]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame_incremental - arch/x86/kvm/vmx.c
projects / mirror_ubuntu-bionic-kernel.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
KVM: ignore reads to perfctr msrs
[mirror_ubuntu-bionic-kernel.git] / arch / x86 / kvm / vmx.c
... / ...
CommitLineData
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 *
9 * Authors:
10 * Avi Kivity <avi@qumranet.com>
11 * Yaniv Kamay <yaniv@qumranet.com>
12 *
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
15 *
16 */
17
18#include "irq.h"
19#include "mmu.h"
20
21#include <linux/kvm_host.h>
22#include <linux/module.h>
23#include <linux/kernel.h>
24#include <linux/mm.h>
25#include <linux/highmem.h>
26#include <linux/sched.h>
27#include <linux/moduleparam.h>
28#include <linux/ftrace_event.h>
29#include "kvm_cache_regs.h"
30#include "x86.h"
31
32#include <asm/io.h>
33#include <asm/desc.h>
34#include <asm/vmx.h>
35#include <asm/virtext.h>
36#include <asm/mce.h>
37
38#include "trace.h"
39
40#define __ex(x) __kvm_handle_fault_on_reboot(x)
41
42MODULE_AUTHOR("Qumranet");
43MODULE_LICENSE("GPL");
44
45static int __read_mostly bypass_guest_pf = 1;
46module_param(bypass_guest_pf, bool, S_IRUGO);
47
48static int __read_mostly enable_vpid = 1;
49module_param_named(vpid, enable_vpid, bool, 0444);
50
51static int __read_mostly flexpriority_enabled = 1;
52module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
53
54static int __read_mostly enable_ept = 1;
55module_param_named(ept, enable_ept, bool, S_IRUGO);
56
57static int __read_mostly enable_unrestricted_guest = 1;
58module_param_named(unrestricted_guest,
59 enable_unrestricted_guest, bool, S_IRUGO);
60
61static int __read_mostly emulate_invalid_guest_state = 0;
62module_param(emulate_invalid_guest_state, bool, S_IRUGO);
63
64struct vmcs {
65 u32 revision_id;
66 u32 abort;
67 char data[0];
68};
69
70struct vcpu_vmx {
71 struct kvm_vcpu vcpu;
72 struct list_head local_vcpus_link;
73 unsigned long host_rsp;
74 int launched;
75 u8 fail;
76 u32 idt_vectoring_info;
77 struct kvm_msr_entry *guest_msrs;
78 struct kvm_msr_entry *host_msrs;
79 int nmsrs;
80 int save_nmsrs;
81 int msr_offset_efer;
82#ifdef CONFIG_X86_64
83 int msr_offset_kernel_gs_base;
84#endif
85 struct vmcs *vmcs;
86 struct {
87 int loaded;
88 u16 fs_sel, gs_sel, ldt_sel;
89 int gs_ldt_reload_needed;
90 int fs_reload_needed;
91 int guest_efer_loaded;
92 } host_state;
93 struct {
94 int vm86_active;
95 u8 save_iopl;
96 struct kvm_save_segment {
97 u16 selector;
98 unsigned long base;
99 u32 limit;
100 u32 ar;
101 } tr, es, ds, fs, gs;
102 struct {
103 bool pending;
104 u8 vector;
105 unsigned rip;
106 } irq;
107 } rmode;
108 int vpid;
109 bool emulation_required;
110 enum emulation_result invalid_state_emulation_result;
111
112 /* Support for vnmi-less CPUs */
113 int soft_vnmi_blocked;
114 ktime_t entry_time;
115 s64 vnmi_blocked_time;
116 u32 exit_reason;
117};
118
119static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
120{
121 return container_of(vcpu, struct vcpu_vmx, vcpu);
122}
123
124static int init_rmode(struct kvm *kvm);
125static u64 construct_eptp(unsigned long root_hpa);
126
127static DEFINE_PER_CPU(struct vmcs *, vmxarea);
128static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
129static DEFINE_PER_CPU(struct list_head, vcpus_on_cpu);
130
131static unsigned long *vmx_io_bitmap_a;
132static unsigned long *vmx_io_bitmap_b;
133static unsigned long *vmx_msr_bitmap_legacy;
134static unsigned long *vmx_msr_bitmap_longmode;
135
136static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
137static DEFINE_SPINLOCK(vmx_vpid_lock);
138
139static struct vmcs_config {
140 int size;
141 int order;
142 u32 revision_id;
143 u32 pin_based_exec_ctrl;
144 u32 cpu_based_exec_ctrl;
145 u32 cpu_based_2nd_exec_ctrl;
146 u32 vmexit_ctrl;
147 u32 vmentry_ctrl;
148} vmcs_config;
149
150static struct vmx_capability {
151 u32 ept;
152 u32 vpid;
153} vmx_capability;
154
155#define VMX_SEGMENT_FIELD(seg) \
156 [VCPU_SREG_##seg] = { \
157 .selector = GUEST_##seg##_SELECTOR, \
158 .base = GUEST_##seg##_BASE, \
159 .limit = GUEST_##seg##_LIMIT, \
160 .ar_bytes = GUEST_##seg##_AR_BYTES, \
161 }
162
163static struct kvm_vmx_segment_field {
164 unsigned selector;
165 unsigned base;
166 unsigned limit;
167 unsigned ar_bytes;
168} kvm_vmx_segment_fields[] = {
169 VMX_SEGMENT_FIELD(CS),
170 VMX_SEGMENT_FIELD(DS),
171 VMX_SEGMENT_FIELD(ES),
172 VMX_SEGMENT_FIELD(FS),
173 VMX_SEGMENT_FIELD(GS),
174 VMX_SEGMENT_FIELD(SS),
175 VMX_SEGMENT_FIELD(TR),
176 VMX_SEGMENT_FIELD(LDTR),
177};
178
179static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
180
181/*
182 * Keep MSR_K6_STAR at the end, as setup_msrs() will try to optimize it
183 * away by decrementing the array size.
184 */
185static const u32 vmx_msr_index[] = {
186#ifdef CONFIG_X86_64
187 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR, MSR_KERNEL_GS_BASE,
188#endif
189 MSR_EFER, MSR_K6_STAR,
190};
191#define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
192
193static void load_msrs(struct kvm_msr_entry *e, int n)
194{
195 int i;
196
197 for (i = 0; i < n; ++i)
198 wrmsrl(e[i].index, e[i].data);
199}
200
201static void save_msrs(struct kvm_msr_entry *e, int n)
202{
203 int i;
204
205 for (i = 0; i < n; ++i)
206 rdmsrl(e[i].index, e[i].data);
207}
208
209static inline int is_page_fault(u32 intr_info)
210{
211 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
212 INTR_INFO_VALID_MASK)) ==
213 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
214}
215
216static inline int is_no_device(u32 intr_info)
217{
218 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
219 INTR_INFO_VALID_MASK)) ==
220 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
221}
222
223static inline int is_invalid_opcode(u32 intr_info)
224{
225 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
226 INTR_INFO_VALID_MASK)) ==
227 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
228}
229
230static inline int is_external_interrupt(u32 intr_info)
231{
232 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
233 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
234}
235
236static inline int is_machine_check(u32 intr_info)
237{
238 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
239 INTR_INFO_VALID_MASK)) ==
240 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
241}
242
243static inline int cpu_has_vmx_msr_bitmap(void)
244{
245 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
246}
247
248static inline int cpu_has_vmx_tpr_shadow(void)
249{
250 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
251}
252
253static inline int vm_need_tpr_shadow(struct kvm *kvm)
254{
255 return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
256}
257
258static inline int cpu_has_secondary_exec_ctrls(void)
259{
260 return vmcs_config.cpu_based_exec_ctrl &
261 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
262}
263
264static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
265{
266 return vmcs_config.cpu_based_2nd_exec_ctrl &
267 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
268}
269
270static inline bool cpu_has_vmx_flexpriority(void)
271{
272 return cpu_has_vmx_tpr_shadow() &&
273 cpu_has_vmx_virtualize_apic_accesses();
274}
275
276static inline bool cpu_has_vmx_ept_execute_only(void)
277{
278 return !!(vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT);
279}
280
281static inline bool cpu_has_vmx_eptp_uncacheable(void)
282{
283 return !!(vmx_capability.ept & VMX_EPTP_UC_BIT);
284}
285
286static inline bool cpu_has_vmx_eptp_writeback(void)
287{
288 return !!(vmx_capability.ept & VMX_EPTP_WB_BIT);
289}
290
291static inline bool cpu_has_vmx_ept_2m_page(void)
292{
293 return !!(vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT);
294}
295
296static inline int cpu_has_vmx_invept_individual_addr(void)
297{
298 return !!(vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT);
299}
300
301static inline int cpu_has_vmx_invept_context(void)
302{
303 return !!(vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT);
304}
305
306static inline int cpu_has_vmx_invept_global(void)
307{
308 return !!(vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT);
309}
310
311static inline int cpu_has_vmx_ept(void)
312{
313 return vmcs_config.cpu_based_2nd_exec_ctrl &
314 SECONDARY_EXEC_ENABLE_EPT;
315}
316
317static inline int cpu_has_vmx_unrestricted_guest(void)
318{
319 return vmcs_config.cpu_based_2nd_exec_ctrl &
320 SECONDARY_EXEC_UNRESTRICTED_GUEST;
321}
322
323static inline int vm_need_virtualize_apic_accesses(struct kvm *kvm)
324{
325 return flexpriority_enabled &&
326 (cpu_has_vmx_virtualize_apic_accesses()) &&
327 (irqchip_in_kernel(kvm));
328}
329
330static inline int cpu_has_vmx_vpid(void)
331{
332 return vmcs_config.cpu_based_2nd_exec_ctrl &
333 SECONDARY_EXEC_ENABLE_VPID;
334}
335
336static inline int cpu_has_virtual_nmis(void)
337{
338 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
339}
340
341static inline bool report_flexpriority(void)
342{
343 return flexpriority_enabled;
344}
345
346static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
347{
348 int i;
349
350 for (i = 0; i < vmx->nmsrs; ++i)
351 if (vmx->guest_msrs[i].index == msr)
352 return i;
353 return -1;
354}
355
356static inline void __invvpid(int ext, u16 vpid, gva_t gva)
357{
358 struct {
359 u64 vpid : 16;
360 u64 rsvd : 48;
361 u64 gva;
362 } operand = { vpid, 0, gva };
363
364 asm volatile (__ex(ASM_VMX_INVVPID)
365 /* CF==1 or ZF==1 --> rc = -1 */
366 "; ja 1f ; ud2 ; 1:"
367 : : "a"(&operand), "c"(ext) : "cc", "memory");
368}
369
370static inline void __invept(int ext, u64 eptp, gpa_t gpa)
371{
372 struct {
373 u64 eptp, gpa;
374 } operand = {eptp, gpa};
375
376 asm volatile (__ex(ASM_VMX_INVEPT)
377 /* CF==1 or ZF==1 --> rc = -1 */
378 "; ja 1f ; ud2 ; 1:\n"
379 : : "a" (&operand), "c" (ext) : "cc", "memory");
380}
381
382static struct kvm_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
383{
384 int i;
385
386 i = __find_msr_index(vmx, msr);
387 if (i >= 0)
388 return &vmx->guest_msrs[i];
389 return NULL;
390}
391
392static void vmcs_clear(struct vmcs *vmcs)
393{
394 u64 phys_addr = __pa(vmcs);
395 u8 error;
396
397 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
398 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
399 : "cc", "memory");
400 if (error)
401 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
402 vmcs, phys_addr);
403}
404
405static void __vcpu_clear(void *arg)
406{
407 struct vcpu_vmx *vmx = arg;
408 int cpu = raw_smp_processor_id();
409
410 if (vmx->vcpu.cpu == cpu)
411 vmcs_clear(vmx->vmcs);
412 if (per_cpu(current_vmcs, cpu) == vmx->vmcs)
413 per_cpu(current_vmcs, cpu) = NULL;
414 rdtscll(vmx->vcpu.arch.host_tsc);
415 list_del(&vmx->local_vcpus_link);
416 vmx->vcpu.cpu = -1;
417 vmx->launched = 0;
418}
419
420static void vcpu_clear(struct vcpu_vmx *vmx)
421{
422 if (vmx->vcpu.cpu == -1)
423 return;
424 smp_call_function_single(vmx->vcpu.cpu, __vcpu_clear, vmx, 1);
425}
426
427static inline void vpid_sync_vcpu_all(struct vcpu_vmx *vmx)
428{
429 if (vmx->vpid == 0)
430 return;
431
432 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
433}
434
435static inline void ept_sync_global(void)
436{
437 if (cpu_has_vmx_invept_global())
438 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
439}
440
441static inline void ept_sync_context(u64 eptp)
442{
443 if (enable_ept) {
444 if (cpu_has_vmx_invept_context())
445 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
446 else
447 ept_sync_global();
448 }
449}
450
451static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
452{
453 if (enable_ept) {
454 if (cpu_has_vmx_invept_individual_addr())
455 __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
456 eptp, gpa);
457 else
458 ept_sync_context(eptp);
459 }
460}
461
462static unsigned long vmcs_readl(unsigned long field)
463{
464 unsigned long value;
465
466 asm volatile (__ex(ASM_VMX_VMREAD_RDX_RAX)
467 : "=a"(value) : "d"(field) : "cc");
468 return value;
469}
470
471static u16 vmcs_read16(unsigned long field)
472{
473 return vmcs_readl(field);
474}
475
476static u32 vmcs_read32(unsigned long field)
477{
478 return vmcs_readl(field);
479}
480
481static u64 vmcs_read64(unsigned long field)
482{
483#ifdef CONFIG_X86_64
484 return vmcs_readl(field);
485#else
486 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
487#endif
488}
489
490static noinline void vmwrite_error(unsigned long field, unsigned long value)
491{
492 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
493 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
494 dump_stack();
495}
496
497static void vmcs_writel(unsigned long field, unsigned long value)
498{
499 u8 error;
500
501 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
502 : "=q"(error) : "a"(value), "d"(field) : "cc");
503 if (unlikely(error))
504 vmwrite_error(field, value);
505}
506
507static void vmcs_write16(unsigned long field, u16 value)
508{
509 vmcs_writel(field, value);
510}
511
512static void vmcs_write32(unsigned long field, u32 value)
513{
514 vmcs_writel(field, value);
515}
516
517static void vmcs_write64(unsigned long field, u64 value)
518{
519 vmcs_writel(field, value);
520#ifndef CONFIG_X86_64
521 asm volatile ("");
522 vmcs_writel(field+1, value >> 32);
523#endif
524}
525
526static void vmcs_clear_bits(unsigned long field, u32 mask)
527{
528 vmcs_writel(field, vmcs_readl(field) & ~mask);
529}
530
531static void vmcs_set_bits(unsigned long field, u32 mask)
532{
533 vmcs_writel(field, vmcs_readl(field) | mask);
534}
535
536static void update_exception_bitmap(struct kvm_vcpu *vcpu)
537{
538 u32 eb;
539
540 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR);
541 if (!vcpu->fpu_active)
542 eb |= 1u << NM_VECTOR;
543 if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
544 if (vcpu->guest_debug &
545 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
546 eb |= 1u << DB_VECTOR;
547 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
548 eb |= 1u << BP_VECTOR;
549 }
550 if (to_vmx(vcpu)->rmode.vm86_active)
551 eb = ~0;
552 if (enable_ept)
553 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
554 vmcs_write32(EXCEPTION_BITMAP, eb);
555}
556
557static void reload_tss(void)
558{
559 /*
560 * VT restores TR but not its size. Useless.
561 */
562 struct descriptor_table gdt;
563 struct desc_struct *descs;
564
565 kvm_get_gdt(&gdt);
566 descs = (void *)gdt.base;
567 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
568 load_TR_desc();
569}
570
571static void load_transition_efer(struct vcpu_vmx *vmx)
572{
573 int efer_offset = vmx->msr_offset_efer;
574 u64 host_efer = vmx->host_msrs[efer_offset].data;
575 u64 guest_efer = vmx->guest_msrs[efer_offset].data;
576 u64 ignore_bits;
577
578 if (efer_offset < 0)
579 return;
580 /*
581 * NX is emulated; LMA and LME handled by hardware; SCE meaninless
582 * outside long mode
583 */
584 ignore_bits = EFER_NX | EFER_SCE;
585#ifdef CONFIG_X86_64
586 ignore_bits |= EFER_LMA | EFER_LME;
587 /* SCE is meaningful only in long mode on Intel */
588 if (guest_efer & EFER_LMA)
589 ignore_bits &= ~(u64)EFER_SCE;
590#endif
591 if ((guest_efer & ~ignore_bits) == (host_efer & ~ignore_bits))
592 return;
593
594 vmx->host_state.guest_efer_loaded = 1;
595 guest_efer &= ~ignore_bits;
596 guest_efer |= host_efer & ignore_bits;
597 wrmsrl(MSR_EFER, guest_efer);
598 vmx->vcpu.stat.efer_reload++;
599}
600
601static void reload_host_efer(struct vcpu_vmx *vmx)
602{
603 if (vmx->host_state.guest_efer_loaded) {
604 vmx->host_state.guest_efer_loaded = 0;
605 load_msrs(vmx->host_msrs + vmx->msr_offset_efer, 1);
606 }
607}
608
609static void vmx_save_host_state(struct kvm_vcpu *vcpu)
610{
611 struct vcpu_vmx *vmx = to_vmx(vcpu);
612
613 if (vmx->host_state.loaded)
614 return;
615
616 vmx->host_state.loaded = 1;
617 /*
618 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
619 * allow segment selectors with cpl > 0 or ti == 1.
620 */
621 vmx->host_state.ldt_sel = kvm_read_ldt();
622 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
623 vmx->host_state.fs_sel = kvm_read_fs();
624 if (!(vmx->host_state.fs_sel & 7)) {
625 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
626 vmx->host_state.fs_reload_needed = 0;
627 } else {
628 vmcs_write16(HOST_FS_SELECTOR, 0);
629 vmx->host_state.fs_reload_needed = 1;
630 }
631 vmx->host_state.gs_sel = kvm_read_gs();
632 if (!(vmx->host_state.gs_sel & 7))
633 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
634 else {
635 vmcs_write16(HOST_GS_SELECTOR, 0);
636 vmx->host_state.gs_ldt_reload_needed = 1;
637 }
638
639#ifdef CONFIG_X86_64
640 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
641 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
642#else
643 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
644 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
645#endif
646
647#ifdef CONFIG_X86_64
648 if (is_long_mode(&vmx->vcpu))
649 save_msrs(vmx->host_msrs +
650 vmx->msr_offset_kernel_gs_base, 1);
651
652#endif
653 load_msrs(vmx->guest_msrs, vmx->save_nmsrs);
654 load_transition_efer(vmx);
655}
656
657static void __vmx_load_host_state(struct vcpu_vmx *vmx)
658{
659 unsigned long flags;
660
661 if (!vmx->host_state.loaded)
662 return;
663
664 ++vmx->vcpu.stat.host_state_reload;
665 vmx->host_state.loaded = 0;
666 if (vmx->host_state.fs_reload_needed)
667 kvm_load_fs(vmx->host_state.fs_sel);
668 if (vmx->host_state.gs_ldt_reload_needed) {
669 kvm_load_ldt(vmx->host_state.ldt_sel);
670 /*
671 * If we have to reload gs, we must take care to
672 * preserve our gs base.
673 */
674 local_irq_save(flags);
675 kvm_load_gs(vmx->host_state.gs_sel);
676#ifdef CONFIG_X86_64
677 wrmsrl(MSR_GS_BASE, vmcs_readl(HOST_GS_BASE));
678#endif
679 local_irq_restore(flags);
680 }
681 reload_tss();
682 save_msrs(vmx->guest_msrs, vmx->save_nmsrs);
683 load_msrs(vmx->host_msrs, vmx->save_nmsrs);
684 reload_host_efer(vmx);
685}
686
687static void vmx_load_host_state(struct vcpu_vmx *vmx)
688{
689 preempt_disable();
690 __vmx_load_host_state(vmx);
691 preempt_enable();
692}
693
694/*
695 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
696 * vcpu mutex is already taken.
697 */
698static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
699{
700 struct vcpu_vmx *vmx = to_vmx(vcpu);
701 u64 phys_addr = __pa(vmx->vmcs);
702 u64 tsc_this, delta, new_offset;
703
704 if (vcpu->cpu != cpu) {
705 vcpu_clear(vmx);
706 kvm_migrate_timers(vcpu);
707 vpid_sync_vcpu_all(vmx);
708 local_irq_disable();
709 list_add(&vmx->local_vcpus_link,
710 &per_cpu(vcpus_on_cpu, cpu));
711 local_irq_enable();
712 }
713
714 if (per_cpu(current_vmcs, cpu) != vmx->vmcs) {
715 u8 error;
716
717 per_cpu(current_vmcs, cpu) = vmx->vmcs;
718 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
719 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
720 : "cc");
721 if (error)
722 printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
723 vmx->vmcs, phys_addr);
724 }
725
726 if (vcpu->cpu != cpu) {
727 struct descriptor_table dt;
728 unsigned long sysenter_esp;
729
730 vcpu->cpu = cpu;
731 /*
732 * Linux uses per-cpu TSS and GDT, so set these when switching
733 * processors.
734 */
735 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
736 kvm_get_gdt(&dt);
737 vmcs_writel(HOST_GDTR_BASE, dt.base); /* 22.2.4 */
738
739 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
740 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
741
742 /*
743 * Make sure the time stamp counter is monotonous.
744 */
745 rdtscll(tsc_this);
746 if (tsc_this < vcpu->arch.host_tsc) {
747 delta = vcpu->arch.host_tsc - tsc_this;
748 new_offset = vmcs_read64(TSC_OFFSET) + delta;
749 vmcs_write64(TSC_OFFSET, new_offset);
750 }
751 }
752}
753
754static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
755{
756 __vmx_load_host_state(to_vmx(vcpu));
757}
758
759static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
760{
761 if (vcpu->fpu_active)
762 return;
763 vcpu->fpu_active = 1;
764 vmcs_clear_bits(GUEST_CR0, X86_CR0_TS);
765 if (vcpu->arch.cr0 & X86_CR0_TS)
766 vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
767 update_exception_bitmap(vcpu);
768}
769
770static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
771{
772 if (!vcpu->fpu_active)
773 return;
774 vcpu->fpu_active = 0;
775 vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
776 update_exception_bitmap(vcpu);
777}
778
779static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
780{
781 return vmcs_readl(GUEST_RFLAGS);
782}
783
784static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
785{
786 if (to_vmx(vcpu)->rmode.vm86_active)
787 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
788 vmcs_writel(GUEST_RFLAGS, rflags);
789}
790
791static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
792{
793 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
794 int ret = 0;
795
796 if (interruptibility & GUEST_INTR_STATE_STI)
797 ret |= X86_SHADOW_INT_STI;
798 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
799 ret |= X86_SHADOW_INT_MOV_SS;
800
801 return ret & mask;
802}
803
804static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
805{
806 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
807 u32 interruptibility = interruptibility_old;
808
809 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
810
811 if (mask & X86_SHADOW_INT_MOV_SS)
812 interruptibility |= GUEST_INTR_STATE_MOV_SS;
813 if (mask & X86_SHADOW_INT_STI)
814 interruptibility |= GUEST_INTR_STATE_STI;
815
816 if ((interruptibility != interruptibility_old))
817 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
818}
819
820static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
821{
822 unsigned long rip;
823
824 rip = kvm_rip_read(vcpu);
825 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
826 kvm_rip_write(vcpu, rip);
827
828 /* skipping an emulated instruction also counts */
829 vmx_set_interrupt_shadow(vcpu, 0);
830}
831
832static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
833 bool has_error_code, u32 error_code)
834{
835 struct vcpu_vmx *vmx = to_vmx(vcpu);
836 u32 intr_info = nr | INTR_INFO_VALID_MASK;
837
838 if (has_error_code) {
839 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
840 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
841 }
842
843 if (vmx->rmode.vm86_active) {
844 vmx->rmode.irq.pending = true;
845 vmx->rmode.irq.vector = nr;
846 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
847 if (kvm_exception_is_soft(nr))
848 vmx->rmode.irq.rip +=
849 vmx->vcpu.arch.event_exit_inst_len;
850 intr_info |= INTR_TYPE_SOFT_INTR;
851 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
852 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
853 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
854 return;
855 }
856
857 if (kvm_exception_is_soft(nr)) {
858 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
859 vmx->vcpu.arch.event_exit_inst_len);
860 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
861 } else
862 intr_info |= INTR_TYPE_HARD_EXCEPTION;
863
864 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
865}
866
867/*
868 * Swap MSR entry in host/guest MSR entry array.
869 */
870#ifdef CONFIG_X86_64
871static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
872{
873 struct kvm_msr_entry tmp;
874
875 tmp = vmx->guest_msrs[to];
876 vmx->guest_msrs[to] = vmx->guest_msrs[from];
877 vmx->guest_msrs[from] = tmp;
878 tmp = vmx->host_msrs[to];
879 vmx->host_msrs[to] = vmx->host_msrs[from];
880 vmx->host_msrs[from] = tmp;
881}
882#endif
883
884/*
885 * Set up the vmcs to automatically save and restore system
886 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
887 * mode, as fiddling with msrs is very expensive.
888 */
889static void setup_msrs(struct vcpu_vmx *vmx)
890{
891 int save_nmsrs;
892 unsigned long *msr_bitmap;
893
894 vmx_load_host_state(vmx);
895 save_nmsrs = 0;
896#ifdef CONFIG_X86_64
897 if (is_long_mode(&vmx->vcpu)) {
898 int index;
899
900 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
901 if (index >= 0)
902 move_msr_up(vmx, index, save_nmsrs++);
903 index = __find_msr_index(vmx, MSR_LSTAR);
904 if (index >= 0)
905 move_msr_up(vmx, index, save_nmsrs++);
906 index = __find_msr_index(vmx, MSR_CSTAR);
907 if (index >= 0)
908 move_msr_up(vmx, index, save_nmsrs++);
909 index = __find_msr_index(vmx, MSR_KERNEL_GS_BASE);
910 if (index >= 0)
911 move_msr_up(vmx, index, save_nmsrs++);
912 /*
913 * MSR_K6_STAR is only needed on long mode guests, and only
914 * if efer.sce is enabled.
915 */
916 index = __find_msr_index(vmx, MSR_K6_STAR);
917 if ((index >= 0) && (vmx->vcpu.arch.shadow_efer & EFER_SCE))
918 move_msr_up(vmx, index, save_nmsrs++);
919 }
920#endif
921 vmx->save_nmsrs = save_nmsrs;
922
923#ifdef CONFIG_X86_64
924 vmx->msr_offset_kernel_gs_base =
925 __find_msr_index(vmx, MSR_KERNEL_GS_BASE);
926#endif
927 vmx->msr_offset_efer = __find_msr_index(vmx, MSR_EFER);
928
929 if (cpu_has_vmx_msr_bitmap()) {
930 if (is_long_mode(&vmx->vcpu))
931 msr_bitmap = vmx_msr_bitmap_longmode;
932 else
933 msr_bitmap = vmx_msr_bitmap_legacy;
934
935 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
936 }
937}
938
939/*
940 * reads and returns guest's timestamp counter "register"
941 * guest_tsc = host_tsc + tsc_offset -- 21.3
942 */
943static u64 guest_read_tsc(void)
944{
945 u64 host_tsc, tsc_offset;
946
947 rdtscll(host_tsc);
948 tsc_offset = vmcs_read64(TSC_OFFSET);
949 return host_tsc + tsc_offset;
950}
951
952/*
953 * writes 'guest_tsc' into guest's timestamp counter "register"
954 * guest_tsc = host_tsc + tsc_offset ==> tsc_offset = guest_tsc - host_tsc
955 */
956static void guest_write_tsc(u64 guest_tsc, u64 host_tsc)
957{
958 vmcs_write64(TSC_OFFSET, guest_tsc - host_tsc);
959}
960
961/*
962 * Reads an msr value (of 'msr_index') into 'pdata'.
963 * Returns 0 on success, non-0 otherwise.
964 * Assumes vcpu_load() was already called.
965 */
966static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
967{
968 u64 data;
969 struct kvm_msr_entry *msr;
970
971 if (!pdata) {
972 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
973 return -EINVAL;
974 }
975
976 switch (msr_index) {
977#ifdef CONFIG_X86_64
978 case MSR_FS_BASE:
979 data = vmcs_readl(GUEST_FS_BASE);
980 break;
981 case MSR_GS_BASE:
982 data = vmcs_readl(GUEST_GS_BASE);
983 break;
984 case MSR_EFER:
985 return kvm_get_msr_common(vcpu, msr_index, pdata);
986#endif
987 case MSR_IA32_TSC:
988 data = guest_read_tsc();
989 break;
990 case MSR_IA32_SYSENTER_CS:
991 data = vmcs_read32(GUEST_SYSENTER_CS);
992 break;
993 case MSR_IA32_SYSENTER_EIP:
994 data = vmcs_readl(GUEST_SYSENTER_EIP);
995 break;
996 case MSR_IA32_SYSENTER_ESP:
997 data = vmcs_readl(GUEST_SYSENTER_ESP);
998 break;
999 default:
1000 vmx_load_host_state(to_vmx(vcpu));
1001 msr = find_msr_entry(to_vmx(vcpu), msr_index);
1002 if (msr) {
1003 data = msr->data;
1004 break;
1005 }
1006 return kvm_get_msr_common(vcpu, msr_index, pdata);
1007 }
1008
1009 *pdata = data;
1010 return 0;
1011}
1012
1013/*
1014 * Writes msr value into into the appropriate "register".
1015 * Returns 0 on success, non-0 otherwise.
1016 * Assumes vcpu_load() was already called.
1017 */
1018static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1019{
1020 struct vcpu_vmx *vmx = to_vmx(vcpu);
1021 struct kvm_msr_entry *msr;
1022 u64 host_tsc;
1023 int ret = 0;
1024
1025 switch (msr_index) {
1026 case MSR_EFER:
1027 vmx_load_host_state(vmx);
1028 ret = kvm_set_msr_common(vcpu, msr_index, data);
1029 break;
1030#ifdef CONFIG_X86_64
1031 case MSR_FS_BASE:
1032 vmcs_writel(GUEST_FS_BASE, data);
1033 break;
1034 case MSR_GS_BASE:
1035 vmcs_writel(GUEST_GS_BASE, data);
1036 break;
1037#endif
1038 case MSR_IA32_SYSENTER_CS:
1039 vmcs_write32(GUEST_SYSENTER_CS, data);
1040 break;
1041 case MSR_IA32_SYSENTER_EIP:
1042 vmcs_writel(GUEST_SYSENTER_EIP, data);
1043 break;
1044 case MSR_IA32_SYSENTER_ESP:
1045 vmcs_writel(GUEST_SYSENTER_ESP, data);
1046 break;
1047 case MSR_IA32_TSC:
1048 rdtscll(host_tsc);
1049 guest_write_tsc(data, host_tsc);
1050 break;
1051 case MSR_IA32_CR_PAT:
1052 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
1053 vmcs_write64(GUEST_IA32_PAT, data);
1054 vcpu->arch.pat = data;
1055 break;
1056 }
1057 /* Otherwise falls through to kvm_set_msr_common */
1058 default:
1059 vmx_load_host_state(vmx);
1060 msr = find_msr_entry(vmx, msr_index);
1061 if (msr) {
1062 msr->data = data;
1063 break;
1064 }
1065 ret = kvm_set_msr_common(vcpu, msr_index, data);
1066 }
1067
1068 return ret;
1069}
1070
1071static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1072{
1073 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
1074 switch (reg) {
1075 case VCPU_REGS_RSP:
1076 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
1077 break;
1078 case VCPU_REGS_RIP:
1079 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
1080 break;
1081 case VCPU_EXREG_PDPTR:
1082 if (enable_ept)
1083 ept_save_pdptrs(vcpu);
1084 break;
1085 default:
1086 break;
1087 }
1088}
1089
1090static int set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
1091{
1092 int old_debug = vcpu->guest_debug;
1093 unsigned long flags;
1094
1095 vcpu->guest_debug = dbg->control;
1096 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
1097 vcpu->guest_debug = 0;
1098
1099 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1100 vmcs_writel(GUEST_DR7, dbg->arch.debugreg[7]);
1101 else
1102 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
1103
1104 flags = vmcs_readl(GUEST_RFLAGS);
1105 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
1106 flags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
1107 else if (old_debug & KVM_GUESTDBG_SINGLESTEP)
1108 flags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1109 vmcs_writel(GUEST_RFLAGS, flags);
1110
1111 update_exception_bitmap(vcpu);
1112
1113 return 0;
1114}
1115
1116static __init int cpu_has_kvm_support(void)
1117{
1118 return cpu_has_vmx();
1119}
1120
1121static __init int vmx_disabled_by_bios(void)
1122{
1123 u64 msr;
1124
1125 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
1126 return (msr & (FEATURE_CONTROL_LOCKED |
1127 FEATURE_CONTROL_VMXON_ENABLED))
1128 == FEATURE_CONTROL_LOCKED;
1129 /* locked but not enabled */
1130}
1131
1132static void hardware_enable(void *garbage)
1133{
1134 int cpu = raw_smp_processor_id();
1135 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1136 u64 old;
1137
1138 INIT_LIST_HEAD(&per_cpu(vcpus_on_cpu, cpu));
1139 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
1140 if ((old & (FEATURE_CONTROL_LOCKED |
1141 FEATURE_CONTROL_VMXON_ENABLED))
1142 != (FEATURE_CONTROL_LOCKED |
1143 FEATURE_CONTROL_VMXON_ENABLED))
1144 /* enable and lock */
1145 wrmsrl(MSR_IA32_FEATURE_CONTROL, old |
1146 FEATURE_CONTROL_LOCKED |
1147 FEATURE_CONTROL_VMXON_ENABLED);
1148 write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
1149 asm volatile (ASM_VMX_VMXON_RAX
1150 : : "a"(&phys_addr), "m"(phys_addr)
1151 : "memory", "cc");
1152}
1153
1154static void vmclear_local_vcpus(void)
1155{
1156 int cpu = raw_smp_processor_id();
1157 struct vcpu_vmx *vmx, *n;
1158
1159 list_for_each_entry_safe(vmx, n, &per_cpu(vcpus_on_cpu, cpu),
1160 local_vcpus_link)
1161 __vcpu_clear(vmx);
1162}
1163
1164
1165/* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
1166 * tricks.
1167 */
1168static void kvm_cpu_vmxoff(void)
1169{
1170 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
1171 write_cr4(read_cr4() & ~X86_CR4_VMXE);
1172}
1173
1174static void hardware_disable(void *garbage)
1175{
1176 vmclear_local_vcpus();
1177 kvm_cpu_vmxoff();
1178}
1179
1180static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
1181 u32 msr, u32 *result)
1182{
1183 u32 vmx_msr_low, vmx_msr_high;
1184 u32 ctl = ctl_min | ctl_opt;
1185
1186 rdmsr(msr, vmx_msr_low, vmx_msr_high);
1187
1188 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
1189 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
1190
1191 /* Ensure minimum (required) set of control bits are supported. */
1192 if (ctl_min & ~ctl)
1193 return -EIO;
1194
1195 *result = ctl;
1196 return 0;
1197}
1198
1199static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
1200{
1201 u32 vmx_msr_low, vmx_msr_high;
1202 u32 min, opt, min2, opt2;
1203 u32 _pin_based_exec_control = 0;
1204 u32 _cpu_based_exec_control = 0;
1205 u32 _cpu_based_2nd_exec_control = 0;
1206 u32 _vmexit_control = 0;
1207 u32 _vmentry_control = 0;
1208
1209 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
1210 opt = PIN_BASED_VIRTUAL_NMIS;
1211 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
1212 &_pin_based_exec_control) < 0)
1213 return -EIO;
1214
1215 min = CPU_BASED_HLT_EXITING |
1216#ifdef CONFIG_X86_64
1217 CPU_BASED_CR8_LOAD_EXITING |
1218 CPU_BASED_CR8_STORE_EXITING |
1219#endif
1220 CPU_BASED_CR3_LOAD_EXITING |
1221 CPU_BASED_CR3_STORE_EXITING |
1222 CPU_BASED_USE_IO_BITMAPS |
1223 CPU_BASED_MOV_DR_EXITING |
1224 CPU_BASED_USE_TSC_OFFSETING |
1225 CPU_BASED_INVLPG_EXITING;
1226 opt = CPU_BASED_TPR_SHADOW |
1227 CPU_BASED_USE_MSR_BITMAPS |
1228 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1229 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1230 &_cpu_based_exec_control) < 0)
1231 return -EIO;
1232#ifdef CONFIG_X86_64
1233 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
1234 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
1235 ~CPU_BASED_CR8_STORE_EXITING;
1236#endif
1237 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
1238 min2 = 0;
1239 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
1240 SECONDARY_EXEC_WBINVD_EXITING |
1241 SECONDARY_EXEC_ENABLE_VPID |
1242 SECONDARY_EXEC_ENABLE_EPT |
1243 SECONDARY_EXEC_UNRESTRICTED_GUEST;
1244 if (adjust_vmx_controls(min2, opt2,
1245 MSR_IA32_VMX_PROCBASED_CTLS2,
1246 &_cpu_based_2nd_exec_control) < 0)
1247 return -EIO;
1248 }
1249#ifndef CONFIG_X86_64
1250 if (!(_cpu_based_2nd_exec_control &
1251 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
1252 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
1253#endif
1254 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
1255 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
1256 enabled */
1257 min &= ~(CPU_BASED_CR3_LOAD_EXITING |
1258 CPU_BASED_CR3_STORE_EXITING |
1259 CPU_BASED_INVLPG_EXITING);
1260 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1261 &_cpu_based_exec_control) < 0)
1262 return -EIO;
1263 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
1264 vmx_capability.ept, vmx_capability.vpid);
1265 }
1266
1267 min = 0;
1268#ifdef CONFIG_X86_64
1269 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
1270#endif
1271 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
1272 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
1273 &_vmexit_control) < 0)
1274 return -EIO;
1275
1276 min = 0;
1277 opt = VM_ENTRY_LOAD_IA32_PAT;
1278 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
1279 &_vmentry_control) < 0)
1280 return -EIO;
1281
1282 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
1283
1284 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
1285 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
1286 return -EIO;
1287
1288#ifdef CONFIG_X86_64
1289 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
1290 if (vmx_msr_high & (1u<<16))
1291 return -EIO;
1292#endif
1293
1294 /* Require Write-Back (WB) memory type for VMCS accesses. */
1295 if (((vmx_msr_high >> 18) & 15) != 6)
1296 return -EIO;
1297
1298 vmcs_conf->size = vmx_msr_high & 0x1fff;
1299 vmcs_conf->order = get_order(vmcs_config.size);
1300 vmcs_conf->revision_id = vmx_msr_low;
1301
1302 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
1303 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
1304 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
1305 vmcs_conf->vmexit_ctrl = _vmexit_control;
1306 vmcs_conf->vmentry_ctrl = _vmentry_control;
1307
1308 return 0;
1309}
1310
1311static struct vmcs *alloc_vmcs_cpu(int cpu)
1312{
1313 int node = cpu_to_node(cpu);
1314 struct page *pages;
1315 struct vmcs *vmcs;
1316
1317 pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
1318 if (!pages)
1319 return NULL;
1320 vmcs = page_address(pages);
1321 memset(vmcs, 0, vmcs_config.size);
1322 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
1323 return vmcs;
1324}
1325
1326static struct vmcs *alloc_vmcs(void)
1327{
1328 return alloc_vmcs_cpu(raw_smp_processor_id());
1329}
1330
1331static void free_vmcs(struct vmcs *vmcs)
1332{
1333 free_pages((unsigned long)vmcs, vmcs_config.order);
1334}
1335
1336static void free_kvm_area(void)
1337{
1338 int cpu;
1339
1340 for_each_online_cpu(cpu)
1341 free_vmcs(per_cpu(vmxarea, cpu));
1342}
1343
1344static __init int alloc_kvm_area(void)
1345{
1346 int cpu;
1347
1348 for_each_online_cpu(cpu) {
1349 struct vmcs *vmcs;
1350
1351 vmcs = alloc_vmcs_cpu(cpu);
1352 if (!vmcs) {
1353 free_kvm_area();
1354 return -ENOMEM;
1355 }
1356
1357 per_cpu(vmxarea, cpu) = vmcs;
1358 }
1359 return 0;
1360}
1361
1362static __init int hardware_setup(void)
1363{
1364 if (setup_vmcs_config(&vmcs_config) < 0)
1365 return -EIO;
1366
1367 if (boot_cpu_has(X86_FEATURE_NX))
1368 kvm_enable_efer_bits(EFER_NX);
1369
1370 if (!cpu_has_vmx_vpid())
1371 enable_vpid = 0;
1372
1373 if (!cpu_has_vmx_ept()) {
1374 enable_ept = 0;
1375 enable_unrestricted_guest = 0;
1376 }
1377
1378 if (!cpu_has_vmx_unrestricted_guest())
1379 enable_unrestricted_guest = 0;
1380
1381 if (!cpu_has_vmx_flexpriority())
1382 flexpriority_enabled = 0;
1383
1384 if (!cpu_has_vmx_tpr_shadow())
1385 kvm_x86_ops->update_cr8_intercept = NULL;
1386
1387 if (enable_ept && !cpu_has_vmx_ept_2m_page())
1388 kvm_disable_largepages();
1389
1390 return alloc_kvm_area();
1391}
1392
1393static __exit void hardware_unsetup(void)
1394{
1395 free_kvm_area();
1396}
1397
1398static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
1399{
1400 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1401
1402 if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
1403 vmcs_write16(sf->selector, save->selector);
1404 vmcs_writel(sf->base, save->base);
1405 vmcs_write32(sf->limit, save->limit);
1406 vmcs_write32(sf->ar_bytes, save->ar);
1407 } else {
1408 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
1409 << AR_DPL_SHIFT;
1410 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
1411 }
1412}
1413
1414static void enter_pmode(struct kvm_vcpu *vcpu)
1415{
1416 unsigned long flags;
1417 struct vcpu_vmx *vmx = to_vmx(vcpu);
1418
1419 vmx->emulation_required = 1;
1420 vmx->rmode.vm86_active = 0;
1421
1422 vmcs_writel(GUEST_TR_BASE, vmx->rmode.tr.base);
1423 vmcs_write32(GUEST_TR_LIMIT, vmx->rmode.tr.limit);
1424 vmcs_write32(GUEST_TR_AR_BYTES, vmx->rmode.tr.ar);
1425
1426 flags = vmcs_readl(GUEST_RFLAGS);
1427 flags &= ~(X86_EFLAGS_IOPL | X86_EFLAGS_VM);
1428 flags |= (vmx->rmode.save_iopl << IOPL_SHIFT);
1429 vmcs_writel(GUEST_RFLAGS, flags);
1430
1431 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
1432 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
1433
1434 update_exception_bitmap(vcpu);
1435
1436 if (emulate_invalid_guest_state)
1437 return;
1438
1439 fix_pmode_dataseg(VCPU_SREG_ES, &vmx->rmode.es);
1440 fix_pmode_dataseg(VCPU_SREG_DS, &vmx->rmode.ds);
1441 fix_pmode_dataseg(VCPU_SREG_GS, &vmx->rmode.gs);
1442 fix_pmode_dataseg(VCPU_SREG_FS, &vmx->rmode.fs);
1443
1444 vmcs_write16(GUEST_SS_SELECTOR, 0);
1445 vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
1446
1447 vmcs_write16(GUEST_CS_SELECTOR,
1448 vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
1449 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1450}
1451
1452static gva_t rmode_tss_base(struct kvm *kvm)
1453{
1454 if (!kvm->arch.tss_addr) {
1455 gfn_t base_gfn = kvm->memslots[0].base_gfn +
1456 kvm->memslots[0].npages - 3;
1457 return base_gfn << PAGE_SHIFT;
1458 }
1459 return kvm->arch.tss_addr;
1460}
1461
1462static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
1463{
1464 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1465
1466 save->selector = vmcs_read16(sf->selector);
1467 save->base = vmcs_readl(sf->base);
1468 save->limit = vmcs_read32(sf->limit);
1469 save->ar = vmcs_read32(sf->ar_bytes);
1470 vmcs_write16(sf->selector, save->base >> 4);
1471 vmcs_write32(sf->base, save->base & 0xfffff);
1472 vmcs_write32(sf->limit, 0xffff);
1473 vmcs_write32(sf->ar_bytes, 0xf3);
1474}
1475
1476static void enter_rmode(struct kvm_vcpu *vcpu)
1477{
1478 unsigned long flags;
1479 struct vcpu_vmx *vmx = to_vmx(vcpu);
1480
1481 if (enable_unrestricted_guest)
1482 return;
1483
1484 vmx->emulation_required = 1;
1485 vmx->rmode.vm86_active = 1;
1486
1487 vmx->rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
1488 vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
1489
1490 vmx->rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
1491 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
1492
1493 vmx->rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
1494 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1495
1496 flags = vmcs_readl(GUEST_RFLAGS);
1497 vmx->rmode.save_iopl
1498 = (flags & X86_EFLAGS_IOPL) >> IOPL_SHIFT;
1499
1500 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1501
1502 vmcs_writel(GUEST_RFLAGS, flags);
1503 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
1504 update_exception_bitmap(vcpu);
1505
1506 if (emulate_invalid_guest_state)
1507 goto continue_rmode;
1508
1509 vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
1510 vmcs_write32(GUEST_SS_LIMIT, 0xffff);
1511 vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
1512
1513 vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
1514 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1515 if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
1516 vmcs_writel(GUEST_CS_BASE, 0xf0000);
1517 vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
1518
1519 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.es);
1520 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.ds);
1521 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.gs);
1522 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.fs);
1523
1524continue_rmode:
1525 kvm_mmu_reset_context(vcpu);
1526 init_rmode(vcpu->kvm);
1527}
1528
1529static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
1530{
1531 struct vcpu_vmx *vmx = to_vmx(vcpu);
1532 struct kvm_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
1533
1534 vcpu->arch.shadow_efer = efer;
1535 if (!msr)
1536 return;
1537 if (efer & EFER_LMA) {
1538 vmcs_write32(VM_ENTRY_CONTROLS,
1539 vmcs_read32(VM_ENTRY_CONTROLS) |
1540 VM_ENTRY_IA32E_MODE);
1541 msr->data = efer;
1542 } else {
1543 vmcs_write32(VM_ENTRY_CONTROLS,
1544 vmcs_read32(VM_ENTRY_CONTROLS) &
1545 ~VM_ENTRY_IA32E_MODE);
1546
1547 msr->data = efer & ~EFER_LME;
1548 }
1549 setup_msrs(vmx);
1550}
1551
1552#ifdef CONFIG_X86_64
1553
1554static void enter_lmode(struct kvm_vcpu *vcpu)
1555{
1556 u32 guest_tr_ar;
1557
1558 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
1559 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
1560 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
1561 __func__);
1562 vmcs_write32(GUEST_TR_AR_BYTES,
1563 (guest_tr_ar & ~AR_TYPE_MASK)
1564 | AR_TYPE_BUSY_64_TSS);
1565 }
1566 vcpu->arch.shadow_efer |= EFER_LMA;
1567 vmx_set_efer(vcpu, vcpu->arch.shadow_efer);
1568}
1569
1570static void exit_lmode(struct kvm_vcpu *vcpu)
1571{
1572 vcpu->arch.shadow_efer &= ~EFER_LMA;
1573
1574 vmcs_write32(VM_ENTRY_CONTROLS,
1575 vmcs_read32(VM_ENTRY_CONTROLS)
1576 & ~VM_ENTRY_IA32E_MODE);
1577}
1578
1579#endif
1580
1581static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
1582{
1583 vpid_sync_vcpu_all(to_vmx(vcpu));
1584 if (enable_ept)
1585 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
1586}
1587
1588static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1589{
1590 vcpu->arch.cr4 &= KVM_GUEST_CR4_MASK;
1591 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & ~KVM_GUEST_CR4_MASK;
1592}
1593
1594static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
1595{
1596 if (!test_bit(VCPU_EXREG_PDPTR,
1597 (unsigned long *)&vcpu->arch.regs_dirty))
1598 return;
1599
1600 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1601 vmcs_write64(GUEST_PDPTR0, vcpu->arch.pdptrs[0]);
1602 vmcs_write64(GUEST_PDPTR1, vcpu->arch.pdptrs[1]);
1603 vmcs_write64(GUEST_PDPTR2, vcpu->arch.pdptrs[2]);
1604 vmcs_write64(GUEST_PDPTR3, vcpu->arch.pdptrs[3]);
1605 }
1606}
1607
1608static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
1609{
1610 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1611 vcpu->arch.pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
1612 vcpu->arch.pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
1613 vcpu->arch.pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
1614 vcpu->arch.pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
1615 }
1616
1617 __set_bit(VCPU_EXREG_PDPTR,
1618 (unsigned long *)&vcpu->arch.regs_avail);
1619 __set_bit(VCPU_EXREG_PDPTR,
1620 (unsigned long *)&vcpu->arch.regs_dirty);
1621}
1622
1623static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
1624
1625static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
1626 unsigned long cr0,
1627 struct kvm_vcpu *vcpu)
1628{
1629 if (!(cr0 & X86_CR0_PG)) {
1630 /* From paging/starting to nonpaging */
1631 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1632 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
1633 (CPU_BASED_CR3_LOAD_EXITING |
1634 CPU_BASED_CR3_STORE_EXITING));
1635 vcpu->arch.cr0 = cr0;
1636 vmx_set_cr4(vcpu, vcpu->arch.cr4);
1637 *hw_cr0 &= ~X86_CR0_WP;
1638 } else if (!is_paging(vcpu)) {
1639 /* From nonpaging to paging */
1640 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1641 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
1642 ~(CPU_BASED_CR3_LOAD_EXITING |
1643 CPU_BASED_CR3_STORE_EXITING));
1644 vcpu->arch.cr0 = cr0;
1645 vmx_set_cr4(vcpu, vcpu->arch.cr4);
1646 if (!(vcpu->arch.cr0 & X86_CR0_WP))
1647 *hw_cr0 &= ~X86_CR0_WP;
1648 }
1649}
1650
1651static void ept_update_paging_mode_cr4(unsigned long *hw_cr4,
1652 struct kvm_vcpu *vcpu)
1653{
1654 if (!is_paging(vcpu)) {
1655 *hw_cr4 &= ~X86_CR4_PAE;
1656 *hw_cr4 |= X86_CR4_PSE;
1657 } else if (!(vcpu->arch.cr4 & X86_CR4_PAE))
1658 *hw_cr4 &= ~X86_CR4_PAE;
1659}
1660
1661static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1662{
1663 struct vcpu_vmx *vmx = to_vmx(vcpu);
1664 unsigned long hw_cr0;
1665
1666 if (enable_unrestricted_guest)
1667 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST)
1668 | KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
1669 else
1670 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON;
1671
1672 vmx_fpu_deactivate(vcpu);
1673
1674 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
1675 enter_pmode(vcpu);
1676
1677 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
1678 enter_rmode(vcpu);
1679
1680#ifdef CONFIG_X86_64
1681 if (vcpu->arch.shadow_efer & EFER_LME) {
1682 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
1683 enter_lmode(vcpu);
1684 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
1685 exit_lmode(vcpu);
1686 }
1687#endif
1688
1689 if (enable_ept)
1690 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
1691
1692 vmcs_writel(CR0_READ_SHADOW, cr0);
1693 vmcs_writel(GUEST_CR0, hw_cr0);
1694 vcpu->arch.cr0 = cr0;
1695
1696 if (!(cr0 & X86_CR0_TS) || !(cr0 & X86_CR0_PE))
1697 vmx_fpu_activate(vcpu);
1698}
1699
1700static u64 construct_eptp(unsigned long root_hpa)
1701{
1702 u64 eptp;
1703
1704 /* TODO write the value reading from MSR */
1705 eptp = VMX_EPT_DEFAULT_MT |
1706 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
1707 eptp |= (root_hpa & PAGE_MASK);
1708
1709 return eptp;
1710}
1711
1712static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1713{
1714 unsigned long guest_cr3;
1715 u64 eptp;
1716
1717 guest_cr3 = cr3;
1718 if (enable_ept) {
1719 eptp = construct_eptp(cr3);
1720 vmcs_write64(EPT_POINTER, eptp);
1721 guest_cr3 = is_paging(vcpu) ? vcpu->arch.cr3 :
1722 vcpu->kvm->arch.ept_identity_map_addr;
1723 }
1724
1725 vmx_flush_tlb(vcpu);
1726 vmcs_writel(GUEST_CR3, guest_cr3);
1727 if (vcpu->arch.cr0 & X86_CR0_PE)
1728 vmx_fpu_deactivate(vcpu);
1729}
1730
1731static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1732{
1733 unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
1734 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
1735
1736 vcpu->arch.cr4 = cr4;
1737 if (enable_ept)
1738 ept_update_paging_mode_cr4(&hw_cr4, vcpu);
1739
1740 vmcs_writel(CR4_READ_SHADOW, cr4);
1741 vmcs_writel(GUEST_CR4, hw_cr4);
1742}
1743
1744static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1745{
1746 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1747
1748 return vmcs_readl(sf->base);
1749}
1750
1751static void vmx_get_segment(struct kvm_vcpu *vcpu,
1752 struct kvm_segment *var, int seg)
1753{
1754 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1755 u32 ar;
1756
1757 var->base = vmcs_readl(sf->base);
1758 var->limit = vmcs_read32(sf->limit);
1759 var->selector = vmcs_read16(sf->selector);
1760 ar = vmcs_read32(sf->ar_bytes);
1761 if ((ar & AR_UNUSABLE_MASK) && !emulate_invalid_guest_state)
1762 ar = 0;
1763 var->type = ar & 15;
1764 var->s = (ar >> 4) & 1;
1765 var->dpl = (ar >> 5) & 3;
1766 var->present = (ar >> 7) & 1;
1767 var->avl = (ar >> 12) & 1;
1768 var->l = (ar >> 13) & 1;
1769 var->db = (ar >> 14) & 1;
1770 var->g = (ar >> 15) & 1;
1771 var->unusable = (ar >> 16) & 1;
1772}
1773
1774static int vmx_get_cpl(struct kvm_vcpu *vcpu)
1775{
1776 if (!(vcpu->arch.cr0 & X86_CR0_PE)) /* if real mode */
1777 return 0;
1778
1779 if (vmx_get_rflags(vcpu) & X86_EFLAGS_VM) /* if virtual 8086 */
1780 return 3;
1781
1782 return vmcs_read16(GUEST_CS_SELECTOR) & 3;
1783}
1784
1785static u32 vmx_segment_access_rights(struct kvm_segment *var)
1786{
1787 u32 ar;
1788
1789 if (var->unusable)
1790 ar = 1 << 16;
1791 else {
1792 ar = var->type & 15;
1793 ar |= (var->s & 1) << 4;
1794 ar |= (var->dpl & 3) << 5;
1795 ar |= (var->present & 1) << 7;
1796 ar |= (var->avl & 1) << 12;
1797 ar |= (var->l & 1) << 13;
1798 ar |= (var->db & 1) << 14;
1799 ar |= (var->g & 1) << 15;
1800 }
1801 if (ar == 0) /* a 0 value means unusable */
1802 ar = AR_UNUSABLE_MASK;
1803
1804 return ar;
1805}
1806
1807static void vmx_set_segment(struct kvm_vcpu *vcpu,
1808 struct kvm_segment *var, int seg)
1809{
1810 struct vcpu_vmx *vmx = to_vmx(vcpu);
1811 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1812 u32 ar;
1813
1814 if (vmx->rmode.vm86_active && seg == VCPU_SREG_TR) {
1815 vmx->rmode.tr.selector = var->selector;
1816 vmx->rmode.tr.base = var->base;
1817 vmx->rmode.tr.limit = var->limit;
1818 vmx->rmode.tr.ar = vmx_segment_access_rights(var);
1819 return;
1820 }
1821 vmcs_writel(sf->base, var->base);
1822 vmcs_write32(sf->limit, var->limit);
1823 vmcs_write16(sf->selector, var->selector);
1824 if (vmx->rmode.vm86_active && var->s) {
1825 /*
1826 * Hack real-mode segments into vm86 compatibility.
1827 */
1828 if (var->base == 0xffff0000 && var->selector == 0xf000)
1829 vmcs_writel(sf->base, 0xf0000);
1830 ar = 0xf3;
1831 } else
1832 ar = vmx_segment_access_rights(var);
1833
1834 /*
1835 * Fix the "Accessed" bit in AR field of segment registers for older
1836 * qemu binaries.
1837 * IA32 arch specifies that at the time of processor reset the
1838 * "Accessed" bit in the AR field of segment registers is 1. And qemu
1839 * is setting it to 0 in the usedland code. This causes invalid guest
1840 * state vmexit when "unrestricted guest" mode is turned on.
1841 * Fix for this setup issue in cpu_reset is being pushed in the qemu
1842 * tree. Newer qemu binaries with that qemu fix would not need this
1843 * kvm hack.
1844 */
1845 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
1846 ar |= 0x1; /* Accessed */
1847
1848 vmcs_write32(sf->ar_bytes, ar);
1849}
1850
1851static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
1852{
1853 u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
1854
1855 *db = (ar >> 14) & 1;
1856 *l = (ar >> 13) & 1;
1857}
1858
1859static void vmx_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1860{
1861 dt->limit = vmcs_read32(GUEST_IDTR_LIMIT);
1862 dt->base = vmcs_readl(GUEST_IDTR_BASE);
1863}
1864
1865static void vmx_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1866{
1867 vmcs_write32(GUEST_IDTR_LIMIT, dt->limit);
1868 vmcs_writel(GUEST_IDTR_BASE, dt->base);
1869}
1870
1871static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1872{
1873 dt->limit = vmcs_read32(GUEST_GDTR_LIMIT);
1874 dt->base = vmcs_readl(GUEST_GDTR_BASE);
1875}
1876
1877static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1878{
1879 vmcs_write32(GUEST_GDTR_LIMIT, dt->limit);
1880 vmcs_writel(GUEST_GDTR_BASE, dt->base);
1881}
1882
1883static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
1884{
1885 struct kvm_segment var;
1886 u32 ar;
1887
1888 vmx_get_segment(vcpu, &var, seg);
1889 ar = vmx_segment_access_rights(&var);
1890
1891 if (var.base != (var.selector << 4))
1892 return false;
1893 if (var.limit != 0xffff)
1894 return false;
1895 if (ar != 0xf3)
1896 return false;
1897
1898 return true;
1899}
1900
1901static bool code_segment_valid(struct kvm_vcpu *vcpu)
1902{
1903 struct kvm_segment cs;
1904 unsigned int cs_rpl;
1905
1906 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
1907 cs_rpl = cs.selector & SELECTOR_RPL_MASK;
1908
1909 if (cs.unusable)
1910 return false;
1911 if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
1912 return false;
1913 if (!cs.s)
1914 return false;
1915 if (cs.type & AR_TYPE_WRITEABLE_MASK) {
1916 if (cs.dpl > cs_rpl)
1917 return false;
1918 } else {
1919 if (cs.dpl != cs_rpl)
1920 return false;
1921 }
1922 if (!cs.present)
1923 return false;
1924
1925 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
1926 return true;
1927}
1928
1929static bool stack_segment_valid(struct kvm_vcpu *vcpu)
1930{
1931 struct kvm_segment ss;
1932 unsigned int ss_rpl;
1933
1934 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
1935 ss_rpl = ss.selector & SELECTOR_RPL_MASK;
1936
1937 if (ss.unusable)
1938 return true;
1939 if (ss.type != 3 && ss.type != 7)
1940 return false;
1941 if (!ss.s)
1942 return false;
1943 if (ss.dpl != ss_rpl) /* DPL != RPL */
1944 return false;
1945 if (!ss.present)
1946 return false;
1947
1948 return true;
1949}
1950
1951static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
1952{
1953 struct kvm_segment var;
1954 unsigned int rpl;
1955
1956 vmx_get_segment(vcpu, &var, seg);
1957 rpl = var.selector & SELECTOR_RPL_MASK;
1958
1959 if (var.unusable)
1960 return true;
1961 if (!var.s)
1962 return false;
1963 if (!var.present)
1964 return false;
1965 if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
1966 if (var.dpl < rpl) /* DPL < RPL */
1967 return false;
1968 }
1969
1970 /* TODO: Add other members to kvm_segment_field to allow checking for other access
1971 * rights flags
1972 */
1973 return true;
1974}
1975
1976static bool tr_valid(struct kvm_vcpu *vcpu)
1977{
1978 struct kvm_segment tr;
1979
1980 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
1981
1982 if (tr.unusable)
1983 return false;
1984 if (tr.selector & SELECTOR_TI_MASK) /* TI = 1 */
1985 return false;
1986 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
1987 return false;
1988 if (!tr.present)
1989 return false;
1990
1991 return true;
1992}
1993
1994static bool ldtr_valid(struct kvm_vcpu *vcpu)
1995{
1996 struct kvm_segment ldtr;
1997
1998 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
1999
2000 if (ldtr.unusable)
2001 return true;
2002 if (ldtr.selector & SELECTOR_TI_MASK) /* TI = 1 */
2003 return false;
2004 if (ldtr.type != 2)
2005 return false;
2006 if (!ldtr.present)
2007 return false;
2008
2009 return true;
2010}
2011
2012static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
2013{
2014 struct kvm_segment cs, ss;
2015
2016 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
2017 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
2018
2019 return ((cs.selector & SELECTOR_RPL_MASK) ==
2020 (ss.selector & SELECTOR_RPL_MASK));
2021}
2022
2023/*
2024 * Check if guest state is valid. Returns true if valid, false if
2025 * not.
2026 * We assume that registers are always usable
2027 */
2028static bool guest_state_valid(struct kvm_vcpu *vcpu)
2029{
2030 /* real mode guest state checks */
2031 if (!(vcpu->arch.cr0 & X86_CR0_PE)) {
2032 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
2033 return false;
2034 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
2035 return false;
2036 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
2037 return false;
2038 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
2039 return false;
2040 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
2041 return false;
2042 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
2043 return false;
2044 } else {
2045 /* protected mode guest state checks */
2046 if (!cs_ss_rpl_check(vcpu))
2047 return false;
2048 if (!code_segment_valid(vcpu))
2049 return false;
2050 if (!stack_segment_valid(vcpu))
2051 return false;
2052 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
2053 return false;
2054 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
2055 return false;
2056 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
2057 return false;
2058 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
2059 return false;
2060 if (!tr_valid(vcpu))
2061 return false;
2062 if (!ldtr_valid(vcpu))
2063 return false;
2064 }
2065 /* TODO:
2066 * - Add checks on RIP
2067 * - Add checks on RFLAGS
2068 */
2069
2070 return true;
2071}
2072
2073static int init_rmode_tss(struct kvm *kvm)
2074{
2075 gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
2076 u16 data = 0;
2077 int ret = 0;
2078 int r;
2079
2080 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2081 if (r < 0)
2082 goto out;
2083 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
2084 r = kvm_write_guest_page(kvm, fn++, &data,
2085 TSS_IOPB_BASE_OFFSET, sizeof(u16));
2086 if (r < 0)
2087 goto out;
2088 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
2089 if (r < 0)
2090 goto out;
2091 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2092 if (r < 0)
2093 goto out;
2094 data = ~0;
2095 r = kvm_write_guest_page(kvm, fn, &data,
2096 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
2097 sizeof(u8));
2098 if (r < 0)
2099 goto out;
2100
2101 ret = 1;
2102out:
2103 return ret;
2104}
2105
2106static int init_rmode_identity_map(struct kvm *kvm)
2107{
2108 int i, r, ret;
2109 pfn_t identity_map_pfn;
2110 u32 tmp;
2111
2112 if (!enable_ept)
2113 return 1;
2114 if (unlikely(!kvm->arch.ept_identity_pagetable)) {
2115 printk(KERN_ERR "EPT: identity-mapping pagetable "
2116 "haven't been allocated!\n");
2117 return 0;
2118 }
2119 if (likely(kvm->arch.ept_identity_pagetable_done))
2120 return 1;
2121 ret = 0;
2122 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
2123 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
2124 if (r < 0)
2125 goto out;
2126 /* Set up identity-mapping pagetable for EPT in real mode */
2127 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
2128 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
2129 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
2130 r = kvm_write_guest_page(kvm, identity_map_pfn,
2131 &tmp, i * sizeof(tmp), sizeof(tmp));
2132 if (r < 0)
2133 goto out;
2134 }
2135 kvm->arch.ept_identity_pagetable_done = true;
2136 ret = 1;
2137out:
2138 return ret;
2139}
2140
2141static void seg_setup(int seg)
2142{
2143 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2144 unsigned int ar;
2145
2146 vmcs_write16(sf->selector, 0);
2147 vmcs_writel(sf->base, 0);
2148 vmcs_write32(sf->limit, 0xffff);
2149 if (enable_unrestricted_guest) {
2150 ar = 0x93;
2151 if (seg == VCPU_SREG_CS)
2152 ar |= 0x08; /* code segment */
2153 } else
2154 ar = 0xf3;
2155
2156 vmcs_write32(sf->ar_bytes, ar);
2157}
2158
2159static int alloc_apic_access_page(struct kvm *kvm)
2160{
2161 struct kvm_userspace_memory_region kvm_userspace_mem;
2162 int r = 0;
2163
2164 down_write(&kvm->slots_lock);
2165 if (kvm->arch.apic_access_page)
2166 goto out;
2167 kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
2168 kvm_userspace_mem.flags = 0;
2169 kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
2170 kvm_userspace_mem.memory_size = PAGE_SIZE;
2171 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2172 if (r)
2173 goto out;
2174
2175 kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
2176out:
2177 up_write(&kvm->slots_lock);
2178 return r;
2179}
2180
2181static int alloc_identity_pagetable(struct kvm *kvm)
2182{
2183 struct kvm_userspace_memory_region kvm_userspace_mem;
2184 int r = 0;
2185
2186 down_write(&kvm->slots_lock);
2187 if (kvm->arch.ept_identity_pagetable)
2188 goto out;
2189 kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
2190 kvm_userspace_mem.flags = 0;
2191 kvm_userspace_mem.guest_phys_addr =
2192 kvm->arch.ept_identity_map_addr;
2193 kvm_userspace_mem.memory_size = PAGE_SIZE;
2194 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2195 if (r)
2196 goto out;
2197
2198 kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
2199 kvm->arch.ept_identity_map_addr >> PAGE_SHIFT);
2200out:
2201 up_write(&kvm->slots_lock);
2202 return r;
2203}
2204
2205static void allocate_vpid(struct vcpu_vmx *vmx)
2206{
2207 int vpid;
2208
2209 vmx->vpid = 0;
2210 if (!enable_vpid)
2211 return;
2212 spin_lock(&vmx_vpid_lock);
2213 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
2214 if (vpid < VMX_NR_VPIDS) {
2215 vmx->vpid = vpid;
2216 __set_bit(vpid, vmx_vpid_bitmap);
2217 }
2218 spin_unlock(&vmx_vpid_lock);
2219}
2220
2221static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, u32 msr)
2222{
2223 int f = sizeof(unsigned long);
2224
2225 if (!cpu_has_vmx_msr_bitmap())
2226 return;
2227
2228 /*
2229 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
2230 * have the write-low and read-high bitmap offsets the wrong way round.
2231 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
2232 */
2233 if (msr <= 0x1fff) {
2234 __clear_bit(msr, msr_bitmap + 0x000 / f); /* read-low */
2235 __clear_bit(msr, msr_bitmap + 0x800 / f); /* write-low */
2236 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
2237 msr &= 0x1fff;
2238 __clear_bit(msr, msr_bitmap + 0x400 / f); /* read-high */
2239 __clear_bit(msr, msr_bitmap + 0xc00 / f); /* write-high */
2240 }
2241}
2242
2243static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
2244{
2245 if (!longmode_only)
2246 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy, msr);
2247 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode, msr);
2248}
2249
2250/*
2251 * Sets up the vmcs for emulated real mode.
2252 */
2253static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
2254{
2255 u32 host_sysenter_cs, msr_low, msr_high;
2256 u32 junk;
2257 u64 host_pat, tsc_this, tsc_base;
2258 unsigned long a;
2259 struct descriptor_table dt;
2260 int i;
2261 unsigned long kvm_vmx_return;
2262 u32 exec_control;
2263
2264 /* I/O */
2265 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
2266 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
2267
2268 if (cpu_has_vmx_msr_bitmap())
2269 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
2270
2271 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
2272
2273 /* Control */
2274 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
2275 vmcs_config.pin_based_exec_ctrl);
2276
2277 exec_control = vmcs_config.cpu_based_exec_ctrl;
2278 if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
2279 exec_control &= ~CPU_BASED_TPR_SHADOW;
2280#ifdef CONFIG_X86_64
2281 exec_control |= CPU_BASED_CR8_STORE_EXITING |
2282 CPU_BASED_CR8_LOAD_EXITING;
2283#endif
2284 }
2285 if (!enable_ept)
2286 exec_control |= CPU_BASED_CR3_STORE_EXITING |
2287 CPU_BASED_CR3_LOAD_EXITING |
2288 CPU_BASED_INVLPG_EXITING;
2289 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
2290
2291 if (cpu_has_secondary_exec_ctrls()) {
2292 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
2293 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2294 exec_control &=
2295 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
2296 if (vmx->vpid == 0)
2297 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
2298 if (!enable_ept)
2299 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
2300 if (!enable_unrestricted_guest)
2301 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
2302 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
2303 }
2304
2305 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
2306 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
2307 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
2308
2309 vmcs_writel(HOST_CR0, read_cr0()); /* 22.2.3 */
2310 vmcs_writel(HOST_CR4, read_cr4()); /* 22.2.3, 22.2.5 */
2311 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
2312
2313 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
2314 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2315 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2316 vmcs_write16(HOST_FS_SELECTOR, kvm_read_fs()); /* 22.2.4 */
2317 vmcs_write16(HOST_GS_SELECTOR, kvm_read_gs()); /* 22.2.4 */
2318 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2319#ifdef CONFIG_X86_64
2320 rdmsrl(MSR_FS_BASE, a);
2321 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
2322 rdmsrl(MSR_GS_BASE, a);
2323 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
2324#else
2325 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
2326 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
2327#endif
2328
2329 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
2330
2331 kvm_get_idt(&dt);
2332 vmcs_writel(HOST_IDTR_BASE, dt.base); /* 22.2.4 */
2333
2334 asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
2335 vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
2336 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
2337 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
2338 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
2339
2340 rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
2341 vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
2342 rdmsrl(MSR_IA32_SYSENTER_ESP, a);
2343 vmcs_writel(HOST_IA32_SYSENTER_ESP, a); /* 22.2.3 */
2344 rdmsrl(MSR_IA32_SYSENTER_EIP, a);
2345 vmcs_writel(HOST_IA32_SYSENTER_EIP, a); /* 22.2.3 */
2346
2347 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
2348 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2349 host_pat = msr_low | ((u64) msr_high << 32);
2350 vmcs_write64(HOST_IA32_PAT, host_pat);
2351 }
2352 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2353 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2354 host_pat = msr_low | ((u64) msr_high << 32);
2355 /* Write the default value follow host pat */
2356 vmcs_write64(GUEST_IA32_PAT, host_pat);
2357 /* Keep arch.pat sync with GUEST_IA32_PAT */
2358 vmx->vcpu.arch.pat = host_pat;
2359 }
2360
2361 for (i = 0; i < NR_VMX_MSR; ++i) {
2362 u32 index = vmx_msr_index[i];
2363 u32 data_low, data_high;
2364 u64 data;
2365 int j = vmx->nmsrs;
2366
2367 if (rdmsr_safe(index, &data_low, &data_high) < 0)
2368 continue;
2369 if (wrmsr_safe(index, data_low, data_high) < 0)
2370 continue;
2371 data = data_low | ((u64)data_high << 32);
2372 vmx->host_msrs[j].index = index;
2373 vmx->host_msrs[j].reserved = 0;
2374 vmx->host_msrs[j].data = data;
2375 vmx->guest_msrs[j] = vmx->host_msrs[j];
2376 ++vmx->nmsrs;
2377 }
2378
2379 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
2380
2381 /* 22.2.1, 20.8.1 */
2382 vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
2383
2384 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
2385 vmcs_writel(CR4_GUEST_HOST_MASK, KVM_GUEST_CR4_MASK);
2386
2387 tsc_base = vmx->vcpu.kvm->arch.vm_init_tsc;
2388 rdtscll(tsc_this);
2389 if (tsc_this < vmx->vcpu.kvm->arch.vm_init_tsc)
2390 tsc_base = tsc_this;
2391
2392 guest_write_tsc(0, tsc_base);
2393
2394 return 0;
2395}
2396
2397static int init_rmode(struct kvm *kvm)
2398{
2399 if (!init_rmode_tss(kvm))
2400 return 0;
2401 if (!init_rmode_identity_map(kvm))
2402 return 0;
2403 return 1;
2404}
2405
2406static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
2407{
2408 struct vcpu_vmx *vmx = to_vmx(vcpu);
2409 u64 msr;
2410 int ret;
2411
2412 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
2413 down_read(&vcpu->kvm->slots_lock);
2414 if (!init_rmode(vmx->vcpu.kvm)) {
2415 ret = -ENOMEM;
2416 goto out;
2417 }
2418
2419 vmx->rmode.vm86_active = 0;
2420
2421 vmx->soft_vnmi_blocked = 0;
2422
2423 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
2424 kvm_set_cr8(&vmx->vcpu, 0);
2425 msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
2426 if (kvm_vcpu_is_bsp(&vmx->vcpu))
2427 msr |= MSR_IA32_APICBASE_BSP;
2428 kvm_set_apic_base(&vmx->vcpu, msr);
2429
2430 fx_init(&vmx->vcpu);
2431
2432 seg_setup(VCPU_SREG_CS);
2433 /*
2434 * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
2435 * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4. Sigh.
2436 */
2437 if (kvm_vcpu_is_bsp(&vmx->vcpu)) {
2438 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
2439 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
2440 } else {
2441 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
2442 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
2443 }
2444
2445 seg_setup(VCPU_SREG_DS);
2446 seg_setup(VCPU_SREG_ES);
2447 seg_setup(VCPU_SREG_FS);
2448 seg_setup(VCPU_SREG_GS);
2449 seg_setup(VCPU_SREG_SS);
2450
2451 vmcs_write16(GUEST_TR_SELECTOR, 0);
2452 vmcs_writel(GUEST_TR_BASE, 0);
2453 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
2454 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2455
2456 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
2457 vmcs_writel(GUEST_LDTR_BASE, 0);
2458 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
2459 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
2460
2461 vmcs_write32(GUEST_SYSENTER_CS, 0);
2462 vmcs_writel(GUEST_SYSENTER_ESP, 0);
2463 vmcs_writel(GUEST_SYSENTER_EIP, 0);
2464
2465 vmcs_writel(GUEST_RFLAGS, 0x02);
2466 if (kvm_vcpu_is_bsp(&vmx->vcpu))
2467 kvm_rip_write(vcpu, 0xfff0);
2468 else
2469 kvm_rip_write(vcpu, 0);
2470 kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
2471
2472 vmcs_writel(GUEST_DR7, 0x400);
2473
2474 vmcs_writel(GUEST_GDTR_BASE, 0);
2475 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
2476
2477 vmcs_writel(GUEST_IDTR_BASE, 0);
2478 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
2479
2480 vmcs_write32(GUEST_ACTIVITY_STATE, 0);
2481 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
2482 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
2483
2484 /* Special registers */
2485 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
2486
2487 setup_msrs(vmx);
2488
2489 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
2490
2491 if (cpu_has_vmx_tpr_shadow()) {
2492 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
2493 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
2494 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
2495 page_to_phys(vmx->vcpu.arch.apic->regs_page));
2496 vmcs_write32(TPR_THRESHOLD, 0);
2497 }
2498
2499 if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2500 vmcs_write64(APIC_ACCESS_ADDR,
2501 page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
2502
2503 if (vmx->vpid != 0)
2504 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2505
2506 vmx->vcpu.arch.cr0 = 0x60000010;
2507 vmx_set_cr0(&vmx->vcpu, vmx->vcpu.arch.cr0); /* enter rmode */
2508 vmx_set_cr4(&vmx->vcpu, 0);
2509 vmx_set_efer(&vmx->vcpu, 0);
2510 vmx_fpu_activate(&vmx->vcpu);
2511 update_exception_bitmap(&vmx->vcpu);
2512
2513 vpid_sync_vcpu_all(vmx);
2514
2515 ret = 0;
2516
2517 /* HACK: Don't enable emulation on guest boot/reset */
2518 vmx->emulation_required = 0;
2519
2520out:
2521 up_read(&vcpu->kvm->slots_lock);
2522 return ret;
2523}
2524
2525static void enable_irq_window(struct kvm_vcpu *vcpu)
2526{
2527 u32 cpu_based_vm_exec_control;
2528
2529 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2530 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2531 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2532}
2533
2534static void enable_nmi_window(struct kvm_vcpu *vcpu)
2535{
2536 u32 cpu_based_vm_exec_control;
2537
2538 if (!cpu_has_virtual_nmis()) {
2539 enable_irq_window(vcpu);
2540 return;
2541 }
2542
2543 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2544 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
2545 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2546}
2547
2548static void vmx_inject_irq(struct kvm_vcpu *vcpu)
2549{
2550 struct vcpu_vmx *vmx = to_vmx(vcpu);
2551 uint32_t intr;
2552 int irq = vcpu->arch.interrupt.nr;
2553
2554 trace_kvm_inj_virq(irq);
2555
2556 ++vcpu->stat.irq_injections;
2557 if (vmx->rmode.vm86_active) {
2558 vmx->rmode.irq.pending = true;
2559 vmx->rmode.irq.vector = irq;
2560 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
2561 if (vcpu->arch.interrupt.soft)
2562 vmx->rmode.irq.rip +=
2563 vmx->vcpu.arch.event_exit_inst_len;
2564 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2565 irq | INTR_TYPE_SOFT_INTR | INTR_INFO_VALID_MASK);
2566 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
2567 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
2568 return;
2569 }
2570 intr = irq | INTR_INFO_VALID_MASK;
2571 if (vcpu->arch.interrupt.soft) {
2572 intr |= INTR_TYPE_SOFT_INTR;
2573 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2574 vmx->vcpu.arch.event_exit_inst_len);
2575 } else
2576 intr |= INTR_TYPE_EXT_INTR;
2577 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
2578}
2579
2580static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
2581{
2582 struct vcpu_vmx *vmx = to_vmx(vcpu);
2583
2584 if (!cpu_has_virtual_nmis()) {
2585 /*
2586 * Tracking the NMI-blocked state in software is built upon
2587 * finding the next open IRQ window. This, in turn, depends on
2588 * well-behaving guests: They have to keep IRQs disabled at
2589 * least as long as the NMI handler runs. Otherwise we may
2590 * cause NMI nesting, maybe breaking the guest. But as this is
2591 * highly unlikely, we can live with the residual risk.
2592 */
2593 vmx->soft_vnmi_blocked = 1;
2594 vmx->vnmi_blocked_time = 0;
2595 }
2596
2597 ++vcpu->stat.nmi_injections;
2598 if (vmx->rmode.vm86_active) {
2599 vmx->rmode.irq.pending = true;
2600 vmx->rmode.irq.vector = NMI_VECTOR;
2601 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
2602 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2603 NMI_VECTOR | INTR_TYPE_SOFT_INTR |
2604 INTR_INFO_VALID_MASK);
2605 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
2606 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
2607 return;
2608 }
2609 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2610 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
2611}
2612
2613static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
2614{
2615 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
2616 return 0;
2617
2618 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2619 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS |
2620 GUEST_INTR_STATE_NMI));
2621}
2622
2623static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
2624{
2625 return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
2626 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2627 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
2628}
2629
2630static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
2631{
2632 int ret;
2633 struct kvm_userspace_memory_region tss_mem = {
2634 .slot = TSS_PRIVATE_MEMSLOT,
2635 .guest_phys_addr = addr,
2636 .memory_size = PAGE_SIZE * 3,
2637 .flags = 0,
2638 };
2639
2640 ret = kvm_set_memory_region(kvm, &tss_mem, 0);
2641 if (ret)
2642 return ret;
2643 kvm->arch.tss_addr = addr;
2644 return 0;
2645}
2646
2647static int handle_rmode_exception(struct kvm_vcpu *vcpu,
2648 int vec, u32 err_code)
2649{
2650 /*
2651 * Instruction with address size override prefix opcode 0x67
2652 * Cause the #SS fault with 0 error code in VM86 mode.
2653 */
2654 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
2655 if (emulate_instruction(vcpu, NULL, 0, 0, 0) == EMULATE_DONE)
2656 return 1;
2657 /*
2658 * Forward all other exceptions that are valid in real mode.
2659 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
2660 * the required debugging infrastructure rework.
2661 */
2662 switch (vec) {
2663 case DB_VECTOR:
2664 if (vcpu->guest_debug &
2665 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
2666 return 0;
2667 kvm_queue_exception(vcpu, vec);
2668 return 1;
2669 case BP_VECTOR:
2670 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
2671 return 0;
2672 /* fall through */
2673 case DE_VECTOR:
2674 case OF_VECTOR:
2675 case BR_VECTOR:
2676 case UD_VECTOR:
2677 case DF_VECTOR:
2678 case SS_VECTOR:
2679 case GP_VECTOR:
2680 case MF_VECTOR:
2681 kvm_queue_exception(vcpu, vec);
2682 return 1;
2683 }
2684 return 0;
2685}
2686
2687/*
2688 * Trigger machine check on the host. We assume all the MSRs are already set up
2689 * by the CPU and that we still run on the same CPU as the MCE occurred on.
2690 * We pass a fake environment to the machine check handler because we want
2691 * the guest to be always treated like user space, no matter what context
2692 * it used internally.
2693 */
2694static void kvm_machine_check(void)
2695{
2696#if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
2697 struct pt_regs regs = {
2698 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
2699 .flags = X86_EFLAGS_IF,
2700 };
2701
2702 do_machine_check(&regs, 0);
2703#endif
2704}
2705
2706static int handle_machine_check(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2707{
2708 /* already handled by vcpu_run */
2709 return 1;
2710}
2711
2712static int handle_exception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2713{
2714 struct vcpu_vmx *vmx = to_vmx(vcpu);
2715 u32 intr_info, ex_no, error_code;
2716 unsigned long cr2, rip, dr6;
2717 u32 vect_info;
2718 enum emulation_result er;
2719
2720 vect_info = vmx->idt_vectoring_info;
2721 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
2722
2723 if (is_machine_check(intr_info))
2724 return handle_machine_check(vcpu, kvm_run);
2725
2726 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
2727 !is_page_fault(intr_info))
2728 printk(KERN_ERR "%s: unexpected, vectoring info 0x%x "
2729 "intr info 0x%x\n", __func__, vect_info, intr_info);
2730
2731 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
2732 return 1; /* already handled by vmx_vcpu_run() */
2733
2734 if (is_no_device(intr_info)) {
2735 vmx_fpu_activate(vcpu);
2736 return 1;
2737 }
2738
2739 if (is_invalid_opcode(intr_info)) {
2740 er = emulate_instruction(vcpu, kvm_run, 0, 0, EMULTYPE_TRAP_UD);
2741 if (er != EMULATE_DONE)
2742 kvm_queue_exception(vcpu, UD_VECTOR);
2743 return 1;
2744 }
2745
2746 error_code = 0;
2747 rip = kvm_rip_read(vcpu);
2748 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
2749 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
2750 if (is_page_fault(intr_info)) {
2751 /* EPT won't cause page fault directly */
2752 if (enable_ept)
2753 BUG();
2754 cr2 = vmcs_readl(EXIT_QUALIFICATION);
2755 trace_kvm_page_fault(cr2, error_code);
2756
2757 if (kvm_event_needs_reinjection(vcpu))
2758 kvm_mmu_unprotect_page_virt(vcpu, cr2);
2759 return kvm_mmu_page_fault(vcpu, cr2, error_code);
2760 }
2761
2762 if (vmx->rmode.vm86_active &&
2763 handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
2764 error_code)) {
2765 if (vcpu->arch.halt_request) {
2766 vcpu->arch.halt_request = 0;
2767 return kvm_emulate_halt(vcpu);
2768 }
2769 return 1;
2770 }
2771
2772 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
2773 switch (ex_no) {
2774 case DB_VECTOR:
2775 dr6 = vmcs_readl(EXIT_QUALIFICATION);
2776 if (!(vcpu->guest_debug &
2777 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
2778 vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
2779 kvm_queue_exception(vcpu, DB_VECTOR);
2780 return 1;
2781 }
2782 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
2783 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
2784 /* fall through */
2785 case BP_VECTOR:
2786 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2787 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
2788 kvm_run->debug.arch.exception = ex_no;
2789 break;
2790 default:
2791 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
2792 kvm_run->ex.exception = ex_no;
2793 kvm_run->ex.error_code = error_code;
2794 break;
2795 }
2796 return 0;
2797}
2798
2799static int handle_external_interrupt(struct kvm_vcpu *vcpu,
2800 struct kvm_run *kvm_run)
2801{
2802 ++vcpu->stat.irq_exits;
2803 return 1;
2804}
2805
2806static int handle_triple_fault(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2807{
2808 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2809 return 0;
2810}
2811
2812static int handle_io(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2813{
2814 unsigned long exit_qualification;
2815 int size, in, string;
2816 unsigned port;
2817
2818 ++vcpu->stat.io_exits;
2819 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2820 string = (exit_qualification & 16) != 0;
2821
2822 if (string) {
2823 if (emulate_instruction(vcpu,
2824 kvm_run, 0, 0, 0) == EMULATE_DO_MMIO)
2825 return 0;
2826 return 1;
2827 }
2828
2829 size = (exit_qualification & 7) + 1;
2830 in = (exit_qualification & 8) != 0;
2831 port = exit_qualification >> 16;
2832
2833 skip_emulated_instruction(vcpu);
2834 return kvm_emulate_pio(vcpu, kvm_run, in, size, port);
2835}
2836
2837static void
2838vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
2839{
2840 /*
2841 * Patch in the VMCALL instruction:
2842 */
2843 hypercall[0] = 0x0f;
2844 hypercall[1] = 0x01;
2845 hypercall[2] = 0xc1;
2846}
2847
2848static int handle_cr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2849{
2850 unsigned long exit_qualification, val;
2851 int cr;
2852 int reg;
2853
2854 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2855 cr = exit_qualification & 15;
2856 reg = (exit_qualification >> 8) & 15;
2857 switch ((exit_qualification >> 4) & 3) {
2858 case 0: /* mov to cr */
2859 val = kvm_register_read(vcpu, reg);
2860 trace_kvm_cr_write(cr, val);
2861 switch (cr) {
2862 case 0:
2863 kvm_set_cr0(vcpu, val);
2864 skip_emulated_instruction(vcpu);
2865 return 1;
2866 case 3:
2867 kvm_set_cr3(vcpu, val);
2868 skip_emulated_instruction(vcpu);
2869 return 1;
2870 case 4:
2871 kvm_set_cr4(vcpu, val);
2872 skip_emulated_instruction(vcpu);
2873 return 1;
2874 case 8: {
2875 u8 cr8_prev = kvm_get_cr8(vcpu);
2876 u8 cr8 = kvm_register_read(vcpu, reg);
2877 kvm_set_cr8(vcpu, cr8);
2878 skip_emulated_instruction(vcpu);
2879 if (irqchip_in_kernel(vcpu->kvm))
2880 return 1;
2881 if (cr8_prev <= cr8)
2882 return 1;
2883 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
2884 return 0;
2885 }
2886 };
2887 break;
2888 case 2: /* clts */
2889 vmx_fpu_deactivate(vcpu);
2890 vcpu->arch.cr0 &= ~X86_CR0_TS;
2891 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
2892 vmx_fpu_activate(vcpu);
2893 skip_emulated_instruction(vcpu);
2894 return 1;
2895 case 1: /*mov from cr*/
2896 switch (cr) {
2897 case 3:
2898 kvm_register_write(vcpu, reg, vcpu->arch.cr3);
2899 trace_kvm_cr_read(cr, vcpu->arch.cr3);
2900 skip_emulated_instruction(vcpu);
2901 return 1;
2902 case 8:
2903 val = kvm_get_cr8(vcpu);
2904 kvm_register_write(vcpu, reg, val);
2905 trace_kvm_cr_read(cr, val);
2906 skip_emulated_instruction(vcpu);
2907 return 1;
2908 }
2909 break;
2910 case 3: /* lmsw */
2911 kvm_lmsw(vcpu, (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f);
2912
2913 skip_emulated_instruction(vcpu);
2914 return 1;
2915 default:
2916 break;
2917 }
2918 kvm_run->exit_reason = 0;
2919 pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
2920 (int)(exit_qualification >> 4) & 3, cr);
2921 return 0;
2922}
2923
2924static int handle_dr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2925{
2926 unsigned long exit_qualification;
2927 unsigned long val;
2928 int dr, reg;
2929
2930 dr = vmcs_readl(GUEST_DR7);
2931 if (dr & DR7_GD) {
2932 /*
2933 * As the vm-exit takes precedence over the debug trap, we
2934 * need to emulate the latter, either for the host or the
2935 * guest debugging itself.
2936 */
2937 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
2938 kvm_run->debug.arch.dr6 = vcpu->arch.dr6;
2939 kvm_run->debug.arch.dr7 = dr;
2940 kvm_run->debug.arch.pc =
2941 vmcs_readl(GUEST_CS_BASE) +
2942 vmcs_readl(GUEST_RIP);
2943 kvm_run->debug.arch.exception = DB_VECTOR;
2944 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2945 return 0;
2946 } else {
2947 vcpu->arch.dr7 &= ~DR7_GD;
2948 vcpu->arch.dr6 |= DR6_BD;
2949 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
2950 kvm_queue_exception(vcpu, DB_VECTOR);
2951 return 1;
2952 }
2953 }
2954
2955 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2956 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
2957 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
2958 if (exit_qualification & TYPE_MOV_FROM_DR) {
2959 switch (dr) {
2960 case 0 ... 3:
2961 val = vcpu->arch.db[dr];
2962 break;
2963 case 6:
2964 val = vcpu->arch.dr6;
2965 break;
2966 case 7:
2967 val = vcpu->arch.dr7;
2968 break;
2969 default:
2970 val = 0;
2971 }
2972 kvm_register_write(vcpu, reg, val);
2973 } else {
2974 val = vcpu->arch.regs[reg];
2975 switch (dr) {
2976 case 0 ... 3:
2977 vcpu->arch.db[dr] = val;
2978 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
2979 vcpu->arch.eff_db[dr] = val;
2980 break;
2981 case 4 ... 5:
2982 if (vcpu->arch.cr4 & X86_CR4_DE)
2983 kvm_queue_exception(vcpu, UD_VECTOR);
2984 break;
2985 case 6:
2986 if (val & 0xffffffff00000000ULL) {
2987 kvm_queue_exception(vcpu, GP_VECTOR);
2988 break;
2989 }
2990 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
2991 break;
2992 case 7:
2993 if (val & 0xffffffff00000000ULL) {
2994 kvm_queue_exception(vcpu, GP_VECTOR);
2995 break;
2996 }
2997 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
2998 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
2999 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
3000 vcpu->arch.switch_db_regs =
3001 (val & DR7_BP_EN_MASK);
3002 }
3003 break;
3004 }
3005 }
3006 skip_emulated_instruction(vcpu);
3007 return 1;
3008}
3009
3010static int handle_cpuid(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3011{
3012 kvm_emulate_cpuid(vcpu);
3013 return 1;
3014}
3015
3016static int handle_rdmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3017{
3018 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3019 u64 data;
3020
3021 if (vmx_get_msr(vcpu, ecx, &data)) {
3022 kvm_inject_gp(vcpu, 0);
3023 return 1;
3024 }
3025
3026 trace_kvm_msr_read(ecx, data);
3027
3028 /* FIXME: handling of bits 32:63 of rax, rdx */
3029 vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
3030 vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
3031 skip_emulated_instruction(vcpu);
3032 return 1;
3033}
3034
3035static int handle_wrmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3036{
3037 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3038 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
3039 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
3040
3041 trace_kvm_msr_write(ecx, data);
3042
3043 if (vmx_set_msr(vcpu, ecx, data) != 0) {
3044 kvm_inject_gp(vcpu, 0);
3045 return 1;
3046 }
3047
3048 skip_emulated_instruction(vcpu);
3049 return 1;
3050}
3051
3052static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu,
3053 struct kvm_run *kvm_run)
3054{
3055 return 1;
3056}
3057
3058static int handle_interrupt_window(struct kvm_vcpu *vcpu,
3059 struct kvm_run *kvm_run)
3060{
3061 u32 cpu_based_vm_exec_control;
3062
3063 /* clear pending irq */
3064 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3065 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
3066 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3067
3068 ++vcpu->stat.irq_window_exits;
3069
3070 /*
3071 * If the user space waits to inject interrupts, exit as soon as
3072 * possible
3073 */
3074 if (!irqchip_in_kernel(vcpu->kvm) &&
3075 kvm_run->request_interrupt_window &&
3076 !kvm_cpu_has_interrupt(vcpu)) {
3077 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
3078 return 0;
3079 }
3080 return 1;
3081}
3082
3083static int handle_halt(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3084{
3085 skip_emulated_instruction(vcpu);
3086 return kvm_emulate_halt(vcpu);
3087}
3088
3089static int handle_vmcall(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3090{
3091 skip_emulated_instruction(vcpu);
3092 kvm_emulate_hypercall(vcpu);
3093 return 1;
3094}
3095
3096static int handle_vmx_insn(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3097{
3098 kvm_queue_exception(vcpu, UD_VECTOR);
3099 return 1;
3100}
3101
3102static int handle_invlpg(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3103{
3104 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3105
3106 kvm_mmu_invlpg(vcpu, exit_qualification);
3107 skip_emulated_instruction(vcpu);
3108 return 1;
3109}
3110
3111static int handle_wbinvd(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3112{
3113 skip_emulated_instruction(vcpu);
3114 /* TODO: Add support for VT-d/pass-through device */
3115 return 1;
3116}
3117
3118static int handle_apic_access(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3119{
3120 unsigned long exit_qualification;
3121 enum emulation_result er;
3122 unsigned long offset;
3123
3124 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3125 offset = exit_qualification & 0xffful;
3126
3127 er = emulate_instruction(vcpu, kvm_run, 0, 0, 0);
3128
3129 if (er != EMULATE_DONE) {
3130 printk(KERN_ERR
3131 "Fail to handle apic access vmexit! Offset is 0x%lx\n",
3132 offset);
3133 return -ENOEXEC;
3134 }
3135 return 1;
3136}
3137
3138static int handle_task_switch(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3139{
3140 struct vcpu_vmx *vmx = to_vmx(vcpu);
3141 unsigned long exit_qualification;
3142 u16 tss_selector;
3143 int reason, type, idt_v;
3144
3145 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
3146 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
3147
3148 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3149
3150 reason = (u32)exit_qualification >> 30;
3151 if (reason == TASK_SWITCH_GATE && idt_v) {
3152 switch (type) {
3153 case INTR_TYPE_NMI_INTR:
3154 vcpu->arch.nmi_injected = false;
3155 if (cpu_has_virtual_nmis())
3156 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3157 GUEST_INTR_STATE_NMI);
3158 break;
3159 case INTR_TYPE_EXT_INTR:
3160 case INTR_TYPE_SOFT_INTR:
3161 kvm_clear_interrupt_queue(vcpu);
3162 break;
3163 case INTR_TYPE_HARD_EXCEPTION:
3164 case INTR_TYPE_SOFT_EXCEPTION:
3165 kvm_clear_exception_queue(vcpu);
3166 break;
3167 default:
3168 break;
3169 }
3170 }
3171 tss_selector = exit_qualification;
3172
3173 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
3174 type != INTR_TYPE_EXT_INTR &&
3175 type != INTR_TYPE_NMI_INTR))
3176 skip_emulated_instruction(vcpu);
3177
3178 if (!kvm_task_switch(vcpu, tss_selector, reason))
3179 return 0;
3180
3181 /* clear all local breakpoint enable flags */
3182 vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
3183
3184 /*
3185 * TODO: What about debug traps on tss switch?
3186 * Are we supposed to inject them and update dr6?
3187 */
3188
3189 return 1;
3190}
3191
3192static int handle_ept_violation(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3193{
3194 unsigned long exit_qualification;
3195 gpa_t gpa;
3196 int gla_validity;
3197
3198 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3199
3200 if (exit_qualification & (1 << 6)) {
3201 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
3202 return -EINVAL;
3203 }
3204
3205 gla_validity = (exit_qualification >> 7) & 0x3;
3206 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
3207 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
3208 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
3209 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
3210 vmcs_readl(GUEST_LINEAR_ADDRESS));
3211 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
3212 (long unsigned int)exit_qualification);
3213 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
3214 kvm_run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
3215 return 0;
3216 }
3217
3218 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3219 trace_kvm_page_fault(gpa, exit_qualification);
3220 return kvm_mmu_page_fault(vcpu, gpa & PAGE_MASK, 0);
3221}
3222
3223static u64 ept_rsvd_mask(u64 spte, int level)
3224{
3225 int i;
3226 u64 mask = 0;
3227
3228 for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
3229 mask |= (1ULL << i);
3230
3231 if (level > 2)
3232 /* bits 7:3 reserved */
3233 mask |= 0xf8;
3234 else if (level == 2) {
3235 if (spte & (1ULL << 7))
3236 /* 2MB ref, bits 20:12 reserved */
3237 mask |= 0x1ff000;
3238 else
3239 /* bits 6:3 reserved */
3240 mask |= 0x78;
3241 }
3242
3243 return mask;
3244}
3245
3246static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
3247 int level)
3248{
3249 printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
3250
3251 /* 010b (write-only) */
3252 WARN_ON((spte & 0x7) == 0x2);
3253
3254 /* 110b (write/execute) */
3255 WARN_ON((spte & 0x7) == 0x6);
3256
3257 /* 100b (execute-only) and value not supported by logical processor */
3258 if (!cpu_has_vmx_ept_execute_only())
3259 WARN_ON((spte & 0x7) == 0x4);
3260
3261 /* not 000b */
3262 if ((spte & 0x7)) {
3263 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
3264
3265 if (rsvd_bits != 0) {
3266 printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
3267 __func__, rsvd_bits);
3268 WARN_ON(1);
3269 }
3270
3271 if (level == 1 || (level == 2 && (spte & (1ULL << 7)))) {
3272 u64 ept_mem_type = (spte & 0x38) >> 3;
3273
3274 if (ept_mem_type == 2 || ept_mem_type == 3 ||
3275 ept_mem_type == 7) {
3276 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
3277 __func__, ept_mem_type);
3278 WARN_ON(1);
3279 }
3280 }
3281 }
3282}
3283
3284static int handle_ept_misconfig(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3285{
3286 u64 sptes[4];
3287 int nr_sptes, i;
3288 gpa_t gpa;
3289
3290 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3291
3292 printk(KERN_ERR "EPT: Misconfiguration.\n");
3293 printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
3294
3295 nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
3296
3297 for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
3298 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
3299
3300 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
3301 kvm_run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
3302
3303 return 0;
3304}
3305
3306static int handle_nmi_window(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3307{
3308 u32 cpu_based_vm_exec_control;
3309
3310 /* clear pending NMI */
3311 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3312 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
3313 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3314 ++vcpu->stat.nmi_window_exits;
3315
3316 return 1;
3317}
3318
3319static void handle_invalid_guest_state(struct kvm_vcpu *vcpu,
3320 struct kvm_run *kvm_run)
3321{
3322 struct vcpu_vmx *vmx = to_vmx(vcpu);
3323 enum emulation_result err = EMULATE_DONE;
3324
3325 local_irq_enable();
3326 preempt_enable();
3327
3328 while (!guest_state_valid(vcpu)) {
3329 err = emulate_instruction(vcpu, kvm_run, 0, 0, 0);
3330
3331 if (err == EMULATE_DO_MMIO)
3332 break;
3333
3334 if (err != EMULATE_DONE) {
3335 kvm_report_emulation_failure(vcpu, "emulation failure");
3336 break;
3337 }
3338
3339 if (signal_pending(current))
3340 break;
3341 if (need_resched())
3342 schedule();
3343 }
3344
3345 preempt_disable();
3346 local_irq_disable();
3347
3348 vmx->invalid_state_emulation_result = err;
3349}
3350
3351/*
3352 * The exit handlers return 1 if the exit was handled fully and guest execution
3353 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
3354 * to be done to userspace and return 0.
3355 */
3356static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu,
3357 struct kvm_run *kvm_run) = {
3358 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
3359 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
3360 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
3361 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
3362 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
3363 [EXIT_REASON_CR_ACCESS] = handle_cr,
3364 [EXIT_REASON_DR_ACCESS] = handle_dr,
3365 [EXIT_REASON_CPUID] = handle_cpuid,
3366 [EXIT_REASON_MSR_READ] = handle_rdmsr,
3367 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
3368 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
3369 [EXIT_REASON_HLT] = handle_halt,
3370 [EXIT_REASON_INVLPG] = handle_invlpg,
3371 [EXIT_REASON_VMCALL] = handle_vmcall,
3372 [EXIT_REASON_VMCLEAR] = handle_vmx_insn,
3373 [EXIT_REASON_VMLAUNCH] = handle_vmx_insn,
3374 [EXIT_REASON_VMPTRLD] = handle_vmx_insn,
3375 [EXIT_REASON_VMPTRST] = handle_vmx_insn,
3376 [EXIT_REASON_VMREAD] = handle_vmx_insn,
3377 [EXIT_REASON_VMRESUME] = handle_vmx_insn,
3378 [EXIT_REASON_VMWRITE] = handle_vmx_insn,
3379 [EXIT_REASON_VMOFF] = handle_vmx_insn,
3380 [EXIT_REASON_VMON] = handle_vmx_insn,
3381 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
3382 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
3383 [EXIT_REASON_WBINVD] = handle_wbinvd,
3384 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
3385 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
3386 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
3387 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
3388};
3389
3390static const int kvm_vmx_max_exit_handlers =
3391 ARRAY_SIZE(kvm_vmx_exit_handlers);
3392
3393/*
3394 * The guest has exited. See if we can fix it or if we need userspace
3395 * assistance.
3396 */
3397static int vmx_handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
3398{
3399 struct vcpu_vmx *vmx = to_vmx(vcpu);
3400 u32 exit_reason = vmx->exit_reason;
3401 u32 vectoring_info = vmx->idt_vectoring_info;
3402
3403 trace_kvm_exit(exit_reason, kvm_rip_read(vcpu));
3404
3405 /* If we need to emulate an MMIO from handle_invalid_guest_state
3406 * we just return 0 */
3407 if (vmx->emulation_required && emulate_invalid_guest_state) {
3408 if (guest_state_valid(vcpu))
3409 vmx->emulation_required = 0;
3410 return vmx->invalid_state_emulation_result != EMULATE_DO_MMIO;
3411 }
3412
3413 /* Access CR3 don't cause VMExit in paging mode, so we need
3414 * to sync with guest real CR3. */
3415 if (enable_ept && is_paging(vcpu))
3416 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3417
3418 if (unlikely(vmx->fail)) {
3419 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3420 kvm_run->fail_entry.hardware_entry_failure_reason
3421 = vmcs_read32(VM_INSTRUCTION_ERROR);
3422 return 0;
3423 }
3424
3425 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
3426 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
3427 exit_reason != EXIT_REASON_EPT_VIOLATION &&
3428 exit_reason != EXIT_REASON_TASK_SWITCH))
3429 printk(KERN_WARNING "%s: unexpected, valid vectoring info "
3430 "(0x%x) and exit reason is 0x%x\n",
3431 __func__, vectoring_info, exit_reason);
3432
3433 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked)) {
3434 if (vmx_interrupt_allowed(vcpu)) {
3435 vmx->soft_vnmi_blocked = 0;
3436 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
3437 vcpu->arch.nmi_pending) {
3438 /*
3439 * This CPU don't support us in finding the end of an
3440 * NMI-blocked window if the guest runs with IRQs
3441 * disabled. So we pull the trigger after 1 s of
3442 * futile waiting, but inform the user about this.
3443 */
3444 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
3445 "state on VCPU %d after 1 s timeout\n",
3446 __func__, vcpu->vcpu_id);
3447 vmx->soft_vnmi_blocked = 0;
3448 }
3449 }
3450
3451 if (exit_reason < kvm_vmx_max_exit_handlers
3452 && kvm_vmx_exit_handlers[exit_reason])
3453 return kvm_vmx_exit_handlers[exit_reason](vcpu, kvm_run);
3454 else {
3455 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
3456 kvm_run->hw.hardware_exit_reason = exit_reason;
3457 }
3458 return 0;
3459}
3460
3461static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3462{
3463 if (irr == -1 || tpr < irr) {
3464 vmcs_write32(TPR_THRESHOLD, 0);
3465 return;
3466 }
3467
3468 vmcs_write32(TPR_THRESHOLD, irr);
3469}
3470
3471static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
3472{
3473 u32 exit_intr_info;
3474 u32 idt_vectoring_info = vmx->idt_vectoring_info;
3475 bool unblock_nmi;
3476 u8 vector;
3477 int type;
3478 bool idtv_info_valid;
3479
3480 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
3481
3482 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
3483
3484 /* Handle machine checks before interrupts are enabled */
3485 if ((vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
3486 || (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI
3487 && is_machine_check(exit_intr_info)))
3488 kvm_machine_check();
3489
3490 /* We need to handle NMIs before interrupts are enabled */
3491 if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
3492 (exit_intr_info & INTR_INFO_VALID_MASK))
3493 asm("int $2");
3494
3495 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
3496
3497 if (cpu_has_virtual_nmis()) {
3498 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
3499 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
3500 /*
3501 * SDM 3: 27.7.1.2 (September 2008)
3502 * Re-set bit "block by NMI" before VM entry if vmexit caused by
3503 * a guest IRET fault.
3504 * SDM 3: 23.2.2 (September 2008)
3505 * Bit 12 is undefined in any of the following cases:
3506 * If the VM exit sets the valid bit in the IDT-vectoring
3507 * information field.
3508 * If the VM exit is due to a double fault.
3509 */
3510 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
3511 vector != DF_VECTOR && !idtv_info_valid)
3512 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3513 GUEST_INTR_STATE_NMI);
3514 } else if (unlikely(vmx->soft_vnmi_blocked))
3515 vmx->vnmi_blocked_time +=
3516 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
3517
3518 vmx->vcpu.arch.nmi_injected = false;
3519 kvm_clear_exception_queue(&vmx->vcpu);
3520 kvm_clear_interrupt_queue(&vmx->vcpu);
3521
3522 if (!idtv_info_valid)
3523 return;
3524
3525 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
3526 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
3527
3528 switch (type) {
3529 case INTR_TYPE_NMI_INTR:
3530 vmx->vcpu.arch.nmi_injected = true;
3531 /*
3532 * SDM 3: 27.7.1.2 (September 2008)
3533 * Clear bit "block by NMI" before VM entry if a NMI
3534 * delivery faulted.
3535 */
3536 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
3537 GUEST_INTR_STATE_NMI);
3538 break;
3539 case INTR_TYPE_SOFT_EXCEPTION:
3540 vmx->vcpu.arch.event_exit_inst_len =
3541 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3542 /* fall through */
3543 case INTR_TYPE_HARD_EXCEPTION:
3544 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
3545 u32 err = vmcs_read32(IDT_VECTORING_ERROR_CODE);
3546 kvm_queue_exception_e(&vmx->vcpu, vector, err);
3547 } else
3548 kvm_queue_exception(&vmx->vcpu, vector);
3549 break;
3550 case INTR_TYPE_SOFT_INTR:
3551 vmx->vcpu.arch.event_exit_inst_len =
3552 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3553 /* fall through */
3554 case INTR_TYPE_EXT_INTR:
3555 kvm_queue_interrupt(&vmx->vcpu, vector,
3556 type == INTR_TYPE_SOFT_INTR);
3557 break;
3558 default:
3559 break;
3560 }
3561}
3562
3563/*
3564 * Failure to inject an interrupt should give us the information
3565 * in IDT_VECTORING_INFO_FIELD. However, if the failure occurs
3566 * when fetching the interrupt redirection bitmap in the real-mode
3567 * tss, this doesn't happen. So we do it ourselves.
3568 */
3569static void fixup_rmode_irq(struct vcpu_vmx *vmx)
3570{
3571 vmx->rmode.irq.pending = 0;
3572 if (kvm_rip_read(&vmx->vcpu) + 1 != vmx->rmode.irq.rip)
3573 return;
3574 kvm_rip_write(&vmx->vcpu, vmx->rmode.irq.rip);
3575 if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) {
3576 vmx->idt_vectoring_info &= ~VECTORING_INFO_TYPE_MASK;
3577 vmx->idt_vectoring_info |= INTR_TYPE_EXT_INTR;
3578 return;
3579 }
3580 vmx->idt_vectoring_info =
3581 VECTORING_INFO_VALID_MASK
3582 | INTR_TYPE_EXT_INTR
3583 | vmx->rmode.irq.vector;
3584}
3585
3586#ifdef CONFIG_X86_64
3587#define R "r"
3588#define Q "q"
3589#else
3590#define R "e"
3591#define Q "l"
3592#endif
3593
3594static void vmx_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
3595{
3596 struct vcpu_vmx *vmx = to_vmx(vcpu);
3597
3598 if (enable_ept && is_paging(vcpu)) {
3599 vmcs_writel(GUEST_CR3, vcpu->arch.cr3);
3600 ept_load_pdptrs(vcpu);
3601 }
3602 /* Record the guest's net vcpu time for enforced NMI injections. */
3603 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
3604 vmx->entry_time = ktime_get();
3605
3606 /* Handle invalid guest state instead of entering VMX */
3607 if (vmx->emulation_required && emulate_invalid_guest_state) {
3608 handle_invalid_guest_state(vcpu, kvm_run);
3609 return;
3610 }
3611
3612 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
3613 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
3614 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
3615 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
3616
3617 /* When single-stepping over STI and MOV SS, we must clear the
3618 * corresponding interruptibility bits in the guest state. Otherwise
3619 * vmentry fails as it then expects bit 14 (BS) in pending debug
3620 * exceptions being set, but that's not correct for the guest debugging
3621 * case. */
3622 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3623 vmx_set_interrupt_shadow(vcpu, 0);
3624
3625 /*
3626 * Loading guest fpu may have cleared host cr0.ts
3627 */
3628 vmcs_writel(HOST_CR0, read_cr0());
3629
3630 set_debugreg(vcpu->arch.dr6, 6);
3631
3632 asm(
3633 /* Store host registers */
3634 "push %%"R"dx; push %%"R"bp;"
3635 "push %%"R"cx \n\t"
3636 "cmp %%"R"sp, %c[host_rsp](%0) \n\t"
3637 "je 1f \n\t"
3638 "mov %%"R"sp, %c[host_rsp](%0) \n\t"
3639 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
3640 "1: \n\t"
3641 /* Reload cr2 if changed */
3642 "mov %c[cr2](%0), %%"R"ax \n\t"
3643 "mov %%cr2, %%"R"dx \n\t"
3644 "cmp %%"R"ax, %%"R"dx \n\t"
3645 "je 2f \n\t"
3646 "mov %%"R"ax, %%cr2 \n\t"
3647 "2: \n\t"
3648 /* Check if vmlaunch of vmresume is needed */
3649 "cmpl $0, %c[launched](%0) \n\t"
3650 /* Load guest registers. Don't clobber flags. */
3651 "mov %c[rax](%0), %%"R"ax \n\t"
3652 "mov %c[rbx](%0), %%"R"bx \n\t"
3653 "mov %c[rdx](%0), %%"R"dx \n\t"
3654 "mov %c[rsi](%0), %%"R"si \n\t"
3655 "mov %c[rdi](%0), %%"R"di \n\t"
3656 "mov %c[rbp](%0), %%"R"bp \n\t"
3657#ifdef CONFIG_X86_64
3658 "mov %c[r8](%0), %%r8 \n\t"
3659 "mov %c[r9](%0), %%r9 \n\t"
3660 "mov %c[r10](%0), %%r10 \n\t"
3661 "mov %c[r11](%0), %%r11 \n\t"
3662 "mov %c[r12](%0), %%r12 \n\t"
3663 "mov %c[r13](%0), %%r13 \n\t"
3664 "mov %c[r14](%0), %%r14 \n\t"
3665 "mov %c[r15](%0), %%r15 \n\t"
3666#endif
3667 "mov %c[rcx](%0), %%"R"cx \n\t" /* kills %0 (ecx) */
3668
3669 /* Enter guest mode */
3670 "jne .Llaunched \n\t"
3671 __ex(ASM_VMX_VMLAUNCH) "\n\t"
3672 "jmp .Lkvm_vmx_return \n\t"
3673 ".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t"
3674 ".Lkvm_vmx_return: "
3675 /* Save guest registers, load host registers, keep flags */
3676 "xchg %0, (%%"R"sp) \n\t"
3677 "mov %%"R"ax, %c[rax](%0) \n\t"
3678 "mov %%"R"bx, %c[rbx](%0) \n\t"
3679 "push"Q" (%%"R"sp); pop"Q" %c[rcx](%0) \n\t"
3680 "mov %%"R"dx, %c[rdx](%0) \n\t"
3681 "mov %%"R"si, %c[rsi](%0) \n\t"
3682 "mov %%"R"di, %c[rdi](%0) \n\t"
3683 "mov %%"R"bp, %c[rbp](%0) \n\t"
3684#ifdef CONFIG_X86_64
3685 "mov %%r8, %c[r8](%0) \n\t"
3686 "mov %%r9, %c[r9](%0) \n\t"
3687 "mov %%r10, %c[r10](%0) \n\t"
3688 "mov %%r11, %c[r11](%0) \n\t"
3689 "mov %%r12, %c[r12](%0) \n\t"
3690 "mov %%r13, %c[r13](%0) \n\t"
3691 "mov %%r14, %c[r14](%0) \n\t"
3692 "mov %%r15, %c[r15](%0) \n\t"
3693#endif
3694 "mov %%cr2, %%"R"ax \n\t"
3695 "mov %%"R"ax, %c[cr2](%0) \n\t"
3696
3697 "pop %%"R"bp; pop %%"R"bp; pop %%"R"dx \n\t"
3698 "setbe %c[fail](%0) \n\t"
3699 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
3700 [launched]"i"(offsetof(struct vcpu_vmx, launched)),
3701 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
3702 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
3703 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
3704 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
3705 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
3706 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
3707 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
3708 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
3709 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
3710#ifdef CONFIG_X86_64
3711 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
3712 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
3713 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
3714 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
3715 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
3716 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
3717 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
3718 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
3719#endif
3720 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2))
3721 : "cc", "memory"
3722 , R"bx", R"di", R"si"
3723#ifdef CONFIG_X86_64
3724 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
3725#endif
3726 );
3727
3728 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
3729 | (1 << VCPU_EXREG_PDPTR));
3730 vcpu->arch.regs_dirty = 0;
3731
3732 get_debugreg(vcpu->arch.dr6, 6);
3733
3734 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
3735 if (vmx->rmode.irq.pending)
3736 fixup_rmode_irq(vmx);
3737
3738 asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
3739 vmx->launched = 1;
3740
3741 vmx_complete_interrupts(vmx);
3742}
3743
3744#undef R
3745#undef Q
3746
3747static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
3748{
3749 struct vcpu_vmx *vmx = to_vmx(vcpu);
3750
3751 if (vmx->vmcs) {
3752 vcpu_clear(vmx);
3753 free_vmcs(vmx->vmcs);
3754 vmx->vmcs = NULL;
3755 }
3756}
3757
3758static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
3759{
3760 struct vcpu_vmx *vmx = to_vmx(vcpu);
3761
3762 spin_lock(&vmx_vpid_lock);
3763 if (vmx->vpid != 0)
3764 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
3765 spin_unlock(&vmx_vpid_lock);
3766 vmx_free_vmcs(vcpu);
3767 kfree(vmx->host_msrs);
3768 kfree(vmx->guest_msrs);
3769 kvm_vcpu_uninit(vcpu);
3770 kmem_cache_free(kvm_vcpu_cache, vmx);
3771}
3772
3773static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
3774{
3775 int err;
3776 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
3777 int cpu;
3778
3779 if (!vmx)
3780 return ERR_PTR(-ENOMEM);
3781
3782 allocate_vpid(vmx);
3783
3784 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
3785 if (err)
3786 goto free_vcpu;
3787
3788 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
3789 if (!vmx->guest_msrs) {
3790 err = -ENOMEM;
3791 goto uninit_vcpu;
3792 }
3793
3794 vmx->host_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
3795 if (!vmx->host_msrs)
3796 goto free_guest_msrs;
3797
3798 vmx->vmcs = alloc_vmcs();
3799 if (!vmx->vmcs)
3800 goto free_msrs;
3801
3802 vmcs_clear(vmx->vmcs);
3803
3804 cpu = get_cpu();
3805 vmx_vcpu_load(&vmx->vcpu, cpu);
3806 err = vmx_vcpu_setup(vmx);
3807 vmx_vcpu_put(&vmx->vcpu);
3808 put_cpu();
3809 if (err)
3810 goto free_vmcs;
3811 if (vm_need_virtualize_apic_accesses(kvm))
3812 if (alloc_apic_access_page(kvm) != 0)
3813 goto free_vmcs;
3814
3815 if (enable_ept) {
3816 if (!kvm->arch.ept_identity_map_addr)
3817 kvm->arch.ept_identity_map_addr =
3818 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3819 if (alloc_identity_pagetable(kvm) != 0)
3820 goto free_vmcs;
3821 }
3822
3823 return &vmx->vcpu;
3824
3825free_vmcs:
3826 free_vmcs(vmx->vmcs);
3827free_msrs:
3828 kfree(vmx->host_msrs);
3829free_guest_msrs:
3830 kfree(vmx->guest_msrs);
3831uninit_vcpu:
3832 kvm_vcpu_uninit(&vmx->vcpu);
3833free_vcpu:
3834 kmem_cache_free(kvm_vcpu_cache, vmx);
3835 return ERR_PTR(err);
3836}
3837
3838static void __init vmx_check_processor_compat(void *rtn)
3839{
3840 struct vmcs_config vmcs_conf;
3841
3842 *(int *)rtn = 0;
3843 if (setup_vmcs_config(&vmcs_conf) < 0)
3844 *(int *)rtn = -EIO;
3845 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
3846 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
3847 smp_processor_id());
3848 *(int *)rtn = -EIO;
3849 }
3850}
3851
3852static int get_ept_level(void)
3853{
3854 return VMX_EPT_DEFAULT_GAW + 1;
3855}
3856
3857static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
3858{
3859 u64 ret;
3860
3861 /* For VT-d and EPT combination
3862 * 1. MMIO: always map as UC
3863 * 2. EPT with VT-d:
3864 * a. VT-d without snooping control feature: can't guarantee the
3865 * result, try to trust guest.
3866 * b. VT-d with snooping control feature: snooping control feature of
3867 * VT-d engine can guarantee the cache correctness. Just set it
3868 * to WB to keep consistent with host. So the same as item 3.
3869 * 3. EPT without VT-d: always map as WB and set IGMT=1 to keep
3870 * consistent with host MTRR
3871 */
3872 if (is_mmio)
3873 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
3874 else if (vcpu->kvm->arch.iommu_domain &&
3875 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY))
3876 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
3877 VMX_EPT_MT_EPTE_SHIFT;
3878 else
3879 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
3880 | VMX_EPT_IGMT_BIT;
3881
3882 return ret;
3883}
3884
3885static const struct trace_print_flags vmx_exit_reasons_str[] = {
3886 { EXIT_REASON_EXCEPTION_NMI, "exception" },
3887 { EXIT_REASON_EXTERNAL_INTERRUPT, "ext_irq" },
3888 { EXIT_REASON_TRIPLE_FAULT, "triple_fault" },
3889 { EXIT_REASON_NMI_WINDOW, "nmi_window" },
3890 { EXIT_REASON_IO_INSTRUCTION, "io_instruction" },
3891 { EXIT_REASON_CR_ACCESS, "cr_access" },
3892 { EXIT_REASON_DR_ACCESS, "dr_access" },
3893 { EXIT_REASON_CPUID, "cpuid" },
3894 { EXIT_REASON_MSR_READ, "rdmsr" },
3895 { EXIT_REASON_MSR_WRITE, "wrmsr" },
3896 { EXIT_REASON_PENDING_INTERRUPT, "interrupt_window" },
3897 { EXIT_REASON_HLT, "halt" },
3898 { EXIT_REASON_INVLPG, "invlpg" },
3899 { EXIT_REASON_VMCALL, "hypercall" },
3900 { EXIT_REASON_TPR_BELOW_THRESHOLD, "tpr_below_thres" },
3901 { EXIT_REASON_APIC_ACCESS, "apic_access" },
3902 { EXIT_REASON_WBINVD, "wbinvd" },
3903 { EXIT_REASON_TASK_SWITCH, "task_switch" },
3904 { EXIT_REASON_EPT_VIOLATION, "ept_violation" },
3905 { -1, NULL }
3906};
3907
3908static bool vmx_gb_page_enable(void)
3909{
3910 return false;
3911}
3912
3913static struct kvm_x86_ops vmx_x86_ops = {
3914 .cpu_has_kvm_support = cpu_has_kvm_support,
3915 .disabled_by_bios = vmx_disabled_by_bios,
3916 .hardware_setup = hardware_setup,
3917 .hardware_unsetup = hardware_unsetup,
3918 .check_processor_compatibility = vmx_check_processor_compat,
3919 .hardware_enable = hardware_enable,
3920 .hardware_disable = hardware_disable,
3921 .cpu_has_accelerated_tpr = report_flexpriority,
3922
3923 .vcpu_create = vmx_create_vcpu,
3924 .vcpu_free = vmx_free_vcpu,
3925 .vcpu_reset = vmx_vcpu_reset,
3926
3927 .prepare_guest_switch = vmx_save_host_state,
3928 .vcpu_load = vmx_vcpu_load,
3929 .vcpu_put = vmx_vcpu_put,
3930
3931 .set_guest_debug = set_guest_debug,
3932 .get_msr = vmx_get_msr,
3933 .set_msr = vmx_set_msr,
3934 .get_segment_base = vmx_get_segment_base,
3935 .get_segment = vmx_get_segment,
3936 .set_segment = vmx_set_segment,
3937 .get_cpl = vmx_get_cpl,
3938 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
3939 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
3940 .set_cr0 = vmx_set_cr0,
3941 .set_cr3 = vmx_set_cr3,
3942 .set_cr4 = vmx_set_cr4,
3943 .set_efer = vmx_set_efer,
3944 .get_idt = vmx_get_idt,
3945 .set_idt = vmx_set_idt,
3946 .get_gdt = vmx_get_gdt,
3947 .set_gdt = vmx_set_gdt,
3948 .cache_reg = vmx_cache_reg,
3949 .get_rflags = vmx_get_rflags,
3950 .set_rflags = vmx_set_rflags,
3951
3952 .tlb_flush = vmx_flush_tlb,
3953
3954 .run = vmx_vcpu_run,
3955 .handle_exit = vmx_handle_exit,
3956 .skip_emulated_instruction = skip_emulated_instruction,
3957 .set_interrupt_shadow = vmx_set_interrupt_shadow,
3958 .get_interrupt_shadow = vmx_get_interrupt_shadow,
3959 .patch_hypercall = vmx_patch_hypercall,
3960 .set_irq = vmx_inject_irq,
3961 .set_nmi = vmx_inject_nmi,
3962 .queue_exception = vmx_queue_exception,
3963 .interrupt_allowed = vmx_interrupt_allowed,
3964 .nmi_allowed = vmx_nmi_allowed,
3965 .enable_nmi_window = enable_nmi_window,
3966 .enable_irq_window = enable_irq_window,
3967 .update_cr8_intercept = update_cr8_intercept,
3968
3969 .set_tss_addr = vmx_set_tss_addr,
3970 .get_tdp_level = get_ept_level,
3971 .get_mt_mask = vmx_get_mt_mask,
3972
3973 .exit_reasons_str = vmx_exit_reasons_str,
3974 .gb_page_enable = vmx_gb_page_enable,
3975};
3976
3977static int __init vmx_init(void)
3978{
3979 int r;
3980
3981 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
3982 if (!vmx_io_bitmap_a)
3983 return -ENOMEM;
3984
3985 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
3986 if (!vmx_io_bitmap_b) {
3987 r = -ENOMEM;
3988 goto out;
3989 }
3990
3991 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
3992 if (!vmx_msr_bitmap_legacy) {
3993 r = -ENOMEM;
3994 goto out1;
3995 }
3996
3997 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
3998 if (!vmx_msr_bitmap_longmode) {
3999 r = -ENOMEM;
4000 goto out2;
4001 }
4002
4003 /*
4004 * Allow direct access to the PC debug port (it is often used for I/O
4005 * delays, but the vmexits simply slow things down).
4006 */
4007 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
4008 clear_bit(0x80, vmx_io_bitmap_a);
4009
4010 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
4011
4012 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
4013 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
4014
4015 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
4016
4017 r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx), THIS_MODULE);
4018 if (r)
4019 goto out3;
4020
4021 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
4022 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
4023 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
4024 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
4025 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
4026 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
4027
4028 if (enable_ept) {
4029 bypass_guest_pf = 0;
4030 kvm_mmu_set_base_ptes(VMX_EPT_READABLE_MASK |
4031 VMX_EPT_WRITABLE_MASK);
4032 kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull,
4033 VMX_EPT_EXECUTABLE_MASK);
4034 kvm_enable_tdp();
4035 } else
4036 kvm_disable_tdp();
4037
4038 if (bypass_guest_pf)
4039 kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull);
4040
4041 ept_sync_global();
4042
4043 return 0;
4044
4045out3:
4046 free_page((unsigned long)vmx_msr_bitmap_longmode);
4047out2:
4048 free_page((unsigned long)vmx_msr_bitmap_legacy);
4049out1:
4050 free_page((unsigned long)vmx_io_bitmap_b);
4051out:
4052 free_page((unsigned long)vmx_io_bitmap_a);
4053 return r;
4054}
4055
4056static void __exit vmx_exit(void)
4057{
4058 free_page((unsigned long)vmx_msr_bitmap_legacy);
4059 free_page((unsigned long)vmx_msr_bitmap_longmode);
4060 free_page((unsigned long)vmx_io_bitmap_b);
4061 free_page((unsigned long)vmx_io_bitmap_a);
4062
4063 kvm_exit();
4064}
4065
4066module_init(vmx_init)
4067module_exit(vmx_exit)
ubuntu-bionic-kernel mirror