xref: /openbmc/linux/arch/x86/kvm/svm/nested.c (revision 9659281c)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Kernel-based Virtual Machine driver for Linux
4  *
5  * AMD SVM support
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
9  *
10  * Authors:
11  *   Yaniv Kamay  <yaniv@qumranet.com>
12  *   Avi Kivity   <avi@qumranet.com>
13  */
14 
15 #define pr_fmt(fmt) "SVM: " fmt
16 
17 #include <linux/kvm_types.h>
18 #include <linux/kvm_host.h>
19 #include <linux/kernel.h>
20 
21 #include <asm/msr-index.h>
22 #include <asm/debugreg.h>
23 
24 #include "kvm_emulate.h"
25 #include "trace.h"
26 #include "mmu.h"
27 #include "x86.h"
28 #include "cpuid.h"
29 #include "lapic.h"
30 #include "svm.h"
31 
32 #define CC KVM_NESTED_VMENTER_CONSISTENCY_CHECK
33 
34 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
35 				       struct x86_exception *fault)
36 {
37 	struct vcpu_svm *svm = to_svm(vcpu);
38 
39 	if (svm->vmcb->control.exit_code != SVM_EXIT_NPF) {
40 		/*
41 		 * TODO: track the cause of the nested page fault, and
42 		 * correctly fill in the high bits of exit_info_1.
43 		 */
44 		svm->vmcb->control.exit_code = SVM_EXIT_NPF;
45 		svm->vmcb->control.exit_code_hi = 0;
46 		svm->vmcb->control.exit_info_1 = (1ULL << 32);
47 		svm->vmcb->control.exit_info_2 = fault->address;
48 	}
49 
50 	svm->vmcb->control.exit_info_1 &= ~0xffffffffULL;
51 	svm->vmcb->control.exit_info_1 |= fault->error_code;
52 
53 	nested_svm_vmexit(svm);
54 }
55 
56 static void svm_inject_page_fault_nested(struct kvm_vcpu *vcpu, struct x86_exception *fault)
57 {
58        struct vcpu_svm *svm = to_svm(vcpu);
59        WARN_ON(!is_guest_mode(vcpu));
60 
61        if (vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_EXCEPTION_OFFSET + PF_VECTOR) &&
62 	   !svm->nested.nested_run_pending) {
63                svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + PF_VECTOR;
64                svm->vmcb->control.exit_code_hi = 0;
65                svm->vmcb->control.exit_info_1 = fault->error_code;
66                svm->vmcb->control.exit_info_2 = fault->address;
67                nested_svm_vmexit(svm);
68        } else {
69                kvm_inject_page_fault(vcpu, fault);
70        }
71 }
72 
73 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
74 {
75 	struct vcpu_svm *svm = to_svm(vcpu);
76 	u64 cr3 = svm->nested.ctl.nested_cr3;
77 	u64 pdpte;
78 	int ret;
79 
80 	ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(cr3), &pdpte,
81 				       offset_in_page(cr3) + index * 8, 8);
82 	if (ret)
83 		return 0;
84 	return pdpte;
85 }
86 
87 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
88 {
89 	struct vcpu_svm *svm = to_svm(vcpu);
90 
91 	return svm->nested.ctl.nested_cr3;
92 }
93 
94 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
95 {
96 	struct vcpu_svm *svm = to_svm(vcpu);
97 
98 	WARN_ON(mmu_is_nested(vcpu));
99 
100 	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
101 
102 	/*
103 	 * The NPT format depends on L1's CR4 and EFER, which is in vmcb01.  Note,
104 	 * when called via KVM_SET_NESTED_STATE, that state may _not_ match current
105 	 * vCPU state.  CR0.WP is explicitly ignored, while CR0.PG is required.
106 	 */
107 	kvm_init_shadow_npt_mmu(vcpu, X86_CR0_PG, svm->vmcb01.ptr->save.cr4,
108 				svm->vmcb01.ptr->save.efer,
109 				svm->nested.ctl.nested_cr3);
110 	vcpu->arch.mmu->get_guest_pgd     = nested_svm_get_tdp_cr3;
111 	vcpu->arch.mmu->get_pdptr         = nested_svm_get_tdp_pdptr;
112 	vcpu->arch.mmu->inject_page_fault = nested_svm_inject_npf_exit;
113 	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
114 }
115 
116 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
117 {
118 	vcpu->arch.mmu = &vcpu->arch.root_mmu;
119 	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
120 }
121 
122 void recalc_intercepts(struct vcpu_svm *svm)
123 {
124 	struct vmcb_control_area *c, *h, *g;
125 	unsigned int i;
126 
127 	vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
128 
129 	if (!is_guest_mode(&svm->vcpu))
130 		return;
131 
132 	c = &svm->vmcb->control;
133 	h = &svm->vmcb01.ptr->control;
134 	g = &svm->nested.ctl;
135 
136 	for (i = 0; i < MAX_INTERCEPT; i++)
137 		c->intercepts[i] = h->intercepts[i];
138 
139 	if (g->int_ctl & V_INTR_MASKING_MASK) {
140 		/* We only want the cr8 intercept bits of L1 */
141 		vmcb_clr_intercept(c, INTERCEPT_CR8_READ);
142 		vmcb_clr_intercept(c, INTERCEPT_CR8_WRITE);
143 
144 		/*
145 		 * Once running L2 with HF_VINTR_MASK, EFLAGS.IF does not
146 		 * affect any interrupt we may want to inject; therefore,
147 		 * interrupt window vmexits are irrelevant to L0.
148 		 */
149 		vmcb_clr_intercept(c, INTERCEPT_VINTR);
150 	}
151 
152 	/* We don't want to see VMMCALLs from a nested guest */
153 	vmcb_clr_intercept(c, INTERCEPT_VMMCALL);
154 
155 	for (i = 0; i < MAX_INTERCEPT; i++)
156 		c->intercepts[i] |= g->intercepts[i];
157 
158 	/* If SMI is not intercepted, ignore guest SMI intercept as well  */
159 	if (!intercept_smi)
160 		vmcb_clr_intercept(c, INTERCEPT_SMI);
161 }
162 
163 static void copy_vmcb_control_area(struct vmcb_control_area *dst,
164 				   struct vmcb_control_area *from)
165 {
166 	unsigned int i;
167 
168 	for (i = 0; i < MAX_INTERCEPT; i++)
169 		dst->intercepts[i] = from->intercepts[i];
170 
171 	dst->iopm_base_pa         = from->iopm_base_pa;
172 	dst->msrpm_base_pa        = from->msrpm_base_pa;
173 	dst->tsc_offset           = from->tsc_offset;
174 	/* asid not copied, it is handled manually for svm->vmcb.  */
175 	dst->tlb_ctl              = from->tlb_ctl;
176 	dst->int_ctl              = from->int_ctl;
177 	dst->int_vector           = from->int_vector;
178 	dst->int_state            = from->int_state;
179 	dst->exit_code            = from->exit_code;
180 	dst->exit_code_hi         = from->exit_code_hi;
181 	dst->exit_info_1          = from->exit_info_1;
182 	dst->exit_info_2          = from->exit_info_2;
183 	dst->exit_int_info        = from->exit_int_info;
184 	dst->exit_int_info_err    = from->exit_int_info_err;
185 	dst->nested_ctl           = from->nested_ctl;
186 	dst->event_inj            = from->event_inj;
187 	dst->event_inj_err        = from->event_inj_err;
188 	dst->nested_cr3           = from->nested_cr3;
189 	dst->virt_ext              = from->virt_ext;
190 	dst->pause_filter_count   = from->pause_filter_count;
191 	dst->pause_filter_thresh  = from->pause_filter_thresh;
192 }
193 
194 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
195 {
196 	/*
197 	 * This function merges the msr permission bitmaps of kvm and the
198 	 * nested vmcb. It is optimized in that it only merges the parts where
199 	 * the kvm msr permission bitmap may contain zero bits
200 	 */
201 	int i;
202 
203 	if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
204 		return true;
205 
206 	for (i = 0; i < MSRPM_OFFSETS; i++) {
207 		u32 value, p;
208 		u64 offset;
209 
210 		if (msrpm_offsets[i] == 0xffffffff)
211 			break;
212 
213 		p      = msrpm_offsets[i];
214 		offset = svm->nested.ctl.msrpm_base_pa + (p * 4);
215 
216 		if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
217 			return false;
218 
219 		svm->nested.msrpm[p] = svm->msrpm[p] | value;
220 	}
221 
222 	svm->vmcb->control.msrpm_base_pa = __sme_set(__pa(svm->nested.msrpm));
223 
224 	return true;
225 }
226 
227 /*
228  * Bits 11:0 of bitmap address are ignored by hardware
229  */
230 static bool nested_svm_check_bitmap_pa(struct kvm_vcpu *vcpu, u64 pa, u32 size)
231 {
232 	u64 addr = PAGE_ALIGN(pa);
233 
234 	return kvm_vcpu_is_legal_gpa(vcpu, addr) &&
235 	    kvm_vcpu_is_legal_gpa(vcpu, addr + size - 1);
236 }
237 
238 static bool nested_vmcb_check_controls(struct kvm_vcpu *vcpu,
239 				       struct vmcb_control_area *control)
240 {
241 	if (CC(!vmcb_is_intercept(control, INTERCEPT_VMRUN)))
242 		return false;
243 
244 	if (CC(control->asid == 0))
245 		return false;
246 
247 	if (CC((control->nested_ctl & SVM_NESTED_CTL_NP_ENABLE) && !npt_enabled))
248 		return false;
249 
250 	if (CC(!nested_svm_check_bitmap_pa(vcpu, control->msrpm_base_pa,
251 					   MSRPM_SIZE)))
252 		return false;
253 	if (CC(!nested_svm_check_bitmap_pa(vcpu, control->iopm_base_pa,
254 					   IOPM_SIZE)))
255 		return false;
256 
257 	return true;
258 }
259 
260 static bool nested_vmcb_check_cr3_cr4(struct kvm_vcpu *vcpu,
261 				      struct vmcb_save_area *save)
262 {
263 	/*
264 	 * These checks are also performed by KVM_SET_SREGS,
265 	 * except that EFER.LMA is not checked by SVM against
266 	 * CR0.PG && EFER.LME.
267 	 */
268 	if ((save->efer & EFER_LME) && (save->cr0 & X86_CR0_PG)) {
269 		if (CC(!(save->cr4 & X86_CR4_PAE)) ||
270 		    CC(!(save->cr0 & X86_CR0_PE)) ||
271 		    CC(kvm_vcpu_is_illegal_gpa(vcpu, save->cr3)))
272 			return false;
273 	}
274 
275 	if (CC(!kvm_is_valid_cr4(vcpu, save->cr4)))
276 		return false;
277 
278 	return true;
279 }
280 
281 /* Common checks that apply to both L1 and L2 state.  */
282 static bool nested_vmcb_valid_sregs(struct kvm_vcpu *vcpu,
283 				    struct vmcb_save_area *save)
284 {
285 	/*
286 	 * FIXME: these should be done after copying the fields,
287 	 * to avoid TOC/TOU races.  For these save area checks
288 	 * the possible damage is limited since kvm_set_cr0 and
289 	 * kvm_set_cr4 handle failure; EFER_SVME is an exception
290 	 * so it is force-set later in nested_prepare_vmcb_save.
291 	 */
292 	if (CC(!(save->efer & EFER_SVME)))
293 		return false;
294 
295 	if (CC((save->cr0 & X86_CR0_CD) == 0 && (save->cr0 & X86_CR0_NW)) ||
296 	    CC(save->cr0 & ~0xffffffffULL))
297 		return false;
298 
299 	if (CC(!kvm_dr6_valid(save->dr6)) || CC(!kvm_dr7_valid(save->dr7)))
300 		return false;
301 
302 	if (!nested_vmcb_check_cr3_cr4(vcpu, save))
303 		return false;
304 
305 	if (CC(!kvm_valid_efer(vcpu, save->efer)))
306 		return false;
307 
308 	return true;
309 }
310 
311 void nested_load_control_from_vmcb12(struct vcpu_svm *svm,
312 				     struct vmcb_control_area *control)
313 {
314 	copy_vmcb_control_area(&svm->nested.ctl, control);
315 
316 	/* Copy it here because nested_svm_check_controls will check it.  */
317 	svm->nested.ctl.asid           = control->asid;
318 	svm->nested.ctl.msrpm_base_pa &= ~0x0fffULL;
319 	svm->nested.ctl.iopm_base_pa  &= ~0x0fffULL;
320 }
321 
322 /*
323  * Synchronize fields that are written by the processor, so that
324  * they can be copied back into the vmcb12.
325  */
326 void nested_sync_control_from_vmcb02(struct vcpu_svm *svm)
327 {
328 	u32 mask;
329 	svm->nested.ctl.event_inj      = svm->vmcb->control.event_inj;
330 	svm->nested.ctl.event_inj_err  = svm->vmcb->control.event_inj_err;
331 
332 	/* Only a few fields of int_ctl are written by the processor.  */
333 	mask = V_IRQ_MASK | V_TPR_MASK;
334 	if (!(svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK) &&
335 	    svm_is_intercept(svm, INTERCEPT_VINTR)) {
336 		/*
337 		 * In order to request an interrupt window, L0 is usurping
338 		 * svm->vmcb->control.int_ctl and possibly setting V_IRQ
339 		 * even if it was clear in L1's VMCB.  Restoring it would be
340 		 * wrong.  However, in this case V_IRQ will remain true until
341 		 * interrupt_window_interception calls svm_clear_vintr and
342 		 * restores int_ctl.  We can just leave it aside.
343 		 */
344 		mask &= ~V_IRQ_MASK;
345 	}
346 	svm->nested.ctl.int_ctl        &= ~mask;
347 	svm->nested.ctl.int_ctl        |= svm->vmcb->control.int_ctl & mask;
348 }
349 
350 /*
351  * Transfer any event that L0 or L1 wanted to inject into L2 to
352  * EXIT_INT_INFO.
353  */
354 static void nested_save_pending_event_to_vmcb12(struct vcpu_svm *svm,
355 						struct vmcb *vmcb12)
356 {
357 	struct kvm_vcpu *vcpu = &svm->vcpu;
358 	u32 exit_int_info = 0;
359 	unsigned int nr;
360 
361 	if (vcpu->arch.exception.injected) {
362 		nr = vcpu->arch.exception.nr;
363 		exit_int_info = nr | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT;
364 
365 		if (vcpu->arch.exception.has_error_code) {
366 			exit_int_info |= SVM_EVTINJ_VALID_ERR;
367 			vmcb12->control.exit_int_info_err =
368 				vcpu->arch.exception.error_code;
369 		}
370 
371 	} else if (vcpu->arch.nmi_injected) {
372 		exit_int_info = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
373 
374 	} else if (vcpu->arch.interrupt.injected) {
375 		nr = vcpu->arch.interrupt.nr;
376 		exit_int_info = nr | SVM_EVTINJ_VALID;
377 
378 		if (vcpu->arch.interrupt.soft)
379 			exit_int_info |= SVM_EVTINJ_TYPE_SOFT;
380 		else
381 			exit_int_info |= SVM_EVTINJ_TYPE_INTR;
382 	}
383 
384 	vmcb12->control.exit_int_info = exit_int_info;
385 }
386 
387 static inline bool nested_npt_enabled(struct vcpu_svm *svm)
388 {
389 	return svm->nested.ctl.nested_ctl & SVM_NESTED_CTL_NP_ENABLE;
390 }
391 
392 static void nested_svm_transition_tlb_flush(struct kvm_vcpu *vcpu)
393 {
394 	/*
395 	 * TODO: optimize unconditional TLB flush/MMU sync.  A partial list of
396 	 * things to fix before this can be conditional:
397 	 *
398 	 *  - Flush TLBs for both L1 and L2 remote TLB flush
399 	 *  - Honor L1's request to flush an ASID on nested VMRUN
400 	 *  - Sync nested NPT MMU on VMRUN that flushes L2's ASID[*]
401 	 *  - Don't crush a pending TLB flush in vmcb02 on nested VMRUN
402 	 *  - Flush L1's ASID on KVM_REQ_TLB_FLUSH_GUEST
403 	 *
404 	 * [*] Unlike nested EPT, SVM's ASID management can invalidate nested
405 	 *     NPT guest-physical mappings on VMRUN.
406 	 */
407 	kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
408 	kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
409 }
410 
411 /*
412  * Load guest's/host's cr3 on nested vmentry or vmexit. @nested_npt is true
413  * if we are emulating VM-Entry into a guest with NPT enabled.
414  */
415 static int nested_svm_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3,
416 			       bool nested_npt, bool reload_pdptrs)
417 {
418 	if (CC(kvm_vcpu_is_illegal_gpa(vcpu, cr3)))
419 		return -EINVAL;
420 
421 	if (reload_pdptrs && !nested_npt && is_pae_paging(vcpu) &&
422 	    CC(!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)))
423 		return -EINVAL;
424 
425 	if (!nested_npt)
426 		kvm_mmu_new_pgd(vcpu, cr3);
427 
428 	vcpu->arch.cr3 = cr3;
429 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
430 
431 	/* Re-initialize the MMU, e.g. to pick up CR4 MMU role changes. */
432 	kvm_init_mmu(vcpu);
433 
434 	return 0;
435 }
436 
437 void nested_vmcb02_compute_g_pat(struct vcpu_svm *svm)
438 {
439 	if (!svm->nested.vmcb02.ptr)
440 		return;
441 
442 	/* FIXME: merge g_pat from vmcb01 and vmcb12.  */
443 	svm->nested.vmcb02.ptr->save.g_pat = svm->vmcb01.ptr->save.g_pat;
444 }
445 
446 static void nested_vmcb02_prepare_save(struct vcpu_svm *svm, struct vmcb *vmcb12)
447 {
448 	bool new_vmcb12 = false;
449 
450 	nested_vmcb02_compute_g_pat(svm);
451 
452 	/* Load the nested guest state */
453 	if (svm->nested.vmcb12_gpa != svm->nested.last_vmcb12_gpa) {
454 		new_vmcb12 = true;
455 		svm->nested.last_vmcb12_gpa = svm->nested.vmcb12_gpa;
456 	}
457 
458 	if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_SEG))) {
459 		svm->vmcb->save.es = vmcb12->save.es;
460 		svm->vmcb->save.cs = vmcb12->save.cs;
461 		svm->vmcb->save.ss = vmcb12->save.ss;
462 		svm->vmcb->save.ds = vmcb12->save.ds;
463 		svm->vmcb->save.cpl = vmcb12->save.cpl;
464 		vmcb_mark_dirty(svm->vmcb, VMCB_SEG);
465 	}
466 
467 	if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_DT))) {
468 		svm->vmcb->save.gdtr = vmcb12->save.gdtr;
469 		svm->vmcb->save.idtr = vmcb12->save.idtr;
470 		vmcb_mark_dirty(svm->vmcb, VMCB_DT);
471 	}
472 
473 	kvm_set_rflags(&svm->vcpu, vmcb12->save.rflags | X86_EFLAGS_FIXED);
474 
475 	/*
476 	 * Force-set EFER_SVME even though it is checked earlier on the
477 	 * VMCB12, because the guest can flip the bit between the check
478 	 * and now.  Clearing EFER_SVME would call svm_free_nested.
479 	 */
480 	svm_set_efer(&svm->vcpu, vmcb12->save.efer | EFER_SVME);
481 
482 	svm_set_cr0(&svm->vcpu, vmcb12->save.cr0);
483 	svm_set_cr4(&svm->vcpu, vmcb12->save.cr4);
484 
485 	svm->vcpu.arch.cr2 = vmcb12->save.cr2;
486 
487 	kvm_rax_write(&svm->vcpu, vmcb12->save.rax);
488 	kvm_rsp_write(&svm->vcpu, vmcb12->save.rsp);
489 	kvm_rip_write(&svm->vcpu, vmcb12->save.rip);
490 
491 	/* In case we don't even reach vcpu_run, the fields are not updated */
492 	svm->vmcb->save.rax = vmcb12->save.rax;
493 	svm->vmcb->save.rsp = vmcb12->save.rsp;
494 	svm->vmcb->save.rip = vmcb12->save.rip;
495 
496 	/* These bits will be set properly on the first execution when new_vmc12 is true */
497 	if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_DR))) {
498 		svm->vmcb->save.dr7 = vmcb12->save.dr7 | DR7_FIXED_1;
499 		svm->vcpu.arch.dr6  = vmcb12->save.dr6 | DR6_ACTIVE_LOW;
500 		vmcb_mark_dirty(svm->vmcb, VMCB_DR);
501 	}
502 }
503 
504 static void nested_vmcb02_prepare_control(struct vcpu_svm *svm)
505 {
506 	const u32 mask = V_INTR_MASKING_MASK | V_GIF_ENABLE_MASK | V_GIF_MASK;
507 	struct kvm_vcpu *vcpu = &svm->vcpu;
508 
509 	/*
510 	 * Filled at exit: exit_code, exit_code_hi, exit_info_1, exit_info_2,
511 	 * exit_int_info, exit_int_info_err, next_rip, insn_len, insn_bytes.
512 	 */
513 
514 	/*
515 	 * Also covers avic_vapic_bar, avic_backing_page, avic_logical_id,
516 	 * avic_physical_id.
517 	 */
518 	WARN_ON(kvm_apicv_activated(svm->vcpu.kvm));
519 
520 	/* Copied from vmcb01.  msrpm_base can be overwritten later.  */
521 	svm->vmcb->control.nested_ctl = svm->vmcb01.ptr->control.nested_ctl;
522 	svm->vmcb->control.iopm_base_pa = svm->vmcb01.ptr->control.iopm_base_pa;
523 	svm->vmcb->control.msrpm_base_pa = svm->vmcb01.ptr->control.msrpm_base_pa;
524 
525 	/* Done at vmrun: asid.  */
526 
527 	/* Also overwritten later if necessary.  */
528 	svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
529 
530 	/* nested_cr3.  */
531 	if (nested_npt_enabled(svm))
532 		nested_svm_init_mmu_context(vcpu);
533 
534 	svm->vmcb->control.tsc_offset = vcpu->arch.tsc_offset =
535 		vcpu->arch.l1_tsc_offset + svm->nested.ctl.tsc_offset;
536 
537 	svm->vmcb->control.int_ctl             =
538 		(svm->nested.ctl.int_ctl & ~mask) |
539 		(svm->vmcb01.ptr->control.int_ctl & mask);
540 
541 	svm->vmcb->control.virt_ext            = svm->nested.ctl.virt_ext;
542 	svm->vmcb->control.int_vector          = svm->nested.ctl.int_vector;
543 	svm->vmcb->control.int_state           = svm->nested.ctl.int_state;
544 	svm->vmcb->control.event_inj           = svm->nested.ctl.event_inj;
545 	svm->vmcb->control.event_inj_err       = svm->nested.ctl.event_inj_err;
546 
547 	svm->vmcb->control.pause_filter_count  = svm->nested.ctl.pause_filter_count;
548 	svm->vmcb->control.pause_filter_thresh = svm->nested.ctl.pause_filter_thresh;
549 
550 	nested_svm_transition_tlb_flush(vcpu);
551 
552 	/* Enter Guest-Mode */
553 	enter_guest_mode(vcpu);
554 
555 	/*
556 	 * Merge guest and host intercepts - must be called with vcpu in
557 	 * guest-mode to take effect.
558 	 */
559 	recalc_intercepts(svm);
560 }
561 
562 static void nested_svm_copy_common_state(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
563 {
564 	/*
565 	 * Some VMCB state is shared between L1 and L2 and thus has to be
566 	 * moved at the time of nested vmrun and vmexit.
567 	 *
568 	 * VMLOAD/VMSAVE state would also belong in this category, but KVM
569 	 * always performs VMLOAD and VMSAVE from the VMCB01.
570 	 */
571 	to_vmcb->save.spec_ctrl = from_vmcb->save.spec_ctrl;
572 }
573 
574 int enter_svm_guest_mode(struct kvm_vcpu *vcpu, u64 vmcb12_gpa,
575 			 struct vmcb *vmcb12)
576 {
577 	struct vcpu_svm *svm = to_svm(vcpu);
578 	int ret;
579 
580 	trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb12_gpa,
581 			       vmcb12->save.rip,
582 			       vmcb12->control.int_ctl,
583 			       vmcb12->control.event_inj,
584 			       vmcb12->control.nested_ctl);
585 
586 	trace_kvm_nested_intercepts(vmcb12->control.intercepts[INTERCEPT_CR] & 0xffff,
587 				    vmcb12->control.intercepts[INTERCEPT_CR] >> 16,
588 				    vmcb12->control.intercepts[INTERCEPT_EXCEPTION],
589 				    vmcb12->control.intercepts[INTERCEPT_WORD3],
590 				    vmcb12->control.intercepts[INTERCEPT_WORD4],
591 				    vmcb12->control.intercepts[INTERCEPT_WORD5]);
592 
593 
594 	svm->nested.vmcb12_gpa = vmcb12_gpa;
595 
596 	WARN_ON(svm->vmcb == svm->nested.vmcb02.ptr);
597 
598 	nested_svm_copy_common_state(svm->vmcb01.ptr, svm->nested.vmcb02.ptr);
599 
600 	svm_switch_vmcb(svm, &svm->nested.vmcb02);
601 	nested_vmcb02_prepare_control(svm);
602 	nested_vmcb02_prepare_save(svm, vmcb12);
603 
604 	ret = nested_svm_load_cr3(&svm->vcpu, vmcb12->save.cr3,
605 				  nested_npt_enabled(svm), true);
606 	if (ret)
607 		return ret;
608 
609 	if (!npt_enabled)
610 		vcpu->arch.mmu->inject_page_fault = svm_inject_page_fault_nested;
611 
612 	svm_set_gif(svm, true);
613 
614 	return 0;
615 }
616 
617 int nested_svm_vmrun(struct kvm_vcpu *vcpu)
618 {
619 	struct vcpu_svm *svm = to_svm(vcpu);
620 	int ret;
621 	struct vmcb *vmcb12;
622 	struct kvm_host_map map;
623 	u64 vmcb12_gpa;
624 
625 	if (!svm->nested.hsave_msr) {
626 		kvm_inject_gp(vcpu, 0);
627 		return 1;
628 	}
629 
630 	if (is_smm(vcpu)) {
631 		kvm_queue_exception(vcpu, UD_VECTOR);
632 		return 1;
633 	}
634 
635 	vmcb12_gpa = svm->vmcb->save.rax;
636 	ret = kvm_vcpu_map(vcpu, gpa_to_gfn(vmcb12_gpa), &map);
637 	if (ret == -EINVAL) {
638 		kvm_inject_gp(vcpu, 0);
639 		return 1;
640 	} else if (ret) {
641 		return kvm_skip_emulated_instruction(vcpu);
642 	}
643 
644 	ret = kvm_skip_emulated_instruction(vcpu);
645 
646 	vmcb12 = map.hva;
647 
648 	if (WARN_ON_ONCE(!svm->nested.initialized))
649 		return -EINVAL;
650 
651 	nested_load_control_from_vmcb12(svm, &vmcb12->control);
652 
653 	if (!nested_vmcb_valid_sregs(vcpu, &vmcb12->save) ||
654 	    !nested_vmcb_check_controls(vcpu, &svm->nested.ctl)) {
655 		vmcb12->control.exit_code    = SVM_EXIT_ERR;
656 		vmcb12->control.exit_code_hi = 0;
657 		vmcb12->control.exit_info_1  = 0;
658 		vmcb12->control.exit_info_2  = 0;
659 		goto out;
660 	}
661 
662 
663 	/* Clear internal status */
664 	kvm_clear_exception_queue(vcpu);
665 	kvm_clear_interrupt_queue(vcpu);
666 
667 	/*
668 	 * Since vmcb01 is not in use, we can use it to store some of the L1
669 	 * state.
670 	 */
671 	svm->vmcb01.ptr->save.efer   = vcpu->arch.efer;
672 	svm->vmcb01.ptr->save.cr0    = kvm_read_cr0(vcpu);
673 	svm->vmcb01.ptr->save.cr4    = vcpu->arch.cr4;
674 	svm->vmcb01.ptr->save.rflags = kvm_get_rflags(vcpu);
675 	svm->vmcb01.ptr->save.rip    = kvm_rip_read(vcpu);
676 
677 	if (!npt_enabled)
678 		svm->vmcb01.ptr->save.cr3 = kvm_read_cr3(vcpu);
679 
680 	svm->nested.nested_run_pending = 1;
681 
682 	if (enter_svm_guest_mode(vcpu, vmcb12_gpa, vmcb12))
683 		goto out_exit_err;
684 
685 	if (nested_svm_vmrun_msrpm(svm))
686 		goto out;
687 
688 out_exit_err:
689 	svm->nested.nested_run_pending = 0;
690 
691 	svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
692 	svm->vmcb->control.exit_code_hi = 0;
693 	svm->vmcb->control.exit_info_1  = 0;
694 	svm->vmcb->control.exit_info_2  = 0;
695 
696 	nested_svm_vmexit(svm);
697 
698 out:
699 	kvm_vcpu_unmap(vcpu, &map, true);
700 
701 	return ret;
702 }
703 
704 /* Copy state save area fields which are handled by VMRUN */
705 void svm_copy_vmrun_state(struct vmcb_save_area *to_save,
706 			  struct vmcb_save_area *from_save)
707 {
708 	to_save->es = from_save->es;
709 	to_save->cs = from_save->cs;
710 	to_save->ss = from_save->ss;
711 	to_save->ds = from_save->ds;
712 	to_save->gdtr = from_save->gdtr;
713 	to_save->idtr = from_save->idtr;
714 	to_save->rflags = from_save->rflags | X86_EFLAGS_FIXED;
715 	to_save->efer = from_save->efer;
716 	to_save->cr0 = from_save->cr0;
717 	to_save->cr3 = from_save->cr3;
718 	to_save->cr4 = from_save->cr4;
719 	to_save->rax = from_save->rax;
720 	to_save->rsp = from_save->rsp;
721 	to_save->rip = from_save->rip;
722 	to_save->cpl = 0;
723 }
724 
725 void svm_copy_vmloadsave_state(struct vmcb *to_vmcb, struct vmcb *from_vmcb)
726 {
727 	to_vmcb->save.fs = from_vmcb->save.fs;
728 	to_vmcb->save.gs = from_vmcb->save.gs;
729 	to_vmcb->save.tr = from_vmcb->save.tr;
730 	to_vmcb->save.ldtr = from_vmcb->save.ldtr;
731 	to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
732 	to_vmcb->save.star = from_vmcb->save.star;
733 	to_vmcb->save.lstar = from_vmcb->save.lstar;
734 	to_vmcb->save.cstar = from_vmcb->save.cstar;
735 	to_vmcb->save.sfmask = from_vmcb->save.sfmask;
736 	to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
737 	to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
738 	to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
739 }
740 
741 int nested_svm_vmexit(struct vcpu_svm *svm)
742 {
743 	struct kvm_vcpu *vcpu = &svm->vcpu;
744 	struct vmcb *vmcb12;
745 	struct vmcb *vmcb = svm->vmcb;
746 	struct kvm_host_map map;
747 	int rc;
748 
749 	/* Triple faults in L2 should never escape. */
750 	WARN_ON_ONCE(kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu));
751 
752 	rc = kvm_vcpu_map(vcpu, gpa_to_gfn(svm->nested.vmcb12_gpa), &map);
753 	if (rc) {
754 		if (rc == -EINVAL)
755 			kvm_inject_gp(vcpu, 0);
756 		return 1;
757 	}
758 
759 	vmcb12 = map.hva;
760 
761 	/* Exit Guest-Mode */
762 	leave_guest_mode(vcpu);
763 	svm->nested.vmcb12_gpa = 0;
764 	WARN_ON_ONCE(svm->nested.nested_run_pending);
765 
766 	kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
767 
768 	/* in case we halted in L2 */
769 	svm->vcpu.arch.mp_state = KVM_MP_STATE_RUNNABLE;
770 
771 	/* Give the current vmcb to the guest */
772 
773 	vmcb12->save.es     = vmcb->save.es;
774 	vmcb12->save.cs     = vmcb->save.cs;
775 	vmcb12->save.ss     = vmcb->save.ss;
776 	vmcb12->save.ds     = vmcb->save.ds;
777 	vmcb12->save.gdtr   = vmcb->save.gdtr;
778 	vmcb12->save.idtr   = vmcb->save.idtr;
779 	vmcb12->save.efer   = svm->vcpu.arch.efer;
780 	vmcb12->save.cr0    = kvm_read_cr0(vcpu);
781 	vmcb12->save.cr3    = kvm_read_cr3(vcpu);
782 	vmcb12->save.cr2    = vmcb->save.cr2;
783 	vmcb12->save.cr4    = svm->vcpu.arch.cr4;
784 	vmcb12->save.rflags = kvm_get_rflags(vcpu);
785 	vmcb12->save.rip    = kvm_rip_read(vcpu);
786 	vmcb12->save.rsp    = kvm_rsp_read(vcpu);
787 	vmcb12->save.rax    = kvm_rax_read(vcpu);
788 	vmcb12->save.dr7    = vmcb->save.dr7;
789 	vmcb12->save.dr6    = svm->vcpu.arch.dr6;
790 	vmcb12->save.cpl    = vmcb->save.cpl;
791 
792 	vmcb12->control.int_state         = vmcb->control.int_state;
793 	vmcb12->control.exit_code         = vmcb->control.exit_code;
794 	vmcb12->control.exit_code_hi      = vmcb->control.exit_code_hi;
795 	vmcb12->control.exit_info_1       = vmcb->control.exit_info_1;
796 	vmcb12->control.exit_info_2       = vmcb->control.exit_info_2;
797 
798 	if (vmcb12->control.exit_code != SVM_EXIT_ERR)
799 		nested_save_pending_event_to_vmcb12(svm, vmcb12);
800 
801 	if (svm->nrips_enabled)
802 		vmcb12->control.next_rip  = vmcb->control.next_rip;
803 
804 	vmcb12->control.int_ctl           = svm->nested.ctl.int_ctl;
805 	vmcb12->control.tlb_ctl           = svm->nested.ctl.tlb_ctl;
806 	vmcb12->control.event_inj         = svm->nested.ctl.event_inj;
807 	vmcb12->control.event_inj_err     = svm->nested.ctl.event_inj_err;
808 
809 	vmcb12->control.pause_filter_count =
810 		svm->vmcb->control.pause_filter_count;
811 	vmcb12->control.pause_filter_thresh =
812 		svm->vmcb->control.pause_filter_thresh;
813 
814 	nested_svm_copy_common_state(svm->nested.vmcb02.ptr, svm->vmcb01.ptr);
815 
816 	svm_switch_vmcb(svm, &svm->vmcb01);
817 
818 	/*
819 	 * On vmexit the  GIF is set to false and
820 	 * no event can be injected in L1.
821 	 */
822 	svm_set_gif(svm, false);
823 	svm->vmcb->control.exit_int_info = 0;
824 
825 	svm->vcpu.arch.tsc_offset = svm->vcpu.arch.l1_tsc_offset;
826 	if (svm->vmcb->control.tsc_offset != svm->vcpu.arch.tsc_offset) {
827 		svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset;
828 		vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
829 	}
830 
831 	svm->nested.ctl.nested_cr3 = 0;
832 
833 	/*
834 	 * Restore processor state that had been saved in vmcb01
835 	 */
836 	kvm_set_rflags(vcpu, svm->vmcb->save.rflags);
837 	svm_set_efer(vcpu, svm->vmcb->save.efer);
838 	svm_set_cr0(vcpu, svm->vmcb->save.cr0 | X86_CR0_PE);
839 	svm_set_cr4(vcpu, svm->vmcb->save.cr4);
840 	kvm_rax_write(vcpu, svm->vmcb->save.rax);
841 	kvm_rsp_write(vcpu, svm->vmcb->save.rsp);
842 	kvm_rip_write(vcpu, svm->vmcb->save.rip);
843 
844 	svm->vcpu.arch.dr7 = DR7_FIXED_1;
845 	kvm_update_dr7(&svm->vcpu);
846 
847 	trace_kvm_nested_vmexit_inject(vmcb12->control.exit_code,
848 				       vmcb12->control.exit_info_1,
849 				       vmcb12->control.exit_info_2,
850 				       vmcb12->control.exit_int_info,
851 				       vmcb12->control.exit_int_info_err,
852 				       KVM_ISA_SVM);
853 
854 	kvm_vcpu_unmap(vcpu, &map, true);
855 
856 	nested_svm_transition_tlb_flush(vcpu);
857 
858 	nested_svm_uninit_mmu_context(vcpu);
859 
860 	rc = nested_svm_load_cr3(vcpu, svm->vmcb->save.cr3, false, true);
861 	if (rc)
862 		return 1;
863 
864 	/*
865 	 * Drop what we picked up for L2 via svm_complete_interrupts() so it
866 	 * doesn't end up in L1.
867 	 */
868 	svm->vcpu.arch.nmi_injected = false;
869 	kvm_clear_exception_queue(vcpu);
870 	kvm_clear_interrupt_queue(vcpu);
871 
872 	/*
873 	 * If we are here following the completion of a VMRUN that
874 	 * is being single-stepped, queue the pending #DB intercept
875 	 * right now so that it an be accounted for before we execute
876 	 * L1's next instruction.
877 	 */
878 	if (unlikely(svm->vmcb->save.rflags & X86_EFLAGS_TF))
879 		kvm_queue_exception(&(svm->vcpu), DB_VECTOR);
880 
881 	return 0;
882 }
883 
884 static void nested_svm_triple_fault(struct kvm_vcpu *vcpu)
885 {
886 	nested_svm_simple_vmexit(to_svm(vcpu), SVM_EXIT_SHUTDOWN);
887 }
888 
889 int svm_allocate_nested(struct vcpu_svm *svm)
890 {
891 	struct page *vmcb02_page;
892 
893 	if (svm->nested.initialized)
894 		return 0;
895 
896 	vmcb02_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
897 	if (!vmcb02_page)
898 		return -ENOMEM;
899 	svm->nested.vmcb02.ptr = page_address(vmcb02_page);
900 	svm->nested.vmcb02.pa = __sme_set(page_to_pfn(vmcb02_page) << PAGE_SHIFT);
901 
902 	svm->nested.msrpm = svm_vcpu_alloc_msrpm();
903 	if (!svm->nested.msrpm)
904 		goto err_free_vmcb02;
905 	svm_vcpu_init_msrpm(&svm->vcpu, svm->nested.msrpm);
906 
907 	svm->nested.initialized = true;
908 	return 0;
909 
910 err_free_vmcb02:
911 	__free_page(vmcb02_page);
912 	return -ENOMEM;
913 }
914 
915 void svm_free_nested(struct vcpu_svm *svm)
916 {
917 	if (!svm->nested.initialized)
918 		return;
919 
920 	svm_vcpu_free_msrpm(svm->nested.msrpm);
921 	svm->nested.msrpm = NULL;
922 
923 	__free_page(virt_to_page(svm->nested.vmcb02.ptr));
924 	svm->nested.vmcb02.ptr = NULL;
925 
926 	/*
927 	 * When last_vmcb12_gpa matches the current vmcb12 gpa,
928 	 * some vmcb12 fields are not loaded if they are marked clean
929 	 * in the vmcb12, since in this case they are up to date already.
930 	 *
931 	 * When the vmcb02 is freed, this optimization becomes invalid.
932 	 */
933 	svm->nested.last_vmcb12_gpa = INVALID_GPA;
934 
935 	svm->nested.initialized = false;
936 }
937 
938 /*
939  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
940  */
941 void svm_leave_nested(struct vcpu_svm *svm)
942 {
943 	struct kvm_vcpu *vcpu = &svm->vcpu;
944 
945 	if (is_guest_mode(vcpu)) {
946 		svm->nested.nested_run_pending = 0;
947 		svm->nested.vmcb12_gpa = INVALID_GPA;
948 
949 		leave_guest_mode(vcpu);
950 
951 		svm_switch_vmcb(svm, &svm->vmcb01);
952 
953 		nested_svm_uninit_mmu_context(vcpu);
954 		vmcb_mark_all_dirty(svm->vmcb);
955 	}
956 
957 	kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
958 }
959 
960 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
961 {
962 	u32 offset, msr, value;
963 	int write, mask;
964 
965 	if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
966 		return NESTED_EXIT_HOST;
967 
968 	msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
969 	offset = svm_msrpm_offset(msr);
970 	write  = svm->vmcb->control.exit_info_1 & 1;
971 	mask   = 1 << ((2 * (msr & 0xf)) + write);
972 
973 	if (offset == MSR_INVALID)
974 		return NESTED_EXIT_DONE;
975 
976 	/* Offset is in 32 bit units but need in 8 bit units */
977 	offset *= 4;
978 
979 	if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.ctl.msrpm_base_pa + offset, &value, 4))
980 		return NESTED_EXIT_DONE;
981 
982 	return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
983 }
984 
985 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
986 {
987 	unsigned port, size, iopm_len;
988 	u16 val, mask;
989 	u8 start_bit;
990 	u64 gpa;
991 
992 	if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_IOIO_PROT)))
993 		return NESTED_EXIT_HOST;
994 
995 	port = svm->vmcb->control.exit_info_1 >> 16;
996 	size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
997 		SVM_IOIO_SIZE_SHIFT;
998 	gpa  = svm->nested.ctl.iopm_base_pa + (port / 8);
999 	start_bit = port % 8;
1000 	iopm_len = (start_bit + size > 8) ? 2 : 1;
1001 	mask = (0xf >> (4 - size)) << start_bit;
1002 	val = 0;
1003 
1004 	if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
1005 		return NESTED_EXIT_DONE;
1006 
1007 	return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
1008 }
1009 
1010 static int nested_svm_intercept(struct vcpu_svm *svm)
1011 {
1012 	u32 exit_code = svm->vmcb->control.exit_code;
1013 	int vmexit = NESTED_EXIT_HOST;
1014 
1015 	switch (exit_code) {
1016 	case SVM_EXIT_MSR:
1017 		vmexit = nested_svm_exit_handled_msr(svm);
1018 		break;
1019 	case SVM_EXIT_IOIO:
1020 		vmexit = nested_svm_intercept_ioio(svm);
1021 		break;
1022 	case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
1023 		if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
1024 			vmexit = NESTED_EXIT_DONE;
1025 		break;
1026 	}
1027 	case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
1028 		if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
1029 			vmexit = NESTED_EXIT_DONE;
1030 		break;
1031 	}
1032 	case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1033 		/*
1034 		 * Host-intercepted exceptions have been checked already in
1035 		 * nested_svm_exit_special.  There is nothing to do here,
1036 		 * the vmexit is injected by svm_check_nested_events.
1037 		 */
1038 		vmexit = NESTED_EXIT_DONE;
1039 		break;
1040 	}
1041 	case SVM_EXIT_ERR: {
1042 		vmexit = NESTED_EXIT_DONE;
1043 		break;
1044 	}
1045 	default: {
1046 		if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
1047 			vmexit = NESTED_EXIT_DONE;
1048 	}
1049 	}
1050 
1051 	return vmexit;
1052 }
1053 
1054 int nested_svm_exit_handled(struct vcpu_svm *svm)
1055 {
1056 	int vmexit;
1057 
1058 	vmexit = nested_svm_intercept(svm);
1059 
1060 	if (vmexit == NESTED_EXIT_DONE)
1061 		nested_svm_vmexit(svm);
1062 
1063 	return vmexit;
1064 }
1065 
1066 int nested_svm_check_permissions(struct kvm_vcpu *vcpu)
1067 {
1068 	if (!(vcpu->arch.efer & EFER_SVME) || !is_paging(vcpu)) {
1069 		kvm_queue_exception(vcpu, UD_VECTOR);
1070 		return 1;
1071 	}
1072 
1073 	if (to_svm(vcpu)->vmcb->save.cpl) {
1074 		kvm_inject_gp(vcpu, 0);
1075 		return 1;
1076 	}
1077 
1078 	return 0;
1079 }
1080 
1081 static bool nested_exit_on_exception(struct vcpu_svm *svm)
1082 {
1083 	unsigned int nr = svm->vcpu.arch.exception.nr;
1084 
1085 	return (svm->nested.ctl.intercepts[INTERCEPT_EXCEPTION] & BIT(nr));
1086 }
1087 
1088 static void nested_svm_inject_exception_vmexit(struct vcpu_svm *svm)
1089 {
1090 	unsigned int nr = svm->vcpu.arch.exception.nr;
1091 
1092 	svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
1093 	svm->vmcb->control.exit_code_hi = 0;
1094 
1095 	if (svm->vcpu.arch.exception.has_error_code)
1096 		svm->vmcb->control.exit_info_1 = svm->vcpu.arch.exception.error_code;
1097 
1098 	/*
1099 	 * EXITINFO2 is undefined for all exception intercepts other
1100 	 * than #PF.
1101 	 */
1102 	if (nr == PF_VECTOR) {
1103 		if (svm->vcpu.arch.exception.nested_apf)
1104 			svm->vmcb->control.exit_info_2 = svm->vcpu.arch.apf.nested_apf_token;
1105 		else if (svm->vcpu.arch.exception.has_payload)
1106 			svm->vmcb->control.exit_info_2 = svm->vcpu.arch.exception.payload;
1107 		else
1108 			svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
1109 	} else if (nr == DB_VECTOR) {
1110 		/* See inject_pending_event.  */
1111 		kvm_deliver_exception_payload(&svm->vcpu);
1112 		if (svm->vcpu.arch.dr7 & DR7_GD) {
1113 			svm->vcpu.arch.dr7 &= ~DR7_GD;
1114 			kvm_update_dr7(&svm->vcpu);
1115 		}
1116 	} else
1117 		WARN_ON(svm->vcpu.arch.exception.has_payload);
1118 
1119 	nested_svm_vmexit(svm);
1120 }
1121 
1122 static inline bool nested_exit_on_init(struct vcpu_svm *svm)
1123 {
1124 	return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_INIT);
1125 }
1126 
1127 static int svm_check_nested_events(struct kvm_vcpu *vcpu)
1128 {
1129 	struct vcpu_svm *svm = to_svm(vcpu);
1130 	bool block_nested_events =
1131 		kvm_event_needs_reinjection(vcpu) || svm->nested.nested_run_pending;
1132 	struct kvm_lapic *apic = vcpu->arch.apic;
1133 
1134 	if (lapic_in_kernel(vcpu) &&
1135 	    test_bit(KVM_APIC_INIT, &apic->pending_events)) {
1136 		if (block_nested_events)
1137 			return -EBUSY;
1138 		if (!nested_exit_on_init(svm))
1139 			return 0;
1140 		nested_svm_simple_vmexit(svm, SVM_EXIT_INIT);
1141 		return 0;
1142 	}
1143 
1144 	if (vcpu->arch.exception.pending) {
1145 		/*
1146 		 * Only a pending nested run can block a pending exception.
1147 		 * Otherwise an injected NMI/interrupt should either be
1148 		 * lost or delivered to the nested hypervisor in the EXITINTINFO
1149 		 * vmcb field, while delivering the pending exception.
1150 		 */
1151 		if (svm->nested.nested_run_pending)
1152                         return -EBUSY;
1153 		if (!nested_exit_on_exception(svm))
1154 			return 0;
1155 		nested_svm_inject_exception_vmexit(svm);
1156 		return 0;
1157 	}
1158 
1159 	if (vcpu->arch.smi_pending && !svm_smi_blocked(vcpu)) {
1160 		if (block_nested_events)
1161 			return -EBUSY;
1162 		if (!nested_exit_on_smi(svm))
1163 			return 0;
1164 		nested_svm_simple_vmexit(svm, SVM_EXIT_SMI);
1165 		return 0;
1166 	}
1167 
1168 	if (vcpu->arch.nmi_pending && !svm_nmi_blocked(vcpu)) {
1169 		if (block_nested_events)
1170 			return -EBUSY;
1171 		if (!nested_exit_on_nmi(svm))
1172 			return 0;
1173 		nested_svm_simple_vmexit(svm, SVM_EXIT_NMI);
1174 		return 0;
1175 	}
1176 
1177 	if (kvm_cpu_has_interrupt(vcpu) && !svm_interrupt_blocked(vcpu)) {
1178 		if (block_nested_events)
1179 			return -EBUSY;
1180 		if (!nested_exit_on_intr(svm))
1181 			return 0;
1182 		trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
1183 		nested_svm_simple_vmexit(svm, SVM_EXIT_INTR);
1184 		return 0;
1185 	}
1186 
1187 	return 0;
1188 }
1189 
1190 int nested_svm_exit_special(struct vcpu_svm *svm)
1191 {
1192 	u32 exit_code = svm->vmcb->control.exit_code;
1193 
1194 	switch (exit_code) {
1195 	case SVM_EXIT_INTR:
1196 	case SVM_EXIT_NMI:
1197 	case SVM_EXIT_NPF:
1198 		return NESTED_EXIT_HOST;
1199 	case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1200 		u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
1201 
1202 		if (svm->vmcb01.ptr->control.intercepts[INTERCEPT_EXCEPTION] &
1203 		    excp_bits)
1204 			return NESTED_EXIT_HOST;
1205 		else if (exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR &&
1206 			 svm->vcpu.arch.apf.host_apf_flags)
1207 			/* Trap async PF even if not shadowing */
1208 			return NESTED_EXIT_HOST;
1209 		break;
1210 	}
1211 	default:
1212 		break;
1213 	}
1214 
1215 	return NESTED_EXIT_CONTINUE;
1216 }
1217 
1218 static int svm_get_nested_state(struct kvm_vcpu *vcpu,
1219 				struct kvm_nested_state __user *user_kvm_nested_state,
1220 				u32 user_data_size)
1221 {
1222 	struct vcpu_svm *svm;
1223 	struct kvm_nested_state kvm_state = {
1224 		.flags = 0,
1225 		.format = KVM_STATE_NESTED_FORMAT_SVM,
1226 		.size = sizeof(kvm_state),
1227 	};
1228 	struct vmcb __user *user_vmcb = (struct vmcb __user *)
1229 		&user_kvm_nested_state->data.svm[0];
1230 
1231 	if (!vcpu)
1232 		return kvm_state.size + KVM_STATE_NESTED_SVM_VMCB_SIZE;
1233 
1234 	svm = to_svm(vcpu);
1235 
1236 	if (user_data_size < kvm_state.size)
1237 		goto out;
1238 
1239 	/* First fill in the header and copy it out.  */
1240 	if (is_guest_mode(vcpu)) {
1241 		kvm_state.hdr.svm.vmcb_pa = svm->nested.vmcb12_gpa;
1242 		kvm_state.size += KVM_STATE_NESTED_SVM_VMCB_SIZE;
1243 		kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
1244 
1245 		if (svm->nested.nested_run_pending)
1246 			kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
1247 	}
1248 
1249 	if (gif_set(svm))
1250 		kvm_state.flags |= KVM_STATE_NESTED_GIF_SET;
1251 
1252 	if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
1253 		return -EFAULT;
1254 
1255 	if (!is_guest_mode(vcpu))
1256 		goto out;
1257 
1258 	/*
1259 	 * Copy over the full size of the VMCB rather than just the size
1260 	 * of the structs.
1261 	 */
1262 	if (clear_user(user_vmcb, KVM_STATE_NESTED_SVM_VMCB_SIZE))
1263 		return -EFAULT;
1264 	if (copy_to_user(&user_vmcb->control, &svm->nested.ctl,
1265 			 sizeof(user_vmcb->control)))
1266 		return -EFAULT;
1267 	if (copy_to_user(&user_vmcb->save, &svm->vmcb01.ptr->save,
1268 			 sizeof(user_vmcb->save)))
1269 		return -EFAULT;
1270 out:
1271 	return kvm_state.size;
1272 }
1273 
1274 static int svm_set_nested_state(struct kvm_vcpu *vcpu,
1275 				struct kvm_nested_state __user *user_kvm_nested_state,
1276 				struct kvm_nested_state *kvm_state)
1277 {
1278 	struct vcpu_svm *svm = to_svm(vcpu);
1279 	struct vmcb __user *user_vmcb = (struct vmcb __user *)
1280 		&user_kvm_nested_state->data.svm[0];
1281 	struct vmcb_control_area *ctl;
1282 	struct vmcb_save_area *save;
1283 	unsigned long cr0;
1284 	int ret;
1285 
1286 	BUILD_BUG_ON(sizeof(struct vmcb_control_area) + sizeof(struct vmcb_save_area) >
1287 		     KVM_STATE_NESTED_SVM_VMCB_SIZE);
1288 
1289 	if (kvm_state->format != KVM_STATE_NESTED_FORMAT_SVM)
1290 		return -EINVAL;
1291 
1292 	if (kvm_state->flags & ~(KVM_STATE_NESTED_GUEST_MODE |
1293 				 KVM_STATE_NESTED_RUN_PENDING |
1294 				 KVM_STATE_NESTED_GIF_SET))
1295 		return -EINVAL;
1296 
1297 	/*
1298 	 * If in guest mode, vcpu->arch.efer actually refers to the L2 guest's
1299 	 * EFER.SVME, but EFER.SVME still has to be 1 for VMRUN to succeed.
1300 	 */
1301 	if (!(vcpu->arch.efer & EFER_SVME)) {
1302 		/* GIF=1 and no guest mode are required if SVME=0.  */
1303 		if (kvm_state->flags != KVM_STATE_NESTED_GIF_SET)
1304 			return -EINVAL;
1305 	}
1306 
1307 	/* SMM temporarily disables SVM, so we cannot be in guest mode.  */
1308 	if (is_smm(vcpu) && (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
1309 		return -EINVAL;
1310 
1311 	if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE)) {
1312 		svm_leave_nested(svm);
1313 		svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
1314 		return 0;
1315 	}
1316 
1317 	if (!page_address_valid(vcpu, kvm_state->hdr.svm.vmcb_pa))
1318 		return -EINVAL;
1319 	if (kvm_state->size < sizeof(*kvm_state) + KVM_STATE_NESTED_SVM_VMCB_SIZE)
1320 		return -EINVAL;
1321 
1322 	ret  = -ENOMEM;
1323 	ctl  = kzalloc(sizeof(*ctl),  GFP_KERNEL_ACCOUNT);
1324 	save = kzalloc(sizeof(*save), GFP_KERNEL_ACCOUNT);
1325 	if (!ctl || !save)
1326 		goto out_free;
1327 
1328 	ret = -EFAULT;
1329 	if (copy_from_user(ctl, &user_vmcb->control, sizeof(*ctl)))
1330 		goto out_free;
1331 	if (copy_from_user(save, &user_vmcb->save, sizeof(*save)))
1332 		goto out_free;
1333 
1334 	ret = -EINVAL;
1335 	if (!nested_vmcb_check_controls(vcpu, ctl))
1336 		goto out_free;
1337 
1338 	/*
1339 	 * Processor state contains L2 state.  Check that it is
1340 	 * valid for guest mode (see nested_vmcb_check_save).
1341 	 */
1342 	cr0 = kvm_read_cr0(vcpu);
1343         if (((cr0 & X86_CR0_CD) == 0) && (cr0 & X86_CR0_NW))
1344 		goto out_free;
1345 
1346 	/*
1347 	 * Validate host state saved from before VMRUN (see
1348 	 * nested_svm_check_permissions).
1349 	 */
1350 	if (!(save->cr0 & X86_CR0_PG) ||
1351 	    !(save->cr0 & X86_CR0_PE) ||
1352 	    (save->rflags & X86_EFLAGS_VM) ||
1353 	    !nested_vmcb_valid_sregs(vcpu, save))
1354 		goto out_free;
1355 
1356 	/*
1357 	 * While the nested guest CR3 is already checked and set by
1358 	 * KVM_SET_SREGS, it was set when nested state was yet loaded,
1359 	 * thus MMU might not be initialized correctly.
1360 	 * Set it again to fix this.
1361 	 */
1362 
1363 	ret = nested_svm_load_cr3(&svm->vcpu, vcpu->arch.cr3,
1364 				  nested_npt_enabled(svm), false);
1365 	if (WARN_ON_ONCE(ret))
1366 		goto out_free;
1367 
1368 
1369 	/*
1370 	 * All checks done, we can enter guest mode. Userspace provides
1371 	 * vmcb12.control, which will be combined with L1 and stored into
1372 	 * vmcb02, and the L1 save state which we store in vmcb01.
1373 	 * L2 registers if needed are moved from the current VMCB to VMCB02.
1374 	 */
1375 
1376 	if (is_guest_mode(vcpu))
1377 		svm_leave_nested(svm);
1378 	else
1379 		svm->nested.vmcb02.ptr->save = svm->vmcb01.ptr->save;
1380 
1381 	svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
1382 
1383 	svm->nested.nested_run_pending =
1384 		!!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
1385 
1386 	svm->nested.vmcb12_gpa = kvm_state->hdr.svm.vmcb_pa;
1387 
1388 	svm_copy_vmrun_state(&svm->vmcb01.ptr->save, save);
1389 	nested_load_control_from_vmcb12(svm, ctl);
1390 
1391 	svm_switch_vmcb(svm, &svm->nested.vmcb02);
1392 	nested_vmcb02_prepare_control(svm);
1393 	kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
1394 	ret = 0;
1395 out_free:
1396 	kfree(save);
1397 	kfree(ctl);
1398 
1399 	return ret;
1400 }
1401 
1402 static bool svm_get_nested_state_pages(struct kvm_vcpu *vcpu)
1403 {
1404 	struct vcpu_svm *svm = to_svm(vcpu);
1405 
1406 	if (WARN_ON(!is_guest_mode(vcpu)))
1407 		return true;
1408 
1409 	if (!vcpu->arch.pdptrs_from_userspace &&
1410 	    !nested_npt_enabled(svm) && is_pae_paging(vcpu))
1411 		/*
1412 		 * Reload the guest's PDPTRs since after a migration
1413 		 * the guest CR3 might be restored prior to setting the nested
1414 		 * state which can lead to a load of wrong PDPTRs.
1415 		 */
1416 		if (CC(!load_pdptrs(vcpu, vcpu->arch.walk_mmu, vcpu->arch.cr3)))
1417 			return false;
1418 
1419 	if (!nested_svm_vmrun_msrpm(svm)) {
1420 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1421 		vcpu->run->internal.suberror =
1422 			KVM_INTERNAL_ERROR_EMULATION;
1423 		vcpu->run->internal.ndata = 0;
1424 		return false;
1425 	}
1426 
1427 	return true;
1428 }
1429 
1430 struct kvm_x86_nested_ops svm_nested_ops = {
1431 	.check_events = svm_check_nested_events,
1432 	.triple_fault = nested_svm_triple_fault,
1433 	.get_nested_state_pages = svm_get_nested_state_pages,
1434 	.get_state = svm_get_nested_state,
1435 	.set_state = svm_set_nested_state,
1436 };
1437