Detach sched.h from mm.h
[linux-2.6/kvm.git] / drivers / kvm / svm.c
blobfa17d6d4f0cb64f0052e41194ee15cbb775cd070
1 /*
2 * Kernel-based Virtual Machine driver for Linux
4 * AMD SVM support
6 * Copyright (C) 2006 Qumranet, Inc.
8 * Authors:
9 * Yaniv Kamay <yaniv@qumranet.com>
10 * Avi Kivity <avi@qumranet.com>
12 * This work is licensed under the terms of the GNU GPL, version 2. See
13 * the COPYING file in the top-level directory.
17 #include <linux/module.h>
18 #include <linux/kernel.h>
19 #include <linux/vmalloc.h>
20 #include <linux/highmem.h>
21 #include <linux/profile.h>
22 #include <linux/sched.h>
23 #include <asm/desc.h>
25 #include "kvm_svm.h"
26 #include "x86_emulate.h"
28 MODULE_AUTHOR("Qumranet");
29 MODULE_LICENSE("GPL");
31 #define IOPM_ALLOC_ORDER 2
32 #define MSRPM_ALLOC_ORDER 1
34 #define DB_VECTOR 1
35 #define UD_VECTOR 6
36 #define GP_VECTOR 13
38 #define DR7_GD_MASK (1 << 13)
39 #define DR6_BD_MASK (1 << 13)
40 #define CR4_DE_MASK (1UL << 3)
42 #define SEG_TYPE_LDT 2
43 #define SEG_TYPE_BUSY_TSS16 3
45 #define KVM_EFER_LMA (1 << 10)
46 #define KVM_EFER_LME (1 << 8)
48 #define SVM_FEATURE_NPT (1 << 0)
49 #define SVM_FEATURE_LBRV (1 << 1)
50 #define SVM_DEATURE_SVML (1 << 2)
52 unsigned long iopm_base;
53 unsigned long msrpm_base;
55 struct kvm_ldttss_desc {
56 u16 limit0;
57 u16 base0;
58 unsigned base1 : 8, type : 5, dpl : 2, p : 1;
59 unsigned limit1 : 4, zero0 : 3, g : 1, base2 : 8;
60 u32 base3;
61 u32 zero1;
62 } __attribute__((packed));
64 struct svm_cpu_data {
65 int cpu;
67 u64 asid_generation;
68 u32 max_asid;
69 u32 next_asid;
70 struct kvm_ldttss_desc *tss_desc;
72 struct page *save_area;
75 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
76 static uint32_t svm_features;
78 struct svm_init_data {
79 int cpu;
80 int r;
83 static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
85 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
86 #define MSRS_RANGE_SIZE 2048
87 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
89 #define MAX_INST_SIZE 15
91 static inline u32 svm_has(u32 feat)
93 return svm_features & feat;
96 static unsigned get_addr_size(struct kvm_vcpu *vcpu)
98 struct vmcb_save_area *sa = &vcpu->svm->vmcb->save;
99 u16 cs_attrib;
101 if (!(sa->cr0 & CR0_PE_MASK) || (sa->rflags & X86_EFLAGS_VM))
102 return 2;
104 cs_attrib = sa->cs.attrib;
106 return (cs_attrib & SVM_SELECTOR_L_MASK) ? 8 :
107 (cs_attrib & SVM_SELECTOR_DB_MASK) ? 4 : 2;
110 static inline u8 pop_irq(struct kvm_vcpu *vcpu)
112 int word_index = __ffs(vcpu->irq_summary);
113 int bit_index = __ffs(vcpu->irq_pending[word_index]);
114 int irq = word_index * BITS_PER_LONG + bit_index;
116 clear_bit(bit_index, &vcpu->irq_pending[word_index]);
117 if (!vcpu->irq_pending[word_index])
118 clear_bit(word_index, &vcpu->irq_summary);
119 return irq;
122 static inline void push_irq(struct kvm_vcpu *vcpu, u8 irq)
124 set_bit(irq, vcpu->irq_pending);
125 set_bit(irq / BITS_PER_LONG, &vcpu->irq_summary);
128 static inline void clgi(void)
130 asm volatile (SVM_CLGI);
133 static inline void stgi(void)
135 asm volatile (SVM_STGI);
138 static inline void invlpga(unsigned long addr, u32 asid)
140 asm volatile (SVM_INVLPGA :: "a"(addr), "c"(asid));
143 static inline unsigned long kvm_read_cr2(void)
145 unsigned long cr2;
147 asm volatile ("mov %%cr2, %0" : "=r" (cr2));
148 return cr2;
151 static inline void kvm_write_cr2(unsigned long val)
153 asm volatile ("mov %0, %%cr2" :: "r" (val));
156 static inline unsigned long read_dr6(void)
158 unsigned long dr6;
160 asm volatile ("mov %%dr6, %0" : "=r" (dr6));
161 return dr6;
164 static inline void write_dr6(unsigned long val)
166 asm volatile ("mov %0, %%dr6" :: "r" (val));
169 static inline unsigned long read_dr7(void)
171 unsigned long dr7;
173 asm volatile ("mov %%dr7, %0" : "=r" (dr7));
174 return dr7;
177 static inline void write_dr7(unsigned long val)
179 asm volatile ("mov %0, %%dr7" :: "r" (val));
182 static inline void force_new_asid(struct kvm_vcpu *vcpu)
184 vcpu->svm->asid_generation--;
187 static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
189 force_new_asid(vcpu);
192 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
194 if (!(efer & KVM_EFER_LMA))
195 efer &= ~KVM_EFER_LME;
197 vcpu->svm->vmcb->save.efer = efer | MSR_EFER_SVME_MASK;
198 vcpu->shadow_efer = efer;
201 static void svm_inject_gp(struct kvm_vcpu *vcpu, unsigned error_code)
203 vcpu->svm->vmcb->control.event_inj = SVM_EVTINJ_VALID |
204 SVM_EVTINJ_VALID_ERR |
205 SVM_EVTINJ_TYPE_EXEPT |
206 GP_VECTOR;
207 vcpu->svm->vmcb->control.event_inj_err = error_code;
210 static void inject_ud(struct kvm_vcpu *vcpu)
212 vcpu->svm->vmcb->control.event_inj = SVM_EVTINJ_VALID |
213 SVM_EVTINJ_TYPE_EXEPT |
214 UD_VECTOR;
217 static int is_page_fault(uint32_t info)
219 info &= SVM_EVTINJ_VEC_MASK | SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
220 return info == (PF_VECTOR | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT);
223 static int is_external_interrupt(u32 info)
225 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
226 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
229 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
231 if (!vcpu->svm->next_rip) {
232 printk(KERN_DEBUG "%s: NOP\n", __FUNCTION__);
233 return;
235 if (vcpu->svm->next_rip - vcpu->svm->vmcb->save.rip > 15) {
236 printk(KERN_ERR "%s: ip 0x%llx next 0x%llx\n",
237 __FUNCTION__,
238 vcpu->svm->vmcb->save.rip,
239 vcpu->svm->next_rip);
242 vcpu->rip = vcpu->svm->vmcb->save.rip = vcpu->svm->next_rip;
243 vcpu->svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
245 vcpu->interrupt_window_open = 1;
248 static int has_svm(void)
250 uint32_t eax, ebx, ecx, edx;
252 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD) {
253 printk(KERN_INFO "has_svm: not amd\n");
254 return 0;
257 cpuid(0x80000000, &eax, &ebx, &ecx, &edx);
258 if (eax < SVM_CPUID_FUNC) {
259 printk(KERN_INFO "has_svm: can't execute cpuid_8000000a\n");
260 return 0;
263 cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
264 if (!(ecx & (1 << SVM_CPUID_FEATURE_SHIFT))) {
265 printk(KERN_DEBUG "has_svm: svm not available\n");
266 return 0;
268 return 1;
271 static void svm_hardware_disable(void *garbage)
273 struct svm_cpu_data *svm_data
274 = per_cpu(svm_data, raw_smp_processor_id());
276 if (svm_data) {
277 uint64_t efer;
279 wrmsrl(MSR_VM_HSAVE_PA, 0);
280 rdmsrl(MSR_EFER, efer);
281 wrmsrl(MSR_EFER, efer & ~MSR_EFER_SVME_MASK);
282 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
283 __free_page(svm_data->save_area);
284 kfree(svm_data);
288 static void svm_hardware_enable(void *garbage)
291 struct svm_cpu_data *svm_data;
292 uint64_t efer;
293 #ifdef CONFIG_X86_64
294 struct desc_ptr gdt_descr;
295 #else
296 struct Xgt_desc_struct gdt_descr;
297 #endif
298 struct desc_struct *gdt;
299 int me = raw_smp_processor_id();
301 if (!has_svm()) {
302 printk(KERN_ERR "svm_cpu_init: err EOPNOTSUPP on %d\n", me);
303 return;
305 svm_data = per_cpu(svm_data, me);
307 if (!svm_data) {
308 printk(KERN_ERR "svm_cpu_init: svm_data is NULL on %d\n",
309 me);
310 return;
313 svm_data->asid_generation = 1;
314 svm_data->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
315 svm_data->next_asid = svm_data->max_asid + 1;
316 svm_features = cpuid_edx(SVM_CPUID_FUNC);
318 asm volatile ( "sgdt %0" : "=m"(gdt_descr) );
319 gdt = (struct desc_struct *)gdt_descr.address;
320 svm_data->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
322 rdmsrl(MSR_EFER, efer);
323 wrmsrl(MSR_EFER, efer | MSR_EFER_SVME_MASK);
325 wrmsrl(MSR_VM_HSAVE_PA,
326 page_to_pfn(svm_data->save_area) << PAGE_SHIFT);
329 static int svm_cpu_init(int cpu)
331 struct svm_cpu_data *svm_data;
332 int r;
334 svm_data = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
335 if (!svm_data)
336 return -ENOMEM;
337 svm_data->cpu = cpu;
338 svm_data->save_area = alloc_page(GFP_KERNEL);
339 r = -ENOMEM;
340 if (!svm_data->save_area)
341 goto err_1;
343 per_cpu(svm_data, cpu) = svm_data;
345 return 0;
347 err_1:
348 kfree(svm_data);
349 return r;
353 static int set_msr_interception(u32 *msrpm, unsigned msr,
354 int read, int write)
356 int i;
358 for (i = 0; i < NUM_MSR_MAPS; i++) {
359 if (msr >= msrpm_ranges[i] &&
360 msr < msrpm_ranges[i] + MSRS_IN_RANGE) {
361 u32 msr_offset = (i * MSRS_IN_RANGE + msr -
362 msrpm_ranges[i]) * 2;
364 u32 *base = msrpm + (msr_offset / 32);
365 u32 msr_shift = msr_offset % 32;
366 u32 mask = ((write) ? 0 : 2) | ((read) ? 0 : 1);
367 *base = (*base & ~(0x3 << msr_shift)) |
368 (mask << msr_shift);
369 return 1;
372 printk(KERN_DEBUG "%s: not found 0x%x\n", __FUNCTION__, msr);
373 return 0;
376 static __init int svm_hardware_setup(void)
378 int cpu;
379 struct page *iopm_pages;
380 struct page *msrpm_pages;
381 void *msrpm_va;
382 int r;
384 kvm_emulator_want_group7_invlpg();
386 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
388 if (!iopm_pages)
389 return -ENOMEM;
390 memset(page_address(iopm_pages), 0xff,
391 PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
392 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
395 msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
397 r = -ENOMEM;
398 if (!msrpm_pages)
399 goto err_1;
401 msrpm_va = page_address(msrpm_pages);
402 memset(msrpm_va, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
403 msrpm_base = page_to_pfn(msrpm_pages) << PAGE_SHIFT;
405 #ifdef CONFIG_X86_64
406 set_msr_interception(msrpm_va, MSR_GS_BASE, 1, 1);
407 set_msr_interception(msrpm_va, MSR_FS_BASE, 1, 1);
408 set_msr_interception(msrpm_va, MSR_KERNEL_GS_BASE, 1, 1);
409 set_msr_interception(msrpm_va, MSR_LSTAR, 1, 1);
410 set_msr_interception(msrpm_va, MSR_CSTAR, 1, 1);
411 set_msr_interception(msrpm_va, MSR_SYSCALL_MASK, 1, 1);
412 #endif
413 set_msr_interception(msrpm_va, MSR_K6_STAR, 1, 1);
414 set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_CS, 1, 1);
415 set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_ESP, 1, 1);
416 set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_EIP, 1, 1);
418 for_each_online_cpu(cpu) {
419 r = svm_cpu_init(cpu);
420 if (r)
421 goto err_2;
423 return 0;
425 err_2:
426 __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
427 msrpm_base = 0;
428 err_1:
429 __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
430 iopm_base = 0;
431 return r;
434 static __exit void svm_hardware_unsetup(void)
436 __free_pages(pfn_to_page(msrpm_base >> PAGE_SHIFT), MSRPM_ALLOC_ORDER);
437 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
438 iopm_base = msrpm_base = 0;
441 static void init_seg(struct vmcb_seg *seg)
443 seg->selector = 0;
444 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
445 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
446 seg->limit = 0xffff;
447 seg->base = 0;
450 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
452 seg->selector = 0;
453 seg->attrib = SVM_SELECTOR_P_MASK | type;
454 seg->limit = 0xffff;
455 seg->base = 0;
458 static int svm_vcpu_setup(struct kvm_vcpu *vcpu)
460 return 0;
463 static void init_vmcb(struct vmcb *vmcb)
465 struct vmcb_control_area *control = &vmcb->control;
466 struct vmcb_save_area *save = &vmcb->save;
468 control->intercept_cr_read = INTERCEPT_CR0_MASK |
469 INTERCEPT_CR3_MASK |
470 INTERCEPT_CR4_MASK;
472 control->intercept_cr_write = INTERCEPT_CR0_MASK |
473 INTERCEPT_CR3_MASK |
474 INTERCEPT_CR4_MASK;
476 control->intercept_dr_read = INTERCEPT_DR0_MASK |
477 INTERCEPT_DR1_MASK |
478 INTERCEPT_DR2_MASK |
479 INTERCEPT_DR3_MASK;
481 control->intercept_dr_write = INTERCEPT_DR0_MASK |
482 INTERCEPT_DR1_MASK |
483 INTERCEPT_DR2_MASK |
484 INTERCEPT_DR3_MASK |
485 INTERCEPT_DR5_MASK |
486 INTERCEPT_DR7_MASK;
488 control->intercept_exceptions = 1 << PF_VECTOR;
491 control->intercept = (1ULL << INTERCEPT_INTR) |
492 (1ULL << INTERCEPT_NMI) |
493 (1ULL << INTERCEPT_SMI) |
495 * selective cr0 intercept bug?
496 * 0: 0f 22 d8 mov %eax,%cr3
497 * 3: 0f 20 c0 mov %cr0,%eax
498 * 6: 0d 00 00 00 80 or $0x80000000,%eax
499 * b: 0f 22 c0 mov %eax,%cr0
500 * set cr3 ->interception
501 * get cr0 ->interception
502 * set cr0 -> no interception
504 /* (1ULL << INTERCEPT_SELECTIVE_CR0) | */
505 (1ULL << INTERCEPT_CPUID) |
506 (1ULL << INTERCEPT_HLT) |
507 (1ULL << INTERCEPT_INVLPGA) |
508 (1ULL << INTERCEPT_IOIO_PROT) |
509 (1ULL << INTERCEPT_MSR_PROT) |
510 (1ULL << INTERCEPT_TASK_SWITCH) |
511 (1ULL << INTERCEPT_SHUTDOWN) |
512 (1ULL << INTERCEPT_VMRUN) |
513 (1ULL << INTERCEPT_VMMCALL) |
514 (1ULL << INTERCEPT_VMLOAD) |
515 (1ULL << INTERCEPT_VMSAVE) |
516 (1ULL << INTERCEPT_STGI) |
517 (1ULL << INTERCEPT_CLGI) |
518 (1ULL << INTERCEPT_SKINIT) |
519 (1ULL << INTERCEPT_MONITOR) |
520 (1ULL << INTERCEPT_MWAIT);
522 control->iopm_base_pa = iopm_base;
523 control->msrpm_base_pa = msrpm_base;
524 control->tsc_offset = 0;
525 control->int_ctl = V_INTR_MASKING_MASK;
527 init_seg(&save->es);
528 init_seg(&save->ss);
529 init_seg(&save->ds);
530 init_seg(&save->fs);
531 init_seg(&save->gs);
533 save->cs.selector = 0xf000;
534 /* Executable/Readable Code Segment */
535 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
536 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
537 save->cs.limit = 0xffff;
539 * cs.base should really be 0xffff0000, but vmx can't handle that, so
540 * be consistent with it.
542 * Replace when we have real mode working for vmx.
544 save->cs.base = 0xf0000;
546 save->gdtr.limit = 0xffff;
547 save->idtr.limit = 0xffff;
549 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
550 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
552 save->efer = MSR_EFER_SVME_MASK;
554 save->dr6 = 0xffff0ff0;
555 save->dr7 = 0x400;
556 save->rflags = 2;
557 save->rip = 0x0000fff0;
560 * cr0 val on cpu init should be 0x60000010, we enable cpu
561 * cache by default. the orderly way is to enable cache in bios.
563 save->cr0 = 0x00000010 | CR0_PG_MASK | CR0_WP_MASK;
564 save->cr4 = CR4_PAE_MASK;
565 /* rdx = ?? */
568 static int svm_create_vcpu(struct kvm_vcpu *vcpu)
570 struct page *page;
571 int r;
573 r = -ENOMEM;
574 vcpu->svm = kzalloc(sizeof *vcpu->svm, GFP_KERNEL);
575 if (!vcpu->svm)
576 goto out1;
577 page = alloc_page(GFP_KERNEL);
578 if (!page)
579 goto out2;
581 vcpu->svm->vmcb = page_address(page);
582 memset(vcpu->svm->vmcb, 0, PAGE_SIZE);
583 vcpu->svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
584 vcpu->svm->asid_generation = 0;
585 memset(vcpu->svm->db_regs, 0, sizeof(vcpu->svm->db_regs));
586 init_vmcb(vcpu->svm->vmcb);
588 fx_init(vcpu);
589 vcpu->fpu_active = 1;
590 vcpu->apic_base = 0xfee00000 |
591 /*for vcpu 0*/ MSR_IA32_APICBASE_BSP |
592 MSR_IA32_APICBASE_ENABLE;
594 return 0;
596 out2:
597 kfree(vcpu->svm);
598 out1:
599 return r;
602 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
604 if (!vcpu->svm)
605 return;
606 if (vcpu->svm->vmcb)
607 __free_page(pfn_to_page(vcpu->svm->vmcb_pa >> PAGE_SHIFT));
608 kfree(vcpu->svm);
611 static void svm_vcpu_load(struct kvm_vcpu *vcpu)
613 int cpu, i;
615 cpu = get_cpu();
616 if (unlikely(cpu != vcpu->cpu)) {
617 u64 tsc_this, delta;
620 * Make sure that the guest sees a monotonically
621 * increasing TSC.
623 rdtscll(tsc_this);
624 delta = vcpu->host_tsc - tsc_this;
625 vcpu->svm->vmcb->control.tsc_offset += delta;
626 vcpu->cpu = cpu;
629 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
630 rdmsrl(host_save_user_msrs[i], vcpu->svm->host_user_msrs[i]);
633 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
635 int i;
637 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
638 wrmsrl(host_save_user_msrs[i], vcpu->svm->host_user_msrs[i]);
640 rdtscll(vcpu->host_tsc);
641 put_cpu();
644 static void svm_vcpu_decache(struct kvm_vcpu *vcpu)
648 static void svm_cache_regs(struct kvm_vcpu *vcpu)
650 vcpu->regs[VCPU_REGS_RAX] = vcpu->svm->vmcb->save.rax;
651 vcpu->regs[VCPU_REGS_RSP] = vcpu->svm->vmcb->save.rsp;
652 vcpu->rip = vcpu->svm->vmcb->save.rip;
655 static void svm_decache_regs(struct kvm_vcpu *vcpu)
657 vcpu->svm->vmcb->save.rax = vcpu->regs[VCPU_REGS_RAX];
658 vcpu->svm->vmcb->save.rsp = vcpu->regs[VCPU_REGS_RSP];
659 vcpu->svm->vmcb->save.rip = vcpu->rip;
662 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
664 return vcpu->svm->vmcb->save.rflags;
667 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
669 vcpu->svm->vmcb->save.rflags = rflags;
672 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
674 struct vmcb_save_area *save = &vcpu->svm->vmcb->save;
676 switch (seg) {
677 case VCPU_SREG_CS: return &save->cs;
678 case VCPU_SREG_DS: return &save->ds;
679 case VCPU_SREG_ES: return &save->es;
680 case VCPU_SREG_FS: return &save->fs;
681 case VCPU_SREG_GS: return &save->gs;
682 case VCPU_SREG_SS: return &save->ss;
683 case VCPU_SREG_TR: return &save->tr;
684 case VCPU_SREG_LDTR: return &save->ldtr;
686 BUG();
687 return NULL;
690 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
692 struct vmcb_seg *s = svm_seg(vcpu, seg);
694 return s->base;
697 static void svm_get_segment(struct kvm_vcpu *vcpu,
698 struct kvm_segment *var, int seg)
700 struct vmcb_seg *s = svm_seg(vcpu, seg);
702 var->base = s->base;
703 var->limit = s->limit;
704 var->selector = s->selector;
705 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
706 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
707 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
708 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
709 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
710 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
711 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
712 var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
713 var->unusable = !var->present;
716 static void svm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
718 struct vmcb_seg *s = svm_seg(vcpu, VCPU_SREG_CS);
720 *db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
721 *l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
724 static void svm_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
726 dt->limit = vcpu->svm->vmcb->save.idtr.limit;
727 dt->base = vcpu->svm->vmcb->save.idtr.base;
730 static void svm_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
732 vcpu->svm->vmcb->save.idtr.limit = dt->limit;
733 vcpu->svm->vmcb->save.idtr.base = dt->base ;
736 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
738 dt->limit = vcpu->svm->vmcb->save.gdtr.limit;
739 dt->base = vcpu->svm->vmcb->save.gdtr.base;
742 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
744 vcpu->svm->vmcb->save.gdtr.limit = dt->limit;
745 vcpu->svm->vmcb->save.gdtr.base = dt->base ;
748 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
752 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
754 #ifdef CONFIG_X86_64
755 if (vcpu->shadow_efer & KVM_EFER_LME) {
756 if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
757 vcpu->shadow_efer |= KVM_EFER_LMA;
758 vcpu->svm->vmcb->save.efer |= KVM_EFER_LMA | KVM_EFER_LME;
761 if (is_paging(vcpu) && !(cr0 & CR0_PG_MASK) ) {
762 vcpu->shadow_efer &= ~KVM_EFER_LMA;
763 vcpu->svm->vmcb->save.efer &= ~(KVM_EFER_LMA | KVM_EFER_LME);
766 #endif
767 if ((vcpu->cr0 & CR0_TS_MASK) && !(cr0 & CR0_TS_MASK)) {
768 vcpu->svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
769 vcpu->fpu_active = 1;
772 vcpu->cr0 = cr0;
773 cr0 |= CR0_PG_MASK | CR0_WP_MASK;
774 cr0 &= ~(CR0_CD_MASK | CR0_NW_MASK);
775 vcpu->svm->vmcb->save.cr0 = cr0;
778 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
780 vcpu->cr4 = cr4;
781 vcpu->svm->vmcb->save.cr4 = cr4 | CR4_PAE_MASK;
784 static void svm_set_segment(struct kvm_vcpu *vcpu,
785 struct kvm_segment *var, int seg)
787 struct vmcb_seg *s = svm_seg(vcpu, seg);
789 s->base = var->base;
790 s->limit = var->limit;
791 s->selector = var->selector;
792 if (var->unusable)
793 s->attrib = 0;
794 else {
795 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
796 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
797 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
798 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
799 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
800 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
801 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
802 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
804 if (seg == VCPU_SREG_CS)
805 vcpu->svm->vmcb->save.cpl
806 = (vcpu->svm->vmcb->save.cs.attrib
807 >> SVM_SELECTOR_DPL_SHIFT) & 3;
811 /* FIXME:
813 vcpu->svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
814 vcpu->svm->vmcb->control.int_ctl |= (sregs->cr8 & V_TPR_MASK);
818 static int svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
820 return -EOPNOTSUPP;
823 static void load_host_msrs(struct kvm_vcpu *vcpu)
825 #ifdef CONFIG_X86_64
826 wrmsrl(MSR_GS_BASE, vcpu->svm->host_gs_base);
827 #endif
830 static void save_host_msrs(struct kvm_vcpu *vcpu)
832 #ifdef CONFIG_X86_64
833 rdmsrl(MSR_GS_BASE, vcpu->svm->host_gs_base);
834 #endif
837 static void new_asid(struct kvm_vcpu *vcpu, struct svm_cpu_data *svm_data)
839 if (svm_data->next_asid > svm_data->max_asid) {
840 ++svm_data->asid_generation;
841 svm_data->next_asid = 1;
842 vcpu->svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
845 vcpu->cpu = svm_data->cpu;
846 vcpu->svm->asid_generation = svm_data->asid_generation;
847 vcpu->svm->vmcb->control.asid = svm_data->next_asid++;
850 static void svm_invlpg(struct kvm_vcpu *vcpu, gva_t address)
852 invlpga(address, vcpu->svm->vmcb->control.asid); // is needed?
855 static unsigned long svm_get_dr(struct kvm_vcpu *vcpu, int dr)
857 return vcpu->svm->db_regs[dr];
860 static void svm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long value,
861 int *exception)
863 *exception = 0;
865 if (vcpu->svm->vmcb->save.dr7 & DR7_GD_MASK) {
866 vcpu->svm->vmcb->save.dr7 &= ~DR7_GD_MASK;
867 vcpu->svm->vmcb->save.dr6 |= DR6_BD_MASK;
868 *exception = DB_VECTOR;
869 return;
872 switch (dr) {
873 case 0 ... 3:
874 vcpu->svm->db_regs[dr] = value;
875 return;
876 case 4 ... 5:
877 if (vcpu->cr4 & CR4_DE_MASK) {
878 *exception = UD_VECTOR;
879 return;
881 case 7: {
882 if (value & ~((1ULL << 32) - 1)) {
883 *exception = GP_VECTOR;
884 return;
886 vcpu->svm->vmcb->save.dr7 = value;
887 return;
889 default:
890 printk(KERN_DEBUG "%s: unexpected dr %u\n",
891 __FUNCTION__, dr);
892 *exception = UD_VECTOR;
893 return;
897 static int pf_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
899 u32 exit_int_info = vcpu->svm->vmcb->control.exit_int_info;
900 u64 fault_address;
901 u32 error_code;
902 enum emulation_result er;
903 int r;
905 if (is_external_interrupt(exit_int_info))
906 push_irq(vcpu, exit_int_info & SVM_EVTINJ_VEC_MASK);
908 spin_lock(&vcpu->kvm->lock);
910 fault_address = vcpu->svm->vmcb->control.exit_info_2;
911 error_code = vcpu->svm->vmcb->control.exit_info_1;
912 r = kvm_mmu_page_fault(vcpu, fault_address, error_code);
913 if (r < 0) {
914 spin_unlock(&vcpu->kvm->lock);
915 return r;
917 if (!r) {
918 spin_unlock(&vcpu->kvm->lock);
919 return 1;
921 er = emulate_instruction(vcpu, kvm_run, fault_address, error_code);
922 spin_unlock(&vcpu->kvm->lock);
924 switch (er) {
925 case EMULATE_DONE:
926 return 1;
927 case EMULATE_DO_MMIO:
928 ++vcpu->stat.mmio_exits;
929 kvm_run->exit_reason = KVM_EXIT_MMIO;
930 return 0;
931 case EMULATE_FAIL:
932 vcpu_printf(vcpu, "%s: emulate fail\n", __FUNCTION__);
933 break;
934 default:
935 BUG();
938 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
939 return 0;
942 static int nm_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
944 vcpu->svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
945 if (!(vcpu->cr0 & CR0_TS_MASK))
946 vcpu->svm->vmcb->save.cr0 &= ~CR0_TS_MASK;
947 vcpu->fpu_active = 1;
949 return 1;
952 static int shutdown_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
955 * VMCB is undefined after a SHUTDOWN intercept
956 * so reinitialize it.
958 memset(vcpu->svm->vmcb, 0, PAGE_SIZE);
959 init_vmcb(vcpu->svm->vmcb);
961 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
962 return 0;
965 static int io_get_override(struct kvm_vcpu *vcpu,
966 struct vmcb_seg **seg,
967 int *addr_override)
969 u8 inst[MAX_INST_SIZE];
970 unsigned ins_length;
971 gva_t rip;
972 int i;
974 rip = vcpu->svm->vmcb->save.rip;
975 ins_length = vcpu->svm->next_rip - rip;
976 rip += vcpu->svm->vmcb->save.cs.base;
978 if (ins_length > MAX_INST_SIZE)
979 printk(KERN_DEBUG
980 "%s: inst length err, cs base 0x%llx rip 0x%llx "
981 "next rip 0x%llx ins_length %u\n",
982 __FUNCTION__,
983 vcpu->svm->vmcb->save.cs.base,
984 vcpu->svm->vmcb->save.rip,
985 vcpu->svm->vmcb->control.exit_info_2,
986 ins_length);
988 if (kvm_read_guest(vcpu, rip, ins_length, inst) != ins_length)
989 /* #PF */
990 return 0;
992 *addr_override = 0;
993 *seg = NULL;
994 for (i = 0; i < ins_length; i++)
995 switch (inst[i]) {
996 case 0xf0:
997 case 0xf2:
998 case 0xf3:
999 case 0x66:
1000 continue;
1001 case 0x67:
1002 *addr_override = 1;
1003 continue;
1004 case 0x2e:
1005 *seg = &vcpu->svm->vmcb->save.cs;
1006 continue;
1007 case 0x36:
1008 *seg = &vcpu->svm->vmcb->save.ss;
1009 continue;
1010 case 0x3e:
1011 *seg = &vcpu->svm->vmcb->save.ds;
1012 continue;
1013 case 0x26:
1014 *seg = &vcpu->svm->vmcb->save.es;
1015 continue;
1016 case 0x64:
1017 *seg = &vcpu->svm->vmcb->save.fs;
1018 continue;
1019 case 0x65:
1020 *seg = &vcpu->svm->vmcb->save.gs;
1021 continue;
1022 default:
1023 return 1;
1025 printk(KERN_DEBUG "%s: unexpected\n", __FUNCTION__);
1026 return 0;
1029 static unsigned long io_adress(struct kvm_vcpu *vcpu, int ins, gva_t *address)
1031 unsigned long addr_mask;
1032 unsigned long *reg;
1033 struct vmcb_seg *seg;
1034 int addr_override;
1035 struct vmcb_save_area *save_area = &vcpu->svm->vmcb->save;
1036 u16 cs_attrib = save_area->cs.attrib;
1037 unsigned addr_size = get_addr_size(vcpu);
1039 if (!io_get_override(vcpu, &seg, &addr_override))
1040 return 0;
1042 if (addr_override)
1043 addr_size = (addr_size == 2) ? 4: (addr_size >> 1);
1045 if (ins) {
1046 reg = &vcpu->regs[VCPU_REGS_RDI];
1047 seg = &vcpu->svm->vmcb->save.es;
1048 } else {
1049 reg = &vcpu->regs[VCPU_REGS_RSI];
1050 seg = (seg) ? seg : &vcpu->svm->vmcb->save.ds;
1053 addr_mask = ~0ULL >> (64 - (addr_size * 8));
1055 if ((cs_attrib & SVM_SELECTOR_L_MASK) &&
1056 !(vcpu->svm->vmcb->save.rflags & X86_EFLAGS_VM)) {
1057 *address = (*reg & addr_mask);
1058 return addr_mask;
1061 if (!(seg->attrib & SVM_SELECTOR_P_SHIFT)) {
1062 svm_inject_gp(vcpu, 0);
1063 return 0;
1066 *address = (*reg & addr_mask) + seg->base;
1067 return addr_mask;
1070 static int io_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1072 u32 io_info = vcpu->svm->vmcb->control.exit_info_1; //address size bug?
1073 int size, down, in, string, rep;
1074 unsigned port;
1075 unsigned long count;
1076 gva_t address = 0;
1078 ++vcpu->stat.io_exits;
1080 vcpu->svm->next_rip = vcpu->svm->vmcb->control.exit_info_2;
1082 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1083 port = io_info >> 16;
1084 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1085 string = (io_info & SVM_IOIO_STR_MASK) != 0;
1086 rep = (io_info & SVM_IOIO_REP_MASK) != 0;
1087 count = 1;
1088 down = (vcpu->svm->vmcb->save.rflags & X86_EFLAGS_DF) != 0;
1090 if (string) {
1091 unsigned addr_mask;
1093 addr_mask = io_adress(vcpu, in, &address);
1094 if (!addr_mask) {
1095 printk(KERN_DEBUG "%s: get io address failed\n",
1096 __FUNCTION__);
1097 return 1;
1100 if (rep)
1101 count = vcpu->regs[VCPU_REGS_RCX] & addr_mask;
1103 return kvm_setup_pio(vcpu, kvm_run, in, size, count, string, down,
1104 address, rep, port);
1107 static int nop_on_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1109 return 1;
1112 static int halt_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1114 vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 1;
1115 skip_emulated_instruction(vcpu);
1116 if (vcpu->irq_summary)
1117 return 1;
1119 kvm_run->exit_reason = KVM_EXIT_HLT;
1120 ++vcpu->stat.halt_exits;
1121 return 0;
1124 static int vmmcall_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1126 vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 3;
1127 skip_emulated_instruction(vcpu);
1128 return kvm_hypercall(vcpu, kvm_run);
1131 static int invalid_op_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1133 inject_ud(vcpu);
1134 return 1;
1137 static int task_switch_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1139 printk(KERN_DEBUG "%s: task swiche is unsupported\n", __FUNCTION__);
1140 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1141 return 0;
1144 static int cpuid_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1146 vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 2;
1147 kvm_emulate_cpuid(vcpu);
1148 return 1;
1151 static int emulate_on_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1153 if (emulate_instruction(vcpu, NULL, 0, 0) != EMULATE_DONE)
1154 printk(KERN_ERR "%s: failed\n", __FUNCTION__);
1155 return 1;
1158 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
1160 switch (ecx) {
1161 case MSR_IA32_TIME_STAMP_COUNTER: {
1162 u64 tsc;
1164 rdtscll(tsc);
1165 *data = vcpu->svm->vmcb->control.tsc_offset + tsc;
1166 break;
1168 case MSR_K6_STAR:
1169 *data = vcpu->svm->vmcb->save.star;
1170 break;
1171 #ifdef CONFIG_X86_64
1172 case MSR_LSTAR:
1173 *data = vcpu->svm->vmcb->save.lstar;
1174 break;
1175 case MSR_CSTAR:
1176 *data = vcpu->svm->vmcb->save.cstar;
1177 break;
1178 case MSR_KERNEL_GS_BASE:
1179 *data = vcpu->svm->vmcb->save.kernel_gs_base;
1180 break;
1181 case MSR_SYSCALL_MASK:
1182 *data = vcpu->svm->vmcb->save.sfmask;
1183 break;
1184 #endif
1185 case MSR_IA32_SYSENTER_CS:
1186 *data = vcpu->svm->vmcb->save.sysenter_cs;
1187 break;
1188 case MSR_IA32_SYSENTER_EIP:
1189 *data = vcpu->svm->vmcb->save.sysenter_eip;
1190 break;
1191 case MSR_IA32_SYSENTER_ESP:
1192 *data = vcpu->svm->vmcb->save.sysenter_esp;
1193 break;
1194 default:
1195 return kvm_get_msr_common(vcpu, ecx, data);
1197 return 0;
1200 static int rdmsr_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1202 u32 ecx = vcpu->regs[VCPU_REGS_RCX];
1203 u64 data;
1205 if (svm_get_msr(vcpu, ecx, &data))
1206 svm_inject_gp(vcpu, 0);
1207 else {
1208 vcpu->svm->vmcb->save.rax = data & 0xffffffff;
1209 vcpu->regs[VCPU_REGS_RDX] = data >> 32;
1210 vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 2;
1211 skip_emulated_instruction(vcpu);
1213 return 1;
1216 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
1218 switch (ecx) {
1219 case MSR_IA32_TIME_STAMP_COUNTER: {
1220 u64 tsc;
1222 rdtscll(tsc);
1223 vcpu->svm->vmcb->control.tsc_offset = data - tsc;
1224 break;
1226 case MSR_K6_STAR:
1227 vcpu->svm->vmcb->save.star = data;
1228 break;
1229 #ifdef CONFIG_X86_64
1230 case MSR_LSTAR:
1231 vcpu->svm->vmcb->save.lstar = data;
1232 break;
1233 case MSR_CSTAR:
1234 vcpu->svm->vmcb->save.cstar = data;
1235 break;
1236 case MSR_KERNEL_GS_BASE:
1237 vcpu->svm->vmcb->save.kernel_gs_base = data;
1238 break;
1239 case MSR_SYSCALL_MASK:
1240 vcpu->svm->vmcb->save.sfmask = data;
1241 break;
1242 #endif
1243 case MSR_IA32_SYSENTER_CS:
1244 vcpu->svm->vmcb->save.sysenter_cs = data;
1245 break;
1246 case MSR_IA32_SYSENTER_EIP:
1247 vcpu->svm->vmcb->save.sysenter_eip = data;
1248 break;
1249 case MSR_IA32_SYSENTER_ESP:
1250 vcpu->svm->vmcb->save.sysenter_esp = data;
1251 break;
1252 default:
1253 return kvm_set_msr_common(vcpu, ecx, data);
1255 return 0;
1258 static int wrmsr_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1260 u32 ecx = vcpu->regs[VCPU_REGS_RCX];
1261 u64 data = (vcpu->svm->vmcb->save.rax & -1u)
1262 | ((u64)(vcpu->regs[VCPU_REGS_RDX] & -1u) << 32);
1263 vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 2;
1264 if (svm_set_msr(vcpu, ecx, data))
1265 svm_inject_gp(vcpu, 0);
1266 else
1267 skip_emulated_instruction(vcpu);
1268 return 1;
1271 static int msr_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1273 if (vcpu->svm->vmcb->control.exit_info_1)
1274 return wrmsr_interception(vcpu, kvm_run);
1275 else
1276 return rdmsr_interception(vcpu, kvm_run);
1279 static int interrupt_window_interception(struct kvm_vcpu *vcpu,
1280 struct kvm_run *kvm_run)
1283 * If the user space waits to inject interrupts, exit as soon as
1284 * possible
1286 if (kvm_run->request_interrupt_window &&
1287 !vcpu->irq_summary) {
1288 ++vcpu->stat.irq_window_exits;
1289 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
1290 return 0;
1293 return 1;
1296 static int (*svm_exit_handlers[])(struct kvm_vcpu *vcpu,
1297 struct kvm_run *kvm_run) = {
1298 [SVM_EXIT_READ_CR0] = emulate_on_interception,
1299 [SVM_EXIT_READ_CR3] = emulate_on_interception,
1300 [SVM_EXIT_READ_CR4] = emulate_on_interception,
1301 /* for now: */
1302 [SVM_EXIT_WRITE_CR0] = emulate_on_interception,
1303 [SVM_EXIT_WRITE_CR3] = emulate_on_interception,
1304 [SVM_EXIT_WRITE_CR4] = emulate_on_interception,
1305 [SVM_EXIT_READ_DR0] = emulate_on_interception,
1306 [SVM_EXIT_READ_DR1] = emulate_on_interception,
1307 [SVM_EXIT_READ_DR2] = emulate_on_interception,
1308 [SVM_EXIT_READ_DR3] = emulate_on_interception,
1309 [SVM_EXIT_WRITE_DR0] = emulate_on_interception,
1310 [SVM_EXIT_WRITE_DR1] = emulate_on_interception,
1311 [SVM_EXIT_WRITE_DR2] = emulate_on_interception,
1312 [SVM_EXIT_WRITE_DR3] = emulate_on_interception,
1313 [SVM_EXIT_WRITE_DR5] = emulate_on_interception,
1314 [SVM_EXIT_WRITE_DR7] = emulate_on_interception,
1315 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
1316 [SVM_EXIT_EXCP_BASE + NM_VECTOR] = nm_interception,
1317 [SVM_EXIT_INTR] = nop_on_interception,
1318 [SVM_EXIT_NMI] = nop_on_interception,
1319 [SVM_EXIT_SMI] = nop_on_interception,
1320 [SVM_EXIT_INIT] = nop_on_interception,
1321 [SVM_EXIT_VINTR] = interrupt_window_interception,
1322 /* [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception, */
1323 [SVM_EXIT_CPUID] = cpuid_interception,
1324 [SVM_EXIT_HLT] = halt_interception,
1325 [SVM_EXIT_INVLPG] = emulate_on_interception,
1326 [SVM_EXIT_INVLPGA] = invalid_op_interception,
1327 [SVM_EXIT_IOIO] = io_interception,
1328 [SVM_EXIT_MSR] = msr_interception,
1329 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
1330 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
1331 [SVM_EXIT_VMRUN] = invalid_op_interception,
1332 [SVM_EXIT_VMMCALL] = vmmcall_interception,
1333 [SVM_EXIT_VMLOAD] = invalid_op_interception,
1334 [SVM_EXIT_VMSAVE] = invalid_op_interception,
1335 [SVM_EXIT_STGI] = invalid_op_interception,
1336 [SVM_EXIT_CLGI] = invalid_op_interception,
1337 [SVM_EXIT_SKINIT] = invalid_op_interception,
1338 [SVM_EXIT_MONITOR] = invalid_op_interception,
1339 [SVM_EXIT_MWAIT] = invalid_op_interception,
1343 static int handle_exit(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1345 u32 exit_code = vcpu->svm->vmcb->control.exit_code;
1347 if (is_external_interrupt(vcpu->svm->vmcb->control.exit_int_info) &&
1348 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR)
1349 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
1350 "exit_code 0x%x\n",
1351 __FUNCTION__, vcpu->svm->vmcb->control.exit_int_info,
1352 exit_code);
1354 if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
1355 || svm_exit_handlers[exit_code] == 0) {
1356 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1357 kvm_run->hw.hardware_exit_reason = exit_code;
1358 return 0;
1361 return svm_exit_handlers[exit_code](vcpu, kvm_run);
1364 static void reload_tss(struct kvm_vcpu *vcpu)
1366 int cpu = raw_smp_processor_id();
1368 struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1369 svm_data->tss_desc->type = 9; //available 32/64-bit TSS
1370 load_TR_desc();
1373 static void pre_svm_run(struct kvm_vcpu *vcpu)
1375 int cpu = raw_smp_processor_id();
1377 struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1379 vcpu->svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
1380 if (vcpu->cpu != cpu ||
1381 vcpu->svm->asid_generation != svm_data->asid_generation)
1382 new_asid(vcpu, svm_data);
1386 static inline void kvm_do_inject_irq(struct kvm_vcpu *vcpu)
1388 struct vmcb_control_area *control;
1390 control = &vcpu->svm->vmcb->control;
1391 control->int_vector = pop_irq(vcpu);
1392 control->int_ctl &= ~V_INTR_PRIO_MASK;
1393 control->int_ctl |= V_IRQ_MASK |
1394 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
1397 static void kvm_reput_irq(struct kvm_vcpu *vcpu)
1399 struct vmcb_control_area *control = &vcpu->svm->vmcb->control;
1401 if (control->int_ctl & V_IRQ_MASK) {
1402 control->int_ctl &= ~V_IRQ_MASK;
1403 push_irq(vcpu, control->int_vector);
1406 vcpu->interrupt_window_open =
1407 !(control->int_state & SVM_INTERRUPT_SHADOW_MASK);
1410 static void do_interrupt_requests(struct kvm_vcpu *vcpu,
1411 struct kvm_run *kvm_run)
1413 struct vmcb_control_area *control = &vcpu->svm->vmcb->control;
1415 vcpu->interrupt_window_open =
1416 (!(control->int_state & SVM_INTERRUPT_SHADOW_MASK) &&
1417 (vcpu->svm->vmcb->save.rflags & X86_EFLAGS_IF));
1419 if (vcpu->interrupt_window_open && vcpu->irq_summary)
1421 * If interrupts enabled, and not blocked by sti or mov ss. Good.
1423 kvm_do_inject_irq(vcpu);
1426 * Interrupts blocked. Wait for unblock.
1428 if (!vcpu->interrupt_window_open &&
1429 (vcpu->irq_summary || kvm_run->request_interrupt_window)) {
1430 control->intercept |= 1ULL << INTERCEPT_VINTR;
1431 } else
1432 control->intercept &= ~(1ULL << INTERCEPT_VINTR);
1435 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1436 struct kvm_run *kvm_run)
1438 kvm_run->ready_for_interrupt_injection = (vcpu->interrupt_window_open &&
1439 vcpu->irq_summary == 0);
1440 kvm_run->if_flag = (vcpu->svm->vmcb->save.rflags & X86_EFLAGS_IF) != 0;
1441 kvm_run->cr8 = vcpu->cr8;
1442 kvm_run->apic_base = vcpu->apic_base;
1446 * Check if userspace requested an interrupt window, and that the
1447 * interrupt window is open.
1449 * No need to exit to userspace if we already have an interrupt queued.
1451 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1452 struct kvm_run *kvm_run)
1454 return (!vcpu->irq_summary &&
1455 kvm_run->request_interrupt_window &&
1456 vcpu->interrupt_window_open &&
1457 (vcpu->svm->vmcb->save.rflags & X86_EFLAGS_IF));
1460 static void save_db_regs(unsigned long *db_regs)
1462 asm volatile ("mov %%dr0, %0" : "=r"(db_regs[0]));
1463 asm volatile ("mov %%dr1, %0" : "=r"(db_regs[1]));
1464 asm volatile ("mov %%dr2, %0" : "=r"(db_regs[2]));
1465 asm volatile ("mov %%dr3, %0" : "=r"(db_regs[3]));
1468 static void load_db_regs(unsigned long *db_regs)
1470 asm volatile ("mov %0, %%dr0" : : "r"(db_regs[0]));
1471 asm volatile ("mov %0, %%dr1" : : "r"(db_regs[1]));
1472 asm volatile ("mov %0, %%dr2" : : "r"(db_regs[2]));
1473 asm volatile ("mov %0, %%dr3" : : "r"(db_regs[3]));
1476 static int svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1478 u16 fs_selector;
1479 u16 gs_selector;
1480 u16 ldt_selector;
1481 int r;
1483 again:
1484 if (!vcpu->mmio_read_completed)
1485 do_interrupt_requests(vcpu, kvm_run);
1487 clgi();
1489 pre_svm_run(vcpu);
1491 save_host_msrs(vcpu);
1492 fs_selector = read_fs();
1493 gs_selector = read_gs();
1494 ldt_selector = read_ldt();
1495 vcpu->svm->host_cr2 = kvm_read_cr2();
1496 vcpu->svm->host_dr6 = read_dr6();
1497 vcpu->svm->host_dr7 = read_dr7();
1498 vcpu->svm->vmcb->save.cr2 = vcpu->cr2;
1500 if (vcpu->svm->vmcb->save.dr7 & 0xff) {
1501 write_dr7(0);
1502 save_db_regs(vcpu->svm->host_db_regs);
1503 load_db_regs(vcpu->svm->db_regs);
1506 if (vcpu->fpu_active) {
1507 fx_save(vcpu->host_fx_image);
1508 fx_restore(vcpu->guest_fx_image);
1511 asm volatile (
1512 #ifdef CONFIG_X86_64
1513 "push %%rbx; push %%rcx; push %%rdx;"
1514 "push %%rsi; push %%rdi; push %%rbp;"
1515 "push %%r8; push %%r9; push %%r10; push %%r11;"
1516 "push %%r12; push %%r13; push %%r14; push %%r15;"
1517 #else
1518 "push %%ebx; push %%ecx; push %%edx;"
1519 "push %%esi; push %%edi; push %%ebp;"
1520 #endif
1522 #ifdef CONFIG_X86_64
1523 "mov %c[rbx](%[vcpu]), %%rbx \n\t"
1524 "mov %c[rcx](%[vcpu]), %%rcx \n\t"
1525 "mov %c[rdx](%[vcpu]), %%rdx \n\t"
1526 "mov %c[rsi](%[vcpu]), %%rsi \n\t"
1527 "mov %c[rdi](%[vcpu]), %%rdi \n\t"
1528 "mov %c[rbp](%[vcpu]), %%rbp \n\t"
1529 "mov %c[r8](%[vcpu]), %%r8 \n\t"
1530 "mov %c[r9](%[vcpu]), %%r9 \n\t"
1531 "mov %c[r10](%[vcpu]), %%r10 \n\t"
1532 "mov %c[r11](%[vcpu]), %%r11 \n\t"
1533 "mov %c[r12](%[vcpu]), %%r12 \n\t"
1534 "mov %c[r13](%[vcpu]), %%r13 \n\t"
1535 "mov %c[r14](%[vcpu]), %%r14 \n\t"
1536 "mov %c[r15](%[vcpu]), %%r15 \n\t"
1537 #else
1538 "mov %c[rbx](%[vcpu]), %%ebx \n\t"
1539 "mov %c[rcx](%[vcpu]), %%ecx \n\t"
1540 "mov %c[rdx](%[vcpu]), %%edx \n\t"
1541 "mov %c[rsi](%[vcpu]), %%esi \n\t"
1542 "mov %c[rdi](%[vcpu]), %%edi \n\t"
1543 "mov %c[rbp](%[vcpu]), %%ebp \n\t"
1544 #endif
1546 #ifdef CONFIG_X86_64
1547 /* Enter guest mode */
1548 "push %%rax \n\t"
1549 "mov %c[svm](%[vcpu]), %%rax \n\t"
1550 "mov %c[vmcb](%%rax), %%rax \n\t"
1551 SVM_VMLOAD "\n\t"
1552 SVM_VMRUN "\n\t"
1553 SVM_VMSAVE "\n\t"
1554 "pop %%rax \n\t"
1555 #else
1556 /* Enter guest mode */
1557 "push %%eax \n\t"
1558 "mov %c[svm](%[vcpu]), %%eax \n\t"
1559 "mov %c[vmcb](%%eax), %%eax \n\t"
1560 SVM_VMLOAD "\n\t"
1561 SVM_VMRUN "\n\t"
1562 SVM_VMSAVE "\n\t"
1563 "pop %%eax \n\t"
1564 #endif
1566 /* Save guest registers, load host registers */
1567 #ifdef CONFIG_X86_64
1568 "mov %%rbx, %c[rbx](%[vcpu]) \n\t"
1569 "mov %%rcx, %c[rcx](%[vcpu]) \n\t"
1570 "mov %%rdx, %c[rdx](%[vcpu]) \n\t"
1571 "mov %%rsi, %c[rsi](%[vcpu]) \n\t"
1572 "mov %%rdi, %c[rdi](%[vcpu]) \n\t"
1573 "mov %%rbp, %c[rbp](%[vcpu]) \n\t"
1574 "mov %%r8, %c[r8](%[vcpu]) \n\t"
1575 "mov %%r9, %c[r9](%[vcpu]) \n\t"
1576 "mov %%r10, %c[r10](%[vcpu]) \n\t"
1577 "mov %%r11, %c[r11](%[vcpu]) \n\t"
1578 "mov %%r12, %c[r12](%[vcpu]) \n\t"
1579 "mov %%r13, %c[r13](%[vcpu]) \n\t"
1580 "mov %%r14, %c[r14](%[vcpu]) \n\t"
1581 "mov %%r15, %c[r15](%[vcpu]) \n\t"
1583 "pop %%r15; pop %%r14; pop %%r13; pop %%r12;"
1584 "pop %%r11; pop %%r10; pop %%r9; pop %%r8;"
1585 "pop %%rbp; pop %%rdi; pop %%rsi;"
1586 "pop %%rdx; pop %%rcx; pop %%rbx; \n\t"
1587 #else
1588 "mov %%ebx, %c[rbx](%[vcpu]) \n\t"
1589 "mov %%ecx, %c[rcx](%[vcpu]) \n\t"
1590 "mov %%edx, %c[rdx](%[vcpu]) \n\t"
1591 "mov %%esi, %c[rsi](%[vcpu]) \n\t"
1592 "mov %%edi, %c[rdi](%[vcpu]) \n\t"
1593 "mov %%ebp, %c[rbp](%[vcpu]) \n\t"
1595 "pop %%ebp; pop %%edi; pop %%esi;"
1596 "pop %%edx; pop %%ecx; pop %%ebx; \n\t"
1597 #endif
1599 : [vcpu]"a"(vcpu),
1600 [svm]"i"(offsetof(struct kvm_vcpu, svm)),
1601 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
1602 [rbx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RBX])),
1603 [rcx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RCX])),
1604 [rdx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RDX])),
1605 [rsi]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RSI])),
1606 [rdi]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RDI])),
1607 [rbp]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RBP]))
1608 #ifdef CONFIG_X86_64
1609 ,[r8 ]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R8 ])),
1610 [r9 ]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R9 ])),
1611 [r10]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R10])),
1612 [r11]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R11])),
1613 [r12]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R12])),
1614 [r13]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R13])),
1615 [r14]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R14])),
1616 [r15]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R15]))
1617 #endif
1618 : "cc", "memory" );
1620 if (vcpu->fpu_active) {
1621 fx_save(vcpu->guest_fx_image);
1622 fx_restore(vcpu->host_fx_image);
1625 if ((vcpu->svm->vmcb->save.dr7 & 0xff))
1626 load_db_regs(vcpu->svm->host_db_regs);
1628 vcpu->cr2 = vcpu->svm->vmcb->save.cr2;
1630 write_dr6(vcpu->svm->host_dr6);
1631 write_dr7(vcpu->svm->host_dr7);
1632 kvm_write_cr2(vcpu->svm->host_cr2);
1634 load_fs(fs_selector);
1635 load_gs(gs_selector);
1636 load_ldt(ldt_selector);
1637 load_host_msrs(vcpu);
1639 reload_tss(vcpu);
1642 * Profile KVM exit RIPs:
1644 if (unlikely(prof_on == KVM_PROFILING))
1645 profile_hit(KVM_PROFILING,
1646 (void *)(unsigned long)vcpu->svm->vmcb->save.rip);
1648 stgi();
1650 kvm_reput_irq(vcpu);
1652 vcpu->svm->next_rip = 0;
1654 if (vcpu->svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
1655 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
1656 kvm_run->fail_entry.hardware_entry_failure_reason
1657 = vcpu->svm->vmcb->control.exit_code;
1658 post_kvm_run_save(vcpu, kvm_run);
1659 return 0;
1662 r = handle_exit(vcpu, kvm_run);
1663 if (r > 0) {
1664 if (signal_pending(current)) {
1665 ++vcpu->stat.signal_exits;
1666 post_kvm_run_save(vcpu, kvm_run);
1667 kvm_run->exit_reason = KVM_EXIT_INTR;
1668 return -EINTR;
1671 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
1672 ++vcpu->stat.request_irq_exits;
1673 post_kvm_run_save(vcpu, kvm_run);
1674 kvm_run->exit_reason = KVM_EXIT_INTR;
1675 return -EINTR;
1677 kvm_resched(vcpu);
1678 goto again;
1680 post_kvm_run_save(vcpu, kvm_run);
1681 return r;
1684 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
1686 force_new_asid(vcpu);
1689 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
1691 vcpu->svm->vmcb->save.cr3 = root;
1692 force_new_asid(vcpu);
1694 if (vcpu->fpu_active) {
1695 vcpu->svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
1696 vcpu->svm->vmcb->save.cr0 |= CR0_TS_MASK;
1697 vcpu->fpu_active = 0;
1701 static void svm_inject_page_fault(struct kvm_vcpu *vcpu,
1702 unsigned long addr,
1703 uint32_t err_code)
1705 uint32_t exit_int_info = vcpu->svm->vmcb->control.exit_int_info;
1707 ++vcpu->stat.pf_guest;
1709 if (is_page_fault(exit_int_info)) {
1711 vcpu->svm->vmcb->control.event_inj_err = 0;
1712 vcpu->svm->vmcb->control.event_inj = SVM_EVTINJ_VALID |
1713 SVM_EVTINJ_VALID_ERR |
1714 SVM_EVTINJ_TYPE_EXEPT |
1715 DF_VECTOR;
1716 return;
1718 vcpu->cr2 = addr;
1719 vcpu->svm->vmcb->save.cr2 = addr;
1720 vcpu->svm->vmcb->control.event_inj = SVM_EVTINJ_VALID |
1721 SVM_EVTINJ_VALID_ERR |
1722 SVM_EVTINJ_TYPE_EXEPT |
1723 PF_VECTOR;
1724 vcpu->svm->vmcb->control.event_inj_err = err_code;
1728 static int is_disabled(void)
1730 return 0;
1733 static void
1734 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
1737 * Patch in the VMMCALL instruction:
1739 hypercall[0] = 0x0f;
1740 hypercall[1] = 0x01;
1741 hypercall[2] = 0xd9;
1742 hypercall[3] = 0xc3;
1745 static struct kvm_arch_ops svm_arch_ops = {
1746 .cpu_has_kvm_support = has_svm,
1747 .disabled_by_bios = is_disabled,
1748 .hardware_setup = svm_hardware_setup,
1749 .hardware_unsetup = svm_hardware_unsetup,
1750 .hardware_enable = svm_hardware_enable,
1751 .hardware_disable = svm_hardware_disable,
1753 .vcpu_create = svm_create_vcpu,
1754 .vcpu_free = svm_free_vcpu,
1756 .vcpu_load = svm_vcpu_load,
1757 .vcpu_put = svm_vcpu_put,
1758 .vcpu_decache = svm_vcpu_decache,
1760 .set_guest_debug = svm_guest_debug,
1761 .get_msr = svm_get_msr,
1762 .set_msr = svm_set_msr,
1763 .get_segment_base = svm_get_segment_base,
1764 .get_segment = svm_get_segment,
1765 .set_segment = svm_set_segment,
1766 .get_cs_db_l_bits = svm_get_cs_db_l_bits,
1767 .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
1768 .set_cr0 = svm_set_cr0,
1769 .set_cr3 = svm_set_cr3,
1770 .set_cr4 = svm_set_cr4,
1771 .set_efer = svm_set_efer,
1772 .get_idt = svm_get_idt,
1773 .set_idt = svm_set_idt,
1774 .get_gdt = svm_get_gdt,
1775 .set_gdt = svm_set_gdt,
1776 .get_dr = svm_get_dr,
1777 .set_dr = svm_set_dr,
1778 .cache_regs = svm_cache_regs,
1779 .decache_regs = svm_decache_regs,
1780 .get_rflags = svm_get_rflags,
1781 .set_rflags = svm_set_rflags,
1783 .invlpg = svm_invlpg,
1784 .tlb_flush = svm_flush_tlb,
1785 .inject_page_fault = svm_inject_page_fault,
1787 .inject_gp = svm_inject_gp,
1789 .run = svm_vcpu_run,
1790 .skip_emulated_instruction = skip_emulated_instruction,
1791 .vcpu_setup = svm_vcpu_setup,
1792 .patch_hypercall = svm_patch_hypercall,
1795 static int __init svm_init(void)
1797 return kvm_init_arch(&svm_arch_ops, THIS_MODULE);
1800 static void __exit svm_exit(void)
1802 kvm_exit_arch();
1805 module_init(svm_init)
1806 module_exit(svm_exit)