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>
36 #include <asm/spec-ctrl.h>
37 #include <asm/cpu_device_id.h>
38 #include <asm/traps.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 initially cleared */
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_EFER, .always = false },
112 { .index = MSR_IA32_CR_PAT, .always = false },
113 { .index = MSR_AMD64_SEV_ES_GHCB, .always = true },
114 { .index = MSR_INVALID, .always = false },
117 /* enable NPT for AMD64 and X86 with PAE */
118 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
119 bool npt_enabled = true;
125 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
126 * pause_filter_count: On processors that support Pause filtering(indicated
127 * by CPUID Fn8000_000A_EDX), the VMCB provides a 16 bit pause filter
128 * count value. On VMRUN this value is loaded into an internal counter.
129 * Each time a pause instruction is executed, this counter is decremented
130 * until it reaches zero at which time a #VMEXIT is generated if pause
131 * intercept is enabled. Refer to AMD APM Vol 2 Section 15.14.4 Pause
132 * Intercept Filtering for more details.
133 * This also indicate if ple logic enabled.
135 * pause_filter_thresh: In addition, some processor families support advanced
136 * pause filtering (indicated by CPUID Fn8000_000A_EDX) upper bound on
137 * the amount of time a guest is allowed to execute in a pause loop.
138 * In this mode, a 16-bit pause filter threshold field is added in the
139 * VMCB. The threshold value is a cycle count that is used to reset the
140 * pause counter. As with simple pause filtering, VMRUN loads the pause
141 * count value from VMCB into an internal counter. Then, on each pause
142 * instruction the hardware checks the elapsed number of cycles since
143 * the most recent pause instruction against the pause filter threshold.
144 * If the elapsed cycle count is greater than the pause filter threshold,
145 * then the internal pause count is reloaded from the VMCB and execution
146 * continues. If the elapsed cycle count is less than the pause filter
147 * threshold, then the internal pause count is decremented. If the count
148 * value is less than zero and PAUSE intercept is enabled, a #VMEXIT is
149 * triggered. If advanced pause filtering is supported and pause filter
150 * threshold field is set to zero, the filter will operate in the simpler,
154 static unsigned short pause_filter_thresh = KVM_DEFAULT_PLE_GAP;
155 module_param(pause_filter_thresh, ushort, 0444);
157 static unsigned short pause_filter_count = KVM_SVM_DEFAULT_PLE_WINDOW;
158 module_param(pause_filter_count, ushort, 0444);
160 /* Default doubles per-vcpu window every exit. */
161 static unsigned short pause_filter_count_grow = KVM_DEFAULT_PLE_WINDOW_GROW;
162 module_param(pause_filter_count_grow, ushort, 0444);
164 /* Default resets per-vcpu window every exit to pause_filter_count. */
165 static unsigned short pause_filter_count_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK;
166 module_param(pause_filter_count_shrink, ushort, 0444);
168 /* Default is to compute the maximum so we can never overflow. */
169 static unsigned short pause_filter_count_max = KVM_SVM_DEFAULT_PLE_WINDOW_MAX;
170 module_param(pause_filter_count_max, ushort, 0444);
172 /* allow nested paging (virtualized MMU) for all guests */
173 static int npt = true;
174 module_param(npt, int, S_IRUGO);
176 /* allow nested virtualization in KVM/SVM */
177 static int nested = true;
178 module_param(nested, int, S_IRUGO);
180 /* enable/disable Next RIP Save */
181 static int nrips = true;
182 module_param(nrips, int, 0444);
184 /* enable/disable Virtual VMLOAD VMSAVE */
185 static int vls = true;
186 module_param(vls, int, 0444);
188 /* enable/disable Virtual GIF */
189 static int vgif = true;
190 module_param(vgif, int, 0444);
192 /* enable/disable SEV support */
193 int sev = IS_ENABLED(CONFIG_AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT);
194 module_param(sev, int, 0444);
196 /* enable/disable SEV-ES support */
197 int sev_es = IS_ENABLED(CONFIG_AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT);
198 module_param(sev_es, int, 0444);
200 bool __read_mostly dump_invalid_vmcb;
201 module_param(dump_invalid_vmcb, bool, 0644);
203 static u8 rsm_ins_bytes[] = "\x0f\xaa";
205 static void svm_complete_interrupts(struct vcpu_svm *svm);
207 static unsigned long iopm_base;
209 struct kvm_ldttss_desc {
212 unsigned base1:8, type:5, dpl:2, p:1;
213 unsigned limit1:4, zero0:3, g:1, base2:8;
216 } __attribute__((packed));
218 DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
220 static const u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
222 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
223 #define MSRS_RANGE_SIZE 2048
224 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
226 u32 svm_msrpm_offset(u32 msr)
231 for (i = 0; i < NUM_MSR_MAPS; i++) {
232 if (msr < msrpm_ranges[i] ||
233 msr >= msrpm_ranges[i] + MSRS_IN_RANGE)
236 offset = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */
237 offset += (i * MSRS_RANGE_SIZE); /* add range offset */
239 /* Now we have the u8 offset - but need the u32 offset */
243 /* MSR not in any range */
247 #define MAX_INST_SIZE 15
249 static inline void clgi(void)
251 asm volatile (__ex("clgi"));
254 static inline void stgi(void)
256 asm volatile (__ex("stgi"));
259 static inline void invlpga(unsigned long addr, u32 asid)
261 asm volatile (__ex("invlpga %1, %0") : : "c"(asid), "a"(addr));
264 static int get_max_npt_level(void)
267 return PT64_ROOT_4LEVEL;
269 return PT32E_ROOT_LEVEL;
273 int svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
275 struct vcpu_svm *svm = to_svm(vcpu);
276 u64 old_efer = vcpu->arch.efer;
277 vcpu->arch.efer = efer;
280 /* Shadow paging assumes NX to be available. */
283 if (!(efer & EFER_LMA))
287 if ((old_efer & EFER_SVME) != (efer & EFER_SVME)) {
288 if (!(efer & EFER_SVME)) {
289 svm_leave_nested(svm);
290 svm_set_gif(svm, true);
293 * Free the nested guest state, unless we are in SMM.
294 * In this case we will return to the nested guest
295 * as soon as we leave SMM.
297 if (!is_smm(&svm->vcpu))
298 svm_free_nested(svm);
301 int ret = svm_allocate_nested(svm);
304 vcpu->arch.efer = old_efer;
310 svm->vmcb->save.efer = efer | EFER_SVME;
311 vmcb_mark_dirty(svm->vmcb, VMCB_CR);
315 static int is_external_interrupt(u32 info)
317 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
318 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
321 static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu)
323 struct vcpu_svm *svm = to_svm(vcpu);
326 if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
327 ret = KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS;
331 static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
333 struct vcpu_svm *svm = to_svm(vcpu);
336 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
338 svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
342 static int skip_emulated_instruction(struct kvm_vcpu *vcpu)
344 struct vcpu_svm *svm = to_svm(vcpu);
347 * SEV-ES does not expose the next RIP. The RIP update is controlled by
348 * the type of exit and the #VC handler in the guest.
350 if (sev_es_guest(vcpu->kvm))
353 if (nrips && svm->vmcb->control.next_rip != 0) {
354 WARN_ON_ONCE(!static_cpu_has(X86_FEATURE_NRIPS));
355 svm->next_rip = svm->vmcb->control.next_rip;
358 if (!svm->next_rip) {
359 if (!kvm_emulate_instruction(vcpu, EMULTYPE_SKIP))
362 kvm_rip_write(vcpu, svm->next_rip);
366 svm_set_interrupt_shadow(vcpu, 0);
371 static void svm_queue_exception(struct kvm_vcpu *vcpu)
373 struct vcpu_svm *svm = to_svm(vcpu);
374 unsigned nr = vcpu->arch.exception.nr;
375 bool has_error_code = vcpu->arch.exception.has_error_code;
376 u32 error_code = vcpu->arch.exception.error_code;
378 kvm_deliver_exception_payload(&svm->vcpu);
380 if (nr == BP_VECTOR && !nrips) {
381 unsigned long rip, old_rip = kvm_rip_read(&svm->vcpu);
384 * For guest debugging where we have to reinject #BP if some
385 * INT3 is guest-owned:
386 * Emulate nRIP by moving RIP forward. Will fail if injection
387 * raises a fault that is not intercepted. Still better than
388 * failing in all cases.
390 (void)skip_emulated_instruction(&svm->vcpu);
391 rip = kvm_rip_read(&svm->vcpu);
392 svm->int3_rip = rip + svm->vmcb->save.cs.base;
393 svm->int3_injected = rip - old_rip;
396 svm->vmcb->control.event_inj = nr
398 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
399 | SVM_EVTINJ_TYPE_EXEPT;
400 svm->vmcb->control.event_inj_err = error_code;
403 static void svm_init_erratum_383(void)
409 if (!static_cpu_has_bug(X86_BUG_AMD_TLB_MMATCH))
412 /* Use _safe variants to not break nested virtualization */
413 val = native_read_msr_safe(MSR_AMD64_DC_CFG, &err);
419 low = lower_32_bits(val);
420 high = upper_32_bits(val);
422 native_write_msr_safe(MSR_AMD64_DC_CFG, low, high);
424 erratum_383_found = true;
427 static void svm_init_osvw(struct kvm_vcpu *vcpu)
430 * Guests should see errata 400 and 415 as fixed (assuming that
431 * HLT and IO instructions are intercepted).
433 vcpu->arch.osvw.length = (osvw_len >= 3) ? (osvw_len) : 3;
434 vcpu->arch.osvw.status = osvw_status & ~(6ULL);
437 * By increasing VCPU's osvw.length to 3 we are telling the guest that
438 * all osvw.status bits inside that length, including bit 0 (which is
439 * reserved for erratum 298), are valid. However, if host processor's
440 * osvw_len is 0 then osvw_status[0] carries no information. We need to
441 * be conservative here and therefore we tell the guest that erratum 298
442 * is present (because we really don't know).
444 if (osvw_len == 0 && boot_cpu_data.x86 == 0x10)
445 vcpu->arch.osvw.status |= 1;
448 static int has_svm(void)
452 if (!cpu_has_svm(&msg)) {
453 printk(KERN_INFO "has_svm: %s\n", msg);
460 static void svm_hardware_disable(void)
462 /* Make sure we clean up behind us */
463 if (static_cpu_has(X86_FEATURE_TSCRATEMSR))
464 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
468 amd_pmu_disable_virt();
471 static int svm_hardware_enable(void)
474 struct svm_cpu_data *sd;
476 struct desc_struct *gdt;
477 int me = raw_smp_processor_id();
479 rdmsrl(MSR_EFER, efer);
480 if (efer & EFER_SVME)
484 pr_err("%s: err EOPNOTSUPP on %d\n", __func__, me);
487 sd = per_cpu(svm_data, me);
489 pr_err("%s: svm_data is NULL on %d\n", __func__, me);
493 sd->asid_generation = 1;
494 sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
495 sd->next_asid = sd->max_asid + 1;
496 sd->min_asid = max_sev_asid + 1;
498 gdt = get_current_gdt_rw();
499 sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
501 wrmsrl(MSR_EFER, efer | EFER_SVME);
503 wrmsrl(MSR_VM_HSAVE_PA, __sme_page_pa(sd->save_area));
505 if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
506 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
507 __this_cpu_write(current_tsc_ratio, TSC_RATIO_DEFAULT);
514 * Note that it is possible to have a system with mixed processor
515 * revisions and therefore different OSVW bits. If bits are not the same
516 * on different processors then choose the worst case (i.e. if erratum
517 * is present on one processor and not on another then assume that the
518 * erratum is present everywhere).
520 if (cpu_has(&boot_cpu_data, X86_FEATURE_OSVW)) {
521 uint64_t len, status = 0;
524 len = native_read_msr_safe(MSR_AMD64_OSVW_ID_LENGTH, &err);
526 status = native_read_msr_safe(MSR_AMD64_OSVW_STATUS,
530 osvw_status = osvw_len = 0;
534 osvw_status |= status;
535 osvw_status &= (1ULL << osvw_len) - 1;
538 osvw_status = osvw_len = 0;
540 svm_init_erratum_383();
542 amd_pmu_enable_virt();
547 static void svm_cpu_uninit(int cpu)
549 struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id());
554 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
555 kfree(sd->sev_vmcbs);
556 __free_page(sd->save_area);
560 static int svm_cpu_init(int cpu)
562 struct svm_cpu_data *sd;
564 sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
568 sd->save_area = alloc_page(GFP_KERNEL);
571 clear_page(page_address(sd->save_area));
573 if (svm_sev_enabled()) {
574 sd->sev_vmcbs = kmalloc_array(max_sev_asid + 1,
581 per_cpu(svm_data, cpu) = sd;
586 __free_page(sd->save_area);
593 static int direct_access_msr_slot(u32 msr)
597 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++)
598 if (direct_access_msrs[i].index == msr)
604 static void set_shadow_msr_intercept(struct kvm_vcpu *vcpu, u32 msr, int read,
607 struct vcpu_svm *svm = to_svm(vcpu);
608 int slot = direct_access_msr_slot(msr);
613 /* Set the shadow bitmaps to the desired intercept states */
615 set_bit(slot, svm->shadow_msr_intercept.read);
617 clear_bit(slot, svm->shadow_msr_intercept.read);
620 set_bit(slot, svm->shadow_msr_intercept.write);
622 clear_bit(slot, svm->shadow_msr_intercept.write);
625 static bool valid_msr_intercept(u32 index)
627 return direct_access_msr_slot(index) != -ENOENT;
630 static bool msr_write_intercepted(struct kvm_vcpu *vcpu, u32 msr)
637 msrpm = is_guest_mode(vcpu) ? to_svm(vcpu)->nested.msrpm:
640 offset = svm_msrpm_offset(msr);
641 bit_write = 2 * (msr & 0x0f) + 1;
644 BUG_ON(offset == MSR_INVALID);
646 return !!test_bit(bit_write, &tmp);
649 static void set_msr_interception_bitmap(struct kvm_vcpu *vcpu, u32 *msrpm,
650 u32 msr, int read, int write)
652 u8 bit_read, bit_write;
657 * If this warning triggers extend the direct_access_msrs list at the
658 * beginning of the file
660 WARN_ON(!valid_msr_intercept(msr));
662 /* Enforce non allowed MSRs to trap */
663 if (read && !kvm_msr_allowed(vcpu, msr, KVM_MSR_FILTER_READ))
666 if (write && !kvm_msr_allowed(vcpu, msr, KVM_MSR_FILTER_WRITE))
669 offset = svm_msrpm_offset(msr);
670 bit_read = 2 * (msr & 0x0f);
671 bit_write = 2 * (msr & 0x0f) + 1;
674 BUG_ON(offset == MSR_INVALID);
676 read ? clear_bit(bit_read, &tmp) : set_bit(bit_read, &tmp);
677 write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp);
682 void set_msr_interception(struct kvm_vcpu *vcpu, u32 *msrpm, u32 msr,
685 set_shadow_msr_intercept(vcpu, msr, read, write);
686 set_msr_interception_bitmap(vcpu, msrpm, msr, read, write);
689 u32 *svm_vcpu_alloc_msrpm(void)
691 struct page *pages = alloc_pages(GFP_KERNEL_ACCOUNT, MSRPM_ALLOC_ORDER);
697 msrpm = page_address(pages);
698 memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
703 void svm_vcpu_init_msrpm(struct kvm_vcpu *vcpu, u32 *msrpm)
707 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
708 if (!direct_access_msrs[i].always)
710 set_msr_interception(vcpu, msrpm, direct_access_msrs[i].index, 1, 1);
715 void svm_vcpu_free_msrpm(u32 *msrpm)
717 __free_pages(virt_to_page(msrpm), MSRPM_ALLOC_ORDER);
720 static void svm_msr_filter_changed(struct kvm_vcpu *vcpu)
722 struct vcpu_svm *svm = to_svm(vcpu);
726 * Set intercept permissions for all direct access MSRs again. They
727 * will automatically get filtered through the MSR filter, so we are
728 * back in sync after this.
730 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
731 u32 msr = direct_access_msrs[i].index;
732 u32 read = test_bit(i, svm->shadow_msr_intercept.read);
733 u32 write = test_bit(i, svm->shadow_msr_intercept.write);
735 set_msr_interception_bitmap(vcpu, svm->msrpm, msr, read, write);
739 static void add_msr_offset(u32 offset)
743 for (i = 0; i < MSRPM_OFFSETS; ++i) {
745 /* Offset already in list? */
746 if (msrpm_offsets[i] == offset)
749 /* Slot used by another offset? */
750 if (msrpm_offsets[i] != MSR_INVALID)
753 /* Add offset to list */
754 msrpm_offsets[i] = offset;
760 * If this BUG triggers the msrpm_offsets table has an overflow. Just
761 * increase MSRPM_OFFSETS in this case.
766 static void init_msrpm_offsets(void)
770 memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets));
772 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
775 offset = svm_msrpm_offset(direct_access_msrs[i].index);
776 BUG_ON(offset == MSR_INVALID);
778 add_msr_offset(offset);
782 static void svm_enable_lbrv(struct kvm_vcpu *vcpu)
784 struct vcpu_svm *svm = to_svm(vcpu);
786 svm->vmcb->control.virt_ext |= LBR_CTL_ENABLE_MASK;
787 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
788 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
789 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
790 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
793 static void svm_disable_lbrv(struct kvm_vcpu *vcpu)
795 struct vcpu_svm *svm = to_svm(vcpu);
797 svm->vmcb->control.virt_ext &= ~LBR_CTL_ENABLE_MASK;
798 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
799 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
800 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
801 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
804 void disable_nmi_singlestep(struct vcpu_svm *svm)
806 svm->nmi_singlestep = false;
808 if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP)) {
809 /* Clear our flags if they were not set by the guest */
810 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF))
811 svm->vmcb->save.rflags &= ~X86_EFLAGS_TF;
812 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF))
813 svm->vmcb->save.rflags &= ~X86_EFLAGS_RF;
817 static void grow_ple_window(struct kvm_vcpu *vcpu)
819 struct vcpu_svm *svm = to_svm(vcpu);
820 struct vmcb_control_area *control = &svm->vmcb->control;
821 int old = control->pause_filter_count;
823 control->pause_filter_count = __grow_ple_window(old,
825 pause_filter_count_grow,
826 pause_filter_count_max);
828 if (control->pause_filter_count != old) {
829 vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
830 trace_kvm_ple_window_update(vcpu->vcpu_id,
831 control->pause_filter_count, old);
835 static void shrink_ple_window(struct kvm_vcpu *vcpu)
837 struct vcpu_svm *svm = to_svm(vcpu);
838 struct vmcb_control_area *control = &svm->vmcb->control;
839 int old = control->pause_filter_count;
841 control->pause_filter_count =
842 __shrink_ple_window(old,
844 pause_filter_count_shrink,
846 if (control->pause_filter_count != old) {
847 vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
848 trace_kvm_ple_window_update(vcpu->vcpu_id,
849 control->pause_filter_count, old);
854 * The default MMIO mask is a single bit (excluding the present bit),
855 * which could conflict with the memory encryption bit. Check for
856 * memory encryption support and override the default MMIO mask if
857 * memory encryption is enabled.
859 static __init void svm_adjust_mmio_mask(void)
861 unsigned int enc_bit, mask_bit;
864 /* If there is no memory encryption support, use existing mask */
865 if (cpuid_eax(0x80000000) < 0x8000001f)
868 /* If memory encryption is not enabled, use existing mask */
869 rdmsrl(MSR_K8_SYSCFG, msr);
870 if (!(msr & MSR_K8_SYSCFG_MEM_ENCRYPT))
873 enc_bit = cpuid_ebx(0x8000001f) & 0x3f;
874 mask_bit = boot_cpu_data.x86_phys_bits;
876 /* Increment the mask bit if it is the same as the encryption bit */
877 if (enc_bit == mask_bit)
881 * If the mask bit location is below 52, then some bits above the
882 * physical addressing limit will always be reserved, so use the
883 * rsvd_bits() function to generate the mask. This mask, along with
884 * the present bit, will be used to generate a page fault with
887 * If the mask bit location is 52 (or above), then clear the mask.
889 mask = (mask_bit < 52) ? rsvd_bits(mask_bit, 51) | PT_PRESENT_MASK : 0;
891 kvm_mmu_set_mmio_spte_mask(mask, PT_WRITABLE_MASK | PT_USER_MASK);
894 static void svm_hardware_teardown(void)
898 if (svm_sev_enabled())
899 sev_hardware_teardown();
901 for_each_possible_cpu(cpu)
904 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
908 static __init void svm_set_cpu_caps(void)
914 /* CPUID 0x80000001 and 0x8000000A (SVM features) */
916 kvm_cpu_cap_set(X86_FEATURE_SVM);
919 kvm_cpu_cap_set(X86_FEATURE_NRIPS);
922 kvm_cpu_cap_set(X86_FEATURE_NPT);
925 /* CPUID 0x80000008 */
926 if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) ||
927 boot_cpu_has(X86_FEATURE_AMD_SSBD))
928 kvm_cpu_cap_set(X86_FEATURE_VIRT_SSBD);
930 /* Enable INVPCID feature */
931 kvm_cpu_cap_check_and_set(X86_FEATURE_INVPCID);
934 static __init int svm_hardware_setup(void)
937 struct page *iopm_pages;
941 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
946 iopm_va = page_address(iopm_pages);
947 memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
948 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
950 init_msrpm_offsets();
952 supported_xcr0 &= ~(XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
954 if (boot_cpu_has(X86_FEATURE_NX))
955 kvm_enable_efer_bits(EFER_NX);
957 if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
958 kvm_enable_efer_bits(EFER_FFXSR);
960 if (boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
961 kvm_has_tsc_control = true;
962 kvm_max_tsc_scaling_ratio = TSC_RATIO_MAX;
963 kvm_tsc_scaling_ratio_frac_bits = 32;
966 /* Check for pause filtering support */
967 if (!boot_cpu_has(X86_FEATURE_PAUSEFILTER)) {
968 pause_filter_count = 0;
969 pause_filter_thresh = 0;
970 } else if (!boot_cpu_has(X86_FEATURE_PFTHRESHOLD)) {
971 pause_filter_thresh = 0;
975 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
976 kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE);
979 if (IS_ENABLED(CONFIG_KVM_AMD_SEV) && sev) {
980 sev_hardware_setup();
986 svm_adjust_mmio_mask();
988 for_each_possible_cpu(cpu) {
989 r = svm_cpu_init(cpu);
994 if (!boot_cpu_has(X86_FEATURE_NPT))
997 if (npt_enabled && !npt)
1000 kvm_configure_mmu(npt_enabled, get_max_npt_level(), PG_LEVEL_1G);
1001 pr_info("kvm: Nested Paging %sabled\n", npt_enabled ? "en" : "dis");
1004 if (!boot_cpu_has(X86_FEATURE_NRIPS))
1010 !boot_cpu_has(X86_FEATURE_AVIC) ||
1011 !IS_ENABLED(CONFIG_X86_LOCAL_APIC)) {
1014 pr_info("AVIC enabled\n");
1016 amd_iommu_register_ga_log_notifier(&avic_ga_log_notifier);
1022 !boot_cpu_has(X86_FEATURE_V_VMSAVE_VMLOAD) ||
1023 !IS_ENABLED(CONFIG_X86_64)) {
1026 pr_info("Virtual VMLOAD VMSAVE supported\n");
1031 if (!boot_cpu_has(X86_FEATURE_VGIF))
1034 pr_info("Virtual GIF supported\n");
1040 * It seems that on AMD processors PTE's accessed bit is
1041 * being set by the CPU hardware before the NPF vmexit.
1042 * This is not expected behaviour and our tests fail because
1044 * A workaround here is to disable support for
1045 * GUEST_MAXPHYADDR < HOST_MAXPHYADDR if NPT is enabled.
1046 * In this case userspace can know if there is support using
1047 * KVM_CAP_SMALLER_MAXPHYADDR extension and decide how to handle
1049 * If future AMD CPU models change the behaviour described above,
1050 * this variable can be changed accordingly
1052 allow_smaller_maxphyaddr = !npt_enabled;
1057 svm_hardware_teardown();
1061 static void init_seg(struct vmcb_seg *seg)
1064 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
1065 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
1066 seg->limit = 0xffff;
1070 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
1073 seg->attrib = SVM_SELECTOR_P_MASK | type;
1074 seg->limit = 0xffff;
1078 static u64 svm_write_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1080 struct vcpu_svm *svm = to_svm(vcpu);
1081 u64 g_tsc_offset = 0;
1083 if (is_guest_mode(vcpu)) {
1084 /* Write L1's TSC offset. */
1085 g_tsc_offset = svm->vmcb->control.tsc_offset -
1086 svm->nested.hsave->control.tsc_offset;
1087 svm->nested.hsave->control.tsc_offset = offset;
1090 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
1091 svm->vmcb->control.tsc_offset - g_tsc_offset,
1094 svm->vmcb->control.tsc_offset = offset + g_tsc_offset;
1096 vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
1097 return svm->vmcb->control.tsc_offset;
1100 static void svm_check_invpcid(struct vcpu_svm *svm)
1103 * Intercept INVPCID instruction only if shadow page table is
1104 * enabled. Interception is not required with nested page table
1107 if (kvm_cpu_cap_has(X86_FEATURE_INVPCID)) {
1109 svm_set_intercept(svm, INTERCEPT_INVPCID);
1111 svm_clr_intercept(svm, INTERCEPT_INVPCID);
1115 static void init_vmcb(struct vcpu_svm *svm)
1117 struct vmcb_control_area *control = &svm->vmcb->control;
1118 struct vmcb_save_area *save = &svm->vmcb->save;
1120 svm->vcpu.arch.hflags = 0;
1122 svm_set_intercept(svm, INTERCEPT_CR0_READ);
1123 svm_set_intercept(svm, INTERCEPT_CR3_READ);
1124 svm_set_intercept(svm, INTERCEPT_CR4_READ);
1125 svm_set_intercept(svm, INTERCEPT_CR0_WRITE);
1126 svm_set_intercept(svm, INTERCEPT_CR3_WRITE);
1127 svm_set_intercept(svm, INTERCEPT_CR4_WRITE);
1128 if (!kvm_vcpu_apicv_active(&svm->vcpu))
1129 svm_set_intercept(svm, INTERCEPT_CR8_WRITE);
1131 set_dr_intercepts(svm);
1133 set_exception_intercept(svm, PF_VECTOR);
1134 set_exception_intercept(svm, UD_VECTOR);
1135 set_exception_intercept(svm, MC_VECTOR);
1136 set_exception_intercept(svm, AC_VECTOR);
1137 set_exception_intercept(svm, DB_VECTOR);
1139 * Guest access to VMware backdoor ports could legitimately
1140 * trigger #GP because of TSS I/O permission bitmap.
1141 * We intercept those #GP and allow access to them anyway
1144 if (enable_vmware_backdoor)
1145 set_exception_intercept(svm, GP_VECTOR);
1147 svm_set_intercept(svm, INTERCEPT_INTR);
1148 svm_set_intercept(svm, INTERCEPT_NMI);
1149 svm_set_intercept(svm, INTERCEPT_SMI);
1150 svm_set_intercept(svm, INTERCEPT_SELECTIVE_CR0);
1151 svm_set_intercept(svm, INTERCEPT_RDPMC);
1152 svm_set_intercept(svm, INTERCEPT_CPUID);
1153 svm_set_intercept(svm, INTERCEPT_INVD);
1154 svm_set_intercept(svm, INTERCEPT_INVLPG);
1155 svm_set_intercept(svm, INTERCEPT_INVLPGA);
1156 svm_set_intercept(svm, INTERCEPT_IOIO_PROT);
1157 svm_set_intercept(svm, INTERCEPT_MSR_PROT);
1158 svm_set_intercept(svm, INTERCEPT_TASK_SWITCH);
1159 svm_set_intercept(svm, INTERCEPT_SHUTDOWN);
1160 svm_set_intercept(svm, INTERCEPT_VMRUN);
1161 svm_set_intercept(svm, INTERCEPT_VMMCALL);
1162 svm_set_intercept(svm, INTERCEPT_VMLOAD);
1163 svm_set_intercept(svm, INTERCEPT_VMSAVE);
1164 svm_set_intercept(svm, INTERCEPT_STGI);
1165 svm_set_intercept(svm, INTERCEPT_CLGI);
1166 svm_set_intercept(svm, INTERCEPT_SKINIT);
1167 svm_set_intercept(svm, INTERCEPT_WBINVD);
1168 svm_set_intercept(svm, INTERCEPT_XSETBV);
1169 svm_set_intercept(svm, INTERCEPT_RDPRU);
1170 svm_set_intercept(svm, INTERCEPT_RSM);
1172 if (!kvm_mwait_in_guest(svm->vcpu.kvm)) {
1173 svm_set_intercept(svm, INTERCEPT_MONITOR);
1174 svm_set_intercept(svm, INTERCEPT_MWAIT);
1177 if (!kvm_hlt_in_guest(svm->vcpu.kvm))
1178 svm_set_intercept(svm, INTERCEPT_HLT);
1180 control->iopm_base_pa = __sme_set(iopm_base);
1181 control->msrpm_base_pa = __sme_set(__pa(svm->msrpm));
1182 control->int_ctl = V_INTR_MASKING_MASK;
1184 init_seg(&save->es);
1185 init_seg(&save->ss);
1186 init_seg(&save->ds);
1187 init_seg(&save->fs);
1188 init_seg(&save->gs);
1190 save->cs.selector = 0xf000;
1191 save->cs.base = 0xffff0000;
1192 /* Executable/Readable Code Segment */
1193 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
1194 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
1195 save->cs.limit = 0xffff;
1197 save->gdtr.limit = 0xffff;
1198 save->idtr.limit = 0xffff;
1200 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
1201 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
1203 svm_set_efer(&svm->vcpu, 0);
1204 save->dr6 = 0xffff0ff0;
1205 kvm_set_rflags(&svm->vcpu, 2);
1206 save->rip = 0x0000fff0;
1207 svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
1210 * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
1211 * It also updates the guest-visible cr0 value.
1213 svm_set_cr0(&svm->vcpu, X86_CR0_NW | X86_CR0_CD | X86_CR0_ET);
1214 kvm_mmu_reset_context(&svm->vcpu);
1216 save->cr4 = X86_CR4_PAE;
1220 /* Setup VMCB for Nested Paging */
1221 control->nested_ctl |= SVM_NESTED_CTL_NP_ENABLE;
1222 svm_clr_intercept(svm, INTERCEPT_INVLPG);
1223 clr_exception_intercept(svm, PF_VECTOR);
1224 svm_clr_intercept(svm, INTERCEPT_CR3_READ);
1225 svm_clr_intercept(svm, INTERCEPT_CR3_WRITE);
1226 save->g_pat = svm->vcpu.arch.pat;
1230 svm->asid_generation = 0;
1233 svm->nested.vmcb12_gpa = 0;
1234 svm->vcpu.arch.hflags = 0;
1236 if (!kvm_pause_in_guest(svm->vcpu.kvm)) {
1237 control->pause_filter_count = pause_filter_count;
1238 if (pause_filter_thresh)
1239 control->pause_filter_thresh = pause_filter_thresh;
1240 svm_set_intercept(svm, INTERCEPT_PAUSE);
1242 svm_clr_intercept(svm, INTERCEPT_PAUSE);
1245 svm_check_invpcid(svm);
1247 if (kvm_vcpu_apicv_active(&svm->vcpu))
1248 avic_init_vmcb(svm);
1251 * If hardware supports Virtual VMLOAD VMSAVE then enable it
1252 * in VMCB and clear intercepts to avoid #VMEXIT.
1255 svm_clr_intercept(svm, INTERCEPT_VMLOAD);
1256 svm_clr_intercept(svm, INTERCEPT_VMSAVE);
1257 svm->vmcb->control.virt_ext |= VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK;
1261 svm_clr_intercept(svm, INTERCEPT_STGI);
1262 svm_clr_intercept(svm, INTERCEPT_CLGI);
1263 svm->vmcb->control.int_ctl |= V_GIF_ENABLE_MASK;
1266 if (sev_guest(svm->vcpu.kvm)) {
1267 svm->vmcb->control.nested_ctl |= SVM_NESTED_CTL_SEV_ENABLE;
1268 clr_exception_intercept(svm, UD_VECTOR);
1270 if (sev_es_guest(svm->vcpu.kvm)) {
1271 /* Perform SEV-ES specific VMCB updates */
1272 sev_es_init_vmcb(svm);
1276 vmcb_mark_all_dirty(svm->vmcb);
1282 static void svm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
1284 struct vcpu_svm *svm = to_svm(vcpu);
1289 svm->virt_spec_ctrl = 0;
1292 svm->vcpu.arch.apic_base = APIC_DEFAULT_PHYS_BASE |
1293 MSR_IA32_APICBASE_ENABLE;
1294 if (kvm_vcpu_is_reset_bsp(&svm->vcpu))
1295 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
1299 kvm_cpuid(vcpu, &eax, &dummy, &dummy, &dummy, false);
1300 kvm_rdx_write(vcpu, eax);
1302 if (kvm_vcpu_apicv_active(vcpu) && !init_event)
1303 avic_update_vapic_bar(svm, APIC_DEFAULT_PHYS_BASE);
1306 static int svm_create_vcpu(struct kvm_vcpu *vcpu)
1308 struct vcpu_svm *svm;
1309 struct page *vmcb_page;
1310 struct page *vmsa_page = NULL;
1313 BUILD_BUG_ON(offsetof(struct vcpu_svm, vcpu) != 0);
1317 vmcb_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
1321 if (sev_es_guest(svm->vcpu.kvm)) {
1323 * SEV-ES guests require a separate VMSA page used to contain
1324 * the encrypted register state of the guest.
1326 vmsa_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
1328 goto error_free_vmcb_page;
1331 * SEV-ES guests maintain an encrypted version of their FPU
1332 * state which is restored and saved on VMRUN and VMEXIT.
1333 * Free the fpu structure to prevent KVM from attempting to
1334 * access the FPU state.
1336 kvm_free_guest_fpu(vcpu);
1339 err = avic_init_vcpu(svm);
1341 goto error_free_vmsa_page;
1343 /* We initialize this flag to true to make sure that the is_running
1344 * bit would be set the first time the vcpu is loaded.
1346 if (irqchip_in_kernel(vcpu->kvm) && kvm_apicv_activated(vcpu->kvm))
1347 svm->avic_is_running = true;
1349 svm->msrpm = svm_vcpu_alloc_msrpm();
1351 goto error_free_vmsa_page;
1353 svm_vcpu_init_msrpm(vcpu, svm->msrpm);
1355 svm->vmcb = page_address(vmcb_page);
1356 svm->vmcb_pa = __sme_set(page_to_pfn(vmcb_page) << PAGE_SHIFT);
1359 svm->vmsa = page_address(vmsa_page);
1361 svm->asid_generation = 0;
1364 svm_init_osvw(vcpu);
1365 vcpu->arch.microcode_version = 0x01000065;
1367 if (sev_es_guest(svm->vcpu.kvm))
1368 /* Perform SEV-ES specific VMCB creation updates */
1369 sev_es_create_vcpu(svm);
1373 error_free_vmsa_page:
1375 __free_page(vmsa_page);
1376 error_free_vmcb_page:
1377 __free_page(vmcb_page);
1382 static void svm_clear_current_vmcb(struct vmcb *vmcb)
1386 for_each_online_cpu(i)
1387 cmpxchg(&per_cpu(svm_data, i)->current_vmcb, vmcb, NULL);
1390 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
1392 struct vcpu_svm *svm = to_svm(vcpu);
1395 * The vmcb page can be recycled, causing a false negative in
1396 * svm_vcpu_load(). So, ensure that no logical CPU has this
1397 * vmcb page recorded as its current vmcb.
1399 svm_clear_current_vmcb(svm->vmcb);
1401 svm_free_nested(svm);
1403 sev_free_vcpu(vcpu);
1405 __free_page(pfn_to_page(__sme_clr(svm->vmcb_pa) >> PAGE_SHIFT));
1406 __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
1409 static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1411 struct vcpu_svm *svm = to_svm(vcpu);
1412 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
1415 if (unlikely(cpu != vcpu->cpu)) {
1416 svm->asid_generation = 0;
1417 vmcb_mark_all_dirty(svm->vmcb);
1420 if (sev_es_guest(svm->vcpu.kvm)) {
1421 sev_es_vcpu_load(svm, cpu);
1423 #ifdef CONFIG_X86_64
1424 rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host.gs_base);
1426 savesegment(fs, svm->host.fs);
1427 savesegment(gs, svm->host.gs);
1428 svm->host.ldt = kvm_read_ldt();
1430 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1431 rdmsrl(host_save_user_msrs[i].index,
1432 svm->host_user_msrs[i]);
1435 if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
1436 u64 tsc_ratio = vcpu->arch.tsc_scaling_ratio;
1437 if (tsc_ratio != __this_cpu_read(current_tsc_ratio)) {
1438 __this_cpu_write(current_tsc_ratio, tsc_ratio);
1439 wrmsrl(MSR_AMD64_TSC_RATIO, tsc_ratio);
1442 /* This assumes that the kernel never uses MSR_TSC_AUX */
1443 if (static_cpu_has(X86_FEATURE_RDTSCP))
1444 wrmsrl(MSR_TSC_AUX, svm->tsc_aux);
1446 if (sd->current_vmcb != svm->vmcb) {
1447 sd->current_vmcb = svm->vmcb;
1448 indirect_branch_prediction_barrier();
1450 avic_vcpu_load(vcpu, cpu);
1453 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
1455 struct vcpu_svm *svm = to_svm(vcpu);
1458 avic_vcpu_put(vcpu);
1460 ++vcpu->stat.host_state_reload;
1461 if (sev_es_guest(svm->vcpu.kvm)) {
1462 sev_es_vcpu_put(svm);
1464 kvm_load_ldt(svm->host.ldt);
1465 #ifdef CONFIG_X86_64
1466 loadsegment(fs, svm->host.fs);
1467 wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gsbase);
1468 load_gs_index(svm->host.gs);
1470 #ifdef CONFIG_X86_32_LAZY_GS
1471 loadsegment(gs, svm->host.gs);
1475 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
1476 wrmsrl(host_save_user_msrs[i].index,
1477 svm->host_user_msrs[i]);
1481 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
1483 struct vcpu_svm *svm = to_svm(vcpu);
1484 unsigned long rflags = svm->vmcb->save.rflags;
1486 if (svm->nmi_singlestep) {
1487 /* Hide our flags if they were not set by the guest */
1488 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF))
1489 rflags &= ~X86_EFLAGS_TF;
1490 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF))
1491 rflags &= ~X86_EFLAGS_RF;
1496 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1498 if (to_svm(vcpu)->nmi_singlestep)
1499 rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
1502 * Any change of EFLAGS.VM is accompanied by a reload of SS
1503 * (caused by either a task switch or an inter-privilege IRET),
1504 * so we do not need to update the CPL here.
1506 to_svm(vcpu)->vmcb->save.rflags = rflags;
1509 static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1512 case VCPU_EXREG_PDPTR:
1513 BUG_ON(!npt_enabled);
1514 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
1521 static void svm_set_vintr(struct vcpu_svm *svm)
1523 struct vmcb_control_area *control;
1525 /* The following fields are ignored when AVIC is enabled */
1526 WARN_ON(kvm_vcpu_apicv_active(&svm->vcpu));
1527 svm_set_intercept(svm, INTERCEPT_VINTR);
1530 * This is just a dummy VINTR to actually cause a vmexit to happen.
1531 * Actual injection of virtual interrupts happens through EVENTINJ.
1533 control = &svm->vmcb->control;
1534 control->int_vector = 0x0;
1535 control->int_ctl &= ~V_INTR_PRIO_MASK;
1536 control->int_ctl |= V_IRQ_MASK |
1537 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
1538 vmcb_mark_dirty(svm->vmcb, VMCB_INTR);
1541 static void svm_clear_vintr(struct vcpu_svm *svm)
1543 const u32 mask = V_TPR_MASK | V_GIF_ENABLE_MASK | V_GIF_MASK | V_INTR_MASKING_MASK;
1544 svm_clr_intercept(svm, INTERCEPT_VINTR);
1546 /* Drop int_ctl fields related to VINTR injection. */
1547 svm->vmcb->control.int_ctl &= mask;
1548 if (is_guest_mode(&svm->vcpu)) {
1549 svm->nested.hsave->control.int_ctl &= mask;
1551 WARN_ON((svm->vmcb->control.int_ctl & V_TPR_MASK) !=
1552 (svm->nested.ctl.int_ctl & V_TPR_MASK));
1553 svm->vmcb->control.int_ctl |= svm->nested.ctl.int_ctl & ~mask;
1556 vmcb_mark_dirty(svm->vmcb, VMCB_INTR);
1559 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
1561 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1564 case VCPU_SREG_CS: return &save->cs;
1565 case VCPU_SREG_DS: return &save->ds;
1566 case VCPU_SREG_ES: return &save->es;
1567 case VCPU_SREG_FS: return &save->fs;
1568 case VCPU_SREG_GS: return &save->gs;
1569 case VCPU_SREG_SS: return &save->ss;
1570 case VCPU_SREG_TR: return &save->tr;
1571 case VCPU_SREG_LDTR: return &save->ldtr;
1577 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1579 struct vmcb_seg *s = svm_seg(vcpu, seg);
1584 static void svm_get_segment(struct kvm_vcpu *vcpu,
1585 struct kvm_segment *var, int seg)
1587 struct vmcb_seg *s = svm_seg(vcpu, seg);
1589 var->base = s->base;
1590 var->limit = s->limit;
1591 var->selector = s->selector;
1592 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
1593 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
1594 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
1595 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
1596 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
1597 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
1598 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
1601 * AMD CPUs circa 2014 track the G bit for all segments except CS.
1602 * However, the SVM spec states that the G bit is not observed by the
1603 * CPU, and some VMware virtual CPUs drop the G bit for all segments.
1604 * So let's synthesize a legal G bit for all segments, this helps
1605 * running KVM nested. It also helps cross-vendor migration, because
1606 * Intel's vmentry has a check on the 'G' bit.
1608 var->g = s->limit > 0xfffff;
1611 * AMD's VMCB does not have an explicit unusable field, so emulate it
1612 * for cross vendor migration purposes by "not present"
1614 var->unusable = !var->present;
1619 * Work around a bug where the busy flag in the tr selector
1629 * The accessed bit must always be set in the segment
1630 * descriptor cache, although it can be cleared in the
1631 * descriptor, the cached bit always remains at 1. Since
1632 * Intel has a check on this, set it here to support
1633 * cross-vendor migration.
1640 * On AMD CPUs sometimes the DB bit in the segment
1641 * descriptor is left as 1, although the whole segment has
1642 * been made unusable. Clear it here to pass an Intel VMX
1643 * entry check when cross vendor migrating.
1647 /* This is symmetric with svm_set_segment() */
1648 var->dpl = to_svm(vcpu)->vmcb->save.cpl;
1653 static int svm_get_cpl(struct kvm_vcpu *vcpu)
1655 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1660 static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1662 struct vcpu_svm *svm = to_svm(vcpu);
1664 dt->size = svm->vmcb->save.idtr.limit;
1665 dt->address = svm->vmcb->save.idtr.base;
1668 static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1670 struct vcpu_svm *svm = to_svm(vcpu);
1672 svm->vmcb->save.idtr.limit = dt->size;
1673 svm->vmcb->save.idtr.base = dt->address ;
1674 vmcb_mark_dirty(svm->vmcb, VMCB_DT);
1677 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1679 struct vcpu_svm *svm = to_svm(vcpu);
1681 dt->size = svm->vmcb->save.gdtr.limit;
1682 dt->address = svm->vmcb->save.gdtr.base;
1685 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
1687 struct vcpu_svm *svm = to_svm(vcpu);
1689 svm->vmcb->save.gdtr.limit = dt->size;
1690 svm->vmcb->save.gdtr.base = dt->address ;
1691 vmcb_mark_dirty(svm->vmcb, VMCB_DT);
1694 static void update_cr0_intercept(struct vcpu_svm *svm)
1700 * SEV-ES guests must always keep the CR intercepts cleared. CR
1701 * tracking is done using the CR write traps.
1703 if (sev_es_guest(svm->vcpu.kvm))
1706 gcr0 = svm->vcpu.arch.cr0;
1707 hcr0 = &svm->vmcb->save.cr0;
1708 *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
1709 | (gcr0 & SVM_CR0_SELECTIVE_MASK);
1711 vmcb_mark_dirty(svm->vmcb, VMCB_CR);
1713 if (gcr0 == *hcr0) {
1714 svm_clr_intercept(svm, INTERCEPT_CR0_READ);
1715 svm_clr_intercept(svm, INTERCEPT_CR0_WRITE);
1717 svm_set_intercept(svm, INTERCEPT_CR0_READ);
1718 svm_set_intercept(svm, INTERCEPT_CR0_WRITE);
1722 void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1724 struct vcpu_svm *svm = to_svm(vcpu);
1726 #ifdef CONFIG_X86_64
1727 if (vcpu->arch.efer & EFER_LME && !vcpu->arch.guest_state_protected) {
1728 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
1729 vcpu->arch.efer |= EFER_LMA;
1730 svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
1733 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
1734 vcpu->arch.efer &= ~EFER_LMA;
1735 svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
1739 vcpu->arch.cr0 = cr0;
1742 cr0 |= X86_CR0_PG | X86_CR0_WP;
1745 * re-enable caching here because the QEMU bios
1746 * does not do it - this results in some delay at
1749 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
1750 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
1751 svm->vmcb->save.cr0 = cr0;
1752 vmcb_mark_dirty(svm->vmcb, VMCB_CR);
1753 update_cr0_intercept(svm);
1756 static bool svm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1761 void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1763 unsigned long host_cr4_mce = cr4_read_shadow() & X86_CR4_MCE;
1764 unsigned long old_cr4 = vcpu->arch.cr4;
1766 if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
1767 svm_flush_tlb(vcpu);
1769 vcpu->arch.cr4 = cr4;
1772 cr4 |= host_cr4_mce;
1773 to_svm(vcpu)->vmcb->save.cr4 = cr4;
1774 vmcb_mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
1776 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
1777 kvm_update_cpuid_runtime(vcpu);
1780 static void svm_set_segment(struct kvm_vcpu *vcpu,
1781 struct kvm_segment *var, int seg)
1783 struct vcpu_svm *svm = to_svm(vcpu);
1784 struct vmcb_seg *s = svm_seg(vcpu, seg);
1786 s->base = var->base;
1787 s->limit = var->limit;
1788 s->selector = var->selector;
1789 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
1790 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
1791 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
1792 s->attrib |= ((var->present & 1) && !var->unusable) << SVM_SELECTOR_P_SHIFT;
1793 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
1794 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
1795 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
1796 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
1799 * This is always accurate, except if SYSRET returned to a segment
1800 * with SS.DPL != 3. Intel does not have this quirk, and always
1801 * forces SS.DPL to 3 on sysret, so we ignore that case; fixing it
1802 * would entail passing the CPL to userspace and back.
1804 if (seg == VCPU_SREG_SS)
1805 /* This is symmetric with svm_get_segment() */
1806 svm->vmcb->save.cpl = (var->dpl & 3);
1808 vmcb_mark_dirty(svm->vmcb, VMCB_SEG);
1811 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1813 struct vcpu_svm *svm = to_svm(vcpu);
1815 clr_exception_intercept(svm, BP_VECTOR);
1817 if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
1818 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
1819 set_exception_intercept(svm, BP_VECTOR);
1823 static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
1825 if (sd->next_asid > sd->max_asid) {
1826 ++sd->asid_generation;
1827 sd->next_asid = sd->min_asid;
1828 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
1829 vmcb_mark_dirty(svm->vmcb, VMCB_ASID);
1832 svm->asid_generation = sd->asid_generation;
1833 svm->asid = sd->next_asid++;
1836 static void svm_set_dr6(struct vcpu_svm *svm, unsigned long value)
1838 struct vmcb *vmcb = svm->vmcb;
1840 if (svm->vcpu.arch.guest_state_protected)
1843 if (unlikely(value != vmcb->save.dr6)) {
1844 vmcb->save.dr6 = value;
1845 vmcb_mark_dirty(vmcb, VMCB_DR);
1849 static void svm_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
1851 struct vcpu_svm *svm = to_svm(vcpu);
1853 if (vcpu->arch.guest_state_protected)
1856 get_debugreg(vcpu->arch.db[0], 0);
1857 get_debugreg(vcpu->arch.db[1], 1);
1858 get_debugreg(vcpu->arch.db[2], 2);
1859 get_debugreg(vcpu->arch.db[3], 3);
1861 * We cannot reset svm->vmcb->save.dr6 to DR6_FIXED_1|DR6_RTM here,
1862 * because db_interception might need it. We can do it before vmentry.
1864 vcpu->arch.dr6 = svm->vmcb->save.dr6;
1865 vcpu->arch.dr7 = svm->vmcb->save.dr7;
1866 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
1867 set_dr_intercepts(svm);
1870 static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value)
1872 struct vcpu_svm *svm = to_svm(vcpu);
1874 if (vcpu->arch.guest_state_protected)
1877 svm->vmcb->save.dr7 = value;
1878 vmcb_mark_dirty(svm->vmcb, VMCB_DR);
1881 static int pf_interception(struct vcpu_svm *svm)
1883 u64 fault_address = __sme_clr(svm->vmcb->control.exit_info_2);
1884 u64 error_code = svm->vmcb->control.exit_info_1;
1886 return kvm_handle_page_fault(&svm->vcpu, error_code, fault_address,
1887 static_cpu_has(X86_FEATURE_DECODEASSISTS) ?
1888 svm->vmcb->control.insn_bytes : NULL,
1889 svm->vmcb->control.insn_len);
1892 static int npf_interception(struct vcpu_svm *svm)
1894 u64 fault_address = __sme_clr(svm->vmcb->control.exit_info_2);
1895 u64 error_code = svm->vmcb->control.exit_info_1;
1897 trace_kvm_page_fault(fault_address, error_code);
1898 return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code,
1899 static_cpu_has(X86_FEATURE_DECODEASSISTS) ?
1900 svm->vmcb->control.insn_bytes : NULL,
1901 svm->vmcb->control.insn_len);
1904 static int db_interception(struct vcpu_svm *svm)
1906 struct kvm_run *kvm_run = svm->vcpu.run;
1907 struct kvm_vcpu *vcpu = &svm->vcpu;
1909 if (!(svm->vcpu.guest_debug &
1910 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
1911 !svm->nmi_singlestep) {
1912 u32 payload = (svm->vmcb->save.dr6 ^ DR6_RTM) & ~DR6_FIXED_1;
1913 kvm_queue_exception_p(&svm->vcpu, DB_VECTOR, payload);
1917 if (svm->nmi_singlestep) {
1918 disable_nmi_singlestep(svm);
1919 /* Make sure we check for pending NMIs upon entry */
1920 kvm_make_request(KVM_REQ_EVENT, vcpu);
1923 if (svm->vcpu.guest_debug &
1924 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
1925 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1926 kvm_run->debug.arch.dr6 = svm->vmcb->save.dr6;
1927 kvm_run->debug.arch.dr7 = svm->vmcb->save.dr7;
1928 kvm_run->debug.arch.pc =
1929 svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1930 kvm_run->debug.arch.exception = DB_VECTOR;
1937 static int bp_interception(struct vcpu_svm *svm)
1939 struct kvm_run *kvm_run = svm->vcpu.run;
1941 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1942 kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1943 kvm_run->debug.arch.exception = BP_VECTOR;
1947 static int ud_interception(struct vcpu_svm *svm)
1949 return handle_ud(&svm->vcpu);
1952 static int ac_interception(struct vcpu_svm *svm)
1954 kvm_queue_exception_e(&svm->vcpu, AC_VECTOR, 0);
1958 static int gp_interception(struct vcpu_svm *svm)
1960 struct kvm_vcpu *vcpu = &svm->vcpu;
1961 u32 error_code = svm->vmcb->control.exit_info_1;
1963 WARN_ON_ONCE(!enable_vmware_backdoor);
1966 * VMware backdoor emulation on #GP interception only handles IN{S},
1967 * OUT{S}, and RDPMC, none of which generate a non-zero error code.
1970 kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
1973 return kvm_emulate_instruction(vcpu, EMULTYPE_VMWARE_GP);
1976 static bool is_erratum_383(void)
1981 if (!erratum_383_found)
1984 value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err);
1988 /* Bit 62 may or may not be set for this mce */
1989 value &= ~(1ULL << 62);
1991 if (value != 0xb600000000010015ULL)
1994 /* Clear MCi_STATUS registers */
1995 for (i = 0; i < 6; ++i)
1996 native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0);
1998 value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err);
2002 value &= ~(1ULL << 2);
2003 low = lower_32_bits(value);
2004 high = upper_32_bits(value);
2006 native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high);
2009 /* Flush tlb to evict multi-match entries */
2015 static void svm_handle_mce(struct vcpu_svm *svm)
2017 if (is_erratum_383()) {
2019 * Erratum 383 triggered. Guest state is corrupt so kill the
2022 pr_err("KVM: Guest triggered AMD Erratum 383\n");
2024 kvm_make_request(KVM_REQ_TRIPLE_FAULT, &svm->vcpu);
2030 * On an #MC intercept the MCE handler is not called automatically in
2031 * the host. So do it by hand here.
2033 kvm_machine_check();
2036 static int mc_interception(struct vcpu_svm *svm)
2041 static int shutdown_interception(struct vcpu_svm *svm)
2043 struct kvm_run *kvm_run = svm->vcpu.run;
2046 * The VM save area has already been encrypted so it
2047 * cannot be reinitialized - just terminate.
2049 if (sev_es_guest(svm->vcpu.kvm))
2053 * VMCB is undefined after a SHUTDOWN intercept
2054 * so reinitialize it.
2056 clear_page(svm->vmcb);
2059 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
2063 static int io_interception(struct vcpu_svm *svm)
2065 struct kvm_vcpu *vcpu = &svm->vcpu;
2066 u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
2067 int size, in, string;
2070 ++svm->vcpu.stat.io_exits;
2071 string = (io_info & SVM_IOIO_STR_MASK) != 0;
2072 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
2073 port = io_info >> 16;
2074 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
2077 if (sev_es_guest(vcpu->kvm))
2078 return sev_es_string_io(svm, size, port, in);
2080 return kvm_emulate_instruction(vcpu, 0);
2083 svm->next_rip = svm->vmcb->control.exit_info_2;
2085 return kvm_fast_pio(&svm->vcpu, size, port, in);
2088 static int nmi_interception(struct vcpu_svm *svm)
2093 static int intr_interception(struct vcpu_svm *svm)
2095 ++svm->vcpu.stat.irq_exits;
2099 static int nop_on_interception(struct vcpu_svm *svm)
2104 static int halt_interception(struct vcpu_svm *svm)
2106 return kvm_emulate_halt(&svm->vcpu);
2109 static int vmmcall_interception(struct vcpu_svm *svm)
2111 return kvm_emulate_hypercall(&svm->vcpu);
2114 static int vmload_interception(struct vcpu_svm *svm)
2116 struct vmcb *nested_vmcb;
2117 struct kvm_host_map map;
2120 if (nested_svm_check_permissions(svm))
2123 ret = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(svm->vmcb->save.rax), &map);
2126 kvm_inject_gp(&svm->vcpu, 0);
2130 nested_vmcb = map.hva;
2132 ret = kvm_skip_emulated_instruction(&svm->vcpu);
2134 nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
2135 kvm_vcpu_unmap(&svm->vcpu, &map, true);
2140 static int vmsave_interception(struct vcpu_svm *svm)
2142 struct vmcb *nested_vmcb;
2143 struct kvm_host_map map;
2146 if (nested_svm_check_permissions(svm))
2149 ret = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(svm->vmcb->save.rax), &map);
2152 kvm_inject_gp(&svm->vcpu, 0);
2156 nested_vmcb = map.hva;
2158 ret = kvm_skip_emulated_instruction(&svm->vcpu);
2160 nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
2161 kvm_vcpu_unmap(&svm->vcpu, &map, true);
2166 static int vmrun_interception(struct vcpu_svm *svm)
2168 if (nested_svm_check_permissions(svm))
2171 return nested_svm_vmrun(svm);
2174 void svm_set_gif(struct vcpu_svm *svm, bool value)
2178 * If VGIF is enabled, the STGI intercept is only added to
2179 * detect the opening of the SMI/NMI window; remove it now.
2180 * Likewise, clear the VINTR intercept, we will set it
2181 * again while processing KVM_REQ_EVENT if needed.
2183 if (vgif_enabled(svm))
2184 svm_clr_intercept(svm, INTERCEPT_STGI);
2185 if (svm_is_intercept(svm, INTERCEPT_VINTR))
2186 svm_clear_vintr(svm);
2189 if (svm->vcpu.arch.smi_pending ||
2190 svm->vcpu.arch.nmi_pending ||
2191 kvm_cpu_has_injectable_intr(&svm->vcpu))
2192 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2197 * After a CLGI no interrupts should come. But if vGIF is
2198 * in use, we still rely on the VINTR intercept (rather than
2199 * STGI) to detect an open interrupt window.
2201 if (!vgif_enabled(svm))
2202 svm_clear_vintr(svm);
2206 static int stgi_interception(struct vcpu_svm *svm)
2210 if (nested_svm_check_permissions(svm))
2213 ret = kvm_skip_emulated_instruction(&svm->vcpu);
2214 svm_set_gif(svm, true);
2218 static int clgi_interception(struct vcpu_svm *svm)
2222 if (nested_svm_check_permissions(svm))
2225 ret = kvm_skip_emulated_instruction(&svm->vcpu);
2226 svm_set_gif(svm, false);
2230 static int invlpga_interception(struct vcpu_svm *svm)
2232 struct kvm_vcpu *vcpu = &svm->vcpu;
2234 trace_kvm_invlpga(svm->vmcb->save.rip, kvm_rcx_read(&svm->vcpu),
2235 kvm_rax_read(&svm->vcpu));
2237 /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
2238 kvm_mmu_invlpg(vcpu, kvm_rax_read(&svm->vcpu));
2240 return kvm_skip_emulated_instruction(&svm->vcpu);
2243 static int skinit_interception(struct vcpu_svm *svm)
2245 trace_kvm_skinit(svm->vmcb->save.rip, kvm_rax_read(&svm->vcpu));
2247 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2251 static int wbinvd_interception(struct vcpu_svm *svm)
2253 return kvm_emulate_wbinvd(&svm->vcpu);
2256 static int xsetbv_interception(struct vcpu_svm *svm)
2258 u64 new_bv = kvm_read_edx_eax(&svm->vcpu);
2259 u32 index = kvm_rcx_read(&svm->vcpu);
2261 if (kvm_set_xcr(&svm->vcpu, index, new_bv) == 0) {
2262 return kvm_skip_emulated_instruction(&svm->vcpu);
2268 static int rdpru_interception(struct vcpu_svm *svm)
2270 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2274 static int task_switch_interception(struct vcpu_svm *svm)
2278 int int_type = svm->vmcb->control.exit_int_info &
2279 SVM_EXITINTINFO_TYPE_MASK;
2280 int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
2282 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
2284 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
2285 bool has_error_code = false;
2288 tss_selector = (u16)svm->vmcb->control.exit_info_1;
2290 if (svm->vmcb->control.exit_info_2 &
2291 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
2292 reason = TASK_SWITCH_IRET;
2293 else if (svm->vmcb->control.exit_info_2 &
2294 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
2295 reason = TASK_SWITCH_JMP;
2297 reason = TASK_SWITCH_GATE;
2299 reason = TASK_SWITCH_CALL;
2301 if (reason == TASK_SWITCH_GATE) {
2303 case SVM_EXITINTINFO_TYPE_NMI:
2304 svm->vcpu.arch.nmi_injected = false;
2306 case SVM_EXITINTINFO_TYPE_EXEPT:
2307 if (svm->vmcb->control.exit_info_2 &
2308 (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) {
2309 has_error_code = true;
2311 (u32)svm->vmcb->control.exit_info_2;
2313 kvm_clear_exception_queue(&svm->vcpu);
2315 case SVM_EXITINTINFO_TYPE_INTR:
2316 kvm_clear_interrupt_queue(&svm->vcpu);
2323 if (reason != TASK_SWITCH_GATE ||
2324 int_type == SVM_EXITINTINFO_TYPE_SOFT ||
2325 (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
2326 (int_vec == OF_VECTOR || int_vec == BP_VECTOR))) {
2327 if (!skip_emulated_instruction(&svm->vcpu))
2331 if (int_type != SVM_EXITINTINFO_TYPE_SOFT)
2334 return kvm_task_switch(&svm->vcpu, tss_selector, int_vec, reason,
2335 has_error_code, error_code);
2338 static int cpuid_interception(struct vcpu_svm *svm)
2340 return kvm_emulate_cpuid(&svm->vcpu);
2343 static int iret_interception(struct vcpu_svm *svm)
2345 ++svm->vcpu.stat.nmi_window_exits;
2346 svm->vcpu.arch.hflags |= HF_IRET_MASK;
2347 if (!sev_es_guest(svm->vcpu.kvm)) {
2348 svm_clr_intercept(svm, INTERCEPT_IRET);
2349 svm->nmi_iret_rip = kvm_rip_read(&svm->vcpu);
2351 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2355 static int invd_interception(struct vcpu_svm *svm)
2357 /* Treat an INVD instruction as a NOP and just skip it. */
2358 return kvm_skip_emulated_instruction(&svm->vcpu);
2361 static int invlpg_interception(struct vcpu_svm *svm)
2363 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2364 return kvm_emulate_instruction(&svm->vcpu, 0);
2366 kvm_mmu_invlpg(&svm->vcpu, svm->vmcb->control.exit_info_1);
2367 return kvm_skip_emulated_instruction(&svm->vcpu);
2370 static int emulate_on_interception(struct vcpu_svm *svm)
2372 return kvm_emulate_instruction(&svm->vcpu, 0);
2375 static int rsm_interception(struct vcpu_svm *svm)
2377 return kvm_emulate_instruction_from_buffer(&svm->vcpu, rsm_ins_bytes, 2);
2380 static int rdpmc_interception(struct vcpu_svm *svm)
2385 return emulate_on_interception(svm);
2387 err = kvm_rdpmc(&svm->vcpu);
2388 return kvm_complete_insn_gp(&svm->vcpu, err);
2391 static bool check_selective_cr0_intercepted(struct vcpu_svm *svm,
2394 unsigned long cr0 = svm->vcpu.arch.cr0;
2397 if (!is_guest_mode(&svm->vcpu) ||
2398 (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_SELECTIVE_CR0))))
2401 cr0 &= ~SVM_CR0_SELECTIVE_MASK;
2402 val &= ~SVM_CR0_SELECTIVE_MASK;
2405 svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
2406 ret = (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE);
2412 #define CR_VALID (1ULL << 63)
2414 static int cr_interception(struct vcpu_svm *svm)
2420 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2421 return emulate_on_interception(svm);
2423 if (unlikely((svm->vmcb->control.exit_info_1 & CR_VALID) == 0))
2424 return emulate_on_interception(svm);
2426 reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
2427 if (svm->vmcb->control.exit_code == SVM_EXIT_CR0_SEL_WRITE)
2428 cr = SVM_EXIT_WRITE_CR0 - SVM_EXIT_READ_CR0;
2430 cr = svm->vmcb->control.exit_code - SVM_EXIT_READ_CR0;
2433 if (cr >= 16) { /* mov to cr */
2435 val = kvm_register_read(&svm->vcpu, reg);
2436 trace_kvm_cr_write(cr, val);
2439 if (!check_selective_cr0_intercepted(svm, val))
2440 err = kvm_set_cr0(&svm->vcpu, val);
2446 err = kvm_set_cr3(&svm->vcpu, val);
2449 err = kvm_set_cr4(&svm->vcpu, val);
2452 err = kvm_set_cr8(&svm->vcpu, val);
2455 WARN(1, "unhandled write to CR%d", cr);
2456 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2459 } else { /* mov from cr */
2462 val = kvm_read_cr0(&svm->vcpu);
2465 val = svm->vcpu.arch.cr2;
2468 val = kvm_read_cr3(&svm->vcpu);
2471 val = kvm_read_cr4(&svm->vcpu);
2474 val = kvm_get_cr8(&svm->vcpu);
2477 WARN(1, "unhandled read from CR%d", cr);
2478 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2481 kvm_register_write(&svm->vcpu, reg, val);
2482 trace_kvm_cr_read(cr, val);
2484 return kvm_complete_insn_gp(&svm->vcpu, err);
2487 static int cr_trap(struct vcpu_svm *svm)
2489 struct kvm_vcpu *vcpu = &svm->vcpu;
2490 unsigned long old_value, new_value;
2494 new_value = (unsigned long)svm->vmcb->control.exit_info_1;
2496 cr = svm->vmcb->control.exit_code - SVM_EXIT_CR0_WRITE_TRAP;
2499 old_value = kvm_read_cr0(vcpu);
2500 svm_set_cr0(vcpu, new_value);
2502 kvm_post_set_cr0(vcpu, old_value, new_value);
2505 old_value = kvm_read_cr4(vcpu);
2506 svm_set_cr4(vcpu, new_value);
2508 kvm_post_set_cr4(vcpu, old_value, new_value);
2511 ret = kvm_set_cr8(&svm->vcpu, new_value);
2514 WARN(1, "unhandled CR%d write trap", cr);
2515 kvm_queue_exception(vcpu, UD_VECTOR);
2519 return kvm_complete_insn_gp(vcpu, ret);
2522 static int dr_interception(struct vcpu_svm *svm)
2527 if (svm->vcpu.guest_debug == 0) {
2529 * No more DR vmexits; force a reload of the debug registers
2530 * and reenter on this instruction. The next vmexit will
2531 * retrieve the full state of the debug registers.
2533 clr_dr_intercepts(svm);
2534 svm->vcpu.arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
2538 if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS))
2539 return emulate_on_interception(svm);
2541 reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
2542 dr = svm->vmcb->control.exit_code - SVM_EXIT_READ_DR0;
2544 if (dr >= 16) { /* mov to DRn */
2545 if (!kvm_require_dr(&svm->vcpu, dr - 16))
2547 val = kvm_register_read(&svm->vcpu, reg);
2548 kvm_set_dr(&svm->vcpu, dr - 16, val);
2550 if (!kvm_require_dr(&svm->vcpu, dr))
2552 kvm_get_dr(&svm->vcpu, dr, &val);
2553 kvm_register_write(&svm->vcpu, reg, val);
2556 return kvm_skip_emulated_instruction(&svm->vcpu);
2559 static int cr8_write_interception(struct vcpu_svm *svm)
2561 struct kvm_run *kvm_run = svm->vcpu.run;
2564 u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
2565 /* instruction emulation calls kvm_set_cr8() */
2566 r = cr_interception(svm);
2567 if (lapic_in_kernel(&svm->vcpu))
2569 if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
2571 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
2575 static int efer_trap(struct vcpu_svm *svm)
2577 struct msr_data msr_info;
2581 * Clear the EFER_SVME bit from EFER. The SVM code always sets this
2582 * bit in svm_set_efer(), but __kvm_valid_efer() checks it against
2583 * whether the guest has X86_FEATURE_SVM - this avoids a failure if
2584 * the guest doesn't have X86_FEATURE_SVM.
2586 msr_info.host_initiated = false;
2587 msr_info.index = MSR_EFER;
2588 msr_info.data = svm->vmcb->control.exit_info_1 & ~EFER_SVME;
2589 ret = kvm_set_msr_common(&svm->vcpu, &msr_info);
2591 return kvm_complete_insn_gp(&svm->vcpu, ret);
2594 static int svm_get_msr_feature(struct kvm_msr_entry *msr)
2598 switch (msr->index) {
2599 case MSR_F10H_DECFG:
2600 if (boot_cpu_has(X86_FEATURE_LFENCE_RDTSC))
2601 msr->data |= MSR_F10H_DECFG_LFENCE_SERIALIZE;
2603 case MSR_IA32_PERF_CAPABILITIES:
2606 return KVM_MSR_RET_INVALID;
2612 static int svm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2614 struct vcpu_svm *svm = to_svm(vcpu);
2616 switch (msr_info->index) {
2618 msr_info->data = svm->vmcb->save.star;
2620 #ifdef CONFIG_X86_64
2622 msr_info->data = svm->vmcb->save.lstar;
2625 msr_info->data = svm->vmcb->save.cstar;
2627 case MSR_KERNEL_GS_BASE:
2628 msr_info->data = svm->vmcb->save.kernel_gs_base;
2630 case MSR_SYSCALL_MASK:
2631 msr_info->data = svm->vmcb->save.sfmask;
2634 case MSR_IA32_SYSENTER_CS:
2635 msr_info->data = svm->vmcb->save.sysenter_cs;
2637 case MSR_IA32_SYSENTER_EIP:
2638 msr_info->data = svm->sysenter_eip;
2640 case MSR_IA32_SYSENTER_ESP:
2641 msr_info->data = svm->sysenter_esp;
2644 if (!boot_cpu_has(X86_FEATURE_RDTSCP))
2646 msr_info->data = svm->tsc_aux;
2649 * Nobody will change the following 5 values in the VMCB so we can
2650 * safely return them on rdmsr. They will always be 0 until LBRV is
2653 case MSR_IA32_DEBUGCTLMSR:
2654 msr_info->data = svm->vmcb->save.dbgctl;
2656 case MSR_IA32_LASTBRANCHFROMIP:
2657 msr_info->data = svm->vmcb->save.br_from;
2659 case MSR_IA32_LASTBRANCHTOIP:
2660 msr_info->data = svm->vmcb->save.br_to;
2662 case MSR_IA32_LASTINTFROMIP:
2663 msr_info->data = svm->vmcb->save.last_excp_from;
2665 case MSR_IA32_LASTINTTOIP:
2666 msr_info->data = svm->vmcb->save.last_excp_to;
2668 case MSR_VM_HSAVE_PA:
2669 msr_info->data = svm->nested.hsave_msr;
2672 msr_info->data = svm->nested.vm_cr_msr;
2674 case MSR_IA32_SPEC_CTRL:
2675 if (!msr_info->host_initiated &&
2676 !guest_has_spec_ctrl_msr(vcpu))
2679 msr_info->data = svm->spec_ctrl;
2681 case MSR_AMD64_VIRT_SPEC_CTRL:
2682 if (!msr_info->host_initiated &&
2683 !guest_cpuid_has(vcpu, X86_FEATURE_VIRT_SSBD))
2686 msr_info->data = svm->virt_spec_ctrl;
2688 case MSR_F15H_IC_CFG: {
2692 family = guest_cpuid_family(vcpu);
2693 model = guest_cpuid_model(vcpu);
2695 if (family < 0 || model < 0)
2696 return kvm_get_msr_common(vcpu, msr_info);
2700 if (family == 0x15 &&
2701 (model >= 0x2 && model < 0x20))
2702 msr_info->data = 0x1E;
2705 case MSR_F10H_DECFG:
2706 msr_info->data = svm->msr_decfg;
2709 return kvm_get_msr_common(vcpu, msr_info);
2714 static int svm_complete_emulated_msr(struct kvm_vcpu *vcpu, int err)
2716 struct vcpu_svm *svm = to_svm(vcpu);
2717 if (!sev_es_guest(svm->vcpu.kvm) || !err)
2718 return kvm_complete_insn_gp(&svm->vcpu, err);
2720 ghcb_set_sw_exit_info_1(svm->ghcb, 1);
2721 ghcb_set_sw_exit_info_2(svm->ghcb,
2723 SVM_EVTINJ_TYPE_EXEPT |
2728 static int rdmsr_interception(struct vcpu_svm *svm)
2730 return kvm_emulate_rdmsr(&svm->vcpu);
2733 static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
2735 struct vcpu_svm *svm = to_svm(vcpu);
2736 int svm_dis, chg_mask;
2738 if (data & ~SVM_VM_CR_VALID_MASK)
2741 chg_mask = SVM_VM_CR_VALID_MASK;
2743 if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
2744 chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
2746 svm->nested.vm_cr_msr &= ~chg_mask;
2747 svm->nested.vm_cr_msr |= (data & chg_mask);
2749 svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
2751 /* check for svm_disable while efer.svme is set */
2752 if (svm_dis && (vcpu->arch.efer & EFER_SVME))
2758 static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2760 struct vcpu_svm *svm = to_svm(vcpu);
2762 u32 ecx = msr->index;
2763 u64 data = msr->data;
2765 case MSR_IA32_CR_PAT:
2766 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
2768 vcpu->arch.pat = data;
2769 svm->vmcb->save.g_pat = data;
2770 vmcb_mark_dirty(svm->vmcb, VMCB_NPT);
2772 case MSR_IA32_SPEC_CTRL:
2773 if (!msr->host_initiated &&
2774 !guest_has_spec_ctrl_msr(vcpu))
2777 if (kvm_spec_ctrl_test_value(data))
2780 svm->spec_ctrl = data;
2786 * When it's written (to non-zero) for the first time, pass
2790 * The handling of the MSR bitmap for L2 guests is done in
2791 * nested_svm_vmrun_msrpm.
2792 * We update the L1 MSR bit as well since it will end up
2793 * touching the MSR anyway now.
2795 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_SPEC_CTRL, 1, 1);
2797 case MSR_IA32_PRED_CMD:
2798 if (!msr->host_initiated &&
2799 !guest_has_pred_cmd_msr(vcpu))
2802 if (data & ~PRED_CMD_IBPB)
2804 if (!boot_cpu_has(X86_FEATURE_IBPB))
2809 wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
2810 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_PRED_CMD, 0, 1);
2812 case MSR_AMD64_VIRT_SPEC_CTRL:
2813 if (!msr->host_initiated &&
2814 !guest_cpuid_has(vcpu, X86_FEATURE_VIRT_SSBD))
2817 if (data & ~SPEC_CTRL_SSBD)
2820 svm->virt_spec_ctrl = data;
2823 svm->vmcb->save.star = data;
2825 #ifdef CONFIG_X86_64
2827 svm->vmcb->save.lstar = data;
2830 svm->vmcb->save.cstar = data;
2832 case MSR_KERNEL_GS_BASE:
2833 svm->vmcb->save.kernel_gs_base = data;
2835 case MSR_SYSCALL_MASK:
2836 svm->vmcb->save.sfmask = data;
2839 case MSR_IA32_SYSENTER_CS:
2840 svm->vmcb->save.sysenter_cs = data;
2842 case MSR_IA32_SYSENTER_EIP:
2843 svm->sysenter_eip = data;
2844 svm->vmcb->save.sysenter_eip = data;
2846 case MSR_IA32_SYSENTER_ESP:
2847 svm->sysenter_esp = data;
2848 svm->vmcb->save.sysenter_esp = data;
2851 if (!boot_cpu_has(X86_FEATURE_RDTSCP))
2855 * This is rare, so we update the MSR here instead of using
2856 * direct_access_msrs. Doing that would require a rdmsr in
2859 svm->tsc_aux = data;
2860 wrmsrl(MSR_TSC_AUX, svm->tsc_aux);
2862 case MSR_IA32_DEBUGCTLMSR:
2863 if (!boot_cpu_has(X86_FEATURE_LBRV)) {
2864 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
2868 if (data & DEBUGCTL_RESERVED_BITS)
2871 svm->vmcb->save.dbgctl = data;
2872 vmcb_mark_dirty(svm->vmcb, VMCB_LBR);
2873 if (data & (1ULL<<0))
2874 svm_enable_lbrv(vcpu);
2876 svm_disable_lbrv(vcpu);
2878 case MSR_VM_HSAVE_PA:
2879 svm->nested.hsave_msr = data;
2882 return svm_set_vm_cr(vcpu, data);
2884 vcpu_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
2886 case MSR_F10H_DECFG: {
2887 struct kvm_msr_entry msr_entry;
2889 msr_entry.index = msr->index;
2890 if (svm_get_msr_feature(&msr_entry))
2893 /* Check the supported bits */
2894 if (data & ~msr_entry.data)
2897 /* Don't allow the guest to change a bit, #GP */
2898 if (!msr->host_initiated && (data ^ msr_entry.data))
2901 svm->msr_decfg = data;
2904 case MSR_IA32_APICBASE:
2905 if (kvm_vcpu_apicv_active(vcpu))
2906 avic_update_vapic_bar(to_svm(vcpu), data);
2909 return kvm_set_msr_common(vcpu, msr);
2914 static int wrmsr_interception(struct vcpu_svm *svm)
2916 return kvm_emulate_wrmsr(&svm->vcpu);
2919 static int msr_interception(struct vcpu_svm *svm)
2921 if (svm->vmcb->control.exit_info_1)
2922 return wrmsr_interception(svm);
2924 return rdmsr_interception(svm);
2927 static int interrupt_window_interception(struct vcpu_svm *svm)
2929 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2930 svm_clear_vintr(svm);
2933 * For AVIC, the only reason to end up here is ExtINTs.
2934 * In this case AVIC was temporarily disabled for
2935 * requesting the IRQ window and we have to re-enable it.
2937 svm_toggle_avic_for_irq_window(&svm->vcpu, true);
2939 ++svm->vcpu.stat.irq_window_exits;
2943 static int pause_interception(struct vcpu_svm *svm)
2945 struct kvm_vcpu *vcpu = &svm->vcpu;
2949 * CPL is not made available for an SEV-ES guest, therefore
2950 * vcpu->arch.preempted_in_kernel can never be true. Just
2951 * set in_kernel to false as well.
2953 in_kernel = !sev_es_guest(svm->vcpu.kvm) && svm_get_cpl(vcpu) == 0;
2955 if (!kvm_pause_in_guest(vcpu->kvm))
2956 grow_ple_window(vcpu);
2958 kvm_vcpu_on_spin(vcpu, in_kernel);
2962 static int nop_interception(struct vcpu_svm *svm)
2964 return kvm_skip_emulated_instruction(&(svm->vcpu));
2967 static int monitor_interception(struct vcpu_svm *svm)
2969 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
2970 return nop_interception(svm);
2973 static int mwait_interception(struct vcpu_svm *svm)
2975 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
2976 return nop_interception(svm);
2979 static int invpcid_interception(struct vcpu_svm *svm)
2981 struct kvm_vcpu *vcpu = &svm->vcpu;
2985 if (!guest_cpuid_has(vcpu, X86_FEATURE_INVPCID)) {
2986 kvm_queue_exception(vcpu, UD_VECTOR);
2991 * For an INVPCID intercept:
2992 * EXITINFO1 provides the linear address of the memory operand.
2993 * EXITINFO2 provides the contents of the register operand.
2995 type = svm->vmcb->control.exit_info_2;
2996 gva = svm->vmcb->control.exit_info_1;
2999 kvm_inject_gp(vcpu, 0);
3003 return kvm_handle_invpcid(vcpu, type, gva);
3006 static int (*const svm_exit_handlers[])(struct vcpu_svm *svm) = {
3007 [SVM_EXIT_READ_CR0] = cr_interception,
3008 [SVM_EXIT_READ_CR3] = cr_interception,
3009 [SVM_EXIT_READ_CR4] = cr_interception,
3010 [SVM_EXIT_READ_CR8] = cr_interception,
3011 [SVM_EXIT_CR0_SEL_WRITE] = cr_interception,
3012 [SVM_EXIT_WRITE_CR0] = cr_interception,
3013 [SVM_EXIT_WRITE_CR3] = cr_interception,
3014 [SVM_EXIT_WRITE_CR4] = cr_interception,
3015 [SVM_EXIT_WRITE_CR8] = cr8_write_interception,
3016 [SVM_EXIT_READ_DR0] = dr_interception,
3017 [SVM_EXIT_READ_DR1] = dr_interception,
3018 [SVM_EXIT_READ_DR2] = dr_interception,
3019 [SVM_EXIT_READ_DR3] = dr_interception,
3020 [SVM_EXIT_READ_DR4] = dr_interception,
3021 [SVM_EXIT_READ_DR5] = dr_interception,
3022 [SVM_EXIT_READ_DR6] = dr_interception,
3023 [SVM_EXIT_READ_DR7] = dr_interception,
3024 [SVM_EXIT_WRITE_DR0] = dr_interception,
3025 [SVM_EXIT_WRITE_DR1] = dr_interception,
3026 [SVM_EXIT_WRITE_DR2] = dr_interception,
3027 [SVM_EXIT_WRITE_DR3] = dr_interception,
3028 [SVM_EXIT_WRITE_DR4] = dr_interception,
3029 [SVM_EXIT_WRITE_DR5] = dr_interception,
3030 [SVM_EXIT_WRITE_DR6] = dr_interception,
3031 [SVM_EXIT_WRITE_DR7] = dr_interception,
3032 [SVM_EXIT_EXCP_BASE + DB_VECTOR] = db_interception,
3033 [SVM_EXIT_EXCP_BASE + BP_VECTOR] = bp_interception,
3034 [SVM_EXIT_EXCP_BASE + UD_VECTOR] = ud_interception,
3035 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
3036 [SVM_EXIT_EXCP_BASE + MC_VECTOR] = mc_interception,
3037 [SVM_EXIT_EXCP_BASE + AC_VECTOR] = ac_interception,
3038 [SVM_EXIT_EXCP_BASE + GP_VECTOR] = gp_interception,
3039 [SVM_EXIT_INTR] = intr_interception,
3040 [SVM_EXIT_NMI] = nmi_interception,
3041 [SVM_EXIT_SMI] = nop_on_interception,
3042 [SVM_EXIT_INIT] = nop_on_interception,
3043 [SVM_EXIT_VINTR] = interrupt_window_interception,
3044 [SVM_EXIT_RDPMC] = rdpmc_interception,
3045 [SVM_EXIT_CPUID] = cpuid_interception,
3046 [SVM_EXIT_IRET] = iret_interception,
3047 [SVM_EXIT_INVD] = invd_interception,
3048 [SVM_EXIT_PAUSE] = pause_interception,
3049 [SVM_EXIT_HLT] = halt_interception,
3050 [SVM_EXIT_INVLPG] = invlpg_interception,
3051 [SVM_EXIT_INVLPGA] = invlpga_interception,
3052 [SVM_EXIT_IOIO] = io_interception,
3053 [SVM_EXIT_MSR] = msr_interception,
3054 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
3055 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
3056 [SVM_EXIT_VMRUN] = vmrun_interception,
3057 [SVM_EXIT_VMMCALL] = vmmcall_interception,
3058 [SVM_EXIT_VMLOAD] = vmload_interception,
3059 [SVM_EXIT_VMSAVE] = vmsave_interception,
3060 [SVM_EXIT_STGI] = stgi_interception,
3061 [SVM_EXIT_CLGI] = clgi_interception,
3062 [SVM_EXIT_SKINIT] = skinit_interception,
3063 [SVM_EXIT_WBINVD] = wbinvd_interception,
3064 [SVM_EXIT_MONITOR] = monitor_interception,
3065 [SVM_EXIT_MWAIT] = mwait_interception,
3066 [SVM_EXIT_XSETBV] = xsetbv_interception,
3067 [SVM_EXIT_RDPRU] = rdpru_interception,
3068 [SVM_EXIT_EFER_WRITE_TRAP] = efer_trap,
3069 [SVM_EXIT_CR0_WRITE_TRAP] = cr_trap,
3070 [SVM_EXIT_CR4_WRITE_TRAP] = cr_trap,
3071 [SVM_EXIT_CR8_WRITE_TRAP] = cr_trap,
3072 [SVM_EXIT_INVPCID] = invpcid_interception,
3073 [SVM_EXIT_NPF] = npf_interception,
3074 [SVM_EXIT_RSM] = rsm_interception,
3075 [SVM_EXIT_AVIC_INCOMPLETE_IPI] = avic_incomplete_ipi_interception,
3076 [SVM_EXIT_AVIC_UNACCELERATED_ACCESS] = avic_unaccelerated_access_interception,
3077 [SVM_EXIT_VMGEXIT] = sev_handle_vmgexit,
3080 static void dump_vmcb(struct kvm_vcpu *vcpu)
3082 struct vcpu_svm *svm = to_svm(vcpu);
3083 struct vmcb_control_area *control = &svm->vmcb->control;
3084 struct vmcb_save_area *save = &svm->vmcb->save;
3086 if (!dump_invalid_vmcb) {
3087 pr_warn_ratelimited("set kvm_amd.dump_invalid_vmcb=1 to dump internal KVM state.\n");
3091 pr_err("VMCB Control Area:\n");
3092 pr_err("%-20s%04x\n", "cr_read:", control->intercepts[INTERCEPT_CR] & 0xffff);
3093 pr_err("%-20s%04x\n", "cr_write:", control->intercepts[INTERCEPT_CR] >> 16);
3094 pr_err("%-20s%04x\n", "dr_read:", control->intercepts[INTERCEPT_DR] & 0xffff);
3095 pr_err("%-20s%04x\n", "dr_write:", control->intercepts[INTERCEPT_DR] >> 16);
3096 pr_err("%-20s%08x\n", "exceptions:", control->intercepts[INTERCEPT_EXCEPTION]);
3097 pr_err("%-20s%08x %08x\n", "intercepts:",
3098 control->intercepts[INTERCEPT_WORD3],
3099 control->intercepts[INTERCEPT_WORD4]);
3100 pr_err("%-20s%d\n", "pause filter count:", control->pause_filter_count);
3101 pr_err("%-20s%d\n", "pause filter threshold:",
3102 control->pause_filter_thresh);
3103 pr_err("%-20s%016llx\n", "iopm_base_pa:", control->iopm_base_pa);
3104 pr_err("%-20s%016llx\n", "msrpm_base_pa:", control->msrpm_base_pa);
3105 pr_err("%-20s%016llx\n", "tsc_offset:", control->tsc_offset);
3106 pr_err("%-20s%d\n", "asid:", control->asid);
3107 pr_err("%-20s%d\n", "tlb_ctl:", control->tlb_ctl);
3108 pr_err("%-20s%08x\n", "int_ctl:", control->int_ctl);
3109 pr_err("%-20s%08x\n", "int_vector:", control->int_vector);
3110 pr_err("%-20s%08x\n", "int_state:", control->int_state);
3111 pr_err("%-20s%08x\n", "exit_code:", control->exit_code);
3112 pr_err("%-20s%016llx\n", "exit_info1:", control->exit_info_1);
3113 pr_err("%-20s%016llx\n", "exit_info2:", control->exit_info_2);
3114 pr_err("%-20s%08x\n", "exit_int_info:", control->exit_int_info);
3115 pr_err("%-20s%08x\n", "exit_int_info_err:", control->exit_int_info_err);
3116 pr_err("%-20s%lld\n", "nested_ctl:", control->nested_ctl);
3117 pr_err("%-20s%016llx\n", "nested_cr3:", control->nested_cr3);
3118 pr_err("%-20s%016llx\n", "avic_vapic_bar:", control->avic_vapic_bar);
3119 pr_err("%-20s%016llx\n", "ghcb:", control->ghcb_gpa);
3120 pr_err("%-20s%08x\n", "event_inj:", control->event_inj);
3121 pr_err("%-20s%08x\n", "event_inj_err:", control->event_inj_err);
3122 pr_err("%-20s%lld\n", "virt_ext:", control->virt_ext);
3123 pr_err("%-20s%016llx\n", "next_rip:", control->next_rip);
3124 pr_err("%-20s%016llx\n", "avic_backing_page:", control->avic_backing_page);
3125 pr_err("%-20s%016llx\n", "avic_logical_id:", control->avic_logical_id);
3126 pr_err("%-20s%016llx\n", "avic_physical_id:", control->avic_physical_id);
3127 pr_err("%-20s%016llx\n", "vmsa_pa:", control->vmsa_pa);
3128 pr_err("VMCB State Save Area:\n");
3129 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3131 save->es.selector, save->es.attrib,
3132 save->es.limit, save->es.base);
3133 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3135 save->cs.selector, save->cs.attrib,
3136 save->cs.limit, save->cs.base);
3137 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3139 save->ss.selector, save->ss.attrib,
3140 save->ss.limit, save->ss.base);
3141 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3143 save->ds.selector, save->ds.attrib,
3144 save->ds.limit, save->ds.base);
3145 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3147 save->fs.selector, save->fs.attrib,
3148 save->fs.limit, save->fs.base);
3149 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3151 save->gs.selector, save->gs.attrib,
3152 save->gs.limit, save->gs.base);
3153 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3155 save->gdtr.selector, save->gdtr.attrib,
3156 save->gdtr.limit, save->gdtr.base);
3157 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3159 save->ldtr.selector, save->ldtr.attrib,
3160 save->ldtr.limit, save->ldtr.base);
3161 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3163 save->idtr.selector, save->idtr.attrib,
3164 save->idtr.limit, save->idtr.base);
3165 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
3167 save->tr.selector, save->tr.attrib,
3168 save->tr.limit, save->tr.base);
3169 pr_err("cpl: %d efer: %016llx\n",
3170 save->cpl, save->efer);
3171 pr_err("%-15s %016llx %-13s %016llx\n",
3172 "cr0:", save->cr0, "cr2:", save->cr2);
3173 pr_err("%-15s %016llx %-13s %016llx\n",
3174 "cr3:", save->cr3, "cr4:", save->cr4);
3175 pr_err("%-15s %016llx %-13s %016llx\n",
3176 "dr6:", save->dr6, "dr7:", save->dr7);
3177 pr_err("%-15s %016llx %-13s %016llx\n",
3178 "rip:", save->rip, "rflags:", save->rflags);
3179 pr_err("%-15s %016llx %-13s %016llx\n",
3180 "rsp:", save->rsp, "rax:", save->rax);
3181 pr_err("%-15s %016llx %-13s %016llx\n",
3182 "star:", save->star, "lstar:", save->lstar);
3183 pr_err("%-15s %016llx %-13s %016llx\n",
3184 "cstar:", save->cstar, "sfmask:", save->sfmask);
3185 pr_err("%-15s %016llx %-13s %016llx\n",
3186 "kernel_gs_base:", save->kernel_gs_base,
3187 "sysenter_cs:", save->sysenter_cs);
3188 pr_err("%-15s %016llx %-13s %016llx\n",
3189 "sysenter_esp:", save->sysenter_esp,
3190 "sysenter_eip:", save->sysenter_eip);
3191 pr_err("%-15s %016llx %-13s %016llx\n",
3192 "gpat:", save->g_pat, "dbgctl:", save->dbgctl);
3193 pr_err("%-15s %016llx %-13s %016llx\n",
3194 "br_from:", save->br_from, "br_to:", save->br_to);
3195 pr_err("%-15s %016llx %-13s %016llx\n",
3196 "excp_from:", save->last_excp_from,
3197 "excp_to:", save->last_excp_to);
3200 static int svm_handle_invalid_exit(struct kvm_vcpu *vcpu, u64 exit_code)
3202 if (exit_code < ARRAY_SIZE(svm_exit_handlers) &&
3203 svm_exit_handlers[exit_code])
3206 vcpu_unimpl(vcpu, "svm: unexpected exit reason 0x%llx\n", exit_code);
3208 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3209 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
3210 vcpu->run->internal.ndata = 2;
3211 vcpu->run->internal.data[0] = exit_code;
3212 vcpu->run->internal.data[1] = vcpu->arch.last_vmentry_cpu;
3217 int svm_invoke_exit_handler(struct vcpu_svm *svm, u64 exit_code)
3219 if (svm_handle_invalid_exit(&svm->vcpu, exit_code))
3222 #ifdef CONFIG_RETPOLINE
3223 if (exit_code == SVM_EXIT_MSR)
3224 return msr_interception(svm);
3225 else if (exit_code == SVM_EXIT_VINTR)
3226 return interrupt_window_interception(svm);
3227 else if (exit_code == SVM_EXIT_INTR)
3228 return intr_interception(svm);
3229 else if (exit_code == SVM_EXIT_HLT)
3230 return halt_interception(svm);
3231 else if (exit_code == SVM_EXIT_NPF)
3232 return npf_interception(svm);
3234 return svm_exit_handlers[exit_code](svm);
3237 static void svm_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2,
3238 u32 *intr_info, u32 *error_code)
3240 struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control;
3242 *info1 = control->exit_info_1;
3243 *info2 = control->exit_info_2;
3244 *intr_info = control->exit_int_info;
3245 if ((*intr_info & SVM_EXITINTINFO_VALID) &&
3246 (*intr_info & SVM_EXITINTINFO_VALID_ERR))
3247 *error_code = control->exit_int_info_err;
3252 static int handle_exit(struct kvm_vcpu *vcpu, fastpath_t exit_fastpath)
3254 struct vcpu_svm *svm = to_svm(vcpu);
3255 struct kvm_run *kvm_run = vcpu->run;
3256 u32 exit_code = svm->vmcb->control.exit_code;
3258 trace_kvm_exit(exit_code, vcpu, KVM_ISA_SVM);
3260 /* SEV-ES guests must use the CR write traps to track CR registers. */
3261 if (!sev_es_guest(vcpu->kvm)) {
3262 if (!svm_is_intercept(svm, INTERCEPT_CR0_WRITE))
3263 vcpu->arch.cr0 = svm->vmcb->save.cr0;
3265 vcpu->arch.cr3 = svm->vmcb->save.cr3;
3268 if (is_guest_mode(vcpu)) {
3271 trace_kvm_nested_vmexit(exit_code, vcpu, KVM_ISA_SVM);
3273 vmexit = nested_svm_exit_special(svm);
3275 if (vmexit == NESTED_EXIT_CONTINUE)
3276 vmexit = nested_svm_exit_handled(svm);
3278 if (vmexit == NESTED_EXIT_DONE)
3282 if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
3283 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3284 kvm_run->fail_entry.hardware_entry_failure_reason
3285 = svm->vmcb->control.exit_code;
3286 kvm_run->fail_entry.cpu = vcpu->arch.last_vmentry_cpu;
3291 if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
3292 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
3293 exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH &&
3294 exit_code != SVM_EXIT_INTR && exit_code != SVM_EXIT_NMI)
3295 printk(KERN_ERR "%s: unexpected exit_int_info 0x%x "
3297 __func__, svm->vmcb->control.exit_int_info,
3300 if (exit_fastpath != EXIT_FASTPATH_NONE)
3303 return svm_invoke_exit_handler(svm, exit_code);
3306 static void reload_tss(struct kvm_vcpu *vcpu)
3308 struct svm_cpu_data *sd = per_cpu(svm_data, vcpu->cpu);
3310 sd->tss_desc->type = 9; /* available 32/64-bit TSS */
3314 static void pre_svm_run(struct vcpu_svm *svm)
3316 struct svm_cpu_data *sd = per_cpu(svm_data, svm->vcpu.cpu);
3318 if (sev_guest(svm->vcpu.kvm))
3319 return pre_sev_run(svm, svm->vcpu.cpu);
3321 /* FIXME: handle wraparound of asid_generation */
3322 if (svm->asid_generation != sd->asid_generation)
3326 static void svm_inject_nmi(struct kvm_vcpu *vcpu)
3328 struct vcpu_svm *svm = to_svm(vcpu);
3330 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
3331 vcpu->arch.hflags |= HF_NMI_MASK;
3332 if (!sev_es_guest(svm->vcpu.kvm))
3333 svm_set_intercept(svm, INTERCEPT_IRET);
3334 ++vcpu->stat.nmi_injections;
3337 static void svm_set_irq(struct kvm_vcpu *vcpu)
3339 struct vcpu_svm *svm = to_svm(vcpu);
3341 BUG_ON(!(gif_set(svm)));
3343 trace_kvm_inj_virq(vcpu->arch.interrupt.nr);
3344 ++vcpu->stat.irq_injections;
3346 svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
3347 SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
3350 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3352 struct vcpu_svm *svm = to_svm(vcpu);
3355 * SEV-ES guests must always keep the CR intercepts cleared. CR
3356 * tracking is done using the CR write traps.
3358 if (sev_es_guest(vcpu->kvm))
3361 if (nested_svm_virtualize_tpr(vcpu))
3364 svm_clr_intercept(svm, INTERCEPT_CR8_WRITE);
3370 svm_set_intercept(svm, INTERCEPT_CR8_WRITE);
3373 bool svm_nmi_blocked(struct kvm_vcpu *vcpu)
3375 struct vcpu_svm *svm = to_svm(vcpu);
3376 struct vmcb *vmcb = svm->vmcb;
3382 if (is_guest_mode(vcpu) && nested_exit_on_nmi(svm))
3385 ret = (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) ||
3386 (svm->vcpu.arch.hflags & HF_NMI_MASK);
3391 static int svm_nmi_allowed(struct kvm_vcpu *vcpu, bool for_injection)
3393 struct vcpu_svm *svm = to_svm(vcpu);
3394 if (svm->nested.nested_run_pending)
3397 /* An NMI must not be injected into L2 if it's supposed to VM-Exit. */
3398 if (for_injection && is_guest_mode(vcpu) && nested_exit_on_nmi(svm))
3401 return !svm_nmi_blocked(vcpu);
3404 static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
3406 struct vcpu_svm *svm = to_svm(vcpu);
3408 return !!(svm->vcpu.arch.hflags & HF_NMI_MASK);
3411 static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
3413 struct vcpu_svm *svm = to_svm(vcpu);
3416 svm->vcpu.arch.hflags |= HF_NMI_MASK;
3417 if (!sev_es_guest(svm->vcpu.kvm))
3418 svm_set_intercept(svm, INTERCEPT_IRET);
3420 svm->vcpu.arch.hflags &= ~HF_NMI_MASK;
3421 if (!sev_es_guest(svm->vcpu.kvm))
3422 svm_clr_intercept(svm, INTERCEPT_IRET);
3426 bool svm_interrupt_blocked(struct kvm_vcpu *vcpu)
3428 struct vcpu_svm *svm = to_svm(vcpu);
3429 struct vmcb *vmcb = svm->vmcb;
3434 if (sev_es_guest(svm->vcpu.kvm)) {
3436 * SEV-ES guests to not expose RFLAGS. Use the VMCB interrupt mask
3437 * bit to determine the state of the IF flag.
3439 if (!(vmcb->control.int_state & SVM_GUEST_INTERRUPT_MASK))
3441 } else if (is_guest_mode(vcpu)) {
3442 /* As long as interrupts are being delivered... */
3443 if ((svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK)
3444 ? !(svm->nested.hsave->save.rflags & X86_EFLAGS_IF)
3445 : !(kvm_get_rflags(vcpu) & X86_EFLAGS_IF))
3448 /* ... vmexits aren't blocked by the interrupt shadow */
3449 if (nested_exit_on_intr(svm))
3452 if (!(kvm_get_rflags(vcpu) & X86_EFLAGS_IF))
3456 return (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK);
3459 static int svm_interrupt_allowed(struct kvm_vcpu *vcpu, bool for_injection)
3461 struct vcpu_svm *svm = to_svm(vcpu);
3462 if (svm->nested.nested_run_pending)
3466 * An IRQ must not be injected into L2 if it's supposed to VM-Exit,
3467 * e.g. if the IRQ arrived asynchronously after checking nested events.
3469 if (for_injection && is_guest_mode(vcpu) && nested_exit_on_intr(svm))
3472 return !svm_interrupt_blocked(vcpu);
3475 static void enable_irq_window(struct kvm_vcpu *vcpu)
3477 struct vcpu_svm *svm = to_svm(vcpu);
3480 * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
3481 * 1, because that's a separate STGI/VMRUN intercept. The next time we
3482 * get that intercept, this function will be called again though and
3483 * we'll get the vintr intercept. However, if the vGIF feature is
3484 * enabled, the STGI interception will not occur. Enable the irq
3485 * window under the assumption that the hardware will set the GIF.
3487 if (vgif_enabled(svm) || gif_set(svm)) {
3489 * IRQ window is not needed when AVIC is enabled,
3490 * unless we have pending ExtINT since it cannot be injected
3491 * via AVIC. In such case, we need to temporarily disable AVIC,
3492 * and fallback to injecting IRQ via V_IRQ.
3494 svm_toggle_avic_for_irq_window(vcpu, false);
3499 static void enable_nmi_window(struct kvm_vcpu *vcpu)
3501 struct vcpu_svm *svm = to_svm(vcpu);
3503 if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
3505 return; /* IRET will cause a vm exit */
3507 if (!gif_set(svm)) {
3508 if (vgif_enabled(svm))
3509 svm_set_intercept(svm, INTERCEPT_STGI);
3510 return; /* STGI will cause a vm exit */
3514 * Something prevents NMI from been injected. Single step over possible
3515 * problem (IRET or exception injection or interrupt shadow)
3517 svm->nmi_singlestep_guest_rflags = svm_get_rflags(vcpu);
3518 svm->nmi_singlestep = true;
3519 svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
3522 static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
3527 static int svm_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
3532 void svm_flush_tlb(struct kvm_vcpu *vcpu)
3534 struct vcpu_svm *svm = to_svm(vcpu);
3537 * Flush only the current ASID even if the TLB flush was invoked via
3538 * kvm_flush_remote_tlbs(). Although flushing remote TLBs requires all
3539 * ASIDs to be flushed, KVM uses a single ASID for L1 and L2, and
3540 * unconditionally does a TLB flush on both nested VM-Enter and nested
3541 * VM-Exit (via kvm_mmu_reset_context()).
3543 if (static_cpu_has(X86_FEATURE_FLUSHBYASID))
3544 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
3546 svm->asid_generation--;
3549 static void svm_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t gva)
3551 struct vcpu_svm *svm = to_svm(vcpu);
3553 invlpga(gva, svm->vmcb->control.asid);
3556 static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
3560 static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
3562 struct vcpu_svm *svm = to_svm(vcpu);
3564 if (nested_svm_virtualize_tpr(vcpu))
3567 if (!svm_is_intercept(svm, INTERCEPT_CR8_WRITE)) {
3568 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
3569 kvm_set_cr8(vcpu, cr8);
3573 static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
3575 struct vcpu_svm *svm = to_svm(vcpu);
3578 if (nested_svm_virtualize_tpr(vcpu) ||
3579 kvm_vcpu_apicv_active(vcpu))
3582 cr8 = kvm_get_cr8(vcpu);
3583 svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
3584 svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
3587 static void svm_complete_interrupts(struct vcpu_svm *svm)
3591 u32 exitintinfo = svm->vmcb->control.exit_int_info;
3592 unsigned int3_injected = svm->int3_injected;
3594 svm->int3_injected = 0;
3597 * If we've made progress since setting HF_IRET_MASK, we've
3598 * executed an IRET and can allow NMI injection.
3600 if ((svm->vcpu.arch.hflags & HF_IRET_MASK) &&
3601 (sev_es_guest(svm->vcpu.kvm) ||
3602 kvm_rip_read(&svm->vcpu) != svm->nmi_iret_rip)) {
3603 svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
3604 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3607 svm->vcpu.arch.nmi_injected = false;
3608 kvm_clear_exception_queue(&svm->vcpu);
3609 kvm_clear_interrupt_queue(&svm->vcpu);
3611 if (!(exitintinfo & SVM_EXITINTINFO_VALID))
3614 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3616 vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
3617 type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
3620 case SVM_EXITINTINFO_TYPE_NMI:
3621 svm->vcpu.arch.nmi_injected = true;
3623 case SVM_EXITINTINFO_TYPE_EXEPT:
3625 * Never re-inject a #VC exception.
3627 if (vector == X86_TRAP_VC)
3631 * In case of software exceptions, do not reinject the vector,
3632 * but re-execute the instruction instead. Rewind RIP first
3633 * if we emulated INT3 before.
3635 if (kvm_exception_is_soft(vector)) {
3636 if (vector == BP_VECTOR && int3_injected &&
3637 kvm_is_linear_rip(&svm->vcpu, svm->int3_rip))
3638 kvm_rip_write(&svm->vcpu,
3639 kvm_rip_read(&svm->vcpu) -
3643 if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
3644 u32 err = svm->vmcb->control.exit_int_info_err;
3645 kvm_requeue_exception_e(&svm->vcpu, vector, err);
3648 kvm_requeue_exception(&svm->vcpu, vector);
3650 case SVM_EXITINTINFO_TYPE_INTR:
3651 kvm_queue_interrupt(&svm->vcpu, vector, false);
3658 static void svm_cancel_injection(struct kvm_vcpu *vcpu)
3660 struct vcpu_svm *svm = to_svm(vcpu);
3661 struct vmcb_control_area *control = &svm->vmcb->control;
3663 control->exit_int_info = control->event_inj;
3664 control->exit_int_info_err = control->event_inj_err;
3665 control->event_inj = 0;
3666 svm_complete_interrupts(svm);
3669 static fastpath_t svm_exit_handlers_fastpath(struct kvm_vcpu *vcpu)
3671 if (to_svm(vcpu)->vmcb->control.exit_code == SVM_EXIT_MSR &&
3672 to_svm(vcpu)->vmcb->control.exit_info_1)
3673 return handle_fastpath_set_msr_irqoff(vcpu);
3675 return EXIT_FASTPATH_NONE;
3678 void __svm_vcpu_run(unsigned long vmcb_pa, unsigned long *regs);
3680 static noinstr void svm_vcpu_enter_exit(struct kvm_vcpu *vcpu,
3681 struct vcpu_svm *svm)
3684 * VMENTER enables interrupts (host state), but the kernel state is
3685 * interrupts disabled when this is invoked. Also tell RCU about
3686 * it. This is the same logic as for exit_to_user_mode().
3688 * This ensures that e.g. latency analysis on the host observes
3689 * guest mode as interrupt enabled.
3691 * guest_enter_irqoff() informs context tracking about the
3692 * transition to guest mode and if enabled adjusts RCU state
3695 instrumentation_begin();
3696 trace_hardirqs_on_prepare();
3697 lockdep_hardirqs_on_prepare(CALLER_ADDR0);
3698 instrumentation_end();
3700 guest_enter_irqoff();
3701 lockdep_hardirqs_on(CALLER_ADDR0);
3703 if (sev_es_guest(svm->vcpu.kvm)) {
3704 __svm_sev_es_vcpu_run(svm->vmcb_pa);
3706 __svm_vcpu_run(svm->vmcb_pa, (unsigned long *)&svm->vcpu.arch.regs);
3708 #ifdef CONFIG_X86_64
3709 native_wrmsrl(MSR_GS_BASE, svm->host.gs_base);
3711 loadsegment(fs, svm->host.fs);
3712 #ifndef CONFIG_X86_32_LAZY_GS
3713 loadsegment(gs, svm->host.gs);
3719 * VMEXIT disables interrupts (host state), but tracing and lockdep
3720 * have them in state 'on' as recorded before entering guest mode.
3721 * Same as enter_from_user_mode().
3723 * guest_exit_irqoff() restores host context and reinstates RCU if
3724 * enabled and required.
3726 * This needs to be done before the below as native_read_msr()
3727 * contains a tracepoint and x86_spec_ctrl_restore_host() calls
3728 * into world and some more.
3730 lockdep_hardirqs_off(CALLER_ADDR0);
3731 guest_exit_irqoff();
3733 instrumentation_begin();
3734 trace_hardirqs_off_finish();
3735 instrumentation_end();
3738 static __no_kcsan fastpath_t svm_vcpu_run(struct kvm_vcpu *vcpu)
3740 struct vcpu_svm *svm = to_svm(vcpu);
3742 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
3743 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
3744 svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
3747 * Disable singlestep if we're injecting an interrupt/exception.
3748 * We don't want our modified rflags to be pushed on the stack where
3749 * we might not be able to easily reset them if we disabled NMI
3752 if (svm->nmi_singlestep && svm->vmcb->control.event_inj) {
3754 * Event injection happens before external interrupts cause a
3755 * vmexit and interrupts are disabled here, so smp_send_reschedule
3756 * is enough to force an immediate vmexit.
3758 disable_nmi_singlestep(svm);
3759 smp_send_reschedule(vcpu->cpu);
3764 sync_lapic_to_cr8(vcpu);
3766 if (unlikely(svm->asid != svm->vmcb->control.asid)) {
3767 svm->vmcb->control.asid = svm->asid;
3768 vmcb_mark_dirty(svm->vmcb, VMCB_ASID);
3770 svm->vmcb->save.cr2 = vcpu->arch.cr2;
3773 * Run with all-zero DR6 unless needed, so that we can get the exact cause
3776 if (unlikely(svm->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT))
3777 svm_set_dr6(svm, vcpu->arch.dr6);
3779 svm_set_dr6(svm, DR6_FIXED_1 | DR6_RTM);
3782 kvm_load_guest_xsave_state(vcpu);
3784 kvm_wait_lapic_expire(vcpu);
3787 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
3788 * it's non-zero. Since vmentry is serialising on affected CPUs, there
3789 * is no need to worry about the conditional branch over the wrmsr
3790 * being speculatively taken.
3792 x86_spec_ctrl_set_guest(svm->spec_ctrl, svm->virt_spec_ctrl);
3794 svm_vcpu_enter_exit(vcpu, svm);
3797 * We do not use IBRS in the kernel. If this vCPU has used the
3798 * SPEC_CTRL MSR it may have left it on; save the value and
3799 * turn it off. This is much more efficient than blindly adding
3800 * it to the atomic save/restore list. Especially as the former
3801 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
3803 * For non-nested case:
3804 * If the L01 MSR bitmap does not intercept the MSR, then we need to
3808 * If the L02 MSR bitmap does not intercept the MSR, then we need to
3811 if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
3812 svm->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
3814 if (!sev_es_guest(svm->vcpu.kvm))
3817 x86_spec_ctrl_restore_host(svm->spec_ctrl, svm->virt_spec_ctrl);
3819 if (!sev_es_guest(svm->vcpu.kvm)) {
3820 vcpu->arch.cr2 = svm->vmcb->save.cr2;
3821 vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
3822 vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
3823 vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
3826 if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
3827 kvm_before_interrupt(&svm->vcpu);
3829 kvm_load_host_xsave_state(vcpu);
3832 /* Any pending NMI will happen here */
3834 if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
3835 kvm_after_interrupt(&svm->vcpu);
3837 sync_cr8_to_lapic(vcpu);
3840 if (is_guest_mode(&svm->vcpu)) {
3841 sync_nested_vmcb_control(svm);
3842 svm->nested.nested_run_pending = 0;
3845 svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
3846 vmcb_mark_all_clean(svm->vmcb);
3848 /* if exit due to PF check for async PF */
3849 if (svm->vmcb->control.exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR)
3850 svm->vcpu.arch.apf.host_apf_flags =
3851 kvm_read_and_reset_apf_flags();
3854 vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
3855 vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
3859 * We need to handle MC intercepts here before the vcpu has a chance to
3860 * change the physical cpu
3862 if (unlikely(svm->vmcb->control.exit_code ==
3863 SVM_EXIT_EXCP_BASE + MC_VECTOR))
3864 svm_handle_mce(svm);
3866 svm_complete_interrupts(svm);
3868 if (is_guest_mode(vcpu))
3869 return EXIT_FASTPATH_NONE;
3871 return svm_exit_handlers_fastpath(vcpu);
3874 static void svm_load_mmu_pgd(struct kvm_vcpu *vcpu, unsigned long root,
3877 struct vcpu_svm *svm = to_svm(vcpu);
3880 cr3 = __sme_set(root);
3882 svm->vmcb->control.nested_cr3 = cr3;
3883 vmcb_mark_dirty(svm->vmcb, VMCB_NPT);
3885 /* Loading L2's CR3 is handled by enter_svm_guest_mode. */
3886 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3888 cr3 = vcpu->arch.cr3;
3891 svm->vmcb->save.cr3 = cr3;
3892 vmcb_mark_dirty(svm->vmcb, VMCB_CR);
3895 static int is_disabled(void)
3899 rdmsrl(MSR_VM_CR, vm_cr);
3900 if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
3907 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
3910 * Patch in the VMMCALL instruction:
3912 hypercall[0] = 0x0f;
3913 hypercall[1] = 0x01;
3914 hypercall[2] = 0xd9;
3917 static int __init svm_check_processor_compat(void)
3922 static bool svm_cpu_has_accelerated_tpr(void)
3928 * The kvm parameter can be NULL (module initialization, or invocation before
3929 * VM creation). Be sure to check the kvm parameter before using it.
3931 static bool svm_has_emulated_msr(struct kvm *kvm, u32 index)
3934 case MSR_IA32_MCG_EXT_CTL:
3935 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3937 case MSR_IA32_SMBASE:
3938 /* SEV-ES guests do not support SMM, so report false */
3939 if (kvm && sev_es_guest(kvm))
3949 static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
3954 static void svm_vcpu_after_set_cpuid(struct kvm_vcpu *vcpu)
3956 struct vcpu_svm *svm = to_svm(vcpu);
3957 struct kvm_cpuid_entry2 *best;
3959 vcpu->arch.xsaves_enabled = guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
3960 boot_cpu_has(X86_FEATURE_XSAVE) &&
3961 boot_cpu_has(X86_FEATURE_XSAVES);
3963 /* Update nrips enabled cache */
3964 svm->nrips_enabled = kvm_cpu_cap_has(X86_FEATURE_NRIPS) &&
3965 guest_cpuid_has(&svm->vcpu, X86_FEATURE_NRIPS);
3967 /* Check again if INVPCID interception if required */
3968 svm_check_invpcid(svm);
3970 /* For sev guests, the memory encryption bit is not reserved in CR3. */
3971 if (sev_guest(vcpu->kvm)) {
3972 best = kvm_find_cpuid_entry(vcpu, 0x8000001F, 0);
3974 vcpu->arch.cr3_lm_rsvd_bits &= ~(1UL << (best->ebx & 0x3f));
3977 if (!kvm_vcpu_apicv_active(vcpu))
3981 * AVIC does not work with an x2APIC mode guest. If the X2APIC feature
3982 * is exposed to the guest, disable AVIC.
3984 if (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC))
3985 kvm_request_apicv_update(vcpu->kvm, false,
3986 APICV_INHIBIT_REASON_X2APIC);
3989 * Currently, AVIC does not work with nested virtualization.
3990 * So, we disable AVIC when cpuid for SVM is set in the L1 guest.
3992 if (nested && guest_cpuid_has(vcpu, X86_FEATURE_SVM))
3993 kvm_request_apicv_update(vcpu->kvm, false,
3994 APICV_INHIBIT_REASON_NESTED);
3997 static bool svm_has_wbinvd_exit(void)
4002 #define PRE_EX(exit) { .exit_code = (exit), \
4003 .stage = X86_ICPT_PRE_EXCEPT, }
4004 #define POST_EX(exit) { .exit_code = (exit), \
4005 .stage = X86_ICPT_POST_EXCEPT, }
4006 #define POST_MEM(exit) { .exit_code = (exit), \
4007 .stage = X86_ICPT_POST_MEMACCESS, }
4009 static const struct __x86_intercept {
4011 enum x86_intercept_stage stage;
4012 } x86_intercept_map[] = {
4013 [x86_intercept_cr_read] = POST_EX(SVM_EXIT_READ_CR0),
4014 [x86_intercept_cr_write] = POST_EX(SVM_EXIT_WRITE_CR0),
4015 [x86_intercept_clts] = POST_EX(SVM_EXIT_WRITE_CR0),
4016 [x86_intercept_lmsw] = POST_EX(SVM_EXIT_WRITE_CR0),
4017 [x86_intercept_smsw] = POST_EX(SVM_EXIT_READ_CR0),
4018 [x86_intercept_dr_read] = POST_EX(SVM_EXIT_READ_DR0),
4019 [x86_intercept_dr_write] = POST_EX(SVM_EXIT_WRITE_DR0),
4020 [x86_intercept_sldt] = POST_EX(SVM_EXIT_LDTR_READ),
4021 [x86_intercept_str] = POST_EX(SVM_EXIT_TR_READ),
4022 [x86_intercept_lldt] = POST_EX(SVM_EXIT_LDTR_WRITE),
4023 [x86_intercept_ltr] = POST_EX(SVM_EXIT_TR_WRITE),
4024 [x86_intercept_sgdt] = POST_EX(SVM_EXIT_GDTR_READ),
4025 [x86_intercept_sidt] = POST_EX(SVM_EXIT_IDTR_READ),
4026 [x86_intercept_lgdt] = POST_EX(SVM_EXIT_GDTR_WRITE),
4027 [x86_intercept_lidt] = POST_EX(SVM_EXIT_IDTR_WRITE),
4028 [x86_intercept_vmrun] = POST_EX(SVM_EXIT_VMRUN),
4029 [x86_intercept_vmmcall] = POST_EX(SVM_EXIT_VMMCALL),
4030 [x86_intercept_vmload] = POST_EX(SVM_EXIT_VMLOAD),
4031 [x86_intercept_vmsave] = POST_EX(SVM_EXIT_VMSAVE),
4032 [x86_intercept_stgi] = POST_EX(SVM_EXIT_STGI),
4033 [x86_intercept_clgi] = POST_EX(SVM_EXIT_CLGI),
4034 [x86_intercept_skinit] = POST_EX(SVM_EXIT_SKINIT),
4035 [x86_intercept_invlpga] = POST_EX(SVM_EXIT_INVLPGA),
4036 [x86_intercept_rdtscp] = POST_EX(SVM_EXIT_RDTSCP),
4037 [x86_intercept_monitor] = POST_MEM(SVM_EXIT_MONITOR),
4038 [x86_intercept_mwait] = POST_EX(SVM_EXIT_MWAIT),
4039 [x86_intercept_invlpg] = POST_EX(SVM_EXIT_INVLPG),
4040 [x86_intercept_invd] = POST_EX(SVM_EXIT_INVD),
4041 [x86_intercept_wbinvd] = POST_EX(SVM_EXIT_WBINVD),
4042 [x86_intercept_wrmsr] = POST_EX(SVM_EXIT_MSR),
4043 [x86_intercept_rdtsc] = POST_EX(SVM_EXIT_RDTSC),
4044 [x86_intercept_rdmsr] = POST_EX(SVM_EXIT_MSR),
4045 [x86_intercept_rdpmc] = POST_EX(SVM_EXIT_RDPMC),
4046 [x86_intercept_cpuid] = PRE_EX(SVM_EXIT_CPUID),
4047 [x86_intercept_rsm] = PRE_EX(SVM_EXIT_RSM),
4048 [x86_intercept_pause] = PRE_EX(SVM_EXIT_PAUSE),
4049 [x86_intercept_pushf] = PRE_EX(SVM_EXIT_PUSHF),
4050 [x86_intercept_popf] = PRE_EX(SVM_EXIT_POPF),
4051 [x86_intercept_intn] = PRE_EX(SVM_EXIT_SWINT),
4052 [x86_intercept_iret] = PRE_EX(SVM_EXIT_IRET),
4053 [x86_intercept_icebp] = PRE_EX(SVM_EXIT_ICEBP),
4054 [x86_intercept_hlt] = POST_EX(SVM_EXIT_HLT),
4055 [x86_intercept_in] = POST_EX(SVM_EXIT_IOIO),
4056 [x86_intercept_ins] = POST_EX(SVM_EXIT_IOIO),
4057 [x86_intercept_out] = POST_EX(SVM_EXIT_IOIO),
4058 [x86_intercept_outs] = POST_EX(SVM_EXIT_IOIO),
4059 [x86_intercept_xsetbv] = PRE_EX(SVM_EXIT_XSETBV),
4066 static int svm_check_intercept(struct kvm_vcpu *vcpu,
4067 struct x86_instruction_info *info,
4068 enum x86_intercept_stage stage,
4069 struct x86_exception *exception)
4071 struct vcpu_svm *svm = to_svm(vcpu);
4072 int vmexit, ret = X86EMUL_CONTINUE;
4073 struct __x86_intercept icpt_info;
4074 struct vmcb *vmcb = svm->vmcb;
4076 if (info->intercept >= ARRAY_SIZE(x86_intercept_map))
4079 icpt_info = x86_intercept_map[info->intercept];
4081 if (stage != icpt_info.stage)
4084 switch (icpt_info.exit_code) {
4085 case SVM_EXIT_READ_CR0:
4086 if (info->intercept == x86_intercept_cr_read)
4087 icpt_info.exit_code += info->modrm_reg;
4089 case SVM_EXIT_WRITE_CR0: {
4090 unsigned long cr0, val;
4092 if (info->intercept == x86_intercept_cr_write)
4093 icpt_info.exit_code += info->modrm_reg;
4095 if (icpt_info.exit_code != SVM_EXIT_WRITE_CR0 ||
4096 info->intercept == x86_intercept_clts)
4099 if (!(vmcb_is_intercept(&svm->nested.ctl,
4100 INTERCEPT_SELECTIVE_CR0)))
4103 cr0 = vcpu->arch.cr0 & ~SVM_CR0_SELECTIVE_MASK;
4104 val = info->src_val & ~SVM_CR0_SELECTIVE_MASK;
4106 if (info->intercept == x86_intercept_lmsw) {
4109 /* lmsw can't clear PE - catch this here */
4110 if (cr0 & X86_CR0_PE)
4115 icpt_info.exit_code = SVM_EXIT_CR0_SEL_WRITE;
4119 case SVM_EXIT_READ_DR0:
4120 case SVM_EXIT_WRITE_DR0:
4121 icpt_info.exit_code += info->modrm_reg;
4124 if (info->intercept == x86_intercept_wrmsr)
4125 vmcb->control.exit_info_1 = 1;
4127 vmcb->control.exit_info_1 = 0;
4129 case SVM_EXIT_PAUSE:
4131 * We get this for NOP only, but pause
4132 * is rep not, check this here
4134 if (info->rep_prefix != REPE_PREFIX)
4137 case SVM_EXIT_IOIO: {
4141 if (info->intercept == x86_intercept_in ||
4142 info->intercept == x86_intercept_ins) {
4143 exit_info = ((info->src_val & 0xffff) << 16) |
4145 bytes = info->dst_bytes;
4147 exit_info = (info->dst_val & 0xffff) << 16;
4148 bytes = info->src_bytes;
4151 if (info->intercept == x86_intercept_outs ||
4152 info->intercept == x86_intercept_ins)
4153 exit_info |= SVM_IOIO_STR_MASK;
4155 if (info->rep_prefix)
4156 exit_info |= SVM_IOIO_REP_MASK;
4158 bytes = min(bytes, 4u);
4160 exit_info |= bytes << SVM_IOIO_SIZE_SHIFT;
4162 exit_info |= (u32)info->ad_bytes << (SVM_IOIO_ASIZE_SHIFT - 1);
4164 vmcb->control.exit_info_1 = exit_info;
4165 vmcb->control.exit_info_2 = info->next_rip;
4173 /* TODO: Advertise NRIPS to guest hypervisor unconditionally */
4174 if (static_cpu_has(X86_FEATURE_NRIPS))
4175 vmcb->control.next_rip = info->next_rip;
4176 vmcb->control.exit_code = icpt_info.exit_code;
4177 vmexit = nested_svm_exit_handled(svm);
4179 ret = (vmexit == NESTED_EXIT_DONE) ? X86EMUL_INTERCEPTED
4186 static void svm_handle_exit_irqoff(struct kvm_vcpu *vcpu)
4190 static void svm_sched_in(struct kvm_vcpu *vcpu, int cpu)
4192 if (!kvm_pause_in_guest(vcpu->kvm))
4193 shrink_ple_window(vcpu);
4196 static void svm_setup_mce(struct kvm_vcpu *vcpu)
4198 /* [63:9] are reserved. */
4199 vcpu->arch.mcg_cap &= 0x1ff;
4202 bool svm_smi_blocked(struct kvm_vcpu *vcpu)
4204 struct vcpu_svm *svm = to_svm(vcpu);
4206 /* Per APM Vol.2 15.22.2 "Response to SMI" */
4210 return is_smm(vcpu);
4213 static int svm_smi_allowed(struct kvm_vcpu *vcpu, bool for_injection)
4215 struct vcpu_svm *svm = to_svm(vcpu);
4216 if (svm->nested.nested_run_pending)
4219 /* An SMI must not be injected into L2 if it's supposed to VM-Exit. */
4220 if (for_injection && is_guest_mode(vcpu) && nested_exit_on_smi(svm))
4223 return !svm_smi_blocked(vcpu);
4226 static int svm_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
4228 struct vcpu_svm *svm = to_svm(vcpu);
4231 if (is_guest_mode(vcpu)) {
4232 /* FED8h - SVM Guest */
4233 put_smstate(u64, smstate, 0x7ed8, 1);
4234 /* FEE0h - SVM Guest VMCB Physical Address */
4235 put_smstate(u64, smstate, 0x7ee0, svm->nested.vmcb12_gpa);
4237 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
4238 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
4239 svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
4241 ret = nested_svm_vmexit(svm);
4248 static int svm_pre_leave_smm(struct kvm_vcpu *vcpu, const char *smstate)
4250 struct vcpu_svm *svm = to_svm(vcpu);
4251 struct kvm_host_map map;
4254 if (guest_cpuid_has(vcpu, X86_FEATURE_LM)) {
4255 u64 saved_efer = GET_SMSTATE(u64, smstate, 0x7ed0);
4256 u64 guest = GET_SMSTATE(u64, smstate, 0x7ed8);
4257 u64 vmcb12_gpa = GET_SMSTATE(u64, smstate, 0x7ee0);
4260 if (!guest_cpuid_has(vcpu, X86_FEATURE_SVM))
4263 if (!(saved_efer & EFER_SVME))
4266 if (kvm_vcpu_map(&svm->vcpu,
4267 gpa_to_gfn(vmcb12_gpa), &map) == -EINVAL)
4270 if (svm_allocate_nested(svm))
4273 ret = enter_svm_guest_mode(svm, vmcb12_gpa, map.hva);
4274 kvm_vcpu_unmap(&svm->vcpu, &map, true);
4281 static void enable_smi_window(struct kvm_vcpu *vcpu)
4283 struct vcpu_svm *svm = to_svm(vcpu);
4285 if (!gif_set(svm)) {
4286 if (vgif_enabled(svm))
4287 svm_set_intercept(svm, INTERCEPT_STGI);
4288 /* STGI will cause a vm exit */
4290 /* We must be in SMM; RSM will cause a vmexit anyway. */
4294 static bool svm_can_emulate_instruction(struct kvm_vcpu *vcpu, void *insn, int insn_len)
4296 bool smep, smap, is_user;
4300 * When the guest is an SEV-ES guest, emulation is not possible.
4302 if (sev_es_guest(vcpu->kvm))
4306 * Detect and workaround Errata 1096 Fam_17h_00_0Fh.
4309 * When CPU raise #NPF on guest data access and vCPU CR4.SMAP=1, it is
4310 * possible that CPU microcode implementing DecodeAssist will fail
4311 * to read bytes of instruction which caused #NPF. In this case,
4312 * GuestIntrBytes field of the VMCB on a VMEXIT will incorrectly
4313 * return 0 instead of the correct guest instruction bytes.
4315 * This happens because CPU microcode reading instruction bytes
4316 * uses a special opcode which attempts to read data using CPL=0
4317 * priviledges. The microcode reads CS:RIP and if it hits a SMAP
4318 * fault, it gives up and returns no instruction bytes.
4321 * We reach here in case CPU supports DecodeAssist, raised #NPF and
4322 * returned 0 in GuestIntrBytes field of the VMCB.
4323 * First, errata can only be triggered in case vCPU CR4.SMAP=1.
4324 * Second, if vCPU CR4.SMEP=1, errata could only be triggered
4325 * in case vCPU CPL==3 (Because otherwise guest would have triggered
4326 * a SMEP fault instead of #NPF).
4327 * Otherwise, vCPU CR4.SMEP=0, errata could be triggered by any vCPU CPL.
4328 * As most guests enable SMAP if they have also enabled SMEP, use above
4329 * logic in order to attempt minimize false-positive of detecting errata
4330 * while still preserving all cases semantic correctness.
4333 * To determine what instruction the guest was executing, the hypervisor
4334 * will have to decode the instruction at the instruction pointer.
4336 * In non SEV guest, hypervisor will be able to read the guest
4337 * memory to decode the instruction pointer when insn_len is zero
4338 * so we return true to indicate that decoding is possible.
4340 * But in the SEV guest, the guest memory is encrypted with the
4341 * guest specific key and hypervisor will not be able to decode the
4342 * instruction pointer so we will not able to workaround it. Lets
4343 * print the error and request to kill the guest.
4345 if (likely(!insn || insn_len))
4349 * If RIP is invalid, go ahead with emulation which will cause an
4350 * internal error exit.
4352 if (!kvm_vcpu_gfn_to_memslot(vcpu, kvm_rip_read(vcpu) >> PAGE_SHIFT))
4355 cr4 = kvm_read_cr4(vcpu);
4356 smep = cr4 & X86_CR4_SMEP;
4357 smap = cr4 & X86_CR4_SMAP;
4358 is_user = svm_get_cpl(vcpu) == 3;
4359 if (smap && (!smep || is_user)) {
4360 if (!sev_guest(vcpu->kvm))
4363 pr_err_ratelimited("KVM: SEV Guest triggered AMD Erratum 1096\n");
4364 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4370 static bool svm_apic_init_signal_blocked(struct kvm_vcpu *vcpu)
4372 struct vcpu_svm *svm = to_svm(vcpu);
4375 * TODO: Last condition latch INIT signals on vCPU when
4376 * vCPU is in guest-mode and vmcb12 defines intercept on INIT.
4377 * To properly emulate the INIT intercept,
4378 * svm_check_nested_events() should call nested_svm_vmexit()
4379 * if an INIT signal is pending.
4381 return !gif_set(svm) ||
4382 (vmcb_is_intercept(&svm->vmcb->control, INTERCEPT_INIT));
4385 static void svm_vm_destroy(struct kvm *kvm)
4387 avic_vm_destroy(kvm);
4388 sev_vm_destroy(kvm);
4391 static int svm_vm_init(struct kvm *kvm)
4393 if (!pause_filter_count || !pause_filter_thresh)
4394 kvm->arch.pause_in_guest = true;
4397 int ret = avic_vm_init(kvm);
4402 kvm_apicv_init(kvm, avic);
4406 static struct kvm_x86_ops svm_x86_ops __initdata = {
4407 .hardware_unsetup = svm_hardware_teardown,
4408 .hardware_enable = svm_hardware_enable,
4409 .hardware_disable = svm_hardware_disable,
4410 .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
4411 .has_emulated_msr = svm_has_emulated_msr,
4413 .vcpu_create = svm_create_vcpu,
4414 .vcpu_free = svm_free_vcpu,
4415 .vcpu_reset = svm_vcpu_reset,
4417 .vm_size = sizeof(struct kvm_svm),
4418 .vm_init = svm_vm_init,
4419 .vm_destroy = svm_vm_destroy,
4421 .prepare_guest_switch = svm_prepare_guest_switch,
4422 .vcpu_load = svm_vcpu_load,
4423 .vcpu_put = svm_vcpu_put,
4424 .vcpu_blocking = svm_vcpu_blocking,
4425 .vcpu_unblocking = svm_vcpu_unblocking,
4427 .update_exception_bitmap = update_exception_bitmap,
4428 .get_msr_feature = svm_get_msr_feature,
4429 .get_msr = svm_get_msr,
4430 .set_msr = svm_set_msr,
4431 .get_segment_base = svm_get_segment_base,
4432 .get_segment = svm_get_segment,
4433 .set_segment = svm_set_segment,
4434 .get_cpl = svm_get_cpl,
4435 .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
4436 .set_cr0 = svm_set_cr0,
4437 .is_valid_cr4 = svm_is_valid_cr4,
4438 .set_cr4 = svm_set_cr4,
4439 .set_efer = svm_set_efer,
4440 .get_idt = svm_get_idt,
4441 .set_idt = svm_set_idt,
4442 .get_gdt = svm_get_gdt,
4443 .set_gdt = svm_set_gdt,
4444 .set_dr7 = svm_set_dr7,
4445 .sync_dirty_debug_regs = svm_sync_dirty_debug_regs,
4446 .cache_reg = svm_cache_reg,
4447 .get_rflags = svm_get_rflags,
4448 .set_rflags = svm_set_rflags,
4450 .tlb_flush_all = svm_flush_tlb,
4451 .tlb_flush_current = svm_flush_tlb,
4452 .tlb_flush_gva = svm_flush_tlb_gva,
4453 .tlb_flush_guest = svm_flush_tlb,
4455 .run = svm_vcpu_run,
4456 .handle_exit = handle_exit,
4457 .skip_emulated_instruction = skip_emulated_instruction,
4458 .update_emulated_instruction = NULL,
4459 .set_interrupt_shadow = svm_set_interrupt_shadow,
4460 .get_interrupt_shadow = svm_get_interrupt_shadow,
4461 .patch_hypercall = svm_patch_hypercall,
4462 .set_irq = svm_set_irq,
4463 .set_nmi = svm_inject_nmi,
4464 .queue_exception = svm_queue_exception,
4465 .cancel_injection = svm_cancel_injection,
4466 .interrupt_allowed = svm_interrupt_allowed,
4467 .nmi_allowed = svm_nmi_allowed,
4468 .get_nmi_mask = svm_get_nmi_mask,
4469 .set_nmi_mask = svm_set_nmi_mask,
4470 .enable_nmi_window = enable_nmi_window,
4471 .enable_irq_window = enable_irq_window,
4472 .update_cr8_intercept = update_cr8_intercept,
4473 .set_virtual_apic_mode = svm_set_virtual_apic_mode,
4474 .refresh_apicv_exec_ctrl = svm_refresh_apicv_exec_ctrl,
4475 .check_apicv_inhibit_reasons = svm_check_apicv_inhibit_reasons,
4476 .pre_update_apicv_exec_ctrl = svm_pre_update_apicv_exec_ctrl,
4477 .load_eoi_exitmap = svm_load_eoi_exitmap,
4478 .hwapic_irr_update = svm_hwapic_irr_update,
4479 .hwapic_isr_update = svm_hwapic_isr_update,
4480 .sync_pir_to_irr = kvm_lapic_find_highest_irr,
4481 .apicv_post_state_restore = avic_post_state_restore,
4483 .set_tss_addr = svm_set_tss_addr,
4484 .set_identity_map_addr = svm_set_identity_map_addr,
4485 .get_mt_mask = svm_get_mt_mask,
4487 .get_exit_info = svm_get_exit_info,
4489 .vcpu_after_set_cpuid = svm_vcpu_after_set_cpuid,
4491 .has_wbinvd_exit = svm_has_wbinvd_exit,
4493 .write_l1_tsc_offset = svm_write_l1_tsc_offset,
4495 .load_mmu_pgd = svm_load_mmu_pgd,
4497 .check_intercept = svm_check_intercept,
4498 .handle_exit_irqoff = svm_handle_exit_irqoff,
4500 .request_immediate_exit = __kvm_request_immediate_exit,
4502 .sched_in = svm_sched_in,
4504 .pmu_ops = &amd_pmu_ops,
4505 .nested_ops = &svm_nested_ops,
4507 .deliver_posted_interrupt = svm_deliver_avic_intr,
4508 .dy_apicv_has_pending_interrupt = svm_dy_apicv_has_pending_interrupt,
4509 .update_pi_irte = svm_update_pi_irte,
4510 .setup_mce = svm_setup_mce,
4512 .smi_allowed = svm_smi_allowed,
4513 .pre_enter_smm = svm_pre_enter_smm,
4514 .pre_leave_smm = svm_pre_leave_smm,
4515 .enable_smi_window = enable_smi_window,
4517 .mem_enc_op = svm_mem_enc_op,
4518 .mem_enc_reg_region = svm_register_enc_region,
4519 .mem_enc_unreg_region = svm_unregister_enc_region,
4521 .can_emulate_instruction = svm_can_emulate_instruction,
4523 .apic_init_signal_blocked = svm_apic_init_signal_blocked,
4525 .msr_filter_changed = svm_msr_filter_changed,
4526 .complete_emulated_msr = svm_complete_emulated_msr,
4529 static struct kvm_x86_init_ops svm_init_ops __initdata = {
4530 .cpu_has_kvm_support = has_svm,
4531 .disabled_by_bios = is_disabled,
4532 .hardware_setup = svm_hardware_setup,
4533 .check_processor_compatibility = svm_check_processor_compat,
4535 .runtime_ops = &svm_x86_ops,
4538 static int __init svm_init(void)
4540 __unused_size_checks();
4542 return kvm_init(&svm_init_ops, sizeof(struct vcpu_svm),
4543 __alignof__(struct vcpu_svm), THIS_MODULE);
4546 static void __exit svm_exit(void)
4551 module_init(svm_init)
4552 module_exit(svm_exit)