KVM: VMX: Respect interrupt window in big real mode
[linux-2.6.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 affilates.
9  *
10  * Authors:
11  *   Avi Kivity   <avi@qumranet.com>
12  *   Yaniv Kamay  <yaniv@qumranet.com>
13  *
14  * This work is licensed under the terms of the GNU GPL, version 2.  See
15  * the COPYING file in the top-level directory.
16  *
17  */
18
19 #include "irq.h"
20 #include "mmu.h"
21
22 #include <linux/kvm_host.h>
23 #include <linux/module.h>
24 #include <linux/kernel.h>
25 #include <linux/mm.h>
26 #include <linux/highmem.h>
27 #include <linux/sched.h>
28 #include <linux/moduleparam.h>
29 #include <linux/ftrace_event.h>
30 #include <linux/slab.h>
31 #include <linux/tboot.h>
32 #include "kvm_cache_regs.h"
33 #include "x86.h"
34
35 #include <asm/io.h>
36 #include <asm/desc.h>
37 #include <asm/vmx.h>
38 #include <asm/virtext.h>
39 #include <asm/mce.h>
40 #include <asm/i387.h>
41 #include <asm/xcr.h>
42
43 #include "trace.h"
44
45 #define __ex(x) __kvm_handle_fault_on_reboot(x)
46
47 MODULE_AUTHOR("Qumranet");
48 MODULE_LICENSE("GPL");
49
50 static int __read_mostly bypass_guest_pf = 1;
51 module_param(bypass_guest_pf, bool, S_IRUGO);
52
53 static int __read_mostly enable_vpid = 1;
54 module_param_named(vpid, enable_vpid, bool, 0444);
55
56 static int __read_mostly flexpriority_enabled = 1;
57 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
58
59 static int __read_mostly enable_ept = 1;
60 module_param_named(ept, enable_ept, bool, S_IRUGO);
61
62 static int __read_mostly enable_unrestricted_guest = 1;
63 module_param_named(unrestricted_guest,
64                         enable_unrestricted_guest, bool, S_IRUGO);
65
66 static int __read_mostly emulate_invalid_guest_state = 0;
67 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
68
69 static int __read_mostly vmm_exclusive = 1;
70 module_param(vmm_exclusive, bool, S_IRUGO);
71
72 #define KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST                           \
73         (X86_CR0_WP | X86_CR0_NE | X86_CR0_NW | X86_CR0_CD)
74 #define KVM_GUEST_CR0_MASK                                              \
75         (KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
76 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST                         \
77         (X86_CR0_WP | X86_CR0_NE)
78 #define KVM_VM_CR0_ALWAYS_ON                                            \
79         (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
80 #define KVM_CR4_GUEST_OWNED_BITS                                      \
81         (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR      \
82          | X86_CR4_OSXMMEXCPT)
83
84 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
85 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
86
87 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
88
89 /*
90  * These 2 parameters are used to config the controls for Pause-Loop Exiting:
91  * ple_gap:    upper bound on the amount of time between two successive
92  *             executions of PAUSE in a loop. Also indicate if ple enabled.
93  *             According to test, this time is usually small than 41 cycles.
94  * ple_window: upper bound on the amount of time a guest is allowed to execute
95  *             in a PAUSE loop. Tests indicate that most spinlocks are held for
96  *             less than 2^12 cycles
97  * Time is measured based on a counter that runs at the same rate as the TSC,
98  * refer SDM volume 3b section 21.6.13 & 22.1.3.
99  */
100 #define KVM_VMX_DEFAULT_PLE_GAP    41
101 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
102 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
103 module_param(ple_gap, int, S_IRUGO);
104
105 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
106 module_param(ple_window, int, S_IRUGO);
107
108 #define NR_AUTOLOAD_MSRS 1
109
110 struct vmcs {
111         u32 revision_id;
112         u32 abort;
113         char data[0];
114 };
115
116 struct shared_msr_entry {
117         unsigned index;
118         u64 data;
119         u64 mask;
120 };
121
122 struct vcpu_vmx {
123         struct kvm_vcpu       vcpu;
124         struct list_head      local_vcpus_link;
125         unsigned long         host_rsp;
126         int                   launched;
127         u8                    fail;
128         u32                   exit_intr_info;
129         u32                   idt_vectoring_info;
130         struct shared_msr_entry *guest_msrs;
131         int                   nmsrs;
132         int                   save_nmsrs;
133 #ifdef CONFIG_X86_64
134         u64                   msr_host_kernel_gs_base;
135         u64                   msr_guest_kernel_gs_base;
136 #endif
137         struct vmcs          *vmcs;
138         struct msr_autoload {
139                 unsigned nr;
140                 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
141                 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
142         } msr_autoload;
143         struct {
144                 int           loaded;
145                 u16           fs_sel, gs_sel, ldt_sel;
146                 int           gs_ldt_reload_needed;
147                 int           fs_reload_needed;
148         } host_state;
149         struct {
150                 int vm86_active;
151                 ulong save_rflags;
152                 struct kvm_save_segment {
153                         u16 selector;
154                         unsigned long base;
155                         u32 limit;
156                         u32 ar;
157                 } tr, es, ds, fs, gs;
158         } rmode;
159         int vpid;
160         bool emulation_required;
161
162         /* Support for vnmi-less CPUs */
163         int soft_vnmi_blocked;
164         ktime_t entry_time;
165         s64 vnmi_blocked_time;
166         u32 exit_reason;
167
168         bool rdtscp_enabled;
169 };
170
171 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
172 {
173         return container_of(vcpu, struct vcpu_vmx, vcpu);
174 }
175
176 static int init_rmode(struct kvm *kvm);
177 static u64 construct_eptp(unsigned long root_hpa);
178 static void kvm_cpu_vmxon(u64 addr);
179 static void kvm_cpu_vmxoff(void);
180
181 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
182 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
183 static DEFINE_PER_CPU(struct list_head, vcpus_on_cpu);
184 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
185
186 static unsigned long *vmx_io_bitmap_a;
187 static unsigned long *vmx_io_bitmap_b;
188 static unsigned long *vmx_msr_bitmap_legacy;
189 static unsigned long *vmx_msr_bitmap_longmode;
190
191 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
192 static DEFINE_SPINLOCK(vmx_vpid_lock);
193
194 static struct vmcs_config {
195         int size;
196         int order;
197         u32 revision_id;
198         u32 pin_based_exec_ctrl;
199         u32 cpu_based_exec_ctrl;
200         u32 cpu_based_2nd_exec_ctrl;
201         u32 vmexit_ctrl;
202         u32 vmentry_ctrl;
203 } vmcs_config;
204
205 static struct vmx_capability {
206         u32 ept;
207         u32 vpid;
208 } vmx_capability;
209
210 #define VMX_SEGMENT_FIELD(seg)                                  \
211         [VCPU_SREG_##seg] = {                                   \
212                 .selector = GUEST_##seg##_SELECTOR,             \
213                 .base = GUEST_##seg##_BASE,                     \
214                 .limit = GUEST_##seg##_LIMIT,                   \
215                 .ar_bytes = GUEST_##seg##_AR_BYTES,             \
216         }
217
218 static struct kvm_vmx_segment_field {
219         unsigned selector;
220         unsigned base;
221         unsigned limit;
222         unsigned ar_bytes;
223 } kvm_vmx_segment_fields[] = {
224         VMX_SEGMENT_FIELD(CS),
225         VMX_SEGMENT_FIELD(DS),
226         VMX_SEGMENT_FIELD(ES),
227         VMX_SEGMENT_FIELD(FS),
228         VMX_SEGMENT_FIELD(GS),
229         VMX_SEGMENT_FIELD(SS),
230         VMX_SEGMENT_FIELD(TR),
231         VMX_SEGMENT_FIELD(LDTR),
232 };
233
234 static u64 host_efer;
235
236 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
237
238 /*
239  * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
240  * away by decrementing the array size.
241  */
242 static const u32 vmx_msr_index[] = {
243 #ifdef CONFIG_X86_64
244         MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
245 #endif
246         MSR_EFER, MSR_TSC_AUX, MSR_STAR,
247 };
248 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
249
250 static inline bool is_page_fault(u32 intr_info)
251 {
252         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
253                              INTR_INFO_VALID_MASK)) ==
254                 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
255 }
256
257 static inline bool is_no_device(u32 intr_info)
258 {
259         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
260                              INTR_INFO_VALID_MASK)) ==
261                 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
262 }
263
264 static inline bool is_invalid_opcode(u32 intr_info)
265 {
266         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
267                              INTR_INFO_VALID_MASK)) ==
268                 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
269 }
270
271 static inline bool is_external_interrupt(u32 intr_info)
272 {
273         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
274                 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
275 }
276
277 static inline bool is_machine_check(u32 intr_info)
278 {
279         return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
280                              INTR_INFO_VALID_MASK)) ==
281                 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
282 }
283
284 static inline bool cpu_has_vmx_msr_bitmap(void)
285 {
286         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
287 }
288
289 static inline bool cpu_has_vmx_tpr_shadow(void)
290 {
291         return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
292 }
293
294 static inline bool vm_need_tpr_shadow(struct kvm *kvm)
295 {
296         return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
297 }
298
299 static inline bool cpu_has_secondary_exec_ctrls(void)
300 {
301         return vmcs_config.cpu_based_exec_ctrl &
302                 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
303 }
304
305 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
306 {
307         return vmcs_config.cpu_based_2nd_exec_ctrl &
308                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
309 }
310
311 static inline bool cpu_has_vmx_flexpriority(void)
312 {
313         return cpu_has_vmx_tpr_shadow() &&
314                 cpu_has_vmx_virtualize_apic_accesses();
315 }
316
317 static inline bool cpu_has_vmx_ept_execute_only(void)
318 {
319         return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
320 }
321
322 static inline bool cpu_has_vmx_eptp_uncacheable(void)
323 {
324         return vmx_capability.ept & VMX_EPTP_UC_BIT;
325 }
326
327 static inline bool cpu_has_vmx_eptp_writeback(void)
328 {
329         return vmx_capability.ept & VMX_EPTP_WB_BIT;
330 }
331
332 static inline bool cpu_has_vmx_ept_2m_page(void)
333 {
334         return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
335 }
336
337 static inline bool cpu_has_vmx_ept_1g_page(void)
338 {
339         return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
340 }
341
342 static inline bool cpu_has_vmx_ept_4levels(void)
343 {
344         return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
345 }
346
347 static inline bool cpu_has_vmx_invept_individual_addr(void)
348 {
349         return vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT;
350 }
351
352 static inline bool cpu_has_vmx_invept_context(void)
353 {
354         return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
355 }
356
357 static inline bool cpu_has_vmx_invept_global(void)
358 {
359         return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
360 }
361
362 static inline bool cpu_has_vmx_invvpid_single(void)
363 {
364         return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
365 }
366
367 static inline bool cpu_has_vmx_invvpid_global(void)
368 {
369         return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
370 }
371
372 static inline bool cpu_has_vmx_ept(void)
373 {
374         return vmcs_config.cpu_based_2nd_exec_ctrl &
375                 SECONDARY_EXEC_ENABLE_EPT;
376 }
377
378 static inline bool cpu_has_vmx_unrestricted_guest(void)
379 {
380         return vmcs_config.cpu_based_2nd_exec_ctrl &
381                 SECONDARY_EXEC_UNRESTRICTED_GUEST;
382 }
383
384 static inline bool cpu_has_vmx_ple(void)
385 {
386         return vmcs_config.cpu_based_2nd_exec_ctrl &
387                 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
388 }
389
390 static inline bool vm_need_virtualize_apic_accesses(struct kvm *kvm)
391 {
392         return flexpriority_enabled && irqchip_in_kernel(kvm);
393 }
394
395 static inline bool cpu_has_vmx_vpid(void)
396 {
397         return vmcs_config.cpu_based_2nd_exec_ctrl &
398                 SECONDARY_EXEC_ENABLE_VPID;
399 }
400
401 static inline bool cpu_has_vmx_rdtscp(void)
402 {
403         return vmcs_config.cpu_based_2nd_exec_ctrl &
404                 SECONDARY_EXEC_RDTSCP;
405 }
406
407 static inline bool cpu_has_virtual_nmis(void)
408 {
409         return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
410 }
411
412 static inline bool cpu_has_vmx_wbinvd_exit(void)
413 {
414         return vmcs_config.cpu_based_2nd_exec_ctrl &
415                 SECONDARY_EXEC_WBINVD_EXITING;
416 }
417
418 static inline bool report_flexpriority(void)
419 {
420         return flexpriority_enabled;
421 }
422
423 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
424 {
425         int i;
426
427         for (i = 0; i < vmx->nmsrs; ++i)
428                 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
429                         return i;
430         return -1;
431 }
432
433 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
434 {
435     struct {
436         u64 vpid : 16;
437         u64 rsvd : 48;
438         u64 gva;
439     } operand = { vpid, 0, gva };
440
441     asm volatile (__ex(ASM_VMX_INVVPID)
442                   /* CF==1 or ZF==1 --> rc = -1 */
443                   "; ja 1f ; ud2 ; 1:"
444                   : : "a"(&operand), "c"(ext) : "cc", "memory");
445 }
446
447 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
448 {
449         struct {
450                 u64 eptp, gpa;
451         } operand = {eptp, gpa};
452
453         asm volatile (__ex(ASM_VMX_INVEPT)
454                         /* CF==1 or ZF==1 --> rc = -1 */
455                         "; ja 1f ; ud2 ; 1:\n"
456                         : : "a" (&operand), "c" (ext) : "cc", "memory");
457 }
458
459 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
460 {
461         int i;
462
463         i = __find_msr_index(vmx, msr);
464         if (i >= 0)
465                 return &vmx->guest_msrs[i];
466         return NULL;
467 }
468
469 static void vmcs_clear(struct vmcs *vmcs)
470 {
471         u64 phys_addr = __pa(vmcs);
472         u8 error;
473
474         asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
475                       : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
476                       : "cc", "memory");
477         if (error)
478                 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
479                        vmcs, phys_addr);
480 }
481
482 static void vmcs_load(struct vmcs *vmcs)
483 {
484         u64 phys_addr = __pa(vmcs);
485         u8 error;
486
487         asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
488                         : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
489                         : "cc", "memory");
490         if (error)
491                 printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
492                        vmcs, phys_addr);
493 }
494
495 static void __vcpu_clear(void *arg)
496 {
497         struct vcpu_vmx *vmx = arg;
498         int cpu = raw_smp_processor_id();
499
500         if (vmx->vcpu.cpu == cpu)
501                 vmcs_clear(vmx->vmcs);
502         if (per_cpu(current_vmcs, cpu) == vmx->vmcs)
503                 per_cpu(current_vmcs, cpu) = NULL;
504         list_del(&vmx->local_vcpus_link);
505         vmx->vcpu.cpu = -1;
506         vmx->launched = 0;
507 }
508
509 static void vcpu_clear(struct vcpu_vmx *vmx)
510 {
511         if (vmx->vcpu.cpu == -1)
512                 return;
513         smp_call_function_single(vmx->vcpu.cpu, __vcpu_clear, vmx, 1);
514 }
515
516 static inline void vpid_sync_vcpu_single(struct vcpu_vmx *vmx)
517 {
518         if (vmx->vpid == 0)
519                 return;
520
521         if (cpu_has_vmx_invvpid_single())
522                 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
523 }
524
525 static inline void vpid_sync_vcpu_global(void)
526 {
527         if (cpu_has_vmx_invvpid_global())
528                 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
529 }
530
531 static inline void vpid_sync_context(struct vcpu_vmx *vmx)
532 {
533         if (cpu_has_vmx_invvpid_single())
534                 vpid_sync_vcpu_single(vmx);
535         else
536                 vpid_sync_vcpu_global();
537 }
538
539 static inline void ept_sync_global(void)
540 {
541         if (cpu_has_vmx_invept_global())
542                 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
543 }
544
545 static inline void ept_sync_context(u64 eptp)
546 {
547         if (enable_ept) {
548                 if (cpu_has_vmx_invept_context())
549                         __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
550                 else
551                         ept_sync_global();
552         }
553 }
554
555 static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
556 {
557         if (enable_ept) {
558                 if (cpu_has_vmx_invept_individual_addr())
559                         __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
560                                         eptp, gpa);
561                 else
562                         ept_sync_context(eptp);
563         }
564 }
565
566 static unsigned long vmcs_readl(unsigned long field)
567 {
568         unsigned long value;
569
570         asm volatile (__ex(ASM_VMX_VMREAD_RDX_RAX)
571                       : "=a"(value) : "d"(field) : "cc");
572         return value;
573 }
574
575 static u16 vmcs_read16(unsigned long field)
576 {
577         return vmcs_readl(field);
578 }
579
580 static u32 vmcs_read32(unsigned long field)
581 {
582         return vmcs_readl(field);
583 }
584
585 static u64 vmcs_read64(unsigned long field)
586 {
587 #ifdef CONFIG_X86_64
588         return vmcs_readl(field);
589 #else
590         return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
591 #endif
592 }
593
594 static noinline void vmwrite_error(unsigned long field, unsigned long value)
595 {
596         printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
597                field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
598         dump_stack();
599 }
600
601 static void vmcs_writel(unsigned long field, unsigned long value)
602 {
603         u8 error;
604
605         asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
606                        : "=q"(error) : "a"(value), "d"(field) : "cc");
607         if (unlikely(error))
608                 vmwrite_error(field, value);
609 }
610
611 static void vmcs_write16(unsigned long field, u16 value)
612 {
613         vmcs_writel(field, value);
614 }
615
616 static void vmcs_write32(unsigned long field, u32 value)
617 {
618         vmcs_writel(field, value);
619 }
620
621 static void vmcs_write64(unsigned long field, u64 value)
622 {
623         vmcs_writel(field, value);
624 #ifndef CONFIG_X86_64
625         asm volatile ("");
626         vmcs_writel(field+1, value >> 32);
627 #endif
628 }
629
630 static void vmcs_clear_bits(unsigned long field, u32 mask)
631 {
632         vmcs_writel(field, vmcs_readl(field) & ~mask);
633 }
634
635 static void vmcs_set_bits(unsigned long field, u32 mask)
636 {
637         vmcs_writel(field, vmcs_readl(field) | mask);
638 }
639
640 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
641 {
642         u32 eb;
643
644         eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
645              (1u << NM_VECTOR) | (1u << DB_VECTOR);
646         if ((vcpu->guest_debug &
647              (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
648             (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
649                 eb |= 1u << BP_VECTOR;
650         if (to_vmx(vcpu)->rmode.vm86_active)
651                 eb = ~0;
652         if (enable_ept)
653                 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
654         if (vcpu->fpu_active)
655                 eb &= ~(1u << NM_VECTOR);
656         vmcs_write32(EXCEPTION_BITMAP, eb);
657 }
658
659 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
660 {
661         unsigned i;
662         struct msr_autoload *m = &vmx->msr_autoload;
663
664         for (i = 0; i < m->nr; ++i)
665                 if (m->guest[i].index == msr)
666                         break;
667
668         if (i == m->nr)
669                 return;
670         --m->nr;
671         m->guest[i] = m->guest[m->nr];
672         m->host[i] = m->host[m->nr];
673         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
674         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
675 }
676
677 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
678                                   u64 guest_val, u64 host_val)
679 {
680         unsigned i;
681         struct msr_autoload *m = &vmx->msr_autoload;
682
683         for (i = 0; i < m->nr; ++i)
684                 if (m->guest[i].index == msr)
685                         break;
686
687         if (i == m->nr) {
688                 ++m->nr;
689                 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
690                 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
691         }
692
693         m->guest[i].index = msr;
694         m->guest[i].value = guest_val;
695         m->host[i].index = msr;
696         m->host[i].value = host_val;
697 }
698
699 static void reload_tss(void)
700 {
701         /*
702          * VT restores TR but not its size.  Useless.
703          */
704         struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
705         struct desc_struct *descs;
706
707         descs = (void *)gdt->address;
708         descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
709         load_TR_desc();
710 }
711
712 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
713 {
714         u64 guest_efer;
715         u64 ignore_bits;
716
717         guest_efer = vmx->vcpu.arch.efer;
718
719         /*
720          * NX is emulated; LMA and LME handled by hardware; SCE meaninless
721          * outside long mode
722          */
723         ignore_bits = EFER_NX | EFER_SCE;
724 #ifdef CONFIG_X86_64
725         ignore_bits |= EFER_LMA | EFER_LME;
726         /* SCE is meaningful only in long mode on Intel */
727         if (guest_efer & EFER_LMA)
728                 ignore_bits &= ~(u64)EFER_SCE;
729 #endif
730         guest_efer &= ~ignore_bits;
731         guest_efer |= host_efer & ignore_bits;
732         vmx->guest_msrs[efer_offset].data = guest_efer;
733         vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
734
735         clear_atomic_switch_msr(vmx, MSR_EFER);
736         /* On ept, can't emulate nx, and must switch nx atomically */
737         if (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX)) {
738                 guest_efer = vmx->vcpu.arch.efer;
739                 if (!(guest_efer & EFER_LMA))
740                         guest_efer &= ~EFER_LME;
741                 add_atomic_switch_msr(vmx, MSR_EFER, guest_efer, host_efer);
742                 return false;
743         }
744
745         return true;
746 }
747
748 static unsigned long segment_base(u16 selector)
749 {
750         struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
751         struct desc_struct *d;
752         unsigned long table_base;
753         unsigned long v;
754
755         if (!(selector & ~3))
756                 return 0;
757
758         table_base = gdt->address;
759
760         if (selector & 4) {           /* from ldt */
761                 u16 ldt_selector = kvm_read_ldt();
762
763                 if (!(ldt_selector & ~3))
764                         return 0;
765
766                 table_base = segment_base(ldt_selector);
767         }
768         d = (struct desc_struct *)(table_base + (selector & ~7));
769         v = get_desc_base(d);
770 #ifdef CONFIG_X86_64
771        if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
772                v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
773 #endif
774         return v;
775 }
776
777 static inline unsigned long kvm_read_tr_base(void)
778 {
779         u16 tr;
780         asm("str %0" : "=g"(tr));
781         return segment_base(tr);
782 }
783
784 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
785 {
786         struct vcpu_vmx *vmx = to_vmx(vcpu);
787         int i;
788
789         if (vmx->host_state.loaded)
790                 return;
791
792         vmx->host_state.loaded = 1;
793         /*
794          * Set host fs and gs selectors.  Unfortunately, 22.2.3 does not
795          * allow segment selectors with cpl > 0 or ti == 1.
796          */
797         vmx->host_state.ldt_sel = kvm_read_ldt();
798         vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
799         savesegment(fs, vmx->host_state.fs_sel);
800         if (!(vmx->host_state.fs_sel & 7)) {
801                 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
802                 vmx->host_state.fs_reload_needed = 0;
803         } else {
804                 vmcs_write16(HOST_FS_SELECTOR, 0);
805                 vmx->host_state.fs_reload_needed = 1;
806         }
807         savesegment(gs, vmx->host_state.gs_sel);
808         if (!(vmx->host_state.gs_sel & 7))
809                 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
810         else {
811                 vmcs_write16(HOST_GS_SELECTOR, 0);
812                 vmx->host_state.gs_ldt_reload_needed = 1;
813         }
814
815 #ifdef CONFIG_X86_64
816         vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
817         vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
818 #else
819         vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
820         vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
821 #endif
822
823 #ifdef CONFIG_X86_64
824         if (is_long_mode(&vmx->vcpu)) {
825                 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
826                 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
827         }
828 #endif
829         for (i = 0; i < vmx->save_nmsrs; ++i)
830                 kvm_set_shared_msr(vmx->guest_msrs[i].index,
831                                    vmx->guest_msrs[i].data,
832                                    vmx->guest_msrs[i].mask);
833 }
834
835 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
836 {
837         if (!vmx->host_state.loaded)
838                 return;
839
840         ++vmx->vcpu.stat.host_state_reload;
841         vmx->host_state.loaded = 0;
842         if (vmx->host_state.fs_reload_needed)
843                 loadsegment(fs, vmx->host_state.fs_sel);
844         if (vmx->host_state.gs_ldt_reload_needed) {
845                 kvm_load_ldt(vmx->host_state.ldt_sel);
846 #ifdef CONFIG_X86_64
847                 load_gs_index(vmx->host_state.gs_sel);
848                 wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gs);
849 #else
850                 loadsegment(gs, vmx->host_state.gs_sel);
851 #endif
852         }
853         reload_tss();
854 #ifdef CONFIG_X86_64
855         if (is_long_mode(&vmx->vcpu)) {
856                 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
857                 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
858         }
859 #endif
860         if (current_thread_info()->status & TS_USEDFPU)
861                 clts();
862         load_gdt(&__get_cpu_var(host_gdt));
863 }
864
865 static void vmx_load_host_state(struct vcpu_vmx *vmx)
866 {
867         preempt_disable();
868         __vmx_load_host_state(vmx);
869         preempt_enable();
870 }
871
872 /*
873  * Switches to specified vcpu, until a matching vcpu_put(), but assumes
874  * vcpu mutex is already taken.
875  */
876 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
877 {
878         struct vcpu_vmx *vmx = to_vmx(vcpu);
879         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
880
881         if (!vmm_exclusive)
882                 kvm_cpu_vmxon(phys_addr);
883         else if (vcpu->cpu != cpu)
884                 vcpu_clear(vmx);
885
886         if (per_cpu(current_vmcs, cpu) != vmx->vmcs) {
887                 per_cpu(current_vmcs, cpu) = vmx->vmcs;
888                 vmcs_load(vmx->vmcs);
889         }
890
891         if (vcpu->cpu != cpu) {
892                 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
893                 unsigned long sysenter_esp;
894
895                 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
896                 local_irq_disable();
897                 list_add(&vmx->local_vcpus_link,
898                          &per_cpu(vcpus_on_cpu, cpu));
899                 local_irq_enable();
900
901                 /*
902                  * Linux uses per-cpu TSS and GDT, so set these when switching
903                  * processors.
904                  */
905                 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
906                 vmcs_writel(HOST_GDTR_BASE, gdt->address);   /* 22.2.4 */
907
908                 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
909                 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
910         }
911 }
912
913 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
914 {
915         __vmx_load_host_state(to_vmx(vcpu));
916         if (!vmm_exclusive) {
917                 __vcpu_clear(to_vmx(vcpu));
918                 kvm_cpu_vmxoff();
919         }
920 }
921
922 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
923 {
924         ulong cr0;
925
926         if (vcpu->fpu_active)
927                 return;
928         vcpu->fpu_active = 1;
929         cr0 = vmcs_readl(GUEST_CR0);
930         cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
931         cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
932         vmcs_writel(GUEST_CR0, cr0);
933         update_exception_bitmap(vcpu);
934         vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
935         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
936 }
937
938 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
939
940 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
941 {
942         vmx_decache_cr0_guest_bits(vcpu);
943         vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
944         update_exception_bitmap(vcpu);
945         vcpu->arch.cr0_guest_owned_bits = 0;
946         vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
947         vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
948 }
949
950 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
951 {
952         unsigned long rflags, save_rflags;
953
954         rflags = vmcs_readl(GUEST_RFLAGS);
955         if (to_vmx(vcpu)->rmode.vm86_active) {
956                 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
957                 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
958                 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
959         }
960         return rflags;
961 }
962
963 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
964 {
965         if (to_vmx(vcpu)->rmode.vm86_active) {
966                 to_vmx(vcpu)->rmode.save_rflags = rflags;
967                 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
968         }
969         vmcs_writel(GUEST_RFLAGS, rflags);
970 }
971
972 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
973 {
974         u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
975         int ret = 0;
976
977         if (interruptibility & GUEST_INTR_STATE_STI)
978                 ret |= KVM_X86_SHADOW_INT_STI;
979         if (interruptibility & GUEST_INTR_STATE_MOV_SS)
980                 ret |= KVM_X86_SHADOW_INT_MOV_SS;
981
982         return ret & mask;
983 }
984
985 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
986 {
987         u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
988         u32 interruptibility = interruptibility_old;
989
990         interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
991
992         if (mask & KVM_X86_SHADOW_INT_MOV_SS)
993                 interruptibility |= GUEST_INTR_STATE_MOV_SS;
994         else if (mask & KVM_X86_SHADOW_INT_STI)
995                 interruptibility |= GUEST_INTR_STATE_STI;
996
997         if ((interruptibility != interruptibility_old))
998                 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
999 }
1000
1001 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
1002 {
1003         unsigned long rip;
1004
1005         rip = kvm_rip_read(vcpu);
1006         rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1007         kvm_rip_write(vcpu, rip);
1008
1009         /* skipping an emulated instruction also counts */
1010         vmx_set_interrupt_shadow(vcpu, 0);
1011 }
1012
1013 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
1014                                 bool has_error_code, u32 error_code,
1015                                 bool reinject)
1016 {
1017         struct vcpu_vmx *vmx = to_vmx(vcpu);
1018         u32 intr_info = nr | INTR_INFO_VALID_MASK;
1019
1020         if (has_error_code) {
1021                 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1022                 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1023         }
1024
1025         if (vmx->rmode.vm86_active) {
1026                 if (kvm_inject_realmode_interrupt(vcpu, nr) != EMULATE_DONE)
1027                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
1028                 return;
1029         }
1030
1031         if (kvm_exception_is_soft(nr)) {
1032                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1033                              vmx->vcpu.arch.event_exit_inst_len);
1034                 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1035         } else
1036                 intr_info |= INTR_TYPE_HARD_EXCEPTION;
1037
1038         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1039 }
1040
1041 static bool vmx_rdtscp_supported(void)
1042 {
1043         return cpu_has_vmx_rdtscp();
1044 }
1045
1046 /*
1047  * Swap MSR entry in host/guest MSR entry array.
1048  */
1049 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1050 {
1051         struct shared_msr_entry tmp;
1052
1053         tmp = vmx->guest_msrs[to];
1054         vmx->guest_msrs[to] = vmx->guest_msrs[from];
1055         vmx->guest_msrs[from] = tmp;
1056 }
1057
1058 /*
1059  * Set up the vmcs to automatically save and restore system
1060  * msrs.  Don't touch the 64-bit msrs if the guest is in legacy
1061  * mode, as fiddling with msrs is very expensive.
1062  */
1063 static void setup_msrs(struct vcpu_vmx *vmx)
1064 {
1065         int save_nmsrs, index;
1066         unsigned long *msr_bitmap;
1067
1068         vmx_load_host_state(vmx);
1069         save_nmsrs = 0;
1070 #ifdef CONFIG_X86_64
1071         if (is_long_mode(&vmx->vcpu)) {
1072                 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1073                 if (index >= 0)
1074                         move_msr_up(vmx, index, save_nmsrs++);
1075                 index = __find_msr_index(vmx, MSR_LSTAR);
1076                 if (index >= 0)
1077                         move_msr_up(vmx, index, save_nmsrs++);
1078                 index = __find_msr_index(vmx, MSR_CSTAR);
1079                 if (index >= 0)
1080                         move_msr_up(vmx, index, save_nmsrs++);
1081                 index = __find_msr_index(vmx, MSR_TSC_AUX);
1082                 if (index >= 0 && vmx->rdtscp_enabled)
1083                         move_msr_up(vmx, index, save_nmsrs++);
1084                 /*
1085                  * MSR_STAR is only needed on long mode guests, and only
1086                  * if efer.sce is enabled.
1087                  */
1088                 index = __find_msr_index(vmx, MSR_STAR);
1089                 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
1090                         move_msr_up(vmx, index, save_nmsrs++);
1091         }
1092 #endif
1093         index = __find_msr_index(vmx, MSR_EFER);
1094         if (index >= 0 && update_transition_efer(vmx, index))
1095                 move_msr_up(vmx, index, save_nmsrs++);
1096
1097         vmx->save_nmsrs = save_nmsrs;
1098
1099         if (cpu_has_vmx_msr_bitmap()) {
1100                 if (is_long_mode(&vmx->vcpu))
1101                         msr_bitmap = vmx_msr_bitmap_longmode;
1102                 else
1103                         msr_bitmap = vmx_msr_bitmap_legacy;
1104
1105                 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
1106         }
1107 }
1108
1109 /*
1110  * reads and returns guest's timestamp counter "register"
1111  * guest_tsc = host_tsc + tsc_offset    -- 21.3
1112  */
1113 static u64 guest_read_tsc(void)
1114 {
1115         u64 host_tsc, tsc_offset;
1116
1117         rdtscll(host_tsc);
1118         tsc_offset = vmcs_read64(TSC_OFFSET);
1119         return host_tsc + tsc_offset;
1120 }
1121
1122 /*
1123  * writes 'offset' into guest's timestamp counter offset register
1124  */
1125 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1126 {
1127         vmcs_write64(TSC_OFFSET, offset);
1128 }
1129
1130 static void vmx_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment)
1131 {
1132         u64 offset = vmcs_read64(TSC_OFFSET);
1133         vmcs_write64(TSC_OFFSET, offset + adjustment);
1134 }
1135
1136 /*
1137  * Reads an msr value (of 'msr_index') into 'pdata'.
1138  * Returns 0 on success, non-0 otherwise.
1139  * Assumes vcpu_load() was already called.
1140  */
1141 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1142 {
1143         u64 data;
1144         struct shared_msr_entry *msr;
1145
1146         if (!pdata) {
1147                 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
1148                 return -EINVAL;
1149         }
1150
1151         switch (msr_index) {
1152 #ifdef CONFIG_X86_64
1153         case MSR_FS_BASE:
1154                 data = vmcs_readl(GUEST_FS_BASE);
1155                 break;
1156         case MSR_GS_BASE:
1157                 data = vmcs_readl(GUEST_GS_BASE);
1158                 break;
1159         case MSR_KERNEL_GS_BASE:
1160                 vmx_load_host_state(to_vmx(vcpu));
1161                 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
1162                 break;
1163 #endif
1164         case MSR_EFER:
1165                 return kvm_get_msr_common(vcpu, msr_index, pdata);
1166         case MSR_IA32_TSC:
1167                 data = guest_read_tsc();
1168                 break;
1169         case MSR_IA32_SYSENTER_CS:
1170                 data = vmcs_read32(GUEST_SYSENTER_CS);
1171                 break;
1172         case MSR_IA32_SYSENTER_EIP:
1173                 data = vmcs_readl(GUEST_SYSENTER_EIP);
1174                 break;
1175         case MSR_IA32_SYSENTER_ESP:
1176                 data = vmcs_readl(GUEST_SYSENTER_ESP);
1177                 break;
1178         case MSR_TSC_AUX:
1179                 if (!to_vmx(vcpu)->rdtscp_enabled)
1180                         return 1;
1181                 /* Otherwise falls through */
1182         default:
1183                 vmx_load_host_state(to_vmx(vcpu));
1184                 msr = find_msr_entry(to_vmx(vcpu), msr_index);
1185                 if (msr) {
1186                         vmx_load_host_state(to_vmx(vcpu));
1187                         data = msr->data;
1188                         break;
1189                 }
1190                 return kvm_get_msr_common(vcpu, msr_index, pdata);
1191         }
1192
1193         *pdata = data;
1194         return 0;
1195 }
1196
1197 /*
1198  * Writes msr value into into the appropriate "register".
1199  * Returns 0 on success, non-0 otherwise.
1200  * Assumes vcpu_load() was already called.
1201  */
1202 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1203 {
1204         struct vcpu_vmx *vmx = to_vmx(vcpu);
1205         struct shared_msr_entry *msr;
1206         int ret = 0;
1207
1208         switch (msr_index) {
1209         case MSR_EFER:
1210                 vmx_load_host_state(vmx);
1211                 ret = kvm_set_msr_common(vcpu, msr_index, data);
1212                 break;
1213 #ifdef CONFIG_X86_64
1214         case MSR_FS_BASE:
1215                 vmcs_writel(GUEST_FS_BASE, data);
1216                 break;
1217         case MSR_GS_BASE:
1218                 vmcs_writel(GUEST_GS_BASE, data);
1219                 break;
1220         case MSR_KERNEL_GS_BASE:
1221                 vmx_load_host_state(vmx);
1222                 vmx->msr_guest_kernel_gs_base = data;
1223                 break;
1224 #endif
1225         case MSR_IA32_SYSENTER_CS:
1226                 vmcs_write32(GUEST_SYSENTER_CS, data);
1227                 break;
1228         case MSR_IA32_SYSENTER_EIP:
1229                 vmcs_writel(GUEST_SYSENTER_EIP, data);
1230                 break;
1231         case MSR_IA32_SYSENTER_ESP:
1232                 vmcs_writel(GUEST_SYSENTER_ESP, data);
1233                 break;
1234         case MSR_IA32_TSC:
1235                 kvm_write_tsc(vcpu, data);
1236                 break;
1237         case MSR_IA32_CR_PAT:
1238                 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
1239                         vmcs_write64(GUEST_IA32_PAT, data);
1240                         vcpu->arch.pat = data;
1241                         break;
1242                 }
1243                 ret = kvm_set_msr_common(vcpu, msr_index, data);
1244                 break;
1245         case MSR_TSC_AUX:
1246                 if (!vmx->rdtscp_enabled)
1247                         return 1;
1248                 /* Check reserved bit, higher 32 bits should be zero */
1249                 if ((data >> 32) != 0)
1250                         return 1;
1251                 /* Otherwise falls through */
1252         default:
1253                 msr = find_msr_entry(vmx, msr_index);
1254                 if (msr) {
1255                         vmx_load_host_state(vmx);
1256                         msr->data = data;
1257                         break;
1258                 }
1259                 ret = kvm_set_msr_common(vcpu, msr_index, data);
1260         }
1261
1262         return ret;
1263 }
1264
1265 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1266 {
1267         __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
1268         switch (reg) {
1269         case VCPU_REGS_RSP:
1270                 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
1271                 break;
1272         case VCPU_REGS_RIP:
1273                 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
1274                 break;
1275         case VCPU_EXREG_PDPTR:
1276                 if (enable_ept)
1277                         ept_save_pdptrs(vcpu);
1278                 break;
1279         default:
1280                 break;
1281         }
1282 }
1283
1284 static void set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
1285 {
1286         if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1287                 vmcs_writel(GUEST_DR7, dbg->arch.debugreg[7]);
1288         else
1289                 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
1290
1291         update_exception_bitmap(vcpu);
1292 }
1293
1294 static __init int cpu_has_kvm_support(void)
1295 {
1296         return cpu_has_vmx();
1297 }
1298
1299 static __init int vmx_disabled_by_bios(void)
1300 {
1301         u64 msr;
1302
1303         rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
1304         if (msr & FEATURE_CONTROL_LOCKED) {
1305                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
1306                         && tboot_enabled())
1307                         return 1;
1308                 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
1309                         && !tboot_enabled())
1310                         return 1;
1311         }
1312
1313         return 0;
1314         /* locked but not enabled */
1315 }
1316
1317 static void kvm_cpu_vmxon(u64 addr)
1318 {
1319         asm volatile (ASM_VMX_VMXON_RAX
1320                         : : "a"(&addr), "m"(addr)
1321                         : "memory", "cc");
1322 }
1323
1324 static int hardware_enable(void *garbage)
1325 {
1326         int cpu = raw_smp_processor_id();
1327         u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1328         u64 old, test_bits;
1329
1330         if (read_cr4() & X86_CR4_VMXE)
1331                 return -EBUSY;
1332
1333         INIT_LIST_HEAD(&per_cpu(vcpus_on_cpu, cpu));
1334         rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
1335
1336         test_bits = FEATURE_CONTROL_LOCKED;
1337         test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
1338         if (tboot_enabled())
1339                 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
1340
1341         if ((old & test_bits) != test_bits) {
1342                 /* enable and lock */
1343                 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
1344         }
1345         write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
1346
1347         if (vmm_exclusive) {
1348                 kvm_cpu_vmxon(phys_addr);
1349                 ept_sync_global();
1350         }
1351
1352         store_gdt(&__get_cpu_var(host_gdt));
1353
1354         return 0;
1355 }
1356
1357 static void vmclear_local_vcpus(void)
1358 {
1359         int cpu = raw_smp_processor_id();
1360         struct vcpu_vmx *vmx, *n;
1361
1362         list_for_each_entry_safe(vmx, n, &per_cpu(vcpus_on_cpu, cpu),
1363                                  local_vcpus_link)
1364                 __vcpu_clear(vmx);
1365 }
1366
1367
1368 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
1369  * tricks.
1370  */
1371 static void kvm_cpu_vmxoff(void)
1372 {
1373         asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
1374 }
1375
1376 static void hardware_disable(void *garbage)
1377 {
1378         if (vmm_exclusive) {
1379                 vmclear_local_vcpus();
1380                 kvm_cpu_vmxoff();
1381         }
1382         write_cr4(read_cr4() & ~X86_CR4_VMXE);
1383 }
1384
1385 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
1386                                       u32 msr, u32 *result)
1387 {
1388         u32 vmx_msr_low, vmx_msr_high;
1389         u32 ctl = ctl_min | ctl_opt;
1390
1391         rdmsr(msr, vmx_msr_low, vmx_msr_high);
1392
1393         ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
1394         ctl |= vmx_msr_low;  /* bit == 1 in low word  ==> must be one  */
1395
1396         /* Ensure minimum (required) set of control bits are supported. */
1397         if (ctl_min & ~ctl)
1398                 return -EIO;
1399
1400         *result = ctl;
1401         return 0;
1402 }
1403
1404 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
1405 {
1406         u32 vmx_msr_low, vmx_msr_high;
1407         u32 min, opt, min2, opt2;
1408         u32 _pin_based_exec_control = 0;
1409         u32 _cpu_based_exec_control = 0;
1410         u32 _cpu_based_2nd_exec_control = 0;
1411         u32 _vmexit_control = 0;
1412         u32 _vmentry_control = 0;
1413
1414         min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
1415         opt = PIN_BASED_VIRTUAL_NMIS;
1416         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
1417                                 &_pin_based_exec_control) < 0)
1418                 return -EIO;
1419
1420         min = CPU_BASED_HLT_EXITING |
1421 #ifdef CONFIG_X86_64
1422               CPU_BASED_CR8_LOAD_EXITING |
1423               CPU_BASED_CR8_STORE_EXITING |
1424 #endif
1425               CPU_BASED_CR3_LOAD_EXITING |
1426               CPU_BASED_CR3_STORE_EXITING |
1427               CPU_BASED_USE_IO_BITMAPS |
1428               CPU_BASED_MOV_DR_EXITING |
1429               CPU_BASED_USE_TSC_OFFSETING |
1430               CPU_BASED_MWAIT_EXITING |
1431               CPU_BASED_MONITOR_EXITING |
1432               CPU_BASED_INVLPG_EXITING;
1433         opt = CPU_BASED_TPR_SHADOW |
1434               CPU_BASED_USE_MSR_BITMAPS |
1435               CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1436         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1437                                 &_cpu_based_exec_control) < 0)
1438                 return -EIO;
1439 #ifdef CONFIG_X86_64
1440         if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
1441                 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
1442                                            ~CPU_BASED_CR8_STORE_EXITING;
1443 #endif
1444         if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
1445                 min2 = 0;
1446                 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
1447                         SECONDARY_EXEC_WBINVD_EXITING |
1448                         SECONDARY_EXEC_ENABLE_VPID |
1449                         SECONDARY_EXEC_ENABLE_EPT |
1450                         SECONDARY_EXEC_UNRESTRICTED_GUEST |
1451                         SECONDARY_EXEC_PAUSE_LOOP_EXITING |
1452                         SECONDARY_EXEC_RDTSCP;
1453                 if (adjust_vmx_controls(min2, opt2,
1454                                         MSR_IA32_VMX_PROCBASED_CTLS2,
1455                                         &_cpu_based_2nd_exec_control) < 0)
1456                         return -EIO;
1457         }
1458 #ifndef CONFIG_X86_64
1459         if (!(_cpu_based_2nd_exec_control &
1460                                 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
1461                 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
1462 #endif
1463         if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
1464                 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
1465                    enabled */
1466                 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
1467                                              CPU_BASED_CR3_STORE_EXITING |
1468                                              CPU_BASED_INVLPG_EXITING);
1469                 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
1470                       vmx_capability.ept, vmx_capability.vpid);
1471         }
1472
1473         min = 0;
1474 #ifdef CONFIG_X86_64
1475         min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
1476 #endif
1477         opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
1478         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
1479                                 &_vmexit_control) < 0)
1480                 return -EIO;
1481
1482         min = 0;
1483         opt = VM_ENTRY_LOAD_IA32_PAT;
1484         if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
1485                                 &_vmentry_control) < 0)
1486                 return -EIO;
1487
1488         rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
1489
1490         /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
1491         if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
1492                 return -EIO;
1493
1494 #ifdef CONFIG_X86_64
1495         /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
1496         if (vmx_msr_high & (1u<<16))
1497                 return -EIO;
1498 #endif
1499
1500         /* Require Write-Back (WB) memory type for VMCS accesses. */
1501         if (((vmx_msr_high >> 18) & 15) != 6)
1502                 return -EIO;
1503
1504         vmcs_conf->size = vmx_msr_high & 0x1fff;
1505         vmcs_conf->order = get_order(vmcs_config.size);
1506         vmcs_conf->revision_id = vmx_msr_low;
1507
1508         vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
1509         vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
1510         vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
1511         vmcs_conf->vmexit_ctrl         = _vmexit_control;
1512         vmcs_conf->vmentry_ctrl        = _vmentry_control;
1513
1514         return 0;
1515 }
1516
1517 static struct vmcs *alloc_vmcs_cpu(int cpu)
1518 {
1519         int node = cpu_to_node(cpu);
1520         struct page *pages;
1521         struct vmcs *vmcs;
1522
1523         pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
1524         if (!pages)
1525                 return NULL;
1526         vmcs = page_address(pages);
1527         memset(vmcs, 0, vmcs_config.size);
1528         vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
1529         return vmcs;
1530 }
1531
1532 static struct vmcs *alloc_vmcs(void)
1533 {
1534         return alloc_vmcs_cpu(raw_smp_processor_id());
1535 }
1536
1537 static void free_vmcs(struct vmcs *vmcs)
1538 {
1539         free_pages((unsigned long)vmcs, vmcs_config.order);
1540 }
1541
1542 static void free_kvm_area(void)
1543 {
1544         int cpu;
1545
1546         for_each_possible_cpu(cpu) {
1547                 free_vmcs(per_cpu(vmxarea, cpu));
1548                 per_cpu(vmxarea, cpu) = NULL;
1549         }
1550 }
1551
1552 static __init int alloc_kvm_area(void)
1553 {
1554         int cpu;
1555
1556         for_each_possible_cpu(cpu) {
1557                 struct vmcs *vmcs;
1558
1559                 vmcs = alloc_vmcs_cpu(cpu);
1560                 if (!vmcs) {
1561                         free_kvm_area();
1562                         return -ENOMEM;
1563                 }
1564
1565                 per_cpu(vmxarea, cpu) = vmcs;
1566         }
1567         return 0;
1568 }
1569
1570 static __init int hardware_setup(void)
1571 {
1572         if (setup_vmcs_config(&vmcs_config) < 0)
1573                 return -EIO;
1574
1575         if (boot_cpu_has(X86_FEATURE_NX))
1576                 kvm_enable_efer_bits(EFER_NX);
1577
1578         if (!cpu_has_vmx_vpid())
1579                 enable_vpid = 0;
1580
1581         if (!cpu_has_vmx_ept() ||
1582             !cpu_has_vmx_ept_4levels()) {
1583                 enable_ept = 0;
1584                 enable_unrestricted_guest = 0;
1585         }
1586
1587         if (!cpu_has_vmx_unrestricted_guest())
1588                 enable_unrestricted_guest = 0;
1589
1590         if (!cpu_has_vmx_flexpriority())
1591                 flexpriority_enabled = 0;
1592
1593         if (!cpu_has_vmx_tpr_shadow())
1594                 kvm_x86_ops->update_cr8_intercept = NULL;
1595
1596         if (enable_ept && !cpu_has_vmx_ept_2m_page())
1597                 kvm_disable_largepages();
1598
1599         if (!cpu_has_vmx_ple())
1600                 ple_gap = 0;
1601
1602         return alloc_kvm_area();
1603 }
1604
1605 static __exit void hardware_unsetup(void)
1606 {
1607         free_kvm_area();
1608 }
1609
1610 static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
1611 {
1612         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1613
1614         if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
1615                 vmcs_write16(sf->selector, save->selector);
1616                 vmcs_writel(sf->base, save->base);
1617                 vmcs_write32(sf->limit, save->limit);
1618                 vmcs_write32(sf->ar_bytes, save->ar);
1619         } else {
1620                 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
1621                         << AR_DPL_SHIFT;
1622                 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
1623         }
1624 }
1625
1626 static void enter_pmode(struct kvm_vcpu *vcpu)
1627 {
1628         unsigned long flags;
1629         struct vcpu_vmx *vmx = to_vmx(vcpu);
1630
1631         vmx->emulation_required = 1;
1632         vmx->rmode.vm86_active = 0;
1633
1634         vmcs_writel(GUEST_TR_BASE, vmx->rmode.tr.base);
1635         vmcs_write32(GUEST_TR_LIMIT, vmx->rmode.tr.limit);
1636         vmcs_write32(GUEST_TR_AR_BYTES, vmx->rmode.tr.ar);
1637
1638         flags = vmcs_readl(GUEST_RFLAGS);
1639         flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1640         flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1641         vmcs_writel(GUEST_RFLAGS, flags);
1642
1643         vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
1644                         (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
1645
1646         update_exception_bitmap(vcpu);
1647
1648         if (emulate_invalid_guest_state)
1649                 return;
1650
1651         fix_pmode_dataseg(VCPU_SREG_ES, &vmx->rmode.es);
1652         fix_pmode_dataseg(VCPU_SREG_DS, &vmx->rmode.ds);
1653         fix_pmode_dataseg(VCPU_SREG_GS, &vmx->rmode.gs);
1654         fix_pmode_dataseg(VCPU_SREG_FS, &vmx->rmode.fs);
1655
1656         vmcs_write16(GUEST_SS_SELECTOR, 0);
1657         vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
1658
1659         vmcs_write16(GUEST_CS_SELECTOR,
1660                      vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
1661         vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1662 }
1663
1664 static gva_t rmode_tss_base(struct kvm *kvm)
1665 {
1666         if (!kvm->arch.tss_addr) {
1667                 struct kvm_memslots *slots;
1668                 gfn_t base_gfn;
1669
1670                 slots = kvm_memslots(kvm);
1671                 base_gfn = slots->memslots[0].base_gfn +
1672                                  kvm->memslots->memslots[0].npages - 3;
1673                 return base_gfn << PAGE_SHIFT;
1674         }
1675         return kvm->arch.tss_addr;
1676 }
1677
1678 static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
1679 {
1680         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1681
1682         save->selector = vmcs_read16(sf->selector);
1683         save->base = vmcs_readl(sf->base);
1684         save->limit = vmcs_read32(sf->limit);
1685         save->ar = vmcs_read32(sf->ar_bytes);
1686         vmcs_write16(sf->selector, save->base >> 4);
1687         vmcs_write32(sf->base, save->base & 0xfffff);
1688         vmcs_write32(sf->limit, 0xffff);
1689         vmcs_write32(sf->ar_bytes, 0xf3);
1690 }
1691
1692 static void enter_rmode(struct kvm_vcpu *vcpu)
1693 {
1694         unsigned long flags;
1695         struct vcpu_vmx *vmx = to_vmx(vcpu);
1696
1697         if (enable_unrestricted_guest)
1698                 return;
1699
1700         vmx->emulation_required = 1;
1701         vmx->rmode.vm86_active = 1;
1702
1703         vmx->rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
1704         vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
1705
1706         vmx->rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
1707         vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
1708
1709         vmx->rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
1710         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1711
1712         flags = vmcs_readl(GUEST_RFLAGS);
1713         vmx->rmode.save_rflags = flags;
1714
1715         flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1716
1717         vmcs_writel(GUEST_RFLAGS, flags);
1718         vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
1719         update_exception_bitmap(vcpu);
1720
1721         if (emulate_invalid_guest_state)
1722                 goto continue_rmode;
1723
1724         vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
1725         vmcs_write32(GUEST_SS_LIMIT, 0xffff);
1726         vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
1727
1728         vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
1729         vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1730         if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
1731                 vmcs_writel(GUEST_CS_BASE, 0xf0000);
1732         vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
1733
1734         fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.es);
1735         fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.ds);
1736         fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.gs);
1737         fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.fs);
1738
1739 continue_rmode:
1740         kvm_mmu_reset_context(vcpu);
1741         init_rmode(vcpu->kvm);
1742 }
1743
1744 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
1745 {
1746         struct vcpu_vmx *vmx = to_vmx(vcpu);
1747         struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
1748
1749         if (!msr)
1750                 return;
1751
1752         /*
1753          * Force kernel_gs_base reloading before EFER changes, as control
1754          * of this msr depends on is_long_mode().
1755          */
1756         vmx_load_host_state(to_vmx(vcpu));
1757         vcpu->arch.efer = efer;
1758         if (efer & EFER_LMA) {
1759                 vmcs_write32(VM_ENTRY_CONTROLS,
1760                              vmcs_read32(VM_ENTRY_CONTROLS) |
1761                              VM_ENTRY_IA32E_MODE);
1762                 msr->data = efer;
1763         } else {
1764                 vmcs_write32(VM_ENTRY_CONTROLS,
1765                              vmcs_read32(VM_ENTRY_CONTROLS) &
1766                              ~VM_ENTRY_IA32E_MODE);
1767
1768                 msr->data = efer & ~EFER_LME;
1769         }
1770         setup_msrs(vmx);
1771 }
1772
1773 #ifdef CONFIG_X86_64
1774
1775 static void enter_lmode(struct kvm_vcpu *vcpu)
1776 {
1777         u32 guest_tr_ar;
1778
1779         guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
1780         if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
1781                 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
1782                        __func__);
1783                 vmcs_write32(GUEST_TR_AR_BYTES,
1784                              (guest_tr_ar & ~AR_TYPE_MASK)
1785                              | AR_TYPE_BUSY_64_TSS);
1786         }
1787         vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
1788 }
1789
1790 static void exit_lmode(struct kvm_vcpu *vcpu)
1791 {
1792         vmcs_write32(VM_ENTRY_CONTROLS,
1793                      vmcs_read32(VM_ENTRY_CONTROLS)
1794                      & ~VM_ENTRY_IA32E_MODE);
1795         vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
1796 }
1797
1798 #endif
1799
1800 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
1801 {
1802         vpid_sync_context(to_vmx(vcpu));
1803         if (enable_ept) {
1804                 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
1805                         return;
1806                 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
1807         }
1808 }
1809
1810 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
1811 {
1812         ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
1813
1814         vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
1815         vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
1816 }
1817
1818 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1819 {
1820         ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
1821
1822         vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
1823         vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
1824 }
1825
1826 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
1827 {
1828         if (!test_bit(VCPU_EXREG_PDPTR,
1829                       (unsigned long *)&vcpu->arch.regs_dirty))
1830                 return;
1831
1832         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1833                 vmcs_write64(GUEST_PDPTR0, vcpu->arch.mmu.pdptrs[0]);
1834                 vmcs_write64(GUEST_PDPTR1, vcpu->arch.mmu.pdptrs[1]);
1835                 vmcs_write64(GUEST_PDPTR2, vcpu->arch.mmu.pdptrs[2]);
1836                 vmcs_write64(GUEST_PDPTR3, vcpu->arch.mmu.pdptrs[3]);
1837         }
1838 }
1839
1840 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
1841 {
1842         if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1843                 vcpu->arch.mmu.pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
1844                 vcpu->arch.mmu.pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
1845                 vcpu->arch.mmu.pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
1846                 vcpu->arch.mmu.pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
1847         }
1848
1849         __set_bit(VCPU_EXREG_PDPTR,
1850                   (unsigned long *)&vcpu->arch.regs_avail);
1851         __set_bit(VCPU_EXREG_PDPTR,
1852                   (unsigned long *)&vcpu->arch.regs_dirty);
1853 }
1854
1855 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
1856
1857 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
1858                                         unsigned long cr0,
1859                                         struct kvm_vcpu *vcpu)
1860 {
1861         if (!(cr0 & X86_CR0_PG)) {
1862                 /* From paging/starting to nonpaging */
1863                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1864                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
1865                              (CPU_BASED_CR3_LOAD_EXITING |
1866                               CPU_BASED_CR3_STORE_EXITING));
1867                 vcpu->arch.cr0 = cr0;
1868                 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
1869         } else if (!is_paging(vcpu)) {
1870                 /* From nonpaging to paging */
1871                 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1872                              vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
1873                              ~(CPU_BASED_CR3_LOAD_EXITING |
1874                                CPU_BASED_CR3_STORE_EXITING));
1875                 vcpu->arch.cr0 = cr0;
1876                 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
1877         }
1878
1879         if (!(cr0 & X86_CR0_WP))
1880                 *hw_cr0 &= ~X86_CR0_WP;
1881 }
1882
1883 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1884 {
1885         struct vcpu_vmx *vmx = to_vmx(vcpu);
1886         unsigned long hw_cr0;
1887
1888         if (enable_unrestricted_guest)
1889                 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST)
1890                         | KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
1891         else
1892                 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON;
1893
1894         if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
1895                 enter_pmode(vcpu);
1896
1897         if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
1898                 enter_rmode(vcpu);
1899
1900 #ifdef CONFIG_X86_64
1901         if (vcpu->arch.efer & EFER_LME) {
1902                 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
1903                         enter_lmode(vcpu);
1904                 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
1905                         exit_lmode(vcpu);
1906         }
1907 #endif
1908
1909         if (enable_ept)
1910                 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
1911
1912         if (!vcpu->fpu_active)
1913                 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
1914
1915         vmcs_writel(CR0_READ_SHADOW, cr0);
1916         vmcs_writel(GUEST_CR0, hw_cr0);
1917         vcpu->arch.cr0 = cr0;
1918 }
1919
1920 static u64 construct_eptp(unsigned long root_hpa)
1921 {
1922         u64 eptp;
1923
1924         /* TODO write the value reading from MSR */
1925         eptp = VMX_EPT_DEFAULT_MT |
1926                 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
1927         eptp |= (root_hpa & PAGE_MASK);
1928
1929         return eptp;
1930 }
1931
1932 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1933 {
1934         unsigned long guest_cr3;
1935         u64 eptp;
1936
1937         guest_cr3 = cr3;
1938         if (enable_ept) {
1939                 eptp = construct_eptp(cr3);
1940                 vmcs_write64(EPT_POINTER, eptp);
1941                 guest_cr3 = is_paging(vcpu) ? vcpu->arch.cr3 :
1942                         vcpu->kvm->arch.ept_identity_map_addr;
1943                 ept_load_pdptrs(vcpu);
1944         }
1945
1946         vmx_flush_tlb(vcpu);
1947         vmcs_writel(GUEST_CR3, guest_cr3);
1948 }
1949
1950 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1951 {
1952         unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
1953                     KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
1954
1955         vcpu->arch.cr4 = cr4;
1956         if (enable_ept) {
1957                 if (!is_paging(vcpu)) {
1958                         hw_cr4 &= ~X86_CR4_PAE;
1959                         hw_cr4 |= X86_CR4_PSE;
1960                 } else if (!(cr4 & X86_CR4_PAE)) {
1961                         hw_cr4 &= ~X86_CR4_PAE;
1962                 }
1963         }
1964
1965         vmcs_writel(CR4_READ_SHADOW, cr4);
1966         vmcs_writel(GUEST_CR4, hw_cr4);
1967 }
1968
1969 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1970 {
1971         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1972
1973         return vmcs_readl(sf->base);
1974 }
1975
1976 static void vmx_get_segment(struct kvm_vcpu *vcpu,
1977                             struct kvm_segment *var, int seg)
1978 {
1979         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1980         u32 ar;
1981
1982         var->base = vmcs_readl(sf->base);
1983         var->limit = vmcs_read32(sf->limit);
1984         var->selector = vmcs_read16(sf->selector);
1985         ar = vmcs_read32(sf->ar_bytes);
1986         if ((ar & AR_UNUSABLE_MASK) && !emulate_invalid_guest_state)
1987                 ar = 0;
1988         var->type = ar & 15;
1989         var->s = (ar >> 4) & 1;
1990         var->dpl = (ar >> 5) & 3;
1991         var->present = (ar >> 7) & 1;
1992         var->avl = (ar >> 12) & 1;
1993         var->l = (ar >> 13) & 1;
1994         var->db = (ar >> 14) & 1;
1995         var->g = (ar >> 15) & 1;
1996         var->unusable = (ar >> 16) & 1;
1997 }
1998
1999 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
2000 {
2001         if (!is_protmode(vcpu))
2002                 return 0;
2003
2004         if (vmx_get_rflags(vcpu) & X86_EFLAGS_VM) /* if virtual 8086 */
2005                 return 3;
2006
2007         return vmcs_read16(GUEST_CS_SELECTOR) & 3;
2008 }
2009
2010 static u32 vmx_segment_access_rights(struct kvm_segment *var)
2011 {
2012         u32 ar;
2013
2014         if (var->unusable)
2015                 ar = 1 << 16;
2016         else {
2017                 ar = var->type & 15;
2018                 ar |= (var->s & 1) << 4;
2019                 ar |= (var->dpl & 3) << 5;
2020                 ar |= (var->present & 1) << 7;
2021                 ar |= (var->avl & 1) << 12;
2022                 ar |= (var->l & 1) << 13;
2023                 ar |= (var->db & 1) << 14;
2024                 ar |= (var->g & 1) << 15;
2025         }
2026         if (ar == 0) /* a 0 value means unusable */
2027                 ar = AR_UNUSABLE_MASK;
2028
2029         return ar;
2030 }
2031
2032 static void vmx_set_segment(struct kvm_vcpu *vcpu,
2033                             struct kvm_segment *var, int seg)
2034 {
2035         struct vcpu_vmx *vmx = to_vmx(vcpu);
2036         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2037         u32 ar;
2038
2039         if (vmx->rmode.vm86_active && seg == VCPU_SREG_TR) {
2040                 vmx->rmode.tr.selector = var->selector;
2041                 vmx->rmode.tr.base = var->base;
2042                 vmx->rmode.tr.limit = var->limit;
2043                 vmx->rmode.tr.ar = vmx_segment_access_rights(var);
2044                 return;
2045         }
2046         vmcs_writel(sf->base, var->base);
2047         vmcs_write32(sf->limit, var->limit);
2048         vmcs_write16(sf->selector, var->selector);
2049         if (vmx->rmode.vm86_active && var->s) {
2050                 /*
2051                  * Hack real-mode segments into vm86 compatibility.
2052                  */
2053                 if (var->base == 0xffff0000 && var->selector == 0xf000)
2054                         vmcs_writel(sf->base, 0xf0000);
2055                 ar = 0xf3;
2056         } else
2057                 ar = vmx_segment_access_rights(var);
2058
2059         /*
2060          *   Fix the "Accessed" bit in AR field of segment registers for older
2061          * qemu binaries.
2062          *   IA32 arch specifies that at the time of processor reset the
2063          * "Accessed" bit in the AR field of segment registers is 1. And qemu
2064          * is setting it to 0 in the usedland code. This causes invalid guest
2065          * state vmexit when "unrestricted guest" mode is turned on.
2066          *    Fix for this setup issue in cpu_reset is being pushed in the qemu
2067          * tree. Newer qemu binaries with that qemu fix would not need this
2068          * kvm hack.
2069          */
2070         if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
2071                 ar |= 0x1; /* Accessed */
2072
2073         vmcs_write32(sf->ar_bytes, ar);
2074 }
2075
2076 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2077 {
2078         u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
2079
2080         *db = (ar >> 14) & 1;
2081         *l = (ar >> 13) & 1;
2082 }
2083
2084 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2085 {
2086         dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
2087         dt->address = vmcs_readl(GUEST_IDTR_BASE);
2088 }
2089
2090 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2091 {
2092         vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
2093         vmcs_writel(GUEST_IDTR_BASE, dt->address);
2094 }
2095
2096 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2097 {
2098         dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
2099         dt->address = vmcs_readl(GUEST_GDTR_BASE);
2100 }
2101
2102 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2103 {
2104         vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
2105         vmcs_writel(GUEST_GDTR_BASE, dt->address);
2106 }
2107
2108 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
2109 {
2110         struct kvm_segment var;
2111         u32 ar;
2112
2113         vmx_get_segment(vcpu, &var, seg);
2114         ar = vmx_segment_access_rights(&var);
2115
2116         if (var.base != (var.selector << 4))
2117                 return false;
2118         if (var.limit != 0xffff)
2119                 return false;
2120         if (ar != 0xf3)
2121                 return false;
2122
2123         return true;
2124 }
2125
2126 static bool code_segment_valid(struct kvm_vcpu *vcpu)
2127 {
2128         struct kvm_segment cs;
2129         unsigned int cs_rpl;
2130
2131         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
2132         cs_rpl = cs.selector & SELECTOR_RPL_MASK;
2133
2134         if (cs.unusable)
2135                 return false;
2136         if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
2137                 return false;
2138         if (!cs.s)
2139                 return false;
2140         if (cs.type & AR_TYPE_WRITEABLE_MASK) {
2141                 if (cs.dpl > cs_rpl)
2142                         return false;
2143         } else {
2144                 if (cs.dpl != cs_rpl)
2145                         return false;
2146         }
2147         if (!cs.present)
2148                 return false;
2149
2150         /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
2151         return true;
2152 }
2153
2154 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
2155 {
2156         struct kvm_segment ss;
2157         unsigned int ss_rpl;
2158
2159         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
2160         ss_rpl = ss.selector & SELECTOR_RPL_MASK;
2161
2162         if (ss.unusable)
2163                 return true;
2164         if (ss.type != 3 && ss.type != 7)
2165                 return false;
2166         if (!ss.s)
2167                 return false;
2168         if (ss.dpl != ss_rpl) /* DPL != RPL */
2169                 return false;
2170         if (!ss.present)
2171                 return false;
2172
2173         return true;
2174 }
2175
2176 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
2177 {
2178         struct kvm_segment var;
2179         unsigned int rpl;
2180
2181         vmx_get_segment(vcpu, &var, seg);
2182         rpl = var.selector & SELECTOR_RPL_MASK;
2183
2184         if (var.unusable)
2185                 return true;
2186         if (!var.s)
2187                 return false;
2188         if (!var.present)
2189                 return false;
2190         if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
2191                 if (var.dpl < rpl) /* DPL < RPL */
2192                         return false;
2193         }
2194
2195         /* TODO: Add other members to kvm_segment_field to allow checking for other access
2196          * rights flags
2197          */
2198         return true;
2199 }
2200
2201 static bool tr_valid(struct kvm_vcpu *vcpu)
2202 {
2203         struct kvm_segment tr;
2204
2205         vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
2206
2207         if (tr.unusable)
2208                 return false;
2209         if (tr.selector & SELECTOR_TI_MASK)     /* TI = 1 */
2210                 return false;
2211         if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
2212                 return false;
2213         if (!tr.present)
2214                 return false;
2215
2216         return true;
2217 }
2218
2219 static bool ldtr_valid(struct kvm_vcpu *vcpu)
2220 {
2221         struct kvm_segment ldtr;
2222
2223         vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
2224
2225         if (ldtr.unusable)
2226                 return true;
2227         if (ldtr.selector & SELECTOR_TI_MASK)   /* TI = 1 */
2228                 return false;
2229         if (ldtr.type != 2)
2230                 return false;
2231         if (!ldtr.present)
2232                 return false;
2233
2234         return true;
2235 }
2236
2237 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
2238 {
2239         struct kvm_segment cs, ss;
2240
2241         vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
2242         vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
2243
2244         return ((cs.selector & SELECTOR_RPL_MASK) ==
2245                  (ss.selector & SELECTOR_RPL_MASK));
2246 }
2247
2248 /*
2249  * Check if guest state is valid. Returns true if valid, false if
2250  * not.
2251  * We assume that registers are always usable
2252  */
2253 static bool guest_state_valid(struct kvm_vcpu *vcpu)
2254 {
2255         /* real mode guest state checks */
2256         if (!is_protmode(vcpu)) {
2257                 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
2258                         return false;
2259                 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
2260                         return false;
2261                 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
2262                         return false;
2263                 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
2264                         return false;
2265                 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
2266                         return false;
2267                 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
2268                         return false;
2269         } else {
2270         /* protected mode guest state checks */
2271                 if (!cs_ss_rpl_check(vcpu))
2272                         return false;
2273                 if (!code_segment_valid(vcpu))
2274                         return false;
2275                 if (!stack_segment_valid(vcpu))
2276                         return false;
2277                 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
2278                         return false;
2279                 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
2280                         return false;
2281                 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
2282                         return false;
2283                 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
2284                         return false;
2285                 if (!tr_valid(vcpu))
2286                         return false;
2287                 if (!ldtr_valid(vcpu))
2288                         return false;
2289         }
2290         /* TODO:
2291          * - Add checks on RIP
2292          * - Add checks on RFLAGS
2293          */
2294
2295         return true;
2296 }
2297
2298 static int init_rmode_tss(struct kvm *kvm)
2299 {
2300         gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
2301         u16 data = 0;
2302         int ret = 0;
2303         int r;
2304
2305         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2306         if (r < 0)
2307                 goto out;
2308         data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
2309         r = kvm_write_guest_page(kvm, fn++, &data,
2310                         TSS_IOPB_BASE_OFFSET, sizeof(u16));
2311         if (r < 0)
2312                 goto out;
2313         r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
2314         if (r < 0)
2315                 goto out;
2316         r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2317         if (r < 0)
2318                 goto out;
2319         data = ~0;
2320         r = kvm_write_guest_page(kvm, fn, &data,
2321                                  RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
2322                                  sizeof(u8));
2323         if (r < 0)
2324                 goto out;
2325
2326         ret = 1;
2327 out:
2328         return ret;
2329 }
2330
2331 static int init_rmode_identity_map(struct kvm *kvm)
2332 {
2333         int i, r, ret;
2334         pfn_t identity_map_pfn;
2335         u32 tmp;
2336
2337         if (!enable_ept)
2338                 return 1;
2339         if (unlikely(!kvm->arch.ept_identity_pagetable)) {
2340                 printk(KERN_ERR "EPT: identity-mapping pagetable "
2341                         "haven't been allocated!\n");
2342                 return 0;
2343         }
2344         if (likely(kvm->arch.ept_identity_pagetable_done))
2345                 return 1;
2346         ret = 0;
2347         identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
2348         r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
2349         if (r < 0)
2350                 goto out;
2351         /* Set up identity-mapping pagetable for EPT in real mode */
2352         for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
2353                 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
2354                         _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
2355                 r = kvm_write_guest_page(kvm, identity_map_pfn,
2356                                 &tmp, i * sizeof(tmp), sizeof(tmp));
2357                 if (r < 0)
2358                         goto out;
2359         }
2360         kvm->arch.ept_identity_pagetable_done = true;
2361         ret = 1;
2362 out:
2363         return ret;
2364 }
2365
2366 static void seg_setup(int seg)
2367 {
2368         struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2369         unsigned int ar;
2370
2371         vmcs_write16(sf->selector, 0);
2372         vmcs_writel(sf->base, 0);
2373         vmcs_write32(sf->limit, 0xffff);
2374         if (enable_unrestricted_guest) {
2375                 ar = 0x93;
2376                 if (seg == VCPU_SREG_CS)
2377                         ar |= 0x08; /* code segment */
2378         } else
2379                 ar = 0xf3;
2380
2381         vmcs_write32(sf->ar_bytes, ar);
2382 }
2383
2384 static int alloc_apic_access_page(struct kvm *kvm)
2385 {
2386         struct kvm_userspace_memory_region kvm_userspace_mem;
2387         int r = 0;
2388
2389         mutex_lock(&kvm->slots_lock);
2390         if (kvm->arch.apic_access_page)
2391                 goto out;
2392         kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
2393         kvm_userspace_mem.flags = 0;
2394         kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
2395         kvm_userspace_mem.memory_size = PAGE_SIZE;
2396         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2397         if (r)
2398                 goto out;
2399
2400         kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
2401 out:
2402         mutex_unlock(&kvm->slots_lock);
2403         return r;
2404 }
2405
2406 static int alloc_identity_pagetable(struct kvm *kvm)
2407 {
2408         struct kvm_userspace_memory_region kvm_userspace_mem;
2409         int r = 0;
2410
2411         mutex_lock(&kvm->slots_lock);
2412         if (kvm->arch.ept_identity_pagetable)
2413                 goto out;
2414         kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
2415         kvm_userspace_mem.flags = 0;
2416         kvm_userspace_mem.guest_phys_addr =
2417                 kvm->arch.ept_identity_map_addr;
2418         kvm_userspace_mem.memory_size = PAGE_SIZE;
2419         r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2420         if (r)
2421                 goto out;
2422
2423         kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
2424                         kvm->arch.ept_identity_map_addr >> PAGE_SHIFT);
2425 out:
2426         mutex_unlock(&kvm->slots_lock);
2427         return r;
2428 }
2429
2430 static void allocate_vpid(struct vcpu_vmx *vmx)
2431 {
2432         int vpid;
2433
2434         vmx->vpid = 0;
2435         if (!enable_vpid)
2436                 return;
2437         spin_lock(&vmx_vpid_lock);
2438         vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
2439         if (vpid < VMX_NR_VPIDS) {
2440                 vmx->vpid = vpid;
2441                 __set_bit(vpid, vmx_vpid_bitmap);
2442         }
2443         spin_unlock(&vmx_vpid_lock);
2444 }
2445
2446 static void free_vpid(struct vcpu_vmx *vmx)
2447 {
2448         if (!enable_vpid)
2449                 return;
2450         spin_lock(&vmx_vpid_lock);
2451         if (vmx->vpid != 0)
2452                 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
2453         spin_unlock(&vmx_vpid_lock);
2454 }
2455
2456 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, u32 msr)
2457 {
2458         int f = sizeof(unsigned long);
2459
2460         if (!cpu_has_vmx_msr_bitmap())
2461                 return;
2462
2463         /*
2464          * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
2465          * have the write-low and read-high bitmap offsets the wrong way round.
2466          * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
2467          */
2468         if (msr <= 0x1fff) {
2469                 __clear_bit(msr, msr_bitmap + 0x000 / f); /* read-low */
2470                 __clear_bit(msr, msr_bitmap + 0x800 / f); /* write-low */
2471         } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
2472                 msr &= 0x1fff;
2473                 __clear_bit(msr, msr_bitmap + 0x400 / f); /* read-high */
2474                 __clear_bit(msr, msr_bitmap + 0xc00 / f); /* write-high */
2475         }
2476 }
2477
2478 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
2479 {
2480         if (!longmode_only)
2481                 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy, msr);
2482         __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode, msr);
2483 }
2484
2485 /*
2486  * Sets up the vmcs for emulated real mode.
2487  */
2488 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
2489 {
2490         u32 host_sysenter_cs, msr_low, msr_high;
2491         u32 junk;
2492         u64 host_pat;
2493         unsigned long a;
2494         struct desc_ptr dt;
2495         int i;
2496         unsigned long kvm_vmx_return;
2497         u32 exec_control;
2498
2499         /* I/O */
2500         vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
2501         vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
2502
2503         if (cpu_has_vmx_msr_bitmap())
2504                 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
2505
2506         vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
2507
2508         /* Control */
2509         vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
2510                 vmcs_config.pin_based_exec_ctrl);
2511
2512         exec_control = vmcs_config.cpu_based_exec_ctrl;
2513         if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
2514                 exec_control &= ~CPU_BASED_TPR_SHADOW;
2515 #ifdef CONFIG_X86_64
2516                 exec_control |= CPU_BASED_CR8_STORE_EXITING |
2517                                 CPU_BASED_CR8_LOAD_EXITING;
2518 #endif
2519         }
2520         if (!enable_ept)
2521                 exec_control |= CPU_BASED_CR3_STORE_EXITING |
2522                                 CPU_BASED_CR3_LOAD_EXITING  |
2523                                 CPU_BASED_INVLPG_EXITING;
2524         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
2525
2526         if (cpu_has_secondary_exec_ctrls()) {
2527                 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
2528                 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2529                         exec_control &=
2530                                 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
2531                 if (vmx->vpid == 0)
2532                         exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
2533                 if (!enable_ept) {
2534                         exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
2535                         enable_unrestricted_guest = 0;
2536                 }
2537                 if (!enable_unrestricted_guest)
2538                         exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
2539                 if (!ple_gap)
2540                         exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
2541                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
2542         }
2543
2544         if (ple_gap) {
2545                 vmcs_write32(PLE_GAP, ple_gap);
2546                 vmcs_write32(PLE_WINDOW, ple_window);
2547         }
2548
2549         vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
2550         vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
2551         vmcs_write32(CR3_TARGET_COUNT, 0);           /* 22.2.1 */
2552
2553         vmcs_writel(HOST_CR0, read_cr0() | X86_CR0_TS);  /* 22.2.3 */
2554         vmcs_writel(HOST_CR4, read_cr4());  /* 22.2.3, 22.2.5 */
2555         vmcs_writel(HOST_CR3, read_cr3());  /* 22.2.3  FIXME: shadow tables */
2556
2557         vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
2558         vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
2559         vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
2560         vmcs_write16(HOST_FS_SELECTOR, 0);            /* 22.2.4 */
2561         vmcs_write16(HOST_GS_SELECTOR, 0);            /* 22.2.4 */
2562         vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS);  /* 22.2.4 */
2563 #ifdef CONFIG_X86_64
2564         rdmsrl(MSR_FS_BASE, a);
2565         vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
2566         rdmsrl(MSR_GS_BASE, a);
2567         vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
2568 #else
2569         vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
2570         vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
2571 #endif
2572
2573         vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8);  /* 22.2.4 */
2574
2575         native_store_idt(&dt);
2576         vmcs_writel(HOST_IDTR_BASE, dt.address);   /* 22.2.4 */
2577
2578         asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
2579         vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
2580         vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
2581         vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
2582         vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
2583         vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
2584         vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
2585
2586         rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
2587         vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
2588         rdmsrl(MSR_IA32_SYSENTER_ESP, a);
2589         vmcs_writel(HOST_IA32_SYSENTER_ESP, a);   /* 22.2.3 */
2590         rdmsrl(MSR_IA32_SYSENTER_EIP, a);
2591         vmcs_writel(HOST_IA32_SYSENTER_EIP, a);   /* 22.2.3 */
2592
2593         if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
2594                 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2595                 host_pat = msr_low | ((u64) msr_high << 32);
2596                 vmcs_write64(HOST_IA32_PAT, host_pat);
2597         }
2598         if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2599                 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2600                 host_pat = msr_low | ((u64) msr_high << 32);
2601                 /* Write the default value follow host pat */
2602                 vmcs_write64(GUEST_IA32_PAT, host_pat);
2603                 /* Keep arch.pat sync with GUEST_IA32_PAT */
2604                 vmx->vcpu.arch.pat = host_pat;
2605         }
2606
2607         for (i = 0; i < NR_VMX_MSR; ++i) {
2608                 u32 index = vmx_msr_index[i];
2609                 u32 data_low, data_high;
2610                 int j = vmx->nmsrs;
2611
2612                 if (rdmsr_safe(index, &data_low, &data_high) < 0)
2613                         continue;
2614                 if (wrmsr_safe(index, data_low, data_high) < 0)
2615                         continue;
2616                 vmx->guest_msrs[j].index = i;
2617                 vmx->guest_msrs[j].data = 0;
2618                 vmx->guest_msrs[j].mask = -1ull;
2619                 ++vmx->nmsrs;
2620         }
2621
2622         vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
2623
2624         /* 22.2.1, 20.8.1 */
2625         vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
2626
2627         vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
2628         vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
2629         if (enable_ept)
2630                 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
2631         vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
2632
2633         kvm_write_tsc(&vmx->vcpu, 0);
2634
2635         return 0;
2636 }
2637
2638 static int init_rmode(struct kvm *kvm)
2639 {
2640         int idx, ret = 0;
2641
2642         idx = srcu_read_lock(&kvm->srcu);
2643         if (!init_rmode_tss(kvm))
2644                 goto exit;
2645         if (!init_rmode_identity_map(kvm))
2646                 goto exit;
2647
2648         ret = 1;
2649 exit:
2650         srcu_read_unlock(&kvm->srcu, idx);
2651         return ret;
2652 }
2653
2654 static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
2655 {
2656         struct vcpu_vmx *vmx = to_vmx(vcpu);
2657         u64 msr;
2658         int ret;
2659
2660         vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
2661         if (!init_rmode(vmx->vcpu.kvm)) {
2662                 ret = -ENOMEM;
2663                 goto out;
2664         }
2665
2666         vmx->rmode.vm86_active = 0;
2667
2668         vmx->soft_vnmi_blocked = 0;
2669
2670         vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
2671         kvm_set_cr8(&vmx->vcpu, 0);
2672         msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
2673         if (kvm_vcpu_is_bsp(&vmx->vcpu))
2674                 msr |= MSR_IA32_APICBASE_BSP;
2675         kvm_set_apic_base(&vmx->vcpu, msr);
2676
2677         ret = fx_init(&vmx->vcpu);
2678         if (ret != 0)
2679                 goto out;
2680
2681         seg_setup(VCPU_SREG_CS);
2682         /*
2683          * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
2684          * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4.  Sigh.
2685          */
2686         if (kvm_vcpu_is_bsp(&vmx->vcpu)) {
2687                 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
2688                 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
2689         } else {
2690                 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
2691                 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
2692         }
2693
2694         seg_setup(VCPU_SREG_DS);
2695         seg_setup(VCPU_SREG_ES);
2696         seg_setup(VCPU_SREG_FS);
2697         seg_setup(VCPU_SREG_GS);
2698         seg_setup(VCPU_SREG_SS);
2699
2700         vmcs_write16(GUEST_TR_SELECTOR, 0);
2701         vmcs_writel(GUEST_TR_BASE, 0);
2702         vmcs_write32(GUEST_TR_LIMIT, 0xffff);
2703         vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2704
2705         vmcs_write16(GUEST_LDTR_SELECTOR, 0);
2706         vmcs_writel(GUEST_LDTR_BASE, 0);
2707         vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
2708         vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
2709
2710         vmcs_write32(GUEST_SYSENTER_CS, 0);
2711         vmcs_writel(GUEST_SYSENTER_ESP, 0);
2712         vmcs_writel(GUEST_SYSENTER_EIP, 0);
2713
2714         vmcs_writel(GUEST_RFLAGS, 0x02);
2715         if (kvm_vcpu_is_bsp(&vmx->vcpu))
2716                 kvm_rip_write(vcpu, 0xfff0);
2717         else
2718                 kvm_rip_write(vcpu, 0);
2719         kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
2720
2721         vmcs_writel(GUEST_DR7, 0x400);
2722
2723         vmcs_writel(GUEST_GDTR_BASE, 0);
2724         vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
2725
2726         vmcs_writel(GUEST_IDTR_BASE, 0);
2727         vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
2728
2729         vmcs_write32(GUEST_ACTIVITY_STATE, 0);
2730         vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
2731         vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
2732
2733         /* Special registers */
2734         vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
2735
2736         setup_msrs(vmx);
2737
2738         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);  /* 22.2.1 */
2739
2740         if (cpu_has_vmx_tpr_shadow()) {
2741                 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
2742                 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
2743                         vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
2744                                 page_to_phys(vmx->vcpu.arch.apic->regs_page));
2745                 vmcs_write32(TPR_THRESHOLD, 0);
2746         }
2747
2748         if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2749                 vmcs_write64(APIC_ACCESS_ADDR,
2750                              page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
2751
2752         if (vmx->vpid != 0)
2753                 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2754
2755         vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
2756         vmx_set_cr0(&vmx->vcpu, kvm_read_cr0(vcpu)); /* enter rmode */
2757         vmx_set_cr4(&vmx->vcpu, 0);
2758         vmx_set_efer(&vmx->vcpu, 0);
2759         vmx_fpu_activate(&vmx->vcpu);
2760         update_exception_bitmap(&vmx->vcpu);
2761
2762         vpid_sync_context(vmx);
2763
2764         ret = 0;
2765
2766         /* HACK: Don't enable emulation on guest boot/reset */
2767         vmx->emulation_required = 0;
2768
2769 out:
2770         return ret;
2771 }
2772
2773 static void enable_irq_window(struct kvm_vcpu *vcpu)
2774 {
2775         u32 cpu_based_vm_exec_control;
2776
2777         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2778         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2779         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2780 }
2781
2782 static void enable_nmi_window(struct kvm_vcpu *vcpu)
2783 {
2784         u32 cpu_based_vm_exec_control;
2785
2786         if (!cpu_has_virtual_nmis()) {
2787                 enable_irq_window(vcpu);
2788                 return;
2789         }
2790
2791         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2792         cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
2793         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2794 }
2795
2796 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
2797 {
2798         struct vcpu_vmx *vmx = to_vmx(vcpu);
2799         uint32_t intr;
2800         int irq = vcpu->arch.interrupt.nr;
2801
2802         trace_kvm_inj_virq(irq);
2803
2804         ++vcpu->stat.irq_injections;
2805         if (vmx->rmode.vm86_active) {
2806                 if (kvm_inject_realmode_interrupt(vcpu, irq) != EMULATE_DONE)
2807                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2808                 return;
2809         }
2810         intr = irq | INTR_INFO_VALID_MASK;
2811         if (vcpu->arch.interrupt.soft) {
2812                 intr |= INTR_TYPE_SOFT_INTR;
2813                 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2814                              vmx->vcpu.arch.event_exit_inst_len);
2815         } else
2816                 intr |= INTR_TYPE_EXT_INTR;
2817         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
2818 }
2819
2820 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
2821 {
2822         struct vcpu_vmx *vmx = to_vmx(vcpu);
2823
2824         if (!cpu_has_virtual_nmis()) {
2825                 /*
2826                  * Tracking the NMI-blocked state in software is built upon
2827                  * finding the next open IRQ window. This, in turn, depends on
2828                  * well-behaving guests: They have to keep IRQs disabled at
2829                  * least as long as the NMI handler runs. Otherwise we may
2830                  * cause NMI nesting, maybe breaking the guest. But as this is
2831                  * highly unlikely, we can live with the residual risk.
2832                  */
2833                 vmx->soft_vnmi_blocked = 1;
2834                 vmx->vnmi_blocked_time = 0;
2835         }
2836
2837         ++vcpu->stat.nmi_injections;
2838         if (vmx->rmode.vm86_active) {
2839                 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR) != EMULATE_DONE)
2840                         kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2841                 return;
2842         }
2843         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2844                         INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
2845 }
2846
2847 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
2848 {
2849         if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
2850                 return 0;
2851
2852         return  !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2853                         (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_NMI));
2854 }
2855
2856 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
2857 {
2858         if (!cpu_has_virtual_nmis())
2859                 return to_vmx(vcpu)->soft_vnmi_blocked;
2860         return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
2861 }
2862
2863 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
2864 {
2865         struct vcpu_vmx *vmx = to_vmx(vcpu);
2866
2867         if (!cpu_has_virtual_nmis()) {
2868                 if (vmx->soft_vnmi_blocked != masked) {
2869                         vmx->soft_vnmi_blocked = masked;
2870                         vmx->vnmi_blocked_time = 0;
2871                 }
2872         } else {
2873                 if (masked)
2874                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
2875                                       GUEST_INTR_STATE_NMI);
2876                 else
2877                         vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
2878                                         GUEST_INTR_STATE_NMI);
2879         }
2880 }
2881
2882 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
2883 {
2884         return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
2885                 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2886                         (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
2887 }
2888
2889 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
2890 {
2891         int ret;
2892         struct kvm_userspace_memory_region tss_mem = {
2893                 .slot = TSS_PRIVATE_MEMSLOT,
2894                 .guest_phys_addr = addr,
2895                 .memory_size = PAGE_SIZE * 3,
2896                 .flags = 0,
2897         };
2898
2899         ret = kvm_set_memory_region(kvm, &tss_mem, 0);
2900         if (ret)
2901                 return ret;
2902         kvm->arch.tss_addr = addr;
2903         return 0;
2904 }
2905
2906 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
2907                                   int vec, u32 err_code)
2908 {
2909         /*
2910          * Instruction with address size override prefix opcode 0x67
2911          * Cause the #SS fault with 0 error code in VM86 mode.
2912          */
2913         if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
2914                 if (emulate_instruction(vcpu, 0, 0, 0) == EMULATE_DONE)
2915                         return 1;
2916         /*
2917          * Forward all other exceptions that are valid in real mode.
2918          * FIXME: Breaks guest debugging in real mode, needs to be fixed with
2919          *        the required debugging infrastructure rework.
2920          */
2921         switch (vec) {
2922         case DB_VECTOR:
2923                 if (vcpu->guest_debug &
2924                     (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
2925                         return 0;
2926                 kvm_queue_exception(vcpu, vec);
2927                 return 1;
2928         case BP_VECTOR:
2929                 /*
2930                  * Update instruction length as we may reinject the exception
2931                  * from user space while in guest debugging mode.
2932                  */
2933                 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
2934                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
2935                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
2936                         return 0;
2937                 /* fall through */
2938         case DE_VECTOR:
2939         case OF_VECTOR:
2940         case BR_VECTOR:
2941         case UD_VECTOR:
2942         case DF_VECTOR:
2943         case SS_VECTOR:
2944         case GP_VECTOR:
2945         case MF_VECTOR:
2946                 kvm_queue_exception(vcpu, vec);
2947                 return 1;
2948         }
2949         return 0;
2950 }
2951
2952 /*
2953  * Trigger machine check on the host. We assume all the MSRs are already set up
2954  * by the CPU and that we still run on the same CPU as the MCE occurred on.
2955  * We pass a fake environment to the machine check handler because we want
2956  * the guest to be always treated like user space, no matter what context
2957  * it used internally.
2958  */
2959 static void kvm_machine_check(void)
2960 {
2961 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
2962         struct pt_regs regs = {
2963                 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
2964                 .flags = X86_EFLAGS_IF,
2965         };
2966
2967         do_machine_check(&regs, 0);
2968 #endif
2969 }
2970
2971 static int handle_machine_check(struct kvm_vcpu *vcpu)
2972 {
2973         /* already handled by vcpu_run */
2974         return 1;
2975 }
2976
2977 static int handle_exception(struct kvm_vcpu *vcpu)
2978 {
2979         struct vcpu_vmx *vmx = to_vmx(vcpu);
2980         struct kvm_run *kvm_run = vcpu->run;
2981         u32 intr_info, ex_no, error_code;
2982         unsigned long cr2, rip, dr6;
2983         u32 vect_info;
2984         enum emulation_result er;
2985
2986         vect_info = vmx->idt_vectoring_info;
2987         intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
2988
2989         if (is_machine_check(intr_info))
2990                 return handle_machine_check(vcpu);
2991
2992         if ((vect_info & VECTORING_INFO_VALID_MASK) &&
2993             !is_page_fault(intr_info)) {
2994                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2995                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
2996                 vcpu->run->internal.ndata = 2;
2997                 vcpu->run->internal.data[0] = vect_info;
2998                 vcpu->run->internal.data[1] = intr_info;
2999                 return 0;
3000         }
3001
3002         if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
3003                 return 1;  /* already handled by vmx_vcpu_run() */
3004
3005         if (is_no_device(intr_info)) {
3006                 vmx_fpu_activate(vcpu);
3007                 return 1;
3008         }
3009
3010         if (is_invalid_opcode(intr_info)) {
3011                 er = emulate_instruction(vcpu, 0, 0, EMULTYPE_TRAP_UD);
3012                 if (er != EMULATE_DONE)
3013                         kvm_queue_exception(vcpu, UD_VECTOR);
3014                 return 1;
3015         }
3016
3017         error_code = 0;
3018         rip = kvm_rip_read(vcpu);
3019         if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
3020                 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
3021         if (is_page_fault(intr_info)) {
3022                 /* EPT won't cause page fault directly */
3023                 if (enable_ept)
3024                         BUG();
3025                 cr2 = vmcs_readl(EXIT_QUALIFICATION);
3026                 trace_kvm_page_fault(cr2, error_code);
3027
3028                 if (kvm_event_needs_reinjection(vcpu))
3029                         kvm_mmu_unprotect_page_virt(vcpu, cr2);
3030                 return kvm_mmu_page_fault(vcpu, cr2, error_code);
3031         }
3032
3033         if (vmx->rmode.vm86_active &&
3034             handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
3035                                                                 error_code)) {
3036                 if (vcpu->arch.halt_request) {
3037                         vcpu->arch.halt_request = 0;
3038                         return kvm_emulate_halt(vcpu);
3039                 }
3040                 return 1;
3041         }
3042
3043         ex_no = intr_info & INTR_INFO_VECTOR_MASK;
3044         switch (ex_no) {
3045         case DB_VECTOR:
3046                 dr6 = vmcs_readl(EXIT_QUALIFICATION);
3047                 if (!(vcpu->guest_debug &
3048                       (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
3049                         vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
3050                         kvm_queue_exception(vcpu, DB_VECTOR);
3051                         return 1;
3052                 }
3053                 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
3054                 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
3055                 /* fall through */
3056         case BP_VECTOR:
3057                 /*
3058                  * Update instruction length as we may reinject #BP from
3059                  * user space while in guest debugging mode. Reading it for
3060                  * #DB as well causes no harm, it is not used in that case.
3061                  */
3062                 vmx->vcpu.arch.event_exit_inst_len =
3063                         vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3064                 kvm_run->exit_reason = KVM_EXIT_DEBUG;
3065                 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
3066                 kvm_run->debug.arch.exception = ex_no;
3067                 break;
3068         default:
3069                 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
3070                 kvm_run->ex.exception = ex_no;
3071                 kvm_run->ex.error_code = error_code;
3072                 break;
3073         }
3074         return 0;
3075 }
3076
3077 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
3078 {
3079         ++vcpu->stat.irq_exits;
3080         return 1;
3081 }
3082
3083 static int handle_triple_fault(struct kvm_vcpu *vcpu)
3084 {
3085         vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
3086         return 0;
3087 }
3088
3089 static int handle_io(struct kvm_vcpu *vcpu)
3090 {
3091         unsigned long exit_qualification;
3092         int size, in, string;
3093         unsigned port;
3094
3095         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3096         string = (exit_qualification & 16) != 0;
3097         in = (exit_qualification & 8) != 0;
3098
3099         ++vcpu->stat.io_exits;
3100
3101         if (string || in)
3102                 return emulate_instruction(vcpu, 0, 0, 0) == EMULATE_DONE;
3103
3104         port = exit_qualification >> 16;
3105         size = (exit_qualification & 7) + 1;
3106         skip_emulated_instruction(vcpu);
3107
3108         return kvm_fast_pio_out(vcpu, size, port);
3109 }
3110
3111 static void
3112 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
3113 {
3114         /*
3115          * Patch in the VMCALL instruction:
3116          */
3117         hypercall[0] = 0x0f;
3118         hypercall[1] = 0x01;
3119         hypercall[2] = 0xc1;
3120 }
3121
3122 static void complete_insn_gp(struct kvm_vcpu *vcpu, int err)
3123 {
3124         if (err)
3125                 kvm_inject_gp(vcpu, 0);
3126         else
3127                 skip_emulated_instruction(vcpu);
3128 }
3129
3130 static int handle_cr(struct kvm_vcpu *vcpu)
3131 {
3132         unsigned long exit_qualification, val;
3133         int cr;
3134         int reg;
3135         int err;
3136
3137         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3138         cr = exit_qualification & 15;
3139         reg = (exit_qualification >> 8) & 15;
3140         switch ((exit_qualification >> 4) & 3) {
3141         case 0: /* mov to cr */
3142                 val = kvm_register_read(vcpu, reg);
3143                 trace_kvm_cr_write(cr, val);
3144                 switch (cr) {
3145                 case 0:
3146                         err = kvm_set_cr0(vcpu, val);
3147                         complete_insn_gp(vcpu, err);
3148                         return 1;
3149                 case 3:
3150                         err = kvm_set_cr3(vcpu, val);
3151                         complete_insn_gp(vcpu, err);
3152                         return 1;
3153                 case 4:
3154                         err = kvm_set_cr4(vcpu, val);
3155                         complete_insn_gp(vcpu, err);
3156                         return 1;
3157                 case 8: {
3158                                 u8 cr8_prev = kvm_get_cr8(vcpu);
3159                                 u8 cr8 = kvm_register_read(vcpu, reg);
3160                                 kvm_set_cr8(vcpu, cr8);
3161                                 skip_emulated_instruction(vcpu);
3162                                 if (irqchip_in_kernel(vcpu->kvm))
3163                                         return 1;
3164                                 if (cr8_prev <= cr8)
3165                                         return 1;
3166                                 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
3167                                 return 0;
3168                         }
3169                 };
3170                 break;
3171         case 2: /* clts */
3172                 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
3173                 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
3174                 skip_emulated_instruction(vcpu);
3175                 vmx_fpu_activate(vcpu);
3176                 return 1;
3177         case 1: /*mov from cr*/
3178                 switch (cr) {
3179                 case 3:
3180                         kvm_register_write(vcpu, reg, vcpu->arch.cr3);
3181                         trace_kvm_cr_read(cr, vcpu->arch.cr3);
3182                         skip_emulated_instruction(vcpu);
3183                         return 1;
3184                 case 8:
3185                         val = kvm_get_cr8(vcpu);
3186                         kvm_register_write(vcpu, reg, val);
3187                         trace_kvm_cr_read(cr, val);
3188                         skip_emulated_instruction(vcpu);
3189                         return 1;
3190                 }
3191                 break;
3192         case 3: /* lmsw */
3193                 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
3194                 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
3195                 kvm_lmsw(vcpu, val);
3196
3197                 skip_emulated_instruction(vcpu);
3198                 return 1;
3199         default:
3200                 break;
3201         }
3202         vcpu->run->exit_reason = 0;
3203         pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
3204                (int)(exit_qualification >> 4) & 3, cr);
3205         return 0;
3206 }
3207
3208 static int handle_dr(struct kvm_vcpu *vcpu)
3209 {
3210         unsigned long exit_qualification;
3211         int dr, reg;
3212
3213         /* Do not handle if the CPL > 0, will trigger GP on re-entry */
3214         if (!kvm_require_cpl(vcpu, 0))
3215                 return 1;
3216         dr = vmcs_readl(GUEST_DR7);
3217         if (dr & DR7_GD) {
3218                 /*
3219                  * As the vm-exit takes precedence over the debug trap, we
3220                  * need to emulate the latter, either for the host or the
3221                  * guest debugging itself.
3222                  */
3223                 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
3224                         vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
3225                         vcpu->run->debug.arch.dr7 = dr;
3226                         vcpu->run->debug.arch.pc =
3227                                 vmcs_readl(GUEST_CS_BASE) +
3228                                 vmcs_readl(GUEST_RIP);
3229                         vcpu->run->debug.arch.exception = DB_VECTOR;
3230                         vcpu->run->exit_reason = KVM_EXIT_DEBUG;
3231                         return 0;
3232                 } else {
3233                         vcpu->arch.dr7 &= ~DR7_GD;
3234                         vcpu->arch.dr6 |= DR6_BD;
3235                         vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
3236                         kvm_queue_exception(vcpu, DB_VECTOR);
3237                         return 1;
3238                 }
3239         }
3240
3241         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3242         dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
3243         reg = DEBUG_REG_ACCESS_REG(exit_qualification);
3244         if (exit_qualification & TYPE_MOV_FROM_DR) {
3245                 unsigned long val;
3246                 if (!kvm_get_dr(vcpu, dr, &val))
3247                         kvm_register_write(vcpu, reg, val);
3248         } else
3249                 kvm_set_dr(vcpu, dr, vcpu->arch.regs[reg]);
3250         skip_emulated_instruction(vcpu);
3251         return 1;
3252 }
3253
3254 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
3255 {
3256         vmcs_writel(GUEST_DR7, val);
3257 }
3258
3259 static int handle_cpuid(struct kvm_vcpu *vcpu)
3260 {
3261         kvm_emulate_cpuid(vcpu);
3262         return 1;
3263 }
3264
3265 static int handle_rdmsr(struct kvm_vcpu *vcpu)
3266 {
3267         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3268         u64 data;
3269
3270         if (vmx_get_msr(vcpu, ecx, &data)) {
3271                 trace_kvm_msr_read_ex(ecx);
3272                 kvm_inject_gp(vcpu, 0);
3273                 return 1;
3274         }
3275
3276         trace_kvm_msr_read(ecx, data);
3277
3278         /* FIXME: handling of bits 32:63 of rax, rdx */
3279         vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
3280         vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
3281         skip_emulated_instruction(vcpu);
3282         return 1;
3283 }
3284
3285 static int handle_wrmsr(struct kvm_vcpu *vcpu)
3286 {
3287         u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3288         u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
3289                 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
3290
3291         if (vmx_set_msr(vcpu, ecx, data) != 0) {
3292                 trace_kvm_msr_write_ex(ecx, data);
3293                 kvm_inject_gp(vcpu, 0);
3294                 return 1;
3295         }
3296
3297         trace_kvm_msr_write(ecx, data);
3298         skip_emulated_instruction(vcpu);
3299         return 1;
3300 }
3301
3302 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
3303 {
3304         kvm_make_request(KVM_REQ_EVENT, vcpu);
3305         return 1;
3306 }
3307
3308 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
3309 {
3310         u32 cpu_based_vm_exec_control;
3311
3312         /* clear pending irq */
3313         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3314         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
3315         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3316
3317         kvm_make_request(KVM_REQ_EVENT, vcpu);
3318
3319         ++vcpu->stat.irq_window_exits;
3320
3321         /*
3322          * If the user space waits to inject interrupts, exit as soon as
3323          * possible
3324          */
3325         if (!irqchip_in_kernel(vcpu->kvm) &&
3326             vcpu->run->request_interrupt_window &&
3327             !kvm_cpu_has_interrupt(vcpu)) {
3328                 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
3329                 return 0;
3330         }
3331         return 1;
3332 }
3333
3334 static int handle_halt(struct kvm_vcpu *vcpu)
3335 {
3336         skip_emulated_instruction(vcpu);
3337         return kvm_emulate_halt(vcpu);
3338 }
3339
3340 static int handle_vmcall(struct kvm_vcpu *vcpu)
3341 {
3342         skip_emulated_instruction(vcpu);
3343         kvm_emulate_hypercall(vcpu);
3344         return 1;
3345 }
3346
3347 static int handle_vmx_insn(struct kvm_vcpu *vcpu)
3348 {
3349         kvm_queue_exception(vcpu, UD_VECTOR);
3350         return 1;
3351 }
3352
3353 static int handle_invlpg(struct kvm_vcpu *vcpu)
3354 {
3355         unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3356
3357         kvm_mmu_invlpg(vcpu, exit_qualification);
3358         skip_emulated_instruction(vcpu);
3359         return 1;
3360 }
3361
3362 static int handle_wbinvd(struct kvm_vcpu *vcpu)
3363 {
3364         skip_emulated_instruction(vcpu);
3365         kvm_emulate_wbinvd(vcpu);
3366         return 1;
3367 }
3368
3369 static int handle_xsetbv(struct kvm_vcpu *vcpu)
3370 {
3371         u64 new_bv = kvm_read_edx_eax(vcpu);
3372         u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3373
3374         if (kvm_set_xcr(vcpu, index, new_bv) == 0)
3375                 skip_emulated_instruction(vcpu);
3376         return 1;
3377 }
3378
3379 static int handle_apic_access(struct kvm_vcpu *vcpu)
3380 {
3381         return emulate_instruction(vcpu, 0, 0, 0) == EMULATE_DONE;
3382 }
3383
3384 static int handle_task_switch(struct kvm_vcpu *vcpu)
3385 {
3386         struct vcpu_vmx *vmx = to_vmx(vcpu);
3387         unsigned long exit_qualification;
3388         bool has_error_code = false;
3389         u32 error_code = 0;
3390         u16 tss_selector;
3391         int reason, type, idt_v;
3392
3393         idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
3394         type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
3395
3396         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3397
3398         reason = (u32)exit_qualification >> 30;
3399         if (reason == TASK_SWITCH_GATE && idt_v) {
3400                 switch (type) {
3401                 case INTR_TYPE_NMI_INTR:
3402                         vcpu->arch.nmi_injected = false;
3403                         if (cpu_has_virtual_nmis())
3404                                 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3405                                               GUEST_INTR_STATE_NMI);
3406                         break;
3407                 case INTR_TYPE_EXT_INTR:
3408                 case INTR_TYPE_SOFT_INTR:
3409                         kvm_clear_interrupt_queue(vcpu);
3410                         break;
3411                 case INTR_TYPE_HARD_EXCEPTION:
3412                         if (vmx->idt_vectoring_info &
3413                             VECTORING_INFO_DELIVER_CODE_MASK) {
3414                                 has_error_code = true;
3415                                 error_code =
3416                                         vmcs_read32(IDT_VECTORING_ERROR_CODE);
3417                         }
3418                         /* fall through */
3419                 case INTR_TYPE_SOFT_EXCEPTION:
3420                         kvm_clear_exception_queue(vcpu);
3421                         break;
3422                 default:
3423                         break;
3424                 }
3425         }
3426         tss_selector = exit_qualification;
3427
3428         if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
3429                        type != INTR_TYPE_EXT_INTR &&
3430                        type != INTR_TYPE_NMI_INTR))
3431                 skip_emulated_instruction(vcpu);
3432
3433         if (kvm_task_switch(vcpu, tss_selector, reason,
3434                                 has_error_code, error_code) == EMULATE_FAIL) {
3435                 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3436                 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
3437                 vcpu->run->internal.ndata = 0;
3438                 return 0;
3439         }
3440
3441         /* clear all local breakpoint enable flags */
3442         vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
3443
3444         /*
3445          * TODO: What about debug traps on tss switch?
3446          *       Are we supposed to inject them and update dr6?
3447          */
3448
3449         return 1;
3450 }
3451
3452 static int handle_ept_violation(struct kvm_vcpu *vcpu)
3453 {
3454         unsigned long exit_qualification;
3455         gpa_t gpa;
3456         int gla_validity;
3457
3458         exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3459
3460         if (exit_qualification & (1 << 6)) {
3461                 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
3462                 return -EINVAL;
3463         }
3464
3465         gla_validity = (exit_qualification >> 7) & 0x3;
3466         if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
3467                 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
3468                 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
3469                         (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
3470                         vmcs_readl(GUEST_LINEAR_ADDRESS));
3471                 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
3472                         (long unsigned int)exit_qualification);
3473                 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3474                 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
3475                 return 0;
3476         }
3477
3478         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3479         trace_kvm_page_fault(gpa, exit_qualification);
3480         return kvm_mmu_page_fault(vcpu, gpa & PAGE_MASK, 0);
3481 }
3482
3483 static u64 ept_rsvd_mask(u64 spte, int level)
3484 {
3485         int i;
3486         u64 mask = 0;
3487
3488         for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
3489                 mask |= (1ULL << i);
3490
3491         if (level > 2)
3492                 /* bits 7:3 reserved */
3493                 mask |= 0xf8;
3494         else if (level == 2) {
3495                 if (spte & (1ULL << 7))
3496                         /* 2MB ref, bits 20:12 reserved */
3497                         mask |= 0x1ff000;
3498                 else
3499                         /* bits 6:3 reserved */
3500                         mask |= 0x78;
3501         }
3502
3503         return mask;
3504 }
3505
3506 static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
3507                                        int level)
3508 {
3509         printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
3510
3511         /* 010b (write-only) */
3512         WARN_ON((spte & 0x7) == 0x2);
3513
3514         /* 110b (write/execute) */
3515         WARN_ON((spte & 0x7) == 0x6);
3516
3517         /* 100b (execute-only) and value not supported by logical processor */
3518         if (!cpu_has_vmx_ept_execute_only())
3519                 WARN_ON((spte & 0x7) == 0x4);
3520
3521         /* not 000b */
3522         if ((spte & 0x7)) {
3523                 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
3524
3525                 if (rsvd_bits != 0) {
3526                         printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
3527                                          __func__, rsvd_bits);
3528                         WARN_ON(1);
3529                 }
3530
3531                 if (level == 1 || (level == 2 && (spte & (1ULL << 7)))) {
3532                         u64 ept_mem_type = (spte & 0x38) >> 3;
3533
3534                         if (ept_mem_type == 2 || ept_mem_type == 3 ||
3535                             ept_mem_type == 7) {
3536                                 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
3537                                                 __func__, ept_mem_type);
3538                                 WARN_ON(1);
3539                         }
3540                 }
3541         }
3542 }
3543
3544 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
3545 {
3546         u64 sptes[4];
3547         int nr_sptes, i;
3548         gpa_t gpa;
3549
3550         gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3551
3552         printk(KERN_ERR "EPT: Misconfiguration.\n");
3553         printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
3554
3555         nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
3556
3557         for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
3558                 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
3559
3560         vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3561         vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
3562
3563         return 0;
3564 }
3565
3566 static int handle_nmi_window(struct kvm_vcpu *vcpu)
3567 {
3568         u32 cpu_based_vm_exec_control;
3569
3570         /* clear pending NMI */
3571         cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3572         cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
3573         vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3574         ++vcpu->stat.nmi_window_exits;
3575         kvm_make_request(KVM_REQ_EVENT, vcpu);
3576
3577         return 1;
3578 }
3579
3580 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
3581 {
3582         struct vcpu_vmx *vmx = to_vmx(vcpu);
3583         enum emulation_result err = EMULATE_DONE;
3584         int ret = 1;
3585         u32 cpu_exec_ctrl;
3586         bool intr_window_requested;
3587
3588         cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3589         intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
3590
3591         while (!guest_state_valid(vcpu)) {
3592                 if (intr_window_requested
3593                     && (kvm_get_rflags(&vmx->vcpu) & X86_EFLAGS_IF))
3594                         return handle_interrupt_window(&vmx->vcpu);
3595
3596                 err = emulate_instruction(vcpu, 0, 0, 0);
3597
3598                 if (err == EMULATE_DO_MMIO) {
3599                         ret = 0;
3600                         goto out;
3601                 }
3602
3603                 if (err != EMULATE_DONE)
3604                         return 0;
3605
3606                 if (signal_pending(current))
3607                         goto out;
3608                 if (need_resched())
3609                         schedule();
3610         }
3611
3612         vmx->emulation_required = 0;
3613 out:
3614         return ret;
3615 }
3616
3617 /*
3618  * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
3619  * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
3620  */
3621 static int handle_pause(struct kvm_vcpu *vcpu)
3622 {
3623         skip_emulated_instruction(vcpu);
3624         kvm_vcpu_on_spin(vcpu);
3625
3626         return 1;
3627 }
3628
3629 static int handle_invalid_op(struct kvm_vcpu *vcpu)
3630 {
3631         kvm_queue_exception(vcpu, UD_VECTOR);
3632         return 1;
3633 }
3634
3635 /*
3636  * The exit handlers return 1 if the exit was handled fully and guest execution
3637  * may resume.  Otherwise they set the kvm_run parameter to indicate what needs
3638  * to be done to userspace and return 0.
3639  */
3640 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
3641         [EXIT_REASON_EXCEPTION_NMI]           = handle_exception,
3642         [EXIT_REASON_EXTERNAL_INTERRUPT]      = handle_external_interrupt,
3643         [EXIT_REASON_TRIPLE_FAULT]            = handle_triple_fault,
3644         [EXIT_REASON_NMI_WINDOW]              = handle_nmi_window,
3645         [EXIT_REASON_IO_INSTRUCTION]          = handle_io,
3646         [EXIT_REASON_CR_ACCESS]               = handle_cr,
3647         [EXIT_REASON_DR_ACCESS]               = handle_dr,
3648         [EXIT_REASON_CPUID]                   = handle_cpuid,
3649         [EXIT_REASON_MSR_READ]                = handle_rdmsr,
3650         [EXIT_REASON_MSR_WRITE]               = handle_wrmsr,
3651         [EXIT_REASON_PENDING_INTERRUPT]       = handle_interrupt_window,
3652         [EXIT_REASON_HLT]                     = handle_halt,
3653         [EXIT_REASON_INVLPG]                  = handle_invlpg,
3654         [EXIT_REASON_VMCALL]                  = handle_vmcall,
3655         [EXIT_REASON_VMCLEAR]                 = handle_vmx_insn,
3656         [EXIT_REASON_VMLAUNCH]                = handle_vmx_insn,
3657         [EXIT_REASON_VMPTRLD]                 = handle_vmx_insn,
3658         [EXIT_REASON_VMPTRST]                 = handle_vmx_insn,
3659         [EXIT_REASON_VMREAD]                  = handle_vmx_insn,
3660         [EXIT_REASON_VMRESUME]                = handle_vmx_insn,
3661         [EXIT_REASON_VMWRITE]                 = handle_vmx_insn,
3662         [EXIT_REASON_VMOFF]                   = handle_vmx_insn,
3663         [EXIT_REASON_VMON]                    = handle_vmx_insn,
3664         [EXIT_REASON_TPR_BELOW_THRESHOLD]     = handle_tpr_below_threshold,
3665         [EXIT_REASON_APIC_ACCESS]             = handle_apic_access,
3666         [EXIT_REASON_WBINVD]                  = handle_wbinvd,
3667         [EXIT_REASON_XSETBV]                  = handle_xsetbv,
3668         [EXIT_REASON_TASK_SWITCH]             = handle_task_switch,
3669         [EXIT_REASON_MCE_DURING_VMENTRY]      = handle_machine_check,
3670         [EXIT_REASON_EPT_VIOLATION]           = handle_ept_violation,
3671         [EXIT_REASON_EPT_MISCONFIG]           = handle_ept_misconfig,
3672         [EXIT_REASON_PAUSE_INSTRUCTION]       = handle_pause,
3673         [EXIT_REASON_MWAIT_INSTRUCTION]       = handle_invalid_op,
3674         [EXIT_REASON_MONITOR_INSTRUCTION]     = handle_invalid_op,
3675 };
3676
3677 static const int kvm_vmx_max_exit_handlers =
3678         ARRAY_SIZE(kvm_vmx_exit_handlers);
3679
3680 /*
3681  * The guest has exited.  See if we can fix it or if we need userspace
3682  * assistance.
3683  */
3684 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
3685 {
3686         struct vcpu_vmx *vmx = to_vmx(vcpu);
3687         u32 exit_reason = vmx->exit_reason;
3688         u32 vectoring_info = vmx->idt_vectoring_info;
3689
3690         trace_kvm_exit(exit_reason, vcpu);
3691
3692         /* If guest state is invalid, start emulating */
3693         if (vmx->emulation_required && emulate_invalid_guest_state)
3694                 return handle_invalid_guest_state(vcpu);
3695
3696         /* Access CR3 don't cause VMExit in paging mode, so we need
3697          * to sync with guest real CR3. */
3698         if (enable_ept && is_paging(vcpu))
3699                 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3700
3701         if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
3702                 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3703                 vcpu->run->fail_entry.hardware_entry_failure_reason
3704                         = exit_reason;
3705                 return 0;
3706         }
3707
3708         if (unlikely(vmx->fail)) {
3709                 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3710                 vcpu->run->fail_entry.hardware_entry_failure_reason
3711                         = vmcs_read32(VM_INSTRUCTION_ERROR);
3712                 return 0;
3713         }
3714
3715         if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
3716                         (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
3717                         exit_reason != EXIT_REASON_EPT_VIOLATION &&
3718                         exit_reason != EXIT_REASON_TASK_SWITCH))
3719                 printk(KERN_WARNING "%s: unexpected, valid vectoring info "
3720                        "(0x%x) and exit reason is 0x%x\n",
3721                        __func__, vectoring_info, exit_reason);
3722
3723         if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked)) {
3724                 if (vmx_interrupt_allowed(vcpu)) {
3725                         vmx->soft_vnmi_blocked = 0;
3726                 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
3727                            vcpu->arch.nmi_pending) {
3728                         /*
3729                          * This CPU don't support us in finding the end of an
3730                          * NMI-blocked window if the guest runs with IRQs
3731                          * disabled. So we pull the trigger after 1 s of
3732                          * futile waiting, but inform the user about this.
3733                          */
3734                         printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
3735                                "state on VCPU %d after 1 s timeout\n",
3736                                __func__, vcpu->vcpu_id);
3737                         vmx->soft_vnmi_blocked = 0;
3738                 }
3739         }
3740
3741         if (exit_reason < kvm_vmx_max_exit_handlers
3742             && kvm_vmx_exit_handlers[exit_reason])
3743                 return kvm_vmx_exit_handlers[exit_reason](vcpu);
3744         else {
3745                 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3746                 vcpu->run->hw.hardware_exit_reason = exit_reason;
3747         }
3748         return 0;
3749 }
3750
3751 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3752 {
3753         if (irr == -1 || tpr < irr) {
3754                 vmcs_write32(TPR_THRESHOLD, 0);
3755                 return;
3756         }
3757
3758         vmcs_write32(TPR_THRESHOLD, irr);
3759 }
3760
3761 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
3762 {
3763         u32 exit_intr_info = vmx->exit_intr_info;
3764
3765         /* Handle machine checks before interrupts are enabled */
3766         if ((vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
3767             || (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI
3768                 && is_machine_check(exit_intr_info)))
3769                 kvm_machine_check();
3770
3771         /* We need to handle NMIs before interrupts are enabled */
3772         if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
3773             (exit_intr_info & INTR_INFO_VALID_MASK)) {
3774                 kvm_before_handle_nmi(&vmx->vcpu);
3775                 asm("int $2");
3776                 kvm_after_handle_nmi(&vmx->vcpu);
3777         }
3778 }
3779
3780 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
3781 {
3782         u32 exit_intr_info = vmx->exit_intr_info;
3783         bool unblock_nmi;
3784         u8 vector;
3785         bool idtv_info_valid;
3786
3787         idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
3788
3789         if (cpu_has_virtual_nmis()) {
3790                 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
3791                 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
3792                 /*
3793                  * SDM 3: 27.7.1.2 (September 2008)
3794                  * Re-set bit "block by NMI" before VM entry if vmexit caused by
3795                  * a guest IRET fault.
3796                  * SDM 3: 23.2.2 (September 2008)
3797                  * Bit 12 is undefined in any of the following cases:
3798                  *  If the VM exit sets the valid bit in the IDT-vectoring
3799                  *   information field.
3800                  *  If the VM exit is due to a double fault.
3801                  */
3802                 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
3803                     vector != DF_VECTOR && !idtv_info_valid)
3804                         vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3805                                       GUEST_INTR_STATE_NMI);
3806         } else if (unlikely(vmx->soft_vnmi_blocked))
3807                 vmx->vnmi_blocked_time +=
3808                         ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
3809 }
3810
3811 static void __vmx_complete_interrupts(struct vcpu_vmx *vmx,
3812                                       u32 idt_vectoring_info,
3813                                       int instr_len_field,
3814                                       int error_code_field)
3815 {
3816         u8 vector;
3817         int type;
3818         bool idtv_info_valid;
3819
3820         idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
3821
3822         vmx->vcpu.arch.nmi_injected = false;
3823         kvm_clear_exception_queue(&vmx->vcpu);
3824         kvm_clear_interrupt_queue(&vmx->vcpu);
3825
3826         if (!idtv_info_valid)
3827                 return;
3828
3829         kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
3830
3831         vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
3832         type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
3833
3834         switch (type) {
3835         case INTR_TYPE_NMI_INTR:
3836                 vmx->vcpu.arch.nmi_injected = true;
3837                 /*
3838                  * SDM 3: 27.7.1.2 (September 2008)
3839                  * Clear bit "block by NMI" before VM entry if a NMI
3840                  * delivery faulted.
3841                  */
3842                 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
3843                                 GUEST_INTR_STATE_NMI);
3844                 break;
3845         case INTR_TYPE_SOFT_EXCEPTION:
3846                 vmx->vcpu.arch.event_exit_inst_len =
3847                         vmcs_read32(instr_len_field);
3848                 /* fall through */
3849         case INTR_TYPE_HARD_EXCEPTION:
3850                 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
3851                         u32 err = vmcs_read32(error_code_field);
3852                         kvm_queue_exception_e(&vmx->vcpu, vector, err);
3853                 } else
3854                         kvm_queue_exception(&vmx->vcpu, vector);
3855                 break;
3856         case INTR_TYPE_SOFT_INTR:
3857                 vmx->vcpu.arch.event_exit_inst_len =
3858                         vmcs_read32(instr_len_field);
3859                 /* fall through */
3860         case INTR_TYPE_EXT_INTR:
3861                 kvm_queue_interrupt(&vmx->vcpu, vector,
3862                         type == INTR_TYPE_SOFT_INTR);
3863                 break;
3864         default:
3865                 break;
3866         }
3867 }
3868
3869 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
3870 {
3871         __vmx_complete_interrupts(vmx, vmx->idt_vectoring_info,
3872                                   VM_EXIT_INSTRUCTION_LEN,
3873                                   IDT_VECTORING_ERROR_CODE);
3874 }
3875
3876 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
3877 {
3878         __vmx_complete_interrupts(to_vmx(vcpu),
3879                                   vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
3880                                   VM_ENTRY_INSTRUCTION_LEN,
3881                                   VM_ENTRY_EXCEPTION_ERROR_CODE);
3882
3883         vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
3884 }
3885
3886 #ifdef CONFIG_X86_64
3887 #define R "r"
3888 #define Q "q"
3889 #else
3890 #define R "e"
3891 #define Q "l"
3892 #endif
3893
3894 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
3895 {
3896         struct vcpu_vmx *vmx = to_vmx(vcpu);
3897
3898         /* Record the guest's net vcpu time for enforced NMI injections. */
3899         if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
3900                 vmx->entry_time = ktime_get();
3901
3902         /* Don't enter VMX if guest state is invalid, let the exit handler
3903            start emulation until we arrive back to a valid state */
3904         if (vmx->emulation_required && emulate_invalid_guest_state)
3905                 return;
3906
3907         if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
3908                 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
3909         if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
3910                 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
3911
3912         /* When single-stepping over STI and MOV SS, we must clear the
3913          * corresponding interruptibility bits in the guest state. Otherwise
3914          * vmentry fails as it then expects bit 14 (BS) in pending debug
3915          * exceptions being set, but that's not correct for the guest debugging
3916          * case. */
3917         if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3918                 vmx_set_interrupt_shadow(vcpu, 0);
3919
3920         asm(
3921                 /* Store host registers */
3922                 "push %%"R"dx; push %%"R"bp;"
3923                 "push %%"R"cx \n\t"
3924                 "cmp %%"R"sp, %c[host_rsp](%0) \n\t"
3925                 "je 1f \n\t"
3926                 "mov %%"R"sp, %c[host_rsp](%0) \n\t"
3927                 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
3928                 "1: \n\t"
3929                 /* Reload cr2 if changed */
3930                 "mov %c[cr2](%0), %%"R"ax \n\t"
3931                 "mov %%cr2, %%"R"dx \n\t"
3932                 "cmp %%"R"ax, %%"R"dx \n\t"
3933                 "je 2f \n\t"
3934                 "mov %%"R"ax, %%cr2 \n\t"
3935                 "2: \n\t"
3936                 /* Check if vmlaunch of vmresume is needed */
3937                 "cmpl $0, %c[launched](%0) \n\t"
3938                 /* Load guest registers.  Don't clobber flags. */
3939                 "mov %c[rax](%0), %%"R"ax \n\t"
3940                 "mov %c[rbx](%0), %%"R"bx \n\t"
3941                 "mov %c[rdx](%0), %%"R"dx \n\t"
3942                 "mov %c[rsi](%0), %%"R"si \n\t"
3943                 "mov %c[rdi](%0), %%"R"di \n\t"
3944                 "mov %c[rbp](%0), %%"R"bp \n\t"
3945 #ifdef CONFIG_X86_64
3946                 "mov %c[r8](%0),  %%r8  \n\t"
3947                 "mov %c[r9](%0),  %%r9  \n\t"
3948                 "mov %c[r10](%0), %%r10 \n\t"
3949                 "mov %c[r11](%0), %%r11 \n\t"
3950                 "mov %c[r12](%0), %%r12 \n\t"
3951                 "mov %c[r13](%0), %%r13 \n\t"
3952                 "mov %c[r14](%0), %%r14 \n\t"
3953                 "mov %c[r15](%0), %%r15 \n\t"
3954 #endif
3955                 "mov %c[rcx](%0), %%"R"cx \n\t" /* kills %0 (ecx) */
3956
3957                 /* Enter guest mode */
3958                 "jne .Llaunched \n\t"
3959                 __ex(ASM_VMX_VMLAUNCH) "\n\t"
3960                 "jmp .Lkvm_vmx_return \n\t"
3961                 ".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t"
3962                 ".Lkvm_vmx_return: "
3963                 /* Save guest registers, load host registers, keep flags */
3964                 "xchg %0,     (%%"R"sp) \n\t"
3965                 "mov %%"R"ax, %c[rax](%0) \n\t"
3966                 "mov %%"R"bx, %c[rbx](%0) \n\t"
3967                 "push"Q" (%%"R"sp); pop"Q" %c[rcx](%0) \n\t"
3968                 "mov %%"R"dx, %c[rdx](%0) \n\t"
3969                 "mov %%"R"si, %c[rsi](%0) \n\t"
3970                 "mov %%"R"di, %c[rdi](%0) \n\t"
3971                 "mov %%"R"bp, %c[rbp](%0) \n\t"
3972 #ifdef CONFIG_X86_64
3973                 "mov %%r8,  %c[r8](%0) \n\t"
3974                 "mov %%r9,  %c[r9](%0) \n\t"
3975                 "mov %%r10, %c[r10](%0) \n\t"
3976                 "mov %%r11, %c[r11](%0) \n\t"
3977                 "mov %%r12, %c[r12](%0) \n\t"
3978                 "mov %%r13, %c[r13](%0) \n\t"
3979                 "mov %%r14, %c[r14](%0) \n\t"
3980                 "mov %%r15, %c[r15](%0) \n\t"
3981 #endif
3982                 "mov %%cr2, %%"R"ax   \n\t"
3983                 "mov %%"R"ax, %c[cr2](%0) \n\t"
3984
3985                 "pop  %%"R"bp; pop  %%"R"bp; pop  %%"R"dx \n\t"
3986                 "setbe %c[fail](%0) \n\t"
3987               : : "c"(vmx), "d"((unsigned long)HOST_RSP),
3988                 [launched]"i"(offsetof(struct vcpu_vmx, launched)),
3989                 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
3990                 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
3991                 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
3992                 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
3993                 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
3994                 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
3995                 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
3996                 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
3997                 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
3998 #ifdef CONFIG_X86_64
3999                 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
4000                 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
4001                 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
4002                 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
4003                 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
4004                 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
4005                 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
4006                 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
4007 #endif
4008                 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2))
4009               : "cc", "memory"
4010                 , R"bx", R"di", R"si"
4011 #ifdef CONFIG_X86_64
4012                 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
4013 #endif
4014               );
4015
4016         vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
4017                                   | (1 << VCPU_EXREG_PDPTR));
4018         vcpu->arch.regs_dirty = 0;
4019
4020         vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
4021
4022         asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
4023         vmx->launched = 1;
4024
4025         vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
4026         vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
4027
4028         vmx_complete_atomic_exit(vmx);
4029         vmx_recover_nmi_blocking(vmx);
4030         vmx_complete_interrupts(vmx);
4031 }
4032
4033 #undef R
4034 #undef Q
4035
4036 static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
4037 {
4038         struct vcpu_vmx *vmx = to_vmx(vcpu);
4039
4040         if (vmx->vmcs) {
4041                 vcpu_clear(vmx);
4042                 free_vmcs(vmx->vmcs);
4043                 vmx->vmcs = NULL;
4044         }
4045 }
4046
4047 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
4048 {
4049         struct vcpu_vmx *vmx = to_vmx(vcpu);
4050
4051         free_vpid(vmx);
4052         vmx_free_vmcs(vcpu);
4053         kfree(vmx->guest_msrs);
4054         kvm_vcpu_uninit(vcpu);
4055         kmem_cache_free(kvm_vcpu_cache, vmx);
4056 }
4057
4058 static inline void vmcs_init(struct vmcs *vmcs)
4059 {
4060         u64 phys_addr = __pa(per_cpu(vmxarea, raw_smp_processor_id()));
4061
4062         if (!vmm_exclusive)
4063                 kvm_cpu_vmxon(phys_addr);
4064
4065         vmcs_clear(vmcs);
4066
4067         if (!vmm_exclusive)
4068                 kvm_cpu_vmxoff();
4069 }
4070
4071 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
4072 {
4073         int err;
4074         struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
4075         int cpu;
4076
4077         if (!vmx)
4078                 return ERR_PTR(-ENOMEM);
4079
4080         allocate_vpid(vmx);
4081
4082         err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
4083         if (err)
4084                 goto free_vcpu;
4085
4086         vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
4087         if (!vmx->guest_msrs) {
4088                 err = -ENOMEM;
4089                 goto uninit_vcpu;
4090         }
4091
4092         vmx->vmcs = alloc_vmcs();
4093         if (!vmx->vmcs)
4094                 goto free_msrs;
4095
4096         vmcs_init(vmx->vmcs);
4097
4098         cpu = get_cpu();
4099         vmx_vcpu_load(&vmx->vcpu, cpu);
4100         vmx->vcpu.cpu = cpu;
4101         err = vmx_vcpu_setup(vmx);
4102         vmx_vcpu_put(&vmx->vcpu);
4103         put_cpu();
4104         if (err)
4105                 goto free_vmcs;
4106         if (vm_need_virtualize_apic_accesses(kvm))
4107                 if (alloc_apic_access_page(kvm) != 0)
4108                         goto free_vmcs;
4109
4110         if (enable_ept) {
4111                 if (!kvm->arch.ept_identity_map_addr)
4112                         kvm->arch.ept_identity_map_addr =
4113                                 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
4114                 if (alloc_identity_pagetable(kvm) != 0)
4115                         goto free_vmcs;
4116         }
4117
4118         return &vmx->vcpu;
4119
4120 free_vmcs:
4121         free_vmcs(vmx->vmcs);
4122 free_msrs:
4123         kfree(vmx->guest_msrs);
4124 uninit_vcpu:
4125         kvm_vcpu_uninit(&vmx->vcpu);
4126 free_vcpu:
4127         free_vpid(vmx);
4128         kmem_cache_free(kvm_vcpu_cache, vmx);
4129         return ERR_PTR(err);
4130 }
4131
4132 static void __init vmx_check_processor_compat(void *rtn)
4133 {
4134         struct vmcs_config vmcs_conf;
4135
4136         *(int *)rtn = 0;
4137         if (setup_vmcs_config(&vmcs_conf) < 0)
4138                 *(int *)rtn = -EIO;
4139         if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
4140                 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
4141                                 smp_processor_id());
4142                 *(int *)rtn = -EIO;
4143         }
4144 }
4145
4146 static int get_ept_level(void)
4147 {
4148         return VMX_EPT_DEFAULT_GAW + 1;
4149 }
4150
4151 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
4152 {
4153         u64 ret;
4154
4155         /* For VT-d and EPT combination
4156          * 1. MMIO: always map as UC
4157          * 2. EPT with VT-d:
4158          *   a. VT-d without snooping control feature: can't guarantee the
4159          *      result, try to trust guest.
4160          *   b. VT-d with snooping control feature: snooping control feature of
4161          *      VT-d engine can guarantee the cache correctness. Just set it
4162          *      to WB to keep consistent with host. So the same as item 3.
4163          * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
4164          *    consistent with host MTRR
4165          */
4166         if (is_mmio)
4167                 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
4168         else if (vcpu->kvm->arch.iommu_domain &&
4169                 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY))
4170                 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
4171                       VMX_EPT_MT_EPTE_SHIFT;
4172         else
4173                 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
4174                         | VMX_EPT_IPAT_BIT;
4175
4176         return ret;
4177 }
4178
4179 #define _ER(x) { EXIT_REASON_##x, #x }
4180
4181 static const struct trace_print_flags vmx_exit_reasons_str[] = {
4182         _ER(EXCEPTION_NMI),
4183         _ER(EXTERNAL_INTERRUPT),
4184         _ER(TRIPLE_FAULT),
4185         _ER(PENDING_INTERRUPT),
4186         _ER(NMI_WINDOW),
4187         _ER(TASK_SWITCH),
4188         _ER(CPUID),
4189         _ER(HLT),
4190         _ER(INVLPG),
4191         _ER(RDPMC),
4192         _ER(RDTSC),
4193         _ER(VMCALL),
4194         _ER(VMCLEAR),
4195         _ER(VMLAUNCH),
4196         _ER(VMPTRLD),
4197         _ER(VMPTRST),
4198         _ER(VMREAD),
4199         _ER(VMRESUME),
4200         _ER(VMWRITE),
4201         _ER(VMOFF),
4202         _ER(VMON),
4203         _ER(CR_ACCESS),
4204         _ER(DR_ACCESS),
4205         _ER(IO_INSTRUCTION),
4206         _ER(MSR_READ),
4207         _ER(MSR_WRITE),
4208         _ER(MWAIT_INSTRUCTION),
4209         _ER(MONITOR_INSTRUCTION),
4210         _ER(PAUSE_INSTRUCTION),
4211         _ER(MCE_DURING_VMENTRY),
4212         _ER(TPR_BELOW_THRESHOLD),
4213         _ER(APIC_ACCESS),
4214         _ER(EPT_VIOLATION),
4215         _ER(EPT_MISCONFIG),
4216         _ER(WBINVD),
4217         { -1, NULL }
4218 };
4219
4220 #undef _ER
4221
4222 static int vmx_get_lpage_level(void)
4223 {
4224         if (enable_ept && !cpu_has_vmx_ept_1g_page())
4225                 return PT_DIRECTORY_LEVEL;
4226         else
4227                 /* For shadow and EPT supported 1GB page */
4228                 return PT_PDPE_LEVEL;
4229 }
4230
4231 static inline u32 bit(int bitno)
4232 {
4233         return 1 << (bitno & 31);
4234 }
4235
4236 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
4237 {
4238         struct kvm_cpuid_entry2 *best;
4239         struct vcpu_vmx *vmx = to_vmx(vcpu);
4240         u32 exec_control;
4241
4242         vmx->rdtscp_enabled = false;
4243         if (vmx_rdtscp_supported()) {
4244                 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
4245                 if (exec_control & SECONDARY_EXEC_RDTSCP) {
4246                         best = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
4247                         if (best && (best->edx & bit(X86_FEATURE_RDTSCP)))
4248                                 vmx->rdtscp_enabled = true;
4249                         else {
4250                                 exec_control &= ~SECONDARY_EXEC_RDTSCP;
4251                                 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
4252                                                 exec_control);
4253                         }
4254                 }
4255         }
4256 }
4257
4258 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
4259 {
4260 }
4261
4262 static struct kvm_x86_ops vmx_x86_ops = {
4263         .cpu_has_kvm_support = cpu_has_kvm_support,
4264         .disabled_by_bios = vmx_disabled_by_bios,
4265         .hardware_setup = hardware_setup,
4266         .hardware_unsetup = hardware_unsetup,
4267         .check_processor_compatibility = vmx_check_processor_compat,
4268         .hardware_enable = hardware_enable,
4269         .hardware_disable = hardware_disable,
4270         .cpu_has_accelerated_tpr = report_flexpriority,
4271
4272         .vcpu_create = vmx_create_vcpu,
4273         .vcpu_free = vmx_free_vcpu,
4274         .vcpu_reset = vmx_vcpu_reset,
4275
4276         .prepare_guest_switch = vmx_save_host_state,
4277         .vcpu_load = vmx_vcpu_load,
4278         .vcpu_put = vmx_vcpu_put,
4279
4280         .set_guest_debug = set_guest_debug,
4281         .get_msr = vmx_get_msr,
4282         .set_msr = vmx_set_msr,
4283         .get_segment_base = vmx_get_segment_base,
4284         .get_segment = vmx_get_segment,
4285         .set_segment = vmx_set_segment,
4286         .get_cpl = vmx_get_cpl,
4287         .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
4288         .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
4289         .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
4290         .set_cr0 = vmx_set_cr0,
4291         .set_cr3 = vmx_set_cr3,
4292         .set_cr4 = vmx_set_cr4,
4293         .set_efer = vmx_set_efer,
4294         .get_idt = vmx_get_idt,
4295         .set_idt = vmx_set_idt,
4296         .get_gdt = vmx_get_gdt,
4297         .set_gdt = vmx_set_gdt,
4298         .set_dr7 = vmx_set_dr7,
4299         .cache_reg = vmx_cache_reg,
4300         .get_rflags = vmx_get_rflags,
4301         .set_rflags = vmx_set_rflags,
4302         .fpu_activate = vmx_fpu_activate,
4303         .fpu_deactivate = vmx_fpu_deactivate,
4304
4305         .tlb_flush = vmx_flush_tlb,
4306
4307         .run = vmx_vcpu_run,
4308         .handle_exit = vmx_handle_exit,
4309         .skip_emulated_instruction = skip_emulated_instruction,
4310         .set_interrupt_shadow = vmx_set_interrupt_shadow,
4311         .get_interrupt_shadow = vmx_get_interrupt_shadow,
4312         .patch_hypercall = vmx_patch_hypercall,
4313         .set_irq = vmx_inject_irq,
4314         .set_nmi = vmx_inject_nmi,
4315         .queue_exception = vmx_queue_exception,
4316         .cancel_injection = vmx_cancel_injection,
4317         .interrupt_allowed = vmx_interrupt_allowed,
4318         .nmi_allowed = vmx_nmi_allowed,
4319         .get_nmi_mask = vmx_get_nmi_mask,
4320         .set_nmi_mask = vmx_set_nmi_mask,
4321         .enable_nmi_window = enable_nmi_window,
4322         .enable_irq_window = enable_irq_window,
4323         .update_cr8_intercept = update_cr8_intercept,
4324
4325         .set_tss_addr = vmx_set_tss_addr,
4326         .get_tdp_level = get_ept_level,
4327         .get_mt_mask = vmx_get_mt_mask,
4328
4329         .exit_reasons_str = vmx_exit_reasons_str,
4330         .get_lpage_level = vmx_get_lpage_level,
4331
4332         .cpuid_update = vmx_cpuid_update,
4333
4334         .rdtscp_supported = vmx_rdtscp_supported,
4335
4336         .set_supported_cpuid = vmx_set_supported_cpuid,
4337
4338         .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
4339
4340         .write_tsc_offset = vmx_write_tsc_offset,
4341         .adjust_tsc_offset = vmx_adjust_tsc_offset,
4342
4343         .set_tdp_cr3 = vmx_set_cr3,
4344 };
4345
4346 static int __init vmx_init(void)
4347 {
4348         int r, i;
4349
4350         rdmsrl_safe(MSR_EFER, &host_efer);
4351
4352         for (i = 0; i < NR_VMX_MSR; ++i)
4353                 kvm_define_shared_msr(i, vmx_msr_index[i]);
4354
4355         vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
4356         if (!vmx_io_bitmap_a)
4357                 return -ENOMEM;
4358
4359         vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
4360         if (!vmx_io_bitmap_b) {
4361                 r = -ENOMEM;
4362                 goto out;
4363         }
4364
4365         vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
4366         if (!vmx_msr_bitmap_legacy) {
4367                 r = -ENOMEM;
4368                 goto out1;
4369         }
4370
4371         vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
4372         if (!vmx_msr_bitmap_longmode) {
4373                 r = -ENOMEM;
4374                 goto out2;
4375         }
4376
4377         /*
4378          * Allow direct access to the PC debug port (it is often used for I/O
4379          * delays, but the vmexits simply slow things down).
4380          */
4381         memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
4382         clear_bit(0x80, vmx_io_bitmap_a);
4383
4384         memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
4385
4386         memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
4387         memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
4388
4389         set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
4390
4391         r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
4392                      __alignof__(struct vcpu_vmx), THIS_MODULE);
4393         if (r)
4394                 goto out3;
4395
4396         vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
4397         vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
4398         vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
4399         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
4400         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
4401         vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
4402
4403         if (enable_ept) {
4404                 bypass_guest_pf = 0;
4405                 kvm_mmu_set_base_ptes(VMX_EPT_READABLE_MASK |
4406                         VMX_EPT_WRITABLE_MASK);
4407                 kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull,
4408                                 VMX_EPT_EXECUTABLE_MASK);
4409                 kvm_enable_tdp();
4410         } else
4411                 kvm_disable_tdp();
4412
4413         if (bypass_guest_pf)
4414                 kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull);
4415
4416         return 0;
4417
4418 out3:
4419         free_page((unsigned long)vmx_msr_bitmap_longmode);
4420 out2:
4421         free_page((unsigned long)vmx_msr_bitmap_legacy);
4422 out1:
4423         free_page((unsigned long)vmx_io_bitmap_b);
4424 out:
4425         free_page((unsigned long)vmx_io_bitmap_a);
4426         return r;
4427 }
4428
4429 static void __exit vmx_exit(void)
4430 {
4431         free_page((unsigned long)vmx_msr_bitmap_legacy);
4432         free_page((unsigned long)vmx_msr_bitmap_longmode);
4433         free_page((unsigned long)vmx_io_bitmap_b);
4434         free_page((unsigned long)vmx_io_bitmap_a);
4435
4436         kvm_exit();
4437 }
4438
4439 module_init(vmx_init)
4440 module_exit(vmx_exit)