2 * Kernel-based Virtual Machine driver for Linux
6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
10 * Yaniv Kamay <yaniv@qumranet.com>
11 * Avi Kivity <avi@qumranet.com>
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
17 #include <linux/kvm_host.h>
21 #include "kvm_cache_regs.h"
25 #include <linux/module.h>
26 #include <linux/mod_devicetable.h>
27 #include <linux/kernel.h>
28 #include <linux/vmalloc.h>
29 #include <linux/highmem.h>
30 #include <linux/sched.h>
31 #include <linux/ftrace_event.h>
32 #include <linux/slab.h>
34 #include <asm/perf_event.h>
35 #include <asm/tlbflush.h>
37 #include <asm/debugreg.h>
38 #include <asm/kvm_para.h>
40 #include <asm/virtext.h>
43 #define __ex(x) __kvm_handle_fault_on_reboot(x)
45 MODULE_AUTHOR("Qumranet");
46 MODULE_LICENSE("GPL");
48 static const struct x86_cpu_id svm_cpu_id[] = {
49 X86_FEATURE_MATCH(X86_FEATURE_SVM),
52 MODULE_DEVICE_TABLE(x86cpu, svm_cpu_id);
54 #define IOPM_ALLOC_ORDER 2
55 #define MSRPM_ALLOC_ORDER 1
57 #define SEG_TYPE_LDT 2
58 #define SEG_TYPE_BUSY_TSS16 3
60 #define SVM_FEATURE_NPT (1 << 0)
61 #define SVM_FEATURE_LBRV (1 << 1)
62 #define SVM_FEATURE_SVML (1 << 2)
63 #define SVM_FEATURE_NRIP (1 << 3)
64 #define SVM_FEATURE_TSC_RATE (1 << 4)
65 #define SVM_FEATURE_VMCB_CLEAN (1 << 5)
66 #define SVM_FEATURE_FLUSH_ASID (1 << 6)
67 #define SVM_FEATURE_DECODE_ASSIST (1 << 7)
68 #define SVM_FEATURE_PAUSE_FILTER (1 << 10)
70 #define NESTED_EXIT_HOST 0 /* Exit handled on host level */
71 #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */
72 #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */
74 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
76 #define TSC_RATIO_RSVD 0xffffff0000000000ULL
77 #define TSC_RATIO_MIN 0x0000000000000001ULL
78 #define TSC_RATIO_MAX 0x000000ffffffffffULL
80 static bool erratum_383_found __read_mostly;
82 static const u32 host_save_user_msrs[] = {
84 MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE,
87 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
90 #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
100 /* These are the merged vectors */
103 /* gpa pointers to the real vectors */
107 /* A VMEXIT is required but not yet emulated */
110 /* cache for intercepts of the guest */
113 u32 intercept_exceptions;
116 /* Nested Paging related state */
120 #define MSRPM_OFFSETS 16
121 static u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
124 * Set osvw_len to higher value when updated Revision Guides
125 * are published and we know what the new status bits are
127 static uint64_t osvw_len = 4, osvw_status;
130 struct kvm_vcpu vcpu;
132 unsigned long vmcb_pa;
133 struct svm_cpu_data *svm_data;
134 uint64_t asid_generation;
135 uint64_t sysenter_esp;
136 uint64_t sysenter_eip;
140 u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS];
152 struct nested_state nested;
156 unsigned int3_injected;
157 unsigned long int3_rip;
163 static DEFINE_PER_CPU(u64, current_tsc_ratio);
164 #define TSC_RATIO_DEFAULT 0x0100000000ULL
166 #define MSR_INVALID 0xffffffffU
168 static const struct svm_direct_access_msrs {
169 u32 index; /* Index of the MSR */
170 bool always; /* True if intercept is always on */
171 } direct_access_msrs[] = {
172 { .index = MSR_STAR, .always = true },
173 { .index = MSR_IA32_SYSENTER_CS, .always = true },
175 { .index = MSR_GS_BASE, .always = true },
176 { .index = MSR_FS_BASE, .always = true },
177 { .index = MSR_KERNEL_GS_BASE, .always = true },
178 { .index = MSR_LSTAR, .always = true },
179 { .index = MSR_CSTAR, .always = true },
180 { .index = MSR_SYSCALL_MASK, .always = true },
182 { .index = MSR_IA32_LASTBRANCHFROMIP, .always = false },
183 { .index = MSR_IA32_LASTBRANCHTOIP, .always = false },
184 { .index = MSR_IA32_LASTINTFROMIP, .always = false },
185 { .index = MSR_IA32_LASTINTTOIP, .always = false },
186 { .index = MSR_INVALID, .always = false },
189 /* enable NPT for AMD64 and X86 with PAE */
190 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
191 static bool npt_enabled = true;
193 static bool npt_enabled;
196 /* allow nested paging (virtualized MMU) for all guests */
197 static int npt = true;
198 module_param(npt, int, S_IRUGO);
200 /* allow nested virtualization in KVM/SVM */
201 static int nested = true;
202 module_param(nested, int, S_IRUGO);
204 static void svm_flush_tlb(struct kvm_vcpu *vcpu);
205 static void svm_complete_interrupts(struct vcpu_svm *svm);
207 static int nested_svm_exit_handled(struct vcpu_svm *svm);
208 static int nested_svm_intercept(struct vcpu_svm *svm);
209 static int nested_svm_vmexit(struct vcpu_svm *svm);
210 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
211 bool has_error_code, u32 error_code);
212 static u64 __scale_tsc(u64 ratio, u64 tsc);
215 VMCB_INTERCEPTS, /* Intercept vectors, TSC offset,
216 pause filter count */
217 VMCB_PERM_MAP, /* IOPM Base and MSRPM Base */
218 VMCB_ASID, /* ASID */
219 VMCB_INTR, /* int_ctl, int_vector */
220 VMCB_NPT, /* npt_en, nCR3, gPAT */
221 VMCB_CR, /* CR0, CR3, CR4, EFER */
222 VMCB_DR, /* DR6, DR7 */
223 VMCB_DT, /* GDT, IDT */
224 VMCB_SEG, /* CS, DS, SS, ES, CPL */
225 VMCB_CR2, /* CR2 only */
226 VMCB_LBR, /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */
230 /* TPR and CR2 are always written before VMRUN */
231 #define VMCB_ALWAYS_DIRTY_MASK ((1U << VMCB_INTR) | (1U << VMCB_CR2))
233 static inline void mark_all_dirty(struct vmcb *vmcb)
235 vmcb->control.clean = 0;
238 static inline void mark_all_clean(struct vmcb *vmcb)
240 vmcb->control.clean = ((1 << VMCB_DIRTY_MAX) - 1)
241 & ~VMCB_ALWAYS_DIRTY_MASK;
244 static inline void mark_dirty(struct vmcb *vmcb, int bit)
246 vmcb->control.clean &= ~(1 << bit);
249 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
251 return container_of(vcpu, struct vcpu_svm, vcpu);
254 static void recalc_intercepts(struct vcpu_svm *svm)
256 struct vmcb_control_area *c, *h;
257 struct nested_state *g;
259 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
261 if (!is_guest_mode(&svm->vcpu))
264 c = &svm->vmcb->control;
265 h = &svm->nested.hsave->control;
268 c->intercept_cr = h->intercept_cr | g->intercept_cr;
269 c->intercept_dr = h->intercept_dr | g->intercept_dr;
270 c->intercept_exceptions = h->intercept_exceptions | g->intercept_exceptions;
271 c->intercept = h->intercept | g->intercept;
274 static inline struct vmcb *get_host_vmcb(struct vcpu_svm *svm)
276 if (is_guest_mode(&svm->vcpu))
277 return svm->nested.hsave;
282 static inline void set_cr_intercept(struct vcpu_svm *svm, int bit)
284 struct vmcb *vmcb = get_host_vmcb(svm);
286 vmcb->control.intercept_cr |= (1U << bit);
288 recalc_intercepts(svm);
291 static inline void clr_cr_intercept(struct vcpu_svm *svm, int bit)
293 struct vmcb *vmcb = get_host_vmcb(svm);
295 vmcb->control.intercept_cr &= ~(1U << bit);
297 recalc_intercepts(svm);
300 static inline bool is_cr_intercept(struct vcpu_svm *svm, int bit)
302 struct vmcb *vmcb = get_host_vmcb(svm);
304 return vmcb->control.intercept_cr & (1U << bit);
307 static inline void set_dr_intercepts(struct vcpu_svm *svm)
309 struct vmcb *vmcb = get_host_vmcb(svm);
311 vmcb->control.intercept_dr = (1 << INTERCEPT_DR0_READ)
312 | (1 << INTERCEPT_DR1_READ)
313 | (1 << INTERCEPT_DR2_READ)
314 | (1 << INTERCEPT_DR3_READ)
315 | (1 << INTERCEPT_DR4_READ)
316 | (1 << INTERCEPT_DR5_READ)
317 | (1 << INTERCEPT_DR6_READ)
318 | (1 << INTERCEPT_DR7_READ)
319 | (1 << INTERCEPT_DR0_WRITE)
320 | (1 << INTERCEPT_DR1_WRITE)
321 | (1 << INTERCEPT_DR2_WRITE)
322 | (1 << INTERCEPT_DR3_WRITE)
323 | (1 << INTERCEPT_DR4_WRITE)
324 | (1 << INTERCEPT_DR5_WRITE)
325 | (1 << INTERCEPT_DR6_WRITE)
326 | (1 << INTERCEPT_DR7_WRITE);
328 recalc_intercepts(svm);
331 static inline void clr_dr_intercepts(struct vcpu_svm *svm)
333 struct vmcb *vmcb = get_host_vmcb(svm);
335 vmcb->control.intercept_dr = 0;
337 recalc_intercepts(svm);
340 static inline void set_exception_intercept(struct vcpu_svm *svm, int bit)
342 struct vmcb *vmcb = get_host_vmcb(svm);
344 vmcb->control.intercept_exceptions |= (1U << bit);
346 recalc_intercepts(svm);
349 static inline void clr_exception_intercept(struct vcpu_svm *svm, int bit)
351 struct vmcb *vmcb = get_host_vmcb(svm);
353 vmcb->control.intercept_exceptions &= ~(1U << bit);
355 recalc_intercepts(svm);
358 static inline void set_intercept(struct vcpu_svm *svm, int bit)
360 struct vmcb *vmcb = get_host_vmcb(svm);
362 vmcb->control.intercept |= (1ULL << bit);
364 recalc_intercepts(svm);
367 static inline void clr_intercept(struct vcpu_svm *svm, int bit)
369 struct vmcb *vmcb = get_host_vmcb(svm);
371 vmcb->control.intercept &= ~(1ULL << bit);
373 recalc_intercepts(svm);
376 static inline void enable_gif(struct vcpu_svm *svm)
378 svm->vcpu.arch.hflags |= HF_GIF_MASK;
381 static inline void disable_gif(struct vcpu_svm *svm)
383 svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
386 static inline bool gif_set(struct vcpu_svm *svm)
388 return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
391 static unsigned long iopm_base;
393 struct kvm_ldttss_desc {
396 unsigned base1:8, type:5, dpl:2, p:1;
397 unsigned limit1:4, zero0:3, g:1, base2:8;
400 } __attribute__((packed));
402 struct svm_cpu_data {
408 struct kvm_ldttss_desc *tss_desc;
410 struct page *save_area;
413 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
415 struct svm_init_data {
420 static const u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
422 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
423 #define MSRS_RANGE_SIZE 2048
424 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
426 static u32 svm_msrpm_offset(u32 msr)
431 for (i = 0; i < NUM_MSR_MAPS; i++) {
432 if (msr < msrpm_ranges[i] ||
433 msr >= msrpm_ranges[i] + MSRS_IN_RANGE)
436 offset = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */
437 offset += (i * MSRS_RANGE_SIZE); /* add range offset */
439 /* Now we have the u8 offset - but need the u32 offset */
443 /* MSR not in any range */
447 #define MAX_INST_SIZE 15
449 static inline void clgi(void)
451 asm volatile (__ex(SVM_CLGI));
454 static inline void stgi(void)
456 asm volatile (__ex(SVM_STGI));
459 static inline void invlpga(unsigned long addr, u32 asid)
461 asm volatile (__ex(SVM_INVLPGA) : : "a"(addr), "c"(asid));
464 static int get_npt_level(void)
467 return PT64_ROOT_LEVEL;
469 return PT32E_ROOT_LEVEL;
473 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
475 vcpu->arch.efer = efer;
476 if (!npt_enabled && !(efer & EFER_LMA))
479 to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME;
480 mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
483 static int is_external_interrupt(u32 info)
485 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
486 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
489 static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu)
491 struct vcpu_svm *svm = to_svm(vcpu);
494 if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
495 ret = KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS;
499 static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
501 struct vcpu_svm *svm = to_svm(vcpu);
504 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
506 svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
510 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
512 struct vcpu_svm *svm = to_svm(vcpu);
514 if (svm->vmcb->control.next_rip != 0)
515 svm->next_rip = svm->vmcb->control.next_rip;
517 if (!svm->next_rip) {
518 if (emulate_instruction(vcpu, EMULTYPE_SKIP) !=
520 printk(KERN_DEBUG "%s: NOP\n", __func__);
523 if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE)
524 printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
525 __func__, kvm_rip_read(vcpu), svm->next_rip);
527 kvm_rip_write(vcpu, svm->next_rip);
528 svm_set_interrupt_shadow(vcpu, 0);
531 static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
532 bool has_error_code, u32 error_code,
535 struct vcpu_svm *svm = to_svm(vcpu);
538 * If we are within a nested VM we'd better #VMEXIT and let the guest
539 * handle the exception
542 nested_svm_check_exception(svm, nr, has_error_code, error_code))
545 if (nr == BP_VECTOR && !static_cpu_has(X86_FEATURE_NRIPS)) {
546 unsigned long rip, old_rip = kvm_rip_read(&svm->vcpu);
549 * For guest debugging where we have to reinject #BP if some
550 * INT3 is guest-owned:
551 * Emulate nRIP by moving RIP forward. Will fail if injection
552 * raises a fault that is not intercepted. Still better than
553 * failing in all cases.
555 skip_emulated_instruction(&svm->vcpu);
556 rip = kvm_rip_read(&svm->vcpu);
557 svm->int3_rip = rip + svm->vmcb->save.cs.base;
558 svm->int3_injected = rip - old_rip;
561 svm->vmcb->control.event_inj = nr
563 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
564 | SVM_EVTINJ_TYPE_EXEPT;
565 svm->vmcb->control.event_inj_err = error_code;
568 static void svm_init_erratum_383(void)
574 if (!static_cpu_has_bug(X86_BUG_AMD_TLB_MMATCH))
577 /* Use _safe variants to not break nested virtualization */
578 val = native_read_msr_safe(MSR_AMD64_DC_CFG, &err);
584 low = lower_32_bits(val);
585 high = upper_32_bits(val);
587 native_write_msr_safe(MSR_AMD64_DC_CFG, low, high);
589 erratum_383_found = true;
592 static void svm_init_osvw(struct kvm_vcpu *vcpu)
595 * Guests should see errata 400 and 415 as fixed (assuming that
596 * HLT and IO instructions are intercepted).
598 vcpu->arch.osvw.length = (osvw_len >= 3) ? (osvw_len) : 3;
599 vcpu->arch.osvw.status = osvw_status & ~(6ULL);
602 * By increasing VCPU's osvw.length to 3 we are telling the guest that
603 * all osvw.status bits inside that length, including bit 0 (which is
604 * reserved for erratum 298), are valid. However, if host processor's
605 * osvw_len is 0 then osvw_status[0] carries no information. We need to
606 * be conservative here and therefore we tell the guest that erratum 298
607 * is present (because we really don't know).
609 if (osvw_len == 0 && boot_cpu_data.x86 == 0x10)
610 vcpu->arch.osvw.status |= 1;
613 static int has_svm(void)
617 if (!cpu_has_svm(&msg)) {
618 printk(KERN_INFO "has_svm: %s\n", msg);
625 static void svm_hardware_disable(void)
627 /* Make sure we clean up behind us */
628 if (static_cpu_has(X86_FEATURE_TSCRATEMSR))
629 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
633 amd_pmu_disable_virt();
636 static int svm_hardware_enable(void)
639 struct svm_cpu_data *sd;
641 struct desc_ptr gdt_descr;
642 struct desc_struct *gdt;
643 int me = raw_smp_processor_id();
645 rdmsrl(MSR_EFER, efer);
646 if (efer & EFER_SVME)
650 pr_err("%s: err EOPNOTSUPP on %d\n", __func__, me);
653 sd = per_cpu(svm_data, me);
655 pr_err("%s: svm_data is NULL on %d\n", __func__, me);
659 sd->asid_generation = 1;
660 sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
661 sd->next_asid = sd->max_asid + 1;
663 native_store_gdt(&gdt_descr);
664 gdt = (struct desc_struct *)gdt_descr.address;
665 sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
667 wrmsrl(MSR_EFER, efer | EFER_SVME);
669 wrmsrl(MSR_VM_HSAVE_PA, page_to_pfn(sd->save_area) << PAGE_SHIFT);
671 if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
672 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
673 __this_cpu_write(current_tsc_ratio, TSC_RATIO_DEFAULT);
680 * Note that it is possible to have a system with mixed processor
681 * revisions and therefore different OSVW bits. If bits are not the same
682 * on different processors then choose the worst case (i.e. if erratum
683 * is present on one processor and not on another then assume that the
684 * erratum is present everywhere).
686 if (cpu_has(&boot_cpu_data, X86_FEATURE_OSVW)) {
687 uint64_t len, status = 0;
690 len = native_read_msr_safe(MSR_AMD64_OSVW_ID_LENGTH, &err);
692 status = native_read_msr_safe(MSR_AMD64_OSVW_STATUS,
696 osvw_status = osvw_len = 0;
700 osvw_status |= status;
701 osvw_status &= (1ULL << osvw_len) - 1;
704 osvw_status = osvw_len = 0;
706 svm_init_erratum_383();
708 amd_pmu_enable_virt();
713 static void svm_cpu_uninit(int cpu)
715 struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id());
720 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
721 __free_page(sd->save_area);
725 static int svm_cpu_init(int cpu)
727 struct svm_cpu_data *sd;
730 sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
734 sd->save_area = alloc_page(GFP_KERNEL);
739 per_cpu(svm_data, cpu) = sd;
749 static bool valid_msr_intercept(u32 index)
753 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++)
754 if (direct_access_msrs[i].index == index)
760 static void set_msr_interception(u32 *msrpm, unsigned msr,
763 u8 bit_read, bit_write;
768 * If this warning triggers extend the direct_access_msrs list at the
769 * beginning of the file
771 WARN_ON(!valid_msr_intercept(msr));
773 offset = svm_msrpm_offset(msr);
774 bit_read = 2 * (msr & 0x0f);
775 bit_write = 2 * (msr & 0x0f) + 1;
778 BUG_ON(offset == MSR_INVALID);
780 read ? clear_bit(bit_read, &tmp) : set_bit(bit_read, &tmp);
781 write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp);
786 static void svm_vcpu_init_msrpm(u32 *msrpm)
790 memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
792 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
793 if (!direct_access_msrs[i].always)
796 set_msr_interception(msrpm, direct_access_msrs[i].index, 1, 1);
800 static void add_msr_offset(u32 offset)
804 for (i = 0; i < MSRPM_OFFSETS; ++i) {
806 /* Offset already in list? */
807 if (msrpm_offsets[i] == offset)
810 /* Slot used by another offset? */
811 if (msrpm_offsets[i] != MSR_INVALID)
814 /* Add offset to list */
815 msrpm_offsets[i] = offset;
821 * If this BUG triggers the msrpm_offsets table has an overflow. Just
822 * increase MSRPM_OFFSETS in this case.
827 static void init_msrpm_offsets(void)
831 memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets));
833 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
836 offset = svm_msrpm_offset(direct_access_msrs[i].index);
837 BUG_ON(offset == MSR_INVALID);
839 add_msr_offset(offset);
843 static void svm_enable_lbrv(struct vcpu_svm *svm)
845 u32 *msrpm = svm->msrpm;
847 svm->vmcb->control.lbr_ctl = 1;
848 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
849 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
850 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
851 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
854 static void svm_disable_lbrv(struct vcpu_svm *svm)
856 u32 *msrpm = svm->msrpm;
858 svm->vmcb->control.lbr_ctl = 0;
859 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
860 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
861 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
862 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
865 static __init int svm_hardware_setup(void)
868 struct page *iopm_pages;
872 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
877 iopm_va = page_address(iopm_pages);
878 memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
879 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
881 init_msrpm_offsets();
883 if (boot_cpu_has(X86_FEATURE_NX))
884 kvm_enable_efer_bits(EFER_NX);
886 if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
887 kvm_enable_efer_bits(EFER_FFXSR);
889 if (boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
892 kvm_has_tsc_control = true;
895 * Make sure the user can only configure tsc_khz values that
896 * fit into a signed integer.
897 * A min value is not calculated needed because it will always
898 * be 1 on all machines and a value of 0 is used to disable
899 * tsc-scaling for the vcpu.
901 max = min(0x7fffffffULL, __scale_tsc(tsc_khz, TSC_RATIO_MAX));
903 kvm_max_guest_tsc_khz = max;
907 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
908 kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE);
911 for_each_possible_cpu(cpu) {
912 r = svm_cpu_init(cpu);
917 if (!boot_cpu_has(X86_FEATURE_NPT))
920 if (npt_enabled && !npt) {
921 printk(KERN_INFO "kvm: Nested Paging disabled\n");
926 printk(KERN_INFO "kvm: Nested Paging enabled\n");
934 __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
939 static __exit void svm_hardware_unsetup(void)
943 for_each_possible_cpu(cpu)
946 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
950 static void init_seg(struct vmcb_seg *seg)
953 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
954 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
959 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
962 seg->attrib = SVM_SELECTOR_P_MASK | type;
967 static u64 __scale_tsc(u64 ratio, u64 tsc)
969 u64 mult, frac, _tsc;
972 frac = ratio & ((1ULL << 32) - 1);
976 _tsc += (tsc >> 32) * frac;
977 _tsc += ((tsc & ((1ULL << 32) - 1)) * frac) >> 32;
982 static u64 svm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
984 struct vcpu_svm *svm = to_svm(vcpu);
987 if (svm->tsc_ratio != TSC_RATIO_DEFAULT)
988 _tsc = __scale_tsc(svm->tsc_ratio, tsc);
993 static void svm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
995 struct vcpu_svm *svm = to_svm(vcpu);
999 /* Guest TSC same frequency as host TSC? */
1001 svm->tsc_ratio = TSC_RATIO_DEFAULT;
1005 /* TSC scaling supported? */
1006 if (!boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
1007 if (user_tsc_khz > tsc_khz) {
1008 vcpu->arch.tsc_catchup = 1;
1009 vcpu->arch.tsc_always_catchup = 1;
1011 WARN(1, "user requested TSC rate below hardware speed\n");
1017 /* TSC scaling required - calculate ratio */
1019 do_div(ratio, tsc_khz);
1021 if (ratio == 0 || ratio & TSC_RATIO_RSVD) {
1022 WARN_ONCE(1, "Invalid TSC ratio - virtual-tsc-khz=%u\n",
1026 svm->tsc_ratio = ratio;
1029 static u64 svm_read_tsc_offset(struct kvm_vcpu *vcpu)
1031 struct vcpu_svm *svm = to_svm(vcpu);
1033 return svm->vmcb->control.tsc_offset;
1036 static void svm_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1038 struct vcpu_svm *svm = to_svm(vcpu);
1039 u64 g_tsc_offset = 0;
1041 if (is_guest_mode(vcpu)) {
1042 g_tsc_offset = svm->vmcb->control.tsc_offset -
1043 svm->nested.hsave->control.tsc_offset;
1044 svm->nested.hsave->control.tsc_offset = offset;
1046 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1047 svm->vmcb->control.tsc_offset,
1050 svm->vmcb->control.tsc_offset = offset + g_tsc_offset;
1052 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1055 static void svm_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment, bool host)
1057 struct vcpu_svm *svm = to_svm(vcpu);
1060 if (svm->tsc_ratio != TSC_RATIO_DEFAULT)
1061 WARN_ON(adjustment < 0);
1062 adjustment = svm_scale_tsc(vcpu, (u64)adjustment);
1065 svm->vmcb->control.tsc_offset += adjustment;
1066 if (is_guest_mode(vcpu))
1067 svm->nested.hsave->control.tsc_offset += adjustment;
1069 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1070 svm->vmcb->control.tsc_offset - adjustment,
1071 svm->vmcb->control.tsc_offset);
1073 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1076 static u64 svm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1080 tsc = svm_scale_tsc(vcpu, native_read_tsc());
1082 return target_tsc - tsc;
1085 static void init_vmcb(struct vcpu_svm *svm)
1087 struct vmcb_control_area *control = &svm->vmcb->control;
1088 struct vmcb_save_area *save = &svm->vmcb->save;
1090 svm->vcpu.fpu_active = 1;
1091 svm->vcpu.arch.hflags = 0;
1093 set_cr_intercept(svm, INTERCEPT_CR0_READ);
1094 set_cr_intercept(svm, INTERCEPT_CR3_READ);
1095 set_cr_intercept(svm, INTERCEPT_CR4_READ);
1096 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1097 set_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1098 set_cr_intercept(svm, INTERCEPT_CR4_WRITE);
1099 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
1101 set_dr_intercepts(svm);
1103 set_exception_intercept(svm, PF_VECTOR);
1104 set_exception_intercept(svm, UD_VECTOR);
1105 set_exception_intercept(svm, MC_VECTOR);
1107 set_intercept(svm, INTERCEPT_INTR);
1108 set_intercept(svm, INTERCEPT_NMI);
1109 set_intercept(svm, INTERCEPT_SMI);
1110 set_intercept(svm, INTERCEPT_SELECTIVE_CR0);
1111 set_intercept(svm, INTERCEPT_RDPMC);
1112 set_intercept(svm, INTERCEPT_CPUID);
1113 set_intercept(svm, INTERCEPT_INVD);
1114 set_intercept(svm, INTERCEPT_HLT);
1115 set_intercept(svm, INTERCEPT_INVLPG);
1116 set_intercept(svm, INTERCEPT_INVLPGA);
1117 set_intercept(svm, INTERCEPT_IOIO_PROT);
1118 set_intercept(svm, INTERCEPT_MSR_PROT);
1119 set_intercept(svm, INTERCEPT_TASK_SWITCH);
1120 set_intercept(svm, INTERCEPT_SHUTDOWN);
1121 set_intercept(svm, INTERCEPT_VMRUN);
1122 set_intercept(svm, INTERCEPT_VMMCALL);
1123 set_intercept(svm, INTERCEPT_VMLOAD);
1124 set_intercept(svm, INTERCEPT_VMSAVE);
1125 set_intercept(svm, INTERCEPT_STGI);
1126 set_intercept(svm, INTERCEPT_CLGI);
1127 set_intercept(svm, INTERCEPT_SKINIT);
1128 set_intercept(svm, INTERCEPT_WBINVD);
1129 set_intercept(svm, INTERCEPT_MONITOR);
1130 set_intercept(svm, INTERCEPT_MWAIT);
1131 set_intercept(svm, INTERCEPT_XSETBV);
1133 control->iopm_base_pa = iopm_base;
1134 control->msrpm_base_pa = __pa(svm->msrpm);
1135 control->int_ctl = V_INTR_MASKING_MASK;
1137 init_seg(&save->es);
1138 init_seg(&save->ss);
1139 init_seg(&save->ds);
1140 init_seg(&save->fs);
1141 init_seg(&save->gs);
1143 save->cs.selector = 0xf000;
1144 save->cs.base = 0xffff0000;
1145 /* Executable/Readable Code Segment */
1146 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
1147 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
1148 save->cs.limit = 0xffff;
1150 save->gdtr.limit = 0xffff;
1151 save->idtr.limit = 0xffff;
1153 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
1154 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
1156 svm_set_efer(&svm->vcpu, 0);
1157 save->dr6 = 0xffff0ff0;
1158 kvm_set_rflags(&svm->vcpu, 2);
1159 save->rip = 0x0000fff0;
1160 svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
1163 * This is the guest-visible cr0 value.
1164 * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
1166 svm->vcpu.arch.cr0 = 0;
1167 (void)kvm_set_cr0(&svm->vcpu, X86_CR0_NW | X86_CR0_CD | X86_CR0_ET);
1169 save->cr4 = X86_CR4_PAE;
1173 /* Setup VMCB for Nested Paging */
1174 control->nested_ctl = 1;
1175 clr_intercept(svm, INTERCEPT_INVLPG);
1176 clr_exception_intercept(svm, PF_VECTOR);
1177 clr_cr_intercept(svm, INTERCEPT_CR3_READ);
1178 clr_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1179 save->g_pat = 0x0007040600070406ULL;
1183 svm->asid_generation = 0;
1185 svm->nested.vmcb = 0;
1186 svm->vcpu.arch.hflags = 0;
1188 if (boot_cpu_has(X86_FEATURE_PAUSEFILTER)) {
1189 control->pause_filter_count = 3000;
1190 set_intercept(svm, INTERCEPT_PAUSE);
1193 mark_all_dirty(svm->vmcb);
1198 static void svm_vcpu_reset(struct kvm_vcpu *vcpu)
1200 struct vcpu_svm *svm = to_svm(vcpu);
1206 kvm_cpuid(vcpu, &eax, &dummy, &dummy, &dummy);
1207 kvm_register_write(vcpu, VCPU_REGS_RDX, eax);
1210 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
1212 struct vcpu_svm *svm;
1214 struct page *msrpm_pages;
1215 struct page *hsave_page;
1216 struct page *nested_msrpm_pages;
1219 svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1225 svm->tsc_ratio = TSC_RATIO_DEFAULT;
1227 err = kvm_vcpu_init(&svm->vcpu, kvm, id);
1232 page = alloc_page(GFP_KERNEL);
1236 msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
1240 nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
1241 if (!nested_msrpm_pages)
1244 hsave_page = alloc_page(GFP_KERNEL);
1248 svm->nested.hsave = page_address(hsave_page);
1250 svm->msrpm = page_address(msrpm_pages);
1251 svm_vcpu_init_msrpm(svm->msrpm);
1253 svm->nested.msrpm = page_address(nested_msrpm_pages);
1254 svm_vcpu_init_msrpm(svm->nested.msrpm);
1256 svm->vmcb = page_address(page);
1257 clear_page(svm->vmcb);
1258 svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
1259 svm->asid_generation = 0;
1262 svm->vcpu.arch.apic_base = APIC_DEFAULT_PHYS_BASE |
1263 MSR_IA32_APICBASE_ENABLE;
1264 if (kvm_vcpu_is_reset_bsp(&svm->vcpu))
1265 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
1267 svm_init_osvw(&svm->vcpu);
1272 __free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER);
1274 __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
1278 kvm_vcpu_uninit(&svm->vcpu);
1280 kmem_cache_free(kvm_vcpu_cache, svm);
1282 return ERR_PTR(err);
1285 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
1287 struct vcpu_svm *svm = to_svm(vcpu);
1289 __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
1290 __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
1291 __free_page(virt_to_page(svm->nested.hsave));
1292 __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER);
1293 kvm_vcpu_uninit(vcpu);
1294 kmem_cache_free(kvm_vcpu_cache, svm);
1297 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1299 struct vcpu_svm *svm = to_svm(vcpu);
1302 if (unlikely(cpu != vcpu->cpu)) {
1303 svm->asid_generation = 0;
1304 mark_all_dirty(svm->vmcb);
1307 #ifdef CONFIG_X86_64
1308 rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host.gs_base);
1310 savesegment(fs, svm->host.fs);
1311 savesegment(gs, svm->host.gs);
1312 svm->host.ldt = kvm_read_ldt();
1314 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1315 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1317 if (static_cpu_has(X86_FEATURE_TSCRATEMSR) &&
1318 svm->tsc_ratio != __this_cpu_read(current_tsc_ratio)) {
1319 __this_cpu_write(current_tsc_ratio, svm->tsc_ratio);
1320 wrmsrl(MSR_AMD64_TSC_RATIO, svm->tsc_ratio);
1324 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
1326 struct vcpu_svm *svm = to_svm(vcpu);
1329 ++vcpu->stat.host_state_reload;
1330 kvm_load_ldt(svm->host.ldt);
1331 #ifdef CONFIG_X86_64
1332 loadsegment(fs, svm->host.fs);
1333 wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gs);
1334 load_gs_index(svm->host.gs);
1336 #ifdef CONFIG_X86_32_LAZY_GS
1337 loadsegment(gs, svm->host.gs);
1340 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1341 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1344 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
1346 return to_svm(vcpu)->vmcb->save.rflags;
1349 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1352 * Any change of EFLAGS.VM is accompained by a reload of SS
1353 * (caused by either a task switch or an inter-privilege IRET),
1354 * so we do not need to update the CPL here.
1356 to_svm(vcpu)->vmcb->save.rflags = rflags;
1359 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1362 case VCPU_EXREG_PDPTR:
1363 BUG_ON(!npt_enabled);
1364 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
1371 static void svm_set_vintr(struct vcpu_svm *svm)
1373 set_intercept(svm, INTERCEPT_VINTR);
1376 static void svm_clear_vintr(struct vcpu_svm *svm)
1378 clr_intercept(svm, INTERCEPT_VINTR);
1381 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
1383 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1386 case VCPU_SREG_CS: return &save->cs;
1387 case VCPU_SREG_DS: return &save->ds;
1388 case VCPU_SREG_ES: return &save->es;
1389 case VCPU_SREG_FS: return &save->fs;
1390 case VCPU_SREG_GS: return &save->gs;
1391 case VCPU_SREG_SS: return &save->ss;
1392 case VCPU_SREG_TR: return &save->tr;
1393 case VCPU_SREG_LDTR: return &save->ldtr;
1399 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1401 struct vmcb_seg *s = svm_seg(vcpu, seg);
1406 static void svm_get_segment(struct kvm_vcpu *vcpu,
1407 struct kvm_segment *var, int seg)
1409 struct vmcb_seg *s = svm_seg(vcpu, seg);
1411 var->base = s->base;
1412 var->limit = s->limit;
1413 var->selector = s->selector;
1414 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
1415 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
1416 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
1417 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
1418 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
1419 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
1420 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
1423 * AMD CPUs circa 2014 track the G bit for all segments except CS.
1424 * However, the SVM spec states that the G bit is not observed by the
1425 * CPU, and some VMware virtual CPUs drop the G bit for all segments.
1426 * So let's synthesize a legal G bit for all segments, this helps
1427 * running KVM nested. It also helps cross-vendor migration, because
1428 * Intel's vmentry has a check on the 'G' bit.
1430 var->g = s->limit > 0xfffff;
1433 * AMD's VMCB does not have an explicit unusable field, so emulate it
1434 * for cross vendor migration purposes by "not present"
1436 var->unusable = !var->present || (var->type == 0);
1441 * Work around a bug where the busy flag in the tr selector
1451 * The accessed bit must always be set in the segment
1452 * descriptor cache, although it can be cleared in the
1453 * descriptor, the cached bit always remains at 1. Since
1454 * Intel has a check on this, set it here to support
1455 * cross-vendor migration.
1462 * On AMD CPUs sometimes the DB bit in the segment
1463 * descriptor is left as 1, although the whole segment has
1464 * been made unusable. Clear it here to pass an Intel VMX
1465 * entry check when cross vendor migrating.
1469 var->dpl = to_svm(vcpu)->vmcb->save.cpl;
1474 static int svm_get_cpl(struct kvm_vcpu *vcpu)
1476 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1481 static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1483 struct vcpu_svm *svm = to_svm(vcpu);
1485 dt->size = svm->vmcb->save.idtr.limit;
1486 dt->address = svm->vmcb->save.idtr.base;
1489 static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1491 struct vcpu_svm *svm = to_svm(vcpu);
1493 svm->vmcb->save.idtr.limit = dt->size;
1494 svm->vmcb->save.idtr.base = dt->address ;
1495 mark_dirty(svm->vmcb, VMCB_DT);
1498 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1500 struct vcpu_svm *svm = to_svm(vcpu);
1502 dt->size = svm->vmcb->save.gdtr.limit;
1503 dt->address = svm->vmcb->save.gdtr.base;
1506 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1508 struct vcpu_svm *svm = to_svm(vcpu);
1510 svm->vmcb->save.gdtr.limit = dt->size;
1511 svm->vmcb->save.gdtr.base = dt->address ;
1512 mark_dirty(svm->vmcb, VMCB_DT);
1515 static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
1519 static void svm_decache_cr3(struct kvm_vcpu *vcpu)
1523 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1527 static void update_cr0_intercept(struct vcpu_svm *svm)
1529 ulong gcr0 = svm->vcpu.arch.cr0;
1530 u64 *hcr0 = &svm->vmcb->save.cr0;
1532 if (!svm->vcpu.fpu_active)
1533 *hcr0 |= SVM_CR0_SELECTIVE_MASK;
1535 *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
1536 | (gcr0 & SVM_CR0_SELECTIVE_MASK);
1538 mark_dirty(svm->vmcb, VMCB_CR);
1540 if (gcr0 == *hcr0 && svm->vcpu.fpu_active) {
1541 clr_cr_intercept(svm, INTERCEPT_CR0_READ);
1542 clr_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1544 set_cr_intercept(svm, INTERCEPT_CR0_READ);
1545 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1549 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1551 struct vcpu_svm *svm = to_svm(vcpu);
1553 #ifdef CONFIG_X86_64
1554 if (vcpu->arch.efer & EFER_LME) {
1555 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
1556 vcpu->arch.efer |= EFER_LMA;
1557 svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
1560 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
1561 vcpu->arch.efer &= ~EFER_LMA;
1562 svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
1566 vcpu->arch.cr0 = cr0;
1569 cr0 |= X86_CR0_PG | X86_CR0_WP;
1571 if (!vcpu->fpu_active)
1574 * re-enable caching here because the QEMU bios
1575 * does not do it - this results in some delay at
1578 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
1579 svm->vmcb->save.cr0 = cr0;
1580 mark_dirty(svm->vmcb, VMCB_CR);
1581 update_cr0_intercept(svm);
1584 static int svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1586 unsigned long host_cr4_mce = cr4_read_shadow() & X86_CR4_MCE;
1587 unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
1589 if (cr4 & X86_CR4_VMXE)
1592 if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
1593 svm_flush_tlb(vcpu);
1595 vcpu->arch.cr4 = cr4;
1598 cr4 |= host_cr4_mce;
1599 to_svm(vcpu)->vmcb->save.cr4 = cr4;
1600 mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
1604 static void svm_set_segment(struct kvm_vcpu *vcpu,
1605 struct kvm_segment *var, int seg)
1607 struct vcpu_svm *svm = to_svm(vcpu);
1608 struct vmcb_seg *s = svm_seg(vcpu, seg);
1610 s->base = var->base;
1611 s->limit = var->limit;
1612 s->selector = var->selector;
1616 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
1617 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
1618 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
1619 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
1620 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
1621 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
1622 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
1623 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
1627 * This is always accurate, except if SYSRET returned to a segment
1628 * with SS.DPL != 3. Intel does not have this quirk, and always
1629 * forces SS.DPL to 3 on sysret, so we ignore that case; fixing it
1630 * would entail passing the CPL to userspace and back.
1632 if (seg == VCPU_SREG_SS)
1633 svm->vmcb->save.cpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
1635 mark_dirty(svm->vmcb, VMCB_SEG);
1638 static void update_db_bp_intercept(struct kvm_vcpu *vcpu)
1640 struct vcpu_svm *svm = to_svm(vcpu);
1642 clr_exception_intercept(svm, DB_VECTOR);
1643 clr_exception_intercept(svm, BP_VECTOR);
1645 if (svm->nmi_singlestep)
1646 set_exception_intercept(svm, DB_VECTOR);
1648 if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
1649 if (vcpu->guest_debug &
1650 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
1651 set_exception_intercept(svm, DB_VECTOR);
1652 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
1653 set_exception_intercept(svm, BP_VECTOR);
1655 vcpu->guest_debug = 0;
1658 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
1660 if (sd->next_asid > sd->max_asid) {
1661 ++sd->asid_generation;
1663 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
1666 svm->asid_generation = sd->asid_generation;
1667 svm->vmcb->control.asid = sd->next_asid++;
1669 mark_dirty(svm->vmcb, VMCB_ASID);
1672 static u64 svm_get_dr6(struct kvm_vcpu *vcpu)
1674 return to_svm(vcpu)->vmcb->save.dr6;
1677 static void svm_set_dr6(struct kvm_vcpu *vcpu, unsigned long value)
1679 struct vcpu_svm *svm = to_svm(vcpu);
1681 svm->vmcb->save.dr6 = value;
1682 mark_dirty(svm->vmcb, VMCB_DR);
1685 static void svm_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
1687 struct vcpu_svm *svm = to_svm(vcpu);
1689 get_debugreg(vcpu->arch.db[0], 0);
1690 get_debugreg(vcpu->arch.db[1], 1);
1691 get_debugreg(vcpu->arch.db[2], 2);
1692 get_debugreg(vcpu->arch.db[3], 3);
1693 vcpu->arch.dr6 = svm_get_dr6(vcpu);
1694 vcpu->arch.dr7 = svm->vmcb->save.dr7;
1696 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
1697 set_dr_intercepts(svm);
1700 static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value)
1702 struct vcpu_svm *svm = to_svm(vcpu);
1704 svm->vmcb->save.dr7 = value;
1705 mark_dirty(svm->vmcb, VMCB_DR);
1708 static int pf_interception(struct vcpu_svm *svm)
1710 u64 fault_address = svm->vmcb->control.exit_info_2;
1714 switch (svm->apf_reason) {
1716 error_code = svm->vmcb->control.exit_info_1;
1718 trace_kvm_page_fault(fault_address, error_code);
1719 if (!npt_enabled && kvm_event_needs_reinjection(&svm->vcpu))
1720 kvm_mmu_unprotect_page_virt(&svm->vcpu, fault_address);
1721 r = kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code,
1722 svm->vmcb->control.insn_bytes,
1723 svm->vmcb->control.insn_len);
1725 case KVM_PV_REASON_PAGE_NOT_PRESENT:
1726 svm->apf_reason = 0;
1727 local_irq_disable();
1728 kvm_async_pf_task_wait(fault_address);
1731 case KVM_PV_REASON_PAGE_READY:
1732 svm->apf_reason = 0;
1733 local_irq_disable();
1734 kvm_async_pf_task_wake(fault_address);
1741 static int db_interception(struct vcpu_svm *svm)
1743 struct kvm_run *kvm_run = svm->vcpu.run;
1745 if (!(svm->vcpu.guest_debug &
1746 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
1747 !svm->nmi_singlestep) {
1748 kvm_queue_exception(&svm->vcpu, DB_VECTOR);
1752 if (svm->nmi_singlestep) {
1753 svm->nmi_singlestep = false;
1754 if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP))
1755 svm->vmcb->save.rflags &=
1756 ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1757 update_db_bp_intercept(&svm->vcpu);
1760 if (svm->vcpu.guest_debug &
1761 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
1762 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1763 kvm_run->debug.arch.pc =
1764 svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1765 kvm_run->debug.arch.exception = DB_VECTOR;
1772 static int bp_interception(struct vcpu_svm *svm)
1774 struct kvm_run *kvm_run = svm->vcpu.run;
1776 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1777 kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1778 kvm_run->debug.arch.exception = BP_VECTOR;
1782 static int ud_interception(struct vcpu_svm *svm)
1786 er = emulate_instruction(&svm->vcpu, EMULTYPE_TRAP_UD);
1787 if (er != EMULATE_DONE)
1788 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1792 static void svm_fpu_activate(struct kvm_vcpu *vcpu)
1794 struct vcpu_svm *svm = to_svm(vcpu);
1796 clr_exception_intercept(svm, NM_VECTOR);
1798 svm->vcpu.fpu_active = 1;
1799 update_cr0_intercept(svm);
1802 static int nm_interception(struct vcpu_svm *svm)
1804 svm_fpu_activate(&svm->vcpu);
1808 static bool is_erratum_383(void)
1813 if (!erratum_383_found)
1816 value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err);
1820 /* Bit 62 may or may not be set for this mce */
1821 value &= ~(1ULL << 62);
1823 if (value != 0xb600000000010015ULL)
1826 /* Clear MCi_STATUS registers */
1827 for (i = 0; i < 6; ++i)
1828 native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0);
1830 value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err);
1834 value &= ~(1ULL << 2);
1835 low = lower_32_bits(value);
1836 high = upper_32_bits(value);
1838 native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high);
1841 /* Flush tlb to evict multi-match entries */
1847 static void svm_handle_mce(struct vcpu_svm *svm)
1849 if (is_erratum_383()) {
1851 * Erratum 383 triggered. Guest state is corrupt so kill the
1854 pr_err("KVM: Guest triggered AMD Erratum 383\n");
1856 kvm_make_request(KVM_REQ_TRIPLE_FAULT, &svm->vcpu);
1862 * On an #MC intercept the MCE handler is not called automatically in
1863 * the host. So do it by hand here.
1867 /* not sure if we ever come back to this point */
1872 static int mc_interception(struct vcpu_svm *svm)
1877 static int shutdown_interception(struct vcpu_svm *svm)
1879 struct kvm_run *kvm_run = svm->vcpu.run;
1882 * VMCB is undefined after a SHUTDOWN intercept
1883 * so reinitialize it.
1885 clear_page(svm->vmcb);
1888 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1892 static int io_interception(struct vcpu_svm *svm)
1894 struct kvm_vcpu *vcpu = &svm->vcpu;
1895 u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
1896 int size, in, string;
1899 ++svm->vcpu.stat.io_exits;
1900 string = (io_info & SVM_IOIO_STR_MASK) != 0;
1901 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
1903 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
1905 port = io_info >> 16;
1906 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
1907 svm->next_rip = svm->vmcb->control.exit_info_2;
1908 skip_emulated_instruction(&svm->vcpu);
1910 return kvm_fast_pio_out(vcpu, size, port);
1913 static int nmi_interception(struct vcpu_svm *svm)
1918 static int intr_interception(struct vcpu_svm *svm)
1920 ++svm->vcpu.stat.irq_exits;
1924 static int nop_on_interception(struct vcpu_svm *svm)
1929 static int halt_interception(struct vcpu_svm *svm)
1931 svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
1932 return kvm_emulate_halt(&svm->vcpu);
1935 static int vmmcall_interception(struct vcpu_svm *svm)
1937 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
1938 kvm_emulate_hypercall(&svm->vcpu);
1942 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
1944 struct vcpu_svm *svm = to_svm(vcpu);
1946 return svm->nested.nested_cr3;
1949 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
1951 struct vcpu_svm *svm = to_svm(vcpu);
1952 u64 cr3 = svm->nested.nested_cr3;
1956 ret = kvm_read_guest_page(vcpu->kvm, gpa_to_gfn(cr3), &pdpte,
1957 offset_in_page(cr3) + index * 8, 8);
1963 static void nested_svm_set_tdp_cr3(struct kvm_vcpu *vcpu,
1966 struct vcpu_svm *svm = to_svm(vcpu);
1968 svm->vmcb->control.nested_cr3 = root;
1969 mark_dirty(svm->vmcb, VMCB_NPT);
1970 svm_flush_tlb(vcpu);
1973 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
1974 struct x86_exception *fault)
1976 struct vcpu_svm *svm = to_svm(vcpu);
1978 if (svm->vmcb->control.exit_code != SVM_EXIT_NPF) {
1980 * TODO: track the cause of the nested page fault, and
1981 * correctly fill in the high bits of exit_info_1.
1983 svm->vmcb->control.exit_code = SVM_EXIT_NPF;
1984 svm->vmcb->control.exit_code_hi = 0;
1985 svm->vmcb->control.exit_info_1 = (1ULL << 32);
1986 svm->vmcb->control.exit_info_2 = fault->address;
1989 svm->vmcb->control.exit_info_1 &= ~0xffffffffULL;
1990 svm->vmcb->control.exit_info_1 |= fault->error_code;
1993 * The present bit is always zero for page structure faults on real
1996 if (svm->vmcb->control.exit_info_1 & (2ULL << 32))
1997 svm->vmcb->control.exit_info_1 &= ~1;
1999 nested_svm_vmexit(svm);
2002 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
2004 WARN_ON(mmu_is_nested(vcpu));
2005 kvm_init_shadow_mmu(vcpu);
2006 vcpu->arch.mmu.set_cr3 = nested_svm_set_tdp_cr3;
2007 vcpu->arch.mmu.get_cr3 = nested_svm_get_tdp_cr3;
2008 vcpu->arch.mmu.get_pdptr = nested_svm_get_tdp_pdptr;
2009 vcpu->arch.mmu.inject_page_fault = nested_svm_inject_npf_exit;
2010 vcpu->arch.mmu.shadow_root_level = get_npt_level();
2011 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
2014 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
2016 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
2019 static int nested_svm_check_permissions(struct vcpu_svm *svm)
2021 if (!(svm->vcpu.arch.efer & EFER_SVME)
2022 || !is_paging(&svm->vcpu)) {
2023 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2027 if (svm->vmcb->save.cpl) {
2028 kvm_inject_gp(&svm->vcpu, 0);
2035 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
2036 bool has_error_code, u32 error_code)
2040 if (!is_guest_mode(&svm->vcpu))
2043 svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
2044 svm->vmcb->control.exit_code_hi = 0;
2045 svm->vmcb->control.exit_info_1 = error_code;
2046 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
2048 vmexit = nested_svm_intercept(svm);
2049 if (vmexit == NESTED_EXIT_DONE)
2050 svm->nested.exit_required = true;
2055 /* This function returns true if it is save to enable the irq window */
2056 static inline bool nested_svm_intr(struct vcpu_svm *svm)
2058 if (!is_guest_mode(&svm->vcpu))
2061 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
2064 if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
2068 * if vmexit was already requested (by intercepted exception
2069 * for instance) do not overwrite it with "external interrupt"
2072 if (svm->nested.exit_required)
2075 svm->vmcb->control.exit_code = SVM_EXIT_INTR;
2076 svm->vmcb->control.exit_info_1 = 0;
2077 svm->vmcb->control.exit_info_2 = 0;
2079 if (svm->nested.intercept & 1ULL) {
2081 * The #vmexit can't be emulated here directly because this
2082 * code path runs with irqs and preemption disabled. A
2083 * #vmexit emulation might sleep. Only signal request for
2086 svm->nested.exit_required = true;
2087 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
2094 /* This function returns true if it is save to enable the nmi window */
2095 static inline bool nested_svm_nmi(struct vcpu_svm *svm)
2097 if (!is_guest_mode(&svm->vcpu))
2100 if (!(svm->nested.intercept & (1ULL << INTERCEPT_NMI)))
2103 svm->vmcb->control.exit_code = SVM_EXIT_NMI;
2104 svm->nested.exit_required = true;
2109 static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, struct page **_page)
2115 page = gfn_to_page(svm->vcpu.kvm, gpa >> PAGE_SHIFT);
2116 if (is_error_page(page))
2124 kvm_inject_gp(&svm->vcpu, 0);
2129 static void nested_svm_unmap(struct page *page)
2132 kvm_release_page_dirty(page);
2135 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
2137 unsigned port, size, iopm_len;
2142 if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT)))
2143 return NESTED_EXIT_HOST;
2145 port = svm->vmcb->control.exit_info_1 >> 16;
2146 size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
2147 SVM_IOIO_SIZE_SHIFT;
2148 gpa = svm->nested.vmcb_iopm + (port / 8);
2149 start_bit = port % 8;
2150 iopm_len = (start_bit + size > 8) ? 2 : 1;
2151 mask = (0xf >> (4 - size)) << start_bit;
2154 if (kvm_read_guest(svm->vcpu.kvm, gpa, &val, iopm_len))
2155 return NESTED_EXIT_DONE;
2157 return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
2160 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
2162 u32 offset, msr, value;
2165 if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
2166 return NESTED_EXIT_HOST;
2168 msr = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2169 offset = svm_msrpm_offset(msr);
2170 write = svm->vmcb->control.exit_info_1 & 1;
2171 mask = 1 << ((2 * (msr & 0xf)) + write);
2173 if (offset == MSR_INVALID)
2174 return NESTED_EXIT_DONE;
2176 /* Offset is in 32 bit units but need in 8 bit units */
2179 if (kvm_read_guest(svm->vcpu.kvm, svm->nested.vmcb_msrpm + offset, &value, 4))
2180 return NESTED_EXIT_DONE;
2182 return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
2185 static int nested_svm_exit_special(struct vcpu_svm *svm)
2187 u32 exit_code = svm->vmcb->control.exit_code;
2189 switch (exit_code) {
2192 case SVM_EXIT_EXCP_BASE + MC_VECTOR:
2193 return NESTED_EXIT_HOST;
2195 /* For now we are always handling NPFs when using them */
2197 return NESTED_EXIT_HOST;
2199 case SVM_EXIT_EXCP_BASE + PF_VECTOR:
2200 /* When we're shadowing, trap PFs, but not async PF */
2201 if (!npt_enabled && svm->apf_reason == 0)
2202 return NESTED_EXIT_HOST;
2204 case SVM_EXIT_EXCP_BASE + NM_VECTOR:
2205 nm_interception(svm);
2211 return NESTED_EXIT_CONTINUE;
2215 * If this function returns true, this #vmexit was already handled
2217 static int nested_svm_intercept(struct vcpu_svm *svm)
2219 u32 exit_code = svm->vmcb->control.exit_code;
2220 int vmexit = NESTED_EXIT_HOST;
2222 switch (exit_code) {
2224 vmexit = nested_svm_exit_handled_msr(svm);
2227 vmexit = nested_svm_intercept_ioio(svm);
2229 case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
2230 u32 bit = 1U << (exit_code - SVM_EXIT_READ_CR0);
2231 if (svm->nested.intercept_cr & bit)
2232 vmexit = NESTED_EXIT_DONE;
2235 case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
2236 u32 bit = 1U << (exit_code - SVM_EXIT_READ_DR0);
2237 if (svm->nested.intercept_dr & bit)
2238 vmexit = NESTED_EXIT_DONE;
2241 case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
2242 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
2243 if (svm->nested.intercept_exceptions & excp_bits)
2244 vmexit = NESTED_EXIT_DONE;
2245 /* async page fault always cause vmexit */
2246 else if ((exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) &&
2247 svm->apf_reason != 0)
2248 vmexit = NESTED_EXIT_DONE;
2251 case SVM_EXIT_ERR: {
2252 vmexit = NESTED_EXIT_DONE;
2256 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
2257 if (svm->nested.intercept & exit_bits)
2258 vmexit = NESTED_EXIT_DONE;
2265 static int nested_svm_exit_handled(struct vcpu_svm *svm)
2269 vmexit = nested_svm_intercept(svm);
2271 if (vmexit == NESTED_EXIT_DONE)
2272 nested_svm_vmexit(svm);
2277 static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
2279 struct vmcb_control_area *dst = &dst_vmcb->control;
2280 struct vmcb_control_area *from = &from_vmcb->control;
2282 dst->intercept_cr = from->intercept_cr;
2283 dst->intercept_dr = from->intercept_dr;
2284 dst->intercept_exceptions = from->intercept_exceptions;
2285 dst->intercept = from->intercept;
2286 dst->iopm_base_pa = from->iopm_base_pa;
2287 dst->msrpm_base_pa = from->msrpm_base_pa;
2288 dst->tsc_offset = from->tsc_offset;
2289 dst->asid = from->asid;
2290 dst->tlb_ctl = from->tlb_ctl;
2291 dst->int_ctl = from->int_ctl;
2292 dst->int_vector = from->int_vector;
2293 dst->int_state = from->int_state;
2294 dst->exit_code = from->exit_code;
2295 dst->exit_code_hi = from->exit_code_hi;
2296 dst->exit_info_1 = from->exit_info_1;
2297 dst->exit_info_2 = from->exit_info_2;
2298 dst->exit_int_info = from->exit_int_info;
2299 dst->exit_int_info_err = from->exit_int_info_err;
2300 dst->nested_ctl = from->nested_ctl;
2301 dst->event_inj = from->event_inj;
2302 dst->event_inj_err = from->event_inj_err;
2303 dst->nested_cr3 = from->nested_cr3;
2304 dst->lbr_ctl = from->lbr_ctl;
2307 static int nested_svm_vmexit(struct vcpu_svm *svm)
2309 struct vmcb *nested_vmcb;
2310 struct vmcb *hsave = svm->nested.hsave;
2311 struct vmcb *vmcb = svm->vmcb;
2314 trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
2315 vmcb->control.exit_info_1,
2316 vmcb->control.exit_info_2,
2317 vmcb->control.exit_int_info,
2318 vmcb->control.exit_int_info_err,
2321 nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, &page);
2325 /* Exit Guest-Mode */
2326 leave_guest_mode(&svm->vcpu);
2327 svm->nested.vmcb = 0;
2329 /* Give the current vmcb to the guest */
2332 nested_vmcb->save.es = vmcb->save.es;
2333 nested_vmcb->save.cs = vmcb->save.cs;
2334 nested_vmcb->save.ss = vmcb->save.ss;
2335 nested_vmcb->save.ds = vmcb->save.ds;
2336 nested_vmcb->save.gdtr = vmcb->save.gdtr;
2337 nested_vmcb->save.idtr = vmcb->save.idtr;
2338 nested_vmcb->save.efer = svm->vcpu.arch.efer;
2339 nested_vmcb->save.cr0 = kvm_read_cr0(&svm->vcpu);
2340 nested_vmcb->save.cr3 = kvm_read_cr3(&svm->vcpu);
2341 nested_vmcb->save.cr2 = vmcb->save.cr2;
2342 nested_vmcb->save.cr4 = svm->vcpu.arch.cr4;
2343 nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu);
2344 nested_vmcb->save.rip = vmcb->save.rip;
2345 nested_vmcb->save.rsp = vmcb->save.rsp;
2346 nested_vmcb->save.rax = vmcb->save.rax;
2347 nested_vmcb->save.dr7 = vmcb->save.dr7;
2348 nested_vmcb->save.dr6 = vmcb->save.dr6;
2349 nested_vmcb->save.cpl = vmcb->save.cpl;
2351 nested_vmcb->control.int_ctl = vmcb->control.int_ctl;
2352 nested_vmcb->control.int_vector = vmcb->control.int_vector;
2353 nested_vmcb->control.int_state = vmcb->control.int_state;
2354 nested_vmcb->control.exit_code = vmcb->control.exit_code;
2355 nested_vmcb->control.exit_code_hi = vmcb->control.exit_code_hi;
2356 nested_vmcb->control.exit_info_1 = vmcb->control.exit_info_1;
2357 nested_vmcb->control.exit_info_2 = vmcb->control.exit_info_2;
2358 nested_vmcb->control.exit_int_info = vmcb->control.exit_int_info;
2359 nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
2360 nested_vmcb->control.next_rip = vmcb->control.next_rip;
2363 * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
2364 * to make sure that we do not lose injected events. So check event_inj
2365 * here and copy it to exit_int_info if it is valid.
2366 * Exit_int_info and event_inj can't be both valid because the case
2367 * below only happens on a VMRUN instruction intercept which has
2368 * no valid exit_int_info set.
2370 if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
2371 struct vmcb_control_area *nc = &nested_vmcb->control;
2373 nc->exit_int_info = vmcb->control.event_inj;
2374 nc->exit_int_info_err = vmcb->control.event_inj_err;
2377 nested_vmcb->control.tlb_ctl = 0;
2378 nested_vmcb->control.event_inj = 0;
2379 nested_vmcb->control.event_inj_err = 0;
2381 /* We always set V_INTR_MASKING and remember the old value in hflags */
2382 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
2383 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
2385 /* Restore the original control entries */
2386 copy_vmcb_control_area(vmcb, hsave);
2388 kvm_clear_exception_queue(&svm->vcpu);
2389 kvm_clear_interrupt_queue(&svm->vcpu);
2391 svm->nested.nested_cr3 = 0;
2393 /* Restore selected save entries */
2394 svm->vmcb->save.es = hsave->save.es;
2395 svm->vmcb->save.cs = hsave->save.cs;
2396 svm->vmcb->save.ss = hsave->save.ss;
2397 svm->vmcb->save.ds = hsave->save.ds;
2398 svm->vmcb->save.gdtr = hsave->save.gdtr;
2399 svm->vmcb->save.idtr = hsave->save.idtr;
2400 kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
2401 svm_set_efer(&svm->vcpu, hsave->save.efer);
2402 svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
2403 svm_set_cr4(&svm->vcpu, hsave->save.cr4);
2405 svm->vmcb->save.cr3 = hsave->save.cr3;
2406 svm->vcpu.arch.cr3 = hsave->save.cr3;
2408 (void)kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
2410 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
2411 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
2412 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
2413 svm->vmcb->save.dr7 = 0;
2414 svm->vmcb->save.cpl = 0;
2415 svm->vmcb->control.exit_int_info = 0;
2417 mark_all_dirty(svm->vmcb);
2419 nested_svm_unmap(page);
2421 nested_svm_uninit_mmu_context(&svm->vcpu);
2422 kvm_mmu_reset_context(&svm->vcpu);
2423 kvm_mmu_load(&svm->vcpu);
2428 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
2431 * This function merges the msr permission bitmaps of kvm and the
2432 * nested vmcb. It is optimized in that it only merges the parts where
2433 * the kvm msr permission bitmap may contain zero bits
2437 if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
2440 for (i = 0; i < MSRPM_OFFSETS; i++) {
2444 if (msrpm_offsets[i] == 0xffffffff)
2447 p = msrpm_offsets[i];
2448 offset = svm->nested.vmcb_msrpm + (p * 4);
2450 if (kvm_read_guest(svm->vcpu.kvm, offset, &value, 4))
2453 svm->nested.msrpm[p] = svm->msrpm[p] | value;
2456 svm->vmcb->control.msrpm_base_pa = __pa(svm->nested.msrpm);
2461 static bool nested_vmcb_checks(struct vmcb *vmcb)
2463 if ((vmcb->control.intercept & (1ULL << INTERCEPT_VMRUN)) == 0)
2466 if (vmcb->control.asid == 0)
2469 if (vmcb->control.nested_ctl && !npt_enabled)
2475 static bool nested_svm_vmrun(struct vcpu_svm *svm)
2477 struct vmcb *nested_vmcb;
2478 struct vmcb *hsave = svm->nested.hsave;
2479 struct vmcb *vmcb = svm->vmcb;
2483 vmcb_gpa = svm->vmcb->save.rax;
2485 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2489 if (!nested_vmcb_checks(nested_vmcb)) {
2490 nested_vmcb->control.exit_code = SVM_EXIT_ERR;
2491 nested_vmcb->control.exit_code_hi = 0;
2492 nested_vmcb->control.exit_info_1 = 0;
2493 nested_vmcb->control.exit_info_2 = 0;
2495 nested_svm_unmap(page);
2500 trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa,
2501 nested_vmcb->save.rip,
2502 nested_vmcb->control.int_ctl,
2503 nested_vmcb->control.event_inj,
2504 nested_vmcb->control.nested_ctl);
2506 trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr & 0xffff,
2507 nested_vmcb->control.intercept_cr >> 16,
2508 nested_vmcb->control.intercept_exceptions,
2509 nested_vmcb->control.intercept);
2511 /* Clear internal status */
2512 kvm_clear_exception_queue(&svm->vcpu);
2513 kvm_clear_interrupt_queue(&svm->vcpu);
2516 * Save the old vmcb, so we don't need to pick what we save, but can
2517 * restore everything when a VMEXIT occurs
2519 hsave->save.es = vmcb->save.es;
2520 hsave->save.cs = vmcb->save.cs;
2521 hsave->save.ss = vmcb->save.ss;
2522 hsave->save.ds = vmcb->save.ds;
2523 hsave->save.gdtr = vmcb->save.gdtr;
2524 hsave->save.idtr = vmcb->save.idtr;
2525 hsave->save.efer = svm->vcpu.arch.efer;
2526 hsave->save.cr0 = kvm_read_cr0(&svm->vcpu);
2527 hsave->save.cr4 = svm->vcpu.arch.cr4;
2528 hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
2529 hsave->save.rip = kvm_rip_read(&svm->vcpu);
2530 hsave->save.rsp = vmcb->save.rsp;
2531 hsave->save.rax = vmcb->save.rax;
2533 hsave->save.cr3 = vmcb->save.cr3;
2535 hsave->save.cr3 = kvm_read_cr3(&svm->vcpu);
2537 copy_vmcb_control_area(hsave, vmcb);
2539 if (kvm_get_rflags(&svm->vcpu) & X86_EFLAGS_IF)
2540 svm->vcpu.arch.hflags |= HF_HIF_MASK;
2542 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
2544 if (nested_vmcb->control.nested_ctl) {
2545 kvm_mmu_unload(&svm->vcpu);
2546 svm->nested.nested_cr3 = nested_vmcb->control.nested_cr3;
2547 nested_svm_init_mmu_context(&svm->vcpu);
2550 /* Load the nested guest state */
2551 svm->vmcb->save.es = nested_vmcb->save.es;
2552 svm->vmcb->save.cs = nested_vmcb->save.cs;
2553 svm->vmcb->save.ss = nested_vmcb->save.ss;
2554 svm->vmcb->save.ds = nested_vmcb->save.ds;
2555 svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
2556 svm->vmcb->save.idtr = nested_vmcb->save.idtr;
2557 kvm_set_rflags(&svm->vcpu, nested_vmcb->save.rflags);
2558 svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
2559 svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
2560 svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
2562 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
2563 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
2565 (void)kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
2567 /* Guest paging mode is active - reset mmu */
2568 kvm_mmu_reset_context(&svm->vcpu);
2570 svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
2571 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
2572 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
2573 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
2575 /* In case we don't even reach vcpu_run, the fields are not updated */
2576 svm->vmcb->save.rax = nested_vmcb->save.rax;
2577 svm->vmcb->save.rsp = nested_vmcb->save.rsp;
2578 svm->vmcb->save.rip = nested_vmcb->save.rip;
2579 svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
2580 svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
2581 svm->vmcb->save.cpl = nested_vmcb->save.cpl;
2583 svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL;
2584 svm->nested.vmcb_iopm = nested_vmcb->control.iopm_base_pa & ~0x0fffULL;
2586 /* cache intercepts */
2587 svm->nested.intercept_cr = nested_vmcb->control.intercept_cr;
2588 svm->nested.intercept_dr = nested_vmcb->control.intercept_dr;
2589 svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
2590 svm->nested.intercept = nested_vmcb->control.intercept;
2592 svm_flush_tlb(&svm->vcpu);
2593 svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
2594 if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
2595 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
2597 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
2599 if (svm->vcpu.arch.hflags & HF_VINTR_MASK) {
2600 /* We only want the cr8 intercept bits of the guest */
2601 clr_cr_intercept(svm, INTERCEPT_CR8_READ);
2602 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
2605 /* We don't want to see VMMCALLs from a nested guest */
2606 clr_intercept(svm, INTERCEPT_VMMCALL);
2608 svm->vmcb->control.lbr_ctl = nested_vmcb->control.lbr_ctl;
2609 svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
2610 svm->vmcb->control.int_state = nested_vmcb->control.int_state;
2611 svm->vmcb->control.tsc_offset += nested_vmcb->control.tsc_offset;
2612 svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
2613 svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
2615 nested_svm_unmap(page);
2617 /* Enter Guest-Mode */
2618 enter_guest_mode(&svm->vcpu);
2621 * Merge guest and host intercepts - must be called with vcpu in
2622 * guest-mode to take affect here
2624 recalc_intercepts(svm);
2626 svm->nested.vmcb = vmcb_gpa;
2630 mark_all_dirty(svm->vmcb);
2635 static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
2637 to_vmcb->save.fs = from_vmcb->save.fs;
2638 to_vmcb->save.gs = from_vmcb->save.gs;
2639 to_vmcb->save.tr = from_vmcb->save.tr;
2640 to_vmcb->save.ldtr = from_vmcb->save.ldtr;
2641 to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
2642 to_vmcb->save.star = from_vmcb->save.star;
2643 to_vmcb->save.lstar = from_vmcb->save.lstar;
2644 to_vmcb->save.cstar = from_vmcb->save.cstar;
2645 to_vmcb->save.sfmask = from_vmcb->save.sfmask;
2646 to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
2647 to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
2648 to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
2651 static int vmload_interception(struct vcpu_svm *svm)
2653 struct vmcb *nested_vmcb;
2656 if (nested_svm_check_permissions(svm))
2659 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2663 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2664 skip_emulated_instruction(&svm->vcpu);
2666 nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
2667 nested_svm_unmap(page);
2672 static int vmsave_interception(struct vcpu_svm *svm)
2674 struct vmcb *nested_vmcb;
2677 if (nested_svm_check_permissions(svm))
2680 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2684 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2685 skip_emulated_instruction(&svm->vcpu);
2687 nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
2688 nested_svm_unmap(page);
2693 static int vmrun_interception(struct vcpu_svm *svm)
2695 if (nested_svm_check_permissions(svm))
2698 /* Save rip after vmrun instruction */
2699 kvm_rip_write(&svm->vcpu, kvm_rip_read(&svm->vcpu) + 3);
2701 if (!nested_svm_vmrun(svm))
2704 if (!nested_svm_vmrun_msrpm(svm))
2711 svm->vmcb->control.exit_code = SVM_EXIT_ERR;
2712 svm->vmcb->control.exit_code_hi = 0;
2713 svm->vmcb->control.exit_info_1 = 0;
2714 svm->vmcb->control.exit_info_2 = 0;
2716 nested_svm_vmexit(svm);
2721 static int stgi_interception(struct vcpu_svm *svm)
2723 if (nested_svm_check_permissions(svm))
2726 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2727 skip_emulated_instruction(&svm->vcpu);
2728 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2735 static int clgi_interception(struct vcpu_svm *svm)
2737 if (nested_svm_check_permissions(svm))
2740 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2741 skip_emulated_instruction(&svm->vcpu);
2745 /* After a CLGI no interrupts should come */
2746 svm_clear_vintr(svm);
2747 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
2749 mark_dirty(svm->vmcb, VMCB_INTR);
2754 static int invlpga_interception(struct vcpu_svm *svm)
2756 struct kvm_vcpu *vcpu = &svm->vcpu;
2758 trace_kvm_invlpga(svm->vmcb->save.rip, kvm_register_read(&svm->vcpu, VCPU_REGS_RCX),
2759 kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
2761 /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
2762 kvm_mmu_invlpg(vcpu, kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
2764 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2765 skip_emulated_instruction(&svm->vcpu);
2769 static int skinit_interception(struct vcpu_svm *svm)
2771 trace_kvm_skinit(svm->vmcb->save.rip, kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
2773 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2777 static int wbinvd_interception(struct vcpu_svm *svm)
2779 kvm_emulate_wbinvd(&svm->vcpu);
2783 static int xsetbv_interception(struct vcpu_svm *svm)
2785 u64 new_bv = kvm_read_edx_eax(&svm->vcpu);
2786 u32 index = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
2788 if (kvm_set_xcr(&svm->vcpu, index, new_bv) == 0) {
2789 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2790 skip_emulated_instruction(&svm->vcpu);
2796 static int task_switch_interception(struct vcpu_svm *svm)
2800 int int_type = svm->vmcb->control.exit_int_info &
2801 SVM_EXITINTINFO_TYPE_MASK;
2802 int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
2804 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
2806 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
2807 bool has_error_code = false;
2810 tss_selector = (u16)svm->vmcb->control.exit_info_1;
2812 if (svm->vmcb->control.exit_info_2 &
2813 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
2814 reason = TASK_SWITCH_IRET;
2815 else if (svm->vmcb->control.exit_info_2 &
2816 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
2817 reason = TASK_SWITCH_JMP;
2819 reason = TASK_SWITCH_GATE;
2821 reason = TASK_SWITCH_CALL;
2823 if (reason == TASK_SWITCH_GATE) {
2825 case SVM_EXITINTINFO_TYPE_NMI:
2826 svm->vcpu.arch.nmi_injected = false;
2828 case SVM_EXITINTINFO_TYPE_EXEPT:
2829 if (svm->vmcb->control.exit_info_2 &
2830 (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) {
2831 has_error_code = true;
2833 (u32)svm->vmcb->control.exit_info_2;
2835 kvm_clear_exception_queue(&svm->vcpu);
2837 case SVM_EXITINTINFO_TYPE_INTR:
2838 kvm_clear_interrupt_queue(&svm->vcpu);
2845 if (reason != TASK_SWITCH_GATE ||
2846 int_type == SVM_EXITINTINFO_TYPE_SOFT ||
2847 (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
2848 (int_vec == OF_VECTOR || int_vec == BP_VECTOR)))
2849 skip_emulated_instruction(&svm->vcpu);
2851 if (int_type != SVM_EXITINTINFO_TYPE_SOFT)
2854 if (kvm_task_switch(&svm->vcpu, tss_selector, int_vec, reason,
2855 has_error_code, error_code) == EMULATE_FAIL) {
2856 svm->vcpu.run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2857 svm->vcpu.run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
2858 svm->vcpu.run->internal.ndata = 0;
2864 static int cpuid_interception(struct vcpu_svm *svm)
2866 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2867 kvm_emulate_cpuid(&svm->vcpu);
2871 static int iret_interception(struct vcpu_svm *svm)
2873 ++svm->vcpu.stat.nmi_window_exits;
2874 clr_intercept(svm, INTERCEPT_IRET);
2875 svm->vcpu.arch.hflags |= HF_IRET_MASK;
2876 svm->nmi_iret_rip = kvm_rip_read(&svm->vcpu);
2877 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2881 static int invlpg_interception(struct vcpu_svm *svm)
2883 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2884 return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
2886 kvm_mmu_invlpg(&svm->vcpu, svm->vmcb->control.exit_info_1);
2887 skip_emulated_instruction(&svm->vcpu);
2891 static int emulate_on_interception(struct vcpu_svm *svm)
2893 return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
2896 static int rdpmc_interception(struct vcpu_svm *svm)
2900 if (!static_cpu_has(X86_FEATURE_NRIPS))
2901 return emulate_on_interception(svm);
2903 err = kvm_rdpmc(&svm->vcpu);
2904 kvm_complete_insn_gp(&svm->vcpu, err);
2909 static bool check_selective_cr0_intercepted(struct vcpu_svm *svm,
2912 unsigned long cr0 = svm->vcpu.arch.cr0;
2916 intercept = svm->nested.intercept;
2918 if (!is_guest_mode(&svm->vcpu) ||
2919 (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0))))
2922 cr0 &= ~SVM_CR0_SELECTIVE_MASK;
2923 val &= ~SVM_CR0_SELECTIVE_MASK;
2926 svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
2927 ret = (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE);
2933 #define CR_VALID (1ULL << 63)
2935 static int cr_interception(struct vcpu_svm *svm)
2941 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2942 return emulate_on_interception(svm);
2944 if (unlikely((svm->vmcb->control.exit_info_1 & CR_VALID) == 0))
2945 return emulate_on_interception(svm);
2947 reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
2948 if (svm->vmcb->control.exit_code == SVM_EXIT_CR0_SEL_WRITE)
2949 cr = SVM_EXIT_WRITE_CR0 - SVM_EXIT_READ_CR0;
2951 cr = svm->vmcb->control.exit_code - SVM_EXIT_READ_CR0;
2954 if (cr >= 16) { /* mov to cr */
2956 val = kvm_register_read(&svm->vcpu, reg);
2959 if (!check_selective_cr0_intercepted(svm, val))
2960 err = kvm_set_cr0(&svm->vcpu, val);
2966 err = kvm_set_cr3(&svm->vcpu, val);
2969 err = kvm_set_cr4(&svm->vcpu, val);
2972 err = kvm_set_cr8(&svm->vcpu, val);
2975 WARN(1, "unhandled write to CR%d", cr);
2976 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2979 } else { /* mov from cr */
2982 val = kvm_read_cr0(&svm->vcpu);
2985 val = svm->vcpu.arch.cr2;
2988 val = kvm_read_cr3(&svm->vcpu);
2991 val = kvm_read_cr4(&svm->vcpu);
2994 val = kvm_get_cr8(&svm->vcpu);
2997 WARN(1, "unhandled read from CR%d", cr);
2998 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
3001 kvm_register_write(&svm->vcpu, reg, val);
3003 kvm_complete_insn_gp(&svm->vcpu, err);
3008 static int dr_interception(struct vcpu_svm *svm)
3013 if (svm->vcpu.guest_debug == 0) {
3015 * No more DR vmexits; force a reload of the debug registers
3016 * and reenter on this instruction. The next vmexit will
3017 * retrieve the full state of the debug registers.
3019 clr_dr_intercepts(svm);
3020 svm->vcpu.arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
3024 if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS))
3025 return emulate_on_interception(svm);
3027 reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
3028 dr = svm->vmcb->control.exit_code - SVM_EXIT_READ_DR0;
3030 if (dr >= 16) { /* mov to DRn */
3031 if (!kvm_require_dr(&svm->vcpu, dr - 16))
3033 val = kvm_register_read(&svm->vcpu, reg);
3034 kvm_set_dr(&svm->vcpu, dr - 16, val);
3036 if (!kvm_require_dr(&svm->vcpu, dr))
3038 kvm_get_dr(&svm->vcpu, dr, &val);
3039 kvm_register_write(&svm->vcpu, reg, val);
3042 skip_emulated_instruction(&svm->vcpu);
3047 static int cr8_write_interception(struct vcpu_svm *svm)
3049 struct kvm_run *kvm_run = svm->vcpu.run;
3052 u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
3053 /* instruction emulation calls kvm_set_cr8() */
3054 r = cr_interception(svm);
3055 if (irqchip_in_kernel(svm->vcpu.kvm))
3057 if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
3059 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
3063 static u64 svm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
3065 struct vmcb *vmcb = get_host_vmcb(to_svm(vcpu));
3066 return vmcb->control.tsc_offset +
3067 svm_scale_tsc(vcpu, host_tsc);
3070 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
3072 struct vcpu_svm *svm = to_svm(vcpu);
3075 case MSR_IA32_TSC: {
3076 *data = svm->vmcb->control.tsc_offset +
3077 svm_scale_tsc(vcpu, native_read_tsc());
3082 *data = svm->vmcb->save.star;
3084 #ifdef CONFIG_X86_64
3086 *data = svm->vmcb->save.lstar;
3089 *data = svm->vmcb->save.cstar;
3091 case MSR_KERNEL_GS_BASE:
3092 *data = svm->vmcb->save.kernel_gs_base;
3094 case MSR_SYSCALL_MASK:
3095 *data = svm->vmcb->save.sfmask;
3098 case MSR_IA32_SYSENTER_CS:
3099 *data = svm->vmcb->save.sysenter_cs;
3101 case MSR_IA32_SYSENTER_EIP:
3102 *data = svm->sysenter_eip;
3104 case MSR_IA32_SYSENTER_ESP:
3105 *data = svm->sysenter_esp;
3108 * Nobody will change the following 5 values in the VMCB so we can
3109 * safely return them on rdmsr. They will always be 0 until LBRV is
3112 case MSR_IA32_DEBUGCTLMSR:
3113 *data = svm->vmcb->save.dbgctl;
3115 case MSR_IA32_LASTBRANCHFROMIP:
3116 *data = svm->vmcb->save.br_from;
3118 case MSR_IA32_LASTBRANCHTOIP:
3119 *data = svm->vmcb->save.br_to;
3121 case MSR_IA32_LASTINTFROMIP:
3122 *data = svm->vmcb->save.last_excp_from;
3124 case MSR_IA32_LASTINTTOIP:
3125 *data = svm->vmcb->save.last_excp_to;
3127 case MSR_VM_HSAVE_PA:
3128 *data = svm->nested.hsave_msr;
3131 *data = svm->nested.vm_cr_msr;
3133 case MSR_IA32_UCODE_REV:
3137 return kvm_get_msr_common(vcpu, ecx, data);
3142 static int rdmsr_interception(struct vcpu_svm *svm)
3144 u32 ecx = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
3147 if (svm_get_msr(&svm->vcpu, ecx, &data)) {
3148 trace_kvm_msr_read_ex(ecx);
3149 kvm_inject_gp(&svm->vcpu, 0);
3151 trace_kvm_msr_read(ecx, data);
3153 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, data & 0xffffffff);
3154 kvm_register_write(&svm->vcpu, VCPU_REGS_RDX, data >> 32);
3155 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3156 skip_emulated_instruction(&svm->vcpu);
3161 static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
3163 struct vcpu_svm *svm = to_svm(vcpu);
3164 int svm_dis, chg_mask;
3166 if (data & ~SVM_VM_CR_VALID_MASK)
3169 chg_mask = SVM_VM_CR_VALID_MASK;
3171 if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
3172 chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
3174 svm->nested.vm_cr_msr &= ~chg_mask;
3175 svm->nested.vm_cr_msr |= (data & chg_mask);
3177 svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
3179 /* check for svm_disable while efer.svme is set */
3180 if (svm_dis && (vcpu->arch.efer & EFER_SVME))
3186 static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
3188 struct vcpu_svm *svm = to_svm(vcpu);
3190 u32 ecx = msr->index;
3191 u64 data = msr->data;
3194 kvm_write_tsc(vcpu, msr);
3197 svm->vmcb->save.star = data;
3199 #ifdef CONFIG_X86_64
3201 svm->vmcb->save.lstar = data;
3204 svm->vmcb->save.cstar = data;
3206 case MSR_KERNEL_GS_BASE:
3207 svm->vmcb->save.kernel_gs_base = data;
3209 case MSR_SYSCALL_MASK:
3210 svm->vmcb->save.sfmask = data;
3213 case MSR_IA32_SYSENTER_CS:
3214 svm->vmcb->save.sysenter_cs = data;
3216 case MSR_IA32_SYSENTER_EIP:
3217 svm->sysenter_eip = data;
3218 svm->vmcb->save.sysenter_eip = data;
3220 case MSR_IA32_SYSENTER_ESP:
3221 svm->sysenter_esp = data;
3222 svm->vmcb->save.sysenter_esp = data;
3224 case MSR_IA32_DEBUGCTLMSR:
3225 if (!boot_cpu_has(X86_FEATURE_LBRV)) {
3226 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
3230 if (data & DEBUGCTL_RESERVED_BITS)
3233 svm->vmcb->save.dbgctl = data;
3234 mark_dirty(svm->vmcb, VMCB_LBR);
3235 if (data & (1ULL<<0))
3236 svm_enable_lbrv(svm);
3238 svm_disable_lbrv(svm);
3240 case MSR_VM_HSAVE_PA:
3241 svm->nested.hsave_msr = data;
3244 return svm_set_vm_cr(vcpu, data);
3246 vcpu_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
3249 return kvm_set_msr_common(vcpu, msr);
3254 static int wrmsr_interception(struct vcpu_svm *svm)
3256 struct msr_data msr;
3257 u32 ecx = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
3258 u64 data = kvm_read_edx_eax(&svm->vcpu);
3262 msr.host_initiated = false;
3264 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3265 if (kvm_set_msr(&svm->vcpu, &msr)) {
3266 trace_kvm_msr_write_ex(ecx, data);
3267 kvm_inject_gp(&svm->vcpu, 0);
3269 trace_kvm_msr_write(ecx, data);
3270 skip_emulated_instruction(&svm->vcpu);
3275 static int msr_interception(struct vcpu_svm *svm)
3277 if (svm->vmcb->control.exit_info_1)
3278 return wrmsr_interception(svm);
3280 return rdmsr_interception(svm);
3283 static int interrupt_window_interception(struct vcpu_svm *svm)
3285 struct kvm_run *kvm_run = svm->vcpu.run;
3287 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3288 svm_clear_vintr(svm);
3289 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
3290 mark_dirty(svm->vmcb, VMCB_INTR);
3291 ++svm->vcpu.stat.irq_window_exits;
3293 * If the user space waits to inject interrupts, exit as soon as
3296 if (!irqchip_in_kernel(svm->vcpu.kvm) &&
3297 kvm_run->request_interrupt_window &&
3298 !kvm_cpu_has_interrupt(&svm->vcpu)) {
3299 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
3306 static int pause_interception(struct vcpu_svm *svm)
3308 kvm_vcpu_on_spin(&(svm->vcpu));
3312 static int nop_interception(struct vcpu_svm *svm)
3314 skip_emulated_instruction(&(svm->vcpu));
3318 static int monitor_interception(struct vcpu_svm *svm)
3320 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
3321 return nop_interception(svm);
3324 static int mwait_interception(struct vcpu_svm *svm)
3326 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
3327 return nop_interception(svm);
3330 static int (*const svm_exit_handlers[])(struct vcpu_svm *svm) = {
3331 [SVM_EXIT_READ_CR0] = cr_interception,
3332 [SVM_EXIT_READ_CR3] = cr_interception,
3333 [SVM_EXIT_READ_CR4] = cr_interception,
3334 [SVM_EXIT_READ_CR8] = cr_interception,
3335 [SVM_EXIT_CR0_SEL_WRITE] = cr_interception,
3336 [SVM_EXIT_WRITE_CR0] = cr_interception,
3337 [SVM_EXIT_WRITE_CR3] = cr_interception,
3338 [SVM_EXIT_WRITE_CR4] = cr_interception,
3339 [SVM_EXIT_WRITE_CR8] = cr8_write_interception,
3340 [SVM_EXIT_READ_DR0] = dr_interception,
3341 [SVM_EXIT_READ_DR1] = dr_interception,
3342 [SVM_EXIT_READ_DR2] = dr_interception,
3343 [SVM_EXIT_READ_DR3] = dr_interception,
3344 [SVM_EXIT_READ_DR4] = dr_interception,
3345 [SVM_EXIT_READ_DR5] = dr_interception,
3346 [SVM_EXIT_READ_DR6] = dr_interception,
3347 [SVM_EXIT_READ_DR7] = dr_interception,
3348 [SVM_EXIT_WRITE_DR0] = dr_interception,
3349 [SVM_EXIT_WRITE_DR1] = dr_interception,
3350 [SVM_EXIT_WRITE_DR2] = dr_interception,
3351 [SVM_EXIT_WRITE_DR3] = dr_interception,
3352 [SVM_EXIT_WRITE_DR4] = dr_interception,
3353 [SVM_EXIT_WRITE_DR5] = dr_interception,
3354 [SVM_EXIT_WRITE_DR6] = dr_interception,
3355 [SVM_EXIT_WRITE_DR7] = dr_interception,
3356 [SVM_EXIT_EXCP_BASE + DB_VECTOR] = db_interception,
3357 [SVM_EXIT_EXCP_BASE + BP_VECTOR] = bp_interception,
3358 [SVM_EXIT_EXCP_BASE + UD_VECTOR] = ud_interception,
3359 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
3360 [SVM_EXIT_EXCP_BASE + NM_VECTOR] = nm_interception,
3361 [SVM_EXIT_EXCP_BASE + MC_VECTOR] = mc_interception,
3362 [SVM_EXIT_INTR] = intr_interception,
3363 [SVM_EXIT_NMI] = nmi_interception,
3364 [SVM_EXIT_SMI] = nop_on_interception,
3365 [SVM_EXIT_INIT] = nop_on_interception,
3366 [SVM_EXIT_VINTR] = interrupt_window_interception,
3367 [SVM_EXIT_RDPMC] = rdpmc_interception,
3368 [SVM_EXIT_CPUID] = cpuid_interception,
3369 [SVM_EXIT_IRET] = iret_interception,
3370 [SVM_EXIT_INVD] = emulate_on_interception,
3371 [SVM_EXIT_PAUSE] = pause_interception,
3372 [SVM_EXIT_HLT] = halt_interception,
3373 [SVM_EXIT_INVLPG] = invlpg_interception,
3374 [SVM_EXIT_INVLPGA] = invlpga_interception,
3375 [SVM_EXIT_IOIO] = io_interception,
3376 [SVM_EXIT_MSR] = msr_interception,
3377 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
3378 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
3379 [SVM_EXIT_VMRUN] = vmrun_interception,
3380 [SVM_EXIT_VMMCALL] = vmmcall_interception,
3381 [SVM_EXIT_VMLOAD] = vmload_interception,
3382 [SVM_EXIT_VMSAVE] = vmsave_interception,
3383 [SVM_EXIT_STGI] = stgi_interception,
3384 [SVM_EXIT_CLGI] = clgi_interception,
3385 [SVM_EXIT_SKINIT] = skinit_interception,
3386 [SVM_EXIT_WBINVD] = wbinvd_interception,
3387 [SVM_EXIT_MONITOR] = monitor_interception,
3388 [SVM_EXIT_MWAIT] = mwait_interception,
3389 [SVM_EXIT_XSETBV] = xsetbv_interception,
3390 [SVM_EXIT_NPF] = pf_interception,
3393 static void dump_vmcb(struct kvm_vcpu *vcpu)
3395 struct vcpu_svm *svm = to_svm(vcpu);
3396 struct vmcb_control_area *control = &svm->vmcb->control;
3397 struct vmcb_save_area *save = &svm->vmcb->save;
3399 pr_err("VMCB Control Area:\n");
3400 pr_err("%-20s%04x\n", "cr_read:", control->intercept_cr & 0xffff);
3401 pr_err("%-20s%04x\n", "cr_write:", control->intercept_cr >> 16);
3402 pr_err("%-20s%04x\n", "dr_read:", control->intercept_dr & 0xffff);
3403 pr_err("%-20s%04x\n", "dr_write:", control->intercept_dr >> 16);
3404 pr_err("%-20s%08x\n", "exceptions:", control->intercept_exceptions);
3405 pr_err("%-20s%016llx\n", "intercepts:", control->intercept);
3406 pr_err("%-20s%d\n", "pause filter count:", control->pause_filter_count);
3407 pr_err("%-20s%016llx\n", "iopm_base_pa:", control->iopm_base_pa);
3408 pr_err("%-20s%016llx\n", "msrpm_base_pa:", control->msrpm_base_pa);
3409 pr_err("%-20s%016llx\n", "tsc_offset:", control->tsc_offset);
3410 pr_err("%-20s%d\n", "asid:", control->asid);
3411 pr_err("%-20s%d\n", "tlb_ctl:", control->tlb_ctl);
3412 pr_err("%-20s%08x\n", "int_ctl:", control->int_ctl);
3413 pr_err("%-20s%08x\n", "int_vector:", control->int_vector);
3414 pr_err("%-20s%08x\n", "int_state:", control->int_state);
3415 pr_err("%-20s%08x\n", "exit_code:", control->exit_code);
3416 pr_err("%-20s%016llx\n", "exit_info1:", control->exit_info_1);
3417 pr_err("%-20s%016llx\n", "exit_info2:", control->exit_info_2);
3418 pr_err("%-20s%08x\n", "exit_int_info:", control->exit_int_info);
3419 pr_err("%-20s%08x\n", "exit_int_info_err:", control->exit_int_info_err);
3420 pr_err("%-20s%lld\n", "nested_ctl:", control->nested_ctl);
3421 pr_err("%-20s%016llx\n", "nested_cr3:", control->nested_cr3);
3422 pr_err("%-20s%08x\n", "event_inj:", control->event_inj);
3423 pr_err("%-20s%08x\n", "event_inj_err:", control->event_inj_err);
3424 pr_err("%-20s%lld\n", "lbr_ctl:", control->lbr_ctl);
3425 pr_err("%-20s%016llx\n", "next_rip:", control->next_rip);
3426 pr_err("VMCB State Save Area:\n");
3427 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3429 save->es.selector, save->es.attrib,
3430 save->es.limit, save->es.base);
3431 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3433 save->cs.selector, save->cs.attrib,
3434 save->cs.limit, save->cs.base);
3435 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3437 save->ss.selector, save->ss.attrib,
3438 save->ss.limit, save->ss.base);
3439 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3441 save->ds.selector, save->ds.attrib,
3442 save->ds.limit, save->ds.base);
3443 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3445 save->fs.selector, save->fs.attrib,
3446 save->fs.limit, save->fs.base);
3447 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3449 save->gs.selector, save->gs.attrib,
3450 save->gs.limit, save->gs.base);
3451 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3453 save->gdtr.selector, save->gdtr.attrib,
3454 save->gdtr.limit, save->gdtr.base);
3455 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3457 save->ldtr.selector, save->ldtr.attrib,
3458 save->ldtr.limit, save->ldtr.base);
3459 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3461 save->idtr.selector, save->idtr.attrib,
3462 save->idtr.limit, save->idtr.base);
3463 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3465 save->tr.selector, save->tr.attrib,
3466 save->tr.limit, save->tr.base);
3467 pr_err("cpl: %d efer: %016llx\n",
3468 save->cpl, save->efer);
3469 pr_err("%-15s %016llx %-13s %016llx\n",
3470 "cr0:", save->cr0, "cr2:", save->cr2);
3471 pr_err("%-15s %016llx %-13s %016llx\n",
3472 "cr3:", save->cr3, "cr4:", save->cr4);
3473 pr_err("%-15s %016llx %-13s %016llx\n",
3474 "dr6:", save->dr6, "dr7:", save->dr7);
3475 pr_err("%-15s %016llx %-13s %016llx\n",
3476 "rip:", save->rip, "rflags:", save->rflags);
3477 pr_err("%-15s %016llx %-13s %016llx\n",
3478 "rsp:", save->rsp, "rax:", save->rax);
3479 pr_err("%-15s %016llx %-13s %016llx\n",
3480 "star:", save->star, "lstar:", save->lstar);
3481 pr_err("%-15s %016llx %-13s %016llx\n",
3482 "cstar:", save->cstar, "sfmask:", save->sfmask);
3483 pr_err("%-15s %016llx %-13s %016llx\n",
3484 "kernel_gs_base:", save->kernel_gs_base,
3485 "sysenter_cs:", save->sysenter_cs);
3486 pr_err("%-15s %016llx %-13s %016llx\n",
3487 "sysenter_esp:", save->sysenter_esp,
3488 "sysenter_eip:", save->sysenter_eip);
3489 pr_err("%-15s %016llx %-13s %016llx\n",
3490 "gpat:", save->g_pat, "dbgctl:", save->dbgctl);
3491 pr_err("%-15s %016llx %-13s %016llx\n",
3492 "br_from:", save->br_from, "br_to:", save->br_to);
3493 pr_err("%-15s %016llx %-13s %016llx\n",
3494 "excp_from:", save->last_excp_from,
3495 "excp_to:", save->last_excp_to);
3498 static void svm_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
3500 struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control;
3502 *info1 = control->exit_info_1;
3503 *info2 = control->exit_info_2;
3506 static int handle_exit(struct kvm_vcpu *vcpu)
3508 struct vcpu_svm *svm = to_svm(vcpu);
3509 struct kvm_run *kvm_run = vcpu->run;
3510 u32 exit_code = svm->vmcb->control.exit_code;
3512 if (!is_cr_intercept(svm, INTERCEPT_CR0_WRITE))
3513 vcpu->arch.cr0 = svm->vmcb->save.cr0;
3515 vcpu->arch.cr3 = svm->vmcb->save.cr3;
3517 if (unlikely(svm->nested.exit_required)) {
3518 nested_svm_vmexit(svm);
3519 svm->nested.exit_required = false;
3524 if (is_guest_mode(vcpu)) {
3527 trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code,
3528 svm->vmcb->control.exit_info_1,
3529 svm->vmcb->control.exit_info_2,
3530 svm->vmcb->control.exit_int_info,
3531 svm->vmcb->control.exit_int_info_err,
3534 vmexit = nested_svm_exit_special(svm);
3536 if (vmexit == NESTED_EXIT_CONTINUE)
3537 vmexit = nested_svm_exit_handled(svm);
3539 if (vmexit == NESTED_EXIT_DONE)
3543 svm_complete_interrupts(svm);
3545 if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
3546 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3547 kvm_run->fail_entry.hardware_entry_failure_reason
3548 = svm->vmcb->control.exit_code;
3549 pr_err("KVM: FAILED VMRUN WITH VMCB:\n");
3554 if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
3555 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
3556 exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH &&
3557 exit_code != SVM_EXIT_INTR && exit_code != SVM_EXIT_NMI)
3558 printk(KERN_ERR "%s: unexpected exit_int_info 0x%x "
3560 __func__, svm->vmcb->control.exit_int_info,
3563 if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
3564 || !svm_exit_handlers[exit_code]) {
3565 WARN_ONCE(1, "svm: unexpected exit reason 0x%x\n", exit_code);
3566 kvm_queue_exception(vcpu, UD_VECTOR);
3570 return svm_exit_handlers[exit_code](svm);
3573 static void reload_tss(struct kvm_vcpu *vcpu)
3575 int cpu = raw_smp_processor_id();
3577 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
3578 sd->tss_desc->type = 9; /* available 32/64-bit TSS */
3582 static void pre_svm_run(struct vcpu_svm *svm)
3584 int cpu = raw_smp_processor_id();
3586 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
3588 /* FIXME: handle wraparound of asid_generation */
3589 if (svm->asid_generation != sd->asid_generation)
3593 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
3595 struct vcpu_svm *svm = to_svm(vcpu);
3597 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
3598 vcpu->arch.hflags |= HF_NMI_MASK;
3599 set_intercept(svm, INTERCEPT_IRET);
3600 ++vcpu->stat.nmi_injections;
3603 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
3605 struct vmcb_control_area *control;
3607 control = &svm->vmcb->control;
3608 control->int_vector = irq;
3609 control->int_ctl &= ~V_INTR_PRIO_MASK;
3610 control->int_ctl |= V_IRQ_MASK |
3611 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
3612 mark_dirty(svm->vmcb, VMCB_INTR);
3615 static void svm_set_irq(struct kvm_vcpu *vcpu)
3617 struct vcpu_svm *svm = to_svm(vcpu);
3619 BUG_ON(!(gif_set(svm)));
3621 trace_kvm_inj_virq(vcpu->arch.interrupt.nr);
3622 ++vcpu->stat.irq_injections;
3624 svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
3625 SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
3628 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3630 struct vcpu_svm *svm = to_svm(vcpu);
3632 if (is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK))
3635 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
3641 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
3644 static void svm_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
3649 static int svm_vm_has_apicv(struct kvm *kvm)
3654 static void svm_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
3659 static void svm_sync_pir_to_irr(struct kvm_vcpu *vcpu)
3664 static int svm_nmi_allowed(struct kvm_vcpu *vcpu)
3666 struct vcpu_svm *svm = to_svm(vcpu);
3667 struct vmcb *vmcb = svm->vmcb;
3669 ret = !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
3670 !(svm->vcpu.arch.hflags & HF_NMI_MASK);
3671 ret = ret && gif_set(svm) && nested_svm_nmi(svm);
3676 static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
3678 struct vcpu_svm *svm = to_svm(vcpu);
3680 return !!(svm->vcpu.arch.hflags & HF_NMI_MASK);
3683 static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
3685 struct vcpu_svm *svm = to_svm(vcpu);
3688 svm->vcpu.arch.hflags |= HF_NMI_MASK;
3689 set_intercept(svm, INTERCEPT_IRET);
3691 svm->vcpu.arch.hflags &= ~HF_NMI_MASK;
3692 clr_intercept(svm, INTERCEPT_IRET);
3696 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu)
3698 struct vcpu_svm *svm = to_svm(vcpu);
3699 struct vmcb *vmcb = svm->vmcb;
3702 if (!gif_set(svm) ||
3703 (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK))
3706 ret = !!(kvm_get_rflags(vcpu) & X86_EFLAGS_IF);
3708 if (is_guest_mode(vcpu))
3709 return ret && !(svm->vcpu.arch.hflags & HF_VINTR_MASK);
3714 static void enable_irq_window(struct kvm_vcpu *vcpu)
3716 struct vcpu_svm *svm = to_svm(vcpu);
3719 * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
3720 * 1, because that's a separate STGI/VMRUN intercept. The next time we
3721 * get that intercept, this function will be called again though and
3722 * we'll get the vintr intercept.
3724 if (gif_set(svm) && nested_svm_intr(svm)) {
3726 svm_inject_irq(svm, 0x0);
3730 static void enable_nmi_window(struct kvm_vcpu *vcpu)
3732 struct vcpu_svm *svm = to_svm(vcpu);
3734 if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
3736 return; /* IRET will cause a vm exit */
3739 * Something prevents NMI from been injected. Single step over possible
3740 * problem (IRET or exception injection or interrupt shadow)
3742 svm->nmi_singlestep = true;
3743 svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
3744 update_db_bp_intercept(vcpu);
3747 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
3752 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
3754 struct vcpu_svm *svm = to_svm(vcpu);
3756 if (static_cpu_has(X86_FEATURE_FLUSHBYASID))
3757 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
3759 svm->asid_generation--;
3762 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
3766 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
3768 struct vcpu_svm *svm = to_svm(vcpu);
3770 if (is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK))
3773 if (!is_cr_intercept(svm, INTERCEPT_CR8_WRITE)) {
3774 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
3775 kvm_set_cr8(vcpu, cr8);
3779 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
3781 struct vcpu_svm *svm = to_svm(vcpu);
3784 if (is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK))
3787 cr8 = kvm_get_cr8(vcpu);
3788 svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
3789 svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
3792 static void svm_complete_interrupts(struct vcpu_svm *svm)
3796 u32 exitintinfo = svm->vmcb->control.exit_int_info;
3797 unsigned int3_injected = svm->int3_injected;
3799 svm->int3_injected = 0;
3802 * If we've made progress since setting HF_IRET_MASK, we've
3803 * executed an IRET and can allow NMI injection.
3805 if ((svm->vcpu.arch.hflags & HF_IRET_MASK)
3806 && kvm_rip_read(&svm->vcpu) != svm->nmi_iret_rip) {
3807 svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
3808 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3811 svm->vcpu.arch.nmi_injected = false;
3812 kvm_clear_exception_queue(&svm->vcpu);
3813 kvm_clear_interrupt_queue(&svm->vcpu);
3815 if (!(exitintinfo & SVM_EXITINTINFO_VALID))
3818 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3820 vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
3821 type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
3824 case SVM_EXITINTINFO_TYPE_NMI:
3825 svm->vcpu.arch.nmi_injected = true;
3827 case SVM_EXITINTINFO_TYPE_EXEPT:
3829 * In case of software exceptions, do not reinject the vector,
3830 * but re-execute the instruction instead. Rewind RIP first
3831 * if we emulated INT3 before.
3833 if (kvm_exception_is_soft(vector)) {
3834 if (vector == BP_VECTOR && int3_injected &&
3835 kvm_is_linear_rip(&svm->vcpu, svm->int3_rip))
3836 kvm_rip_write(&svm->vcpu,
3837 kvm_rip_read(&svm->vcpu) -
3841 if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
3842 u32 err = svm->vmcb->control.exit_int_info_err;
3843 kvm_requeue_exception_e(&svm->vcpu, vector, err);
3846 kvm_requeue_exception(&svm->vcpu, vector);
3848 case SVM_EXITINTINFO_TYPE_INTR:
3849 kvm_queue_interrupt(&svm->vcpu, vector, false);
3856 static void svm_cancel_injection(struct kvm_vcpu *vcpu)
3858 struct vcpu_svm *svm = to_svm(vcpu);
3859 struct vmcb_control_area *control = &svm->vmcb->control;
3861 control->exit_int_info = control->event_inj;
3862 control->exit_int_info_err = control->event_inj_err;
3863 control->event_inj = 0;
3864 svm_complete_interrupts(svm);
3867 static void svm_vcpu_run(struct kvm_vcpu *vcpu)
3869 struct vcpu_svm *svm = to_svm(vcpu);
3871 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
3872 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
3873 svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
3876 * A vmexit emulation is required before the vcpu can be executed
3879 if (unlikely(svm->nested.exit_required))
3884 sync_lapic_to_cr8(vcpu);
3886 svm->vmcb->save.cr2 = vcpu->arch.cr2;
3893 "push %%" _ASM_BP "; \n\t"
3894 "mov %c[rbx](%[svm]), %%" _ASM_BX " \n\t"
3895 "mov %c[rcx](%[svm]), %%" _ASM_CX " \n\t"
3896 "mov %c[rdx](%[svm]), %%" _ASM_DX " \n\t"
3897 "mov %c[rsi](%[svm]), %%" _ASM_SI " \n\t"
3898 "mov %c[rdi](%[svm]), %%" _ASM_DI " \n\t"
3899 "mov %c[rbp](%[svm]), %%" _ASM_BP " \n\t"
3900 #ifdef CONFIG_X86_64
3901 "mov %c[r8](%[svm]), %%r8 \n\t"
3902 "mov %c[r9](%[svm]), %%r9 \n\t"
3903 "mov %c[r10](%[svm]), %%r10 \n\t"
3904 "mov %c[r11](%[svm]), %%r11 \n\t"
3905 "mov %c[r12](%[svm]), %%r12 \n\t"
3906 "mov %c[r13](%[svm]), %%r13 \n\t"
3907 "mov %c[r14](%[svm]), %%r14 \n\t"
3908 "mov %c[r15](%[svm]), %%r15 \n\t"
3911 /* Enter guest mode */
3912 "push %%" _ASM_AX " \n\t"
3913 "mov %c[vmcb](%[svm]), %%" _ASM_AX " \n\t"
3914 __ex(SVM_VMLOAD) "\n\t"
3915 __ex(SVM_VMRUN) "\n\t"
3916 __ex(SVM_VMSAVE) "\n\t"
3917 "pop %%" _ASM_AX " \n\t"
3919 /* Save guest registers, load host registers */
3920 "mov %%" _ASM_BX ", %c[rbx](%[svm]) \n\t"
3921 "mov %%" _ASM_CX ", %c[rcx](%[svm]) \n\t"
3922 "mov %%" _ASM_DX ", %c[rdx](%[svm]) \n\t"
3923 "mov %%" _ASM_SI ", %c[rsi](%[svm]) \n\t"
3924 "mov %%" _ASM_DI ", %c[rdi](%[svm]) \n\t"
3925 "mov %%" _ASM_BP ", %c[rbp](%[svm]) \n\t"
3926 #ifdef CONFIG_X86_64
3927 "mov %%r8, %c[r8](%[svm]) \n\t"
3928 "mov %%r9, %c[r9](%[svm]) \n\t"
3929 "mov %%r10, %c[r10](%[svm]) \n\t"
3930 "mov %%r11, %c[r11](%[svm]) \n\t"
3931 "mov %%r12, %c[r12](%[svm]) \n\t"
3932 "mov %%r13, %c[r13](%[svm]) \n\t"
3933 "mov %%r14, %c[r14](%[svm]) \n\t"
3934 "mov %%r15, %c[r15](%[svm]) \n\t"
3939 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
3940 [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
3941 [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
3942 [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
3943 [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
3944 [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
3945 [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
3946 #ifdef CONFIG_X86_64
3947 , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
3948 [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
3949 [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
3950 [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
3951 [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
3952 [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
3953 [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
3954 [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
3957 #ifdef CONFIG_X86_64
3958 , "rbx", "rcx", "rdx", "rsi", "rdi"
3959 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
3961 , "ebx", "ecx", "edx", "esi", "edi"
3965 #ifdef CONFIG_X86_64
3966 wrmsrl(MSR_GS_BASE, svm->host.gs_base);
3968 loadsegment(fs, svm->host.fs);
3969 #ifndef CONFIG_X86_32_LAZY_GS
3970 loadsegment(gs, svm->host.gs);
3976 local_irq_disable();
3978 vcpu->arch.cr2 = svm->vmcb->save.cr2;
3979 vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
3980 vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
3981 vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
3983 trace_kvm_exit(svm->vmcb->control.exit_code, vcpu, KVM_ISA_SVM);
3985 if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
3986 kvm_before_handle_nmi(&svm->vcpu);
3990 /* Any pending NMI will happen here */
3992 if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
3993 kvm_after_handle_nmi(&svm->vcpu);
3995 sync_cr8_to_lapic(vcpu);
3999 svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
4001 /* if exit due to PF check for async PF */
4002 if (svm->vmcb->control.exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR)
4003 svm->apf_reason = kvm_read_and_reset_pf_reason();
4006 vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
4007 vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
4011 * We need to handle MC intercepts here before the vcpu has a chance to
4012 * change the physical cpu
4014 if (unlikely(svm->vmcb->control.exit_code ==
4015 SVM_EXIT_EXCP_BASE + MC_VECTOR))
4016 svm_handle_mce(svm);
4018 mark_all_clean(svm->vmcb);
4021 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
4023 struct vcpu_svm *svm = to_svm(vcpu);
4025 svm->vmcb->save.cr3 = root;
4026 mark_dirty(svm->vmcb, VMCB_CR);
4027 svm_flush_tlb(vcpu);
4030 static void set_tdp_cr3(struct kvm_vcpu *vcpu, unsigned long root)
4032 struct vcpu_svm *svm = to_svm(vcpu);
4034 svm->vmcb->control.nested_cr3 = root;
4035 mark_dirty(svm->vmcb, VMCB_NPT);
4037 /* Also sync guest cr3 here in case we live migrate */
4038 svm->vmcb->save.cr3 = kvm_read_cr3(vcpu);
4039 mark_dirty(svm->vmcb, VMCB_CR);
4041 svm_flush_tlb(vcpu);
4044 static int is_disabled(void)
4048 rdmsrl(MSR_VM_CR, vm_cr);
4049 if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
4056 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4059 * Patch in the VMMCALL instruction:
4061 hypercall[0] = 0x0f;
4062 hypercall[1] = 0x01;
4063 hypercall[2] = 0xd9;
4066 static void svm_check_processor_compat(void *rtn)
4071 static bool svm_cpu_has_accelerated_tpr(void)
4076 static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
4081 static void svm_cpuid_update(struct kvm_vcpu *vcpu)
4085 static void svm_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
4090 entry->ecx |= (1 << 2); /* Set SVM bit */
4093 entry->eax = 1; /* SVM revision 1 */
4094 entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
4095 ASID emulation to nested SVM */
4096 entry->ecx = 0; /* Reserved */
4097 entry->edx = 0; /* Per default do not support any
4098 additional features */
4100 /* Support next_rip if host supports it */
4101 if (boot_cpu_has(X86_FEATURE_NRIPS))
4102 entry->edx |= SVM_FEATURE_NRIP;
4104 /* Support NPT for the guest if enabled */
4106 entry->edx |= SVM_FEATURE_NPT;
4112 static int svm_get_lpage_level(void)
4114 return PT_PDPE_LEVEL;
4117 static bool svm_rdtscp_supported(void)
4122 static bool svm_invpcid_supported(void)
4127 static bool svm_mpx_supported(void)
4132 static bool svm_xsaves_supported(void)
4137 static bool svm_has_wbinvd_exit(void)
4142 static void svm_fpu_deactivate(struct kvm_vcpu *vcpu)
4144 struct vcpu_svm *svm = to_svm(vcpu);
4146 set_exception_intercept(svm, NM_VECTOR);
4147 update_cr0_intercept(svm);
4150 #define PRE_EX(exit) { .exit_code = (exit), \
4151 .stage = X86_ICPT_PRE_EXCEPT, }
4152 #define POST_EX(exit) { .exit_code = (exit), \
4153 .stage = X86_ICPT_POST_EXCEPT, }
4154 #define POST_MEM(exit) { .exit_code = (exit), \
4155 .stage = X86_ICPT_POST_MEMACCESS, }
4157 static const struct __x86_intercept {
4159 enum x86_intercept_stage stage;
4160 } x86_intercept_map[] = {
4161 [x86_intercept_cr_read] = POST_EX(SVM_EXIT_READ_CR0),
4162 [x86_intercept_cr_write] = POST_EX(SVM_EXIT_WRITE_CR0),
4163 [x86_intercept_clts] = POST_EX(SVM_EXIT_WRITE_CR0),
4164 [x86_intercept_lmsw] = POST_EX(SVM_EXIT_WRITE_CR0),
4165 [x86_intercept_smsw] = POST_EX(SVM_EXIT_READ_CR0),
4166 [x86_intercept_dr_read] = POST_EX(SVM_EXIT_READ_DR0),
4167 [x86_intercept_dr_write] = POST_EX(SVM_EXIT_WRITE_DR0),
4168 [x86_intercept_sldt] = POST_EX(SVM_EXIT_LDTR_READ),
4169 [x86_intercept_str] = POST_EX(SVM_EXIT_TR_READ),
4170 [x86_intercept_lldt] = POST_EX(SVM_EXIT_LDTR_WRITE),
4171 [x86_intercept_ltr] = POST_EX(SVM_EXIT_TR_WRITE),
4172 [x86_intercept_sgdt] = POST_EX(SVM_EXIT_GDTR_READ),
4173 [x86_intercept_sidt] = POST_EX(SVM_EXIT_IDTR_READ),
4174 [x86_intercept_lgdt] = POST_EX(SVM_EXIT_GDTR_WRITE),
4175 [x86_intercept_lidt] = POST_EX(SVM_EXIT_IDTR_WRITE),
4176 [x86_intercept_vmrun] = POST_EX(SVM_EXIT_VMRUN),
4177 [x86_intercept_vmmcall] = POST_EX(SVM_EXIT_VMMCALL),
4178 [x86_intercept_vmload] = POST_EX(SVM_EXIT_VMLOAD),
4179 [x86_intercept_vmsave] = POST_EX(SVM_EXIT_VMSAVE),
4180 [x86_intercept_stgi] = POST_EX(SVM_EXIT_STGI),
4181 [x86_intercept_clgi] = POST_EX(SVM_EXIT_CLGI),
4182 [x86_intercept_skinit] = POST_EX(SVM_EXIT_SKINIT),
4183 [x86_intercept_invlpga] = POST_EX(SVM_EXIT_INVLPGA),
4184 [x86_intercept_rdtscp] = POST_EX(SVM_EXIT_RDTSCP),
4185 [x86_intercept_monitor] = POST_MEM(SVM_EXIT_MONITOR),
4186 [x86_intercept_mwait] = POST_EX(SVM_EXIT_MWAIT),
4187 [x86_intercept_invlpg] = POST_EX(SVM_EXIT_INVLPG),
4188 [x86_intercept_invd] = POST_EX(SVM_EXIT_INVD),
4189 [x86_intercept_wbinvd] = POST_EX(SVM_EXIT_WBINVD),
4190 [x86_intercept_wrmsr] = POST_EX(SVM_EXIT_MSR),
4191 [x86_intercept_rdtsc] = POST_EX(SVM_EXIT_RDTSC),
4192 [x86_intercept_rdmsr] = POST_EX(SVM_EXIT_MSR),
4193 [x86_intercept_rdpmc] = POST_EX(SVM_EXIT_RDPMC),
4194 [x86_intercept_cpuid] = PRE_EX(SVM_EXIT_CPUID),
4195 [x86_intercept_rsm] = PRE_EX(SVM_EXIT_RSM),
4196 [x86_intercept_pause] = PRE_EX(SVM_EXIT_PAUSE),
4197 [x86_intercept_pushf] = PRE_EX(SVM_EXIT_PUSHF),
4198 [x86_intercept_popf] = PRE_EX(SVM_EXIT_POPF),
4199 [x86_intercept_intn] = PRE_EX(SVM_EXIT_SWINT),
4200 [x86_intercept_iret] = PRE_EX(SVM_EXIT_IRET),
4201 [x86_intercept_icebp] = PRE_EX(SVM_EXIT_ICEBP),
4202 [x86_intercept_hlt] = POST_EX(SVM_EXIT_HLT),
4203 [x86_intercept_in] = POST_EX(SVM_EXIT_IOIO),
4204 [x86_intercept_ins] = POST_EX(SVM_EXIT_IOIO),
4205 [x86_intercept_out] = POST_EX(SVM_EXIT_IOIO),
4206 [x86_intercept_outs] = POST_EX(SVM_EXIT_IOIO),
4213 static int svm_check_intercept(struct kvm_vcpu *vcpu,
4214 struct x86_instruction_info *info,
4215 enum x86_intercept_stage stage)
4217 struct vcpu_svm *svm = to_svm(vcpu);
4218 int vmexit, ret = X86EMUL_CONTINUE;
4219 struct __x86_intercept icpt_info;
4220 struct vmcb *vmcb = svm->vmcb;
4222 if (info->intercept >= ARRAY_SIZE(x86_intercept_map))
4225 icpt_info = x86_intercept_map[info->intercept];
4227 if (stage != icpt_info.stage)
4230 switch (icpt_info.exit_code) {
4231 case SVM_EXIT_READ_CR0:
4232 if (info->intercept == x86_intercept_cr_read)
4233 icpt_info.exit_code += info->modrm_reg;
4235 case SVM_EXIT_WRITE_CR0: {
4236 unsigned long cr0, val;
4239 if (info->intercept == x86_intercept_cr_write)
4240 icpt_info.exit_code += info->modrm_reg;
4242 if (icpt_info.exit_code != SVM_EXIT_WRITE_CR0 ||
4243 info->intercept == x86_intercept_clts)
4246 intercept = svm->nested.intercept;
4248 if (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0)))
4251 cr0 = vcpu->arch.cr0 & ~SVM_CR0_SELECTIVE_MASK;
4252 val = info->src_val & ~SVM_CR0_SELECTIVE_MASK;
4254 if (info->intercept == x86_intercept_lmsw) {
4257 /* lmsw can't clear PE - catch this here */
4258 if (cr0 & X86_CR0_PE)
4263 icpt_info.exit_code = SVM_EXIT_CR0_SEL_WRITE;
4267 case SVM_EXIT_READ_DR0:
4268 case SVM_EXIT_WRITE_DR0:
4269 icpt_info.exit_code += info->modrm_reg;
4272 if (info->intercept == x86_intercept_wrmsr)
4273 vmcb->control.exit_info_1 = 1;
4275 vmcb->control.exit_info_1 = 0;
4277 case SVM_EXIT_PAUSE:
4279 * We get this for NOP only, but pause
4280 * is rep not, check this here
4282 if (info->rep_prefix != REPE_PREFIX)
4284 case SVM_EXIT_IOIO: {
4288 if (info->intercept == x86_intercept_in ||
4289 info->intercept == x86_intercept_ins) {
4290 exit_info = ((info->src_val & 0xffff) << 16) |
4292 bytes = info->dst_bytes;
4294 exit_info = (info->dst_val & 0xffff) << 16;
4295 bytes = info->src_bytes;
4298 if (info->intercept == x86_intercept_outs ||
4299 info->intercept == x86_intercept_ins)
4300 exit_info |= SVM_IOIO_STR_MASK;
4302 if (info->rep_prefix)
4303 exit_info |= SVM_IOIO_REP_MASK;
4305 bytes = min(bytes, 4u);
4307 exit_info |= bytes << SVM_IOIO_SIZE_SHIFT;
4309 exit_info |= (u32)info->ad_bytes << (SVM_IOIO_ASIZE_SHIFT - 1);
4311 vmcb->control.exit_info_1 = exit_info;
4312 vmcb->control.exit_info_2 = info->next_rip;
4320 vmcb->control.next_rip = info->next_rip;
4321 vmcb->control.exit_code = icpt_info.exit_code;
4322 vmexit = nested_svm_exit_handled(svm);
4324 ret = (vmexit == NESTED_EXIT_DONE) ? X86EMUL_INTERCEPTED
4331 static void svm_handle_external_intr(struct kvm_vcpu *vcpu)
4336 static void svm_sched_in(struct kvm_vcpu *vcpu, int cpu)
4340 static struct kvm_x86_ops svm_x86_ops = {
4341 .cpu_has_kvm_support = has_svm,
4342 .disabled_by_bios = is_disabled,
4343 .hardware_setup = svm_hardware_setup,
4344 .hardware_unsetup = svm_hardware_unsetup,
4345 .check_processor_compatibility = svm_check_processor_compat,
4346 .hardware_enable = svm_hardware_enable,
4347 .hardware_disable = svm_hardware_disable,
4348 .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
4350 .vcpu_create = svm_create_vcpu,
4351 .vcpu_free = svm_free_vcpu,
4352 .vcpu_reset = svm_vcpu_reset,
4354 .prepare_guest_switch = svm_prepare_guest_switch,
4355 .vcpu_load = svm_vcpu_load,
4356 .vcpu_put = svm_vcpu_put,
4358 .update_db_bp_intercept = update_db_bp_intercept,
4359 .get_msr = svm_get_msr,
4360 .set_msr = svm_set_msr,
4361 .get_segment_base = svm_get_segment_base,
4362 .get_segment = svm_get_segment,
4363 .set_segment = svm_set_segment,
4364 .get_cpl = svm_get_cpl,
4365 .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
4366 .decache_cr0_guest_bits = svm_decache_cr0_guest_bits,
4367 .decache_cr3 = svm_decache_cr3,
4368 .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
4369 .set_cr0 = svm_set_cr0,
4370 .set_cr3 = svm_set_cr3,
4371 .set_cr4 = svm_set_cr4,
4372 .set_efer = svm_set_efer,
4373 .get_idt = svm_get_idt,
4374 .set_idt = svm_set_idt,
4375 .get_gdt = svm_get_gdt,
4376 .set_gdt = svm_set_gdt,
4377 .get_dr6 = svm_get_dr6,
4378 .set_dr6 = svm_set_dr6,
4379 .set_dr7 = svm_set_dr7,
4380 .sync_dirty_debug_regs = svm_sync_dirty_debug_regs,
4381 .cache_reg = svm_cache_reg,
4382 .get_rflags = svm_get_rflags,
4383 .set_rflags = svm_set_rflags,
4384 .fpu_activate = svm_fpu_activate,
4385 .fpu_deactivate = svm_fpu_deactivate,
4387 .tlb_flush = svm_flush_tlb,
4389 .run = svm_vcpu_run,
4390 .handle_exit = handle_exit,
4391 .skip_emulated_instruction = skip_emulated_instruction,
4392 .set_interrupt_shadow = svm_set_interrupt_shadow,
4393 .get_interrupt_shadow = svm_get_interrupt_shadow,
4394 .patch_hypercall = svm_patch_hypercall,
4395 .set_irq = svm_set_irq,
4396 .set_nmi = svm_inject_nmi,
4397 .queue_exception = svm_queue_exception,
4398 .cancel_injection = svm_cancel_injection,
4399 .interrupt_allowed = svm_interrupt_allowed,
4400 .nmi_allowed = svm_nmi_allowed,
4401 .get_nmi_mask = svm_get_nmi_mask,
4402 .set_nmi_mask = svm_set_nmi_mask,
4403 .enable_nmi_window = enable_nmi_window,
4404 .enable_irq_window = enable_irq_window,
4405 .update_cr8_intercept = update_cr8_intercept,
4406 .set_virtual_x2apic_mode = svm_set_virtual_x2apic_mode,
4407 .vm_has_apicv = svm_vm_has_apicv,
4408 .load_eoi_exitmap = svm_load_eoi_exitmap,
4409 .sync_pir_to_irr = svm_sync_pir_to_irr,
4411 .set_tss_addr = svm_set_tss_addr,
4412 .get_tdp_level = get_npt_level,
4413 .get_mt_mask = svm_get_mt_mask,
4415 .get_exit_info = svm_get_exit_info,
4417 .get_lpage_level = svm_get_lpage_level,
4419 .cpuid_update = svm_cpuid_update,
4421 .rdtscp_supported = svm_rdtscp_supported,
4422 .invpcid_supported = svm_invpcid_supported,
4423 .mpx_supported = svm_mpx_supported,
4424 .xsaves_supported = svm_xsaves_supported,
4426 .set_supported_cpuid = svm_set_supported_cpuid,
4428 .has_wbinvd_exit = svm_has_wbinvd_exit,
4430 .set_tsc_khz = svm_set_tsc_khz,
4431 .read_tsc_offset = svm_read_tsc_offset,
4432 .write_tsc_offset = svm_write_tsc_offset,
4433 .adjust_tsc_offset = svm_adjust_tsc_offset,
4434 .compute_tsc_offset = svm_compute_tsc_offset,
4435 .read_l1_tsc = svm_read_l1_tsc,
4437 .set_tdp_cr3 = set_tdp_cr3,
4439 .check_intercept = svm_check_intercept,
4440 .handle_external_intr = svm_handle_external_intr,
4442 .sched_in = svm_sched_in,
4445 static int __init svm_init(void)
4447 return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
4448 __alignof__(struct vcpu_svm), THIS_MODULE);
4451 static void __exit svm_exit(void)
4456 module_init(svm_init)
4457 module_exit(svm_exit)