xref: /openbmc/linux/arch/x86/kvm/svm/nested.c (revision d9565bf4)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Kernel-based Virtual Machine driver for Linux
4  *
5  * AMD SVM support
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
9  *
10  * Authors:
11  *   Yaniv Kamay  <yaniv@qumranet.com>
12  *   Avi Kivity   <avi@qumranet.com>
13  */
14 
15 #define pr_fmt(fmt) "SVM: " fmt
16 
17 #include <linux/kvm_types.h>
18 #include <linux/kvm_host.h>
19 #include <linux/kernel.h>
20 
21 #include <asm/msr-index.h>
22 #include <asm/debugreg.h>
23 
24 #include "kvm_emulate.h"
25 #include "trace.h"
26 #include "mmu.h"
27 #include "x86.h"
28 #include "cpuid.h"
29 #include "lapic.h"
30 #include "svm.h"
31 
32 #define CC KVM_NESTED_VMENTER_CONSISTENCY_CHECK
33 
34 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
35 				       struct x86_exception *fault)
36 {
37 	struct vcpu_svm *svm = to_svm(vcpu);
38 
39 	if (svm->vmcb->control.exit_code != SVM_EXIT_NPF) {
40 		/*
41 		 * TODO: track the cause of the nested page fault, and
42 		 * correctly fill in the high bits of exit_info_1.
43 		 */
44 		svm->vmcb->control.exit_code = SVM_EXIT_NPF;
45 		svm->vmcb->control.exit_code_hi = 0;
46 		svm->vmcb->control.exit_info_1 = (1ULL << 32);
47 		svm->vmcb->control.exit_info_2 = fault->address;
48 	}
49 
50 	svm->vmcb->control.exit_info_1 &= ~0xffffffffULL;
51 	svm->vmcb->control.exit_info_1 |= fault->error_code;
52 
53 	nested_svm_vmexit(svm);
54 }
55 
56 static void svm_inject_page_fault_nested(struct kvm_vcpu *vcpu, struct x86_exception *fault)
57 {
58        struct vcpu_svm *svm = to_svm(vcpu);
59        WARN_ON(!is_guest_mode(vcpu));
60 
61 	if (vmcb12_is_intercept(&svm->nested.ctl,
62 				INTERCEPT_EXCEPTION_OFFSET + PF_VECTOR) &&
63 	    !svm->nested.nested_run_pending) {
64                svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + PF_VECTOR;
65                svm->vmcb->control.exit_code_hi = 0;
66                svm->vmcb->control.exit_info_1 = fault->error_code;
67                svm->vmcb->control.exit_info_2 = fault->address;
68                nested_svm_vmexit(svm);
69        } else {
70                kvm_inject_page_fault(vcpu, fault);
71        }
72 }
73 
74 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
75 {
76 	struct vcpu_svm *svm = to_svm(vcpu);
77 	u64 cr3 = svm->nested.ctl.nested_cr3;
78 	u64 pdpte;
79 	int ret;
80 
81 	ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(cr3), &pdpte,
82 				       offset_in_page(cr3) + index * 8, 8);
83 	if (ret)
84 		return 0;
85 	return pdpte;
86 }
87 
88 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
89 {
90 	struct vcpu_svm *svm = to_svm(vcpu);
91 
92 	return svm->nested.ctl.nested_cr3;
93 }
94 
95 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
96 {
97 	struct vcpu_svm *svm = to_svm(vcpu);
98 
99 	WARN_ON(mmu_is_nested(vcpu));
100 
101 	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
102 
103 	/*
104 	 * The NPT format depends on L1's CR4 and EFER, which is in vmcb01.  Note,
105 	 * when called via KVM_SET_NESTED_STATE, that state may _not_ match current
106 	 * vCPU state.  CR0.WP is explicitly ignored, while CR0.PG is required.
107 	 */
108 	kvm_init_shadow_npt_mmu(vcpu, X86_CR0_PG, svm->vmcb01.ptr->save.cr4,
109 				svm->vmcb01.ptr->save.efer,
110 				svm->nested.ctl.nested_cr3);
111 	vcpu->arch.mmu->get_guest_pgd     = nested_svm_get_tdp_cr3;
112 	vcpu->arch.mmu->get_pdptr         = nested_svm_get_tdp_pdptr;
113 	vcpu->arch.mmu->inject_page_fault = nested_svm_inject_npf_exit;
114 	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
115 }
116 
117 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
118 {
119 	vcpu->arch.mmu = &vcpu->arch.root_mmu;
120 	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
121 }
122 
123 void recalc_intercepts(struct vcpu_svm *svm)
124 {
125 	struct vmcb_control_area *c, *h;
126 	struct vmcb_ctrl_area_cached *g;
127 	unsigned int i;
128 
129 	vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
130 
131 	if (!is_guest_mode(&svm->vcpu))
132 		return;
133 
134 	c = &svm->vmcb->control;
135 	h = &svm->vmcb01.ptr->control;
136 	g = &svm->nested.ctl;
137 
138 	for (i = 0; i < MAX_INTERCEPT; i++)
139 		c->intercepts[i] = h->intercepts[i];
140 
141 	if (g->int_ctl & V_INTR_MASKING_MASK) {
142 		/* We only want the cr8 intercept bits of L1 */
143 		vmcb_clr_intercept(c, INTERCEPT_CR8_READ);
144 		vmcb_clr_intercept(c, INTERCEPT_CR8_WRITE);
145 
146 		/*
147 		 * Once running L2 with HF_VINTR_MASK, EFLAGS.IF does not
148 		 * affect any interrupt we may want to inject; therefore,
149 		 * interrupt window vmexits are irrelevant to L0.
150 		 */
151 		vmcb_clr_intercept(c, INTERCEPT_VINTR);
152 	}
153 
154 	/* We don't want to see VMMCALLs from a nested guest */
155 	vmcb_clr_intercept(c, INTERCEPT_VMMCALL);
156 
157 	for (i = 0; i < MAX_INTERCEPT; i++)
158 		c->intercepts[i] |= g->intercepts[i];
159 
160 	/* If SMI is not intercepted, ignore guest SMI intercept as well  */
161 	if (!intercept_smi)
162 		vmcb_clr_intercept(c, INTERCEPT_SMI);
163 
164 	vmcb_set_intercept(c, INTERCEPT_VMLOAD);
165 	vmcb_set_intercept(c, INTERCEPT_VMSAVE);
166 }
167 
168 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
169 {
170 	/*
171 	 * This function merges the msr permission bitmaps of kvm and the
172 	 * nested vmcb. It is optimized in that it only merges the parts where
173 	 * the kvm msr permission bitmap may contain zero bits
174 	 */
175 	int i;
176 
177 	if (!(vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
178 		return true;
179 
180 	for (i = 0; i < MSRPM_OFFSETS; i++) {
181 		u32 value, p;
182 		u64 offset;
183 
184 		if (msrpm_offsets[i] == 0xffffffff)
185 			break;
186 
187 		p      = msrpm_offsets[i];
188 		offset = svm->nested.ctl.msrpm_base_pa + (p * 4);
189 
190 		if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
191 			return false;
192 
193 		svm->nested.msrpm[p] = svm->msrpm[p] | value;
194 	}
195 
196 	svm->vmcb->control.msrpm_base_pa = __sme_set(__pa(svm->nested.msrpm));
197 
198 	return true;
199 }
200 
201 /*
202  * Bits 11:0 of bitmap address are ignored by hardware
203  */
204 static bool nested_svm_check_bitmap_pa(struct kvm_vcpu *vcpu, u64 pa, u32 size)
205 {
206 	u64 addr = PAGE_ALIGN(pa);
207 
208 	return kvm_vcpu_is_legal_gpa(vcpu, addr) &&
209 	    kvm_vcpu_is_legal_gpa(vcpu, addr + size - 1);
210 }
211 
212 static bool nested_svm_check_tlb_ctl(struct kvm_vcpu *vcpu, u8 tlb_ctl)
213 {
214 	/* Nested FLUSHBYASID is not supported yet.  */
215 	switch(tlb_ctl) {
216 		case TLB_CONTROL_DO_NOTHING:
217 		case TLB_CONTROL_FLUSH_ALL_ASID:
218 			return true;
219 		default:
220 			return false;
221 	}
222 }
223 
224 static bool __nested_vmcb_check_controls(struct kvm_vcpu *vcpu,
225 					 struct vmcb_ctrl_area_cached *control)
226 {
227 	if (CC(!vmcb12_is_intercept(control, INTERCEPT_VMRUN)))
228 		return false;
229 
230 	if (CC(control->asid == 0))
231 		return false;
232 
233 	if (CC((control->nested_ctl & SVM_NESTED_CTL_NP_ENABLE) && !npt_enabled))
234 		return false;
235 
236 	if (CC(!nested_svm_check_bitmap_pa(vcpu, control->msrpm_base_pa,
237 					   MSRPM_SIZE)))
238 		return false;
239 	if (CC(!nested_svm_check_bitmap_pa(vcpu, control->iopm_base_pa,
240 					   IOPM_SIZE)))
241 		return false;
242 
243 	if (CC(!nested_svm_check_tlb_ctl(vcpu, control->tlb_ctl)))
244 		return false;
245 
246 	return true;
247 }
248 
249 /* Common checks that apply to both L1 and L2 state.  */
250 static bool __nested_vmcb_check_save(struct kvm_vcpu *vcpu,
251 				     struct vmcb_save_area_cached *save)
252 {
253 	if (CC(!(save->efer & EFER_SVME)))
254 		return false;
255 
256 	if (CC((save->cr0 & X86_CR0_CD) == 0 && (save->cr0 & X86_CR0_NW)) ||
257 	    CC(save->cr0 & ~0xffffffffULL))
258 		return false;
259 
260 	if (CC(!kvm_dr6_valid(save->dr6)) || CC(!kvm_dr7_valid(save->dr7)))
261 		return false;
262 
263 	/*
264 	 * These checks are also performed by KVM_SET_SREGS,
265 	 * except that EFER.LMA is not checked by SVM against
266 	 * CR0.PG && EFER.LME.
267 	 */
268 	if ((save->efer & EFER_LME) && (save->cr0 & X86_CR0_PG)) {
269 		if (CC(!(save->cr4 & X86_CR4_PAE)) ||
270 		    CC(!(save->cr0 & X86_CR0_PE)) ||
271 		    CC(kvm_vcpu_is_illegal_gpa(vcpu, save->cr3)))
272 			return false;
273 	}
274 
275 	if (CC(!kvm_is_valid_cr4(vcpu, save->cr4)))
276 		return false;
277 
278 	if (CC(!kvm_valid_efer(vcpu, save->efer)))
279 		return false;
280 
281 	return true;
282 }
283 
284 static bool nested_vmcb_check_save(struct kvm_vcpu *vcpu)
285 {
286 	struct vcpu_svm *svm = to_svm(vcpu);
287 	struct vmcb_save_area_cached *save = &svm->nested.save;
288 
289 	return __nested_vmcb_check_save(vcpu, save);
290 }
291 
292 static bool nested_vmcb_check_controls(struct kvm_vcpu *vcpu)
293 {
294 	struct vcpu_svm *svm = to_svm(vcpu);
295 	struct vmcb_ctrl_area_cached *ctl = &svm->nested.ctl;
296 
297 	return __nested_vmcb_check_controls(vcpu, ctl);
298 }
299 
300 static
301 void __nested_copy_vmcb_control_to_cache(struct vmcb_ctrl_area_cached *to,
302 					 struct vmcb_control_area *from)
303 {
304 	unsigned int i;
305 
306 	for (i = 0; i < MAX_INTERCEPT; i++)
307 		to->intercepts[i] = from->intercepts[i];
308 
309 	to->iopm_base_pa        = from->iopm_base_pa;
310 	to->msrpm_base_pa       = from->msrpm_base_pa;
311 	to->tsc_offset          = from->tsc_offset;
312 	to->tlb_ctl             = from->tlb_ctl;
313 	to->int_ctl             = from->int_ctl;
314 	to->int_vector          = from->int_vector;
315 	to->int_state           = from->int_state;
316 	to->exit_code           = from->exit_code;
317 	to->exit_code_hi        = from->exit_code_hi;
318 	to->exit_info_1         = from->exit_info_1;
319 	to->exit_info_2         = from->exit_info_2;
320 	to->exit_int_info       = from->exit_int_info;
321 	to->exit_int_info_err   = from->exit_int_info_err;
322 	to->nested_ctl          = from->nested_ctl;
323 	to->event_inj           = from->event_inj;
324 	to->event_inj_err       = from->event_inj_err;
325 	to->nested_cr3          = from->nested_cr3;
326 	to->virt_ext            = from->virt_ext;
327 	to->pause_filter_count  = from->pause_filter_count;
328 	to->pause_filter_thresh = from->pause_filter_thresh;
329 
330 	/* Copy asid here because nested_vmcb_check_controls will check it.  */
331 	to->asid           = from->asid;
332 	to->msrpm_base_pa &= ~0x0fffULL;
333 	to->iopm_base_pa  &= ~0x0fffULL;
334 }
335 
336 void nested_copy_vmcb_control_to_cache(struct vcpu_svm *svm,
337 				       struct vmcb_control_area *control)
338 {
339 	__nested_copy_vmcb_control_to_cache(&svm->nested.ctl, control);
340 }
341 
342 static void __nested_copy_vmcb_save_to_cache(struct vmcb_save_area_cached *to,
343 					     struct vmcb_save_area *from)
344 {
345 	/*
346 	 * Copy only fields that are validated, as we need them
347 	 * to avoid TOC/TOU races.
348 	 */
349 	to->efer = from->efer;
350 	to->cr0 = from->cr0;
351 	to->cr3 = from->cr3;
352 	to->cr4 = from->cr4;
353 
354 	to->dr6 = from->dr6;
355 	to->dr7 = from->dr7;
356 }
357 
358 void nested_copy_vmcb_save_to_cache(struct vcpu_svm *svm,
359 				    struct vmcb_save_area *save)
360 {
361 	__nested_copy_vmcb_save_to_cache(&svm->nested.save, save);
362 }
363 
364 /*
365  * Synchronize fields that are written by the processor, so that
366  * they can be copied back into the vmcb12.
367  */
368 void nested_sync_control_from_vmcb02(struct vcpu_svm *svm)
369 {
370 	u32 mask;
371 	svm->nested.ctl.event_inj      = svm->vmcb->control.event_inj;
372 	svm->nested.ctl.event_inj_err  = svm->vmcb->control.event_inj_err;
373 
374 	/* Only a few fields of int_ctl are written by the processor.  */
375 	mask = V_IRQ_MASK | V_TPR_MASK;
376 	if (!(svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK) &&
377 	    svm_is_intercept(svm, INTERCEPT_VINTR)) {
378 		/*
379 		 * In order to request an interrupt window, L0 is usurping
380 		 * svm->vmcb->control.int_ctl and possibly setting V_IRQ
381 		 * even if it was clear in L1's VMCB.  Restoring it would be
382 		 * wrong.  However, in this case V_IRQ will remain true until
383 		 * interrupt_window_interception calls svm_clear_vintr and
384 		 * restores int_ctl.  We can just leave it aside.
385 		 */
386 		mask &= ~V_IRQ_MASK;
387 	}
388 	svm->nested.ctl.int_ctl        &= ~mask;
389 	svm->nested.ctl.int_ctl        |= svm->vmcb->control.int_ctl & mask;
390 }
391 
392 /*
393  * Transfer any event that L0 or L1 wanted to inject into L2 to
394  * EXIT_INT_INFO.
395  */
396 static void nested_save_pending_event_to_vmcb12(struct vcpu_svm *svm,
397 						struct vmcb *vmcb12)
398 {
399 	struct kvm_vcpu *vcpu = &svm->vcpu;
400 	u32 exit_int_info = 0;
401 	unsigned int nr;
402 
403 	if (vcpu->arch.exception.injected) {
404 		nr = vcpu->arch.exception.nr;
405 		exit_int_info = nr | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT;
406 
407 		if (vcpu->arch.exception.has_error_code) {
408 			exit_int_info |= SVM_EVTINJ_VALID_ERR;
409 			vmcb12->control.exit_int_info_err =
410 				vcpu->arch.exception.error_code;
411 		}
412 
413 	} else if (vcpu->arch.nmi_injected) {
414 		exit_int_info = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
415 
416 	} else if (vcpu->arch.interrupt.injected) {
417 		nr = vcpu->arch.interrupt.nr;
418 		exit_int_info = nr | SVM_EVTINJ_VALID;
419 
420 		if (vcpu->arch.interrupt.soft)
421 			exit_int_info |= SVM_EVTINJ_TYPE_SOFT;
422 		else
423 			exit_int_info |= SVM_EVTINJ_TYPE_INTR;
424 	}
425 
426 	vmcb12->control.exit_int_info = exit_int_info;
427 }
428 
429 static inline bool nested_npt_enabled(struct vcpu_svm *svm)
430 {
431 	return svm->nested.ctl.nested_ctl & SVM_NESTED_CTL_NP_ENABLE;
432 }
433 
434 static void nested_svm_transition_tlb_flush(struct kvm_vcpu *vcpu)
435 {
436 	/*
437 	 * TODO: optimize unconditional TLB flush/MMU sync.  A partial list of
438 	 * things to fix before this can be conditional:
439 	 *
440 	 *  - Flush TLBs for both L1 and L2 remote TLB flush
441 	 *  - Honor L1's request to flush an ASID on nested VMRUN
442 	 *  - Sync nested NPT MMU on VMRUN that flushes L2's ASID[*]
443 	 *  - Don't crush a pending TLB flush in vmcb02 on nested VMRUN
444 	 *  - Flush L1's ASID on KVM_REQ_TLB_FLUSH_GUEST
445 	 *
446 	 * [*] Unlike nested EPT, SVM's ASID management can invalidate nested
447 	 *     NPT guest-physical mappings on VMRUN.
448 	 */
449 	kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
450 	kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
451 }
452 
453 /*
454  * Load guest's/host's cr3 on nested vmentry or vmexit. @nested_npt is true
455  * if we are emulating VM-Entry into a guest with NPT enabled.
456  */
457 static int nested_svm_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3,
458 			       bool nested_npt, bool reload_pdptrs)
459 {
460 	if (CC(kvm_vcpu_is_illegal_gpa(vcpu, cr3)))
461 		return -EINVAL;
462 
463 	if (reload_pdptrs && !nested_npt && is_pae_paging(vcpu) &&
464 	    CC(!load_pdptrs(vcpu, cr3)))
465 		return -EINVAL;
466 
467 	if (!nested_npt)
468 		kvm_mmu_new_pgd(vcpu, cr3);
469 
470 	vcpu->arch.cr3 = cr3;
471 
472 	/* Re-initialize the MMU, e.g. to pick up CR4 MMU role changes. */
473 	kvm_init_mmu(vcpu);
474 
475 	return 0;
476 }
477 
478 void nested_vmcb02_compute_g_pat(struct vcpu_svm *svm)
479 {
480 	if (!svm->nested.vmcb02.ptr)
481 		return;
482 
483 	/* FIXME: merge g_pat from vmcb01 and vmcb12.  */
484 	svm->nested.vmcb02.ptr->save.g_pat = svm->vmcb01.ptr->save.g_pat;
485 }
486 
487 static void nested_vmcb02_prepare_save(struct vcpu_svm *svm, struct vmcb *vmcb12)
488 {
489 	bool new_vmcb12 = false;
490 
491 	nested_vmcb02_compute_g_pat(svm);
492 
493 	/* Load the nested guest state */
494 	if (svm->nested.vmcb12_gpa != svm->nested.last_vmcb12_gpa) {
495 		new_vmcb12 = true;
496 		svm->nested.last_vmcb12_gpa = svm->nested.vmcb12_gpa;
497 	}
498 
499 	if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_SEG))) {
500 		svm->vmcb->save.es = vmcb12->save.es;
501 		svm->vmcb->save.cs = vmcb12->save.cs;
502 		svm->vmcb->save.ss = vmcb12->save.ss;
503 		svm->vmcb->save.ds = vmcb12->save.ds;
504 		svm->vmcb->save.cpl = vmcb12->save.cpl;
505 		vmcb_mark_dirty(svm->vmcb, VMCB_SEG);
506 	}
507 
508 	if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_DT))) {
509 		svm->vmcb->save.gdtr = vmcb12->save.gdtr;
510 		svm->vmcb->save.idtr = vmcb12->save.idtr;
511 		vmcb_mark_dirty(svm->vmcb, VMCB_DT);
512 	}
513 
514 	kvm_set_rflags(&svm->vcpu, vmcb12->save.rflags | X86_EFLAGS_FIXED);
515 
516 	svm_set_efer(&svm->vcpu, svm->nested.save.efer);
517 
518 	svm_set_cr0(&svm->vcpu, svm->nested.save.cr0);
519 	svm_set_cr4(&svm->vcpu, svm->nested.save.cr4);
520 
521 	svm->vcpu.arch.cr2 = vmcb12->save.cr2;
522 
523 	kvm_rax_write(&svm->vcpu, vmcb12->save.rax);
524 	kvm_rsp_write(&svm->vcpu, vmcb12->save.rsp);
525 	kvm_rip_write(&svm->vcpu, vmcb12->save.rip);
526 
527 	/* In case we don't even reach vcpu_run, the fields are not updated */
528 	svm->vmcb->save.rax = vmcb12->save.rax;
529 	svm->vmcb->save.rsp = vmcb12->save.rsp;
530 	svm->vmcb->save.rip = vmcb12->save.rip;
531 
532 	/* These bits will be set properly on the first execution when new_vmc12 is true */
533 	if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_DR))) {
534 		svm->vmcb->save.dr7 = svm->nested.save.dr7 | DR7_FIXED_1;
535 		svm->vcpu.arch.dr6  = svm->nested.save.dr6 | DR6_ACTIVE_LOW;
536 		vmcb_mark_dirty(svm->vmcb, VMCB_DR);
537 	}
538 }
539 
540 static void nested_vmcb02_prepare_control(struct vcpu_svm *svm)
541 {
542 	const u32 int_ctl_vmcb01_bits =
543 		V_INTR_MASKING_MASK | V_GIF_MASK | V_GIF_ENABLE_MASK;
544 
545 	const u32 int_ctl_vmcb12_bits = V_TPR_MASK | V_IRQ_INJECTION_BITS_MASK;
546 
547 	struct kvm_vcpu *vcpu = &svm->vcpu;
548 
549 	/*
550 	 * Filled at exit: exit_code, exit_code_hi, exit_info_1, exit_info_2,
551 	 * exit_int_info, exit_int_info_err, next_rip, insn_len, insn_bytes.
552 	 */
553 
554 	/*
555 	 * Also covers avic_vapic_bar, avic_backing_page, avic_logical_id,
556 	 * avic_physical_id.
557 	 */
558 	WARN_ON(kvm_apicv_activated(svm->vcpu.kvm));
559 
560 	/* Copied from vmcb01.  msrpm_base can be overwritten later.  */
561 	svm->vmcb->control.nested_ctl = svm->vmcb01.ptr->control.nested_ctl;
562 	svm->vmcb->control.iopm_base_pa = svm->vmcb01.ptr->control.iopm_base_pa;
563 	svm->vmcb->control.msrpm_base_pa = svm->vmcb01.ptr->control.msrpm_base_pa;
564 
565 	/* Done at vmrun: asid.  */
566 
567 	/* Also overwritten later if necessary.  */
568 	svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
569 
570 	/* nested_cr3.  */
571 	if (nested_npt_enabled(svm))
572 		nested_svm_init_mmu_context(vcpu);
573 
574 	vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
575 			vcpu->arch.l1_tsc_offset,
576 			svm->nested.ctl.tsc_offset,
577 			svm->tsc_ratio_msr);
578 
579 	svm->vmcb->control.tsc_offset = vcpu->arch.tsc_offset;
580 
581 	if (svm->tsc_ratio_msr != kvm_default_tsc_scaling_ratio) {
582 		WARN_ON(!svm->tsc_scaling_enabled);
583 		nested_svm_update_tsc_ratio_msr(vcpu);
584 	}
585 
586 	svm->vmcb->control.int_ctl             =
587 		(svm->nested.ctl.int_ctl & int_ctl_vmcb12_bits) |
588 		(svm->vmcb01.ptr->control.int_ctl & int_ctl_vmcb01_bits);
589 
590 	svm->vmcb->control.int_vector          = svm->nested.ctl.int_vector;
591 	svm->vmcb->control.int_state           = svm->nested.ctl.int_state;
592 	svm->vmcb->control.event_inj           = svm->nested.ctl.event_inj;
593 	svm->vmcb->control.event_inj_err       = svm->nested.ctl.event_inj_err;
594 
595 	nested_svm_transition_tlb_flush(vcpu);
596 
597 	/* Enter Guest-Mode */
598 	enter_guest_mode(vcpu);
599 
600 	/*
601 	 * Merge guest and host intercepts - must be called with vcpu in
602 	 * guest-mode to take effect.
603 	 */
604 	recalc_intercepts(svm);
605 }
606 
607 static void nested_svm_copy_common_state(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
608 {
609 	/*
610 	 * Some VMCB state is shared between L1 and L2 and thus has to be
611 	 * moved at the time of nested vmrun and vmexit.
612 	 *
613 	 * VMLOAD/VMSAVE state would also belong in this category, but KVM
614 	 * always performs VMLOAD and VMSAVE from the VMCB01.
615 	 */
616 	to_vmcb->save.spec_ctrl = from_vmcb->save.spec_ctrl;
617 }
618 
619 int enter_svm_guest_mode(struct kvm_vcpu *vcpu, u64 vmcb12_gpa,
620 			 struct vmcb *vmcb12, bool from_vmrun)
621 {
622 	struct vcpu_svm *svm = to_svm(vcpu);
623 	int ret;
624 
625 	trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb12_gpa,
626 			       vmcb12->save.rip,
627 			       vmcb12->control.int_ctl,
628 			       vmcb12->control.event_inj,
629 			       vmcb12->control.nested_ctl);
630 
631 	trace_kvm_nested_intercepts(vmcb12->control.intercepts[INTERCEPT_CR] & 0xffff,
632 				    vmcb12->control.intercepts[INTERCEPT_CR] >> 16,
633 				    vmcb12->control.intercepts[INTERCEPT_EXCEPTION],
634 				    vmcb12->control.intercepts[INTERCEPT_WORD3],
635 				    vmcb12->control.intercepts[INTERCEPT_WORD4],
636 				    vmcb12->control.intercepts[INTERCEPT_WORD5]);
637 
638 
639 	svm->nested.vmcb12_gpa = vmcb12_gpa;
640 
641 	WARN_ON(svm->vmcb == svm->nested.vmcb02.ptr);
642 
643 	nested_svm_copy_common_state(svm->vmcb01.ptr, svm->nested.vmcb02.ptr);
644 
645 	svm_switch_vmcb(svm, &svm->nested.vmcb02);
646 	nested_vmcb02_prepare_control(svm);
647 	nested_vmcb02_prepare_save(svm, vmcb12);
648 
649 	ret = nested_svm_load_cr3(&svm->vcpu, svm->nested.save.cr3,
650 				  nested_npt_enabled(svm), from_vmrun);
651 	if (ret)
652 		return ret;
653 
654 	if (!npt_enabled)
655 		vcpu->arch.mmu->inject_page_fault = svm_inject_page_fault_nested;
656 
657 	if (!from_vmrun)
658 		kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
659 
660 	svm_set_gif(svm, true);
661 
662 	return 0;
663 }
664 
665 int nested_svm_vmrun(struct kvm_vcpu *vcpu)
666 {
667 	struct vcpu_svm *svm = to_svm(vcpu);
668 	int ret;
669 	struct vmcb *vmcb12;
670 	struct kvm_host_map map;
671 	u64 vmcb12_gpa;
672 
673 	if (!svm->nested.hsave_msr) {
674 		kvm_inject_gp(vcpu, 0);
675 		return 1;
676 	}
677 
678 	if (is_smm(vcpu)) {
679 		kvm_queue_exception(vcpu, UD_VECTOR);
680 		return 1;
681 	}
682 
683 	vmcb12_gpa = svm->vmcb->save.rax;
684 	ret = kvm_vcpu_map(vcpu, gpa_to_gfn(vmcb12_gpa), &map);
685 	if (ret == -EINVAL) {
686 		kvm_inject_gp(vcpu, 0);
687 		return 1;
688 	} else if (ret) {
689 		return kvm_skip_emulated_instruction(vcpu);
690 	}
691 
692 	ret = kvm_skip_emulated_instruction(vcpu);
693 
694 	vmcb12 = map.hva;
695 
696 	if (WARN_ON_ONCE(!svm->nested.initialized))
697 		return -EINVAL;
698 
699 	nested_copy_vmcb_control_to_cache(svm, &vmcb12->control);
700 	nested_copy_vmcb_save_to_cache(svm, &vmcb12->save);
701 
702 	if (!nested_vmcb_check_save(vcpu) ||
703 	    !nested_vmcb_check_controls(vcpu)) {
704 		vmcb12->control.exit_code    = SVM_EXIT_ERR;
705 		vmcb12->control.exit_code_hi = 0;
706 		vmcb12->control.exit_info_1  = 0;
707 		vmcb12->control.exit_info_2  = 0;
708 		goto out;
709 	}
710 
711 	/*
712 	 * Since vmcb01 is not in use, we can use it to store some of the L1
713 	 * state.
714 	 */
715 	svm->vmcb01.ptr->save.efer   = vcpu->arch.efer;
716 	svm->vmcb01.ptr->save.cr0    = kvm_read_cr0(vcpu);
717 	svm->vmcb01.ptr->save.cr4    = vcpu->arch.cr4;
718 	svm->vmcb01.ptr->save.rflags = kvm_get_rflags(vcpu);
719 	svm->vmcb01.ptr->save.rip    = kvm_rip_read(vcpu);
720 
721 	if (!npt_enabled)
722 		svm->vmcb01.ptr->save.cr3 = kvm_read_cr3(vcpu);
723 
724 	svm->nested.nested_run_pending = 1;
725 
726 	if (enter_svm_guest_mode(vcpu, vmcb12_gpa, vmcb12, true))
727 		goto out_exit_err;
728 
729 	if (nested_svm_vmrun_msrpm(svm))
730 		goto out;
731 
732 out_exit_err:
733 	svm->nested.nested_run_pending = 0;
734 
735 	svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
736 	svm->vmcb->control.exit_code_hi = 0;
737 	svm->vmcb->control.exit_info_1  = 0;
738 	svm->vmcb->control.exit_info_2  = 0;
739 
740 	nested_svm_vmexit(svm);
741 
742 out:
743 	kvm_vcpu_unmap(vcpu, &map, true);
744 
745 	return ret;
746 }
747 
748 /* Copy state save area fields which are handled by VMRUN */
749 void svm_copy_vmrun_state(struct vmcb_save_area *to_save,
750 			  struct vmcb_save_area *from_save)
751 {
752 	to_save->es = from_save->es;
753 	to_save->cs = from_save->cs;
754 	to_save->ss = from_save->ss;
755 	to_save->ds = from_save->ds;
756 	to_save->gdtr = from_save->gdtr;
757 	to_save->idtr = from_save->idtr;
758 	to_save->rflags = from_save->rflags | X86_EFLAGS_FIXED;
759 	to_save->efer = from_save->efer;
760 	to_save->cr0 = from_save->cr0;
761 	to_save->cr3 = from_save->cr3;
762 	to_save->cr4 = from_save->cr4;
763 	to_save->rax = from_save->rax;
764 	to_save->rsp = from_save->rsp;
765 	to_save->rip = from_save->rip;
766 	to_save->cpl = 0;
767 }
768 
769 void svm_copy_vmloadsave_state(struct vmcb *to_vmcb, struct vmcb *from_vmcb)
770 {
771 	to_vmcb->save.fs = from_vmcb->save.fs;
772 	to_vmcb->save.gs = from_vmcb->save.gs;
773 	to_vmcb->save.tr = from_vmcb->save.tr;
774 	to_vmcb->save.ldtr = from_vmcb->save.ldtr;
775 	to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
776 	to_vmcb->save.star = from_vmcb->save.star;
777 	to_vmcb->save.lstar = from_vmcb->save.lstar;
778 	to_vmcb->save.cstar = from_vmcb->save.cstar;
779 	to_vmcb->save.sfmask = from_vmcb->save.sfmask;
780 	to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
781 	to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
782 	to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
783 }
784 
785 int nested_svm_vmexit(struct vcpu_svm *svm)
786 {
787 	struct kvm_vcpu *vcpu = &svm->vcpu;
788 	struct vmcb *vmcb12;
789 	struct vmcb *vmcb = svm->vmcb;
790 	struct kvm_host_map map;
791 	int rc;
792 
793 	/* Triple faults in L2 should never escape. */
794 	WARN_ON_ONCE(kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu));
795 
796 	rc = kvm_vcpu_map(vcpu, gpa_to_gfn(svm->nested.vmcb12_gpa), &map);
797 	if (rc) {
798 		if (rc == -EINVAL)
799 			kvm_inject_gp(vcpu, 0);
800 		return 1;
801 	}
802 
803 	vmcb12 = map.hva;
804 
805 	/* Exit Guest-Mode */
806 	leave_guest_mode(vcpu);
807 	svm->nested.vmcb12_gpa = 0;
808 	WARN_ON_ONCE(svm->nested.nested_run_pending);
809 
810 	kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
811 
812 	/* in case we halted in L2 */
813 	svm->vcpu.arch.mp_state = KVM_MP_STATE_RUNNABLE;
814 
815 	/* Give the current vmcb to the guest */
816 
817 	vmcb12->save.es     = vmcb->save.es;
818 	vmcb12->save.cs     = vmcb->save.cs;
819 	vmcb12->save.ss     = vmcb->save.ss;
820 	vmcb12->save.ds     = vmcb->save.ds;
821 	vmcb12->save.gdtr   = vmcb->save.gdtr;
822 	vmcb12->save.idtr   = vmcb->save.idtr;
823 	vmcb12->save.efer   = svm->vcpu.arch.efer;
824 	vmcb12->save.cr0    = kvm_read_cr0(vcpu);
825 	vmcb12->save.cr3    = kvm_read_cr3(vcpu);
826 	vmcb12->save.cr2    = vmcb->save.cr2;
827 	vmcb12->save.cr4    = svm->vcpu.arch.cr4;
828 	vmcb12->save.rflags = kvm_get_rflags(vcpu);
829 	vmcb12->save.rip    = kvm_rip_read(vcpu);
830 	vmcb12->save.rsp    = kvm_rsp_read(vcpu);
831 	vmcb12->save.rax    = kvm_rax_read(vcpu);
832 	vmcb12->save.dr7    = vmcb->save.dr7;
833 	vmcb12->save.dr6    = svm->vcpu.arch.dr6;
834 	vmcb12->save.cpl    = vmcb->save.cpl;
835 
836 	vmcb12->control.int_state         = vmcb->control.int_state;
837 	vmcb12->control.exit_code         = vmcb->control.exit_code;
838 	vmcb12->control.exit_code_hi      = vmcb->control.exit_code_hi;
839 	vmcb12->control.exit_info_1       = vmcb->control.exit_info_1;
840 	vmcb12->control.exit_info_2       = vmcb->control.exit_info_2;
841 
842 	if (vmcb12->control.exit_code != SVM_EXIT_ERR)
843 		nested_save_pending_event_to_vmcb12(svm, vmcb12);
844 
845 	if (svm->nrips_enabled)
846 		vmcb12->control.next_rip  = vmcb->control.next_rip;
847 
848 	vmcb12->control.int_ctl           = svm->nested.ctl.int_ctl;
849 	vmcb12->control.tlb_ctl           = svm->nested.ctl.tlb_ctl;
850 	vmcb12->control.event_inj         = svm->nested.ctl.event_inj;
851 	vmcb12->control.event_inj_err     = svm->nested.ctl.event_inj_err;
852 
853 	nested_svm_copy_common_state(svm->nested.vmcb02.ptr, svm->vmcb01.ptr);
854 
855 	svm_switch_vmcb(svm, &svm->vmcb01);
856 
857 	/*
858 	 * On vmexit the  GIF is set to false and
859 	 * no event can be injected in L1.
860 	 */
861 	svm_set_gif(svm, false);
862 	svm->vmcb->control.exit_int_info = 0;
863 
864 	svm->vcpu.arch.tsc_offset = svm->vcpu.arch.l1_tsc_offset;
865 	if (svm->vmcb->control.tsc_offset != svm->vcpu.arch.tsc_offset) {
866 		svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset;
867 		vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
868 	}
869 
870 	if (svm->tsc_ratio_msr != kvm_default_tsc_scaling_ratio) {
871 		WARN_ON(!svm->tsc_scaling_enabled);
872 		vcpu->arch.tsc_scaling_ratio = vcpu->arch.l1_tsc_scaling_ratio;
873 		svm_write_tsc_multiplier(vcpu, vcpu->arch.tsc_scaling_ratio);
874 	}
875 
876 	svm->nested.ctl.nested_cr3 = 0;
877 
878 	/*
879 	 * Restore processor state that had been saved in vmcb01
880 	 */
881 	kvm_set_rflags(vcpu, svm->vmcb->save.rflags);
882 	svm_set_efer(vcpu, svm->vmcb->save.efer);
883 	svm_set_cr0(vcpu, svm->vmcb->save.cr0 | X86_CR0_PE);
884 	svm_set_cr4(vcpu, svm->vmcb->save.cr4);
885 	kvm_rax_write(vcpu, svm->vmcb->save.rax);
886 	kvm_rsp_write(vcpu, svm->vmcb->save.rsp);
887 	kvm_rip_write(vcpu, svm->vmcb->save.rip);
888 
889 	svm->vcpu.arch.dr7 = DR7_FIXED_1;
890 	kvm_update_dr7(&svm->vcpu);
891 
892 	trace_kvm_nested_vmexit_inject(vmcb12->control.exit_code,
893 				       vmcb12->control.exit_info_1,
894 				       vmcb12->control.exit_info_2,
895 				       vmcb12->control.exit_int_info,
896 				       vmcb12->control.exit_int_info_err,
897 				       KVM_ISA_SVM);
898 
899 	kvm_vcpu_unmap(vcpu, &map, true);
900 
901 	nested_svm_transition_tlb_flush(vcpu);
902 
903 	nested_svm_uninit_mmu_context(vcpu);
904 
905 	rc = nested_svm_load_cr3(vcpu, svm->vmcb->save.cr3, false, true);
906 	if (rc)
907 		return 1;
908 
909 	/*
910 	 * Drop what we picked up for L2 via svm_complete_interrupts() so it
911 	 * doesn't end up in L1.
912 	 */
913 	svm->vcpu.arch.nmi_injected = false;
914 	kvm_clear_exception_queue(vcpu);
915 	kvm_clear_interrupt_queue(vcpu);
916 
917 	/*
918 	 * If we are here following the completion of a VMRUN that
919 	 * is being single-stepped, queue the pending #DB intercept
920 	 * right now so that it an be accounted for before we execute
921 	 * L1's next instruction.
922 	 */
923 	if (unlikely(svm->vmcb->save.rflags & X86_EFLAGS_TF))
924 		kvm_queue_exception(&(svm->vcpu), DB_VECTOR);
925 
926 	return 0;
927 }
928 
929 static void nested_svm_triple_fault(struct kvm_vcpu *vcpu)
930 {
931 	nested_svm_simple_vmexit(to_svm(vcpu), SVM_EXIT_SHUTDOWN);
932 }
933 
934 int svm_allocate_nested(struct vcpu_svm *svm)
935 {
936 	struct page *vmcb02_page;
937 
938 	if (svm->nested.initialized)
939 		return 0;
940 
941 	vmcb02_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
942 	if (!vmcb02_page)
943 		return -ENOMEM;
944 	svm->nested.vmcb02.ptr = page_address(vmcb02_page);
945 	svm->nested.vmcb02.pa = __sme_set(page_to_pfn(vmcb02_page) << PAGE_SHIFT);
946 
947 	svm->nested.msrpm = svm_vcpu_alloc_msrpm();
948 	if (!svm->nested.msrpm)
949 		goto err_free_vmcb02;
950 	svm_vcpu_init_msrpm(&svm->vcpu, svm->nested.msrpm);
951 
952 	svm->nested.initialized = true;
953 	return 0;
954 
955 err_free_vmcb02:
956 	__free_page(vmcb02_page);
957 	return -ENOMEM;
958 }
959 
960 void svm_free_nested(struct vcpu_svm *svm)
961 {
962 	if (!svm->nested.initialized)
963 		return;
964 
965 	svm_vcpu_free_msrpm(svm->nested.msrpm);
966 	svm->nested.msrpm = NULL;
967 
968 	__free_page(virt_to_page(svm->nested.vmcb02.ptr));
969 	svm->nested.vmcb02.ptr = NULL;
970 
971 	/*
972 	 * When last_vmcb12_gpa matches the current vmcb12 gpa,
973 	 * some vmcb12 fields are not loaded if they are marked clean
974 	 * in the vmcb12, since in this case they are up to date already.
975 	 *
976 	 * When the vmcb02 is freed, this optimization becomes invalid.
977 	 */
978 	svm->nested.last_vmcb12_gpa = INVALID_GPA;
979 
980 	svm->nested.initialized = false;
981 }
982 
983 /*
984  * Forcibly leave nested mode in order to be able to reset the VCPU later on.
985  */
986 void svm_leave_nested(struct kvm_vcpu *vcpu)
987 {
988 	struct vcpu_svm *svm = to_svm(vcpu);
989 
990 	if (is_guest_mode(vcpu)) {
991 		svm->nested.nested_run_pending = 0;
992 		svm->nested.vmcb12_gpa = INVALID_GPA;
993 
994 		leave_guest_mode(vcpu);
995 
996 		svm_switch_vmcb(svm, &svm->vmcb01);
997 
998 		nested_svm_uninit_mmu_context(vcpu);
999 		vmcb_mark_all_dirty(svm->vmcb);
1000 	}
1001 
1002 	kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
1003 }
1004 
1005 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
1006 {
1007 	u32 offset, msr, value;
1008 	int write, mask;
1009 
1010 	if (!(vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
1011 		return NESTED_EXIT_HOST;
1012 
1013 	msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
1014 	offset = svm_msrpm_offset(msr);
1015 	write  = svm->vmcb->control.exit_info_1 & 1;
1016 	mask   = 1 << ((2 * (msr & 0xf)) + write);
1017 
1018 	if (offset == MSR_INVALID)
1019 		return NESTED_EXIT_DONE;
1020 
1021 	/* Offset is in 32 bit units but need in 8 bit units */
1022 	offset *= 4;
1023 
1024 	if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.ctl.msrpm_base_pa + offset, &value, 4))
1025 		return NESTED_EXIT_DONE;
1026 
1027 	return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
1028 }
1029 
1030 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
1031 {
1032 	unsigned port, size, iopm_len;
1033 	u16 val, mask;
1034 	u8 start_bit;
1035 	u64 gpa;
1036 
1037 	if (!(vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_IOIO_PROT)))
1038 		return NESTED_EXIT_HOST;
1039 
1040 	port = svm->vmcb->control.exit_info_1 >> 16;
1041 	size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
1042 		SVM_IOIO_SIZE_SHIFT;
1043 	gpa  = svm->nested.ctl.iopm_base_pa + (port / 8);
1044 	start_bit = port % 8;
1045 	iopm_len = (start_bit + size > 8) ? 2 : 1;
1046 	mask = (0xf >> (4 - size)) << start_bit;
1047 	val = 0;
1048 
1049 	if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
1050 		return NESTED_EXIT_DONE;
1051 
1052 	return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
1053 }
1054 
1055 static int nested_svm_intercept(struct vcpu_svm *svm)
1056 {
1057 	u32 exit_code = svm->vmcb->control.exit_code;
1058 	int vmexit = NESTED_EXIT_HOST;
1059 
1060 	switch (exit_code) {
1061 	case SVM_EXIT_MSR:
1062 		vmexit = nested_svm_exit_handled_msr(svm);
1063 		break;
1064 	case SVM_EXIT_IOIO:
1065 		vmexit = nested_svm_intercept_ioio(svm);
1066 		break;
1067 	case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
1068 		if (vmcb12_is_intercept(&svm->nested.ctl, exit_code))
1069 			vmexit = NESTED_EXIT_DONE;
1070 		break;
1071 	}
1072 	case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
1073 		if (vmcb12_is_intercept(&svm->nested.ctl, exit_code))
1074 			vmexit = NESTED_EXIT_DONE;
1075 		break;
1076 	}
1077 	case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1078 		/*
1079 		 * Host-intercepted exceptions have been checked already in
1080 		 * nested_svm_exit_special.  There is nothing to do here,
1081 		 * the vmexit is injected by svm_check_nested_events.
1082 		 */
1083 		vmexit = NESTED_EXIT_DONE;
1084 		break;
1085 	}
1086 	case SVM_EXIT_ERR: {
1087 		vmexit = NESTED_EXIT_DONE;
1088 		break;
1089 	}
1090 	default: {
1091 		if (vmcb12_is_intercept(&svm->nested.ctl, exit_code))
1092 			vmexit = NESTED_EXIT_DONE;
1093 	}
1094 	}
1095 
1096 	return vmexit;
1097 }
1098 
1099 int nested_svm_exit_handled(struct vcpu_svm *svm)
1100 {
1101 	int vmexit;
1102 
1103 	vmexit = nested_svm_intercept(svm);
1104 
1105 	if (vmexit == NESTED_EXIT_DONE)
1106 		nested_svm_vmexit(svm);
1107 
1108 	return vmexit;
1109 }
1110 
1111 int nested_svm_check_permissions(struct kvm_vcpu *vcpu)
1112 {
1113 	if (!(vcpu->arch.efer & EFER_SVME) || !is_paging(vcpu)) {
1114 		kvm_queue_exception(vcpu, UD_VECTOR);
1115 		return 1;
1116 	}
1117 
1118 	if (to_svm(vcpu)->vmcb->save.cpl) {
1119 		kvm_inject_gp(vcpu, 0);
1120 		return 1;
1121 	}
1122 
1123 	return 0;
1124 }
1125 
1126 static bool nested_exit_on_exception(struct vcpu_svm *svm)
1127 {
1128 	unsigned int nr = svm->vcpu.arch.exception.nr;
1129 
1130 	return (svm->nested.ctl.intercepts[INTERCEPT_EXCEPTION] & BIT(nr));
1131 }
1132 
1133 static void nested_svm_inject_exception_vmexit(struct vcpu_svm *svm)
1134 {
1135 	unsigned int nr = svm->vcpu.arch.exception.nr;
1136 
1137 	svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
1138 	svm->vmcb->control.exit_code_hi = 0;
1139 
1140 	if (svm->vcpu.arch.exception.has_error_code)
1141 		svm->vmcb->control.exit_info_1 = svm->vcpu.arch.exception.error_code;
1142 
1143 	/*
1144 	 * EXITINFO2 is undefined for all exception intercepts other
1145 	 * than #PF.
1146 	 */
1147 	if (nr == PF_VECTOR) {
1148 		if (svm->vcpu.arch.exception.nested_apf)
1149 			svm->vmcb->control.exit_info_2 = svm->vcpu.arch.apf.nested_apf_token;
1150 		else if (svm->vcpu.arch.exception.has_payload)
1151 			svm->vmcb->control.exit_info_2 = svm->vcpu.arch.exception.payload;
1152 		else
1153 			svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
1154 	} else if (nr == DB_VECTOR) {
1155 		/* See inject_pending_event.  */
1156 		kvm_deliver_exception_payload(&svm->vcpu);
1157 		if (svm->vcpu.arch.dr7 & DR7_GD) {
1158 			svm->vcpu.arch.dr7 &= ~DR7_GD;
1159 			kvm_update_dr7(&svm->vcpu);
1160 		}
1161 	} else
1162 		WARN_ON(svm->vcpu.arch.exception.has_payload);
1163 
1164 	nested_svm_vmexit(svm);
1165 }
1166 
1167 static inline bool nested_exit_on_init(struct vcpu_svm *svm)
1168 {
1169 	return vmcb12_is_intercept(&svm->nested.ctl, INTERCEPT_INIT);
1170 }
1171 
1172 static int svm_check_nested_events(struct kvm_vcpu *vcpu)
1173 {
1174 	struct vcpu_svm *svm = to_svm(vcpu);
1175 	bool block_nested_events =
1176 		kvm_event_needs_reinjection(vcpu) || svm->nested.nested_run_pending;
1177 	struct kvm_lapic *apic = vcpu->arch.apic;
1178 
1179 	if (lapic_in_kernel(vcpu) &&
1180 	    test_bit(KVM_APIC_INIT, &apic->pending_events)) {
1181 		if (block_nested_events)
1182 			return -EBUSY;
1183 		if (!nested_exit_on_init(svm))
1184 			return 0;
1185 		nested_svm_simple_vmexit(svm, SVM_EXIT_INIT);
1186 		return 0;
1187 	}
1188 
1189 	if (vcpu->arch.exception.pending) {
1190 		/*
1191 		 * Only a pending nested run can block a pending exception.
1192 		 * Otherwise an injected NMI/interrupt should either be
1193 		 * lost or delivered to the nested hypervisor in the EXITINTINFO
1194 		 * vmcb field, while delivering the pending exception.
1195 		 */
1196 		if (svm->nested.nested_run_pending)
1197                         return -EBUSY;
1198 		if (!nested_exit_on_exception(svm))
1199 			return 0;
1200 		nested_svm_inject_exception_vmexit(svm);
1201 		return 0;
1202 	}
1203 
1204 	if (vcpu->arch.smi_pending && !svm_smi_blocked(vcpu)) {
1205 		if (block_nested_events)
1206 			return -EBUSY;
1207 		if (!nested_exit_on_smi(svm))
1208 			return 0;
1209 		nested_svm_simple_vmexit(svm, SVM_EXIT_SMI);
1210 		return 0;
1211 	}
1212 
1213 	if (vcpu->arch.nmi_pending && !svm_nmi_blocked(vcpu)) {
1214 		if (block_nested_events)
1215 			return -EBUSY;
1216 		if (!nested_exit_on_nmi(svm))
1217 			return 0;
1218 		nested_svm_simple_vmexit(svm, SVM_EXIT_NMI);
1219 		return 0;
1220 	}
1221 
1222 	if (kvm_cpu_has_interrupt(vcpu) && !svm_interrupt_blocked(vcpu)) {
1223 		if (block_nested_events)
1224 			return -EBUSY;
1225 		if (!nested_exit_on_intr(svm))
1226 			return 0;
1227 		trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
1228 		nested_svm_simple_vmexit(svm, SVM_EXIT_INTR);
1229 		return 0;
1230 	}
1231 
1232 	return 0;
1233 }
1234 
1235 int nested_svm_exit_special(struct vcpu_svm *svm)
1236 {
1237 	u32 exit_code = svm->vmcb->control.exit_code;
1238 
1239 	switch (exit_code) {
1240 	case SVM_EXIT_INTR:
1241 	case SVM_EXIT_NMI:
1242 	case SVM_EXIT_NPF:
1243 		return NESTED_EXIT_HOST;
1244 	case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1245 		u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
1246 
1247 		if (svm->vmcb01.ptr->control.intercepts[INTERCEPT_EXCEPTION] &
1248 		    excp_bits)
1249 			return NESTED_EXIT_HOST;
1250 		else if (exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR &&
1251 			 svm->vcpu.arch.apf.host_apf_flags)
1252 			/* Trap async PF even if not shadowing */
1253 			return NESTED_EXIT_HOST;
1254 		break;
1255 	}
1256 	default:
1257 		break;
1258 	}
1259 
1260 	return NESTED_EXIT_CONTINUE;
1261 }
1262 
1263 void nested_svm_update_tsc_ratio_msr(struct kvm_vcpu *vcpu)
1264 {
1265 	struct vcpu_svm *svm = to_svm(vcpu);
1266 
1267 	vcpu->arch.tsc_scaling_ratio =
1268 		kvm_calc_nested_tsc_multiplier(vcpu->arch.l1_tsc_scaling_ratio,
1269 					       svm->tsc_ratio_msr);
1270 	svm_write_tsc_multiplier(vcpu, vcpu->arch.tsc_scaling_ratio);
1271 }
1272 
1273 /* Inverse operation of nested_copy_vmcb_control_to_cache(). asid is copied too. */
1274 static void nested_copy_vmcb_cache_to_control(struct vmcb_control_area *dst,
1275 					      struct vmcb_ctrl_area_cached *from)
1276 {
1277 	unsigned int i;
1278 
1279 	memset(dst, 0, sizeof(struct vmcb_control_area));
1280 
1281 	for (i = 0; i < MAX_INTERCEPT; i++)
1282 		dst->intercepts[i] = from->intercepts[i];
1283 
1284 	dst->iopm_base_pa         = from->iopm_base_pa;
1285 	dst->msrpm_base_pa        = from->msrpm_base_pa;
1286 	dst->tsc_offset           = from->tsc_offset;
1287 	dst->asid                 = from->asid;
1288 	dst->tlb_ctl              = from->tlb_ctl;
1289 	dst->int_ctl              = from->int_ctl;
1290 	dst->int_vector           = from->int_vector;
1291 	dst->int_state            = from->int_state;
1292 	dst->exit_code            = from->exit_code;
1293 	dst->exit_code_hi         = from->exit_code_hi;
1294 	dst->exit_info_1          = from->exit_info_1;
1295 	dst->exit_info_2          = from->exit_info_2;
1296 	dst->exit_int_info        = from->exit_int_info;
1297 	dst->exit_int_info_err    = from->exit_int_info_err;
1298 	dst->nested_ctl           = from->nested_ctl;
1299 	dst->event_inj            = from->event_inj;
1300 	dst->event_inj_err        = from->event_inj_err;
1301 	dst->nested_cr3           = from->nested_cr3;
1302 	dst->virt_ext              = from->virt_ext;
1303 	dst->pause_filter_count   = from->pause_filter_count;
1304 	dst->pause_filter_thresh  = from->pause_filter_thresh;
1305 }
1306 
1307 static int svm_get_nested_state(struct kvm_vcpu *vcpu,
1308 				struct kvm_nested_state __user *user_kvm_nested_state,
1309 				u32 user_data_size)
1310 {
1311 	struct vcpu_svm *svm;
1312 	struct vmcb_control_area *ctl;
1313 	unsigned long r;
1314 	struct kvm_nested_state kvm_state = {
1315 		.flags = 0,
1316 		.format = KVM_STATE_NESTED_FORMAT_SVM,
1317 		.size = sizeof(kvm_state),
1318 	};
1319 	struct vmcb __user *user_vmcb = (struct vmcb __user *)
1320 		&user_kvm_nested_state->data.svm[0];
1321 
1322 	if (!vcpu)
1323 		return kvm_state.size + KVM_STATE_NESTED_SVM_VMCB_SIZE;
1324 
1325 	svm = to_svm(vcpu);
1326 
1327 	if (user_data_size < kvm_state.size)
1328 		goto out;
1329 
1330 	/* First fill in the header and copy it out.  */
1331 	if (is_guest_mode(vcpu)) {
1332 		kvm_state.hdr.svm.vmcb_pa = svm->nested.vmcb12_gpa;
1333 		kvm_state.size += KVM_STATE_NESTED_SVM_VMCB_SIZE;
1334 		kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
1335 
1336 		if (svm->nested.nested_run_pending)
1337 			kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
1338 	}
1339 
1340 	if (gif_set(svm))
1341 		kvm_state.flags |= KVM_STATE_NESTED_GIF_SET;
1342 
1343 	if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
1344 		return -EFAULT;
1345 
1346 	if (!is_guest_mode(vcpu))
1347 		goto out;
1348 
1349 	/*
1350 	 * Copy over the full size of the VMCB rather than just the size
1351 	 * of the structs.
1352 	 */
1353 	if (clear_user(user_vmcb, KVM_STATE_NESTED_SVM_VMCB_SIZE))
1354 		return -EFAULT;
1355 
1356 	ctl = kzalloc(sizeof(*ctl), GFP_KERNEL);
1357 	if (!ctl)
1358 		return -ENOMEM;
1359 
1360 	nested_copy_vmcb_cache_to_control(ctl, &svm->nested.ctl);
1361 	r = copy_to_user(&user_vmcb->control, ctl,
1362 			 sizeof(user_vmcb->control));
1363 	kfree(ctl);
1364 	if (r)
1365 		return -EFAULT;
1366 
1367 	if (copy_to_user(&user_vmcb->save, &svm->vmcb01.ptr->save,
1368 			 sizeof(user_vmcb->save)))
1369 		return -EFAULT;
1370 out:
1371 	return kvm_state.size;
1372 }
1373 
1374 static int svm_set_nested_state(struct kvm_vcpu *vcpu,
1375 				struct kvm_nested_state __user *user_kvm_nested_state,
1376 				struct kvm_nested_state *kvm_state)
1377 {
1378 	struct vcpu_svm *svm = to_svm(vcpu);
1379 	struct vmcb __user *user_vmcb = (struct vmcb __user *)
1380 		&user_kvm_nested_state->data.svm[0];
1381 	struct vmcb_control_area *ctl;
1382 	struct vmcb_save_area *save;
1383 	struct vmcb_save_area_cached save_cached;
1384 	struct vmcb_ctrl_area_cached ctl_cached;
1385 	unsigned long cr0;
1386 	int ret;
1387 
1388 	BUILD_BUG_ON(sizeof(struct vmcb_control_area) + sizeof(struct vmcb_save_area) >
1389 		     KVM_STATE_NESTED_SVM_VMCB_SIZE);
1390 
1391 	if (kvm_state->format != KVM_STATE_NESTED_FORMAT_SVM)
1392 		return -EINVAL;
1393 
1394 	if (kvm_state->flags & ~(KVM_STATE_NESTED_GUEST_MODE |
1395 				 KVM_STATE_NESTED_RUN_PENDING |
1396 				 KVM_STATE_NESTED_GIF_SET))
1397 		return -EINVAL;
1398 
1399 	/*
1400 	 * If in guest mode, vcpu->arch.efer actually refers to the L2 guest's
1401 	 * EFER.SVME, but EFER.SVME still has to be 1 for VMRUN to succeed.
1402 	 */
1403 	if (!(vcpu->arch.efer & EFER_SVME)) {
1404 		/* GIF=1 and no guest mode are required if SVME=0.  */
1405 		if (kvm_state->flags != KVM_STATE_NESTED_GIF_SET)
1406 			return -EINVAL;
1407 	}
1408 
1409 	/* SMM temporarily disables SVM, so we cannot be in guest mode.  */
1410 	if (is_smm(vcpu) && (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
1411 		return -EINVAL;
1412 
1413 	if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE)) {
1414 		svm_leave_nested(vcpu);
1415 		svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
1416 		return 0;
1417 	}
1418 
1419 	if (!page_address_valid(vcpu, kvm_state->hdr.svm.vmcb_pa))
1420 		return -EINVAL;
1421 	if (kvm_state->size < sizeof(*kvm_state) + KVM_STATE_NESTED_SVM_VMCB_SIZE)
1422 		return -EINVAL;
1423 
1424 	ret  = -ENOMEM;
1425 	ctl  = kzalloc(sizeof(*ctl),  GFP_KERNEL_ACCOUNT);
1426 	save = kzalloc(sizeof(*save), GFP_KERNEL_ACCOUNT);
1427 	if (!ctl || !save)
1428 		goto out_free;
1429 
1430 	ret = -EFAULT;
1431 	if (copy_from_user(ctl, &user_vmcb->control, sizeof(*ctl)))
1432 		goto out_free;
1433 	if (copy_from_user(save, &user_vmcb->save, sizeof(*save)))
1434 		goto out_free;
1435 
1436 	ret = -EINVAL;
1437 	__nested_copy_vmcb_control_to_cache(&ctl_cached, ctl);
1438 	if (!__nested_vmcb_check_controls(vcpu, &ctl_cached))
1439 		goto out_free;
1440 
1441 	/*
1442 	 * Processor state contains L2 state.  Check that it is
1443 	 * valid for guest mode (see nested_vmcb_check_save).
1444 	 */
1445 	cr0 = kvm_read_cr0(vcpu);
1446         if (((cr0 & X86_CR0_CD) == 0) && (cr0 & X86_CR0_NW))
1447 		goto out_free;
1448 
1449 	/*
1450 	 * Validate host state saved from before VMRUN (see
1451 	 * nested_svm_check_permissions).
1452 	 */
1453 	__nested_copy_vmcb_save_to_cache(&save_cached, save);
1454 	if (!(save->cr0 & X86_CR0_PG) ||
1455 	    !(save->cr0 & X86_CR0_PE) ||
1456 	    (save->rflags & X86_EFLAGS_VM) ||
1457 	    !__nested_vmcb_check_save(vcpu, &save_cached))
1458 		goto out_free;
1459 
1460 	/*
1461 	 * While the nested guest CR3 is already checked and set by
1462 	 * KVM_SET_SREGS, it was set when nested state was yet loaded,
1463 	 * thus MMU might not be initialized correctly.
1464 	 * Set it again to fix this.
1465 	 */
1466 
1467 	ret = nested_svm_load_cr3(&svm->vcpu, vcpu->arch.cr3,
1468 				  nested_npt_enabled(svm), false);
1469 	if (WARN_ON_ONCE(ret))
1470 		goto out_free;
1471 
1472 
1473 	/*
1474 	 * All checks done, we can enter guest mode. Userspace provides
1475 	 * vmcb12.control, which will be combined with L1 and stored into
1476 	 * vmcb02, and the L1 save state which we store in vmcb01.
1477 	 * L2 registers if needed are moved from the current VMCB to VMCB02.
1478 	 */
1479 
1480 	if (is_guest_mode(vcpu))
1481 		svm_leave_nested(vcpu);
1482 	else
1483 		svm->nested.vmcb02.ptr->save = svm->vmcb01.ptr->save;
1484 
1485 	svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
1486 
1487 	svm->nested.nested_run_pending =
1488 		!!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
1489 
1490 	svm->nested.vmcb12_gpa = kvm_state->hdr.svm.vmcb_pa;
1491 
1492 	svm_copy_vmrun_state(&svm->vmcb01.ptr->save, save);
1493 	nested_copy_vmcb_control_to_cache(svm, ctl);
1494 
1495 	svm_switch_vmcb(svm, &svm->nested.vmcb02);
1496 	nested_vmcb02_prepare_control(svm);
1497 	kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
1498 	ret = 0;
1499 out_free:
1500 	kfree(save);
1501 	kfree(ctl);
1502 
1503 	return ret;
1504 }
1505 
1506 static bool svm_get_nested_state_pages(struct kvm_vcpu *vcpu)
1507 {
1508 	struct vcpu_svm *svm = to_svm(vcpu);
1509 
1510 	if (WARN_ON(!is_guest_mode(vcpu)))
1511 		return true;
1512 
1513 	if (!vcpu->arch.pdptrs_from_userspace &&
1514 	    !nested_npt_enabled(svm) && is_pae_paging(vcpu))
1515 		/*
1516 		 * Reload the guest's PDPTRs since after a migration
1517 		 * the guest CR3 might be restored prior to setting the nested
1518 		 * state which can lead to a load of wrong PDPTRs.
1519 		 */
1520 		if (CC(!load_pdptrs(vcpu, vcpu->arch.cr3)))
1521 			return false;
1522 
1523 	if (!nested_svm_vmrun_msrpm(svm)) {
1524 		vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1525 		vcpu->run->internal.suberror =
1526 			KVM_INTERNAL_ERROR_EMULATION;
1527 		vcpu->run->internal.ndata = 0;
1528 		return false;
1529 	}
1530 
1531 	return true;
1532 }
1533 
1534 struct kvm_x86_nested_ops svm_nested_ops = {
1535 	.leave_nested = svm_leave_nested,
1536 	.check_events = svm_check_nested_events,
1537 	.triple_fault = nested_svm_triple_fault,
1538 	.get_nested_state_pages = svm_get_nested_state_pages,
1539 	.get_state = svm_get_nested_state,
1540 	.set_state = svm_set_nested_state,
1541 };
1542