1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Yaniv Kamay <yaniv@qumranet.com>
12 * Avi Kivity <avi@qumranet.com>
15 #define pr_fmt(fmt) "SVM: " fmt
17 #include <linux/kvm_types.h>
18 #include <linux/kvm_host.h>
19 #include <linux/kernel.h>
21 #include <asm/msr-index.h>
22 #include <asm/debugreg.h>
24 #include "kvm_emulate.h"
32 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
33 struct x86_exception *fault)
35 struct vcpu_svm *svm = to_svm(vcpu);
37 if (svm->vmcb->control.exit_code != SVM_EXIT_NPF) {
39 * TODO: track the cause of the nested page fault, and
40 * correctly fill in the high bits of exit_info_1.
42 svm->vmcb->control.exit_code = SVM_EXIT_NPF;
43 svm->vmcb->control.exit_code_hi = 0;
44 svm->vmcb->control.exit_info_1 = (1ULL << 32);
45 svm->vmcb->control.exit_info_2 = fault->address;
48 svm->vmcb->control.exit_info_1 &= ~0xffffffffULL;
49 svm->vmcb->control.exit_info_1 |= fault->error_code;
51 nested_svm_vmexit(svm);
54 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
56 struct vcpu_svm *svm = to_svm(vcpu);
57 u64 cr3 = svm->nested.ctl.nested_cr3;
61 ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(__sme_clr(cr3)), &pdpte,
62 offset_in_page(cr3) + index * 8, 8);
68 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
70 struct vcpu_svm *svm = to_svm(vcpu);
72 return svm->nested.ctl.nested_cr3;
75 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
77 struct vcpu_svm *svm = to_svm(vcpu);
78 struct vmcb *hsave = svm->nested.hsave;
80 WARN_ON(mmu_is_nested(vcpu));
82 vcpu->arch.mmu = &vcpu->arch.guest_mmu;
83 kvm_init_shadow_npt_mmu(vcpu, X86_CR0_PG, hsave->save.cr4, hsave->save.efer,
84 svm->nested.ctl.nested_cr3);
85 vcpu->arch.mmu->get_guest_pgd = nested_svm_get_tdp_cr3;
86 vcpu->arch.mmu->get_pdptr = nested_svm_get_tdp_pdptr;
87 vcpu->arch.mmu->inject_page_fault = nested_svm_inject_npf_exit;
88 reset_shadow_zero_bits_mask(vcpu, vcpu->arch.mmu);
89 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
92 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
94 vcpu->arch.mmu = &vcpu->arch.root_mmu;
95 vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
98 void recalc_intercepts(struct vcpu_svm *svm)
100 struct vmcb_control_area *c, *h, *g;
103 vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
105 if (!is_guest_mode(&svm->vcpu))
108 c = &svm->vmcb->control;
109 h = &svm->nested.hsave->control;
110 g = &svm->nested.ctl;
112 for (i = 0; i < MAX_INTERCEPT; i++)
113 c->intercepts[i] = h->intercepts[i];
115 if (g->int_ctl & V_INTR_MASKING_MASK) {
116 /* We only want the cr8 intercept bits of L1 */
117 vmcb_clr_intercept(c, INTERCEPT_CR8_READ);
118 vmcb_clr_intercept(c, INTERCEPT_CR8_WRITE);
121 * Once running L2 with HF_VINTR_MASK, EFLAGS.IF does not
122 * affect any interrupt we may want to inject; therefore,
123 * interrupt window vmexits are irrelevant to L0.
125 vmcb_clr_intercept(c, INTERCEPT_VINTR);
128 /* We don't want to see VMMCALLs from a nested guest */
129 vmcb_clr_intercept(c, INTERCEPT_VMMCALL);
131 for (i = 0; i < MAX_INTERCEPT; i++)
132 c->intercepts[i] |= g->intercepts[i];
135 static void copy_vmcb_control_area(struct vmcb_control_area *dst,
136 struct vmcb_control_area *from)
140 for (i = 0; i < MAX_INTERCEPT; i++)
141 dst->intercepts[i] = from->intercepts[i];
143 dst->iopm_base_pa = from->iopm_base_pa;
144 dst->msrpm_base_pa = from->msrpm_base_pa;
145 dst->tsc_offset = from->tsc_offset;
146 /* asid not copied, it is handled manually for svm->vmcb. */
147 dst->tlb_ctl = from->tlb_ctl;
148 dst->int_ctl = from->int_ctl;
149 dst->int_vector = from->int_vector;
150 dst->int_state = from->int_state;
151 dst->exit_code = from->exit_code;
152 dst->exit_code_hi = from->exit_code_hi;
153 dst->exit_info_1 = from->exit_info_1;
154 dst->exit_info_2 = from->exit_info_2;
155 dst->exit_int_info = from->exit_int_info;
156 dst->exit_int_info_err = from->exit_int_info_err;
157 dst->nested_ctl = from->nested_ctl;
158 dst->event_inj = from->event_inj;
159 dst->event_inj_err = from->event_inj_err;
160 dst->nested_cr3 = from->nested_cr3;
161 dst->virt_ext = from->virt_ext;
162 dst->pause_filter_count = from->pause_filter_count;
163 dst->pause_filter_thresh = from->pause_filter_thresh;
166 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
169 * This function merges the msr permission bitmaps of kvm and the
170 * nested vmcb. It is optimized in that it only merges the parts where
171 * the kvm msr permission bitmap may contain zero bits
175 if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
178 for (i = 0; i < MSRPM_OFFSETS; i++) {
182 if (msrpm_offsets[i] == 0xffffffff)
185 p = msrpm_offsets[i];
186 offset = svm->nested.ctl.msrpm_base_pa + (p * 4);
188 if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
191 svm->nested.msrpm[p] = svm->msrpm[p] | value;
194 svm->vmcb->control.msrpm_base_pa = __sme_set(__pa(svm->nested.msrpm));
199 static bool svm_get_nested_state_pages(struct kvm_vcpu *vcpu)
201 struct vcpu_svm *svm = to_svm(vcpu);
202 if (!nested_svm_vmrun_msrpm(svm)) {
203 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
204 vcpu->run->internal.suberror =
205 KVM_INTERNAL_ERROR_EMULATION;
206 vcpu->run->internal.ndata = 0;
213 static bool nested_vmcb_check_controls(struct vmcb_control_area *control)
215 if ((vmcb_is_intercept(control, INTERCEPT_VMRUN)) == 0)
218 if (control->asid == 0)
221 if ((control->nested_ctl & SVM_NESTED_CTL_NP_ENABLE) &&
228 static bool nested_vmcb_check_save(struct vcpu_svm *svm, struct vmcb *vmcb12)
233 * FIXME: these should be done after copying the fields,
234 * to avoid TOC/TOU races. For these save area checks
235 * the possible damage is limited since kvm_set_cr0 and
236 * kvm_set_cr4 handle failure; EFER_SVME is an exception
237 * so it is force-set later in nested_prepare_vmcb_save.
239 if ((vmcb12->save.efer & EFER_SVME) == 0)
242 if (((vmcb12->save.cr0 & X86_CR0_CD) == 0) && (vmcb12->save.cr0 & X86_CR0_NW))
245 if (!kvm_dr6_valid(vmcb12->save.dr6) || !kvm_dr7_valid(vmcb12->save.dr7))
248 vmcb12_lma = (vmcb12->save.efer & EFER_LME) && (vmcb12->save.cr0 & X86_CR0_PG);
251 if (vmcb12->save.cr4 & X86_CR4_PAE) {
252 if (vmcb12->save.cr3 & MSR_CR3_LEGACY_PAE_RESERVED_MASK)
255 if (vmcb12->save.cr3 & MSR_CR3_LEGACY_RESERVED_MASK)
259 if (!(vmcb12->save.cr4 & X86_CR4_PAE) ||
260 !(vmcb12->save.cr0 & X86_CR0_PE) ||
261 (vmcb12->save.cr3 & MSR_CR3_LONG_MBZ_MASK))
264 if (kvm_valid_cr4(&svm->vcpu, vmcb12->save.cr4))
270 static void load_nested_vmcb_control(struct vcpu_svm *svm,
271 struct vmcb_control_area *control)
273 copy_vmcb_control_area(&svm->nested.ctl, control);
275 /* Copy it here because nested_svm_check_controls will check it. */
276 svm->nested.ctl.asid = control->asid;
277 svm->nested.ctl.msrpm_base_pa &= ~0x0fffULL;
278 svm->nested.ctl.iopm_base_pa &= ~0x0fffULL;
282 * Synchronize fields that are written by the processor, so that
283 * they can be copied back into the nested_vmcb.
285 void sync_nested_vmcb_control(struct vcpu_svm *svm)
288 svm->nested.ctl.event_inj = svm->vmcb->control.event_inj;
289 svm->nested.ctl.event_inj_err = svm->vmcb->control.event_inj_err;
291 /* Only a few fields of int_ctl are written by the processor. */
292 mask = V_IRQ_MASK | V_TPR_MASK;
293 if (!(svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK) &&
294 svm_is_intercept(svm, INTERCEPT_VINTR)) {
296 * In order to request an interrupt window, L0 is usurping
297 * svm->vmcb->control.int_ctl and possibly setting V_IRQ
298 * even if it was clear in L1's VMCB. Restoring it would be
299 * wrong. However, in this case V_IRQ will remain true until
300 * interrupt_window_interception calls svm_clear_vintr and
301 * restores int_ctl. We can just leave it aside.
305 svm->nested.ctl.int_ctl &= ~mask;
306 svm->nested.ctl.int_ctl |= svm->vmcb->control.int_ctl & mask;
310 * Transfer any event that L0 or L1 wanted to inject into L2 to
313 static void nested_vmcb_save_pending_event(struct vcpu_svm *svm,
316 struct kvm_vcpu *vcpu = &svm->vcpu;
317 u32 exit_int_info = 0;
320 if (vcpu->arch.exception.injected) {
321 nr = vcpu->arch.exception.nr;
322 exit_int_info = nr | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT;
324 if (vcpu->arch.exception.has_error_code) {
325 exit_int_info |= SVM_EVTINJ_VALID_ERR;
326 vmcb12->control.exit_int_info_err =
327 vcpu->arch.exception.error_code;
330 } else if (vcpu->arch.nmi_injected) {
331 exit_int_info = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
333 } else if (vcpu->arch.interrupt.injected) {
334 nr = vcpu->arch.interrupt.nr;
335 exit_int_info = nr | SVM_EVTINJ_VALID;
337 if (vcpu->arch.interrupt.soft)
338 exit_int_info |= SVM_EVTINJ_TYPE_SOFT;
340 exit_int_info |= SVM_EVTINJ_TYPE_INTR;
343 vmcb12->control.exit_int_info = exit_int_info;
346 static inline bool nested_npt_enabled(struct vcpu_svm *svm)
348 return svm->nested.ctl.nested_ctl & SVM_NESTED_CTL_NP_ENABLE;
352 * Load guest's/host's cr3 on nested vmentry or vmexit. @nested_npt is true
353 * if we are emulating VM-Entry into a guest with NPT enabled.
355 static int nested_svm_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3,
358 if (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63))
361 if (!nested_npt && is_pae_paging(vcpu) &&
362 (cr3 != kvm_read_cr3(vcpu) || pdptrs_changed(vcpu))) {
363 if (!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
368 * TODO: optimize unconditional TLB flush/MMU sync here and in
369 * kvm_init_shadow_npt_mmu().
372 kvm_mmu_new_pgd(vcpu, cr3, false, false);
374 vcpu->arch.cr3 = cr3;
375 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
377 kvm_init_mmu(vcpu, false);
382 static void nested_prepare_vmcb_save(struct vcpu_svm *svm, struct vmcb *vmcb12)
384 /* Load the nested guest state */
385 svm->vmcb->save.es = vmcb12->save.es;
386 svm->vmcb->save.cs = vmcb12->save.cs;
387 svm->vmcb->save.ss = vmcb12->save.ss;
388 svm->vmcb->save.ds = vmcb12->save.ds;
389 svm->vmcb->save.gdtr = vmcb12->save.gdtr;
390 svm->vmcb->save.idtr = vmcb12->save.idtr;
391 kvm_set_rflags(&svm->vcpu, vmcb12->save.rflags);
394 * Force-set EFER_SVME even though it is checked earlier on the
395 * VMCB12, because the guest can flip the bit between the check
396 * and now. Clearing EFER_SVME would call svm_free_nested.
398 svm_set_efer(&svm->vcpu, vmcb12->save.efer | EFER_SVME);
400 svm_set_cr0(&svm->vcpu, vmcb12->save.cr0);
401 svm_set_cr4(&svm->vcpu, vmcb12->save.cr4);
402 svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = vmcb12->save.cr2;
403 kvm_rax_write(&svm->vcpu, vmcb12->save.rax);
404 kvm_rsp_write(&svm->vcpu, vmcb12->save.rsp);
405 kvm_rip_write(&svm->vcpu, vmcb12->save.rip);
407 /* In case we don't even reach vcpu_run, the fields are not updated */
408 svm->vmcb->save.rax = vmcb12->save.rax;
409 svm->vmcb->save.rsp = vmcb12->save.rsp;
410 svm->vmcb->save.rip = vmcb12->save.rip;
411 svm->vmcb->save.dr7 = vmcb12->save.dr7;
412 svm->vcpu.arch.dr6 = vmcb12->save.dr6;
413 svm->vmcb->save.cpl = vmcb12->save.cpl;
416 static void nested_prepare_vmcb_control(struct vcpu_svm *svm)
418 const u32 mask = V_INTR_MASKING_MASK | V_GIF_ENABLE_MASK | V_GIF_MASK;
420 if (nested_npt_enabled(svm))
421 nested_svm_init_mmu_context(&svm->vcpu);
423 svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset =
424 svm->vcpu.arch.l1_tsc_offset + svm->nested.ctl.tsc_offset;
426 svm->vmcb->control.int_ctl =
427 (svm->nested.ctl.int_ctl & ~mask) |
428 (svm->nested.hsave->control.int_ctl & mask);
430 svm->vmcb->control.virt_ext = svm->nested.ctl.virt_ext;
431 svm->vmcb->control.int_vector = svm->nested.ctl.int_vector;
432 svm->vmcb->control.int_state = svm->nested.ctl.int_state;
433 svm->vmcb->control.event_inj = svm->nested.ctl.event_inj;
434 svm->vmcb->control.event_inj_err = svm->nested.ctl.event_inj_err;
436 svm->vmcb->control.pause_filter_count = svm->nested.ctl.pause_filter_count;
437 svm->vmcb->control.pause_filter_thresh = svm->nested.ctl.pause_filter_thresh;
439 /* Enter Guest-Mode */
440 enter_guest_mode(&svm->vcpu);
443 * Merge guest and host intercepts - must be called with vcpu in
444 * guest-mode to take affect here
446 recalc_intercepts(svm);
448 vmcb_mark_all_dirty(svm->vmcb);
451 int enter_svm_guest_mode(struct vcpu_svm *svm, u64 vmcb12_gpa,
456 svm->nested.vmcb12_gpa = vmcb12_gpa;
457 nested_prepare_vmcb_save(svm, vmcb12);
458 nested_prepare_vmcb_control(svm);
460 ret = nested_svm_load_cr3(&svm->vcpu, vmcb12->save.cr3,
461 nested_npt_enabled(svm));
465 svm_set_gif(svm, true);
470 int nested_svm_vmrun(struct vcpu_svm *svm)
474 struct vmcb *hsave = svm->nested.hsave;
475 struct vmcb *vmcb = svm->vmcb;
476 struct kvm_host_map map;
479 if (is_smm(&svm->vcpu)) {
480 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
484 vmcb12_gpa = svm->vmcb->save.rax;
485 ret = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(vmcb12_gpa), &map);
486 if (ret == -EINVAL) {
487 kvm_inject_gp(&svm->vcpu, 0);
490 return kvm_skip_emulated_instruction(&svm->vcpu);
493 ret = kvm_skip_emulated_instruction(&svm->vcpu);
497 if (WARN_ON_ONCE(!svm->nested.initialized))
500 load_nested_vmcb_control(svm, &vmcb12->control);
502 if (!nested_vmcb_check_save(svm, vmcb12) ||
503 !nested_vmcb_check_controls(&svm->nested.ctl)) {
504 vmcb12->control.exit_code = SVM_EXIT_ERR;
505 vmcb12->control.exit_code_hi = 0;
506 vmcb12->control.exit_info_1 = 0;
507 vmcb12->control.exit_info_2 = 0;
511 trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb12_gpa,
513 vmcb12->control.int_ctl,
514 vmcb12->control.event_inj,
515 vmcb12->control.nested_ctl);
517 trace_kvm_nested_intercepts(vmcb12->control.intercepts[INTERCEPT_CR] & 0xffff,
518 vmcb12->control.intercepts[INTERCEPT_CR] >> 16,
519 vmcb12->control.intercepts[INTERCEPT_EXCEPTION],
520 vmcb12->control.intercepts[INTERCEPT_WORD3],
521 vmcb12->control.intercepts[INTERCEPT_WORD4],
522 vmcb12->control.intercepts[INTERCEPT_WORD5]);
524 /* Clear internal status */
525 kvm_clear_exception_queue(&svm->vcpu);
526 kvm_clear_interrupt_queue(&svm->vcpu);
529 * Save the old vmcb, so we don't need to pick what we save, but can
530 * restore everything when a VMEXIT occurs
532 hsave->save.es = vmcb->save.es;
533 hsave->save.cs = vmcb->save.cs;
534 hsave->save.ss = vmcb->save.ss;
535 hsave->save.ds = vmcb->save.ds;
536 hsave->save.gdtr = vmcb->save.gdtr;
537 hsave->save.idtr = vmcb->save.idtr;
538 hsave->save.efer = svm->vcpu.arch.efer;
539 hsave->save.cr0 = kvm_read_cr0(&svm->vcpu);
540 hsave->save.cr4 = svm->vcpu.arch.cr4;
541 hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
542 hsave->save.rip = kvm_rip_read(&svm->vcpu);
543 hsave->save.rsp = vmcb->save.rsp;
544 hsave->save.rax = vmcb->save.rax;
546 hsave->save.cr3 = vmcb->save.cr3;
548 hsave->save.cr3 = kvm_read_cr3(&svm->vcpu);
550 copy_vmcb_control_area(&hsave->control, &vmcb->control);
552 svm->nested.nested_run_pending = 1;
554 if (enter_svm_guest_mode(svm, vmcb12_gpa, vmcb12))
557 if (nested_svm_vmrun_msrpm(svm))
561 svm->nested.nested_run_pending = 0;
563 svm->vmcb->control.exit_code = SVM_EXIT_ERR;
564 svm->vmcb->control.exit_code_hi = 0;
565 svm->vmcb->control.exit_info_1 = 0;
566 svm->vmcb->control.exit_info_2 = 0;
568 nested_svm_vmexit(svm);
571 kvm_vcpu_unmap(&svm->vcpu, &map, true);
576 void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
578 to_vmcb->save.fs = from_vmcb->save.fs;
579 to_vmcb->save.gs = from_vmcb->save.gs;
580 to_vmcb->save.tr = from_vmcb->save.tr;
581 to_vmcb->save.ldtr = from_vmcb->save.ldtr;
582 to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
583 to_vmcb->save.star = from_vmcb->save.star;
584 to_vmcb->save.lstar = from_vmcb->save.lstar;
585 to_vmcb->save.cstar = from_vmcb->save.cstar;
586 to_vmcb->save.sfmask = from_vmcb->save.sfmask;
587 to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
588 to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
589 to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
592 int nested_svm_vmexit(struct vcpu_svm *svm)
596 struct vmcb *hsave = svm->nested.hsave;
597 struct vmcb *vmcb = svm->vmcb;
598 struct kvm_host_map map;
600 rc = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(svm->nested.vmcb12_gpa), &map);
603 kvm_inject_gp(&svm->vcpu, 0);
609 /* Exit Guest-Mode */
610 leave_guest_mode(&svm->vcpu);
611 svm->nested.vmcb12_gpa = 0;
612 WARN_ON_ONCE(svm->nested.nested_run_pending);
614 /* in case we halted in L2 */
615 svm->vcpu.arch.mp_state = KVM_MP_STATE_RUNNABLE;
617 /* Give the current vmcb to the guest */
619 vmcb12->save.es = vmcb->save.es;
620 vmcb12->save.cs = vmcb->save.cs;
621 vmcb12->save.ss = vmcb->save.ss;
622 vmcb12->save.ds = vmcb->save.ds;
623 vmcb12->save.gdtr = vmcb->save.gdtr;
624 vmcb12->save.idtr = vmcb->save.idtr;
625 vmcb12->save.efer = svm->vcpu.arch.efer;
626 vmcb12->save.cr0 = kvm_read_cr0(&svm->vcpu);
627 vmcb12->save.cr3 = kvm_read_cr3(&svm->vcpu);
628 vmcb12->save.cr2 = vmcb->save.cr2;
629 vmcb12->save.cr4 = svm->vcpu.arch.cr4;
630 vmcb12->save.rflags = kvm_get_rflags(&svm->vcpu);
631 vmcb12->save.rip = kvm_rip_read(&svm->vcpu);
632 vmcb12->save.rsp = kvm_rsp_read(&svm->vcpu);
633 vmcb12->save.rax = kvm_rax_read(&svm->vcpu);
634 vmcb12->save.dr7 = vmcb->save.dr7;
635 vmcb12->save.dr6 = svm->vcpu.arch.dr6;
636 vmcb12->save.cpl = vmcb->save.cpl;
638 vmcb12->control.int_state = vmcb->control.int_state;
639 vmcb12->control.exit_code = vmcb->control.exit_code;
640 vmcb12->control.exit_code_hi = vmcb->control.exit_code_hi;
641 vmcb12->control.exit_info_1 = vmcb->control.exit_info_1;
642 vmcb12->control.exit_info_2 = vmcb->control.exit_info_2;
644 if (vmcb12->control.exit_code != SVM_EXIT_ERR)
645 nested_vmcb_save_pending_event(svm, vmcb12);
647 if (svm->nrips_enabled)
648 vmcb12->control.next_rip = vmcb->control.next_rip;
650 vmcb12->control.int_ctl = svm->nested.ctl.int_ctl;
651 vmcb12->control.tlb_ctl = svm->nested.ctl.tlb_ctl;
652 vmcb12->control.event_inj = svm->nested.ctl.event_inj;
653 vmcb12->control.event_inj_err = svm->nested.ctl.event_inj_err;
655 vmcb12->control.pause_filter_count =
656 svm->vmcb->control.pause_filter_count;
657 vmcb12->control.pause_filter_thresh =
658 svm->vmcb->control.pause_filter_thresh;
660 /* Restore the original control entries */
661 copy_vmcb_control_area(&vmcb->control, &hsave->control);
663 /* On vmexit the GIF is set to false */
664 svm_set_gif(svm, false);
666 svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset =
667 svm->vcpu.arch.l1_tsc_offset;
669 svm->nested.ctl.nested_cr3 = 0;
671 /* Restore selected save entries */
672 svm->vmcb->save.es = hsave->save.es;
673 svm->vmcb->save.cs = hsave->save.cs;
674 svm->vmcb->save.ss = hsave->save.ss;
675 svm->vmcb->save.ds = hsave->save.ds;
676 svm->vmcb->save.gdtr = hsave->save.gdtr;
677 svm->vmcb->save.idtr = hsave->save.idtr;
678 kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
679 svm_set_efer(&svm->vcpu, hsave->save.efer);
680 svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
681 svm_set_cr4(&svm->vcpu, hsave->save.cr4);
682 kvm_rax_write(&svm->vcpu, hsave->save.rax);
683 kvm_rsp_write(&svm->vcpu, hsave->save.rsp);
684 kvm_rip_write(&svm->vcpu, hsave->save.rip);
685 svm->vmcb->save.dr7 = 0;
686 svm->vmcb->save.cpl = 0;
687 svm->vmcb->control.exit_int_info = 0;
689 vmcb_mark_all_dirty(svm->vmcb);
691 trace_kvm_nested_vmexit_inject(vmcb12->control.exit_code,
692 vmcb12->control.exit_info_1,
693 vmcb12->control.exit_info_2,
694 vmcb12->control.exit_int_info,
695 vmcb12->control.exit_int_info_err,
698 kvm_vcpu_unmap(&svm->vcpu, &map, true);
700 nested_svm_uninit_mmu_context(&svm->vcpu);
702 rc = nested_svm_load_cr3(&svm->vcpu, hsave->save.cr3, false);
707 svm->vmcb->save.cr3 = hsave->save.cr3;
710 * Drop what we picked up for L2 via svm_complete_interrupts() so it
711 * doesn't end up in L1.
713 svm->vcpu.arch.nmi_injected = false;
714 kvm_clear_exception_queue(&svm->vcpu);
715 kvm_clear_interrupt_queue(&svm->vcpu);
720 int svm_allocate_nested(struct vcpu_svm *svm)
722 struct page *hsave_page;
724 if (svm->nested.initialized)
727 hsave_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
730 svm->nested.hsave = page_address(hsave_page);
732 svm->nested.msrpm = svm_vcpu_alloc_msrpm();
733 if (!svm->nested.msrpm)
735 svm_vcpu_init_msrpm(&svm->vcpu, svm->nested.msrpm);
737 svm->nested.initialized = true;
741 __free_page(hsave_page);
745 void svm_free_nested(struct vcpu_svm *svm)
747 if (!svm->nested.initialized)
750 svm_vcpu_free_msrpm(svm->nested.msrpm);
751 svm->nested.msrpm = NULL;
753 __free_page(virt_to_page(svm->nested.hsave));
754 svm->nested.hsave = NULL;
756 svm->nested.initialized = false;
760 * Forcibly leave nested mode in order to be able to reset the VCPU later on.
762 void svm_leave_nested(struct vcpu_svm *svm)
764 if (is_guest_mode(&svm->vcpu)) {
765 struct vmcb *hsave = svm->nested.hsave;
766 struct vmcb *vmcb = svm->vmcb;
768 svm->nested.nested_run_pending = 0;
769 leave_guest_mode(&svm->vcpu);
770 copy_vmcb_control_area(&vmcb->control, &hsave->control);
771 nested_svm_uninit_mmu_context(&svm->vcpu);
774 kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, &svm->vcpu);
777 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
779 u32 offset, msr, value;
782 if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
783 return NESTED_EXIT_HOST;
785 msr = svm->vcpu.arch.regs[VCPU_REGS_RCX];
786 offset = svm_msrpm_offset(msr);
787 write = svm->vmcb->control.exit_info_1 & 1;
788 mask = 1 << ((2 * (msr & 0xf)) + write);
790 if (offset == MSR_INVALID)
791 return NESTED_EXIT_DONE;
793 /* Offset is in 32 bit units but need in 8 bit units */
796 if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.ctl.msrpm_base_pa + offset, &value, 4))
797 return NESTED_EXIT_DONE;
799 return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
802 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
804 unsigned port, size, iopm_len;
809 if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_IOIO_PROT)))
810 return NESTED_EXIT_HOST;
812 port = svm->vmcb->control.exit_info_1 >> 16;
813 size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
815 gpa = svm->nested.ctl.iopm_base_pa + (port / 8);
816 start_bit = port % 8;
817 iopm_len = (start_bit + size > 8) ? 2 : 1;
818 mask = (0xf >> (4 - size)) << start_bit;
821 if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
822 return NESTED_EXIT_DONE;
824 return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
827 static int nested_svm_intercept(struct vcpu_svm *svm)
829 u32 exit_code = svm->vmcb->control.exit_code;
830 int vmexit = NESTED_EXIT_HOST;
834 vmexit = nested_svm_exit_handled_msr(svm);
837 vmexit = nested_svm_intercept_ioio(svm);
839 case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
840 if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
841 vmexit = NESTED_EXIT_DONE;
844 case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
845 if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
846 vmexit = NESTED_EXIT_DONE;
849 case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
851 * Host-intercepted exceptions have been checked already in
852 * nested_svm_exit_special. There is nothing to do here,
853 * the vmexit is injected by svm_check_nested_events.
855 vmexit = NESTED_EXIT_DONE;
859 vmexit = NESTED_EXIT_DONE;
863 if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
864 vmexit = NESTED_EXIT_DONE;
871 int nested_svm_exit_handled(struct vcpu_svm *svm)
875 vmexit = nested_svm_intercept(svm);
877 if (vmexit == NESTED_EXIT_DONE)
878 nested_svm_vmexit(svm);
883 int nested_svm_check_permissions(struct vcpu_svm *svm)
885 if (!(svm->vcpu.arch.efer & EFER_SVME) ||
886 !is_paging(&svm->vcpu)) {
887 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
891 if (svm->vmcb->save.cpl) {
892 kvm_inject_gp(&svm->vcpu, 0);
899 static bool nested_exit_on_exception(struct vcpu_svm *svm)
901 unsigned int nr = svm->vcpu.arch.exception.nr;
903 return (svm->nested.ctl.intercepts[INTERCEPT_EXCEPTION] & BIT(nr));
906 static void nested_svm_inject_exception_vmexit(struct vcpu_svm *svm)
908 unsigned int nr = svm->vcpu.arch.exception.nr;
910 svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
911 svm->vmcb->control.exit_code_hi = 0;
913 if (svm->vcpu.arch.exception.has_error_code)
914 svm->vmcb->control.exit_info_1 = svm->vcpu.arch.exception.error_code;
917 * EXITINFO2 is undefined for all exception intercepts other
920 if (nr == PF_VECTOR) {
921 if (svm->vcpu.arch.exception.nested_apf)
922 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.apf.nested_apf_token;
923 else if (svm->vcpu.arch.exception.has_payload)
924 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.exception.payload;
926 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
927 } else if (nr == DB_VECTOR) {
928 /* See inject_pending_event. */
929 kvm_deliver_exception_payload(&svm->vcpu);
930 if (svm->vcpu.arch.dr7 & DR7_GD) {
931 svm->vcpu.arch.dr7 &= ~DR7_GD;
932 kvm_update_dr7(&svm->vcpu);
935 WARN_ON(svm->vcpu.arch.exception.has_payload);
937 nested_svm_vmexit(svm);
940 static void nested_svm_smi(struct vcpu_svm *svm)
942 svm->vmcb->control.exit_code = SVM_EXIT_SMI;
943 svm->vmcb->control.exit_info_1 = 0;
944 svm->vmcb->control.exit_info_2 = 0;
946 nested_svm_vmexit(svm);
949 static void nested_svm_nmi(struct vcpu_svm *svm)
951 svm->vmcb->control.exit_code = SVM_EXIT_NMI;
952 svm->vmcb->control.exit_info_1 = 0;
953 svm->vmcb->control.exit_info_2 = 0;
955 nested_svm_vmexit(svm);
958 static void nested_svm_intr(struct vcpu_svm *svm)
960 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
962 svm->vmcb->control.exit_code = SVM_EXIT_INTR;
963 svm->vmcb->control.exit_info_1 = 0;
964 svm->vmcb->control.exit_info_2 = 0;
966 nested_svm_vmexit(svm);
969 static inline bool nested_exit_on_init(struct vcpu_svm *svm)
971 return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_INIT);
974 static void nested_svm_init(struct vcpu_svm *svm)
976 svm->vmcb->control.exit_code = SVM_EXIT_INIT;
977 svm->vmcb->control.exit_info_1 = 0;
978 svm->vmcb->control.exit_info_2 = 0;
980 nested_svm_vmexit(svm);
984 static int svm_check_nested_events(struct kvm_vcpu *vcpu)
986 struct vcpu_svm *svm = to_svm(vcpu);
987 bool block_nested_events =
988 kvm_event_needs_reinjection(vcpu) || svm->nested.nested_run_pending;
989 struct kvm_lapic *apic = vcpu->arch.apic;
991 if (lapic_in_kernel(vcpu) &&
992 test_bit(KVM_APIC_INIT, &apic->pending_events)) {
993 if (block_nested_events)
995 if (!nested_exit_on_init(svm))
997 nested_svm_init(svm);
1001 if (vcpu->arch.exception.pending) {
1002 if (block_nested_events)
1004 if (!nested_exit_on_exception(svm))
1006 nested_svm_inject_exception_vmexit(svm);
1010 if (vcpu->arch.smi_pending && !svm_smi_blocked(vcpu)) {
1011 if (block_nested_events)
1013 if (!nested_exit_on_smi(svm))
1015 nested_svm_smi(svm);
1019 if (vcpu->arch.nmi_pending && !svm_nmi_blocked(vcpu)) {
1020 if (block_nested_events)
1022 if (!nested_exit_on_nmi(svm))
1024 nested_svm_nmi(svm);
1028 if (kvm_cpu_has_interrupt(vcpu) && !svm_interrupt_blocked(vcpu)) {
1029 if (block_nested_events)
1031 if (!nested_exit_on_intr(svm))
1033 nested_svm_intr(svm);
1040 int nested_svm_exit_special(struct vcpu_svm *svm)
1042 u32 exit_code = svm->vmcb->control.exit_code;
1044 switch (exit_code) {
1048 return NESTED_EXIT_HOST;
1049 case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1050 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
1052 if (get_host_vmcb(svm)->control.intercepts[INTERCEPT_EXCEPTION] &
1054 return NESTED_EXIT_HOST;
1055 else if (exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR &&
1056 svm->vcpu.arch.apf.host_apf_flags)
1057 /* Trap async PF even if not shadowing */
1058 return NESTED_EXIT_HOST;
1065 return NESTED_EXIT_CONTINUE;
1068 static int svm_get_nested_state(struct kvm_vcpu *vcpu,
1069 struct kvm_nested_state __user *user_kvm_nested_state,
1072 struct vcpu_svm *svm;
1073 struct kvm_nested_state kvm_state = {
1075 .format = KVM_STATE_NESTED_FORMAT_SVM,
1076 .size = sizeof(kvm_state),
1078 struct vmcb __user *user_vmcb = (struct vmcb __user *)
1079 &user_kvm_nested_state->data.svm[0];
1082 return kvm_state.size + KVM_STATE_NESTED_SVM_VMCB_SIZE;
1086 if (user_data_size < kvm_state.size)
1089 /* First fill in the header and copy it out. */
1090 if (is_guest_mode(vcpu)) {
1091 kvm_state.hdr.svm.vmcb_pa = svm->nested.vmcb12_gpa;
1092 kvm_state.size += KVM_STATE_NESTED_SVM_VMCB_SIZE;
1093 kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
1095 if (svm->nested.nested_run_pending)
1096 kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
1100 kvm_state.flags |= KVM_STATE_NESTED_GIF_SET;
1102 if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
1105 if (!is_guest_mode(vcpu))
1109 * Copy over the full size of the VMCB rather than just the size
1112 if (clear_user(user_vmcb, KVM_STATE_NESTED_SVM_VMCB_SIZE))
1114 if (copy_to_user(&user_vmcb->control, &svm->nested.ctl,
1115 sizeof(user_vmcb->control)))
1117 if (copy_to_user(&user_vmcb->save, &svm->nested.hsave->save,
1118 sizeof(user_vmcb->save)))
1122 return kvm_state.size;
1125 static int svm_set_nested_state(struct kvm_vcpu *vcpu,
1126 struct kvm_nested_state __user *user_kvm_nested_state,
1127 struct kvm_nested_state *kvm_state)
1129 struct vcpu_svm *svm = to_svm(vcpu);
1130 struct vmcb *hsave = svm->nested.hsave;
1131 struct vmcb __user *user_vmcb = (struct vmcb __user *)
1132 &user_kvm_nested_state->data.svm[0];
1133 struct vmcb_control_area *ctl;
1134 struct vmcb_save_area *save;
1138 BUILD_BUG_ON(sizeof(struct vmcb_control_area) + sizeof(struct vmcb_save_area) >
1139 KVM_STATE_NESTED_SVM_VMCB_SIZE);
1141 if (kvm_state->format != KVM_STATE_NESTED_FORMAT_SVM)
1144 if (kvm_state->flags & ~(KVM_STATE_NESTED_GUEST_MODE |
1145 KVM_STATE_NESTED_RUN_PENDING |
1146 KVM_STATE_NESTED_GIF_SET))
1150 * If in guest mode, vcpu->arch.efer actually refers to the L2 guest's
1151 * EFER.SVME, but EFER.SVME still has to be 1 for VMRUN to succeed.
1153 if (!(vcpu->arch.efer & EFER_SVME)) {
1154 /* GIF=1 and no guest mode are required if SVME=0. */
1155 if (kvm_state->flags != KVM_STATE_NESTED_GIF_SET)
1159 /* SMM temporarily disables SVM, so we cannot be in guest mode. */
1160 if (is_smm(vcpu) && (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
1163 if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE)) {
1164 svm_leave_nested(svm);
1165 svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
1169 if (!page_address_valid(vcpu, kvm_state->hdr.svm.vmcb_pa))
1171 if (kvm_state->size < sizeof(*kvm_state) + KVM_STATE_NESTED_SVM_VMCB_SIZE)
1175 ctl = kzalloc(sizeof(*ctl), GFP_KERNEL);
1176 save = kzalloc(sizeof(*save), GFP_KERNEL);
1181 if (copy_from_user(ctl, &user_vmcb->control, sizeof(*ctl)))
1183 if (copy_from_user(save, &user_vmcb->save, sizeof(*save)))
1187 if (!nested_vmcb_check_controls(ctl))
1191 * Processor state contains L2 state. Check that it is
1192 * valid for guest mode (see nested_vmcb_checks).
1194 cr0 = kvm_read_cr0(vcpu);
1195 if (((cr0 & X86_CR0_CD) == 0) && (cr0 & X86_CR0_NW))
1199 * Validate host state saved from before VMRUN (see
1200 * nested_svm_check_permissions).
1201 * TODO: validate reserved bits for all saved state.
1203 if (!(save->cr0 & X86_CR0_PG))
1205 if (!(save->efer & EFER_SVME))
1209 * All checks done, we can enter guest mode. L1 control fields
1210 * come from the nested save state. Guest state is already
1211 * in the registers, the save area of the nested state instead
1212 * contains saved L1 state.
1214 copy_vmcb_control_area(&hsave->control, &svm->vmcb->control);
1215 hsave->save = *save;
1217 svm->nested.vmcb12_gpa = kvm_state->hdr.svm.vmcb_pa;
1218 load_nested_vmcb_control(svm, ctl);
1219 nested_prepare_vmcb_control(svm);
1221 kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
1230 struct kvm_x86_nested_ops svm_nested_ops = {
1231 .check_events = svm_check_nested_events,
1232 .get_nested_state_pages = svm_get_nested_state_pages,
1233 .get_state = svm_get_nested_state,
1234 .set_state = svm_set_nested_state,