Merge tag 'kvm-3.7-1' of git://git.kernel.org/pub/scm/virt/kvm/kvm
[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 #include "cpuid.h"
22
23 #include <linux/kvm_host.h>
24 #include <linux/module.h>
25 #include <linux/kernel.h>
26 #include <linux/mm.h>
27 #include <linux/highmem.h>
28 #include <linux/sched.h>
29 #include <linux/moduleparam.h>
30 #include <linux/mod_devicetable.h>
31 #include <linux/ftrace_event.h>
32 #include <linux/slab.h>
33 #include <linux/tboot.h>
34 #include "kvm_cache_regs.h"
35 #include "x86.h"
36
37 #include <asm/io.h>
38 #include <asm/desc.h>
39 #include <asm/vmx.h>
40 #include <asm/virtext.h>
41 #include <asm/mce.h>
42 #include <asm/i387.h>
43 #include <asm/xcr.h>
44 #include <asm/perf_event.h>
45
46 #include "trace.h"
47
48 #define __ex(x) __kvm_handle_fault_on_reboot(x)
49 #define __ex_clear(x, reg) \
50         ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
51
52 MODULE_AUTHOR("Qumranet");
53 MODULE_LICENSE("GPL");
54
55 static const struct x86_cpu_id vmx_cpu_id[] = {
56         X86_FEATURE_MATCH(X86_FEATURE_VMX),
57         {}
58 };
59 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
60
61 static bool __read_mostly enable_vpid = 1;
62 module_param_named(vpid, enable_vpid, bool, 0444);
63
64 static bool __read_mostly flexpriority_enabled = 1;
65 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
66
67 static bool __read_mostly enable_ept = 1;
68 module_param_named(ept, enable_ept, bool, S_IRUGO);
69
70 static bool __read_mostly enable_unrestricted_guest = 1;
71 module_param_named(unrestricted_guest,
72                         enable_unrestricted_guest, bool, S_IRUGO);
73
74 static bool __read_mostly enable_ept_ad_bits = 1;
75 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
76
77 static bool __read_mostly emulate_invalid_guest_state = true;
78 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
79
80 static bool __read_mostly vmm_exclusive = 1;
81 module_param(vmm_exclusive, bool, S_IRUGO);
82
83 static bool __read_mostly fasteoi = 1;
84 module_param(fasteoi, bool, S_IRUGO);
85
86 /*
87  * If nested=1, nested virtualization is supported, i.e., guests may use
88  * VMX and be a hypervisor for its own guests. If nested=0, guests may not
89  * use VMX instructions.
90  */
91 static bool __read_mostly nested = 0;
92 module_param(nested, bool, S_IRUGO);
93
94 #define KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST                           \
95         (X86_CR0_WP | X86_CR0_NE | X86_CR0_NW | X86_CR0_CD)
96 #define KVM_GUEST_CR0_MASK                                              \
97         (KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
98 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST                         \
99         (X86_CR0_WP | X86_CR0_NE)
100 #define KVM_VM_CR0_ALWAYS_ON                                            \
101         (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
102 #define KVM_CR4_GUEST_OWNED_BITS                                      \
103         (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR      \
104          | X86_CR4_OSXMMEXCPT)
105
106 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
107 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
108
109 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
110
111 /*
112  * These 2 parameters are used to config the controls for Pause-Loop Exiting:
113  * ple_gap:    upper bound on the amount of time between two successive
114  *             executions of PAUSE in a loop. Also indicate if ple enabled.
115  *             According to test, this time is usually smaller than 128 cycles.
116  * ple_window: upper bound on the amount of time a guest is allowed to execute
117  *             in a PAUSE loop. Tests indicate that most spinlocks are held for
118  *             less than 2^12 cycles
119  * Time is measured based on a counter that runs at the same rate as the TSC,
120  * refer SDM volume 3b section 21.6.13 & 22.1.3.
121  */
122 #define KVM_VMX_DEFAULT_PLE_GAP    128
123 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
124 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
125 module_param(ple_gap, int, S_IRUGO);
126
127 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
128 module_param(ple_window, int, S_IRUGO);
129
130 extern const ulong vmx_return;
131
132 #define NR_AUTOLOAD_MSRS 8
133 #define VMCS02_POOL_SIZE 1
134
135 struct vmcs {
136         u32 revision_id;
137         u32 abort;
138         char data[0];
139 };
140
141 /*
142  * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
143  * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
144  * loaded on this CPU (so we can clear them if the CPU goes down).
145  */
146 struct loaded_vmcs {
147         struct vmcs *vmcs;
148         int cpu;
149         int launched;
150         struct list_head loaded_vmcss_on_cpu_link;
151 };
152
153 struct shared_msr_entry {
154         unsigned index;
155         u64 data;
156         u64 mask;
157 };
158
159 /*
160  * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
161  * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
162  * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
163  * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
164  * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
165  * More than one of these structures may exist, if L1 runs multiple L2 guests.
166  * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
167  * underlying hardware which will be used to run L2.
168  * This structure is packed to ensure that its layout is identical across
169  * machines (necessary for live migration).
170  * If there are changes in this struct, VMCS12_REVISION must be changed.
171  */
172 typedef u64 natural_width;
173 struct __packed vmcs12 {
174         /* According to the Intel spec, a VMCS region must start with the
175          * following two fields. Then follow implementation-specific data.
176          */
177         u32 revision_id;
178         u32 abort;
179
180         u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
181         u32 padding[7]; /* room for future expansion */
182
183         u64 io_bitmap_a;
184         u64 io_bitmap_b;
185         u64 msr_bitmap;
186         u64 vm_exit_msr_store_addr;
187         u64 vm_exit_msr_load_addr;
188         u64 vm_entry_msr_load_addr;
189         u64 tsc_offset;
190         u64 virtual_apic_page_addr;
191         u64 apic_access_addr;
192         u64 ept_pointer;
193         u64 guest_physical_address;
194         u64 vmcs_link_pointer;
195         u64 guest_ia32_debugctl;
196         u64 guest_ia32_pat;
197         u64 guest_ia32_efer;
198         u64 guest_ia32_perf_global_ctrl;
199         u64 guest_pdptr0;
200         u64 guest_pdptr1;
201         u64 guest_pdptr2;
202         u64 guest_pdptr3;
203         u64 host_ia32_pat;
204         u64 host_ia32_efer;
205         u64 host_ia32_perf_global_ctrl;
206         u64 padding64[8]; /* room for future expansion */
207         /*
208          * To allow migration of L1 (complete with its L2 guests) between
209          * machines of different natural widths (32 or 64 bit), we cannot have
210          * unsigned long fields with no explict size. We use u64 (aliased
211          * natural_width) instead. Luckily, x86 is little-endian.
212          */
213         natural_width cr0_guest_host_mask;
214         natural_width cr4_guest_host_mask;
215         natural_width cr0_read_shadow;
216         natural_width cr4_read_shadow;
217         natural_width cr3_target_value0;
218         natural_width cr3_target_value1;
219         natural_width cr3_target_value2;
220         natural_width cr3_target_value3;
221         natural_width exit_qualification;
222         natural_width guest_linear_address;
223         natural_width guest_cr0;
224         natural_width guest_cr3;
225         natural_width guest_cr4;
226         natural_width guest_es_base;
227         natural_width guest_cs_base;
228         natural_width guest_ss_base;
229         natural_width guest_ds_base;
230         natural_width guest_fs_base;
231         natural_width guest_gs_base;
232         natural_width guest_ldtr_base;
233         natural_width guest_tr_base;
234         natural_width guest_gdtr_base;
235         natural_width guest_idtr_base;
236         natural_width guest_dr7;
237         natural_width guest_rsp;
238         natural_width guest_rip;
239         natural_width guest_rflags;
240         natural_width guest_pending_dbg_exceptions;
241         natural_width guest_sysenter_esp;
242         natural_width guest_sysenter_eip;
243         natural_width host_cr0;
244         natural_width host_cr3;
245         natural_width host_cr4;
246         natural_width host_fs_base;
247         natural_width host_gs_base;
248         natural_width host_tr_base;
249         natural_width host_gdtr_base;
250         natural_width host_idtr_base;
251         natural_width host_ia32_sysenter_esp;
252         natural_width host_ia32_sysenter_eip;
253         natural_width host_rsp;
254         natural_width host_rip;
255         natural_width paddingl[8]; /* room for future expansion */
256         u32 pin_based_vm_exec_control;
257         u32 cpu_based_vm_exec_control;
258         u32 exception_bitmap;
259         u32 page_fault_error_code_mask;
260         u32 page_fault_error_code_match;
261         u32 cr3_target_count;
262         u32 vm_exit_controls;
263         u32 vm_exit_msr_store_count;
264         u32 vm_exit_msr_load_count;
265         u32 vm_entry_controls;
266         u32 vm_entry_msr_load_count;
267         u32 vm_entry_intr_info_field;
268         u32 vm_entry_exception_error_code;
269         u32 vm_entry_instruction_len;
270         u32 tpr_threshold;
271         u32 secondary_vm_exec_control;
272         u32 vm_instruction_error;
273         u32 vm_exit_reason;
274         u32 vm_exit_intr_info;
275         u32 vm_exit_intr_error_code;
276         u32 idt_vectoring_info_field;
277         u32 idt_vectoring_error_code;
278         u32 vm_exit_instruction_len;
279         u32 vmx_instruction_info;
280         u32 guest_es_limit;
281         u32 guest_cs_limit;
282         u32 guest_ss_limit;
283         u32 guest_ds_limit;
284         u32 guest_fs_limit;
285         u32 guest_gs_limit;
286         u32 guest_ldtr_limit;
287         u32 guest_tr_limit;
288         u32 guest_gdtr_limit;
289         u32 guest_idtr_limit;
290         u32 guest_es_ar_bytes;
291         u32 guest_cs_ar_bytes;
292         u32 guest_ss_ar_bytes;
293         u32 guest_ds_ar_bytes;
294         u32 guest_fs_ar_bytes;
295         u32 guest_gs_ar_bytes;
296         u32 guest_ldtr_ar_bytes;
297         u32 guest_tr_ar_bytes;
298         u32 guest_interruptibility_info;
299         u32 guest_activity_state;
300         u32 guest_sysenter_cs;
301         u32 host_ia32_sysenter_cs;
302         u32 padding32[8]; /* room for future expansion */
303         u16 virtual_processor_id;
304         u16 guest_es_selector;
305         u16 guest_cs_selector;
306         u16 guest_ss_selector;
307         u16 guest_ds_selector;
308         u16 guest_fs_selector;
309         u16 guest_gs_selector;
310         u16 guest_ldtr_selector;
311         u16 guest_tr_selector;
312         u16 host_es_selector;
313         u16 host_cs_selector;
314         u16 host_ss_selector;
315         u16 host_ds_selector;
316         u16 host_fs_selector;
317         u16 host_gs_selector;
318         u16 host_tr_selector;
319 };
320
321 /*
322  * VMCS12_REVISION is an arbitrary id that should be changed if the content or
323  * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
324  * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
325  */
326 #define VMCS12_REVISION 0x11e57ed0
327
328 /*
329  * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
330  * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
331  * current implementation, 4K are reserved to avoid future complications.
332  */
333 #define VMCS12_SIZE 0x1000
334
335 /* Used to remember the last vmcs02 used for some recently used vmcs12s */
336 struct vmcs02_list {
337         struct list_head list;
338         gpa_t vmptr;
339         struct loaded_vmcs vmcs02;
340 };
341
342 /*
343  * The nested_vmx structure is part of vcpu_vmx, and holds information we need
344  * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
345  */
346 struct nested_vmx {
347         /* Has the level1 guest done vmxon? */
348         bool vmxon;
349
350         /* The guest-physical address of the current VMCS L1 keeps for L2 */
351         gpa_t current_vmptr;
352         /* The host-usable pointer to the above */
353         struct page *current_vmcs12_page;
354         struct vmcs12 *current_vmcs12;
355
356         /* vmcs02_list cache of VMCSs recently used to run L2 guests */
357         struct list_head vmcs02_pool;
358         int vmcs02_num;
359         u64 vmcs01_tsc_offset;
360         /* L2 must run next, and mustn't decide to exit to L1. */
361         bool nested_run_pending;
362         /*
363          * Guest pages referred to in vmcs02 with host-physical pointers, so
364          * we must keep them pinned while L2 runs.
365          */
366         struct page *apic_access_page;
367 };
368
369 struct vcpu_vmx {
370         struct kvm_vcpu       vcpu;
371         unsigned long         host_rsp;
372         u8                    fail;
373         u8                    cpl;
374         bool                  nmi_known_unmasked;
375         u32                   exit_intr_info;
376         u32                   idt_vectoring_info;
377         ulong                 rflags;
378         struct shared_msr_entry *guest_msrs;
379         int                   nmsrs;
380         int                   save_nmsrs;
381 #ifdef CONFIG_X86_64
382         u64                   msr_host_kernel_gs_base;
383         u64                   msr_guest_kernel_gs_base;
384 #endif
385         /*
386          * loaded_vmcs points to the VMCS currently used in this vcpu. For a
387          * non-nested (L1) guest, it always points to vmcs01. For a nested
388          * guest (L2), it points to a different VMCS.
389          */
390         struct loaded_vmcs    vmcs01;
391         struct loaded_vmcs   *loaded_vmcs;
392         bool                  __launched; /* temporary, used in vmx_vcpu_run */
393         struct msr_autoload {
394                 unsigned nr;
395                 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
396                 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
397         } msr_autoload;
398         struct {
399                 int           loaded;
400                 u16           fs_sel, gs_sel, ldt_sel;
401 #ifdef CONFIG_X86_64
402                 u16           ds_sel, es_sel;
403 #endif
404                 int           gs_ldt_reload_needed;
405                 int           fs_reload_needed;
406         } host_state;
407         struct {
408                 int vm86_active;
409                 ulong save_rflags;
410                 struct kvm_segment segs[8];
411         } rmode;
412         struct {
413                 u32 bitmask; /* 4 bits per segment (1 bit per field) */
414                 struct kvm_save_segment {
415                         u16 selector;
416                         unsigned long base;
417                         u32 limit;
418                         u32 ar;
419                 } seg[8];
420         } segment_cache;
421         int vpid;
422         bool emulation_required;
423
424         /* Support for vnmi-less CPUs */
425         int soft_vnmi_blocked;
426         ktime_t entry_time;
427         s64 vnmi_blocked_time;
428         u32 exit_reason;
429
430         bool rdtscp_enabled;
431
432         /* Support for a guest hypervisor (nested VMX) */
433         struct nested_vmx nested;
434 };
435
436 enum segment_cache_field {
437         SEG_FIELD_SEL = 0,
438         SEG_FIELD_BASE = 1,
439         SEG_FIELD_LIMIT = 2,
440         SEG_FIELD_AR = 3,
441
442         SEG_FIELD_NR = 4
443 };
444
445 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
446 {
447         return container_of(vcpu, struct vcpu_vmx, vcpu);
448 }
449
450 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
451 #define FIELD(number, name)     [number] = VMCS12_OFFSET(name)
452 #define FIELD64(number, name)   [number] = VMCS12_OFFSET(name), \
453                                 [number##_HIGH] = VMCS12_OFFSET(name)+4
454
455 static const unsigned short vmcs_field_to_offset_table[] = {
456         FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
457         FIELD(GUEST_ES_SELECTOR, guest_es_selector),
458         FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
459         FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
460         FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
461         FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
462         FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
463         FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
464         FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
465         FIELD(HOST_ES_SELECTOR, host_es_selector),
466         FIELD(HOST_CS_SELECTOR, host_cs_selector),
467         FIELD(HOST_SS_SELECTOR, host_ss_selector),
468         FIELD(HOST_DS_SELECTOR, host_ds_selector),
469         FIELD(HOST_FS_SELECTOR, host_fs_selector),
470         FIELD(HOST_GS_SELECTOR, host_gs_selector),
471         FIELD(HOST_TR_SELECTOR, host_tr_selector),
472         FIELD64(IO_BITMAP_A, io_bitmap_a),
473         FIELD64(IO_BITMAP_B, io_bitmap_b),
474         FIELD64(MSR_BITMAP, msr_bitmap),
475         FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
476         FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
477         FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
478         FIELD64(TSC_OFFSET, tsc_offset),
479         FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
480         FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
481         FIELD64(EPT_POINTER, ept_pointer),
482         FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
483         FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
484         FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
485         FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
486         FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
487         FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
488         FIELD64(GUEST_PDPTR0, guest_pdptr0),
489         FIELD64(GUEST_PDPTR1, guest_pdptr1),
490         FIELD64(GUEST_PDPTR2, guest_pdptr2),
491         FIELD64(GUEST_PDPTR3, guest_pdptr3),
492         FIELD64(HOST_IA32_PAT, host_ia32_pat),
493         FIELD64(HOST_IA32_EFER, host_ia32_efer),
494         FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
495         FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
496         FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
497         FIELD(EXCEPTION_BITMAP, exception_bitmap),
498         FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
499         FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
500         FIELD(CR3_TARGET_COUNT, cr3_target_count),
501         FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
502         FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
503         FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
504         FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
505         FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
506         FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
507         FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
508         FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
509         FIELD(TPR_THRESHOLD, tpr_threshold),
510         FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
511         FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
512         FIELD(VM_EXIT_REASON, vm_exit_reason),
513         FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
514         FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
515         FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
516         FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
517         FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
518         FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
519         FIELD(GUEST_ES_LIMIT, guest_es_limit),
520         FIELD(GUEST_CS_LIMIT, guest_cs_limit),
521         FIELD(GUEST_SS_LIMIT, guest_ss_limit),
522         FIELD(GUEST_DS_LIMIT, guest_ds_limit),
523         FIELD(GUEST_FS_LIMIT, guest_fs_limit),
524         FIELD(GUEST_GS_LIMIT, guest_gs_limit),
525         FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
526         FIELD(GUEST_TR_LIMIT, guest_tr_limit),
527         FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
528         FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
529         FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
530         FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
531         FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
532         FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
533         FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
534         FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
535         FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
536         FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
537         FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
538         FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
539         FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
540         FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
541         FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
542         FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
543         FIELD(CR0_READ_SHADOW, cr0_read_shadow),
544         FIELD(CR4_READ_SHADOW, cr4_read_shadow),
545         FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
546         FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
547         FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
548         FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
549         FIELD(EXIT_QUALIFICATION, exit_qualification),
550         FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
551         FIELD(GUEST_CR0, guest_cr0),
552         FIELD(GUEST_CR3, guest_cr3),
553         FIELD(GUEST_CR4, guest_cr4),
554         FIELD(GUEST_ES_BASE, guest_es_base),
555         FIELD(GUEST_CS_BASE, guest_cs_base),
556         FIELD(GUEST_SS_BASE, guest_ss_base),
557         FIELD(GUEST_DS_BASE, guest_ds_base),
558         FIELD(GUEST_FS_BASE, guest_fs_base),
559         FIELD(GUEST_GS_BASE, guest_gs_base),
560         FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
561         FIELD(GUEST_TR_BASE, guest_tr_base),
562         FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
563         FIELD(GUEST_IDTR_BASE, guest_idtr_base),
564         FIELD(GUEST_DR7, guest_dr7),
565         FIELD(GUEST_RSP, guest_rsp),
566         FIELD(GUEST_RIP, guest_rip),
567         FIELD(GUEST_RFLAGS, guest_rflags),
568         FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
569         FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
570         FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
571         FIELD(HOST_CR0, host_cr0),
572         FIELD(HOST_CR3, host_cr3),
573         FIELD(HOST_CR4, host_cr4),
574         FIELD(HOST_FS_BASE, host_fs_base),
575         FIELD(HOST_GS_BASE, host_gs_base),
576         FIELD(HOST_TR_BASE, host_tr_base),
577         FIELD(HOST_GDTR_BASE, host_gdtr_base),
578         FIELD(HOST_IDTR_BASE, host_idtr_base),
579         FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
580         FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
581         FIELD(HOST_RSP, host_rsp),
582         FIELD(HOST_RIP, host_rip),
583 };
584 static const int max_vmcs_field = ARRAY_SIZE(vmcs_field_to_offset_table);
585
586 static inline short vmcs_field_to_offset(unsigned long field)
587 {
588         if (field >= max_vmcs_field || vmcs_field_to_offset_table[field] == 0)
589                 return -1;
590         return vmcs_field_to_offset_table[field];
591 }
592
593 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
594 {
595         return to_vmx(vcpu)->nested.current_vmcs12;
596 }
597
598 static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
599 {
600         struct page *page = gfn_to_page(vcpu->kvm, addr >> PAGE_SHIFT);
601         if (is_error_page(page))
602                 return NULL;
603
604         return page;
605 }
606
607 static void nested_release_page(struct page *page)
608 {
609         kvm_release_page_dirty(page);
610 }
611
612 static void nested_release_page_clean(struct page *page)
613 {
614         kvm_release_page_clean(page);
615 }
616
617 static u64 construct_eptp(unsigned long root_hpa);
618 static void kvm_cpu_vmxon(u64 addr);
619 static void kvm_cpu_vmxoff(void);
620 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3);
621 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
622 static void vmx_set_segment(struct kvm_vcpu *vcpu,
623                             struct kvm_segment *var, int seg);
624 static void vmx_get_segment(struct kvm_vcpu *vcpu,
625                             struct kvm_segment *var, int seg);
626
627 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
628 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
629 /*
630  * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
631  * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
632  */
633 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
634 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
635
636 static unsigned long *vmx_io_bitmap_a;
637 static unsigned long *vmx_io_bitmap_b;
638 static unsigned long *vmx_msr_bitmap_legacy;
639 static unsigned long *vmx_msr_bitmap_longmode;
640
641 static bool cpu_has_load_ia32_efer;
642 static bool cpu_has_load_perf_global_ctrl;
643
644 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
645 static DEFINE_SPINLOCK(vmx_vpid_lock);
646
647 static struct vmcs_config {
648         int size;
649         int order;
650         u32 revision_id;
651         u32 pin_based_exec_ctrl;
652         u32 cpu_based_exec_ctrl;
653         u32 cpu_based_2nd_exec_ctrl;
654         u32 vmexit_ctrl;
655         u32 vmentry_ctrl;
656 } vmcs_config;
657
658 static struct vmx_capability {
659         u32 ept;
660         u32 vpid;
661 } vmx_capability;
662
663 #define VMX_SEGMENT_FIELD(seg)                                  \
664         [VCPU_SREG_##seg] = {                                   \
665                 .selector = GUEST_##seg##_SELECTOR,             \
666                 .base = GUEST_##seg##_BASE,                     \
667                 .limit = GUEST_##seg##_LIMIT,                   \
668                 .ar_bytes = GUEST_##seg##_AR_BYTES,             \
669         }
670
671 static const struct kvm_vmx_segment_field {
672         unsigned selector;
673         unsigned base;
674         unsigned limit;
675         unsigned ar_bytes;
676 } kvm_vmx_segment_fields[] = {
677         VMX_SEGMENT_FIELD(CS),
678         VMX_SEGMENT_FIELD(DS),
679         VMX_SEGMENT_FIELD(ES),
680         VMX_SEGMENT_FIELD(FS),
681         VMX_SEGMENT_FIELD(GS),
682         VMX_SEGMENT_FIELD(SS),
683         VMX_SEGMENT_FIELD(TR),
684         VMX_SEGMENT_FIELD(LDTR),
685 };
686
687 static u64 host_efer;
688
689 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
690
691 /*
692  * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
693  * away by decrementing the array size.
694  */
695 static const u32 vmx_msr_index[] = {
696 #ifdef CONFIG_X86_64
697         MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
698 #endif
699         MSR_EFER, MSR_TSC_AUX, MSR_STAR,
700 };
701 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
702
703 static inline bool is_page_fault(u32 intr_info)
704 {
705         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
706                              INTR_INFO_VALID_MASK)) ==
707                 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
708 }
709
710 static inline bool is_no_device(u32 intr_info)
711 {
712         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
713                              INTR_INFO_VALID_MASK)) ==
714                 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
715 }
716
717 static inline bool is_invalid_opcode(u32 intr_info)
718 {
719         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
720                              INTR_INFO_VALID_MASK)) ==
721                 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
722 }
723
724 static inline bool is_external_interrupt(u32 intr_info)
725 {
726         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
727                 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
728 }
729
730 static inline bool is_machine_check(u32 intr_info)
731 {
732         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
733                              INTR_INFO_VALID_MASK)) ==
734                 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
735 }
736
737 static inline bool cpu_has_vmx_msr_bitmap(void)
738 {
739         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
740 }
741
742 static inline bool cpu_has_vmx_tpr_shadow(void)
743 {
744         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
745 }
746
747 static inline bool vm_need_tpr_shadow(struct kvm *kvm)
748 {
749         return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
750 }
751
752 static inline bool cpu_has_secondary_exec_ctrls(void)
753 {
754         return vmcs_config.cpu_based_exec_ctrl &
755                 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
756 }
757
758 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
759 {
760         return vmcs_config.cpu_based_2nd_exec_ctrl &
761                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
762 }
763
764 static inline bool cpu_has_vmx_flexpriority(void)
765 {
766         return cpu_has_vmx_tpr_shadow() &&
767                 cpu_has_vmx_virtualize_apic_accesses();
768 }
769
770 static inline bool cpu_has_vmx_ept_execute_only(void)
771 {
772         return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
773 }
774
775 static inline bool cpu_has_vmx_eptp_uncacheable(void)
776 {
777         return vmx_capability.ept & VMX_EPTP_UC_BIT;
778 }
779
780 static inline bool cpu_has_vmx_eptp_writeback(void)
781 {
782         return vmx_capability.ept & VMX_EPTP_WB_BIT;
783 }
784
785 static inline bool cpu_has_vmx_ept_2m_page(void)
786 {
787         return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
788 }
789
790 static inline bool cpu_has_vmx_ept_1g_page(void)
791 {
792         return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
793 }
794
795 static inline bool cpu_has_vmx_ept_4levels(void)
796 {
797         return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
798 }
799
800 static inline bool cpu_has_vmx_ept_ad_bits(void)
801 {
802         return vmx_capability.ept & VMX_EPT_AD_BIT;
803 }
804
805 static inline bool cpu_has_vmx_invept_individual_addr(void)
806 {
807         return vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT;
808 }
809
810 static inline bool cpu_has_vmx_invept_context(void)
811 {
812         return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
813 }
814
815 static inline bool cpu_has_vmx_invept_global(void)
816 {
817         return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
818 }
819
820 static inline bool cpu_has_vmx_invvpid_single(void)
821 {
822         return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
823 }
824
825 static inline bool cpu_has_vmx_invvpid_global(void)
826 {
827         return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
828 }
829
830 static inline bool cpu_has_vmx_ept(void)
831 {
832         return vmcs_config.cpu_based_2nd_exec_ctrl &
833                 SECONDARY_EXEC_ENABLE_EPT;
834 }
835
836 static inline bool cpu_has_vmx_unrestricted_guest(void)
837 {
838         return vmcs_config.cpu_based_2nd_exec_ctrl &
839                 SECONDARY_EXEC_UNRESTRICTED_GUEST;
840 }
841
842 static inline bool cpu_has_vmx_ple(void)
843 {
844         return vmcs_config.cpu_based_2nd_exec_ctrl &
845                 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
846 }
847
848 static inline bool vm_need_virtualize_apic_accesses(struct kvm *kvm)
849 {
850         return flexpriority_enabled && irqchip_in_kernel(kvm);
851 }
852
853 static inline bool cpu_has_vmx_vpid(void)
854 {
855         return vmcs_config.cpu_based_2nd_exec_ctrl &
856                 SECONDARY_EXEC_ENABLE_VPID;
857 }
858
859 static inline bool cpu_has_vmx_rdtscp(void)
860 {
861         return vmcs_config.cpu_based_2nd_exec_ctrl &
862                 SECONDARY_EXEC_RDTSCP;
863 }
864
865 static inline bool cpu_has_vmx_invpcid(void)
866 {
867         return vmcs_config.cpu_based_2nd_exec_ctrl &
868                 SECONDARY_EXEC_ENABLE_INVPCID;
869 }
870
871 static inline bool cpu_has_virtual_nmis(void)
872 {
873         return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
874 }
875
876 static inline bool cpu_has_vmx_wbinvd_exit(void)
877 {
878         return vmcs_config.cpu_based_2nd_exec_ctrl &
879                 SECONDARY_EXEC_WBINVD_EXITING;
880 }
881
882 static inline bool report_flexpriority(void)
883 {
884         return flexpriority_enabled;
885 }
886
887 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
888 {
889         return vmcs12->cpu_based_vm_exec_control & bit;
890 }
891
892 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
893 {
894         return (vmcs12->cpu_based_vm_exec_control &
895                         CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
896                 (vmcs12->secondary_vm_exec_control & bit);
897 }
898
899 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12,
900         struct kvm_vcpu *vcpu)
901 {
902         return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
903 }
904
905 static inline bool is_exception(u32 intr_info)
906 {
907         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
908                 == (INTR_TYPE_HARD_EXCEPTION | INTR_INFO_VALID_MASK);
909 }
910
911 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu);
912 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
913                         struct vmcs12 *vmcs12,
914                         u32 reason, unsigned long qualification);
915
916 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
917 {
918         int i;
919
920         for (i = 0; i < vmx->nmsrs; ++i)
921                 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
922                         return i;
923         return -1;
924 }
925
926 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
927 {
928     struct {
929         u64 vpid : 16;
930         u64 rsvd : 48;
931         u64 gva;
932     } operand = { vpid, 0, gva };
933
934     asm volatile (__ex(ASM_VMX_INVVPID)
935                   /* CF==1 or ZF==1 --> rc = -1 */
936                   "; ja 1f ; ud2 ; 1:"
937                   : : "a"(&operand), "c"(ext) : "cc", "memory");
938 }
939
940 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
941 {
942         struct {
943                 u64 eptp, gpa;
944         } operand = {eptp, gpa};
945
946         asm volatile (__ex(ASM_VMX_INVEPT)
947                         /* CF==1 or ZF==1 --> rc = -1 */
948                         "; ja 1f ; ud2 ; 1:\n"
949                         : : "a" (&operand), "c" (ext) : "cc", "memory");
950 }
951
952 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
953 {
954         int i;
955
956         i = __find_msr_index(vmx, msr);
957         if (i >= 0)
958                 return &vmx->guest_msrs[i];
959         return NULL;
960 }
961
962 static void vmcs_clear(struct vmcs *vmcs)
963 {
964         u64 phys_addr = __pa(vmcs);
965         u8 error;
966
967         asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
968                       : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
969                       : "cc", "memory");
970         if (error)
971                 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
972                        vmcs, phys_addr);
973 }
974
975 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
976 {
977         vmcs_clear(loaded_vmcs->vmcs);
978         loaded_vmcs->cpu = -1;
979         loaded_vmcs->launched = 0;
980 }
981
982 static void vmcs_load(struct vmcs *vmcs)
983 {
984         u64 phys_addr = __pa(vmcs);
985         u8 error;
986
987         asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
988                         : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
989                         : "cc", "memory");
990         if (error)
991                 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
992                        vmcs, phys_addr);
993 }
994
995 static void __loaded_vmcs_clear(void *arg)
996 {
997         struct loaded_vmcs *loaded_vmcs = arg;
998         int cpu = raw_smp_processor_id();
999
1000         if (loaded_vmcs->cpu != cpu)
1001                 return; /* vcpu migration can race with cpu offline */
1002         if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1003                 per_cpu(current_vmcs, cpu) = NULL;
1004         list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1005         loaded_vmcs_init(loaded_vmcs);
1006 }
1007
1008 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1009 {
1010         if (loaded_vmcs->cpu != -1)
1011                 smp_call_function_single(
1012                         loaded_vmcs->cpu, __loaded_vmcs_clear, loaded_vmcs, 1);
1013 }
1014
1015 static inline void vpid_sync_vcpu_single(struct vcpu_vmx *vmx)
1016 {
1017         if (vmx->vpid == 0)
1018                 return;
1019
1020         if (cpu_has_vmx_invvpid_single())
1021                 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
1022 }
1023
1024 static inline void vpid_sync_vcpu_global(void)
1025 {
1026         if (cpu_has_vmx_invvpid_global())
1027                 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1028 }
1029
1030 static inline void vpid_sync_context(struct vcpu_vmx *vmx)
1031 {
1032         if (cpu_has_vmx_invvpid_single())
1033                 vpid_sync_vcpu_single(vmx);
1034         else
1035                 vpid_sync_vcpu_global();
1036 }
1037
1038 static inline void ept_sync_global(void)
1039 {
1040         if (cpu_has_vmx_invept_global())
1041                 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1042 }
1043
1044 static inline void ept_sync_context(u64 eptp)
1045 {
1046         if (enable_ept) {
1047                 if (cpu_has_vmx_invept_context())
1048                         __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1049                 else
1050                         ept_sync_global();
1051         }
1052 }
1053
1054 static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
1055 {
1056         if (enable_ept) {
1057                 if (cpu_has_vmx_invept_individual_addr())
1058                         __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
1059                                         eptp, gpa);
1060                 else
1061                         ept_sync_context(eptp);
1062         }
1063 }
1064
1065 static __always_inline unsigned long vmcs_readl(unsigned long field)
1066 {
1067         unsigned long value;
1068
1069         asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1070                       : "=a"(value) : "d"(field) : "cc");
1071         return value;
1072 }
1073
1074 static __always_inline u16 vmcs_read16(unsigned long field)
1075 {
1076         return vmcs_readl(field);
1077 }
1078
1079 static __always_inline u32 vmcs_read32(unsigned long field)
1080 {
1081         return vmcs_readl(field);
1082 }
1083
1084 static __always_inline u64 vmcs_read64(unsigned long field)
1085 {
1086 #ifdef CONFIG_X86_64
1087         return vmcs_readl(field);
1088 #else
1089         return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
1090 #endif
1091 }
1092
1093 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1094 {
1095         printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1096                field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1097         dump_stack();
1098 }
1099
1100 static void vmcs_writel(unsigned long field, unsigned long value)
1101 {
1102         u8 error;
1103
1104         asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1105                        : "=q"(error) : "a"(value), "d"(field) : "cc");
1106         if (unlikely(error))
1107                 vmwrite_error(field, value);
1108 }
1109
1110 static void vmcs_write16(unsigned long field, u16 value)
1111 {
1112         vmcs_writel(field, value);
1113 }
1114
1115 static void vmcs_write32(unsigned long field, u32 value)
1116 {
1117         vmcs_writel(field, value);
1118 }
1119
1120 static void vmcs_write64(unsigned long field, u64 value)
1121 {
1122         vmcs_writel(field, value);
1123 #ifndef CONFIG_X86_64
1124         asm volatile ("");
1125         vmcs_writel(field+1, value >> 32);
1126 #endif
1127 }
1128
1129 static void vmcs_clear_bits(unsigned long field, u32 mask)
1130 {
1131         vmcs_writel(field, vmcs_readl(field) & ~mask);
1132 }
1133
1134 static void vmcs_set_bits(unsigned long field, u32 mask)
1135 {
1136         vmcs_writel(field, vmcs_readl(field) | mask);
1137 }
1138
1139 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1140 {
1141         vmx->segment_cache.bitmask = 0;
1142 }
1143
1144 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1145                                        unsigned field)
1146 {
1147         bool ret;
1148         u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1149
1150         if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1151                 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1152                 vmx->segment_cache.bitmask = 0;
1153         }
1154         ret = vmx->segment_cache.bitmask & mask;
1155         vmx->segment_cache.bitmask |= mask;
1156         return ret;
1157 }
1158
1159 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1160 {
1161         u16 *p = &vmx->segment_cache.seg[seg].selector;
1162
1163         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1164                 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1165         return *p;
1166 }
1167
1168 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1169 {
1170         ulong *p = &vmx->segment_cache.seg[seg].base;
1171
1172         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1173                 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1174         return *p;
1175 }
1176
1177 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1178 {
1179         u32 *p = &vmx->segment_cache.seg[seg].limit;
1180
1181         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1182                 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1183         return *p;
1184 }
1185
1186 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1187 {
1188         u32 *p = &vmx->segment_cache.seg[seg].ar;
1189
1190         if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1191                 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1192         return *p;
1193 }
1194
1195 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1196 {
1197         u32 eb;
1198
1199         eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1200              (1u << NM_VECTOR) | (1u << DB_VECTOR);
1201         if ((vcpu->guest_debug &
1202              (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1203             (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1204                 eb |= 1u << BP_VECTOR;
1205         if (to_vmx(vcpu)->rmode.vm86_active)
1206                 eb = ~0;
1207         if (enable_ept)
1208                 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1209         if (vcpu->fpu_active)
1210                 eb &= ~(1u << NM_VECTOR);
1211
1212         /* When we are running a nested L2 guest and L1 specified for it a
1213          * certain exception bitmap, we must trap the same exceptions and pass
1214          * them to L1. When running L2, we will only handle the exceptions
1215          * specified above if L1 did not want them.
1216          */
1217         if (is_guest_mode(vcpu))
1218                 eb |= get_vmcs12(vcpu)->exception_bitmap;
1219
1220         vmcs_write32(EXCEPTION_BITMAP, eb);
1221 }
1222
1223 static void clear_atomic_switch_msr_special(unsigned long entry,
1224                 unsigned long exit)
1225 {
1226         vmcs_clear_bits(VM_ENTRY_CONTROLS, entry);
1227         vmcs_clear_bits(VM_EXIT_CONTROLS, exit);
1228 }
1229
1230 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1231 {
1232         unsigned i;
1233         struct msr_autoload *m = &vmx->msr_autoload;
1234
1235         switch (msr) {
1236         case MSR_EFER:
1237                 if (cpu_has_load_ia32_efer) {
1238                         clear_atomic_switch_msr_special(VM_ENTRY_LOAD_IA32_EFER,
1239                                         VM_EXIT_LOAD_IA32_EFER);
1240                         return;
1241                 }
1242                 break;
1243         case MSR_CORE_PERF_GLOBAL_CTRL:
1244                 if (cpu_has_load_perf_global_ctrl) {
1245                         clear_atomic_switch_msr_special(
1246                                         VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1247                                         VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1248                         return;
1249                 }
1250                 break;
1251         }
1252
1253         for (i = 0; i < m->nr; ++i)
1254                 if (m->guest[i].index == msr)
1255                         break;
1256
1257         if (i == m->nr)
1258                 return;
1259         --m->nr;
1260         m->guest[i] = m->guest[m->nr];
1261         m->host[i] = m->host[m->nr];
1262         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1263         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1264 }
1265
1266 static void add_atomic_switch_msr_special(unsigned long entry,
1267                 unsigned long exit, unsigned long guest_val_vmcs,
1268                 unsigned long host_val_vmcs, u64 guest_val, u64 host_val)
1269 {
1270         vmcs_write64(guest_val_vmcs, guest_val);
1271         vmcs_write64(host_val_vmcs, host_val);
1272         vmcs_set_bits(VM_ENTRY_CONTROLS, entry);
1273         vmcs_set_bits(VM_EXIT_CONTROLS, exit);
1274 }
1275
1276 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1277                                   u64 guest_val, u64 host_val)
1278 {
1279         unsigned i;
1280         struct msr_autoload *m = &vmx->msr_autoload;
1281
1282         switch (msr) {
1283         case MSR_EFER:
1284                 if (cpu_has_load_ia32_efer) {
1285                         add_atomic_switch_msr_special(VM_ENTRY_LOAD_IA32_EFER,
1286                                         VM_EXIT_LOAD_IA32_EFER,
1287                                         GUEST_IA32_EFER,
1288                                         HOST_IA32_EFER,
1289                                         guest_val, host_val);
1290                         return;
1291                 }
1292                 break;
1293         case MSR_CORE_PERF_GLOBAL_CTRL:
1294                 if (cpu_has_load_perf_global_ctrl) {
1295                         add_atomic_switch_msr_special(
1296                                         VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1297                                         VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1298                                         GUEST_IA32_PERF_GLOBAL_CTRL,
1299                                         HOST_IA32_PERF_GLOBAL_CTRL,
1300                                         guest_val, host_val);
1301                         return;
1302                 }
1303                 break;
1304         }
1305
1306         for (i = 0; i < m->nr; ++i)
1307                 if (m->guest[i].index == msr)
1308                         break;
1309
1310         if (i == NR_AUTOLOAD_MSRS) {
1311                 printk_once(KERN_WARNING"Not enough mst switch entries. "
1312                                 "Can't add msr %x\n", msr);
1313                 return;
1314         } else if (i == m->nr) {
1315                 ++m->nr;
1316                 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1317                 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1318         }
1319
1320         m->guest[i].index = msr;
1321         m->guest[i].value = guest_val;
1322         m->host[i].index = msr;
1323         m->host[i].value = host_val;
1324 }
1325
1326 static void reload_tss(void)
1327 {
1328         /*
1329          * VT restores TR but not its size.  Useless.
1330          */
1331         struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1332         struct desc_struct *descs;
1333
1334         descs = (void *)gdt->address;
1335         descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
1336         load_TR_desc();
1337 }
1338
1339 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
1340 {
1341         u64 guest_efer;
1342         u64 ignore_bits;
1343
1344         guest_efer = vmx->vcpu.arch.efer;
1345
1346         /*
1347          * NX is emulated; LMA and LME handled by hardware; SCE meaningless
1348          * outside long mode
1349          */
1350         ignore_bits = EFER_NX | EFER_SCE;
1351 #ifdef CONFIG_X86_64
1352         ignore_bits |= EFER_LMA | EFER_LME;
1353         /* SCE is meaningful only in long mode on Intel */
1354         if (guest_efer & EFER_LMA)
1355                 ignore_bits &= ~(u64)EFER_SCE;
1356 #endif
1357         guest_efer &= ~ignore_bits;
1358         guest_efer |= host_efer & ignore_bits;
1359         vmx->guest_msrs[efer_offset].data = guest_efer;
1360         vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1361
1362         clear_atomic_switch_msr(vmx, MSR_EFER);
1363         /* On ept, can't emulate nx, and must switch nx atomically */
1364         if (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX)) {
1365                 guest_efer = vmx->vcpu.arch.efer;
1366                 if (!(guest_efer & EFER_LMA))
1367                         guest_efer &= ~EFER_LME;
1368                 add_atomic_switch_msr(vmx, MSR_EFER, guest_efer, host_efer);
1369                 return false;
1370         }
1371
1372         return true;
1373 }
1374
1375 static unsigned long segment_base(u16 selector)
1376 {
1377         struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1378         struct desc_struct *d;
1379         unsigned long table_base;
1380         unsigned long v;
1381
1382         if (!(selector & ~3))
1383                 return 0;
1384
1385         table_base = gdt->address;
1386
1387         if (selector & 4) {           /* from ldt */
1388                 u16 ldt_selector = kvm_read_ldt();
1389
1390                 if (!(ldt_selector & ~3))
1391                         return 0;
1392
1393                 table_base = segment_base(ldt_selector);
1394         }
1395         d = (struct desc_struct *)(table_base + (selector & ~7));
1396         v = get_desc_base(d);
1397 #ifdef CONFIG_X86_64
1398        if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
1399                v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
1400 #endif
1401         return v;
1402 }
1403
1404 static inline unsigned long kvm_read_tr_base(void)
1405 {
1406         u16 tr;
1407         asm("str %0" : "=g"(tr));
1408         return segment_base(tr);
1409 }
1410
1411 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
1412 {
1413         struct vcpu_vmx *vmx = to_vmx(vcpu);
1414         int i;
1415
1416         if (vmx->host_state.loaded)
1417                 return;
1418
1419         vmx->host_state.loaded = 1;
1420         /*
1421          * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
1422          * allow segment selectors with cpl > 0 or ti == 1.
1423          */
1424         vmx->host_state.ldt_sel = kvm_read_ldt();
1425         vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
1426         savesegment(fs, vmx->host_state.fs_sel);
1427         if (!(vmx->host_state.fs_sel & 7)) {
1428                 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
1429                 vmx->host_state.fs_reload_needed = 0;
1430         } else {
1431                 vmcs_write16(HOST_FS_SELECTOR, 0);
1432                 vmx->host_state.fs_reload_needed = 1;
1433         }
1434         savesegment(gs, vmx->host_state.gs_sel);
1435         if (!(vmx->host_state.gs_sel & 7))
1436                 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
1437         else {
1438                 vmcs_write16(HOST_GS_SELECTOR, 0);
1439                 vmx->host_state.gs_ldt_reload_needed = 1;
1440         }
1441
1442 #ifdef CONFIG_X86_64
1443         savesegment(ds, vmx->host_state.ds_sel);
1444         savesegment(es, vmx->host_state.es_sel);
1445 #endif
1446
1447 #ifdef CONFIG_X86_64
1448         vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
1449         vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
1450 #else
1451         vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
1452         vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
1453 #endif
1454
1455 #ifdef CONFIG_X86_64
1456         rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1457         if (is_long_mode(&vmx->vcpu))
1458                 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1459 #endif
1460         for (i = 0; i < vmx->save_nmsrs; ++i)
1461                 kvm_set_shared_msr(vmx->guest_msrs[i].index,
1462                                    vmx->guest_msrs[i].data,
1463                                    vmx->guest_msrs[i].mask);
1464 }
1465
1466 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
1467 {
1468         if (!vmx->host_state.loaded)
1469                 return;
1470
1471         ++vmx->vcpu.stat.host_state_reload;
1472         vmx->host_state.loaded = 0;
1473 #ifdef CONFIG_X86_64
1474         if (is_long_mode(&vmx->vcpu))
1475                 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1476 #endif
1477         if (vmx->host_state.gs_ldt_reload_needed) {
1478                 kvm_load_ldt(vmx->host_state.ldt_sel);
1479 #ifdef CONFIG_X86_64
1480                 load_gs_index(vmx->host_state.gs_sel);
1481 #else
1482                 loadsegment(gs, vmx->host_state.gs_sel);
1483 #endif
1484         }
1485         if (vmx->host_state.fs_reload_needed)
1486                 loadsegment(fs, vmx->host_state.fs_sel);
1487 #ifdef CONFIG_X86_64
1488         if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
1489                 loadsegment(ds, vmx->host_state.ds_sel);
1490                 loadsegment(es, vmx->host_state.es_sel);
1491         }
1492 #endif
1493         reload_tss();
1494 #ifdef CONFIG_X86_64
1495         wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1496 #endif
1497         /*
1498          * If the FPU is not active (through the host task or
1499          * the guest vcpu), then restore the cr0.TS bit.
1500          */
1501         if (!user_has_fpu() && !vmx->vcpu.guest_fpu_loaded)
1502                 stts();
1503         load_gdt(&__get_cpu_var(host_gdt));
1504 }
1505
1506 static void vmx_load_host_state(struct vcpu_vmx *vmx)
1507 {
1508         preempt_disable();
1509         __vmx_load_host_state(vmx);
1510         preempt_enable();
1511 }
1512
1513 /*
1514  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1515  * vcpu mutex is already taken.
1516  */
1517 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1518 {
1519         struct vcpu_vmx *vmx = to_vmx(vcpu);
1520         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1521
1522         if (!vmm_exclusive)
1523                 kvm_cpu_vmxon(phys_addr);
1524         else if (vmx->loaded_vmcs->cpu != cpu)
1525                 loaded_vmcs_clear(vmx->loaded_vmcs);
1526
1527         if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1528                 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1529                 vmcs_load(vmx->loaded_vmcs->vmcs);
1530         }
1531
1532         if (vmx->loaded_vmcs->cpu != cpu) {
1533                 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1534                 unsigned long sysenter_esp;
1535
1536                 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1537                 local_irq_disable();
1538                 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1539                          &per_cpu(loaded_vmcss_on_cpu, cpu));
1540                 local_irq_enable();
1541
1542                 /*
1543                  * Linux uses per-cpu TSS and GDT, so set these when switching
1544                  * processors.
1545                  */
1546                 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
1547                 vmcs_writel(HOST_GDTR_BASE, gdt->address);   /* 22.2.4 */
1548
1549                 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1550                 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1551                 vmx->loaded_vmcs->cpu = cpu;
1552         }
1553 }
1554
1555 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1556 {
1557         __vmx_load_host_state(to_vmx(vcpu));
1558         if (!vmm_exclusive) {
1559                 __loaded_vmcs_clear(to_vmx(vcpu)->loaded_vmcs);
1560                 vcpu->cpu = -1;
1561                 kvm_cpu_vmxoff();
1562         }
1563 }
1564
1565 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
1566 {
1567         ulong cr0;
1568
1569         if (vcpu->fpu_active)
1570                 return;
1571         vcpu->fpu_active = 1;
1572         cr0 = vmcs_readl(GUEST_CR0);
1573         cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
1574         cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
1575         vmcs_writel(GUEST_CR0, cr0);
1576         update_exception_bitmap(vcpu);
1577         vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
1578         if (is_guest_mode(vcpu))
1579                 vcpu->arch.cr0_guest_owned_bits &=
1580                         ~get_vmcs12(vcpu)->cr0_guest_host_mask;
1581         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1582 }
1583
1584 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
1585
1586 /*
1587  * Return the cr0 value that a nested guest would read. This is a combination
1588  * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
1589  * its hypervisor (cr0_read_shadow).
1590  */
1591 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
1592 {
1593         return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
1594                 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
1595 }
1596 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
1597 {
1598         return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
1599                 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
1600 }
1601
1602 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
1603 {
1604         /* Note that there is no vcpu->fpu_active = 0 here. The caller must
1605          * set this *before* calling this function.
1606          */
1607         vmx_decache_cr0_guest_bits(vcpu);
1608         vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
1609         update_exception_bitmap(vcpu);
1610         vcpu->arch.cr0_guest_owned_bits = 0;
1611         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1612         if (is_guest_mode(vcpu)) {
1613                 /*
1614                  * L1's specified read shadow might not contain the TS bit,
1615                  * so now that we turned on shadowing of this bit, we need to
1616                  * set this bit of the shadow. Like in nested_vmx_run we need
1617                  * nested_read_cr0(vmcs12), but vmcs12->guest_cr0 is not yet
1618                  * up-to-date here because we just decached cr0.TS (and we'll
1619                  * only update vmcs12->guest_cr0 on nested exit).
1620                  */
1621                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1622                 vmcs12->guest_cr0 = (vmcs12->guest_cr0 & ~X86_CR0_TS) |
1623                         (vcpu->arch.cr0 & X86_CR0_TS);
1624                 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
1625         } else
1626                 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
1627 }
1628
1629 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1630 {
1631         unsigned long rflags, save_rflags;
1632
1633         if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
1634                 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1635                 rflags = vmcs_readl(GUEST_RFLAGS);
1636                 if (to_vmx(vcpu)->rmode.vm86_active) {
1637                         rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1638                         save_rflags = to_vmx(vcpu)->rmode.save_rflags;
1639                         rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1640                 }
1641                 to_vmx(vcpu)->rflags = rflags;
1642         }
1643         return to_vmx(vcpu)->rflags;
1644 }
1645
1646 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1647 {
1648         __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1649         __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
1650         to_vmx(vcpu)->rflags = rflags;
1651         if (to_vmx(vcpu)->rmode.vm86_active) {
1652                 to_vmx(vcpu)->rmode.save_rflags = rflags;
1653                 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1654         }
1655         vmcs_writel(GUEST_RFLAGS, rflags);
1656 }
1657
1658 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1659 {
1660         u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1661         int ret = 0;
1662
1663         if (interruptibility & GUEST_INTR_STATE_STI)
1664                 ret |= KVM_X86_SHADOW_INT_STI;
1665         if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1666                 ret |= KVM_X86_SHADOW_INT_MOV_SS;
1667
1668         return ret & mask;
1669 }
1670
1671 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1672 {
1673         u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1674         u32 interruptibility = interruptibility_old;
1675
1676         interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1677
1678         if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1679                 interruptibility |= GUEST_INTR_STATE_MOV_SS;
1680         else if (mask & KVM_X86_SHADOW_INT_STI)
1681                 interruptibility |= GUEST_INTR_STATE_STI;
1682
1683         if ((interruptibility != interruptibility_old))
1684                 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1685 }
1686
1687 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
1688 {
1689         unsigned long rip;
1690
1691         rip = kvm_rip_read(vcpu);
1692         rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1693         kvm_rip_write(vcpu, rip);
1694
1695         /* skipping an emulated instruction also counts */
1696         vmx_set_interrupt_shadow(vcpu, 0);
1697 }
1698
1699 /*
1700  * KVM wants to inject page-faults which it got to the guest. This function
1701  * checks whether in a nested guest, we need to inject them to L1 or L2.
1702  * This function assumes it is called with the exit reason in vmcs02 being
1703  * a #PF exception (this is the only case in which KVM injects a #PF when L2
1704  * is running).
1705  */
1706 static int nested_pf_handled(struct kvm_vcpu *vcpu)
1707 {
1708         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1709
1710         /* TODO: also check PFEC_MATCH/MASK, not just EB.PF. */
1711         if (!(vmcs12->exception_bitmap & (1u << PF_VECTOR)))
1712                 return 0;
1713
1714         nested_vmx_vmexit(vcpu);
1715         return 1;
1716 }
1717
1718 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
1719                                 bool has_error_code, u32 error_code,
1720                                 bool reinject)
1721 {
1722         struct vcpu_vmx *vmx = to_vmx(vcpu);
1723         u32 intr_info = nr | INTR_INFO_VALID_MASK;
1724
1725         if (nr == PF_VECTOR && is_guest_mode(vcpu) &&
1726                 nested_pf_handled(vcpu))
1727                 return;
1728
1729         if (has_error_code) {
1730                 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1731                 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1732         }
1733
1734         if (vmx->rmode.vm86_active) {
1735                 int inc_eip = 0;
1736                 if (kvm_exception_is_soft(nr))
1737                         inc_eip = vcpu->arch.event_exit_inst_len;
1738                 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
1739                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
1740                 return;
1741         }
1742
1743         if (kvm_exception_is_soft(nr)) {
1744                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1745                              vmx->vcpu.arch.event_exit_inst_len);
1746                 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1747         } else
1748                 intr_info |= INTR_TYPE_HARD_EXCEPTION;
1749
1750         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1751 }
1752
1753 static bool vmx_rdtscp_supported(void)
1754 {
1755         return cpu_has_vmx_rdtscp();
1756 }
1757
1758 static bool vmx_invpcid_supported(void)
1759 {
1760         return cpu_has_vmx_invpcid() && enable_ept;
1761 }
1762
1763 /*
1764  * Swap MSR entry in host/guest MSR entry array.
1765  */
1766 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1767 {
1768         struct shared_msr_entry tmp;
1769
1770         tmp = vmx->guest_msrs[to];
1771         vmx->guest_msrs[to] = vmx->guest_msrs[from];
1772         vmx->guest_msrs[from] = tmp;
1773 }
1774
1775 /*
1776  * Set up the vmcs to automatically save and restore system
1777  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
1778  * mode, as fiddling with msrs is very expensive.
1779  */
1780 static void setup_msrs(struct vcpu_vmx *vmx)
1781 {
1782         int save_nmsrs, index;
1783         unsigned long *msr_bitmap;
1784
1785         save_nmsrs = 0;
1786 #ifdef CONFIG_X86_64
1787         if (is_long_mode(&vmx->vcpu)) {
1788                 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1789                 if (index >= 0)
1790                         move_msr_up(vmx, index, save_nmsrs++);
1791                 index = __find_msr_index(vmx, MSR_LSTAR);
1792                 if (index >= 0)
1793                         move_msr_up(vmx, index, save_nmsrs++);
1794                 index = __find_msr_index(vmx, MSR_CSTAR);
1795                 if (index >= 0)
1796                         move_msr_up(vmx, index, save_nmsrs++);
1797                 index = __find_msr_index(vmx, MSR_TSC_AUX);
1798                 if (index >= 0 && vmx->rdtscp_enabled)
1799                         move_msr_up(vmx, index, save_nmsrs++);
1800                 /*
1801                  * MSR_STAR is only needed on long mode guests, and only
1802                  * if efer.sce is enabled.
1803                  */
1804                 index = __find_msr_index(vmx, MSR_STAR);
1805                 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
1806                         move_msr_up(vmx, index, save_nmsrs++);
1807         }
1808 #endif
1809         index = __find_msr_index(vmx, MSR_EFER);
1810         if (index >= 0 && update_transition_efer(vmx, index))
1811                 move_msr_up(vmx, index, save_nmsrs++);
1812
1813         vmx->save_nmsrs = save_nmsrs;
1814
1815         if (cpu_has_vmx_msr_bitmap()) {
1816                 if (is_long_mode(&vmx->vcpu))
1817                         msr_bitmap = vmx_msr_bitmap_longmode;
1818                 else
1819                         msr_bitmap = vmx_msr_bitmap_legacy;
1820
1821                 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
1822         }
1823 }
1824
1825 /*
1826  * reads and returns guest's timestamp counter "register"
1827  * guest_tsc = host_tsc + tsc_offset    -- 21.3
1828  */
1829 static u64 guest_read_tsc(void)
1830 {
1831         u64 host_tsc, tsc_offset;
1832
1833         rdtscll(host_tsc);
1834         tsc_offset = vmcs_read64(TSC_OFFSET);
1835         return host_tsc + tsc_offset;
1836 }
1837
1838 /*
1839  * Like guest_read_tsc, but always returns L1's notion of the timestamp
1840  * counter, even if a nested guest (L2) is currently running.
1841  */
1842 u64 vmx_read_l1_tsc(struct kvm_vcpu *vcpu)
1843 {
1844         u64 host_tsc, tsc_offset;
1845
1846         rdtscll(host_tsc);
1847         tsc_offset = is_guest_mode(vcpu) ?
1848                 to_vmx(vcpu)->nested.vmcs01_tsc_offset :
1849                 vmcs_read64(TSC_OFFSET);
1850         return host_tsc + tsc_offset;
1851 }
1852
1853 /*
1854  * Engage any workarounds for mis-matched TSC rates.  Currently limited to
1855  * software catchup for faster rates on slower CPUs.
1856  */
1857 static void vmx_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1858 {
1859         if (!scale)
1860                 return;
1861
1862         if (user_tsc_khz > tsc_khz) {
1863                 vcpu->arch.tsc_catchup = 1;
1864                 vcpu->arch.tsc_always_catchup = 1;
1865         } else
1866                 WARN(1, "user requested TSC rate below hardware speed\n");
1867 }
1868
1869 /*
1870  * writes 'offset' into guest's timestamp counter offset register
1871  */
1872 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1873 {
1874         if (is_guest_mode(vcpu)) {
1875                 /*
1876                  * We're here if L1 chose not to trap WRMSR to TSC. According
1877                  * to the spec, this should set L1's TSC; The offset that L1
1878                  * set for L2 remains unchanged, and still needs to be added
1879                  * to the newly set TSC to get L2's TSC.
1880                  */
1881                 struct vmcs12 *vmcs12;
1882                 to_vmx(vcpu)->nested.vmcs01_tsc_offset = offset;
1883                 /* recalculate vmcs02.TSC_OFFSET: */
1884                 vmcs12 = get_vmcs12(vcpu);
1885                 vmcs_write64(TSC_OFFSET, offset +
1886                         (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
1887                          vmcs12->tsc_offset : 0));
1888         } else {
1889                 vmcs_write64(TSC_OFFSET, offset);
1890         }
1891 }
1892
1893 static void vmx_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment, bool host)
1894 {
1895         u64 offset = vmcs_read64(TSC_OFFSET);
1896         vmcs_write64(TSC_OFFSET, offset + adjustment);
1897         if (is_guest_mode(vcpu)) {
1898                 /* Even when running L2, the adjustment needs to apply to L1 */
1899                 to_vmx(vcpu)->nested.vmcs01_tsc_offset += adjustment;
1900         }
1901 }
1902
1903 static u64 vmx_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1904 {
1905         return target_tsc - native_read_tsc();
1906 }
1907
1908 static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
1909 {
1910         struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
1911         return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
1912 }
1913
1914 /*
1915  * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1916  * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1917  * all guests if the "nested" module option is off, and can also be disabled
1918  * for a single guest by disabling its VMX cpuid bit.
1919  */
1920 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1921 {
1922         return nested && guest_cpuid_has_vmx(vcpu);
1923 }
1924
1925 /*
1926  * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
1927  * returned for the various VMX controls MSRs when nested VMX is enabled.
1928  * The same values should also be used to verify that vmcs12 control fields are
1929  * valid during nested entry from L1 to L2.
1930  * Each of these control msrs has a low and high 32-bit half: A low bit is on
1931  * if the corresponding bit in the (32-bit) control field *must* be on, and a
1932  * bit in the high half is on if the corresponding bit in the control field
1933  * may be on. See also vmx_control_verify().
1934  * TODO: allow these variables to be modified (downgraded) by module options
1935  * or other means.
1936  */
1937 static u32 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high;
1938 static u32 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high;
1939 static u32 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high;
1940 static u32 nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high;
1941 static u32 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high;
1942 static __init void nested_vmx_setup_ctls_msrs(void)
1943 {
1944         /*
1945          * Note that as a general rule, the high half of the MSRs (bits in
1946          * the control fields which may be 1) should be initialized by the
1947          * intersection of the underlying hardware's MSR (i.e., features which
1948          * can be supported) and the list of features we want to expose -
1949          * because they are known to be properly supported in our code.
1950          * Also, usually, the low half of the MSRs (bits which must be 1) can
1951          * be set to 0, meaning that L1 may turn off any of these bits. The
1952          * reason is that if one of these bits is necessary, it will appear
1953          * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
1954          * fields of vmcs01 and vmcs02, will turn these bits off - and
1955          * nested_vmx_exit_handled() will not pass related exits to L1.
1956          * These rules have exceptions below.
1957          */
1958
1959         /* pin-based controls */
1960         /*
1961          * According to the Intel spec, if bit 55 of VMX_BASIC is off (as it is
1962          * in our case), bits 1, 2 and 4 (i.e., 0x16) must be 1 in this MSR.
1963          */
1964         nested_vmx_pinbased_ctls_low = 0x16 ;
1965         nested_vmx_pinbased_ctls_high = 0x16 |
1966                 PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING |
1967                 PIN_BASED_VIRTUAL_NMIS;
1968
1969         /* exit controls */
1970         nested_vmx_exit_ctls_low = 0;
1971         /* Note that guest use of VM_EXIT_ACK_INTR_ON_EXIT is not supported. */
1972 #ifdef CONFIG_X86_64
1973         nested_vmx_exit_ctls_high = VM_EXIT_HOST_ADDR_SPACE_SIZE;
1974 #else
1975         nested_vmx_exit_ctls_high = 0;
1976 #endif
1977
1978         /* entry controls */
1979         rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
1980                 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high);
1981         nested_vmx_entry_ctls_low = 0;
1982         nested_vmx_entry_ctls_high &=
1983                 VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_IA32E_MODE;
1984
1985         /* cpu-based controls */
1986         rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
1987                 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high);
1988         nested_vmx_procbased_ctls_low = 0;
1989         nested_vmx_procbased_ctls_high &=
1990                 CPU_BASED_VIRTUAL_INTR_PENDING | CPU_BASED_USE_TSC_OFFSETING |
1991                 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
1992                 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
1993                 CPU_BASED_CR3_STORE_EXITING |
1994 #ifdef CONFIG_X86_64
1995                 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
1996 #endif
1997                 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
1998                 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_EXITING |
1999                 CPU_BASED_RDPMC_EXITING | CPU_BASED_RDTSC_EXITING |
2000                 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2001         /*
2002          * We can allow some features even when not supported by the
2003          * hardware. For example, L1 can specify an MSR bitmap - and we
2004          * can use it to avoid exits to L1 - even when L0 runs L2
2005          * without MSR bitmaps.
2006          */
2007         nested_vmx_procbased_ctls_high |= CPU_BASED_USE_MSR_BITMAPS;
2008
2009         /* secondary cpu-based controls */
2010         rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2011                 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high);
2012         nested_vmx_secondary_ctls_low = 0;
2013         nested_vmx_secondary_ctls_high &=
2014                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
2015 }
2016
2017 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2018 {
2019         /*
2020          * Bits 0 in high must be 0, and bits 1 in low must be 1.
2021          */
2022         return ((control & high) | low) == control;
2023 }
2024
2025 static inline u64 vmx_control_msr(u32 low, u32 high)
2026 {
2027         return low | ((u64)high << 32);
2028 }
2029
2030 /*
2031  * If we allow our guest to use VMX instructions (i.e., nested VMX), we should
2032  * also let it use VMX-specific MSRs.
2033  * vmx_get_vmx_msr() and vmx_set_vmx_msr() return 1 when we handled a
2034  * VMX-specific MSR, or 0 when we haven't (and the caller should handle it
2035  * like all other MSRs).
2036  */
2037 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2038 {
2039         if (!nested_vmx_allowed(vcpu) && msr_index >= MSR_IA32_VMX_BASIC &&
2040                      msr_index <= MSR_IA32_VMX_TRUE_ENTRY_CTLS) {
2041                 /*
2042                  * According to the spec, processors which do not support VMX
2043                  * should throw a #GP(0) when VMX capability MSRs are read.
2044                  */
2045                 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
2046                 return 1;
2047         }
2048
2049         switch (msr_index) {
2050         case MSR_IA32_FEATURE_CONTROL:
2051                 *pdata = 0;
2052                 break;
2053         case MSR_IA32_VMX_BASIC:
2054                 /*
2055                  * This MSR reports some information about VMX support. We
2056                  * should return information about the VMX we emulate for the
2057                  * guest, and the VMCS structure we give it - not about the
2058                  * VMX support of the underlying hardware.
2059                  */
2060                 *pdata = VMCS12_REVISION |
2061                            ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2062                            (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2063                 break;
2064         case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2065         case MSR_IA32_VMX_PINBASED_CTLS:
2066                 *pdata = vmx_control_msr(nested_vmx_pinbased_ctls_low,
2067                                         nested_vmx_pinbased_ctls_high);
2068                 break;
2069         case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2070         case MSR_IA32_VMX_PROCBASED_CTLS:
2071                 *pdata = vmx_control_msr(nested_vmx_procbased_ctls_low,
2072                                         nested_vmx_procbased_ctls_high);
2073                 break;
2074         case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2075         case MSR_IA32_VMX_EXIT_CTLS:
2076                 *pdata = vmx_control_msr(nested_vmx_exit_ctls_low,
2077                                         nested_vmx_exit_ctls_high);
2078                 break;
2079         case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2080         case MSR_IA32_VMX_ENTRY_CTLS:
2081                 *pdata = vmx_control_msr(nested_vmx_entry_ctls_low,
2082                                         nested_vmx_entry_ctls_high);
2083                 break;
2084         case MSR_IA32_VMX_MISC:
2085                 *pdata = 0;
2086                 break;
2087         /*
2088          * These MSRs specify bits which the guest must keep fixed (on or off)
2089          * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2090          * We picked the standard core2 setting.
2091          */
2092 #define VMXON_CR0_ALWAYSON      (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2093 #define VMXON_CR4_ALWAYSON      X86_CR4_VMXE
2094         case MSR_IA32_VMX_CR0_FIXED0:
2095                 *pdata = VMXON_CR0_ALWAYSON;
2096                 break;
2097         case MSR_IA32_VMX_CR0_FIXED1:
2098                 *pdata = -1ULL;
2099                 break;
2100         case MSR_IA32_VMX_CR4_FIXED0:
2101                 *pdata = VMXON_CR4_ALWAYSON;
2102                 break;
2103         case MSR_IA32_VMX_CR4_FIXED1:
2104                 *pdata = -1ULL;
2105                 break;
2106         case MSR_IA32_VMX_VMCS_ENUM:
2107                 *pdata = 0x1f;
2108                 break;
2109         case MSR_IA32_VMX_PROCBASED_CTLS2:
2110                 *pdata = vmx_control_msr(nested_vmx_secondary_ctls_low,
2111                                         nested_vmx_secondary_ctls_high);
2112                 break;
2113         case MSR_IA32_VMX_EPT_VPID_CAP:
2114                 /* Currently, no nested ept or nested vpid */
2115                 *pdata = 0;
2116                 break;
2117         default:
2118                 return 0;
2119         }
2120
2121         return 1;
2122 }
2123
2124 static int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
2125 {
2126         if (!nested_vmx_allowed(vcpu))
2127                 return 0;
2128
2129         if (msr_index == MSR_IA32_FEATURE_CONTROL)
2130                 /* TODO: the right thing. */
2131                 return 1;
2132         /*
2133          * No need to treat VMX capability MSRs specially: If we don't handle
2134          * them, handle_wrmsr will #GP(0), which is correct (they are readonly)
2135          */
2136         return 0;
2137 }
2138
2139 /*
2140  * Reads an msr value (of 'msr_index') into 'pdata'.
2141  * Returns 0 on success, non-0 otherwise.
2142  * Assumes vcpu_load() was already called.
2143  */
2144 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2145 {
2146         u64 data;
2147         struct shared_msr_entry *msr;
2148
2149         if (!pdata) {
2150                 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
2151                 return -EINVAL;
2152         }
2153
2154         switch (msr_index) {
2155 #ifdef CONFIG_X86_64
2156         case MSR_FS_BASE:
2157                 data = vmcs_readl(GUEST_FS_BASE);
2158                 break;
2159         case MSR_GS_BASE:
2160                 data = vmcs_readl(GUEST_GS_BASE);
2161                 break;
2162         case MSR_KERNEL_GS_BASE:
2163                 vmx_load_host_state(to_vmx(vcpu));
2164                 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
2165                 break;
2166 #endif
2167         case MSR_EFER:
2168                 return kvm_get_msr_common(vcpu, msr_index, pdata);
2169         case MSR_IA32_TSC:
2170                 data = guest_read_tsc();
2171                 break;
2172         case MSR_IA32_SYSENTER_CS:
2173                 data = vmcs_read32(GUEST_SYSENTER_CS);
2174                 break;
2175         case MSR_IA32_SYSENTER_EIP:
2176                 data = vmcs_readl(GUEST_SYSENTER_EIP);
2177                 break;
2178         case MSR_IA32_SYSENTER_ESP:
2179                 data = vmcs_readl(GUEST_SYSENTER_ESP);
2180                 break;
2181         case MSR_TSC_AUX:
2182                 if (!to_vmx(vcpu)->rdtscp_enabled)
2183                         return 1;
2184                 /* Otherwise falls through */
2185         default:
2186                 if (vmx_get_vmx_msr(vcpu, msr_index, pdata))
2187                         return 0;
2188                 msr = find_msr_entry(to_vmx(vcpu), msr_index);
2189                 if (msr) {
2190                         data = msr->data;
2191                         break;
2192                 }
2193                 return kvm_get_msr_common(vcpu, msr_index, pdata);
2194         }
2195
2196         *pdata = data;
2197         return 0;
2198 }
2199
2200 /*
2201  * Writes msr value into into the appropriate "register".
2202  * Returns 0 on success, non-0 otherwise.
2203  * Assumes vcpu_load() was already called.
2204  */
2205 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
2206 {
2207         struct vcpu_vmx *vmx = to_vmx(vcpu);
2208         struct shared_msr_entry *msr;
2209         int ret = 0;
2210
2211         switch (msr_index) {
2212         case MSR_EFER:
2213                 ret = kvm_set_msr_common(vcpu, msr_index, data);
2214                 break;
2215 #ifdef CONFIG_X86_64
2216         case MSR_FS_BASE:
2217                 vmx_segment_cache_clear(vmx);
2218                 vmcs_writel(GUEST_FS_BASE, data);
2219                 break;
2220         case MSR_GS_BASE:
2221                 vmx_segment_cache_clear(vmx);
2222                 vmcs_writel(GUEST_GS_BASE, data);
2223                 break;
2224         case MSR_KERNEL_GS_BASE:
2225                 vmx_load_host_state(vmx);
2226                 vmx->msr_guest_kernel_gs_base = data;
2227                 break;
2228 #endif
2229         case MSR_IA32_SYSENTER_CS:
2230                 vmcs_write32(GUEST_SYSENTER_CS, data);
2231                 break;
2232         case MSR_IA32_SYSENTER_EIP:
2233                 vmcs_writel(GUEST_SYSENTER_EIP, data);
2234                 break;
2235         case MSR_IA32_SYSENTER_ESP:
2236                 vmcs_writel(GUEST_SYSENTER_ESP, data);
2237                 break;
2238         case MSR_IA32_TSC:
2239                 kvm_write_tsc(vcpu, data);
2240                 break;
2241         case MSR_IA32_CR_PAT:
2242                 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2243                         vmcs_write64(GUEST_IA32_PAT, data);
2244                         vcpu->arch.pat = data;
2245                         break;
2246                 }
2247                 ret = kvm_set_msr_common(vcpu, msr_index, data);
2248                 break;
2249         case MSR_TSC_AUX:
2250                 if (!vmx->rdtscp_enabled)
2251                         return 1;
2252                 /* Check reserved bit, higher 32 bits should be zero */
2253                 if ((data >> 32) != 0)
2254                         return 1;
2255                 /* Otherwise falls through */
2256         default:
2257                 if (vmx_set_vmx_msr(vcpu, msr_index, data))
2258                         break;
2259                 msr = find_msr_entry(vmx, msr_index);
2260                 if (msr) {
2261                         msr->data = data;
2262                         if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
2263                                 preempt_disable();
2264                                 kvm_set_shared_msr(msr->index, msr->data,
2265                                                    msr->mask);
2266                                 preempt_enable();
2267                         }
2268                         break;
2269                 }
2270                 ret = kvm_set_msr_common(vcpu, msr_index, data);
2271         }
2272
2273         return ret;
2274 }
2275
2276 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2277 {
2278         __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
2279         switch (reg) {
2280         case VCPU_REGS_RSP:
2281                 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2282                 break;
2283         case VCPU_REGS_RIP:
2284                 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2285                 break;
2286         case VCPU_EXREG_PDPTR:
2287                 if (enable_ept)
2288                         ept_save_pdptrs(vcpu);
2289                 break;
2290         default:
2291                 break;
2292         }
2293 }
2294
2295 static __init int cpu_has_kvm_support(void)
2296 {
2297         return cpu_has_vmx();
2298 }
2299
2300 static __init int vmx_disabled_by_bios(void)
2301 {
2302         u64 msr;
2303
2304         rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
2305         if (msr & FEATURE_CONTROL_LOCKED) {
2306                 /* launched w/ TXT and VMX disabled */
2307                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2308                         && tboot_enabled())
2309                         return 1;
2310                 /* launched w/o TXT and VMX only enabled w/ TXT */
2311                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2312                         && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2313                         && !tboot_enabled()) {
2314                         printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
2315                                 "activate TXT before enabling KVM\n");
2316                         return 1;
2317                 }
2318                 /* launched w/o TXT and VMX disabled */
2319                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2320                         && !tboot_enabled())
2321                         return 1;
2322         }
2323
2324         return 0;
2325 }
2326
2327 static void kvm_cpu_vmxon(u64 addr)
2328 {
2329         asm volatile (ASM_VMX_VMXON_RAX
2330                         : : "a"(&addr), "m"(addr)
2331                         : "memory", "cc");
2332 }
2333
2334 static int hardware_enable(void *garbage)
2335 {
2336         int cpu = raw_smp_processor_id();
2337         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2338         u64 old, test_bits;
2339
2340         if (read_cr4() & X86_CR4_VMXE)
2341                 return -EBUSY;
2342
2343         INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2344         rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
2345
2346         test_bits = FEATURE_CONTROL_LOCKED;
2347         test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
2348         if (tboot_enabled())
2349                 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
2350
2351         if ((old & test_bits) != test_bits) {
2352                 /* enable and lock */
2353                 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
2354         }
2355         write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
2356
2357         if (vmm_exclusive) {
2358                 kvm_cpu_vmxon(phys_addr);
2359                 ept_sync_global();
2360         }
2361
2362         store_gdt(&__get_cpu_var(host_gdt));
2363
2364         return 0;
2365 }
2366
2367 static void vmclear_local_loaded_vmcss(void)
2368 {
2369         int cpu = raw_smp_processor_id();
2370         struct loaded_vmcs *v, *n;
2371
2372         list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2373                                  loaded_vmcss_on_cpu_link)
2374                 __loaded_vmcs_clear(v);
2375 }
2376
2377
2378 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2379  * tricks.
2380  */
2381 static void kvm_cpu_vmxoff(void)
2382 {
2383         asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
2384 }
2385
2386 static void hardware_disable(void *garbage)
2387 {
2388         if (vmm_exclusive) {
2389                 vmclear_local_loaded_vmcss();
2390                 kvm_cpu_vmxoff();
2391         }
2392         write_cr4(read_cr4() & ~X86_CR4_VMXE);
2393 }
2394
2395 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2396                                       u32 msr, u32 *result)
2397 {
2398         u32 vmx_msr_low, vmx_msr_high;
2399         u32 ctl = ctl_min | ctl_opt;
2400
2401         rdmsr(msr, vmx_msr_low, vmx_msr_high);
2402
2403         ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2404         ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
2405
2406         /* Ensure minimum (required) set of control bits are supported. */
2407         if (ctl_min & ~ctl)
2408                 return -EIO;
2409
2410         *result = ctl;
2411         return 0;
2412 }
2413
2414 static __init bool allow_1_setting(u32 msr, u32 ctl)
2415 {
2416         u32 vmx_msr_low, vmx_msr_high;
2417
2418         rdmsr(msr, vmx_msr_low, vmx_msr_high);
2419         return vmx_msr_high & ctl;
2420 }
2421
2422 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
2423 {
2424         u32 vmx_msr_low, vmx_msr_high;
2425         u32 min, opt, min2, opt2;
2426         u32 _pin_based_exec_control = 0;
2427         u32 _cpu_based_exec_control = 0;
2428         u32 _cpu_based_2nd_exec_control = 0;
2429         u32 _vmexit_control = 0;
2430         u32 _vmentry_control = 0;
2431
2432         min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2433         opt = PIN_BASED_VIRTUAL_NMIS;
2434         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2435                                 &_pin_based_exec_control) < 0)
2436                 return -EIO;
2437
2438         min = CPU_BASED_HLT_EXITING |
2439 #ifdef CONFIG_X86_64
2440               CPU_BASED_CR8_LOAD_EXITING |
2441               CPU_BASED_CR8_STORE_EXITING |
2442 #endif
2443               CPU_BASED_CR3_LOAD_EXITING |
2444               CPU_BASED_CR3_STORE_EXITING |
2445               CPU_BASED_USE_IO_BITMAPS |
2446               CPU_BASED_MOV_DR_EXITING |
2447               CPU_BASED_USE_TSC_OFFSETING |
2448               CPU_BASED_MWAIT_EXITING |
2449               CPU_BASED_MONITOR_EXITING |
2450               CPU_BASED_INVLPG_EXITING |
2451               CPU_BASED_RDPMC_EXITING;
2452
2453         opt = CPU_BASED_TPR_SHADOW |
2454               CPU_BASED_USE_MSR_BITMAPS |
2455               CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2456         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2457                                 &_cpu_based_exec_control) < 0)
2458                 return -EIO;
2459 #ifdef CONFIG_X86_64
2460         if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2461                 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2462                                            ~CPU_BASED_CR8_STORE_EXITING;
2463 #endif
2464         if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2465                 min2 = 0;
2466                 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2467                         SECONDARY_EXEC_WBINVD_EXITING |
2468                         SECONDARY_EXEC_ENABLE_VPID |
2469                         SECONDARY_EXEC_ENABLE_EPT |
2470                         SECONDARY_EXEC_UNRESTRICTED_GUEST |
2471                         SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2472                         SECONDARY_EXEC_RDTSCP |
2473                         SECONDARY_EXEC_ENABLE_INVPCID;
2474                 if (adjust_vmx_controls(min2, opt2,
2475                                         MSR_IA32_VMX_PROCBASED_CTLS2,
2476                                         &_cpu_based_2nd_exec_control) < 0)
2477                         return -EIO;
2478         }
2479 #ifndef CONFIG_X86_64
2480         if (!(_cpu_based_2nd_exec_control &
2481                                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2482                 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2483 #endif
2484         if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2485                 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2486                    enabled */
2487                 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2488                                              CPU_BASED_CR3_STORE_EXITING |
2489                                              CPU_BASED_INVLPG_EXITING);
2490                 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
2491                       vmx_capability.ept, vmx_capability.vpid);
2492         }
2493
2494         min = 0;
2495 #ifdef CONFIG_X86_64
2496         min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2497 #endif
2498         opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
2499         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2500                                 &_vmexit_control) < 0)
2501                 return -EIO;
2502
2503         min = 0;
2504         opt = VM_ENTRY_LOAD_IA32_PAT;
2505         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2506                                 &_vmentry_control) < 0)
2507                 return -EIO;
2508
2509         rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2510
2511         /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2512         if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2513                 return -EIO;
2514
2515 #ifdef CONFIG_X86_64
2516         /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2517         if (vmx_msr_high & (1u<<16))
2518                 return -EIO;
2519 #endif
2520
2521         /* Require Write-Back (WB) memory type for VMCS accesses. */
2522         if (((vmx_msr_high >> 18) & 15) != 6)
2523                 return -EIO;
2524
2525         vmcs_conf->size = vmx_msr_high & 0x1fff;
2526         vmcs_conf->order = get_order(vmcs_config.size);
2527         vmcs_conf->revision_id = vmx_msr_low;
2528
2529         vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2530         vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2531         vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2532         vmcs_conf->vmexit_ctrl         = _vmexit_control;
2533         vmcs_conf->vmentry_ctrl        = _vmentry_control;
2534
2535         cpu_has_load_ia32_efer =
2536                 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2537                                 VM_ENTRY_LOAD_IA32_EFER)
2538                 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2539                                    VM_EXIT_LOAD_IA32_EFER);
2540
2541         cpu_has_load_perf_global_ctrl =
2542                 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2543                                 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
2544                 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2545                                    VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
2546
2547         /*
2548          * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
2549          * but due to arrata below it can't be used. Workaround is to use
2550          * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2551          *
2552          * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
2553          *
2554          * AAK155             (model 26)
2555          * AAP115             (model 30)
2556          * AAT100             (model 37)
2557          * BC86,AAY89,BD102   (model 44)
2558          * BA97               (model 46)
2559          *
2560          */
2561         if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
2562                 switch (boot_cpu_data.x86_model) {
2563                 case 26:
2564                 case 30:
2565                 case 37:
2566                 case 44:
2567                 case 46:
2568                         cpu_has_load_perf_global_ctrl = false;
2569                         printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2570                                         "does not work properly. Using workaround\n");
2571                         break;
2572                 default:
2573                         break;
2574                 }
2575         }
2576
2577         return 0;
2578 }
2579
2580 static struct vmcs *alloc_vmcs_cpu(int cpu)
2581 {
2582         int node = cpu_to_node(cpu);
2583         struct page *pages;
2584         struct vmcs *vmcs;
2585
2586         pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
2587         if (!pages)
2588                 return NULL;
2589         vmcs = page_address(pages);
2590         memset(vmcs, 0, vmcs_config.size);
2591         vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
2592         return vmcs;
2593 }
2594
2595 static struct vmcs *alloc_vmcs(void)
2596 {
2597         return alloc_vmcs_cpu(raw_smp_processor_id());
2598 }
2599
2600 static void free_vmcs(struct vmcs *vmcs)
2601 {
2602         free_pages((unsigned long)vmcs, vmcs_config.order);
2603 }
2604
2605 /*
2606  * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2607  */
2608 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2609 {
2610         if (!loaded_vmcs->vmcs)
2611                 return;
2612         loaded_vmcs_clear(loaded_vmcs);
2613         free_vmcs(loaded_vmcs->vmcs);
2614         loaded_vmcs->vmcs = NULL;
2615 }
2616
2617 static void free_kvm_area(void)
2618 {
2619         int cpu;
2620
2621         for_each_possible_cpu(cpu) {
2622                 free_vmcs(per_cpu(vmxarea, cpu));
2623                 per_cpu(vmxarea, cpu) = NULL;
2624         }
2625 }
2626
2627 static __init int alloc_kvm_area(void)
2628 {
2629         int cpu;
2630
2631         for_each_possible_cpu(cpu) {
2632                 struct vmcs *vmcs;
2633
2634                 vmcs = alloc_vmcs_cpu(cpu);
2635                 if (!vmcs) {
2636                         free_kvm_area();
2637                         return -ENOMEM;
2638                 }
2639
2640                 per_cpu(vmxarea, cpu) = vmcs;
2641         }
2642         return 0;
2643 }
2644
2645 static __init int hardware_setup(void)
2646 {
2647         if (setup_vmcs_config(&vmcs_config) < 0)
2648                 return -EIO;
2649
2650         if (boot_cpu_has(X86_FEATURE_NX))
2651                 kvm_enable_efer_bits(EFER_NX);
2652
2653         if (!cpu_has_vmx_vpid())
2654                 enable_vpid = 0;
2655
2656         if (!cpu_has_vmx_ept() ||
2657             !cpu_has_vmx_ept_4levels()) {
2658                 enable_ept = 0;
2659                 enable_unrestricted_guest = 0;
2660                 enable_ept_ad_bits = 0;
2661         }
2662
2663         if (!cpu_has_vmx_ept_ad_bits())
2664                 enable_ept_ad_bits = 0;
2665
2666         if (!cpu_has_vmx_unrestricted_guest())
2667                 enable_unrestricted_guest = 0;
2668
2669         if (!cpu_has_vmx_flexpriority())
2670                 flexpriority_enabled = 0;
2671
2672         if (!cpu_has_vmx_tpr_shadow())
2673                 kvm_x86_ops->update_cr8_intercept = NULL;
2674
2675         if (enable_ept && !cpu_has_vmx_ept_2m_page())
2676                 kvm_disable_largepages();
2677
2678         if (!cpu_has_vmx_ple())
2679                 ple_gap = 0;
2680
2681         if (nested)
2682                 nested_vmx_setup_ctls_msrs();
2683
2684         return alloc_kvm_area();
2685 }
2686
2687 static __exit void hardware_unsetup(void)
2688 {
2689         free_kvm_area();
2690 }
2691
2692 static void fix_pmode_dataseg(struct kvm_vcpu *vcpu, int seg, struct kvm_segment *save)
2693 {
2694         const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2695         struct kvm_segment tmp = *save;
2696
2697         if (!(vmcs_readl(sf->base) == tmp.base && tmp.s)) {
2698                 tmp.base = vmcs_readl(sf->base);
2699                 tmp.selector = vmcs_read16(sf->selector);
2700                 tmp.s = 1;
2701         }
2702         vmx_set_segment(vcpu, &tmp, seg);
2703 }
2704
2705 static void enter_pmode(struct kvm_vcpu *vcpu)
2706 {
2707         unsigned long flags;
2708         struct vcpu_vmx *vmx = to_vmx(vcpu);
2709
2710         vmx->emulation_required = 1;
2711         vmx->rmode.vm86_active = 0;
2712
2713         vmx_segment_cache_clear(vmx);
2714
2715         vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2716
2717         flags = vmcs_readl(GUEST_RFLAGS);
2718         flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2719         flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2720         vmcs_writel(GUEST_RFLAGS, flags);
2721
2722         vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2723                         (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2724
2725         update_exception_bitmap(vcpu);
2726
2727         if (emulate_invalid_guest_state)
2728                 return;
2729
2730         fix_pmode_dataseg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2731         fix_pmode_dataseg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2732         fix_pmode_dataseg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2733         fix_pmode_dataseg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2734
2735         vmx_segment_cache_clear(vmx);
2736
2737         vmcs_write16(GUEST_SS_SELECTOR, 0);
2738         vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
2739
2740         vmcs_write16(GUEST_CS_SELECTOR,
2741                      vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
2742         vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
2743 }
2744
2745 static gva_t rmode_tss_base(struct kvm *kvm)
2746 {
2747         if (!kvm->arch.tss_addr) {
2748                 struct kvm_memslots *slots;
2749                 struct kvm_memory_slot *slot;
2750                 gfn_t base_gfn;
2751
2752                 slots = kvm_memslots(kvm);
2753                 slot = id_to_memslot(slots, 0);
2754                 base_gfn = slot->base_gfn + slot->npages - 3;
2755
2756                 return base_gfn << PAGE_SHIFT;
2757         }
2758         return kvm->arch.tss_addr;
2759 }
2760
2761 static void fix_rmode_seg(int seg, struct kvm_segment *save)
2762 {
2763         const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2764
2765         vmcs_write16(sf->selector, save->base >> 4);
2766         vmcs_write32(sf->base, save->base & 0xffff0);
2767         vmcs_write32(sf->limit, 0xffff);
2768         vmcs_write32(sf->ar_bytes, 0xf3);
2769         if (save->base & 0xf)
2770                 printk_once(KERN_WARNING "kvm: segment base is not paragraph"
2771                             " aligned when entering protected mode (seg=%d)",
2772                             seg);
2773 }
2774
2775 static void enter_rmode(struct kvm_vcpu *vcpu)
2776 {
2777         unsigned long flags;
2778         struct vcpu_vmx *vmx = to_vmx(vcpu);
2779         struct kvm_segment var;
2780
2781         if (enable_unrestricted_guest)
2782                 return;
2783
2784         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2785         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2786         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2787         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2788         vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2789
2790         vmx->emulation_required = 1;
2791         vmx->rmode.vm86_active = 1;
2792
2793
2794         /*
2795          * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2796          * vcpu. Call it here with phys address pointing 16M below 4G.
2797          */
2798         if (!vcpu->kvm->arch.tss_addr) {
2799                 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2800                              "called before entering vcpu\n");
2801                 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
2802                 vmx_set_tss_addr(vcpu->kvm, 0xfeffd000);
2803                 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2804         }
2805
2806         vmx_segment_cache_clear(vmx);
2807
2808         vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
2809         vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2810         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2811
2812         flags = vmcs_readl(GUEST_RFLAGS);
2813         vmx->rmode.save_rflags = flags;
2814
2815         flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2816
2817         vmcs_writel(GUEST_RFLAGS, flags);
2818         vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2819         update_exception_bitmap(vcpu);
2820
2821         if (emulate_invalid_guest_state)
2822                 goto continue_rmode;
2823
2824         vmx_get_segment(vcpu, &var, VCPU_SREG_SS);
2825         vmx_set_segment(vcpu, &var, VCPU_SREG_SS);
2826
2827         vmx_get_segment(vcpu, &var, VCPU_SREG_CS);
2828         vmx_set_segment(vcpu, &var, VCPU_SREG_CS);
2829
2830         vmx_get_segment(vcpu, &var, VCPU_SREG_ES);
2831         vmx_set_segment(vcpu, &var, VCPU_SREG_ES);
2832
2833         vmx_get_segment(vcpu, &var, VCPU_SREG_DS);
2834         vmx_set_segment(vcpu, &var, VCPU_SREG_DS);
2835
2836         vmx_get_segment(vcpu, &var, VCPU_SREG_GS);
2837         vmx_set_segment(vcpu, &var, VCPU_SREG_GS);
2838
2839         vmx_get_segment(vcpu, &var, VCPU_SREG_FS);
2840         vmx_set_segment(vcpu, &var, VCPU_SREG_FS);
2841
2842 continue_rmode:
2843         kvm_mmu_reset_context(vcpu);
2844 }
2845
2846 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2847 {
2848         struct vcpu_vmx *vmx = to_vmx(vcpu);
2849         struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2850
2851         if (!msr)
2852                 return;
2853
2854         /*
2855          * Force kernel_gs_base reloading before EFER changes, as control
2856          * of this msr depends on is_long_mode().
2857          */
2858         vmx_load_host_state(to_vmx(vcpu));
2859         vcpu->arch.efer = efer;
2860         if (efer & EFER_LMA) {
2861                 vmcs_write32(VM_ENTRY_CONTROLS,
2862                              vmcs_read32(VM_ENTRY_CONTROLS) |
2863                              VM_ENTRY_IA32E_MODE);
2864                 msr->data = efer;
2865         } else {
2866                 vmcs_write32(VM_ENTRY_CONTROLS,
2867                              vmcs_read32(VM_ENTRY_CONTROLS) &
2868                              ~VM_ENTRY_IA32E_MODE);
2869
2870                 msr->data = efer & ~EFER_LME;
2871         }
2872         setup_msrs(vmx);
2873 }
2874
2875 #ifdef CONFIG_X86_64
2876
2877 static void enter_lmode(struct kvm_vcpu *vcpu)
2878 {
2879         u32 guest_tr_ar;
2880
2881         vmx_segment_cache_clear(to_vmx(vcpu));
2882
2883         guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2884         if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
2885                 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
2886                                      __func__);
2887                 vmcs_write32(GUEST_TR_AR_BYTES,
2888                              (guest_tr_ar & ~AR_TYPE_MASK)
2889                              | AR_TYPE_BUSY_64_TSS);
2890         }
2891         vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2892 }
2893
2894 static void exit_lmode(struct kvm_vcpu *vcpu)
2895 {
2896         vmcs_write32(VM_ENTRY_CONTROLS,
2897                      vmcs_read32(VM_ENTRY_CONTROLS)
2898                      & ~VM_ENTRY_IA32E_MODE);
2899         vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2900 }
2901
2902 #endif
2903
2904 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
2905 {
2906         vpid_sync_context(to_vmx(vcpu));
2907         if (enable_ept) {
2908                 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
2909                         return;
2910                 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
2911         }
2912 }
2913
2914 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2915 {
2916         ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2917
2918         vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
2919         vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
2920 }
2921
2922 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
2923 {
2924         if (enable_ept && is_paging(vcpu))
2925                 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2926         __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
2927 }
2928
2929 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2930 {
2931         ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2932
2933         vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
2934         vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
2935 }
2936
2937 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2938 {
2939         if (!test_bit(VCPU_EXREG_PDPTR,
2940                       (unsigned long *)&vcpu->arch.regs_dirty))
2941                 return;
2942
2943         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
2944                 vmcs_write64(GUEST_PDPTR0, vcpu->arch.mmu.pdptrs[0]);
2945                 vmcs_write64(GUEST_PDPTR1, vcpu->arch.mmu.pdptrs[1]);
2946                 vmcs_write64(GUEST_PDPTR2, vcpu->arch.mmu.pdptrs[2]);
2947                 vmcs_write64(GUEST_PDPTR3, vcpu->arch.mmu.pdptrs[3]);
2948         }
2949 }
2950
2951 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2952 {
2953         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
2954                 vcpu->arch.mmu.pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2955                 vcpu->arch.mmu.pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2956                 vcpu->arch.mmu.pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2957                 vcpu->arch.mmu.pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2958         }
2959
2960         __set_bit(VCPU_EXREG_PDPTR,
2961                   (unsigned long *)&vcpu->arch.regs_avail);
2962         __set_bit(VCPU_EXREG_PDPTR,
2963                   (unsigned long *)&vcpu->arch.regs_dirty);
2964 }
2965
2966 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
2967
2968 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
2969                                         unsigned long cr0,
2970                                         struct kvm_vcpu *vcpu)
2971 {
2972         if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
2973                 vmx_decache_cr3(vcpu);
2974         if (!(cr0 & X86_CR0_PG)) {
2975                 /* From paging/starting to nonpaging */
2976                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
2977                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
2978                              (CPU_BASED_CR3_LOAD_EXITING |
2979                               CPU_BASED_CR3_STORE_EXITING));
2980                 vcpu->arch.cr0 = cr0;
2981                 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2982         } else if (!is_paging(vcpu)) {
2983                 /* From nonpaging to paging */
2984                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
2985                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
2986                              ~(CPU_BASED_CR3_LOAD_EXITING |
2987                                CPU_BASED_CR3_STORE_EXITING));
2988                 vcpu->arch.cr0 = cr0;
2989                 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2990         }
2991
2992         if (!(cr0 & X86_CR0_WP))
2993                 *hw_cr0 &= ~X86_CR0_WP;
2994 }
2995
2996 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2997 {
2998         struct vcpu_vmx *vmx = to_vmx(vcpu);
2999         unsigned long hw_cr0;
3000
3001         if (enable_unrestricted_guest)
3002                 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST)
3003                         | KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
3004         else
3005                 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON;
3006
3007         if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
3008                 enter_pmode(vcpu);
3009
3010         if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
3011                 enter_rmode(vcpu);
3012
3013 #ifdef CONFIG_X86_64
3014         if (vcpu->arch.efer & EFER_LME) {
3015                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
3016                         enter_lmode(vcpu);
3017                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
3018                         exit_lmode(vcpu);
3019         }
3020 #endif
3021
3022         if (enable_ept)
3023                 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
3024
3025         if (!vcpu->fpu_active)
3026                 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
3027
3028         vmcs_writel(CR0_READ_SHADOW, cr0);
3029         vmcs_writel(GUEST_CR0, hw_cr0);
3030         vcpu->arch.cr0 = cr0;
3031         __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
3032 }
3033
3034 static u64 construct_eptp(unsigned long root_hpa)
3035 {
3036         u64 eptp;
3037
3038         /* TODO write the value reading from MSR */
3039         eptp = VMX_EPT_DEFAULT_MT |
3040                 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
3041         if (enable_ept_ad_bits)
3042                 eptp |= VMX_EPT_AD_ENABLE_BIT;
3043         eptp |= (root_hpa & PAGE_MASK);
3044
3045         return eptp;
3046 }
3047
3048 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
3049 {
3050         unsigned long guest_cr3;
3051         u64 eptp;
3052
3053         guest_cr3 = cr3;
3054         if (enable_ept) {
3055                 eptp = construct_eptp(cr3);
3056                 vmcs_write64(EPT_POINTER, eptp);
3057                 guest_cr3 = is_paging(vcpu) ? kvm_read_cr3(vcpu) :
3058                         vcpu->kvm->arch.ept_identity_map_addr;
3059                 ept_load_pdptrs(vcpu);
3060         }
3061
3062         vmx_flush_tlb(vcpu);
3063         vmcs_writel(GUEST_CR3, guest_cr3);
3064 }
3065
3066 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3067 {
3068         unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
3069                     KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
3070
3071         if (cr4 & X86_CR4_VMXE) {
3072                 /*
3073                  * To use VMXON (and later other VMX instructions), a guest
3074                  * must first be able to turn on cr4.VMXE (see handle_vmon()).
3075                  * So basically the check on whether to allow nested VMX
3076                  * is here.
3077                  */
3078                 if (!nested_vmx_allowed(vcpu))
3079                         return 1;
3080         } else if (to_vmx(vcpu)->nested.vmxon)
3081                 return 1;
3082
3083         vcpu->arch.cr4 = cr4;
3084         if (enable_ept) {
3085                 if (!is_paging(vcpu)) {
3086                         hw_cr4 &= ~X86_CR4_PAE;
3087                         hw_cr4 |= X86_CR4_PSE;
3088                 } else if (!(cr4 & X86_CR4_PAE)) {
3089                         hw_cr4 &= ~X86_CR4_PAE;
3090                 }
3091         }
3092
3093         vmcs_writel(CR4_READ_SHADOW, cr4);
3094         vmcs_writel(GUEST_CR4, hw_cr4);
3095         return 0;
3096 }
3097
3098 static void vmx_get_segment(struct kvm_vcpu *vcpu,
3099                             struct kvm_segment *var, int seg)
3100 {
3101         struct vcpu_vmx *vmx = to_vmx(vcpu);
3102         u32 ar;
3103
3104         if (vmx->rmode.vm86_active
3105             && (seg == VCPU_SREG_TR || seg == VCPU_SREG_ES
3106                 || seg == VCPU_SREG_DS || seg == VCPU_SREG_FS
3107                 || seg == VCPU_SREG_GS)) {
3108                 *var = vmx->rmode.segs[seg];
3109                 if (seg == VCPU_SREG_TR
3110                     || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3111                         return;
3112                 var->base = vmx_read_guest_seg_base(vmx, seg);
3113                 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3114                 return;
3115         }
3116         var->base = vmx_read_guest_seg_base(vmx, seg);
3117         var->limit = vmx_read_guest_seg_limit(vmx, seg);
3118         var->selector = vmx_read_guest_seg_selector(vmx, seg);
3119         ar = vmx_read_guest_seg_ar(vmx, seg);
3120         if ((ar & AR_UNUSABLE_MASK) && !emulate_invalid_guest_state)
3121                 ar = 0;
3122         var->type = ar & 15;
3123         var->s = (ar >> 4) & 1;
3124         var->dpl = (ar >> 5) & 3;
3125         var->present = (ar >> 7) & 1;
3126         var->avl = (ar >> 12) & 1;
3127         var->l = (ar >> 13) & 1;
3128         var->db = (ar >> 14) & 1;
3129         var->g = (ar >> 15) & 1;
3130         var->unusable = (ar >> 16) & 1;
3131 }
3132
3133 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3134 {
3135         struct kvm_segment s;
3136
3137         if (to_vmx(vcpu)->rmode.vm86_active) {
3138                 vmx_get_segment(vcpu, &s, seg);
3139                 return s.base;
3140         }
3141         return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3142 }
3143
3144 static int __vmx_get_cpl(struct kvm_vcpu *vcpu)
3145 {
3146         if (!is_protmode(vcpu))
3147                 return 0;
3148
3149         if (!is_long_mode(vcpu)
3150             && (kvm_get_rflags(vcpu) & X86_EFLAGS_VM)) /* if virtual 8086 */
3151                 return 3;
3152
3153         return vmx_read_guest_seg_selector(to_vmx(vcpu), VCPU_SREG_CS) & 3;
3154 }
3155
3156 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
3157 {
3158         struct vcpu_vmx *vmx = to_vmx(vcpu);
3159
3160         /*
3161          * If we enter real mode with cs.sel & 3 != 0, the normal CPL calculations
3162          * fail; use the cache instead.
3163          */
3164         if (unlikely(vmx->emulation_required && emulate_invalid_guest_state)) {
3165                 return vmx->cpl;
3166         }
3167
3168         if (!test_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail)) {
3169                 __set_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
3170                 vmx->cpl = __vmx_get_cpl(vcpu);
3171         }
3172
3173         return vmx->cpl;
3174 }
3175
3176
3177 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3178 {
3179         u32 ar;
3180
3181         if (var->unusable || !var->present)
3182                 ar = 1 << 16;
3183         else {
3184                 ar = var->type & 15;
3185                 ar |= (var->s & 1) << 4;
3186                 ar |= (var->dpl & 3) << 5;
3187                 ar |= (var->present & 1) << 7;
3188                 ar |= (var->avl & 1) << 12;
3189                 ar |= (var->l & 1) << 13;
3190                 ar |= (var->db & 1) << 14;
3191                 ar |= (var->g & 1) << 15;
3192         }
3193
3194         return ar;
3195 }
3196
3197 static void vmx_set_segment(struct kvm_vcpu *vcpu,
3198                             struct kvm_segment *var, int seg)
3199 {
3200         struct vcpu_vmx *vmx = to_vmx(vcpu);
3201         const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3202         u32 ar;
3203
3204         vmx_segment_cache_clear(vmx);
3205
3206         if (vmx->rmode.vm86_active && seg == VCPU_SREG_TR) {
3207                 vmcs_write16(sf->selector, var->selector);
3208                 vmx->rmode.segs[VCPU_SREG_TR] = *var;
3209                 return;
3210         }
3211         vmcs_writel(sf->base, var->base);
3212         vmcs_write32(sf->limit, var->limit);
3213         vmcs_write16(sf->selector, var->selector);
3214         if (vmx->rmode.vm86_active && var->s) {
3215                 vmx->rmode.segs[seg] = *var;
3216                 /*
3217                  * Hack real-mode segments into vm86 compatibility.
3218                  */
3219                 if (var->base == 0xffff0000 && var->selector == 0xf000)
3220                         vmcs_writel(sf->base, 0xf0000);
3221                 ar = 0xf3;
3222         } else
3223                 ar = vmx_segment_access_rights(var);
3224
3225         /*
3226          *   Fix the "Accessed" bit in AR field of segment registers for older
3227          * qemu binaries.
3228          *   IA32 arch specifies that at the time of processor reset the
3229          * "Accessed" bit in the AR field of segment registers is 1. And qemu
3230          * is setting it to 0 in the userland code. This causes invalid guest
3231          * state vmexit when "unrestricted guest" mode is turned on.
3232          *    Fix for this setup issue in cpu_reset is being pushed in the qemu
3233          * tree. Newer qemu binaries with that qemu fix would not need this
3234          * kvm hack.
3235          */
3236         if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3237                 ar |= 0x1; /* Accessed */
3238
3239         vmcs_write32(sf->ar_bytes, ar);
3240         __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
3241
3242         /*
3243          * Fix segments for real mode guest in hosts that don't have
3244          * "unrestricted_mode" or it was disabled.
3245          * This is done to allow migration of the guests from hosts with
3246          * unrestricted guest like Westmere to older host that don't have
3247          * unrestricted guest like Nehelem.
3248          */
3249         if (!enable_unrestricted_guest && vmx->rmode.vm86_active) {
3250                 switch (seg) {
3251                 case VCPU_SREG_CS:
3252                         vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
3253                         vmcs_write32(GUEST_CS_LIMIT, 0xffff);
3254                         if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
3255                                 vmcs_writel(GUEST_CS_BASE, 0xf0000);
3256                         vmcs_write16(GUEST_CS_SELECTOR,
3257                                      vmcs_readl(GUEST_CS_BASE) >> 4);
3258                         break;
3259                 case VCPU_SREG_ES:
3260                 case VCPU_SREG_DS:
3261                 case VCPU_SREG_GS:
3262                 case VCPU_SREG_FS:
3263                         fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3264                         break;
3265                 case VCPU_SREG_SS:
3266                         vmcs_write16(GUEST_SS_SELECTOR,
3267                                      vmcs_readl(GUEST_SS_BASE) >> 4);
3268                         vmcs_write32(GUEST_SS_LIMIT, 0xffff);
3269                         vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
3270                         break;
3271                 }
3272         }
3273 }
3274
3275 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3276 {
3277         u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3278
3279         *db = (ar >> 14) & 1;
3280         *l = (ar >> 13) & 1;
3281 }
3282
3283 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3284 {
3285         dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3286         dt->address = vmcs_readl(GUEST_IDTR_BASE);
3287 }
3288
3289 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3290 {
3291         vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3292         vmcs_writel(GUEST_IDTR_BASE, dt->address);
3293 }
3294
3295 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3296 {
3297         dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3298         dt->address = vmcs_readl(GUEST_GDTR_BASE);
3299 }
3300
3301 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3302 {
3303         vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3304         vmcs_writel(GUEST_GDTR_BASE, dt->address);
3305 }
3306
3307 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3308 {
3309         struct kvm_segment var;
3310         u32 ar;
3311
3312         vmx_get_segment(vcpu, &var, seg);
3313         ar = vmx_segment_access_rights(&var);
3314
3315         if (var.base != (var.selector << 4))
3316                 return false;
3317         if (var.limit < 0xffff)
3318                 return false;
3319         if (((ar | (3 << AR_DPL_SHIFT)) & ~(AR_G_MASK | AR_DB_MASK)) != 0xf3)
3320                 return false;
3321
3322         return true;
3323 }
3324
3325 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3326 {
3327         struct kvm_segment cs;
3328         unsigned int cs_rpl;
3329
3330         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3331         cs_rpl = cs.selector & SELECTOR_RPL_MASK;
3332
3333         if (cs.unusable)
3334                 return false;
3335         if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
3336                 return false;
3337         if (!cs.s)
3338                 return false;
3339         if (cs.type & AR_TYPE_WRITEABLE_MASK) {
3340                 if (cs.dpl > cs_rpl)
3341                         return false;
3342         } else {
3343                 if (cs.dpl != cs_rpl)
3344                         return false;
3345         }
3346         if (!cs.present)
3347                 return false;
3348
3349         /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3350         return true;
3351 }
3352
3353 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3354 {
3355         struct kvm_segment ss;
3356         unsigned int ss_rpl;
3357
3358         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3359         ss_rpl = ss.selector & SELECTOR_RPL_MASK;
3360
3361         if (ss.unusable)
3362                 return true;
3363         if (ss.type != 3 && ss.type != 7)
3364                 return false;
3365         if (!ss.s)
3366                 return false;
3367         if (ss.dpl != ss_rpl) /* DPL != RPL */
3368                 return false;
3369         if (!ss.present)
3370                 return false;
3371
3372         return true;
3373 }
3374
3375 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3376 {
3377         struct kvm_segment var;
3378         unsigned int rpl;
3379
3380         vmx_get_segment(vcpu, &var, seg);
3381         rpl = var.selector & SELECTOR_RPL_MASK;
3382
3383         if (var.unusable)
3384                 return true;
3385         if (!var.s)
3386                 return false;
3387         if (!var.present)
3388                 return false;
3389         if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
3390                 if (var.dpl < rpl) /* DPL < RPL */
3391                         return false;
3392         }
3393
3394         /* TODO: Add other members to kvm_segment_field to allow checking for other access
3395          * rights flags
3396          */
3397         return true;
3398 }
3399
3400 static bool tr_valid(struct kvm_vcpu *vcpu)
3401 {
3402         struct kvm_segment tr;
3403
3404         vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3405
3406         if (tr.unusable)
3407                 return false;
3408         if (tr.selector & SELECTOR_TI_MASK)     /* TI = 1 */
3409                 return false;
3410         if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3411                 return false;
3412         if (!tr.present)
3413                 return false;
3414
3415         return true;
3416 }
3417
3418 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3419 {
3420         struct kvm_segment ldtr;
3421
3422         vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3423
3424         if (ldtr.unusable)
3425                 return true;
3426         if (ldtr.selector & SELECTOR_TI_MASK)   /* TI = 1 */
3427                 return false;
3428         if (ldtr.type != 2)
3429                 return false;
3430         if (!ldtr.present)
3431                 return false;
3432
3433         return true;
3434 }
3435
3436 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3437 {
3438         struct kvm_segment cs, ss;
3439
3440         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3441         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3442
3443         return ((cs.selector & SELECTOR_RPL_MASK) ==
3444                  (ss.selector & SELECTOR_RPL_MASK));
3445 }
3446
3447 /*
3448  * Check if guest state is valid. Returns true if valid, false if
3449  * not.
3450  * We assume that registers are always usable
3451  */
3452 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3453 {
3454         /* real mode guest state checks */
3455         if (!is_protmode(vcpu)) {
3456                 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3457                         return false;
3458                 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3459                         return false;
3460                 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3461                         return false;
3462                 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3463                         return false;
3464                 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3465                         return false;
3466                 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3467                         return false;
3468         } else {
3469         /* protected mode guest state checks */
3470                 if (!cs_ss_rpl_check(vcpu))
3471                         return false;
3472                 if (!code_segment_valid(vcpu))
3473                         return false;
3474                 if (!stack_segment_valid(vcpu))
3475                         return false;
3476                 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3477                         return false;
3478                 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3479                         return false;
3480                 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3481                         return false;
3482                 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3483                         return false;
3484                 if (!tr_valid(vcpu))
3485                         return false;
3486                 if (!ldtr_valid(vcpu))
3487                         return false;
3488         }
3489         /* TODO:
3490          * - Add checks on RIP
3491          * - Add checks on RFLAGS
3492          */
3493
3494         return true;
3495 }
3496
3497 static int init_rmode_tss(struct kvm *kvm)
3498 {
3499         gfn_t fn;
3500         u16 data = 0;
3501         int r, idx, ret = 0;
3502
3503         idx = srcu_read_lock(&kvm->srcu);
3504         fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
3505         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3506         if (r < 0)
3507                 goto out;
3508         data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3509         r = kvm_write_guest_page(kvm, fn++, &data,
3510                         TSS_IOPB_BASE_OFFSET, sizeof(u16));
3511         if (r < 0)
3512                 goto out;
3513         r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3514         if (r < 0)
3515                 goto out;
3516         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3517         if (r < 0)
3518                 goto out;
3519         data = ~0;
3520         r = kvm_write_guest_page(kvm, fn, &data,
3521                                  RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3522                                  sizeof(u8));
3523         if (r < 0)
3524                 goto out;
3525
3526         ret = 1;
3527 out:
3528         srcu_read_unlock(&kvm->srcu, idx);
3529         return ret;
3530 }
3531
3532 static int init_rmode_identity_map(struct kvm *kvm)
3533 {
3534         int i, idx, r, ret;
3535         pfn_t identity_map_pfn;
3536         u32 tmp;
3537
3538         if (!enable_ept)
3539                 return 1;
3540         if (unlikely(!kvm->arch.ept_identity_pagetable)) {
3541                 printk(KERN_ERR "EPT: identity-mapping pagetable "
3542                         "haven't been allocated!\n");
3543                 return 0;
3544         }
3545         if (likely(kvm->arch.ept_identity_pagetable_done))
3546                 return 1;
3547         ret = 0;
3548         identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
3549         idx = srcu_read_lock(&kvm->srcu);
3550         r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3551         if (r < 0)
3552                 goto out;
3553         /* Set up identity-mapping pagetable for EPT in real mode */
3554         for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3555                 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3556                         _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3557                 r = kvm_write_guest_page(kvm, identity_map_pfn,
3558                                 &tmp, i * sizeof(tmp), sizeof(tmp));
3559                 if (r < 0)
3560                         goto out;
3561         }
3562         kvm->arch.ept_identity_pagetable_done = true;
3563         ret = 1;
3564 out:
3565         srcu_read_unlock(&kvm->srcu, idx);
3566         return ret;
3567 }
3568
3569 static void seg_setup(int seg)
3570 {
3571         const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3572         unsigned int ar;
3573
3574         vmcs_write16(sf->selector, 0);
3575         vmcs_writel(sf->base, 0);
3576         vmcs_write32(sf->limit, 0xffff);
3577         if (enable_unrestricted_guest) {
3578                 ar = 0x93;
3579                 if (seg == VCPU_SREG_CS)
3580                         ar |= 0x08; /* code segment */
3581         } else
3582                 ar = 0xf3;
3583
3584         vmcs_write32(sf->ar_bytes, ar);
3585 }
3586
3587 static int alloc_apic_access_page(struct kvm *kvm)
3588 {
3589         struct page *page;
3590         struct kvm_userspace_memory_region kvm_userspace_mem;
3591         int r = 0;
3592
3593         mutex_lock(&kvm->slots_lock);
3594         if (kvm->arch.apic_access_page)
3595                 goto out;
3596         kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
3597         kvm_userspace_mem.flags = 0;
3598         kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
3599         kvm_userspace_mem.memory_size = PAGE_SIZE;
3600         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
3601         if (r)
3602                 goto out;
3603
3604         page = gfn_to_page(kvm, 0xfee00);
3605         if (is_error_page(page)) {
3606                 r = -EFAULT;
3607                 goto out;
3608         }
3609
3610         kvm->arch.apic_access_page = page;
3611 out:
3612         mutex_unlock(&kvm->slots_lock);
3613         return r;
3614 }
3615
3616 static int alloc_identity_pagetable(struct kvm *kvm)
3617 {
3618         struct page *page;
3619         struct kvm_userspace_memory_region kvm_userspace_mem;
3620         int r = 0;
3621
3622         mutex_lock(&kvm->slots_lock);
3623         if (kvm->arch.ept_identity_pagetable)
3624                 goto out;
3625         kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
3626         kvm_userspace_mem.flags = 0;
3627         kvm_userspace_mem.guest_phys_addr =
3628                 kvm->arch.ept_identity_map_addr;
3629         kvm_userspace_mem.memory_size = PAGE_SIZE;
3630         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
3631         if (r)
3632                 goto out;
3633
3634         page = gfn_to_page(kvm, kvm->arch.ept_identity_map_addr >> PAGE_SHIFT);
3635         if (is_error_page(page)) {
3636                 r = -EFAULT;
3637                 goto out;
3638         }
3639
3640         kvm->arch.ept_identity_pagetable = page;
3641 out:
3642         mutex_unlock(&kvm->slots_lock);
3643         return r;
3644 }
3645
3646 static void allocate_vpid(struct vcpu_vmx *vmx)
3647 {
3648         int vpid;
3649
3650         vmx->vpid = 0;
3651         if (!enable_vpid)
3652                 return;
3653         spin_lock(&vmx_vpid_lock);
3654         vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3655         if (vpid < VMX_NR_VPIDS) {
3656                 vmx->vpid = vpid;
3657                 __set_bit(vpid, vmx_vpid_bitmap);
3658         }
3659         spin_unlock(&vmx_vpid_lock);
3660 }
3661
3662 static void free_vpid(struct vcpu_vmx *vmx)
3663 {
3664         if (!enable_vpid)
3665                 return;
3666         spin_lock(&vmx_vpid_lock);
3667         if (vmx->vpid != 0)
3668                 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
3669         spin_unlock(&vmx_vpid_lock);
3670 }
3671
3672 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, u32 msr)
3673 {
3674         int f = sizeof(unsigned long);
3675
3676         if (!cpu_has_vmx_msr_bitmap())
3677                 return;
3678
3679         /*
3680          * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3681          * have the write-low and read-high bitmap offsets the wrong way round.
3682          * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3683          */
3684         if (msr <= 0x1fff) {
3685                 __clear_bit(msr, msr_bitmap + 0x000 / f); /* read-low */
3686                 __clear_bit(msr, msr_bitmap + 0x800 / f); /* write-low */
3687         } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3688                 msr &= 0x1fff;
3689                 __clear_bit(msr, msr_bitmap + 0x400 / f); /* read-high */
3690                 __clear_bit(msr, msr_bitmap + 0xc00 / f); /* write-high */
3691         }
3692 }
3693
3694 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
3695 {
3696         if (!longmode_only)
3697                 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy, msr);
3698         __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode, msr);
3699 }
3700
3701 /*
3702  * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3703  * will not change in the lifetime of the guest.
3704  * Note that host-state that does change is set elsewhere. E.g., host-state
3705  * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3706  */
3707 static void vmx_set_constant_host_state(void)
3708 {
3709         u32 low32, high32;
3710         unsigned long tmpl;
3711         struct desc_ptr dt;
3712
3713         vmcs_writel(HOST_CR0, read_cr0() & ~X86_CR0_TS);  /* 22.2.3 */
3714         vmcs_writel(HOST_CR4, read_cr4());  /* 22.2.3, 22.2.5 */
3715         vmcs_writel(HOST_CR3, read_cr3());  /* 22.2.3  FIXME: shadow tables */
3716
3717         vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
3718 #ifdef CONFIG_X86_64
3719         /*
3720          * Load null selectors, so we can avoid reloading them in
3721          * __vmx_load_host_state(), in case userspace uses the null selectors
3722          * too (the expected case).
3723          */
3724         vmcs_write16(HOST_DS_SELECTOR, 0);
3725         vmcs_write16(HOST_ES_SELECTOR, 0);
3726 #else
3727         vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3728         vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3729 #endif
3730         vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
3731         vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
3732
3733         native_store_idt(&dt);
3734         vmcs_writel(HOST_IDTR_BASE, dt.address);   /* 22.2.4 */
3735
3736         vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
3737
3738         rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
3739         vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
3740         rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
3741         vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl);   /* 22.2.3 */
3742
3743         if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
3744                 rdmsr(MSR_IA32_CR_PAT, low32, high32);
3745                 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
3746         }
3747 }
3748
3749 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
3750 {
3751         vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
3752         if (enable_ept)
3753                 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
3754         if (is_guest_mode(&vmx->vcpu))
3755                 vmx->vcpu.arch.cr4_guest_owned_bits &=
3756                         ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
3757         vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
3758 }
3759
3760 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
3761 {
3762         u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
3763         if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
3764                 exec_control &= ~CPU_BASED_TPR_SHADOW;
3765 #ifdef CONFIG_X86_64
3766                 exec_control |= CPU_BASED_CR8_STORE_EXITING |
3767                                 CPU_BASED_CR8_LOAD_EXITING;
3768 #endif
3769         }
3770         if (!enable_ept)
3771                 exec_control |= CPU_BASED_CR3_STORE_EXITING |
3772                                 CPU_BASED_CR3_LOAD_EXITING  |
3773                                 CPU_BASED_INVLPG_EXITING;
3774         return exec_control;
3775 }
3776
3777 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
3778 {
3779         u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
3780         if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
3781                 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
3782         if (vmx->vpid == 0)
3783                 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
3784         if (!enable_ept) {
3785                 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
3786                 enable_unrestricted_guest = 0;
3787                 /* Enable INVPCID for non-ept guests may cause performance regression. */
3788                 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
3789         }
3790         if (!enable_unrestricted_guest)
3791                 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
3792         if (!ple_gap)
3793                 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
3794         return exec_control;
3795 }
3796
3797 static void ept_set_mmio_spte_mask(void)
3798 {
3799         /*
3800          * EPT Misconfigurations can be generated if the value of bits 2:0
3801          * of an EPT paging-structure entry is 110b (write/execute).
3802          * Also, magic bits (0xffull << 49) is set to quickly identify mmio
3803          * spte.
3804          */
3805         kvm_mmu_set_mmio_spte_mask(0xffull << 49 | 0x6ull);
3806 }
3807
3808 /*
3809  * Sets up the vmcs for emulated real mode.
3810  */
3811 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
3812 {
3813 #ifdef CONFIG_X86_64
3814         unsigned long a;
3815 #endif
3816         int i;
3817
3818         /* I/O */
3819         vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
3820         vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
3821
3822         if (cpu_has_vmx_msr_bitmap())
3823                 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
3824
3825         vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
3826
3827         /* Control */
3828         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
3829                 vmcs_config.pin_based_exec_ctrl);
3830
3831         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
3832
3833         if (cpu_has_secondary_exec_ctrls()) {
3834                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
3835                                 vmx_secondary_exec_control(vmx));
3836         }
3837
3838         if (ple_gap) {
3839                 vmcs_write32(PLE_GAP, ple_gap);
3840                 vmcs_write32(PLE_WINDOW, ple_window);
3841         }
3842
3843         vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
3844         vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
3845         vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
3846
3847         vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
3848         vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
3849         vmx_set_constant_host_state();
3850 #ifdef CONFIG_X86_64
3851         rdmsrl(MSR_FS_BASE, a);
3852         vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
3853         rdmsrl(MSR_GS_BASE, a);
3854         vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
3855 #else
3856         vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
3857         vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
3858 #endif
3859
3860         vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
3861         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
3862         vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
3863         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
3864         vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
3865
3866         if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
3867                 u32 msr_low, msr_high;
3868                 u64 host_pat;
3869                 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
3870                 host_pat = msr_low | ((u64) msr_high << 32);
3871                 /* Write the default value follow host pat */
3872                 vmcs_write64(GUEST_IA32_PAT, host_pat);
3873                 /* Keep arch.pat sync with GUEST_IA32_PAT */
3874                 vmx->vcpu.arch.pat = host_pat;
3875         }
3876
3877         for (i = 0; i < NR_VMX_MSR; ++i) {
3878                 u32 index = vmx_msr_index[i];
3879                 u32 data_low, data_high;
3880                 int j = vmx->nmsrs;
3881
3882                 if (rdmsr_safe(index, &data_low, &data_high) < 0)
3883                         continue;
3884                 if (wrmsr_safe(index, data_low, data_high) < 0)
3885                         continue;
3886                 vmx->guest_msrs[j].index = i;
3887                 vmx->guest_msrs[j].data = 0;
3888                 vmx->guest_msrs[j].mask = -1ull;
3889                 ++vmx->nmsrs;
3890         }
3891
3892         vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
3893
3894         /* 22.2.1, 20.8.1 */
3895         vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
3896
3897         vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
3898         set_cr4_guest_host_mask(vmx);
3899
3900         kvm_write_tsc(&vmx->vcpu, 0);
3901
3902         return 0;
3903 }
3904
3905 static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
3906 {
3907         struct vcpu_vmx *vmx = to_vmx(vcpu);
3908         u64 msr;
3909         int ret;
3910
3911         vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
3912
3913         vmx->rmode.vm86_active = 0;
3914
3915         vmx->soft_vnmi_blocked = 0;
3916
3917         vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
3918         kvm_set_cr8(&vmx->vcpu, 0);
3919         msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
3920         if (kvm_vcpu_is_bsp(&vmx->vcpu))
3921                 msr |= MSR_IA32_APICBASE_BSP;
3922         kvm_set_apic_base(&vmx->vcpu, msr);
3923
3924         ret = fx_init(&vmx->vcpu);
3925         if (ret != 0)
3926                 goto out;
3927
3928         vmx_segment_cache_clear(vmx);
3929
3930         seg_setup(VCPU_SREG_CS);
3931         /*
3932          * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
3933          * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4.  Sigh.
3934          */
3935         if (kvm_vcpu_is_bsp(&vmx->vcpu)) {
3936                 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
3937                 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
3938         } else {
3939                 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
3940                 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
3941         }
3942
3943         seg_setup(VCPU_SREG_DS);
3944         seg_setup(VCPU_SREG_ES);
3945         seg_setup(VCPU_SREG_FS);
3946         seg_setup(VCPU_SREG_GS);
3947         seg_setup(VCPU_SREG_SS);
3948
3949         vmcs_write16(GUEST_TR_SELECTOR, 0);
3950         vmcs_writel(GUEST_TR_BASE, 0);
3951         vmcs_write32(GUEST_TR_LIMIT, 0xffff);
3952         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3953
3954         vmcs_write16(GUEST_LDTR_SELECTOR, 0);
3955         vmcs_writel(GUEST_LDTR_BASE, 0);
3956         vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
3957         vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
3958
3959         vmcs_write32(GUEST_SYSENTER_CS, 0);
3960         vmcs_writel(GUEST_SYSENTER_ESP, 0);
3961         vmcs_writel(GUEST_SYSENTER_EIP, 0);
3962
3963         vmcs_writel(GUEST_RFLAGS, 0x02);
3964         if (kvm_vcpu_is_bsp(&vmx->vcpu))
3965                 kvm_rip_write(vcpu, 0xfff0);
3966         else
3967                 kvm_rip_write(vcpu, 0);
3968         kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
3969
3970         vmcs_writel(GUEST_GDTR_BASE, 0);
3971         vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
3972
3973         vmcs_writel(GUEST_IDTR_BASE, 0);
3974         vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
3975
3976         vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
3977         vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
3978         vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
3979
3980         /* Special registers */
3981         vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
3982
3983         setup_msrs(vmx);
3984
3985         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
3986
3987         if (cpu_has_vmx_tpr_shadow()) {
3988                 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
3989                 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
3990                         vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
3991                                      __pa(vmx->vcpu.arch.apic->regs));
3992                 vmcs_write32(TPR_THRESHOLD, 0);
3993         }
3994
3995         if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
3996                 vmcs_write64(APIC_ACCESS_ADDR,
3997                              page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
3998
3999         if (vmx->vpid != 0)
4000                 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4001
4002         vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4003         vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4004         vmx_set_cr0(&vmx->vcpu, kvm_read_cr0(vcpu)); /* enter rmode */
4005         srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4006         vmx_set_cr4(&vmx->vcpu, 0);
4007         vmx_set_efer(&vmx->vcpu, 0);
4008         vmx_fpu_activate(&vmx->vcpu);
4009         update_exception_bitmap(&vmx->vcpu);
4010
4011         vpid_sync_context(vmx);
4012
4013         ret = 0;
4014
4015         /* HACK: Don't enable emulation on guest boot/reset */
4016         vmx->emulation_required = 0;
4017
4018 out:
4019         return ret;
4020 }
4021
4022 /*
4023  * In nested virtualization, check if L1 asked to exit on external interrupts.
4024  * For most existing hypervisors, this will always return true.
4025  */
4026 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
4027 {
4028         return get_vmcs12(vcpu)->pin_based_vm_exec_control &
4029                 PIN_BASED_EXT_INTR_MASK;
4030 }
4031
4032 static void enable_irq_window(struct kvm_vcpu *vcpu)
4033 {
4034         u32 cpu_based_vm_exec_control;
4035         if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu)) {
4036                 /*
4037                  * We get here if vmx_interrupt_allowed() said we can't
4038                  * inject to L1 now because L2 must run. Ask L2 to exit
4039                  * right after entry, so we can inject to L1 more promptly.
4040                  */
4041                 kvm_make_request(KVM_REQ_IMMEDIATE_EXIT, vcpu);
4042                 return;
4043         }
4044
4045         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4046         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
4047         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4048 }
4049
4050 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4051 {
4052         u32 cpu_based_vm_exec_control;
4053
4054         if (!cpu_has_virtual_nmis()) {
4055                 enable_irq_window(vcpu);
4056                 return;
4057         }
4058
4059         if (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4060                 enable_irq_window(vcpu);
4061                 return;
4062         }
4063         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4064         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
4065         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4066 }
4067
4068 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4069 {
4070         struct vcpu_vmx *vmx = to_vmx(vcpu);
4071         uint32_t intr;
4072         int irq = vcpu->arch.interrupt.nr;
4073
4074         trace_kvm_inj_virq(irq);
4075
4076         ++vcpu->stat.irq_injections;
4077         if (vmx->rmode.vm86_active) {
4078                 int inc_eip = 0;
4079                 if (vcpu->arch.interrupt.soft)
4080                         inc_eip = vcpu->arch.event_exit_inst_len;
4081                 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
4082                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4083                 return;
4084         }
4085         intr = irq | INTR_INFO_VALID_MASK;
4086         if (vcpu->arch.interrupt.soft) {
4087                 intr |= INTR_TYPE_SOFT_INTR;
4088                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4089                              vmx->vcpu.arch.event_exit_inst_len);
4090         } else
4091                 intr |= INTR_TYPE_EXT_INTR;
4092         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4093 }
4094
4095 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4096 {
4097         struct vcpu_vmx *vmx = to_vmx(vcpu);
4098
4099         if (is_guest_mode(vcpu))
4100                 return;
4101
4102         if (!cpu_has_virtual_nmis()) {
4103                 /*
4104                  * Tracking the NMI-blocked state in software is built upon
4105                  * finding the next open IRQ window. This, in turn, depends on
4106                  * well-behaving guests: They have to keep IRQs disabled at
4107                  * least as long as the NMI handler runs. Otherwise we may
4108                  * cause NMI nesting, maybe breaking the guest. But as this is
4109                  * highly unlikely, we can live with the residual risk.
4110                  */
4111                 vmx->soft_vnmi_blocked = 1;
4112                 vmx->vnmi_blocked_time = 0;
4113         }
4114
4115         ++vcpu->stat.nmi_injections;
4116         vmx->nmi_known_unmasked = false;
4117         if (vmx->rmode.vm86_active) {
4118                 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
4119                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4120                 return;
4121         }
4122         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4123                         INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4124 }
4125
4126 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4127 {
4128         if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
4129                 return 0;
4130
4131         return  !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4132                   (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4133                    | GUEST_INTR_STATE_NMI));
4134 }
4135
4136 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4137 {
4138         if (!cpu_has_virtual_nmis())
4139                 return to_vmx(vcpu)->soft_vnmi_blocked;
4140         if (to_vmx(vcpu)->nmi_known_unmasked)
4141                 return false;
4142         return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4143 }
4144
4145 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4146 {
4147         struct vcpu_vmx *vmx = to_vmx(vcpu);
4148
4149         if (!cpu_has_virtual_nmis()) {
4150                 if (vmx->soft_vnmi_blocked != masked) {
4151                         vmx->soft_vnmi_blocked = masked;
4152                         vmx->vnmi_blocked_time = 0;
4153                 }
4154         } else {
4155                 vmx->nmi_known_unmasked = !masked;
4156                 if (masked)
4157                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4158                                       GUEST_INTR_STATE_NMI);
4159                 else
4160                         vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4161                                         GUEST_INTR_STATE_NMI);
4162         }
4163 }
4164
4165 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4166 {
4167         if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu)) {
4168                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4169                 if (to_vmx(vcpu)->nested.nested_run_pending ||
4170                     (vmcs12->idt_vectoring_info_field &
4171                      VECTORING_INFO_VALID_MASK))
4172                         return 0;
4173                 nested_vmx_vmexit(vcpu);
4174                 vmcs12->vm_exit_reason = EXIT_REASON_EXTERNAL_INTERRUPT;
4175                 vmcs12->vm_exit_intr_info = 0;
4176                 /* fall through to normal code, but now in L1, not L2 */
4177         }
4178
4179         return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4180                 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4181                         (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4182 }
4183
4184 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4185 {
4186         int ret;
4187         struct kvm_userspace_memory_region tss_mem = {
4188                 .slot = TSS_PRIVATE_MEMSLOT,
4189                 .guest_phys_addr = addr,
4190                 .memory_size = PAGE_SIZE * 3,
4191                 .flags = 0,
4192         };
4193
4194         ret = kvm_set_memory_region(kvm, &tss_mem, 0);
4195         if (ret)
4196                 return ret;
4197         kvm->arch.tss_addr = addr;
4198         if (!init_rmode_tss(kvm))
4199                 return  -ENOMEM;
4200
4201         return 0;
4202 }
4203
4204 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4205                                   int vec, u32 err_code)
4206 {
4207         /*
4208          * Instruction with address size override prefix opcode 0x67
4209          * Cause the #SS fault with 0 error code in VM86 mode.
4210          */
4211         if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
4212                 if (emulate_instruction(vcpu, 0) == EMULATE_DONE)
4213                         return 1;
4214         /*
4215          * Forward all other exceptions that are valid in real mode.
4216          * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4217          *        the required debugging infrastructure rework.
4218          */
4219         switch (vec) {
4220         case DB_VECTOR:
4221                 if (vcpu->guest_debug &
4222                     (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4223                         return 0;
4224                 kvm_queue_exception(vcpu, vec);
4225                 return 1;
4226         case BP_VECTOR:
4227                 /*
4228                  * Update instruction length as we may reinject the exception
4229                  * from user space while in guest debugging mode.
4230                  */
4231                 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4232                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4233                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4234                         return 0;
4235                 /* fall through */
4236         case DE_VECTOR:
4237         case OF_VECTOR:
4238         case BR_VECTOR:
4239         case UD_VECTOR:
4240         case DF_VECTOR:
4241         case SS_VECTOR:
4242         case GP_VECTOR:
4243         case MF_VECTOR:
4244                 kvm_queue_exception(vcpu, vec);
4245                 return 1;
4246         }
4247         return 0;
4248 }
4249
4250 /*
4251  * Trigger machine check on the host. We assume all the MSRs are already set up
4252  * by the CPU and that we still run on the same CPU as the MCE occurred on.
4253  * We pass a fake environment to the machine check handler because we want
4254  * the guest to be always treated like user space, no matter what context
4255  * it used internally.
4256  */
4257 static void kvm_machine_check(void)
4258 {
4259 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4260         struct pt_regs regs = {
4261                 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
4262                 .flags = X86_EFLAGS_IF,
4263         };
4264
4265         do_machine_check(&regs, 0);
4266 #endif
4267 }
4268
4269 static int handle_machine_check(struct kvm_vcpu *vcpu)
4270 {
4271         /* already handled by vcpu_run */
4272         return 1;
4273 }
4274
4275 static int handle_exception(struct kvm_vcpu *vcpu)
4276 {
4277         struct vcpu_vmx *vmx = to_vmx(vcpu);
4278         struct kvm_run *kvm_run = vcpu->run;
4279         u32 intr_info, ex_no, error_code;
4280         unsigned long cr2, rip, dr6;
4281         u32 vect_info;
4282         enum emulation_result er;
4283
4284         vect_info = vmx->idt_vectoring_info;
4285         intr_info = vmx->exit_intr_info;
4286
4287         if (is_machine_check(intr_info))
4288                 return handle_machine_check(vcpu);
4289
4290         if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4291             !is_page_fault(intr_info)) {
4292                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4293                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4294                 vcpu->run->internal.ndata = 2;
4295                 vcpu->run->internal.data[0] = vect_info;
4296                 vcpu->run->internal.data[1] = intr_info;
4297                 return 0;
4298         }
4299
4300         if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
4301                 return 1;  /* already handled by vmx_vcpu_run() */
4302
4303         if (is_no_device(intr_info)) {
4304                 vmx_fpu_activate(vcpu);
4305                 return 1;
4306         }
4307
4308         if (is_invalid_opcode(intr_info)) {
4309                 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
4310                 if (er != EMULATE_DONE)
4311                         kvm_queue_exception(vcpu, UD_VECTOR);
4312                 return 1;
4313         }
4314
4315         error_code = 0;
4316         if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4317                 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4318         if (is_page_fault(intr_info)) {
4319                 /* EPT won't cause page fault directly */
4320                 BUG_ON(enable_ept);
4321                 cr2 = vmcs_readl(EXIT_QUALIFICATION);
4322                 trace_kvm_page_fault(cr2, error_code);
4323
4324                 if (kvm_event_needs_reinjection(vcpu))
4325                         kvm_mmu_unprotect_page_virt(vcpu, cr2);
4326                 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
4327         }
4328
4329         if (vmx->rmode.vm86_active &&
4330             handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
4331                                                                 error_code)) {
4332                 if (vcpu->arch.halt_request) {
4333                         vcpu->arch.halt_request = 0;
4334                         return kvm_emulate_halt(vcpu);
4335                 }
4336                 return 1;
4337         }
4338
4339         ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4340         switch (ex_no) {
4341         case DB_VECTOR:
4342                 dr6 = vmcs_readl(EXIT_QUALIFICATION);
4343                 if (!(vcpu->guest_debug &
4344                       (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4345                         vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
4346                         kvm_queue_exception(vcpu, DB_VECTOR);
4347                         return 1;
4348                 }
4349                 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4350                 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4351                 /* fall through */
4352         case BP_VECTOR:
4353                 /*
4354                  * Update instruction length as we may reinject #BP from
4355                  * user space while in guest debugging mode. Reading it for
4356                  * #DB as well causes no harm, it is not used in that case.
4357                  */
4358                 vmx->vcpu.arch.event_exit_inst_len =
4359                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4360                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
4361                 rip = kvm_rip_read(vcpu);
4362                 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4363                 kvm_run->debug.arch.exception = ex_no;
4364                 break;
4365         default:
4366                 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4367                 kvm_run->ex.exception = ex_no;
4368                 kvm_run->ex.error_code = error_code;
4369                 break;
4370         }
4371         return 0;
4372 }
4373
4374 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
4375 {
4376         ++vcpu->stat.irq_exits;
4377         return 1;
4378 }
4379
4380 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4381 {
4382         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4383         return 0;
4384 }
4385
4386 static int handle_io(struct kvm_vcpu *vcpu)
4387 {
4388         unsigned long exit_qualification;
4389         int size, in, string;
4390         unsigned port;
4391
4392         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4393         string = (exit_qualification & 16) != 0;
4394         in = (exit_qualification & 8) != 0;
4395
4396         ++vcpu->stat.io_exits;
4397
4398         if (string || in)
4399                 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4400
4401         port = exit_qualification >> 16;
4402         size = (exit_qualification & 7) + 1;
4403         skip_emulated_instruction(vcpu);
4404
4405         return kvm_fast_pio_out(vcpu, size, port);
4406 }
4407
4408 static void
4409 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4410 {
4411         /*
4412          * Patch in the VMCALL instruction:
4413          */
4414         hypercall[0] = 0x0f;
4415         hypercall[1] = 0x01;
4416         hypercall[2] = 0xc1;
4417 }
4418
4419 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4420 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4421 {
4422         if (to_vmx(vcpu)->nested.vmxon &&
4423             ((val & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON))
4424                 return 1;
4425
4426         if (is_guest_mode(vcpu)) {
4427                 /*
4428                  * We get here when L2 changed cr0 in a way that did not change
4429                  * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4430                  * but did change L0 shadowed bits. This can currently happen
4431                  * with the TS bit: L0 may want to leave TS on (for lazy fpu
4432                  * loading) while pretending to allow the guest to change it.
4433                  */
4434                 if (kvm_set_cr0(vcpu, (val & vcpu->arch.cr0_guest_owned_bits) |
4435                          (vcpu->arch.cr0 & ~vcpu->arch.cr0_guest_owned_bits)))
4436                         return 1;
4437                 vmcs_writel(CR0_READ_SHADOW, val);
4438                 return 0;
4439         } else
4440                 return kvm_set_cr0(vcpu, val);
4441 }
4442
4443 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4444 {
4445         if (is_guest_mode(vcpu)) {
4446                 if (kvm_set_cr4(vcpu, (val & vcpu->arch.cr4_guest_owned_bits) |
4447                          (vcpu->arch.cr4 & ~vcpu->arch.cr4_guest_owned_bits)))
4448                         return 1;
4449                 vmcs_writel(CR4_READ_SHADOW, val);
4450                 return 0;
4451         } else
4452                 return kvm_set_cr4(vcpu, val);
4453 }
4454
4455 /* called to set cr0 as approriate for clts instruction exit. */
4456 static void handle_clts(struct kvm_vcpu *vcpu)
4457 {
4458         if (is_guest_mode(vcpu)) {
4459                 /*
4460                  * We get here when L2 did CLTS, and L1 didn't shadow CR0.TS
4461                  * but we did (!fpu_active). We need to keep GUEST_CR0.TS on,
4462                  * just pretend it's off (also in arch.cr0 for fpu_activate).
4463                  */
4464                 vmcs_writel(CR0_READ_SHADOW,
4465                         vmcs_readl(CR0_READ_SHADOW) & ~X86_CR0_TS);
4466                 vcpu->arch.cr0 &= ~X86_CR0_TS;
4467         } else
4468                 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4469 }
4470
4471 static int handle_cr(struct kvm_vcpu *vcpu)
4472 {
4473         unsigned long exit_qualification, val;
4474         int cr;
4475         int reg;
4476         int err;
4477
4478         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4479         cr = exit_qualification & 15;
4480         reg = (exit_qualification >> 8) & 15;
4481         switch ((exit_qualification >> 4) & 3) {
4482         case 0: /* mov to cr */
4483                 val = kvm_register_read(vcpu, reg);
4484                 trace_kvm_cr_write(cr, val);
4485                 switch (cr) {
4486                 case 0:
4487                         err = handle_set_cr0(vcpu, val);
4488                         kvm_complete_insn_gp(vcpu, err);
4489                         return 1;
4490                 case 3:
4491                         err = kvm_set_cr3(vcpu, val);
4492                         kvm_complete_insn_gp(vcpu, err);
4493                         return 1;
4494                 case 4:
4495                         err = handle_set_cr4(vcpu, val);
4496                         kvm_complete_insn_gp(vcpu, err);
4497                         return 1;
4498                 case 8: {
4499                                 u8 cr8_prev = kvm_get_cr8(vcpu);
4500                                 u8 cr8 = kvm_register_read(vcpu, reg);
4501                                 err = kvm_set_cr8(vcpu, cr8);
4502                                 kvm_complete_insn_gp(vcpu, err);
4503                                 if (irqchip_in_kernel(vcpu->kvm))
4504                                         return 1;
4505                                 if (cr8_prev <= cr8)
4506                                         return 1;
4507                                 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4508                                 return 0;
4509                         }
4510                 }
4511                 break;
4512         case 2: /* clts */
4513                 handle_clts(vcpu);
4514                 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4515                 skip_emulated_instruction(vcpu);
4516                 vmx_fpu_activate(vcpu);
4517                 return 1;
4518         case 1: /*mov from cr*/
4519                 switch (cr) {
4520                 case 3:
4521                         val = kvm_read_cr3(vcpu);
4522                         kvm_register_write(vcpu, reg, val);
4523                         trace_kvm_cr_read(cr, val);
4524                         skip_emulated_instruction(vcpu);
4525                         return 1;
4526                 case 8:
4527                         val = kvm_get_cr8(vcpu);
4528                         kvm_register_write(vcpu, reg, val);
4529                         trace_kvm_cr_read(cr, val);
4530                         skip_emulated_instruction(vcpu);
4531                         return 1;
4532                 }
4533                 break;
4534         case 3: /* lmsw */
4535                 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4536                 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
4537                 kvm_lmsw(vcpu, val);
4538
4539                 skip_emulated_instruction(vcpu);
4540                 return 1;
4541         default:
4542                 break;
4543         }
4544         vcpu->run->exit_reason = 0;
4545         vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
4546                (int)(exit_qualification >> 4) & 3, cr);
4547         return 0;
4548 }
4549
4550 static int handle_dr(struct kvm_vcpu *vcpu)
4551 {
4552         unsigned long exit_qualification;
4553         int dr, reg;
4554
4555         /* Do not handle if the CPL > 0, will trigger GP on re-entry */
4556         if (!kvm_require_cpl(vcpu, 0))
4557                 return 1;
4558         dr = vmcs_readl(GUEST_DR7);
4559         if (dr & DR7_GD) {
4560                 /*
4561                  * As the vm-exit takes precedence over the debug trap, we
4562                  * need to emulate the latter, either for the host or the
4563                  * guest debugging itself.
4564                  */
4565                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4566                         vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
4567                         vcpu->run->debug.arch.dr7 = dr;
4568                         vcpu->run->debug.arch.pc =
4569                                 vmcs_readl(GUEST_CS_BASE) +
4570                                 vmcs_readl(GUEST_RIP);
4571                         vcpu->run->debug.arch.exception = DB_VECTOR;
4572                         vcpu->run->exit_reason = KVM_EXIT_DEBUG;
4573                         return 0;
4574                 } else {
4575                         vcpu->arch.dr7 &= ~DR7_GD;
4576                         vcpu->arch.dr6 |= DR6_BD;
4577                         vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
4578                         kvm_queue_exception(vcpu, DB_VECTOR);
4579                         return 1;
4580                 }
4581         }
4582
4583         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4584         dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
4585         reg = DEBUG_REG_ACCESS_REG(exit_qualification);
4586         if (exit_qualification & TYPE_MOV_FROM_DR) {
4587                 unsigned long val;
4588                 if (!kvm_get_dr(vcpu, dr, &val))
4589                         kvm_register_write(vcpu, reg, val);
4590         } else
4591                 kvm_set_dr(vcpu, dr, vcpu->arch.regs[reg]);
4592         skip_emulated_instruction(vcpu);
4593         return 1;
4594 }
4595
4596 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
4597 {
4598         vmcs_writel(GUEST_DR7, val);
4599 }
4600
4601 static int handle_cpuid(struct kvm_vcpu *vcpu)
4602 {
4603         kvm_emulate_cpuid(vcpu);
4604         return 1;
4605 }
4606
4607 static int handle_rdmsr(struct kvm_vcpu *vcpu)
4608 {
4609         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
4610         u64 data;
4611
4612         if (vmx_get_msr(vcpu, ecx, &data)) {
4613                 trace_kvm_msr_read_ex(ecx);
4614                 kvm_inject_gp(vcpu, 0);
4615                 return 1;
4616         }
4617
4618         trace_kvm_msr_read(ecx, data);
4619
4620         /* FIXME: handling of bits 32:63 of rax, rdx */
4621         vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
4622         vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
4623         skip_emulated_instruction(vcpu);
4624         return 1;
4625 }
4626
4627 static int handle_wrmsr(struct kvm_vcpu *vcpu)
4628 {
4629         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
4630         u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
4631                 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
4632
4633         if (vmx_set_msr(vcpu, ecx, data) != 0) {
4634                 trace_kvm_msr_write_ex(ecx, data);
4635                 kvm_inject_gp(vcpu, 0);
4636                 return 1;
4637         }
4638
4639         trace_kvm_msr_write(ecx, data);
4640         skip_emulated_instruction(vcpu);
4641         return 1;
4642 }
4643
4644 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
4645 {
4646         kvm_make_request(KVM_REQ_EVENT, vcpu);
4647         return 1;
4648 }
4649
4650 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
4651 {
4652         u32 cpu_based_vm_exec_control;
4653
4654         /* clear pending irq */
4655         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4656         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
4657         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4658
4659         kvm_make_request(KVM_REQ_EVENT, vcpu);
4660
4661         ++vcpu->stat.irq_window_exits;
4662
4663         /*
4664          * If the user space waits to inject interrupts, exit as soon as
4665          * possible
4666          */
4667         if (!irqchip_in_kernel(vcpu->kvm) &&
4668             vcpu->run->request_interrupt_window &&
4669             !kvm_cpu_has_interrupt(vcpu)) {
4670                 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
4671                 return 0;
4672         }
4673         return 1;
4674 }
4675
4676 static int handle_halt(struct kvm_vcpu *vcpu)
4677 {
4678         skip_emulated_instruction(vcpu);
4679         return kvm_emulate_halt(vcpu);
4680 }
4681
4682 static int handle_vmcall(struct kvm_vcpu *vcpu)
4683 {
4684         skip_emulated_instruction(vcpu);
4685         kvm_emulate_hypercall(vcpu);
4686         return 1;
4687 }
4688
4689 static int handle_invd(struct kvm_vcpu *vcpu)
4690 {
4691         return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4692 }
4693
4694 static int handle_invlpg(struct kvm_vcpu *vcpu)
4695 {
4696         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4697
4698         kvm_mmu_invlpg(vcpu, exit_qualification);
4699         skip_emulated_instruction(vcpu);
4700         return 1;
4701 }
4702
4703 static int handle_rdpmc(struct kvm_vcpu *vcpu)
4704 {
4705         int err;
4706
4707         err = kvm_rdpmc(vcpu);
4708         kvm_complete_insn_gp(vcpu, err);
4709
4710         return 1;
4711 }
4712
4713 static int handle_wbinvd(struct kvm_vcpu *vcpu)
4714 {
4715         skip_emulated_instruction(vcpu);
4716         kvm_emulate_wbinvd(vcpu);
4717         return 1;
4718 }
4719
4720 static int handle_xsetbv(struct kvm_vcpu *vcpu)
4721 {
4722         u64 new_bv = kvm_read_edx_eax(vcpu);
4723         u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4724
4725         if (kvm_set_xcr(vcpu, index, new_bv) == 0)
4726                 skip_emulated_instruction(vcpu);
4727         return 1;
4728 }
4729
4730 static int handle_apic_access(struct kvm_vcpu *vcpu)
4731 {
4732         if (likely(fasteoi)) {
4733                 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4734                 int access_type, offset;
4735
4736                 access_type = exit_qualification & APIC_ACCESS_TYPE;
4737                 offset = exit_qualification & APIC_ACCESS_OFFSET;
4738                 /*
4739                  * Sane guest uses MOV to write EOI, with written value
4740                  * not cared. So make a short-circuit here by avoiding
4741                  * heavy instruction emulation.
4742                  */
4743                 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
4744                     (offset == APIC_EOI)) {
4745                         kvm_lapic_set_eoi(vcpu);
4746                         skip_emulated_instruction(vcpu);
4747                         return 1;
4748                 }
4749         }
4750         return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4751 }
4752
4753 static int handle_task_switch(struct kvm_vcpu *vcpu)
4754 {
4755         struct vcpu_vmx *vmx = to_vmx(vcpu);
4756         unsigned long exit_qualification;
4757         bool has_error_code = false;
4758         u32 error_code = 0;
4759         u16 tss_selector;
4760         int reason, type, idt_v, idt_index;
4761
4762         idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
4763         idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
4764         type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
4765
4766         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4767
4768         reason = (u32)exit_qualification >> 30;
4769         if (reason == TASK_SWITCH_GATE && idt_v) {
4770                 switch (type) {
4771                 case INTR_TYPE_NMI_INTR:
4772                         vcpu->arch.nmi_injected = false;
4773                         vmx_set_nmi_mask(vcpu, true);
4774                         break;
4775                 case INTR_TYPE_EXT_INTR:
4776                 case INTR_TYPE_SOFT_INTR:
4777                         kvm_clear_interrupt_queue(vcpu);
4778                         break;
4779                 case INTR_TYPE_HARD_EXCEPTION:
4780                         if (vmx->idt_vectoring_info &
4781                             VECTORING_INFO_DELIVER_CODE_MASK) {
4782                                 has_error_code = true;
4783                                 error_code =
4784                                         vmcs_read32(IDT_VECTORING_ERROR_CODE);
4785                         }
4786                         /* fall through */
4787                 case INTR_TYPE_SOFT_EXCEPTION:
4788                         kvm_clear_exception_queue(vcpu);
4789                         break;
4790                 default:
4791                         break;
4792                 }
4793         }
4794         tss_selector = exit_qualification;
4795
4796         if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
4797                        type != INTR_TYPE_EXT_INTR &&
4798                        type != INTR_TYPE_NMI_INTR))
4799                 skip_emulated_instruction(vcpu);
4800
4801         if (kvm_task_switch(vcpu, tss_selector,
4802                             type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
4803                             has_error_code, error_code) == EMULATE_FAIL) {
4804                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4805                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4806                 vcpu->run->internal.ndata = 0;
4807                 return 0;
4808         }
4809
4810         /* clear all local breakpoint enable flags */
4811         vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
4812
4813         /*
4814          * TODO: What about debug traps on tss switch?
4815          *       Are we supposed to inject them and update dr6?
4816          */
4817
4818         return 1;
4819 }
4820
4821 static int handle_ept_violation(struct kvm_vcpu *vcpu)
4822 {
4823         unsigned long exit_qualification;
4824         gpa_t gpa;
4825         u32 error_code;
4826         int gla_validity;
4827
4828         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4829
4830         if (exit_qualification & (1 << 6)) {
4831                 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
4832                 return -EINVAL;
4833         }
4834
4835         gla_validity = (exit_qualification >> 7) & 0x3;
4836         if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
4837                 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
4838                 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
4839                         (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
4840                         vmcs_readl(GUEST_LINEAR_ADDRESS));
4841                 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
4842                         (long unsigned int)exit_qualification);
4843                 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
4844                 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
4845                 return 0;
4846         }
4847
4848         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
4849         trace_kvm_page_fault(gpa, exit_qualification);
4850
4851         /* It is a write fault? */
4852         error_code = exit_qualification & (1U << 1);
4853         /* ept page table is present? */
4854         error_code |= (exit_qualification >> 3) & 0x1;
4855
4856         return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
4857 }
4858
4859 static u64 ept_rsvd_mask(u64 spte, int level)
4860 {
4861         int i;
4862         u64 mask = 0;
4863
4864         for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
4865                 mask |= (1ULL << i);
4866
4867         if (level > 2)
4868                 /* bits 7:3 reserved */
4869                 mask |= 0xf8;
4870         else if (level == 2) {
4871                 if (spte & (1ULL << 7))
4872                         /* 2MB ref, bits 20:12 reserved */
4873                         mask |= 0x1ff000;
4874                 else
4875                         /* bits 6:3 reserved */
4876                         mask |= 0x78;
4877         }
4878
4879         return mask;
4880 }
4881
4882 static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
4883                                        int level)
4884 {
4885         printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
4886
4887         /* 010b (write-only) */
4888         WARN_ON((spte & 0x7) == 0x2);
4889
4890         /* 110b (write/execute) */
4891         WARN_ON((spte & 0x7) == 0x6);
4892
4893         /* 100b (execute-only) and value not supported by logical processor */
4894         if (!cpu_has_vmx_ept_execute_only())
4895                 WARN_ON((spte & 0x7) == 0x4);
4896
4897         /* not 000b */
4898         if ((spte & 0x7)) {
4899                 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
4900
4901                 if (rsvd_bits != 0) {
4902                         printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
4903                                          __func__, rsvd_bits);
4904                         WARN_ON(1);
4905                 }
4906
4907                 if (level == 1 || (level == 2 && (spte & (1ULL << 7)))) {
4908                         u64 ept_mem_type = (spte & 0x38) >> 3;
4909
4910                         if (ept_mem_type == 2 || ept_mem_type == 3 ||
4911                             ept_mem_type == 7) {
4912                                 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
4913                                                 __func__, ept_mem_type);
4914                                 WARN_ON(1);
4915                         }
4916                 }
4917         }
4918 }
4919
4920 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
4921 {
4922         u64 sptes[4];
4923         int nr_sptes, i, ret;
4924         gpa_t gpa;
4925
4926         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
4927
4928         ret = handle_mmio_page_fault_common(vcpu, gpa, true);
4929         if (likely(ret == 1))
4930                 return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) ==
4931                                               EMULATE_DONE;
4932         if (unlikely(!ret))
4933                 return 1;
4934
4935         /* It is the real ept misconfig */
4936         printk(KERN_ERR "EPT: Misconfiguration.\n");
4937         printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
4938
4939         nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
4940
4941         for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
4942                 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
4943
4944         vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
4945         vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
4946
4947         return 0;
4948 }
4949
4950 static int handle_nmi_window(struct kvm_vcpu *vcpu)
4951 {
4952         u32 cpu_based_vm_exec_control;
4953
4954         /* clear pending NMI */
4955         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4956         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
4957         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4958         ++vcpu->stat.nmi_window_exits;
4959         kvm_make_request(KVM_REQ_EVENT, vcpu);
4960
4961         return 1;
4962 }
4963
4964 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
4965 {
4966         struct vcpu_vmx *vmx = to_vmx(vcpu);
4967         enum emulation_result err = EMULATE_DONE;
4968         int ret = 1;
4969         u32 cpu_exec_ctrl;
4970         bool intr_window_requested;
4971         unsigned count = 130;
4972
4973         cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4974         intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
4975
4976         while (!guest_state_valid(vcpu) && count-- != 0) {
4977                 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
4978                         return handle_interrupt_window(&vmx->vcpu);
4979
4980                 if (test_bit(KVM_REQ_EVENT, &vcpu->requests))
4981                         return 1;
4982
4983                 err = emulate_instruction(vcpu, 0);
4984
4985                 if (err == EMULATE_DO_MMIO) {
4986                         ret = 0;
4987                         goto out;
4988                 }
4989
4990                 if (err != EMULATE_DONE) {
4991                         vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4992                         vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4993                         vcpu->run->internal.ndata = 0;
4994                         return 0;
4995                 }
4996
4997                 if (signal_pending(current))
4998                         goto out;
4999                 if (need_resched())
5000                         schedule();
5001         }
5002
5003         vmx->emulation_required = !guest_state_valid(vcpu);
5004 out:
5005         return ret;
5006 }
5007
5008 /*
5009  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5010  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5011  */
5012 static int handle_pause(struct kvm_vcpu *vcpu)
5013 {
5014         skip_emulated_instruction(vcpu);
5015         kvm_vcpu_on_spin(vcpu);
5016
5017         return 1;
5018 }
5019
5020 static int handle_invalid_op(struct kvm_vcpu *vcpu)
5021 {
5022         kvm_queue_exception(vcpu, UD_VECTOR);
5023         return 1;
5024 }
5025
5026 /*
5027  * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
5028  * We could reuse a single VMCS for all the L2 guests, but we also want the
5029  * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
5030  * allows keeping them loaded on the processor, and in the future will allow
5031  * optimizations where prepare_vmcs02 doesn't need to set all the fields on
5032  * every entry if they never change.
5033  * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
5034  * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
5035  *
5036  * The following functions allocate and free a vmcs02 in this pool.
5037  */
5038
5039 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
5040 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
5041 {
5042         struct vmcs02_list *item;
5043         list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
5044                 if (item->vmptr == vmx->nested.current_vmptr) {
5045                         list_move(&item->list, &vmx->nested.vmcs02_pool);
5046                         return &item->vmcs02;
5047                 }
5048
5049         if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
5050                 /* Recycle the least recently used VMCS. */
5051                 item = list_entry(vmx->nested.vmcs02_pool.prev,
5052                         struct vmcs02_list, list);
5053                 item->vmptr = vmx->nested.current_vmptr;
5054                 list_move(&item->list, &vmx->nested.vmcs02_pool);
5055                 return &item->vmcs02;
5056         }
5057
5058         /* Create a new VMCS */
5059         item = (struct vmcs02_list *)
5060                 kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
5061         if (!item)
5062                 return NULL;
5063         item->vmcs02.vmcs = alloc_vmcs();
5064         if (!item->vmcs02.vmcs) {
5065                 kfree(item);
5066                 return NULL;
5067         }
5068         loaded_vmcs_init(&item->vmcs02);
5069         item->vmptr = vmx->nested.current_vmptr;
5070         list_add(&(item->list), &(vmx->nested.vmcs02_pool));
5071         vmx->nested.vmcs02_num++;
5072         return &item->vmcs02;
5073 }
5074
5075 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
5076 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
5077 {
5078         struct vmcs02_list *item;
5079         list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
5080                 if (item->vmptr == vmptr) {
5081                         free_loaded_vmcs(&item->vmcs02);
5082                         list_del(&item->list);
5083                         kfree(item);
5084                         vmx->nested.vmcs02_num--;
5085                         return;
5086                 }
5087 }
5088
5089 /*
5090  * Free all VMCSs saved for this vcpu, except the one pointed by
5091  * vmx->loaded_vmcs. These include the VMCSs in vmcs02_pool (except the one
5092  * currently used, if running L2), and vmcs01 when running L2.
5093  */
5094 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
5095 {
5096         struct vmcs02_list *item, *n;
5097         list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
5098                 if (vmx->loaded_vmcs != &item->vmcs02)
5099                         free_loaded_vmcs(&item->vmcs02);
5100                 list_del(&item->list);
5101                 kfree(item);
5102         }
5103         vmx->nested.vmcs02_num = 0;
5104
5105         if (vmx->loaded_vmcs != &vmx->vmcs01)
5106                 free_loaded_vmcs(&vmx->vmcs01);
5107 }
5108
5109 /*
5110  * Emulate the VMXON instruction.
5111  * Currently, we just remember that VMX is active, and do not save or even
5112  * inspect the argument to VMXON (the so-called "VMXON pointer") because we
5113  * do not currently need to store anything in that guest-allocated memory
5114  * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
5115  * argument is different from the VMXON pointer (which the spec says they do).
5116  */
5117 static int handle_vmon(struct kvm_vcpu *vcpu)
5118 {
5119         struct kvm_segment cs;
5120         struct vcpu_vmx *vmx = to_vmx(vcpu);
5121
5122         /* The Intel VMX Instruction Reference lists a bunch of bits that
5123          * are prerequisite to running VMXON, most notably cr4.VMXE must be
5124          * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
5125          * Otherwise, we should fail with #UD. We test these now:
5126          */
5127         if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
5128             !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
5129             (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
5130                 kvm_queue_exception(vcpu, UD_VECTOR);
5131                 return 1;
5132         }
5133
5134         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
5135         if (is_long_mode(vcpu) && !cs.l) {
5136                 kvm_queue_exception(vcpu, UD_VECTOR);
5137                 return 1;
5138         }
5139
5140         if (vmx_get_cpl(vcpu)) {
5141                 kvm_inject_gp(vcpu, 0);
5142                 return 1;
5143         }
5144
5145         INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
5146         vmx->nested.vmcs02_num = 0;
5147
5148         vmx->nested.vmxon = true;
5149
5150         skip_emulated_instruction(vcpu);
5151         return 1;
5152 }
5153
5154 /*
5155  * Intel's VMX Instruction Reference specifies a common set of prerequisites
5156  * for running VMX instructions (except VMXON, whose prerequisites are
5157  * slightly different). It also specifies what exception to inject otherwise.
5158  */
5159 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
5160 {
5161         struct kvm_segment cs;
5162         struct vcpu_vmx *vmx = to_vmx(vcpu);
5163
5164         if (!vmx->nested.vmxon) {
5165                 kvm_queue_exception(vcpu, UD_VECTOR);
5166                 return 0;
5167         }
5168
5169         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
5170         if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
5171             (is_long_mode(vcpu) && !cs.l)) {
5172                 kvm_queue_exception(vcpu, UD_VECTOR);
5173                 return 0;
5174         }
5175
5176         if (vmx_get_cpl(vcpu)) {
5177                 kvm_inject_gp(vcpu, 0);
5178                 return 0;
5179         }
5180
5181         return 1;
5182 }
5183
5184 /*
5185  * Free whatever needs to be freed from vmx->nested when L1 goes down, or
5186  * just stops using VMX.
5187  */
5188 static void free_nested(struct vcpu_vmx *vmx)
5189 {
5190         if (!vmx->nested.vmxon)
5191                 return;
5192         vmx->nested.vmxon = false;
5193         if (vmx->nested.current_vmptr != -1ull) {
5194                 kunmap(vmx->nested.current_vmcs12_page);
5195                 nested_release_page(vmx->nested.current_vmcs12_page);
5196                 vmx->nested.current_vmptr = -1ull;
5197                 vmx->nested.current_vmcs12 = NULL;
5198         }
5199         /* Unpin physical memory we referred to in current vmcs02 */
5200         if (vmx->nested.apic_access_page) {
5201                 nested_release_page(vmx->nested.apic_access_page);
5202                 vmx->nested.apic_access_page = 0;
5203         }
5204
5205         nested_free_all_saved_vmcss(vmx);
5206 }
5207
5208 /* Emulate the VMXOFF instruction */
5209 static int handle_vmoff(struct kvm_vcpu *vcpu)
5210 {
5211         if (!nested_vmx_check_permission(vcpu))
5212                 return 1;
5213         free_nested(to_vmx(vcpu));
5214         skip_emulated_instruction(vcpu);
5215         return 1;
5216 }
5217
5218 /*
5219  * Decode the memory-address operand of a vmx instruction, as recorded on an
5220  * exit caused by such an instruction (run by a guest hypervisor).
5221  * On success, returns 0. When the operand is invalid, returns 1 and throws
5222  * #UD or #GP.
5223  */
5224 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
5225                                  unsigned long exit_qualification,
5226                                  u32 vmx_instruction_info, gva_t *ret)
5227 {
5228         /*
5229          * According to Vol. 3B, "Information for VM Exits Due to Instruction
5230          * Execution", on an exit, vmx_instruction_info holds most of the
5231          * addressing components of the operand. Only the displacement part
5232          * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
5233          * For how an actual address is calculated from all these components,
5234          * refer to Vol. 1, "Operand Addressing".
5235          */
5236         int  scaling = vmx_instruction_info & 3;
5237         int  addr_size = (vmx_instruction_info >> 7) & 7;
5238         bool is_reg = vmx_instruction_info & (1u << 10);
5239         int  seg_reg = (vmx_instruction_info >> 15) & 7;
5240         int  index_reg = (vmx_instruction_info >> 18) & 0xf;
5241         bool index_is_valid = !(vmx_instruction_info & (1u << 22));
5242         int  base_reg       = (vmx_instruction_info >> 23) & 0xf;
5243         bool base_is_valid  = !(vmx_instruction_info & (1u << 27));
5244
5245         if (is_reg) {
5246                 kvm_queue_exception(vcpu, UD_VECTOR);
5247                 return 1;
5248         }
5249
5250         /* Addr = segment_base + offset */
5251         /* offset = base + [index * scale] + displacement */
5252         *ret = vmx_get_segment_base(vcpu, seg_reg);
5253         if (base_is_valid)
5254                 *ret += kvm_register_read(vcpu, base_reg);
5255         if (index_is_valid)
5256                 *ret += kvm_register_read(vcpu, index_reg)<<scaling;
5257         *ret += exit_qualification; /* holds the displacement */
5258
5259         if (addr_size == 1) /* 32 bit */
5260                 *ret &= 0xffffffff;
5261
5262         /*
5263          * TODO: throw #GP (and return 1) in various cases that the VM*
5264          * instructions require it - e.g., offset beyond segment limit,
5265          * unusable or unreadable/unwritable segment, non-canonical 64-bit
5266          * address, and so on. Currently these are not checked.
5267          */
5268         return 0;
5269 }
5270
5271 /*
5272  * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
5273  * set the success or error code of an emulated VMX instruction, as specified
5274  * by Vol 2B, VMX Instruction Reference, "Conventions".
5275  */
5276 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
5277 {
5278         vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
5279                         & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
5280                             X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
5281 }
5282
5283 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
5284 {
5285         vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
5286                         & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
5287                             X86_EFLAGS_SF | X86_EFLAGS_OF))
5288                         | X86_EFLAGS_CF);
5289 }
5290
5291 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
5292                                         u32 vm_instruction_error)
5293 {
5294         if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
5295                 /*
5296                  * failValid writes the error number to the current VMCS, which
5297                  * can't be done there isn't a current VMCS.
5298                  */
5299                 nested_vmx_failInvalid(vcpu);
5300                 return;
5301         }
5302         vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
5303                         & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
5304                             X86_EFLAGS_SF | X86_EFLAGS_OF))
5305                         | X86_EFLAGS_ZF);
5306         get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
5307 }
5308
5309 /* Emulate the VMCLEAR instruction */
5310 static int handle_vmclear(struct kvm_vcpu *vcpu)
5311 {
5312         struct vcpu_vmx *vmx = to_vmx(vcpu);
5313         gva_t gva;
5314         gpa_t vmptr;
5315         struct vmcs12 *vmcs12;
5316         struct page *page;
5317         struct x86_exception e;
5318
5319         if (!nested_vmx_check_permission(vcpu))
5320                 return 1;
5321
5322         if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5323                         vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
5324                 return 1;
5325
5326         if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
5327                                 sizeof(vmptr), &e)) {
5328                 kvm_inject_page_fault(vcpu, &e);
5329                 return 1;
5330         }
5331
5332         if (!IS_ALIGNED(vmptr, PAGE_SIZE)) {
5333                 nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
5334                 skip_emulated_instruction(vcpu);
5335                 return 1;
5336         }
5337
5338         if (vmptr == vmx->nested.current_vmptr) {
5339                 kunmap(vmx->nested.current_vmcs12_page);
5340                 nested_release_page(vmx->nested.current_vmcs12_page);
5341                 vmx->nested.current_vmptr = -1ull;
5342                 vmx->nested.current_vmcs12 = NULL;
5343         }
5344
5345         page = nested_get_page(vcpu, vmptr);
5346         if (page == NULL) {
5347                 /*
5348                  * For accurate processor emulation, VMCLEAR beyond available
5349                  * physical memory should do nothing at all. However, it is
5350                  * possible that a nested vmx bug, not a guest hypervisor bug,
5351                  * resulted in this case, so let's shut down before doing any
5352                  * more damage:
5353                  */
5354                 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5355                 return 1;
5356         }
5357         vmcs12 = kmap(page);
5358         vmcs12->launch_state = 0;
5359         kunmap(page);
5360         nested_release_page(page);
5361
5362         nested_free_vmcs02(vmx, vmptr);
5363
5364         skip_emulated_instruction(vcpu);
5365         nested_vmx_succeed(vcpu);
5366         return 1;
5367 }
5368
5369 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
5370
5371 /* Emulate the VMLAUNCH instruction */
5372 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
5373 {
5374         return nested_vmx_run(vcpu, true);
5375 }
5376
5377 /* Emulate the VMRESUME instruction */
5378 static int handle_vmresume(struct kvm_vcpu *vcpu)
5379 {
5380
5381         return nested_vmx_run(vcpu, false);
5382 }
5383
5384 enum vmcs_field_type {
5385         VMCS_FIELD_TYPE_U16 = 0,
5386         VMCS_FIELD_TYPE_U64 = 1,
5387         VMCS_FIELD_TYPE_U32 = 2,
5388         VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
5389 };
5390
5391 static inline int vmcs_field_type(unsigned long field)
5392 {
5393         if (0x1 & field)        /* the *_HIGH fields are all 32 bit */
5394                 return VMCS_FIELD_TYPE_U32;
5395         return (field >> 13) & 0x3 ;
5396 }
5397
5398 static inline int vmcs_field_readonly(unsigned long field)
5399 {
5400         return (((field >> 10) & 0x3) == 1);
5401 }
5402
5403 /*
5404  * Read a vmcs12 field. Since these can have varying lengths and we return
5405  * one type, we chose the biggest type (u64) and zero-extend the return value
5406  * to that size. Note that the caller, handle_vmread, might need to use only
5407  * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
5408  * 64-bit fields are to be returned).
5409  */
5410 static inline bool vmcs12_read_any(struct kvm_vcpu *vcpu,
5411                                         unsigned long field, u64 *ret)
5412 {
5413         short offset = vmcs_field_to_offset(field);
5414         char *p;
5415
5416         if (offset < 0)
5417                 return 0;
5418
5419         p = ((char *)(get_vmcs12(vcpu))) + offset;
5420
5421         switch (vmcs_field_type(field)) {
5422         case VMCS_FIELD_TYPE_NATURAL_WIDTH:
5423                 *ret = *((natural_width *)p);
5424                 return 1;
5425         case VMCS_FIELD_TYPE_U16:
5426                 *ret = *((u16 *)p);
5427                 return 1;
5428         case VMCS_FIELD_TYPE_U32:
5429                 *ret = *((u32 *)p);
5430                 return 1;
5431         case VMCS_FIELD_TYPE_U64:
5432                 *ret = *((u64 *)p);
5433                 return 1;
5434         default:
5435                 return 0; /* can never happen. */
5436         }
5437 }
5438
5439 /*
5440  * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
5441  * used before) all generate the same failure when it is missing.
5442  */
5443 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
5444 {
5445         struct vcpu_vmx *vmx = to_vmx(vcpu);
5446         if (vmx->nested.current_vmptr == -1ull) {
5447                 nested_vmx_failInvalid(vcpu);
5448                 skip_emulated_instruction(vcpu);
5449                 return 0;
5450         }
5451         return 1;
5452 }
5453
5454 static int handle_vmread(struct kvm_vcpu *vcpu)
5455 {
5456         unsigned long field;
5457         u64 field_value;
5458         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5459         u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5460         gva_t gva = 0;
5461
5462         if (!nested_vmx_check_permission(vcpu) ||
5463             !nested_vmx_check_vmcs12(vcpu))
5464                 return 1;
5465
5466         /* Decode instruction info and find the field to read */
5467         field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5468         /* Read the field, zero-extended to a u64 field_value */
5469         if (!vmcs12_read_any(vcpu, field, &field_value)) {
5470                 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5471                 skip_emulated_instruction(vcpu);
5472                 return 1;
5473         }
5474         /*
5475          * Now copy part of this value to register or memory, as requested.
5476          * Note that the number of bits actually copied is 32 or 64 depending
5477          * on the guest's mode (32 or 64 bit), not on the given field's length.
5478          */
5479         if (vmx_instruction_info & (1u << 10)) {
5480                 kvm_register_write(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
5481                         field_value);
5482         } else {
5483                 if (get_vmx_mem_address(vcpu, exit_qualification,
5484                                 vmx_instruction_info, &gva))
5485                         return 1;
5486                 /* _system ok, as nested_vmx_check_permission verified cpl=0 */
5487                 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
5488                              &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
5489         }
5490
5491         nested_vmx_succeed(vcpu);
5492         skip_emulated_instruction(vcpu);
5493         return 1;
5494 }
5495
5496
5497 static int handle_vmwrite(struct kvm_vcpu *vcpu)
5498 {
5499         unsigned long field;
5500         gva_t gva;
5501         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5502         u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5503         char *p;
5504         short offset;
5505         /* The value to write might be 32 or 64 bits, depending on L1's long
5506          * mode, and eventually we need to write that into a field of several
5507          * possible lengths. The code below first zero-extends the value to 64
5508          * bit (field_value), and then copies only the approriate number of
5509          * bits into the vmcs12 field.
5510          */
5511         u64 field_value = 0;
5512         struct x86_exception e;
5513
5514         if (!nested_vmx_check_permission(vcpu) ||
5515             !nested_vmx_check_vmcs12(vcpu))
5516                 return 1;
5517
5518         if (vmx_instruction_info & (1u << 10))
5519                 field_value = kvm_register_read(vcpu,
5520                         (((vmx_instruction_info) >> 3) & 0xf));
5521         else {
5522                 if (get_vmx_mem_address(vcpu, exit_qualification,
5523                                 vmx_instruction_info, &gva))
5524                         return 1;
5525                 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
5526                            &field_value, (is_long_mode(vcpu) ? 8 : 4), &e)) {
5527                         kvm_inject_page_fault(vcpu, &e);
5528                         return 1;
5529                 }
5530         }
5531
5532
5533         field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5534         if (vmcs_field_readonly(field)) {
5535                 nested_vmx_failValid(vcpu,
5536                         VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
5537                 skip_emulated_instruction(vcpu);
5538                 return 1;
5539         }
5540
5541         offset = vmcs_field_to_offset(field);
5542         if (offset < 0) {
5543                 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5544                 skip_emulated_instruction(vcpu);
5545                 return 1;
5546         }
5547         p = ((char *) get_vmcs12(vcpu)) + offset;
5548
5549         switch (vmcs_field_type(field)) {
5550         case VMCS_FIELD_TYPE_U16:
5551                 *(u16 *)p = field_value;
5552                 break;
5553         case VMCS_FIELD_TYPE_U32:
5554                 *(u32 *)p = field_value;
5555                 break;
5556         case VMCS_FIELD_TYPE_U64:
5557                 *(u64 *)p = field_value;
5558                 break;
5559         case VMCS_FIELD_TYPE_NATURAL_WIDTH:
5560                 *(natural_width *)p = field_value;
5561                 break;
5562         default:
5563                 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5564                 skip_emulated_instruction(vcpu);
5565                 return 1;
5566         }
5567
5568         nested_vmx_succeed(vcpu);
5569         skip_emulated_instruction(vcpu);
5570         return 1;
5571 }
5572
5573 /* Emulate the VMPTRLD instruction */
5574 static int handle_vmptrld(struct kvm_vcpu *vcpu)
5575 {
5576         struct vcpu_vmx *vmx = to_vmx(vcpu);
5577         gva_t gva;
5578         gpa_t vmptr;
5579         struct x86_exception e;
5580
5581         if (!nested_vmx_check_permission(vcpu))
5582                 return 1;
5583
5584         if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5585                         vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
5586                 return 1;
5587
5588         if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
5589                                 sizeof(vmptr), &e)) {
5590                 kvm_inject_page_fault(vcpu, &e);
5591                 return 1;
5592         }
5593
5594         if (!IS_ALIGNED(vmptr, PAGE_SIZE)) {
5595                 nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
5596                 skip_emulated_instruction(vcpu);
5597                 return 1;
5598         }
5599
5600         if (vmx->nested.current_vmptr != vmptr) {
5601                 struct vmcs12 *new_vmcs12;
5602                 struct page *page;
5603                 page = nested_get_page(vcpu, vmptr);
5604                 if (page == NULL) {
5605                         nested_vmx_failInvalid(vcpu);
5606                         skip_emulated_instruction(vcpu);
5607                         return 1;
5608                 }
5609                 new_vmcs12 = kmap(page);
5610                 if (new_vmcs12->revision_id != VMCS12_REVISION) {
5611                         kunmap(page);
5612                         nested_release_page_clean(page);
5613                         nested_vmx_failValid(vcpu,
5614                                 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5615                         skip_emulated_instruction(vcpu);
5616                         return 1;
5617                 }
5618                 if (vmx->nested.current_vmptr != -1ull) {
5619                         kunmap(vmx->nested.current_vmcs12_page);
5620                         nested_release_page(vmx->nested.current_vmcs12_page);
5621                 }
5622
5623                 vmx->nested.current_vmptr = vmptr;
5624                 vmx->nested.current_vmcs12 = new_vmcs12;
5625                 vmx->nested.current_vmcs12_page = page;
5626         }
5627
5628         nested_vmx_succeed(vcpu);
5629         skip_emulated_instruction(vcpu);
5630         return 1;
5631 }
5632
5633 /* Emulate the VMPTRST instruction */
5634 static int handle_vmptrst(struct kvm_vcpu *vcpu)
5635 {
5636         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5637         u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5638         gva_t vmcs_gva;
5639         struct x86_exception e;
5640
5641         if (!nested_vmx_check_permission(vcpu))
5642                 return 1;
5643
5644         if (get_vmx_mem_address(vcpu, exit_qualification,
5645                         vmx_instruction_info, &vmcs_gva))
5646                 return 1;
5647         /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
5648         if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
5649                                  (void *)&to_vmx(vcpu)->nested.current_vmptr,
5650                                  sizeof(u64), &e)) {
5651                 kvm_inject_page_fault(vcpu, &e);
5652                 return 1;
5653         }
5654         nested_vmx_succeed(vcpu);
5655         skip_emulated_instruction(vcpu);
5656         return 1;
5657 }
5658
5659 /*
5660  * The exit handlers return 1 if the exit was handled fully and guest execution
5661  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
5662  * to be done to userspace and return 0.
5663  */
5664 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5665         [EXIT_REASON_EXCEPTION_NMI]           = handle_exception,
5666         [EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
5667         [EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
5668         [EXIT_REASON_NMI_WINDOW]              = handle_nmi_window,
5669         [EXIT_REASON_IO_INSTRUCTION]          = handle_io,
5670         [EXIT_REASON_CR_ACCESS]               = handle_cr,
5671         [EXIT_REASON_DR_ACCESS]               = handle_dr,
5672         [EXIT_REASON_CPUID]                   = handle_cpuid,
5673         [EXIT_REASON_MSR_READ]                = handle_rdmsr,
5674         [EXIT_REASON_MSR_WRITE]               = handle_wrmsr,
5675         [EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
5676         [EXIT_REASON_HLT]                     = handle_halt,
5677         [EXIT_REASON_INVD]                    = handle_invd,
5678         [EXIT_REASON_INVLPG]                  = handle_invlpg,
5679         [EXIT_REASON_RDPMC]                   = handle_rdpmc,
5680         [EXIT_REASON_VMCALL]                  = handle_vmcall,
5681         [EXIT_REASON_VMCLEAR]                 = handle_vmclear,
5682         [EXIT_REASON_VMLAUNCH]                = handle_vmlaunch,
5683         [EXIT_REASON_VMPTRLD]                 = handle_vmptrld,
5684         [EXIT_REASON_VMPTRST]                 = handle_vmptrst,
5685         [EXIT_REASON_VMREAD]                  = handle_vmread,
5686         [EXIT_REASON_VMRESUME]                = handle_vmresume,
5687         [EXIT_REASON_VMWRITE]                 = handle_vmwrite,
5688         [EXIT_REASON_VMOFF]                   = handle_vmoff,
5689         [EXIT_REASON_VMON]                    = handle_vmon,
5690         [EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
5691         [EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
5692         [EXIT_REASON_WBINVD]                  = handle_wbinvd,
5693         [EXIT_REASON_XSETBV]                  = handle_xsetbv,
5694         [EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
5695         [EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
5696         [EXIT_REASON_EPT_VIOLATION]           = handle_ept_violation,
5697         [EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
5698         [EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
5699         [EXIT_REASON_MWAIT_INSTRUCTION]       = handle_invalid_op,
5700         [EXIT_REASON_MONITOR_INSTRUCTION]     = handle_invalid_op,
5701 };
5702
5703 static const int kvm_vmx_max_exit_handlers =
5704         ARRAY_SIZE(kvm_vmx_exit_handlers);
5705
5706 /*
5707  * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
5708  * rather than handle it ourselves in L0. I.e., check whether L1 expressed
5709  * disinterest in the current event (read or write a specific MSR) by using an
5710  * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
5711  */
5712 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
5713         struct vmcs12 *vmcs12, u32 exit_reason)
5714 {
5715         u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
5716         gpa_t bitmap;
5717
5718         if (!nested_cpu_has(get_vmcs12(vcpu), CPU_BASED_USE_MSR_BITMAPS))
5719                 return 1;
5720
5721         /*
5722          * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
5723          * for the four combinations of read/write and low/high MSR numbers.
5724          * First we need to figure out which of the four to use:
5725          */
5726         bitmap = vmcs12->msr_bitmap;
5727         if (exit_reason == EXIT_REASON_MSR_WRITE)
5728                 bitmap += 2048;
5729         if (msr_index >= 0xc0000000) {
5730                 msr_index -= 0xc0000000;
5731                 bitmap += 1024;
5732         }
5733
5734         /* Then read the msr_index'th bit from this bitmap: */
5735         if (msr_index < 1024*8) {
5736                 unsigned char b;
5737                 kvm_read_guest(vcpu->kvm, bitmap + msr_index/8, &b, 1);
5738                 return 1 & (b >> (msr_index & 7));
5739         } else
5740                 return 1; /* let L1 handle the wrong parameter */
5741 }
5742
5743 /*
5744  * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
5745  * rather than handle it ourselves in L0. I.e., check if L1 wanted to
5746  * intercept (via guest_host_mask etc.) the current event.
5747  */
5748 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
5749         struct vmcs12 *vmcs12)
5750 {
5751         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5752         int cr = exit_qualification & 15;
5753         int reg = (exit_qualification >> 8) & 15;
5754         unsigned long val = kvm_register_read(vcpu, reg);
5755
5756         switch ((exit_qualification >> 4) & 3) {
5757         case 0: /* mov to cr */
5758                 switch (cr) {
5759                 case 0:
5760                         if (vmcs12->cr0_guest_host_mask &
5761                             (val ^ vmcs12->cr0_read_shadow))
5762                                 return 1;
5763                         break;
5764                 case 3:
5765                         if ((vmcs12->cr3_target_count >= 1 &&
5766                                         vmcs12->cr3_target_value0 == val) ||
5767                                 (vmcs12->cr3_target_count >= 2 &&
5768                                         vmcs12->cr3_target_value1 == val) ||
5769                                 (vmcs12->cr3_target_count >= 3 &&
5770                                         vmcs12->cr3_target_value2 == val) ||
5771                                 (vmcs12->cr3_target_count >= 4 &&
5772                                         vmcs12->cr3_target_value3 == val))
5773                                 return 0;
5774                         if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
5775                                 return 1;
5776                         break;
5777                 case 4:
5778                         if (vmcs12->cr4_guest_host_mask &
5779                             (vmcs12->cr4_read_shadow ^ val))
5780                                 return 1;
5781                         break;
5782                 case 8:
5783                         if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
5784                                 return 1;
5785                         break;
5786                 }
5787                 break;
5788         case 2: /* clts */
5789                 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
5790                     (vmcs12->cr0_read_shadow & X86_CR0_TS))
5791                         return 1;
5792                 break;
5793         case 1: /* mov from cr */
5794                 switch (cr) {
5795                 case 3:
5796                         if (vmcs12->cpu_based_vm_exec_control &
5797                             CPU_BASED_CR3_STORE_EXITING)
5798                                 return 1;
5799                         break;
5800                 case 8:
5801                         if (vmcs12->cpu_based_vm_exec_control &
5802                             CPU_BASED_CR8_STORE_EXITING)
5803                                 return 1;
5804                         break;
5805                 }
5806                 break;
5807         case 3: /* lmsw */
5808                 /*
5809                  * lmsw can change bits 1..3 of cr0, and only set bit 0 of
5810                  * cr0. Other attempted changes are ignored, with no exit.
5811                  */
5812                 if (vmcs12->cr0_guest_host_mask & 0xe &
5813                     (val ^ vmcs12->cr0_read_shadow))
5814                         return 1;
5815                 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
5816                     !(vmcs12->cr0_read_shadow & 0x1) &&
5817                     (val & 0x1))
5818                         return 1;
5819                 break;
5820         }
5821         return 0;
5822 }
5823
5824 /*
5825  * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
5826  * should handle it ourselves in L0 (and then continue L2). Only call this
5827  * when in is_guest_mode (L2).
5828  */
5829 static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu)
5830 {
5831         u32 exit_reason = vmcs_read32(VM_EXIT_REASON);
5832         u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
5833         struct vcpu_vmx *vmx = to_vmx(vcpu);
5834         struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5835
5836         if (vmx->nested.nested_run_pending)
5837                 return 0;
5838
5839         if (unlikely(vmx->fail)) {
5840                 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
5841                                     vmcs_read32(VM_INSTRUCTION_ERROR));
5842                 return 1;
5843         }
5844
5845         switch (exit_reason) {
5846         case EXIT_REASON_EXCEPTION_NMI:
5847                 if (!is_exception(intr_info))
5848                         return 0;
5849                 else if (is_page_fault(intr_info))
5850                         return enable_ept;
5851                 return vmcs12->exception_bitmap &
5852                                 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
5853         case EXIT_REASON_EXTERNAL_INTERRUPT:
5854                 return 0;
5855         case EXIT_REASON_TRIPLE_FAULT:
5856                 return 1;
5857         case EXIT_REASON_PENDING_INTERRUPT:
5858         case EXIT_REASON_NMI_WINDOW:
5859                 /*
5860                  * prepare_vmcs02() set the CPU_BASED_VIRTUAL_INTR_PENDING bit
5861                  * (aka Interrupt Window Exiting) only when L1 turned it on,
5862                  * so if we got a PENDING_INTERRUPT exit, this must be for L1.
5863                  * Same for NMI Window Exiting.
5864                  */
5865                 return 1;
5866         case EXIT_REASON_TASK_SWITCH:
5867                 return 1;
5868         case EXIT_REASON_CPUID:
5869                 return 1;
5870         case EXIT_REASON_HLT:
5871                 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
5872         case EXIT_REASON_INVD:
5873                 return 1;
5874         case EXIT_REASON_INVLPG:
5875                 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5876         case EXIT_REASON_RDPMC:
5877                 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
5878         case EXIT_REASON_RDTSC:
5879                 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
5880         case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
5881         case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
5882         case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
5883         case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
5884         case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
5885                 /*
5886                  * VMX instructions trap unconditionally. This allows L1 to
5887                  * emulate them for its L2 guest, i.e., allows 3-level nesting!
5888                  */
5889                 return 1;
5890         case EXIT_REASON_CR_ACCESS:
5891                 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
5892         case EXIT_REASON_DR_ACCESS:
5893                 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
5894         case EXIT_REASON_IO_INSTRUCTION:
5895                 /* TODO: support IO bitmaps */
5896                 return 1;
5897         case EXIT_REASON_MSR_READ:
5898         case EXIT_REASON_MSR_WRITE:
5899                 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
5900         case EXIT_REASON_INVALID_STATE:
5901                 return 1;
5902         case EXIT_REASON_MWAIT_INSTRUCTION:
5903                 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
5904         case EXIT_REASON_MONITOR_INSTRUCTION:
5905                 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
5906         case EXIT_REASON_PAUSE_INSTRUCTION:
5907                 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
5908                         nested_cpu_has2(vmcs12,
5909                                 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
5910         case EXIT_REASON_MCE_DURING_VMENTRY:
5911                 return 0;
5912         case EXIT_REASON_TPR_BELOW_THRESHOLD:
5913                 return 1;
5914         case EXIT_REASON_APIC_ACCESS:
5915                 return nested_cpu_has2(vmcs12,
5916                         SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
5917         case EXIT_REASON_EPT_VIOLATION:
5918         case EXIT_REASON_EPT_MISCONFIG:
5919                 return 0;
5920         case EXIT_REASON_WBINVD:
5921                 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
5922         case EXIT_REASON_XSETBV:
5923                 return 1;
5924         default:
5925                 return 1;
5926         }
5927 }
5928
5929 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
5930 {
5931         *info1 = vmcs_readl(EXIT_QUALIFICATION);
5932         *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
5933 }
5934
5935 /*
5936  * The guest has exited.  See if we can fix it or if we need userspace
5937  * assistance.
5938  */
5939 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
5940 {
5941         struct vcpu_vmx *vmx = to_vmx(vcpu);
5942         u32 exit_reason = vmx->exit_reason;
5943         u32 vectoring_info = vmx->idt_vectoring_info;
5944
5945         /* If guest state is invalid, start emulating */
5946         if (vmx->emulation_required && emulate_invalid_guest_state)
5947                 return handle_invalid_guest_state(vcpu);
5948
5949         /*
5950          * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
5951          * we did not inject a still-pending event to L1 now because of
5952          * nested_run_pending, we need to re-enable this bit.
5953          */
5954         if (vmx->nested.nested_run_pending)
5955                 kvm_make_request(KVM_REQ_EVENT, vcpu);
5956
5957         if (!is_guest_mode(vcpu) && (exit_reason == EXIT_REASON_VMLAUNCH ||
5958             exit_reason == EXIT_REASON_VMRESUME))
5959                 vmx->nested.nested_run_pending = 1;
5960         else
5961                 vmx->nested.nested_run_pending = 0;
5962
5963         if (is_guest_mode(vcpu) && nested_vmx_exit_handled(vcpu)) {
5964                 nested_vmx_vmexit(vcpu);
5965                 return 1;
5966         }
5967
5968         if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
5969                 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5970                 vcpu->run->fail_entry.hardware_entry_failure_reason
5971                         = exit_reason;
5972                 return 0;
5973         }
5974
5975         if (unlikely(vmx->fail)) {
5976                 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5977                 vcpu->run->fail_entry.hardware_entry_failure_reason
5978                         = vmcs_read32(VM_INSTRUCTION_ERROR);
5979                 return 0;
5980         }
5981
5982         if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
5983                         (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
5984                         exit_reason != EXIT_REASON_EPT_VIOLATION &&
5985                         exit_reason != EXIT_REASON_TASK_SWITCH))
5986                 printk(KERN_WARNING "%s: unexpected, valid vectoring info "
5987                        "(0x%x) and exit reason is 0x%x\n",
5988                        __func__, vectoring_info, exit_reason);
5989
5990         if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked &&
5991             !(is_guest_mode(vcpu) && nested_cpu_has_virtual_nmis(
5992                                         get_vmcs12(vcpu), vcpu)))) {
5993                 if (vmx_interrupt_allowed(vcpu)) {
5994                         vmx->soft_vnmi_blocked = 0;
5995                 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
5996                            vcpu->arch.nmi_pending) {
5997                         /*
5998                          * This CPU don't support us in finding the end of an
5999                          * NMI-blocked window if the guest runs with IRQs
6000                          * disabled. So we pull the trigger after 1 s of
6001                          * futile waiting, but inform the user about this.
6002                          */
6003                         printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
6004                                "state on VCPU %d after 1 s timeout\n",
6005                                __func__, vcpu->vcpu_id);
6006                         vmx->soft_vnmi_blocked = 0;
6007                 }
6008         }
6009
6010         if (exit_reason < kvm_vmx_max_exit_handlers
6011             && kvm_vmx_exit_handlers[exit_reason])
6012                 return kvm_vmx_exit_handlers[exit_reason](vcpu);
6013         else {
6014                 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
6015                 vcpu->run->hw.hardware_exit_reason = exit_reason;
6016         }
6017         return 0;
6018 }
6019
6020 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
6021 {
6022         if (irr == -1 || tpr < irr) {
6023                 vmcs_write32(TPR_THRESHOLD, 0);
6024                 return;
6025         }
6026
6027         vmcs_write32(TPR_THRESHOLD, irr);
6028 }
6029
6030 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
6031 {
6032         u32 exit_intr_info;
6033
6034         if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
6035               || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
6036                 return;
6037
6038         vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6039         exit_intr_info = vmx->exit_intr_info;
6040
6041         /* Handle machine checks before interrupts are enabled */
6042         if (is_machine_check(exit_intr_info))
6043                 kvm_machine_check();
6044
6045         /* We need to handle NMIs before interrupts are enabled */
6046         if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
6047             (exit_intr_info & INTR_INFO_VALID_MASK)) {
6048                 kvm_before_handle_nmi(&vmx->vcpu);
6049                 asm("int $2");
6050                 kvm_after_handle_nmi(&vmx->vcpu);
6051         }
6052 }
6053
6054 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6055 {
6056         u32 exit_intr_info;
6057         bool unblock_nmi;
6058         u8 vector;
6059         bool idtv_info_valid;
6060
6061         idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6062
6063         if (cpu_has_virtual_nmis()) {
6064                 if (vmx->nmi_known_unmasked)
6065                         return;
6066                 /*
6067                  * Can't use vmx->exit_intr_info since we're not sure what
6068                  * the exit reason is.
6069                  */
6070                 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6071                 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6072                 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6073                 /*
6074                  * SDM 3: 27.7.1.2 (September 2008)
6075                  * Re-set bit "block by NMI" before VM entry if vmexit caused by
6076                  * a guest IRET fault.
6077                  * SDM 3: 23.2.2 (September 2008)
6078                  * Bit 12 is undefined in any of the following cases:
6079                  *  If the VM exit sets the valid bit in the IDT-vectoring
6080                  *   information field.
6081                  *  If the VM exit is due to a double fault.
6082                  */
6083                 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
6084                     vector != DF_VECTOR && !idtv_info_valid)
6085                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6086                                       GUEST_INTR_STATE_NMI);
6087                 else
6088                         vmx->nmi_known_unmasked =
6089                                 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
6090                                   & GUEST_INTR_STATE_NMI);
6091         } else if (unlikely(vmx->soft_vnmi_blocked))
6092                 vmx->vnmi_blocked_time +=
6093                         ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
6094 }
6095
6096 static void __vmx_complete_interrupts(struct vcpu_vmx *vmx,
6097                                       u32 idt_vectoring_info,
6098                                       int instr_len_field,
6099                                       int error_code_field)
6100 {
6101         u8 vector;
6102         int type;
6103         bool idtv_info_valid;
6104
6105         idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6106
6107         vmx->vcpu.arch.nmi_injected = false;
6108         kvm_clear_exception_queue(&vmx->vcpu);
6109         kvm_clear_interrupt_queue(&vmx->vcpu);
6110
6111         if (!idtv_info_valid)
6112                 return;
6113
6114         kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
6115
6116         vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6117         type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
6118
6119         switch (type) {
6120         case INTR_TYPE_NMI_INTR:
6121                 vmx->vcpu.arch.nmi_injected = true;
6122                 /*
6123                  * SDM 3: 27.7.1.2 (September 2008)
6124                  * Clear bit "block by NMI" before VM entry if a NMI
6125                  * delivery faulted.
6126                  */
6127                 vmx_set_nmi_mask(&vmx->vcpu, false);
6128                 break;
6129         case INTR_TYPE_SOFT_EXCEPTION:
6130                 vmx->vcpu.arch.event_exit_inst_len =
6131                         vmcs_read32(instr_len_field);
6132                 /* fall through */
6133         case INTR_TYPE_HARD_EXCEPTION:
6134                 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
6135                         u32 err = vmcs_read32(error_code_field);
6136                         kvm_queue_exception_e(&vmx->vcpu, vector, err);
6137                 } else
6138                         kvm_queue_exception(&vmx->vcpu, vector);
6139                 break;
6140         case INTR_TYPE_SOFT_INTR:
6141                 vmx->vcpu.arch.event_exit_inst_len =
6142                         vmcs_read32(instr_len_field);
6143                 /* fall through */
6144         case INTR_TYPE_EXT_INTR:
6145                 kvm_queue_interrupt(&vmx->vcpu, vector,
6146                         type == INTR_TYPE_SOFT_INTR);
6147                 break;
6148         default:
6149                 break;
6150         }
6151 }
6152
6153 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6154 {
6155         if (is_guest_mode(&vmx->vcpu))
6156                 return;
6157         __vmx_complete_interrupts(vmx, vmx->idt_vectoring_info,
6158                                   VM_EXIT_INSTRUCTION_LEN,
6159                                   IDT_VECTORING_ERROR_CODE);
6160 }
6161
6162 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6163 {
6164         if (is_guest_mode(vcpu))
6165                 return;
6166         __vmx_complete_interrupts(to_vmx(vcpu),
6167                                   vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6168                                   VM_ENTRY_INSTRUCTION_LEN,
6169                                   VM_ENTRY_EXCEPTION_ERROR_CODE);
6170
6171         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6172 }
6173
6174 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6175 {
6176         int i, nr_msrs;
6177         struct perf_guest_switch_msr *msrs;
6178
6179         msrs = perf_guest_get_msrs(&nr_msrs);
6180
6181         if (!msrs)
6182                 return;
6183
6184         for (i = 0; i < nr_msrs; i++)
6185                 if (msrs[i].host == msrs[i].guest)
6186                         clear_atomic_switch_msr(vmx, msrs[i].msr);
6187                 else
6188                         add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6189                                         msrs[i].host);
6190 }
6191
6192 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
6193 {
6194         struct vcpu_vmx *vmx = to_vmx(vcpu);
6195         unsigned long debugctlmsr;
6196
6197         if (is_guest_mode(vcpu) && !vmx->nested.nested_run_pending) {
6198                 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6199                 if (vmcs12->idt_vectoring_info_field &
6200                                 VECTORING_INFO_VALID_MASK) {
6201                         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
6202                                 vmcs12->idt_vectoring_info_field);
6203                         vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
6204                                 vmcs12->vm_exit_instruction_len);
6205                         if (vmcs12->idt_vectoring_info_field &
6206                                         VECTORING_INFO_DELIVER_CODE_MASK)
6207                                 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
6208                                         vmcs12->idt_vectoring_error_code);
6209                 }
6210         }
6211
6212         /* Record the guest's net vcpu time for enforced NMI injections. */
6213         if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
6214                 vmx->entry_time = ktime_get();
6215
6216         /* Don't enter VMX if guest state is invalid, let the exit handler
6217            start emulation until we arrive back to a valid state */
6218         if (vmx->emulation_required && emulate_invalid_guest_state)
6219                 return;
6220
6221         if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
6222                 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
6223         if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
6224                 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
6225
6226         /* When single-stepping over STI and MOV SS, we must clear the
6227          * corresponding interruptibility bits in the guest state. Otherwise
6228          * vmentry fails as it then expects bit 14 (BS) in pending debug
6229          * exceptions being set, but that's not correct for the guest debugging
6230          * case. */
6231         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6232                 vmx_set_interrupt_shadow(vcpu, 0);
6233
6234         atomic_switch_perf_msrs(vmx);
6235         debugctlmsr = get_debugctlmsr();
6236
6237         vmx->__launched = vmx->loaded_vmcs->launched;
6238         asm(
6239                 /* Store host registers */
6240                 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
6241                 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
6242                 "push %%" _ASM_CX " \n\t"
6243                 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
6244                 "je 1f \n\t"
6245                 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
6246                 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
6247                 "1: \n\t"
6248                 /* Reload cr2 if changed */
6249                 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
6250                 "mov %%cr2, %%" _ASM_DX " \n\t"
6251                 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
6252                 "je 2f \n\t"
6253                 "mov %%" _ASM_AX", %%cr2 \n\t"
6254                 "2: \n\t"
6255                 /* Check if vmlaunch of vmresume is needed */
6256                 "cmpl $0, %c[launched](%0) \n\t"
6257                 /* Load guest registers.  Don't clobber flags. */
6258                 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
6259                 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
6260                 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
6261                 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
6262                 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
6263                 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
6264 #ifdef CONFIG_X86_64
6265                 "mov %c[r8](%0),  %%r8  \n\t"
6266                 "mov %c[r9](%0),  %%r9  \n\t"
6267                 "mov %c[r10](%0), %%r10 \n\t"
6268                 "mov %c[r11](%0), %%r11 \n\t"
6269                 "mov %c[r12](%0), %%r12 \n\t"
6270                 "mov %c[r13](%0), %%r13 \n\t"
6271                 "mov %c[r14](%0), %%r14 \n\t"
6272                 "mov %c[r15](%0), %%r15 \n\t"
6273 #endif
6274                 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
6275
6276                 /* Enter guest mode */
6277                 "jne 1f \n\t"
6278                 __ex(ASM_VMX_VMLAUNCH) "\n\t"
6279                 "jmp 2f \n\t"
6280                 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
6281                 "2: "
6282                 /* Save guest registers, load host registers, keep flags */
6283                 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
6284                 "pop %0 \n\t"
6285                 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
6286                 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
6287                 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
6288                 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
6289                 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
6290                 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
6291                 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
6292 #ifdef CONFIG_X86_64
6293                 "mov %%r8,  %c[r8](%0) \n\t"
6294                 "mov %%r9,  %c[r9](%0) \n\t"
6295                 "mov %%r10, %c[r10](%0) \n\t"
6296                 "mov %%r11, %c[r11](%0) \n\t"
6297                 "mov %%r12, %c[r12](%0) \n\t"
6298                 "mov %%r13, %c[r13](%0) \n\t"
6299                 "mov %%r14, %c[r14](%0) \n\t"
6300                 "mov %%r15, %c[r15](%0) \n\t"
6301 #endif
6302                 "mov %%cr2, %%" _ASM_AX "   \n\t"
6303                 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
6304
6305                 "pop  %%" _ASM_BP "; pop  %%" _ASM_DX " \n\t"
6306                 "setbe %c[fail](%0) \n\t"
6307                 ".pushsection .rodata \n\t"
6308                 ".global vmx_return \n\t"
6309     &