sh: hp6xx pata_platform support.
[linux-2.6/btrfs-unstable.git] / drivers / kvm / svm.c
blob9c15f32eea1852def2f890e6d99c35248597619a
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 <asm/desc.h>
24 #include "kvm_svm.h"
25 #include "x86_emulate.h"
27 MODULE_AUTHOR("Qumranet");
28 MODULE_LICENSE("GPL");
30 #define IOPM_ALLOC_ORDER 2
31 #define MSRPM_ALLOC_ORDER 1
33 #define DB_VECTOR 1
34 #define UD_VECTOR 6
35 #define GP_VECTOR 13
37 #define DR7_GD_MASK (1 << 13)
38 #define DR6_BD_MASK (1 << 13)
39 #define CR4_DE_MASK (1UL << 3)
41 #define SEG_TYPE_LDT 2
42 #define SEG_TYPE_BUSY_TSS16 3
44 #define KVM_EFER_LMA (1 << 10)
45 #define KVM_EFER_LME (1 << 8)
47 #define SVM_FEATURE_NPT (1 << 0)
48 #define SVM_FEATURE_LBRV (1 << 1)
49 #define SVM_DEATURE_SVML (1 << 2)
51 unsigned long iopm_base;
52 unsigned long msrpm_base;
54 struct kvm_ldttss_desc {
55 u16 limit0;
56 u16 base0;
57 unsigned base1 : 8, type : 5, dpl : 2, p : 1;
58 unsigned limit1 : 4, zero0 : 3, g : 1, base2 : 8;
59 u32 base3;
60 u32 zero1;
61 } __attribute__((packed));
63 struct svm_cpu_data {
64 int cpu;
66 u64 asid_generation;
67 u32 max_asid;
68 u32 next_asid;
69 struct kvm_ldttss_desc *tss_desc;
71 struct page *save_area;
74 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
75 static uint32_t svm_features;
77 struct svm_init_data {
78 int cpu;
79 int r;
82 static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
84 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
85 #define MSRS_RANGE_SIZE 2048
86 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
88 #define MAX_INST_SIZE 15
90 static inline u32 svm_has(u32 feat)
92 return svm_features & feat;
95 static unsigned get_addr_size(struct kvm_vcpu *vcpu)
97 struct vmcb_save_area *sa = &vcpu->svm->vmcb->save;
98 u16 cs_attrib;
100 if (!(sa->cr0 & CR0_PE_MASK) || (sa->rflags & X86_EFLAGS_VM))
101 return 2;
103 cs_attrib = sa->cs.attrib;
105 return (cs_attrib & SVM_SELECTOR_L_MASK) ? 8 :
106 (cs_attrib & SVM_SELECTOR_DB_MASK) ? 4 : 2;
109 static inline u8 pop_irq(struct kvm_vcpu *vcpu)
111 int word_index = __ffs(vcpu->irq_summary);
112 int bit_index = __ffs(vcpu->irq_pending[word_index]);
113 int irq = word_index * BITS_PER_LONG + bit_index;
115 clear_bit(bit_index, &vcpu->irq_pending[word_index]);
116 if (!vcpu->irq_pending[word_index])
117 clear_bit(word_index, &vcpu->irq_summary);
118 return irq;
121 static inline void push_irq(struct kvm_vcpu *vcpu, u8 irq)
123 set_bit(irq, vcpu->irq_pending);
124 set_bit(irq / BITS_PER_LONG, &vcpu->irq_summary);
127 static inline void clgi(void)
129 asm volatile (SVM_CLGI);
132 static inline void stgi(void)
134 asm volatile (SVM_STGI);
137 static inline void invlpga(unsigned long addr, u32 asid)
139 asm volatile (SVM_INVLPGA :: "a"(addr), "c"(asid));
142 static inline unsigned long kvm_read_cr2(void)
144 unsigned long cr2;
146 asm volatile ("mov %%cr2, %0" : "=r" (cr2));
147 return cr2;
150 static inline void kvm_write_cr2(unsigned long val)
152 asm volatile ("mov %0, %%cr2" :: "r" (val));
155 static inline unsigned long read_dr6(void)
157 unsigned long dr6;
159 asm volatile ("mov %%dr6, %0" : "=r" (dr6));
160 return dr6;
163 static inline void write_dr6(unsigned long val)
165 asm volatile ("mov %0, %%dr6" :: "r" (val));
168 static inline unsigned long read_dr7(void)
170 unsigned long dr7;
172 asm volatile ("mov %%dr7, %0" : "=r" (dr7));
173 return dr7;
176 static inline void write_dr7(unsigned long val)
178 asm volatile ("mov %0, %%dr7" :: "r" (val));
181 static inline void force_new_asid(struct kvm_vcpu *vcpu)
183 vcpu->svm->asid_generation--;
186 static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
188 force_new_asid(vcpu);
191 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
193 if (!(efer & KVM_EFER_LMA))
194 efer &= ~KVM_EFER_LME;
196 vcpu->svm->vmcb->save.efer = efer | MSR_EFER_SVME_MASK;
197 vcpu->shadow_efer = efer;
200 static void svm_inject_gp(struct kvm_vcpu *vcpu, unsigned error_code)
202 vcpu->svm->vmcb->control.event_inj = SVM_EVTINJ_VALID |
203 SVM_EVTINJ_VALID_ERR |
204 SVM_EVTINJ_TYPE_EXEPT |
205 GP_VECTOR;
206 vcpu->svm->vmcb->control.event_inj_err = error_code;
209 static void inject_ud(struct kvm_vcpu *vcpu)
211 vcpu->svm->vmcb->control.event_inj = SVM_EVTINJ_VALID |
212 SVM_EVTINJ_TYPE_EXEPT |
213 UD_VECTOR;
216 static int is_page_fault(uint32_t info)
218 info &= SVM_EVTINJ_VEC_MASK | SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
219 return info == (PF_VECTOR | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT);
222 static int is_external_interrupt(u32 info)
224 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
225 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
228 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
230 if (!vcpu->svm->next_rip) {
231 printk(KERN_DEBUG "%s: NOP\n", __FUNCTION__);
232 return;
234 if (vcpu->svm->next_rip - vcpu->svm->vmcb->save.rip > 15) {
235 printk(KERN_ERR "%s: ip 0x%llx next 0x%llx\n",
236 __FUNCTION__,
237 vcpu->svm->vmcb->save.rip,
238 vcpu->svm->next_rip);
241 vcpu->rip = vcpu->svm->vmcb->save.rip = vcpu->svm->next_rip;
242 vcpu->svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
244 vcpu->interrupt_window_open = 1;
247 static int has_svm(void)
249 uint32_t eax, ebx, ecx, edx;
251 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD) {
252 printk(KERN_INFO "has_svm: not amd\n");
253 return 0;
256 cpuid(0x80000000, &eax, &ebx, &ecx, &edx);
257 if (eax < SVM_CPUID_FUNC) {
258 printk(KERN_INFO "has_svm: can't execute cpuid_8000000a\n");
259 return 0;
262 cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
263 if (!(ecx & (1 << SVM_CPUID_FEATURE_SHIFT))) {
264 printk(KERN_DEBUG "has_svm: svm not available\n");
265 return 0;
267 return 1;
270 static void svm_hardware_disable(void *garbage)
272 struct svm_cpu_data *svm_data
273 = per_cpu(svm_data, raw_smp_processor_id());
275 if (svm_data) {
276 uint64_t efer;
278 wrmsrl(MSR_VM_HSAVE_PA, 0);
279 rdmsrl(MSR_EFER, efer);
280 wrmsrl(MSR_EFER, efer & ~MSR_EFER_SVME_MASK);
281 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
282 __free_page(svm_data->save_area);
283 kfree(svm_data);
287 static void svm_hardware_enable(void *garbage)
290 struct svm_cpu_data *svm_data;
291 uint64_t efer;
292 #ifdef CONFIG_X86_64
293 struct desc_ptr gdt_descr;
294 #else
295 struct Xgt_desc_struct gdt_descr;
296 #endif
297 struct desc_struct *gdt;
298 int me = raw_smp_processor_id();
300 if (!has_svm()) {
301 printk(KERN_ERR "svm_cpu_init: err EOPNOTSUPP on %d\n", me);
302 return;
304 svm_data = per_cpu(svm_data, me);
306 if (!svm_data) {
307 printk(KERN_ERR "svm_cpu_init: svm_data is NULL on %d\n",
308 me);
309 return;
312 svm_data->asid_generation = 1;
313 svm_data->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
314 svm_data->next_asid = svm_data->max_asid + 1;
315 svm_features = cpuid_edx(SVM_CPUID_FUNC);
317 asm volatile ( "sgdt %0" : "=m"(gdt_descr) );
318 gdt = (struct desc_struct *)gdt_descr.address;
319 svm_data->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
321 rdmsrl(MSR_EFER, efer);
322 wrmsrl(MSR_EFER, efer | MSR_EFER_SVME_MASK);
324 wrmsrl(MSR_VM_HSAVE_PA,
325 page_to_pfn(svm_data->save_area) << PAGE_SHIFT);
328 static int svm_cpu_init(int cpu)
330 struct svm_cpu_data *svm_data;
331 int r;
333 svm_data = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
334 if (!svm_data)
335 return -ENOMEM;
336 svm_data->cpu = cpu;
337 svm_data->save_area = alloc_page(GFP_KERNEL);
338 r = -ENOMEM;
339 if (!svm_data->save_area)
340 goto err_1;
342 per_cpu(svm_data, cpu) = svm_data;
344 return 0;
346 err_1:
347 kfree(svm_data);
348 return r;
352 static int set_msr_interception(u32 *msrpm, unsigned msr,
353 int read, int write)
355 int i;
357 for (i = 0; i < NUM_MSR_MAPS; i++) {
358 if (msr >= msrpm_ranges[i] &&
359 msr < msrpm_ranges[i] + MSRS_IN_RANGE) {
360 u32 msr_offset = (i * MSRS_IN_RANGE + msr -
361 msrpm_ranges[i]) * 2;
363 u32 *base = msrpm + (msr_offset / 32);
364 u32 msr_shift = msr_offset % 32;
365 u32 mask = ((write) ? 0 : 2) | ((read) ? 0 : 1);
366 *base = (*base & ~(0x3 << msr_shift)) |
367 (mask << msr_shift);
368 return 1;
371 printk(KERN_DEBUG "%s: not found 0x%x\n", __FUNCTION__, msr);
372 return 0;
375 static __init int svm_hardware_setup(void)
377 int cpu;
378 struct page *iopm_pages;
379 struct page *msrpm_pages;
380 void *msrpm_va;
381 int r;
383 kvm_emulator_want_group7_invlpg();
385 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
387 if (!iopm_pages)
388 return -ENOMEM;
389 memset(page_address(iopm_pages), 0xff,
390 PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
391 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
394 msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
396 r = -ENOMEM;
397 if (!msrpm_pages)
398 goto err_1;
400 msrpm_va = page_address(msrpm_pages);
401 memset(msrpm_va, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
402 msrpm_base = page_to_pfn(msrpm_pages) << PAGE_SHIFT;
404 #ifdef CONFIG_X86_64
405 set_msr_interception(msrpm_va, MSR_GS_BASE, 1, 1);
406 set_msr_interception(msrpm_va, MSR_FS_BASE, 1, 1);
407 set_msr_interception(msrpm_va, MSR_KERNEL_GS_BASE, 1, 1);
408 set_msr_interception(msrpm_va, MSR_LSTAR, 1, 1);
409 set_msr_interception(msrpm_va, MSR_CSTAR, 1, 1);
410 set_msr_interception(msrpm_va, MSR_SYSCALL_MASK, 1, 1);
411 #endif
412 set_msr_interception(msrpm_va, MSR_K6_STAR, 1, 1);
413 set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_CS, 1, 1);
414 set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_ESP, 1, 1);
415 set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_EIP, 1, 1);
417 for_each_online_cpu(cpu) {
418 r = svm_cpu_init(cpu);
419 if (r)
420 goto err_2;
422 return 0;
424 err_2:
425 __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
426 msrpm_base = 0;
427 err_1:
428 __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
429 iopm_base = 0;
430 return r;
433 static __exit void svm_hardware_unsetup(void)
435 __free_pages(pfn_to_page(msrpm_base >> PAGE_SHIFT), MSRPM_ALLOC_ORDER);
436 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
437 iopm_base = msrpm_base = 0;
440 static void init_seg(struct vmcb_seg *seg)
442 seg->selector = 0;
443 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
444 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
445 seg->limit = 0xffff;
446 seg->base = 0;
449 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
451 seg->selector = 0;
452 seg->attrib = SVM_SELECTOR_P_MASK | type;
453 seg->limit = 0xffff;
454 seg->base = 0;
457 static int svm_vcpu_setup(struct kvm_vcpu *vcpu)
459 return 0;
462 static void init_vmcb(struct vmcb *vmcb)
464 struct vmcb_control_area *control = &vmcb->control;
465 struct vmcb_save_area *save = &vmcb->save;
467 control->intercept_cr_read = INTERCEPT_CR0_MASK |
468 INTERCEPT_CR3_MASK |
469 INTERCEPT_CR4_MASK;
471 control->intercept_cr_write = INTERCEPT_CR0_MASK |
472 INTERCEPT_CR3_MASK |
473 INTERCEPT_CR4_MASK;
475 control->intercept_dr_read = INTERCEPT_DR0_MASK |
476 INTERCEPT_DR1_MASK |
477 INTERCEPT_DR2_MASK |
478 INTERCEPT_DR3_MASK;
480 control->intercept_dr_write = INTERCEPT_DR0_MASK |
481 INTERCEPT_DR1_MASK |
482 INTERCEPT_DR2_MASK |
483 INTERCEPT_DR3_MASK |
484 INTERCEPT_DR5_MASK |
485 INTERCEPT_DR7_MASK;
487 control->intercept_exceptions = 1 << PF_VECTOR;
490 control->intercept = (1ULL << INTERCEPT_INTR) |
491 (1ULL << INTERCEPT_NMI) |
492 (1ULL << INTERCEPT_SMI) |
494 * selective cr0 intercept bug?
495 * 0: 0f 22 d8 mov %eax,%cr3
496 * 3: 0f 20 c0 mov %cr0,%eax
497 * 6: 0d 00 00 00 80 or $0x80000000,%eax
498 * b: 0f 22 c0 mov %eax,%cr0
499 * set cr3 ->interception
500 * get cr0 ->interception
501 * set cr0 -> no interception
503 /* (1ULL << INTERCEPT_SELECTIVE_CR0) | */
504 (1ULL << INTERCEPT_CPUID) |
505 (1ULL << INTERCEPT_HLT) |
506 (1ULL << INTERCEPT_INVLPGA) |
507 (1ULL << INTERCEPT_IOIO_PROT) |
508 (1ULL << INTERCEPT_MSR_PROT) |
509 (1ULL << INTERCEPT_TASK_SWITCH) |
510 (1ULL << INTERCEPT_SHUTDOWN) |
511 (1ULL << INTERCEPT_VMRUN) |
512 (1ULL << INTERCEPT_VMMCALL) |
513 (1ULL << INTERCEPT_VMLOAD) |
514 (1ULL << INTERCEPT_VMSAVE) |
515 (1ULL << INTERCEPT_STGI) |
516 (1ULL << INTERCEPT_CLGI) |
517 (1ULL << INTERCEPT_SKINIT) |
518 (1ULL << INTERCEPT_MONITOR) |
519 (1ULL << INTERCEPT_MWAIT);
521 control->iopm_base_pa = iopm_base;
522 control->msrpm_base_pa = msrpm_base;
523 control->tsc_offset = 0;
524 control->int_ctl = V_INTR_MASKING_MASK;
526 init_seg(&save->es);
527 init_seg(&save->ss);
528 init_seg(&save->ds);
529 init_seg(&save->fs);
530 init_seg(&save->gs);
532 save->cs.selector = 0xf000;
533 /* Executable/Readable Code Segment */
534 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
535 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
536 save->cs.limit = 0xffff;
538 * cs.base should really be 0xffff0000, but vmx can't handle that, so
539 * be consistent with it.
541 * Replace when we have real mode working for vmx.
543 save->cs.base = 0xf0000;
545 save->gdtr.limit = 0xffff;
546 save->idtr.limit = 0xffff;
548 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
549 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
551 save->efer = MSR_EFER_SVME_MASK;
553 save->dr6 = 0xffff0ff0;
554 save->dr7 = 0x400;
555 save->rflags = 2;
556 save->rip = 0x0000fff0;
559 * cr0 val on cpu init should be 0x60000010, we enable cpu
560 * cache by default. the orderly way is to enable cache in bios.
562 save->cr0 = 0x00000010 | CR0_PG_MASK | CR0_WP_MASK;
563 save->cr4 = CR4_PAE_MASK;
564 /* rdx = ?? */
567 static int svm_create_vcpu(struct kvm_vcpu *vcpu)
569 struct page *page;
570 int r;
572 r = -ENOMEM;
573 vcpu->svm = kzalloc(sizeof *vcpu->svm, GFP_KERNEL);
574 if (!vcpu->svm)
575 goto out1;
576 page = alloc_page(GFP_KERNEL);
577 if (!page)
578 goto out2;
580 vcpu->svm->vmcb = page_address(page);
581 memset(vcpu->svm->vmcb, 0, PAGE_SIZE);
582 vcpu->svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
583 vcpu->svm->asid_generation = 0;
584 memset(vcpu->svm->db_regs, 0, sizeof(vcpu->svm->db_regs));
585 init_vmcb(vcpu->svm->vmcb);
587 fx_init(vcpu);
588 vcpu->fpu_active = 1;
589 vcpu->apic_base = 0xfee00000 |
590 /*for vcpu 0*/ MSR_IA32_APICBASE_BSP |
591 MSR_IA32_APICBASE_ENABLE;
593 return 0;
595 out2:
596 kfree(vcpu->svm);
597 out1:
598 return r;
601 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
603 if (!vcpu->svm)
604 return;
605 if (vcpu->svm->vmcb)
606 __free_page(pfn_to_page(vcpu->svm->vmcb_pa >> PAGE_SHIFT));
607 kfree(vcpu->svm);
610 static void svm_vcpu_load(struct kvm_vcpu *vcpu)
612 int cpu, i;
614 cpu = get_cpu();
615 if (unlikely(cpu != vcpu->cpu)) {
616 u64 tsc_this, delta;
619 * Make sure that the guest sees a monotonically
620 * increasing TSC.
622 rdtscll(tsc_this);
623 delta = vcpu->host_tsc - tsc_this;
624 vcpu->svm->vmcb->control.tsc_offset += delta;
625 vcpu->cpu = cpu;
628 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
629 rdmsrl(host_save_user_msrs[i], vcpu->svm->host_user_msrs[i]);
632 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
634 int i;
636 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
637 wrmsrl(host_save_user_msrs[i], vcpu->svm->host_user_msrs[i]);
639 rdtscll(vcpu->host_tsc);
640 put_cpu();
643 static void svm_vcpu_decache(struct kvm_vcpu *vcpu)
647 static void svm_cache_regs(struct kvm_vcpu *vcpu)
649 vcpu->regs[VCPU_REGS_RAX] = vcpu->svm->vmcb->save.rax;
650 vcpu->regs[VCPU_REGS_RSP] = vcpu->svm->vmcb->save.rsp;
651 vcpu->rip = vcpu->svm->vmcb->save.rip;
654 static void svm_decache_regs(struct kvm_vcpu *vcpu)
656 vcpu->svm->vmcb->save.rax = vcpu->regs[VCPU_REGS_RAX];
657 vcpu->svm->vmcb->save.rsp = vcpu->regs[VCPU_REGS_RSP];
658 vcpu->svm->vmcb->save.rip = vcpu->rip;
661 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
663 return vcpu->svm->vmcb->save.rflags;
666 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
668 vcpu->svm->vmcb->save.rflags = rflags;
671 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
673 struct vmcb_save_area *save = &vcpu->svm->vmcb->save;
675 switch (seg) {
676 case VCPU_SREG_CS: return &save->cs;
677 case VCPU_SREG_DS: return &save->ds;
678 case VCPU_SREG_ES: return &save->es;
679 case VCPU_SREG_FS: return &save->fs;
680 case VCPU_SREG_GS: return &save->gs;
681 case VCPU_SREG_SS: return &save->ss;
682 case VCPU_SREG_TR: return &save->tr;
683 case VCPU_SREG_LDTR: return &save->ldtr;
685 BUG();
686 return NULL;
689 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
691 struct vmcb_seg *s = svm_seg(vcpu, seg);
693 return s->base;
696 static void svm_get_segment(struct kvm_vcpu *vcpu,
697 struct kvm_segment *var, int seg)
699 struct vmcb_seg *s = svm_seg(vcpu, seg);
701 var->base = s->base;
702 var->limit = s->limit;
703 var->selector = s->selector;
704 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
705 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
706 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
707 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
708 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
709 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
710 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
711 var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
712 var->unusable = !var->present;
715 static void svm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
717 struct vmcb_seg *s = svm_seg(vcpu, VCPU_SREG_CS);
719 *db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
720 *l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
723 static void svm_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
725 dt->limit = vcpu->svm->vmcb->save.idtr.limit;
726 dt->base = vcpu->svm->vmcb->save.idtr.base;
729 static void svm_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
731 vcpu->svm->vmcb->save.idtr.limit = dt->limit;
732 vcpu->svm->vmcb->save.idtr.base = dt->base ;
735 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
737 dt->limit = vcpu->svm->vmcb->save.gdtr.limit;
738 dt->base = vcpu->svm->vmcb->save.gdtr.base;
741 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
743 vcpu->svm->vmcb->save.gdtr.limit = dt->limit;
744 vcpu->svm->vmcb->save.gdtr.base = dt->base ;
747 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
751 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
753 #ifdef CONFIG_X86_64
754 if (vcpu->shadow_efer & KVM_EFER_LME) {
755 if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
756 vcpu->shadow_efer |= KVM_EFER_LMA;
757 vcpu->svm->vmcb->save.efer |= KVM_EFER_LMA | KVM_EFER_LME;
760 if (is_paging(vcpu) && !(cr0 & CR0_PG_MASK) ) {
761 vcpu->shadow_efer &= ~KVM_EFER_LMA;
762 vcpu->svm->vmcb->save.efer &= ~(KVM_EFER_LMA | KVM_EFER_LME);
765 #endif
766 if ((vcpu->cr0 & CR0_TS_MASK) && !(cr0 & CR0_TS_MASK)) {
767 vcpu->svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
768 vcpu->fpu_active = 1;
771 vcpu->cr0 = cr0;
772 cr0 |= CR0_PG_MASK | CR0_WP_MASK;
773 cr0 &= ~(CR0_CD_MASK | CR0_NW_MASK);
774 vcpu->svm->vmcb->save.cr0 = cr0;
777 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
779 vcpu->cr4 = cr4;
780 vcpu->svm->vmcb->save.cr4 = cr4 | CR4_PAE_MASK;
783 static void svm_set_segment(struct kvm_vcpu *vcpu,
784 struct kvm_segment *var, int seg)
786 struct vmcb_seg *s = svm_seg(vcpu, seg);
788 s->base = var->base;
789 s->limit = var->limit;
790 s->selector = var->selector;
791 if (var->unusable)
792 s->attrib = 0;
793 else {
794 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
795 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
796 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
797 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
798 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
799 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
800 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
801 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
803 if (seg == VCPU_SREG_CS)
804 vcpu->svm->vmcb->save.cpl
805 = (vcpu->svm->vmcb->save.cs.attrib
806 >> SVM_SELECTOR_DPL_SHIFT) & 3;
810 /* FIXME:
812 vcpu->svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
813 vcpu->svm->vmcb->control.int_ctl |= (sregs->cr8 & V_TPR_MASK);
817 static int svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
819 return -EOPNOTSUPP;
822 static void load_host_msrs(struct kvm_vcpu *vcpu)
824 #ifdef CONFIG_X86_64
825 wrmsrl(MSR_GS_BASE, vcpu->svm->host_gs_base);
826 #endif
829 static void save_host_msrs(struct kvm_vcpu *vcpu)
831 #ifdef CONFIG_X86_64
832 rdmsrl(MSR_GS_BASE, vcpu->svm->host_gs_base);
833 #endif
836 static void new_asid(struct kvm_vcpu *vcpu, struct svm_cpu_data *svm_data)
838 if (svm_data->next_asid > svm_data->max_asid) {
839 ++svm_data->asid_generation;
840 svm_data->next_asid = 1;
841 vcpu->svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
844 vcpu->cpu = svm_data->cpu;
845 vcpu->svm->asid_generation = svm_data->asid_generation;
846 vcpu->svm->vmcb->control.asid = svm_data->next_asid++;
849 static void svm_invlpg(struct kvm_vcpu *vcpu, gva_t address)
851 invlpga(address, vcpu->svm->vmcb->control.asid); // is needed?
854 static unsigned long svm_get_dr(struct kvm_vcpu *vcpu, int dr)
856 return vcpu->svm->db_regs[dr];
859 static void svm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long value,
860 int *exception)
862 *exception = 0;
864 if (vcpu->svm->vmcb->save.dr7 & DR7_GD_MASK) {
865 vcpu->svm->vmcb->save.dr7 &= ~DR7_GD_MASK;
866 vcpu->svm->vmcb->save.dr6 |= DR6_BD_MASK;
867 *exception = DB_VECTOR;
868 return;
871 switch (dr) {
872 case 0 ... 3:
873 vcpu->svm->db_regs[dr] = value;
874 return;
875 case 4 ... 5:
876 if (vcpu->cr4 & CR4_DE_MASK) {
877 *exception = UD_VECTOR;
878 return;
880 case 7: {
881 if (value & ~((1ULL << 32) - 1)) {
882 *exception = GP_VECTOR;
883 return;
885 vcpu->svm->vmcb->save.dr7 = value;
886 return;
888 default:
889 printk(KERN_DEBUG "%s: unexpected dr %u\n",
890 __FUNCTION__, dr);
891 *exception = UD_VECTOR;
892 return;
896 static int pf_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
898 u32 exit_int_info = vcpu->svm->vmcb->control.exit_int_info;
899 u64 fault_address;
900 u32 error_code;
901 enum emulation_result er;
902 int r;
904 if (is_external_interrupt(exit_int_info))
905 push_irq(vcpu, exit_int_info & SVM_EVTINJ_VEC_MASK);
907 spin_lock(&vcpu->kvm->lock);
909 fault_address = vcpu->svm->vmcb->control.exit_info_2;
910 error_code = vcpu->svm->vmcb->control.exit_info_1;
911 r = kvm_mmu_page_fault(vcpu, fault_address, error_code);
912 if (r < 0) {
913 spin_unlock(&vcpu->kvm->lock);
914 return r;
916 if (!r) {
917 spin_unlock(&vcpu->kvm->lock);
918 return 1;
920 er = emulate_instruction(vcpu, kvm_run, fault_address, error_code);
921 spin_unlock(&vcpu->kvm->lock);
923 switch (er) {
924 case EMULATE_DONE:
925 return 1;
926 case EMULATE_DO_MMIO:
927 ++vcpu->stat.mmio_exits;
928 kvm_run->exit_reason = KVM_EXIT_MMIO;
929 return 0;
930 case EMULATE_FAIL:
931 vcpu_printf(vcpu, "%s: emulate fail\n", __FUNCTION__);
932 break;
933 default:
934 BUG();
937 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
938 return 0;
941 static int nm_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
943 vcpu->svm->vmcb->control.intercept_exceptions &= ~(1 << NM_VECTOR);
944 if (!(vcpu->cr0 & CR0_TS_MASK))
945 vcpu->svm->vmcb->save.cr0 &= ~CR0_TS_MASK;
946 vcpu->fpu_active = 1;
948 return 1;
951 static int shutdown_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
954 * VMCB is undefined after a SHUTDOWN intercept
955 * so reinitialize it.
957 memset(vcpu->svm->vmcb, 0, PAGE_SIZE);
958 init_vmcb(vcpu->svm->vmcb);
960 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
961 return 0;
964 static int io_get_override(struct kvm_vcpu *vcpu,
965 struct vmcb_seg **seg,
966 int *addr_override)
968 u8 inst[MAX_INST_SIZE];
969 unsigned ins_length;
970 gva_t rip;
971 int i;
973 rip = vcpu->svm->vmcb->save.rip;
974 ins_length = vcpu->svm->next_rip - rip;
975 rip += vcpu->svm->vmcb->save.cs.base;
977 if (ins_length > MAX_INST_SIZE)
978 printk(KERN_DEBUG
979 "%s: inst length err, cs base 0x%llx rip 0x%llx "
980 "next rip 0x%llx ins_length %u\n",
981 __FUNCTION__,
982 vcpu->svm->vmcb->save.cs.base,
983 vcpu->svm->vmcb->save.rip,
984 vcpu->svm->vmcb->control.exit_info_2,
985 ins_length);
987 if (kvm_read_guest(vcpu, rip, ins_length, inst) != ins_length)
988 /* #PF */
989 return 0;
991 *addr_override = 0;
992 *seg = NULL;
993 for (i = 0; i < ins_length; i++)
994 switch (inst[i]) {
995 case 0xf0:
996 case 0xf2:
997 case 0xf3:
998 case 0x66:
999 continue;
1000 case 0x67:
1001 *addr_override = 1;
1002 continue;
1003 case 0x2e:
1004 *seg = &vcpu->svm->vmcb->save.cs;
1005 continue;
1006 case 0x36:
1007 *seg = &vcpu->svm->vmcb->save.ss;
1008 continue;
1009 case 0x3e:
1010 *seg = &vcpu->svm->vmcb->save.ds;
1011 continue;
1012 case 0x26:
1013 *seg = &vcpu->svm->vmcb->save.es;
1014 continue;
1015 case 0x64:
1016 *seg = &vcpu->svm->vmcb->save.fs;
1017 continue;
1018 case 0x65:
1019 *seg = &vcpu->svm->vmcb->save.gs;
1020 continue;
1021 default:
1022 return 1;
1024 printk(KERN_DEBUG "%s: unexpected\n", __FUNCTION__);
1025 return 0;
1028 static unsigned long io_adress(struct kvm_vcpu *vcpu, int ins, gva_t *address)
1030 unsigned long addr_mask;
1031 unsigned long *reg;
1032 struct vmcb_seg *seg;
1033 int addr_override;
1034 struct vmcb_save_area *save_area = &vcpu->svm->vmcb->save;
1035 u16 cs_attrib = save_area->cs.attrib;
1036 unsigned addr_size = get_addr_size(vcpu);
1038 if (!io_get_override(vcpu, &seg, &addr_override))
1039 return 0;
1041 if (addr_override)
1042 addr_size = (addr_size == 2) ? 4: (addr_size >> 1);
1044 if (ins) {
1045 reg = &vcpu->regs[VCPU_REGS_RDI];
1046 seg = &vcpu->svm->vmcb->save.es;
1047 } else {
1048 reg = &vcpu->regs[VCPU_REGS_RSI];
1049 seg = (seg) ? seg : &vcpu->svm->vmcb->save.ds;
1052 addr_mask = ~0ULL >> (64 - (addr_size * 8));
1054 if ((cs_attrib & SVM_SELECTOR_L_MASK) &&
1055 !(vcpu->svm->vmcb->save.rflags & X86_EFLAGS_VM)) {
1056 *address = (*reg & addr_mask);
1057 return addr_mask;
1060 if (!(seg->attrib & SVM_SELECTOR_P_SHIFT)) {
1061 svm_inject_gp(vcpu, 0);
1062 return 0;
1065 *address = (*reg & addr_mask) + seg->base;
1066 return addr_mask;
1069 static int io_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1071 u32 io_info = vcpu->svm->vmcb->control.exit_info_1; //address size bug?
1072 int size, down, in, string, rep;
1073 unsigned port;
1074 unsigned long count;
1075 gva_t address = 0;
1077 ++vcpu->stat.io_exits;
1079 vcpu->svm->next_rip = vcpu->svm->vmcb->control.exit_info_2;
1081 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1082 port = io_info >> 16;
1083 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1084 string = (io_info & SVM_IOIO_STR_MASK) != 0;
1085 rep = (io_info & SVM_IOIO_REP_MASK) != 0;
1086 count = 1;
1087 down = (vcpu->svm->vmcb->save.rflags & X86_EFLAGS_DF) != 0;
1089 if (string) {
1090 unsigned addr_mask;
1092 addr_mask = io_adress(vcpu, in, &address);
1093 if (!addr_mask) {
1094 printk(KERN_DEBUG "%s: get io address failed\n",
1095 __FUNCTION__);
1096 return 1;
1099 if (rep)
1100 count = vcpu->regs[VCPU_REGS_RCX] & addr_mask;
1102 return kvm_setup_pio(vcpu, kvm_run, in, size, count, string, down,
1103 address, rep, port);
1106 static int nop_on_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1108 return 1;
1111 static int halt_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1113 vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 1;
1114 skip_emulated_instruction(vcpu);
1115 if (vcpu->irq_summary)
1116 return 1;
1118 kvm_run->exit_reason = KVM_EXIT_HLT;
1119 ++vcpu->stat.halt_exits;
1120 return 0;
1123 static int vmmcall_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1125 vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 3;
1126 skip_emulated_instruction(vcpu);
1127 return kvm_hypercall(vcpu, kvm_run);
1130 static int invalid_op_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1132 inject_ud(vcpu);
1133 return 1;
1136 static int task_switch_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1138 printk(KERN_DEBUG "%s: task swiche is unsupported\n", __FUNCTION__);
1139 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1140 return 0;
1143 static int cpuid_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1145 vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 2;
1146 kvm_emulate_cpuid(vcpu);
1147 return 1;
1150 static int emulate_on_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1152 if (emulate_instruction(vcpu, NULL, 0, 0) != EMULATE_DONE)
1153 printk(KERN_ERR "%s: failed\n", __FUNCTION__);
1154 return 1;
1157 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
1159 switch (ecx) {
1160 case MSR_IA32_TIME_STAMP_COUNTER: {
1161 u64 tsc;
1163 rdtscll(tsc);
1164 *data = vcpu->svm->vmcb->control.tsc_offset + tsc;
1165 break;
1167 case MSR_K6_STAR:
1168 *data = vcpu->svm->vmcb->save.star;
1169 break;
1170 #ifdef CONFIG_X86_64
1171 case MSR_LSTAR:
1172 *data = vcpu->svm->vmcb->save.lstar;
1173 break;
1174 case MSR_CSTAR:
1175 *data = vcpu->svm->vmcb->save.cstar;
1176 break;
1177 case MSR_KERNEL_GS_BASE:
1178 *data = vcpu->svm->vmcb->save.kernel_gs_base;
1179 break;
1180 case MSR_SYSCALL_MASK:
1181 *data = vcpu->svm->vmcb->save.sfmask;
1182 break;
1183 #endif
1184 case MSR_IA32_SYSENTER_CS:
1185 *data = vcpu->svm->vmcb->save.sysenter_cs;
1186 break;
1187 case MSR_IA32_SYSENTER_EIP:
1188 *data = vcpu->svm->vmcb->save.sysenter_eip;
1189 break;
1190 case MSR_IA32_SYSENTER_ESP:
1191 *data = vcpu->svm->vmcb->save.sysenter_esp;
1192 break;
1193 default:
1194 return kvm_get_msr_common(vcpu, ecx, data);
1196 return 0;
1199 static int rdmsr_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1201 u32 ecx = vcpu->regs[VCPU_REGS_RCX];
1202 u64 data;
1204 if (svm_get_msr(vcpu, ecx, &data))
1205 svm_inject_gp(vcpu, 0);
1206 else {
1207 vcpu->svm->vmcb->save.rax = data & 0xffffffff;
1208 vcpu->regs[VCPU_REGS_RDX] = data >> 32;
1209 vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 2;
1210 skip_emulated_instruction(vcpu);
1212 return 1;
1215 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
1217 switch (ecx) {
1218 case MSR_IA32_TIME_STAMP_COUNTER: {
1219 u64 tsc;
1221 rdtscll(tsc);
1222 vcpu->svm->vmcb->control.tsc_offset = data - tsc;
1223 break;
1225 case MSR_K6_STAR:
1226 vcpu->svm->vmcb->save.star = data;
1227 break;
1228 #ifdef CONFIG_X86_64
1229 case MSR_LSTAR:
1230 vcpu->svm->vmcb->save.lstar = data;
1231 break;
1232 case MSR_CSTAR:
1233 vcpu->svm->vmcb->save.cstar = data;
1234 break;
1235 case MSR_KERNEL_GS_BASE:
1236 vcpu->svm->vmcb->save.kernel_gs_base = data;
1237 break;
1238 case MSR_SYSCALL_MASK:
1239 vcpu->svm->vmcb->save.sfmask = data;
1240 break;
1241 #endif
1242 case MSR_IA32_SYSENTER_CS:
1243 vcpu->svm->vmcb->save.sysenter_cs = data;
1244 break;
1245 case MSR_IA32_SYSENTER_EIP:
1246 vcpu->svm->vmcb->save.sysenter_eip = data;
1247 break;
1248 case MSR_IA32_SYSENTER_ESP:
1249 vcpu->svm->vmcb->save.sysenter_esp = data;
1250 break;
1251 default:
1252 return kvm_set_msr_common(vcpu, ecx, data);
1254 return 0;
1257 static int wrmsr_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1259 u32 ecx = vcpu->regs[VCPU_REGS_RCX];
1260 u64 data = (vcpu->svm->vmcb->save.rax & -1u)
1261 | ((u64)(vcpu->regs[VCPU_REGS_RDX] & -1u) << 32);
1262 vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 2;
1263 if (svm_set_msr(vcpu, ecx, data))
1264 svm_inject_gp(vcpu, 0);
1265 else
1266 skip_emulated_instruction(vcpu);
1267 return 1;
1270 static int msr_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1272 if (vcpu->svm->vmcb->control.exit_info_1)
1273 return wrmsr_interception(vcpu, kvm_run);
1274 else
1275 return rdmsr_interception(vcpu, kvm_run);
1278 static int interrupt_window_interception(struct kvm_vcpu *vcpu,
1279 struct kvm_run *kvm_run)
1282 * If the user space waits to inject interrupts, exit as soon as
1283 * possible
1285 if (kvm_run->request_interrupt_window &&
1286 !vcpu->irq_summary) {
1287 ++vcpu->stat.irq_window_exits;
1288 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
1289 return 0;
1292 return 1;
1295 static int (*svm_exit_handlers[])(struct kvm_vcpu *vcpu,
1296 struct kvm_run *kvm_run) = {
1297 [SVM_EXIT_READ_CR0] = emulate_on_interception,
1298 [SVM_EXIT_READ_CR3] = emulate_on_interception,
1299 [SVM_EXIT_READ_CR4] = emulate_on_interception,
1300 /* for now: */
1301 [SVM_EXIT_WRITE_CR0] = emulate_on_interception,
1302 [SVM_EXIT_WRITE_CR3] = emulate_on_interception,
1303 [SVM_EXIT_WRITE_CR4] = emulate_on_interception,
1304 [SVM_EXIT_READ_DR0] = emulate_on_interception,
1305 [SVM_EXIT_READ_DR1] = emulate_on_interception,
1306 [SVM_EXIT_READ_DR2] = emulate_on_interception,
1307 [SVM_EXIT_READ_DR3] = emulate_on_interception,
1308 [SVM_EXIT_WRITE_DR0] = emulate_on_interception,
1309 [SVM_EXIT_WRITE_DR1] = emulate_on_interception,
1310 [SVM_EXIT_WRITE_DR2] = emulate_on_interception,
1311 [SVM_EXIT_WRITE_DR3] = emulate_on_interception,
1312 [SVM_EXIT_WRITE_DR5] = emulate_on_interception,
1313 [SVM_EXIT_WRITE_DR7] = emulate_on_interception,
1314 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
1315 [SVM_EXIT_EXCP_BASE + NM_VECTOR] = nm_interception,
1316 [SVM_EXIT_INTR] = nop_on_interception,
1317 [SVM_EXIT_NMI] = nop_on_interception,
1318 [SVM_EXIT_SMI] = nop_on_interception,
1319 [SVM_EXIT_INIT] = nop_on_interception,
1320 [SVM_EXIT_VINTR] = interrupt_window_interception,
1321 /* [SVM_EXIT_CR0_SEL_WRITE] = emulate_on_interception, */
1322 [SVM_EXIT_CPUID] = cpuid_interception,
1323 [SVM_EXIT_HLT] = halt_interception,
1324 [SVM_EXIT_INVLPG] = emulate_on_interception,
1325 [SVM_EXIT_INVLPGA] = invalid_op_interception,
1326 [SVM_EXIT_IOIO] = io_interception,
1327 [SVM_EXIT_MSR] = msr_interception,
1328 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
1329 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
1330 [SVM_EXIT_VMRUN] = invalid_op_interception,
1331 [SVM_EXIT_VMMCALL] = vmmcall_interception,
1332 [SVM_EXIT_VMLOAD] = invalid_op_interception,
1333 [SVM_EXIT_VMSAVE] = invalid_op_interception,
1334 [SVM_EXIT_STGI] = invalid_op_interception,
1335 [SVM_EXIT_CLGI] = invalid_op_interception,
1336 [SVM_EXIT_SKINIT] = invalid_op_interception,
1337 [SVM_EXIT_MONITOR] = invalid_op_interception,
1338 [SVM_EXIT_MWAIT] = invalid_op_interception,
1342 static int handle_exit(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1344 u32 exit_code = vcpu->svm->vmcb->control.exit_code;
1346 if (is_external_interrupt(vcpu->svm->vmcb->control.exit_int_info) &&
1347 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR)
1348 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
1349 "exit_code 0x%x\n",
1350 __FUNCTION__, vcpu->svm->vmcb->control.exit_int_info,
1351 exit_code);
1353 if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
1354 || svm_exit_handlers[exit_code] == 0) {
1355 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1356 kvm_run->hw.hardware_exit_reason = exit_code;
1357 return 0;
1360 return svm_exit_handlers[exit_code](vcpu, kvm_run);
1363 static void reload_tss(struct kvm_vcpu *vcpu)
1365 int cpu = raw_smp_processor_id();
1367 struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1368 svm_data->tss_desc->type = 9; //available 32/64-bit TSS
1369 load_TR_desc();
1372 static void pre_svm_run(struct kvm_vcpu *vcpu)
1374 int cpu = raw_smp_processor_id();
1376 struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1378 vcpu->svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
1379 if (vcpu->cpu != cpu ||
1380 vcpu->svm->asid_generation != svm_data->asid_generation)
1381 new_asid(vcpu, svm_data);
1385 static inline void kvm_do_inject_irq(struct kvm_vcpu *vcpu)
1387 struct vmcb_control_area *control;
1389 control = &vcpu->svm->vmcb->control;
1390 control->int_vector = pop_irq(vcpu);
1391 control->int_ctl &= ~V_INTR_PRIO_MASK;
1392 control->int_ctl |= V_IRQ_MASK |
1393 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
1396 static void kvm_reput_irq(struct kvm_vcpu *vcpu)
1398 struct vmcb_control_area *control = &vcpu->svm->vmcb->control;
1400 if (control->int_ctl & V_IRQ_MASK) {
1401 control->int_ctl &= ~V_IRQ_MASK;
1402 push_irq(vcpu, control->int_vector);
1405 vcpu->interrupt_window_open =
1406 !(control->int_state & SVM_INTERRUPT_SHADOW_MASK);
1409 static void do_interrupt_requests(struct kvm_vcpu *vcpu,
1410 struct kvm_run *kvm_run)
1412 struct vmcb_control_area *control = &vcpu->svm->vmcb->control;
1414 vcpu->interrupt_window_open =
1415 (!(control->int_state & SVM_INTERRUPT_SHADOW_MASK) &&
1416 (vcpu->svm->vmcb->save.rflags & X86_EFLAGS_IF));
1418 if (vcpu->interrupt_window_open && vcpu->irq_summary)
1420 * If interrupts enabled, and not blocked by sti or mov ss. Good.
1422 kvm_do_inject_irq(vcpu);
1425 * Interrupts blocked. Wait for unblock.
1427 if (!vcpu->interrupt_window_open &&
1428 (vcpu->irq_summary || kvm_run->request_interrupt_window)) {
1429 control->intercept |= 1ULL << INTERCEPT_VINTR;
1430 } else
1431 control->intercept &= ~(1ULL << INTERCEPT_VINTR);
1434 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1435 struct kvm_run *kvm_run)
1437 kvm_run->ready_for_interrupt_injection = (vcpu->interrupt_window_open &&
1438 vcpu->irq_summary == 0);
1439 kvm_run->if_flag = (vcpu->svm->vmcb->save.rflags & X86_EFLAGS_IF) != 0;
1440 kvm_run->cr8 = vcpu->cr8;
1441 kvm_run->apic_base = vcpu->apic_base;
1445 * Check if userspace requested an interrupt window, and that the
1446 * interrupt window is open.
1448 * No need to exit to userspace if we already have an interrupt queued.
1450 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1451 struct kvm_run *kvm_run)
1453 return (!vcpu->irq_summary &&
1454 kvm_run->request_interrupt_window &&
1455 vcpu->interrupt_window_open &&
1456 (vcpu->svm->vmcb->save.rflags & X86_EFLAGS_IF));
1459 static void save_db_regs(unsigned long *db_regs)
1461 asm volatile ("mov %%dr0, %0" : "=r"(db_regs[0]));
1462 asm volatile ("mov %%dr1, %0" : "=r"(db_regs[1]));
1463 asm volatile ("mov %%dr2, %0" : "=r"(db_regs[2]));
1464 asm volatile ("mov %%dr3, %0" : "=r"(db_regs[3]));
1467 static void load_db_regs(unsigned long *db_regs)
1469 asm volatile ("mov %0, %%dr0" : : "r"(db_regs[0]));
1470 asm volatile ("mov %0, %%dr1" : : "r"(db_regs[1]));
1471 asm volatile ("mov %0, %%dr2" : : "r"(db_regs[2]));
1472 asm volatile ("mov %0, %%dr3" : : "r"(db_regs[3]));
1475 static int svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1477 u16 fs_selector;
1478 u16 gs_selector;
1479 u16 ldt_selector;
1480 int r;
1482 again:
1483 if (!vcpu->mmio_read_completed)
1484 do_interrupt_requests(vcpu, kvm_run);
1486 clgi();
1488 pre_svm_run(vcpu);
1490 save_host_msrs(vcpu);
1491 fs_selector = read_fs();
1492 gs_selector = read_gs();
1493 ldt_selector = read_ldt();
1494 vcpu->svm->host_cr2 = kvm_read_cr2();
1495 vcpu->svm->host_dr6 = read_dr6();
1496 vcpu->svm->host_dr7 = read_dr7();
1497 vcpu->svm->vmcb->save.cr2 = vcpu->cr2;
1499 if (vcpu->svm->vmcb->save.dr7 & 0xff) {
1500 write_dr7(0);
1501 save_db_regs(vcpu->svm->host_db_regs);
1502 load_db_regs(vcpu->svm->db_regs);
1505 if (vcpu->fpu_active) {
1506 fx_save(vcpu->host_fx_image);
1507 fx_restore(vcpu->guest_fx_image);
1510 asm volatile (
1511 #ifdef CONFIG_X86_64
1512 "push %%rbx; push %%rcx; push %%rdx;"
1513 "push %%rsi; push %%rdi; push %%rbp;"
1514 "push %%r8; push %%r9; push %%r10; push %%r11;"
1515 "push %%r12; push %%r13; push %%r14; push %%r15;"
1516 #else
1517 "push %%ebx; push %%ecx; push %%edx;"
1518 "push %%esi; push %%edi; push %%ebp;"
1519 #endif
1521 #ifdef CONFIG_X86_64
1522 "mov %c[rbx](%[vcpu]), %%rbx \n\t"
1523 "mov %c[rcx](%[vcpu]), %%rcx \n\t"
1524 "mov %c[rdx](%[vcpu]), %%rdx \n\t"
1525 "mov %c[rsi](%[vcpu]), %%rsi \n\t"
1526 "mov %c[rdi](%[vcpu]), %%rdi \n\t"
1527 "mov %c[rbp](%[vcpu]), %%rbp \n\t"
1528 "mov %c[r8](%[vcpu]), %%r8 \n\t"
1529 "mov %c[r9](%[vcpu]), %%r9 \n\t"
1530 "mov %c[r10](%[vcpu]), %%r10 \n\t"
1531 "mov %c[r11](%[vcpu]), %%r11 \n\t"
1532 "mov %c[r12](%[vcpu]), %%r12 \n\t"
1533 "mov %c[r13](%[vcpu]), %%r13 \n\t"
1534 "mov %c[r14](%[vcpu]), %%r14 \n\t"
1535 "mov %c[r15](%[vcpu]), %%r15 \n\t"
1536 #else
1537 "mov %c[rbx](%[vcpu]), %%ebx \n\t"
1538 "mov %c[rcx](%[vcpu]), %%ecx \n\t"
1539 "mov %c[rdx](%[vcpu]), %%edx \n\t"
1540 "mov %c[rsi](%[vcpu]), %%esi \n\t"
1541 "mov %c[rdi](%[vcpu]), %%edi \n\t"
1542 "mov %c[rbp](%[vcpu]), %%ebp \n\t"
1543 #endif
1545 #ifdef CONFIG_X86_64
1546 /* Enter guest mode */
1547 "push %%rax \n\t"
1548 "mov %c[svm](%[vcpu]), %%rax \n\t"
1549 "mov %c[vmcb](%%rax), %%rax \n\t"
1550 SVM_VMLOAD "\n\t"
1551 SVM_VMRUN "\n\t"
1552 SVM_VMSAVE "\n\t"
1553 "pop %%rax \n\t"
1554 #else
1555 /* Enter guest mode */
1556 "push %%eax \n\t"
1557 "mov %c[svm](%[vcpu]), %%eax \n\t"
1558 "mov %c[vmcb](%%eax), %%eax \n\t"
1559 SVM_VMLOAD "\n\t"
1560 SVM_VMRUN "\n\t"
1561 SVM_VMSAVE "\n\t"
1562 "pop %%eax \n\t"
1563 #endif
1565 /* Save guest registers, load host registers */
1566 #ifdef CONFIG_X86_64
1567 "mov %%rbx, %c[rbx](%[vcpu]) \n\t"
1568 "mov %%rcx, %c[rcx](%[vcpu]) \n\t"
1569 "mov %%rdx, %c[rdx](%[vcpu]) \n\t"
1570 "mov %%rsi, %c[rsi](%[vcpu]) \n\t"
1571 "mov %%rdi, %c[rdi](%[vcpu]) \n\t"
1572 "mov %%rbp, %c[rbp](%[vcpu]) \n\t"
1573 "mov %%r8, %c[r8](%[vcpu]) \n\t"
1574 "mov %%r9, %c[r9](%[vcpu]) \n\t"
1575 "mov %%r10, %c[r10](%[vcpu]) \n\t"
1576 "mov %%r11, %c[r11](%[vcpu]) \n\t"
1577 "mov %%r12, %c[r12](%[vcpu]) \n\t"
1578 "mov %%r13, %c[r13](%[vcpu]) \n\t"
1579 "mov %%r14, %c[r14](%[vcpu]) \n\t"
1580 "mov %%r15, %c[r15](%[vcpu]) \n\t"
1582 "pop %%r15; pop %%r14; pop %%r13; pop %%r12;"
1583 "pop %%r11; pop %%r10; pop %%r9; pop %%r8;"
1584 "pop %%rbp; pop %%rdi; pop %%rsi;"
1585 "pop %%rdx; pop %%rcx; pop %%rbx; \n\t"
1586 #else
1587 "mov %%ebx, %c[rbx](%[vcpu]) \n\t"
1588 "mov %%ecx, %c[rcx](%[vcpu]) \n\t"
1589 "mov %%edx, %c[rdx](%[vcpu]) \n\t"
1590 "mov %%esi, %c[rsi](%[vcpu]) \n\t"
1591 "mov %%edi, %c[rdi](%[vcpu]) \n\t"
1592 "mov %%ebp, %c[rbp](%[vcpu]) \n\t"
1594 "pop %%ebp; pop %%edi; pop %%esi;"
1595 "pop %%edx; pop %%ecx; pop %%ebx; \n\t"
1596 #endif
1598 : [vcpu]"a"(vcpu),
1599 [svm]"i"(offsetof(struct kvm_vcpu, svm)),
1600 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
1601 [rbx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RBX])),
1602 [rcx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RCX])),
1603 [rdx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RDX])),
1604 [rsi]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RSI])),
1605 [rdi]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RDI])),
1606 [rbp]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RBP]))
1607 #ifdef CONFIG_X86_64
1608 ,[r8 ]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R8 ])),
1609 [r9 ]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R9 ])),
1610 [r10]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R10])),
1611 [r11]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R11])),
1612 [r12]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R12])),
1613 [r13]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R13])),
1614 [r14]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R14])),
1615 [r15]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R15]))
1616 #endif
1617 : "cc", "memory" );
1619 if (vcpu->fpu_active) {
1620 fx_save(vcpu->guest_fx_image);
1621 fx_restore(vcpu->host_fx_image);
1624 if ((vcpu->svm->vmcb->save.dr7 & 0xff))
1625 load_db_regs(vcpu->svm->host_db_regs);
1627 vcpu->cr2 = vcpu->svm->vmcb->save.cr2;
1629 write_dr6(vcpu->svm->host_dr6);
1630 write_dr7(vcpu->svm->host_dr7);
1631 kvm_write_cr2(vcpu->svm->host_cr2);
1633 load_fs(fs_selector);
1634 load_gs(gs_selector);
1635 load_ldt(ldt_selector);
1636 load_host_msrs(vcpu);
1638 reload_tss(vcpu);
1641 * Profile KVM exit RIPs:
1643 if (unlikely(prof_on == KVM_PROFILING))
1644 profile_hit(KVM_PROFILING,
1645 (void *)(unsigned long)vcpu->svm->vmcb->save.rip);
1647 stgi();
1649 kvm_reput_irq(vcpu);
1651 vcpu->svm->next_rip = 0;
1653 if (vcpu->svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
1654 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
1655 kvm_run->fail_entry.hardware_entry_failure_reason
1656 = vcpu->svm->vmcb->control.exit_code;
1657 post_kvm_run_save(vcpu, kvm_run);
1658 return 0;
1661 r = handle_exit(vcpu, kvm_run);
1662 if (r > 0) {
1663 if (signal_pending(current)) {
1664 ++vcpu->stat.signal_exits;
1665 post_kvm_run_save(vcpu, kvm_run);
1666 kvm_run->exit_reason = KVM_EXIT_INTR;
1667 return -EINTR;
1670 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
1671 ++vcpu->stat.request_irq_exits;
1672 post_kvm_run_save(vcpu, kvm_run);
1673 kvm_run->exit_reason = KVM_EXIT_INTR;
1674 return -EINTR;
1676 kvm_resched(vcpu);
1677 goto again;
1679 post_kvm_run_save(vcpu, kvm_run);
1680 return r;
1683 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
1685 force_new_asid(vcpu);
1688 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
1690 vcpu->svm->vmcb->save.cr3 = root;
1691 force_new_asid(vcpu);
1693 if (vcpu->fpu_active) {
1694 vcpu->svm->vmcb->control.intercept_exceptions |= (1 << NM_VECTOR);
1695 vcpu->svm->vmcb->save.cr0 |= CR0_TS_MASK;
1696 vcpu->fpu_active = 0;
1700 static void svm_inject_page_fault(struct kvm_vcpu *vcpu,
1701 unsigned long addr,
1702 uint32_t err_code)
1704 uint32_t exit_int_info = vcpu->svm->vmcb->control.exit_int_info;
1706 ++vcpu->stat.pf_guest;
1708 if (is_page_fault(exit_int_info)) {
1710 vcpu->svm->vmcb->control.event_inj_err = 0;
1711 vcpu->svm->vmcb->control.event_inj = SVM_EVTINJ_VALID |
1712 SVM_EVTINJ_VALID_ERR |
1713 SVM_EVTINJ_TYPE_EXEPT |
1714 DF_VECTOR;
1715 return;
1717 vcpu->cr2 = addr;
1718 vcpu->svm->vmcb->save.cr2 = addr;
1719 vcpu->svm->vmcb->control.event_inj = SVM_EVTINJ_VALID |
1720 SVM_EVTINJ_VALID_ERR |
1721 SVM_EVTINJ_TYPE_EXEPT |
1722 PF_VECTOR;
1723 vcpu->svm->vmcb->control.event_inj_err = err_code;
1727 static int is_disabled(void)
1729 return 0;
1732 static void
1733 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
1736 * Patch in the VMMCALL instruction:
1738 hypercall[0] = 0x0f;
1739 hypercall[1] = 0x01;
1740 hypercall[2] = 0xd9;
1741 hypercall[3] = 0xc3;
1744 static struct kvm_arch_ops svm_arch_ops = {
1745 .cpu_has_kvm_support = has_svm,
1746 .disabled_by_bios = is_disabled,
1747 .hardware_setup = svm_hardware_setup,
1748 .hardware_unsetup = svm_hardware_unsetup,
1749 .hardware_enable = svm_hardware_enable,
1750 .hardware_disable = svm_hardware_disable,
1752 .vcpu_create = svm_create_vcpu,
1753 .vcpu_free = svm_free_vcpu,
1755 .vcpu_load = svm_vcpu_load,
1756 .vcpu_put = svm_vcpu_put,
1757 .vcpu_decache = svm_vcpu_decache,
1759 .set_guest_debug = svm_guest_debug,
1760 .get_msr = svm_get_msr,
1761 .set_msr = svm_set_msr,
1762 .get_segment_base = svm_get_segment_base,
1763 .get_segment = svm_get_segment,
1764 .set_segment = svm_set_segment,
1765 .get_cs_db_l_bits = svm_get_cs_db_l_bits,
1766 .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
1767 .set_cr0 = svm_set_cr0,
1768 .set_cr3 = svm_set_cr3,
1769 .set_cr4 = svm_set_cr4,
1770 .set_efer = svm_set_efer,
1771 .get_idt = svm_get_idt,
1772 .set_idt = svm_set_idt,
1773 .get_gdt = svm_get_gdt,
1774 .set_gdt = svm_set_gdt,
1775 .get_dr = svm_get_dr,
1776 .set_dr = svm_set_dr,
1777 .cache_regs = svm_cache_regs,
1778 .decache_regs = svm_decache_regs,
1779 .get_rflags = svm_get_rflags,
1780 .set_rflags = svm_set_rflags,
1782 .invlpg = svm_invlpg,
1783 .tlb_flush = svm_flush_tlb,
1784 .inject_page_fault = svm_inject_page_fault,
1786 .inject_gp = svm_inject_gp,
1788 .run = svm_vcpu_run,
1789 .skip_emulated_instruction = skip_emulated_instruction,
1790 .vcpu_setup = svm_vcpu_setup,
1791 .patch_hypercall = svm_patch_hypercall,
1794 static int __init svm_init(void)
1796 return kvm_init_arch(&svm_arch_ops, THIS_MODULE);
1799 static void __exit svm_exit(void)
1801 kvm_exit_arch();
1804 module_init(svm_init)
1805 module_exit(svm_exit)