xref: /openbmc/linux/arch/x86/kvm/svm/nested.c (revision 0bc92e7f)
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 	if (!nested_svm_vmrun_msrpm(svm)) {
203 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
204 		vcpu->run->internal.suberror =
205 			KVM_INTERNAL_ERROR_EMULATION;
206 		vcpu->run->internal.ndata = 0;
207 		return false;
208 	}
209 
210 	return true;
211 }
212 
213 static bool nested_vmcb_check_controls(struct vmcb_control_area *control)
214 {
215 	if ((vmcb_is_intercept(control, INTERCEPT_VMRUN)) == 0)
216 		return false;
217 
218 	if (control->asid == 0)
219 		return false;
220 
221 	if ((control->nested_ctl & SVM_NESTED_CTL_NP_ENABLE) &&
222 	    !npt_enabled)
223 		return false;
224 
225 	return true;
226 }
227 
228 static bool nested_vmcb_checks(struct vcpu_svm *svm, struct vmcb *vmcb12)
229 {
230 	bool vmcb12_lma;
231 
232 	if ((vmcb12->save.efer & EFER_SVME) == 0)
233 		return false;
234 
235 	if (((vmcb12->save.cr0 & X86_CR0_CD) == 0) && (vmcb12->save.cr0 & X86_CR0_NW))
236 		return false;
237 
238 	if (!kvm_dr6_valid(vmcb12->save.dr6) || !kvm_dr7_valid(vmcb12->save.dr7))
239 		return false;
240 
241 	vmcb12_lma = (vmcb12->save.efer & EFER_LME) && (vmcb12->save.cr0 & X86_CR0_PG);
242 
243 	if (!vmcb12_lma) {
244 		if (vmcb12->save.cr4 & X86_CR4_PAE) {
245 			if (vmcb12->save.cr3 & MSR_CR3_LEGACY_PAE_RESERVED_MASK)
246 				return false;
247 		} else {
248 			if (vmcb12->save.cr3 & MSR_CR3_LEGACY_RESERVED_MASK)
249 				return false;
250 		}
251 	} else {
252 		if (!(vmcb12->save.cr4 & X86_CR4_PAE) ||
253 		    !(vmcb12->save.cr0 & X86_CR0_PE) ||
254 		    (vmcb12->save.cr3 & MSR_CR3_LONG_MBZ_MASK))
255 			return false;
256 	}
257 	if (!kvm_is_valid_cr4(&svm->vcpu, vmcb12->save.cr4))
258 		return false;
259 
260 	return nested_vmcb_check_controls(&vmcb12->control);
261 }
262 
263 static void load_nested_vmcb_control(struct vcpu_svm *svm,
264 				     struct vmcb_control_area *control)
265 {
266 	copy_vmcb_control_area(&svm->nested.ctl, control);
267 
268 	/* Copy it here because nested_svm_check_controls will check it.  */
269 	svm->nested.ctl.asid           = control->asid;
270 	svm->nested.ctl.msrpm_base_pa &= ~0x0fffULL;
271 	svm->nested.ctl.iopm_base_pa  &= ~0x0fffULL;
272 }
273 
274 /*
275  * Synchronize fields that are written by the processor, so that
276  * they can be copied back into the nested_vmcb.
277  */
278 void sync_nested_vmcb_control(struct vcpu_svm *svm)
279 {
280 	u32 mask;
281 	svm->nested.ctl.event_inj      = svm->vmcb->control.event_inj;
282 	svm->nested.ctl.event_inj_err  = svm->vmcb->control.event_inj_err;
283 
284 	/* Only a few fields of int_ctl are written by the processor.  */
285 	mask = V_IRQ_MASK | V_TPR_MASK;
286 	if (!(svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK) &&
287 	    svm_is_intercept(svm, INTERCEPT_VINTR)) {
288 		/*
289 		 * In order to request an interrupt window, L0 is usurping
290 		 * svm->vmcb->control.int_ctl and possibly setting V_IRQ
291 		 * even if it was clear in L1's VMCB.  Restoring it would be
292 		 * wrong.  However, in this case V_IRQ will remain true until
293 		 * interrupt_window_interception calls svm_clear_vintr and
294 		 * restores int_ctl.  We can just leave it aside.
295 		 */
296 		mask &= ~V_IRQ_MASK;
297 	}
298 	svm->nested.ctl.int_ctl        &= ~mask;
299 	svm->nested.ctl.int_ctl        |= svm->vmcb->control.int_ctl & mask;
300 }
301 
302 /*
303  * Transfer any event that L0 or L1 wanted to inject into L2 to
304  * EXIT_INT_INFO.
305  */
306 static void nested_vmcb_save_pending_event(struct vcpu_svm *svm,
307 					   struct vmcb *vmcb12)
308 {
309 	struct kvm_vcpu *vcpu = &svm->vcpu;
310 	u32 exit_int_info = 0;
311 	unsigned int nr;
312 
313 	if (vcpu->arch.exception.injected) {
314 		nr = vcpu->arch.exception.nr;
315 		exit_int_info = nr | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT;
316 
317 		if (vcpu->arch.exception.has_error_code) {
318 			exit_int_info |= SVM_EVTINJ_VALID_ERR;
319 			vmcb12->control.exit_int_info_err =
320 				vcpu->arch.exception.error_code;
321 		}
322 
323 	} else if (vcpu->arch.nmi_injected) {
324 		exit_int_info = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
325 
326 	} else if (vcpu->arch.interrupt.injected) {
327 		nr = vcpu->arch.interrupt.nr;
328 		exit_int_info = nr | SVM_EVTINJ_VALID;
329 
330 		if (vcpu->arch.interrupt.soft)
331 			exit_int_info |= SVM_EVTINJ_TYPE_SOFT;
332 		else
333 			exit_int_info |= SVM_EVTINJ_TYPE_INTR;
334 	}
335 
336 	vmcb12->control.exit_int_info = exit_int_info;
337 }
338 
339 static inline bool nested_npt_enabled(struct vcpu_svm *svm)
340 {
341 	return svm->nested.ctl.nested_ctl & SVM_NESTED_CTL_NP_ENABLE;
342 }
343 
344 /*
345  * Load guest's/host's cr3 on nested vmentry or vmexit. @nested_npt is true
346  * if we are emulating VM-Entry into a guest with NPT enabled.
347  */
348 static int nested_svm_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3,
349 			       bool nested_npt)
350 {
351 	if (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63))
352 		return -EINVAL;
353 
354 	if (!nested_npt && is_pae_paging(vcpu) &&
355 	    (cr3 != kvm_read_cr3(vcpu) || pdptrs_changed(vcpu))) {
356 		if (!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
357 			return -EINVAL;
358 	}
359 
360 	/*
361 	 * TODO: optimize unconditional TLB flush/MMU sync here and in
362 	 * kvm_init_shadow_npt_mmu().
363 	 */
364 	if (!nested_npt)
365 		kvm_mmu_new_pgd(vcpu, cr3, false, false);
366 
367 	vcpu->arch.cr3 = cr3;
368 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
369 
370 	kvm_init_mmu(vcpu, false);
371 
372 	return 0;
373 }
374 
375 static void nested_prepare_vmcb_save(struct vcpu_svm *svm, struct vmcb *vmcb12)
376 {
377 	/* Load the nested guest state */
378 	svm->vmcb->save.es = vmcb12->save.es;
379 	svm->vmcb->save.cs = vmcb12->save.cs;
380 	svm->vmcb->save.ss = vmcb12->save.ss;
381 	svm->vmcb->save.ds = vmcb12->save.ds;
382 	svm->vmcb->save.gdtr = vmcb12->save.gdtr;
383 	svm->vmcb->save.idtr = vmcb12->save.idtr;
384 	kvm_set_rflags(&svm->vcpu, vmcb12->save.rflags | X86_EFLAGS_FIXED);
385 	svm_set_efer(&svm->vcpu, vmcb12->save.efer);
386 	svm_set_cr0(&svm->vcpu, vmcb12->save.cr0);
387 	svm_set_cr4(&svm->vcpu, vmcb12->save.cr4);
388 	svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = vmcb12->save.cr2;
389 	kvm_rax_write(&svm->vcpu, vmcb12->save.rax);
390 	kvm_rsp_write(&svm->vcpu, vmcb12->save.rsp);
391 	kvm_rip_write(&svm->vcpu, vmcb12->save.rip);
392 
393 	/* In case we don't even reach vcpu_run, the fields are not updated */
394 	svm->vmcb->save.rax = vmcb12->save.rax;
395 	svm->vmcb->save.rsp = vmcb12->save.rsp;
396 	svm->vmcb->save.rip = vmcb12->save.rip;
397 	svm->vmcb->save.dr7 = vmcb12->save.dr7 | DR7_FIXED_1;
398 	svm->vcpu.arch.dr6  = vmcb12->save.dr6 | DR6_FIXED_1 | DR6_RTM;
399 	svm->vmcb->save.cpl = vmcb12->save.cpl;
400 }
401 
402 static void nested_prepare_vmcb_control(struct vcpu_svm *svm)
403 {
404 	const u32 mask = V_INTR_MASKING_MASK | V_GIF_ENABLE_MASK | V_GIF_MASK;
405 
406 	if (nested_npt_enabled(svm))
407 		nested_svm_init_mmu_context(&svm->vcpu);
408 
409 	svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset =
410 		svm->vcpu.arch.l1_tsc_offset + svm->nested.ctl.tsc_offset;
411 
412 	svm->vmcb->control.int_ctl             =
413 		(svm->nested.ctl.int_ctl & ~mask) |
414 		(svm->nested.hsave->control.int_ctl & mask);
415 
416 	svm->vmcb->control.virt_ext            = svm->nested.ctl.virt_ext;
417 	svm->vmcb->control.int_vector          = svm->nested.ctl.int_vector;
418 	svm->vmcb->control.int_state           = svm->nested.ctl.int_state;
419 	svm->vmcb->control.event_inj           = svm->nested.ctl.event_inj;
420 	svm->vmcb->control.event_inj_err       = svm->nested.ctl.event_inj_err;
421 
422 	svm->vmcb->control.pause_filter_count  = svm->nested.ctl.pause_filter_count;
423 	svm->vmcb->control.pause_filter_thresh = svm->nested.ctl.pause_filter_thresh;
424 
425 	/* Enter Guest-Mode */
426 	enter_guest_mode(&svm->vcpu);
427 
428 	/*
429 	 * Merge guest and host intercepts - must be called  with vcpu in
430 	 * guest-mode to take affect here
431 	 */
432 	recalc_intercepts(svm);
433 
434 	vmcb_mark_all_dirty(svm->vmcb);
435 }
436 
437 int enter_svm_guest_mode(struct vcpu_svm *svm, u64 vmcb12_gpa,
438 			 struct vmcb *vmcb12)
439 {
440 	int ret;
441 
442 	svm->nested.vmcb12_gpa = vmcb12_gpa;
443 	load_nested_vmcb_control(svm, &vmcb12->control);
444 	nested_prepare_vmcb_save(svm, vmcb12);
445 	nested_prepare_vmcb_control(svm);
446 
447 	ret = nested_svm_load_cr3(&svm->vcpu, vmcb12->save.cr3,
448 				  nested_npt_enabled(svm));
449 	if (ret)
450 		return ret;
451 
452 	svm_set_gif(svm, true);
453 
454 	return 0;
455 }
456 
457 int nested_svm_vmrun(struct vcpu_svm *svm)
458 {
459 	int ret;
460 	struct vmcb *vmcb12;
461 	struct vmcb *hsave = svm->nested.hsave;
462 	struct vmcb *vmcb = svm->vmcb;
463 	struct kvm_host_map map;
464 	u64 vmcb12_gpa;
465 
466 	if (is_smm(&svm->vcpu)) {
467 		kvm_queue_exception(&svm->vcpu, UD_VECTOR);
468 		return 1;
469 	}
470 
471 	vmcb12_gpa = svm->vmcb->save.rax;
472 	ret = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(vmcb12_gpa), &map);
473 	if (ret == -EINVAL) {
474 		kvm_inject_gp(&svm->vcpu, 0);
475 		return 1;
476 	} else if (ret) {
477 		return kvm_skip_emulated_instruction(&svm->vcpu);
478 	}
479 
480 	ret = kvm_skip_emulated_instruction(&svm->vcpu);
481 
482 	vmcb12 = map.hva;
483 
484 	if (WARN_ON_ONCE(!svm->nested.initialized))
485 		return -EINVAL;
486 
487 	if (!nested_vmcb_checks(svm, vmcb12)) {
488 		vmcb12->control.exit_code    = SVM_EXIT_ERR;
489 		vmcb12->control.exit_code_hi = 0;
490 		vmcb12->control.exit_info_1  = 0;
491 		vmcb12->control.exit_info_2  = 0;
492 		goto out;
493 	}
494 
495 	trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb12_gpa,
496 			       vmcb12->save.rip,
497 			       vmcb12->control.int_ctl,
498 			       vmcb12->control.event_inj,
499 			       vmcb12->control.nested_ctl);
500 
501 	trace_kvm_nested_intercepts(vmcb12->control.intercepts[INTERCEPT_CR] & 0xffff,
502 				    vmcb12->control.intercepts[INTERCEPT_CR] >> 16,
503 				    vmcb12->control.intercepts[INTERCEPT_EXCEPTION],
504 				    vmcb12->control.intercepts[INTERCEPT_WORD3],
505 				    vmcb12->control.intercepts[INTERCEPT_WORD4],
506 				    vmcb12->control.intercepts[INTERCEPT_WORD5]);
507 
508 	/* Clear internal status */
509 	kvm_clear_exception_queue(&svm->vcpu);
510 	kvm_clear_interrupt_queue(&svm->vcpu);
511 
512 	/*
513 	 * Save the old vmcb, so we don't need to pick what we save, but can
514 	 * restore everything when a VMEXIT occurs
515 	 */
516 	hsave->save.es     = vmcb->save.es;
517 	hsave->save.cs     = vmcb->save.cs;
518 	hsave->save.ss     = vmcb->save.ss;
519 	hsave->save.ds     = vmcb->save.ds;
520 	hsave->save.gdtr   = vmcb->save.gdtr;
521 	hsave->save.idtr   = vmcb->save.idtr;
522 	hsave->save.efer   = svm->vcpu.arch.efer;
523 	hsave->save.cr0    = kvm_read_cr0(&svm->vcpu);
524 	hsave->save.cr4    = svm->vcpu.arch.cr4;
525 	hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
526 	hsave->save.rip    = kvm_rip_read(&svm->vcpu);
527 	hsave->save.rsp    = vmcb->save.rsp;
528 	hsave->save.rax    = vmcb->save.rax;
529 	if (npt_enabled)
530 		hsave->save.cr3    = vmcb->save.cr3;
531 	else
532 		hsave->save.cr3    = kvm_read_cr3(&svm->vcpu);
533 
534 	copy_vmcb_control_area(&hsave->control, &vmcb->control);
535 
536 	svm->nested.nested_run_pending = 1;
537 
538 	if (enter_svm_guest_mode(svm, vmcb12_gpa, vmcb12))
539 		goto out_exit_err;
540 
541 	if (nested_svm_vmrun_msrpm(svm))
542 		goto out;
543 
544 out_exit_err:
545 	svm->nested.nested_run_pending = 0;
546 
547 	svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
548 	svm->vmcb->control.exit_code_hi = 0;
549 	svm->vmcb->control.exit_info_1  = 0;
550 	svm->vmcb->control.exit_info_2  = 0;
551 
552 	nested_svm_vmexit(svm);
553 
554 out:
555 	kvm_vcpu_unmap(&svm->vcpu, &map, true);
556 
557 	return ret;
558 }
559 
560 void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
561 {
562 	to_vmcb->save.fs = from_vmcb->save.fs;
563 	to_vmcb->save.gs = from_vmcb->save.gs;
564 	to_vmcb->save.tr = from_vmcb->save.tr;
565 	to_vmcb->save.ldtr = from_vmcb->save.ldtr;
566 	to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
567 	to_vmcb->save.star = from_vmcb->save.star;
568 	to_vmcb->save.lstar = from_vmcb->save.lstar;
569 	to_vmcb->save.cstar = from_vmcb->save.cstar;
570 	to_vmcb->save.sfmask = from_vmcb->save.sfmask;
571 	to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
572 	to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
573 	to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
574 }
575 
576 int nested_svm_vmexit(struct vcpu_svm *svm)
577 {
578 	int rc;
579 	struct vmcb *vmcb12;
580 	struct vmcb *hsave = svm->nested.hsave;
581 	struct vmcb *vmcb = svm->vmcb;
582 	struct kvm_host_map map;
583 
584 	rc = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(svm->nested.vmcb12_gpa), &map);
585 	if (rc) {
586 		if (rc == -EINVAL)
587 			kvm_inject_gp(&svm->vcpu, 0);
588 		return 1;
589 	}
590 
591 	vmcb12 = map.hva;
592 
593 	/* Exit Guest-Mode */
594 	leave_guest_mode(&svm->vcpu);
595 	svm->nested.vmcb12_gpa = 0;
596 	WARN_ON_ONCE(svm->nested.nested_run_pending);
597 
598 	/* in case we halted in L2 */
599 	svm->vcpu.arch.mp_state = KVM_MP_STATE_RUNNABLE;
600 
601 	/* Give the current vmcb to the guest */
602 
603 	vmcb12->save.es     = vmcb->save.es;
604 	vmcb12->save.cs     = vmcb->save.cs;
605 	vmcb12->save.ss     = vmcb->save.ss;
606 	vmcb12->save.ds     = vmcb->save.ds;
607 	vmcb12->save.gdtr   = vmcb->save.gdtr;
608 	vmcb12->save.idtr   = vmcb->save.idtr;
609 	vmcb12->save.efer   = svm->vcpu.arch.efer;
610 	vmcb12->save.cr0    = kvm_read_cr0(&svm->vcpu);
611 	vmcb12->save.cr3    = kvm_read_cr3(&svm->vcpu);
612 	vmcb12->save.cr2    = vmcb->save.cr2;
613 	vmcb12->save.cr4    = svm->vcpu.arch.cr4;
614 	vmcb12->save.rflags = kvm_get_rflags(&svm->vcpu);
615 	vmcb12->save.rip    = kvm_rip_read(&svm->vcpu);
616 	vmcb12->save.rsp    = kvm_rsp_read(&svm->vcpu);
617 	vmcb12->save.rax    = kvm_rax_read(&svm->vcpu);
618 	vmcb12->save.dr7    = vmcb->save.dr7;
619 	vmcb12->save.dr6    = svm->vcpu.arch.dr6;
620 	vmcb12->save.cpl    = vmcb->save.cpl;
621 
622 	vmcb12->control.int_state         = vmcb->control.int_state;
623 	vmcb12->control.exit_code         = vmcb->control.exit_code;
624 	vmcb12->control.exit_code_hi      = vmcb->control.exit_code_hi;
625 	vmcb12->control.exit_info_1       = vmcb->control.exit_info_1;
626 	vmcb12->control.exit_info_2       = vmcb->control.exit_info_2;
627 
628 	if (vmcb12->control.exit_code != SVM_EXIT_ERR)
629 		nested_vmcb_save_pending_event(svm, vmcb12);
630 
631 	if (svm->nrips_enabled)
632 		vmcb12->control.next_rip  = vmcb->control.next_rip;
633 
634 	vmcb12->control.int_ctl           = svm->nested.ctl.int_ctl;
635 	vmcb12->control.tlb_ctl           = svm->nested.ctl.tlb_ctl;
636 	vmcb12->control.event_inj         = svm->nested.ctl.event_inj;
637 	vmcb12->control.event_inj_err     = svm->nested.ctl.event_inj_err;
638 
639 	vmcb12->control.pause_filter_count =
640 		svm->vmcb->control.pause_filter_count;
641 	vmcb12->control.pause_filter_thresh =
642 		svm->vmcb->control.pause_filter_thresh;
643 
644 	/* Restore the original control entries */
645 	copy_vmcb_control_area(&vmcb->control, &hsave->control);
646 
647 	/* On vmexit the  GIF is set to false */
648 	svm_set_gif(svm, false);
649 
650 	svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset =
651 		svm->vcpu.arch.l1_tsc_offset;
652 
653 	svm->nested.ctl.nested_cr3 = 0;
654 
655 	/* Restore selected save entries */
656 	svm->vmcb->save.es = hsave->save.es;
657 	svm->vmcb->save.cs = hsave->save.cs;
658 	svm->vmcb->save.ss = hsave->save.ss;
659 	svm->vmcb->save.ds = hsave->save.ds;
660 	svm->vmcb->save.gdtr = hsave->save.gdtr;
661 	svm->vmcb->save.idtr = hsave->save.idtr;
662 	kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
663 	kvm_set_rflags(&svm->vcpu, hsave->save.rflags | X86_EFLAGS_FIXED);
664 	svm_set_efer(&svm->vcpu, hsave->save.efer);
665 	svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
666 	svm_set_cr4(&svm->vcpu, hsave->save.cr4);
667 	kvm_rax_write(&svm->vcpu, hsave->save.rax);
668 	kvm_rsp_write(&svm->vcpu, hsave->save.rsp);
669 	kvm_rip_write(&svm->vcpu, hsave->save.rip);
670 	svm->vmcb->save.dr7 = DR7_FIXED_1;
671 	svm->vmcb->save.cpl = 0;
672 	svm->vmcb->control.exit_int_info = 0;
673 
674 	vmcb_mark_all_dirty(svm->vmcb);
675 
676 	trace_kvm_nested_vmexit_inject(vmcb12->control.exit_code,
677 				       vmcb12->control.exit_info_1,
678 				       vmcb12->control.exit_info_2,
679 				       vmcb12->control.exit_int_info,
680 				       vmcb12->control.exit_int_info_err,
681 				       KVM_ISA_SVM);
682 
683 	kvm_vcpu_unmap(&svm->vcpu, &map, true);
684 
685 	nested_svm_uninit_mmu_context(&svm->vcpu);
686 
687 	rc = nested_svm_load_cr3(&svm->vcpu, hsave->save.cr3, false);
688 	if (rc)
689 		return 1;
690 
691 	if (npt_enabled)
692 		svm->vmcb->save.cr3 = hsave->save.cr3;
693 
694 	/*
695 	 * Drop what we picked up for L2 via svm_complete_interrupts() so it
696 	 * doesn't end up in L1.
697 	 */
698 	svm->vcpu.arch.nmi_injected = false;
699 	kvm_clear_exception_queue(&svm->vcpu);
700 	kvm_clear_interrupt_queue(&svm->vcpu);
701 
702 	return 0;
703 }
704 
705 int svm_allocate_nested(struct vcpu_svm *svm)
706 {
707 	struct page *hsave_page;
708 
709 	if (svm->nested.initialized)
710 		return 0;
711 
712 	hsave_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
713 	if (!hsave_page)
714 		return -ENOMEM;
715 	svm->nested.hsave = page_address(hsave_page);
716 
717 	svm->nested.msrpm = svm_vcpu_alloc_msrpm();
718 	if (!svm->nested.msrpm)
719 		goto err_free_hsave;
720 	svm_vcpu_init_msrpm(&svm->vcpu, svm->nested.msrpm);
721 
722 	svm->nested.initialized = true;
723 	return 0;
724 
725 err_free_hsave:
726 	__free_page(hsave_page);
727 	return -ENOMEM;
728 }
729 
730 void svm_free_nested(struct vcpu_svm *svm)
731 {
732 	if (!svm->nested.initialized)
733 		return;
734 
735 	svm_vcpu_free_msrpm(svm->nested.msrpm);
736 	svm->nested.msrpm = NULL;
737 
738 	__free_page(virt_to_page(svm->nested.hsave));
739 	svm->nested.hsave = NULL;
740 
741 	svm->nested.initialized = false;
742 }
743 
744 /*
745  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
746  */
747 void svm_leave_nested(struct vcpu_svm *svm)
748 {
749 	if (is_guest_mode(&svm->vcpu)) {
750 		struct vmcb *hsave = svm->nested.hsave;
751 		struct vmcb *vmcb = svm->vmcb;
752 
753 		svm->nested.nested_run_pending = 0;
754 		leave_guest_mode(&svm->vcpu);
755 		copy_vmcb_control_area(&vmcb->control, &hsave->control);
756 		nested_svm_uninit_mmu_context(&svm->vcpu);
757 	}
758 
759 	kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, &svm->vcpu);
760 }
761 
762 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
763 {
764 	u32 offset, msr, value;
765 	int write, mask;
766 
767 	if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
768 		return NESTED_EXIT_HOST;
769 
770 	msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
771 	offset = svm_msrpm_offset(msr);
772 	write  = svm->vmcb->control.exit_info_1 & 1;
773 	mask   = 1 << ((2 * (msr & 0xf)) + write);
774 
775 	if (offset == MSR_INVALID)
776 		return NESTED_EXIT_DONE;
777 
778 	/* Offset is in 32 bit units but need in 8 bit units */
779 	offset *= 4;
780 
781 	if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.ctl.msrpm_base_pa + offset, &value, 4))
782 		return NESTED_EXIT_DONE;
783 
784 	return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
785 }
786 
787 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
788 {
789 	unsigned port, size, iopm_len;
790 	u16 val, mask;
791 	u8 start_bit;
792 	u64 gpa;
793 
794 	if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_IOIO_PROT)))
795 		return NESTED_EXIT_HOST;
796 
797 	port = svm->vmcb->control.exit_info_1 >> 16;
798 	size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
799 		SVM_IOIO_SIZE_SHIFT;
800 	gpa  = svm->nested.ctl.iopm_base_pa + (port / 8);
801 	start_bit = port % 8;
802 	iopm_len = (start_bit + size > 8) ? 2 : 1;
803 	mask = (0xf >> (4 - size)) << start_bit;
804 	val = 0;
805 
806 	if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
807 		return NESTED_EXIT_DONE;
808 
809 	return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
810 }
811 
812 static int nested_svm_intercept(struct vcpu_svm *svm)
813 {
814 	u32 exit_code = svm->vmcb->control.exit_code;
815 	int vmexit = NESTED_EXIT_HOST;
816 
817 	switch (exit_code) {
818 	case SVM_EXIT_MSR:
819 		vmexit = nested_svm_exit_handled_msr(svm);
820 		break;
821 	case SVM_EXIT_IOIO:
822 		vmexit = nested_svm_intercept_ioio(svm);
823 		break;
824 	case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
825 		if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
826 			vmexit = NESTED_EXIT_DONE;
827 		break;
828 	}
829 	case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
830 		if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
831 			vmexit = NESTED_EXIT_DONE;
832 		break;
833 	}
834 	case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
835 		/*
836 		 * Host-intercepted exceptions have been checked already in
837 		 * nested_svm_exit_special.  There is nothing to do here,
838 		 * the vmexit is injected by svm_check_nested_events.
839 		 */
840 		vmexit = NESTED_EXIT_DONE;
841 		break;
842 	}
843 	case SVM_EXIT_ERR: {
844 		vmexit = NESTED_EXIT_DONE;
845 		break;
846 	}
847 	default: {
848 		if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
849 			vmexit = NESTED_EXIT_DONE;
850 	}
851 	}
852 
853 	return vmexit;
854 }
855 
856 int nested_svm_exit_handled(struct vcpu_svm *svm)
857 {
858 	int vmexit;
859 
860 	vmexit = nested_svm_intercept(svm);
861 
862 	if (vmexit == NESTED_EXIT_DONE)
863 		nested_svm_vmexit(svm);
864 
865 	return vmexit;
866 }
867 
868 int nested_svm_check_permissions(struct vcpu_svm *svm)
869 {
870 	if (!(svm->vcpu.arch.efer & EFER_SVME) ||
871 	    !is_paging(&svm->vcpu)) {
872 		kvm_queue_exception(&svm->vcpu, UD_VECTOR);
873 		return 1;
874 	}
875 
876 	if (svm->vmcb->save.cpl) {
877 		kvm_inject_gp(&svm->vcpu, 0);
878 		return 1;
879 	}
880 
881 	return 0;
882 }
883 
884 static bool nested_exit_on_exception(struct vcpu_svm *svm)
885 {
886 	unsigned int nr = svm->vcpu.arch.exception.nr;
887 
888 	return (svm->nested.ctl.intercepts[INTERCEPT_EXCEPTION] & BIT(nr));
889 }
890 
891 static void nested_svm_inject_exception_vmexit(struct vcpu_svm *svm)
892 {
893 	unsigned int nr = svm->vcpu.arch.exception.nr;
894 
895 	svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
896 	svm->vmcb->control.exit_code_hi = 0;
897 
898 	if (svm->vcpu.arch.exception.has_error_code)
899 		svm->vmcb->control.exit_info_1 = svm->vcpu.arch.exception.error_code;
900 
901 	/*
902 	 * EXITINFO2 is undefined for all exception intercepts other
903 	 * than #PF.
904 	 */
905 	if (nr == PF_VECTOR) {
906 		if (svm->vcpu.arch.exception.nested_apf)
907 			svm->vmcb->control.exit_info_2 = svm->vcpu.arch.apf.nested_apf_token;
908 		else if (svm->vcpu.arch.exception.has_payload)
909 			svm->vmcb->control.exit_info_2 = svm->vcpu.arch.exception.payload;
910 		else
911 			svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
912 	} else if (nr == DB_VECTOR) {
913 		/* See inject_pending_event.  */
914 		kvm_deliver_exception_payload(&svm->vcpu);
915 		if (svm->vcpu.arch.dr7 & DR7_GD) {
916 			svm->vcpu.arch.dr7 &= ~DR7_GD;
917 			kvm_update_dr7(&svm->vcpu);
918 		}
919 	} else
920 		WARN_ON(svm->vcpu.arch.exception.has_payload);
921 
922 	nested_svm_vmexit(svm);
923 }
924 
925 static void nested_svm_smi(struct vcpu_svm *svm)
926 {
927 	svm->vmcb->control.exit_code = SVM_EXIT_SMI;
928 	svm->vmcb->control.exit_info_1 = 0;
929 	svm->vmcb->control.exit_info_2 = 0;
930 
931 	nested_svm_vmexit(svm);
932 }
933 
934 static void nested_svm_nmi(struct vcpu_svm *svm)
935 {
936 	svm->vmcb->control.exit_code = SVM_EXIT_NMI;
937 	svm->vmcb->control.exit_info_1 = 0;
938 	svm->vmcb->control.exit_info_2 = 0;
939 
940 	nested_svm_vmexit(svm);
941 }
942 
943 static void nested_svm_intr(struct vcpu_svm *svm)
944 {
945 	trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
946 
947 	svm->vmcb->control.exit_code   = SVM_EXIT_INTR;
948 	svm->vmcb->control.exit_info_1 = 0;
949 	svm->vmcb->control.exit_info_2 = 0;
950 
951 	nested_svm_vmexit(svm);
952 }
953 
954 static inline bool nested_exit_on_init(struct vcpu_svm *svm)
955 {
956 	return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_INIT);
957 }
958 
959 static void nested_svm_init(struct vcpu_svm *svm)
960 {
961 	svm->vmcb->control.exit_code   = SVM_EXIT_INIT;
962 	svm->vmcb->control.exit_info_1 = 0;
963 	svm->vmcb->control.exit_info_2 = 0;
964 
965 	nested_svm_vmexit(svm);
966 }
967 
968 
969 static int svm_check_nested_events(struct kvm_vcpu *vcpu)
970 {
971 	struct vcpu_svm *svm = to_svm(vcpu);
972 	bool block_nested_events =
973 		kvm_event_needs_reinjection(vcpu) || svm->nested.nested_run_pending;
974 	struct kvm_lapic *apic = vcpu->arch.apic;
975 
976 	if (lapic_in_kernel(vcpu) &&
977 	    test_bit(KVM_APIC_INIT, &apic->pending_events)) {
978 		if (block_nested_events)
979 			return -EBUSY;
980 		if (!nested_exit_on_init(svm))
981 			return 0;
982 		nested_svm_init(svm);
983 		return 0;
984 	}
985 
986 	if (vcpu->arch.exception.pending) {
987 		if (block_nested_events)
988                         return -EBUSY;
989 		if (!nested_exit_on_exception(svm))
990 			return 0;
991 		nested_svm_inject_exception_vmexit(svm);
992 		return 0;
993 	}
994 
995 	if (vcpu->arch.smi_pending && !svm_smi_blocked(vcpu)) {
996 		if (block_nested_events)
997 			return -EBUSY;
998 		if (!nested_exit_on_smi(svm))
999 			return 0;
1000 		nested_svm_smi(svm);
1001 		return 0;
1002 	}
1003 
1004 	if (vcpu->arch.nmi_pending && !svm_nmi_blocked(vcpu)) {
1005 		if (block_nested_events)
1006 			return -EBUSY;
1007 		if (!nested_exit_on_nmi(svm))
1008 			return 0;
1009 		nested_svm_nmi(svm);
1010 		return 0;
1011 	}
1012 
1013 	if (kvm_cpu_has_interrupt(vcpu) && !svm_interrupt_blocked(vcpu)) {
1014 		if (block_nested_events)
1015 			return -EBUSY;
1016 		if (!nested_exit_on_intr(svm))
1017 			return 0;
1018 		nested_svm_intr(svm);
1019 		return 0;
1020 	}
1021 
1022 	return 0;
1023 }
1024 
1025 int nested_svm_exit_special(struct vcpu_svm *svm)
1026 {
1027 	u32 exit_code = svm->vmcb->control.exit_code;
1028 
1029 	switch (exit_code) {
1030 	case SVM_EXIT_INTR:
1031 	case SVM_EXIT_NMI:
1032 	case SVM_EXIT_NPF:
1033 		return NESTED_EXIT_HOST;
1034 	case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1035 		u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
1036 
1037 		if (get_host_vmcb(svm)->control.intercepts[INTERCEPT_EXCEPTION] &
1038 				excp_bits)
1039 			return NESTED_EXIT_HOST;
1040 		else if (exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR &&
1041 			 svm->vcpu.arch.apf.host_apf_flags)
1042 			/* Trap async PF even if not shadowing */
1043 			return NESTED_EXIT_HOST;
1044 		break;
1045 	}
1046 	default:
1047 		break;
1048 	}
1049 
1050 	return NESTED_EXIT_CONTINUE;
1051 }
1052 
1053 static int svm_get_nested_state(struct kvm_vcpu *vcpu,
1054 				struct kvm_nested_state __user *user_kvm_nested_state,
1055 				u32 user_data_size)
1056 {
1057 	struct vcpu_svm *svm;
1058 	struct kvm_nested_state kvm_state = {
1059 		.flags = 0,
1060 		.format = KVM_STATE_NESTED_FORMAT_SVM,
1061 		.size = sizeof(kvm_state),
1062 	};
1063 	struct vmcb __user *user_vmcb = (struct vmcb __user *)
1064 		&user_kvm_nested_state->data.svm[0];
1065 
1066 	if (!vcpu)
1067 		return kvm_state.size + KVM_STATE_NESTED_SVM_VMCB_SIZE;
1068 
1069 	svm = to_svm(vcpu);
1070 
1071 	if (user_data_size < kvm_state.size)
1072 		goto out;
1073 
1074 	/* First fill in the header and copy it out.  */
1075 	if (is_guest_mode(vcpu)) {
1076 		kvm_state.hdr.svm.vmcb_pa = svm->nested.vmcb12_gpa;
1077 		kvm_state.size += KVM_STATE_NESTED_SVM_VMCB_SIZE;
1078 		kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
1079 
1080 		if (svm->nested.nested_run_pending)
1081 			kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
1082 	}
1083 
1084 	if (gif_set(svm))
1085 		kvm_state.flags |= KVM_STATE_NESTED_GIF_SET;
1086 
1087 	if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
1088 		return -EFAULT;
1089 
1090 	if (!is_guest_mode(vcpu))
1091 		goto out;
1092 
1093 	/*
1094 	 * Copy over the full size of the VMCB rather than just the size
1095 	 * of the structs.
1096 	 */
1097 	if (clear_user(user_vmcb, KVM_STATE_NESTED_SVM_VMCB_SIZE))
1098 		return -EFAULT;
1099 	if (copy_to_user(&user_vmcb->control, &svm->nested.ctl,
1100 			 sizeof(user_vmcb->control)))
1101 		return -EFAULT;
1102 	if (copy_to_user(&user_vmcb->save, &svm->nested.hsave->save,
1103 			 sizeof(user_vmcb->save)))
1104 		return -EFAULT;
1105 
1106 out:
1107 	return kvm_state.size;
1108 }
1109 
1110 static int svm_set_nested_state(struct kvm_vcpu *vcpu,
1111 				struct kvm_nested_state __user *user_kvm_nested_state,
1112 				struct kvm_nested_state *kvm_state)
1113 {
1114 	struct vcpu_svm *svm = to_svm(vcpu);
1115 	struct vmcb *hsave = svm->nested.hsave;
1116 	struct vmcb __user *user_vmcb = (struct vmcb __user *)
1117 		&user_kvm_nested_state->data.svm[0];
1118 	struct vmcb_control_area *ctl;
1119 	struct vmcb_save_area *save;
1120 	int ret;
1121 	u32 cr0;
1122 
1123 	BUILD_BUG_ON(sizeof(struct vmcb_control_area) + sizeof(struct vmcb_save_area) >
1124 		     KVM_STATE_NESTED_SVM_VMCB_SIZE);
1125 
1126 	if (kvm_state->format != KVM_STATE_NESTED_FORMAT_SVM)
1127 		return -EINVAL;
1128 
1129 	if (kvm_state->flags & ~(KVM_STATE_NESTED_GUEST_MODE |
1130 				 KVM_STATE_NESTED_RUN_PENDING |
1131 				 KVM_STATE_NESTED_GIF_SET))
1132 		return -EINVAL;
1133 
1134 	/*
1135 	 * If in guest mode, vcpu->arch.efer actually refers to the L2 guest's
1136 	 * EFER.SVME, but EFER.SVME still has to be 1 for VMRUN to succeed.
1137 	 */
1138 	if (!(vcpu->arch.efer & EFER_SVME)) {
1139 		/* GIF=1 and no guest mode are required if SVME=0.  */
1140 		if (kvm_state->flags != KVM_STATE_NESTED_GIF_SET)
1141 			return -EINVAL;
1142 	}
1143 
1144 	/* SMM temporarily disables SVM, so we cannot be in guest mode.  */
1145 	if (is_smm(vcpu) && (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
1146 		return -EINVAL;
1147 
1148 	if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE)) {
1149 		svm_leave_nested(svm);
1150 		svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
1151 		return 0;
1152 	}
1153 
1154 	if (!page_address_valid(vcpu, kvm_state->hdr.svm.vmcb_pa))
1155 		return -EINVAL;
1156 	if (kvm_state->size < sizeof(*kvm_state) + KVM_STATE_NESTED_SVM_VMCB_SIZE)
1157 		return -EINVAL;
1158 
1159 	ret  = -ENOMEM;
1160 	ctl  = kzalloc(sizeof(*ctl),  GFP_KERNEL);
1161 	save = kzalloc(sizeof(*save), GFP_KERNEL);
1162 	if (!ctl || !save)
1163 		goto out_free;
1164 
1165 	ret = -EFAULT;
1166 	if (copy_from_user(ctl, &user_vmcb->control, sizeof(*ctl)))
1167 		goto out_free;
1168 	if (copy_from_user(save, &user_vmcb->save, sizeof(*save)))
1169 		goto out_free;
1170 
1171 	ret = -EINVAL;
1172 	if (!nested_vmcb_check_controls(ctl))
1173 		goto out_free;
1174 
1175 	/*
1176 	 * Processor state contains L2 state.  Check that it is
1177 	 * valid for guest mode (see nested_vmcb_checks).
1178 	 */
1179 	cr0 = kvm_read_cr0(vcpu);
1180         if (((cr0 & X86_CR0_CD) == 0) && (cr0 & X86_CR0_NW))
1181 		goto out_free;
1182 
1183 	/*
1184 	 * Validate host state saved from before VMRUN (see
1185 	 * nested_svm_check_permissions).
1186 	 * TODO: validate reserved bits for all saved state.
1187 	 */
1188 	if (!(save->cr0 & X86_CR0_PG))
1189 		goto out_free;
1190 
1191 	/*
1192 	 * All checks done, we can enter guest mode.  L1 control fields
1193 	 * come from the nested save state.  Guest state is already
1194 	 * in the registers, the save area of the nested state instead
1195 	 * contains saved L1 state.
1196 	 */
1197 	copy_vmcb_control_area(&hsave->control, &svm->vmcb->control);
1198 	hsave->save = *save;
1199 
1200 	svm->nested.vmcb12_gpa = kvm_state->hdr.svm.vmcb_pa;
1201 	load_nested_vmcb_control(svm, ctl);
1202 	nested_prepare_vmcb_control(svm);
1203 
1204 	kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
1205 	ret = 0;
1206 out_free:
1207 	kfree(save);
1208 	kfree(ctl);
1209 
1210 	return ret;
1211 }
1212 
1213 struct kvm_x86_nested_ops svm_nested_ops = {
1214 	.check_events = svm_check_nested_events,
1215 	.get_nested_state_pages = svm_get_nested_state_pages,
1216 	.get_state = svm_get_nested_state,
1217 	.set_state = svm_set_nested_state,
1218 };
1219