xref: /linux/arch/x86/kvm/svm/nested.c (revision e85d3e7b495bb6c0b847a693f5f6d4bd429fae55)
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 	vmcb_set_intercept(c, INTERCEPT_VMLOAD);
163 	vmcb_set_intercept(c, INTERCEPT_VMSAVE);
164 }
165 
166 static void copy_vmcb_control_area(struct vmcb_control_area *dst,
167 				   struct vmcb_control_area *from)
168 {
169 	unsigned int i;
170 
171 	for (i = 0; i < MAX_INTERCEPT; i++)
172 		dst->intercepts[i] = from->intercepts[i];
173 
174 	dst->iopm_base_pa         = from->iopm_base_pa;
175 	dst->msrpm_base_pa        = from->msrpm_base_pa;
176 	dst->tsc_offset           = from->tsc_offset;
177 	/* asid not copied, it is handled manually for svm->vmcb.  */
178 	dst->tlb_ctl              = from->tlb_ctl;
179 	dst->int_ctl              = from->int_ctl;
180 	dst->int_vector           = from->int_vector;
181 	dst->int_state            = from->int_state;
182 	dst->exit_code            = from->exit_code;
183 	dst->exit_code_hi         = from->exit_code_hi;
184 	dst->exit_info_1          = from->exit_info_1;
185 	dst->exit_info_2          = from->exit_info_2;
186 	dst->exit_int_info        = from->exit_int_info;
187 	dst->exit_int_info_err    = from->exit_int_info_err;
188 	dst->nested_ctl           = from->nested_ctl;
189 	dst->event_inj            = from->event_inj;
190 	dst->event_inj_err        = from->event_inj_err;
191 	dst->nested_cr3           = from->nested_cr3;
192 	dst->virt_ext              = from->virt_ext;
193 	dst->pause_filter_count   = from->pause_filter_count;
194 	dst->pause_filter_thresh  = from->pause_filter_thresh;
195 }
196 
197 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
198 {
199 	/*
200 	 * This function merges the msr permission bitmaps of kvm and the
201 	 * nested vmcb. It is optimized in that it only merges the parts where
202 	 * the kvm msr permission bitmap may contain zero bits
203 	 */
204 	int i;
205 
206 	if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
207 		return true;
208 
209 	for (i = 0; i < MSRPM_OFFSETS; i++) {
210 		u32 value, p;
211 		u64 offset;
212 
213 		if (msrpm_offsets[i] == 0xffffffff)
214 			break;
215 
216 		p      = msrpm_offsets[i];
217 		offset = svm->nested.ctl.msrpm_base_pa + (p * 4);
218 
219 		if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
220 			return false;
221 
222 		svm->nested.msrpm[p] = svm->msrpm[p] | value;
223 	}
224 
225 	svm->vmcb->control.msrpm_base_pa = __sme_set(__pa(svm->nested.msrpm));
226 
227 	return true;
228 }
229 
230 /*
231  * Bits 11:0 of bitmap address are ignored by hardware
232  */
233 static bool nested_svm_check_bitmap_pa(struct kvm_vcpu *vcpu, u64 pa, u32 size)
234 {
235 	u64 addr = PAGE_ALIGN(pa);
236 
237 	return kvm_vcpu_is_legal_gpa(vcpu, addr) &&
238 	    kvm_vcpu_is_legal_gpa(vcpu, addr + size - 1);
239 }
240 
241 static bool nested_vmcb_check_controls(struct kvm_vcpu *vcpu,
242 				       struct vmcb_control_area *control)
243 {
244 	if (CC(!vmcb_is_intercept(control, INTERCEPT_VMRUN)))
245 		return false;
246 
247 	if (CC(control->asid == 0))
248 		return false;
249 
250 	if (CC((control->nested_ctl & SVM_NESTED_CTL_NP_ENABLE) && !npt_enabled))
251 		return false;
252 
253 	if (CC(!nested_svm_check_bitmap_pa(vcpu, control->msrpm_base_pa,
254 					   MSRPM_SIZE)))
255 		return false;
256 	if (CC(!nested_svm_check_bitmap_pa(vcpu, control->iopm_base_pa,
257 					   IOPM_SIZE)))
258 		return false;
259 
260 	return true;
261 }
262 
263 static bool nested_vmcb_check_cr3_cr4(struct kvm_vcpu *vcpu,
264 				      struct vmcb_save_area *save)
265 {
266 	/*
267 	 * These checks are also performed by KVM_SET_SREGS,
268 	 * except that EFER.LMA is not checked by SVM against
269 	 * CR0.PG && EFER.LME.
270 	 */
271 	if ((save->efer & EFER_LME) && (save->cr0 & X86_CR0_PG)) {
272 		if (CC(!(save->cr4 & X86_CR4_PAE)) ||
273 		    CC(!(save->cr0 & X86_CR0_PE)) ||
274 		    CC(kvm_vcpu_is_illegal_gpa(vcpu, save->cr3)))
275 			return false;
276 	}
277 
278 	if (CC(!kvm_is_valid_cr4(vcpu, save->cr4)))
279 		return false;
280 
281 	return true;
282 }
283 
284 /* Common checks that apply to both L1 and L2 state.  */
285 static bool nested_vmcb_valid_sregs(struct kvm_vcpu *vcpu,
286 				    struct vmcb_save_area *save)
287 {
288 	/*
289 	 * FIXME: these should be done after copying the fields,
290 	 * to avoid TOC/TOU races.  For these save area checks
291 	 * the possible damage is limited since kvm_set_cr0 and
292 	 * kvm_set_cr4 handle failure; EFER_SVME is an exception
293 	 * so it is force-set later in nested_prepare_vmcb_save.
294 	 */
295 	if (CC(!(save->efer & EFER_SVME)))
296 		return false;
297 
298 	if (CC((save->cr0 & X86_CR0_CD) == 0 && (save->cr0 & X86_CR0_NW)) ||
299 	    CC(save->cr0 & ~0xffffffffULL))
300 		return false;
301 
302 	if (CC(!kvm_dr6_valid(save->dr6)) || CC(!kvm_dr7_valid(save->dr7)))
303 		return false;
304 
305 	if (!nested_vmcb_check_cr3_cr4(vcpu, save))
306 		return false;
307 
308 	if (CC(!kvm_valid_efer(vcpu, save->efer)))
309 		return false;
310 
311 	return true;
312 }
313 
314 void nested_load_control_from_vmcb12(struct vcpu_svm *svm,
315 				     struct vmcb_control_area *control)
316 {
317 	copy_vmcb_control_area(&svm->nested.ctl, control);
318 
319 	/* Copy it here because nested_svm_check_controls will check it.  */
320 	svm->nested.ctl.asid           = control->asid;
321 	svm->nested.ctl.msrpm_base_pa &= ~0x0fffULL;
322 	svm->nested.ctl.iopm_base_pa  &= ~0x0fffULL;
323 }
324 
325 /*
326  * Synchronize fields that are written by the processor, so that
327  * they can be copied back into the vmcb12.
328  */
329 void nested_sync_control_from_vmcb02(struct vcpu_svm *svm)
330 {
331 	u32 mask;
332 	svm->nested.ctl.event_inj      = svm->vmcb->control.event_inj;
333 	svm->nested.ctl.event_inj_err  = svm->vmcb->control.event_inj_err;
334 
335 	/* Only a few fields of int_ctl are written by the processor.  */
336 	mask = V_IRQ_MASK | V_TPR_MASK;
337 	if (!(svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK) &&
338 	    svm_is_intercept(svm, INTERCEPT_VINTR)) {
339 		/*
340 		 * In order to request an interrupt window, L0 is usurping
341 		 * svm->vmcb->control.int_ctl and possibly setting V_IRQ
342 		 * even if it was clear in L1's VMCB.  Restoring it would be
343 		 * wrong.  However, in this case V_IRQ will remain true until
344 		 * interrupt_window_interception calls svm_clear_vintr and
345 		 * restores int_ctl.  We can just leave it aside.
346 		 */
347 		mask &= ~V_IRQ_MASK;
348 	}
349 	svm->nested.ctl.int_ctl        &= ~mask;
350 	svm->nested.ctl.int_ctl        |= svm->vmcb->control.int_ctl & mask;
351 }
352 
353 /*
354  * Transfer any event that L0 or L1 wanted to inject into L2 to
355  * EXIT_INT_INFO.
356  */
357 static void nested_save_pending_event_to_vmcb12(struct vcpu_svm *svm,
358 						struct vmcb *vmcb12)
359 {
360 	struct kvm_vcpu *vcpu = &svm->vcpu;
361 	u32 exit_int_info = 0;
362 	unsigned int nr;
363 
364 	if (vcpu->arch.exception.injected) {
365 		nr = vcpu->arch.exception.nr;
366 		exit_int_info = nr | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT;
367 
368 		if (vcpu->arch.exception.has_error_code) {
369 			exit_int_info |= SVM_EVTINJ_VALID_ERR;
370 			vmcb12->control.exit_int_info_err =
371 				vcpu->arch.exception.error_code;
372 		}
373 
374 	} else if (vcpu->arch.nmi_injected) {
375 		exit_int_info = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
376 
377 	} else if (vcpu->arch.interrupt.injected) {
378 		nr = vcpu->arch.interrupt.nr;
379 		exit_int_info = nr | SVM_EVTINJ_VALID;
380 
381 		if (vcpu->arch.interrupt.soft)
382 			exit_int_info |= SVM_EVTINJ_TYPE_SOFT;
383 		else
384 			exit_int_info |= SVM_EVTINJ_TYPE_INTR;
385 	}
386 
387 	vmcb12->control.exit_int_info = exit_int_info;
388 }
389 
390 static inline bool nested_npt_enabled(struct vcpu_svm *svm)
391 {
392 	return svm->nested.ctl.nested_ctl & SVM_NESTED_CTL_NP_ENABLE;
393 }
394 
395 static void nested_svm_transition_tlb_flush(struct kvm_vcpu *vcpu)
396 {
397 	/*
398 	 * TODO: optimize unconditional TLB flush/MMU sync.  A partial list of
399 	 * things to fix before this can be conditional:
400 	 *
401 	 *  - Flush TLBs for both L1 and L2 remote TLB flush
402 	 *  - Honor L1's request to flush an ASID on nested VMRUN
403 	 *  - Sync nested NPT MMU on VMRUN that flushes L2's ASID[*]
404 	 *  - Don't crush a pending TLB flush in vmcb02 on nested VMRUN
405 	 *  - Flush L1's ASID on KVM_REQ_TLB_FLUSH_GUEST
406 	 *
407 	 * [*] Unlike nested EPT, SVM's ASID management can invalidate nested
408 	 *     NPT guest-physical mappings on VMRUN.
409 	 */
410 	kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
411 	kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
412 }
413 
414 /*
415  * Load guest's/host's cr3 on nested vmentry or vmexit. @nested_npt is true
416  * if we are emulating VM-Entry into a guest with NPT enabled.
417  */
418 static int nested_svm_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3,
419 			       bool nested_npt, bool reload_pdptrs)
420 {
421 	if (CC(kvm_vcpu_is_illegal_gpa(vcpu, cr3)))
422 		return -EINVAL;
423 
424 	if (reload_pdptrs && !nested_npt && is_pae_paging(vcpu) &&
425 	    CC(!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)))
426 		return -EINVAL;
427 
428 	if (!nested_npt)
429 		kvm_mmu_new_pgd(vcpu, cr3);
430 
431 	vcpu->arch.cr3 = cr3;
432 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
433 
434 	/* Re-initialize the MMU, e.g. to pick up CR4 MMU role changes. */
435 	kvm_init_mmu(vcpu);
436 
437 	return 0;
438 }
439 
440 void nested_vmcb02_compute_g_pat(struct vcpu_svm *svm)
441 {
442 	if (!svm->nested.vmcb02.ptr)
443 		return;
444 
445 	/* FIXME: merge g_pat from vmcb01 and vmcb12.  */
446 	svm->nested.vmcb02.ptr->save.g_pat = svm->vmcb01.ptr->save.g_pat;
447 }
448 
449 static void nested_vmcb02_prepare_save(struct vcpu_svm *svm, struct vmcb *vmcb12)
450 {
451 	bool new_vmcb12 = false;
452 
453 	nested_vmcb02_compute_g_pat(svm);
454 
455 	/* Load the nested guest state */
456 	if (svm->nested.vmcb12_gpa != svm->nested.last_vmcb12_gpa) {
457 		new_vmcb12 = true;
458 		svm->nested.last_vmcb12_gpa = svm->nested.vmcb12_gpa;
459 	}
460 
461 	if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_SEG))) {
462 		svm->vmcb->save.es = vmcb12->save.es;
463 		svm->vmcb->save.cs = vmcb12->save.cs;
464 		svm->vmcb->save.ss = vmcb12->save.ss;
465 		svm->vmcb->save.ds = vmcb12->save.ds;
466 		svm->vmcb->save.cpl = vmcb12->save.cpl;
467 		vmcb_mark_dirty(svm->vmcb, VMCB_SEG);
468 	}
469 
470 	if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_DT))) {
471 		svm->vmcb->save.gdtr = vmcb12->save.gdtr;
472 		svm->vmcb->save.idtr = vmcb12->save.idtr;
473 		vmcb_mark_dirty(svm->vmcb, VMCB_DT);
474 	}
475 
476 	kvm_set_rflags(&svm->vcpu, vmcb12->save.rflags | X86_EFLAGS_FIXED);
477 
478 	/*
479 	 * Force-set EFER_SVME even though it is checked earlier on the
480 	 * VMCB12, because the guest can flip the bit between the check
481 	 * and now.  Clearing EFER_SVME would call svm_free_nested.
482 	 */
483 	svm_set_efer(&svm->vcpu, vmcb12->save.efer | EFER_SVME);
484 
485 	svm_set_cr0(&svm->vcpu, vmcb12->save.cr0);
486 	svm_set_cr4(&svm->vcpu, vmcb12->save.cr4);
487 
488 	svm->vcpu.arch.cr2 = vmcb12->save.cr2;
489 
490 	kvm_rax_write(&svm->vcpu, vmcb12->save.rax);
491 	kvm_rsp_write(&svm->vcpu, vmcb12->save.rsp);
492 	kvm_rip_write(&svm->vcpu, vmcb12->save.rip);
493 
494 	/* In case we don't even reach vcpu_run, the fields are not updated */
495 	svm->vmcb->save.rax = vmcb12->save.rax;
496 	svm->vmcb->save.rsp = vmcb12->save.rsp;
497 	svm->vmcb->save.rip = vmcb12->save.rip;
498 
499 	/* These bits will be set properly on the first execution when new_vmc12 is true */
500 	if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_DR))) {
501 		svm->vmcb->save.dr7 = vmcb12->save.dr7 | DR7_FIXED_1;
502 		svm->vcpu.arch.dr6  = vmcb12->save.dr6 | DR6_ACTIVE_LOW;
503 		vmcb_mark_dirty(svm->vmcb, VMCB_DR);
504 	}
505 }
506 
507 static void nested_vmcb02_prepare_control(struct vcpu_svm *svm)
508 {
509 	const u32 int_ctl_vmcb01_bits =
510 		V_INTR_MASKING_MASK | V_GIF_MASK | V_GIF_ENABLE_MASK;
511 
512 	const u32 int_ctl_vmcb12_bits = V_TPR_MASK | V_IRQ_INJECTION_BITS_MASK;
513 
514 	struct kvm_vcpu *vcpu = &svm->vcpu;
515 
516 	/*
517 	 * Filled at exit: exit_code, exit_code_hi, exit_info_1, exit_info_2,
518 	 * exit_int_info, exit_int_info_err, next_rip, insn_len, insn_bytes.
519 	 */
520 
521 	/*
522 	 * Also covers avic_vapic_bar, avic_backing_page, avic_logical_id,
523 	 * avic_physical_id.
524 	 */
525 	WARN_ON(kvm_apicv_activated(svm->vcpu.kvm));
526 
527 	/* Copied from vmcb01.  msrpm_base can be overwritten later.  */
528 	svm->vmcb->control.nested_ctl = svm->vmcb01.ptr->control.nested_ctl;
529 	svm->vmcb->control.iopm_base_pa = svm->vmcb01.ptr->control.iopm_base_pa;
530 	svm->vmcb->control.msrpm_base_pa = svm->vmcb01.ptr->control.msrpm_base_pa;
531 
532 	/* Done at vmrun: asid.  */
533 
534 	/* Also overwritten later if necessary.  */
535 	svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
536 
537 	/* nested_cr3.  */
538 	if (nested_npt_enabled(svm))
539 		nested_svm_init_mmu_context(vcpu);
540 
541 	svm->vmcb->control.tsc_offset = vcpu->arch.tsc_offset =
542 		vcpu->arch.l1_tsc_offset + svm->nested.ctl.tsc_offset;
543 
544 	svm->vmcb->control.int_ctl             =
545 		(svm->nested.ctl.int_ctl & int_ctl_vmcb12_bits) |
546 		(svm->vmcb01.ptr->control.int_ctl & int_ctl_vmcb01_bits);
547 
548 	svm->vmcb->control.virt_ext            = svm->nested.ctl.virt_ext;
549 	svm->vmcb->control.int_vector          = svm->nested.ctl.int_vector;
550 	svm->vmcb->control.int_state           = svm->nested.ctl.int_state;
551 	svm->vmcb->control.event_inj           = svm->nested.ctl.event_inj;
552 	svm->vmcb->control.event_inj_err       = svm->nested.ctl.event_inj_err;
553 
554 	svm->vmcb->control.pause_filter_count  = svm->nested.ctl.pause_filter_count;
555 	svm->vmcb->control.pause_filter_thresh = svm->nested.ctl.pause_filter_thresh;
556 
557 	nested_svm_transition_tlb_flush(vcpu);
558 
559 	/* Enter Guest-Mode */
560 	enter_guest_mode(vcpu);
561 
562 	/*
563 	 * Merge guest and host intercepts - must be called with vcpu in
564 	 * guest-mode to take effect.
565 	 */
566 	recalc_intercepts(svm);
567 }
568 
569 static void nested_svm_copy_common_state(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
570 {
571 	/*
572 	 * Some VMCB state is shared between L1 and L2 and thus has to be
573 	 * moved at the time of nested vmrun and vmexit.
574 	 *
575 	 * VMLOAD/VMSAVE state would also belong in this category, but KVM
576 	 * always performs VMLOAD and VMSAVE from the VMCB01.
577 	 */
578 	to_vmcb->save.spec_ctrl = from_vmcb->save.spec_ctrl;
579 }
580 
581 int enter_svm_guest_mode(struct kvm_vcpu *vcpu, u64 vmcb12_gpa,
582 			 struct vmcb *vmcb12, bool from_vmrun)
583 {
584 	struct vcpu_svm *svm = to_svm(vcpu);
585 	int ret;
586 
587 	trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb12_gpa,
588 			       vmcb12->save.rip,
589 			       vmcb12->control.int_ctl,
590 			       vmcb12->control.event_inj,
591 			       vmcb12->control.nested_ctl);
592 
593 	trace_kvm_nested_intercepts(vmcb12->control.intercepts[INTERCEPT_CR] & 0xffff,
594 				    vmcb12->control.intercepts[INTERCEPT_CR] >> 16,
595 				    vmcb12->control.intercepts[INTERCEPT_EXCEPTION],
596 				    vmcb12->control.intercepts[INTERCEPT_WORD3],
597 				    vmcb12->control.intercepts[INTERCEPT_WORD4],
598 				    vmcb12->control.intercepts[INTERCEPT_WORD5]);
599 
600 
601 	svm->nested.vmcb12_gpa = vmcb12_gpa;
602 
603 	WARN_ON(svm->vmcb == svm->nested.vmcb02.ptr);
604 
605 	nested_svm_copy_common_state(svm->vmcb01.ptr, svm->nested.vmcb02.ptr);
606 
607 	svm_switch_vmcb(svm, &svm->nested.vmcb02);
608 	nested_vmcb02_prepare_control(svm);
609 	nested_vmcb02_prepare_save(svm, vmcb12);
610 
611 	ret = nested_svm_load_cr3(&svm->vcpu, vmcb12->save.cr3,
612 				  nested_npt_enabled(svm), from_vmrun);
613 	if (ret)
614 		return ret;
615 
616 	if (!npt_enabled)
617 		vcpu->arch.mmu->inject_page_fault = svm_inject_page_fault_nested;
618 
619 	if (!from_vmrun)
620 		kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
621 
622 	svm_set_gif(svm, true);
623 
624 	return 0;
625 }
626 
627 int nested_svm_vmrun(struct kvm_vcpu *vcpu)
628 {
629 	struct vcpu_svm *svm = to_svm(vcpu);
630 	int ret;
631 	struct vmcb *vmcb12;
632 	struct kvm_host_map map;
633 	u64 vmcb12_gpa;
634 
635 	if (!svm->nested.hsave_msr) {
636 		kvm_inject_gp(vcpu, 0);
637 		return 1;
638 	}
639 
640 	if (is_smm(vcpu)) {
641 		kvm_queue_exception(vcpu, UD_VECTOR);
642 		return 1;
643 	}
644 
645 	vmcb12_gpa = svm->vmcb->save.rax;
646 	ret = kvm_vcpu_map(vcpu, gpa_to_gfn(vmcb12_gpa), &map);
647 	if (ret == -EINVAL) {
648 		kvm_inject_gp(vcpu, 0);
649 		return 1;
650 	} else if (ret) {
651 		return kvm_skip_emulated_instruction(vcpu);
652 	}
653 
654 	ret = kvm_skip_emulated_instruction(vcpu);
655 
656 	vmcb12 = map.hva;
657 
658 	if (WARN_ON_ONCE(!svm->nested.initialized))
659 		return -EINVAL;
660 
661 	nested_load_control_from_vmcb12(svm, &vmcb12->control);
662 
663 	if (!nested_vmcb_valid_sregs(vcpu, &vmcb12->save) ||
664 	    !nested_vmcb_check_controls(vcpu, &svm->nested.ctl)) {
665 		vmcb12->control.exit_code    = SVM_EXIT_ERR;
666 		vmcb12->control.exit_code_hi = 0;
667 		vmcb12->control.exit_info_1  = 0;
668 		vmcb12->control.exit_info_2  = 0;
669 		goto out;
670 	}
671 
672 	/*
673 	 * Since vmcb01 is not in use, we can use it to store some of the L1
674 	 * state.
675 	 */
676 	svm->vmcb01.ptr->save.efer   = vcpu->arch.efer;
677 	svm->vmcb01.ptr->save.cr0    = kvm_read_cr0(vcpu);
678 	svm->vmcb01.ptr->save.cr4    = vcpu->arch.cr4;
679 	svm->vmcb01.ptr->save.rflags = kvm_get_rflags(vcpu);
680 	svm->vmcb01.ptr->save.rip    = kvm_rip_read(vcpu);
681 
682 	if (!npt_enabled)
683 		svm->vmcb01.ptr->save.cr3 = kvm_read_cr3(vcpu);
684 
685 	svm->nested.nested_run_pending = 1;
686 
687 	if (enter_svm_guest_mode(vcpu, vmcb12_gpa, vmcb12, true))
688 		goto out_exit_err;
689 
690 	if (nested_svm_vmrun_msrpm(svm))
691 		goto out;
692 
693 out_exit_err:
694 	svm->nested.nested_run_pending = 0;
695 
696 	svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
697 	svm->vmcb->control.exit_code_hi = 0;
698 	svm->vmcb->control.exit_info_1  = 0;
699 	svm->vmcb->control.exit_info_2  = 0;
700 
701 	nested_svm_vmexit(svm);
702 
703 out:
704 	kvm_vcpu_unmap(vcpu, &map, true);
705 
706 	return ret;
707 }
708 
709 /* Copy state save area fields which are handled by VMRUN */
710 void svm_copy_vmrun_state(struct vmcb_save_area *to_save,
711 			  struct vmcb_save_area *from_save)
712 {
713 	to_save->es = from_save->es;
714 	to_save->cs = from_save->cs;
715 	to_save->ss = from_save->ss;
716 	to_save->ds = from_save->ds;
717 	to_save->gdtr = from_save->gdtr;
718 	to_save->idtr = from_save->idtr;
719 	to_save->rflags = from_save->rflags | X86_EFLAGS_FIXED;
720 	to_save->efer = from_save->efer;
721 	to_save->cr0 = from_save->cr0;
722 	to_save->cr3 = from_save->cr3;
723 	to_save->cr4 = from_save->cr4;
724 	to_save->rax = from_save->rax;
725 	to_save->rsp = from_save->rsp;
726 	to_save->rip = from_save->rip;
727 	to_save->cpl = 0;
728 }
729 
730 void svm_copy_vmloadsave_state(struct vmcb *to_vmcb, struct vmcb *from_vmcb)
731 {
732 	to_vmcb->save.fs = from_vmcb->save.fs;
733 	to_vmcb->save.gs = from_vmcb->save.gs;
734 	to_vmcb->save.tr = from_vmcb->save.tr;
735 	to_vmcb->save.ldtr = from_vmcb->save.ldtr;
736 	to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
737 	to_vmcb->save.star = from_vmcb->save.star;
738 	to_vmcb->save.lstar = from_vmcb->save.lstar;
739 	to_vmcb->save.cstar = from_vmcb->save.cstar;
740 	to_vmcb->save.sfmask = from_vmcb->save.sfmask;
741 	to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
742 	to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
743 	to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
744 }
745 
746 int nested_svm_vmexit(struct vcpu_svm *svm)
747 {
748 	struct kvm_vcpu *vcpu = &svm->vcpu;
749 	struct vmcb *vmcb12;
750 	struct vmcb *vmcb = svm->vmcb;
751 	struct kvm_host_map map;
752 	int rc;
753 
754 	/* Triple faults in L2 should never escape. */
755 	WARN_ON_ONCE(kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu));
756 
757 	rc = kvm_vcpu_map(vcpu, gpa_to_gfn(svm->nested.vmcb12_gpa), &map);
758 	if (rc) {
759 		if (rc == -EINVAL)
760 			kvm_inject_gp(vcpu, 0);
761 		return 1;
762 	}
763 
764 	vmcb12 = map.hva;
765 
766 	/* Exit Guest-Mode */
767 	leave_guest_mode(vcpu);
768 	svm->nested.vmcb12_gpa = 0;
769 	WARN_ON_ONCE(svm->nested.nested_run_pending);
770 
771 	kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
772 
773 	/* in case we halted in L2 */
774 	svm->vcpu.arch.mp_state = KVM_MP_STATE_RUNNABLE;
775 
776 	/* Give the current vmcb to the guest */
777 
778 	vmcb12->save.es     = vmcb->save.es;
779 	vmcb12->save.cs     = vmcb->save.cs;
780 	vmcb12->save.ss     = vmcb->save.ss;
781 	vmcb12->save.ds     = vmcb->save.ds;
782 	vmcb12->save.gdtr   = vmcb->save.gdtr;
783 	vmcb12->save.idtr   = vmcb->save.idtr;
784 	vmcb12->save.efer   = svm->vcpu.arch.efer;
785 	vmcb12->save.cr0    = kvm_read_cr0(vcpu);
786 	vmcb12->save.cr3    = kvm_read_cr3(vcpu);
787 	vmcb12->save.cr2    = vmcb->save.cr2;
788 	vmcb12->save.cr4    = svm->vcpu.arch.cr4;
789 	vmcb12->save.rflags = kvm_get_rflags(vcpu);
790 	vmcb12->save.rip    = kvm_rip_read(vcpu);
791 	vmcb12->save.rsp    = kvm_rsp_read(vcpu);
792 	vmcb12->save.rax    = kvm_rax_read(vcpu);
793 	vmcb12->save.dr7    = vmcb->save.dr7;
794 	vmcb12->save.dr6    = svm->vcpu.arch.dr6;
795 	vmcb12->save.cpl    = vmcb->save.cpl;
796 
797 	vmcb12->control.int_state         = vmcb->control.int_state;
798 	vmcb12->control.exit_code         = vmcb->control.exit_code;
799 	vmcb12->control.exit_code_hi      = vmcb->control.exit_code_hi;
800 	vmcb12->control.exit_info_1       = vmcb->control.exit_info_1;
801 	vmcb12->control.exit_info_2       = vmcb->control.exit_info_2;
802 
803 	if (vmcb12->control.exit_code != SVM_EXIT_ERR)
804 		nested_save_pending_event_to_vmcb12(svm, vmcb12);
805 
806 	if (svm->nrips_enabled)
807 		vmcb12->control.next_rip  = vmcb->control.next_rip;
808 
809 	vmcb12->control.int_ctl           = svm->nested.ctl.int_ctl;
810 	vmcb12->control.tlb_ctl           = svm->nested.ctl.tlb_ctl;
811 	vmcb12->control.event_inj         = svm->nested.ctl.event_inj;
812 	vmcb12->control.event_inj_err     = svm->nested.ctl.event_inj_err;
813 
814 	vmcb12->control.pause_filter_count =
815 		svm->vmcb->control.pause_filter_count;
816 	vmcb12->control.pause_filter_thresh =
817 		svm->vmcb->control.pause_filter_thresh;
818 
819 	nested_svm_copy_common_state(svm->nested.vmcb02.ptr, svm->vmcb01.ptr);
820 
821 	svm_switch_vmcb(svm, &svm->vmcb01);
822 
823 	/*
824 	 * On vmexit the  GIF is set to false and
825 	 * no event can be injected in L1.
826 	 */
827 	svm_set_gif(svm, false);
828 	svm->vmcb->control.exit_int_info = 0;
829 
830 	svm->vcpu.arch.tsc_offset = svm->vcpu.arch.l1_tsc_offset;
831 	if (svm->vmcb->control.tsc_offset != svm->vcpu.arch.tsc_offset) {
832 		svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset;
833 		vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
834 	}
835 
836 	svm->nested.ctl.nested_cr3 = 0;
837 
838 	/*
839 	 * Restore processor state that had been saved in vmcb01
840 	 */
841 	kvm_set_rflags(vcpu, svm->vmcb->save.rflags);
842 	svm_set_efer(vcpu, svm->vmcb->save.efer);
843 	svm_set_cr0(vcpu, svm->vmcb->save.cr0 | X86_CR0_PE);
844 	svm_set_cr4(vcpu, svm->vmcb->save.cr4);
845 	kvm_rax_write(vcpu, svm->vmcb->save.rax);
846 	kvm_rsp_write(vcpu, svm->vmcb->save.rsp);
847 	kvm_rip_write(vcpu, svm->vmcb->save.rip);
848 
849 	svm->vcpu.arch.dr7 = DR7_FIXED_1;
850 	kvm_update_dr7(&svm->vcpu);
851 
852 	trace_kvm_nested_vmexit_inject(vmcb12->control.exit_code,
853 				       vmcb12->control.exit_info_1,
854 				       vmcb12->control.exit_info_2,
855 				       vmcb12->control.exit_int_info,
856 				       vmcb12->control.exit_int_info_err,
857 				       KVM_ISA_SVM);
858 
859 	kvm_vcpu_unmap(vcpu, &map, true);
860 
861 	nested_svm_transition_tlb_flush(vcpu);
862 
863 	nested_svm_uninit_mmu_context(vcpu);
864 
865 	rc = nested_svm_load_cr3(vcpu, svm->vmcb->save.cr3, false, true);
866 	if (rc)
867 		return 1;
868 
869 	/*
870 	 * Drop what we picked up for L2 via svm_complete_interrupts() so it
871 	 * doesn't end up in L1.
872 	 */
873 	svm->vcpu.arch.nmi_injected = false;
874 	kvm_clear_exception_queue(vcpu);
875 	kvm_clear_interrupt_queue(vcpu);
876 
877 	/*
878 	 * If we are here following the completion of a VMRUN that
879 	 * is being single-stepped, queue the pending #DB intercept
880 	 * right now so that it an be accounted for before we execute
881 	 * L1's next instruction.
882 	 */
883 	if (unlikely(svm->vmcb->save.rflags & X86_EFLAGS_TF))
884 		kvm_queue_exception(&(svm->vcpu), DB_VECTOR);
885 
886 	return 0;
887 }
888 
889 static void nested_svm_triple_fault(struct kvm_vcpu *vcpu)
890 {
891 	nested_svm_simple_vmexit(to_svm(vcpu), SVM_EXIT_SHUTDOWN);
892 }
893 
894 int svm_allocate_nested(struct vcpu_svm *svm)
895 {
896 	struct page *vmcb02_page;
897 
898 	if (svm->nested.initialized)
899 		return 0;
900 
901 	vmcb02_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
902 	if (!vmcb02_page)
903 		return -ENOMEM;
904 	svm->nested.vmcb02.ptr = page_address(vmcb02_page);
905 	svm->nested.vmcb02.pa = __sme_set(page_to_pfn(vmcb02_page) << PAGE_SHIFT);
906 
907 	svm->nested.msrpm = svm_vcpu_alloc_msrpm();
908 	if (!svm->nested.msrpm)
909 		goto err_free_vmcb02;
910 	svm_vcpu_init_msrpm(&svm->vcpu, svm->nested.msrpm);
911 
912 	svm->nested.initialized = true;
913 	return 0;
914 
915 err_free_vmcb02:
916 	__free_page(vmcb02_page);
917 	return -ENOMEM;
918 }
919 
920 void svm_free_nested(struct vcpu_svm *svm)
921 {
922 	if (!svm->nested.initialized)
923 		return;
924 
925 	svm_vcpu_free_msrpm(svm->nested.msrpm);
926 	svm->nested.msrpm = NULL;
927 
928 	__free_page(virt_to_page(svm->nested.vmcb02.ptr));
929 	svm->nested.vmcb02.ptr = NULL;
930 
931 	/*
932 	 * When last_vmcb12_gpa matches the current vmcb12 gpa,
933 	 * some vmcb12 fields are not loaded if they are marked clean
934 	 * in the vmcb12, since in this case they are up to date already.
935 	 *
936 	 * When the vmcb02 is freed, this optimization becomes invalid.
937 	 */
938 	svm->nested.last_vmcb12_gpa = INVALID_GPA;
939 
940 	svm->nested.initialized = false;
941 }
942 
943 /*
944  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
945  */
946 void svm_leave_nested(struct vcpu_svm *svm)
947 {
948 	struct kvm_vcpu *vcpu = &svm->vcpu;
949 
950 	if (is_guest_mode(vcpu)) {
951 		svm->nested.nested_run_pending = 0;
952 		svm->nested.vmcb12_gpa = INVALID_GPA;
953 
954 		leave_guest_mode(vcpu);
955 
956 		svm_switch_vmcb(svm, &svm->vmcb01);
957 
958 		nested_svm_uninit_mmu_context(vcpu);
959 		vmcb_mark_all_dirty(svm->vmcb);
960 	}
961 
962 	kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
963 }
964 
965 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
966 {
967 	u32 offset, msr, value;
968 	int write, mask;
969 
970 	if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
971 		return NESTED_EXIT_HOST;
972 
973 	msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
974 	offset = svm_msrpm_offset(msr);
975 	write  = svm->vmcb->control.exit_info_1 & 1;
976 	mask   = 1 << ((2 * (msr & 0xf)) + write);
977 
978 	if (offset == MSR_INVALID)
979 		return NESTED_EXIT_DONE;
980 
981 	/* Offset is in 32 bit units but need in 8 bit units */
982 	offset *= 4;
983 
984 	if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.ctl.msrpm_base_pa + offset, &value, 4))
985 		return NESTED_EXIT_DONE;
986 
987 	return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
988 }
989 
990 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
991 {
992 	unsigned port, size, iopm_len;
993 	u16 val, mask;
994 	u8 start_bit;
995 	u64 gpa;
996 
997 	if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_IOIO_PROT)))
998 		return NESTED_EXIT_HOST;
999 
1000 	port = svm->vmcb->control.exit_info_1 >> 16;
1001 	size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
1002 		SVM_IOIO_SIZE_SHIFT;
1003 	gpa  = svm->nested.ctl.iopm_base_pa + (port / 8);
1004 	start_bit = port % 8;
1005 	iopm_len = (start_bit + size > 8) ? 2 : 1;
1006 	mask = (0xf >> (4 - size)) << start_bit;
1007 	val = 0;
1008 
1009 	if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
1010 		return NESTED_EXIT_DONE;
1011 
1012 	return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
1013 }
1014 
1015 static int nested_svm_intercept(struct vcpu_svm *svm)
1016 {
1017 	u32 exit_code = svm->vmcb->control.exit_code;
1018 	int vmexit = NESTED_EXIT_HOST;
1019 
1020 	switch (exit_code) {
1021 	case SVM_EXIT_MSR:
1022 		vmexit = nested_svm_exit_handled_msr(svm);
1023 		break;
1024 	case SVM_EXIT_IOIO:
1025 		vmexit = nested_svm_intercept_ioio(svm);
1026 		break;
1027 	case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
1028 		if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
1029 			vmexit = NESTED_EXIT_DONE;
1030 		break;
1031 	}
1032 	case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
1033 		if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
1034 			vmexit = NESTED_EXIT_DONE;
1035 		break;
1036 	}
1037 	case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1038 		/*
1039 		 * Host-intercepted exceptions have been checked already in
1040 		 * nested_svm_exit_special.  There is nothing to do here,
1041 		 * the vmexit is injected by svm_check_nested_events.
1042 		 */
1043 		vmexit = NESTED_EXIT_DONE;
1044 		break;
1045 	}
1046 	case SVM_EXIT_ERR: {
1047 		vmexit = NESTED_EXIT_DONE;
1048 		break;
1049 	}
1050 	default: {
1051 		if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
1052 			vmexit = NESTED_EXIT_DONE;
1053 	}
1054 	}
1055 
1056 	return vmexit;
1057 }
1058 
1059 int nested_svm_exit_handled(struct vcpu_svm *svm)
1060 {
1061 	int vmexit;
1062 
1063 	vmexit = nested_svm_intercept(svm);
1064 
1065 	if (vmexit == NESTED_EXIT_DONE)
1066 		nested_svm_vmexit(svm);
1067 
1068 	return vmexit;
1069 }
1070 
1071 int nested_svm_check_permissions(struct kvm_vcpu *vcpu)
1072 {
1073 	if (!(vcpu->arch.efer & EFER_SVME) || !is_paging(vcpu)) {
1074 		kvm_queue_exception(vcpu, UD_VECTOR);
1075 		return 1;
1076 	}
1077 
1078 	if (to_svm(vcpu)->vmcb->save.cpl) {
1079 		kvm_inject_gp(vcpu, 0);
1080 		return 1;
1081 	}
1082 
1083 	return 0;
1084 }
1085 
1086 static bool nested_exit_on_exception(struct vcpu_svm *svm)
1087 {
1088 	unsigned int nr = svm->vcpu.arch.exception.nr;
1089 
1090 	return (svm->nested.ctl.intercepts[INTERCEPT_EXCEPTION] & BIT(nr));
1091 }
1092 
1093 static void nested_svm_inject_exception_vmexit(struct vcpu_svm *svm)
1094 {
1095 	unsigned int nr = svm->vcpu.arch.exception.nr;
1096 
1097 	svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
1098 	svm->vmcb->control.exit_code_hi = 0;
1099 
1100 	if (svm->vcpu.arch.exception.has_error_code)
1101 		svm->vmcb->control.exit_info_1 = svm->vcpu.arch.exception.error_code;
1102 
1103 	/*
1104 	 * EXITINFO2 is undefined for all exception intercepts other
1105 	 * than #PF.
1106 	 */
1107 	if (nr == PF_VECTOR) {
1108 		if (svm->vcpu.arch.exception.nested_apf)
1109 			svm->vmcb->control.exit_info_2 = svm->vcpu.arch.apf.nested_apf_token;
1110 		else if (svm->vcpu.arch.exception.has_payload)
1111 			svm->vmcb->control.exit_info_2 = svm->vcpu.arch.exception.payload;
1112 		else
1113 			svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
1114 	} else if (nr == DB_VECTOR) {
1115 		/* See inject_pending_event.  */
1116 		kvm_deliver_exception_payload(&svm->vcpu);
1117 		if (svm->vcpu.arch.dr7 & DR7_GD) {
1118 			svm->vcpu.arch.dr7 &= ~DR7_GD;
1119 			kvm_update_dr7(&svm->vcpu);
1120 		}
1121 	} else
1122 		WARN_ON(svm->vcpu.arch.exception.has_payload);
1123 
1124 	nested_svm_vmexit(svm);
1125 }
1126 
1127 static inline bool nested_exit_on_init(struct vcpu_svm *svm)
1128 {
1129 	return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_INIT);
1130 }
1131 
1132 static int svm_check_nested_events(struct kvm_vcpu *vcpu)
1133 {
1134 	struct vcpu_svm *svm = to_svm(vcpu);
1135 	bool block_nested_events =
1136 		kvm_event_needs_reinjection(vcpu) || svm->nested.nested_run_pending;
1137 	struct kvm_lapic *apic = vcpu->arch.apic;
1138 
1139 	if (lapic_in_kernel(vcpu) &&
1140 	    test_bit(KVM_APIC_INIT, &apic->pending_events)) {
1141 		if (block_nested_events)
1142 			return -EBUSY;
1143 		if (!nested_exit_on_init(svm))
1144 			return 0;
1145 		nested_svm_simple_vmexit(svm, SVM_EXIT_INIT);
1146 		return 0;
1147 	}
1148 
1149 	if (vcpu->arch.exception.pending) {
1150 		/*
1151 		 * Only a pending nested run can block a pending exception.
1152 		 * Otherwise an injected NMI/interrupt should either be
1153 		 * lost or delivered to the nested hypervisor in the EXITINTINFO
1154 		 * vmcb field, while delivering the pending exception.
1155 		 */
1156 		if (svm->nested.nested_run_pending)
1157                         return -EBUSY;
1158 		if (!nested_exit_on_exception(svm))
1159 			return 0;
1160 		nested_svm_inject_exception_vmexit(svm);
1161 		return 0;
1162 	}
1163 
1164 	if (vcpu->arch.smi_pending && !svm_smi_blocked(vcpu)) {
1165 		if (block_nested_events)
1166 			return -EBUSY;
1167 		if (!nested_exit_on_smi(svm))
1168 			return 0;
1169 		nested_svm_simple_vmexit(svm, SVM_EXIT_SMI);
1170 		return 0;
1171 	}
1172 
1173 	if (vcpu->arch.nmi_pending && !svm_nmi_blocked(vcpu)) {
1174 		if (block_nested_events)
1175 			return -EBUSY;
1176 		if (!nested_exit_on_nmi(svm))
1177 			return 0;
1178 		nested_svm_simple_vmexit(svm, SVM_EXIT_NMI);
1179 		return 0;
1180 	}
1181 
1182 	if (kvm_cpu_has_interrupt(vcpu) && !svm_interrupt_blocked(vcpu)) {
1183 		if (block_nested_events)
1184 			return -EBUSY;
1185 		if (!nested_exit_on_intr(svm))
1186 			return 0;
1187 		trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
1188 		nested_svm_simple_vmexit(svm, SVM_EXIT_INTR);
1189 		return 0;
1190 	}
1191 
1192 	return 0;
1193 }
1194 
1195 int nested_svm_exit_special(struct vcpu_svm *svm)
1196 {
1197 	u32 exit_code = svm->vmcb->control.exit_code;
1198 
1199 	switch (exit_code) {
1200 	case SVM_EXIT_INTR:
1201 	case SVM_EXIT_NMI:
1202 	case SVM_EXIT_NPF:
1203 		return NESTED_EXIT_HOST;
1204 	case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1205 		u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
1206 
1207 		if (svm->vmcb01.ptr->control.intercepts[INTERCEPT_EXCEPTION] &
1208 		    excp_bits)
1209 			return NESTED_EXIT_HOST;
1210 		else if (exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR &&
1211 			 svm->vcpu.arch.apf.host_apf_flags)
1212 			/* Trap async PF even if not shadowing */
1213 			return NESTED_EXIT_HOST;
1214 		break;
1215 	}
1216 	default:
1217 		break;
1218 	}
1219 
1220 	return NESTED_EXIT_CONTINUE;
1221 }
1222 
1223 static int svm_get_nested_state(struct kvm_vcpu *vcpu,
1224 				struct kvm_nested_state __user *user_kvm_nested_state,
1225 				u32 user_data_size)
1226 {
1227 	struct vcpu_svm *svm;
1228 	struct kvm_nested_state kvm_state = {
1229 		.flags = 0,
1230 		.format = KVM_STATE_NESTED_FORMAT_SVM,
1231 		.size = sizeof(kvm_state),
1232 	};
1233 	struct vmcb __user *user_vmcb = (struct vmcb __user *)
1234 		&user_kvm_nested_state->data.svm[0];
1235 
1236 	if (!vcpu)
1237 		return kvm_state.size + KVM_STATE_NESTED_SVM_VMCB_SIZE;
1238 
1239 	svm = to_svm(vcpu);
1240 
1241 	if (user_data_size < kvm_state.size)
1242 		goto out;
1243 
1244 	/* First fill in the header and copy it out.  */
1245 	if (is_guest_mode(vcpu)) {
1246 		kvm_state.hdr.svm.vmcb_pa = svm->nested.vmcb12_gpa;
1247 		kvm_state.size += KVM_STATE_NESTED_SVM_VMCB_SIZE;
1248 		kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
1249 
1250 		if (svm->nested.nested_run_pending)
1251 			kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
1252 	}
1253 
1254 	if (gif_set(svm))
1255 		kvm_state.flags |= KVM_STATE_NESTED_GIF_SET;
1256 
1257 	if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
1258 		return -EFAULT;
1259 
1260 	if (!is_guest_mode(vcpu))
1261 		goto out;
1262 
1263 	/*
1264 	 * Copy over the full size of the VMCB rather than just the size
1265 	 * of the structs.
1266 	 */
1267 	if (clear_user(user_vmcb, KVM_STATE_NESTED_SVM_VMCB_SIZE))
1268 		return -EFAULT;
1269 	if (copy_to_user(&user_vmcb->control, &svm->nested.ctl,
1270 			 sizeof(user_vmcb->control)))
1271 		return -EFAULT;
1272 	if (copy_to_user(&user_vmcb->save, &svm->vmcb01.ptr->save,
1273 			 sizeof(user_vmcb->save)))
1274 		return -EFAULT;
1275 out:
1276 	return kvm_state.size;
1277 }
1278 
1279 static int svm_set_nested_state(struct kvm_vcpu *vcpu,
1280 				struct kvm_nested_state __user *user_kvm_nested_state,
1281 				struct kvm_nested_state *kvm_state)
1282 {
1283 	struct vcpu_svm *svm = to_svm(vcpu);
1284 	struct vmcb __user *user_vmcb = (struct vmcb __user *)
1285 		&user_kvm_nested_state->data.svm[0];
1286 	struct vmcb_control_area *ctl;
1287 	struct vmcb_save_area *save;
1288 	unsigned long cr0;
1289 	int ret;
1290 
1291 	BUILD_BUG_ON(sizeof(struct vmcb_control_area) + sizeof(struct vmcb_save_area) >
1292 		     KVM_STATE_NESTED_SVM_VMCB_SIZE);
1293 
1294 	if (kvm_state->format != KVM_STATE_NESTED_FORMAT_SVM)
1295 		return -EINVAL;
1296 
1297 	if (kvm_state->flags & ~(KVM_STATE_NESTED_GUEST_MODE |
1298 				 KVM_STATE_NESTED_RUN_PENDING |
1299 				 KVM_STATE_NESTED_GIF_SET))
1300 		return -EINVAL;
1301 
1302 	/*
1303 	 * If in guest mode, vcpu->arch.efer actually refers to the L2 guest's
1304 	 * EFER.SVME, but EFER.SVME still has to be 1 for VMRUN to succeed.
1305 	 */
1306 	if (!(vcpu->arch.efer & EFER_SVME)) {
1307 		/* GIF=1 and no guest mode are required if SVME=0.  */
1308 		if (kvm_state->flags != KVM_STATE_NESTED_GIF_SET)
1309 			return -EINVAL;
1310 	}
1311 
1312 	/* SMM temporarily disables SVM, so we cannot be in guest mode.  */
1313 	if (is_smm(vcpu) && (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
1314 		return -EINVAL;
1315 
1316 	if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE)) {
1317 		svm_leave_nested(svm);
1318 		svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
1319 		return 0;
1320 	}
1321 
1322 	if (!page_address_valid(vcpu, kvm_state->hdr.svm.vmcb_pa))
1323 		return -EINVAL;
1324 	if (kvm_state->size < sizeof(*kvm_state) + KVM_STATE_NESTED_SVM_VMCB_SIZE)
1325 		return -EINVAL;
1326 
1327 	ret  = -ENOMEM;
1328 	ctl  = kzalloc(sizeof(*ctl),  GFP_KERNEL_ACCOUNT);
1329 	save = kzalloc(sizeof(*save), GFP_KERNEL_ACCOUNT);
1330 	if (!ctl || !save)
1331 		goto out_free;
1332 
1333 	ret = -EFAULT;
1334 	if (copy_from_user(ctl, &user_vmcb->control, sizeof(*ctl)))
1335 		goto out_free;
1336 	if (copy_from_user(save, &user_vmcb->save, sizeof(*save)))
1337 		goto out_free;
1338 
1339 	ret = -EINVAL;
1340 	if (!nested_vmcb_check_controls(vcpu, ctl))
1341 		goto out_free;
1342 
1343 	/*
1344 	 * Processor state contains L2 state.  Check that it is
1345 	 * valid for guest mode (see nested_vmcb_check_save).
1346 	 */
1347 	cr0 = kvm_read_cr0(vcpu);
1348         if (((cr0 & X86_CR0_CD) == 0) && (cr0 & X86_CR0_NW))
1349 		goto out_free;
1350 
1351 	/*
1352 	 * Validate host state saved from before VMRUN (see
1353 	 * nested_svm_check_permissions).
1354 	 */
1355 	if (!(save->cr0 & X86_CR0_PG) ||
1356 	    !(save->cr0 & X86_CR0_PE) ||
1357 	    (save->rflags & X86_EFLAGS_VM) ||
1358 	    !nested_vmcb_valid_sregs(vcpu, save))
1359 		goto out_free;
1360 
1361 	/*
1362 	 * While the nested guest CR3 is already checked and set by
1363 	 * KVM_SET_SREGS, it was set when nested state was yet loaded,
1364 	 * thus MMU might not be initialized correctly.
1365 	 * Set it again to fix this.
1366 	 */
1367 
1368 	ret = nested_svm_load_cr3(&svm->vcpu, vcpu->arch.cr3,
1369 				  nested_npt_enabled(svm), false);
1370 	if (WARN_ON_ONCE(ret))
1371 		goto out_free;
1372 
1373 
1374 	/*
1375 	 * All checks done, we can enter guest mode. Userspace provides
1376 	 * vmcb12.control, which will be combined with L1 and stored into
1377 	 * vmcb02, and the L1 save state which we store in vmcb01.
1378 	 * L2 registers if needed are moved from the current VMCB to VMCB02.
1379 	 */
1380 
1381 	if (is_guest_mode(vcpu))
1382 		svm_leave_nested(svm);
1383 	else
1384 		svm->nested.vmcb02.ptr->save = svm->vmcb01.ptr->save;
1385 
1386 	svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
1387 
1388 	svm->nested.nested_run_pending =
1389 		!!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
1390 
1391 	svm->nested.vmcb12_gpa = kvm_state->hdr.svm.vmcb_pa;
1392 
1393 	svm_copy_vmrun_state(&svm->vmcb01.ptr->save, save);
1394 	nested_load_control_from_vmcb12(svm, ctl);
1395 
1396 	svm_switch_vmcb(svm, &svm->nested.vmcb02);
1397 	nested_vmcb02_prepare_control(svm);
1398 	kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
1399 	ret = 0;
1400 out_free:
1401 	kfree(save);
1402 	kfree(ctl);
1403 
1404 	return ret;
1405 }
1406 
1407 static bool svm_get_nested_state_pages(struct kvm_vcpu *vcpu)
1408 {
1409 	struct vcpu_svm *svm = to_svm(vcpu);
1410 
1411 	if (WARN_ON(!is_guest_mode(vcpu)))
1412 		return true;
1413 
1414 	if (!vcpu->arch.pdptrs_from_userspace &&
1415 	    !nested_npt_enabled(svm) && is_pae_paging(vcpu))
1416 		/*
1417 		 * Reload the guest's PDPTRs since after a migration
1418 		 * the guest CR3 might be restored prior to setting the nested
1419 		 * state which can lead to a load of wrong PDPTRs.
1420 		 */
1421 		if (CC(!load_pdptrs(vcpu, vcpu->arch.walk_mmu, vcpu->arch.cr3)))
1422 			return false;
1423 
1424 	if (!nested_svm_vmrun_msrpm(svm)) {
1425 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1426 		vcpu->run->internal.suberror =
1427 			KVM_INTERNAL_ERROR_EMULATION;
1428 		vcpu->run->internal.ndata = 0;
1429 		return false;
1430 	}
1431 
1432 	return true;
1433 }
1434 
1435 struct kvm_x86_nested_ops svm_nested_ops = {
1436 	.check_events = svm_check_nested_events,
1437 	.triple_fault = nested_svm_triple_fault,
1438 	.get_nested_state_pages = svm_get_nested_state_pages,
1439 	.get_state = svm_get_nested_state,
1440 	.set_state = svm_set_nested_state,
1441 };
1442