1 #define pr_fmt(fmt) "SVM: " fmt
3 #include <linux/kvm_host.h>
7 #include "kvm_cache_regs.h"
12 #include <linux/module.h>
13 #include <linux/mod_devicetable.h>
14 #include <linux/kernel.h>
15 #include <linux/vmalloc.h>
16 #include <linux/highmem.h>
17 #include <linux/amd-iommu.h>
18 #include <linux/sched.h>
19 #include <linux/trace_events.h>
20 #include <linux/slab.h>
21 #include <linux/hashtable.h>
22 #include <linux/objtool.h>
23 #include <linux/psp-sev.h>
24 #include <linux/file.h>
25 #include <linux/pagemap.h>
26 #include <linux/swap.h>
27 #include <linux/rwsem.h>
30 #include <asm/perf_event.h>
31 #include <asm/tlbflush.h>
33 #include <asm/debugreg.h>
34 #include <asm/kvm_para.h>
35 #include <asm/irq_remapping.h>
37 #include <asm/spec-ctrl.h>
38 #include <asm/cpu_device_id.h>
40 #include <asm/virtext.h>
45 #define __ex(x) __kvm_handle_fault_on_reboot(x)
47 MODULE_AUTHOR("Qumranet");
48 MODULE_LICENSE("GPL");
51 static const struct x86_cpu_id svm_cpu_id[] = {
52 X86_MATCH_FEATURE(X86_FEATURE_SVM, NULL),
55 MODULE_DEVICE_TABLE(x86cpu, svm_cpu_id);
58 #define IOPM_ALLOC_ORDER 2
59 #define MSRPM_ALLOC_ORDER 1
61 #define SEG_TYPE_LDT 2
62 #define SEG_TYPE_BUSY_TSS16 3
64 #define SVM_FEATURE_LBRV (1 << 1)
65 #define SVM_FEATURE_SVML (1 << 2)
66 #define SVM_FEATURE_TSC_RATE (1 << 4)
67 #define SVM_FEATURE_VMCB_CLEAN (1 << 5)
68 #define SVM_FEATURE_FLUSH_ASID (1 << 6)
69 #define SVM_FEATURE_DECODE_ASSIST (1 << 7)
70 #define SVM_FEATURE_PAUSE_FILTER (1 << 10)
72 #define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
74 #define TSC_RATIO_RSVD 0xffffff0000000000ULL
75 #define TSC_RATIO_MIN 0x0000000000000001ULL
76 #define TSC_RATIO_MAX 0x000000ffffffffffULL
78 static bool erratum_383_found __read_mostly;
80 u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
83 * Set osvw_len to higher value when updated Revision Guides
84 * are published and we know what the new status bits are
86 static uint64_t osvw_len = 4, osvw_status;
88 static DEFINE_PER_CPU(u64, current_tsc_ratio);
89 #define TSC_RATIO_DEFAULT 0x0100000000ULL
91 static const struct svm_direct_access_msrs {
92 u32 index; /* Index of the MSR */
93 bool always; /* True if intercept is always on */
94 } direct_access_msrs[MAX_DIRECT_ACCESS_MSRS] = {
95 { .index = MSR_STAR, .always = true },
96 { .index = MSR_IA32_SYSENTER_CS, .always = true },
98 { .index = MSR_GS_BASE, .always = true },
99 { .index = MSR_FS_BASE, .always = true },
100 { .index = MSR_KERNEL_GS_BASE, .always = true },
101 { .index = MSR_LSTAR, .always = true },
102 { .index = MSR_CSTAR, .always = true },
103 { .index = MSR_SYSCALL_MASK, .always = true },
105 { .index = MSR_IA32_SPEC_CTRL, .always = false },
106 { .index = MSR_IA32_PRED_CMD, .always = false },
107 { .index = MSR_IA32_LASTBRANCHFROMIP, .always = false },
108 { .index = MSR_IA32_LASTBRANCHTOIP, .always = false },
109 { .index = MSR_IA32_LASTINTFROMIP, .always = false },
110 { .index = MSR_IA32_LASTINTTOIP, .always = false },
111 { .index = MSR_INVALID, .always = false },
114 /* enable NPT for AMD64 and X86 with PAE */
115 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
116 bool npt_enabled = true;
122 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
123 * pause_filter_count: On processors that support Pause filtering(indicated
124 * by CPUID Fn8000_000A_EDX), the VMCB provides a 16 bit pause filter
125 * count value. On VMRUN this value is loaded into an internal counter.
126 * Each time a pause instruction is executed, this counter is decremented
127 * until it reaches zero at which time a #VMEXIT is generated if pause
128 * intercept is enabled. Refer to AMD APM Vol 2 Section 15.14.4 Pause
129 * Intercept Filtering for more details.
130 * This also indicate if ple logic enabled.
132 * pause_filter_thresh: In addition, some processor families support advanced
133 * pause filtering (indicated by CPUID Fn8000_000A_EDX) upper bound on
134 * the amount of time a guest is allowed to execute in a pause loop.
135 * In this mode, a 16-bit pause filter threshold field is added in the
136 * VMCB. The threshold value is a cycle count that is used to reset the
137 * pause counter. As with simple pause filtering, VMRUN loads the pause
138 * count value from VMCB into an internal counter. Then, on each pause
139 * instruction the hardware checks the elapsed number of cycles since
140 * the most recent pause instruction against the pause filter threshold.
141 * If the elapsed cycle count is greater than the pause filter threshold,
142 * then the internal pause count is reloaded from the VMCB and execution
143 * continues. If the elapsed cycle count is less than the pause filter
144 * threshold, then the internal pause count is decremented. If the count
145 * value is less than zero and PAUSE intercept is enabled, a #VMEXIT is
146 * triggered. If advanced pause filtering is supported and pause filter
147 * threshold field is set to zero, the filter will operate in the simpler,
151 static unsigned short pause_filter_thresh = KVM_DEFAULT_PLE_GAP;
152 module_param(pause_filter_thresh, ushort, 0444);
154 static unsigned short pause_filter_count = KVM_SVM_DEFAULT_PLE_WINDOW;
155 module_param(pause_filter_count, ushort, 0444);
157 /* Default doubles per-vcpu window every exit. */
158 static unsigned short pause_filter_count_grow = KVM_DEFAULT_PLE_WINDOW_GROW;
159 module_param(pause_filter_count_grow, ushort, 0444);
161 /* Default resets per-vcpu window every exit to pause_filter_count. */
162 static unsigned short pause_filter_count_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK;
163 module_param(pause_filter_count_shrink, ushort, 0444);
165 /* Default is to compute the maximum so we can never overflow. */
166 static unsigned short pause_filter_count_max = KVM_SVM_DEFAULT_PLE_WINDOW_MAX;
167 module_param(pause_filter_count_max, ushort, 0444);
169 /* allow nested paging (virtualized MMU) for all guests */
170 static int npt = true;
171 module_param(npt, int, S_IRUGO);
173 /* allow nested virtualization in KVM/SVM */
174 static int nested = true;
175 module_param(nested, int, S_IRUGO);
177 /* enable/disable Next RIP Save */
178 static int nrips = true;
179 module_param(nrips, int, 0444);
181 /* enable/disable Virtual VMLOAD VMSAVE */
182 static int vls = true;
183 module_param(vls, int, 0444);
185 /* enable/disable Virtual GIF */
186 static int vgif = true;
187 module_param(vgif, int, 0444);
189 /* enable/disable SEV support */
190 static int sev = IS_ENABLED(CONFIG_AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT);
191 module_param(sev, int, 0444);
193 static bool __read_mostly dump_invalid_vmcb = 0;
194 module_param(dump_invalid_vmcb, bool, 0644);
196 static u8 rsm_ins_bytes[] = "\x0f\xaa";
198 static void svm_complete_interrupts(struct vcpu_svm *svm);
200 static unsigned long iopm_base;
202 struct kvm_ldttss_desc {
205 unsigned base1:8, type:5, dpl:2, p:1;
206 unsigned limit1:4, zero0:3, g:1, base2:8;
209 } __attribute__((packed));
211 DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
213 static const u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
215 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
216 #define MSRS_RANGE_SIZE 2048
217 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
219 u32 svm_msrpm_offset(u32 msr)
224 for (i = 0; i < NUM_MSR_MAPS; i++) {
225 if (msr < msrpm_ranges[i] ||
226 msr >= msrpm_ranges[i] + MSRS_IN_RANGE)
229 offset = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */
230 offset += (i * MSRS_RANGE_SIZE); /* add range offset */
232 /* Now we have the u8 offset - but need the u32 offset */
236 /* MSR not in any range */
240 #define MAX_INST_SIZE 15
242 static inline void clgi(void)
244 asm volatile (__ex("clgi"));
247 static inline void stgi(void)
249 asm volatile (__ex("stgi"));
252 static inline void invlpga(unsigned long addr, u32 asid)
254 asm volatile (__ex("invlpga %1, %0") : : "c"(asid), "a"(addr));
257 static int get_max_npt_level(void)
260 return PT64_ROOT_4LEVEL;
262 return PT32E_ROOT_LEVEL;
266 int svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
268 struct vcpu_svm *svm = to_svm(vcpu);
269 u64 old_efer = vcpu->arch.efer;
270 vcpu->arch.efer = efer;
273 /* Shadow paging assumes NX to be available. */
276 if (!(efer & EFER_LMA))
280 if ((old_efer & EFER_SVME) != (efer & EFER_SVME)) {
281 if (!(efer & EFER_SVME)) {
282 svm_leave_nested(svm);
283 svm_set_gif(svm, true);
286 * Free the nested guest state, unless we are in SMM.
287 * In this case we will return to the nested guest
288 * as soon as we leave SMM.
290 if (!is_smm(&svm->vcpu))
291 svm_free_nested(svm);
294 int ret = svm_allocate_nested(svm);
297 vcpu->arch.efer = old_efer;
303 svm->vmcb->save.efer = efer | EFER_SVME;
304 vmcb_mark_dirty(svm->vmcb, VMCB_CR);
308 static int is_external_interrupt(u32 info)
310 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
311 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
314 static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu)
316 struct vcpu_svm *svm = to_svm(vcpu);
319 if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
320 ret = KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS;
324 static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
326 struct vcpu_svm *svm = to_svm(vcpu);
329 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
331 svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
335 static int skip_emulated_instruction(struct kvm_vcpu *vcpu)
337 struct vcpu_svm *svm = to_svm(vcpu);
339 if (nrips && svm->vmcb->control.next_rip != 0) {
340 WARN_ON_ONCE(!static_cpu_has(X86_FEATURE_NRIPS));
341 svm->next_rip = svm->vmcb->control.next_rip;
344 if (!svm->next_rip) {
345 if (!kvm_emulate_instruction(vcpu, EMULTYPE_SKIP))
348 kvm_rip_write(vcpu, svm->next_rip);
350 svm_set_interrupt_shadow(vcpu, 0);
355 static void svm_queue_exception(struct kvm_vcpu *vcpu)
357 struct vcpu_svm *svm = to_svm(vcpu);
358 unsigned nr = vcpu->arch.exception.nr;
359 bool has_error_code = vcpu->arch.exception.has_error_code;
360 u32 error_code = vcpu->arch.exception.error_code;
362 kvm_deliver_exception_payload(&svm->vcpu);
364 if (nr == BP_VECTOR && !nrips) {
365 unsigned long rip, old_rip = kvm_rip_read(&svm->vcpu);
368 * For guest debugging where we have to reinject #BP if some
369 * INT3 is guest-owned:
370 * Emulate nRIP by moving RIP forward. Will fail if injection
371 * raises a fault that is not intercepted. Still better than
372 * failing in all cases.
374 (void)skip_emulated_instruction(&svm->vcpu);
375 rip = kvm_rip_read(&svm->vcpu);
376 svm->int3_rip = rip + svm->vmcb->save.cs.base;
377 svm->int3_injected = rip - old_rip;
380 svm->vmcb->control.event_inj = nr
382 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
383 | SVM_EVTINJ_TYPE_EXEPT;
384 svm->vmcb->control.event_inj_err = error_code;
387 static void svm_init_erratum_383(void)
393 if (!static_cpu_has_bug(X86_BUG_AMD_TLB_MMATCH))
396 /* Use _safe variants to not break nested virtualization */
397 val = native_read_msr_safe(MSR_AMD64_DC_CFG, &err);
403 low = lower_32_bits(val);
404 high = upper_32_bits(val);
406 native_write_msr_safe(MSR_AMD64_DC_CFG, low, high);
408 erratum_383_found = true;
411 static void svm_init_osvw(struct kvm_vcpu *vcpu)
414 * Guests should see errata 400 and 415 as fixed (assuming that
415 * HLT and IO instructions are intercepted).
417 vcpu->arch.osvw.length = (osvw_len >= 3) ? (osvw_len) : 3;
418 vcpu->arch.osvw.status = osvw_status & ~(6ULL);
421 * By increasing VCPU's osvw.length to 3 we are telling the guest that
422 * all osvw.status bits inside that length, including bit 0 (which is
423 * reserved for erratum 298), are valid. However, if host processor's
424 * osvw_len is 0 then osvw_status[0] carries no information. We need to
425 * be conservative here and therefore we tell the guest that erratum 298
426 * is present (because we really don't know).
428 if (osvw_len == 0 && boot_cpu_data.x86 == 0x10)
429 vcpu->arch.osvw.status |= 1;
432 static int has_svm(void)
436 if (!cpu_has_svm(&msg)) {
437 printk(KERN_INFO "has_svm: %s\n", msg);
444 static void svm_hardware_disable(void)
446 /* Make sure we clean up behind us */
447 if (static_cpu_has(X86_FEATURE_TSCRATEMSR))
448 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
452 amd_pmu_disable_virt();
455 static int svm_hardware_enable(void)
458 struct svm_cpu_data *sd;
460 struct desc_struct *gdt;
461 int me = raw_smp_processor_id();
463 rdmsrl(MSR_EFER, efer);
464 if (efer & EFER_SVME)
468 pr_err("%s: err EOPNOTSUPP on %d\n", __func__, me);
471 sd = per_cpu(svm_data, me);
473 pr_err("%s: svm_data is NULL on %d\n", __func__, me);
477 sd->asid_generation = 1;
478 sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
479 sd->next_asid = sd->max_asid + 1;
480 sd->min_asid = max_sev_asid + 1;
482 gdt = get_current_gdt_rw();
483 sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
485 wrmsrl(MSR_EFER, efer | EFER_SVME);
487 wrmsrl(MSR_VM_HSAVE_PA, page_to_pfn(sd->save_area) << PAGE_SHIFT);
489 if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
490 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
491 __this_cpu_write(current_tsc_ratio, TSC_RATIO_DEFAULT);
498 * Note that it is possible to have a system with mixed processor
499 * revisions and therefore different OSVW bits. If bits are not the same
500 * on different processors then choose the worst case (i.e. if erratum
501 * is present on one processor and not on another then assume that the
502 * erratum is present everywhere).
504 if (cpu_has(&boot_cpu_data, X86_FEATURE_OSVW)) {
505 uint64_t len, status = 0;
508 len = native_read_msr_safe(MSR_AMD64_OSVW_ID_LENGTH, &err);
510 status = native_read_msr_safe(MSR_AMD64_OSVW_STATUS,
514 osvw_status = osvw_len = 0;
518 osvw_status |= status;
519 osvw_status &= (1ULL << osvw_len) - 1;
522 osvw_status = osvw_len = 0;
524 svm_init_erratum_383();
526 amd_pmu_enable_virt();
531 static void svm_cpu_uninit(int cpu)
533 struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id());
538 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
539 kfree(sd->sev_vmcbs);
540 __free_page(sd->save_area);
544 static int svm_cpu_init(int cpu)
546 struct svm_cpu_data *sd;
548 sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
552 sd->save_area = alloc_page(GFP_KERNEL);
556 if (svm_sev_enabled()) {
557 sd->sev_vmcbs = kmalloc_array(max_sev_asid + 1,
564 per_cpu(svm_data, cpu) = sd;
569 __free_page(sd->save_area);
576 static int direct_access_msr_slot(u32 msr)
580 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++)
581 if (direct_access_msrs[i].index == msr)
587 static void set_shadow_msr_intercept(struct kvm_vcpu *vcpu, u32 msr, int read,
590 struct vcpu_svm *svm = to_svm(vcpu);
591 int slot = direct_access_msr_slot(msr);
596 /* Set the shadow bitmaps to the desired intercept states */
598 set_bit(slot, svm->shadow_msr_intercept.read);
600 clear_bit(slot, svm->shadow_msr_intercept.read);
603 set_bit(slot, svm->shadow_msr_intercept.write);
605 clear_bit(slot, svm->shadow_msr_intercept.write);
608 static bool valid_msr_intercept(u32 index)
610 return direct_access_msr_slot(index) != -ENOENT;
613 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr)
620 msrpm = is_guest_mode(vcpu) ? to_svm(vcpu)->nested.msrpm:
623 offset = svm_msrpm_offset(msr);
624 bit_write = 2 * (msr & 0x0f) + 1;
627 BUG_ON(offset == MSR_INVALID);
629 return !!test_bit(bit_write, &tmp);
632 static void set_msr_interception_bitmap(struct kvm_vcpu *vcpu, u32 *msrpm,
633 u32 msr, int read, int write)
635 u8 bit_read, bit_write;
640 * If this warning triggers extend the direct_access_msrs list at the
641 * beginning of the file
643 WARN_ON(!valid_msr_intercept(msr));
645 /* Enforce non allowed MSRs to trap */
646 if (read && !kvm_msr_allowed(vcpu, msr, KVM_MSR_FILTER_READ))
649 if (write && !kvm_msr_allowed(vcpu, msr, KVM_MSR_FILTER_WRITE))
652 offset = svm_msrpm_offset(msr);
653 bit_read = 2 * (msr & 0x0f);
654 bit_write = 2 * (msr & 0x0f) + 1;
657 BUG_ON(offset == MSR_INVALID);
659 read ? clear_bit(bit_read, &tmp) : set_bit(bit_read, &tmp);
660 write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp);
665 static void set_msr_interception(struct kvm_vcpu *vcpu, u32 *msrpm, u32 msr,
668 set_shadow_msr_intercept(vcpu, msr, read, write);
669 set_msr_interception_bitmap(vcpu, msrpm, msr, read, write);
672 u32 *svm_vcpu_alloc_msrpm(void)
674 struct page *pages = alloc_pages(GFP_KERNEL_ACCOUNT, MSRPM_ALLOC_ORDER);
680 msrpm = page_address(pages);
681 memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
686 void svm_vcpu_init_msrpm(struct kvm_vcpu *vcpu, u32 *msrpm)
690 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
691 if (!direct_access_msrs[i].always)
693 set_msr_interception(vcpu, msrpm, direct_access_msrs[i].index, 1, 1);
698 void svm_vcpu_free_msrpm(u32 *msrpm)
700 __free_pages(virt_to_page(msrpm), MSRPM_ALLOC_ORDER);
703 static void svm_msr_filter_changed(struct kvm_vcpu *vcpu)
705 struct vcpu_svm *svm = to_svm(vcpu);
709 * Set intercept permissions for all direct access MSRs again. They
710 * will automatically get filtered through the MSR filter, so we are
711 * back in sync after this.
713 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
714 u32 msr = direct_access_msrs[i].index;
715 u32 read = test_bit(i, svm->shadow_msr_intercept.read);
716 u32 write = test_bit(i, svm->shadow_msr_intercept.write);
718 set_msr_interception_bitmap(vcpu, svm->msrpm, msr, read, write);
722 static void add_msr_offset(u32 offset)
726 for (i = 0; i < MSRPM_OFFSETS; ++i) {
728 /* Offset already in list? */
729 if (msrpm_offsets[i] == offset)
732 /* Slot used by another offset? */
733 if (msrpm_offsets[i] != MSR_INVALID)
736 /* Add offset to list */
737 msrpm_offsets[i] = offset;
743 * If this BUG triggers the msrpm_offsets table has an overflow. Just
744 * increase MSRPM_OFFSETS in this case.
749 static void init_msrpm_offsets(void)
753 memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets));
755 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
758 offset = svm_msrpm_offset(direct_access_msrs[i].index);
759 BUG_ON(offset == MSR_INVALID);
761 add_msr_offset(offset);
765 static void svm_enable_lbrv(struct kvm_vcpu *vcpu)
767 struct vcpu_svm *svm = to_svm(vcpu);
769 svm->vmcb->control.virt_ext |= LBR_CTL_ENABLE_MASK;
770 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
771 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
772 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
773 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
776 static void svm_disable_lbrv(struct kvm_vcpu *vcpu)
778 struct vcpu_svm *svm = to_svm(vcpu);
780 svm->vmcb->control.virt_ext &= ~LBR_CTL_ENABLE_MASK;
781 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
782 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
783 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
784 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
787 void disable_nmi_singlestep(struct vcpu_svm *svm)
789 svm->nmi_singlestep = false;
791 if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP)) {
792 /* Clear our flags if they were not set by the guest */
793 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF))
794 svm->vmcb->save.rflags &= ~X86_EFLAGS_TF;
795 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF))
796 svm->vmcb->save.rflags &= ~X86_EFLAGS_RF;
800 static void grow_ple_window(struct kvm_vcpu *vcpu)
802 struct vcpu_svm *svm = to_svm(vcpu);
803 struct vmcb_control_area *control = &svm->vmcb->control;
804 int old = control->pause_filter_count;
806 control->pause_filter_count = __grow_ple_window(old,
808 pause_filter_count_grow,
809 pause_filter_count_max);
811 if (control->pause_filter_count != old) {
812 vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
813 trace_kvm_ple_window_update(vcpu->vcpu_id,
814 control->pause_filter_count, old);
818 static void shrink_ple_window(struct kvm_vcpu *vcpu)
820 struct vcpu_svm *svm = to_svm(vcpu);
821 struct vmcb_control_area *control = &svm->vmcb->control;
822 int old = control->pause_filter_count;
824 control->pause_filter_count =
825 __shrink_ple_window(old,
827 pause_filter_count_shrink,
829 if (control->pause_filter_count != old) {
830 vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
831 trace_kvm_ple_window_update(vcpu->vcpu_id,
832 control->pause_filter_count, old);
837 * The default MMIO mask is a single bit (excluding the present bit),
838 * which could conflict with the memory encryption bit. Check for
839 * memory encryption support and override the default MMIO mask if
840 * memory encryption is enabled.
842 static __init void svm_adjust_mmio_mask(void)
844 unsigned int enc_bit, mask_bit;
847 /* If there is no memory encryption support, use existing mask */
848 if (cpuid_eax(0x80000000) < 0x8000001f)
851 /* If memory encryption is not enabled, use existing mask */
852 rdmsrl(MSR_K8_SYSCFG, msr);
853 if (!(msr & MSR_K8_SYSCFG_MEM_ENCRYPT))
856 enc_bit = cpuid_ebx(0x8000001f) & 0x3f;
857 mask_bit = boot_cpu_data.x86_phys_bits;
859 /* Increment the mask bit if it is the same as the encryption bit */
860 if (enc_bit == mask_bit)
864 * If the mask bit location is below 52, then some bits above the
865 * physical addressing limit will always be reserved, so use the
866 * rsvd_bits() function to generate the mask. This mask, along with
867 * the present bit, will be used to generate a page fault with
870 * If the mask bit location is 52 (or above), then clear the mask.
872 mask = (mask_bit < 52) ? rsvd_bits(mask_bit, 51) | PT_PRESENT_MASK : 0;
874 kvm_mmu_set_mmio_spte_mask(mask, PT_WRITABLE_MASK | PT_USER_MASK);
877 static void svm_hardware_teardown(void)
881 if (svm_sev_enabled())
882 sev_hardware_teardown();
884 for_each_possible_cpu(cpu)
887 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
891 static __init void svm_set_cpu_caps(void)
897 /* CPUID 0x80000001 and 0x8000000A (SVM features) */
899 kvm_cpu_cap_set(X86_FEATURE_SVM);
902 kvm_cpu_cap_set(X86_FEATURE_NRIPS);
905 kvm_cpu_cap_set(X86_FEATURE_NPT);
908 /* CPUID 0x80000008 */
909 if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) ||
910 boot_cpu_has(X86_FEATURE_AMD_SSBD))
911 kvm_cpu_cap_set(X86_FEATURE_VIRT_SSBD);
913 /* Enable INVPCID feature */
914 kvm_cpu_cap_check_and_set(X86_FEATURE_INVPCID);
917 static __init int svm_hardware_setup(void)
920 struct page *iopm_pages;
924 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
929 iopm_va = page_address(iopm_pages);
930 memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
931 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
933 init_msrpm_offsets();
935 supported_xcr0 &= ~(XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
937 if (boot_cpu_has(X86_FEATURE_NX))
938 kvm_enable_efer_bits(EFER_NX);
940 if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
941 kvm_enable_efer_bits(EFER_FFXSR);
943 if (boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
944 kvm_has_tsc_control = true;
945 kvm_max_tsc_scaling_ratio = TSC_RATIO_MAX;
946 kvm_tsc_scaling_ratio_frac_bits = 32;
949 /* Check for pause filtering support */
950 if (!boot_cpu_has(X86_FEATURE_PAUSEFILTER)) {
951 pause_filter_count = 0;
952 pause_filter_thresh = 0;
953 } else if (!boot_cpu_has(X86_FEATURE_PFTHRESHOLD)) {
954 pause_filter_thresh = 0;
958 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
959 kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE);
963 if (boot_cpu_has(X86_FEATURE_SEV) &&
964 IS_ENABLED(CONFIG_KVM_AMD_SEV)) {
965 r = sev_hardware_setup();
973 svm_adjust_mmio_mask();
975 for_each_possible_cpu(cpu) {
976 r = svm_cpu_init(cpu);
981 if (!boot_cpu_has(X86_FEATURE_NPT))
984 if (npt_enabled && !npt)
987 kvm_configure_mmu(npt_enabled, get_max_npt_level(), PG_LEVEL_1G);
988 pr_info("kvm: Nested Paging %sabled\n", npt_enabled ? "en" : "dis");
991 if (!boot_cpu_has(X86_FEATURE_NRIPS))
997 !boot_cpu_has(X86_FEATURE_AVIC) ||
998 !IS_ENABLED(CONFIG_X86_LOCAL_APIC)) {
1001 pr_info("AVIC enabled\n");
1003 amd_iommu_register_ga_log_notifier(&avic_ga_log_notifier);
1009 !boot_cpu_has(X86_FEATURE_V_VMSAVE_VMLOAD) ||
1010 !IS_ENABLED(CONFIG_X86_64)) {
1013 pr_info("Virtual VMLOAD VMSAVE supported\n");
1018 if (!boot_cpu_has(X86_FEATURE_VGIF))
1021 pr_info("Virtual GIF supported\n");
1027 * It seems that on AMD processors PTE's accessed bit is
1028 * being set by the CPU hardware before the NPF vmexit.
1029 * This is not expected behaviour and our tests fail because
1031 * A workaround here is to disable support for
1032 * GUEST_MAXPHYADDR < HOST_MAXPHYADDR if NPT is enabled.
1033 * In this case userspace can know if there is support using
1034 * KVM_CAP_SMALLER_MAXPHYADDR extension and decide how to handle
1036 * If future AMD CPU models change the behaviour described above,
1037 * this variable can be changed accordingly
1039 allow_smaller_maxphyaddr = !npt_enabled;
1044 svm_hardware_teardown();
1048 static void init_seg(struct vmcb_seg *seg)
1051 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
1052 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
1053 seg->limit = 0xffff;
1057 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
1060 seg->attrib = SVM_SELECTOR_P_MASK | type;
1061 seg->limit = 0xffff;
1065 static u64 svm_write_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1067 struct vcpu_svm *svm = to_svm(vcpu);
1068 u64 g_tsc_offset = 0;
1070 if (is_guest_mode(vcpu)) {
1071 /* Write L1's TSC offset. */
1072 g_tsc_offset = svm->vmcb->control.tsc_offset -
1073 svm->nested.hsave->control.tsc_offset;
1074 svm->nested.hsave->control.tsc_offset = offset;
1077 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1078 svm->vmcb->control.tsc_offset - g_tsc_offset,
1081 svm->vmcb->control.tsc_offset = offset + g_tsc_offset;
1083 vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1084 return svm->vmcb->control.tsc_offset;
1087 static void svm_check_invpcid(struct vcpu_svm *svm)
1090 * Intercept INVPCID instruction only if shadow page table is
1091 * enabled. Interception is not required with nested page table
1094 if (kvm_cpu_cap_has(X86_FEATURE_INVPCID)) {
1096 svm_set_intercept(svm, INTERCEPT_INVPCID);
1098 svm_clr_intercept(svm, INTERCEPT_INVPCID);
1102 static void init_vmcb(struct vcpu_svm *svm)
1104 struct vmcb_control_area *control = &svm->vmcb->control;
1105 struct vmcb_save_area *save = &svm->vmcb->save;
1107 svm->vcpu.arch.hflags = 0;
1109 svm_set_intercept(svm, INTERCEPT_CR0_READ);
1110 svm_set_intercept(svm, INTERCEPT_CR3_READ);
1111 svm_set_intercept(svm, INTERCEPT_CR4_READ);
1112 svm_set_intercept(svm, INTERCEPT_CR0_WRITE);
1113 svm_set_intercept(svm, INTERCEPT_CR3_WRITE);
1114 svm_set_intercept(svm, INTERCEPT_CR4_WRITE);
1115 if (!kvm_vcpu_apicv_active(&svm->vcpu))
1116 svm_set_intercept(svm, INTERCEPT_CR8_WRITE);
1118 set_dr_intercepts(svm);
1120 set_exception_intercept(svm, PF_VECTOR);
1121 set_exception_intercept(svm, UD_VECTOR);
1122 set_exception_intercept(svm, MC_VECTOR);
1123 set_exception_intercept(svm, AC_VECTOR);
1124 set_exception_intercept(svm, DB_VECTOR);
1126 * Guest access to VMware backdoor ports could legitimately
1127 * trigger #GP because of TSS I/O permission bitmap.
1128 * We intercept those #GP and allow access to them anyway
1131 if (enable_vmware_backdoor)
1132 set_exception_intercept(svm, GP_VECTOR);
1134 svm_set_intercept(svm, INTERCEPT_INTR);
1135 svm_set_intercept(svm, INTERCEPT_NMI);
1136 svm_set_intercept(svm, INTERCEPT_SMI);
1137 svm_set_intercept(svm, INTERCEPT_SELECTIVE_CR0);
1138 svm_set_intercept(svm, INTERCEPT_RDPMC);
1139 svm_set_intercept(svm, INTERCEPT_CPUID);
1140 svm_set_intercept(svm, INTERCEPT_INVD);
1141 svm_set_intercept(svm, INTERCEPT_INVLPG);
1142 svm_set_intercept(svm, INTERCEPT_INVLPGA);
1143 svm_set_intercept(svm, INTERCEPT_IOIO_PROT);
1144 svm_set_intercept(svm, INTERCEPT_MSR_PROT);
1145 svm_set_intercept(svm, INTERCEPT_TASK_SWITCH);
1146 svm_set_intercept(svm, INTERCEPT_SHUTDOWN);
1147 svm_set_intercept(svm, INTERCEPT_VMRUN);
1148 svm_set_intercept(svm, INTERCEPT_VMMCALL);
1149 svm_set_intercept(svm, INTERCEPT_VMLOAD);
1150 svm_set_intercept(svm, INTERCEPT_VMSAVE);
1151 svm_set_intercept(svm, INTERCEPT_STGI);
1152 svm_set_intercept(svm, INTERCEPT_CLGI);
1153 svm_set_intercept(svm, INTERCEPT_SKINIT);
1154 svm_set_intercept(svm, INTERCEPT_WBINVD);
1155 svm_set_intercept(svm, INTERCEPT_XSETBV);
1156 svm_set_intercept(svm, INTERCEPT_RDPRU);
1157 svm_set_intercept(svm, INTERCEPT_RSM);
1159 if (!kvm_mwait_in_guest(svm->vcpu.kvm)) {
1160 svm_set_intercept(svm, INTERCEPT_MONITOR);
1161 svm_set_intercept(svm, INTERCEPT_MWAIT);
1164 if (!kvm_hlt_in_guest(svm->vcpu.kvm))
1165 svm_set_intercept(svm, INTERCEPT_HLT);
1167 control->iopm_base_pa = __sme_set(iopm_base);
1168 control->msrpm_base_pa = __sme_set(__pa(svm->msrpm));
1169 control->int_ctl = V_INTR_MASKING_MASK;
1171 init_seg(&save->es);
1172 init_seg(&save->ss);
1173 init_seg(&save->ds);
1174 init_seg(&save->fs);
1175 init_seg(&save->gs);
1177 save->cs.selector = 0xf000;
1178 save->cs.base = 0xffff0000;
1179 /* Executable/Readable Code Segment */
1180 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
1181 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
1182 save->cs.limit = 0xffff;
1184 save->gdtr.limit = 0xffff;
1185 save->idtr.limit = 0xffff;
1187 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
1188 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
1190 svm_set_efer(&svm->vcpu, 0);
1191 save->dr6 = 0xffff0ff0;
1192 kvm_set_rflags(&svm->vcpu, 2);
1193 save->rip = 0x0000fff0;
1194 svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
1197 * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
1198 * It also updates the guest-visible cr0 value.
1200 svm_set_cr0(&svm->vcpu, X86_CR0_NW | X86_CR0_CD | X86_CR0_ET);
1201 kvm_mmu_reset_context(&svm->vcpu);
1203 save->cr4 = X86_CR4_PAE;
1207 /* Setup VMCB for Nested Paging */
1208 control->nested_ctl |= SVM_NESTED_CTL_NP_ENABLE;
1209 svm_clr_intercept(svm, INTERCEPT_INVLPG);
1210 clr_exception_intercept(svm, PF_VECTOR);
1211 svm_clr_intercept(svm, INTERCEPT_CR3_READ);
1212 svm_clr_intercept(svm, INTERCEPT_CR3_WRITE);
1213 save->g_pat = svm->vcpu.arch.pat;
1217 svm->asid_generation = 0;
1220 svm->nested.vmcb12_gpa = 0;
1221 svm->vcpu.arch.hflags = 0;
1223 if (!kvm_pause_in_guest(svm->vcpu.kvm)) {
1224 control->pause_filter_count = pause_filter_count;
1225 if (pause_filter_thresh)
1226 control->pause_filter_thresh = pause_filter_thresh;
1227 svm_set_intercept(svm, INTERCEPT_PAUSE);
1229 svm_clr_intercept(svm, INTERCEPT_PAUSE);
1232 svm_check_invpcid(svm);
1234 if (kvm_vcpu_apicv_active(&svm->vcpu))
1235 avic_init_vmcb(svm);
1238 * If hardware supports Virtual VMLOAD VMSAVE then enable it
1239 * in VMCB and clear intercepts to avoid #VMEXIT.
1242 svm_clr_intercept(svm, INTERCEPT_VMLOAD);
1243 svm_clr_intercept(svm, INTERCEPT_VMSAVE);
1244 svm->vmcb->control.virt_ext |= VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK;
1248 svm_clr_intercept(svm, INTERCEPT_STGI);
1249 svm_clr_intercept(svm, INTERCEPT_CLGI);
1250 svm->vmcb->control.int_ctl |= V_GIF_ENABLE_MASK;
1253 if (sev_guest(svm->vcpu.kvm)) {
1254 svm->vmcb->control.nested_ctl |= SVM_NESTED_CTL_SEV_ENABLE;
1255 clr_exception_intercept(svm, UD_VECTOR);
1258 vmcb_mark_all_dirty(svm->vmcb);
1264 static void svm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
1266 struct vcpu_svm *svm = to_svm(vcpu);
1271 svm->virt_spec_ctrl = 0;
1274 svm->vcpu.arch.apic_base = APIC_DEFAULT_PHYS_BASE |
1275 MSR_IA32_APICBASE_ENABLE;
1276 if (kvm_vcpu_is_reset_bsp(&svm->vcpu))
1277 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
1281 kvm_cpuid(vcpu, &eax, &dummy, &dummy, &dummy, false);
1282 kvm_rdx_write(vcpu, eax);
1284 if (kvm_vcpu_apicv_active(vcpu) && !init_event)
1285 avic_update_vapic_bar(svm, APIC_DEFAULT_PHYS_BASE);
1288 static int svm_create_vcpu(struct kvm_vcpu *vcpu)
1290 struct vcpu_svm *svm;
1291 struct page *vmcb_page;
1294 BUILD_BUG_ON(offsetof(struct vcpu_svm, vcpu) != 0);
1298 vmcb_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
1302 err = avic_init_vcpu(svm);
1304 goto error_free_vmcb_page;
1306 /* We initialize this flag to true to make sure that the is_running
1307 * bit would be set the first time the vcpu is loaded.
1309 if (irqchip_in_kernel(vcpu->kvm) && kvm_apicv_activated(vcpu->kvm))
1310 svm->avic_is_running = true;
1312 svm->msrpm = svm_vcpu_alloc_msrpm();
1314 goto error_free_vmcb_page;
1316 svm_vcpu_init_msrpm(vcpu, svm->msrpm);
1318 svm->vmcb = page_address(vmcb_page);
1319 svm->vmcb_pa = __sme_set(page_to_pfn(vmcb_page) << PAGE_SHIFT);
1320 svm->asid_generation = 0;
1323 svm_init_osvw(vcpu);
1324 vcpu->arch.microcode_version = 0x01000065;
1328 error_free_vmcb_page:
1329 __free_page(vmcb_page);
1334 static void svm_clear_current_vmcb(struct vmcb *vmcb)
1338 for_each_online_cpu(i)
1339 cmpxchg(&per_cpu(svm_data, i)->current_vmcb, vmcb, NULL);
1342 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
1344 struct vcpu_svm *svm = to_svm(vcpu);
1347 * The vmcb page can be recycled, causing a false negative in
1348 * svm_vcpu_load(). So, ensure that no logical CPU has this
1349 * vmcb page recorded as its current vmcb.
1351 svm_clear_current_vmcb(svm->vmcb);
1353 svm_free_nested(svm);
1355 __free_page(pfn_to_page(__sme_clr(svm->vmcb_pa) >> PAGE_SHIFT));
1356 __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
1359 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1361 struct vcpu_svm *svm = to_svm(vcpu);
1362 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
1365 if (unlikely(cpu != vcpu->cpu)) {
1366 svm->asid_generation = 0;
1367 vmcb_mark_all_dirty(svm->vmcb);
1370 #ifdef CONFIG_X86_64
1371 rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host.gs_base);
1373 savesegment(fs, svm->host.fs);
1374 savesegment(gs, svm->host.gs);
1375 svm->host.ldt = kvm_read_ldt();
1377 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1378 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1380 if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
1381 u64 tsc_ratio = vcpu->arch.tsc_scaling_ratio;
1382 if (tsc_ratio != __this_cpu_read(current_tsc_ratio)) {
1383 __this_cpu_write(current_tsc_ratio, tsc_ratio);
1384 wrmsrl(MSR_AMD64_TSC_RATIO, tsc_ratio);
1387 /* This assumes that the kernel never uses MSR_TSC_AUX */
1388 if (static_cpu_has(X86_FEATURE_RDTSCP))
1389 wrmsrl(MSR_TSC_AUX, svm->tsc_aux);
1391 if (sd->current_vmcb != svm->vmcb) {
1392 sd->current_vmcb = svm->vmcb;
1393 indirect_branch_prediction_barrier();
1395 avic_vcpu_load(vcpu, cpu);
1398 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
1400 struct vcpu_svm *svm = to_svm(vcpu);
1403 avic_vcpu_put(vcpu);
1405 ++vcpu->stat.host_state_reload;
1406 kvm_load_ldt(svm->host.ldt);
1407 #ifdef CONFIG_X86_64
1408 loadsegment(fs, svm->host.fs);
1409 wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gsbase);
1410 load_gs_index(svm->host.gs);
1412 #ifdef CONFIG_X86_32_LAZY_GS
1413 loadsegment(gs, svm->host.gs);
1416 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1417 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
1420 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
1422 struct vcpu_svm *svm = to_svm(vcpu);
1423 unsigned long rflags = svm->vmcb->save.rflags;
1425 if (svm->nmi_singlestep) {
1426 /* Hide our flags if they were not set by the guest */
1427 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF))
1428 rflags &= ~X86_EFLAGS_TF;
1429 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF))
1430 rflags &= ~X86_EFLAGS_RF;
1435 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1437 if (to_svm(vcpu)->nmi_singlestep)
1438 rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
1441 * Any change of EFLAGS.VM is accompanied by a reload of SS
1442 * (caused by either a task switch or an inter-privilege IRET),
1443 * so we do not need to update the CPL here.
1445 to_svm(vcpu)->vmcb->save.rflags = rflags;
1448 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1451 case VCPU_EXREG_PDPTR:
1452 BUG_ON(!npt_enabled);
1453 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
1460 static void svm_set_vintr(struct vcpu_svm *svm)
1462 struct vmcb_control_area *control;
1464 /* The following fields are ignored when AVIC is enabled */
1465 WARN_ON(kvm_vcpu_apicv_active(&svm->vcpu));
1466 svm_set_intercept(svm, INTERCEPT_VINTR);
1469 * This is just a dummy VINTR to actually cause a vmexit to happen.
1470 * Actual injection of virtual interrupts happens through EVENTINJ.
1472 control = &svm->vmcb->control;
1473 control->int_vector = 0x0;
1474 control->int_ctl &= ~V_INTR_PRIO_MASK;
1475 control->int_ctl |= V_IRQ_MASK |
1476 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
1477 vmcb_mark_dirty(svm->vmcb, VMCB_INTR);
1480 static void svm_clear_vintr(struct vcpu_svm *svm)
1482 const u32 mask = V_TPR_MASK | V_GIF_ENABLE_MASK | V_GIF_MASK | V_INTR_MASKING_MASK;
1483 svm_clr_intercept(svm, INTERCEPT_VINTR);
1485 /* Drop int_ctl fields related to VINTR injection. */
1486 svm->vmcb->control.int_ctl &= mask;
1487 if (is_guest_mode(&svm->vcpu)) {
1488 svm->nested.hsave->control.int_ctl &= mask;
1490 WARN_ON((svm->vmcb->control.int_ctl & V_TPR_MASK) !=
1491 (svm->nested.ctl.int_ctl & V_TPR_MASK));
1492 svm->vmcb->control.int_ctl |= svm->nested.ctl.int_ctl & ~mask;
1495 vmcb_mark_dirty(svm->vmcb, VMCB_INTR);
1498 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
1500 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1503 case VCPU_SREG_CS: return &save->cs;
1504 case VCPU_SREG_DS: return &save->ds;
1505 case VCPU_SREG_ES: return &save->es;
1506 case VCPU_SREG_FS: return &save->fs;
1507 case VCPU_SREG_GS: return &save->gs;
1508 case VCPU_SREG_SS: return &save->ss;
1509 case VCPU_SREG_TR: return &save->tr;
1510 case VCPU_SREG_LDTR: return &save->ldtr;
1516 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1518 struct vmcb_seg *s = svm_seg(vcpu, seg);
1523 static void svm_get_segment(struct kvm_vcpu *vcpu,
1524 struct kvm_segment *var, int seg)
1526 struct vmcb_seg *s = svm_seg(vcpu, seg);
1528 var->base = s->base;
1529 var->limit = s->limit;
1530 var->selector = s->selector;
1531 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
1532 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
1533 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
1534 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
1535 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
1536 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
1537 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
1540 * AMD CPUs circa 2014 track the G bit for all segments except CS.
1541 * However, the SVM spec states that the G bit is not observed by the
1542 * CPU, and some VMware virtual CPUs drop the G bit for all segments.
1543 * So let's synthesize a legal G bit for all segments, this helps
1544 * running KVM nested. It also helps cross-vendor migration, because
1545 * Intel's vmentry has a check on the 'G' bit.
1547 var->g = s->limit > 0xfffff;
1550 * AMD's VMCB does not have an explicit unusable field, so emulate it
1551 * for cross vendor migration purposes by "not present"
1553 var->unusable = !var->present;
1558 * Work around a bug where the busy flag in the tr selector
1568 * The accessed bit must always be set in the segment
1569 * descriptor cache, although it can be cleared in the
1570 * descriptor, the cached bit always remains at 1. Since
1571 * Intel has a check on this, set it here to support
1572 * cross-vendor migration.
1579 * On AMD CPUs sometimes the DB bit in the segment
1580 * descriptor is left as 1, although the whole segment has
1581 * been made unusable. Clear it here to pass an Intel VMX
1582 * entry check when cross vendor migrating.
1586 /* This is symmetric with svm_set_segment() */
1587 var->dpl = to_svm(vcpu)->vmcb->save.cpl;
1592 static int svm_get_cpl(struct kvm_vcpu *vcpu)
1594 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1599 static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1601 struct vcpu_svm *svm = to_svm(vcpu);
1603 dt->size = svm->vmcb->save.idtr.limit;
1604 dt->address = svm->vmcb->save.idtr.base;
1607 static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1609 struct vcpu_svm *svm = to_svm(vcpu);
1611 svm->vmcb->save.idtr.limit = dt->size;
1612 svm->vmcb->save.idtr.base = dt->address ;
1613 vmcb_mark_dirty(svm->vmcb, VMCB_DT);
1616 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1618 struct vcpu_svm *svm = to_svm(vcpu);
1620 dt->size = svm->vmcb->save.gdtr.limit;
1621 dt->address = svm->vmcb->save.gdtr.base;
1624 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1626 struct vcpu_svm *svm = to_svm(vcpu);
1628 svm->vmcb->save.gdtr.limit = dt->size;
1629 svm->vmcb->save.gdtr.base = dt->address ;
1630 vmcb_mark_dirty(svm->vmcb, VMCB_DT);
1633 static void update_cr0_intercept(struct vcpu_svm *svm)
1635 ulong gcr0 = svm->vcpu.arch.cr0;
1636 u64 *hcr0 = &svm->vmcb->save.cr0;
1638 *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
1639 | (gcr0 & SVM_CR0_SELECTIVE_MASK);
1641 vmcb_mark_dirty(svm->vmcb, VMCB_CR);
1643 if (gcr0 == *hcr0) {
1644 svm_clr_intercept(svm, INTERCEPT_CR0_READ);
1645 svm_clr_intercept(svm, INTERCEPT_CR0_WRITE);
1647 svm_set_intercept(svm, INTERCEPT_CR0_READ);
1648 svm_set_intercept(svm, INTERCEPT_CR0_WRITE);
1652 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;
1675 * re-enable caching here because the QEMU bios
1676 * does not do it - this results in some delay at
1679 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
1680 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
1681 svm->vmcb->save.cr0 = cr0;
1682 vmcb_mark_dirty(svm->vmcb, VMCB_CR);
1683 update_cr0_intercept(svm);
1686 static bool svm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1691 void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1693 unsigned long host_cr4_mce = cr4_read_shadow() & X86_CR4_MCE;
1694 unsigned long old_cr4 = vcpu->arch.cr4;
1696 if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
1697 svm_flush_tlb(vcpu);
1699 vcpu->arch.cr4 = cr4;
1702 cr4 |= host_cr4_mce;
1703 to_svm(vcpu)->vmcb->save.cr4 = cr4;
1704 vmcb_mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
1706 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
1707 kvm_update_cpuid_runtime(vcpu);
1710 static void svm_set_segment(struct kvm_vcpu *vcpu,
1711 struct kvm_segment *var, int seg)
1713 struct vcpu_svm *svm = to_svm(vcpu);
1714 struct vmcb_seg *s = svm_seg(vcpu, seg);
1716 s->base = var->base;
1717 s->limit = var->limit;
1718 s->selector = var->selector;
1719 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
1720 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
1721 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
1722 s->attrib |= ((var->present & 1) && !var->unusable) << SVM_SELECTOR_P_SHIFT;
1723 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
1724 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
1725 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
1726 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
1729 * This is always accurate, except if SYSRET returned to a segment
1730 * with SS.DPL != 3. Intel does not have this quirk, and always
1731 * forces SS.DPL to 3 on sysret, so we ignore that case; fixing it
1732 * would entail passing the CPL to userspace and back.
1734 if (seg == VCPU_SREG_SS)
1735 /* This is symmetric with svm_get_segment() */
1736 svm->vmcb->save.cpl = (var->dpl & 3);
1738 vmcb_mark_dirty(svm->vmcb, VMCB_SEG);
1741 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1743 struct vcpu_svm *svm = to_svm(vcpu);
1745 clr_exception_intercept(svm, BP_VECTOR);
1747 if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
1748 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
1749 set_exception_intercept(svm, BP_VECTOR);
1753 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
1755 if (sd->next_asid > sd->max_asid) {
1756 ++sd->asid_generation;
1757 sd->next_asid = sd->min_asid;
1758 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
1759 vmcb_mark_dirty(svm->vmcb, VMCB_ASID);
1762 svm->asid_generation = sd->asid_generation;
1763 svm->asid = sd->next_asid++;
1766 static void svm_set_dr6(struct vcpu_svm *svm, unsigned long value)
1768 struct vmcb *vmcb = svm->vmcb;
1770 if (unlikely(value != vmcb->save.dr6)) {
1771 vmcb->save.dr6 = value;
1772 vmcb_mark_dirty(vmcb, VMCB_DR);
1776 static void svm_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
1778 struct vcpu_svm *svm = to_svm(vcpu);
1780 get_debugreg(vcpu->arch.db[0], 0);
1781 get_debugreg(vcpu->arch.db[1], 1);
1782 get_debugreg(vcpu->arch.db[2], 2);
1783 get_debugreg(vcpu->arch.db[3], 3);
1785 * We cannot reset svm->vmcb->save.dr6 to DR6_FIXED_1|DR6_RTM here,
1786 * because db_interception might need it. We can do it before vmentry.
1788 vcpu->arch.dr6 = svm->vmcb->save.dr6;
1789 vcpu->arch.dr7 = svm->vmcb->save.dr7;
1790 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
1791 set_dr_intercepts(svm);
1794 static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value)
1796 struct vcpu_svm *svm = to_svm(vcpu);
1798 svm->vmcb->save.dr7 = value;
1799 vmcb_mark_dirty(svm->vmcb, VMCB_DR);
1802 static int pf_interception(struct vcpu_svm *svm)
1804 u64 fault_address = __sme_clr(svm->vmcb->control.exit_info_2);
1805 u64 error_code = svm->vmcb->control.exit_info_1;
1807 return kvm_handle_page_fault(&svm->vcpu, error_code, fault_address,
1808 static_cpu_has(X86_FEATURE_DECODEASSISTS) ?
1809 svm->vmcb->control.insn_bytes : NULL,
1810 svm->vmcb->control.insn_len);
1813 static int npf_interception(struct vcpu_svm *svm)
1815 u64 fault_address = __sme_clr(svm->vmcb->control.exit_info_2);
1816 u64 error_code = svm->vmcb->control.exit_info_1;
1818 trace_kvm_page_fault(fault_address, error_code);
1819 return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code,
1820 static_cpu_has(X86_FEATURE_DECODEASSISTS) ?
1821 svm->vmcb->control.insn_bytes : NULL,
1822 svm->vmcb->control.insn_len);
1825 static int db_interception(struct vcpu_svm *svm)
1827 struct kvm_run *kvm_run = svm->vcpu.run;
1828 struct kvm_vcpu *vcpu = &svm->vcpu;
1830 if (!(svm->vcpu.guest_debug &
1831 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
1832 !svm->nmi_singlestep) {
1833 u32 payload = (svm->vmcb->save.dr6 ^ DR6_RTM) & ~DR6_FIXED_1;
1834 kvm_queue_exception_p(&svm->vcpu, DB_VECTOR, payload);
1838 if (svm->nmi_singlestep) {
1839 disable_nmi_singlestep(svm);
1840 /* Make sure we check for pending NMIs upon entry */
1841 kvm_make_request(KVM_REQ_EVENT, vcpu);
1844 if (svm->vcpu.guest_debug &
1845 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
1846 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1847 kvm_run->debug.arch.dr6 = svm->vmcb->save.dr6;
1848 kvm_run->debug.arch.dr7 = svm->vmcb->save.dr7;
1849 kvm_run->debug.arch.pc =
1850 svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1851 kvm_run->debug.arch.exception = DB_VECTOR;
1858 static int bp_interception(struct vcpu_svm *svm)
1860 struct kvm_run *kvm_run = svm->vcpu.run;
1862 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1863 kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1864 kvm_run->debug.arch.exception = BP_VECTOR;
1868 static int ud_interception(struct vcpu_svm *svm)
1870 return handle_ud(&svm->vcpu);
1873 static int ac_interception(struct vcpu_svm *svm)
1875 kvm_queue_exception_e(&svm->vcpu, AC_VECTOR, 0);
1879 static int gp_interception(struct vcpu_svm *svm)
1881 struct kvm_vcpu *vcpu = &svm->vcpu;
1882 u32 error_code = svm->vmcb->control.exit_info_1;
1884 WARN_ON_ONCE(!enable_vmware_backdoor);
1887 * VMware backdoor emulation on #GP interception only handles IN{S},
1888 * OUT{S}, and RDPMC, none of which generate a non-zero error code.
1891 kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
1894 return kvm_emulate_instruction(vcpu, EMULTYPE_VMWARE_GP);
1897 static bool is_erratum_383(void)
1902 if (!erratum_383_found)
1905 value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err);
1909 /* Bit 62 may or may not be set for this mce */
1910 value &= ~(1ULL << 62);
1912 if (value != 0xb600000000010015ULL)
1915 /* Clear MCi_STATUS registers */
1916 for (i = 0; i < 6; ++i)
1917 native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0);
1919 value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err);
1923 value &= ~(1ULL << 2);
1924 low = lower_32_bits(value);
1925 high = upper_32_bits(value);
1927 native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high);
1930 /* Flush tlb to evict multi-match entries */
1937 * Trigger machine check on the host. We assume all the MSRs are already set up
1938 * by the CPU and that we still run on the same CPU as the MCE occurred on.
1939 * We pass a fake environment to the machine check handler because we want
1940 * the guest to be always treated like user space, no matter what context
1941 * it used internally.
1943 static void kvm_machine_check(void)
1945 #if defined(CONFIG_X86_MCE)
1946 struct pt_regs regs = {
1947 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
1948 .flags = X86_EFLAGS_IF,
1951 do_machine_check(®s);
1955 static void svm_handle_mce(struct vcpu_svm *svm)
1957 if (is_erratum_383()) {
1959 * Erratum 383 triggered. Guest state is corrupt so kill the
1962 pr_err("KVM: Guest triggered AMD Erratum 383\n");
1964 kvm_make_request(KVM_REQ_TRIPLE_FAULT, &svm->vcpu);
1970 * On an #MC intercept the MCE handler is not called automatically in
1971 * the host. So do it by hand here.
1973 kvm_machine_check();
1976 static int mc_interception(struct vcpu_svm *svm)
1981 static int shutdown_interception(struct vcpu_svm *svm)
1983 struct kvm_run *kvm_run = svm->vcpu.run;
1986 * VMCB is undefined after a SHUTDOWN intercept
1987 * so reinitialize it.
1989 clear_page(svm->vmcb);
1992 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1996 static int io_interception(struct vcpu_svm *svm)
1998 struct kvm_vcpu *vcpu = &svm->vcpu;
1999 u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
2000 int size, in, string;
2003 ++svm->vcpu.stat.io_exits;
2004 string = (io_info & SVM_IOIO_STR_MASK) != 0;
2005 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
2007 return kvm_emulate_instruction(vcpu, 0);
2009 port = io_info >> 16;
2010 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
2011 svm->next_rip = svm->vmcb->control.exit_info_2;
2013 return kvm_fast_pio(&svm->vcpu, size, port, in);
2016 static int nmi_interception(struct vcpu_svm *svm)
2021 static int intr_interception(struct vcpu_svm *svm)
2023 ++svm->vcpu.stat.irq_exits;
2027 static int nop_on_interception(struct vcpu_svm *svm)
2032 static int halt_interception(struct vcpu_svm *svm)
2034 return kvm_emulate_halt(&svm->vcpu);
2037 static int vmmcall_interception(struct vcpu_svm *svm)
2039 return kvm_emulate_hypercall(&svm->vcpu);
2042 static int vmload_interception(struct vcpu_svm *svm)
2044 struct vmcb *nested_vmcb;
2045 struct kvm_host_map map;
2048 if (nested_svm_check_permissions(svm))
2051 ret = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(svm->vmcb->save.rax), &map);
2054 kvm_inject_gp(&svm->vcpu, 0);
2058 nested_vmcb = map.hva;
2060 ret = kvm_skip_emulated_instruction(&svm->vcpu);
2062 nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
2063 kvm_vcpu_unmap(&svm->vcpu, &map, true);
2068 static int vmsave_interception(struct vcpu_svm *svm)
2070 struct vmcb *nested_vmcb;
2071 struct kvm_host_map map;
2074 if (nested_svm_check_permissions(svm))
2077 ret = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(svm->vmcb->save.rax), &map);
2080 kvm_inject_gp(&svm->vcpu, 0);
2084 nested_vmcb = map.hva;
2086 ret = kvm_skip_emulated_instruction(&svm->vcpu);
2088 nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
2089 kvm_vcpu_unmap(&svm->vcpu, &map, true);
2094 static int vmrun_interception(struct vcpu_svm *svm)
2096 if (nested_svm_check_permissions(svm))
2099 return nested_svm_vmrun(svm);
2102 void svm_set_gif(struct vcpu_svm *svm, bool value)
2106 * If VGIF is enabled, the STGI intercept is only added to
2107 * detect the opening of the SMI/NMI window; remove it now.
2108 * Likewise, clear the VINTR intercept, we will set it
2109 * again while processing KVM_REQ_EVENT if needed.
2111 if (vgif_enabled(svm))
2112 svm_clr_intercept(svm, INTERCEPT_STGI);
2113 if (svm_is_intercept(svm, INTERCEPT_VINTR))
2114 svm_clear_vintr(svm);
2117 if (svm->vcpu.arch.smi_pending ||
2118 svm->vcpu.arch.nmi_pending ||
2119 kvm_cpu_has_injectable_intr(&svm->vcpu))
2120 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2125 * After a CLGI no interrupts should come. But if vGIF is
2126 * in use, we still rely on the VINTR intercept (rather than
2127 * STGI) to detect an open interrupt window.
2129 if (!vgif_enabled(svm))
2130 svm_clear_vintr(svm);
2134 static int stgi_interception(struct vcpu_svm *svm)
2138 if (nested_svm_check_permissions(svm))
2141 ret = kvm_skip_emulated_instruction(&svm->vcpu);
2142 svm_set_gif(svm, true);
2146 static int clgi_interception(struct vcpu_svm *svm)
2150 if (nested_svm_check_permissions(svm))
2153 ret = kvm_skip_emulated_instruction(&svm->vcpu);
2154 svm_set_gif(svm, false);
2158 static int invlpga_interception(struct vcpu_svm *svm)
2160 struct kvm_vcpu *vcpu = &svm->vcpu;
2162 trace_kvm_invlpga(svm->vmcb->save.rip, kvm_rcx_read(&svm->vcpu),
2163 kvm_rax_read(&svm->vcpu));
2165 /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
2166 kvm_mmu_invlpg(vcpu, kvm_rax_read(&svm->vcpu));
2168 return kvm_skip_emulated_instruction(&svm->vcpu);
2171 static int skinit_interception(struct vcpu_svm *svm)
2173 trace_kvm_skinit(svm->vmcb->save.rip, kvm_rax_read(&svm->vcpu));
2175 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2179 static int wbinvd_interception(struct vcpu_svm *svm)
2181 return kvm_emulate_wbinvd(&svm->vcpu);
2184 static int xsetbv_interception(struct vcpu_svm *svm)
2186 u64 new_bv = kvm_read_edx_eax(&svm->vcpu);
2187 u32 index = kvm_rcx_read(&svm->vcpu);
2189 if (kvm_set_xcr(&svm->vcpu, index, new_bv) == 0) {
2190 return kvm_skip_emulated_instruction(&svm->vcpu);
2196 static int rdpru_interception(struct vcpu_svm *svm)
2198 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2202 static int task_switch_interception(struct vcpu_svm *svm)
2206 int int_type = svm->vmcb->control.exit_int_info &
2207 SVM_EXITINTINFO_TYPE_MASK;
2208 int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
2210 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
2212 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
2213 bool has_error_code = false;
2216 tss_selector = (u16)svm->vmcb->control.exit_info_1;
2218 if (svm->vmcb->control.exit_info_2 &
2219 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
2220 reason = TASK_SWITCH_IRET;
2221 else if (svm->vmcb->control.exit_info_2 &
2222 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
2223 reason = TASK_SWITCH_JMP;
2225 reason = TASK_SWITCH_GATE;
2227 reason = TASK_SWITCH_CALL;
2229 if (reason == TASK_SWITCH_GATE) {
2231 case SVM_EXITINTINFO_TYPE_NMI:
2232 svm->vcpu.arch.nmi_injected = false;
2234 case SVM_EXITINTINFO_TYPE_EXEPT:
2235 if (svm->vmcb->control.exit_info_2 &
2236 (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) {
2237 has_error_code = true;
2239 (u32)svm->vmcb->control.exit_info_2;
2241 kvm_clear_exception_queue(&svm->vcpu);
2243 case SVM_EXITINTINFO_TYPE_INTR:
2244 kvm_clear_interrupt_queue(&svm->vcpu);
2251 if (reason != TASK_SWITCH_GATE ||
2252 int_type == SVM_EXITINTINFO_TYPE_SOFT ||
2253 (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
2254 (int_vec == OF_VECTOR || int_vec == BP_VECTOR))) {
2255 if (!skip_emulated_instruction(&svm->vcpu))
2259 if (int_type != SVM_EXITINTINFO_TYPE_SOFT)
2262 return kvm_task_switch(&svm->vcpu, tss_selector, int_vec, reason,
2263 has_error_code, error_code);
2266 static int cpuid_interception(struct vcpu_svm *svm)
2268 return kvm_emulate_cpuid(&svm->vcpu);
2271 static int iret_interception(struct vcpu_svm *svm)
2273 ++svm->vcpu.stat.nmi_window_exits;
2274 svm_clr_intercept(svm, INTERCEPT_IRET);
2275 svm->vcpu.arch.hflags |= HF_IRET_MASK;
2276 svm->nmi_iret_rip = kvm_rip_read(&svm->vcpu);
2277 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2281 static int invd_interception(struct vcpu_svm *svm)
2283 /* Treat an INVD instruction as a NOP and just skip it. */
2284 return kvm_skip_emulated_instruction(&svm->vcpu);
2287 static int invlpg_interception(struct vcpu_svm *svm)
2289 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2290 return kvm_emulate_instruction(&svm->vcpu, 0);
2292 kvm_mmu_invlpg(&svm->vcpu, svm->vmcb->control.exit_info_1);
2293 return kvm_skip_emulated_instruction(&svm->vcpu);
2296 static int emulate_on_interception(struct vcpu_svm *svm)
2298 return kvm_emulate_instruction(&svm->vcpu, 0);
2301 static int rsm_interception(struct vcpu_svm *svm)
2303 return kvm_emulate_instruction_from_buffer(&svm->vcpu, rsm_ins_bytes, 2);
2306 static int rdpmc_interception(struct vcpu_svm *svm)
2311 return emulate_on_interception(svm);
2313 err = kvm_rdpmc(&svm->vcpu);
2314 return kvm_complete_insn_gp(&svm->vcpu, err);
2317 static bool check_selective_cr0_intercepted(struct vcpu_svm *svm,
2320 unsigned long cr0 = svm->vcpu.arch.cr0;
2323 if (!is_guest_mode(&svm->vcpu) ||
2324 (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_SELECTIVE_CR0))))
2327 cr0 &= ~SVM_CR0_SELECTIVE_MASK;
2328 val &= ~SVM_CR0_SELECTIVE_MASK;
2331 svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
2332 ret = (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE);
2338 #define CR_VALID (1ULL << 63)
2340 static int cr_interception(struct vcpu_svm *svm)
2346 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2347 return emulate_on_interception(svm);
2349 if (unlikely((svm->vmcb->control.exit_info_1 & CR_VALID) == 0))
2350 return emulate_on_interception(svm);
2352 reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
2353 if (svm->vmcb->control.exit_code == SVM_EXIT_CR0_SEL_WRITE)
2354 cr = SVM_EXIT_WRITE_CR0 - SVM_EXIT_READ_CR0;
2356 cr = svm->vmcb->control.exit_code - SVM_EXIT_READ_CR0;
2359 if (cr >= 16) { /* mov to cr */
2361 val = kvm_register_read(&svm->vcpu, reg);
2362 trace_kvm_cr_write(cr, val);
2365 if (!check_selective_cr0_intercepted(svm, val))
2366 err = kvm_set_cr0(&svm->vcpu, val);
2372 err = kvm_set_cr3(&svm->vcpu, val);
2375 err = kvm_set_cr4(&svm->vcpu, val);
2378 err = kvm_set_cr8(&svm->vcpu, val);
2381 WARN(1, "unhandled write to CR%d", cr);
2382 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2385 } else { /* mov from cr */
2388 val = kvm_read_cr0(&svm->vcpu);
2391 val = svm->vcpu.arch.cr2;
2394 val = kvm_read_cr3(&svm->vcpu);
2397 val = kvm_read_cr4(&svm->vcpu);
2400 val = kvm_get_cr8(&svm->vcpu);
2403 WARN(1, "unhandled read from CR%d", cr);
2404 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2407 kvm_register_write(&svm->vcpu, reg, val);
2408 trace_kvm_cr_read(cr, val);
2410 return kvm_complete_insn_gp(&svm->vcpu, err);
2413 static int dr_interception(struct vcpu_svm *svm)
2418 if (svm->vcpu.guest_debug == 0) {
2420 * No more DR vmexits; force a reload of the debug registers
2421 * and reenter on this instruction. The next vmexit will
2422 * retrieve the full state of the debug registers.
2424 clr_dr_intercepts(svm);
2425 svm->vcpu.arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
2429 if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS))
2430 return emulate_on_interception(svm);
2432 reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
2433 dr = svm->vmcb->control.exit_code - SVM_EXIT_READ_DR0;
2435 if (dr >= 16) { /* mov to DRn */
2436 if (!kvm_require_dr(&svm->vcpu, dr - 16))
2438 val = kvm_register_read(&svm->vcpu, reg);
2439 kvm_set_dr(&svm->vcpu, dr - 16, val);
2441 if (!kvm_require_dr(&svm->vcpu, dr))
2443 kvm_get_dr(&svm->vcpu, dr, &val);
2444 kvm_register_write(&svm->vcpu, reg, val);
2447 return kvm_skip_emulated_instruction(&svm->vcpu);
2450 static int cr8_write_interception(struct vcpu_svm *svm)
2452 struct kvm_run *kvm_run = svm->vcpu.run;
2455 u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
2456 /* instruction emulation calls kvm_set_cr8() */
2457 r = cr_interception(svm);
2458 if (lapic_in_kernel(&svm->vcpu))
2460 if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
2462 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
2466 static int svm_get_msr_feature(struct kvm_msr_entry *msr)
2470 switch (msr->index) {
2471 case MSR_F10H_DECFG:
2472 if (boot_cpu_has(X86_FEATURE_LFENCE_RDTSC))
2473 msr->data |= MSR_F10H_DECFG_LFENCE_SERIALIZE;
2475 case MSR_IA32_PERF_CAPABILITIES:
2478 return KVM_MSR_RET_INVALID;
2484 static int svm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2486 struct vcpu_svm *svm = to_svm(vcpu);
2488 switch (msr_info->index) {
2490 msr_info->data = svm->vmcb->save.star;
2492 #ifdef CONFIG_X86_64
2494 msr_info->data = svm->vmcb->save.lstar;
2497 msr_info->data = svm->vmcb->save.cstar;
2499 case MSR_KERNEL_GS_BASE:
2500 msr_info->data = svm->vmcb->save.kernel_gs_base;
2502 case MSR_SYSCALL_MASK:
2503 msr_info->data = svm->vmcb->save.sfmask;
2506 case MSR_IA32_SYSENTER_CS:
2507 msr_info->data = svm->vmcb->save.sysenter_cs;
2509 case MSR_IA32_SYSENTER_EIP:
2510 msr_info->data = svm->sysenter_eip;
2512 case MSR_IA32_SYSENTER_ESP:
2513 msr_info->data = svm->sysenter_esp;
2516 if (!boot_cpu_has(X86_FEATURE_RDTSCP))
2518 msr_info->data = svm->tsc_aux;
2521 * Nobody will change the following 5 values in the VMCB so we can
2522 * safely return them on rdmsr. They will always be 0 until LBRV is
2525 case MSR_IA32_DEBUGCTLMSR:
2526 msr_info->data = svm->vmcb->save.dbgctl;
2528 case MSR_IA32_LASTBRANCHFROMIP:
2529 msr_info->data = svm->vmcb->save.br_from;
2531 case MSR_IA32_LASTBRANCHTOIP:
2532 msr_info->data = svm->vmcb->save.br_to;
2534 case MSR_IA32_LASTINTFROMIP:
2535 msr_info->data = svm->vmcb->save.last_excp_from;
2537 case MSR_IA32_LASTINTTOIP:
2538 msr_info->data = svm->vmcb->save.last_excp_to;
2540 case MSR_VM_HSAVE_PA:
2541 msr_info->data = svm->nested.hsave_msr;
2544 msr_info->data = svm->nested.vm_cr_msr;
2546 case MSR_IA32_SPEC_CTRL:
2547 if (!msr_info->host_initiated &&
2548 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL) &&
2549 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_STIBP) &&
2550 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBRS) &&
2551 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_SSBD))
2554 msr_info->data = svm->spec_ctrl;
2556 case MSR_AMD64_VIRT_SPEC_CTRL:
2557 if (!msr_info->host_initiated &&
2558 !guest_cpuid_has(vcpu, X86_FEATURE_VIRT_SSBD))
2561 msr_info->data = svm->virt_spec_ctrl;
2563 case MSR_F15H_IC_CFG: {
2567 family = guest_cpuid_family(vcpu);
2568 model = guest_cpuid_model(vcpu);
2570 if (family < 0 || model < 0)
2571 return kvm_get_msr_common(vcpu, msr_info);
2575 if (family == 0x15 &&
2576 (model >= 0x2 && model < 0x20))
2577 msr_info->data = 0x1E;
2580 case MSR_F10H_DECFG:
2581 msr_info->data = svm->msr_decfg;
2584 return kvm_get_msr_common(vcpu, msr_info);
2589 static int rdmsr_interception(struct vcpu_svm *svm)
2591 return kvm_emulate_rdmsr(&svm->vcpu);
2594 static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
2596 struct vcpu_svm *svm = to_svm(vcpu);
2597 int svm_dis, chg_mask;
2599 if (data & ~SVM_VM_CR_VALID_MASK)
2602 chg_mask = SVM_VM_CR_VALID_MASK;
2604 if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
2605 chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
2607 svm->nested.vm_cr_msr &= ~chg_mask;
2608 svm->nested.vm_cr_msr |= (data & chg_mask);
2610 svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
2612 /* check for svm_disable while efer.svme is set */
2613 if (svm_dis && (vcpu->arch.efer & EFER_SVME))
2619 static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2621 struct vcpu_svm *svm = to_svm(vcpu);
2623 u32 ecx = msr->index;
2624 u64 data = msr->data;
2626 case MSR_IA32_CR_PAT:
2627 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
2629 vcpu->arch.pat = data;
2630 svm->vmcb->save.g_pat = data;
2631 vmcb_mark_dirty(svm->vmcb, VMCB_NPT);
2633 case MSR_IA32_SPEC_CTRL:
2634 if (!msr->host_initiated &&
2635 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL) &&
2636 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_STIBP) &&
2637 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBRS) &&
2638 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_SSBD))
2641 if (kvm_spec_ctrl_test_value(data))
2644 svm->spec_ctrl = data;
2650 * When it's written (to non-zero) for the first time, pass
2654 * The handling of the MSR bitmap for L2 guests is done in
2655 * nested_svm_vmrun_msrpm.
2656 * We update the L1 MSR bit as well since it will end up
2657 * touching the MSR anyway now.
2659 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_SPEC_CTRL, 1, 1);
2661 case MSR_IA32_PRED_CMD:
2662 if (!msr->host_initiated &&
2663 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBPB))
2666 if (data & ~PRED_CMD_IBPB)
2668 if (!boot_cpu_has(X86_FEATURE_AMD_IBPB))
2673 wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
2674 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_PRED_CMD, 0, 1);
2676 case MSR_AMD64_VIRT_SPEC_CTRL:
2677 if (!msr->host_initiated &&
2678 !guest_cpuid_has(vcpu, X86_FEATURE_VIRT_SSBD))
2681 if (data & ~SPEC_CTRL_SSBD)
2684 svm->virt_spec_ctrl = data;
2687 svm->vmcb->save.star = data;
2689 #ifdef CONFIG_X86_64
2691 svm->vmcb->save.lstar = data;
2694 svm->vmcb->save.cstar = data;
2696 case MSR_KERNEL_GS_BASE:
2697 svm->vmcb->save.kernel_gs_base = data;
2699 case MSR_SYSCALL_MASK:
2700 svm->vmcb->save.sfmask = data;
2703 case MSR_IA32_SYSENTER_CS:
2704 svm->vmcb->save.sysenter_cs = data;
2706 case MSR_IA32_SYSENTER_EIP:
2707 svm->sysenter_eip = data;
2708 svm->vmcb->save.sysenter_eip = data;
2710 case MSR_IA32_SYSENTER_ESP:
2711 svm->sysenter_esp = data;
2712 svm->vmcb->save.sysenter_esp = data;
2715 if (!boot_cpu_has(X86_FEATURE_RDTSCP))
2719 * This is rare, so we update the MSR here instead of using
2720 * direct_access_msrs. Doing that would require a rdmsr in
2723 svm->tsc_aux = data;
2724 wrmsrl(MSR_TSC_AUX, svm->tsc_aux);
2726 case MSR_IA32_DEBUGCTLMSR:
2727 if (!boot_cpu_has(X86_FEATURE_LBRV)) {
2728 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
2732 if (data & DEBUGCTL_RESERVED_BITS)
2735 svm->vmcb->save.dbgctl = data;
2736 vmcb_mark_dirty(svm->vmcb, VMCB_LBR);
2737 if (data & (1ULL<<0))
2738 svm_enable_lbrv(vcpu);
2740 svm_disable_lbrv(vcpu);
2742 case MSR_VM_HSAVE_PA:
2743 svm->nested.hsave_msr = data;
2746 return svm_set_vm_cr(vcpu, data);
2748 vcpu_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
2750 case MSR_F10H_DECFG: {
2751 struct kvm_msr_entry msr_entry;
2753 msr_entry.index = msr->index;
2754 if (svm_get_msr_feature(&msr_entry))
2757 /* Check the supported bits */
2758 if (data & ~msr_entry.data)
2761 /* Don't allow the guest to change a bit, #GP */
2762 if (!msr->host_initiated && (data ^ msr_entry.data))
2765 svm->msr_decfg = data;
2768 case MSR_IA32_APICBASE:
2769 if (kvm_vcpu_apicv_active(vcpu))
2770 avic_update_vapic_bar(to_svm(vcpu), data);
2773 return kvm_set_msr_common(vcpu, msr);
2778 static int wrmsr_interception(struct vcpu_svm *svm)
2780 return kvm_emulate_wrmsr(&svm->vcpu);
2783 static int msr_interception(struct vcpu_svm *svm)
2785 if (svm->vmcb->control.exit_info_1)
2786 return wrmsr_interception(svm);
2788 return rdmsr_interception(svm);
2791 static int interrupt_window_interception(struct vcpu_svm *svm)
2793 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2794 svm_clear_vintr(svm);
2797 * For AVIC, the only reason to end up here is ExtINTs.
2798 * In this case AVIC was temporarily disabled for
2799 * requesting the IRQ window and we have to re-enable it.
2801 svm_toggle_avic_for_irq_window(&svm->vcpu, true);
2803 ++svm->vcpu.stat.irq_window_exits;
2807 static int pause_interception(struct vcpu_svm *svm)
2809 struct kvm_vcpu *vcpu = &svm->vcpu;
2810 bool in_kernel = (svm_get_cpl(vcpu) == 0);
2812 if (!kvm_pause_in_guest(vcpu->kvm))
2813 grow_ple_window(vcpu);
2815 kvm_vcpu_on_spin(vcpu, in_kernel);
2819 static int nop_interception(struct vcpu_svm *svm)
2821 return kvm_skip_emulated_instruction(&(svm->vcpu));
2824 static int monitor_interception(struct vcpu_svm *svm)
2826 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
2827 return nop_interception(svm);
2830 static int mwait_interception(struct vcpu_svm *svm)
2832 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
2833 return nop_interception(svm);
2836 static int invpcid_interception(struct vcpu_svm *svm)
2838 struct kvm_vcpu *vcpu = &svm->vcpu;
2842 if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
2843 kvm_queue_exception(vcpu, UD_VECTOR);
2848 * For an INVPCID intercept:
2849 * EXITINFO1 provides the linear address of the memory operand.
2850 * EXITINFO2 provides the contents of the register operand.
2852 type = svm->vmcb->control.exit_info_2;
2853 gva = svm->vmcb->control.exit_info_1;
2856 kvm_inject_gp(vcpu, 0);
2860 return kvm_handle_invpcid(vcpu, type, gva);
2863 static int (*const svm_exit_handlers[])(struct vcpu_svm *svm) = {
2864 [SVM_EXIT_READ_CR0] = cr_interception,
2865 [SVM_EXIT_READ_CR3] = cr_interception,
2866 [SVM_EXIT_READ_CR4] = cr_interception,
2867 [SVM_EXIT_READ_CR8] = cr_interception,
2868 [SVM_EXIT_CR0_SEL_WRITE] = cr_interception,
2869 [SVM_EXIT_WRITE_CR0] = cr_interception,
2870 [SVM_EXIT_WRITE_CR3] = cr_interception,
2871 [SVM_EXIT_WRITE_CR4] = cr_interception,
2872 [SVM_EXIT_WRITE_CR8] = cr8_write_interception,
2873 [SVM_EXIT_READ_DR0] = dr_interception,
2874 [SVM_EXIT_READ_DR1] = dr_interception,
2875 [SVM_EXIT_READ_DR2] = dr_interception,
2876 [SVM_EXIT_READ_DR3] = dr_interception,
2877 [SVM_EXIT_READ_DR4] = dr_interception,
2878 [SVM_EXIT_READ_DR5] = dr_interception,
2879 [SVM_EXIT_READ_DR6] = dr_interception,
2880 [SVM_EXIT_READ_DR7] = dr_interception,
2881 [SVM_EXIT_WRITE_DR0] = dr_interception,
2882 [SVM_EXIT_WRITE_DR1] = dr_interception,
2883 [SVM_EXIT_WRITE_DR2] = dr_interception,
2884 [SVM_EXIT_WRITE_DR3] = dr_interception,
2885 [SVM_EXIT_WRITE_DR4] = dr_interception,
2886 [SVM_EXIT_WRITE_DR5] = dr_interception,
2887 [SVM_EXIT_WRITE_DR6] = dr_interception,
2888 [SVM_EXIT_WRITE_DR7] = dr_interception,
2889 [SVM_EXIT_EXCP_BASE + DB_VECTOR] = db_interception,
2890 [SVM_EXIT_EXCP_BASE + BP_VECTOR] = bp_interception,
2891 [SVM_EXIT_EXCP_BASE + UD_VECTOR] = ud_interception,
2892 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
2893 [SVM_EXIT_EXCP_BASE + MC_VECTOR] = mc_interception,
2894 [SVM_EXIT_EXCP_BASE + AC_VECTOR] = ac_interception,
2895 [SVM_EXIT_EXCP_BASE + GP_VECTOR] = gp_interception,
2896 [SVM_EXIT_INTR] = intr_interception,
2897 [SVM_EXIT_NMI] = nmi_interception,
2898 [SVM_EXIT_SMI] = nop_on_interception,
2899 [SVM_EXIT_INIT] = nop_on_interception,
2900 [SVM_EXIT_VINTR] = interrupt_window_interception,
2901 [SVM_EXIT_RDPMC] = rdpmc_interception,
2902 [SVM_EXIT_CPUID] = cpuid_interception,
2903 [SVM_EXIT_IRET] = iret_interception,
2904 [SVM_EXIT_INVD] = invd_interception,
2905 [SVM_EXIT_PAUSE] = pause_interception,
2906 [SVM_EXIT_HLT] = halt_interception,
2907 [SVM_EXIT_INVLPG] = invlpg_interception,
2908 [SVM_EXIT_INVLPGA] = invlpga_interception,
2909 [SVM_EXIT_IOIO] = io_interception,
2910 [SVM_EXIT_MSR] = msr_interception,
2911 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
2912 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
2913 [SVM_EXIT_VMRUN] = vmrun_interception,
2914 [SVM_EXIT_VMMCALL] = vmmcall_interception,
2915 [SVM_EXIT_VMLOAD] = vmload_interception,
2916 [SVM_EXIT_VMSAVE] = vmsave_interception,
2917 [SVM_EXIT_STGI] = stgi_interception,
2918 [SVM_EXIT_CLGI] = clgi_interception,
2919 [SVM_EXIT_SKINIT] = skinit_interception,
2920 [SVM_EXIT_WBINVD] = wbinvd_interception,
2921 [SVM_EXIT_MONITOR] = monitor_interception,
2922 [SVM_EXIT_MWAIT] = mwait_interception,
2923 [SVM_EXIT_XSETBV] = xsetbv_interception,
2924 [SVM_EXIT_RDPRU] = rdpru_interception,
2925 [SVM_EXIT_INVPCID] = invpcid_interception,
2926 [SVM_EXIT_NPF] = npf_interception,
2927 [SVM_EXIT_RSM] = rsm_interception,
2928 [SVM_EXIT_AVIC_INCOMPLETE_IPI] = avic_incomplete_ipi_interception,
2929 [SVM_EXIT_AVIC_UNACCELERATED_ACCESS] = avic_unaccelerated_access_interception,
2932 static void dump_vmcb(struct kvm_vcpu *vcpu)
2934 struct vcpu_svm *svm = to_svm(vcpu);
2935 struct vmcb_control_area *control = &svm->vmcb->control;
2936 struct vmcb_save_area *save = &svm->vmcb->save;
2938 if (!dump_invalid_vmcb) {
2939 pr_warn_ratelimited("set kvm_amd.dump_invalid_vmcb=1 to dump internal KVM state.\n");
2943 pr_err("VMCB Control Area:\n");
2944 pr_err("%-20s%04x\n", "cr_read:", control->intercepts[INTERCEPT_CR] & 0xffff);
2945 pr_err("%-20s%04x\n", "cr_write:", control->intercepts[INTERCEPT_CR] >> 16);
2946 pr_err("%-20s%04x\n", "dr_read:", control->intercepts[INTERCEPT_DR] & 0xffff);
2947 pr_err("%-20s%04x\n", "dr_write:", control->intercepts[INTERCEPT_DR] >> 16);
2948 pr_err("%-20s%08x\n", "exceptions:", control->intercepts[INTERCEPT_EXCEPTION]);
2949 pr_err("%-20s%08x %08x\n", "intercepts:",
2950 control->intercepts[INTERCEPT_WORD3],
2951 control->intercepts[INTERCEPT_WORD4]);
2952 pr_err("%-20s%d\n", "pause filter count:", control->pause_filter_count);
2953 pr_err("%-20s%d\n", "pause filter threshold:",
2954 control->pause_filter_thresh);
2955 pr_err("%-20s%016llx\n", "iopm_base_pa:", control->iopm_base_pa);
2956 pr_err("%-20s%016llx\n", "msrpm_base_pa:", control->msrpm_base_pa);
2957 pr_err("%-20s%016llx\n", "tsc_offset:", control->tsc_offset);
2958 pr_err("%-20s%d\n", "asid:", control->asid);
2959 pr_err("%-20s%d\n", "tlb_ctl:", control->tlb_ctl);
2960 pr_err("%-20s%08x\n", "int_ctl:", control->int_ctl);
2961 pr_err("%-20s%08x\n", "int_vector:", control->int_vector);
2962 pr_err("%-20s%08x\n", "int_state:", control->int_state);
2963 pr_err("%-20s%08x\n", "exit_code:", control->exit_code);
2964 pr_err("%-20s%016llx\n", "exit_info1:", control->exit_info_1);
2965 pr_err("%-20s%016llx\n", "exit_info2:", control->exit_info_2);
2966 pr_err("%-20s%08x\n", "exit_int_info:", control->exit_int_info);
2967 pr_err("%-20s%08x\n", "exit_int_info_err:", control->exit_int_info_err);
2968 pr_err("%-20s%lld\n", "nested_ctl:", control->nested_ctl);
2969 pr_err("%-20s%016llx\n", "nested_cr3:", control->nested_cr3);
2970 pr_err("%-20s%016llx\n", "avic_vapic_bar:", control->avic_vapic_bar);
2971 pr_err("%-20s%08x\n", "event_inj:", control->event_inj);
2972 pr_err("%-20s%08x\n", "event_inj_err:", control->event_inj_err);
2973 pr_err("%-20s%lld\n", "virt_ext:", control->virt_ext);
2974 pr_err("%-20s%016llx\n", "next_rip:", control->next_rip);
2975 pr_err("%-20s%016llx\n", "avic_backing_page:", control->avic_backing_page);
2976 pr_err("%-20s%016llx\n", "avic_logical_id:", control->avic_logical_id);
2977 pr_err("%-20s%016llx\n", "avic_physical_id:", control->avic_physical_id);
2978 pr_err("VMCB State Save Area:\n");
2979 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
2981 save->es.selector, save->es.attrib,
2982 save->es.limit, save->es.base);
2983 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
2985 save->cs.selector, save->cs.attrib,
2986 save->cs.limit, save->cs.base);
2987 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
2989 save->ss.selector, save->ss.attrib,
2990 save->ss.limit, save->ss.base);
2991 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
2993 save->ds.selector, save->ds.attrib,
2994 save->ds.limit, save->ds.base);
2995 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
2997 save->fs.selector, save->fs.attrib,
2998 save->fs.limit, save->fs.base);
2999 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3001 save->gs.selector, save->gs.attrib,
3002 save->gs.limit, save->gs.base);
3003 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3005 save->gdtr.selector, save->gdtr.attrib,
3006 save->gdtr.limit, save->gdtr.base);
3007 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3009 save->ldtr.selector, save->ldtr.attrib,
3010 save->ldtr.limit, save->ldtr.base);
3011 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3013 save->idtr.selector, save->idtr.attrib,
3014 save->idtr.limit, save->idtr.base);
3015 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3017 save->tr.selector, save->tr.attrib,
3018 save->tr.limit, save->tr.base);
3019 pr_err("cpl: %d efer: %016llx\n",
3020 save->cpl, save->efer);
3021 pr_err("%-15s %016llx %-13s %016llx\n",
3022 "cr0:", save->cr0, "cr2:", save->cr2);
3023 pr_err("%-15s %016llx %-13s %016llx\n",
3024 "cr3:", save->cr3, "cr4:", save->cr4);
3025 pr_err("%-15s %016llx %-13s %016llx\n",
3026 "dr6:", save->dr6, "dr7:", save->dr7);
3027 pr_err("%-15s %016llx %-13s %016llx\n",
3028 "rip:", save->rip, "rflags:", save->rflags);
3029 pr_err("%-15s %016llx %-13s %016llx\n",
3030 "rsp:", save->rsp, "rax:", save->rax);
3031 pr_err("%-15s %016llx %-13s %016llx\n",
3032 "star:", save->star, "lstar:", save->lstar);
3033 pr_err("%-15s %016llx %-13s %016llx\n",
3034 "cstar:", save->cstar, "sfmask:", save->sfmask);
3035 pr_err("%-15s %016llx %-13s %016llx\n",
3036 "kernel_gs_base:", save->kernel_gs_base,
3037 "sysenter_cs:", save->sysenter_cs);
3038 pr_err("%-15s %016llx %-13s %016llx\n",
3039 "sysenter_esp:", save->sysenter_esp,
3040 "sysenter_eip:", save->sysenter_eip);
3041 pr_err("%-15s %016llx %-13s %016llx\n",
3042 "gpat:", save->g_pat, "dbgctl:", save->dbgctl);
3043 pr_err("%-15s %016llx %-13s %016llx\n",
3044 "br_from:", save->br_from, "br_to:", save->br_to);
3045 pr_err("%-15s %016llx %-13s %016llx\n",
3046 "excp_from:", save->last_excp_from,
3047 "excp_to:", save->last_excp_to);
3050 static void svm_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2,
3051 u32 *intr_info, u32 *error_code)
3053 struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control;
3055 *info1 = control->exit_info_1;
3056 *info2 = control->exit_info_2;
3057 *intr_info = control->exit_int_info;
3058 if ((*intr_info & SVM_EXITINTINFO_VALID) &&
3059 (*intr_info & SVM_EXITINTINFO_VALID_ERR))
3060 *error_code = control->exit_int_info_err;
3065 static int handle_exit(struct kvm_vcpu *vcpu, fastpath_t exit_fastpath)
3067 struct vcpu_svm *svm = to_svm(vcpu);
3068 struct kvm_run *kvm_run = vcpu->run;
3069 u32 exit_code = svm->vmcb->control.exit_code;
3071 trace_kvm_exit(exit_code, vcpu, KVM_ISA_SVM);
3073 if (!svm_is_intercept(svm, INTERCEPT_CR0_WRITE))
3074 vcpu->arch.cr0 = svm->vmcb->save.cr0;
3076 vcpu->arch.cr3 = svm->vmcb->save.cr3;
3078 if (is_guest_mode(vcpu)) {
3081 trace_kvm_nested_vmexit(exit_code, vcpu, KVM_ISA_SVM);
3083 vmexit = nested_svm_exit_special(svm);
3085 if (vmexit == NESTED_EXIT_CONTINUE)
3086 vmexit = nested_svm_exit_handled(svm);
3088 if (vmexit == NESTED_EXIT_DONE)
3092 if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
3093 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3094 kvm_run->fail_entry.hardware_entry_failure_reason
3095 = svm->vmcb->control.exit_code;
3096 kvm_run->fail_entry.cpu = vcpu->arch.last_vmentry_cpu;
3101 if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
3102 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
3103 exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH &&
3104 exit_code != SVM_EXIT_INTR && exit_code != SVM_EXIT_NMI)
3105 printk(KERN_ERR "%s: unexpected exit_int_info 0x%x "
3107 __func__, svm->vmcb->control.exit_int_info,
3110 if (exit_fastpath != EXIT_FASTPATH_NONE)
3113 if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
3114 || !svm_exit_handlers[exit_code]) {
3115 vcpu_unimpl(vcpu, "svm: unexpected exit reason 0x%x\n", exit_code);
3117 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3118 vcpu->run->internal.suberror =
3119 KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
3120 vcpu->run->internal.ndata = 2;
3121 vcpu->run->internal.data[0] = exit_code;
3122 vcpu->run->internal.data[1] = vcpu->arch.last_vmentry_cpu;
3126 #ifdef CONFIG_RETPOLINE
3127 if (exit_code == SVM_EXIT_MSR)
3128 return msr_interception(svm);
3129 else if (exit_code == SVM_EXIT_VINTR)
3130 return interrupt_window_interception(svm);
3131 else if (exit_code == SVM_EXIT_INTR)
3132 return intr_interception(svm);
3133 else if (exit_code == SVM_EXIT_HLT)
3134 return halt_interception(svm);
3135 else if (exit_code == SVM_EXIT_NPF)
3136 return npf_interception(svm);
3138 return svm_exit_handlers[exit_code](svm);
3141 static void reload_tss(struct kvm_vcpu *vcpu)
3143 struct svm_cpu_data *sd = per_cpu(svm_data, vcpu->cpu);
3145 sd->tss_desc->type = 9; /* available 32/64-bit TSS */
3149 static void pre_svm_run(struct vcpu_svm *svm)
3151 struct svm_cpu_data *sd = per_cpu(svm_data, svm->vcpu.cpu);
3153 if (sev_guest(svm->vcpu.kvm))
3154 return pre_sev_run(svm, svm->vcpu.cpu);
3156 /* FIXME: handle wraparound of asid_generation */
3157 if (svm->asid_generation != sd->asid_generation)
3161 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
3163 struct vcpu_svm *svm = to_svm(vcpu);
3165 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
3166 vcpu->arch.hflags |= HF_NMI_MASK;
3167 svm_set_intercept(svm, INTERCEPT_IRET);
3168 ++vcpu->stat.nmi_injections;
3171 static void svm_set_irq(struct kvm_vcpu *vcpu)
3173 struct vcpu_svm *svm = to_svm(vcpu);
3175 BUG_ON(!(gif_set(svm)));
3177 trace_kvm_inj_virq(vcpu->arch.interrupt.nr);
3178 ++vcpu->stat.irq_injections;
3180 svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
3181 SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
3184 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3186 struct vcpu_svm *svm = to_svm(vcpu);
3188 if (nested_svm_virtualize_tpr(vcpu))
3191 svm_clr_intercept(svm, INTERCEPT_CR8_WRITE);
3197 svm_set_intercept(svm, INTERCEPT_CR8_WRITE);
3200 bool svm_nmi_blocked(struct kvm_vcpu *vcpu)
3202 struct vcpu_svm *svm = to_svm(vcpu);
3203 struct vmcb *vmcb = svm->vmcb;
3209 if (is_guest_mode(vcpu) && nested_exit_on_nmi(svm))
3212 ret = (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) ||
3213 (svm->vcpu.arch.hflags & HF_NMI_MASK);
3218 static int svm_nmi_allowed(struct kvm_vcpu *vcpu, bool for_injection)
3220 struct vcpu_svm *svm = to_svm(vcpu);
3221 if (svm->nested.nested_run_pending)
3224 /* An NMI must not be injected into L2 if it's supposed to VM-Exit. */
3225 if (for_injection && is_guest_mode(vcpu) && nested_exit_on_nmi(svm))
3228 return !svm_nmi_blocked(vcpu);
3231 static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
3233 struct vcpu_svm *svm = to_svm(vcpu);
3235 return !!(svm->vcpu.arch.hflags & HF_NMI_MASK);
3238 static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
3240 struct vcpu_svm *svm = to_svm(vcpu);
3243 svm->vcpu.arch.hflags |= HF_NMI_MASK;
3244 svm_set_intercept(svm, INTERCEPT_IRET);
3246 svm->vcpu.arch.hflags &= ~HF_NMI_MASK;
3247 svm_clr_intercept(svm, INTERCEPT_IRET);
3251 bool svm_interrupt_blocked(struct kvm_vcpu *vcpu)
3253 struct vcpu_svm *svm = to_svm(vcpu);
3254 struct vmcb *vmcb = svm->vmcb;
3259 if (is_guest_mode(vcpu)) {
3260 /* As long as interrupts are being delivered... */
3261 if ((svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK)
3262 ? !(svm->nested.hsave->save.rflags & X86_EFLAGS_IF)
3263 : !(kvm_get_rflags(vcpu) & X86_EFLAGS_IF))
3266 /* ... vmexits aren't blocked by the interrupt shadow */
3267 if (nested_exit_on_intr(svm))
3270 if (!(kvm_get_rflags(vcpu) & X86_EFLAGS_IF))
3274 return (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK);
3277 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu, bool for_injection)
3279 struct vcpu_svm *svm = to_svm(vcpu);
3280 if (svm->nested.nested_run_pending)
3284 * An IRQ must not be injected into L2 if it's supposed to VM-Exit,
3285 * e.g. if the IRQ arrived asynchronously after checking nested events.
3287 if (for_injection && is_guest_mode(vcpu) && nested_exit_on_intr(svm))
3290 return !svm_interrupt_blocked(vcpu);
3293 static void enable_irq_window(struct kvm_vcpu *vcpu)
3295 struct vcpu_svm *svm = to_svm(vcpu);
3298 * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
3299 * 1, because that's a separate STGI/VMRUN intercept. The next time we
3300 * get that intercept, this function will be called again though and
3301 * we'll get the vintr intercept. However, if the vGIF feature is
3302 * enabled, the STGI interception will not occur. Enable the irq
3303 * window under the assumption that the hardware will set the GIF.
3305 if (vgif_enabled(svm) || gif_set(svm)) {
3307 * IRQ window is not needed when AVIC is enabled,
3308 * unless we have pending ExtINT since it cannot be injected
3309 * via AVIC. In such case, we need to temporarily disable AVIC,
3310 * and fallback to injecting IRQ via V_IRQ.
3312 svm_toggle_avic_for_irq_window(vcpu, false);
3317 static void enable_nmi_window(struct kvm_vcpu *vcpu)
3319 struct vcpu_svm *svm = to_svm(vcpu);
3321 if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
3323 return; /* IRET will cause a vm exit */
3325 if (!gif_set(svm)) {
3326 if (vgif_enabled(svm))
3327 svm_set_intercept(svm, INTERCEPT_STGI);
3328 return; /* STGI will cause a vm exit */
3332 * Something prevents NMI from been injected. Single step over possible
3333 * problem (IRET or exception injection or interrupt shadow)
3335 svm->nmi_singlestep_guest_rflags = svm_get_rflags(vcpu);
3336 svm->nmi_singlestep = true;
3337 svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
3340 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
3345 static int svm_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
3350 void svm_flush_tlb(struct kvm_vcpu *vcpu)
3352 struct vcpu_svm *svm = to_svm(vcpu);
3355 * Flush only the current ASID even if the TLB flush was invoked via
3356 * kvm_flush_remote_tlbs(). Although flushing remote TLBs requires all
3357 * ASIDs to be flushed, KVM uses a single ASID for L1 and L2, and
3358 * unconditionally does a TLB flush on both nested VM-Enter and nested
3359 * VM-Exit (via kvm_mmu_reset_context()).
3361 if (static_cpu_has(X86_FEATURE_FLUSHBYASID))
3362 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
3364 svm->asid_generation--;
3367 static void svm_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t gva)
3369 struct vcpu_svm *svm = to_svm(vcpu);
3371 invlpga(gva, svm->vmcb->control.asid);
3374 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
3378 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
3380 struct vcpu_svm *svm = to_svm(vcpu);
3382 if (nested_svm_virtualize_tpr(vcpu))
3385 if (!svm_is_intercept(svm, INTERCEPT_CR8_WRITE)) {
3386 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
3387 kvm_set_cr8(vcpu, cr8);
3391 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
3393 struct vcpu_svm *svm = to_svm(vcpu);
3396 if (nested_svm_virtualize_tpr(vcpu) ||
3397 kvm_vcpu_apicv_active(vcpu))
3400 cr8 = kvm_get_cr8(vcpu);
3401 svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
3402 svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
3405 static void svm_complete_interrupts(struct vcpu_svm *svm)
3409 u32 exitintinfo = svm->vmcb->control.exit_int_info;
3410 unsigned int3_injected = svm->int3_injected;
3412 svm->int3_injected = 0;
3415 * If we've made progress since setting HF_IRET_MASK, we've
3416 * executed an IRET and can allow NMI injection.
3418 if ((svm->vcpu.arch.hflags & HF_IRET_MASK)
3419 && kvm_rip_read(&svm->vcpu) != svm->nmi_iret_rip) {
3420 svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
3421 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3424 svm->vcpu.arch.nmi_injected = false;
3425 kvm_clear_exception_queue(&svm->vcpu);
3426 kvm_clear_interrupt_queue(&svm->vcpu);
3428 if (!(exitintinfo & SVM_EXITINTINFO_VALID))
3431 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3433 vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
3434 type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
3437 case SVM_EXITINTINFO_TYPE_NMI:
3438 svm->vcpu.arch.nmi_injected = true;
3440 case SVM_EXITINTINFO_TYPE_EXEPT:
3442 * In case of software exceptions, do not reinject the vector,
3443 * but re-execute the instruction instead. Rewind RIP first
3444 * if we emulated INT3 before.
3446 if (kvm_exception_is_soft(vector)) {
3447 if (vector == BP_VECTOR && int3_injected &&
3448 kvm_is_linear_rip(&svm->vcpu, svm->int3_rip))
3449 kvm_rip_write(&svm->vcpu,
3450 kvm_rip_read(&svm->vcpu) -
3454 if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
3455 u32 err = svm->vmcb->control.exit_int_info_err;
3456 kvm_requeue_exception_e(&svm->vcpu, vector, err);
3459 kvm_requeue_exception(&svm->vcpu, vector);
3461 case SVM_EXITINTINFO_TYPE_INTR:
3462 kvm_queue_interrupt(&svm->vcpu, vector, false);
3469 static void svm_cancel_injection(struct kvm_vcpu *vcpu)
3471 struct vcpu_svm *svm = to_svm(vcpu);
3472 struct vmcb_control_area *control = &svm->vmcb->control;
3474 control->exit_int_info = control->event_inj;
3475 control->exit_int_info_err = control->event_inj_err;
3476 control->event_inj = 0;
3477 svm_complete_interrupts(svm);
3480 static fastpath_t svm_exit_handlers_fastpath(struct kvm_vcpu *vcpu)
3482 if (to_svm(vcpu)->vmcb->control.exit_code == SVM_EXIT_MSR &&
3483 to_svm(vcpu)->vmcb->control.exit_info_1)
3484 return handle_fastpath_set_msr_irqoff(vcpu);
3486 return EXIT_FASTPATH_NONE;
3489 void __svm_vcpu_run(unsigned long vmcb_pa, unsigned long *regs);
3491 static noinstr void svm_vcpu_enter_exit(struct kvm_vcpu *vcpu,
3492 struct vcpu_svm *svm)
3495 * VMENTER enables interrupts (host state), but the kernel state is
3496 * interrupts disabled when this is invoked. Also tell RCU about
3497 * it. This is the same logic as for exit_to_user_mode().
3499 * This ensures that e.g. latency analysis on the host observes
3500 * guest mode as interrupt enabled.
3502 * guest_enter_irqoff() informs context tracking about the
3503 * transition to guest mode and if enabled adjusts RCU state
3506 instrumentation_begin();
3507 trace_hardirqs_on_prepare();
3508 lockdep_hardirqs_on_prepare(CALLER_ADDR0);
3509 instrumentation_end();
3511 guest_enter_irqoff();
3512 lockdep_hardirqs_on(CALLER_ADDR0);
3514 __svm_vcpu_run(svm->vmcb_pa, (unsigned long *)&svm->vcpu.arch.regs);
3516 #ifdef CONFIG_X86_64
3517 native_wrmsrl(MSR_GS_BASE, svm->host.gs_base);
3519 loadsegment(fs, svm->host.fs);
3520 #ifndef CONFIG_X86_32_LAZY_GS
3521 loadsegment(gs, svm->host.gs);
3526 * VMEXIT disables interrupts (host state), but tracing and lockdep
3527 * have them in state 'on' as recorded before entering guest mode.
3528 * Same as enter_from_user_mode().
3530 * guest_exit_irqoff() restores host context and reinstates RCU if
3531 * enabled and required.
3533 * This needs to be done before the below as native_read_msr()
3534 * contains a tracepoint and x86_spec_ctrl_restore_host() calls
3535 * into world and some more.
3537 lockdep_hardirqs_off(CALLER_ADDR0);
3538 guest_exit_irqoff();
3540 instrumentation_begin();
3541 trace_hardirqs_off_finish();
3542 instrumentation_end();
3545 static __no_kcsan fastpath_t svm_vcpu_run(struct kvm_vcpu *vcpu)
3547 struct vcpu_svm *svm = to_svm(vcpu);
3549 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
3550 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
3551 svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
3554 * Disable singlestep if we're injecting an interrupt/exception.
3555 * We don't want our modified rflags to be pushed on the stack where
3556 * we might not be able to easily reset them if we disabled NMI
3559 if (svm->nmi_singlestep && svm->vmcb->control.event_inj) {
3561 * Event injection happens before external interrupts cause a
3562 * vmexit and interrupts are disabled here, so smp_send_reschedule
3563 * is enough to force an immediate vmexit.
3565 disable_nmi_singlestep(svm);
3566 smp_send_reschedule(vcpu->cpu);
3571 sync_lapic_to_cr8(vcpu);
3573 if (unlikely(svm->asid != svm->vmcb->control.asid)) {
3574 svm->vmcb->control.asid = svm->asid;
3575 vmcb_mark_dirty(svm->vmcb, VMCB_ASID);
3577 svm->vmcb->save.cr2 = vcpu->arch.cr2;
3580 * Run with all-zero DR6 unless needed, so that we can get the exact cause
3583 if (unlikely(svm->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT))
3584 svm_set_dr6(svm, vcpu->arch.dr6);
3586 svm_set_dr6(svm, DR6_FIXED_1 | DR6_RTM);
3589 kvm_load_guest_xsave_state(vcpu);
3591 kvm_wait_lapic_expire(vcpu);
3594 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
3595 * it's non-zero. Since vmentry is serialising on affected CPUs, there
3596 * is no need to worry about the conditional branch over the wrmsr
3597 * being speculatively taken.
3599 x86_spec_ctrl_set_guest(svm->spec_ctrl, svm->virt_spec_ctrl);
3601 svm_vcpu_enter_exit(vcpu, svm);
3604 * We do not use IBRS in the kernel. If this vCPU has used the
3605 * SPEC_CTRL MSR it may have left it on; save the value and
3606 * turn it off. This is much more efficient than blindly adding
3607 * it to the atomic save/restore list. Especially as the former
3608 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
3610 * For non-nested case:
3611 * If the L01 MSR bitmap does not intercept the MSR, then we need to
3615 * If the L02 MSR bitmap does not intercept the MSR, then we need to
3618 if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
3619 svm->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
3623 x86_spec_ctrl_restore_host(svm->spec_ctrl, svm->virt_spec_ctrl);
3625 vcpu->arch.cr2 = svm->vmcb->save.cr2;
3626 vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
3627 vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
3628 vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
3630 if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
3631 kvm_before_interrupt(&svm->vcpu);
3633 kvm_load_host_xsave_state(vcpu);
3636 /* Any pending NMI will happen here */
3638 if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
3639 kvm_after_interrupt(&svm->vcpu);
3641 sync_cr8_to_lapic(vcpu);
3644 if (is_guest_mode(&svm->vcpu)) {
3645 sync_nested_vmcb_control(svm);
3646 svm->nested.nested_run_pending = 0;
3649 svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
3650 vmcb_mark_all_clean(svm->vmcb);
3652 /* if exit due to PF check for async PF */
3653 if (svm->vmcb->control.exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR)
3654 svm->vcpu.arch.apf.host_apf_flags =
3655 kvm_read_and_reset_apf_flags();
3658 vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
3659 vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
3663 * We need to handle MC intercepts here before the vcpu has a chance to
3664 * change the physical cpu
3666 if (unlikely(svm->vmcb->control.exit_code ==
3667 SVM_EXIT_EXCP_BASE + MC_VECTOR))
3668 svm_handle_mce(svm);
3670 svm_complete_interrupts(svm);
3672 if (is_guest_mode(vcpu))
3673 return EXIT_FASTPATH_NONE;
3675 return svm_exit_handlers_fastpath(vcpu);
3678 static void svm_load_mmu_pgd(struct kvm_vcpu *vcpu, unsigned long root,
3681 struct vcpu_svm *svm = to_svm(vcpu);
3684 cr3 = __sme_set(root);
3686 svm->vmcb->control.nested_cr3 = cr3;
3687 vmcb_mark_dirty(svm->vmcb, VMCB_NPT);
3689 /* Loading L2's CR3 is handled by enter_svm_guest_mode. */
3690 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3692 cr3 = vcpu->arch.cr3;
3695 svm->vmcb->save.cr3 = cr3;
3696 vmcb_mark_dirty(svm->vmcb, VMCB_CR);
3699 static int is_disabled(void)
3703 rdmsrl(MSR_VM_CR, vm_cr);
3704 if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
3711 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
3714 * Patch in the VMMCALL instruction:
3716 hypercall[0] = 0x0f;
3717 hypercall[1] = 0x01;
3718 hypercall[2] = 0xd9;
3721 static int __init svm_check_processor_compat(void)
3726 static bool svm_cpu_has_accelerated_tpr(void)
3731 static bool svm_has_emulated_msr(u32 index)
3734 case MSR_IA32_MCG_EXT_CTL:
3735 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3744 static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
3749 static void svm_vcpu_after_set_cpuid(struct kvm_vcpu *vcpu)
3751 struct vcpu_svm *svm = to_svm(vcpu);
3752 struct kvm_cpuid_entry2 *best;
3754 vcpu->arch.xsaves_enabled = guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
3755 boot_cpu_has(X86_FEATURE_XSAVE) &&
3756 boot_cpu_has(X86_FEATURE_XSAVES);
3758 /* Update nrips enabled cache */
3759 svm->nrips_enabled = kvm_cpu_cap_has(X86_FEATURE_NRIPS) &&
3760 guest_cpuid_has(&svm->vcpu, X86_FEATURE_NRIPS);
3762 /* Check again if INVPCID interception if required */
3763 svm_check_invpcid(svm);
3765 /* For sev guests, the memory encryption bit is not reserved in CR3. */
3766 if (sev_guest(vcpu->kvm)) {
3767 best = kvm_find_cpuid_entry(vcpu, 0x8000001F, 0);
3769 vcpu->arch.cr3_lm_rsvd_bits &= ~(1UL << (best->ebx & 0x3f));
3772 if (!kvm_vcpu_apicv_active(vcpu))
3776 * AVIC does not work with an x2APIC mode guest. If the X2APIC feature
3777 * is exposed to the guest, disable AVIC.
3779 if (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC))
3780 kvm_request_apicv_update(vcpu->kvm, false,
3781 APICV_INHIBIT_REASON_X2APIC);
3784 * Currently, AVIC does not work with nested virtualization.
3785 * So, we disable AVIC when cpuid for SVM is set in the L1 guest.
3787 if (nested && guest_cpuid_has(vcpu, X86_FEATURE_SVM))
3788 kvm_request_apicv_update(vcpu->kvm, false,
3789 APICV_INHIBIT_REASON_NESTED);
3792 static bool svm_has_wbinvd_exit(void)
3797 #define PRE_EX(exit) { .exit_code = (exit), \
3798 .stage = X86_ICPT_PRE_EXCEPT, }
3799 #define POST_EX(exit) { .exit_code = (exit), \
3800 .stage = X86_ICPT_POST_EXCEPT, }
3801 #define POST_MEM(exit) { .exit_code = (exit), \
3802 .stage = X86_ICPT_POST_MEMACCESS, }
3804 static const struct __x86_intercept {
3806 enum x86_intercept_stage stage;
3807 } x86_intercept_map[] = {
3808 [x86_intercept_cr_read] = POST_EX(SVM_EXIT_READ_CR0),
3809 [x86_intercept_cr_write] = POST_EX(SVM_EXIT_WRITE_CR0),
3810 [x86_intercept_clts] = POST_EX(SVM_EXIT_WRITE_CR0),
3811 [x86_intercept_lmsw] = POST_EX(SVM_EXIT_WRITE_CR0),
3812 [x86_intercept_smsw] = POST_EX(SVM_EXIT_READ_CR0),
3813 [x86_intercept_dr_read] = POST_EX(SVM_EXIT_READ_DR0),
3814 [x86_intercept_dr_write] = POST_EX(SVM_EXIT_WRITE_DR0),
3815 [x86_intercept_sldt] = POST_EX(SVM_EXIT_LDTR_READ),
3816 [x86_intercept_str] = POST_EX(SVM_EXIT_TR_READ),
3817 [x86_intercept_lldt] = POST_EX(SVM_EXIT_LDTR_WRITE),
3818 [x86_intercept_ltr] = POST_EX(SVM_EXIT_TR_WRITE),
3819 [x86_intercept_sgdt] = POST_EX(SVM_EXIT_GDTR_READ),
3820 [x86_intercept_sidt] = POST_EX(SVM_EXIT_IDTR_READ),
3821 [x86_intercept_lgdt] = POST_EX(SVM_EXIT_GDTR_WRITE),
3822 [x86_intercept_lidt] = POST_EX(SVM_EXIT_IDTR_WRITE),
3823 [x86_intercept_vmrun] = POST_EX(SVM_EXIT_VMRUN),
3824 [x86_intercept_vmmcall] = POST_EX(SVM_EXIT_VMMCALL),
3825 [x86_intercept_vmload] = POST_EX(SVM_EXIT_VMLOAD),
3826 [x86_intercept_vmsave] = POST_EX(SVM_EXIT_VMSAVE),
3827 [x86_intercept_stgi] = POST_EX(SVM_EXIT_STGI),
3828 [x86_intercept_clgi] = POST_EX(SVM_EXIT_CLGI),
3829 [x86_intercept_skinit] = POST_EX(SVM_EXIT_SKINIT),
3830 [x86_intercept_invlpga] = POST_EX(SVM_EXIT_INVLPGA),
3831 [x86_intercept_rdtscp] = POST_EX(SVM_EXIT_RDTSCP),
3832 [x86_intercept_monitor] = POST_MEM(SVM_EXIT_MONITOR),
3833 [x86_intercept_mwait] = POST_EX(SVM_EXIT_MWAIT),
3834 [x86_intercept_invlpg] = POST_EX(SVM_EXIT_INVLPG),
3835 [x86_intercept_invd] = POST_EX(SVM_EXIT_INVD),
3836 [x86_intercept_wbinvd] = POST_EX(SVM_EXIT_WBINVD),
3837 [x86_intercept_wrmsr] = POST_EX(SVM_EXIT_MSR),
3838 [x86_intercept_rdtsc] = POST_EX(SVM_EXIT_RDTSC),
3839 [x86_intercept_rdmsr] = POST_EX(SVM_EXIT_MSR),
3840 [x86_intercept_rdpmc] = POST_EX(SVM_EXIT_RDPMC),
3841 [x86_intercept_cpuid] = PRE_EX(SVM_EXIT_CPUID),
3842 [x86_intercept_rsm] = PRE_EX(SVM_EXIT_RSM),
3843 [x86_intercept_pause] = PRE_EX(SVM_EXIT_PAUSE),
3844 [x86_intercept_pushf] = PRE_EX(SVM_EXIT_PUSHF),
3845 [x86_intercept_popf] = PRE_EX(SVM_EXIT_POPF),
3846 [x86_intercept_intn] = PRE_EX(SVM_EXIT_SWINT),
3847 [x86_intercept_iret] = PRE_EX(SVM_EXIT_IRET),
3848 [x86_intercept_icebp] = PRE_EX(SVM_EXIT_ICEBP),
3849 [x86_intercept_hlt] = POST_EX(SVM_EXIT_HLT),
3850 [x86_intercept_in] = POST_EX(SVM_EXIT_IOIO),
3851 [x86_intercept_ins] = POST_EX(SVM_EXIT_IOIO),
3852 [x86_intercept_out] = POST_EX(SVM_EXIT_IOIO),
3853 [x86_intercept_outs] = POST_EX(SVM_EXIT_IOIO),
3854 [x86_intercept_xsetbv] = PRE_EX(SVM_EXIT_XSETBV),
3861 static int svm_check_intercept(struct kvm_vcpu *vcpu,
3862 struct x86_instruction_info *info,
3863 enum x86_intercept_stage stage,
3864 struct x86_exception *exception)
3866 struct vcpu_svm *svm = to_svm(vcpu);
3867 int vmexit, ret = X86EMUL_CONTINUE;
3868 struct __x86_intercept icpt_info;
3869 struct vmcb *vmcb = svm->vmcb;
3871 if (info->intercept >= ARRAY_SIZE(x86_intercept_map))
3874 icpt_info = x86_intercept_map[info->intercept];
3876 if (stage != icpt_info.stage)
3879 switch (icpt_info.exit_code) {
3880 case SVM_EXIT_READ_CR0:
3881 if (info->intercept == x86_intercept_cr_read)
3882 icpt_info.exit_code += info->modrm_reg;
3884 case SVM_EXIT_WRITE_CR0: {
3885 unsigned long cr0, val;
3887 if (info->intercept == x86_intercept_cr_write)
3888 icpt_info.exit_code += info->modrm_reg;
3890 if (icpt_info.exit_code != SVM_EXIT_WRITE_CR0 ||
3891 info->intercept == x86_intercept_clts)
3894 if (!(vmcb_is_intercept(&svm->nested.ctl,
3895 INTERCEPT_SELECTIVE_CR0)))
3898 cr0 = vcpu->arch.cr0 & ~SVM_CR0_SELECTIVE_MASK;
3899 val = info->src_val & ~SVM_CR0_SELECTIVE_MASK;
3901 if (info->intercept == x86_intercept_lmsw) {
3904 /* lmsw can't clear PE - catch this here */
3905 if (cr0 & X86_CR0_PE)
3910 icpt_info.exit_code = SVM_EXIT_CR0_SEL_WRITE;
3914 case SVM_EXIT_READ_DR0:
3915 case SVM_EXIT_WRITE_DR0:
3916 icpt_info.exit_code += info->modrm_reg;
3919 if (info->intercept == x86_intercept_wrmsr)
3920 vmcb->control.exit_info_1 = 1;
3922 vmcb->control.exit_info_1 = 0;
3924 case SVM_EXIT_PAUSE:
3926 * We get this for NOP only, but pause
3927 * is rep not, check this here
3929 if (info->rep_prefix != REPE_PREFIX)
3932 case SVM_EXIT_IOIO: {
3936 if (info->intercept == x86_intercept_in ||
3937 info->intercept == x86_intercept_ins) {
3938 exit_info = ((info->src_val & 0xffff) << 16) |
3940 bytes = info->dst_bytes;
3942 exit_info = (info->dst_val & 0xffff) << 16;
3943 bytes = info->src_bytes;
3946 if (info->intercept == x86_intercept_outs ||
3947 info->intercept == x86_intercept_ins)
3948 exit_info |= SVM_IOIO_STR_MASK;
3950 if (info->rep_prefix)
3951 exit_info |= SVM_IOIO_REP_MASK;
3953 bytes = min(bytes, 4u);
3955 exit_info |= bytes << SVM_IOIO_SIZE_SHIFT;
3957 exit_info |= (u32)info->ad_bytes << (SVM_IOIO_ASIZE_SHIFT - 1);
3959 vmcb->control.exit_info_1 = exit_info;
3960 vmcb->control.exit_info_2 = info->next_rip;
3968 /* TODO: Advertise NRIPS to guest hypervisor unconditionally */
3969 if (static_cpu_has(X86_FEATURE_NRIPS))
3970 vmcb->control.next_rip = info->next_rip;
3971 vmcb->control.exit_code = icpt_info.exit_code;
3972 vmexit = nested_svm_exit_handled(svm);
3974 ret = (vmexit == NESTED_EXIT_DONE) ? X86EMUL_INTERCEPTED
3981 static void svm_handle_exit_irqoff(struct kvm_vcpu *vcpu)
3985 static void svm_sched_in(struct kvm_vcpu *vcpu, int cpu)
3987 if (!kvm_pause_in_guest(vcpu->kvm))
3988 shrink_ple_window(vcpu);
3991 static void svm_setup_mce(struct kvm_vcpu *vcpu)
3993 /* [63:9] are reserved. */
3994 vcpu->arch.mcg_cap &= 0x1ff;
3997 bool svm_smi_blocked(struct kvm_vcpu *vcpu)
3999 struct vcpu_svm *svm = to_svm(vcpu);
4001 /* Per APM Vol.2 15.22.2 "Response to SMI" */
4005 return is_smm(vcpu);
4008 static int svm_smi_allowed(struct kvm_vcpu *vcpu, bool for_injection)
4010 struct vcpu_svm *svm = to_svm(vcpu);
4011 if (svm->nested.nested_run_pending)
4014 /* An SMI must not be injected into L2 if it's supposed to VM-Exit. */
4015 if (for_injection && is_guest_mode(vcpu) && nested_exit_on_smi(svm))
4018 return !svm_smi_blocked(vcpu);
4021 static int svm_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
4023 struct vcpu_svm *svm = to_svm(vcpu);
4026 if (is_guest_mode(vcpu)) {
4027 /* FED8h - SVM Guest */
4028 put_smstate(u64, smstate, 0x7ed8, 1);
4029 /* FEE0h - SVM Guest VMCB Physical Address */
4030 put_smstate(u64, smstate, 0x7ee0, svm->nested.vmcb12_gpa);
4032 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
4033 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
4034 svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
4036 ret = nested_svm_vmexit(svm);
4043 static int svm_pre_leave_smm(struct kvm_vcpu *vcpu, const char *smstate)
4045 struct vcpu_svm *svm = to_svm(vcpu);
4046 struct kvm_host_map map;
4049 if (guest_cpuid_has(vcpu, X86_FEATURE_LM)) {
4050 u64 saved_efer = GET_SMSTATE(u64, smstate, 0x7ed0);
4051 u64 guest = GET_SMSTATE(u64, smstate, 0x7ed8);
4052 u64 vmcb12_gpa = GET_SMSTATE(u64, smstate, 0x7ee0);
4055 if (!guest_cpuid_has(vcpu, X86_FEATURE_SVM))
4058 if (!(saved_efer & EFER_SVME))
4061 if (kvm_vcpu_map(&svm->vcpu,
4062 gpa_to_gfn(vmcb12_gpa), &map) == -EINVAL)
4065 if (svm_allocate_nested(svm))
4068 ret = enter_svm_guest_mode(svm, vmcb12_gpa, map.hva);
4069 kvm_vcpu_unmap(&svm->vcpu, &map, true);
4076 static void enable_smi_window(struct kvm_vcpu *vcpu)
4078 struct vcpu_svm *svm = to_svm(vcpu);
4080 if (!gif_set(svm)) {
4081 if (vgif_enabled(svm))
4082 svm_set_intercept(svm, INTERCEPT_STGI);
4083 /* STGI will cause a vm exit */
4085 /* We must be in SMM; RSM will cause a vmexit anyway. */
4089 static bool svm_can_emulate_instruction(struct kvm_vcpu *vcpu, void *insn, int insn_len)
4091 bool smep, smap, is_user;
4095 * Detect and workaround Errata 1096 Fam_17h_00_0Fh.
4098 * When CPU raise #NPF on guest data access and vCPU CR4.SMAP=1, it is
4099 * possible that CPU microcode implementing DecodeAssist will fail
4100 * to read bytes of instruction which caused #NPF. In this case,
4101 * GuestIntrBytes field of the VMCB on a VMEXIT will incorrectly
4102 * return 0 instead of the correct guest instruction bytes.
4104 * This happens because CPU microcode reading instruction bytes
4105 * uses a special opcode which attempts to read data using CPL=0
4106 * priviledges. The microcode reads CS:RIP and if it hits a SMAP
4107 * fault, it gives up and returns no instruction bytes.
4110 * We reach here in case CPU supports DecodeAssist, raised #NPF and
4111 * returned 0 in GuestIntrBytes field of the VMCB.
4112 * First, errata can only be triggered in case vCPU CR4.SMAP=1.
4113 * Second, if vCPU CR4.SMEP=1, errata could only be triggered
4114 * in case vCPU CPL==3 (Because otherwise guest would have triggered
4115 * a SMEP fault instead of #NPF).
4116 * Otherwise, vCPU CR4.SMEP=0, errata could be triggered by any vCPU CPL.
4117 * As most guests enable SMAP if they have also enabled SMEP, use above
4118 * logic in order to attempt minimize false-positive of detecting errata
4119 * while still preserving all cases semantic correctness.
4122 * To determine what instruction the guest was executing, the hypervisor
4123 * will have to decode the instruction at the instruction pointer.
4125 * In non SEV guest, hypervisor will be able to read the guest
4126 * memory to decode the instruction pointer when insn_len is zero
4127 * so we return true to indicate that decoding is possible.
4129 * But in the SEV guest, the guest memory is encrypted with the
4130 * guest specific key and hypervisor will not be able to decode the
4131 * instruction pointer so we will not able to workaround it. Lets
4132 * print the error and request to kill the guest.
4134 if (likely(!insn || insn_len))
4138 * If RIP is invalid, go ahead with emulation which will cause an
4139 * internal error exit.
4141 if (!kvm_vcpu_gfn_to_memslot(vcpu, kvm_rip_read(vcpu) >> PAGE_SHIFT))
4144 cr4 = kvm_read_cr4(vcpu);
4145 smep = cr4 & X86_CR4_SMEP;
4146 smap = cr4 & X86_CR4_SMAP;
4147 is_user = svm_get_cpl(vcpu) == 3;
4148 if (smap && (!smep || is_user)) {
4149 if (!sev_guest(vcpu->kvm))
4152 pr_err_ratelimited("KVM: SEV Guest triggered AMD Erratum 1096\n");
4153 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4159 static bool svm_apic_init_signal_blocked(struct kvm_vcpu *vcpu)
4161 struct vcpu_svm *svm = to_svm(vcpu);
4164 * TODO: Last condition latch INIT signals on vCPU when
4165 * vCPU is in guest-mode and vmcb12 defines intercept on INIT.
4166 * To properly emulate the INIT intercept,
4167 * svm_check_nested_events() should call nested_svm_vmexit()
4168 * if an INIT signal is pending.
4170 return !gif_set(svm) ||
4171 (vmcb_is_intercept(&svm->vmcb->control, INTERCEPT_INIT));
4174 static void svm_vm_destroy(struct kvm *kvm)
4176 avic_vm_destroy(kvm);
4177 sev_vm_destroy(kvm);
4180 static int svm_vm_init(struct kvm *kvm)
4182 if (!pause_filter_count || !pause_filter_thresh)
4183 kvm->arch.pause_in_guest = true;
4186 int ret = avic_vm_init(kvm);
4191 kvm_apicv_init(kvm, avic);
4195 static struct kvm_x86_ops svm_x86_ops __initdata = {
4196 .hardware_unsetup = svm_hardware_teardown,
4197 .hardware_enable = svm_hardware_enable,
4198 .hardware_disable = svm_hardware_disable,
4199 .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
4200 .has_emulated_msr = svm_has_emulated_msr,
4202 .vcpu_create = svm_create_vcpu,
4203 .vcpu_free = svm_free_vcpu,
4204 .vcpu_reset = svm_vcpu_reset,
4206 .vm_size = sizeof(struct kvm_svm),
4207 .vm_init = svm_vm_init,
4208 .vm_destroy = svm_vm_destroy,
4210 .prepare_guest_switch = svm_prepare_guest_switch,
4211 .vcpu_load = svm_vcpu_load,
4212 .vcpu_put = svm_vcpu_put,
4213 .vcpu_blocking = svm_vcpu_blocking,
4214 .vcpu_unblocking = svm_vcpu_unblocking,
4216 .update_exception_bitmap = update_exception_bitmap,
4217 .get_msr_feature = svm_get_msr_feature,
4218 .get_msr = svm_get_msr,
4219 .set_msr = svm_set_msr,
4220 .get_segment_base = svm_get_segment_base,
4221 .get_segment = svm_get_segment,
4222 .set_segment = svm_set_segment,
4223 .get_cpl = svm_get_cpl,
4224 .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
4225 .set_cr0 = svm_set_cr0,
4226 .is_valid_cr4 = svm_is_valid_cr4,
4227 .set_cr4 = svm_set_cr4,
4228 .set_efer = svm_set_efer,
4229 .get_idt = svm_get_idt,
4230 .set_idt = svm_set_idt,
4231 .get_gdt = svm_get_gdt,
4232 .set_gdt = svm_set_gdt,
4233 .set_dr7 = svm_set_dr7,
4234 .sync_dirty_debug_regs = svm_sync_dirty_debug_regs,
4235 .cache_reg = svm_cache_reg,
4236 .get_rflags = svm_get_rflags,
4237 .set_rflags = svm_set_rflags,
4239 .tlb_flush_all = svm_flush_tlb,
4240 .tlb_flush_current = svm_flush_tlb,
4241 .tlb_flush_gva = svm_flush_tlb_gva,
4242 .tlb_flush_guest = svm_flush_tlb,
4244 .run = svm_vcpu_run,
4245 .handle_exit = handle_exit,
4246 .skip_emulated_instruction = skip_emulated_instruction,
4247 .update_emulated_instruction = NULL,
4248 .set_interrupt_shadow = svm_set_interrupt_shadow,
4249 .get_interrupt_shadow = svm_get_interrupt_shadow,
4250 .patch_hypercall = svm_patch_hypercall,
4251 .set_irq = svm_set_irq,
4252 .set_nmi = svm_inject_nmi,
4253 .queue_exception = svm_queue_exception,
4254 .cancel_injection = svm_cancel_injection,
4255 .interrupt_allowed = svm_interrupt_allowed,
4256 .nmi_allowed = svm_nmi_allowed,
4257 .get_nmi_mask = svm_get_nmi_mask,
4258 .set_nmi_mask = svm_set_nmi_mask,
4259 .enable_nmi_window = enable_nmi_window,
4260 .enable_irq_window = enable_irq_window,
4261 .update_cr8_intercept = update_cr8_intercept,
4262 .set_virtual_apic_mode = svm_set_virtual_apic_mode,
4263 .refresh_apicv_exec_ctrl = svm_refresh_apicv_exec_ctrl,
4264 .check_apicv_inhibit_reasons = svm_check_apicv_inhibit_reasons,
4265 .pre_update_apicv_exec_ctrl = svm_pre_update_apicv_exec_ctrl,
4266 .load_eoi_exitmap = svm_load_eoi_exitmap,
4267 .hwapic_irr_update = svm_hwapic_irr_update,
4268 .hwapic_isr_update = svm_hwapic_isr_update,
4269 .sync_pir_to_irr = kvm_lapic_find_highest_irr,
4270 .apicv_post_state_restore = avic_post_state_restore,
4272 .set_tss_addr = svm_set_tss_addr,
4273 .set_identity_map_addr = svm_set_identity_map_addr,
4274 .get_mt_mask = svm_get_mt_mask,
4276 .get_exit_info = svm_get_exit_info,
4278 .vcpu_after_set_cpuid = svm_vcpu_after_set_cpuid,
4280 .has_wbinvd_exit = svm_has_wbinvd_exit,
4282 .write_l1_tsc_offset = svm_write_l1_tsc_offset,
4284 .load_mmu_pgd = svm_load_mmu_pgd,
4286 .check_intercept = svm_check_intercept,
4287 .handle_exit_irqoff = svm_handle_exit_irqoff,
4289 .request_immediate_exit = __kvm_request_immediate_exit,
4291 .sched_in = svm_sched_in,
4293 .pmu_ops = &amd_pmu_ops,
4294 .nested_ops = &svm_nested_ops,
4296 .deliver_posted_interrupt = svm_deliver_avic_intr,
4297 .dy_apicv_has_pending_interrupt = svm_dy_apicv_has_pending_interrupt,
4298 .update_pi_irte = svm_update_pi_irte,
4299 .setup_mce = svm_setup_mce,
4301 .smi_allowed = svm_smi_allowed,
4302 .pre_enter_smm = svm_pre_enter_smm,
4303 .pre_leave_smm = svm_pre_leave_smm,
4304 .enable_smi_window = enable_smi_window,
4306 .mem_enc_op = svm_mem_enc_op,
4307 .mem_enc_reg_region = svm_register_enc_region,
4308 .mem_enc_unreg_region = svm_unregister_enc_region,
4310 .can_emulate_instruction = svm_can_emulate_instruction,
4312 .apic_init_signal_blocked = svm_apic_init_signal_blocked,
4314 .msr_filter_changed = svm_msr_filter_changed,
4317 static struct kvm_x86_init_ops svm_init_ops __initdata = {
4318 .cpu_has_kvm_support = has_svm,
4319 .disabled_by_bios = is_disabled,
4320 .hardware_setup = svm_hardware_setup,
4321 .check_processor_compatibility = svm_check_processor_compat,
4323 .runtime_ops = &svm_x86_ops,
4326 static int __init svm_init(void)
4328 __unused_size_checks();
4330 return kvm_init(&svm_init_ops, sizeof(struct vcpu_svm),
4331 __alignof__(struct vcpu_svm), THIS_MODULE);
4334 static void __exit svm_exit(void)
4339 module_init(svm_init)
4340 module_exit(svm_exit)