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"
26 #include <linux/module.h>
27 #include <linux/mod_devicetable.h>
28 #include <linux/kernel.h>
29 #include <linux/vmalloc.h>
30 #include <linux/highmem.h>
31 #include <linux/sched.h>
32 #include <linux/trace_events.h>
33 #include <linux/slab.h>
35 #include <asm/perf_event.h>
36 #include <asm/tlbflush.h>
38 #include <asm/debugreg.h>
39 #include <asm/kvm_para.h>
41 #include <asm/virtext.h>
44 #define __ex(x) __kvm_handle_fault_on_reboot(x)
46 MODULE_AUTHOR("Qumranet");
47 MODULE_LICENSE("GPL");
49 static const struct x86_cpu_id svm_cpu_id[] = {
50 X86_FEATURE_MATCH(X86_FEATURE_SVM),
53 MODULE_DEVICE_TABLE(x86cpu, svm_cpu_id);
55 #define IOPM_ALLOC_ORDER 2
56 #define MSRPM_ALLOC_ORDER 1
58 #define SEG_TYPE_LDT 2
59 #define SEG_TYPE_BUSY_TSS16 3
61 #define SVM_FEATURE_NPT (1 << 0)
62 #define SVM_FEATURE_LBRV (1 << 1)
63 #define SVM_FEATURE_SVML (1 << 2)
64 #define SVM_FEATURE_NRIP (1 << 3)
65 #define SVM_FEATURE_TSC_RATE (1 << 4)
66 #define SVM_FEATURE_VMCB_CLEAN (1 << 5)
67 #define SVM_FEATURE_FLUSH_ASID (1 << 6)
68 #define SVM_FEATURE_DECODE_ASSIST (1 << 7)
69 #define SVM_FEATURE_PAUSE_FILTER (1 << 10)
71 #define NESTED_EXIT_HOST 0 /* Exit handled on host level */
72 #define NESTED_EXIT_DONE 1 /* Exit caused nested vmexit */
73 #define NESTED_EXIT_CONTINUE 2 /* Further checks needed */
75 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
77 #define TSC_RATIO_RSVD 0xffffff0000000000ULL
78 #define TSC_RATIO_MIN 0x0000000000000001ULL
79 #define TSC_RATIO_MAX 0x000000ffffffffffULL
81 static bool erratum_383_found __read_mostly;
83 static const u32 host_save_user_msrs[] = {
85 MSR_STAR, MSR_LSTAR, MSR_CSTAR, MSR_SYSCALL_MASK, MSR_KERNEL_GS_BASE,
88 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
91 #define NR_HOST_SAVE_USER_MSRS ARRAY_SIZE(host_save_user_msrs)
101 /* These are the merged vectors */
104 /* gpa pointers to the real vectors */
108 /* A VMEXIT is required but not yet emulated */
111 /* cache for intercepts of the guest */
114 u32 intercept_exceptions;
117 /* Nested Paging related state */
121 #define MSRPM_OFFSETS 16
122 static u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
125 * Set osvw_len to higher value when updated Revision Guides
126 * are published and we know what the new status bits are
128 static uint64_t osvw_len = 4, osvw_status;
131 struct kvm_vcpu vcpu;
133 unsigned long vmcb_pa;
134 struct svm_cpu_data *svm_data;
135 uint64_t asid_generation;
136 uint64_t sysenter_esp;
137 uint64_t sysenter_eip;
141 u64 host_user_msrs[NR_HOST_SAVE_USER_MSRS];
153 struct nested_state nested;
157 unsigned int3_injected;
158 unsigned long int3_rip;
164 static DEFINE_PER_CPU(u64, current_tsc_ratio);
165 #define TSC_RATIO_DEFAULT 0x0100000000ULL
167 #define MSR_INVALID 0xffffffffU
169 static const struct svm_direct_access_msrs {
170 u32 index; /* Index of the MSR */
171 bool always; /* True if intercept is always on */
172 } direct_access_msrs[] = {
173 { .index = MSR_STAR, .always = true },
174 { .index = MSR_IA32_SYSENTER_CS, .always = true },
176 { .index = MSR_GS_BASE, .always = true },
177 { .index = MSR_FS_BASE, .always = true },
178 { .index = MSR_KERNEL_GS_BASE, .always = true },
179 { .index = MSR_LSTAR, .always = true },
180 { .index = MSR_CSTAR, .always = true },
181 { .index = MSR_SYSCALL_MASK, .always = true },
183 { .index = MSR_IA32_LASTBRANCHFROMIP, .always = false },
184 { .index = MSR_IA32_LASTBRANCHTOIP, .always = false },
185 { .index = MSR_IA32_LASTINTFROMIP, .always = false },
186 { .index = MSR_IA32_LASTINTTOIP, .always = false },
187 { .index = MSR_INVALID, .always = false },
190 /* enable NPT for AMD64 and X86 with PAE */
191 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
192 static bool npt_enabled = true;
194 static bool npt_enabled;
197 /* allow nested paging (virtualized MMU) for all guests */
198 static int npt = true;
199 module_param(npt, int, S_IRUGO);
201 /* allow nested virtualization in KVM/SVM */
202 static int nested = true;
203 module_param(nested, int, S_IRUGO);
205 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0);
206 static void svm_flush_tlb(struct kvm_vcpu *vcpu);
207 static void svm_complete_interrupts(struct vcpu_svm *svm);
209 static int nested_svm_exit_handled(struct vcpu_svm *svm);
210 static int nested_svm_intercept(struct vcpu_svm *svm);
211 static int nested_svm_vmexit(struct vcpu_svm *svm);
212 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
213 bool has_error_code, u32 error_code);
214 static u64 __scale_tsc(u64 ratio, u64 tsc);
217 VMCB_INTERCEPTS, /* Intercept vectors, TSC offset,
218 pause filter count */
219 VMCB_PERM_MAP, /* IOPM Base and MSRPM Base */
220 VMCB_ASID, /* ASID */
221 VMCB_INTR, /* int_ctl, int_vector */
222 VMCB_NPT, /* npt_en, nCR3, gPAT */
223 VMCB_CR, /* CR0, CR3, CR4, EFER */
224 VMCB_DR, /* DR6, DR7 */
225 VMCB_DT, /* GDT, IDT */
226 VMCB_SEG, /* CS, DS, SS, ES, CPL */
227 VMCB_CR2, /* CR2 only */
228 VMCB_LBR, /* DBGCTL, BR_FROM, BR_TO, LAST_EX_FROM, LAST_EX_TO */
232 /* TPR and CR2 are always written before VMRUN */
233 #define VMCB_ALWAYS_DIRTY_MASK ((1U << VMCB_INTR) | (1U << VMCB_CR2))
235 static inline void mark_all_dirty(struct vmcb *vmcb)
237 vmcb->control.clean = 0;
240 static inline void mark_all_clean(struct vmcb *vmcb)
242 vmcb->control.clean = ((1 << VMCB_DIRTY_MAX) - 1)
243 & ~VMCB_ALWAYS_DIRTY_MASK;
246 static inline void mark_dirty(struct vmcb *vmcb, int bit)
248 vmcb->control.clean &= ~(1 << bit);
251 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
253 return container_of(vcpu, struct vcpu_svm, vcpu);
256 static void recalc_intercepts(struct vcpu_svm *svm)
258 struct vmcb_control_area *c, *h;
259 struct nested_state *g;
261 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
263 if (!is_guest_mode(&svm->vcpu))
266 c = &svm->vmcb->control;
267 h = &svm->nested.hsave->control;
270 c->intercept_cr = h->intercept_cr | g->intercept_cr;
271 c->intercept_dr = h->intercept_dr | g->intercept_dr;
272 c->intercept_exceptions = h->intercept_exceptions | g->intercept_exceptions;
273 c->intercept = h->intercept | g->intercept;
276 static inline struct vmcb *get_host_vmcb(struct vcpu_svm *svm)
278 if (is_guest_mode(&svm->vcpu))
279 return svm->nested.hsave;
284 static inline void set_cr_intercept(struct vcpu_svm *svm, int bit)
286 struct vmcb *vmcb = get_host_vmcb(svm);
288 vmcb->control.intercept_cr |= (1U << bit);
290 recalc_intercepts(svm);
293 static inline void clr_cr_intercept(struct vcpu_svm *svm, int bit)
295 struct vmcb *vmcb = get_host_vmcb(svm);
297 vmcb->control.intercept_cr &= ~(1U << bit);
299 recalc_intercepts(svm);
302 static inline bool is_cr_intercept(struct vcpu_svm *svm, int bit)
304 struct vmcb *vmcb = get_host_vmcb(svm);
306 return vmcb->control.intercept_cr & (1U << bit);
309 static inline void set_dr_intercepts(struct vcpu_svm *svm)
311 struct vmcb *vmcb = get_host_vmcb(svm);
313 vmcb->control.intercept_dr = (1 << INTERCEPT_DR0_READ)
314 | (1 << INTERCEPT_DR1_READ)
315 | (1 << INTERCEPT_DR2_READ)
316 | (1 << INTERCEPT_DR3_READ)
317 | (1 << INTERCEPT_DR4_READ)
318 | (1 << INTERCEPT_DR5_READ)
319 | (1 << INTERCEPT_DR6_READ)
320 | (1 << INTERCEPT_DR7_READ)
321 | (1 << INTERCEPT_DR0_WRITE)
322 | (1 << INTERCEPT_DR1_WRITE)
323 | (1 << INTERCEPT_DR2_WRITE)
324 | (1 << INTERCEPT_DR3_WRITE)
325 | (1 << INTERCEPT_DR4_WRITE)
326 | (1 << INTERCEPT_DR5_WRITE)
327 | (1 << INTERCEPT_DR6_WRITE)
328 | (1 << INTERCEPT_DR7_WRITE);
330 recalc_intercepts(svm);
333 static inline void clr_dr_intercepts(struct vcpu_svm *svm)
335 struct vmcb *vmcb = get_host_vmcb(svm);
337 vmcb->control.intercept_dr = 0;
339 recalc_intercepts(svm);
342 static inline void set_exception_intercept(struct vcpu_svm *svm, int bit)
344 struct vmcb *vmcb = get_host_vmcb(svm);
346 vmcb->control.intercept_exceptions |= (1U << bit);
348 recalc_intercepts(svm);
351 static inline void clr_exception_intercept(struct vcpu_svm *svm, int bit)
353 struct vmcb *vmcb = get_host_vmcb(svm);
355 vmcb->control.intercept_exceptions &= ~(1U << bit);
357 recalc_intercepts(svm);
360 static inline void set_intercept(struct vcpu_svm *svm, int bit)
362 struct vmcb *vmcb = get_host_vmcb(svm);
364 vmcb->control.intercept |= (1ULL << bit);
366 recalc_intercepts(svm);
369 static inline void clr_intercept(struct vcpu_svm *svm, int bit)
371 struct vmcb *vmcb = get_host_vmcb(svm);
373 vmcb->control.intercept &= ~(1ULL << bit);
375 recalc_intercepts(svm);
378 static inline void enable_gif(struct vcpu_svm *svm)
380 svm->vcpu.arch.hflags |= HF_GIF_MASK;
383 static inline void disable_gif(struct vcpu_svm *svm)
385 svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
388 static inline bool gif_set(struct vcpu_svm *svm)
390 return !!(svm->vcpu.arch.hflags & HF_GIF_MASK);
393 static unsigned long iopm_base;
395 struct kvm_ldttss_desc {
398 unsigned base1:8, type:5, dpl:2, p:1;
399 unsigned limit1:4, zero0:3, g:1, base2:8;
402 } __attribute__((packed));
404 struct svm_cpu_data {
410 struct kvm_ldttss_desc *tss_desc;
412 struct page *save_area;
415 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
417 struct svm_init_data {
422 static const u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
424 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
425 #define MSRS_RANGE_SIZE 2048
426 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
428 static u32 svm_msrpm_offset(u32 msr)
433 for (i = 0; i < NUM_MSR_MAPS; i++) {
434 if (msr < msrpm_ranges[i] ||
435 msr >= msrpm_ranges[i] + MSRS_IN_RANGE)
438 offset = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */
439 offset += (i * MSRS_RANGE_SIZE); /* add range offset */
441 /* Now we have the u8 offset - but need the u32 offset */
445 /* MSR not in any range */
449 #define MAX_INST_SIZE 15
451 static inline void clgi(void)
453 asm volatile (__ex(SVM_CLGI));
456 static inline void stgi(void)
458 asm volatile (__ex(SVM_STGI));
461 static inline void invlpga(unsigned long addr, u32 asid)
463 asm volatile (__ex(SVM_INVLPGA) : : "a"(addr), "c"(asid));
466 static int get_npt_level(void)
469 return PT64_ROOT_LEVEL;
471 return PT32E_ROOT_LEVEL;
475 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
477 vcpu->arch.efer = efer;
478 if (!npt_enabled && !(efer & EFER_LMA))
481 to_svm(vcpu)->vmcb->save.efer = efer | EFER_SVME;
482 mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
485 static int is_external_interrupt(u32 info)
487 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
488 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
491 static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu)
493 struct vcpu_svm *svm = to_svm(vcpu);
496 if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
497 ret = KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS;
501 static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
503 struct vcpu_svm *svm = to_svm(vcpu);
506 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
508 svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
512 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
514 struct vcpu_svm *svm = to_svm(vcpu);
516 if (svm->vmcb->control.next_rip != 0) {
517 WARN_ON(!static_cpu_has(X86_FEATURE_NRIPS));
518 svm->next_rip = svm->vmcb->control.next_rip;
521 if (!svm->next_rip) {
522 if (emulate_instruction(vcpu, EMULTYPE_SKIP) !=
524 printk(KERN_DEBUG "%s: NOP\n", __func__);
527 if (svm->next_rip - kvm_rip_read(vcpu) > MAX_INST_SIZE)
528 printk(KERN_ERR "%s: ip 0x%lx next 0x%llx\n",
529 __func__, kvm_rip_read(vcpu), svm->next_rip);
531 kvm_rip_write(vcpu, svm->next_rip);
532 svm_set_interrupt_shadow(vcpu, 0);
535 static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
536 bool has_error_code, u32 error_code,
539 struct vcpu_svm *svm = to_svm(vcpu);
542 * If we are within a nested VM we'd better #VMEXIT and let the guest
543 * handle the exception
546 nested_svm_check_exception(svm, nr, has_error_code, error_code))
549 if (nr == BP_VECTOR && !static_cpu_has(X86_FEATURE_NRIPS)) {
550 unsigned long rip, old_rip = kvm_rip_read(&svm->vcpu);
553 * For guest debugging where we have to reinject #BP if some
554 * INT3 is guest-owned:
555 * Emulate nRIP by moving RIP forward. Will fail if injection
556 * raises a fault that is not intercepted. Still better than
557 * failing in all cases.
559 skip_emulated_instruction(&svm->vcpu);
560 rip = kvm_rip_read(&svm->vcpu);
561 svm->int3_rip = rip + svm->vmcb->save.cs.base;
562 svm->int3_injected = rip - old_rip;
565 svm->vmcb->control.event_inj = nr
567 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
568 | SVM_EVTINJ_TYPE_EXEPT;
569 svm->vmcb->control.event_inj_err = error_code;
572 static void svm_init_erratum_383(void)
578 if (!static_cpu_has_bug(X86_BUG_AMD_TLB_MMATCH))
581 /* Use _safe variants to not break nested virtualization */
582 val = native_read_msr_safe(MSR_AMD64_DC_CFG, &err);
588 low = lower_32_bits(val);
589 high = upper_32_bits(val);
591 native_write_msr_safe(MSR_AMD64_DC_CFG, low, high);
593 erratum_383_found = true;
596 static void svm_init_osvw(struct kvm_vcpu *vcpu)
599 * Guests should see errata 400 and 415 as fixed (assuming that
600 * HLT and IO instructions are intercepted).
602 vcpu->arch.osvw.length = (osvw_len >= 3) ? (osvw_len) : 3;
603 vcpu->arch.osvw.status = osvw_status & ~(6ULL);
606 * By increasing VCPU's osvw.length to 3 we are telling the guest that
607 * all osvw.status bits inside that length, including bit 0 (which is
608 * reserved for erratum 298), are valid. However, if host processor's
609 * osvw_len is 0 then osvw_status[0] carries no information. We need to
610 * be conservative here and therefore we tell the guest that erratum 298
611 * is present (because we really don't know).
613 if (osvw_len == 0 && boot_cpu_data.x86 == 0x10)
614 vcpu->arch.osvw.status |= 1;
617 static int has_svm(void)
621 if (!cpu_has_svm(&msg)) {
622 printk(KERN_INFO "has_svm: %s\n", msg);
629 static void svm_hardware_disable(void)
631 /* Make sure we clean up behind us */
632 if (static_cpu_has(X86_FEATURE_TSCRATEMSR))
633 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
637 amd_pmu_disable_virt();
640 static int svm_hardware_enable(void)
643 struct svm_cpu_data *sd;
645 struct desc_ptr gdt_descr;
646 struct desc_struct *gdt;
647 int me = raw_smp_processor_id();
649 rdmsrl(MSR_EFER, efer);
650 if (efer & EFER_SVME)
654 pr_err("%s: err EOPNOTSUPP on %d\n", __func__, me);
657 sd = per_cpu(svm_data, me);
659 pr_err("%s: svm_data is NULL on %d\n", __func__, me);
663 sd->asid_generation = 1;
664 sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
665 sd->next_asid = sd->max_asid + 1;
667 native_store_gdt(&gdt_descr);
668 gdt = (struct desc_struct *)gdt_descr.address;
669 sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
671 wrmsrl(MSR_EFER, efer | EFER_SVME);
673 wrmsrl(MSR_VM_HSAVE_PA, page_to_pfn(sd->save_area) << PAGE_SHIFT);
675 if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
676 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
677 __this_cpu_write(current_tsc_ratio, TSC_RATIO_DEFAULT);
684 * Note that it is possible to have a system with mixed processor
685 * revisions and therefore different OSVW bits. If bits are not the same
686 * on different processors then choose the worst case (i.e. if erratum
687 * is present on one processor and not on another then assume that the
688 * erratum is present everywhere).
690 if (cpu_has(&boot_cpu_data, X86_FEATURE_OSVW)) {
691 uint64_t len, status = 0;
694 len = native_read_msr_safe(MSR_AMD64_OSVW_ID_LENGTH, &err);
696 status = native_read_msr_safe(MSR_AMD64_OSVW_STATUS,
700 osvw_status = osvw_len = 0;
704 osvw_status |= status;
705 osvw_status &= (1ULL << osvw_len) - 1;
708 osvw_status = osvw_len = 0;
710 svm_init_erratum_383();
712 amd_pmu_enable_virt();
717 static void svm_cpu_uninit(int cpu)
719 struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id());
724 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
725 __free_page(sd->save_area);
729 static int svm_cpu_init(int cpu)
731 struct svm_cpu_data *sd;
734 sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
738 sd->save_area = alloc_page(GFP_KERNEL);
743 per_cpu(svm_data, cpu) = sd;
753 static bool valid_msr_intercept(u32 index)
757 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++)
758 if (direct_access_msrs[i].index == index)
764 static void set_msr_interception(u32 *msrpm, unsigned msr,
767 u8 bit_read, bit_write;
772 * If this warning triggers extend the direct_access_msrs list at the
773 * beginning of the file
775 WARN_ON(!valid_msr_intercept(msr));
777 offset = svm_msrpm_offset(msr);
778 bit_read = 2 * (msr & 0x0f);
779 bit_write = 2 * (msr & 0x0f) + 1;
782 BUG_ON(offset == MSR_INVALID);
784 read ? clear_bit(bit_read, &tmp) : set_bit(bit_read, &tmp);
785 write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp);
790 static void svm_vcpu_init_msrpm(u32 *msrpm)
794 memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
796 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
797 if (!direct_access_msrs[i].always)
800 set_msr_interception(msrpm, direct_access_msrs[i].index, 1, 1);
804 static void add_msr_offset(u32 offset)
808 for (i = 0; i < MSRPM_OFFSETS; ++i) {
810 /* Offset already in list? */
811 if (msrpm_offsets[i] == offset)
814 /* Slot used by another offset? */
815 if (msrpm_offsets[i] != MSR_INVALID)
818 /* Add offset to list */
819 msrpm_offsets[i] = offset;
825 * If this BUG triggers the msrpm_offsets table has an overflow. Just
826 * increase MSRPM_OFFSETS in this case.
831 static void init_msrpm_offsets(void)
835 memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets));
837 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
840 offset = svm_msrpm_offset(direct_access_msrs[i].index);
841 BUG_ON(offset == MSR_INVALID);
843 add_msr_offset(offset);
847 static void svm_enable_lbrv(struct vcpu_svm *svm)
849 u32 *msrpm = svm->msrpm;
851 svm->vmcb->control.lbr_ctl = 1;
852 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
853 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
854 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
855 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
858 static void svm_disable_lbrv(struct vcpu_svm *svm)
860 u32 *msrpm = svm->msrpm;
862 svm->vmcb->control.lbr_ctl = 0;
863 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
864 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
865 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
866 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
869 #define MTRR_TYPE_UC_MINUS 7
870 #define MTRR2PROTVAL_INVALID 0xff
872 static u8 mtrr2protval[8];
874 static u8 fallback_mtrr_type(int mtrr)
877 * WT and WP aren't always available in the host PAT. Treat
878 * them as UC and UC- respectively. Everything else should be
883 case MTRR_TYPE_WRTHROUGH:
884 return MTRR_TYPE_UNCACHABLE;
885 case MTRR_TYPE_WRPROT:
886 return MTRR_TYPE_UC_MINUS;
892 static void build_mtrr2protval(void)
897 for (i = 0; i < 8; i++)
898 mtrr2protval[i] = MTRR2PROTVAL_INVALID;
900 /* Ignore the invalid MTRR types. */
905 * Use host PAT value to figure out the mapping from guest MTRR
906 * values to nested page table PAT/PCD/PWT values. We do not
907 * want to change the host PAT value every time we enter the
910 rdmsrl(MSR_IA32_CR_PAT, pat);
911 for (i = 0; i < 8; i++) {
912 u8 mtrr = pat >> (8 * i);
914 if (mtrr2protval[mtrr] == MTRR2PROTVAL_INVALID)
915 mtrr2protval[mtrr] = __cm_idx2pte(i);
918 for (i = 0; i < 8; i++) {
919 if (mtrr2protval[i] == MTRR2PROTVAL_INVALID) {
920 u8 fallback = fallback_mtrr_type(i);
921 mtrr2protval[i] = mtrr2protval[fallback];
922 BUG_ON(mtrr2protval[i] == MTRR2PROTVAL_INVALID);
927 static __init int svm_hardware_setup(void)
930 struct page *iopm_pages;
934 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
939 iopm_va = page_address(iopm_pages);
940 memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
941 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
943 init_msrpm_offsets();
945 if (boot_cpu_has(X86_FEATURE_NX))
946 kvm_enable_efer_bits(EFER_NX);
948 if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
949 kvm_enable_efer_bits(EFER_FFXSR);
951 if (boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
954 kvm_has_tsc_control = true;
957 * Make sure the user can only configure tsc_khz values that
958 * fit into a signed integer.
959 * A min value is not calculated needed because it will always
960 * be 1 on all machines and a value of 0 is used to disable
961 * tsc-scaling for the vcpu.
963 max = min(0x7fffffffULL, __scale_tsc(tsc_khz, TSC_RATIO_MAX));
965 kvm_max_guest_tsc_khz = max;
969 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
970 kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE);
973 for_each_possible_cpu(cpu) {
974 r = svm_cpu_init(cpu);
979 if (!boot_cpu_has(X86_FEATURE_NPT))
982 if (npt_enabled && !npt) {
983 printk(KERN_INFO "kvm: Nested Paging disabled\n");
988 printk(KERN_INFO "kvm: Nested Paging enabled\n");
993 build_mtrr2protval();
997 __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
1002 static __exit void svm_hardware_unsetup(void)
1006 for_each_possible_cpu(cpu)
1007 svm_cpu_uninit(cpu);
1009 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
1013 static void init_seg(struct vmcb_seg *seg)
1016 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
1017 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
1018 seg->limit = 0xffff;
1022 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
1025 seg->attrib = SVM_SELECTOR_P_MASK | type;
1026 seg->limit = 0xffff;
1030 static u64 __scale_tsc(u64 ratio, u64 tsc)
1032 u64 mult, frac, _tsc;
1035 frac = ratio & ((1ULL << 32) - 1);
1039 _tsc += (tsc >> 32) * frac;
1040 _tsc += ((tsc & ((1ULL << 32) - 1)) * frac) >> 32;
1045 static u64 svm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1047 struct vcpu_svm *svm = to_svm(vcpu);
1050 if (svm->tsc_ratio != TSC_RATIO_DEFAULT)
1051 _tsc = __scale_tsc(svm->tsc_ratio, tsc);
1056 static void svm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1058 struct vcpu_svm *svm = to_svm(vcpu);
1062 /* Guest TSC same frequency as host TSC? */
1064 svm->tsc_ratio = TSC_RATIO_DEFAULT;
1068 /* TSC scaling supported? */
1069 if (!boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
1070 if (user_tsc_khz > tsc_khz) {
1071 vcpu->arch.tsc_catchup = 1;
1072 vcpu->arch.tsc_always_catchup = 1;
1074 WARN(1, "user requested TSC rate below hardware speed\n");
1080 /* TSC scaling required - calculate ratio */
1082 do_div(ratio, tsc_khz);
1084 if (ratio == 0 || ratio & TSC_RATIO_RSVD) {
1085 WARN_ONCE(1, "Invalid TSC ratio - virtual-tsc-khz=%u\n",
1089 svm->tsc_ratio = ratio;
1092 static u64 svm_read_tsc_offset(struct kvm_vcpu *vcpu)
1094 struct vcpu_svm *svm = to_svm(vcpu);
1096 return svm->vmcb->control.tsc_offset;
1099 static void svm_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1101 struct vcpu_svm *svm = to_svm(vcpu);
1102 u64 g_tsc_offset = 0;
1104 if (is_guest_mode(vcpu)) {
1105 g_tsc_offset = svm->vmcb->control.tsc_offset -
1106 svm->nested.hsave->control.tsc_offset;
1107 svm->nested.hsave->control.tsc_offset = offset;
1109 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1110 svm->vmcb->control.tsc_offset,
1113 svm->vmcb->control.tsc_offset = offset + g_tsc_offset;
1115 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1118 static void svm_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment, bool host)
1120 struct vcpu_svm *svm = to_svm(vcpu);
1123 if (svm->tsc_ratio != TSC_RATIO_DEFAULT)
1124 WARN_ON(adjustment < 0);
1125 adjustment = svm_scale_tsc(vcpu, (u64)adjustment);
1128 svm->vmcb->control.tsc_offset += adjustment;
1129 if (is_guest_mode(vcpu))
1130 svm->nested.hsave->control.tsc_offset += adjustment;
1132 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1133 svm->vmcb->control.tsc_offset - adjustment,
1134 svm->vmcb->control.tsc_offset);
1136 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1139 static u64 svm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1143 tsc = svm_scale_tsc(vcpu, rdtsc());
1145 return target_tsc - tsc;
1148 static void svm_set_guest_pat(struct vcpu_svm *svm, u64 *g_pat)
1150 struct kvm_vcpu *vcpu = &svm->vcpu;
1152 /* Unlike Intel, AMD takes the guest's CR0.CD into account.
1154 * AMD does not have IPAT. To emulate it for the case of guests
1155 * with no assigned devices, just set everything to WB. If guests
1156 * have assigned devices, however, we cannot force WB for RAM
1157 * pages only, so use the guest PAT directly.
1159 if (!kvm_arch_has_assigned_device(vcpu->kvm))
1160 *g_pat = 0x0606060606060606;
1162 *g_pat = vcpu->arch.pat;
1165 static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
1170 * 1. MMIO: trust guest MTRR, so same as item 3.
1171 * 2. No passthrough: always map as WB, and force guest PAT to WB as well
1172 * 3. Passthrough: can't guarantee the result, try to trust guest.
1174 if (!is_mmio && !kvm_arch_has_assigned_device(vcpu->kvm))
1177 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED) &&
1178 kvm_read_cr0(vcpu) & X86_CR0_CD)
1179 return _PAGE_NOCACHE;
1181 mtrr = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
1182 return mtrr2protval[mtrr];
1185 static void init_vmcb(struct vcpu_svm *svm, bool init_event)
1187 struct vmcb_control_area *control = &svm->vmcb->control;
1188 struct vmcb_save_area *save = &svm->vmcb->save;
1190 svm->vcpu.fpu_active = 1;
1191 svm->vcpu.arch.hflags = 0;
1193 set_cr_intercept(svm, INTERCEPT_CR0_READ);
1194 set_cr_intercept(svm, INTERCEPT_CR3_READ);
1195 set_cr_intercept(svm, INTERCEPT_CR4_READ);
1196 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1197 set_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1198 set_cr_intercept(svm, INTERCEPT_CR4_WRITE);
1199 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
1201 set_dr_intercepts(svm);
1203 set_exception_intercept(svm, PF_VECTOR);
1204 set_exception_intercept(svm, UD_VECTOR);
1205 set_exception_intercept(svm, MC_VECTOR);
1207 set_intercept(svm, INTERCEPT_INTR);
1208 set_intercept(svm, INTERCEPT_NMI);
1209 set_intercept(svm, INTERCEPT_SMI);
1210 set_intercept(svm, INTERCEPT_SELECTIVE_CR0);
1211 set_intercept(svm, INTERCEPT_RDPMC);
1212 set_intercept(svm, INTERCEPT_CPUID);
1213 set_intercept(svm, INTERCEPT_INVD);
1214 set_intercept(svm, INTERCEPT_HLT);
1215 set_intercept(svm, INTERCEPT_INVLPG);
1216 set_intercept(svm, INTERCEPT_INVLPGA);
1217 set_intercept(svm, INTERCEPT_IOIO_PROT);
1218 set_intercept(svm, INTERCEPT_MSR_PROT);
1219 set_intercept(svm, INTERCEPT_TASK_SWITCH);
1220 set_intercept(svm, INTERCEPT_SHUTDOWN);
1221 set_intercept(svm, INTERCEPT_VMRUN);
1222 set_intercept(svm, INTERCEPT_VMMCALL);
1223 set_intercept(svm, INTERCEPT_VMLOAD);
1224 set_intercept(svm, INTERCEPT_VMSAVE);
1225 set_intercept(svm, INTERCEPT_STGI);
1226 set_intercept(svm, INTERCEPT_CLGI);
1227 set_intercept(svm, INTERCEPT_SKINIT);
1228 set_intercept(svm, INTERCEPT_WBINVD);
1229 set_intercept(svm, INTERCEPT_MONITOR);
1230 set_intercept(svm, INTERCEPT_MWAIT);
1231 set_intercept(svm, INTERCEPT_XSETBV);
1233 control->iopm_base_pa = iopm_base;
1234 control->msrpm_base_pa = __pa(svm->msrpm);
1235 control->int_ctl = V_INTR_MASKING_MASK;
1237 init_seg(&save->es);
1238 init_seg(&save->ss);
1239 init_seg(&save->ds);
1240 init_seg(&save->fs);
1241 init_seg(&save->gs);
1243 save->cs.selector = 0xf000;
1244 save->cs.base = 0xffff0000;
1245 /* Executable/Readable Code Segment */
1246 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
1247 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
1248 save->cs.limit = 0xffff;
1250 save->gdtr.limit = 0xffff;
1251 save->idtr.limit = 0xffff;
1253 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
1254 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
1257 svm_set_efer(&svm->vcpu, 0);
1258 save->dr6 = 0xffff0ff0;
1259 kvm_set_rflags(&svm->vcpu, 2);
1260 save->rip = 0x0000fff0;
1261 svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
1264 * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
1265 * It also updates the guest-visible cr0 value.
1267 svm_set_cr0(&svm->vcpu, X86_CR0_NW | X86_CR0_CD | X86_CR0_ET);
1268 kvm_mmu_reset_context(&svm->vcpu);
1270 save->cr4 = X86_CR4_PAE;
1274 /* Setup VMCB for Nested Paging */
1275 control->nested_ctl = 1;
1276 clr_intercept(svm, INTERCEPT_INVLPG);
1277 clr_exception_intercept(svm, PF_VECTOR);
1278 clr_cr_intercept(svm, INTERCEPT_CR3_READ);
1279 clr_cr_intercept(svm, INTERCEPT_CR3_WRITE);
1280 save->g_pat = svm->vcpu.arch.pat;
1281 svm_set_guest_pat(svm, &save->g_pat);
1285 svm->asid_generation = 0;
1287 svm->nested.vmcb = 0;
1288 svm->vcpu.arch.hflags = 0;
1290 if (boot_cpu_has(X86_FEATURE_PAUSEFILTER)) {
1291 control->pause_filter_count = 3000;
1292 set_intercept(svm, INTERCEPT_PAUSE);
1295 mark_all_dirty(svm->vmcb);
1300 static void svm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
1302 struct vcpu_svm *svm = to_svm(vcpu);
1307 svm->vcpu.arch.apic_base = APIC_DEFAULT_PHYS_BASE |
1308 MSR_IA32_APICBASE_ENABLE;
1309 if (kvm_vcpu_is_reset_bsp(&svm->vcpu))
1310 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
1312 init_vmcb(svm, init_event);
1314 kvm_cpuid(vcpu, &eax, &dummy, &dummy, &dummy);
1315 kvm_register_write(vcpu, VCPU_REGS_RDX, eax);
1318 static struct kvm_vcpu *svm_create_vcpu(struct kvm *kvm, unsigned int id)
1320 struct vcpu_svm *svm;
1322 struct page *msrpm_pages;
1323 struct page *hsave_page;
1324 struct page *nested_msrpm_pages;
1327 svm = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
1333 svm->tsc_ratio = TSC_RATIO_DEFAULT;
1335 err = kvm_vcpu_init(&svm->vcpu, kvm, id);
1340 page = alloc_page(GFP_KERNEL);
1344 msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
1348 nested_msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
1349 if (!nested_msrpm_pages)
1352 hsave_page = alloc_page(GFP_KERNEL);
1356 svm->nested.hsave = page_address(hsave_page);
1358 svm->msrpm = page_address(msrpm_pages);
1359 svm_vcpu_init_msrpm(svm->msrpm);
1361 svm->nested.msrpm = page_address(nested_msrpm_pages);
1362 svm_vcpu_init_msrpm(svm->nested.msrpm);
1364 svm->vmcb = page_address(page);
1365 clear_page(svm->vmcb);
1366 svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
1367 svm->asid_generation = 0;
1368 init_vmcb(svm, false);
1370 svm_init_osvw(&svm->vcpu);
1375 __free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER);
1377 __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
1381 kvm_vcpu_uninit(&svm->vcpu);
1383 kmem_cache_free(kvm_vcpu_cache, svm);
1385 return ERR_PTR(err);
1388 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
1390 struct vcpu_svm *svm = to_svm(vcpu);
1392 __free_page(pfn_to_page(svm->vmcb_pa >> PAGE_SHIFT));
1393 __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
1394 __free_page(virt_to_page(svm->nested.hsave));
1395 __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER);
1396 kvm_vcpu_uninit(vcpu);
1397 kmem_cache_free(kvm_vcpu_cache, svm);
1400 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1402 struct vcpu_svm *svm = to_svm(vcpu);
1405 if (unlikely(cpu != vcpu->cpu)) {
1406 svm->asid_generation = 0;
1407 mark_all_dirty(svm->vmcb);
1410 #ifdef CONFIG_X86_64
1411 rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host.gs_base);
1413 savesegment(fs, svm->host.fs);
1414 savesegment(gs, svm->host.gs);
1415 svm->host.ldt = kvm_read_ldt();
1417 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1418 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1420 if (static_cpu_has(X86_FEATURE_TSCRATEMSR) &&
1421 svm->tsc_ratio != __this_cpu_read(current_tsc_ratio)) {
1422 __this_cpu_write(current_tsc_ratio, svm->tsc_ratio);
1423 wrmsrl(MSR_AMD64_TSC_RATIO, svm->tsc_ratio);
1427 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
1429 struct vcpu_svm *svm = to_svm(vcpu);
1432 ++vcpu->stat.host_state_reload;
1433 kvm_load_ldt(svm->host.ldt);
1434 #ifdef CONFIG_X86_64
1435 loadsegment(fs, svm->host.fs);
1436 wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gs);
1437 load_gs_index(svm->host.gs);
1439 #ifdef CONFIG_X86_32_LAZY_GS
1440 loadsegment(gs, svm->host.gs);
1443 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1444 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1447 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
1449 return to_svm(vcpu)->vmcb->save.rflags;
1452 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1455 * Any change of EFLAGS.VM is accompained by a reload of SS
1456 * (caused by either a task switch or an inter-privilege IRET),
1457 * so we do not need to update the CPL here.
1459 to_svm(vcpu)->vmcb->save.rflags = rflags;
1462 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1465 case VCPU_EXREG_PDPTR:
1466 BUG_ON(!npt_enabled);
1467 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
1474 static void svm_set_vintr(struct vcpu_svm *svm)
1476 set_intercept(svm, INTERCEPT_VINTR);
1479 static void svm_clear_vintr(struct vcpu_svm *svm)
1481 clr_intercept(svm, INTERCEPT_VINTR);
1484 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
1486 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1489 case VCPU_SREG_CS: return &save->cs;
1490 case VCPU_SREG_DS: return &save->ds;
1491 case VCPU_SREG_ES: return &save->es;
1492 case VCPU_SREG_FS: return &save->fs;
1493 case VCPU_SREG_GS: return &save->gs;
1494 case VCPU_SREG_SS: return &save->ss;
1495 case VCPU_SREG_TR: return &save->tr;
1496 case VCPU_SREG_LDTR: return &save->ldtr;
1502 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1504 struct vmcb_seg *s = svm_seg(vcpu, seg);
1509 static void svm_get_segment(struct kvm_vcpu *vcpu,
1510 struct kvm_segment *var, int seg)
1512 struct vmcb_seg *s = svm_seg(vcpu, seg);
1514 var->base = s->base;
1515 var->limit = s->limit;
1516 var->selector = s->selector;
1517 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
1518 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
1519 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
1520 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
1521 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
1522 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
1523 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
1526 * AMD CPUs circa 2014 track the G bit for all segments except CS.
1527 * However, the SVM spec states that the G bit is not observed by the
1528 * CPU, and some VMware virtual CPUs drop the G bit for all segments.
1529 * So let's synthesize a legal G bit for all segments, this helps
1530 * running KVM nested. It also helps cross-vendor migration, because
1531 * Intel's vmentry has a check on the 'G' bit.
1533 var->g = s->limit > 0xfffff;
1536 * AMD's VMCB does not have an explicit unusable field, so emulate it
1537 * for cross vendor migration purposes by "not present"
1539 var->unusable = !var->present || (var->type == 0);
1544 * Work around a bug where the busy flag in the tr selector
1554 * The accessed bit must always be set in the segment
1555 * descriptor cache, although it can be cleared in the
1556 * descriptor, the cached bit always remains at 1. Since
1557 * Intel has a check on this, set it here to support
1558 * cross-vendor migration.
1565 * On AMD CPUs sometimes the DB bit in the segment
1566 * descriptor is left as 1, although the whole segment has
1567 * been made unusable. Clear it here to pass an Intel VMX
1568 * entry check when cross vendor migrating.
1572 var->dpl = to_svm(vcpu)->vmcb->save.cpl;
1577 static int svm_get_cpl(struct kvm_vcpu *vcpu)
1579 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1584 static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1586 struct vcpu_svm *svm = to_svm(vcpu);
1588 dt->size = svm->vmcb->save.idtr.limit;
1589 dt->address = svm->vmcb->save.idtr.base;
1592 static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1594 struct vcpu_svm *svm = to_svm(vcpu);
1596 svm->vmcb->save.idtr.limit = dt->size;
1597 svm->vmcb->save.idtr.base = dt->address ;
1598 mark_dirty(svm->vmcb, VMCB_DT);
1601 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1603 struct vcpu_svm *svm = to_svm(vcpu);
1605 dt->size = svm->vmcb->save.gdtr.limit;
1606 dt->address = svm->vmcb->save.gdtr.base;
1609 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1611 struct vcpu_svm *svm = to_svm(vcpu);
1613 svm->vmcb->save.gdtr.limit = dt->size;
1614 svm->vmcb->save.gdtr.base = dt->address ;
1615 mark_dirty(svm->vmcb, VMCB_DT);
1618 static void svm_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
1622 static void svm_decache_cr3(struct kvm_vcpu *vcpu)
1626 static void svm_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1630 static void update_cr0_intercept(struct vcpu_svm *svm)
1632 ulong gcr0 = svm->vcpu.arch.cr0;
1633 u64 *hcr0 = &svm->vmcb->save.cr0;
1635 if (!svm->vcpu.fpu_active)
1636 *hcr0 |= SVM_CR0_SELECTIVE_MASK;
1638 *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
1639 | (gcr0 & SVM_CR0_SELECTIVE_MASK);
1641 mark_dirty(svm->vmcb, VMCB_CR);
1643 if (gcr0 == *hcr0 && svm->vcpu.fpu_active) {
1644 clr_cr_intercept(svm, INTERCEPT_CR0_READ);
1645 clr_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1647 set_cr_intercept(svm, INTERCEPT_CR0_READ);
1648 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
1652 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1654 struct vcpu_svm *svm = to_svm(vcpu);
1656 #ifdef CONFIG_X86_64
1657 if (vcpu->arch.efer & EFER_LME) {
1658 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
1659 vcpu->arch.efer |= EFER_LMA;
1660 svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
1663 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
1664 vcpu->arch.efer &= ~EFER_LMA;
1665 svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
1669 vcpu->arch.cr0 = cr0;
1672 cr0 |= X86_CR0_PG | X86_CR0_WP;
1674 if (!vcpu->fpu_active)
1677 /* These are emulated via page tables. */
1678 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
1680 svm->vmcb->save.cr0 = cr0;
1681 mark_dirty(svm->vmcb, VMCB_CR);
1682 update_cr0_intercept(svm);
1685 static int svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1687 unsigned long host_cr4_mce = cr4_read_shadow() & X86_CR4_MCE;
1688 unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
1690 if (cr4 & X86_CR4_VMXE)
1693 if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
1694 svm_flush_tlb(vcpu);
1696 vcpu->arch.cr4 = cr4;
1699 cr4 |= host_cr4_mce;
1700 to_svm(vcpu)->vmcb->save.cr4 = cr4;
1701 mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
1705 static void svm_set_segment(struct kvm_vcpu *vcpu,
1706 struct kvm_segment *var, int seg)
1708 struct vcpu_svm *svm = to_svm(vcpu);
1709 struct vmcb_seg *s = svm_seg(vcpu, seg);
1711 s->base = var->base;
1712 s->limit = var->limit;
1713 s->selector = var->selector;
1717 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
1718 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
1719 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
1720 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
1721 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
1722 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
1723 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
1724 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
1728 * This is always accurate, except if SYSRET returned to a segment
1729 * with SS.DPL != 3. Intel does not have this quirk, and always
1730 * forces SS.DPL to 3 on sysret, so we ignore that case; fixing it
1731 * would entail passing the CPL to userspace and back.
1733 if (seg == VCPU_SREG_SS)
1734 svm->vmcb->save.cpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
1736 mark_dirty(svm->vmcb, VMCB_SEG);
1739 static void update_db_bp_intercept(struct kvm_vcpu *vcpu)
1741 struct vcpu_svm *svm = to_svm(vcpu);
1743 clr_exception_intercept(svm, DB_VECTOR);
1744 clr_exception_intercept(svm, BP_VECTOR);
1746 if (svm->nmi_singlestep)
1747 set_exception_intercept(svm, DB_VECTOR);
1749 if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
1750 if (vcpu->guest_debug &
1751 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
1752 set_exception_intercept(svm, DB_VECTOR);
1753 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
1754 set_exception_intercept(svm, BP_VECTOR);
1756 vcpu->guest_debug = 0;
1759 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
1761 if (sd->next_asid > sd->max_asid) {
1762 ++sd->asid_generation;
1764 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
1767 svm->asid_generation = sd->asid_generation;
1768 svm->vmcb->control.asid = sd->next_asid++;
1770 mark_dirty(svm->vmcb, VMCB_ASID);
1773 static u64 svm_get_dr6(struct kvm_vcpu *vcpu)
1775 return to_svm(vcpu)->vmcb->save.dr6;
1778 static void svm_set_dr6(struct kvm_vcpu *vcpu, unsigned long value)
1780 struct vcpu_svm *svm = to_svm(vcpu);
1782 svm->vmcb->save.dr6 = value;
1783 mark_dirty(svm->vmcb, VMCB_DR);
1786 static void svm_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
1788 struct vcpu_svm *svm = to_svm(vcpu);
1790 get_debugreg(vcpu->arch.db[0], 0);
1791 get_debugreg(vcpu->arch.db[1], 1);
1792 get_debugreg(vcpu->arch.db[2], 2);
1793 get_debugreg(vcpu->arch.db[3], 3);
1794 vcpu->arch.dr6 = svm_get_dr6(vcpu);
1795 vcpu->arch.dr7 = svm->vmcb->save.dr7;
1797 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
1798 set_dr_intercepts(svm);
1801 static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value)
1803 struct vcpu_svm *svm = to_svm(vcpu);
1805 svm->vmcb->save.dr7 = value;
1806 mark_dirty(svm->vmcb, VMCB_DR);
1809 static int pf_interception(struct vcpu_svm *svm)
1811 u64 fault_address = svm->vmcb->control.exit_info_2;
1815 switch (svm->apf_reason) {
1817 error_code = svm->vmcb->control.exit_info_1;
1819 trace_kvm_page_fault(fault_address, error_code);
1820 if (!npt_enabled && kvm_event_needs_reinjection(&svm->vcpu))
1821 kvm_mmu_unprotect_page_virt(&svm->vcpu, fault_address);
1822 r = kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code,
1823 svm->vmcb->control.insn_bytes,
1824 svm->vmcb->control.insn_len);
1826 case KVM_PV_REASON_PAGE_NOT_PRESENT:
1827 svm->apf_reason = 0;
1828 local_irq_disable();
1829 kvm_async_pf_task_wait(fault_address);
1832 case KVM_PV_REASON_PAGE_READY:
1833 svm->apf_reason = 0;
1834 local_irq_disable();
1835 kvm_async_pf_task_wake(fault_address);
1842 static int db_interception(struct vcpu_svm *svm)
1844 struct kvm_run *kvm_run = svm->vcpu.run;
1846 if (!(svm->vcpu.guest_debug &
1847 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
1848 !svm->nmi_singlestep) {
1849 kvm_queue_exception(&svm->vcpu, DB_VECTOR);
1853 if (svm->nmi_singlestep) {
1854 svm->nmi_singlestep = false;
1855 if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP))
1856 svm->vmcb->save.rflags &=
1857 ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
1858 update_db_bp_intercept(&svm->vcpu);
1861 if (svm->vcpu.guest_debug &
1862 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
1863 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1864 kvm_run->debug.arch.pc =
1865 svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1866 kvm_run->debug.arch.exception = DB_VECTOR;
1873 static int bp_interception(struct vcpu_svm *svm)
1875 struct kvm_run *kvm_run = svm->vcpu.run;
1877 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1878 kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1879 kvm_run->debug.arch.exception = BP_VECTOR;
1883 static int ud_interception(struct vcpu_svm *svm)
1887 er = emulate_instruction(&svm->vcpu, EMULTYPE_TRAP_UD);
1888 if (er != EMULATE_DONE)
1889 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
1893 static void svm_fpu_activate(struct kvm_vcpu *vcpu)
1895 struct vcpu_svm *svm = to_svm(vcpu);
1897 clr_exception_intercept(svm, NM_VECTOR);
1899 svm->vcpu.fpu_active = 1;
1900 update_cr0_intercept(svm);
1903 static int nm_interception(struct vcpu_svm *svm)
1905 svm_fpu_activate(&svm->vcpu);
1909 static bool is_erratum_383(void)
1914 if (!erratum_383_found)
1917 value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err);
1921 /* Bit 62 may or may not be set for this mce */
1922 value &= ~(1ULL << 62);
1924 if (value != 0xb600000000010015ULL)
1927 /* Clear MCi_STATUS registers */
1928 for (i = 0; i < 6; ++i)
1929 native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0);
1931 value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err);
1935 value &= ~(1ULL << 2);
1936 low = lower_32_bits(value);
1937 high = upper_32_bits(value);
1939 native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high);
1942 /* Flush tlb to evict multi-match entries */
1948 static void svm_handle_mce(struct vcpu_svm *svm)
1950 if (is_erratum_383()) {
1952 * Erratum 383 triggered. Guest state is corrupt so kill the
1955 pr_err("KVM: Guest triggered AMD Erratum 383\n");
1957 kvm_make_request(KVM_REQ_TRIPLE_FAULT, &svm->vcpu);
1963 * On an #MC intercept the MCE handler is not called automatically in
1964 * the host. So do it by hand here.
1968 /* not sure if we ever come back to this point */
1973 static int mc_interception(struct vcpu_svm *svm)
1978 static int shutdown_interception(struct vcpu_svm *svm)
1980 struct kvm_run *kvm_run = svm->vcpu.run;
1983 * VMCB is undefined after a SHUTDOWN intercept
1984 * so reinitialize it.
1986 clear_page(svm->vmcb);
1987 init_vmcb(svm, false);
1989 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1993 static int io_interception(struct vcpu_svm *svm)
1995 struct kvm_vcpu *vcpu = &svm->vcpu;
1996 u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
1997 int size, in, string;
2000 ++svm->vcpu.stat.io_exits;
2001 string = (io_info & SVM_IOIO_STR_MASK) != 0;
2002 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
2004 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
2006 port = io_info >> 16;
2007 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
2008 svm->next_rip = svm->vmcb->control.exit_info_2;
2009 skip_emulated_instruction(&svm->vcpu);
2011 return kvm_fast_pio_out(vcpu, size, port);
2014 static int nmi_interception(struct vcpu_svm *svm)
2019 static int intr_interception(struct vcpu_svm *svm)
2021 ++svm->vcpu.stat.irq_exits;
2025 static int nop_on_interception(struct vcpu_svm *svm)
2030 static int halt_interception(struct vcpu_svm *svm)
2032 svm->next_rip = kvm_rip_read(&svm->vcpu) + 1;
2033 return kvm_emulate_halt(&svm->vcpu);
2036 static int vmmcall_interception(struct vcpu_svm *svm)
2038 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2039 kvm_emulate_hypercall(&svm->vcpu);
2043 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
2045 struct vcpu_svm *svm = to_svm(vcpu);
2047 return svm->nested.nested_cr3;
2050 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
2052 struct vcpu_svm *svm = to_svm(vcpu);
2053 u64 cr3 = svm->nested.nested_cr3;
2057 ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(cr3), &pdpte,
2058 offset_in_page(cr3) + index * 8, 8);
2064 static void nested_svm_set_tdp_cr3(struct kvm_vcpu *vcpu,
2067 struct vcpu_svm *svm = to_svm(vcpu);
2069 svm->vmcb->control.nested_cr3 = root;
2070 mark_dirty(svm->vmcb, VMCB_NPT);
2071 svm_flush_tlb(vcpu);
2074 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
2075 struct x86_exception *fault)
2077 struct vcpu_svm *svm = to_svm(vcpu);
2079 if (svm->vmcb->control.exit_code != SVM_EXIT_NPF) {
2081 * TODO: track the cause of the nested page fault, and
2082 * correctly fill in the high bits of exit_info_1.
2084 svm->vmcb->control.exit_code = SVM_EXIT_NPF;
2085 svm->vmcb->control.exit_code_hi = 0;
2086 svm->vmcb->control.exit_info_1 = (1ULL << 32);
2087 svm->vmcb->control.exit_info_2 = fault->address;
2090 svm->vmcb->control.exit_info_1 &= ~0xffffffffULL;
2091 svm->vmcb->control.exit_info_1 |= fault->error_code;
2094 * The present bit is always zero for page structure faults on real
2097 if (svm->vmcb->control.exit_info_1 & (2ULL << 32))
2098 svm->vmcb->control.exit_info_1 &= ~1;
2100 nested_svm_vmexit(svm);
2103 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
2105 WARN_ON(mmu_is_nested(vcpu));
2106 kvm_init_shadow_mmu(vcpu);
2107 vcpu->arch.mmu.set_cr3 = nested_svm_set_tdp_cr3;
2108 vcpu->arch.mmu.get_cr3 = nested_svm_get_tdp_cr3;
2109 vcpu->arch.mmu.get_pdptr = nested_svm_get_tdp_pdptr;
2110 vcpu->arch.mmu.inject_page_fault = nested_svm_inject_npf_exit;
2111 vcpu->arch.mmu.shadow_root_level = get_npt_level();
2112 reset_shadow_zero_bits_mask(vcpu, &vcpu->arch.mmu);
2113 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
2116 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
2118 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
2121 static int nested_svm_check_permissions(struct vcpu_svm *svm)
2123 if (!(svm->vcpu.arch.efer & EFER_SVME)
2124 || !is_paging(&svm->vcpu)) {
2125 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2129 if (svm->vmcb->save.cpl) {
2130 kvm_inject_gp(&svm->vcpu, 0);
2137 static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
2138 bool has_error_code, u32 error_code)
2142 if (!is_guest_mode(&svm->vcpu))
2145 svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
2146 svm->vmcb->control.exit_code_hi = 0;
2147 svm->vmcb->control.exit_info_1 = error_code;
2148 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
2150 vmexit = nested_svm_intercept(svm);
2151 if (vmexit == NESTED_EXIT_DONE)
2152 svm->nested.exit_required = true;
2157 /* This function returns true if it is save to enable the irq window */
2158 static inline bool nested_svm_intr(struct vcpu_svm *svm)
2160 if (!is_guest_mode(&svm->vcpu))
2163 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
2166 if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
2170 * if vmexit was already requested (by intercepted exception
2171 * for instance) do not overwrite it with "external interrupt"
2174 if (svm->nested.exit_required)
2177 svm->vmcb->control.exit_code = SVM_EXIT_INTR;
2178 svm->vmcb->control.exit_info_1 = 0;
2179 svm->vmcb->control.exit_info_2 = 0;
2181 if (svm->nested.intercept & 1ULL) {
2183 * The #vmexit can't be emulated here directly because this
2184 * code path runs with irqs and preemption disabled. A
2185 * #vmexit emulation might sleep. Only signal request for
2188 svm->nested.exit_required = true;
2189 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
2196 /* This function returns true if it is save to enable the nmi window */
2197 static inline bool nested_svm_nmi(struct vcpu_svm *svm)
2199 if (!is_guest_mode(&svm->vcpu))
2202 if (!(svm->nested.intercept & (1ULL << INTERCEPT_NMI)))
2205 svm->vmcb->control.exit_code = SVM_EXIT_NMI;
2206 svm->nested.exit_required = true;
2211 static void *nested_svm_map(struct vcpu_svm *svm, u64 gpa, struct page **_page)
2217 page = kvm_vcpu_gfn_to_page(&svm->vcpu, gpa >> PAGE_SHIFT);
2218 if (is_error_page(page))
2226 kvm_inject_gp(&svm->vcpu, 0);
2231 static void nested_svm_unmap(struct page *page)
2234 kvm_release_page_dirty(page);
2237 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
2239 unsigned port, size, iopm_len;
2244 if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT)))
2245 return NESTED_EXIT_HOST;
2247 port = svm->vmcb->control.exit_info_1 >> 16;
2248 size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
2249 SVM_IOIO_SIZE_SHIFT;
2250 gpa = svm->nested.vmcb_iopm + (port / 8);
2251 start_bit = port % 8;
2252 iopm_len = (start_bit + size > 8) ? 2 : 1;
2253 mask = (0xf >> (4 - size)) << start_bit;
2256 if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
2257 return NESTED_EXIT_DONE;
2259 return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
2262 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
2264 u32 offset, msr, value;
2267 if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
2268 return NESTED_EXIT_HOST;
2270 msr = svm->vcpu.arch.regs[VCPU_REGS_RCX];
2271 offset = svm_msrpm_offset(msr);
2272 write = svm->vmcb->control.exit_info_1 & 1;
2273 mask = 1 << ((2 * (msr & 0xf)) + write);
2275 if (offset == MSR_INVALID)
2276 return NESTED_EXIT_DONE;
2278 /* Offset is in 32 bit units but need in 8 bit units */
2281 if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.vmcb_msrpm + offset, &value, 4))
2282 return NESTED_EXIT_DONE;
2284 return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
2287 static int nested_svm_exit_special(struct vcpu_svm *svm)
2289 u32 exit_code = svm->vmcb->control.exit_code;
2291 switch (exit_code) {
2294 case SVM_EXIT_EXCP_BASE + MC_VECTOR:
2295 return NESTED_EXIT_HOST;
2297 /* For now we are always handling NPFs when using them */
2299 return NESTED_EXIT_HOST;
2301 case SVM_EXIT_EXCP_BASE + PF_VECTOR:
2302 /* When we're shadowing, trap PFs, but not async PF */
2303 if (!npt_enabled && svm->apf_reason == 0)
2304 return NESTED_EXIT_HOST;
2306 case SVM_EXIT_EXCP_BASE + NM_VECTOR:
2307 nm_interception(svm);
2313 return NESTED_EXIT_CONTINUE;
2317 * If this function returns true, this #vmexit was already handled
2319 static int nested_svm_intercept(struct vcpu_svm *svm)
2321 u32 exit_code = svm->vmcb->control.exit_code;
2322 int vmexit = NESTED_EXIT_HOST;
2324 switch (exit_code) {
2326 vmexit = nested_svm_exit_handled_msr(svm);
2329 vmexit = nested_svm_intercept_ioio(svm);
2331 case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
2332 u32 bit = 1U << (exit_code - SVM_EXIT_READ_CR0);
2333 if (svm->nested.intercept_cr & bit)
2334 vmexit = NESTED_EXIT_DONE;
2337 case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
2338 u32 bit = 1U << (exit_code - SVM_EXIT_READ_DR0);
2339 if (svm->nested.intercept_dr & bit)
2340 vmexit = NESTED_EXIT_DONE;
2343 case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
2344 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
2345 if (svm->nested.intercept_exceptions & excp_bits)
2346 vmexit = NESTED_EXIT_DONE;
2347 /* async page fault always cause vmexit */
2348 else if ((exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) &&
2349 svm->apf_reason != 0)
2350 vmexit = NESTED_EXIT_DONE;
2353 case SVM_EXIT_ERR: {
2354 vmexit = NESTED_EXIT_DONE;
2358 u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
2359 if (svm->nested.intercept & exit_bits)
2360 vmexit = NESTED_EXIT_DONE;
2367 static int nested_svm_exit_handled(struct vcpu_svm *svm)
2371 vmexit = nested_svm_intercept(svm);
2373 if (vmexit == NESTED_EXIT_DONE)
2374 nested_svm_vmexit(svm);
2379 static inline void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
2381 struct vmcb_control_area *dst = &dst_vmcb->control;
2382 struct vmcb_control_area *from = &from_vmcb->control;
2384 dst->intercept_cr = from->intercept_cr;
2385 dst->intercept_dr = from->intercept_dr;
2386 dst->intercept_exceptions = from->intercept_exceptions;
2387 dst->intercept = from->intercept;
2388 dst->iopm_base_pa = from->iopm_base_pa;
2389 dst->msrpm_base_pa = from->msrpm_base_pa;
2390 dst->tsc_offset = from->tsc_offset;
2391 dst->asid = from->asid;
2392 dst->tlb_ctl = from->tlb_ctl;
2393 dst->int_ctl = from->int_ctl;
2394 dst->int_vector = from->int_vector;
2395 dst->int_state = from->int_state;
2396 dst->exit_code = from->exit_code;
2397 dst->exit_code_hi = from->exit_code_hi;
2398 dst->exit_info_1 = from->exit_info_1;
2399 dst->exit_info_2 = from->exit_info_2;
2400 dst->exit_int_info = from->exit_int_info;
2401 dst->exit_int_info_err = from->exit_int_info_err;
2402 dst->nested_ctl = from->nested_ctl;
2403 dst->event_inj = from->event_inj;
2404 dst->event_inj_err = from->event_inj_err;
2405 dst->nested_cr3 = from->nested_cr3;
2406 dst->lbr_ctl = from->lbr_ctl;
2409 static int nested_svm_vmexit(struct vcpu_svm *svm)
2411 struct vmcb *nested_vmcb;
2412 struct vmcb *hsave = svm->nested.hsave;
2413 struct vmcb *vmcb = svm->vmcb;
2416 trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
2417 vmcb->control.exit_info_1,
2418 vmcb->control.exit_info_2,
2419 vmcb->control.exit_int_info,
2420 vmcb->control.exit_int_info_err,
2423 nested_vmcb = nested_svm_map(svm, svm->nested.vmcb, &page);
2427 /* Exit Guest-Mode */
2428 leave_guest_mode(&svm->vcpu);
2429 svm->nested.vmcb = 0;
2431 /* Give the current vmcb to the guest */
2434 nested_vmcb->save.es = vmcb->save.es;
2435 nested_vmcb->save.cs = vmcb->save.cs;
2436 nested_vmcb->save.ss = vmcb->save.ss;
2437 nested_vmcb->save.ds = vmcb->save.ds;
2438 nested_vmcb->save.gdtr = vmcb->save.gdtr;
2439 nested_vmcb->save.idtr = vmcb->save.idtr;
2440 nested_vmcb->save.efer = svm->vcpu.arch.efer;
2441 nested_vmcb->save.cr0 = kvm_read_cr0(&svm->vcpu);
2442 nested_vmcb->save.cr3 = kvm_read_cr3(&svm->vcpu);
2443 nested_vmcb->save.cr2 = vmcb->save.cr2;
2444 nested_vmcb->save.cr4 = svm->vcpu.arch.cr4;
2445 nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu);
2446 nested_vmcb->save.rip = vmcb->save.rip;
2447 nested_vmcb->save.rsp = vmcb->save.rsp;
2448 nested_vmcb->save.rax = vmcb->save.rax;
2449 nested_vmcb->save.dr7 = vmcb->save.dr7;
2450 nested_vmcb->save.dr6 = vmcb->save.dr6;
2451 nested_vmcb->save.cpl = vmcb->save.cpl;
2453 nested_vmcb->control.int_ctl = vmcb->control.int_ctl;
2454 nested_vmcb->control.int_vector = vmcb->control.int_vector;
2455 nested_vmcb->control.int_state = vmcb->control.int_state;
2456 nested_vmcb->control.exit_code = vmcb->control.exit_code;
2457 nested_vmcb->control.exit_code_hi = vmcb->control.exit_code_hi;
2458 nested_vmcb->control.exit_info_1 = vmcb->control.exit_info_1;
2459 nested_vmcb->control.exit_info_2 = vmcb->control.exit_info_2;
2460 nested_vmcb->control.exit_int_info = vmcb->control.exit_int_info;
2461 nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
2462 nested_vmcb->control.next_rip = vmcb->control.next_rip;
2465 * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
2466 * to make sure that we do not lose injected events. So check event_inj
2467 * here and copy it to exit_int_info if it is valid.
2468 * Exit_int_info and event_inj can't be both valid because the case
2469 * below only happens on a VMRUN instruction intercept which has
2470 * no valid exit_int_info set.
2472 if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
2473 struct vmcb_control_area *nc = &nested_vmcb->control;
2475 nc->exit_int_info = vmcb->control.event_inj;
2476 nc->exit_int_info_err = vmcb->control.event_inj_err;
2479 nested_vmcb->control.tlb_ctl = 0;
2480 nested_vmcb->control.event_inj = 0;
2481 nested_vmcb->control.event_inj_err = 0;
2483 /* We always set V_INTR_MASKING and remember the old value in hflags */
2484 if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
2485 nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
2487 /* Restore the original control entries */
2488 copy_vmcb_control_area(vmcb, hsave);
2490 kvm_clear_exception_queue(&svm->vcpu);
2491 kvm_clear_interrupt_queue(&svm->vcpu);
2493 svm->nested.nested_cr3 = 0;
2495 /* Restore selected save entries */
2496 svm->vmcb->save.es = hsave->save.es;
2497 svm->vmcb->save.cs = hsave->save.cs;
2498 svm->vmcb->save.ss = hsave->save.ss;
2499 svm->vmcb->save.ds = hsave->save.ds;
2500 svm->vmcb->save.gdtr = hsave->save.gdtr;
2501 svm->vmcb->save.idtr = hsave->save.idtr;
2502 kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
2503 svm_set_efer(&svm->vcpu, hsave->save.efer);
2504 svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
2505 svm_set_cr4(&svm->vcpu, hsave->save.cr4);
2507 svm->vmcb->save.cr3 = hsave->save.cr3;
2508 svm->vcpu.arch.cr3 = hsave->save.cr3;
2510 (void)kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
2512 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
2513 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
2514 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
2515 svm->vmcb->save.dr7 = 0;
2516 svm->vmcb->save.cpl = 0;
2517 svm->vmcb->control.exit_int_info = 0;
2519 mark_all_dirty(svm->vmcb);
2521 nested_svm_unmap(page);
2523 nested_svm_uninit_mmu_context(&svm->vcpu);
2524 kvm_mmu_reset_context(&svm->vcpu);
2525 kvm_mmu_load(&svm->vcpu);
2530 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
2533 * This function merges the msr permission bitmaps of kvm and the
2534 * nested vmcb. It is optimized in that it only merges the parts where
2535 * the kvm msr permission bitmap may contain zero bits
2539 if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
2542 for (i = 0; i < MSRPM_OFFSETS; i++) {
2546 if (msrpm_offsets[i] == 0xffffffff)
2549 p = msrpm_offsets[i];
2550 offset = svm->nested.vmcb_msrpm + (p * 4);
2552 if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
2555 svm->nested.msrpm[p] = svm->msrpm[p] | value;
2558 svm->vmcb->control.msrpm_base_pa = __pa(svm->nested.msrpm);
2563 static bool nested_vmcb_checks(struct vmcb *vmcb)
2565 if ((vmcb->control.intercept & (1ULL << INTERCEPT_VMRUN)) == 0)
2568 if (vmcb->control.asid == 0)
2571 if (vmcb->control.nested_ctl && !npt_enabled)
2577 static bool nested_svm_vmrun(struct vcpu_svm *svm)
2579 struct vmcb *nested_vmcb;
2580 struct vmcb *hsave = svm->nested.hsave;
2581 struct vmcb *vmcb = svm->vmcb;
2585 vmcb_gpa = svm->vmcb->save.rax;
2587 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2591 if (!nested_vmcb_checks(nested_vmcb)) {
2592 nested_vmcb->control.exit_code = SVM_EXIT_ERR;
2593 nested_vmcb->control.exit_code_hi = 0;
2594 nested_vmcb->control.exit_info_1 = 0;
2595 nested_vmcb->control.exit_info_2 = 0;
2597 nested_svm_unmap(page);
2602 trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa,
2603 nested_vmcb->save.rip,
2604 nested_vmcb->control.int_ctl,
2605 nested_vmcb->control.event_inj,
2606 nested_vmcb->control.nested_ctl);
2608 trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr & 0xffff,
2609 nested_vmcb->control.intercept_cr >> 16,
2610 nested_vmcb->control.intercept_exceptions,
2611 nested_vmcb->control.intercept);
2613 /* Clear internal status */
2614 kvm_clear_exception_queue(&svm->vcpu);
2615 kvm_clear_interrupt_queue(&svm->vcpu);
2618 * Save the old vmcb, so we don't need to pick what we save, but can
2619 * restore everything when a VMEXIT occurs
2621 hsave->save.es = vmcb->save.es;
2622 hsave->save.cs = vmcb->save.cs;
2623 hsave->save.ss = vmcb->save.ss;
2624 hsave->save.ds = vmcb->save.ds;
2625 hsave->save.gdtr = vmcb->save.gdtr;
2626 hsave->save.idtr = vmcb->save.idtr;
2627 hsave->save.efer = svm->vcpu.arch.efer;
2628 hsave->save.cr0 = kvm_read_cr0(&svm->vcpu);
2629 hsave->save.cr4 = svm->vcpu.arch.cr4;
2630 hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
2631 hsave->save.rip = kvm_rip_read(&svm->vcpu);
2632 hsave->save.rsp = vmcb->save.rsp;
2633 hsave->save.rax = vmcb->save.rax;
2635 hsave->save.cr3 = vmcb->save.cr3;
2637 hsave->save.cr3 = kvm_read_cr3(&svm->vcpu);
2639 copy_vmcb_control_area(hsave, vmcb);
2641 if (kvm_get_rflags(&svm->vcpu) & X86_EFLAGS_IF)
2642 svm->vcpu.arch.hflags |= HF_HIF_MASK;
2644 svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
2646 if (nested_vmcb->control.nested_ctl) {
2647 kvm_mmu_unload(&svm->vcpu);
2648 svm->nested.nested_cr3 = nested_vmcb->control.nested_cr3;
2649 nested_svm_init_mmu_context(&svm->vcpu);
2652 /* Load the nested guest state */
2653 svm->vmcb->save.es = nested_vmcb->save.es;
2654 svm->vmcb->save.cs = nested_vmcb->save.cs;
2655 svm->vmcb->save.ss = nested_vmcb->save.ss;
2656 svm->vmcb->save.ds = nested_vmcb->save.ds;
2657 svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
2658 svm->vmcb->save.idtr = nested_vmcb->save.idtr;
2659 kvm_set_rflags(&svm->vcpu, nested_vmcb->save.rflags);
2660 svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
2661 svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
2662 svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
2664 svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
2665 svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
2667 (void)kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
2669 /* Guest paging mode is active - reset mmu */
2670 kvm_mmu_reset_context(&svm->vcpu);
2672 svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
2673 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, nested_vmcb->save.rax);
2674 kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, nested_vmcb->save.rsp);
2675 kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, nested_vmcb->save.rip);
2677 /* In case we don't even reach vcpu_run, the fields are not updated */
2678 svm->vmcb->save.rax = nested_vmcb->save.rax;
2679 svm->vmcb->save.rsp = nested_vmcb->save.rsp;
2680 svm->vmcb->save.rip = nested_vmcb->save.rip;
2681 svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
2682 svm->vmcb->save.dr6 = nested_vmcb->save.dr6;
2683 svm->vmcb->save.cpl = nested_vmcb->save.cpl;
2685 svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL;
2686 svm->nested.vmcb_iopm = nested_vmcb->control.iopm_base_pa & ~0x0fffULL;
2688 /* cache intercepts */
2689 svm->nested.intercept_cr = nested_vmcb->control.intercept_cr;
2690 svm->nested.intercept_dr = nested_vmcb->control.intercept_dr;
2691 svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
2692 svm->nested.intercept = nested_vmcb->control.intercept;
2694 svm_flush_tlb(&svm->vcpu);
2695 svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
2696 if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
2697 svm->vcpu.arch.hflags |= HF_VINTR_MASK;
2699 svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
2701 if (svm->vcpu.arch.hflags & HF_VINTR_MASK) {
2702 /* We only want the cr8 intercept bits of the guest */
2703 clr_cr_intercept(svm, INTERCEPT_CR8_READ);
2704 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
2707 /* We don't want to see VMMCALLs from a nested guest */
2708 clr_intercept(svm, INTERCEPT_VMMCALL);
2710 svm->vmcb->control.lbr_ctl = nested_vmcb->control.lbr_ctl;
2711 svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
2712 svm->vmcb->control.int_state = nested_vmcb->control.int_state;
2713 svm->vmcb->control.tsc_offset += nested_vmcb->control.tsc_offset;
2714 svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
2715 svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
2717 nested_svm_unmap(page);
2719 /* Enter Guest-Mode */
2720 enter_guest_mode(&svm->vcpu);
2723 * Merge guest and host intercepts - must be called with vcpu in
2724 * guest-mode to take affect here
2726 recalc_intercepts(svm);
2728 svm->nested.vmcb = vmcb_gpa;
2732 mark_all_dirty(svm->vmcb);
2737 static void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
2739 to_vmcb->save.fs = from_vmcb->save.fs;
2740 to_vmcb->save.gs = from_vmcb->save.gs;
2741 to_vmcb->save.tr = from_vmcb->save.tr;
2742 to_vmcb->save.ldtr = from_vmcb->save.ldtr;
2743 to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
2744 to_vmcb->save.star = from_vmcb->save.star;
2745 to_vmcb->save.lstar = from_vmcb->save.lstar;
2746 to_vmcb->save.cstar = from_vmcb->save.cstar;
2747 to_vmcb->save.sfmask = from_vmcb->save.sfmask;
2748 to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
2749 to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
2750 to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
2753 static int vmload_interception(struct vcpu_svm *svm)
2755 struct vmcb *nested_vmcb;
2758 if (nested_svm_check_permissions(svm))
2761 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2765 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2766 skip_emulated_instruction(&svm->vcpu);
2768 nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
2769 nested_svm_unmap(page);
2774 static int vmsave_interception(struct vcpu_svm *svm)
2776 struct vmcb *nested_vmcb;
2779 if (nested_svm_check_permissions(svm))
2782 nested_vmcb = nested_svm_map(svm, svm->vmcb->save.rax, &page);
2786 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2787 skip_emulated_instruction(&svm->vcpu);
2789 nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
2790 nested_svm_unmap(page);
2795 static int vmrun_interception(struct vcpu_svm *svm)
2797 if (nested_svm_check_permissions(svm))
2800 /* Save rip after vmrun instruction */
2801 kvm_rip_write(&svm->vcpu, kvm_rip_read(&svm->vcpu) + 3);
2803 if (!nested_svm_vmrun(svm))
2806 if (!nested_svm_vmrun_msrpm(svm))
2813 svm->vmcb->control.exit_code = SVM_EXIT_ERR;
2814 svm->vmcb->control.exit_code_hi = 0;
2815 svm->vmcb->control.exit_info_1 = 0;
2816 svm->vmcb->control.exit_info_2 = 0;
2818 nested_svm_vmexit(svm);
2823 static int stgi_interception(struct vcpu_svm *svm)
2825 if (nested_svm_check_permissions(svm))
2828 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2829 skip_emulated_instruction(&svm->vcpu);
2830 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2837 static int clgi_interception(struct vcpu_svm *svm)
2839 if (nested_svm_check_permissions(svm))
2842 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2843 skip_emulated_instruction(&svm->vcpu);
2847 /* After a CLGI no interrupts should come */
2848 svm_clear_vintr(svm);
2849 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
2851 mark_dirty(svm->vmcb, VMCB_INTR);
2856 static int invlpga_interception(struct vcpu_svm *svm)
2858 struct kvm_vcpu *vcpu = &svm->vcpu;
2860 trace_kvm_invlpga(svm->vmcb->save.rip, kvm_register_read(&svm->vcpu, VCPU_REGS_RCX),
2861 kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
2863 /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
2864 kvm_mmu_invlpg(vcpu, kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
2866 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2867 skip_emulated_instruction(&svm->vcpu);
2871 static int skinit_interception(struct vcpu_svm *svm)
2873 trace_kvm_skinit(svm->vmcb->save.rip, kvm_register_read(&svm->vcpu, VCPU_REGS_RAX));
2875 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2879 static int wbinvd_interception(struct vcpu_svm *svm)
2881 kvm_emulate_wbinvd(&svm->vcpu);
2885 static int xsetbv_interception(struct vcpu_svm *svm)
2887 u64 new_bv = kvm_read_edx_eax(&svm->vcpu);
2888 u32 index = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
2890 if (kvm_set_xcr(&svm->vcpu, index, new_bv) == 0) {
2891 svm->next_rip = kvm_rip_read(&svm->vcpu) + 3;
2892 skip_emulated_instruction(&svm->vcpu);
2898 static int task_switch_interception(struct vcpu_svm *svm)
2902 int int_type = svm->vmcb->control.exit_int_info &
2903 SVM_EXITINTINFO_TYPE_MASK;
2904 int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
2906 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
2908 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
2909 bool has_error_code = false;
2912 tss_selector = (u16)svm->vmcb->control.exit_info_1;
2914 if (svm->vmcb->control.exit_info_2 &
2915 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
2916 reason = TASK_SWITCH_IRET;
2917 else if (svm->vmcb->control.exit_info_2 &
2918 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
2919 reason = TASK_SWITCH_JMP;
2921 reason = TASK_SWITCH_GATE;
2923 reason = TASK_SWITCH_CALL;
2925 if (reason == TASK_SWITCH_GATE) {
2927 case SVM_EXITINTINFO_TYPE_NMI:
2928 svm->vcpu.arch.nmi_injected = false;
2930 case SVM_EXITINTINFO_TYPE_EXEPT:
2931 if (svm->vmcb->control.exit_info_2 &
2932 (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) {
2933 has_error_code = true;
2935 (u32)svm->vmcb->control.exit_info_2;
2937 kvm_clear_exception_queue(&svm->vcpu);
2939 case SVM_EXITINTINFO_TYPE_INTR:
2940 kvm_clear_interrupt_queue(&svm->vcpu);
2947 if (reason != TASK_SWITCH_GATE ||
2948 int_type == SVM_EXITINTINFO_TYPE_SOFT ||
2949 (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
2950 (int_vec == OF_VECTOR || int_vec == BP_VECTOR)))
2951 skip_emulated_instruction(&svm->vcpu);
2953 if (int_type != SVM_EXITINTINFO_TYPE_SOFT)
2956 if (kvm_task_switch(&svm->vcpu, tss_selector, int_vec, reason,
2957 has_error_code, error_code) == EMULATE_FAIL) {
2958 svm->vcpu.run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2959 svm->vcpu.run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
2960 svm->vcpu.run->internal.ndata = 0;
2966 static int cpuid_interception(struct vcpu_svm *svm)
2968 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
2969 kvm_emulate_cpuid(&svm->vcpu);
2973 static int iret_interception(struct vcpu_svm *svm)
2975 ++svm->vcpu.stat.nmi_window_exits;
2976 clr_intercept(svm, INTERCEPT_IRET);
2977 svm->vcpu.arch.hflags |= HF_IRET_MASK;
2978 svm->nmi_iret_rip = kvm_rip_read(&svm->vcpu);
2979 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2983 static int invlpg_interception(struct vcpu_svm *svm)
2985 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2986 return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
2988 kvm_mmu_invlpg(&svm->vcpu, svm->vmcb->control.exit_info_1);
2989 skip_emulated_instruction(&svm->vcpu);
2993 static int emulate_on_interception(struct vcpu_svm *svm)
2995 return emulate_instruction(&svm->vcpu, 0) == EMULATE_DONE;
2998 static int rdpmc_interception(struct vcpu_svm *svm)
3002 if (!static_cpu_has(X86_FEATURE_NRIPS))
3003 return emulate_on_interception(svm);
3005 err = kvm_rdpmc(&svm->vcpu);
3006 kvm_complete_insn_gp(&svm->vcpu, err);
3011 static bool check_selective_cr0_intercepted(struct vcpu_svm *svm,
3014 unsigned long cr0 = svm->vcpu.arch.cr0;
3018 intercept = svm->nested.intercept;
3020 if (!is_guest_mode(&svm->vcpu) ||
3021 (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0))))
3024 cr0 &= ~SVM_CR0_SELECTIVE_MASK;
3025 val &= ~SVM_CR0_SELECTIVE_MASK;
3028 svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
3029 ret = (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE);
3035 #define CR_VALID (1ULL << 63)
3037 static int cr_interception(struct vcpu_svm *svm)
3043 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
3044 return emulate_on_interception(svm);
3046 if (unlikely((svm->vmcb->control.exit_info_1 & CR_VALID) == 0))
3047 return emulate_on_interception(svm);
3049 reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
3050 if (svm->vmcb->control.exit_code == SVM_EXIT_CR0_SEL_WRITE)
3051 cr = SVM_EXIT_WRITE_CR0 - SVM_EXIT_READ_CR0;
3053 cr = svm->vmcb->control.exit_code - SVM_EXIT_READ_CR0;
3056 if (cr >= 16) { /* mov to cr */
3058 val = kvm_register_read(&svm->vcpu, reg);
3061 if (!check_selective_cr0_intercepted(svm, val))
3062 err = kvm_set_cr0(&svm->vcpu, val);
3068 err = kvm_set_cr3(&svm->vcpu, val);
3071 err = kvm_set_cr4(&svm->vcpu, val);
3074 err = kvm_set_cr8(&svm->vcpu, val);
3077 WARN(1, "unhandled write to CR%d", cr);
3078 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
3081 } else { /* mov from cr */
3084 val = kvm_read_cr0(&svm->vcpu);
3087 val = svm->vcpu.arch.cr2;
3090 val = kvm_read_cr3(&svm->vcpu);
3093 val = kvm_read_cr4(&svm->vcpu);
3096 val = kvm_get_cr8(&svm->vcpu);
3099 WARN(1, "unhandled read from CR%d", cr);
3100 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
3103 kvm_register_write(&svm->vcpu, reg, val);
3105 kvm_complete_insn_gp(&svm->vcpu, err);
3110 static int dr_interception(struct vcpu_svm *svm)
3115 if (svm->vcpu.guest_debug == 0) {
3117 * No more DR vmexits; force a reload of the debug registers
3118 * and reenter on this instruction. The next vmexit will
3119 * retrieve the full state of the debug registers.
3121 clr_dr_intercepts(svm);
3122 svm->vcpu.arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
3126 if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS))
3127 return emulate_on_interception(svm);
3129 reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
3130 dr = svm->vmcb->control.exit_code - SVM_EXIT_READ_DR0;
3132 if (dr >= 16) { /* mov to DRn */
3133 if (!kvm_require_dr(&svm->vcpu, dr - 16))
3135 val = kvm_register_read(&svm->vcpu, reg);
3136 kvm_set_dr(&svm->vcpu, dr - 16, val);
3138 if (!kvm_require_dr(&svm->vcpu, dr))
3140 kvm_get_dr(&svm->vcpu, dr, &val);
3141 kvm_register_write(&svm->vcpu, reg, val);
3144 skip_emulated_instruction(&svm->vcpu);
3149 static int cr8_write_interception(struct vcpu_svm *svm)
3151 struct kvm_run *kvm_run = svm->vcpu.run;
3154 u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
3155 /* instruction emulation calls kvm_set_cr8() */
3156 r = cr_interception(svm);
3157 if (irqchip_in_kernel(svm->vcpu.kvm))
3159 if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
3161 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
3165 static u64 svm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
3167 struct vmcb *vmcb = get_host_vmcb(to_svm(vcpu));
3168 return vmcb->control.tsc_offset +
3169 svm_scale_tsc(vcpu, host_tsc);
3172 static int svm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3174 struct vcpu_svm *svm = to_svm(vcpu);
3176 switch (msr_info->index) {
3177 case MSR_IA32_TSC: {
3178 msr_info->data = svm->vmcb->control.tsc_offset +
3179 svm_scale_tsc(vcpu, rdtsc());
3184 msr_info->data = svm->vmcb->save.star;
3186 #ifdef CONFIG_X86_64
3188 msr_info->data = svm->vmcb->save.lstar;
3191 msr_info->data = svm->vmcb->save.cstar;
3193 case MSR_KERNEL_GS_BASE:
3194 msr_info->data = svm->vmcb->save.kernel_gs_base;
3196 case MSR_SYSCALL_MASK:
3197 msr_info->data = svm->vmcb->save.sfmask;
3200 case MSR_IA32_SYSENTER_CS:
3201 msr_info->data = svm->vmcb->save.sysenter_cs;
3203 case MSR_IA32_SYSENTER_EIP:
3204 msr_info->data = svm->sysenter_eip;
3206 case MSR_IA32_SYSENTER_ESP:
3207 msr_info->data = svm->sysenter_esp;
3210 * Nobody will change the following 5 values in the VMCB so we can
3211 * safely return them on rdmsr. They will always be 0 until LBRV is
3214 case MSR_IA32_DEBUGCTLMSR:
3215 msr_info->data = svm->vmcb->save.dbgctl;
3217 case MSR_IA32_LASTBRANCHFROMIP:
3218 msr_info->data = svm->vmcb->save.br_from;
3220 case MSR_IA32_LASTBRANCHTOIP:
3221 msr_info->data = svm->vmcb->save.br_to;
3223 case MSR_IA32_LASTINTFROMIP:
3224 msr_info->data = svm->vmcb->save.last_excp_from;
3226 case MSR_IA32_LASTINTTOIP:
3227 msr_info->data = svm->vmcb->save.last_excp_to;
3229 case MSR_VM_HSAVE_PA:
3230 msr_info->data = svm->nested.hsave_msr;
3233 msr_info->data = svm->nested.vm_cr_msr;
3235 case MSR_IA32_UCODE_REV:
3236 msr_info->data = 0x01000065;
3239 return kvm_get_msr_common(vcpu, msr_info);
3244 static int rdmsr_interception(struct vcpu_svm *svm)
3246 u32 ecx = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
3247 struct msr_data msr_info;
3249 msr_info.index = ecx;
3250 msr_info.host_initiated = false;
3251 if (svm_get_msr(&svm->vcpu, &msr_info)) {
3252 trace_kvm_msr_read_ex(ecx);
3253 kvm_inject_gp(&svm->vcpu, 0);
3255 trace_kvm_msr_read(ecx, msr_info.data);
3257 kvm_register_write(&svm->vcpu, VCPU_REGS_RAX,
3258 msr_info.data & 0xffffffff);
3259 kvm_register_write(&svm->vcpu, VCPU_REGS_RDX,
3260 msr_info.data >> 32);
3261 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3262 skip_emulated_instruction(&svm->vcpu);
3267 static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
3269 struct vcpu_svm *svm = to_svm(vcpu);
3270 int svm_dis, chg_mask;
3272 if (data & ~SVM_VM_CR_VALID_MASK)
3275 chg_mask = SVM_VM_CR_VALID_MASK;
3277 if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
3278 chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
3280 svm->nested.vm_cr_msr &= ~chg_mask;
3281 svm->nested.vm_cr_msr |= (data & chg_mask);
3283 svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
3285 /* check for svm_disable while efer.svme is set */
3286 if (svm_dis && (vcpu->arch.efer & EFER_SVME))
3292 static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
3294 struct vcpu_svm *svm = to_svm(vcpu);
3296 u32 ecx = msr->index;
3297 u64 data = msr->data;
3300 kvm_write_tsc(vcpu, msr);
3303 svm->vmcb->save.star = data;
3305 #ifdef CONFIG_X86_64
3307 svm->vmcb->save.lstar = data;
3310 svm->vmcb->save.cstar = data;
3312 case MSR_KERNEL_GS_BASE:
3313 svm->vmcb->save.kernel_gs_base = data;
3315 case MSR_SYSCALL_MASK:
3316 svm->vmcb->save.sfmask = data;
3319 case MSR_IA32_SYSENTER_CS:
3320 svm->vmcb->save.sysenter_cs = data;
3322 case MSR_IA32_SYSENTER_EIP:
3323 svm->sysenter_eip = data;
3324 svm->vmcb->save.sysenter_eip = data;
3326 case MSR_IA32_SYSENTER_ESP:
3327 svm->sysenter_esp = data;
3328 svm->vmcb->save.sysenter_esp = data;
3330 case MSR_IA32_DEBUGCTLMSR:
3331 if (!boot_cpu_has(X86_FEATURE_LBRV)) {
3332 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
3336 if (data & DEBUGCTL_RESERVED_BITS)
3339 svm->vmcb->save.dbgctl = data;
3340 mark_dirty(svm->vmcb, VMCB_LBR);
3341 if (data & (1ULL<<0))
3342 svm_enable_lbrv(svm);
3344 svm_disable_lbrv(svm);
3346 case MSR_VM_HSAVE_PA:
3347 svm->nested.hsave_msr = data;
3350 return svm_set_vm_cr(vcpu, data);
3352 vcpu_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
3354 case MSR_IA32_CR_PAT:
3356 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
3358 vcpu->arch.pat = data;
3359 svm_set_guest_pat(svm, &svm->vmcb->save.g_pat);
3360 mark_dirty(svm->vmcb, VMCB_NPT);
3365 return kvm_set_msr_common(vcpu, msr);
3370 static int wrmsr_interception(struct vcpu_svm *svm)
3372 struct msr_data msr;
3373 u32 ecx = kvm_register_read(&svm->vcpu, VCPU_REGS_RCX);
3374 u64 data = kvm_read_edx_eax(&svm->vcpu);
3378 msr.host_initiated = false;
3380 svm->next_rip = kvm_rip_read(&svm->vcpu) + 2;
3381 if (kvm_set_msr(&svm->vcpu, &msr)) {
3382 trace_kvm_msr_write_ex(ecx, data);
3383 kvm_inject_gp(&svm->vcpu, 0);
3385 trace_kvm_msr_write(ecx, data);
3386 skip_emulated_instruction(&svm->vcpu);
3391 static int msr_interception(struct vcpu_svm *svm)
3393 if (svm->vmcb->control.exit_info_1)
3394 return wrmsr_interception(svm);
3396 return rdmsr_interception(svm);
3399 static int interrupt_window_interception(struct vcpu_svm *svm)
3401 struct kvm_run *kvm_run = svm->vcpu.run;
3403 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3404 svm_clear_vintr(svm);
3405 svm->vmcb->control.int_ctl &= ~V_IRQ_MASK;
3406 mark_dirty(svm->vmcb, VMCB_INTR);
3407 ++svm->vcpu.stat.irq_window_exits;
3409 * If the user space waits to inject interrupts, exit as soon as
3412 if (!irqchip_in_kernel(svm->vcpu.kvm) &&
3413 kvm_run->request_interrupt_window &&
3414 !kvm_cpu_has_interrupt(&svm->vcpu)) {
3415 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
3422 static int pause_interception(struct vcpu_svm *svm)
3424 kvm_vcpu_on_spin(&(svm->vcpu));
3428 static int nop_interception(struct vcpu_svm *svm)
3430 skip_emulated_instruction(&(svm->vcpu));
3434 static int monitor_interception(struct vcpu_svm *svm)
3436 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
3437 return nop_interception(svm);
3440 static int mwait_interception(struct vcpu_svm *svm)
3442 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
3443 return nop_interception(svm);
3446 static int (*const svm_exit_handlers[])(struct vcpu_svm *svm) = {
3447 [SVM_EXIT_READ_CR0] = cr_interception,
3448 [SVM_EXIT_READ_CR3] = cr_interception,
3449 [SVM_EXIT_READ_CR4] = cr_interception,
3450 [SVM_EXIT_READ_CR8] = cr_interception,
3451 [SVM_EXIT_CR0_SEL_WRITE] = cr_interception,
3452 [SVM_EXIT_WRITE_CR0] = cr_interception,
3453 [SVM_EXIT_WRITE_CR3] = cr_interception,
3454 [SVM_EXIT_WRITE_CR4] = cr_interception,
3455 [SVM_EXIT_WRITE_CR8] = cr8_write_interception,
3456 [SVM_EXIT_READ_DR0] = dr_interception,
3457 [SVM_EXIT_READ_DR1] = dr_interception,
3458 [SVM_EXIT_READ_DR2] = dr_interception,
3459 [SVM_EXIT_READ_DR3] = dr_interception,
3460 [SVM_EXIT_READ_DR4] = dr_interception,
3461 [SVM_EXIT_READ_DR5] = dr_interception,
3462 [SVM_EXIT_READ_DR6] = dr_interception,
3463 [SVM_EXIT_READ_DR7] = dr_interception,
3464 [SVM_EXIT_WRITE_DR0] = dr_interception,
3465 [SVM_EXIT_WRITE_DR1] = dr_interception,
3466 [SVM_EXIT_WRITE_DR2] = dr_interception,
3467 [SVM_EXIT_WRITE_DR3] = dr_interception,
3468 [SVM_EXIT_WRITE_DR4] = dr_interception,
3469 [SVM_EXIT_WRITE_DR5] = dr_interception,
3470 [SVM_EXIT_WRITE_DR6] = dr_interception,
3471 [SVM_EXIT_WRITE_DR7] = dr_interception,
3472 [SVM_EXIT_EXCP_BASE + DB_VECTOR] = db_interception,
3473 [SVM_EXIT_EXCP_BASE + BP_VECTOR] = bp_interception,
3474 [SVM_EXIT_EXCP_BASE + UD_VECTOR] = ud_interception,
3475 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
3476 [SVM_EXIT_EXCP_BASE + NM_VECTOR] = nm_interception,
3477 [SVM_EXIT_EXCP_BASE + MC_VECTOR] = mc_interception,
3478 [SVM_EXIT_INTR] = intr_interception,
3479 [SVM_EXIT_NMI] = nmi_interception,
3480 [SVM_EXIT_SMI] = nop_on_interception,
3481 [SVM_EXIT_INIT] = nop_on_interception,
3482 [SVM_EXIT_VINTR] = interrupt_window_interception,
3483 [SVM_EXIT_RDPMC] = rdpmc_interception,
3484 [SVM_EXIT_CPUID] = cpuid_interception,
3485 [SVM_EXIT_IRET] = iret_interception,
3486 [SVM_EXIT_INVD] = emulate_on_interception,
3487 [SVM_EXIT_PAUSE] = pause_interception,
3488 [SVM_EXIT_HLT] = halt_interception,
3489 [SVM_EXIT_INVLPG] = invlpg_interception,
3490 [SVM_EXIT_INVLPGA] = invlpga_interception,
3491 [SVM_EXIT_IOIO] = io_interception,
3492 [SVM_EXIT_MSR] = msr_interception,
3493 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
3494 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
3495 [SVM_EXIT_VMRUN] = vmrun_interception,
3496 [SVM_EXIT_VMMCALL] = vmmcall_interception,
3497 [SVM_EXIT_VMLOAD] = vmload_interception,
3498 [SVM_EXIT_VMSAVE] = vmsave_interception,
3499 [SVM_EXIT_STGI] = stgi_interception,
3500 [SVM_EXIT_CLGI] = clgi_interception,
3501 [SVM_EXIT_SKINIT] = skinit_interception,
3502 [SVM_EXIT_WBINVD] = wbinvd_interception,
3503 [SVM_EXIT_MONITOR] = monitor_interception,
3504 [SVM_EXIT_MWAIT] = mwait_interception,
3505 [SVM_EXIT_XSETBV] = xsetbv_interception,
3506 [SVM_EXIT_NPF] = pf_interception,
3507 [SVM_EXIT_RSM] = emulate_on_interception,
3510 static void dump_vmcb(struct kvm_vcpu *vcpu)
3512 struct vcpu_svm *svm = to_svm(vcpu);
3513 struct vmcb_control_area *control = &svm->vmcb->control;
3514 struct vmcb_save_area *save = &svm->vmcb->save;
3516 pr_err("VMCB Control Area:\n");
3517 pr_err("%-20s%04x\n", "cr_read:", control->intercept_cr & 0xffff);
3518 pr_err("%-20s%04x\n", "cr_write:", control->intercept_cr >> 16);
3519 pr_err("%-20s%04x\n", "dr_read:", control->intercept_dr & 0xffff);
3520 pr_err("%-20s%04x\n", "dr_write:", control->intercept_dr >> 16);
3521 pr_err("%-20s%08x\n", "exceptions:", control->intercept_exceptions);
3522 pr_err("%-20s%016llx\n", "intercepts:", control->intercept);
3523 pr_err("%-20s%d\n", "pause filter count:", control->pause_filter_count);
3524 pr_err("%-20s%016llx\n", "iopm_base_pa:", control->iopm_base_pa);
3525 pr_err("%-20s%016llx\n", "msrpm_base_pa:", control->msrpm_base_pa);
3526 pr_err("%-20s%016llx\n", "tsc_offset:", control->tsc_offset);
3527 pr_err("%-20s%d\n", "asid:", control->asid);
3528 pr_err("%-20s%d\n", "tlb_ctl:", control->tlb_ctl);
3529 pr_err("%-20s%08x\n", "int_ctl:", control->int_ctl);
3530 pr_err("%-20s%08x\n", "int_vector:", control->int_vector);
3531 pr_err("%-20s%08x\n", "int_state:", control->int_state);
3532 pr_err("%-20s%08x\n", "exit_code:", control->exit_code);
3533 pr_err("%-20s%016llx\n", "exit_info1:", control->exit_info_1);
3534 pr_err("%-20s%016llx\n", "exit_info2:", control->exit_info_2);
3535 pr_err("%-20s%08x\n", "exit_int_info:", control->exit_int_info);
3536 pr_err("%-20s%08x\n", "exit_int_info_err:", control->exit_int_info_err);
3537 pr_err("%-20s%lld\n", "nested_ctl:", control->nested_ctl);
3538 pr_err("%-20s%016llx\n", "nested_cr3:", control->nested_cr3);
3539 pr_err("%-20s%08x\n", "event_inj:", control->event_inj);
3540 pr_err("%-20s%08x\n", "event_inj_err:", control->event_inj_err);
3541 pr_err("%-20s%lld\n", "lbr_ctl:", control->lbr_ctl);
3542 pr_err("%-20s%016llx\n", "next_rip:", control->next_rip);
3543 pr_err("VMCB State Save Area:\n");
3544 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3546 save->es.selector, save->es.attrib,
3547 save->es.limit, save->es.base);
3548 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3550 save->cs.selector, save->cs.attrib,
3551 save->cs.limit, save->cs.base);
3552 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3554 save->ss.selector, save->ss.attrib,
3555 save->ss.limit, save->ss.base);
3556 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3558 save->ds.selector, save->ds.attrib,
3559 save->ds.limit, save->ds.base);
3560 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3562 save->fs.selector, save->fs.attrib,
3563 save->fs.limit, save->fs.base);
3564 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3566 save->gs.selector, save->gs.attrib,
3567 save->gs.limit, save->gs.base);
3568 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3570 save->gdtr.selector, save->gdtr.attrib,
3571 save->gdtr.limit, save->gdtr.base);
3572 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3574 save->ldtr.selector, save->ldtr.attrib,
3575 save->ldtr.limit, save->ldtr.base);
3576 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3578 save->idtr.selector, save->idtr.attrib,
3579 save->idtr.limit, save->idtr.base);
3580 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3582 save->tr.selector, save->tr.attrib,
3583 save->tr.limit, save->tr.base);
3584 pr_err("cpl: %d efer: %016llx\n",
3585 save->cpl, save->efer);
3586 pr_err("%-15s %016llx %-13s %016llx\n",
3587 "cr0:", save->cr0, "cr2:", save->cr2);
3588 pr_err("%-15s %016llx %-13s %016llx\n",
3589 "cr3:", save->cr3, "cr4:", save->cr4);
3590 pr_err("%-15s %016llx %-13s %016llx\n",
3591 "dr6:", save->dr6, "dr7:", save->dr7);
3592 pr_err("%-15s %016llx %-13s %016llx\n",
3593 "rip:", save->rip, "rflags:", save->rflags);
3594 pr_err("%-15s %016llx %-13s %016llx\n",
3595 "rsp:", save->rsp, "rax:", save->rax);
3596 pr_err("%-15s %016llx %-13s %016llx\n",
3597 "star:", save->star, "lstar:", save->lstar);
3598 pr_err("%-15s %016llx %-13s %016llx\n",
3599 "cstar:", save->cstar, "sfmask:", save->sfmask);
3600 pr_err("%-15s %016llx %-13s %016llx\n",
3601 "kernel_gs_base:", save->kernel_gs_base,
3602 "sysenter_cs:", save->sysenter_cs);
3603 pr_err("%-15s %016llx %-13s %016llx\n",
3604 "sysenter_esp:", save->sysenter_esp,
3605 "sysenter_eip:", save->sysenter_eip);
3606 pr_err("%-15s %016llx %-13s %016llx\n",
3607 "gpat:", save->g_pat, "dbgctl:", save->dbgctl);
3608 pr_err("%-15s %016llx %-13s %016llx\n",
3609 "br_from:", save->br_from, "br_to:", save->br_to);
3610 pr_err("%-15s %016llx %-13s %016llx\n",
3611 "excp_from:", save->last_excp_from,
3612 "excp_to:", save->last_excp_to);
3615 static void svm_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
3617 struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control;
3619 *info1 = control->exit_info_1;
3620 *info2 = control->exit_info_2;
3623 static int handle_exit(struct kvm_vcpu *vcpu)
3625 struct vcpu_svm *svm = to_svm(vcpu);
3626 struct kvm_run *kvm_run = vcpu->run;
3627 u32 exit_code = svm->vmcb->control.exit_code;
3629 if (!is_cr_intercept(svm, INTERCEPT_CR0_WRITE))
3630 vcpu->arch.cr0 = svm->vmcb->save.cr0;
3632 vcpu->arch.cr3 = svm->vmcb->save.cr3;
3634 if (unlikely(svm->nested.exit_required)) {
3635 nested_svm_vmexit(svm);
3636 svm->nested.exit_required = false;
3641 if (is_guest_mode(vcpu)) {
3644 trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code,
3645 svm->vmcb->control.exit_info_1,
3646 svm->vmcb->control.exit_info_2,
3647 svm->vmcb->control.exit_int_info,
3648 svm->vmcb->control.exit_int_info_err,
3651 vmexit = nested_svm_exit_special(svm);
3653 if (vmexit == NESTED_EXIT_CONTINUE)
3654 vmexit = nested_svm_exit_handled(svm);
3656 if (vmexit == NESTED_EXIT_DONE)
3660 svm_complete_interrupts(svm);
3662 if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
3663 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3664 kvm_run->fail_entry.hardware_entry_failure_reason
3665 = svm->vmcb->control.exit_code;
3666 pr_err("KVM: FAILED VMRUN WITH VMCB:\n");
3671 if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
3672 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
3673 exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH &&
3674 exit_code != SVM_EXIT_INTR && exit_code != SVM_EXIT_NMI)
3675 printk(KERN_ERR "%s: unexpected exit_int_info 0x%x "
3677 __func__, svm->vmcb->control.exit_int_info,
3680 if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
3681 || !svm_exit_handlers[exit_code]) {
3682 WARN_ONCE(1, "svm: unexpected exit reason 0x%x\n", exit_code);
3683 kvm_queue_exception(vcpu, UD_VECTOR);
3687 return svm_exit_handlers[exit_code](svm);
3690 static void reload_tss(struct kvm_vcpu *vcpu)
3692 int cpu = raw_smp_processor_id();
3694 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
3695 sd->tss_desc->type = 9; /* available 32/64-bit TSS */
3699 static void pre_svm_run(struct vcpu_svm *svm)
3701 int cpu = raw_smp_processor_id();
3703 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
3705 /* FIXME: handle wraparound of asid_generation */
3706 if (svm->asid_generation != sd->asid_generation)
3710 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
3712 struct vcpu_svm *svm = to_svm(vcpu);
3714 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
3715 vcpu->arch.hflags |= HF_NMI_MASK;
3716 set_intercept(svm, INTERCEPT_IRET);
3717 ++vcpu->stat.nmi_injections;
3720 static inline void svm_inject_irq(struct vcpu_svm *svm, int irq)
3722 struct vmcb_control_area *control;
3724 control = &svm->vmcb->control;
3725 control->int_vector = irq;
3726 control->int_ctl &= ~V_INTR_PRIO_MASK;
3727 control->int_ctl |= V_IRQ_MASK |
3728 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
3729 mark_dirty(svm->vmcb, VMCB_INTR);
3732 static void svm_set_irq(struct kvm_vcpu *vcpu)
3734 struct vcpu_svm *svm = to_svm(vcpu);
3736 BUG_ON(!(gif_set(svm)));
3738 trace_kvm_inj_virq(vcpu->arch.interrupt.nr);
3739 ++vcpu->stat.irq_injections;
3741 svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
3742 SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
3745 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3747 struct vcpu_svm *svm = to_svm(vcpu);
3749 if (is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK))
3752 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
3758 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
3761 static void svm_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
3766 static int svm_vm_has_apicv(struct kvm *kvm)
3771 static void svm_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
3776 static void svm_sync_pir_to_irr(struct kvm_vcpu *vcpu)
3781 static int svm_nmi_allowed(struct kvm_vcpu *vcpu)
3783 struct vcpu_svm *svm = to_svm(vcpu);
3784 struct vmcb *vmcb = svm->vmcb;
3786 ret = !(vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) &&
3787 !(svm->vcpu.arch.hflags & HF_NMI_MASK);
3788 ret = ret && gif_set(svm) && nested_svm_nmi(svm);
3793 static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
3795 struct vcpu_svm *svm = to_svm(vcpu);
3797 return !!(svm->vcpu.arch.hflags & HF_NMI_MASK);
3800 static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
3802 struct vcpu_svm *svm = to_svm(vcpu);
3805 svm->vcpu.arch.hflags |= HF_NMI_MASK;
3806 set_intercept(svm, INTERCEPT_IRET);
3808 svm->vcpu.arch.hflags &= ~HF_NMI_MASK;
3809 clr_intercept(svm, INTERCEPT_IRET);
3813 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu)
3815 struct vcpu_svm *svm = to_svm(vcpu);
3816 struct vmcb *vmcb = svm->vmcb;
3819 if (!gif_set(svm) ||
3820 (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK))
3823 ret = !!(kvm_get_rflags(vcpu) & X86_EFLAGS_IF);
3825 if (is_guest_mode(vcpu))
3826 return ret && !(svm->vcpu.arch.hflags & HF_VINTR_MASK);
3831 static void enable_irq_window(struct kvm_vcpu *vcpu)
3833 struct vcpu_svm *svm = to_svm(vcpu);
3836 * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
3837 * 1, because that's a separate STGI/VMRUN intercept. The next time we
3838 * get that intercept, this function will be called again though and
3839 * we'll get the vintr intercept.
3841 if (gif_set(svm) && nested_svm_intr(svm)) {
3843 svm_inject_irq(svm, 0x0);
3847 static void enable_nmi_window(struct kvm_vcpu *vcpu)
3849 struct vcpu_svm *svm = to_svm(vcpu);
3851 if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
3853 return; /* IRET will cause a vm exit */
3856 * Something prevents NMI from been injected. Single step over possible
3857 * problem (IRET or exception injection or interrupt shadow)
3859 svm->nmi_singlestep = true;
3860 svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
3861 update_db_bp_intercept(vcpu);
3864 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
3869 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
3871 struct vcpu_svm *svm = to_svm(vcpu);
3873 if (static_cpu_has(X86_FEATURE_FLUSHBYASID))
3874 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
3876 svm->asid_generation--;
3879 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
3883 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
3885 struct vcpu_svm *svm = to_svm(vcpu);
3887 if (is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK))
3890 if (!is_cr_intercept(svm, INTERCEPT_CR8_WRITE)) {
3891 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
3892 kvm_set_cr8(vcpu, cr8);
3896 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
3898 struct vcpu_svm *svm = to_svm(vcpu);
3901 if (is_guest_mode(vcpu) && (vcpu->arch.hflags & HF_VINTR_MASK))
3904 cr8 = kvm_get_cr8(vcpu);
3905 svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
3906 svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
3909 static void svm_complete_interrupts(struct vcpu_svm *svm)
3913 u32 exitintinfo = svm->vmcb->control.exit_int_info;
3914 unsigned int3_injected = svm->int3_injected;
3916 svm->int3_injected = 0;
3919 * If we've made progress since setting HF_IRET_MASK, we've
3920 * executed an IRET and can allow NMI injection.
3922 if ((svm->vcpu.arch.hflags & HF_IRET_MASK)
3923 && kvm_rip_read(&svm->vcpu) != svm->nmi_iret_rip) {
3924 svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
3925 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3928 svm->vcpu.arch.nmi_injected = false;
3929 kvm_clear_exception_queue(&svm->vcpu);
3930 kvm_clear_interrupt_queue(&svm->vcpu);
3932 if (!(exitintinfo & SVM_EXITINTINFO_VALID))
3935 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3937 vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
3938 type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
3941 case SVM_EXITINTINFO_TYPE_NMI:
3942 svm->vcpu.arch.nmi_injected = true;
3944 case SVM_EXITINTINFO_TYPE_EXEPT:
3946 * In case of software exceptions, do not reinject the vector,
3947 * but re-execute the instruction instead. Rewind RIP first
3948 * if we emulated INT3 before.
3950 if (kvm_exception_is_soft(vector)) {
3951 if (vector == BP_VECTOR && int3_injected &&
3952 kvm_is_linear_rip(&svm->vcpu, svm->int3_rip))
3953 kvm_rip_write(&svm->vcpu,
3954 kvm_rip_read(&svm->vcpu) -
3958 if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
3959 u32 err = svm->vmcb->control.exit_int_info_err;
3960 kvm_requeue_exception_e(&svm->vcpu, vector, err);
3963 kvm_requeue_exception(&svm->vcpu, vector);
3965 case SVM_EXITINTINFO_TYPE_INTR:
3966 kvm_queue_interrupt(&svm->vcpu, vector, false);
3973 static void svm_cancel_injection(struct kvm_vcpu *vcpu)
3975 struct vcpu_svm *svm = to_svm(vcpu);
3976 struct vmcb_control_area *control = &svm->vmcb->control;
3978 control->exit_int_info = control->event_inj;
3979 control->exit_int_info_err = control->event_inj_err;
3980 control->event_inj = 0;
3981 svm_complete_interrupts(svm);
3984 static void svm_vcpu_run(struct kvm_vcpu *vcpu)
3986 struct vcpu_svm *svm = to_svm(vcpu);
3988 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
3989 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
3990 svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
3993 * A vmexit emulation is required before the vcpu can be executed
3996 if (unlikely(svm->nested.exit_required))
4001 sync_lapic_to_cr8(vcpu);
4003 svm->vmcb->save.cr2 = vcpu->arch.cr2;
4010 "push %%" _ASM_BP "; \n\t"
4011 "mov %c[rbx](%[svm]), %%" _ASM_BX " \n\t"
4012 "mov %c[rcx](%[svm]), %%" _ASM_CX " \n\t"
4013 "mov %c[rdx](%[svm]), %%" _ASM_DX " \n\t"
4014 "mov %c[rsi](%[svm]), %%" _ASM_SI " \n\t"
4015 "mov %c[rdi](%[svm]), %%" _ASM_DI " \n\t"
4016 "mov %c[rbp](%[svm]), %%" _ASM_BP " \n\t"
4017 #ifdef CONFIG_X86_64
4018 "mov %c[r8](%[svm]), %%r8 \n\t"
4019 "mov %c[r9](%[svm]), %%r9 \n\t"
4020 "mov %c[r10](%[svm]), %%r10 \n\t"
4021 "mov %c[r11](%[svm]), %%r11 \n\t"
4022 "mov %c[r12](%[svm]), %%r12 \n\t"
4023 "mov %c[r13](%[svm]), %%r13 \n\t"
4024 "mov %c[r14](%[svm]), %%r14 \n\t"
4025 "mov %c[r15](%[svm]), %%r15 \n\t"
4028 /* Enter guest mode */
4029 "push %%" _ASM_AX " \n\t"
4030 "mov %c[vmcb](%[svm]), %%" _ASM_AX " \n\t"
4031 __ex(SVM_VMLOAD) "\n\t"
4032 __ex(SVM_VMRUN) "\n\t"
4033 __ex(SVM_VMSAVE) "\n\t"
4034 "pop %%" _ASM_AX " \n\t"
4036 /* Save guest registers, load host registers */
4037 "mov %%" _ASM_BX ", %c[rbx](%[svm]) \n\t"
4038 "mov %%" _ASM_CX ", %c[rcx](%[svm]) \n\t"
4039 "mov %%" _ASM_DX ", %c[rdx](%[svm]) \n\t"
4040 "mov %%" _ASM_SI ", %c[rsi](%[svm]) \n\t"
4041 "mov %%" _ASM_DI ", %c[rdi](%[svm]) \n\t"
4042 "mov %%" _ASM_BP ", %c[rbp](%[svm]) \n\t"
4043 #ifdef CONFIG_X86_64
4044 "mov %%r8, %c[r8](%[svm]) \n\t"
4045 "mov %%r9, %c[r9](%[svm]) \n\t"
4046 "mov %%r10, %c[r10](%[svm]) \n\t"
4047 "mov %%r11, %c[r11](%[svm]) \n\t"
4048 "mov %%r12, %c[r12](%[svm]) \n\t"
4049 "mov %%r13, %c[r13](%[svm]) \n\t"
4050 "mov %%r14, %c[r14](%[svm]) \n\t"
4051 "mov %%r15, %c[r15](%[svm]) \n\t"
4056 [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
4057 [rbx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBX])),
4058 [rcx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RCX])),
4059 [rdx]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDX])),
4060 [rsi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RSI])),
4061 [rdi]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RDI])),
4062 [rbp]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_RBP]))
4063 #ifdef CONFIG_X86_64
4064 , [r8]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R8])),
4065 [r9]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R9])),
4066 [r10]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R10])),
4067 [r11]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R11])),
4068 [r12]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R12])),
4069 [r13]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R13])),
4070 [r14]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R14])),
4071 [r15]"i"(offsetof(struct vcpu_svm, vcpu.arch.regs[VCPU_REGS_R15]))
4074 #ifdef CONFIG_X86_64
4075 , "rbx", "rcx", "rdx", "rsi", "rdi"
4076 , "r8", "r9", "r10", "r11" , "r12", "r13", "r14", "r15"
4078 , "ebx", "ecx", "edx", "esi", "edi"
4082 #ifdef CONFIG_X86_64
4083 wrmsrl(MSR_GS_BASE, svm->host.gs_base);
4085 loadsegment(fs, svm->host.fs);
4086 #ifndef CONFIG_X86_32_LAZY_GS
4087 loadsegment(gs, svm->host.gs);
4093 local_irq_disable();
4095 vcpu->arch.cr2 = svm->vmcb->save.cr2;
4096 vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
4097 vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
4098 vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
4100 trace_kvm_exit(svm->vmcb->control.exit_code, vcpu, KVM_ISA_SVM);
4102 if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
4103 kvm_before_handle_nmi(&svm->vcpu);
4107 /* Any pending NMI will happen here */
4109 if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
4110 kvm_after_handle_nmi(&svm->vcpu);
4112 sync_cr8_to_lapic(vcpu);
4116 svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
4118 /* if exit due to PF check for async PF */
4119 if (svm->vmcb->control.exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR)
4120 svm->apf_reason = kvm_read_and_reset_pf_reason();
4123 vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
4124 vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
4128 * We need to handle MC intercepts here before the vcpu has a chance to
4129 * change the physical cpu
4131 if (unlikely(svm->vmcb->control.exit_code ==
4132 SVM_EXIT_EXCP_BASE + MC_VECTOR))
4133 svm_handle_mce(svm);
4135 mark_all_clean(svm->vmcb);
4138 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
4140 struct vcpu_svm *svm = to_svm(vcpu);
4142 svm->vmcb->save.cr3 = root;
4143 mark_dirty(svm->vmcb, VMCB_CR);
4144 svm_flush_tlb(vcpu);
4147 static void set_tdp_cr3(struct kvm_vcpu *vcpu, unsigned long root)
4149 struct vcpu_svm *svm = to_svm(vcpu);
4151 svm->vmcb->control.nested_cr3 = root;
4152 mark_dirty(svm->vmcb, VMCB_NPT);
4154 /* Also sync guest cr3 here in case we live migrate */
4155 svm->vmcb->save.cr3 = kvm_read_cr3(vcpu);
4156 mark_dirty(svm->vmcb, VMCB_CR);
4158 svm_flush_tlb(vcpu);
4161 static int is_disabled(void)
4165 rdmsrl(MSR_VM_CR, vm_cr);
4166 if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
4173 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4176 * Patch in the VMMCALL instruction:
4178 hypercall[0] = 0x0f;
4179 hypercall[1] = 0x01;
4180 hypercall[2] = 0xd9;
4183 static void svm_check_processor_compat(void *rtn)
4188 static bool svm_cpu_has_accelerated_tpr(void)
4193 static bool svm_has_high_real_mode_segbase(void)
4198 static void svm_cpuid_update(struct kvm_vcpu *vcpu)
4202 static void svm_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
4207 entry->ecx |= (1 << 2); /* Set SVM bit */
4210 entry->eax = 1; /* SVM revision 1 */
4211 entry->ebx = 8; /* Lets support 8 ASIDs in case we add proper
4212 ASID emulation to nested SVM */
4213 entry->ecx = 0; /* Reserved */
4214 entry->edx = 0; /* Per default do not support any
4215 additional features */
4217 /* Support next_rip if host supports it */
4218 if (boot_cpu_has(X86_FEATURE_NRIPS))
4219 entry->edx |= SVM_FEATURE_NRIP;
4221 /* Support NPT for the guest if enabled */
4223 entry->edx |= SVM_FEATURE_NPT;
4229 static int svm_get_lpage_level(void)
4231 return PT_PDPE_LEVEL;
4234 static bool svm_rdtscp_supported(void)
4239 static bool svm_invpcid_supported(void)
4244 static bool svm_mpx_supported(void)
4249 static bool svm_xsaves_supported(void)
4254 static bool svm_has_wbinvd_exit(void)
4259 static void svm_fpu_deactivate(struct kvm_vcpu *vcpu)
4261 struct vcpu_svm *svm = to_svm(vcpu);
4263 set_exception_intercept(svm, NM_VECTOR);
4264 update_cr0_intercept(svm);
4267 #define PRE_EX(exit) { .exit_code = (exit), \
4268 .stage = X86_ICPT_PRE_EXCEPT, }
4269 #define POST_EX(exit) { .exit_code = (exit), \
4270 .stage = X86_ICPT_POST_EXCEPT, }
4271 #define POST_MEM(exit) { .exit_code = (exit), \
4272 .stage = X86_ICPT_POST_MEMACCESS, }
4274 static const struct __x86_intercept {
4276 enum x86_intercept_stage stage;
4277 } x86_intercept_map[] = {
4278 [x86_intercept_cr_read] = POST_EX(SVM_EXIT_READ_CR0),
4279 [x86_intercept_cr_write] = POST_EX(SVM_EXIT_WRITE_CR0),
4280 [x86_intercept_clts] = POST_EX(SVM_EXIT_WRITE_CR0),
4281 [x86_intercept_lmsw] = POST_EX(SVM_EXIT_WRITE_CR0),
4282 [x86_intercept_smsw] = POST_EX(SVM_EXIT_READ_CR0),
4283 [x86_intercept_dr_read] = POST_EX(SVM_EXIT_READ_DR0),
4284 [x86_intercept_dr_write] = POST_EX(SVM_EXIT_WRITE_DR0),
4285 [x86_intercept_sldt] = POST_EX(SVM_EXIT_LDTR_READ),
4286 [x86_intercept_str] = POST_EX(SVM_EXIT_TR_READ),
4287 [x86_intercept_lldt] = POST_EX(SVM_EXIT_LDTR_WRITE),
4288 [x86_intercept_ltr] = POST_EX(SVM_EXIT_TR_WRITE),
4289 [x86_intercept_sgdt] = POST_EX(SVM_EXIT_GDTR_READ),
4290 [x86_intercept_sidt] = POST_EX(SVM_EXIT_IDTR_READ),
4291 [x86_intercept_lgdt] = POST_EX(SVM_EXIT_GDTR_WRITE),
4292 [x86_intercept_lidt] = POST_EX(SVM_EXIT_IDTR_WRITE),
4293 [x86_intercept_vmrun] = POST_EX(SVM_EXIT_VMRUN),
4294 [x86_intercept_vmmcall] = POST_EX(SVM_EXIT_VMMCALL),
4295 [x86_intercept_vmload] = POST_EX(SVM_EXIT_VMLOAD),
4296 [x86_intercept_vmsave] = POST_EX(SVM_EXIT_VMSAVE),
4297 [x86_intercept_stgi] = POST_EX(SVM_EXIT_STGI),
4298 [x86_intercept_clgi] = POST_EX(SVM_EXIT_CLGI),
4299 [x86_intercept_skinit] = POST_EX(SVM_EXIT_SKINIT),
4300 [x86_intercept_invlpga] = POST_EX(SVM_EXIT_INVLPGA),
4301 [x86_intercept_rdtscp] = POST_EX(SVM_EXIT_RDTSCP),
4302 [x86_intercept_monitor] = POST_MEM(SVM_EXIT_MONITOR),
4303 [x86_intercept_mwait] = POST_EX(SVM_EXIT_MWAIT),
4304 [x86_intercept_invlpg] = POST_EX(SVM_EXIT_INVLPG),
4305 [x86_intercept_invd] = POST_EX(SVM_EXIT_INVD),
4306 [x86_intercept_wbinvd] = POST_EX(SVM_EXIT_WBINVD),
4307 [x86_intercept_wrmsr] = POST_EX(SVM_EXIT_MSR),
4308 [x86_intercept_rdtsc] = POST_EX(SVM_EXIT_RDTSC),
4309 [x86_intercept_rdmsr] = POST_EX(SVM_EXIT_MSR),
4310 [x86_intercept_rdpmc] = POST_EX(SVM_EXIT_RDPMC),
4311 [x86_intercept_cpuid] = PRE_EX(SVM_EXIT_CPUID),
4312 [x86_intercept_rsm] = PRE_EX(SVM_EXIT_RSM),
4313 [x86_intercept_pause] = PRE_EX(SVM_EXIT_PAUSE),
4314 [x86_intercept_pushf] = PRE_EX(SVM_EXIT_PUSHF),
4315 [x86_intercept_popf] = PRE_EX(SVM_EXIT_POPF),
4316 [x86_intercept_intn] = PRE_EX(SVM_EXIT_SWINT),
4317 [x86_intercept_iret] = PRE_EX(SVM_EXIT_IRET),
4318 [x86_intercept_icebp] = PRE_EX(SVM_EXIT_ICEBP),
4319 [x86_intercept_hlt] = POST_EX(SVM_EXIT_HLT),
4320 [x86_intercept_in] = POST_EX(SVM_EXIT_IOIO),
4321 [x86_intercept_ins] = POST_EX(SVM_EXIT_IOIO),
4322 [x86_intercept_out] = POST_EX(SVM_EXIT_IOIO),
4323 [x86_intercept_outs] = POST_EX(SVM_EXIT_IOIO),
4330 static int svm_check_intercept(struct kvm_vcpu *vcpu,
4331 struct x86_instruction_info *info,
4332 enum x86_intercept_stage stage)
4334 struct vcpu_svm *svm = to_svm(vcpu);
4335 int vmexit, ret = X86EMUL_CONTINUE;
4336 struct __x86_intercept icpt_info;
4337 struct vmcb *vmcb = svm->vmcb;
4339 if (info->intercept >= ARRAY_SIZE(x86_intercept_map))
4342 icpt_info = x86_intercept_map[info->intercept];
4344 if (stage != icpt_info.stage)
4347 switch (icpt_info.exit_code) {
4348 case SVM_EXIT_READ_CR0:
4349 if (info->intercept == x86_intercept_cr_read)
4350 icpt_info.exit_code += info->modrm_reg;
4352 case SVM_EXIT_WRITE_CR0: {
4353 unsigned long cr0, val;
4356 if (info->intercept == x86_intercept_cr_write)
4357 icpt_info.exit_code += info->modrm_reg;
4359 if (icpt_info.exit_code != SVM_EXIT_WRITE_CR0 ||
4360 info->intercept == x86_intercept_clts)
4363 intercept = svm->nested.intercept;
4365 if (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0)))
4368 cr0 = vcpu->arch.cr0 & ~SVM_CR0_SELECTIVE_MASK;
4369 val = info->src_val & ~SVM_CR0_SELECTIVE_MASK;
4371 if (info->intercept == x86_intercept_lmsw) {
4374 /* lmsw can't clear PE - catch this here */
4375 if (cr0 & X86_CR0_PE)
4380 icpt_info.exit_code = SVM_EXIT_CR0_SEL_WRITE;
4384 case SVM_EXIT_READ_DR0:
4385 case SVM_EXIT_WRITE_DR0:
4386 icpt_info.exit_code += info->modrm_reg;
4389 if (info->intercept == x86_intercept_wrmsr)
4390 vmcb->control.exit_info_1 = 1;
4392 vmcb->control.exit_info_1 = 0;
4394 case SVM_EXIT_PAUSE:
4396 * We get this for NOP only, but pause
4397 * is rep not, check this here
4399 if (info->rep_prefix != REPE_PREFIX)
4401 case SVM_EXIT_IOIO: {
4405 if (info->intercept == x86_intercept_in ||
4406 info->intercept == x86_intercept_ins) {
4407 exit_info = ((info->src_val & 0xffff) << 16) |
4409 bytes = info->dst_bytes;
4411 exit_info = (info->dst_val & 0xffff) << 16;
4412 bytes = info->src_bytes;
4415 if (info->intercept == x86_intercept_outs ||
4416 info->intercept == x86_intercept_ins)
4417 exit_info |= SVM_IOIO_STR_MASK;
4419 if (info->rep_prefix)
4420 exit_info |= SVM_IOIO_REP_MASK;
4422 bytes = min(bytes, 4u);
4424 exit_info |= bytes << SVM_IOIO_SIZE_SHIFT;
4426 exit_info |= (u32)info->ad_bytes << (SVM_IOIO_ASIZE_SHIFT - 1);
4428 vmcb->control.exit_info_1 = exit_info;
4429 vmcb->control.exit_info_2 = info->next_rip;
4437 /* TODO: Advertise NRIPS to guest hypervisor unconditionally */
4438 if (static_cpu_has(X86_FEATURE_NRIPS))
4439 vmcb->control.next_rip = info->next_rip;
4440 vmcb->control.exit_code = icpt_info.exit_code;
4441 vmexit = nested_svm_exit_handled(svm);
4443 ret = (vmexit == NESTED_EXIT_DONE) ? X86EMUL_INTERCEPTED
4450 static void svm_handle_external_intr(struct kvm_vcpu *vcpu)
4455 static void svm_sched_in(struct kvm_vcpu *vcpu, int cpu)
4459 static struct kvm_x86_ops svm_x86_ops = {
4460 .cpu_has_kvm_support = has_svm,
4461 .disabled_by_bios = is_disabled,
4462 .hardware_setup = svm_hardware_setup,
4463 .hardware_unsetup = svm_hardware_unsetup,
4464 .check_processor_compatibility = svm_check_processor_compat,
4465 .hardware_enable = svm_hardware_enable,
4466 .hardware_disable = svm_hardware_disable,
4467 .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
4468 .cpu_has_high_real_mode_segbase = svm_has_high_real_mode_segbase,
4470 .vcpu_create = svm_create_vcpu,
4471 .vcpu_free = svm_free_vcpu,
4472 .vcpu_reset = svm_vcpu_reset,
4474 .prepare_guest_switch = svm_prepare_guest_switch,
4475 .vcpu_load = svm_vcpu_load,
4476 .vcpu_put = svm_vcpu_put,
4478 .update_db_bp_intercept = update_db_bp_intercept,
4479 .get_msr = svm_get_msr,
4480 .set_msr = svm_set_msr,
4481 .get_segment_base = svm_get_segment_base,
4482 .get_segment = svm_get_segment,
4483 .set_segment = svm_set_segment,
4484 .get_cpl = svm_get_cpl,
4485 .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
4486 .decache_cr0_guest_bits = svm_decache_cr0_guest_bits,
4487 .decache_cr3 = svm_decache_cr3,
4488 .decache_cr4_guest_bits = svm_decache_cr4_guest_bits,
4489 .set_cr0 = svm_set_cr0,
4490 .set_cr3 = svm_set_cr3,
4491 .set_cr4 = svm_set_cr4,
4492 .set_efer = svm_set_efer,
4493 .get_idt = svm_get_idt,
4494 .set_idt = svm_set_idt,
4495 .get_gdt = svm_get_gdt,
4496 .set_gdt = svm_set_gdt,
4497 .get_dr6 = svm_get_dr6,
4498 .set_dr6 = svm_set_dr6,
4499 .set_dr7 = svm_set_dr7,
4500 .sync_dirty_debug_regs = svm_sync_dirty_debug_regs,
4501 .cache_reg = svm_cache_reg,
4502 .get_rflags = svm_get_rflags,
4503 .set_rflags = svm_set_rflags,
4504 .fpu_activate = svm_fpu_activate,
4505 .fpu_deactivate = svm_fpu_deactivate,
4507 .tlb_flush = svm_flush_tlb,
4509 .run = svm_vcpu_run,
4510 .handle_exit = handle_exit,
4511 .skip_emulated_instruction = skip_emulated_instruction,
4512 .set_interrupt_shadow = svm_set_interrupt_shadow,
4513 .get_interrupt_shadow = svm_get_interrupt_shadow,
4514 .patch_hypercall = svm_patch_hypercall,
4515 .set_irq = svm_set_irq,
4516 .set_nmi = svm_inject_nmi,
4517 .queue_exception = svm_queue_exception,
4518 .cancel_injection = svm_cancel_injection,
4519 .interrupt_allowed = svm_interrupt_allowed,
4520 .nmi_allowed = svm_nmi_allowed,
4521 .get_nmi_mask = svm_get_nmi_mask,
4522 .set_nmi_mask = svm_set_nmi_mask,
4523 .enable_nmi_window = enable_nmi_window,
4524 .enable_irq_window = enable_irq_window,
4525 .update_cr8_intercept = update_cr8_intercept,
4526 .set_virtual_x2apic_mode = svm_set_virtual_x2apic_mode,
4527 .vm_has_apicv = svm_vm_has_apicv,
4528 .load_eoi_exitmap = svm_load_eoi_exitmap,
4529 .sync_pir_to_irr = svm_sync_pir_to_irr,
4531 .set_tss_addr = svm_set_tss_addr,
4532 .get_tdp_level = get_npt_level,
4533 .get_mt_mask = svm_get_mt_mask,
4535 .get_exit_info = svm_get_exit_info,
4537 .get_lpage_level = svm_get_lpage_level,
4539 .cpuid_update = svm_cpuid_update,
4541 .rdtscp_supported = svm_rdtscp_supported,
4542 .invpcid_supported = svm_invpcid_supported,
4543 .mpx_supported = svm_mpx_supported,
4544 .xsaves_supported = svm_xsaves_supported,
4546 .set_supported_cpuid = svm_set_supported_cpuid,
4548 .has_wbinvd_exit = svm_has_wbinvd_exit,
4550 .set_tsc_khz = svm_set_tsc_khz,
4551 .read_tsc_offset = svm_read_tsc_offset,
4552 .write_tsc_offset = svm_write_tsc_offset,
4553 .adjust_tsc_offset = svm_adjust_tsc_offset,
4554 .compute_tsc_offset = svm_compute_tsc_offset,
4555 .read_l1_tsc = svm_read_l1_tsc,
4557 .set_tdp_cr3 = set_tdp_cr3,
4559 .check_intercept = svm_check_intercept,
4560 .handle_external_intr = svm_handle_external_intr,
4562 .sched_in = svm_sched_in,
4564 .pmu_ops = &amd_pmu_ops,
4567 static int __init svm_init(void)
4569 return kvm_init(&svm_x86_ops, sizeof(struct vcpu_svm),
4570 __alignof__(struct vcpu_svm), THIS_MODULE);
4573 static void __exit svm_exit(void)
4578 module_init(svm_init)
4579 module_exit(svm_exit)