]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame_incremental - arch/x86/kvm/svm.c
projects / mirror_ubuntu-artful-kernel.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
KVM: SVM: Coding style cleanup
[mirror_ubuntu-artful-kernel.git] / arch / x86 / kvm / svm.c
... / ...
CommitLineData
1/*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * AMD SVM support
5 *
6 * Copyright (C) 2006 Qumranet, Inc.
7 *
8 * Authors:
9 * Yaniv Kamay <yaniv@qumranet.com>
10 * Avi Kivity <avi@qumranet.com>
11 *
12 * This work is licensed under the terms of the GNU GPL, version 2. See
13 * the COPYING file in the top-level directory.
14 *
15 */
16#include <linux/kvm_host.h>
17
18#include "irq.h"
19#include "mmu.h"
20#include "kvm_cache_regs.h"
21#include "x86.h"
22
23#include <linux/module.h>
24#include <linux/kernel.h>
25#include <linux/vmalloc.h>
26#include <linux/highmem.h>
27#include <linux/sched.h>
28#include <linux/ftrace_event.h>
29#include <linux/slab.h>
30
31#include <asm/desc.h>
32
33#include <asm/virtext.h>
34#include "trace.h"
35
36#define __ex(x) __kvm_handle_fault_on_reboot(x)
37
38MODULE_AUTHOR("Qumranet");
39MODULE_LICENSE("GPL");
40
41#define IOPM_ALLOC_ORDER 2
42#define MSRPM_ALLOC_ORDER 1
43
44#define SEG_TYPE_LDT 2
45#define SEG_TYPE_BUSY_TSS16 3
46
47#define SVM_FEATURE_NPT (1 << 0)
48#define SVM_FEATURE_LBRV (1 << 1)
49#define SVM_FEATURE_SVML (1 << 2)
50#define SVM_FEATURE_NRIP (1 << 3)
51#define SVM_FEATURE_PAUSE_FILTER (1 << 10)
52
53#define NESTED_EXIT_HOST 0 /* Exit handled on host level */
54#define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */
55#define NESTED_EXIT_CONTINUE 2 /* Further checks needed */
56
57#define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
58
59static const u32 host_save_user_msrs[] = {
60#ifdef CONFIG_X86_64
61 MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE,
62 MSR_FS_BASE,
63#endif
64 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
65};
66
67#define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
68
69struct kvm_vcpu;
70
71struct nested_state {
72 struct vmcb *hsave;
73 u64 hsave_msr;
74 u64 vmcb;
75
76 /* These are the merged vectors */
77 u32 *msrpm;
78
79 /* gpa pointers to the real vectors */
80 u64 vmcb_msrpm;
81
82 /* A VMEXIT is required but not yet emulated */
83 bool exit_required;
84
85 /* cache for intercepts of the guest */
86 u16 intercept_cr_read;
87 u16 intercept_cr_write;
88 u16 intercept_dr_read;
89 u16 intercept_dr_write;
90 u32 intercept_exceptions;
91 u64 intercept;
92
93};
94
95struct vcpu_svm {
96 struct kvm_vcpu vcpu;
97 struct vmcb *vmcb;
98 unsigned long vmcb_pa;
99 struct svm_cpu_data *svm_data;
100 uint64_t asid_generation;
101 uint64_t sysenter_esp;
102 uint64_t sysenter_eip;
103
104 u64 next_rip;
105
106 u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS];
107 u64 host_gs_base;
108
109 u32 *msrpm;
110
111 struct nested_state nested;
112
113 bool nmi_singlestep;
114
115 unsigned int3_injected;
116 unsigned long int3_rip;
117};
118
119/* enable NPT for AMD64 and X86 with PAE */
120#if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
121static bool npt_enabled = true;
122#else
123static bool npt_enabled;
124#endif
125static int npt = 1;
126
127module_param(npt, int, S_IRUGO);
128
129static int nested = 1;
130module_param(nested, int, S_IRUGO);
131
132static void svm_flush_tlb(struct kvm_vcpu *vcpu);
133static void svm_complete_interrupts(struct vcpu_svm *svm);
134
135static int nested_svm_exit_handled(struct vcpu_svm *svm);
136static int nested_svm_intercept(struct vcpu_svm *svm);
137static int nested_svm_vmexit(struct vcpu_svm *svm);
138static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
139 bool has_error_code, u32 error_code);
140
141static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
142{
143 return container_of(vcpu, struct vcpu_svm, vcpu);
144}
145
146static inline bool is_nested(struct vcpu_svm *svm)
147{
148 return svm->nested.vmcb;
149}
150
151static inline void enable_gif(struct vcpu_svm *svm)
152{
153 svm->vcpu.arch.hflags |= HF_GIF_MASK;
154}
155
156static inline void disable_gif(struct vcpu_svm *svm)
157{
158 svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
159}
160
161static inline bool gif_set(struct vcpu_svm *svm)
162{
163 return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
164}
165
166static unsigned long iopm_base;
167
168struct kvm_ldttss_desc {
169 u16 limit0;
170 u16 base0;
171 unsigned base1:8, type:5, dpl:2, p:1;
172 unsigned limit1:4, zero0:3, g:1, base2:8;
173 u32 base3;
174 u32 zero1;
175} __attribute__((packed));
176
177struct svm_cpu_data {
178 int cpu;
179
180 u64 asid_generation;
181 u32 max_asid;
182 u32 next_asid;
183 struct kvm_ldttss_desc *tss_desc;
184
185 struct page *save_area;
186};
187
188static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
189static uint32_t svm_features;
190
191struct svm_init_data {
192 int cpu;
193 int r;
194};
195
196static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
197
198#define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
199#define MSRS_RANGE_SIZE 2048
200#define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
201
202#define MAX_INST_SIZE 15
203
204static inline u32 svm_has(u32 feat)
205{
206 return svm_features & feat;
207}
208
209static inline void clgi(void)
210{
211 asm volatile (__ex(SVM_CLGI));
212}
213
214static inline void stgi(void)
215{
216 asm volatile (__ex(SVM_STGI));
217}
218
219static inline void invlpga(unsigned long addr, u32 asid)
220{
221 asm volatile (__ex(SVM_INVLPGA) : : "a"(addr), "c"(asid));
222}
223
224static inline void force_new_asid(struct kvm_vcpu *vcpu)
225{
226 to_svm(vcpu)->asid_generation--;
227}
228
229static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
230{
231 force_new_asid(vcpu);
232}
233
234static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
235{
236 if (!npt_enabled && !(efer & EFER_LMA))
237 efer &= ~EFER_LME;
238
239 to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME;
240 vcpu->arch.efer = efer;
241}
242
243static int is_external_interrupt(u32 info)
244{
245 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
246 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
247}
248
249static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
250{
251 struct vcpu_svm *svm = to_svm(vcpu);
252 u32 ret = 0;
253
254 if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
255 ret |= KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS;
256 return ret & mask;
257}
258
259static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
260{
261 struct vcpu_svm *svm = to_svm(vcpu);
262
263 if (mask == 0)
264 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
265 else
266 svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
267
268}
269
270static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
271{
272 struct vcpu_svm *svm = to_svm(vcpu);
273
274 if (!svm->next_rip) {
275 if (emulate_instruction(vcpu, 0, 0, EMULTYPE_SKIP) !=
276 EMULATE_DONE)
277 printk(KERN_DEBUG "%s: NOP\n", __func__);
278 return;
279 }
280 if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE)
281 printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
282 __func__, kvm_rip_read(vcpu), svm->next_rip);
283
284 kvm_rip_write(vcpu, svm->next_rip);
285 svm_set_interrupt_shadow(vcpu, 0);
286}
287
288static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
289 bool has_error_code, u32 error_code)
290{
291 struct vcpu_svm *svm = to_svm(vcpu);
292
293 /*
294 * If we are within a nested VM we'd better #VMEXIT and let the guest
295 * handle the exception
296 */
297 if (nested_svm_check_exception(svm, nr, has_error_code, error_code))
298 return;
299
300 if (nr == BP_VECTOR && !svm_has(SVM_FEATURE_NRIP)) {
301 unsigned long rip, old_rip = kvm_rip_read(&svm->vcpu);
302
303 /*
304 * For guest debugging where we have to reinject #BP if some
305 * INT3 is guest-owned:
306 * Emulate nRIP by moving RIP forward. Will fail if injection
307 * raises a fault that is not intercepted. Still better than
308 * failing in all cases.
309 */
310 skip_emulated_instruction(&svm->vcpu);
311 rip = kvm_rip_read(&svm->vcpu);
312 svm->int3_rip = rip + svm->vmcb->save.cs.base;
313 svm->int3_injected = rip - old_rip;
314 }
315
316 svm->vmcb->control.event_inj = nr
317 | SVM_EVTINJ_VALID
318 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
319 | SVM_EVTINJ_TYPE_EXEPT;
320 svm->vmcb->control.event_inj_err = error_code;
321}
322
323static int has_svm(void)
324{
325 const char *msg;
326
327 if (!cpu_has_svm(&msg)) {
328 printk(KERN_INFO "has_svm: %s\n", msg);
329 return 0;
330 }
331
332 return 1;
333}
334
335static void svm_hardware_disable(void *garbage)
336{
337 cpu_svm_disable();
338}
339
340static int svm_hardware_enable(void *garbage)
341{
342
343 struct svm_cpu_data *sd;
344 uint64_t efer;
345 struct desc_ptr gdt_descr;
346 struct desc_struct *gdt;
347 int me = raw_smp_processor_id();
348
349 rdmsrl(MSR_EFER, efer);
350 if (efer & EFER_SVME)
351 return -EBUSY;
352
353 if (!has_svm()) {
354 printk(KERN_ERR "svm_hardware_enable: err EOPNOTSUPP on %d\n",
355 me);
356 return -EINVAL;
357 }
358 sd = per_cpu(svm_data, me);
359
360 if (!sd) {
361 printk(KERN_ERR "svm_hardware_enable: svm_data is NULL on %d\n",
362 me);
363 return -EINVAL;
364 }
365
366 sd->asid_generation = 1;
367 sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
368 sd->next_asid = sd->max_asid + 1;
369
370 kvm_get_gdt(&gdt_descr);
371 gdt = (struct desc_struct *)gdt_descr.address;
372 sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
373
374 wrmsrl(MSR_EFER, efer | EFER_SVME);
375
376 wrmsrl(MSR_VM_HSAVE_PA, page_to_pfn(sd->save_area) << PAGE_SHIFT);
377
378 return 0;
379}
380
381static void svm_cpu_uninit(int cpu)
382{
383 struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id());
384
385 if (!sd)
386 return;
387
388 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
389 __free_page(sd->save_area);
390 kfree(sd);
391}
392
393static int svm_cpu_init(int cpu)
394{
395 struct svm_cpu_data *sd;
396 int r;
397
398 sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
399 if (!sd)
400 return -ENOMEM;
401 sd->cpu = cpu;
402 sd->save_area = alloc_page(GFP_KERNEL);
403 r = -ENOMEM;
404 if (!sd->save_area)
405 goto err_1;
406
407 per_cpu(svm_data, cpu) = sd;
408
409 return 0;
410
411err_1:
412 kfree(sd);
413 return r;
414
415}
416
417static void set_msr_interception(u32 *msrpm, unsigned msr,
418 int read, int write)
419{
420 int i;
421
422 for (i = 0; i < NUM_MSR_MAPS; i++) {
423 if (msr >= msrpm_ranges[i] &&
424 msr < msrpm_ranges[i] + MSRS_IN_RANGE) {
425 u32 msr_offset = (i * MSRS_IN_RANGE + msr -
426 msrpm_ranges[i]) * 2;
427
428 u32 *base = msrpm + (msr_offset / 32);
429 u32 msr_shift = msr_offset % 32;
430 u32 mask = ((write) ? 0 : 2) | ((read) ? 0 : 1);
431 *base = (*base & ~(0x3 << msr_shift)) |
432 (mask << msr_shift);
433 return;
434 }
435 }
436 BUG();
437}
438
439static void svm_vcpu_init_msrpm(u32 *msrpm)
440{
441 memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
442
443#ifdef CONFIG_X86_64
444 set_msr_interception(msrpm, MSR_GS_BASE, 1, 1);
445 set_msr_interception(msrpm, MSR_FS_BASE, 1, 1);
446 set_msr_interception(msrpm, MSR_KERNEL_GS_BASE, 1, 1);
447 set_msr_interception(msrpm, MSR_LSTAR, 1, 1);
448 set_msr_interception(msrpm, MSR_CSTAR, 1, 1);
449 set_msr_interception(msrpm, MSR_SYSCALL_MASK, 1, 1);
450#endif
451 set_msr_interception(msrpm, MSR_K6_STAR, 1, 1);
452 set_msr_interception(msrpm, MSR_IA32_SYSENTER_CS, 1, 1);
453}
454
455static void svm_enable_lbrv(struct vcpu_svm *svm)
456{
457 u32 *msrpm = svm->msrpm;
458
459 svm->vmcb->control.lbr_ctl = 1;
460 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
461 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
462 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
463 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
464}
465
466static void svm_disable_lbrv(struct vcpu_svm *svm)
467{
468 u32 *msrpm = svm->msrpm;
469
470 svm->vmcb->control.lbr_ctl = 0;
471 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
472 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
473 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
474 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
475}
476
477static __init int svm_hardware_setup(void)
478{
479 int cpu;
480 struct page *iopm_pages;
481 void *iopm_va;
482 int r;
483
484 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
485
486 if (!iopm_pages)
487 return -ENOMEM;
488
489 iopm_va = page_address(iopm_pages);
490 memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
491 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
492
493 if (boot_cpu_has(X86_FEATURE_NX))
494 kvm_enable_efer_bits(EFER_NX);
495
496 if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
497 kvm_enable_efer_bits(EFER_FFXSR);
498
499 if (nested) {
500 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
501 kvm_enable_efer_bits(EFER_SVME);
502 }
503
504 for_each_possible_cpu(cpu) {
505 r = svm_cpu_init(cpu);
506 if (r)
507 goto err;
508 }
509
510 svm_features = cpuid_edx(SVM_CPUID_FUNC);
511
512 if (!svm_has(SVM_FEATURE_NPT))
513 npt_enabled = false;
514
515 if (npt_enabled && !npt) {
516 printk(KERN_INFO "kvm: Nested Paging disabled\n");
517 npt_enabled = false;
518 }
519
520 if (npt_enabled) {
521 printk(KERN_INFO "kvm: Nested Paging enabled\n");
522 kvm_enable_tdp();
523 } else
524 kvm_disable_tdp();
525
526 return 0;
527
528err:
529 __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
530 iopm_base = 0;
531 return r;
532}
533
534static __exit void svm_hardware_unsetup(void)
535{
536 int cpu;
537
538 for_each_possible_cpu(cpu)
539 svm_cpu_uninit(cpu);
540
541 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
542 iopm_base = 0;
543}
544
545static void init_seg(struct vmcb_seg *seg)
546{
547 seg->selector = 0;
548 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
549 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
550 seg->limit = 0xffff;
551 seg->base = 0;
552}
553
554static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
555{
556 seg->selector = 0;
557 seg->attrib = SVM_SELECTOR_P_MASK | type;
558 seg->limit = 0xffff;
559 seg->base = 0;
560}
561
562static void init_vmcb(struct vcpu_svm *svm)
563{
564 struct vmcb_control_area *control = &svm->vmcb->control;
565 struct vmcb_save_area *save = &svm->vmcb->save;
566
567 svm->vcpu.fpu_active = 1;
568
569 control->intercept_cr_read = INTERCEPT_CR0_MASK |
570 INTERCEPT_CR3_MASK |
571 INTERCEPT_CR4_MASK;
572
573 control->intercept_cr_write = INTERCEPT_CR0_MASK |
574 INTERCEPT_CR3_MASK |
575 INTERCEPT_CR4_MASK |
576 INTERCEPT_CR8_MASK;
577
578 control->intercept_dr_read = INTERCEPT_DR0_MASK |
579 INTERCEPT_DR1_MASK |
580 INTERCEPT_DR2_MASK |
581 INTERCEPT_DR3_MASK |
582 INTERCEPT_DR4_MASK |
583 INTERCEPT_DR5_MASK |
584 INTERCEPT_DR6_MASK |
585 INTERCEPT_DR7_MASK;
586
587 control->intercept_dr_write = INTERCEPT_DR0_MASK |
588 INTERCEPT_DR1_MASK |
589 INTERCEPT_DR2_MASK |
590 INTERCEPT_DR3_MASK |
591 INTERCEPT_DR4_MASK |
592 INTERCEPT_DR5_MASK |
593 INTERCEPT_DR6_MASK |
594 INTERCEPT_DR7_MASK;
595
596 control->intercept_exceptions = (1 << PF_VECTOR) |
597 (1 << UD_VECTOR) |
598 (1 << MC_VECTOR);
599
600
601 control->intercept = (1ULL << INTERCEPT_INTR) |
602 (1ULL << INTERCEPT_NMI) |
603 (1ULL << INTERCEPT_SMI) |
604 (1ULL << INTERCEPT_SELECTIVE_CR0) |
605 (1ULL << INTERCEPT_CPUID) |
606 (1ULL << INTERCEPT_INVD) |
607 (1ULL << INTERCEPT_HLT) |
608 (1ULL << INTERCEPT_INVLPG) |
609 (1ULL << INTERCEPT_INVLPGA) |
610 (1ULL << INTERCEPT_IOIO_PROT) |
611 (1ULL << INTERCEPT_MSR_PROT) |
612 (1ULL << INTERCEPT_TASK_SWITCH) |
613 (1ULL << INTERCEPT_SHUTDOWN) |
614 (1ULL << INTERCEPT_VMRUN) |
615 (1ULL << INTERCEPT_VMMCALL) |
616 (1ULL << INTERCEPT_VMLOAD) |
617 (1ULL << INTERCEPT_VMSAVE) |
618 (1ULL << INTERCEPT_STGI) |
619 (1ULL << INTERCEPT_CLGI) |
620 (1ULL << INTERCEPT_SKINIT) |
621 (1ULL << INTERCEPT_WBINVD) |
622 (1ULL << INTERCEPT_MONITOR) |
623 (1ULL << INTERCEPT_MWAIT);
624
625 control->iopm_base_pa = iopm_base;
626 control->msrpm_base_pa = __pa(svm->msrpm);
627 control->tsc_offset = 0;
628 control->int_ctl = V_INTR_MASKING_MASK;
629
630 init_seg(&save->es);
631 init_seg(&save->ss);
632 init_seg(&save->ds);
633 init_seg(&save->fs);
634 init_seg(&save->gs);
635
636 save->cs.selector = 0xf000;
637 /* Executable/Readable Code Segment */
638 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
639 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
640 save->cs.limit = 0xffff;
641 /*
642 * cs.base should really be 0xffff0000, but vmx can't handle that, so
643 * be consistent with it.
644 *
645 * Replace when we have real mode working for vmx.
646 */
647 save->cs.base = 0xf0000;
648
649 save->gdtr.limit = 0xffff;
650 save->idtr.limit = 0xffff;
651
652 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
653 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
654
655 save->efer = EFER_SVME;
656 save->dr6 = 0xffff0ff0;
657 save->dr7 = 0x400;
658 save->rflags = 2;
659 save->rip = 0x0000fff0;
660 svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
661
662 /*
663 * This is the guest-visible cr0 value.
664 * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
665 */
666 svm->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
667 kvm_set_cr0(&svm->vcpu, svm->vcpu.arch.cr0);
668
669 save->cr4 = X86_CR4_PAE;
670 /* rdx = ?? */
671
672 if (npt_enabled) {
673 /* Setup VMCB for Nested Paging */
674 control->nested_ctl = 1;
675 control->intercept &= ~((1ULL << INTERCEPT_TASK_SWITCH) |
676 (1ULL << INTERCEPT_INVLPG));
677 control->intercept_exceptions &= ~(1 << PF_VECTOR);
678 control->intercept_cr_read &= ~INTERCEPT_CR3_MASK;
679 control->intercept_cr_write &= ~INTERCEPT_CR3_MASK;
680 save->g_pat = 0x0007040600070406ULL;
681 save->cr3 = 0;
682 save->cr4 = 0;
683 }
684 force_new_asid(&svm->vcpu);
685
686 svm->nested.vmcb = 0;
687 svm->vcpu.arch.hflags = 0;
688
689 if (svm_has(SVM_FEATURE_PAUSE_FILTER)) {
690 control->pause_filter_count = 3000;
691 control->intercept |= (1ULL << INTERCEPT_PAUSE);
692 }
693
694 enable_gif(svm);
695}
696
697static int svm_vcpu_reset(struct kvm_vcpu *vcpu)
698{
699 struct vcpu_svm *svm = to_svm(vcpu);
700
701 init_vmcb(svm);
702
703 if (!kvm_vcpu_is_bsp(vcpu)) {
704 kvm_rip_write(vcpu, 0);
705 svm->vmcb->save.cs.base = svm->vcpu.arch.sipi_vector << 12;
706 svm->vmcb->save.cs.selector = svm->vcpu.arch.sipi_vector << 8;
707 }
708 vcpu->arch.regs_avail = ~0;
709 vcpu->arch.regs_dirty = ~0;
710
711 return 0;
712}
713
714static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
715{
716 struct vcpu_svm *svm;
717 struct page *page;
718 struct page *msrpm_pages;
719 struct page *hsave_page;
720 struct page *nested_msrpm_pages;
721 int err;
722
723 svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
724 if (!svm) {
725 err = -ENOMEM;
726 goto out;
727 }
728
729 err = kvm_vcpu_init(&svm->vcpu, kvm, id);
730 if (err)
731 goto free_svm;
732
733 err = -ENOMEM;
734 page = alloc_page(GFP_KERNEL);
735 if (!page)
736 goto uninit;
737
738 msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
739 if (!msrpm_pages)
740 goto free_page1;
741
742 nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
743 if (!nested_msrpm_pages)
744 goto free_page2;
745
746 hsave_page = alloc_page(GFP_KERNEL);
747 if (!hsave_page)
748 goto free_page3;
749
750 svm->nested.hsave = page_address(hsave_page);
751
752 svm->msrpm = page_address(msrpm_pages);
753 svm_vcpu_init_msrpm(svm->msrpm);
754
755 svm->nested.msrpm = page_address(nested_msrpm_pages);
756
757 svm->vmcb = page_address(page);
758 clear_page(svm->vmcb);
759 svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
760 svm->asid_generation = 0;
761 init_vmcb(svm);
762
763 fx_init(&svm->vcpu);
764 svm->vcpu.arch.apic_base = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
765 if (kvm_vcpu_is_bsp(&svm->vcpu))
766 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
767
768 return &svm->vcpu;
769
770free_page3:
771 __free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER);
772free_page2:
773 __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
774free_page1:
775 __free_page(page);
776uninit:
777 kvm_vcpu_uninit(&svm->vcpu);
778free_svm:
779 kmem_cache_free(kvm_vcpu_cache, svm);
780out:
781 return ERR_PTR(err);
782}
783
784static void svm_free_vcpu(struct kvm_vcpu *vcpu)
785{
786 struct vcpu_svm *svm = to_svm(vcpu);
787
788 __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
789 __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
790 __free_page(virt_to_page(svm->nested.hsave));
791 __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER);
792 kvm_vcpu_uninit(vcpu);
793 kmem_cache_free(kvm_vcpu_cache, svm);
794}
795
796static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
797{
798 struct vcpu_svm *svm = to_svm(vcpu);
799 int i;
800
801 if (unlikely(cpu != vcpu->cpu)) {
802 u64 delta;
803
804 if (check_tsc_unstable()) {
805 /*
806 * Make sure that the guest sees a monotonically
807 * increasing TSC.
808 */
809 delta = vcpu->arch.host_tsc - native_read_tsc();
810 svm->vmcb->control.tsc_offset += delta;
811 if (is_nested(svm))
812 svm->nested.hsave->control.tsc_offset += delta;
813 }
814 vcpu->cpu = cpu;
815 kvm_migrate_timers(vcpu);
816 svm->asid_generation = 0;
817 }
818
819 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
820 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
821}
822
823static void svm_vcpu_put(struct kvm_vcpu *vcpu)
824{
825 struct vcpu_svm *svm = to_svm(vcpu);
826 int i;
827
828 ++vcpu->stat.host_state_reload;
829 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
830 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
831
832 vcpu->arch.host_tsc = native_read_tsc();
833}
834
835static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
836{
837 return to_svm(vcpu)->vmcb->save.rflags;
838}
839
840static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
841{
842 to_svm(vcpu)->vmcb->save.rflags = rflags;
843}
844
845static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
846{
847 switch (reg) {
848 case VCPU_EXREG_PDPTR:
849 BUG_ON(!npt_enabled);
850 load_pdptrs(vcpu, vcpu->arch.cr3);
851 break;
852 default:
853 BUG();
854 }
855}
856
857static void svm_set_vintr(struct vcpu_svm *svm)
858{
859 svm->vmcb->control.intercept |= 1ULL << INTERCEPT_VINTR;
860}
861
862static void svm_clear_vintr(struct vcpu_svm *svm)
863{
864 svm->vmcb->control.intercept &= ~(1ULL << INTERCEPT_VINTR);
865}
866
867static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
868{
869 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
870
871 switch (seg) {
872 case VCPU_SREG_CS: return &save->cs;
873 case VCPU_SREG_DS: return &save->ds;
874 case VCPU_SREG_ES: return &save->es;
875 case VCPU_SREG_FS: return &save->fs;
876 case VCPU_SREG_GS: return &save->gs;
877 case VCPU_SREG_SS: return &save->ss;
878 case VCPU_SREG_TR: return &save->tr;
879 case VCPU_SREG_LDTR: return &save->ldtr;
880 }
881 BUG();
882 return NULL;
883}
884
885static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
886{
887 struct vmcb_seg *s = svm_seg(vcpu, seg);
888
889 return s->base;
890}
891
892static void svm_get_segment(struct kvm_vcpu *vcpu,
893 struct kvm_segment *var, int seg)
894{
895 struct vmcb_seg *s = svm_seg(vcpu, seg);
896
897 var->base = s->base;
898 var->limit = s->limit;
899 var->selector = s->selector;
900 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
901 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
902 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
903 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
904 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
905 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
906 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
907 var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
908
909 /*
910 * AMD's VMCB does not have an explicit unusable field, so emulate it
911 * for cross vendor migration purposes by "not present"
912 */
913 var->unusable = !var->present || (var->type == 0);
914
915 switch (seg) {
916 case VCPU_SREG_CS:
917 /*
918 * SVM always stores 0 for the 'G' bit in the CS selector in
919 * the VMCB on a VMEXIT. This hurts cross-vendor migration:
920 * Intel's VMENTRY has a check on the 'G' bit.
921 */
922 var->g = s->limit > 0xfffff;
923 break;
924 case VCPU_SREG_TR:
925 /*
926 * Work around a bug where the busy flag in the tr selector
927 * isn't exposed
928 */
929 var->type |= 0x2;
930 break;
931 case VCPU_SREG_DS:
932 case VCPU_SREG_ES:
933 case VCPU_SREG_FS:
934 case VCPU_SREG_GS:
935 /*
936 * The accessed bit must always be set in the segment
937 * descriptor cache, although it can be cleared in the
938 * descriptor, the cached bit always remains at 1. Since
939 * Intel has a check on this, set it here to support
940 * cross-vendor migration.
941 */
942 if (!var->unusable)
943 var->type |= 0x1;
944 break;
945 case VCPU_SREG_SS:
946 /*
947 * On AMD CPUs sometimes the DB bit in the segment
948 * descriptor is left as 1, although the whole segment has
949 * been made unusable. Clear it here to pass an Intel VMX
950 * entry check when cross vendor migrating.
951 */
952 if (var->unusable)
953 var->db = 0;
954 break;
955 }
956}
957
958static int svm_get_cpl(struct kvm_vcpu *vcpu)
959{
960 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
961
962 return save->cpl;
963}
964
965static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
966{
967 struct vcpu_svm *svm = to_svm(vcpu);
968
969 dt->size = svm->vmcb->save.idtr.limit;
970 dt->address = svm->vmcb->save.idtr.base;
971}
972
973static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
974{
975 struct vcpu_svm *svm = to_svm(vcpu);
976
977 svm->vmcb->save.idtr.limit = dt->size;
978 svm->vmcb->save.idtr.base = dt->address ;
979}
980
981static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
982{
983 struct vcpu_svm *svm = to_svm(vcpu);
984
985 dt->size = svm->vmcb->save.gdtr.limit;
986 dt->address = svm->vmcb->save.gdtr.base;
987}
988
989static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
990{
991 struct vcpu_svm *svm = to_svm(vcpu);
992
993 svm->vmcb->save.gdtr.limit = dt->size;
994 svm->vmcb->save.gdtr.base = dt->address ;
995}
996
997static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
998{
999}
1000
1001static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1002{
1003}
1004
1005static void update_cr0_intercept(struct vcpu_svm *svm)
1006{
1007 struct vmcb *vmcb = svm->vmcb;
1008 ulong gcr0 = svm->vcpu.arch.cr0;
1009 u64 *hcr0 = &svm->vmcb->save.cr0;
1010
1011 if (!svm->vcpu.fpu_active)
1012 *hcr0 |= SVM_CR0_SELECTIVE_MASK;
1013 else
1014 *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
1015 | (gcr0 & SVM_CR0_SELECTIVE_MASK);
1016
1017
1018 if (gcr0 == *hcr0 && svm->vcpu.fpu_active) {
1019 vmcb->control.intercept_cr_read &= ~INTERCEPT_CR0_MASK;
1020 vmcb->control.intercept_cr_write &= ~INTERCEPT_CR0_MASK;
1021 if (is_nested(svm)) {
1022 struct vmcb *hsave = svm->nested.hsave;
1023
1024 hsave->control.intercept_cr_read &= ~INTERCEPT_CR0_MASK;
1025 hsave->control.intercept_cr_write &= ~INTERCEPT_CR0_MASK;
1026 vmcb->control.intercept_cr_read |= svm->nested.intercept_cr_read;
1027 vmcb->control.intercept_cr_write |= svm->nested.intercept_cr_write;
1028 }
1029 } else {
1030 svm->vmcb->control.intercept_cr_read |= INTERCEPT_CR0_MASK;
1031 svm->vmcb->control.intercept_cr_write |= INTERCEPT_CR0_MASK;
1032 if (is_nested(svm)) {
1033 struct vmcb *hsave = svm->nested.hsave;
1034
1035 hsave->control.intercept_cr_read |= INTERCEPT_CR0_MASK;
1036 hsave->control.intercept_cr_write |= INTERCEPT_CR0_MASK;
1037 }
1038 }
1039}
1040
1041static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1042{
1043 struct vcpu_svm *svm = to_svm(vcpu);
1044
1045#ifdef CONFIG_X86_64
1046 if (vcpu->arch.efer & EFER_LME) {
1047 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
1048 vcpu->arch.efer |= EFER_LMA;
1049 svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
1050 }
1051
1052 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
1053 vcpu->arch.efer &= ~EFER_LMA;
1054 svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
1055 }
1056 }
1057#endif
1058 vcpu->arch.cr0 = cr0;
1059
1060 if (!npt_enabled)
1061 cr0 |= X86_CR0_PG | X86_CR0_WP;
1062
1063 if (!vcpu->fpu_active)
1064 cr0 |= X86_CR0_TS;
1065 /*
1066 * re-enable caching here because the QEMU bios
1067 * does not do it - this results in some delay at
1068 * reboot
1069 */
1070 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
1071 svm->vmcb->save.cr0 = cr0;
1072 update_cr0_intercept(svm);
1073}
1074
1075static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1076{
1077 unsigned long host_cr4_mce = read_cr4() & X86_CR4_MCE;
1078 unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
1079
1080 if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
1081 force_new_asid(vcpu);
1082
1083 vcpu->arch.cr4 = cr4;
1084 if (!npt_enabled)
1085 cr4 |= X86_CR4_PAE;
1086 cr4 |= host_cr4_mce;
1087 to_svm(vcpu)->vmcb->save.cr4 = cr4;
1088}
1089
1090static void svm_set_segment(struct kvm_vcpu *vcpu,
1091 struct kvm_segment *var, int seg)
1092{
1093 struct vcpu_svm *svm = to_svm(vcpu);
1094 struct vmcb_seg *s = svm_seg(vcpu, seg);
1095
1096 s->base = var->base;
1097 s->limit = var->limit;
1098 s->selector = var->selector;
1099 if (var->unusable)
1100 s->attrib = 0;
1101 else {
1102 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
1103 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
1104 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
1105 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
1106 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
1107 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
1108 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
1109 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
1110 }
1111 if (seg == VCPU_SREG_CS)
1112 svm->vmcb->save.cpl
1113 = (svm->vmcb->save.cs.attrib
1114 >> SVM_SELECTOR_DPL_SHIFT) & 3;
1115
1116}
1117
1118static void update_db_intercept(struct kvm_vcpu *vcpu)
1119{
1120 struct vcpu_svm *svm = to_svm(vcpu);
1121
1122 svm->vmcb->control.intercept_exceptions &=
1123 ~((1 << DB_VECTOR) | (1 << BP_VECTOR));
1124
1125 if (svm->nmi_singlestep)
1126 svm->vmcb->control.intercept_exceptions |= (1 << DB_VECTOR);
1127
1128 if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
1129 if (vcpu->guest_debug &
1130 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
1131 svm->vmcb->control.intercept_exceptions |=
1132 1 << DB_VECTOR;
1133 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
1134 svm->vmcb->control.intercept_exceptions |=
1135 1 << BP_VECTOR;
1136 } else
1137 vcpu->guest_debug = 0;
1138}
1139
1140static void svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
1141{
1142 struct vcpu_svm *svm = to_svm(vcpu);
1143
1144 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1145 svm->vmcb->save.dr7 = dbg->arch.debugreg[7];
1146 else
1147 svm->vmcb->save.dr7 = vcpu->arch.dr7;
1148
1149 update_db_intercept(vcpu);
1150}
1151
1152static void load_host_msrs(struct kvm_vcpu *vcpu)
1153{
1154#ifdef CONFIG_X86_64
1155 wrmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
1156#endif
1157}
1158
1159static void save_host_msrs(struct kvm_vcpu *vcpu)
1160{
1161#ifdef CONFIG_X86_64
1162 rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host_gs_base);
1163#endif
1164}
1165
1166static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
1167{
1168 if (sd->next_asid > sd->max_asid) {
1169 ++sd->asid_generation;
1170 sd->next_asid = 1;
1171 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
1172 }
1173
1174 svm->asid_generation = sd->asid_generation;
1175 svm->vmcb->control.asid = sd->next_asid++;
1176}
1177
1178static int svm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *dest)
1179{
1180 struct vcpu_svm *svm = to_svm(vcpu);
1181
1182 switch (dr) {
1183 case 0 ... 3:
1184 *dest = vcpu->arch.db[dr];
1185 break;
1186 case 4:
1187 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
1188 return EMULATE_FAIL; /* will re-inject UD */
1189 /* fall through */
1190 case 6:
1191 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1192 *dest = vcpu->arch.dr6;
1193 else
1194 *dest = svm->vmcb->save.dr6;
1195 break;
1196 case 5:
1197 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
1198 return EMULATE_FAIL; /* will re-inject UD */
1199 /* fall through */
1200 case 7:
1201 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1202 *dest = vcpu->arch.dr7;
1203 else
1204 *dest = svm->vmcb->save.dr7;
1205 break;
1206 }
1207
1208 return EMULATE_DONE;
1209}
1210
1211static int svm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long value)
1212{
1213 struct vcpu_svm *svm = to_svm(vcpu);
1214
1215 switch (dr) {
1216 case 0 ... 3:
1217 vcpu->arch.db[dr] = value;
1218 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1219 vcpu->arch.eff_db[dr] = value;
1220 break;
1221 case 4:
1222 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
1223 return EMULATE_FAIL; /* will re-inject UD */
1224 /* fall through */
1225 case 6:
1226 vcpu->arch.dr6 = (value & DR6_VOLATILE) | DR6_FIXED_1;
1227 break;
1228 case 5:
1229 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE))
1230 return EMULATE_FAIL; /* will re-inject UD */
1231 /* fall through */
1232 case 7:
1233 vcpu->arch.dr7 = (value & DR7_VOLATILE) | DR7_FIXED_1;
1234 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1235 svm->vmcb->save.dr7 = vcpu->arch.dr7;
1236 vcpu->arch.switch_db_regs = (value & DR7_BP_EN_MASK);
1237 }
1238 break;
1239 }
1240
1241 return EMULATE_DONE;
1242}
1243
1244static int pf_interception(struct vcpu_svm *svm)
1245{
1246 u64 fault_address;
1247 u32 error_code;
1248
1249 fault_address = svm->vmcb->control.exit_info_2;
1250 error_code = svm->vmcb->control.exit_info_1;
1251
1252 trace_kvm_page_fault(fault_address, error_code);
1253 if (!npt_enabled && kvm_event_needs_reinjection(&svm->vcpu))
1254 kvm_mmu_unprotect_page_virt(&svm->vcpu, fault_address);
1255 return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code);
1256}
1257
1258static int db_interception(struct vcpu_svm *svm)
1259{
1260 struct kvm_run *kvm_run = svm->vcpu.run;
1261
1262 if (!(svm->vcpu.guest_debug &
1263 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
1264 !svm->nmi_singlestep) {
1265 kvm_queue_exception(&svm->vcpu, DB_VECTOR);
1266 return 1;
1267 }
1268
1269 if (svm->nmi_singlestep) {
1270 svm->nmi_singlestep = false;
1271 if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP))
1272 svm->vmcb->save.rflags &=
1273 ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1274 update_db_intercept(&svm->vcpu);
1275 }
1276
1277 if (svm->vcpu.guest_debug &
1278 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
1279 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1280 kvm_run->debug.arch.pc =
1281 svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1282 kvm_run->debug.arch.exception = DB_VECTOR;
1283 return 0;
1284 }
1285
1286 return 1;
1287}
1288
1289static int bp_interception(struct vcpu_svm *svm)
1290{
1291 struct kvm_run *kvm_run = svm->vcpu.run;
1292
1293 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1294 kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1295 kvm_run->debug.arch.exception = BP_VECTOR;
1296 return 0;
1297}
1298
1299static int ud_interception(struct vcpu_svm *svm)
1300{
1301 int er;
1302
1303 er = emulate_instruction(&svm->vcpu, 0, 0, EMULTYPE_TRAP_UD);
1304 if (er != EMULATE_DONE)
1305 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1306 return 1;
1307}
1308
1309static void svm_fpu_activate(struct kvm_vcpu *vcpu)
1310{
1311 struct vcpu_svm *svm = to_svm(vcpu);
1312 u32 excp;
1313
1314 if (is_nested(svm)) {
1315 u32 h_excp, n_excp;
1316
1317 h_excp = svm->nested.hsave->control.intercept_exceptions;
1318 n_excp = svm->nested.intercept_exceptions;
1319 h_excp &= ~(1 << NM_VECTOR);
1320 excp = h_excp | n_excp;
1321 } else {
1322 excp = svm->vmcb->control.intercept_exceptions;
1323 excp &= ~(1 << NM_VECTOR);
1324 }
1325
1326 svm->vmcb->control.intercept_exceptions = excp;
1327
1328 svm->vcpu.fpu_active = 1;
1329 update_cr0_intercept(svm);
1330}
1331
1332static int nm_interception(struct vcpu_svm *svm)
1333{
1334 svm_fpu_activate(&svm->vcpu);
1335 return 1;
1336}
1337
1338static int mc_interception(struct vcpu_svm *svm)
1339{
1340 /*
1341 * On an #MC intercept the MCE handler is not called automatically in
1342 * the host. So do it by hand here.
1343 */
1344 asm volatile (
1345 "int $0x12\n");
1346 /* not sure if we ever come back to this point */
1347
1348 return 1;
1349}
1350
1351static int shutdown_interception(struct vcpu_svm *svm)
1352{
1353 struct kvm_run *kvm_run = svm->vcpu.run;
1354
1355 /*
1356 * VMCB is undefined after a SHUTDOWN intercept
1357 * so reinitialize it.
1358 */
1359 clear_page(svm->vmcb);
1360 init_vmcb(svm);
1361
1362 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1363 return 0;
1364}
1365
1366static int io_interception(struct vcpu_svm *svm)
1367{
1368 u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
1369 int size, in, string;
1370 unsigned port;
1371
1372 ++svm->vcpu.stat.io_exits;
1373
1374 svm->next_rip = svm->vmcb->control.exit_info_2;
1375
1376 string = (io_info & SVM_IOIO_STR_MASK) != 0;
1377
1378 if (string) {
1379 if (emulate_instruction(&svm->vcpu,
1380 0, 0, 0) == EMULATE_DO_MMIO)
1381 return 0;
1382 return 1;
1383 }
1384
1385 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1386 port = io_info >> 16;
1387 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1388
1389 skip_emulated_instruction(&svm->vcpu);
1390 return kvm_emulate_pio(&svm->vcpu, in, size, port);
1391}
1392
1393static int nmi_interception(struct vcpu_svm *svm)
1394{
1395 return 1;
1396}
1397
1398static int intr_interception(struct vcpu_svm *svm)
1399{
1400 ++svm->vcpu.stat.irq_exits;
1401 return 1;
1402}
1403
1404static int nop_on_interception(struct vcpu_svm *svm)
1405{
1406 return 1;
1407}
1408
1409static int halt_interception(struct vcpu_svm *svm)
1410{
1411 svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
1412 skip_emulated_instruction(&svm->vcpu);
1413 return kvm_emulate_halt(&svm->vcpu);
1414}
1415
1416static int vmmcall_interception(struct vcpu_svm *svm)
1417{
1418 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1419 skip_emulated_instruction(&svm->vcpu);
1420 kvm_emulate_hypercall(&svm->vcpu);
1421 return 1;
1422}
1423
1424static int nested_svm_check_permissions(struct vcpu_svm *svm)
1425{
1426 if (!(svm->vcpu.arch.efer & EFER_SVME)
1427 || !is_paging(&svm->vcpu)) {
1428 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1429 return 1;
1430 }
1431
1432 if (svm->vmcb->save.cpl) {
1433 kvm_inject_gp(&svm->vcpu, 0);
1434 return 1;
1435 }
1436
1437 return 0;
1438}
1439
1440static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
1441 bool has_error_code, u32 error_code)
1442{
1443 int vmexit;
1444
1445 if (!is_nested(svm))
1446 return 0;
1447
1448 svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
1449 svm->vmcb->control.exit_code_hi = 0;
1450 svm->vmcb->control.exit_info_1 = error_code;
1451 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
1452
1453 vmexit = nested_svm_intercept(svm);
1454 if (vmexit == NESTED_EXIT_DONE)
1455 svm->nested.exit_required = true;
1456
1457 return vmexit;
1458}
1459
1460/* This function returns true if it is save to enable the irq window */
1461static inline bool nested_svm_intr(struct vcpu_svm *svm)
1462{
1463 if (!is_nested(svm))
1464 return true;
1465
1466 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
1467 return true;
1468
1469 if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
1470 return false;
1471
1472 svm->vmcb->control.exit_code = SVM_EXIT_INTR;
1473
1474 if (svm->nested.intercept & 1ULL) {
1475 /*
1476 * The #vmexit can't be emulated here directly because this
1477 * code path runs with irqs and preemtion disabled. A
1478 * #vmexit emulation might sleep. Only signal request for
1479 * the #vmexit here.
1480 */
1481 svm->nested.exit_required = true;
1482 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
1483 return false;
1484 }
1485
1486 return true;
1487}
1488
1489static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, struct page **_page)
1490{
1491 struct page *page;
1492
1493 might_sleep();
1494
1495 page = gfn_to_page(svm->vcpu.kvm, gpa >> PAGE_SHIFT);
1496 if (is_error_page(page))
1497 goto error;
1498
1499 *_page = page;
1500
1501 return kmap(page);
1502
1503error:
1504 kvm_release_page_clean(page);
1505 kvm_inject_gp(&svm->vcpu, 0);
1506
1507 return NULL;
1508}
1509
1510static void nested_svm_unmap(struct page *page)
1511{
1512 kunmap(page);
1513 kvm_release_page_dirty(page);
1514}
1515
1516static bool nested_svm_exit_handled_msr(struct vcpu_svm *svm)
1517{
1518 u32 param = svm->vmcb->control.exit_info_1 & 1;
1519 u32 msr = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1520 bool ret = false;
1521 u32 t0, t1;
1522 u8 val;
1523
1524 if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
1525 return false;
1526
1527 switch (msr) {
1528 case 0 ... 0x1fff:
1529 t0 = (msr * 2) % 8;
1530 t1 = msr / 8;
1531 break;
1532 case 0xc0000000 ... 0xc0001fff:
1533 t0 = (8192 + msr - 0xc0000000) * 2;
1534 t1 = (t0 / 8);
1535 t0 %= 8;
1536 break;
1537 case 0xc0010000 ... 0xc0011fff:
1538 t0 = (16384 + msr - 0xc0010000) * 2;
1539 t1 = (t0 / 8);
1540 t0 %= 8;
1541 break;
1542 default:
1543 ret = true;
1544 goto out;
1545 }
1546
1547 if (!kvm_read_guest(svm->vcpu.kvm, svm->nested.vmcb_msrpm + t1, &val, 1))
1548 ret = val & ((1 << param) << t0);
1549
1550out:
1551 return ret;
1552}
1553
1554static int nested_svm_exit_special(struct vcpu_svm *svm)
1555{
1556 u32 exit_code = svm->vmcb->control.exit_code;
1557
1558 switch (exit_code) {
1559 case SVM_EXIT_INTR:
1560 case SVM_EXIT_NMI:
1561 return NESTED_EXIT_HOST;
1562 case SVM_EXIT_NPF:
1563 /* For now we are always handling NPFs when using them */
1564 if (npt_enabled)
1565 return NESTED_EXIT_HOST;
1566 break;
1567 case SVM_EXIT_EXCP_BASE + PF_VECTOR:
1568 /* When we're shadowing, trap PFs */
1569 if (!npt_enabled)
1570 return NESTED_EXIT_HOST;
1571 break;
1572 case SVM_EXIT_EXCP_BASE + NM_VECTOR:
1573 nm_interception(svm);
1574 break;
1575 default:
1576 break;
1577 }
1578
1579 return NESTED_EXIT_CONTINUE;
1580}
1581
1582/*
1583 * If this function returns true, this #vmexit was already handled
1584 */
1585static int nested_svm_intercept(struct vcpu_svm *svm)
1586{
1587 u32 exit_code = svm->vmcb->control.exit_code;
1588 int vmexit = NESTED_EXIT_HOST;
1589
1590 switch (exit_code) {
1591 case SVM_EXIT_MSR:
1592 vmexit = nested_svm_exit_handled_msr(svm);
1593 break;
1594 case SVM_EXIT_READ_CR0 ... SVM_EXIT_READ_CR8: {
1595 u32 cr_bits = 1 << (exit_code - SVM_EXIT_READ_CR0);
1596 if (svm->nested.intercept_cr_read & cr_bits)
1597 vmexit = NESTED_EXIT_DONE;
1598 break;
1599 }
1600 case SVM_EXIT_WRITE_CR0 ... SVM_EXIT_WRITE_CR8: {
1601 u32 cr_bits = 1 << (exit_code - SVM_EXIT_WRITE_CR0);
1602 if (svm->nested.intercept_cr_write & cr_bits)
1603 vmexit = NESTED_EXIT_DONE;
1604 break;
1605 }
1606 case SVM_EXIT_READ_DR0 ... SVM_EXIT_READ_DR7: {
1607 u32 dr_bits = 1 << (exit_code - SVM_EXIT_READ_DR0);
1608 if (svm->nested.intercept_dr_read & dr_bits)
1609 vmexit = NESTED_EXIT_DONE;
1610 break;
1611 }
1612 case SVM_EXIT_WRITE_DR0 ... SVM_EXIT_WRITE_DR7: {
1613 u32 dr_bits = 1 << (exit_code - SVM_EXIT_WRITE_DR0);
1614 if (svm->nested.intercept_dr_write & dr_bits)
1615 vmexit = NESTED_EXIT_DONE;
1616 break;
1617 }
1618 case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1619 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
1620 if (svm->nested.intercept_exceptions & excp_bits)
1621 vmexit = NESTED_EXIT_DONE;
1622 break;
1623 }
1624 default: {
1625 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
1626 if (svm->nested.intercept & exit_bits)
1627 vmexit = NESTED_EXIT_DONE;
1628 }
1629 }
1630
1631 return vmexit;
1632}
1633
1634static int nested_svm_exit_handled(struct vcpu_svm *svm)
1635{
1636 int vmexit;
1637
1638 vmexit = nested_svm_intercept(svm);
1639
1640 if (vmexit == NESTED_EXIT_DONE)
1641 nested_svm_vmexit(svm);
1642
1643 return vmexit;
1644}
1645
1646static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
1647{
1648 struct vmcb_control_area *dst = &dst_vmcb->control;
1649 struct vmcb_control_area *from = &from_vmcb->control;
1650
1651 dst->intercept_cr_read = from->intercept_cr_read;
1652 dst->intercept_cr_write = from->intercept_cr_write;
1653 dst->intercept_dr_read = from->intercept_dr_read;
1654 dst->intercept_dr_write = from->intercept_dr_write;
1655 dst->intercept_exceptions = from->intercept_exceptions;
1656 dst->intercept = from->intercept;
1657 dst->iopm_base_pa = from->iopm_base_pa;
1658 dst->msrpm_base_pa = from->msrpm_base_pa;
1659 dst->tsc_offset = from->tsc_offset;
1660 dst->asid = from->asid;
1661 dst->tlb_ctl = from->tlb_ctl;
1662 dst->int_ctl = from->int_ctl;
1663 dst->int_vector = from->int_vector;
1664 dst->int_state = from->int_state;
1665 dst->exit_code = from->exit_code;
1666 dst->exit_code_hi = from->exit_code_hi;
1667 dst->exit_info_1 = from->exit_info_1;
1668 dst->exit_info_2 = from->exit_info_2;
1669 dst->exit_int_info = from->exit_int_info;
1670 dst->exit_int_info_err = from->exit_int_info_err;
1671 dst->nested_ctl = from->nested_ctl;
1672 dst->event_inj = from->event_inj;
1673 dst->event_inj_err = from->event_inj_err;
1674 dst->nested_cr3 = from->nested_cr3;
1675 dst->lbr_ctl = from->lbr_ctl;
1676}
1677
1678static int nested_svm_vmexit(struct vcpu_svm *svm)
1679{
1680 struct vmcb *nested_vmcb;
1681 struct vmcb *hsave = svm->nested.hsave;
1682 struct vmcb *vmcb = svm->vmcb;
1683 struct page *page;
1684
1685 trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
1686 vmcb->control.exit_info_1,
1687 vmcb->control.exit_info_2,
1688 vmcb->control.exit_int_info,
1689 vmcb->control.exit_int_info_err);
1690
1691 nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, &page);
1692 if (!nested_vmcb)
1693 return 1;
1694
1695 /* Exit nested SVM mode */
1696 svm->nested.vmcb = 0;
1697
1698 /* Give the current vmcb to the guest */
1699 disable_gif(svm);
1700
1701 nested_vmcb->save.es = vmcb->save.es;
1702 nested_vmcb->save.cs = vmcb->save.cs;
1703 nested_vmcb->save.ss = vmcb->save.ss;
1704 nested_vmcb->save.ds = vmcb->save.ds;
1705 nested_vmcb->save.gdtr = vmcb->save.gdtr;
1706 nested_vmcb->save.idtr = vmcb->save.idtr;
1707 nested_vmcb->save.cr0 = kvm_read_cr0(&svm->vcpu);
1708 if (npt_enabled)
1709 nested_vmcb->save.cr3 = vmcb->save.cr3;
1710 else
1711 nested_vmcb->save.cr3 = svm->vcpu.arch.cr3;
1712 nested_vmcb->save.cr2 = vmcb->save.cr2;
1713 nested_vmcb->save.cr4 = svm->vcpu.arch.cr4;
1714 nested_vmcb->save.rflags = vmcb->save.rflags;
1715 nested_vmcb->save.rip = vmcb->save.rip;
1716 nested_vmcb->save.rsp = vmcb->save.rsp;
1717 nested_vmcb->save.rax = vmcb->save.rax;
1718 nested_vmcb->save.dr7 = vmcb->save.dr7;
1719 nested_vmcb->save.dr6 = vmcb->save.dr6;
1720 nested_vmcb->save.cpl = vmcb->save.cpl;
1721
1722 nested_vmcb->control.int_ctl = vmcb->control.int_ctl;
1723 nested_vmcb->control.int_vector = vmcb->control.int_vector;
1724 nested_vmcb->control.int_state = vmcb->control.int_state;
1725 nested_vmcb->control.exit_code = vmcb->control.exit_code;
1726 nested_vmcb->control.exit_code_hi = vmcb->control.exit_code_hi;
1727 nested_vmcb->control.exit_info_1 = vmcb->control.exit_info_1;
1728 nested_vmcb->control.exit_info_2 = vmcb->control.exit_info_2;
1729 nested_vmcb->control.exit_int_info = vmcb->control.exit_int_info;
1730 nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
1731
1732 /*
1733 * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
1734 * to make sure that we do not lose injected events. So check event_inj
1735 * here and copy it to exit_int_info if it is valid.
1736 * Exit_int_info and event_inj can't be both valid because the case
1737 * below only happens on a VMRUN instruction intercept which has
1738 * no valid exit_int_info set.
1739 */
1740 if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
1741 struct vmcb_control_area *nc = &nested_vmcb->control;
1742
1743 nc->exit_int_info = vmcb->control.event_inj;
1744 nc->exit_int_info_err = vmcb->control.event_inj_err;
1745 }
1746
1747 nested_vmcb->control.tlb_ctl = 0;
1748 nested_vmcb->control.event_inj = 0;
1749 nested_vmcb->control.event_inj_err = 0;
1750
1751 /* We always set V_INTR_MASKING and remember the old value in hflags */
1752 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
1753 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
1754
1755 /* Restore the original control entries */
1756 copy_vmcb_control_area(vmcb, hsave);
1757
1758 kvm_clear_exception_queue(&svm->vcpu);
1759 kvm_clear_interrupt_queue(&svm->vcpu);
1760
1761 /* Restore selected save entries */
1762 svm->vmcb->save.es = hsave->save.es;
1763 svm->vmcb->save.cs = hsave->save.cs;
1764 svm->vmcb->save.ss = hsave->save.ss;
1765 svm->vmcb->save.ds = hsave->save.ds;
1766 svm->vmcb->save.gdtr = hsave->save.gdtr;
1767 svm->vmcb->save.idtr = hsave->save.idtr;
1768 svm->vmcb->save.rflags = hsave->save.rflags;
1769 svm_set_efer(&svm->vcpu, hsave->save.efer);
1770 svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
1771 svm_set_cr4(&svm->vcpu, hsave->save.cr4);
1772 if (npt_enabled) {
1773 svm->vmcb->save.cr3 = hsave->save.cr3;
1774 svm->vcpu.arch.cr3 = hsave->save.cr3;
1775 } else {
1776 kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
1777 }
1778 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
1779 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
1780 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
1781 svm->vmcb->save.dr7 = 0;
1782 svm->vmcb->save.cpl = 0;
1783 svm->vmcb->control.exit_int_info = 0;
1784
1785 nested_svm_unmap(page);
1786
1787 kvm_mmu_reset_context(&svm->vcpu);
1788 kvm_mmu_load(&svm->vcpu);
1789
1790 return 0;
1791}
1792
1793static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
1794{
1795 u32 *nested_msrpm;
1796 struct page *page;
1797 int i;
1798
1799 nested_msrpm = nested_svm_map(svm, svm->nested.vmcb_msrpm, &page);
1800 if (!nested_msrpm)
1801 return false;
1802
1803 for (i = 0; i < PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER) / 4; i++)
1804 svm->nested.msrpm[i] = svm->msrpm[i] | nested_msrpm[i];
1805
1806 svm->vmcb->control.msrpm_base_pa = __pa(svm->nested.msrpm);
1807
1808 nested_svm_unmap(page);
1809
1810 return true;
1811}
1812
1813static bool nested_svm_vmrun(struct vcpu_svm *svm)
1814{
1815 struct vmcb *nested_vmcb;
1816 struct vmcb *hsave = svm->nested.hsave;
1817 struct vmcb *vmcb = svm->vmcb;
1818 struct page *page;
1819 u64 vmcb_gpa;
1820
1821 vmcb_gpa = svm->vmcb->save.rax;
1822
1823 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
1824 if (!nested_vmcb)
1825 return false;
1826
1827 trace_kvm_nested_vmrun(svm->vmcb->save.rip - 3, svm->nested.vmcb,
1828 nested_vmcb->save.rip,
1829 nested_vmcb->control.int_ctl,
1830 nested_vmcb->control.event_inj,
1831 nested_vmcb->control.nested_ctl);
1832
1833 /* Clear internal status */
1834 kvm_clear_exception_queue(&svm->vcpu);
1835 kvm_clear_interrupt_queue(&svm->vcpu);
1836
1837 /*
1838 * Save the old vmcb, so we don't need to pick what we save, but can
1839 * restore everything when a VMEXIT occurs
1840 */
1841 hsave->save.es = vmcb->save.es;
1842 hsave->save.cs = vmcb->save.cs;
1843 hsave->save.ss = vmcb->save.ss;
1844 hsave->save.ds = vmcb->save.ds;
1845 hsave->save.gdtr = vmcb->save.gdtr;
1846 hsave->save.idtr = vmcb->save.idtr;
1847 hsave->save.efer = svm->vcpu.arch.efer;
1848 hsave->save.cr0 = kvm_read_cr0(&svm->vcpu);
1849 hsave->save.cr4 = svm->vcpu.arch.cr4;
1850 hsave->save.rflags = vmcb->save.rflags;
1851 hsave->save.rip = svm->next_rip;
1852 hsave->save.rsp = vmcb->save.rsp;
1853 hsave->save.rax = vmcb->save.rax;
1854 if (npt_enabled)
1855 hsave->save.cr3 = vmcb->save.cr3;
1856 else
1857 hsave->save.cr3 = svm->vcpu.arch.cr3;
1858
1859 copy_vmcb_control_area(hsave, vmcb);
1860
1861 if (svm->vmcb->save.rflags & X86_EFLAGS_IF)
1862 svm->vcpu.arch.hflags |= HF_HIF_MASK;
1863 else
1864 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
1865
1866 /* Load the nested guest state */
1867 svm->vmcb->save.es = nested_vmcb->save.es;
1868 svm->vmcb->save.cs = nested_vmcb->save.cs;
1869 svm->vmcb->save.ss = nested_vmcb->save.ss;
1870 svm->vmcb->save.ds = nested_vmcb->save.ds;
1871 svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
1872 svm->vmcb->save.idtr = nested_vmcb->save.idtr;
1873 svm->vmcb->save.rflags = nested_vmcb->save.rflags;
1874 svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
1875 svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
1876 svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
1877 if (npt_enabled) {
1878 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
1879 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
1880 } else {
1881 kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
1882 kvm_mmu_reset_context(&svm->vcpu);
1883 }
1884 svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
1885 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
1886 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
1887 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
1888
1889 /* In case we don't even reach vcpu_run, the fields are not updated */
1890 svm->vmcb->save.rax = nested_vmcb->save.rax;
1891 svm->vmcb->save.rsp = nested_vmcb->save.rsp;
1892 svm->vmcb->save.rip = nested_vmcb->save.rip;
1893 svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
1894 svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
1895 svm->vmcb->save.cpl = nested_vmcb->save.cpl;
1896
1897 svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa;
1898
1899 /* cache intercepts */
1900 svm->nested.intercept_cr_read = nested_vmcb->control.intercept_cr_read;
1901 svm->nested.intercept_cr_write = nested_vmcb->control.intercept_cr_write;
1902 svm->nested.intercept_dr_read = nested_vmcb->control.intercept_dr_read;
1903 svm->nested.intercept_dr_write = nested_vmcb->control.intercept_dr_write;
1904 svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
1905 svm->nested.intercept = nested_vmcb->control.intercept;
1906
1907 force_new_asid(&svm->vcpu);
1908 svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
1909 if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
1910 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
1911 else
1912 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
1913
1914 if (svm->vcpu.arch.hflags & HF_VINTR_MASK) {
1915 /* We only want the cr8 intercept bits of the guest */
1916 svm->vmcb->control.intercept_cr_read &= ~INTERCEPT_CR8_MASK;
1917 svm->vmcb->control.intercept_cr_write &= ~INTERCEPT_CR8_MASK;
1918 }
1919
1920 /*
1921 * We don't want a nested guest to be more powerful than the guest, so
1922 * all intercepts are ORed
1923 */
1924 svm->vmcb->control.intercept_cr_read |=
1925 nested_vmcb->control.intercept_cr_read;
1926 svm->vmcb->control.intercept_cr_write |=
1927 nested_vmcb->control.intercept_cr_write;
1928 svm->vmcb->control.intercept_dr_read |=
1929 nested_vmcb->control.intercept_dr_read;
1930 svm->vmcb->control.intercept_dr_write |=
1931 nested_vmcb->control.intercept_dr_write;
1932 svm->vmcb->control.intercept_exceptions |=
1933 nested_vmcb->control.intercept_exceptions;
1934
1935 svm->vmcb->control.intercept |= nested_vmcb->control.intercept;
1936
1937 svm->vmcb->control.lbr_ctl = nested_vmcb->control.lbr_ctl;
1938 svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
1939 svm->vmcb->control.int_state = nested_vmcb->control.int_state;
1940 svm->vmcb->control.tsc_offset += nested_vmcb->control.tsc_offset;
1941 svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
1942 svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
1943
1944 nested_svm_unmap(page);
1945
1946 /* nested_vmcb is our indicator if nested SVM is activated */
1947 svm->nested.vmcb = vmcb_gpa;
1948
1949 enable_gif(svm);
1950
1951 return true;
1952}
1953
1954static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
1955{
1956 to_vmcb->save.fs = from_vmcb->save.fs;
1957 to_vmcb->save.gs = from_vmcb->save.gs;
1958 to_vmcb->save.tr = from_vmcb->save.tr;
1959 to_vmcb->save.ldtr = from_vmcb->save.ldtr;
1960 to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
1961 to_vmcb->save.star = from_vmcb->save.star;
1962 to_vmcb->save.lstar = from_vmcb->save.lstar;
1963 to_vmcb->save.cstar = from_vmcb->save.cstar;
1964 to_vmcb->save.sfmask = from_vmcb->save.sfmask;
1965 to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
1966 to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
1967 to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
1968}
1969
1970static int vmload_interception(struct vcpu_svm *svm)
1971{
1972 struct vmcb *nested_vmcb;
1973 struct page *page;
1974
1975 if (nested_svm_check_permissions(svm))
1976 return 1;
1977
1978 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1979 skip_emulated_instruction(&svm->vcpu);
1980
1981 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
1982 if (!nested_vmcb)
1983 return 1;
1984
1985 nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
1986 nested_svm_unmap(page);
1987
1988 return 1;
1989}
1990
1991static int vmsave_interception(struct vcpu_svm *svm)
1992{
1993 struct vmcb *nested_vmcb;
1994 struct page *page;
1995
1996 if (nested_svm_check_permissions(svm))
1997 return 1;
1998
1999 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2000 skip_emulated_instruction(&svm->vcpu);
2001
2002 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2003 if (!nested_vmcb)
2004 return 1;
2005
2006 nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
2007 nested_svm_unmap(page);
2008
2009 return 1;
2010}
2011
2012static int vmrun_interception(struct vcpu_svm *svm)
2013{
2014 if (nested_svm_check_permissions(svm))
2015 return 1;
2016
2017 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2018 skip_emulated_instruction(&svm->vcpu);
2019
2020 if (!nested_svm_vmrun(svm))
2021 return 1;
2022
2023 if (!nested_svm_vmrun_msrpm(svm))
2024 goto failed;
2025
2026 return 1;
2027
2028failed:
2029
2030 svm->vmcb->control.exit_code = SVM_EXIT_ERR;
2031 svm->vmcb->control.exit_code_hi = 0;
2032 svm->vmcb->control.exit_info_1 = 0;
2033 svm->vmcb->control.exit_info_2 = 0;
2034
2035 nested_svm_vmexit(svm);
2036
2037 return 1;
2038}
2039
2040static int stgi_interception(struct vcpu_svm *svm)
2041{
2042 if (nested_svm_check_permissions(svm))
2043 return 1;
2044
2045 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2046 skip_emulated_instruction(&svm->vcpu);
2047
2048 enable_gif(svm);
2049
2050 return 1;
2051}
2052
2053static int clgi_interception(struct vcpu_svm *svm)
2054{
2055 if (nested_svm_check_permissions(svm))
2056 return 1;
2057
2058 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2059 skip_emulated_instruction(&svm->vcpu);
2060
2061 disable_gif(svm);
2062
2063 /* After a CLGI no interrupts should come */
2064 svm_clear_vintr(svm);
2065 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
2066
2067 return 1;
2068}
2069
2070static int invlpga_interception(struct vcpu_svm *svm)
2071{
2072 struct kvm_vcpu *vcpu = &svm->vcpu;
2073
2074 trace_kvm_invlpga(svm->vmcb->save.rip, vcpu->arch.regs[VCPU_REGS_RCX],
2075 vcpu->arch.regs[VCPU_REGS_RAX]);
2076
2077 /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
2078 kvm_mmu_invlpg(vcpu, vcpu->arch.regs[VCPU_REGS_RAX]);
2079
2080 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2081 skip_emulated_instruction(&svm->vcpu);
2082 return 1;
2083}
2084
2085static int skinit_interception(struct vcpu_svm *svm)
2086{
2087 trace_kvm_skinit(svm->vmcb->save.rip, svm->vcpu.arch.regs[VCPU_REGS_RAX]);
2088
2089 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2090 return 1;
2091}
2092
2093static int invalid_op_interception(struct vcpu_svm *svm)
2094{
2095 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2096 return 1;
2097}
2098
2099static int task_switch_interception(struct vcpu_svm *svm)
2100{
2101 u16 tss_selector;
2102 int reason;
2103 int int_type = svm->vmcb->control.exit_int_info &
2104 SVM_EXITINTINFO_TYPE_MASK;
2105 int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
2106 uint32_t type =
2107 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
2108 uint32_t idt_v =
2109 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
2110
2111 tss_selector = (u16)svm->vmcb->control.exit_info_1;
2112
2113 if (svm->vmcb->control.exit_info_2 &
2114 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
2115 reason = TASK_SWITCH_IRET;
2116 else if (svm->vmcb->control.exit_info_2 &
2117 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
2118 reason = TASK_SWITCH_JMP;
2119 else if (idt_v)
2120 reason = TASK_SWITCH_GATE;
2121 else
2122 reason = TASK_SWITCH_CALL;
2123
2124 if (reason == TASK_SWITCH_GATE) {
2125 switch (type) {
2126 case SVM_EXITINTINFO_TYPE_NMI:
2127 svm->vcpu.arch.nmi_injected = false;
2128 break;
2129 case SVM_EXITINTINFO_TYPE_EXEPT:
2130 kvm_clear_exception_queue(&svm->vcpu);
2131 break;
2132 case SVM_EXITINTINFO_TYPE_INTR:
2133 kvm_clear_interrupt_queue(&svm->vcpu);
2134 break;
2135 default:
2136 break;
2137 }
2138 }
2139
2140 if (reason != TASK_SWITCH_GATE ||
2141 int_type == SVM_EXITINTINFO_TYPE_SOFT ||
2142 (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
2143 (int_vec == OF_VECTOR || int_vec == BP_VECTOR)))
2144 skip_emulated_instruction(&svm->vcpu);
2145
2146 return kvm_task_switch(&svm->vcpu, tss_selector, reason);
2147}
2148
2149static int cpuid_interception(struct vcpu_svm *svm)
2150{
2151 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2152 kvm_emulate_cpuid(&svm->vcpu);
2153 return 1;
2154}
2155
2156static int iret_interception(struct vcpu_svm *svm)
2157{
2158 ++svm->vcpu.stat.nmi_window_exits;
2159 svm->vmcb->control.intercept &= ~(1UL << INTERCEPT_IRET);
2160 svm->vcpu.arch.hflags |= HF_IRET_MASK;
2161 return 1;
2162}
2163
2164static int invlpg_interception(struct vcpu_svm *svm)
2165{
2166 if (emulate_instruction(&svm->vcpu, 0, 0, 0) != EMULATE_DONE)
2167 pr_unimpl(&svm->vcpu, "%s: failed\n", __func__);
2168 return 1;
2169}
2170
2171static int emulate_on_interception(struct vcpu_svm *svm)
2172{
2173 if (emulate_instruction(&svm->vcpu, 0, 0, 0) != EMULATE_DONE)
2174 pr_unimpl(&svm->vcpu, "%s: failed\n", __func__);
2175 return 1;
2176}
2177
2178static int cr8_write_interception(struct vcpu_svm *svm)
2179{
2180 struct kvm_run *kvm_run = svm->vcpu.run;
2181
2182 u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
2183 /* instruction emulation calls kvm_set_cr8() */
2184 emulate_instruction(&svm->vcpu, 0, 0, 0);
2185 if (irqchip_in_kernel(svm->vcpu.kvm)) {
2186 svm->vmcb->control.intercept_cr_write &= ~INTERCEPT_CR8_MASK;
2187 return 1;
2188 }
2189 if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
2190 return 1;
2191 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
2192 return 0;
2193}
2194
2195static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
2196{
2197 struct vcpu_svm *svm = to_svm(vcpu);
2198
2199 switch (ecx) {
2200 case MSR_IA32_TSC: {
2201 u64 tsc_offset;
2202
2203 if (is_nested(svm))
2204 tsc_offset = svm->nested.hsave->control.tsc_offset;
2205 else
2206 tsc_offset = svm->vmcb->control.tsc_offset;
2207
2208 *data = tsc_offset + native_read_tsc();
2209 break;
2210 }
2211 case MSR_K6_STAR:
2212 *data = svm->vmcb->save.star;
2213 break;
2214#ifdef CONFIG_X86_64
2215 case MSR_LSTAR:
2216 *data = svm->vmcb->save.lstar;
2217 break;
2218 case MSR_CSTAR:
2219 *data = svm->vmcb->save.cstar;
2220 break;
2221 case MSR_KERNEL_GS_BASE:
2222 *data = svm->vmcb->save.kernel_gs_base;
2223 break;
2224 case MSR_SYSCALL_MASK:
2225 *data = svm->vmcb->save.sfmask;
2226 break;
2227#endif
2228 case MSR_IA32_SYSENTER_CS:
2229 *data = svm->vmcb->save.sysenter_cs;
2230 break;
2231 case MSR_IA32_SYSENTER_EIP:
2232 *data = svm->sysenter_eip;
2233 break;
2234 case MSR_IA32_SYSENTER_ESP:
2235 *data = svm->sysenter_esp;
2236 break;
2237 /*
2238 * Nobody will change the following 5 values in the VMCB so we can
2239 * safely return them on rdmsr. They will always be 0 until LBRV is
2240 * implemented.
2241 */
2242 case MSR_IA32_DEBUGCTLMSR:
2243 *data = svm->vmcb->save.dbgctl;
2244 break;
2245 case MSR_IA32_LASTBRANCHFROMIP:
2246 *data = svm->vmcb->save.br_from;
2247 break;
2248 case MSR_IA32_LASTBRANCHTOIP:
2249 *data = svm->vmcb->save.br_to;
2250 break;
2251 case MSR_IA32_LASTINTFROMIP:
2252 *data = svm->vmcb->save.last_excp_from;
2253 break;
2254 case MSR_IA32_LASTINTTOIP:
2255 *data = svm->vmcb->save.last_excp_to;
2256 break;
2257 case MSR_VM_HSAVE_PA:
2258 *data = svm->nested.hsave_msr;
2259 break;
2260 case MSR_VM_CR:
2261 *data = 0;
2262 break;
2263 case MSR_IA32_UCODE_REV:
2264 *data = 0x01000065;
2265 break;
2266 default:
2267 return kvm_get_msr_common(vcpu, ecx, data);
2268 }
2269 return 0;
2270}
2271
2272static int rdmsr_interception(struct vcpu_svm *svm)
2273{
2274 u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2275 u64 data;
2276
2277 if (svm_get_msr(&svm->vcpu, ecx, &data)) {
2278 trace_kvm_msr_read_ex(ecx);
2279 kvm_inject_gp(&svm->vcpu, 0);
2280 } else {
2281 trace_kvm_msr_read(ecx, data);
2282
2283 svm->vcpu.arch.regs[VCPU_REGS_RAX] = data & 0xffffffff;
2284 svm->vcpu.arch.regs[VCPU_REGS_RDX] = data >> 32;
2285 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2286 skip_emulated_instruction(&svm->vcpu);
2287 }
2288 return 1;
2289}
2290
2291static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
2292{
2293 struct vcpu_svm *svm = to_svm(vcpu);
2294
2295 switch (ecx) {
2296 case MSR_IA32_TSC: {
2297 u64 tsc_offset = data - native_read_tsc();
2298 u64 g_tsc_offset = 0;
2299
2300 if (is_nested(svm)) {
2301 g_tsc_offset = svm->vmcb->control.tsc_offset -
2302 svm->nested.hsave->control.tsc_offset;
2303 svm->nested.hsave->control.tsc_offset = tsc_offset;
2304 }
2305
2306 svm->vmcb->control.tsc_offset = tsc_offset + g_tsc_offset;
2307
2308 break;
2309 }
2310 case MSR_K6_STAR:
2311 svm->vmcb->save.star = data;
2312 break;
2313#ifdef CONFIG_X86_64
2314 case MSR_LSTAR:
2315 svm->vmcb->save.lstar = data;
2316 break;
2317 case MSR_CSTAR:
2318 svm->vmcb->save.cstar = data;
2319 break;
2320 case MSR_KERNEL_GS_BASE:
2321 svm->vmcb->save.kernel_gs_base = data;
2322 break;
2323 case MSR_SYSCALL_MASK:
2324 svm->vmcb->save.sfmask = data;
2325 break;
2326#endif
2327 case MSR_IA32_SYSENTER_CS:
2328 svm->vmcb->save.sysenter_cs = data;
2329 break;
2330 case MSR_IA32_SYSENTER_EIP:
2331 svm->sysenter_eip = data;
2332 svm->vmcb->save.sysenter_eip = data;
2333 break;
2334 case MSR_IA32_SYSENTER_ESP:
2335 svm->sysenter_esp = data;
2336 svm->vmcb->save.sysenter_esp = data;
2337 break;
2338 case MSR_IA32_DEBUGCTLMSR:
2339 if (!svm_has(SVM_FEATURE_LBRV)) {
2340 pr_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
2341 __func__, data);
2342 break;
2343 }
2344 if (data & DEBUGCTL_RESERVED_BITS)
2345 return 1;
2346
2347 svm->vmcb->save.dbgctl = data;
2348 if (data & (1ULL<<0))
2349 svm_enable_lbrv(svm);
2350 else
2351 svm_disable_lbrv(svm);
2352 break;
2353 case MSR_VM_HSAVE_PA:
2354 svm->nested.hsave_msr = data;
2355 break;
2356 case MSR_VM_CR:
2357 case MSR_VM_IGNNE:
2358 pr_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
2359 break;
2360 default:
2361 return kvm_set_msr_common(vcpu, ecx, data);
2362 }
2363 return 0;
2364}
2365
2366static int wrmsr_interception(struct vcpu_svm *svm)
2367{
2368 u32 ecx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2369 u64 data = (svm->vcpu.arch.regs[VCPU_REGS_RAX] & -1u)
2370 | ((u64)(svm->vcpu.arch.regs[VCPU_REGS_RDX] & -1u) << 32);
2371
2372
2373 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2374 if (svm_set_msr(&svm->vcpu, ecx, data)) {
2375 trace_kvm_msr_write_ex(ecx, data);
2376 kvm_inject_gp(&svm->vcpu, 0);
2377 } else {
2378 trace_kvm_msr_write(ecx, data);
2379 skip_emulated_instruction(&svm->vcpu);
2380 }
2381 return 1;
2382}
2383
2384static int msr_interception(struct vcpu_svm *svm)
2385{
2386 if (svm->vmcb->control.exit_info_1)
2387 return wrmsr_interception(svm);
2388 else
2389 return rdmsr_interception(svm);
2390}
2391
2392static int interrupt_window_interception(struct vcpu_svm *svm)
2393{
2394 struct kvm_run *kvm_run = svm->vcpu.run;
2395
2396 svm_clear_vintr(svm);
2397 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
2398 /*
2399 * If the user space waits to inject interrupts, exit as soon as
2400 * possible
2401 */
2402 if (!irqchip_in_kernel(svm->vcpu.kvm) &&
2403 kvm_run->request_interrupt_window &&
2404 !kvm_cpu_has_interrupt(&svm->vcpu)) {
2405 ++svm->vcpu.stat.irq_window_exits;
2406 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
2407 return 0;
2408 }
2409
2410 return 1;
2411}
2412
2413static int pause_interception(struct vcpu_svm *svm)
2414{
2415 kvm_vcpu_on_spin(&(svm->vcpu));
2416 return 1;
2417}
2418
2419static int (*svm_exit_handlers[])(struct vcpu_svm *svm) = {
2420 [SVM_EXIT_READ_CR0] = emulate_on_interception,
2421 [SVM_EXIT_READ_CR3] = emulate_on_interception,
2422 [SVM_EXIT_READ_CR4] = emulate_on_interception,
2423 [SVM_EXIT_READ_CR8] = emulate_on_interception,
2424 [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception,
2425 [SVM_EXIT_WRITE_CR0] = emulate_on_interception,
2426 [SVM_EXIT_WRITE_CR3] = emulate_on_interception,
2427 [SVM_EXIT_WRITE_CR4] = emulate_on_interception,
2428 [SVM_EXIT_WRITE_CR8] = cr8_write_interception,
2429 [SVM_EXIT_READ_DR0] = emulate_on_interception,
2430 [SVM_EXIT_READ_DR1] = emulate_on_interception,
2431 [SVM_EXIT_READ_DR2] = emulate_on_interception,
2432 [SVM_EXIT_READ_DR3] = emulate_on_interception,
2433 [SVM_EXIT_READ_DR4] = emulate_on_interception,
2434 [SVM_EXIT_READ_DR5] = emulate_on_interception,
2435 [SVM_EXIT_READ_DR6] = emulate_on_interception,
2436 [SVM_EXIT_READ_DR7] = emulate_on_interception,
2437 [SVM_EXIT_WRITE_DR0] = emulate_on_interception,
2438 [SVM_EXIT_WRITE_DR1] = emulate_on_interception,
2439 [SVM_EXIT_WRITE_DR2] = emulate_on_interception,
2440 [SVM_EXIT_WRITE_DR3] = emulate_on_interception,
2441 [SVM_EXIT_WRITE_DR4] = emulate_on_interception,
2442 [SVM_EXIT_WRITE_DR5] = emulate_on_interception,
2443 [SVM_EXIT_WRITE_DR6] = emulate_on_interception,
2444 [SVM_EXIT_WRITE_DR7] = emulate_on_interception,
2445 [SVM_EXIT_EXCP_BASE + DB_VECTOR] = db_interception,
2446 [SVM_EXIT_EXCP_BASE + BP_VECTOR] = bp_interception,
2447 [SVM_EXIT_EXCP_BASE + UD_VECTOR] = ud_interception,
2448 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
2449 [SVM_EXIT_EXCP_BASE + NM_VECTOR] = nm_interception,
2450 [SVM_EXIT_EXCP_BASE + MC_VECTOR] = mc_interception,
2451 [SVM_EXIT_INTR] = intr_interception,
2452 [SVM_EXIT_NMI] = nmi_interception,
2453 [SVM_EXIT_SMI] = nop_on_interception,
2454 [SVM_EXIT_INIT] = nop_on_interception,
2455 [SVM_EXIT_VINTR] = interrupt_window_interception,
2456 [SVM_EXIT_CPUID] = cpuid_interception,
2457 [SVM_EXIT_IRET] = iret_interception,
2458 [SVM_EXIT_INVD] = emulate_on_interception,
2459 [SVM_EXIT_PAUSE] = pause_interception,
2460 [SVM_EXIT_HLT] = halt_interception,
2461 [SVM_EXIT_INVLPG] = invlpg_interception,
2462 [SVM_EXIT_INVLPGA] = invlpga_interception,
2463 [SVM_EXIT_IOIO] = io_interception,
2464 [SVM_EXIT_MSR] = msr_interception,
2465 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
2466 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
2467 [SVM_EXIT_VMRUN] = vmrun_interception,
2468 [SVM_EXIT_VMMCALL] = vmmcall_interception,
2469 [SVM_EXIT_VMLOAD] = vmload_interception,
2470 [SVM_EXIT_VMSAVE] = vmsave_interception,
2471 [SVM_EXIT_STGI] = stgi_interception,
2472 [SVM_EXIT_CLGI] = clgi_interception,
2473 [SVM_EXIT_SKINIT] = skinit_interception,
2474 [SVM_EXIT_WBINVD] = emulate_on_interception,
2475 [SVM_EXIT_MONITOR] = invalid_op_interception,
2476 [SVM_EXIT_MWAIT] = invalid_op_interception,
2477 [SVM_EXIT_NPF] = pf_interception,
2478};
2479
2480static int handle_exit(struct kvm_vcpu *vcpu)
2481{
2482 struct vcpu_svm *svm = to_svm(vcpu);
2483 struct kvm_run *kvm_run = vcpu->run;
2484 u32 exit_code = svm->vmcb->control.exit_code;
2485
2486 trace_kvm_exit(exit_code, svm->vmcb->save.rip);
2487
2488 if (unlikely(svm->nested.exit_required)) {
2489 nested_svm_vmexit(svm);
2490 svm->nested.exit_required = false;
2491
2492 return 1;
2493 }
2494
2495 if (is_nested(svm)) {
2496 int vmexit;
2497
2498 trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code,
2499 svm->vmcb->control.exit_info_1,
2500 svm->vmcb->control.exit_info_2,
2501 svm->vmcb->control.exit_int_info,
2502 svm->vmcb->control.exit_int_info_err);
2503
2504 vmexit = nested_svm_exit_special(svm);
2505
2506 if (vmexit == NESTED_EXIT_CONTINUE)
2507 vmexit = nested_svm_exit_handled(svm);
2508
2509 if (vmexit == NESTED_EXIT_DONE)
2510 return 1;
2511 }
2512
2513 svm_complete_interrupts(svm);
2514
2515 if (!(svm->vmcb->control.intercept_cr_write & INTERCEPT_CR0_MASK))
2516 vcpu->arch.cr0 = svm->vmcb->save.cr0;
2517 if (npt_enabled)
2518 vcpu->arch.cr3 = svm->vmcb->save.cr3;
2519
2520 if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
2521 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
2522 kvm_run->fail_entry.hardware_entry_failure_reason
2523 = svm->vmcb->control.exit_code;
2524 return 0;
2525 }
2526
2527 if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
2528 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
2529 exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH)
2530 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
2531 "exit_code 0x%x\n",
2532 __func__, svm->vmcb->control.exit_int_info,
2533 exit_code);
2534
2535 if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
2536 || !svm_exit_handlers[exit_code]) {
2537 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
2538 kvm_run->hw.hardware_exit_reason = exit_code;
2539 return 0;
2540 }
2541
2542 return svm_exit_handlers[exit_code](svm);
2543}
2544
2545static void reload_tss(struct kvm_vcpu *vcpu)
2546{
2547 int cpu = raw_smp_processor_id();
2548
2549 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2550 sd->tss_desc->type = 9; /* available 32/64-bit TSS */
2551 load_TR_desc();
2552}
2553
2554static void pre_svm_run(struct vcpu_svm *svm)
2555{
2556 int cpu = raw_smp_processor_id();
2557
2558 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2559
2560 svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
2561 /* FIXME: handle wraparound of asid_generation */
2562 if (svm->asid_generation != sd->asid_generation)
2563 new_asid(svm, sd);
2564}
2565
2566static void svm_inject_nmi(struct kvm_vcpu *vcpu)
2567{
2568 struct vcpu_svm *svm = to_svm(vcpu);
2569
2570 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
2571 vcpu->arch.hflags |= HF_NMI_MASK;
2572 svm->vmcb->control.intercept |= (1UL << INTERCEPT_IRET);
2573 ++vcpu->stat.nmi_injections;
2574}
2575
2576static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
2577{
2578 struct vmcb_control_area *control;
2579
2580 trace_kvm_inj_virq(irq);
2581
2582 ++svm->vcpu.stat.irq_injections;
2583 control = &svm->vmcb->control;
2584 control->int_vector = irq;
2585 control->int_ctl &= ~V_INTR_PRIO_MASK;
2586 control->int_ctl |= V_IRQ_MASK |
2587 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
2588}
2589
2590static void svm_set_irq(struct kvm_vcpu *vcpu)
2591{
2592 struct vcpu_svm *svm = to_svm(vcpu);
2593
2594 BUG_ON(!(gif_set(svm)));
2595
2596 svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
2597 SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
2598}
2599
2600static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
2601{
2602 struct vcpu_svm *svm = to_svm(vcpu);
2603
2604 if (is_nested(svm) && (vcpu->arch.hflags & HF_VINTR_MASK))
2605 return;
2606
2607 if (irr == -1)
2608 return;
2609
2610 if (tpr >= irr)
2611 svm->vmcb->control.intercept_cr_write |= INTERCEPT_CR8_MASK;
2612}
2613
2614static int svm_nmi_allowed(struct kvm_vcpu *vcpu)
2615{
2616 struct vcpu_svm *svm = to_svm(vcpu);
2617 struct vmcb *vmcb = svm->vmcb;
2618 return !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
2619 !(svm->vcpu.arch.hflags & HF_NMI_MASK);
2620}
2621
2622static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
2623{
2624 struct vcpu_svm *svm = to_svm(vcpu);
2625
2626 return !!(svm->vcpu.arch.hflags & HF_NMI_MASK);
2627}
2628
2629static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
2630{
2631 struct vcpu_svm *svm = to_svm(vcpu);
2632
2633 if (masked) {
2634 svm->vcpu.arch.hflags |= HF_NMI_MASK;
2635 svm->vmcb->control.intercept |= (1UL << INTERCEPT_IRET);
2636 } else {
2637 svm->vcpu.arch.hflags &= ~HF_NMI_MASK;
2638 svm->vmcb->control.intercept &= ~(1UL << INTERCEPT_IRET);
2639 }
2640}
2641
2642static int svm_interrupt_allowed(struct kvm_vcpu *vcpu)
2643{
2644 struct vcpu_svm *svm = to_svm(vcpu);
2645 struct vmcb *vmcb = svm->vmcb;
2646 int ret;
2647
2648 if (!gif_set(svm) ||
2649 (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK))
2650 return 0;
2651
2652 ret = !!(vmcb->save.rflags & X86_EFLAGS_IF);
2653
2654 if (is_nested(svm))
2655 return ret && !(svm->vcpu.arch.hflags & HF_VINTR_MASK);
2656
2657 return ret;
2658}
2659
2660static void enable_irq_window(struct kvm_vcpu *vcpu)
2661{
2662 struct vcpu_svm *svm = to_svm(vcpu);
2663
2664 /*
2665 * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
2666 * 1, because that's a separate STGI/VMRUN intercept. The next time we
2667 * get that intercept, this function will be called again though and
2668 * we'll get the vintr intercept.
2669 */
2670 if (gif_set(svm) && nested_svm_intr(svm)) {
2671 svm_set_vintr(svm);
2672 svm_inject_irq(svm, 0x0);
2673 }
2674}
2675
2676static void enable_nmi_window(struct kvm_vcpu *vcpu)
2677{
2678 struct vcpu_svm *svm = to_svm(vcpu);
2679
2680 if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
2681 == HF_NMI_MASK)
2682 return; /* IRET will cause a vm exit */
2683
2684 /*
2685 * Something prevents NMI from been injected. Single step over possible
2686 * problem (IRET or exception injection or interrupt shadow)
2687 */
2688 svm->nmi_singlestep = true;
2689 svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
2690 update_db_intercept(vcpu);
2691}
2692
2693static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
2694{
2695 return 0;
2696}
2697
2698static void svm_flush_tlb(struct kvm_vcpu *vcpu)
2699{
2700 force_new_asid(vcpu);
2701}
2702
2703static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
2704{
2705}
2706
2707static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
2708{
2709 struct vcpu_svm *svm = to_svm(vcpu);
2710
2711 if (is_nested(svm) && (vcpu->arch.hflags & HF_VINTR_MASK))
2712 return;
2713
2714 if (!(svm->vmcb->control.intercept_cr_write & INTERCEPT_CR8_MASK)) {
2715 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
2716 kvm_set_cr8(vcpu, cr8);
2717 }
2718}
2719
2720static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
2721{
2722 struct vcpu_svm *svm = to_svm(vcpu);
2723 u64 cr8;
2724
2725 if (is_nested(svm) && (vcpu->arch.hflags & HF_VINTR_MASK))
2726 return;
2727
2728 cr8 = kvm_get_cr8(vcpu);
2729 svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
2730 svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
2731}
2732
2733static void svm_complete_interrupts(struct vcpu_svm *svm)
2734{
2735 u8 vector;
2736 int type;
2737 u32 exitintinfo = svm->vmcb->control.exit_int_info;
2738 unsigned int3_injected = svm->int3_injected;
2739
2740 svm->int3_injected = 0;
2741
2742 if (svm->vcpu.arch.hflags & HF_IRET_MASK)
2743 svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
2744
2745 svm->vcpu.arch.nmi_injected = false;
2746 kvm_clear_exception_queue(&svm->vcpu);
2747 kvm_clear_interrupt_queue(&svm->vcpu);
2748
2749 if (!(exitintinfo & SVM_EXITINTINFO_VALID))
2750 return;
2751
2752 vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
2753 type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
2754
2755 switch (type) {
2756 case SVM_EXITINTINFO_TYPE_NMI:
2757 svm->vcpu.arch.nmi_injected = true;
2758 break;
2759 case SVM_EXITINTINFO_TYPE_EXEPT:
2760 if (is_nested(svm))
2761 break;
2762 /*
2763 * In case of software exceptions, do not reinject the vector,
2764 * but re-execute the instruction instead. Rewind RIP first
2765 * if we emulated INT3 before.
2766 */
2767 if (kvm_exception_is_soft(vector)) {
2768 if (vector == BP_VECTOR && int3_injected &&
2769 kvm_is_linear_rip(&svm->vcpu, svm->int3_rip))
2770 kvm_rip_write(&svm->vcpu,
2771 kvm_rip_read(&svm->vcpu) -
2772 int3_injected);
2773 break;
2774 }
2775 if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
2776 u32 err = svm->vmcb->control.exit_int_info_err;
2777 kvm_queue_exception_e(&svm->vcpu, vector, err);
2778
2779 } else
2780 kvm_queue_exception(&svm->vcpu, vector);
2781 break;
2782 case SVM_EXITINTINFO_TYPE_INTR:
2783 kvm_queue_interrupt(&svm->vcpu, vector, false);
2784 break;
2785 default:
2786 break;
2787 }
2788}
2789
2790#ifdef CONFIG_X86_64
2791#define R "r"
2792#else
2793#define R "e"
2794#endif
2795
2796static void svm_vcpu_run(struct kvm_vcpu *vcpu)
2797{
2798 struct vcpu_svm *svm = to_svm(vcpu);
2799 u16 fs_selector;
2800 u16 gs_selector;
2801 u16 ldt_selector;
2802
2803 /*
2804 * A vmexit emulation is required before the vcpu can be executed
2805 * again.
2806 */
2807 if (unlikely(svm->nested.exit_required))
2808 return;
2809
2810 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
2811 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
2812 svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
2813
2814 pre_svm_run(svm);
2815
2816 sync_lapic_to_cr8(vcpu);
2817
2818 save_host_msrs(vcpu);
2819 fs_selector = kvm_read_fs();
2820 gs_selector = kvm_read_gs();
2821 ldt_selector = kvm_read_ldt();
2822 svm->vmcb->save.cr2 = vcpu->arch.cr2;
2823 /* required for live migration with NPT */
2824 if (npt_enabled)
2825 svm->vmcb->save.cr3 = vcpu->arch.cr3;
2826
2827 clgi();
2828
2829 local_irq_enable();
2830
2831 asm volatile (
2832 "push %%"R"bp; \n\t"
2833 "mov %c[rbx](%[svm]), %%"R"bx \n\t"
2834 "mov %c[rcx](%[svm]), %%"R"cx \n\t"
2835 "mov %c[rdx](%[svm]), %%"R"dx \n\t"
2836 "mov %c[rsi](%[svm]), %%"R"si \n\t"
2837 "mov %c[rdi](%[svm]), %%"R"di \n\t"
2838 "mov %c[rbp](%[svm]), %%"R"bp \n\t"
2839#ifdef CONFIG_X86_64
2840 "mov %c[r8](%[svm]), %%r8 \n\t"
2841 "mov %c[r9](%[svm]), %%r9 \n\t"
2842 "mov %c[r10](%[svm]), %%r10 \n\t"
2843 "mov %c[r11](%[svm]), %%r11 \n\t"
2844 "mov %c[r12](%[svm]), %%r12 \n\t"
2845 "mov %c[r13](%[svm]), %%r13 \n\t"
2846 "mov %c[r14](%[svm]), %%r14 \n\t"
2847 "mov %c[r15](%[svm]), %%r15 \n\t"
2848#endif
2849
2850 /* Enter guest mode */
2851 "push %%"R"ax \n\t"
2852 "mov %c[vmcb](%[svm]), %%"R"ax \n\t"
2853 __ex(SVM_VMLOAD) "\n\t"
2854 __ex(SVM_VMRUN) "\n\t"
2855 __ex(SVM_VMSAVE) "\n\t"
2856 "pop %%"R"ax \n\t"
2857
2858 /* Save guest registers, load host registers */
2859 "mov %%"R"bx, %c[rbx](%[svm]) \n\t"
2860 "mov %%"R"cx, %c[rcx](%[svm]) \n\t"
2861 "mov %%"R"dx, %c[rdx](%[svm]) \n\t"
2862 "mov %%"R"si, %c[rsi](%[svm]) \n\t"
2863 "mov %%"R"di, %c[rdi](%[svm]) \n\t"
2864 "mov %%"R"bp, %c[rbp](%[svm]) \n\t"
2865#ifdef CONFIG_X86_64
2866 "mov %%r8, %c[r8](%[svm]) \n\t"
2867 "mov %%r9, %c[r9](%[svm]) \n\t"
2868 "mov %%r10, %c[r10](%[svm]) \n\t"
2869 "mov %%r11, %c[r11](%[svm]) \n\t"
2870 "mov %%r12, %c[r12](%[svm]) \n\t"
2871 "mov %%r13, %c[r13](%[svm]) \n\t"
2872 "mov %%r14, %c[r14](%[svm]) \n\t"
2873 "mov %%r15, %c[r15](%[svm]) \n\t"
2874#endif
2875 "pop %%"R"bp"
2876 :
2877 : [svm]"a"(svm),
2878 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
2879 [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
2880 [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
2881 [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
2882 [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
2883 [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
2884 [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
2885#ifdef CONFIG_X86_64
2886 , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
2887 [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
2888 [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
2889 [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
2890 [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
2891 [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
2892 [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
2893 [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
2894#endif
2895 : "cc", "memory"
2896 , R"bx", R"cx", R"dx", R"si", R"di"
2897#ifdef CONFIG_X86_64
2898 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
2899#endif
2900 );
2901
2902 vcpu->arch.cr2 = svm->vmcb->save.cr2;
2903 vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
2904 vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
2905 vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
2906
2907 kvm_load_fs(fs_selector);
2908 kvm_load_gs(gs_selector);
2909 kvm_load_ldt(ldt_selector);
2910 load_host_msrs(vcpu);
2911
2912 reload_tss(vcpu);
2913
2914 local_irq_disable();
2915
2916 stgi();
2917
2918 sync_cr8_to_lapic(vcpu);
2919
2920 svm->next_rip = 0;
2921
2922 if (npt_enabled) {
2923 vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
2924 vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
2925 }
2926}
2927
2928#undef R
2929
2930static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
2931{
2932 struct vcpu_svm *svm = to_svm(vcpu);
2933
2934 if (npt_enabled) {
2935 svm->vmcb->control.nested_cr3 = root;
2936 force_new_asid(vcpu);
2937 return;
2938 }
2939
2940 svm->vmcb->save.cr3 = root;
2941 force_new_asid(vcpu);
2942}
2943
2944static int is_disabled(void)
2945{
2946 u64 vm_cr;
2947
2948 rdmsrl(MSR_VM_CR, vm_cr);
2949 if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
2950 return 1;
2951
2952 return 0;
2953}
2954
2955static void
2956svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
2957{
2958 /*
2959 * Patch in the VMMCALL instruction:
2960 */
2961 hypercall[0] = 0x0f;
2962 hypercall[1] = 0x01;
2963 hypercall[2] = 0xd9;
2964}
2965
2966static void svm_check_processor_compat(void *rtn)
2967{
2968 *(int *)rtn = 0;
2969}
2970
2971static bool svm_cpu_has_accelerated_tpr(void)
2972{
2973 return false;
2974}
2975
2976static int get_npt_level(void)
2977{
2978#ifdef CONFIG_X86_64
2979 return PT64_ROOT_LEVEL;
2980#else
2981 return PT32E_ROOT_LEVEL;
2982#endif
2983}
2984
2985static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
2986{
2987 return 0;
2988}
2989
2990static void svm_cpuid_update(struct kvm_vcpu *vcpu)
2991{
2992}
2993
2994static const struct trace_print_flags svm_exit_reasons_str[] = {
2995 { SVM_EXIT_READ_CR0, "read_cr0" },
2996 { SVM_EXIT_READ_CR3, "read_cr3" },
2997 { SVM_EXIT_READ_CR4, "read_cr4" },
2998 { SVM_EXIT_READ_CR8, "read_cr8" },
2999 { SVM_EXIT_WRITE_CR0, "write_cr0" },
3000 { SVM_EXIT_WRITE_CR3, "write_cr3" },
3001 { SVM_EXIT_WRITE_CR4, "write_cr4" },
3002 { SVM_EXIT_WRITE_CR8, "write_cr8" },
3003 { SVM_EXIT_READ_DR0, "read_dr0" },
3004 { SVM_EXIT_READ_DR1, "read_dr1" },
3005 { SVM_EXIT_READ_DR2, "read_dr2" },
3006 { SVM_EXIT_READ_DR3, "read_dr3" },
3007 { SVM_EXIT_WRITE_DR0, "write_dr0" },
3008 { SVM_EXIT_WRITE_DR1, "write_dr1" },
3009 { SVM_EXIT_WRITE_DR2, "write_dr2" },
3010 { SVM_EXIT_WRITE_DR3, "write_dr3" },
3011 { SVM_EXIT_WRITE_DR5, "write_dr5" },
3012 { SVM_EXIT_WRITE_DR7, "write_dr7" },
3013 { SVM_EXIT_EXCP_BASE + DB_VECTOR, "DB excp" },
3014 { SVM_EXIT_EXCP_BASE + BP_VECTOR, "BP excp" },
3015 { SVM_EXIT_EXCP_BASE + UD_VECTOR, "UD excp" },
3016 { SVM_EXIT_EXCP_BASE + PF_VECTOR, "PF excp" },
3017 { SVM_EXIT_EXCP_BASE + NM_VECTOR, "NM excp" },
3018 { SVM_EXIT_EXCP_BASE + MC_VECTOR, "MC excp" },
3019 { SVM_EXIT_INTR, "interrupt" },
3020 { SVM_EXIT_NMI, "nmi" },
3021 { SVM_EXIT_SMI, "smi" },
3022 { SVM_EXIT_INIT, "init" },
3023 { SVM_EXIT_VINTR, "vintr" },
3024 { SVM_EXIT_CPUID, "cpuid" },
3025 { SVM_EXIT_INVD, "invd" },
3026 { SVM_EXIT_HLT, "hlt" },
3027 { SVM_EXIT_INVLPG, "invlpg" },
3028 { SVM_EXIT_INVLPGA, "invlpga" },
3029 { SVM_EXIT_IOIO, "io" },
3030 { SVM_EXIT_MSR, "msr" },
3031 { SVM_EXIT_TASK_SWITCH, "task_switch" },
3032 { SVM_EXIT_SHUTDOWN, "shutdown" },
3033 { SVM_EXIT_VMRUN, "vmrun" },
3034 { SVM_EXIT_VMMCALL, "hypercall" },
3035 { SVM_EXIT_VMLOAD, "vmload" },
3036 { SVM_EXIT_VMSAVE, "vmsave" },
3037 { SVM_EXIT_STGI, "stgi" },
3038 { SVM_EXIT_CLGI, "clgi" },
3039 { SVM_EXIT_SKINIT, "skinit" },
3040 { SVM_EXIT_WBINVD, "wbinvd" },
3041 { SVM_EXIT_MONITOR, "monitor" },
3042 { SVM_EXIT_MWAIT, "mwait" },
3043 { SVM_EXIT_NPF, "npf" },
3044 { -1, NULL }
3045};
3046
3047static int svm_get_lpage_level(void)
3048{
3049 return PT_PDPE_LEVEL;
3050}
3051
3052static bool svm_rdtscp_supported(void)
3053{
3054 return false;
3055}
3056
3057static void svm_fpu_deactivate(struct kvm_vcpu *vcpu)
3058{
3059 struct vcpu_svm *svm = to_svm(vcpu);
3060
3061 svm->vmcb->control.intercept_exceptions |= 1 << NM_VECTOR;
3062 if (is_nested(svm))
3063 svm->nested.hsave->control.intercept_exceptions |= 1 << NM_VECTOR;
3064 update_cr0_intercept(svm);
3065}
3066
3067static struct kvm_x86_ops svm_x86_ops = {
3068 .cpu_has_kvm_support = has_svm,
3069 .disabled_by_bios = is_disabled,
3070 .hardware_setup = svm_hardware_setup,
3071 .hardware_unsetup = svm_hardware_unsetup,
3072 .check_processor_compatibility = svm_check_processor_compat,
3073 .hardware_enable = svm_hardware_enable,
3074 .hardware_disable = svm_hardware_disable,
3075 .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
3076
3077 .vcpu_create = svm_create_vcpu,
3078 .vcpu_free = svm_free_vcpu,
3079 .vcpu_reset = svm_vcpu_reset,
3080
3081 .prepare_guest_switch = svm_prepare_guest_switch,
3082 .vcpu_load = svm_vcpu_load,
3083 .vcpu_put = svm_vcpu_put,
3084
3085 .set_guest_debug = svm_guest_debug,
3086 .get_msr = svm_get_msr,
3087 .set_msr = svm_set_msr,
3088 .get_segment_base = svm_get_segment_base,
3089 .get_segment = svm_get_segment,
3090 .set_segment = svm_set_segment,
3091 .get_cpl = svm_get_cpl,
3092 .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
3093 .decache_cr0_guest_bits = svm_decache_cr0_guest_bits,
3094 .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
3095 .set_cr0 = svm_set_cr0,
3096 .set_cr3 = svm_set_cr3,
3097 .set_cr4 = svm_set_cr4,
3098 .set_efer = svm_set_efer,
3099 .get_idt = svm_get_idt,
3100 .set_idt = svm_set_idt,
3101 .get_gdt = svm_get_gdt,
3102 .set_gdt = svm_set_gdt,
3103 .get_dr = svm_get_dr,
3104 .set_dr = svm_set_dr,
3105 .cache_reg = svm_cache_reg,
3106 .get_rflags = svm_get_rflags,
3107 .set_rflags = svm_set_rflags,
3108 .fpu_activate = svm_fpu_activate,
3109 .fpu_deactivate = svm_fpu_deactivate,
3110
3111 .tlb_flush = svm_flush_tlb,
3112
3113 .run = svm_vcpu_run,
3114 .handle_exit = handle_exit,
3115 .skip_emulated_instruction = skip_emulated_instruction,
3116 .set_interrupt_shadow = svm_set_interrupt_shadow,
3117 .get_interrupt_shadow = svm_get_interrupt_shadow,
3118 .patch_hypercall = svm_patch_hypercall,
3119 .set_irq = svm_set_irq,
3120 .set_nmi = svm_inject_nmi,
3121 .queue_exception = svm_queue_exception,
3122 .interrupt_allowed = svm_interrupt_allowed,
3123 .nmi_allowed = svm_nmi_allowed,
3124 .get_nmi_mask = svm_get_nmi_mask,
3125 .set_nmi_mask = svm_set_nmi_mask,
3126 .enable_nmi_window = enable_nmi_window,
3127 .enable_irq_window = enable_irq_window,
3128 .update_cr8_intercept = update_cr8_intercept,
3129
3130 .set_tss_addr = svm_set_tss_addr,
3131 .get_tdp_level = get_npt_level,
3132 .get_mt_mask = svm_get_mt_mask,
3133
3134 .exit_reasons_str = svm_exit_reasons_str,
3135 .get_lpage_level = svm_get_lpage_level,
3136
3137 .cpuid_update = svm_cpuid_update,
3138
3139 .rdtscp_supported = svm_rdtscp_supported,
3140};
3141
3142static int __init svm_init(void)
3143{
3144 return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
3145 THIS_MODULE);
3146}
3147
3148static void __exit svm_exit(void)
3149{
3150 kvm_exit();
3151}
3152
3153module_init(svm_init)
3154module_exit(svm_exit)
Ubuntu Artful kernel mirror