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
nested_svm_inject_npf_exit(struct kvm_vcpu * vcpu,struct x86_exception * fault)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
svm_inject_page_fault_nested(struct kvm_vcpu * vcpu,struct x86_exception * fault)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 (vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_EXCEPTION_OFFSET + PF_VECTOR) &&
62 !svm->nested.nested_run_pending) {
63 svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + PF_VECTOR;
64 svm->vmcb->control.exit_code_hi = 0;
65 svm->vmcb->control.exit_info_1 = fault->error_code;
66 svm->vmcb->control.exit_info_2 = fault->address;
67 nested_svm_vmexit(svm);
68 } else {
69 kvm_inject_page_fault(vcpu, fault);
70 }
71 }
72
nested_svm_get_tdp_pdptr(struct kvm_vcpu * vcpu,int index)73 static u64 nested_svm_get_tdp_pdptr(struct kvm_vcpu *vcpu, int index)
74 {
75 struct vcpu_svm *svm = to_svm(vcpu);
76 u64 cr3 = svm->nested.ctl.nested_cr3;
77 u64 pdpte;
78 int ret;
79
80 ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(cr3), &pdpte,
81 offset_in_page(cr3) + index * 8, 8);
82 if (ret)
83 return 0;
84 return pdpte;
85 }
86
nested_svm_get_tdp_cr3(struct kvm_vcpu * vcpu)87 static unsigned long nested_svm_get_tdp_cr3(struct kvm_vcpu *vcpu)
88 {
89 struct vcpu_svm *svm = to_svm(vcpu);
90
91 return svm->nested.ctl.nested_cr3;
92 }
93
nested_svm_init_mmu_context(struct kvm_vcpu * vcpu)94 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
95 {
96 struct vcpu_svm *svm = to_svm(vcpu);
97
98 WARN_ON(mmu_is_nested(vcpu));
99
100 vcpu->arch.mmu = &vcpu->arch.guest_mmu;
101
102 /*
103 * The NPT format depends on L1's CR4 and EFER, which is in vmcb01. Note,
104 * when called via KVM_SET_NESTED_STATE, that state may _not_ match current
105 * vCPU state. CR0.WP is explicitly ignored, while CR0.PG is required.
106 */
107 kvm_init_shadow_npt_mmu(vcpu, X86_CR0_PG, svm->vmcb01.ptr->save.cr4,
108 svm->vmcb01.ptr->save.efer,
109 svm->nested.ctl.nested_cr3);
110 vcpu->arch.mmu->get_guest_pgd = nested_svm_get_tdp_cr3;
111 vcpu->arch.mmu->get_pdptr = nested_svm_get_tdp_pdptr;
112 vcpu->arch.mmu->inject_page_fault = nested_svm_inject_npf_exit;
113 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
114 }
115
nested_svm_uninit_mmu_context(struct kvm_vcpu * vcpu)116 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
117 {
118 vcpu->arch.mmu = &vcpu->arch.root_mmu;
119 vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
120 }
121
recalc_intercepts(struct vcpu_svm * svm)122 void recalc_intercepts(struct vcpu_svm *svm)
123 {
124 struct vmcb_control_area *c, *h, *g;
125 unsigned int i;
126
127 vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
128
129 if (!is_guest_mode(&svm->vcpu))
130 return;
131
132 c = &svm->vmcb->control;
133 h = &svm->vmcb01.ptr->control;
134 g = &svm->nested.ctl;
135
136 for (i = 0; i < MAX_INTERCEPT; i++)
137 c->intercepts[i] = h->intercepts[i];
138
139 if (g->int_ctl & V_INTR_MASKING_MASK) {
140 /* We only want the cr8 intercept bits of L1 */
141 vmcb_clr_intercept(c, INTERCEPT_CR8_READ);
142 vmcb_clr_intercept(c, INTERCEPT_CR8_WRITE);
143
144 /*
145 * Once running L2 with HF_VINTR_MASK, EFLAGS.IF does not
146 * affect any interrupt we may want to inject; therefore,
147 * interrupt window vmexits are irrelevant to L0.
148 */
149 vmcb_clr_intercept(c, INTERCEPT_VINTR);
150 }
151
152 /* We don't want to see VMMCALLs from a nested guest */
153 vmcb_clr_intercept(c, INTERCEPT_VMMCALL);
154
155 for (i = 0; i < MAX_INTERCEPT; i++)
156 c->intercepts[i] |= g->intercepts[i];
157
158 /* If SMI is not intercepted, ignore guest SMI intercept as well */
159 if (!intercept_smi)
160 vmcb_clr_intercept(c, INTERCEPT_SMI);
161
162 vmcb_set_intercept(c, INTERCEPT_VMLOAD);
163 vmcb_set_intercept(c, INTERCEPT_VMSAVE);
164 }
165
copy_vmcb_control_area(struct vmcb_control_area * dst,struct vmcb_control_area * from)166 static void copy_vmcb_control_area(struct vmcb_control_area *dst,
167 struct vmcb_control_area *from)
168 {
169 unsigned int i;
170
171 for (i = 0; i < MAX_INTERCEPT; i++)
172 dst->intercepts[i] = from->intercepts[i];
173
174 dst->iopm_base_pa = from->iopm_base_pa;
175 dst->msrpm_base_pa = from->msrpm_base_pa;
176 dst->tsc_offset = from->tsc_offset;
177 /* asid not copied, it is handled manually for svm->vmcb. */
178 dst->tlb_ctl = from->tlb_ctl;
179 dst->int_ctl = from->int_ctl;
180 dst->int_vector = from->int_vector;
181 dst->int_state = from->int_state;
182 dst->exit_code = from->exit_code;
183 dst->exit_code_hi = from->exit_code_hi;
184 dst->exit_info_1 = from->exit_info_1;
185 dst->exit_info_2 = from->exit_info_2;
186 dst->exit_int_info = from->exit_int_info;
187 dst->exit_int_info_err = from->exit_int_info_err;
188 dst->nested_ctl = from->nested_ctl;
189 dst->event_inj = from->event_inj;
190 dst->event_inj_err = from->event_inj_err;
191 dst->nested_cr3 = from->nested_cr3;
192 dst->virt_ext = from->virt_ext;
193 dst->pause_filter_count = from->pause_filter_count;
194 dst->pause_filter_thresh = from->pause_filter_thresh;
195 }
196
nested_svm_vmrun_msrpm(struct vcpu_svm * svm)197 static bool nested_svm_vmrun_msrpm(struct vcpu_svm *svm)
198 {
199 /*
200 * This function merges the msr permission bitmaps of kvm and the
201 * nested vmcb. It is optimized in that it only merges the parts where
202 * the kvm msr permission bitmap may contain zero bits
203 */
204 int i;
205
206 if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
207 return true;
208
209 for (i = 0; i < MSRPM_OFFSETS; i++) {
210 u32 value, p;
211 u64 offset;
212
213 if (msrpm_offsets[i] == 0xffffffff)
214 break;
215
216 p = msrpm_offsets[i];
217 offset = svm->nested.ctl.msrpm_base_pa + (p * 4);
218
219 if (kvm_vcpu_read_guest(&svm->vcpu, offset, &value, 4))
220 return false;
221
222 svm->nested.msrpm[p] = svm->msrpm[p] | value;
223 }
224
225 svm->vmcb->control.msrpm_base_pa = __sme_set(__pa(svm->nested.msrpm));
226
227 return true;
228 }
229
230 /*
231 * Bits 11:0 of bitmap address are ignored by hardware
232 */
nested_svm_check_bitmap_pa(struct kvm_vcpu * vcpu,u64 pa,u32 size)233 static bool nested_svm_check_bitmap_pa(struct kvm_vcpu *vcpu, u64 pa, u32 size)
234 {
235 u64 addr = PAGE_ALIGN(pa);
236
237 return kvm_vcpu_is_legal_gpa(vcpu, addr) &&
238 kvm_vcpu_is_legal_gpa(vcpu, addr + size - 1);
239 }
240
nested_vmcb_check_controls(struct kvm_vcpu * vcpu,struct vmcb_control_area * control)241 static bool nested_vmcb_check_controls(struct kvm_vcpu *vcpu,
242 struct vmcb_control_area *control)
243 {
244 if (CC(!vmcb_is_intercept(control, INTERCEPT_VMRUN)))
245 return false;
246
247 if (CC(control->asid == 0))
248 return false;
249
250 if (CC((control->nested_ctl & SVM_NESTED_CTL_NP_ENABLE) && !npt_enabled))
251 return false;
252
253 if (CC(!nested_svm_check_bitmap_pa(vcpu, control->msrpm_base_pa,
254 MSRPM_SIZE)))
255 return false;
256 if (CC(!nested_svm_check_bitmap_pa(vcpu, control->iopm_base_pa,
257 IOPM_SIZE)))
258 return false;
259
260 return true;
261 }
262
nested_vmcb_check_cr3_cr4(struct kvm_vcpu * vcpu,struct vmcb_save_area * save)263 static bool nested_vmcb_check_cr3_cr4(struct kvm_vcpu *vcpu,
264 struct vmcb_save_area *save)
265 {
266 /*
267 * These checks are also performed by KVM_SET_SREGS,
268 * except that EFER.LMA is not checked by SVM against
269 * CR0.PG && EFER.LME.
270 */
271 if ((save->efer & EFER_LME) && (save->cr0 & X86_CR0_PG)) {
272 if (CC(!(save->cr4 & X86_CR4_PAE)) ||
273 CC(!(save->cr0 & X86_CR0_PE)) ||
274 CC(kvm_vcpu_is_illegal_gpa(vcpu, save->cr3)))
275 return false;
276 }
277
278 if (CC(!kvm_is_valid_cr4(vcpu, save->cr4)))
279 return false;
280
281 return true;
282 }
283
284 /* Common checks that apply to both L1 and L2 state. */
nested_vmcb_valid_sregs(struct kvm_vcpu * vcpu,struct vmcb_save_area * save)285 static bool nested_vmcb_valid_sregs(struct kvm_vcpu *vcpu,
286 struct vmcb_save_area *save)
287 {
288 /*
289 * FIXME: these should be done after copying the fields,
290 * to avoid TOC/TOU races. For these save area checks
291 * the possible damage is limited since kvm_set_cr0 and
292 * kvm_set_cr4 handle failure; EFER_SVME is an exception
293 * so it is force-set later in nested_prepare_vmcb_save.
294 */
295 if (CC(!(save->efer & EFER_SVME)))
296 return false;
297
298 if (CC((save->cr0 & X86_CR0_CD) == 0 && (save->cr0 & X86_CR0_NW)) ||
299 CC(save->cr0 & ~0xffffffffULL))
300 return false;
301
302 if (CC(!kvm_dr6_valid(save->dr6)) || CC(!kvm_dr7_valid(save->dr7)))
303 return false;
304
305 if (!nested_vmcb_check_cr3_cr4(vcpu, save))
306 return false;
307
308 if (CC(!kvm_valid_efer(vcpu, save->efer)))
309 return false;
310
311 return true;
312 }
313
nested_load_control_from_vmcb12(struct vcpu_svm * svm,struct vmcb_control_area * control)314 void nested_load_control_from_vmcb12(struct vcpu_svm *svm,
315 struct vmcb_control_area *control)
316 {
317 copy_vmcb_control_area(&svm->nested.ctl, control);
318
319 /* Copy it here because nested_svm_check_controls will check it. */
320 svm->nested.ctl.asid = control->asid;
321 svm->nested.ctl.msrpm_base_pa &= ~0x0fffULL;
322 svm->nested.ctl.iopm_base_pa &= ~0x0fffULL;
323 }
324
325 /*
326 * Synchronize fields that are written by the processor, so that
327 * they can be copied back into the vmcb12.
328 */
nested_sync_control_from_vmcb02(struct vcpu_svm * svm)329 void nested_sync_control_from_vmcb02(struct vcpu_svm *svm)
330 {
331 u32 mask;
332 svm->nested.ctl.event_inj = svm->vmcb->control.event_inj;
333 svm->nested.ctl.event_inj_err = svm->vmcb->control.event_inj_err;
334
335 /* Only a few fields of int_ctl are written by the processor. */
336 mask = V_IRQ_MASK | V_TPR_MASK;
337 if (!(svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK) &&
338 svm_is_intercept(svm, INTERCEPT_VINTR)) {
339 /*
340 * In order to request an interrupt window, L0 is usurping
341 * svm->vmcb->control.int_ctl and possibly setting V_IRQ
342 * even if it was clear in L1's VMCB. Restoring it would be
343 * wrong. However, in this case V_IRQ will remain true until
344 * interrupt_window_interception calls svm_clear_vintr and
345 * restores int_ctl. We can just leave it aside.
346 */
347 mask &= ~V_IRQ_MASK;
348 }
349 svm->nested.ctl.int_ctl &= ~mask;
350 svm->nested.ctl.int_ctl |= svm->vmcb->control.int_ctl & mask;
351 }
352
353 /*
354 * Transfer any event that L0 or L1 wanted to inject into L2 to
355 * EXIT_INT_INFO.
356 */
nested_save_pending_event_to_vmcb12(struct vcpu_svm * svm,struct vmcb * vmcb12)357 static void nested_save_pending_event_to_vmcb12(struct vcpu_svm *svm,
358 struct vmcb *vmcb12)
359 {
360 struct kvm_vcpu *vcpu = &svm->vcpu;
361 u32 exit_int_info = 0;
362 unsigned int nr;
363
364 if (vcpu->arch.exception.injected) {
365 nr = vcpu->arch.exception.nr;
366 exit_int_info = nr | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT;
367
368 if (vcpu->arch.exception.has_error_code) {
369 exit_int_info |= SVM_EVTINJ_VALID_ERR;
370 vmcb12->control.exit_int_info_err =
371 vcpu->arch.exception.error_code;
372 }
373
374 } else if (vcpu->arch.nmi_injected) {
375 exit_int_info = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
376
377 } else if (vcpu->arch.interrupt.injected) {
378 nr = vcpu->arch.interrupt.nr;
379 exit_int_info = nr | SVM_EVTINJ_VALID;
380
381 if (vcpu->arch.interrupt.soft)
382 exit_int_info |= SVM_EVTINJ_TYPE_SOFT;
383 else
384 exit_int_info |= SVM_EVTINJ_TYPE_INTR;
385 }
386
387 vmcb12->control.exit_int_info = exit_int_info;
388 }
389
nested_npt_enabled(struct vcpu_svm * svm)390 static inline bool nested_npt_enabled(struct vcpu_svm *svm)
391 {
392 return svm->nested.ctl.nested_ctl & SVM_NESTED_CTL_NP_ENABLE;
393 }
394
nested_svm_transition_tlb_flush(struct kvm_vcpu * vcpu)395 static void nested_svm_transition_tlb_flush(struct kvm_vcpu *vcpu)
396 {
397 /*
398 * TODO: optimize unconditional TLB flush/MMU sync. A partial list of
399 * things to fix before this can be conditional:
400 *
401 * - Flush TLBs for both L1 and L2 remote TLB flush
402 * - Honor L1's request to flush an ASID on nested VMRUN
403 * - Sync nested NPT MMU on VMRUN that flushes L2's ASID[*]
404 * - Don't crush a pending TLB flush in vmcb02 on nested VMRUN
405 * - Flush L1's ASID on KVM_REQ_TLB_FLUSH_GUEST
406 *
407 * [*] Unlike nested EPT, SVM's ASID management can invalidate nested
408 * NPT guest-physical mappings on VMRUN.
409 */
410 kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
411 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
412 }
413
414 /*
415 * Load guest's/host's cr3 on nested vmentry or vmexit. @nested_npt is true
416 * if we are emulating VM-Entry into a guest with NPT enabled.
417 */
nested_svm_load_cr3(struct kvm_vcpu * vcpu,unsigned long cr3,bool nested_npt,bool reload_pdptrs)418 static int nested_svm_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3,
419 bool nested_npt, bool reload_pdptrs)
420 {
421 if (CC(kvm_vcpu_is_illegal_gpa(vcpu, cr3)))
422 return -EINVAL;
423
424 if (reload_pdptrs && !nested_npt && is_pae_paging(vcpu) &&
425 CC(!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)))
426 return -EINVAL;
427
428 if (!nested_npt)
429 kvm_mmu_new_pgd(vcpu, cr3);
430
431 vcpu->arch.cr3 = cr3;
432 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
433
434 /* Re-initialize the MMU, e.g. to pick up CR4 MMU role changes. */
435 kvm_init_mmu(vcpu);
436
437 return 0;
438 }
439
nested_vmcb02_compute_g_pat(struct vcpu_svm * svm)440 void nested_vmcb02_compute_g_pat(struct vcpu_svm *svm)
441 {
442 if (!svm->nested.vmcb02.ptr)
443 return;
444
445 /* FIXME: merge g_pat from vmcb01 and vmcb12. */
446 svm->nested.vmcb02.ptr->save.g_pat = svm->vmcb01.ptr->save.g_pat;
447 }
448
nested_vmcb02_prepare_save(struct vcpu_svm * svm,struct vmcb * vmcb12)449 static void nested_vmcb02_prepare_save(struct vcpu_svm *svm, struct vmcb *vmcb12)
450 {
451 bool new_vmcb12 = false;
452
453 nested_vmcb02_compute_g_pat(svm);
454
455 /* Load the nested guest state */
456 if (svm->nested.vmcb12_gpa != svm->nested.last_vmcb12_gpa) {
457 new_vmcb12 = true;
458 svm->nested.last_vmcb12_gpa = svm->nested.vmcb12_gpa;
459 }
460
461 if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_SEG))) {
462 svm->vmcb->save.es = vmcb12->save.es;
463 svm->vmcb->save.cs = vmcb12->save.cs;
464 svm->vmcb->save.ss = vmcb12->save.ss;
465 svm->vmcb->save.ds = vmcb12->save.ds;
466 svm->vmcb->save.cpl = vmcb12->save.cpl;
467 vmcb_mark_dirty(svm->vmcb, VMCB_SEG);
468 }
469
470 if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_DT))) {
471 svm->vmcb->save.gdtr = vmcb12->save.gdtr;
472 svm->vmcb->save.idtr = vmcb12->save.idtr;
473 vmcb_mark_dirty(svm->vmcb, VMCB_DT);
474 }
475
476 kvm_set_rflags(&svm->vcpu, vmcb12->save.rflags | X86_EFLAGS_FIXED);
477
478 /*
479 * Force-set EFER_SVME even though it is checked earlier on the
480 * VMCB12, because the guest can flip the bit between the check
481 * and now. Clearing EFER_SVME would call svm_free_nested.
482 */
483 svm_set_efer(&svm->vcpu, vmcb12->save.efer | EFER_SVME);
484
485 svm_set_cr0(&svm->vcpu, vmcb12->save.cr0);
486 svm_set_cr4(&svm->vcpu, vmcb12->save.cr4);
487
488 svm->vcpu.arch.cr2 = vmcb12->save.cr2;
489
490 kvm_rax_write(&svm->vcpu, vmcb12->save.rax);
491 kvm_rsp_write(&svm->vcpu, vmcb12->save.rsp);
492 kvm_rip_write(&svm->vcpu, vmcb12->save.rip);
493
494 /* In case we don't even reach vcpu_run, the fields are not updated */
495 svm->vmcb->save.rax = vmcb12->save.rax;
496 svm->vmcb->save.rsp = vmcb12->save.rsp;
497 svm->vmcb->save.rip = vmcb12->save.rip;
498
499 /* These bits will be set properly on the first execution when new_vmc12 is true */
500 if (unlikely(new_vmcb12 || vmcb_is_dirty(vmcb12, VMCB_DR))) {
501 svm->vmcb->save.dr7 = vmcb12->save.dr7 | DR7_FIXED_1;
502 svm->vcpu.arch.dr6 = vmcb12->save.dr6 | DR6_ACTIVE_LOW;
503 vmcb_mark_dirty(svm->vmcb, VMCB_DR);
504 }
505 }
506
nested_vmcb02_prepare_control(struct vcpu_svm * svm)507 static void nested_vmcb02_prepare_control(struct vcpu_svm *svm)
508 {
509 const u32 int_ctl_vmcb01_bits =
510 V_INTR_MASKING_MASK | V_GIF_MASK | V_GIF_ENABLE_MASK;
511
512 const u32 int_ctl_vmcb12_bits = V_TPR_MASK | V_IRQ_INJECTION_BITS_MASK;
513
514 struct kvm_vcpu *vcpu = &svm->vcpu;
515
516 /*
517 * Filled at exit: exit_code, exit_code_hi, exit_info_1, exit_info_2,
518 * exit_int_info, exit_int_info_err, next_rip, insn_len, insn_bytes.
519 */
520
521 /*
522 * Also covers avic_vapic_bar, avic_backing_page, avic_logical_id,
523 * avic_physical_id.
524 */
525 WARN_ON(kvm_apicv_activated(svm->vcpu.kvm));
526
527 /* Copied from vmcb01. msrpm_base can be overwritten later. */
528 svm->vmcb->control.nested_ctl = svm->vmcb01.ptr->control.nested_ctl;
529 svm->vmcb->control.iopm_base_pa = svm->vmcb01.ptr->control.iopm_base_pa;
530 svm->vmcb->control.msrpm_base_pa = svm->vmcb01.ptr->control.msrpm_base_pa;
531
532 /* Done at vmrun: asid. */
533
534 /* Also overwritten later if necessary. */
535 svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
536
537 /* nested_cr3. */
538 if (nested_npt_enabled(svm))
539 nested_svm_init_mmu_context(vcpu);
540
541 svm->vmcb->control.tsc_offset = vcpu->arch.tsc_offset =
542 vcpu->arch.l1_tsc_offset + svm->nested.ctl.tsc_offset;
543
544 svm->vmcb->control.int_ctl =
545 (svm->nested.ctl.int_ctl & int_ctl_vmcb12_bits) |
546 (svm->vmcb01.ptr->control.int_ctl & int_ctl_vmcb01_bits);
547
548 svm->vmcb->control.int_vector = svm->nested.ctl.int_vector;
549 svm->vmcb->control.int_state = svm->nested.ctl.int_state;
550 svm->vmcb->control.event_inj = svm->nested.ctl.event_inj;
551 svm->vmcb->control.event_inj_err = svm->nested.ctl.event_inj_err;
552
553 svm->vmcb->control.pause_filter_count = svm->nested.ctl.pause_filter_count;
554 svm->vmcb->control.pause_filter_thresh = svm->nested.ctl.pause_filter_thresh;
555
556 nested_svm_transition_tlb_flush(vcpu);
557
558 /* Enter Guest-Mode */
559 enter_guest_mode(vcpu);
560
561 /*
562 * Merge guest and host intercepts - must be called with vcpu in
563 * guest-mode to take effect.
564 */
565 recalc_intercepts(svm);
566 }
567
nested_svm_copy_common_state(struct vmcb * from_vmcb,struct vmcb * to_vmcb)568 static void nested_svm_copy_common_state(struct vmcb *from_vmcb, struct vmcb *to_vmcb)
569 {
570 /*
571 * Some VMCB state is shared between L1 and L2 and thus has to be
572 * moved at the time of nested vmrun and vmexit.
573 *
574 * VMLOAD/VMSAVE state would also belong in this category, but KVM
575 * always performs VMLOAD and VMSAVE from the VMCB01.
576 */
577 to_vmcb->save.spec_ctrl = from_vmcb->save.spec_ctrl;
578 }
579
enter_svm_guest_mode(struct kvm_vcpu * vcpu,u64 vmcb12_gpa,struct vmcb * vmcb12,bool from_vmrun)580 int enter_svm_guest_mode(struct kvm_vcpu *vcpu, u64 vmcb12_gpa,
581 struct vmcb *vmcb12, bool from_vmrun)
582 {
583 struct vcpu_svm *svm = to_svm(vcpu);
584 int ret;
585
586 trace_kvm_nested_vmrun(svm->vmcb->save.rip, vmcb12_gpa,
587 vmcb12->save.rip,
588 vmcb12->control.int_ctl,
589 vmcb12->control.event_inj,
590 vmcb12->control.nested_ctl);
591
592 trace_kvm_nested_intercepts(vmcb12->control.intercepts[INTERCEPT_CR] & 0xffff,
593 vmcb12->control.intercepts[INTERCEPT_CR] >> 16,
594 vmcb12->control.intercepts[INTERCEPT_EXCEPTION],
595 vmcb12->control.intercepts[INTERCEPT_WORD3],
596 vmcb12->control.intercepts[INTERCEPT_WORD4],
597 vmcb12->control.intercepts[INTERCEPT_WORD5]);
598
599
600 svm->nested.vmcb12_gpa = vmcb12_gpa;
601
602 WARN_ON(svm->vmcb == svm->nested.vmcb02.ptr);
603
604 nested_svm_copy_common_state(svm->vmcb01.ptr, svm->nested.vmcb02.ptr);
605
606 svm_switch_vmcb(svm, &svm->nested.vmcb02);
607 nested_vmcb02_prepare_control(svm);
608 nested_vmcb02_prepare_save(svm, vmcb12);
609
610 ret = nested_svm_load_cr3(&svm->vcpu, vmcb12->save.cr3,
611 nested_npt_enabled(svm), from_vmrun);
612 if (ret)
613 return ret;
614
615 if (!npt_enabled)
616 vcpu->arch.mmu->inject_page_fault = svm_inject_page_fault_nested;
617
618 if (!from_vmrun)
619 kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
620
621 svm_set_gif(svm, true);
622
623 return 0;
624 }
625
nested_svm_vmrun(struct kvm_vcpu * vcpu)626 int nested_svm_vmrun(struct kvm_vcpu *vcpu)
627 {
628 struct vcpu_svm *svm = to_svm(vcpu);
629 int ret;
630 struct vmcb *vmcb12;
631 struct kvm_host_map map;
632 u64 vmcb12_gpa;
633
634 if (!svm->nested.hsave_msr) {
635 kvm_inject_gp(vcpu, 0);
636 return 1;
637 }
638
639 if (is_smm(vcpu)) {
640 kvm_queue_exception(vcpu, UD_VECTOR);
641 return 1;
642 }
643
644 vmcb12_gpa = svm->vmcb->save.rax;
645 ret = kvm_vcpu_map(vcpu, gpa_to_gfn(vmcb12_gpa), &map);
646 if (ret == -EINVAL) {
647 kvm_inject_gp(vcpu, 0);
648 return 1;
649 } else if (ret) {
650 return kvm_skip_emulated_instruction(vcpu);
651 }
652
653 ret = kvm_skip_emulated_instruction(vcpu);
654
655 vmcb12 = map.hva;
656
657 if (WARN_ON_ONCE(!svm->nested.initialized))
658 return -EINVAL;
659
660 nested_load_control_from_vmcb12(svm, &vmcb12->control);
661
662 if (!nested_vmcb_valid_sregs(vcpu, &vmcb12->save) ||
663 !nested_vmcb_check_controls(vcpu, &svm->nested.ctl)) {
664 vmcb12->control.exit_code = SVM_EXIT_ERR;
665 vmcb12->control.exit_code_hi = 0;
666 vmcb12->control.exit_info_1 = 0;
667 vmcb12->control.exit_info_2 = 0;
668 goto out;
669 }
670
671 /*
672 * Since vmcb01 is not in use, we can use it to store some of the L1
673 * state.
674 */
675 svm->vmcb01.ptr->save.efer = vcpu->arch.efer;
676 svm->vmcb01.ptr->save.cr0 = kvm_read_cr0(vcpu);
677 svm->vmcb01.ptr->save.cr4 = vcpu->arch.cr4;
678 svm->vmcb01.ptr->save.rflags = kvm_get_rflags(vcpu);
679 svm->vmcb01.ptr->save.rip = kvm_rip_read(vcpu);
680
681 if (!npt_enabled)
682 svm->vmcb01.ptr->save.cr3 = kvm_read_cr3(vcpu);
683
684 svm->nested.nested_run_pending = 1;
685
686 if (enter_svm_guest_mode(vcpu, vmcb12_gpa, vmcb12, true))
687 goto out_exit_err;
688
689 if (nested_svm_vmrun_msrpm(svm))
690 goto out;
691
692 out_exit_err:
693 svm->nested.nested_run_pending = 0;
694
695 svm->vmcb->control.exit_code = SVM_EXIT_ERR;
696 svm->vmcb->control.exit_code_hi = 0;
697 svm->vmcb->control.exit_info_1 = 0;
698 svm->vmcb->control.exit_info_2 = 0;
699
700 nested_svm_vmexit(svm);
701
702 out:
703 kvm_vcpu_unmap(vcpu, &map, true);
704
705 return ret;
706 }
707
708 /* Copy state save area fields which are handled by VMRUN */
svm_copy_vmrun_state(struct vmcb_save_area * to_save,struct vmcb_save_area * from_save)709 void svm_copy_vmrun_state(struct vmcb_save_area *to_save,
710 struct vmcb_save_area *from_save)
711 {
712 to_save->es = from_save->es;
713 to_save->cs = from_save->cs;
714 to_save->ss = from_save->ss;
715 to_save->ds = from_save->ds;
716 to_save->gdtr = from_save->gdtr;
717 to_save->idtr = from_save->idtr;
718 to_save->rflags = from_save->rflags | X86_EFLAGS_FIXED;
719 to_save->efer = from_save->efer;
720 to_save->cr0 = from_save->cr0;
721 to_save->cr3 = from_save->cr3;
722 to_save->cr4 = from_save->cr4;
723 to_save->rax = from_save->rax;
724 to_save->rsp = from_save->rsp;
725 to_save->rip = from_save->rip;
726 to_save->cpl = 0;
727 }
728
svm_copy_vmloadsave_state(struct vmcb * to_vmcb,struct vmcb * from_vmcb)729 void svm_copy_vmloadsave_state(struct vmcb *to_vmcb, struct vmcb *from_vmcb)
730 {
731 to_vmcb->save.fs = from_vmcb->save.fs;
732 to_vmcb->save.gs = from_vmcb->save.gs;
733 to_vmcb->save.tr = from_vmcb->save.tr;
734 to_vmcb->save.ldtr = from_vmcb->save.ldtr;
735 to_vmcb->save.kernel_gs_base = from_vmcb->save.kernel_gs_base;
736 to_vmcb->save.star = from_vmcb->save.star;
737 to_vmcb->save.lstar = from_vmcb->save.lstar;
738 to_vmcb->save.cstar = from_vmcb->save.cstar;
739 to_vmcb->save.sfmask = from_vmcb->save.sfmask;
740 to_vmcb->save.sysenter_cs = from_vmcb->save.sysenter_cs;
741 to_vmcb->save.sysenter_esp = from_vmcb->save.sysenter_esp;
742 to_vmcb->save.sysenter_eip = from_vmcb->save.sysenter_eip;
743 }
744
nested_svm_vmexit(struct vcpu_svm * svm)745 int nested_svm_vmexit(struct vcpu_svm *svm)
746 {
747 struct kvm_vcpu *vcpu = &svm->vcpu;
748 struct vmcb *vmcb12;
749 struct vmcb *vmcb = svm->vmcb;
750 struct kvm_host_map map;
751 int rc;
752
753 /* Triple faults in L2 should never escape. */
754 WARN_ON_ONCE(kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu));
755
756 rc = kvm_vcpu_map(vcpu, gpa_to_gfn(svm->nested.vmcb12_gpa), &map);
757 if (rc) {
758 if (rc == -EINVAL)
759 kvm_inject_gp(vcpu, 0);
760 return 1;
761 }
762
763 vmcb12 = map.hva;
764
765 /* Exit Guest-Mode */
766 leave_guest_mode(vcpu);
767 svm->nested.vmcb12_gpa = 0;
768 WARN_ON_ONCE(svm->nested.nested_run_pending);
769
770 kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
771
772 /* in case we halted in L2 */
773 svm->vcpu.arch.mp_state = KVM_MP_STATE_RUNNABLE;
774
775 /* Give the current vmcb to the guest */
776
777 vmcb12->save.es = vmcb->save.es;
778 vmcb12->save.cs = vmcb->save.cs;
779 vmcb12->save.ss = vmcb->save.ss;
780 vmcb12->save.ds = vmcb->save.ds;
781 vmcb12->save.gdtr = vmcb->save.gdtr;
782 vmcb12->save.idtr = vmcb->save.idtr;
783 vmcb12->save.efer = svm->vcpu.arch.efer;
784 vmcb12->save.cr0 = kvm_read_cr0(vcpu);
785 vmcb12->save.cr3 = kvm_read_cr3(vcpu);
786 vmcb12->save.cr2 = vmcb->save.cr2;
787 vmcb12->save.cr4 = svm->vcpu.arch.cr4;
788 vmcb12->save.rflags = kvm_get_rflags(vcpu);
789 vmcb12->save.rip = kvm_rip_read(vcpu);
790 vmcb12->save.rsp = kvm_rsp_read(vcpu);
791 vmcb12->save.rax = kvm_rax_read(vcpu);
792 vmcb12->save.dr7 = vmcb->save.dr7;
793 vmcb12->save.dr6 = svm->vcpu.arch.dr6;
794 vmcb12->save.cpl = vmcb->save.cpl;
795
796 vmcb12->control.int_state = vmcb->control.int_state;
797 vmcb12->control.exit_code = vmcb->control.exit_code;
798 vmcb12->control.exit_code_hi = vmcb->control.exit_code_hi;
799 vmcb12->control.exit_info_1 = vmcb->control.exit_info_1;
800 vmcb12->control.exit_info_2 = vmcb->control.exit_info_2;
801
802 if (vmcb12->control.exit_code != SVM_EXIT_ERR)
803 nested_save_pending_event_to_vmcb12(svm, vmcb12);
804
805 if (svm->nrips_enabled)
806 vmcb12->control.next_rip = vmcb->control.next_rip;
807
808 vmcb12->control.int_ctl = svm->nested.ctl.int_ctl;
809 vmcb12->control.tlb_ctl = svm->nested.ctl.tlb_ctl;
810 vmcb12->control.event_inj = svm->nested.ctl.event_inj;
811 vmcb12->control.event_inj_err = svm->nested.ctl.event_inj_err;
812
813 vmcb12->control.pause_filter_count =
814 svm->vmcb->control.pause_filter_count;
815 vmcb12->control.pause_filter_thresh =
816 svm->vmcb->control.pause_filter_thresh;
817
818 nested_svm_copy_common_state(svm->nested.vmcb02.ptr, svm->vmcb01.ptr);
819
820 svm_switch_vmcb(svm, &svm->vmcb01);
821
822 /*
823 * On vmexit the GIF is set to false and
824 * no event can be injected in L1.
825 */
826 svm_set_gif(svm, false);
827 svm->vmcb->control.exit_int_info = 0;
828
829 svm->vcpu.arch.tsc_offset = svm->vcpu.arch.l1_tsc_offset;
830 if (svm->vmcb->control.tsc_offset != svm->vcpu.arch.tsc_offset) {
831 svm->vmcb->control.tsc_offset = svm->vcpu.arch.tsc_offset;
832 vmcb_mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
833 }
834
835 svm->nested.ctl.nested_cr3 = 0;
836
837 /*
838 * Restore processor state that had been saved in vmcb01
839 */
840 kvm_set_rflags(vcpu, svm->vmcb->save.rflags);
841 svm_set_efer(vcpu, svm->vmcb->save.efer);
842 svm_set_cr0(vcpu, svm->vmcb->save.cr0 | X86_CR0_PE);
843 svm_set_cr4(vcpu, svm->vmcb->save.cr4);
844 kvm_rax_write(vcpu, svm->vmcb->save.rax);
845 kvm_rsp_write(vcpu, svm->vmcb->save.rsp);
846 kvm_rip_write(vcpu, svm->vmcb->save.rip);
847
848 svm->vcpu.arch.dr7 = DR7_FIXED_1;
849 kvm_update_dr7(&svm->vcpu);
850
851 trace_kvm_nested_vmexit_inject(vmcb12->control.exit_code,
852 vmcb12->control.exit_info_1,
853 vmcb12->control.exit_info_2,
854 vmcb12->control.exit_int_info,
855 vmcb12->control.exit_int_info_err,
856 KVM_ISA_SVM);
857
858 kvm_vcpu_unmap(vcpu, &map, true);
859
860 nested_svm_transition_tlb_flush(vcpu);
861
862 nested_svm_uninit_mmu_context(vcpu);
863
864 rc = nested_svm_load_cr3(vcpu, svm->vmcb->save.cr3, false, true);
865 if (rc)
866 return 1;
867
868 /*
869 * Drop what we picked up for L2 via svm_complete_interrupts() so it
870 * doesn't end up in L1.
871 */
872 svm->vcpu.arch.nmi_injected = false;
873 kvm_clear_exception_queue(vcpu);
874 kvm_clear_interrupt_queue(vcpu);
875
876 /*
877 * If we are here following the completion of a VMRUN that
878 * is being single-stepped, queue the pending #DB intercept
879 * right now so that it an be accounted for before we execute
880 * L1's next instruction.
881 */
882 if (unlikely(svm->vmcb->save.rflags & X86_EFLAGS_TF))
883 kvm_queue_exception(&(svm->vcpu), DB_VECTOR);
884
885 return 0;
886 }
887
nested_svm_triple_fault(struct kvm_vcpu * vcpu)888 static void nested_svm_triple_fault(struct kvm_vcpu *vcpu)
889 {
890 nested_svm_simple_vmexit(to_svm(vcpu), SVM_EXIT_SHUTDOWN);
891 }
892
svm_allocate_nested(struct vcpu_svm * svm)893 int svm_allocate_nested(struct vcpu_svm *svm)
894 {
895 struct page *vmcb02_page;
896
897 if (svm->nested.initialized)
898 return 0;
899
900 vmcb02_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
901 if (!vmcb02_page)
902 return -ENOMEM;
903 svm->nested.vmcb02.ptr = page_address(vmcb02_page);
904 svm->nested.vmcb02.pa = __sme_set(page_to_pfn(vmcb02_page) << PAGE_SHIFT);
905
906 svm->nested.msrpm = svm_vcpu_alloc_msrpm();
907 if (!svm->nested.msrpm)
908 goto err_free_vmcb02;
909 svm_vcpu_init_msrpm(&svm->vcpu, svm->nested.msrpm);
910
911 svm->nested.initialized = true;
912 return 0;
913
914 err_free_vmcb02:
915 __free_page(vmcb02_page);
916 return -ENOMEM;
917 }
918
svm_free_nested(struct vcpu_svm * svm)919 void svm_free_nested(struct vcpu_svm *svm)
920 {
921 if (!svm->nested.initialized)
922 return;
923
924 svm_vcpu_free_msrpm(svm->nested.msrpm);
925 svm->nested.msrpm = NULL;
926
927 __free_page(virt_to_page(svm->nested.vmcb02.ptr));
928 svm->nested.vmcb02.ptr = NULL;
929
930 /*
931 * When last_vmcb12_gpa matches the current vmcb12 gpa,
932 * some vmcb12 fields are not loaded if they are marked clean
933 * in the vmcb12, since in this case they are up to date already.
934 *
935 * When the vmcb02 is freed, this optimization becomes invalid.
936 */
937 svm->nested.last_vmcb12_gpa = INVALID_GPA;
938
939 svm->nested.initialized = false;
940 }
941
942 /*
943 * Forcibly leave nested mode in order to be able to reset the VCPU later on.
944 */
svm_leave_nested(struct vcpu_svm * svm)945 void svm_leave_nested(struct vcpu_svm *svm)
946 {
947 struct kvm_vcpu *vcpu = &svm->vcpu;
948
949 if (is_guest_mode(vcpu)) {
950 svm->nested.nested_run_pending = 0;
951 svm->nested.vmcb12_gpa = INVALID_GPA;
952
953 leave_guest_mode(vcpu);
954
955 svm_switch_vmcb(svm, &svm->vmcb01);
956
957 nested_svm_uninit_mmu_context(vcpu);
958 vmcb_mark_all_dirty(svm->vmcb);
959 }
960
961 kvm_clear_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
962 }
963
nested_svm_exit_handled_msr(struct vcpu_svm * svm)964 static int nested_svm_exit_handled_msr(struct vcpu_svm *svm)
965 {
966 u32 offset, msr, value;
967 int write, mask;
968
969 if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_MSR_PROT)))
970 return NESTED_EXIT_HOST;
971
972 msr = svm->vcpu.arch.regs[VCPU_REGS_RCX];
973 offset = svm_msrpm_offset(msr);
974 write = svm->vmcb->control.exit_info_1 & 1;
975 mask = 1 << ((2 * (msr & 0xf)) + write);
976
977 if (offset == MSR_INVALID)
978 return NESTED_EXIT_DONE;
979
980 /* Offset is in 32 bit units but need in 8 bit units */
981 offset *= 4;
982
983 if (kvm_vcpu_read_guest(&svm->vcpu, svm->nested.ctl.msrpm_base_pa + offset, &value, 4))
984 return NESTED_EXIT_DONE;
985
986 return (value & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
987 }
988
nested_svm_intercept_ioio(struct vcpu_svm * svm)989 static int nested_svm_intercept_ioio(struct vcpu_svm *svm)
990 {
991 unsigned port, size, iopm_len;
992 u16 val, mask;
993 u8 start_bit;
994 u64 gpa;
995
996 if (!(vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_IOIO_PROT)))
997 return NESTED_EXIT_HOST;
998
999 port = svm->vmcb->control.exit_info_1 >> 16;
1000 size = (svm->vmcb->control.exit_info_1 & SVM_IOIO_SIZE_MASK) >>
1001 SVM_IOIO_SIZE_SHIFT;
1002 gpa = svm->nested.ctl.iopm_base_pa + (port / 8);
1003 start_bit = port % 8;
1004 iopm_len = (start_bit + size > 8) ? 2 : 1;
1005 mask = (0xf >> (4 - size)) << start_bit;
1006 val = 0;
1007
1008 if (kvm_vcpu_read_guest(&svm->vcpu, gpa, &val, iopm_len))
1009 return NESTED_EXIT_DONE;
1010
1011 return (val & mask) ? NESTED_EXIT_DONE : NESTED_EXIT_HOST;
1012 }
1013
nested_svm_intercept(struct vcpu_svm * svm)1014 static int nested_svm_intercept(struct vcpu_svm *svm)
1015 {
1016 u32 exit_code = svm->vmcb->control.exit_code;
1017 int vmexit = NESTED_EXIT_HOST;
1018
1019 switch (exit_code) {
1020 case SVM_EXIT_MSR:
1021 vmexit = nested_svm_exit_handled_msr(svm);
1022 break;
1023 case SVM_EXIT_IOIO:
1024 vmexit = nested_svm_intercept_ioio(svm);
1025 break;
1026 case SVM_EXIT_READ_CR0 ... SVM_EXIT_WRITE_CR8: {
1027 if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
1028 vmexit = NESTED_EXIT_DONE;
1029 break;
1030 }
1031 case SVM_EXIT_READ_DR0 ... SVM_EXIT_WRITE_DR7: {
1032 if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
1033 vmexit = NESTED_EXIT_DONE;
1034 break;
1035 }
1036 case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1037 /*
1038 * Host-intercepted exceptions have been checked already in
1039 * nested_svm_exit_special. There is nothing to do here,
1040 * the vmexit is injected by svm_check_nested_events.
1041 */
1042 vmexit = NESTED_EXIT_DONE;
1043 break;
1044 }
1045 case SVM_EXIT_ERR: {
1046 vmexit = NESTED_EXIT_DONE;
1047 break;
1048 }
1049 default: {
1050 if (vmcb_is_intercept(&svm->nested.ctl, exit_code))
1051 vmexit = NESTED_EXIT_DONE;
1052 }
1053 }
1054
1055 return vmexit;
1056 }
1057
nested_svm_exit_handled(struct vcpu_svm * svm)1058 int nested_svm_exit_handled(struct vcpu_svm *svm)
1059 {
1060 int vmexit;
1061
1062 vmexit = nested_svm_intercept(svm);
1063
1064 if (vmexit == NESTED_EXIT_DONE)
1065 nested_svm_vmexit(svm);
1066
1067 return vmexit;
1068 }
1069
nested_svm_check_permissions(struct kvm_vcpu * vcpu)1070 int nested_svm_check_permissions(struct kvm_vcpu *vcpu)
1071 {
1072 if (!(vcpu->arch.efer & EFER_SVME) || !is_paging(vcpu)) {
1073 kvm_queue_exception(vcpu, UD_VECTOR);
1074 return 1;
1075 }
1076
1077 if (to_svm(vcpu)->vmcb->save.cpl) {
1078 kvm_inject_gp(vcpu, 0);
1079 return 1;
1080 }
1081
1082 return 0;
1083 }
1084
nested_exit_on_exception(struct vcpu_svm * svm)1085 static bool nested_exit_on_exception(struct vcpu_svm *svm)
1086 {
1087 unsigned int nr = svm->vcpu.arch.exception.nr;
1088
1089 return (svm->nested.ctl.intercepts[INTERCEPT_EXCEPTION] & BIT(nr));
1090 }
1091
nested_svm_inject_exception_vmexit(struct vcpu_svm * svm)1092 static void nested_svm_inject_exception_vmexit(struct vcpu_svm *svm)
1093 {
1094 unsigned int nr = svm->vcpu.arch.exception.nr;
1095
1096 svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
1097 svm->vmcb->control.exit_code_hi = 0;
1098
1099 if (svm->vcpu.arch.exception.has_error_code)
1100 svm->vmcb->control.exit_info_1 = svm->vcpu.arch.exception.error_code;
1101
1102 /*
1103 * EXITINFO2 is undefined for all exception intercepts other
1104 * than #PF.
1105 */
1106 if (nr == PF_VECTOR) {
1107 if (svm->vcpu.arch.exception.nested_apf)
1108 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.apf.nested_apf_token;
1109 else if (svm->vcpu.arch.exception.has_payload)
1110 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.exception.payload;
1111 else
1112 svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
1113 } else if (nr == DB_VECTOR) {
1114 /* See inject_pending_event. */
1115 kvm_deliver_exception_payload(&svm->vcpu);
1116 if (svm->vcpu.arch.dr7 & DR7_GD) {
1117 svm->vcpu.arch.dr7 &= ~DR7_GD;
1118 kvm_update_dr7(&svm->vcpu);
1119 }
1120 } else
1121 WARN_ON(svm->vcpu.arch.exception.has_payload);
1122
1123 nested_svm_vmexit(svm);
1124 }
1125
nested_exit_on_init(struct vcpu_svm * svm)1126 static inline bool nested_exit_on_init(struct vcpu_svm *svm)
1127 {
1128 return vmcb_is_intercept(&svm->nested.ctl, INTERCEPT_INIT);
1129 }
1130
svm_check_nested_events(struct kvm_vcpu * vcpu)1131 static int svm_check_nested_events(struct kvm_vcpu *vcpu)
1132 {
1133 struct vcpu_svm *svm = to_svm(vcpu);
1134 bool block_nested_events =
1135 kvm_event_needs_reinjection(vcpu) || svm->nested.nested_run_pending;
1136 struct kvm_lapic *apic = vcpu->arch.apic;
1137
1138 if (lapic_in_kernel(vcpu) &&
1139 test_bit(KVM_APIC_INIT, &apic->pending_events)) {
1140 if (block_nested_events)
1141 return -EBUSY;
1142 if (!nested_exit_on_init(svm))
1143 return 0;
1144 nested_svm_simple_vmexit(svm, SVM_EXIT_INIT);
1145 return 0;
1146 }
1147
1148 if (vcpu->arch.exception.pending) {
1149 /*
1150 * Only a pending nested run can block a pending exception.
1151 * Otherwise an injected NMI/interrupt should either be
1152 * lost or delivered to the nested hypervisor in the EXITINTINFO
1153 * vmcb field, while delivering the pending exception.
1154 */
1155 if (svm->nested.nested_run_pending)
1156 return -EBUSY;
1157 if (!nested_exit_on_exception(svm))
1158 return 0;
1159 nested_svm_inject_exception_vmexit(svm);
1160 return 0;
1161 }
1162
1163 if (vcpu->arch.smi_pending && !svm_smi_blocked(vcpu)) {
1164 if (block_nested_events)
1165 return -EBUSY;
1166 if (!nested_exit_on_smi(svm))
1167 return 0;
1168 nested_svm_simple_vmexit(svm, SVM_EXIT_SMI);
1169 return 0;
1170 }
1171
1172 if (vcpu->arch.nmi_pending && !svm_nmi_blocked(vcpu)) {
1173 if (block_nested_events)
1174 return -EBUSY;
1175 if (!nested_exit_on_nmi(svm))
1176 return 0;
1177 nested_svm_simple_vmexit(svm, SVM_EXIT_NMI);
1178 return 0;
1179 }
1180
1181 if (kvm_cpu_has_interrupt(vcpu) && !svm_interrupt_blocked(vcpu)) {
1182 if (block_nested_events)
1183 return -EBUSY;
1184 if (!nested_exit_on_intr(svm))
1185 return 0;
1186 trace_kvm_nested_intr_vmexit(svm->vmcb->save.rip);
1187 nested_svm_simple_vmexit(svm, SVM_EXIT_INTR);
1188 return 0;
1189 }
1190
1191 return 0;
1192 }
1193
nested_svm_exit_special(struct vcpu_svm * svm)1194 int nested_svm_exit_special(struct vcpu_svm *svm)
1195 {
1196 u32 exit_code = svm->vmcb->control.exit_code;
1197
1198 switch (exit_code) {
1199 case SVM_EXIT_INTR:
1200 case SVM_EXIT_NMI:
1201 case SVM_EXIT_NPF:
1202 return NESTED_EXIT_HOST;
1203 case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
1204 u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
1205
1206 if (svm->vmcb01.ptr->control.intercepts[INTERCEPT_EXCEPTION] &
1207 excp_bits)
1208 return NESTED_EXIT_HOST;
1209 else if (exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR &&
1210 svm->vcpu.arch.apf.host_apf_flags)
1211 /* Trap async PF even if not shadowing */
1212 return NESTED_EXIT_HOST;
1213 break;
1214 }
1215 default:
1216 break;
1217 }
1218
1219 return NESTED_EXIT_CONTINUE;
1220 }
1221
svm_get_nested_state(struct kvm_vcpu * vcpu,struct kvm_nested_state __user * user_kvm_nested_state,u32 user_data_size)1222 static int svm_get_nested_state(struct kvm_vcpu *vcpu,
1223 struct kvm_nested_state __user *user_kvm_nested_state,
1224 u32 user_data_size)
1225 {
1226 struct vcpu_svm *svm;
1227 struct kvm_nested_state kvm_state = {
1228 .flags = 0,
1229 .format = KVM_STATE_NESTED_FORMAT_SVM,
1230 .size = sizeof(kvm_state),
1231 };
1232 struct vmcb __user *user_vmcb = (struct vmcb __user *)
1233 &user_kvm_nested_state->data.svm[0];
1234
1235 if (!vcpu)
1236 return kvm_state.size + KVM_STATE_NESTED_SVM_VMCB_SIZE;
1237
1238 svm = to_svm(vcpu);
1239
1240 if (user_data_size < kvm_state.size)
1241 goto out;
1242
1243 /* First fill in the header and copy it out. */
1244 if (is_guest_mode(vcpu)) {
1245 kvm_state.hdr.svm.vmcb_pa = svm->nested.vmcb12_gpa;
1246 kvm_state.size += KVM_STATE_NESTED_SVM_VMCB_SIZE;
1247 kvm_state.flags |= KVM_STATE_NESTED_GUEST_MODE;
1248
1249 if (svm->nested.nested_run_pending)
1250 kvm_state.flags |= KVM_STATE_NESTED_RUN_PENDING;
1251 }
1252
1253 if (gif_set(svm))
1254 kvm_state.flags |= KVM_STATE_NESTED_GIF_SET;
1255
1256 if (copy_to_user(user_kvm_nested_state, &kvm_state, sizeof(kvm_state)))
1257 return -EFAULT;
1258
1259 if (!is_guest_mode(vcpu))
1260 goto out;
1261
1262 /*
1263 * Copy over the full size of the VMCB rather than just the size
1264 * of the structs.
1265 */
1266 if (clear_user(user_vmcb, KVM_STATE_NESTED_SVM_VMCB_SIZE))
1267 return -EFAULT;
1268 if (copy_to_user(&user_vmcb->control, &svm->nested.ctl,
1269 sizeof(user_vmcb->control)))
1270 return -EFAULT;
1271 if (copy_to_user(&user_vmcb->save, &svm->vmcb01.ptr->save,
1272 sizeof(user_vmcb->save)))
1273 return -EFAULT;
1274 out:
1275 return kvm_state.size;
1276 }
1277
svm_set_nested_state(struct kvm_vcpu * vcpu,struct kvm_nested_state __user * user_kvm_nested_state,struct kvm_nested_state * kvm_state)1278 static int svm_set_nested_state(struct kvm_vcpu *vcpu,
1279 struct kvm_nested_state __user *user_kvm_nested_state,
1280 struct kvm_nested_state *kvm_state)
1281 {
1282 struct vcpu_svm *svm = to_svm(vcpu);
1283 struct vmcb __user *user_vmcb = (struct vmcb __user *)
1284 &user_kvm_nested_state->data.svm[0];
1285 struct vmcb_control_area *ctl;
1286 struct vmcb_save_area *save;
1287 unsigned long cr0;
1288 int ret;
1289
1290 BUILD_BUG_ON(sizeof(struct vmcb_control_area) + sizeof(struct vmcb_save_area) >
1291 KVM_STATE_NESTED_SVM_VMCB_SIZE);
1292
1293 if (kvm_state->format != KVM_STATE_NESTED_FORMAT_SVM)
1294 return -EINVAL;
1295
1296 if (kvm_state->flags & ~(KVM_STATE_NESTED_GUEST_MODE |
1297 KVM_STATE_NESTED_RUN_PENDING |
1298 KVM_STATE_NESTED_GIF_SET))
1299 return -EINVAL;
1300
1301 /*
1302 * If in guest mode, vcpu->arch.efer actually refers to the L2 guest's
1303 * EFER.SVME, but EFER.SVME still has to be 1 for VMRUN to succeed.
1304 */
1305 if (!(vcpu->arch.efer & EFER_SVME)) {
1306 /* GIF=1 and no guest mode are required if SVME=0. */
1307 if (kvm_state->flags != KVM_STATE_NESTED_GIF_SET)
1308 return -EINVAL;
1309 }
1310
1311 /* SMM temporarily disables SVM, so we cannot be in guest mode. */
1312 if (is_smm(vcpu) && (kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE))
1313 return -EINVAL;
1314
1315 if (!(kvm_state->flags & KVM_STATE_NESTED_GUEST_MODE)) {
1316 svm_leave_nested(svm);
1317 svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
1318 return 0;
1319 }
1320
1321 if (!page_address_valid(vcpu, kvm_state->hdr.svm.vmcb_pa))
1322 return -EINVAL;
1323 if (kvm_state->size < sizeof(*kvm_state) + KVM_STATE_NESTED_SVM_VMCB_SIZE)
1324 return -EINVAL;
1325
1326 ret = -ENOMEM;
1327 ctl = kzalloc(sizeof(*ctl), GFP_KERNEL_ACCOUNT);
1328 save = kzalloc(sizeof(*save), GFP_KERNEL_ACCOUNT);
1329 if (!ctl || !save)
1330 goto out_free;
1331
1332 ret = -EFAULT;
1333 if (copy_from_user(ctl, &user_vmcb->control, sizeof(*ctl)))
1334 goto out_free;
1335 if (copy_from_user(save, &user_vmcb->save, sizeof(*save)))
1336 goto out_free;
1337
1338 ret = -EINVAL;
1339 if (!nested_vmcb_check_controls(vcpu, ctl))
1340 goto out_free;
1341
1342 /*
1343 * Processor state contains L2 state. Check that it is
1344 * valid for guest mode (see nested_vmcb_check_save).
1345 */
1346 cr0 = kvm_read_cr0(vcpu);
1347 if (((cr0 & X86_CR0_CD) == 0) && (cr0 & X86_CR0_NW))
1348 goto out_free;
1349
1350 /*
1351 * Validate host state saved from before VMRUN (see
1352 * nested_svm_check_permissions).
1353 */
1354 if (!(save->cr0 & X86_CR0_PG) ||
1355 !(save->cr0 & X86_CR0_PE) ||
1356 (save->rflags & X86_EFLAGS_VM) ||
1357 !nested_vmcb_valid_sregs(vcpu, save))
1358 goto out_free;
1359
1360 /*
1361 * While the nested guest CR3 is already checked and set by
1362 * KVM_SET_SREGS, it was set when nested state was yet loaded,
1363 * thus MMU might not be initialized correctly.
1364 * Set it again to fix this.
1365 */
1366
1367 ret = nested_svm_load_cr3(&svm->vcpu, vcpu->arch.cr3,
1368 nested_npt_enabled(svm), false);
1369 if (WARN_ON_ONCE(ret))
1370 goto out_free;
1371
1372
1373 /*
1374 * All checks done, we can enter guest mode. Userspace provides
1375 * vmcb12.control, which will be combined with L1 and stored into
1376 * vmcb02, and the L1 save state which we store in vmcb01.
1377 * L2 registers if needed are moved from the current VMCB to VMCB02.
1378 */
1379
1380 if (is_guest_mode(vcpu))
1381 svm_leave_nested(svm);
1382 else
1383 svm->nested.vmcb02.ptr->save = svm->vmcb01.ptr->save;
1384
1385 svm_set_gif(svm, !!(kvm_state->flags & KVM_STATE_NESTED_GIF_SET));
1386
1387 svm->nested.nested_run_pending =
1388 !!(kvm_state->flags & KVM_STATE_NESTED_RUN_PENDING);
1389
1390 svm->nested.vmcb12_gpa = kvm_state->hdr.svm.vmcb_pa;
1391
1392 svm_copy_vmrun_state(&svm->vmcb01.ptr->save, save);
1393 nested_load_control_from_vmcb12(svm, ctl);
1394
1395 svm_switch_vmcb(svm, &svm->nested.vmcb02);
1396 nested_vmcb02_prepare_control(svm);
1397 kvm_make_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu);
1398 ret = 0;
1399 out_free:
1400 kfree(save);
1401 kfree(ctl);
1402
1403 return ret;
1404 }
1405
svm_get_nested_state_pages(struct kvm_vcpu * vcpu)1406 static bool svm_get_nested_state_pages(struct kvm_vcpu *vcpu)
1407 {
1408 struct vcpu_svm *svm = to_svm(vcpu);
1409
1410 if (WARN_ON(!is_guest_mode(vcpu)))
1411 return true;
1412
1413 if (!vcpu->arch.pdptrs_from_userspace &&
1414 !nested_npt_enabled(svm) && is_pae_paging(vcpu))
1415 /*
1416 * Reload the guest's PDPTRs since after a migration
1417 * the guest CR3 might be restored prior to setting the nested
1418 * state which can lead to a load of wrong PDPTRs.
1419 */
1420 if (CC(!load_pdptrs(vcpu, vcpu->arch.walk_mmu, vcpu->arch.cr3)))
1421 return false;
1422
1423 if (!nested_svm_vmrun_msrpm(svm)) {
1424 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
1425 vcpu->run->internal.suberror =
1426 KVM_INTERNAL_ERROR_EMULATION;
1427 vcpu->run->internal.ndata = 0;
1428 return false;
1429 }
1430
1431 return true;
1432 }
1433
1434 struct kvm_x86_nested_ops svm_nested_ops = {
1435 .check_events = svm_check_nested_events,
1436 .triple_fault = nested_svm_triple_fault,
1437 .get_nested_state_pages = svm_get_nested_state_pages,
1438 .get_state = svm_get_nested_state,
1439 .set_state = svm_set_nested_state,
1440 };
1441
