]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - drivers/kvm/svm.c
[PATCH] KVM: MMU: Report nx faults to the guest
[mirror_ubuntu-bionic-kernel.git] / drivers / kvm / svm.c
CommitLineData
6aa8b732
AK
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
17#include <linux/module.h>
18#include <linux/vmalloc.h>
19#include <linux/highmem.h>
07031e14 20#include <linux/profile.h>
6aa8b732
AK
21#include <asm/desc.h>
22
23#include "kvm_svm.h"
24#include "x86_emulate.h"
25
26MODULE_AUTHOR("Qumranet");
27MODULE_LICENSE("GPL");
28
29#define IOPM_ALLOC_ORDER 2
30#define MSRPM_ALLOC_ORDER 1
31
32#define DB_VECTOR 1
33#define UD_VECTOR 6
34#define GP_VECTOR 13
35
36#define DR7_GD_MASK (1 << 13)
37#define DR6_BD_MASK (1 << 13)
38#define CR4_DE_MASK (1UL << 3)
39
40#define SEG_TYPE_LDT 2
41#define SEG_TYPE_BUSY_TSS16 3
42
43#define KVM_EFER_LMA (1 << 10)
44#define KVM_EFER_LME (1 << 8)
45
46unsigned long iopm_base;
47unsigned long msrpm_base;
48
49struct kvm_ldttss_desc {
50 u16 limit0;
51 u16 base0;
52 unsigned base1 : 8, type : 5, dpl : 2, p : 1;
53 unsigned limit1 : 4, zero0 : 3, g : 1, base2 : 8;
54 u32 base3;
55 u32 zero1;
56} __attribute__((packed));
57
58struct svm_cpu_data {
59 int cpu;
60
61 uint64_t asid_generation;
62 uint32_t max_asid;
63 uint32_t next_asid;
64 struct kvm_ldttss_desc *tss_desc;
65
66 struct page *save_area;
67};
68
69static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
70
71struct svm_init_data {
72 int cpu;
73 int r;
74};
75
76static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
77
78#define NUM_MSR_MAPS (sizeof(msrpm_ranges) / sizeof(*msrpm_ranges))
79#define MSRS_RANGE_SIZE 2048
80#define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
81
82#define MAX_INST_SIZE 15
83
84static unsigned get_addr_size(struct kvm_vcpu *vcpu)
85{
86 struct vmcb_save_area *sa = &vcpu->svm->vmcb->save;
87 u16 cs_attrib;
88
89 if (!(sa->cr0 & CR0_PE_MASK) || (sa->rflags & X86_EFLAGS_VM))
90 return 2;
91
92 cs_attrib = sa->cs.attrib;
93
94 return (cs_attrib & SVM_SELECTOR_L_MASK) ? 8 :
95 (cs_attrib & SVM_SELECTOR_DB_MASK) ? 4 : 2;
96}
97
98static inline u8 pop_irq(struct kvm_vcpu *vcpu)
99{
100 int word_index = __ffs(vcpu->irq_summary);
101 int bit_index = __ffs(vcpu->irq_pending[word_index]);
102 int irq = word_index * BITS_PER_LONG + bit_index;
103
104 clear_bit(bit_index, &vcpu->irq_pending[word_index]);
105 if (!vcpu->irq_pending[word_index])
106 clear_bit(word_index, &vcpu->irq_summary);
107 return irq;
108}
109
110static inline void push_irq(struct kvm_vcpu *vcpu, u8 irq)
111{
112 set_bit(irq, vcpu->irq_pending);
113 set_bit(irq / BITS_PER_LONG, &vcpu->irq_summary);
114}
115
116static inline void clgi(void)
117{
118 asm volatile (SVM_CLGI);
119}
120
121static inline void stgi(void)
122{
123 asm volatile (SVM_STGI);
124}
125
126static inline void invlpga(unsigned long addr, u32 asid)
127{
128 asm volatile (SVM_INVLPGA :: "a"(addr), "c"(asid));
129}
130
131static inline unsigned long kvm_read_cr2(void)
132{
133 unsigned long cr2;
134
135 asm volatile ("mov %%cr2, %0" : "=r" (cr2));
136 return cr2;
137}
138
139static inline void kvm_write_cr2(unsigned long val)
140{
141 asm volatile ("mov %0, %%cr2" :: "r" (val));
142}
143
144static inline unsigned long read_dr6(void)
145{
146 unsigned long dr6;
147
148 asm volatile ("mov %%dr6, %0" : "=r" (dr6));
149 return dr6;
150}
151
152static inline void write_dr6(unsigned long val)
153{
154 asm volatile ("mov %0, %%dr6" :: "r" (val));
155}
156
157static inline unsigned long read_dr7(void)
158{
159 unsigned long dr7;
160
161 asm volatile ("mov %%dr7, %0" : "=r" (dr7));
162 return dr7;
163}
164
165static inline void write_dr7(unsigned long val)
166{
167 asm volatile ("mov %0, %%dr7" :: "r" (val));
168}
169
6aa8b732
AK
170static inline void force_new_asid(struct kvm_vcpu *vcpu)
171{
172 vcpu->svm->asid_generation--;
173}
174
175static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
176{
177 force_new_asid(vcpu);
178}
179
180static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
181{
182 if (!(efer & KVM_EFER_LMA))
183 efer &= ~KVM_EFER_LME;
184
185 vcpu->svm->vmcb->save.efer = efer | MSR_EFER_SVME_MASK;
186 vcpu->shadow_efer = efer;
187}
188
189static void svm_inject_gp(struct kvm_vcpu *vcpu, unsigned error_code)
190{
191 vcpu->svm->vmcb->control.event_inj = SVM_EVTINJ_VALID |
192 SVM_EVTINJ_VALID_ERR |
193 SVM_EVTINJ_TYPE_EXEPT |
194 GP_VECTOR;
195 vcpu->svm->vmcb->control.event_inj_err = error_code;
196}
197
198static void inject_ud(struct kvm_vcpu *vcpu)
199{
200 vcpu->svm->vmcb->control.event_inj = SVM_EVTINJ_VALID |
201 SVM_EVTINJ_TYPE_EXEPT |
202 UD_VECTOR;
203}
204
205static void inject_db(struct kvm_vcpu *vcpu)
206{
207 vcpu->svm->vmcb->control.event_inj = SVM_EVTINJ_VALID |
208 SVM_EVTINJ_TYPE_EXEPT |
209 DB_VECTOR;
210}
211
212static int is_page_fault(uint32_t info)
213{
214 info &= SVM_EVTINJ_VEC_MASK | SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
215 return info == (PF_VECTOR | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT);
216}
217
218static int is_external_interrupt(u32 info)
219{
220 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
221 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
222}
223
224static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
225{
226 if (!vcpu->svm->next_rip) {
227 printk(KERN_DEBUG "%s: NOP\n", __FUNCTION__);
228 return;
229 }
230 if (vcpu->svm->next_rip - vcpu->svm->vmcb->save.rip > 15) {
231 printk(KERN_ERR "%s: ip 0x%llx next 0x%llx\n",
232 __FUNCTION__,
233 vcpu->svm->vmcb->save.rip,
234 vcpu->svm->next_rip);
235 }
236
237 vcpu->rip = vcpu->svm->vmcb->save.rip = vcpu->svm->next_rip;
238 vcpu->svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
c1150d8c
DL
239
240 vcpu->interrupt_window_open = 1;
6aa8b732
AK
241}
242
243static int has_svm(void)
244{
245 uint32_t eax, ebx, ecx, edx;
246
1e885461 247 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD) {
6aa8b732
AK
248 printk(KERN_INFO "has_svm: not amd\n");
249 return 0;
250 }
251
252 cpuid(0x80000000, &eax, &ebx, &ecx, &edx);
253 if (eax < SVM_CPUID_FUNC) {
254 printk(KERN_INFO "has_svm: can't execute cpuid_8000000a\n");
255 return 0;
256 }
257
258 cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
259 if (!(ecx & (1 << SVM_CPUID_FEATURE_SHIFT))) {
260 printk(KERN_DEBUG "has_svm: svm not available\n");
261 return 0;
262 }
263 return 1;
264}
265
266static void svm_hardware_disable(void *garbage)
267{
268 struct svm_cpu_data *svm_data
269 = per_cpu(svm_data, raw_smp_processor_id());
270
271 if (svm_data) {
272 uint64_t efer;
273
274 wrmsrl(MSR_VM_HSAVE_PA, 0);
275 rdmsrl(MSR_EFER, efer);
276 wrmsrl(MSR_EFER, efer & ~MSR_EFER_SVME_MASK);
277 per_cpu(svm_data, raw_smp_processor_id()) = 0;
278 __free_page(svm_data->save_area);
279 kfree(svm_data);
280 }
281}
282
283static void svm_hardware_enable(void *garbage)
284{
285
286 struct svm_cpu_data *svm_data;
287 uint64_t efer;
05b3e0c2 288#ifdef CONFIG_X86_64
6aa8b732
AK
289 struct desc_ptr gdt_descr;
290#else
291 struct Xgt_desc_struct gdt_descr;
292#endif
293 struct desc_struct *gdt;
294 int me = raw_smp_processor_id();
295
296 if (!has_svm()) {
297 printk(KERN_ERR "svm_cpu_init: err EOPNOTSUPP on %d\n", me);
298 return;
299 }
300 svm_data = per_cpu(svm_data, me);
301
302 if (!svm_data) {
303 printk(KERN_ERR "svm_cpu_init: svm_data is NULL on %d\n",
304 me);
305 return;
306 }
307
308 svm_data->asid_generation = 1;
309 svm_data->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
310 svm_data->next_asid = svm_data->max_asid + 1;
311
312 asm volatile ( "sgdt %0" : "=m"(gdt_descr) );
313 gdt = (struct desc_struct *)gdt_descr.address;
314 svm_data->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
315
316 rdmsrl(MSR_EFER, efer);
317 wrmsrl(MSR_EFER, efer | MSR_EFER_SVME_MASK);
318
319 wrmsrl(MSR_VM_HSAVE_PA,
320 page_to_pfn(svm_data->save_area) << PAGE_SHIFT);
321}
322
323static int svm_cpu_init(int cpu)
324{
325 struct svm_cpu_data *svm_data;
326 int r;
327
328 svm_data = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
329 if (!svm_data)
330 return -ENOMEM;
331 svm_data->cpu = cpu;
332 svm_data->save_area = alloc_page(GFP_KERNEL);
333 r = -ENOMEM;
334 if (!svm_data->save_area)
335 goto err_1;
336
337 per_cpu(svm_data, cpu) = svm_data;
338
339 return 0;
340
341err_1:
342 kfree(svm_data);
343 return r;
344
345}
346
347static int set_msr_interception(u32 *msrpm, unsigned msr,
348 int read, int write)
349{
350 int i;
351
352 for (i = 0; i < NUM_MSR_MAPS; i++) {
353 if (msr >= msrpm_ranges[i] &&
354 msr < msrpm_ranges[i] + MSRS_IN_RANGE) {
355 u32 msr_offset = (i * MSRS_IN_RANGE + msr -
356 msrpm_ranges[i]) * 2;
357
358 u32 *base = msrpm + (msr_offset / 32);
359 u32 msr_shift = msr_offset % 32;
360 u32 mask = ((write) ? 0 : 2) | ((read) ? 0 : 1);
361 *base = (*base & ~(0x3 << msr_shift)) |
362 (mask << msr_shift);
363 return 1;
364 }
365 }
366 printk(KERN_DEBUG "%s: not found 0x%x\n", __FUNCTION__, msr);
367 return 0;
368}
369
370static __init int svm_hardware_setup(void)
371{
372 int cpu;
373 struct page *iopm_pages;
374 struct page *msrpm_pages;
375 void *msrpm_va;
376 int r;
377
873a7c42 378 kvm_emulator_want_group7_invlpg();
6aa8b732
AK
379
380 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
381
382 if (!iopm_pages)
383 return -ENOMEM;
384 memset(page_address(iopm_pages), 0xff,
385 PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
386 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
387
388
389 msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
390
391 r = -ENOMEM;
392 if (!msrpm_pages)
393 goto err_1;
394
395 msrpm_va = page_address(msrpm_pages);
396 memset(msrpm_va, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
397 msrpm_base = page_to_pfn(msrpm_pages) << PAGE_SHIFT;
398
05b3e0c2 399#ifdef CONFIG_X86_64
6aa8b732
AK
400 set_msr_interception(msrpm_va, MSR_GS_BASE, 1, 1);
401 set_msr_interception(msrpm_va, MSR_FS_BASE, 1, 1);
402 set_msr_interception(msrpm_va, MSR_KERNEL_GS_BASE, 1, 1);
6aa8b732
AK
403 set_msr_interception(msrpm_va, MSR_LSTAR, 1, 1);
404 set_msr_interception(msrpm_va, MSR_CSTAR, 1, 1);
405 set_msr_interception(msrpm_va, MSR_SYSCALL_MASK, 1, 1);
406#endif
0e859cac 407 set_msr_interception(msrpm_va, MSR_K6_STAR, 1, 1);
6aa8b732
AK
408 set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_CS, 1, 1);
409 set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_ESP, 1, 1);
410 set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_EIP, 1, 1);
411
412 for_each_online_cpu(cpu) {
413 r = svm_cpu_init(cpu);
414 if (r)
415 goto err_2;
416 }
417 return 0;
418
419err_2:
420 __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
421 msrpm_base = 0;
422err_1:
423 __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
424 iopm_base = 0;
425 return r;
426}
427
428static __exit void svm_hardware_unsetup(void)
429{
430 __free_pages(pfn_to_page(msrpm_base >> PAGE_SHIFT), MSRPM_ALLOC_ORDER);
431 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
432 iopm_base = msrpm_base = 0;
433}
434
435static void init_seg(struct vmcb_seg *seg)
436{
437 seg->selector = 0;
438 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
439 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
440 seg->limit = 0xffff;
441 seg->base = 0;
442}
443
444static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
445{
446 seg->selector = 0;
447 seg->attrib = SVM_SELECTOR_P_MASK | type;
448 seg->limit = 0xffff;
449 seg->base = 0;
450}
451
452static int svm_vcpu_setup(struct kvm_vcpu *vcpu)
453{
454 return 0;
455}
456
457static void init_vmcb(struct vmcb *vmcb)
458{
459 struct vmcb_control_area *control = &vmcb->control;
460 struct vmcb_save_area *save = &vmcb->save;
461 u64 tsc;
462
463 control->intercept_cr_read = INTERCEPT_CR0_MASK |
464 INTERCEPT_CR3_MASK |
465 INTERCEPT_CR4_MASK;
466
467 control->intercept_cr_write = INTERCEPT_CR0_MASK |
468 INTERCEPT_CR3_MASK |
469 INTERCEPT_CR4_MASK;
470
471 control->intercept_dr_read = INTERCEPT_DR0_MASK |
472 INTERCEPT_DR1_MASK |
473 INTERCEPT_DR2_MASK |
474 INTERCEPT_DR3_MASK;
475
476 control->intercept_dr_write = INTERCEPT_DR0_MASK |
477 INTERCEPT_DR1_MASK |
478 INTERCEPT_DR2_MASK |
479 INTERCEPT_DR3_MASK |
480 INTERCEPT_DR5_MASK |
481 INTERCEPT_DR7_MASK;
482
483 control->intercept_exceptions = 1 << PF_VECTOR;
484
485
486 control->intercept = (1ULL << INTERCEPT_INTR) |
487 (1ULL << INTERCEPT_NMI) |
488 /*
489 * selective cr0 intercept bug?
490 * 0: 0f 22 d8 mov %eax,%cr3
491 * 3: 0f 20 c0 mov %cr0,%eax
492 * 6: 0d 00 00 00 80 or $0x80000000,%eax
493 * b: 0f 22 c0 mov %eax,%cr0
494 * set cr3 ->interception
495 * get cr0 ->interception
496 * set cr0 -> no interception
497 */
498 /* (1ULL << INTERCEPT_SELECTIVE_CR0) | */
499 (1ULL << INTERCEPT_CPUID) |
500 (1ULL << INTERCEPT_HLT) |
6aa8b732
AK
501 (1ULL << INTERCEPT_INVLPGA) |
502 (1ULL << INTERCEPT_IOIO_PROT) |
503 (1ULL << INTERCEPT_MSR_PROT) |
504 (1ULL << INTERCEPT_TASK_SWITCH) |
505 (1ULL << INTERCEPT_VMRUN) |
506 (1ULL << INTERCEPT_VMMCALL) |
507 (1ULL << INTERCEPT_VMLOAD) |
508 (1ULL << INTERCEPT_VMSAVE) |
509 (1ULL << INTERCEPT_STGI) |
510 (1ULL << INTERCEPT_CLGI) |
511 (1ULL << INTERCEPT_SKINIT);
512
513 control->iopm_base_pa = iopm_base;
514 control->msrpm_base_pa = msrpm_base;
515 rdtscll(tsc);
516 control->tsc_offset = -tsc;
517 control->int_ctl = V_INTR_MASKING_MASK;
518
519 init_seg(&save->es);
520 init_seg(&save->ss);
521 init_seg(&save->ds);
522 init_seg(&save->fs);
523 init_seg(&save->gs);
524
525 save->cs.selector = 0xf000;
526 /* Executable/Readable Code Segment */
527 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
528 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
529 save->cs.limit = 0xffff;
530 save->cs.base = 0xffff0000;
531
532 save->gdtr.limit = 0xffff;
533 save->idtr.limit = 0xffff;
534
535 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
536 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
537
538 save->efer = MSR_EFER_SVME_MASK;
539
540 save->dr6 = 0xffff0ff0;
541 save->dr7 = 0x400;
542 save->rflags = 2;
543 save->rip = 0x0000fff0;
544
545 /*
546 * cr0 val on cpu init should be 0x60000010, we enable cpu
547 * cache by default. the orderly way is to enable cache in bios.
548 */
549 save->cr0 = 0x00000010 | CR0_PG_MASK;
550 save->cr4 = CR4_PAE_MASK;
551 /* rdx = ?? */
552}
553
554static int svm_create_vcpu(struct kvm_vcpu *vcpu)
555{
556 struct page *page;
557 int r;
558
559 r = -ENOMEM;
560 vcpu->svm = kzalloc(sizeof *vcpu->svm, GFP_KERNEL);
561 if (!vcpu->svm)
562 goto out1;
563 page = alloc_page(GFP_KERNEL);
564 if (!page)
565 goto out2;
566
567 vcpu->svm->vmcb = page_address(page);
568 memset(vcpu->svm->vmcb, 0, PAGE_SIZE);
569 vcpu->svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
570 vcpu->svm->cr0 = 0x00000010;
571 vcpu->svm->asid_generation = 0;
572 memset(vcpu->svm->db_regs, 0, sizeof(vcpu->svm->db_regs));
573 init_vmcb(vcpu->svm->vmcb);
574
36241b8c
AK
575 fx_init(vcpu);
576
6aa8b732
AK
577 return 0;
578
579out2:
580 kfree(vcpu->svm);
581out1:
582 return r;
583}
584
585static void svm_free_vcpu(struct kvm_vcpu *vcpu)
586{
587 if (!vcpu->svm)
588 return;
589 if (vcpu->svm->vmcb)
590 __free_page(pfn_to_page(vcpu->svm->vmcb_pa >> PAGE_SHIFT));
591 kfree(vcpu->svm);
592}
593
594static struct kvm_vcpu *svm_vcpu_load(struct kvm_vcpu *vcpu)
595{
596 get_cpu();
597 return vcpu;
598}
599
600static void svm_vcpu_put(struct kvm_vcpu *vcpu)
601{
602 put_cpu();
603}
604
605static void svm_cache_regs(struct kvm_vcpu *vcpu)
606{
607 vcpu->regs[VCPU_REGS_RAX] = vcpu->svm->vmcb->save.rax;
608 vcpu->regs[VCPU_REGS_RSP] = vcpu->svm->vmcb->save.rsp;
609 vcpu->rip = vcpu->svm->vmcb->save.rip;
610}
611
612static void svm_decache_regs(struct kvm_vcpu *vcpu)
613{
614 vcpu->svm->vmcb->save.rax = vcpu->regs[VCPU_REGS_RAX];
615 vcpu->svm->vmcb->save.rsp = vcpu->regs[VCPU_REGS_RSP];
616 vcpu->svm->vmcb->save.rip = vcpu->rip;
617}
618
619static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
620{
621 return vcpu->svm->vmcb->save.rflags;
622}
623
624static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
625{
626 vcpu->svm->vmcb->save.rflags = rflags;
627}
628
629static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
630{
631 struct vmcb_save_area *save = &vcpu->svm->vmcb->save;
632
633 switch (seg) {
634 case VCPU_SREG_CS: return &save->cs;
635 case VCPU_SREG_DS: return &save->ds;
636 case VCPU_SREG_ES: return &save->es;
637 case VCPU_SREG_FS: return &save->fs;
638 case VCPU_SREG_GS: return &save->gs;
639 case VCPU_SREG_SS: return &save->ss;
640 case VCPU_SREG_TR: return &save->tr;
641 case VCPU_SREG_LDTR: return &save->ldtr;
642 }
643 BUG();
644 return 0;
645}
646
647static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
648{
649 struct vmcb_seg *s = svm_seg(vcpu, seg);
650
651 return s->base;
652}
653
654static void svm_get_segment(struct kvm_vcpu *vcpu,
655 struct kvm_segment *var, int seg)
656{
657 struct vmcb_seg *s = svm_seg(vcpu, seg);
658
659 var->base = s->base;
660 var->limit = s->limit;
661 var->selector = s->selector;
662 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
663 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
664 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
665 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
666 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
667 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
668 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
669 var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
670 var->unusable = !var->present;
671}
672
673static void svm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
674{
675 struct vmcb_seg *s = svm_seg(vcpu, VCPU_SREG_CS);
676
677 *db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
678 *l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
679}
680
681static void svm_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
682{
bce66ca4
LN
683 dt->limit = vcpu->svm->vmcb->save.idtr.limit;
684 dt->base = vcpu->svm->vmcb->save.idtr.base;
6aa8b732
AK
685}
686
687static void svm_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
688{
bce66ca4
LN
689 vcpu->svm->vmcb->save.idtr.limit = dt->limit;
690 vcpu->svm->vmcb->save.idtr.base = dt->base ;
6aa8b732
AK
691}
692
693static void svm_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
694{
695 dt->limit = vcpu->svm->vmcb->save.gdtr.limit;
696 dt->base = vcpu->svm->vmcb->save.gdtr.base;
697}
698
699static void svm_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
700{
701 vcpu->svm->vmcb->save.gdtr.limit = dt->limit;
702 vcpu->svm->vmcb->save.gdtr.base = dt->base ;
703}
704
399badf3
AK
705static void svm_decache_cr0_cr4_guest_bits(struct kvm_vcpu *vcpu)
706{
707}
708
6aa8b732
AK
709static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
710{
05b3e0c2 711#ifdef CONFIG_X86_64
6aa8b732
AK
712 if (vcpu->shadow_efer & KVM_EFER_LME) {
713 if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
714 vcpu->shadow_efer |= KVM_EFER_LMA;
715 vcpu->svm->vmcb->save.efer |= KVM_EFER_LMA | KVM_EFER_LME;
716 }
717
718 if (is_paging(vcpu) && !(cr0 & CR0_PG_MASK) ) {
719 vcpu->shadow_efer &= ~KVM_EFER_LMA;
720 vcpu->svm->vmcb->save.efer &= ~(KVM_EFER_LMA | KVM_EFER_LME);
721 }
722 }
723#endif
724 vcpu->svm->cr0 = cr0;
725 vcpu->svm->vmcb->save.cr0 = cr0 | CR0_PG_MASK;
726 vcpu->cr0 = cr0;
727}
728
729static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
730{
731 vcpu->cr4 = cr4;
732 vcpu->svm->vmcb->save.cr4 = cr4 | CR4_PAE_MASK;
733}
734
735static void svm_set_segment(struct kvm_vcpu *vcpu,
736 struct kvm_segment *var, int seg)
737{
738 struct vmcb_seg *s = svm_seg(vcpu, seg);
739
740 s->base = var->base;
741 s->limit = var->limit;
742 s->selector = var->selector;
743 if (var->unusable)
744 s->attrib = 0;
745 else {
746 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
747 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
748 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
749 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
750 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
751 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
752 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
753 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
754 }
755 if (seg == VCPU_SREG_CS)
756 vcpu->svm->vmcb->save.cpl
757 = (vcpu->svm->vmcb->save.cs.attrib
758 >> SVM_SELECTOR_DPL_SHIFT) & 3;
759
760}
761
762/* FIXME:
763
764 vcpu->svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
765 vcpu->svm->vmcb->control.int_ctl |= (sregs->cr8 & V_TPR_MASK);
766
767*/
768
769static int svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
770{
771 return -EOPNOTSUPP;
772}
773
774static void load_host_msrs(struct kvm_vcpu *vcpu)
775{
776 int i;
777
778 for ( i = 0; i < NR_HOST_SAVE_MSRS; i++)
779 wrmsrl(host_save_msrs[i], vcpu->svm->host_msrs[i]);
780}
781
782static void save_host_msrs(struct kvm_vcpu *vcpu)
783{
784 int i;
785
786 for ( i = 0; i < NR_HOST_SAVE_MSRS; i++)
787 rdmsrl(host_save_msrs[i], vcpu->svm->host_msrs[i]);
788}
789
790static void new_asid(struct kvm_vcpu *vcpu, struct svm_cpu_data *svm_data)
791{
792 if (svm_data->next_asid > svm_data->max_asid) {
793 ++svm_data->asid_generation;
794 svm_data->next_asid = 1;
795 vcpu->svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
796 }
797
798 vcpu->cpu = svm_data->cpu;
799 vcpu->svm->asid_generation = svm_data->asid_generation;
800 vcpu->svm->vmcb->control.asid = svm_data->next_asid++;
801}
802
803static void svm_invlpg(struct kvm_vcpu *vcpu, gva_t address)
804{
805 invlpga(address, vcpu->svm->vmcb->control.asid); // is needed?
806}
807
808static unsigned long svm_get_dr(struct kvm_vcpu *vcpu, int dr)
809{
810 return vcpu->svm->db_regs[dr];
811}
812
813static void svm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long value,
814 int *exception)
815{
816 *exception = 0;
817
818 if (vcpu->svm->vmcb->save.dr7 & DR7_GD_MASK) {
819 vcpu->svm->vmcb->save.dr7 &= ~DR7_GD_MASK;
820 vcpu->svm->vmcb->save.dr6 |= DR6_BD_MASK;
821 *exception = DB_VECTOR;
822 return;
823 }
824
825 switch (dr) {
826 case 0 ... 3:
827 vcpu->svm->db_regs[dr] = value;
828 return;
829 case 4 ... 5:
830 if (vcpu->cr4 & CR4_DE_MASK) {
831 *exception = UD_VECTOR;
832 return;
833 }
834 case 7: {
835 if (value & ~((1ULL << 32) - 1)) {
836 *exception = GP_VECTOR;
837 return;
838 }
839 vcpu->svm->vmcb->save.dr7 = value;
840 return;
841 }
842 default:
843 printk(KERN_DEBUG "%s: unexpected dr %u\n",
844 __FUNCTION__, dr);
845 *exception = UD_VECTOR;
846 return;
847 }
848}
849
850static int pf_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
851{
852 u32 exit_int_info = vcpu->svm->vmcb->control.exit_int_info;
853 u64 fault_address;
854 u32 error_code;
855 enum emulation_result er;
e2dec939 856 int r;
6aa8b732
AK
857
858 if (is_external_interrupt(exit_int_info))
859 push_irq(vcpu, exit_int_info & SVM_EVTINJ_VEC_MASK);
860
861 spin_lock(&vcpu->kvm->lock);
862
863 fault_address = vcpu->svm->vmcb->control.exit_info_2;
864 error_code = vcpu->svm->vmcb->control.exit_info_1;
e2dec939
AK
865 r = kvm_mmu_page_fault(vcpu, fault_address, error_code);
866 if (r < 0) {
867 spin_unlock(&vcpu->kvm->lock);
868 return r;
869 }
870 if (!r) {
6aa8b732
AK
871 spin_unlock(&vcpu->kvm->lock);
872 return 1;
873 }
874 er = emulate_instruction(vcpu, kvm_run, fault_address, error_code);
875 spin_unlock(&vcpu->kvm->lock);
876
877 switch (er) {
878 case EMULATE_DONE:
879 return 1;
880 case EMULATE_DO_MMIO:
881 ++kvm_stat.mmio_exits;
882 kvm_run->exit_reason = KVM_EXIT_MMIO;
883 return 0;
884 case EMULATE_FAIL:
885 vcpu_printf(vcpu, "%s: emulate fail\n", __FUNCTION__);
886 break;
887 default:
888 BUG();
889 }
890
891 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
892 return 0;
893}
894
895static int io_get_override(struct kvm_vcpu *vcpu,
896 struct vmcb_seg **seg,
897 int *addr_override)
898{
899 u8 inst[MAX_INST_SIZE];
900 unsigned ins_length;
901 gva_t rip;
902 int i;
903
904 rip = vcpu->svm->vmcb->save.rip;
905 ins_length = vcpu->svm->next_rip - rip;
906 rip += vcpu->svm->vmcb->save.cs.base;
907
908 if (ins_length > MAX_INST_SIZE)
909 printk(KERN_DEBUG
910 "%s: inst length err, cs base 0x%llx rip 0x%llx "
911 "next rip 0x%llx ins_length %u\n",
912 __FUNCTION__,
913 vcpu->svm->vmcb->save.cs.base,
914 vcpu->svm->vmcb->save.rip,
915 vcpu->svm->vmcb->control.exit_info_2,
916 ins_length);
917
918 if (kvm_read_guest(vcpu, rip, ins_length, inst) != ins_length)
919 /* #PF */
920 return 0;
921
922 *addr_override = 0;
923 *seg = 0;
924 for (i = 0; i < ins_length; i++)
925 switch (inst[i]) {
926 case 0xf0:
927 case 0xf2:
928 case 0xf3:
929 case 0x66:
930 continue;
931 case 0x67:
932 *addr_override = 1;
933 continue;
934 case 0x2e:
935 *seg = &vcpu->svm->vmcb->save.cs;
936 continue;
937 case 0x36:
938 *seg = &vcpu->svm->vmcb->save.ss;
939 continue;
940 case 0x3e:
941 *seg = &vcpu->svm->vmcb->save.ds;
942 continue;
943 case 0x26:
944 *seg = &vcpu->svm->vmcb->save.es;
945 continue;
946 case 0x64:
947 *seg = &vcpu->svm->vmcb->save.fs;
948 continue;
949 case 0x65:
950 *seg = &vcpu->svm->vmcb->save.gs;
951 continue;
952 default:
953 return 1;
954 }
955 printk(KERN_DEBUG "%s: unexpected\n", __FUNCTION__);
956 return 0;
957}
958
959static unsigned long io_adress(struct kvm_vcpu *vcpu, int ins, u64 *address)
960{
961 unsigned long addr_mask;
962 unsigned long *reg;
963 struct vmcb_seg *seg;
964 int addr_override;
965 struct vmcb_save_area *save_area = &vcpu->svm->vmcb->save;
966 u16 cs_attrib = save_area->cs.attrib;
967 unsigned addr_size = get_addr_size(vcpu);
968
969 if (!io_get_override(vcpu, &seg, &addr_override))
970 return 0;
971
972 if (addr_override)
973 addr_size = (addr_size == 2) ? 4: (addr_size >> 1);
974
975 if (ins) {
976 reg = &vcpu->regs[VCPU_REGS_RDI];
977 seg = &vcpu->svm->vmcb->save.es;
978 } else {
979 reg = &vcpu->regs[VCPU_REGS_RSI];
980 seg = (seg) ? seg : &vcpu->svm->vmcb->save.ds;
981 }
982
983 addr_mask = ~0ULL >> (64 - (addr_size * 8));
984
985 if ((cs_attrib & SVM_SELECTOR_L_MASK) &&
986 !(vcpu->svm->vmcb->save.rflags & X86_EFLAGS_VM)) {
987 *address = (*reg & addr_mask);
988 return addr_mask;
989 }
990
991 if (!(seg->attrib & SVM_SELECTOR_P_SHIFT)) {
992 svm_inject_gp(vcpu, 0);
993 return 0;
994 }
995
996 *address = (*reg & addr_mask) + seg->base;
997 return addr_mask;
998}
999
1000static int io_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1001{
1002 u32 io_info = vcpu->svm->vmcb->control.exit_info_1; //address size bug?
1003 int _in = io_info & SVM_IOIO_TYPE_MASK;
1004
1005 ++kvm_stat.io_exits;
1006
1007 vcpu->svm->next_rip = vcpu->svm->vmcb->control.exit_info_2;
1008
1009 kvm_run->exit_reason = KVM_EXIT_IO;
1010 kvm_run->io.port = io_info >> 16;
1011 kvm_run->io.direction = (_in) ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1012 kvm_run->io.size = ((io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT);
1013 kvm_run->io.string = (io_info & SVM_IOIO_STR_MASK) != 0;
1014 kvm_run->io.rep = (io_info & SVM_IOIO_REP_MASK) != 0;
1015
1016 if (kvm_run->io.string) {
1017 unsigned addr_mask;
1018
1019 addr_mask = io_adress(vcpu, _in, &kvm_run->io.address);
1020 if (!addr_mask) {
1021 printk(KERN_DEBUG "%s: get io address failed\n", __FUNCTION__);
1022 return 1;
1023 }
1024
1025 if (kvm_run->io.rep) {
1026 kvm_run->io.count = vcpu->regs[VCPU_REGS_RCX] & addr_mask;
1027 kvm_run->io.string_down = (vcpu->svm->vmcb->save.rflags
1028 & X86_EFLAGS_DF) != 0;
1029 }
1030 } else {
1031 kvm_run->io.value = vcpu->svm->vmcb->save.rax;
1032 }
1033 return 0;
1034}
1035
1036
1037static int nop_on_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1038{
1039 return 1;
1040}
1041
1042static int halt_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1043{
1044 vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 1;
1045 skip_emulated_instruction(vcpu);
c1150d8c 1046 if (vcpu->irq_summary)
6aa8b732
AK
1047 return 1;
1048
1049 kvm_run->exit_reason = KVM_EXIT_HLT;
c1150d8c 1050 ++kvm_stat.halt_exits;
6aa8b732
AK
1051 return 0;
1052}
1053
1054static int invalid_op_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1055{
1056 inject_ud(vcpu);
1057 return 1;
1058}
1059
1060static int task_switch_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1061{
1062 printk(KERN_DEBUG "%s: task swiche is unsupported\n", __FUNCTION__);
1063 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1064 return 0;
1065}
1066
1067static int cpuid_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1068{
1069 vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 2;
1070 kvm_run->exit_reason = KVM_EXIT_CPUID;
1071 return 0;
1072}
1073
1074static int emulate_on_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1075{
1076 if (emulate_instruction(vcpu, 0, 0, 0) != EMULATE_DONE)
1077 printk(KERN_ERR "%s: failed\n", __FUNCTION__);
1078 return 1;
1079}
1080
1081static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
1082{
1083 switch (ecx) {
6aa8b732
AK
1084 case MSR_IA32_TIME_STAMP_COUNTER: {
1085 u64 tsc;
1086
1087 rdtscll(tsc);
1088 *data = vcpu->svm->vmcb->control.tsc_offset + tsc;
1089 break;
1090 }
0e859cac 1091 case MSR_K6_STAR:
6aa8b732
AK
1092 *data = vcpu->svm->vmcb->save.star;
1093 break;
0e859cac 1094#ifdef CONFIG_X86_64
6aa8b732
AK
1095 case MSR_LSTAR:
1096 *data = vcpu->svm->vmcb->save.lstar;
1097 break;
1098 case MSR_CSTAR:
1099 *data = vcpu->svm->vmcb->save.cstar;
1100 break;
1101 case MSR_KERNEL_GS_BASE:
1102 *data = vcpu->svm->vmcb->save.kernel_gs_base;
1103 break;
1104 case MSR_SYSCALL_MASK:
1105 *data = vcpu->svm->vmcb->save.sfmask;
1106 break;
1107#endif
1108 case MSR_IA32_SYSENTER_CS:
1109 *data = vcpu->svm->vmcb->save.sysenter_cs;
1110 break;
1111 case MSR_IA32_SYSENTER_EIP:
1112 *data = vcpu->svm->vmcb->save.sysenter_eip;
1113 break;
1114 case MSR_IA32_SYSENTER_ESP:
1115 *data = vcpu->svm->vmcb->save.sysenter_esp;
1116 break;
1117 default:
3bab1f5d 1118 return kvm_get_msr_common(vcpu, ecx, data);
6aa8b732
AK
1119 }
1120 return 0;
1121}
1122
1123static int rdmsr_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1124{
1125 u32 ecx = vcpu->regs[VCPU_REGS_RCX];
1126 u64 data;
1127
1128 if (svm_get_msr(vcpu, ecx, &data))
1129 svm_inject_gp(vcpu, 0);
1130 else {
1131 vcpu->svm->vmcb->save.rax = data & 0xffffffff;
1132 vcpu->regs[VCPU_REGS_RDX] = data >> 32;
1133 vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 2;
1134 skip_emulated_instruction(vcpu);
1135 }
1136 return 1;
1137}
1138
1139static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
1140{
1141 switch (ecx) {
6aa8b732
AK
1142 case MSR_IA32_TIME_STAMP_COUNTER: {
1143 u64 tsc;
1144
1145 rdtscll(tsc);
1146 vcpu->svm->vmcb->control.tsc_offset = data - tsc;
1147 break;
1148 }
0e859cac 1149 case MSR_K6_STAR:
6aa8b732
AK
1150 vcpu->svm->vmcb->save.star = data;
1151 break;
0e859cac 1152#ifdef CONFIG_X86_64_
6aa8b732
AK
1153 case MSR_LSTAR:
1154 vcpu->svm->vmcb->save.lstar = data;
1155 break;
1156 case MSR_CSTAR:
1157 vcpu->svm->vmcb->save.cstar = data;
1158 break;
1159 case MSR_KERNEL_GS_BASE:
1160 vcpu->svm->vmcb->save.kernel_gs_base = data;
1161 break;
1162 case MSR_SYSCALL_MASK:
1163 vcpu->svm->vmcb->save.sfmask = data;
1164 break;
1165#endif
1166 case MSR_IA32_SYSENTER_CS:
1167 vcpu->svm->vmcb->save.sysenter_cs = data;
1168 break;
1169 case MSR_IA32_SYSENTER_EIP:
1170 vcpu->svm->vmcb->save.sysenter_eip = data;
1171 break;
1172 case MSR_IA32_SYSENTER_ESP:
1173 vcpu->svm->vmcb->save.sysenter_esp = data;
1174 break;
1175 default:
3bab1f5d 1176 return kvm_set_msr_common(vcpu, ecx, data);
6aa8b732
AK
1177 }
1178 return 0;
1179}
1180
1181static int wrmsr_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1182{
1183 u32 ecx = vcpu->regs[VCPU_REGS_RCX];
1184 u64 data = (vcpu->svm->vmcb->save.rax & -1u)
1185 | ((u64)(vcpu->regs[VCPU_REGS_RDX] & -1u) << 32);
1186 vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 2;
1187 if (svm_set_msr(vcpu, ecx, data))
1188 svm_inject_gp(vcpu, 0);
1189 else
1190 skip_emulated_instruction(vcpu);
1191 return 1;
1192}
1193
1194static int msr_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1195{
1196 if (vcpu->svm->vmcb->control.exit_info_1)
1197 return wrmsr_interception(vcpu, kvm_run);
1198 else
1199 return rdmsr_interception(vcpu, kvm_run);
1200}
1201
c1150d8c
DL
1202static int interrupt_window_interception(struct kvm_vcpu *vcpu,
1203 struct kvm_run *kvm_run)
1204{
1205 /*
1206 * If the user space waits to inject interrupts, exit as soon as
1207 * possible
1208 */
1209 if (kvm_run->request_interrupt_window &&
022a9308 1210 !vcpu->irq_summary) {
c1150d8c
DL
1211 ++kvm_stat.irq_window_exits;
1212 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
1213 return 0;
1214 }
1215
1216 return 1;
1217}
1218
6aa8b732
AK
1219static int (*svm_exit_handlers[])(struct kvm_vcpu *vcpu,
1220 struct kvm_run *kvm_run) = {
1221 [SVM_EXIT_READ_CR0] = emulate_on_interception,
1222 [SVM_EXIT_READ_CR3] = emulate_on_interception,
1223 [SVM_EXIT_READ_CR4] = emulate_on_interception,
1224 /* for now: */
1225 [SVM_EXIT_WRITE_CR0] = emulate_on_interception,
1226 [SVM_EXIT_WRITE_CR3] = emulate_on_interception,
1227 [SVM_EXIT_WRITE_CR4] = emulate_on_interception,
1228 [SVM_EXIT_READ_DR0] = emulate_on_interception,
1229 [SVM_EXIT_READ_DR1] = emulate_on_interception,
1230 [SVM_EXIT_READ_DR2] = emulate_on_interception,
1231 [SVM_EXIT_READ_DR3] = emulate_on_interception,
1232 [SVM_EXIT_WRITE_DR0] = emulate_on_interception,
1233 [SVM_EXIT_WRITE_DR1] = emulate_on_interception,
1234 [SVM_EXIT_WRITE_DR2] = emulate_on_interception,
1235 [SVM_EXIT_WRITE_DR3] = emulate_on_interception,
1236 [SVM_EXIT_WRITE_DR5] = emulate_on_interception,
1237 [SVM_EXIT_WRITE_DR7] = emulate_on_interception,
1238 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
1239 [SVM_EXIT_INTR] = nop_on_interception,
1240 [SVM_EXIT_NMI] = nop_on_interception,
1241 [SVM_EXIT_SMI] = nop_on_interception,
1242 [SVM_EXIT_INIT] = nop_on_interception,
c1150d8c 1243 [SVM_EXIT_VINTR] = interrupt_window_interception,
6aa8b732
AK
1244 /* [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception, */
1245 [SVM_EXIT_CPUID] = cpuid_interception,
1246 [SVM_EXIT_HLT] = halt_interception,
1247 [SVM_EXIT_INVLPG] = emulate_on_interception,
1248 [SVM_EXIT_INVLPGA] = invalid_op_interception,
1249 [SVM_EXIT_IOIO] = io_interception,
1250 [SVM_EXIT_MSR] = msr_interception,
1251 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
1252 [SVM_EXIT_VMRUN] = invalid_op_interception,
1253 [SVM_EXIT_VMMCALL] = invalid_op_interception,
1254 [SVM_EXIT_VMLOAD] = invalid_op_interception,
1255 [SVM_EXIT_VMSAVE] = invalid_op_interception,
1256 [SVM_EXIT_STGI] = invalid_op_interception,
1257 [SVM_EXIT_CLGI] = invalid_op_interception,
1258 [SVM_EXIT_SKINIT] = invalid_op_interception,
1259};
1260
1261
1262static int handle_exit(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1263{
1264 u32 exit_code = vcpu->svm->vmcb->control.exit_code;
1265
1266 kvm_run->exit_type = KVM_EXIT_TYPE_VM_EXIT;
1267
1268 if (is_external_interrupt(vcpu->svm->vmcb->control.exit_int_info) &&
1269 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR)
1270 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
1271 "exit_code 0x%x\n",
1272 __FUNCTION__, vcpu->svm->vmcb->control.exit_int_info,
1273 exit_code);
1274
1275 if (exit_code >= sizeof(svm_exit_handlers) / sizeof(*svm_exit_handlers)
1276 || svm_exit_handlers[exit_code] == 0) {
1277 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1278 printk(KERN_ERR "%s: 0x%x @ 0x%llx cr0 0x%lx rflags 0x%llx\n",
1279 __FUNCTION__,
1280 exit_code,
1281 vcpu->svm->vmcb->save.rip,
1282 vcpu->cr0,
1283 vcpu->svm->vmcb->save.rflags);
1284 return 0;
1285 }
1286
1287 return svm_exit_handlers[exit_code](vcpu, kvm_run);
1288}
1289
1290static void reload_tss(struct kvm_vcpu *vcpu)
1291{
1292 int cpu = raw_smp_processor_id();
1293
1294 struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1295 svm_data->tss_desc->type = 9; //available 32/64-bit TSS
1296 load_TR_desc();
1297}
1298
1299static void pre_svm_run(struct kvm_vcpu *vcpu)
1300{
1301 int cpu = raw_smp_processor_id();
1302
1303 struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1304
1305 vcpu->svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
1306 if (vcpu->cpu != cpu ||
1307 vcpu->svm->asid_generation != svm_data->asid_generation)
1308 new_asid(vcpu, svm_data);
1309}
1310
1311
c1150d8c 1312static inline void kvm_do_inject_irq(struct kvm_vcpu *vcpu)
6aa8b732
AK
1313{
1314 struct vmcb_control_area *control;
1315
6aa8b732 1316 control = &vcpu->svm->vmcb->control;
6aa8b732
AK
1317 control->int_vector = pop_irq(vcpu);
1318 control->int_ctl &= ~V_INTR_PRIO_MASK;
1319 control->int_ctl |= V_IRQ_MASK |
1320 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
1321}
1322
1323static void kvm_reput_irq(struct kvm_vcpu *vcpu)
1324{
1325 struct vmcb_control_area *control = &vcpu->svm->vmcb->control;
1326
1327 if (control->int_ctl & V_IRQ_MASK) {
1328 control->int_ctl &= ~V_IRQ_MASK;
1329 push_irq(vcpu, control->int_vector);
1330 }
c1150d8c
DL
1331
1332 vcpu->interrupt_window_open =
1333 !(control->int_state & SVM_INTERRUPT_SHADOW_MASK);
1334}
1335
1336static void do_interrupt_requests(struct kvm_vcpu *vcpu,
1337 struct kvm_run *kvm_run)
1338{
1339 struct vmcb_control_area *control = &vcpu->svm->vmcb->control;
1340
1341 vcpu->interrupt_window_open =
1342 (!(control->int_state & SVM_INTERRUPT_SHADOW_MASK) &&
1343 (vcpu->svm->vmcb->save.rflags & X86_EFLAGS_IF));
1344
1345 if (vcpu->interrupt_window_open && vcpu->irq_summary)
1346 /*
1347 * If interrupts enabled, and not blocked by sti or mov ss. Good.
1348 */
1349 kvm_do_inject_irq(vcpu);
1350
1351 /*
1352 * Interrupts blocked. Wait for unblock.
1353 */
1354 if (!vcpu->interrupt_window_open &&
1355 (vcpu->irq_summary || kvm_run->request_interrupt_window)) {
1356 control->intercept |= 1ULL << INTERCEPT_VINTR;
1357 } else
1358 control->intercept &= ~(1ULL << INTERCEPT_VINTR);
1359}
1360
1361static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1362 struct kvm_run *kvm_run)
1363{
1364 kvm_run->ready_for_interrupt_injection = (vcpu->interrupt_window_open &&
1365 vcpu->irq_summary == 0);
1366 kvm_run->if_flag = (vcpu->svm->vmcb->save.rflags & X86_EFLAGS_IF) != 0;
1367 kvm_run->cr8 = vcpu->cr8;
1368 kvm_run->apic_base = vcpu->apic_base;
1369}
1370
1371/*
1372 * Check if userspace requested an interrupt window, and that the
1373 * interrupt window is open.
1374 *
1375 * No need to exit to userspace if we already have an interrupt queued.
1376 */
1377static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1378 struct kvm_run *kvm_run)
1379{
1380 return (!vcpu->irq_summary &&
1381 kvm_run->request_interrupt_window &&
1382 vcpu->interrupt_window_open &&
1383 (vcpu->svm->vmcb->save.rflags & X86_EFLAGS_IF));
6aa8b732
AK
1384}
1385
1386static void save_db_regs(unsigned long *db_regs)
1387{
5aff458e
AK
1388 asm volatile ("mov %%dr0, %0" : "=r"(db_regs[0]));
1389 asm volatile ("mov %%dr1, %0" : "=r"(db_regs[1]));
1390 asm volatile ("mov %%dr2, %0" : "=r"(db_regs[2]));
1391 asm volatile ("mov %%dr3, %0" : "=r"(db_regs[3]));
6aa8b732
AK
1392}
1393
1394static void load_db_regs(unsigned long *db_regs)
1395{
5aff458e
AK
1396 asm volatile ("mov %0, %%dr0" : : "r"(db_regs[0]));
1397 asm volatile ("mov %0, %%dr1" : : "r"(db_regs[1]));
1398 asm volatile ("mov %0, %%dr2" : : "r"(db_regs[2]));
1399 asm volatile ("mov %0, %%dr3" : : "r"(db_regs[3]));
6aa8b732
AK
1400}
1401
1402static int svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1403{
1404 u16 fs_selector;
1405 u16 gs_selector;
1406 u16 ldt_selector;
e2dec939 1407 int r;
6aa8b732
AK
1408
1409again:
cccf748b
AK
1410 if (!vcpu->mmio_read_completed)
1411 do_interrupt_requests(vcpu, kvm_run);
6aa8b732
AK
1412
1413 clgi();
1414
1415 pre_svm_run(vcpu);
1416
1417 save_host_msrs(vcpu);
1418 fs_selector = read_fs();
1419 gs_selector = read_gs();
1420 ldt_selector = read_ldt();
1421 vcpu->svm->host_cr2 = kvm_read_cr2();
1422 vcpu->svm->host_dr6 = read_dr6();
1423 vcpu->svm->host_dr7 = read_dr7();
1424 vcpu->svm->vmcb->save.cr2 = vcpu->cr2;
1425
1426 if (vcpu->svm->vmcb->save.dr7 & 0xff) {
1427 write_dr7(0);
1428 save_db_regs(vcpu->svm->host_db_regs);
1429 load_db_regs(vcpu->svm->db_regs);
1430 }
36241b8c
AK
1431
1432 fx_save(vcpu->host_fx_image);
1433 fx_restore(vcpu->guest_fx_image);
1434
6aa8b732 1435 asm volatile (
05b3e0c2 1436#ifdef CONFIG_X86_64
6aa8b732
AK
1437 "push %%rbx; push %%rcx; push %%rdx;"
1438 "push %%rsi; push %%rdi; push %%rbp;"
1439 "push %%r8; push %%r9; push %%r10; push %%r11;"
1440 "push %%r12; push %%r13; push %%r14; push %%r15;"
1441#else
1442 "push %%ebx; push %%ecx; push %%edx;"
1443 "push %%esi; push %%edi; push %%ebp;"
1444#endif
1445
05b3e0c2 1446#ifdef CONFIG_X86_64
6aa8b732
AK
1447 "mov %c[rbx](%[vcpu]), %%rbx \n\t"
1448 "mov %c[rcx](%[vcpu]), %%rcx \n\t"
1449 "mov %c[rdx](%[vcpu]), %%rdx \n\t"
1450 "mov %c[rsi](%[vcpu]), %%rsi \n\t"
1451 "mov %c[rdi](%[vcpu]), %%rdi \n\t"
1452 "mov %c[rbp](%[vcpu]), %%rbp \n\t"
1453 "mov %c[r8](%[vcpu]), %%r8 \n\t"
1454 "mov %c[r9](%[vcpu]), %%r9 \n\t"
1455 "mov %c[r10](%[vcpu]), %%r10 \n\t"
1456 "mov %c[r11](%[vcpu]), %%r11 \n\t"
1457 "mov %c[r12](%[vcpu]), %%r12 \n\t"
1458 "mov %c[r13](%[vcpu]), %%r13 \n\t"
1459 "mov %c[r14](%[vcpu]), %%r14 \n\t"
1460 "mov %c[r15](%[vcpu]), %%r15 \n\t"
1461#else
1462 "mov %c[rbx](%[vcpu]), %%ebx \n\t"
1463 "mov %c[rcx](%[vcpu]), %%ecx \n\t"
1464 "mov %c[rdx](%[vcpu]), %%edx \n\t"
1465 "mov %c[rsi](%[vcpu]), %%esi \n\t"
1466 "mov %c[rdi](%[vcpu]), %%edi \n\t"
1467 "mov %c[rbp](%[vcpu]), %%ebp \n\t"
1468#endif
1469
05b3e0c2 1470#ifdef CONFIG_X86_64
6aa8b732
AK
1471 /* Enter guest mode */
1472 "push %%rax \n\t"
1473 "mov %c[svm](%[vcpu]), %%rax \n\t"
1474 "mov %c[vmcb](%%rax), %%rax \n\t"
1475 SVM_VMLOAD "\n\t"
1476 SVM_VMRUN "\n\t"
1477 SVM_VMSAVE "\n\t"
1478 "pop %%rax \n\t"
1479#else
1480 /* Enter guest mode */
1481 "push %%eax \n\t"
1482 "mov %c[svm](%[vcpu]), %%eax \n\t"
1483 "mov %c[vmcb](%%eax), %%eax \n\t"
1484 SVM_VMLOAD "\n\t"
1485 SVM_VMRUN "\n\t"
1486 SVM_VMSAVE "\n\t"
1487 "pop %%eax \n\t"
1488#endif
1489
1490 /* Save guest registers, load host registers */
05b3e0c2 1491#ifdef CONFIG_X86_64
6aa8b732
AK
1492 "mov %%rbx, %c[rbx](%[vcpu]) \n\t"
1493 "mov %%rcx, %c[rcx](%[vcpu]) \n\t"
1494 "mov %%rdx, %c[rdx](%[vcpu]) \n\t"
1495 "mov %%rsi, %c[rsi](%[vcpu]) \n\t"
1496 "mov %%rdi, %c[rdi](%[vcpu]) \n\t"
1497 "mov %%rbp, %c[rbp](%[vcpu]) \n\t"
1498 "mov %%r8, %c[r8](%[vcpu]) \n\t"
1499 "mov %%r9, %c[r9](%[vcpu]) \n\t"
1500 "mov %%r10, %c[r10](%[vcpu]) \n\t"
1501 "mov %%r11, %c[r11](%[vcpu]) \n\t"
1502 "mov %%r12, %c[r12](%[vcpu]) \n\t"
1503 "mov %%r13, %c[r13](%[vcpu]) \n\t"
1504 "mov %%r14, %c[r14](%[vcpu]) \n\t"
1505 "mov %%r15, %c[r15](%[vcpu]) \n\t"
1506
1507 "pop %%r15; pop %%r14; pop %%r13; pop %%r12;"
1508 "pop %%r11; pop %%r10; pop %%r9; pop %%r8;"
1509 "pop %%rbp; pop %%rdi; pop %%rsi;"
1510 "pop %%rdx; pop %%rcx; pop %%rbx; \n\t"
1511#else
1512 "mov %%ebx, %c[rbx](%[vcpu]) \n\t"
1513 "mov %%ecx, %c[rcx](%[vcpu]) \n\t"
1514 "mov %%edx, %c[rdx](%[vcpu]) \n\t"
1515 "mov %%esi, %c[rsi](%[vcpu]) \n\t"
1516 "mov %%edi, %c[rdi](%[vcpu]) \n\t"
1517 "mov %%ebp, %c[rbp](%[vcpu]) \n\t"
1518
1519 "pop %%ebp; pop %%edi; pop %%esi;"
1520 "pop %%edx; pop %%ecx; pop %%ebx; \n\t"
1521#endif
1522 :
1523 : [vcpu]"a"(vcpu),
1524 [svm]"i"(offsetof(struct kvm_vcpu, svm)),
1525 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
1526 [rbx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RBX])),
1527 [rcx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RCX])),
1528 [rdx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RDX])),
1529 [rsi]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RSI])),
1530 [rdi]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RDI])),
1531 [rbp]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RBP]))
05b3e0c2 1532#ifdef CONFIG_X86_64
6aa8b732
AK
1533 ,[r8 ]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R8 ])),
1534 [r9 ]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R9 ])),
1535 [r10]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R10])),
1536 [r11]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R11])),
1537 [r12]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R12])),
1538 [r13]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R13])),
1539 [r14]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R14])),
1540 [r15]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R15]))
1541#endif
1542 : "cc", "memory" );
1543
36241b8c
AK
1544 fx_save(vcpu->guest_fx_image);
1545 fx_restore(vcpu->host_fx_image);
1546
6aa8b732
AK
1547 if ((vcpu->svm->vmcb->save.dr7 & 0xff))
1548 load_db_regs(vcpu->svm->host_db_regs);
1549
1550 vcpu->cr2 = vcpu->svm->vmcb->save.cr2;
1551
1552 write_dr6(vcpu->svm->host_dr6);
1553 write_dr7(vcpu->svm->host_dr7);
1554 kvm_write_cr2(vcpu->svm->host_cr2);
1555
1556 load_fs(fs_selector);
1557 load_gs(gs_selector);
1558 load_ldt(ldt_selector);
1559 load_host_msrs(vcpu);
1560
1561 reload_tss(vcpu);
1562
07031e14
IM
1563 /*
1564 * Profile KVM exit RIPs:
1565 */
1566 if (unlikely(prof_on == KVM_PROFILING))
1567 profile_hit(KVM_PROFILING,
1568 (void *)(unsigned long)vcpu->svm->vmcb->save.rip);
1569
6aa8b732
AK
1570 stgi();
1571
1572 kvm_reput_irq(vcpu);
1573
1574 vcpu->svm->next_rip = 0;
1575
1576 if (vcpu->svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
1577 kvm_run->exit_type = KVM_EXIT_TYPE_FAIL_ENTRY;
1578 kvm_run->exit_reason = vcpu->svm->vmcb->control.exit_code;
c1150d8c 1579 post_kvm_run_save(vcpu, kvm_run);
6aa8b732
AK
1580 return 0;
1581 }
1582
e2dec939
AK
1583 r = handle_exit(vcpu, kvm_run);
1584 if (r > 0) {
6aa8b732
AK
1585 if (signal_pending(current)) {
1586 ++kvm_stat.signal_exits;
c1150d8c
DL
1587 post_kvm_run_save(vcpu, kvm_run);
1588 return -EINTR;
1589 }
1590
1591 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
1592 ++kvm_stat.request_irq_exits;
1593 post_kvm_run_save(vcpu, kvm_run);
6aa8b732
AK
1594 return -EINTR;
1595 }
1596 kvm_resched(vcpu);
1597 goto again;
1598 }
c1150d8c 1599 post_kvm_run_save(vcpu, kvm_run);
e2dec939 1600 return r;
6aa8b732
AK
1601}
1602
1603static void svm_flush_tlb(struct kvm_vcpu *vcpu)
1604{
1605 force_new_asid(vcpu);
1606}
1607
1608static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
1609{
1610 vcpu->svm->vmcb->save.cr3 = root;
1611 force_new_asid(vcpu);
1612}
1613
1614static void svm_inject_page_fault(struct kvm_vcpu *vcpu,
1615 unsigned long addr,
1616 uint32_t err_code)
1617{
1618 uint32_t exit_int_info = vcpu->svm->vmcb->control.exit_int_info;
1619
1620 ++kvm_stat.pf_guest;
1621
1622 if (is_page_fault(exit_int_info)) {
1623
1624 vcpu->svm->vmcb->control.event_inj_err = 0;
1625 vcpu->svm->vmcb->control.event_inj = SVM_EVTINJ_VALID |
1626 SVM_EVTINJ_VALID_ERR |
1627 SVM_EVTINJ_TYPE_EXEPT |
1628 DF_VECTOR;
1629 return;
1630 }
1631 vcpu->cr2 = addr;
1632 vcpu->svm->vmcb->save.cr2 = addr;
1633 vcpu->svm->vmcb->control.event_inj = SVM_EVTINJ_VALID |
1634 SVM_EVTINJ_VALID_ERR |
1635 SVM_EVTINJ_TYPE_EXEPT |
1636 PF_VECTOR;
1637 vcpu->svm->vmcb->control.event_inj_err = err_code;
1638}
1639
1640
1641static int is_disabled(void)
1642{
1643 return 0;
1644}
1645
1646static struct kvm_arch_ops svm_arch_ops = {
1647 .cpu_has_kvm_support = has_svm,
1648 .disabled_by_bios = is_disabled,
1649 .hardware_setup = svm_hardware_setup,
1650 .hardware_unsetup = svm_hardware_unsetup,
1651 .hardware_enable = svm_hardware_enable,
1652 .hardware_disable = svm_hardware_disable,
1653
1654 .vcpu_create = svm_create_vcpu,
1655 .vcpu_free = svm_free_vcpu,
1656
1657 .vcpu_load = svm_vcpu_load,
1658 .vcpu_put = svm_vcpu_put,
1659
1660 .set_guest_debug = svm_guest_debug,
1661 .get_msr = svm_get_msr,
1662 .set_msr = svm_set_msr,
1663 .get_segment_base = svm_get_segment_base,
1664 .get_segment = svm_get_segment,
1665 .set_segment = svm_set_segment,
6aa8b732 1666 .get_cs_db_l_bits = svm_get_cs_db_l_bits,
399badf3 1667 .decache_cr0_cr4_guest_bits = svm_decache_cr0_cr4_guest_bits,
6aa8b732
AK
1668 .set_cr0 = svm_set_cr0,
1669 .set_cr0_no_modeswitch = svm_set_cr0,
1670 .set_cr3 = svm_set_cr3,
1671 .set_cr4 = svm_set_cr4,
1672 .set_efer = svm_set_efer,
1673 .get_idt = svm_get_idt,
1674 .set_idt = svm_set_idt,
1675 .get_gdt = svm_get_gdt,
1676 .set_gdt = svm_set_gdt,
1677 .get_dr = svm_get_dr,
1678 .set_dr = svm_set_dr,
1679 .cache_regs = svm_cache_regs,
1680 .decache_regs = svm_decache_regs,
1681 .get_rflags = svm_get_rflags,
1682 .set_rflags = svm_set_rflags,
1683
1684 .invlpg = svm_invlpg,
1685 .tlb_flush = svm_flush_tlb,
1686 .inject_page_fault = svm_inject_page_fault,
1687
1688 .inject_gp = svm_inject_gp,
1689
1690 .run = svm_vcpu_run,
1691 .skip_emulated_instruction = skip_emulated_instruction,
1692 .vcpu_setup = svm_vcpu_setup,
1693};
1694
1695static int __init svm_init(void)
1696{
873a7c42 1697 return kvm_init_arch(&svm_arch_ops, THIS_MODULE);
6aa8b732
AK
1698}
1699
1700static void __exit svm_exit(void)
1701{
1702 kvm_exit_arch();
1703}
1704
1705module_init(svm_init)
1706module_exit(svm_exit)