xref: /openbmc/linux/arch/x86/kvm/svm/nested.c (revision c1cf3d89)
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 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
33 				       struct x86_exception *fault)
34 {
35 	struct vcpu_svm *svm = to_svm(vcpu);
36 
37 	if (svm->vmcb->control.exit_code != SVM_EXIT_NPF) {
38 		/*
39 		 * TODO: track the cause of the nested page fault, and
40 		 * correctly fill in the high bits of exit_info_1.
41 		 */
42 		svm->vmcb->control.exit_code = SVM_EXIT_NPF;
43 		svm->vmcb->control.exit_code_hi = 0;
44 		svm->vmcb->control.exit_info_1 = (1ULL << 32);
45 		svm->vmcb->control.exit_info_2 = fault->address;
46 	}
47 
48 	svm->vmcb->control.exit_info_1 &= ~0xffffffffULL;
49 	svm->vmcb->control.exit_info_1 |= fault->error_code;
50 
51 	nested_svm_vmexit(svm);
52 }
53 
54 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
55 {
56 	struct vcpu_svm *svm = to_svm(vcpu);
57 	u64 cr3 = svm->nested.ctl.nested_cr3;
58 	u64 pdpte;
59 	int ret;
60 
61 	ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(__sme_clr(cr3)), &pdpte,
62 				       offset_in_page(cr3) + index * 8, 8);
63 	if (ret)
64 		return 0;
65 	return pdpte;
66 }
67 
68 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
69 {
70 	struct vcpu_svm *svm = to_svm(vcpu);
71 
72 	return svm->nested.ctl.nested_cr3;
73 }
74 
75 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
76 {
77 	struct vcpu_svm *svm = to_svm(vcpu);
78 	struct vmcb *hsave = svm->nested.hsave;
79 
80 	WARN_ON(mmu_is_nested(vcpu));
81 
82 	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
83 	kvm_init_shadow_npt_mmu(vcpu, X86_CR0_PG, hsave->save.cr4, hsave->save.efer,
84 				svm->nested.ctl.nested_cr3);
85 	vcpu->arch.mmu->get_guest_pgd     = nested_svm_get_tdp_cr3;
86 	vcpu->arch.mmu->get_pdptr         = nested_svm_get_tdp_pdptr;
87 	vcpu->arch.mmu->inject_page_fault = nested_svm_inject_npf_exit;
88 	reset_shadow_zero_bits_mask(vcpu, vcpu->arch.mmu);
89 	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
90 }
91 
92 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
93 {
94 	vcpu->arch.mmu = &vcpu->arch.root_mmu;
95 	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
96 }
97 
98 void recalc_intercepts(struct vcpu_svm *svm)
99 {
100 	struct vmcb_control_area *c, *h, *g;
101 	unsigned int i;
102 
103 	vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
104 
105 	if (!is_guest_mode(&svm->vcpu))
106 		return;
107 
108 	c = &svm->vmcb->control;
109 	h = &svm->nested.hsave->control;
110 	g = &svm->nested.ctl;
111 
112 	for (i = 0; i < MAX_INTERCEPT; i++)
113 		c->intercepts[i] = h->intercepts[i];
114 
115 	if (g->int_ctl & V_INTR_MASKING_MASK) {
116 		/* We only want the cr8 intercept bits of L1 */
117 		vmcb_clr_intercept(c, INTERCEPT_CR8_READ);
118 		vmcb_clr_intercept(c, INTERCEPT_CR8_WRITE);
119 
120 		/*
121 		 * Once running L2 with HF_VINTR_MASK, EFLAGS.IF does not
122 		 * affect any interrupt we may want to inject; therefore,
123 		 * interrupt window vmexits are irrelevant to L0.
124 		 */
125 		vmcb_clr_intercept(c, INTERCEPT_VINTR);
126 	}
127 
128 	/* We don't want to see VMMCALLs from a nested guest */
129 	vmcb_clr_intercept(c, INTERCEPT_VMMCALL);
130 
131 	for (i = 0; i < MAX_INTERCEPT; i++)
132 		c->intercepts[i] |= g->intercepts[i];
133 }
134 
135 static void copy_vmcb_control_area(struct vmcb_control_area *dst,
136 				   struct vmcb_control_area *from)
137 {
138 	unsigned int i;
139 
140 	for (i = 0; i < MAX_INTERCEPT; i++)
141 		dst->intercepts[i] = from->intercepts[i];
142 
143 	dst->iopm_base_pa         = from->iopm_base_pa;
144 	dst->msrpm_base_pa        = from->msrpm_base_pa;
145 	dst->tsc_offset           = from->tsc_offset;
146 	/* asid not copied, it is handled manually for svm->vmcb.  */
147 	dst->tlb_ctl              = from->tlb_ctl;
148 	dst->int_ctl              = from->int_ctl;
149 	dst->int_vector           = from->int_vector;
150 	dst->int_state            = from->int_state;
151 	dst->exit_code            = from->exit_code;
152 	dst->exit_code_hi         = from->exit_code_hi;
153 	dst->exit_info_1          = from->exit_info_1;
154 	dst->exit_info_2          = from->exit_info_2;
155 	dst->exit_int_info        = from->exit_int_info;
156 	dst->exit_int_info_err    = from->exit_int_info_err;
157 	dst->nested_ctl           = from->nested_ctl;
158 	dst->event_inj            = from->event_inj;
159 	dst->event_inj_err        = from->event_inj_err;
160 	dst->nested_cr3           = from->nested_cr3;
161 	dst->virt_ext              = from->virt_ext;
162 	dst->pause_filter_count   = from->pause_filter_count;
163 	dst->pause_filter_thresh  = from->pause_filter_thresh;
164 }
165 
166 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
167 {
168 	/*
169 	 * This function merges the msr permission bitmaps of kvm and the
170 	 * nested vmcb. It is optimized in that it only merges the parts where
171 	 * the kvm msr permission bitmap may contain zero bits
172 	 */
173 	int i;
174 
175 	if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
176 		return true;
177 
178 	for (i = 0; i < MSRPM_OFFSETS; i++) {
179 		u32 value, p;
180 		u64 offset;
181 
182 		if (msrpm_offsets[i] == 0xffffffff)
183 			break;
184 
185 		p      = msrpm_offsets[i];
186 		offset = svm->nested.ctl.msrpm_base_pa + (p * 4);
187 
188 		if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
189 			return false;
190 
191 		svm->nested.msrpm[p] = svm->msrpm[p] | value;
192 	}
193 
194 	svm->vmcb->control.msrpm_base_pa = __sme_set(__pa(svm->nested.msrpm));
195 
196 	return true;
197 }
198 
199 static bool svm_get_nested_state_pages(struct kvm_vcpu *vcpu)
200 {
201 	struct vcpu_svm *svm = to_svm(vcpu);
202 
203 	if (WARN_ON(!is_guest_mode(vcpu)))
204 		return true;
205 
206 	if (!nested_svm_vmrun_msrpm(svm)) {
207 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
208 		vcpu->run->internal.suberror =
209 			KVM_INTERNAL_ERROR_EMULATION;
210 		vcpu->run->internal.ndata = 0;
211 		return false;
212 	}
213 
214 	return true;
215 }
216 
217 static bool nested_vmcb_check_controls(struct vmcb_control_area *control)
218 {
219 	if ((vmcb_is_intercept(control, INTERCEPT_VMRUN)) == 0)
220 		return false;
221 
222 	if (control->asid == 0)
223 		return false;
224 
225 	if ((control->nested_ctl & SVM_NESTED_CTL_NP_ENABLE) &&
226 	    !npt_enabled)
227 		return false;
228 
229 	return true;
230 }
231 
232 static bool nested_vmcb_checks(struct vcpu_svm *svm, struct vmcb *vmcb12)
233 {
234 	bool vmcb12_lma;
235 
236 	if ((vmcb12->save.efer & EFER_SVME) == 0)
237 		return false;
238 
239 	if (((vmcb12->save.cr0 & X86_CR0_CD) == 0) && (vmcb12->save.cr0 & X86_CR0_NW))
240 		return false;
241 
242 	if (!kvm_dr6_valid(vmcb12->save.dr6) || !kvm_dr7_valid(vmcb12->save.dr7))
243 		return false;
244 
245 	vmcb12_lma = (vmcb12->save.efer & EFER_LME) && (vmcb12->save.cr0 & X86_CR0_PG);
246 
247 	if (!vmcb12_lma) {
248 		if (vmcb12->save.cr4 & X86_CR4_PAE) {
249 			if (vmcb12->save.cr3 & MSR_CR3_LEGACY_PAE_RESERVED_MASK)
250 				return false;
251 		} else {
252 			if (vmcb12->save.cr3 & MSR_CR3_LEGACY_RESERVED_MASK)
253 				return false;
254 		}
255 	} else {
256 		if (!(vmcb12->save.cr4 & X86_CR4_PAE) ||
257 		    !(vmcb12->save.cr0 & X86_CR0_PE) ||
258 		    (vmcb12->save.cr3 & MSR_CR3_LONG_MBZ_MASK))
259 			return false;
260 	}
261 	if (!kvm_is_valid_cr4(&svm->vcpu, vmcb12->save.cr4))
262 		return false;
263 
264 	return nested_vmcb_check_controls(&vmcb12->control);
265 }
266 
267 static void load_nested_vmcb_control(struct vcpu_svm *svm,
268 				     struct vmcb_control_area *control)
269 {
270 	copy_vmcb_control_area(&svm->nested.ctl, control);
271 
272 	/* Copy it here because nested_svm_check_controls will check it.  */
273 	svm->nested.ctl.asid           = control->asid;
274 	svm->nested.ctl.msrpm_base_pa &= ~0x0fffULL;
275 	svm->nested.ctl.iopm_base_pa  &= ~0x0fffULL;
276 }
277 
278 /*
279  * Synchronize fields that are written by the processor, so that
280  * they can be copied back into the nested_vmcb.
281  */
282 void sync_nested_vmcb_control(struct vcpu_svm *svm)
283 {
284 	u32 mask;
285 	svm->nested.ctl.event_inj      = svm->vmcb->control.event_inj;
286 	svm->nested.ctl.event_inj_err  = svm->vmcb->control.event_inj_err;
287 
288 	/* Only a few fields of int_ctl are written by the processor.  */
289 	mask = V_IRQ_MASK | V_TPR_MASK;
290 	if (!(svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK) &&
291 	    svm_is_intercept(svm, INTERCEPT_VINTR)) {
292 		/*
293 		 * In order to request an interrupt window, L0 is usurping
294 		 * svm->vmcb->control.int_ctl and possibly setting V_IRQ
295 		 * even if it was clear in L1's VMCB.  Restoring it would be
296 		 * wrong.  However, in this case V_IRQ will remain true until
297 		 * interrupt_window_interception calls svm_clear_vintr and
298 		 * restores int_ctl.  We can just leave it aside.
299 		 */
300 		mask &= ~V_IRQ_MASK;
301 	}
302 	svm->nested.ctl.int_ctl        &= ~mask;
303 	svm->nested.ctl.int_ctl        |= svm->vmcb->control.int_ctl & mask;
304 }
305 
306 /*
307  * Transfer any event that L0 or L1 wanted to inject into L2 to
308  * EXIT_INT_INFO.
309  */
310 static void nested_vmcb_save_pending_event(struct vcpu_svm *svm,
311 					   struct vmcb *vmcb12)
312 {
313 	struct kvm_vcpu *vcpu = &svm->vcpu;
314 	u32 exit_int_info = 0;
315 	unsigned int nr;
316 
317 	if (vcpu->arch.exception.injected) {
318 		nr = vcpu->arch.exception.nr;
319 		exit_int_info = nr | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT;
320 
321 		if (vcpu->arch.exception.has_error_code) {
322 			exit_int_info |= SVM_EVTINJ_VALID_ERR;
323 			vmcb12->control.exit_int_info_err =
324 				vcpu->arch.exception.error_code;
325 		}
326 
327 	} else if (vcpu->arch.nmi_injected) {
328 		exit_int_info = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
329 
330 	} else if (vcpu->arch.interrupt.injected) {
331 		nr = vcpu->arch.interrupt.nr;
332 		exit_int_info = nr | SVM_EVTINJ_VALID;
333 
334 		if (vcpu->arch.interrupt.soft)
335 			exit_int_info |= SVM_EVTINJ_TYPE_SOFT;
336 		else
337 			exit_int_info |= SVM_EVTINJ_TYPE_INTR;
338 	}
339 
340 	vmcb12->control.exit_int_info = exit_int_info;
341 }
342 
343 static inline bool nested_npt_enabled(struct vcpu_svm *svm)
344 {
345 	return svm->nested.ctl.nested_ctl & SVM_NESTED_CTL_NP_ENABLE;
346 }
347 
348 /*
349  * Load guest's/host's cr3 on nested vmentry or vmexit. @nested_npt is true
350  * if we are emulating VM-Entry into a guest with NPT enabled.
351  */
352 static int nested_svm_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3,
353 			       bool nested_npt)
354 {
355 	if (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63))
356 		return -EINVAL;
357 
358 	if (!nested_npt && is_pae_paging(vcpu) &&
359 	    (cr3 != kvm_read_cr3(vcpu) || pdptrs_changed(vcpu))) {
360 		if (!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
361 			return -EINVAL;
362 	}
363 
364 	/*
365 	 * TODO: optimize unconditional TLB flush/MMU sync here and in
366 	 * kvm_init_shadow_npt_mmu().
367 	 */
368 	if (!nested_npt)
369 		kvm_mmu_new_pgd(vcpu, cr3, false, false);
370 
371 	vcpu->arch.cr3 = cr3;
372 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
373 
374 	kvm_init_mmu(vcpu, false);
375 
376 	return 0;
377 }
378 
379 static void nested_prepare_vmcb_save(struct vcpu_svm *svm, struct vmcb *vmcb12)
380 {
381 	/* Load the nested guest state */
382 	svm->vmcb->save.es = vmcb12->save.es;
383 	svm->vmcb->save.cs = vmcb12->save.cs;
384 	svm->vmcb->save.ss = vmcb12->save.ss;
385 	svm->vmcb->save.ds = vmcb12->save.ds;
386 	svm->vmcb->save.gdtr = vmcb12->save.gdtr;
387 	svm->vmcb->save.idtr = vmcb12->save.idtr;
388 	kvm_set_rflags(&svm->vcpu, vmcb12->save.rflags | X86_EFLAGS_FIXED);
389 	svm_set_efer(&svm->vcpu, vmcb12->save.efer);
390 	svm_set_cr0(&svm->vcpu, vmcb12->save.cr0);
391 	svm_set_cr4(&svm->vcpu, vmcb12->save.cr4);
392 	svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = vmcb12->save.cr2;
393 	kvm_rax_write(&svm->vcpu, vmcb12->save.rax);
394 	kvm_rsp_write(&svm->vcpu, vmcb12->save.rsp);
395 	kvm_rip_write(&svm->vcpu, vmcb12->save.rip);
396 
397 	/* In case we don't even reach vcpu_run, the fields are not updated */
398 	svm->vmcb->save.rax = vmcb12->save.rax;
399 	svm->vmcb->save.rsp = vmcb12->save.rsp;
400 	svm->vmcb->save.rip = vmcb12->save.rip;
401 	svm->vmcb->save.dr7 = vmcb12->save.dr7 | DR7_FIXED_1;
402 	svm->vcpu.arch.dr6  = vmcb12->save.dr6 | DR6_FIXED_1 | DR6_RTM;
403 	svm->vmcb->save.cpl = vmcb12->save.cpl;
404 }
405 
406 static void nested_prepare_vmcb_control(struct vcpu_svm *svm)
407 {
408 	const u32 mask = V_INTR_MASKING_MASK | V_GIF_ENABLE_MASK | V_GIF_MASK;
409 
410 	if (nested_npt_enabled(svm))
411 		nested_svm_init_mmu_context(&svm->vcpu);
412 
413 	svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset =
414 		svm->vcpu.arch.l1_tsc_offset + svm->nested.ctl.tsc_offset;
415 
416 	svm->vmcb->control.int_ctl             =
417 		(svm->nested.ctl.int_ctl & ~mask) |
418 		(svm->nested.hsave->control.int_ctl & mask);
419 
420 	svm->vmcb->control.virt_ext            = svm->nested.ctl.virt_ext;
421 	svm->vmcb->control.int_vector          = svm->nested.ctl.int_vector;
422 	svm->vmcb->control.int_state           = svm->nested.ctl.int_state;
423 	svm->vmcb->control.event_inj           = svm->nested.ctl.event_inj;
424 	svm->vmcb->control.event_inj_err       = svm->nested.ctl.event_inj_err;
425 
426 	svm->vmcb->control.pause_filter_count  = svm->nested.ctl.pause_filter_count;
427 	svm->vmcb->control.pause_filter_thresh = svm->nested.ctl.pause_filter_thresh;
428 
429 	/* Enter Guest-Mode */
430 	enter_guest_mode(&svm->vcpu);
431 
432 	/*
433 	 * Merge guest and host intercepts - must be called  with vcpu in
434 	 * guest-mode to take affect here
435 	 */
436 	recalc_intercepts(svm);
437 
438 	vmcb_mark_all_dirty(svm->vmcb);
439 }
440 
441 int enter_svm_guest_mode(struct vcpu_svm *svm, u64 vmcb12_gpa,
442 			 struct vmcb *vmcb12)
443 {
444 	int ret;
445 
446 	svm->nested.vmcb12_gpa = vmcb12_gpa;
447 	load_nested_vmcb_control(svm, &vmcb12->control);
448 	nested_prepare_vmcb_save(svm, vmcb12);
449 	nested_prepare_vmcb_control(svm);
450 
451 	ret = nested_svm_load_cr3(&svm->vcpu, vmcb12->save.cr3,
452 				  nested_npt_enabled(svm));
453 	if (ret)
454 		return ret;
455 
456 	svm_set_gif(svm, true);
457 
458 	return 0;
459 }
460 
461 int nested_svm_vmrun(struct vcpu_svm *svm)
462 {
463 	int ret;
464 	struct vmcb *vmcb12;
465 	struct vmcb *hsave = svm->nested.hsave;
466 	struct vmcb *vmcb = svm->vmcb;
467 	struct kvm_host_map map;
468 	u64 vmcb12_gpa;
469 
470 	if (is_smm(&svm->vcpu)) {
471 		kvm_queue_exception(&svm->vcpu, UD_VECTOR);
472 		return 1;
473 	}
474 
475 	vmcb12_gpa = svm->vmcb->save.rax;
476 	ret = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(vmcb12_gpa), &map);
477 	if (ret == -EINVAL) {
478 		kvm_inject_gp(&svm->vcpu, 0);
479 		return 1;
480 	} else if (ret) {
481 		return kvm_skip_emulated_instruction(&svm->vcpu);
482 	}
483 
484 	ret = kvm_skip_emulated_instruction(&svm->vcpu);
485 
486 	vmcb12 = map.hva;
487 
488 	if (WARN_ON_ONCE(!svm->nested.initialized))
489 		return -EINVAL;
490 
491 	if (!nested_vmcb_checks(svm, vmcb12)) {
492 		vmcb12->control.exit_code    = SVM_EXIT_ERR;
493 		vmcb12->control.exit_code_hi = 0;
494 		vmcb12->control.exit_info_1  = 0;
495 		vmcb12->control.exit_info_2  = 0;
496 		goto out;
497 	}
498 
499 	trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb12_gpa,
500 			       vmcb12->save.rip,
501 			       vmcb12->control.int_ctl,
502 			       vmcb12->control.event_inj,
503 			       vmcb12->control.nested_ctl);
504 
505 	trace_kvm_nested_intercepts(vmcb12->control.intercepts[INTERCEPT_CR] & 0xffff,
506 				    vmcb12->control.intercepts[INTERCEPT_CR] >> 16,
507 				    vmcb12->control.intercepts[INTERCEPT_EXCEPTION],
508 				    vmcb12->control.intercepts[INTERCEPT_WORD3],
509 				    vmcb12->control.intercepts[INTERCEPT_WORD4],
510 				    vmcb12->control.intercepts[INTERCEPT_WORD5]);
511 
512 	/* Clear internal status */
513 	kvm_clear_exception_queue(&svm->vcpu);
514 	kvm_clear_interrupt_queue(&svm->vcpu);
515 
516 	/*
517 	 * Save the old vmcb, so we don't need to pick what we save, but can
518 	 * restore everything when a VMEXIT occurs
519 	 */
520 	hsave->save.es     = vmcb->save.es;
521 	hsave->save.cs     = vmcb->save.cs;
522 	hsave->save.ss     = vmcb->save.ss;
523 	hsave->save.ds     = vmcb->save.ds;
524 	hsave->save.gdtr   = vmcb->save.gdtr;
525 	hsave->save.idtr   = vmcb->save.idtr;
526 	hsave->save.efer   = svm->vcpu.arch.efer;
527 	hsave->save.cr0    = kvm_read_cr0(&svm->vcpu);
528 	hsave->save.cr4    = svm->vcpu.arch.cr4;
529 	hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
530 	hsave->save.rip    = kvm_rip_read(&svm->vcpu);
531 	hsave->save.rsp    = vmcb->save.rsp;
532 	hsave->save.rax    = vmcb->save.rax;
533 	if (npt_enabled)
534 		hsave->save.cr3    = vmcb->save.cr3;
535 	else
536 		hsave->save.cr3    = kvm_read_cr3(&svm->vcpu);
537 
538 	copy_vmcb_control_area(&hsave->control, &vmcb->control);
539 
540 	svm->nested.nested_run_pending = 1;
541 
542 	if (enter_svm_guest_mode(svm, vmcb12_gpa, vmcb12))
543 		goto out_exit_err;
544 
545 	if (nested_svm_vmrun_msrpm(svm))
546 		goto out;
547 
548 out_exit_err:
549 	svm->nested.nested_run_pending = 0;
550 
551 	svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
552 	svm->vmcb->control.exit_code_hi = 0;
553 	svm->vmcb->control.exit_info_1  = 0;
554 	svm->vmcb->control.exit_info_2  = 0;
555 
556 	nested_svm_vmexit(svm);
557 
558 out:
559 	kvm_vcpu_unmap(&svm->vcpu, &map, true);
560 
561 	return ret;
562 }
563 
564 void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
565 {
566 	to_vmcb->save.fs = from_vmcb->save.fs;
567 	to_vmcb->save.gs = from_vmcb->save.gs;
568 	to_vmcb->save.tr = from_vmcb->save.tr;
569 	to_vmcb->save.ldtr = from_vmcb->save.ldtr;
570 	to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
571 	to_vmcb->save.star = from_vmcb->save.star;
572 	to_vmcb->save.lstar = from_vmcb->save.lstar;
573 	to_vmcb->save.cstar = from_vmcb->save.cstar;
574 	to_vmcb->save.sfmask = from_vmcb->save.sfmask;
575 	to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
576 	to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
577 	to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
578 }
579 
580 int nested_svm_vmexit(struct vcpu_svm *svm)
581 {
582 	int rc;
583 	struct vmcb *vmcb12;
584 	struct vmcb *hsave = svm->nested.hsave;
585 	struct vmcb *vmcb = svm->vmcb;
586 	struct kvm_host_map map;
587 
588 	rc = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(svm->nested.vmcb12_gpa), &map);
589 	if (rc) {
590 		if (rc == -EINVAL)
591 			kvm_inject_gp(&svm->vcpu, 0);
592 		return 1;
593 	}
594 
595 	vmcb12 = map.hva;
596 
597 	/* Exit Guest-Mode */
598 	leave_guest_mode(&svm->vcpu);
599 	svm->nested.vmcb12_gpa = 0;
600 	WARN_ON_ONCE(svm->nested.nested_run_pending);
601 
602 	kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, &svm->vcpu);
603 
604 	/* in case we halted in L2 */
605 	svm->vcpu.arch.mp_state = KVM_MP_STATE_RUNNABLE;
606 
607 	/* Give the current vmcb to the guest */
608 
609 	vmcb12->save.es     = vmcb->save.es;
610 	vmcb12->save.cs     = vmcb->save.cs;
611 	vmcb12->save.ss     = vmcb->save.ss;
612 	vmcb12->save.ds     = vmcb->save.ds;
613 	vmcb12->save.gdtr   = vmcb->save.gdtr;
614 	vmcb12->save.idtr   = vmcb->save.idtr;
615 	vmcb12->save.efer   = svm->vcpu.arch.efer;
616 	vmcb12->save.cr0    = kvm_read_cr0(&svm->vcpu);
617 	vmcb12->save.cr3    = kvm_read_cr3(&svm->vcpu);
618 	vmcb12->save.cr2    = vmcb->save.cr2;
619 	vmcb12->save.cr4    = svm->vcpu.arch.cr4;
620 	vmcb12->save.rflags = kvm_get_rflags(&svm->vcpu);
621 	vmcb12->save.rip    = kvm_rip_read(&svm->vcpu);
622 	vmcb12->save.rsp    = kvm_rsp_read(&svm->vcpu);
623 	vmcb12->save.rax    = kvm_rax_read(&svm->vcpu);
624 	vmcb12->save.dr7    = vmcb->save.dr7;
625 	vmcb12->save.dr6    = svm->vcpu.arch.dr6;
626 	vmcb12->save.cpl    = vmcb->save.cpl;
627 
628 	vmcb12->control.int_state         = vmcb->control.int_state;
629 	vmcb12->control.exit_code         = vmcb->control.exit_code;
630 	vmcb12->control.exit_code_hi      = vmcb->control.exit_code_hi;
631 	vmcb12->control.exit_info_1       = vmcb->control.exit_info_1;
632 	vmcb12->control.exit_info_2       = vmcb->control.exit_info_2;
633 
634 	if (vmcb12->control.exit_code != SVM_EXIT_ERR)
635 		nested_vmcb_save_pending_event(svm, vmcb12);
636 
637 	if (svm->nrips_enabled)
638 		vmcb12->control.next_rip  = vmcb->control.next_rip;
639 
640 	vmcb12->control.int_ctl           = svm->nested.ctl.int_ctl;
641 	vmcb12->control.tlb_ctl           = svm->nested.ctl.tlb_ctl;
642 	vmcb12->control.event_inj         = svm->nested.ctl.event_inj;
643 	vmcb12->control.event_inj_err     = svm->nested.ctl.event_inj_err;
644 
645 	vmcb12->control.pause_filter_count =
646 		svm->vmcb->control.pause_filter_count;
647 	vmcb12->control.pause_filter_thresh =
648 		svm->vmcb->control.pause_filter_thresh;
649 
650 	/* Restore the original control entries */
651 	copy_vmcb_control_area(&vmcb->control, &hsave->control);
652 
653 	/* On vmexit the  GIF is set to false */
654 	svm_set_gif(svm, false);
655 
656 	svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset =
657 		svm->vcpu.arch.l1_tsc_offset;
658 
659 	svm->nested.ctl.nested_cr3 = 0;
660 
661 	/* Restore selected save entries */
662 	svm->vmcb->save.es = hsave->save.es;
663 	svm->vmcb->save.cs = hsave->save.cs;
664 	svm->vmcb->save.ss = hsave->save.ss;
665 	svm->vmcb->save.ds = hsave->save.ds;
666 	svm->vmcb->save.gdtr = hsave->save.gdtr;
667 	svm->vmcb->save.idtr = hsave->save.idtr;
668 	kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
669 	kvm_set_rflags(&svm->vcpu, hsave->save.rflags | X86_EFLAGS_FIXED);
670 	svm_set_efer(&svm->vcpu, hsave->save.efer);
671 	svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
672 	svm_set_cr4(&svm->vcpu, hsave->save.cr4);
673 	kvm_rax_write(&svm->vcpu, hsave->save.rax);
674 	kvm_rsp_write(&svm->vcpu, hsave->save.rsp);
675 	kvm_rip_write(&svm->vcpu, hsave->save.rip);
676 	svm->vmcb->save.dr7 = DR7_FIXED_1;
677 	svm->vmcb->save.cpl = 0;
678 	svm->vmcb->control.exit_int_info = 0;
679 
680 	vmcb_mark_all_dirty(svm->vmcb);
681 
682 	trace_kvm_nested_vmexit_inject(vmcb12->control.exit_code,
683 				       vmcb12->control.exit_info_1,
684 				       vmcb12->control.exit_info_2,
685 				       vmcb12->control.exit_int_info,
686 				       vmcb12->control.exit_int_info_err,
687 				       KVM_ISA_SVM);
688 
689 	kvm_vcpu_unmap(&svm->vcpu, &map, true);
690 
691 	nested_svm_uninit_mmu_context(&svm->vcpu);
692 
693 	rc = nested_svm_load_cr3(&svm->vcpu, hsave->save.cr3, false);
694 	if (rc)
695 		return 1;
696 
697 	if (npt_enabled)
698 		svm->vmcb->save.cr3 = hsave->save.cr3;
699 
700 	/*
701 	 * Drop what we picked up for L2 via svm_complete_interrupts() so it
702 	 * doesn't end up in L1.
703 	 */
704 	svm->vcpu.arch.nmi_injected = false;
705 	kvm_clear_exception_queue(&svm->vcpu);
706 	kvm_clear_interrupt_queue(&svm->vcpu);
707 
708 	return 0;
709 }
710 
711 int svm_allocate_nested(struct vcpu_svm *svm)
712 {
713 	struct page *hsave_page;
714 
715 	if (svm->nested.initialized)
716 		return 0;
717 
718 	hsave_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
719 	if (!hsave_page)
720 		return -ENOMEM;
721 	svm->nested.hsave = page_address(hsave_page);
722 
723 	svm->nested.msrpm = svm_vcpu_alloc_msrpm();
724 	if (!svm->nested.msrpm)
725 		goto err_free_hsave;
726 	svm_vcpu_init_msrpm(&svm->vcpu, svm->nested.msrpm);
727 
728 	svm->nested.initialized = true;
729 	return 0;
730 
731 err_free_hsave:
732 	__free_page(hsave_page);
733 	return -ENOMEM;
734 }
735 
736 void svm_free_nested(struct vcpu_svm *svm)
737 {
738 	if (!svm->nested.initialized)
739 		return;
740 
741 	svm_vcpu_free_msrpm(svm->nested.msrpm);
742 	svm->nested.msrpm = NULL;
743 
744 	__free_page(virt_to_page(svm->nested.hsave));
745 	svm->nested.hsave = NULL;
746 
747 	svm->nested.initialized = false;
748 }
749 
750 /*
751  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
752  */
753 void svm_leave_nested(struct vcpu_svm *svm)
754 {
755 	if (is_guest_mode(&svm->vcpu)) {
756 		struct vmcb *hsave = svm->nested.hsave;
757 		struct vmcb *vmcb = svm->vmcb;
758 
759 		svm->nested.nested_run_pending = 0;
760 		leave_guest_mode(&svm->vcpu);
761 		copy_vmcb_control_area(&vmcb->control, &hsave->control);
762 		nested_svm_uninit_mmu_context(&svm->vcpu);
763 		vmcb_mark_all_dirty(svm->vmcb);
764 	}
765 
766 	kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, &svm->vcpu);
767 }
768 
769 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
770 {
771 	u32 offset, msr, value;
772 	int write, mask;
773 
774 	if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
775 		return NESTED_EXIT_HOST;
776 
777 	msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
778 	offset = svm_msrpm_offset(msr);
779 	write  = svm->vmcb->control.exit_info_1 & 1;
780 	mask   = 1 << ((2 * (msr & 0xf)) + write);
781 
782 	if (offset == MSR_INVALID)
783 		return NESTED_EXIT_DONE;
784 
785 	/* Offset is in 32 bit units but need in 8 bit units */
786 	offset *= 4;
787 
788 	if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.ctl.msrpm_base_pa + offset, &value, 4))
789 		return NESTED_EXIT_DONE;
790 
791 	return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
792 }
793 
794 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
795 {
796 	unsigned port, size, iopm_len;
797 	u16 val, mask;
798 	u8 start_bit;
799 	u64 gpa;
800 
801 	if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_IOIO_PROT)))
802 		return NESTED_EXIT_HOST;
803 
804 	port = svm->vmcb->control.exit_info_1 >> 16;
805 	size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
806 		SVM_IOIO_SIZE_SHIFT;
807 	gpa  = svm->nested.ctl.iopm_base_pa + (port / 8);
808 	start_bit = port % 8;
809 	iopm_len = (start_bit + size > 8) ? 2 : 1;
810 	mask = (0xf >> (4 - size)) << start_bit;
811 	val = 0;
812 
813 	if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
814 		return NESTED_EXIT_DONE;
815 
816 	return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
817 }
818 
819 static int nested_svm_intercept(struct vcpu_svm *svm)
820 {
821 	u32 exit_code = svm->vmcb->control.exit_code;
822 	int vmexit = NESTED_EXIT_HOST;
823 
824 	switch (exit_code) {
825 	case SVM_EXIT_MSR:
826 		vmexit = nested_svm_exit_handled_msr(svm);
827 		break;
828 	case SVM_EXIT_IOIO:
829 		vmexit = nested_svm_intercept_ioio(svm);
830 		break;
831 	case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
832 		if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
833 			vmexit = NESTED_EXIT_DONE;
834 		break;
835 	}
836 	case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
837 		if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
838 			vmexit = NESTED_EXIT_DONE;
839 		break;
840 	}
841 	case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
842 		/*
843 		 * Host-intercepted exceptions have been checked already in
844 		 * nested_svm_exit_special.  There is nothing to do here,
845 		 * the vmexit is injected by svm_check_nested_events.
846 		 */
847 		vmexit = NESTED_EXIT_DONE;
848 		break;
849 	}
850 	case SVM_EXIT_ERR: {
851 		vmexit = NESTED_EXIT_DONE;
852 		break;
853 	}
854 	default: {
855 		if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
856 			vmexit = NESTED_EXIT_DONE;
857 	}
858 	}
859 
860 	return vmexit;
861 }
862 
863 int nested_svm_exit_handled(struct vcpu_svm *svm)
864 {
865 	int vmexit;
866 
867 	vmexit = nested_svm_intercept(svm);
868 
869 	if (vmexit == NESTED_EXIT_DONE)
870 		nested_svm_vmexit(svm);
871 
872 	return vmexit;
873 }
874 
875 int nested_svm_check_permissions(struct vcpu_svm *svm)
876 {
877 	if (!(svm->vcpu.arch.efer & EFER_SVME) ||
878 	    !is_paging(&svm->vcpu)) {
879 		kvm_queue_exception(&svm->vcpu, UD_VECTOR);
880 		return 1;
881 	}
882 
883 	if (svm->vmcb->save.cpl) {
884 		kvm_inject_gp(&svm->vcpu, 0);
885 		return 1;
886 	}
887 
888 	return 0;
889 }
890 
891 static bool nested_exit_on_exception(struct vcpu_svm *svm)
892 {
893 	unsigned int nr = svm->vcpu.arch.exception.nr;
894 
895 	return (svm->nested.ctl.intercepts[INTERCEPT_EXCEPTION] & BIT(nr));
896 }
897 
898 static void nested_svm_inject_exception_vmexit(struct vcpu_svm *svm)
899 {
900 	unsigned int nr = svm->vcpu.arch.exception.nr;
901 
902 	svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
903 	svm->vmcb->control.exit_code_hi = 0;
904 
905 	if (svm->vcpu.arch.exception.has_error_code)
906 		svm->vmcb->control.exit_info_1 = svm->vcpu.arch.exception.error_code;
907 
908 	/*
909 	 * EXITINFO2 is undefined for all exception intercepts other
910 	 * than #PF.
911 	 */
912 	if (nr == PF_VECTOR) {
913 		if (svm->vcpu.arch.exception.nested_apf)
914 			svm->vmcb->control.exit_info_2 = svm->vcpu.arch.apf.nested_apf_token;
915 		else if (svm->vcpu.arch.exception.has_payload)
916 			svm->vmcb->control.exit_info_2 = svm->vcpu.arch.exception.payload;
917 		else
918 			svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
919 	} else if (nr == DB_VECTOR) {
920 		/* See inject_pending_event.  */
921 		kvm_deliver_exception_payload(&svm->vcpu);
922 		if (svm->vcpu.arch.dr7 & DR7_GD) {
923 			svm->vcpu.arch.dr7 &= ~DR7_GD;
924 			kvm_update_dr7(&svm->vcpu);
925 		}
926 	} else
927 		WARN_ON(svm->vcpu.arch.exception.has_payload);
928 
929 	nested_svm_vmexit(svm);
930 }
931 
932 static void nested_svm_smi(struct vcpu_svm *svm)
933 {
934 	svm->vmcb->control.exit_code = SVM_EXIT_SMI;
935 	svm->vmcb->control.exit_info_1 = 0;
936 	svm->vmcb->control.exit_info_2 = 0;
937 
938 	nested_svm_vmexit(svm);
939 }
940 
941 static void nested_svm_nmi(struct vcpu_svm *svm)
942 {
943 	svm->vmcb->control.exit_code = SVM_EXIT_NMI;
944 	svm->vmcb->control.exit_info_1 = 0;
945 	svm->vmcb->control.exit_info_2 = 0;
946 
947 	nested_svm_vmexit(svm);
948 }
949 
950 static void nested_svm_intr(struct vcpu_svm *svm)
951 {
952 	trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
953 
954 	svm->vmcb->control.exit_code   = SVM_EXIT_INTR;
955 	svm->vmcb->control.exit_info_1 = 0;
956 	svm->vmcb->control.exit_info_2 = 0;
957 
958 	nested_svm_vmexit(svm);
959 }
960 
961 static inline bool nested_exit_on_init(struct vcpu_svm *svm)
962 {
963 	return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_INIT);
964 }
965 
966 static void nested_svm_init(struct vcpu_svm *svm)
967 {
968 	svm->vmcb->control.exit_code   = SVM_EXIT_INIT;
969 	svm->vmcb->control.exit_info_1 = 0;
970 	svm->vmcb->control.exit_info_2 = 0;
971 
972 	nested_svm_vmexit(svm);
973 }
974 
975 
976 static int svm_check_nested_events(struct kvm_vcpu *vcpu)
977 {
978 	struct vcpu_svm *svm = to_svm(vcpu);
979 	bool block_nested_events =
980 		kvm_event_needs_reinjection(vcpu) || svm->nested.nested_run_pending;
981 	struct kvm_lapic *apic = vcpu->arch.apic;
982 
983 	if (lapic_in_kernel(vcpu) &&
984 	    test_bit(KVM_APIC_INIT, &apic->pending_events)) {
985 		if (block_nested_events)
986 			return -EBUSY;
987 		if (!nested_exit_on_init(svm))
988 			return 0;
989 		nested_svm_init(svm);
990 		return 0;
991 	}
992 
993 	if (vcpu->arch.exception.pending) {
994 		if (block_nested_events)
995                         return -EBUSY;
996 		if (!nested_exit_on_exception(svm))
997 			return 0;
998 		nested_svm_inject_exception_vmexit(svm);
999 		return 0;
1000 	}
1001 
1002 	if (vcpu->arch.smi_pending && !svm_smi_blocked(vcpu)) {
1003 		if (block_nested_events)
1004 			return -EBUSY;
1005 		if (!nested_exit_on_smi(svm))
1006 			return 0;
1007 		nested_svm_smi(svm);
1008 		return 0;
1009 	}
1010 
1011 	if (vcpu->arch.nmi_pending && !svm_nmi_blocked(vcpu)) {
1012 		if (block_nested_events)
1013 			return -EBUSY;
1014 		if (!nested_exit_on_nmi(svm))
1015 			return 0;
1016 		nested_svm_nmi(svm);
1017 		return 0;
1018 	}
1019 
1020 	if (kvm_cpu_has_interrupt(vcpu) && !svm_interrupt_blocked(vcpu)) {
1021 		if (block_nested_events)
1022 			return -EBUSY;
1023 		if (!nested_exit_on_intr(svm))
1024 			return 0;
1025 		nested_svm_intr(svm);
1026 		return 0;
1027 	}
1028 
1029 	return 0;
1030 }
1031 
1032 int nested_svm_exit_special(struct vcpu_svm *svm)
1033 {
1034 	u32 exit_code = svm->vmcb->control.exit_code;
1035 
1036 	switch (exit_code) {
1037 	case SVM_EXIT_INTR:
1038 	case SVM_EXIT_NMI:
1039 	case SVM_EXIT_NPF:
1040 		return NESTED_EXIT_HOST;
1041 	case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1042 		u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
1043 
1044 		if (get_host_vmcb(svm)->control.intercepts[INTERCEPT_EXCEPTION] &
1045 				excp_bits)
1046 			return NESTED_EXIT_HOST;
1047 		else if (exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR &&
1048 			 svm->vcpu.arch.apf.host_apf_flags)
1049 			/* Trap async PF even if not shadowing */
1050 			return NESTED_EXIT_HOST;
1051 		break;
1052 	}
1053 	default:
1054 		break;
1055 	}
1056 
1057 	return NESTED_EXIT_CONTINUE;
1058 }
1059 
1060 static int svm_get_nested_state(struct kvm_vcpu *vcpu,
1061 				struct kvm_nested_state __user *user_kvm_nested_state,
1062 				u32 user_data_size)
1063 {
1064 	struct vcpu_svm *svm;
1065 	struct kvm_nested_state kvm_state = {
1066 		.flags = 0,
1067 		.format = KVM_STATE_NESTED_FORMAT_SVM,
1068 		.size = sizeof(kvm_state),
1069 	};
1070 	struct vmcb __user *user_vmcb = (struct vmcb __user *)
1071 		&user_kvm_nested_state->data.svm[0];
1072 
1073 	if (!vcpu)
1074 		return kvm_state.size + KVM_STATE_NESTED_SVM_VMCB_SIZE;
1075 
1076 	svm = to_svm(vcpu);
1077 
1078 	if (user_data_size < kvm_state.size)
1079 		goto out;
1080 
1081 	/* First fill in the header and copy it out.  */
1082 	if (is_guest_mode(vcpu)) {
1083 		kvm_state.hdr.svm.vmcb_pa = svm->nested.vmcb12_gpa;
1084 		kvm_state.size += KVM_STATE_NESTED_SVM_VMCB_SIZE;
1085 		kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
1086 
1087 		if (svm->nested.nested_run_pending)
1088 			kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
1089 	}
1090 
1091 	if (gif_set(svm))
1092 		kvm_state.flags |= KVM_STATE_NESTED_GIF_SET;
1093 
1094 	if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
1095 		return -EFAULT;
1096 
1097 	if (!is_guest_mode(vcpu))
1098 		goto out;
1099 
1100 	/*
1101 	 * Copy over the full size of the VMCB rather than just the size
1102 	 * of the structs.
1103 	 */
1104 	if (clear_user(user_vmcb, KVM_STATE_NESTED_SVM_VMCB_SIZE))
1105 		return -EFAULT;
1106 	if (copy_to_user(&user_vmcb->control, &svm->nested.ctl,
1107 			 sizeof(user_vmcb->control)))
1108 		return -EFAULT;
1109 	if (copy_to_user(&user_vmcb->save, &svm->nested.hsave->save,
1110 			 sizeof(user_vmcb->save)))
1111 		return -EFAULT;
1112 
1113 out:
1114 	return kvm_state.size;
1115 }
1116 
1117 static int svm_set_nested_state(struct kvm_vcpu *vcpu,
1118 				struct kvm_nested_state __user *user_kvm_nested_state,
1119 				struct kvm_nested_state *kvm_state)
1120 {
1121 	struct vcpu_svm *svm = to_svm(vcpu);
1122 	struct vmcb *hsave = svm->nested.hsave;
1123 	struct vmcb __user *user_vmcb = (struct vmcb __user *)
1124 		&user_kvm_nested_state->data.svm[0];
1125 	struct vmcb_control_area *ctl;
1126 	struct vmcb_save_area *save;
1127 	int ret;
1128 	u32 cr0;
1129 
1130 	BUILD_BUG_ON(sizeof(struct vmcb_control_area) + sizeof(struct vmcb_save_area) >
1131 		     KVM_STATE_NESTED_SVM_VMCB_SIZE);
1132 
1133 	if (kvm_state->format != KVM_STATE_NESTED_FORMAT_SVM)
1134 		return -EINVAL;
1135 
1136 	if (kvm_state->flags & ~(KVM_STATE_NESTED_GUEST_MODE |
1137 				 KVM_STATE_NESTED_RUN_PENDING |
1138 				 KVM_STATE_NESTED_GIF_SET))
1139 		return -EINVAL;
1140 
1141 	/*
1142 	 * If in guest mode, vcpu->arch.efer actually refers to the L2 guest's
1143 	 * EFER.SVME, but EFER.SVME still has to be 1 for VMRUN to succeed.
1144 	 */
1145 	if (!(vcpu->arch.efer & EFER_SVME)) {
1146 		/* GIF=1 and no guest mode are required if SVME=0.  */
1147 		if (kvm_state->flags != KVM_STATE_NESTED_GIF_SET)
1148 			return -EINVAL;
1149 	}
1150 
1151 	/* SMM temporarily disables SVM, so we cannot be in guest mode.  */
1152 	if (is_smm(vcpu) && (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
1153 		return -EINVAL;
1154 
1155 	if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE)) {
1156 		svm_leave_nested(svm);
1157 		svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
1158 		return 0;
1159 	}
1160 
1161 	if (!page_address_valid(vcpu, kvm_state->hdr.svm.vmcb_pa))
1162 		return -EINVAL;
1163 	if (kvm_state->size < sizeof(*kvm_state) + KVM_STATE_NESTED_SVM_VMCB_SIZE)
1164 		return -EINVAL;
1165 
1166 	ret  = -ENOMEM;
1167 	ctl  = kzalloc(sizeof(*ctl),  GFP_KERNEL);
1168 	save = kzalloc(sizeof(*save), GFP_KERNEL);
1169 	if (!ctl || !save)
1170 		goto out_free;
1171 
1172 	ret = -EFAULT;
1173 	if (copy_from_user(ctl, &user_vmcb->control, sizeof(*ctl)))
1174 		goto out_free;
1175 	if (copy_from_user(save, &user_vmcb->save, sizeof(*save)))
1176 		goto out_free;
1177 
1178 	ret = -EINVAL;
1179 	if (!nested_vmcb_check_controls(ctl))
1180 		goto out_free;
1181 
1182 	/*
1183 	 * Processor state contains L2 state.  Check that it is
1184 	 * valid for guest mode (see nested_vmcb_checks).
1185 	 */
1186 	cr0 = kvm_read_cr0(vcpu);
1187         if (((cr0 & X86_CR0_CD) == 0) && (cr0 & X86_CR0_NW))
1188 		goto out_free;
1189 
1190 	/*
1191 	 * Validate host state saved from before VMRUN (see
1192 	 * nested_svm_check_permissions).
1193 	 * TODO: validate reserved bits for all saved state.
1194 	 */
1195 	if (!(save->cr0 & X86_CR0_PG))
1196 		goto out_free;
1197 
1198 	/*
1199 	 * All checks done, we can enter guest mode.  L1 control fields
1200 	 * come from the nested save state.  Guest state is already
1201 	 * in the registers, the save area of the nested state instead
1202 	 * contains saved L1 state.
1203 	 */
1204 
1205 	svm->nested.nested_run_pending =
1206 		!!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
1207 
1208 	copy_vmcb_control_area(&hsave->control, &svm->vmcb->control);
1209 	hsave->save = *save;
1210 
1211 	svm->nested.vmcb12_gpa = kvm_state->hdr.svm.vmcb_pa;
1212 	load_nested_vmcb_control(svm, ctl);
1213 	nested_prepare_vmcb_control(svm);
1214 
1215 	kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
1216 	ret = 0;
1217 out_free:
1218 	kfree(save);
1219 	kfree(ctl);
1220 
1221 	return ret;
1222 }
1223 
1224 struct kvm_x86_nested_ops svm_nested_ops = {
1225 	.check_events = svm_check_nested_events,
1226 	.get_nested_state_pages = svm_get_nested_state_pages,
1227 	.get_state = svm_get_nested_state,
1228 	.set_state = svm_set_nested_state,
1229 };
1230