xref: /openbmc/linux/arch/x86/kvm/svm/nested.c (revision a01822e94ee53e8ebc9632fe2764048b81921254)
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 "svm.h"
29 
30 static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
31 				       struct x86_exception *fault)
32 {
33 	struct vcpu_svm *svm = to_svm(vcpu);
34 
35 	if (svm->vmcb->control.exit_code != SVM_EXIT_NPF) {
36 		/*
37 		 * TODO: track the cause of the nested page fault, and
38 		 * correctly fill in the high bits of exit_info_1.
39 		 */
40 		svm->vmcb->control.exit_code = SVM_EXIT_NPF;
41 		svm->vmcb->control.exit_code_hi = 0;
42 		svm->vmcb->control.exit_info_1 = (1ULL << 32);
43 		svm->vmcb->control.exit_info_2 = fault->address;
44 	}
45 
46 	svm->vmcb->control.exit_info_1 &= ~0xffffffffULL;
47 	svm->vmcb->control.exit_info_1 |= fault->error_code;
48 
49 	/*
50 	 * The present bit is always zero for page structure faults on real
51 	 * hardware.
52 	 */
53 	if (svm->vmcb->control.exit_info_1 & (2ULL << 32))
54 		svm->vmcb->control.exit_info_1 &= ~1;
55 
56 	nested_svm_vmexit(svm);
57 }
58 
59 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
60 {
61 	struct vcpu_svm *svm = to_svm(vcpu);
62 	u64 cr3 = svm->nested.nested_cr3;
63 	u64 pdpte;
64 	int ret;
65 
66 	ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(__sme_clr(cr3)), &pdpte,
67 				       offset_in_page(cr3) + index * 8, 8);
68 	if (ret)
69 		return 0;
70 	return pdpte;
71 }
72 
73 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
74 {
75 	struct vcpu_svm *svm = to_svm(vcpu);
76 
77 	return svm->nested.nested_cr3;
78 }
79 
80 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
81 {
82 	WARN_ON(mmu_is_nested(vcpu));
83 
84 	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
85 	kvm_init_shadow_mmu(vcpu);
86 	vcpu->arch.mmu->get_guest_pgd     = nested_svm_get_tdp_cr3;
87 	vcpu->arch.mmu->get_pdptr         = nested_svm_get_tdp_pdptr;
88 	vcpu->arch.mmu->inject_page_fault = nested_svm_inject_npf_exit;
89 	vcpu->arch.mmu->shadow_root_level = kvm_x86_ops.get_tdp_level(vcpu);
90 	reset_shadow_zero_bits_mask(vcpu, vcpu->arch.mmu);
91 	vcpu->arch.walk_mmu              = &vcpu->arch.nested_mmu;
92 }
93 
94 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
95 {
96 	vcpu->arch.mmu = &vcpu->arch.root_mmu;
97 	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
98 }
99 
100 void recalc_intercepts(struct vcpu_svm *svm)
101 {
102 	struct vmcb_control_area *c, *h;
103 	struct nested_state *g;
104 
105 	mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
106 
107 	if (!is_guest_mode(&svm->vcpu))
108 		return;
109 
110 	c = &svm->vmcb->control;
111 	h = &svm->nested.hsave->control;
112 	g = &svm->nested;
113 
114 	c->intercept_cr = h->intercept_cr;
115 	c->intercept_dr = h->intercept_dr;
116 	c->intercept_exceptions = h->intercept_exceptions;
117 	c->intercept = h->intercept;
118 
119 	if (svm->vcpu.arch.hflags & HF_VINTR_MASK) {
120 		/* We only want the cr8 intercept bits of L1 */
121 		c->intercept_cr &= ~(1U << INTERCEPT_CR8_READ);
122 		c->intercept_cr &= ~(1U << INTERCEPT_CR8_WRITE);
123 
124 		/*
125 		 * Once running L2 with HF_VINTR_MASK, EFLAGS.IF does not
126 		 * affect any interrupt we may want to inject; therefore,
127 		 * interrupt window vmexits are irrelevant to L0.
128 		 */
129 		c->intercept &= ~(1ULL << INTERCEPT_VINTR);
130 	}
131 
132 	/* We don't want to see VMMCALLs from a nested guest */
133 	c->intercept &= ~(1ULL << INTERCEPT_VMMCALL);
134 
135 	c->intercept_cr |= g->intercept_cr;
136 	c->intercept_dr |= g->intercept_dr;
137 	c->intercept_exceptions |= g->intercept_exceptions;
138 	c->intercept |= g->intercept;
139 }
140 
141 static void copy_vmcb_control_area(struct vmcb *dst_vmcb, struct vmcb *from_vmcb)
142 {
143 	struct vmcb_control_area *dst  = &dst_vmcb->control;
144 	struct vmcb_control_area *from = &from_vmcb->control;
145 
146 	dst->intercept_cr         = from->intercept_cr;
147 	dst->intercept_dr         = from->intercept_dr;
148 	dst->intercept_exceptions = from->intercept_exceptions;
149 	dst->intercept            = from->intercept;
150 	dst->iopm_base_pa         = from->iopm_base_pa;
151 	dst->msrpm_base_pa        = from->msrpm_base_pa;
152 	dst->tsc_offset           = from->tsc_offset;
153 	dst->asid                 = from->asid;
154 	dst->tlb_ctl              = from->tlb_ctl;
155 	dst->int_ctl              = from->int_ctl;
156 	dst->int_vector           = from->int_vector;
157 	dst->int_state            = from->int_state;
158 	dst->exit_code            = from->exit_code;
159 	dst->exit_code_hi         = from->exit_code_hi;
160 	dst->exit_info_1          = from->exit_info_1;
161 	dst->exit_info_2          = from->exit_info_2;
162 	dst->exit_int_info        = from->exit_int_info;
163 	dst->exit_int_info_err    = from->exit_int_info_err;
164 	dst->nested_ctl           = from->nested_ctl;
165 	dst->event_inj            = from->event_inj;
166 	dst->event_inj_err        = from->event_inj_err;
167 	dst->nested_cr3           = from->nested_cr3;
168 	dst->virt_ext              = from->virt_ext;
169 	dst->pause_filter_count   = from->pause_filter_count;
170 	dst->pause_filter_thresh  = from->pause_filter_thresh;
171 }
172 
173 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
174 {
175 	/*
176 	 * This function merges the msr permission bitmaps of kvm and the
177 	 * nested vmcb. It is optimized in that it only merges the parts where
178 	 * the kvm msr permission bitmap may contain zero bits
179 	 */
180 	int i;
181 
182 	if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
183 		return true;
184 
185 	for (i = 0; i < MSRPM_OFFSETS; i++) {
186 		u32 value, p;
187 		u64 offset;
188 
189 		if (msrpm_offsets[i] == 0xffffffff)
190 			break;
191 
192 		p      = msrpm_offsets[i];
193 		offset = svm->nested.vmcb_msrpm + (p * 4);
194 
195 		if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
196 			return false;
197 
198 		svm->nested.msrpm[p] = svm->msrpm[p] | value;
199 	}
200 
201 	svm->vmcb->control.msrpm_base_pa = __sme_set(__pa(svm->nested.msrpm));
202 
203 	return true;
204 }
205 
206 static bool nested_vmcb_checks(struct vmcb *vmcb)
207 {
208 	if ((vmcb->save.efer & EFER_SVME) == 0)
209 		return false;
210 
211 	if ((vmcb->control.intercept & (1ULL << INTERCEPT_VMRUN)) == 0)
212 		return false;
213 
214 	if (vmcb->control.asid == 0)
215 		return false;
216 
217 	if ((vmcb->control.nested_ctl & SVM_NESTED_CTL_NP_ENABLE) &&
218 	    !npt_enabled)
219 		return false;
220 
221 	return true;
222 }
223 
224 void enter_svm_guest_mode(struct vcpu_svm *svm, u64 vmcb_gpa,
225 			  struct vmcb *nested_vmcb, struct kvm_host_map *map)
226 {
227 	bool evaluate_pending_interrupts =
228 		is_intercept(svm, INTERCEPT_VINTR) ||
229 		is_intercept(svm, INTERCEPT_IRET);
230 
231 	if (kvm_get_rflags(&svm->vcpu) & X86_EFLAGS_IF)
232 		svm->vcpu.arch.hflags |= HF_HIF_MASK;
233 	else
234 		svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
235 
236 	if (nested_vmcb->control.nested_ctl & SVM_NESTED_CTL_NP_ENABLE) {
237 		svm->nested.nested_cr3 = nested_vmcb->control.nested_cr3;
238 		nested_svm_init_mmu_context(&svm->vcpu);
239 	}
240 
241 	/* Load the nested guest state */
242 	svm->vmcb->save.es = nested_vmcb->save.es;
243 	svm->vmcb->save.cs = nested_vmcb->save.cs;
244 	svm->vmcb->save.ss = nested_vmcb->save.ss;
245 	svm->vmcb->save.ds = nested_vmcb->save.ds;
246 	svm->vmcb->save.gdtr = nested_vmcb->save.gdtr;
247 	svm->vmcb->save.idtr = nested_vmcb->save.idtr;
248 	kvm_set_rflags(&svm->vcpu, nested_vmcb->save.rflags);
249 	svm_set_efer(&svm->vcpu, nested_vmcb->save.efer);
250 	svm_set_cr0(&svm->vcpu, nested_vmcb->save.cr0);
251 	svm_set_cr4(&svm->vcpu, nested_vmcb->save.cr4);
252 	if (npt_enabled) {
253 		svm->vmcb->save.cr3 = nested_vmcb->save.cr3;
254 		svm->vcpu.arch.cr3 = nested_vmcb->save.cr3;
255 	} else
256 		(void)kvm_set_cr3(&svm->vcpu, nested_vmcb->save.cr3);
257 
258 	/* Guest paging mode is active - reset mmu */
259 	kvm_mmu_reset_context(&svm->vcpu);
260 
261 	svm->vmcb->save.cr2 = svm->vcpu.arch.cr2 = nested_vmcb->save.cr2;
262 	kvm_rax_write(&svm->vcpu, nested_vmcb->save.rax);
263 	kvm_rsp_write(&svm->vcpu, nested_vmcb->save.rsp);
264 	kvm_rip_write(&svm->vcpu, nested_vmcb->save.rip);
265 
266 	/* In case we don't even reach vcpu_run, the fields are not updated */
267 	svm->vmcb->save.rax = nested_vmcb->save.rax;
268 	svm->vmcb->save.rsp = nested_vmcb->save.rsp;
269 	svm->vmcb->save.rip = nested_vmcb->save.rip;
270 	svm->vmcb->save.dr7 = nested_vmcb->save.dr7;
271 	svm->vcpu.arch.dr6  = nested_vmcb->save.dr6;
272 	svm->vmcb->save.cpl = nested_vmcb->save.cpl;
273 
274 	svm->nested.vmcb_msrpm = nested_vmcb->control.msrpm_base_pa & ~0x0fffULL;
275 	svm->nested.vmcb_iopm  = nested_vmcb->control.iopm_base_pa  & ~0x0fffULL;
276 
277 	/* cache intercepts */
278 	svm->nested.intercept_cr         = nested_vmcb->control.intercept_cr;
279 	svm->nested.intercept_dr         = nested_vmcb->control.intercept_dr;
280 	svm->nested.intercept_exceptions = nested_vmcb->control.intercept_exceptions;
281 	svm->nested.intercept            = nested_vmcb->control.intercept;
282 
283 	svm_flush_tlb(&svm->vcpu, true);
284 	svm->vmcb->control.int_ctl = nested_vmcb->control.int_ctl | V_INTR_MASKING_MASK;
285 	if (nested_vmcb->control.int_ctl & V_INTR_MASKING_MASK)
286 		svm->vcpu.arch.hflags |= HF_VINTR_MASK;
287 	else
288 		svm->vcpu.arch.hflags &= ~HF_VINTR_MASK;
289 
290 	svm->vcpu.arch.tsc_offset += nested_vmcb->control.tsc_offset;
291 	svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset;
292 
293 	svm->vmcb->control.virt_ext = nested_vmcb->control.virt_ext;
294 	svm->vmcb->control.int_vector = nested_vmcb->control.int_vector;
295 	svm->vmcb->control.int_state = nested_vmcb->control.int_state;
296 	svm->vmcb->control.event_inj = nested_vmcb->control.event_inj;
297 	svm->vmcb->control.event_inj_err = nested_vmcb->control.event_inj_err;
298 
299 	svm->vmcb->control.pause_filter_count =
300 		nested_vmcb->control.pause_filter_count;
301 	svm->vmcb->control.pause_filter_thresh =
302 		nested_vmcb->control.pause_filter_thresh;
303 
304 	kvm_vcpu_unmap(&svm->vcpu, map, true);
305 
306 	/* Enter Guest-Mode */
307 	enter_guest_mode(&svm->vcpu);
308 
309 	/*
310 	 * Merge guest and host intercepts - must be called  with vcpu in
311 	 * guest-mode to take affect here
312 	 */
313 	recalc_intercepts(svm);
314 
315 	svm->nested.vmcb = vmcb_gpa;
316 
317 	/*
318 	 * If L1 had a pending IRQ/NMI before executing VMRUN,
319 	 * which wasn't delivered because it was disallowed (e.g.
320 	 * interrupts disabled), L0 needs to evaluate if this pending
321 	 * event should cause an exit from L2 to L1 or be delivered
322 	 * directly to L2.
323 	 *
324 	 * Usually this would be handled by the processor noticing an
325 	 * IRQ/NMI window request.  However, VMRUN can unblock interrupts
326 	 * by implicitly setting GIF, so force L0 to perform pending event
327 	 * evaluation by requesting a KVM_REQ_EVENT.
328 	 */
329 	enable_gif(svm);
330 	if (unlikely(evaluate_pending_interrupts))
331 		kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
332 
333 	mark_all_dirty(svm->vmcb);
334 }
335 
336 int nested_svm_vmrun(struct vcpu_svm *svm)
337 {
338 	int ret;
339 	struct vmcb *nested_vmcb;
340 	struct vmcb *hsave = svm->nested.hsave;
341 	struct vmcb *vmcb = svm->vmcb;
342 	struct kvm_host_map map;
343 	u64 vmcb_gpa;
344 
345 	vmcb_gpa = svm->vmcb->save.rax;
346 
347 	ret = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(vmcb_gpa), &map);
348 	if (ret == -EINVAL) {
349 		kvm_inject_gp(&svm->vcpu, 0);
350 		return 1;
351 	} else if (ret) {
352 		return kvm_skip_emulated_instruction(&svm->vcpu);
353 	}
354 
355 	ret = kvm_skip_emulated_instruction(&svm->vcpu);
356 
357 	nested_vmcb = map.hva;
358 
359 	if (!nested_vmcb_checks(nested_vmcb)) {
360 		nested_vmcb->control.exit_code    = SVM_EXIT_ERR;
361 		nested_vmcb->control.exit_code_hi = 0;
362 		nested_vmcb->control.exit_info_1  = 0;
363 		nested_vmcb->control.exit_info_2  = 0;
364 
365 		kvm_vcpu_unmap(&svm->vcpu, &map, true);
366 
367 		return ret;
368 	}
369 
370 	trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb_gpa,
371 			       nested_vmcb->save.rip,
372 			       nested_vmcb->control.int_ctl,
373 			       nested_vmcb->control.event_inj,
374 			       nested_vmcb->control.nested_ctl);
375 
376 	trace_kvm_nested_intercepts(nested_vmcb->control.intercept_cr & 0xffff,
377 				    nested_vmcb->control.intercept_cr >> 16,
378 				    nested_vmcb->control.intercept_exceptions,
379 				    nested_vmcb->control.intercept);
380 
381 	/* Clear internal status */
382 	kvm_clear_exception_queue(&svm->vcpu);
383 	kvm_clear_interrupt_queue(&svm->vcpu);
384 
385 	/*
386 	 * Save the old vmcb, so we don't need to pick what we save, but can
387 	 * restore everything when a VMEXIT occurs
388 	 */
389 	hsave->save.es     = vmcb->save.es;
390 	hsave->save.cs     = vmcb->save.cs;
391 	hsave->save.ss     = vmcb->save.ss;
392 	hsave->save.ds     = vmcb->save.ds;
393 	hsave->save.gdtr   = vmcb->save.gdtr;
394 	hsave->save.idtr   = vmcb->save.idtr;
395 	hsave->save.efer   = svm->vcpu.arch.efer;
396 	hsave->save.cr0    = kvm_read_cr0(&svm->vcpu);
397 	hsave->save.cr4    = svm->vcpu.arch.cr4;
398 	hsave->save.rflags = kvm_get_rflags(&svm->vcpu);
399 	hsave->save.rip    = kvm_rip_read(&svm->vcpu);
400 	hsave->save.rsp    = vmcb->save.rsp;
401 	hsave->save.rax    = vmcb->save.rax;
402 	if (npt_enabled)
403 		hsave->save.cr3    = vmcb->save.cr3;
404 	else
405 		hsave->save.cr3    = kvm_read_cr3(&svm->vcpu);
406 
407 	copy_vmcb_control_area(hsave, vmcb);
408 
409 	enter_svm_guest_mode(svm, vmcb_gpa, nested_vmcb, &map);
410 
411 	if (!nested_svm_vmrun_msrpm(svm)) {
412 		svm->vmcb->control.exit_code    = SVM_EXIT_ERR;
413 		svm->vmcb->control.exit_code_hi = 0;
414 		svm->vmcb->control.exit_info_1  = 0;
415 		svm->vmcb->control.exit_info_2  = 0;
416 
417 		nested_svm_vmexit(svm);
418 	}
419 
420 	return ret;
421 }
422 
423 void nested_svm_vmloadsave(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
424 {
425 	to_vmcb->save.fs = from_vmcb->save.fs;
426 	to_vmcb->save.gs = from_vmcb->save.gs;
427 	to_vmcb->save.tr = from_vmcb->save.tr;
428 	to_vmcb->save.ldtr = from_vmcb->save.ldtr;
429 	to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
430 	to_vmcb->save.star = from_vmcb->save.star;
431 	to_vmcb->save.lstar = from_vmcb->save.lstar;
432 	to_vmcb->save.cstar = from_vmcb->save.cstar;
433 	to_vmcb->save.sfmask = from_vmcb->save.sfmask;
434 	to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
435 	to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
436 	to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
437 }
438 
439 int nested_svm_vmexit(struct vcpu_svm *svm)
440 {
441 	int rc;
442 	struct vmcb *nested_vmcb;
443 	struct vmcb *hsave = svm->nested.hsave;
444 	struct vmcb *vmcb = svm->vmcb;
445 	struct kvm_host_map map;
446 
447 	trace_kvm_nested_vmexit_inject(vmcb->control.exit_code,
448 				       vmcb->control.exit_info_1,
449 				       vmcb->control.exit_info_2,
450 				       vmcb->control.exit_int_info,
451 				       vmcb->control.exit_int_info_err,
452 				       KVM_ISA_SVM);
453 
454 	rc = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(svm->nested.vmcb), &map);
455 	if (rc) {
456 		if (rc == -EINVAL)
457 			kvm_inject_gp(&svm->vcpu, 0);
458 		return 1;
459 	}
460 
461 	nested_vmcb = map.hva;
462 
463 	/* Exit Guest-Mode */
464 	leave_guest_mode(&svm->vcpu);
465 	svm->nested.vmcb = 0;
466 
467 	/* Give the current vmcb to the guest */
468 	disable_gif(svm);
469 
470 	nested_vmcb->save.es     = vmcb->save.es;
471 	nested_vmcb->save.cs     = vmcb->save.cs;
472 	nested_vmcb->save.ss     = vmcb->save.ss;
473 	nested_vmcb->save.ds     = vmcb->save.ds;
474 	nested_vmcb->save.gdtr   = vmcb->save.gdtr;
475 	nested_vmcb->save.idtr   = vmcb->save.idtr;
476 	nested_vmcb->save.efer   = svm->vcpu.arch.efer;
477 	nested_vmcb->save.cr0    = kvm_read_cr0(&svm->vcpu);
478 	nested_vmcb->save.cr3    = kvm_read_cr3(&svm->vcpu);
479 	nested_vmcb->save.cr2    = vmcb->save.cr2;
480 	nested_vmcb->save.cr4    = svm->vcpu.arch.cr4;
481 	nested_vmcb->save.rflags = kvm_get_rflags(&svm->vcpu);
482 	nested_vmcb->save.rip    = vmcb->save.rip;
483 	nested_vmcb->save.rsp    = vmcb->save.rsp;
484 	nested_vmcb->save.rax    = vmcb->save.rax;
485 	nested_vmcb->save.dr7    = vmcb->save.dr7;
486 	nested_vmcb->save.dr6    = svm->vcpu.arch.dr6;
487 	nested_vmcb->save.cpl    = vmcb->save.cpl;
488 
489 	nested_vmcb->control.int_ctl           = vmcb->control.int_ctl;
490 	nested_vmcb->control.int_vector        = vmcb->control.int_vector;
491 	nested_vmcb->control.int_state         = vmcb->control.int_state;
492 	nested_vmcb->control.exit_code         = vmcb->control.exit_code;
493 	nested_vmcb->control.exit_code_hi      = vmcb->control.exit_code_hi;
494 	nested_vmcb->control.exit_info_1       = vmcb->control.exit_info_1;
495 	nested_vmcb->control.exit_info_2       = vmcb->control.exit_info_2;
496 	nested_vmcb->control.exit_int_info     = vmcb->control.exit_int_info;
497 	nested_vmcb->control.exit_int_info_err = vmcb->control.exit_int_info_err;
498 
499 	if (svm->nrips_enabled)
500 		nested_vmcb->control.next_rip  = vmcb->control.next_rip;
501 
502 	/*
503 	 * If we emulate a VMRUN/#VMEXIT in the same host #vmexit cycle we have
504 	 * to make sure that we do not lose injected events. So check event_inj
505 	 * here and copy it to exit_int_info if it is valid.
506 	 * Exit_int_info and event_inj can't be both valid because the case
507 	 * below only happens on a VMRUN instruction intercept which has
508 	 * no valid exit_int_info set.
509 	 */
510 	if (vmcb->control.event_inj & SVM_EVTINJ_VALID) {
511 		struct vmcb_control_area *nc = &nested_vmcb->control;
512 
513 		nc->exit_int_info     = vmcb->control.event_inj;
514 		nc->exit_int_info_err = vmcb->control.event_inj_err;
515 	}
516 
517 	nested_vmcb->control.tlb_ctl           = 0;
518 	nested_vmcb->control.event_inj         = 0;
519 	nested_vmcb->control.event_inj_err     = 0;
520 
521 	nested_vmcb->control.pause_filter_count =
522 		svm->vmcb->control.pause_filter_count;
523 	nested_vmcb->control.pause_filter_thresh =
524 		svm->vmcb->control.pause_filter_thresh;
525 
526 	/* We always set V_INTR_MASKING and remember the old value in hflags */
527 	if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
528 		nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
529 
530 	/* Restore the original control entries */
531 	copy_vmcb_control_area(vmcb, hsave);
532 
533 	svm->vcpu.arch.tsc_offset = svm->vmcb->control.tsc_offset;
534 	kvm_clear_exception_queue(&svm->vcpu);
535 	kvm_clear_interrupt_queue(&svm->vcpu);
536 
537 	svm->nested.nested_cr3 = 0;
538 
539 	/* Restore selected save entries */
540 	svm->vmcb->save.es = hsave->save.es;
541 	svm->vmcb->save.cs = hsave->save.cs;
542 	svm->vmcb->save.ss = hsave->save.ss;
543 	svm->vmcb->save.ds = hsave->save.ds;
544 	svm->vmcb->save.gdtr = hsave->save.gdtr;
545 	svm->vmcb->save.idtr = hsave->save.idtr;
546 	kvm_set_rflags(&svm->vcpu, hsave->save.rflags);
547 	svm_set_efer(&svm->vcpu, hsave->save.efer);
548 	svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
549 	svm_set_cr4(&svm->vcpu, hsave->save.cr4);
550 	if (npt_enabled) {
551 		svm->vmcb->save.cr3 = hsave->save.cr3;
552 		svm->vcpu.arch.cr3 = hsave->save.cr3;
553 	} else {
554 		(void)kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
555 	}
556 	kvm_rax_write(&svm->vcpu, hsave->save.rax);
557 	kvm_rsp_write(&svm->vcpu, hsave->save.rsp);
558 	kvm_rip_write(&svm->vcpu, hsave->save.rip);
559 	svm->vmcb->save.dr7 = 0;
560 	svm->vmcb->save.cpl = 0;
561 	svm->vmcb->control.exit_int_info = 0;
562 
563 	mark_all_dirty(svm->vmcb);
564 
565 	kvm_vcpu_unmap(&svm->vcpu, &map, true);
566 
567 	nested_svm_uninit_mmu_context(&svm->vcpu);
568 	kvm_mmu_reset_context(&svm->vcpu);
569 	kvm_mmu_load(&svm->vcpu);
570 
571 	/*
572 	 * Drop what we picked up for L2 via svm_complete_interrupts() so it
573 	 * doesn't end up in L1.
574 	 */
575 	svm->vcpu.arch.nmi_injected = false;
576 	kvm_clear_exception_queue(&svm->vcpu);
577 	kvm_clear_interrupt_queue(&svm->vcpu);
578 
579 	return 0;
580 }
581 
582 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
583 {
584 	u32 offset, msr, value;
585 	int write, mask;
586 
587 	if (!(svm->nested.intercept & (1ULL << INTERCEPT_MSR_PROT)))
588 		return NESTED_EXIT_HOST;
589 
590 	msr    = svm->vcpu.arch.regs[VCPU_REGS_RCX];
591 	offset = svm_msrpm_offset(msr);
592 	write  = svm->vmcb->control.exit_info_1 & 1;
593 	mask   = 1 << ((2 * (msr & 0xf)) + write);
594 
595 	if (offset == MSR_INVALID)
596 		return NESTED_EXIT_DONE;
597 
598 	/* Offset is in 32 bit units but need in 8 bit units */
599 	offset *= 4;
600 
601 	if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.vmcb_msrpm + offset, &value, 4))
602 		return NESTED_EXIT_DONE;
603 
604 	return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
605 }
606 
607 /* DB exceptions for our internal use must not cause vmexit */
608 static int nested_svm_intercept_db(struct vcpu_svm *svm)
609 {
610 	unsigned long dr6 = svm->vmcb->save.dr6;
611 
612 	/* Always catch it and pass it to userspace if debugging.  */
613 	if (svm->vcpu.guest_debug &
614 	    (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
615 		return NESTED_EXIT_HOST;
616 
617 	/* if we're not singlestepping, it's not ours */
618 	if (!svm->nmi_singlestep)
619 		goto reflected_db;
620 
621 	/* if it's not a singlestep exception, it's not ours */
622 	if (!(dr6 & DR6_BS))
623 		goto reflected_db;
624 
625 	/* if the guest is singlestepping, it should get the vmexit */
626 	if (svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF) {
627 		disable_nmi_singlestep(svm);
628 		goto reflected_db;
629 	}
630 
631 	/* it's ours, the nested hypervisor must not see this one */
632 	return NESTED_EXIT_HOST;
633 
634 reflected_db:
635 	/*
636 	 * Synchronize guest DR6 here just like in kvm_deliver_exception_payload;
637 	 * it will be moved into the nested VMCB by nested_svm_vmexit.  Once
638 	 * exceptions will be moved to svm_check_nested_events, all this stuff
639 	 * will just go away and we could just return NESTED_EXIT_HOST
640 	 * unconditionally.  db_interception will queue the exception, which
641 	 * will be processed by svm_check_nested_events if a nested vmexit is
642 	 * required, and we will just use kvm_deliver_exception_payload to copy
643 	 * the payload to DR6 before vmexit.
644 	 */
645 	WARN_ON(svm->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT);
646 	svm->vcpu.arch.dr6 &= ~(DR_TRAP_BITS | DR6_RTM);
647 	svm->vcpu.arch.dr6 |= dr6 & ~DR6_FIXED_1;
648 	return NESTED_EXIT_DONE;
649 }
650 
651 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
652 {
653 	unsigned port, size, iopm_len;
654 	u16 val, mask;
655 	u8 start_bit;
656 	u64 gpa;
657 
658 	if (!(svm->nested.intercept & (1ULL << INTERCEPT_IOIO_PROT)))
659 		return NESTED_EXIT_HOST;
660 
661 	port = svm->vmcb->control.exit_info_1 >> 16;
662 	size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
663 		SVM_IOIO_SIZE_SHIFT;
664 	gpa  = svm->nested.vmcb_iopm + (port / 8);
665 	start_bit = port % 8;
666 	iopm_len = (start_bit + size > 8) ? 2 : 1;
667 	mask = (0xf >> (4 - size)) << start_bit;
668 	val = 0;
669 
670 	if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
671 		return NESTED_EXIT_DONE;
672 
673 	return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
674 }
675 
676 static int nested_svm_intercept(struct vcpu_svm *svm)
677 {
678 	u32 exit_code = svm->vmcb->control.exit_code;
679 	int vmexit = NESTED_EXIT_HOST;
680 
681 	switch (exit_code) {
682 	case SVM_EXIT_MSR:
683 		vmexit = nested_svm_exit_handled_msr(svm);
684 		break;
685 	case SVM_EXIT_IOIO:
686 		vmexit = nested_svm_intercept_ioio(svm);
687 		break;
688 	case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
689 		u32 bit = 1U << (exit_code - SVM_EXIT_READ_CR0);
690 		if (svm->nested.intercept_cr & bit)
691 			vmexit = NESTED_EXIT_DONE;
692 		break;
693 	}
694 	case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
695 		u32 bit = 1U << (exit_code - SVM_EXIT_READ_DR0);
696 		if (svm->nested.intercept_dr & bit)
697 			vmexit = NESTED_EXIT_DONE;
698 		break;
699 	}
700 	case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
701 		u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
702 		if (svm->nested.intercept_exceptions & excp_bits) {
703 			if (exit_code == SVM_EXIT_EXCP_BASE + DB_VECTOR)
704 				vmexit = nested_svm_intercept_db(svm);
705 			else if (exit_code == SVM_EXIT_EXCP_BASE + BP_VECTOR &&
706 				 svm->vcpu.guest_debug & KVM_GUESTDBG_USE_SW_BP)
707 				vmexit = NESTED_EXIT_HOST;
708 			else
709 				vmexit = NESTED_EXIT_DONE;
710 		}
711 		/* async page fault always cause vmexit */
712 		else if ((exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR) &&
713 			 svm->vcpu.arch.exception.nested_apf != 0)
714 			vmexit = NESTED_EXIT_DONE;
715 		break;
716 	}
717 	case SVM_EXIT_ERR: {
718 		vmexit = NESTED_EXIT_DONE;
719 		break;
720 	}
721 	default: {
722 		u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
723 		if (svm->nested.intercept & exit_bits)
724 			vmexit = NESTED_EXIT_DONE;
725 	}
726 	}
727 
728 	return vmexit;
729 }
730 
731 int nested_svm_exit_handled(struct vcpu_svm *svm)
732 {
733 	int vmexit;
734 
735 	vmexit = nested_svm_intercept(svm);
736 
737 	if (vmexit == NESTED_EXIT_DONE)
738 		nested_svm_vmexit(svm);
739 
740 	return vmexit;
741 }
742 
743 int nested_svm_check_permissions(struct vcpu_svm *svm)
744 {
745 	if (!(svm->vcpu.arch.efer & EFER_SVME) ||
746 	    !is_paging(&svm->vcpu)) {
747 		kvm_queue_exception(&svm->vcpu, UD_VECTOR);
748 		return 1;
749 	}
750 
751 	if (svm->vmcb->save.cpl) {
752 		kvm_inject_gp(&svm->vcpu, 0);
753 		return 1;
754 	}
755 
756 	return 0;
757 }
758 
759 int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
760 			       bool has_error_code, u32 error_code)
761 {
762 	int vmexit;
763 
764 	if (!is_guest_mode(&svm->vcpu))
765 		return 0;
766 
767 	vmexit = nested_svm_intercept(svm);
768 	if (vmexit != NESTED_EXIT_DONE)
769 		return 0;
770 
771 	svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
772 	svm->vmcb->control.exit_code_hi = 0;
773 	svm->vmcb->control.exit_info_1 = error_code;
774 
775 	/*
776 	 * EXITINFO2 is undefined for all exception intercepts other
777 	 * than #PF.
778 	 */
779 	if (svm->vcpu.arch.exception.nested_apf)
780 		svm->vmcb->control.exit_info_2 = svm->vcpu.arch.apf.nested_apf_token;
781 	else if (svm->vcpu.arch.exception.has_payload)
782 		svm->vmcb->control.exit_info_2 = svm->vcpu.arch.exception.payload;
783 	else
784 		svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
785 
786 	svm->nested.exit_required = true;
787 	return vmexit;
788 }
789 
790 static void nested_svm_intr(struct vcpu_svm *svm)
791 {
792 	svm->vmcb->control.exit_code   = SVM_EXIT_INTR;
793 	svm->vmcb->control.exit_info_1 = 0;
794 	svm->vmcb->control.exit_info_2 = 0;
795 
796 	/* nested_svm_vmexit this gets called afterwards from handle_exit */
797 	svm->nested.exit_required = true;
798 	trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
799 }
800 
801 static bool nested_exit_on_intr(struct vcpu_svm *svm)
802 {
803 	return (svm->nested.intercept & 1ULL);
804 }
805 
806 int svm_check_nested_events(struct kvm_vcpu *vcpu)
807 {
808 	struct vcpu_svm *svm = to_svm(vcpu);
809 	bool block_nested_events =
810 		kvm_event_needs_reinjection(vcpu) || svm->nested.exit_required;
811 
812 	if (kvm_cpu_has_interrupt(vcpu) && nested_exit_on_intr(svm)) {
813 		if (block_nested_events)
814 			return -EBUSY;
815 		nested_svm_intr(svm);
816 		return 0;
817 	}
818 
819 	return 0;
820 }
821 
822 int nested_svm_exit_special(struct vcpu_svm *svm)
823 {
824 	u32 exit_code = svm->vmcb->control.exit_code;
825 
826 	switch (exit_code) {
827 	case SVM_EXIT_INTR:
828 	case SVM_EXIT_NMI:
829 	case SVM_EXIT_EXCP_BASE + MC_VECTOR:
830 		return NESTED_EXIT_HOST;
831 	case SVM_EXIT_NPF:
832 		/* For now we are always handling NPFs when using them */
833 		if (npt_enabled)
834 			return NESTED_EXIT_HOST;
835 		break;
836 	case SVM_EXIT_EXCP_BASE + PF_VECTOR:
837 		/* When we're shadowing, trap PFs, but not async PF */
838 		if (!npt_enabled && svm->vcpu.arch.apf.host_apf_reason == 0)
839 			return NESTED_EXIT_HOST;
840 		break;
841 	default:
842 		break;
843 	}
844 
845 	return NESTED_EXIT_CONTINUE;
846 }
847