579a0b51696ac560b1768ae445bb8f2084172f8c
[linux-3.10.git] / arch / x86 / kvm / vmx.c
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * This module enables machines with Intel VT-x extensions to run virtual
5  * machines without emulation or binary translation.
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
9  *
10  * Authors:
11  *   Avi Kivity   <avi@qumranet.com>
12  *   Yaniv Kamay  <yaniv@qumranet.com>
13  *
14  * This work is licensed under the terms of the GNU GPL, version 2.  See
15  * the COPYING file in the top-level directory.
16  *
17  */
18
19 #include "irq.h"
20 #include "mmu.h"
21
22 #include <linux/kvm_host.h>
23 #include <linux/module.h>
24 #include <linux/kernel.h>
25 #include <linux/mm.h>
26 #include <linux/highmem.h>
27 #include <linux/sched.h>
28 #include <linux/moduleparam.h>
29 #include <linux/ftrace_event.h>
30 #include <linux/slab.h>
31 #include <linux/tboot.h>
32 #include "kvm_cache_regs.h"
33 #include "x86.h"
34
35 #include <asm/io.h>
36 #include <asm/desc.h>
37 #include <asm/vmx.h>
38 #include <asm/virtext.h>
39 #include <asm/mce.h>
40 #include <asm/i387.h>
41 #include <asm/xcr.h>
42 #include <asm/perf_event.h>
43
44 #include "trace.h"
45
46 #define __ex(x) __kvm_handle_fault_on_reboot(x)
47 #define __ex_clear(x, reg) \
48         ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
49
50 MODULE_AUTHOR("Qumranet");
51 MODULE_LICENSE("GPL");
52
53 static int __read_mostly enable_vpid = 1;
54 module_param_named(vpid, enable_vpid, bool, 0444);
55
56 static int __read_mostly flexpriority_enabled = 1;
57 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
58
59 static int __read_mostly enable_ept = 1;
60 module_param_named(ept, enable_ept, bool, S_IRUGO);
61
62 static int __read_mostly enable_unrestricted_guest = 1;
63 module_param_named(unrestricted_guest,
64                         enable_unrestricted_guest, bool, S_IRUGO);
65
66 static int __read_mostly emulate_invalid_guest_state = 0;
67 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
68
69 static int __read_mostly vmm_exclusive = 1;
70 module_param(vmm_exclusive, bool, S_IRUGO);
71
72 static int __read_mostly yield_on_hlt = 1;
73 module_param(yield_on_hlt, bool, S_IRUGO);
74
75 static int __read_mostly fasteoi = 1;
76 module_param(fasteoi, bool, S_IRUGO);
77
78 /*
79  * If nested=1, nested virtualization is supported, i.e., guests may use
80  * VMX and be a hypervisor for its own guests. If nested=0, guests may not
81  * use VMX instructions.
82  */
83 static int __read_mostly nested = 0;
84 module_param(nested, bool, S_IRUGO);
85
86 #define KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST                           \
87         (X86_CR0_WP | X86_CR0_NE | X86_CR0_NW | X86_CR0_CD)
88 #define KVM_GUEST_CR0_MASK                                              \
89         (KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
90 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST                         \
91         (X86_CR0_WP | X86_CR0_NE)
92 #define KVM_VM_CR0_ALWAYS_ON                                            \
93         (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
94 #define KVM_CR4_GUEST_OWNED_BITS                                      \
95         (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR      \
96          | X86_CR4_OSXMMEXCPT)
97
98 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
99 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
100
101 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
102
103 /*
104  * These 2 parameters are used to config the controls for Pause-Loop Exiting:
105  * ple_gap:    upper bound on the amount of time between two successive
106  *             executions of PAUSE in a loop. Also indicate if ple enabled.
107  *             According to test, this time is usually smaller than 128 cycles.
108  * ple_window: upper bound on the amount of time a guest is allowed to execute
109  *             in a PAUSE loop. Tests indicate that most spinlocks are held for
110  *             less than 2^12 cycles
111  * Time is measured based on a counter that runs at the same rate as the TSC,
112  * refer SDM volume 3b section 21.6.13 & 22.1.3.
113  */
114 #define KVM_VMX_DEFAULT_PLE_GAP    128
115 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
116 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
117 module_param(ple_gap, int, S_IRUGO);
118
119 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
120 module_param(ple_window, int, S_IRUGO);
121
122 #define NR_AUTOLOAD_MSRS 8
123 #define VMCS02_POOL_SIZE 1
124
125 struct vmcs {
126         u32 revision_id;
127         u32 abort;
128         char data[0];
129 };
130
131 /*
132  * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
133  * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
134  * loaded on this CPU (so we can clear them if the CPU goes down).
135  */
136 struct loaded_vmcs {
137         struct vmcs *vmcs;
138         int cpu;
139         int launched;
140         struct list_head loaded_vmcss_on_cpu_link;
141 };
142
143 struct shared_msr_entry {
144         unsigned index;
145         u64 data;
146         u64 mask;
147 };
148
149 /*
150  * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
151  * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
152  * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
153  * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
154  * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
155  * More than one of these structures may exist, if L1 runs multiple L2 guests.
156  * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
157  * underlying hardware which will be used to run L2.
158  * This structure is packed to ensure that its layout is identical across
159  * machines (necessary for live migration).
160  * If there are changes in this struct, VMCS12_REVISION must be changed.
161  */
162 typedef u64 natural_width;
163 struct __packed vmcs12 {
164         /* According to the Intel spec, a VMCS region must start with the
165          * following two fields. Then follow implementation-specific data.
166          */
167         u32 revision_id;
168         u32 abort;
169
170         u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
171         u32 padding[7]; /* room for future expansion */
172
173         u64 io_bitmap_a;
174         u64 io_bitmap_b;
175         u64 msr_bitmap;
176         u64 vm_exit_msr_store_addr;
177         u64 vm_exit_msr_load_addr;
178         u64 vm_entry_msr_load_addr;
179         u64 tsc_offset;
180         u64 virtual_apic_page_addr;
181         u64 apic_access_addr;
182         u64 ept_pointer;
183         u64 guest_physical_address;
184         u64 vmcs_link_pointer;
185         u64 guest_ia32_debugctl;
186         u64 guest_ia32_pat;
187         u64 guest_ia32_efer;
188         u64 guest_ia32_perf_global_ctrl;
189         u64 guest_pdptr0;
190         u64 guest_pdptr1;
191         u64 guest_pdptr2;
192         u64 guest_pdptr3;
193         u64 host_ia32_pat;
194         u64 host_ia32_efer;
195         u64 host_ia32_perf_global_ctrl;
196         u64 padding64[8]; /* room for future expansion */
197         /*
198          * To allow migration of L1 (complete with its L2 guests) between
199          * machines of different natural widths (32 or 64 bit), we cannot have
200          * unsigned long fields with no explict size. We use u64 (aliased
201          * natural_width) instead. Luckily, x86 is little-endian.
202          */
203         natural_width cr0_guest_host_mask;
204         natural_width cr4_guest_host_mask;
205         natural_width cr0_read_shadow;
206         natural_width cr4_read_shadow;
207         natural_width cr3_target_value0;
208         natural_width cr3_target_value1;
209         natural_width cr3_target_value2;
210         natural_width cr3_target_value3;
211         natural_width exit_qualification;
212         natural_width guest_linear_address;
213         natural_width guest_cr0;
214         natural_width guest_cr3;
215         natural_width guest_cr4;
216         natural_width guest_es_base;
217         natural_width guest_cs_base;
218         natural_width guest_ss_base;
219         natural_width guest_ds_base;
220         natural_width guest_fs_base;
221         natural_width guest_gs_base;
222         natural_width guest_ldtr_base;
223         natural_width guest_tr_base;
224         natural_width guest_gdtr_base;
225         natural_width guest_idtr_base;
226         natural_width guest_dr7;
227         natural_width guest_rsp;
228         natural_width guest_rip;
229         natural_width guest_rflags;
230         natural_width guest_pending_dbg_exceptions;
231         natural_width guest_sysenter_esp;
232         natural_width guest_sysenter_eip;
233         natural_width host_cr0;
234         natural_width host_cr3;
235         natural_width host_cr4;
236         natural_width host_fs_base;
237         natural_width host_gs_base;
238         natural_width host_tr_base;
239         natural_width host_gdtr_base;
240         natural_width host_idtr_base;
241         natural_width host_ia32_sysenter_esp;
242         natural_width host_ia32_sysenter_eip;
243         natural_width host_rsp;
244         natural_width host_rip;
245         natural_width paddingl[8]; /* room for future expansion */
246         u32 pin_based_vm_exec_control;
247         u32 cpu_based_vm_exec_control;
248         u32 exception_bitmap;
249         u32 page_fault_error_code_mask;
250         u32 page_fault_error_code_match;
251         u32 cr3_target_count;
252         u32 vm_exit_controls;
253         u32 vm_exit_msr_store_count;
254         u32 vm_exit_msr_load_count;
255         u32 vm_entry_controls;
256         u32 vm_entry_msr_load_count;
257         u32 vm_entry_intr_info_field;
258         u32 vm_entry_exception_error_code;
259         u32 vm_entry_instruction_len;
260         u32 tpr_threshold;
261         u32 secondary_vm_exec_control;
262         u32 vm_instruction_error;
263         u32 vm_exit_reason;
264         u32 vm_exit_intr_info;
265         u32 vm_exit_intr_error_code;
266         u32 idt_vectoring_info_field;
267         u32 idt_vectoring_error_code;
268         u32 vm_exit_instruction_len;
269         u32 vmx_instruction_info;
270         u32 guest_es_limit;
271         u32 guest_cs_limit;
272         u32 guest_ss_limit;
273         u32 guest_ds_limit;
274         u32 guest_fs_limit;
275         u32 guest_gs_limit;
276         u32 guest_ldtr_limit;
277         u32 guest_tr_limit;
278         u32 guest_gdtr_limit;
279         u32 guest_idtr_limit;
280         u32 guest_es_ar_bytes;
281         u32 guest_cs_ar_bytes;
282         u32 guest_ss_ar_bytes;
283         u32 guest_ds_ar_bytes;
284         u32 guest_fs_ar_bytes;
285         u32 guest_gs_ar_bytes;
286         u32 guest_ldtr_ar_bytes;
287         u32 guest_tr_ar_bytes;
288         u32 guest_interruptibility_info;
289         u32 guest_activity_state;
290         u32 guest_sysenter_cs;
291         u32 host_ia32_sysenter_cs;
292         u32 padding32[8]; /* room for future expansion */
293         u16 virtual_processor_id;
294         u16 guest_es_selector;
295         u16 guest_cs_selector;
296         u16 guest_ss_selector;
297         u16 guest_ds_selector;
298         u16 guest_fs_selector;
299         u16 guest_gs_selector;
300         u16 guest_ldtr_selector;
301         u16 guest_tr_selector;
302         u16 host_es_selector;
303         u16 host_cs_selector;
304         u16 host_ss_selector;
305         u16 host_ds_selector;
306         u16 host_fs_selector;
307         u16 host_gs_selector;
308         u16 host_tr_selector;
309 };
310
311 /*
312  * VMCS12_REVISION is an arbitrary id that should be changed if the content or
313  * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
314  * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
315  */
316 #define VMCS12_REVISION 0x11e57ed0
317
318 /*
319  * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
320  * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
321  * current implementation, 4K are reserved to avoid future complications.
322  */
323 #define VMCS12_SIZE 0x1000
324
325 /* Used to remember the last vmcs02 used for some recently used vmcs12s */
326 struct vmcs02_list {
327         struct list_head list;
328         gpa_t vmptr;
329         struct loaded_vmcs vmcs02;
330 };
331
332 /*
333  * The nested_vmx structure is part of vcpu_vmx, and holds information we need
334  * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
335  */
336 struct nested_vmx {
337         /* Has the level1 guest done vmxon? */
338         bool vmxon;
339
340         /* The guest-physical address of the current VMCS L1 keeps for L2 */
341         gpa_t current_vmptr;
342         /* The host-usable pointer to the above */
343         struct page *current_vmcs12_page;
344         struct vmcs12 *current_vmcs12;
345
346         /* vmcs02_list cache of VMCSs recently used to run L2 guests */
347         struct list_head vmcs02_pool;
348         int vmcs02_num;
349         u64 vmcs01_tsc_offset;
350         /* L2 must run next, and mustn't decide to exit to L1. */
351         bool nested_run_pending;
352         /*
353          * Guest pages referred to in vmcs02 with host-physical pointers, so
354          * we must keep them pinned while L2 runs.
355          */
356         struct page *apic_access_page;
357 };
358
359 struct vcpu_vmx {
360         struct kvm_vcpu       vcpu;
361         unsigned long         host_rsp;
362         u8                    fail;
363         u8                    cpl;
364         bool                  nmi_known_unmasked;
365         u32                   exit_intr_info;
366         u32                   idt_vectoring_info;
367         ulong                 rflags;
368         struct shared_msr_entry *guest_msrs;
369         int                   nmsrs;
370         int                   save_nmsrs;
371 #ifdef CONFIG_X86_64
372         u64                   msr_host_kernel_gs_base;
373         u64                   msr_guest_kernel_gs_base;
374 #endif
375         /*
376          * loaded_vmcs points to the VMCS currently used in this vcpu. For a
377          * non-nested (L1) guest, it always points to vmcs01. For a nested
378          * guest (L2), it points to a different VMCS.
379          */
380         struct loaded_vmcs    vmcs01;
381         struct loaded_vmcs   *loaded_vmcs;
382         bool                  __launched; /* temporary, used in vmx_vcpu_run */
383         struct msr_autoload {
384                 unsigned nr;
385                 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
386                 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
387         } msr_autoload;
388         struct {
389                 int           loaded;
390                 u16           fs_sel, gs_sel, ldt_sel;
391                 int           gs_ldt_reload_needed;
392                 int           fs_reload_needed;
393         } host_state;
394         struct {
395                 int vm86_active;
396                 ulong save_rflags;
397                 struct kvm_save_segment {
398                         u16 selector;
399                         unsigned long base;
400                         u32 limit;
401                         u32 ar;
402                 } tr, es, ds, fs, gs;
403         } rmode;
404         struct {
405                 u32 bitmask; /* 4 bits per segment (1 bit per field) */
406                 struct kvm_save_segment seg[8];
407         } segment_cache;
408         int vpid;
409         bool emulation_required;
410
411         /* Support for vnmi-less CPUs */
412         int soft_vnmi_blocked;
413         ktime_t entry_time;
414         s64 vnmi_blocked_time;
415         u32 exit_reason;
416
417         bool rdtscp_enabled;
418
419         /* Support for a guest hypervisor (nested VMX) */
420         struct nested_vmx nested;
421 };
422
423 enum segment_cache_field {
424         SEG_FIELD_SEL = 0,
425         SEG_FIELD_BASE = 1,
426         SEG_FIELD_LIMIT = 2,
427         SEG_FIELD_AR = 3,
428
429         SEG_FIELD_NR = 4
430 };
431
432 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
433 {
434         return container_of(vcpu, struct vcpu_vmx, vcpu);
435 }
436
437 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
438 #define FIELD(number, name)     [number] = VMCS12_OFFSET(name)
439 #define FIELD64(number, name)   [number] = VMCS12_OFFSET(name), \
440                                 [number##_HIGH] = VMCS12_OFFSET(name)+4
441
442 static unsigned short vmcs_field_to_offset_table[] = {
443         FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
444         FIELD(GUEST_ES_SELECTOR, guest_es_selector),
445         FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
446         FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
447         FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
448         FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
449         FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
450         FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
451         FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
452         FIELD(HOST_ES_SELECTOR, host_es_selector),
453         FIELD(HOST_CS_SELECTOR, host_cs_selector),
454         FIELD(HOST_SS_SELECTOR, host_ss_selector),
455         FIELD(HOST_DS_SELECTOR, host_ds_selector),
456         FIELD(HOST_FS_SELECTOR, host_fs_selector),
457         FIELD(HOST_GS_SELECTOR, host_gs_selector),
458         FIELD(HOST_TR_SELECTOR, host_tr_selector),
459         FIELD64(IO_BITMAP_A, io_bitmap_a),
460         FIELD64(IO_BITMAP_B, io_bitmap_b),
461         FIELD64(MSR_BITMAP, msr_bitmap),
462         FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
463         FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
464         FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
465         FIELD64(TSC_OFFSET, tsc_offset),
466         FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
467         FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
468         FIELD64(EPT_POINTER, ept_pointer),
469         FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
470         FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
471         FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
472         FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
473         FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
474         FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
475         FIELD64(GUEST_PDPTR0, guest_pdptr0),
476         FIELD64(GUEST_PDPTR1, guest_pdptr1),
477         FIELD64(GUEST_PDPTR2, guest_pdptr2),
478         FIELD64(GUEST_PDPTR3, guest_pdptr3),
479         FIELD64(HOST_IA32_PAT, host_ia32_pat),
480         FIELD64(HOST_IA32_EFER, host_ia32_efer),
481         FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
482         FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
483         FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
484         FIELD(EXCEPTION_BITMAP, exception_bitmap),
485         FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
486         FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
487         FIELD(CR3_TARGET_COUNT, cr3_target_count),
488         FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
489         FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
490         FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
491         FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
492         FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
493         FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
494         FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
495         FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
496         FIELD(TPR_THRESHOLD, tpr_threshold),
497         FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
498         FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
499         FIELD(VM_EXIT_REASON, vm_exit_reason),
500         FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
501         FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
502         FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
503         FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
504         FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
505         FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
506         FIELD(GUEST_ES_LIMIT, guest_es_limit),
507         FIELD(GUEST_CS_LIMIT, guest_cs_limit),
508         FIELD(GUEST_SS_LIMIT, guest_ss_limit),
509         FIELD(GUEST_DS_LIMIT, guest_ds_limit),
510         FIELD(GUEST_FS_LIMIT, guest_fs_limit),
511         FIELD(GUEST_GS_LIMIT, guest_gs_limit),
512         FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
513         FIELD(GUEST_TR_LIMIT, guest_tr_limit),
514         FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
515         FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
516         FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
517         FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
518         FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
519         FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
520         FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
521         FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
522         FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
523         FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
524         FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
525         FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
526         FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
527         FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
528         FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
529         FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
530         FIELD(CR0_READ_SHADOW, cr0_read_shadow),
531         FIELD(CR4_READ_SHADOW, cr4_read_shadow),
532         FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
533         FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
534         FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
535         FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
536         FIELD(EXIT_QUALIFICATION, exit_qualification),
537         FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
538         FIELD(GUEST_CR0, guest_cr0),
539         FIELD(GUEST_CR3, guest_cr3),
540         FIELD(GUEST_CR4, guest_cr4),
541         FIELD(GUEST_ES_BASE, guest_es_base),
542         FIELD(GUEST_CS_BASE, guest_cs_base),
543         FIELD(GUEST_SS_BASE, guest_ss_base),
544         FIELD(GUEST_DS_BASE, guest_ds_base),
545         FIELD(GUEST_FS_BASE, guest_fs_base),
546         FIELD(GUEST_GS_BASE, guest_gs_base),
547         FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
548         FIELD(GUEST_TR_BASE, guest_tr_base),
549         FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
550         FIELD(GUEST_IDTR_BASE, guest_idtr_base),
551         FIELD(GUEST_DR7, guest_dr7),
552         FIELD(GUEST_RSP, guest_rsp),
553         FIELD(GUEST_RIP, guest_rip),
554         FIELD(GUEST_RFLAGS, guest_rflags),
555         FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
556         FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
557         FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
558         FIELD(HOST_CR0, host_cr0),
559         FIELD(HOST_CR3, host_cr3),
560         FIELD(HOST_CR4, host_cr4),
561         FIELD(HOST_FS_BASE, host_fs_base),
562         FIELD(HOST_GS_BASE, host_gs_base),
563         FIELD(HOST_TR_BASE, host_tr_base),
564         FIELD(HOST_GDTR_BASE, host_gdtr_base),
565         FIELD(HOST_IDTR_BASE, host_idtr_base),
566         FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
567         FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
568         FIELD(HOST_RSP, host_rsp),
569         FIELD(HOST_RIP, host_rip),
570 };
571 static const int max_vmcs_field = ARRAY_SIZE(vmcs_field_to_offset_table);
572
573 static inline short vmcs_field_to_offset(unsigned long field)
574 {
575         if (field >= max_vmcs_field || vmcs_field_to_offset_table[field] == 0)
576                 return -1;
577         return vmcs_field_to_offset_table[field];
578 }
579
580 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
581 {
582         return to_vmx(vcpu)->nested.current_vmcs12;
583 }
584
585 static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
586 {
587         struct page *page = gfn_to_page(vcpu->kvm, addr >> PAGE_SHIFT);
588         if (is_error_page(page)) {
589                 kvm_release_page_clean(page);
590                 return NULL;
591         }
592         return page;
593 }
594
595 static void nested_release_page(struct page *page)
596 {
597         kvm_release_page_dirty(page);
598 }
599
600 static void nested_release_page_clean(struct page *page)
601 {
602         kvm_release_page_clean(page);
603 }
604
605 static u64 construct_eptp(unsigned long root_hpa);
606 static void kvm_cpu_vmxon(u64 addr);
607 static void kvm_cpu_vmxoff(void);
608 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3);
609 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
610
611 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
612 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
613 /*
614  * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
615  * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
616  */
617 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
618 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
619
620 static unsigned long *vmx_io_bitmap_a;
621 static unsigned long *vmx_io_bitmap_b;
622 static unsigned long *vmx_msr_bitmap_legacy;
623 static unsigned long *vmx_msr_bitmap_longmode;
624
625 static bool cpu_has_load_ia32_efer;
626 static bool cpu_has_load_perf_global_ctrl;
627
628 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
629 static DEFINE_SPINLOCK(vmx_vpid_lock);
630
631 static struct vmcs_config {
632         int size;
633         int order;
634         u32 revision_id;
635         u32 pin_based_exec_ctrl;
636         u32 cpu_based_exec_ctrl;
637         u32 cpu_based_2nd_exec_ctrl;
638         u32 vmexit_ctrl;
639         u32 vmentry_ctrl;
640 } vmcs_config;
641
642 static struct vmx_capability {
643         u32 ept;
644         u32 vpid;
645 } vmx_capability;
646
647 #define VMX_SEGMENT_FIELD(seg)                                  \
648         [VCPU_SREG_##seg] = {                                   \
649                 .selector = GUEST_##seg##_SELECTOR,             \
650                 .base = GUEST_##seg##_BASE,                     \
651                 .limit = GUEST_##seg##_LIMIT,                   \
652                 .ar_bytes = GUEST_##seg##_AR_BYTES,             \
653         }
654
655 static struct kvm_vmx_segment_field {
656         unsigned selector;
657         unsigned base;
658         unsigned limit;
659         unsigned ar_bytes;
660 } kvm_vmx_segment_fields[] = {
661         VMX_SEGMENT_FIELD(CS),
662         VMX_SEGMENT_FIELD(DS),
663         VMX_SEGMENT_FIELD(ES),
664         VMX_SEGMENT_FIELD(FS),
665         VMX_SEGMENT_FIELD(GS),
666         VMX_SEGMENT_FIELD(SS),
667         VMX_SEGMENT_FIELD(TR),
668         VMX_SEGMENT_FIELD(LDTR),
669 };
670
671 static u64 host_efer;
672
673 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
674
675 /*
676  * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
677  * away by decrementing the array size.
678  */
679 static const u32 vmx_msr_index[] = {
680 #ifdef CONFIG_X86_64
681         MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
682 #endif
683         MSR_EFER, MSR_TSC_AUX, MSR_STAR,
684 };
685 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
686
687 static inline bool is_page_fault(u32 intr_info)
688 {
689         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
690                              INTR_INFO_VALID_MASK)) ==
691                 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
692 }
693
694 static inline bool is_no_device(u32 intr_info)
695 {
696         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
697                              INTR_INFO_VALID_MASK)) ==
698                 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
699 }
700
701 static inline bool is_invalid_opcode(u32 intr_info)
702 {
703         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
704                              INTR_INFO_VALID_MASK)) ==
705                 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
706 }
707
708 static inline bool is_external_interrupt(u32 intr_info)
709 {
710         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
711                 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
712 }
713
714 static inline bool is_machine_check(u32 intr_info)
715 {
716         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
717                              INTR_INFO_VALID_MASK)) ==
718                 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
719 }
720
721 static inline bool cpu_has_vmx_msr_bitmap(void)
722 {
723         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
724 }
725
726 static inline bool cpu_has_vmx_tpr_shadow(void)
727 {
728         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
729 }
730
731 static inline bool vm_need_tpr_shadow(struct kvm *kvm)
732 {
733         return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
734 }
735
736 static inline bool cpu_has_secondary_exec_ctrls(void)
737 {
738         return vmcs_config.cpu_based_exec_ctrl &
739                 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
740 }
741
742 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
743 {
744         return vmcs_config.cpu_based_2nd_exec_ctrl &
745                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
746 }
747
748 static inline bool cpu_has_vmx_flexpriority(void)
749 {
750         return cpu_has_vmx_tpr_shadow() &&
751                 cpu_has_vmx_virtualize_apic_accesses();
752 }
753
754 static inline bool cpu_has_vmx_ept_execute_only(void)
755 {
756         return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
757 }
758
759 static inline bool cpu_has_vmx_eptp_uncacheable(void)
760 {
761         return vmx_capability.ept & VMX_EPTP_UC_BIT;
762 }
763
764 static inline bool cpu_has_vmx_eptp_writeback(void)
765 {
766         return vmx_capability.ept & VMX_EPTP_WB_BIT;
767 }
768
769 static inline bool cpu_has_vmx_ept_2m_page(void)
770 {
771         return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
772 }
773
774 static inline bool cpu_has_vmx_ept_1g_page(void)
775 {
776         return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
777 }
778
779 static inline bool cpu_has_vmx_ept_4levels(void)
780 {
781         return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
782 }
783
784 static inline bool cpu_has_vmx_invept_individual_addr(void)
785 {
786         return vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT;
787 }
788
789 static inline bool cpu_has_vmx_invept_context(void)
790 {
791         return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
792 }
793
794 static inline bool cpu_has_vmx_invept_global(void)
795 {
796         return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
797 }
798
799 static inline bool cpu_has_vmx_invvpid_single(void)
800 {
801         return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
802 }
803
804 static inline bool cpu_has_vmx_invvpid_global(void)
805 {
806         return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
807 }
808
809 static inline bool cpu_has_vmx_ept(void)
810 {
811         return vmcs_config.cpu_based_2nd_exec_ctrl &
812                 SECONDARY_EXEC_ENABLE_EPT;
813 }
814
815 static inline bool cpu_has_vmx_unrestricted_guest(void)
816 {
817         return vmcs_config.cpu_based_2nd_exec_ctrl &
818                 SECONDARY_EXEC_UNRESTRICTED_GUEST;
819 }
820
821 static inline bool cpu_has_vmx_ple(void)
822 {
823         return vmcs_config.cpu_based_2nd_exec_ctrl &
824                 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
825 }
826
827 static inline bool vm_need_virtualize_apic_accesses(struct kvm *kvm)
828 {
829         return flexpriority_enabled && irqchip_in_kernel(kvm);
830 }
831
832 static inline bool cpu_has_vmx_vpid(void)
833 {
834         return vmcs_config.cpu_based_2nd_exec_ctrl &
835                 SECONDARY_EXEC_ENABLE_VPID;
836 }
837
838 static inline bool cpu_has_vmx_rdtscp(void)
839 {
840         return vmcs_config.cpu_based_2nd_exec_ctrl &
841                 SECONDARY_EXEC_RDTSCP;
842 }
843
844 static inline bool cpu_has_virtual_nmis(void)
845 {
846         return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
847 }
848
849 static inline bool cpu_has_vmx_wbinvd_exit(void)
850 {
851         return vmcs_config.cpu_based_2nd_exec_ctrl &
852                 SECONDARY_EXEC_WBINVD_EXITING;
853 }
854
855 static inline bool report_flexpriority(void)
856 {
857         return flexpriority_enabled;
858 }
859
860 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
861 {
862         return vmcs12->cpu_based_vm_exec_control & bit;
863 }
864
865 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
866 {
867         return (vmcs12->cpu_based_vm_exec_control &
868                         CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
869                 (vmcs12->secondary_vm_exec_control & bit);
870 }
871
872 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12,
873         struct kvm_vcpu *vcpu)
874 {
875         return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
876 }
877
878 static inline bool is_exception(u32 intr_info)
879 {
880         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
881                 == (INTR_TYPE_HARD_EXCEPTION | INTR_INFO_VALID_MASK);
882 }
883
884 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu);
885 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
886                         struct vmcs12 *vmcs12,
887                         u32 reason, unsigned long qualification);
888
889 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
890 {
891         int i;
892
893         for (i = 0; i < vmx->nmsrs; ++i)
894                 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
895                         return i;
896         return -1;
897 }
898
899 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
900 {
901     struct {
902         u64 vpid : 16;
903         u64 rsvd : 48;
904         u64 gva;
905     } operand = { vpid, 0, gva };
906
907     asm volatile (__ex(ASM_VMX_INVVPID)
908                   /* CF==1 or ZF==1 --> rc = -1 */
909                   "; ja 1f ; ud2 ; 1:"
910                   : : "a"(&operand), "c"(ext) : "cc", "memory");
911 }
912
913 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
914 {
915         struct {
916                 u64 eptp, gpa;
917         } operand = {eptp, gpa};
918
919         asm volatile (__ex(ASM_VMX_INVEPT)
920                         /* CF==1 or ZF==1 --> rc = -1 */
921                         "; ja 1f ; ud2 ; 1:\n"
922                         : : "a" (&operand), "c" (ext) : "cc", "memory");
923 }
924
925 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
926 {
927         int i;
928
929         i = __find_msr_index(vmx, msr);
930         if (i >= 0)
931                 return &vmx->guest_msrs[i];
932         return NULL;
933 }
934
935 static void vmcs_clear(struct vmcs *vmcs)
936 {
937         u64 phys_addr = __pa(vmcs);
938         u8 error;
939
940         asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
941                       : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
942                       : "cc", "memory");
943         if (error)
944                 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
945                        vmcs, phys_addr);
946 }
947
948 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
949 {
950         vmcs_clear(loaded_vmcs->vmcs);
951         loaded_vmcs->cpu = -1;
952         loaded_vmcs->launched = 0;
953 }
954
955 static void vmcs_load(struct vmcs *vmcs)
956 {
957         u64 phys_addr = __pa(vmcs);
958         u8 error;
959
960         asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
961                         : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
962                         : "cc", "memory");
963         if (error)
964                 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
965                        vmcs, phys_addr);
966 }
967
968 static void __loaded_vmcs_clear(void *arg)
969 {
970         struct loaded_vmcs *loaded_vmcs = arg;
971         int cpu = raw_smp_processor_id();
972
973         if (loaded_vmcs->cpu != cpu)
974                 return; /* vcpu migration can race with cpu offline */
975         if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
976                 per_cpu(current_vmcs, cpu) = NULL;
977         list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
978         loaded_vmcs_init(loaded_vmcs);
979 }
980
981 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
982 {
983         if (loaded_vmcs->cpu != -1)
984                 smp_call_function_single(
985                         loaded_vmcs->cpu, __loaded_vmcs_clear, loaded_vmcs, 1);
986 }
987
988 static inline void vpid_sync_vcpu_single(struct vcpu_vmx *vmx)
989 {
990         if (vmx->vpid == 0)
991                 return;
992
993         if (cpu_has_vmx_invvpid_single())
994                 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
995 }
996
997 static inline void vpid_sync_vcpu_global(void)
998 {
999         if (cpu_has_vmx_invvpid_global())
1000                 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1001 }
1002
1003 static inline void vpid_sync_context(struct vcpu_vmx *vmx)
1004 {
1005         if (cpu_has_vmx_invvpid_single())
1006                 vpid_sync_vcpu_single(vmx);
1007         else
1008                 vpid_sync_vcpu_global();
1009 }
1010
1011 static inline void ept_sync_global(void)
1012 {
1013         if (cpu_has_vmx_invept_global())
1014                 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1015 }
1016
1017 static inline void ept_sync_context(u64 eptp)
1018 {
1019         if (enable_ept) {
1020                 if (cpu_has_vmx_invept_context())
1021                         __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1022                 else
1023                         ept_sync_global();
1024         }
1025 }
1026
1027 static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
1028 {
1029         if (enable_ept) {
1030                 if (cpu_has_vmx_invept_individual_addr())
1031                         __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
1032                                         eptp, gpa);
1033                 else
1034                         ept_sync_context(eptp);
1035         }
1036 }
1037
1038 static __always_inline unsigned long vmcs_readl(unsigned long field)
1039 {
1040         unsigned long value;
1041
1042         asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1043                       : "=a"(value) : "d"(field) : "cc");
1044         return value;
1045 }
1046
1047 static __always_inline u16 vmcs_read16(unsigned long field)
1048 {
1049         return vmcs_readl(field);
1050 }
1051
1052 static __always_inline u32 vmcs_read32(unsigned long field)
1053 {
1054         return vmcs_readl(field);
1055 }
1056
1057 static __always_inline u64 vmcs_read64(unsigned long field)
1058 {
1059 #ifdef CONFIG_X86_64
1060         return vmcs_readl(field);
1061 #else
1062         return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
1063 #endif
1064 }
1065
1066 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1067 {
1068         printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1069                field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1070         dump_stack();
1071 }
1072
1073 static void vmcs_writel(unsigned long field, unsigned long value)
1074 {
1075         u8 error;
1076
1077         asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1078                        : "=q"(error) : "a"(value), "d"(field) : "cc");
1079         if (unlikely(error))
1080                 vmwrite_error(field, value);
1081 }
1082
1083 static void vmcs_write16(unsigned long field, u16 value)
1084 {
1085         vmcs_writel(field, value);
1086 }
1087
1088 static void vmcs_write32(unsigned long field, u32 value)
1089 {
1090         vmcs_writel(field, value);
1091 }
1092
1093 static void vmcs_write64(unsigned long field, u64 value)
1094 {
1095         vmcs_writel(field, value);
1096 #ifndef CONFIG_X86_64
1097         asm volatile ("");
1098         vmcs_writel(field+1, value >> 32);
1099 #endif
1100 }
1101
1102 static void vmcs_clear_bits(unsigned long field, u32 mask)
1103 {
1104         vmcs_writel(field, vmcs_readl(field) & ~mask);
1105 }
1106
1107 static void vmcs_set_bits(unsigned long field, u32 mask)
1108 {
1109         vmcs_writel(field, vmcs_readl(field) | mask);
1110 }
1111
1112 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1113 {
1114         vmx->segment_cache.bitmask = 0;
1115 }
1116
1117 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1118                                        unsigned field)
1119 {
1120         bool ret;
1121         u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1122
1123         if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1124                 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1125                 vmx->segment_cache.bitmask = 0;
1126         }
1127         ret = vmx->segment_cache.bitmask & mask;
1128         vmx->segment_cache.bitmask |= mask;
1129         return ret;
1130 }
1131
1132 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1133 {
1134         u16 *p = &vmx->segment_cache.seg[seg].selector;
1135
1136         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1137                 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1138         return *p;
1139 }
1140
1141 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1142 {
1143         ulong *p = &vmx->segment_cache.seg[seg].base;
1144
1145         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1146                 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1147         return *p;
1148 }
1149
1150 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1151 {
1152         u32 *p = &vmx->segment_cache.seg[seg].limit;
1153
1154         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1155                 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1156         return *p;
1157 }
1158
1159 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1160 {
1161         u32 *p = &vmx->segment_cache.seg[seg].ar;
1162
1163         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1164                 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1165         return *p;
1166 }
1167
1168 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1169 {
1170         u32 eb;
1171
1172         eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1173              (1u << NM_VECTOR) | (1u << DB_VECTOR);
1174         if ((vcpu->guest_debug &
1175              (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1176             (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1177                 eb |= 1u << BP_VECTOR;
1178         if (to_vmx(vcpu)->rmode.vm86_active)
1179                 eb = ~0;
1180         if (enable_ept)
1181                 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1182         if (vcpu->fpu_active)
1183                 eb &= ~(1u << NM_VECTOR);
1184
1185         /* When we are running a nested L2 guest and L1 specified for it a
1186          * certain exception bitmap, we must trap the same exceptions and pass
1187          * them to L1. When running L2, we will only handle the exceptions
1188          * specified above if L1 did not want them.
1189          */
1190         if (is_guest_mode(vcpu))
1191                 eb |= get_vmcs12(vcpu)->exception_bitmap;
1192
1193         vmcs_write32(EXCEPTION_BITMAP, eb);
1194 }
1195
1196 static void clear_atomic_switch_msr_special(unsigned long entry,
1197                 unsigned long exit)
1198 {
1199         vmcs_clear_bits(VM_ENTRY_CONTROLS, entry);
1200         vmcs_clear_bits(VM_EXIT_CONTROLS, exit);
1201 }
1202
1203 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1204 {
1205         unsigned i;
1206         struct msr_autoload *m = &vmx->msr_autoload;
1207
1208         switch (msr) {
1209         case MSR_EFER:
1210                 if (cpu_has_load_ia32_efer) {
1211                         clear_atomic_switch_msr_special(VM_ENTRY_LOAD_IA32_EFER,
1212                                         VM_EXIT_LOAD_IA32_EFER);
1213                         return;
1214                 }
1215                 break;
1216         case MSR_CORE_PERF_GLOBAL_CTRL:
1217                 if (cpu_has_load_perf_global_ctrl) {
1218                         clear_atomic_switch_msr_special(
1219                                         VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1220                                         VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1221                         return;
1222                 }
1223                 break;
1224         }
1225
1226         for (i = 0; i < m->nr; ++i)
1227                 if (m->guest[i].index == msr)
1228                         break;
1229
1230         if (i == m->nr)
1231                 return;
1232         --m->nr;
1233         m->guest[i] = m->guest[m->nr];
1234         m->host[i] = m->host[m->nr];
1235         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1236         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1237 }
1238
1239 static void add_atomic_switch_msr_special(unsigned long entry,
1240                 unsigned long exit, unsigned long guest_val_vmcs,
1241                 unsigned long host_val_vmcs, u64 guest_val, u64 host_val)
1242 {
1243         vmcs_write64(guest_val_vmcs, guest_val);
1244         vmcs_write64(host_val_vmcs, host_val);
1245         vmcs_set_bits(VM_ENTRY_CONTROLS, entry);
1246         vmcs_set_bits(VM_EXIT_CONTROLS, exit);
1247 }
1248
1249 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1250                                   u64 guest_val, u64 host_val)
1251 {
1252         unsigned i;
1253         struct msr_autoload *m = &vmx->msr_autoload;
1254
1255         switch (msr) {
1256         case MSR_EFER:
1257                 if (cpu_has_load_ia32_efer) {
1258                         add_atomic_switch_msr_special(VM_ENTRY_LOAD_IA32_EFER,
1259                                         VM_EXIT_LOAD_IA32_EFER,
1260                                         GUEST_IA32_EFER,
1261                                         HOST_IA32_EFER,
1262                                         guest_val, host_val);
1263                         return;
1264                 }
1265                 break;
1266         case MSR_CORE_PERF_GLOBAL_CTRL:
1267                 if (cpu_has_load_perf_global_ctrl) {
1268                         add_atomic_switch_msr_special(
1269                                         VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1270                                         VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1271                                         GUEST_IA32_PERF_GLOBAL_CTRL,
1272                                         HOST_IA32_PERF_GLOBAL_CTRL,
1273                                         guest_val, host_val);
1274                         return;
1275                 }
1276                 break;
1277         }
1278
1279         for (i = 0; i < m->nr; ++i)
1280                 if (m->guest[i].index == msr)
1281                         break;
1282
1283         if (i == NR_AUTOLOAD_MSRS) {
1284                 printk_once(KERN_WARNING"Not enough mst switch entries. "
1285                                 "Can't add msr %x\n", msr);
1286                 return;
1287         } else if (i == m->nr) {
1288                 ++m->nr;
1289                 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1290                 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1291         }
1292
1293         m->guest[i].index = msr;
1294         m->guest[i].value = guest_val;
1295         m->host[i].index = msr;
1296         m->host[i].value = host_val;
1297 }
1298
1299 static void reload_tss(void)
1300 {
1301         /*
1302          * VT restores TR but not its size.  Useless.
1303          */
1304         struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1305         struct desc_struct *descs;
1306
1307         descs = (void *)gdt->address;
1308         descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
1309         load_TR_desc();
1310 }
1311
1312 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
1313 {
1314         u64 guest_efer;
1315         u64 ignore_bits;
1316
1317         guest_efer = vmx->vcpu.arch.efer;
1318
1319         /*
1320          * NX is emulated; LMA and LME handled by hardware; SCE meaninless
1321          * outside long mode
1322          */
1323         ignore_bits = EFER_NX | EFER_SCE;
1324 #ifdef CONFIG_X86_64
1325         ignore_bits |= EFER_LMA | EFER_LME;
1326         /* SCE is meaningful only in long mode on Intel */
1327         if (guest_efer & EFER_LMA)
1328                 ignore_bits &= ~(u64)EFER_SCE;
1329 #endif
1330         guest_efer &= ~ignore_bits;
1331         guest_efer |= host_efer & ignore_bits;
1332         vmx->guest_msrs[efer_offset].data = guest_efer;
1333         vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1334
1335         clear_atomic_switch_msr(vmx, MSR_EFER);
1336         /* On ept, can't emulate nx, and must switch nx atomically */
1337         if (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX)) {
1338                 guest_efer = vmx->vcpu.arch.efer;
1339                 if (!(guest_efer & EFER_LMA))
1340                         guest_efer &= ~EFER_LME;
1341                 add_atomic_switch_msr(vmx, MSR_EFER, guest_efer, host_efer);
1342                 return false;
1343         }
1344
1345         return true;
1346 }
1347
1348 static unsigned long segment_base(u16 selector)
1349 {
1350         struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1351         struct desc_struct *d;
1352         unsigned long table_base;
1353         unsigned long v;
1354
1355         if (!(selector & ~3))
1356                 return 0;
1357
1358         table_base = gdt->address;
1359
1360         if (selector & 4) {           /* from ldt */
1361                 u16 ldt_selector = kvm_read_ldt();
1362
1363                 if (!(ldt_selector & ~3))
1364                         return 0;
1365
1366                 table_base = segment_base(ldt_selector);
1367         }
1368         d = (struct desc_struct *)(table_base + (selector & ~7));
1369         v = get_desc_base(d);
1370 #ifdef CONFIG_X86_64
1371        if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
1372                v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
1373 #endif
1374         return v;
1375 }
1376
1377 static inline unsigned long kvm_read_tr_base(void)
1378 {
1379         u16 tr;
1380         asm("str %0" : "=g"(tr));
1381         return segment_base(tr);
1382 }
1383
1384 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
1385 {
1386         struct vcpu_vmx *vmx = to_vmx(vcpu);
1387         int i;
1388
1389         if (vmx->host_state.loaded)
1390                 return;
1391
1392         vmx->host_state.loaded = 1;
1393         /*
1394          * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
1395          * allow segment selectors with cpl > 0 or ti == 1.
1396          */
1397         vmx->host_state.ldt_sel = kvm_read_ldt();
1398         vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
1399         savesegment(fs, vmx->host_state.fs_sel);
1400         if (!(vmx->host_state.fs_sel & 7)) {
1401                 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
1402                 vmx->host_state.fs_reload_needed = 0;
1403         } else {
1404                 vmcs_write16(HOST_FS_SELECTOR, 0);
1405                 vmx->host_state.fs_reload_needed = 1;
1406         }
1407         savesegment(gs, vmx->host_state.gs_sel);
1408         if (!(vmx->host_state.gs_sel & 7))
1409                 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
1410         else {
1411                 vmcs_write16(HOST_GS_SELECTOR, 0);
1412                 vmx->host_state.gs_ldt_reload_needed = 1;
1413         }
1414
1415 #ifdef CONFIG_X86_64
1416         vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
1417         vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
1418 #else
1419         vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
1420         vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
1421 #endif
1422
1423 #ifdef CONFIG_X86_64
1424         rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1425         if (is_long_mode(&vmx->vcpu))
1426                 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1427 #endif
1428         for (i = 0; i < vmx->save_nmsrs; ++i)
1429                 kvm_set_shared_msr(vmx->guest_msrs[i].index,
1430                                    vmx->guest_msrs[i].data,
1431                                    vmx->guest_msrs[i].mask);
1432 }
1433
1434 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
1435 {
1436         if (!vmx->host_state.loaded)
1437                 return;
1438
1439         ++vmx->vcpu.stat.host_state_reload;
1440         vmx->host_state.loaded = 0;
1441 #ifdef CONFIG_X86_64
1442         if (is_long_mode(&vmx->vcpu))
1443                 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1444 #endif
1445         if (vmx->host_state.gs_ldt_reload_needed) {
1446                 kvm_load_ldt(vmx->host_state.ldt_sel);
1447 #ifdef CONFIG_X86_64
1448                 load_gs_index(vmx->host_state.gs_sel);
1449 #else
1450                 loadsegment(gs, vmx->host_state.gs_sel);
1451 #endif
1452         }
1453         if (vmx->host_state.fs_reload_needed)
1454                 loadsegment(fs, vmx->host_state.fs_sel);
1455         reload_tss();
1456 #ifdef CONFIG_X86_64
1457         wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1458 #endif
1459         if (current_thread_info()->status & TS_USEDFPU)
1460                 clts();
1461         load_gdt(&__get_cpu_var(host_gdt));
1462 }
1463
1464 static void vmx_load_host_state(struct vcpu_vmx *vmx)
1465 {
1466         preempt_disable();
1467         __vmx_load_host_state(vmx);
1468         preempt_enable();
1469 }
1470
1471 /*
1472  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1473  * vcpu mutex is already taken.
1474  */
1475 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1476 {
1477         struct vcpu_vmx *vmx = to_vmx(vcpu);
1478         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1479
1480         if (!vmm_exclusive)
1481                 kvm_cpu_vmxon(phys_addr);
1482         else if (vmx->loaded_vmcs->cpu != cpu)
1483                 loaded_vmcs_clear(vmx->loaded_vmcs);
1484
1485         if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1486                 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1487                 vmcs_load(vmx->loaded_vmcs->vmcs);
1488         }
1489
1490         if (vmx->loaded_vmcs->cpu != cpu) {
1491                 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1492                 unsigned long sysenter_esp;
1493
1494                 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1495                 local_irq_disable();
1496                 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1497                          &per_cpu(loaded_vmcss_on_cpu, cpu));
1498                 local_irq_enable();
1499
1500                 /*
1501                  * Linux uses per-cpu TSS and GDT, so set these when switching
1502                  * processors.
1503                  */
1504                 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
1505                 vmcs_writel(HOST_GDTR_BASE, gdt->address);   /* 22.2.4 */
1506
1507                 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1508                 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1509                 vmx->loaded_vmcs->cpu = cpu;
1510         }
1511 }
1512
1513 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1514 {
1515         __vmx_load_host_state(to_vmx(vcpu));
1516         if (!vmm_exclusive) {
1517                 __loaded_vmcs_clear(to_vmx(vcpu)->loaded_vmcs);
1518                 vcpu->cpu = -1;
1519                 kvm_cpu_vmxoff();
1520         }
1521 }
1522
1523 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
1524 {
1525         ulong cr0;
1526
1527         if (vcpu->fpu_active)
1528                 return;
1529         vcpu->fpu_active = 1;
1530         cr0 = vmcs_readl(GUEST_CR0);
1531         cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
1532         cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
1533         vmcs_writel(GUEST_CR0, cr0);
1534         update_exception_bitmap(vcpu);
1535         vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
1536         if (is_guest_mode(vcpu))
1537                 vcpu->arch.cr0_guest_owned_bits &=
1538                         ~get_vmcs12(vcpu)->cr0_guest_host_mask;
1539         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1540 }
1541
1542 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
1543
1544 /*
1545  * Return the cr0 value that a nested guest would read. This is a combination
1546  * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
1547  * its hypervisor (cr0_read_shadow).
1548  */
1549 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
1550 {
1551         return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
1552                 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
1553 }
1554 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
1555 {
1556         return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
1557                 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
1558 }
1559
1560 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
1561 {
1562         /* Note that there is no vcpu->fpu_active = 0 here. The caller must
1563          * set this *before* calling this function.
1564          */
1565         vmx_decache_cr0_guest_bits(vcpu);
1566         vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
1567         update_exception_bitmap(vcpu);
1568         vcpu->arch.cr0_guest_owned_bits = 0;
1569         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1570         if (is_guest_mode(vcpu)) {
1571                 /*
1572                  * L1's specified read shadow might not contain the TS bit,
1573                  * so now that we turned on shadowing of this bit, we need to
1574                  * set this bit of the shadow. Like in nested_vmx_run we need
1575                  * nested_read_cr0(vmcs12), but vmcs12->guest_cr0 is not yet
1576                  * up-to-date here because we just decached cr0.TS (and we'll
1577                  * only update vmcs12->guest_cr0 on nested exit).
1578                  */
1579                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1580                 vmcs12->guest_cr0 = (vmcs12->guest_cr0 & ~X86_CR0_TS) |
1581                         (vcpu->arch.cr0 & X86_CR0_TS);
1582                 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
1583         } else
1584                 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
1585 }
1586
1587 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1588 {
1589         unsigned long rflags, save_rflags;
1590
1591         if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
1592                 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1593                 rflags = vmcs_readl(GUEST_RFLAGS);
1594                 if (to_vmx(vcpu)->rmode.vm86_active) {
1595                         rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1596                         save_rflags = to_vmx(vcpu)->rmode.save_rflags;
1597                         rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1598                 }
1599                 to_vmx(vcpu)->rflags = rflags;
1600         }
1601         return to_vmx(vcpu)->rflags;
1602 }
1603
1604 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1605 {
1606         __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1607         __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
1608         to_vmx(vcpu)->rflags = rflags;
1609         if (to_vmx(vcpu)->rmode.vm86_active) {
1610                 to_vmx(vcpu)->rmode.save_rflags = rflags;
1611                 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1612         }
1613         vmcs_writel(GUEST_RFLAGS, rflags);
1614 }
1615
1616 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1617 {
1618         u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1619         int ret = 0;
1620
1621         if (interruptibility & GUEST_INTR_STATE_STI)
1622                 ret |= KVM_X86_SHADOW_INT_STI;
1623         if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1624                 ret |= KVM_X86_SHADOW_INT_MOV_SS;
1625
1626         return ret & mask;
1627 }
1628
1629 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1630 {
1631         u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1632         u32 interruptibility = interruptibility_old;
1633
1634         interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1635
1636         if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1637                 interruptibility |= GUEST_INTR_STATE_MOV_SS;
1638         else if (mask & KVM_X86_SHADOW_INT_STI)
1639                 interruptibility |= GUEST_INTR_STATE_STI;
1640
1641         if ((interruptibility != interruptibility_old))
1642                 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1643 }
1644
1645 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
1646 {
1647         unsigned long rip;
1648
1649         rip = kvm_rip_read(vcpu);
1650         rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1651         kvm_rip_write(vcpu, rip);
1652
1653         /* skipping an emulated instruction also counts */
1654         vmx_set_interrupt_shadow(vcpu, 0);
1655 }
1656
1657 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1658 {
1659         /* Ensure that we clear the HLT state in the VMCS.  We don't need to
1660          * explicitly skip the instruction because if the HLT state is set, then
1661          * the instruction is already executing and RIP has already been
1662          * advanced. */
1663         if (!yield_on_hlt &&
1664             vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1665                 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1666 }
1667
1668 /*
1669  * KVM wants to inject page-faults which it got to the guest. This function
1670  * checks whether in a nested guest, we need to inject them to L1 or L2.
1671  * This function assumes it is called with the exit reason in vmcs02 being
1672  * a #PF exception (this is the only case in which KVM injects a #PF when L2
1673  * is running).
1674  */
1675 static int nested_pf_handled(struct kvm_vcpu *vcpu)
1676 {
1677         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1678
1679         /* TODO: also check PFEC_MATCH/MASK, not just EB.PF. */
1680         if (!(vmcs12->exception_bitmap & PF_VECTOR))
1681                 return 0;
1682
1683         nested_vmx_vmexit(vcpu);
1684         return 1;
1685 }
1686
1687 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
1688                                 bool has_error_code, u32 error_code,
1689                                 bool reinject)
1690 {
1691         struct vcpu_vmx *vmx = to_vmx(vcpu);
1692         u32 intr_info = nr | INTR_INFO_VALID_MASK;
1693
1694         if (nr == PF_VECTOR && is_guest_mode(vcpu) &&
1695                 nested_pf_handled(vcpu))
1696                 return;
1697
1698         if (has_error_code) {
1699                 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1700                 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1701         }
1702
1703         if (vmx->rmode.vm86_active) {
1704                 int inc_eip = 0;
1705                 if (kvm_exception_is_soft(nr))
1706                         inc_eip = vcpu->arch.event_exit_inst_len;
1707                 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
1708                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
1709                 return;
1710         }
1711
1712         if (kvm_exception_is_soft(nr)) {
1713                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1714                              vmx->vcpu.arch.event_exit_inst_len);
1715                 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1716         } else
1717                 intr_info |= INTR_TYPE_HARD_EXCEPTION;
1718
1719         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1720         vmx_clear_hlt(vcpu);
1721 }
1722
1723 static bool vmx_rdtscp_supported(void)
1724 {
1725         return cpu_has_vmx_rdtscp();
1726 }
1727
1728 /*
1729  * Swap MSR entry in host/guest MSR entry array.
1730  */
1731 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1732 {
1733         struct shared_msr_entry tmp;
1734
1735         tmp = vmx->guest_msrs[to];
1736         vmx->guest_msrs[to] = vmx->guest_msrs[from];
1737         vmx->guest_msrs[from] = tmp;
1738 }
1739
1740 /*
1741  * Set up the vmcs to automatically save and restore system
1742  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
1743  * mode, as fiddling with msrs is very expensive.
1744  */
1745 static void setup_msrs(struct vcpu_vmx *vmx)
1746 {
1747         int save_nmsrs, index;
1748         unsigned long *msr_bitmap;
1749
1750         vmx_load_host_state(vmx);
1751         save_nmsrs = 0;
1752 #ifdef CONFIG_X86_64
1753         if (is_long_mode(&vmx->vcpu)) {
1754                 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1755                 if (index >= 0)
1756                         move_msr_up(vmx, index, save_nmsrs++);
1757                 index = __find_msr_index(vmx, MSR_LSTAR);
1758                 if (index >= 0)
1759                         move_msr_up(vmx, index, save_nmsrs++);
1760                 index = __find_msr_index(vmx, MSR_CSTAR);
1761                 if (index >= 0)
1762                         move_msr_up(vmx, index, save_nmsrs++);
1763                 index = __find_msr_index(vmx, MSR_TSC_AUX);
1764                 if (index >= 0 && vmx->rdtscp_enabled)
1765                         move_msr_up(vmx, index, save_nmsrs++);
1766                 /*
1767                  * MSR_STAR is only needed on long mode guests, and only
1768                  * if efer.sce is enabled.
1769                  */
1770                 index = __find_msr_index(vmx, MSR_STAR);
1771                 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
1772                         move_msr_up(vmx, index, save_nmsrs++);
1773         }
1774 #endif
1775         index = __find_msr_index(vmx, MSR_EFER);
1776         if (index >= 0 && update_transition_efer(vmx, index))
1777                 move_msr_up(vmx, index, save_nmsrs++);
1778
1779         vmx->save_nmsrs = save_nmsrs;
1780
1781         if (cpu_has_vmx_msr_bitmap()) {
1782                 if (is_long_mode(&vmx->vcpu))
1783                         msr_bitmap = vmx_msr_bitmap_longmode;
1784                 else
1785                         msr_bitmap = vmx_msr_bitmap_legacy;
1786
1787                 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
1788         }
1789 }
1790
1791 /*
1792  * reads and returns guest's timestamp counter "register"
1793  * guest_tsc = host_tsc + tsc_offset    -- 21.3
1794  */
1795 static u64 guest_read_tsc(void)
1796 {
1797         u64 host_tsc, tsc_offset;
1798
1799         rdtscll(host_tsc);
1800         tsc_offset = vmcs_read64(TSC_OFFSET);
1801         return host_tsc + tsc_offset;
1802 }
1803
1804 /*
1805  * Like guest_read_tsc, but always returns L1's notion of the timestamp
1806  * counter, even if a nested guest (L2) is currently running.
1807  */
1808 u64 vmx_read_l1_tsc(struct kvm_vcpu *vcpu)
1809 {
1810         u64 host_tsc, tsc_offset;
1811
1812         rdtscll(host_tsc);
1813         tsc_offset = is_guest_mode(vcpu) ?
1814                 to_vmx(vcpu)->nested.vmcs01_tsc_offset :
1815                 vmcs_read64(TSC_OFFSET);
1816         return host_tsc + tsc_offset;
1817 }
1818
1819 /*
1820  * Empty call-back. Needs to be implemented when VMX enables the SET_TSC_KHZ
1821  * ioctl. In this case the call-back should update internal vmx state to make
1822  * the changes effective.
1823  */
1824 static void vmx_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1825 {
1826         /* Nothing to do here */
1827 }
1828
1829 /*
1830  * writes 'offset' into guest's timestamp counter offset register
1831  */
1832 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1833 {
1834         if (is_guest_mode(vcpu)) {
1835                 /*
1836                  * We're here if L1 chose not to trap WRMSR to TSC. According
1837                  * to the spec, this should set L1's TSC; The offset that L1
1838                  * set for L2 remains unchanged, and still needs to be added
1839                  * to the newly set TSC to get L2's TSC.
1840                  */
1841                 struct vmcs12 *vmcs12;
1842                 to_vmx(vcpu)->nested.vmcs01_tsc_offset = offset;
1843                 /* recalculate vmcs02.TSC_OFFSET: */
1844                 vmcs12 = get_vmcs12(vcpu);
1845                 vmcs_write64(TSC_OFFSET, offset +
1846                         (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
1847                          vmcs12->tsc_offset : 0));
1848         } else {
1849                 vmcs_write64(TSC_OFFSET, offset);
1850         }
1851 }
1852
1853 static void vmx_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment)
1854 {
1855         u64 offset = vmcs_read64(TSC_OFFSET);
1856         vmcs_write64(TSC_OFFSET, offset + adjustment);
1857         if (is_guest_mode(vcpu)) {
1858                 /* Even when running L2, the adjustment needs to apply to L1 */
1859                 to_vmx(vcpu)->nested.vmcs01_tsc_offset += adjustment;
1860         }
1861 }
1862
1863 static u64 vmx_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1864 {
1865         return target_tsc - native_read_tsc();
1866 }
1867
1868 static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
1869 {
1870         struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
1871         return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
1872 }
1873
1874 /*
1875  * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1876  * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1877  * all guests if the "nested" module option is off, and can also be disabled
1878  * for a single guest by disabling its VMX cpuid bit.
1879  */
1880 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1881 {
1882         return nested && guest_cpuid_has_vmx(vcpu);
1883 }
1884
1885 /*
1886  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
1887  * returned for the various VMX controls MSRs when nested VMX is enabled.
1888  * The same values should also be used to verify that vmcs12 control fields are
1889  * valid during nested entry from L1 to L2.
1890  * Each of these control msrs has a low and high 32-bit half: A low bit is on
1891  * if the corresponding bit in the (32-bit) control field *must* be on, and a
1892  * bit in the high half is on if the corresponding bit in the control field
1893  * may be on. See also vmx_control_verify().
1894  * TODO: allow these variables to be modified (downgraded) by module options
1895  * or other means.
1896  */
1897 static u32 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high;
1898 static u32 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high;
1899 static u32 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high;
1900 static u32 nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high;
1901 static u32 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high;
1902 static __init void nested_vmx_setup_ctls_msrs(void)
1903 {
1904         /*
1905          * Note that as a general rule, the high half of the MSRs (bits in
1906          * the control fields which may be 1) should be initialized by the
1907          * intersection of the underlying hardware's MSR (i.e., features which
1908          * can be supported) and the list of features we want to expose -
1909          * because they are known to be properly supported in our code.
1910          * Also, usually, the low half of the MSRs (bits which must be 1) can
1911          * be set to 0, meaning that L1 may turn off any of these bits. The
1912          * reason is that if one of these bits is necessary, it will appear
1913          * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
1914          * fields of vmcs01 and vmcs02, will turn these bits off - and
1915          * nested_vmx_exit_handled() will not pass related exits to L1.
1916          * These rules have exceptions below.
1917          */
1918
1919         /* pin-based controls */
1920         /*
1921          * According to the Intel spec, if bit 55 of VMX_BASIC is off (as it is
1922          * in our case), bits 1, 2 and 4 (i.e., 0x16) must be 1 in this MSR.
1923          */
1924         nested_vmx_pinbased_ctls_low = 0x16 ;
1925         nested_vmx_pinbased_ctls_high = 0x16 |
1926                 PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING |
1927                 PIN_BASED_VIRTUAL_NMIS;
1928
1929         /* exit controls */
1930         nested_vmx_exit_ctls_low = 0;
1931         /* Note that guest use of VM_EXIT_ACK_INTR_ON_EXIT is not supported. */
1932 #ifdef CONFIG_X86_64
1933         nested_vmx_exit_ctls_high = VM_EXIT_HOST_ADDR_SPACE_SIZE;
1934 #else
1935         nested_vmx_exit_ctls_high = 0;
1936 #endif
1937
1938         /* entry controls */
1939         rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
1940                 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high);
1941         nested_vmx_entry_ctls_low = 0;
1942         nested_vmx_entry_ctls_high &=
1943                 VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_IA32E_MODE;
1944
1945         /* cpu-based controls */
1946         rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
1947                 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high);
1948         nested_vmx_procbased_ctls_low = 0;
1949         nested_vmx_procbased_ctls_high &=
1950                 CPU_BASED_VIRTUAL_INTR_PENDING | CPU_BASED_USE_TSC_OFFSETING |
1951                 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
1952                 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
1953                 CPU_BASED_CR3_STORE_EXITING |
1954 #ifdef CONFIG_X86_64
1955                 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
1956 #endif
1957                 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
1958                 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_EXITING |
1959                 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1960         /*
1961          * We can allow some features even when not supported by the
1962          * hardware. For example, L1 can specify an MSR bitmap - and we
1963          * can use it to avoid exits to L1 - even when L0 runs L2
1964          * without MSR bitmaps.
1965          */
1966         nested_vmx_procbased_ctls_high |= CPU_BASED_USE_MSR_BITMAPS;
1967
1968         /* secondary cpu-based controls */
1969         rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
1970                 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high);
1971         nested_vmx_secondary_ctls_low = 0;
1972         nested_vmx_secondary_ctls_high &=
1973                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1974 }
1975
1976 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
1977 {
1978         /*
1979          * Bits 0 in high must be 0, and bits 1 in low must be 1.
1980          */
1981         return ((control & high) | low) == control;
1982 }
1983
1984 static inline u64 vmx_control_msr(u32 low, u32 high)
1985 {
1986         return low | ((u64)high << 32);
1987 }
1988
1989 /*
1990  * If we allow our guest to use VMX instructions (i.e., nested VMX), we should
1991  * also let it use VMX-specific MSRs.
1992  * vmx_get_vmx_msr() and vmx_set_vmx_msr() return 1 when we handled a
1993  * VMX-specific MSR, or 0 when we haven't (and the caller should handle it
1994  * like all other MSRs).
1995  */
1996 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1997 {
1998         if (!nested_vmx_allowed(vcpu) && msr_index >= MSR_IA32_VMX_BASIC &&
1999                      msr_index <= MSR_IA32_VMX_TRUE_ENTRY_CTLS) {
2000                 /*
2001                  * According to the spec, processors which do not support VMX
2002                  * should throw a #GP(0) when VMX capability MSRs are read.
2003                  */
2004                 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
2005                 return 1;
2006         }
2007
2008         switch (msr_index) {
2009         case MSR_IA32_FEATURE_CONTROL:
2010                 *pdata = 0;
2011                 break;
2012         case MSR_IA32_VMX_BASIC:
2013                 /*
2014                  * This MSR reports some information about VMX support. We
2015                  * should return information about the VMX we emulate for the
2016                  * guest, and the VMCS structure we give it - not about the
2017                  * VMX support of the underlying hardware.
2018                  */
2019                 *pdata = VMCS12_REVISION |
2020                            ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2021                            (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2022                 break;
2023         case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2024         case MSR_IA32_VMX_PINBASED_CTLS:
2025                 *pdata = vmx_control_msr(nested_vmx_pinbased_ctls_low,
2026                                         nested_vmx_pinbased_ctls_high);
2027                 break;
2028         case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2029         case MSR_IA32_VMX_PROCBASED_CTLS:
2030                 *pdata = vmx_control_msr(nested_vmx_procbased_ctls_low,
2031                                         nested_vmx_procbased_ctls_high);
2032                 break;
2033         case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2034         case MSR_IA32_VMX_EXIT_CTLS:
2035                 *pdata = vmx_control_msr(nested_vmx_exit_ctls_low,
2036                                         nested_vmx_exit_ctls_high);
2037                 break;
2038         case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2039         case MSR_IA32_VMX_ENTRY_CTLS:
2040                 *pdata = vmx_control_msr(nested_vmx_entry_ctls_low,
2041                                         nested_vmx_entry_ctls_high);
2042                 break;
2043         case MSR_IA32_VMX_MISC:
2044                 *pdata = 0;
2045                 break;
2046         /*
2047          * These MSRs specify bits which the guest must keep fixed (on or off)
2048          * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2049          * We picked the standard core2 setting.
2050          */
2051 #define VMXON_CR0_ALWAYSON      (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2052 #define VMXON_CR4_ALWAYSON      X86_CR4_VMXE
2053         case MSR_IA32_VMX_CR0_FIXED0:
2054                 *pdata = VMXON_CR0_ALWAYSON;
2055                 break;
2056         case MSR_IA32_VMX_CR0_FIXED1:
2057                 *pdata = -1ULL;
2058                 break;
2059         case MSR_IA32_VMX_CR4_FIXED0:
2060                 *pdata = VMXON_CR4_ALWAYSON;
2061                 break;
2062         case MSR_IA32_VMX_CR4_FIXED1:
2063                 *pdata = -1ULL;
2064                 break;
2065         case MSR_IA32_VMX_VMCS_ENUM:
2066                 *pdata = 0x1f;
2067                 break;
2068         case MSR_IA32_VMX_PROCBASED_CTLS2:
2069                 *pdata = vmx_control_msr(nested_vmx_secondary_ctls_low,
2070                                         nested_vmx_secondary_ctls_high);
2071                 break;
2072         case MSR_IA32_VMX_EPT_VPID_CAP:
2073                 /* Currently, no nested ept or nested vpid */
2074                 *pdata = 0;
2075                 break;
2076         default:
2077                 return 0;
2078         }
2079
2080         return 1;
2081 }
2082
2083 static int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
2084 {
2085         if (!nested_vmx_allowed(vcpu))
2086                 return 0;
2087
2088         if (msr_index == MSR_IA32_FEATURE_CONTROL)
2089                 /* TODO: the right thing. */
2090                 return 1;
2091         /*
2092          * No need to treat VMX capability MSRs specially: If we don't handle
2093          * them, handle_wrmsr will #GP(0), which is correct (they are readonly)
2094          */
2095         return 0;
2096 }
2097
2098 /*
2099  * Reads an msr value (of 'msr_index') into 'pdata'.
2100  * Returns 0 on success, non-0 otherwise.
2101  * Assumes vcpu_load() was already called.
2102  */
2103 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2104 {
2105         u64 data;
2106         struct shared_msr_entry *msr;
2107
2108         if (!pdata) {
2109                 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
2110                 return -EINVAL;
2111         }
2112
2113         switch (msr_index) {
2114 #ifdef CONFIG_X86_64
2115         case MSR_FS_BASE:
2116                 data = vmcs_readl(GUEST_FS_BASE);
2117                 break;
2118         case MSR_GS_BASE:
2119                 data = vmcs_readl(GUEST_GS_BASE);
2120                 break;
2121         case MSR_KERNEL_GS_BASE:
2122                 vmx_load_host_state(to_vmx(vcpu));
2123                 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
2124                 break;
2125 #endif
2126         case MSR_EFER:
2127                 return kvm_get_msr_common(vcpu, msr_index, pdata);
2128         case MSR_IA32_TSC:
2129                 data = guest_read_tsc();
2130                 break;
2131         case MSR_IA32_SYSENTER_CS:
2132                 data = vmcs_read32(GUEST_SYSENTER_CS);
2133                 break;
2134         case MSR_IA32_SYSENTER_EIP:
2135                 data = vmcs_readl(GUEST_SYSENTER_EIP);
2136                 break;
2137         case MSR_IA32_SYSENTER_ESP:
2138                 data = vmcs_readl(GUEST_SYSENTER_ESP);
2139                 break;
2140         case MSR_TSC_AUX:
2141                 if (!to_vmx(vcpu)->rdtscp_enabled)
2142                         return 1;
2143                 /* Otherwise falls through */
2144         default:
2145                 vmx_load_host_state(to_vmx(vcpu));
2146                 if (vmx_get_vmx_msr(vcpu, msr_index, pdata))
2147                         return 0;
2148                 msr = find_msr_entry(to_vmx(vcpu), msr_index);
2149                 if (msr) {
2150                         vmx_load_host_state(to_vmx(vcpu));
2151                         data = msr->data;
2152                         break;
2153                 }
2154                 return kvm_get_msr_common(vcpu, msr_index, pdata);
2155         }
2156
2157         *pdata = data;
2158         return 0;
2159 }
2160
2161 /*
2162  * Writes msr value into into the appropriate "register".
2163  * Returns 0 on success, non-0 otherwise.
2164  * Assumes vcpu_load() was already called.
2165  */
2166 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
2167 {
2168         struct vcpu_vmx *vmx = to_vmx(vcpu);
2169         struct shared_msr_entry *msr;
2170         int ret = 0;
2171
2172         switch (msr_index) {
2173         case MSR_EFER:
2174                 vmx_load_host_state(vmx);
2175                 ret = kvm_set_msr_common(vcpu, msr_index, data);
2176                 break;
2177 #ifdef CONFIG_X86_64
2178         case MSR_FS_BASE:
2179                 vmx_segment_cache_clear(vmx);
2180                 vmcs_writel(GUEST_FS_BASE, data);
2181                 break;
2182         case MSR_GS_BASE:
2183                 vmx_segment_cache_clear(vmx);
2184                 vmcs_writel(GUEST_GS_BASE, data);
2185                 break;
2186         case MSR_KERNEL_GS_BASE:
2187                 vmx_load_host_state(vmx);
2188                 vmx->msr_guest_kernel_gs_base = data;
2189                 break;
2190 #endif
2191         case MSR_IA32_SYSENTER_CS:
2192                 vmcs_write32(GUEST_SYSENTER_CS, data);
2193                 break;
2194         case MSR_IA32_SYSENTER_EIP:
2195                 vmcs_writel(GUEST_SYSENTER_EIP, data);
2196                 break;
2197         case MSR_IA32_SYSENTER_ESP:
2198                 vmcs_writel(GUEST_SYSENTER_ESP, data);
2199                 break;
2200         case MSR_IA32_TSC:
2201                 kvm_write_tsc(vcpu, data);
2202                 break;
2203         case MSR_IA32_CR_PAT:
2204                 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2205                         vmcs_write64(GUEST_IA32_PAT, data);
2206                         vcpu->arch.pat = data;
2207                         break;
2208                 }
2209                 ret = kvm_set_msr_common(vcpu, msr_index, data);
2210                 break;
2211         case MSR_TSC_AUX:
2212                 if (!vmx->rdtscp_enabled)
2213                         return 1;
2214                 /* Check reserved bit, higher 32 bits should be zero */
2215                 if ((data >> 32) != 0)
2216                         return 1;
2217                 /* Otherwise falls through */
2218         default:
2219                 if (vmx_set_vmx_msr(vcpu, msr_index, data))
2220                         break;
2221                 msr = find_msr_entry(vmx, msr_index);
2222                 if (msr) {
2223                         vmx_load_host_state(vmx);
2224                         msr->data = data;
2225                         break;
2226                 }
2227                 ret = kvm_set_msr_common(vcpu, msr_index, data);
2228         }
2229
2230         return ret;
2231 }
2232
2233 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2234 {
2235         __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
2236         switch (reg) {
2237         case VCPU_REGS_RSP:
2238                 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2239                 break;
2240         case VCPU_REGS_RIP:
2241                 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2242                 break;
2243         case VCPU_EXREG_PDPTR:
2244                 if (enable_ept)
2245                         ept_save_pdptrs(vcpu);
2246                 break;
2247         default:
2248                 break;
2249         }
2250 }
2251
2252 static void set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
2253 {
2254         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
2255                 vmcs_writel(GUEST_DR7, dbg->arch.debugreg[7]);
2256         else
2257                 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
2258
2259         update_exception_bitmap(vcpu);
2260 }
2261
2262 static __init int cpu_has_kvm_support(void)
2263 {
2264         return cpu_has_vmx();
2265 }
2266
2267 static __init int vmx_disabled_by_bios(void)
2268 {
2269         u64 msr;
2270
2271         rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
2272         if (msr & FEATURE_CONTROL_LOCKED) {
2273                 /* launched w/ TXT and VMX disabled */
2274                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2275                         && tboot_enabled())
2276                         return 1;
2277                 /* launched w/o TXT and VMX only enabled w/ TXT */
2278                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2279                         && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2280                         && !tboot_enabled()) {
2281                         printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
2282                                 "activate TXT before enabling KVM\n");
2283                         return 1;
2284                 }
2285                 /* launched w/o TXT and VMX disabled */
2286                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2287                         && !tboot_enabled())
2288                         return 1;
2289         }
2290
2291         return 0;
2292 }
2293
2294 static void kvm_cpu_vmxon(u64 addr)
2295 {
2296         asm volatile (ASM_VMX_VMXON_RAX
2297                         : : "a"(&addr), "m"(addr)
2298                         : "memory", "cc");
2299 }
2300
2301 static int hardware_enable(void *garbage)
2302 {
2303         int cpu = raw_smp_processor_id();
2304         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2305         u64 old, test_bits;
2306
2307         if (read_cr4() & X86_CR4_VMXE)
2308                 return -EBUSY;
2309
2310         INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2311         rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
2312
2313         test_bits = FEATURE_CONTROL_LOCKED;
2314         test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
2315         if (tboot_enabled())
2316                 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
2317
2318         if ((old & test_bits) != test_bits) {
2319                 /* enable and lock */
2320                 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
2321         }
2322         write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
2323
2324         if (vmm_exclusive) {
2325                 kvm_cpu_vmxon(phys_addr);
2326                 ept_sync_global();
2327         }
2328
2329         store_gdt(&__get_cpu_var(host_gdt));
2330
2331         return 0;
2332 }
2333
2334 static void vmclear_local_loaded_vmcss(void)
2335 {
2336         int cpu = raw_smp_processor_id();
2337         struct loaded_vmcs *v, *n;
2338
2339         list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2340                                  loaded_vmcss_on_cpu_link)
2341                 __loaded_vmcs_clear(v);
2342 }
2343
2344
2345 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2346  * tricks.
2347  */
2348 static void kvm_cpu_vmxoff(void)
2349 {
2350         asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
2351 }
2352
2353 static void hardware_disable(void *garbage)
2354 {
2355         if (vmm_exclusive) {
2356                 vmclear_local_loaded_vmcss();
2357                 kvm_cpu_vmxoff();
2358         }
2359         write_cr4(read_cr4() & ~X86_CR4_VMXE);
2360 }
2361
2362 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2363                                       u32 msr, u32 *result)
2364 {
2365         u32 vmx_msr_low, vmx_msr_high;
2366         u32 ctl = ctl_min | ctl_opt;
2367
2368         rdmsr(msr, vmx_msr_low, vmx_msr_high);
2369
2370         ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2371         ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
2372
2373         /* Ensure minimum (required) set of control bits are supported. */
2374         if (ctl_min & ~ctl)
2375                 return -EIO;
2376
2377         *result = ctl;
2378         return 0;
2379 }
2380
2381 static __init bool allow_1_setting(u32 msr, u32 ctl)
2382 {
2383         u32 vmx_msr_low, vmx_msr_high;
2384
2385         rdmsr(msr, vmx_msr_low, vmx_msr_high);
2386         return vmx_msr_high & ctl;
2387 }
2388
2389 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
2390 {
2391         u32 vmx_msr_low, vmx_msr_high;
2392         u32 min, opt, min2, opt2;
2393         u32 _pin_based_exec_control = 0;
2394         u32 _cpu_based_exec_control = 0;
2395         u32 _cpu_based_2nd_exec_control = 0;
2396         u32 _vmexit_control = 0;
2397         u32 _vmentry_control = 0;
2398
2399         min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2400         opt = PIN_BASED_VIRTUAL_NMIS;
2401         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2402                                 &_pin_based_exec_control) < 0)
2403                 return -EIO;
2404
2405         min =
2406 #ifdef CONFIG_X86_64
2407               CPU_BASED_CR8_LOAD_EXITING |
2408               CPU_BASED_CR8_STORE_EXITING |
2409 #endif
2410               CPU_BASED_CR3_LOAD_EXITING |
2411               CPU_BASED_CR3_STORE_EXITING |
2412               CPU_BASED_USE_IO_BITMAPS |
2413               CPU_BASED_MOV_DR_EXITING |
2414               CPU_BASED_USE_TSC_OFFSETING |
2415               CPU_BASED_MWAIT_EXITING |
2416               CPU_BASED_MONITOR_EXITING |
2417               CPU_BASED_INVLPG_EXITING;
2418
2419         if (yield_on_hlt)
2420                 min |= CPU_BASED_HLT_EXITING;
2421
2422         opt = CPU_BASED_TPR_SHADOW |
2423               CPU_BASED_USE_MSR_BITMAPS |
2424               CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2425         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2426                                 &_cpu_based_exec_control) < 0)
2427                 return -EIO;
2428 #ifdef CONFIG_X86_64
2429         if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2430                 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2431                                            ~CPU_BASED_CR8_STORE_EXITING;
2432 #endif
2433         if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2434                 min2 = 0;
2435                 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2436                         SECONDARY_EXEC_WBINVD_EXITING |
2437                         SECONDARY_EXEC_ENABLE_VPID |
2438                         SECONDARY_EXEC_ENABLE_EPT |
2439                         SECONDARY_EXEC_UNRESTRICTED_GUEST |
2440                         SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2441                         SECONDARY_EXEC_RDTSCP;
2442                 if (adjust_vmx_controls(min2, opt2,
2443                                         MSR_IA32_VMX_PROCBASED_CTLS2,
2444                                         &_cpu_based_2nd_exec_control) < 0)
2445                         return -EIO;
2446         }
2447 #ifndef CONFIG_X86_64
2448         if (!(_cpu_based_2nd_exec_control &
2449                                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2450                 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2451 #endif
2452         if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2453                 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2454                    enabled */
2455                 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2456                                              CPU_BASED_CR3_STORE_EXITING |
2457                                              CPU_BASED_INVLPG_EXITING);
2458                 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
2459                       vmx_capability.ept, vmx_capability.vpid);
2460         }
2461
2462         min = 0;
2463 #ifdef CONFIG_X86_64
2464         min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2465 #endif
2466         opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
2467         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2468                                 &_vmexit_control) < 0)
2469                 return -EIO;
2470
2471         min = 0;
2472         opt = VM_ENTRY_LOAD_IA32_PAT;
2473         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2474                                 &_vmentry_control) < 0)
2475                 return -EIO;
2476
2477         rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2478
2479         /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2480         if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2481                 return -EIO;
2482
2483 #ifdef CONFIG_X86_64
2484         /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2485         if (vmx_msr_high & (1u<<16))
2486                 return -EIO;
2487 #endif
2488
2489         /* Require Write-Back (WB) memory type for VMCS accesses. */
2490         if (((vmx_msr_high >> 18) & 15) != 6)
2491                 return -EIO;
2492
2493         vmcs_conf->size = vmx_msr_high & 0x1fff;
2494         vmcs_conf->order = get_order(vmcs_config.size);
2495         vmcs_conf->revision_id = vmx_msr_low;
2496
2497         vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2498         vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2499         vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2500         vmcs_conf->vmexit_ctrl         = _vmexit_control;
2501         vmcs_conf->vmentry_ctrl        = _vmentry_control;
2502
2503         cpu_has_load_ia32_efer =
2504                 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2505                                 VM_ENTRY_LOAD_IA32_EFER)
2506                 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2507                                    VM_EXIT_LOAD_IA32_EFER);
2508
2509         cpu_has_load_perf_global_ctrl =
2510                 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2511                                 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
2512                 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2513                                    VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
2514
2515         /*
2516          * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
2517          * but due to arrata below it can't be used. Workaround is to use
2518          * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2519          *
2520          * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
2521          *
2522          * AAK155             (model 26)
2523          * AAP115             (model 30)
2524          * AAT100             (model 37)
2525          * BC86,AAY89,BD102   (model 44)
2526          * BA97               (model 46)
2527          *
2528          */
2529         if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
2530                 switch (boot_cpu_data.x86_model) {
2531                 case 26:
2532                 case 30:
2533                 case 37:
2534                 case 44:
2535                 case 46:
2536                         cpu_has_load_perf_global_ctrl = false;
2537                         printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2538                                         "does not work properly. Using workaround\n");
2539                         break;
2540                 default:
2541                         break;
2542                 }
2543         }
2544
2545         return 0;
2546 }
2547
2548 static struct vmcs *alloc_vmcs_cpu(int cpu)
2549 {
2550         int node = cpu_to_node(cpu);
2551         struct page *pages;
2552         struct vmcs *vmcs;
2553
2554         pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
2555         if (!pages)
2556                 return NULL;
2557         vmcs = page_address(pages);
2558         memset(vmcs, 0, vmcs_config.size);
2559         vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
2560         return vmcs;
2561 }
2562
2563 static struct vmcs *alloc_vmcs(void)
2564 {
2565         return alloc_vmcs_cpu(raw_smp_processor_id());
2566 }
2567
2568 static void free_vmcs(struct vmcs *vmcs)
2569 {
2570         free_pages((unsigned long)vmcs, vmcs_config.order);
2571 }
2572
2573 /*
2574  * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2575  */
2576 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2577 {
2578         if (!loaded_vmcs->vmcs)
2579                 return;
2580         loaded_vmcs_clear(loaded_vmcs);
2581         free_vmcs(loaded_vmcs->vmcs);
2582         loaded_vmcs->vmcs = NULL;
2583 }
2584
2585 static void free_kvm_area(void)
2586 {
2587         int cpu;
2588
2589         for_each_possible_cpu(cpu) {
2590                 free_vmcs(per_cpu(vmxarea, cpu));
2591                 per_cpu(vmxarea, cpu) = NULL;
2592         }
2593 }
2594
2595 static __init int alloc_kvm_area(void)
2596 {
2597         int cpu;
2598
2599         for_each_possible_cpu(cpu) {
2600                 struct vmcs *vmcs;
2601
2602                 vmcs = alloc_vmcs_cpu(cpu);
2603                 if (!vmcs) {
2604                         free_kvm_area();
2605                         return -ENOMEM;
2606                 }
2607
2608                 per_cpu(vmxarea, cpu) = vmcs;
2609         }
2610         return 0;
2611 }
2612
2613 static __init int hardware_setup(void)
2614 {
2615         if (setup_vmcs_config(&vmcs_config) < 0)
2616                 return -EIO;
2617
2618         if (boot_cpu_has(X86_FEATURE_NX))
2619                 kvm_enable_efer_bits(EFER_NX);
2620
2621         if (!cpu_has_vmx_vpid())
2622                 enable_vpid = 0;
2623
2624         if (!cpu_has_vmx_ept() ||
2625             !cpu_has_vmx_ept_4levels()) {
2626                 enable_ept = 0;
2627                 enable_unrestricted_guest = 0;
2628         }
2629
2630         if (!cpu_has_vmx_unrestricted_guest())
2631                 enable_unrestricted_guest = 0;
2632
2633         if (!cpu_has_vmx_flexpriority())
2634                 flexpriority_enabled = 0;
2635
2636         if (!cpu_has_vmx_tpr_shadow())
2637                 kvm_x86_ops->update_cr8_intercept = NULL;
2638
2639         if (enable_ept && !cpu_has_vmx_ept_2m_page())
2640                 kvm_disable_largepages();
2641
2642         if (!cpu_has_vmx_ple())
2643                 ple_gap = 0;
2644
2645         if (nested)
2646                 nested_vmx_setup_ctls_msrs();
2647
2648         return alloc_kvm_area();
2649 }
2650
2651 static __exit void hardware_unsetup(void)
2652 {
2653         free_kvm_area();
2654 }
2655
2656 static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
2657 {
2658         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2659
2660         if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
2661                 vmcs_write16(sf->selector, save->selector);
2662                 vmcs_writel(sf->base, save->base);
2663                 vmcs_write32(sf->limit, save->limit);
2664                 vmcs_write32(sf->ar_bytes, save->ar);
2665         } else {
2666                 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
2667                         << AR_DPL_SHIFT;
2668                 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
2669         }
2670 }
2671
2672 static void enter_pmode(struct kvm_vcpu *vcpu)
2673 {
2674         unsigned long flags;
2675         struct vcpu_vmx *vmx = to_vmx(vcpu);
2676
2677         vmx->emulation_required = 1;
2678         vmx->rmode.vm86_active = 0;
2679
2680         vmx_segment_cache_clear(vmx);
2681
2682         vmcs_write16(GUEST_TR_SELECTOR, vmx->rmode.tr.selector);
2683         vmcs_writel(GUEST_TR_BASE, vmx->rmode.tr.base);
2684         vmcs_write32(GUEST_TR_LIMIT, vmx->rmode.tr.limit);
2685         vmcs_write32(GUEST_TR_AR_BYTES, vmx->rmode.tr.ar);
2686
2687         flags = vmcs_readl(GUEST_RFLAGS);
2688         flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2689         flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2690         vmcs_writel(GUEST_RFLAGS, flags);
2691
2692         vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2693                         (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2694
2695         update_exception_bitmap(vcpu);
2696
2697         if (emulate_invalid_guest_state)
2698                 return;
2699
2700         fix_pmode_dataseg(VCPU_SREG_ES, &vmx->rmode.es);
2701         fix_pmode_dataseg(VCPU_SREG_DS, &vmx->rmode.ds);
2702         fix_pmode_dataseg(VCPU_SREG_GS, &vmx->rmode.gs);
2703         fix_pmode_dataseg(VCPU_SREG_FS, &vmx->rmode.fs);
2704
2705         vmx_segment_cache_clear(vmx);
2706
2707         vmcs_write16(GUEST_SS_SELECTOR, 0);
2708         vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
2709
2710         vmcs_write16(GUEST_CS_SELECTOR,
2711                      vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
2712         vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
2713 }
2714
2715 static gva_t rmode_tss_base(struct kvm *kvm)
2716 {
2717         if (!kvm->arch.tss_addr) {
2718                 struct kvm_memslots *slots;
2719                 gfn_t base_gfn;
2720
2721                 slots = kvm_memslots(kvm);
2722                 base_gfn = slots->memslots[0].base_gfn +
2723                                  kvm->memslots->memslots[0].npages - 3;
2724                 return base_gfn << PAGE_SHIFT;
2725         }
2726         return kvm->arch.tss_addr;
2727 }
2728
2729 static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
2730 {
2731         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2732
2733         save->selector = vmcs_read16(sf->selector);
2734         save->base = vmcs_readl(sf->base);
2735         save->limit = vmcs_read32(sf->limit);
2736         save->ar = vmcs_read32(sf->ar_bytes);
2737         vmcs_write16(sf->selector, save->base >> 4);
2738         vmcs_write32(sf->base, save->base & 0xffff0);
2739         vmcs_write32(sf->limit, 0xffff);
2740         vmcs_write32(sf->ar_bytes, 0xf3);
2741         if (save->base & 0xf)
2742                 printk_once(KERN_WARNING "kvm: segment base is not paragraph"
2743                             " aligned when entering protected mode (seg=%d)",
2744                             seg);
2745 }
2746
2747 static void enter_rmode(struct kvm_vcpu *vcpu)
2748 {
2749         unsigned long flags;
2750         struct vcpu_vmx *vmx = to_vmx(vcpu);
2751
2752         if (enable_unrestricted_guest)
2753                 return;
2754
2755         vmx->emulation_required = 1;
2756         vmx->rmode.vm86_active = 1;
2757
2758         /*
2759          * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2760          * vcpu. Call it here with phys address pointing 16M below 4G.
2761          */
2762         if (!vcpu->kvm->arch.tss_addr) {
2763                 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2764                              "called before entering vcpu\n");
2765                 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
2766                 vmx_set_tss_addr(vcpu->kvm, 0xfeffd000);
2767                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2768         }
2769
2770         vmx_segment_cache_clear(vmx);
2771
2772         vmx->rmode.tr.selector = vmcs_read16(GUEST_TR_SELECTOR);
2773         vmx->rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
2774         vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
2775
2776         vmx->rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
2777         vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2778
2779         vmx->rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
2780         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2781
2782         flags = vmcs_readl(GUEST_RFLAGS);
2783         vmx->rmode.save_rflags = flags;
2784
2785         flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2786
2787         vmcs_writel(GUEST_RFLAGS, flags);
2788         vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2789         update_exception_bitmap(vcpu);
2790
2791         if (emulate_invalid_guest_state)
2792                 goto continue_rmode;
2793
2794         vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
2795         vmcs_write32(GUEST_SS_LIMIT, 0xffff);
2796         vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
2797
2798         vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
2799         vmcs_write32(GUEST_CS_LIMIT, 0xffff);
2800         if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
2801                 vmcs_writel(GUEST_CS_BASE, 0xf0000);
2802         vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
2803
2804         fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.es);
2805         fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.ds);
2806         fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.gs);
2807         fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.fs);
2808
2809 continue_rmode:
2810         kvm_mmu_reset_context(vcpu);
2811 }
2812
2813 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2814 {
2815         struct vcpu_vmx *vmx = to_vmx(vcpu);
2816         struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2817
2818         if (!msr)
2819                 return;
2820
2821         /*
2822          * Force kernel_gs_base reloading before EFER changes, as control
2823          * of this msr depends on is_long_mode().
2824          */
2825         vmx_load_host_state(to_vmx(vcpu));
2826         vcpu->arch.efer = efer;
2827         if (efer & EFER_LMA) {
2828                 vmcs_write32(VM_ENTRY_CONTROLS,
2829                              vmcs_read32(VM_ENTRY_CONTROLS) |
2830                              VM_ENTRY_IA32E_MODE);
2831                 msr->data = efer;
2832         } else {
2833                 vmcs_write32(VM_ENTRY_CONTROLS,
2834                              vmcs_read32(VM_ENTRY_CONTROLS) &
2835                              ~VM_ENTRY_IA32E_MODE);
2836
2837                 msr->data = efer & ~EFER_LME;
2838         }
2839         setup_msrs(vmx);
2840 }
2841
2842 #ifdef CONFIG_X86_64
2843
2844 static void enter_lmode(struct kvm_vcpu *vcpu)
2845 {
2846         u32 guest_tr_ar;
2847
2848         vmx_segment_cache_clear(to_vmx(vcpu));
2849
2850         guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2851         if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
2852                 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
2853                                      __func__);
2854                 vmcs_write32(GUEST_TR_AR_BYTES,
2855                              (guest_tr_ar & ~AR_TYPE_MASK)
2856                              | AR_TYPE_BUSY_64_TSS);
2857         }
2858         vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2859 }
2860
2861 static void exit_lmode(struct kvm_vcpu *vcpu)
2862 {
2863         vmcs_write32(VM_ENTRY_CONTROLS,
2864                      vmcs_read32(VM_ENTRY_CONTROLS)
2865                      & ~VM_ENTRY_IA32E_MODE);
2866         vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2867 }
2868
2869 #endif
2870
2871 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
2872 {
2873         vpid_sync_context(to_vmx(vcpu));
2874         if (enable_ept) {
2875                 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
2876                         return;
2877                 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
2878         }
2879 }
2880
2881 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2882 {
2883         ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2884
2885         vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
2886         vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
2887 }
2888
2889 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
2890 {
2891         if (enable_ept && is_paging(vcpu))
2892                 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2893         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
2894 }
2895
2896 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2897 {
2898         ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2899
2900         vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
2901         vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
2902 }
2903
2904 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2905 {
2906         if (!test_bit(VCPU_EXREG_PDPTR,
2907                       (unsigned long *)&vcpu->arch.regs_dirty))
2908                 return;
2909
2910         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
2911                 vmcs_write64(GUEST_PDPTR0, vcpu->arch.mmu.pdptrs[0]);
2912                 vmcs_write64(GUEST_PDPTR1, vcpu->arch.mmu.pdptrs[1]);
2913                 vmcs_write64(GUEST_PDPTR2, vcpu->arch.mmu.pdptrs[2]);
2914                 vmcs_write64(GUEST_PDPTR3, vcpu->arch.mmu.pdptrs[3]);
2915         }
2916 }
2917
2918 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2919 {
2920         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
2921                 vcpu->arch.mmu.pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2922                 vcpu->arch.mmu.pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2923                 vcpu->arch.mmu.pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2924                 vcpu->arch.mmu.pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2925         }
2926
2927         __set_bit(VCPU_EXREG_PDPTR,
2928                   (unsigned long *)&vcpu->arch.regs_avail);
2929         __set_bit(VCPU_EXREG_PDPTR,
2930                   (unsigned long *)&vcpu->arch.regs_dirty);
2931 }
2932
2933 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
2934
2935 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
2936                                         unsigned long cr0,
2937                                         struct kvm_vcpu *vcpu)
2938 {
2939         if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
2940                 vmx_decache_cr3(vcpu);
2941         if (!(cr0 & X86_CR0_PG)) {
2942                 /* From paging/starting to nonpaging */
2943                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
2944                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
2945                              (CPU_BASED_CR3_LOAD_EXITING |
2946                               CPU_BASED_CR3_STORE_EXITING));
2947                 vcpu->arch.cr0 = cr0;
2948                 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2949         } else if (!is_paging(vcpu)) {
2950                 /* From nonpaging to paging */
2951                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
2952                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
2953                              ~(CPU_BASED_CR3_LOAD_EXITING |
2954                                CPU_BASED_CR3_STORE_EXITING));
2955                 vcpu->arch.cr0 = cr0;
2956                 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2957         }
2958
2959         if (!(cr0 & X86_CR0_WP))
2960                 *hw_cr0 &= ~X86_CR0_WP;
2961 }
2962
2963 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2964 {
2965         struct vcpu_vmx *vmx = to_vmx(vcpu);
2966         unsigned long hw_cr0;
2967
2968         if (enable_unrestricted_guest)
2969                 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST)
2970                         | KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
2971         else
2972                 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON;
2973
2974         if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
2975                 enter_pmode(vcpu);
2976
2977         if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
2978                 enter_rmode(vcpu);
2979
2980 #ifdef CONFIG_X86_64
2981         if (vcpu->arch.efer & EFER_LME) {
2982                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
2983                         enter_lmode(vcpu);
2984                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
2985                         exit_lmode(vcpu);
2986         }
2987 #endif
2988
2989         if (enable_ept)
2990                 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
2991
2992         if (!vcpu->fpu_active)
2993                 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
2994
2995         vmcs_writel(CR0_READ_SHADOW, cr0);
2996         vmcs_writel(GUEST_CR0, hw_cr0);
2997         vcpu->arch.cr0 = cr0;
2998         __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
2999 }
3000
3001 static u64 construct_eptp(unsigned long root_hpa)
3002 {
3003         u64 eptp;
3004
3005         /* TODO write the value reading from MSR */
3006         eptp = VMX_EPT_DEFAULT_MT |
3007                 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
3008         eptp |= (root_hpa & PAGE_MASK);
3009
3010         return eptp;
3011 }
3012
3013 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
3014 {
3015         unsigned long guest_cr3;
3016         u64 eptp;
3017
3018         guest_cr3 = cr3;
3019         if (enable_ept) {
3020                 eptp = construct_eptp(cr3);
3021                 vmcs_write64(EPT_POINTER, eptp);
3022                 guest_cr3 = is_paging(vcpu) ? kvm_read_cr3(vcpu) :
3023                         vcpu->kvm->arch.ept_identity_map_addr;
3024                 ept_load_pdptrs(vcpu);
3025         }
3026
3027         vmx_flush_tlb(vcpu);
3028         vmcs_writel(GUEST_CR3, guest_cr3);
3029 }
3030
3031 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3032 {
3033         unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
3034                     KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
3035
3036         if (cr4 & X86_CR4_VMXE) {
3037                 /*
3038                  * To use VMXON (and later other VMX instructions), a guest
3039                  * must first be able to turn on cr4.VMXE (see handle_vmon()).
3040                  * So basically the check on whether to allow nested VMX
3041                  * is here.
3042                  */
3043                 if (!nested_vmx_allowed(vcpu))
3044                         return 1;
3045         } else if (to_vmx(vcpu)->nested.vmxon)
3046                 return 1;
3047
3048         vcpu->arch.cr4 = cr4;
3049         if (enable_ept) {
3050                 if (!is_paging(vcpu)) {
3051                         hw_cr4 &= ~X86_CR4_PAE;
3052                         hw_cr4 |= X86_CR4_PSE;
3053                 } else if (!(cr4 & X86_CR4_PAE)) {
3054                         hw_cr4 &= ~X86_CR4_PAE;
3055                 }
3056         }
3057
3058         vmcs_writel(CR4_READ_SHADOW, cr4);
3059         vmcs_writel(GUEST_CR4, hw_cr4);
3060         return 0;
3061 }
3062
3063 static void vmx_get_segment(struct kvm_vcpu *vcpu,
3064                             struct kvm_segment *var, int seg)
3065 {
3066         struct vcpu_vmx *vmx = to_vmx(vcpu);
3067         struct kvm_save_segment *save;
3068         u32 ar;
3069
3070         if (vmx->rmode.vm86_active
3071             && (seg == VCPU_SREG_TR || seg == VCPU_SREG_ES
3072                 || seg == VCPU_SREG_DS || seg == VCPU_SREG_FS
3073                 || seg == VCPU_SREG_GS)
3074             && !emulate_invalid_guest_state) {
3075                 switch (seg) {
3076                 case VCPU_SREG_TR: save = &vmx->rmode.tr; break;
3077                 case VCPU_SREG_ES: save = &vmx->rmode.es; break;
3078                 case VCPU_SREG_DS: save = &vmx->rmode.ds; break;
3079                 case VCPU_SREG_FS: save = &vmx->rmode.fs; break;
3080                 case VCPU_SREG_GS: save = &vmx->rmode.gs; break;
3081                 default: BUG();
3082                 }
3083                 var->selector = save->selector;
3084                 var->base = save->base;
3085                 var->limit = save->limit;
3086                 ar = save->ar;
3087                 if (seg == VCPU_SREG_TR
3088                     || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3089                         goto use_saved_rmode_seg;
3090         }
3091         var->base = vmx_read_guest_seg_base(vmx, seg);
3092         var->limit = vmx_read_guest_seg_limit(vmx, seg);
3093         var->selector = vmx_read_guest_seg_selector(vmx, seg);
3094         ar = vmx_read_guest_seg_ar(vmx, seg);
3095 use_saved_rmode_seg:
3096         if ((ar & AR_UNUSABLE_MASK) && !emulate_invalid_guest_state)
3097                 ar = 0;
3098         var->type = ar & 15;
3099         var->s = (ar >> 4) & 1;
3100         var->dpl = (ar >> 5) & 3;
3101         var->present = (ar >> 7) & 1;
3102         var->avl = (ar >> 12) & 1;
3103         var->l = (ar >> 13) & 1;
3104         var->db = (ar >> 14) & 1;
3105         var->g = (ar >> 15) & 1;
3106         var->unusable = (ar >> 16) & 1;
3107 }
3108
3109 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3110 {
3111         struct kvm_segment s;
3112
3113         if (to_vmx(vcpu)->rmode.vm86_active) {
3114                 vmx_get_segment(vcpu, &s, seg);
3115                 return s.base;
3116         }
3117         return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3118 }
3119
3120 static int __vmx_get_cpl(struct kvm_vcpu *vcpu)
3121 {
3122         if (!is_protmode(vcpu))
3123                 return 0;
3124
3125         if (!is_long_mode(vcpu)
3126             && (kvm_get_rflags(vcpu) & X86_EFLAGS_VM)) /* if virtual 8086 */
3127                 return 3;
3128
3129         return vmx_read_guest_seg_selector(to_vmx(vcpu), VCPU_SREG_CS) & 3;
3130 }
3131
3132 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
3133 {
3134         if (!test_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail)) {
3135                 __set_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
3136                 to_vmx(vcpu)->cpl = __vmx_get_cpl(vcpu);
3137         }
3138         return to_vmx(vcpu)->cpl;
3139 }
3140
3141
3142 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3143 {
3144         u32 ar;
3145
3146         if (var->unusable)
3147                 ar = 1 << 16;
3148         else {
3149                 ar = var->type & 15;
3150                 ar |= (var->s & 1) << 4;
3151                 ar |= (var->dpl & 3) << 5;
3152                 ar |= (var->present & 1) << 7;
3153                 ar |= (var->avl & 1) << 12;
3154                 ar |= (var->l & 1) << 13;
3155                 ar |= (var->db & 1) << 14;
3156                 ar |= (var->g & 1) << 15;
3157         }
3158         if (ar == 0) /* a 0 value means unusable */
3159                 ar = AR_UNUSABLE_MASK;
3160
3161         return ar;
3162 }
3163
3164 static void vmx_set_segment(struct kvm_vcpu *vcpu,
3165                             struct kvm_segment *var, int seg)
3166 {
3167         struct vcpu_vmx *vmx = to_vmx(vcpu);
3168         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3169         u32 ar;
3170
3171         vmx_segment_cache_clear(vmx);
3172
3173         if (vmx->rmode.vm86_active && seg == VCPU_SREG_TR) {
3174                 vmcs_write16(sf->selector, var->selector);
3175                 vmx->rmode.tr.selector = var->selector;
3176                 vmx->rmode.tr.base = var->base;
3177                 vmx->rmode.tr.limit = var->limit;
3178                 vmx->rmode.tr.ar = vmx_segment_access_rights(var);
3179                 return;
3180         }
3181         vmcs_writel(sf->base, var->base);
3182         vmcs_write32(sf->limit, var->limit);
3183         vmcs_write16(sf->selector, var->selector);
3184         if (vmx->rmode.vm86_active && var->s) {
3185                 /*
3186                  * Hack real-mode segments into vm86 compatibility.
3187                  */
3188                 if (var->base == 0xffff0000 && var->selector == 0xf000)
3189                         vmcs_writel(sf->base, 0xf0000);
3190                 ar = 0xf3;
3191         } else
3192                 ar = vmx_segment_access_rights(var);
3193
3194         /*
3195          *   Fix the "Accessed" bit in AR field of segment registers for older
3196          * qemu binaries.
3197          *   IA32 arch specifies that at the time of processor reset the
3198          * "Accessed" bit in the AR field of segment registers is 1. And qemu
3199          * is setting it to 0 in the usedland code. This causes invalid guest
3200          * state vmexit when "unrestricted guest" mode is turned on.
3201          *    Fix for this setup issue in cpu_reset is being pushed in the qemu
3202          * tree. Newer qemu binaries with that qemu fix would not need this
3203          * kvm hack.
3204          */
3205         if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3206                 ar |= 0x1; /* Accessed */
3207
3208         vmcs_write32(sf->ar_bytes, ar);
3209         __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
3210 }
3211
3212 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3213 {
3214         u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3215
3216         *db = (ar >> 14) & 1;
3217         *l = (ar >> 13) & 1;
3218 }
3219
3220 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3221 {
3222         dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3223         dt->address = vmcs_readl(GUEST_IDTR_BASE);
3224 }
3225
3226 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3227 {
3228         vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3229         vmcs_writel(GUEST_IDTR_BASE, dt->address);
3230 }
3231
3232 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3233 {
3234         dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3235         dt->address = vmcs_readl(GUEST_GDTR_BASE);
3236 }
3237
3238 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3239 {
3240         vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3241         vmcs_writel(GUEST_GDTR_BASE, dt->address);
3242 }
3243
3244 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3245 {
3246         struct kvm_segment var;
3247         u32 ar;
3248
3249         vmx_get_segment(vcpu, &var, seg);
3250         ar = vmx_segment_access_rights(&var);
3251
3252         if (var.base != (var.selector << 4))
3253                 return false;
3254         if (var.limit != 0xffff)
3255                 return false;
3256         if (ar != 0xf3)
3257                 return false;
3258
3259         return true;
3260 }
3261
3262 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3263 {
3264         struct kvm_segment cs;
3265         unsigned int cs_rpl;
3266
3267         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3268         cs_rpl = cs.selector & SELECTOR_RPL_MASK;
3269
3270         if (cs.unusable)
3271                 return false;
3272         if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
3273                 return false;
3274         if (!cs.s)
3275                 return false;
3276         if (cs.type & AR_TYPE_WRITEABLE_MASK) {
3277                 if (cs.dpl > cs_rpl)
3278                         return false;
3279         } else {
3280                 if (cs.dpl != cs_rpl)
3281                         return false;
3282         }
3283         if (!cs.present)
3284                 return false;
3285
3286         /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3287         return true;
3288 }
3289
3290 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3291 {
3292         struct kvm_segment ss;
3293         unsigned int ss_rpl;
3294
3295         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3296         ss_rpl = ss.selector & SELECTOR_RPL_MASK;
3297
3298         if (ss.unusable)
3299                 return true;
3300         if (ss.type != 3 && ss.type != 7)
3301                 return false;
3302         if (!ss.s)
3303                 return false;
3304         if (ss.dpl != ss_rpl) /* DPL != RPL */
3305                 return false;
3306         if (!ss.present)
3307                 return false;
3308
3309         return true;
3310 }
3311
3312 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3313 {
3314         struct kvm_segment var;
3315         unsigned int rpl;
3316
3317         vmx_get_segment(vcpu, &var, seg);
3318         rpl = var.selector & SELECTOR_RPL_MASK;
3319
3320         if (var.unusable)
3321                 return true;
3322         if (!var.s)
3323                 return false;
3324         if (!var.present)
3325                 return false;
3326         if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
3327                 if (var.dpl < rpl) /* DPL < RPL */
3328                         return false;
3329         }
3330
3331         /* TODO: Add other members to kvm_segment_field to allow checking for other access
3332          * rights flags
3333          */
3334         return true;
3335 }
3336
3337 static bool tr_valid(struct kvm_vcpu *vcpu)
3338 {
3339         struct kvm_segment tr;
3340
3341         vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3342
3343         if (tr.unusable)
3344                 return false;
3345         if (tr.selector & SELECTOR_TI_MASK)     /* TI = 1 */
3346                 return false;
3347         if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3348                 return false;
3349         if (!tr.present)
3350                 return false;
3351
3352         return true;
3353 }
3354
3355 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3356 {
3357         struct kvm_segment ldtr;
3358
3359         vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3360
3361         if (ldtr.unusable)
3362                 return true;
3363         if (ldtr.selector & SELECTOR_TI_MASK)   /* TI = 1 */
3364                 return false;
3365         if (ldtr.type != 2)
3366                 return false;
3367         if (!ldtr.present)
3368                 return false;
3369
3370         return true;
3371 }
3372
3373 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3374 {
3375         struct kvm_segment cs, ss;
3376
3377         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3378         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3379
3380         return ((cs.selector & SELECTOR_RPL_MASK) ==
3381                  (ss.selector & SELECTOR_RPL_MASK));
3382 }
3383
3384 /*
3385  * Check if guest state is valid. Returns true if valid, false if
3386  * not.
3387  * We assume that registers are always usable
3388  */
3389 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3390 {
3391         /* real mode guest state checks */
3392         if (!is_protmode(vcpu)) {
3393                 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3394                         return false;
3395                 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3396                         return false;
3397                 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3398                         return false;
3399                 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3400                         return false;
3401                 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3402                         return false;
3403                 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3404                         return false;
3405         } else {
3406         /* protected mode guest state checks */
3407                 if (!cs_ss_rpl_check(vcpu))
3408                         return false;
3409                 if (!code_segment_valid(vcpu))
3410                         return false;
3411                 if (!stack_segment_valid(vcpu))
3412                         return false;
3413                 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3414                         return false;
3415                 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3416                         return false;
3417                 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3418                         return false;
3419                 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3420                         return false;
3421                 if (!tr_valid(vcpu))
3422                         return false;
3423                 if (!ldtr_valid(vcpu))
3424                         return false;
3425         }
3426         /* TODO:
3427          * - Add checks on RIP
3428          * - Add checks on RFLAGS
3429          */
3430
3431         return true;
3432 }
3433
3434 static int init_rmode_tss(struct kvm *kvm)
3435 {
3436         gfn_t fn;
3437         u16 data = 0;
3438         int r, idx, ret = 0;
3439
3440         idx = srcu_read_lock(&kvm->srcu);
3441         fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
3442         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3443         if (r < 0)
3444                 goto out;
3445         data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3446         r = kvm_write_guest_page(kvm, fn++, &data,
3447                         TSS_IOPB_BASE_OFFSET, sizeof(u16));
3448         if (r < 0)
3449                 goto out;
3450         r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3451         if (r < 0)
3452                 goto out;
3453         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3454         if (r < 0)
3455                 goto out;
3456         data = ~0;
3457         r = kvm_write_guest_page(kvm, fn, &data,
3458                                  RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3459                                  sizeof(u8));
3460         if (r < 0)
3461                 goto out;
3462
3463         ret = 1;
3464 out:
3465         srcu_read_unlock(&kvm->srcu, idx);
3466         return ret;
3467 }
3468
3469 static int init_rmode_identity_map(struct kvm *kvm)
3470 {
3471         int i, idx, r, ret;
3472         pfn_t identity_map_pfn;
3473         u32 tmp;
3474
3475         if (!enable_ept)
3476                 return 1;
3477         if (unlikely(!kvm->arch.ept_identity_pagetable)) {
3478                 printk(KERN_ERR "EPT: identity-mapping pagetable "
3479                         "haven't been allocated!\n");
3480                 return 0;
3481         }
3482         if (likely(kvm->arch.ept_identity_pagetable_done))
3483                 return 1;
3484         ret = 0;
3485         identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
3486         idx = srcu_read_lock(&kvm->srcu);
3487         r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3488         if (r < 0)
3489                 goto out;
3490         /* Set up identity-mapping pagetable for EPT in real mode */
3491         for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3492                 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3493                         _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3494                 r = kvm_write_guest_page(kvm, identity_map_pfn,
3495                                 &tmp, i * sizeof(tmp), sizeof(tmp));
3496                 if (r < 0)
3497                         goto out;
3498         }
3499         kvm->arch.ept_identity_pagetable_done = true;
3500         ret = 1;
3501 out:
3502         srcu_read_unlock(&kvm->srcu, idx);
3503         return ret;
3504 }
3505
3506 static void seg_setup(int seg)
3507 {
3508         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3509         unsigned int ar;
3510
3511         vmcs_write16(sf->selector, 0);
3512         vmcs_writel(sf->base, 0);
3513         vmcs_write32(sf->limit, 0xffff);
3514         if (enable_unrestricted_guest) {
3515                 ar = 0x93;
3516                 if (seg == VCPU_SREG_CS)
3517                         ar |= 0x08; /* code segment */
3518         } else
3519                 ar = 0xf3;
3520
3521         vmcs_write32(sf->ar_bytes, ar);
3522 }
3523
3524 static int alloc_apic_access_page(struct kvm *kvm)
3525 {
3526         struct kvm_userspace_memory_region kvm_userspace_mem;
3527         int r = 0;
3528
3529         mutex_lock(&kvm->slots_lock);
3530         if (kvm->arch.apic_access_page)
3531                 goto out;
3532         kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
3533         kvm_userspace_mem.flags = 0;
3534         kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
3535         kvm_userspace_mem.memory_size = PAGE_SIZE;
3536         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
3537         if (r)
3538                 goto out;
3539
3540         kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
3541 out:
3542         mutex_unlock(&kvm->slots_lock);
3543         return r;
3544 }
3545
3546 static int alloc_identity_pagetable(struct kvm *kvm)
3547 {
3548         struct kvm_userspace_memory_region kvm_userspace_mem;
3549         int r = 0;
3550
3551         mutex_lock(&kvm->slots_lock);
3552         if (kvm->arch.ept_identity_pagetable)
3553                 goto out;
3554         kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
3555         kvm_userspace_mem.flags = 0;
3556         kvm_userspace_mem.guest_phys_addr =
3557                 kvm->arch.ept_identity_map_addr;
3558         kvm_userspace_mem.memory_size = PAGE_SIZE;
3559         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
3560         if (r)
3561                 goto out;
3562
3563         kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
3564                         kvm->arch.ept_identity_map_addr >> PAGE_SHIFT);
3565 out:
3566         mutex_unlock(&kvm->slots_lock);
3567         return r;
3568 }
3569
3570 static void allocate_vpid(struct vcpu_vmx *vmx)
3571 {
3572         int vpid;
3573
3574         vmx->vpid = 0;
3575         if (!enable_vpid)
3576                 return;
3577         spin_lock(&vmx_vpid_lock);
3578         vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3579         if (vpid < VMX_NR_VPIDS) {
3580                 vmx->vpid = vpid;
3581                 __set_bit(vpid, vmx_vpid_bitmap);
3582         }
3583         spin_unlock(&vmx_vpid_lock);
3584 }
3585
3586 static void free_vpid(struct vcpu_vmx *vmx)
3587 {
3588         if (!enable_vpid)
3589                 return;
3590         spin_lock(&vmx_vpid_lock);
3591         if (vmx->vpid != 0)
3592                 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
3593         spin_unlock(&vmx_vpid_lock);
3594 }
3595
3596 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, u32 msr)
3597 {
3598         int f = sizeof(unsigned long);
3599
3600         if (!cpu_has_vmx_msr_bitmap())
3601                 return;
3602
3603         /*
3604          * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3605          * have the write-low and read-high bitmap offsets the wrong way round.
3606          * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3607          */
3608         if (msr <= 0x1fff) {
3609                 __clear_bit(msr, msr_bitmap + 0x000 / f); /* read-low */
3610                 __clear_bit(msr, msr_bitmap + 0x800 / f); /* write-low */
3611         } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3612                 msr &= 0x1fff;
3613                 __clear_bit(msr, msr_bitmap + 0x400 / f); /* read-high */
3614                 __clear_bit(msr, msr_bitmap + 0xc00 / f); /* write-high */
3615         }
3616 }
3617
3618 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
3619 {
3620         if (!longmode_only)
3621                 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy, msr);
3622         __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode, msr);
3623 }
3624
3625 /*
3626  * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3627  * will not change in the lifetime of the guest.
3628  * Note that host-state that does change is set elsewhere. E.g., host-state
3629  * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3630  */
3631 static void vmx_set_constant_host_state(void)
3632 {
3633         u32 low32, high32;
3634         unsigned long tmpl;
3635         struct desc_ptr dt;
3636
3637         vmcs_writel(HOST_CR0, read_cr0() | X86_CR0_TS);  /* 22.2.3 */
3638         vmcs_writel(HOST_CR4, read_cr4());  /* 22.2.3, 22.2.5 */
3639         vmcs_writel(HOST_CR3, read_cr3());  /* 22.2.3  FIXME: shadow tables */
3640
3641         vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
3642         vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3643         vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3644         vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3645         vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
3646
3647         native_store_idt(&dt);
3648         vmcs_writel(HOST_IDTR_BASE, dt.address);   /* 22.2.4 */
3649
3650         asm("mov $.Lkvm_vmx_return, %0" : "=r"(tmpl));
3651         vmcs_writel(HOST_RIP, tmpl); /* 22.2.5 */
3652
3653         rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
3654         vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
3655         rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
3656         vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl);   /* 22.2.3 */
3657
3658         if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
3659                 rdmsr(MSR_IA32_CR_PAT, low32, high32);
3660                 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
3661         }
3662 }
3663
3664 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
3665 {
3666         vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
3667         if (enable_ept)
3668                 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
3669         if (is_guest_mode(&vmx->vcpu))
3670                 vmx->vcpu.arch.cr4_guest_owned_bits &=
3671                         ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
3672         vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
3673 }
3674
3675 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
3676 {
3677         u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
3678         if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
3679                 exec_control &= ~CPU_BASED_TPR_SHADOW;
3680 #ifdef CONFIG_X86_64
3681                 exec_control |= CPU_BASED_CR8_STORE_EXITING |
3682                                 CPU_BASED_CR8_LOAD_EXITING;
3683 #endif
3684         }
3685         if (!enable_ept)
3686                 exec_control |= CPU_BASED_CR3_STORE_EXITING |
3687                                 CPU_BASED_CR3_LOAD_EXITING  |
3688                                 CPU_BASED_INVLPG_EXITING;
3689         return exec_control;
3690 }
3691
3692 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
3693 {
3694         u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
3695         if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
3696                 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
3697         if (vmx->vpid == 0)
3698                 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
3699         if (!enable_ept) {
3700                 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
3701                 enable_unrestricted_guest = 0;
3702         }
3703         if (!enable_unrestricted_guest)
3704                 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
3705         if (!ple_gap)
3706                 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
3707         return exec_control;
3708 }
3709
3710 static void ept_set_mmio_spte_mask(void)
3711 {
3712         /*
3713          * EPT Misconfigurations can be generated if the value of bits 2:0
3714          * of an EPT paging-structure entry is 110b (write/execute).
3715          * Also, magic bits (0xffull << 49) is set to quickly identify mmio
3716          * spte.
3717          */
3718         kvm_mmu_set_mmio_spte_mask(0xffull << 49 | 0x6ull);
3719 }
3720
3721 /*
3722  * Sets up the vmcs for emulated real mode.
3723  */
3724 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
3725 {
3726 #ifdef CONFIG_X86_64
3727         unsigned long a;
3728 #endif
3729         int i;
3730
3731         /* I/O */
3732         vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
3733         vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
3734
3735         if (cpu_has_vmx_msr_bitmap())
3736                 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
3737
3738         vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
3739
3740         /* Control */
3741         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
3742                 vmcs_config.pin_based_exec_ctrl);
3743
3744         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
3745
3746         if (cpu_has_secondary_exec_ctrls()) {
3747                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
3748                                 vmx_secondary_exec_control(vmx));
3749         }
3750
3751         if (ple_gap) {
3752                 vmcs_write32(PLE_GAP, ple_gap);
3753                 vmcs_write32(PLE_WINDOW, ple_window);
3754         }
3755
3756         vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
3757         vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
3758         vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
3759
3760         vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
3761         vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
3762         vmx_set_constant_host_state();
3763 #ifdef CONFIG_X86_64
3764         rdmsrl(MSR_FS_BASE, a);
3765         vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
3766         rdmsrl(MSR_GS_BASE, a);
3767         vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
3768 #else
3769         vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
3770         vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
3771 #endif
3772
3773         vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
3774         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
3775         vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
3776         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
3777         vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
3778
3779         if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
3780                 u32 msr_low, msr_high;
3781                 u64 host_pat;
3782                 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
3783                 host_pat = msr_low | ((u64) msr_high << 32);
3784                 /* Write the default value follow host pat */
3785                 vmcs_write64(GUEST_IA32_PAT, host_pat);
3786                 /* Keep arch.pat sync with GUEST_IA32_PAT */
3787                 vmx->vcpu.arch.pat = host_pat;
3788         }
3789
3790         for (i = 0; i < NR_VMX_MSR; ++i) {
3791                 u32 index = vmx_msr_index[i];
3792                 u32 data_low, data_high;
3793                 int j = vmx->nmsrs;
3794
3795                 if (rdmsr_safe(index, &data_low, &data_high) < 0)
3796                         continue;
3797                 if (wrmsr_safe(index, data_low, data_high) < 0)
3798                         continue;
3799                 vmx->guest_msrs[j].index = i;
3800                 vmx->guest_msrs[j].data = 0;
3801                 vmx->guest_msrs[j].mask = -1ull;
3802                 ++vmx->nmsrs;
3803         }
3804
3805         vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
3806
3807         /* 22.2.1, 20.8.1 */
3808         vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
3809
3810         vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
3811         set_cr4_guest_host_mask(vmx);
3812
3813         kvm_write_tsc(&vmx->vcpu, 0);
3814
3815         return 0;
3816 }
3817
3818 static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
3819 {
3820         struct vcpu_vmx *vmx = to_vmx(vcpu);
3821         u64 msr;
3822         int ret;
3823
3824         vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
3825
3826         vmx->rmode.vm86_active = 0;
3827
3828         vmx->soft_vnmi_blocked = 0;
3829
3830         vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
3831         kvm_set_cr8(&vmx->vcpu, 0);
3832         msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
3833         if (kvm_vcpu_is_bsp(&vmx->vcpu))
3834                 msr |= MSR_IA32_APICBASE_BSP;
3835         kvm_set_apic_base(&vmx->vcpu, msr);
3836
3837         ret = fx_init(&vmx->vcpu);
3838         if (ret != 0)
3839                 goto out;
3840
3841         vmx_segment_cache_clear(vmx);
3842
3843         seg_setup(VCPU_SREG_CS);
3844         /*
3845          * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
3846          * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4.  Sigh.
3847          */
3848         if (kvm_vcpu_is_bsp(&vmx->vcpu)) {
3849                 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
3850                 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
3851         } else {
3852                 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
3853                 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
3854         }
3855
3856         seg_setup(VCPU_SREG_DS);
3857         seg_setup(VCPU_SREG_ES);
3858         seg_setup(VCPU_SREG_FS);
3859         seg_setup(VCPU_SREG_GS);
3860         seg_setup(VCPU_SREG_SS);
3861
3862         vmcs_write16(GUEST_TR_SELECTOR, 0);
3863         vmcs_writel(GUEST_TR_BASE, 0);
3864         vmcs_write32(GUEST_TR_LIMIT, 0xffff);
3865         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3866
3867         vmcs_write16(GUEST_LDTR_SELECTOR, 0);
3868         vmcs_writel(GUEST_LDTR_BASE, 0);
3869         vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
3870         vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
3871
3872         vmcs_write32(GUEST_SYSENTER_CS, 0);
3873         vmcs_writel(GUEST_SYSENTER_ESP, 0);
3874         vmcs_writel(GUEST_SYSENTER_EIP, 0);
3875
3876         vmcs_writel(GUEST_RFLAGS, 0x02);
3877         if (kvm_vcpu_is_bsp(&vmx->vcpu))
3878                 kvm_rip_write(vcpu, 0xfff0);
3879         else
3880                 kvm_rip_write(vcpu, 0);
3881         kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
3882
3883         vmcs_writel(GUEST_DR7, 0x400);
3884
3885         vmcs_writel(GUEST_GDTR_BASE, 0);
3886         vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
3887
3888         vmcs_writel(GUEST_IDTR_BASE, 0);
3889         vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
3890
3891         vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
3892         vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
3893         vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
3894
3895         /* Special registers */
3896         vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
3897
3898         setup_msrs(vmx);
3899
3900         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
3901
3902         if (cpu_has_vmx_tpr_shadow()) {
3903                 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
3904                 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
3905                         vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
3906                                      __pa(vmx->vcpu.arch.apic->regs));
3907                 vmcs_write32(TPR_THRESHOLD, 0);
3908         }
3909
3910         if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
3911                 vmcs_write64(APIC_ACCESS_ADDR,
3912                              page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
3913
3914         if (vmx->vpid != 0)
3915                 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
3916
3917         vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
3918         vmx_set_cr0(&vmx->vcpu, kvm_read_cr0(vcpu)); /* enter rmode */
3919         vmx_set_cr4(&vmx->vcpu, 0);
3920         vmx_set_efer(&vmx->vcpu, 0);
3921         vmx_fpu_activate(&vmx->vcpu);
3922         update_exception_bitmap(&vmx->vcpu);
3923
3924         vpid_sync_context(vmx);
3925
3926         ret = 0;
3927
3928         /* HACK: Don't enable emulation on guest boot/reset */
3929         vmx->emulation_required = 0;
3930
3931 out:
3932         return ret;
3933 }
3934
3935 /*
3936  * In nested virtualization, check if L1 asked to exit on external interrupts.
3937  * For most existing hypervisors, this will always return true.
3938  */
3939 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
3940 {
3941         return get_vmcs12(vcpu)->pin_based_vm_exec_control &
3942                 PIN_BASED_EXT_INTR_MASK;
3943 }
3944
3945 static void enable_irq_window(struct kvm_vcpu *vcpu)
3946 {
3947         u32 cpu_based_vm_exec_control;
3948         if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu))
3949                 /* We can get here when nested_run_pending caused
3950                  * vmx_interrupt_allowed() to return false. In this case, do
3951                  * nothing - the interrupt will be injected later.
3952                  */
3953                 return;
3954
3955         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3956         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
3957         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3958 }
3959
3960 static void enable_nmi_window(struct kvm_vcpu *vcpu)
3961 {
3962         u32 cpu_based_vm_exec_control;
3963
3964         if (!cpu_has_virtual_nmis()) {
3965                 enable_irq_window(vcpu);
3966                 return;
3967         }
3968
3969         if (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
3970                 enable_irq_window(vcpu);
3971                 return;
3972         }
3973         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3974         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
3975         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3976 }
3977
3978 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
3979 {
3980         struct vcpu_vmx *vmx = to_vmx(vcpu);
3981         uint32_t intr;
3982         int irq = vcpu->arch.interrupt.nr;
3983
3984         trace_kvm_inj_virq(irq);
3985
3986         ++vcpu->stat.irq_injections;
3987         if (vmx->rmode.vm86_active) {
3988                 int inc_eip = 0;
3989                 if (vcpu->arch.interrupt.soft)
3990                         inc_eip = vcpu->arch.event_exit_inst_len;
3991                 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
3992                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3993                 return;
3994         }
3995         intr = irq | INTR_INFO_VALID_MASK;
3996         if (vcpu->arch.interrupt.soft) {
3997                 intr |= INTR_TYPE_SOFT_INTR;
3998                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
3999                              vmx->vcpu.arch.event_exit_inst_len);
4000         } else
4001                 intr |= INTR_TYPE_EXT_INTR;
4002         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4003         vmx_clear_hlt(vcpu);
4004 }
4005
4006 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4007 {
4008         struct vcpu_vmx *vmx = to_vmx(vcpu);
4009
4010         if (is_guest_mode(vcpu))
4011                 return;
4012
4013         if (!cpu_has_virtual_nmis()) {
4014                 /*
4015                  * Tracking the NMI-blocked state in software is built upon
4016                  * finding the next open IRQ window. This, in turn, depends on
4017                  * well-behaving guests: They have to keep IRQs disabled at
4018                  * least as long as the NMI handler runs. Otherwise we may
4019                  * cause NMI nesting, maybe breaking the guest. But as this is
4020                  * highly unlikely, we can live with the residual risk.
4021                  */
4022                 vmx->soft_vnmi_blocked = 1;
4023                 vmx->vnmi_blocked_time = 0;
4024         }
4025
4026         ++vcpu->stat.nmi_injections;
4027         vmx->nmi_known_unmasked = false;
4028         if (vmx->rmode.vm86_active) {
4029                 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
4030                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4031                 return;
4032         }
4033         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4034                         INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4035         vmx_clear_hlt(vcpu);
4036 }
4037
4038 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4039 {
4040         if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
4041                 return 0;
4042
4043         return  !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4044                   (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4045                    | GUEST_INTR_STATE_NMI));
4046 }
4047
4048 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4049 {
4050         if (!cpu_has_virtual_nmis())
4051                 return to_vmx(vcpu)->soft_vnmi_blocked;
4052         if (to_vmx(vcpu)->nmi_known_unmasked)
4053                 return false;
4054         return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4055 }
4056
4057 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4058 {
4059         struct vcpu_vmx *vmx = to_vmx(vcpu);
4060
4061         if (!cpu_has_virtual_nmis()) {
4062                 if (vmx->soft_vnmi_blocked != masked) {
4063                         vmx->soft_vnmi_blocked = masked;
4064                         vmx->vnmi_blocked_time = 0;
4065                 }
4066         } else {
4067                 vmx->nmi_known_unmasked = !masked;
4068                 if (masked)
4069                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4070                                       GUEST_INTR_STATE_NMI);
4071                 else
4072                         vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4073                                         GUEST_INTR_STATE_NMI);
4074         }
4075 }
4076
4077 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4078 {
4079         if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu)) {
4080                 struct vmcs12 *vmcs12;
4081                 if (to_vmx(vcpu)->nested.nested_run_pending)
4082                         return 0;
4083                 nested_vmx_vmexit(vcpu);
4084                 vmcs12 = get_vmcs12(vcpu);
4085                 vmcs12->vm_exit_reason = EXIT_REASON_EXTERNAL_INTERRUPT;
4086                 vmcs12->vm_exit_intr_info = 0;
4087                 /* fall through to normal code, but now in L1, not L2 */
4088         }
4089
4090         return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4091                 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4092                         (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4093 }
4094
4095 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4096 {
4097         int ret;
4098         struct kvm_userspace_memory_region tss_mem = {
4099                 .slot = TSS_PRIVATE_MEMSLOT,
4100                 .guest_phys_addr = addr,
4101                 .memory_size = PAGE_SIZE * 3,
4102                 .flags = 0,
4103         };
4104
4105         ret = kvm_set_memory_region(kvm, &tss_mem, 0);
4106         if (ret)
4107                 return ret;
4108         kvm->arch.tss_addr = addr;
4109         if (!init_rmode_tss(kvm))
4110                 return  -ENOMEM;
4111
4112         return 0;
4113 }
4114
4115 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4116                                   int vec, u32 err_code)
4117 {
4118         /*
4119          * Instruction with address size override prefix opcode 0x67
4120          * Cause the #SS fault with 0 error code in VM86 mode.
4121          */
4122         if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
4123                 if (emulate_instruction(vcpu, 0) == EMULATE_DONE)
4124                         return 1;
4125         /*
4126          * Forward all other exceptions that are valid in real mode.
4127          * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4128          *        the required debugging infrastructure rework.
4129          */
4130         switch (vec) {
4131         case DB_VECTOR:
4132                 if (vcpu->guest_debug &
4133                     (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4134                         return 0;
4135                 kvm_queue_exception(vcpu, vec);
4136                 return 1;
4137         case BP_VECTOR:
4138                 /*
4139                  * Update instruction length as we may reinject the exception
4140                  * from user space while in guest debugging mode.
4141                  */
4142                 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4143                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4144                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4145                         return 0;
4146                 /* fall through */
4147         case DE_VECTOR:
4148         case OF_VECTOR:
4149         case BR_VECTOR:
4150         case UD_VECTOR:
4151         case DF_VECTOR:
4152         case SS_VECTOR:
4153         case GP_VECTOR:
4154         case MF_VECTOR:
4155                 kvm_queue_exception(vcpu, vec);
4156                 return 1;
4157         }
4158         return 0;
4159 }
4160
4161 /*
4162  * Trigger machine check on the host. We assume all the MSRs are already set up
4163  * by the CPU and that we still run on the same CPU as the MCE occurred on.
4164  * We pass a fake environment to the machine check handler because we want
4165  * the guest to be always treated like user space, no matter what context
4166  * it used internally.
4167  */
4168 static void kvm_machine_check(void)
4169 {
4170 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4171         struct pt_regs regs = {
4172                 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
4173                 .flags = X86_EFLAGS_IF,
4174         };
4175
4176         do_machine_check(&regs, 0);
4177 #endif
4178 }
4179
4180 static int handle_machine_check(struct kvm_vcpu *vcpu)
4181 {
4182         /* already handled by vcpu_run */
4183         return 1;
4184 }
4185
4186 static int handle_exception(struct kvm_vcpu *vcpu)
4187 {
4188         struct vcpu_vmx *vmx = to_vmx(vcpu);
4189         struct kvm_run *kvm_run = vcpu->run;
4190         u32 intr_info, ex_no, error_code;
4191         unsigned long cr2, rip, dr6;
4192         u32 vect_info;
4193         enum emulation_result er;
4194
4195         vect_info = vmx->idt_vectoring_info;
4196         intr_info = vmx->exit_intr_info;
4197
4198         if (is_machine_check(intr_info))
4199                 return handle_machine_check(vcpu);
4200
4201         if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4202             !is_page_fault(intr_info)) {
4203                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4204                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4205                 vcpu->run->internal.ndata = 2;
4206                 vcpu->run->internal.data[0] = vect_info;
4207                 vcpu->run->internal.data[1] = intr_info;
4208                 return 0;
4209         }
4210
4211         if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
4212                 return 1;  /* already handled by vmx_vcpu_run() */
4213
4214         if (is_no_device(intr_info)) {
4215                 vmx_fpu_activate(vcpu);
4216                 return 1;
4217         }
4218
4219         if (is_invalid_opcode(intr_info)) {
4220                 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
4221                 if (er != EMULATE_DONE)
4222                         kvm_queue_exception(vcpu, UD_VECTOR);
4223                 return 1;
4224         }
4225
4226         error_code = 0;
4227         if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4228                 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4229         if (is_page_fault(intr_info)) {
4230                 /* EPT won't cause page fault directly */
4231                 BUG_ON(enable_ept);
4232                 cr2 = vmcs_readl(EXIT_QUALIFICATION);
4233                 trace_kvm_page_fault(cr2, error_code);
4234
4235                 if (kvm_event_needs_reinjection(vcpu))
4236                         kvm_mmu_unprotect_page_virt(vcpu, cr2);
4237                 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
4238         }
4239
4240         if (vmx->rmode.vm86_active &&
4241             handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
4242                                                                 error_code)) {
4243                 if (vcpu->arch.halt_request) {
4244                         vcpu->arch.halt_request = 0;
4245                         return kvm_emulate_halt(vcpu);
4246                 }
4247                 return 1;
4248         }
4249
4250         ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4251         switch (ex_no) {
4252         case DB_VECTOR:
4253                 dr6 = vmcs_readl(EXIT_QUALIFICATION);
4254                 if (!(vcpu->guest_debug &
4255                       (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4256                         vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
4257                         kvm_queue_exception(vcpu, DB_VECTOR);
4258                         return 1;
4259                 }
4260                 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4261                 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4262                 /* fall through */
4263         case BP_VECTOR:
4264                 /*
4265                  * Update instruction length as we may reinject #BP from
4266                  * user space while in guest debugging mode. Reading it for
4267                  * #DB as well causes no harm, it is not used in that case.
4268                  */
4269                 vmx->vcpu.arch.event_exit_inst_len =
4270                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4271                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
4272                 rip = kvm_rip_read(vcpu);
4273                 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4274                 kvm_run->debug.arch.exception = ex_no;
4275                 break;
4276         default:
4277                 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4278                 kvm_run->ex.exception = ex_no;
4279                 kvm_run->ex.error_code = error_code;
4280                 break;
4281         }
4282         return 0;
4283 }
4284
4285 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
4286 {
4287         ++vcpu->stat.irq_exits;
4288         return 1;
4289 }
4290
4291 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4292 {
4293         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4294         return 0;
4295 }
4296
4297 static int handle_io(struct kvm_vcpu *vcpu)
4298 {
4299         unsigned long exit_qualification;
4300         int size, in, string;
4301         unsigned port;
4302
4303         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4304         string = (exit_qualification & 16) != 0;
4305         in = (exit_qualification & 8) != 0;
4306
4307         ++vcpu->stat.io_exits;
4308
4309         if (string || in)
4310                 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4311
4312         port = exit_qualification >> 16;
4313         size = (exit_qualification & 7) + 1;
4314         skip_emulated_instruction(vcpu);
4315
4316         return kvm_fast_pio_out(vcpu, size, port);
4317 }
4318
4319 static void
4320 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4321 {
4322         /*
4323          * Patch in the VMCALL instruction:
4324          */
4325         hypercall[0] = 0x0f;
4326         hypercall[1] = 0x01;
4327         hypercall[2] = 0xc1;
4328 }
4329
4330 /* called to set cr0 as approriate for a mov-to-cr0 exit. */
4331 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4332 {
4333         if (to_vmx(vcpu)->nested.vmxon &&
4334             ((val & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON))
4335                 return 1;
4336
4337         if (is_guest_mode(vcpu)) {
4338                 /*
4339                  * We get here when L2 changed cr0 in a way that did not change
4340                  * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4341                  * but did change L0 shadowed bits. This can currently happen
4342                  * with the TS bit: L0 may want to leave TS on (for lazy fpu
4343                  * loading) while pretending to allow the guest to change it.
4344                  */
4345                 if (kvm_set_cr0(vcpu, (val & vcpu->arch.cr0_guest_owned_bits) |
4346                          (vcpu->arch.cr0 & ~vcpu->arch.cr0_guest_owned_bits)))
4347                         return 1;
4348                 vmcs_writel(CR0_READ_SHADOW, val);
4349                 return 0;
4350         } else
4351                 return kvm_set_cr0(vcpu, val);
4352 }
4353
4354 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4355 {
4356         if (is_guest_mode(vcpu)) {
4357                 if (kvm_set_cr4(vcpu, (val & vcpu->arch.cr4_guest_owned_bits) |
4358                          (vcpu->arch.cr4 & ~vcpu->arch.cr4_guest_owned_bits)))
4359                         return 1;
4360                 vmcs_writel(CR4_READ_SHADOW, val);
4361                 return 0;
4362         } else
4363                 return kvm_set_cr4(vcpu, val);
4364 }
4365
4366 /* called to set cr0 as approriate for clts instruction exit. */
4367 static void handle_clts(struct kvm_vcpu *vcpu)
4368 {
4369         if (is_guest_mode(vcpu)) {
4370                 /*
4371                  * We get here when L2 did CLTS, and L1 didn't shadow CR0.TS
4372                  * but we did (!fpu_active). We need to keep GUEST_CR0.TS on,
4373                  * just pretend it's off (also in arch.cr0 for fpu_activate).
4374                  */
4375                 vmcs_writel(CR0_READ_SHADOW,
4376                         vmcs_readl(CR0_READ_SHADOW) & ~X86_CR0_TS);
4377                 vcpu->arch.cr0 &= ~X86_CR0_TS;
4378         } else
4379                 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4380 }
4381
4382 static int handle_cr(struct kvm_vcpu *vcpu)
4383 {
4384         unsigned long exit_qualification, val;
4385         int cr;
4386         int reg;
4387         int err;
4388
4389         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4390         cr = exit_qualification & 15;
4391         reg = (exit_qualification >> 8) & 15;
4392         switch ((exit_qualification >> 4) & 3) {
4393         case 0: /* mov to cr */
4394                 val = kvm_register_read(vcpu, reg);
4395                 trace_kvm_cr_write(cr, val);
4396                 switch (cr) {
4397                 case 0:
4398                         err = handle_set_cr0(vcpu, val);
4399                         kvm_complete_insn_gp(vcpu, err);
4400                         return 1;
4401                 case 3:
4402                         err = kvm_set_cr3(vcpu, val);
4403                         kvm_complete_insn_gp(vcpu, err);
4404                         return 1;
4405                 case 4:
4406                         err = handle_set_cr4(vcpu, val);
4407                         kvm_complete_insn_gp(vcpu, err);
4408                         return 1;
4409                 case 8: {
4410                                 u8 cr8_prev = kvm_get_cr8(vcpu);
4411                                 u8 cr8 = kvm_register_read(vcpu, reg);
4412                                 err = kvm_set_cr8(vcpu, cr8);
4413                                 kvm_complete_insn_gp(vcpu, err);
4414                                 if (irqchip_in_kernel(vcpu->kvm))
4415                                         return 1;
4416                                 if (cr8_prev <= cr8)
4417                                         return 1;
4418                                 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4419                                 return 0;
4420                         }
4421                 };
4422                 break;
4423         case 2: /* clts */
4424                 handle_clts(vcpu);
4425                 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4426                 skip_emulated_instruction(vcpu);
4427                 vmx_fpu_activate(vcpu);
4428                 return 1;
4429         case 1: /*mov from cr*/
4430                 switch (cr) {
4431                 case 3:
4432                         val = kvm_read_cr3(vcpu);
4433                         kvm_register_write(vcpu, reg, val);
4434                         trace_kvm_cr_read(cr, val);
4435                         skip_emulated_instruction(vcpu);
4436                         return 1;
4437                 case 8:
4438                         val = kvm_get_cr8(vcpu);
4439                         kvm_register_write(vcpu, reg, val);
4440                         trace_kvm_cr_read(cr, val);
4441                         skip_emulated_instruction(vcpu);
4442                         return 1;
4443                 }
4444                 break;
4445         case 3: /* lmsw */
4446                 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4447                 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
4448                 kvm_lmsw(vcpu, val);
4449
4450                 skip_emulated_instruction(vcpu);
4451                 return 1;
4452         default:
4453                 break;
4454         }
4455         vcpu->run->exit_reason = 0;
4456         pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
4457                (int)(exit_qualification >> 4) & 3, cr);
4458         return 0;
4459 }
4460
4461 static int handle_dr(struct kvm_vcpu *vcpu)
4462 {
4463         unsigned long exit_qualification;
4464         int dr, reg;
4465
4466         /* Do not handle if the CPL > 0, will trigger GP on re-entry */
4467         if (!kvm_require_cpl(vcpu, 0))
4468                 return 1;
4469         dr = vmcs_readl(GUEST_DR7);
4470         if (dr & DR7_GD) {
4471                 /*
4472                  * As the vm-exit takes precedence over the debug trap, we
4473                  * need to emulate the latter, either for the host or the
4474                  * guest debugging itself.
4475                  */
4476                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4477                         vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
4478                         vcpu->run->debug.arch.dr7 = dr;
4479                         vcpu->run->debug.arch.pc =
4480                                 vmcs_readl(GUEST_CS_BASE) +
4481                                 vmcs_readl(GUEST_RIP);
4482                         vcpu->run->debug.arch.exception = DB_VECTOR;
4483                         vcpu->run->exit_reason = KVM_EXIT_DEBUG;
4484                         return 0;
4485                 } else {
4486                         vcpu->arch.dr7 &= ~DR7_GD;
4487                         vcpu->arch.dr6 |= DR6_BD;
4488                         vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
4489                         kvm_queue_exception(vcpu, DB_VECTOR);
4490                         return 1;
4491                 }
4492         }
4493
4494         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4495         dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
4496         reg = DEBUG_REG_ACCESS_REG(exit_qualification);
4497         if (exit_qualification & TYPE_MOV_FROM_DR) {
4498                 unsigned long val;
4499                 if (!kvm_get_dr(vcpu, dr, &val))
4500                         kvm_register_write(vcpu, reg, val);
4501         } else
4502                 kvm_set_dr(vcpu, dr, vcpu->arch.regs[reg]);
4503         skip_emulated_instruction(vcpu);
4504         return 1;
4505 }
4506
4507 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
4508 {
4509         vmcs_writel(GUEST_DR7, val);
4510 }
4511
4512 static int handle_cpuid(struct kvm_vcpu *vcpu)
4513 {
4514         kvm_emulate_cpuid(vcpu);
4515         return 1;
4516 }
4517
4518 static int handle_rdmsr(struct kvm_vcpu *vcpu)
4519 {
4520         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
4521         u64 data;
4522
4523         if (vmx_get_msr(vcpu, ecx, &data)) {
4524                 trace_kvm_msr_read_ex(ecx);
4525                 kvm_inject_gp(vcpu, 0);
4526                 return 1;
4527         }
4528
4529         trace_kvm_msr_read(ecx, data);
4530
4531         /* FIXME: handling of bits 32:63 of rax, rdx */
4532         vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
4533         vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
4534         skip_emulated_instruction(vcpu);
4535         return 1;
4536 }
4537
4538 static int handle_wrmsr(struct kvm_vcpu *vcpu)
4539 {
4540         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
4541         u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
4542                 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
4543
4544         if (vmx_set_msr(vcpu, ecx, data) != 0) {
4545                 trace_kvm_msr_write_ex(ecx, data);
4546                 kvm_inject_gp(vcpu, 0);
4547                 return 1;
4548         }
4549
4550         trace_kvm_msr_write(ecx, data);
4551         skip_emulated_instruction(vcpu);
4552         return 1;
4553 }
4554
4555 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
4556 {
4557         kvm_make_request(KVM_REQ_EVENT, vcpu);
4558         return 1;
4559 }
4560
4561 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
4562 {
4563         u32 cpu_based_vm_exec_control;
4564
4565         /* clear pending irq */
4566         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4567         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
4568         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4569
4570         kvm_make_request(KVM_REQ_EVENT, vcpu);
4571
4572         ++vcpu->stat.irq_window_exits;
4573
4574         /*
4575          * If the user space waits to inject interrupts, exit as soon as
4576          * possible
4577          */
4578         if (!irqchip_in_kernel(vcpu->kvm) &&
4579             vcpu->run->request_interrupt_window &&
4580             !kvm_cpu_has_interrupt(vcpu)) {
4581                 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
4582                 return 0;
4583         }
4584         return 1;
4585 }
4586
4587 static int handle_halt(struct kvm_vcpu *vcpu)
4588 {
4589         skip_emulated_instruction(vcpu);
4590         return kvm_emulate_halt(vcpu);
4591 }
4592
4593 static int handle_vmcall(struct kvm_vcpu *vcpu)
4594 {
4595         skip_emulated_instruction(vcpu);
4596         kvm_emulate_hypercall(vcpu);
4597         return 1;
4598 }
4599
4600 static int handle_invd(struct kvm_vcpu *vcpu)
4601 {
4602         return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4603 }
4604
4605 static int handle_invlpg(struct kvm_vcpu *vcpu)
4606 {
4607         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4608
4609         kvm_mmu_invlpg(vcpu, exit_qualification);
4610         skip_emulated_instruction(vcpu);
4611         return 1;
4612 }
4613
4614 static int handle_wbinvd(struct kvm_vcpu *vcpu)
4615 {
4616         skip_emulated_instruction(vcpu);
4617         kvm_emulate_wbinvd(vcpu);
4618         return 1;
4619 }
4620
4621 static int handle_xsetbv(struct kvm_vcpu *vcpu)
4622 {
4623         u64 new_bv = kvm_read_edx_eax(vcpu);
4624         u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4625
4626         if (kvm_set_xcr(vcpu, index, new_bv) == 0)
4627                 skip_emulated_instruction(vcpu);
4628         return 1;
4629 }
4630
4631 static int handle_apic_access(struct kvm_vcpu *vcpu)
4632 {
4633         if (likely(fasteoi)) {
4634                 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4635                 int access_type, offset;
4636
4637                 access_type = exit_qualification & APIC_ACCESS_TYPE;
4638                 offset = exit_qualification & APIC_ACCESS_OFFSET;
4639                 /*
4640                  * Sane guest uses MOV to write EOI, with written value
4641                  * not cared. So make a short-circuit here by avoiding
4642                  * heavy instruction emulation.
4643                  */
4644                 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
4645                     (offset == APIC_EOI)) {
4646                         kvm_lapic_set_eoi(vcpu);
4647                         skip_emulated_instruction(vcpu);
4648                         return 1;
4649                 }
4650         }
4651         return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4652 }
4653
4654 static int handle_task_switch(struct kvm_vcpu *vcpu)
4655 {
4656         struct vcpu_vmx *vmx = to_vmx(vcpu);
4657         unsigned long exit_qualification;
4658         bool has_error_code = false;
4659         u32 error_code = 0;
4660         u16 tss_selector;
4661         int reason, type, idt_v;
4662
4663         idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
4664         type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
4665
4666         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4667
4668         reason = (u32)exit_qualification >> 30;
4669         if (reason == TASK_SWITCH_GATE && idt_v) {
4670                 switch (type) {
4671                 case INTR_TYPE_NMI_INTR:
4672                         vcpu->arch.nmi_injected = false;
4673                         vmx_set_nmi_mask(vcpu, true);
4674                         break;
4675                 case INTR_TYPE_EXT_INTR:
4676                 case INTR_TYPE_SOFT_INTR:
4677                         kvm_clear_interrupt_queue(vcpu);
4678                         break;
4679                 case INTR_TYPE_HARD_EXCEPTION:
4680                         if (vmx->idt_vectoring_info &
4681                             VECTORING_INFO_DELIVER_CODE_MASK) {
4682                                 has_error_code = true;
4683                                 error_code =
4684                                         vmcs_read32(IDT_VECTORING_ERROR_CODE);
4685                         }
4686                         /* fall through */
4687                 case INTR_TYPE_SOFT_EXCEPTION:
4688                         kvm_clear_exception_queue(vcpu);
4689                         break;
4690                 default:
4691                         break;
4692                 }
4693         }
4694         tss_selector = exit_qualification;
4695
4696         if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
4697                        type != INTR_TYPE_EXT_INTR &&
4698                        type != INTR_TYPE_NMI_INTR))
4699                 skip_emulated_instruction(vcpu);
4700
4701         if (kvm_task_switch(vcpu, tss_selector, reason,
4702                                 has_error_code, error_code) == EMULATE_FAIL) {
4703                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4704                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4705                 vcpu->run->internal.ndata = 0;
4706                 return 0;
4707         }
4708
4709         /* clear all local breakpoint enable flags */
4710         vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
4711
4712         /*
4713          * TODO: What about debug traps on tss switch?
4714          *       Are we supposed to inject them and update dr6?
4715          */
4716
4717         return 1;
4718 }
4719
4720 static int handle_ept_violation(struct kvm_vcpu *vcpu)
4721 {
4722         unsigned long exit_qualification;
4723         gpa_t gpa;
4724         int gla_validity;
4725
4726         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4727
4728         if (exit_qualification & (1 << 6)) {
4729                 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
4730                 return -EINVAL;
4731         }
4732
4733         gla_validity = (exit_qualification >> 7) & 0x3;
4734         if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
4735                 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
4736                 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
4737                         (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
4738                         vmcs_readl(GUEST_LINEAR_ADDRESS));
4739                 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
4740                         (long unsigned int)exit_qualification);
4741                 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
4742                 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
4743                 return 0;
4744         }
4745
4746         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
4747         trace_kvm_page_fault(gpa, exit_qualification);
4748         return kvm_mmu_page_fault(vcpu, gpa, exit_qualification & 0x3, NULL, 0);
4749 }
4750
4751 static u64 ept_rsvd_mask(u64 spte, int level)
4752 {
4753         int i;
4754         u64 mask = 0;
4755
4756         for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
4757                 mask |= (1ULL << i);
4758
4759         if (level > 2)
4760                 /* bits 7:3 reserved */
4761                 mask |= 0xf8;
4762         else if (level == 2) {
4763                 if (spte & (1ULL << 7))
4764                         /* 2MB ref, bits 20:12 reserved */
4765                         mask |= 0x1ff000;
4766                 else
4767                         /* bits 6:3 reserved */
4768                         mask |= 0x78;
4769         }
4770
4771         return mask;
4772 }
4773
4774 static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
4775                                        int level)
4776 {
4777         printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
4778
4779         /* 010b (write-only) */
4780         WARN_ON((spte & 0x7) == 0x2);
4781
4782         /* 110b (write/execute) */
4783         WARN_ON((spte & 0x7) == 0x6);
4784
4785         /* 100b (execute-only) and value not supported by logical processor */
4786         if (!cpu_has_vmx_ept_execute_only())
4787                 WARN_ON((spte & 0x7) == 0x4);
4788
4789         /* not 000b */
4790         if ((spte & 0x7)) {
4791                 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
4792
4793                 if (rsvd_bits != 0) {
4794                         printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
4795                                          __func__, rsvd_bits);
4796                         WARN_ON(1);
4797                 }
4798
4799                 if (level == 1 || (level == 2 && (spte & (1ULL << 7)))) {
4800                         u64 ept_mem_type = (spte & 0x38) >> 3;
4801
4802                         if (ept_mem_type == 2 || ept_mem_type == 3 ||
4803                             ept_mem_type == 7) {
4804                                 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
4805                                                 __func__, ept_mem_type);
4806                                 WARN_ON(1);
4807                         }
4808                 }
4809         }
4810 }
4811
4812 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
4813 {
4814         u64 sptes[4];
4815         int nr_sptes, i, ret;
4816         gpa_t gpa;
4817
4818         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
4819
4820         ret = handle_mmio_page_fault_common(vcpu, gpa, true);
4821         if (likely(ret == 1))
4822                 return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) ==
4823                                               EMULATE_DONE;
4824         if (unlikely(!ret))
4825                 return 1;
4826
4827         /* It is the real ept misconfig */
4828         printk(KERN_ERR "EPT: Misconfiguration.\n");
4829         printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
4830
4831         nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
4832
4833         for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
4834                 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
4835
4836         vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
4837         vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
4838
4839         return 0;
4840 }
4841
4842 static int handle_nmi_window(struct kvm_vcpu *vcpu)
4843 {
4844         u32 cpu_based_vm_exec_control;
4845
4846         /* clear pending NMI */
4847         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4848         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
4849         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4850         ++vcpu->stat.nmi_window_exits;
4851         kvm_make_request(KVM_REQ_EVENT, vcpu);
4852
4853         return 1;
4854 }
4855
4856 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
4857 {
4858         struct vcpu_vmx *vmx = to_vmx(vcpu);
4859         enum emulation_result err = EMULATE_DONE;
4860         int ret = 1;
4861         u32 cpu_exec_ctrl;
4862         bool intr_window_requested;
4863
4864         cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4865         intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
4866
4867         while (!guest_state_valid(vcpu)) {
4868                 if (intr_window_requested
4869                     && (kvm_get_rflags(&vmx->vcpu) & X86_EFLAGS_IF))
4870                         return handle_interrupt_window(&vmx->vcpu);
4871
4872                 err = emulate_instruction(vcpu, 0);
4873
4874                 if (err == EMULATE_DO_MMIO) {
4875                         ret = 0;
4876                         goto out;
4877                 }
4878
4879                 if (err != EMULATE_DONE)
4880                         return 0;
4881
4882                 if (signal_pending(current))
4883                         goto out;
4884                 if (need_resched())
4885                         schedule();
4886         }
4887
4888         vmx->emulation_required = 0;
4889 out:
4890         return ret;
4891 }
4892
4893 /*
4894  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
4895  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
4896  */
4897 static int handle_pause(struct kvm_vcpu *vcpu)
4898 {
4899         skip_emulated_instruction(vcpu);
4900         kvm_vcpu_on_spin(vcpu);
4901
4902         return 1;
4903 }
4904
4905 static int handle_invalid_op(struct kvm_vcpu *vcpu)
4906 {
4907         kvm_queue_exception(vcpu, UD_VECTOR);
4908         return 1;
4909 }
4910
4911 /*
4912  * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
4913  * We could reuse a single VMCS for all the L2 guests, but we also want the
4914  * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
4915  * allows keeping them loaded on the processor, and in the future will allow
4916  * optimizations where prepare_vmcs02 doesn't need to set all the fields on
4917  * every entry if they never change.
4918  * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
4919  * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
4920  *
4921  * The following functions allocate and free a vmcs02 in this pool.
4922  */
4923
4924 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
4925 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
4926 {
4927         struct vmcs02_list *item;
4928         list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
4929                 if (item->vmptr == vmx->nested.current_vmptr) {
4930                         list_move(&item->list, &vmx->nested.vmcs02_pool);
4931                         return &item->vmcs02;
4932                 }
4933
4934         if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
4935                 /* Recycle the least recently used VMCS. */
4936                 item = list_entry(vmx->nested.vmcs02_pool.prev,
4937                         struct vmcs02_list, list);
4938                 item->vmptr = vmx->nested.current_vmptr;
4939                 list_move(&item->list, &vmx->nested.vmcs02_pool);
4940                 return &item->vmcs02;
4941         }
4942
4943         /* Create a new VMCS */
4944         item = (struct vmcs02_list *)
4945                 kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
4946         if (!item)
4947                 return NULL;
4948         item->vmcs02.vmcs = alloc_vmcs();
4949         if (!item->vmcs02.vmcs) {
4950                 kfree(item);
4951                 return NULL;
4952         }
4953         loaded_vmcs_init(&item->vmcs02);
4954         item->vmptr = vmx->nested.current_vmptr;
4955         list_add(&(item->list), &(vmx->nested.vmcs02_pool));
4956         vmx->nested.vmcs02_num++;
4957         return &item->vmcs02;
4958 }
4959
4960 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
4961 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
4962 {
4963         struct vmcs02_list *item;
4964         list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
4965                 if (item->vmptr == vmptr) {
4966                         free_loaded_vmcs(&item->vmcs02);
4967                         list_del(&item->list);
4968                         kfree(item);
4969                         vmx->nested.vmcs02_num--;
4970                         return;
4971                 }
4972 }
4973
4974 /*
4975  * Free all VMCSs saved for this vcpu, except the one pointed by
4976  * vmx->loaded_vmcs. These include the VMCSs in vmcs02_pool (except the one
4977  * currently used, if running L2), and vmcs01 when running L2.
4978  */
4979 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
4980 {
4981         struct vmcs02_list *item, *n;
4982         list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
4983                 if (vmx->loaded_vmcs != &item->vmcs02)
4984                         free_loaded_vmcs(&item->vmcs02);
4985                 list_del(&item->list);
4986                 kfree(item);
4987         }
4988         vmx->nested.vmcs02_num = 0;
4989
4990         if (vmx->loaded_vmcs != &vmx->vmcs01)
4991                 free_loaded_vmcs(&vmx->vmcs01);
4992 }
4993
4994 /*
4995  * Emulate the VMXON instruction.
4996  * Currently, we just remember that VMX is active, and do not save or even
4997  * inspect the argument to VMXON (the so-called "VMXON pointer") because we
4998  * do not currently need to store anything in that guest-allocated memory
4999  * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
5000  * argument is different from the VMXON pointer (which the spec says they do).
5001  */
5002 static int handle_vmon(struct kvm_vcpu *vcpu)
5003 {
5004         struct kvm_segment cs;
5005         struct vcpu_vmx *vmx = to_vmx(vcpu);
5006
5007         /* The Intel VMX Instruction Reference lists a bunch of bits that
5008          * are prerequisite to running VMXON, most notably cr4.VMXE must be
5009          * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
5010          * Otherwise, we should fail with #UD. We test these now:
5011          */
5012         if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
5013             !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
5014             (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
5015                 kvm_queue_exception(vcpu, UD_VECTOR);
5016                 return 1;
5017         }
5018
5019         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
5020         if (is_long_mode(vcpu) && !cs.l) {
5021                 kvm_queue_exception(vcpu, UD_VECTOR);
5022                 return 1;
5023         }
5024
5025         if (vmx_get_cpl(vcpu)) {
5026                 kvm_inject_gp(vcpu, 0);
5027                 return 1;
5028         }
5029
5030         INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
5031         vmx->nested.vmcs02_num = 0;
5032
5033         vmx->nested.vmxon = true;
5034
5035         skip_emulated_instruction(vcpu);
5036         return 1;
5037 }
5038
5039 /*
5040  * Intel's VMX Instruction Reference specifies a common set of prerequisites
5041  * for running VMX instructions (except VMXON, whose prerequisites are
5042  * slightly different). It also specifies what exception to inject otherwise.
5043  */
5044 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
5045 {
5046         struct kvm_segment cs;
5047         struct vcpu_vmx *vmx = to_vmx(vcpu);
5048
5049         if (!vmx->nested.vmxon) {
5050                 kvm_queue_exception(vcpu, UD_VECTOR);
5051                 return 0;
5052         }
5053
5054         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
5055         if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
5056             (is_long_mode(vcpu) && !cs.l)) {
5057                 kvm_queue_exception(vcpu, UD_VECTOR);
5058                 return 0;
5059         }
5060
5061         if (vmx_get_cpl(vcpu)) {
5062                 kvm_inject_gp(vcpu, 0);
5063                 return 0;
5064         }
5065
5066         return 1;
5067 }
5068
5069 /*
5070  * Free whatever needs to be freed from vmx->nested when L1 goes down, or
5071  * just stops using VMX.
5072  */
5073 static void free_nested(struct vcpu_vmx *vmx)
5074 {
5075         if (!vmx->nested.vmxon)
5076                 return;
5077         vmx->nested.vmxon = false;
5078         if (vmx->nested.current_vmptr != -1ull) {
5079                 kunmap(vmx->nested.current_vmcs12_page);
5080                 nested_release_page(vmx->nested.current_vmcs12_page);
5081                 vmx->nested.current_vmptr = -1ull;
5082                 vmx->nested.current_vmcs12 = NULL;
5083         }
5084         /* Unpin physical memory we referred to in current vmcs02 */
5085         if (vmx->nested.apic_access_page) {
5086                 nested_release_page(vmx->nested.apic_access_page);
5087                 vmx->nested.apic_access_page = 0;
5088         }
5089
5090         nested_free_all_saved_vmcss(vmx);
5091 }
5092
5093 /* Emulate the VMXOFF instruction */
5094 static int handle_vmoff(struct kvm_vcpu *vcpu)
5095 {
5096         if (!nested_vmx_check_permission(vcpu))
5097                 return 1;
5098         free_nested(to_vmx(vcpu));
5099         skip_emulated_instruction(vcpu);
5100         return 1;
5101 }
5102
5103 /*
5104  * Decode the memory-address operand of a vmx instruction, as recorded on an
5105  * exit caused by such an instruction (run by a guest hypervisor).
5106  * On success, returns 0. When the operand is invalid, returns 1 and throws
5107  * #UD or #GP.
5108  */
5109 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
5110                                  unsigned long exit_qualification,
5111                                  u32 vmx_instruction_info, gva_t *ret)
5112 {
5113         /*
5114          * According to Vol. 3B, "Information for VM Exits Due to Instruction
5115          * Execution", on an exit, vmx_instruction_info holds most of the
5116          * addressing components of the operand. Only the displacement part
5117          * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
5118          * For how an actual address is calculated from all these components,
5119          * refer to Vol. 1, "Operand Addressing".
5120          */
5121         int  scaling = vmx_instruction_info & 3;
5122         int  addr_size = (vmx_instruction_info >> 7) & 7;
5123         bool is_reg = vmx_instruction_info & (1u << 10);
5124         int  seg_reg = (vmx_instruction_info >> 15) & 7;
5125         int  index_reg = (vmx_instruction_info >> 18) & 0xf;
5126         bool index_is_valid = !(vmx_instruction_info & (1u << 22));
5127         int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
5128         bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
5129
5130         if (is_reg) {
5131                 kvm_queue_exception(vcpu, UD_VECTOR);
5132                 return 1;
5133         }
5134
5135         /* Addr = segment_base + offset */
5136         /* offset = base + [index * scale] + displacement */
5137         *ret = vmx_get_segment_base(vcpu, seg_reg);
5138         if (base_is_valid)
5139                 *ret += kvm_register_read(vcpu, base_reg);
5140         if (index_is_valid)
5141                 *ret += kvm_register_read(vcpu, index_reg)<<scaling;
5142         *ret += exit_qualification; /* holds the displacement */
5143
5144         if (addr_size == 1) /* 32 bit */
5145                 *ret &= 0xffffffff;
5146
5147         /*
5148          * TODO: throw #GP (and return 1) in various cases that the VM*
5149          * instructions require it - e.g., offset beyond segment limit,
5150          * unusable or unreadable/unwritable segment, non-canonical 64-bit
5151          * address, and so on. Currently these are not checked.
5152          */
5153         return 0;
5154 }
5155
5156 /*
5157  * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
5158  * set the success or error code of an emulated VMX instruction, as specified
5159  * by Vol 2B, VMX Instruction Reference, "Conventions".
5160  */
5161 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
5162 {
5163         vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
5164                         & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
5165                             X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
5166 }
5167
5168 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
5169 {
5170         vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
5171                         & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
5172                             X86_EFLAGS_SF | X86_EFLAGS_OF))
5173                         | X86_EFLAGS_CF);
5174 }
5175
5176 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
5177                                         u32 vm_instruction_error)
5178 {
5179         if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
5180                 /*
5181                  * failValid writes the error number to the current VMCS, which
5182                  * can't be done there isn't a current VMCS.
5183                  */
5184                 nested_vmx_failInvalid(vcpu);
5185                 return;
5186         }
5187         vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
5188                         & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
5189                             X86_EFLAGS_SF | X86_EFLAGS_OF))
5190                         | X86_EFLAGS_ZF);
5191         get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
5192 }
5193
5194 /* Emulate the VMCLEAR instruction */
5195 static int handle_vmclear(struct kvm_vcpu *vcpu)
5196 {
5197         struct vcpu_vmx *vmx = to_vmx(vcpu);
5198         gva_t gva;
5199         gpa_t vmptr;
5200         struct vmcs12 *vmcs12;
5201         struct page *page;
5202         struct x86_exception e;
5203
5204         if (!nested_vmx_check_permission(vcpu))
5205                 return 1;
5206
5207         if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5208                         vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
5209                 return 1;
5210
5211         if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
5212                                 sizeof(vmptr), &e)) {
5213                 kvm_inject_page_fault(vcpu, &e);
5214                 return 1;
5215         }
5216
5217         if (!IS_ALIGNED(vmptr, PAGE_SIZE)) {
5218                 nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
5219                 skip_emulated_instruction(vcpu);
5220                 return 1;
5221         }
5222
5223         if (vmptr == vmx->nested.current_vmptr) {
5224                 kunmap(vmx->nested.current_vmcs12_page);
5225                 nested_release_page(vmx->nested.current_vmcs12_page);
5226                 vmx->nested.current_vmptr = -1ull;
5227                 vmx->nested.current_vmcs12 = NULL;
5228         }
5229
5230         page = nested_get_page(vcpu, vmptr);
5231         if (page == NULL) {
5232                 /*
5233                  * For accurate processor emulation, VMCLEAR beyond available
5234                  * physical memory should do nothing at all. However, it is
5235                  * possible that a nested vmx bug, not a guest hypervisor bug,
5236                  * resulted in this case, so let's shut down before doing any
5237                  * more damage:
5238                  */
5239                 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5240                 return 1;
5241         }
5242         vmcs12 = kmap(page);
5243         vmcs12->launch_state = 0;
5244         kunmap(page);
5245         nested_release_page(page);
5246
5247         nested_free_vmcs02(vmx, vmptr);
5248
5249         skip_emulated_instruction(vcpu);
5250         nested_vmx_succeed(vcpu);
5251         return 1;
5252 }
5253
5254 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
5255
5256 /* Emulate the VMLAUNCH instruction */
5257 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
5258 {
5259         return nested_vmx_run(vcpu, true);
5260 }
5261
5262 /* Emulate the VMRESUME instruction */
5263 static int handle_vmresume(struct kvm_vcpu *vcpu)
5264 {
5265
5266         return nested_vmx_run(vcpu, false);
5267 }
5268
5269 enum vmcs_field_type {
5270         VMCS_FIELD_TYPE_U16 = 0,
5271         VMCS_FIELD_TYPE_U64 = 1,
5272         VMCS_FIELD_TYPE_U32 = 2,
5273         VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
5274 };
5275
5276 static inline int vmcs_field_type(unsigned long field)
5277 {
5278         if (0x1 & field)        /* the *_HIGH fields are all 32 bit */
5279                 return VMCS_FIELD_TYPE_U32;
5280         return (field >> 13) & 0x3 ;
5281 }
5282
5283 static inline int vmcs_field_readonly(unsigned long field)
5284 {
5285         return (((field >> 10) & 0x3) == 1);
5286 }
5287
5288 /*
5289  * Read a vmcs12 field. Since these can have varying lengths and we return
5290  * one type, we chose the biggest type (u64) and zero-extend the return value
5291  * to that size. Note that the caller, handle_vmread, might need to use only
5292  * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
5293  * 64-bit fields are to be returned).
5294  */
5295 static inline bool vmcs12_read_any(struct kvm_vcpu *vcpu,
5296                                         unsigned long field, u64 *ret)
5297 {
5298         short offset = vmcs_field_to_offset(field);
5299         char *p;
5300
5301         if (offset < 0)
5302                 return 0;
5303
5304         p = ((char *)(get_vmcs12(vcpu))) + offset;
5305
5306         switch (vmcs_field_type(field)) {
5307         case VMCS_FIELD_TYPE_NATURAL_WIDTH:
5308                 *ret = *((natural_width *)p);
5309                 return 1;
5310         case VMCS_FIELD_TYPE_U16:
5311                 *ret = *((u16 *)p);
5312                 return 1;
5313         case VMCS_FIELD_TYPE_U32:
5314                 *ret = *((u32 *)p);
5315                 return 1;
5316         case VMCS_FIELD_TYPE_U64:
5317                 *ret = *((u64 *)p);
5318                 return 1;
5319         default:
5320                 return 0; /* can never happen. */
5321         }
5322 }
5323
5324 /*
5325  * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
5326  * used before) all generate the same failure when it is missing.
5327  */
5328 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
5329 {
5330         struct vcpu_vmx *vmx = to_vmx(vcpu);
5331         if (vmx->nested.current_vmptr == -1ull) {
5332                 nested_vmx_failInvalid(vcpu);
5333                 skip_emulated_instruction(vcpu);
5334                 return 0;
5335         }
5336         return 1;
5337 }
5338
5339 static int handle_vmread(struct kvm_vcpu *vcpu)
5340 {
5341         unsigned long field;
5342         u64 field_value;
5343         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5344         u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5345         gva_t gva = 0;
5346
5347         if (!nested_vmx_check_permission(vcpu) ||
5348             !nested_vmx_check_vmcs12(vcpu))
5349                 return 1;
5350
5351         /* Decode instruction info and find the field to read */
5352         field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5353         /* Read the field, zero-extended to a u64 field_value */
5354         if (!vmcs12_read_any(vcpu, field, &field_value)) {
5355                 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5356                 skip_emulated_instruction(vcpu);
5357                 return 1;
5358         }
5359         /*
5360          * Now copy part of this value to register or memory, as requested.
5361          * Note that the number of bits actually copied is 32 or 64 depending
5362          * on the guest's mode (32 or 64 bit), not on the given field's length.
5363          */
5364         if (vmx_instruction_info & (1u << 10)) {
5365                 kvm_register_write(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
5366                         field_value);
5367         } else {
5368                 if (get_vmx_mem_address(vcpu, exit_qualification,
5369                                 vmx_instruction_info, &gva))
5370                         return 1;
5371                 /* _system ok, as nested_vmx_check_permission verified cpl=0 */
5372                 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
5373                              &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
5374         }
5375
5376         nested_vmx_succeed(vcpu);
5377         skip_emulated_instruction(vcpu);
5378         return 1;
5379 }
5380
5381
5382 static int handle_vmwrite(struct kvm_vcpu *vcpu)
5383 {
5384         unsigned long field;
5385         gva_t gva;
5386         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5387         u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5388         char *p;
5389         short offset;
5390         /* The value to write might be 32 or 64 bits, depending on L1's long
5391          * mode, and eventually we need to write that into a field of several
5392          * possible lengths. The code below first zero-extends the value to 64
5393          * bit (field_value), and then copies only the approriate number of
5394          * bits into the vmcs12 field.
5395          */
5396         u64 field_value = 0;
5397         struct x86_exception e;
5398
5399         if (!nested_vmx_check_permission(vcpu) ||
5400             !nested_vmx_check_vmcs12(vcpu))
5401                 return 1;
5402
5403         if (vmx_instruction_info & (1u << 10))
5404                 field_value = kvm_register_read(vcpu,
5405                         (((vmx_instruction_info) >> 3) & 0xf));
5406         else {
5407                 if (get_vmx_mem_address(vcpu, exit_qualification,
5408                                 vmx_instruction_info, &gva))
5409                         return 1;
5410                 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
5411                            &field_value, (is_long_mode(vcpu) ? 8 : 4), &e)) {
5412                         kvm_inject_page_fault(vcpu, &e);
5413                         return 1;
5414                 }
5415         }
5416
5417
5418         field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5419         if (vmcs_field_readonly(field)) {
5420                 nested_vmx_failValid(vcpu,
5421                         VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
5422                 skip_emulated_instruction(vcpu);
5423                 return 1;
5424         }
5425
5426         offset = vmcs_field_to_offset(field);
5427         if (offset < 0) {
5428                 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5429                 skip_emulated_instruction(vcpu);
5430                 return 1;
5431         }
5432         p = ((char *) get_vmcs12(vcpu)) + offset;
5433
5434         switch (vmcs_field_type(field)) {
5435         case VMCS_FIELD_TYPE_U16:
5436                 *(u16 *)p = field_value;
5437                 break;
5438         case VMCS_FIELD_TYPE_U32:
5439                 *(u32 *)p = field_value;
5440                 break;
5441         case VMCS_FIELD_TYPE_U64:
5442                 *(u64 *)p = field_value;
5443                 break;
5444         case VMCS_FIELD_TYPE_NATURAL_WIDTH:
5445                 *(natural_width *)p = field_value;
5446                 break;
5447         default:
5448                 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5449                 skip_emulated_instruction(vcpu);
5450                 return 1;
5451         }
5452
5453         nested_vmx_succeed(vcpu);
5454         skip_emulated_instruction(vcpu);
5455         return 1;
5456 }
5457
5458 /* Emulate the VMPTRLD instruction */
5459 static int handle_vmptrld(struct kvm_vcpu *vcpu)
5460 {
5461         struct vcpu_vmx *vmx = to_vmx(vcpu);
5462         gva_t gva;
5463         gpa_t vmptr;
5464         struct x86_exception e;
5465
5466         if (!nested_vmx_check_permission(vcpu))
5467                 return 1;
5468
5469         if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5470                         vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
5471                 return 1;
5472
5473         if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
5474                                 sizeof(vmptr), &e)) {
5475                 kvm_inject_page_fault(vcpu, &e);
5476                 return 1;
5477         }
5478
5479         if (!IS_ALIGNED(vmptr, PAGE_SIZE)) {
5480                 nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
5481                 skip_emulated_instruction(vcpu);
5482                 return 1;
5483         }
5484
5485         if (vmx->nested.current_vmptr != vmptr) {
5486                 struct vmcs12 *new_vmcs12;
5487                 struct page *page;
5488                 page = nested_get_page(vcpu, vmptr);
5489                 if (page == NULL) {
5490                         nested_vmx_failInvalid(vcpu);
5491                         skip_emulated_instruction(vcpu);
5492                         return 1;
5493                 }
5494                 new_vmcs12 = kmap(page);
5495                 if (new_vmcs12->revision_id != VMCS12_REVISION) {
5496                         kunmap(page);
5497                         nested_release_page_clean(page);
5498                         nested_vmx_failValid(vcpu,
5499                                 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5500                         skip_emulated_instruction(vcpu);
5501                         return 1;
5502                 }
5503                 if (vmx->nested.current_vmptr != -1ull) {
5504                         kunmap(vmx->nested.current_vmcs12_page);
5505                         nested_release_page(vmx->nested.current_vmcs12_page);
5506                 }
5507
5508                 vmx->nested.current_vmptr = vmptr;
5509                 vmx->nested.current_vmcs12 = new_vmcs12;
5510                 vmx->nested.current_vmcs12_page = page;
5511         }
5512
5513         nested_vmx_succeed(vcpu);
5514         skip_emulated_instruction(vcpu);
5515         return 1;
5516 }
5517
5518 /* Emulate the VMPTRST instruction */
5519 static int handle_vmptrst(struct kvm_vcpu *vcpu)
5520 {
5521         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5522         u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5523         gva_t vmcs_gva;
5524         struct x86_exception e;
5525
5526         if (!nested_vmx_check_permission(vcpu))
5527                 return 1;
5528
5529         if (get_vmx_mem_address(vcpu, exit_qualification,
5530                         vmx_instruction_info, &vmcs_gva))
5531                 return 1;
5532         /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
5533         if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
5534                                  (void *)&to_vmx(vcpu)->nested.current_vmptr,
5535                                  sizeof(u64), &e)) {
5536                 kvm_inject_page_fault(vcpu, &e);
5537                 return 1;
5538         }
5539         nested_vmx_succeed(vcpu);
5540         skip_emulated_instruction(vcpu);
5541         return 1;
5542 }
5543
5544 /*
5545  * The exit handlers return 1 if the exit was handled fully and guest execution
5546  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
5547  * to be done to userspace and return 0.
5548  */
5549 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5550         [EXIT_REASON_EXCEPTION_NMI]           = handle_exception,
5551         [EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
5552         [EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
5553         [EXIT_REASON_NMI_WINDOW]              = handle_nmi_window,
5554         [EXIT_REASON_IO_INSTRUCTION]          = handle_io,
5555         [EXIT_REASON_CR_ACCESS]               = handle_cr,
5556         [EXIT_REASON_DR_ACCESS]               = handle_dr,
5557         [EXIT_REASON_CPUID]                   = handle_cpuid,
5558         [EXIT_REASON_MSR_READ]                = handle_rdmsr,
5559         [EXIT_REASON_MSR_WRITE]               = handle_wrmsr,
5560         [EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
5561         [EXIT_REASON_HLT]                     = handle_halt,
5562         [EXIT_REASON_INVD]                    = handle_invd,
5563         [EXIT_REASON_INVLPG]                  = handle_invlpg,
5564         [EXIT_REASON_VMCALL]                  = handle_vmcall,
5565         [EXIT_REASON_VMCLEAR]                 = handle_vmclear,
5566         [EXIT_REASON_VMLAUNCH]                = handle_vmlaunch,
5567         [EXIT_REASON_VMPTRLD]                 = handle_vmptrld,
5568         [EXIT_REASON_VMPTRST]                 = handle_vmptrst,
5569         [EXIT_REASON_VMREAD]                  = handle_vmread,
5570         [EXIT_REASON_VMRESUME]                = handle_vmresume,
5571         [EXIT_REASON_VMWRITE]                 = handle_vmwrite,
5572         [EXIT_REASON_VMOFF]                   = handle_vmoff,
5573         [EXIT_REASON_VMON]                    = handle_vmon,
5574         [EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
5575         [EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
5576         [EXIT_REASON_WBINVD]                  = handle_wbinvd,
5577         [EXIT_REASON_XSETBV]                  = handle_xsetbv,
5578         [EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
5579         [EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
5580         [EXIT_REASON_EPT_VIOLATION]           = handle_ept_violation,
5581         [EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
5582         [EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
5583         [EXIT_REASON_MWAIT_INSTRUCTION]       = handle_invalid_op,
5584         [EXIT_REASON_MONITOR_INSTRUCTION]     = handle_invalid_op,
5585 };
5586
5587 static const int kvm_vmx_max_exit_handlers =
5588         ARRAY_SIZE(kvm_vmx_exit_handlers);
5589
5590 /*
5591  * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
5592  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
5593  * disinterest in the current event (read or write a specific MSR) by using an
5594  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
5595  */
5596 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
5597         struct vmcs12 *vmcs12, u32 exit_reason)
5598 {
5599         u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
5600         gpa_t bitmap;
5601
5602         if (!nested_cpu_has(get_vmcs12(vcpu), CPU_BASED_USE_MSR_BITMAPS))
5603                 return 1;
5604
5605         /*
5606          * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
5607          * for the four combinations of read/write and low/high MSR numbers.
5608          * First we need to figure out which of the four to use:
5609          */
5610         bitmap = vmcs12->msr_bitmap;
5611         if (exit_reason == EXIT_REASON_MSR_WRITE)
5612                 bitmap += 2048;
5613         if (msr_index >= 0xc0000000) {
5614                 msr_index -= 0xc0000000;
5615                 bitmap += 1024;
5616         }
5617
5618         /* Then read the msr_index'th bit from this bitmap: */
5619         if (msr_index < 1024*8) {
5620                 unsigned char b;
5621                 kvm_read_guest(vcpu->kvm, bitmap + msr_index/8, &b, 1);
5622                 return 1 & (b >> (msr_index & 7));
5623         } else
5624                 return 1; /* let L1 handle the wrong parameter */
5625 }
5626
5627 /*
5628  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
5629  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
5630  * intercept (via guest_host_mask etc.) the current event.
5631  */
5632 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
5633         struct vmcs12 *vmcs12)
5634 {
5635         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5636         int cr = exit_qualification & 15;
5637         int reg = (exit_qualification >> 8) & 15;
5638         unsigned long val = kvm_register_read(vcpu, reg);
5639
5640         switch ((exit_qualification >> 4) & 3) {
5641         case 0: /* mov to cr */
5642                 switch (cr) {
5643                 case 0:
5644                         if (vmcs12->cr0_guest_host_mask &
5645                             (val ^ vmcs12->cr0_read_shadow))
5646                                 return 1;
5647                         break;
5648                 case 3:
5649                         if ((vmcs12->cr3_target_count >= 1 &&
5650                                         vmcs12->cr3_target_value0 == val) ||
5651                                 (vmcs12->cr3_target_count >= 2 &&
5652                                         vmcs12->cr3_target_value1 == val) ||
5653                                 (vmcs12->cr3_target_count >= 3 &&
5654                                         vmcs12->cr3_target_value2 == val) ||
5655                                 (vmcs12->cr3_target_count >= 4 &&
5656                                         vmcs12->cr3_target_value3 == val))
5657                                 return 0;
5658                         if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
5659                                 return 1;
5660                         break;
5661                 case 4:
5662                         if (vmcs12->cr4_guest_host_mask &
5663                             (vmcs12->cr4_read_shadow ^ val))
5664                                 return 1;
5665                         break;
5666                 case 8:
5667                         if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
5668                                 return 1;
5669                         break;
5670                 }
5671                 break;
5672         case 2: /* clts */
5673                 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
5674                     (vmcs12->cr0_read_shadow & X86_CR0_TS))
5675                         return 1;
5676                 break;
5677         case 1: /* mov from cr */
5678                 switch (cr) {
5679                 case 3:
5680                         if (vmcs12->cpu_based_vm_exec_control &
5681                             CPU_BASED_CR3_STORE_EXITING)
5682                                 return 1;
5683                         break;
5684                 case 8:
5685                         if (vmcs12->cpu_based_vm_exec_control &
5686                             CPU_BASED_CR8_STORE_EXITING)
5687                                 return 1;
5688                         break;
5689                 }
5690                 break;
5691         case 3: /* lmsw */
5692                 /*
5693                  * lmsw can change bits 1..3 of cr0, and only set bit 0 of
5694                  * cr0. Other attempted changes are ignored, with no exit.
5695                  */
5696                 if (vmcs12->cr0_guest_host_mask & 0xe &
5697                     (val ^ vmcs12->cr0_read_shadow))
5698                         return 1;
5699                 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
5700                     !(vmcs12->cr0_read_shadow & 0x1) &&
5701                     (val & 0x1))
5702                         return 1;
5703                 break;
5704         }
5705         return 0;
5706 }
5707
5708 /*
5709  * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
5710  * should handle it ourselves in L0 (and then continue L2). Only call this
5711  * when in is_guest_mode (L2).
5712  */
5713 static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu)
5714 {
5715         u32 exit_reason = vmcs_read32(VM_EXIT_REASON);
5716         u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
5717         struct vcpu_vmx *vmx = to_vmx(vcpu);
5718         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5719
5720         if (vmx->nested.nested_run_pending)
5721                 return 0;
5722
5723         if (unlikely(vmx->fail)) {
5724                 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
5725                                     vmcs_read32(VM_INSTRUCTION_ERROR));
5726                 return 1;
5727         }
5728
5729         switch (exit_reason) {
5730         case EXIT_REASON_EXCEPTION_NMI:
5731                 if (!is_exception(intr_info))
5732                         return 0;
5733                 else if (is_page_fault(intr_info))
5734                         return enable_ept;
5735                 return vmcs12->exception_bitmap &
5736                                 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
5737         case EXIT_REASON_EXTERNAL_INTERRUPT:
5738                 return 0;
5739         case EXIT_REASON_TRIPLE_FAULT:
5740                 return 1;
5741         case EXIT_REASON_PENDING_INTERRUPT:
5742         case EXIT_REASON_NMI_WINDOW:
5743                 /*
5744                  * prepare_vmcs02() set the CPU_BASED_VIRTUAL_INTR_PENDING bit
5745                  * (aka Interrupt Window Exiting) only when L1 turned it on,
5746                  * so if we got a PENDING_INTERRUPT exit, this must be for L1.
5747                  * Same for NMI Window Exiting.
5748                  */
5749                 return 1;
5750         case EXIT_REASON_TASK_SWITCH:
5751                 return 1;
5752         case EXIT_REASON_CPUID:
5753                 return 1;
5754         case EXIT_REASON_HLT:
5755                 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
5756         case EXIT_REASON_INVD:
5757                 return 1;
5758         case EXIT_REASON_INVLPG:
5759                 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5760         case EXIT_REASON_RDPMC:
5761                 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
5762         case EXIT_REASON_RDTSC:
5763                 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
5764         case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
5765         case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
5766         case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
5767         case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
5768         case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
5769                 /*
5770                  * VMX instructions trap unconditionally. This allows L1 to
5771                  * emulate them for its L2 guest, i.e., allows 3-level nesting!
5772                  */
5773                 return 1;
5774         case EXIT_REASON_CR_ACCESS:
5775                 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
5776         case EXIT_REASON_DR_ACCESS:
5777                 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
5778         case EXIT_REASON_IO_INSTRUCTION:
5779                 /* TODO: support IO bitmaps */
5780                 return 1;
5781         case EXIT_REASON_MSR_READ:
5782         case EXIT_REASON_MSR_WRITE:
5783                 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
5784         case EXIT_REASON_INVALID_STATE:
5785                 return 1;
5786         case EXIT_REASON_MWAIT_INSTRUCTION:
5787                 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
5788         case EXIT_REASON_MONITOR_INSTRUCTION:
5789                 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
5790         case EXIT_REASON_PAUSE_INSTRUCTION:
5791                 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
5792                         nested_cpu_has2(vmcs12,
5793                                 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
5794         case EXIT_REASON_MCE_DURING_VMENTRY:
5795                 return 0;
5796         case EXIT_REASON_TPR_BELOW_THRESHOLD:
5797                 return 1;
5798         case EXIT_REASON_APIC_ACCESS:
5799                 return nested_cpu_has2(vmcs12,
5800                         SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
5801         case EXIT_REASON_EPT_VIOLATION:
5802         case EXIT_REASON_EPT_MISCONFIG:
5803                 return 0;
5804         case EXIT_REASON_WBINVD:
5805                 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
5806         case EXIT_REASON_XSETBV:
5807                 return 1;
5808         default:
5809                 return 1;
5810         }
5811 }
5812
5813 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
5814 {
5815         *info1 = vmcs_readl(EXIT_QUALIFICATION);
5816         *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
5817 }
5818
5819 /*
5820  * The guest has exited.  See if we can fix it or if we need userspace
5821  * assistance.
5822  */
5823 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
5824 {
5825         struct vcpu_vmx *vmx = to_vmx(vcpu);
5826         u32 exit_reason = vmx->exit_reason;
5827         u32 vectoring_info = vmx->idt_vectoring_info;
5828
5829         /* If guest state is invalid, start emulating */
5830         if (vmx->emulation_required && emulate_invalid_guest_state)
5831                 return handle_invalid_guest_state(vcpu);
5832
5833         /*
5834          * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
5835          * we did not inject a still-pending event to L1 now because of
5836          * nested_run_pending, we need to re-enable this bit.
5837          */
5838         if (vmx->nested.nested_run_pending)
5839                 kvm_make_request(KVM_REQ_EVENT, vcpu);
5840
5841         if (!is_guest_mode(vcpu) && (exit_reason == EXIT_REASON_VMLAUNCH ||
5842             exit_reason == EXIT_REASON_VMRESUME))
5843                 vmx->nested.nested_run_pending = 1;
5844         else
5845                 vmx->nested.nested_run_pending = 0;
5846
5847         if (is_guest_mode(vcpu) && nested_vmx_exit_handled(vcpu)) {
5848                 nested_vmx_vmexit(vcpu);
5849                 return 1;
5850         }
5851
5852         if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
5853                 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5854                 vcpu->run->fail_entry.hardware_entry_failure_reason
5855                         = exit_reason;
5856                 return 0;
5857         }
5858
5859         if (unlikely(vmx->fail)) {
5860                 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5861                 vcpu->run->fail_entry.hardware_entry_failure_reason
5862                         = vmcs_read32(VM_INSTRUCTION_ERROR);
5863                 return 0;
5864         }
5865
5866         if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
5867                         (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
5868                         exit_reason != EXIT_REASON_EPT_VIOLATION &&
5869                         exit_reason != EXIT_REASON_TASK_SWITCH))
5870                 printk(KERN_WARNING "%s: unexpected, valid vectoring info "
5871                        "(0x%x) and exit reason is 0x%x\n",
5872                        __func__, vectoring_info, exit_reason);
5873
5874         if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked &&
5875             !(is_guest_mode(vcpu) && nested_cpu_has_virtual_nmis(
5876                                         get_vmcs12(vcpu), vcpu)))) {
5877                 if (vmx_interrupt_allowed(vcpu)) {
5878                         vmx->soft_vnmi_blocked = 0;
5879                 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
5880                            vcpu->arch.nmi_pending) {
5881                         /*
5882                          * This CPU don't support us in finding the end of an
5883                          * NMI-blocked window if the guest runs with IRQs
5884                          * disabled. So we pull the trigger after 1 s of
5885                          * futile waiting, but inform the user about this.
5886                          */
5887                         printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
5888                                "state on VCPU %d after 1 s timeout\n",
5889                                __func__, vcpu->vcpu_id);
5890                         vmx->soft_vnmi_blocked = 0;
5891                 }
5892         }
5893
5894         if (exit_reason < kvm_vmx_max_exit_handlers
5895             && kvm_vmx_exit_handlers[exit_reason])
5896                 return kvm_vmx_exit_handlers[exit_reason](vcpu);
5897         else {
5898                 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5899                 vcpu->run->hw.hardware_exit_reason = exit_reason;
5900         }
5901         return 0;
5902 }
5903
5904 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
5905 {
5906         if (irr == -1 || tpr < irr) {
5907                 vmcs_write32(TPR_THRESHOLD, 0);
5908                 return;
5909         }
5910
5911         vmcs_write32(TPR_THRESHOLD, irr);
5912 }
5913
5914 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
5915 {
5916         u32 exit_intr_info;
5917
5918         if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
5919               || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
5920                 return;
5921
5922         vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
5923         exit_intr_info = vmx->exit_intr_info;
5924
5925         /* Handle machine checks before interrupts are enabled */
5926         if (is_machine_check(exit_intr_info))
5927                 kvm_machine_check();
5928
5929         /* We need to handle NMIs before interrupts are enabled */
5930         if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
5931             (exit_intr_info & INTR_INFO_VALID_MASK)) {
5932                 kvm_before_handle_nmi(&vmx->vcpu);
5933                 asm("int $2");
5934                 kvm_after_handle_nmi(&vmx->vcpu);
5935         }
5936 }
5937
5938 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
5939 {
5940         u32 exit_intr_info;
5941         bool unblock_nmi;
5942         u8 vector;
5943         bool idtv_info_valid;
5944
5945         idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
5946
5947         if (cpu_has_virtual_nmis()) {
5948                 if (vmx->nmi_known_unmasked)
5949                         return;
5950                 /*
5951                  * Can't use vmx->exit_intr_info since we're not sure what
5952                  * the exit reason is.
5953                  */
5954                 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
5955                 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
5956                 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
5957                 /*
5958                  * SDM 3: 27.7.1.2 (September 2008)
5959                  * Re-set bit "block by NMI" before VM entry if vmexit caused by
5960                  * a guest IRET fault.
5961                  * SDM 3: 23.2.2 (September 2008)
5962                  * Bit 12 is undefined in any of the following cases:
5963                  *  If the VM exit sets the valid bit in the IDT-vectoring
5964                  *   information field.
5965                  *  If the VM exit is due to a double fault.
5966                  */
5967                 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
5968                     vector != DF_VECTOR && !idtv_info_valid)
5969                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5970                                       GUEST_INTR_STATE_NMI);
5971                 else
5972                         vmx->nmi_known_unmasked =
5973                                 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
5974                                   & GUEST_INTR_STATE_NMI);
5975         } else if (unlikely(vmx->soft_vnmi_blocked))
5976                 vmx->vnmi_blocked_time +=
5977                         ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
5978 }
5979
5980 static void __vmx_complete_interrupts(struct vcpu_vmx *vmx,
5981                                       u32 idt_vectoring_info,
5982                                       int instr_len_field,
5983                                       int error_code_field)
5984 {
5985         u8 vector;
5986         int type;
5987         bool idtv_info_valid;
5988
5989         idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
5990
5991         vmx->vcpu.arch.nmi_injected = false;
5992         kvm_clear_exception_queue(&vmx->vcpu);
5993         kvm_clear_interrupt_queue(&vmx->vcpu);
5994
5995         if (!idtv_info_valid)
5996                 return;
5997
5998         kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
5999
6000         vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6001         type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
6002
6003         switch (type) {
6004         case INTR_TYPE_NMI_INTR:
6005                 vmx->vcpu.arch.nmi_injected = true;
6006                 /*
6007                  * SDM 3: 27.7.1.2 (September 2008)
6008                  * Clear bit "block by NMI" before VM entry if a NMI
6009                  * delivery faulted.
6010                  */
6011                 vmx_set_nmi_mask(&vmx->vcpu, false);
6012                 break;
6013         case INTR_TYPE_SOFT_EXCEPTION:
6014                 vmx->vcpu.arch.event_exit_inst_len =
6015                         vmcs_read32(instr_len_field);
6016                 /* fall through */
6017         case INTR_TYPE_HARD_EXCEPTION:
6018                 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
6019                         u32 err = vmcs_read32(error_code_field);
6020                         kvm_queue_exception_e(&vmx->vcpu, vector, err);
6021                 } else
6022                         kvm_queue_exception(&vmx->vcpu, vector);
6023                 break;
6024         case INTR_TYPE_SOFT_INTR:
6025                 vmx->vcpu.arch.event_exit_inst_len =
6026                         vmcs_read32(instr_len_field);
6027                 /* fall through */
6028         case INTR_TYPE_EXT_INTR:
6029                 kvm_queue_interrupt(&vmx->vcpu, vector,
6030                         type == INTR_TYPE_SOFT_INTR);
6031                 break;
6032         default:
6033                 break;
6034         }
6035 }
6036
6037 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6038 {
6039         if (is_guest_mode(&vmx->vcpu))
6040                 return;
6041         __vmx_complete_interrupts(vmx, vmx->idt_vectoring_info,
6042                                   VM_EXIT_INSTRUCTION_LEN,
6043                                   IDT_VECTORING_ERROR_CODE);
6044 }
6045
6046 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6047 {
6048         if (is_guest_mode(vcpu))
6049                 return;
6050         __vmx_complete_interrupts(to_vmx(vcpu),
6051                                   vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6052                                   VM_ENTRY_INSTRUCTION_LEN,
6053                                   VM_ENTRY_EXCEPTION_ERROR_CODE);
6054
6055         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6056 }
6057
6058 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6059 {
6060         int i, nr_msrs;
6061         struct perf_guest_switch_msr *msrs;
6062
6063         msrs = perf_guest_get_msrs(&nr_msrs);
6064
6065         if (!msrs)
6066                 return;
6067
6068         for (i = 0; i < nr_msrs; i++)
6069                 if (msrs[i].host == msrs[i].guest)
6070                         clear_atomic_switch_msr(vmx, msrs[i].msr);
6071                 else
6072                         add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6073                                         msrs[i].host);
6074 }
6075
6076 #ifdef CONFIG_X86_64
6077 #define R "r"
6078 #define Q "q"
6079 #else
6080 #define R "e"
6081 #define Q "l"
6082 #endif
6083
6084 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
6085 {
6086         struct vcpu_vmx *vmx = to_vmx(vcpu);
6087
6088         if (is_guest_mode(vcpu) && !vmx->nested.nested_run_pending) {
6089                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6090                 if (vmcs12->idt_vectoring_info_field &
6091                                 VECTORING_INFO_VALID_MASK) {
6092                         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
6093                                 vmcs12->idt_vectoring_info_field);
6094                         vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
6095                                 vmcs12->vm_exit_instruction_len);
6096                         if (vmcs12->idt_vectoring_info_field &
6097                                         VECTORING_INFO_DELIVER_CODE_MASK)
6098                                 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
6099                                         vmcs12->idt_vectoring_error_code);
6100                 }
6101         }
6102
6103         /* Record the guest's net vcpu time for enforced NMI injections. */
6104         if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
6105                 vmx->entry_time = ktime_get();
6106
6107         /* Don't enter VMX if guest state is invalid, let the exit handler
6108            start emulation until we arrive back to a valid state */
6109         if (vmx->emulation_required && emulate_invalid_guest_state)
6110                 return;
6111
6112         if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
6113                 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
6114         if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
6115                 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
6116
6117         /* When single-stepping over STI and MOV SS, we must clear the
6118          * corresponding interruptibility bits in the guest state. Otherwise
6119          * vmentry fails as it then expects bit 14 (BS) in pending debug
6120          * exceptions being set, but that's not correct for the guest debugging
6121          * case. */
6122         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6123                 vmx_set_interrupt_shadow(vcpu, 0);
6124
6125         atomic_switch_perf_msrs(vmx);
6126
6127         vmx->__launched = vmx->loaded_vmcs->launched;
6128         asm(
6129                 /* Store host registers */
6130                 "push %%"R"dx; push %%"R"bp;"
6131                 "push %%"R"cx \n\t" /* placeholder for guest rcx */
6132                 "push %%"R"cx \n\t"
6133                 "cmp %%"R"sp, %c[host_rsp](%0) \n\t"
6134                 "je 1f \n\t"
6135                 "mov %%"R"sp, %c[host_rsp](%0) \n\t"
6136                 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
6137                 "1: \n\t"
6138                 /* Reload cr2 if changed */
6139                 "mov %c[cr2](%0), %%"R"ax \n\t"
6140                 "mov %%cr2, %%"R"dx \n\t"
6141                 "cmp %%"R"ax, %%"R"dx \n\t"
6142                 "je 2f \n\t"
6143                 "mov %%"R"ax, %%cr2 \n\t"
6144                 "2: \n\t"
6145                 /* Check if vmlaunch of vmresume is needed */
6146                 "cmpl $0, %c[launched](%0) \n\t"
6147                 /* Load guest registers.  Don't clobber flags. */
6148                 "mov %c[rax](%0), %%"R"ax \n\t"
6149                 "mov %c[rbx](%0), %%"R"bx \n\t"
6150                 "mov %c[rdx](%0), %%"R"dx \n\t"
6151                 "mov %c[rsi](%0), %%"R"si \n\t"
6152                 "mov %c[rdi](%0), %%"R"di \n\t"
6153                 "mov %c[rbp](%0), %%"R"bp \n\t"
6154 #ifdef CONFIG_X86_64
6155                 "mov %c[r8](%0),  %%r8  \n\t"
6156                 "mov %c[r9](%0),  %%r9  \n\t"
6157                 "mov %c[r10](%0), %%r10 \n\t"
6158                 "mov %c[r11](%0), %%r11 \n\t"
6159                 "mov %c[r12](%0), %%r12 \n\t"
6160                 "mov %c[r13](%0), %%r13 \n\t"
6161                 "mov %c[r14](%0), %%r14 \n\t"
6162                 "mov %c[r15](%0), %%r15 \n\t"
6163 #endif
6164                 "mov %c[rcx](%0), %%"R"cx \n\t" /* kills %0 (ecx) */
6165
6166                 /* Enter guest mode */
6167                 "jne .Llaunched \n\t"
6168                 __ex(ASM_VMX_VMLAUNCH) "\n\t"
6169                 "jmp .Lkvm_vmx_return \n\t"
6170                 ".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t"
6171                 ".Lkvm_vmx_return: "
6172                 /* Save guest registers, load host registers, keep flags */
6173                 "mov %0, %c[wordsize](%%"R"sp) \n\t"
6174                 "pop %0 \n\t"
6175                 "mov %%"R"ax, %c[rax](%0) \n\t"
6176                 "mov %%"R"bx, %c[rbx](%0) \n\t"
6177                 "pop"Q" %c[rcx](%0) \n\t"
6178                 "mov %%"R"dx, %c[rdx](%0) \n\t"
6179                 "mov %%"R"si, %c[rsi](%0) \n\t"
6180                 "mov %%"R"di, %c[rdi](%0) \n\t"
6181                 "mov %%"R"bp, %c[rbp](%0) \n\t"
6182 #ifdef CONFIG_X86_64
6183                 "mov %%r8,  %c[r8](%0) \n\t"
6184                 "mov %%r9,  %c[r9](%0) \n\t"
6185                 "mov %%r10, %c[r10](%0) \n\t"
6186                 "mov %%r11, %c[r11](%0) \n\t"
6187                 "mov %%r12, %c[r12](%0) \n\t"
6188                 "mov %%r13, %c[r13](%0) \n\t"
6189                 "mov %%r14, %c[r14](%0) \n\t"
6190                 "mov %%r15, %c[r15](%0) \n\t"
6191 #endif
6192                 "mov %%cr2, %%"R"ax   \n\t"
6193                 "mov %%"R"ax, %c[cr2](%0) \n\t"
6194
6195                 "pop  %%"R"bp; pop  %%"R"dx \n\t"
6196                 "setbe %c[fail](%0) \n\t"
6197               : : "c"(vmx), "d"((unsigned long)HOST_RSP),
6198                 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
6199                 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
6200                 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
6201                 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
6202                 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
6203                 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
6204                 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
6205                 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
6206                 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
6207                 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
6208 #ifdef CONFIG_X86_64
6209                 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
6210                 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
6211                 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
6212                 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
6213                 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
6214                 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
6215                 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
6216                 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
6217 #endif
6218                 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
6219                 [wordsize]"i"(sizeof(ulong))
6220               : "cc", "memory"
6221                 , R"ax", R"bx", R"di", R"si"
6222 #ifdef CONFIG_X86_64
6223                 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
6224 #endif
6225               );
6226
6227         vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
6228                                   | (1 << VCPU_EXREG_RFLAGS)
6229                                   | (1 << VCPU_EXREG_CPL)
6230                                   | (1 << VCPU_EXREG_PDPTR)
6231                                   | (1 << VCPU_EXREG_SEGMENTS)
6232                                   | (1 << VCPU_EXREG_CR3));
6233         vcpu->arch.regs_dirty = 0;
6234
6235         vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
6236
6237         if (is_guest_mode(vcpu)) {
6238                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6239                 vmcs12->idt_vectoring_info_field = vmx->idt_vectoring_info;
6240                 if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) {
6241                         vmcs12->idt_vectoring_error_code =
6242                                 vmcs_read32(IDT_VECTORING_ERROR_CODE);
6243                         vmcs12->vm_exit_instruction_len =
6244                                 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
6245                 }
6246         }
6247
6248         asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
6249         vmx->loaded_vmcs->launched = 1;
6250
6251         vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
6252         trace_kvm_exit(vmx->exit_reason, vcpu, KVM_ISA_VMX);
6253
6254         vmx_complete_atomic_exit(vmx);
6255         vmx_recover_nmi_blocking(vmx);
6256         vmx_complete_interrupts(vmx);
6257 }
6258
6259 #undef R
6260 #undef Q
6261
6262 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
6263 {
6264         struct vcpu_vmx *vmx = to_vmx(vcpu);
6265
6266         free_vpid(vmx);
6267         free_nested(vmx);
6268         free_loaded_vmcs(vmx->loaded_vmcs);
6269         kfree(vmx->guest_msrs);
6270         kvm_vcpu_uninit(vcpu);
6271         kmem_cache_free(kvm_vcpu_cache, vmx);
6272 }
6273
6274 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
6275 {
6276         int err;
6277         struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
6278         int cpu;
6279
6280         if (!vmx)
6281                 return ERR_PTR(-ENOMEM);
6282
6283         allocate_vpid(vmx);
6284
6285         err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
6286         if (err)
6287                 goto free_vcpu;
6288
6289         vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
6290         err = -ENOMEM;
6291         if (!vmx->guest_msrs) {
6292                 goto uninit_vcpu;
6293         }
6294
6295         vmx->loaded_vmcs = &vmx->vmcs01;
6296         vmx->loaded_vmcs->vmcs = alloc_vmcs();
6297         if (!vmx->loaded_vmcs->vmcs)
6298                 goto free_msrs;
6299         if (!vmm_exclusive)
6300                 kvm_cpu_vmxon(__pa(per_cpu(vmxarea, raw_smp_processor_id())));
6301         loaded_vmcs_init(vmx->loaded_vmcs);
6302         if (!vmm_exclusive)
6303                 kvm_cpu_vmxoff();
6304
6305         cpu = get_cpu();
6306         vmx_vcpu_load(&vmx->vcpu, cpu);
6307         vmx->vcpu.cpu = cpu;
6308         err = vmx_vcpu_setup(vmx);
6309         vmx_vcpu_put(&vmx->vcpu);
6310         put_cpu();
6311         if (err)
6312                 goto free_vmcs;
6313         if (vm_need_virtualize_apic_accesses(kvm))
6314                 err = alloc_apic_access_page(kvm);
6315                 if (err)
6316      &