xref: /openbmc/linux/arch/x86/kvm/svm/nested.c (revision cd6d421e)
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 void svm_inject_page_fault_nested(struct kvm_vcpu *vcpu, struct x86_exception *fault)
55 {
56        struct vcpu_svm *svm = to_svm(vcpu);
57        WARN_ON(!is_guest_mode(vcpu));
58 
59        if (vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_EXCEPTION_OFFSET + PF_VECTOR) &&
60 	   !svm->nested.nested_run_pending) {
61                svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + PF_VECTOR;
62                svm->vmcb->control.exit_code_hi = 0;
63                svm->vmcb->control.exit_info_1 = fault->error_code;
64                svm->vmcb->control.exit_info_2 = fault->address;
65                nested_svm_vmexit(svm);
66        } else {
67                kvm_inject_page_fault(vcpu, fault);
68        }
69 }
70 
71 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
72 {
73 	struct vcpu_svm *svm = to_svm(vcpu);
74 	u64 cr3 = svm->nested.ctl.nested_cr3;
75 	u64 pdpte;
76 	int ret;
77 
78 	ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(cr3), &pdpte,
79 				       offset_in_page(cr3) + index * 8, 8);
80 	if (ret)
81 		return 0;
82 	return pdpte;
83 }
84 
85 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
86 {
87 	struct vcpu_svm *svm = to_svm(vcpu);
88 
89 	return svm->nested.ctl.nested_cr3;
90 }
91 
92 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
93 {
94 	struct vcpu_svm *svm = to_svm(vcpu);
95 	struct vmcb *hsave = svm->nested.hsave;
96 
97 	WARN_ON(mmu_is_nested(vcpu));
98 
99 	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
100 	kvm_init_shadow_npt_mmu(vcpu, X86_CR0_PG, hsave->save.cr4, hsave->save.efer,
101 				svm->nested.ctl.nested_cr3);
102 	vcpu->arch.mmu->get_guest_pgd     = nested_svm_get_tdp_cr3;
103 	vcpu->arch.mmu->get_pdptr         = nested_svm_get_tdp_pdptr;
104 	vcpu->arch.mmu->inject_page_fault = nested_svm_inject_npf_exit;
105 	reset_shadow_zero_bits_mask(vcpu, vcpu->arch.mmu);
106 	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
107 }
108 
109 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
110 {
111 	vcpu->arch.mmu = &vcpu->arch.root_mmu;
112 	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
113 }
114 
115 void recalc_intercepts(struct vcpu_svm *svm)
116 {
117 	struct vmcb_control_area *c, *h, *g;
118 	unsigned int i;
119 
120 	vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
121 
122 	if (!is_guest_mode(&svm->vcpu))
123 		return;
124 
125 	c = &svm->vmcb->control;
126 	h = &svm->nested.hsave->control;
127 	g = &svm->nested.ctl;
128 
129 	for (i = 0; i < MAX_INTERCEPT; i++)
130 		c->intercepts[i] = h->intercepts[i];
131 
132 	if (g->int_ctl & V_INTR_MASKING_MASK) {
133 		/* We only want the cr8 intercept bits of L1 */
134 		vmcb_clr_intercept(c, INTERCEPT_CR8_READ);
135 		vmcb_clr_intercept(c, INTERCEPT_CR8_WRITE);
136 
137 		/*
138 		 * Once running L2 with HF_VINTR_MASK, EFLAGS.IF does not
139 		 * affect any interrupt we may want to inject; therefore,
140 		 * interrupt window vmexits are irrelevant to L0.
141 		 */
142 		vmcb_clr_intercept(c, INTERCEPT_VINTR);
143 	}
144 
145 	/* We don't want to see VMMCALLs from a nested guest */
146 	vmcb_clr_intercept(c, INTERCEPT_VMMCALL);
147 
148 	for (i = 0; i < MAX_INTERCEPT; i++)
149 		c->intercepts[i] |= g->intercepts[i];
150 }
151 
152 static void copy_vmcb_control_area(struct vmcb_control_area *dst,
153 				   struct vmcb_control_area *from)
154 {
155 	unsigned int i;
156 
157 	for (i = 0; i < MAX_INTERCEPT; i++)
158 		dst->intercepts[i] = from->intercepts[i];
159 
160 	dst->iopm_base_pa         = from->iopm_base_pa;
161 	dst->msrpm_base_pa        = from->msrpm_base_pa;
162 	dst->tsc_offset           = from->tsc_offset;
163 	/* asid not copied, it is handled manually for svm->vmcb.  */
164 	dst->tlb_ctl              = from->tlb_ctl;
165 	dst->int_ctl              = from->int_ctl;
166 	dst->int_vector           = from->int_vector;
167 	dst->int_state            = from->int_state;
168 	dst->exit_code            = from->exit_code;
169 	dst->exit_code_hi         = from->exit_code_hi;
170 	dst->exit_info_1          = from->exit_info_1;
171 	dst->exit_info_2          = from->exit_info_2;
172 	dst->exit_int_info        = from->exit_int_info;
173 	dst->exit_int_info_err    = from->exit_int_info_err;
174 	dst->nested_ctl           = from->nested_ctl;
175 	dst->event_inj            = from->event_inj;
176 	dst->event_inj_err        = from->event_inj_err;
177 	dst->nested_cr3           = from->nested_cr3;
178 	dst->virt_ext              = from->virt_ext;
179 	dst->pause_filter_count   = from->pause_filter_count;
180 	dst->pause_filter_thresh  = from->pause_filter_thresh;
181 }
182 
183 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
184 {
185 	/*
186 	 * This function merges the msr permission bitmaps of kvm and the
187 	 * nested vmcb. It is optimized in that it only merges the parts where
188 	 * the kvm msr permission bitmap may contain zero bits
189 	 */
190 	int i;
191 
192 	if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
193 		return true;
194 
195 	for (i = 0; i < MSRPM_OFFSETS; i++) {
196 		u32 value, p;
197 		u64 offset;
198 
199 		if (msrpm_offsets[i] == 0xffffffff)
200 			break;
201 
202 		p      = msrpm_offsets[i];
203 		offset = svm->nested.ctl.msrpm_base_pa + (p * 4);
204 
205 		if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
206 			return false;
207 
208 		svm->nested.msrpm[p] = svm->msrpm[p] | value;
209 	}
210 
211 	svm->vmcb->control.msrpm_base_pa = __sme_set(__pa(svm->nested.msrpm));
212 
213 	return true;
214 }
215 
216 static bool svm_get_nested_state_pages(struct kvm_vcpu *vcpu)
217 {
218 	struct vcpu_svm *svm = to_svm(vcpu);
219 
220 	if (WARN_ON(!is_guest_mode(vcpu)))
221 		return true;
222 
223 	if (!nested_svm_vmrun_msrpm(svm)) {
224 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
225 		vcpu->run->internal.suberror =
226 			KVM_INTERNAL_ERROR_EMULATION;
227 		vcpu->run->internal.ndata = 0;
228 		return false;
229 	}
230 
231 	return true;
232 }
233 
234 static bool nested_vmcb_check_controls(struct vmcb_control_area *control)
235 {
236 	if ((vmcb_is_intercept(control, INTERCEPT_VMRUN)) == 0)
237 		return false;
238 
239 	if (control->asid == 0)
240 		return false;
241 
242 	if ((control->nested_ctl & SVM_NESTED_CTL_NP_ENABLE) &&
243 	    !npt_enabled)
244 		return false;
245 
246 	return true;
247 }
248 
249 static bool nested_vmcb_check_save(struct vcpu_svm *svm, struct vmcb *vmcb12)
250 {
251 	struct kvm_vcpu *vcpu = &svm->vcpu;
252 	bool vmcb12_lma;
253 
254 	/*
255 	 * FIXME: these should be done after copying the fields,
256 	 * to avoid TOC/TOU races.  For these save area checks
257 	 * the possible damage is limited since kvm_set_cr0 and
258 	 * kvm_set_cr4 handle failure; EFER_SVME is an exception
259 	 * so it is force-set later in nested_prepare_vmcb_save.
260 	 */
261 	if ((vmcb12->save.efer & EFER_SVME) == 0)
262 		return false;
263 
264 	if (((vmcb12->save.cr0 & X86_CR0_CD) == 0) && (vmcb12->save.cr0 & X86_CR0_NW))
265 		return false;
266 
267 	if (!kvm_dr6_valid(vmcb12->save.dr6) || !kvm_dr7_valid(vmcb12->save.dr7))
268 		return false;
269 
270 	vmcb12_lma = (vmcb12->save.efer & EFER_LME) && (vmcb12->save.cr0 & X86_CR0_PG);
271 
272 	if (vmcb12_lma) {
273 		if (!(vmcb12->save.cr4 & X86_CR4_PAE) ||
274 		    !(vmcb12->save.cr0 & X86_CR0_PE) ||
275 		    kvm_vcpu_is_illegal_gpa(vcpu, vmcb12->save.cr3))
276 			return false;
277 	}
278 	if (!kvm_is_valid_cr4(&svm->vcpu, vmcb12->save.cr4))
279 		return false;
280 
281 	return true;
282 }
283 
284 static void load_nested_vmcb_control(struct vcpu_svm *svm,
285 				     struct vmcb_control_area *control)
286 {
287 	copy_vmcb_control_area(&svm->nested.ctl, control);
288 
289 	/* Copy it here because nested_svm_check_controls will check it.  */
290 	svm->nested.ctl.asid           = control->asid;
291 	svm->nested.ctl.msrpm_base_pa &= ~0x0fffULL;
292 	svm->nested.ctl.iopm_base_pa  &= ~0x0fffULL;
293 }
294 
295 /*
296  * Synchronize fields that are written by the processor, so that
297  * they can be copied back into the nested_vmcb.
298  */
299 void sync_nested_vmcb_control(struct vcpu_svm *svm)
300 {
301 	u32 mask;
302 	svm->nested.ctl.event_inj      = svm->vmcb->control.event_inj;
303 	svm->nested.ctl.event_inj_err  = svm->vmcb->control.event_inj_err;
304 
305 	/* Only a few fields of int_ctl are written by the processor.  */
306 	mask = V_IRQ_MASK | V_TPR_MASK;
307 	if (!(svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK) &&
308 	    svm_is_intercept(svm, INTERCEPT_VINTR)) {
309 		/*
310 		 * In order to request an interrupt window, L0 is usurping
311 		 * svm->vmcb->control.int_ctl and possibly setting V_IRQ
312 		 * even if it was clear in L1's VMCB.  Restoring it would be
313 		 * wrong.  However, in this case V_IRQ will remain true until
314 		 * interrupt_window_interception calls svm_clear_vintr and
315 		 * restores int_ctl.  We can just leave it aside.
316 		 */
317 		mask &= ~V_IRQ_MASK;
318 	}
319 	svm->nested.ctl.int_ctl        &= ~mask;
320 	svm->nested.ctl.int_ctl        |= svm->vmcb->control.int_ctl & mask;
321 }
322 
323 /*
324  * Transfer any event that L0 or L1 wanted to inject into L2 to
325  * EXIT_INT_INFO.
326  */
327 static void nested_vmcb_save_pending_event(struct vcpu_svm *svm,
328 					   struct vmcb *vmcb12)
329 {
330 	struct kvm_vcpu *vcpu = &svm->vcpu;
331 	u32 exit_int_info = 0;
332 	unsigned int nr;
333 
334 	if (vcpu->arch.exception.injected) {
335 		nr = vcpu->arch.exception.nr;
336 		exit_int_info = nr | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT;
337 
338 		if (vcpu->arch.exception.has_error_code) {
339 			exit_int_info |= SVM_EVTINJ_VALID_ERR;
340 			vmcb12->control.exit_int_info_err =
341 				vcpu->arch.exception.error_code;
342 		}
343 
344 	} else if (vcpu->arch.nmi_injected) {
345 		exit_int_info = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
346 
347 	} else if (vcpu->arch.interrupt.injected) {
348 		nr = vcpu->arch.interrupt.nr;
349 		exit_int_info = nr | SVM_EVTINJ_VALID;
350 
351 		if (vcpu->arch.interrupt.soft)
352 			exit_int_info |= SVM_EVTINJ_TYPE_SOFT;
353 		else
354 			exit_int_info |= SVM_EVTINJ_TYPE_INTR;
355 	}
356 
357 	vmcb12->control.exit_int_info = exit_int_info;
358 }
359 
360 static inline bool nested_npt_enabled(struct vcpu_svm *svm)
361 {
362 	return svm->nested.ctl.nested_ctl & SVM_NESTED_CTL_NP_ENABLE;
363 }
364 
365 /*
366  * Load guest's/host's cr3 on nested vmentry or vmexit. @nested_npt is true
367  * if we are emulating VM-Entry into a guest with NPT enabled.
368  */
369 static int nested_svm_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3,
370 			       bool nested_npt)
371 {
372 	if (kvm_vcpu_is_illegal_gpa(vcpu, cr3))
373 		return -EINVAL;
374 
375 	if (!nested_npt && is_pae_paging(vcpu) &&
376 	    (cr3 != kvm_read_cr3(vcpu) || pdptrs_changed(vcpu))) {
377 		if (!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
378 			return -EINVAL;
379 	}
380 
381 	/*
382 	 * TODO: optimize unconditional TLB flush/MMU sync here and in
383 	 * kvm_init_shadow_npt_mmu().
384 	 */
385 	if (!nested_npt)
386 		kvm_mmu_new_pgd(vcpu, cr3, false, false);
387 
388 	vcpu->arch.cr3 = cr3;
389 	kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
390 
391 	kvm_init_mmu(vcpu, false);
392 
393 	return 0;
394 }
395 
396 static void nested_prepare_vmcb_save(struct vcpu_svm *svm, struct vmcb *vmcb12)
397 {
398 	/* Load the nested guest state */
399 	svm->vmcb->save.es = vmcb12->save.es;
400 	svm->vmcb->save.cs = vmcb12->save.cs;
401 	svm->vmcb->save.ss = vmcb12->save.ss;
402 	svm->vmcb->save.ds = vmcb12->save.ds;
403 	svm->vmcb->save.gdtr = vmcb12->save.gdtr;
404 	svm->vmcb->save.idtr = vmcb12->save.idtr;
405 	kvm_set_rflags(&svm->vcpu, vmcb12->save.rflags | X86_EFLAGS_FIXED);
406 
407 	/*
408 	 * Force-set EFER_SVME even though it is checked earlier on the
409 	 * VMCB12, because the guest can flip the bit between the check
410 	 * and now.  Clearing EFER_SVME would call svm_free_nested.
411 	 */
412 	svm_set_efer(&svm->vcpu, vmcb12->save.efer | EFER_SVME);
413 
414 	svm_set_cr0(&svm->vcpu, vmcb12->save.cr0);
415 	svm_set_cr4(&svm->vcpu, vmcb12->save.cr4);
416 	svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = vmcb12->save.cr2;
417 	kvm_rax_write(&svm->vcpu, vmcb12->save.rax);
418 	kvm_rsp_write(&svm->vcpu, vmcb12->save.rsp);
419 	kvm_rip_write(&svm->vcpu, vmcb12->save.rip);
420 
421 	/* In case we don't even reach vcpu_run, the fields are not updated */
422 	svm->vmcb->save.rax = vmcb12->save.rax;
423 	svm->vmcb->save.rsp = vmcb12->save.rsp;
424 	svm->vmcb->save.rip = vmcb12->save.rip;
425 	svm->vmcb->save.dr7 = vmcb12->save.dr7 | DR7_FIXED_1;
426 	svm->vcpu.arch.dr6  = vmcb12->save.dr6 | DR6_ACTIVE_LOW;
427 	svm->vmcb->save.cpl = vmcb12->save.cpl;
428 }
429 
430 static void nested_prepare_vmcb_control(struct vcpu_svm *svm)
431 {
432 	const u32 mask = V_INTR_MASKING_MASK | V_GIF_ENABLE_MASK | V_GIF_MASK;
433 
434 	if (nested_npt_enabled(svm))
435 		nested_svm_init_mmu_context(&svm->vcpu);
436 
437 	svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset =
438 		svm->vcpu.arch.l1_tsc_offset + svm->nested.ctl.tsc_offset;
439 
440 	svm->vmcb->control.int_ctl             =
441 		(svm->nested.ctl.int_ctl & ~mask) |
442 		(svm->nested.hsave->control.int_ctl & mask);
443 
444 	svm->vmcb->control.virt_ext            = svm->nested.ctl.virt_ext;
445 	svm->vmcb->control.int_vector          = svm->nested.ctl.int_vector;
446 	svm->vmcb->control.int_state           = svm->nested.ctl.int_state;
447 	svm->vmcb->control.event_inj           = svm->nested.ctl.event_inj;
448 	svm->vmcb->control.event_inj_err       = svm->nested.ctl.event_inj_err;
449 
450 	svm->vmcb->control.pause_filter_count  = svm->nested.ctl.pause_filter_count;
451 	svm->vmcb->control.pause_filter_thresh = svm->nested.ctl.pause_filter_thresh;
452 
453 	/* Enter Guest-Mode */
454 	enter_guest_mode(&svm->vcpu);
455 
456 	/*
457 	 * Merge guest and host intercepts - must be called  with vcpu in
458 	 * guest-mode to take affect here
459 	 */
460 	recalc_intercepts(svm);
461 
462 	vmcb_mark_all_dirty(svm->vmcb);
463 }
464 
465 int enter_svm_guest_mode(struct vcpu_svm *svm, u64 vmcb12_gpa,
466 			 struct vmcb *vmcb12)
467 {
468 	int ret;
469 
470 	trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb12_gpa,
471 			       vmcb12->save.rip,
472 			       vmcb12->control.int_ctl,
473 			       vmcb12->control.event_inj,
474 			       vmcb12->control.nested_ctl);
475 
476 	trace_kvm_nested_intercepts(vmcb12->control.intercepts[INTERCEPT_CR] & 0xffff,
477 				    vmcb12->control.intercepts[INTERCEPT_CR] >> 16,
478 				    vmcb12->control.intercepts[INTERCEPT_EXCEPTION],
479 				    vmcb12->control.intercepts[INTERCEPT_WORD3],
480 				    vmcb12->control.intercepts[INTERCEPT_WORD4],
481 				    vmcb12->control.intercepts[INTERCEPT_WORD5]);
482 
483 
484 	svm->nested.vmcb12_gpa = vmcb12_gpa;
485 	nested_prepare_vmcb_control(svm);
486 	nested_prepare_vmcb_save(svm, vmcb12);
487 
488 	ret = nested_svm_load_cr3(&svm->vcpu, vmcb12->save.cr3,
489 				  nested_npt_enabled(svm));
490 	if (ret)
491 		return ret;
492 
493 	if (!npt_enabled)
494 		svm->vcpu.arch.mmu->inject_page_fault = svm_inject_page_fault_nested;
495 
496 	svm_set_gif(svm, true);
497 
498 	return 0;
499 }
500 
501 int nested_svm_vmrun(struct vcpu_svm *svm)
502 {
503 	int ret;
504 	struct vmcb *vmcb12;
505 	struct vmcb *hsave = svm->nested.hsave;
506 	struct vmcb *vmcb = svm->vmcb;
507 	struct kvm_host_map map;
508 	u64 vmcb12_gpa;
509 
510 	if (is_smm(&svm->vcpu)) {
511 		kvm_queue_exception(&svm->vcpu, UD_VECTOR);
512 		return 1;
513 	}
514 
515 	vmcb12_gpa = svm->vmcb->save.rax;
516 	ret = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(vmcb12_gpa), &map);
517 	if (ret == -EINVAL) {
518 		kvm_inject_gp(&svm->vcpu, 0);
519 		return 1;
520 	} else if (ret) {
521 		return kvm_skip_emulated_instruction(&svm->vcpu);
522 	}
523 
524 	ret = kvm_skip_emulated_instruction(&svm->vcpu);
525 
526 	vmcb12 = map.hva;
527 
528 	if (WARN_ON_ONCE(!svm->nested.initialized))
529 		return -EINVAL;
530 
531 	load_nested_vmcb_control(svm, &vmcb12->control);
532 
533 	if (!nested_vmcb_check_save(svm, vmcb12) ||
534 	    !nested_vmcb_check_controls(&svm->nested.ctl)) {
535 		vmcb12->control.exit_code    = SVM_EXIT_ERR;
536 		vmcb12->control.exit_code_hi = 0;
537 		vmcb12->control.exit_info_1  = 0;
538 		vmcb12->control.exit_info_2  = 0;
539 		goto out;
540 	}
541 
542 
543 	/* Clear internal status */
544 	kvm_clear_exception_queue(&svm->vcpu);
545 	kvm_clear_interrupt_queue(&svm->vcpu);
546 
547 	/*
548 	 * Save the old vmcb, so we don't need to pick what we save, but can
549 	 * restore everything when a VMEXIT occurs
550 	 */
551 	hsave->save.es     = vmcb->save.es;
552 	hsave->save.cs     = vmcb->save.cs;
553 	hsave->save.ss     = vmcb->save.ss;
554 	hsave->save.ds     = vmcb->save.ds;
555 	hsave->save.gdtr   = vmcb->save.gdtr;
556 	hsave->save.idtr   = vmcb->save.idtr;
557 	hsave->save.efer   = svm->vcpu.arch.efer;
558 	hsave->save.cr0    = kvm_read_cr0(&svm->vcpu);
559 	hsave->save.cr4    = svm->vcpu.arch.cr4;
560 	hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
561 	hsave->save.rip    = kvm_rip_read(&svm->vcpu);
562 	hsave->save.rsp    = vmcb->save.rsp;
563 	hsave->save.rax    = vmcb->save.rax;
564 	if (npt_enabled)
565 		hsave->save.cr3    = vmcb->save.cr3;
566 	else
567 		hsave->save.cr3    = kvm_read_cr3(&svm->vcpu);
568 
569 	copy_vmcb_control_area(&hsave->control, &vmcb->control);
570 
571 	svm->nested.nested_run_pending = 1;
572 
573 	if (enter_svm_guest_mode(svm, vmcb12_gpa, vmcb12))
574 		goto out_exit_err;
575 
576 	if (nested_svm_vmrun_msrpm(svm))
577 		goto out;
578 
579 out_exit_err:
580 	svm->nested.nested_run_pending = 0;
581 
582 	svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
583 	svm->vmcb->control.exit_code_hi = 0;
584 	svm->vmcb->control.exit_info_1  = 0;
585 	svm->vmcb->control.exit_info_2  = 0;
586 
587 	nested_svm_vmexit(svm);
588 
589 out:
590 	kvm_vcpu_unmap(&svm->vcpu, &map, true);
591 
592 	return ret;
593 }
594 
595 void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
596 {
597 	to_vmcb->save.fs = from_vmcb->save.fs;
598 	to_vmcb->save.gs = from_vmcb->save.gs;
599 	to_vmcb->save.tr = from_vmcb->save.tr;
600 	to_vmcb->save.ldtr = from_vmcb->save.ldtr;
601 	to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
602 	to_vmcb->save.star = from_vmcb->save.star;
603 	to_vmcb->save.lstar = from_vmcb->save.lstar;
604 	to_vmcb->save.cstar = from_vmcb->save.cstar;
605 	to_vmcb->save.sfmask = from_vmcb->save.sfmask;
606 	to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
607 	to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
608 	to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
609 }
610 
611 int nested_svm_vmexit(struct vcpu_svm *svm)
612 {
613 	int rc;
614 	struct vmcb *vmcb12;
615 	struct vmcb *hsave = svm->nested.hsave;
616 	struct vmcb *vmcb = svm->vmcb;
617 	struct kvm_host_map map;
618 
619 	rc = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(svm->nested.vmcb12_gpa), &map);
620 	if (rc) {
621 		if (rc == -EINVAL)
622 			kvm_inject_gp(&svm->vcpu, 0);
623 		return 1;
624 	}
625 
626 	vmcb12 = map.hva;
627 
628 	/* Exit Guest-Mode */
629 	leave_guest_mode(&svm->vcpu);
630 	svm->nested.vmcb12_gpa = 0;
631 	WARN_ON_ONCE(svm->nested.nested_run_pending);
632 
633 	kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, &svm->vcpu);
634 
635 	/* in case we halted in L2 */
636 	svm->vcpu.arch.mp_state = KVM_MP_STATE_RUNNABLE;
637 
638 	/* Give the current vmcb to the guest */
639 
640 	vmcb12->save.es     = vmcb->save.es;
641 	vmcb12->save.cs     = vmcb->save.cs;
642 	vmcb12->save.ss     = vmcb->save.ss;
643 	vmcb12->save.ds     = vmcb->save.ds;
644 	vmcb12->save.gdtr   = vmcb->save.gdtr;
645 	vmcb12->save.idtr   = vmcb->save.idtr;
646 	vmcb12->save.efer   = svm->vcpu.arch.efer;
647 	vmcb12->save.cr0    = kvm_read_cr0(&svm->vcpu);
648 	vmcb12->save.cr3    = kvm_read_cr3(&svm->vcpu);
649 	vmcb12->save.cr2    = vmcb->save.cr2;
650 	vmcb12->save.cr4    = svm->vcpu.arch.cr4;
651 	vmcb12->save.rflags = kvm_get_rflags(&svm->vcpu);
652 	vmcb12->save.rip    = kvm_rip_read(&svm->vcpu);
653 	vmcb12->save.rsp    = kvm_rsp_read(&svm->vcpu);
654 	vmcb12->save.rax    = kvm_rax_read(&svm->vcpu);
655 	vmcb12->save.dr7    = vmcb->save.dr7;
656 	vmcb12->save.dr6    = svm->vcpu.arch.dr6;
657 	vmcb12->save.cpl    = vmcb->save.cpl;
658 
659 	vmcb12->control.int_state         = vmcb->control.int_state;
660 	vmcb12->control.exit_code         = vmcb->control.exit_code;
661 	vmcb12->control.exit_code_hi      = vmcb->control.exit_code_hi;
662 	vmcb12->control.exit_info_1       = vmcb->control.exit_info_1;
663 	vmcb12->control.exit_info_2       = vmcb->control.exit_info_2;
664 
665 	if (vmcb12->control.exit_code != SVM_EXIT_ERR)
666 		nested_vmcb_save_pending_event(svm, vmcb12);
667 
668 	if (svm->nrips_enabled)
669 		vmcb12->control.next_rip  = vmcb->control.next_rip;
670 
671 	vmcb12->control.int_ctl           = svm->nested.ctl.int_ctl;
672 	vmcb12->control.tlb_ctl           = svm->nested.ctl.tlb_ctl;
673 	vmcb12->control.event_inj         = svm->nested.ctl.event_inj;
674 	vmcb12->control.event_inj_err     = svm->nested.ctl.event_inj_err;
675 
676 	vmcb12->control.pause_filter_count =
677 		svm->vmcb->control.pause_filter_count;
678 	vmcb12->control.pause_filter_thresh =
679 		svm->vmcb->control.pause_filter_thresh;
680 
681 	/* Restore the original control entries */
682 	copy_vmcb_control_area(&vmcb->control, &hsave->control);
683 
684 	/* On vmexit the  GIF is set to false */
685 	svm_set_gif(svm, false);
686 
687 	svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset =
688 		svm->vcpu.arch.l1_tsc_offset;
689 
690 	svm->nested.ctl.nested_cr3 = 0;
691 
692 	/* Restore selected save entries */
693 	svm->vmcb->save.es = hsave->save.es;
694 	svm->vmcb->save.cs = hsave->save.cs;
695 	svm->vmcb->save.ss = hsave->save.ss;
696 	svm->vmcb->save.ds = hsave->save.ds;
697 	svm->vmcb->save.gdtr = hsave->save.gdtr;
698 	svm->vmcb->save.idtr = hsave->save.idtr;
699 	kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
700 	kvm_set_rflags(&svm->vcpu, hsave->save.rflags | X86_EFLAGS_FIXED);
701 	svm_set_efer(&svm->vcpu, hsave->save.efer);
702 	svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
703 	svm_set_cr4(&svm->vcpu, hsave->save.cr4);
704 	kvm_rax_write(&svm->vcpu, hsave->save.rax);
705 	kvm_rsp_write(&svm->vcpu, hsave->save.rsp);
706 	kvm_rip_write(&svm->vcpu, hsave->save.rip);
707 	svm->vmcb->save.dr7 = DR7_FIXED_1;
708 	svm->vmcb->save.cpl = 0;
709 	svm->vmcb->control.exit_int_info = 0;
710 
711 	vmcb_mark_all_dirty(svm->vmcb);
712 
713 	trace_kvm_nested_vmexit_inject(vmcb12->control.exit_code,
714 				       vmcb12->control.exit_info_1,
715 				       vmcb12->control.exit_info_2,
716 				       vmcb12->control.exit_int_info,
717 				       vmcb12->control.exit_int_info_err,
718 				       KVM_ISA_SVM);
719 
720 	kvm_vcpu_unmap(&svm->vcpu, &map, true);
721 
722 	nested_svm_uninit_mmu_context(&svm->vcpu);
723 
724 	rc = nested_svm_load_cr3(&svm->vcpu, hsave->save.cr3, false);
725 	if (rc)
726 		return 1;
727 
728 	if (npt_enabled)
729 		svm->vmcb->save.cr3 = hsave->save.cr3;
730 
731 	/*
732 	 * Drop what we picked up for L2 via svm_complete_interrupts() so it
733 	 * doesn't end up in L1.
734 	 */
735 	svm->vcpu.arch.nmi_injected = false;
736 	kvm_clear_exception_queue(&svm->vcpu);
737 	kvm_clear_interrupt_queue(&svm->vcpu);
738 
739 	return 0;
740 }
741 
742 int svm_allocate_nested(struct vcpu_svm *svm)
743 {
744 	struct page *hsave_page;
745 
746 	if (svm->nested.initialized)
747 		return 0;
748 
749 	hsave_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
750 	if (!hsave_page)
751 		return -ENOMEM;
752 	svm->nested.hsave = page_address(hsave_page);
753 
754 	svm->nested.msrpm = svm_vcpu_alloc_msrpm();
755 	if (!svm->nested.msrpm)
756 		goto err_free_hsave;
757 	svm_vcpu_init_msrpm(&svm->vcpu, svm->nested.msrpm);
758 
759 	svm->nested.initialized = true;
760 	return 0;
761 
762 err_free_hsave:
763 	__free_page(hsave_page);
764 	return -ENOMEM;
765 }
766 
767 void svm_free_nested(struct vcpu_svm *svm)
768 {
769 	if (!svm->nested.initialized)
770 		return;
771 
772 	svm_vcpu_free_msrpm(svm->nested.msrpm);
773 	svm->nested.msrpm = NULL;
774 
775 	__free_page(virt_to_page(svm->nested.hsave));
776 	svm->nested.hsave = NULL;
777 
778 	svm->nested.initialized = false;
779 }
780 
781 /*
782  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
783  */
784 void svm_leave_nested(struct vcpu_svm *svm)
785 {
786 	if (is_guest_mode(&svm->vcpu)) {
787 		struct vmcb *hsave = svm->nested.hsave;
788 		struct vmcb *vmcb = svm->vmcb;
789 
790 		svm->nested.nested_run_pending = 0;
791 		leave_guest_mode(&svm->vcpu);
792 		copy_vmcb_control_area(&vmcb->control, &hsave->control);
793 		nested_svm_uninit_mmu_context(&svm->vcpu);
794 		vmcb_mark_all_dirty(svm->vmcb);
795 	}
796 
797 	kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, &svm->vcpu);
798 }
799 
800 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
801 {
802 	u32 offset, msr, value;
803 	int write, mask;
804 
805 	if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
806 		return NESTED_EXIT_HOST;
807 
808 	msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
809 	offset = svm_msrpm_offset(msr);
810 	write  = svm->vmcb->control.exit_info_1 & 1;
811 	mask   = 1 << ((2 * (msr & 0xf)) + write);
812 
813 	if (offset == MSR_INVALID)
814 		return NESTED_EXIT_DONE;
815 
816 	/* Offset is in 32 bit units but need in 8 bit units */
817 	offset *= 4;
818 
819 	if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.ctl.msrpm_base_pa + offset, &value, 4))
820 		return NESTED_EXIT_DONE;
821 
822 	return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
823 }
824 
825 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
826 {
827 	unsigned port, size, iopm_len;
828 	u16 val, mask;
829 	u8 start_bit;
830 	u64 gpa;
831 
832 	if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_IOIO_PROT)))
833 		return NESTED_EXIT_HOST;
834 
835 	port = svm->vmcb->control.exit_info_1 >> 16;
836 	size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
837 		SVM_IOIO_SIZE_SHIFT;
838 	gpa  = svm->nested.ctl.iopm_base_pa + (port / 8);
839 	start_bit = port % 8;
840 	iopm_len = (start_bit + size > 8) ? 2 : 1;
841 	mask = (0xf >> (4 - size)) << start_bit;
842 	val = 0;
843 
844 	if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
845 		return NESTED_EXIT_DONE;
846 
847 	return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
848 }
849 
850 static int nested_svm_intercept(struct vcpu_svm *svm)
851 {
852 	u32 exit_code = svm->vmcb->control.exit_code;
853 	int vmexit = NESTED_EXIT_HOST;
854 
855 	switch (exit_code) {
856 	case SVM_EXIT_MSR:
857 		vmexit = nested_svm_exit_handled_msr(svm);
858 		break;
859 	case SVM_EXIT_IOIO:
860 		vmexit = nested_svm_intercept_ioio(svm);
861 		break;
862 	case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
863 		if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
864 			vmexit = NESTED_EXIT_DONE;
865 		break;
866 	}
867 	case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
868 		if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
869 			vmexit = NESTED_EXIT_DONE;
870 		break;
871 	}
872 	case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
873 		/*
874 		 * Host-intercepted exceptions have been checked already in
875 		 * nested_svm_exit_special.  There is nothing to do here,
876 		 * the vmexit is injected by svm_check_nested_events.
877 		 */
878 		vmexit = NESTED_EXIT_DONE;
879 		break;
880 	}
881 	case SVM_EXIT_ERR: {
882 		vmexit = NESTED_EXIT_DONE;
883 		break;
884 	}
885 	default: {
886 		if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
887 			vmexit = NESTED_EXIT_DONE;
888 	}
889 	}
890 
891 	return vmexit;
892 }
893 
894 int nested_svm_exit_handled(struct vcpu_svm *svm)
895 {
896 	int vmexit;
897 
898 	vmexit = nested_svm_intercept(svm);
899 
900 	if (vmexit == NESTED_EXIT_DONE)
901 		nested_svm_vmexit(svm);
902 
903 	return vmexit;
904 }
905 
906 int nested_svm_check_permissions(struct vcpu_svm *svm)
907 {
908 	if (!(svm->vcpu.arch.efer & EFER_SVME) ||
909 	    !is_paging(&svm->vcpu)) {
910 		kvm_queue_exception(&svm->vcpu, UD_VECTOR);
911 		return 1;
912 	}
913 
914 	if (svm->vmcb->save.cpl) {
915 		kvm_inject_gp(&svm->vcpu, 0);
916 		return 1;
917 	}
918 
919 	return 0;
920 }
921 
922 static bool nested_exit_on_exception(struct vcpu_svm *svm)
923 {
924 	unsigned int nr = svm->vcpu.arch.exception.nr;
925 
926 	return (svm->nested.ctl.intercepts[INTERCEPT_EXCEPTION] & BIT(nr));
927 }
928 
929 static void nested_svm_inject_exception_vmexit(struct vcpu_svm *svm)
930 {
931 	unsigned int nr = svm->vcpu.arch.exception.nr;
932 
933 	svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
934 	svm->vmcb->control.exit_code_hi = 0;
935 
936 	if (svm->vcpu.arch.exception.has_error_code)
937 		svm->vmcb->control.exit_info_1 = svm->vcpu.arch.exception.error_code;
938 
939 	/*
940 	 * EXITINFO2 is undefined for all exception intercepts other
941 	 * than #PF.
942 	 */
943 	if (nr == PF_VECTOR) {
944 		if (svm->vcpu.arch.exception.nested_apf)
945 			svm->vmcb->control.exit_info_2 = svm->vcpu.arch.apf.nested_apf_token;
946 		else if (svm->vcpu.arch.exception.has_payload)
947 			svm->vmcb->control.exit_info_2 = svm->vcpu.arch.exception.payload;
948 		else
949 			svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
950 	} else if (nr == DB_VECTOR) {
951 		/* See inject_pending_event.  */
952 		kvm_deliver_exception_payload(&svm->vcpu);
953 		if (svm->vcpu.arch.dr7 & DR7_GD) {
954 			svm->vcpu.arch.dr7 &= ~DR7_GD;
955 			kvm_update_dr7(&svm->vcpu);
956 		}
957 	} else
958 		WARN_ON(svm->vcpu.arch.exception.has_payload);
959 
960 	nested_svm_vmexit(svm);
961 }
962 
963 static void nested_svm_smi(struct vcpu_svm *svm)
964 {
965 	svm->vmcb->control.exit_code = SVM_EXIT_SMI;
966 	svm->vmcb->control.exit_info_1 = 0;
967 	svm->vmcb->control.exit_info_2 = 0;
968 
969 	nested_svm_vmexit(svm);
970 }
971 
972 static void nested_svm_nmi(struct vcpu_svm *svm)
973 {
974 	svm->vmcb->control.exit_code = SVM_EXIT_NMI;
975 	svm->vmcb->control.exit_info_1 = 0;
976 	svm->vmcb->control.exit_info_2 = 0;
977 
978 	nested_svm_vmexit(svm);
979 }
980 
981 static void nested_svm_intr(struct vcpu_svm *svm)
982 {
983 	trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
984 
985 	svm->vmcb->control.exit_code   = SVM_EXIT_INTR;
986 	svm->vmcb->control.exit_info_1 = 0;
987 	svm->vmcb->control.exit_info_2 = 0;
988 
989 	nested_svm_vmexit(svm);
990 }
991 
992 static inline bool nested_exit_on_init(struct vcpu_svm *svm)
993 {
994 	return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_INIT);
995 }
996 
997 static void nested_svm_init(struct vcpu_svm *svm)
998 {
999 	svm->vmcb->control.exit_code   = SVM_EXIT_INIT;
1000 	svm->vmcb->control.exit_info_1 = 0;
1001 	svm->vmcb->control.exit_info_2 = 0;
1002 
1003 	nested_svm_vmexit(svm);
1004 }
1005 
1006 
1007 static int svm_check_nested_events(struct kvm_vcpu *vcpu)
1008 {
1009 	struct vcpu_svm *svm = to_svm(vcpu);
1010 	bool block_nested_events =
1011 		kvm_event_needs_reinjection(vcpu) || svm->nested.nested_run_pending;
1012 	struct kvm_lapic *apic = vcpu->arch.apic;
1013 
1014 	if (lapic_in_kernel(vcpu) &&
1015 	    test_bit(KVM_APIC_INIT, &apic->pending_events)) {
1016 		if (block_nested_events)
1017 			return -EBUSY;
1018 		if (!nested_exit_on_init(svm))
1019 			return 0;
1020 		nested_svm_init(svm);
1021 		return 0;
1022 	}
1023 
1024 	if (vcpu->arch.exception.pending) {
1025 		if (block_nested_events)
1026                         return -EBUSY;
1027 		if (!nested_exit_on_exception(svm))
1028 			return 0;
1029 		nested_svm_inject_exception_vmexit(svm);
1030 		return 0;
1031 	}
1032 
1033 	if (vcpu->arch.smi_pending && !svm_smi_blocked(vcpu)) {
1034 		if (block_nested_events)
1035 			return -EBUSY;
1036 		if (!nested_exit_on_smi(svm))
1037 			return 0;
1038 		nested_svm_smi(svm);
1039 		return 0;
1040 	}
1041 
1042 	if (vcpu->arch.nmi_pending && !svm_nmi_blocked(vcpu)) {
1043 		if (block_nested_events)
1044 			return -EBUSY;
1045 		if (!nested_exit_on_nmi(svm))
1046 			return 0;
1047 		nested_svm_nmi(svm);
1048 		return 0;
1049 	}
1050 
1051 	if (kvm_cpu_has_interrupt(vcpu) && !svm_interrupt_blocked(vcpu)) {
1052 		if (block_nested_events)
1053 			return -EBUSY;
1054 		if (!nested_exit_on_intr(svm))
1055 			return 0;
1056 		nested_svm_intr(svm);
1057 		return 0;
1058 	}
1059 
1060 	return 0;
1061 }
1062 
1063 int nested_svm_exit_special(struct vcpu_svm *svm)
1064 {
1065 	u32 exit_code = svm->vmcb->control.exit_code;
1066 
1067 	switch (exit_code) {
1068 	case SVM_EXIT_INTR:
1069 	case SVM_EXIT_NMI:
1070 	case SVM_EXIT_NPF:
1071 		return NESTED_EXIT_HOST;
1072 	case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1073 		u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
1074 
1075 		if (get_host_vmcb(svm)->control.intercepts[INTERCEPT_EXCEPTION] &
1076 				excp_bits)
1077 			return NESTED_EXIT_HOST;
1078 		else if (exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR &&
1079 			 svm->vcpu.arch.apf.host_apf_flags)
1080 			/* Trap async PF even if not shadowing */
1081 			return NESTED_EXIT_HOST;
1082 		break;
1083 	}
1084 	default:
1085 		break;
1086 	}
1087 
1088 	return NESTED_EXIT_CONTINUE;
1089 }
1090 
1091 static int svm_get_nested_state(struct kvm_vcpu *vcpu,
1092 				struct kvm_nested_state __user *user_kvm_nested_state,
1093 				u32 user_data_size)
1094 {
1095 	struct vcpu_svm *svm;
1096 	struct kvm_nested_state kvm_state = {
1097 		.flags = 0,
1098 		.format = KVM_STATE_NESTED_FORMAT_SVM,
1099 		.size = sizeof(kvm_state),
1100 	};
1101 	struct vmcb __user *user_vmcb = (struct vmcb __user *)
1102 		&user_kvm_nested_state->data.svm[0];
1103 
1104 	if (!vcpu)
1105 		return kvm_state.size + KVM_STATE_NESTED_SVM_VMCB_SIZE;
1106 
1107 	svm = to_svm(vcpu);
1108 
1109 	if (user_data_size < kvm_state.size)
1110 		goto out;
1111 
1112 	/* First fill in the header and copy it out.  */
1113 	if (is_guest_mode(vcpu)) {
1114 		kvm_state.hdr.svm.vmcb_pa = svm->nested.vmcb12_gpa;
1115 		kvm_state.size += KVM_STATE_NESTED_SVM_VMCB_SIZE;
1116 		kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
1117 
1118 		if (svm->nested.nested_run_pending)
1119 			kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
1120 	}
1121 
1122 	if (gif_set(svm))
1123 		kvm_state.flags |= KVM_STATE_NESTED_GIF_SET;
1124 
1125 	if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
1126 		return -EFAULT;
1127 
1128 	if (!is_guest_mode(vcpu))
1129 		goto out;
1130 
1131 	/*
1132 	 * Copy over the full size of the VMCB rather than just the size
1133 	 * of the structs.
1134 	 */
1135 	if (clear_user(user_vmcb, KVM_STATE_NESTED_SVM_VMCB_SIZE))
1136 		return -EFAULT;
1137 	if (copy_to_user(&user_vmcb->control, &svm->nested.ctl,
1138 			 sizeof(user_vmcb->control)))
1139 		return -EFAULT;
1140 	if (copy_to_user(&user_vmcb->save, &svm->nested.hsave->save,
1141 			 sizeof(user_vmcb->save)))
1142 		return -EFAULT;
1143 
1144 out:
1145 	return kvm_state.size;
1146 }
1147 
1148 static int svm_set_nested_state(struct kvm_vcpu *vcpu,
1149 				struct kvm_nested_state __user *user_kvm_nested_state,
1150 				struct kvm_nested_state *kvm_state)
1151 {
1152 	struct vcpu_svm *svm = to_svm(vcpu);
1153 	struct vmcb *hsave = svm->nested.hsave;
1154 	struct vmcb __user *user_vmcb = (struct vmcb __user *)
1155 		&user_kvm_nested_state->data.svm[0];
1156 	struct vmcb_control_area *ctl;
1157 	struct vmcb_save_area *save;
1158 	int ret;
1159 	u32 cr0;
1160 
1161 	BUILD_BUG_ON(sizeof(struct vmcb_control_area) + sizeof(struct vmcb_save_area) >
1162 		     KVM_STATE_NESTED_SVM_VMCB_SIZE);
1163 
1164 	if (kvm_state->format != KVM_STATE_NESTED_FORMAT_SVM)
1165 		return -EINVAL;
1166 
1167 	if (kvm_state->flags & ~(KVM_STATE_NESTED_GUEST_MODE |
1168 				 KVM_STATE_NESTED_RUN_PENDING |
1169 				 KVM_STATE_NESTED_GIF_SET))
1170 		return -EINVAL;
1171 
1172 	/*
1173 	 * If in guest mode, vcpu->arch.efer actually refers to the L2 guest's
1174 	 * EFER.SVME, but EFER.SVME still has to be 1 for VMRUN to succeed.
1175 	 */
1176 	if (!(vcpu->arch.efer & EFER_SVME)) {
1177 		/* GIF=1 and no guest mode are required if SVME=0.  */
1178 		if (kvm_state->flags != KVM_STATE_NESTED_GIF_SET)
1179 			return -EINVAL;
1180 	}
1181 
1182 	/* SMM temporarily disables SVM, so we cannot be in guest mode.  */
1183 	if (is_smm(vcpu) && (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
1184 		return -EINVAL;
1185 
1186 	if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE)) {
1187 		svm_leave_nested(svm);
1188 		svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
1189 		return 0;
1190 	}
1191 
1192 	if (!page_address_valid(vcpu, kvm_state->hdr.svm.vmcb_pa))
1193 		return -EINVAL;
1194 	if (kvm_state->size < sizeof(*kvm_state) + KVM_STATE_NESTED_SVM_VMCB_SIZE)
1195 		return -EINVAL;
1196 
1197 	ret  = -ENOMEM;
1198 	ctl  = kzalloc(sizeof(*ctl),  GFP_KERNEL);
1199 	save = kzalloc(sizeof(*save), GFP_KERNEL);
1200 	if (!ctl || !save)
1201 		goto out_free;
1202 
1203 	ret = -EFAULT;
1204 	if (copy_from_user(ctl, &user_vmcb->control, sizeof(*ctl)))
1205 		goto out_free;
1206 	if (copy_from_user(save, &user_vmcb->save, sizeof(*save)))
1207 		goto out_free;
1208 
1209 	ret = -EINVAL;
1210 	if (!nested_vmcb_check_controls(ctl))
1211 		goto out_free;
1212 
1213 	/*
1214 	 * Processor state contains L2 state.  Check that it is
1215 	 * valid for guest mode (see nested_vmcb_checks).
1216 	 */
1217 	cr0 = kvm_read_cr0(vcpu);
1218         if (((cr0 & X86_CR0_CD) == 0) && (cr0 & X86_CR0_NW))
1219 		goto out_free;
1220 
1221 	/*
1222 	 * Validate host state saved from before VMRUN (see
1223 	 * nested_svm_check_permissions).
1224 	 * TODO: validate reserved bits for all saved state.
1225 	 */
1226 	if (!(save->cr0 & X86_CR0_PG))
1227 		goto out_free;
1228 	if (!(save->efer & EFER_SVME))
1229 		goto out_free;
1230 
1231 	/*
1232 	 * All checks done, we can enter guest mode.  L1 control fields
1233 	 * come from the nested save state.  Guest state is already
1234 	 * in the registers, the save area of the nested state instead
1235 	 * contains saved L1 state.
1236 	 */
1237 
1238 	svm->nested.nested_run_pending =
1239 		!!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
1240 
1241 	copy_vmcb_control_area(&hsave->control, &svm->vmcb->control);
1242 	hsave->save = *save;
1243 
1244 	svm->nested.vmcb12_gpa = kvm_state->hdr.svm.vmcb_pa;
1245 	load_nested_vmcb_control(svm, ctl);
1246 	nested_prepare_vmcb_control(svm);
1247 
1248 	kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
1249 	ret = 0;
1250 out_free:
1251 	kfree(save);
1252 	kfree(ctl);
1253 
1254 	return ret;
1255 }
1256 
1257 struct kvm_x86_nested_ops svm_nested_ops = {
1258 	.check_events = svm_check_nested_events,
1259 	.get_nested_state_pages = svm_get_nested_state_pages,
1260 	.get_state = svm_get_nested_state,
1261 	.set_state = svm_set_nested_state,
1262 };
1263