cr0_guest_host_mask;
251 natural_width cr4_guest_host_mask;
252 natural_width cr0_read_shadow;
253 natural_width cr4_read_shadow;
254 natural_width cr3_target_value0;
255 natural_width cr3_target_value1;
256 natural_width cr3_target_value2;
257 natural_width cr3_target_value3;
258 natural_width exit_qualification;
259 natural_width guest_linear_address;
260 natural_width guest_cr0;
261 natural_width guest_cr3;
262 natural_width guest_cr4;
263 natural_width guest_es_base;
264 natural_width guest_cs_base;
265 natural_width guest_ss_base;
266 natural_width guest_ds_base;
267 natural_width guest_fs_base;
268 natural_width guest_gs_base;
269 natural_width guest_ldtr_base;
270 natural_width guest_tr_base;
271 natural_width guest_gdtr_base;
272 natural_width guest_idtr_base;
273 natural_width guest_dr7;
274 natural_width guest_rsp;
275 natural_width guest_rip;
276 natural_width guest_rflags;
277 natural_width guest_pending_dbg_exceptions;
278 natural_width guest_sysenter_esp;
279 natural_width guest_sysenter_eip;
280 natural_width host_cr0;
281 natural_width host_cr3;
282 natural_width host_cr4;
283 natural_width host_fs_base;
284 natural_width host_gs_base;
285 natural_width host_tr_base;
286 natural_width host_gdtr_base;
287 natural_width host_idtr_base;
288 natural_width host_ia32_sysenter_esp;
289 natural_width host_ia32_sysenter_eip;
290 natural_width host_rsp;
291 natural_width host_rip;
292 natural_width paddingl[8]; /* room for future expansion */
293 u32 pin_based_vm_exec_control;
294 u32 cpu_based_vm_exec_control;
295 u32 exception_bitmap;
296 u32 page_fault_error_code_mask;
297 u32 page_fault_error_code_match;
298 u32 cr3_target_count;
299 u32 vm_exit_controls;
300 u32 vm_exit_msr_store_count;
301 u32 vm_exit_msr_load_count;
302 u32 vm_entry_controls;
303 u32 vm_entry_msr_load_count;
304 u32 vm_entry_intr_info_field;
305 u32 vm_entry_exception_error_code;
306 u32 vm_entry_instruction_len;
308 u32 secondary_vm_exec_control;
309 u32 vm_instruction_error;
311 u32 vm_exit_intr_info;
312 u32 vm_exit_intr_error_code;
313 u32 idt_vectoring_info_field;
314 u32 idt_vectoring_error_code;
315 u32 vm_exit_instruction_len;
316 u32 vmx_instruction_info;
323 u32 guest_ldtr_limit;
325 u32 guest_gdtr_limit;
326 u32 guest_idtr_limit;
327 u32 guest_es_ar_bytes;
328 u32 guest_cs_ar_bytes;
329 u32 guest_ss_ar_bytes;
330 u32 guest_ds_ar_bytes;
331 u32 guest_fs_ar_bytes;
332 u32 guest_gs_ar_bytes;
333 u32 guest_ldtr_ar_bytes;
334 u32 guest_tr_ar_bytes;
335 u32 guest_interruptibility_info;
336 u32 guest_activity_state;
337 u32 guest_sysenter_cs;
338 u32 host_ia32_sysenter_cs;
339 u32 vmx_preemption_timer_value;
340 u32 padding32[7]; /* room for future expansion */
341 u16 virtual_processor_id;
343 u16 guest_es_selector;
344 u16 guest_cs_selector;
345 u16 guest_ss_selector;
346 u16 guest_ds_selector;
347 u16 guest_fs_selector;
348 u16 guest_gs_selector;
349 u16 guest_ldtr_selector;
350 u16 guest_tr_selector;
351 u16 guest_intr_status;
352 u16 host_es_selector;
353 u16 host_cs_selector;
354 u16 host_ss_selector;
355 u16 host_ds_selector;
356 u16 host_fs_selector;
357 u16 host_gs_selector;
358 u16 host_tr_selector;
362 * VMCS12_REVISION is an arbitrary id that should be changed if the content or
363 * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
364 * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
366 #define VMCS12_REVISION 0x11e57ed0
369 * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
370 * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
371 * current implementation, 4K are reserved to avoid future complications.
373 #define VMCS12_SIZE 0x1000
375 /* Used to remember the last vmcs02 used for some recently used vmcs12s */
377 struct list_head list;
379 struct loaded_vmcs vmcs02;
383 * The nested_vmx structure is part of vcpu_vmx, and holds information we need
384 * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
387 /* Has the level1 guest done vmxon? */
391 /* The guest-physical address of the current VMCS L1 keeps for L2 */
393 /* The host-usable pointer to the above */
394 struct page *current_vmcs12_page;
395 struct vmcs12 *current_vmcs12;
396 struct vmcs *current_shadow_vmcs;
398 * Indicates if the shadow vmcs must be updated with the
399 * data hold by vmcs12
401 bool sync_shadow_vmcs;
403 /* vmcs02_list cache of VMCSs recently used to run L2 guests */
404 struct list_head vmcs02_pool;
406 u64 vmcs01_tsc_offset;
407 /* L2 must run next, and mustn't decide to exit to L1. */
408 bool nested_run_pending;
410 * Guest pages referred to in vmcs02 with host-physical pointers, so
411 * we must keep them pinned while L2 runs.
413 struct page *apic_access_page;
414 struct page *virtual_apic_page;
415 struct page *pi_desc_page;
416 struct pi_desc *pi_desc;
419 u64 msr_ia32_feature_control;
421 struct hrtimer preemption_timer;
422 bool preemption_timer_expired;
424 /* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
427 u32 nested_vmx_procbased_ctls_low;
428 u32 nested_vmx_procbased_ctls_high;
429 u32 nested_vmx_true_procbased_ctls_low;
430 u32 nested_vmx_secondary_ctls_low;
431 u32 nested_vmx_secondary_ctls_high;
432 u32 nested_vmx_pinbased_ctls_low;
433 u32 nested_vmx_pinbased_ctls_high;
434 u32 nested_vmx_exit_ctls_low;
435 u32 nested_vmx_exit_ctls_high;
436 u32 nested_vmx_true_exit_ctls_low;
437 u32 nested_vmx_entry_ctls_low;
438 u32 nested_vmx_entry_ctls_high;
439 u32 nested_vmx_true_entry_ctls_low;
440 u32 nested_vmx_misc_low;
441 u32 nested_vmx_misc_high;
442 u32 nested_vmx_ept_caps;
445 #define POSTED_INTR_ON 0
446 /* Posted-Interrupt Descriptor */
448 u32 pir[8]; /* Posted interrupt requested */
449 u32 control; /* bit 0 of control is outstanding notification bit */
453 static bool pi_test_and_set_on(struct pi_desc *pi_desc)
455 return test_and_set_bit(POSTED_INTR_ON,
456 (unsigned long *)&pi_desc->control);
459 static bool pi_test_and_clear_on(struct pi_desc *pi_desc)
461 return test_and_clear_bit(POSTED_INTR_ON,
462 (unsigned long *)&pi_desc->control);
465 static int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
467 return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
471 struct kvm_vcpu vcpu;
472 unsigned long host_rsp;
474 bool nmi_known_unmasked;
476 u32 idt_vectoring_info;
478 struct shared_msr_entry *guest_msrs;
481 unsigned long host_idt_base;
483 u64 msr_host_kernel_gs_base;
484 u64 msr_guest_kernel_gs_base;
486 u32 vm_entry_controls_shadow;
487 u32 vm_exit_controls_shadow;
489 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
490 * non-nested (L1) guest, it always points to vmcs01. For a nested
491 * guest (L2), it points to a different VMCS.
493 struct loaded_vmcs vmcs01;
494 struct loaded_vmcs *loaded_vmcs;
495 bool __launched; /* temporary, used in vmx_vcpu_run */
496 struct msr_autoload {
498 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
499 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
503 u16 fs_sel, gs_sel, ldt_sel;
507 int gs_ldt_reload_needed;
508 int fs_reload_needed;
509 u64 msr_host_bndcfgs;
510 unsigned long vmcs_host_cr4; /* May not match real cr4 */
515 struct kvm_segment segs[8];
518 u32 bitmask; /* 4 bits per segment (1 bit per field) */
519 struct kvm_save_segment {
527 bool emulation_required;
529 /* Support for vnmi-less CPUs */
530 int soft_vnmi_blocked;
532 s64 vnmi_blocked_time;
537 /* Posted interrupt descriptor */
538 struct pi_desc pi_desc;
540 /* Support for a guest hypervisor (nested VMX) */
541 struct nested_vmx nested;
543 /* Dynamic PLE window. */
545 bool ple_window_dirty;
547 /* Support for PML */
548 #define PML_ENTITY_NUM 512
552 enum segment_cache_field {
561 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
563 return container_of(vcpu, struct vcpu_vmx, vcpu);
566 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
567 #define FIELD(number, name) [number] = VMCS12_OFFSET(name)
568 #define FIELD64(number, name) [number] = VMCS12_OFFSET(name), \
569 [number##_HIGH] = VMCS12_OFFSET(name)+4
572 static unsigned long shadow_read_only_fields[] = {
574 * We do NOT shadow fields that are modified when L0
575 * traps and emulates any vmx instruction (e.g. VMPTRLD,
576 * VMXON...) executed by L1.
577 * For example, VM_INSTRUCTION_ERROR is read
578 * by L1 if a vmx instruction fails (part of the error path).
579 * Note the code assumes this logic. If for some reason
580 * we start shadowing these fields then we need to
581 * force a shadow sync when L0 emulates vmx instructions
582 * (e.g. force a sync if VM_INSTRUCTION_ERROR is modified
583 * by nested_vmx_failValid)
587 VM_EXIT_INSTRUCTION_LEN,
588 IDT_VECTORING_INFO_FIELD,
589 IDT_VECTORING_ERROR_CODE,
590 VM_EXIT_INTR_ERROR_CODE,
592 GUEST_LINEAR_ADDRESS,
593 GUEST_PHYSICAL_ADDRESS
595 static int max_shadow_read_only_fields =
596 ARRAY_SIZE(shadow_read_only_fields);
598 static unsigned long shadow_read_write_fields[] = {
605 GUEST_INTERRUPTIBILITY_INFO,
618 CPU_BASED_VM_EXEC_CONTROL,
619 VM_ENTRY_EXCEPTION_ERROR_CODE,
620 VM_ENTRY_INTR_INFO_FIELD,
621 VM_ENTRY_INSTRUCTION_LEN,
622 VM_ENTRY_EXCEPTION_ERROR_CODE,
628 static int max_shadow_read_write_fields =
629 ARRAY_SIZE(shadow_read_write_fields);
631 static const unsigned short vmcs_field_to_offset_table[] = {
632 FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
633 FIELD(POSTED_INTR_NV, posted_intr_nv),
634 FIELD(GUEST_ES_SELECTOR, guest_es_selector),
635 FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
636 FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
637 FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
638 FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
639 FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
640 FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
641 FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
642 FIELD(GUEST_INTR_STATUS, guest_intr_status),
643 FIELD(HOST_ES_SELECTOR, host_es_selector),
644 FIELD(HOST_CS_SELECTOR, host_cs_selector),
645 FIELD(HOST_SS_SELECTOR, host_ss_selector),
646 FIELD(HOST_DS_SELECTOR, host_ds_selector),
647 FIELD(HOST_FS_SELECTOR, host_fs_selector),
648 FIELD(HOST_GS_SELECTOR, host_gs_selector),
649 FIELD(HOST_TR_SELECTOR, host_tr_selector),
650 FIELD64(IO_BITMAP_A, io_bitmap_a),
651 FIELD64(IO_BITMAP_B, io_bitmap_b),
652 FIELD64(MSR_BITMAP, msr_bitmap),
653 FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
654 FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
655 FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
656 FIELD64(TSC_OFFSET, tsc_offset),
657 FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
658 FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
659 FIELD64(POSTED_INTR_DESC_ADDR, posted_intr_desc_addr),
660 FIELD64(EPT_POINTER, ept_pointer),
661 FIELD64(EOI_EXIT_BITMAP0, eoi_exit_bitmap0),
662 FIELD64(EOI_EXIT_BITMAP1, eoi_exit_bitmap1),
663 FIELD64(EOI_EXIT_BITMAP2, eoi_exit_bitmap2),
664 FIELD64(EOI_EXIT_BITMAP3, eoi_exit_bitmap3),
665 FIELD64(XSS_EXIT_BITMAP, xss_exit_bitmap),
666 FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
667 FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
668 FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
669 FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
670 FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
671 FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
672 FIELD64(GUEST_PDPTR0, guest_pdptr0),
673 FIELD64(GUEST_PDPTR1, guest_pdptr1),
674 FIELD64(GUEST_PDPTR2, guest_pdptr2),
675 FIELD64(GUEST_PDPTR3, guest_pdptr3),
676 FIELD64(GUEST_BNDCFGS, guest_bndcfgs),
677 FIELD64(HOST_IA32_PAT, host_ia32_pat),
678 FIELD64(HOST_IA32_EFER, host_ia32_efer),
679 FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
680 FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
681 FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
682 FIELD(EXCEPTION_BITMAP, exception_bitmap),
683 FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
684 FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
685 FIELD(CR3_TARGET_COUNT, cr3_target_count),
686 FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
687 FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
688 FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
689 FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
690 FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
691 FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
692 FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
693 FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
694 FIELD(TPR_THRESHOLD, tpr_threshold),
695 FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
696 FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
697 FIELD(VM_EXIT_REASON, vm_exit_reason),
698 FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
699 FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
700 FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
701 FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
702 FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
703 FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
704 FIELD(GUEST_ES_LIMIT, guest_es_limit),
705 FIELD(GUEST_CS_LIMIT, guest_cs_limit),
706 FIELD(GUEST_SS_LIMIT, guest_ss_limit),
707 FIELD(GUEST_DS_LIMIT, guest_ds_limit),
708 FIELD(GUEST_FS_LIMIT, guest_fs_limit),
709 FIELD(GUEST_GS_LIMIT, guest_gs_limit),
710 FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
711 FIELD(GUEST_TR_LIMIT, guest_tr_limit),
712 FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
713 FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
714 FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
715 FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
716 FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
717 FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
718 FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
719 FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
720 FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
721 FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
722 FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
723 FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
724 FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
725 FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
726 FIELD(VMX_PREEMPTION_TIMER_VALUE, vmx_preemption_timer_value),
727 FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
728 FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
729 FIELD(CR0_READ_SHADOW, cr0_read_shadow),
730 FIELD(CR4_READ_SHADOW, cr4_read_shadow),
731 FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
732 FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
733 FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
734 FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
735 FIELD(EXIT_QUALIFICATION, exit_qualification),
736 FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
737 FIELD(GUEST_CR0, guest_cr0),
738 FIELD(GUEST_CR3, guest_cr3),
739 FIELD(GUEST_CR4, guest_cr4),
740 FIELD(GUEST_ES_BASE, guest_es_base),
741 FIELD(GUEST_CS_BASE, guest_cs_base),
742 FIELD(GUEST_SS_BASE, guest_ss_base),
743 FIELD(GUEST_DS_BASE, guest_ds_base),
744 FIELD(GUEST_FS_BASE, guest_fs_base),
745 FIELD(GUEST_GS_BASE, guest_gs_base),
746 FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
747 FIELD(GUEST_TR_BASE, guest_tr_base),
748 FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
749 FIELD(GUEST_IDTR_BASE, guest_idtr_base),
750 FIELD(GUEST_DR7, guest_dr7),
751 FIELD(GUEST_RSP, guest_rsp),
752 FIELD(GUEST_RIP, guest_rip),
753 FIELD(GUEST_RFLAGS, guest_rflags),
754 FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
755 FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
756 FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
757 FIELD(HOST_CR0, host_cr0),
758 FIELD(HOST_CR3, host_cr3),
759 FIELD(HOST_CR4, host_cr4),
760 FIELD(HOST_FS_BASE, host_fs_base),
761 FIELD(HOST_GS_BASE, host_gs_base),
762 FIELD(HOST_TR_BASE, host_tr_base),
763 FIELD(HOST_GDTR_BASE, host_gdtr_base),
764 FIELD(HOST_IDTR_BASE, host_idtr_base),
765 FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
766 FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
767 FIELD(HOST_RSP, host_rsp),
768 FIELD(HOST_RIP, host_rip),
771 static inline short vmcs_field_to_offset(unsigned long field)
773 BUILD_BUG_ON(ARRAY_SIZE(vmcs_field_to_offset_table) > SHRT_MAX);
775 if (field >= ARRAY_SIZE(vmcs_field_to_offset_table) ||
776 vmcs_field_to_offset_table[field] == 0)
779 return vmcs_field_to_offset_table[field];
782 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
784 return to_vmx(vcpu)->nested.current_vmcs12;
787 static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
789 struct page *page = gfn_to_page(vcpu->kvm, addr >> PAGE_SHIFT);
790 if (is_error_page(page))
796 static void nested_release_page(struct page *page)
798 kvm_release_page_dirty(page);
801 static void nested_release_page_clean(struct page *page)
803 kvm_release_page_clean(page);
806 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu);
807 static u64 construct_eptp(unsigned long root_hpa);
808 static void kvm_cpu_vmxon(u64 addr);
809 static void kvm_cpu_vmxoff(void);
810 static bool vmx_mpx_supported(void);
811 static bool vmx_xsaves_supported(void);
812 static int vmx_vm_has_apicv(struct kvm *kvm);
813 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
814 static void vmx_set_segment(struct kvm_vcpu *vcpu,
815 struct kvm_segment *var, int seg);
816 static void vmx_get_segment(struct kvm_vcpu *vcpu,
817 struct kvm_segment *var, int seg);
818 static bool guest_state_valid(struct kvm_vcpu *vcpu);
819 static u32 vmx_segment_access_rights(struct kvm_segment *var);
820 static void vmx_sync_pir_to_irr_dummy(struct kvm_vcpu *vcpu);
821 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx);
822 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx);
823 static int alloc_identity_pagetable(struct kvm *kvm);
825 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
826 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
828 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
829 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
831 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
832 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
834 static unsigned long *vmx_io_bitmap_a;
835 static unsigned long *vmx_io_bitmap_b;
836 static unsigned long *vmx_msr_bitmap_legacy;
837 static unsigned long *vmx_msr_bitmap_longmode;
838 static unsigned long *vmx_msr_bitmap_legacy_x2apic;
839 static unsigned long *vmx_msr_bitmap_longmode_x2apic;
840 static unsigned long *vmx_msr_bitmap_nested;
841 static unsigned long *vmx_vmread_bitmap;
842 static unsigned long *vmx_vmwrite_bitmap;
844 static bool cpu_has_load_ia32_efer;
845 static bool cpu_has_load_perf_global_ctrl;
847 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
848 static DEFINE_SPINLOCK(vmx_vpid_lock);
850 static struct vmcs_config {
854 u32 pin_based_exec_ctrl;
855 u32 cpu_based_exec_ctrl;
856 u32 cpu_based_2nd_exec_ctrl;
861 static struct vmx_capability {
866 #define VMX_SEGMENT_FIELD(seg) \
867 [VCPU_SREG_##seg] = { \
868 .selector = GUEST_##seg##_SELECTOR, \
869 .base = GUEST_##seg##_BASE, \
870 .limit = GUEST_##seg##_LIMIT, \
871 .ar_bytes = GUEST_##seg##_AR_BYTES, \
874 static const struct kvm_vmx_segment_field {
879 } kvm_vmx_segment_fields[] = {
880 VMX_SEGMENT_FIELD(CS),
881 VMX_SEGMENT_FIELD(DS),
882 VMX_SEGMENT_FIELD(ES),
883 VMX_SEGMENT_FIELD(FS),
884 VMX_SEGMENT_FIELD(GS),
885 VMX_SEGMENT_FIELD(SS),
886 VMX_SEGMENT_FIELD(TR),
887 VMX_SEGMENT_FIELD(LDTR),
890 static u64 host_efer;
892 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
895 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
896 * away by decrementing the array size.
898 static const u32 vmx_msr_index[] = {
900 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
902 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
905 static inline bool is_page_fault(u32 intr_info)
907 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
908 INTR_INFO_VALID_MASK)) ==
909 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
912 static inline bool is_no_device(u32 intr_info)
914 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
915 INTR_INFO_VALID_MASK)) ==
916 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
919 static inline bool is_invalid_opcode(u32 intr_info)
921 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
922 INTR_INFO_VALID_MASK)) ==
923 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
926 static inline bool is_external_interrupt(u32 intr_info)
928 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
929 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
932 static inline bool is_machine_check(u32 intr_info)
934 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
935 INTR_INFO_VALID_MASK)) ==
936 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
939 static inline bool cpu_has_vmx_msr_bitmap(void)
941 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
944 static inline bool cpu_has_vmx_tpr_shadow(void)
946 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
949 static inline bool vm_need_tpr_shadow(struct kvm *kvm)
951 return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
954 static inline bool cpu_has_secondary_exec_ctrls(void)
956 return vmcs_config.cpu_based_exec_ctrl &
957 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
960 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
962 return vmcs_config.cpu_based_2nd_exec_ctrl &
963 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
966 static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
968 return vmcs_config.cpu_based_2nd_exec_ctrl &
969 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
972 static inline bool cpu_has_vmx_apic_register_virt(void)
974 return vmcs_config.cpu_based_2nd_exec_ctrl &
975 SECONDARY_EXEC_APIC_REGISTER_VIRT;
978 static inline bool cpu_has_vmx_virtual_intr_delivery(void)
980 return vmcs_config.cpu_based_2nd_exec_ctrl &
981 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
984 static inline bool cpu_has_vmx_posted_intr(void)
986 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_POSTED_INTR;
989 static inline bool cpu_has_vmx_apicv(void)
991 return cpu_has_vmx_apic_register_virt() &&
992 cpu_has_vmx_virtual_intr_delivery() &&
993 cpu_has_vmx_posted_intr();
996 static inline bool cpu_has_vmx_flexpriority(void)
998 return cpu_has_vmx_tpr_shadow() &&
999 cpu_has_vmx_virtualize_apic_accesses();
1002 static inline bool cpu_has_vmx_ept_execute_only(void)
1004 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
1007 static inline bool cpu_has_vmx_ept_2m_page(void)
1009 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
1012 static inline bool cpu_has_vmx_ept_1g_page(void)
1014 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
1017 static inline bool cpu_has_vmx_ept_4levels(void)
1019 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
1022 static inline bool cpu_has_vmx_ept_ad_bits(void)
1024 return vmx_capability.ept & VMX_EPT_AD_BIT;
1027 static inline bool cpu_has_vmx_invept_context(void)
1029 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
1032 static inline bool cpu_has_vmx_invept_global(void)
1034 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
1037 static inline bool cpu_has_vmx_invvpid_single(void)
1039 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
1042 static inline bool cpu_has_vmx_invvpid_global(void)
1044 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
1047 static inline bool cpu_has_vmx_ept(void)
1049 return vmcs_config.cpu_based_2nd_exec_ctrl &
1050 SECONDARY_EXEC_ENABLE_EPT;
1053 static inline bool cpu_has_vmx_unrestricted_guest(void)
1055 return vmcs_config.cpu_based_2nd_exec_ctrl &
1056 SECONDARY_EXEC_UNRESTRICTED_GUEST;
1059 static inline bool cpu_has_vmx_ple(void)
1061 return vmcs_config.cpu_based_2nd_exec_ctrl &
1062 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
1065 static inline bool vm_need_virtualize_apic_accesses(struct kvm *kvm)
1067 return flexpriority_enabled && irqchip_in_kernel(kvm);
1070 static inline bool cpu_has_vmx_vpid(void)
1072 return vmcs_config.cpu_based_2nd_exec_ctrl &
1073 SECONDARY_EXEC_ENABLE_VPID;
1076 static inline bool cpu_has_vmx_rdtscp(void)
1078 return vmcs_config.cpu_based_2nd_exec_ctrl &
1079 SECONDARY_EXEC_RDTSCP;
1082 static inline bool cpu_has_vmx_invpcid(void)
1084 return vmcs_config.cpu_based_2nd_exec_ctrl &
1085 SECONDARY_EXEC_ENABLE_INVPCID;
1088 static inline bool cpu_has_virtual_nmis(void)
1090 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
1093 static inline bool cpu_has_vmx_wbinvd_exit(void)
1095 return vmcs_config.cpu_based_2nd_exec_ctrl &
1096 SECONDARY_EXEC_WBINVD_EXITING;
1099 static inline bool cpu_has_vmx_shadow_vmcs(void)
1102 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
1103 /* check if the cpu supports writing r/o exit information fields */
1104 if (!(vmx_msr & MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS))
1107 return vmcs_config.cpu_based_2nd_exec_ctrl &
1108 SECONDARY_EXEC_SHADOW_VMCS;
1111 static inline bool cpu_has_vmx_pml(void)
1113 return vmcs_config.cpu_based_2nd_exec_ctrl & SECONDARY_EXEC_ENABLE_PML;
1116 static inline bool report_flexpriority(void)
1118 return flexpriority_enabled;
1121 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
1123 return vmcs12->cpu_based_vm_exec_control & bit;
1126 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
1128 return (vmcs12->cpu_based_vm_exec_control &
1129 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
1130 (vmcs12->secondary_vm_exec_control & bit);
1133 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12)
1135 return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
1138 static inline bool nested_cpu_has_preemption_timer(struct vmcs12 *vmcs12)
1140 return vmcs12->pin_based_vm_exec_control &
1141 PIN_BASED_VMX_PREEMPTION_TIMER;
1144 static inline int nested_cpu_has_ept(struct vmcs12 *vmcs12)
1146 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_EPT);
1149 static inline bool nested_cpu_has_xsaves(struct vmcs12 *vmcs12)
1151 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES) &&
1152 vmx_xsaves_supported();
1155 static inline bool nested_cpu_has_virt_x2apic_mode(struct vmcs12 *vmcs12)
1157 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
1160 static inline bool nested_cpu_has_apic_reg_virt(struct vmcs12 *vmcs12)
1162 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_APIC_REGISTER_VIRT);
1165 static inline bool nested_cpu_has_vid(struct vmcs12 *vmcs12)
1167 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
1170 static inline bool nested_cpu_has_posted_intr(struct vmcs12 *vmcs12)
1172 return vmcs12->pin_based_vm_exec_control & PIN_BASED_POSTED_INTR;
1175 static inline bool is_exception(u32 intr_info)
1177 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1178 == (INTR_TYPE_HARD_EXCEPTION | INTR_INFO_VALID_MASK);
1181 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
1183 unsigned long exit_qualification);
1184 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
1185 struct vmcs12 *vmcs12,
1186 u32 reason, unsigned long qualification);
1188 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
1192 for (i = 0; i < vmx->nmsrs; ++i)
1193 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
1198 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
1204 } operand = { vpid, 0, gva };
1206 asm volatile (__ex(ASM_VMX_INVVPID)
1207 /* CF==1 or ZF==1 --> rc = -1 */
1208 "; ja 1f ; ud2 ; 1:"
1209 : : "a"(&operand), "c"(ext) : "cc", "memory");
1212 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
1216 } operand = {eptp, gpa};
1218 asm volatile (__ex(ASM_VMX_INVEPT)
1219 /* CF==1 or ZF==1 --> rc = -1 */
1220 "; ja 1f ; ud2 ; 1:\n"
1221 : : "a" (&operand), "c" (ext) : "cc", "memory");
1224 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
1228 i = __find_msr_index(vmx, msr);
1230 return &vmx->guest_msrs[i];
1234 static void vmcs_clear(struct vmcs *vmcs)
1236 u64 phys_addr = __pa(vmcs);
1239 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
1240 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1243 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
1247 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
1249 vmcs_clear(loaded_vmcs->vmcs);
1250 loaded_vmcs->cpu = -1;
1251 loaded_vmcs->launched = 0;
1254 static void vmcs_load(struct vmcs *vmcs)
1256 u64 phys_addr = __pa(vmcs);
1259 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
1260 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1263 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
1269 * This bitmap is used to indicate whether the vmclear
1270 * operation is enabled on all cpus. All disabled by
1273 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
1275 static inline void crash_enable_local_vmclear(int cpu)
1277 cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
1280 static inline void crash_disable_local_vmclear(int cpu)
1282 cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
1285 static inline int crash_local_vmclear_enabled(int cpu)
1287 return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
1290 static void crash_vmclear_local_loaded_vmcss(void)
1292 int cpu = raw_smp_processor_id();
1293 struct loaded_vmcs *v;
1295 if (!crash_local_vmclear_enabled(cpu))
1298 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
1299 loaded_vmcss_on_cpu_link)
1300 vmcs_clear(v->vmcs);
1303 static inline void crash_enable_local_vmclear(int cpu) { }
1304 static inline void crash_disable_local_vmclear(int cpu) { }
1305 #endif /* CONFIG_KEXEC */
1307 static void __loaded_vmcs_clear(void *arg)
1309 struct loaded_vmcs *loaded_vmcs = arg;
1310 int cpu = raw_smp_processor_id();
1312 if (loaded_vmcs->cpu != cpu)
1313 return; /* vcpu migration can race with cpu offline */
1314 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1315 per_cpu(current_vmcs, cpu) = NULL;
1316 crash_disable_local_vmclear(cpu);
1317 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1320 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1321 * is before setting loaded_vmcs->vcpu to -1 which is done in
1322 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1323 * then adds the vmcs into percpu list before it is deleted.
1327 loaded_vmcs_init(loaded_vmcs);
1328 crash_enable_local_vmclear(cpu);
1331 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1333 int cpu = loaded_vmcs->cpu;
1336 smp_call_function_single(cpu,
1337 __loaded_vmcs_clear, loaded_vmcs, 1);
1340 static inline void vpid_sync_vcpu_single(struct vcpu_vmx *vmx)
1345 if (cpu_has_vmx_invvpid_single())
1346 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
1349 static inline void vpid_sync_vcpu_global(void)
1351 if (cpu_has_vmx_invvpid_global())
1352 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1355 static inline void vpid_sync_context(struct vcpu_vmx *vmx)
1357 if (cpu_has_vmx_invvpid_single())
1358 vpid_sync_vcpu_single(vmx);
1360 vpid_sync_vcpu_global();
1363 static inline void ept_sync_global(void)
1365 if (cpu_has_vmx_invept_global())
1366 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1369 static inline void ept_sync_context(u64 eptp)
1372 if (cpu_has_vmx_invept_context())
1373 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1379 static __always_inline unsigned long vmcs_readl(unsigned long field)
1381 unsigned long value;
1383 asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1384 : "=a"(value) : "d"(field) : "cc");
1388 static __always_inline u16 vmcs_read16(unsigned long field)
1390 return vmcs_readl(field);
1393 static __always_inline u32 vmcs_read32(unsigned long field)
1395 return vmcs_readl(field);
1398 static __always_inline u64 vmcs_read64(unsigned long field)
1400 #ifdef CONFIG_X86_64
1401 return vmcs_readl(field);
1403 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
1407 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1409 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1410 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1414 static void vmcs_writel(unsigned long field, unsigned long value)
1418 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1419 : "=q"(error) : "a"(value), "d"(field) : "cc");
1420 if (unlikely(error))
1421 vmwrite_error(field, value);
1424 static void vmcs_write16(unsigned long field, u16 value)
1426 vmcs_writel(field, value);
1429 static void vmcs_write32(unsigned long field, u32 value)
1431 vmcs_writel(field, value);
1434 static void vmcs_write64(unsigned long field, u64 value)
1436 vmcs_writel(field, value);
1437 #ifndef CONFIG_X86_64
1439 vmcs_writel(field+1, value >> 32);
1443 static void vmcs_clear_bits(unsigned long field, u32 mask)
1445 vmcs_writel(field, vmcs_readl(field) & ~mask);
1448 static void vmcs_set_bits(unsigned long field, u32 mask)
1450 vmcs_writel(field, vmcs_readl(field) | mask);
1453 static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val)
1455 vmcs_write32(VM_ENTRY_CONTROLS, val);
1456 vmx->vm_entry_controls_shadow = val;
1459 static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val)
1461 if (vmx->vm_entry_controls_shadow != val)
1462 vm_entry_controls_init(vmx, val);
1465 static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx)
1467 return vmx->vm_entry_controls_shadow;
1471 static inline void vm_entry_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1473 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) | val);
1476 static inline void vm_entry_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1478 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) & ~val);
1481 static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val)
1483 vmcs_write32(VM_EXIT_CONTROLS, val);
1484 vmx->vm_exit_controls_shadow = val;
1487 static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val)
1489 if (vmx->vm_exit_controls_shadow != val)
1490 vm_exit_controls_init(vmx, val);
1493 static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx)
1495 return vmx->vm_exit_controls_shadow;
1499 static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1501 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val);
1504 static inline void vm_exit_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1506 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) & ~val);
1509 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1511 vmx->segment_cache.bitmask = 0;
1514 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1518 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1520 if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1521 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1522 vmx->segment_cache.bitmask = 0;
1524 ret = vmx->segment_cache.bitmask & mask;
1525 vmx->segment_cache.bitmask |= mask;
1529 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1531 u16 *p = &vmx->segment_cache.seg[seg].selector;
1533 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1534 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1538 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1540 ulong *p = &vmx->segment_cache.seg[seg].base;
1542 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1543 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1547 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1549 u32 *p = &vmx->segment_cache.seg[seg].limit;
1551 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1552 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1556 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1558 u32 *p = &vmx->segment_cache.seg[seg].ar;
1560 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1561 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1565 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1569 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1570 (1u << NM_VECTOR) | (1u << DB_VECTOR);
1571 if ((vcpu->guest_debug &
1572 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1573 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1574 eb |= 1u << BP_VECTOR;
1575 if (to_vmx(vcpu)->rmode.vm86_active)
1578 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1579 if (vcpu->fpu_active)
1580 eb &= ~(1u << NM_VECTOR);
1582 /* When we are running a nested L2 guest and L1 specified for it a
1583 * certain exception bitmap, we must trap the same exceptions and pass
1584 * them to L1. When running L2, we will only handle the exceptions
1585 * specified above if L1 did not want them.
1587 if (is_guest_mode(vcpu))
1588 eb |= get_vmcs12(vcpu)->exception_bitmap;
1590 vmcs_write32(EXCEPTION_BITMAP, eb);
1593 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1594 unsigned long entry, unsigned long exit)
1596 vm_entry_controls_clearbit(vmx, entry);
1597 vm_exit_controls_clearbit(vmx, exit);
1600 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1603 struct msr_autoload *m = &vmx->msr_autoload;
1607 if (cpu_has_load_ia32_efer) {
1608 clear_atomic_switch_msr_special(vmx,
1609 VM_ENTRY_LOAD_IA32_EFER,
1610 VM_EXIT_LOAD_IA32_EFER);
1614 case MSR_CORE_PERF_GLOBAL_CTRL:
1615 if (cpu_has_load_perf_global_ctrl) {
1616 clear_atomic_switch_msr_special(vmx,
1617 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1618 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1624 for (i = 0; i < m->nr; ++i)
1625 if (m->guest[i].index == msr)
1631 m->guest[i] = m->guest[m->nr];
1632 m->host[i] = m->host[m->nr];
1633 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1634 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1637 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1638 unsigned long entry, unsigned long exit,
1639 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
1640 u64 guest_val, u64 host_val)
1642 vmcs_write64(guest_val_vmcs, guest_val);
1643 vmcs_write64(host_val_vmcs, host_val);
1644 vm_entry_controls_setbit(vmx, entry);
1645 vm_exit_controls_setbit(vmx, exit);
1648 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1649 u64 guest_val, u64 host_val)
1652 struct msr_autoload *m = &vmx->msr_autoload;
1656 if (cpu_has_load_ia32_efer) {
1657 add_atomic_switch_msr_special(vmx,
1658 VM_ENTRY_LOAD_IA32_EFER,
1659 VM_EXIT_LOAD_IA32_EFER,
1662 guest_val, host_val);
1666 case MSR_CORE_PERF_GLOBAL_CTRL:
1667 if (cpu_has_load_perf_global_ctrl) {
1668 add_atomic_switch_msr_special(vmx,
1669 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1670 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1671 GUEST_IA32_PERF_GLOBAL_CTRL,
1672 HOST_IA32_PERF_GLOBAL_CTRL,
1673 guest_val, host_val);
1679 for (i = 0; i < m->nr; ++i)
1680 if (m->guest[i].index == msr)
1683 if (i == NR_AUTOLOAD_MSRS) {
1684 printk_once(KERN_WARNING "Not enough msr switch entries. "
1685 "Can't add msr %x\n", msr);
1687 } else if (i == m->nr) {
1689 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1690 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1693 m->guest[i].index = msr;
1694 m->guest[i].value = guest_val;
1695 m->host[i].index = msr;
1696 m->host[i].value = host_val;
1699 static void reload_tss(void)
1702 * VT restores TR but not its size. Useless.
1704 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
1705 struct desc_struct *descs;
1707 descs = (void *)gdt->address;
1708 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
1712 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
1717 guest_efer = vmx->vcpu.arch.efer;
1720 * NX is emulated; LMA and LME handled by hardware; SCE meaningless
1723 ignore_bits = EFER_NX | EFER_SCE;
1724 #ifdef CONFIG_X86_64
1725 ignore_bits |= EFER_LMA | EFER_LME;
1726 /* SCE is meaningful only in long mode on Intel */
1727 if (guest_efer & EFER_LMA)
1728 ignore_bits &= ~(u64)EFER_SCE;
1730 guest_efer &= ~ignore_bits;
1731 guest_efer |= host_efer & ignore_bits;
1732 vmx->guest_msrs[efer_offset].data = guest_efer;
1733 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1735 clear_atomic_switch_msr(vmx, MSR_EFER);
1738 * On EPT, we can't emulate NX, so we must switch EFER atomically.
1739 * On CPUs that support "load IA32_EFER", always switch EFER
1740 * atomically, since it's faster than switching it manually.
1742 if (cpu_has_load_ia32_efer ||
1743 (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
1744 guest_efer = vmx->vcpu.arch.efer;
1745 if (!(guest_efer & EFER_LMA))
1746 guest_efer &= ~EFER_LME;
1747 if (guest_efer != host_efer)
1748 add_atomic_switch_msr(vmx, MSR_EFER,
1749 guest_efer, host_efer);
1756 static unsigned long segment_base(u16 selector)
1758 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
1759 struct desc_struct *d;
1760 unsigned long table_base;
1763 if (!(selector & ~3))
1766 table_base = gdt->address;
1768 if (selector & 4) { /* from ldt */
1769 u16 ldt_selector = kvm_read_ldt();
1771 if (!(ldt_selector & ~3))
1774 table_base = segment_base(ldt_selector);
1776 d = (struct desc_struct *)(table_base + (selector & ~7));
1777 v = get_desc_base(d);
1778 #ifdef CONFIG_X86_64
1779 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
1780 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
1785 static inline unsigned long kvm_read_tr_base(void)
1788 asm("str %0" : "=g"(tr));
1789 return segment_base(tr);
1792 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
1794 struct vcpu_vmx *vmx = to_vmx(vcpu);
1797 if (vmx->host_state.loaded)
1800 vmx->host_state.loaded = 1;
1802 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
1803 * allow segment selectors with cpl > 0 or ti == 1.
1805 vmx->host_state.ldt_sel = kvm_read_ldt();
1806 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
1807 savesegment(fs, vmx->host_state.fs_sel);
1808 if (!(vmx->host_state.fs_sel & 7)) {
1809 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
1810 vmx->host_state.fs_reload_needed = 0;
1812 vmcs_write16(HOST_FS_SELECTOR, 0);
1813 vmx->host_state.fs_reload_needed = 1;
1815 savesegment(gs, vmx->host_state.gs_sel);
1816 if (!(vmx->host_state.gs_sel & 7))
1817 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
1819 vmcs_write16(HOST_GS_SELECTOR, 0);
1820 vmx->host_state.gs_ldt_reload_needed = 1;
1823 #ifdef CONFIG_X86_64
1824 savesegment(ds, vmx->host_state.ds_sel);
1825 savesegment(es, vmx->host_state.es_sel);
1828 #ifdef CONFIG_X86_64
1829 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
1830 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
1832 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
1833 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
1836 #ifdef CONFIG_X86_64
1837 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1838 if (is_long_mode(&vmx->vcpu))
1839 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1841 if (boot_cpu_has(X86_FEATURE_MPX))
1842 rdmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
1843 for (i = 0; i < vmx->save_nmsrs; ++i)
1844 kvm_set_shared_msr(vmx->guest_msrs[i].index,
1845 vmx->guest_msrs[i].data,
1846 vmx->guest_msrs[i].mask);
1849 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
1851 if (!vmx->host_state.loaded)
1854 ++vmx->vcpu.stat.host_state_reload;
1855 vmx->host_state.loaded = 0;
1856 #ifdef CONFIG_X86_64
1857 if (is_long_mode(&vmx->vcpu))
1858 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1860 if (vmx->host_state.gs_ldt_reload_needed) {
1861 kvm_load_ldt(vmx->host_state.ldt_sel);
1862 #ifdef CONFIG_X86_64
1863 load_gs_index(vmx->host_state.gs_sel);
1865 loadsegment(gs, vmx->host_state.gs_sel);
1868 if (vmx->host_state.fs_reload_needed)
1869 loadsegment(fs, vmx->host_state.fs_sel);
1870 #ifdef CONFIG_X86_64
1871 if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
1872 loadsegment(ds, vmx->host_state.ds_sel);
1873 loadsegment(es, vmx->host_state.es_sel);
1877 #ifdef CONFIG_X86_64
1878 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1880 if (vmx->host_state.msr_host_bndcfgs)
1881 wrmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
1883 * If the FPU is not active (through the host task or
1884 * the guest vcpu), then restore the cr0.TS bit.
1886 if (!user_has_fpu() && !vmx->vcpu.guest_fpu_loaded)
1888 load_gdt(this_cpu_ptr(&host_gdt));
1891 static void vmx_load_host_state(struct vcpu_vmx *vmx)
1894 __vmx_load_host_state(vmx);
1899 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1900 * vcpu mutex is already taken.
1902 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1904 struct vcpu_vmx *vmx = to_vmx(vcpu);
1905 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1908 kvm_cpu_vmxon(phys_addr);
1909 else if (vmx->loaded_vmcs->cpu != cpu)
1910 loaded_vmcs_clear(vmx->loaded_vmcs);
1912 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1913 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1914 vmcs_load(vmx->loaded_vmcs->vmcs);
1917 if (vmx->loaded_vmcs->cpu != cpu) {
1918 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
1919 unsigned long sysenter_esp;
1921 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1922 local_irq_disable();
1923 crash_disable_local_vmclear(cpu);
1926 * Read loaded_vmcs->cpu should be before fetching
1927 * loaded_vmcs->loaded_vmcss_on_cpu_link.
1928 * See the comments in __loaded_vmcs_clear().
1932 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1933 &per_cpu(loaded_vmcss_on_cpu, cpu));
1934 crash_enable_local_vmclear(cpu);
1938 * Linux uses per-cpu TSS and GDT, so set these when switching
1941 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
1942 vmcs_writel(HOST_GDTR_BASE, gdt->address); /* 22.2.4 */
1944 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1945 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1946 vmx->loaded_vmcs->cpu = cpu;
1950 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1952 __vmx_load_host_state(to_vmx(vcpu));
1953 if (!vmm_exclusive) {
1954 __loaded_vmcs_clear(to_vmx(vcpu)->loaded_vmcs);
1960 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
1964 if (vcpu->fpu_active)
1966 vcpu->fpu_active = 1;
1967 cr0 = vmcs_readl(GUEST_CR0);
1968 cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
1969 cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
1970 vmcs_writel(GUEST_CR0, cr0);
1971 update_exception_bitmap(vcpu);
1972 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
1973 if (is_guest_mode(vcpu))
1974 vcpu->arch.cr0_guest_owned_bits &=
1975 ~get_vmcs12(vcpu)->cr0_guest_host_mask;
1976 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1979 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
1982 * Return the cr0 value that a nested guest would read. This is a combination
1983 * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
1984 * its hypervisor (cr0_read_shadow).
1986 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
1988 return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
1989 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
1991 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
1993 return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
1994 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
1997 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
1999 /* Note that there is no vcpu->fpu_active = 0 here. The caller must
2000 * set this *before* calling this function.
2002 vmx_decache_cr0_guest_bits(vcpu);
2003 vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
2004 update_exception_bitmap(vcpu);
2005 vcpu->arch.cr0_guest_owned_bits = 0;
2006 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2007 if (is_guest_mode(vcpu)) {
2009 * L1's specified read shadow might not contain the TS bit,
2010 * so now that we turned on shadowing of this bit, we need to
2011 * set this bit of the shadow. Like in nested_vmx_run we need
2012 * nested_read_cr0(vmcs12), but vmcs12->guest_cr0 is not yet
2013 * up-to-date here because we just decached cr0.TS (and we'll
2014 * only update vmcs12->guest_cr0 on nested exit).
2016 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2017 vmcs12->guest_cr0 = (vmcs12->guest_cr0 & ~X86_CR0_TS) |
2018 (vcpu->arch.cr0 & X86_CR0_TS);
2019 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2021 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
2024 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
2026 unsigned long rflags, save_rflags;
2028 if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
2029 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2030 rflags = vmcs_readl(GUEST_RFLAGS);
2031 if (to_vmx(vcpu)->rmode.vm86_active) {
2032 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2033 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
2034 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2036 to_vmx(vcpu)->rflags = rflags;
2038 return to_vmx(vcpu)->rflags;
2041 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
2043 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2044 to_vmx(vcpu)->rflags = rflags;
2045 if (to_vmx(vcpu)->rmode.vm86_active) {
2046 to_vmx(vcpu)->rmode.save_rflags = rflags;
2047 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2049 vmcs_writel(GUEST_RFLAGS, rflags);
2052 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
2054 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2057 if (interruptibility & GUEST_INTR_STATE_STI)
2058 ret |= KVM_X86_SHADOW_INT_STI;
2059 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
2060 ret |= KVM_X86_SHADOW_INT_MOV_SS;
2065 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
2067 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2068 u32 interruptibility = interruptibility_old;
2070 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
2072 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
2073 interruptibility |= GUEST_INTR_STATE_MOV_SS;
2074 else if (mask & KVM_X86_SHADOW_INT_STI)
2075 interruptibility |= GUEST_INTR_STATE_STI;
2077 if ((interruptibility != interruptibility_old))
2078 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
2081 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
2085 rip = kvm_rip_read(vcpu);
2086 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
2087 kvm_rip_write(vcpu, rip);
2089 /* skipping an emulated instruction also counts */
2090 vmx_set_interrupt_shadow(vcpu, 0);
2094 * KVM wants to inject page-faults which it got to the guest. This function
2095 * checks whether in a nested guest, we need to inject them to L1 or L2.
2097 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned nr)
2099 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2101 if (!(vmcs12->exception_bitmap & (1u << nr)))
2104 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
2105 vmcs_read32(VM_EXIT_INTR_INFO),
2106 vmcs_readl(EXIT_QUALIFICATION));
2110 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
2111 bool has_error_code, u32 error_code,
2114 struct vcpu_vmx *vmx = to_vmx(vcpu);
2115 u32 intr_info = nr | INTR_INFO_VALID_MASK;
2117 if (!reinject && is_guest_mode(vcpu) &&
2118 nested_vmx_check_exception(vcpu, nr))
2121 if (has_error_code) {
2122 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
2123 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2126 if (vmx->rmode.vm86_active) {
2128 if (kvm_exception_is_soft(nr))
2129 inc_eip = vcpu->arch.event_exit_inst_len;
2130 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
2131 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2135 if (kvm_exception_is_soft(nr)) {
2136 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2137 vmx->vcpu.arch.event_exit_inst_len);
2138 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2140 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2142 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
2145 static bool vmx_rdtscp_supported(void)
2147 return cpu_has_vmx_rdtscp();
2150 static bool vmx_invpcid_supported(void)
2152 return cpu_has_vmx_invpcid() && enable_ept;
2156 * Swap MSR entry in host/guest MSR entry array.
2158 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
2160 struct shared_msr_entry tmp;
2162 tmp = vmx->guest_msrs[to];
2163 vmx->guest_msrs[to] = vmx->guest_msrs[from];
2164 vmx->guest_msrs[from] = tmp;
2167 static void vmx_set_msr_bitmap(struct kvm_vcpu *vcpu)
2169 unsigned long *msr_bitmap;
2171 if (is_guest_mode(vcpu))
2172 msr_bitmap = vmx_msr_bitmap_nested;
2173 else if (vcpu->arch.apic_base & X2APIC_ENABLE) {
2174 if (is_long_mode(vcpu))
2175 msr_bitmap = vmx_msr_bitmap_longmode_x2apic;
2177 msr_bitmap = vmx_msr_bitmap_legacy_x2apic;
2179 if (is_long_mode(vcpu))
2180 msr_bitmap = vmx_msr_bitmap_longmode;
2182 msr_bitmap = vmx_msr_bitmap_legacy;
2185 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
2189 * Set up the vmcs to automatically save and restore system
2190 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
2191 * mode, as fiddling with msrs is very expensive.
2193 static void setup_msrs(struct vcpu_vmx *vmx)
2195 int save_nmsrs, index;
2198 #ifdef CONFIG_X86_64
2199 if (is_long_mode(&vmx->vcpu)) {
2200 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
2202 move_msr_up(vmx, index, save_nmsrs++);
2203 index = __find_msr_index(vmx, MSR_LSTAR);
2205 move_msr_up(vmx, index, save_nmsrs++);
2206 index = __find_msr_index(vmx, MSR_CSTAR);
2208 move_msr_up(vmx, index, save_nmsrs++);
2209 index = __find_msr_index(vmx, MSR_TSC_AUX);
2210 if (index >= 0 && vmx->rdtscp_enabled)
2211 move_msr_up(vmx, index, save_nmsrs++);
2213 * MSR_STAR is only needed on long mode guests, and only
2214 * if efer.sce is enabled.
2216 index = __find_msr_index(vmx, MSR_STAR);
2217 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
2218 move_msr_up(vmx, index, save_nmsrs++);
2221 index = __find_msr_index(vmx, MSR_EFER);
2222 if (index >= 0 && update_transition_efer(vmx, index))
2223 move_msr_up(vmx, index, save_nmsrs++);
2225 vmx->save_nmsrs = save_nmsrs;
2227 if (cpu_has_vmx_msr_bitmap())
2228 vmx_set_msr_bitmap(&vmx->vcpu);
2232 * reads and returns guest's timestamp counter "register"
2233 * guest_tsc = host_tsc + tsc_offset -- 21.3
2235 static u64 guest_read_tsc(void)
2237 u64 host_tsc, tsc_offset;
2240 tsc_offset = vmcs_read64(TSC_OFFSET);
2241 return host_tsc + tsc_offset;
2245 * Like guest_read_tsc, but always returns L1's notion of the timestamp
2246 * counter, even if a nested guest (L2) is currently running.
2248 static u64 vmx_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2252 tsc_offset = is_guest_mode(vcpu) ?
2253 to_vmx(vcpu)->nested.vmcs01_tsc_offset :
2254 vmcs_read64(TSC_OFFSET);
2255 return host_tsc + tsc_offset;
2259 * Engage any workarounds for mis-matched TSC rates. Currently limited to
2260 * software catchup for faster rates on slower CPUs.
2262 static void vmx_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2267 if (user_tsc_khz > tsc_khz) {
2268 vcpu->arch.tsc_catchup = 1;
2269 vcpu->arch.tsc_always_catchup = 1;
2271 WARN(1, "user requested TSC rate below hardware speed\n");
2274 static u64 vmx_read_tsc_offset(struct kvm_vcpu *vcpu)
2276 return vmcs_read64(TSC_OFFSET);
2280 * writes 'offset' into guest's timestamp counter offset register
2282 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2284 if (is_guest_mode(vcpu)) {
2286 * We're here if L1 chose not to trap WRMSR to TSC. According
2287 * to the spec, this should set L1's TSC; The offset that L1
2288 * set for L2 remains unchanged, and still needs to be added
2289 * to the newly set TSC to get L2's TSC.
2291 struct vmcs12 *vmcs12;
2292 to_vmx(vcpu)->nested.vmcs01_tsc_offset = offset;
2293 /* recalculate vmcs02.TSC_OFFSET: */
2294 vmcs12 = get_vmcs12(vcpu);
2295 vmcs_write64(TSC_OFFSET, offset +
2296 (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
2297 vmcs12->tsc_offset : 0));
2299 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2300 vmcs_read64(TSC_OFFSET), offset);
2301 vmcs_write64(TSC_OFFSET, offset);
2305 static void vmx_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment, bool host)
2307 u64 offset = vmcs_read64(TSC_OFFSET);
2309 vmcs_write64(TSC_OFFSET, offset + adjustment);
2310 if (is_guest_mode(vcpu)) {
2311 /* Even when running L2, the adjustment needs to apply to L1 */
2312 to_vmx(vcpu)->nested.vmcs01_tsc_offset += adjustment;
2314 trace_kvm_write_tsc_offset(vcpu->vcpu_id, offset,
2315 offset + adjustment);
2318 static u64 vmx_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2320 return target_tsc - native_read_tsc();
2323 static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
2325 struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
2326 return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
2330 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
2331 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
2332 * all guests if the "nested" module option is off, and can also be disabled
2333 * for a single guest by disabling its VMX cpuid bit.
2335 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
2337 return nested && guest_cpuid_has_vmx(vcpu);
2341 * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
2342 * returned for the various VMX controls MSRs when nested VMX is enabled.
2343 * The same values should also be used to verify that vmcs12 control fields are
2344 * valid during nested entry from L1 to L2.
2345 * Each of these control msrs has a low and high 32-bit half: A low bit is on
2346 * if the corresponding bit in the (32-bit) control field *must* be on, and a
2347 * bit in the high half is on if the corresponding bit in the control field
2348 * may be on. See also vmx_control_verify().
2350 static void nested_vmx_setup_ctls_msrs(struct vcpu_vmx *vmx)
2353 * Note that as a general rule, the high half of the MSRs (bits in
2354 * the control fields which may be 1) should be initialized by the
2355 * intersection of the underlying hardware's MSR (i.e., features which
2356 * can be supported) and the list of features we want to expose -
2357 * because they are known to be properly supported in our code.
2358 * Also, usually, the low half of the MSRs (bits which must be 1) can
2359 * be set to 0, meaning that L1 may turn off any of these bits. The
2360 * reason is that if one of these bits is necessary, it will appear
2361 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
2362 * fields of vmcs01 and vmcs02, will turn these bits off - and
2363 * nested_vmx_exit_handled() will not pass related exits to L1.
2364 * These rules have exceptions below.
2367 /* pin-based controls */
2368 rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
2369 vmx->nested.nested_vmx_pinbased_ctls_low,
2370 vmx->nested.nested_vmx_pinbased_ctls_high);
2371 vmx->nested.nested_vmx_pinbased_ctls_low |=
2372 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2373 vmx->nested.nested_vmx_pinbased_ctls_high &=
2374 PIN_BASED_EXT_INTR_MASK |
2375 PIN_BASED_NMI_EXITING |
2376 PIN_BASED_VIRTUAL_NMIS;
2377 vmx->nested.nested_vmx_pinbased_ctls_high |=
2378 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2379 PIN_BASED_VMX_PREEMPTION_TIMER;
2380 if (vmx_vm_has_apicv(vmx->vcpu.kvm))
2381 vmx->nested.nested_vmx_pinbased_ctls_high |=
2382 PIN_BASED_POSTED_INTR;
2385 rdmsr(MSR_IA32_VMX_EXIT_CTLS,
2386 vmx->nested.nested_vmx_exit_ctls_low,
2387 vmx->nested.nested_vmx_exit_ctls_high);
2388 vmx->nested.nested_vmx_exit_ctls_low =
2389 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
2391 vmx->nested.nested_vmx_exit_ctls_high &=
2392 #ifdef CONFIG_X86_64
2393 VM_EXIT_HOST_ADDR_SPACE_SIZE |
2395 VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
2396 vmx->nested.nested_vmx_exit_ctls_high |=
2397 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
2398 VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
2399 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
2401 if (vmx_mpx_supported())
2402 vmx->nested.nested_vmx_exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
2404 /* We support free control of debug control saving. */
2405 vmx->nested.nested_vmx_true_exit_ctls_low =
2406 vmx->nested.nested_vmx_exit_ctls_low &
2407 ~VM_EXIT_SAVE_DEBUG_CONTROLS;
2409 /* entry controls */
2410 rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
2411 vmx->nested.nested_vmx_entry_ctls_low,
2412 vmx->nested.nested_vmx_entry_ctls_high);
2413 vmx->nested.nested_vmx_entry_ctls_low =
2414 VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
2415 vmx->nested.nested_vmx_entry_ctls_high &=
2416 #ifdef CONFIG_X86_64
2417 VM_ENTRY_IA32E_MODE |
2419 VM_ENTRY_LOAD_IA32_PAT;
2420 vmx->nested.nested_vmx_entry_ctls_high |=
2421 (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
2422 if (vmx_mpx_supported())
2423 vmx->nested.nested_vmx_entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
2425 /* We support free control of debug control loading. */
2426 vmx->nested.nested_vmx_true_entry_ctls_low =
2427 vmx->nested.nested_vmx_entry_ctls_low &
2428 ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
2430 /* cpu-based controls */
2431 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
2432 vmx->nested.nested_vmx_procbased_ctls_low,
2433 vmx->nested.nested_vmx_procbased_ctls_high);
2434 vmx->nested.nested_vmx_procbased_ctls_low =
2435 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2436 vmx->nested.nested_vmx_procbased_ctls_high &=
2437 CPU_BASED_VIRTUAL_INTR_PENDING |
2438 CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
2439 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
2440 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
2441 CPU_BASED_CR3_STORE_EXITING |
2442 #ifdef CONFIG_X86_64
2443 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
2445 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
2446 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_EXITING |
2447 CPU_BASED_RDPMC_EXITING | CPU_BASED_RDTSC_EXITING |
2448 CPU_BASED_PAUSE_EXITING | CPU_BASED_TPR_SHADOW |
2449 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2451 * We can allow some features even when not supported by the
2452 * hardware. For example, L1 can specify an MSR bitmap - and we
2453 * can use it to avoid exits to L1 - even when L0 runs L2
2454 * without MSR bitmaps.
2456 vmx->nested.nested_vmx_procbased_ctls_high |=
2457 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2458 CPU_BASED_USE_MSR_BITMAPS;
2460 /* We support free control of CR3 access interception. */
2461 vmx->nested.nested_vmx_true_procbased_ctls_low =
2462 vmx->nested.nested_vmx_procbased_ctls_low &
2463 ~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
2465 /* secondary cpu-based controls */
2466 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2467 vmx->nested.nested_vmx_secondary_ctls_low,
2468 vmx->nested.nested_vmx_secondary_ctls_high);
2469 vmx->nested.nested_vmx_secondary_ctls_low = 0;
2470 vmx->nested.nested_vmx_secondary_ctls_high &=
2471 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2472 SECONDARY_EXEC_RDTSCP |
2473 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2474 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2475 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2476 SECONDARY_EXEC_WBINVD_EXITING |
2477 SECONDARY_EXEC_XSAVES;
2480 /* nested EPT: emulate EPT also to L1 */
2481 vmx->nested.nested_vmx_secondary_ctls_high |=
2482 SECONDARY_EXEC_ENABLE_EPT;
2483 vmx->nested.nested_vmx_ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
2484 VMX_EPTP_WB_BIT | VMX_EPT_2MB_PAGE_BIT |
2486 vmx->nested.nested_vmx_ept_caps &= vmx_capability.ept;
2488 * For nested guests, we don't do anything specific
2489 * for single context invalidation. Hence, only advertise
2490 * support for global context invalidation.
2492 vmx->nested.nested_vmx_ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT;
2494 vmx->nested.nested_vmx_ept_caps = 0;
2496 if (enable_unrestricted_guest)
2497 vmx->nested.nested_vmx_secondary_ctls_high |=
2498 SECONDARY_EXEC_UNRESTRICTED_GUEST;
2500 /* miscellaneous data */
2501 rdmsr(MSR_IA32_VMX_MISC,
2502 vmx->nested.nested_vmx_misc_low,
2503 vmx->nested.nested_vmx_misc_high);
2504 vmx->nested.nested_vmx_misc_low &= VMX_MISC_SAVE_EFER_LMA;
2505 vmx->nested.nested_vmx_misc_low |=
2506 VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
2507 VMX_MISC_ACTIVITY_HLT;
2508 vmx->nested.nested_vmx_misc_high = 0;
2511 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2514 * Bits 0 in high must be 0, and bits 1 in low must be 1.
2516 return ((control & high) | low) == control;
2519 static inline u64 vmx_control_msr(u32 low, u32 high)
2521 return low | ((u64)high << 32);
2524 /* Returns 0 on success, non-0 otherwise. */
2525 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2527 struct vcpu_vmx *vmx = to_vmx(vcpu);
2529 switch (msr_index) {
2530 case MSR_IA32_VMX_BASIC:
2532 * This MSR reports some information about VMX support. We
2533 * should return information about the VMX we emulate for the
2534 * guest, and the VMCS structure we give it - not about the
2535 * VMX support of the underlying hardware.
2537 *pdata = VMCS12_REVISION | VMX_BASIC_TRUE_CTLS |
2538 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2539 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2541 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2542 case MSR_IA32_VMX_PINBASED_CTLS:
2543 *pdata = vmx_control_msr(
2544 vmx->nested.nested_vmx_pinbased_ctls_low,
2545 vmx->nested.nested_vmx_pinbased_ctls_high);
2547 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2548 *pdata = vmx_control_msr(
2549 vmx->nested.nested_vmx_true_procbased_ctls_low,
2550 vmx->nested.nested_vmx_procbased_ctls_high);
2552 case MSR_IA32_VMX_PROCBASED_CTLS:
2553 *pdata = vmx_control_msr(
2554 vmx->nested.nested_vmx_procbased_ctls_low,
2555 vmx->nested.nested_vmx_procbased_ctls_high);
2557 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2558 *pdata = vmx_control_msr(
2559 vmx->nested.nested_vmx_true_exit_ctls_low,
2560 vmx->nested.nested_vmx_exit_ctls_high);
2562 case MSR_IA32_VMX_EXIT_CTLS:
2563 *pdata = vmx_control_msr(
2564 vmx->nested.nested_vmx_exit_ctls_low,
2565 vmx->nested.nested_vmx_exit_ctls_high);
2567 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2568 *pdata = vmx_control_msr(
2569 vmx->nested.nested_vmx_true_entry_ctls_low,
2570 vmx->nested.nested_vmx_entry_ctls_high);
2572 case MSR_IA32_VMX_ENTRY_CTLS:
2573 *pdata = vmx_control_msr(
2574 vmx->nested.nested_vmx_entry_ctls_low,
2575 vmx->nested.nested_vmx_entry_ctls_high);
2577 case MSR_IA32_VMX_MISC:
2578 *pdata = vmx_control_msr(
2579 vmx->nested.nested_vmx_misc_low,
2580 vmx->nested.nested_vmx_misc_high);
2583 * These MSRs specify bits which the guest must keep fixed (on or off)
2584 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2585 * We picked the standard core2 setting.
2587 #define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2588 #define VMXON_CR4_ALWAYSON X86_CR4_VMXE
2589 case MSR_IA32_VMX_CR0_FIXED0:
2590 *pdata = VMXON_CR0_ALWAYSON;
2592 case MSR_IA32_VMX_CR0_FIXED1:
2595 case MSR_IA32_VMX_CR4_FIXED0:
2596 *pdata = VMXON_CR4_ALWAYSON;
2598 case MSR_IA32_VMX_CR4_FIXED1:
2601 case MSR_IA32_VMX_VMCS_ENUM:
2602 *pdata = 0x2e; /* highest index: VMX_PREEMPTION_TIMER_VALUE */
2604 case MSR_IA32_VMX_PROCBASED_CTLS2:
2605 *pdata = vmx_control_msr(
2606 vmx->nested.nested_vmx_secondary_ctls_low,
2607 vmx->nested.nested_vmx_secondary_ctls_high);
2609 case MSR_IA32_VMX_EPT_VPID_CAP:
2610 /* Currently, no nested vpid support */
2611 *pdata = vmx->nested.nested_vmx_ept_caps;
2621 * Reads an msr value (of 'msr_index') into 'pdata'.
2622 * Returns 0 on success, non-0 otherwise.
2623 * Assumes vcpu_load() was already called.
2625 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2628 struct shared_msr_entry *msr;
2631 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
2635 switch (msr_index) {
2636 #ifdef CONFIG_X86_64
2638 data = vmcs_readl(GUEST_FS_BASE);
2641 data = vmcs_readl(GUEST_GS_BASE);
2643 case MSR_KERNEL_GS_BASE:
2644 vmx_load_host_state(to_vmx(vcpu));
2645 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
2649 return kvm_get_msr_common(vcpu, msr_index, pdata);
2651 data = guest_read_tsc();
2653 case MSR_IA32_SYSENTER_CS:
2654 data = vmcs_read32(GUEST_SYSENTER_CS);
2656 case MSR_IA32_SYSENTER_EIP:
2657 data = vmcs_readl(GUEST_SYSENTER_EIP);
2659 case MSR_IA32_SYSENTER_ESP:
2660 data = vmcs_readl(GUEST_SYSENTER_ESP);
2662 case MSR_IA32_BNDCFGS:
2663 if (!vmx_mpx_supported())
2665 data = vmcs_read64(GUEST_BNDCFGS);
2667 case MSR_IA32_FEATURE_CONTROL:
2668 if (!nested_vmx_allowed(vcpu))
2670 data = to_vmx(vcpu)->nested.msr_ia32_feature_control;
2672 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2673 if (!nested_vmx_allowed(vcpu))
2675 return vmx_get_vmx_msr(vcpu, msr_index, pdata);
2677 if (!vmx_xsaves_supported())
2679 data = vcpu->arch.ia32_xss;
2682 if (!to_vmx(vcpu)->rdtscp_enabled)
2684 /* Otherwise falls through */
2686 msr = find_msr_entry(to_vmx(vcpu), msr_index);
2691 return kvm_get_msr_common(vcpu, msr_index, pdata);
2698 static void vmx_leave_nested(struct kvm_vcpu *vcpu);
2701 * Writes msr value into into the appropriate "register".
2702 * Returns 0 on success, non-0 otherwise.
2703 * Assumes vcpu_load() was already called.
2705 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2707 struct vcpu_vmx *vmx = to_vmx(vcpu);
2708 struct shared_msr_entry *msr;
2710 u32 msr_index = msr_info->index;
2711 u64 data = msr_info->data;
2713 switch (msr_index) {
2715 ret = kvm_set_msr_common(vcpu, msr_info);
2717 #ifdef CONFIG_X86_64
2719 vmx_segment_cache_clear(vmx);
2720 vmcs_writel(GUEST_FS_BASE, data);
2723 vmx_segment_cache_clear(vmx);
2724 vmcs_writel(GUEST_GS_BASE, data);
2726 case MSR_KERNEL_GS_BASE:
2727 vmx_load_host_state(vmx);
2728 vmx->msr_guest_kernel_gs_base = data;
2731 case MSR_IA32_SYSENTER_CS:
2732 vmcs_write32(GUEST_SYSENTER_CS, data);
2734 case MSR_IA32_SYSENTER_EIP:
2735 vmcs_writel(GUEST_SYSENTER_EIP, data);
2737 case MSR_IA32_SYSENTER_ESP:
2738 vmcs_writel(GUEST_SYSENTER_ESP, data);
2740 case MSR_IA32_BNDCFGS:
2741 if (!vmx_mpx_supported())
2743 vmcs_write64(GUEST_BNDCFGS, data);
2746 kvm_write_tsc(vcpu, msr_info);
2748 case MSR_IA32_CR_PAT:
2749 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2750 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
2752 vmcs_write64(GUEST_IA32_PAT, data);
2753 vcpu->arch.pat = data;
2756 ret = kvm_set_msr_common(vcpu, msr_info);
2758 case MSR_IA32_TSC_ADJUST:
2759 ret = kvm_set_msr_common(vcpu, msr_info);
2761 case MSR_IA32_FEATURE_CONTROL:
2762 if (!nested_vmx_allowed(vcpu) ||
2763 (to_vmx(vcpu)->nested.msr_ia32_feature_control &
2764 FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
2766 vmx->nested.msr_ia32_feature_control = data;
2767 if (msr_info->host_initiated && data == 0)
2768 vmx_leave_nested(vcpu);
2770 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2771 return 1; /* they are read-only */
2773 if (!vmx_xsaves_supported())
2776 * The only supported bit as of Skylake is bit 8, but
2777 * it is not supported on KVM.
2781 vcpu->arch.ia32_xss = data;
2782 if (vcpu->arch.ia32_xss != host_xss)
2783 add_atomic_switch_msr(vmx, MSR_IA32_XSS,
2784 vcpu->arch.ia32_xss, host_xss);
2786 clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
2789 if (!vmx->rdtscp_enabled)
2791 /* Check reserved bit, higher 32 bits should be zero */
2792 if ((data >> 32) != 0)
2794 /* Otherwise falls through */
2796 msr = find_msr_entry(vmx, msr_index);
2798 u64 old_msr_data = msr->data;
2800 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
2802 ret = kvm_set_shared_msr(msr->index, msr->data,
2806 msr->data = old_msr_data;
2810 ret = kvm_set_msr_common(vcpu, msr_info);
2816 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2818 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
2821 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2824 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2826 case VCPU_EXREG_PDPTR:
2828 ept_save_pdptrs(vcpu);
2835 static __init int cpu_has_kvm_support(void)
2837 return cpu_has_vmx();
2840 static __init int vmx_disabled_by_bios(void)
2844 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
2845 if (msr & FEATURE_CONTROL_LOCKED) {
2846 /* launched w/ TXT and VMX disabled */
2847 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2850 /* launched w/o TXT and VMX only enabled w/ TXT */
2851 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2852 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2853 && !tboot_enabled()) {
2854 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
2855 "activate TXT before enabling KVM\n");
2858 /* launched w/o TXT and VMX disabled */
2859 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2860 && !tboot_enabled())
2867 static void kvm_cpu_vmxon(u64 addr)
2869 asm volatile (ASM_VMX_VMXON_RAX
2870 : : "a"(&addr), "m"(addr)
2874 static int hardware_enable(void)
2876 int cpu = raw_smp_processor_id();
2877 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2880 if (cr4_read_shadow() & X86_CR4_VMXE)
2883 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2886 * Now we can enable the vmclear operation in kdump
2887 * since the loaded_vmcss_on_cpu list on this cpu
2888 * has been initialized.
2890 * Though the cpu is not in VMX operation now, there
2891 * is no problem to enable the vmclear operation
2892 * for the loaded_vmcss_on_cpu list is empty!
2894 crash_enable_local_vmclear(cpu);
2896 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
2898 test_bits = FEATURE_CONTROL_LOCKED;
2899 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
2900 if (tboot_enabled())
2901 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
2903 if ((old & test_bits) != test_bits) {
2904 /* enable and lock */
2905 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
2907 cr4_set_bits(X86_CR4_VMXE);
2909 if (vmm_exclusive) {
2910 kvm_cpu_vmxon(phys_addr);
2914 native_store_gdt(this_cpu_ptr(&host_gdt));
2919 static void vmclear_local_loaded_vmcss(void)
2921 int cpu = raw_smp_processor_id();
2922 struct loaded_vmcs *v, *n;
2924 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2925 loaded_vmcss_on_cpu_link)
2926 __loaded_vmcs_clear(v);
2930 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2933 static void kvm_cpu_vmxoff(void)
2935 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
2938 static void hardware_disable(void)
2940 if (vmm_exclusive) {
2941 vmclear_local_loaded_vmcss();
2944 cr4_clear_bits(X86_CR4_VMXE);
2947 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2948 u32 msr, u32 *result)
2950 u32 vmx_msr_low, vmx_msr_high;
2951 u32 ctl = ctl_min | ctl_opt;
2953 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2955 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2956 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
2958 /* Ensure minimum (required) set of control bits are supported. */
2966 static __init bool allow_1_setting(u32 msr, u32 ctl)
2968 u32 vmx_msr_low, vmx_msr_high;
2970 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2971 return vmx_msr_high & ctl;
2974 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
2976 u32 vmx_msr_low, vmx_msr_high;
2977 u32 min, opt, min2, opt2;
2978 u32 _pin_based_exec_control = 0;
2979 u32 _cpu_based_exec_control = 0;
2980 u32 _cpu_based_2nd_exec_control = 0;
2981 u32 _vmexit_control = 0;
2982 u32 _vmentry_control = 0;
2984 min = CPU_BASED_HLT_EXITING |
2985 #ifdef CONFIG_X86_64
2986 CPU_BASED_CR8_LOAD_EXITING |
2987 CPU_BASED_CR8_STORE_EXITING |
2989 CPU_BASED_CR3_LOAD_EXITING |
2990 CPU_BASED_CR3_STORE_EXITING |
2991 CPU_BASED_USE_IO_BITMAPS |
2992 CPU_BASED_MOV_DR_EXITING |
2993 CPU_BASED_USE_TSC_OFFSETING |
2994 CPU_BASED_MWAIT_EXITING |
2995 CPU_BASED_MONITOR_EXITING |
2996 CPU_BASED_INVLPG_EXITING |
2997 CPU_BASED_RDPMC_EXITING;
2999 opt = CPU_BASED_TPR_SHADOW |
3000 CPU_BASED_USE_MSR_BITMAPS |
3001 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
3002 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
3003 &_cpu_based_exec_control) < 0)
3005 #ifdef CONFIG_X86_64
3006 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3007 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
3008 ~CPU_BASED_CR8_STORE_EXITING;
3010 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
3012 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
3013 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3014 SECONDARY_EXEC_WBINVD_EXITING |
3015 SECONDARY_EXEC_ENABLE_VPID |
3016 SECONDARY_EXEC_ENABLE_EPT |
3017 SECONDARY_EXEC_UNRESTRICTED_GUEST |
3018 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
3019 SECONDARY_EXEC_RDTSCP |
3020 SECONDARY_EXEC_ENABLE_INVPCID |
3021 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3022 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
3023 SECONDARY_EXEC_SHADOW_VMCS |
3024 SECONDARY_EXEC_XSAVES |
3025 SECONDARY_EXEC_ENABLE_PML;
3026 if (adjust_vmx_controls(min2, opt2,
3027 MSR_IA32_VMX_PROCBASED_CTLS2,
3028 &_cpu_based_2nd_exec_control) < 0)
3031 #ifndef CONFIG_X86_64
3032 if (!(_cpu_based_2nd_exec_control &
3033 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
3034 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
3037 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3038 _cpu_based_2nd_exec_control &= ~(
3039 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3040 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3041 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3043 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
3044 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
3046 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
3047 CPU_BASED_CR3_STORE_EXITING |
3048 CPU_BASED_INVLPG_EXITING);
3049 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
3050 vmx_capability.ept, vmx_capability.vpid);
3053 min = VM_EXIT_SAVE_DEBUG_CONTROLS;
3054 #ifdef CONFIG_X86_64
3055 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
3057 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT |
3058 VM_EXIT_ACK_INTR_ON_EXIT | VM_EXIT_CLEAR_BNDCFGS;
3059 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
3060 &_vmexit_control) < 0)
3063 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
3064 opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR;
3065 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
3066 &_pin_based_exec_control) < 0)
3069 if (!(_cpu_based_2nd_exec_control &
3070 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) ||
3071 !(_vmexit_control & VM_EXIT_ACK_INTR_ON_EXIT))
3072 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
3074 min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
3075 opt = VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS;
3076 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
3077 &_vmentry_control) < 0)
3080 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
3082 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
3083 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
3086 #ifdef CONFIG_X86_64
3087 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
3088 if (vmx_msr_high & (1u<<16))
3092 /* Require Write-Back (WB) memory type for VMCS accesses. */
3093 if (((vmx_msr_high >> 18) & 15) != 6)
3096 vmcs_conf->size = vmx_msr_high & 0x1fff;
3097 vmcs_conf->order = get_order(vmcs_config.size);
3098 vmcs_conf->revision_id = vmx_msr_low;
3100 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
3101 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
3102 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
3103 vmcs_conf->vmexit_ctrl = _vmexit_control;
3104 vmcs_conf->vmentry_ctrl = _vmentry_control;
3106 cpu_has_load_ia32_efer =
3107 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3108 VM_ENTRY_LOAD_IA32_EFER)
3109 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3110 VM_EXIT_LOAD_IA32_EFER);
3112 cpu_has_load_perf_global_ctrl =
3113 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3114 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
3115 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3116 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
3119 * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
3120 * but due to arrata below it can't be used. Workaround is to use
3121 * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
3123 * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
3128 * BC86,AAY89,BD102 (model 44)
3132 if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
3133 switch (boot_cpu_data.x86_model) {
3139 cpu_has_load_perf_global_ctrl = false;
3140 printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
3141 "does not work properly. Using workaround\n");
3149 rdmsrl(MSR_IA32_XSS, host_xss);
3154 static struct vmcs *alloc_vmcs_cpu(int cpu)
3156 int node = cpu_to_node(cpu);
3160 pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
3163 vmcs = page_address(pages);
3164 memset(vmcs, 0, vmcs_config.size);
3165 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
3169 static struct vmcs *alloc_vmcs(void)
3171 return alloc_vmcs_cpu(raw_smp_processor_id());
3174 static void free_vmcs(struct vmcs *vmcs)
3176 free_pages((unsigned long)vmcs, vmcs_config.order);
3180 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
3182 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
3184 if (!loaded_vmcs->vmcs)
3186 loaded_vmcs_clear(loaded_vmcs);
3187 free_vmcs(loaded_vmcs->vmcs);
3188 loaded_vmcs->vmcs = NULL;
3191 static void free_kvm_area(void)
3195 for_each_possible_cpu(cpu) {
3196 free_vmcs(per_cpu(vmxarea, cpu));
3197 per_cpu(vmxarea, cpu) = NULL;
3201 static void init_vmcs_shadow_fields(void)
3205 /* No checks for read only fields yet */
3207 for (i = j = 0; i < max_shadow_read_write_fields; i++) {
3208 switch (shadow_read_write_fields[i]) {
3210 if (!vmx_mpx_supported())
3218 shadow_read_write_fields[j] =
3219 shadow_read_write_fields[i];
3222 max_shadow_read_write_fields = j;
3224 /* shadowed fields guest access without vmexit */
3225 for (i = 0; i < max_shadow_read_write_fields; i++) {
3226 clear_bit(shadow_read_write_fields[i],
3227 vmx_vmwrite_bitmap);
3228 clear_bit(shadow_read_write_fields[i],
3231 for (i = 0; i < max_shadow_read_only_fields; i++)
3232 clear_bit(shadow_read_only_fields[i],
3236 static __init int alloc_kvm_area(void)
3240 for_each_possible_cpu(cpu) {
3243 vmcs = alloc_vmcs_cpu(cpu);
3249 per_cpu(vmxarea, cpu) = vmcs;
3254 static bool emulation_required(struct kvm_vcpu *vcpu)
3256 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
3259 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
3260 struct kvm_segment *save)
3262 if (!emulate_invalid_guest_state) {
3264 * CS and SS RPL should be equal during guest entry according
3265 * to VMX spec, but in reality it is not always so. Since vcpu
3266 * is in the middle of the transition from real mode to
3267 * protected mode it is safe to assume that RPL 0 is a good
3270 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
3271 save->selector &= ~SEGMENT_RPL_MASK;
3272 save->dpl = save->selector & SEGMENT_RPL_MASK;
3275 vmx_set_segment(vcpu, save, seg);
3278 static void enter_pmode(struct kvm_vcpu *vcpu)
3280 unsigned long flags;
3281 struct vcpu_vmx *vmx = to_vmx(vcpu);
3284 * Update real mode segment cache. It may be not up-to-date if sement
3285 * register was written while vcpu was in a guest mode.
3287 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3288 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3289 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3290 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3291 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3292 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3294 vmx->rmode.vm86_active = 0;
3296 vmx_segment_cache_clear(vmx);
3298 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3300 flags = vmcs_readl(GUEST_RFLAGS);
3301 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
3302 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
3303 vmcs_writel(GUEST_RFLAGS, flags);
3305 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
3306 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
3308 update_exception_bitmap(vcpu);
3310 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3311 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3312 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3313 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3314 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3315 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3318 static void fix_rmode_seg(int seg, struct kvm_segment *save)
3320 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3321 struct kvm_segment var = *save;
3324 if (seg == VCPU_SREG_CS)
3327 if (!emulate_invalid_guest_state) {
3328 var.selector = var.base >> 4;
3329 var.base = var.base & 0xffff0;
3339 if (save->base & 0xf)
3340 printk_once(KERN_WARNING "kvm: segment base is not "
3341 "paragraph aligned when entering "
3342 "protected mode (seg=%d)", seg);
3345 vmcs_write16(sf->selector, var.selector);
3346 vmcs_write32(sf->base, var.base);
3347 vmcs_write32(sf->limit, var.limit);
3348 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
3351 static void enter_rmode(struct kvm_vcpu *vcpu)
3353 unsigned long flags;
3354 struct vcpu_vmx *vmx = to_vmx(vcpu);
3356 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3357 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3358 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3359 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3360 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3361 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3362 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3364 vmx->rmode.vm86_active = 1;
3367 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
3368 * vcpu. Warn the user that an update is overdue.
3370 if (!vcpu->kvm->arch.tss_addr)
3371 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
3372 "called before entering vcpu\n");
3374 vmx_segment_cache_clear(vmx);
3376 vmcs_writel(GUEST_TR_BASE, vcpu->kvm->arch.tss_addr);
3377 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
3378 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3380 flags = vmcs_readl(GUEST_RFLAGS);
3381 vmx->rmode.save_rflags = flags;
3383 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
3385 vmcs_writel(GUEST_RFLAGS, flags);
3386 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
3387 update_exception_bitmap(vcpu);
3389 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3390 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3391 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3392 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3393 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3394 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3396 kvm_mmu_reset_context(vcpu);
3399 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
3401 struct vcpu_vmx *vmx = to_vmx(vcpu);
3402 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
3408 * Force kernel_gs_base reloading before EFER changes, as control
3409 * of this msr depends on is_long_mode().
3411 vmx_load_host_state(to_vmx(vcpu));
3412 vcpu->arch.efer = efer;
3413 if (efer & EFER_LMA) {
3414 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3417 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3419 msr->data = efer & ~EFER_LME;
3424 #ifdef CONFIG_X86_64
3426 static void enter_lmode(struct kvm_vcpu *vcpu)
3430 vmx_segment_cache_clear(to_vmx(vcpu));
3432 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
3433 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
3434 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
3436 vmcs_write32(GUEST_TR_AR_BYTES,
3437 (guest_tr_ar & ~AR_TYPE_MASK)
3438 | AR_TYPE_BUSY_64_TSS);
3440 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
3443 static void exit_lmode(struct kvm_vcpu *vcpu)
3445 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3446 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
3451 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
3453 vpid_sync_context(to_vmx(vcpu));
3455 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
3457 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
3461 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
3463 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
3465 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
3466 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
3469 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
3471 if (enable_ept && is_paging(vcpu))
3472 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3473 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
3476 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
3478 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
3480 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
3481 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
3484 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
3486 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3488 if (!test_bit(VCPU_EXREG_PDPTR,
3489 (unsigned long *)&vcpu->arch.regs_dirty))
3492 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3493 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
3494 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
3495 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
3496 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
3500 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
3502 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3504 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3505 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
3506 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
3507 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
3508 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
3511 __set_bit(VCPU_EXREG_PDPTR,
3512 (unsigned long *)&vcpu->arch.regs_avail);
3513 __set_bit(VCPU_EXREG_PDPTR,
3514 (unsigned long *)&vcpu->arch.regs_dirty);
3517 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
3519 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
3521 struct kvm_vcpu *vcpu)
3523 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3524 vmx_decache_cr3(vcpu);
3525 if (!(cr0 & X86_CR0_PG)) {
3526 /* From paging/starting to nonpaging */
3527 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3528 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
3529 (CPU_BASED_CR3_LOAD_EXITING |
3530 CPU_BASED_CR3_STORE_EXITING));
3531 vcpu->arch.cr0 = cr0;
3532 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3533 } else if (!is_paging(vcpu)) {
3534 /* From nonpaging to paging */
3535 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3536 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
3537 ~(CPU_BASED_CR3_LOAD_EXITING |
3538 CPU_BASED_CR3_STORE_EXITING));
3539 vcpu->arch.cr0 = cr0;
3540 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3543 if (!(cr0 & X86_CR0_WP))
3544 *hw_cr0 &= ~X86_CR0_WP;
3547 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
3549 struct vcpu_vmx *vmx = to_vmx(vcpu);
3550 unsigned long hw_cr0;
3552 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK);
3553 if (enable_unrestricted_guest)
3554 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
3556 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
3558 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
3561 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
3565 #ifdef CONFIG_X86_64
3566 if (vcpu->arch.efer & EFER_LME) {
3567 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
3569 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
3575 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
3577 if (!vcpu->fpu_active)
3578 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
3580 vmcs_writel(CR0_READ_SHADOW, cr0);
3581 vmcs_writel(GUEST_CR0, hw_cr0);
3582 vcpu->arch.cr0 = cr0;
3584 /* depends on vcpu->arch.cr0 to be set to a new value */
3585 vmx->emulation_required = emulation_required(vcpu);
3588 static u64 construct_eptp(unsigned long root_hpa)
3592 /* TODO write the value reading from MSR */
3593 eptp = VMX_EPT_DEFAULT_MT |
3594 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
3595 if (enable_ept_ad_bits)
3596 eptp |= VMX_EPT_AD_ENABLE_BIT;
3597 eptp |= (root_hpa & PAGE_MASK);
3602 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
3604 unsigned long guest_cr3;
3609 eptp = construct_eptp(cr3);
3610 vmcs_write64(EPT_POINTER, eptp);
3611 if (is_paging(vcpu) || is_guest_mode(vcpu))
3612 guest_cr3 = kvm_read_cr3(vcpu);
3614 guest_cr3 = vcpu->kvm->arch.ept_identity_map_addr;
3615 ept_load_pdptrs(vcpu);
3618 vmx_flush_tlb(vcpu);
3619 vmcs_writel(GUEST_CR3, guest_cr3);
3622 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3625 * Pass through host's Machine Check Enable value to hw_cr4, which
3626 * is in force while we are in guest mode. Do not let guests control
3627 * this bit, even if host CR4.MCE == 0.
3629 unsigned long hw_cr4 =
3630 (cr4_read_shadow() & X86_CR4_MCE) |
3631 (cr4 & ~X86_CR4_MCE) |
3632 (to_vmx(vcpu)->rmode.vm86_active ?
3633 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
3635 if (cr4 & X86_CR4_VMXE) {
3637 * To use VMXON (and later other VMX instructions), a guest
3638 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3639 * So basically the check on whether to allow nested VMX
3642 if (!nested_vmx_allowed(vcpu))
3645 if (to_vmx(vcpu)->nested.vmxon &&
3646 ((cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON))
3649 vcpu->arch.cr4 = cr4;
3651 if (!is_paging(vcpu)) {
3652 hw_cr4 &= ~X86_CR4_PAE;
3653 hw_cr4 |= X86_CR4_PSE;
3655 * SMEP/SMAP is disabled if CPU is in non-paging mode
3656 * in hardware. However KVM always uses paging mode to
3657 * emulate guest non-paging mode with TDP.
3658 * To emulate this behavior, SMEP/SMAP needs to be
3659 * manually disabled when guest switches to non-paging
3662 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP);
3663 } else if (!(cr4 & X86_CR4_PAE)) {
3664 hw_cr4 &= ~X86_CR4_PAE;
3668 vmcs_writel(CR4_READ_SHADOW, cr4);
3669 vmcs_writel(GUEST_CR4, hw_cr4);
3673 static void vmx_get_segment(struct kvm_vcpu *vcpu,
3674 struct kvm_segment *var, int seg)
3676 struct vcpu_vmx *vmx = to_vmx(vcpu);
3679 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3680 *var = vmx->rmode.segs[seg];
3681 if (seg == VCPU_SREG_TR
3682 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3684 var->base = vmx_read_guest_seg_base(vmx, seg);
3685 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3688 var->base = vmx_read_guest_seg_base(vmx, seg);
3689 var->limit = vmx_read_guest_seg_limit(vmx, seg);
3690 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3691 ar = vmx_read_guest_seg_ar(vmx, seg);
3692 var->unusable = (ar >> 16) & 1;
3693 var->type = ar & 15;
3694 var->s = (ar >> 4) & 1;
3695 var->dpl = (ar >> 5) & 3;
3697 * Some userspaces do not preserve unusable property. Since usable
3698 * segment has to be present according to VMX spec we can use present
3699 * property to amend userspace bug by making unusable segment always
3700 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3701 * segment as unusable.
3703 var->present = !var->unusable;
3704 var->avl = (ar >> 12) & 1;
3705 var->l = (ar >> 13) & 1;
3706 var->db = (ar >> 14) & 1;
3707 var->g = (ar >> 15) & 1;
3710 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3712 struct kvm_segment s;
3714 if (to_vmx(vcpu)->rmode.vm86_active) {
3715 vmx_get_segment(vcpu, &s, seg);
3718 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3721 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
3723 struct vcpu_vmx *vmx = to_vmx(vcpu);
3725 if (unlikely(vmx->rmode.vm86_active))
3728 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3733 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3737 if (var->unusable || !var->present)
3740 ar = var->type & 15;
3741 ar |= (var->s & 1) << 4;
3742 ar |= (var->dpl & 3) << 5;
3743 ar |= (var->present & 1) << 7;
3744 ar |= (var->avl & 1) << 12;
3745 ar |= (var->l & 1) << 13;
3746 ar |= (var->db & 1) << 14;
3747 ar |= (var->g & 1) << 15;
3753 static void vmx_set_segment(struct kvm_vcpu *vcpu,
3754 struct kvm_segment *var, int seg)
3756 struct vcpu_vmx *vmx = to_vmx(vcpu);
3757 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3759 vmx_segment_cache_clear(vmx);
3761 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3762 vmx->rmode.segs[seg] = *var;
3763 if (seg == VCPU_SREG_TR)
3764 vmcs_write16(sf->selector, var->selector);
3766 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3770 vmcs_writel(sf->base, var->base);
3771 vmcs_write32(sf->limit, var->limit);
3772 vmcs_write16(sf->selector, var->selector);
3775 * Fix the "Accessed" bit in AR field of segment registers for older
3777 * IA32 arch specifies that at the time of processor reset the
3778 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3779 * is setting it to 0 in the userland code. This causes invalid guest
3780 * state vmexit when "unrestricted guest" mode is turned on.
3781 * Fix for this setup issue in cpu_reset is being pushed in the qemu
3782 * tree. Newer qemu binaries with that qemu fix would not need this
3785 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3786 var->type |= 0x1; /* Accessed */
3788 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3791 vmx->emulation_required = emulation_required(vcpu);
3794 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3796 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3798 *db = (ar >> 14) & 1;
3799 *l = (ar >> 13) & 1;
3802 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3804 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3805 dt->address = vmcs_readl(GUEST_IDTR_BASE);
3808 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3810 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3811 vmcs_writel(GUEST_IDTR_BASE, dt->address);
3814 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3816 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3817 dt->address = vmcs_readl(GUEST_GDTR_BASE);
3820 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3822 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3823 vmcs_writel(GUEST_GDTR_BASE, dt->address);
3826 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3828 struct kvm_segment var;
3831 vmx_get_segment(vcpu, &var, seg);
3833 if (seg == VCPU_SREG_CS)
3835 ar = vmx_segment_access_rights(&var);
3837 if (var.base != (var.selector << 4))
3839 if (var.limit != 0xffff)
3847 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3849 struct kvm_segment cs;
3850 unsigned int cs_rpl;
3852 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3853 cs_rpl = cs.selector & SEGMENT_RPL_MASK;
3857 if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
3861 if (cs.type & AR_TYPE_WRITEABLE_MASK) {
3862 if (cs.dpl > cs_rpl)
3865 if (cs.dpl != cs_rpl)
3871 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3875 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3877 struct kvm_segment ss;
3878 unsigned int ss_rpl;
3880 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3881 ss_rpl = ss.selector & SEGMENT_RPL_MASK;
3885 if (ss.type != 3 && ss.type != 7)
3889 if (ss.dpl != ss_rpl) /* DPL != RPL */
3897 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3899 struct kvm_segment var;
3902 vmx_get_segment(vcpu, &var, seg);
3903 rpl = var.selector & SEGMENT_RPL_MASK;
3911 if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
3912 if (var.dpl < rpl) /* DPL < RPL */
3916 /* TODO: Add other members to kvm_segment_field to allow checking for other access
3922 static bool tr_valid(struct kvm_vcpu *vcpu)
3924 struct kvm_segment tr;
3926 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3930 if (tr.selector & SEGMENT_TI_MASK) /* TI = 1 */
3932 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3940 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3942 struct kvm_segment ldtr;
3944 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3948 if (ldtr.selector & SEGMENT_TI_MASK) /* TI = 1 */
3958 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3960 struct kvm_segment cs, ss;
3962 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3963 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3965 return ((cs.selector & SEGMENT_RPL_MASK) ==
3966 (ss.selector & SEGMENT_RPL_MASK));
3970 * Check if guest state is valid. Returns true if valid, false if
3972 * We assume that registers are always usable
3974 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3976 if (enable_unrestricted_guest)
3979 /* real mode guest state checks */
3980 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3981 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3983 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3985 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3987 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3989 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3991 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3994 /* protected mode guest state checks */
3995 if (!cs_ss_rpl_check(vcpu))
3997 if (!code_segment_valid(vcpu))
3999 if (!stack_segment_valid(vcpu))
4001 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
4003 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
4005 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
4007 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
4009 if (!tr_valid(vcpu))
4011 if (!ldtr_valid(vcpu))
4015 * - Add checks on RIP
4016 * - Add checks on RFLAGS
4022 static int init_rmode_tss(struct kvm *kvm)
4028 idx = srcu_read_lock(&kvm->srcu);
4029 fn = kvm->arch.tss_addr >> PAGE_SHIFT;
4030 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4033 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
4034 r = kvm_write_guest_page(kvm, fn++, &data,
4035 TSS_IOPB_BASE_OFFSET, sizeof(u16));
4038 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
4041 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4045 r = kvm_write_guest_page(kvm, fn, &data,
4046 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
4049 srcu_read_unlock(&kvm->srcu, idx);
4053 static int init_rmode_identity_map(struct kvm *kvm)
4056 pfn_t identity_map_pfn;
4062 /* Protect kvm->arch.ept_identity_pagetable_done. */
4063 mutex_lock(&kvm->slots_lock);
4065 if (likely(kvm->arch.ept_identity_pagetable_done))
4068 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
4070 r = alloc_identity_pagetable(kvm);
4074 idx = srcu_read_lock(&kvm->srcu);
4075 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
4078 /* Set up identity-mapping pagetable for EPT in real mode */
4079 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
4080 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
4081 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
4082 r = kvm_write_guest_page(kvm, identity_map_pfn,
4083 &tmp, i * sizeof(tmp), sizeof(tmp));
4087 kvm->arch.ept_identity_pagetable_done = true;
4090 srcu_read_unlock(&kvm->srcu, idx);
4093 mutex_unlock(&kvm->slots_lock);
4097 static void seg_setup(int seg)
4099 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4102 vmcs_write16(sf->selector, 0);
4103 vmcs_writel(sf->base, 0);
4104 vmcs_write32(sf->limit, 0xffff);
4106 if (seg == VCPU_SREG_CS)
4107 ar |= 0x08; /* code segment */
4109 vmcs_write32(sf->ar_bytes, ar);
4112 static int alloc_apic_access_page(struct kvm *kvm)
4115 struct kvm_userspace_memory_region kvm_userspace_mem;
4118 mutex_lock(&kvm->slots_lock);
4119 if (kvm->arch.apic_access_page_done)
4121 kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
4122 kvm_userspace_mem.flags = 0;
4123 kvm_userspace_mem.guest_phys_addr = APIC_DEFAULT_PHYS_BASE;
4124 kvm_userspace_mem.memory_size = PAGE_SIZE;
4125 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem);
4129 page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
4130 if (is_error_page(page)) {
4136 * Do not pin the page in memory, so that memory hot-unplug
4137 * is able to migrate it.
4140 kvm->arch.apic_access_page_done = true;
4142 mutex_unlock(&kvm->slots_lock);
4146 static int alloc_identity_pagetable(struct kvm *kvm)
4148 /* Called with kvm->slots_lock held. */
4150 struct kvm_userspace_memory_region kvm_userspace_mem;
4153 BUG_ON(kvm->arch.ept_identity_pagetable_done);
4155 kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
4156 kvm_userspace_mem.flags = 0;
4157 kvm_userspace_mem.guest_phys_addr =
4158 kvm->arch.ept_identity_map_addr;
4159 kvm_userspace_mem.memory_size = PAGE_SIZE;
4160 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem);
4165 static void allocate_vpid(struct vcpu_vmx *vmx)
4172 spin_lock(&vmx_vpid_lock);
4173 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
4174 if (vpid < VMX_NR_VPIDS) {
4176 __set_bit(vpid, vmx_vpid_bitmap);
4178 spin_unlock(&vmx_vpid_lock);
4181 static void free_vpid(struct vcpu_vmx *vmx)
4185 spin_lock(&vmx_vpid_lock);
4187 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
4188 spin_unlock(&vmx_vpid_lock);
4191 #define MSR_TYPE_R 1
4192 #define MSR_TYPE_W 2
4193 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
4196 int f = sizeof(unsigned long);
4198 if (!cpu_has_vmx_msr_bitmap())
4202 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4203 * have the write-low and read-high bitmap offsets the wrong way round.
4204 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4206 if (msr <= 0x1fff) {
4207 if (type & MSR_TYPE_R)
4209 __clear_bit(msr, msr_bitmap + 0x000 / f);
4211 if (type & MSR_TYPE_W)
4213 __clear_bit(msr, msr_bitmap + 0x800 / f);
4215 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4217 if (type & MSR_TYPE_R)
4219 __clear_bit(msr, msr_bitmap + 0x400 / f);
4221 if (type & MSR_TYPE_W)
4223 __clear_bit(msr, msr_bitmap + 0xc00 / f);
4228 static void __vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
4231 int f = sizeof(unsigned long);
4233 if (!cpu_has_vmx_msr_bitmap())
4237 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4238 * have the write-low and read-high bitmap offsets the wrong way round.
4239 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4241 if (msr <= 0x1fff) {
4242 if (type & MSR_TYPE_R)
4244 __set_bit(msr, msr_bitmap + 0x000 / f);
4246 if (type & MSR_TYPE_W)
4248 __set_bit(msr, msr_bitmap + 0x800 / f);
4250 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4252 if (type & MSR_TYPE_R)
4254 __set_bit(msr, msr_bitmap + 0x400 / f);
4256 if (type & MSR_TYPE_W)
4258 __set_bit(msr, msr_bitmap + 0xc00 / f);
4264 * If a msr is allowed by L0, we should check whether it is allowed by L1.
4265 * The corresponding bit will be cleared unless both of L0 and L1 allow it.
4267 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
4268 unsigned long *msr_bitmap_nested,
4271 int f = sizeof(unsigned long);
4273 if (!cpu_has_vmx_msr_bitmap()) {
4279 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4280 * have the write-low and read-high bitmap offsets the wrong way round.
4281 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4283 if (msr <= 0x1fff) {
4284 if (type & MSR_TYPE_R &&
4285 !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
4287 __clear_bit(msr, msr_bitmap_nested + 0x000 / f);
4289 if (type & MSR_TYPE_W &&
4290 !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
4292 __clear_bit(msr, msr_bitmap_nested + 0x800 / f);
4294 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4296 if (type & MSR_TYPE_R &&
4297 !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
4299 __clear_bit(msr, msr_bitmap_nested + 0x400 / f);
4301 if (type & MSR_TYPE_W &&
4302 !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
4304 __clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
4309 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
4312 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy,
4313 msr, MSR_TYPE_R | MSR_TYPE_W);
4314 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode,
4315 msr, MSR_TYPE_R | MSR_TYPE_W);
4318 static void vmx_enable_intercept_msr_read_x2apic(u32 msr)
4320 __vmx_enable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4322 __vmx_enable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4326 static void vmx_disable_intercept_msr_read_x2apic(u32 msr)
4328 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4330 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4334 static void vmx_disable_intercept_msr_write_x2apic(u32 msr)
4336 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4338 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4342 static int vmx_vm_has_apicv(struct kvm *kvm)
4344 return enable_apicv && irqchip_in_kernel(kvm);
4347 static int vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
4349 struct vcpu_vmx *vmx = to_vmx(vcpu);
4354 if (vmx->nested.pi_desc &&
4355 vmx->nested.pi_pending) {
4356 vmx->nested.pi_pending = false;
4357 if (!pi_test_and_clear_on(vmx->nested.pi_desc))
4360 max_irr = find_last_bit(
4361 (unsigned long *)vmx->nested.pi_desc->pir, 256);
4366 vapic_page = kmap(vmx->nested.virtual_apic_page);
4371 __kvm_apic_update_irr(vmx->nested.pi_desc->pir, vapic_page);
4372 kunmap(vmx->nested.virtual_apic_page);
4374 status = vmcs_read16(GUEST_INTR_STATUS);
4375 if ((u8)max_irr > ((u8)status & 0xff)) {
4377 status |= (u8)max_irr;
4378 vmcs_write16(GUEST_INTR_STATUS, status);
4384 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu)
4387 if (vcpu->mode == IN_GUEST_MODE) {
4388 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu),
4389 POSTED_INTR_VECTOR);
4396 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
4399 struct vcpu_vmx *vmx = to_vmx(vcpu);
4401 if (is_guest_mode(vcpu) &&
4402 vector == vmx->nested.posted_intr_nv) {
4403 /* the PIR and ON have been set by L1. */
4404 kvm_vcpu_trigger_posted_interrupt(vcpu);
4406 * If a posted intr is not recognized by hardware,
4407 * we will accomplish it in the next vmentry.
4409 vmx->nested.pi_pending = true;
4410 kvm_make_request(KVM_REQ_EVENT, vcpu);
4416 * Send interrupt to vcpu via posted interrupt way.
4417 * 1. If target vcpu is running(non-root mode), send posted interrupt
4418 * notification to vcpu and hardware will sync PIR to vIRR atomically.
4419 * 2. If target vcpu isn't running(root mode), kick it to pick up the
4420 * interrupt from PIR in next vmentry.
4422 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
4424 struct vcpu_vmx *vmx = to_vmx(vcpu);
4427 r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
4431 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
4434 r = pi_test_and_set_on(&vmx->pi_desc);
4435 kvm_make_request(KVM_REQ_EVENT, vcpu);
4436 if (r || !kvm_vcpu_trigger_posted_interrupt(vcpu))
4437 kvm_vcpu_kick(vcpu);
4440 static void vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
4442 struct vcpu_vmx *vmx = to_vmx(vcpu);
4444 if (!pi_test_and_clear_on(&vmx->pi_desc))
4447 kvm_apic_update_irr(vcpu, vmx->pi_desc.pir);
4450 static void vmx_sync_pir_to_irr_dummy(struct kvm_vcpu *vcpu)
4456 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
4457 * will not change in the lifetime of the guest.
4458 * Note that host-state that does change is set elsewhere. E.g., host-state
4459 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
4461 static void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
4468 vmcs_writel(HOST_CR0, read_cr0() & ~X86_CR0_TS); /* 22.2.3 */
4469 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
4471 /* Save the most likely value for this task's CR4 in the VMCS. */
4472 cr4 = cr4_read_shadow();
4473 vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
4474 vmx->host_state.vmcs_host_cr4 = cr4;
4476 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
4477 #ifdef CONFIG_X86_64
4479 * Load null selectors, so we can avoid reloading them in
4480 * __vmx_load_host_state(), in case userspace uses the null selectors
4481 * too (the expected case).
4483 vmcs_write16(HOST_DS_SELECTOR, 0);
4484 vmcs_write16(HOST_ES_SELECTOR, 0);
4486 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4487 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4489 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4490 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
4492 native_store_idt(&dt);
4493 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
4494 vmx->host_idt_base = dt.address;
4496 vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
4498 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
4499 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
4500 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
4501 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
4503 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
4504 rdmsr(MSR_IA32_CR_PAT, low32, high32);
4505 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
4509 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
4511 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
4513 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
4514 if (is_guest_mode(&vmx->vcpu))
4515 vmx->vcpu.arch.cr4_guest_owned_bits &=
4516 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
4517 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
4520 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
4522 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
4524 if (!vmx_vm_has_apicv(vmx->vcpu.kvm))
4525 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
4526 return pin_based_exec_ctrl;
4529 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
4531 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
4533 if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
4534 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
4536 if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
4537 exec_control &= ~CPU_BASED_TPR_SHADOW;
4538 #ifdef CONFIG_X86_64
4539 exec_control |= CPU_BASED_CR8_STORE_EXITING |
4540 CPU_BASED_CR8_LOAD_EXITING;
4544 exec_control |= CPU_BASED_CR3_STORE_EXITING |
4545 CPU_BASED_CR3_LOAD_EXITING |
4546 CPU_BASED_INVLPG_EXITING;
4547 return exec_control;
4550 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
4552 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4553 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
4554 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4556 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4558 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4559 enable_unrestricted_guest = 0;
4560 /* Enable INVPCID for non-ept guests may cause performance regression. */
4561 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4563 if (!enable_unrestricted_guest)
4564 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4566 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4567 if (!vmx_vm_has_apicv(vmx->vcpu.kvm))
4568 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4569 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4570 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4571 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4573 We can NOT enable shadow_vmcs here because we don't have yet
4576 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4577 /* PML is enabled/disabled in creating/destorying vcpu */
4578 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4580 return exec_control;
4583 static void ept_set_mmio_spte_mask(void)
4586 * EPT Misconfigurations can be generated if the value of bits 2:0
4587 * of an EPT paging-structure entry is 110b (write/execute).
4588 * Also, magic bits (0x3ull << 62) is set to quickly identify mmio
4591 kvm_mmu_set_mmio_spte_mask((0x3ull << 62) | 0x6ull);
4594 #define VMX_XSS_EXIT_BITMAP 0
4596 * Sets up the vmcs for emulated real mode.
4598 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
4600 #ifdef CONFIG_X86_64
4606 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
4607 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
4609 if (enable_shadow_vmcs) {
4610 vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
4611 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
4613 if (cpu_has_vmx_msr_bitmap())
4614 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
4616 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4619 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
4621 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
4623 if (cpu_has_secondary_exec_ctrls()) {
4624 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
4625 vmx_secondary_exec_control(vmx));
4628 if (vmx_vm_has_apicv(vmx->vcpu.kvm)) {
4629 vmcs_write64(EOI_EXIT_BITMAP0, 0);
4630 vmcs_write64(EOI_EXIT_BITMAP1, 0);
4631 vmcs_write64(EOI_EXIT_BITMAP2, 0);
4632 vmcs_write64(EOI_EXIT_BITMAP3, 0);
4634 vmcs_write16(GUEST_INTR_STATUS, 0);
4636 vmcs_write64(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4637 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4641 vmcs_write32(PLE_GAP, ple_gap);
4642 vmx->ple_window = ple_window;
4643 vmx->ple_window_dirty = true;
4646 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4647 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4648 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
4650 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
4651 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
4652 vmx_set_constant_host_state(vmx);
4653 #ifdef CONFIG_X86_64
4654 rdmsrl(MSR_FS_BASE, a);
4655 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
4656 rdmsrl(MSR_GS_BASE, a);
4657 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
4659 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4660 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4663 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4664 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4665 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
4666 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4667 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
4669 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4670 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4672 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
4673 u32 index = vmx_msr_index[i];
4674 u32 data_low, data_high;
4677 if (rdmsr_safe(index, &data_low, &data_high) < 0)
4679 if (wrmsr_safe(index, data_low, data_high) < 0)
4681 vmx->guest_msrs[j].index = i;
4682 vmx->guest_msrs[j].data = 0;
4683 vmx->guest_msrs[j].mask = -1ull;
4688 vm_exit_controls_init(vmx, vmcs_config.vmexit_ctrl);
4690 /* 22.2.1, 20.8.1 */
4691 vm_entry_controls_init(vmx, vmcs_config.vmentry_ctrl);
4693 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
4694 set_cr4_guest_host_mask(vmx);
4696 if (vmx_xsaves_supported())
4697 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4702 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4704 struct vcpu_vmx *vmx = to_vmx(vcpu);
4705 struct msr_data apic_base_msr;
4708 vmx->rmode.vm86_active = 0;
4710 vmx->soft_vnmi_blocked = 0;
4712 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4713 kvm_set_cr8(vcpu, 0);
4716 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
4717 MSR_IA32_APICBASE_ENABLE;
4718 if (kvm_vcpu_is_reset_bsp(vcpu))
4719 apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4720 apic_base_msr.host_initiated = true;
4721 kvm_set_apic_base(vcpu, &apic_base_msr);
4724 vmx_segment_cache_clear(vmx);
4726 seg_setup(VCPU_SREG_CS);
4727 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4728 vmcs_write32(GUEST_CS_BASE, 0xffff0000);
4730 seg_setup(VCPU_SREG_DS);
4731 seg_setup(VCPU_SREG_ES);
4732 seg_setup(VCPU_SREG_FS);
4733 seg_setup(VCPU_SREG_GS);
4734 seg_setup(VCPU_SREG_SS);
4736 vmcs_write16(GUEST_TR_SELECTOR, 0);
4737 vmcs_writel(GUEST_TR_BASE, 0);
4738 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4739 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4741 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4742 vmcs_writel(GUEST_LDTR_BASE, 0);
4743 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4744 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4747 vmcs_write32(GUEST_SYSENTER_CS, 0);
4748 vmcs_writel(GUEST_SYSENTER_ESP, 0);
4749 vmcs_writel(GUEST_SYSENTER_EIP, 0);
4750 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4753 vmcs_writel(GUEST_RFLAGS, 0x02);
4754 kvm_rip_write(vcpu, 0xfff0);
4756 vmcs_writel(GUEST_GDTR_BASE, 0);
4757 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4759 vmcs_writel(GUEST_IDTR_BASE, 0);
4760 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4762 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4763 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4764 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4768 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
4770 if (cpu_has_vmx_tpr_shadow() && !init_event) {
4771 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4772 if (vm_need_tpr_shadow(vcpu->kvm))
4773 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4774 __pa(vcpu->arch.apic->regs));
4775 vmcs_write32(TPR_THRESHOLD, 0);
4778 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4780 if (vmx_vm_has_apicv(vcpu->kvm))
4781 memset(&vmx->pi_desc, 0, sizeof(struct pi_desc));
4784 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4786 cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4787 vmx_set_cr0(vcpu, cr0); /* enter rmode */
4788 vmx->vcpu.arch.cr0 = cr0;
4789 vmx_set_cr4(vcpu, 0);
4791 vmx_set_efer(vcpu, 0);
4792 vmx_fpu_activate(vcpu);
4793 update_exception_bitmap(vcpu);
4795 vpid_sync_context(vmx);
4799 * In nested virtualization, check if L1 asked to exit on external interrupts.
4800 * For most existing hypervisors, this will always return true.
4802 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
4804 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
4805 PIN_BASED_EXT_INTR_MASK;
4809 * In nested virtualization, check if L1 has set
4810 * VM_EXIT_ACK_INTR_ON_EXIT
4812 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
4814 return get_vmcs12(vcpu)->vm_exit_controls &
4815 VM_EXIT_ACK_INTR_ON_EXIT;
4818 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
4820 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
4821 PIN_BASED_NMI_EXITING;
4824 static void enable_irq_window(struct kvm_vcpu *vcpu)
4826 u32 cpu_based_vm_exec_control;
4828 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4829 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
4830 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4833 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4835 u32 cpu_based_vm_exec_control;
4837 if (!cpu_has_virtual_nmis() ||
4838 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4839 enable_irq_window(vcpu);
4843 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4844 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
4845 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4848 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4850 struct vcpu_vmx *vmx = to_vmx(vcpu);
4852 int irq = vcpu->arch.interrupt.nr;
4854 trace_kvm_inj_virq(irq);
4856 ++vcpu->stat.irq_injections;
4857 if (vmx->rmode.vm86_active) {
4859 if (vcpu->arch.interrupt.soft)
4860 inc_eip = vcpu->arch.event_exit_inst_len;
4861 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
4862 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4865 intr = irq | INTR_INFO_VALID_MASK;
4866 if (vcpu->arch.interrupt.soft) {
4867 intr |= INTR_TYPE_SOFT_INTR;
4868 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4869 vmx->vcpu.arch.event_exit_inst_len);
4871 intr |= INTR_TYPE_EXT_INTR;
4872 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4875 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4877 struct vcpu_vmx *vmx = to_vmx(vcpu);
4879 if (is_guest_mode(vcpu))
4882 if (!cpu_has_virtual_nmis()) {
4884 * Tracking the NMI-blocked state in software is built upon
4885 * finding the next open IRQ window. This, in turn, depends on
4886 * well-behaving guests: They have to keep IRQs disabled at
4887 * least as long as the NMI handler runs. Otherwise we may
4888 * cause NMI nesting, maybe breaking the guest. But as this is
4889 * highly unlikely, we can live with the residual risk.
4891 vmx->soft_vnmi_blocked = 1;
4892 vmx->vnmi_blocked_time = 0;
4895 ++vcpu->stat.nmi_injections;
4896 vmx->nmi_known_unmasked = false;
4897 if (vmx->rmode.vm86_active) {
4898 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
4899 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4902 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4903 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4906 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4908 if (!cpu_has_virtual_nmis())
4909 return to_vmx(vcpu)->soft_vnmi_blocked;
4910 if (to_vmx(vcpu)->nmi_known_unmasked)
4912 return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4915 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4917 struct vcpu_vmx *vmx = to_vmx(vcpu);
4919 if (!cpu_has_virtual_nmis()) {
4920 if (vmx->soft_vnmi_blocked != masked) {
4921 vmx->soft_vnmi_blocked = masked;
4922 vmx->vnmi_blocked_time = 0;
4925 vmx->nmi_known_unmasked = !masked;
4927 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4928 GUEST_INTR_STATE_NMI);
4930 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4931 GUEST_INTR_STATE_NMI);
4935 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4937 if (to_vmx(vcpu)->nested.nested_run_pending)
4940 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
4943 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4944 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4945 | GUEST_INTR_STATE_NMI));
4948 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4950 return (!to_vmx(vcpu)->nested.nested_run_pending &&
4951 vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4952 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4953 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4956 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4959 struct kvm_userspace_memory_region tss_mem = {
4960 .slot = TSS_PRIVATE_MEMSLOT,
4961 .guest_phys_addr = addr,
4962 .memory_size = PAGE_SIZE * 3,
4966 ret = kvm_set_memory_region(kvm, &tss_mem);
4969 kvm->arch.tss_addr = addr;
4970 return init_rmode_tss(kvm);
4973 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4978 * Update instruction length as we may reinject the exception
4979 * from user space while in guest debugging mode.
4981 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4982 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4983 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4987 if (vcpu->guest_debug &
4988 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5005 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
5006 int vec, u32 err_code)
5009 * Instruction with address size override prefix opcode 0x67
5010 * Cause the #SS fault with 0 error code in VM86 mode.
5012 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
5013 if (emulate_instruction(vcpu, 0) == EMULATE_DONE) {
5014 if (vcpu->arch.halt_request) {
5015 vcpu->arch.halt_request = 0;
5016 return kvm_vcpu_halt(vcpu);
5024 * Forward all other exceptions that are valid in real mode.
5025 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
5026 * the required debugging infrastructure rework.
5028 kvm_queue_exception(vcpu, vec);
5033 * Trigger machine check on the host. We assume all the MSRs are already set up
5034 * by the CPU and that we still run on the same CPU as the MCE occurred on.
5035 * We pass a fake environment to the machine check handler because we want
5036 * the guest to be always treated like user space, no matter what context
5037 * it used internally.
5039 static void kvm_machine_check(void)
5041 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
5042 struct pt_regs regs = {
5043 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
5044 .flags = X86_EFLAGS_IF,
5047 do_machine_check(®s, 0);
5051 static int handle_machine_check(struct kvm_vcpu *vcpu)
5053 /* already handled by vcpu_run */
5057 static int handle_exception(struct kvm_vcpu *vcpu)
5059 struct vcpu_vmx *vmx = to_vmx(vcpu);
5060 struct kvm_run *kvm_run = vcpu->run;
5061 u32 intr_info, ex_no, error_code;
5062 unsigned long cr2, rip, dr6;
5064 enum emulation_result er;
5066 vect_info = vmx->idt_vectoring_info;
5067 intr_info = vmx->exit_intr_info;
5069 if (is_machine_check(intr_info))
5070 return handle_machine_check(vcpu);
5072 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
5073 return 1; /* already handled by vmx_vcpu_run() */
5075 if (is_no_device(intr_info)) {
5076 vmx_fpu_activate(vcpu);
5080 if (is_invalid_opcode(intr_info)) {
5081 if (is_guest_mode(vcpu)) {
5082 kvm_queue_exception(vcpu, UD_VECTOR);
5085 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
5086 if (er != EMULATE_DONE)
5087 kvm_queue_exception(vcpu, UD_VECTOR);
5092 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
5093 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
5096 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
5097 * MMIO, it is better to report an internal error.
5098 * See the comments in vmx_handle_exit.
5100 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
5101 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
5102 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5103 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
5104 vcpu->run->internal.ndata = 3;
5105 vcpu->run->internal.data[0] = vect_info;
5106 vcpu->run->internal.data[1] = intr_info;
5107 vcpu->run->internal.data[2] = error_code;
5111 if (is_page_fault(intr_info)) {
5112 /* EPT won't cause page fault directly */
5114 cr2 = vmcs_readl(EXIT_QUALIFICATION);
5115 trace_kvm_page_fault(cr2, error_code);
5117 if (kvm_event_needs_reinjection(vcpu))
5118 kvm_mmu_unprotect_page_virt(vcpu, cr2);
5119 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
5122 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
5124 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
5125 return handle_rmode_exception(vcpu, ex_no, error_code);
5129 dr6 = vmcs_readl(EXIT_QUALIFICATION);
5130 if (!(vcpu->guest_debug &
5131 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
5132 vcpu->arch.dr6 &= ~15;
5133 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5134 if (!(dr6 & ~DR6_RESERVED)) /* icebp */
5135 skip_emulated_instruction(vcpu);
5137 kvm_queue_exception(vcpu, DB_VECTOR);
5140 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
5141 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
5145 * Update instruction length as we may reinject #BP from
5146 * user space while in guest debugging mode. Reading it for
5147 * #DB as well causes no harm, it is not used in that case.
5149 vmx->vcpu.arch.event_exit_inst_len =
5150 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5151 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5152 rip = kvm_rip_read(vcpu);
5153 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
5154 kvm_run->debug.arch.exception = ex_no;
5157 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
5158 kvm_run->ex.exception = ex_no;
5159 kvm_run->ex.error_code = error_code;
5165 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
5167 ++vcpu->stat.irq_exits;
5171 static int handle_triple_fault(struct kvm_vcpu *vcpu)
5173 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5177 static int handle_io(struct kvm_vcpu *vcpu)
5179 unsigned long exit_qualification;
5180 int size, in, string;
5183 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5184 string = (exit_qualification & 16) != 0;
5185 in = (exit_qualification & 8) != 0;
5187 ++vcpu->stat.io_exits;
5190 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5192 port = exit_qualification >> 16;
5193 size = (exit_qualification & 7) + 1;
5194 skip_emulated_instruction(vcpu);
5196 return kvm_fast_pio_out(vcpu, size, port);
5200 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
5203 * Patch in the VMCALL instruction:
5205 hypercall[0] = 0x0f;
5206 hypercall[1] = 0x01;
5207 hypercall[2] = 0xc1;
5210 static bool nested_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
5212 unsigned long always_on = VMXON_CR0_ALWAYSON;
5213 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5215 if (to_vmx(vcpu)->nested.nested_vmx_secondary_ctls_high &
5216 SECONDARY_EXEC_UNRESTRICTED_GUEST &&
5217 nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
5218 always_on &= ~(X86_CR0_PE | X86_CR0_PG);
5219 return (val & always_on) == always_on;
5222 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
5223 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
5225 if (is_guest_mode(vcpu)) {
5226 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5227 unsigned long orig_val = val;
5230 * We get here when L2 changed cr0 in a way that did not change
5231 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
5232 * but did change L0 shadowed bits. So we first calculate the
5233 * effective cr0 value that L1 would like to write into the
5234 * hardware. It consists of the L2-owned bits from the new
5235 * value combined with the L1-owned bits from L1's guest_cr0.
5237 val = (val & ~vmcs12->cr0_guest_host_mask) |
5238 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
5240 if (!nested_cr0_valid(vcpu, val))
5243 if (kvm_set_cr0(vcpu, val))
5245 vmcs_writel(CR0_READ_SHADOW, orig_val);
5248 if (to_vmx(vcpu)->nested.vmxon &&
5249 ((val & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON))
5251 return kvm_set_cr0(vcpu, val);
5255 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
5257 if (is_guest_mode(vcpu)) {
5258 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5259 unsigned long orig_val = val;
5261 /* analogously to handle_set_cr0 */
5262 val = (val & ~vmcs12->cr4_guest_host_mask) |
5263 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
5264 if (kvm_set_cr4(vcpu, val))
5266 vmcs_writel(CR4_READ_SHADOW, orig_val);
5269 return kvm_set_cr4(vcpu, val);
5272 /* called to set cr0 as approriate for clts instruction exit. */
5273 static void handle_clts(struct kvm_vcpu *vcpu)
5275 if (is_guest_mode(vcpu)) {
5277 * We get here when L2 did CLTS, and L1 didn't shadow CR0.TS
5278 * but we did (!fpu_active). We need to keep GUEST_CR0.TS on,
5279 * just pretend it's off (also in arch.cr0 for fpu_activate).
5281 vmcs_writel(CR0_READ_SHADOW,
5282 vmcs_readl(CR0_READ_SHADOW) & ~X86_CR0_TS);
5283 vcpu->arch.cr0 &= ~X86_CR0_TS;
5285 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
5288 static int handle_cr(struct kvm_vcpu *vcpu)
5290 unsigned long exit_qualification, val;
5295 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5296 cr = exit_qualification & 15;
5297 reg = (exit_qualification >> 8) & 15;
5298 switch ((exit_qualification >> 4) & 3) {
5299 case 0: /* mov to cr */
5300 val = kvm_register_readl(vcpu, reg);
5301 trace_kvm_cr_write(cr, val);
5304 err = handle_set_cr0(vcpu, val);
5305 kvm_complete_insn_gp(vcpu, err);
5308 err = kvm_set_cr3(vcpu, val);
5309 kvm_complete_insn_gp(vcpu, err);
5312 err = handle_set_cr4(vcpu, val);
5313 kvm_complete_insn_gp(vcpu, err);
5316 u8 cr8_prev = kvm_get_cr8(vcpu);
5318 err = kvm_set_cr8(vcpu, cr8);
5319 kvm_complete_insn_gp(vcpu, err);
5320 if (irqchip_in_kernel(vcpu->kvm))
5322 if (cr8_prev <= cr8)
5324 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
5331 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
5332 skip_emulated_instruction(vcpu);
5333 vmx_fpu_activate(vcpu);
5335 case 1: /*mov from cr*/
5338 val = kvm_read_cr3(vcpu);
5339 kvm_register_write(vcpu, reg, val);
5340 trace_kvm_cr_read(cr, val);
5341 skip_emulated_instruction(vcpu);
5344 val = kvm_get_cr8(vcpu);
5345 kvm_register_write(vcpu, reg, val);
5346 trace_kvm_cr_read(cr, val);
5347 skip_emulated_instruction(vcpu);
5352 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5353 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
5354 kvm_lmsw(vcpu, val);
5356 skip_emulated_instruction(vcpu);
5361 vcpu->run->exit_reason = 0;
5362 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
5363 (int)(exit_qualification >> 4) & 3, cr);
5367 static int handle_dr(struct kvm_vcpu *vcpu)
5369 unsigned long exit_qualification;
5372 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5373 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
5375 /* First, if DR does not exist, trigger UD */
5376 if (!kvm_require_dr(vcpu, dr))
5379 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
5380 if (!kvm_require_cpl(vcpu, 0))
5382 dr7 = vmcs_readl(GUEST_DR7);
5385 * As the vm-exit takes precedence over the debug trap, we
5386 * need to emulate the latter, either for the host or the
5387 * guest debugging itself.
5389 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5390 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
5391 vcpu->run->debug.arch.dr7 = dr7;
5392 vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
5393 vcpu->run->debug.arch.exception = DB_VECTOR;
5394 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
5397 vcpu->arch.dr6 &= ~15;
5398 vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
5399 kvm_queue_exception(vcpu, DB_VECTOR);
5404 if (vcpu->guest_debug == 0) {
5405 u32 cpu_based_vm_exec_control;
5407 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5408 cpu_based_vm_exec_control &= ~CPU_BASED_MOV_DR_EXITING;
5409 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5412 * No more DR vmexits; force a reload of the debug registers
5413 * and reenter on this instruction. The next vmexit will
5414 * retrieve the full state of the debug registers.
5416 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
5420 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
5421 if (exit_qualification & TYPE_MOV_FROM_DR) {
5424 if (kvm_get_dr(vcpu, dr, &val))
5426 kvm_register_write(vcpu, reg, val);
5428 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
5431 skip_emulated_instruction(vcpu);
5435 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
5437 return vcpu->arch.dr6;
5440 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
5444 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
5446 u32 cpu_based_vm_exec_control;
5448 get_debugreg(vcpu->arch.db[0], 0);
5449 get_debugreg(vcpu->arch.db[1], 1);
5450 get_debugreg(vcpu->arch.db[2], 2);
5451 get_debugreg(vcpu->arch.db[3], 3);
5452 get_debugreg(vcpu->arch.dr6, 6);
5453 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
5455 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
5457 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5458 cpu_based_vm_exec_control |= CPU_BASED_MOV_DR_EXITING;
5459 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5462 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
5464 vmcs_writel(GUEST_DR7, val);
5467 static int handle_cpuid(struct kvm_vcpu *vcpu)
5469 kvm_emulate_cpuid(vcpu);
5473 static int handle_rdmsr(struct kvm_vcpu *vcpu)
5475 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5478 if (vmx_get_msr(vcpu, ecx, &data)) {
5479 trace_kvm_msr_read_ex(ecx);
5480 kvm_inject_gp(vcpu, 0);
5484 trace_kvm_msr_read(ecx, data);
5486 /* FIXME: handling of bits 32:63 of rax, rdx */
5487 vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
5488 vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
5489 skip_emulated_instruction(vcpu);
5493 static int handle_wrmsr(struct kvm_vcpu *vcpu)
5495 struct msr_data msr;
5496 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5497 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
5498 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
5502 msr.host_initiated = false;
5503 if (kvm_set_msr(vcpu, &msr) != 0) {
5504 trace_kvm_msr_write_ex(ecx, data);
5505 kvm_inject_gp(vcpu, 0);
5509 trace_kvm_msr_write(ecx, data);
5510 skip_emulated_instruction(vcpu);
5514 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
5516 kvm_make_request(KVM_REQ_EVENT, vcpu);
5520 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
5522 u32 cpu_based_vm_exec_control;
5524 /* clear pending irq */
5525 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5526 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
5527 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5529 kvm_make_request(KVM_REQ_EVENT, vcpu);
5531 ++vcpu->stat.irq_window_exits;
5534 * If the user space waits to inject interrupts, exit as soon as
5537 if (!irqchip_in_kernel(vcpu->kvm) &&
5538 vcpu->run->request_interrupt_window &&
5539 !kvm_cpu_has_interrupt(vcpu)) {
5540 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
5546 static int handle_halt(struct kvm_vcpu *vcpu)
5548 return kvm_emulate_halt(vcpu);
5551 static int handle_vmcall(struct kvm_vcpu *vcpu)
5553 kvm_emulate_hypercall(vcpu);
5557 static int handle_invd(struct kvm_vcpu *vcpu)
5559 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5562 static int handle_invlpg(struct kvm_vcpu *vcpu)
5564 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5566 kvm_mmu_invlpg(vcpu, exit_qualification);
5567 skip_emulated_instruction(vcpu);
5571 static int handle_rdpmc(struct kvm_vcpu *vcpu)
5575 err = kvm_rdpmc(vcpu);
5576 kvm_complete_insn_gp(vcpu, err);
5581 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5583 kvm_emulate_wbinvd(vcpu);
5587 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5589 u64 new_bv = kvm_read_edx_eax(vcpu);
5590 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5592 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5593 skip_emulated_instruction(vcpu);
5597 static int handle_xsaves(struct kvm_vcpu *vcpu)
5599 skip_emulated_instruction(vcpu);
5600 WARN(1, "this should never happen\n");
5604 static int handle_xrstors(struct kvm_vcpu *vcpu)
5606 skip_emulated_instruction(vcpu);
5607 WARN(1, "this should never happen\n");
5611 static int handle_apic_access(struct kvm_vcpu *vcpu)
5613 if (likely(fasteoi)) {
5614 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5615 int access_type, offset;
5617 access_type = exit_qualification & APIC_ACCESS_TYPE;
5618 offset = exit_qualification & APIC_ACCESS_OFFSET;
5620 * Sane guest uses MOV to write EOI, with written value
5621 * not cared. So make a short-circuit here by avoiding
5622 * heavy instruction emulation.
5624 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5625 (offset == APIC_EOI)) {
5626 kvm_lapic_set_eoi(vcpu);
5627 skip_emulated_instruction(vcpu);
5631 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5634 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5636 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5637 int vector = exit_qualification & 0xff;
5639 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5640 kvm_apic_set_eoi_accelerated(vcpu, vector);
5644 static int handle_apic_write(struct kvm_vcpu *vcpu)
5646 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5647 u32 offset = exit_qualification & 0xfff;
5649 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
5650 kvm_apic_write_nodecode(vcpu, offset);
5654 static int handle_task_switch(struct kvm_vcpu *vcpu)
5656 struct vcpu_vmx *vmx = to_vmx(vcpu);
5657 unsigned long exit_qualification;
5658 bool has_error_code = false;
5661 int reason, type, idt_v, idt_index;
5663 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5664 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5665 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5667 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5669 reason = (u32)exit_qualification >> 30;
5670 if (reason == TASK_SWITCH_GATE && idt_v) {
5672 case INTR_TYPE_NMI_INTR:
5673 vcpu->arch.nmi_injected = false;
5674 vmx_set_nmi_mask(vcpu, true);
5676 case INTR_TYPE_EXT_INTR:
5677 case INTR_TYPE_SOFT_INTR:
5678 kvm_clear_interrupt_queue(vcpu);
5680 case INTR_TYPE_HARD_EXCEPTION:
5681 if (vmx->idt_vectoring_info &
5682 VECTORING_INFO_DELIVER_CODE_MASK) {
5683 has_error_code = true;
5685 vmcs_read32(IDT_VECTORING_ERROR_CODE);
5688 case INTR_TYPE_SOFT_EXCEPTION:
5689 kvm_clear_exception_queue(vcpu);
5695 tss_selector = exit_qualification;
5697 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5698 type != INTR_TYPE_EXT_INTR &&
5699 type != INTR_TYPE_NMI_INTR))
5700 skip_emulated_instruction(vcpu);
5702 if (kvm_task_switch(vcpu, tss_selector,
5703 type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
5704 has_error_code, error_code) == EMULATE_FAIL) {
5705 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5706 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5707 vcpu->run->internal.ndata = 0;
5711 /* clear all local breakpoint enable flags */
5712 vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~0x155);
5715 * TODO: What about debug traps on tss switch?
5716 * Are we supposed to inject them and update dr6?
5722 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5724 unsigned long exit_qualification;
5729 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5731 gla_validity = (exit_qualification >> 7) & 0x3;
5732 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
5733 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
5734 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
5735 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
5736 vmcs_readl(GUEST_LINEAR_ADDRESS));
5737 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
5738 (long unsigned int)exit_qualification);
5739 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5740 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
5745 * EPT violation happened while executing iret from NMI,
5746 * "blocked by NMI" bit has to be set before next VM entry.
5747 * There are errata that may cause this bit to not be set:
5750 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5751 cpu_has_virtual_nmis() &&
5752 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
5753 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5755 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5756 trace_kvm_page_fault(gpa, exit_qualification);
5758 /* It is a write fault? */
5759 error_code = exit_qualification & PFERR_WRITE_MASK;
5760 /* It is a fetch fault? */
5761 error_code |= (exit_qualification << 2) & PFERR_FETCH_MASK;
5762 /* ept page table is present? */
5763 error_code |= (exit_qualification >> 3) & PFERR_PRESENT_MASK;
5765 vcpu->arch.exit_qualification = exit_qualification;
5767 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5770 static u64 ept_rsvd_mask(u64 spte, int level)
5775 for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
5776 mask |= (1ULL << i);
5779 /* bits 7:3 reserved */
5781 else if (spte & (1ULL << 7))
5783 * 1GB/2MB page, bits 29:12 or 20:12 reserved respectively,
5784 * level == 1 if the hypervisor is using the ignored bit 7.
5786 mask |= (PAGE_SIZE << ((level - 1) * 9)) - PAGE_SIZE;
5788 /* bits 6:3 reserved */
5794 static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
5797 printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
5799 /* 010b (write-only) */
5800 WARN_ON((spte & 0x7) == 0x2);
5802 /* 110b (write/execute) */
5803 WARN_ON((spte & 0x7) == 0x6);
5805 /* 100b (execute-only) and value not supported by logical processor */
5806 if (!cpu_has_vmx_ept_execute_only())
5807 WARN_ON((spte & 0x7) == 0x4);
5811 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
5813 if (rsvd_bits != 0) {
5814 printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
5815 __func__, rsvd_bits);
5819 /* bits 5:3 are _not_ reserved for large page or leaf page */
5820 if ((rsvd_bits & 0x38) == 0) {
5821 u64 ept_mem_type = (spte & 0x38) >> 3;
5823 if (ept_mem_type == 2 || ept_mem_type == 3 ||
5824 ept_mem_type == 7) {
5825 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
5826 __func__, ept_mem_type);
5833 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5836 int nr_sptes, i, ret;
5839 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5840 if (!kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5841 skip_emulated_instruction(vcpu);
5845 ret = handle_mmio_page_fault_common(vcpu, gpa, true);
5846 if (likely(ret == RET_MMIO_PF_EMULATE))
5847 return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) ==
5850 if (unlikely(ret == RET_MMIO_PF_INVALID))
5851 return kvm_mmu_page_fault(vcpu, gpa, 0, NULL, 0);
5853 if (unlikely(ret == RET_MMIO_PF_RETRY))
5856 /* It is the real ept misconfig */
5857 printk(KERN_ERR "EPT: Misconfiguration.\n");
5858 printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
5860 nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
5862 for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
5863 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
5865 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5866 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
5871 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5873 u32 cpu_based_vm_exec_control;
5875 /* clear pending NMI */
5876 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5877 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
5878 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5879 ++vcpu->stat.nmi_window_exits;
5880 kvm_make_request(KVM_REQ_EVENT, vcpu);
5885 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5887 struct vcpu_vmx *vmx = to_vmx(vcpu);
5888 enum emulation_result err = EMULATE_DONE;
5891 bool intr_window_requested;
5892 unsigned count = 130;
5894 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5895 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
5897 while (vmx->emulation_required && count-- != 0) {
5898 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5899 return handle_interrupt_window(&vmx->vcpu);
5901 if (test_bit(KVM_REQ_EVENT, &vcpu->requests))
5904 err = emulate_instruction(vcpu, EMULTYPE_NO_REEXECUTE);
5906 if (err == EMULATE_USER_EXIT) {
5907 ++vcpu->stat.mmio_exits;
5912 if (err != EMULATE_DONE) {
5913 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5914 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5915 vcpu->run->internal.ndata = 0;
5919 if (vcpu->arch.halt_request) {
5920 vcpu->arch.halt_request = 0;
5921 ret = kvm_vcpu_halt(vcpu);
5925 if (signal_pending(current))
5935 static int __grow_ple_window(int val)
5937 if (ple_window_grow < 1)
5940 val = min(val, ple_window_actual_max);
5942 if (ple_window_grow < ple_window)
5943 val *= ple_window_grow;
5945 val += ple_window_grow;
5950 static int __shrink_ple_window(int val, int modifier, int minimum)
5955 if (modifier < ple_window)
5960 return max(val, minimum);
5963 static void grow_ple_window(struct kvm_vcpu *vcpu)
5965 struct vcpu_vmx *vmx = to_vmx(vcpu);
5966 int old = vmx->ple_window;
5968 vmx->ple_window = __grow_ple_window(old);
5970 if (vmx->ple_window != old)
5971 vmx->ple_window_dirty = true;
5973 trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old);
5976 static void shrink_ple_window(struct kvm_vcpu *vcpu)
5978 struct vcpu_vmx *vmx = to_vmx(vcpu);
5979 int old = vmx->ple_window;
5981 vmx->ple_window = __shrink_ple_window(old,
5982 ple_window_shrink, ple_window);
5984 if (vmx->ple_window != old)
5985 vmx->ple_window_dirty = true;
5987 trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old);
5991 * ple_window_actual_max is computed to be one grow_ple_window() below
5992 * ple_window_max. (See __grow_ple_window for the reason.)
5993 * This prevents overflows, because ple_window_max is int.
5994 * ple_window_max effectively rounded down to a multiple of ple_window_grow in
5996 * ple_window_max is also prevented from setting vmx->ple_window < ple_window.
5998 static void update_ple_window_actual_max(void)
6000 ple_window_actual_max =
6001 __shrink_ple_window(max(ple_window_max, ple_window),
6002 ple_window_grow, INT_MIN);
6005 static __init int hardware_setup(void)
6007 int r = -ENOMEM, i, msr;
6009 rdmsrl_safe(MSR_EFER, &host_efer);
6011 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
6012 kvm_define_shared_msr(i, vmx_msr_index[i]);
6014 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
6015 if (!vmx_io_bitmap_a)
6018 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
6019 if (!vmx_io_bitmap_b)
6022 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
6023 if (!vmx_msr_bitmap_legacy)
6026 vmx_msr_bitmap_legacy_x2apic =
6027 (unsigned long *)__get_free_page(GFP_KERNEL);
6028 if (!vmx_msr_bitmap_legacy_x2apic)
6031 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
6032 if (!vmx_msr_bitmap_longmode)
6035 vmx_msr_bitmap_longmode_x2apic =
6036 (unsigned long *)__get_free_page(GFP_KERNEL);
6037 if (!vmx_msr_bitmap_longmode_x2apic)
6041 vmx_msr_bitmap_nested =
6042 (unsigned long *)__get_free_page(GFP_KERNEL);
6043 if (!vmx_msr_bitmap_nested)
6047 vmx_vmread_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
6048 if (!vmx_vmread_bitmap)
6051 vmx_vmwrite_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
6052 if (!vmx_vmwrite_bitmap)
6055 memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
6056 memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
6059 * Allow direct access to the PC debug port (it is often used for I/O
6060 * delays, but the vmexits simply slow things down).
6062 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
6063 clear_bit(0x80, vmx_io_bitmap_a);
6065 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
6067 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
6068 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
6070 memset(vmx_msr_bitmap_nested, 0xff, PAGE_SIZE);
6072 if (setup_vmcs_config(&vmcs_config) < 0) {
6077 if (boot_cpu_has(X86_FEATURE_NX))
6078 kvm_enable_efer_bits(EFER_NX);
6080 if (!cpu_has_vmx_vpid())
6082 if (!cpu_has_vmx_shadow_vmcs())
6083 enable_shadow_vmcs = 0;
6084 if (enable_shadow_vmcs)
6085 init_vmcs_shadow_fields();
6087 if (!cpu_has_vmx_ept() ||
6088 !cpu_has_vmx_ept_4levels()) {
6090 enable_unrestricted_guest = 0;
6091 enable_ept_ad_bits = 0;
6094 if (!cpu_has_vmx_ept_ad_bits())
6095 enable_ept_ad_bits = 0;
6097 if (!cpu_has_vmx_unrestricted_guest())
6098 enable_unrestricted_guest = 0;
6100 if (!cpu_has_vmx_flexpriority())
6101 flexpriority_enabled = 0;
6104 * set_apic_access_page_addr() is used to reload apic access
6105 * page upon invalidation. No need to do anything if not
6106 * using the APIC_ACCESS_ADDR VMCS field.
6108 if (!flexpriority_enabled)
6109 kvm_x86_ops->set_apic_access_page_addr = NULL;
6111 if (!cpu_has_vmx_tpr_shadow())
6112 kvm_x86_ops->update_cr8_intercept = NULL;
6114 if (enable_ept && !cpu_has_vmx_ept_2m_page())
6115 kvm_disable_largepages();
6117 if (!cpu_has_vmx_ple())
6120 if (!cpu_has_vmx_apicv())
6124 kvm_x86_ops->update_cr8_intercept = NULL;
6126 kvm_x86_ops->hwapic_irr_update = NULL;
6127 kvm_x86_ops->hwapic_isr_update = NULL;
6128 kvm_x86_ops->deliver_posted_interrupt = NULL;
6129 kvm_x86_ops->sync_pir_to_irr = vmx_sync_pir_to_irr_dummy;
6132 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
6133 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
6134 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
6135 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
6136 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
6137 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
6138 vmx_disable_intercept_for_msr(MSR_IA32_BNDCFGS, true);
6140 memcpy(vmx_msr_bitmap_legacy_x2apic,
6141 vmx_msr_bitmap_legacy, PAGE_SIZE);
6142 memcpy(vmx_msr_bitmap_longmode_x2apic,
6143 vmx_msr_bitmap_longmode, PAGE_SIZE);
6146 for (msr = 0x800; msr <= 0x8ff; msr++)
6147 vmx_disable_intercept_msr_read_x2apic(msr);
6149 /* According SDM, in x2apic mode, the whole id reg is used.
6150 * But in KVM, it only use the highest eight bits. Need to
6152 vmx_enable_intercept_msr_read_x2apic(0x802);
6154 vmx_enable_intercept_msr_read_x2apic(0x839);
6156 vmx_disable_intercept_msr_write_x2apic(0x808);
6158 vmx_disable_intercept_msr_write_x2apic(0x80b);
6160 vmx_disable_intercept_msr_write_x2apic(0x83f);
6164 kvm_mmu_set_mask_ptes(0ull,
6165 (enable_ept_ad_bits) ? VMX_EPT_ACCESS_BIT : 0ull,
6166 (enable_ept_ad_bits) ? VMX_EPT_DIRTY_BIT : 0ull,
6167 0ull, VMX_EPT_EXECUTABLE_MASK);
6168 ept_set_mmio_spte_mask();
6173 update_ple_window_actual_max();
6176 * Only enable PML when hardware supports PML feature, and both EPT
6177 * and EPT A/D bit features are enabled -- PML depends on them to work.
6179 if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
6183 kvm_x86_ops->slot_enable_log_dirty = NULL;
6184 kvm_x86_ops->slot_disable_log_dirty = NULL;
6185 kvm_x86_ops->flush_log_dirty = NULL;
6186 kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
6189 return alloc_kvm_area();
6192 free_page((unsigned long)vmx_vmwrite_bitmap);
6194 free_page((unsigned long)vmx_vmread_bitmap);
6197 free_page((unsigned long)vmx_msr_bitmap_nested);
6199 free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
6201 free_page((unsigned long)vmx_msr_bitmap_longmode);
6203 free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
6205 free_page((unsigned long)vmx_msr_bitmap_legacy);
6207 free_page((unsigned long)vmx_io_bitmap_b);
6209 free_page((unsigned long)vmx_io_bitmap_a);
6214 static __exit void hardware_unsetup(void)
6216 free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
6217 free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
6218 free_page((unsigned long)vmx_msr_bitmap_legacy);
6219 free_page((unsigned long)vmx_msr_bitmap_longmode);
6220 free_page((unsigned long)vmx_io_bitmap_b);
6221 free_page((unsigned long)vmx_io_bitmap_a);
6222 free_page((unsigned long)vmx_vmwrite_bitmap);
6223 free_page((unsigned long)vmx_vmread_bitmap);
6225 free_page((unsigned long)vmx_msr_bitmap_nested);
6231 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
6232 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
6234 static int handle_pause(struct kvm_vcpu *vcpu)
6237 grow_ple_window(vcpu);
6239 skip_emulated_instruction(vcpu);
6240 kvm_vcpu_on_spin(vcpu);
6245 static int handle_nop(struct kvm_vcpu *vcpu)
6247 skip_emulated_instruction(vcpu);
6251 static int handle_mwait(struct kvm_vcpu *vcpu)
6253 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
6254 return handle_nop(vcpu);
6257 static int handle_monitor(struct kvm_vcpu *vcpu)
6259 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
6260 return handle_nop(vcpu);
6264 * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
6265 * We could reuse a single VMCS for all the L2 guests, but we also want the
6266 * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
6267 * allows keeping them loaded on the processor, and in the future will allow
6268 * optimizations where prepare_vmcs02 doesn't need to set all the fields on
6269 * every entry if they never change.
6270 * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
6271 * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
6273 * The following functions allocate and free a vmcs02 in this pool.
6276 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
6277 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
6279 struct vmcs02_list *item;
6280 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6281 if (item->vmptr == vmx->nested.current_vmptr) {
6282 list_move(&item->list, &vmx->nested.vmcs02_pool);
6283 return &item->vmcs02;
6286 if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
6287 /* Recycle the least recently used VMCS. */
6288 item = list_entry(vmx->nested.vmcs02_pool.prev,
6289 struct vmcs02_list, list);
6290 item->vmptr = vmx->nested.current_vmptr;
6291 list_move(&item->list, &vmx->nested.vmcs02_pool);
6292 return &item->vmcs02;
6295 /* Create a new VMCS */
6296 item = kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
6299 item->vmcs02.vmcs = alloc_vmcs();
6300 if (!item->vmcs02.vmcs) {
6304 loaded_vmcs_init(&item->vmcs02);
6305 item->vmptr = vmx->nested.current_vmptr;
6306 list_add(&(item->list), &(vmx->nested.vmcs02_pool));
6307 vmx->nested.vmcs02_num++;
6308 return &item->vmcs02;
6311 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
6312 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
6314 struct vmcs02_list *item;
6315 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6316 if (item->vmptr == vmptr) {
6317 free_loaded_vmcs(&item->vmcs02);
6318 list_del(&item->list);
6320 vmx->nested.vmcs02_num--;
6326 * Free all VMCSs saved for this vcpu, except the one pointed by
6327 * vmx->loaded_vmcs. We must be running L1, so vmx->loaded_vmcs
6328 * must be &vmx->vmcs01.
6330 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
6332 struct vmcs02_list *item, *n;
6334 WARN_ON(vmx->loaded_vmcs != &vmx->vmcs01);
6335 list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
6337 * Something will leak if the above WARN triggers. Better than
6340 if (vmx->loaded_vmcs == &item->vmcs02)
6343 free_loaded_vmcs(&item->vmcs02);
6344 list_del(&item->list);
6346 vmx->nested.vmcs02_num--;
6351 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
6352 * set the success or error code of an emulated VMX instruction, as specified
6353 * by Vol 2B, VMX Instruction Reference, "Conventions".
6355 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
6357 vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
6358 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6359 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
6362 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
6364 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6365 & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
6366 X86_EFLAGS_SF | X86_EFLAGS_OF))
6370 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
6371 u32 vm_instruction_error)
6373 if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
6375 * failValid writes the error number to the current VMCS, which
6376 * can't be done there isn't a current VMCS.
6378 nested_vmx_failInvalid(vcpu);
6381 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6382 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6383 X86_EFLAGS_SF | X86_EFLAGS_OF))
6385 get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
6387 * We don't need to force a shadow sync because
6388 * VM_INSTRUCTION_ERROR is not shadowed
6392 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
6394 /* TODO: not to reset guest simply here. */
6395 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6396 pr_warn("kvm: nested vmx abort, indicator %d\n", indicator);
6399 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
6401 struct vcpu_vmx *vmx =
6402 container_of(timer, struct vcpu_vmx, nested.preemption_timer);
6404 vmx->nested.preemption_timer_expired = true;
6405 kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
6406 kvm_vcpu_kick(&vmx->vcpu);
6408 return HRTIMER_NORESTART;
6412 * Decode the memory-address operand of a vmx instruction, as recorded on an
6413 * exit caused by such an instruction (run by a guest hypervisor).
6414 * On success, returns 0. When the operand is invalid, returns 1 and throws
6417 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
6418 unsigned long exit_qualification,
6419 u32 vmx_instruction_info, gva_t *ret)
6422 * According to Vol. 3B, "Information for VM Exits Due to Instruction
6423 * Execution", on an exit, vmx_instruction_info holds most of the
6424 * addressing components of the operand. Only the displacement part
6425 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
6426 * For how an actual address is calculated from all these components,
6427 * refer to Vol. 1, "Operand Addressing".
6429 int scaling = vmx_instruction_info & 3;
6430 int addr_size = (vmx_instruction_info >> 7) & 7;
6431 bool is_reg = vmx_instruction_info & (1u << 10);
6432 int seg_reg = (vmx_instruction_info >> 15) & 7;
6433 int index_reg = (vmx_instruction_info >> 18) & 0xf;
6434 bool index_is_valid = !(vmx_instruction_info & (1u << 22));
6435 int base_reg = (vmx_instruction_info >> 23) & 0xf;
6436 bool base_is_valid = !(vmx_instruction_info & (1u << 27));
6439 kvm_queue_exception(vcpu, UD_VECTOR);
6443 /* Addr = segment_base + offset */
6444 /* offset = base + [index * scale] + displacement */
6445 *ret = vmx_get_segment_base(vcpu, seg_reg);
6447 *ret += kvm_register_read(vcpu, base_reg);
6449 *ret += kvm_register_read(vcpu, index_reg)<<scaling;
6450 *ret += exit_qualification; /* holds the displacement */
6452 if (addr_size == 1) /* 32 bit */
6456 * TODO: throw #GP (and return 1) in various cases that the VM*
6457 * instructions require it - e.g., offset beyond segment limit,
6458 * unusable or unreadable/unwritable segment, non-canonical 64-bit
6459 * address, and so on. Currently these are not checked.
6465 * This function performs the various checks including
6466 * - if it's 4KB aligned
6467 * - No bits beyond the physical address width are set
6468 * - Returns 0 on success or else 1
6469 * (Intel SDM Section 30.3)
6471 static int nested_vmx_check_vmptr(struct kvm_vcpu *vcpu, int exit_reason,
6476 struct x86_exception e;
6478 struct vcpu_vmx *vmx = to_vmx(vcpu);
6479 int maxphyaddr = cpuid_maxphyaddr(vcpu);
6481 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
6482 vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
6485 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
6486 sizeof(vmptr), &e)) {
6487 kvm_inject_page_fault(vcpu, &e);
6491 switch (exit_reason) {
6492 case EXIT_REASON_VMON:
6495 * The first 4 bytes of VMXON region contain the supported
6496 * VMCS revision identifier
6498 * Note - IA32_VMX_BASIC[48] will never be 1
6499 * for the nested case;
6500 * which replaces physical address width with 32
6503 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6504 nested_vmx_failInvalid(vcpu);
6505 skip_emulated_instruction(vcpu);
6509 page = nested_get_page(vcpu, vmptr);
6511 *(u32 *)kmap(page) != VMCS12_REVISION) {
6512 nested_vmx_failInvalid(vcpu);
6514 skip_emulated_instruction(vcpu);
6518 vmx->nested.vmxon_ptr = vmptr;
6520 case EXIT_REASON_VMCLEAR:
6521 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6522 nested_vmx_failValid(vcpu,
6523 VMXERR_VMCLEAR_INVALID_ADDRESS);
6524 skip_emulated_instruction(vcpu);
6528 if (vmptr == vmx->nested.vmxon_ptr) {
6529 nested_vmx_failValid(vcpu,
6530 VMXERR_VMCLEAR_VMXON_POINTER);
6531 skip_emulated_instruction(vcpu);
6535 case EXIT_REASON_VMPTRLD:
6536 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6537 nested_vmx_failValid(vcpu,
6538 VMXERR_VMPTRLD_INVALID_ADDRESS);
6539 skip_emulated_instruction(vcpu);
6543 if (vmptr == vmx->nested.vmxon_ptr) {
6544 nested_vmx_failValid(vcpu,
6545 VMXERR_VMCLEAR_VMXON_POINTER);
6546 skip_emulated_instruction(vcpu);
6551 return 1; /* shouldn't happen */
6560 * Emulate the VMXON instruction.
6561 * Currently, we just remember that VMX is active, and do not save or even
6562 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
6563 * do not currently need to store anything in that guest-allocated memory
6564 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
6565 * argument is different from the VMXON pointer (which the spec says they do).
6567 static int handle_vmon(struct kvm_vcpu *vcpu)
6569 struct kvm_segment cs;
6570 struct vcpu_vmx *vmx = to_vmx(vcpu);
6571 struct vmcs *shadow_vmcs;
6572 const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
6573 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
6575 /* The Intel VMX Instruction Reference lists a bunch of bits that
6576 * are prerequisite to running VMXON, most notably cr4.VMXE must be
6577 * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
6578 * Otherwise, we should fail with #UD. We test these now:
6580 if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
6581 !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
6582 (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
6583 kvm_queue_exception(vcpu, UD_VECTOR);
6587 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
6588 if (is_long_mode(vcpu) && !cs.l) {
6589 kvm_queue_exception(vcpu, UD_VECTOR);
6593 if (vmx_get_cpl(vcpu)) {
6594 kvm_inject_gp(vcpu, 0);
6598 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMON, NULL))
6601 if (vmx->nested.vmxon) {
6602 nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
6603 skip_emulated_instruction(vcpu);
6607 if ((vmx->nested.msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
6608 != VMXON_NEEDED_FEATURES) {
6609 kvm_inject_gp(vcpu, 0);
6613 if (enable_shadow_vmcs) {
6614 shadow_vmcs = alloc_vmcs();
6617 /* mark vmcs as shadow */
6618 shadow_vmcs->revision_id |= (1u << 31);
6619 /* init shadow vmcs */
6620 vmcs_clear(shadow_vmcs);
6621 vmx->nested.current_shadow_vmcs = shadow_vmcs;
6624 INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
6625 vmx->nested.vmcs02_num = 0;
6627 hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
6629 vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
6631 vmx->nested.vmxon = true;
6633 skip_emulated_instruction(vcpu);
6634 nested_vmx_succeed(vcpu);
6639 * Intel's VMX Instruction Reference specifies a common set of prerequisites
6640 * for running VMX instructions (except VMXON, whose prerequisites are
6641 * slightly different). It also specifies what exception to inject otherwise.
6643 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
6645 struct kvm_segment cs;
6646 struct vcpu_vmx *vmx = to_vmx(vcpu);
6648 if (!vmx->nested.vmxon) {
6649 kvm_queue_exception(vcpu, UD_VECTOR);
6653 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
6654 if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
6655 (is_long_mode(vcpu) && !cs.l)) {
6656 kvm_queue_exception(vcpu, UD_VECTOR);
6660 if (vmx_get_cpl(vcpu)) {
6661 kvm_inject_gp(vcpu, 0);
6668 static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
6671 if (vmx->nested.current_vmptr == -1ull)
6674 /* current_vmptr and current_vmcs12 are always set/reset together */
6675 if (WARN_ON(vmx->nested.current_vmcs12 == NULL))
6678 if (enable_shadow_vmcs) {
6679 /* copy to memory all shadowed fields in case
6680 they were modified */
6681 copy_shadow_to_vmcs12(vmx);
6682 vmx->nested.sync_shadow_vmcs = false;
6683 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6684 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
6685 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
6686 vmcs_write64(VMCS_LINK_POINTER, -1ull);
6688 vmx->nested.posted_intr_nv = -1;
6689 kunmap(vmx->nested.current_vmcs12_page);
6690 nested_release_page(vmx->nested.current_vmcs12_page);
6691 vmx->nested.current_vmptr = -1ull;
6692 vmx->nested.current_vmcs12 = NULL;
6696 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
6697 * just stops using VMX.
6699 static void free_nested(struct vcpu_vmx *vmx)
6701 if (!vmx->nested.vmxon)
6704 vmx->nested.vmxon = false;
6705 nested_release_vmcs12(vmx);
6706 if (enable_shadow_vmcs)
6707 free_vmcs(vmx->nested.current_shadow_vmcs);
6708 /* Unpin physical memory we referred to in current vmcs02 */
6709 if (vmx->nested.apic_access_page) {
6710 nested_release_page(vmx->nested.apic_access_page);
6711 vmx->nested.apic_access_page = NULL;
6713 if (vmx->nested.virtual_apic_page) {
6714 nested_release_page(vmx->nested.virtual_apic_page);
6715 vmx->nested.virtual_apic_page = NULL;
6717 if (vmx->nested.pi_desc_page) {
6718 kunmap(vmx->nested.pi_desc_page);
6719 nested_release_page(vmx->nested.pi_desc_page);
6720 vmx->nested.pi_desc_page = NULL;
6721 vmx->nested.pi_desc = NULL;
6724 nested_free_all_saved_vmcss(vmx);
6727 /* Emulate the VMXOFF instruction */
6728 static int handle_vmoff(struct kvm_vcpu *vcpu)
6730 if (!nested_vmx_check_permission(vcpu))
6732 free_nested(to_vmx(vcpu));
6733 skip_emulated_instruction(vcpu);
6734 nested_vmx_succeed(vcpu);
6738 /* Emulate the VMCLEAR instruction */
6739 static int handle_vmclear(struct kvm_vcpu *vcpu)
6741 struct vcpu_vmx *vmx = to_vmx(vcpu);
6743 struct vmcs12 *vmcs12;
6746 if (!nested_vmx_check_permission(vcpu))
6749 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMCLEAR, &vmptr))
6752 if (vmptr == vmx->nested.current_vmptr)
6753 nested_release_vmcs12(vmx);
6755 page = nested_get_page(vcpu, vmptr);
6758 * For accurate processor emulation, VMCLEAR beyond available
6759 * physical memory should do nothing at all. However, it is
6760 * possible that a nested vmx bug, not a guest hypervisor bug,
6761 * resulted in this case, so let's shut down before doing any
6764 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6767 vmcs12 = kmap(page);
6768 vmcs12->launch_state = 0;
6770 nested_release_page(page);
6772 nested_free_vmcs02(vmx, vmptr);
6774 skip_emulated_instruction(vcpu);
6775 nested_vmx_succeed(vcpu);
6779 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
6781 /* Emulate the VMLAUNCH instruction */
6782 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
6784 return nested_vmx_run(vcpu, true);
6787 /* Emulate the VMRESUME instruction */
6788 static int handle_vmresume(struct kvm_vcpu *vcpu)
6791 return nested_vmx_run(vcpu, false);
6794 enum vmcs_field_type {
6795 VMCS_FIELD_TYPE_U16 = 0,
6796 VMCS_FIELD_TYPE_U64 = 1,
6797 VMCS_FIELD_TYPE_U32 = 2,
6798 VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
6801 static inline int vmcs_field_type(unsigned long field)
6803 if (0x1 & field) /* the *_HIGH fields are all 32 bit */
6804 return VMCS_FIELD_TYPE_U32;
6805 return (field >> 13) & 0x3 ;
6808 static inline int vmcs_field_readonly(unsigned long field)
6810 return (((field >> 10) & 0x3) == 1);
6814 * Read a vmcs12 field. Since these can have varying lengths and we return
6815 * one type, we chose the biggest type (u64) and zero-extend the return value
6816 * to that size. Note that the caller, handle_vmread, might need to use only
6817 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
6818 * 64-bit fields are to be returned).
6820 static inline int vmcs12_read_any(struct kvm_vcpu *vcpu,
6821 unsigned long field, u64 *ret)
6823 short offset = vmcs_field_to_offset(field);
6829 p = ((char *)(get_vmcs12(vcpu))) + offset;
6831 switch (vmcs_field_type(field)) {
6832 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6833 *ret = *((natural_width *)p);
6835 case VMCS_FIELD_TYPE_U16:
6838 case VMCS_FIELD_TYPE_U32:
6841 case VMCS_FIELD_TYPE_U64:
6851 static inline int vmcs12_write_any(struct kvm_vcpu *vcpu,
6852 unsigned long field, u64 field_value){
6853 short offset = vmcs_field_to_offset(field);
6854 char *p = ((char *) get_vmcs12(vcpu)) + offset;
6858 switch (vmcs_field_type(field)) {
6859 case VMCS_FIELD_TYPE_U16:
6860 *(u16 *)p = field_value;
6862 case VMCS_FIELD_TYPE_U32:
6863 *(u32 *)p = field_value;
6865 case VMCS_FIELD_TYPE_U64:
6866 *(u64 *)p = field_value;
6868 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6869 *(natural_width *)p = field_value;
6878 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
6881 unsigned long field;
6883 struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
6884 const unsigned long *fields = shadow_read_write_fields;
6885 const int num_fields = max_shadow_read_write_fields;
6889 vmcs_load(shadow_vmcs);
6891 for (i = 0; i < num_fields; i++) {
6893 switch (vmcs_field_type(field)) {
6894 case VMCS_FIELD_TYPE_U16:
6895 field_value = vmcs_read16(field);
6897 case VMCS_FIELD_TYPE_U32:
6898 field_value = vmcs_read32(field);
6900 case VMCS_FIELD_TYPE_U64:
6901 field_value = vmcs_read64(field);
6903 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6904 field_value = vmcs_readl(field);
6910 vmcs12_write_any(&vmx->vcpu, field, field_value);
6913 vmcs_clear(shadow_vmcs);
6914 vmcs_load(vmx->loaded_vmcs->vmcs);
6919 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
6921 const unsigned long *fields[] = {
6922 shadow_read_write_fields,
6923 shadow_read_only_fields
6925 const int max_fields[] = {
6926 max_shadow_read_write_fields,
6927 max_shadow_read_only_fields
6930 unsigned long field;
6931 u64 field_value = 0;
6932 struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
6934 vmcs_load(shadow_vmcs);
6936 for (q = 0; q < ARRAY_SIZE(fields); q++) {
6937 for (i = 0; i < max_fields[q]; i++) {
6938 field = fields[q][i];
6939 vmcs12_read_any(&vmx->vcpu, field, &field_value);
6941 switch (vmcs_field_type(field)) {
6942 case VMCS_FIELD_TYPE_U16:
6943 vmcs_write16(field, (u16)field_value);
6945 case VMCS_FIELD_TYPE_U32:
6946 vmcs_write32(field, (u32)field_value);
6948 case VMCS_FIELD_TYPE_U64:
6949 vmcs_write64(field, (u64)field_value);
6951 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6952 vmcs_writel(field, (long)field_value);
6961 vmcs_clear(shadow_vmcs);
6962 vmcs_load(vmx->loaded_vmcs->vmcs);
6966 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
6967 * used before) all generate the same failure when it is missing.
6969 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
6971 struct vcpu_vmx *vmx = to_vmx(vcpu);
6972 if (vmx->nested.current_vmptr == -1ull) {
6973 nested_vmx_failInvalid(vcpu);
6974 skip_emulated_instruction(vcpu);
6980 static int handle_vmread(struct kvm_vcpu *vcpu)
6982 unsigned long field;
6984 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6985 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6988 if (!nested_vmx_check_permission(vcpu) ||
6989 !nested_vmx_check_vmcs12(vcpu))
6992 /* Decode instruction info and find the field to read */
6993 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
6994 /* Read the field, zero-extended to a u64 field_value */
6995 if (vmcs12_read_any(vcpu, field, &field_value) < 0) {
6996 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
6997 skip_emulated_instruction(vcpu);
7001 * Now copy part of this value to register or memory, as requested.
7002 * Note that the number of bits actually copied is 32 or 64 depending
7003 * on the guest's mode (32 or 64 bit), not on the given field's length.
7005 if (vmx_instruction_info & (1u << 10)) {
7006 kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
7009 if (get_vmx_mem_address(vcpu, exit_qualification,
7010 vmx_instruction_info, &gva))
7012 /* _system ok, as nested_vmx_check_permission verified cpl=0 */
7013 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
7014 &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
7017 nested_vmx_succeed(vcpu);
7018 skip_emulated_instruction(vcpu);
7023 static int handle_vmwrite(struct kvm_vcpu *vcpu)
7025 unsigned long field;
7027 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7028 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7029 /* The value to write might be 32 or 64 bits, depending on L1's long
7030 * mode, and eventually we need to write that into a field of several
7031 * possible lengths. The code below first zero-extends the value to 64
7032 * bit (field_value), and then copies only the approriate number of
7033 * bits into the vmcs12 field.
7035 u64 field_value = 0;
7036 struct x86_exception e;
7038 if (!nested_vmx_check_permission(vcpu) ||
7039 !nested_vmx_check_vmcs12(vcpu))
7042 if (vmx_instruction_info & (1u << 10))
7043 field_value = kvm_register_readl(vcpu,
7044 (((vmx_instruction_info) >> 3) & 0xf));
7046 if (get_vmx_mem_address(vcpu, exit_qualification,
7047 vmx_instruction_info, &gva))
7049 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
7050 &field_value, (is_64_bit_mode(vcpu) ? 8 : 4), &e)) {
7051 kvm_inject_page_fault(vcpu, &e);
7057 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7058 if (vmcs_field_readonly(field)) {
7059 nested_vmx_failValid(vcpu,
7060 VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
7061 skip_emulated_instruction(vcpu);
7065 if (vmcs12_write_any(vcpu, field, field_value) < 0) {
7066 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7067 skip_emulated_instruction(vcpu);
7071 nested_vmx_succeed(vcpu);
7072 skip_emulated_instruction(vcpu);
7076 /* Emulate the VMPTRLD instruction */
7077 static int handle_vmptrld(struct kvm_vcpu *vcpu)
7079 struct vcpu_vmx *vmx = to_vmx(vcpu);
7083 if (!nested_vmx_check_permission(vcpu))
7086 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMPTRLD, &vmptr))
7089 if (vmx->nested.current_vmptr != vmptr) {
7090 struct vmcs12 *new_vmcs12;
7092 page = nested_get_page(vcpu, vmptr);
7094 nested_vmx_failInvalid(vcpu);
7095 skip_emulated_instruction(vcpu);
7098 new_vmcs12 = kmap(page);
7099 if (new_vmcs12->revision_id != VMCS12_REVISION) {
7101 nested_release_page_clean(page);
7102 nested_vmx_failValid(vcpu,
7103 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
7104 skip_emulated_instruction(vcpu);
7108 nested_release_vmcs12(vmx);
7109 vmx->nested.current_vmptr = vmptr;
7110 vmx->nested.current_vmcs12 = new_vmcs12;
7111 vmx->nested.current_vmcs12_page = page;
7112 if (enable_shadow_vmcs) {
7113 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7114 exec_control |= SECONDARY_EXEC_SHADOW_VMCS;
7115 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
7116 vmcs_write64(VMCS_LINK_POINTER,
7117 __pa(vmx->nested.current_shadow_vmcs));
7118 vmx->nested.sync_shadow_vmcs = true;
7122 nested_vmx_succeed(vcpu);
7123 skip_emulated_instruction(vcpu);
7127 /* Emulate the VMPTRST instruction */
7128 static int handle_vmptrst(struct kvm_vcpu *vcpu)
7130 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7131 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7133 struct x86_exception e;
7135 if (!nested_vmx_check_permission(vcpu))
7138 if (get_vmx_mem_address(vcpu, exit_qualification,
7139 vmx_instruction_info, &vmcs_gva))
7141 /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
7142 if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
7143 (void *)&to_vmx(vcpu)->nested.current_vmptr,
7145 kvm_inject_page_fault(vcpu, &e);
7148 nested_vmx_succeed(vcpu);
7149 skip_emulated_instruction(vcpu);
7153 /* Emulate the INVEPT instruction */
7154 static int handle_invept(struct kvm_vcpu *vcpu)
7156 struct vcpu_vmx *vmx = to_vmx(vcpu);
7157 u32 vmx_instruction_info, types;
7160 struct x86_exception e;
7165 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7166 SECONDARY_EXEC_ENABLE_EPT) ||
7167 !(vmx->nested.nested_vmx_ept_caps & VMX_EPT_INVEPT_BIT)) {
7168 kvm_queue_exception(vcpu, UD_VECTOR);
7172 if (!nested_vmx_check_permission(vcpu))
7175 if (!kvm_read_cr0_bits(vcpu, X86_CR0_PE)) {
7176 kvm_queue_exception(vcpu, UD_VECTOR);
7180 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7181 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7183 types = (vmx->nested.nested_vmx_ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
7185 if (!(types & (1UL << type))) {
7186 nested_vmx_failValid(vcpu,
7187 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7191 /* According to the Intel VMX instruction reference, the memory
7192 * operand is read even if it isn't needed (e.g., for type==global)
7194 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7195 vmx_instruction_info, &gva))
7197 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
7198 sizeof(operand), &e)) {
7199 kvm_inject_page_fault(vcpu, &e);
7204 case VMX_EPT_EXTENT_GLOBAL:
7205 kvm_mmu_sync_roots(vcpu);
7206 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
7207 nested_vmx_succeed(vcpu);
7210 /* Trap single context invalidation invept calls */
7215 skip_emulated_instruction(vcpu);
7219 static int handle_invvpid(struct kvm_vcpu *vcpu)
7221 kvm_queue_exception(vcpu, UD_VECTOR);
7225 static int handle_pml_full(struct kvm_vcpu *vcpu)
7227 unsigned long exit_qualification;
7229 trace_kvm_pml_full(vcpu->vcpu_id);
7231 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7234 * PML buffer FULL happened while executing iret from NMI,
7235 * "blocked by NMI" bit has to be set before next VM entry.
7237 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
7238 cpu_has_virtual_nmis() &&
7239 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
7240 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
7241 GUEST_INTR_STATE_NMI);
7244 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
7245 * here.., and there's no userspace involvement needed for PML.
7251 * The exit handlers return 1 if the exit was handled fully and guest execution
7252 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
7253 * to be done to userspace and return 0.
7255 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
7256 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
7257 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
7258 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
7259 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
7260 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
7261 [EXIT_REASON_CR_ACCESS] = handle_cr,
7262 [EXIT_REASON_DR_ACCESS] = handle_dr,
7263 [EXIT_REASON_CPUID] = handle_cpuid,
7264 [EXIT_REASON_MSR_READ] = handle_rdmsr,
7265 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
7266 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
7267 [EXIT_REASON_HLT] = handle_halt,
7268 [EXIT_REASON_INVD] = handle_invd,
7269 [EXIT_REASON_INVLPG] = handle_invlpg,
7270 [EXIT_REASON_RDPMC] = handle_rdpmc,
7271 [EXIT_REASON_VMCALL] = handle_vmcall,
7272 [EXIT_REASON_VMCLEAR] = handle_vmclear,
7273 [EXIT_REASON_VMLAUNCH] = handle_vmlaunch,
7274 [EXIT_REASON_VMPTRLD] = handle_vmptrld,
7275 [EXIT_REASON_VMPTRST] = handle_vmptrst,
7276 [EXIT_REASON_VMREAD] = handle_vmread,
7277 [EXIT_REASON_VMRESUME] = handle_vmresume,
7278 [EXIT_REASON_VMWRITE] = handle_vmwrite,
7279 [EXIT_REASON_VMOFF] = handle_vmoff,
7280 [EXIT_REASON_VMON] = handle_vmon,
7281 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
7282 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
7283 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
7284 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
7285 [EXIT_REASON_WBINVD] = handle_wbinvd,
7286 [EXIT_REASON_XSETBV] = handle_xsetbv,
7287 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
7288 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
7289 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
7290 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
7291 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
7292 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait,
7293 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor,
7294 [EXIT_REASON_INVEPT] = handle_invept,
7295 [EXIT_REASON_INVVPID] = handle_invvpid,
7296 [EXIT_REASON_XSAVES] = handle_xsaves,
7297 [EXIT_REASON_XRSTORS] = handle_xrstors,
7298 [EXIT_REASON_PML_FULL] = handle_pml_full,
7301 static const int kvm_vmx_max_exit_handlers =
7302 ARRAY_SIZE(kvm_vmx_exit_handlers);
7304 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
7305 struct vmcs12 *vmcs12)
7307 unsigned long exit_qualification;
7308 gpa_t bitmap, last_bitmap;
7313 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7314 return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
7316 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7318 port = exit_qualification >> 16;
7319 size = (exit_qualification & 7) + 1;
7321 last_bitmap = (gpa_t)-1;
7326 bitmap = vmcs12->io_bitmap_a;
7327 else if (port < 0x10000)
7328 bitmap = vmcs12->io_bitmap_b;
7331 bitmap += (port & 0x7fff) / 8;
7333 if (last_bitmap != bitmap)
7334 if (kvm_read_guest(vcpu->kvm, bitmap, &b, 1))
7336 if (b & (1 << (port & 7)))
7341 last_bitmap = bitmap;
7348 * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
7349 * rather than handle it ourselves in L0. I.e., check whether L1 expressed
7350 * disinterest in the current event (read or write a specific MSR) by using an
7351 * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
7353 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
7354 struct vmcs12 *vmcs12, u32 exit_reason)
7356 u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
7359 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
7363 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
7364 * for the four combinations of read/write and low/high MSR numbers.
7365 * First we need to figure out which of the four to use:
7367 bitmap = vmcs12->msr_bitmap;
7368 if (exit_reason == EXIT_REASON_MSR_WRITE)
7370 if (msr_index >= 0xc0000000) {
7371 msr_index -= 0xc0000000;
7375 /* Then read the msr_index'th bit from this bitmap: */
7376 if (msr_index < 1024*8) {
7378 if (kvm_read_guest(vcpu->kvm, bitmap + msr_index/8, &b, 1))
7380 return 1 & (b >> (msr_index & 7));
7382 return true; /* let L1 handle the wrong parameter */
7386 * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
7387 * rather than handle it ourselves in L0. I.e., check if L1 wanted to
7388 * intercept (via guest_host_mask etc.) the current event.
7390 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
7391 struct vmcs12 *vmcs12)
7393 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7394 int cr = exit_qualification & 15;
7395 int reg = (exit_qualification >> 8) & 15;
7396 unsigned long val = kvm_register_readl(vcpu, reg);
7398 switch ((exit_qualification >> 4) & 3) {
7399 case 0: /* mov to cr */
7402 if (vmcs12->cr0_guest_host_mask &
7403 (val ^ vmcs12->cr0_read_shadow))
7407 if ((vmcs12->cr3_target_count >= 1 &&
7408 vmcs12->cr3_target_value0 == val) ||
7409 (vmcs12->cr3_target_count >= 2 &&
7410 vmcs12->cr3_target_value1 == val) ||
7411 (vmcs12->cr3_target_count >= 3 &&
7412 vmcs12->cr3_target_value2 == val) ||
7413 (vmcs12->cr3_target_count >= 4 &&
7414 vmcs12->cr3_target_value3 == val))
7416 if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
7420 if (vmcs12->cr4_guest_host_mask &
7421 (vmcs12->cr4_read_shadow ^ val))
7425 if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
7431 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
7432 (vmcs12->cr0_read_shadow & X86_CR0_TS))
7435 case 1: /* mov from cr */
7438 if (vmcs12->cpu_based_vm_exec_control &
7439 CPU_BASED_CR3_STORE_EXITING)
7443 if (vmcs12->cpu_based_vm_exec_control &
7444 CPU_BASED_CR8_STORE_EXITING)
7451 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
7452 * cr0. Other attempted changes are ignored, with no exit.
7454 if (vmcs12->cr0_guest_host_mask & 0xe &
7455 (val ^ vmcs12->cr0_read_shadow))
7457 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
7458 !(vmcs12->cr0_read_shadow & 0x1) &&
7467 * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
7468 * should handle it ourselves in L0 (and then continue L2). Only call this
7469 * when in is_guest_mode (L2).
7471 static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu)
7473 u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
7474 struct vcpu_vmx *vmx = to_vmx(vcpu);
7475 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7476 u32 exit_reason = vmx->exit_reason;
7478 trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
7479 vmcs_readl(EXIT_QUALIFICATION),
7480 vmx->idt_vectoring_info,
7482 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
7485 if (vmx->nested.nested_run_pending)
7488 if (unlikely(vmx->fail)) {
7489 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
7490 vmcs_read32(VM_INSTRUCTION_ERROR));
7494 switch (exit_reason) {
7495 case EXIT_REASON_EXCEPTION_NMI:
7496 if (!is_exception(intr_info))
7498 else if (is_page_fault(intr_info))
7500 else if (is_no_device(intr_info) &&
7501 !(vmcs12->guest_cr0 & X86_CR0_TS))
7503 return vmcs12->exception_bitmap &
7504 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
7505 case EXIT_REASON_EXTERNAL_INTERRUPT:
7507 case EXIT_REASON_TRIPLE_FAULT:
7509 case EXIT_REASON_PENDING_INTERRUPT:
7510 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
7511 case EXIT_REASON_NMI_WINDOW:
7512 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
7513 case EXIT_REASON_TASK_SWITCH:
7515 case EXIT_REASON_CPUID:
7516 if (kvm_register_read(vcpu, VCPU_REGS_RAX) == 0xa)
7519 case EXIT_REASON_HLT:
7520 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
7521 case EXIT_REASON_INVD:
7523 case EXIT_REASON_INVLPG:
7524 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
7525 case EXIT_REASON_RDPMC:
7526 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
7527 case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
7528 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
7529 case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
7530 case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
7531 case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
7532 case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
7533 case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
7534 case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
7536 * VMX instructions trap unconditionally. This allows L1 to
7537 * emulate them for its L2 guest, i.e., allows 3-level nesting!
7540 case EXIT_REASON_CR_ACCESS:
7541 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
7542 case EXIT_REASON_DR_ACCESS:
7543 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
7544 case EXIT_REASON_IO_INSTRUCTION:
7545 return nested_vmx_exit_handled_io(vcpu, vmcs12);
7546 case EXIT_REASON_MSR_READ:
7547 case EXIT_REASON_MSR_WRITE:
7548 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
7549 case EXIT_REASON_INVALID_STATE:
7551 case EXIT_REASON_MWAIT_INSTRUCTION:
7552 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
7553 case EXIT_REASON_MONITOR_INSTRUCTION:
7554 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
7555 case EXIT_REASON_PAUSE_INSTRUCTION:
7556 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
7557 nested_cpu_has2(vmcs12,
7558 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
7559 case EXIT_REASON_MCE_DURING_VMENTRY:
7561 case EXIT_REASON_TPR_BELOW_THRESHOLD:
7562 return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
7563 case EXIT_REASON_APIC_ACCESS:
7564 return nested_cpu_has2(vmcs12,
7565 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
7566 case EXIT_REASON_APIC_WRITE:
7567 case EXIT_REASON_EOI_INDUCED:
7568 /* apic_write and eoi_induced should exit unconditionally. */
7570 case EXIT_REASON_EPT_VIOLATION:
7572 * L0 always deals with the EPT violation. If nested EPT is
7573 * used, and the nested mmu code discovers that the address is
7574 * missing in the guest EPT table (EPT12), the EPT violation
7575 * will be injected with nested_ept_inject_page_fault()
7578 case EXIT_REASON_EPT_MISCONFIG:
7580 * L2 never uses directly L1's EPT, but rather L0's own EPT
7581 * table (shadow on EPT) or a merged EPT table that L0 built
7582 * (EPT on EPT). So any problems with the structure of the
7583 * table is L0's fault.
7586 case EXIT_REASON_WBINVD:
7587 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
7588 case EXIT_REASON_XSETBV:
7590 case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
7592 * This should never happen, since it is not possible to
7593 * set XSS to a non-zero value---neither in L1 nor in L2.
7594 * If if it were, XSS would have to be checked against
7595 * the XSS exit bitmap in vmcs12.
7597 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
7603 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
7605 *info1 = vmcs_readl(EXIT_QUALIFICATION);
7606 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
7609 static int vmx_enable_pml(struct vcpu_vmx *vmx)
7611 struct page *pml_pg;
7614 pml_pg = alloc_page(GFP_KERNEL | __GFP_ZERO);
7618 vmx->pml_pg = pml_pg;
7620 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
7621 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
7623 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7624 exec_control |= SECONDARY_EXEC_ENABLE_PML;
7625 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
7630 static void vmx_disable_pml(struct vcpu_vmx *vmx)
7634 ASSERT(vmx->pml_pg);
7635 __free_page(vmx->pml_pg);
7638 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7639 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
7640 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
7643 static void vmx_flush_pml_buffer(struct vcpu_vmx *vmx)
7645 struct kvm *kvm = vmx->vcpu.kvm;
7649 pml_idx = vmcs_read16(GUEST_PML_INDEX);
7651 /* Do nothing if PML buffer is empty */
7652 if (pml_idx == (PML_ENTITY_NUM - 1))
7655 /* PML index always points to next available PML buffer entity */
7656 if (pml_idx >= PML_ENTITY_NUM)
7661 pml_buf = page_address(vmx->pml_pg);
7662 for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
7665 gpa = pml_buf[pml_idx];
7666 WARN_ON(gpa & (PAGE_SIZE - 1));
7667 mark_page_dirty(kvm, gpa >> PAGE_SHIFT);
7670 /* reset PML index */
7671 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
7675 * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
7676 * Called before reporting dirty_bitmap to userspace.
7678 static void kvm_flush_pml_buffers(struct kvm *kvm)
7681 struct kvm_vcpu *vcpu;
7683 * We only need to kick vcpu out of guest mode here, as PML buffer
7684 * is flushed at beginning of all VMEXITs, and it's obvious that only
7685 * vcpus running in guest are possible to have unflushed GPAs in PML
7688 kvm_for_each_vcpu(i, vcpu, kvm)
7689 kvm_vcpu_kick(vcpu);
7692 static void vmx_dump_sel(char *name, uint32_t sel)
7694 pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
7695 name, vmcs_read32(sel),
7696 vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
7697 vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
7698 vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
7701 static void vmx_dump_dtsel(char *name, uint32_t limit)
7703 pr_err("%s limit=0x%08x, base=0x%016lx\n",
7704 name, vmcs_read32(limit),
7705 vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
7708 static void dump_vmcs(void)
7710 u32 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
7711 u32 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
7712 u32 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
7713 u32 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
7714 u32 secondary_exec_control = 0;
7715 unsigned long cr4 = vmcs_readl(GUEST_CR4);
7716 u64 efer = vmcs_readl(GUEST_IA32_EFER);
7719 if (cpu_has_secondary_exec_ctrls())
7720 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7722 pr_err("*** Guest State ***\n");
7723 pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
7724 vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
7725 vmcs_readl(CR0_GUEST_HOST_MASK));
7726 pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
7727 cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
7728 pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
7729 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
7730 (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
7732 pr_err("PDPTR0 = 0x%016lx PDPTR1 = 0x%016lx\n",
7733 vmcs_readl(GUEST_PDPTR0), vmcs_readl(GUEST_PDPTR1));
7734 pr_err("PDPTR2 = 0x%016lx PDPTR3 = 0x%016lx\n",
7735 vmcs_readl(GUEST_PDPTR2), vmcs_readl(GUEST_PDPTR3));
7737 pr_err("RSP = 0x%016lx RIP = 0x%016lx\n",
7738 vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
7739 pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n",
7740 vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
7741 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
7742 vmcs_readl(GUEST_SYSENTER_ESP),
7743 vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
7744 vmx_dump_sel("CS: ", GUEST_CS_SELECTOR);
7745 vmx_dump_sel("DS: ", GUEST_DS_SELECTOR);
7746 vmx_dump_sel("SS: ", GUEST_SS_SELECTOR);
7747 vmx_dump_sel("ES: ", GUEST_ES_SELECTOR);
7748 vmx_dump_sel("FS: ", GUEST_FS_SELECTOR);
7749 vmx_dump_sel("GS: ", GUEST_GS_SELECTOR);
7750 vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
7751 vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
7752 vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
7753 vmx_dump_sel("TR: ", GUEST_TR_SELECTOR);
7754 if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
7755 (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
7756 pr_err("EFER = 0x%016llx PAT = 0x%016lx\n",
7757 efer, vmcs_readl(GUEST_IA32_PAT));
7758 pr_err("DebugCtl = 0x%016lx DebugExceptions = 0x%016lx\n",
7759 vmcs_readl(GUEST_IA32_DEBUGCTL),
7760 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
7761 if (vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
7762 pr_err("PerfGlobCtl = 0x%016lx\n",
7763 vmcs_readl(GUEST_IA32_PERF_GLOBAL_CTRL));
7764 if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
7765 pr_err("BndCfgS = 0x%016lx\n", vmcs_readl(GUEST_BNDCFGS));
7766 pr_err("Interruptibility = %08x ActivityState = %08x\n",
7767 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
7768 vmcs_read32(GUEST_ACTIVITY_STATE));
7769 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
7770 pr_err("InterruptStatus = %04x\n",
7771 vmcs_read16(GUEST_INTR_STATUS));
7773 pr_err("*** Host State ***\n");
7774 pr_err("RIP = 0x%016lx RSP = 0x%016lx\n",
7775 vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
7776 pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
7777 vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
7778 vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
7779 vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
7780 vmcs_read16(HOST_TR_SELECTOR));
7781 pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
7782 vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
7783 vmcs_readl(HOST_TR_BASE));
7784 pr_err("GDTBase=%016lx IDTBase=%016lx\n",
7785 vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
7786 pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
7787 vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
7788 vmcs_readl(HOST_CR4));
7789 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
7790 vmcs_readl(HOST_IA32_SYSENTER_ESP),
7791 vmcs_read32(HOST_IA32_SYSENTER_CS),
7792 vmcs_readl(HOST_IA32_SYSENTER_EIP));
7793 if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
7794 pr_err("EFER = 0x%016lx PAT = 0x%016lx\n",
7795 vmcs_readl(HOST_IA32_EFER), vmcs_readl(HOST_IA32_PAT));
7796 if (vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
7797 pr_err("PerfGlobCtl = 0x%016lx\n",
7798 vmcs_readl(HOST_IA32_PERF_GLOBAL_CTRL));
7800 pr_err("*** Control State ***\n");
7801 pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
7802 pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
7803 pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
7804 pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
7805 vmcs_read32(EXCEPTION_BITMAP),
7806 vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
7807 vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
7808 pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
7809 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
7810 vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
7811 vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
7812 pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
7813 vmcs_read32(VM_EXIT_INTR_INFO),
7814 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
7815 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
7816 pr_err(" reason=%08x qualification=%016lx\n",
7817 vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
7818 pr_err("IDTVectoring: info=%08x errcode=%08x\n",
7819 vmcs_read32(IDT_VECTORING_INFO_FIELD),
7820 vmcs_read32(IDT_VECTORING_ERROR_CODE));
7821 pr_err("TSC Offset = 0x%016lx\n", vmcs_readl(TSC_OFFSET));
7822 if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW)
7823 pr_err("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
7824 if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
7825 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
7826 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
7827 pr_err("EPT pointer = 0x%016lx\n", vmcs_readl(EPT_POINTER));
7828 n = vmcs_read32(CR3_TARGET_COUNT);
7829 for (i = 0; i + 1 < n; i += 4)
7830 pr_err("CR3 target%u=%016lx target%u=%016lx\n",
7831 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
7832 i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
7834 pr_err("CR3 target%u=%016lx\n",
7835 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
7836 if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
7837 pr_err("PLE Gap=%08x Window=%08x\n",
7838 vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
7839 if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
7840 pr_err("Virtual processor ID = 0x%04x\n",
7841 vmcs_read16(VIRTUAL_PROCESSOR_ID));
7845 * The guest has exited. See if we can fix it or if we need userspace
7848 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
7850 struct vcpu_vmx *vmx = to_vmx(vcpu);
7851 u32 exit_reason = vmx->exit_reason;
7852 u32 vectoring_info = vmx->idt_vectoring_info;
7855 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
7856 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
7857 * querying dirty_bitmap, we only need to kick all vcpus out of guest
7858 * mode as if vcpus is in root mode, the PML buffer must has been
7862 vmx_flush_pml_buffer(vmx);
7864 /* If guest state is invalid, start emulating */
7865 if (vmx->emulation_required)
7866 return handle_invalid_guest_state(vcpu);
7868 if (is_guest_mode(vcpu) && nested_vmx_exit_handled(vcpu)) {
7869 nested_vmx_vmexit(vcpu, exit_reason,
7870 vmcs_read32(VM_EXIT_INTR_INFO),
7871 vmcs_readl(EXIT_QUALIFICATION));
7875 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
7877 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
7878 vcpu->run->fail_entry.hardware_entry_failure_reason
7883 if (unlikely(vmx->fail)) {
7884 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
7885 vcpu->run->fail_entry.hardware_entry_failure_reason
7886 = vmcs_read32(VM_INSTRUCTION_ERROR);
7892 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
7893 * delivery event since it indicates guest is accessing MMIO.
7894 * The vm-exit can be triggered again after return to guest that
7895 * will cause infinite loop.
7897 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
7898 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
7899 exit_reason != EXIT_REASON_EPT_VIOLATION &&
7900 exit_reason != EXIT_REASON_TASK_SWITCH)) {
7901 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7902 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
7903 vcpu->run->internal.ndata = 2;
7904 vcpu->run->internal.data[0] = vectoring_info;
7905 vcpu->run->internal.data[1] = exit_reason;
7909 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked &&
7910 !(is_guest_mode(vcpu) && nested_cpu_has_virtual_nmis(
7911 get_vmcs12(vcpu))))) {
7912 if (vmx_interrupt_allowed(vcpu)) {
7913 vmx->soft_vnmi_blocked = 0;
7914 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
7915 vcpu->arch.nmi_pending) {
7917 * This CPU don't support us in finding the end of an
7918 * NMI-blocked window if the guest runs with IRQs
7919 * disabled. So we pull the trigger after 1 s of
7920 * futile waiting, but inform the user about this.
7922 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
7923 "state on VCPU %d after 1 s timeout\n",
7924 __func__, vcpu->vcpu_id);
7925 vmx->soft_vnmi_blocked = 0;
7929 if (exit_reason < kvm_vmx_max_exit_handlers
7930 && kvm_vmx_exit_handlers[exit_reason])
7931 return kvm_vmx_exit_handlers[exit_reason](vcpu);
7933 WARN_ONCE(1, "vmx: unexpected exit reason 0x%x\n", exit_reason);
7934 kvm_queue_exception(vcpu, UD_VECTOR);
7939 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
7941 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7943 if (is_guest_mode(vcpu) &&
7944 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
7947 if (irr == -1 || tpr < irr) {
7948 vmcs_write32(TPR_THRESHOLD, 0);
7952 vmcs_write32(TPR_THRESHOLD, irr);
7955 static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
7957 u32 sec_exec_control;
7960 * There is not point to enable virtualize x2apic without enable
7963 if (!cpu_has_vmx_virtualize_x2apic_mode() ||
7964 !vmx_vm_has_apicv(vcpu->kvm))
7967 if (!vm_need_tpr_shadow(vcpu->kvm))
7970 sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7973 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
7974 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
7976 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
7977 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
7979 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
7981 vmx_set_msr_bitmap(vcpu);
7984 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
7986 struct vcpu_vmx *vmx = to_vmx(vcpu);
7989 * Currently we do not handle the nested case where L2 has an
7990 * APIC access page of its own; that page is still pinned.
7991 * Hence, we skip the case where the VCPU is in guest mode _and_
7992 * L1 prepared an APIC access page for L2.
7994 * For the case where L1 and L2 share the same APIC access page
7995 * (flexpriority=Y but SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES clear
7996 * in the vmcs12), this function will only update either the vmcs01
7997 * or the vmcs02. If the former, the vmcs02 will be updated by
7998 * prepare_vmcs02. If the latter, the vmcs01 will be updated in
7999 * the next L2->L1 exit.
8001 if (!is_guest_mode(vcpu) ||
8002 !nested_cpu_has2(vmx->nested.current_vmcs12,
8003 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
8004 vmcs_write64(APIC_ACCESS_ADDR, hpa);
8007 static void vmx_hwapic_isr_update(struct kvm *kvm, int isr)
8015 status = vmcs_read16(GUEST_INTR_STATUS);
8020 vmcs_write16(GUEST_INTR_STATUS, status);
8024 static void vmx_set_rvi(int vector)
8032 status = vmcs_read16(GUEST_INTR_STATUS);
8033 old = (u8)status & 0xff;
8034 if ((u8)vector != old) {
8036 status |= (u8)vector;
8037 vmcs_write16(GUEST_INTR_STATUS, status);
8041 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
8043 if (!is_guest_mode(vcpu)) {
8044 vmx_set_rvi(max_irr);
8052 * In guest mode. If a vmexit is needed, vmx_check_nested_events
8055 if (nested_exit_on_intr(vcpu))
8059 * Else, fall back to pre-APICv interrupt injection since L2
8060 * is run without virtual interrupt delivery.
8062 if (!kvm_event_needs_reinjection(vcpu) &&
8063 vmx_interrupt_allowed(vcpu)) {
8064 kvm_queue_interrupt(vcpu, max_irr, false);
8065 vmx_inject_irq(vcpu);
8069 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
8071 if (!vmx_vm_has_apicv(vcpu->kvm))
8074 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
8075 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
8076 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
8077 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
8080 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
8084 if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
8085 || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
8088 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8089 exit_intr_info = vmx->exit_intr_info;
8091 /* Handle machine checks before interrupts are enabled */
8092 if (is_machine_check(exit_intr_info))
8093 kvm_machine_check();
8095 /* We need to handle NMIs before interrupts are enabled */
8096 if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
8097 (exit_intr_info & INTR_INFO_VALID_MASK)) {
8098 kvm_before_handle_nmi(&vmx->vcpu);
8100 kvm_after_handle_nmi(&vmx->vcpu);
8104 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
8106 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8109 * If external interrupt exists, IF bit is set in rflags/eflags on the
8110 * interrupt stack frame, and interrupt will be enabled on a return
8111 * from interrupt handler.
8113 if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
8114 == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
8115 unsigned int vector;
8116 unsigned long entry;
8118 struct vcpu_vmx *vmx = to_vmx(vcpu);
8119 #ifdef CONFIG_X86_64
8123 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
8124 desc = (gate_desc *)vmx->host_idt_base + vector;
8125 entry = gate_offset(*desc);
8127 #ifdef CONFIG_X86_64
8128 "mov %%" _ASM_SP ", %[sp]\n\t"
8129 "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
8134 "orl $0x200, (%%" _ASM_SP ")\n\t"
8135 __ASM_SIZE(push) " $%c[cs]\n\t"
8136 "call *%[entry]\n\t"
8138 #ifdef CONFIG_X86_64
8143 [ss]"i"(__KERNEL_DS),
8144 [cs]"i"(__KERNEL_CS)
8150 static bool vmx_mpx_supported(void)
8152 return (vmcs_config.vmexit_ctrl & VM_EXIT_CLEAR_BNDCFGS) &&
8153 (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_BNDCFGS);
8156 static bool vmx_xsaves_supported(void)
8158 return vmcs_config.cpu_based_2nd_exec_ctrl &
8159 SECONDARY_EXEC_XSAVES;
8162 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
8167 bool idtv_info_valid;
8169 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8171 if (cpu_has_virtual_nmis()) {
8172 if (vmx->nmi_known_unmasked)
8175 * Can't use vmx->exit_intr_info since we're not sure what
8176 * the exit reason is.
8178 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8179 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
8180 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
8182 * SDM 3: 27.7.1.2 (September 2008)
8183 * Re-set bit "block by NMI" before VM entry if vmexit caused by
8184 * a guest IRET fault.
8185 * SDM 3: 23.2.2 (September 2008)
8186 * Bit 12 is undefined in any of the following cases:
8187 * If the VM exit sets the valid bit in the IDT-vectoring
8188 * information field.
8189 * If the VM exit is due to a double fault.
8191 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
8192 vector != DF_VECTOR && !idtv_info_valid)
8193 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
8194 GUEST_INTR_STATE_NMI);
8196 vmx->nmi_known_unmasked =
8197 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
8198 & GUEST_INTR_STATE_NMI);
8199 } else if (unlikely(vmx->soft_vnmi_blocked))
8200 vmx->vnmi_blocked_time +=
8201 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
8204 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
8205 u32 idt_vectoring_info,
8206 int instr_len_field,
8207 int error_code_field)
8211 bool idtv_info_valid;
8213 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8215 vcpu->arch.nmi_injected = false;
8216 kvm_clear_exception_queue(vcpu);
8217 kvm_clear_interrupt_queue(vcpu);
8219 if (!idtv_info_valid)
8222 kvm_make_request(KVM_REQ_EVENT, vcpu);
8224 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
8225 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
8228 case INTR_TYPE_NMI_INTR:
8229 vcpu->arch.nmi_injected = true;
8231 * SDM 3: 27.7.1.2 (September 2008)
8232 * Clear bit "block by NMI" before VM entry if a NMI
8235 vmx_set_nmi_mask(vcpu, false);
8237 case INTR_TYPE_SOFT_EXCEPTION:
8238 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8240 case INTR_TYPE_HARD_EXCEPTION:
8241 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
8242 u32 err = vmcs_read32(error_code_field);
8243 kvm_requeue_exception_e(vcpu, vector, err);
8245 kvm_requeue_exception(vcpu, vector);
8247 case INTR_TYPE_SOFT_INTR:
8248 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8250 case INTR_TYPE_EXT_INTR:
8251 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
8258 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
8260 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
8261 VM_EXIT_INSTRUCTION_LEN,
8262 IDT_VECTORING_ERROR_CODE);
8265 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
8267 __vmx_complete_interrupts(vcpu,
8268 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8269 VM_ENTRY_INSTRUCTION_LEN,
8270 VM_ENTRY_EXCEPTION_ERROR_CODE);
8272 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
8275 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
8278 struct perf_guest_switch_msr *msrs;
8280 msrs = perf_guest_get_msrs(&nr_msrs);
8285 for (i = 0; i < nr_msrs; i++)
8286 if (msrs[i].host == msrs[i].guest)
8287 clear_atomic_switch_msr(vmx, msrs[i].msr);
8289 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
8293 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
8295 struct vcpu_vmx *vmx = to_vmx(vcpu);
8296 unsigned long debugctlmsr, cr4;
8298 /* Record the guest's net vcpu time for enforced NMI injections. */
8299 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
8300 vmx->entry_time = ktime_get();
8302 /* Don't enter VMX if guest state is invalid, let the exit handler
8303 start emulation until we arrive back to a valid state */
8304 if (vmx->emulation_required)
8307 if (vmx->ple_window_dirty) {
8308 vmx->ple_window_dirty = false;
8309 vmcs_write32(PLE_WINDOW, vmx->ple_window);
8312 if (vmx->nested.sync_shadow_vmcs) {
8313 copy_vmcs12_to_shadow(vmx);
8314 vmx->nested.sync_shadow_vmcs = false;
8317 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
8318 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
8319 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
8320 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
8322 cr4 = cr4_read_shadow();
8323 if (unlikely(cr4 != vmx->host_state.vmcs_host_cr4)) {
8324 vmcs_writel(HOST_CR4, cr4);
8325 vmx->host_state.vmcs_host_cr4 = cr4;
8328 /* When single-stepping over STI and MOV SS, we must clear the
8329 * corresponding interruptibility bits in the guest state. Otherwise
8330 * vmentry fails as it then expects bit 14 (BS) in pending debug
8331 * exceptions being set, but that's not correct for the guest debugging
8333 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8334 vmx_set_interrupt_shadow(vcpu, 0);
8336 atomic_switch_perf_msrs(vmx);
8337 debugctlmsr = get_debugctlmsr();
8339 vmx->__launched = vmx->loaded_vmcs->launched;
8341 /* Store host registers */
8342 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
8343 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
8344 "push %%" _ASM_CX " \n\t"
8345 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
8347 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
8348 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
8350 /* Reload cr2 if changed */
8351 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
8352 "mov %%cr2, %%" _ASM_DX " \n\t"
8353 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
8355 "mov %%" _ASM_AX", %%cr2 \n\t"
8357 /* Check if vmlaunch of vmresume is needed */
8358 "cmpl $0, %c[launched](%0) \n\t"
8359 /* Load guest registers. Don't clobber flags. */
8360 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
8361 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
8362 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
8363 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
8364 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
8365 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
8366 #ifdef CONFIG_X86_64
8367 "mov %c[r8](%0), %%r8 \n\t"
8368 "mov %c[r9](%0), %%r9 \n\t"
8369 "mov %c[r10](%0), %%r10 \n\t"
8370 "mov %c[r11](%0), %%r11 \n\t"
8371 "mov %c[r12](%0), %%r12 \n\t"
8372 "mov %c[r13](%0), %%r13 \n\t"
8373 "mov %c[r14](%0), %%r14 \n\t"
8374 "mov %c[r15](%0), %%r15 \n\t"
8376 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
8378 /* Enter guest mode */
8380 __ex(ASM_VMX_VMLAUNCH) "\n\t"
8382 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
8384 /* Save guest registers, load host registers, keep flags */
8385 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
8387 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
8388 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
8389 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
8390 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
8391 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
8392 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
8393 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
8394 #ifdef CONFIG_X86_64
8395 "mov %%r8, %c[r8](%0) \n\t"
8396 "mov %%r9, %c[r9](%0) \n\t"
8397 "mov %%r10, %c[r10](%0) \n\t"
8398 "mov %%r11, %c[r11](%0) \n\t"
8399 "mov %%r12, %c[r12](%0) \n\t"
8400 "mov %%r13, %c[r13](%0) \n\t"
8401 "mov %%r14, %c[r14](%0) \n\t"
8402 "mov %%r15, %c[r15](%0) \n\t"
8404 "mov %%cr2, %%" _ASM_AX " \n\t"
8405 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
8407 "pop %%" _ASM_BP "; pop %%" _ASM_DX " \n\t"
8408 "setbe %c[fail](%0) \n\t"
8409 ".pushsection .rodata \n\t"
8410 ".global vmx_return \n\t"
8411 "vmx_return: " _ASM_PTR " 2b \n\t"
8413 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
8414 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
8415 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
8416 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
8417 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
8418 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
8419 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
8420 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
8421 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
8422 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
8423 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
8424 #ifdef CONFIG_X86_64
8425 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
8426 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
8427 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
8428 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
8429 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
8430 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
8431 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
8432 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
8434 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
8435 [wordsize]"i"(sizeof(ulong))
8437 #ifdef CONFIG_X86_64
8438 , "rax", "rbx", "rdi", "rsi"
8439 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
8441 , "eax", "ebx", "edi", "esi"
8445 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
8447 update_debugctlmsr(debugctlmsr);
8449 #ifndef CONFIG_X86_64
8451 * The sysexit path does not restore ds/es, so we must set them to
8452 * a reasonable value ourselves.
8454 * We can't defer this to vmx_load_host_state() since that function
8455 * may be executed in interrupt context, which saves and restore segments
8456 * around it, nullifying its effect.
8458 loadsegment(ds, __USER_DS);
8459 loadsegment(es, __USER_DS);
8462 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
8463 | (1 << VCPU_EXREG_RFLAGS)
8464 | (1 << VCPU_EXREG_PDPTR)
8465 | (1 << VCPU_EXREG_SEGMENTS)
8466 | (1 << VCPU_EXREG_CR3));
8467 vcpu->arch.regs_dirty = 0;
8469 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
8471 vmx->loaded_vmcs->launched = 1;
8473 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
8474 trace_kvm_exit(vmx->exit_reason, vcpu, KVM_ISA_VMX);
8477 * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
8478 * we did not inject a still-pending event to L1 now because of
8479 * nested_run_pending, we need to re-enable this bit.
8481 if (vmx->nested.nested_run_pending)
8482 kvm_make_request(KVM_REQ_EVENT, vcpu);
8484 vmx->nested.nested_run_pending = 0;
8486 vmx_complete_atomic_exit(vmx);
8487 vmx_recover_nmi_blocking(vmx);
8488 vmx_complete_interrupts(vmx);
8491 static void vmx_load_vmcs01(struct kvm_vcpu *vcpu)
8493 struct vcpu_vmx *vmx = to_vmx(vcpu);
8496 if (vmx->loaded_vmcs == &vmx->vmcs01)
8500 vmx->loaded_vmcs = &vmx->vmcs01;
8502 vmx_vcpu_load(vcpu, cpu);
8507 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
8509 struct vcpu_vmx *vmx = to_vmx(vcpu);
8512 vmx_disable_pml(vmx);
8514 leave_guest_mode(vcpu);
8515 vmx_load_vmcs01(vcpu);
8517 free_loaded_vmcs(vmx->loaded_vmcs);
8518 kfree(vmx->guest_msrs);
8519 kvm_vcpu_uninit(vcpu);
8520 kmem_cache_free(kvm_vcpu_cache, vmx);
8523 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
8526 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
8530 return ERR_PTR(-ENOMEM);
8534 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
8538 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
8539 BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
8543 if (!vmx->guest_msrs) {
8547 vmx->loaded_vmcs = &vmx->vmcs01;
8548 vmx->loaded_vmcs->vmcs = alloc_vmcs();
8549 if (!vmx->loaded_vmcs->vmcs)
8552 kvm_cpu_vmxon(__pa(per_cpu(vmxarea, raw_smp_processor_id())));
8553 loaded_vmcs_init(vmx->loaded_vmcs);
8558 vmx_vcpu_load(&vmx->vcpu, cpu);
8559 vmx->vcpu.cpu = cpu;
8560 err = vmx_vcpu_setup(vmx);
8561 vmx_vcpu_put(&vmx->vcpu);
8565 if (vm_need_virtualize_apic_accesses(kvm)) {
8566 err = alloc_apic_access_page(kvm);
8572 if (!kvm->arch.ept_identity_map_addr)
8573 kvm->arch.ept_identity_map_addr =
8574 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
8575 err = init_rmode_identity_map(kvm);
8581 nested_vmx_setup_ctls_msrs(vmx);
8583 vmx->nested.posted_intr_nv = -1;
8584 vmx->nested.current_vmptr = -1ull;
8585 vmx->nested.current_vmcs12 = NULL;
8588 * If PML is turned on, failure on enabling PML just results in failure
8589 * of creating the vcpu, therefore we can simplify PML logic (by
8590 * avoiding dealing with cases, such as enabling PML partially on vcpus
8591 * for the guest, etc.
8594 err = vmx_enable_pml(vmx);
8602 free_loaded_vmcs(vmx->loaded_vmcs);
8604 kfree(vmx->guest_msrs);
8606 kvm_vcpu_uninit(&vmx->vcpu);
8609 kmem_cache_free(kvm_vcpu_cache, vmx);
8610 return ERR_PTR(err);
8613 static void __init vmx_check_processor_compat(void *rtn)
8615 struct vmcs_config vmcs_conf;
8618 if (setup_vmcs_config(&vmcs_conf) < 0)
8620 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
8621 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
8622 smp_processor_id());
8627 static int get_ept_level(void)
8629 return VMX_EPT_DEFAULT_GAW + 1;
8632 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
8636 /* For VT-d and EPT combination
8637 * 1. MMIO: always map as UC
8639 * a. VT-d without snooping control feature: can't guarantee the
8640 * result, try to trust guest.
8641 * b. VT-d with snooping control feature: snooping control feature of
8642 * VT-d engine can guarantee the cache correctness. Just set it
8643 * to WB to keep consistent with host. So the same as item 3.
8644 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
8645 * consistent with host MTRR
8648 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
8649 else if (kvm_arch_has_noncoherent_dma(vcpu->kvm))
8650 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
8651 VMX_EPT_MT_EPTE_SHIFT;
8653 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
8659 static int vmx_get_lpage_level(void)
8661 if (enable_ept && !cpu_has_vmx_ept_1g_page())
8662 return PT_DIRECTORY_LEVEL;
8664 /* For shadow and EPT supported 1GB page */
8665 return PT_PDPE_LEVEL;
8668 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
8670 struct kvm_cpuid_entry2 *best;
8671 struct vcpu_vmx *vmx = to_vmx(vcpu);
8674 vmx->rdtscp_enabled = false;
8675 if (vmx_rdtscp_supported()) {
8676 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8677 if (exec_control & SECONDARY_EXEC_RDTSCP) {
8678 best = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
8679 if (best && (best->edx & bit(X86_FEATURE_RDTSCP)))
8680 vmx->rdtscp_enabled = true;
8682 exec_control &= ~SECONDARY_EXEC_RDTSCP;
8683 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
8687 if (nested && !vmx->rdtscp_enabled)
8688 vmx->nested.nested_vmx_secondary_ctls_high &=
8689 ~SECONDARY_EXEC_RDTSCP;
8692 /* Exposing INVPCID only when PCID is exposed */
8693 best = kvm_find_cpuid_entry(vcpu, 0x7, 0);
8694 if (vmx_invpcid_supported() &&
8695 best && (best->ebx & bit(X86_FEATURE_INVPCID)) &&
8696 guest_cpuid_has_pcid(vcpu)) {
8697 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8698 exec_control |= SECONDARY_EXEC_ENABLE_INVPCID;
8699 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
8702 if (cpu_has_secondary_exec_ctrls()) {
8703 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8704 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
8705 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
8709 best->ebx &= ~bit(X86_FEATURE_INVPCID);
8713 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
8715 if (func == 1 && nested)
8716 entry->ecx |= bit(X86_FEATURE_VMX);
8719 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
8720 struct x86_exception *fault)
8722 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8725 if (fault->error_code & PFERR_RSVD_MASK)
8726 exit_reason = EXIT_REASON_EPT_MISCONFIG;
8728 exit_reason = EXIT_REASON_EPT_VIOLATION;
8729 nested_vmx_vmexit(vcpu, exit_reason, 0, vcpu->arch.exit_qualification);
8730 vmcs12->guest_physical_address = fault->address;
8733 /* Callbacks for nested_ept_init_mmu_context: */
8735 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
8737 /* return the page table to be shadowed - in our case, EPT12 */
8738 return get_vmcs12(vcpu)->ept_pointer;
8741 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
8743 WARN_ON(mmu_is_nested(vcpu));
8744 kvm_init_shadow_ept_mmu(vcpu,
8745 to_vmx(vcpu)->nested.nested_vmx_ept_caps &
8746 VMX_EPT_EXECUTE_ONLY_BIT);
8747 vcpu->arch.mmu.set_cr3 = vmx_set_cr3;
8748 vcpu->arch.mmu.get_cr3 = nested_ept_get_cr3;
8749 vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
8751 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
8754 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
8756 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
8759 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
8762 bool inequality, bit;
8764 bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
8766 (error_code & vmcs12->page_fault_error_code_mask) !=
8767 vmcs12->page_fault_error_code_match;
8768 return inequality ^ bit;
8771 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
8772 struct x86_exception *fault)
8774 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8776 WARN_ON(!is_guest_mode(vcpu));
8778 if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code))
8779 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
8780 vmcs_read32(VM_EXIT_INTR_INFO),
8781 vmcs_readl(EXIT_QUALIFICATION));
8783 kvm_inject_page_fault(vcpu, fault);
8786 static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu,
8787 struct vmcs12 *vmcs12)
8789 struct vcpu_vmx *vmx = to_vmx(vcpu);
8790 int maxphyaddr = cpuid_maxphyaddr(vcpu);
8792 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
8793 if (!PAGE_ALIGNED(vmcs12->apic_access_addr) ||
8794 vmcs12->apic_access_addr >> maxphyaddr)
8798 * Translate L1 physical address to host physical
8799 * address for vmcs02. Keep the page pinned, so this
8800 * physical address remains valid. We keep a reference
8801 * to it so we can release it later.
8803 if (vmx->nested.apic_access_page) /* shouldn't happen */
8804 nested_release_page(vmx->nested.apic_access_page);
8805 vmx->nested.apic_access_page =
8806 nested_get_page(vcpu, vmcs12->apic_access_addr);
8809 if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
8810 if (!PAGE_ALIGNED(vmcs12->virtual_apic_page_addr) ||
8811 vmcs12->virtual_apic_page_addr >> maxphyaddr)
8814 if (vmx->nested.virtual_apic_page) /* shouldn't happen */
8815 nested_release_page(vmx->nested.virtual_apic_page);
8816 vmx->nested.virtual_apic_page =
8817 nested_get_page(vcpu, vmcs12->virtual_apic_page_addr);
8820 * Failing the vm entry is _not_ what the processor does
8821 * but it's basically the only possibility we have.
8822 * We could still enter the guest if CR8 load exits are
8823 * enabled, CR8 store exits are enabled, and virtualize APIC
8824 * access is disabled; in this case the processor would never
8825 * use the TPR shadow and we could simply clear the bit from
8826 * the execution control. But such a configuration is useless,
8827 * so let's keep the code simple.
8829 if (!vmx->nested.virtual_apic_page)
8833 if (nested_cpu_has_posted_intr(vmcs12)) {
8834 if (!IS_ALIGNED(vmcs12->posted_intr_desc_addr, 64) ||
8835 vmcs12->posted_intr_desc_addr >> maxphyaddr)
8838 if (vmx->nested.pi_desc_page) { /* shouldn't happen */
8839 kunmap(vmx->nested.pi_desc_page);
8840 nested_release_page(vmx->nested.pi_desc_page);
8842 vmx->nested.pi_desc_page =
8843 nested_get_page(vcpu, vmcs12->posted_intr_desc_addr);
8844 if (!vmx->nested.pi_desc_page)
8847 vmx->nested.pi_desc =
8848 (struct pi_desc *)kmap(vmx->nested.pi_desc_page);
8849 if (!vmx->nested.pi_desc) {
8850 nested_release_page_clean(vmx->nested.pi_desc_page);
8853 vmx->nested.pi_desc =
8854 (struct pi_desc *)((void *)vmx->nested.pi_desc +
8855 (unsigned long)(vmcs12->posted_intr_desc_addr &
8862 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
8864 u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
8865 struct vcpu_vmx *vmx = to_vmx(vcpu);
8867 if (vcpu->arch.virtual_tsc_khz == 0)
8870 /* Make sure short timeouts reliably trigger an immediate vmexit.
8871 * hrtimer_start does not guarantee this. */
8872 if (preemption_timeout <= 1) {
8873 vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
8877 preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
8878 preemption_timeout *= 1000000;
8879 do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
8880 hrtimer_start(&vmx->nested.preemption_timer,
8881 ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
8884 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
8885 struct vmcs12 *vmcs12)
8890 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
8893 if (vmcs12_read_any(vcpu, MSR_BITMAP, &addr)) {
8897 maxphyaddr = cpuid_maxphyaddr(vcpu);
8899 if (!PAGE_ALIGNED(vmcs12->msr_bitmap) ||
8900 ((addr + PAGE_SIZE) >> maxphyaddr))
8907 * Merge L0's and L1's MSR bitmap, return false to indicate that
8908 * we do not use the hardware.
8910 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
8911 struct vmcs12 *vmcs12)
8915 unsigned long *msr_bitmap;
8917 if (!nested_cpu_has_virt_x2apic_mode(vmcs12))
8920 page = nested_get_page(vcpu, vmcs12->msr_bitmap);
8925 msr_bitmap = (unsigned long *)kmap(page);
8927 nested_release_page_clean(page);
8932 if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
8933 if (nested_cpu_has_apic_reg_virt(vmcs12))
8934 for (msr = 0x800; msr <= 0x8ff; msr++)
8935 nested_vmx_disable_intercept_for_msr(
8937 vmx_msr_bitmap_nested,
8939 /* TPR is allowed */
8940 nested_vmx_disable_intercept_for_msr(msr_bitmap,
8941 vmx_msr_bitmap_nested,
8942 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
8943 MSR_TYPE_R | MSR_TYPE_W);
8944 if (nested_cpu_has_vid(vmcs12)) {
8945 /* EOI and self-IPI are allowed */
8946 nested_vmx_disable_intercept_for_msr(
8948 vmx_msr_bitmap_nested,
8949 APIC_BASE_MSR + (APIC_EOI >> 4),
8951 nested_vmx_disable_intercept_for_msr(
8953 vmx_msr_bitmap_nested,
8954 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
8959 * Enable reading intercept of all the x2apic
8960 * MSRs. We should not rely on vmcs12 to do any
8961 * optimizations here, it may have been modified
8964 for (msr = 0x800; msr <= 0x8ff; msr++)
8965 __vmx_enable_intercept_for_msr(
8966 vmx_msr_bitmap_nested,
8970 __vmx_enable_intercept_for_msr(
8971 vmx_msr_bitmap_nested,
8972 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
8974 __vmx_enable_intercept_for_msr(
8975 vmx_msr_bitmap_nested,
8976 APIC_BASE_MSR + (APIC_EOI >> 4),
8978 __vmx_enable_intercept_for_msr(
8979 vmx_msr_bitmap_nested,
8980 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
8984 nested_release_page_clean(page);
8989 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
8990 struct vmcs12 *vmcs12)
8992 if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
8993 !nested_cpu_has_apic_reg_virt(vmcs12) &&
8994 !nested_cpu_has_vid(vmcs12) &&
8995 !nested_cpu_has_posted_intr(vmcs12))
8999 * If virtualize x2apic mode is enabled,
9000 * virtualize apic access must be disabled.
9002 if (nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9003 nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
9007 * If virtual interrupt delivery is enabled,
9008 * we must exit on external interrupts.
9010 if (nested_cpu_has_vid(vmcs12) &&
9011 !nested_exit_on_intr(vcpu))
9015 * bits 15:8 should be zero in posted_intr_nv,
9016 * the descriptor address has been already checked
9017 * in nested_get_vmcs12_pages.
9019 if (nested_cpu_has_posted_intr(vmcs12) &&
9020 (!nested_cpu_has_vid(vmcs12) ||
9021 !nested_exit_intr_ack_set(vcpu) ||
9022 vmcs12->posted_intr_nv & 0xff00))
9025 /* tpr shadow is needed by all apicv features. */
9026 if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
9032 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
9033 unsigned long count_field,
9034 unsigned long addr_field)
9039 if (vmcs12_read_any(vcpu, count_field, &count) ||
9040 vmcs12_read_any(vcpu, addr_field, &addr)) {
9046 maxphyaddr = cpuid_maxphyaddr(vcpu);
9047 if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
9048 (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr) {
9049 pr_warn_ratelimited(
9050 "nVMX: invalid MSR switch (0x%lx, %d, %llu, 0x%08llx)",
9051 addr_field, maxphyaddr, count, addr);
9057 static int nested_vmx_check_msr_switch_controls(struct kvm_vcpu *vcpu,
9058 struct vmcs12 *vmcs12)
9060 if (vmcs12->vm_exit_msr_load_count == 0 &&
9061 vmcs12->vm_exit_msr_store_count == 0 &&
9062 vmcs12->vm_entry_msr_load_count == 0)
9063 return 0; /* Fast path */
9064 if (nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_LOAD_COUNT,
9065 VM_EXIT_MSR_LOAD_ADDR) ||
9066 nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_STORE_COUNT,
9067 VM_EXIT_MSR_STORE_ADDR) ||
9068 nested_vmx_check_msr_switch(vcpu, VM_ENTRY_MSR_LOAD_COUNT,
9069 VM_ENTRY_MSR_LOAD_ADDR))
9074 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
9075 struct vmx_msr_entry *e)
9077 /* x2APIC MSR accesses are not allowed */
9078 if (vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8)
9080 if (e->index == MSR_IA32_UCODE_WRITE || /* SDM Table 35-2 */
9081 e->index == MSR_IA32_UCODE_REV)
9083 if (e->reserved != 0)
9088 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
9089 struct vmx_msr_entry *e)
9091 if (e->index == MSR_FS_BASE ||
9092 e->index == MSR_GS_BASE ||
9093 e->index == MSR_IA32_SMM_MONITOR_CTL || /* SMM is not supported */
9094 nested_vmx_msr_check_common(vcpu, e))
9099 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
9100 struct vmx_msr_entry *e)
9102 if (e->index == MSR_IA32_SMBASE || /* SMM is not supported */
9103 nested_vmx_msr_check_common(vcpu, e))
9109 * Load guest's/host's msr at nested entry/exit.
9110 * return 0 for success, entry index for failure.
9112 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9115 struct vmx_msr_entry e;
9116 struct msr_data msr;
9118 msr.host_initiated = false;
9119 for (i = 0; i < count; i++) {
9120 if (kvm_read_guest(vcpu->kvm, gpa + i * sizeof(e),
9122 pr_warn_ratelimited(
9123 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9124 __func__, i, gpa + i * sizeof(e));
9127 if (nested_vmx_load_msr_check(vcpu, &e)) {
9128 pr_warn_ratelimited(
9129 "%s check failed (%u, 0x%x, 0x%x)\n",
9130 __func__, i, e.index, e.reserved);
9133 msr.index = e.index;
9135 if (kvm_set_msr(vcpu, &msr)) {
9136 pr_warn_ratelimited(
9137 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9138 __func__, i, e.index, e.value);
9147 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9150 struct vmx_msr_entry e;
9152 for (i = 0; i < count; i++) {
9153 if (kvm_read_guest(vcpu->kvm,
9154 gpa + i * sizeof(e),
9155 &e, 2 * sizeof(u32))) {
9156 pr_warn_ratelimited(
9157 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9158 __func__, i, gpa + i * sizeof(e));
9161 if (nested_vmx_store_msr_check(vcpu, &e)) {
9162 pr_warn_ratelimited(
9163 "%s check failed (%u, 0x%x, 0x%x)\n",
9164 __func__, i, e.index, e.reserved);
9167 if (kvm_get_msr(vcpu, e.index, &e.value)) {
9168 pr_warn_ratelimited(
9169 "%s cannot read MSR (%u, 0x%x)\n",
9170 __func__, i, e.index);
9173 if (kvm_write_guest(vcpu->kvm,
9174 gpa + i * sizeof(e) +
9175 offsetof(struct vmx_msr_entry, value),
9176 &e.value, sizeof(e.value))) {
9177 pr_warn_ratelimited(
9178 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9179 __func__, i, e.index, e.value);
9187 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
9188 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
9189 * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
9190 * guest in a way that will both be appropriate to L1's requests, and our
9191 * needs. In addition to modifying the active vmcs (which is vmcs02), this
9192 * function also has additional necessary side-effects, like setting various
9193 * vcpu->arch fields.
9195 static void prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
9197 struct vcpu_vmx *vmx = to_vmx(vcpu);
9200 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
9201 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
9202 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
9203 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
9204 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
9205 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
9206 vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
9207 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
9208 vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
9209 vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
9210 vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
9211 vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
9212 vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
9213 vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
9214 vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
9215 vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
9216 vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
9217 vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
9218 vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
9219 vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
9220 vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
9221 vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
9222 vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
9223 vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
9224 vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
9225 vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
9226 vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
9227 vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
9228 vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
9229 vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
9230 vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
9231 vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
9232 vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
9233 vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
9234 vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
9235 vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
9237 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
9238 kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
9239 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
9241 kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
9242 vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
9244 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
9245 vmcs12->vm_entry_intr_info_field);
9246 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
9247 vmcs12->vm_entry_exception_error_code);
9248 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
9249 vmcs12->vm_entry_instruction_len);
9250 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
9251 vmcs12->guest_interruptibility_info);
9252 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
9253 vmx_set_rflags(vcpu, vmcs12->guest_rflags);
9254 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
9255 vmcs12->guest_pending_dbg_exceptions);
9256 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
9257 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
9259 if (nested_cpu_has_xsaves(vmcs12))
9260 vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
9261 vmcs_write64(VMCS_LINK_POINTER, -1ull);
9263 exec_control = vmcs12->pin_based_vm_exec_control;
9264 exec_control |= vmcs_config.pin_based_exec_ctrl;
9265 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
9267 if (nested_cpu_has_posted_intr(vmcs12)) {
9269 * Note that we use L0's vector here and in
9270 * vmx_deliver_nested_posted_interrupt.
9272 vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
9273 vmx->nested.pi_pending = false;
9274 vmcs_write64(POSTED_INTR_NV, POSTED_INTR_VECTOR);
9275 vmcs_write64(POSTED_INTR_DESC_ADDR,
9276 page_to_phys(vmx->nested.pi_desc_page) +
9277 (unsigned long)(vmcs12->posted_intr_desc_addr &
9280 exec_control &= ~PIN_BASED_POSTED_INTR;
9282 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
9284 vmx->nested.preemption_timer_expired = false;
9285 if (nested_cpu_has_preemption_timer(vmcs12))
9286 vmx_start_preemption_timer(vcpu);
9289 * Whether page-faults are trapped is determined by a combination of
9290 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
9291 * If enable_ept, L0 doesn't care about page faults and we should
9292 * set all of these to L1's desires. However, if !enable_ept, L0 does
9293 * care about (at least some) page faults, and because it is not easy
9294 * (if at all possible?) to merge L0 and L1's desires, we simply ask
9295 * to exit on each and every L2 page fault. This is done by setting
9296 * MASK=MATCH=0 and (see below) EB.PF=1.
9297 * Note that below we don't need special code to set EB.PF beyond the
9298 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
9299 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
9300 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
9302 * A problem with this approach (when !enable_ept) is that L1 may be
9303 * injected with more page faults than it asked for. This could have
9304 * caused problems, but in practice existing hypervisors don't care.
9305 * To fix this, we will need to emulate the PFEC checking (on the L1
9306 * page tables), using walk_addr(), when injecting PFs to L1.
9308 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
9309 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
9310 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
9311 enable_ept ? vmcs12->page_fault_error_code_match : 0);
9313 if (cpu_has_secondary_exec_ctrls()) {
9314 exec_control = vmx_secondary_exec_control(vmx);
9315 if (!vmx->rdtscp_enabled)
9316 exec_control &= ~SECONDARY_EXEC_RDTSCP;
9317 /* Take the following fields only from vmcs12 */
9318 exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
9319 SECONDARY_EXEC_RDTSCP |
9320 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
9321 SECONDARY_EXEC_APIC_REGISTER_VIRT);
9322 if (nested_cpu_has(vmcs12,
9323 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
9324 exec_control |= vmcs12->secondary_vm_exec_control;
9326 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) {
9328 * If translation failed, no matter: This feature asks
9329 * to exit when accessing the given address, and if it
9330 * can never be accessed, this feature won't do
9333 if (!vmx->nested.apic_access_page)
9335 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9337 vmcs_write64(APIC_ACCESS_ADDR,
9338 page_to_phys(vmx->nested.apic_access_page));
9339 } else if (!(nested_cpu_has_virt_x2apic_mode(vmcs12)) &&
9340 (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))) {
9342 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9343 kvm_vcpu_reload_apic_access_page(vcpu);
9346 if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
9347 vmcs_write64(EOI_EXIT_BITMAP0,
9348 vmcs12->eoi_exit_bitmap0);
9349 vmcs_write64(EOI_EXIT_BITMAP1,
9350 vmcs12->eoi_exit_bitmap1);
9351 vmcs_write64(EOI_EXIT_BITMAP2,
9352 vmcs12->eoi_exit_bitmap2);
9353 vmcs_write64(EOI_EXIT_BITMAP3,
9354 vmcs12->eoi_exit_bitmap3);
9355 vmcs_write16(GUEST_INTR_STATUS,
9356 vmcs12->guest_intr_status);
9359 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
9364 * Set host-state according to L0's settings (vmcs12 is irrelevant here)
9365 * Some constant fields are set here by vmx_set_constant_host_state().
9366 * Other fields are different per CPU, and will be set later when
9367 * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
9369 vmx_set_constant_host_state(vmx);
9372 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
9373 * entry, but only if the current (host) sp changed from the value
9374 * we wrote last (vmx->host_rsp). This cache is no longer relevant
9375 * if we switch vmcs, and rather than hold a separate cache per vmcs,
9376 * here we just force the write to happen on entry.
9380 exec_control = vmx_exec_control(vmx); /* L0's desires */
9381 exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
9382 exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
9383 exec_control &= ~CPU_BASED_TPR_SHADOW;
9384 exec_control |= vmcs12->cpu_based_vm_exec_control;
9386 if (exec_control & CPU_BASED_TPR_SHADOW) {
9387 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
9388 page_to_phys(vmx->nested.virtual_apic_page));
9389 vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
9392 if (cpu_has_vmx_msr_bitmap() &&
9393 exec_control & CPU_BASED_USE_MSR_BITMAPS) {
9394 nested_vmx_merge_msr_bitmap(vcpu, vmcs12);
9395 /* MSR_BITMAP will be set by following vmx_set_efer. */
9397 exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
9400 * Merging of IO bitmap not currently supported.
9401 * Rather, exit every time.
9403 exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
9404 exec_control |= CPU_BASED_UNCOND_IO_EXITING;
9406 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
9408 /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
9409 * bitwise-or of what L1 wants to trap for L2, and what we want to
9410 * trap. Note that CR0.TS also needs updating - we do this later.
9412 update_exception_bitmap(vcpu);
9413 vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
9414 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
9416 /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
9417 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
9418 * bits are further modified by vmx_set_efer() below.
9420 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
9422 /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
9423 * emulated by vmx_set_efer(), below.
9425 vm_entry_controls_init(vmx,
9426 (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER &
9427 ~VM_ENTRY_IA32E_MODE) |
9428 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
9430 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) {
9431 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
9432 vcpu->arch.pat = vmcs12->guest_ia32_pat;
9433 } else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
9434 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
9437 set_cr4_guest_host_mask(vmx);
9439 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)
9440 vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
9442 if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
9443 vmcs_write64(TSC_OFFSET,
9444 vmx->nested.vmcs01_tsc_offset + vmcs12->tsc_offset);
9446 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
9450 * Trivially support vpid by letting L2s share their parent
9451 * L1's vpid. TODO: move to a more elaborate solution, giving
9452 * each L2 its own vpid and exposing the vpid feature to L1.
9454 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
9455 vmx_flush_tlb(vcpu);
9458 if (nested_cpu_has_ept(vmcs12)) {
9459 kvm_mmu_unload(vcpu);
9460 nested_ept_init_mmu_context(vcpu);
9463 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)
9464 vcpu->arch.efer = vmcs12->guest_ia32_efer;
9465 else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
9466 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
9468 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
9469 /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
9470 vmx_set_efer(vcpu, vcpu->arch.efer);
9473 * This sets GUEST_CR0 to vmcs12->guest_cr0, with possibly a modified
9474 * TS bit (for lazy fpu) and bits which we consider mandatory enabled.
9475 * The CR0_READ_SHADOW is what L2 should have expected to read given
9476 * the specifications by L1; It's not enough to take
9477 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
9478 * have more bits than L1 expected.
9480 vmx_set_cr0(vcpu, vmcs12->guest_cr0);
9481 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
9483 vmx_set_cr4(vcpu, vmcs12->guest_cr4);
9484 vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
9486 /* shadow page tables on either EPT or shadow page tables */
9487 kvm_set_cr3(vcpu, vmcs12->guest_cr3);
9488 kvm_mmu_reset_context(vcpu);
9491 vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
9494 * L1 may access the L2's PDPTR, so save them to construct vmcs12
9497 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
9498 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
9499 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
9500 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
9503 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
9504 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
9508 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
9509 * for running an L2 nested guest.
9511 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
9513 struct vmcs12 *vmcs12;
9514 struct vcpu_vmx *vmx = to_vmx(vcpu);
9516 struct loaded_vmcs *vmcs02;
9520 if (!nested_vmx_check_permission(vcpu) ||
9521 !nested_vmx_check_vmcs12(vcpu))
9524 skip_emulated_instruction(vcpu);
9525 vmcs12 = get_vmcs12(vcpu);
9527 if (enable_shadow_vmcs)
9528 copy_shadow_to_vmcs12(vmx);
9531 * The nested entry process starts with enforcing various prerequisites
9532 * on vmcs12 as required by the Intel SDM, and act appropriately when
9533 * they fail: As the SDM explains, some conditions should cause the
9534 * instruction to fail, while others will cause the instruction to seem
9535 * to succeed, but return an EXIT_REASON_INVALID_STATE.
9536 * To speed up the normal (success) code path, we should avoid checking
9537 * for misconfigurations which will anyway be caught by the processor
9538 * when using the merged vmcs02.
9540 if (vmcs12->launch_state == launch) {
9541 nested_vmx_failValid(vcpu,
9542 launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
9543 : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
9547 if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
9548 vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT) {
9549 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9553 if (!nested_get_vmcs12_pages(vcpu, vmcs12)) {
9554 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9558 if (nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12)) {
9559 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9563 if (nested_vmx_check_apicv_controls(vcpu, vmcs12)) {
9564 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9568 if (nested_vmx_check_msr_switch_controls(vcpu, vmcs12)) {
9569 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9573 if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
9574 vmx->nested.nested_vmx_true_procbased_ctls_low,
9575 vmx->nested.nested_vmx_procbased_ctls_high) ||
9576 !vmx_control_verify(vmcs12->secondary_vm_exec_control,
9577 vmx->nested.nested_vmx_secondary_ctls_low,
9578 vmx->nested.nested_vmx_secondary_ctls_high) ||
9579 !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
9580 vmx->nested.nested_vmx_pinbased_ctls_low,
9581 vmx->nested.nested_vmx_pinbased_ctls_high) ||
9582 !vmx_control_verify(vmcs12->vm_exit_controls,
9583 vmx->nested.nested_vmx_true_exit_ctls_low,
9584 vmx->nested.nested_vmx_exit_ctls_high) ||
9585 !vmx_control_verify(vmcs12->vm_entry_controls,
9586 vmx->nested.nested_vmx_true_entry_ctls_low,
9587 vmx->nested.nested_vmx_entry_ctls_high))
9589 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9593 if (((vmcs12->host_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
9594 ((vmcs12->host_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
9595 nested_vmx_failValid(vcpu,
9596 VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
9600 if (!nested_cr0_valid(vcpu, vmcs12->guest_cr0) ||
9601 ((vmcs12->guest_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
9602 nested_vmx_entry_failure(vcpu, vmcs12,
9603 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
9606 if (vmcs12->vmcs_link_pointer != -1ull) {
9607 nested_vmx_entry_failure(vcpu, vmcs12,
9608 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_VMCS_LINK_PTR);
9613 * If the load IA32_EFER VM-entry control is 1, the following checks
9614 * are performed on the field for the IA32_EFER MSR:
9615 * - Bits reserved in the IA32_EFER MSR must be 0.
9616 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
9617 * the IA-32e mode guest VM-exit control. It must also be identical
9618 * to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
9621 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER) {
9622 ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
9623 if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
9624 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
9625 ((vmcs12->guest_cr0 & X86_CR0_PG) &&
9626 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))) {
9627 nested_vmx_entry_failure(vcpu, vmcs12,
9628 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
9634 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
9635 * IA32_EFER MSR must be 0 in the field for that register. In addition,
9636 * the values of the LMA and LME bits in the field must each be that of
9637 * the host address-space size VM-exit control.
9639 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
9640 ia32e = (vmcs12->vm_exit_controls &
9641 VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
9642 if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
9643 ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
9644 ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)) {
9645 nested_vmx_entry_failure(vcpu, vmcs12,
9646 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
9652 * We're finally done with prerequisite checking, and can start with
9656 vmcs02 = nested_get_current_vmcs02(vmx);
9660 enter_guest_mode(vcpu);
9662 vmx->nested.vmcs01_tsc_offset = vmcs_read64(TSC_OFFSET);
9664 if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
9665 vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
9668 vmx->loaded_vmcs = vmcs02;
9670 vmx_vcpu_load(vcpu, cpu);
9674 vmx_segment_cache_clear(vmx);
9676 prepare_vmcs02(vcpu, vmcs12);
9678 msr_entry_idx = nested_vmx_load_msr(vcpu,
9679 vmcs12->vm_entry_msr_load_addr,
9680 vmcs12->vm_entry_msr_load_count);
9681 if (msr_entry_idx) {
9682 leave_guest_mode(vcpu);
9683 vmx_load_vmcs01(vcpu);
9684 nested_vmx_entry_failure(vcpu, vmcs12,
9685 EXIT_REASON_MSR_LOAD_FAIL, msr_entry_idx);
9689 vmcs12->launch_state = 1;
9691 if (vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT)
9692 return kvm_vcpu_halt(vcpu);
9694 vmx->nested.nested_run_pending = 1;
9697 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
9698 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
9699 * returned as far as L1 is concerned. It will only return (and set
9700 * the success flag) when L2 exits (see nested_vmx_vmexit()).
9706 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
9707 * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
9708 * This function returns the new value we should put in vmcs12.guest_cr0.
9709 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
9710 * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
9711 * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
9712 * didn't trap the bit, because if L1 did, so would L0).
9713 * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have
9714 * been modified by L2, and L1 knows it. So just leave the old value of
9715 * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
9716 * isn't relevant, because if L0 traps this bit it can set it to anything.
9717 * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
9718 * changed these bits, and therefore they need to be updated, but L0
9719 * didn't necessarily allow them to be changed in GUEST_CR0 - and rather
9720 * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
9722 static inline unsigned long
9723 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
9726 /*1*/ (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
9727 /*2*/ (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
9728 /*3*/ (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
9729 vcpu->arch.cr0_guest_owned_bits));
9732 static inline unsigned long
9733 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
9736 /*1*/ (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
9737 /*2*/ (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
9738 /*3*/ (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
9739 vcpu->arch.cr4_guest_owned_bits));
9742 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
9743 struct vmcs12 *vmcs12)
9748 if (vcpu->arch.exception.pending && vcpu->arch.exception.reinject) {
9749 nr = vcpu->arch.exception.nr;
9750 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
9752 if (kvm_exception_is_soft(nr)) {
9753 vmcs12->vm_exit_instruction_len =
9754 vcpu->arch.event_exit_inst_len;
9755 idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
9757 idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
9759 if (vcpu->arch.exception.has_error_code) {
9760 idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
9761 vmcs12->idt_vectoring_error_code =
9762 vcpu->arch.exception.error_code;
9765 vmcs12->idt_vectoring_info_field = idt_vectoring;
9766 } else if (vcpu->arch.nmi_injected) {
9767 vmcs12->idt_vectoring_info_field =
9768 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
9769 } else if (vcpu->arch.interrupt.pending) {
9770 nr = vcpu->arch.interrupt.nr;
9771 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
9773 if (vcpu->arch.interrupt.soft) {
9774 idt_vectoring |= INTR_TYPE_SOFT_INTR;
9775 vmcs12->vm_entry_instruction_len =
9776 vcpu->arch.event_exit_inst_len;
9778 idt_vectoring |= INTR_TYPE_EXT_INTR;
9780 vmcs12->idt_vectoring_info_field = idt_vectoring;
9784 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
9786 struct vcpu_vmx *vmx = to_vmx(vcpu);
9788 if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
9789 vmx->nested.preemption_timer_expired) {
9790 if (vmx->nested.nested_run_pending)
9792 nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
9796 if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
9797 if (vmx->nested.nested_run_pending ||
9798 vcpu->arch.interrupt.pending)
9800 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
9801 NMI_VECTOR | INTR_TYPE_NMI_INTR |
9802 INTR_INFO_VALID_MASK, 0);
9804 * The NMI-triggered VM exit counts as injection:
9805 * clear this one and block further NMIs.
9807 vcpu->arch.nmi_pending = 0;
9808 vmx_set_nmi_mask(vcpu, true);
9812 if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
9813 nested_exit_on_intr(vcpu)) {
9814 if (vmx->nested.nested_run_pending)
9816 nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
9820 return vmx_complete_nested_posted_interrupt(vcpu);
9823 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
9826 hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
9829 if (ktime_to_ns(remaining) <= 0)
9832 value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
9833 do_div(value, 1000000);
9834 return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
9838 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
9839 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
9840 * and this function updates it to reflect the changes to the guest state while
9841 * L2 was running (and perhaps made some exits which were handled directly by L0
9842 * without going back to L1), and to reflect the exit reason.
9843 * Note that we do not have to copy here all VMCS fields, just those that
9844 * could have changed by the L2 guest or the exit - i.e., the guest-state and
9845 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
9846 * which already writes to vmcs12 directly.
9848 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
9849 u32 exit_reason, u32 exit_intr_info,
9850 unsigned long exit_qualification)
9852 /* update guest state fields: */
9853 vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
9854 vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
9856 vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
9857 vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
9858 vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
9860 vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
9861 vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
9862 vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
9863 vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
9864 vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
9865 vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
9866 vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
9867 vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
9868 vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
9869 vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
9870 vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
9871 vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
9872 vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
9873 vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
9874 vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
9875 vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
9876 vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
9877 vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
9878 vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
9879 vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
9880 vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
9881 vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
9882 vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
9883 vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
9884 vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
9885 vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
9886 vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
9887 vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
9888 vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
9889 vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
9890 vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
9891 vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
9892 vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
9893 vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
9894 vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
9895 vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
9897 vmcs12->guest_interruptibility_info =
9898 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
9899 vmcs12->guest_pending_dbg_exceptions =
9900 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
9901 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
9902 vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
9904 vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
9906 if (nested_cpu_has_preemption_timer(vmcs12)) {
9907 if (vmcs12->vm_exit_controls &
9908 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
9909 vmcs12->vmx_preemption_timer_value =
9910 vmx_get_preemption_timer_value(vcpu);
9911 hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
9915 * In some cases (usually, nested EPT), L2 is allowed to change its
9916 * own CR3 without exiting. If it has changed it, we must keep it.
9917 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
9918 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
9920 * Additionally, restore L2's PDPTR to vmcs12.
9923 vmcs12->guest_cr3 = vmcs_read64(GUEST_CR3);
9924 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
9925 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
9926 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
9927 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
9930 if (nested_cpu_has_vid(vmcs12))
9931 vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
9933 vmcs12->vm_entry_controls =
9934 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
9935 (vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
9937 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) {
9938 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
9939 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
9942 /* TODO: These cannot have changed unless we have MSR bitmaps and
9943 * the relevant bit asks not to trap the change */
9944 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
9945 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
9946 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
9947 vmcs12->guest_ia32_efer = vcpu->arch.efer;
9948 vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
9949 vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
9950 vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
9951 if (vmx_mpx_supported())
9952 vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
9953 if (nested_cpu_has_xsaves(vmcs12))
9954 vmcs12->xss_exit_bitmap = vmcs_read64(XSS_EXIT_BITMAP);
9956 /* update exit information fields: */
9958 vmcs12->vm_exit_reason = exit_reason;
9959 vmcs12->exit_qualification = exit_qualification;
9961 vmcs12->vm_exit_intr_info = exit_intr_info;
9962 if ((vmcs12->vm_exit_intr_info &
9963 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
9964 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK))
9965 vmcs12->vm_exit_intr_error_code =
9966 vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
9967 vmcs12->idt_vectoring_info_field = 0;
9968 vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
9969 vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
9971 if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
9972 /* vm_entry_intr_info_field is cleared on exit. Emulate this
9973 * instead of reading the real value. */
9974 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
9977 * Transfer the event that L0 or L1 may wanted to inject into
9978 * L2 to IDT_VECTORING_INFO_FIELD.
9980 vmcs12_save_pending_event(vcpu, vmcs12);
9984 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
9985 * preserved above and would only end up incorrectly in L1.
9987 vcpu->arch.nmi_injected = false;
9988 kvm_clear_exception_queue(vcpu);
9989 kvm_clear_interrupt_queue(vcpu);
9993 * A part of what we need to when the nested L2 guest exits and we want to
9994 * run its L1 parent, is to reset L1's guest state to the host state specified
9996 * This function is to be called not only on normal nested exit, but also on
9997 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
9998 * Failures During or After Loading Guest State").
9999 * This function should be called when the active VMCS is L1's (vmcs01).
10001 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
10002 struct vmcs12 *vmcs12)
10004 struct kvm_segment seg;
10006 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
10007 vcpu->arch.efer = vmcs12->host_ia32_efer;
10008 else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
10009 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10011 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10012 vmx_set_efer(vcpu, vcpu->arch.efer);
10014 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
10015 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
10016 vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
10018 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
10019 * actually changed, because it depends on the current state of
10020 * fpu_active (which may have changed).
10021 * Note that vmx_set_cr0 refers to efer set above.
10023 vmx_set_cr0(vcpu, vmcs12->host_cr0);
10025 * If we did fpu_activate()/fpu_deactivate() during L2's run, we need
10026 * to apply the same changes to L1's vmcs. We just set cr0 correctly,
10027 * but we also need to update cr0_guest_host_mask and exception_bitmap.
10029 update_exception_bitmap(vcpu);
10030 vcpu->arch.cr0_guest_owned_bits = (vcpu->fpu_active ? X86_CR0_TS : 0);
10031 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
10034 * Note that CR4_GUEST_HOST_MASK is already set in the original vmcs01
10035 * (KVM doesn't change it)- no reason to call set_cr4_guest_host_mask();
10037 vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
10038 kvm_set_cr4(vcpu, vmcs12->host_cr4);
10040 nested_ept_uninit_mmu_context(vcpu);
10042 kvm_set_cr3(vcpu, vmcs12->host_cr3);
10043 kvm_mmu_reset_context(vcpu);
10046 vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
10050 * Trivially support vpid by letting L2s share their parent
10051 * L1's vpid. TODO: move to a more elaborate solution, giving
10052 * each L2 its own vpid and exposing the vpid feature to L1.
10054 vmx_flush_tlb(vcpu);
10058 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
10059 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
10060 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
10061 vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
10062 vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
10064 /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1. */
10065 if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
10066 vmcs_write64(GUEST_BNDCFGS, 0);
10068 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
10069 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
10070 vcpu->arch.pat = vmcs12->host_ia32_pat;
10072 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
10073 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
10074 vmcs12->host_ia32_perf_global_ctrl);
10076 /* Set L1 segment info according to Intel SDM
10077 27.5.2 Loading Host Segment and Descriptor-Table Registers */
10078 seg = (struct kvm_segment) {
10080 .limit = 0xFFFFFFFF,
10081 .selector = vmcs12->host_cs_selector,
10087 if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
10091 vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
10092 seg = (struct kvm_segment) {
10094 .limit = 0xFFFFFFFF,
10101 seg.selector = vmcs12->host_ds_selector;
10102 vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
10103 seg.selector = vmcs12->host_es_selector;
10104 vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
10105 seg.selector = vmcs12->host_ss_selector;
10106 vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
10107 seg.selector = vmcs12->host_fs_selector;
10108 seg.base = vmcs12->host_fs_base;
10109 vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
10110 seg.selector = vmcs12->host_gs_selector;
10111 seg.base = vmcs12->host_gs_base;
10112 vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
10113 seg = (struct kvm_segment) {
10114 .base = vmcs12->host_tr_base,
10116 .selector = vmcs12->host_tr_selector,
10120 vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
10122 kvm_set_dr(vcpu, 7, 0x400);
10123 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
10125 if (cpu_has_vmx_msr_bitmap())
10126 vmx_set_msr_bitmap(vcpu);
10128 if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
10129 vmcs12->vm_exit_msr_load_count))
10130 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
10134 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
10135 * and modify vmcs12 to make it see what it would expect to see there if
10136 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
10138 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
10139 u32 exit_intr_info,
10140 unsigned long exit_qualification)
10142 struct vcpu_vmx *vmx = to_vmx(vcpu);
10143 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
10145 /* trying to cancel vmlaunch/vmresume is a bug */
10146 WARN_ON_ONCE(vmx->nested.nested_run_pending);
10148 leave_guest_mode(vcpu);
10149 prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
10150 exit_qualification);
10152 if (nested_vmx_store_msr(vcpu, vmcs12->vm_exit_msr_store_addr,
10153 vmcs12->vm_exit_msr_store_count))
10154 nested_vmx_abort(vcpu, VMX_ABORT_SAVE_GUEST_MSR_FAIL);
10156 vmx_load_vmcs01(vcpu);
10158 if ((exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
10159 && nested_exit_intr_ack_set(vcpu)) {
10160 int irq = kvm_cpu_get_interrupt(vcpu);
10162 vmcs12->vm_exit_intr_info = irq |
10163 INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
10166 trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
10167 vmcs12->exit_qualification,
10168 vmcs12->idt_vectoring_info_field,
10169 vmcs12->vm_exit_intr_info,
10170 vmcs12->vm_exit_intr_error_code,
10173 vm_entry_controls_init(vmx, vmcs_read32(VM_ENTRY_CONTROLS));
10174 vm_exit_controls_init(vmx, vmcs_read32(VM_EXIT_CONTROLS));
10175 vmx_segment_cache_clear(vmx);
10177 /* if no vmcs02 cache requested, remove the one we used */
10178 if (VMCS02_POOL_SIZE == 0)
10179 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
10181 load_vmcs12_host_state(vcpu, vmcs12);
10183 /* Update TSC_OFFSET if TSC was changed while L2 ran */
10184 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
10186 /* This is needed for same reason as it was needed in prepare_vmcs02 */
10189 /* Unpin physical memory we referred to in vmcs02 */
10190 if (vmx->nested.apic_access_page) {
10191 nested_release_page(vmx->nested.apic_access_page);
10192 vmx->nested.apic_access_page = NULL;
10194 if (vmx->nested.virtual_apic_page) {
10195 nested_release_page(vmx->nested.virtual_apic_page);
10196 vmx->nested.virtual_apic_page = NULL;
10198 if (vmx->nested.pi_desc_page) {
10199 kunmap(vmx->nested.pi_desc_page);
10200 nested_release_page(vmx->nested.pi_desc_page);
10201 vmx->nested.pi_desc_page = NULL;
10202 vmx->nested.pi_desc = NULL;
10206 * We are now running in L2, mmu_notifier will force to reload the
10207 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
10209 kvm_vcpu_reload_apic_access_page(vcpu);
10212 * Exiting from L2 to L1, we're now back to L1 which thinks it just
10213 * finished a VMLAUNCH or VMRESUME instruction, so we need to set the
10214 * success or failure flag accordingly.
10216 if (unlikely(vmx->fail)) {
10218 nested_vmx_failValid(vcpu, vmcs_read32(VM_INSTRUCTION_ERROR));
10220 nested_vmx_succeed(vcpu);
10221 if (enable_shadow_vmcs)
10222 vmx->nested.sync_shadow_vmcs = true;
10224 /* in case we halted in L2 */
10225 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10229 * Forcibly leave nested mode in order to be able to reset the VCPU later on.
10231 static void vmx_leave_nested(struct kvm_vcpu *vcpu)
10233 if (is_guest_mode(vcpu))
10234 nested_vmx_vmexit(vcpu, -1, 0, 0);
10235 free_nested(to_vmx(vcpu));
10239 * L1's failure to enter L2 is a subset of a normal exit, as explained in
10240 * 23.7 "VM-entry failures during or after loading guest state" (this also
10241 * lists the acceptable exit-reason and exit-qualification parameters).
10242 * It should only be called before L2 actually succeeded to run, and when
10243 * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
10245 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
10246 struct vmcs12 *vmcs12,
10247 u32 reason, unsigned long qualification)
10249 load_vmcs12_host_state(vcpu, vmcs12);
10250 vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
10251 vmcs12->exit_qualification = qualification;
10252 nested_vmx_succeed(vcpu);
10253 if (enable_shadow_vmcs)
10254 to_vmx(vcpu)->nested.sync_shadow_vmcs = true;
10257 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
10258 struct x86_instruction_info *info,
10259 enum x86_intercept_stage stage)
10261 return X86EMUL_CONTINUE;
10264 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
10267 shrink_ple_window(vcpu);
10270 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
10271 struct kvm_memory_slot *slot)
10273 kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
10274 kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
10277 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
10278 struct kvm_memory_slot *slot)
10280 kvm_mmu_slot_set_dirty(kvm, slot);
10283 static void vmx_flush_log_dirty(struct kvm *kvm)
10285 kvm_flush_pml_buffers(kvm);
10288 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
10289 struct kvm_memory_slot *memslot,
10290 gfn_t offset, unsigned long mask)
10292 kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
10295 static struct kvm_x86_ops vmx_x86_ops = {
10296 .cpu_has_kvm_support = cpu_has_kvm_support,
10297 .disabled_by_bios = vmx_disabled_by_bios,
10298 .hardware_setup = hardware_setup,
10299 .hardware_unsetup = hardware_unsetup,
10300 .check_processor_compatibility = vmx_check_processor_compat,
10301 .hardware_enable = hardware_enable,
10302 .hardware_disable = hardware_disable,
10303 .cpu_has_accelerated_tpr = report_flexpriority,
10305 .vcpu_create = vmx_create_vcpu,
10306 .vcpu_free = vmx_free_vcpu,
10307 .vcpu_reset = vmx_vcpu_reset,
10309 .prepare_guest_switch = vmx_save_host_state,
10310 .vcpu_load = vmx_vcpu_load,
10311 .vcpu_put = vmx_vcpu_put,
10313 .update_db_bp_intercept = update_exception_bitmap,
10314 .get_msr = vmx_get_msr,
10315 .set_msr = vmx_set_msr,
10316 .get_segment_base = vmx_get_segment_base,
10317 .get_segment = vmx_get_segment,
10318 .set_segment = vmx_set_segment,
10319 .get_cpl = vmx_get_cpl,
10320 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
10321 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
10322 .decache_cr3 = vmx_decache_cr3,
10323 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
10324 .set_cr0 = vmx_set_cr0,
10325 .set_cr3 = vmx_set_cr3,
10326 .set_cr4 = vmx_set_cr4,
10327 .set_efer = vmx_set_efer,
10328 .get_idt = vmx_get_idt,
10329 .set_idt = vmx_set_idt,
10330 .get_gdt = vmx_get_gdt,
10331 .set_gdt = vmx_set_gdt,
10332 .get_dr6 = vmx_get_dr6,
10333 .set_dr6 = vmx_set_dr6,
10334 .set_dr7 = vmx_set_dr7,
10335 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
10336 .cache_reg = vmx_cache_reg,
10337 .get_rflags = vmx_get_rflags,
10338 .set_rflags = vmx_set_rflags,
10339 .fpu_deactivate = vmx_fpu_deactivate,
10341 .tlb_flush = vmx_flush_tlb,
10343 .run = vmx_vcpu_run,
10344 .handle_exit = vmx_handle_exit,
10345 .skip_emulated_instruction = skip_emulated_instruction,
10346 .set_interrupt_shadow = vmx_set_interrupt_shadow,
10347 .get_interrupt_shadow = vmx_get_interrupt_shadow,
10348 .patch_hypercall = vmx_patch_hypercall,
10349 .set_irq = vmx_inject_irq,
10350 .set_nmi = vmx_inject_nmi,
10351 .queue_exception = vmx_queue_exception,
10352 .cancel_injection = vmx_cancel_injection,
10353 .interrupt_allowed = vmx_interrupt_allowed,
10354 .nmi_allowed = vmx_nmi_allowed,
10355 .get_nmi_mask = vmx_get_nmi_mask,
10356 .set_nmi_mask = vmx_set_nmi_mask,
10357 .enable_nmi_window = enable_nmi_window,
10358 .enable_irq_window = enable_irq_window,
10359 .update_cr8_intercept = update_cr8_intercept,
10360 .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode,
10361 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
10362 .vm_has_apicv = vmx_vm_has_apicv,
10363 .load_eoi_exitmap = vmx_load_eoi_exitmap,
10364 .hwapic_irr_update = vmx_hwapic_irr_update,
10365 .hwapic_isr_update = vmx_hwapic_isr_update,
10366 .sync_pir_to_irr = vmx_sync_pir_to_irr,
10367 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
10369 .set_tss_addr = vmx_set_tss_addr,
10370 .get_tdp_level = get_ept_level,
10371 .get_mt_mask = vmx_get_mt_mask,
10373 .get_exit_info = vmx_get_exit_info,
10375 .get_lpage_level = vmx_get_lpage_level,
10377 .cpuid_update = vmx_cpuid_update,
10379 .rdtscp_supported = vmx_rdtscp_supported,
10380 .invpcid_supported = vmx_invpcid_supported,
10382 .set_supported_cpuid = vmx_set_supported_cpuid,
10384 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
10386 .set_tsc_khz = vmx_set_tsc_khz,
10387 .read_tsc_offset = vmx_read_tsc_offset,
10388 .write_tsc_offset = vmx_write_tsc_offset,
10389 .adjust_tsc_offset = vmx_adjust_tsc_offset,
10390 .compute_tsc_offset = vmx_compute_tsc_offset,
10391 .read_l1_tsc = vmx_read_l1_tsc,
10393 .set_tdp_cr3 = vmx_set_cr3,
10395 .check_intercept = vmx_check_intercept,
10396 .handle_external_intr = vmx_handle_external_intr,
10397 .mpx_supported = vmx_mpx_supported,
10398 .xsaves_supported = vmx_xsaves_supported,
10400 .check_nested_events = vmx_check_nested_events,
10402 .sched_in = vmx_sched_in,
10404 .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
10405 .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
10406 .flush_log_dirty = vmx_flush_log_dirty,
10407 .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
10410 static int __init vmx_init(void)
10412 int r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
10413 __alignof__(struct vcpu_vmx), THIS_MODULE);
10417 #ifdef CONFIG_KEXEC
10418 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
10419 crash_vmclear_local_loaded_vmcss);
10425 static void __exit vmx_exit(void)
10427 #ifdef CONFIG_KEXEC
10428 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
10435 module_init(vmx_init)
10436 module_exit(vmx_exit)