VirtualBox

source: vbox/trunk/src/VBox/VMM/VMMAll/IEMAllCImplVmxInstr.cpp.h@ 82812

Last change on this file since 82812 was 82705, checked in by vboxsync, 5 years ago

VMM/IEM: Removed obsolete todo.
  • Property svn:eol-style set to native
  • Property svn:keywords set to Author Date Id Revision
File size: 353.4 KB
Line 
1/* $Id: IEMAllCImplVmxInstr.cpp.h 82705 2020-01-10 05:26:47Z vboxsync $ */
2/** @file
3 * IEM - VT-x instruction implementation.
4 */
5
6/*
7 * Copyright (C) 2011-2019 Oracle Corporation
8 *
9 * This file is part of VirtualBox Open Source Edition (OSE), as
10 * available from http://www.virtualbox.org. This file is free software;
11 * you can redistribute it and/or modify it under the terms of the GNU
12 * General Public License (GPL) as published by the Free Software
13 * Foundation, in version 2 as it comes in the "COPYING" file of the
14 * VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15 * hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16 */
17
18
19/*********************************************************************************************************************************
20* Defined Constants And Macros *
21*********************************************************************************************************************************/
22#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
23/**
24 * Gets the ModR/M, SIB and displacement byte(s) from decoded opcodes given their
25 * relative offsets.
26 */
27# ifdef IEM_WITH_CODE_TLB
28# define IEM_MODRM_GET_U8(a_pVCpu, a_bModRm, a_offModRm) do { } while (0)
29# define IEM_SIB_GET_U8(a_pVCpu, a_bSib, a_offSib) do { } while (0)
30# define IEM_DISP_GET_U16(a_pVCpu, a_u16Disp, a_offDisp) do { } while (0)
31# define IEM_DISP_GET_S8_SX_U16(a_pVCpu, a_u16Disp, a_offDisp) do { } while (0)
32# define IEM_DISP_GET_U32(a_pVCpu, a_u32Disp, a_offDisp) do { } while (0)
33# define IEM_DISP_GET_S8_SX_U32(a_pVCpu, a_u32Disp, a_offDisp) do { } while (0)
34# define IEM_DISP_GET_S32_SX_U64(a_pVCpu, a_u64Disp, a_offDisp) do { } while (0)
35# define IEM_DISP_GET_S8_SX_U64(a_pVCpu, a_u64Disp, a_offDisp) do { } while (0)
36# error "Implement me: Getting ModR/M, SIB, displacement needs to work even when instruction crosses a page boundary."
37# else /* !IEM_WITH_CODE_TLB */
38# define IEM_MODRM_GET_U8(a_pVCpu, a_bModRm, a_offModRm) \
39 do \
40 { \
41 Assert((a_offModRm) < (a_pVCpu)->iem.s.cbOpcode); \
42 (a_bModRm) = (a_pVCpu)->iem.s.abOpcode[(a_offModRm)]; \
43 } while (0)
44
45# define IEM_SIB_GET_U8(a_pVCpu, a_bSib, a_offSib) IEM_MODRM_GET_U8(a_pVCpu, a_bSib, a_offSib)
46
47# define IEM_DISP_GET_U16(a_pVCpu, a_u16Disp, a_offDisp) \
48 do \
49 { \
50 Assert((a_offDisp) + 1 < (a_pVCpu)->iem.s.cbOpcode); \
51 uint8_t const bTmpLo = (a_pVCpu)->iem.s.abOpcode[(a_offDisp)]; \
52 uint8_t const bTmpHi = (a_pVCpu)->iem.s.abOpcode[(a_offDisp) + 1]; \
53 (a_u16Disp) = RT_MAKE_U16(bTmpLo, bTmpHi); \
54 } while (0)
55
56# define IEM_DISP_GET_S8_SX_U16(a_pVCpu, a_u16Disp, a_offDisp) \
57 do \
58 { \
59 Assert((a_offDisp) < (a_pVCpu)->iem.s.cbOpcode); \
60 (a_u16Disp) = (int8_t)((a_pVCpu)->iem.s.abOpcode[(a_offDisp)]); \
61 } while (0)
62
63# define IEM_DISP_GET_U32(a_pVCpu, a_u32Disp, a_offDisp) \
64 do \
65 { \
66 Assert((a_offDisp) + 3 < (a_pVCpu)->iem.s.cbOpcode); \
67 uint8_t const bTmp0 = (a_pVCpu)->iem.s.abOpcode[(a_offDisp)]; \
68 uint8_t const bTmp1 = (a_pVCpu)->iem.s.abOpcode[(a_offDisp) + 1]; \
69 uint8_t const bTmp2 = (a_pVCpu)->iem.s.abOpcode[(a_offDisp) + 2]; \
70 uint8_t const bTmp3 = (a_pVCpu)->iem.s.abOpcode[(a_offDisp) + 3]; \
71 (a_u32Disp) = RT_MAKE_U32_FROM_U8(bTmp0, bTmp1, bTmp2, bTmp3); \
72 } while (0)
73
74# define IEM_DISP_GET_S8_SX_U32(a_pVCpu, a_u32Disp, a_offDisp) \
75 do \
76 { \
77 Assert((a_offDisp) + 1 < (a_pVCpu)->iem.s.cbOpcode); \
78 (a_u32Disp) = (int8_t)((a_pVCpu)->iem.s.abOpcode[(a_offDisp)]); \
79 } while (0)
80
81# define IEM_DISP_GET_S8_SX_U64(a_pVCpu, a_u64Disp, a_offDisp) \
82 do \
83 { \
84 Assert((a_offDisp) + 1 < (a_pVCpu)->iem.s.cbOpcode); \
85 (a_u64Disp) = (int8_t)((a_pVCpu)->iem.s.abOpcode[(a_offDisp)]); \
86 } while (0)
87
88# define IEM_DISP_GET_S32_SX_U64(a_pVCpu, a_u64Disp, a_offDisp) \
89 do \
90 { \
91 Assert((a_offDisp) + 3 < (a_pVCpu)->iem.s.cbOpcode); \
92 uint8_t const bTmp0 = (a_pVCpu)->iem.s.abOpcode[(a_offDisp)]; \
93 uint8_t const bTmp1 = (a_pVCpu)->iem.s.abOpcode[(a_offDisp) + 1]; \
94 uint8_t const bTmp2 = (a_pVCpu)->iem.s.abOpcode[(a_offDisp) + 2]; \
95 uint8_t const bTmp3 = (a_pVCpu)->iem.s.abOpcode[(a_offDisp) + 3]; \
96 (a_u64Disp) = (int32_t)RT_MAKE_U32_FROM_U8(bTmp0, bTmp1, bTmp2, bTmp3); \
97 } while (0)
98# endif /* !IEM_WITH_CODE_TLB */
99
100/** Gets the guest-physical address of the shadows VMCS for the given VCPU. */
101# define IEM_VMX_GET_SHADOW_VMCS(a_pVCpu) ((a_pVCpu)->cpum.GstCtx.hwvirt.vmx.GCPhysShadowVmcs)
102
103/** Whether a shadow VMCS is present for the given VCPU. */
104# define IEM_VMX_HAS_SHADOW_VMCS(a_pVCpu) RT_BOOL(IEM_VMX_GET_SHADOW_VMCS(a_pVCpu) != NIL_RTGCPHYS)
105
106/** Gets the VMXON region pointer. */
107# define IEM_VMX_GET_VMXON_PTR(a_pVCpu) ((a_pVCpu)->cpum.GstCtx.hwvirt.vmx.GCPhysVmxon)
108
109/** Gets the guest-physical address of the current VMCS for the given VCPU. */
110# define IEM_VMX_GET_CURRENT_VMCS(a_pVCpu) ((a_pVCpu)->cpum.GstCtx.hwvirt.vmx.GCPhysVmcs)
111
112/** Whether a current VMCS is present for the given VCPU. */
113# define IEM_VMX_HAS_CURRENT_VMCS(a_pVCpu) RT_BOOL(IEM_VMX_GET_CURRENT_VMCS(a_pVCpu) != NIL_RTGCPHYS)
114
115/** Assigns the guest-physical address of the current VMCS for the given VCPU. */
116# define IEM_VMX_SET_CURRENT_VMCS(a_pVCpu, a_GCPhysVmcs) \
117 do \
118 { \
119 Assert((a_GCPhysVmcs) != NIL_RTGCPHYS); \
120 (a_pVCpu)->cpum.GstCtx.hwvirt.vmx.GCPhysVmcs = (a_GCPhysVmcs); \
121 } while (0)
122
123/** Clears any current VMCS for the given VCPU. */
124# define IEM_VMX_CLEAR_CURRENT_VMCS(a_pVCpu) \
125 do \
126 { \
127 (a_pVCpu)->cpum.GstCtx.hwvirt.vmx.GCPhysVmcs = NIL_RTGCPHYS; \
128 } while (0)
129
130/** Check for VMX instructions requiring to be in VMX operation.
131 * @note Any changes here, check if IEMOP_HLP_IN_VMX_OPERATION needs updating. */
132# define IEM_VMX_IN_VMX_OPERATION(a_pVCpu, a_szInstr, a_InsDiagPrefix) \
133 do \
134 { \
135 if (IEM_VMX_IS_ROOT_MODE(a_pVCpu)) \
136 { /* likely */ } \
137 else \
138 { \
139 Log((a_szInstr ": Not in VMX operation (root mode) -> #UD\n")); \
140 (a_pVCpu)->cpum.GstCtx.hwvirt.vmx.enmDiag = a_InsDiagPrefix##_VmxRoot; \
141 return iemRaiseUndefinedOpcode(a_pVCpu); \
142 } \
143 } while (0)
144
145/** Marks a VM-entry failure with a diagnostic reason, logs and returns. */
146# define IEM_VMX_VMENTRY_FAILED_RET(a_pVCpu, a_pszInstr, a_pszFailure, a_VmxDiag) \
147 do \
148 { \
149 LogRel(("%s: VM-entry failed! enmDiag=%u (%s) -> %s\n", (a_pszInstr), (a_VmxDiag), \
150 HMGetVmxDiagDesc(a_VmxDiag), (a_pszFailure))); \
151 (a_pVCpu)->cpum.GstCtx.hwvirt.vmx.enmDiag = (a_VmxDiag); \
152 return VERR_VMX_VMENTRY_FAILED; \
153 } while (0)
154
155/** Marks a VM-exit failure with a diagnostic reason, logs and returns. */
156# define IEM_VMX_VMEXIT_FAILED_RET(a_pVCpu, a_uExitReason, a_pszFailure, a_VmxDiag) \
157 do \
158 { \
159 LogRel(("VM-exit failed! uExitReason=%u enmDiag=%u (%s) -> %s\n", (a_uExitReason), (a_VmxDiag), \
160 HMGetVmxDiagDesc(a_VmxDiag), (a_pszFailure))); \
161 (a_pVCpu)->cpum.GstCtx.hwvirt.vmx.enmDiag = (a_VmxDiag); \
162 return VERR_VMX_VMEXIT_FAILED; \
163 } while (0)
164
165
166/*********************************************************************************************************************************
167* Global Variables *
168*********************************************************************************************************************************/
169/** @todo NSTVMX: The following VM-exit intercepts are pending:
170 * VMX_EXIT_IO_SMI
171 * VMX_EXIT_SMI
172 * VMX_EXIT_GETSEC
173 * VMX_EXIT_RSM
174 * VMX_EXIT_MONITOR (APIC access VM-exit caused by MONITOR pending)
175 * VMX_EXIT_ERR_MACHINE_CHECK (we never need to raise this?)
176 * VMX_EXIT_EPT_VIOLATION
177 * VMX_EXIT_EPT_MISCONFIG
178 * VMX_EXIT_INVEPT
179 * VMX_EXIT_RDRAND
180 * VMX_EXIT_VMFUNC
181 * VMX_EXIT_ENCLS
182 * VMX_EXIT_RDSEED
183 * VMX_EXIT_PML_FULL
184 * VMX_EXIT_XSAVES
185 * VMX_EXIT_XRSTORS
186 */
187/**
188 * Map of VMCS field encodings to their virtual-VMCS structure offsets.
189 *
190 * The first array dimension is VMCS field encoding of Width OR'ed with Type and the
191 * second dimension is the Index, see VMXVMCSFIELD.
192 */
193uint16_t const g_aoffVmcsMap[16][VMX_V_VMCS_MAX_INDEX + 1] =
194{
195 /* VMX_VMCSFIELD_WIDTH_16BIT | VMX_VMCSFIELD_TYPE_CONTROL: */
196 {
197 /* 0 */ RT_UOFFSETOF(VMXVVMCS, u16Vpid),
198 /* 1 */ RT_UOFFSETOF(VMXVVMCS, u16PostIntNotifyVector),
199 /* 2 */ RT_UOFFSETOF(VMXVVMCS, u16EptpIndex),
200 /* 3-10 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX,
201 /* 11-18 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX,
202 /* 19-25 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX
203 },
204 /* VMX_VMCSFIELD_WIDTH_16BIT | VMX_VMCSFIELD_TYPE_VMEXIT_INFO: */
205 {
206 /* 0-7 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX,
207 /* 8-15 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX,
208 /* 16-23 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX,
209 /* 24-25 */ UINT16_MAX, UINT16_MAX
210 },
211 /* VMX_VMCSFIELD_WIDTH_16BIT | VMX_VMCSFIELD_TYPE_GUEST_STATE: */
212 {
213 /* 0 */ RT_UOFFSETOF(VMXVVMCS, GuestEs),
214 /* 1 */ RT_UOFFSETOF(VMXVVMCS, GuestCs),
215 /* 2 */ RT_UOFFSETOF(VMXVVMCS, GuestSs),
216 /* 3 */ RT_UOFFSETOF(VMXVVMCS, GuestDs),
217 /* 4 */ RT_UOFFSETOF(VMXVVMCS, GuestFs),
218 /* 5 */ RT_UOFFSETOF(VMXVVMCS, GuestGs),
219 /* 6 */ RT_UOFFSETOF(VMXVVMCS, GuestLdtr),
220 /* 7 */ RT_UOFFSETOF(VMXVVMCS, GuestTr),
221 /* 8 */ RT_UOFFSETOF(VMXVVMCS, u16GuestIntStatus),
222 /* 9 */ RT_UOFFSETOF(VMXVVMCS, u16PmlIndex),
223 /* 10-17 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX,
224 /* 18-25 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX
225 },
226 /* VMX_VMCSFIELD_WIDTH_16BIT | VMX_VMCSFIELD_TYPE_HOST_STATE: */
227 {
228 /* 0 */ RT_UOFFSETOF(VMXVVMCS, HostEs),
229 /* 1 */ RT_UOFFSETOF(VMXVVMCS, HostCs),
230 /* 2 */ RT_UOFFSETOF(VMXVVMCS, HostSs),
231 /* 3 */ RT_UOFFSETOF(VMXVVMCS, HostDs),
232 /* 4 */ RT_UOFFSETOF(VMXVVMCS, HostFs),
233 /* 5 */ RT_UOFFSETOF(VMXVVMCS, HostGs),
234 /* 6 */ RT_UOFFSETOF(VMXVVMCS, HostTr),
235 /* 7-14 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX,
236 /* 15-22 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX,
237 /* 23-25 */ UINT16_MAX, UINT16_MAX, UINT16_MAX
238 },
239 /* VMX_VMCSFIELD_WIDTH_64BIT | VMX_VMCSFIELD_TYPE_CONTROL: */
240 {
241 /* 0 */ RT_UOFFSETOF(VMXVVMCS, u64AddrIoBitmapA),
242 /* 1 */ RT_UOFFSETOF(VMXVVMCS, u64AddrIoBitmapB),
243 /* 2 */ RT_UOFFSETOF(VMXVVMCS, u64AddrMsrBitmap),
244 /* 3 */ RT_UOFFSETOF(VMXVVMCS, u64AddrExitMsrStore),
245 /* 4 */ RT_UOFFSETOF(VMXVVMCS, u64AddrExitMsrLoad),
246 /* 5 */ RT_UOFFSETOF(VMXVVMCS, u64AddrEntryMsrLoad),
247 /* 6 */ RT_UOFFSETOF(VMXVVMCS, u64ExecVmcsPtr),
248 /* 7 */ RT_UOFFSETOF(VMXVVMCS, u64AddrPml),
249 /* 8 */ RT_UOFFSETOF(VMXVVMCS, u64TscOffset),
250 /* 9 */ RT_UOFFSETOF(VMXVVMCS, u64AddrVirtApic),
251 /* 10 */ RT_UOFFSETOF(VMXVVMCS, u64AddrApicAccess),
252 /* 11 */ RT_UOFFSETOF(VMXVVMCS, u64AddrPostedIntDesc),
253 /* 12 */ RT_UOFFSETOF(VMXVVMCS, u64VmFuncCtls),
254 /* 13 */ RT_UOFFSETOF(VMXVVMCS, u64EptpPtr),
255 /* 14 */ RT_UOFFSETOF(VMXVVMCS, u64EoiExitBitmap0),
256 /* 15 */ RT_UOFFSETOF(VMXVVMCS, u64EoiExitBitmap1),
257 /* 16 */ RT_UOFFSETOF(VMXVVMCS, u64EoiExitBitmap2),
258 /* 17 */ RT_UOFFSETOF(VMXVVMCS, u64EoiExitBitmap3),
259 /* 18 */ RT_UOFFSETOF(VMXVVMCS, u64AddrEptpList),
260 /* 19 */ RT_UOFFSETOF(VMXVVMCS, u64AddrVmreadBitmap),
261 /* 20 */ RT_UOFFSETOF(VMXVVMCS, u64AddrVmwriteBitmap),
262 /* 21 */ RT_UOFFSETOF(VMXVVMCS, u64AddrXcptVeInfo),
263 /* 22 */ RT_UOFFSETOF(VMXVVMCS, u64XssBitmap),
264 /* 23 */ RT_UOFFSETOF(VMXVVMCS, u64EnclsBitmap),
265 /* 24 */ RT_UOFFSETOF(VMXVVMCS, u64SpptPtr),
266 /* 25 */ RT_UOFFSETOF(VMXVVMCS, u64TscMultiplier)
267 },
268 /* VMX_VMCSFIELD_WIDTH_64BIT | VMX_VMCSFIELD_TYPE_VMEXIT_INFO: */
269 {
270 /* 0 */ RT_UOFFSETOF(VMXVVMCS, u64RoGuestPhysAddr),
271 /* 1-8 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX,
272 /* 9-16 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX,
273 /* 17-24 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX,
274 /* 25 */ UINT16_MAX
275 },
276 /* VMX_VMCSFIELD_WIDTH_64BIT | VMX_VMCSFIELD_TYPE_GUEST_STATE: */
277 {
278 /* 0 */ RT_UOFFSETOF(VMXVVMCS, u64VmcsLinkPtr),
279 /* 1 */ RT_UOFFSETOF(VMXVVMCS, u64GuestDebugCtlMsr),
280 /* 2 */ RT_UOFFSETOF(VMXVVMCS, u64GuestPatMsr),
281 /* 3 */ RT_UOFFSETOF(VMXVVMCS, u64GuestEferMsr),
282 /* 4 */ RT_UOFFSETOF(VMXVVMCS, u64GuestPerfGlobalCtlMsr),
283 /* 5 */ RT_UOFFSETOF(VMXVVMCS, u64GuestPdpte0),
284 /* 6 */ RT_UOFFSETOF(VMXVVMCS, u64GuestPdpte1),
285 /* 7 */ RT_UOFFSETOF(VMXVVMCS, u64GuestPdpte2),
286 /* 8 */ RT_UOFFSETOF(VMXVVMCS, u64GuestPdpte3),
287 /* 9 */ RT_UOFFSETOF(VMXVVMCS, u64GuestBndcfgsMsr),
288 /* 10 */ RT_UOFFSETOF(VMXVVMCS, u64GuestRtitCtlMsr),
289 /* 11-18 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX,
290 /* 19-25 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX
291 },
292 /* VMX_VMCSFIELD_WIDTH_64BIT | VMX_VMCSFIELD_TYPE_HOST_STATE: */
293 {
294 /* 0 */ RT_UOFFSETOF(VMXVVMCS, u64HostPatMsr),
295 /* 1 */ RT_UOFFSETOF(VMXVVMCS, u64HostEferMsr),
296 /* 2 */ RT_UOFFSETOF(VMXVVMCS, u64HostPerfGlobalCtlMsr),
297 /* 3-10 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX,
298 /* 11-18 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX,
299 /* 19-25 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX
300 },
301 /* VMX_VMCSFIELD_WIDTH_32BIT | VMX_VMCSFIELD_TYPE_CONTROL: */
302 {
303 /* 0 */ RT_UOFFSETOF(VMXVVMCS, u32PinCtls),
304 /* 1 */ RT_UOFFSETOF(VMXVVMCS, u32ProcCtls),
305 /* 2 */ RT_UOFFSETOF(VMXVVMCS, u32XcptBitmap),
306 /* 3 */ RT_UOFFSETOF(VMXVVMCS, u32XcptPFMask),
307 /* 4 */ RT_UOFFSETOF(VMXVVMCS, u32XcptPFMatch),
308 /* 5 */ RT_UOFFSETOF(VMXVVMCS, u32Cr3TargetCount),
309 /* 6 */ RT_UOFFSETOF(VMXVVMCS, u32ExitCtls),
310 /* 7 */ RT_UOFFSETOF(VMXVVMCS, u32ExitMsrStoreCount),
311 /* 8 */ RT_UOFFSETOF(VMXVVMCS, u32ExitMsrLoadCount),
312 /* 9 */ RT_UOFFSETOF(VMXVVMCS, u32EntryCtls),
313 /* 10 */ RT_UOFFSETOF(VMXVVMCS, u32EntryMsrLoadCount),
314 /* 11 */ RT_UOFFSETOF(VMXVVMCS, u32EntryIntInfo),
315 /* 12 */ RT_UOFFSETOF(VMXVVMCS, u32EntryXcptErrCode),
316 /* 13 */ RT_UOFFSETOF(VMXVVMCS, u32EntryInstrLen),
317 /* 14 */ RT_UOFFSETOF(VMXVVMCS, u32TprThreshold),
318 /* 15 */ RT_UOFFSETOF(VMXVVMCS, u32ProcCtls2),
319 /* 16 */ RT_UOFFSETOF(VMXVVMCS, u32PleGap),
320 /* 17 */ RT_UOFFSETOF(VMXVVMCS, u32PleWindow),
321 /* 18-25 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX
322 },
323 /* VMX_VMCSFIELD_WIDTH_32BIT | VMX_VMCSFIELD_TYPE_VMEXIT_INFO: */
324 {
325 /* 0 */ RT_UOFFSETOF(VMXVVMCS, u32RoVmInstrError),
326 /* 1 */ RT_UOFFSETOF(VMXVVMCS, u32RoExitReason),
327 /* 2 */ RT_UOFFSETOF(VMXVVMCS, u32RoExitIntInfo),
328 /* 3 */ RT_UOFFSETOF(VMXVVMCS, u32RoExitIntErrCode),
329 /* 4 */ RT_UOFFSETOF(VMXVVMCS, u32RoIdtVectoringInfo),
330 /* 5 */ RT_UOFFSETOF(VMXVVMCS, u32RoIdtVectoringErrCode),
331 /* 6 */ RT_UOFFSETOF(VMXVVMCS, u32RoExitInstrLen),
332 /* 7 */ RT_UOFFSETOF(VMXVVMCS, u32RoExitInstrInfo),
333 /* 8-15 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX,
334 /* 16-23 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX,
335 /* 24-25 */ UINT16_MAX, UINT16_MAX
336 },
337 /* VMX_VMCSFIELD_WIDTH_32BIT | VMX_VMCSFIELD_TYPE_GUEST_STATE: */
338 {
339 /* 0 */ RT_UOFFSETOF(VMXVVMCS, u32GuestEsLimit),
340 /* 1 */ RT_UOFFSETOF(VMXVVMCS, u32GuestCsLimit),
341 /* 2 */ RT_UOFFSETOF(VMXVVMCS, u32GuestSsLimit),
342 /* 3 */ RT_UOFFSETOF(VMXVVMCS, u32GuestDsLimit),
343 /* 4 */ RT_UOFFSETOF(VMXVVMCS, u32GuestFsLimit),
344 /* 5 */ RT_UOFFSETOF(VMXVVMCS, u32GuestGsLimit),
345 /* 6 */ RT_UOFFSETOF(VMXVVMCS, u32GuestLdtrLimit),
346 /* 7 */ RT_UOFFSETOF(VMXVVMCS, u32GuestTrLimit),
347 /* 8 */ RT_UOFFSETOF(VMXVVMCS, u32GuestGdtrLimit),
348 /* 9 */ RT_UOFFSETOF(VMXVVMCS, u32GuestIdtrLimit),
349 /* 10 */ RT_UOFFSETOF(VMXVVMCS, u32GuestEsAttr),
350 /* 11 */ RT_UOFFSETOF(VMXVVMCS, u32GuestCsAttr),
351 /* 12 */ RT_UOFFSETOF(VMXVVMCS, u32GuestSsAttr),
352 /* 13 */ RT_UOFFSETOF(VMXVVMCS, u32GuestDsAttr),
353 /* 14 */ RT_UOFFSETOF(VMXVVMCS, u32GuestFsAttr),
354 /* 15 */ RT_UOFFSETOF(VMXVVMCS, u32GuestGsAttr),
355 /* 16 */ RT_UOFFSETOF(VMXVVMCS, u32GuestLdtrAttr),
356 /* 17 */ RT_UOFFSETOF(VMXVVMCS, u32GuestTrAttr),
357 /* 18 */ RT_UOFFSETOF(VMXVVMCS, u32GuestIntrState),
358 /* 19 */ RT_UOFFSETOF(VMXVVMCS, u32GuestActivityState),
359 /* 20 */ RT_UOFFSETOF(VMXVVMCS, u32GuestSmBase),
360 /* 21 */ RT_UOFFSETOF(VMXVVMCS, u32GuestSysenterCS),
361 /* 22 */ UINT16_MAX,
362 /* 23 */ RT_UOFFSETOF(VMXVVMCS, u32PreemptTimer),
363 /* 24-25 */ UINT16_MAX, UINT16_MAX
364 },
365 /* VMX_VMCSFIELD_WIDTH_32BIT | VMX_VMCSFIELD_TYPE_HOST_STATE: */
366 {
367 /* 0 */ RT_UOFFSETOF(VMXVVMCS, u32HostSysenterCs),
368 /* 1-8 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX,
369 /* 9-16 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX,
370 /* 17-24 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX,
371 /* 25 */ UINT16_MAX
372 },
373 /* VMX_VMCSFIELD_WIDTH_NATURAL | VMX_VMCSFIELD_TYPE_CONTROL: */
374 {
375 /* 0 */ RT_UOFFSETOF(VMXVVMCS, u64Cr0Mask),
376 /* 1 */ RT_UOFFSETOF(VMXVVMCS, u64Cr4Mask),
377 /* 2 */ RT_UOFFSETOF(VMXVVMCS, u64Cr0ReadShadow),
378 /* 3 */ RT_UOFFSETOF(VMXVVMCS, u64Cr4ReadShadow),
379 /* 4 */ RT_UOFFSETOF(VMXVVMCS, u64Cr3Target0),
380 /* 5 */ RT_UOFFSETOF(VMXVVMCS, u64Cr3Target1),
381 /* 6 */ RT_UOFFSETOF(VMXVVMCS, u64Cr3Target2),
382 /* 7 */ RT_UOFFSETOF(VMXVVMCS, u64Cr3Target3),
383 /* 8-15 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX,
384 /* 16-23 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX,
385 /* 24-25 */ UINT16_MAX, UINT16_MAX
386 },
387 /* VMX_VMCSFIELD_WIDTH_NATURAL | VMX_VMCSFIELD_TYPE_VMEXIT_INFO: */
388 {
389 /* 0 */ RT_UOFFSETOF(VMXVVMCS, u64RoExitQual),
390 /* 1 */ RT_UOFFSETOF(VMXVVMCS, u64RoIoRcx),
391 /* 2 */ RT_UOFFSETOF(VMXVVMCS, u64RoIoRsi),
392 /* 3 */ RT_UOFFSETOF(VMXVVMCS, u64RoIoRdi),
393 /* 4 */ RT_UOFFSETOF(VMXVVMCS, u64RoIoRip),
394 /* 5 */ RT_UOFFSETOF(VMXVVMCS, u64RoGuestLinearAddr),
395 /* 6-13 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX,
396 /* 14-21 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX,
397 /* 22-25 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX
398 },
399 /* VMX_VMCSFIELD_WIDTH_NATURAL | VMX_VMCSFIELD_TYPE_GUEST_STATE: */
400 {
401 /* 0 */ RT_UOFFSETOF(VMXVVMCS, u64GuestCr0),
402 /* 1 */ RT_UOFFSETOF(VMXVVMCS, u64GuestCr3),
403 /* 2 */ RT_UOFFSETOF(VMXVVMCS, u64GuestCr4),
404 /* 3 */ RT_UOFFSETOF(VMXVVMCS, u64GuestEsBase),
405 /* 4 */ RT_UOFFSETOF(VMXVVMCS, u64GuestCsBase),
406 /* 5 */ RT_UOFFSETOF(VMXVVMCS, u64GuestSsBase),
407 /* 6 */ RT_UOFFSETOF(VMXVVMCS, u64GuestDsBase),
408 /* 7 */ RT_UOFFSETOF(VMXVVMCS, u64GuestFsBase),
409 /* 8 */ RT_UOFFSETOF(VMXVVMCS, u64GuestGsBase),
410 /* 9 */ RT_UOFFSETOF(VMXVVMCS, u64GuestLdtrBase),
411 /* 10 */ RT_UOFFSETOF(VMXVVMCS, u64GuestTrBase),
412 /* 11 */ RT_UOFFSETOF(VMXVVMCS, u64GuestGdtrBase),
413 /* 12 */ RT_UOFFSETOF(VMXVVMCS, u64GuestIdtrBase),
414 /* 13 */ RT_UOFFSETOF(VMXVVMCS, u64GuestDr7),
415 /* 14 */ RT_UOFFSETOF(VMXVVMCS, u64GuestRsp),
416 /* 15 */ RT_UOFFSETOF(VMXVVMCS, u64GuestRip),
417 /* 16 */ RT_UOFFSETOF(VMXVVMCS, u64GuestRFlags),
418 /* 17 */ RT_UOFFSETOF(VMXVVMCS, u64GuestPendingDbgXcpts),
419 /* 18 */ RT_UOFFSETOF(VMXVVMCS, u64GuestSysenterEsp),
420 /* 19 */ RT_UOFFSETOF(VMXVVMCS, u64GuestSysenterEip),
421 /* 20-25 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX
422 },
423 /* VMX_VMCSFIELD_WIDTH_NATURAL | VMX_VMCSFIELD_TYPE_HOST_STATE: */
424 {
425 /* 0 */ RT_UOFFSETOF(VMXVVMCS, u64HostCr0),
426 /* 1 */ RT_UOFFSETOF(VMXVVMCS, u64HostCr3),
427 /* 2 */ RT_UOFFSETOF(VMXVVMCS, u64HostCr4),
428 /* 3 */ RT_UOFFSETOF(VMXVVMCS, u64HostFsBase),
429 /* 4 */ RT_UOFFSETOF(VMXVVMCS, u64HostGsBase),
430 /* 5 */ RT_UOFFSETOF(VMXVVMCS, u64HostTrBase),
431 /* 6 */ RT_UOFFSETOF(VMXVVMCS, u64HostGdtrBase),
432 /* 7 */ RT_UOFFSETOF(VMXVVMCS, u64HostIdtrBase),
433 /* 8 */ RT_UOFFSETOF(VMXVVMCS, u64HostSysenterEsp),
434 /* 9 */ RT_UOFFSETOF(VMXVVMCS, u64HostSysenterEip),
435 /* 10 */ RT_UOFFSETOF(VMXVVMCS, u64HostRsp),
436 /* 11 */ RT_UOFFSETOF(VMXVVMCS, u64HostRip),
437 /* 12-19 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX,
438 /* 20-25 */ UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX, UINT16_MAX
439 }
440};
441
442
443/**
444 * Gets a host selector from the VMCS.
445 *
446 * @param pVmcs Pointer to the virtual VMCS.
447 * @param iSelReg The index of the segment register (X86_SREG_XXX).
448 */
449DECLINLINE(RTSEL) iemVmxVmcsGetHostSelReg(PCVMXVVMCS pVmcs, uint8_t iSegReg)
450{
451 Assert(iSegReg < X86_SREG_COUNT);
452 RTSEL HostSel;
453 uint8_t const uWidth = VMX_VMCSFIELD_WIDTH_16BIT;
454 uint8_t const uType = VMX_VMCSFIELD_TYPE_HOST_STATE;
455 uint8_t const uWidthType = (uWidth << 2) | uType;
456 uint8_t const uIndex = iSegReg + RT_BF_GET(VMX_VMCS16_HOST_ES_SEL, VMX_BF_VMCSFIELD_INDEX);
457 Assert(uIndex <= VMX_V_VMCS_MAX_INDEX);
458 uint16_t const offField = g_aoffVmcsMap[uWidthType][uIndex];
459 uint8_t const *pbVmcs = (uint8_t *)pVmcs;
460 uint8_t const *pbField = pbVmcs + offField;
461 HostSel = *(uint16_t *)pbField;
462 return HostSel;
463}
464
465
466/**
467 * Sets a guest segment register in the VMCS.
468 *
469 * @param pVmcs Pointer to the virtual VMCS.
470 * @param iSegReg The index of the segment register (X86_SREG_XXX).
471 * @param pSelReg Pointer to the segment register.
472 */
473IEM_STATIC void iemVmxVmcsSetGuestSegReg(PCVMXVVMCS pVmcs, uint8_t iSegReg, PCCPUMSELREG pSelReg)
474{
475 Assert(pSelReg);
476 Assert(iSegReg < X86_SREG_COUNT);
477
478 /* Selector. */
479 {
480 uint8_t const uWidth = VMX_VMCSFIELD_WIDTH_16BIT;
481 uint8_t const uType = VMX_VMCSFIELD_TYPE_GUEST_STATE;
482 uint8_t const uWidthType = (uWidth << 2) | uType;
483 uint8_t const uIndex = iSegReg + RT_BF_GET(VMX_VMCS16_GUEST_ES_SEL, VMX_BF_VMCSFIELD_INDEX);
484 Assert(uIndex <= VMX_V_VMCS_MAX_INDEX);
485 uint16_t const offField = g_aoffVmcsMap[uWidthType][uIndex];
486 uint8_t *pbVmcs = (uint8_t *)pVmcs;
487 uint8_t *pbField = pbVmcs + offField;
488 *(uint16_t *)pbField = pSelReg->Sel;
489 }
490
491 /* Limit. */
492 {
493 uint8_t const uWidth = VMX_VMCSFIELD_WIDTH_32BIT;
494 uint8_t const uType = VMX_VMCSFIELD_TYPE_GUEST_STATE;
495 uint8_t const uWidthType = (uWidth << 2) | uType;
496 uint8_t const uIndex = iSegReg + RT_BF_GET(VMX_VMCS32_GUEST_ES_LIMIT, VMX_BF_VMCSFIELD_INDEX);
497 Assert(uIndex <= VMX_V_VMCS_MAX_INDEX);
498 uint16_t const offField = g_aoffVmcsMap[uWidthType][uIndex];
499 uint8_t *pbVmcs = (uint8_t *)pVmcs;
500 uint8_t *pbField = pbVmcs + offField;
501 *(uint32_t *)pbField = pSelReg->u32Limit;
502 }
503
504 /* Base. */
505 {
506 uint8_t const uWidth = VMX_VMCSFIELD_WIDTH_NATURAL;
507 uint8_t const uType = VMX_VMCSFIELD_TYPE_GUEST_STATE;
508 uint8_t const uWidthType = (uWidth << 2) | uType;
509 uint8_t const uIndex = iSegReg + RT_BF_GET(VMX_VMCS_GUEST_ES_BASE, VMX_BF_VMCSFIELD_INDEX);
510 Assert(uIndex <= VMX_V_VMCS_MAX_INDEX);
511 uint16_t const offField = g_aoffVmcsMap[uWidthType][uIndex];
512 uint8_t const *pbVmcs = (uint8_t *)pVmcs;
513 uint8_t const *pbField = pbVmcs + offField;
514 *(uint64_t *)pbField = pSelReg->u64Base;
515 }
516
517 /* Attributes. */
518 {
519 uint32_t const fValidAttrMask = X86DESCATTR_TYPE | X86DESCATTR_DT | X86DESCATTR_DPL | X86DESCATTR_P
520 | X86DESCATTR_AVL | X86DESCATTR_L | X86DESCATTR_D | X86DESCATTR_G
521 | X86DESCATTR_UNUSABLE;
522 uint8_t const uWidth = VMX_VMCSFIELD_WIDTH_32BIT;
523 uint8_t const uType = VMX_VMCSFIELD_TYPE_GUEST_STATE;
524 uint8_t const uWidthType = (uWidth << 2) | uType;
525 uint8_t const uIndex = iSegReg + RT_BF_GET(VMX_VMCS32_GUEST_ES_ACCESS_RIGHTS, VMX_BF_VMCSFIELD_INDEX);
526 Assert(uIndex <= VMX_V_VMCS_MAX_INDEX);
527 uint16_t const offField = g_aoffVmcsMap[uWidthType][uIndex];
528 uint8_t *pbVmcs = (uint8_t *)pVmcs;
529 uint8_t *pbField = pbVmcs + offField;
530 *(uint32_t *)pbField = pSelReg->Attr.u & fValidAttrMask;
531 }
532}
533
534
535/**
536 * Gets a guest segment register from the VMCS.
537 *
538 * @returns VBox status code.
539 * @param pVmcs Pointer to the virtual VMCS.
540 * @param iSegReg The index of the segment register (X86_SREG_XXX).
541 * @param pSelReg Where to store the segment register (only updated when
542 * VINF_SUCCESS is returned).
543 *
544 * @remarks Warning! This does not validate the contents of the retrieved segment
545 * register.
546 */
547IEM_STATIC int iemVmxVmcsGetGuestSegReg(PCVMXVVMCS pVmcs, uint8_t iSegReg, PCPUMSELREG pSelReg)
548{
549 Assert(pSelReg);
550 Assert(iSegReg < X86_SREG_COUNT);
551
552 /* Selector. */
553 uint16_t u16Sel;
554 {
555 uint8_t const uWidth = VMX_VMCSFIELD_WIDTH_16BIT;
556 uint8_t const uType = VMX_VMCSFIELD_TYPE_GUEST_STATE;
557 uint8_t const uWidthType = (uWidth << 2) | uType;
558 uint8_t const uIndex = iSegReg + RT_BF_GET(VMX_VMCS16_GUEST_ES_SEL, VMX_BF_VMCSFIELD_INDEX);
559 AssertReturn(uIndex <= VMX_V_VMCS_MAX_INDEX, VERR_IEM_IPE_3);
560 uint16_t const offField = g_aoffVmcsMap[uWidthType][uIndex];
561 uint8_t const *pbVmcs = (uint8_t *)pVmcs;
562 uint8_t const *pbField = pbVmcs + offField;
563 u16Sel = *(uint16_t *)pbField;
564 }
565
566 /* Limit. */
567 uint32_t u32Limit;
568 {
569 uint8_t const uWidth = VMX_VMCSFIELD_WIDTH_32BIT;
570 uint8_t const uType = VMX_VMCSFIELD_TYPE_GUEST_STATE;
571 uint8_t const uWidthType = (uWidth << 2) | uType;
572 uint8_t const uIndex = iSegReg + RT_BF_GET(VMX_VMCS32_GUEST_ES_LIMIT, VMX_BF_VMCSFIELD_INDEX);
573 AssertReturn(uIndex <= VMX_V_VMCS_MAX_INDEX, VERR_IEM_IPE_3);
574 uint16_t const offField = g_aoffVmcsMap[uWidthType][uIndex];
575 uint8_t const *pbVmcs = (uint8_t *)pVmcs;
576 uint8_t const *pbField = pbVmcs + offField;
577 u32Limit = *(uint32_t *)pbField;
578 }
579
580 /* Base. */
581 uint64_t u64Base;
582 {
583 uint8_t const uWidth = VMX_VMCSFIELD_WIDTH_NATURAL;
584 uint8_t const uType = VMX_VMCSFIELD_TYPE_GUEST_STATE;
585 uint8_t const uWidthType = (uWidth << 2) | uType;
586 uint8_t const uIndex = iSegReg + RT_BF_GET(VMX_VMCS_GUEST_ES_BASE, VMX_BF_VMCSFIELD_INDEX);
587 AssertReturn(uIndex <= VMX_V_VMCS_MAX_INDEX, VERR_IEM_IPE_3);
588 uint16_t const offField = g_aoffVmcsMap[uWidthType][uIndex];
589 uint8_t const *pbVmcs = (uint8_t *)pVmcs;
590 uint8_t const *pbField = pbVmcs + offField;
591 u64Base = *(uint64_t *)pbField;
592 /** @todo NSTVMX: Should we zero out high bits here for 32-bit virtual CPUs? */
593 }
594
595 /* Attributes. */
596 uint32_t u32Attr;
597 {
598 uint8_t const uWidth = VMX_VMCSFIELD_WIDTH_32BIT;
599 uint8_t const uType = VMX_VMCSFIELD_TYPE_GUEST_STATE;
600 uint8_t const uWidthType = (uWidth << 2) | uType;
601 uint8_t const uIndex = iSegReg + RT_BF_GET(VMX_VMCS32_GUEST_ES_ACCESS_RIGHTS, VMX_BF_VMCSFIELD_INDEX);
602 AssertReturn(uIndex <= VMX_V_VMCS_MAX_INDEX, VERR_IEM_IPE_3);
603 uint16_t const offField = g_aoffVmcsMap[uWidthType][uIndex];
604 uint8_t const *pbVmcs = (uint8_t *)pVmcs;
605 uint8_t const *pbField = pbVmcs + offField;
606 u32Attr = *(uint32_t *)pbField;
607 }
608
609 pSelReg->Sel = u16Sel;
610 pSelReg->ValidSel = u16Sel;
611 pSelReg->fFlags = CPUMSELREG_FLAGS_VALID;
612 pSelReg->u32Limit = u32Limit;
613 pSelReg->u64Base = u64Base;
614 pSelReg->Attr.u = u32Attr;
615 return VINF_SUCCESS;
616}
617
618
619/**
620 * Converts an IEM exception event type to a VMX event type.
621 *
622 * @returns The VMX event type.
623 * @param uVector The interrupt / exception vector.
624 * @param fFlags The IEM event flag (see IEM_XCPT_FLAGS_XXX).
625 */
626DECLINLINE(uint8_t) iemVmxGetEventType(uint32_t uVector, uint32_t fFlags)
627{
628 /* Paranoia (callers may use these interchangeably). */
629 AssertCompile(VMX_EXIT_INT_INFO_TYPE_NMI == VMX_IDT_VECTORING_INFO_TYPE_NMI);
630 AssertCompile(VMX_EXIT_INT_INFO_TYPE_HW_XCPT == VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT);
631 AssertCompile(VMX_EXIT_INT_INFO_TYPE_EXT_INT == VMX_IDT_VECTORING_INFO_TYPE_EXT_INT);
632 AssertCompile(VMX_EXIT_INT_INFO_TYPE_SW_XCPT == VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT);
633 AssertCompile(VMX_EXIT_INT_INFO_TYPE_SW_INT == VMX_IDT_VECTORING_INFO_TYPE_SW_INT);
634 AssertCompile(VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT == VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT);
635 AssertCompile(VMX_EXIT_INT_INFO_TYPE_NMI == VMX_ENTRY_INT_INFO_TYPE_NMI);
636 AssertCompile(VMX_EXIT_INT_INFO_TYPE_HW_XCPT == VMX_ENTRY_INT_INFO_TYPE_HW_XCPT);
637 AssertCompile(VMX_EXIT_INT_INFO_TYPE_EXT_INT == VMX_ENTRY_INT_INFO_TYPE_EXT_INT);
638 AssertCompile(VMX_EXIT_INT_INFO_TYPE_SW_XCPT == VMX_ENTRY_INT_INFO_TYPE_SW_XCPT);
639 AssertCompile(VMX_EXIT_INT_INFO_TYPE_SW_INT == VMX_ENTRY_INT_INFO_TYPE_SW_INT);
640 AssertCompile(VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT == VMX_ENTRY_INT_INFO_TYPE_PRIV_SW_XCPT);
641
642 if (fFlags & IEM_XCPT_FLAGS_T_CPU_XCPT)
643 {
644 if (uVector == X86_XCPT_NMI)
645 return VMX_EXIT_INT_INFO_TYPE_NMI;
646 return VMX_EXIT_INT_INFO_TYPE_HW_XCPT;
647 }
648
649 if (fFlags & IEM_XCPT_FLAGS_T_SOFT_INT)
650 {
651 if (fFlags & (IEM_XCPT_FLAGS_BP_INSTR | IEM_XCPT_FLAGS_OF_INSTR))
652 return VMX_EXIT_INT_INFO_TYPE_SW_XCPT;
653 if (fFlags & IEM_XCPT_FLAGS_ICEBP_INSTR)
654 return VMX_EXIT_INT_INFO_TYPE_PRIV_SW_XCPT;
655 return VMX_EXIT_INT_INFO_TYPE_SW_INT;
656 }
657
658 Assert(fFlags & IEM_XCPT_FLAGS_T_EXT_INT);
659 return VMX_EXIT_INT_INFO_TYPE_EXT_INT;
660}
661
662
663/**
664 * Sets the Exit qualification VMCS field.
665 *
666 * @param pVCpu The cross context virtual CPU structure.
667 * @param u64ExitQual The Exit qualification.
668 */
669DECL_FORCE_INLINE(void) iemVmxVmcsSetExitQual(PVMCPUCC pVCpu, uint64_t u64ExitQual)
670{
671 PVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
672 pVmcs->u64RoExitQual.u = u64ExitQual;
673}
674
675
676/**
677 * Sets the VM-exit interruption information field.
678 *
679 * @param pVCpu The cross context virtual CPU structure.
680 * @param uExitIntInfo The VM-exit interruption information.
681 */
682DECL_FORCE_INLINE(void) iemVmxVmcsSetExitIntInfo(PVMCPUCC pVCpu, uint32_t uExitIntInfo)
683{
684 PVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
685 pVmcs->u32RoExitIntInfo = uExitIntInfo;
686}
687
688
689/**
690 * Sets the VM-exit interruption error code.
691 *
692 * @param pVCpu The cross context virtual CPU structure.
693 * @param uErrCode The error code.
694 */
695DECL_FORCE_INLINE(void) iemVmxVmcsSetExitIntErrCode(PVMCPUCC pVCpu, uint32_t uErrCode)
696{
697 PVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
698 pVmcs->u32RoExitIntErrCode = uErrCode;
699}
700
701
702/**
703 * Sets the IDT-vectoring information field.
704 *
705 * @param pVCpu The cross context virtual CPU structure.
706 * @param uIdtVectorInfo The IDT-vectoring information.
707 */
708DECL_FORCE_INLINE(void) iemVmxVmcsSetIdtVectoringInfo(PVMCPUCC pVCpu, uint32_t uIdtVectorInfo)
709{
710 PVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
711 pVmcs->u32RoIdtVectoringInfo = uIdtVectorInfo;
712}
713
714
715/**
716 * Sets the IDT-vectoring error code field.
717 *
718 * @param pVCpu The cross context virtual CPU structure.
719 * @param uErrCode The error code.
720 */
721DECL_FORCE_INLINE(void) iemVmxVmcsSetIdtVectoringErrCode(PVMCPUCC pVCpu, uint32_t uErrCode)
722{
723 PVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
724 pVmcs->u32RoIdtVectoringErrCode = uErrCode;
725}
726
727
728/**
729 * Sets the VM-exit guest-linear address VMCS field.
730 *
731 * @param pVCpu The cross context virtual CPU structure.
732 * @param uGuestLinearAddr The VM-exit guest-linear address.
733 */
734DECL_FORCE_INLINE(void) iemVmxVmcsSetExitGuestLinearAddr(PVMCPUCC pVCpu, uint64_t uGuestLinearAddr)
735{
736 PVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
737 pVmcs->u64RoGuestLinearAddr.u = uGuestLinearAddr;
738}
739
740
741/**
742 * Sets the VM-exit guest-physical address VMCS field.
743 *
744 * @param pVCpu The cross context virtual CPU structure.
745 * @param uGuestPhysAddr The VM-exit guest-physical address.
746 */
747DECL_FORCE_INLINE(void) iemVmxVmcsSetExitGuestPhysAddr(PVMCPUCC pVCpu, uint64_t uGuestPhysAddr)
748{
749 PVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
750 pVmcs->u64RoGuestPhysAddr.u = uGuestPhysAddr;
751}
752
753
754/**
755 * Sets the VM-exit instruction length VMCS field.
756 *
757 * @param pVCpu The cross context virtual CPU structure.
758 * @param cbInstr The VM-exit instruction length in bytes.
759 *
760 * @remarks Callers may clear this field to 0. Hence, this function does not check
761 * the validity of the instruction length.
762 */
763DECL_FORCE_INLINE(void) iemVmxVmcsSetExitInstrLen(PVMCPUCC pVCpu, uint32_t cbInstr)
764{
765 PVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
766 pVmcs->u32RoExitInstrLen = cbInstr;
767}
768
769
770/**
771 * Sets the VM-exit instruction info. VMCS field.
772 *
773 * @param pVCpu The cross context virtual CPU structure.
774 * @param uExitInstrInfo The VM-exit instruction information.
775 */
776DECL_FORCE_INLINE(void) iemVmxVmcsSetExitInstrInfo(PVMCPUCC pVCpu, uint32_t uExitInstrInfo)
777{
778 PVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
779 pVmcs->u32RoExitInstrInfo = uExitInstrInfo;
780}
781
782
783/**
784 * Sets the guest pending-debug exceptions field.
785 *
786 * @param pVCpu The cross context virtual CPU structure.
787 * @param uGuestPendingDbgXcpts The guest pending-debug exceptions.
788 */
789DECL_FORCE_INLINE(void) iemVmxVmcsSetGuestPendingDbgXcpts(PVMCPUCC pVCpu, uint64_t uGuestPendingDbgXcpts)
790{
791 PVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
792 Assert(!(uGuestPendingDbgXcpts & VMX_VMCS_GUEST_PENDING_DEBUG_VALID_MASK));
793 pVmcs->u64GuestPendingDbgXcpts.u = uGuestPendingDbgXcpts;
794}
795
796
797/**
798 * Implements VMSucceed for VMX instruction success.
799 *
800 * @param pVCpu The cross context virtual CPU structure.
801 */
802DECL_FORCE_INLINE(void) iemVmxVmSucceed(PVMCPUCC pVCpu)
803{
804 return CPUMSetGuestVmxVmSucceed(&pVCpu->cpum.GstCtx);
805}
806
807
808/**
809 * Implements VMFailInvalid for VMX instruction failure.
810 *
811 * @param pVCpu The cross context virtual CPU structure.
812 */
813DECL_FORCE_INLINE(void) iemVmxVmFailInvalid(PVMCPUCC pVCpu)
814{
815 return CPUMSetGuestVmxVmFailInvalid(&pVCpu->cpum.GstCtx);
816}
817
818
819/**
820 * Implements VMFail for VMX instruction failure.
821 *
822 * @param pVCpu The cross context virtual CPU structure.
823 * @param enmInsErr The VM instruction error.
824 */
825DECL_FORCE_INLINE(void) iemVmxVmFail(PVMCPUCC pVCpu, VMXINSTRERR enmInsErr)
826{
827 return CPUMSetGuestVmxVmFail(&pVCpu->cpum.GstCtx, enmInsErr);
828}
829
830
831/**
832 * Checks if the given auto-load/store MSR area count is valid for the
833 * implementation.
834 *
835 * @returns @c true if it's within the valid limit, @c false otherwise.
836 * @param pVCpu The cross context virtual CPU structure.
837 * @param uMsrCount The MSR area count to check.
838 */
839DECL_FORCE_INLINE(bool) iemVmxIsAutoMsrCountValid(PCVMCPU pVCpu, uint32_t uMsrCount)
840{
841 uint64_t const u64VmxMiscMsr = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Misc;
842 uint32_t const cMaxSupportedMsrs = VMX_MISC_MAX_MSRS(u64VmxMiscMsr);
843 Assert(cMaxSupportedMsrs <= VMX_V_AUTOMSR_AREA_SIZE / sizeof(VMXAUTOMSR));
844 if (uMsrCount <= cMaxSupportedMsrs)
845 return true;
846 return false;
847}
848
849
850/**
851 * Flushes the current VMCS contents back to guest memory.
852 *
853 * @returns VBox status code.
854 * @param pVCpu The cross context virtual CPU structure.
855 */
856DECL_FORCE_INLINE(int) iemVmxWriteCurrentVmcsToGstMem(PVMCPUCC pVCpu)
857{
858 Assert(pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs));
859 Assert(IEM_VMX_HAS_CURRENT_VMCS(pVCpu));
860 int rc = PGMPhysSimpleWriteGCPhys(pVCpu->CTX_SUFF(pVM), IEM_VMX_GET_CURRENT_VMCS(pVCpu),
861 pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs), sizeof(VMXVVMCS));
862 return rc;
863}
864
865
866/**
867 * Populates the current VMCS contents from guest memory.
868 *
869 * @returns VBox status code.
870 * @param pVCpu The cross context virtual CPU structure.
871 */
872DECL_FORCE_INLINE(int) iemVmxReadCurrentVmcsFromGstMem(PVMCPUCC pVCpu)
873{
874 Assert(pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs));
875 Assert(IEM_VMX_HAS_CURRENT_VMCS(pVCpu));
876 int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), (void *)pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs),
877 IEM_VMX_GET_CURRENT_VMCS(pVCpu), sizeof(VMXVVMCS));
878 return rc;
879}
880
881
882/**
883 * Implements VMSucceed for the VMREAD instruction and increments the guest RIP.
884 *
885 * @param pVCpu The cross context virtual CPU structure.
886 */
887DECL_FORCE_INLINE(void) iemVmxVmreadSuccess(PVMCPUCC pVCpu, uint8_t cbInstr)
888{
889 iemVmxVmSucceed(pVCpu);
890 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
891}
892
893
894/**
895 * Gets the instruction diagnostic for segment base checks during VM-entry of a
896 * nested-guest.
897 *
898 * @param iSegReg The segment index (X86_SREG_XXX).
899 */
900IEM_STATIC VMXVDIAG iemVmxGetDiagVmentrySegBase(unsigned iSegReg)
901{
902 switch (iSegReg)
903 {
904 case X86_SREG_CS: return kVmxVDiag_Vmentry_GuestSegBaseCs;
905 case X86_SREG_DS: return kVmxVDiag_Vmentry_GuestSegBaseDs;
906 case X86_SREG_ES: return kVmxVDiag_Vmentry_GuestSegBaseEs;
907 case X86_SREG_FS: return kVmxVDiag_Vmentry_GuestSegBaseFs;
908 case X86_SREG_GS: return kVmxVDiag_Vmentry_GuestSegBaseGs;
909 case X86_SREG_SS: return kVmxVDiag_Vmentry_GuestSegBaseSs;
910 IEM_NOT_REACHED_DEFAULT_CASE_RET2(kVmxVDiag_Ipe_1);
911 }
912}
913
914
915/**
916 * Gets the instruction diagnostic for segment base checks during VM-entry of a
917 * nested-guest that is in Virtual-8086 mode.
918 *
919 * @param iSegReg The segment index (X86_SREG_XXX).
920 */
921IEM_STATIC VMXVDIAG iemVmxGetDiagVmentrySegBaseV86(unsigned iSegReg)
922{
923 switch (iSegReg)
924 {
925 case X86_SREG_CS: return kVmxVDiag_Vmentry_GuestSegBaseV86Cs;
926 case X86_SREG_DS: return kVmxVDiag_Vmentry_GuestSegBaseV86Ds;
927 case X86_SREG_ES: return kVmxVDiag_Vmentry_GuestSegBaseV86Es;
928 case X86_SREG_FS: return kVmxVDiag_Vmentry_GuestSegBaseV86Fs;
929 case X86_SREG_GS: return kVmxVDiag_Vmentry_GuestSegBaseV86Gs;
930 case X86_SREG_SS: return kVmxVDiag_Vmentry_GuestSegBaseV86Ss;
931 IEM_NOT_REACHED_DEFAULT_CASE_RET2(kVmxVDiag_Ipe_2);
932 }
933}
934
935
936/**
937 * Gets the instruction diagnostic for segment limit checks during VM-entry of a
938 * nested-guest that is in Virtual-8086 mode.
939 *
940 * @param iSegReg The segment index (X86_SREG_XXX).
941 */
942IEM_STATIC VMXVDIAG iemVmxGetDiagVmentrySegLimitV86(unsigned iSegReg)
943{
944 switch (iSegReg)
945 {
946 case X86_SREG_CS: return kVmxVDiag_Vmentry_GuestSegLimitV86Cs;
947 case X86_SREG_DS: return kVmxVDiag_Vmentry_GuestSegLimitV86Ds;
948 case X86_SREG_ES: return kVmxVDiag_Vmentry_GuestSegLimitV86Es;
949 case X86_SREG_FS: return kVmxVDiag_Vmentry_GuestSegLimitV86Fs;
950 case X86_SREG_GS: return kVmxVDiag_Vmentry_GuestSegLimitV86Gs;
951 case X86_SREG_SS: return kVmxVDiag_Vmentry_GuestSegLimitV86Ss;
952 IEM_NOT_REACHED_DEFAULT_CASE_RET2(kVmxVDiag_Ipe_3);
953 }
954}
955
956
957/**
958 * Gets the instruction diagnostic for segment attribute checks during VM-entry of a
959 * nested-guest that is in Virtual-8086 mode.
960 *
961 * @param iSegReg The segment index (X86_SREG_XXX).
962 */
963IEM_STATIC VMXVDIAG iemVmxGetDiagVmentrySegAttrV86(unsigned iSegReg)
964{
965 switch (iSegReg)
966 {
967 case X86_SREG_CS: return kVmxVDiag_Vmentry_GuestSegAttrV86Cs;
968 case X86_SREG_DS: return kVmxVDiag_Vmentry_GuestSegAttrV86Ds;
969 case X86_SREG_ES: return kVmxVDiag_Vmentry_GuestSegAttrV86Es;
970 case X86_SREG_FS: return kVmxVDiag_Vmentry_GuestSegAttrV86Fs;
971 case X86_SREG_GS: return kVmxVDiag_Vmentry_GuestSegAttrV86Gs;
972 case X86_SREG_SS: return kVmxVDiag_Vmentry_GuestSegAttrV86Ss;
973 IEM_NOT_REACHED_DEFAULT_CASE_RET2(kVmxVDiag_Ipe_4);
974 }
975}
976
977
978/**
979 * Gets the instruction diagnostic for segment attributes reserved bits failure
980 * during VM-entry of a nested-guest.
981 *
982 * @param iSegReg The segment index (X86_SREG_XXX).
983 */
984IEM_STATIC VMXVDIAG iemVmxGetDiagVmentrySegAttrRsvd(unsigned iSegReg)
985{
986 switch (iSegReg)
987 {
988 case X86_SREG_CS: return kVmxVDiag_Vmentry_GuestSegAttrRsvdCs;
989 case X86_SREG_DS: return kVmxVDiag_Vmentry_GuestSegAttrRsvdDs;
990 case X86_SREG_ES: return kVmxVDiag_Vmentry_GuestSegAttrRsvdEs;
991 case X86_SREG_FS: return kVmxVDiag_Vmentry_GuestSegAttrRsvdFs;
992 case X86_SREG_GS: return kVmxVDiag_Vmentry_GuestSegAttrRsvdGs;
993 case X86_SREG_SS: return kVmxVDiag_Vmentry_GuestSegAttrRsvdSs;
994 IEM_NOT_REACHED_DEFAULT_CASE_RET2(kVmxVDiag_Ipe_5);
995 }
996}
997
998
999/**
1000 * Gets the instruction diagnostic for segment attributes descriptor-type
1001 * (code/segment or system) failure during VM-entry of a nested-guest.
1002 *
1003 * @param iSegReg The segment index (X86_SREG_XXX).
1004 */
1005IEM_STATIC VMXVDIAG iemVmxGetDiagVmentrySegAttrDescType(unsigned iSegReg)
1006{
1007 switch (iSegReg)
1008 {
1009 case X86_SREG_CS: return kVmxVDiag_Vmentry_GuestSegAttrDescTypeCs;
1010 case X86_SREG_DS: return kVmxVDiag_Vmentry_GuestSegAttrDescTypeDs;
1011 case X86_SREG_ES: return kVmxVDiag_Vmentry_GuestSegAttrDescTypeEs;
1012 case X86_SREG_FS: return kVmxVDiag_Vmentry_GuestSegAttrDescTypeFs;
1013 case X86_SREG_GS: return kVmxVDiag_Vmentry_GuestSegAttrDescTypeGs;
1014 case X86_SREG_SS: return kVmxVDiag_Vmentry_GuestSegAttrDescTypeSs;
1015 IEM_NOT_REACHED_DEFAULT_CASE_RET2(kVmxVDiag_Ipe_6);
1016 }
1017}
1018
1019
1020/**
1021 * Gets the instruction diagnostic for segment attributes descriptor-type
1022 * (code/segment or system) failure during VM-entry of a nested-guest.
1023 *
1024 * @param iSegReg The segment index (X86_SREG_XXX).
1025 */
1026IEM_STATIC VMXVDIAG iemVmxGetDiagVmentrySegAttrPresent(unsigned iSegReg)
1027{
1028 switch (iSegReg)
1029 {
1030 case X86_SREG_CS: return kVmxVDiag_Vmentry_GuestSegAttrPresentCs;
1031 case X86_SREG_DS: return kVmxVDiag_Vmentry_GuestSegAttrPresentDs;
1032 case X86_SREG_ES: return kVmxVDiag_Vmentry_GuestSegAttrPresentEs;
1033 case X86_SREG_FS: return kVmxVDiag_Vmentry_GuestSegAttrPresentFs;
1034 case X86_SREG_GS: return kVmxVDiag_Vmentry_GuestSegAttrPresentGs;
1035 case X86_SREG_SS: return kVmxVDiag_Vmentry_GuestSegAttrPresentSs;
1036 IEM_NOT_REACHED_DEFAULT_CASE_RET2(kVmxVDiag_Ipe_7);
1037 }
1038}
1039
1040
1041/**
1042 * Gets the instruction diagnostic for segment attribute granularity failure during
1043 * VM-entry of a nested-guest.
1044 *
1045 * @param iSegReg The segment index (X86_SREG_XXX).
1046 */
1047IEM_STATIC VMXVDIAG iemVmxGetDiagVmentrySegAttrGran(unsigned iSegReg)
1048{
1049 switch (iSegReg)
1050 {
1051 case X86_SREG_CS: return kVmxVDiag_Vmentry_GuestSegAttrGranCs;
1052 case X86_SREG_DS: return kVmxVDiag_Vmentry_GuestSegAttrGranDs;
1053 case X86_SREG_ES: return kVmxVDiag_Vmentry_GuestSegAttrGranEs;
1054 case X86_SREG_FS: return kVmxVDiag_Vmentry_GuestSegAttrGranFs;
1055 case X86_SREG_GS: return kVmxVDiag_Vmentry_GuestSegAttrGranGs;
1056 case X86_SREG_SS: return kVmxVDiag_Vmentry_GuestSegAttrGranSs;
1057 IEM_NOT_REACHED_DEFAULT_CASE_RET2(kVmxVDiag_Ipe_8);
1058 }
1059}
1060
1061/**
1062 * Gets the instruction diagnostic for segment attribute DPL/RPL failure during
1063 * VM-entry of a nested-guest.
1064 *
1065 * @param iSegReg The segment index (X86_SREG_XXX).
1066 */
1067IEM_STATIC VMXVDIAG iemVmxGetDiagVmentrySegAttrDplRpl(unsigned iSegReg)
1068{
1069 switch (iSegReg)
1070 {
1071 case X86_SREG_CS: return kVmxVDiag_Vmentry_GuestSegAttrDplRplCs;
1072 case X86_SREG_DS: return kVmxVDiag_Vmentry_GuestSegAttrDplRplDs;
1073 case X86_SREG_ES: return kVmxVDiag_Vmentry_GuestSegAttrDplRplEs;
1074 case X86_SREG_FS: return kVmxVDiag_Vmentry_GuestSegAttrDplRplFs;
1075 case X86_SREG_GS: return kVmxVDiag_Vmentry_GuestSegAttrDplRplGs;
1076 case X86_SREG_SS: return kVmxVDiag_Vmentry_GuestSegAttrDplRplSs;
1077 IEM_NOT_REACHED_DEFAULT_CASE_RET2(kVmxVDiag_Ipe_9);
1078 }
1079}
1080
1081
1082/**
1083 * Gets the instruction diagnostic for segment attribute type accessed failure
1084 * during VM-entry of a nested-guest.
1085 *
1086 * @param iSegReg The segment index (X86_SREG_XXX).
1087 */
1088IEM_STATIC VMXVDIAG iemVmxGetDiagVmentrySegAttrTypeAcc(unsigned iSegReg)
1089{
1090 switch (iSegReg)
1091 {
1092 case X86_SREG_CS: return kVmxVDiag_Vmentry_GuestSegAttrTypeAccCs;
1093 case X86_SREG_DS: return kVmxVDiag_Vmentry_GuestSegAttrTypeAccDs;
1094 case X86_SREG_ES: return kVmxVDiag_Vmentry_GuestSegAttrTypeAccEs;
1095 case X86_SREG_FS: return kVmxVDiag_Vmentry_GuestSegAttrTypeAccFs;
1096 case X86_SREG_GS: return kVmxVDiag_Vmentry_GuestSegAttrTypeAccGs;
1097 case X86_SREG_SS: return kVmxVDiag_Vmentry_GuestSegAttrTypeAccSs;
1098 IEM_NOT_REACHED_DEFAULT_CASE_RET2(kVmxVDiag_Ipe_10);
1099 }
1100}
1101
1102
1103/**
1104 * Gets the instruction diagnostic for guest CR3 referenced PDPTE reserved bits
1105 * failure during VM-entry of a nested-guest.
1106 *
1107 * @param iSegReg The PDPTE entry index.
1108 */
1109IEM_STATIC VMXVDIAG iemVmxGetDiagVmentryPdpteRsvd(unsigned iPdpte)
1110{
1111 Assert(iPdpte < X86_PG_PAE_PDPE_ENTRIES);
1112 switch (iPdpte)
1113 {
1114 case 0: return kVmxVDiag_Vmentry_GuestPdpte0Rsvd;
1115 case 1: return kVmxVDiag_Vmentry_GuestPdpte1Rsvd;
1116 case 2: return kVmxVDiag_Vmentry_GuestPdpte2Rsvd;
1117 case 3: return kVmxVDiag_Vmentry_GuestPdpte3Rsvd;
1118 IEM_NOT_REACHED_DEFAULT_CASE_RET2(kVmxVDiag_Ipe_11);
1119 }
1120}
1121
1122
1123/**
1124 * Gets the instruction diagnostic for host CR3 referenced PDPTE reserved bits
1125 * failure during VM-exit of a nested-guest.
1126 *
1127 * @param iSegReg The PDPTE entry index.
1128 */
1129IEM_STATIC VMXVDIAG iemVmxGetDiagVmexitPdpteRsvd(unsigned iPdpte)
1130{
1131 Assert(iPdpte < X86_PG_PAE_PDPE_ENTRIES);
1132 switch (iPdpte)
1133 {
1134 case 0: return kVmxVDiag_Vmexit_HostPdpte0Rsvd;
1135 case 1: return kVmxVDiag_Vmexit_HostPdpte1Rsvd;
1136 case 2: return kVmxVDiag_Vmexit_HostPdpte2Rsvd;
1137 case 3: return kVmxVDiag_Vmexit_HostPdpte3Rsvd;
1138 IEM_NOT_REACHED_DEFAULT_CASE_RET2(kVmxVDiag_Ipe_12);
1139 }
1140}
1141
1142
1143/**
1144 * Saves the guest control registers, debug registers and some MSRs are part of
1145 * VM-exit.
1146 *
1147 * @param pVCpu The cross context virtual CPU structure.
1148 */
1149IEM_STATIC void iemVmxVmexitSaveGuestControlRegsMsrs(PVMCPUCC pVCpu)
1150{
1151 /*
1152 * Saves the guest control registers, debug registers and some MSRs.
1153 * See Intel spec. 27.3.1 "Saving Control Registers, Debug Registers and MSRs".
1154 */
1155 PVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
1156
1157 /* Save control registers. */
1158 pVmcs->u64GuestCr0.u = pVCpu->cpum.GstCtx.cr0;
1159 pVmcs->u64GuestCr3.u = pVCpu->cpum.GstCtx.cr3;
1160 pVmcs->u64GuestCr4.u = pVCpu->cpum.GstCtx.cr4;
1161
1162 /* Save SYSENTER CS, ESP, EIP. */
1163 pVmcs->u32GuestSysenterCS = pVCpu->cpum.GstCtx.SysEnter.cs;
1164 if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode)
1165 {
1166 pVmcs->u64GuestSysenterEsp.u = pVCpu->cpum.GstCtx.SysEnter.esp;
1167 pVmcs->u64GuestSysenterEip.u = pVCpu->cpum.GstCtx.SysEnter.eip;
1168 }
1169 else
1170 {
1171 pVmcs->u64GuestSysenterEsp.s.Lo = pVCpu->cpum.GstCtx.SysEnter.esp;
1172 pVmcs->u64GuestSysenterEip.s.Lo = pVCpu->cpum.GstCtx.SysEnter.eip;
1173 }
1174
1175 /* Save debug registers (DR7 and IA32_DEBUGCTL MSR). */
1176 if (pVmcs->u32ExitCtls & VMX_EXIT_CTLS_SAVE_DEBUG)
1177 {
1178 pVmcs->u64GuestDr7.u = pVCpu->cpum.GstCtx.dr[7];
1179 /** @todo NSTVMX: Support IA32_DEBUGCTL MSR */
1180 }
1181
1182 /* Save PAT MSR. */
1183 if (pVmcs->u32ExitCtls & VMX_EXIT_CTLS_SAVE_PAT_MSR)
1184 pVmcs->u64GuestPatMsr.u = pVCpu->cpum.GstCtx.msrPAT;
1185
1186 /* Save EFER MSR. */
1187 if (pVmcs->u32ExitCtls & VMX_EXIT_CTLS_SAVE_EFER_MSR)
1188 pVmcs->u64GuestEferMsr.u = pVCpu->cpum.GstCtx.msrEFER;
1189
1190 /* We don't support clearing IA32_BNDCFGS MSR yet. */
1191 Assert(!(pVmcs->u32ExitCtls & VMX_EXIT_CTLS_CLEAR_BNDCFGS_MSR));
1192
1193 /* Nothing to do for SMBASE register - We don't support SMM yet. */
1194}
1195
1196
1197/**
1198 * Saves the guest force-flags in preparation of entering the nested-guest.
1199 *
1200 * @param pVCpu The cross context virtual CPU structure.
1201 */
1202IEM_STATIC void iemVmxVmentrySaveNmiBlockingFF(PVMCPUCC pVCpu)
1203{
1204 /* We shouldn't be called multiple times during VM-entry. */
1205 Assert(pVCpu->cpum.GstCtx.hwvirt.fLocalForcedActions == 0);
1206
1207 /* MTF should not be set outside VMX non-root mode. */
1208 Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_VMX_MTF));
1209
1210 /*
1211 * Preserve the required force-flags.
1212 *
1213 * We cache and clear force-flags that would affect the execution of the
1214 * nested-guest. Cached flags are then restored while returning to the guest
1215 * if necessary.
1216 *
1217 * - VMCPU_FF_INHIBIT_INTERRUPTS need not be cached as it only affects
1218 * interrupts until the completion of the current VMLAUNCH/VMRESUME
1219 * instruction. Interrupt inhibition for any nested-guest instruction
1220 * is supplied by the guest-interruptibility state VMCS field and will
1221 * be set up as part of loading the guest state.
1222 *
1223 * - VMCPU_FF_BLOCK_NMIS needs to be cached as VM-exits caused before
1224 * successful VM-entry (due to invalid guest-state) need to continue
1225 * blocking NMIs if it was in effect before VM-entry.
1226 *
1227 * - MTF need not be preserved as it's used only in VMX non-root mode and
1228 * is supplied through the VM-execution controls.
1229 *
1230 * The remaining FFs (e.g. timers, APIC updates) can stay in place so that
1231 * we will be able to generate interrupts that may cause VM-exits for
1232 * the nested-guest.
1233 */
1234 pVCpu->cpum.GstCtx.hwvirt.fLocalForcedActions = pVCpu->fLocalForcedActions & VMCPU_FF_BLOCK_NMIS;
1235}
1236
1237
1238/**
1239 * Restores the guest force-flags in preparation of exiting the nested-guest.
1240 *
1241 * @param pVCpu The cross context virtual CPU structure.
1242 */
1243IEM_STATIC void iemVmxVmexitRestoreNmiBlockingFF(PVMCPUCC pVCpu)
1244{
1245 if (pVCpu->cpum.GstCtx.hwvirt.fLocalForcedActions)
1246 {
1247 VMCPU_FF_SET_MASK(pVCpu, pVCpu->cpum.GstCtx.hwvirt.fLocalForcedActions);
1248 pVCpu->cpum.GstCtx.hwvirt.fLocalForcedActions = 0;
1249 }
1250}
1251
1252
1253/**
1254 * Perform a VMX transition updated PGM, IEM and CPUM.
1255 *
1256 * @param pVCpu The cross context virtual CPU structure.
1257 */
1258IEM_STATIC int iemVmxWorldSwitch(PVMCPUCC pVCpu)
1259{
1260 /*
1261 * Inform PGM about paging mode changes.
1262 * We include X86_CR0_PE because PGM doesn't handle paged-real mode yet,
1263 * see comment in iemMemPageTranslateAndCheckAccess().
1264 */
1265 int rc = PGMChangeMode(pVCpu, pVCpu->cpum.GstCtx.cr0 | X86_CR0_PE, pVCpu->cpum.GstCtx.cr4, pVCpu->cpum.GstCtx.msrEFER);
1266# ifdef IN_RING3
1267 Assert(rc != VINF_PGM_CHANGE_MODE);
1268# endif
1269 AssertRCReturn(rc, rc);
1270
1271 /* Inform CPUM (recompiler), can later be removed. */
1272 CPUMSetChangedFlags(pVCpu, CPUM_CHANGED_ALL);
1273
1274 /*
1275 * Flush the TLB with new CR3. This is required in case the PGM mode change
1276 * above doesn't actually change anything.
1277 */
1278 if (rc == VINF_SUCCESS)
1279 {
1280 rc = PGMFlushTLB(pVCpu, pVCpu->cpum.GstCtx.cr3, true);
1281 AssertRCReturn(rc, rc);
1282 }
1283
1284 /* Re-initialize IEM cache/state after the drastic mode switch. */
1285 iemReInitExec(pVCpu);
1286 return rc;
1287}
1288
1289
1290/**
1291 * Calculates the current VMX-preemption timer value.
1292 *
1293 * @returns The current VMX-preemption timer value.
1294 * @param pVCpu The cross context virtual CPU structure.
1295 */
1296IEM_STATIC uint32_t iemVmxCalcPreemptTimer(PVMCPUCC pVCpu)
1297{
1298 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
1299 Assert(pVmcs);
1300
1301 /*
1302 * Assume the following:
1303 * PreemptTimerShift = 5
1304 * VmcsPreemptTimer = 2 (i.e. need to decrement by 1 every 2 * RT_BIT(5) = 20000 TSC ticks)
1305 * EntryTick = 50000 (TSC at time of VM-entry)
1306 *
1307 * CurTick Delta PreemptTimerVal
1308 * ----------------------------------
1309 * 60000 10000 2
1310 * 80000 30000 1
1311 * 90000 40000 0 -> VM-exit.
1312 *
1313 * If Delta >= VmcsPreemptTimer * RT_BIT(PreemptTimerShift) cause a VMX-preemption timer VM-exit.
1314 * The saved VMX-preemption timer value is calculated as follows:
1315 * PreemptTimerVal = VmcsPreemptTimer - (Delta / (VmcsPreemptTimer * RT_BIT(PreemptTimerShift)))
1316 * E.g.:
1317 * Delta = 10000
1318 * Tmp = 10000 / (2 * 10000) = 0.5
1319 * NewPt = 2 - 0.5 = 2
1320 * Delta = 30000
1321 * Tmp = 30000 / (2 * 10000) = 1.5
1322 * NewPt = 2 - 1.5 = 1
1323 * Delta = 40000
1324 * Tmp = 40000 / 20000 = 2
1325 * NewPt = 2 - 2 = 0
1326 */
1327 IEM_CTX_ASSERT(pVCpu, CPUMCTX_EXTRN_HWVIRT);
1328 uint32_t const uVmcsPreemptVal = pVmcs->u32PreemptTimer;
1329 if (uVmcsPreemptVal > 0)
1330 {
1331 uint64_t const uCurTick = TMCpuTickGetNoCheck(pVCpu);
1332 uint64_t const uEntryTick = pVCpu->cpum.GstCtx.hwvirt.vmx.uEntryTick;
1333 uint64_t const uDelta = uCurTick - uEntryTick;
1334 uint32_t const uPreemptTimer = uVmcsPreemptVal
1335 - ASMDivU64ByU32RetU32(uDelta, uVmcsPreemptVal * RT_BIT(VMX_V_PREEMPT_TIMER_SHIFT));
1336 return uPreemptTimer;
1337 }
1338 return 0;
1339}
1340
1341
1342/**
1343 * Saves guest segment registers, GDTR, IDTR, LDTR, TR as part of VM-exit.
1344 *
1345 * @param pVCpu The cross context virtual CPU structure.
1346 */
1347IEM_STATIC void iemVmxVmexitSaveGuestSegRegs(PVMCPUCC pVCpu)
1348{
1349 /*
1350 * Save guest segment registers, GDTR, IDTR, LDTR, TR.
1351 * See Intel spec 27.3.2 "Saving Segment Registers and Descriptor-Table Registers".
1352 */
1353 /* CS, SS, ES, DS, FS, GS. */
1354 PVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
1355 for (unsigned iSegReg = 0; iSegReg < X86_SREG_COUNT; iSegReg++)
1356 {
1357 PCCPUMSELREG pSelReg = &pVCpu->cpum.GstCtx.aSRegs[iSegReg];
1358 if (!pSelReg->Attr.n.u1Unusable)
1359 iemVmxVmcsSetGuestSegReg(pVmcs, iSegReg, pSelReg);
1360 else
1361 {
1362 /*
1363 * For unusable segments the attributes are undefined except for CS and SS.
1364 * For the rest we don't bother preserving anything but the unusable bit.
1365 */
1366 switch (iSegReg)
1367 {
1368 case X86_SREG_CS:
1369 pVmcs->GuestCs = pSelReg->Sel;
1370 pVmcs->u64GuestCsBase.u = pSelReg->u64Base;
1371 pVmcs->u32GuestCsLimit = pSelReg->u32Limit;
1372 pVmcs->u32GuestCsAttr = pSelReg->Attr.u & ( X86DESCATTR_L | X86DESCATTR_D | X86DESCATTR_G
1373 | X86DESCATTR_UNUSABLE);
1374 break;
1375
1376 case X86_SREG_SS:
1377 pVmcs->GuestSs = pSelReg->Sel;
1378 if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode)
1379 pVmcs->u64GuestSsBase.u &= UINT32_C(0xffffffff);
1380 pVmcs->u32GuestSsAttr = pSelReg->Attr.u & (X86DESCATTR_DPL | X86DESCATTR_UNUSABLE);
1381 break;
1382
1383 case X86_SREG_DS:
1384 pVmcs->GuestDs = pSelReg->Sel;
1385 if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode)
1386 pVmcs->u64GuestDsBase.u &= UINT32_C(0xffffffff);
1387 pVmcs->u32GuestDsAttr = X86DESCATTR_UNUSABLE;
1388 break;
1389
1390 case X86_SREG_ES:
1391 pVmcs->GuestEs = pSelReg->Sel;
1392 if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode)
1393 pVmcs->u64GuestEsBase.u &= UINT32_C(0xffffffff);
1394 pVmcs->u32GuestEsAttr = X86DESCATTR_UNUSABLE;
1395 break;
1396
1397 case X86_SREG_FS:
1398 pVmcs->GuestFs = pSelReg->Sel;
1399 pVmcs->u64GuestFsBase.u = pSelReg->u64Base;
1400 pVmcs->u32GuestFsAttr = X86DESCATTR_UNUSABLE;
1401 break;
1402
1403 case X86_SREG_GS:
1404 pVmcs->GuestGs = pSelReg->Sel;
1405 pVmcs->u64GuestGsBase.u = pSelReg->u64Base;
1406 pVmcs->u32GuestGsAttr = X86DESCATTR_UNUSABLE;
1407 break;
1408 }
1409 }
1410 }
1411
1412 /* Segment attribute bits 31:17 and 11:8 MBZ. */
1413 uint32_t const fValidAttrMask = X86DESCATTR_TYPE | X86DESCATTR_DT | X86DESCATTR_DPL | X86DESCATTR_P
1414 | X86DESCATTR_AVL | X86DESCATTR_L | X86DESCATTR_D | X86DESCATTR_G
1415 | X86DESCATTR_UNUSABLE;
1416 /* LDTR. */
1417 {
1418 PCCPUMSELREG pSelReg = &pVCpu->cpum.GstCtx.ldtr;
1419 pVmcs->GuestLdtr = pSelReg->Sel;
1420 pVmcs->u64GuestLdtrBase.u = pSelReg->u64Base;
1421 Assert(X86_IS_CANONICAL(pSelReg->u64Base));
1422 pVmcs->u32GuestLdtrLimit = pSelReg->u32Limit;
1423 pVmcs->u32GuestLdtrAttr = pSelReg->Attr.u & fValidAttrMask;
1424 }
1425
1426 /* TR. */
1427 {
1428 PCCPUMSELREG pSelReg = &pVCpu->cpum.GstCtx.tr;
1429 pVmcs->GuestTr = pSelReg->Sel;
1430 pVmcs->u64GuestTrBase.u = pSelReg->u64Base;
1431 pVmcs->u32GuestTrLimit = pSelReg->u32Limit;
1432 pVmcs->u32GuestTrAttr = pSelReg->Attr.u & fValidAttrMask;
1433 }
1434
1435 /* GDTR. */
1436 pVmcs->u64GuestGdtrBase.u = pVCpu->cpum.GstCtx.gdtr.pGdt;
1437 pVmcs->u32GuestGdtrLimit = pVCpu->cpum.GstCtx.gdtr.cbGdt;
1438
1439 /* IDTR. */
1440 pVmcs->u64GuestIdtrBase.u = pVCpu->cpum.GstCtx.idtr.pIdt;
1441 pVmcs->u32GuestIdtrLimit = pVCpu->cpum.GstCtx.idtr.cbIdt;
1442}
1443
1444
1445/**
1446 * Saves guest non-register state as part of VM-exit.
1447 *
1448 * @param pVCpu The cross context virtual CPU structure.
1449 * @param uExitReason The VM-exit reason.
1450 */
1451IEM_STATIC void iemVmxVmexitSaveGuestNonRegState(PVMCPUCC pVCpu, uint32_t uExitReason)
1452{
1453 /*
1454 * Save guest non-register state.
1455 * See Intel spec. 27.3.4 "Saving Non-Register State".
1456 */
1457 PVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
1458
1459 /*
1460 * Activity state.
1461 * Most VM-exits will occur in the active state. However, if the first instruction
1462 * following the VM-entry is a HLT instruction, and the MTF VM-execution control is set,
1463 * the VM-exit will be from the HLT activity state.
1464 *
1465 * See Intel spec. 25.5.2 "Monitor Trap Flag".
1466 */
1467 /** @todo NSTVMX: Does triple-fault VM-exit reflect a shutdown activity state or
1468 * not? */
1469 EMSTATE const enmActivityState = EMGetState(pVCpu);
1470 switch (enmActivityState)
1471 {
1472 case EMSTATE_HALTED: pVmcs->u32GuestActivityState = VMX_VMCS_GUEST_ACTIVITY_HLT; break;
1473 default: pVmcs->u32GuestActivityState = VMX_VMCS_GUEST_ACTIVITY_ACTIVE; break;
1474 }
1475
1476 /*
1477 * Interruptibility-state.
1478 */
1479 /* NMI. */
1480 pVmcs->u32GuestIntrState = 0;
1481 if (pVmcs->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI)
1482 {
1483 if (pVCpu->cpum.GstCtx.hwvirt.vmx.fVirtNmiBlocking)
1484 pVmcs->u32GuestIntrState |= VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI;
1485 }
1486 else
1487 {
1488 if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_BLOCK_NMIS))
1489 pVmcs->u32GuestIntrState |= VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI;
1490 }
1491
1492 /* Blocking-by-STI. */
1493 if ( VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS)
1494 && pVCpu->cpum.GstCtx.rip == EMGetInhibitInterruptsPC(pVCpu))
1495 {
1496 /** @todo NSTVMX: We can't distinguish between blocking-by-MovSS and blocking-by-STI
1497 * currently. */
1498 pVmcs->u32GuestIntrState |= VMX_VMCS_GUEST_INT_STATE_BLOCK_STI;
1499 }
1500 /* Nothing to do for SMI/enclave. We don't support enclaves or SMM yet. */
1501
1502 /*
1503 * Pending debug exceptions.
1504 *
1505 * For VM-exits where it is not applicable, we can safely zero out the field.
1506 * For VM-exits where it is applicable, it's expected to be updated by the caller already.
1507 */
1508 if ( uExitReason != VMX_EXIT_INIT_SIGNAL
1509 && uExitReason != VMX_EXIT_SMI
1510 && uExitReason != VMX_EXIT_ERR_MACHINE_CHECK
1511 && !VMXIsVmexitTrapLike(uExitReason))
1512 {
1513 /** @todo NSTVMX: also must exclude VM-exits caused by debug exceptions when
1514 * block-by-MovSS is in effect. */
1515 pVmcs->u64GuestPendingDbgXcpts.u = 0;
1516 }
1517
1518 /*
1519 * Save the VMX-preemption timer value back into the VMCS if the feature is enabled.
1520 *
1521 * For VMX-preemption timer VM-exits, we should have already written back 0 if the
1522 * feature is supported back into the VMCS, and thus there is nothing further to do here.
1523 */
1524 if ( uExitReason != VMX_EXIT_PREEMPT_TIMER
1525 && (pVmcs->u32ExitCtls & VMX_EXIT_CTLS_SAVE_PREEMPT_TIMER))
1526 pVmcs->u32PreemptTimer = iemVmxCalcPreemptTimer(pVCpu);
1527
1528 /* PDPTEs. */
1529 /* We don't support EPT yet. */
1530 Assert(!(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_EPT));
1531 pVmcs->u64GuestPdpte0.u = 0;
1532 pVmcs->u64GuestPdpte1.u = 0;
1533 pVmcs->u64GuestPdpte2.u = 0;
1534 pVmcs->u64GuestPdpte3.u = 0;
1535}
1536
1537
1538/**
1539 * Saves the guest-state as part of VM-exit.
1540 *
1541 * @returns VBox status code.
1542 * @param pVCpu The cross context virtual CPU structure.
1543 * @param uExitReason The VM-exit reason.
1544 */
1545IEM_STATIC void iemVmxVmexitSaveGuestState(PVMCPUCC pVCpu, uint32_t uExitReason)
1546{
1547 PVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
1548 Assert(pVmcs);
1549
1550 iemVmxVmexitSaveGuestControlRegsMsrs(pVCpu);
1551 iemVmxVmexitSaveGuestSegRegs(pVCpu);
1552
1553 pVmcs->u64GuestRip.u = pVCpu->cpum.GstCtx.rip;
1554 pVmcs->u64GuestRsp.u = pVCpu->cpum.GstCtx.rsp;
1555 pVmcs->u64GuestRFlags.u = pVCpu->cpum.GstCtx.rflags.u; /** @todo NSTVMX: Check RFLAGS.RF handling. */
1556
1557 iemVmxVmexitSaveGuestNonRegState(pVCpu, uExitReason);
1558}
1559
1560
1561/**
1562 * Saves the guest MSRs into the VM-exit MSR-store area as part of VM-exit.
1563 *
1564 * @returns VBox status code.
1565 * @param pVCpu The cross context virtual CPU structure.
1566 * @param uExitReason The VM-exit reason (for diagnostic purposes).
1567 */
1568IEM_STATIC int iemVmxVmexitSaveGuestAutoMsrs(PVMCPUCC pVCpu, uint32_t uExitReason)
1569{
1570 /*
1571 * Save guest MSRs.
1572 * See Intel spec. 27.4 "Saving MSRs".
1573 */
1574 PVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
1575 const char *const pszFailure = "VMX-abort";
1576
1577 /*
1578 * The VM-exit MSR-store area address need not be a valid guest-physical address if the
1579 * VM-exit MSR-store count is 0. If this is the case, bail early without reading it.
1580 * See Intel spec. 24.7.2 "VM-Exit Controls for MSRs".
1581 */
1582 uint32_t const cMsrs = pVmcs->u32ExitMsrStoreCount;
1583 if (!cMsrs)
1584 return VINF_SUCCESS;
1585
1586 /*
1587 * Verify the MSR auto-store count. Physical CPUs can behave unpredictably if the count
1588 * is exceeded including possibly raising #MC exceptions during VMX transition. Our
1589 * implementation causes a VMX-abort followed by a triple-fault.
1590 */
1591 bool const fIsMsrCountValid = iemVmxIsAutoMsrCountValid(pVCpu, cMsrs);
1592 if (fIsMsrCountValid)
1593 { /* likely */ }
1594 else
1595 IEM_VMX_VMEXIT_FAILED_RET(pVCpu, uExitReason, pszFailure, kVmxVDiag_Vmexit_MsrStoreCount);
1596
1597 /*
1598 * Optimization if the nested hypervisor is using the same guest-physical page for both
1599 * the VM-entry MSR-load area as well as the VM-exit MSR store area.
1600 */
1601 PVMXAUTOMSR pMsrArea;
1602 RTGCPHYS const GCPhysVmEntryMsrLoadArea = pVmcs->u64AddrEntryMsrLoad.u;
1603 RTGCPHYS const GCPhysVmExitMsrStoreArea = pVmcs->u64AddrExitMsrStore.u;
1604 if (GCPhysVmEntryMsrLoadArea == GCPhysVmExitMsrStoreArea)
1605 pMsrArea = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pEntryMsrLoadArea);
1606 else
1607 {
1608 int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), (void *)pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pExitMsrStoreArea),
1609 GCPhysVmExitMsrStoreArea, cMsrs * sizeof(VMXAUTOMSR));
1610 if (RT_SUCCESS(rc))
1611 pMsrArea = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pExitMsrStoreArea);
1612 else
1613 {
1614 AssertMsgFailed(("VM-exit: Failed to read MSR auto-store area at %#RGp, rc=%Rrc\n", GCPhysVmExitMsrStoreArea, rc));
1615 IEM_VMX_VMEXIT_FAILED_RET(pVCpu, uExitReason, pszFailure, kVmxVDiag_Vmexit_MsrStorePtrReadPhys);
1616 }
1617 }
1618
1619 /*
1620 * Update VM-exit MSR store area.
1621 */
1622 PVMXAUTOMSR pMsr = pMsrArea;
1623 Assert(pMsr);
1624 for (uint32_t idxMsr = 0; idxMsr < cMsrs; idxMsr++, pMsr++)
1625 {
1626 if ( !pMsr->u32Reserved
1627 && pMsr->u32Msr != MSR_IA32_SMBASE
1628 && pMsr->u32Msr >> 8 != MSR_IA32_X2APIC_START >> 8)
1629 {
1630 VBOXSTRICTRC rcStrict = CPUMQueryGuestMsr(pVCpu, pMsr->u32Msr, &pMsr->u64Value);
1631 if (rcStrict == VINF_SUCCESS)
1632 continue;
1633
1634 /*
1635 * If we're in ring-0, we cannot handle returns to ring-3 at this point and continue VM-exit.
1636 * If any nested hypervisor loads MSRs that require ring-3 handling, we cause a VMX-abort
1637 * recording the MSR index in the auxiliary info. field and indicated further by our
1638 * own, specific diagnostic code. Later, we can try implement handling of the MSR in ring-0
1639 * if possible, or come up with a better, generic solution.
1640 */
1641 pVCpu->cpum.GstCtx.hwvirt.vmx.uAbortAux = pMsr->u32Msr;
1642 VMXVDIAG const enmDiag = rcStrict == VINF_CPUM_R3_MSR_READ
1643 ? kVmxVDiag_Vmexit_MsrStoreRing3
1644 : kVmxVDiag_Vmexit_MsrStore;
1645 IEM_VMX_VMEXIT_FAILED_RET(pVCpu, uExitReason, pszFailure, enmDiag);
1646 }
1647 else
1648 {
1649 pVCpu->cpum.GstCtx.hwvirt.vmx.uAbortAux = pMsr->u32Msr;
1650 IEM_VMX_VMEXIT_FAILED_RET(pVCpu, uExitReason, pszFailure, kVmxVDiag_Vmexit_MsrStoreRsvd);
1651 }
1652 }
1653
1654 /*
1655 * Commit the VM-exit MSR store are to guest memory.
1656 */
1657 int rc = PGMPhysSimpleWriteGCPhys(pVCpu->CTX_SUFF(pVM), GCPhysVmExitMsrStoreArea, pMsrArea, cMsrs * sizeof(VMXAUTOMSR));
1658 if (RT_SUCCESS(rc))
1659 return VINF_SUCCESS;
1660
1661 NOREF(uExitReason);
1662 NOREF(pszFailure);
1663
1664 AssertMsgFailed(("VM-exit: Failed to write MSR auto-store area at %#RGp, rc=%Rrc\n", GCPhysVmExitMsrStoreArea, rc));
1665 IEM_VMX_VMEXIT_FAILED_RET(pVCpu, uExitReason, pszFailure, kVmxVDiag_Vmexit_MsrStorePtrWritePhys);
1666}
1667
1668
1669/**
1670 * Performs a VMX abort (due to an fatal error during VM-exit).
1671 *
1672 * @returns Strict VBox status code.
1673 * @param pVCpu The cross context virtual CPU structure.
1674 * @param enmAbort The VMX abort reason.
1675 */
1676IEM_STATIC VBOXSTRICTRC iemVmxAbort(PVMCPUCC pVCpu, VMXABORT enmAbort)
1677{
1678 /*
1679 * Perform the VMX abort.
1680 * See Intel spec. 27.7 "VMX Aborts".
1681 */
1682 LogFunc(("enmAbort=%u (%s) -> RESET\n", enmAbort, VMXGetAbortDesc(enmAbort)));
1683
1684 /* We don't support SMX yet. */
1685 pVCpu->cpum.GstCtx.hwvirt.vmx.enmAbort = enmAbort;
1686 if (IEM_VMX_HAS_CURRENT_VMCS(pVCpu))
1687 {
1688 RTGCPHYS const GCPhysVmcs = IEM_VMX_GET_CURRENT_VMCS(pVCpu);
1689 uint32_t const offVmxAbort = RT_UOFFSETOF(VMXVVMCS, enmVmxAbort);
1690 PGMPhysSimpleWriteGCPhys(pVCpu->CTX_SUFF(pVM), GCPhysVmcs + offVmxAbort, &enmAbort, sizeof(enmAbort));
1691 }
1692
1693 return VINF_EM_TRIPLE_FAULT;
1694}
1695
1696
1697/**
1698 * Loads host control registers, debug registers and MSRs as part of VM-exit.
1699 *
1700 * @param pVCpu The cross context virtual CPU structure.
1701 */
1702IEM_STATIC void iemVmxVmexitLoadHostControlRegsMsrs(PVMCPUCC pVCpu)
1703{
1704 /*
1705 * Load host control registers, debug registers and MSRs.
1706 * See Intel spec. 27.5.1 "Loading Host Control Registers, Debug Registers, MSRs".
1707 */
1708 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
1709 bool const fHostInLongMode = RT_BOOL(pVmcs->u32ExitCtls & VMX_EXIT_CTLS_HOST_ADDR_SPACE_SIZE);
1710
1711 /* CR0. */
1712 {
1713 /* Bits 63:32, 28:19, 17, 15:6, ET, CD, NW and CR0 fixed bits are not modified. */
1714 uint64_t const uCr0Mb1 = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Cr0Fixed0;
1715 uint64_t const uCr0Mb0 = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Cr0Fixed1;
1716 uint64_t const fCr0IgnMask = VMX_EXIT_HOST_CR0_IGNORE_MASK | uCr0Mb1 | ~uCr0Mb0;
1717 uint64_t const uHostCr0 = pVmcs->u64HostCr0.u;
1718 uint64_t const uGuestCr0 = pVCpu->cpum.GstCtx.cr0;
1719 uint64_t const uValidHostCr0 = (uHostCr0 & ~fCr0IgnMask) | (uGuestCr0 & fCr0IgnMask);
1720
1721 /* Verify we have not modified CR0 fixed bits in VMX non-root operation. */
1722 Assert((uGuestCr0 & uCr0Mb1) == uCr0Mb1);
1723 Assert((uGuestCr0 & ~uCr0Mb0) == 0);
1724 CPUMSetGuestCR0(pVCpu, uValidHostCr0);
1725 }
1726
1727 /* CR4. */
1728 {
1729 /* CR4 fixed bits are not modified. */
1730 uint64_t const uCr4Mb1 = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Cr4Fixed0;
1731 uint64_t const uCr4Mb0 = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Cr4Fixed1;
1732 uint64_t const fCr4IgnMask = uCr4Mb1 | ~uCr4Mb0;
1733 uint64_t const uHostCr4 = pVmcs->u64HostCr4.u;
1734 uint64_t const uGuestCr4 = pVCpu->cpum.GstCtx.cr4;
1735 uint64_t uValidHostCr4 = (uHostCr4 & ~fCr4IgnMask) | (uGuestCr4 & fCr4IgnMask);
1736 if (fHostInLongMode)
1737 uValidHostCr4 |= X86_CR4_PAE;
1738 else
1739 uValidHostCr4 &= ~(uint64_t)X86_CR4_PCIDE;
1740
1741 /* Verify we have not modified CR4 fixed bits in VMX non-root operation. */
1742 Assert((uGuestCr4 & uCr4Mb1) == uCr4Mb1);
1743 Assert((uGuestCr4 & ~uCr4Mb0) == 0);
1744 CPUMSetGuestCR4(pVCpu, uValidHostCr4);
1745 }
1746
1747 /* CR3 (host value validated while checking host-state during VM-entry). */
1748 pVCpu->cpum.GstCtx.cr3 = pVmcs->u64HostCr3.u;
1749
1750 /* DR7. */
1751 pVCpu->cpum.GstCtx.dr[7] = X86_DR7_INIT_VAL;
1752
1753 /** @todo NSTVMX: Support IA32_DEBUGCTL MSR */
1754
1755 /* Save SYSENTER CS, ESP, EIP (host value validated while checking host-state during VM-entry). */
1756 pVCpu->cpum.GstCtx.SysEnter.eip = pVmcs->u64HostSysenterEip.u;
1757 pVCpu->cpum.GstCtx.SysEnter.esp = pVmcs->u64HostSysenterEsp.u;
1758 pVCpu->cpum.GstCtx.SysEnter.cs = pVmcs->u32HostSysenterCs;
1759
1760 /* FS, GS bases are loaded later while we load host segment registers. */
1761
1762 /* EFER MSR (host value validated while checking host-state during VM-entry). */
1763 if (pVmcs->u32ExitCtls & VMX_EXIT_CTLS_LOAD_EFER_MSR)
1764 pVCpu->cpum.GstCtx.msrEFER = pVmcs->u64HostEferMsr.u;
1765 else if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode)
1766 {
1767 if (fHostInLongMode)
1768 pVCpu->cpum.GstCtx.msrEFER |= (MSR_K6_EFER_LMA | MSR_K6_EFER_LME);
1769 else
1770 pVCpu->cpum.GstCtx.msrEFER &= ~(MSR_K6_EFER_LMA | MSR_K6_EFER_LME);
1771 }
1772
1773 /* We don't support IA32_PERF_GLOBAL_CTRL MSR yet. */
1774
1775 /* PAT MSR (host value is validated while checking host-state during VM-entry). */
1776 if (pVmcs->u32ExitCtls & VMX_EXIT_CTLS_LOAD_PAT_MSR)
1777 pVCpu->cpum.GstCtx.msrPAT = pVmcs->u64HostPatMsr.u;
1778
1779 /* We don't support IA32_BNDCFGS MSR yet. */
1780}
1781
1782
1783/**
1784 * Loads host segment registers, GDTR, IDTR, LDTR and TR as part of VM-exit.
1785 *
1786 * @param pVCpu The cross context virtual CPU structure.
1787 */
1788IEM_STATIC void iemVmxVmexitLoadHostSegRegs(PVMCPUCC pVCpu)
1789{
1790 /*
1791 * Load host segment registers, GDTR, IDTR, LDTR and TR.
1792 * See Intel spec. 27.5.2 "Loading Host Segment and Descriptor-Table Registers".
1793 *
1794 * Warning! Be careful to not touch fields that are reserved by VT-x,
1795 * e.g. segment limit high bits stored in segment attributes (in bits 11:8).
1796 */
1797 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
1798 bool const fHostInLongMode = RT_BOOL(pVmcs->u32ExitCtls & VMX_EXIT_CTLS_HOST_ADDR_SPACE_SIZE);
1799
1800 /* CS, SS, ES, DS, FS, GS. */
1801 for (unsigned iSegReg = 0; iSegReg < X86_SREG_COUNT; iSegReg++)
1802 {
1803 RTSEL const HostSel = iemVmxVmcsGetHostSelReg(pVmcs, iSegReg);
1804 bool const fUnusable = RT_BOOL(HostSel == 0);
1805 PCPUMSELREG pSelReg = &pVCpu->cpum.GstCtx.aSRegs[iSegReg];
1806
1807 /* Selector. */
1808 pSelReg->Sel = HostSel;
1809 pSelReg->ValidSel = HostSel;
1810 pSelReg->fFlags = CPUMSELREG_FLAGS_VALID;
1811
1812 /* Limit. */
1813 pSelReg->u32Limit = 0xffffffff;
1814
1815 /* Base. */
1816 pSelReg->u64Base = 0;
1817
1818 /* Attributes. */
1819 if (iSegReg == X86_SREG_CS)
1820 {
1821 pSelReg->Attr.n.u4Type = X86_SEL_TYPE_CODE | X86_SEL_TYPE_READ | X86_SEL_TYPE_ACCESSED;
1822 pSelReg->Attr.n.u1DescType = 1;
1823 pSelReg->Attr.n.u2Dpl = 0;
1824 pSelReg->Attr.n.u1Present = 1;
1825 pSelReg->Attr.n.u1Long = fHostInLongMode;
1826 pSelReg->Attr.n.u1DefBig = !fHostInLongMode;
1827 pSelReg->Attr.n.u1Granularity = 1;
1828 Assert(!pSelReg->Attr.n.u1Unusable);
1829 Assert(!fUnusable);
1830 }
1831 else
1832 {
1833 pSelReg->Attr.n.u4Type = X86_SEL_TYPE_RW | X86_SEL_TYPE_ACCESSED;
1834 pSelReg->Attr.n.u1DescType = 1;
1835 pSelReg->Attr.n.u2Dpl = 0;
1836 pSelReg->Attr.n.u1Present = 1;
1837 pSelReg->Attr.n.u1DefBig = 1;
1838 pSelReg->Attr.n.u1Granularity = 1;
1839 pSelReg->Attr.n.u1Unusable = fUnusable;
1840 }
1841 }
1842
1843 /* FS base. */
1844 if ( !pVCpu->cpum.GstCtx.fs.Attr.n.u1Unusable
1845 || fHostInLongMode)
1846 {
1847 Assert(X86_IS_CANONICAL(pVmcs->u64HostFsBase.u));
1848 pVCpu->cpum.GstCtx.fs.u64Base = pVmcs->u64HostFsBase.u;
1849 }
1850
1851 /* GS base. */
1852 if ( !pVCpu->cpum.GstCtx.gs.Attr.n.u1Unusable
1853 || fHostInLongMode)
1854 {
1855 Assert(X86_IS_CANONICAL(pVmcs->u64HostGsBase.u));
1856 pVCpu->cpum.GstCtx.gs.u64Base = pVmcs->u64HostGsBase.u;
1857 }
1858
1859 /* TR. */
1860 Assert(X86_IS_CANONICAL(pVmcs->u64HostTrBase.u));
1861 Assert(!pVCpu->cpum.GstCtx.tr.Attr.n.u1Unusable);
1862 pVCpu->cpum.GstCtx.tr.Sel = pVmcs->HostTr;
1863 pVCpu->cpum.GstCtx.tr.ValidSel = pVmcs->HostTr;
1864 pVCpu->cpum.GstCtx.tr.fFlags = CPUMSELREG_FLAGS_VALID;
1865 pVCpu->cpum.GstCtx.tr.u32Limit = X86_SEL_TYPE_SYS_386_TSS_LIMIT_MIN;
1866 pVCpu->cpum.GstCtx.tr.u64Base = pVmcs->u64HostTrBase.u;
1867 pVCpu->cpum.GstCtx.tr.Attr.n.u4Type = X86_SEL_TYPE_SYS_386_TSS_BUSY;
1868 pVCpu->cpum.GstCtx.tr.Attr.n.u1DescType = 0;
1869 pVCpu->cpum.GstCtx.tr.Attr.n.u2Dpl = 0;
1870 pVCpu->cpum.GstCtx.tr.Attr.n.u1Present = 1;
1871 pVCpu->cpum.GstCtx.tr.Attr.n.u1DefBig = 0;
1872 pVCpu->cpum.GstCtx.tr.Attr.n.u1Granularity = 0;
1873
1874 /* LDTR (Warning! do not touch the base and limits here). */
1875 pVCpu->cpum.GstCtx.ldtr.Sel = 0;
1876 pVCpu->cpum.GstCtx.ldtr.ValidSel = 0;
1877 pVCpu->cpum.GstCtx.ldtr.fFlags = CPUMSELREG_FLAGS_VALID;
1878 pVCpu->cpum.GstCtx.ldtr.Attr.u = X86DESCATTR_UNUSABLE;
1879
1880 /* GDTR. */
1881 Assert(X86_IS_CANONICAL(pVmcs->u64HostGdtrBase.u));
1882 pVCpu->cpum.GstCtx.gdtr.pGdt = pVmcs->u64HostGdtrBase.u;
1883 pVCpu->cpum.GstCtx.gdtr.cbGdt = 0xffff;
1884
1885 /* IDTR.*/
1886 Assert(X86_IS_CANONICAL(pVmcs->u64HostIdtrBase.u));
1887 pVCpu->cpum.GstCtx.idtr.pIdt = pVmcs->u64HostIdtrBase.u;
1888 pVCpu->cpum.GstCtx.idtr.cbIdt = 0xffff;
1889}
1890
1891
1892/**
1893 * Checks host PDPTes as part of VM-exit.
1894 *
1895 * @param pVCpu The cross context virtual CPU structure.
1896 * @param uExitReason The VM-exit reason (for logging purposes).
1897 */
1898IEM_STATIC int iemVmxVmexitCheckHostPdptes(PVMCPUCC pVCpu, uint32_t uExitReason)
1899{
1900 /*
1901 * Check host PDPTEs.
1902 * See Intel spec. 27.5.4 "Checking and Loading Host Page-Directory-Pointer-Table Entries".
1903 */
1904 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
1905 const char *const pszFailure = "VMX-abort";
1906 bool const fHostInLongMode = RT_BOOL(pVmcs->u32ExitCtls & VMX_EXIT_CTLS_HOST_ADDR_SPACE_SIZE);
1907
1908 if ( (pVCpu->cpum.GstCtx.cr4 & X86_CR4_PAE)
1909 && !fHostInLongMode)
1910 {
1911 uint64_t const uHostCr3 = pVCpu->cpum.GstCtx.cr3 & X86_CR3_PAE_PAGE_MASK;
1912 X86PDPE aPdptes[X86_PG_PAE_PDPE_ENTRIES];
1913 int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), (void *)&aPdptes[0], uHostCr3, sizeof(aPdptes));
1914 if (RT_SUCCESS(rc))
1915 {
1916 for (unsigned iPdpte = 0; iPdpte < RT_ELEMENTS(aPdptes); iPdpte++)
1917 {
1918 if ( !(aPdptes[iPdpte].u & X86_PDPE_P)
1919 || !(aPdptes[iPdpte].u & X86_PDPE_PAE_MBZ_MASK))
1920 { /* likely */ }
1921 else
1922 {
1923 VMXVDIAG const enmDiag = iemVmxGetDiagVmexitPdpteRsvd(iPdpte);
1924 IEM_VMX_VMEXIT_FAILED_RET(pVCpu, uExitReason, pszFailure, enmDiag);
1925 }
1926 }
1927 }
1928 else
1929 IEM_VMX_VMEXIT_FAILED_RET(pVCpu, uExitReason, pszFailure, kVmxVDiag_Vmexit_HostPdpteCr3ReadPhys);
1930 }
1931
1932 NOREF(pszFailure);
1933 NOREF(uExitReason);
1934 return VINF_SUCCESS;
1935}
1936
1937
1938/**
1939 * Loads the host MSRs from the VM-exit MSR-load area as part of VM-exit.
1940 *
1941 * @returns VBox status code.
1942 * @param pVCpu The cross context virtual CPU structure.
1943 * @param pszInstr The VMX instruction name (for logging purposes).
1944 */
1945IEM_STATIC int iemVmxVmexitLoadHostAutoMsrs(PVMCPUCC pVCpu, uint32_t uExitReason)
1946{
1947 /*
1948 * Load host MSRs.
1949 * See Intel spec. 27.6 "Loading MSRs".
1950 */
1951 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
1952 const char *const pszFailure = "VMX-abort";
1953
1954 /*
1955 * The VM-exit MSR-load area address need not be a valid guest-physical address if the
1956 * VM-exit MSR load count is 0. If this is the case, bail early without reading it.
1957 * See Intel spec. 24.7.2 "VM-Exit Controls for MSRs".
1958 */
1959 uint32_t const cMsrs = pVmcs->u32ExitMsrLoadCount;
1960 if (!cMsrs)
1961 return VINF_SUCCESS;
1962
1963 /*
1964 * Verify the MSR auto-load count. Physical CPUs can behave unpredictably if the count
1965 * is exceeded including possibly raising #MC exceptions during VMX transition. Our
1966 * implementation causes a VMX-abort followed by a triple-fault.
1967 */
1968 bool const fIsMsrCountValid = iemVmxIsAutoMsrCountValid(pVCpu, cMsrs);
1969 if (fIsMsrCountValid)
1970 { /* likely */ }
1971 else
1972 IEM_VMX_VMEXIT_FAILED_RET(pVCpu, uExitReason, pszFailure, kVmxVDiag_Vmexit_MsrLoadCount);
1973
1974 RTGCPHYS const GCPhysVmExitMsrLoadArea = pVmcs->u64AddrExitMsrLoad.u;
1975 int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), (void *)pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pExitMsrLoadArea),
1976 GCPhysVmExitMsrLoadArea, cMsrs * sizeof(VMXAUTOMSR));
1977 if (RT_SUCCESS(rc))
1978 {
1979 PCVMXAUTOMSR pMsr = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pExitMsrLoadArea);
1980 Assert(pMsr);
1981 for (uint32_t idxMsr = 0; idxMsr < cMsrs; idxMsr++, pMsr++)
1982 {
1983 if ( !pMsr->u32Reserved
1984 && pMsr->u32Msr != MSR_K8_FS_BASE
1985 && pMsr->u32Msr != MSR_K8_GS_BASE
1986 && pMsr->u32Msr != MSR_K6_EFER
1987 && pMsr->u32Msr != MSR_IA32_SMM_MONITOR_CTL
1988 && pMsr->u32Msr >> 8 != MSR_IA32_X2APIC_START >> 8)
1989 {
1990 VBOXSTRICTRC rcStrict = CPUMSetGuestMsr(pVCpu, pMsr->u32Msr, pMsr->u64Value);
1991 if (rcStrict == VINF_SUCCESS)
1992 continue;
1993
1994 /*
1995 * If we're in ring-0, we cannot handle returns to ring-3 at this point and continue VM-exit.
1996 * If any nested hypervisor loads MSRs that require ring-3 handling, we cause a VMX-abort
1997 * recording the MSR index in the auxiliary info. field and indicated further by our
1998 * own, specific diagnostic code. Later, we can try implement handling of the MSR in ring-0
1999 * if possible, or come up with a better, generic solution.
2000 */
2001 pVCpu->cpum.GstCtx.hwvirt.vmx.uAbortAux = pMsr->u32Msr;
2002 VMXVDIAG const enmDiag = rcStrict == VINF_CPUM_R3_MSR_WRITE
2003 ? kVmxVDiag_Vmexit_MsrLoadRing3
2004 : kVmxVDiag_Vmexit_MsrLoad;
2005 IEM_VMX_VMEXIT_FAILED_RET(pVCpu, uExitReason, pszFailure, enmDiag);
2006 }
2007 else
2008 IEM_VMX_VMEXIT_FAILED_RET(pVCpu, uExitReason, pszFailure, kVmxVDiag_Vmexit_MsrLoadRsvd);
2009 }
2010 }
2011 else
2012 {
2013 AssertMsgFailed(("VM-exit: Failed to read MSR auto-load area at %#RGp, rc=%Rrc\n", GCPhysVmExitMsrLoadArea, rc));
2014 IEM_VMX_VMEXIT_FAILED_RET(pVCpu, uExitReason, pszFailure, kVmxVDiag_Vmexit_MsrLoadPtrReadPhys);
2015 }
2016
2017 NOREF(uExitReason);
2018 NOREF(pszFailure);
2019 return VINF_SUCCESS;
2020}
2021
2022
2023/**
2024 * Loads the host state as part of VM-exit.
2025 *
2026 * @returns Strict VBox status code.
2027 * @param pVCpu The cross context virtual CPU structure.
2028 * @param uExitReason The VM-exit reason (for logging purposes).
2029 */
2030IEM_STATIC VBOXSTRICTRC iemVmxVmexitLoadHostState(PVMCPUCC pVCpu, uint32_t uExitReason)
2031{
2032 /*
2033 * Load host state.
2034 * See Intel spec. 27.5 "Loading Host State".
2035 */
2036 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
2037 bool const fHostInLongMode = RT_BOOL(pVmcs->u32ExitCtls & VMX_EXIT_CTLS_HOST_ADDR_SPACE_SIZE);
2038
2039 /* We cannot return from a long-mode guest to a host that is not in long mode. */
2040 if ( CPUMIsGuestInLongMode(pVCpu)
2041 && !fHostInLongMode)
2042 {
2043 Log(("VM-exit from long-mode guest to host not in long-mode -> VMX-Abort\n"));
2044 return iemVmxAbort(pVCpu, VMXABORT_HOST_NOT_IN_LONG_MODE);
2045 }
2046
2047 iemVmxVmexitLoadHostControlRegsMsrs(pVCpu);
2048 iemVmxVmexitLoadHostSegRegs(pVCpu);
2049
2050 /*
2051 * Load host RIP, RSP and RFLAGS.
2052 * See Intel spec. 27.5.3 "Loading Host RIP, RSP and RFLAGS"
2053 */
2054 pVCpu->cpum.GstCtx.rip = pVmcs->u64HostRip.u;
2055 pVCpu->cpum.GstCtx.rsp = pVmcs->u64HostRsp.u;
2056 pVCpu->cpum.GstCtx.rflags.u = X86_EFL_1;
2057
2058 /* Clear address range monitoring. */
2059 EMMonitorWaitClear(pVCpu);
2060
2061 /* Perform the VMX transition (PGM updates). */
2062 VBOXSTRICTRC rcStrict = iemVmxWorldSwitch(pVCpu);
2063 if (rcStrict == VINF_SUCCESS)
2064 {
2065 /* Check host PDPTEs (only when we've fully switched page tables_. */
2066 /** @todo r=ramshankar: I don't know if PGM does this for us already or not... */
2067 int rc = iemVmxVmexitCheckHostPdptes(pVCpu, uExitReason);
2068 if (RT_FAILURE(rc))
2069 {
2070 Log(("VM-exit failed while restoring host PDPTEs -> VMX-Abort\n"));
2071 return iemVmxAbort(pVCpu, VMXBOART_HOST_PDPTE);
2072 }
2073 }
2074 else if (RT_SUCCESS(rcStrict))
2075 {
2076 Log3(("VM-exit: iemVmxWorldSwitch returns %Rrc (uExitReason=%u) -> Setting passup status\n", VBOXSTRICTRC_VAL(rcStrict),
2077 uExitReason));
2078 rcStrict = iemSetPassUpStatus(pVCpu, rcStrict);
2079 }
2080 else
2081 {
2082 Log3(("VM-exit: iemVmxWorldSwitch failed! rc=%Rrc (uExitReason=%u)\n", VBOXSTRICTRC_VAL(rcStrict), uExitReason));
2083 return VBOXSTRICTRC_VAL(rcStrict);
2084 }
2085
2086 Assert(rcStrict == VINF_SUCCESS);
2087
2088 /* Load MSRs from the VM-exit auto-load MSR area. */
2089 int rc = iemVmxVmexitLoadHostAutoMsrs(pVCpu, uExitReason);
2090 if (RT_FAILURE(rc))
2091 {
2092 Log(("VM-exit failed while loading host MSRs -> VMX-Abort\n"));
2093 return iemVmxAbort(pVCpu, VMXABORT_LOAD_HOST_MSR);
2094 }
2095 return VINF_SUCCESS;
2096}
2097
2098
2099/**
2100 * Gets VM-exit instruction information along with any displacement for an
2101 * instruction VM-exit.
2102 *
2103 * @returns The VM-exit instruction information.
2104 * @param pVCpu The cross context virtual CPU structure.
2105 * @param uExitReason The VM-exit reason.
2106 * @param uInstrId The VM-exit instruction identity (VMXINSTRID_XXX).
2107 * @param pGCPtrDisp Where to store the displacement field. Optional, can be
2108 * NULL.
2109 */
2110IEM_STATIC uint32_t iemVmxGetExitInstrInfo(PVMCPUCC pVCpu, uint32_t uExitReason, VMXINSTRID uInstrId, PRTGCPTR pGCPtrDisp)
2111{
2112 RTGCPTR GCPtrDisp;
2113 VMXEXITINSTRINFO ExitInstrInfo;
2114 ExitInstrInfo.u = 0;
2115
2116 /*
2117 * Get and parse the ModR/M byte from our decoded opcodes.
2118 */
2119 uint8_t bRm;
2120 uint8_t const offModRm = pVCpu->iem.s.offModRm;
2121 IEM_MODRM_GET_U8(pVCpu, bRm, offModRm);
2122 if ((bRm & X86_MODRM_MOD_MASK) == (3 << X86_MODRM_MOD_SHIFT))
2123 {
2124 /*
2125 * ModR/M indicates register addressing.
2126 *
2127 * The primary/secondary register operands are reported in the iReg1 or iReg2
2128 * fields depending on whether it is a read/write form.
2129 */
2130 uint8_t idxReg1;
2131 uint8_t idxReg2;
2132 if (!VMXINSTRID_IS_MODRM_PRIMARY_OP_W(uInstrId))
2133 {
2134 idxReg1 = ((bRm >> X86_MODRM_REG_SHIFT) & X86_MODRM_REG_SMASK) | pVCpu->iem.s.uRexReg;
2135 idxReg2 = (bRm & X86_MODRM_RM_MASK) | pVCpu->iem.s.uRexB;
2136 }
2137 else
2138 {
2139 idxReg1 = (bRm & X86_MODRM_RM_MASK) | pVCpu->iem.s.uRexB;
2140 idxReg2 = ((bRm >> X86_MODRM_REG_SHIFT) & X86_MODRM_REG_SMASK) | pVCpu->iem.s.uRexReg;
2141 }
2142 ExitInstrInfo.All.u2Scaling = 0;
2143 ExitInstrInfo.All.iReg1 = idxReg1;
2144 ExitInstrInfo.All.u3AddrSize = pVCpu->iem.s.enmEffAddrMode;
2145 ExitInstrInfo.All.fIsRegOperand = 1;
2146 ExitInstrInfo.All.uOperandSize = pVCpu->iem.s.enmEffOpSize;
2147 ExitInstrInfo.All.iSegReg = 0;
2148 ExitInstrInfo.All.iIdxReg = 0;
2149 ExitInstrInfo.All.fIdxRegInvalid = 1;
2150 ExitInstrInfo.All.iBaseReg = 0;
2151 ExitInstrInfo.All.fBaseRegInvalid = 1;
2152 ExitInstrInfo.All.iReg2 = idxReg2;
2153
2154 /* Displacement not applicable for register addressing. */
2155 GCPtrDisp = 0;
2156 }
2157 else
2158 {
2159 /*
2160 * ModR/M indicates memory addressing.
2161 */
2162 uint8_t uScale = 0;
2163 bool fBaseRegValid = false;
2164 bool fIdxRegValid = false;
2165 uint8_t iBaseReg = 0;
2166 uint8_t iIdxReg = 0;
2167 if (pVCpu->iem.s.enmEffAddrMode == IEMMODE_16BIT)
2168 {
2169 /*
2170 * Parse the ModR/M, displacement for 16-bit addressing mode.
2171 * See Intel instruction spec. Table 2-1. "16-Bit Addressing Forms with the ModR/M Byte".
2172 */
2173 uint16_t u16Disp = 0;
2174 uint8_t const offDisp = offModRm + sizeof(bRm);
2175 if ((bRm & (X86_MODRM_MOD_MASK | X86_MODRM_RM_MASK)) == 6)
2176 {
2177 /* Displacement without any registers. */
2178 IEM_DISP_GET_U16(pVCpu, u16Disp, offDisp);
2179 }
2180 else
2181 {
2182 /* Register (index and base). */
2183 switch (bRm & X86_MODRM_RM_MASK)
2184 {
2185 case 0: fBaseRegValid = true; iBaseReg = X86_GREG_xBX; fIdxRegValid = true; iIdxReg = X86_GREG_xSI; break;
2186 case 1: fBaseRegValid = true; iBaseReg = X86_GREG_xBX; fIdxRegValid = true; iIdxReg = X86_GREG_xDI; break;
2187 case 2: fBaseRegValid = true; iBaseReg = X86_GREG_xBP; fIdxRegValid = true; iIdxReg = X86_GREG_xSI; break;
2188 case 3: fBaseRegValid = true; iBaseReg = X86_GREG_xBP; fIdxRegValid = true; iIdxReg = X86_GREG_xDI; break;
2189 case 4: fIdxRegValid = true; iIdxReg = X86_GREG_xSI; break;
2190 case 5: fIdxRegValid = true; iIdxReg = X86_GREG_xDI; break;
2191 case 6: fBaseRegValid = true; iBaseReg = X86_GREG_xBP; break;
2192 case 7: fBaseRegValid = true; iBaseReg = X86_GREG_xBX; break;
2193 }
2194
2195 /* Register + displacement. */
2196 switch ((bRm >> X86_MODRM_MOD_SHIFT) & X86_MODRM_MOD_SMASK)
2197 {
2198 case 0: break;
2199 case 1: IEM_DISP_GET_S8_SX_U16(pVCpu, u16Disp, offDisp); break;
2200 case 2: IEM_DISP_GET_U16(pVCpu, u16Disp, offDisp); break;
2201 default:
2202 {
2203 /* Register addressing, handled at the beginning. */
2204 AssertMsgFailed(("ModR/M %#x implies register addressing, memory addressing expected!", bRm));
2205 break;
2206 }
2207 }
2208 }
2209
2210 Assert(!uScale); /* There's no scaling/SIB byte for 16-bit addressing. */
2211 GCPtrDisp = (int16_t)u16Disp; /* Sign-extend the displacement. */
2212 }
2213 else if (pVCpu->iem.s.enmEffAddrMode == IEMMODE_32BIT)
2214 {
2215 /*
2216 * Parse the ModR/M, SIB, displacement for 32-bit addressing mode.
2217 * See Intel instruction spec. Table 2-2. "32-Bit Addressing Forms with the ModR/M Byte".
2218 */
2219 uint32_t u32Disp = 0;
2220 if ((bRm & (X86_MODRM_MOD_MASK | X86_MODRM_RM_MASK)) == 5)
2221 {
2222 /* Displacement without any registers. */
2223 uint8_t const offDisp = offModRm + sizeof(bRm);
2224 IEM_DISP_GET_U32(pVCpu, u32Disp, offDisp);
2225 }
2226 else
2227 {
2228 /* Register (and perhaps scale, index and base). */
2229 uint8_t offDisp = offModRm + sizeof(bRm);
2230 iBaseReg = (bRm & X86_MODRM_RM_MASK);
2231 if (iBaseReg == 4)
2232 {
2233 /* An SIB byte follows the ModR/M byte, parse it. */
2234 uint8_t bSib;
2235 uint8_t const offSib = offModRm + sizeof(bRm);
2236 IEM_SIB_GET_U8(pVCpu, bSib, offSib);
2237
2238 /* A displacement may follow SIB, update its offset. */
2239 offDisp += sizeof(bSib);
2240
2241 /* Get the scale. */
2242 uScale = (bSib >> X86_SIB_SCALE_SHIFT) & X86_SIB_SCALE_SMASK;
2243
2244 /* Get the index register. */
2245 iIdxReg = (bSib >> X86_SIB_INDEX_SHIFT) & X86_SIB_INDEX_SMASK;
2246 fIdxRegValid = RT_BOOL(iIdxReg != 4);
2247
2248 /* Get the base register. */
2249 iBaseReg = bSib & X86_SIB_BASE_MASK;
2250 fBaseRegValid = true;
2251 if (iBaseReg == 5)
2252 {
2253 if ((bRm & X86_MODRM_MOD_MASK) == 0)
2254 {
2255 /* Mod is 0 implies a 32-bit displacement with no base. */
2256 fBaseRegValid = false;
2257 IEM_DISP_GET_U32(pVCpu, u32Disp, offDisp);
2258 }
2259 else
2260 {
2261 /* Mod is not 0 implies an 8-bit/32-bit displacement (handled below) with an EBP base. */
2262 iBaseReg = X86_GREG_xBP;
2263 }
2264 }
2265 }
2266
2267 /* Register + displacement. */
2268 switch ((bRm >> X86_MODRM_MOD_SHIFT) & X86_MODRM_MOD_SMASK)
2269 {
2270 case 0: /* Handled above */ break;
2271 case 1: IEM_DISP_GET_S8_SX_U32(pVCpu, u32Disp, offDisp); break;
2272 case 2: IEM_DISP_GET_U32(pVCpu, u32Disp, offDisp); break;
2273 default:
2274 {
2275 /* Register addressing, handled at the beginning. */
2276 AssertMsgFailed(("ModR/M %#x implies register addressing, memory addressing expected!", bRm));
2277 break;
2278 }
2279 }
2280 }
2281
2282 GCPtrDisp = (int32_t)u32Disp; /* Sign-extend the displacement. */
2283 }
2284 else
2285 {
2286 Assert(pVCpu->iem.s.enmEffAddrMode == IEMMODE_64BIT);
2287
2288 /*
2289 * Parse the ModR/M, SIB, displacement for 64-bit addressing mode.
2290 * See Intel instruction spec. 2.2 "IA-32e Mode".
2291 */
2292 uint64_t u64Disp = 0;
2293 bool const fRipRelativeAddr = RT_BOOL((bRm & (X86_MODRM_MOD_MASK | X86_MODRM_RM_MASK)) == 5);
2294 if (fRipRelativeAddr)
2295 {
2296 /*
2297 * RIP-relative addressing mode.
2298 *
2299 * The displacement is 32-bit signed implying an offset range of +/-2G.
2300 * See Intel instruction spec. 2.2.1.6 "RIP-Relative Addressing".
2301 */
2302 uint8_t const offDisp = offModRm + sizeof(bRm);
2303 IEM_DISP_GET_S32_SX_U64(pVCpu, u64Disp, offDisp);
2304 }
2305 else
2306 {
2307 uint8_t offDisp = offModRm + sizeof(bRm);
2308
2309 /*
2310 * Register (and perhaps scale, index and base).
2311 *
2312 * REX.B extends the most-significant bit of the base register. However, REX.B
2313 * is ignored while determining whether an SIB follows the opcode. Hence, we
2314 * shall OR any REX.B bit -after- inspecting for an SIB byte below.
2315 *
2316 * See Intel instruction spec. Table 2-5. "Special Cases of REX Encodings".
2317 */
2318 iBaseReg = (bRm & X86_MODRM_RM_MASK);
2319 if (iBaseReg == 4)
2320 {
2321 /* An SIB byte follows the ModR/M byte, parse it. Displacement (if any) follows SIB. */
2322 uint8_t bSib;
2323 uint8_t const offSib = offModRm + sizeof(bRm);
2324 IEM_SIB_GET_U8(pVCpu, bSib, offSib);
2325
2326 /* Displacement may follow SIB, update its offset. */
2327 offDisp += sizeof(bSib);
2328
2329 /* Get the scale. */
2330 uScale = (bSib >> X86_SIB_SCALE_SHIFT) & X86_SIB_SCALE_SMASK;
2331
2332 /* Get the index. */
2333 iIdxReg = ((bSib >> X86_SIB_INDEX_SHIFT) & X86_SIB_INDEX_SMASK) | pVCpu->iem.s.uRexIndex;
2334 fIdxRegValid = RT_BOOL(iIdxReg != 4); /* R12 -can- be used as an index register. */
2335
2336 /* Get the base. */
2337 iBaseReg = (bSib & X86_SIB_BASE_MASK);
2338 fBaseRegValid = true;
2339 if (iBaseReg == 5)
2340 {
2341 if ((bRm & X86_MODRM_MOD_MASK) == 0)
2342 {
2343 /* Mod is 0 implies a signed 32-bit displacement with no base. */
2344 IEM_DISP_GET_S32_SX_U64(pVCpu, u64Disp, offDisp);
2345 }
2346 else
2347 {
2348 /* Mod is non-zero implies an 8-bit/32-bit displacement (handled below) with RBP or R13 as base. */
2349 iBaseReg = pVCpu->iem.s.uRexB ? X86_GREG_x13 : X86_GREG_xBP;
2350 }
2351 }
2352 }
2353 iBaseReg |= pVCpu->iem.s.uRexB;
2354
2355 /* Register + displacement. */
2356 switch ((bRm >> X86_MODRM_MOD_SHIFT) & X86_MODRM_MOD_SMASK)
2357 {
2358 case 0: /* Handled above */ break;
2359 case 1: IEM_DISP_GET_S8_SX_U64(pVCpu, u64Disp, offDisp); break;
2360 case 2: IEM_DISP_GET_S32_SX_U64(pVCpu, u64Disp, offDisp); break;
2361 default:
2362 {
2363 /* Register addressing, handled at the beginning. */
2364 AssertMsgFailed(("ModR/M %#x implies register addressing, memory addressing expected!", bRm));
2365 break;
2366 }
2367 }
2368 }
2369
2370 GCPtrDisp = fRipRelativeAddr ? pVCpu->cpum.GstCtx.rip + u64Disp : u64Disp;
2371 }
2372
2373 /*
2374 * The primary or secondary register operand is reported in iReg2 depending
2375 * on whether the primary operand is in read/write form.
2376 */
2377 uint8_t idxReg2;
2378 if (!VMXINSTRID_IS_MODRM_PRIMARY_OP_W(uInstrId))
2379 {
2380 idxReg2 = bRm & X86_MODRM_RM_MASK;
2381 if (pVCpu->iem.s.enmEffAddrMode == IEMMODE_64BIT)
2382 idxReg2 |= pVCpu->iem.s.uRexB;
2383 }
2384 else
2385 {
2386 idxReg2 = (bRm >> X86_MODRM_REG_SHIFT) & X86_MODRM_REG_SMASK;
2387 if (pVCpu->iem.s.enmEffAddrMode == IEMMODE_64BIT)
2388 idxReg2 |= pVCpu->iem.s.uRexReg;
2389 }
2390 ExitInstrInfo.All.u2Scaling = uScale;
2391 ExitInstrInfo.All.iReg1 = 0; /* Not applicable for memory addressing. */
2392 ExitInstrInfo.All.u3AddrSize = pVCpu->iem.s.enmEffAddrMode;
2393 ExitInstrInfo.All.fIsRegOperand = 0;
2394 ExitInstrInfo.All.uOperandSize = pVCpu->iem.s.enmEffOpSize;
2395 ExitInstrInfo.All.iSegReg = pVCpu->iem.s.iEffSeg;
2396 ExitInstrInfo.All.iIdxReg = iIdxReg;
2397 ExitInstrInfo.All.fIdxRegInvalid = !fIdxRegValid;
2398 ExitInstrInfo.All.iBaseReg = iBaseReg;
2399 ExitInstrInfo.All.iIdxReg = !fBaseRegValid;
2400 ExitInstrInfo.All.iReg2 = idxReg2;
2401 }
2402
2403 /*
2404 * Handle exceptions to the norm for certain instructions.
2405 * (e.g. some instructions convey an instruction identity in place of iReg2).
2406 */
2407 switch (uExitReason)
2408 {
2409 case VMX_EXIT_GDTR_IDTR_ACCESS:
2410 {
2411 Assert(VMXINSTRID_IS_VALID(uInstrId));
2412 Assert(VMXINSTRID_GET_ID(uInstrId) == (uInstrId & 0x3));
2413 ExitInstrInfo.GdtIdt.u2InstrId = VMXINSTRID_GET_ID(uInstrId);
2414 ExitInstrInfo.GdtIdt.u2Undef0 = 0;
2415 break;
2416 }
2417
2418 case VMX_EXIT_LDTR_TR_ACCESS:
2419 {
2420 Assert(VMXINSTRID_IS_VALID(uInstrId));
2421 Assert(VMXINSTRID_GET_ID(uInstrId) == (uInstrId & 0x3));
2422 ExitInstrInfo.LdtTr.u2InstrId = VMXINSTRID_GET_ID(uInstrId);
2423 ExitInstrInfo.LdtTr.u2Undef0 = 0;
2424 break;
2425 }
2426
2427 case VMX_EXIT_RDRAND:
2428 case VMX_EXIT_RDSEED:
2429 {
2430 Assert(ExitInstrInfo.RdrandRdseed.u2OperandSize != 3);
2431 break;
2432 }
2433 }
2434
2435 /* Update displacement and return the constructed VM-exit instruction information field. */
2436 if (pGCPtrDisp)
2437 *pGCPtrDisp = GCPtrDisp;
2438
2439 return ExitInstrInfo.u;
2440}
2441
2442
2443/**
2444 * VMX VM-exit handler.
2445 *
2446 * @returns Strict VBox status code.
2447 * @retval VINF_VMX_VMEXIT when the VM-exit is successful.
2448 * @retval VINF_EM_TRIPLE_FAULT when VM-exit is unsuccessful and leads to a
2449 * triple-fault.
2450 *
2451 * @param pVCpu The cross context virtual CPU structure.
2452 * @param uExitReason The VM-exit reason.
2453 * @param u64ExitQual The Exit qualification.
2454 */
2455IEM_STATIC VBOXSTRICTRC iemVmxVmexit(PVMCPUCC pVCpu, uint32_t uExitReason, uint64_t u64ExitQual)
2456{
2457# if defined(VBOX_WITH_NESTED_HWVIRT_ONLY_IN_IEM) && !defined(IN_RING3)
2458 RT_NOREF3(pVCpu, uExitReason, u64ExitQual);
2459 AssertMsgFailed(("VM-exit should only be invoked from ring-3 when nested-guest executes only in ring-3!\n"));
2460 return VERR_IEM_IPE_7;
2461# else
2462 PVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
2463 Assert(pVmcs);
2464
2465 /*
2466 * Import all the guest-CPU state.
2467 *
2468 * HM on returning to guest execution would have to reset up a whole lot of state
2469 * anyway, (e.g., VM-entry/VM-exit controls) and we do not ever import a part of
2470 * the state and flag reloading the entire state on re-entry. So import the entire
2471 * state here, see HMNotifyVmxNstGstVmexit() for more comments.
2472 */
2473 IEM_CTX_IMPORT_RET(pVCpu, CPUMCTX_EXTRN_ALL);
2474
2475 /*
2476 * Ensure VM-entry interruption information valid bit is cleared.
2477 *
2478 * We do it here on every VM-exit so that even premature VM-exits (e.g. those caused
2479 * by invalid-guest state or machine-check exceptions) also clear this bit.
2480 *
2481 * See Intel spec. 27.2 "Recording VM-exit Information And Updating VM-entry control fields".
2482 */
2483 if (VMX_ENTRY_INT_INFO_IS_VALID(pVmcs->u32EntryIntInfo))
2484 pVmcs->u32EntryIntInfo &= ~VMX_ENTRY_INT_INFO_VALID;
2485
2486 /*
2487 * Update the VM-exit reason and Exit qualification.
2488 * Other VMCS read-only data fields are expected to be updated by the caller already.
2489 */
2490 pVmcs->u32RoExitReason = uExitReason;
2491 pVmcs->u64RoExitQual.u = u64ExitQual;
2492
2493 Log3(("vmexit: reason=%#RX32 qual=%#RX64 cs:rip=%04x:%#RX64 cr0=%#RX64 cr3=%#RX64 cr4=%#RX64\n", uExitReason,
2494 pVmcs->u64RoExitQual.u, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip, pVCpu->cpum.GstCtx.cr0,
2495 pVCpu->cpum.GstCtx.cr3, pVCpu->cpum.GstCtx.cr4));
2496
2497 /*
2498 * Update the IDT-vectoring information fields if the VM-exit is triggered during delivery of an event.
2499 * See Intel spec. 27.2.4 "Information for VM Exits During Event Delivery".
2500 */
2501 {
2502 uint8_t uVector;
2503 uint32_t fFlags;
2504 uint32_t uErrCode;
2505 bool const fInEventDelivery = IEMGetCurrentXcpt(pVCpu, &uVector, &fFlags, &uErrCode, NULL /* puCr2 */);
2506 if (fInEventDelivery)
2507 {
2508 /*
2509 * A VM-exit is not considered to occur during event delivery when the VM-exit is
2510 * caused by a triple-fault or the original event results in a double-fault that
2511 * causes the VM exit directly (exception bitmap). Therefore, we must not set the
2512 * original event information into the IDT-vectoring information fields.
2513 *
2514 * See Intel spec. 27.2.4 "Information for VM Exits During Event Delivery".
2515 */
2516 if ( uExitReason != VMX_EXIT_TRIPLE_FAULT
2517 && ( uExitReason != VMX_EXIT_XCPT_OR_NMI
2518 || !VMX_EXIT_INT_INFO_IS_XCPT_DF(pVmcs->u32RoExitIntInfo)))
2519 {
2520 uint8_t const uIdtVectoringType = iemVmxGetEventType(uVector, fFlags);
2521 uint8_t const fErrCodeValid = RT_BOOL(fFlags & IEM_XCPT_FLAGS_ERR);
2522 uint32_t const uIdtVectoringInfo = RT_BF_MAKE(VMX_BF_IDT_VECTORING_INFO_VECTOR, uVector)
2523 | RT_BF_MAKE(VMX_BF_IDT_VECTORING_INFO_TYPE, uIdtVectoringType)
2524 | RT_BF_MAKE(VMX_BF_IDT_VECTORING_INFO_ERR_CODE_VALID, fErrCodeValid)
2525 | RT_BF_MAKE(VMX_BF_IDT_VECTORING_INFO_VALID, 1);
2526 iemVmxVmcsSetIdtVectoringInfo(pVCpu, uIdtVectoringInfo);
2527 iemVmxVmcsSetIdtVectoringErrCode(pVCpu, uErrCode);
2528 LogFlow(("vmexit: idt_info=%#RX32 idt_err_code=%#RX32 cr2=%#RX64\n", uIdtVectoringInfo, uErrCode,
2529 pVCpu->cpum.GstCtx.cr2));
2530 }
2531 }
2532 }
2533
2534 /* The following VMCS fields should always be zero since we don't support injecting SMIs into a guest. */
2535 Assert(pVmcs->u64RoIoRcx.u == 0);
2536 Assert(pVmcs->u64RoIoRsi.u == 0);
2537 Assert(pVmcs->u64RoIoRdi.u == 0);
2538 Assert(pVmcs->u64RoIoRip.u == 0);
2539
2540 /* We should not cause an NMI-window/interrupt-window VM-exit when injecting events as part of VM-entry. */
2541 if (!CPUMIsGuestVmxInterceptEvents(&pVCpu->cpum.GstCtx))
2542 {
2543 Assert(uExitReason != VMX_EXIT_NMI_WINDOW);
2544 Assert(uExitReason != VMX_EXIT_INT_WINDOW);
2545 }
2546
2547 /* For exception or NMI VM-exits the VM-exit interruption info. field must be valid. */
2548 Assert(uExitReason != VMX_EXIT_XCPT_OR_NMI || VMX_EXIT_INT_INFO_IS_VALID(pVmcs->u32RoExitIntInfo));
2549
2550 /*
2551 * Save the guest state back into the VMCS.
2552 * We only need to save the state when the VM-entry was successful.
2553 */
2554 bool const fVmentryFailed = VMX_EXIT_REASON_HAS_ENTRY_FAILED(uExitReason);
2555 if (!fVmentryFailed)
2556 {
2557 /*
2558 * If we support storing EFER.LMA into IA32e-mode guest field on VM-exit, we need to do that now.
2559 * See Intel spec. 27.2 "Recording VM-exit Information And Updating VM-entry Control".
2560 *
2561 * It is not clear from the Intel spec. if this is done only when VM-entry succeeds.
2562 * If a VM-exit happens before loading guest EFER, we risk restoring the host EFER.LMA
2563 * as guest-CPU state would not been modified. Hence for now, we do this only when
2564 * the VM-entry succeeded.
2565 */
2566 /** @todo r=ramshankar: Figure out if this bit gets set to host EFER.LMA on real
2567 * hardware when VM-exit fails during VM-entry (e.g. VERR_VMX_INVALID_GUEST_STATE). */
2568 if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fVmxExitSaveEferLma)
2569 {
2570 if (pVCpu->cpum.GstCtx.msrEFER & MSR_K6_EFER_LMA)
2571 pVmcs->u32EntryCtls |= VMX_ENTRY_CTLS_IA32E_MODE_GUEST;
2572 else
2573 pVmcs->u32EntryCtls &= ~VMX_ENTRY_CTLS_IA32E_MODE_GUEST;
2574 }
2575
2576 /*
2577 * The rest of the high bits of the VM-exit reason are only relevant when the VM-exit
2578 * occurs in enclave mode/SMM which we don't support yet.
2579 *
2580 * If we ever add support for it, we can pass just the lower bits to the functions
2581 * below, till then an assert should suffice.
2582 */
2583 Assert(!RT_HI_U16(uExitReason));
2584
2585 /* Save the guest state into the VMCS and restore guest MSRs from the auto-store guest MSR area. */
2586 iemVmxVmexitSaveGuestState(pVCpu, uExitReason);
2587 int rc = iemVmxVmexitSaveGuestAutoMsrs(pVCpu, uExitReason);
2588 if (RT_SUCCESS(rc))
2589 { /* likely */ }
2590 else
2591 return iemVmxAbort(pVCpu, VMXABORT_SAVE_GUEST_MSRS);
2592
2593 /* Clear any saved NMI-blocking state so we don't assert on next VM-entry (if it was in effect on the previous one). */
2594 pVCpu->cpum.GstCtx.hwvirt.fLocalForcedActions &= ~VMCPU_FF_BLOCK_NMIS;
2595 }
2596 else
2597 {
2598 /* Restore the NMI-blocking state if VM-entry failed due to invalid guest state or while loading MSRs. */
2599 uint32_t const uExitReasonBasic = VMX_EXIT_REASON_BASIC(uExitReason);
2600 if ( uExitReasonBasic == VMX_EXIT_ERR_INVALID_GUEST_STATE
2601 || uExitReasonBasic == VMX_EXIT_ERR_MSR_LOAD)
2602 iemVmxVmexitRestoreNmiBlockingFF(pVCpu);
2603 }
2604
2605 /*
2606 * Stop any running VMX-preemption timer if necessary.
2607 */
2608 if (pVmcs->u32PinCtls & VMX_PIN_CTLS_PREEMPT_TIMER)
2609 CPUMStopGuestVmxPremptTimer(pVCpu);
2610
2611 /*
2612 * Clear any pending VMX nested-guest force-flags.
2613 * These force-flags have no effect on (outer) guest execution and will
2614 * be re-evaluated and setup on the next nested-guest VM-entry.
2615 */
2616 VMCPU_FF_CLEAR_MASK(pVCpu, VMCPU_FF_VMX_ALL_MASK);
2617
2618 /* Restore the host (outer guest) state. */
2619 VBOXSTRICTRC rcStrict = iemVmxVmexitLoadHostState(pVCpu, uExitReason);
2620 if (RT_SUCCESS(rcStrict))
2621 {
2622 Assert(rcStrict == VINF_SUCCESS);
2623 rcStrict = VINF_VMX_VMEXIT;
2624 }
2625 else
2626 Log3(("vmexit: Loading host-state failed. uExitReason=%u rc=%Rrc\n", uExitReason, VBOXSTRICTRC_VAL(rcStrict)));
2627
2628 /* We're no longer in nested-guest execution mode. */
2629 pVCpu->cpum.GstCtx.hwvirt.vmx.fInVmxNonRootMode = false;
2630
2631 /* Notify HM that the current VMCS fields have been modified. */
2632 HMNotifyVmxNstGstCurrentVmcsChanged(pVCpu);
2633
2634 /* Notify HM that we've completed the VM-exit. */
2635 HMNotifyVmxNstGstVmexit(pVCpu);
2636
2637# if defined(VBOX_WITH_NESTED_HWVIRT_ONLY_IN_IEM) && defined(IN_RING3)
2638 /* Revert any IEM-only nested-guest execution policy, otherwise return rcStrict. */
2639 Log(("vmexit: Disabling IEM-only EM execution policy!\n"));
2640 int rcSched = EMR3SetExecutionPolicy(pVCpu->CTX_SUFF(pVM)->pUVM, EMEXECPOLICY_IEM_ALL, false);
2641 if (rcSched != VINF_SUCCESS)
2642 iemSetPassUpStatus(pVCpu, rcSched);
2643# endif
2644 return rcStrict;
2645# endif
2646}
2647
2648
2649/**
2650 * VMX VM-exit handler for VM-exits due to instruction execution.
2651 *
2652 * This is intended for instructions where the caller provides all the relevant
2653 * VM-exit information.
2654 *
2655 * @returns Strict VBox status code.
2656 * @param pVCpu The cross context virtual CPU structure.
2657 * @param pExitInfo Pointer to the VM-exit information.
2658 */
2659IEM_STATIC VBOXSTRICTRC iemVmxVmexitInstrWithInfo(PVMCPUCC pVCpu, PCVMXVEXITINFO pExitInfo)
2660{
2661 /*
2662 * For instructions where any of the following fields are not applicable:
2663 * - Exit qualification must be cleared.
2664 * - VM-exit instruction info. is undefined.
2665 * - Guest-linear address is undefined.
2666 * - Guest-physical address is undefined.
2667 *
2668 * The VM-exit instruction length is mandatory for all VM-exits that are caused by
2669 * instruction execution. For VM-exits that are not due to instruction execution this
2670 * field is undefined.
2671 *
2672 * In our implementation in IEM, all undefined fields are generally cleared. However,
2673 * if the caller supplies information (from say the physical CPU directly) it is
2674 * then possible that the undefined fields are not cleared.
2675 *
2676 * See Intel spec. 27.2.1 "Basic VM-Exit Information".
2677 * See Intel spec. 27.2.4 "Information for VM Exits Due to Instruction Execution".
2678 */
2679 Assert(pExitInfo);
2680 AssertMsg(pExitInfo->uReason <= VMX_EXIT_MAX, ("uReason=%u\n", pExitInfo->uReason));
2681 AssertMsg(pExitInfo->cbInstr >= 1 && pExitInfo->cbInstr <= 15,
2682 ("uReason=%u cbInstr=%u\n", pExitInfo->uReason, pExitInfo->cbInstr));
2683
2684 /* Update all the relevant fields from the VM-exit instruction information struct. */
2685 iemVmxVmcsSetExitInstrInfo(pVCpu, pExitInfo->InstrInfo.u);
2686 iemVmxVmcsSetExitGuestLinearAddr(pVCpu, pExitInfo->u64GuestLinearAddr);
2687 iemVmxVmcsSetExitGuestPhysAddr(pVCpu, pExitInfo->u64GuestPhysAddr);
2688 iemVmxVmcsSetExitInstrLen(pVCpu, pExitInfo->cbInstr);
2689
2690 /* Perform the VM-exit. */
2691 return iemVmxVmexit(pVCpu, pExitInfo->uReason, pExitInfo->u64Qual);
2692}
2693
2694
2695/**
2696 * VMX VM-exit handler for VM-exits due to instruction execution.
2697 *
2698 * This is intended for instructions that only provide the VM-exit instruction
2699 * length.
2700 *
2701 * @param pVCpu The cross context virtual CPU structure.
2702 * @param uExitReason The VM-exit reason.
2703 * @param cbInstr The instruction length in bytes.
2704 */
2705IEM_STATIC VBOXSTRICTRC iemVmxVmexitInstr(PVMCPUCC pVCpu, uint32_t uExitReason, uint8_t cbInstr)
2706{
2707 VMXVEXITINFO ExitInfo;
2708 RT_ZERO(ExitInfo);
2709 ExitInfo.uReason = uExitReason;
2710 ExitInfo.cbInstr = cbInstr;
2711
2712#ifdef VBOX_STRICT
2713 /*
2714 * To prevent us from shooting ourselves in the foot.
2715 * The follow instructions should convey more than just the instruction length.
2716 */
2717 switch (uExitReason)
2718 {
2719 case VMX_EXIT_INVEPT:
2720 case VMX_EXIT_INVPCID:
2721 case VMX_EXIT_INVVPID:
2722 case VMX_EXIT_LDTR_TR_ACCESS:
2723 case VMX_EXIT_GDTR_IDTR_ACCESS:
2724 case VMX_EXIT_VMCLEAR:
2725 case VMX_EXIT_VMPTRLD:
2726 case VMX_EXIT_VMPTRST:
2727 case VMX_EXIT_VMREAD:
2728 case VMX_EXIT_VMWRITE:
2729 case VMX_EXIT_VMXON:
2730 case VMX_EXIT_XRSTORS:
2731 case VMX_EXIT_XSAVES:
2732 case VMX_EXIT_RDRAND:
2733 case VMX_EXIT_RDSEED:
2734 case VMX_EXIT_IO_INSTR:
2735 AssertMsgFailedReturn(("Use iemVmxVmexitInstrNeedsInfo for uExitReason=%u\n", uExitReason), VERR_IEM_IPE_5);
2736 break;
2737 }
2738#endif
2739
2740 return iemVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
2741}
2742
2743
2744/**
2745 * VMX VM-exit handler for VM-exits due to instruction execution.
2746 *
2747 * This is intended for instructions that have a ModR/M byte and update the VM-exit
2748 * instruction information and Exit qualification fields.
2749 *
2750 * @param pVCpu The cross context virtual CPU structure.
2751 * @param uExitReason The VM-exit reason.
2752 * @param uInstrid The instruction identity (VMXINSTRID_XXX).
2753 * @param cbInstr The instruction length in bytes.
2754 *
2755 * @remarks Do not use this for INS/OUTS instruction.
2756 */
2757IEM_STATIC VBOXSTRICTRC iemVmxVmexitInstrNeedsInfo(PVMCPUCC pVCpu, uint32_t uExitReason, VMXINSTRID uInstrId, uint8_t cbInstr)
2758{
2759 VMXVEXITINFO ExitInfo;
2760 RT_ZERO(ExitInfo);
2761 ExitInfo.uReason = uExitReason;
2762 ExitInfo.cbInstr = cbInstr;
2763
2764 /*
2765 * Update the Exit qualification field with displacement bytes.
2766 * See Intel spec. 27.2.1 "Basic VM-Exit Information".
2767 */
2768 switch (uExitReason)
2769 {
2770 case VMX_EXIT_INVEPT:
2771 case VMX_EXIT_INVPCID:
2772 case VMX_EXIT_INVVPID:
2773 case VMX_EXIT_LDTR_TR_ACCESS:
2774 case VMX_EXIT_GDTR_IDTR_ACCESS:
2775 case VMX_EXIT_VMCLEAR:
2776 case VMX_EXIT_VMPTRLD:
2777 case VMX_EXIT_VMPTRST:
2778 case VMX_EXIT_VMREAD:
2779 case VMX_EXIT_VMWRITE:
2780 case VMX_EXIT_VMXON:
2781 case VMX_EXIT_XRSTORS:
2782 case VMX_EXIT_XSAVES:
2783 case VMX_EXIT_RDRAND:
2784 case VMX_EXIT_RDSEED:
2785 {
2786 /* Construct the VM-exit instruction information. */
2787 RTGCPTR GCPtrDisp;
2788 uint32_t const uInstrInfo = iemVmxGetExitInstrInfo(pVCpu, uExitReason, uInstrId, &GCPtrDisp);
2789
2790 /* Update the VM-exit instruction information. */
2791 ExitInfo.InstrInfo.u = uInstrInfo;
2792
2793 /* Update the Exit qualification. */
2794 ExitInfo.u64Qual = GCPtrDisp;
2795 break;
2796 }
2797
2798 default:
2799 AssertMsgFailedReturn(("Use instruction-specific handler\n"), VERR_IEM_IPE_5);
2800 break;
2801 }
2802
2803 return iemVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
2804}
2805
2806
2807/**
2808 * VMX VM-exit handler for VM-exits due to INVLPG.
2809 *
2810 * @returns Strict VBox status code.
2811 * @param pVCpu The cross context virtual CPU structure.
2812 * @param GCPtrPage The guest-linear address of the page being invalidated.
2813 * @param cbInstr The instruction length in bytes.
2814 */
2815IEM_STATIC VBOXSTRICTRC iemVmxVmexitInstrInvlpg(PVMCPUCC pVCpu, RTGCPTR GCPtrPage, uint8_t cbInstr)
2816{
2817 VMXVEXITINFO ExitInfo;
2818 RT_ZERO(ExitInfo);
2819 ExitInfo.uReason = VMX_EXIT_INVLPG;
2820 ExitInfo.cbInstr = cbInstr;
2821 ExitInfo.u64Qual = GCPtrPage;
2822 Assert(IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode || !RT_HI_U32(ExitInfo.u64Qual));
2823
2824 return iemVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
2825}
2826
2827
2828/**
2829 * VMX VM-exit handler for VM-exits due to LMSW.
2830 *
2831 * @returns Strict VBox status code.
2832 * @param pVCpu The cross context virtual CPU structure.
2833 * @param uGuestCr0 The current guest CR0.
2834 * @param pu16NewMsw The machine-status word specified in LMSW's source
2835 * operand. This will be updated depending on the VMX
2836 * guest/host CR0 mask if LMSW is not intercepted.
2837 * @param GCPtrEffDst The guest-linear address of the source operand in case
2838 * of a memory operand. For register operand, pass
2839 * NIL_RTGCPTR.
2840 * @param cbInstr The instruction length in bytes.
2841 */
2842IEM_STATIC VBOXSTRICTRC iemVmxVmexitInstrLmsw(PVMCPUCC pVCpu, uint32_t uGuestCr0, uint16_t *pu16NewMsw, RTGCPTR GCPtrEffDst,
2843 uint8_t cbInstr)
2844{
2845 Assert(pu16NewMsw);
2846
2847 uint16_t const uNewMsw = *pu16NewMsw;
2848 if (CPUMIsGuestVmxLmswInterceptSet(&pVCpu->cpum.GstCtx, uNewMsw))
2849 {
2850 Log2(("lmsw: Guest intercept -> VM-exit\n"));
2851
2852 VMXVEXITINFO ExitInfo;
2853 RT_ZERO(ExitInfo);
2854 ExitInfo.uReason = VMX_EXIT_MOV_CRX;
2855 ExitInfo.cbInstr = cbInstr;
2856
2857 bool const fMemOperand = RT_BOOL(GCPtrEffDst != NIL_RTGCPTR);
2858 if (fMemOperand)
2859 {
2860 Assert(IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode || !RT_HI_U32(GCPtrEffDst));
2861 ExitInfo.u64GuestLinearAddr = GCPtrEffDst;
2862 }
2863
2864 ExitInfo.u64Qual = RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_REGISTER, 0) /* CR0 */
2865 | RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_ACCESS, VMX_EXIT_QUAL_CRX_ACCESS_LMSW)
2866 | RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_LMSW_OP, fMemOperand)
2867 | RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_LMSW_DATA, uNewMsw);
2868
2869 return iemVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
2870 }
2871
2872 /*
2873 * If LMSW did not cause a VM-exit, any CR0 bits in the range 0:3 that is set in the
2874 * CR0 guest/host mask must be left unmodified.
2875 *
2876 * See Intel Spec. 25.3 "Changes To Instruction Behavior In VMX Non-root Operation".
2877 */
2878 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
2879 Assert(pVmcs);
2880 uint32_t const fGstHostMask = pVmcs->u64Cr0Mask.u;
2881 uint32_t const fGstHostLmswMask = fGstHostMask & (X86_CR0_PE | X86_CR0_MP | X86_CR0_EM | X86_CR0_TS);
2882 *pu16NewMsw = (uGuestCr0 & fGstHostLmswMask) | (uNewMsw & ~fGstHostLmswMask);
2883
2884 return VINF_VMX_INTERCEPT_NOT_ACTIVE;
2885}
2886
2887
2888/**
2889 * VMX VM-exit handler for VM-exits due to CLTS.
2890 *
2891 * @returns Strict VBox status code.
2892 * @retval VINF_VMX_MODIFIES_BEHAVIOR if the CLTS instruction did not cause a
2893 * VM-exit but must not modify the guest CR0.TS bit.
2894 * @retval VINF_VMX_INTERCEPT_NOT_ACTIVE if the CLTS instruction did not cause a
2895 * VM-exit and modification to the guest CR0.TS bit is allowed (subject to
2896 * CR0 fixed bits in VMX operation).
2897 * @param pVCpu The cross context virtual CPU structure.
2898 * @param cbInstr The instruction length in bytes.
2899 */
2900IEM_STATIC VBOXSTRICTRC iemVmxVmexitInstrClts(PVMCPUCC pVCpu, uint8_t cbInstr)
2901{
2902 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
2903 Assert(pVmcs);
2904
2905 uint32_t const fGstHostMask = pVmcs->u64Cr0Mask.u;
2906 uint32_t const fReadShadow = pVmcs->u64Cr0ReadShadow.u;
2907
2908 /*
2909 * If CR0.TS is owned by the host:
2910 * - If CR0.TS is set in the read-shadow, we must cause a VM-exit.
2911 * - If CR0.TS is cleared in the read-shadow, no VM-exit is caused and the
2912 * CLTS instruction completes without clearing CR0.TS.
2913 *
2914 * See Intel spec. 25.1.3 "Instructions That Cause VM Exits Conditionally".
2915 */
2916 if (fGstHostMask & X86_CR0_TS)
2917 {
2918 if (fReadShadow & X86_CR0_TS)
2919 {
2920 Log2(("clts: Guest intercept -> VM-exit\n"));
2921
2922 VMXVEXITINFO ExitInfo;
2923 RT_ZERO(ExitInfo);
2924 ExitInfo.uReason = VMX_EXIT_MOV_CRX;
2925 ExitInfo.cbInstr = cbInstr;
2926 ExitInfo.u64Qual = RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_REGISTER, 0) /* CR0 */
2927 | RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_ACCESS, VMX_EXIT_QUAL_CRX_ACCESS_CLTS);
2928 return iemVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
2929 }
2930
2931 return VINF_VMX_MODIFIES_BEHAVIOR;
2932 }
2933
2934 /*
2935 * If CR0.TS is not owned by the host, the CLTS instructions operates normally
2936 * and may modify CR0.TS (subject to CR0 fixed bits in VMX operation).
2937 */
2938 return VINF_VMX_INTERCEPT_NOT_ACTIVE;
2939}
2940
2941
2942/**
2943 * VMX VM-exit handler for VM-exits due to 'Mov CR0,GReg' and 'Mov CR4,GReg'
2944 * (CR0/CR4 write).
2945 *
2946 * @returns Strict VBox status code.
2947 * @param pVCpu The cross context virtual CPU structure.
2948 * @param iCrReg The control register (either CR0 or CR4).
2949 * @param uGuestCrX The current guest CR0/CR4.
2950 * @param puNewCrX Pointer to the new CR0/CR4 value. Will be updated if no
2951 * VM-exit is caused.
2952 * @param iGReg The general register from which the CR0/CR4 value is being
2953 * loaded.
2954 * @param cbInstr The instruction length in bytes.
2955 */
2956IEM_STATIC VBOXSTRICTRC iemVmxVmexitInstrMovToCr0Cr4(PVMCPUCC pVCpu, uint8_t iCrReg, uint64_t *puNewCrX, uint8_t iGReg,
2957 uint8_t cbInstr)
2958{
2959 Assert(puNewCrX);
2960 Assert(iCrReg == 0 || iCrReg == 4);
2961 Assert(iGReg < X86_GREG_COUNT);
2962
2963 uint64_t const uNewCrX = *puNewCrX;
2964 if (CPUMIsGuestVmxMovToCr0Cr4InterceptSet(&pVCpu->cpum.GstCtx, iCrReg, uNewCrX))
2965 {
2966 Log2(("mov_Cr_Rd: (CR%u) Guest intercept -> VM-exit\n", iCrReg));
2967
2968 VMXVEXITINFO ExitInfo;
2969 RT_ZERO(ExitInfo);
2970 ExitInfo.uReason = VMX_EXIT_MOV_CRX;
2971 ExitInfo.cbInstr = cbInstr;
2972 ExitInfo.u64Qual = RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_REGISTER, iCrReg)
2973 | RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_ACCESS, VMX_EXIT_QUAL_CRX_ACCESS_WRITE)
2974 | RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_GENREG, iGReg);
2975 return iemVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
2976 }
2977
2978 /*
2979 * If the Mov-to-CR0/CR4 did not cause a VM-exit, any bits owned by the host
2980 * must not be modified the instruction.
2981 *
2982 * See Intel Spec. 25.3 "Changes To Instruction Behavior In VMX Non-root Operation".
2983 */
2984 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
2985 Assert(pVmcs);
2986 uint64_t uGuestCrX;
2987 uint64_t fGstHostMask;
2988 if (iCrReg == 0)
2989 {
2990 IEM_CTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR0);
2991 uGuestCrX = pVCpu->cpum.GstCtx.cr0;
2992 fGstHostMask = pVmcs->u64Cr0Mask.u;
2993 }
2994 else
2995 {
2996 IEM_CTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR4);
2997 uGuestCrX = pVCpu->cpum.GstCtx.cr4;
2998 fGstHostMask = pVmcs->u64Cr4Mask.u;
2999 }
3000
3001 *puNewCrX = (uGuestCrX & fGstHostMask) | (*puNewCrX & ~fGstHostMask);
3002 return VINF_VMX_INTERCEPT_NOT_ACTIVE;
3003}
3004
3005
3006/**
3007 * VMX VM-exit handler for VM-exits due to 'Mov GReg,CR3' (CR3 read).
3008 *
3009 * @returns VBox strict status code.
3010 * @param pVCpu The cross context virtual CPU structure.
3011 * @param iGReg The general register to which the CR3 value is being stored.
3012 * @param cbInstr The instruction length in bytes.
3013 */
3014IEM_STATIC VBOXSTRICTRC iemVmxVmexitInstrMovFromCr3(PVMCPUCC pVCpu, uint8_t iGReg, uint8_t cbInstr)
3015{
3016 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
3017 Assert(pVmcs);
3018 Assert(iGReg < X86_GREG_COUNT);
3019 IEM_CTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR3);
3020
3021 /*
3022 * If the CR3-store exiting control is set, we must cause a VM-exit.
3023 * See Intel spec. 25.1.3 "Instructions That Cause VM Exits Conditionally".
3024 */
3025 if (pVmcs->u32ProcCtls & VMX_PROC_CTLS_CR3_STORE_EXIT)
3026 {
3027 Log2(("mov_Rd_Cr: (CR3) Guest intercept -> VM-exit\n"));
3028
3029 VMXVEXITINFO ExitInfo;
3030 RT_ZERO(ExitInfo);
3031 ExitInfo.uReason = VMX_EXIT_MOV_CRX;
3032 ExitInfo.cbInstr = cbInstr;
3033 ExitInfo.u64Qual = RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_REGISTER, 3) /* CR3 */
3034 | RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_ACCESS, VMX_EXIT_QUAL_CRX_ACCESS_READ)
3035 | RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_GENREG, iGReg);
3036 return iemVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
3037 }
3038
3039 return VINF_VMX_INTERCEPT_NOT_ACTIVE;
3040}
3041
3042
3043/**
3044 * VMX VM-exit handler for VM-exits due to 'Mov CR3,GReg' (CR3 write).
3045 *
3046 * @returns VBox strict status code.
3047 * @param pVCpu The cross context virtual CPU structure.
3048 * @param uNewCr3 The new CR3 value.
3049 * @param iGReg The general register from which the CR3 value is being
3050 * loaded.
3051 * @param cbInstr The instruction length in bytes.
3052 */
3053IEM_STATIC VBOXSTRICTRC iemVmxVmexitInstrMovToCr3(PVMCPUCC pVCpu, uint64_t uNewCr3, uint8_t iGReg, uint8_t cbInstr)
3054{
3055 Assert(iGReg < X86_GREG_COUNT);
3056
3057 /*
3058 * If the CR3-load exiting control is set and the new CR3 value does not
3059 * match any of the CR3-target values in the VMCS, we must cause a VM-exit.
3060 *
3061 * See Intel spec. 25.1.3 "Instructions That Cause VM Exits Conditionally".
3062 */
3063 if (CPUMIsGuestVmxMovToCr3InterceptSet(pVCpu, uNewCr3))
3064 {
3065 Log2(("mov_Cr_Rd: (CR3) Guest intercept -> VM-exit\n"));
3066
3067 VMXVEXITINFO ExitInfo;
3068 RT_ZERO(ExitInfo);
3069 ExitInfo.uReason = VMX_EXIT_MOV_CRX;
3070 ExitInfo.cbInstr = cbInstr;
3071 ExitInfo.u64Qual = RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_REGISTER, 3) /* CR3 */
3072 | RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_ACCESS, VMX_EXIT_QUAL_CRX_ACCESS_WRITE)
3073 | RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_GENREG, iGReg);
3074 return iemVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
3075 }
3076
3077 return VINF_VMX_INTERCEPT_NOT_ACTIVE;
3078}
3079
3080
3081/**
3082 * VMX VM-exit handler for VM-exits due to 'Mov GReg,CR8' (CR8 read).
3083 *
3084 * @returns VBox strict status code.
3085 * @param pVCpu The cross context virtual CPU structure.
3086 * @param iGReg The general register to which the CR8 value is being stored.
3087 * @param cbInstr The instruction length in bytes.
3088 */
3089IEM_STATIC VBOXSTRICTRC iemVmxVmexitInstrMovFromCr8(PVMCPUCC pVCpu, uint8_t iGReg, uint8_t cbInstr)
3090{
3091 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
3092 Assert(pVmcs);
3093 Assert(iGReg < X86_GREG_COUNT);
3094
3095 /*
3096 * If the CR8-store exiting control is set, we must cause a VM-exit.
3097 * See Intel spec. 25.1.3 "Instructions That Cause VM Exits Conditionally".
3098 */
3099 if (pVmcs->u32ProcCtls & VMX_PROC_CTLS_CR8_STORE_EXIT)
3100 {
3101 Log2(("mov_Rd_Cr: (CR8) Guest intercept -> VM-exit\n"));
3102
3103 VMXVEXITINFO ExitInfo;
3104 RT_ZERO(ExitInfo);
3105 ExitInfo.uReason = VMX_EXIT_MOV_CRX;
3106 ExitInfo.cbInstr = cbInstr;
3107 ExitInfo.u64Qual = RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_REGISTER, 8) /* CR8 */
3108 | RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_ACCESS, VMX_EXIT_QUAL_CRX_ACCESS_READ)
3109 | RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_GENREG, iGReg);
3110 return iemVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
3111 }
3112
3113 return VINF_VMX_INTERCEPT_NOT_ACTIVE;
3114}
3115
3116
3117/**
3118 * VMX VM-exit handler for VM-exits due to 'Mov CR8,GReg' (CR8 write).
3119 *
3120 * @returns VBox strict status code.
3121 * @param pVCpu The cross context virtual CPU structure.
3122 * @param iGReg The general register from which the CR8 value is being
3123 * loaded.
3124 * @param cbInstr The instruction length in bytes.
3125 */
3126IEM_STATIC VBOXSTRICTRC iemVmxVmexitInstrMovToCr8(PVMCPUCC pVCpu, uint8_t iGReg, uint8_t cbInstr)
3127{
3128 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
3129 Assert(pVmcs);
3130 Assert(iGReg < X86_GREG_COUNT);
3131
3132 /*
3133 * If the CR8-load exiting control is set, we must cause a VM-exit.
3134 * See Intel spec. 25.1.3 "Instructions That Cause VM Exits Conditionally".
3135 */
3136 if (pVmcs->u32ProcCtls & VMX_PROC_CTLS_CR8_LOAD_EXIT)
3137 {
3138 Log2(("mov_Cr_Rd: (CR8) Guest intercept -> VM-exit\n"));
3139
3140 VMXVEXITINFO ExitInfo;
3141 RT_ZERO(ExitInfo);
3142 ExitInfo.uReason = VMX_EXIT_MOV_CRX;
3143 ExitInfo.cbInstr = cbInstr;
3144 ExitInfo.u64Qual = RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_REGISTER, 8) /* CR8 */
3145 | RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_ACCESS, VMX_EXIT_QUAL_CRX_ACCESS_WRITE)
3146 | RT_BF_MAKE(VMX_BF_EXIT_QUAL_CRX_GENREG, iGReg);
3147 return iemVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
3148 }
3149
3150 return VINF_VMX_INTERCEPT_NOT_ACTIVE;
3151}
3152
3153
3154/**
3155 * VMX VM-exit handler for VM-exits due to 'Mov DRx,GReg' (DRx write) and 'Mov
3156 * GReg,DRx' (DRx read).
3157 *
3158 * @returns VBox strict status code.
3159 * @param pVCpu The cross context virtual CPU structure.
3160 * @param uInstrid The instruction identity (VMXINSTRID_MOV_TO_DRX or
3161 * VMXINSTRID_MOV_FROM_DRX).
3162 * @param iDrReg The debug register being accessed.
3163 * @param iGReg The general register to/from which the DRx value is being
3164 * store/loaded.
3165 * @param cbInstr The instruction length in bytes.
3166 */
3167IEM_STATIC VBOXSTRICTRC iemVmxVmexitInstrMovDrX(PVMCPUCC pVCpu, VMXINSTRID uInstrId, uint8_t iDrReg, uint8_t iGReg,
3168 uint8_t cbInstr)
3169{
3170 Assert(iDrReg <= 7);
3171 Assert(uInstrId == VMXINSTRID_MOV_TO_DRX || uInstrId == VMXINSTRID_MOV_FROM_DRX);
3172 Assert(iGReg < X86_GREG_COUNT);
3173
3174 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
3175 Assert(pVmcs);
3176
3177 if (pVmcs->u32ProcCtls & VMX_PROC_CTLS_MOV_DR_EXIT)
3178 {
3179 uint32_t const uDirection = uInstrId == VMXINSTRID_MOV_TO_DRX ? VMX_EXIT_QUAL_DRX_DIRECTION_WRITE
3180 : VMX_EXIT_QUAL_DRX_DIRECTION_READ;
3181 VMXVEXITINFO ExitInfo;
3182 RT_ZERO(ExitInfo);
3183 ExitInfo.uReason = VMX_EXIT_MOV_DRX;
3184 ExitInfo.cbInstr = cbInstr;
3185 ExitInfo.u64Qual = RT_BF_MAKE(VMX_BF_EXIT_QUAL_DRX_REGISTER, iDrReg)
3186 | RT_BF_MAKE(VMX_BF_EXIT_QUAL_DRX_DIRECTION, uDirection)
3187 | RT_BF_MAKE(VMX_BF_EXIT_QUAL_DRX_GENREG, iGReg);
3188 return iemVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
3189 }
3190
3191 return VINF_VMX_INTERCEPT_NOT_ACTIVE;
3192}
3193
3194
3195/**
3196 * VMX VM-exit handler for VM-exits due to I/O instructions (IN and OUT).
3197 *
3198 * @returns VBox strict status code.
3199 * @param pVCpu The cross context virtual CPU structure.
3200 * @param uInstrId The VM-exit instruction identity (VMXINSTRID_IO_IN or
3201 * VMXINSTRID_IO_OUT).
3202 * @param u16Port The I/O port being accessed.
3203 * @param fImm Whether the I/O port was encoded using an immediate operand
3204 * or the implicit DX register.
3205 * @param cbAccess The size of the I/O access in bytes (1, 2 or 4 bytes).
3206 * @param cbInstr The instruction length in bytes.
3207 */
3208IEM_STATIC VBOXSTRICTRC iemVmxVmexitInstrIo(PVMCPUCC pVCpu, VMXINSTRID uInstrId, uint16_t u16Port, bool fImm, uint8_t cbAccess,
3209 uint8_t cbInstr)
3210{
3211 Assert(uInstrId == VMXINSTRID_IO_IN || uInstrId == VMXINSTRID_IO_OUT);
3212 Assert(cbAccess == 1 || cbAccess == 2 || cbAccess == 4);
3213
3214 bool const fIntercept = CPUMIsGuestVmxIoInterceptSet(pVCpu, u16Port, cbAccess);
3215 if (fIntercept)
3216 {
3217 uint32_t const uDirection = uInstrId == VMXINSTRID_IO_IN ? VMX_EXIT_QUAL_IO_DIRECTION_IN
3218 : VMX_EXIT_QUAL_IO_DIRECTION_OUT;
3219 VMXVEXITINFO ExitInfo;
3220 RT_ZERO(ExitInfo);
3221 ExitInfo.uReason = VMX_EXIT_IO_INSTR;
3222 ExitInfo.cbInstr = cbInstr;
3223 ExitInfo.u64Qual = RT_BF_MAKE(VMX_BF_EXIT_QUAL_IO_WIDTH, cbAccess - 1)
3224 | RT_BF_MAKE(VMX_BF_EXIT_QUAL_IO_DIRECTION, uDirection)
3225 | RT_BF_MAKE(VMX_BF_EXIT_QUAL_IO_ENCODING, fImm)
3226 | RT_BF_MAKE(VMX_BF_EXIT_QUAL_IO_PORT, u16Port);
3227 return iemVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
3228 }
3229
3230 return VINF_VMX_INTERCEPT_NOT_ACTIVE;
3231}
3232
3233
3234/**
3235 * VMX VM-exit handler for VM-exits due to string I/O instructions (INS and OUTS).
3236 *
3237 * @returns VBox strict status code.
3238 * @param pVCpu The cross context virtual CPU structure.
3239 * @param uInstrId The VM-exit instruction identity (VMXINSTRID_IO_INS or
3240 * VMXINSTRID_IO_OUTS).
3241 * @param u16Port The I/O port being accessed.
3242 * @param cbAccess The size of the I/O access in bytes (1, 2 or 4 bytes).
3243 * @param fRep Whether the instruction has a REP prefix or not.
3244 * @param ExitInstrInfo The VM-exit instruction info. field.
3245 * @param cbInstr The instruction length in bytes.
3246 */
3247IEM_STATIC VBOXSTRICTRC iemVmxVmexitInstrStrIo(PVMCPUCC pVCpu, VMXINSTRID uInstrId, uint16_t u16Port, uint8_t cbAccess, bool fRep,
3248 VMXEXITINSTRINFO ExitInstrInfo, uint8_t cbInstr)
3249{
3250 Assert(uInstrId == VMXINSTRID_IO_INS || uInstrId == VMXINSTRID_IO_OUTS);
3251 Assert(cbAccess == 1 || cbAccess == 2 || cbAccess == 4);
3252 Assert(ExitInstrInfo.StrIo.iSegReg < X86_SREG_COUNT);
3253 Assert(ExitInstrInfo.StrIo.u3AddrSize == 0 || ExitInstrInfo.StrIo.u3AddrSize == 1 || ExitInstrInfo.StrIo.u3AddrSize == 2);
3254 Assert(uInstrId != VMXINSTRID_IO_INS || ExitInstrInfo.StrIo.iSegReg == X86_SREG_ES);
3255
3256 bool const fIntercept = CPUMIsGuestVmxIoInterceptSet(pVCpu, u16Port, cbAccess);
3257 if (fIntercept)
3258 {
3259 /*
3260 * Figure out the guest-linear address and the direction bit (INS/OUTS).
3261 */
3262 /** @todo r=ramshankar: Is there something in IEM that already does this? */
3263 static uint64_t const s_auAddrSizeMasks[] = { UINT64_C(0xffff), UINT64_C(0xffffffff), UINT64_C(0xffffffffffffffff) };
3264 uint8_t const iSegReg = ExitInstrInfo.StrIo.iSegReg;
3265 uint8_t const uAddrSize = ExitInstrInfo.StrIo.u3AddrSize;
3266 uint64_t const uAddrSizeMask = s_auAddrSizeMasks[uAddrSize];
3267
3268 uint32_t uDirection;
3269 uint64_t uGuestLinearAddr;
3270 if (uInstrId == VMXINSTRID_IO_INS)
3271 {
3272 uDirection = VMX_EXIT_QUAL_IO_DIRECTION_IN;
3273 uGuestLinearAddr = pVCpu->cpum.GstCtx.aSRegs[iSegReg].u64Base + (pVCpu->cpum.GstCtx.rdi & uAddrSizeMask);
3274 }
3275 else
3276 {
3277 uDirection = VMX_EXIT_QUAL_IO_DIRECTION_OUT;
3278 uGuestLinearAddr = pVCpu->cpum.GstCtx.aSRegs[iSegReg].u64Base + (pVCpu->cpum.GstCtx.rsi & uAddrSizeMask);
3279 }
3280
3281 /*
3282 * If the segment is unusable, the guest-linear address in undefined.
3283 * We shall clear it for consistency.
3284 *
3285 * See Intel spec. 27.2.1 "Basic VM-Exit Information".
3286 */
3287 if (pVCpu->cpum.GstCtx.aSRegs[iSegReg].Attr.n.u1Unusable)
3288 uGuestLinearAddr = 0;
3289
3290 VMXVEXITINFO ExitInfo;
3291 RT_ZERO(ExitInfo);
3292 ExitInfo.uReason = VMX_EXIT_IO_INSTR;
3293 ExitInfo.cbInstr = cbInstr;
3294 ExitInfo.u64GuestLinearAddr = uGuestLinearAddr;
3295 ExitInfo.u64Qual = RT_BF_MAKE(VMX_BF_EXIT_QUAL_IO_WIDTH, cbAccess - 1)
3296 | RT_BF_MAKE(VMX_BF_EXIT_QUAL_IO_DIRECTION, uDirection)
3297 | RT_BF_MAKE(VMX_BF_EXIT_QUAL_IO_IS_STRING, 1)
3298 | RT_BF_MAKE(VMX_BF_EXIT_QUAL_IO_IS_REP, fRep)
3299 | RT_BF_MAKE(VMX_BF_EXIT_QUAL_IO_ENCODING, VMX_EXIT_QUAL_IO_ENCODING_DX)
3300 | RT_BF_MAKE(VMX_BF_EXIT_QUAL_IO_PORT, u16Port);
3301 if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fVmxInsOutInfo)
3302 ExitInfo.InstrInfo = ExitInstrInfo;
3303 return iemVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
3304 }
3305
3306 return VINF_VMX_INTERCEPT_NOT_ACTIVE;
3307}
3308
3309
3310/**
3311 * VMX VM-exit handler for VM-exits due to MWAIT.
3312 *
3313 * @returns VBox strict status code.
3314 * @param pVCpu The cross context virtual CPU structure.
3315 * @param fMonitorHwArmed Whether the address-range monitor hardware is armed.
3316 * @param cbInstr The instruction length in bytes.
3317 */
3318IEM_STATIC VBOXSTRICTRC iemVmxVmexitInstrMwait(PVMCPUCC pVCpu, bool fMonitorHwArmed, uint8_t cbInstr)
3319{
3320 VMXVEXITINFO ExitInfo;
3321 RT_ZERO(ExitInfo);
3322 ExitInfo.uReason = VMX_EXIT_MWAIT;
3323 ExitInfo.cbInstr = cbInstr;
3324 ExitInfo.u64Qual = fMonitorHwArmed;
3325 return iemVmxVmexitInstrWithInfo(pVCpu, &ExitInfo);
3326}
3327
3328
3329/**
3330 * VMX VM-exit handler for VM-exits due to PAUSE.
3331 *
3332 * @returns VBox strict status code.
3333 * @param pVCpu The cross context virtual CPU structure.
3334 * @param cbInstr The instruction length in bytes.
3335 */
3336IEM_STATIC VBOXSTRICTRC iemVmxVmexitInstrPause(PVMCPUCC pVCpu, uint8_t cbInstr)
3337{
3338 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
3339 Assert(pVmcs);
3340
3341 /*
3342 * The PAUSE VM-exit is controlled by the "PAUSE exiting" control and the
3343 * "PAUSE-loop exiting" control.
3344 *
3345 * The PLE-Gap is the maximum number of TSC ticks between two successive executions of
3346 * the PAUSE instruction before we cause a VM-exit. The PLE-Window is the maximum amount
3347 * of TSC ticks the guest is allowed to execute in a pause loop before we must cause
3348 * a VM-exit.
3349 *
3350 * See Intel spec. 24.6.13 "Controls for PAUSE-Loop Exiting".
3351 * See Intel spec. 25.1.3 "Instructions That Cause VM Exits Conditionally".
3352 */
3353 bool fIntercept = false;
3354 if (pVmcs->u32ProcCtls & VMX_PROC_CTLS_PAUSE_EXIT)
3355 fIntercept = true;
3356 else if ( (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_PAUSE_LOOP_EXIT)
3357 && pVCpu->iem.s.uCpl == 0)
3358 {
3359 IEM_CTX_IMPORT_RET(pVCpu, CPUMCTX_EXTRN_HWVIRT);
3360
3361 /*
3362 * A previous-PAUSE-tick value of 0 is used to identify the first time
3363 * execution of a PAUSE instruction after VM-entry at CPL 0. We must
3364 * consider this to be the first execution of PAUSE in a loop according
3365 * to the Intel.
3366 *
3367 * All subsequent records for the previous-PAUSE-tick we ensure that it
3368 * cannot be zero by OR'ing 1 to rule out the TSC wrap-around cases at 0.
3369 */
3370 uint64_t *puFirstPauseLoopTick = &pVCpu->cpum.GstCtx.hwvirt.vmx.uFirstPauseLoopTick;
3371 uint64_t *puPrevPauseTick = &pVCpu->cpum.GstCtx.hwvirt.vmx.uPrevPauseTick;
3372 uint64_t const uTick = TMCpuTickGet(pVCpu);
3373 uint32_t const uPleGap = pVmcs->u32PleGap;
3374 uint32_t const uPleWindow = pVmcs->u32PleWindow;
3375 if ( *puPrevPauseTick == 0
3376 || uTick - *puPrevPauseTick > uPleGap)
3377 *puFirstPauseLoopTick = uTick;
3378 else if (uTick - *puFirstPauseLoopTick > uPleWindow)
3379 fIntercept = true;
3380
3381 *puPrevPauseTick = uTick | 1;
3382 }
3383
3384 if (fIntercept)
3385 return iemVmxVmexitInstr(pVCpu, VMX_EXIT_PAUSE, cbInstr);
3386
3387 return VINF_VMX_INTERCEPT_NOT_ACTIVE;
3388}
3389
3390
3391/**
3392 * VMX VM-exit handler for VM-exits due to task switches.
3393 *
3394 * @returns VBox strict status code.
3395 * @param pVCpu The cross context virtual CPU structure.
3396 * @param enmTaskSwitch The cause of the task switch.
3397 * @param SelNewTss The selector of the new TSS.
3398 * @param cbInstr The instruction length in bytes.
3399 */
3400IEM_STATIC VBOXSTRICTRC iemVmxVmexitTaskSwitch(PVMCPUCC pVCpu, IEMTASKSWITCH enmTaskSwitch, RTSEL SelNewTss, uint8_t cbInstr)
3401{
3402 /*
3403 * Task-switch VM-exits are unconditional and provide the Exit qualification.
3404 *
3405 * If the cause of the task switch is due to execution of CALL, IRET or the JMP
3406 * instruction or delivery of the exception generated by one of these instructions
3407 * lead to a task switch through a task gate in the IDT, we need to provide the
3408 * VM-exit instruction length. Any other means of invoking a task switch VM-exit
3409 * leaves the VM-exit instruction length field undefined.
3410 *
3411 * See Intel spec. 25.2 "Other Causes Of VM Exits".
3412 * See Intel spec. 27.2.4 "Information for VM Exits Due to Instruction Execution".
3413 */
3414 Assert(cbInstr <= 15);
3415
3416 uint8_t uType;
3417 switch (enmTaskSwitch)
3418 {
3419 case IEMTASKSWITCH_CALL: uType = VMX_EXIT_QUAL_TASK_SWITCH_TYPE_CALL; break;
3420 case IEMTASKSWITCH_IRET: uType = VMX_EXIT_QUAL_TASK_SWITCH_TYPE_IRET; break;
3421 case IEMTASKSWITCH_JUMP: uType = VMX_EXIT_QUAL_TASK_SWITCH_TYPE_JMP; break;
3422 case IEMTASKSWITCH_INT_XCPT: uType = VMX_EXIT_QUAL_TASK_SWITCH_TYPE_IDT; break;
3423 IEM_NOT_REACHED_DEFAULT_CASE_RET();
3424 }
3425
3426 uint64_t const u64ExitQual = RT_BF_MAKE(VMX_BF_EXIT_QUAL_TASK_SWITCH_NEW_TSS, SelNewTss)
3427 | RT_BF_MAKE(VMX_BF_EXIT_QUAL_TASK_SWITCH_SOURCE, uType);
3428 iemVmxVmcsSetExitInstrLen(pVCpu, cbInstr);
3429 return iemVmxVmexit(pVCpu, VMX_EXIT_TASK_SWITCH, u64ExitQual);
3430}
3431
3432
3433/**
3434 * VMX VM-exit handler for trap-like VM-exits.
3435 *
3436 * @returns VBox strict status code.
3437 * @param pVCpu The cross context virtual CPU structure.
3438 * @param pExitInfo Pointer to the VM-exit information.
3439 * @param pExitEventInfo Pointer to the VM-exit event information.
3440 */
3441IEM_STATIC VBOXSTRICTRC iemVmxVmexitTrapLikeWithInfo(PVMCPUCC pVCpu, PCVMXVEXITINFO pExitInfo)
3442{
3443 Assert(VMXIsVmexitTrapLike(pExitInfo->uReason));
3444 iemVmxVmcsSetGuestPendingDbgXcpts(pVCpu, pExitInfo->u64GuestPendingDbgXcpts);
3445 return iemVmxVmexit(pVCpu, pExitInfo->uReason, pExitInfo->u64Qual);
3446}
3447
3448
3449/**
3450 * VMX VM-exit handler for VM-exits due to task switches.
3451 *
3452 * This is intended for task switches where the caller provides all the relevant
3453 * VM-exit information.
3454 *
3455 * @returns VBox strict status code.
3456 * @param pVCpu The cross context virtual CPU structure.
3457 * @param pExitInfo Pointer to the VM-exit information.
3458 * @param pExitEventInfo Pointer to the VM-exit event information.
3459 */
3460IEM_STATIC VBOXSTRICTRC iemVmxVmexitTaskSwitchWithInfo(PVMCPUCC pVCpu, PCVMXVEXITINFO pExitInfo,
3461 PCVMXVEXITEVENTINFO pExitEventInfo)
3462{
3463 Assert(pExitInfo->uReason == VMX_EXIT_TASK_SWITCH);
3464 iemVmxVmcsSetExitInstrLen(pVCpu, pExitInfo->cbInstr);
3465 iemVmxVmcsSetIdtVectoringInfo(pVCpu, pExitEventInfo->uIdtVectoringInfo);
3466 iemVmxVmcsSetIdtVectoringErrCode(pVCpu, pExitEventInfo->uIdtVectoringErrCode);
3467 return iemVmxVmexit(pVCpu, VMX_EXIT_TASK_SWITCH, pExitInfo->u64Qual);
3468}
3469
3470
3471/**
3472 * VMX VM-exit handler for VM-exits due to expiring of the preemption timer.
3473 *
3474 * @returns VBox strict status code.
3475 * @param pVCpu The cross context virtual CPU structure.
3476 */
3477IEM_STATIC VBOXSTRICTRC iemVmxVmexitPreemptTimer(PVMCPUCC pVCpu)
3478{
3479 PVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
3480 Assert(pVmcs);
3481 Assert(VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_VMX_PREEMPT_TIMER));
3482 Assert(pVmcs->u32PinCtls & VMX_PIN_CTLS_PREEMPT_TIMER);
3483
3484 /* Import the hardware virtualization state (for nested-guest VM-entry TSC-tick). */
3485 IEM_CTX_IMPORT_RET(pVCpu, CPUMCTX_EXTRN_HWVIRT);
3486
3487 /* Save the VMX-preemption timer value (of 0) back in to the VMCS if the CPU supports this feature. */
3488 if (pVmcs->u32ExitCtls & VMX_EXIT_CTLS_SAVE_PREEMPT_TIMER)
3489 pVmcs->u32PreemptTimer = 0;
3490
3491 /* Cause the VMX-preemption timer VM-exit. The Exit qualification MBZ. */
3492 return iemVmxVmexit(pVCpu, VMX_EXIT_PREEMPT_TIMER, 0 /* u64ExitQual */);
3493}
3494
3495
3496/**
3497 * VMX VM-exit handler for VM-exits due to external interrupts.
3498 *
3499 * @returns VBox strict status code.
3500 * @param pVCpu The cross context virtual CPU structure.
3501 * @param uVector The external interrupt vector (pass 0 if the interrupt
3502 * is still pending since we typically won't know the
3503 * vector).
3504 * @param fIntPending Whether the external interrupt is pending or
3505 * acknowledged in the interrupt controller.
3506 */
3507IEM_STATIC VBOXSTRICTRC iemVmxVmexitExtInt(PVMCPUCC pVCpu, uint8_t uVector, bool fIntPending)
3508{
3509 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
3510 Assert(pVmcs);
3511 Assert(!fIntPending || uVector == 0);
3512
3513 /* The VM-exit is subject to "External interrupt exiting" being set. */
3514 if (pVmcs->u32PinCtls & VMX_PIN_CTLS_EXT_INT_EXIT)
3515 {
3516 if (fIntPending)
3517 {
3518 /*
3519 * If the interrupt is pending and we don't need to acknowledge the
3520 * interrupt on VM-exit, cause the VM-exit immediately.
3521 *
3522 * See Intel spec 25.2 "Other Causes Of VM Exits".
3523 */
3524 if (!(pVmcs->u32ExitCtls & VMX_EXIT_CTLS_ACK_EXT_INT))
3525 return iemVmxVmexit(pVCpu, VMX_EXIT_EXT_INT, 0 /* u64ExitQual */);
3526
3527 /*
3528 * If the interrupt is pending and we -do- need to acknowledge the interrupt
3529 * on VM-exit, postpone VM-exit till after the interrupt controller has been
3530 * acknowledged that the interrupt has been consumed.
3531 */
3532 return VINF_VMX_INTERCEPT_NOT_ACTIVE;
3533 }
3534
3535 /*
3536 * If the interrupt is no longer pending (i.e. it has been acknowledged) and the
3537 * "External interrupt exiting" and "Acknowledge interrupt on VM-exit" controls are
3538 * all set, we cause the VM-exit now. We need to record the external interrupt that
3539 * just occurred in the VM-exit interruption information field.
3540 *
3541 * See Intel spec. 27.2.2 "Information for VM Exits Due to Vectored Events".
3542 */
3543 if (pVmcs->u32ExitCtls & VMX_EXIT_CTLS_ACK_EXT_INT)
3544 {
3545 bool const fNmiUnblocking = pVCpu->cpum.GstCtx.hwvirt.vmx.fNmiUnblockingIret;
3546 uint32_t const uExitIntInfo = RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_VECTOR, uVector)
3547 | RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_EXT_INT)
3548 | RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_NMI_UNBLOCK_IRET, fNmiUnblocking)
3549 | RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_VALID, 1);
3550 iemVmxVmcsSetExitIntInfo(pVCpu, uExitIntInfo);
3551 return iemVmxVmexit(pVCpu, VMX_EXIT_EXT_INT, 0 /* u64ExitQual */);
3552 }
3553 }
3554
3555 return VINF_VMX_INTERCEPT_NOT_ACTIVE;
3556}
3557
3558
3559/**
3560 * VMX VM-exit handler for VM-exits due to a double fault caused during delivery of
3561 * an event.
3562 *
3563 * @returns VBox strict status code.
3564 * @param pVCpu The cross context virtual CPU structure.
3565 */
3566IEM_STATIC VBOXSTRICTRC iemVmxVmexitEventDoubleFault(PVMCPUCC pVCpu)
3567{
3568 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
3569 Assert(pVmcs);
3570
3571 uint32_t const fXcptBitmap = pVmcs->u32XcptBitmap;
3572 if (fXcptBitmap & RT_BIT(X86_XCPT_DF))
3573 {
3574 /*
3575 * The NMI-unblocking due to IRET field need not be set for double faults.
3576 * See Intel spec. 31.7.1.2 "Resuming Guest Software After Handling An Exception".
3577 */
3578 uint32_t const uExitIntInfo = RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_VECTOR, X86_XCPT_DF)
3579 | RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_TYPE, VMX_EXIT_INT_INFO_TYPE_HW_XCPT)
3580 | RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_ERR_CODE_VALID, 1)
3581 | RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_NMI_UNBLOCK_IRET, 0)
3582 | RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_VALID, 1);
3583 iemVmxVmcsSetExitIntInfo(pVCpu, uExitIntInfo);
3584 return iemVmxVmexit(pVCpu, VMX_EXIT_XCPT_OR_NMI, 0 /* u64ExitQual */);
3585 }
3586
3587 return VINF_VMX_INTERCEPT_NOT_ACTIVE;
3588}
3589
3590
3591/**
3592 * VMX VM-exit handler for VM-exit due to delivery of an events.
3593 *
3594 * This is intended for VM-exit due to exceptions or NMIs where the caller provides
3595 * all the relevant VM-exit information.
3596 *
3597 * @returns VBox strict status code.
3598 * @param pVCpu The cross context virtual CPU structure.
3599 * @param pExitInfo Pointer to the VM-exit information.
3600 * @param pExitEventInfo Pointer to the VM-exit event information.
3601 */
3602IEM_STATIC VBOXSTRICTRC iemVmxVmexitEventWithInfo(PVMCPUCC pVCpu, PCVMXVEXITINFO pExitInfo, PCVMXVEXITEVENTINFO pExitEventInfo)
3603{
3604 Assert(pExitInfo);
3605 Assert(pExitEventInfo);
3606 Assert(pExitInfo->uReason == VMX_EXIT_XCPT_OR_NMI);
3607 Assert(VMX_EXIT_INT_INFO_IS_VALID(pExitEventInfo->uExitIntInfo));
3608
3609 iemVmxVmcsSetExitInstrLen(pVCpu, pExitInfo->cbInstr);
3610 iemVmxVmcsSetExitIntInfo(pVCpu, pExitEventInfo->uExitIntInfo);
3611 iemVmxVmcsSetExitIntErrCode(pVCpu, pExitEventInfo->uExitIntErrCode);
3612 iemVmxVmcsSetIdtVectoringInfo(pVCpu, pExitEventInfo->uIdtVectoringInfo);
3613 iemVmxVmcsSetIdtVectoringErrCode(pVCpu, pExitEventInfo->uIdtVectoringErrCode);
3614 return iemVmxVmexit(pVCpu, VMX_EXIT_XCPT_OR_NMI, pExitInfo->u64Qual);
3615}
3616
3617
3618/**
3619 * VMX VM-exit handler for VM-exits due to delivery of an event.
3620 *
3621 * @returns VBox strict status code.
3622 * @param pVCpu The cross context virtual CPU structure.
3623 * @param uVector The interrupt / exception vector.
3624 * @param fFlags The flags (see IEM_XCPT_FLAGS_XXX).
3625 * @param uErrCode The error code associated with the event.
3626 * @param uCr2 The CR2 value in case of a \#PF exception.
3627 * @param cbInstr The instruction length in bytes.
3628 */
3629IEM_STATIC VBOXSTRICTRC iemVmxVmexitEvent(PVMCPUCC pVCpu, uint8_t uVector, uint32_t fFlags, uint32_t uErrCode, uint64_t uCr2,
3630 uint8_t cbInstr)
3631{
3632 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
3633 Assert(pVmcs);
3634
3635 /*
3636 * If the event is being injected as part of VM-entry, it is -not- subject to event
3637 * intercepts in the nested-guest. However, secondary exceptions that occur during
3638 * injection of any event -are- subject to event interception.
3639 *
3640 * See Intel spec. 26.5.1.2 "VM Exits During Event Injection".
3641 */
3642 if (!CPUMIsGuestVmxInterceptEvents(&pVCpu->cpum.GstCtx))
3643 {
3644 /*
3645 * If the event is a virtual-NMI (which is an NMI being inject during VM-entry)
3646 * virtual-NMI blocking must be set in effect rather than physical NMI blocking.
3647 *
3648 * See Intel spec. 24.6.1 "Pin-Based VM-Execution Controls".
3649 */
3650 if ( uVector == X86_XCPT_NMI
3651 && (fFlags & IEM_XCPT_FLAGS_T_CPU_XCPT)
3652 && (pVmcs->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI))
3653 pVCpu->cpum.GstCtx.hwvirt.vmx.fVirtNmiBlocking = true;
3654 else
3655 Assert(!pVCpu->cpum.GstCtx.hwvirt.vmx.fVirtNmiBlocking);
3656
3657 CPUMSetGuestVmxInterceptEvents(&pVCpu->cpum.GstCtx, true);
3658 return VINF_VMX_INTERCEPT_NOT_ACTIVE;
3659 }
3660
3661 /*
3662 * We are injecting an external interrupt, check if we need to cause a VM-exit now.
3663 * If not, the caller will continue delivery of the external interrupt as it would
3664 * normally. The interrupt is no longer pending in the interrupt controller at this
3665 * point.
3666 */
3667 if (fFlags & IEM_XCPT_FLAGS_T_EXT_INT)
3668 {
3669 Assert(!VMX_IDT_VECTORING_INFO_IS_VALID(pVmcs->u32RoIdtVectoringInfo));
3670 return iemVmxVmexitExtInt(pVCpu, uVector, false /* fIntPending */);
3671 }
3672
3673 /*
3674 * Evaluate intercepts for hardware exceptions, software exceptions (#BP, #OF),
3675 * and privileged software exceptions (#DB generated by INT1/ICEBP) and software
3676 * interrupts.
3677 */
3678 Assert(fFlags & (IEM_XCPT_FLAGS_T_CPU_XCPT | IEM_XCPT_FLAGS_T_SOFT_INT));
3679 bool fIntercept;
3680 if ( !(fFlags & IEM_XCPT_FLAGS_T_SOFT_INT)
3681 || (fFlags & (IEM_XCPT_FLAGS_BP_INSTR | IEM_XCPT_FLAGS_OF_INSTR | IEM_XCPT_FLAGS_ICEBP_INSTR)))
3682 {
3683 fIntercept = CPUMIsGuestVmxXcptInterceptSet(&pVCpu->cpum.GstCtx, uVector, uErrCode);
3684 }
3685 else
3686 {
3687 /* Software interrupts cannot be intercepted and therefore do not cause a VM-exit. */
3688 fIntercept = false;
3689 }
3690
3691 /*
3692 * Now that we've determined whether the event causes a VM-exit, we need to construct the
3693 * relevant VM-exit information and cause the VM-exit.
3694 */
3695 if (fIntercept)
3696 {
3697 Assert(!(fFlags & IEM_XCPT_FLAGS_T_EXT_INT));
3698
3699 /* Construct the rest of the event related information fields and cause the VM-exit. */
3700 uint64_t u64ExitQual;
3701 if (uVector == X86_XCPT_PF)
3702 {
3703 Assert(fFlags & IEM_XCPT_FLAGS_CR2);
3704 u64ExitQual = uCr2;
3705 }
3706 else if (uVector == X86_XCPT_DB)
3707 {
3708 IEM_CTX_IMPORT_RET(pVCpu, CPUMCTX_EXTRN_DR6);
3709 u64ExitQual = pVCpu->cpum.GstCtx.dr[6] & VMX_VMCS_EXIT_QUAL_VALID_MASK;
3710 }
3711 else
3712 u64ExitQual = 0;
3713
3714 uint8_t const fNmiUnblocking = pVCpu->cpum.GstCtx.hwvirt.vmx.fNmiUnblockingIret;
3715 bool const fErrCodeValid = RT_BOOL(fFlags & IEM_XCPT_FLAGS_ERR);
3716 uint8_t const uIntInfoType = iemVmxGetEventType(uVector, fFlags);
3717 uint32_t const uExitIntInfo = RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_VECTOR, uVector)
3718 | RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_TYPE, uIntInfoType)
3719 | RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_ERR_CODE_VALID, fErrCodeValid)
3720 | RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_NMI_UNBLOCK_IRET, fNmiUnblocking)
3721 | RT_BF_MAKE(VMX_BF_EXIT_INT_INFO_VALID, 1);
3722 iemVmxVmcsSetExitIntInfo(pVCpu, uExitIntInfo);
3723 iemVmxVmcsSetExitIntErrCode(pVCpu, uErrCode);
3724
3725 /*
3726 * For VM-exits due to software exceptions (those generated by INT3 or INTO) or privileged
3727 * software exceptions (those generated by INT1/ICEBP) we need to supply the VM-exit instruction
3728 * length.
3729 */
3730 if ( (fFlags & IEM_XCPT_FLAGS_T_SOFT_INT)
3731 || (fFlags & (IEM_XCPT_FLAGS_BP_INSTR | IEM_XCPT_FLAGS_OF_INSTR | IEM_XCPT_FLAGS_ICEBP_INSTR)))
3732 iemVmxVmcsSetExitInstrLen(pVCpu, cbInstr);
3733 else
3734 iemVmxVmcsSetExitInstrLen(pVCpu, 0);
3735
3736 return iemVmxVmexit(pVCpu, VMX_EXIT_XCPT_OR_NMI, u64ExitQual);
3737 }
3738
3739 return VINF_VMX_INTERCEPT_NOT_ACTIVE;
3740}
3741
3742
3743/**
3744 * VMX VM-exit handler for APIC accesses.
3745 *
3746 * @param pVCpu The cross context virtual CPU structure.
3747 * @param offAccess The offset of the register being accessed.
3748 * @param fAccess The type of access (must contain IEM_ACCESS_TYPE_READ or
3749 * IEM_ACCESS_TYPE_WRITE or IEM_ACCESS_INSTRUCTION).
3750 */
3751IEM_STATIC VBOXSTRICTRC iemVmxVmexitApicAccess(PVMCPUCC pVCpu, uint16_t offAccess, uint32_t fAccess)
3752{
3753 Assert((fAccess & IEM_ACCESS_TYPE_READ) || (fAccess & IEM_ACCESS_TYPE_WRITE) || (fAccess & IEM_ACCESS_INSTRUCTION));
3754
3755 VMXAPICACCESS enmAccess;
3756 bool const fInEventDelivery = IEMGetCurrentXcpt(pVCpu, NULL, NULL, NULL, NULL);
3757 if (fInEventDelivery)
3758 enmAccess = VMXAPICACCESS_LINEAR_EVENT_DELIVERY;
3759 else if (fAccess & IEM_ACCESS_INSTRUCTION)
3760 enmAccess = VMXAPICACCESS_LINEAR_INSTR_FETCH;
3761 else if (fAccess & IEM_ACCESS_TYPE_WRITE)
3762 enmAccess = VMXAPICACCESS_LINEAR_WRITE;
3763 else
3764 enmAccess = VMXAPICACCESS_LINEAR_READ;
3765
3766 uint64_t const u64ExitQual = RT_BF_MAKE(VMX_BF_EXIT_QUAL_APIC_ACCESS_OFFSET, offAccess)
3767 | RT_BF_MAKE(VMX_BF_EXIT_QUAL_APIC_ACCESS_TYPE, enmAccess);
3768 return iemVmxVmexit(pVCpu, VMX_EXIT_APIC_ACCESS, u64ExitQual);
3769}
3770
3771
3772/**
3773 * VMX VM-exit handler for APIC accesses.
3774 *
3775 * This is intended for APIC accesses where the caller provides all the
3776 * relevant VM-exit information.
3777 *
3778 * @returns VBox strict status code.
3779 * @param pVCpu The cross context virtual CPU structure.
3780 * @param pExitInfo Pointer to the VM-exit information.
3781 * @param pExitEventInfo Pointer to the VM-exit event information.
3782 */
3783IEM_STATIC VBOXSTRICTRC iemVmxVmexitApicAccessWithInfo(PVMCPUCC pVCpu, PCVMXVEXITINFO pExitInfo,
3784 PCVMXVEXITEVENTINFO pExitEventInfo)
3785{
3786 /* VM-exit interruption information should not be valid for APIC-access VM-exits. */
3787 Assert(!VMX_EXIT_INT_INFO_IS_VALID(pExitEventInfo->uExitIntInfo));
3788 Assert(pExitInfo->uReason == VMX_EXIT_APIC_ACCESS);
3789 iemVmxVmcsSetExitIntInfo(pVCpu, 0);
3790 iemVmxVmcsSetExitIntErrCode(pVCpu, 0);
3791 iemVmxVmcsSetExitInstrLen(pVCpu, pExitInfo->cbInstr);
3792 iemVmxVmcsSetIdtVectoringInfo(pVCpu, pExitEventInfo->uIdtVectoringInfo);
3793 iemVmxVmcsSetIdtVectoringErrCode(pVCpu, pExitEventInfo->uIdtVectoringErrCode);
3794 return iemVmxVmexit(pVCpu, VMX_EXIT_APIC_ACCESS, pExitInfo->u64Qual);
3795}
3796
3797
3798/**
3799 * VMX VM-exit handler for APIC-write VM-exits.
3800 *
3801 * @param pVCpu The cross context virtual CPU structure.
3802 * @param offApic The write to the virtual-APIC page offset that caused this
3803 * VM-exit.
3804 */
3805IEM_STATIC VBOXSTRICTRC iemVmxVmexitApicWrite(PVMCPUCC pVCpu, uint16_t offApic)
3806{
3807 Assert(offApic < XAPIC_OFF_END + 4);
3808 /* Write only bits 11:0 of the APIC offset into the Exit qualification field. */
3809 offApic &= UINT16_C(0xfff);
3810 return iemVmxVmexit(pVCpu, VMX_EXIT_APIC_WRITE, offApic);
3811}
3812
3813
3814/**
3815 * Sets virtual-APIC write emulation as pending.
3816 *
3817 * @param pVCpu The cross context virtual CPU structure.
3818 * @param offApic The offset in the virtual-APIC page that was written.
3819 */
3820DECLINLINE(void) iemVmxVirtApicSetPendingWrite(PVMCPUCC pVCpu, uint16_t offApic)
3821{
3822 Assert(offApic < XAPIC_OFF_END + 4);
3823
3824 /*
3825 * Record the currently updated APIC offset, as we need this later for figuring
3826 * out whether to perform TPR, EOI or self-IPI virtualization as well as well
3827 * as for supplying the exit qualification when causing an APIC-write VM-exit.
3828 */
3829 pVCpu->cpum.GstCtx.hwvirt.vmx.offVirtApicWrite = offApic;
3830
3831 /*
3832 * Flag that we need to perform virtual-APIC write emulation (TPR/PPR/EOI/Self-IPI
3833 * virtualization or APIC-write emulation).
3834 */
3835 if (!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_VMX_APIC_WRITE))
3836 VMCPU_FF_SET(pVCpu, VMCPU_FF_VMX_APIC_WRITE);
3837}
3838
3839
3840/**
3841 * Clears any pending virtual-APIC write emulation.
3842 *
3843 * @returns The virtual-APIC offset that was written before clearing it.
3844 * @param pVCpu The cross context virtual CPU structure.
3845 */
3846DECLINLINE(uint16_t) iemVmxVirtApicClearPendingWrite(PVMCPUCC pVCpu)
3847{
3848 IEM_CTX_ASSERT(pVCpu, CPUMCTX_EXTRN_HWVIRT);
3849 uint8_t const offVirtApicWrite = pVCpu->cpum.GstCtx.hwvirt.vmx.offVirtApicWrite;
3850 pVCpu->cpum.GstCtx.hwvirt.vmx.offVirtApicWrite = 0;
3851 Assert(VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_VMX_APIC_WRITE));
3852 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_VMX_APIC_WRITE);
3853 return offVirtApicWrite;
3854}
3855
3856
3857/**
3858 * Reads a 32-bit register from the virtual-APIC page at the given offset.
3859 *
3860 * @returns The register from the virtual-APIC page.
3861 * @param pVCpu The cross context virtual CPU structure.
3862 * @param offReg The offset of the register being read.
3863 */
3864IEM_STATIC uint32_t iemVmxVirtApicReadRaw32(PVMCPUCC pVCpu, uint16_t offReg)
3865{
3866 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
3867 Assert(pVmcs);
3868 Assert(offReg <= VMX_V_VIRT_APIC_SIZE - sizeof(uint32_t));
3869
3870 uint32_t uReg;
3871 RTGCPHYS const GCPhysVirtApic = pVmcs->u64AddrVirtApic.u;
3872 int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), &uReg, GCPhysVirtApic + offReg, sizeof(uReg));
3873 if (RT_SUCCESS(rc))
3874 { /* likely */ }
3875 else
3876 {
3877 AssertMsgFailed(("Failed to read %u bytes at offset %#x of the virtual-APIC page at %#RGp\n", sizeof(uReg), offReg,
3878 GCPhysVirtApic));
3879 uReg = 0;
3880 }
3881 return uReg;
3882}
3883
3884
3885/**
3886 * Reads a 64-bit register from the virtual-APIC page at the given offset.
3887 *
3888 * @returns The register from the virtual-APIC page.
3889 * @param pVCpu The cross context virtual CPU structure.
3890 * @param offReg The offset of the register being read.
3891 */
3892IEM_STATIC uint64_t iemVmxVirtApicReadRaw64(PVMCPUCC pVCpu, uint16_t offReg)
3893{
3894 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
3895 Assert(pVmcs);
3896 Assert(offReg <= VMX_V_VIRT_APIC_SIZE - sizeof(uint64_t));
3897
3898 uint64_t uReg;
3899 RTGCPHYS const GCPhysVirtApic = pVmcs->u64AddrVirtApic.u;
3900 int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), &uReg, GCPhysVirtApic + offReg, sizeof(uReg));
3901 if (RT_SUCCESS(rc))
3902 { /* likely */ }
3903 else
3904 {
3905 AssertMsgFailed(("Failed to read %u bytes at offset %#x of the virtual-APIC page at %#RGp\n", sizeof(uReg), offReg,
3906 GCPhysVirtApic));
3907 uReg = 0;
3908 }
3909 return uReg;
3910}
3911
3912
3913/**
3914 * Writes a 32-bit register to the virtual-APIC page at the given offset.
3915 *
3916 * @param pVCpu The cross context virtual CPU structure.
3917 * @param offReg The offset of the register being written.
3918 * @param uReg The register value to write.
3919 */
3920IEM_STATIC void iemVmxVirtApicWriteRaw32(PVMCPUCC pVCpu, uint16_t offReg, uint32_t uReg)
3921{
3922 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
3923 Assert(pVmcs);
3924 Assert(offReg <= VMX_V_VIRT_APIC_SIZE - sizeof(uint32_t));
3925
3926 RTGCPHYS const GCPhysVirtApic = pVmcs->u64AddrVirtApic.u;
3927 int rc = PGMPhysSimpleWriteGCPhys(pVCpu->CTX_SUFF(pVM), GCPhysVirtApic + offReg, &uReg, sizeof(uReg));
3928 if (RT_SUCCESS(rc))
3929 { /* likely */ }
3930 else
3931 {
3932 AssertMsgFailed(("Failed to write %u bytes at offset %#x of the virtual-APIC page at %#RGp\n", sizeof(uReg), offReg,
3933 GCPhysVirtApic));
3934 }
3935}
3936
3937
3938/**
3939 * Writes a 64-bit register to the virtual-APIC page at the given offset.
3940 *
3941 * @param pVCpu The cross context virtual CPU structure.
3942 * @param offReg The offset of the register being written.
3943 * @param uReg The register value to write.
3944 */
3945IEM_STATIC void iemVmxVirtApicWriteRaw64(PVMCPUCC pVCpu, uint16_t offReg, uint64_t uReg)
3946{
3947 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
3948 Assert(pVmcs);
3949 Assert(offReg <= VMX_V_VIRT_APIC_SIZE - sizeof(uint64_t));
3950
3951 RTGCPHYS const GCPhysVirtApic = pVmcs->u64AddrVirtApic.u;
3952 int rc = PGMPhysSimpleWriteGCPhys(pVCpu->CTX_SUFF(pVM), GCPhysVirtApic + offReg, &uReg, sizeof(uReg));
3953 if (RT_SUCCESS(rc))
3954 { /* likely */ }
3955 else
3956 {
3957 AssertMsgFailed(("Failed to write %u bytes at offset %#x of the virtual-APIC page at %#RGp\n", sizeof(uReg), offReg,
3958 GCPhysVirtApic));
3959 }
3960}
3961
3962
3963/**
3964 * Sets the vector in a virtual-APIC 256-bit sparse register.
3965 *
3966 * @param pVCpu The cross context virtual CPU structure.
3967 * @param offReg The offset of the 256-bit spare register.
3968 * @param uVector The vector to set.
3969 *
3970 * @remarks This is based on our APIC device code.
3971 */
3972IEM_STATIC void iemVmxVirtApicSetVectorInReg(PVMCPUCC pVCpu, uint16_t offReg, uint8_t uVector)
3973{
3974 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
3975 Assert(pVmcs);
3976
3977 /* Determine the vector offset within the chunk. */
3978 uint16_t const offVector = (uVector & UINT32_C(0xe0)) >> 1;
3979
3980 /* Read the chunk at the offset. */
3981 uint32_t uReg;
3982 RTGCPHYS const GCPhysVirtApic = pVmcs->u64AddrVirtApic.u;
3983 int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), &uReg, GCPhysVirtApic + offReg + offVector, sizeof(uReg));
3984 if (RT_SUCCESS(rc))
3985 {
3986 /* Modify the chunk. */
3987 uint16_t const idxVectorBit = uVector & UINT32_C(0x1f);
3988 uReg |= RT_BIT(idxVectorBit);
3989
3990 /* Write the chunk. */
3991 rc = PGMPhysSimpleWriteGCPhys(pVCpu->CTX_SUFF(pVM), GCPhysVirtApic + offReg + offVector, &uReg, sizeof(uReg));
3992 if (RT_SUCCESS(rc))
3993 { /* likely */ }
3994 else
3995 {
3996 AssertMsgFailed(("Failed to set vector %#x in 256-bit register at %#x of the virtual-APIC page at %#RGp\n",
3997 uVector, offReg, GCPhysVirtApic));
3998 }
3999 }
4000 else
4001 {
4002 AssertMsgFailed(("Failed to get vector %#x in 256-bit register at %#x of the virtual-APIC page at %#RGp\n",
4003 uVector, offReg, GCPhysVirtApic));
4004 }
4005}
4006
4007
4008/**
4009 * Clears the vector in a virtual-APIC 256-bit sparse register.
4010 *
4011 * @param pVCpu The cross context virtual CPU structure.
4012 * @param offReg The offset of the 256-bit spare register.
4013 * @param uVector The vector to clear.
4014 *
4015 * @remarks This is based on our APIC device code.
4016 */
4017IEM_STATIC void iemVmxVirtApicClearVectorInReg(PVMCPUCC pVCpu, uint16_t offReg, uint8_t uVector)
4018{
4019 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
4020 Assert(pVmcs);
4021
4022 /* Determine the vector offset within the chunk. */
4023 uint16_t const offVector = (uVector & UINT32_C(0xe0)) >> 1;
4024
4025 /* Read the chunk at the offset. */
4026 uint32_t uReg;
4027 RTGCPHYS const GCPhysVirtApic = pVmcs->u64AddrVirtApic.u;
4028 int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), &uReg, GCPhysVirtApic + offReg + offVector, sizeof(uReg));
4029 if (RT_SUCCESS(rc))
4030 {
4031 /* Modify the chunk. */
4032 uint16_t const idxVectorBit = uVector & UINT32_C(0x1f);
4033 uReg &= ~RT_BIT(idxVectorBit);
4034
4035 /* Write the chunk. */
4036 rc = PGMPhysSimpleWriteGCPhys(pVCpu->CTX_SUFF(pVM), GCPhysVirtApic + offReg + offVector, &uReg, sizeof(uReg));
4037 if (RT_SUCCESS(rc))
4038 { /* likely */ }
4039 else
4040 {
4041 AssertMsgFailed(("Failed to clear vector %#x in 256-bit register at %#x of the virtual-APIC page at %#RGp\n",
4042 uVector, offReg, GCPhysVirtApic));
4043 }
4044 }
4045 else
4046 {
4047 AssertMsgFailed(("Failed to get vector %#x in 256-bit register at %#x of the virtual-APIC page at %#RGp\n",
4048 uVector, offReg, GCPhysVirtApic));
4049 }
4050}
4051
4052
4053/**
4054 * Checks if a memory access to the APIC-access page must causes an APIC-access
4055 * VM-exit.
4056 *
4057 * @param pVCpu The cross context virtual CPU structure.
4058 * @param offAccess The offset of the register being accessed.
4059 * @param cbAccess The size of the access in bytes.
4060 * @param fAccess The type of access (must be IEM_ACCESS_TYPE_READ or
4061 * IEM_ACCESS_TYPE_WRITE).
4062 *
4063 * @remarks This must not be used for MSR-based APIC-access page accesses!
4064 * @sa iemVmxVirtApicAccessMsrWrite, iemVmxVirtApicAccessMsrRead.
4065 */
4066IEM_STATIC bool iemVmxVirtApicIsMemAccessIntercepted(PVMCPUCC pVCpu, uint16_t offAccess, size_t cbAccess, uint32_t fAccess)
4067{
4068 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
4069 Assert(pVmcs);
4070 Assert(fAccess == IEM_ACCESS_TYPE_READ || fAccess == IEM_ACCESS_TYPE_WRITE);
4071
4072 /*
4073 * We must cause a VM-exit if any of the following are true:
4074 * - TPR shadowing isn't active.
4075 * - The access size exceeds 32-bits.
4076 * - The access is not contained within low 4 bytes of a 16-byte aligned offset.
4077 *
4078 * See Intel spec. 29.4.2 "Virtualizing Reads from the APIC-Access Page".
4079 * See Intel spec. 29.4.3.1 "Determining Whether a Write Access is Virtualized".
4080 */
4081 if ( !(pVmcs->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW)
4082 || cbAccess > sizeof(uint32_t)
4083 || ((offAccess + cbAccess - 1) & 0xc)
4084 || offAccess >= XAPIC_OFF_END + 4)
4085 return true;
4086
4087 /*
4088 * If the access is part of an operation where we have already
4089 * virtualized a virtual-APIC write, we must cause a VM-exit.
4090 */
4091 if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_VMX_APIC_WRITE))
4092 return true;
4093
4094 /*
4095 * Check write accesses to the APIC-access page that cause VM-exits.
4096 */
4097 if (fAccess & IEM_ACCESS_TYPE_WRITE)
4098 {
4099 if (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_APIC_REG_VIRT)
4100 {
4101 /*
4102 * With APIC-register virtualization, a write access to any of the
4103 * following registers are virtualized. Accessing any other register
4104 * causes a VM-exit.
4105 */
4106 uint16_t const offAlignedAccess = offAccess & 0xfffc;
4107 switch (offAlignedAccess)
4108 {
4109 case XAPIC_OFF_ID:
4110 case XAPIC_OFF_TPR:
4111 case XAPIC_OFF_EOI:
4112 case XAPIC_OFF_LDR:
4113 case XAPIC_OFF_DFR:
4114 case XAPIC_OFF_SVR:
4115 case XAPIC_OFF_ESR:
4116 case XAPIC_OFF_ICR_LO:
4117 case XAPIC_OFF_ICR_HI:
4118 case XAPIC_OFF_LVT_TIMER:
4119 case XAPIC_OFF_LVT_THERMAL:
4120 case XAPIC_OFF_LVT_PERF:
4121 case XAPIC_OFF_LVT_LINT0:
4122 case XAPIC_OFF_LVT_LINT1:
4123 case XAPIC_OFF_LVT_ERROR:
4124 case XAPIC_OFF_TIMER_ICR:
4125 case XAPIC_OFF_TIMER_DCR:
4126 break;
4127 default:
4128 return true;
4129 }
4130 }
4131 else if (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_INT_DELIVERY)
4132 {
4133 /*
4134 * With virtual-interrupt delivery, a write access to any of the
4135 * following registers are virtualized. Accessing any other register
4136 * causes a VM-exit.
4137 *
4138 * Note! The specification does not allow writing to offsets in-between
4139 * these registers (e.g. TPR + 1 byte) unlike read accesses.
4140 */
4141 switch (offAccess)
4142 {
4143 case XAPIC_OFF_TPR:
4144 case XAPIC_OFF_EOI:
4145 case XAPIC_OFF_ICR_LO:
4146 break;
4147 default:
4148 return true;
4149 }
4150 }
4151 else
4152 {
4153 /*
4154 * Without APIC-register virtualization or virtual-interrupt delivery,
4155 * only TPR accesses are virtualized.
4156 */
4157 if (offAccess == XAPIC_OFF_TPR)
4158 { /* likely */ }
4159 else
4160 return true;
4161 }
4162 }
4163 else
4164 {
4165 /*
4166 * Check read accesses to the APIC-access page that cause VM-exits.
4167 */
4168 if (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_APIC_REG_VIRT)
4169 {
4170 /*
4171 * With APIC-register virtualization, a read access to any of the
4172 * following registers are virtualized. Accessing any other register
4173 * causes a VM-exit.
4174 */
4175 uint16_t const offAlignedAccess = offAccess & 0xfffc;
4176 switch (offAlignedAccess)
4177 {
4178 /** @todo r=ramshankar: What about XAPIC_OFF_LVT_CMCI? */
4179 case XAPIC_OFF_ID:
4180 case XAPIC_OFF_VERSION:
4181 case XAPIC_OFF_TPR:
4182 case XAPIC_OFF_EOI:
4183 case XAPIC_OFF_LDR:
4184 case XAPIC_OFF_DFR:
4185 case XAPIC_OFF_SVR:
4186 case XAPIC_OFF_ISR0: case XAPIC_OFF_ISR1: case XAPIC_OFF_ISR2: case XAPIC_OFF_ISR3:
4187 case XAPIC_OFF_ISR4: case XAPIC_OFF_ISR5: case XAPIC_OFF_ISR6: case XAPIC_OFF_ISR7:
4188 case XAPIC_OFF_TMR0: case XAPIC_OFF_TMR1: case XAPIC_OFF_TMR2: case XAPIC_OFF_TMR3:
4189 case XAPIC_OFF_TMR4: case XAPIC_OFF_TMR5: case XAPIC_OFF_TMR6: case XAPIC_OFF_TMR7:
4190 case XAPIC_OFF_IRR0: case XAPIC_OFF_IRR1: case XAPIC_OFF_IRR2: case XAPIC_OFF_IRR3:
4191 case XAPIC_OFF_IRR4: case XAPIC_OFF_IRR5: case XAPIC_OFF_IRR6: case XAPIC_OFF_IRR7:
4192 case XAPIC_OFF_ESR:
4193 case XAPIC_OFF_ICR_LO:
4194 case XAPIC_OFF_ICR_HI:
4195 case XAPIC_OFF_LVT_TIMER:
4196 case XAPIC_OFF_LVT_THERMAL:
4197 case XAPIC_OFF_LVT_PERF:
4198 case XAPIC_OFF_LVT_LINT0:
4199 case XAPIC_OFF_LVT_LINT1:
4200 case XAPIC_OFF_LVT_ERROR:
4201 case XAPIC_OFF_TIMER_ICR:
4202 case XAPIC_OFF_TIMER_DCR:
4203 break;
4204 default:
4205 return true;
4206 }
4207 }
4208 else
4209 {
4210 /* Without APIC-register virtualization, only TPR accesses are virtualized. */
4211 if (offAccess == XAPIC_OFF_TPR)
4212 { /* likely */ }
4213 else
4214 return true;
4215 }
4216 }
4217
4218 /* The APIC access is virtualized, does not cause a VM-exit. */
4219 return false;
4220}
4221
4222
4223/**
4224 * Virtualizes a memory-based APIC access where the address is not used to access
4225 * memory.
4226 *
4227 * This is for instructions like MONITOR, CLFLUSH, CLFLUSHOPT, ENTER which may cause
4228 * page-faults but do not use the address to access memory.
4229 *
4230 * @param pVCpu The cross context virtual CPU structure.
4231 * @param pGCPhysAccess Pointer to the guest-physical address used.
4232 */
4233IEM_STATIC VBOXSTRICTRC iemVmxVirtApicAccessUnused(PVMCPUCC pVCpu, PRTGCPHYS pGCPhysAccess)
4234{
4235 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
4236 Assert(pVmcs);
4237 Assert(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS);
4238 Assert(pGCPhysAccess);
4239
4240 RTGCPHYS const GCPhysAccess = *pGCPhysAccess & ~(RTGCPHYS)PAGE_OFFSET_MASK;
4241 RTGCPHYS const GCPhysApic = pVmcs->u64AddrApicAccess.u;
4242 Assert(!(GCPhysApic & PAGE_OFFSET_MASK));
4243
4244 if (GCPhysAccess == GCPhysApic)
4245 {
4246 uint16_t const offAccess = *pGCPhysAccess & PAGE_OFFSET_MASK;
4247 uint32_t const fAccess = IEM_ACCESS_TYPE_READ;
4248 uint16_t const cbAccess = 1;
4249 bool const fIntercept = iemVmxVirtApicIsMemAccessIntercepted(pVCpu, offAccess, cbAccess, fAccess);
4250 if (fIntercept)
4251 return iemVmxVmexitApicAccess(pVCpu, offAccess, fAccess);
4252
4253 *pGCPhysAccess = GCPhysApic | offAccess;
4254 return VINF_VMX_MODIFIES_BEHAVIOR;
4255 }
4256
4257 return VINF_VMX_INTERCEPT_NOT_ACTIVE;
4258}
4259
4260
4261/**
4262 * Virtualizes a memory-based APIC access.
4263 *
4264 * @returns VBox strict status code.
4265 * @retval VINF_VMX_MODIFIES_BEHAVIOR if the access was virtualized.
4266 * @retval VINF_VMX_VMEXIT if the access causes a VM-exit.
4267 *
4268 * @param pVCpu The cross context virtual CPU structure.
4269 * @param offAccess The offset of the register being accessed (within the
4270 * APIC-access page).
4271 * @param cbAccess The size of the access in bytes.
4272 * @param pvData Pointer to the data being written or where to store the data
4273 * being read.
4274 * @param fAccess The type of access (must contain IEM_ACCESS_TYPE_READ or
4275 * IEM_ACCESS_TYPE_WRITE or IEM_ACCESS_INSTRUCTION).
4276 */
4277IEM_STATIC VBOXSTRICTRC iemVmxVirtApicAccessMem(PVMCPUCC pVCpu, uint16_t offAccess, size_t cbAccess, void *pvData,
4278 uint32_t fAccess)
4279{
4280 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
4281 Assert(pVmcs);
4282 Assert(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS); NOREF(pVmcs);
4283 Assert(pvData);
4284 Assert( (fAccess & IEM_ACCESS_TYPE_READ)
4285 || (fAccess & IEM_ACCESS_TYPE_WRITE)
4286 || (fAccess & IEM_ACCESS_INSTRUCTION));
4287
4288 bool const fIntercept = iemVmxVirtApicIsMemAccessIntercepted(pVCpu, offAccess, cbAccess, fAccess);
4289 if (fIntercept)
4290 return iemVmxVmexitApicAccess(pVCpu, offAccess, fAccess);
4291
4292 if (fAccess & IEM_ACCESS_TYPE_WRITE)
4293 {
4294 /*
4295 * A write access to the APIC-access page that is virtualized (rather than
4296 * causing a VM-exit) writes data to the virtual-APIC page.
4297 */
4298 uint32_t const u32Data = *(uint32_t *)pvData;
4299 iemVmxVirtApicWriteRaw32(pVCpu, offAccess, u32Data);
4300
4301 /*
4302 * Record the currently updated APIC offset, as we need this later for figuring
4303 * out whether to perform TPR, EOI or self-IPI virtualization as well as well
4304 * as for supplying the exit qualification when causing an APIC-write VM-exit.
4305 *
4306 * After completion of the current operation, we need to perform TPR virtualization,
4307 * EOI virtualization or APIC-write VM-exit depending on which register was written.
4308 *
4309 * The current operation may be a REP-prefixed string instruction, execution of any
4310 * other instruction, or delivery of an event through the IDT.
4311 *
4312 * Thus things like clearing bytes 3:1 of the VTPR, clearing VEOI are not to be
4313 * performed now but later after completion of the current operation.
4314 *
4315 * See Intel spec. 29.4.3.2 "APIC-Write Emulation".
4316 */
4317 iemVmxVirtApicSetPendingWrite(pVCpu, offAccess);
4318 }
4319 else
4320 {
4321 /*
4322 * A read access from the APIC-access page that is virtualized (rather than
4323 * causing a VM-exit) returns data from the virtual-APIC page.
4324 *
4325 * See Intel spec. 29.4.2 "Virtualizing Reads from the APIC-Access Page".
4326 */
4327 Assert(cbAccess <= 4);
4328 Assert(offAccess < XAPIC_OFF_END + 4);
4329 static uint32_t const s_auAccessSizeMasks[] = { 0, 0xff, 0xffff, 0xffffff, 0xffffffff };
4330
4331 uint32_t u32Data = iemVmxVirtApicReadRaw32(pVCpu, offAccess);
4332 u32Data &= s_auAccessSizeMasks[cbAccess];
4333 *(uint32_t *)pvData = u32Data;
4334 }
4335
4336 return VINF_VMX_MODIFIES_BEHAVIOR;
4337}
4338
4339
4340/**
4341 * Virtualizes an MSR-based APIC read access.
4342 *
4343 * @returns VBox strict status code.
4344 * @retval VINF_VMX_MODIFIES_BEHAVIOR if the MSR read was virtualized.
4345 * @retval VINF_VMX_INTERCEPT_NOT_ACTIVE if the MSR read access must be
4346 * handled by the x2APIC device.
4347 * @retval VERR_OUT_RANGE if the MSR read was supposed to be virtualized but was
4348 * not within the range of valid MSRs, caller must raise \#GP(0).
4349 * @param pVCpu The cross context virtual CPU structure.
4350 * @param idMsr The x2APIC MSR being read.
4351 * @param pu64Value Where to store the read x2APIC MSR value (only valid when
4352 * VINF_VMX_MODIFIES_BEHAVIOR is returned).
4353 */
4354IEM_STATIC VBOXSTRICTRC iemVmxVirtApicAccessMsrRead(PVMCPUCC pVCpu, uint32_t idMsr, uint64_t *pu64Value)
4355{
4356 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
4357 Assert(pVmcs);
4358 Assert(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_X2APIC_MODE);
4359 Assert(pu64Value);
4360
4361 if (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_APIC_REG_VIRT)
4362 {
4363 /*
4364 * Intel has different ideas in the x2APIC spec. vs the VT-x spec. as to
4365 * what the end of the valid x2APIC MSR range is. Hence the use of different
4366 * macros here.
4367 *
4368 * See Intel spec. 10.12.1.2 "x2APIC Register Address Space".
4369 * See Intel spec. 29.5 "Virtualizing MSR-based APIC Accesses".
4370 */
4371 if ( idMsr >= VMX_V_VIRT_APIC_MSR_START
4372 && idMsr <= VMX_V_VIRT_APIC_MSR_END)
4373 {
4374 uint16_t const offReg = (idMsr & 0xff) << 4;
4375 uint64_t const u64Value = iemVmxVirtApicReadRaw64(pVCpu, offReg);
4376 *pu64Value = u64Value;
4377 return VINF_VMX_MODIFIES_BEHAVIOR;
4378 }
4379 return VERR_OUT_OF_RANGE;
4380 }
4381
4382 if (idMsr == MSR_IA32_X2APIC_TPR)
4383 {
4384 uint16_t const offReg = (idMsr & 0xff) << 4;
4385 uint64_t const u64Value = iemVmxVirtApicReadRaw64(pVCpu, offReg);
4386 *pu64Value = u64Value;
4387 return VINF_VMX_MODIFIES_BEHAVIOR;
4388 }
4389
4390 return VINF_VMX_INTERCEPT_NOT_ACTIVE;
4391}
4392
4393
4394/**
4395 * Virtualizes an MSR-based APIC write access.
4396 *
4397 * @returns VBox strict status code.
4398 * @retval VINF_VMX_MODIFIES_BEHAVIOR if the MSR write was virtualized.
4399 * @retval VERR_OUT_RANGE if the MSR read was supposed to be virtualized but was
4400 * not within the range of valid MSRs, caller must raise \#GP(0).
4401 * @retval VINF_VMX_INTERCEPT_NOT_ACTIVE if the MSR must be written normally.
4402 *
4403 * @param pVCpu The cross context virtual CPU structure.
4404 * @param idMsr The x2APIC MSR being written.
4405 * @param u64Value The value of the x2APIC MSR being written.
4406 */
4407IEM_STATIC VBOXSTRICTRC iemVmxVirtApicAccessMsrWrite(PVMCPUCC pVCpu, uint32_t idMsr, uint64_t u64Value)
4408{
4409 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
4410 Assert(pVmcs);
4411
4412 /*
4413 * Check if the access is to be virtualized.
4414 * See Intel spec. 29.5 "Virtualizing MSR-based APIC Accesses".
4415 */
4416 if ( idMsr == MSR_IA32_X2APIC_TPR
4417 || ( (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_INT_DELIVERY)
4418 && ( idMsr == MSR_IA32_X2APIC_EOI
4419 || idMsr == MSR_IA32_X2APIC_SELF_IPI)))
4420 {
4421 /* Validate the MSR write depending on the register. */
4422 switch (idMsr)
4423 {
4424 case MSR_IA32_X2APIC_TPR:
4425 case MSR_IA32_X2APIC_SELF_IPI:
4426 {
4427 if (u64Value & UINT64_C(0xffffffffffffff00))
4428 return VERR_OUT_OF_RANGE;
4429 break;
4430 }
4431 case MSR_IA32_X2APIC_EOI:
4432 {
4433 if (u64Value != 0)
4434 return VERR_OUT_OF_RANGE;
4435 break;
4436 }
4437 }
4438
4439 /* Write the MSR to the virtual-APIC page. */
4440 uint16_t const offReg = (idMsr & 0xff) << 4;
4441 iemVmxVirtApicWriteRaw64(pVCpu, offReg, u64Value);
4442
4443 /*
4444 * Record the currently updated APIC offset, as we need this later for figuring
4445 * out whether to perform TPR, EOI or self-IPI virtualization as well as well
4446 * as for supplying the exit qualification when causing an APIC-write VM-exit.
4447 */
4448 iemVmxVirtApicSetPendingWrite(pVCpu, offReg);
4449
4450 return VINF_VMX_MODIFIES_BEHAVIOR;
4451 }
4452
4453 return VINF_VMX_INTERCEPT_NOT_ACTIVE;
4454}
4455
4456
4457/**
4458 * Finds the most significant set bit in a virtual-APIC 256-bit sparse register.
4459 *
4460 * @returns VBox status code.
4461 * @retval VINF_SUCCESS when the highest set bit is found.
4462 * @retval VERR_NOT_FOUND when no bit is set.
4463 *
4464 * @param pVCpu The cross context virtual CPU structure.
4465 * @param offReg The offset of the APIC 256-bit sparse register.
4466 * @param pidxHighestBit Where to store the highest bit (most significant bit)
4467 * set in the register. Only valid when VINF_SUCCESS is
4468 * returned.
4469 *
4470 * @remarks The format of the 256-bit sparse register here mirrors that found in
4471 * real APIC hardware.
4472 */
4473static int iemVmxVirtApicGetHighestSetBitInReg(PVMCPUCC pVCpu, uint16_t offReg, uint8_t *pidxHighestBit)
4474{
4475 Assert(offReg < XAPIC_OFF_END + 4);
4476 Assert(pidxHighestBit);
4477 Assert(pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs));
4478
4479 /*
4480 * There are 8 contiguous fragments (of 16-bytes each) in the sparse register.
4481 * However, in each fragment only the first 4 bytes are used.
4482 */
4483 uint8_t const cFrags = 8;
4484 for (int8_t iFrag = cFrags; iFrag >= 0; iFrag--)
4485 {
4486 uint16_t const offFrag = iFrag * 16;
4487 uint32_t const u32Frag = iemVmxVirtApicReadRaw32(pVCpu, offReg + offFrag);
4488 if (!u32Frag)
4489 continue;
4490
4491 unsigned idxHighestBit = ASMBitLastSetU32(u32Frag);
4492 Assert(idxHighestBit > 0);
4493 --idxHighestBit;
4494 Assert(idxHighestBit <= UINT8_MAX);
4495 *pidxHighestBit = idxHighestBit;
4496 return VINF_SUCCESS;
4497 }
4498 return VERR_NOT_FOUND;
4499}
4500
4501
4502/**
4503 * Evaluates pending virtual interrupts.
4504 *
4505 * @param pVCpu The cross context virtual CPU structure.
4506 */
4507IEM_STATIC void iemVmxEvalPendingVirtIntrs(PVMCPUCC pVCpu)
4508{
4509 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
4510 Assert(pVmcs);
4511 Assert(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_INT_DELIVERY);
4512
4513 if (!(pVmcs->u32ProcCtls & VMX_PROC_CTLS_INT_WINDOW_EXIT))
4514 {
4515 uint8_t const uRvi = RT_LO_U8(pVmcs->u16GuestIntStatus);
4516 uint8_t const uPpr = iemVmxVirtApicReadRaw32(pVCpu, XAPIC_OFF_PPR);
4517
4518 if ((uRvi >> 4) > (uPpr >> 4))
4519 {
4520 Log2(("eval_virt_intrs: uRvi=%#x uPpr=%#x - Signalling pending interrupt\n", uRvi, uPpr));
4521 VMCPU_FF_SET(pVCpu, VMCPU_FF_INTERRUPT_NESTED_GUEST);
4522 }
4523 else
4524 Log2(("eval_virt_intrs: uRvi=%#x uPpr=%#x - Nothing to do\n", uRvi, uPpr));
4525 }
4526}
4527
4528
4529/**
4530 * Performs PPR virtualization.
4531 *
4532 * @returns VBox strict status code.
4533 * @param pVCpu The cross context virtual CPU structure.
4534 */
4535IEM_STATIC void iemVmxPprVirtualization(PVMCPUCC pVCpu)
4536{
4537 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
4538 Assert(pVmcs);
4539 Assert(pVmcs->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW);
4540 Assert(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_INT_DELIVERY);
4541
4542 /*
4543 * PPR virtualization is caused in response to a VM-entry, TPR-virtualization,
4544 * or EOI-virtualization.
4545 *
4546 * See Intel spec. 29.1.3 "PPR Virtualization".
4547 */
4548 uint32_t const uTpr = iemVmxVirtApicReadRaw32(pVCpu, XAPIC_OFF_TPR);
4549 uint32_t const uSvi = RT_HI_U8(pVmcs->u16GuestIntStatus);
4550
4551 uint32_t uPpr;
4552 if (((uTpr >> 4) & 0xf) >= ((uSvi >> 4) & 0xf))
4553 uPpr = uTpr & 0xff;
4554 else
4555 uPpr = uSvi & 0xf0;
4556
4557 Log2(("ppr_virt: uTpr=%#x uSvi=%#x uPpr=%#x\n", uTpr, uSvi, uPpr));
4558 iemVmxVirtApicWriteRaw32(pVCpu, XAPIC_OFF_PPR, uPpr);
4559}
4560
4561
4562/**
4563 * Performs VMX TPR virtualization.
4564 *
4565 * @returns VBox strict status code.
4566 * @param pVCpu The cross context virtual CPU structure.
4567 */
4568IEM_STATIC VBOXSTRICTRC iemVmxTprVirtualization(PVMCPUCC pVCpu)
4569{
4570 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
4571 Assert(pVmcs);
4572 Assert(pVmcs->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW);
4573
4574 /*
4575 * We should have already performed the virtual-APIC write to the TPR offset
4576 * in the virtual-APIC page. We now perform TPR virtualization.
4577 *
4578 * See Intel spec. 29.1.2 "TPR Virtualization".
4579 */
4580 if (!(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_INT_DELIVERY))
4581 {
4582 uint32_t const uTprThreshold = pVmcs->u32TprThreshold;
4583 uint32_t const uTpr = iemVmxVirtApicReadRaw32(pVCpu, XAPIC_OFF_TPR);
4584
4585 /*
4586 * If the VTPR falls below the TPR threshold, we must cause a VM-exit.
4587 * See Intel spec. 29.1.2 "TPR Virtualization".
4588 */
4589 if (((uTpr >> 4) & 0xf) < uTprThreshold)
4590 {
4591 Log2(("tpr_virt: uTpr=%u uTprThreshold=%u -> VM-exit\n", uTpr, uTprThreshold));
4592 return iemVmxVmexit(pVCpu, VMX_EXIT_TPR_BELOW_THRESHOLD, 0 /* u64ExitQual */);
4593 }
4594 }
4595 else
4596 {
4597 iemVmxPprVirtualization(pVCpu);
4598 iemVmxEvalPendingVirtIntrs(pVCpu);
4599 }
4600
4601 return VINF_SUCCESS;
4602}
4603
4604
4605/**
4606 * Checks whether an EOI write for the given interrupt vector causes a VM-exit or
4607 * not.
4608 *
4609 * @returns @c true if the EOI write is intercepted, @c false otherwise.
4610 * @param pVCpu The cross context virtual CPU structure.
4611 * @param uVector The interrupt that was acknowledged using an EOI.
4612 */
4613IEM_STATIC bool iemVmxIsEoiInterceptSet(PCVMCPU pVCpu, uint8_t uVector)
4614{
4615 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
4616 Assert(pVmcs);
4617 Assert(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_INT_DELIVERY);
4618
4619 if (uVector < 64)
4620 return RT_BOOL(pVmcs->u64EoiExitBitmap0.u & RT_BIT_64(uVector));
4621 if (uVector < 128)
4622 return RT_BOOL(pVmcs->u64EoiExitBitmap1.u & RT_BIT_64(uVector));
4623 if (uVector < 192)
4624 return RT_BOOL(pVmcs->u64EoiExitBitmap2.u & RT_BIT_64(uVector));
4625 return RT_BOOL(pVmcs->u64EoiExitBitmap3.u & RT_BIT_64(uVector));
4626}
4627
4628
4629/**
4630 * Performs EOI virtualization.
4631 *
4632 * @returns VBox strict status code.
4633 * @param pVCpu The cross context virtual CPU structure.
4634 */
4635IEM_STATIC VBOXSTRICTRC iemVmxEoiVirtualization(PVMCPUCC pVCpu)
4636{
4637 PVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
4638 Assert(pVmcs);
4639 Assert(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_INT_DELIVERY);
4640
4641 /*
4642 * Clear the interrupt guest-interrupt as no longer in-service (ISR)
4643 * and get the next guest-interrupt that's in-service (if any).
4644 *
4645 * See Intel spec. 29.1.4 "EOI Virtualization".
4646 */
4647 uint8_t const uRvi = RT_LO_U8(pVmcs->u16GuestIntStatus);
4648 uint8_t const uSvi = RT_HI_U8(pVmcs->u16GuestIntStatus);
4649 Log2(("eoi_virt: uRvi=%#x uSvi=%#x\n", uRvi, uSvi));
4650
4651 uint8_t uVector = uSvi;
4652 iemVmxVirtApicClearVectorInReg(pVCpu, XAPIC_OFF_ISR0, uVector);
4653
4654 uVector = 0;
4655 iemVmxVirtApicGetHighestSetBitInReg(pVCpu, XAPIC_OFF_ISR0, &uVector);
4656
4657 if (uVector)
4658 Log2(("eoi_virt: next interrupt %#x\n", uVector));
4659 else
4660 Log2(("eoi_virt: no interrupt pending in ISR\n"));
4661
4662 /* Update guest-interrupt status SVI (leave RVI portion as it is) in the VMCS. */
4663 pVmcs->u16GuestIntStatus = RT_MAKE_U16(uRvi, uVector);
4664
4665 iemVmxPprVirtualization(pVCpu);
4666 if (iemVmxIsEoiInterceptSet(pVCpu, uVector))
4667 return iemVmxVmexit(pVCpu, VMX_EXIT_VIRTUALIZED_EOI, uVector);
4668 iemVmxEvalPendingVirtIntrs(pVCpu);
4669 return VINF_SUCCESS;
4670}
4671
4672
4673/**
4674 * Performs self-IPI virtualization.
4675 *
4676 * @returns VBox strict status code.
4677 * @param pVCpu The cross context virtual CPU structure.
4678 */
4679IEM_STATIC VBOXSTRICTRC iemVmxSelfIpiVirtualization(PVMCPUCC pVCpu)
4680{
4681 PVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
4682 Assert(pVmcs);
4683 Assert(pVmcs->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW);
4684
4685 /*
4686 * We should have already performed the virtual-APIC write to the self-IPI offset
4687 * in the virtual-APIC page. We now perform self-IPI virtualization.
4688 *
4689 * See Intel spec. 29.1.5 "Self-IPI Virtualization".
4690 */
4691 uint8_t const uVector = iemVmxVirtApicReadRaw32(pVCpu, XAPIC_OFF_ICR_LO);
4692 Log2(("self_ipi_virt: uVector=%#x\n", uVector));
4693 iemVmxVirtApicSetVectorInReg(pVCpu, XAPIC_OFF_IRR0, uVector);
4694 uint8_t const uRvi = RT_LO_U8(pVmcs->u16GuestIntStatus);
4695 uint8_t const uSvi = RT_HI_U8(pVmcs->u16GuestIntStatus);
4696 if (uVector > uRvi)
4697 pVmcs->u16GuestIntStatus = RT_MAKE_U16(uVector, uSvi);
4698 iemVmxEvalPendingVirtIntrs(pVCpu);
4699 return VINF_SUCCESS;
4700}
4701
4702
4703/**
4704 * Performs VMX APIC-write emulation.
4705 *
4706 * @returns VBox strict status code.
4707 * @param pVCpu The cross context virtual CPU structure.
4708 */
4709IEM_STATIC VBOXSTRICTRC iemVmxApicWriteEmulation(PVMCPUCC pVCpu)
4710{
4711 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
4712 Assert(pVmcs);
4713
4714 /* Import the virtual-APIC write offset (part of the hardware-virtualization state). */
4715 IEM_CTX_IMPORT_RET(pVCpu, CPUMCTX_EXTRN_HWVIRT);
4716
4717 /*
4718 * Perform APIC-write emulation based on the virtual-APIC register written.
4719 * See Intel spec. 29.4.3.2 "APIC-Write Emulation".
4720 */
4721 uint16_t const offApicWrite = iemVmxVirtApicClearPendingWrite(pVCpu);
4722 VBOXSTRICTRC rcStrict;
4723 switch (offApicWrite)
4724 {
4725 case XAPIC_OFF_TPR:
4726 {
4727 /* Clear bytes 3:1 of the VTPR and perform TPR virtualization. */
4728 uint32_t uTpr = iemVmxVirtApicReadRaw32(pVCpu, XAPIC_OFF_TPR);
4729 uTpr &= UINT32_C(0x000000ff);
4730 iemVmxVirtApicWriteRaw32(pVCpu, XAPIC_OFF_TPR, uTpr);
4731 Log2(("iemVmxApicWriteEmulation: TPR write %#x\n", uTpr));
4732 rcStrict = iemVmxTprVirtualization(pVCpu);
4733 break;
4734 }
4735
4736 case XAPIC_OFF_EOI:
4737 {
4738 if (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_INT_DELIVERY)
4739 {
4740 /* Clear VEOI and perform EOI virtualization. */
4741 iemVmxVirtApicWriteRaw32(pVCpu, XAPIC_OFF_EOI, 0);
4742 Log2(("iemVmxApicWriteEmulation: EOI write\n"));
4743 rcStrict = iemVmxEoiVirtualization(pVCpu);
4744 }
4745 else
4746 rcStrict = iemVmxVmexitApicWrite(pVCpu, offApicWrite);
4747 break;
4748 }
4749
4750 case XAPIC_OFF_ICR_LO:
4751 {
4752 if (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_INT_DELIVERY)
4753 {
4754 /* If the ICR_LO is valid, write it and perform self-IPI virtualization. */
4755 uint32_t const uIcrLo = iemVmxVirtApicReadRaw32(pVCpu, XAPIC_OFF_TPR);
4756 uint32_t const fIcrLoMb0 = UINT32_C(0xfffbb700);
4757 uint32_t const fIcrLoMb1 = UINT32_C(0x000000f0);
4758 if ( !(uIcrLo & fIcrLoMb0)
4759 && (uIcrLo & fIcrLoMb1))
4760 {
4761 Log2(("iemVmxApicWriteEmulation: Self-IPI virtualization with vector %#x\n", (uIcrLo & 0xff)));
4762 rcStrict = iemVmxSelfIpiVirtualization(pVCpu);
4763 }
4764 else
4765 rcStrict = iemVmxVmexitApicWrite(pVCpu, offApicWrite);
4766 }
4767 else
4768 rcStrict = iemVmxVmexitApicWrite(pVCpu, offApicWrite);
4769 break;
4770 }
4771
4772 case XAPIC_OFF_ICR_HI:
4773 {
4774 /* Clear bytes 2:0 of VICR_HI. No other virtualization or VM-exit must occur. */
4775 uint32_t uIcrHi = iemVmxVirtApicReadRaw32(pVCpu, XAPIC_OFF_ICR_HI);
4776 uIcrHi &= UINT32_C(0xff000000);
4777 iemVmxVirtApicWriteRaw32(pVCpu, XAPIC_OFF_ICR_HI, uIcrHi);
4778 rcStrict = VINF_SUCCESS;
4779 break;
4780 }
4781
4782 default:
4783 {
4784 /* Writes to any other virtual-APIC register causes an APIC-write VM-exit. */
4785 rcStrict = iemVmxVmexitApicWrite(pVCpu, offApicWrite);
4786 break;
4787 }
4788 }
4789
4790 return rcStrict;
4791}
4792
4793
4794/**
4795 * Checks guest control registers, debug registers and MSRs as part of VM-entry.
4796 *
4797 * @param pVCpu The cross context virtual CPU structure.
4798 * @param pszInstr The VMX instruction name (for logging purposes).
4799 */
4800DECLINLINE(int) iemVmxVmentryCheckGuestControlRegsMsrs(PVMCPUCC pVCpu, const char *pszInstr)
4801{
4802 /*
4803 * Guest Control Registers, Debug Registers, and MSRs.
4804 * See Intel spec. 26.3.1.1 "Checks on Guest Control Registers, Debug Registers, and MSRs".
4805 */
4806 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
4807 const char *const pszFailure = "VM-exit";
4808 bool const fUnrestrictedGuest = RT_BOOL(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_UNRESTRICTED_GUEST);
4809
4810 /* CR0 reserved bits. */
4811 {
4812 /* CR0 MB1 bits. */
4813 uint64_t u64Cr0Fixed0 = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Cr0Fixed0;
4814 Assert(!(u64Cr0Fixed0 & (X86_CR0_NW | X86_CR0_CD)));
4815 if (fUnrestrictedGuest)
4816 u64Cr0Fixed0 &= ~(X86_CR0_PE | X86_CR0_PG);
4817 if ((pVmcs->u64GuestCr0.u & u64Cr0Fixed0) == u64Cr0Fixed0)
4818 { /* likely */ }
4819 else
4820 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestCr0Fixed0);
4821
4822 /* CR0 MBZ bits. */
4823 uint64_t const u64Cr0Fixed1 = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Cr0Fixed1;
4824 if (!(pVmcs->u64GuestCr0.u & ~u64Cr0Fixed1))
4825 { /* likely */ }
4826 else
4827 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestCr0Fixed1);
4828
4829 /* Without unrestricted guest support, VT-x supports does not support unpaged protected mode. */
4830 if ( !fUnrestrictedGuest
4831 && (pVmcs->u64GuestCr0.u & X86_CR0_PG)
4832 && !(pVmcs->u64GuestCr0.u & X86_CR0_PE))
4833 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestCr0PgPe);
4834 }
4835
4836 /* CR4 reserved bits. */
4837 {
4838 /* CR4 MB1 bits. */
4839 uint64_t const u64Cr4Fixed0 = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Cr4Fixed0;
4840 if ((pVmcs->u64GuestCr4.u & u64Cr4Fixed0) == u64Cr4Fixed0)
4841 { /* likely */ }
4842 else
4843 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestCr4Fixed0);
4844
4845 /* CR4 MBZ bits. */
4846 uint64_t const u64Cr4Fixed1 = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Cr4Fixed1;
4847 if (!(pVmcs->u64GuestCr4.u & ~u64Cr4Fixed1))
4848 { /* likely */ }
4849 else
4850 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestCr4Fixed1);
4851 }
4852
4853 /* DEBUGCTL MSR. */
4854 if ( !(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_DEBUG)
4855 || !(pVmcs->u64GuestDebugCtlMsr.u & ~MSR_IA32_DEBUGCTL_VALID_MASK_INTEL))
4856 { /* likely */ }
4857 else
4858 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestDebugCtl);
4859
4860 /* 64-bit CPU checks. */
4861 bool const fGstInLongMode = RT_BOOL(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_IA32E_MODE_GUEST);
4862 if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode)
4863 {
4864 if (fGstInLongMode)
4865 {
4866 /* PAE must be set. */
4867 if ( (pVmcs->u64GuestCr0.u & X86_CR0_PG)
4868 && (pVmcs->u64GuestCr0.u & X86_CR4_PAE))
4869 { /* likely */ }
4870 else
4871 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestPae);
4872 }
4873 else
4874 {
4875 /* PCIDE should not be set. */
4876 if (!(pVmcs->u64GuestCr4.u & X86_CR4_PCIDE))
4877 { /* likely */ }
4878 else
4879 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestPcide);
4880 }
4881
4882 /* CR3. */
4883 if (!(pVmcs->u64GuestCr3.u >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cMaxPhysAddrWidth))
4884 { /* likely */ }
4885 else
4886 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestCr3);
4887
4888 /* DR7. */
4889 if ( !(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_DEBUG)
4890 || !(pVmcs->u64GuestDr7.u & X86_DR7_MBZ_MASK))
4891 { /* likely */ }
4892 else
4893 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestDr7);
4894
4895 /* SYSENTER ESP and SYSENTER EIP. */
4896 if ( X86_IS_CANONICAL(pVmcs->u64GuestSysenterEsp.u)
4897 && X86_IS_CANONICAL(pVmcs->u64GuestSysenterEip.u))
4898 { /* likely */ }
4899 else
4900 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSysenterEspEip);
4901 }
4902
4903 /* We don't support IA32_PERF_GLOBAL_CTRL MSR yet. */
4904 Assert(!(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_PERF_MSR));
4905
4906 /* PAT MSR. */
4907 if ( !(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_PAT_MSR)
4908 || CPUMIsPatMsrValid(pVmcs->u64GuestPatMsr.u))
4909 { /* likely */ }
4910 else
4911 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestPatMsr);
4912
4913 /* EFER MSR. */
4914 if (pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_EFER_MSR)
4915 {
4916 uint64_t const uValidEferMask = CPUMGetGuestEferMsrValidMask(pVCpu->CTX_SUFF(pVM));
4917 if (!(pVmcs->u64GuestEferMsr.u & ~uValidEferMask))
4918 { /* likely */ }
4919 else
4920 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestEferMsrRsvd);
4921
4922 bool const fGstLma = RT_BOOL(pVmcs->u64GuestEferMsr.u & MSR_K6_EFER_LMA);
4923 bool const fGstLme = RT_BOOL(pVmcs->u64GuestEferMsr.u & MSR_K6_EFER_LME);
4924 if ( fGstLma == fGstInLongMode
4925 && ( !(pVmcs->u64GuestCr0.u & X86_CR0_PG)
4926 || fGstLma == fGstLme))
4927 { /* likely */ }
4928 else
4929 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestEferMsr);
4930 }
4931
4932 /* We don't support IA32_BNDCFGS MSR yet. */
4933 Assert(!(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_BNDCFGS_MSR));
4934
4935 NOREF(pszInstr);
4936 NOREF(pszFailure);
4937 return VINF_SUCCESS;
4938}
4939
4940
4941/**
4942 * Checks guest segment registers, LDTR and TR as part of VM-entry.
4943 *
4944 * @param pVCpu The cross context virtual CPU structure.
4945 * @param pszInstr The VMX instruction name (for logging purposes).
4946 */
4947DECLINLINE(int) iemVmxVmentryCheckGuestSegRegs(PVMCPUCC pVCpu, const char *pszInstr)
4948{
4949 /*
4950 * Segment registers.
4951 * See Intel spec. 26.3.1.2 "Checks on Guest Segment Registers".
4952 */
4953 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
4954 const char *const pszFailure = "VM-exit";
4955 bool const fGstInV86Mode = RT_BOOL(pVmcs->u64GuestRFlags.u & X86_EFL_VM);
4956 bool const fUnrestrictedGuest = RT_BOOL(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_UNRESTRICTED_GUEST);
4957 bool const fGstInLongMode = RT_BOOL(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_IA32E_MODE_GUEST);
4958
4959 /* Selectors. */
4960 if ( !fGstInV86Mode
4961 && !fUnrestrictedGuest
4962 && (pVmcs->GuestSs & X86_SEL_RPL) != (pVmcs->GuestCs & X86_SEL_RPL))
4963 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegSelCsSsRpl);
4964
4965 for (unsigned iSegReg = 0; iSegReg < X86_SREG_COUNT; iSegReg++)
4966 {
4967 CPUMSELREG SelReg;
4968 int rc = iemVmxVmcsGetGuestSegReg(pVmcs, iSegReg, &SelReg);
4969 if (RT_LIKELY(rc == VINF_SUCCESS))
4970 { /* likely */ }
4971 else
4972 return rc;
4973
4974 /*
4975 * Virtual-8086 mode checks.
4976 */
4977 if (fGstInV86Mode)
4978 {
4979 /* Base address. */
4980 if (SelReg.u64Base == (uint64_t)SelReg.Sel << 4)
4981 { /* likely */ }
4982 else
4983 {
4984 VMXVDIAG const enmDiag = iemVmxGetDiagVmentrySegBaseV86(iSegReg);
4985 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, enmDiag);
4986 }
4987
4988 /* Limit. */
4989 if (SelReg.u32Limit == 0xffff)
4990 { /* likely */ }
4991 else
4992 {
4993 VMXVDIAG const enmDiag = iemVmxGetDiagVmentrySegLimitV86(iSegReg);
4994 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, enmDiag);
4995 }
4996
4997 /* Attribute. */
4998 if (SelReg.Attr.u == 0xf3)
4999 { /* likely */ }
5000 else
5001 {
5002 VMXVDIAG const enmDiag = iemVmxGetDiagVmentrySegAttrV86(iSegReg);
5003 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, enmDiag);
5004 }
5005
5006 /* We're done; move to checking the next segment. */
5007 continue;
5008 }
5009
5010 /* Checks done by 64-bit CPUs. */
5011 if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode)
5012 {
5013 /* Base address. */
5014 if ( iSegReg == X86_SREG_FS
5015 || iSegReg == X86_SREG_GS)
5016 {
5017 if (X86_IS_CANONICAL(SelReg.u64Base))
5018 { /* likely */ }
5019 else
5020 {
5021 VMXVDIAG const enmDiag = iemVmxGetDiagVmentrySegBase(iSegReg);
5022 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, enmDiag);
5023 }
5024 }
5025 else if (iSegReg == X86_SREG_CS)
5026 {
5027 if (!RT_HI_U32(SelReg.u64Base))
5028 { /* likely */ }
5029 else
5030 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegBaseCs);
5031 }
5032 else
5033 {
5034 if ( SelReg.Attr.n.u1Unusable
5035 || !RT_HI_U32(SelReg.u64Base))
5036 { /* likely */ }
5037 else
5038 {
5039 VMXVDIAG const enmDiag = iemVmxGetDiagVmentrySegBase(iSegReg);
5040 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, enmDiag);
5041 }
5042 }
5043 }
5044
5045 /*
5046 * Checks outside Virtual-8086 mode.
5047 */
5048 uint8_t const uSegType = SelReg.Attr.n.u4Type;
5049 uint8_t const fCodeDataSeg = SelReg.Attr.n.u1DescType;
5050 uint8_t const fUsable = !SelReg.Attr.n.u1Unusable;
5051 uint8_t const uDpl = SelReg.Attr.n.u2Dpl;
5052 uint8_t const fPresent = SelReg.Attr.n.u1Present;
5053 uint8_t const uGranularity = SelReg.Attr.n.u1Granularity;
5054 uint8_t const uDefBig = SelReg.Attr.n.u1DefBig;
5055 uint8_t const fSegLong = SelReg.Attr.n.u1Long;
5056
5057 /* Code or usable segment. */
5058 if ( iSegReg == X86_SREG_CS
5059 || fUsable)
5060 {
5061 /* Reserved bits (bits 31:17 and bits 11:8). */
5062 if (!(SelReg.Attr.u & 0xfffe0f00))
5063 { /* likely */ }
5064 else
5065 {
5066 VMXVDIAG const enmDiag = iemVmxGetDiagVmentrySegAttrRsvd(iSegReg);
5067 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, enmDiag);
5068 }
5069
5070 /* Descriptor type. */
5071 if (fCodeDataSeg)
5072 { /* likely */ }
5073 else
5074 {
5075 VMXVDIAG const enmDiag = iemVmxGetDiagVmentrySegAttrDescType(iSegReg);
5076 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, enmDiag);
5077 }
5078
5079 /* Present. */
5080 if (fPresent)
5081 { /* likely */ }
5082 else
5083 {
5084 VMXVDIAG const enmDiag = iemVmxGetDiagVmentrySegAttrPresent(iSegReg);
5085 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, enmDiag);
5086 }
5087
5088 /* Granularity. */
5089 if ( ((SelReg.u32Limit & 0x00000fff) == 0x00000fff || !uGranularity)
5090 && ((SelReg.u32Limit & 0xfff00000) == 0x00000000 || uGranularity))
5091 { /* likely */ }
5092 else
5093 {
5094 VMXVDIAG const enmDiag = iemVmxGetDiagVmentrySegAttrGran(iSegReg);
5095 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, enmDiag);
5096 }
5097 }
5098
5099 if (iSegReg == X86_SREG_CS)
5100 {
5101 /* Segment Type and DPL. */
5102 if ( uSegType == (X86_SEL_TYPE_RW | X86_SEL_TYPE_ACCESSED)
5103 && fUnrestrictedGuest)
5104 {
5105 if (uDpl == 0)
5106 { /* likely */ }
5107 else
5108 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrCsDplZero);
5109 }
5110 else if ( uSegType == (X86_SEL_TYPE_CODE | X86_SEL_TYPE_ACCESSED)
5111 || uSegType == (X86_SEL_TYPE_CODE | X86_SEL_TYPE_READ | X86_SEL_TYPE_ACCESSED))
5112 {
5113 X86DESCATTR AttrSs; AttrSs.u = pVmcs->u32GuestSsAttr;
5114 if (uDpl == AttrSs.n.u2Dpl)
5115 { /* likely */ }
5116 else
5117 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrCsDplEqSs);
5118 }
5119 else if ((uSegType & (X86_SEL_TYPE_CODE | X86_SEL_TYPE_CONF | X86_SEL_TYPE_ACCESSED))
5120 == (X86_SEL_TYPE_CODE | X86_SEL_TYPE_CONF | X86_SEL_TYPE_ACCESSED))
5121 {
5122 X86DESCATTR AttrSs; AttrSs.u = pVmcs->u32GuestSsAttr;
5123 if (uDpl <= AttrSs.n.u2Dpl)
5124 { /* likely */ }
5125 else
5126 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrCsDplLtSs);
5127 }
5128 else
5129 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrCsType);
5130
5131 /* Def/Big. */
5132 if ( fGstInLongMode
5133 && fSegLong)
5134 {
5135 if (uDefBig == 0)
5136 { /* likely */ }
5137 else
5138 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrCsDefBig);
5139 }
5140 }
5141 else if (iSegReg == X86_SREG_SS)
5142 {
5143 /* Segment Type. */
5144 if ( !fUsable
5145 || uSegType == (X86_SEL_TYPE_RW | X86_SEL_TYPE_ACCESSED)
5146 || uSegType == (X86_SEL_TYPE_DOWN | X86_SEL_TYPE_RW | X86_SEL_TYPE_ACCESSED))
5147 { /* likely */ }
5148 else
5149 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrSsType);
5150
5151 /* DPL. */
5152 if (!fUnrestrictedGuest)
5153 {
5154 if (uDpl == (SelReg.Sel & X86_SEL_RPL))
5155 { /* likely */ }
5156 else
5157 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrSsDplEqRpl);
5158 }
5159 X86DESCATTR AttrCs; AttrCs.u = pVmcs->u32GuestCsAttr;
5160 if ( AttrCs.n.u4Type == (X86_SEL_TYPE_RW | X86_SEL_TYPE_ACCESSED)
5161 || !(pVmcs->u64GuestCr0.u & X86_CR0_PE))
5162 {
5163 if (uDpl == 0)
5164 { /* likely */ }
5165 else
5166 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrSsDplZero);
5167 }
5168 }
5169 else
5170 {
5171 /* DS, ES, FS, GS. */
5172 if (fUsable)
5173 {
5174 /* Segment type. */
5175 if (uSegType & X86_SEL_TYPE_ACCESSED)
5176 { /* likely */ }
5177 else
5178 {
5179 VMXVDIAG const enmDiag = iemVmxGetDiagVmentrySegAttrTypeAcc(iSegReg);
5180 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, enmDiag);
5181 }
5182
5183 if ( !(uSegType & X86_SEL_TYPE_CODE)
5184 || (uSegType & X86_SEL_TYPE_READ))
5185 { /* likely */ }
5186 else
5187 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrCsTypeRead);
5188
5189 /* DPL. */
5190 if ( !fUnrestrictedGuest
5191 && uSegType <= (X86_SEL_TYPE_CODE | X86_SEL_TYPE_READ | X86_SEL_TYPE_ACCESSED))
5192 {
5193 if (uDpl >= (SelReg.Sel & X86_SEL_RPL))
5194 { /* likely */ }
5195 else
5196 {
5197 VMXVDIAG const enmDiag = iemVmxGetDiagVmentrySegAttrDplRpl(iSegReg);
5198 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, enmDiag);
5199 }
5200 }
5201 }
5202 }
5203 }
5204
5205 /*
5206 * LDTR.
5207 */
5208 {
5209 CPUMSELREG Ldtr;
5210 Ldtr.Sel = pVmcs->GuestLdtr;
5211 Ldtr.u32Limit = pVmcs->u32GuestLdtrLimit;
5212 Ldtr.u64Base = pVmcs->u64GuestLdtrBase.u;
5213 Ldtr.Attr.u = pVmcs->u32GuestLdtrAttr;
5214
5215 if (!Ldtr.Attr.n.u1Unusable)
5216 {
5217 /* Selector. */
5218 if (!(Ldtr.Sel & X86_SEL_LDT))
5219 { /* likely */ }
5220 else
5221 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegSelLdtr);
5222
5223 /* Base. */
5224 if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode)
5225 {
5226 if (X86_IS_CANONICAL(Ldtr.u64Base))
5227 { /* likely */ }
5228 else
5229 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegBaseLdtr);
5230 }
5231
5232 /* Attributes. */
5233 /* Reserved bits (bits 31:17 and bits 11:8). */
5234 if (!(Ldtr.Attr.u & 0xfffe0f00))
5235 { /* likely */ }
5236 else
5237 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrLdtrRsvd);
5238
5239 if (Ldtr.Attr.n.u4Type == X86_SEL_TYPE_SYS_LDT)
5240 { /* likely */ }
5241 else
5242 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrLdtrType);
5243
5244 if (!Ldtr.Attr.n.u1DescType)
5245 { /* likely */ }
5246 else
5247 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrLdtrDescType);
5248
5249 if (Ldtr.Attr.n.u1Present)
5250 { /* likely */ }
5251 else
5252 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrLdtrPresent);
5253
5254 if ( ((Ldtr.u32Limit & 0x00000fff) == 0x00000fff || !Ldtr.Attr.n.u1Granularity)
5255 && ((Ldtr.u32Limit & 0xfff00000) == 0x00000000 || Ldtr.Attr.n.u1Granularity))
5256 { /* likely */ }
5257 else
5258 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrLdtrGran);
5259 }
5260 }
5261
5262 /*
5263 * TR.
5264 */
5265 {
5266 CPUMSELREG Tr;
5267 Tr.Sel = pVmcs->GuestTr;
5268 Tr.u32Limit = pVmcs->u32GuestTrLimit;
5269 Tr.u64Base = pVmcs->u64GuestTrBase.u;
5270 Tr.Attr.u = pVmcs->u32GuestTrAttr;
5271
5272 /* Selector. */
5273 if (!(Tr.Sel & X86_SEL_LDT))
5274 { /* likely */ }
5275 else
5276 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegSelTr);
5277
5278 /* Base. */
5279 if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode)
5280 {
5281 if (X86_IS_CANONICAL(Tr.u64Base))
5282 { /* likely */ }
5283 else
5284 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegBaseTr);
5285 }
5286
5287 /* Attributes. */
5288 /* Reserved bits (bits 31:17 and bits 11:8). */
5289 if (!(Tr.Attr.u & 0xfffe0f00))
5290 { /* likely */ }
5291 else
5292 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrTrRsvd);
5293
5294 if (!Tr.Attr.n.u1Unusable)
5295 { /* likely */ }
5296 else
5297 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrTrUnusable);
5298
5299 if ( Tr.Attr.n.u4Type == X86_SEL_TYPE_SYS_386_TSS_BUSY
5300 || ( !fGstInLongMode
5301 && Tr.Attr.n.u4Type == X86_SEL_TYPE_SYS_286_TSS_BUSY))
5302 { /* likely */ }
5303 else
5304 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrTrType);
5305
5306 if (!Tr.Attr.n.u1DescType)
5307 { /* likely */ }
5308 else
5309 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrTrDescType);
5310
5311 if (Tr.Attr.n.u1Present)
5312 { /* likely */ }
5313 else
5314 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrTrPresent);
5315
5316 if ( ((Tr.u32Limit & 0x00000fff) == 0x00000fff || !Tr.Attr.n.u1Granularity)
5317 && ((Tr.u32Limit & 0xfff00000) == 0x00000000 || Tr.Attr.n.u1Granularity))
5318 { /* likely */ }
5319 else
5320 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestSegAttrTrGran);
5321 }
5322
5323 NOREF(pszInstr);
5324 NOREF(pszFailure);
5325 return VINF_SUCCESS;
5326}
5327
5328
5329/**
5330 * Checks guest GDTR and IDTR as part of VM-entry.
5331 *
5332 * @param pVCpu The cross context virtual CPU structure.
5333 * @param pszInstr The VMX instruction name (for logging purposes).
5334 */
5335DECLINLINE(int) iemVmxVmentryCheckGuestGdtrIdtr(PVMCPUCC pVCpu, const char *pszInstr)
5336{
5337 /*
5338 * GDTR and IDTR.
5339 * See Intel spec. 26.3.1.3 "Checks on Guest Descriptor-Table Registers".
5340 */
5341 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
5342 const char *const pszFailure = "VM-exit";
5343
5344 if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode)
5345 {
5346 /* Base. */
5347 if (X86_IS_CANONICAL(pVmcs->u64GuestGdtrBase.u))
5348 { /* likely */ }
5349 else
5350 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestGdtrBase);
5351
5352 if (X86_IS_CANONICAL(pVmcs->u64GuestIdtrBase.u))
5353 { /* likely */ }
5354 else
5355 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestIdtrBase);
5356 }
5357
5358 /* Limit. */
5359 if (!RT_HI_U16(pVmcs->u32GuestGdtrLimit))
5360 { /* likely */ }
5361 else
5362 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestGdtrLimit);
5363
5364 if (!RT_HI_U16(pVmcs->u32GuestIdtrLimit))
5365 { /* likely */ }
5366 else
5367 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestIdtrLimit);
5368
5369 NOREF(pszInstr);
5370 NOREF(pszFailure);
5371 return VINF_SUCCESS;
5372}
5373
5374
5375/**
5376 * Checks guest RIP and RFLAGS as part of VM-entry.
5377 *
5378 * @param pVCpu The cross context virtual CPU structure.
5379 * @param pszInstr The VMX instruction name (for logging purposes).
5380 */
5381DECLINLINE(int) iemVmxVmentryCheckGuestRipRFlags(PVMCPUCC pVCpu, const char *pszInstr)
5382{
5383 /*
5384 * RIP and RFLAGS.
5385 * See Intel spec. 26.3.1.4 "Checks on Guest RIP and RFLAGS".
5386 */
5387 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
5388 const char *const pszFailure = "VM-exit";
5389 bool const fGstInLongMode = RT_BOOL(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_IA32E_MODE_GUEST);
5390
5391 /* RIP. */
5392 if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode)
5393 {
5394 X86DESCATTR AttrCs;
5395 AttrCs.u = pVmcs->u32GuestCsAttr;
5396 if ( !fGstInLongMode
5397 || !AttrCs.n.u1Long)
5398 {
5399 if (!RT_HI_U32(pVmcs->u64GuestRip.u))
5400 { /* likely */ }
5401 else
5402 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestRipRsvd);
5403 }
5404
5405 if ( fGstInLongMode
5406 && AttrCs.n.u1Long)
5407 {
5408 Assert(IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cMaxLinearAddrWidth == 48); /* Canonical. */
5409 if ( IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cMaxLinearAddrWidth < 64
5410 && X86_IS_CANONICAL(pVmcs->u64GuestRip.u))
5411 { /* likely */ }
5412 else
5413 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestRip);
5414 }
5415 }
5416
5417 /* RFLAGS (bits 63:22 (or 31:22), bits 15, 5, 3 are reserved, bit 1 MB1). */
5418 uint64_t const uGuestRFlags = IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode ? pVmcs->u64GuestRFlags.u
5419 : pVmcs->u64GuestRFlags.s.Lo;
5420 if ( !(uGuestRFlags & ~(X86_EFL_LIVE_MASK | X86_EFL_RA1_MASK))
5421 && (uGuestRFlags & X86_EFL_RA1_MASK) == X86_EFL_RA1_MASK)
5422 { /* likely */ }
5423 else
5424 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestRFlagsRsvd);
5425
5426 if ( fGstInLongMode
5427 || !(pVmcs->u64GuestCr0.u & X86_CR0_PE))
5428 {
5429 if (!(uGuestRFlags & X86_EFL_VM))
5430 { /* likely */ }
5431 else
5432 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestRFlagsVm);
5433 }
5434
5435 if (VMX_ENTRY_INT_INFO_IS_EXT_INT(pVmcs->u32EntryIntInfo))
5436 {
5437 if (uGuestRFlags & X86_EFL_IF)
5438 { /* likely */ }
5439 else
5440 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestRFlagsIf);
5441 }
5442
5443 NOREF(pszInstr);
5444 NOREF(pszFailure);
5445 return VINF_SUCCESS;
5446}
5447
5448
5449/**
5450 * Checks guest non-register state as part of VM-entry.
5451 *
5452 * @param pVCpu The cross context virtual CPU structure.
5453 * @param pszInstr The VMX instruction name (for logging purposes).
5454 */
5455DECLINLINE(int) iemVmxVmentryCheckGuestNonRegState(PVMCPUCC pVCpu, const char *pszInstr)
5456{
5457 /*
5458 * Guest non-register state.
5459 * See Intel spec. 26.3.1.5 "Checks on Guest Non-Register State".
5460 */
5461 PVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
5462 const char *const pszFailure = "VM-exit";
5463
5464 /*
5465 * Activity state.
5466 */
5467 uint64_t const u64GuestVmxMiscMsr = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Misc;
5468 uint32_t const fActivityStateMask = RT_BF_GET(u64GuestVmxMiscMsr, VMX_BF_MISC_ACTIVITY_STATES);
5469 if (!(pVmcs->u32GuestActivityState & fActivityStateMask))
5470 { /* likely */ }
5471 else
5472 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestActStateRsvd);
5473
5474 X86DESCATTR AttrSs; AttrSs.u = pVmcs->u32GuestSsAttr;
5475 if ( !AttrSs.n.u2Dpl
5476 || pVmcs->u32GuestActivityState != VMX_VMCS_GUEST_ACTIVITY_HLT)
5477 { /* likely */ }
5478 else
5479 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestActStateSsDpl);
5480
5481 if ( pVmcs->u32GuestIntrState == VMX_VMCS_GUEST_INT_STATE_BLOCK_STI
5482 || pVmcs->u32GuestIntrState == VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS)
5483 {
5484 if (pVmcs->u32GuestActivityState == VMX_VMCS_GUEST_ACTIVITY_ACTIVE)
5485 { /* likely */ }
5486 else
5487 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestActStateStiMovSs);
5488 }
5489
5490 if (VMX_ENTRY_INT_INFO_IS_VALID(pVmcs->u32EntryIntInfo))
5491 {
5492 uint8_t const uType = VMX_ENTRY_INT_INFO_TYPE(pVmcs->u32EntryIntInfo);
5493 uint8_t const uVector = VMX_ENTRY_INT_INFO_VECTOR(pVmcs->u32EntryIntInfo);
5494 AssertCompile(VMX_V_GUEST_ACTIVITY_STATE_MASK == (VMX_VMCS_GUEST_ACTIVITY_HLT | VMX_VMCS_GUEST_ACTIVITY_SHUTDOWN));
5495 switch (pVmcs->u32GuestActivityState)
5496 {
5497 case VMX_VMCS_GUEST_ACTIVITY_HLT:
5498 {
5499 if ( uType == VMX_ENTRY_INT_INFO_TYPE_EXT_INT
5500 || uType == VMX_ENTRY_INT_INFO_TYPE_NMI
5501 || ( uType == VMX_ENTRY_INT_INFO_TYPE_HW_XCPT
5502 && ( uVector == X86_XCPT_DB
5503 || uVector == X86_XCPT_MC))
5504 || ( uType == VMX_ENTRY_INT_INFO_TYPE_OTHER_EVENT
5505 && uVector == VMX_ENTRY_INT_INFO_VECTOR_MTF))
5506 { /* likely */ }
5507 else
5508 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestActStateHlt);
5509 break;
5510 }
5511
5512 case VMX_VMCS_GUEST_ACTIVITY_SHUTDOWN:
5513 {
5514 if ( uType == VMX_ENTRY_INT_INFO_TYPE_NMI
5515 || ( uType == VMX_ENTRY_INT_INFO_TYPE_HW_XCPT
5516 && uVector == X86_XCPT_MC))
5517 { /* likely */ }
5518 else
5519 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestActStateShutdown);
5520 break;
5521 }
5522
5523 case VMX_VMCS_GUEST_ACTIVITY_ACTIVE:
5524 default:
5525 break;
5526 }
5527 }
5528
5529 /*
5530 * Interruptibility state.
5531 */
5532 if (!(pVmcs->u32GuestIntrState & ~VMX_VMCS_GUEST_INT_STATE_MASK))
5533 { /* likely */ }
5534 else
5535 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestIntStateRsvd);
5536
5537 if ((pVmcs->u32GuestIntrState & (VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS | VMX_VMCS_GUEST_INT_STATE_BLOCK_STI))
5538 != (VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS | VMX_VMCS_GUEST_INT_STATE_BLOCK_STI))
5539 { /* likely */ }
5540 else
5541 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestIntStateStiMovSs);
5542
5543 if ( (pVmcs->u64GuestRFlags.u & X86_EFL_IF)
5544 || !(pVmcs->u32GuestIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_STI))
5545 { /* likely */ }
5546 else
5547 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestIntStateRFlagsSti);
5548
5549 if (VMX_ENTRY_INT_INFO_IS_VALID(pVmcs->u32EntryIntInfo))
5550 {
5551 uint8_t const uType = VMX_ENTRY_INT_INFO_TYPE(pVmcs->u32EntryIntInfo);
5552 if (uType == VMX_ENTRY_INT_INFO_TYPE_EXT_INT)
5553 {
5554 if (!(pVmcs->u32GuestIntrState & (VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS | VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)))
5555 { /* likely */ }
5556 else
5557 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestIntStateExtInt);
5558 }
5559 else if (uType == VMX_ENTRY_INT_INFO_TYPE_NMI)
5560 {
5561 if (!(pVmcs->u32GuestIntrState & (VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS | VMX_VMCS_GUEST_INT_STATE_BLOCK_STI)))
5562 { /* likely */ }
5563 else
5564 {
5565 /*
5566 * We don't support injecting NMIs when blocking-by-STI would be in effect.
5567 * We update the Exit qualification only when blocking-by-STI is set
5568 * without blocking-by-MovSS being set. Although in practise it does not
5569 * make much difference since the order of checks are implementation defined.
5570 */
5571 if (!(pVmcs->u32GuestIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS))
5572 iemVmxVmcsSetExitQual(pVCpu, VMX_ENTRY_FAIL_QUAL_NMI_INJECT);
5573 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestIntStateNmi);
5574 }
5575
5576 if ( !(pVmcs->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI)
5577 || !(pVmcs->u32GuestIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI))
5578 { /* likely */ }
5579 else
5580 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestIntStateVirtNmi);
5581 }
5582 }
5583
5584 /* We don't support SMM yet. So blocking-by-SMIs must not be set. */
5585 if (!(pVmcs->u32GuestIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_SMI))
5586 { /* likely */ }
5587 else
5588 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestIntStateSmi);
5589
5590 /* We don't support SGX yet. So enclave-interruption must not be set. */
5591 if (!(pVmcs->u32GuestIntrState & VMX_VMCS_GUEST_INT_STATE_ENCLAVE))
5592 { /* likely */ }
5593 else
5594 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestIntStateEnclave);
5595
5596 /*
5597 * Pending debug exceptions.
5598 */
5599 uint64_t const uPendingDbgXcpts = IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode
5600 ? pVmcs->u64GuestPendingDbgXcpts.u
5601 : pVmcs->u64GuestPendingDbgXcpts.s.Lo;
5602 if (!(uPendingDbgXcpts & ~VMX_VMCS_GUEST_PENDING_DEBUG_VALID_MASK))
5603 { /* likely */ }
5604 else
5605 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestPndDbgXcptRsvd);
5606
5607 if ( (pVmcs->u32GuestIntrState & (VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS | VMX_VMCS_GUEST_INT_STATE_BLOCK_STI))
5608 || pVmcs->u32GuestActivityState == VMX_VMCS_GUEST_ACTIVITY_HLT)
5609 {
5610 if ( (pVmcs->u64GuestRFlags.u & X86_EFL_TF)
5611 && !(pVmcs->u64GuestDebugCtlMsr.u & MSR_IA32_DEBUGCTL_BTF)
5612 && !(uPendingDbgXcpts & VMX_VMCS_GUEST_PENDING_DEBUG_XCPT_BS))
5613 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestPndDbgXcptBsTf);
5614
5615 if ( ( !(pVmcs->u64GuestRFlags.u & X86_EFL_TF)
5616 || (pVmcs->u64GuestDebugCtlMsr.u & MSR_IA32_DEBUGCTL_BTF))
5617 && (uPendingDbgXcpts & VMX_VMCS_GUEST_PENDING_DEBUG_XCPT_BS))
5618 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestPndDbgXcptBsNoTf);
5619 }
5620
5621 /* We don't support RTM (Real-time Transactional Memory) yet. */
5622 if (!(uPendingDbgXcpts & VMX_VMCS_GUEST_PENDING_DEBUG_RTM))
5623 { /* likely */ }
5624 else
5625 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestPndDbgXcptRtm);
5626
5627 /*
5628 * VMCS link pointer.
5629 */
5630 if (pVmcs->u64VmcsLinkPtr.u != UINT64_C(0xffffffffffffffff))
5631 {
5632 RTGCPHYS const GCPhysShadowVmcs = pVmcs->u64VmcsLinkPtr.u;
5633 /* We don't support SMM yet (so VMCS link pointer cannot be the current VMCS). */
5634 if (GCPhysShadowVmcs != IEM_VMX_GET_CURRENT_VMCS(pVCpu))
5635 { /* likely */ }
5636 else
5637 {
5638 iemVmxVmcsSetExitQual(pVCpu, VMX_ENTRY_FAIL_QUAL_VMCS_LINK_PTR);
5639 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_VmcsLinkPtrCurVmcs);
5640 }
5641
5642 /* Validate the address. */
5643 if ( !(GCPhysShadowVmcs & X86_PAGE_4K_OFFSET_MASK)
5644 && !(GCPhysShadowVmcs >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cVmxMaxPhysAddrWidth)
5645 && PGMPhysIsGCPhysNormal(pVCpu->CTX_SUFF(pVM), GCPhysShadowVmcs))
5646 { /* likely */ }
5647 else
5648 {
5649 iemVmxVmcsSetExitQual(pVCpu, VMX_ENTRY_FAIL_QUAL_VMCS_LINK_PTR);
5650 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_AddrVmcsLinkPtr);
5651 }
5652 }
5653
5654 NOREF(pszInstr);
5655 NOREF(pszFailure);
5656 return VINF_SUCCESS;
5657}
5658
5659
5660/**
5661 * Checks if the PDPTEs referenced by the nested-guest CR3 are valid as part of
5662 * VM-entry.
5663 *
5664 * @returns @c true if all PDPTEs are valid, @c false otherwise.
5665 * @param pVCpu The cross context virtual CPU structure.
5666 * @param pszInstr The VMX instruction name (for logging purposes).
5667 * @param pVmcs Pointer to the virtual VMCS.
5668 */
5669IEM_STATIC int iemVmxVmentryCheckGuestPdptesForCr3(PVMCPUCC pVCpu, const char *pszInstr, PVMXVVMCS pVmcs)
5670{
5671 /*
5672 * Check PDPTEs.
5673 * See Intel spec. 4.4.1 "PDPTE Registers".
5674 */
5675 uint64_t const uGuestCr3 = pVmcs->u64GuestCr3.u & X86_CR3_PAE_PAGE_MASK;
5676 const char *const pszFailure = "VM-exit";
5677
5678 X86PDPE aPdptes[X86_PG_PAE_PDPE_ENTRIES];
5679 int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), (void *)&aPdptes[0], uGuestCr3, sizeof(aPdptes));
5680 if (RT_SUCCESS(rc))
5681 {
5682 for (unsigned iPdpte = 0; iPdpte < RT_ELEMENTS(aPdptes); iPdpte++)
5683 {
5684 if ( !(aPdptes[iPdpte].u & X86_PDPE_P)
5685 || !(aPdptes[iPdpte].u & X86_PDPE_PAE_MBZ_MASK))
5686 { /* likely */ }
5687 else
5688 {
5689 iemVmxVmcsSetExitQual(pVCpu, VMX_ENTRY_FAIL_QUAL_PDPTE);
5690 VMXVDIAG const enmDiag = iemVmxGetDiagVmentryPdpteRsvd(iPdpte);
5691 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, enmDiag);
5692 }
5693 }
5694 }
5695 else
5696 {
5697 iemVmxVmcsSetExitQual(pVCpu, VMX_ENTRY_FAIL_QUAL_PDPTE);
5698 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_GuestPdpteCr3ReadPhys);
5699 }
5700
5701 NOREF(pszFailure);
5702 NOREF(pszInstr);
5703 return rc;
5704}
5705
5706
5707/**
5708 * Checks guest PDPTEs as part of VM-entry.
5709 *
5710 * @param pVCpu The cross context virtual CPU structure.
5711 * @param pszInstr The VMX instruction name (for logging purposes).
5712 */
5713DECLINLINE(int) iemVmxVmentryCheckGuestPdptes(PVMCPUCC pVCpu, const char *pszInstr)
5714{
5715 /*
5716 * Guest PDPTEs.
5717 * See Intel spec. 26.3.1.5 "Checks on Guest Page-Directory-Pointer-Table Entries".
5718 */
5719 PVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
5720 bool const fGstInLongMode = RT_BOOL(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_IA32E_MODE_GUEST);
5721
5722 /* Check PDPTes if the VM-entry is to a guest using PAE paging. */
5723 int rc;
5724 if ( !fGstInLongMode
5725 && (pVmcs->u64GuestCr4.u & X86_CR4_PAE)
5726 && (pVmcs->u64GuestCr0.u & X86_CR0_PG))
5727 {
5728 /*
5729 * We don't support nested-paging for nested-guests yet.
5730 *
5731 * Without nested-paging for nested-guests, PDPTEs in the VMCS are not used,
5732 * rather we need to check the PDPTEs referenced by the guest CR3.
5733 */
5734 rc = iemVmxVmentryCheckGuestPdptesForCr3(pVCpu, pszInstr, pVmcs);
5735 }
5736 else
5737 rc = VINF_SUCCESS;
5738 return rc;
5739}
5740
5741
5742/**
5743 * Checks guest-state as part of VM-entry.
5744 *
5745 * @returns VBox status code.
5746 * @param pVCpu The cross context virtual CPU structure.
5747 * @param pszInstr The VMX instruction name (for logging purposes).
5748 */
5749IEM_STATIC int iemVmxVmentryCheckGuestState(PVMCPUCC pVCpu, const char *pszInstr)
5750{
5751 int rc = iemVmxVmentryCheckGuestControlRegsMsrs(pVCpu, pszInstr);
5752 if (RT_SUCCESS(rc))
5753 {
5754 rc = iemVmxVmentryCheckGuestSegRegs(pVCpu, pszInstr);
5755 if (RT_SUCCESS(rc))
5756 {
5757 rc = iemVmxVmentryCheckGuestGdtrIdtr(pVCpu, pszInstr);
5758 if (RT_SUCCESS(rc))
5759 {
5760 rc = iemVmxVmentryCheckGuestRipRFlags(pVCpu, pszInstr);
5761 if (RT_SUCCESS(rc))
5762 {
5763 rc = iemVmxVmentryCheckGuestNonRegState(pVCpu, pszInstr);
5764 if (RT_SUCCESS(rc))
5765 return iemVmxVmentryCheckGuestPdptes(pVCpu, pszInstr);
5766 }
5767 }
5768 }
5769 }
5770 return rc;
5771}
5772
5773
5774/**
5775 * Checks host-state as part of VM-entry.
5776 *
5777 * @returns VBox status code.
5778 * @param pVCpu The cross context virtual CPU structure.
5779 * @param pszInstr The VMX instruction name (for logging purposes).
5780 */
5781IEM_STATIC int iemVmxVmentryCheckHostState(PVMCPUCC pVCpu, const char *pszInstr)
5782{
5783 /*
5784 * Host Control Registers and MSRs.
5785 * See Intel spec. 26.2.2 "Checks on Host Control Registers and MSRs".
5786 */
5787 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
5788 const char * const pszFailure = "VMFail";
5789
5790 /* CR0 reserved bits. */
5791 {
5792 /* CR0 MB1 bits. */
5793 uint64_t const u64Cr0Fixed0 = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Cr0Fixed0;
5794 if ((pVmcs->u64HostCr0.u & u64Cr0Fixed0) == u64Cr0Fixed0)
5795 { /* likely */ }
5796 else
5797 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostCr0Fixed0);
5798
5799 /* CR0 MBZ bits. */
5800 uint64_t const u64Cr0Fixed1 = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Cr0Fixed1;
5801 if (!(pVmcs->u64HostCr0.u & ~u64Cr0Fixed1))
5802 { /* likely */ }
5803 else
5804 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostCr0Fixed1);
5805 }
5806
5807 /* CR4 reserved bits. */
5808 {
5809 /* CR4 MB1 bits. */
5810 uint64_t const u64Cr4Fixed0 = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Cr4Fixed0;
5811 if ((pVmcs->u64HostCr4.u & u64Cr4Fixed0) == u64Cr4Fixed0)
5812 { /* likely */ }
5813 else
5814 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostCr4Fixed0);
5815
5816 /* CR4 MBZ bits. */
5817 uint64_t const u64Cr4Fixed1 = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Cr4Fixed1;
5818 if (!(pVmcs->u64HostCr4.u & ~u64Cr4Fixed1))
5819 { /* likely */ }
5820 else
5821 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostCr4Fixed1);
5822 }
5823
5824 if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode)
5825 {
5826 /* CR3 reserved bits. */
5827 if (!(pVmcs->u64HostCr3.u >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cMaxPhysAddrWidth))
5828 { /* likely */ }
5829 else
5830 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostCr3);
5831
5832 /* SYSENTER ESP and SYSENTER EIP. */
5833 if ( X86_IS_CANONICAL(pVmcs->u64HostSysenterEsp.u)
5834 && X86_IS_CANONICAL(pVmcs->u64HostSysenterEip.u))
5835 { /* likely */ }
5836 else
5837 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostSysenterEspEip);
5838 }
5839
5840 /* We don't support IA32_PERF_GLOBAL_CTRL MSR yet. */
5841 Assert(!(pVmcs->u32ExitCtls & VMX_EXIT_CTLS_LOAD_PERF_MSR));
5842
5843 /* PAT MSR. */
5844 if ( !(pVmcs->u32ExitCtls & VMX_EXIT_CTLS_LOAD_PAT_MSR)
5845 || CPUMIsPatMsrValid(pVmcs->u64HostPatMsr.u))
5846 { /* likely */ }
5847 else
5848 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostPatMsr);
5849
5850 /* EFER MSR. */
5851 uint64_t const uValidEferMask = CPUMGetGuestEferMsrValidMask(pVCpu->CTX_SUFF(pVM));
5852 if ( !(pVmcs->u32ExitCtls & VMX_EXIT_CTLS_LOAD_EFER_MSR)
5853 || !(pVmcs->u64HostEferMsr.u & ~uValidEferMask))
5854 { /* likely */ }
5855 else
5856 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostEferMsrRsvd);
5857
5858 bool const fHostInLongMode = RT_BOOL(pVmcs->u32ExitCtls & VMX_EXIT_CTLS_HOST_ADDR_SPACE_SIZE);
5859 bool const fHostLma = RT_BOOL(pVmcs->u64HostEferMsr.u & MSR_K6_EFER_LMA);
5860 bool const fHostLme = RT_BOOL(pVmcs->u64HostEferMsr.u & MSR_K6_EFER_LME);
5861 if ( fHostInLongMode == fHostLma
5862 && fHostInLongMode == fHostLme)
5863 { /* likely */ }
5864 else
5865 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostEferMsr);
5866
5867 /*
5868 * Host Segment and Descriptor-Table Registers.
5869 * See Intel spec. 26.2.3 "Checks on Host Segment and Descriptor-Table Registers".
5870 */
5871 /* Selector RPL and TI. */
5872 if ( !(pVmcs->HostCs & (X86_SEL_RPL | X86_SEL_LDT))
5873 && !(pVmcs->HostSs & (X86_SEL_RPL | X86_SEL_LDT))
5874 && !(pVmcs->HostDs & (X86_SEL_RPL | X86_SEL_LDT))
5875 && !(pVmcs->HostEs & (X86_SEL_RPL | X86_SEL_LDT))
5876 && !(pVmcs->HostFs & (X86_SEL_RPL | X86_SEL_LDT))
5877 && !(pVmcs->HostGs & (X86_SEL_RPL | X86_SEL_LDT))
5878 && !(pVmcs->HostTr & (X86_SEL_RPL | X86_SEL_LDT)))
5879 { /* likely */ }
5880 else
5881 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostSel);
5882
5883 /* CS and TR selectors cannot be 0. */
5884 if ( pVmcs->HostCs
5885 && pVmcs->HostTr)
5886 { /* likely */ }
5887 else
5888 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostCsTr);
5889
5890 /* SS cannot be 0 if 32-bit host. */
5891 if ( fHostInLongMode
5892 || pVmcs->HostSs)
5893 { /* likely */ }
5894 else
5895 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostSs);
5896
5897 if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode)
5898 {
5899 /* FS, GS, GDTR, IDTR, TR base address. */
5900 if ( X86_IS_CANONICAL(pVmcs->u64HostFsBase.u)
5901 && X86_IS_CANONICAL(pVmcs->u64HostFsBase.u)
5902 && X86_IS_CANONICAL(pVmcs->u64HostGdtrBase.u)
5903 && X86_IS_CANONICAL(pVmcs->u64HostIdtrBase.u)
5904 && X86_IS_CANONICAL(pVmcs->u64HostTrBase.u))
5905 { /* likely */ }
5906 else
5907 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostSegBase);
5908 }
5909
5910 /*
5911 * Host address-space size for 64-bit CPUs.
5912 * See Intel spec. 26.2.4 "Checks Related to Address-Space Size".
5913 */
5914 bool const fGstInLongMode = RT_BOOL(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_IA32E_MODE_GUEST);
5915 if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode)
5916 {
5917 bool const fCpuInLongMode = CPUMIsGuestInLongMode(pVCpu);
5918
5919 /* Logical processor in IA-32e mode. */
5920 if (fCpuInLongMode)
5921 {
5922 if (fHostInLongMode)
5923 {
5924 /* PAE must be set. */
5925 if (pVmcs->u64HostCr4.u & X86_CR4_PAE)
5926 { /* likely */ }
5927 else
5928 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostCr4Pae);
5929
5930 /* RIP must be canonical. */
5931 if (X86_IS_CANONICAL(pVmcs->u64HostRip.u))
5932 { /* likely */ }
5933 else
5934 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostRip);
5935 }
5936 else
5937 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostLongMode);
5938 }
5939 else
5940 {
5941 /* Logical processor is outside IA-32e mode. */
5942 if ( !fGstInLongMode
5943 && !fHostInLongMode)
5944 {
5945 /* PCIDE should not be set. */
5946 if (!(pVmcs->u64HostCr4.u & X86_CR4_PCIDE))
5947 { /* likely */ }
5948 else
5949 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostCr4Pcide);
5950
5951 /* The high 32-bits of RIP MBZ. */
5952 if (!pVmcs->u64HostRip.s.Hi)
5953 { /* likely */ }
5954 else
5955 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostRipRsvd);
5956 }
5957 else
5958 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostGuestLongMode);
5959 }
5960 }
5961 else
5962 {
5963 /* Host address-space size for 32-bit CPUs. */
5964 if ( !fGstInLongMode
5965 && !fHostInLongMode)
5966 { /* likely */ }
5967 else
5968 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_HostGuestLongModeNoCpu);
5969 }
5970
5971 NOREF(pszInstr);
5972 NOREF(pszFailure);
5973 return VINF_SUCCESS;
5974}
5975
5976
5977/**
5978 * Checks VMCS controls fields as part of VM-entry.
5979 *
5980 * @returns VBox status code.
5981 * @param pVCpu The cross context virtual CPU structure.
5982 * @param pszInstr The VMX instruction name (for logging purposes).
5983 *
5984 * @remarks This may update secondary-processor based VM-execution control fields
5985 * in the current VMCS if necessary.
5986 */
5987IEM_STATIC int iemVmxVmentryCheckCtls(PVMCPUCC pVCpu, const char *pszInstr)
5988{
5989 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
5990 const char * const pszFailure = "VMFail";
5991
5992 /*
5993 * VM-execution controls.
5994 * See Intel spec. 26.2.1.1 "VM-Execution Control Fields".
5995 */
5996 {
5997 /* Pin-based VM-execution controls. */
5998 {
5999 VMXCTLSMSR const PinCtls = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.PinCtls;
6000 if (!(~pVmcs->u32PinCtls & PinCtls.n.allowed0))
6001 { /* likely */ }
6002 else
6003 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_PinCtlsDisallowed0);
6004
6005 if (!(pVmcs->u32PinCtls & ~PinCtls.n.allowed1))
6006 { /* likely */ }
6007 else
6008 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_PinCtlsAllowed1);
6009 }
6010
6011 /* Processor-based VM-execution controls. */
6012 {
6013 VMXCTLSMSR const ProcCtls = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.ProcCtls;
6014 if (!(~pVmcs->u32ProcCtls & ProcCtls.n.allowed0))
6015 { /* likely */ }
6016 else
6017 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_ProcCtlsDisallowed0);
6018
6019 if (!(pVmcs->u32ProcCtls & ~ProcCtls.n.allowed1))
6020 { /* likely */ }
6021 else
6022 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_ProcCtlsAllowed1);
6023 }
6024
6025 /* Secondary processor-based VM-execution controls. */
6026 if (pVmcs->u32ProcCtls & VMX_PROC_CTLS_USE_SECONDARY_CTLS)
6027 {
6028 VMXCTLSMSR const ProcCtls2 = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.ProcCtls2;
6029 if (!(~pVmcs->u32ProcCtls2 & ProcCtls2.n.allowed0))
6030 { /* likely */ }
6031 else
6032 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_ProcCtls2Disallowed0);
6033
6034 if (!(pVmcs->u32ProcCtls2 & ~ProcCtls2.n.allowed1))
6035 { /* likely */ }
6036 else
6037 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_ProcCtls2Allowed1);
6038 }
6039 else
6040 Assert(!pVmcs->u32ProcCtls2);
6041
6042 /* CR3-target count. */
6043 if (pVmcs->u32Cr3TargetCount <= VMX_V_CR3_TARGET_COUNT)
6044 { /* likely */ }
6045 else
6046 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_Cr3TargetCount);
6047
6048 /* I/O bitmaps physical addresses. */
6049 if (pVmcs->u32ProcCtls & VMX_PROC_CTLS_USE_IO_BITMAPS)
6050 {
6051 RTGCPHYS const GCPhysIoBitmapA = pVmcs->u64AddrIoBitmapA.u;
6052 if ( !(GCPhysIoBitmapA & X86_PAGE_4K_OFFSET_MASK)
6053 && !(GCPhysIoBitmapA >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cVmxMaxPhysAddrWidth)
6054 && PGMPhysIsGCPhysNormal(pVCpu->CTX_SUFF(pVM), GCPhysIoBitmapA))
6055 { /* likely */ }
6056 else
6057 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_AddrIoBitmapA);
6058
6059 RTGCPHYS const GCPhysIoBitmapB = pVmcs->u64AddrIoBitmapB.u;
6060 if ( !(GCPhysIoBitmapB & X86_PAGE_4K_OFFSET_MASK)
6061 && !(GCPhysIoBitmapB >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cVmxMaxPhysAddrWidth)
6062 && PGMPhysIsGCPhysNormal(pVCpu->CTX_SUFF(pVM), GCPhysIoBitmapB))
6063 { /* likely */ }
6064 else
6065 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_AddrIoBitmapB);
6066 }
6067
6068 /* MSR bitmap physical address. */
6069 if (pVmcs->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
6070 {
6071 RTGCPHYS const GCPhysMsrBitmap = pVmcs->u64AddrMsrBitmap.u;
6072 if ( !(GCPhysMsrBitmap & X86_PAGE_4K_OFFSET_MASK)
6073 && !(GCPhysMsrBitmap >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cVmxMaxPhysAddrWidth)
6074 && PGMPhysIsGCPhysNormal(pVCpu->CTX_SUFF(pVM), GCPhysMsrBitmap))
6075 { /* likely */ }
6076 else
6077 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_AddrMsrBitmap);
6078 }
6079
6080 /* TPR shadow related controls. */
6081 if (pVmcs->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW)
6082 {
6083 /* Virtual-APIC page physical address. */
6084 RTGCPHYS const GCPhysVirtApic = pVmcs->u64AddrVirtApic.u;
6085 if ( !(GCPhysVirtApic & X86_PAGE_4K_OFFSET_MASK)
6086 && !(GCPhysVirtApic >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cVmxMaxPhysAddrWidth)
6087 && PGMPhysIsGCPhysNormal(pVCpu->CTX_SUFF(pVM), GCPhysVirtApic))
6088 { /* likely */ }
6089 else
6090 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_AddrVirtApicPage);
6091
6092 /* TPR threshold bits 31:4 MBZ without virtual-interrupt delivery. */
6093 if ( !(pVmcs->u32TprThreshold & ~VMX_TPR_THRESHOLD_MASK)
6094 || (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_INT_DELIVERY))
6095 { /* likely */ }
6096 else
6097 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_TprThresholdRsvd);
6098
6099 /* The rest done XXX document */
6100 }
6101 else
6102 {
6103 if ( !(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_X2APIC_MODE)
6104 && !(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_APIC_REG_VIRT)
6105 && !(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_INT_DELIVERY))
6106 { /* likely */ }
6107 else
6108 {
6109 if (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_X2APIC_MODE)
6110 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_VirtX2ApicTprShadow);
6111 if (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_APIC_REG_VIRT)
6112 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_ApicRegVirt);
6113 Assert(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_INT_DELIVERY);
6114 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_VirtIntDelivery);
6115 }
6116 }
6117
6118 /* NMI exiting and virtual-NMIs. */
6119 if ( (pVmcs->u32PinCtls & VMX_PIN_CTLS_NMI_EXIT)
6120 || !(pVmcs->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI))
6121 { /* likely */ }
6122 else
6123 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_VirtNmi);
6124
6125 /* Virtual-NMIs and NMI-window exiting. */
6126 if ( (pVmcs->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI)
6127 || !(pVmcs->u32ProcCtls & VMX_PROC_CTLS_NMI_WINDOW_EXIT))
6128 { /* likely */ }
6129 else
6130 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_NmiWindowExit);
6131
6132 /* Virtualize APIC accesses. */
6133 if (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
6134 {
6135 /* APIC-access physical address. */
6136 RTGCPHYS const GCPhysApicAccess = pVmcs->u64AddrApicAccess.u;
6137 if ( !(GCPhysApicAccess & X86_PAGE_4K_OFFSET_MASK)
6138 && !(GCPhysApicAccess >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cVmxMaxPhysAddrWidth)
6139 && PGMPhysIsGCPhysNormal(pVCpu->CTX_SUFF(pVM), GCPhysApicAccess))
6140 { /* likely */ }
6141 else
6142 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_AddrApicAccess);
6143
6144 /*
6145 * Disallow APIC-access page and virtual-APIC page from being the same address.
6146 * Note! This is not an Intel requirement, but one imposed by our implementation.
6147 */
6148 /** @todo r=ramshankar: This is done primarily to simplify recursion scenarios while
6149 * redirecting accesses between the APIC-access page and the virtual-APIC
6150 * page. If any nested hypervisor requires this, we can implement it later. */
6151 if (pVmcs->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW)
6152 {
6153 RTGCPHYS const GCPhysVirtApic = pVmcs->u64AddrVirtApic.u;
6154 if (GCPhysVirtApic != GCPhysApicAccess)
6155 { /* likely */ }
6156 else
6157 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_AddrApicAccessEqVirtApic);
6158 }
6159 }
6160
6161 /* Virtualize-x2APIC mode is mutually exclusive with virtualize-APIC accesses. */
6162 if ( !(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_X2APIC_MODE)
6163 || !(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS))
6164 { /* likely */ }
6165 else
6166 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_VirtX2ApicVirtApic);
6167
6168 /* Virtual-interrupt delivery requires external interrupt exiting. */
6169 if ( !(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_INT_DELIVERY)
6170 || (pVmcs->u32PinCtls & VMX_PIN_CTLS_EXT_INT_EXIT))
6171 { /* likely */ }
6172 else
6173 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_VirtX2ApicVirtApic);
6174
6175 /* VPID. */
6176 if ( !(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VPID)
6177 || pVmcs->u16Vpid != 0)
6178 { /* likely */ }
6179 else
6180 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_Vpid);
6181
6182 Assert(!(pVmcs->u32PinCtls & VMX_PIN_CTLS_POSTED_INT)); /* We don't support posted interrupts yet. */
6183 Assert(!(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_EPT)); /* We don't support EPT yet. */
6184 Assert(!(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_PML)); /* We don't support PML yet. */
6185 Assert(!(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_UNRESTRICTED_GUEST)); /* We don't support Unrestricted-guests yet. */
6186 Assert(!(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VMFUNC)); /* We don't support VM functions yet. */
6187 Assert(!(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_EPT_VE)); /* We don't support EPT-violation #VE yet. */
6188 Assert(!(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_PAUSE_LOOP_EXIT)); /* We don't support Pause-loop exiting yet. */
6189 Assert(!(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_TSC_SCALING)); /* We don't support TSC-scaling yet. */
6190
6191 /* VMCS shadowing. */
6192 if (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING)
6193 {
6194 /* VMREAD-bitmap physical address. */
6195 RTGCPHYS const GCPhysVmreadBitmap = pVmcs->u64AddrVmreadBitmap.u;
6196 if ( !(GCPhysVmreadBitmap & X86_PAGE_4K_OFFSET_MASK)
6197 && !(GCPhysVmreadBitmap >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cVmxMaxPhysAddrWidth)
6198 && PGMPhysIsGCPhysNormal(pVCpu->CTX_SUFF(pVM), GCPhysVmreadBitmap))
6199 { /* likely */ }
6200 else
6201 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_AddrVmreadBitmap);
6202
6203 /* VMWRITE-bitmap physical address. */
6204 RTGCPHYS const GCPhysVmwriteBitmap = pVmcs->u64AddrVmreadBitmap.u;
6205 if ( !(GCPhysVmwriteBitmap & X86_PAGE_4K_OFFSET_MASK)
6206 && !(GCPhysVmwriteBitmap >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cVmxMaxPhysAddrWidth)
6207 && PGMPhysIsGCPhysNormal(pVCpu->CTX_SUFF(pVM), GCPhysVmwriteBitmap))
6208 { /* likely */ }
6209 else
6210 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_AddrVmwriteBitmap);
6211 }
6212 }
6213
6214 /*
6215 * VM-exit controls.
6216 * See Intel spec. 26.2.1.2 "VM-Exit Control Fields".
6217 */
6218 {
6219 VMXCTLSMSR const ExitCtls = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.ExitCtls;
6220 if (!(~pVmcs->u32ExitCtls & ExitCtls.n.allowed0))
6221 { /* likely */ }
6222 else
6223 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_ExitCtlsDisallowed0);
6224
6225 if (!(pVmcs->u32ExitCtls & ~ExitCtls.n.allowed1))
6226 { /* likely */ }
6227 else
6228 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_ExitCtlsAllowed1);
6229
6230 /* Save preemption timer without activating it. */
6231 if ( (pVmcs->u32PinCtls & VMX_PIN_CTLS_PREEMPT_TIMER)
6232 || !(pVmcs->u32ProcCtls & VMX_EXIT_CTLS_SAVE_PREEMPT_TIMER))
6233 { /* likely */ }
6234 else
6235 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_SavePreemptTimer);
6236
6237 /* VM-exit MSR-store count and VM-exit MSR-store area address. */
6238 if (pVmcs->u32ExitMsrStoreCount)
6239 {
6240 if ( !(pVmcs->u64AddrExitMsrStore.u & VMX_AUTOMSR_OFFSET_MASK)
6241 && !(pVmcs->u64AddrExitMsrStore.u >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cVmxMaxPhysAddrWidth)
6242 && PGMPhysIsGCPhysNormal(pVCpu->CTX_SUFF(pVM), pVmcs->u64AddrExitMsrStore.u))
6243 { /* likely */ }
6244 else
6245 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_AddrExitMsrStore);
6246 }
6247
6248 /* VM-exit MSR-load count and VM-exit MSR-load area address. */
6249 if (pVmcs->u32ExitMsrLoadCount)
6250 {
6251 if ( !(pVmcs->u64AddrExitMsrLoad.u & VMX_AUTOMSR_OFFSET_MASK)
6252 && !(pVmcs->u64AddrExitMsrLoad.u >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cVmxMaxPhysAddrWidth)
6253 && PGMPhysIsGCPhysNormal(pVCpu->CTX_SUFF(pVM), pVmcs->u64AddrExitMsrLoad.u))
6254 { /* likely */ }
6255 else
6256 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_AddrExitMsrLoad);
6257 }
6258 }
6259
6260 /*
6261 * VM-entry controls.
6262 * See Intel spec. 26.2.1.3 "VM-Entry Control Fields".
6263 */
6264 {
6265 VMXCTLSMSR const EntryCtls = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.EntryCtls;
6266 if (!(~pVmcs->u32EntryCtls & EntryCtls.n.allowed0))
6267 { /* likely */ }
6268 else
6269 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_EntryCtlsDisallowed0);
6270
6271 if (!(pVmcs->u32EntryCtls & ~EntryCtls.n.allowed1))
6272 { /* likely */ }
6273 else
6274 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_EntryCtlsAllowed1);
6275
6276 /* Event injection. */
6277 uint32_t const uIntInfo = pVmcs->u32EntryIntInfo;
6278 if (RT_BF_GET(uIntInfo, VMX_BF_ENTRY_INT_INFO_VALID))
6279 {
6280 /* Type and vector. */
6281 uint8_t const uType = RT_BF_GET(uIntInfo, VMX_BF_ENTRY_INT_INFO_TYPE);
6282 uint8_t const uVector = RT_BF_GET(uIntInfo, VMX_BF_ENTRY_INT_INFO_VECTOR);
6283 uint8_t const uRsvd = RT_BF_GET(uIntInfo, VMX_BF_ENTRY_INT_INFO_RSVD_12_30);
6284 if ( !uRsvd
6285 && VMXIsEntryIntInfoTypeValid(IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fVmxMonitorTrapFlag, uType)
6286 && VMXIsEntryIntInfoVectorValid(uVector, uType))
6287 { /* likely */ }
6288 else
6289 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_EntryIntInfoTypeVecRsvd);
6290
6291 /* Exception error code. */
6292 if (RT_BF_GET(uIntInfo, VMX_BF_ENTRY_INT_INFO_ERR_CODE_VALID))
6293 {
6294 /* Delivery possible only in Unrestricted-guest mode when CR0.PE is set. */
6295 if ( !(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_UNRESTRICTED_GUEST)
6296 || (pVmcs->u64GuestCr0.s.Lo & X86_CR0_PE))
6297 { /* likely */ }
6298 else
6299 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_EntryIntInfoErrCodePe);
6300
6301 /* Exceptions that provide an error code. */
6302 if ( uType == VMX_ENTRY_INT_INFO_TYPE_HW_XCPT
6303 && ( uVector == X86_XCPT_DF
6304 || uVector == X86_XCPT_TS
6305 || uVector == X86_XCPT_NP
6306 || uVector == X86_XCPT_SS
6307 || uVector == X86_XCPT_GP
6308 || uVector == X86_XCPT_PF
6309 || uVector == X86_XCPT_AC))
6310 { /* likely */ }
6311 else
6312 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_EntryIntInfoErrCodeVec);
6313
6314 /* Exception error-code reserved bits. */
6315 if (!(pVmcs->u32EntryXcptErrCode & ~VMX_ENTRY_INT_XCPT_ERR_CODE_VALID_MASK))
6316 { /* likely */ }
6317 else
6318 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_EntryXcptErrCodeRsvd);
6319
6320 /* Injecting a software interrupt, software exception or privileged software exception. */
6321 if ( uType == VMX_ENTRY_INT_INFO_TYPE_SW_INT
6322 || uType == VMX_ENTRY_INT_INFO_TYPE_SW_XCPT
6323 || uType == VMX_ENTRY_INT_INFO_TYPE_PRIV_SW_XCPT)
6324 {
6325 /* Instruction length must be in the range 0-15. */
6326 if (pVmcs->u32EntryInstrLen <= VMX_ENTRY_INSTR_LEN_MAX)
6327 { /* likely */ }
6328 else
6329 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_EntryInstrLen);
6330
6331 /* However, instruction length of 0 is allowed only when its CPU feature is present. */
6332 if ( pVmcs->u32EntryInstrLen != 0
6333 || IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fVmxEntryInjectSoftInt)
6334 { /* likely */ }
6335 else
6336 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_EntryInstrLenZero);
6337 }
6338 }
6339 }
6340
6341 /* VM-entry MSR-load count and VM-entry MSR-load area address. */
6342 if (pVmcs->u32EntryMsrLoadCount)
6343 {
6344 if ( !(pVmcs->u64AddrEntryMsrLoad.u & VMX_AUTOMSR_OFFSET_MASK)
6345 && !(pVmcs->u64AddrEntryMsrLoad.u >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cVmxMaxPhysAddrWidth)
6346 && PGMPhysIsGCPhysNormal(pVCpu->CTX_SUFF(pVM), pVmcs->u64AddrEntryMsrLoad.u))
6347 { /* likely */ }
6348 else
6349 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_AddrEntryMsrLoad);
6350 }
6351
6352 Assert(!(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_ENTRY_TO_SMM)); /* We don't support SMM yet. */
6353 Assert(!(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_DEACTIVATE_DUAL_MON)); /* We don't support dual-monitor treatment yet. */
6354 }
6355
6356 NOREF(pszInstr);
6357 NOREF(pszFailure);
6358 return VINF_SUCCESS;
6359}
6360
6361
6362/**
6363 * Loads the guest control registers, debug register and some MSRs as part of
6364 * VM-entry.
6365 *
6366 * @param pVCpu The cross context virtual CPU structure.
6367 */
6368IEM_STATIC void iemVmxVmentryLoadGuestControlRegsMsrs(PVMCPUCC pVCpu)
6369{
6370 /*
6371 * Load guest control registers, debug registers and MSRs.
6372 * See Intel spec. 26.3.2.1 "Loading Guest Control Registers, Debug Registers and MSRs".
6373 */
6374 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
6375
6376 IEM_CTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR0);
6377 uint64_t const uGstCr0 = (pVmcs->u64GuestCr0.u & ~VMX_ENTRY_GUEST_CR0_IGNORE_MASK)
6378 | (pVCpu->cpum.GstCtx.cr0 & VMX_ENTRY_GUEST_CR0_IGNORE_MASK);
6379 CPUMSetGuestCR0(pVCpu, uGstCr0);
6380 CPUMSetGuestCR4(pVCpu, pVmcs->u64GuestCr4.u);
6381 pVCpu->cpum.GstCtx.cr3 = pVmcs->u64GuestCr3.u;
6382
6383 if (pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_DEBUG)
6384 pVCpu->cpum.GstCtx.dr[7] = (pVmcs->u64GuestDr7.u & ~VMX_ENTRY_GUEST_DR7_MBZ_MASK) | VMX_ENTRY_GUEST_DR7_MB1_MASK;
6385
6386 pVCpu->cpum.GstCtx.SysEnter.eip = pVmcs->u64GuestSysenterEip.s.Lo;
6387 pVCpu->cpum.GstCtx.SysEnter.esp = pVmcs->u64GuestSysenterEsp.s.Lo;
6388 pVCpu->cpum.GstCtx.SysEnter.cs = pVmcs->u32GuestSysenterCS;
6389
6390 if (IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fLongMode)
6391 {
6392 /* FS base and GS base are loaded while loading the rest of the guest segment registers. */
6393
6394 /* EFER MSR. */
6395 if (!(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_EFER_MSR))
6396 {
6397 IEM_CTX_ASSERT(pVCpu, CPUMCTX_EXTRN_EFER);
6398 uint64_t const uHostEfer = pVCpu->cpum.GstCtx.msrEFER;
6399 bool const fGstInLongMode = RT_BOOL(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_IA32E_MODE_GUEST);
6400 bool const fGstPaging = RT_BOOL(uGstCr0 & X86_CR0_PG);
6401 if (fGstInLongMode)
6402 {
6403 /* If the nested-guest is in long mode, LMA and LME are both set. */
6404 Assert(fGstPaging);
6405 pVCpu->cpum.GstCtx.msrEFER = uHostEfer | (MSR_K6_EFER_LMA | MSR_K6_EFER_LME);
6406 }
6407 else
6408 {
6409 /*
6410 * If the nested-guest is outside long mode:
6411 * - With paging: LMA is cleared, LME is cleared.
6412 * - Without paging: LMA is cleared, LME is left unmodified.
6413 */
6414 uint64_t const fLmaLmeMask = MSR_K6_EFER_LMA | (fGstPaging ? MSR_K6_EFER_LME : 0);
6415 pVCpu->cpum.GstCtx.msrEFER = uHostEfer & ~fLmaLmeMask;
6416 }
6417 }
6418 /* else: see below. */
6419 }
6420
6421 /* PAT MSR. */
6422 if (pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_PAT_MSR)
6423 pVCpu->cpum.GstCtx.msrPAT = pVmcs->u64GuestPatMsr.u;
6424
6425 /* EFER MSR. */
6426 if (pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_EFER_MSR)
6427 pVCpu->cpum.GstCtx.msrEFER = pVmcs->u64GuestEferMsr.u;
6428
6429 /* We don't support IA32_PERF_GLOBAL_CTRL MSR yet. */
6430 Assert(!(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_PERF_MSR));
6431
6432 /* We don't support IA32_BNDCFGS MSR yet. */
6433 Assert(!(pVmcs->u32EntryCtls & VMX_ENTRY_CTLS_LOAD_BNDCFGS_MSR));
6434
6435 /* Nothing to do for SMBASE register - We don't support SMM yet. */
6436}
6437
6438
6439/**
6440 * Loads the guest segment registers, GDTR, IDTR, LDTR and TR as part of VM-entry.
6441 *
6442 * @param pVCpu The cross context virtual CPU structure.
6443 */
6444IEM_STATIC void iemVmxVmentryLoadGuestSegRegs(PVMCPUCC pVCpu)
6445{
6446 /*
6447 * Load guest segment registers, GDTR, IDTR, LDTR and TR.
6448 * See Intel spec. 26.3.2.2 "Loading Guest Segment Registers and Descriptor-Table Registers".
6449 */
6450 /* CS, SS, ES, DS, FS, GS. */
6451 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
6452 for (unsigned iSegReg = 0; iSegReg < X86_SREG_COUNT; iSegReg++)
6453 {
6454 PCPUMSELREG pGstSelReg = &pVCpu->cpum.GstCtx.aSRegs[iSegReg];
6455 CPUMSELREG VmcsSelReg;
6456 int rc = iemVmxVmcsGetGuestSegReg(pVmcs, iSegReg, &VmcsSelReg);
6457 AssertRC(rc); NOREF(rc);
6458 if (!(VmcsSelReg.Attr.u & X86DESCATTR_UNUSABLE))
6459 {
6460 pGstSelReg->Sel = VmcsSelReg.Sel;
6461 pGstSelReg->ValidSel = VmcsSelReg.Sel;
6462 pGstSelReg->fFlags = CPUMSELREG_FLAGS_VALID;
6463 pGstSelReg->u64Base = VmcsSelReg.u64Base;
6464 pGstSelReg->u32Limit = VmcsSelReg.u32Limit;
6465 pGstSelReg->Attr.u = VmcsSelReg.Attr.u;
6466 }
6467 else
6468 {
6469 pGstSelReg->Sel = VmcsSelReg.Sel;
6470 pGstSelReg->ValidSel = VmcsSelReg.Sel;
6471 pGstSelReg->fFlags = CPUMSELREG_FLAGS_VALID;
6472 switch (iSegReg)
6473 {
6474 case X86_SREG_CS:
6475 pGstSelReg->u64Base = VmcsSelReg.u64Base;
6476 pGstSelReg->u32Limit = VmcsSelReg.u32Limit;
6477 pGstSelReg->Attr.u = VmcsSelReg.Attr.u;
6478 break;
6479
6480 case X86_SREG_SS:
6481 pGstSelReg->u64Base = VmcsSelReg.u64Base & UINT32_C(0xfffffff0);
6482 pGstSelReg->u32Limit = 0;
6483 pGstSelReg->Attr.u = (VmcsSelReg.Attr.u & X86DESCATTR_DPL) | X86DESCATTR_D | X86DESCATTR_UNUSABLE;
6484 break;
6485
6486 case X86_SREG_ES:
6487 case X86_SREG_DS:
6488 pGstSelReg->u64Base = 0;
6489 pGstSelReg->u32Limit = 0;
6490 pGstSelReg->Attr.u = X86DESCATTR_UNUSABLE;
6491 break;
6492
6493 case X86_SREG_FS:
6494 case X86_SREG_GS:
6495 pGstSelReg->u64Base = VmcsSelReg.u64Base;
6496 pGstSelReg->u32Limit = 0;
6497 pGstSelReg->Attr.u = X86DESCATTR_UNUSABLE;
6498 break;
6499 }
6500 Assert(pGstSelReg->Attr.n.u1Unusable);
6501 }
6502 }
6503
6504 /* LDTR. */
6505 pVCpu->cpum.GstCtx.ldtr.Sel = pVmcs->GuestLdtr;
6506 pVCpu->cpum.GstCtx.ldtr.ValidSel = pVmcs->GuestLdtr;
6507 pVCpu->cpum.GstCtx.ldtr.fFlags = CPUMSELREG_FLAGS_VALID;
6508 if (!(pVmcs->u32GuestLdtrAttr & X86DESCATTR_UNUSABLE))
6509 {
6510 pVCpu->cpum.GstCtx.ldtr.u64Base = pVmcs->u64GuestLdtrBase.u;
6511 pVCpu->cpum.GstCtx.ldtr.u32Limit = pVmcs->u32GuestLdtrLimit;
6512 pVCpu->cpum.GstCtx.ldtr.Attr.u = pVmcs->u32GuestLdtrAttr;
6513 }
6514 else
6515 {
6516 pVCpu->cpum.GstCtx.ldtr.u64Base = 0;
6517 pVCpu->cpum.GstCtx.ldtr.u32Limit = 0;
6518 pVCpu->cpum.GstCtx.ldtr.Attr.u = X86DESCATTR_UNUSABLE;
6519 }
6520
6521 /* TR. */
6522 Assert(!(pVmcs->u32GuestTrAttr & X86DESCATTR_UNUSABLE));
6523 pVCpu->cpum.GstCtx.tr.Sel = pVmcs->GuestTr;
6524 pVCpu->cpum.GstCtx.tr.ValidSel = pVmcs->GuestTr;
6525 pVCpu->cpum.GstCtx.tr.fFlags = CPUMSELREG_FLAGS_VALID;
6526 pVCpu->cpum.GstCtx.tr.u64Base = pVmcs->u64GuestTrBase.u;
6527 pVCpu->cpum.GstCtx.tr.u32Limit = pVmcs->u32GuestTrLimit;
6528 pVCpu->cpum.GstCtx.tr.Attr.u = pVmcs->u32GuestTrAttr;
6529
6530 /* GDTR. */
6531 pVCpu->cpum.GstCtx.gdtr.cbGdt = pVmcs->u32GuestGdtrLimit;
6532 pVCpu->cpum.GstCtx.gdtr.pGdt = pVmcs->u64GuestGdtrBase.u;
6533
6534 /* IDTR. */
6535 pVCpu->cpum.GstCtx.idtr.cbIdt = pVmcs->u32GuestIdtrLimit;
6536 pVCpu->cpum.GstCtx.idtr.pIdt = pVmcs->u64GuestIdtrBase.u;
6537}
6538
6539
6540/**
6541 * Loads the guest MSRs from the VM-entry MSR-load area as part of VM-entry.
6542 *
6543 * @returns VBox status code.
6544 * @param pVCpu The cross context virtual CPU structure.
6545 * @param pszInstr The VMX instruction name (for logging purposes).
6546 */
6547IEM_STATIC int iemVmxVmentryLoadGuestAutoMsrs(PVMCPUCC pVCpu, const char *pszInstr)
6548{
6549 /*
6550 * Load guest MSRs.
6551 * See Intel spec. 26.4 "Loading MSRs".
6552 */
6553 PVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
6554 const char *const pszFailure = "VM-exit";
6555
6556 /*
6557 * The VM-entry MSR-load area address need not be a valid guest-physical address if the
6558 * VM-entry MSR load count is 0. If this is the case, bail early without reading it.
6559 * See Intel spec. 24.8.2 "VM-Entry Controls for MSRs".
6560 */
6561 uint32_t const cMsrs = pVmcs->u32EntryMsrLoadCount;
6562 if (!cMsrs)
6563 return VINF_SUCCESS;
6564
6565 /*
6566 * Verify the MSR auto-load count. Physical CPUs can behave unpredictably if the count is
6567 * exceeded including possibly raising #MC exceptions during VMX transition. Our
6568 * implementation shall fail VM-entry with an VMX_EXIT_ERR_MSR_LOAD VM-exit.
6569 */
6570 bool const fIsMsrCountValid = iemVmxIsAutoMsrCountValid(pVCpu, cMsrs);
6571 if (fIsMsrCountValid)
6572 { /* likely */ }
6573 else
6574 {
6575 iemVmxVmcsSetExitQual(pVCpu, VMX_V_AUTOMSR_AREA_SIZE / sizeof(VMXAUTOMSR));
6576 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_MsrLoadCount);
6577 }
6578
6579 RTGCPHYS const GCPhysVmEntryMsrLoadArea = pVmcs->u64AddrEntryMsrLoad.u;
6580 int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), (void *)pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pEntryMsrLoadArea),
6581 GCPhysVmEntryMsrLoadArea, cMsrs * sizeof(VMXAUTOMSR));
6582 if (RT_SUCCESS(rc))
6583 {
6584 PCVMXAUTOMSR pMsr = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pEntryMsrLoadArea);
6585 Assert(pMsr);
6586 for (uint32_t idxMsr = 0; idxMsr < cMsrs; idxMsr++, pMsr++)
6587 {
6588 if ( !pMsr->u32Reserved
6589 && pMsr->u32Msr != MSR_K8_FS_BASE
6590 && pMsr->u32Msr != MSR_K8_GS_BASE
6591 && pMsr->u32Msr != MSR_K6_EFER
6592 && pMsr->u32Msr != MSR_IA32_SMM_MONITOR_CTL
6593 && pMsr->u32Msr >> 8 != MSR_IA32_X2APIC_START >> 8)
6594 {
6595 VBOXSTRICTRC rcStrict = CPUMSetGuestMsr(pVCpu, pMsr->u32Msr, pMsr->u64Value);
6596 if (rcStrict == VINF_SUCCESS)
6597 continue;
6598
6599 /*
6600 * If we're in ring-0, we cannot handle returns to ring-3 at this point and continue VM-entry.
6601 * If any nested hypervisor loads MSRs that require ring-3 handling, we cause a VM-entry failure
6602 * recording the MSR index in the Exit qualification (as per the Intel spec.) and indicated
6603 * further by our own, specific diagnostic code. Later, we can try implement handling of the
6604 * MSR in ring-0 if possible, or come up with a better, generic solution.
6605 */
6606 iemVmxVmcsSetExitQual(pVCpu, idxMsr);
6607 VMXVDIAG const enmDiag = rcStrict == VINF_CPUM_R3_MSR_WRITE
6608 ? kVmxVDiag_Vmentry_MsrLoadRing3
6609 : kVmxVDiag_Vmentry_MsrLoad;
6610 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, enmDiag);
6611 }
6612 else
6613 {
6614 iemVmxVmcsSetExitQual(pVCpu, idxMsr);
6615 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_MsrLoadRsvd);
6616 }
6617 }
6618 }
6619 else
6620 {
6621 AssertMsgFailed(("%s: Failed to read MSR auto-load area at %#RGp, rc=%Rrc\n", pszInstr, GCPhysVmEntryMsrLoadArea, rc));
6622 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_MsrLoadPtrReadPhys);
6623 }
6624
6625 NOREF(pszInstr);
6626 NOREF(pszFailure);
6627 return VINF_SUCCESS;
6628}
6629
6630
6631/**
6632 * Loads the guest-state non-register state as part of VM-entry.
6633 *
6634 * @returns VBox status code.
6635 * @param pVCpu The cross context virtual CPU structure.
6636 *
6637 * @remarks This must be called only after loading the nested-guest register state
6638 * (especially nested-guest RIP).
6639 */
6640IEM_STATIC void iemVmxVmentryLoadGuestNonRegState(PVMCPUCC pVCpu)
6641{
6642 /*
6643 * Load guest non-register state.
6644 * See Intel spec. 26.6 "Special Features of VM Entry"
6645 */
6646 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
6647
6648 /*
6649 * If VM-entry is not vectoring, block-by-STI and block-by-MovSS state must be loaded.
6650 * If VM-entry is vectoring, there is no block-by-STI or block-by-MovSS.
6651 *
6652 * See Intel spec. 26.6.1 "Interruptibility State".
6653 */
6654 bool const fEntryVectoring = VMXIsVmentryVectoring(pVmcs->u32EntryIntInfo, NULL /* puEntryIntInfoType */);
6655 if ( !fEntryVectoring
6656 && (pVmcs->u32GuestIntrState & (VMX_VMCS_GUEST_INT_STATE_BLOCK_STI | VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS)))
6657 EMSetInhibitInterruptsPC(pVCpu, pVmcs->u64GuestRip.u);
6658 else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
6659 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
6660
6661 /* NMI blocking. */
6662 if (pVmcs->u32GuestIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_NMI)
6663 {
6664 if (pVmcs->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI)
6665 pVCpu->cpum.GstCtx.hwvirt.vmx.fVirtNmiBlocking = true;
6666 else
6667 {
6668 pVCpu->cpum.GstCtx.hwvirt.vmx.fVirtNmiBlocking = false;
6669 if (!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_BLOCK_NMIS))
6670 VMCPU_FF_SET(pVCpu, VMCPU_FF_BLOCK_NMIS);
6671 }
6672 }
6673 else
6674 pVCpu->cpum.GstCtx.hwvirt.vmx.fVirtNmiBlocking = false;
6675
6676 /* SMI blocking is irrelevant. We don't support SMIs yet. */
6677
6678 /* Loading PDPTEs will be taken care when we switch modes. We don't support EPT yet. */
6679 Assert(!(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_EPT));
6680
6681 /* VPID is irrelevant. We don't support VPID yet. */
6682
6683 /* Clear address-range monitoring. */
6684 EMMonitorWaitClear(pVCpu);
6685}
6686
6687
6688/**
6689 * Loads the guest VMCS referenced state (such as MSR bitmaps, I/O bitmaps etc).
6690 *
6691 * @param pVCpu The cross context virtual CPU structure.
6692 * @param pszInstr The VMX instruction name (for logging purposes).
6693 *
6694 * @remarks This assumes various VMCS related data structure pointers have already
6695 * been verified prior to calling this function.
6696 */
6697IEM_STATIC int iemVmxVmentryLoadGuestVmcsRefState(PVMCPUCC pVCpu, const char *pszInstr)
6698{
6699 const char *const pszFailure = "VM-exit";
6700 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
6701
6702 /*
6703 * Virtualize APIC accesses.
6704 */
6705 if (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
6706 {
6707 /* APIC-access physical address. */
6708 RTGCPHYS const GCPhysApicAccess = pVmcs->u64AddrApicAccess.u;
6709
6710 /*
6711 * Register the handler for the APIC-access page.
6712 *
6713 * We don't deregister the APIC-access page handler during the VM-exit as a different
6714 * nested-VCPU might be using the same guest-physical address for its APIC-access page.
6715 *
6716 * We leave the page registered until the first access that happens outside VMX non-root
6717 * mode. Guest software is allowed to access structures such as the APIC-access page
6718 * only when no logical processor with a current VMCS references it in VMX non-root mode,
6719 * otherwise it can lead to unpredictable behavior including guest triple-faults.
6720 *
6721 * See Intel spec. 24.11.4 "Software Access to Related Structures".
6722 */
6723 if (!PGMHandlerPhysicalIsRegistered(pVCpu->CTX_SUFF(pVM), GCPhysApicAccess))
6724 {
6725 PVMCC pVM = pVCpu->CTX_SUFF(pVM);
6726 PVMCPUCC pVCpu0 = VMCC_GET_CPU_0(pVM);
6727 int rc = PGMHandlerPhysicalRegister(pVM, GCPhysApicAccess, GCPhysApicAccess + X86_PAGE_4K_SIZE - 1,
6728 pVCpu0->iem.s.hVmxApicAccessPage, NIL_RTR3PTR /* pvUserR3 */,
6729 NIL_RTR0PTR /* pvUserR0 */, NIL_RTRCPTR /* pvUserRC */, NULL /* pszDesc */);
6730 if (RT_SUCCESS(rc))
6731 { /* likely */ }
6732 else
6733 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_AddrApicAccessHandlerReg);
6734 }
6735 }
6736
6737 /*
6738 * VMCS shadowing.
6739 */
6740 if (pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING)
6741 {
6742 /* Read the VMREAD-bitmap. */
6743 RTGCPHYS const GCPhysVmreadBitmap = pVmcs->u64AddrVmreadBitmap.u;
6744 Assert(pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvVmreadBitmap));
6745 int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvVmreadBitmap),
6746 GCPhysVmreadBitmap, VMX_V_VMREAD_VMWRITE_BITMAP_SIZE);
6747 if (RT_SUCCESS(rc))
6748 { /* likely */ }
6749 else
6750 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_VmreadBitmapPtrReadPhys);
6751
6752 /* Read the VMWRITE-bitmap. */
6753 RTGCPHYS const GCPhysVmwriteBitmap = pVmcs->u64AddrVmwriteBitmap.u;
6754 Assert(pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvVmwriteBitmap));
6755 rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvVmwriteBitmap),
6756 GCPhysVmwriteBitmap, VMX_V_VMREAD_VMWRITE_BITMAP_SIZE);
6757 if (RT_SUCCESS(rc))
6758 { /* likely */ }
6759 else
6760 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_VmwriteBitmapPtrReadPhys);
6761 }
6762
6763 /*
6764 * I/O bitmaps.
6765 */
6766 if (pVmcs->u32ProcCtls & VMX_PROC_CTLS_USE_IO_BITMAPS)
6767 {
6768 /* Read the IO bitmap A. */
6769 RTGCPHYS const GCPhysIoBitmapA = pVmcs->u64AddrIoBitmapA.u;
6770 Assert(pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvIoBitmap));
6771 int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvIoBitmap),
6772 GCPhysIoBitmapA, VMX_V_IO_BITMAP_A_SIZE);
6773 if (RT_SUCCESS(rc))
6774 { /* likely */ }
6775 else
6776 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_IoBitmapAPtrReadPhys);
6777
6778 /* Read the IO bitmap B. */
6779 RTGCPHYS const GCPhysIoBitmapB = pVmcs->u64AddrIoBitmapB.u;
6780 uint8_t *pbIoBitmapB = (uint8_t *)pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvIoBitmap) + VMX_V_IO_BITMAP_A_SIZE;
6781 rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), pbIoBitmapB, GCPhysIoBitmapB, VMX_V_IO_BITMAP_B_SIZE);
6782 if (RT_SUCCESS(rc))
6783 { /* likely */ }
6784 else
6785 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_IoBitmapBPtrReadPhys);
6786 }
6787
6788 /*
6789 * TPR shadow and Virtual-APIC page.
6790 */
6791 if (pVmcs->u32ProcCtls & VMX_PROC_CTLS_USE_TPR_SHADOW)
6792 {
6793 /* Verify TPR threshold and VTPR when both virtualize-APIC accesses and virtual-interrupt delivery aren't used. */
6794 if ( !(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_APIC_ACCESS)
6795 && !(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VIRT_INT_DELIVERY))
6796 {
6797 /* Read the VTPR from the virtual-APIC page. */
6798 RTGCPHYS const GCPhysVirtApic = pVmcs->u64AddrVirtApic.u;
6799 uint8_t u8VTpr;
6800 int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), &u8VTpr, GCPhysVirtApic + XAPIC_OFF_TPR, sizeof(u8VTpr));
6801 if (RT_SUCCESS(rc))
6802 { /* likely */ }
6803 else
6804 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_VirtApicPagePtrReadPhys);
6805
6806 /* Bits 3:0 of the TPR-threshold must not be greater than bits 7:4 of VTPR. */
6807 if ((uint8_t)RT_BF_GET(pVmcs->u32TprThreshold, VMX_BF_TPR_THRESHOLD_TPR) <= (u8VTpr & 0xf0))
6808 { /* likely */ }
6809 else
6810 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_TprThresholdVTpr);
6811 }
6812 }
6813
6814 /*
6815 * VMCS link pointer.
6816 */
6817 if (pVmcs->u64VmcsLinkPtr.u != UINT64_C(0xffffffffffffffff))
6818 {
6819 /* Read the VMCS-link pointer from guest memory. */
6820 RTGCPHYS const GCPhysShadowVmcs = pVmcs->u64VmcsLinkPtr.u;
6821 Assert(pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pShadowVmcs));
6822 int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pShadowVmcs),
6823 GCPhysShadowVmcs, VMX_V_SHADOW_VMCS_SIZE);
6824 if (RT_SUCCESS(rc))
6825 { /* likely */ }
6826 else
6827 {
6828 iemVmxVmcsSetExitQual(pVCpu, VMX_ENTRY_FAIL_QUAL_VMCS_LINK_PTR);
6829 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_VmcsLinkPtrReadPhys);
6830 }
6831
6832 /* Verify the VMCS revision specified by the guest matches what we reported to the guest. */
6833 if (pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pShadowVmcs)->u32VmcsRevId.n.u31RevisionId == VMX_V_VMCS_REVISION_ID)
6834 { /* likely */ }
6835 else
6836 {
6837 iemVmxVmcsSetExitQual(pVCpu, VMX_ENTRY_FAIL_QUAL_VMCS_LINK_PTR);
6838 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_VmcsLinkPtrRevId);
6839 }
6840
6841 /* Verify the shadow bit is set if VMCS shadowing is enabled . */
6842 if ( !(pVmcs->u32ProcCtls2 & VMX_PROC_CTLS2_VMCS_SHADOWING)
6843 || pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pShadowVmcs)->u32VmcsRevId.n.fIsShadowVmcs)
6844 { /* likely */ }
6845 else
6846 {
6847 iemVmxVmcsSetExitQual(pVCpu, VMX_ENTRY_FAIL_QUAL_VMCS_LINK_PTR);
6848 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_VmcsLinkPtrShadow);
6849 }
6850
6851 /* Update our cache of the guest physical address of the shadow VMCS. */
6852 pVCpu->cpum.GstCtx.hwvirt.vmx.GCPhysShadowVmcs = GCPhysShadowVmcs;
6853 }
6854
6855 /*
6856 * MSR bitmap.
6857 */
6858 if (pVmcs->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
6859 {
6860 /* Read the MSR bitmap. */
6861 RTGCPHYS const GCPhysMsrBitmap = pVmcs->u64AddrMsrBitmap.u;
6862 Assert(pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvMsrBitmap));
6863 int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvMsrBitmap),
6864 GCPhysMsrBitmap, VMX_V_MSR_BITMAP_SIZE);
6865 if (RT_SUCCESS(rc))
6866 { /* likely */ }
6867 else
6868 IEM_VMX_VMENTRY_FAILED_RET(pVCpu, pszInstr, pszFailure, kVmxVDiag_Vmentry_MsrBitmapPtrReadPhys);
6869 }
6870
6871 NOREF(pszFailure);
6872 NOREF(pszInstr);
6873 return VINF_SUCCESS;
6874}
6875
6876
6877/**
6878 * Loads the guest-state as part of VM-entry.
6879 *
6880 * @returns VBox status code.
6881 * @param pVCpu The cross context virtual CPU structure.
6882 * @param pszInstr The VMX instruction name (for logging purposes).
6883 *
6884 * @remarks This must be done after all the necessary steps prior to loading of
6885 * guest-state (e.g. checking various VMCS state).
6886 */
6887IEM_STATIC int iemVmxVmentryLoadGuestState(PVMCPUCC pVCpu, const char *pszInstr)
6888{
6889 /* Load guest control registers, MSRs (that are directly part of the VMCS). */
6890 iemVmxVmentryLoadGuestControlRegsMsrs(pVCpu);
6891
6892 /* Load guest segment registers. */
6893 iemVmxVmentryLoadGuestSegRegs(pVCpu);
6894
6895 /*
6896 * Load guest RIP, RSP and RFLAGS.
6897 * See Intel spec. 26.3.2.3 "Loading Guest RIP, RSP and RFLAGS".
6898 */
6899 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
6900 pVCpu->cpum.GstCtx.rsp = pVmcs->u64GuestRsp.u;
6901 pVCpu->cpum.GstCtx.rip = pVmcs->u64GuestRip.u;
6902 pVCpu->cpum.GstCtx.rflags.u = pVmcs->u64GuestRFlags.u;
6903
6904 /* Initialize the PAUSE-loop controls as part of VM-entry. */
6905 pVCpu->cpum.GstCtx.hwvirt.vmx.uFirstPauseLoopTick = 0;
6906 pVCpu->cpum.GstCtx.hwvirt.vmx.uPrevPauseTick = 0;
6907
6908 /* Load guest non-register state (such as interrupt shadows, NMI blocking etc). */
6909 iemVmxVmentryLoadGuestNonRegState(pVCpu);
6910
6911 /* Load VMX related structures and state referenced by the VMCS. */
6912 int rc = iemVmxVmentryLoadGuestVmcsRefState(pVCpu, pszInstr);
6913 if (rc == VINF_SUCCESS)
6914 { /* likely */ }
6915 else
6916 return rc;
6917
6918 NOREF(pszInstr);
6919 return VINF_SUCCESS;
6920}
6921
6922
6923/**
6924 * Returns whether there are is a pending debug exception on VM-entry.
6925 *
6926 * @param pVCpu The cross context virtual CPU structure.
6927 * @param pszInstr The VMX instruction name (for logging purposes).
6928 */
6929IEM_STATIC bool iemVmxVmentryIsPendingDebugXcpt(PVMCPUCC pVCpu, const char *pszInstr)
6930{
6931 /*
6932 * Pending debug exceptions.
6933 * See Intel spec. 26.6.3 "Delivery of Pending Debug Exceptions after VM Entry".
6934 */
6935 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
6936 Assert(pVmcs);
6937
6938 bool fPendingDbgXcpt = RT_BOOL(pVmcs->u64GuestPendingDbgXcpts.u & ( VMX_VMCS_GUEST_PENDING_DEBUG_XCPT_BS
6939 | VMX_VMCS_GUEST_PENDING_DEBUG_XCPT_EN_BP));
6940 if (fPendingDbgXcpt)
6941 {
6942 uint8_t uEntryIntInfoType;
6943 bool const fEntryVectoring = VMXIsVmentryVectoring(pVmcs->u32EntryIntInfo, &uEntryIntInfoType);
6944 if (fEntryVectoring)
6945 {
6946 switch (uEntryIntInfoType)
6947 {
6948 case VMX_ENTRY_INT_INFO_TYPE_EXT_INT:
6949 case VMX_ENTRY_INT_INFO_TYPE_NMI:
6950 case VMX_ENTRY_INT_INFO_TYPE_HW_XCPT:
6951 case VMX_ENTRY_INT_INFO_TYPE_PRIV_SW_XCPT:
6952 fPendingDbgXcpt = false;
6953 break;
6954
6955 case VMX_ENTRY_INT_INFO_TYPE_SW_XCPT:
6956 {
6957 /*
6958 * Whether the pending debug exception for software exceptions other than
6959 * #BP and #OF is delivered after injecting the exception or is discard
6960 * is CPU implementation specific. We will discard them (easier).
6961 */
6962 uint8_t const uVector = VMX_ENTRY_INT_INFO_VECTOR(pVmcs->u32EntryIntInfo);
6963 if ( uVector != X86_XCPT_BP
6964 && uVector != X86_XCPT_OF)
6965 fPendingDbgXcpt = false;
6966 RT_FALL_THRU();
6967 }
6968 case VMX_ENTRY_INT_INFO_TYPE_SW_INT:
6969 {
6970 if (!(pVmcs->u32GuestIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS))
6971 fPendingDbgXcpt = false;
6972 break;
6973 }
6974 }
6975 }
6976 else
6977 {
6978 /*
6979 * When the VM-entry is not vectoring but there is blocking-by-MovSS, whether the
6980 * pending debug exception is held pending or is discarded is CPU implementation
6981 * specific. We will discard them (easier).
6982 */
6983 if (pVmcs->u32GuestIntrState & VMX_VMCS_GUEST_INT_STATE_BLOCK_MOVSS)
6984 fPendingDbgXcpt = false;
6985
6986 /* There's no pending debug exception in the shutdown or wait-for-SIPI state. */
6987 if (pVmcs->u32GuestActivityState & (VMX_VMCS_GUEST_ACTIVITY_SHUTDOWN | VMX_VMCS_GUEST_ACTIVITY_SIPI_WAIT))
6988 fPendingDbgXcpt = false;
6989 }
6990 }
6991
6992 NOREF(pszInstr);
6993 return fPendingDbgXcpt;
6994}
6995
6996
6997/**
6998 * Set up the monitor-trap flag (MTF).
6999 *
7000 * @param pVCpu The cross context virtual CPU structure.
7001 * @param pszInstr The VMX instruction name (for logging purposes).
7002 */
7003IEM_STATIC void iemVmxVmentrySetupMtf(PVMCPUCC pVCpu, const char *pszInstr)
7004{
7005 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
7006 Assert(pVmcs);
7007 if (pVmcs->u32ProcCtls & VMX_PROC_CTLS_MONITOR_TRAP_FLAG)
7008 {
7009 VMCPU_FF_SET(pVCpu, VMCPU_FF_VMX_MTF);
7010 Log(("%s: Monitor-trap flag set on VM-entry\n", pszInstr));
7011 }
7012 else
7013 Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_VMX_MTF));
7014 NOREF(pszInstr);
7015}
7016
7017
7018/**
7019 * Sets up NMI-window exiting.
7020 *
7021 * @param pVCpu The cross context virtual CPU structure.
7022 * @param pszInstr The VMX instruction name (for logging purposes).
7023 */
7024IEM_STATIC void iemVmxVmentrySetupNmiWindow(PVMCPUCC pVCpu, const char *pszInstr)
7025{
7026 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
7027 Assert(pVmcs);
7028 if (pVmcs->u32ProcCtls & VMX_PROC_CTLS_NMI_WINDOW_EXIT)
7029 {
7030 Assert(pVmcs->u32PinCtls & VMX_PIN_CTLS_VIRT_NMI);
7031 VMCPU_FF_SET(pVCpu, VMCPU_FF_VMX_NMI_WINDOW);
7032 Log(("%s: NMI-window set on VM-entry\n", pszInstr));
7033 }
7034 else
7035 Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_VMX_NMI_WINDOW));
7036 NOREF(pszInstr);
7037}
7038
7039
7040/**
7041 * Sets up interrupt-window exiting.
7042 *
7043 * @param pVCpu The cross context virtual CPU structure.
7044 * @param pszInstr The VMX instruction name (for logging purposes).
7045 */
7046IEM_STATIC void iemVmxVmentrySetupIntWindow(PVMCPUCC pVCpu, const char *pszInstr)
7047{
7048 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
7049 Assert(pVmcs);
7050 if (pVmcs->u32ProcCtls & VMX_PROC_CTLS_INT_WINDOW_EXIT)
7051 {
7052 VMCPU_FF_SET(pVCpu, VMCPU_FF_VMX_INT_WINDOW);
7053 Log(("%s: Interrupt-window set on VM-entry\n", pszInstr));
7054 }
7055 else
7056 Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_VMX_INT_WINDOW));
7057 NOREF(pszInstr);
7058}
7059
7060
7061/**
7062 * Set up the VMX-preemption timer.
7063 *
7064 * @param pVCpu The cross context virtual CPU structure.
7065 * @param pszInstr The VMX instruction name (for logging purposes).
7066 */
7067IEM_STATIC void iemVmxVmentrySetupPreemptTimer(PVMCPUCC pVCpu, const char *pszInstr)
7068{
7069 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
7070 Assert(pVmcs);
7071 if (pVmcs->u32PinCtls & VMX_PIN_CTLS_PREEMPT_TIMER)
7072 {
7073 /*
7074 * If the timer is 0, we must cause a VM-exit before executing the first
7075 * nested-guest instruction. So we can flag as though the timer has already
7076 * expired and we will check and cause a VM-exit at the right priority elsewhere
7077 * in the code.
7078 */
7079 uint64_t uEntryTick;
7080 uint32_t const uPreemptTimer = pVmcs->u32PreemptTimer;
7081 if (uPreemptTimer)
7082 {
7083 int rc = CPUMStartGuestVmxPremptTimer(pVCpu, uPreemptTimer, VMX_V_PREEMPT_TIMER_SHIFT, &uEntryTick);
7084 AssertRC(rc);
7085 Log(("%s: VM-entry set up VMX-preemption timer at %#RX64\n", pszInstr, uEntryTick));
7086 }
7087 else
7088 {
7089 uEntryTick = TMCpuTickGetNoCheck(pVCpu);
7090 VMCPU_FF_SET(pVCpu, VMCPU_FF_VMX_PREEMPT_TIMER);
7091 Log(("%s: VM-entry set up VMX-preemption timer at %#RX64 to expire immediately!\n", pszInstr, uEntryTick));
7092 }
7093
7094 pVCpu->cpum.GstCtx.hwvirt.vmx.uEntryTick = uEntryTick;
7095 }
7096 else
7097 Assert(!VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_VMX_PREEMPT_TIMER));
7098
7099 NOREF(pszInstr);
7100}
7101
7102
7103/**
7104 * Injects an event using TRPM given a VM-entry interruption info. and related
7105 * fields.
7106 *
7107 * @param pVCpu The cross context virtual CPU structure.
7108 * @param pszInstr The VMX instruction name (for logging purposes).
7109 * @param uEntryIntInfo The VM-entry interruption info.
7110 * @param uErrCode The error code associated with the event if any.
7111 * @param cbInstr The VM-entry instruction length (for software
7112 * interrupts and software exceptions). Pass 0
7113 * otherwise.
7114 * @param GCPtrFaultAddress The guest CR2 if this is a \#PF event.
7115 */
7116IEM_STATIC void iemVmxVmentryInjectTrpmEvent(PVMCPUCC pVCpu, const char *pszInstr, uint32_t uEntryIntInfo, uint32_t uErrCode,
7117 uint32_t cbInstr, RTGCUINTPTR GCPtrFaultAddress)
7118{
7119 Assert(VMX_ENTRY_INT_INFO_IS_VALID(uEntryIntInfo));
7120
7121 uint8_t const uType = VMX_ENTRY_INT_INFO_TYPE(uEntryIntInfo);
7122 uint8_t const uVector = VMX_ENTRY_INT_INFO_VECTOR(uEntryIntInfo);
7123 TRPMEVENT const enmTrpmEvent = HMVmxEventTypeToTrpmEventType(uEntryIntInfo);
7124
7125 Assert(uType != VMX_ENTRY_INT_INFO_TYPE_OTHER_EVENT);
7126
7127 int rc = TRPMAssertTrap(pVCpu, uVector, enmTrpmEvent);
7128 AssertRC(rc);
7129 Log(("%s: Injecting: vector=%#x type=%#x (%s)\n", pszInstr, uVector, uType, VMXGetEntryIntInfoTypeDesc(uType)));
7130
7131 if (VMX_ENTRY_INT_INFO_IS_ERROR_CODE_VALID(uEntryIntInfo))
7132 {
7133 TRPMSetErrorCode(pVCpu, uErrCode);
7134 Log(("%s: Injecting: err_code=%#x\n", pszInstr, uErrCode));
7135 }
7136
7137 if (VMX_ENTRY_INT_INFO_IS_XCPT_PF(uEntryIntInfo))
7138 {
7139 TRPMSetFaultAddress(pVCpu, GCPtrFaultAddress);
7140 Log(("%s: Injecting: fault_addr=%RGp\n", pszInstr, GCPtrFaultAddress));
7141 }
7142 else
7143 {
7144 if ( uType == VMX_ENTRY_INT_INFO_TYPE_SW_INT
7145 || uType == VMX_ENTRY_INT_INFO_TYPE_SW_XCPT
7146 || uType == VMX_ENTRY_INT_INFO_TYPE_PRIV_SW_XCPT)
7147 {
7148 TRPMSetInstrLength(pVCpu, cbInstr);
7149 Log(("%s: Injecting: instr_len=%u\n", pszInstr, cbInstr));
7150 }
7151 }
7152
7153 if (VMX_ENTRY_INT_INFO_TYPE(uEntryIntInfo) == VMX_ENTRY_INT_INFO_TYPE_PRIV_SW_XCPT)
7154 {
7155 TRPMSetTrapDueToIcebp(pVCpu);
7156 Log(("%s: Injecting: icebp\n", pszInstr));
7157 }
7158
7159 NOREF(pszInstr);
7160}
7161
7162
7163/**
7164 * Performs event injection (if any) as part of VM-entry.
7165 *
7166 * @param pVCpu The cross context virtual CPU structure.
7167 * @param pszInstr The VMX instruction name (for logging purposes).
7168 */
7169IEM_STATIC void iemVmxVmentryInjectEvent(PVMCPUCC pVCpu, const char *pszInstr)
7170{
7171 PVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
7172
7173 /*
7174 * Inject events.
7175 * The event that is going to be made pending for injection is not subject to VMX intercepts,
7176 * thus we flag ignoring of intercepts. However, recursive exceptions if any during delivery
7177 * of the current event -are- subject to intercepts, hence this flag will be flipped during
7178 * the actually delivery of this event.
7179 *
7180 * See Intel spec. 26.5 "Event Injection".
7181 */
7182 uint32_t const uEntryIntInfo = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs)->u32EntryIntInfo;
7183 bool const fEntryIntInfoValid = VMX_ENTRY_INT_INFO_IS_VALID(uEntryIntInfo);
7184
7185 CPUMSetGuestVmxInterceptEvents(&pVCpu->cpum.GstCtx, !fEntryIntInfoValid);
7186 if (fEntryIntInfoValid)
7187 {
7188 if (VMX_ENTRY_INT_INFO_TYPE(uEntryIntInfo) == VMX_ENTRY_INT_INFO_TYPE_OTHER_EVENT)
7189 {
7190 Assert(VMX_ENTRY_INT_INFO_VECTOR(uEntryIntInfo) == VMX_ENTRY_INT_INFO_VECTOR_MTF);
7191 VMCPU_FF_SET(pVCpu, VMCPU_FF_VMX_MTF);
7192 }
7193 else
7194 iemVmxVmentryInjectTrpmEvent(pVCpu, pszInstr, uEntryIntInfo, pVmcs->u32EntryXcptErrCode, pVmcs->u32EntryInstrLen,
7195 pVCpu->cpum.GstCtx.cr2);
7196
7197 /*
7198 * We need to clear the VM-entry interruption information field's valid bit on VM-exit.
7199 *
7200 * However, we do it here on VM-entry as well because while it isn't visible to guest
7201 * software until VM-exit, when and if HM looks at the VMCS to continue nested-guest
7202 * execution using hardware-assisted VMX, it will not be try to inject the event again.
7203 *
7204 * See Intel spec. 24.8.3 "VM-Entry Controls for Event Injection".
7205 */
7206 pVmcs->u32EntryIntInfo &= ~VMX_ENTRY_INT_INFO_VALID;
7207 }
7208 else
7209 {
7210 /*
7211 * Inject any pending guest debug exception.
7212 * Unlike injecting events, this #DB injection on VM-entry is subject to #DB VMX intercept.
7213 * See Intel spec. 26.6.3 "Delivery of Pending Debug Exceptions after VM Entry".
7214 */
7215 bool const fPendingDbgXcpt = iemVmxVmentryIsPendingDebugXcpt(pVCpu, pszInstr);
7216 if (fPendingDbgXcpt)
7217 {
7218 uint32_t const uDbgXcptInfo = RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VECTOR, X86_XCPT_DB)
7219 | RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_TYPE, VMX_ENTRY_INT_INFO_TYPE_HW_XCPT)
7220 | RT_BF_MAKE(VMX_BF_ENTRY_INT_INFO_VALID, 1);
7221 iemVmxVmentryInjectTrpmEvent(pVCpu, pszInstr, uDbgXcptInfo, 0 /* uErrCode */, pVmcs->u32EntryInstrLen,
7222 0 /* GCPtrFaultAddress */);
7223 }
7224 }
7225
7226 NOREF(pszInstr);
7227}
7228
7229
7230/**
7231 * Initializes all read-only VMCS fields as part of VM-entry.
7232 *
7233 * @param pVCpu The cross context virtual CPU structure.
7234 */
7235IEM_STATIC void iemVmxVmentryInitReadOnlyFields(PVMCPUCC pVCpu)
7236{
7237 /*
7238 * Any VMCS field which we do not establish on every VM-exit but may potentially
7239 * be used on the VM-exit path of a nested hypervisor -and- is not explicitly
7240 * specified to be undefined, needs to be initialized here.
7241 *
7242 * Thus, it is especially important to clear the Exit qualification field
7243 * since it must be zero for VM-exits where it is not used. Similarly, the
7244 * VM-exit interruption information field's valid bit needs to be cleared for
7245 * the same reasons.
7246 */
7247 PVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
7248 Assert(pVmcs);
7249
7250 /* 16-bit (none currently). */
7251 /* 32-bit. */
7252 pVmcs->u32RoVmInstrError = 0;
7253 pVmcs->u32RoExitReason = 0;
7254 pVmcs->u32RoExitIntInfo = 0;
7255 pVmcs->u32RoExitIntErrCode = 0;
7256 pVmcs->u32RoIdtVectoringInfo = 0;
7257 pVmcs->u32RoIdtVectoringErrCode = 0;
7258 pVmcs->u32RoExitInstrLen = 0;
7259 pVmcs->u32RoExitInstrInfo = 0;
7260
7261 /* 64-bit. */
7262 pVmcs->u64RoGuestPhysAddr.u = 0;
7263
7264 /* Natural-width. */
7265 pVmcs->u64RoExitQual.u = 0;
7266 pVmcs->u64RoIoRcx.u = 0;
7267 pVmcs->u64RoIoRsi.u = 0;
7268 pVmcs->u64RoIoRdi.u = 0;
7269 pVmcs->u64RoIoRip.u = 0;
7270 pVmcs->u64RoGuestLinearAddr.u = 0;
7271}
7272
7273
7274/**
7275 * VMLAUNCH/VMRESUME instruction execution worker.
7276 *
7277 * @returns Strict VBox status code.
7278 * @param pVCpu The cross context virtual CPU structure.
7279 * @param cbInstr The instruction length in bytes.
7280 * @param uInstrId The instruction identity (VMXINSTRID_VMLAUNCH or
7281 * VMXINSTRID_VMRESUME).
7282 *
7283 * @remarks Common VMX instruction checks are already expected to by the caller,
7284 * i.e. CR4.VMXE, Real/V86 mode, EFER/CS.L checks.
7285 */
7286IEM_STATIC VBOXSTRICTRC iemVmxVmlaunchVmresume(PVMCPUCC pVCpu, uint8_t cbInstr, VMXINSTRID uInstrId)
7287{
7288# if defined(VBOX_WITH_NESTED_HWVIRT_ONLY_IN_IEM) && !defined(IN_RING3)
7289 RT_NOREF3(pVCpu, cbInstr, uInstrId);
7290 return VINF_EM_RAW_EMULATE_INSTR;
7291# else
7292 Assert( uInstrId == VMXINSTRID_VMLAUNCH
7293 || uInstrId == VMXINSTRID_VMRESUME);
7294 const char *pszInstr = uInstrId == VMXINSTRID_VMRESUME ? "vmresume" : "vmlaunch";
7295
7296 /* Nested-guest intercept. */
7297 if (IEM_VMX_IS_NON_ROOT_MODE(pVCpu))
7298 return iemVmxVmexitInstr(pVCpu, uInstrId == VMXINSTRID_VMRESUME ? VMX_EXIT_VMRESUME : VMX_EXIT_VMLAUNCH, cbInstr);
7299
7300 Assert(IEM_VMX_IS_ROOT_MODE(pVCpu));
7301
7302 /*
7303 * Basic VM-entry checks.
7304 * The order of the CPL, current and shadow VMCS and block-by-MovSS are important.
7305 * The checks following that do not have to follow a specific order.
7306 *
7307 * See Intel spec. 26.1 "Basic VM-entry Checks".
7308 */
7309
7310 /* CPL. */
7311 if (pVCpu->iem.s.uCpl == 0)
7312 { /* likely */ }
7313 else
7314 {
7315 Log(("%s: CPL %u -> #GP(0)\n", pszInstr, pVCpu->iem.s.uCpl));
7316 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmentry_Cpl;
7317 return iemRaiseGeneralProtectionFault0(pVCpu);
7318 }
7319
7320 /* Current VMCS valid. */
7321 if (IEM_VMX_HAS_CURRENT_VMCS(pVCpu))
7322 { /* likely */ }
7323 else
7324 {
7325 Log(("%s: VMCS pointer %#RGp invalid -> VMFailInvalid\n", pszInstr, IEM_VMX_GET_CURRENT_VMCS(pVCpu)));
7326 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmentry_PtrInvalid;
7327 iemVmxVmFailInvalid(pVCpu);
7328 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
7329 return VINF_SUCCESS;
7330 }
7331
7332 /* Current VMCS is not a shadow VMCS. */
7333 if (!pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs)->u32VmcsRevId.n.fIsShadowVmcs)
7334 { /* likely */ }
7335 else
7336 {
7337 Log(("%s: VMCS pointer %#RGp is a shadow VMCS -> VMFailInvalid\n", pszInstr, IEM_VMX_GET_CURRENT_VMCS(pVCpu)));
7338 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmentry_PtrShadowVmcs;
7339 iemVmxVmFailInvalid(pVCpu);
7340 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
7341 return VINF_SUCCESS;
7342 }
7343
7344 /** @todo Distinguish block-by-MovSS from block-by-STI. Currently we
7345 * use block-by-STI here which is not quite correct. */
7346 if ( !VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS)
7347 || pVCpu->cpum.GstCtx.rip != EMGetInhibitInterruptsPC(pVCpu))
7348 { /* likely */ }
7349 else
7350 {
7351 Log(("%s: VM entry with events blocked by MOV SS -> VMFail\n", pszInstr));
7352 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmentry_BlocKMovSS;
7353 iemVmxVmFail(pVCpu, VMXINSTRERR_VMENTRY_BLOCK_MOVSS);
7354 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
7355 return VINF_SUCCESS;
7356 }
7357
7358 if (uInstrId == VMXINSTRID_VMLAUNCH)
7359 {
7360 /* VMLAUNCH with non-clear VMCS. */
7361 if (pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs)->fVmcsState == VMX_V_VMCS_LAUNCH_STATE_CLEAR)
7362 { /* likely */ }
7363 else
7364 {
7365 Log(("vmlaunch: VMLAUNCH with non-clear VMCS -> VMFail\n"));
7366 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmentry_VmcsClear;
7367 iemVmxVmFail(pVCpu, VMXINSTRERR_VMLAUNCH_NON_CLEAR_VMCS);
7368 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
7369 return VINF_SUCCESS;
7370 }
7371 }
7372 else
7373 {
7374 /* VMRESUME with non-launched VMCS. */
7375 if (pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs)->fVmcsState == VMX_V_VMCS_LAUNCH_STATE_LAUNCHED)
7376 { /* likely */ }
7377 else
7378 {
7379 Log(("vmresume: VMRESUME with non-launched VMCS -> VMFail\n"));
7380 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmentry_VmcsLaunch;
7381 iemVmxVmFail(pVCpu, VMXINSTRERR_VMRESUME_NON_LAUNCHED_VMCS);
7382 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
7383 return VINF_SUCCESS;
7384 }
7385 }
7386
7387 /*
7388 * We are allowed to cache VMCS related data structures (such as I/O bitmaps, MSR bitmaps)
7389 * while entering VMX non-root mode. We do some of this while checking VM-execution
7390 * controls. The nested hypervisor should not make assumptions and cannot expect
7391 * predictable behavior if changes to these structures are made in guest memory while
7392 * executing in VMX non-root mode. As far as VirtualBox is concerned, the guest cannot
7393 * modify them anyway as we cache them in host memory.
7394 *
7395 * See Intel spec. 24.11.4 "Software Access to Related Structures".
7396 */
7397 PVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
7398 Assert(pVmcs);
7399 Assert(IEM_VMX_HAS_CURRENT_VMCS(pVCpu));
7400
7401 int rc = iemVmxVmentryCheckCtls(pVCpu, pszInstr);
7402 if (RT_SUCCESS(rc))
7403 {
7404 rc = iemVmxVmentryCheckHostState(pVCpu, pszInstr);
7405 if (RT_SUCCESS(rc))
7406 {
7407 /*
7408 * Initialize read-only VMCS fields before VM-entry since we don't update all of them
7409 * for every VM-exit. This needs to be done before invoking a VM-exit (even those
7410 * ones that may occur during VM-entry below).
7411 */
7412 iemVmxVmentryInitReadOnlyFields(pVCpu);
7413
7414 /*
7415 * Blocking of NMIs need to be restored if VM-entry fails due to invalid-guest state.
7416 * So we save the VMCPU_FF_BLOCK_NMI force-flag here so we can restore it on
7417 * VM-exit when required.
7418 * See Intel spec. 26.7 "VM-entry Failures During or After Loading Guest State"
7419 */
7420 iemVmxVmentrySaveNmiBlockingFF(pVCpu);
7421
7422 rc = iemVmxVmentryCheckGuestState(pVCpu, pszInstr);
7423 if (RT_SUCCESS(rc))
7424 {
7425 rc = iemVmxVmentryLoadGuestState(pVCpu, pszInstr);
7426 if (RT_SUCCESS(rc))
7427 {
7428 rc = iemVmxVmentryLoadGuestAutoMsrs(pVCpu, pszInstr);
7429 if (RT_SUCCESS(rc))
7430 {
7431 Assert(rc != VINF_CPUM_R3_MSR_WRITE);
7432
7433 /* VMLAUNCH instruction must update the VMCS launch state. */
7434 if (uInstrId == VMXINSTRID_VMLAUNCH)
7435 pVmcs->fVmcsState = VMX_V_VMCS_LAUNCH_STATE_LAUNCHED;
7436
7437 /* Perform the VMX transition (PGM updates). */
7438 VBOXSTRICTRC rcStrict = iemVmxWorldSwitch(pVCpu);
7439 if (rcStrict == VINF_SUCCESS)
7440 { /* likely */ }
7441 else if (RT_SUCCESS(rcStrict))
7442 {
7443 Log3(("%s: iemVmxWorldSwitch returns %Rrc -> Setting passup status\n", pszInstr,
7444 VBOXSTRICTRC_VAL(rcStrict)));
7445 rcStrict = iemSetPassUpStatus(pVCpu, rcStrict);
7446 }
7447 else
7448 {
7449 Log3(("%s: iemVmxWorldSwitch failed! rc=%Rrc\n", pszInstr, VBOXSTRICTRC_VAL(rcStrict)));
7450 return rcStrict;
7451 }
7452
7453 /* Paranoia. */
7454 Assert(rcStrict == VINF_SUCCESS);
7455
7456 /* We've now entered nested-guest execution. */
7457 pVCpu->cpum.GstCtx.hwvirt.vmx.fInVmxNonRootMode = true;
7458
7459 /*
7460 * The priority of potential VM-exits during VM-entry is important.
7461 * The priorities of VM-exits and events are listed from highest
7462 * to lowest as follows:
7463 *
7464 * 1. Event injection.
7465 * 2. Trap on task-switch (T flag set in TSS).
7466 * 3. TPR below threshold / APIC-write.
7467 * 4. SMI, INIT.
7468 * 5. MTF exit.
7469 * 6. Debug-trap exceptions (EFLAGS.TF), pending debug exceptions.
7470 * 7. VMX-preemption timer.
7471 * 9. NMI-window exit.
7472 * 10. NMI injection.
7473 * 11. Interrupt-window exit.
7474 * 12. Virtual-interrupt injection.
7475 * 13. Interrupt injection.
7476 * 14. Process next instruction (fetch, decode, execute).
7477 */
7478
7479 /* Setup VMX-preemption timer. */
7480 iemVmxVmentrySetupPreemptTimer(pVCpu, pszInstr);
7481
7482 /* Setup monitor-trap flag. */
7483 iemVmxVmentrySetupMtf(pVCpu, pszInstr);
7484
7485 /* Setup NMI-window exiting. */
7486 iemVmxVmentrySetupNmiWindow(pVCpu, pszInstr);
7487
7488 /* Setup interrupt-window exiting. */
7489 iemVmxVmentrySetupIntWindow(pVCpu, pszInstr);
7490
7491 /*
7492 * Inject any event that the nested hypervisor wants to inject.
7493 * Note! We cannot immediately perform the event injection here as we may have
7494 * pending PGM operations to perform due to switching page tables and/or
7495 * mode.
7496 */
7497 iemVmxVmentryInjectEvent(pVCpu, pszInstr);
7498
7499# if defined(VBOX_WITH_NESTED_HWVIRT_ONLY_IN_IEM) && defined(IN_RING3)
7500 /* Reschedule to IEM-only execution of the nested-guest. */
7501 Log(("%s: Enabling IEM-only EM execution policy!\n", pszInstr));
7502 int rcSched = EMR3SetExecutionPolicy(pVCpu->CTX_SUFF(pVM)->pUVM, EMEXECPOLICY_IEM_ALL, true);
7503 if (rcSched != VINF_SUCCESS)
7504 iemSetPassUpStatus(pVCpu, rcSched);
7505# endif
7506
7507 /* Finally, done. */
7508 Log3(("%s: cs:rip=%#04x:%#RX64 cr0=%#RX64 (%#RX64) cr4=%#RX64 (%#RX64) efer=%#RX64\n",
7509 pszInstr, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip, pVCpu->cpum.GstCtx.cr0,
7510 pVmcs->u64Cr0ReadShadow.u, pVCpu->cpum.GstCtx.cr4, pVmcs->u64Cr4ReadShadow.u,
7511 pVCpu->cpum.GstCtx.msrEFER));
7512 return VINF_SUCCESS;
7513 }
7514 return iemVmxVmexit(pVCpu, VMX_EXIT_ERR_MSR_LOAD | VMX_EXIT_REASON_ENTRY_FAILED,
7515 pVmcs->u64RoExitQual.u);
7516 }
7517 }
7518 return iemVmxVmexit(pVCpu, VMX_EXIT_ERR_INVALID_GUEST_STATE | VMX_EXIT_REASON_ENTRY_FAILED,
7519 pVmcs->u64RoExitQual.u);
7520 }
7521
7522 iemVmxVmFail(pVCpu, VMXINSTRERR_VMENTRY_INVALID_HOST_STATE);
7523 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
7524 return VINF_SUCCESS;
7525 }
7526
7527 iemVmxVmFail(pVCpu, VMXINSTRERR_VMENTRY_INVALID_CTLS);
7528 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
7529 return VINF_SUCCESS;
7530# endif
7531}
7532
7533
7534/**
7535 * Checks whether an RDMSR or WRMSR instruction for the given MSR is intercepted
7536 * (causes a VM-exit) or not.
7537 *
7538 * @returns @c true if the instruction is intercepted, @c false otherwise.
7539 * @param pVCpu The cross context virtual CPU structure.
7540 * @param uExitReason The VM-exit reason (VMX_EXIT_RDMSR or
7541 * VMX_EXIT_WRMSR).
7542 * @param idMsr The MSR.
7543 */
7544IEM_STATIC bool iemVmxIsRdmsrWrmsrInterceptSet(PCVMCPU pVCpu, uint32_t uExitReason, uint32_t idMsr)
7545{
7546 Assert(IEM_VMX_IS_NON_ROOT_MODE(pVCpu));
7547 Assert( uExitReason == VMX_EXIT_RDMSR
7548 || uExitReason == VMX_EXIT_WRMSR);
7549
7550 /* Consult the MSR bitmap if the feature is supported. */
7551 PCVMXVVMCS pVmcs = pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs);
7552 Assert(pVmcs);
7553 if (pVmcs->u32ProcCtls & VMX_PROC_CTLS_USE_MSR_BITMAPS)
7554 {
7555 Assert(pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvMsrBitmap));
7556 uint32_t const fMsrpm = CPUMGetVmxMsrPermission(pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pvMsrBitmap), idMsr);
7557 if (uExitReason == VMX_EXIT_RDMSR)
7558 return RT_BOOL(fMsrpm & VMXMSRPM_EXIT_RD);
7559 return RT_BOOL(fMsrpm & VMXMSRPM_EXIT_WR);
7560 }
7561
7562 /* Without MSR bitmaps, all MSR accesses are intercepted. */
7563 return true;
7564}
7565
7566
7567/**
7568 * VMREAD instruction execution worker that does not perform any validation checks.
7569 *
7570 * Callers are expected to have performed the necessary checks and to ensure the
7571 * VMREAD will succeed.
7572 *
7573 * @param pVmcs Pointer to the virtual VMCS.
7574 * @param pu64Dst Where to write the VMCS value.
7575 * @param u64VmcsField The VMCS field.
7576 *
7577 * @remarks May be called with interrupts disabled.
7578 */
7579IEM_STATIC void iemVmxVmreadNoCheck(PCVMXVVMCS pVmcs, uint64_t *pu64Dst, uint64_t u64VmcsField)
7580{
7581 VMXVMCSFIELD VmcsField;
7582 VmcsField.u = u64VmcsField;
7583 uint8_t const uWidth = RT_BF_GET(VmcsField.u, VMX_BF_VMCSFIELD_WIDTH);
7584 uint8_t const uType = RT_BF_GET(VmcsField.u, VMX_BF_VMCSFIELD_TYPE);
7585 uint8_t const uWidthType = (uWidth << 2) | uType;
7586 uint8_t const uIndex = RT_BF_GET(VmcsField.u, VMX_BF_VMCSFIELD_INDEX);
7587 Assert(uIndex <= VMX_V_VMCS_MAX_INDEX);
7588 uint16_t const offField = g_aoffVmcsMap[uWidthType][uIndex];
7589 AssertMsg(offField < VMX_V_VMCS_SIZE, ("off=%u field=%#RX64 width=%#x type=%#x index=%#x (%u)\n", offField, u64VmcsField,
7590 uWidth, uType, uIndex, uIndex));
7591 AssertCompile(VMX_V_SHADOW_VMCS_SIZE == VMX_V_VMCS_SIZE);
7592
7593 /*
7594 * Read the VMCS component based on the field's effective width.
7595 *
7596 * The effective width is 64-bit fields adjusted to 32-bits if the access-type
7597 * indicates high bits (little endian).
7598 *
7599 * Note! The caller is responsible to trim the result and update registers
7600 * or memory locations are required. Here we just zero-extend to the largest
7601 * type (i.e. 64-bits).
7602 */
7603 uint8_t const *pbVmcs = (uint8_t const *)pVmcs;
7604 uint8_t const *pbField = pbVmcs + offField;
7605 uint8_t const uEffWidth = VMXGetVmcsFieldWidthEff(VmcsField.u);
7606 switch (uEffWidth)
7607 {
7608 case VMX_VMCSFIELD_WIDTH_64BIT:
7609 case VMX_VMCSFIELD_WIDTH_NATURAL: *pu64Dst = *(uint64_t const *)pbField; break;
7610 case VMX_VMCSFIELD_WIDTH_32BIT: *pu64Dst = *(uint32_t const *)pbField; break;
7611 case VMX_VMCSFIELD_WIDTH_16BIT: *pu64Dst = *(uint16_t const *)pbField; break;
7612 }
7613}
7614
7615
7616/**
7617 * VMREAD common (memory/register) instruction execution worker.
7618 *
7619 * @returns Strict VBox status code.
7620 * @param pVCpu The cross context virtual CPU structure.
7621 * @param cbInstr The instruction length in bytes.
7622 * @param pu64Dst Where to write the VMCS value (only updated when
7623 * VINF_SUCCESS is returned).
7624 * @param u64VmcsField The VMCS field.
7625 * @param pExitInfo Pointer to the VM-exit information. Optional, can be
7626 * NULL.
7627 */
7628IEM_STATIC VBOXSTRICTRC iemVmxVmreadCommon(PVMCPUCC pVCpu, uint8_t cbInstr, uint64_t *pu64Dst, uint64_t u64VmcsField,
7629 PCVMXVEXITINFO pExitInfo)
7630{
7631 /* Nested-guest intercept. */
7632 if ( IEM_VMX_IS_NON_ROOT_MODE(pVCpu)
7633 && CPUMIsGuestVmxVmreadVmwriteInterceptSet(pVCpu, VMX_EXIT_VMREAD, u64VmcsField))
7634 {
7635 if (pExitInfo)
7636 return iemVmxVmexitInstrWithInfo(pVCpu, pExitInfo);
7637 return iemVmxVmexitInstrNeedsInfo(pVCpu, VMX_EXIT_VMREAD, VMXINSTRID_VMREAD, cbInstr);
7638 }
7639
7640 /* CPL. */
7641 if (pVCpu->iem.s.uCpl == 0)
7642 { /* likely */ }
7643 else
7644 {
7645 Log(("vmread: CPL %u -> #GP(0)\n", pVCpu->iem.s.uCpl));
7646 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmread_Cpl;
7647 return iemRaiseGeneralProtectionFault0(pVCpu);
7648 }
7649
7650 /* VMCS pointer in root mode. */
7651 if ( !IEM_VMX_IS_ROOT_MODE(pVCpu)
7652 || IEM_VMX_HAS_CURRENT_VMCS(pVCpu))
7653 { /* likely */ }
7654 else
7655 {
7656 Log(("vmread: VMCS pointer %#RGp invalid -> VMFailInvalid\n", IEM_VMX_GET_CURRENT_VMCS(pVCpu)));
7657 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmread_PtrInvalid;
7658 iemVmxVmFailInvalid(pVCpu);
7659 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
7660 return VINF_SUCCESS;
7661 }
7662
7663 /* VMCS-link pointer in non-root mode. */
7664 if ( !IEM_VMX_IS_NON_ROOT_MODE(pVCpu)
7665 || IEM_VMX_HAS_SHADOW_VMCS(pVCpu))
7666 { /* likely */ }
7667 else
7668 {
7669 Log(("vmread: VMCS-link pointer %#RGp invalid -> VMFailInvalid\n", IEM_VMX_GET_SHADOW_VMCS(pVCpu)));
7670 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmread_LinkPtrInvalid;
7671 iemVmxVmFailInvalid(pVCpu);
7672 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
7673 return VINF_SUCCESS;
7674 }
7675
7676 /* Supported VMCS field. */
7677 if (CPUMIsGuestVmxVmcsFieldValid(pVCpu->CTX_SUFF(pVM), u64VmcsField))
7678 { /* likely */ }
7679 else
7680 {
7681 Log(("vmread: VMCS field %#RX64 invalid -> VMFail\n", u64VmcsField));
7682 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmread_FieldInvalid;
7683 pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = u64VmcsField;
7684 iemVmxVmFail(pVCpu, VMXINSTRERR_VMREAD_INVALID_COMPONENT);
7685 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
7686 return VINF_SUCCESS;
7687 }
7688
7689 /*
7690 * Reading from the current or shadow VMCS.
7691 */
7692 PCVMXVVMCS pVmcs = !IEM_VMX_IS_NON_ROOT_MODE(pVCpu)
7693 ? pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs)
7694 : pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pShadowVmcs);
7695 Assert(pVmcs);
7696 iemVmxVmreadNoCheck(pVmcs, pu64Dst, u64VmcsField);
7697 return VINF_SUCCESS;
7698}
7699
7700
7701/**
7702 * VMREAD (64-bit register) instruction execution worker.
7703 *
7704 * @returns Strict VBox status code.
7705 * @param pVCpu The cross context virtual CPU structure.
7706 * @param cbInstr The instruction length in bytes.
7707 * @param pu64Dst Where to store the VMCS field's value.
7708 * @param u64VmcsField The VMCS field.
7709 * @param pExitInfo Pointer to the VM-exit information. Optional, can be
7710 * NULL.
7711 */
7712IEM_STATIC VBOXSTRICTRC iemVmxVmreadReg64(PVMCPUCC pVCpu, uint8_t cbInstr, uint64_t *pu64Dst, uint64_t u64VmcsField,
7713 PCVMXVEXITINFO pExitInfo)
7714{
7715 VBOXSTRICTRC rcStrict = iemVmxVmreadCommon(pVCpu, cbInstr, pu64Dst, u64VmcsField, pExitInfo);
7716 if (rcStrict == VINF_SUCCESS)
7717 {
7718 iemVmxVmreadSuccess(pVCpu, cbInstr);
7719 return VINF_SUCCESS;
7720 }
7721
7722 Log(("vmread/reg: iemVmxVmreadCommon failed rc=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
7723 return rcStrict;
7724}
7725
7726
7727/**
7728 * VMREAD (32-bit register) instruction execution worker.
7729 *
7730 * @returns Strict VBox status code.
7731 * @param pVCpu The cross context virtual CPU structure.
7732 * @param cbInstr The instruction length in bytes.
7733 * @param pu32Dst Where to store the VMCS field's value.
7734 * @param u32VmcsField The VMCS field.
7735 * @param pExitInfo Pointer to the VM-exit information. Optional, can be
7736 * NULL.
7737 */
7738IEM_STATIC VBOXSTRICTRC iemVmxVmreadReg32(PVMCPUCC pVCpu, uint8_t cbInstr, uint32_t *pu32Dst, uint64_t u32VmcsField,
7739 PCVMXVEXITINFO pExitInfo)
7740{
7741 uint64_t u64Dst;
7742 VBOXSTRICTRC rcStrict = iemVmxVmreadCommon(pVCpu, cbInstr, &u64Dst, u32VmcsField, pExitInfo);
7743 if (rcStrict == VINF_SUCCESS)
7744 {
7745 *pu32Dst = u64Dst;
7746 iemVmxVmreadSuccess(pVCpu, cbInstr);
7747 return VINF_SUCCESS;
7748 }
7749
7750 Log(("vmread/reg: iemVmxVmreadCommon failed rc=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
7751 return rcStrict;
7752}
7753
7754
7755/**
7756 * VMREAD (memory) instruction execution worker.
7757 *
7758 * @returns Strict VBox status code.
7759 * @param pVCpu The cross context virtual CPU structure.
7760 * @param cbInstr The instruction length in bytes.
7761 * @param iEffSeg The effective segment register to use with @a u64Val.
7762 * Pass UINT8_MAX if it is a register access.
7763 * @param GCPtrDst The guest linear address to store the VMCS field's
7764 * value.
7765 * @param u64VmcsField The VMCS field.
7766 * @param pExitInfo Pointer to the VM-exit information. Optional, can be
7767 * NULL.
7768 */
7769IEM_STATIC VBOXSTRICTRC iemVmxVmreadMem(PVMCPUCC pVCpu, uint8_t cbInstr, uint8_t iEffSeg, RTGCPTR GCPtrDst, uint64_t u64VmcsField,
7770 PCVMXVEXITINFO pExitInfo)
7771{
7772 uint64_t u64Dst;
7773 VBOXSTRICTRC rcStrict = iemVmxVmreadCommon(pVCpu, cbInstr, &u64Dst, u64VmcsField, pExitInfo);
7774 if (rcStrict == VINF_SUCCESS)
7775 {
7776 /*
7777 * Write the VMCS field's value to the location specified in guest-memory.
7778 */
7779 if (pVCpu->iem.s.enmCpuMode == IEMMODE_64BIT)
7780 rcStrict = iemMemStoreDataU64(pVCpu, iEffSeg, GCPtrDst, u64Dst);
7781 else
7782 rcStrict = iemMemStoreDataU32(pVCpu, iEffSeg, GCPtrDst, u64Dst);
7783 if (rcStrict == VINF_SUCCESS)
7784 {
7785 iemVmxVmreadSuccess(pVCpu, cbInstr);
7786 return VINF_SUCCESS;
7787 }
7788
7789 Log(("vmread/mem: Failed to write to memory operand at %#RGv, rc=%Rrc\n", GCPtrDst, VBOXSTRICTRC_VAL(rcStrict)));
7790 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmread_PtrMap;
7791 pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPtrDst;
7792 return rcStrict;
7793 }
7794
7795 Log(("vmread/mem: iemVmxVmreadCommon failed rc=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
7796 return rcStrict;
7797}
7798
7799
7800/**
7801 * VMWRITE instruction execution worker that does not perform any validation
7802 * checks.
7803 *
7804 * Callers are expected to have performed the necessary checks and to ensure the
7805 * VMWRITE will succeed.
7806 *
7807 * @param pVmcs Pointer to the virtual VMCS.
7808 * @param u64Val The value to write.
7809 * @param u64VmcsField The VMCS field.
7810 *
7811 * @remarks May be called with interrupts disabled.
7812 */
7813IEM_STATIC void iemVmxVmwriteNoCheck(PVMXVVMCS pVmcs, uint64_t u64Val, uint64_t u64VmcsField)
7814{
7815 VMXVMCSFIELD VmcsField;
7816 VmcsField.u = u64VmcsField;
7817 uint8_t const uWidth = RT_BF_GET(VmcsField.u, VMX_BF_VMCSFIELD_WIDTH);
7818 uint8_t const uType = RT_BF_GET(VmcsField.u, VMX_BF_VMCSFIELD_TYPE);
7819 uint8_t const uWidthType = (uWidth << 2) | uType;
7820 uint8_t const uIndex = RT_BF_GET(VmcsField.u, VMX_BF_VMCSFIELD_INDEX);
7821 Assert(uIndex <= VMX_V_VMCS_MAX_INDEX);
7822 uint16_t const offField = g_aoffVmcsMap[uWidthType][uIndex];
7823 Assert(offField < VMX_V_VMCS_SIZE);
7824 AssertCompile(VMX_V_SHADOW_VMCS_SIZE == VMX_V_VMCS_SIZE);
7825
7826 /*
7827 * Write the VMCS component based on the field's effective width.
7828 *
7829 * The effective width is 64-bit fields adjusted to 32-bits if the access-type
7830 * indicates high bits (little endian).
7831 */
7832 uint8_t *pbVmcs = (uint8_t *)pVmcs;
7833 uint8_t *pbField = pbVmcs + offField;
7834 uint8_t const uEffWidth = VMXGetVmcsFieldWidthEff(VmcsField.u);
7835 switch (uEffWidth)
7836 {
7837 case VMX_VMCSFIELD_WIDTH_64BIT:
7838 case VMX_VMCSFIELD_WIDTH_NATURAL: *(uint64_t *)pbField = u64Val; break;
7839 case VMX_VMCSFIELD_WIDTH_32BIT: *(uint32_t *)pbField = u64Val; break;
7840 case VMX_VMCSFIELD_WIDTH_16BIT: *(uint16_t *)pbField = u64Val; break;
7841 }
7842}
7843
7844
7845/**
7846 * VMWRITE instruction execution worker.
7847 *
7848 * @returns Strict VBox status code.
7849 * @param pVCpu The cross context virtual CPU structure.
7850 * @param cbInstr The instruction length in bytes.
7851 * @param iEffSeg The effective segment register to use with @a u64Val.
7852 * Pass UINT8_MAX if it is a register access.
7853 * @param u64Val The value to write (or guest linear address to the
7854 * value), @a iEffSeg will indicate if it's a memory
7855 * operand.
7856 * @param u64VmcsField The VMCS field.
7857 * @param pExitInfo Pointer to the VM-exit information. Optional, can be
7858 * NULL.
7859 */
7860IEM_STATIC VBOXSTRICTRC iemVmxVmwrite(PVMCPUCC pVCpu, uint8_t cbInstr, uint8_t iEffSeg, uint64_t u64Val, uint64_t u64VmcsField,
7861 PCVMXVEXITINFO pExitInfo)
7862{
7863 /* Nested-guest intercept. */
7864 if ( IEM_VMX_IS_NON_ROOT_MODE(pVCpu)
7865 && CPUMIsGuestVmxVmreadVmwriteInterceptSet(pVCpu, VMX_EXIT_VMWRITE, u64VmcsField))
7866 {
7867 if (pExitInfo)
7868 return iemVmxVmexitInstrWithInfo(pVCpu, pExitInfo);
7869 return iemVmxVmexitInstrNeedsInfo(pVCpu, VMX_EXIT_VMWRITE, VMXINSTRID_VMWRITE, cbInstr);
7870 }
7871
7872 /* CPL. */
7873 if (pVCpu->iem.s.uCpl == 0)
7874 { /* likely */ }
7875 else
7876 {
7877 Log(("vmwrite: CPL %u -> #GP(0)\n", pVCpu->iem.s.uCpl));
7878 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmwrite_Cpl;
7879 return iemRaiseGeneralProtectionFault0(pVCpu);
7880 }
7881
7882 /* VMCS pointer in root mode. */
7883 if ( !IEM_VMX_IS_ROOT_MODE(pVCpu)
7884 || IEM_VMX_HAS_CURRENT_VMCS(pVCpu))
7885 { /* likely */ }
7886 else
7887 {
7888 Log(("vmwrite: VMCS pointer %#RGp invalid -> VMFailInvalid\n", IEM_VMX_GET_CURRENT_VMCS(pVCpu)));
7889 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmwrite_PtrInvalid;
7890 iemVmxVmFailInvalid(pVCpu);
7891 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
7892 return VINF_SUCCESS;
7893 }
7894
7895 /* VMCS-link pointer in non-root mode. */
7896 if ( !IEM_VMX_IS_NON_ROOT_MODE(pVCpu)
7897 || IEM_VMX_HAS_SHADOW_VMCS(pVCpu))
7898 { /* likely */ }
7899 else
7900 {
7901 Log(("vmwrite: VMCS-link pointer %#RGp invalid -> VMFailInvalid\n", IEM_VMX_GET_SHADOW_VMCS(pVCpu)));
7902 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmwrite_LinkPtrInvalid;
7903 iemVmxVmFailInvalid(pVCpu);
7904 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
7905 return VINF_SUCCESS;
7906 }
7907
7908 /* If the VMWRITE instruction references memory, access the specified memory operand. */
7909 bool const fIsRegOperand = iEffSeg == UINT8_MAX;
7910 if (!fIsRegOperand)
7911 {
7912 /* Read the value from the specified guest memory location. */
7913 VBOXSTRICTRC rcStrict;
7914 RTGCPTR const GCPtrVal = u64Val;
7915 if (pVCpu->iem.s.enmCpuMode == IEMMODE_64BIT)
7916 rcStrict = iemMemFetchDataU64(pVCpu, &u64Val, iEffSeg, GCPtrVal);
7917 else
7918 {
7919 uint32_t u32Val;
7920 rcStrict = iemMemFetchDataU32(pVCpu, &u32Val, iEffSeg, GCPtrVal);
7921 u64Val = u32Val;
7922 }
7923 if (RT_UNLIKELY(rcStrict != VINF_SUCCESS))
7924 {
7925 Log(("vmwrite: Failed to read value from memory operand at %#RGv, rc=%Rrc\n", GCPtrVal, VBOXSTRICTRC_VAL(rcStrict)));
7926 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmwrite_PtrMap;
7927 pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPtrVal;
7928 return rcStrict;
7929 }
7930 }
7931 else
7932 Assert(!pExitInfo || pExitInfo->InstrInfo.VmreadVmwrite.fIsRegOperand);
7933
7934 /* Supported VMCS field. */
7935 if (CPUMIsGuestVmxVmcsFieldValid(pVCpu->CTX_SUFF(pVM), u64VmcsField))
7936 { /* likely */ }
7937 else
7938 {
7939 Log(("vmwrite: VMCS field %#RX64 invalid -> VMFail\n", u64VmcsField));
7940 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmwrite_FieldInvalid;
7941 pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = u64VmcsField;
7942 iemVmxVmFail(pVCpu, VMXINSTRERR_VMWRITE_INVALID_COMPONENT);
7943 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
7944 return VINF_SUCCESS;
7945 }
7946
7947 /* Read-only VMCS field. */
7948 bool const fIsFieldReadOnly = VMXIsVmcsFieldReadOnly(u64VmcsField);
7949 if ( !fIsFieldReadOnly
7950 || IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fVmxVmwriteAll)
7951 { /* likely */ }
7952 else
7953 {
7954 Log(("vmwrite: Write to read-only VMCS component %#RX64 -> VMFail\n", u64VmcsField));
7955 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmwrite_FieldRo;
7956 pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = u64VmcsField;
7957 iemVmxVmFail(pVCpu, VMXINSTRERR_VMWRITE_RO_COMPONENT);
7958 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
7959 return VINF_SUCCESS;
7960 }
7961
7962 /*
7963 * Write to the current or shadow VMCS.
7964 */
7965 bool const fInVmxNonRootMode = IEM_VMX_IS_NON_ROOT_MODE(pVCpu);
7966 PVMXVVMCS pVmcs = !fInVmxNonRootMode
7967 ? pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs)
7968 : pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pShadowVmcs);
7969 Assert(pVmcs);
7970 iemVmxVmwriteNoCheck(pVmcs, u64Val, u64VmcsField);
7971
7972 /* Notify HM that the VMCS content might have changed. */
7973 if (!fInVmxNonRootMode)
7974 HMNotifyVmxNstGstCurrentVmcsChanged(pVCpu);
7975
7976 iemVmxVmSucceed(pVCpu);
7977 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
7978 return VINF_SUCCESS;
7979}
7980
7981
7982/**
7983 * VMCLEAR instruction execution worker.
7984 *
7985 * @returns Strict VBox status code.
7986 * @param pVCpu The cross context virtual CPU structure.
7987 * @param cbInstr The instruction length in bytes.
7988 * @param iEffSeg The effective segment register to use with @a GCPtrVmcs.
7989 * @param GCPtrVmcs The linear address of the VMCS pointer.
7990 * @param pExitInfo Pointer to the VM-exit information. Optional, can be NULL.
7991 *
7992 * @remarks Common VMX instruction checks are already expected to by the caller,
7993 * i.e. VMX operation, CR4.VMXE, Real/V86 mode, EFER/CS.L checks.
7994 */
7995IEM_STATIC VBOXSTRICTRC iemVmxVmclear(PVMCPUCC pVCpu, uint8_t cbInstr, uint8_t iEffSeg, RTGCPHYS GCPtrVmcs,
7996 PCVMXVEXITINFO pExitInfo)
7997{
7998 /* Nested-guest intercept. */
7999 if (IEM_VMX_IS_NON_ROOT_MODE(pVCpu))
8000 {
8001 if (pExitInfo)
8002 return iemVmxVmexitInstrWithInfo(pVCpu, pExitInfo);
8003 return iemVmxVmexitInstrNeedsInfo(pVCpu, VMX_EXIT_VMCLEAR, VMXINSTRID_NONE, cbInstr);
8004 }
8005
8006 Assert(IEM_VMX_IS_ROOT_MODE(pVCpu));
8007
8008 /* CPL. */
8009 if (pVCpu->iem.s.uCpl == 0)
8010 { /* likely */ }
8011 else
8012 {
8013 Log(("vmclear: CPL %u -> #GP(0)\n", pVCpu->iem.s.uCpl));
8014 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmclear_Cpl;
8015 return iemRaiseGeneralProtectionFault0(pVCpu);
8016 }
8017
8018 /* Get the VMCS pointer from the location specified by the source memory operand. */
8019 RTGCPHYS GCPhysVmcs;
8020 VBOXSTRICTRC rcStrict = iemMemFetchDataU64(pVCpu, &GCPhysVmcs, iEffSeg, GCPtrVmcs);
8021 if (RT_LIKELY(rcStrict == VINF_SUCCESS))
8022 { /* likely */ }
8023 else
8024 {
8025 Log(("vmclear: Failed to read VMCS physaddr from %#RGv, rc=%Rrc\n", GCPtrVmcs, VBOXSTRICTRC_VAL(rcStrict)));
8026 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmclear_PtrMap;
8027 pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPtrVmcs;
8028 return rcStrict;
8029 }
8030
8031 /* VMCS pointer alignment. */
8032 if (!(GCPhysVmcs & X86_PAGE_4K_OFFSET_MASK))
8033 { /* likely */ }
8034 else
8035 {
8036 Log(("vmclear: VMCS pointer not page-aligned -> VMFail()\n"));
8037 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmclear_PtrAlign;
8038 pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPhysVmcs;
8039 iemVmxVmFail(pVCpu, VMXINSTRERR_VMCLEAR_INVALID_PHYSADDR);
8040 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
8041 return VINF_SUCCESS;
8042 }
8043
8044 /* VMCS physical-address width limits. */
8045 if (!(GCPhysVmcs >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cVmxMaxPhysAddrWidth))
8046 { /* likely */ }
8047 else
8048 {
8049 Log(("vmclear: VMCS pointer extends beyond physical-address width -> VMFail()\n"));
8050 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmclear_PtrWidth;
8051 pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPhysVmcs;
8052 iemVmxVmFail(pVCpu, VMXINSTRERR_VMCLEAR_INVALID_PHYSADDR);
8053 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
8054 return VINF_SUCCESS;
8055 }
8056
8057 /* VMCS is not the VMXON region. */
8058 if (GCPhysVmcs != pVCpu->cpum.GstCtx.hwvirt.vmx.GCPhysVmxon)
8059 { /* likely */ }
8060 else
8061 {
8062 Log(("vmclear: VMCS pointer cannot be identical to VMXON region pointer -> VMFail()\n"));
8063 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmclear_PtrVmxon;
8064 pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPhysVmcs;
8065 iemVmxVmFail(pVCpu, VMXINSTRERR_VMCLEAR_VMXON_PTR);
8066 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
8067 return VINF_SUCCESS;
8068 }
8069
8070 /* Ensure VMCS is not MMIO, ROM etc. This is not an Intel requirement but a
8071 restriction imposed by our implementation. */
8072 if (PGMPhysIsGCPhysNormal(pVCpu->CTX_SUFF(pVM), GCPhysVmcs))
8073 { /* likely */ }
8074 else
8075 {
8076 Log(("vmclear: VMCS not normal memory -> VMFail()\n"));
8077 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmclear_PtrAbnormal;
8078 pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPhysVmcs;
8079 iemVmxVmFail(pVCpu, VMXINSTRERR_VMCLEAR_INVALID_PHYSADDR);
8080 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
8081 return VINF_SUCCESS;
8082 }
8083
8084 /*
8085 * VMCLEAR allows committing and clearing any valid VMCS pointer.
8086 *
8087 * If the current VMCS is the one being cleared, set its state to 'clear' and commit
8088 * to guest memory. Otherwise, set the state of the VMCS referenced in guest memory
8089 * to 'clear'.
8090 */
8091 uint8_t const fVmcsLaunchStateClear = VMX_V_VMCS_LAUNCH_STATE_CLEAR;
8092 if ( IEM_VMX_HAS_CURRENT_VMCS(pVCpu)
8093 && IEM_VMX_GET_CURRENT_VMCS(pVCpu) == GCPhysVmcs)
8094 {
8095 pVCpu->cpum.GstCtx.hwvirt.vmx.CTX_SUFF(pVmcs)->fVmcsState = fVmcsLaunchStateClear;
8096 iemVmxWriteCurrentVmcsToGstMem(pVCpu);
8097 IEM_VMX_CLEAR_CURRENT_VMCS(pVCpu);
8098 }
8099 else
8100 {
8101 AssertCompileMemberSize(VMXVVMCS, fVmcsState, sizeof(fVmcsLaunchStateClear));
8102 rcStrict = PGMPhysSimpleWriteGCPhys(pVCpu->CTX_SUFF(pVM), GCPhysVmcs + RT_UOFFSETOF(VMXVVMCS, fVmcsState),
8103 (const void *)&fVmcsLaunchStateClear, sizeof(fVmcsLaunchStateClear));
8104 if (RT_FAILURE(rcStrict))
8105 return rcStrict;
8106 }
8107
8108 iemVmxVmSucceed(pVCpu);
8109 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
8110 return VINF_SUCCESS;
8111}
8112
8113
8114/**
8115 * VMPTRST instruction execution worker.
8116 *
8117 * @returns Strict VBox status code.
8118 * @param pVCpu The cross context virtual CPU structure.
8119 * @param cbInstr The instruction length in bytes.
8120 * @param iEffSeg The effective segment register to use with @a GCPtrVmcs.
8121 * @param GCPtrVmcs The linear address of where to store the current VMCS
8122 * pointer.
8123 * @param pExitInfo Pointer to the VM-exit information. Optional, can be NULL.
8124 *
8125 * @remarks Common VMX instruction checks are already expected to by the caller,
8126 * i.e. VMX operation, CR4.VMXE, Real/V86 mode, EFER/CS.L checks.
8127 */
8128IEM_STATIC VBOXSTRICTRC iemVmxVmptrst(PVMCPUCC pVCpu, uint8_t cbInstr, uint8_t iEffSeg, RTGCPHYS GCPtrVmcs,
8129 PCVMXVEXITINFO pExitInfo)
8130{
8131 /* Nested-guest intercept. */
8132 if (IEM_VMX_IS_NON_ROOT_MODE(pVCpu))
8133 {
8134 if (pExitInfo)
8135 return iemVmxVmexitInstrWithInfo(pVCpu, pExitInfo);
8136 return iemVmxVmexitInstrNeedsInfo(pVCpu, VMX_EXIT_VMPTRST, VMXINSTRID_NONE, cbInstr);
8137 }
8138
8139 Assert(IEM_VMX_IS_ROOT_MODE(pVCpu));
8140
8141 /* CPL. */
8142 if (pVCpu->iem.s.uCpl == 0)
8143 { /* likely */ }
8144 else
8145 {
8146 Log(("vmptrst: CPL %u -> #GP(0)\n", pVCpu->iem.s.uCpl));
8147 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmptrst_Cpl;
8148 return iemRaiseGeneralProtectionFault0(pVCpu);
8149 }
8150
8151 /* Set the VMCS pointer to the location specified by the destination memory operand. */
8152 AssertCompile(NIL_RTGCPHYS == ~(RTGCPHYS)0U);
8153 VBOXSTRICTRC rcStrict = iemMemStoreDataU64(pVCpu, iEffSeg, GCPtrVmcs, IEM_VMX_GET_CURRENT_VMCS(pVCpu));
8154 if (RT_LIKELY(rcStrict == VINF_SUCCESS))
8155 {
8156 iemVmxVmSucceed(pVCpu);
8157 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
8158 return rcStrict;
8159 }
8160
8161 Log(("vmptrst: Failed to store VMCS pointer to memory at destination operand %#Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
8162 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmptrst_PtrMap;
8163 pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPtrVmcs;
8164 return rcStrict;
8165}
8166
8167
8168/**
8169 * VMPTRLD instruction execution worker.
8170 *
8171 * @returns Strict VBox status code.
8172 * @param pVCpu The cross context virtual CPU structure.
8173 * @param cbInstr The instruction length in bytes.
8174 * @param GCPtrVmcs The linear address of the current VMCS pointer.
8175 * @param pExitInfo Pointer to the VM-exit information. Optional, can be NULL.
8176 *
8177 * @remarks Common VMX instruction checks are already expected to by the caller,
8178 * i.e. VMX operation, CR4.VMXE, Real/V86 mode, EFER/CS.L checks.
8179 */
8180IEM_STATIC VBOXSTRICTRC iemVmxVmptrld(PVMCPUCC pVCpu, uint8_t cbInstr, uint8_t iEffSeg, RTGCPHYS GCPtrVmcs,
8181 PCVMXVEXITINFO pExitInfo)
8182{
8183 /* Nested-guest intercept. */
8184 if (IEM_VMX_IS_NON_ROOT_MODE(pVCpu))
8185 {
8186 if (pExitInfo)
8187 return iemVmxVmexitInstrWithInfo(pVCpu, pExitInfo);
8188 return iemVmxVmexitInstrNeedsInfo(pVCpu, VMX_EXIT_VMPTRLD, VMXINSTRID_NONE, cbInstr);
8189 }
8190
8191 Assert(IEM_VMX_IS_ROOT_MODE(pVCpu));
8192
8193 /* CPL. */
8194 if (pVCpu->iem.s.uCpl == 0)
8195 { /* likely */ }
8196 else
8197 {
8198 Log(("vmptrld: CPL %u -> #GP(0)\n", pVCpu->iem.s.uCpl));
8199 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmptrld_Cpl;
8200 return iemRaiseGeneralProtectionFault0(pVCpu);
8201 }
8202
8203 /* Get the VMCS pointer from the location specified by the source memory operand. */
8204 RTGCPHYS GCPhysVmcs;
8205 VBOXSTRICTRC rcStrict = iemMemFetchDataU64(pVCpu, &GCPhysVmcs, iEffSeg, GCPtrVmcs);
8206 if (RT_LIKELY(rcStrict == VINF_SUCCESS))
8207 { /* likely */ }
8208 else
8209 {
8210 Log(("vmptrld: Failed to read VMCS physaddr from %#RGv, rc=%Rrc\n", GCPtrVmcs, VBOXSTRICTRC_VAL(rcStrict)));
8211 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmptrld_PtrMap;
8212 pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPtrVmcs;
8213 return rcStrict;
8214 }
8215
8216 /* VMCS pointer alignment. */
8217 if (!(GCPhysVmcs & X86_PAGE_4K_OFFSET_MASK))
8218 { /* likely */ }
8219 else
8220 {
8221 Log(("vmptrld: VMCS pointer not page-aligned -> VMFail()\n"));
8222 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmptrld_PtrAlign;
8223 pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPhysVmcs;
8224 iemVmxVmFail(pVCpu, VMXINSTRERR_VMPTRLD_INVALID_PHYSADDR);
8225 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
8226 return VINF_SUCCESS;
8227 }
8228
8229 /* VMCS physical-address width limits. */
8230 if (!(GCPhysVmcs >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cVmxMaxPhysAddrWidth))
8231 { /* likely */ }
8232 else
8233 {
8234 Log(("vmptrld: VMCS pointer extends beyond physical-address width -> VMFail()\n"));
8235 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmptrld_PtrWidth;
8236 pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPhysVmcs;
8237 iemVmxVmFail(pVCpu, VMXINSTRERR_VMPTRLD_INVALID_PHYSADDR);
8238 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
8239 return VINF_SUCCESS;
8240 }
8241
8242 /* VMCS is not the VMXON region. */
8243 if (GCPhysVmcs != pVCpu->cpum.GstCtx.hwvirt.vmx.GCPhysVmxon)
8244 { /* likely */ }
8245 else
8246 {
8247 Log(("vmptrld: VMCS pointer cannot be identical to VMXON region pointer -> VMFail()\n"));
8248 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmptrld_PtrVmxon;
8249 pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPhysVmcs;
8250 iemVmxVmFail(pVCpu, VMXINSTRERR_VMPTRLD_VMXON_PTR);
8251 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
8252 return VINF_SUCCESS;
8253 }
8254
8255 /* Ensure VMCS is not MMIO, ROM etc. This is not an Intel requirement but a
8256 restriction imposed by our implementation. */
8257 if (PGMPhysIsGCPhysNormal(pVCpu->CTX_SUFF(pVM), GCPhysVmcs))
8258 { /* likely */ }
8259 else
8260 {
8261 Log(("vmptrld: VMCS not normal memory -> VMFail()\n"));
8262 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmptrld_PtrAbnormal;
8263 pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPhysVmcs;
8264 iemVmxVmFail(pVCpu, VMXINSTRERR_VMPTRLD_INVALID_PHYSADDR);
8265 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
8266 return VINF_SUCCESS;
8267 }
8268
8269 /* Read just the VMCS revision from the VMCS. */
8270 VMXVMCSREVID VmcsRevId;
8271 int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), &VmcsRevId, GCPhysVmcs, sizeof(VmcsRevId));
8272 if (RT_SUCCESS(rc))
8273 { /* likely */ }
8274 else
8275 {
8276 Log(("vmptrld: Failed to read revision identifier from VMCS at %#RGp, rc=%Rrc\n", GCPhysVmcs, rc));
8277 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmptrld_RevPtrReadPhys;
8278 pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPhysVmcs;
8279 return rc;
8280 }
8281
8282 /*
8283 * Verify the VMCS revision specified by the guest matches what we reported to the guest.
8284 * Verify the VMCS is not a shadow VMCS, if the VMCS shadowing feature is supported.
8285 */
8286 if ( VmcsRevId.n.u31RevisionId == VMX_V_VMCS_REVISION_ID
8287 && ( !VmcsRevId.n.fIsShadowVmcs
8288 || IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fVmxVmcsShadowing))
8289 { /* likely */ }
8290 else
8291 {
8292 if (VmcsRevId.n.u31RevisionId != VMX_V_VMCS_REVISION_ID)
8293 {
8294 Log(("vmptrld: VMCS revision mismatch, expected %#RX32 got %#RX32, GCPtrVmcs=%#RGv GCPhysVmcs=%#RGp -> VMFail()\n",
8295 VMX_V_VMCS_REVISION_ID, VmcsRevId.n.u31RevisionId, GCPtrVmcs, GCPhysVmcs));
8296 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmptrld_VmcsRevId;
8297 iemVmxVmFail(pVCpu, VMXINSTRERR_VMPTRLD_INCORRECT_VMCS_REV);
8298 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
8299 return VINF_SUCCESS;
8300 }
8301
8302 Log(("vmptrld: Shadow VMCS -> VMFail()\n"));
8303 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmptrld_ShadowVmcs;
8304 iemVmxVmFail(pVCpu, VMXINSTRERR_VMPTRLD_INCORRECT_VMCS_REV);
8305 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
8306 return VINF_SUCCESS;
8307 }
8308
8309 /*
8310 * We cache only the current VMCS in CPUMCTX. Therefore, VMPTRLD should always flush
8311 * the cache of an existing, current VMCS back to guest memory before loading a new,
8312 * different current VMCS.
8313 */
8314 if (IEM_VMX_GET_CURRENT_VMCS(pVCpu) != GCPhysVmcs)
8315 {
8316 if (IEM_VMX_HAS_CURRENT_VMCS(pVCpu))
8317 {
8318 iemVmxWriteCurrentVmcsToGstMem(pVCpu);
8319 IEM_VMX_CLEAR_CURRENT_VMCS(pVCpu);
8320 }
8321
8322 /* Set the new VMCS as the current VMCS and read it from guest memory. */
8323 IEM_VMX_SET_CURRENT_VMCS(pVCpu, GCPhysVmcs);
8324 rc = iemVmxReadCurrentVmcsFromGstMem(pVCpu);
8325 if (RT_SUCCESS(rc))
8326 {
8327 /* Notify HM that a new, current VMCS is loaded. */
8328 HMNotifyVmxNstGstCurrentVmcsChanged(pVCpu);
8329 }
8330 else
8331 {
8332 Log(("vmptrld: Failed to read VMCS at %#RGp, rc=%Rrc\n", GCPhysVmcs, rc));
8333 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmptrld_PtrReadPhys;
8334 pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPhysVmcs;
8335 return rc;
8336 }
8337 }
8338
8339 Assert(IEM_VMX_HAS_CURRENT_VMCS(pVCpu));
8340 iemVmxVmSucceed(pVCpu);
8341 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
8342 return VINF_SUCCESS;
8343}
8344
8345
8346/**
8347 * INVVPID instruction execution worker.
8348 *
8349 * @returns Strict VBox status code.
8350 * @param pVCpu The cross context virtual CPU structure.
8351 * @param cbInstr The instruction length in bytes.
8352 * @param iEffSeg The segment of the invvpid descriptor.
8353 * @param GCPtrInvvpidDesc The address of invvpid descriptor.
8354 * @param u64InvvpidType The invalidation type.
8355 * @param pExitInfo Pointer to the VM-exit information. Optional, can be
8356 * NULL.
8357 *
8358 * @remarks Common VMX instruction checks are already expected to by the caller,
8359 * i.e. VMX operation, CR4.VMXE, Real/V86 mode, EFER/CS.L checks.
8360 */
8361IEM_STATIC VBOXSTRICTRC iemVmxInvvpid(PVMCPUCC pVCpu, uint8_t cbInstr, uint8_t iEffSeg, RTGCPTR GCPtrInvvpidDesc,
8362 uint64_t u64InvvpidType, PCVMXVEXITINFO pExitInfo)
8363{
8364 /* Check if INVVPID instruction is supported, otherwise raise #UD. */
8365 if (!IEM_GET_GUEST_CPU_FEATURES(pVCpu)->fVmxVpid)
8366 return iemRaiseUndefinedOpcode(pVCpu);
8367
8368 /* Nested-guest intercept. */
8369 if (IEM_VMX_IS_NON_ROOT_MODE(pVCpu))
8370 {
8371 if (pExitInfo)
8372 return iemVmxVmexitInstrWithInfo(pVCpu, pExitInfo);
8373 return iemVmxVmexitInstrNeedsInfo(pVCpu, VMX_EXIT_INVVPID, VMXINSTRID_NONE, cbInstr);
8374 }
8375
8376 /* CPL. */
8377 if (pVCpu->iem.s.uCpl != 0)
8378 {
8379 Log(("invvpid: CPL != 0 -> #GP(0)\n"));
8380 return iemRaiseGeneralProtectionFault0(pVCpu);
8381 }
8382
8383 /*
8384 * Validate INVVPID invalidation type.
8385 *
8386 * The instruction specifies exactly ONE of the supported invalidation types.
8387 *
8388 * Each of the types has a bit in IA32_VMX_EPT_VPID_CAP MSR specifying if it is
8389 * supported. In theory, it's possible for a CPU to not support flushing individual
8390 * addresses but all the other types or any other combination. We do not take any
8391 * shortcuts here by assuming the types we currently expose to the guest.
8392 */
8393 uint64_t const fCaps = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64EptVpidCaps;
8394 uint8_t const fTypeIndivAddr = RT_BF_GET(fCaps, VMX_BF_EPT_VPID_CAP_INVVPID_INDIV_ADDR);
8395 uint8_t const fTypeSingleCtx = RT_BF_GET(fCaps, VMX_BF_EPT_VPID_CAP_INVVPID_SINGLE_CTX);
8396 uint8_t const fTypeAllCtx = RT_BF_GET(fCaps, VMX_BF_EPT_VPID_CAP_INVVPID_ALL_CTX);
8397 uint8_t const fTypeSingleCtxRetainGlobals = RT_BF_GET(fCaps, VMX_BF_EPT_VPID_CAP_INVVPID_SINGLE_CTX_RETAIN_GLOBALS);
8398 if ( (fTypeIndivAddr && u64InvvpidType == VMXTLBFLUSHVPID_INDIV_ADDR)
8399 || (fTypeSingleCtx && u64InvvpidType == VMXTLBFLUSHVPID_SINGLE_CONTEXT)
8400 || (fTypeAllCtx && u64InvvpidType == VMXTLBFLUSHVPID_ALL_CONTEXTS)
8401 || (fTypeSingleCtxRetainGlobals && u64InvvpidType == VMXTLBFLUSHVPID_SINGLE_CONTEXT_RETAIN_GLOBALS))
8402 { /* likely */ }
8403 else
8404 {
8405 Log(("invvpid: invalid/unsupported invvpid type %#x -> VMFail\n", u64InvvpidType));
8406 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Invvpid_TypeInvalid;
8407 pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = u64InvvpidType;
8408 iemVmxVmFail(pVCpu, VMXINSTRERR_INVEPT_INVVPID_INVALID_OPERAND);
8409 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
8410 return VINF_SUCCESS;
8411 }
8412
8413 /*
8414 * Fetch the invvpid descriptor from guest memory.
8415 */
8416 RTUINT128U uDesc;
8417 VBOXSTRICTRC rcStrict = iemMemFetchDataU128(pVCpu, &uDesc, iEffSeg, GCPtrInvvpidDesc);
8418 if (rcStrict == VINF_SUCCESS)
8419 {
8420 /*
8421 * Validate the descriptor.
8422 */
8423 if (uDesc.s.Lo > 0xfff)
8424 {
8425 Log(("invvpid: reserved bits set in invvpid descriptor %#RX64 -> #GP(0)\n", uDesc.s.Lo));
8426 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Invvpid_DescRsvd;
8427 pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = uDesc.s.Lo;
8428 iemVmxVmFail(pVCpu, VMXINSTRERR_INVEPT_INVVPID_INVALID_OPERAND);
8429 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
8430 return VINF_SUCCESS;
8431 }
8432
8433 IEM_CTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR3);
8434 RTGCUINTPTR64 const GCPtrInvAddr = uDesc.s.Hi;
8435 uint8_t const uVpid = uDesc.s.Lo & UINT64_C(0xfff);
8436 uint64_t const uCr3 = pVCpu->cpum.GstCtx.cr3;
8437 switch (u64InvvpidType)
8438 {
8439 case VMXTLBFLUSHVPID_INDIV_ADDR:
8440 {
8441 if (uVpid != 0)
8442 {
8443 if (IEM_IS_CANONICAL(GCPtrInvAddr))
8444 {
8445 /* Invalidate mappings for the linear address tagged with VPID. */
8446 /** @todo PGM support for VPID? Currently just flush everything. */
8447 PGMFlushTLB(pVCpu, uCr3, true /* fGlobal */);
8448 iemVmxVmSucceed(pVCpu);
8449 }
8450 else
8451 {
8452 Log(("invvpid: invalidation address %#RGP is not canonical -> VMFail\n", GCPtrInvAddr));
8453 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Invvpid_Type0InvalidAddr;
8454 pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPtrInvAddr;
8455 iemVmxVmFail(pVCpu, VMXINSTRERR_INVEPT_INVVPID_INVALID_OPERAND);
8456 }
8457 }
8458 else
8459 {
8460 Log(("invvpid: invalid VPID %#x for invalidation type %u -> VMFail\n", uVpid, u64InvvpidType));
8461 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Invvpid_Type0InvalidVpid;
8462 pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = u64InvvpidType;
8463 iemVmxVmFail(pVCpu, VMXINSTRERR_INVEPT_INVVPID_INVALID_OPERAND);
8464 }
8465 break;
8466 }
8467
8468 case VMXTLBFLUSHVPID_SINGLE_CONTEXT:
8469 {
8470 if (uVpid != 0)
8471 {
8472 /* Invalidate all mappings with VPID. */
8473 /** @todo PGM support for VPID? Currently just flush everything. */
8474 PGMFlushTLB(pVCpu, uCr3, true /* fGlobal */);
8475 iemVmxVmSucceed(pVCpu);
8476 }
8477 else
8478 {
8479 Log(("invvpid: invalid VPID %#x for invalidation type %u -> VMFail\n", uVpid, u64InvvpidType));
8480 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Invvpid_Type1InvalidVpid;
8481 pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = u64InvvpidType;
8482 iemVmxVmFail(pVCpu, VMXINSTRERR_INVEPT_INVVPID_INVALID_OPERAND);
8483 }
8484 break;
8485 }
8486
8487 case VMXTLBFLUSHVPID_ALL_CONTEXTS:
8488 {
8489 /* Invalidate all mappings with non-zero VPIDs. */
8490 /** @todo PGM support for VPID? Currently just flush everything. */
8491 PGMFlushTLB(pVCpu, uCr3, true /* fGlobal */);
8492 iemVmxVmSucceed(pVCpu);
8493 break;
8494 }
8495
8496 case VMXTLBFLUSHVPID_SINGLE_CONTEXT_RETAIN_GLOBALS:
8497 {
8498 if (uVpid != 0)
8499 {
8500 /* Invalidate all mappings with VPID except global translations. */
8501 /** @todo PGM support for VPID? Currently just flush everything. */
8502 PGMFlushTLB(pVCpu, uCr3, true /* fGlobal */);
8503 iemVmxVmSucceed(pVCpu);
8504 }
8505 else
8506 {
8507 Log(("invvpid: invalid VPID %#x for invalidation type %u -> VMFail\n", uVpid, u64InvvpidType));
8508 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Invvpid_Type3InvalidVpid;
8509 pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = uVpid;
8510 iemVmxVmFail(pVCpu, VMXINSTRERR_INVEPT_INVVPID_INVALID_OPERAND);
8511 }
8512 break;
8513 }
8514 IEM_NOT_REACHED_DEFAULT_CASE_RET();
8515 }
8516 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
8517 }
8518 return rcStrict;
8519}
8520
8521
8522/**
8523 * VMXON instruction execution worker.
8524 *
8525 * @returns Strict VBox status code.
8526 * @param pVCpu The cross context virtual CPU structure.
8527 * @param cbInstr The instruction length in bytes.
8528 * @param iEffSeg The effective segment register to use with @a
8529 * GCPtrVmxon.
8530 * @param GCPtrVmxon The linear address of the VMXON pointer.
8531 * @param pExitInfo Pointer to the VM-exit information. Optional, can be NULL.
8532 *
8533 * @remarks Common VMX instruction checks are already expected to by the caller,
8534 * i.e. CR4.VMXE, Real/V86 mode, EFER/CS.L checks.
8535 */
8536IEM_STATIC VBOXSTRICTRC iemVmxVmxon(PVMCPUCC pVCpu, uint8_t cbInstr, uint8_t iEffSeg, RTGCPHYS GCPtrVmxon,
8537 PCVMXVEXITINFO pExitInfo)
8538{
8539 if (!IEM_VMX_IS_ROOT_MODE(pVCpu))
8540 {
8541 /* CPL. */
8542 if (pVCpu->iem.s.uCpl == 0)
8543 { /* likely */ }
8544 else
8545 {
8546 Log(("vmxon: CPL %u -> #GP(0)\n", pVCpu->iem.s.uCpl));
8547 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_Cpl;
8548 return iemRaiseGeneralProtectionFault0(pVCpu);
8549 }
8550
8551 /* A20M (A20 Masked) mode. */
8552 if (PGMPhysIsA20Enabled(pVCpu))
8553 { /* likely */ }
8554 else
8555 {
8556 Log(("vmxon: A20M mode -> #GP(0)\n"));
8557 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_A20M;
8558 return iemRaiseGeneralProtectionFault0(pVCpu);
8559 }
8560
8561 /* CR0. */
8562 {
8563 /* CR0 MB1 bits. */
8564 uint64_t const uCr0Fixed0 = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Cr0Fixed0;
8565 if ((pVCpu->cpum.GstCtx.cr0 & uCr0Fixed0) == uCr0Fixed0)
8566 { /* likely */ }
8567 else
8568 {
8569 Log(("vmxon: CR0 fixed0 bits cleared -> #GP(0)\n"));
8570 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_Cr0Fixed0;
8571 return iemRaiseGeneralProtectionFault0(pVCpu);
8572 }
8573
8574 /* CR0 MBZ bits. */
8575 uint64_t const uCr0Fixed1 = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Cr0Fixed1;
8576 if (!(pVCpu->cpum.GstCtx.cr0 & ~uCr0Fixed1))
8577 { /* likely */ }
8578 else
8579 {
8580 Log(("vmxon: CR0 fixed1 bits set -> #GP(0)\n"));
8581 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_Cr0Fixed1;
8582 return iemRaiseGeneralProtectionFault0(pVCpu);
8583 }
8584 }
8585
8586 /* CR4. */
8587 {
8588 /* CR4 MB1 bits. */
8589 uint64_t const uCr4Fixed0 = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Cr4Fixed0;
8590 if ((pVCpu->cpum.GstCtx.cr4 & uCr4Fixed0) == uCr4Fixed0)
8591 { /* likely */ }
8592 else
8593 {
8594 Log(("vmxon: CR4 fixed0 bits cleared -> #GP(0)\n"));
8595 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_Cr4Fixed0;
8596 return iemRaiseGeneralProtectionFault0(pVCpu);
8597 }
8598
8599 /* CR4 MBZ bits. */
8600 uint64_t const uCr4Fixed1 = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64Cr4Fixed1;
8601 if (!(pVCpu->cpum.GstCtx.cr4 & ~uCr4Fixed1))
8602 { /* likely */ }
8603 else
8604 {
8605 Log(("vmxon: CR4 fixed1 bits set -> #GP(0)\n"));
8606 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_Cr4Fixed1;
8607 return iemRaiseGeneralProtectionFault0(pVCpu);
8608 }
8609 }
8610
8611 /* Feature control MSR's LOCK and VMXON bits. */
8612 uint64_t const uMsrFeatCtl = pVCpu->cpum.GstCtx.hwvirt.vmx.Msrs.u64FeatCtrl;
8613 if ((uMsrFeatCtl & (MSR_IA32_FEATURE_CONTROL_LOCK | MSR_IA32_FEATURE_CONTROL_VMXON))
8614 == (MSR_IA32_FEATURE_CONTROL_LOCK | MSR_IA32_FEATURE_CONTROL_VMXON))
8615 { /* likely */ }
8616 else
8617 {
8618 Log(("vmxon: Feature control lock bit or VMXON bit cleared -> #GP(0)\n"));
8619 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_MsrFeatCtl;
8620 return iemRaiseGeneralProtectionFault0(pVCpu);
8621 }
8622
8623 /* Get the VMXON pointer from the location specified by the source memory operand. */
8624 RTGCPHYS GCPhysVmxon;
8625 VBOXSTRICTRC rcStrict = iemMemFetchDataU64(pVCpu, &GCPhysVmxon, iEffSeg, GCPtrVmxon);
8626 if (RT_LIKELY(rcStrict == VINF_SUCCESS))
8627 { /* likely */ }
8628 else
8629 {
8630 Log(("vmxon: Failed to read VMXON region physaddr from %#RGv, rc=%Rrc\n", GCPtrVmxon, VBOXSTRICTRC_VAL(rcStrict)));
8631 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_PtrMap;
8632 pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPtrVmxon;
8633 return rcStrict;
8634 }
8635
8636 /* VMXON region pointer alignment. */
8637 if (!(GCPhysVmxon & X86_PAGE_4K_OFFSET_MASK))
8638 { /* likely */ }
8639 else
8640 {
8641 Log(("vmxon: VMXON region pointer not page-aligned -> VMFailInvalid\n"));
8642 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_PtrAlign;
8643 pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPhysVmxon;
8644 iemVmxVmFailInvalid(pVCpu);
8645 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
8646 return VINF_SUCCESS;
8647 }
8648
8649 /* VMXON physical-address width limits. */
8650 if (!(GCPhysVmxon >> IEM_GET_GUEST_CPU_FEATURES(pVCpu)->cVmxMaxPhysAddrWidth))
8651 { /* likely */ }
8652 else
8653 {
8654 Log(("vmxon: VMXON region pointer extends beyond physical-address width -> VMFailInvalid\n"));
8655 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_PtrWidth;
8656 pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPhysVmxon;
8657 iemVmxVmFailInvalid(pVCpu);
8658 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
8659 return VINF_SUCCESS;
8660 }
8661
8662 /* Ensure VMXON region is not MMIO, ROM etc. This is not an Intel requirement but a
8663 restriction imposed by our implementation. */
8664 if (PGMPhysIsGCPhysNormal(pVCpu->CTX_SUFF(pVM), GCPhysVmxon))
8665 { /* likely */ }
8666 else
8667 {
8668 Log(("vmxon: VMXON region not normal memory -> VMFailInvalid\n"));
8669 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_PtrAbnormal;
8670 pVCpu->cpum.GstCtx.hwvirt.vmx.uDiagAux = GCPhysVmxon;
8671 iemVmxVmFailInvalid(pVCpu);
8672 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
8673 return VINF_SUCCESS;
8674 }
8675
8676 /* Read the VMCS revision ID from the VMXON region. */
8677 VMXVMCSREVID VmcsRevId;
8678 int rc = PGMPhysSimpleReadGCPhys(pVCpu->CTX_SUFF(pVM), &VmcsRevId, GCPhysVmxon, sizeof(VmcsRevId));
8679 if (RT_SUCCESS(rc))
8680 { /* likely */ }
8681 else
8682 {
8683 Log(("vmxon: Failed to read VMXON region at %#RGp, rc=%Rrc\n", GCPhysVmxon, rc));
8684 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_PtrReadPhys;
8685 return rc;
8686 }
8687
8688 /* Verify the VMCS revision specified by the guest matches what we reported to the guest. */
8689 if (RT_LIKELY(VmcsRevId.u == VMX_V_VMCS_REVISION_ID))
8690 { /* likely */ }
8691 else
8692 {
8693 /* Revision ID mismatch. */
8694 if (!VmcsRevId.n.fIsShadowVmcs)
8695 {
8696 Log(("vmxon: VMCS revision mismatch, expected %#RX32 got %#RX32 -> VMFailInvalid\n", VMX_V_VMCS_REVISION_ID,
8697 VmcsRevId.n.u31RevisionId));
8698 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_VmcsRevId;
8699 iemVmxVmFailInvalid(pVCpu);
8700 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
8701 return VINF_SUCCESS;
8702 }
8703
8704 /* Shadow VMCS disallowed. */
8705 Log(("vmxon: Shadow VMCS -> VMFailInvalid\n"));
8706 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_ShadowVmcs;
8707 iemVmxVmFailInvalid(pVCpu);
8708 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
8709 return VINF_SUCCESS;
8710 }
8711
8712 /*
8713 * Record that we're in VMX operation, block INIT, block and disable A20M.
8714 */
8715 pVCpu->cpum.GstCtx.hwvirt.vmx.GCPhysVmxon = GCPhysVmxon;
8716 IEM_VMX_CLEAR_CURRENT_VMCS(pVCpu);
8717 pVCpu->cpum.GstCtx.hwvirt.vmx.fInVmxRootMode = true;
8718
8719 /* Clear address-range monitoring. */
8720 EMMonitorWaitClear(pVCpu);
8721 /** @todo NSTVMX: Intel PT. */
8722
8723 iemVmxVmSucceed(pVCpu);
8724 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
8725 return VINF_SUCCESS;
8726 }
8727 else if (IEM_VMX_IS_NON_ROOT_MODE(pVCpu))
8728 {
8729 /* Nested-guest intercept. */
8730 if (pExitInfo)
8731 return iemVmxVmexitInstrWithInfo(pVCpu, pExitInfo);
8732 return iemVmxVmexitInstrNeedsInfo(pVCpu, VMX_EXIT_VMXON, VMXINSTRID_NONE, cbInstr);
8733 }
8734
8735 Assert(IEM_VMX_IS_ROOT_MODE(pVCpu));
8736
8737 /* CPL. */
8738 if (pVCpu->iem.s.uCpl > 0)
8739 {
8740 Log(("vmxon: In VMX root mode: CPL %u -> #GP(0)\n", pVCpu->iem.s.uCpl));
8741 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_VmxRootCpl;
8742 return iemRaiseGeneralProtectionFault0(pVCpu);
8743 }
8744
8745 /* VMXON when already in VMX root mode. */
8746 iemVmxVmFail(pVCpu, VMXINSTRERR_VMXON_IN_VMXROOTMODE);
8747 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxon_VmxAlreadyRoot;
8748 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
8749 return VINF_SUCCESS;
8750}
8751
8752
8753/**
8754 * Implements 'VMXOFF'.
8755 *
8756 * @remarks Common VMX instruction checks are already expected to by the caller,
8757 * i.e. CR4.VMXE, Real/V86 mode, EFER/CS.L checks.
8758 */
8759IEM_CIMPL_DEF_0(iemCImpl_vmxoff)
8760{
8761 /* Nested-guest intercept. */
8762 if (IEM_VMX_IS_NON_ROOT_MODE(pVCpu))
8763 return iemVmxVmexitInstr(pVCpu, VMX_EXIT_VMXOFF, cbInstr);
8764
8765 /* CPL. */
8766 if (pVCpu->iem.s.uCpl == 0)
8767 { /* likely */ }
8768 else
8769 {
8770 Log(("vmxoff: CPL %u -> #GP(0)\n", pVCpu->iem.s.uCpl));
8771 pVCpu->cpum.GstCtx.hwvirt.vmx.enmDiag = kVmxVDiag_Vmxoff_Cpl;
8772 return iemRaiseGeneralProtectionFault0(pVCpu);
8773 }
8774
8775 /* Dual monitor treatment of SMIs and SMM. */
8776 uint64_t const fSmmMonitorCtl = CPUMGetGuestIa32SmmMonitorCtl(pVCpu);
8777 if (!(fSmmMonitorCtl & MSR_IA32_SMM_MONITOR_VALID))
8778 { /* likely */ }
8779 else
8780 {
8781 iemVmxVmFail(pVCpu, VMXINSTRERR_VMXOFF_DUAL_MON);
8782 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
8783 return VINF_SUCCESS;
8784 }
8785
8786 /* Record that we're no longer in VMX root operation, block INIT, block and disable A20M. */
8787 pVCpu->cpum.GstCtx.hwvirt.vmx.fInVmxRootMode = false;
8788 Assert(!pVCpu->cpum.GstCtx.hwvirt.vmx.fInVmxNonRootMode);
8789
8790 if (fSmmMonitorCtl & MSR_IA32_SMM_MONITOR_VMXOFF_UNBLOCK_SMI)
8791 { /** @todo NSTVMX: Unblock SMI. */ }
8792
8793 EMMonitorWaitClear(pVCpu);
8794 /** @todo NSTVMX: Unblock and enable A20M. */
8795
8796 iemVmxVmSucceed(pVCpu);
8797 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
8798 return VINF_SUCCESS;
8799}
8800
8801
8802/**
8803 * Implements 'VMXON'.
8804 */
8805IEM_CIMPL_DEF_2(iemCImpl_vmxon, uint8_t, iEffSeg, RTGCPTR, GCPtrVmxon)
8806{
8807 return iemVmxVmxon(pVCpu, cbInstr, iEffSeg, GCPtrVmxon, NULL /* pExitInfo */);
8808}
8809
8810
8811/**
8812 * Implements 'VMLAUNCH'.
8813 */
8814IEM_CIMPL_DEF_0(iemCImpl_vmlaunch)
8815{
8816 return iemVmxVmlaunchVmresume(pVCpu, cbInstr, VMXINSTRID_VMLAUNCH);
8817}
8818
8819
8820/**
8821 * Implements 'VMRESUME'.
8822 */
8823IEM_CIMPL_DEF_0(iemCImpl_vmresume)
8824{
8825 return iemVmxVmlaunchVmresume(pVCpu, cbInstr, VMXINSTRID_VMRESUME);
8826}
8827
8828
8829/**
8830 * Implements 'VMPTRLD'.
8831 */
8832IEM_CIMPL_DEF_2(iemCImpl_vmptrld, uint8_t, iEffSeg, RTGCPTR, GCPtrVmcs)
8833{
8834 return iemVmxVmptrld(pVCpu, cbInstr, iEffSeg, GCPtrVmcs, NULL /* pExitInfo */);
8835}
8836
8837
8838/**
8839 * Implements 'VMPTRST'.
8840 */
8841IEM_CIMPL_DEF_2(iemCImpl_vmptrst, uint8_t, iEffSeg, RTGCPTR, GCPtrVmcs)
8842{
8843 return iemVmxVmptrst(pVCpu, cbInstr, iEffSeg, GCPtrVmcs, NULL /* pExitInfo */);
8844}
8845
8846
8847/**
8848 * Implements 'VMCLEAR'.
8849 */
8850IEM_CIMPL_DEF_2(iemCImpl_vmclear, uint8_t, iEffSeg, RTGCPTR, GCPtrVmcs)
8851{
8852 return iemVmxVmclear(pVCpu, cbInstr, iEffSeg, GCPtrVmcs, NULL /* pExitInfo */);
8853}
8854
8855
8856/**
8857 * Implements 'VMWRITE' register.
8858 */
8859IEM_CIMPL_DEF_2(iemCImpl_vmwrite_reg, uint64_t, u64Val, uint64_t, u64VmcsField)
8860{
8861 return iemVmxVmwrite(pVCpu, cbInstr, UINT8_MAX /* iEffSeg */, u64Val, u64VmcsField, NULL /* pExitInfo */);
8862}
8863
8864
8865/**
8866 * Implements 'VMWRITE' memory.
8867 */
8868IEM_CIMPL_DEF_3(iemCImpl_vmwrite_mem, uint8_t, iEffSeg, RTGCPTR, GCPtrVal, uint32_t, u64VmcsField)
8869{
8870 return iemVmxVmwrite(pVCpu, cbInstr, iEffSeg, GCPtrVal, u64VmcsField, NULL /* pExitInfo */);
8871}
8872
8873
8874/**
8875 * Implements 'VMREAD' register (64-bit).
8876 */
8877IEM_CIMPL_DEF_2(iemCImpl_vmread_reg64, uint64_t *, pu64Dst, uint64_t, u64VmcsField)
8878{
8879 return iemVmxVmreadReg64(pVCpu, cbInstr, pu64Dst, u64VmcsField, NULL /* pExitInfo */);
8880}
8881
8882
8883/**
8884 * Implements 'VMREAD' register (32-bit).
8885 */
8886IEM_CIMPL_DEF_2(iemCImpl_vmread_reg32, uint32_t *, pu32Dst, uint32_t, u32VmcsField)
8887{
8888 return iemVmxVmreadReg32(pVCpu, cbInstr, pu32Dst, u32VmcsField, NULL /* pExitInfo */);
8889}
8890
8891
8892/**
8893 * Implements 'VMREAD' memory, 64-bit register.
8894 */
8895IEM_CIMPL_DEF_3(iemCImpl_vmread_mem_reg64, uint8_t, iEffSeg, RTGCPTR, GCPtrDst, uint32_t, u64VmcsField)
8896{
8897 return iemVmxVmreadMem(pVCpu, cbInstr, iEffSeg, GCPtrDst, u64VmcsField, NULL /* pExitInfo */);
8898}
8899
8900
8901/**
8902 * Implements 'VMREAD' memory, 32-bit register.
8903 */
8904IEM_CIMPL_DEF_3(iemCImpl_vmread_mem_reg32, uint8_t, iEffSeg, RTGCPTR, GCPtrDst, uint32_t, u32VmcsField)
8905{
8906 return iemVmxVmreadMem(pVCpu, cbInstr, iEffSeg, GCPtrDst, u32VmcsField, NULL /* pExitInfo */);
8907}
8908
8909
8910/**
8911 * Implements 'INVVPID'.
8912 */
8913IEM_CIMPL_DEF_3(iemCImpl_invvpid, uint8_t, iEffSeg, RTGCPTR, GCPtrInvvpidDesc, uint64_t, uInvvpidType)
8914{
8915 return iemVmxInvvpid(pVCpu, cbInstr, iEffSeg, GCPtrInvvpidDesc, uInvvpidType, NULL /* pExitInfo */);
8916}
8917
8918
8919/**
8920 * Implements VMX's implementation of PAUSE.
8921 */
8922IEM_CIMPL_DEF_0(iemCImpl_vmx_pause)
8923{
8924 if (IEM_VMX_IS_NON_ROOT_MODE(pVCpu))
8925 {
8926 VBOXSTRICTRC rcStrict = iemVmxVmexitInstrPause(pVCpu, cbInstr);
8927 if (rcStrict != VINF_VMX_INTERCEPT_NOT_ACTIVE)
8928 return rcStrict;
8929 }
8930
8931 /*
8932 * Outside VMX non-root operation or if the PAUSE instruction does not cause
8933 * a VM-exit, the instruction operates normally.
8934 */
8935 iemRegAddToRipAndClearRF(pVCpu, cbInstr);
8936 return VINF_SUCCESS;
8937}
8938
8939#endif /* VBOX_WITH_NESTED_HWVIRT_VMX */
8940
8941
8942/**
8943 * Implements 'VMCALL'.
8944 */
8945IEM_CIMPL_DEF_0(iemCImpl_vmcall)
8946{
8947#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
8948 /* Nested-guest intercept. */
8949 if (IEM_VMX_IS_NON_ROOT_MODE(pVCpu))
8950 return iemVmxVmexitInstr(pVCpu, VMX_EXIT_VMCALL, cbInstr);
8951#endif
8952
8953 /* Join forces with vmmcall. */
8954 return IEM_CIMPL_CALL_1(iemCImpl_Hypercall, OP_VMCALL);
8955}
8956
Note: See TracBrowser for help on using the repository browser.

© 2025 Oracle Support Privacy / Do Not Sell My Info Terms of Use Trademark Policy Automated Access Etiquette