VirtualBox

source: vbox/trunk/src/VBox/VMM/VMMR0/HMSVMR0.cpp@ 97196

Last change on this file since 97196 was 97196, checked in by vboxsync, 2 years ago

VMM/DBGF: Changed the PCPUMCTXCORE arguments to PCPUMCTX for DBGFTrap01Handler and DBGFTrap03Handler.

  • Property svn:eol-style set to native
  • Property svn:keywords set to Author Date Id Revision
File size: 380.6 KB
Line 
1/* $Id: HMSVMR0.cpp 97196 2022-10-18 10:42:52Z vboxsync $ */
2/** @file
3 * HM SVM (AMD-V) - Host Context Ring-0.
4 */
5
6/*
7 * Copyright (C) 2013-2022 Oracle and/or its affiliates.
8 *
9 * This file is part of VirtualBox base platform packages, as
10 * available from https://www.virtualbox.org.
11 *
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public License
14 * as published by the Free Software Foundation, in version 3 of the
15 * License.
16 *
17 * This program is distributed in the hope that it will be useful, but
18 * WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20 * General Public License for more details.
21 *
22 * You should have received a copy of the GNU General Public License
23 * along with this program; if not, see <https://www.gnu.org/licenses>.
24 *
25 * SPDX-License-Identifier: GPL-3.0-only
26 */
27
28
29/*********************************************************************************************************************************
30* Header Files *
31*********************************************************************************************************************************/
32#define LOG_GROUP LOG_GROUP_HM
33#define VMCPU_INCL_CPUM_GST_CTX
34#include <iprt/asm-amd64-x86.h>
35#include <iprt/thread.h>
36
37#include <VBox/vmm/pdmapi.h>
38#include <VBox/vmm/dbgf.h>
39#include <VBox/vmm/iem.h>
40#include <VBox/vmm/iom.h>
41#include <VBox/vmm/tm.h>
42#include <VBox/vmm/em.h>
43#include <VBox/vmm/gcm.h>
44#include <VBox/vmm/gim.h>
45#include <VBox/vmm/apic.h>
46#include "HMInternal.h"
47#include <VBox/vmm/vmcc.h>
48#include <VBox/err.h>
49#include "HMSVMR0.h"
50#include "dtrace/VBoxVMM.h"
51
52#ifdef DEBUG_ramshankar
53# define HMSVM_SYNC_FULL_GUEST_STATE
54# define HMSVM_ALWAYS_TRAP_ALL_XCPTS
55# define HMSVM_ALWAYS_TRAP_PF
56# define HMSVM_ALWAYS_TRAP_TASK_SWITCH
57#endif
58
59
60/*********************************************************************************************************************************
61* Defined Constants And Macros *
62*********************************************************************************************************************************/
63#ifdef VBOX_WITH_STATISTICS
64# define HMSVM_EXITCODE_STAM_COUNTER_INC(u64ExitCode) do { \
65 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitAll); \
66 if ((u64ExitCode) == SVM_EXIT_NPF) \
67 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitReasonNpf); \
68 else \
69 STAM_COUNTER_INC(&pVCpu->hm.s.aStatExitReason[(u64ExitCode) & MASK_EXITREASON_STAT]); \
70 } while (0)
71
72# define HMSVM_DEBUG_EXITCODE_STAM_COUNTER_INC(u64ExitCode) do { \
73 STAM_COUNTER_INC(&pVCpu->hm.s.StatDebugExitAll); \
74 if ((u64ExitCode) == SVM_EXIT_NPF) \
75 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitReasonNpf); \
76 else \
77 STAM_COUNTER_INC(&pVCpu->hm.s.aStatExitReason[(u64ExitCode) & MASK_EXITREASON_STAT]); \
78 } while (0)
79
80# define HMSVM_NESTED_EXITCODE_STAM_COUNTER_INC(u64ExitCode) do { \
81 STAM_COUNTER_INC(&pVCpu->hm.s.StatNestedExitAll); \
82 if ((u64ExitCode) == SVM_EXIT_NPF) \
83 STAM_COUNTER_INC(&pVCpu->hm.s.StatNestedExitReasonNpf); \
84 else \
85 STAM_COUNTER_INC(&pVCpu->hm.s.aStatNestedExitReason[(u64ExitCode) & MASK_EXITREASON_STAT]); \
86 } while (0)
87#else
88# define HMSVM_EXITCODE_STAM_COUNTER_INC(u64ExitCode) do { } while (0)
89# define HMSVM_DEBUG_EXITCODE_STAM_COUNTER_INC(u64ExitCode) do { } while (0)
90# define HMSVM_NESTED_EXITCODE_STAM_COUNTER_INC(u64ExitCode) do { } while (0)
91#endif /* !VBOX_WITH_STATISTICS */
92
93/** If we decide to use a function table approach this can be useful to
94 * switch to a "static DECLCALLBACK(int)". */
95#define HMSVM_EXIT_DECL static VBOXSTRICTRC
96
97/**
98 * Subset of the guest-CPU state that is kept by SVM R0 code while executing the
99 * guest using hardware-assisted SVM.
100 *
101 * This excludes state like TSC AUX, GPRs (other than RSP, RAX) which are always
102 * are swapped and restored across the world-switch and also registers like
103 * EFER, PAT MSR etc. which cannot be modified by the guest without causing a
104 * \#VMEXIT.
105 */
106#define HMSVM_CPUMCTX_EXTRN_ALL ( CPUMCTX_EXTRN_RIP \
107 | CPUMCTX_EXTRN_RFLAGS \
108 | CPUMCTX_EXTRN_RAX \
109 | CPUMCTX_EXTRN_RSP \
110 | CPUMCTX_EXTRN_SREG_MASK \
111 | CPUMCTX_EXTRN_CR0 \
112 | CPUMCTX_EXTRN_CR2 \
113 | CPUMCTX_EXTRN_CR3 \
114 | CPUMCTX_EXTRN_TABLE_MASK \
115 | CPUMCTX_EXTRN_DR6 \
116 | CPUMCTX_EXTRN_DR7 \
117 | CPUMCTX_EXTRN_KERNEL_GS_BASE \
118 | CPUMCTX_EXTRN_SYSCALL_MSRS \
119 | CPUMCTX_EXTRN_SYSENTER_MSRS \
120 | CPUMCTX_EXTRN_HWVIRT \
121 | CPUMCTX_EXTRN_INHIBIT_INT \
122 | CPUMCTX_EXTRN_HM_SVM_MASK)
123
124/**
125 * Subset of the guest-CPU state that is shared between the guest and host.
126 */
127#define HMSVM_CPUMCTX_SHARED_STATE CPUMCTX_EXTRN_DR_MASK
128
129/** Macro for importing guest state from the VMCB back into CPUMCTX. */
130#define HMSVM_CPUMCTX_IMPORT_STATE(a_pVCpu, a_fWhat) \
131 do { \
132 if ((a_pVCpu)->cpum.GstCtx.fExtrn & (a_fWhat)) \
133 hmR0SvmImportGuestState((a_pVCpu), (a_fWhat)); \
134 } while (0)
135
136/** Assert that the required state bits are fetched. */
137#define HMSVM_CPUMCTX_ASSERT(a_pVCpu, a_fExtrnMbz) AssertMsg(!((a_pVCpu)->cpum.GstCtx.fExtrn & (a_fExtrnMbz)), \
138 ("fExtrn=%#RX64 fExtrnMbz=%#RX64\n", \
139 (a_pVCpu)->cpum.GstCtx.fExtrn, (a_fExtrnMbz)))
140
141/** Assert that preemption is disabled or covered by thread-context hooks. */
142#define HMSVM_ASSERT_PREEMPT_SAFE(a_pVCpu) Assert( VMMR0ThreadCtxHookIsEnabled((a_pVCpu)) \
143 || !RTThreadPreemptIsEnabled(NIL_RTTHREAD));
144
145/** Assert that we haven't migrated CPUs when thread-context hooks are not
146 * used. */
147#define HMSVM_ASSERT_CPU_SAFE(a_pVCpu) AssertMsg( VMMR0ThreadCtxHookIsEnabled((a_pVCpu)) \
148 || (a_pVCpu)->hmr0.s.idEnteredCpu == RTMpCpuId(), \
149 ("Illegal migration! Entered on CPU %u Current %u\n", \
150 (a_pVCpu)->hmr0.s.idEnteredCpu, RTMpCpuId()));
151
152/** Assert that we're not executing a nested-guest. */
153#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
154# define HMSVM_ASSERT_NOT_IN_NESTED_GUEST(a_pCtx) Assert(!CPUMIsGuestInSvmNestedHwVirtMode((a_pCtx)))
155#else
156# define HMSVM_ASSERT_NOT_IN_NESTED_GUEST(a_pCtx) do { NOREF((a_pCtx)); } while (0)
157#endif
158
159/** Assert that we're executing a nested-guest. */
160#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
161# define HMSVM_ASSERT_IN_NESTED_GUEST(a_pCtx) Assert(CPUMIsGuestInSvmNestedHwVirtMode((a_pCtx)))
162#else
163# define HMSVM_ASSERT_IN_NESTED_GUEST(a_pCtx) do { NOREF((a_pCtx)); } while (0)
164#endif
165
166/** Macro for checking and returning from the using function for
167 * \#VMEXIT intercepts that maybe caused during delivering of another
168 * event in the guest. */
169#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
170# define HMSVM_CHECK_EXIT_DUE_TO_EVENT_DELIVERY(a_pVCpu, a_pSvmTransient) \
171 do \
172 { \
173 int rc = hmR0SvmCheckExitDueToEventDelivery((a_pVCpu), (a_pSvmTransient)); \
174 if (RT_LIKELY(rc == VINF_SUCCESS)) { /* continue #VMEXIT handling */ } \
175 else if ( rc == VINF_HM_DOUBLE_FAULT) { return VINF_SUCCESS; } \
176 else if ( rc == VINF_EM_RESET \
177 && CPUMIsGuestSvmCtrlInterceptSet((a_pVCpu), &(a_pVCpu)->cpum.GstCtx, SVM_CTRL_INTERCEPT_SHUTDOWN)) \
178 { \
179 HMSVM_CPUMCTX_IMPORT_STATE((a_pVCpu), HMSVM_CPUMCTX_EXTRN_ALL); \
180 return IEMExecSvmVmexit((a_pVCpu), SVM_EXIT_SHUTDOWN, 0, 0); \
181 } \
182 else \
183 return rc; \
184 } while (0)
185#else
186# define HMSVM_CHECK_EXIT_DUE_TO_EVENT_DELIVERY(a_pVCpu, a_pSvmTransient) \
187 do \
188 { \
189 int rc = hmR0SvmCheckExitDueToEventDelivery((a_pVCpu), (a_pSvmTransient)); \
190 if (RT_LIKELY(rc == VINF_SUCCESS)) { /* continue #VMEXIT handling */ } \
191 else if ( rc == VINF_HM_DOUBLE_FAULT) { return VINF_SUCCESS; } \
192 else \
193 return rc; \
194 } while (0)
195#endif
196
197/** Macro for upgrading a @a a_rc to VINF_EM_DBG_STEPPED after emulating an
198 * instruction that exited. */
199#define HMSVM_CHECK_SINGLE_STEP(a_pVCpu, a_rc) \
200 do { \
201 if ((a_pVCpu)->hm.s.fSingleInstruction && (a_rc) == VINF_SUCCESS) \
202 (a_rc) = VINF_EM_DBG_STEPPED; \
203 } while (0)
204
205/** Validate segment descriptor granularity bit. */
206#ifdef VBOX_STRICT
207# define HMSVM_ASSERT_SEG_GRANULARITY(a_pCtx, reg) \
208 AssertMsg( !(a_pCtx)->reg.Attr.n.u1Present \
209 || ( (a_pCtx)->reg.Attr.n.u1Granularity \
210 ? ((a_pCtx)->reg.u32Limit & 0xfff) == 0xfff \
211 : (a_pCtx)->reg.u32Limit <= UINT32_C(0xfffff)), \
212 ("Invalid Segment Attributes Limit=%#RX32 Attr=%#RX32 Base=%#RX64\n", (a_pCtx)->reg.u32Limit, \
213 (a_pCtx)->reg.Attr.u, (a_pCtx)->reg.u64Base))
214#else
215# define HMSVM_ASSERT_SEG_GRANULARITY(a_pCtx, reg) do { } while (0)
216#endif
217
218/**
219 * Exception bitmap mask for all contributory exceptions.
220 *
221 * Page fault is deliberately excluded here as it's conditional as to whether
222 * it's contributory or benign. Page faults are handled separately.
223 */
224#define HMSVM_CONTRIBUTORY_XCPT_MASK ( RT_BIT(X86_XCPT_GP) | RT_BIT(X86_XCPT_NP) | RT_BIT(X86_XCPT_SS) | RT_BIT(X86_XCPT_TS) \
225 | RT_BIT(X86_XCPT_DE))
226
227/**
228 * Mandatory/unconditional guest control intercepts.
229 *
230 * SMIs can and do happen in normal operation. We need not intercept them
231 * while executing the guest (or nested-guest).
232 */
233#define HMSVM_MANDATORY_GUEST_CTRL_INTERCEPTS ( SVM_CTRL_INTERCEPT_INTR \
234 | SVM_CTRL_INTERCEPT_NMI \
235 | SVM_CTRL_INTERCEPT_INIT \
236 | SVM_CTRL_INTERCEPT_RDPMC \
237 | SVM_CTRL_INTERCEPT_CPUID \
238 | SVM_CTRL_INTERCEPT_RSM \
239 | SVM_CTRL_INTERCEPT_HLT \
240 | SVM_CTRL_INTERCEPT_IOIO_PROT \
241 | SVM_CTRL_INTERCEPT_MSR_PROT \
242 | SVM_CTRL_INTERCEPT_INVLPGA \
243 | SVM_CTRL_INTERCEPT_SHUTDOWN \
244 | SVM_CTRL_INTERCEPT_FERR_FREEZE \
245 | SVM_CTRL_INTERCEPT_VMRUN \
246 | SVM_CTRL_INTERCEPT_SKINIT \
247 | SVM_CTRL_INTERCEPT_WBINVD \
248 | SVM_CTRL_INTERCEPT_MONITOR \
249 | SVM_CTRL_INTERCEPT_MWAIT \
250 | SVM_CTRL_INTERCEPT_CR0_SEL_WRITE \
251 | SVM_CTRL_INTERCEPT_XSETBV)
252
253/** @name VMCB Clean Bits.
254 *
255 * These flags are used for VMCB-state caching. A set VMCB Clean bit indicates
256 * AMD-V doesn't need to reload the corresponding value(s) from the VMCB in
257 * memory.
258 *
259 * @{ */
260/** All intercepts vectors, TSC offset, PAUSE filter counter. */
261#define HMSVM_VMCB_CLEAN_INTERCEPTS RT_BIT(0)
262/** I/O permission bitmap, MSR permission bitmap. */
263#define HMSVM_VMCB_CLEAN_IOPM_MSRPM RT_BIT(1)
264/** ASID. */
265#define HMSVM_VMCB_CLEAN_ASID RT_BIT(2)
266/** TRP: V_TPR, V_IRQ, V_INTR_PRIO, V_IGN_TPR, V_INTR_MASKING,
267V_INTR_VECTOR. */
268#define HMSVM_VMCB_CLEAN_INT_CTRL RT_BIT(3)
269/** Nested Paging: Nested CR3 (nCR3), PAT. */
270#define HMSVM_VMCB_CLEAN_NP RT_BIT(4)
271/** Control registers (CR0, CR3, CR4, EFER). */
272#define HMSVM_VMCB_CLEAN_CRX_EFER RT_BIT(5)
273/** Debug registers (DR6, DR7). */
274#define HMSVM_VMCB_CLEAN_DRX RT_BIT(6)
275/** GDT, IDT limit and base. */
276#define HMSVM_VMCB_CLEAN_DT RT_BIT(7)
277/** Segment register: CS, SS, DS, ES limit and base. */
278#define HMSVM_VMCB_CLEAN_SEG RT_BIT(8)
279/** CR2.*/
280#define HMSVM_VMCB_CLEAN_CR2 RT_BIT(9)
281/** Last-branch record (DbgCtlMsr, br_from, br_to, lastint_from, lastint_to) */
282#define HMSVM_VMCB_CLEAN_LBR RT_BIT(10)
283/** AVIC (AVIC APIC_BAR; AVIC APIC_BACKING_PAGE, AVIC
284PHYSICAL_TABLE and AVIC LOGICAL_TABLE Pointers). */
285#define HMSVM_VMCB_CLEAN_AVIC RT_BIT(11)
286/** Mask of all valid VMCB Clean bits. */
287#define HMSVM_VMCB_CLEAN_ALL ( HMSVM_VMCB_CLEAN_INTERCEPTS \
288 | HMSVM_VMCB_CLEAN_IOPM_MSRPM \
289 | HMSVM_VMCB_CLEAN_ASID \
290 | HMSVM_VMCB_CLEAN_INT_CTRL \
291 | HMSVM_VMCB_CLEAN_NP \
292 | HMSVM_VMCB_CLEAN_CRX_EFER \
293 | HMSVM_VMCB_CLEAN_DRX \
294 | HMSVM_VMCB_CLEAN_DT \
295 | HMSVM_VMCB_CLEAN_SEG \
296 | HMSVM_VMCB_CLEAN_CR2 \
297 | HMSVM_VMCB_CLEAN_LBR \
298 | HMSVM_VMCB_CLEAN_AVIC)
299/** @} */
300
301/**
302 * MSRPM (MSR permission bitmap) read permissions (for guest RDMSR).
303 */
304typedef enum SVMMSREXITREAD
305{
306 /** Reading this MSR causes a \#VMEXIT. */
307 SVMMSREXIT_INTERCEPT_READ = 0xb,
308 /** Reading this MSR does not cause a \#VMEXIT. */
309 SVMMSREXIT_PASSTHRU_READ
310} SVMMSREXITREAD;
311
312/**
313 * MSRPM (MSR permission bitmap) write permissions (for guest WRMSR).
314 */
315typedef enum SVMMSREXITWRITE
316{
317 /** Writing to this MSR causes a \#VMEXIT. */
318 SVMMSREXIT_INTERCEPT_WRITE = 0xd,
319 /** Writing to this MSR does not cause a \#VMEXIT. */
320 SVMMSREXIT_PASSTHRU_WRITE
321} SVMMSREXITWRITE;
322
323/**
324 * SVM \#VMEXIT handler.
325 *
326 * @returns Strict VBox status code.
327 * @param pVCpu The cross context virtual CPU structure.
328 * @param pSvmTransient Pointer to the SVM-transient structure.
329 */
330typedef VBOXSTRICTRC FNSVMEXITHANDLER(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient);
331
332
333/*********************************************************************************************************************************
334* Internal Functions *
335*********************************************************************************************************************************/
336static void hmR0SvmPendingEventToTrpmTrap(PVMCPUCC pVCpu);
337static void hmR0SvmLeave(PVMCPUCC pVCpu, bool fImportState);
338
339
340/** @name \#VMEXIT handlers.
341 * @{
342 */
343static FNSVMEXITHANDLER hmR0SvmExitIntr;
344static FNSVMEXITHANDLER hmR0SvmExitWbinvd;
345static FNSVMEXITHANDLER hmR0SvmExitInvd;
346static FNSVMEXITHANDLER hmR0SvmExitCpuid;
347static FNSVMEXITHANDLER hmR0SvmExitRdtsc;
348static FNSVMEXITHANDLER hmR0SvmExitRdtscp;
349static FNSVMEXITHANDLER hmR0SvmExitRdpmc;
350static FNSVMEXITHANDLER hmR0SvmExitInvlpg;
351static FNSVMEXITHANDLER hmR0SvmExitHlt;
352static FNSVMEXITHANDLER hmR0SvmExitMonitor;
353static FNSVMEXITHANDLER hmR0SvmExitMwait;
354static FNSVMEXITHANDLER hmR0SvmExitShutdown;
355static FNSVMEXITHANDLER hmR0SvmExitUnexpected;
356static FNSVMEXITHANDLER hmR0SvmExitReadCRx;
357static FNSVMEXITHANDLER hmR0SvmExitWriteCRx;
358static FNSVMEXITHANDLER hmR0SvmExitMsr;
359static FNSVMEXITHANDLER hmR0SvmExitReadDRx;
360static FNSVMEXITHANDLER hmR0SvmExitWriteDRx;
361static FNSVMEXITHANDLER hmR0SvmExitXsetbv;
362static FNSVMEXITHANDLER hmR0SvmExitIOInstr;
363static FNSVMEXITHANDLER hmR0SvmExitNestedPF;
364static FNSVMEXITHANDLER hmR0SvmExitVIntr;
365static FNSVMEXITHANDLER hmR0SvmExitTaskSwitch;
366static FNSVMEXITHANDLER hmR0SvmExitVmmCall;
367static FNSVMEXITHANDLER hmR0SvmExitPause;
368static FNSVMEXITHANDLER hmR0SvmExitFerrFreeze;
369static FNSVMEXITHANDLER hmR0SvmExitIret;
370static FNSVMEXITHANDLER hmR0SvmExitXcptDE;
371static FNSVMEXITHANDLER hmR0SvmExitXcptPF;
372static FNSVMEXITHANDLER hmR0SvmExitXcptUD;
373static FNSVMEXITHANDLER hmR0SvmExitXcptMF;
374static FNSVMEXITHANDLER hmR0SvmExitXcptDB;
375static FNSVMEXITHANDLER hmR0SvmExitXcptAC;
376static FNSVMEXITHANDLER hmR0SvmExitXcptBP;
377static FNSVMEXITHANDLER hmR0SvmExitXcptGP;
378static FNSVMEXITHANDLER hmR0SvmExitXcptGeneric;
379static FNSVMEXITHANDLER hmR0SvmExitSwInt;
380static FNSVMEXITHANDLER hmR0SvmExitTrRead;
381static FNSVMEXITHANDLER hmR0SvmExitTrWrite;
382#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
383static FNSVMEXITHANDLER hmR0SvmExitClgi;
384static FNSVMEXITHANDLER hmR0SvmExitStgi;
385static FNSVMEXITHANDLER hmR0SvmExitVmload;
386static FNSVMEXITHANDLER hmR0SvmExitVmsave;
387static FNSVMEXITHANDLER hmR0SvmExitInvlpga;
388static FNSVMEXITHANDLER hmR0SvmExitVmrun;
389static FNSVMEXITHANDLER hmR0SvmNestedExitXcptDB;
390static FNSVMEXITHANDLER hmR0SvmNestedExitXcptBP;
391#endif
392/** @} */
393
394static VBOXSTRICTRC hmR0SvmHandleExit(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient);
395#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
396static VBOXSTRICTRC hmR0SvmHandleExitNested(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient);
397#endif
398static VBOXSTRICTRC hmR0SvmRunGuestCodeDebug(PVMCPUCC pVCpu, uint32_t *pcLoops);
399
400
401/*********************************************************************************************************************************
402* Global Variables *
403*********************************************************************************************************************************/
404/** Ring-0 memory object for the IO bitmap. */
405static RTR0MEMOBJ g_hMemObjIOBitmap = NIL_RTR0MEMOBJ;
406/** Physical address of the IO bitmap. */
407static RTHCPHYS g_HCPhysIOBitmap;
408/** Pointer to the IO bitmap. */
409static R0PTRTYPE(void *) g_pvIOBitmap;
410
411#ifdef VBOX_STRICT
412# define HMSVM_LOG_RBP_RSP RT_BIT_32(0)
413# define HMSVM_LOG_CR_REGS RT_BIT_32(1)
414# define HMSVM_LOG_CS RT_BIT_32(2)
415# define HMSVM_LOG_SS RT_BIT_32(3)
416# define HMSVM_LOG_FS RT_BIT_32(4)
417# define HMSVM_LOG_GS RT_BIT_32(5)
418# define HMSVM_LOG_LBR RT_BIT_32(6)
419# define HMSVM_LOG_ALL ( HMSVM_LOG_RBP_RSP \
420 | HMSVM_LOG_CR_REGS \
421 | HMSVM_LOG_CS \
422 | HMSVM_LOG_SS \
423 | HMSVM_LOG_FS \
424 | HMSVM_LOG_GS \
425 | HMSVM_LOG_LBR)
426
427/**
428 * Dumps virtual CPU state and additional info. to the logger for diagnostics.
429 *
430 * @param pVCpu The cross context virtual CPU structure.
431 * @param pVmcb Pointer to the VM control block.
432 * @param pszPrefix Log prefix.
433 * @param fFlags Log flags, see HMSVM_LOG_XXX.
434 * @param uVerbose The verbosity level, currently unused.
435 */
436static void hmR0SvmLogState(PVMCPUCC pVCpu, PCSVMVMCB pVmcb, const char *pszPrefix, uint32_t fFlags, uint8_t uVerbose)
437{
438 RT_NOREF2(pVCpu, uVerbose);
439 PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
440
441 HMSVM_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CS | CPUMCTX_EXTRN_RIP | CPUMCTX_EXTRN_RFLAGS);
442 Log4(("%s: cs:rip=%04x:%RX64 efl=%#RX64\n", pszPrefix, pCtx->cs.Sel, pCtx->rip, pCtx->rflags.u));
443
444 if (fFlags & HMSVM_LOG_RBP_RSP)
445 {
446 HMSVM_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_RSP | CPUMCTX_EXTRN_RBP);
447 Log4(("%s: rsp=%#RX64 rbp=%#RX64\n", pszPrefix, pCtx->rsp, pCtx->rbp));
448 }
449
450 if (fFlags & HMSVM_LOG_CR_REGS)
451 {
452 HMSVM_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR0 | CPUMCTX_EXTRN_CR3 | CPUMCTX_EXTRN_CR4);
453 Log4(("%s: cr0=%#RX64 cr3=%#RX64 cr4=%#RX64\n", pszPrefix, pCtx->cr0, pCtx->cr3, pCtx->cr4));
454 }
455
456 if (fFlags & HMSVM_LOG_CS)
457 {
458 HMSVM_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CS);
459 Log4(("%s: cs={%04x base=%016RX64 limit=%08x flags=%08x}\n", pszPrefix, pCtx->cs.Sel, pCtx->cs.u64Base,
460 pCtx->cs.u32Limit, pCtx->cs.Attr.u));
461 }
462 if (fFlags & HMSVM_LOG_SS)
463 {
464 HMSVM_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_SS);
465 Log4(("%s: ss={%04x base=%016RX64 limit=%08x flags=%08x}\n", pszPrefix, pCtx->ss.Sel, pCtx->ss.u64Base,
466 pCtx->ss.u32Limit, pCtx->ss.Attr.u));
467 }
468 if (fFlags & HMSVM_LOG_FS)
469 {
470 HMSVM_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_FS);
471 Log4(("%s: fs={%04x base=%016RX64 limit=%08x flags=%08x}\n", pszPrefix, pCtx->fs.Sel, pCtx->fs.u64Base,
472 pCtx->fs.u32Limit, pCtx->fs.Attr.u));
473 }
474 if (fFlags & HMSVM_LOG_GS)
475 {
476 HMSVM_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_GS);
477 Log4(("%s: gs={%04x base=%016RX64 limit=%08x flags=%08x}\n", pszPrefix, pCtx->gs.Sel, pCtx->gs.u64Base,
478 pCtx->gs.u32Limit, pCtx->gs.Attr.u));
479 }
480
481 PCSVMVMCBSTATESAVE pVmcbGuest = &pVmcb->guest;
482 if (fFlags & HMSVM_LOG_LBR)
483 {
484 Log4(("%s: br_from=%#RX64 br_to=%#RX64 lastxcpt_from=%#RX64 lastxcpt_to=%#RX64\n", pszPrefix, pVmcbGuest->u64BR_FROM,
485 pVmcbGuest->u64BR_TO, pVmcbGuest->u64LASTEXCPFROM, pVmcbGuest->u64LASTEXCPTO));
486 }
487 NOREF(pszPrefix); NOREF(pVmcbGuest); NOREF(pCtx);
488}
489#endif /* VBOX_STRICT */
490
491
492/**
493 * Sets up and activates AMD-V on the current CPU.
494 *
495 * @returns VBox status code.
496 * @param pHostCpu The HM physical-CPU structure.
497 * @param pVM The cross context VM structure. Can be
498 * NULL after a resume!
499 * @param pvCpuPage Pointer to the global CPU page.
500 * @param HCPhysCpuPage Physical address of the global CPU page.
501 * @param fEnabledByHost Whether the host OS has already initialized AMD-V.
502 * @param pHwvirtMsrs Pointer to the hardware-virtualization MSRs (currently
503 * unused).
504 */
505VMMR0DECL(int) SVMR0EnableCpu(PHMPHYSCPU pHostCpu, PVMCC pVM, void *pvCpuPage, RTHCPHYS HCPhysCpuPage, bool fEnabledByHost,
506 PCSUPHWVIRTMSRS pHwvirtMsrs)
507{
508 Assert(!fEnabledByHost);
509 Assert(HCPhysCpuPage && HCPhysCpuPage != NIL_RTHCPHYS);
510 Assert(RT_ALIGN_T(HCPhysCpuPage, _4K, RTHCPHYS) == HCPhysCpuPage);
511 Assert(pvCpuPage); NOREF(pvCpuPage);
512 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
513
514 RT_NOREF2(fEnabledByHost, pHwvirtMsrs);
515
516 /* Paranoid: Disable interrupt as, in theory, interrupt handlers might mess with EFER. */
517 RTCCUINTREG const fEFlags = ASMIntDisableFlags();
518
519 /*
520 * We must turn on AMD-V and setup the host state physical address, as those MSRs are per CPU.
521 */
522 uint64_t u64HostEfer = ASMRdMsr(MSR_K6_EFER);
523 if (u64HostEfer & MSR_K6_EFER_SVME)
524 {
525 /* If the VBOX_HWVIRTEX_IGNORE_SVM_IN_USE is active, then we blindly use AMD-V. */
526 if ( pVM
527 && pVM->hm.s.svm.fIgnoreInUseError)
528 pHostCpu->fIgnoreAMDVInUseError = true;
529
530 if (!pHostCpu->fIgnoreAMDVInUseError)
531 {
532 ASMSetFlags(fEFlags);
533 return VERR_SVM_IN_USE;
534 }
535 }
536
537 /* Turn on AMD-V in the EFER MSR. */
538 ASMWrMsr(MSR_K6_EFER, u64HostEfer | MSR_K6_EFER_SVME);
539
540 /* Write the physical page address where the CPU will store the host state while executing the VM. */
541 ASMWrMsr(MSR_K8_VM_HSAVE_PA, HCPhysCpuPage);
542
543 /* Restore interrupts. */
544 ASMSetFlags(fEFlags);
545
546 /*
547 * Theoretically, other hypervisors may have used ASIDs, ideally we should flush all
548 * non-zero ASIDs when enabling SVM. AMD doesn't have an SVM instruction to flush all
549 * ASIDs (flushing is done upon VMRUN). Therefore, flag that we need to flush the TLB
550 * entirely with before executing any guest code.
551 */
552 pHostCpu->fFlushAsidBeforeUse = true;
553
554 /*
555 * Ensure each VCPU scheduled on this CPU gets a new ASID on resume. See @bugref{6255}.
556 */
557 ++pHostCpu->cTlbFlushes;
558
559 return VINF_SUCCESS;
560}
561
562
563/**
564 * Deactivates AMD-V on the current CPU.
565 *
566 * @returns VBox status code.
567 * @param pHostCpu The HM physical-CPU structure.
568 * @param pvCpuPage Pointer to the global CPU page.
569 * @param HCPhysCpuPage Physical address of the global CPU page.
570 */
571VMMR0DECL(int) SVMR0DisableCpu(PHMPHYSCPU pHostCpu, void *pvCpuPage, RTHCPHYS HCPhysCpuPage)
572{
573 RT_NOREF1(pHostCpu);
574 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
575 AssertReturn( HCPhysCpuPage
576 && HCPhysCpuPage != NIL_RTHCPHYS, VERR_INVALID_PARAMETER);
577 AssertReturn(pvCpuPage, VERR_INVALID_PARAMETER);
578
579 /* Paranoid: Disable interrupts as, in theory, interrupt handlers might mess with EFER. */
580 RTCCUINTREG const fEFlags = ASMIntDisableFlags();
581
582 /* Turn off AMD-V in the EFER MSR. */
583 uint64_t u64HostEfer = ASMRdMsr(MSR_K6_EFER);
584 ASMWrMsr(MSR_K6_EFER, u64HostEfer & ~MSR_K6_EFER_SVME);
585
586 /* Invalidate host state physical address. */
587 ASMWrMsr(MSR_K8_VM_HSAVE_PA, 0);
588
589 /* Restore interrupts. */
590 ASMSetFlags(fEFlags);
591
592 return VINF_SUCCESS;
593}
594
595
596/**
597 * Does global AMD-V initialization (called during module initialization).
598 *
599 * @returns VBox status code.
600 */
601VMMR0DECL(int) SVMR0GlobalInit(void)
602{
603 /*
604 * Allocate 12 KB (3 pages) for the IO bitmap. Since this is non-optional and we always
605 * intercept all IO accesses, it's done once globally here instead of per-VM.
606 */
607 Assert(g_hMemObjIOBitmap == NIL_RTR0MEMOBJ);
608 int rc = RTR0MemObjAllocCont(&g_hMemObjIOBitmap, SVM_IOPM_PAGES << X86_PAGE_4K_SHIFT, false /* fExecutable */);
609 if (RT_FAILURE(rc))
610 return rc;
611
612 g_pvIOBitmap = RTR0MemObjAddress(g_hMemObjIOBitmap);
613 g_HCPhysIOBitmap = RTR0MemObjGetPagePhysAddr(g_hMemObjIOBitmap, 0 /* iPage */);
614
615 /* Set all bits to intercept all IO accesses. */
616 ASMMemFill32(g_pvIOBitmap, SVM_IOPM_PAGES << X86_PAGE_4K_SHIFT, UINT32_C(0xffffffff));
617
618 return VINF_SUCCESS;
619}
620
621
622/**
623 * Does global AMD-V termination (called during module termination).
624 */
625VMMR0DECL(void) SVMR0GlobalTerm(void)
626{
627 if (g_hMemObjIOBitmap != NIL_RTR0MEMOBJ)
628 {
629 RTR0MemObjFree(g_hMemObjIOBitmap, true /* fFreeMappings */);
630 g_pvIOBitmap = NULL;
631 g_HCPhysIOBitmap = 0;
632 g_hMemObjIOBitmap = NIL_RTR0MEMOBJ;
633 }
634}
635
636
637/**
638 * Frees any allocated per-VCPU structures for a VM.
639 *
640 * @param pVM The cross context VM structure.
641 */
642DECLINLINE(void) hmR0SvmFreeStructs(PVMCC pVM)
643{
644 for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
645 {
646 PVMCPUCC pVCpu = VMCC_GET_CPU(pVM, idCpu);
647 AssertPtr(pVCpu);
648
649 if (pVCpu->hmr0.s.svm.hMemObjVmcbHost != NIL_RTR0MEMOBJ)
650 {
651 RTR0MemObjFree(pVCpu->hmr0.s.svm.hMemObjVmcbHost, false);
652 pVCpu->hmr0.s.svm.HCPhysVmcbHost = 0;
653 pVCpu->hmr0.s.svm.hMemObjVmcbHost = NIL_RTR0MEMOBJ;
654 }
655
656 if (pVCpu->hmr0.s.svm.hMemObjVmcb != NIL_RTR0MEMOBJ)
657 {
658 RTR0MemObjFree(pVCpu->hmr0.s.svm.hMemObjVmcb, false);
659 pVCpu->hmr0.s.svm.pVmcb = NULL;
660 pVCpu->hmr0.s.svm.HCPhysVmcb = 0;
661 pVCpu->hmr0.s.svm.hMemObjVmcb = NIL_RTR0MEMOBJ;
662 }
663
664 if (pVCpu->hmr0.s.svm.hMemObjMsrBitmap != NIL_RTR0MEMOBJ)
665 {
666 RTR0MemObjFree(pVCpu->hmr0.s.svm.hMemObjMsrBitmap, false);
667 pVCpu->hmr0.s.svm.pvMsrBitmap = NULL;
668 pVCpu->hmr0.s.svm.HCPhysMsrBitmap = 0;
669 pVCpu->hmr0.s.svm.hMemObjMsrBitmap = NIL_RTR0MEMOBJ;
670 }
671 }
672}
673
674
675/**
676 * Sets pfnVMRun to the best suited variant.
677 *
678 * This must be called whenever anything changes relative to the SVMR0VMRun
679 * variant selection:
680 * - pVCpu->hm.s.fLoadSaveGuestXcr0
681 * - CPUMCTX_WSF_IBPB_ENTRY in pVCpu->cpum.GstCtx.fWorldSwitcher
682 * - CPUMCTX_WSF_IBPB_EXIT in pVCpu->cpum.GstCtx.fWorldSwitcher
683 * - Perhaps: CPUMIsGuestFPUStateActive() (windows only)
684 * - Perhaps: CPUMCTX.fXStateMask (windows only)
685 *
686 * We currently ASSUME that neither CPUMCTX_WSF_IBPB_ENTRY nor
687 * CPUMCTX_WSF_IBPB_EXIT cannot be changed at runtime.
688 */
689static void hmR0SvmUpdateVmRunFunction(PVMCPUCC pVCpu)
690{
691 static const struct CLANGWORKAROUND { PFNHMSVMVMRUN pfn; } s_aHmR0SvmVmRunFunctions[] =
692 {
693 { hmR0SvmVmRun_SansXcr0_SansIbpbEntry_SansIbpbExit },
694 { hmR0SvmVmRun_WithXcr0_SansIbpbEntry_SansIbpbExit },
695 { hmR0SvmVmRun_SansXcr0_WithIbpbEntry_SansIbpbExit },
696 { hmR0SvmVmRun_WithXcr0_WithIbpbEntry_SansIbpbExit },
697 { hmR0SvmVmRun_SansXcr0_SansIbpbEntry_WithIbpbExit },
698 { hmR0SvmVmRun_WithXcr0_SansIbpbEntry_WithIbpbExit },
699 { hmR0SvmVmRun_SansXcr0_WithIbpbEntry_WithIbpbExit },
700 { hmR0SvmVmRun_WithXcr0_WithIbpbEntry_WithIbpbExit },
701 };
702 uintptr_t const idx = (pVCpu->hmr0.s.fLoadSaveGuestXcr0 ? 1 : 0)
703 | (pVCpu->hmr0.s.fWorldSwitcher & HM_WSF_IBPB_ENTRY ? 2 : 0)
704 | (pVCpu->hmr0.s.fWorldSwitcher & HM_WSF_IBPB_EXIT ? 4 : 0);
705 PFNHMSVMVMRUN const pfnVMRun = s_aHmR0SvmVmRunFunctions[idx].pfn;
706 if (pVCpu->hmr0.s.svm.pfnVMRun != pfnVMRun)
707 pVCpu->hmr0.s.svm.pfnVMRun = pfnVMRun;
708}
709
710
711/**
712 * Selector FNHMSVMVMRUN implementation.
713 */
714static DECLCALLBACK(int) hmR0SvmVMRunSelector(PVMCC pVM, PVMCPUCC pVCpu, RTHCPHYS HCPhysVMCB)
715{
716 hmR0SvmUpdateVmRunFunction(pVCpu);
717 return pVCpu->hmr0.s.svm.pfnVMRun(pVM, pVCpu, HCPhysVMCB);
718}
719
720
721/**
722 * Does per-VM AMD-V initialization.
723 *
724 * @returns VBox status code.
725 * @param pVM The cross context VM structure.
726 */
727VMMR0DECL(int) SVMR0InitVM(PVMCC pVM)
728{
729 int rc = VERR_INTERNAL_ERROR_5;
730
731 /*
732 * Check for an AMD CPU erratum which requires us to flush the TLB before every world-switch.
733 */
734 uint32_t u32Family;
735 uint32_t u32Model;
736 uint32_t u32Stepping;
737 if (HMIsSubjectToSvmErratum170(&u32Family, &u32Model, &u32Stepping))
738 {
739 Log4Func(("AMD cpu with erratum 170 family %#x model %#x stepping %#x\n", u32Family, u32Model, u32Stepping));
740 pVM->hmr0.s.svm.fAlwaysFlushTLB = true;
741 }
742
743 /*
744 * Initialize the R0 memory objects up-front so we can properly cleanup on allocation failures.
745 */
746 for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
747 {
748 PVMCPUCC pVCpu = VMCC_GET_CPU(pVM, idCpu);
749 pVCpu->hmr0.s.svm.hMemObjVmcbHost = NIL_RTR0MEMOBJ;
750 pVCpu->hmr0.s.svm.hMemObjVmcb = NIL_RTR0MEMOBJ;
751 pVCpu->hmr0.s.svm.hMemObjMsrBitmap = NIL_RTR0MEMOBJ;
752 }
753
754 for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
755 {
756 PVMCPUCC pVCpu = VMCC_GET_CPU(pVM, idCpu);
757
758 /*
759 * Initialize the hardware-assisted SVM guest-execution handler.
760 * We now use a single handler for both 32-bit and 64-bit guests, see @bugref{6208#c73}.
761 */
762 pVCpu->hmr0.s.svm.pfnVMRun = hmR0SvmVMRunSelector;
763
764 /*
765 * Allocate one page for the host-context VM control block (VMCB). This is used for additional host-state (such as
766 * FS, GS, Kernel GS Base, etc.) apart from the host-state save area specified in MSR_K8_VM_HSAVE_PA.
767 */
768/** @todo Does this need to be below 4G? */
769 rc = RTR0MemObjAllocCont(&pVCpu->hmr0.s.svm.hMemObjVmcbHost, SVM_VMCB_PAGES << HOST_PAGE_SHIFT, false /* fExecutable */);
770 if (RT_FAILURE(rc))
771 goto failure_cleanup;
772
773 void *pvVmcbHost = RTR0MemObjAddress(pVCpu->hmr0.s.svm.hMemObjVmcbHost);
774 pVCpu->hmr0.s.svm.HCPhysVmcbHost = RTR0MemObjGetPagePhysAddr(pVCpu->hmr0.s.svm.hMemObjVmcbHost, 0 /* iPage */);
775 Assert(pVCpu->hmr0.s.svm.HCPhysVmcbHost < _4G);
776 RT_BZERO(pvVmcbHost, HOST_PAGE_SIZE);
777
778 /*
779 * Allocate one page for the guest-state VMCB.
780 */
781/** @todo Does this need to be below 4G? */
782 rc = RTR0MemObjAllocCont(&pVCpu->hmr0.s.svm.hMemObjVmcb, SVM_VMCB_PAGES << HOST_PAGE_SHIFT, false /* fExecutable */);
783 if (RT_FAILURE(rc))
784 goto failure_cleanup;
785
786 pVCpu->hmr0.s.svm.pVmcb = (PSVMVMCB)RTR0MemObjAddress(pVCpu->hmr0.s.svm.hMemObjVmcb);
787 pVCpu->hmr0.s.svm.HCPhysVmcb = RTR0MemObjGetPagePhysAddr(pVCpu->hmr0.s.svm.hMemObjVmcb, 0 /* iPage */);
788 Assert(pVCpu->hmr0.s.svm.HCPhysVmcb < _4G);
789 RT_BZERO(pVCpu->hmr0.s.svm.pVmcb, HOST_PAGE_SIZE);
790
791 /*
792 * Allocate two pages (8 KB) for the MSR permission bitmap. There doesn't seem to be a way to convince
793 * SVM to not require one.
794 */
795/** @todo Does this need to be below 4G? */
796 rc = RTR0MemObjAllocCont(&pVCpu->hmr0.s.svm.hMemObjMsrBitmap, SVM_MSRPM_PAGES << HOST_PAGE_SHIFT,
797 false /* fExecutable */);
798 if (RT_FAILURE(rc))
799 goto failure_cleanup;
800
801 pVCpu->hmr0.s.svm.pvMsrBitmap = RTR0MemObjAddress(pVCpu->hmr0.s.svm.hMemObjMsrBitmap);
802 pVCpu->hmr0.s.svm.HCPhysMsrBitmap = RTR0MemObjGetPagePhysAddr(pVCpu->hmr0.s.svm.hMemObjMsrBitmap, 0 /* iPage */);
803 /* Set all bits to intercept all MSR accesses (changed later on). */
804 ASMMemFill32(pVCpu->hmr0.s.svm.pvMsrBitmap, SVM_MSRPM_PAGES << HOST_PAGE_SHIFT, UINT32_C(0xffffffff));
805 }
806
807 return VINF_SUCCESS;
808
809failure_cleanup:
810 hmR0SvmFreeStructs(pVM);
811 return rc;
812}
813
814
815/**
816 * Does per-VM AMD-V termination.
817 *
818 * @returns VBox status code.
819 * @param pVM The cross context VM structure.
820 */
821VMMR0DECL(int) SVMR0TermVM(PVMCC pVM)
822{
823 hmR0SvmFreeStructs(pVM);
824 return VINF_SUCCESS;
825}
826
827
828/**
829 * Returns whether the VMCB Clean Bits feature is supported.
830 *
831 * @returns @c true if supported, @c false otherwise.
832 * @param pVCpu The cross context virtual CPU structure.
833 * @param fIsNestedGuest Whether we are currently executing the nested-guest.
834 */
835DECL_FORCE_INLINE(bool) hmR0SvmSupportsVmcbCleanBits(PVMCPUCC pVCpu, bool fIsNestedGuest)
836{
837 PCVMCC pVM = pVCpu->CTX_SUFF(pVM);
838 bool const fHostVmcbCleanBits = RT_BOOL(g_fHmSvmFeatures & X86_CPUID_SVM_FEATURE_EDX_VMCB_CLEAN);
839 if (!fIsNestedGuest)
840 return fHostVmcbCleanBits;
841 return fHostVmcbCleanBits && pVM->cpum.ro.GuestFeatures.fSvmVmcbClean;
842}
843
844
845/**
846 * Returns whether the decode assists feature is supported.
847 *
848 * @returns @c true if supported, @c false otherwise.
849 * @param pVCpu The cross context virtual CPU structure.
850 */
851DECLINLINE(bool) hmR0SvmSupportsDecodeAssists(PVMCPUCC pVCpu)
852{
853 PVMCC pVM = pVCpu->CTX_SUFF(pVM);
854#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
855 if (CPUMIsGuestInSvmNestedHwVirtMode(&pVCpu->cpum.GstCtx))
856 return (g_fHmSvmFeatures & X86_CPUID_SVM_FEATURE_EDX_DECODE_ASSISTS)
857 && pVM->cpum.ro.GuestFeatures.fSvmDecodeAssists;
858#endif
859 return RT_BOOL(g_fHmSvmFeatures & X86_CPUID_SVM_FEATURE_EDX_DECODE_ASSISTS);
860}
861
862
863/**
864 * Returns whether the NRIP_SAVE feature is supported.
865 *
866 * @returns @c true if supported, @c false otherwise.
867 * @param pVCpu The cross context virtual CPU structure.
868 */
869DECLINLINE(bool) hmR0SvmSupportsNextRipSave(PVMCPUCC pVCpu)
870{
871 PVMCC pVM = pVCpu->CTX_SUFF(pVM);
872#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
873 if (CPUMIsGuestInSvmNestedHwVirtMode(&pVCpu->cpum.GstCtx))
874 return (g_fHmSvmFeatures & X86_CPUID_SVM_FEATURE_EDX_NRIP_SAVE)
875 && pVM->cpum.ro.GuestFeatures.fSvmNextRipSave;
876#endif
877 return RT_BOOL(g_fHmSvmFeatures & X86_CPUID_SVM_FEATURE_EDX_NRIP_SAVE);
878}
879
880
881/**
882 * Sets the permission bits for the specified MSR in the MSRPM bitmap.
883 *
884 * @param pVCpu The cross context virtual CPU structure.
885 * @param pbMsrBitmap Pointer to the MSR bitmap.
886 * @param idMsr The MSR for which the permissions are being set.
887 * @param enmRead MSR read permissions.
888 * @param enmWrite MSR write permissions.
889 *
890 * @remarks This function does -not- clear the VMCB clean bits for MSRPM. The
891 * caller needs to take care of this.
892 */
893static void hmR0SvmSetMsrPermission(PVMCPUCC pVCpu, uint8_t *pbMsrBitmap, uint32_t idMsr, SVMMSREXITREAD enmRead,
894 SVMMSREXITWRITE enmWrite)
895{
896 bool const fInNestedGuestMode = CPUMIsGuestInSvmNestedHwVirtMode(&pVCpu->cpum.GstCtx);
897 uint16_t offMsrpm;
898 uint8_t uMsrpmBit;
899 int rc = CPUMGetSvmMsrpmOffsetAndBit(idMsr, &offMsrpm, &uMsrpmBit);
900 AssertRC(rc);
901
902 Assert(uMsrpmBit == 0 || uMsrpmBit == 2 || uMsrpmBit == 4 || uMsrpmBit == 6);
903 Assert(offMsrpm < SVM_MSRPM_PAGES << X86_PAGE_4K_SHIFT);
904
905 pbMsrBitmap += offMsrpm;
906 if (enmRead == SVMMSREXIT_INTERCEPT_READ)
907 *pbMsrBitmap |= RT_BIT(uMsrpmBit);
908 else
909 {
910 if (!fInNestedGuestMode)
911 *pbMsrBitmap &= ~RT_BIT(uMsrpmBit);
912#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
913 else
914 {
915 /* Only clear the bit if the nested-guest is also not intercepting the MSR read.*/
916 if (!(pVCpu->cpum.GstCtx.hwvirt.svm.abMsrBitmap[offMsrpm] & RT_BIT(uMsrpmBit)))
917 *pbMsrBitmap &= ~RT_BIT(uMsrpmBit);
918 else
919 Assert(*pbMsrBitmap & RT_BIT(uMsrpmBit));
920 }
921#endif
922 }
923
924 if (enmWrite == SVMMSREXIT_INTERCEPT_WRITE)
925 *pbMsrBitmap |= RT_BIT(uMsrpmBit + 1);
926 else
927 {
928 if (!fInNestedGuestMode)
929 *pbMsrBitmap &= ~RT_BIT(uMsrpmBit + 1);
930#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
931 else
932 {
933 /* Only clear the bit if the nested-guest is also not intercepting the MSR write.*/
934 if (!(pVCpu->cpum.GstCtx.hwvirt.svm.abMsrBitmap[offMsrpm] & RT_BIT(uMsrpmBit + 1)))
935 *pbMsrBitmap &= ~RT_BIT(uMsrpmBit + 1);
936 else
937 Assert(*pbMsrBitmap & RT_BIT(uMsrpmBit + 1));
938 }
939#endif
940 }
941}
942
943
944/**
945 * Sets up AMD-V for the specified VM.
946 * This function is only called once per-VM during initalization.
947 *
948 * @returns VBox status code.
949 * @param pVM The cross context VM structure.
950 */
951VMMR0DECL(int) SVMR0SetupVM(PVMCC pVM)
952{
953 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
954 AssertReturn(pVM, VERR_INVALID_PARAMETER);
955
956 /*
957 * Validate and copy over some parameters.
958 */
959 AssertReturn(pVM->hm.s.svm.fSupported, VERR_INCOMPATIBLE_CONFIG);
960 bool const fNestedPaging = pVM->hm.s.fNestedPagingCfg;
961 AssertReturn(!fNestedPaging || (g_fHmSvmFeatures & X86_CPUID_SVM_FEATURE_EDX_NESTED_PAGING), VERR_INCOMPATIBLE_CONFIG);
962 pVM->hmr0.s.fNestedPaging = fNestedPaging;
963 pVM->hmr0.s.fAllow64BitGuests = pVM->hm.s.fAllow64BitGuestsCfg;
964
965 /*
966 * Determin some configuration parameters.
967 */
968 bool const fPauseFilter = RT_BOOL(g_fHmSvmFeatures & X86_CPUID_SVM_FEATURE_EDX_PAUSE_FILTER);
969 bool const fPauseFilterThreshold = RT_BOOL(g_fHmSvmFeatures & X86_CPUID_SVM_FEATURE_EDX_PAUSE_FILTER_THRESHOLD);
970 bool const fUsePauseFilter = fPauseFilter && pVM->hm.s.svm.cPauseFilter;
971
972 bool const fLbrVirt = RT_BOOL(g_fHmSvmFeatures & X86_CPUID_SVM_FEATURE_EDX_LBR_VIRT);
973 bool const fUseLbrVirt = fLbrVirt && pVM->hm.s.svm.fLbrVirt; /** @todo IEM implementation etc. */
974
975#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
976 bool const fVirtVmsaveVmload = RT_BOOL(g_fHmSvmFeatures & X86_CPUID_SVM_FEATURE_EDX_VIRT_VMSAVE_VMLOAD);
977 bool const fUseVirtVmsaveVmload = fVirtVmsaveVmload && pVM->hm.s.svm.fVirtVmsaveVmload && fNestedPaging;
978
979 bool const fVGif = RT_BOOL(g_fHmSvmFeatures & X86_CPUID_SVM_FEATURE_EDX_VGIF);
980 bool const fUseVGif = fVGif && pVM->hm.s.svm.fVGif;
981#endif
982
983 PVMCPUCC pVCpu0 = VMCC_GET_CPU_0(pVM);
984 PSVMVMCB pVmcb0 = pVCpu0->hmr0.s.svm.pVmcb;
985 AssertMsgReturn(RT_VALID_PTR(pVmcb0), ("Invalid pVmcb (%p) for vcpu[0]\n", pVmcb0), VERR_SVM_INVALID_PVMCB);
986 PSVMVMCBCTRL pVmcbCtrl0 = &pVmcb0->ctrl;
987
988 /* Always trap #AC for reasons of security. */
989 pVmcbCtrl0->u32InterceptXcpt |= RT_BIT_32(X86_XCPT_AC);
990
991 /* Always trap #DB for reasons of security. */
992 pVmcbCtrl0->u32InterceptXcpt |= RT_BIT_32(X86_XCPT_DB);
993
994 /* Trap exceptions unconditionally (debug purposes). */
995#ifdef HMSVM_ALWAYS_TRAP_PF
996 pVmcbCtrl0->u32InterceptXcpt |= RT_BIT_32(X86_XCPT_PF);
997#endif
998#ifdef HMSVM_ALWAYS_TRAP_ALL_XCPTS
999 /* If you add any exceptions here, make sure to update hmR0SvmHandleExit(). */
1000 pVmcbCtrl0->u32InterceptXcpt |= RT_BIT_32(X86_XCPT_BP)
1001 | RT_BIT_32(X86_XCPT_DE)
1002 | RT_BIT_32(X86_XCPT_NM)
1003 | RT_BIT_32(X86_XCPT_UD)
1004 | RT_BIT_32(X86_XCPT_NP)
1005 | RT_BIT_32(X86_XCPT_SS)
1006 | RT_BIT_32(X86_XCPT_GP)
1007 | RT_BIT_32(X86_XCPT_PF)
1008 | RT_BIT_32(X86_XCPT_MF)
1009 ;
1010#endif
1011
1012 /* Apply the exceptions intercepts needed by the GIM provider. */
1013 if (pVCpu0->hm.s.fGIMTrapXcptUD || pVCpu0->hm.s.svm.fEmulateLongModeSysEnterExit)
1014 pVmcbCtrl0->u32InterceptXcpt |= RT_BIT(X86_XCPT_UD);
1015
1016 /* Apply the exceptions intercepts needed by the GCM fixers. */
1017 if (pVCpu0->hm.s.fGCMTrapXcptDE)
1018 pVmcbCtrl0->u32InterceptXcpt |= RT_BIT(X86_XCPT_DE);
1019
1020 /* The mesa 3d driver hack needs #GP. */
1021 if (pVCpu0->hm.s.fTrapXcptGpForLovelyMesaDrv)
1022 pVmcbCtrl0->u32InterceptXcpt |= RT_BIT(X86_XCPT_GP);
1023
1024 /* Set up unconditional intercepts and conditions. */
1025 pVmcbCtrl0->u64InterceptCtrl = HMSVM_MANDATORY_GUEST_CTRL_INTERCEPTS
1026 | SVM_CTRL_INTERCEPT_VMMCALL
1027 | SVM_CTRL_INTERCEPT_VMSAVE
1028 | SVM_CTRL_INTERCEPT_VMLOAD
1029 | SVM_CTRL_INTERCEPT_CLGI
1030 | SVM_CTRL_INTERCEPT_STGI;
1031
1032#ifdef HMSVM_ALWAYS_TRAP_TASK_SWITCH
1033 pVmcbCtrl0->u64InterceptCtrl |= SVM_CTRL_INTERCEPT_TASK_SWITCH;
1034#endif
1035
1036#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
1037 if (pVCpu0->CTX_SUFF(pVM)->cpum.ro.GuestFeatures.fSvm)
1038 {
1039 /* Virtualized VMSAVE/VMLOAD. */
1040 if (fUseVirtVmsaveVmload)
1041 {
1042 pVmcbCtrl0->LbrVirt.n.u1VirtVmsaveVmload = 1;
1043 pVmcbCtrl0->u64InterceptCtrl &= ~( SVM_CTRL_INTERCEPT_VMSAVE
1044 | SVM_CTRL_INTERCEPT_VMLOAD);
1045 }
1046 else
1047 Assert(!pVmcbCtrl0->LbrVirt.n.u1VirtVmsaveVmload);
1048
1049 /* Virtual GIF. */
1050 if (fUseVGif)
1051 {
1052 pVmcbCtrl0->IntCtrl.n.u1VGifEnable = 1;
1053 pVmcbCtrl0->u64InterceptCtrl &= ~( SVM_CTRL_INTERCEPT_CLGI
1054 | SVM_CTRL_INTERCEPT_STGI);
1055 }
1056 else
1057 Assert(!pVmcbCtrl0->IntCtrl.n.u1VGifEnable);
1058 }
1059 else
1060#endif
1061 {
1062 Assert(!pVCpu0->CTX_SUFF(pVM)->cpum.ro.GuestFeatures.fSvm);
1063 Assert(!pVmcbCtrl0->LbrVirt.n.u1VirtVmsaveVmload);
1064 Assert(!pVmcbCtrl0->IntCtrl.n.u1VGifEnable);
1065 }
1066
1067 /* CR4 writes must always be intercepted for tracking PGM mode changes and
1068 AVX (for XCR0 syncing during worlds switching). */
1069 pVmcbCtrl0->u16InterceptWrCRx = RT_BIT(4);
1070
1071 /* Intercept all DRx reads and writes by default. Changed later on. */
1072 pVmcbCtrl0->u16InterceptRdDRx = 0xffff;
1073 pVmcbCtrl0->u16InterceptWrDRx = 0xffff;
1074
1075 /* Virtualize masking of INTR interrupts. (reads/writes from/to CR8 go to the V_TPR register) */
1076 pVmcbCtrl0->IntCtrl.n.u1VIntrMasking = 1;
1077
1078 /* Ignore the priority in the virtual TPR. This is necessary for delivering PIC style (ExtInt) interrupts
1079 and we currently deliver both PIC and APIC interrupts alike, see hmR0SvmEvaluatePendingEvent() */
1080 pVmcbCtrl0->IntCtrl.n.u1IgnoreTPR = 1;
1081
1082 /* Set the IO permission bitmap physical addresses. */
1083 pVmcbCtrl0->u64IOPMPhysAddr = g_HCPhysIOBitmap;
1084
1085 /* LBR virtualization. */
1086 pVmcbCtrl0->LbrVirt.n.u1LbrVirt = fUseLbrVirt;
1087
1088 /* The host ASID MBZ, for the guest start with 1. */
1089 pVmcbCtrl0->TLBCtrl.n.u32ASID = 1;
1090
1091 /* Setup Nested Paging. This doesn't change throughout the execution time of the VM. */
1092 pVmcbCtrl0->NestedPagingCtrl.n.u1NestedPaging = fNestedPaging;
1093
1094 /* Without Nested Paging, we need additionally intercepts. */
1095 if (!fNestedPaging)
1096 {
1097 /* CR3 reads/writes must be intercepted; our shadow values differ from the guest values. */
1098 pVmcbCtrl0->u16InterceptRdCRx |= RT_BIT(3);
1099 pVmcbCtrl0->u16InterceptWrCRx |= RT_BIT(3);
1100
1101 /* Intercept INVLPG and task switches (may change CR3, EFLAGS, LDT). */
1102 pVmcbCtrl0->u64InterceptCtrl |= SVM_CTRL_INTERCEPT_INVLPG
1103 | SVM_CTRL_INTERCEPT_TASK_SWITCH;
1104
1105 /* Page faults must be intercepted to implement shadow paging. */
1106 pVmcbCtrl0->u32InterceptXcpt |= RT_BIT(X86_XCPT_PF);
1107 }
1108
1109 /* Workaround for missing OS/2 TLB flush, see ticketref:20625. */
1110 if (pVM->hm.s.fMissingOS2TlbFlushWorkaround)
1111 pVmcbCtrl0->u64InterceptCtrl |= SVM_CTRL_INTERCEPT_TR_WRITES;
1112
1113 /* Setup Pause Filter for guest pause-loop (spinlock) exiting. */
1114 if (fUsePauseFilter)
1115 {
1116 Assert(pVM->hm.s.svm.cPauseFilter > 0);
1117 pVmcbCtrl0->u16PauseFilterCount = pVM->hm.s.svm.cPauseFilter;
1118 if (fPauseFilterThreshold)
1119 pVmcbCtrl0->u16PauseFilterThreshold = pVM->hm.s.svm.cPauseFilterThresholdTicks;
1120 pVmcbCtrl0->u64InterceptCtrl |= SVM_CTRL_INTERCEPT_PAUSE;
1121 }
1122
1123 /*
1124 * Setup the MSR permission bitmap.
1125 * The following MSRs are saved/restored automatically during the world-switch.
1126 * Don't intercept guest read/write accesses to these MSRs.
1127 */
1128 uint8_t *pbMsrBitmap0 = (uint8_t *)pVCpu0->hmr0.s.svm.pvMsrBitmap;
1129 hmR0SvmSetMsrPermission(pVCpu0, pbMsrBitmap0, MSR_K8_LSTAR, SVMMSREXIT_PASSTHRU_READ, SVMMSREXIT_PASSTHRU_WRITE);
1130 hmR0SvmSetMsrPermission(pVCpu0, pbMsrBitmap0, MSR_K8_CSTAR, SVMMSREXIT_PASSTHRU_READ, SVMMSREXIT_PASSTHRU_WRITE);
1131 hmR0SvmSetMsrPermission(pVCpu0, pbMsrBitmap0, MSR_K6_STAR, SVMMSREXIT_PASSTHRU_READ, SVMMSREXIT_PASSTHRU_WRITE);
1132 hmR0SvmSetMsrPermission(pVCpu0, pbMsrBitmap0, MSR_K8_SF_MASK, SVMMSREXIT_PASSTHRU_READ, SVMMSREXIT_PASSTHRU_WRITE);
1133 hmR0SvmSetMsrPermission(pVCpu0, pbMsrBitmap0, MSR_K8_FS_BASE, SVMMSREXIT_PASSTHRU_READ, SVMMSREXIT_PASSTHRU_WRITE);
1134 hmR0SvmSetMsrPermission(pVCpu0, pbMsrBitmap0, MSR_K8_GS_BASE, SVMMSREXIT_PASSTHRU_READ, SVMMSREXIT_PASSTHRU_WRITE);
1135 hmR0SvmSetMsrPermission(pVCpu0, pbMsrBitmap0, MSR_K8_KERNEL_GS_BASE, SVMMSREXIT_PASSTHRU_READ, SVMMSREXIT_PASSTHRU_WRITE);
1136 if (!pVCpu0->hm.s.svm.fEmulateLongModeSysEnterExit)
1137 {
1138 hmR0SvmSetMsrPermission(pVCpu0, pbMsrBitmap0, MSR_IA32_SYSENTER_CS, SVMMSREXIT_PASSTHRU_READ, SVMMSREXIT_PASSTHRU_WRITE);
1139 hmR0SvmSetMsrPermission(pVCpu0, pbMsrBitmap0, MSR_IA32_SYSENTER_ESP, SVMMSREXIT_PASSTHRU_READ, SVMMSREXIT_PASSTHRU_WRITE);
1140 hmR0SvmSetMsrPermission(pVCpu0, pbMsrBitmap0, MSR_IA32_SYSENTER_EIP, SVMMSREXIT_PASSTHRU_READ, SVMMSREXIT_PASSTHRU_WRITE);
1141 }
1142 else
1143 {
1144 hmR0SvmSetMsrPermission(pVCpu0, pbMsrBitmap0, MSR_IA32_SYSENTER_CS, SVMMSREXIT_INTERCEPT_READ, SVMMSREXIT_INTERCEPT_WRITE);
1145 hmR0SvmSetMsrPermission(pVCpu0, pbMsrBitmap0, MSR_IA32_SYSENTER_ESP, SVMMSREXIT_INTERCEPT_READ, SVMMSREXIT_INTERCEPT_WRITE);
1146 hmR0SvmSetMsrPermission(pVCpu0, pbMsrBitmap0, MSR_IA32_SYSENTER_EIP, SVMMSREXIT_INTERCEPT_READ, SVMMSREXIT_INTERCEPT_WRITE);
1147 }
1148 pVmcbCtrl0->u64MSRPMPhysAddr = pVCpu0->hmr0.s.svm.HCPhysMsrBitmap;
1149
1150 /* Initially all VMCB clean bits MBZ indicating that everything should be loaded from the VMCB in memory. */
1151 Assert(pVmcbCtrl0->u32VmcbCleanBits == 0);
1152
1153 for (VMCPUID idCpu = 1; idCpu < pVM->cCpus; idCpu++)
1154 {
1155 PVMCPUCC pVCpuCur = VMCC_GET_CPU(pVM, idCpu);
1156 PSVMVMCB pVmcbCur = pVCpuCur->hmr0.s.svm.pVmcb;
1157 AssertMsgReturn(RT_VALID_PTR(pVmcbCur), ("Invalid pVmcb (%p) for vcpu[%u]\n", pVmcbCur, idCpu), VERR_SVM_INVALID_PVMCB);
1158 PSVMVMCBCTRL pVmcbCtrlCur = &pVmcbCur->ctrl;
1159
1160 /* Copy the VMCB control area. */
1161 memcpy(pVmcbCtrlCur, pVmcbCtrl0, sizeof(*pVmcbCtrlCur));
1162
1163 /* Copy the MSR bitmap and setup the VCPU-specific host physical address. */
1164 uint8_t *pbMsrBitmapCur = (uint8_t *)pVCpuCur->hmr0.s.svm.pvMsrBitmap;
1165 memcpy(pbMsrBitmapCur, pbMsrBitmap0, SVM_MSRPM_PAGES << X86_PAGE_4K_SHIFT);
1166 pVmcbCtrlCur->u64MSRPMPhysAddr = pVCpuCur->hmr0.s.svm.HCPhysMsrBitmap;
1167
1168 /* Initially all VMCB clean bits MBZ indicating that everything should be loaded from the VMCB in memory. */
1169 Assert(pVmcbCtrlCur->u32VmcbCleanBits == 0);
1170
1171 /* Verify our assumption that GIM providers trap #UD uniformly across VCPUs initially. */
1172 Assert(pVCpuCur->hm.s.fGIMTrapXcptUD == pVCpu0->hm.s.fGIMTrapXcptUD);
1173 /* Same for GCM, #DE trapping should be uniform across VCPUs. */
1174 Assert(pVCpuCur->hm.s.fGCMTrapXcptDE == pVCpu0->hm.s.fGCMTrapXcptDE);
1175 }
1176
1177#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
1178 LogRel(("HM: fUsePauseFilter=%RTbool fUseLbrVirt=%RTbool fUseVGif=%RTbool fUseVirtVmsaveVmload=%RTbool\n", fUsePauseFilter,
1179 fUseLbrVirt, fUseVGif, fUseVirtVmsaveVmload));
1180#else
1181 LogRel(("HM: fUsePauseFilter=%RTbool fUseLbrVirt=%RTbool\n", fUsePauseFilter, fUseLbrVirt));
1182#endif
1183 return VINF_SUCCESS;
1184}
1185
1186
1187/**
1188 * Gets a pointer to the currently active guest (or nested-guest) VMCB.
1189 *
1190 * @returns Pointer to the current context VMCB.
1191 * @param pVCpu The cross context virtual CPU structure.
1192 */
1193DECLINLINE(PSVMVMCB) hmR0SvmGetCurrentVmcb(PVMCPUCC pVCpu)
1194{
1195#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
1196 if (CPUMIsGuestInSvmNestedHwVirtMode(&pVCpu->cpum.GstCtx))
1197 return &pVCpu->cpum.GstCtx.hwvirt.svm.Vmcb;
1198#endif
1199 return pVCpu->hmr0.s.svm.pVmcb;
1200}
1201
1202
1203/**
1204 * Gets a pointer to the nested-guest VMCB cache.
1205 *
1206 * @returns Pointer to the nested-guest VMCB cache.
1207 * @param pVCpu The cross context virtual CPU structure.
1208 */
1209DECLINLINE(PSVMNESTEDVMCBCACHE) hmR0SvmGetNestedVmcbCache(PVMCPUCC pVCpu)
1210{
1211#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
1212 Assert(pVCpu->hm.s.svm.NstGstVmcbCache.fCacheValid);
1213 return &pVCpu->hm.s.svm.NstGstVmcbCache;
1214#else
1215 RT_NOREF(pVCpu);
1216 return NULL;
1217#endif
1218}
1219
1220
1221/**
1222 * Invalidates a guest page by guest virtual address.
1223 *
1224 * @returns VBox status code.
1225 * @param pVCpu The cross context virtual CPU structure.
1226 * @param GCVirt Guest virtual address of the page to invalidate.
1227 */
1228VMMR0DECL(int) SVMR0InvalidatePage(PVMCPUCC pVCpu, RTGCPTR GCVirt)
1229{
1230 Assert(pVCpu->CTX_SUFF(pVM)->hm.s.svm.fSupported);
1231
1232 bool const fFlushPending = VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_TLB_FLUSH) || pVCpu->CTX_SUFF(pVM)->hmr0.s.svm.fAlwaysFlushTLB;
1233
1234 /* Skip it if a TLB flush is already pending. */
1235 if (!fFlushPending)
1236 {
1237 Log4Func(("%#RGv\n", GCVirt));
1238
1239 PSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
1240 AssertMsgReturn(pVmcb, ("Invalid pVmcb!\n"), VERR_SVM_INVALID_PVMCB);
1241
1242 SVMR0InvlpgA(GCVirt, pVmcb->ctrl.TLBCtrl.n.u32ASID);
1243 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbInvlpgVirt);
1244 }
1245 return VINF_SUCCESS;
1246}
1247
1248
1249/**
1250 * Flushes the appropriate tagged-TLB entries.
1251 *
1252 * @param pHostCpu The HM physical-CPU structure.
1253 * @param pVCpu The cross context virtual CPU structure.
1254 * @param pVmcb Pointer to the VM control block.
1255 */
1256static void hmR0SvmFlushTaggedTlb(PHMPHYSCPU pHostCpu, PVMCPUCC pVCpu, PSVMVMCB pVmcb)
1257{
1258 /*
1259 * Force a TLB flush for the first world switch if the current CPU differs from the one
1260 * we ran on last. This can happen both for start & resume due to long jumps back to
1261 * ring-3.
1262 *
1263 * We also force a TLB flush every time when executing a nested-guest VCPU as there is no
1264 * correlation between it and the physical CPU.
1265 *
1266 * If the TLB flush count changed, another VM (VCPU rather) has hit the ASID limit while
1267 * flushing the TLB, so we cannot reuse the ASIDs without flushing.
1268 */
1269 bool fNewAsid = false;
1270 Assert(pHostCpu->idCpu != NIL_RTCPUID);
1271 if ( pVCpu->hmr0.s.idLastCpu != pHostCpu->idCpu
1272 || pVCpu->hmr0.s.cTlbFlushes != pHostCpu->cTlbFlushes
1273#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
1274 || CPUMIsGuestInSvmNestedHwVirtMode(&pVCpu->cpum.GstCtx)
1275#endif
1276 )
1277 {
1278 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
1279 pVCpu->hmr0.s.fForceTLBFlush = true;
1280 fNewAsid = true;
1281 }
1282
1283 /* Set TLB flush state as checked until we return from the world switch. */
1284 ASMAtomicUoWriteBool(&pVCpu->hm.s.fCheckedTLBFlush, true);
1285
1286 /* Check for explicit TLB flushes. */
1287 if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH))
1288 {
1289 pVCpu->hmr0.s.fForceTLBFlush = true;
1290 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
1291 }
1292
1293 /*
1294 * If the AMD CPU erratum 170, We need to flush the entire TLB for each world switch. Sad.
1295 * This Host CPU requirement takes precedence.
1296 */
1297 PVMCC pVM = pVCpu->CTX_SUFF(pVM);
1298 if (pVM->hmr0.s.svm.fAlwaysFlushTLB)
1299 {
1300 pHostCpu->uCurrentAsid = 1;
1301 pVCpu->hmr0.s.uCurrentAsid = 1;
1302 pVCpu->hmr0.s.cTlbFlushes = pHostCpu->cTlbFlushes;
1303 pVCpu->hmr0.s.idLastCpu = pHostCpu->idCpu;
1304 pVmcb->ctrl.TLBCtrl.n.u8TLBFlush = SVM_TLB_FLUSH_ENTIRE;
1305
1306 /* Clear the VMCB Clean Bit for NP while flushing the TLB. See @bugref{7152}. */
1307 pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_NP;
1308 }
1309 else
1310 {
1311 pVmcb->ctrl.TLBCtrl.n.u8TLBFlush = SVM_TLB_FLUSH_NOTHING;
1312 if (pVCpu->hmr0.s.fForceTLBFlush)
1313 {
1314 /* Clear the VMCB Clean Bit for NP while flushing the TLB. See @bugref{7152}. */
1315 pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_NP;
1316
1317 if (fNewAsid)
1318 {
1319 ++pHostCpu->uCurrentAsid;
1320
1321 bool fHitASIDLimit = false;
1322 if (pHostCpu->uCurrentAsid >= g_uHmMaxAsid)
1323 {
1324 pHostCpu->uCurrentAsid = 1; /* Wraparound at 1; host uses 0 */
1325 pHostCpu->cTlbFlushes++; /* All VCPUs that run on this host CPU must use a new ASID. */
1326 fHitASIDLimit = true;
1327 }
1328
1329 if ( fHitASIDLimit
1330 || pHostCpu->fFlushAsidBeforeUse)
1331 {
1332 pVmcb->ctrl.TLBCtrl.n.u8TLBFlush = SVM_TLB_FLUSH_ENTIRE;
1333 pHostCpu->fFlushAsidBeforeUse = false;
1334 }
1335
1336 pVCpu->hmr0.s.uCurrentAsid = pHostCpu->uCurrentAsid;
1337 pVCpu->hmr0.s.idLastCpu = pHostCpu->idCpu;
1338 pVCpu->hmr0.s.cTlbFlushes = pHostCpu->cTlbFlushes;
1339 }
1340 else
1341 {
1342 if (g_fHmSvmFeatures & X86_CPUID_SVM_FEATURE_EDX_FLUSH_BY_ASID)
1343 pVmcb->ctrl.TLBCtrl.n.u8TLBFlush = SVM_TLB_FLUSH_SINGLE_CONTEXT;
1344 else
1345 pVmcb->ctrl.TLBCtrl.n.u8TLBFlush = SVM_TLB_FLUSH_ENTIRE;
1346 }
1347
1348 pVCpu->hmr0.s.fForceTLBFlush = false;
1349 }
1350 }
1351
1352 /* Update VMCB with the ASID. */
1353 if (pVmcb->ctrl.TLBCtrl.n.u32ASID != pVCpu->hmr0.s.uCurrentAsid)
1354 {
1355 pVmcb->ctrl.TLBCtrl.n.u32ASID = pVCpu->hmr0.s.uCurrentAsid;
1356 pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_ASID;
1357 }
1358
1359 AssertMsg(pVCpu->hmr0.s.idLastCpu == pHostCpu->idCpu,
1360 ("vcpu idLastCpu=%u hostcpu idCpu=%u\n", pVCpu->hmr0.s.idLastCpu, pHostCpu->idCpu));
1361 AssertMsg(pVCpu->hmr0.s.cTlbFlushes == pHostCpu->cTlbFlushes,
1362 ("Flush count mismatch for cpu %u (%u vs %u)\n", pHostCpu->idCpu, pVCpu->hmr0.s.cTlbFlushes, pHostCpu->cTlbFlushes));
1363 AssertMsg(pHostCpu->uCurrentAsid >= 1 && pHostCpu->uCurrentAsid < g_uHmMaxAsid,
1364 ("cpu%d uCurrentAsid = %x\n", pHostCpu->idCpu, pHostCpu->uCurrentAsid));
1365 AssertMsg(pVCpu->hmr0.s.uCurrentAsid >= 1 && pVCpu->hmr0.s.uCurrentAsid < g_uHmMaxAsid,
1366 ("cpu%d VM uCurrentAsid = %x\n", pHostCpu->idCpu, pVCpu->hmr0.s.uCurrentAsid));
1367
1368#ifdef VBOX_WITH_STATISTICS
1369 if (pVmcb->ctrl.TLBCtrl.n.u8TLBFlush == SVM_TLB_FLUSH_NOTHING)
1370 STAM_COUNTER_INC(&pVCpu->hm.s.StatNoFlushTlbWorldSwitch);
1371 else if ( pVmcb->ctrl.TLBCtrl.n.u8TLBFlush == SVM_TLB_FLUSH_SINGLE_CONTEXT
1372 || pVmcb->ctrl.TLBCtrl.n.u8TLBFlush == SVM_TLB_FLUSH_SINGLE_CONTEXT_RETAIN_GLOBALS)
1373 {
1374 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushAsid);
1375 }
1376 else
1377 {
1378 Assert(pVmcb->ctrl.TLBCtrl.n.u8TLBFlush == SVM_TLB_FLUSH_ENTIRE);
1379 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushEntire);
1380 }
1381#endif
1382}
1383
1384
1385/**
1386 * Sets an exception intercept in the specified VMCB.
1387 *
1388 * @param pVmcb Pointer to the VM control block.
1389 * @param uXcpt The exception (X86_XCPT_*).
1390 */
1391DECLINLINE(void) hmR0SvmSetXcptIntercept(PSVMVMCB pVmcb, uint8_t uXcpt)
1392{
1393 if (!(pVmcb->ctrl.u32InterceptXcpt & RT_BIT(uXcpt)))
1394 {
1395 pVmcb->ctrl.u32InterceptXcpt |= RT_BIT(uXcpt);
1396 pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_INTERCEPTS;
1397 }
1398}
1399
1400
1401/**
1402 * Clears an exception intercept in the specified VMCB.
1403 *
1404 * @param pVCpu The cross context virtual CPU structure.
1405 * @param pVmcb Pointer to the VM control block.
1406 * @param uXcpt The exception (X86_XCPT_*).
1407 *
1408 * @remarks This takes into account if we're executing a nested-guest and only
1409 * removes the exception intercept if both the guest -and- nested-guest
1410 * are not intercepting it.
1411 */
1412DECLINLINE(void) hmR0SvmClearXcptIntercept(PVMCPUCC pVCpu, PSVMVMCB pVmcb, uint8_t uXcpt)
1413{
1414 Assert(uXcpt != X86_XCPT_DB);
1415 Assert(uXcpt != X86_XCPT_AC);
1416 Assert(uXcpt != X86_XCPT_GP);
1417#ifndef HMSVM_ALWAYS_TRAP_ALL_XCPTS
1418 if (pVmcb->ctrl.u32InterceptXcpt & RT_BIT(uXcpt))
1419 {
1420 bool fRemove = true;
1421# ifdef VBOX_WITH_NESTED_HWVIRT_SVM
1422 /* Only remove the intercept if the nested-guest is also not intercepting it! */
1423 PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
1424 if (CPUMIsGuestInSvmNestedHwVirtMode(pCtx))
1425 {
1426 PCSVMNESTEDVMCBCACHE pVmcbNstGstCache = hmR0SvmGetNestedVmcbCache(pVCpu);
1427 fRemove = !(pVmcbNstGstCache->u32InterceptXcpt & RT_BIT(uXcpt));
1428 }
1429# else
1430 RT_NOREF(pVCpu);
1431# endif
1432 if (fRemove)
1433 {
1434 pVmcb->ctrl.u32InterceptXcpt &= ~RT_BIT(uXcpt);
1435 pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_INTERCEPTS;
1436 }
1437 }
1438#else
1439 RT_NOREF3(pVCpu, pVmcb, uXcpt);
1440#endif
1441}
1442
1443
1444/**
1445 * Sets a control intercept in the specified VMCB.
1446 *
1447 * @param pVmcb Pointer to the VM control block.
1448 * @param fCtrlIntercept The control intercept (SVM_CTRL_INTERCEPT_*).
1449 */
1450DECLINLINE(void) hmR0SvmSetCtrlIntercept(PSVMVMCB pVmcb, uint64_t fCtrlIntercept)
1451{
1452 if (!(pVmcb->ctrl.u64InterceptCtrl & fCtrlIntercept))
1453 {
1454 pVmcb->ctrl.u64InterceptCtrl |= fCtrlIntercept;
1455 pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_INTERCEPTS;
1456 }
1457}
1458
1459
1460/**
1461 * Clears a control intercept in the specified VMCB.
1462 *
1463 * @returns @c true if the intercept is still set, @c false otherwise.
1464 * @param pVCpu The cross context virtual CPU structure.
1465 * @param pVmcb Pointer to the VM control block.
1466 * @param fCtrlIntercept The control intercept (SVM_CTRL_INTERCEPT_*).
1467 *
1468 * @remarks This takes into account if we're executing a nested-guest and only
1469 * removes the control intercept if both the guest -and- nested-guest
1470 * are not intercepting it.
1471 */
1472static bool hmR0SvmClearCtrlIntercept(PVMCPUCC pVCpu, PSVMVMCB pVmcb, uint64_t fCtrlIntercept)
1473{
1474 if (pVmcb->ctrl.u64InterceptCtrl & fCtrlIntercept)
1475 {
1476 bool fRemove = true;
1477#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
1478 /* Only remove the control intercept if the nested-guest is also not intercepting it! */
1479 if (CPUMIsGuestInSvmNestedHwVirtMode(&pVCpu->cpum.GstCtx))
1480 {
1481 PCSVMNESTEDVMCBCACHE pVmcbNstGstCache = hmR0SvmGetNestedVmcbCache(pVCpu);
1482 fRemove = !(pVmcbNstGstCache->u64InterceptCtrl & fCtrlIntercept);
1483 }
1484#else
1485 RT_NOREF(pVCpu);
1486#endif
1487 if (fRemove)
1488 {
1489 pVmcb->ctrl.u64InterceptCtrl &= ~fCtrlIntercept;
1490 pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_INTERCEPTS;
1491 }
1492 }
1493
1494 return RT_BOOL(pVmcb->ctrl.u64InterceptCtrl & fCtrlIntercept);
1495}
1496
1497
1498/**
1499 * Exports the guest (or nested-guest) CR0 into the VMCB.
1500 *
1501 * @param pVCpu The cross context virtual CPU structure.
1502 * @param pVmcb Pointer to the VM control block.
1503 *
1504 * @remarks This assumes we always pre-load the guest FPU.
1505 * @remarks No-long-jump zone!!!
1506 */
1507static void hmR0SvmExportGuestCR0(PVMCPUCC pVCpu, PSVMVMCB pVmcb)
1508{
1509 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1510
1511 PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
1512 uint64_t const uGuestCr0 = pCtx->cr0;
1513 uint64_t uShadowCr0 = uGuestCr0;
1514
1515 /* Always enable caching. */
1516 uShadowCr0 &= ~(X86_CR0_CD | X86_CR0_NW);
1517
1518 /* When Nested Paging is not available use shadow page tables and intercept #PFs (latter done in SVMR0SetupVM()). */
1519 if (!pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging)
1520 {
1521 uShadowCr0 |= X86_CR0_PG /* Use shadow page tables. */
1522 | X86_CR0_WP; /* Guest CPL 0 writes to its read-only pages should cause a #PF #VMEXIT. */
1523 }
1524
1525 /*
1526 * Use the #MF style of legacy-FPU error reporting for now. Although AMD-V has MSRs that
1527 * lets us isolate the host from it, IEM/REM still needs work to emulate it properly,
1528 * see @bugref{7243#c103}.
1529 */
1530 if (!(uGuestCr0 & X86_CR0_NE))
1531 {
1532 uShadowCr0 |= X86_CR0_NE;
1533 hmR0SvmSetXcptIntercept(pVmcb, X86_XCPT_MF);
1534 }
1535 else
1536 hmR0SvmClearXcptIntercept(pVCpu, pVmcb, X86_XCPT_MF);
1537
1538 /*
1539 * If the shadow and guest CR0 are identical we can avoid intercepting CR0 reads.
1540 *
1541 * CR0 writes still needs interception as PGM requires tracking paging mode changes,
1542 * see @bugref{6944}.
1543 *
1544 * We also don't ever want to honor weird things like cache disable from the guest.
1545 * However, we can avoid intercepting changes to the TS & MP bits by clearing the CR0
1546 * write intercept below and keeping SVM_CTRL_INTERCEPT_CR0_SEL_WRITE instead.
1547 */
1548 if (uShadowCr0 == uGuestCr0)
1549 {
1550 if (!CPUMIsGuestInSvmNestedHwVirtMode(pCtx))
1551 {
1552 pVmcb->ctrl.u16InterceptRdCRx &= ~RT_BIT(0);
1553 pVmcb->ctrl.u16InterceptWrCRx &= ~RT_BIT(0);
1554 Assert(pVmcb->ctrl.u64InterceptCtrl & SVM_CTRL_INTERCEPT_CR0_SEL_WRITE);
1555 }
1556 else
1557 {
1558 /* If the nested-hypervisor intercepts CR0 reads/writes, we need to continue intercepting them. */
1559 PCSVMNESTEDVMCBCACHE pVmcbNstGstCache = hmR0SvmGetNestedVmcbCache(pVCpu);
1560 pVmcb->ctrl.u16InterceptRdCRx = (pVmcb->ctrl.u16InterceptRdCRx & ~RT_BIT(0))
1561 | (pVmcbNstGstCache->u16InterceptRdCRx & RT_BIT(0));
1562 pVmcb->ctrl.u16InterceptWrCRx = (pVmcb->ctrl.u16InterceptWrCRx & ~RT_BIT(0))
1563 | (pVmcbNstGstCache->u16InterceptWrCRx & RT_BIT(0));
1564 }
1565 }
1566 else
1567 {
1568 pVmcb->ctrl.u16InterceptRdCRx |= RT_BIT(0);
1569 pVmcb->ctrl.u16InterceptWrCRx |= RT_BIT(0);
1570 }
1571 pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_INTERCEPTS;
1572
1573 Assert(!RT_HI_U32(uShadowCr0));
1574 if (pVmcb->guest.u64CR0 != uShadowCr0)
1575 {
1576 pVmcb->guest.u64CR0 = uShadowCr0;
1577 pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_CRX_EFER;
1578 }
1579}
1580
1581
1582/**
1583 * Exports the guest (or nested-guest) CR3 into the VMCB.
1584 *
1585 * @param pVCpu The cross context virtual CPU structure.
1586 * @param pVmcb Pointer to the VM control block.
1587 *
1588 * @remarks No-long-jump zone!!!
1589 */
1590static void hmR0SvmExportGuestCR3(PVMCPUCC pVCpu, PSVMVMCB pVmcb)
1591{
1592 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1593
1594 PVMCC pVM = pVCpu->CTX_SUFF(pVM);
1595 PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
1596 if (pVM->hmr0.s.fNestedPaging)
1597 {
1598 pVmcb->ctrl.u64NestedPagingCR3 = PGMGetHyperCR3(pVCpu);
1599 pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_NP;
1600 pVmcb->guest.u64CR3 = pCtx->cr3;
1601 Assert(pVmcb->ctrl.u64NestedPagingCR3);
1602 }
1603 else
1604 pVmcb->guest.u64CR3 = PGMGetHyperCR3(pVCpu);
1605
1606 pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_CRX_EFER;
1607}
1608
1609
1610/**
1611 * Exports the guest (or nested-guest) CR4 into the VMCB.
1612 *
1613 * @param pVCpu The cross context virtual CPU structure.
1614 * @param pVmcb Pointer to the VM control block.
1615 *
1616 * @remarks No-long-jump zone!!!
1617 */
1618static int hmR0SvmExportGuestCR4(PVMCPUCC pVCpu, PSVMVMCB pVmcb)
1619{
1620 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1621
1622 PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
1623 uint64_t uShadowCr4 = pCtx->cr4;
1624 if (!pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging)
1625 {
1626 switch (pVCpu->hm.s.enmShadowMode)
1627 {
1628 case PGMMODE_REAL:
1629 case PGMMODE_PROTECTED: /* Protected mode, no paging. */
1630 return VERR_PGM_UNSUPPORTED_SHADOW_PAGING_MODE;
1631
1632 case PGMMODE_32_BIT: /* 32-bit paging. */
1633 uShadowCr4 &= ~X86_CR4_PAE;
1634 break;
1635
1636 case PGMMODE_PAE: /* PAE paging. */
1637 case PGMMODE_PAE_NX: /* PAE paging with NX enabled. */
1638 /** Must use PAE paging as we could use physical memory > 4 GB */
1639 uShadowCr4 |= X86_CR4_PAE;
1640 break;
1641
1642 case PGMMODE_AMD64: /* 64-bit AMD paging (long mode). */
1643 case PGMMODE_AMD64_NX: /* 64-bit AMD paging (long mode) with NX enabled. */
1644#ifdef VBOX_WITH_64_BITS_GUESTS
1645 break;
1646#else
1647 return VERR_PGM_UNSUPPORTED_SHADOW_PAGING_MODE;
1648#endif
1649
1650 default: /* shut up gcc */
1651 return VERR_PGM_UNSUPPORTED_SHADOW_PAGING_MODE;
1652 }
1653 }
1654
1655 /* Whether to save/load/restore XCR0 during world switch depends on CR4.OSXSAVE and host+guest XCR0. */
1656 bool const fLoadSaveGuestXcr0 = (pCtx->cr4 & X86_CR4_OSXSAVE) && pCtx->aXcr[0] != ASMGetXcr0();
1657 if (fLoadSaveGuestXcr0 != pVCpu->hmr0.s.fLoadSaveGuestXcr0)
1658 {
1659 pVCpu->hmr0.s.fLoadSaveGuestXcr0 = fLoadSaveGuestXcr0;
1660 hmR0SvmUpdateVmRunFunction(pVCpu);
1661 }
1662
1663 /* Avoid intercepting CR4 reads if the guest and shadow CR4 values are identical. */
1664 if (uShadowCr4 == pCtx->cr4)
1665 {
1666 if (!CPUMIsGuestInSvmNestedHwVirtMode(pCtx))
1667 pVmcb->ctrl.u16InterceptRdCRx &= ~RT_BIT(4);
1668 else
1669 {
1670 /* If the nested-hypervisor intercepts CR4 reads, we need to continue intercepting them. */
1671 PCSVMNESTEDVMCBCACHE pVmcbNstGstCache = hmR0SvmGetNestedVmcbCache(pVCpu);
1672 pVmcb->ctrl.u16InterceptRdCRx = (pVmcb->ctrl.u16InterceptRdCRx & ~RT_BIT(4))
1673 | (pVmcbNstGstCache->u16InterceptRdCRx & RT_BIT(4));
1674 }
1675 }
1676 else
1677 pVmcb->ctrl.u16InterceptRdCRx |= RT_BIT(4);
1678
1679 /* CR4 writes are always intercepted (both guest, nested-guest) for tracking
1680 PGM mode changes and AVX (for XCR0 syncing during worlds switching). */
1681 Assert(pVmcb->ctrl.u16InterceptWrCRx & RT_BIT(4));
1682
1683 /* Update VMCB with the shadow CR4 the appropriate VMCB clean bits. */
1684 Assert(!RT_HI_U32(uShadowCr4));
1685 pVmcb->guest.u64CR4 = uShadowCr4;
1686 pVmcb->ctrl.u32VmcbCleanBits &= ~(HMSVM_VMCB_CLEAN_CRX_EFER | HMSVM_VMCB_CLEAN_INTERCEPTS);
1687
1688 return VINF_SUCCESS;
1689}
1690
1691
1692/**
1693 * Exports the guest (or nested-guest) control registers into the VMCB.
1694 *
1695 * @returns VBox status code.
1696 * @param pVCpu The cross context virtual CPU structure.
1697 * @param pVmcb Pointer to the VM control block.
1698 *
1699 * @remarks No-long-jump zone!!!
1700 */
1701static int hmR0SvmExportGuestControlRegs(PVMCPUCC pVCpu, PSVMVMCB pVmcb)
1702{
1703 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1704
1705 if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_CR_MASK)
1706 {
1707 if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_CR0)
1708 hmR0SvmExportGuestCR0(pVCpu, pVmcb);
1709
1710 if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_CR2)
1711 {
1712 pVmcb->guest.u64CR2 = pVCpu->cpum.GstCtx.cr2;
1713 pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_CR2;
1714 }
1715
1716 if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_CR3)
1717 hmR0SvmExportGuestCR3(pVCpu, pVmcb);
1718
1719 /* CR4 re-loading is ASSUMED to be done everytime we get in from ring-3! (XCR0) */
1720 if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_CR4)
1721 {
1722 int rc = hmR0SvmExportGuestCR4(pVCpu, pVmcb);
1723 if (RT_FAILURE(rc))
1724 return rc;
1725 }
1726
1727 pVCpu->hm.s.fCtxChanged &= ~HM_CHANGED_GUEST_CR_MASK;
1728 }
1729 return VINF_SUCCESS;
1730}
1731
1732
1733/**
1734 * Exports the guest (or nested-guest) segment registers into the VMCB.
1735 *
1736 * @returns VBox status code.
1737 * @param pVCpu The cross context virtual CPU structure.
1738 * @param pVmcb Pointer to the VM control block.
1739 *
1740 * @remarks No-long-jump zone!!!
1741 */
1742static void hmR0SvmExportGuestSegmentRegs(PVMCPUCC pVCpu, PSVMVMCB pVmcb)
1743{
1744 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1745 PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
1746
1747 /* Guest segment registers. */
1748 if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_SREG_MASK)
1749 {
1750 if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_CS)
1751 HMSVM_SEG_REG_COPY_TO_VMCB(pCtx, &pVmcb->guest, CS, cs);
1752
1753 if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_SS)
1754 {
1755 HMSVM_SEG_REG_COPY_TO_VMCB(pCtx, &pVmcb->guest, SS, ss);
1756 pVmcb->guest.u8CPL = pCtx->ss.Attr.n.u2Dpl;
1757 }
1758
1759 if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_DS)
1760 HMSVM_SEG_REG_COPY_TO_VMCB(pCtx, &pVmcb->guest, DS, ds);
1761
1762 if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_ES)
1763 HMSVM_SEG_REG_COPY_TO_VMCB(pCtx, &pVmcb->guest, ES, es);
1764
1765 if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_FS)
1766 HMSVM_SEG_REG_COPY_TO_VMCB(pCtx, &pVmcb->guest, FS, fs);
1767
1768 if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_GS)
1769 HMSVM_SEG_REG_COPY_TO_VMCB(pCtx, &pVmcb->guest, GS, gs);
1770
1771 pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_SEG;
1772 }
1773
1774 /* Guest TR. */
1775 if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_TR)
1776 HMSVM_SEG_REG_COPY_TO_VMCB(pCtx, &pVmcb->guest, TR, tr);
1777
1778 /* Guest LDTR. */
1779 if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_LDTR)
1780 HMSVM_SEG_REG_COPY_TO_VMCB(pCtx, &pVmcb->guest, LDTR, ldtr);
1781
1782 /* Guest GDTR. */
1783 if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_GDTR)
1784 {
1785 pVmcb->guest.GDTR.u32Limit = pCtx->gdtr.cbGdt;
1786 pVmcb->guest.GDTR.u64Base = pCtx->gdtr.pGdt;
1787 pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_DT;
1788 }
1789
1790 /* Guest IDTR. */
1791 if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_IDTR)
1792 {
1793 pVmcb->guest.IDTR.u32Limit = pCtx->idtr.cbIdt;
1794 pVmcb->guest.IDTR.u64Base = pCtx->idtr.pIdt;
1795 pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_DT;
1796 }
1797
1798 pVCpu->hm.s.fCtxChanged &= ~( HM_CHANGED_GUEST_SREG_MASK
1799 | HM_CHANGED_GUEST_TABLE_MASK);
1800}
1801
1802
1803/**
1804 * Exports the guest (or nested-guest) MSRs into the VMCB.
1805 *
1806 * @param pVCpu The cross context virtual CPU structure.
1807 * @param pVmcb Pointer to the VM control block.
1808 *
1809 * @remarks No-long-jump zone!!!
1810 */
1811static void hmR0SvmExportGuestMsrs(PVMCPUCC pVCpu, PSVMVMCB pVmcb)
1812{
1813 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1814 PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
1815
1816 /* Guest Sysenter MSRs. */
1817 if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_SYSENTER_MSR_MASK)
1818 {
1819 if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_SYSENTER_CS_MSR)
1820 pVmcb->guest.u64SysEnterCS = pCtx->SysEnter.cs;
1821
1822 if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_SYSENTER_EIP_MSR)
1823 pVmcb->guest.u64SysEnterEIP = pCtx->SysEnter.eip;
1824
1825 if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_SYSENTER_ESP_MSR)
1826 pVmcb->guest.u64SysEnterESP = pCtx->SysEnter.esp;
1827 }
1828
1829 /*
1830 * Guest EFER MSR.
1831 * AMD-V requires guest EFER.SVME to be set. Weird.
1832 * See AMD spec. 15.5.1 "Basic Operation" | "Canonicalization and Consistency Checks".
1833 */
1834 if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_EFER_MSR)
1835 {
1836 pVmcb->guest.u64EFER = pCtx->msrEFER | MSR_K6_EFER_SVME;
1837 pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_CRX_EFER;
1838 }
1839
1840 /* If the guest isn't in 64-bit mode, clear MSR_K6_LME bit, otherwise SVM expects amd64 shadow paging. */
1841 if ( !CPUMIsGuestInLongModeEx(pCtx)
1842 && (pCtx->msrEFER & MSR_K6_EFER_LME))
1843 {
1844 pVmcb->guest.u64EFER &= ~MSR_K6_EFER_LME;
1845 pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_CRX_EFER;
1846 }
1847
1848 if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_SYSCALL_MSRS)
1849 {
1850 pVmcb->guest.u64STAR = pCtx->msrSTAR;
1851 pVmcb->guest.u64LSTAR = pCtx->msrLSTAR;
1852 pVmcb->guest.u64CSTAR = pCtx->msrCSTAR;
1853 pVmcb->guest.u64SFMASK = pCtx->msrSFMASK;
1854 }
1855
1856 if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_KERNEL_GS_BASE)
1857 pVmcb->guest.u64KernelGSBase = pCtx->msrKERNELGSBASE;
1858
1859 pVCpu->hm.s.fCtxChanged &= ~( HM_CHANGED_GUEST_SYSENTER_MSR_MASK
1860 | HM_CHANGED_GUEST_EFER_MSR
1861 | HM_CHANGED_GUEST_SYSCALL_MSRS
1862 | HM_CHANGED_GUEST_KERNEL_GS_BASE);
1863
1864 /*
1865 * Setup the PAT MSR (applicable for Nested Paging only).
1866 *
1867 * The default value should be MSR_IA32_CR_PAT_INIT_VAL, but we treat all guest memory
1868 * as WB, so choose type 6 for all PAT slots, see @bugref{9634}.
1869 *
1870 * While guests can modify and see the modified values through the shadow values,
1871 * we shall not honor any guest modifications of this MSR to ensure caching is always
1872 * enabled similar to how we clear CR0.CD and NW bits.
1873 *
1874 * For nested-guests this needs to always be set as well, see @bugref{7243#c109}.
1875 */
1876 pVmcb->guest.u64PAT = UINT64_C(0x0006060606060606);
1877
1878 /* Enable the last branch record bit if LBR virtualization is enabled. */
1879 if (pVmcb->ctrl.LbrVirt.n.u1LbrVirt)
1880 pVmcb->guest.u64DBGCTL = MSR_IA32_DEBUGCTL_LBR;
1881}
1882
1883
1884/**
1885 * Exports the guest (or nested-guest) debug state into the VMCB and programs
1886 * the necessary intercepts accordingly.
1887 *
1888 * @param pVCpu The cross context virtual CPU structure.
1889 * @param pVmcb Pointer to the VM control block.
1890 *
1891 * @remarks No-long-jump zone!!!
1892 * @remarks Requires EFLAGS to be up-to-date in the VMCB!
1893 */
1894static void hmR0SvmExportSharedDebugState(PVMCPUCC pVCpu, PSVMVMCB pVmcb)
1895{
1896 PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
1897
1898 /** @todo Figure out stepping with nested-guest. */
1899 if (CPUMIsGuestInSvmNestedHwVirtMode(pCtx))
1900 {
1901 /*
1902 * We don't want to always intercept DRx read/writes for nested-guests as it causes
1903 * problems when the nested hypervisor isn't intercepting them, see @bugref{10080}.
1904 * Instead, they are strictly only requested when the nested hypervisor intercepts
1905 * them -- handled while merging VMCB controls.
1906 *
1907 * If neither the outer nor the nested-hypervisor is intercepting DRx read/writes,
1908 * then the nested-guest debug state should be actively loaded on the host so that
1909 * nested-guest reads/writes its own debug registers without causing VM-exits.
1910 */
1911 if ( ( pVmcb->ctrl.u16InterceptRdDRx != 0xffff
1912 || pVmcb->ctrl.u16InterceptWrDRx != 0xffff)
1913 && !CPUMIsGuestDebugStateActive(pVCpu))
1914 {
1915 CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */);
1916 STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxArmed);
1917 Assert(!CPUMIsHyperDebugStateActive(pVCpu));
1918 Assert(CPUMIsGuestDebugStateActive(pVCpu));
1919 }
1920
1921 pVmcb->guest.u64DR6 = pCtx->dr[6];
1922 pVmcb->guest.u64DR7 = pCtx->dr[7];
1923 return;
1924 }
1925
1926 /*
1927 * Anyone single stepping on the host side? If so, we'll have to use the
1928 * trap flag in the guest EFLAGS since AMD-V doesn't have a trap flag on
1929 * the VMM level like the VT-x implementations does.
1930 */
1931 bool fInterceptMovDRx = false;
1932 bool const fStepping = pVCpu->hm.s.fSingleInstruction || DBGFIsStepping(pVCpu);
1933 if (fStepping)
1934 {
1935 pVCpu->hmr0.s.fClearTrapFlag = true;
1936 pVmcb->guest.u64RFlags |= X86_EFL_TF;
1937 fInterceptMovDRx = true; /* Need clean DR6, no guest mess. */
1938 }
1939
1940 if ( fStepping
1941 || (CPUMGetHyperDR7(pVCpu) & X86_DR7_ENABLED_MASK))
1942 {
1943 /*
1944 * Use the combined guest and host DRx values found in the hypervisor
1945 * register set because the debugger has breakpoints active or someone
1946 * is single stepping on the host side.
1947 *
1948 * Note! DBGF expects a clean DR6 state before executing guest code.
1949 */
1950 if (!CPUMIsHyperDebugStateActive(pVCpu))
1951 {
1952 CPUMR0LoadHyperDebugState(pVCpu, false /* include DR6 */);
1953 Assert(!CPUMIsGuestDebugStateActive(pVCpu));
1954 Assert(CPUMIsHyperDebugStateActive(pVCpu));
1955 }
1956
1957 /* Update DR6 & DR7. (The other DRx values are handled by CPUM one way or the other.) */
1958 if ( pVmcb->guest.u64DR6 != X86_DR6_INIT_VAL
1959 || pVmcb->guest.u64DR7 != CPUMGetHyperDR7(pVCpu))
1960 {
1961 pVmcb->guest.u64DR7 = CPUMGetHyperDR7(pVCpu);
1962 pVmcb->guest.u64DR6 = X86_DR6_INIT_VAL;
1963 pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_DRX;
1964 }
1965
1966 /** @todo If we cared, we could optimize to allow the guest to read registers
1967 * with the same values. */
1968 fInterceptMovDRx = true;
1969 pVCpu->hmr0.s.fUsingHyperDR7 = true;
1970 Log5(("hmR0SvmExportSharedDebugState: Loaded hyper DRx\n"));
1971 }
1972 else
1973 {
1974 /*
1975 * Update DR6, DR7 with the guest values if necessary.
1976 */
1977 if ( pVmcb->guest.u64DR7 != pCtx->dr[7]
1978 || pVmcb->guest.u64DR6 != pCtx->dr[6])
1979 {
1980 pVmcb->guest.u64DR7 = pCtx->dr[7];
1981 pVmcb->guest.u64DR6 = pCtx->dr[6];
1982 pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_DRX;
1983 }
1984 pVCpu->hmr0.s.fUsingHyperDR7 = false;
1985
1986 /*
1987 * If the guest has enabled debug registers, we need to load them prior to
1988 * executing guest code so they'll trigger at the right time.
1989 */
1990 if (pCtx->dr[7] & (X86_DR7_ENABLED_MASK | X86_DR7_GD)) /** @todo Why GD? */
1991 {
1992 if (!CPUMIsGuestDebugStateActive(pVCpu))
1993 {
1994 CPUMR0LoadGuestDebugState(pVCpu, false /* include DR6 */);
1995 STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxArmed);
1996 Assert(!CPUMIsHyperDebugStateActive(pVCpu));
1997 Assert(CPUMIsGuestDebugStateActive(pVCpu));
1998 }
1999 Log5(("hmR0SvmExportSharedDebugState: Loaded guest DRx\n"));
2000 }
2001 /*
2002 * If no debugging enabled, we'll lazy load DR0-3. We don't need to
2003 * intercept #DB as DR6 is updated in the VMCB.
2004 *
2005 * Note! If we cared and dared, we could skip intercepting \#DB here.
2006 * However, \#DB shouldn't be performance critical, so we'll play safe
2007 * and keep the code similar to the VT-x code and always intercept it.
2008 */
2009 else if (!CPUMIsGuestDebugStateActive(pVCpu))
2010 fInterceptMovDRx = true;
2011 }
2012
2013 Assert(pVmcb->ctrl.u32InterceptXcpt & RT_BIT_32(X86_XCPT_DB));
2014 if (fInterceptMovDRx)
2015 {
2016 if ( pVmcb->ctrl.u16InterceptRdDRx != 0xffff
2017 || pVmcb->ctrl.u16InterceptWrDRx != 0xffff)
2018 {
2019 pVmcb->ctrl.u16InterceptRdDRx = 0xffff;
2020 pVmcb->ctrl.u16InterceptWrDRx = 0xffff;
2021 pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_INTERCEPTS;
2022 }
2023 }
2024 else
2025 {
2026 if ( pVmcb->ctrl.u16InterceptRdDRx
2027 || pVmcb->ctrl.u16InterceptWrDRx)
2028 {
2029 pVmcb->ctrl.u16InterceptRdDRx = 0;
2030 pVmcb->ctrl.u16InterceptWrDRx = 0;
2031 pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_INTERCEPTS;
2032 }
2033 }
2034 Log4Func(("DR6=%#RX64 DR7=%#RX64\n", pCtx->dr[6], pCtx->dr[7]));
2035}
2036
2037/**
2038 * Exports the hardware virtualization state into the nested-guest
2039 * VMCB.
2040 *
2041 * @param pVCpu The cross context virtual CPU structure.
2042 * @param pVmcb Pointer to the VM control block.
2043 *
2044 * @remarks No-long-jump zone!!!
2045 */
2046static void hmR0SvmExportGuestHwvirtState(PVMCPUCC pVCpu, PSVMVMCB pVmcb)
2047{
2048 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2049
2050 if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_HWVIRT)
2051 {
2052 if (pVmcb->ctrl.IntCtrl.n.u1VGifEnable)
2053 {
2054 PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
2055 PCVMCC pVM = pVCpu->CTX_SUFF(pVM);
2056
2057 HMSVM_ASSERT_NOT_IN_NESTED_GUEST(pCtx); /* Nested VGIF is not supported yet. */
2058 Assert(g_fHmSvmFeatures & X86_CPUID_SVM_FEATURE_EDX_VGIF); /* Physical hardware supports VGIF. */
2059 Assert(HMIsSvmVGifActive(pVM)); /* Outer VM has enabled VGIF. */
2060 NOREF(pVM);
2061
2062 pVmcb->ctrl.IntCtrl.n.u1VGif = CPUMGetGuestGif(pCtx);
2063 }
2064
2065 /*
2066 * Ensure the nested-guest pause-filter counters don't exceed the outer guest values esp.
2067 * since SVM doesn't have a preemption timer.
2068 *
2069 * We do this here rather than in hmR0SvmSetupVmcbNested() as we may have been executing the
2070 * nested-guest in IEM incl. PAUSE instructions which would update the pause-filter counters
2071 * and may continue execution in SVM R0 without a nested-guest #VMEXIT in between.
2072 */
2073 PVMCC pVM = pVCpu->CTX_SUFF(pVM);
2074 PSVMVMCBCTRL pVmcbCtrl = &pVmcb->ctrl;
2075 uint16_t const uGuestPauseFilterCount = pVM->hm.s.svm.cPauseFilter;
2076 uint16_t const uGuestPauseFilterThreshold = pVM->hm.s.svm.cPauseFilterThresholdTicks;
2077 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, &pVCpu->cpum.GstCtx, SVM_CTRL_INTERCEPT_PAUSE))
2078 {
2079 PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
2080 pVmcbCtrl->u16PauseFilterCount = RT_MIN(pCtx->hwvirt.svm.cPauseFilter, uGuestPauseFilterCount);
2081 pVmcbCtrl->u16PauseFilterThreshold = RT_MIN(pCtx->hwvirt.svm.cPauseFilterThreshold, uGuestPauseFilterThreshold);
2082 }
2083 else
2084 {
2085 /** @todo r=ramshankar: We can turn these assignments into assertions. */
2086 pVmcbCtrl->u16PauseFilterCount = uGuestPauseFilterCount;
2087 pVmcbCtrl->u16PauseFilterThreshold = uGuestPauseFilterThreshold;
2088 }
2089 pVmcbCtrl->u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_INTERCEPTS;
2090
2091 pVCpu->hm.s.fCtxChanged &= ~HM_CHANGED_GUEST_HWVIRT;
2092 }
2093}
2094
2095
2096/**
2097 * Exports the guest APIC TPR state into the VMCB.
2098 *
2099 * @returns VBox status code.
2100 * @param pVCpu The cross context virtual CPU structure.
2101 * @param pVmcb Pointer to the VM control block.
2102 */
2103static int hmR0SvmExportGuestApicTpr(PVMCPUCC pVCpu, PSVMVMCB pVmcb)
2104{
2105 HMSVM_ASSERT_NOT_IN_NESTED_GUEST(&pVCpu->cpum.GstCtx);
2106
2107 if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_GUEST_APIC_TPR)
2108 {
2109 PVMCC pVM = pVCpu->CTX_SUFF(pVM);
2110 if ( PDMHasApic(pVM)
2111 && APICIsEnabled(pVCpu))
2112 {
2113 bool fPendingIntr;
2114 uint8_t u8Tpr;
2115 int rc = APICGetTpr(pVCpu, &u8Tpr, &fPendingIntr, NULL /* pu8PendingIrq */);
2116 AssertRCReturn(rc, rc);
2117
2118 /* Assume that we need to trap all TPR accesses and thus need not check on
2119 every #VMEXIT if we should update the TPR. */
2120 Assert(pVmcb->ctrl.IntCtrl.n.u1VIntrMasking);
2121 pVCpu->hmr0.s.svm.fSyncVTpr = false;
2122
2123 if (!pVM->hm.s.fTprPatchingActive)
2124 {
2125 /* Bits 3-0 of the VTPR field correspond to bits 7-4 of the TPR (which is the Task-Priority Class). */
2126 pVmcb->ctrl.IntCtrl.n.u8VTPR = (u8Tpr >> 4);
2127
2128 /* If there are interrupts pending, intercept CR8 writes to evaluate ASAP if we
2129 can deliver the interrupt to the guest. */
2130 if (fPendingIntr)
2131 pVmcb->ctrl.u16InterceptWrCRx |= RT_BIT(8);
2132 else
2133 {
2134 pVmcb->ctrl.u16InterceptWrCRx &= ~RT_BIT(8);
2135 pVCpu->hmr0.s.svm.fSyncVTpr = true;
2136 }
2137
2138 pVmcb->ctrl.u32VmcbCleanBits &= ~(HMSVM_VMCB_CLEAN_INTERCEPTS | HMSVM_VMCB_CLEAN_INT_CTRL);
2139 }
2140 else
2141 {
2142 /* 32-bit guests uses LSTAR MSR for patching guest code which touches the TPR. */
2143 pVmcb->guest.u64LSTAR = u8Tpr;
2144 uint8_t *pbMsrBitmap = (uint8_t *)pVCpu->hmr0.s.svm.pvMsrBitmap;
2145
2146 /* If there are interrupts pending, intercept LSTAR writes, otherwise don't intercept reads or writes. */
2147 if (fPendingIntr)
2148 hmR0SvmSetMsrPermission(pVCpu, pbMsrBitmap, MSR_K8_LSTAR, SVMMSREXIT_PASSTHRU_READ, SVMMSREXIT_INTERCEPT_WRITE);
2149 else
2150 {
2151 hmR0SvmSetMsrPermission(pVCpu, pbMsrBitmap, MSR_K8_LSTAR, SVMMSREXIT_PASSTHRU_READ, SVMMSREXIT_PASSTHRU_WRITE);
2152 pVCpu->hmr0.s.svm.fSyncVTpr = true;
2153 }
2154 pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_IOPM_MSRPM;
2155 }
2156 }
2157 ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_GUEST_APIC_TPR);
2158 }
2159 return VINF_SUCCESS;
2160}
2161
2162
2163/**
2164 * Sets up the exception interrupts required for guest execution in the VMCB.
2165 *
2166 * @param pVCpu The cross context virtual CPU structure.
2167 * @param pVmcb Pointer to the VM control block.
2168 *
2169 * @remarks No-long-jump zone!!!
2170 */
2171static void hmR0SvmExportGuestXcptIntercepts(PVMCPUCC pVCpu, PSVMVMCB pVmcb)
2172{
2173 HMSVM_ASSERT_NOT_IN_NESTED_GUEST(&pVCpu->cpum.GstCtx);
2174
2175 /* If we modify intercepts from here, please check & adjust hmR0SvmMergeVmcbCtrlsNested() if required. */
2176 if (ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged) & HM_CHANGED_SVM_XCPT_INTERCEPTS)
2177 {
2178 /* Trap #UD for GIM provider (e.g. for hypercalls). */
2179 if (pVCpu->hm.s.fGIMTrapXcptUD || pVCpu->hm.s.svm.fEmulateLongModeSysEnterExit)
2180 hmR0SvmSetXcptIntercept(pVmcb, X86_XCPT_UD);
2181 else
2182 hmR0SvmClearXcptIntercept(pVCpu, pVmcb, X86_XCPT_UD);
2183
2184 /* Trap #BP for INT3 debug breakpoints set by the VM debugger. */
2185 if (pVCpu->CTX_SUFF(pVM)->dbgf.ro.cEnabledInt3Breakpoints)
2186 hmR0SvmSetXcptIntercept(pVmcb, X86_XCPT_BP);
2187 else
2188 hmR0SvmClearXcptIntercept(pVCpu, pVmcb, X86_XCPT_BP);
2189
2190 /* The remaining intercepts are handled elsewhere, e.g. in hmR0SvmExportGuestCR0(). */
2191 ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_SVM_XCPT_INTERCEPTS);
2192 }
2193}
2194
2195
2196#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
2197/**
2198 * Merges guest and nested-guest intercepts for executing the nested-guest using
2199 * hardware-assisted SVM.
2200 *
2201 * This merges the guest and nested-guest intercepts in a way that if the outer
2202 * guest intercept is set we need to intercept it in the nested-guest as
2203 * well.
2204 *
2205 * @param pVCpu The cross context virtual CPU structure.
2206 * @param pVmcbNstGst Pointer to the nested-guest VM control block.
2207 */
2208static void hmR0SvmMergeVmcbCtrlsNested(PVMCPUCC pVCpu)
2209{
2210 PVMCC pVM = pVCpu->CTX_SUFF(pVM);
2211 PCSVMVMCB pVmcb = pVCpu->hmr0.s.svm.pVmcb;
2212 PSVMVMCB pVmcbNstGst = &pVCpu->cpum.GstCtx.hwvirt.svm.Vmcb;
2213 PSVMVMCBCTRL pVmcbNstGstCtrl = &pVmcbNstGst->ctrl;
2214
2215 /* Merge the guest's CR intercepts into the nested-guest VMCB. */
2216 pVmcbNstGstCtrl->u16InterceptRdCRx |= pVmcb->ctrl.u16InterceptRdCRx;
2217 pVmcbNstGstCtrl->u16InterceptWrCRx |= pVmcb->ctrl.u16InterceptWrCRx;
2218
2219 /* Always intercept CR4 writes for tracking PGM mode changes and AVX (for
2220 XCR0 syncing during worlds switching). */
2221 pVmcbNstGstCtrl->u16InterceptWrCRx |= RT_BIT(4);
2222
2223 /* Without nested paging, intercept CR3 reads and writes as we load shadow page tables. */
2224 if (!pVM->hmr0.s.fNestedPaging)
2225 {
2226 pVmcbNstGstCtrl->u16InterceptRdCRx |= RT_BIT(3);
2227 pVmcbNstGstCtrl->u16InterceptWrCRx |= RT_BIT(3);
2228 }
2229
2230 /* Merge the guest's DR intercepts into the nested-guest VMCB. */
2231 pVmcbNstGstCtrl->u16InterceptRdDRx |= pVmcb->ctrl.u16InterceptRdDRx;
2232 pVmcbNstGstCtrl->u16InterceptWrDRx |= pVmcb->ctrl.u16InterceptWrDRx;
2233
2234 /*
2235 * Merge the guest's exception intercepts into the nested-guest VMCB.
2236 *
2237 * - #UD: Exclude these as the outer guest's GIM hypercalls are not applicable
2238 * while executing the nested-guest.
2239 *
2240 * - #BP: Exclude breakpoints set by the VM debugger for the outer guest. This can
2241 * be tweaked later depending on how we wish to implement breakpoints.
2242 *
2243 * - #GP: Exclude these as it's the inner VMMs problem to get vmsvga 3d drivers
2244 * loaded into their guests, not ours.
2245 *
2246 * Warning!! This ASSUMES we only intercept \#UD for hypercall purposes and \#BP
2247 * for VM debugger breakpoints, see hmR0SvmExportGuestXcptIntercepts().
2248 */
2249#ifndef HMSVM_ALWAYS_TRAP_ALL_XCPTS
2250 pVmcbNstGstCtrl->u32InterceptXcpt |= pVmcb->ctrl.u32InterceptXcpt
2251 & ~( RT_BIT(X86_XCPT_UD)
2252 | RT_BIT(X86_XCPT_BP)
2253 | (pVCpu->hm.s.fTrapXcptGpForLovelyMesaDrv ? RT_BIT(X86_XCPT_GP) : 0));
2254#else
2255 pVmcbNstGstCtrl->u32InterceptXcpt |= pVmcb->ctrl.u32InterceptXcpt;
2256#endif
2257
2258 /*
2259 * Adjust intercepts while executing the nested-guest that differ from the
2260 * outer guest intercepts.
2261 *
2262 * - VINTR: Exclude the outer guest intercept as we don't need to cause VINTR #VMEXITs
2263 * that belong to the nested-guest to the outer guest.
2264 *
2265 * - VMMCALL: Exclude the outer guest intercept as when it's also not intercepted by
2266 * the nested-guest, the physical CPU raises a \#UD exception as expected.
2267 */
2268 pVmcbNstGstCtrl->u64InterceptCtrl |= (pVmcb->ctrl.u64InterceptCtrl & ~( SVM_CTRL_INTERCEPT_VINTR
2269 | SVM_CTRL_INTERCEPT_VMMCALL))
2270 | HMSVM_MANDATORY_GUEST_CTRL_INTERCEPTS;
2271
2272 Assert( (pVmcbNstGstCtrl->u64InterceptCtrl & HMSVM_MANDATORY_GUEST_CTRL_INTERCEPTS)
2273 == HMSVM_MANDATORY_GUEST_CTRL_INTERCEPTS);
2274
2275 /* Finally, update the VMCB clean bits. */
2276 pVmcbNstGstCtrl->u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_INTERCEPTS;
2277}
2278#endif
2279
2280
2281/**
2282 * Enters the AMD-V session.
2283 *
2284 * @returns VBox status code.
2285 * @param pVCpu The cross context virtual CPU structure.
2286 */
2287VMMR0DECL(int) SVMR0Enter(PVMCPUCC pVCpu)
2288{
2289 AssertPtr(pVCpu);
2290 Assert(pVCpu->CTX_SUFF(pVM)->hm.s.svm.fSupported);
2291 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2292
2293 LogFlowFunc(("pVCpu=%p\n", pVCpu));
2294 Assert((pVCpu->hm.s.fCtxChanged & (HM_CHANGED_HOST_CONTEXT | HM_CHANGED_SVM_HOST_GUEST_SHARED_STATE))
2295 == (HM_CHANGED_HOST_CONTEXT | HM_CHANGED_SVM_HOST_GUEST_SHARED_STATE));
2296
2297 pVCpu->hmr0.s.fLeaveDone = false;
2298 return VINF_SUCCESS;
2299}
2300
2301
2302/**
2303 * Thread-context callback for AMD-V.
2304 *
2305 * This is used together with RTThreadCtxHookCreate() on platforms which
2306 * supports it, and directly from VMMR0EmtPrepareForBlocking() and
2307 * VMMR0EmtResumeAfterBlocking() on platforms which don't.
2308 *
2309 * @param enmEvent The thread-context event.
2310 * @param pVCpu The cross context virtual CPU structure.
2311 * @param fGlobalInit Whether global VT-x/AMD-V init. is used.
2312 * @thread EMT(pVCpu)
2313 */
2314VMMR0DECL(void) SVMR0ThreadCtxCallback(RTTHREADCTXEVENT enmEvent, PVMCPUCC pVCpu, bool fGlobalInit)
2315{
2316 NOREF(fGlobalInit);
2317
2318 switch (enmEvent)
2319 {
2320 case RTTHREADCTXEVENT_OUT:
2321 {
2322 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2323 VMCPU_ASSERT_EMT(pVCpu);
2324
2325 /* No longjmps (log-flush, locks) in this fragile context. */
2326 VMMRZCallRing3Disable(pVCpu);
2327
2328 if (!pVCpu->hmr0.s.fLeaveDone)
2329 {
2330 hmR0SvmLeave(pVCpu, false /* fImportState */);
2331 pVCpu->hmr0.s.fLeaveDone = true;
2332 }
2333
2334 /* Leave HM context, takes care of local init (term). */
2335 int rc = HMR0LeaveCpu(pVCpu);
2336 AssertRC(rc); NOREF(rc);
2337
2338 /* Restore longjmp state. */
2339 VMMRZCallRing3Enable(pVCpu);
2340 STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatSwitchPreempt);
2341 break;
2342 }
2343
2344 case RTTHREADCTXEVENT_IN:
2345 {
2346 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2347 VMCPU_ASSERT_EMT(pVCpu);
2348
2349 /* No longjmps (log-flush, locks) in this fragile context. */
2350 VMMRZCallRing3Disable(pVCpu);
2351
2352 /*
2353 * Initialize the bare minimum state required for HM. This takes care of
2354 * initializing AMD-V if necessary (onlined CPUs, local init etc.)
2355 */
2356 int rc = hmR0EnterCpu(pVCpu);
2357 AssertRC(rc); NOREF(rc);
2358 Assert( (pVCpu->hm.s.fCtxChanged & (HM_CHANGED_HOST_CONTEXT | HM_CHANGED_SVM_HOST_GUEST_SHARED_STATE))
2359 == (HM_CHANGED_HOST_CONTEXT | HM_CHANGED_SVM_HOST_GUEST_SHARED_STATE));
2360
2361 pVCpu->hmr0.s.fLeaveDone = false;
2362
2363 /* Restore longjmp state. */
2364 VMMRZCallRing3Enable(pVCpu);
2365 break;
2366 }
2367
2368 default:
2369 break;
2370 }
2371}
2372
2373
2374/**
2375 * Saves the host state.
2376 *
2377 * @returns VBox status code.
2378 * @param pVCpu The cross context virtual CPU structure.
2379 *
2380 * @remarks No-long-jump zone!!!
2381 */
2382VMMR0DECL(int) SVMR0ExportHostState(PVMCPUCC pVCpu)
2383{
2384 NOREF(pVCpu);
2385
2386 /* Nothing to do here. AMD-V does this for us automatically during the world-switch. */
2387 ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~HM_CHANGED_HOST_CONTEXT);
2388 return VINF_SUCCESS;
2389}
2390
2391
2392/**
2393 * Exports the guest or nested-guest state from the virtual-CPU context into the
2394 * VMCB.
2395 *
2396 * Also sets up the appropriate VMRUN function to execute guest or nested-guest
2397 * code based on the virtual-CPU mode.
2398 *
2399 * @returns VBox status code.
2400 * @param pVCpu The cross context virtual CPU structure.
2401 * @param pSvmTransient Pointer to the SVM-transient structure.
2402 *
2403 * @remarks No-long-jump zone!!!
2404 */
2405static int hmR0SvmExportGuestState(PVMCPUCC pVCpu, PCSVMTRANSIENT pSvmTransient)
2406{
2407 STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExportGuestState, x);
2408
2409 PSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
2410 PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
2411 Assert(pVmcb);
2412
2413 pVmcb->guest.u64RIP = pCtx->rip;
2414 pVmcb->guest.u64RSP = pCtx->rsp;
2415 pVmcb->guest.u64RFlags = pCtx->eflags.u32;
2416 pVmcb->guest.u64RAX = pCtx->rax;
2417
2418 bool const fIsNestedGuest = pSvmTransient->fIsNestedGuest;
2419 RTCCUINTREG const fEFlags = ASMIntDisableFlags();
2420
2421 int rc = hmR0SvmExportGuestControlRegs(pVCpu, pVmcb);
2422 AssertRCReturnStmt(rc, ASMSetFlags(fEFlags), rc);
2423 hmR0SvmExportGuestSegmentRegs(pVCpu, pVmcb);
2424 hmR0SvmExportGuestMsrs(pVCpu, pVmcb);
2425 hmR0SvmExportGuestHwvirtState(pVCpu, pVmcb);
2426
2427 ASMSetFlags(fEFlags);
2428
2429 if (!fIsNestedGuest)
2430 {
2431 /* hmR0SvmExportGuestApicTpr() must be called -after- hmR0SvmExportGuestMsrs() as we
2432 otherwise we would overwrite the LSTAR MSR that we use for TPR patching. */
2433 hmR0SvmExportGuestApicTpr(pVCpu, pVmcb);
2434 hmR0SvmExportGuestXcptIntercepts(pVCpu, pVmcb);
2435 }
2436
2437 /* Clear any bits that may be set but exported unconditionally or unused/reserved bits. */
2438 uint64_t fUnusedMask = HM_CHANGED_GUEST_RIP
2439 | HM_CHANGED_GUEST_RFLAGS
2440 | HM_CHANGED_GUEST_GPRS_MASK
2441 | HM_CHANGED_GUEST_X87
2442 | HM_CHANGED_GUEST_SSE_AVX
2443 | HM_CHANGED_GUEST_OTHER_XSAVE
2444 | HM_CHANGED_GUEST_XCRx
2445 | HM_CHANGED_GUEST_TSC_AUX
2446 | HM_CHANGED_GUEST_OTHER_MSRS;
2447 if (fIsNestedGuest)
2448 fUnusedMask |= HM_CHANGED_SVM_XCPT_INTERCEPTS
2449 | HM_CHANGED_GUEST_APIC_TPR;
2450
2451 ASMAtomicUoAndU64(&pVCpu->hm.s.fCtxChanged, ~( fUnusedMask
2452 | (HM_CHANGED_KEEPER_STATE_MASK & ~HM_CHANGED_SVM_MASK)));
2453
2454#ifdef VBOX_STRICT
2455 /*
2456 * All of the guest-CPU state and SVM keeper bits should be exported here by now,
2457 * except for the host-context and/or shared host-guest context bits.
2458 */
2459 uint64_t const fCtxChanged = ASMAtomicUoReadU64(&pVCpu->hm.s.fCtxChanged);
2460 AssertMsg(!(fCtxChanged & (HM_CHANGED_ALL_GUEST & ~HM_CHANGED_SVM_HOST_GUEST_SHARED_STATE)),
2461 ("fCtxChanged=%#RX64\n", fCtxChanged));
2462
2463 /*
2464 * If we need to log state that isn't always imported, we'll need to import them here.
2465 * See hmR0SvmPostRunGuest() for which part of the state is imported uncondtionally.
2466 */
2467 hmR0SvmLogState(pVCpu, pVmcb, "hmR0SvmExportGuestState", 0 /* fFlags */, 0 /* uVerbose */);
2468#endif
2469
2470 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExportGuestState, x);
2471 return VINF_SUCCESS;
2472}
2473
2474#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
2475
2476/**
2477 * Merges the guest and nested-guest MSR permission bitmap.
2478 *
2479 * If the guest is intercepting an MSR we need to intercept it regardless of
2480 * whether the nested-guest is intercepting it or not.
2481 *
2482 * @param pHostCpu The HM physical-CPU structure.
2483 * @param pVCpu The cross context virtual CPU structure.
2484 *
2485 * @remarks No-long-jmp zone!!!
2486 */
2487DECLINLINE(void) hmR0SvmMergeMsrpmNested(PHMPHYSCPU pHostCpu, PVMCPUCC pVCpu)
2488{
2489 uint64_t const *pu64GstMsrpm = (uint64_t const *)pVCpu->hmr0.s.svm.pvMsrBitmap;
2490 uint64_t const *pu64NstGstMsrpm = (uint64_t const *)&pVCpu->cpum.GstCtx.hwvirt.svm.abMsrBitmap[0];
2491 uint64_t *pu64DstMsrpm = (uint64_t *)pHostCpu->n.svm.pvNstGstMsrpm;
2492
2493 /* MSRPM bytes from offset 0x1800 are reserved, so we stop merging there. */
2494 uint32_t const offRsvdQwords = 0x1800 >> 3;
2495 for (uint32_t i = 0; i < offRsvdQwords; i++)
2496 pu64DstMsrpm[i] = pu64NstGstMsrpm[i] | pu64GstMsrpm[i];
2497}
2498
2499
2500/**
2501 * Caches the nested-guest VMCB fields before we modify them for execution using
2502 * hardware-assisted SVM.
2503 *
2504 * @returns true if the VMCB was previously already cached, false otherwise.
2505 * @param pVCpu The cross context virtual CPU structure.
2506 *
2507 * @sa HMNotifySvmNstGstVmexit.
2508 */
2509static bool hmR0SvmCacheVmcbNested(PVMCPUCC pVCpu)
2510{
2511 /*
2512 * Cache the nested-guest programmed VMCB fields if we have not cached it yet.
2513 * Otherwise we risk re-caching the values we may have modified, see @bugref{7243#c44}.
2514 *
2515 * Nested-paging CR3 is not saved back into the VMCB on #VMEXIT, hence no need to
2516 * cache and restore it, see AMD spec. 15.25.4 "Nested Paging and VMRUN/#VMEXIT".
2517 */
2518 PSVMNESTEDVMCBCACHE pVmcbNstGstCache = &pVCpu->hm.s.svm.NstGstVmcbCache;
2519 bool const fWasCached = pVmcbNstGstCache->fCacheValid;
2520 if (!fWasCached)
2521 {
2522 PCSVMVMCB pVmcbNstGst = &pVCpu->cpum.GstCtx.hwvirt.svm.Vmcb;
2523 PCSVMVMCBCTRL pVmcbNstGstCtrl = &pVmcbNstGst->ctrl;
2524 pVmcbNstGstCache->u16InterceptRdCRx = pVmcbNstGstCtrl->u16InterceptRdCRx;
2525 pVmcbNstGstCache->u16InterceptWrCRx = pVmcbNstGstCtrl->u16InterceptWrCRx;
2526 pVmcbNstGstCache->u16InterceptRdDRx = pVmcbNstGstCtrl->u16InterceptRdDRx;
2527 pVmcbNstGstCache->u16InterceptWrDRx = pVmcbNstGstCtrl->u16InterceptWrDRx;
2528 pVmcbNstGstCache->u16PauseFilterThreshold = pVmcbNstGstCtrl->u16PauseFilterThreshold;
2529 pVmcbNstGstCache->u16PauseFilterCount = pVmcbNstGstCtrl->u16PauseFilterCount;
2530 pVmcbNstGstCache->u32InterceptXcpt = pVmcbNstGstCtrl->u32InterceptXcpt;
2531 pVmcbNstGstCache->u64InterceptCtrl = pVmcbNstGstCtrl->u64InterceptCtrl;
2532 pVmcbNstGstCache->u64TSCOffset = pVmcbNstGstCtrl->u64TSCOffset;
2533 pVmcbNstGstCache->fVIntrMasking = pVmcbNstGstCtrl->IntCtrl.n.u1VIntrMasking;
2534 pVmcbNstGstCache->fNestedPaging = pVmcbNstGstCtrl->NestedPagingCtrl.n.u1NestedPaging;
2535 pVmcbNstGstCache->fLbrVirt = pVmcbNstGstCtrl->LbrVirt.n.u1LbrVirt;
2536 pVmcbNstGstCache->fCacheValid = true;
2537 Log4Func(("Cached VMCB fields\n"));
2538 }
2539
2540 return fWasCached;
2541}
2542
2543
2544/**
2545 * Sets up the nested-guest VMCB for execution using hardware-assisted SVM.
2546 *
2547 * This is done the first time we enter nested-guest execution using SVM R0
2548 * until the nested-guest \#VMEXIT (not to be confused with physical CPU
2549 * \#VMEXITs which may or may not cause a corresponding nested-guest \#VMEXIT).
2550 *
2551 * @param pVCpu The cross context virtual CPU structure.
2552 */
2553static void hmR0SvmSetupVmcbNested(PVMCPUCC pVCpu)
2554{
2555 PSVMVMCB pVmcbNstGst = &pVCpu->cpum.GstCtx.hwvirt.svm.Vmcb;
2556 PSVMVMCBCTRL pVmcbNstGstCtrl = &pVmcbNstGst->ctrl;
2557
2558 HMSVM_ASSERT_IN_NESTED_GUEST(&pVCpu->cpum.GstCtx);
2559
2560 /*
2561 * First cache the nested-guest VMCB fields we may potentially modify.
2562 */
2563 bool const fVmcbCached = hmR0SvmCacheVmcbNested(pVCpu);
2564 if (!fVmcbCached)
2565 {
2566 /*
2567 * The IOPM of the nested-guest can be ignored because the the guest always
2568 * intercepts all IO port accesses. Thus, we'll swap to the guest IOPM rather
2569 * than the nested-guest IOPM and swap the field back on the #VMEXIT.
2570 */
2571 pVmcbNstGstCtrl->u64IOPMPhysAddr = g_HCPhysIOBitmap;
2572
2573 /*
2574 * Use the same nested-paging as the outer guest. We can't dynamically switch off
2575 * nested-paging suddenly while executing a VM (see assertion at the end of
2576 * Trap0eHandler() in PGMAllBth.h).
2577 */
2578 pVmcbNstGstCtrl->NestedPagingCtrl.n.u1NestedPaging = pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging;
2579
2580 /* Always enable V_INTR_MASKING as we do not want to allow access to the physical APIC TPR. */
2581 pVmcbNstGstCtrl->IntCtrl.n.u1VIntrMasking = 1;
2582
2583 /*
2584 * Turn off TPR syncing on #VMEXIT for nested-guests as CR8 intercepts are subject
2585 * to the nested-guest intercepts and we always run with V_INTR_MASKING.
2586 */
2587 pVCpu->hmr0.s.svm.fSyncVTpr = false;
2588
2589# ifdef DEBUG_ramshankar
2590 /* For debugging purposes - copy the LBR info. from outer guest VMCB. */
2591 pVmcbNstGstCtrl->LbrVirt.n.u1LbrVirt = pVmcb->ctrl.LbrVirt.n.u1LbrVirt;
2592# endif
2593
2594 /*
2595 * If we don't expose Virtualized-VMSAVE/VMLOAD feature to the outer guest, we
2596 * need to intercept VMSAVE/VMLOAD instructions executed by the nested-guest.
2597 */
2598 if (!pVCpu->CTX_SUFF(pVM)->cpum.ro.GuestFeatures.fSvmVirtVmsaveVmload)
2599 pVmcbNstGstCtrl->u64InterceptCtrl |= SVM_CTRL_INTERCEPT_VMSAVE
2600 | SVM_CTRL_INTERCEPT_VMLOAD;
2601
2602 /*
2603 * If we don't expose Virtual GIF feature to the outer guest, we need to intercept
2604 * CLGI/STGI instructions executed by the nested-guest.
2605 */
2606 if (!pVCpu->CTX_SUFF(pVM)->cpum.ro.GuestFeatures.fSvmVGif)
2607 pVmcbNstGstCtrl->u64InterceptCtrl |= SVM_CTRL_INTERCEPT_CLGI
2608 | SVM_CTRL_INTERCEPT_STGI;
2609
2610 /* Merge the guest and nested-guest intercepts. */
2611 hmR0SvmMergeVmcbCtrlsNested(pVCpu);
2612
2613 /* Update the VMCB clean bits. */
2614 pVmcbNstGstCtrl->u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_INTERCEPTS;
2615 }
2616 else
2617 {
2618 Assert(!pVCpu->hmr0.s.svm.fSyncVTpr);
2619 Assert(pVmcbNstGstCtrl->u64IOPMPhysAddr == g_HCPhysIOBitmap);
2620 Assert(RT_BOOL(pVmcbNstGstCtrl->NestedPagingCtrl.n.u1NestedPaging) == pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging);
2621 Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPagingCfg == pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging);
2622 }
2623}
2624
2625#endif /* VBOX_WITH_NESTED_HWVIRT_SVM */
2626
2627/**
2628 * Exports the state shared between the host and guest (or nested-guest) into
2629 * the VMCB.
2630 *
2631 * @param pVCpu The cross context virtual CPU structure.
2632 * @param pVmcb Pointer to the VM control block.
2633 *
2634 * @remarks No-long-jump zone!!!
2635 */
2636static void hmR0SvmExportSharedState(PVMCPUCC pVCpu, PSVMVMCB pVmcb)
2637{
2638 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2639 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
2640
2641 if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_GUEST_DR_MASK)
2642 hmR0SvmExportSharedDebugState(pVCpu, pVmcb);
2643
2644 pVCpu->hm.s.fCtxChanged &= ~HM_CHANGED_GUEST_DR_MASK;
2645 AssertMsg(!(pVCpu->hm.s.fCtxChanged & HM_CHANGED_SVM_HOST_GUEST_SHARED_STATE),
2646 ("fCtxChanged=%#RX64\n", pVCpu->hm.s.fCtxChanged));
2647}
2648
2649
2650/**
2651 * Worker for SVMR0ImportStateOnDemand.
2652 *
2653 * @param pVCpu The cross context virtual CPU structure.
2654 * @param fWhat What to import, CPUMCTX_EXTRN_XXX.
2655 */
2656static void hmR0SvmImportGuestState(PVMCPUCC pVCpu, uint64_t fWhat)
2657{
2658 STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatImportGuestState, x);
2659
2660 PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
2661 PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
2662 PCSVMVMCBSTATESAVE pVmcbGuest = &pVmcb->guest;
2663 PCSVMVMCBCTRL pVmcbCtrl = &pVmcb->ctrl;
2664
2665 /*
2666 * We disable interrupts to make the updating of the state and in particular
2667 * the fExtrn modification atomic wrt to preemption hooks.
2668 */
2669 RTCCUINTREG const fEFlags = ASMIntDisableFlags();
2670
2671 fWhat &= pCtx->fExtrn;
2672 if (fWhat)
2673 {
2674#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
2675 if (fWhat & CPUMCTX_EXTRN_HWVIRT)
2676 {
2677 if (pVmcbCtrl->IntCtrl.n.u1VGifEnable)
2678 {
2679 Assert(!CPUMIsGuestInSvmNestedHwVirtMode(pCtx)); /* We don't yet support passing VGIF feature to the guest. */
2680 Assert(HMIsSvmVGifActive(pVCpu->CTX_SUFF(pVM))); /* VM has configured it. */
2681 CPUMSetGuestGif(pCtx, pVmcbCtrl->IntCtrl.n.u1VGif);
2682 }
2683 }
2684
2685 if (fWhat & CPUMCTX_EXTRN_HM_SVM_HWVIRT_VIRQ)
2686 {
2687 if ( !pVmcbCtrl->IntCtrl.n.u1VIrqPending
2688 && VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INTERRUPT_NESTED_GUEST))
2689 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INTERRUPT_NESTED_GUEST);
2690 }
2691#endif
2692
2693 if (fWhat & CPUMCTX_EXTRN_INHIBIT_INT)
2694 CPUMUpdateInterruptShadowEx(pCtx, pVmcbCtrl->IntShadow.n.u1IntShadow, pVmcbGuest->u64RIP);
2695
2696 if (fWhat & CPUMCTX_EXTRN_RIP)
2697 pCtx->rip = pVmcbGuest->u64RIP;
2698
2699 if (fWhat & CPUMCTX_EXTRN_RFLAGS)
2700 {
2701 pCtx->eflags.u32 = pVmcbGuest->u64RFlags;
2702 if (pVCpu->hmr0.s.fClearTrapFlag)
2703 {
2704 pVCpu->hmr0.s.fClearTrapFlag = false;
2705 pCtx->eflags.Bits.u1TF = 0;
2706 }
2707 }
2708
2709 if (fWhat & CPUMCTX_EXTRN_RSP)
2710 pCtx->rsp = pVmcbGuest->u64RSP;
2711
2712 if (fWhat & CPUMCTX_EXTRN_RAX)
2713 pCtx->rax = pVmcbGuest->u64RAX;
2714
2715 if (fWhat & CPUMCTX_EXTRN_SREG_MASK)
2716 {
2717 if (fWhat & CPUMCTX_EXTRN_CS)
2718 {
2719 HMSVM_SEG_REG_COPY_FROM_VMCB(pCtx, pVmcbGuest, CS, cs);
2720 /* Correct the CS granularity bit. Haven't seen it being wrong in any other register (yet). */
2721 /** @todo SELM might need to be fixed as it too should not care about the
2722 * granularity bit. See @bugref{6785}. */
2723 if ( !pCtx->cs.Attr.n.u1Granularity
2724 && pCtx->cs.Attr.n.u1Present
2725 && pCtx->cs.u32Limit > UINT32_C(0xfffff))
2726 {
2727 Assert((pCtx->cs.u32Limit & 0xfff) == 0xfff);
2728 pCtx->cs.Attr.n.u1Granularity = 1;
2729 }
2730 HMSVM_ASSERT_SEG_GRANULARITY(pCtx, cs);
2731 }
2732 if (fWhat & CPUMCTX_EXTRN_SS)
2733 {
2734 HMSVM_SEG_REG_COPY_FROM_VMCB(pCtx, pVmcbGuest, SS, ss);
2735 HMSVM_ASSERT_SEG_GRANULARITY(pCtx, ss);
2736 /*
2737 * Sync the hidden SS DPL field. AMD CPUs have a separate CPL field in the
2738 * VMCB and uses that and thus it's possible that when the CPL changes during
2739 * guest execution that the SS DPL isn't updated by AMD-V. Observed on some
2740 * AMD Fusion CPUs with 64-bit guests.
2741 *
2742 * See AMD spec. 15.5.1 "Basic operation".
2743 */
2744 Assert(!(pVmcbGuest->u8CPL & ~0x3));
2745 uint8_t const uCpl = pVmcbGuest->u8CPL;
2746 if (pCtx->ss.Attr.n.u2Dpl != uCpl)
2747 pCtx->ss.Attr.n.u2Dpl = uCpl & 0x3;
2748 }
2749 if (fWhat & CPUMCTX_EXTRN_DS)
2750 {
2751 HMSVM_SEG_REG_COPY_FROM_VMCB(pCtx, pVmcbGuest, DS, ds);
2752 HMSVM_ASSERT_SEG_GRANULARITY(pCtx, ds);
2753 }
2754 if (fWhat & CPUMCTX_EXTRN_ES)
2755 {
2756 HMSVM_SEG_REG_COPY_FROM_VMCB(pCtx, pVmcbGuest, ES, es);
2757 HMSVM_ASSERT_SEG_GRANULARITY(pCtx, es);
2758 }
2759 if (fWhat & CPUMCTX_EXTRN_FS)
2760 {
2761 HMSVM_SEG_REG_COPY_FROM_VMCB(pCtx, pVmcbGuest, FS, fs);
2762 HMSVM_ASSERT_SEG_GRANULARITY(pCtx, fs);
2763 }
2764 if (fWhat & CPUMCTX_EXTRN_GS)
2765 {
2766 HMSVM_SEG_REG_COPY_FROM_VMCB(pCtx, pVmcbGuest, GS, gs);
2767 HMSVM_ASSERT_SEG_GRANULARITY(pCtx, gs);
2768 }
2769 }
2770
2771 if (fWhat & CPUMCTX_EXTRN_TABLE_MASK)
2772 {
2773 if (fWhat & CPUMCTX_EXTRN_TR)
2774 {
2775 /*
2776 * Fixup TR attributes so it's compatible with Intel. Important when saved-states
2777 * are used between Intel and AMD, see @bugref{6208#c39}.
2778 * ASSUME that it's normally correct and that we're in 32-bit or 64-bit mode.
2779 */
2780 HMSVM_SEG_REG_COPY_FROM_VMCB(pCtx, pVmcbGuest, TR, tr);
2781 if (pCtx->tr.Attr.n.u4Type != X86_SEL_TYPE_SYS_386_TSS_BUSY)
2782 {
2783 if ( pCtx->tr.Attr.n.u4Type == X86_SEL_TYPE_SYS_386_TSS_AVAIL
2784 || CPUMIsGuestInLongModeEx(pCtx))
2785 pCtx->tr.Attr.n.u4Type = X86_SEL_TYPE_SYS_386_TSS_BUSY;
2786 else if (pCtx->tr.Attr.n.u4Type == X86_SEL_TYPE_SYS_286_TSS_AVAIL)
2787 pCtx->tr.Attr.n.u4Type = X86_SEL_TYPE_SYS_286_TSS_BUSY;
2788 }
2789 }
2790
2791 if (fWhat & CPUMCTX_EXTRN_LDTR)
2792 HMSVM_SEG_REG_COPY_FROM_VMCB(pCtx, pVmcbGuest, LDTR, ldtr);
2793
2794 if (fWhat & CPUMCTX_EXTRN_GDTR)
2795 {
2796 pCtx->gdtr.cbGdt = pVmcbGuest->GDTR.u32Limit;
2797 pCtx->gdtr.pGdt = pVmcbGuest->GDTR.u64Base;
2798 }
2799
2800 if (fWhat & CPUMCTX_EXTRN_IDTR)
2801 {
2802 pCtx->idtr.cbIdt = pVmcbGuest->IDTR.u32Limit;
2803 pCtx->idtr.pIdt = pVmcbGuest->IDTR.u64Base;
2804 }
2805 }
2806
2807 if (fWhat & CPUMCTX_EXTRN_SYSCALL_MSRS)
2808 {
2809 pCtx->msrSTAR = pVmcbGuest->u64STAR;
2810 pCtx->msrLSTAR = pVmcbGuest->u64LSTAR;
2811 pCtx->msrCSTAR = pVmcbGuest->u64CSTAR;
2812 pCtx->msrSFMASK = pVmcbGuest->u64SFMASK;
2813 }
2814
2815 if ( (fWhat & CPUMCTX_EXTRN_SYSENTER_MSRS)
2816 && !pVCpu->hm.s.svm.fEmulateLongModeSysEnterExit /* Intercepted. AMD-V would clear the high 32 bits of EIP & ESP. */)
2817 {
2818 pCtx->SysEnter.cs = pVmcbGuest->u64SysEnterCS;
2819 pCtx->SysEnter.eip = pVmcbGuest->u64SysEnterEIP;
2820 pCtx->SysEnter.esp = pVmcbGuest->u64SysEnterESP;
2821 }
2822
2823 if (fWhat & CPUMCTX_EXTRN_KERNEL_GS_BASE)
2824 pCtx->msrKERNELGSBASE = pVmcbGuest->u64KernelGSBase;
2825
2826 if (fWhat & CPUMCTX_EXTRN_DR_MASK)
2827 {
2828 if (fWhat & CPUMCTX_EXTRN_DR6)
2829 {
2830 if (!pVCpu->hmr0.s.fUsingHyperDR7)
2831 pCtx->dr[6] = pVmcbGuest->u64DR6;
2832 else
2833 CPUMSetHyperDR6(pVCpu, pVmcbGuest->u64DR6);
2834 }
2835
2836 if (fWhat & CPUMCTX_EXTRN_DR7)
2837 {
2838 if (!pVCpu->hmr0.s.fUsingHyperDR7)
2839 pCtx->dr[7] = pVmcbGuest->u64DR7;
2840 else
2841 Assert(pVmcbGuest->u64DR7 == CPUMGetHyperDR7(pVCpu));
2842 }
2843 }
2844
2845 if (fWhat & CPUMCTX_EXTRN_CR_MASK)
2846 {
2847 if (fWhat & CPUMCTX_EXTRN_CR0)
2848 {
2849 /* We intercept changes to all CR0 bits except maybe TS & MP bits. */
2850 uint64_t const uCr0 = (pCtx->cr0 & ~(X86_CR0_TS | X86_CR0_MP))
2851 | (pVmcbGuest->u64CR0 & (X86_CR0_TS | X86_CR0_MP));
2852 VMMRZCallRing3Disable(pVCpu); /* Calls into PGM which has Log statements. */
2853 CPUMSetGuestCR0(pVCpu, uCr0);
2854 VMMRZCallRing3Enable(pVCpu);
2855 }
2856
2857 if (fWhat & CPUMCTX_EXTRN_CR2)
2858 pCtx->cr2 = pVmcbGuest->u64CR2;
2859
2860 if (fWhat & CPUMCTX_EXTRN_CR3)
2861 {
2862 if ( pVmcbCtrl->NestedPagingCtrl.n.u1NestedPaging
2863 && pCtx->cr3 != pVmcbGuest->u64CR3)
2864 {
2865 CPUMSetGuestCR3(pVCpu, pVmcbGuest->u64CR3);
2866 VMCPU_FF_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3);
2867 }
2868 }
2869
2870 /* Changes to CR4 are always intercepted. */
2871 }
2872
2873 /* Update fExtrn. */
2874 pCtx->fExtrn &= ~fWhat;
2875
2876 /* If everything has been imported, clear the HM keeper bit. */
2877 if (!(pCtx->fExtrn & HMSVM_CPUMCTX_EXTRN_ALL))
2878 {
2879 pCtx->fExtrn &= ~CPUMCTX_EXTRN_KEEPER_HM;
2880 Assert(!pCtx->fExtrn);
2881 }
2882 }
2883 else
2884 Assert(!pCtx->fExtrn || (pCtx->fExtrn & HMSVM_CPUMCTX_EXTRN_ALL));
2885
2886 ASMSetFlags(fEFlags);
2887
2888 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatImportGuestState, x);
2889
2890 /*
2891 * Honor any pending CR3 updates.
2892 *
2893 * Consider this scenario: #VMEXIT -> VMMRZCallRing3Enable() -> do stuff that causes a longjmp
2894 * -> SVMR0CallRing3Callback() -> VMMRZCallRing3Disable() -> hmR0SvmImportGuestState()
2895 * -> Sets VMCPU_FF_HM_UPDATE_CR3 pending -> return from the longjmp -> continue with #VMEXIT
2896 * handling -> hmR0SvmImportGuestState() and here we are.
2897 *
2898 * The reason for such complicated handling is because VM-exits that call into PGM expect
2899 * CR3 to be up-to-date and thus any CR3-saves -before- the VM-exit (longjmp) would've
2900 * postponed the CR3 update via the force-flag and cleared CR3 from fExtrn. Any SVM R0
2901 * VM-exit handler that requests CR3 to be saved will end up here and we call PGMUpdateCR3().
2902 *
2903 * The longjmp exit path can't check these CR3 force-flags and call code that takes a lock again,
2904 * and does not process force-flag like regular exits to ring-3 either, we cover for it here.
2905 */
2906 if ( VMMRZCallRing3IsEnabled(pVCpu)
2907 && VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3))
2908 {
2909 AssertMsg(pCtx->cr3 == pVmcbGuest->u64CR3, ("cr3=%#RX64 vmcb_cr3=%#RX64\n", pCtx->cr3, pVmcbGuest->u64CR3));
2910 PGMUpdateCR3(pVCpu, pCtx->cr3);
2911 }
2912}
2913
2914
2915/**
2916 * Saves the guest (or nested-guest) state from the VMCB into the guest-CPU
2917 * context.
2918 *
2919 * Currently there is no residual state left in the CPU that is not updated in the
2920 * VMCB.
2921 *
2922 * @returns VBox status code.
2923 * @param pVCpu The cross context virtual CPU structure.
2924 * @param fWhat What to import, CPUMCTX_EXTRN_XXX.
2925 */
2926VMMR0DECL(int) SVMR0ImportStateOnDemand(PVMCPUCC pVCpu, uint64_t fWhat)
2927{
2928 hmR0SvmImportGuestState(pVCpu, fWhat);
2929 return VINF_SUCCESS;
2930}
2931
2932
2933/**
2934 * Gets SVM \#VMEXIT auxiliary information.
2935 *
2936 * @returns VBox status code.
2937 * @param pVCpu The cross context virtual CPU structure.
2938 * @param pSvmExitAux Where to store the auxiliary info.
2939 */
2940VMMR0DECL(int) SVMR0GetExitAuxInfo(PVMCPUCC pVCpu, PSVMEXITAUX pSvmExitAux)
2941{
2942 PCSVMTRANSIENT pSvmTransient = pVCpu->hmr0.s.svm.pSvmTransient;
2943 if (RT_LIKELY(pSvmTransient))
2944 {
2945 PCSVMVMCB pVmcb = pSvmTransient->pVmcb;
2946 if (RT_LIKELY(pVmcb))
2947 {
2948 pSvmExitAux->u64ExitCode = pVmcb->ctrl.u64ExitCode;
2949 pSvmExitAux->u64ExitInfo1 = pVmcb->ctrl.u64ExitInfo1;
2950 pSvmExitAux->u64ExitInfo2 = pVmcb->ctrl.u64ExitInfo2;
2951 pSvmExitAux->ExitIntInfo = pVmcb->ctrl.ExitIntInfo;
2952 return VINF_SUCCESS;
2953 }
2954 return VERR_SVM_IPE_5;
2955 }
2956 return VERR_NOT_AVAILABLE;
2957}
2958
2959
2960/**
2961 * Does the necessary state syncing before returning to ring-3 for any reason
2962 * (longjmp, preemption, voluntary exits to ring-3) from AMD-V.
2963 *
2964 * @param pVCpu The cross context virtual CPU structure.
2965 * @param fImportState Whether to import the guest state from the VMCB back
2966 * to the guest-CPU context.
2967 *
2968 * @remarks No-long-jmp zone!!!
2969 */
2970static void hmR0SvmLeave(PVMCPUCC pVCpu, bool fImportState)
2971{
2972 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2973 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
2974
2975 /*
2976 * !!! IMPORTANT !!!
2977 * If you modify code here, make sure to check whether SVMR0CallRing3Callback() needs to be updated too.
2978 */
2979
2980 /* Save the guest state if necessary. */
2981 if (fImportState)
2982 hmR0SvmImportGuestState(pVCpu, HMSVM_CPUMCTX_EXTRN_ALL);
2983
2984 /* Restore host FPU state if necessary and resync on next R0 reentry. */
2985 CPUMR0FpuStateMaybeSaveGuestAndRestoreHost(pVCpu);
2986 Assert(!CPUMIsGuestFPUStateActive(pVCpu));
2987
2988 /*
2989 * Restore host debug registers if necessary and resync on next R0 reentry.
2990 */
2991#ifdef VBOX_STRICT
2992 if (CPUMIsHyperDebugStateActive(pVCpu))
2993 {
2994 PSVMVMCB pVmcb = pVCpu->hmr0.s.svm.pVmcb; /** @todo nested-guest. */
2995 Assert(pVmcb->ctrl.u16InterceptRdDRx == 0xffff);
2996 Assert(pVmcb->ctrl.u16InterceptWrDRx == 0xffff);
2997 }
2998#endif
2999 CPUMR0DebugStateMaybeSaveGuestAndRestoreHost(pVCpu, false /* save DR6 */);
3000 Assert(!CPUMIsHyperDebugStateActive(pVCpu));
3001 Assert(!CPUMIsGuestDebugStateActive(pVCpu));
3002
3003 STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatEntry);
3004 STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatImportGuestState);
3005 STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExportGuestState);
3006 STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatPreExit);
3007 STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitHandling);
3008 STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitVmentry);
3009 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchLongJmpToR3);
3010
3011 VMCPU_CMPXCHG_STATE(pVCpu, VMCPUSTATE_STARTED_HM, VMCPUSTATE_STARTED_EXEC);
3012}
3013
3014
3015/**
3016 * Leaves the AMD-V session.
3017 *
3018 * Only used while returning to ring-3 either due to longjump or exits to
3019 * ring-3.
3020 *
3021 * @returns VBox status code.
3022 * @param pVCpu The cross context virtual CPU structure.
3023 */
3024static int hmR0SvmLeaveSession(PVMCPUCC pVCpu)
3025{
3026 HM_DISABLE_PREEMPT(pVCpu);
3027 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
3028 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
3029
3030 /* When thread-context hooks are used, we can avoid doing the leave again if we had been preempted before
3031 and done this from the SVMR0ThreadCtxCallback(). */
3032 if (!pVCpu->hmr0.s.fLeaveDone)
3033 {
3034 hmR0SvmLeave(pVCpu, true /* fImportState */);
3035 pVCpu->hmr0.s.fLeaveDone = true;
3036 }
3037
3038 /*
3039 * !!! IMPORTANT !!!
3040 * If you modify code here, make sure to check whether SVMR0CallRing3Callback() needs to be updated too.
3041 */
3042
3043 /** @todo eliminate the need for calling VMMR0ThreadCtxHookDisable here! */
3044 /* Deregister hook now that we've left HM context before re-enabling preemption. */
3045 VMMR0ThreadCtxHookDisable(pVCpu);
3046
3047 /* Leave HM context. This takes care of local init (term). */
3048 int rc = HMR0LeaveCpu(pVCpu);
3049
3050 HM_RESTORE_PREEMPT();
3051 return rc;
3052}
3053
3054
3055/**
3056 * VMMRZCallRing3() callback wrapper which saves the guest state (or restores
3057 * any remaining host state) before we go back to ring-3 due to an assertion.
3058 *
3059 * @param pVCpu The cross context virtual CPU structure.
3060 */
3061VMMR0DECL(int) SVMR0AssertionCallback(PVMCPUCC pVCpu)
3062{
3063 /*
3064 * !!! IMPORTANT !!!
3065 * If you modify code here, make sure to check whether hmR0SvmLeave() and hmR0SvmLeaveSession() needs
3066 * to be updated too. This is a stripped down version which gets out ASAP trying to not trigger any assertion.
3067 */
3068 VMMR0AssertionRemoveNotification(pVCpu);
3069 VMMRZCallRing3Disable(pVCpu);
3070 HM_DISABLE_PREEMPT(pVCpu);
3071
3072 /* Import the entire guest state. */
3073 hmR0SvmImportGuestState(pVCpu, HMSVM_CPUMCTX_EXTRN_ALL);
3074
3075 /* Restore host FPU state if necessary and resync on next R0 reentry. */
3076 CPUMR0FpuStateMaybeSaveGuestAndRestoreHost(pVCpu);
3077
3078 /* Restore host debug registers if necessary and resync on next R0 reentry. */
3079 CPUMR0DebugStateMaybeSaveGuestAndRestoreHost(pVCpu, false /* save DR6 */);
3080
3081 /* Deregister the hook now that we've left HM context before re-enabling preemption. */
3082 /** @todo eliminate the need for calling VMMR0ThreadCtxHookDisable here! */
3083 VMMR0ThreadCtxHookDisable(pVCpu);
3084
3085 /* Leave HM context. This takes care of local init (term). */
3086 HMR0LeaveCpu(pVCpu);
3087
3088 HM_RESTORE_PREEMPT();
3089 return VINF_SUCCESS;
3090}
3091
3092
3093/**
3094 * Take necessary actions before going back to ring-3.
3095 *
3096 * An action requires us to go back to ring-3. This function does the necessary
3097 * steps before we can safely return to ring-3. This is not the same as longjmps
3098 * to ring-3, this is voluntary.
3099 *
3100 * @returns Strict VBox status code.
3101 * @param pVCpu The cross context virtual CPU structure.
3102 * @param rcExit The reason for exiting to ring-3. Can be
3103 * VINF_VMM_UNKNOWN_RING3_CALL.
3104 */
3105static VBOXSTRICTRC hmR0SvmExitToRing3(PVMCPUCC pVCpu, VBOXSTRICTRC rcExit)
3106{
3107 Assert(pVCpu);
3108 HMSVM_ASSERT_PREEMPT_SAFE(pVCpu);
3109
3110 /* Please, no longjumps here (any logging shouldn't flush jump back to ring-3). NO LOGGING BEFORE THIS POINT! */
3111 VMMRZCallRing3Disable(pVCpu);
3112 Log4Func(("rcExit=%d LocalFF=%#RX64 GlobalFF=%#RX32\n", VBOXSTRICTRC_VAL(rcExit), (uint64_t)pVCpu->fLocalForcedActions,
3113 pVCpu->CTX_SUFF(pVM)->fGlobalForcedActions));
3114
3115 /* We need to do this only while truly exiting the "inner loop" back to ring-3 and -not- for any longjmp to ring3. */
3116 if (pVCpu->hm.s.Event.fPending)
3117 {
3118 hmR0SvmPendingEventToTrpmTrap(pVCpu);
3119 Assert(!pVCpu->hm.s.Event.fPending);
3120 }
3121
3122 /* Sync. the necessary state for going back to ring-3. */
3123 hmR0SvmLeaveSession(pVCpu);
3124 STAM_COUNTER_DEC(&pVCpu->hm.s.StatSwitchLongJmpToR3);
3125
3126 /* Thread-context hooks are unregistered at this point!!! */
3127 /* Ring-3 callback notifications are unregistered at this point!!! */
3128
3129 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TO_R3);
3130 CPUMSetChangedFlags(pVCpu, CPUM_CHANGED_SYSENTER_MSR
3131 | CPUM_CHANGED_LDTR
3132 | CPUM_CHANGED_GDTR
3133 | CPUM_CHANGED_IDTR
3134 | CPUM_CHANGED_TR
3135 | CPUM_CHANGED_HIDDEN_SEL_REGS);
3136 if ( pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging
3137 && CPUMIsGuestPagingEnabledEx(&pVCpu->cpum.GstCtx))
3138 {
3139 CPUMSetChangedFlags(pVCpu, CPUM_CHANGED_GLOBAL_TLB_FLUSH);
3140 }
3141
3142 /* Update the exit-to-ring 3 reason. */
3143 pVCpu->hm.s.rcLastExitToR3 = VBOXSTRICTRC_VAL(rcExit);
3144
3145 /* On our way back from ring-3, reload the guest-CPU state if it may change while in ring-3. */
3146 if ( rcExit != VINF_EM_RAW_INTERRUPT
3147 || CPUMIsGuestInSvmNestedHwVirtMode(&pVCpu->cpum.GstCtx))
3148 {
3149 Assert(!(pVCpu->cpum.GstCtx.fExtrn & HMSVM_CPUMCTX_EXTRN_ALL));
3150 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
3151 }
3152
3153 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchExitToR3);
3154 VMMRZCallRing3Enable(pVCpu);
3155
3156 /*
3157 * If we're emulating an instruction, we shouldn't have any TRPM traps pending
3158 * and if we're injecting an event we should have a TRPM trap pending.
3159 */
3160 AssertReturnStmt(rcExit != VINF_EM_RAW_INJECT_TRPM_EVENT || TRPMHasTrap(pVCpu),
3161 pVCpu->hm.s.u32HMError = VBOXSTRICTRC_VAL(rcExit),
3162 VERR_SVM_IPE_5);
3163 AssertReturnStmt(rcExit != VINF_EM_RAW_EMULATE_INSTR || !TRPMHasTrap(pVCpu),
3164 pVCpu->hm.s.u32HMError = VBOXSTRICTRC_VAL(rcExit),
3165 VERR_SVM_IPE_4);
3166
3167 return rcExit;
3168}
3169
3170
3171/**
3172 * Updates the use of TSC offsetting mode for the CPU and adjusts the necessary
3173 * intercepts.
3174 *
3175 * @param pVCpu The cross context virtual CPU structure.
3176 * @param pVmcb Pointer to the VM control block.
3177 *
3178 * @remarks No-long-jump zone!!!
3179 */
3180static void hmR0SvmUpdateTscOffsetting(PVMCPUCC pVCpu, PSVMVMCB pVmcb)
3181{
3182 /*
3183 * Avoid intercepting RDTSC/RDTSCP if we determined the host TSC (++) is stable
3184 * and in case of a nested-guest, if the nested-VMCB specifies it is not intercepting
3185 * RDTSC/RDTSCP as well.
3186 */
3187 bool fParavirtTsc;
3188 uint64_t uTscOffset;
3189 bool const fCanUseRealTsc = TMCpuTickCanUseRealTSC(pVCpu->CTX_SUFF(pVM), pVCpu, &uTscOffset, &fParavirtTsc);
3190
3191 bool fIntercept;
3192 if (fCanUseRealTsc)
3193 fIntercept = hmR0SvmClearCtrlIntercept(pVCpu, pVmcb, SVM_CTRL_INTERCEPT_RDTSC | SVM_CTRL_INTERCEPT_RDTSCP);
3194 else
3195 {
3196 hmR0SvmSetCtrlIntercept(pVmcb, SVM_CTRL_INTERCEPT_RDTSC | SVM_CTRL_INTERCEPT_RDTSCP);
3197 fIntercept = true;
3198 }
3199
3200 if (!fIntercept)
3201 {
3202#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
3203 /* Apply the nested-guest VMCB's TSC offset over the guest TSC offset. */
3204 if (CPUMIsGuestInSvmNestedHwVirtMode(&pVCpu->cpum.GstCtx))
3205 uTscOffset = CPUMApplyNestedGuestTscOffset(pVCpu, uTscOffset);
3206#endif
3207
3208 /* Update the TSC offset in the VMCB and the relevant clean bits. */
3209 pVmcb->ctrl.u64TSCOffset = uTscOffset;
3210 pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_INTERCEPTS;
3211 }
3212
3213 /* Currently neither Hyper-V nor KVM need to update their paravirt. TSC
3214 information before every VM-entry, hence we have nothing to do here at the moment. */
3215 if (fParavirtTsc)
3216 STAM_COUNTER_INC(&pVCpu->hm.s.StatTscParavirt);
3217}
3218
3219
3220/**
3221 * Sets an event as a pending event to be injected into the guest.
3222 *
3223 * @param pVCpu The cross context virtual CPU structure.
3224 * @param pEvent Pointer to the SVM event.
3225 * @param GCPtrFaultAddress The fault-address (CR2) in case it's a
3226 * page-fault.
3227 *
3228 * @remarks Statistics counter assumes this is a guest event being reflected to
3229 * the guest i.e. 'StatInjectPendingReflect' is incremented always.
3230 */
3231DECLINLINE(void) hmR0SvmSetPendingEvent(PVMCPUCC pVCpu, PSVMEVENT pEvent, RTGCUINTPTR GCPtrFaultAddress)
3232{
3233 Assert(!pVCpu->hm.s.Event.fPending);
3234 Assert(pEvent->n.u1Valid);
3235
3236 pVCpu->hm.s.Event.u64IntInfo = pEvent->u;
3237 pVCpu->hm.s.Event.fPending = true;
3238 pVCpu->hm.s.Event.GCPtrFaultAddress = GCPtrFaultAddress;
3239
3240 Log4Func(("u=%#RX64 u8Vector=%#x Type=%#x ErrorCodeValid=%RTbool ErrorCode=%#RX32\n", pEvent->u, pEvent->n.u8Vector,
3241 (uint8_t)pEvent->n.u3Type, !!pEvent->n.u1ErrorCodeValid, pEvent->n.u32ErrorCode));
3242}
3243
3244
3245/**
3246 * Sets an divide error (\#DE) exception as pending-for-injection into the VM.
3247 *
3248 * @param pVCpu The cross context virtual CPU structure.
3249 */
3250DECLINLINE(void) hmR0SvmSetPendingXcptDE(PVMCPUCC pVCpu)
3251{
3252 SVMEVENT Event;
3253 Event.u = 0;
3254 Event.n.u1Valid = 1;
3255 Event.n.u3Type = SVM_EVENT_EXCEPTION;
3256 Event.n.u8Vector = X86_XCPT_DE;
3257 hmR0SvmSetPendingEvent(pVCpu, &Event, 0 /* GCPtrFaultAddress */);
3258}
3259
3260
3261/**
3262 * Sets an invalid-opcode (\#UD) exception as pending-for-injection into the VM.
3263 *
3264 * @param pVCpu The cross context virtual CPU structure.
3265 */
3266DECLINLINE(void) hmR0SvmSetPendingXcptUD(PVMCPUCC pVCpu)
3267{
3268 SVMEVENT Event;
3269 Event.u = 0;
3270 Event.n.u1Valid = 1;
3271 Event.n.u3Type = SVM_EVENT_EXCEPTION;
3272 Event.n.u8Vector = X86_XCPT_UD;
3273 hmR0SvmSetPendingEvent(pVCpu, &Event, 0 /* GCPtrFaultAddress */);
3274}
3275
3276
3277/**
3278 * Sets a debug (\#DB) exception as pending-for-injection into the VM.
3279 *
3280 * @param pVCpu The cross context virtual CPU structure.
3281 */
3282DECLINLINE(void) hmR0SvmSetPendingXcptDB(PVMCPUCC pVCpu)
3283{
3284 SVMEVENT Event;
3285 Event.u = 0;
3286 Event.n.u1Valid = 1;
3287 Event.n.u3Type = SVM_EVENT_EXCEPTION;
3288 Event.n.u8Vector = X86_XCPT_DB;
3289 hmR0SvmSetPendingEvent(pVCpu, &Event, 0 /* GCPtrFaultAddress */);
3290}
3291
3292
3293/**
3294 * Sets a page fault (\#PF) exception as pending-for-injection into the VM.
3295 *
3296 * @param pVCpu The cross context virtual CPU structure.
3297 * @param u32ErrCode The error-code for the page-fault.
3298 * @param uFaultAddress The page fault address (CR2).
3299 *
3300 * @remarks This updates the guest CR2 with @a uFaultAddress!
3301 */
3302DECLINLINE(void) hmR0SvmSetPendingXcptPF(PVMCPUCC pVCpu, uint32_t u32ErrCode, RTGCUINTPTR uFaultAddress)
3303{
3304 SVMEVENT Event;
3305 Event.u = 0;
3306 Event.n.u1Valid = 1;
3307 Event.n.u3Type = SVM_EVENT_EXCEPTION;
3308 Event.n.u8Vector = X86_XCPT_PF;
3309 Event.n.u1ErrorCodeValid = 1;
3310 Event.n.u32ErrorCode = u32ErrCode;
3311
3312 /* Update CR2 of the guest. */
3313 HMSVM_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CR2);
3314 if (pVCpu->cpum.GstCtx.cr2 != uFaultAddress)
3315 {
3316 pVCpu->cpum.GstCtx.cr2 = uFaultAddress;
3317 /* The VMCB clean bit for CR2 will be updated while re-loading the guest state. */
3318 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_CR2);
3319 }
3320
3321 hmR0SvmSetPendingEvent(pVCpu, &Event, uFaultAddress);
3322}
3323
3324
3325/**
3326 * Sets a math-fault (\#MF) exception as pending-for-injection into the VM.
3327 *
3328 * @param pVCpu The cross context virtual CPU structure.
3329 */
3330DECLINLINE(void) hmR0SvmSetPendingXcptMF(PVMCPUCC pVCpu)
3331{
3332 SVMEVENT Event;
3333 Event.u = 0;
3334 Event.n.u1Valid = 1;
3335 Event.n.u3Type = SVM_EVENT_EXCEPTION;
3336 Event.n.u8Vector = X86_XCPT_MF;
3337 hmR0SvmSetPendingEvent(pVCpu, &Event, 0 /* GCPtrFaultAddress */);
3338}
3339
3340
3341/**
3342 * Sets a double fault (\#DF) exception as pending-for-injection into the VM.
3343 *
3344 * @param pVCpu The cross context virtual CPU structure.
3345 */
3346DECLINLINE(void) hmR0SvmSetPendingXcptDF(PVMCPUCC pVCpu)
3347{
3348 SVMEVENT Event;
3349 Event.u = 0;
3350 Event.n.u1Valid = 1;
3351 Event.n.u3Type = SVM_EVENT_EXCEPTION;
3352 Event.n.u8Vector = X86_XCPT_DF;
3353 Event.n.u1ErrorCodeValid = 1;
3354 Event.n.u32ErrorCode = 0;
3355 hmR0SvmSetPendingEvent(pVCpu, &Event, 0 /* GCPtrFaultAddress */);
3356}
3357
3358
3359/**
3360 * Injects an event into the guest upon VMRUN by updating the relevant field
3361 * in the VMCB.
3362 *
3363 * @param pVCpu The cross context virtual CPU structure.
3364 * @param pVmcb Pointer to the guest VM control block.
3365 * @param pEvent Pointer to the event.
3366 *
3367 * @remarks No-long-jump zone!!!
3368 * @remarks Requires CR0!
3369 */
3370DECLINLINE(void) hmR0SvmInjectEventVmcb(PVMCPUCC pVCpu, PSVMVMCB pVmcb, PSVMEVENT pEvent)
3371{
3372 Assert(!pVmcb->ctrl.EventInject.n.u1Valid);
3373 pVmcb->ctrl.EventInject.u = pEvent->u;
3374 if ( pVmcb->ctrl.EventInject.n.u3Type == SVM_EVENT_EXCEPTION
3375 || pVmcb->ctrl.EventInject.n.u3Type == SVM_EVENT_NMI)
3376 {
3377 Assert(pEvent->n.u8Vector <= X86_XCPT_LAST);
3378 STAM_COUNTER_INC(&pVCpu->hm.s.aStatInjectedXcpts[pEvent->n.u8Vector]);
3379 }
3380 else
3381 STAM_COUNTER_INC(&pVCpu->hm.s.aStatInjectedIrqs[pEvent->n.u8Vector & MASK_INJECT_IRQ_STAT]);
3382 RT_NOREF(pVCpu);
3383
3384 Log4Func(("u=%#RX64 u8Vector=%#x Type=%#x ErrorCodeValid=%RTbool ErrorCode=%#RX32\n", pEvent->u, pEvent->n.u8Vector,
3385 (uint8_t)pEvent->n.u3Type, !!pEvent->n.u1ErrorCodeValid, pEvent->n.u32ErrorCode));
3386}
3387
3388
3389
3390/**
3391 * Converts any TRPM trap into a pending HM event. This is typically used when
3392 * entering from ring-3 (not longjmp returns).
3393 *
3394 * @param pVCpu The cross context virtual CPU structure.
3395 */
3396static void hmR0SvmTrpmTrapToPendingEvent(PVMCPUCC pVCpu)
3397{
3398 Assert(TRPMHasTrap(pVCpu));
3399 Assert(!pVCpu->hm.s.Event.fPending);
3400
3401 uint8_t uVector;
3402 TRPMEVENT enmTrpmEvent;
3403 uint32_t uErrCode;
3404 RTGCUINTPTR GCPtrFaultAddress;
3405 uint8_t cbInstr;
3406
3407 int rc = TRPMQueryTrapAll(pVCpu, &uVector, &enmTrpmEvent, &uErrCode, &GCPtrFaultAddress, &cbInstr, NULL /* pfIcebp */);
3408 AssertRC(rc);
3409
3410 SVMEVENT Event;
3411 Event.u = 0;
3412 Event.n.u1Valid = 1;
3413 Event.n.u8Vector = uVector;
3414
3415 /* Refer AMD spec. 15.20 "Event Injection" for the format. */
3416 if (enmTrpmEvent == TRPM_TRAP)
3417 {
3418 Event.n.u3Type = SVM_EVENT_EXCEPTION;
3419 switch (uVector)
3420 {
3421 case X86_XCPT_NMI:
3422 {
3423 Event.n.u3Type = SVM_EVENT_NMI;
3424 break;
3425 }
3426
3427 case X86_XCPT_BP:
3428 case X86_XCPT_OF:
3429 AssertMsgFailed(("Invalid TRPM vector %d for event type %d\n", uVector, enmTrpmEvent));
3430 RT_FALL_THRU();
3431
3432 case X86_XCPT_PF:
3433 case X86_XCPT_DF:
3434 case X86_XCPT_TS:
3435 case X86_XCPT_NP:
3436 case X86_XCPT_SS:
3437 case X86_XCPT_GP:
3438 case X86_XCPT_AC:
3439 {
3440 Event.n.u1ErrorCodeValid = 1;
3441 Event.n.u32ErrorCode = uErrCode;
3442 break;
3443 }
3444 }
3445 }
3446 else if (enmTrpmEvent == TRPM_HARDWARE_INT)
3447 Event.n.u3Type = SVM_EVENT_EXTERNAL_IRQ;
3448 else if (enmTrpmEvent == TRPM_SOFTWARE_INT)
3449 Event.n.u3Type = SVM_EVENT_SOFTWARE_INT;
3450 else
3451 AssertMsgFailed(("Invalid TRPM event type %d\n", enmTrpmEvent));
3452
3453 rc = TRPMResetTrap(pVCpu);
3454 AssertRC(rc);
3455
3456 Log4(("TRPM->HM event: u=%#RX64 u8Vector=%#x uErrorCodeValid=%RTbool uErrorCode=%#RX32\n", Event.u, Event.n.u8Vector,
3457 !!Event.n.u1ErrorCodeValid, Event.n.u32ErrorCode));
3458
3459 hmR0SvmSetPendingEvent(pVCpu, &Event, GCPtrFaultAddress);
3460}
3461
3462
3463/**
3464 * Converts any pending SVM event into a TRPM trap. Typically used when leaving
3465 * AMD-V to execute any instruction.
3466 *
3467 * @param pVCpu The cross context virtual CPU structure.
3468 */
3469static void hmR0SvmPendingEventToTrpmTrap(PVMCPUCC pVCpu)
3470{
3471 Assert(pVCpu->hm.s.Event.fPending);
3472 Assert(TRPMQueryTrap(pVCpu, NULL /* pu8TrapNo */, NULL /* pEnmType */) == VERR_TRPM_NO_ACTIVE_TRAP);
3473
3474 SVMEVENT Event;
3475 Event.u = pVCpu->hm.s.Event.u64IntInfo;
3476
3477 uint8_t uVector = Event.n.u8Vector;
3478 TRPMEVENT enmTrapType = HMSvmEventToTrpmEventType(&Event, uVector);
3479
3480 Log4(("HM event->TRPM: uVector=%#x enmTrapType=%d\n", uVector, Event.n.u3Type));
3481
3482 int rc = TRPMAssertTrap(pVCpu, uVector, enmTrapType);
3483 AssertRC(rc);
3484
3485 if (Event.n.u1ErrorCodeValid)
3486 TRPMSetErrorCode(pVCpu, Event.n.u32ErrorCode);
3487
3488 if ( enmTrapType == TRPM_TRAP
3489 && uVector == X86_XCPT_PF)
3490 {
3491 TRPMSetFaultAddress(pVCpu, pVCpu->hm.s.Event.GCPtrFaultAddress);
3492 Assert(pVCpu->hm.s.Event.GCPtrFaultAddress == CPUMGetGuestCR2(pVCpu));
3493 }
3494 else if (enmTrapType == TRPM_SOFTWARE_INT)
3495 TRPMSetInstrLength(pVCpu, pVCpu->hm.s.Event.cbInstr);
3496 pVCpu->hm.s.Event.fPending = false;
3497}
3498
3499
3500/**
3501 * Sets the virtual interrupt intercept control in the VMCB.
3502 *
3503 * @param pVCpu The cross context virtual CPU structure.
3504 * @param pVmcb Pointer to the VM control block.
3505 */
3506static void hmR0SvmSetIntWindowExiting(PVMCPUCC pVCpu, PSVMVMCB pVmcb)
3507{
3508 HMSVM_ASSERT_NOT_IN_NESTED_GUEST(&pVCpu->cpum.GstCtx); NOREF(pVCpu);
3509
3510 /*
3511 * When AVIC isn't supported, set up an interrupt window to cause a #VMEXIT when the guest
3512 * is ready to accept interrupts. At #VMEXIT, we then get the interrupt from the APIC
3513 * (updating ISR at the right time) and inject the interrupt.
3514 *
3515 * With AVIC is supported, we could make use of the asynchronously delivery without
3516 * #VMEXIT and we would be passing the AVIC page to SVM.
3517 *
3518 * In AMD-V, an interrupt window is achieved using a combination of V_IRQ (an interrupt
3519 * is pending), V_IGN_TPR (ignore TPR priorities) and the VINTR intercept all being set.
3520 */
3521 Assert(pVmcb->ctrl.IntCtrl.n.u1IgnoreTPR);
3522 pVmcb->ctrl.IntCtrl.n.u1VIrqPending = 1;
3523 pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_INT_CTRL;
3524 hmR0SvmSetCtrlIntercept(pVmcb, SVM_CTRL_INTERCEPT_VINTR);
3525 Log4(("Set VINTR intercept\n"));
3526}
3527
3528
3529/**
3530 * Clears the virtual interrupt intercept control in the VMCB as
3531 * we are figured the guest is unable process any interrupts
3532 * at this point of time.
3533 *
3534 * @param pVCpu The cross context virtual CPU structure.
3535 * @param pVmcb Pointer to the VM control block.
3536 */
3537static void hmR0SvmClearIntWindowExiting(PVMCPUCC pVCpu, PSVMVMCB pVmcb)
3538{
3539 HMSVM_ASSERT_NOT_IN_NESTED_GUEST(&pVCpu->cpum.GstCtx); NOREF(pVCpu);
3540
3541 PSVMVMCBCTRL pVmcbCtrl = &pVmcb->ctrl;
3542 if ( pVmcbCtrl->IntCtrl.n.u1VIrqPending
3543 || (pVmcbCtrl->u64InterceptCtrl & SVM_CTRL_INTERCEPT_VINTR))
3544 {
3545 pVmcbCtrl->IntCtrl.n.u1VIrqPending = 0;
3546 pVmcbCtrl->u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_INT_CTRL;
3547 hmR0SvmClearCtrlIntercept(pVCpu, pVmcb, SVM_CTRL_INTERCEPT_VINTR);
3548 Log4(("Cleared VINTR intercept\n"));
3549 }
3550}
3551
3552
3553/**
3554 * Evaluates the event to be delivered to the guest and sets it as the pending
3555 * event.
3556 *
3557 * @returns Strict VBox status code.
3558 * @param pVCpu The cross context virtual CPU structure.
3559 * @param pSvmTransient Pointer to the SVM transient structure.
3560 */
3561static VBOXSTRICTRC hmR0SvmEvaluatePendingEvent(PVMCPUCC pVCpu, PCSVMTRANSIENT pSvmTransient)
3562{
3563 PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
3564 HMSVM_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_HWVIRT
3565 | CPUMCTX_EXTRN_RFLAGS
3566 | CPUMCTX_EXTRN_INHIBIT_INT
3567 | CPUMCTX_EXTRN_HM_SVM_HWVIRT_VIRQ);
3568
3569 Assert(!pVCpu->hm.s.Event.fPending);
3570 PSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
3571 Assert(pVmcb);
3572
3573 bool const fGif = CPUMGetGuestGif(pCtx);
3574 bool const fIntShadow = CPUMIsInInterruptShadowWithUpdate(pCtx);
3575 bool const fBlockNmi = CPUMAreInterruptsInhibitedByNmi(pCtx);
3576
3577 Log4Func(("fGif=%RTbool fBlockNmi=%RTbool fIntShadow=%RTbool fIntPending=%RTbool fNmiPending=%RTbool\n",
3578 fGif, fBlockNmi, fIntShadow, VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_INTERRUPT_APIC | VMCPU_FF_INTERRUPT_PIC),
3579 VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INTERRUPT_NMI)));
3580
3581 /** @todo SMI. SMIs take priority over NMIs. */
3582
3583 /*
3584 * Check if the guest or nested-guest can receive NMIs.
3585 * Nested NMIs are not allowed, see AMD spec. 8.1.4 "Masking External Interrupts".
3586 * NMIs take priority over maskable interrupts, see AMD spec. 8.5 "Priorities".
3587 */
3588 if ( VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INTERRUPT_NMI)
3589 && !fBlockNmi)
3590 {
3591 if ( fGif
3592 && !fIntShadow)
3593 {
3594#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
3595 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_NMI))
3596 {
3597 Log4(("Intercepting NMI -> #VMEXIT\n"));
3598 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, HMSVM_CPUMCTX_EXTRN_ALL);
3599 return IEMExecSvmVmexit(pVCpu, SVM_EXIT_NMI, 0, 0);
3600 }
3601#endif
3602 Log4(("Setting NMI pending for injection\n"));
3603 SVMEVENT Event;
3604 Event.u = 0;
3605 Event.n.u1Valid = 1;
3606 Event.n.u8Vector = X86_XCPT_NMI;
3607 Event.n.u3Type = SVM_EVENT_NMI;
3608 hmR0SvmSetPendingEvent(pVCpu, &Event, 0 /* GCPtrFaultAddress */);
3609 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INTERRUPT_NMI);
3610 }
3611 else if (!fGif)
3612 hmR0SvmSetCtrlIntercept(pVmcb, SVM_CTRL_INTERCEPT_STGI);
3613 else if (!pSvmTransient->fIsNestedGuest)
3614 hmR0SvmSetIntWindowExiting(pVCpu, pVmcb);
3615 /* else: for nested-guests, interrupt-window exiting will be picked up when merging VMCB controls. */
3616 }
3617 /*
3618 * Check if the guest can receive external interrupts (PIC/APIC). Once PDMGetInterrupt()
3619 * returns a valid interrupt we -must- deliver the interrupt. We can no longer re-request
3620 * it from the APIC device.
3621 *
3622 * For nested-guests, physical interrupts always take priority over virtual interrupts.
3623 * We don't need to inject nested-guest virtual interrupts here, we can let the hardware
3624 * do that work when we execute nested-guest code esp. since all the required information
3625 * is in the VMCB, unlike physical interrupts where we need to fetch the interrupt from
3626 * the virtual interrupt controller.
3627 *
3628 * See AMD spec. 15.21.4 "Injecting Virtual (INTR) Interrupts".
3629 */
3630 else if ( VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_INTERRUPT_APIC | VMCPU_FF_INTERRUPT_PIC)
3631 && !pVCpu->hm.s.fSingleInstruction)
3632 {
3633 bool const fBlockInt = !pSvmTransient->fIsNestedGuest ? !(pCtx->eflags.u32 & X86_EFL_IF)
3634 : CPUMIsGuestSvmPhysIntrEnabled(pVCpu, pCtx);
3635 if ( fGif
3636 && !fBlockInt
3637 && !fIntShadow)
3638 {
3639#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
3640 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_INTR))
3641 {
3642 Log4(("Intercepting INTR -> #VMEXIT\n"));
3643 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, HMSVM_CPUMCTX_EXTRN_ALL);
3644 return IEMExecSvmVmexit(pVCpu, SVM_EXIT_INTR, 0, 0);
3645 }
3646#endif
3647 uint8_t u8Interrupt;
3648 int rc = PDMGetInterrupt(pVCpu, &u8Interrupt);
3649 if (RT_SUCCESS(rc))
3650 {
3651 Log4(("Setting external interrupt %#x pending for injection\n", u8Interrupt));
3652 SVMEVENT Event;
3653 Event.u = 0;
3654 Event.n.u1Valid = 1;
3655 Event.n.u8Vector = u8Interrupt;
3656 Event.n.u3Type = SVM_EVENT_EXTERNAL_IRQ;
3657 hmR0SvmSetPendingEvent(pVCpu, &Event, 0 /* GCPtrFaultAddress */);
3658 }
3659 else if (rc == VERR_APIC_INTR_MASKED_BY_TPR)
3660 {
3661 /*
3662 * AMD-V has no TPR thresholding feature. TPR and the force-flag will be
3663 * updated eventually when the TPR is written by the guest.
3664 */
3665 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchTprMaskedIrq);
3666 }
3667 else
3668 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchGuestIrq);
3669 }
3670 else if (!fGif)
3671 hmR0SvmSetCtrlIntercept(pVmcb, SVM_CTRL_INTERCEPT_STGI);
3672 else if (!pSvmTransient->fIsNestedGuest)
3673 hmR0SvmSetIntWindowExiting(pVCpu, pVmcb);
3674 /* else: for nested-guests, interrupt-window exiting will be picked up when merging VMCB controls. */
3675 }
3676
3677 return VINF_SUCCESS;
3678}
3679
3680
3681/**
3682 * Injects any pending events into the guest (or nested-guest).
3683 *
3684 * @param pVCpu The cross context virtual CPU structure.
3685 * @param pVmcb Pointer to the VM control block.
3686 *
3687 * @remarks Must only be called when we are guaranteed to enter
3688 * hardware-assisted SVM execution and not return to ring-3
3689 * prematurely.
3690 */
3691static void hmR0SvmInjectPendingEvent(PVMCPUCC pVCpu, PSVMVMCB pVmcb)
3692{
3693 Assert(!TRPMHasTrap(pVCpu));
3694 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
3695
3696 bool const fIntShadow = CPUMIsInInterruptShadowWithUpdate(&pVCpu->cpum.GstCtx);
3697#ifdef VBOX_STRICT
3698 PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
3699 bool const fGif = CPUMGetGuestGif(pCtx);
3700 bool fAllowInt = fGif;
3701 if (fGif)
3702 {
3703 /*
3704 * For nested-guests we have no way to determine if we're injecting a physical or
3705 * virtual interrupt at this point. Hence the partial verification below.
3706 */
3707 if (CPUMIsGuestInSvmNestedHwVirtMode(pCtx))
3708 fAllowInt = CPUMIsGuestSvmPhysIntrEnabled(pVCpu, pCtx) || CPUMIsGuestSvmVirtIntrEnabled(pVCpu, pCtx);
3709 else
3710 fAllowInt = RT_BOOL(pCtx->eflags.u32 & X86_EFL_IF);
3711 }
3712#endif
3713
3714 if (pVCpu->hm.s.Event.fPending)
3715 {
3716 SVMEVENT Event;
3717 Event.u = pVCpu->hm.s.Event.u64IntInfo;
3718 Assert(Event.n.u1Valid);
3719
3720 /*
3721 * Validate event injection pre-conditions.
3722 */
3723 if (Event.n.u3Type == SVM_EVENT_EXTERNAL_IRQ)
3724 {
3725 Assert(fAllowInt);
3726 Assert(!fIntShadow);
3727 }
3728 else if (Event.n.u3Type == SVM_EVENT_NMI)
3729 {
3730 Assert(fGif);
3731 Assert(!fIntShadow);
3732 }
3733
3734 /*
3735 * Before injecting an NMI we must set VMCPU_FF_BLOCK_NMIS to prevent nested NMIs. We
3736 * do this only when we are surely going to inject the NMI as otherwise if we return
3737 * to ring-3 prematurely we could leave NMIs blocked indefinitely upon re-entry into
3738 * SVM R0.
3739 *
3740 * With VT-x, this is handled by the Guest interruptibility information VMCS field
3741 * which will set the VMCS field after actually delivering the NMI which we read on
3742 * VM-exit to determine the state.
3743 */
3744 if ( Event.n.u3Type == SVM_EVENT_NMI
3745 && Event.n.u8Vector == X86_XCPT_NMI)
3746 CPUMSetInterruptInhibitingByNmi(&pVCpu->cpum.GstCtx);
3747
3748 /*
3749 * Inject it (update VMCB for injection by the hardware).
3750 */
3751 Log4(("Injecting pending HM event\n"));
3752 hmR0SvmInjectEventVmcb(pVCpu, pVmcb, &Event);
3753 pVCpu->hm.s.Event.fPending = false;
3754
3755 if (Event.n.u3Type == SVM_EVENT_EXTERNAL_IRQ)
3756 STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectInterrupt);
3757 else
3758 STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectXcpt);
3759 }
3760 else
3761 Assert(pVmcb->ctrl.EventInject.n.u1Valid == 0);
3762
3763 /*
3764 * We could have injected an NMI through IEM and continue guest execution using
3765 * hardware-assisted SVM. In which case, we would not have any events pending (above)
3766 * but we still need to intercept IRET in order to eventually clear NMI inhibition.
3767 */
3768 if (CPUMAreInterruptsInhibitedByNmi(&pVCpu->cpum.GstCtx))
3769 hmR0SvmSetCtrlIntercept(pVmcb, SVM_CTRL_INTERCEPT_IRET);
3770
3771 /*
3772 * Update the guest interrupt shadow in the guest (or nested-guest) VMCB.
3773 *
3774 * For nested-guests: We need to update it too for the scenario where IEM executes
3775 * the nested-guest but execution later continues here with an interrupt shadow active.
3776 */
3777 pVmcb->ctrl.IntShadow.n.u1IntShadow = fIntShadow;
3778}
3779
3780
3781/**
3782 * Reports world-switch error and dumps some useful debug info.
3783 *
3784 * @param pVCpu The cross context virtual CPU structure.
3785 * @param rcVMRun The return code from VMRUN (or
3786 * VERR_SVM_INVALID_GUEST_STATE for invalid
3787 * guest-state).
3788 */
3789static void hmR0SvmReportWorldSwitchError(PVMCPUCC pVCpu, int rcVMRun)
3790{
3791 HMSVM_ASSERT_PREEMPT_SAFE(pVCpu);
3792 HMSVM_ASSERT_NOT_IN_NESTED_GUEST(&pVCpu->cpum.GstCtx);
3793 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, HMSVM_CPUMCTX_EXTRN_ALL);
3794
3795 if (rcVMRun == VERR_SVM_INVALID_GUEST_STATE)
3796 {
3797#ifdef VBOX_STRICT
3798 hmR0DumpRegs(pVCpu, HM_DUMP_REG_FLAGS_ALL);
3799 PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
3800 Log4(("ctrl.u32VmcbCleanBits %#RX32\n", pVmcb->ctrl.u32VmcbCleanBits));
3801 Log4(("ctrl.u16InterceptRdCRx %#x\n", pVmcb->ctrl.u16InterceptRdCRx));
3802 Log4(("ctrl.u16InterceptWrCRx %#x\n", pVmcb->ctrl.u16InterceptWrCRx));
3803 Log4(("ctrl.u16InterceptRdDRx %#x\n", pVmcb->ctrl.u16InterceptRdDRx));
3804 Log4(("ctrl.u16InterceptWrDRx %#x\n", pVmcb->ctrl.u16InterceptWrDRx));
3805 Log4(("ctrl.u32InterceptXcpt %#x\n", pVmcb->ctrl.u32InterceptXcpt));
3806 Log4(("ctrl.u64InterceptCtrl %#RX64\n", pVmcb->ctrl.u64InterceptCtrl));
3807 Log4(("ctrl.u64IOPMPhysAddr %#RX64\n", pVmcb->ctrl.u64IOPMPhysAddr));
3808 Log4(("ctrl.u64MSRPMPhysAddr %#RX64\n", pVmcb->ctrl.u64MSRPMPhysAddr));
3809 Log4(("ctrl.u64TSCOffset %#RX64\n", pVmcb->ctrl.u64TSCOffset));
3810
3811 Log4(("ctrl.TLBCtrl.u32ASID %#x\n", pVmcb->ctrl.TLBCtrl.n.u32ASID));
3812 Log4(("ctrl.TLBCtrl.u8TLBFlush %#x\n", pVmcb->ctrl.TLBCtrl.n.u8TLBFlush));
3813 Log4(("ctrl.TLBCtrl.u24Reserved %#x\n", pVmcb->ctrl.TLBCtrl.n.u24Reserved));
3814
3815 Log4(("ctrl.IntCtrl.u8VTPR %#x\n", pVmcb->ctrl.IntCtrl.n.u8VTPR));
3816 Log4(("ctrl.IntCtrl.u1VIrqPending %#x\n", pVmcb->ctrl.IntCtrl.n.u1VIrqPending));
3817 Log4(("ctrl.IntCtrl.u1VGif %#x\n", pVmcb->ctrl.IntCtrl.n.u1VGif));
3818 Log4(("ctrl.IntCtrl.u6Reserved0 %#x\n", pVmcb->ctrl.IntCtrl.n.u6Reserved));
3819 Log4(("ctrl.IntCtrl.u4VIntrPrio %#x\n", pVmcb->ctrl.IntCtrl.n.u4VIntrPrio));
3820 Log4(("ctrl.IntCtrl.u1IgnoreTPR %#x\n", pVmcb->ctrl.IntCtrl.n.u1IgnoreTPR));
3821 Log4(("ctrl.IntCtrl.u3Reserved %#x\n", pVmcb->ctrl.IntCtrl.n.u3Reserved));
3822 Log4(("ctrl.IntCtrl.u1VIntrMasking %#x\n", pVmcb->ctrl.IntCtrl.n.u1VIntrMasking));
3823 Log4(("ctrl.IntCtrl.u1VGifEnable %#x\n", pVmcb->ctrl.IntCtrl.n.u1VGifEnable));
3824 Log4(("ctrl.IntCtrl.u5Reserved1 %#x\n", pVmcb->ctrl.IntCtrl.n.u5Reserved));
3825 Log4(("ctrl.IntCtrl.u8VIntrVector %#x\n", pVmcb->ctrl.IntCtrl.n.u8VIntrVector));
3826 Log4(("ctrl.IntCtrl.u24Reserved %#x\n", pVmcb->ctrl.IntCtrl.n.u24Reserved));
3827
3828 Log4(("ctrl.IntShadow.u1IntShadow %#x\n", pVmcb->ctrl.IntShadow.n.u1IntShadow));
3829 Log4(("ctrl.IntShadow.u1GuestIntMask %#x\n", pVmcb->ctrl.IntShadow.n.u1GuestIntMask));
3830 Log4(("ctrl.u64ExitCode %#RX64\n", pVmcb->ctrl.u64ExitCode));
3831 Log4(("ctrl.u64ExitInfo1 %#RX64\n", pVmcb->ctrl.u64ExitInfo1));
3832 Log4(("ctrl.u64ExitInfo2 %#RX64\n", pVmcb->ctrl.u64ExitInfo2));
3833 Log4(("ctrl.ExitIntInfo.u8Vector %#x\n", pVmcb->ctrl.ExitIntInfo.n.u8Vector));
3834 Log4(("ctrl.ExitIntInfo.u3Type %#x\n", pVmcb->ctrl.ExitIntInfo.n.u3Type));
3835 Log4(("ctrl.ExitIntInfo.u1ErrorCodeValid %#x\n", pVmcb->ctrl.ExitIntInfo.n.u1ErrorCodeValid));
3836 Log4(("ctrl.ExitIntInfo.u19Reserved %#x\n", pVmcb->ctrl.ExitIntInfo.n.u19Reserved));
3837 Log4(("ctrl.ExitIntInfo.u1Valid %#x\n", pVmcb->ctrl.ExitIntInfo.n.u1Valid));
3838 Log4(("ctrl.ExitIntInfo.u32ErrorCode %#x\n", pVmcb->ctrl.ExitIntInfo.n.u32ErrorCode));
3839 Log4(("ctrl.NestedPagingCtrl.u1NestedPaging %#x\n", pVmcb->ctrl.NestedPagingCtrl.n.u1NestedPaging));
3840 Log4(("ctrl.NestedPagingCtrl.u1Sev %#x\n", pVmcb->ctrl.NestedPagingCtrl.n.u1Sev));
3841 Log4(("ctrl.NestedPagingCtrl.u1SevEs %#x\n", pVmcb->ctrl.NestedPagingCtrl.n.u1SevEs));
3842 Log4(("ctrl.EventInject.u8Vector %#x\n", pVmcb->ctrl.EventInject.n.u8Vector));
3843 Log4(("ctrl.EventInject.u3Type %#x\n", pVmcb->ctrl.EventInject.n.u3Type));
3844 Log4(("ctrl.EventInject.u1ErrorCodeValid %#x\n", pVmcb->ctrl.EventInject.n.u1ErrorCodeValid));
3845 Log4(("ctrl.EventInject.u19Reserved %#x\n", pVmcb->ctrl.EventInject.n.u19Reserved));
3846 Log4(("ctrl.EventInject.u1Valid %#x\n", pVmcb->ctrl.EventInject.n.u1Valid));
3847 Log4(("ctrl.EventInject.u32ErrorCode %#x\n", pVmcb->ctrl.EventInject.n.u32ErrorCode));
3848
3849 Log4(("ctrl.u64NestedPagingCR3 %#RX64\n", pVmcb->ctrl.u64NestedPagingCR3));
3850
3851 Log4(("ctrl.LbrVirt.u1LbrVirt %#x\n", pVmcb->ctrl.LbrVirt.n.u1LbrVirt));
3852 Log4(("ctrl.LbrVirt.u1VirtVmsaveVmload %#x\n", pVmcb->ctrl.LbrVirt.n.u1VirtVmsaveVmload));
3853
3854 Log4(("guest.CS.u16Sel %RTsel\n", pVmcb->guest.CS.u16Sel));
3855 Log4(("guest.CS.u16Attr %#x\n", pVmcb->guest.CS.u16Attr));
3856 Log4(("guest.CS.u32Limit %#RX32\n", pVmcb->guest.CS.u32Limit));
3857 Log4(("guest.CS.u64Base %#RX64\n", pVmcb->guest.CS.u64Base));
3858 Log4(("guest.DS.u16Sel %#RTsel\n", pVmcb->guest.DS.u16Sel));
3859 Log4(("guest.DS.u16Attr %#x\n", pVmcb->guest.DS.u16Attr));
3860 Log4(("guest.DS.u32Limit %#RX32\n", pVmcb->guest.DS.u32Limit));
3861 Log4(("guest.DS.u64Base %#RX64\n", pVmcb->guest.DS.u64Base));
3862 Log4(("guest.ES.u16Sel %RTsel\n", pVmcb->guest.ES.u16Sel));
3863 Log4(("guest.ES.u16Attr %#x\n", pVmcb->guest.ES.u16Attr));
3864 Log4(("guest.ES.u32Limit %#RX32\n", pVmcb->guest.ES.u32Limit));
3865 Log4(("guest.ES.u64Base %#RX64\n", pVmcb->guest.ES.u64Base));
3866 Log4(("guest.FS.u16Sel %RTsel\n", pVmcb->guest.FS.u16Sel));
3867 Log4(("guest.FS.u16Attr %#x\n", pVmcb->guest.FS.u16Attr));
3868 Log4(("guest.FS.u32Limit %#RX32\n", pVmcb->guest.FS.u32Limit));
3869 Log4(("guest.FS.u64Base %#RX64\n", pVmcb->guest.FS.u64Base));
3870 Log4(("guest.GS.u16Sel %RTsel\n", pVmcb->guest.GS.u16Sel));
3871 Log4(("guest.GS.u16Attr %#x\n", pVmcb->guest.GS.u16Attr));
3872 Log4(("guest.GS.u32Limit %#RX32\n", pVmcb->guest.GS.u32Limit));
3873 Log4(("guest.GS.u64Base %#RX64\n", pVmcb->guest.GS.u64Base));
3874
3875 Log4(("guest.GDTR.u32Limit %#RX32\n", pVmcb->guest.GDTR.u32Limit));
3876 Log4(("guest.GDTR.u64Base %#RX64\n", pVmcb->guest.GDTR.u64Base));
3877
3878 Log4(("guest.LDTR.u16Sel %RTsel\n", pVmcb->guest.LDTR.u16Sel));
3879 Log4(("guest.LDTR.u16Attr %#x\n", pVmcb->guest.LDTR.u16Attr));
3880 Log4(("guest.LDTR.u32Limit %#RX32\n", pVmcb->guest.LDTR.u32Limit));
3881 Log4(("guest.LDTR.u64Base %#RX64\n", pVmcb->guest.LDTR.u64Base));
3882
3883 Log4(("guest.IDTR.u32Limit %#RX32\n", pVmcb->guest.IDTR.u32Limit));
3884 Log4(("guest.IDTR.u64Base %#RX64\n", pVmcb->guest.IDTR.u64Base));
3885
3886 Log4(("guest.TR.u16Sel %RTsel\n", pVmcb->guest.TR.u16Sel));
3887 Log4(("guest.TR.u16Attr %#x\n", pVmcb->guest.TR.u16Attr));
3888 Log4(("guest.TR.u32Limit %#RX32\n", pVmcb->guest.TR.u32Limit));
3889 Log4(("guest.TR.u64Base %#RX64\n", pVmcb->guest.TR.u64Base));
3890
3891 Log4(("guest.u8CPL %#x\n", pVmcb->guest.u8CPL));
3892 Log4(("guest.u64CR0 %#RX64\n", pVmcb->guest.u64CR0));
3893 Log4(("guest.u64CR2 %#RX64\n", pVmcb->guest.u64CR2));
3894 Log4(("guest.u64CR3 %#RX64\n", pVmcb->guest.u64CR3));
3895 Log4(("guest.u64CR4 %#RX64\n", pVmcb->guest.u64CR4));
3896 Log4(("guest.u64DR6 %#RX64\n", pVmcb->guest.u64DR6));
3897 Log4(("guest.u64DR7 %#RX64\n", pVmcb->guest.u64DR7));
3898
3899 Log4(("guest.u64RIP %#RX64\n", pVmcb->guest.u64RIP));
3900 Log4(("guest.u64RSP %#RX64\n", pVmcb->guest.u64RSP));
3901 Log4(("guest.u64RAX %#RX64\n", pVmcb->guest.u64RAX));
3902 Log4(("guest.u64RFlags %#RX64\n", pVmcb->guest.u64RFlags));
3903
3904 Log4(("guest.u64SysEnterCS %#RX64\n", pVmcb->guest.u64SysEnterCS));
3905 Log4(("guest.u64SysEnterEIP %#RX64\n", pVmcb->guest.u64SysEnterEIP));
3906 Log4(("guest.u64SysEnterESP %#RX64\n", pVmcb->guest.u64SysEnterESP));
3907
3908 Log4(("guest.u64EFER %#RX64\n", pVmcb->guest.u64EFER));
3909 Log4(("guest.u64STAR %#RX64\n", pVmcb->guest.u64STAR));
3910 Log4(("guest.u64LSTAR %#RX64\n", pVmcb->guest.u64LSTAR));
3911 Log4(("guest.u64CSTAR %#RX64\n", pVmcb->guest.u64CSTAR));
3912 Log4(("guest.u64SFMASK %#RX64\n", pVmcb->guest.u64SFMASK));
3913 Log4(("guest.u64KernelGSBase %#RX64\n", pVmcb->guest.u64KernelGSBase));
3914 Log4(("guest.u64PAT %#RX64\n", pVmcb->guest.u64PAT));
3915 Log4(("guest.u64DBGCTL %#RX64\n", pVmcb->guest.u64DBGCTL));
3916 Log4(("guest.u64BR_FROM %#RX64\n", pVmcb->guest.u64BR_FROM));
3917 Log4(("guest.u64BR_TO %#RX64\n", pVmcb->guest.u64BR_TO));
3918 Log4(("guest.u64LASTEXCPFROM %#RX64\n", pVmcb->guest.u64LASTEXCPFROM));
3919 Log4(("guest.u64LASTEXCPTO %#RX64\n", pVmcb->guest.u64LASTEXCPTO));
3920
3921 NOREF(pVmcb);
3922#endif /* VBOX_STRICT */
3923 }
3924 else
3925 Log4Func(("rcVMRun=%d\n", rcVMRun));
3926}
3927
3928
3929/**
3930 * Check per-VM and per-VCPU force flag actions that require us to go back to
3931 * ring-3 for one reason or another.
3932 *
3933 * @returns Strict VBox status code (information status code included).
3934 * @retval VINF_SUCCESS if we don't have any actions that require going back to
3935 * ring-3.
3936 * @retval VINF_PGM_SYNC_CR3 if we have pending PGM CR3 sync.
3937 * @retval VINF_EM_PENDING_REQUEST if we have pending requests (like hardware
3938 * interrupts)
3939 * @retval VINF_PGM_POOL_FLUSH_PENDING if PGM is doing a pool flush and requires
3940 * all EMTs to be in ring-3.
3941 * @retval VINF_EM_RAW_TO_R3 if there is pending DMA requests.
3942 * @retval VINF_EM_NO_MEMORY PGM is out of memory, we need to return
3943 * to the EM loop.
3944 *
3945 * @param pVCpu The cross context virtual CPU structure.
3946 */
3947static VBOXSTRICTRC hmR0SvmCheckForceFlags(PVMCPUCC pVCpu)
3948{
3949 Assert(VMMRZCallRing3IsEnabled(pVCpu));
3950
3951 /* Could happen as a result of longjump. */
3952 if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3))
3953 PGMUpdateCR3(pVCpu, CPUMGetGuestCR3(pVCpu));
3954
3955 /* Update pending interrupts into the APIC's IRR. */
3956 if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_UPDATE_APIC))
3957 APICUpdatePendingInterrupts(pVCpu);
3958
3959 PVMCC pVM = pVCpu->CTX_SUFF(pVM);
3960 if ( VM_FF_IS_ANY_SET(pVM, !pVCpu->hm.s.fSingleInstruction
3961 ? VM_FF_HP_R0_PRE_HM_MASK : VM_FF_HP_R0_PRE_HM_STEP_MASK)
3962 || VMCPU_FF_IS_ANY_SET(pVCpu, !pVCpu->hm.s.fSingleInstruction
3963 ? VMCPU_FF_HP_R0_PRE_HM_MASK : VMCPU_FF_HP_R0_PRE_HM_STEP_MASK) )
3964 {
3965 /* Pending PGM C3 sync. */
3966 if (VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3 | VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL))
3967 {
3968 int rc = PGMSyncCR3(pVCpu, pVCpu->cpum.GstCtx.cr0, pVCpu->cpum.GstCtx.cr3, pVCpu->cpum.GstCtx.cr4,
3969 VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3));
3970 if (rc != VINF_SUCCESS)
3971 {
3972 Log4Func(("PGMSyncCR3 forcing us back to ring-3. rc=%d\n", rc));
3973 return rc;
3974 }
3975 }
3976
3977 /* Pending HM-to-R3 operations (critsects, timers, EMT rendezvous etc.) */
3978 /* -XXX- what was that about single stepping? */
3979 if ( VM_FF_IS_ANY_SET(pVM, VM_FF_HM_TO_R3_MASK)
3980 || VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HM_TO_R3_MASK))
3981 {
3982 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHmToR3FF);
3983 int rc = RT_LIKELY(!VM_FF_IS_SET(pVM, VM_FF_PGM_NO_MEMORY)) ? VINF_EM_RAW_TO_R3 : VINF_EM_NO_MEMORY;
3984 Log4Func(("HM_TO_R3 forcing us back to ring-3. rc=%d\n", rc));
3985 return rc;
3986 }
3987
3988 /* Pending VM request packets, such as hardware interrupts. */
3989 if ( VM_FF_IS_SET(pVM, VM_FF_REQUEST)
3990 || VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_REQUEST))
3991 {
3992 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchVmReq);
3993 Log4Func(("Pending VM request forcing us back to ring-3\n"));
3994 return VINF_EM_PENDING_REQUEST;
3995 }
3996
3997 /* Pending PGM pool flushes. */
3998 if (VM_FF_IS_SET(pVM, VM_FF_PGM_POOL_FLUSH_PENDING))
3999 {
4000 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchPgmPoolFlush);
4001 Log4Func(("PGM pool flush pending forcing us back to ring-3\n"));
4002 return VINF_PGM_POOL_FLUSH_PENDING;
4003 }
4004
4005 /* Pending DMA requests. */
4006 if (VM_FF_IS_SET(pVM, VM_FF_PDM_DMA))
4007 {
4008 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchDma);
4009 Log4Func(("Pending DMA request forcing us back to ring-3\n"));
4010 return VINF_EM_RAW_TO_R3;
4011 }
4012 }
4013
4014 return VINF_SUCCESS;
4015}
4016
4017
4018/**
4019 * Does the preparations before executing guest code in AMD-V.
4020 *
4021 * This may cause longjmps to ring-3 and may even result in rescheduling to the
4022 * recompiler. We must be cautious what we do here regarding committing
4023 * guest-state information into the VMCB assuming we assuredly execute the guest
4024 * in AMD-V. If we fall back to the recompiler after updating the VMCB and
4025 * clearing the common-state (TRPM/forceflags), we must undo those changes so
4026 * that the recompiler can (and should) use them when it resumes guest
4027 * execution. Otherwise such operations must be done when we can no longer
4028 * exit to ring-3.
4029 *
4030 * @returns Strict VBox status code (informational status codes included).
4031 * @retval VINF_SUCCESS if we can proceed with running the guest.
4032 * @retval VINF_* scheduling changes, we have to go back to ring-3.
4033 *
4034 * @param pVCpu The cross context virtual CPU structure.
4035 * @param pSvmTransient Pointer to the SVM transient structure.
4036 */
4037static VBOXSTRICTRC hmR0SvmPreRunGuest(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
4038{
4039 HMSVM_ASSERT_PREEMPT_SAFE(pVCpu);
4040
4041#ifdef VBOX_WITH_NESTED_HWVIRT_ONLY_IN_IEM
4042 if (pSvmTransient->fIsNestedGuest)
4043 {
4044 Log2(("hmR0SvmPreRunGuest: Rescheduling to IEM due to nested-hwvirt or forced IEM exec -> VINF_EM_RESCHEDULE_REM\n"));
4045 return VINF_EM_RESCHEDULE_REM;
4046 }
4047#endif
4048
4049 /* Check force flag actions that might require us to go back to ring-3. */
4050 VBOXSTRICTRC rc = hmR0SvmCheckForceFlags(pVCpu);
4051 if (rc != VINF_SUCCESS)
4052 return rc;
4053
4054 if (TRPMHasTrap(pVCpu))
4055 hmR0SvmTrpmTrapToPendingEvent(pVCpu);
4056 else if (!pVCpu->hm.s.Event.fPending)
4057 {
4058 rc = hmR0SvmEvaluatePendingEvent(pVCpu, pSvmTransient);
4059 if ( rc != VINF_SUCCESS
4060 || pSvmTransient->fIsNestedGuest != CPUMIsGuestInSvmNestedHwVirtMode(&pVCpu->cpum.GstCtx))
4061 {
4062 /* If a nested-guest VM-exit occurred, bail. */
4063 if (pSvmTransient->fIsNestedGuest)
4064 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchNstGstVmexit);
4065 return rc;
4066 }
4067 }
4068
4069 /*
4070 * On the oldest AMD-V systems, we may not get enough information to reinject an NMI.
4071 * Just do it in software, see @bugref{8411}.
4072 * NB: If we could continue a task switch exit we wouldn't need to do this.
4073 */
4074 PVMCC pVM = pVCpu->CTX_SUFF(pVM);
4075 if (RT_UNLIKELY( !g_fHmSvmFeatures
4076 && pVCpu->hm.s.Event.fPending
4077 && SVM_EVENT_GET_TYPE(pVCpu->hm.s.Event.u64IntInfo) == SVM_EVENT_NMI))
4078 return VINF_EM_RAW_INJECT_TRPM_EVENT;
4079
4080#ifdef HMSVM_SYNC_FULL_GUEST_STATE
4081 Assert(!(pVCpu->cpum.GstCtx.fExtrn & HMSVM_CPUMCTX_EXTRN_ALL));
4082 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
4083#endif
4084
4085#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
4086 /*
4087 * Set up the nested-guest VMCB for execution using hardware-assisted SVM.
4088 */
4089 if (pSvmTransient->fIsNestedGuest)
4090 hmR0SvmSetupVmcbNested(pVCpu);
4091#endif
4092
4093 /*
4094 * Export the guest state bits that are not shared with the host in any way as we can
4095 * longjmp or get preempted in the midst of exporting some of the state.
4096 */
4097 rc = hmR0SvmExportGuestState(pVCpu, pSvmTransient);
4098 AssertRCReturn(rc, rc);
4099 STAM_COUNTER_INC(&pVCpu->hm.s.StatExportFull);
4100
4101 /* Ensure we've cached (and hopefully modified) the nested-guest VMCB for execution using hardware-assisted SVM. */
4102 Assert(!pSvmTransient->fIsNestedGuest || pVCpu->hm.s.svm.NstGstVmcbCache.fCacheValid);
4103
4104 /*
4105 * If we're not intercepting TPR changes in the guest, save the guest TPR before the
4106 * world-switch so we can update it on the way back if the guest changed the TPR.
4107 */
4108 if (pVCpu->hmr0.s.svm.fSyncVTpr)
4109 {
4110 Assert(!pSvmTransient->fIsNestedGuest);
4111 PCSVMVMCB pVmcb = pVCpu->hmr0.s.svm.pVmcb;
4112 if (pVM->hm.s.fTprPatchingActive)
4113 pSvmTransient->u8GuestTpr = pVmcb->guest.u64LSTAR;
4114 else
4115 pSvmTransient->u8GuestTpr = pVmcb->ctrl.IntCtrl.n.u8VTPR;
4116 }
4117
4118 /*
4119 * No longjmps to ring-3 from this point on!!!
4120 *
4121 * Asserts() will still longjmp to ring-3 (but won't return), which is intentional,
4122 * better than a kernel panic. This also disables flushing of the R0-logger instance.
4123 */
4124 VMMRZCallRing3Disable(pVCpu);
4125
4126 /*
4127 * We disable interrupts so that we don't miss any interrupts that would flag preemption
4128 * (IPI/timers etc.) when thread-context hooks aren't used and we've been running with
4129 * preemption disabled for a while. Since this is purly to aid the
4130 * RTThreadPreemptIsPending() code, it doesn't matter that it may temporarily reenable and
4131 * disable interrupt on NT.
4132 *
4133 * We need to check for force-flags that could've possible been altered since we last
4134 * checked them (e.g. by PDMGetInterrupt() leaving the PDM critical section,
4135 * see @bugref{6398}).
4136 *
4137 * We also check a couple of other force-flags as a last opportunity to get the EMT back
4138 * to ring-3 before executing guest code.
4139 */
4140 pSvmTransient->fEFlags = ASMIntDisableFlags();
4141 if ( VM_FF_IS_ANY_SET(pVM, VM_FF_EMT_RENDEZVOUS | VM_FF_TM_VIRTUAL_SYNC)
4142 || VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_HM_TO_R3_MASK))
4143 {
4144 ASMSetFlags(pSvmTransient->fEFlags);
4145 VMMRZCallRing3Enable(pVCpu);
4146 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHmToR3FF);
4147 return VINF_EM_RAW_TO_R3;
4148 }
4149 if (RTThreadPreemptIsPending(NIL_RTTHREAD))
4150 {
4151 ASMSetFlags(pSvmTransient->fEFlags);
4152 VMMRZCallRing3Enable(pVCpu);
4153 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchPendingHostIrq);
4154 return VINF_EM_RAW_INTERRUPT;
4155 }
4156
4157 return VINF_SUCCESS;
4158}
4159
4160
4161/**
4162 * Prepares to run guest (or nested-guest) code in AMD-V and we've committed to
4163 * doing so.
4164 *
4165 * This means there is no backing out to ring-3 or anywhere else at this point.
4166 *
4167 * @param pVCpu The cross context virtual CPU structure.
4168 * @param pSvmTransient Pointer to the SVM transient structure.
4169 *
4170 * @remarks Called with preemption disabled.
4171 * @remarks No-long-jump zone!!!
4172 */
4173static void hmR0SvmPreRunGuestCommitted(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
4174{
4175 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
4176 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
4177
4178 VMCPU_ASSERT_STATE(pVCpu, VMCPUSTATE_STARTED_HM);
4179 VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED_EXEC); /* Indicate the start of guest execution. */
4180
4181 PVMCC pVM = pVCpu->CTX_SUFF(pVM);
4182 PSVMVMCB pVmcb = pSvmTransient->pVmcb;
4183
4184 hmR0SvmInjectPendingEvent(pVCpu, pVmcb);
4185
4186 if (!CPUMIsGuestFPUStateActive(pVCpu))
4187 {
4188 STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatLoadGuestFpuState, x);
4189 CPUMR0LoadGuestFPU(pVM, pVCpu); /* (Ignore rc, no need to set HM_CHANGED_HOST_CONTEXT for SVM.) */
4190 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatLoadGuestFpuState, x);
4191 STAM_COUNTER_INC(&pVCpu->hm.s.StatLoadGuestFpu);
4192 }
4193
4194 /* Load the state shared between host and guest (FPU, debug). */
4195 if (pVCpu->hm.s.fCtxChanged & HM_CHANGED_SVM_HOST_GUEST_SHARED_STATE)
4196 hmR0SvmExportSharedState(pVCpu, pVmcb);
4197
4198 pVCpu->hm.s.fCtxChanged &= ~HM_CHANGED_HOST_CONTEXT; /* Preemption might set this, nothing to do on AMD-V. */
4199 AssertMsg(!pVCpu->hm.s.fCtxChanged, ("fCtxChanged=%#RX64\n", pVCpu->hm.s.fCtxChanged));
4200
4201 PHMPHYSCPU pHostCpu = hmR0GetCurrentCpu();
4202 RTCPUID const idHostCpu = pHostCpu->idCpu;
4203 bool const fMigratedHostCpu = idHostCpu != pVCpu->hmr0.s.idLastCpu;
4204
4205 /* Setup TSC offsetting. */
4206 if ( pSvmTransient->fUpdateTscOffsetting
4207 || fMigratedHostCpu)
4208 {
4209 hmR0SvmUpdateTscOffsetting(pVCpu, pVmcb);
4210 pSvmTransient->fUpdateTscOffsetting = false;
4211 }
4212
4213 /* Record statistics of how often we use TSC offsetting as opposed to intercepting RDTSC/P. */
4214 if (!(pVmcb->ctrl.u64InterceptCtrl & (SVM_CTRL_INTERCEPT_RDTSC | SVM_CTRL_INTERCEPT_RDTSCP)))
4215 STAM_COUNTER_INC(&pVCpu->hm.s.StatTscOffset);
4216 else
4217 STAM_COUNTER_INC(&pVCpu->hm.s.StatTscIntercept);
4218
4219 /* If we've migrating CPUs, mark the VMCB Clean bits as dirty. */
4220 if (fMigratedHostCpu)
4221 pVmcb->ctrl.u32VmcbCleanBits = 0;
4222
4223 /* Store status of the shared guest-host state at the time of VMRUN. */
4224 pSvmTransient->fWasGuestDebugStateActive = CPUMIsGuestDebugStateActive(pVCpu);
4225 pSvmTransient->fWasHyperDebugStateActive = CPUMIsHyperDebugStateActive(pVCpu);
4226
4227#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
4228 uint8_t *pbMsrBitmap;
4229 if (!pSvmTransient->fIsNestedGuest)
4230 pbMsrBitmap = (uint8_t *)pVCpu->hmr0.s.svm.pvMsrBitmap;
4231 else
4232 {
4233 /** @todo We could perhaps optimize this by monitoring if the guest modifies its
4234 * MSRPM and only perform this if it changed also use EVEX.POR when it
4235 * does. */
4236 hmR0SvmMergeMsrpmNested(pHostCpu, pVCpu);
4237
4238 /* Update the nested-guest VMCB with the newly merged MSRPM (clean bits updated below). */
4239 pVmcb->ctrl.u64MSRPMPhysAddr = pHostCpu->n.svm.HCPhysNstGstMsrpm;
4240 pbMsrBitmap = (uint8_t *)pHostCpu->n.svm.pvNstGstMsrpm;
4241 }
4242#else
4243 uint8_t *pbMsrBitmap = (uint8_t *)pVCpu->hm.s.svm.pvMsrBitmap;
4244#endif
4245
4246 ASMAtomicUoWriteBool(&pVCpu->hm.s.fCheckedTLBFlush, true); /* Used for TLB flushing, set this across the world switch. */
4247 /* Flush the appropriate tagged-TLB entries. */
4248 hmR0SvmFlushTaggedTlb(pHostCpu, pVCpu, pVmcb);
4249 Assert(pVCpu->hmr0.s.idLastCpu == idHostCpu);
4250
4251 STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatEntry, &pVCpu->hm.s.StatInGC, x);
4252
4253 TMNotifyStartOfExecution(pVM, pVCpu); /* Finally, notify TM to resume its clocks as we're about
4254 to start executing. */
4255
4256 /*
4257 * Save the current Host TSC_AUX and write the guest TSC_AUX to the host, so that RDTSCPs
4258 * (that don't cause exits) reads the guest MSR, see @bugref{3324}.
4259 *
4260 * This should be done -after- any RDTSCPs for obtaining the host timestamp (TM, STAM etc).
4261 */
4262 if ( g_CpumHostFeatures.s.fRdTscP
4263 && !(pVmcb->ctrl.u64InterceptCtrl & SVM_CTRL_INTERCEPT_RDTSCP))
4264 {
4265 uint64_t const uGuestTscAux = CPUMGetGuestTscAux(pVCpu);
4266 pVCpu->hmr0.s.svm.u64HostTscAux = ASMRdMsr(MSR_K8_TSC_AUX);
4267 if (uGuestTscAux != pVCpu->hmr0.s.svm.u64HostTscAux)
4268 ASMWrMsr(MSR_K8_TSC_AUX, uGuestTscAux);
4269 hmR0SvmSetMsrPermission(pVCpu, pbMsrBitmap, MSR_K8_TSC_AUX, SVMMSREXIT_PASSTHRU_READ, SVMMSREXIT_PASSTHRU_WRITE);
4270 pSvmTransient->fRestoreTscAuxMsr = true;
4271 }
4272 else
4273 {
4274 hmR0SvmSetMsrPermission(pVCpu, pbMsrBitmap, MSR_K8_TSC_AUX, SVMMSREXIT_INTERCEPT_READ, SVMMSREXIT_INTERCEPT_WRITE);
4275 pSvmTransient->fRestoreTscAuxMsr = false;
4276 }
4277 pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_IOPM_MSRPM;
4278
4279 /*
4280 * If VMCB Clean bits isn't supported by the CPU or exposed to the guest in the nested
4281 * virtualization case, mark all state-bits as dirty indicating to the CPU to re-load
4282 * from the VMCB.
4283 */
4284 bool const fSupportsVmcbCleanBits = hmR0SvmSupportsVmcbCleanBits(pVCpu, pSvmTransient->fIsNestedGuest);
4285 if (!fSupportsVmcbCleanBits)
4286 pVmcb->ctrl.u32VmcbCleanBits = 0;
4287}
4288
4289
4290/**
4291 * Wrapper for running the guest (or nested-guest) code in AMD-V.
4292 *
4293 * @returns VBox strict status code.
4294 * @param pVCpu The cross context virtual CPU structure.
4295 * @param HCPhysVmcb The host physical address of the VMCB.
4296 *
4297 * @remarks No-long-jump zone!!!
4298 */
4299DECLINLINE(int) hmR0SvmRunGuest(PVMCPUCC pVCpu, RTHCPHYS HCPhysVmcb)
4300{
4301 /* Mark that HM is the keeper of all guest-CPU registers now that we're going to execute guest code. */
4302 pVCpu->cpum.GstCtx.fExtrn |= HMSVM_CPUMCTX_EXTRN_ALL | CPUMCTX_EXTRN_KEEPER_HM;
4303 return pVCpu->hmr0.s.svm.pfnVMRun(pVCpu->CTX_SUFF(pVM), pVCpu, HCPhysVmcb);
4304}
4305
4306
4307/**
4308 * Performs some essential restoration of state after running guest (or
4309 * nested-guest) code in AMD-V.
4310 *
4311 * @param pVCpu The cross context virtual CPU structure.
4312 * @param pSvmTransient Pointer to the SVM transient structure.
4313 * @param rcVMRun Return code of VMRUN.
4314 *
4315 * @remarks Called with interrupts disabled.
4316 * @remarks No-long-jump zone!!! This function will however re-enable longjmps
4317 * unconditionally when it is safe to do so.
4318 */
4319static void hmR0SvmPostRunGuest(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient, VBOXSTRICTRC rcVMRun)
4320{
4321 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
4322
4323 ASMAtomicUoWriteBool(&pVCpu->hm.s.fCheckedTLBFlush, false); /* See HMInvalidatePageOnAllVCpus(): used for TLB flushing. */
4324 ASMAtomicIncU32(&pVCpu->hmr0.s.cWorldSwitchExits); /* Initialized in vmR3CreateUVM(): used for EMT poking. */
4325
4326 PSVMVMCB pVmcb = pSvmTransient->pVmcb;
4327 PSVMVMCBCTRL pVmcbCtrl = &pVmcb->ctrl;
4328
4329 /* TSC read must be done early for maximum accuracy. */
4330 if (!(pVmcbCtrl->u64InterceptCtrl & SVM_CTRL_INTERCEPT_RDTSC))
4331 {
4332 if (!pSvmTransient->fIsNestedGuest)
4333 TMCpuTickSetLastSeen(pVCpu, pVCpu->hmr0.s.uTscExit + pVmcbCtrl->u64TSCOffset);
4334#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
4335 else
4336 {
4337 /* The nested-guest VMCB TSC offset shall eventually be restored on #VMEXIT via HMNotifySvmNstGstVmexit(). */
4338 uint64_t const uGstTsc = CPUMRemoveNestedGuestTscOffset(pVCpu, pVCpu->hmr0.s.uTscExit + pVmcbCtrl->u64TSCOffset);
4339 TMCpuTickSetLastSeen(pVCpu, uGstTsc);
4340 }
4341#endif
4342 }
4343
4344 if (pSvmTransient->fRestoreTscAuxMsr)
4345 {
4346 uint64_t u64GuestTscAuxMsr = ASMRdMsr(MSR_K8_TSC_AUX);
4347 CPUMSetGuestTscAux(pVCpu, u64GuestTscAuxMsr);
4348 if (u64GuestTscAuxMsr != pVCpu->hmr0.s.svm.u64HostTscAux)
4349 ASMWrMsr(MSR_K8_TSC_AUX, pVCpu->hmr0.s.svm.u64HostTscAux);
4350 }
4351
4352 STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatInGC, &pVCpu->hm.s.StatPreExit, x);
4353 PVMCC pVM = pVCpu->CTX_SUFF(pVM);
4354 TMNotifyEndOfExecution(pVM, pVCpu, pVCpu->hmr0.s.uTscExit); /* Notify TM that the guest is no longer running. */
4355 VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED_HM);
4356
4357 Assert(!(ASMGetFlags() & X86_EFL_IF));
4358 ASMSetFlags(pSvmTransient->fEFlags); /* Enable interrupts. */
4359 VMMRZCallRing3Enable(pVCpu); /* It is now safe to do longjmps to ring-3!!! */
4360
4361 /* If VMRUN failed, we can bail out early. This does -not- cover SVM_EXIT_INVALID. */
4362 if (RT_UNLIKELY(rcVMRun != VINF_SUCCESS))
4363 {
4364 Log4Func(("VMRUN failure: rcVMRun=%Rrc\n", VBOXSTRICTRC_VAL(rcVMRun)));
4365 return;
4366 }
4367
4368 pSvmTransient->u64ExitCode = pVmcbCtrl->u64ExitCode; /* Save the #VMEXIT reason. */
4369 pSvmTransient->fVectoringDoublePF = false; /* Vectoring double page-fault needs to be determined later. */
4370 pSvmTransient->fVectoringPF = false; /* Vectoring page-fault needs to be determined later. */
4371 pVmcbCtrl->u32VmcbCleanBits = HMSVM_VMCB_CLEAN_ALL; /* Mark the VMCB-state cache as unmodified by VMM. */
4372
4373#ifdef HMSVM_SYNC_FULL_GUEST_STATE
4374 hmR0SvmImportGuestState(pVCpu, HMSVM_CPUMCTX_EXTRN_ALL);
4375 Assert(!(pVCpu->cpum.GstCtx.fExtrn & HMSVM_CPUMCTX_EXTRN_ALL));
4376#else
4377 /*
4378 * Always import the following:
4379 *
4380 * - RIP for exit optimizations and evaluating event injection on re-entry.
4381 * - RFLAGS for evaluating event injection on VM re-entry and for exporting shared debug
4382 * state on preemption.
4383 * - Interrupt shadow, GIF for evaluating event injection on VM re-entry.
4384 * - CS for exit optimizations.
4385 * - RAX, RSP for simplifying assumptions on GPRs. All other GPRs are swapped by the
4386 * assembly switcher code.
4387 * - Shared state (only DR7 currently) for exporting shared debug state on preemption.
4388 */
4389 hmR0SvmImportGuestState(pVCpu, CPUMCTX_EXTRN_RIP
4390 | CPUMCTX_EXTRN_RFLAGS
4391 | CPUMCTX_EXTRN_RAX
4392 | CPUMCTX_EXTRN_RSP
4393 | CPUMCTX_EXTRN_CS
4394 | CPUMCTX_EXTRN_HWVIRT
4395 | CPUMCTX_EXTRN_INHIBIT_INT
4396 | CPUMCTX_EXTRN_HM_SVM_HWVIRT_VIRQ
4397 | HMSVM_CPUMCTX_SHARED_STATE);
4398#endif
4399
4400 if ( pSvmTransient->u64ExitCode != SVM_EXIT_INVALID
4401 && pVCpu->hmr0.s.svm.fSyncVTpr)
4402 {
4403 Assert(!pSvmTransient->fIsNestedGuest);
4404 /* TPR patching (for 32-bit guests) uses LSTAR MSR for holding the TPR value, otherwise uses the VTPR. */
4405 if ( pVM->hm.s.fTprPatchingActive
4406 && (pVmcb->guest.u64LSTAR & 0xff) != pSvmTransient->u8GuestTpr)
4407 {
4408 int rc = APICSetTpr(pVCpu, pVmcb->guest.u64LSTAR & 0xff);
4409 AssertRC(rc);
4410 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_APIC_TPR);
4411 }
4412 /* Sync TPR when we aren't intercepting CR8 writes. */
4413 else if (pSvmTransient->u8GuestTpr != pVmcbCtrl->IntCtrl.n.u8VTPR)
4414 {
4415 int rc = APICSetTpr(pVCpu, pVmcbCtrl->IntCtrl.n.u8VTPR << 4);
4416 AssertRC(rc);
4417 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_APIC_TPR);
4418 }
4419 }
4420
4421#ifdef DEBUG_ramshankar
4422 if (CPUMIsGuestInSvmNestedHwVirtMode(&pVCpu->cpum.GstCtx))
4423 {
4424 hmR0SvmImportGuestState(pVCpu, HMSVM_CPUMCTX_EXTRN_ALL);
4425 hmR0SvmLogState(pVCpu, pVmcb, pVCpu->cpum.GstCtx, "hmR0SvmPostRunGuestNested", HMSVM_LOG_ALL & ~HMSVM_LOG_LBR,
4426 0 /* uVerbose */);
4427 }
4428#endif
4429
4430 HMSVM_CPUMCTX_ASSERT(pVCpu, CPUMCTX_EXTRN_CS | CPUMCTX_EXTRN_RIP);
4431 EMHistoryAddExit(pVCpu, EMEXIT_MAKE_FT(EMEXIT_F_KIND_SVM, pSvmTransient->u64ExitCode & EMEXIT_F_TYPE_MASK),
4432 pVCpu->cpum.GstCtx.cs.u64Base + pVCpu->cpum.GstCtx.rip, pVCpu->hmr0.s.uTscExit);
4433}
4434
4435
4436/**
4437 * Runs the guest code using AMD-V.
4438 *
4439 * @returns Strict VBox status code.
4440 * @param pVCpu The cross context virtual CPU structure.
4441 * @param pcLoops Pointer to the number of executed loops.
4442 */
4443static VBOXSTRICTRC hmR0SvmRunGuestCodeNormal(PVMCPUCC pVCpu, uint32_t *pcLoops)
4444{
4445 uint32_t const cMaxResumeLoops = pVCpu->CTX_SUFF(pVM)->hmr0.s.cMaxResumeLoops;
4446 Assert(pcLoops);
4447 Assert(*pcLoops <= cMaxResumeLoops);
4448
4449 SVMTRANSIENT SvmTransient;
4450 RT_ZERO(SvmTransient);
4451 SvmTransient.fUpdateTscOffsetting = true;
4452 SvmTransient.pVmcb = pVCpu->hmr0.s.svm.pVmcb;
4453
4454 VBOXSTRICTRC rc = VERR_INTERNAL_ERROR_5;
4455 for (;;)
4456 {
4457 Assert(!HMR0SuspendPending());
4458 HMSVM_ASSERT_CPU_SAFE(pVCpu);
4459
4460 /* Preparatory work for running nested-guest code, this may force us to return to
4461 ring-3. This bugger disables interrupts on VINF_SUCCESS! */
4462 STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
4463 rc = hmR0SvmPreRunGuest(pVCpu, &SvmTransient);
4464 if (rc != VINF_SUCCESS)
4465 break;
4466
4467 /*
4468 * No longjmps to ring-3 from this point on!!!
4469 *
4470 * Asserts() will still longjmp to ring-3 (but won't return), which is intentional,
4471 * better than a kernel panic. This also disables flushing of the R0-logger instance.
4472 */
4473 hmR0SvmPreRunGuestCommitted(pVCpu, &SvmTransient);
4474 rc = hmR0SvmRunGuest(pVCpu, pVCpu->hmr0.s.svm.HCPhysVmcb);
4475
4476 /* Restore any residual host-state and save any bits shared between host and guest
4477 into the guest-CPU state. Re-enables interrupts! */
4478 hmR0SvmPostRunGuest(pVCpu, &SvmTransient, rc);
4479
4480 if (RT_UNLIKELY( rc != VINF_SUCCESS /* Check for VMRUN errors. */
4481 || SvmTransient.u64ExitCode == SVM_EXIT_INVALID)) /* Check for invalid guest-state errors. */
4482 {
4483 if (rc == VINF_SUCCESS)
4484 rc = VERR_SVM_INVALID_GUEST_STATE;
4485 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatPreExit, x);
4486 hmR0SvmReportWorldSwitchError(pVCpu, VBOXSTRICTRC_VAL(rc));
4487 break;
4488 }
4489
4490 /* Handle the #VMEXIT. */
4491 HMSVM_EXITCODE_STAM_COUNTER_INC(SvmTransient.u64ExitCode);
4492 STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatPreExit, &pVCpu->hm.s.StatExitHandling, x);
4493 VBOXVMM_R0_HMSVM_VMEXIT(pVCpu, &pVCpu->cpum.GstCtx, SvmTransient.u64ExitCode, pVCpu->hmr0.s.svm.pVmcb);
4494 rc = hmR0SvmHandleExit(pVCpu, &SvmTransient);
4495 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitHandling, x);
4496 if (rc != VINF_SUCCESS)
4497 break;
4498 if (++(*pcLoops) >= cMaxResumeLoops)
4499 {
4500 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchMaxResumeLoops);
4501 rc = VINF_EM_RAW_INTERRUPT;
4502 break;
4503 }
4504 }
4505
4506 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
4507 return rc;
4508}
4509
4510
4511#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
4512/**
4513 * Runs the nested-guest code using AMD-V.
4514 *
4515 * @returns Strict VBox status code.
4516 * @param pVCpu The cross context virtual CPU structure.
4517 * @param pcLoops Pointer to the number of executed loops. If we're switching
4518 * from the guest-code execution loop to this nested-guest
4519 * execution loop pass the remainder value, else pass 0.
4520 */
4521static VBOXSTRICTRC hmR0SvmRunGuestCodeNested(PVMCPUCC pVCpu, uint32_t *pcLoops)
4522{
4523 PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
4524 HMSVM_ASSERT_IN_NESTED_GUEST(pCtx);
4525 Assert(pcLoops);
4526 Assert(*pcLoops <= pVCpu->CTX_SUFF(pVM)->hmr0.s.cMaxResumeLoops);
4527 /** @todo r=bird: Sharing this with ring-3 isn't safe in the long run, I fear... */
4528 RTHCPHYS const HCPhysVmcb = GVMMR0ConvertGVMPtr2HCPhys(pVCpu->pGVM, &pCtx->hwvirt.svm.Vmcb);
4529
4530 SVMTRANSIENT SvmTransient;
4531 RT_ZERO(SvmTransient);
4532 SvmTransient.fUpdateTscOffsetting = true;
4533 SvmTransient.pVmcb = &pCtx->hwvirt.svm.Vmcb;
4534 SvmTransient.fIsNestedGuest = true;
4535
4536 /* Setup pointer so PGM/IEM can query #VMEXIT auxiliary info. on demand in ring-0. */
4537 pVCpu->hmr0.s.svm.pSvmTransient = &SvmTransient;
4538
4539 VBOXSTRICTRC rc = VERR_INTERNAL_ERROR_4;
4540 for (;;)
4541 {
4542 Assert(!HMR0SuspendPending());
4543 HMSVM_ASSERT_CPU_SAFE(pVCpu);
4544
4545 /* Preparatory work for running nested-guest code, this may force us to return to
4546 ring-3. This bugger disables interrupts on VINF_SUCCESS! */
4547 STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
4548 rc = hmR0SvmPreRunGuest(pVCpu, &SvmTransient);
4549 if ( rc != VINF_SUCCESS
4550 || !CPUMIsGuestInSvmNestedHwVirtMode(pCtx))
4551 break;
4552
4553 /*
4554 * No longjmps to ring-3 from this point on!!!
4555 *
4556 * Asserts() will still longjmp to ring-3 (but won't return), which is intentional,
4557 * better than a kernel panic. This also disables flushing of the R0-logger instance.
4558 */
4559 hmR0SvmPreRunGuestCommitted(pVCpu, &SvmTransient);
4560
4561 rc = hmR0SvmRunGuest(pVCpu, HCPhysVmcb);
4562
4563 /* Restore any residual host-state and save any bits shared between host and guest
4564 into the guest-CPU state. Re-enables interrupts! */
4565 hmR0SvmPostRunGuest(pVCpu, &SvmTransient, rc);
4566
4567 if (RT_LIKELY( rc == VINF_SUCCESS
4568 && SvmTransient.u64ExitCode != SVM_EXIT_INVALID))
4569 { /* extremely likely */ }
4570 else
4571 {
4572 /* VMRUN failed, shouldn't really happen, Guru. */
4573 if (rc != VINF_SUCCESS)
4574 break;
4575
4576 /* Invalid nested-guest state. Cause a #VMEXIT but assert on strict builds. */
4577 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, HMSVM_CPUMCTX_EXTRN_ALL);
4578 AssertMsgFailed(("Invalid nested-guest state. rc=%Rrc u64ExitCode=%#RX64\n", rc, SvmTransient.u64ExitCode));
4579 rc = IEMExecSvmVmexit(pVCpu, SVM_EXIT_INVALID, 0, 0);
4580 break;
4581 }
4582
4583 /* Handle the #VMEXIT. */
4584 HMSVM_NESTED_EXITCODE_STAM_COUNTER_INC(SvmTransient.u64ExitCode);
4585 STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatPreExit, &pVCpu->hm.s.StatExitHandling, x);
4586 VBOXVMM_R0_HMSVM_VMEXIT(pVCpu, pCtx, SvmTransient.u64ExitCode, &pCtx->hwvirt.svm.Vmcb);
4587 rc = hmR0SvmHandleExitNested(pVCpu, &SvmTransient);
4588 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitHandling, x);
4589 if (rc == VINF_SUCCESS)
4590 {
4591 if (!CPUMIsGuestInSvmNestedHwVirtMode(pCtx))
4592 {
4593 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchNstGstVmexit);
4594 rc = VINF_SVM_VMEXIT;
4595 }
4596 else
4597 {
4598 if (++(*pcLoops) <= pVCpu->CTX_SUFF(pVM)->hmr0.s.cMaxResumeLoops)
4599 continue;
4600 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchMaxResumeLoops);
4601 rc = VINF_EM_RAW_INTERRUPT;
4602 }
4603 }
4604 else
4605 Assert(rc != VINF_SVM_VMEXIT);
4606 break;
4607 /** @todo NSTSVM: handle single-stepping. */
4608 }
4609
4610 /* Ensure #VMEXIT auxiliary info. is no longer available. */
4611 pVCpu->hmr0.s.svm.pSvmTransient = NULL;
4612
4613 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
4614 return rc;
4615}
4616#endif /* VBOX_WITH_NESTED_HWVIRT_SVM */
4617
4618
4619/**
4620 * Checks if any expensive dtrace probes are enabled and we should go to the
4621 * debug loop.
4622 *
4623 * @returns true if we should use debug loop, false if not.
4624 */
4625static bool hmR0SvmAnyExpensiveProbesEnabled(void)
4626{
4627 /* It's probably faster to OR the raw 32-bit counter variables together.
4628 Since the variables are in an array and the probes are next to one
4629 another (more or less), we have good locality. So, better read
4630 eight-nine cache lines ever time and only have one conditional, than
4631 128+ conditionals, right? */
4632 return ( VBOXVMM_R0_HMSVM_VMEXIT_ENABLED_RAW() /* expensive too due to context */
4633 | VBOXVMM_XCPT_DE_ENABLED_RAW()
4634 | VBOXVMM_XCPT_DB_ENABLED_RAW()
4635 | VBOXVMM_XCPT_BP_ENABLED_RAW()
4636 | VBOXVMM_XCPT_OF_ENABLED_RAW()
4637 | VBOXVMM_XCPT_BR_ENABLED_RAW()
4638 | VBOXVMM_XCPT_UD_ENABLED_RAW()
4639 | VBOXVMM_XCPT_NM_ENABLED_RAW()
4640 | VBOXVMM_XCPT_DF_ENABLED_RAW()
4641 | VBOXVMM_XCPT_TS_ENABLED_RAW()
4642 | VBOXVMM_XCPT_NP_ENABLED_RAW()
4643 | VBOXVMM_XCPT_SS_ENABLED_RAW()
4644 | VBOXVMM_XCPT_GP_ENABLED_RAW()
4645 | VBOXVMM_XCPT_PF_ENABLED_RAW()
4646 | VBOXVMM_XCPT_MF_ENABLED_RAW()
4647 | VBOXVMM_XCPT_AC_ENABLED_RAW()
4648 | VBOXVMM_XCPT_XF_ENABLED_RAW()
4649 | VBOXVMM_XCPT_VE_ENABLED_RAW()
4650 | VBOXVMM_XCPT_SX_ENABLED_RAW()
4651 | VBOXVMM_INT_SOFTWARE_ENABLED_RAW()
4652 | VBOXVMM_INT_HARDWARE_ENABLED_RAW()
4653 ) != 0
4654 || ( VBOXVMM_INSTR_HALT_ENABLED_RAW()
4655 | VBOXVMM_INSTR_MWAIT_ENABLED_RAW()
4656 | VBOXVMM_INSTR_MONITOR_ENABLED_RAW()
4657 | VBOXVMM_INSTR_CPUID_ENABLED_RAW()
4658 | VBOXVMM_INSTR_INVD_ENABLED_RAW()
4659 | VBOXVMM_INSTR_WBINVD_ENABLED_RAW()
4660 | VBOXVMM_INSTR_INVLPG_ENABLED_RAW()
4661 | VBOXVMM_INSTR_RDTSC_ENABLED_RAW()
4662 | VBOXVMM_INSTR_RDTSCP_ENABLED_RAW()
4663 | VBOXVMM_INSTR_RDPMC_ENABLED_RAW()
4664 | VBOXVMM_INSTR_RDMSR_ENABLED_RAW()
4665 | VBOXVMM_INSTR_WRMSR_ENABLED_RAW()
4666 | VBOXVMM_INSTR_CRX_READ_ENABLED_RAW()
4667 | VBOXVMM_INSTR_CRX_WRITE_ENABLED_RAW()
4668 | VBOXVMM_INSTR_DRX_READ_ENABLED_RAW()
4669 | VBOXVMM_INSTR_DRX_WRITE_ENABLED_RAW()
4670 | VBOXVMM_INSTR_PAUSE_ENABLED_RAW()
4671 | VBOXVMM_INSTR_XSETBV_ENABLED_RAW()
4672 | VBOXVMM_INSTR_SIDT_ENABLED_RAW()
4673 | VBOXVMM_INSTR_LIDT_ENABLED_RAW()
4674 | VBOXVMM_INSTR_SGDT_ENABLED_RAW()
4675 | VBOXVMM_INSTR_LGDT_ENABLED_RAW()
4676 | VBOXVMM_INSTR_SLDT_ENABLED_RAW()
4677 | VBOXVMM_INSTR_LLDT_ENABLED_RAW()
4678 | VBOXVMM_INSTR_STR_ENABLED_RAW()
4679 | VBOXVMM_INSTR_LTR_ENABLED_RAW()
4680 //| VBOXVMM_INSTR_GETSEC_ENABLED_RAW()
4681 | VBOXVMM_INSTR_RSM_ENABLED_RAW()
4682 //| VBOXVMM_INSTR_RDRAND_ENABLED_RAW()
4683 //| VBOXVMM_INSTR_RDSEED_ENABLED_RAW()
4684 //| VBOXVMM_INSTR_XSAVES_ENABLED_RAW()
4685 //| VBOXVMM_INSTR_XRSTORS_ENABLED_RAW()
4686 | VBOXVMM_INSTR_VMM_CALL_ENABLED_RAW()
4687 | VBOXVMM_INSTR_SVM_VMRUN_ENABLED_RAW()
4688 | VBOXVMM_INSTR_SVM_VMLOAD_ENABLED_RAW()
4689 | VBOXVMM_INSTR_SVM_VMSAVE_ENABLED_RAW()
4690 | VBOXVMM_INSTR_SVM_STGI_ENABLED_RAW()
4691 | VBOXVMM_INSTR_SVM_CLGI_ENABLED_RAW()
4692 ) != 0
4693 || ( VBOXVMM_EXIT_TASK_SWITCH_ENABLED_RAW()
4694 | VBOXVMM_EXIT_HALT_ENABLED_RAW()
4695 | VBOXVMM_EXIT_MWAIT_ENABLED_RAW()
4696 | VBOXVMM_EXIT_MONITOR_ENABLED_RAW()
4697 | VBOXVMM_EXIT_CPUID_ENABLED_RAW()
4698 | VBOXVMM_EXIT_INVD_ENABLED_RAW()
4699 | VBOXVMM_EXIT_WBINVD_ENABLED_RAW()
4700 | VBOXVMM_EXIT_INVLPG_ENABLED_RAW()
4701 | VBOXVMM_EXIT_RDTSC_ENABLED_RAW()
4702 | VBOXVMM_EXIT_RDTSCP_ENABLED_RAW()
4703 | VBOXVMM_EXIT_RDPMC_ENABLED_RAW()
4704 | VBOXVMM_EXIT_RDMSR_ENABLED_RAW()
4705 | VBOXVMM_EXIT_WRMSR_ENABLED_RAW()
4706 | VBOXVMM_EXIT_CRX_READ_ENABLED_RAW()
4707 | VBOXVMM_EXIT_CRX_WRITE_ENABLED_RAW()
4708 | VBOXVMM_EXIT_DRX_READ_ENABLED_RAW()
4709 | VBOXVMM_EXIT_DRX_WRITE_ENABLED_RAW()
4710 | VBOXVMM_EXIT_PAUSE_ENABLED_RAW()
4711 | VBOXVMM_EXIT_XSETBV_ENABLED_RAW()
4712 | VBOXVMM_EXIT_SIDT_ENABLED_RAW()
4713 | VBOXVMM_EXIT_LIDT_ENABLED_RAW()
4714 | VBOXVMM_EXIT_SGDT_ENABLED_RAW()
4715 | VBOXVMM_EXIT_LGDT_ENABLED_RAW()
4716 | VBOXVMM_EXIT_SLDT_ENABLED_RAW()
4717 | VBOXVMM_EXIT_LLDT_ENABLED_RAW()
4718 | VBOXVMM_EXIT_STR_ENABLED_RAW()
4719 | VBOXVMM_EXIT_LTR_ENABLED_RAW()
4720 //| VBOXVMM_EXIT_GETSEC_ENABLED_RAW()
4721 | VBOXVMM_EXIT_RSM_ENABLED_RAW()
4722 //| VBOXVMM_EXIT_RDRAND_ENABLED_RAW()
4723 //| VBOXVMM_EXIT_RDSEED_ENABLED_RAW()
4724 //| VBOXVMM_EXIT_XSAVES_ENABLED_RAW()
4725 //| VBOXVMM_EXIT_XRSTORS_ENABLED_RAW()
4726 | VBOXVMM_EXIT_VMM_CALL_ENABLED_RAW()
4727 | VBOXVMM_EXIT_SVM_VMRUN_ENABLED_RAW()
4728 | VBOXVMM_EXIT_SVM_VMLOAD_ENABLED_RAW()
4729 | VBOXVMM_EXIT_SVM_VMSAVE_ENABLED_RAW()
4730 | VBOXVMM_EXIT_SVM_STGI_ENABLED_RAW()
4731 | VBOXVMM_EXIT_SVM_CLGI_ENABLED_RAW()
4732 ) != 0;
4733}
4734
4735
4736/**
4737 * Runs the guest code using AMD-V.
4738 *
4739 * @returns Strict VBox status code.
4740 * @param pVCpu The cross context virtual CPU structure.
4741 */
4742VMMR0DECL(VBOXSTRICTRC) SVMR0RunGuestCode(PVMCPUCC pVCpu)
4743{
4744 AssertPtr(pVCpu);
4745 PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
4746 Assert(VMMRZCallRing3IsEnabled(pVCpu));
4747 Assert(!ASMAtomicUoReadU64(&pCtx->fExtrn));
4748 HMSVM_ASSERT_PREEMPT_SAFE(pVCpu);
4749
4750 uint32_t cLoops = 0;
4751 VBOXSTRICTRC rc;
4752 for (;;)
4753 {
4754#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
4755 bool const fInNestedGuestMode = CPUMIsGuestInSvmNestedHwVirtMode(pCtx);
4756#else
4757 NOREF(pCtx);
4758 bool const fInNestedGuestMode = false;
4759#endif
4760 if (!fInNestedGuestMode)
4761 {
4762 if ( !pVCpu->hm.s.fUseDebugLoop
4763 && (!VBOXVMM_ANY_PROBES_ENABLED() || !hmR0SvmAnyExpensiveProbesEnabled())
4764 && !DBGFIsStepping(pVCpu)
4765 && !pVCpu->CTX_SUFF(pVM)->dbgf.ro.cEnabledInt3Breakpoints)
4766 rc = hmR0SvmRunGuestCodeNormal(pVCpu, &cLoops);
4767 else
4768 rc = hmR0SvmRunGuestCodeDebug(pVCpu, &cLoops);
4769 }
4770#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
4771 else
4772 rc = hmR0SvmRunGuestCodeNested(pVCpu, &cLoops);
4773
4774 if (rc == VINF_SVM_VMRUN)
4775 {
4776 Assert(CPUMIsGuestInSvmNestedHwVirtMode(pCtx));
4777 continue;
4778 }
4779 if (rc == VINF_SVM_VMEXIT)
4780 {
4781 Assert(!CPUMIsGuestInSvmNestedHwVirtMode(pCtx));
4782 continue;
4783 }
4784#endif
4785 break;
4786 }
4787
4788 /* Fixup error codes. */
4789 if (rc == VERR_EM_INTERPRETER)
4790 rc = VINF_EM_RAW_EMULATE_INSTR;
4791 else if (rc == VINF_EM_RESET)
4792 rc = VINF_EM_TRIPLE_FAULT;
4793
4794 /* Prepare to return to ring-3. This will remove longjmp notifications. */
4795 rc = hmR0SvmExitToRing3(pVCpu, rc);
4796 Assert(!ASMAtomicUoReadU64(&pCtx->fExtrn));
4797 Assert(!VMMR0AssertionIsNotificationSet(pVCpu));
4798 return rc;
4799}
4800
4801#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
4802
4803/**
4804 * Determines whether the given I/O access should cause a nested-guest \#VMEXIT.
4805 *
4806 * @param pvIoBitmap Pointer to the nested-guest IO bitmap.
4807 * @param pIoExitInfo Pointer to the SVMIOIOEXITINFO.
4808 */
4809static bool hmR0SvmIsIoInterceptSet(void *pvIoBitmap, PSVMIOIOEXITINFO pIoExitInfo)
4810{
4811 const uint16_t u16Port = pIoExitInfo->n.u16Port;
4812 const SVMIOIOTYPE enmIoType = (SVMIOIOTYPE)pIoExitInfo->n.u1Type;
4813 const uint8_t cbReg = (pIoExitInfo->u >> SVM_IOIO_OP_SIZE_SHIFT) & 7;
4814 const uint8_t cAddrSizeBits = ((pIoExitInfo->u >> SVM_IOIO_ADDR_SIZE_SHIFT) & 7) << 4;
4815 const uint8_t iEffSeg = pIoExitInfo->n.u3Seg;
4816 const bool fRep = pIoExitInfo->n.u1Rep;
4817 const bool fStrIo = pIoExitInfo->n.u1Str;
4818
4819 return CPUMIsSvmIoInterceptSet(pvIoBitmap, u16Port, enmIoType, cbReg, cAddrSizeBits, iEffSeg, fRep, fStrIo,
4820 NULL /* pIoExitInfo */);
4821}
4822
4823
4824/**
4825 * Handles a nested-guest \#VMEXIT (for all EXITCODE values except
4826 * SVM_EXIT_INVALID).
4827 *
4828 * @returns VBox status code (informational status codes included).
4829 * @param pVCpu The cross context virtual CPU structure.
4830 * @param pSvmTransient Pointer to the SVM transient structure.
4831 */
4832static VBOXSTRICTRC hmR0SvmHandleExitNested(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
4833{
4834 HMSVM_ASSERT_IN_NESTED_GUEST(&pVCpu->cpum.GstCtx);
4835 Assert(pSvmTransient->u64ExitCode != SVM_EXIT_INVALID);
4836 Assert(pSvmTransient->u64ExitCode <= SVM_EXIT_MAX);
4837
4838 /*
4839 * We import the complete state here because we use separate VMCBs for the guest and the
4840 * nested-guest, and the guest's VMCB is used after the #VMEXIT. We can only save/restore
4841 * the #VMEXIT specific state if we used the same VMCB for both guest and nested-guest.
4842 */
4843#define NST_GST_VMEXIT_CALL_RET(a_pVCpu, a_uExitCode, a_uExitInfo1, a_uExitInfo2) \
4844 do { \
4845 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, HMSVM_CPUMCTX_EXTRN_ALL); \
4846 return IEMExecSvmVmexit((a_pVCpu), (a_uExitCode), (a_uExitInfo1), (a_uExitInfo2)); \
4847 } while (0)
4848
4849 /*
4850 * For all the #VMEXITs here we primarily figure out if the #VMEXIT is expected by the
4851 * nested-guest. If it isn't, it should be handled by the (outer) guest.
4852 */
4853 PSVMVMCB pVmcbNstGst = &pVCpu->cpum.GstCtx.hwvirt.svm.Vmcb;
4854 PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
4855 PSVMVMCBCTRL pVmcbNstGstCtrl = &pVmcbNstGst->ctrl;
4856 uint64_t const uExitCode = pVmcbNstGstCtrl->u64ExitCode;
4857 uint64_t const uExitInfo1 = pVmcbNstGstCtrl->u64ExitInfo1;
4858 uint64_t const uExitInfo2 = pVmcbNstGstCtrl->u64ExitInfo2;
4859
4860 Assert(uExitCode == pVmcbNstGstCtrl->u64ExitCode);
4861 switch (uExitCode)
4862 {
4863 case SVM_EXIT_CPUID:
4864 {
4865 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_CPUID))
4866 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
4867 return hmR0SvmExitCpuid(pVCpu, pSvmTransient);
4868 }
4869
4870 case SVM_EXIT_RDTSC:
4871 {
4872 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_RDTSC))
4873 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
4874 return hmR0SvmExitRdtsc(pVCpu, pSvmTransient);
4875 }
4876
4877 case SVM_EXIT_RDTSCP:
4878 {
4879 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_RDTSCP))
4880 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
4881 return hmR0SvmExitRdtscp(pVCpu, pSvmTransient);
4882 }
4883
4884 case SVM_EXIT_MONITOR:
4885 {
4886 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_MONITOR))
4887 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
4888 return hmR0SvmExitMonitor(pVCpu, pSvmTransient);
4889 }
4890
4891 case SVM_EXIT_MWAIT:
4892 {
4893 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_MWAIT))
4894 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
4895 return hmR0SvmExitMwait(pVCpu, pSvmTransient);
4896 }
4897
4898 case SVM_EXIT_HLT:
4899 {
4900 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_HLT))
4901 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
4902 return hmR0SvmExitHlt(pVCpu, pSvmTransient);
4903 }
4904
4905 case SVM_EXIT_MSR:
4906 {
4907 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_MSR_PROT))
4908 {
4909 uint32_t const idMsr = pVCpu->cpum.GstCtx.ecx;
4910 uint16_t offMsrpm;
4911 uint8_t uMsrpmBit;
4912 int rc = CPUMGetSvmMsrpmOffsetAndBit(idMsr, &offMsrpm, &uMsrpmBit);
4913 if (RT_SUCCESS(rc))
4914 {
4915 Assert(uMsrpmBit == 0 || uMsrpmBit == 2 || uMsrpmBit == 4 || uMsrpmBit == 6);
4916 Assert(offMsrpm < SVM_MSRPM_PAGES << X86_PAGE_4K_SHIFT);
4917
4918 uint8_t const * const pbMsrBitmap = &pVCpu->cpum.GstCtx.hwvirt.svm.abMsrBitmap[offMsrpm];
4919 bool const fInterceptRead = RT_BOOL(*pbMsrBitmap & RT_BIT(uMsrpmBit));
4920 bool const fInterceptWrite = RT_BOOL(*pbMsrBitmap & RT_BIT(uMsrpmBit + 1));
4921
4922 if ( (fInterceptWrite && pVmcbNstGstCtrl->u64ExitInfo1 == SVM_EXIT1_MSR_WRITE)
4923 || (fInterceptRead && pVmcbNstGstCtrl->u64ExitInfo1 == SVM_EXIT1_MSR_READ))
4924 {
4925 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
4926 }
4927 }
4928 else
4929 {
4930 /*
4931 * MSRs not covered by the MSRPM automatically cause an #VMEXIT.
4932 * See AMD-V spec. "15.11 MSR Intercepts".
4933 */
4934 Assert(rc == VERR_OUT_OF_RANGE);
4935 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
4936 }
4937 }
4938 return hmR0SvmExitMsr(pVCpu, pSvmTransient);
4939 }
4940
4941 case SVM_EXIT_IOIO:
4942 {
4943 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_IOIO_PROT))
4944 {
4945 SVMIOIOEXITINFO IoExitInfo;
4946 IoExitInfo.u = pVmcbNstGst->ctrl.u64ExitInfo1;
4947 bool const fIntercept = hmR0SvmIsIoInterceptSet(pVCpu->cpum.GstCtx.hwvirt.svm.abIoBitmap, &IoExitInfo);
4948 if (fIntercept)
4949 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
4950 }
4951 return hmR0SvmExitIOInstr(pVCpu, pSvmTransient);
4952 }
4953
4954 case SVM_EXIT_XCPT_PF:
4955 {
4956 PVMCC pVM = pVCpu->CTX_SUFF(pVM);
4957 if (pVM->hmr0.s.fNestedPaging)
4958 {
4959 uint32_t const u32ErrCode = pVmcbNstGstCtrl->u64ExitInfo1;
4960 uint64_t const uFaultAddress = pVmcbNstGstCtrl->u64ExitInfo2;
4961
4962 /* If the nested-guest is intercepting #PFs, cause a #PF #VMEXIT. */
4963 if (CPUMIsGuestSvmXcptInterceptSet(pVCpu, pCtx, X86_XCPT_PF))
4964 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, u32ErrCode, uFaultAddress);
4965
4966 /* If the nested-guest is not intercepting #PFs, forward the #PF to the guest. */
4967 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, CPUMCTX_EXTRN_CR2);
4968 hmR0SvmSetPendingXcptPF(pVCpu, u32ErrCode, uFaultAddress);
4969 return VINF_SUCCESS;
4970 }
4971 return hmR0SvmExitXcptPF(pVCpu, pSvmTransient);
4972 }
4973
4974 case SVM_EXIT_XCPT_UD:
4975 {
4976 if (CPUMIsGuestSvmXcptInterceptSet(pVCpu, pCtx, X86_XCPT_UD))
4977 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
4978 hmR0SvmSetPendingXcptUD(pVCpu);
4979 return VINF_SUCCESS;
4980 }
4981
4982 case SVM_EXIT_XCPT_MF:
4983 {
4984 if (CPUMIsGuestSvmXcptInterceptSet(pVCpu, pCtx, X86_XCPT_MF))
4985 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
4986 return hmR0SvmExitXcptMF(pVCpu, pSvmTransient);
4987 }
4988
4989 case SVM_EXIT_XCPT_DB:
4990 {
4991 if (CPUMIsGuestSvmXcptInterceptSet(pVCpu, pCtx, X86_XCPT_DB))
4992 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
4993 return hmR0SvmNestedExitXcptDB(pVCpu, pSvmTransient);
4994 }
4995
4996 case SVM_EXIT_XCPT_AC:
4997 {
4998 if (CPUMIsGuestSvmXcptInterceptSet(pVCpu, pCtx, X86_XCPT_AC))
4999 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5000 return hmR0SvmExitXcptAC(pVCpu, pSvmTransient);
5001 }
5002
5003 case SVM_EXIT_XCPT_BP:
5004 {
5005 if (CPUMIsGuestSvmXcptInterceptSet(pVCpu, pCtx, X86_XCPT_BP))
5006 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5007 return hmR0SvmNestedExitXcptBP(pVCpu, pSvmTransient);
5008 }
5009
5010 case SVM_EXIT_READ_CR0:
5011 case SVM_EXIT_READ_CR3:
5012 case SVM_EXIT_READ_CR4:
5013 {
5014 uint8_t const uCr = uExitCode - SVM_EXIT_READ_CR0;
5015 if (CPUMIsGuestSvmReadCRxInterceptSet(pVCpu, pCtx, uCr))
5016 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5017 return hmR0SvmExitReadCRx(pVCpu, pSvmTransient);
5018 }
5019
5020 case SVM_EXIT_CR0_SEL_WRITE:
5021 {
5022 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_CR0_SEL_WRITE))
5023 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5024 return hmR0SvmExitWriteCRx(pVCpu, pSvmTransient);
5025 }
5026
5027 case SVM_EXIT_WRITE_CR0:
5028 case SVM_EXIT_WRITE_CR3:
5029 case SVM_EXIT_WRITE_CR4:
5030 case SVM_EXIT_WRITE_CR8: /* CR8 writes would go to the V_TPR rather than here, since we run with V_INTR_MASKING. */
5031 {
5032 uint8_t const uCr = uExitCode - SVM_EXIT_WRITE_CR0;
5033 Log4Func(("Write CR%u: uExitInfo1=%#RX64 uExitInfo2=%#RX64\n", uCr, uExitInfo1, uExitInfo2));
5034
5035 if (CPUMIsGuestSvmWriteCRxInterceptSet(pVCpu, pCtx, uCr))
5036 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5037 return hmR0SvmExitWriteCRx(pVCpu, pSvmTransient);
5038 }
5039
5040 case SVM_EXIT_PAUSE:
5041 {
5042 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_PAUSE))
5043 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5044 return hmR0SvmExitPause(pVCpu, pSvmTransient);
5045 }
5046
5047 case SVM_EXIT_VINTR:
5048 {
5049 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_VINTR))
5050 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5051 return hmR0SvmExitUnexpected(pVCpu, pSvmTransient);
5052 }
5053
5054 case SVM_EXIT_INTR:
5055 case SVM_EXIT_NMI:
5056 case SVM_EXIT_SMI:
5057 case SVM_EXIT_XCPT_NMI: /* Should not occur, SVM_EXIT_NMI is used instead. */
5058 {
5059 /*
5060 * We shouldn't direct physical interrupts, NMIs, SMIs to the nested-guest.
5061 *
5062 * Although we don't intercept SMIs, the nested-guest might. Therefore, we might
5063 * get an SMI #VMEXIT here so simply ignore rather than causing a corresponding
5064 * nested-guest #VMEXIT.
5065 *
5066 * We shall import the complete state here as we may cause #VMEXITs from ring-3
5067 * while trying to inject interrupts, see comment at the top of this function.
5068 */
5069 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, CPUMCTX_EXTRN_ALL);
5070 return hmR0SvmExitIntr(pVCpu, pSvmTransient);
5071 }
5072
5073 case SVM_EXIT_FERR_FREEZE:
5074 {
5075 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_FERR_FREEZE))
5076 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5077 return hmR0SvmExitFerrFreeze(pVCpu, pSvmTransient);
5078 }
5079
5080 case SVM_EXIT_INVLPG:
5081 {
5082 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_INVLPG))
5083 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5084 return hmR0SvmExitInvlpg(pVCpu, pSvmTransient);
5085 }
5086
5087 case SVM_EXIT_WBINVD:
5088 {
5089 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_WBINVD))
5090 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5091 return hmR0SvmExitWbinvd(pVCpu, pSvmTransient);
5092 }
5093
5094 case SVM_EXIT_INVD:
5095 {
5096 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_INVD))
5097 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5098 return hmR0SvmExitInvd(pVCpu, pSvmTransient);
5099 }
5100
5101 case SVM_EXIT_RDPMC:
5102 {
5103 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_RDPMC))
5104 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5105 return hmR0SvmExitRdpmc(pVCpu, pSvmTransient);
5106 }
5107
5108 default:
5109 {
5110 switch (uExitCode)
5111 {
5112 case SVM_EXIT_READ_DR0: case SVM_EXIT_READ_DR1: case SVM_EXIT_READ_DR2: case SVM_EXIT_READ_DR3:
5113 case SVM_EXIT_READ_DR6: case SVM_EXIT_READ_DR7: case SVM_EXIT_READ_DR8: case SVM_EXIT_READ_DR9:
5114 case SVM_EXIT_READ_DR10: case SVM_EXIT_READ_DR11: case SVM_EXIT_READ_DR12: case SVM_EXIT_READ_DR13:
5115 case SVM_EXIT_READ_DR14: case SVM_EXIT_READ_DR15:
5116 {
5117 uint8_t const uDr = uExitCode - SVM_EXIT_READ_DR0;
5118 if (CPUMIsGuestSvmReadDRxInterceptSet(pVCpu, pCtx, uDr))
5119 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5120 return hmR0SvmExitReadDRx(pVCpu, pSvmTransient);
5121 }
5122
5123 case SVM_EXIT_WRITE_DR0: case SVM_EXIT_WRITE_DR1: case SVM_EXIT_WRITE_DR2: case SVM_EXIT_WRITE_DR3:
5124 case SVM_EXIT_WRITE_DR6: case SVM_EXIT_WRITE_DR7: case SVM_EXIT_WRITE_DR8: case SVM_EXIT_WRITE_DR9:
5125 case SVM_EXIT_WRITE_DR10: case SVM_EXIT_WRITE_DR11: case SVM_EXIT_WRITE_DR12: case SVM_EXIT_WRITE_DR13:
5126 case SVM_EXIT_WRITE_DR14: case SVM_EXIT_WRITE_DR15:
5127 {
5128 uint8_t const uDr = uExitCode - SVM_EXIT_WRITE_DR0;
5129 if (CPUMIsGuestSvmWriteDRxInterceptSet(pVCpu, pCtx, uDr))
5130 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5131 return hmR0SvmExitWriteDRx(pVCpu, pSvmTransient);
5132 }
5133
5134 case SVM_EXIT_XCPT_DE:
5135 /* SVM_EXIT_XCPT_DB: */ /* Handled above. */
5136 /* SVM_EXIT_XCPT_NMI: */ /* Handled above. */
5137 /* SVM_EXIT_XCPT_BP: */ /* Handled above. */
5138 case SVM_EXIT_XCPT_OF:
5139 case SVM_EXIT_XCPT_BR:
5140 /* SVM_EXIT_XCPT_UD: */ /* Handled above. */
5141 case SVM_EXIT_XCPT_NM:
5142 case SVM_EXIT_XCPT_DF:
5143 case SVM_EXIT_XCPT_CO_SEG_OVERRUN:
5144 case SVM_EXIT_XCPT_TS:
5145 case SVM_EXIT_XCPT_NP:
5146 case SVM_EXIT_XCPT_SS:
5147 case SVM_EXIT_XCPT_GP:
5148 /* SVM_EXIT_XCPT_PF: */ /* Handled above. */
5149 case SVM_EXIT_XCPT_15: /* Reserved. */
5150 /* SVM_EXIT_XCPT_MF: */ /* Handled above. */
5151 /* SVM_EXIT_XCPT_AC: */ /* Handled above. */
5152 case SVM_EXIT_XCPT_MC:
5153 case SVM_EXIT_XCPT_XF:
5154 case SVM_EXIT_XCPT_20: case SVM_EXIT_XCPT_21: case SVM_EXIT_XCPT_22: case SVM_EXIT_XCPT_23:
5155 case SVM_EXIT_XCPT_24: case SVM_EXIT_XCPT_25: case SVM_EXIT_XCPT_26: case SVM_EXIT_XCPT_27:
5156 case SVM_EXIT_XCPT_28: case SVM_EXIT_XCPT_29: case SVM_EXIT_XCPT_30: case SVM_EXIT_XCPT_31:
5157 {
5158 uint8_t const uVector = uExitCode - SVM_EXIT_XCPT_0;
5159 if (CPUMIsGuestSvmXcptInterceptSet(pVCpu, pCtx, uVector))
5160 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5161 return hmR0SvmExitXcptGeneric(pVCpu, pSvmTransient);
5162 }
5163
5164 case SVM_EXIT_XSETBV:
5165 {
5166 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_XSETBV))
5167 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5168 return hmR0SvmExitXsetbv(pVCpu, pSvmTransient);
5169 }
5170
5171 case SVM_EXIT_TASK_SWITCH:
5172 {
5173 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_TASK_SWITCH))
5174 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5175 return hmR0SvmExitTaskSwitch(pVCpu, pSvmTransient);
5176 }
5177
5178 case SVM_EXIT_IRET:
5179 {
5180 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_IRET))
5181 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5182 return hmR0SvmExitIret(pVCpu, pSvmTransient);
5183 }
5184
5185 case SVM_EXIT_SHUTDOWN:
5186 {
5187 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_SHUTDOWN))
5188 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5189 return hmR0SvmExitShutdown(pVCpu, pSvmTransient);
5190 }
5191
5192 case SVM_EXIT_VMMCALL:
5193 {
5194 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_VMMCALL))
5195 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5196 return hmR0SvmExitVmmCall(pVCpu, pSvmTransient);
5197 }
5198
5199 case SVM_EXIT_CLGI:
5200 {
5201 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_CLGI))
5202 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5203 return hmR0SvmExitClgi(pVCpu, pSvmTransient);
5204 }
5205
5206 case SVM_EXIT_STGI:
5207 {
5208 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_STGI))
5209 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5210 return hmR0SvmExitStgi(pVCpu, pSvmTransient);
5211 }
5212
5213 case SVM_EXIT_VMLOAD:
5214 {
5215 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_VMLOAD))
5216 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5217 return hmR0SvmExitVmload(pVCpu, pSvmTransient);
5218 }
5219
5220 case SVM_EXIT_VMSAVE:
5221 {
5222 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_VMSAVE))
5223 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5224 return hmR0SvmExitVmsave(pVCpu, pSvmTransient);
5225 }
5226
5227 case SVM_EXIT_INVLPGA:
5228 {
5229 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_INVLPGA))
5230 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5231 return hmR0SvmExitInvlpga(pVCpu, pSvmTransient);
5232 }
5233
5234 case SVM_EXIT_VMRUN:
5235 {
5236 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_VMRUN))
5237 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5238 return hmR0SvmExitVmrun(pVCpu, pSvmTransient);
5239 }
5240
5241 case SVM_EXIT_RSM:
5242 {
5243 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_RSM))
5244 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5245 hmR0SvmSetPendingXcptUD(pVCpu);
5246 return VINF_SUCCESS;
5247 }
5248
5249 case SVM_EXIT_SKINIT:
5250 {
5251 if (CPUMIsGuestSvmCtrlInterceptSet(pVCpu, pCtx, SVM_CTRL_INTERCEPT_SKINIT))
5252 NST_GST_VMEXIT_CALL_RET(pVCpu, uExitCode, uExitInfo1, uExitInfo2);
5253 hmR0SvmSetPendingXcptUD(pVCpu);
5254 return VINF_SUCCESS;
5255 }
5256
5257 case SVM_EXIT_NPF:
5258 {
5259 Assert(pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging);
5260 return hmR0SvmExitNestedPF(pVCpu, pSvmTransient);
5261 }
5262
5263 case SVM_EXIT_INIT: /* We shouldn't get INIT signals while executing a nested-guest. */
5264 return hmR0SvmExitUnexpected(pVCpu, pSvmTransient);
5265
5266 default:
5267 {
5268 AssertMsgFailed(("hmR0SvmHandleExitNested: Unknown exit code %#x\n", pSvmTransient->u64ExitCode));
5269 pVCpu->hm.s.u32HMError = pSvmTransient->u64ExitCode;
5270 return VERR_SVM_UNKNOWN_EXIT;
5271 }
5272 }
5273 }
5274 }
5275 /* not reached */
5276
5277# undef NST_GST_VMEXIT_CALL_RET
5278}
5279
5280#endif /* VBOX_WITH_NESTED_HWVIRT_SVM */
5281
5282/** @def VMEXIT_CALL_RET
5283 * Used by hmR0SvmHandleExit and hmR0SvmDebugHandleExit
5284 */
5285#ifdef DEBUG_ramshankar
5286# define VMEXIT_CALL_RET(a_fDbg, a_CallExpr) \
5287 do { \
5288 if ((a_fDbg) == 1) \
5289 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, HMSVM_CPUMCTX_EXTRN_ALL); \
5290 int rc = a_CallExpr; \
5291 if ((a_fDbg) == 1) \
5292 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST); \
5293 return rc; \
5294 } while (0)
5295#else
5296# define VMEXIT_CALL_RET(a_fDbg, a_CallExpr) return a_CallExpr
5297#endif
5298
5299/**
5300 * Handles a guest \#VMEXIT (for all EXITCODE values except SVM_EXIT_INVALID).
5301 *
5302 * @returns Strict VBox status code (informational status codes included).
5303 * @param pVCpu The cross context virtual CPU structure.
5304 * @param pSvmTransient Pointer to the SVM transient structure.
5305 */
5306static VBOXSTRICTRC hmR0SvmHandleExit(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
5307{
5308 Assert(pSvmTransient->u64ExitCode != SVM_EXIT_INVALID);
5309 Assert(pSvmTransient->u64ExitCode <= SVM_EXIT_MAX);
5310
5311 /*
5312 * The ordering of the case labels is based on most-frequently-occurring #VMEXITs
5313 * for most guests under normal workloads (for some definition of "normal").
5314 */
5315 uint64_t const uExitCode = pSvmTransient->u64ExitCode;
5316 switch (uExitCode)
5317 {
5318 case SVM_EXIT_NPF: VMEXIT_CALL_RET(0, hmR0SvmExitNestedPF(pVCpu, pSvmTransient));
5319 case SVM_EXIT_IOIO: VMEXIT_CALL_RET(0, hmR0SvmExitIOInstr(pVCpu, pSvmTransient));
5320 case SVM_EXIT_RDTSC: VMEXIT_CALL_RET(0, hmR0SvmExitRdtsc(pVCpu, pSvmTransient));
5321 case SVM_EXIT_RDTSCP: VMEXIT_CALL_RET(0, hmR0SvmExitRdtscp(pVCpu, pSvmTransient));
5322 case SVM_EXIT_CPUID: VMEXIT_CALL_RET(0, hmR0SvmExitCpuid(pVCpu, pSvmTransient));
5323 case SVM_EXIT_XCPT_PF: VMEXIT_CALL_RET(0, hmR0SvmExitXcptPF(pVCpu, pSvmTransient));
5324 case SVM_EXIT_MSR: VMEXIT_CALL_RET(0, hmR0SvmExitMsr(pVCpu, pSvmTransient));
5325 case SVM_EXIT_MONITOR: VMEXIT_CALL_RET(0, hmR0SvmExitMonitor(pVCpu, pSvmTransient));
5326 case SVM_EXIT_MWAIT: VMEXIT_CALL_RET(0, hmR0SvmExitMwait(pVCpu, pSvmTransient));
5327 case SVM_EXIT_HLT: VMEXIT_CALL_RET(0, hmR0SvmExitHlt(pVCpu, pSvmTransient));
5328
5329 case SVM_EXIT_XCPT_NMI: /* Should not occur, SVM_EXIT_NMI is used instead. */
5330 case SVM_EXIT_INTR:
5331 case SVM_EXIT_NMI: VMEXIT_CALL_RET(0, hmR0SvmExitIntr(pVCpu, pSvmTransient));
5332
5333 case SVM_EXIT_READ_CR0:
5334 case SVM_EXIT_READ_CR3:
5335 case SVM_EXIT_READ_CR4: VMEXIT_CALL_RET(0, hmR0SvmExitReadCRx(pVCpu, pSvmTransient));
5336
5337 case SVM_EXIT_CR0_SEL_WRITE:
5338 case SVM_EXIT_WRITE_CR0:
5339 case SVM_EXIT_WRITE_CR3:
5340 case SVM_EXIT_WRITE_CR4:
5341 case SVM_EXIT_WRITE_CR8: VMEXIT_CALL_RET(0, hmR0SvmExitWriteCRx(pVCpu, pSvmTransient));
5342
5343 case SVM_EXIT_VINTR: VMEXIT_CALL_RET(0, hmR0SvmExitVIntr(pVCpu, pSvmTransient));
5344 case SVM_EXIT_PAUSE: VMEXIT_CALL_RET(0, hmR0SvmExitPause(pVCpu, pSvmTransient));
5345 case SVM_EXIT_VMMCALL: VMEXIT_CALL_RET(0, hmR0SvmExitVmmCall(pVCpu, pSvmTransient));
5346 case SVM_EXIT_INVLPG: VMEXIT_CALL_RET(0, hmR0SvmExitInvlpg(pVCpu, pSvmTransient));
5347 case SVM_EXIT_WBINVD: VMEXIT_CALL_RET(0, hmR0SvmExitWbinvd(pVCpu, pSvmTransient));
5348 case SVM_EXIT_INVD: VMEXIT_CALL_RET(0, hmR0SvmExitInvd(pVCpu, pSvmTransient));
5349 case SVM_EXIT_RDPMC: VMEXIT_CALL_RET(0, hmR0SvmExitRdpmc(pVCpu, pSvmTransient));
5350 case SVM_EXIT_IRET: VMEXIT_CALL_RET(0, hmR0SvmExitIret(pVCpu, pSvmTransient));
5351 case SVM_EXIT_XCPT_DE: VMEXIT_CALL_RET(0, hmR0SvmExitXcptDE(pVCpu, pSvmTransient));
5352 case SVM_EXIT_XCPT_UD: VMEXIT_CALL_RET(0, hmR0SvmExitXcptUD(pVCpu, pSvmTransient));
5353 case SVM_EXIT_XCPT_MF: VMEXIT_CALL_RET(0, hmR0SvmExitXcptMF(pVCpu, pSvmTransient));
5354 case SVM_EXIT_XCPT_DB: VMEXIT_CALL_RET(0, hmR0SvmExitXcptDB(pVCpu, pSvmTransient));
5355 case SVM_EXIT_XCPT_AC: VMEXIT_CALL_RET(0, hmR0SvmExitXcptAC(pVCpu, pSvmTransient));
5356 case SVM_EXIT_XCPT_BP: VMEXIT_CALL_RET(0, hmR0SvmExitXcptBP(pVCpu, pSvmTransient));
5357 case SVM_EXIT_XCPT_GP: VMEXIT_CALL_RET(0, hmR0SvmExitXcptGP(pVCpu, pSvmTransient));
5358 case SVM_EXIT_XSETBV: VMEXIT_CALL_RET(0, hmR0SvmExitXsetbv(pVCpu, pSvmTransient));
5359 case SVM_EXIT_FERR_FREEZE: VMEXIT_CALL_RET(0, hmR0SvmExitFerrFreeze(pVCpu, pSvmTransient));
5360
5361 default:
5362 {
5363 switch (pSvmTransient->u64ExitCode)
5364 {
5365 case SVM_EXIT_READ_DR0: case SVM_EXIT_READ_DR1: case SVM_EXIT_READ_DR2: case SVM_EXIT_READ_DR3:
5366 case SVM_EXIT_READ_DR6: case SVM_EXIT_READ_DR7: case SVM_EXIT_READ_DR8: case SVM_EXIT_READ_DR9:
5367 case SVM_EXIT_READ_DR10: case SVM_EXIT_READ_DR11: case SVM_EXIT_READ_DR12: case SVM_EXIT_READ_DR13:
5368 case SVM_EXIT_READ_DR14: case SVM_EXIT_READ_DR15:
5369 VMEXIT_CALL_RET(0, hmR0SvmExitReadDRx(pVCpu, pSvmTransient));
5370
5371 case SVM_EXIT_WRITE_DR0: case SVM_EXIT_WRITE_DR1: case SVM_EXIT_WRITE_DR2: case SVM_EXIT_WRITE_DR3:
5372 case SVM_EXIT_WRITE_DR6: case SVM_EXIT_WRITE_DR7: case SVM_EXIT_WRITE_DR8: case SVM_EXIT_WRITE_DR9:
5373 case SVM_EXIT_WRITE_DR10: case SVM_EXIT_WRITE_DR11: case SVM_EXIT_WRITE_DR12: case SVM_EXIT_WRITE_DR13:
5374 case SVM_EXIT_WRITE_DR14: case SVM_EXIT_WRITE_DR15:
5375 VMEXIT_CALL_RET(0, hmR0SvmExitWriteDRx(pVCpu, pSvmTransient));
5376
5377 case SVM_EXIT_TASK_SWITCH: VMEXIT_CALL_RET(0, hmR0SvmExitTaskSwitch(pVCpu, pSvmTransient));
5378 case SVM_EXIT_SHUTDOWN: VMEXIT_CALL_RET(0, hmR0SvmExitShutdown(pVCpu, pSvmTransient));
5379
5380 case SVM_EXIT_SMI:
5381 case SVM_EXIT_INIT:
5382 {
5383 /*
5384 * We don't intercept SMIs. As for INIT signals, it really shouldn't ever happen here.
5385 * If it ever does, we want to know about it so log the exit code and bail.
5386 */
5387 VMEXIT_CALL_RET(0, hmR0SvmExitUnexpected(pVCpu, pSvmTransient));
5388 }
5389
5390#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
5391 case SVM_EXIT_CLGI: VMEXIT_CALL_RET(0, hmR0SvmExitClgi(pVCpu, pSvmTransient));
5392 case SVM_EXIT_STGI: VMEXIT_CALL_RET(0, hmR0SvmExitStgi(pVCpu, pSvmTransient));
5393 case SVM_EXIT_VMLOAD: VMEXIT_CALL_RET(0, hmR0SvmExitVmload(pVCpu, pSvmTransient));
5394 case SVM_EXIT_VMSAVE: VMEXIT_CALL_RET(0, hmR0SvmExitVmsave(pVCpu, pSvmTransient));
5395 case SVM_EXIT_INVLPGA: VMEXIT_CALL_RET(0, hmR0SvmExitInvlpga(pVCpu, pSvmTransient));
5396 case SVM_EXIT_VMRUN: VMEXIT_CALL_RET(0, hmR0SvmExitVmrun(pVCpu, pSvmTransient));
5397#else
5398 case SVM_EXIT_CLGI:
5399 case SVM_EXIT_STGI:
5400 case SVM_EXIT_VMLOAD:
5401 case SVM_EXIT_VMSAVE:
5402 case SVM_EXIT_INVLPGA:
5403 case SVM_EXIT_VMRUN:
5404#endif
5405 case SVM_EXIT_RSM:
5406 case SVM_EXIT_SKINIT:
5407 {
5408 hmR0SvmSetPendingXcptUD(pVCpu);
5409 return VINF_SUCCESS;
5410 }
5411
5412 /*
5413 * The remaining should only be possible when debugging or dtracing.
5414 */
5415 case SVM_EXIT_XCPT_DE:
5416 /* SVM_EXIT_XCPT_DB: */ /* Handled above. */
5417 /* SVM_EXIT_XCPT_NMI: */ /* Handled above. */
5418 /* SVM_EXIT_XCPT_BP: */ /* Handled above. */
5419 case SVM_EXIT_XCPT_OF:
5420 case SVM_EXIT_XCPT_BR:
5421 /* SVM_EXIT_XCPT_UD: */ /* Handled above. */
5422 case SVM_EXIT_XCPT_NM:
5423 case SVM_EXIT_XCPT_DF:
5424 case SVM_EXIT_XCPT_CO_SEG_OVERRUN:
5425 case SVM_EXIT_XCPT_TS:
5426 case SVM_EXIT_XCPT_NP:
5427 case SVM_EXIT_XCPT_SS:
5428 /* SVM_EXIT_XCPT_GP: */ /* Handled above. */
5429 /* SVM_EXIT_XCPT_PF: */
5430 case SVM_EXIT_XCPT_15: /* Reserved. */
5431 /* SVM_EXIT_XCPT_MF: */ /* Handled above. */
5432 /* SVM_EXIT_XCPT_AC: */ /* Handled above. */
5433 case SVM_EXIT_XCPT_MC:
5434 case SVM_EXIT_XCPT_XF:
5435 case SVM_EXIT_XCPT_20: case SVM_EXIT_XCPT_21: case SVM_EXIT_XCPT_22: case SVM_EXIT_XCPT_23:
5436 case SVM_EXIT_XCPT_24: case SVM_EXIT_XCPT_25: case SVM_EXIT_XCPT_26: case SVM_EXIT_XCPT_27:
5437 case SVM_EXIT_XCPT_28: case SVM_EXIT_XCPT_29: case SVM_EXIT_XCPT_30: case SVM_EXIT_XCPT_31:
5438 VMEXIT_CALL_RET(0, hmR0SvmExitXcptGeneric(pVCpu, pSvmTransient));
5439
5440 case SVM_EXIT_SWINT: VMEXIT_CALL_RET(0, hmR0SvmExitSwInt(pVCpu, pSvmTransient));
5441 case SVM_EXIT_TR_READ: VMEXIT_CALL_RET(0, hmR0SvmExitTrRead(pVCpu, pSvmTransient));
5442 case SVM_EXIT_TR_WRITE: VMEXIT_CALL_RET(0, hmR0SvmExitTrWrite(pVCpu, pSvmTransient)); /* Also OS/2 TLB workaround. */
5443
5444 default:
5445 {
5446 AssertMsgFailed(("hmR0SvmHandleExit: Unknown exit code %#RX64\n", uExitCode));
5447 pVCpu->hm.s.u32HMError = uExitCode;
5448 return VERR_SVM_UNKNOWN_EXIT;
5449 }
5450 }
5451 }
5452 }
5453 /* not reached */
5454}
5455
5456
5457/** @name Execution loop for single stepping, DBGF events and expensive Dtrace probes.
5458 *
5459 * The following few functions and associated structure contains the bloat
5460 * necessary for providing detailed debug events and dtrace probes as well as
5461 * reliable host side single stepping. This works on the principle of
5462 * "subclassing" the normal execution loop and workers. We replace the loop
5463 * method completely and override selected helpers to add necessary adjustments
5464 * to their core operation.
5465 *
5466 * The goal is to keep the "parent" code lean and mean, so as not to sacrifice
5467 * any performance for debug and analysis features.
5468 *
5469 * @{
5470 */
5471
5472/**
5473 * Transient per-VCPU debug state of VMCS and related info. we save/restore in
5474 * the debug run loop.
5475 */
5476typedef struct SVMRUNDBGSTATE
5477{
5478 /** The initial SVMVMCBCTRL::u64InterceptCtrl value (helps with restore). */
5479 uint64_t bmInterceptInitial;
5480 /** The initial SVMVMCBCTRL::u32InterceptXcpt value (helps with restore). */
5481 uint32_t bmXcptInitial;
5482 /** The initial SVMVMCBCTRL::u16InterceptRdCRx value (helps with restore). */
5483 uint16_t bmInterceptRdCRxInitial;
5484 /** The initial SVMVMCBCTRL::u16InterceptWrCRx value (helps with restore). */
5485 uint16_t bmInterceptWrCRxInitial;
5486 /** The initial SVMVMCBCTRL::u16InterceptRdDRx value (helps with restore). */
5487 uint16_t bmInterceptRdDRxInitial;
5488 /** The initial SVMVMCBCTRL::u16InterceptWrDRx value (helps with restore). */
5489 uint16_t bmInterceptWrDRxInitial;
5490
5491 /** Whether we've actually modified the intercept control qword. */
5492 bool fModifiedInterceptCtrl : 1;
5493 /** Whether we've actually modified the exception bitmap. */
5494 bool fModifiedXcptBitmap : 1;
5495 /** Whether we've actually modified SVMVMCBCTRL::u16InterceptRdCRx. */
5496 bool fModifiedInterceptRdCRx : 1;
5497 /** Whether we've actually modified SVMVMCBCTRL::u16InterceptWrCRx. */
5498 bool fModifiedInterceptWrCRx : 1;
5499 /** Whether we've actually modified SVMVMCBCTRL::u16InterceptRdDRx. */
5500 bool fModifiedInterceptRdDRx : 1;
5501 /** Whether we've actually modified SVMVMCBCTRL::u16InterceptWrDRx. */
5502 bool fModifiedInterceptWrDRx : 1;
5503
5504 /** The CS we started executing with. */
5505 uint16_t uCsStart;
5506 /** The RIP we started executing at. This is for detecting that we stepped. */
5507 uint64_t uRipStart;
5508
5509 /** The sequence number of the Dtrace provider settings the state was
5510 * configured against. */
5511 uint32_t uDtraceSettingsSeqNo;
5512 /** Extra stuff we need in SVMVMCBCTRL::u32InterceptXcpt. */
5513 uint32_t bmXcptExtra;
5514 /** Extra stuff we need in SVMVMCBCTRL::u64InterceptCtrl. */
5515 uint64_t bmInterceptExtra;
5516 /** Extra stuff we need in SVMVMCBCTRL::u16InterceptRdCRx. */
5517 uint16_t bmInterceptRdCRxExtra;
5518 /** Extra stuff we need in SVMVMCBCTRL::u16InterceptWrCRx. */
5519 uint16_t bmInterceptWrCRxExtra;
5520 /** Extra stuff we need in SVMVMCBCTRL::u16InterceptRdDRx. */
5521 uint16_t bmInterceptRdDRxExtra;
5522 /** Extra stuff we need in SVMVMCBCTRL::u16InterceptWrDRx. */
5523 uint16_t bmInterceptWrDRxExtra;
5524 /** VM-exits to check (one bit per VM-exit). */
5525 uint32_t bmExitsToCheck[33];
5526} SVMRUNDBGSTATE;
5527AssertCompileMemberSize(SVMRUNDBGSTATE, bmExitsToCheck, (SVM_EXIT_MAX + 1 + 31) / 32 * 4);
5528typedef SVMRUNDBGSTATE *PSVMRUNDBGSTATE;
5529
5530
5531/**
5532 * Initializes the SVMRUNDBGSTATE structure.
5533 *
5534 * @param pVCpu The cross context virtual CPU structure of the
5535 * calling EMT.
5536 * @param pSvmTransient The SVM-transient structure.
5537 * @param pDbgState The debug state to initialize.
5538 */
5539static void hmR0SvmRunDebugStateInit(PVMCPUCC pVCpu, PCSVMTRANSIENT pSvmTransient, PSVMRUNDBGSTATE pDbgState)
5540{
5541 PSVMVMCB pVmcb = pSvmTransient->pVmcb;
5542 pDbgState->bmInterceptInitial = pVmcb->ctrl.u64InterceptCtrl;
5543 pDbgState->bmXcptInitial = pVmcb->ctrl.u32InterceptXcpt;
5544 pDbgState->bmInterceptRdCRxInitial = pVmcb->ctrl.u16InterceptRdCRx;
5545 pDbgState->bmInterceptWrCRxInitial = pVmcb->ctrl.u16InterceptWrCRx;
5546 pDbgState->bmInterceptRdDRxInitial = pVmcb->ctrl.u16InterceptRdDRx;
5547 pDbgState->bmInterceptWrDRxInitial = pVmcb->ctrl.u16InterceptWrDRx;
5548
5549 pDbgState->fModifiedInterceptCtrl = false;
5550 pDbgState->fModifiedXcptBitmap = false;
5551 pDbgState->fModifiedInterceptRdCRx = false;
5552 pDbgState->fModifiedInterceptWrCRx = false;
5553 pDbgState->fModifiedInterceptRdDRx = false;
5554 pDbgState->fModifiedInterceptWrDRx = false;
5555
5556 pDbgState->uCsStart = pVCpu->cpum.GstCtx.cs.Sel;
5557 pDbgState->uRipStart = pVCpu->cpum.GstCtx.rip;
5558
5559 /* We don't really need to zero these. */
5560 pDbgState->bmInterceptExtra = 0;
5561 pDbgState->bmXcptExtra = 0;
5562 pDbgState->bmInterceptRdCRxExtra = 0;
5563 pDbgState->bmInterceptWrCRxExtra = 0;
5564 pDbgState->bmInterceptRdDRxExtra = 0;
5565 pDbgState->bmInterceptWrDRxExtra = 0;
5566}
5567
5568
5569/**
5570 * Updates the VMCB fields with changes requested by @a pDbgState.
5571 *
5572 * This is performed after hmR0SvmPreRunGuestDebugStateUpdate as well
5573 * immediately before executing guest code, i.e. when interrupts are disabled.
5574 * We don't check status codes here as we cannot easily assert or return in the
5575 * latter case.
5576 *
5577 * @param pSvmTransient The SVM-transient structure.
5578 * @param pDbgState The debug state.
5579 */
5580static void hmR0SvmPreRunGuestDebugStateApply(PSVMTRANSIENT pSvmTransient, PSVMRUNDBGSTATE pDbgState)
5581{
5582 /*
5583 * Ensure desired flags in VMCS control fields are set.
5584 */
5585 PSVMVMCB const pVmcb = pSvmTransient->pVmcb;
5586#define ADD_EXTRA_INTERCEPTS(a_VmcbCtrlField, a_bmExtra, a_fModified) do { \
5587 if ((pVmcb->ctrl. a_VmcbCtrlField & (a_bmExtra)) != (a_bmExtra)) \
5588 { \
5589 pVmcb->ctrl. a_VmcbCtrlField |= (a_bmExtra); \
5590 pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_INTERCEPTS; \
5591 Log6Func((#a_VmcbCtrlField ": %#RX64\n", pVmcb->ctrl. a_VmcbCtrlField)); \
5592 (a_fModified) = true; \
5593 } \
5594 } while (0)
5595 ADD_EXTRA_INTERCEPTS(u64InterceptCtrl, pDbgState->bmInterceptExtra, pDbgState->fModifiedInterceptCtrl);
5596 ADD_EXTRA_INTERCEPTS(u32InterceptXcpt, pDbgState->bmXcptExtra, pDbgState->fModifiedXcptBitmap);
5597 ADD_EXTRA_INTERCEPTS(u16InterceptRdCRx, pDbgState->bmInterceptRdCRxExtra, pDbgState->fModifiedInterceptRdCRx);
5598 ADD_EXTRA_INTERCEPTS(u16InterceptWrCRx, pDbgState->bmInterceptWrCRxExtra, pDbgState->fModifiedInterceptWrCRx);
5599 ADD_EXTRA_INTERCEPTS(u16InterceptRdDRx, pDbgState->bmInterceptRdDRxExtra, pDbgState->fModifiedInterceptRdDRx);
5600 ADD_EXTRA_INTERCEPTS(u16InterceptWrDRx, pDbgState->bmInterceptWrDRxExtra, pDbgState->fModifiedInterceptWrDRx);
5601#undef ADD_EXTRA_INTERCEPTS
5602}
5603
5604
5605/**
5606 * Restores VMCB fields that were changed by hmR0SvmPreRunGuestDebugStateApply
5607 * for re-entry next time around.
5608 *
5609 * @returns Strict VBox status code (i.e. informational status codes too).
5610 * @param pSvmTransient The SVM-transient structure.
5611 * @param pDbgState The debug state.
5612 */
5613static void hmR0SvmRunDebugStateRevert(PSVMTRANSIENT pSvmTransient, PSVMRUNDBGSTATE pDbgState)
5614{
5615 /*
5616 * Restore VM-exit control settings as we may not reenter this function the
5617 * next time around.
5618 */
5619 PSVMVMCB const pVmcb = pSvmTransient->pVmcb;
5620
5621#define RESTORE_INTERCEPTS(a_VmcbCtrlField, a_bmInitial, a_fModified) do { \
5622 if ((a_fModified)) \
5623 { \
5624 pVmcb->ctrl. a_VmcbCtrlField = (a_bmInitial); \
5625 pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_INTERCEPTS; \
5626 } \
5627 } while (0)
5628 RESTORE_INTERCEPTS(u64InterceptCtrl, pDbgState->bmInterceptInitial, pDbgState->fModifiedInterceptCtrl);
5629 RESTORE_INTERCEPTS(u32InterceptXcpt, pDbgState->bmXcptInitial, pDbgState->fModifiedXcptBitmap);
5630 RESTORE_INTERCEPTS(u16InterceptRdCRx, pDbgState->bmInterceptRdCRxInitial, pDbgState->fModifiedInterceptRdCRx);
5631 RESTORE_INTERCEPTS(u16InterceptWrCRx, pDbgState->bmInterceptWrCRxInitial, pDbgState->fModifiedInterceptWrCRx);
5632 RESTORE_INTERCEPTS(u16InterceptRdDRx, pDbgState->bmInterceptRdDRxInitial, pDbgState->fModifiedInterceptRdDRx);
5633 RESTORE_INTERCEPTS(u16InterceptWrDRx, pDbgState->bmInterceptWrDRxInitial, pDbgState->fModifiedInterceptWrDRx);
5634#undef RESTORE_INTERCEPTS
5635}
5636
5637
5638/**
5639 * Configures VM-exit controls for current DBGF and DTrace settings.
5640 *
5641 * This updates @a pDbgState and the VMCB execution control fields (in the debug
5642 * state) to reflect the necessary VM-exits demanded by DBGF and DTrace.
5643 *
5644 * @param pVCpu The cross context virtual CPU structure.
5645 * @param pSvmTransient The SVM-transient structure. May update
5646 * fUpdatedTscOffsettingAndPreemptTimer.
5647 * @param pDbgState The debug state.
5648 */
5649static void hmR0SvmPreRunGuestDebugStateUpdate(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient, PSVMRUNDBGSTATE pDbgState)
5650{
5651 /*
5652 * Take down the dtrace serial number so we can spot changes.
5653 */
5654 pDbgState->uDtraceSettingsSeqNo = VBOXVMM_GET_SETTINGS_SEQ_NO();
5655 ASMCompilerBarrier();
5656
5657 /*
5658 * Clear data members that we'll be rebuilding here.
5659 */
5660 pDbgState->bmXcptExtra = 0;
5661 pDbgState->bmInterceptExtra = 0;
5662 pDbgState->bmInterceptRdCRxExtra = 0;
5663 pDbgState->bmInterceptWrCRxExtra = 0;
5664 pDbgState->bmInterceptRdDRxExtra = 0;
5665 pDbgState->bmInterceptWrDRxExtra = 0;
5666 for (unsigned i = 0; i < RT_ELEMENTS(pDbgState->bmExitsToCheck); i++)
5667 pDbgState->bmExitsToCheck[i] = 0;
5668
5669 /*
5670 * Software interrupts (INT XXh)
5671 */
5672 PVMCC pVM = pVCpu->CTX_SUFF(pVM);
5673 if ( DBGF_IS_EVENT_ENABLED(pVM, DBGFEVENT_INTERRUPT_SOFTWARE)
5674 || VBOXVMM_INT_SOFTWARE_ENABLED())
5675 {
5676 pDbgState->bmInterceptExtra |= SVM_CTRL_INTERCEPT_INTN;
5677 ASMBitSet(pDbgState->bmExitsToCheck, SVM_EXIT_SWINT);
5678 }
5679
5680 /*
5681 * INT3 breakpoints - triggered by #BP exceptions.
5682 */
5683 if (pVM->dbgf.ro.cEnabledInt3Breakpoints > 0)
5684 pDbgState->bmXcptExtra |= RT_BIT_32(X86_XCPT_BP);
5685
5686 /*
5687 * Exception bitmap and XCPT events+probes.
5688 */
5689#define SET_XCPT(a_iXcpt) do { \
5690 pDbgState->bmXcptExtra |= RT_BIT_32(a_iXcpt); \
5691 ASMBitSet(pDbgState->bmExitsToCheck, SVM_EXIT_XCPT_0 + (a_iXcpt)); \
5692 } while (0)
5693
5694 for (int iXcpt = 0; iXcpt < (DBGFEVENT_XCPT_LAST - DBGFEVENT_XCPT_FIRST + 1); iXcpt++)
5695 if (DBGF_IS_EVENT_ENABLED(pVM, (DBGFEVENTTYPE)(DBGFEVENT_XCPT_FIRST + iXcpt)))
5696 SET_XCPT(iXcpt);
5697
5698 if (VBOXVMM_XCPT_DE_ENABLED()) SET_XCPT(X86_XCPT_DE);
5699 if (VBOXVMM_XCPT_DB_ENABLED()) SET_XCPT(X86_XCPT_DB);
5700 if (VBOXVMM_XCPT_BP_ENABLED()) SET_XCPT(X86_XCPT_BP);
5701 if (VBOXVMM_XCPT_OF_ENABLED()) SET_XCPT(X86_XCPT_OF);
5702 if (VBOXVMM_XCPT_BR_ENABLED()) SET_XCPT(X86_XCPT_BR);
5703 if (VBOXVMM_XCPT_UD_ENABLED()) SET_XCPT(X86_XCPT_UD);
5704 if (VBOXVMM_XCPT_NM_ENABLED()) SET_XCPT(X86_XCPT_NM);
5705 if (VBOXVMM_XCPT_DF_ENABLED()) SET_XCPT(X86_XCPT_DF);
5706 if (VBOXVMM_XCPT_TS_ENABLED()) SET_XCPT(X86_XCPT_TS);
5707 if (VBOXVMM_XCPT_NP_ENABLED()) SET_XCPT(X86_XCPT_NP);
5708 if (VBOXVMM_XCPT_SS_ENABLED()) SET_XCPT(X86_XCPT_SS);
5709 if (VBOXVMM_XCPT_GP_ENABLED()) SET_XCPT(X86_XCPT_GP);
5710 if (VBOXVMM_XCPT_PF_ENABLED()) SET_XCPT(X86_XCPT_PF);
5711 if (VBOXVMM_XCPT_MF_ENABLED()) SET_XCPT(X86_XCPT_MF);
5712 if (VBOXVMM_XCPT_AC_ENABLED()) SET_XCPT(X86_XCPT_AC);
5713 if (VBOXVMM_XCPT_XF_ENABLED()) SET_XCPT(X86_XCPT_XF);
5714 if (VBOXVMM_XCPT_VE_ENABLED()) SET_XCPT(X86_XCPT_VE);
5715 if (VBOXVMM_XCPT_SX_ENABLED()) SET_XCPT(X86_XCPT_SX);
5716
5717#undef SET_XCPT
5718
5719 /*
5720 * Process events and probes for VM-exits, making sure we get the wanted VM-exits.
5721 *
5722 * Note! This is the reverse of what hmR0SvmHandleExitDtraceEvents does.
5723 * So, when adding/changing/removing please don't forget to update it.
5724 *
5725 * Some of the macros are picking up local variables to save horizontal space,
5726 * (being able to see it in a table is the lesser evil here).
5727 */
5728#define IS_EITHER_ENABLED(a_pVM, a_EventSubName) \
5729 ( DBGF_IS_EVENT_ENABLED(a_pVM, RT_CONCAT(DBGFEVENT_, a_EventSubName)) \
5730 || RT_CONCAT3(VBOXVMM_, a_EventSubName, _ENABLED)() )
5731#define SET_ONLY_XBM_IF_EITHER_EN(a_EventSubName, a_uExit) \
5732 if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
5733 { AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
5734 ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
5735 } else do { } while (0)
5736#define SET_INCP_XBM_IF_EITHER_EN(a_EventSubName, a_uExit, a_fInterceptCtrl) \
5737 if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
5738 { \
5739 (pDbgState)->bmInterceptExtra |= (a_fInterceptCtrl); \
5740 AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
5741 ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
5742 } else do { } while (0)
5743
5744 /** @todo double check these */
5745 /** @todo Check what more AMD-V specific we can intercept. */
5746 //SET_INCP_XBM_IF_EITHER_EN(EXIT_TASK_SWITCH, SVM_EXIT_TASK_SWITCH, SVM_CTRL_INTERCEPT_TASK_SWITCH);
5747 SET_ONLY_XBM_IF_EITHER_EN(EXIT_TASK_SWITCH, SVM_EXIT_TASK_SWITCH);
5748 SET_INCP_XBM_IF_EITHER_EN(INSTR_VMM_CALL, SVM_EXIT_VMMCALL, SVM_CTRL_INTERCEPT_VMMCALL);
5749 SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMM_CALL, SVM_EXIT_VMMCALL);
5750 SET_INCP_XBM_IF_EITHER_EN(INSTR_SVM_VMRUN, SVM_EXIT_VMRUN, SVM_CTRL_INTERCEPT_VMRUN);
5751 SET_ONLY_XBM_IF_EITHER_EN( EXIT_SVM_VMRUN, SVM_EXIT_VMRUN);
5752 SET_INCP_XBM_IF_EITHER_EN(INSTR_SVM_VMLOAD, SVM_EXIT_VMLOAD, SVM_CTRL_INTERCEPT_VMLOAD);
5753 SET_ONLY_XBM_IF_EITHER_EN( EXIT_SVM_VMLOAD, SVM_EXIT_VMLOAD);
5754 SET_INCP_XBM_IF_EITHER_EN(INSTR_SVM_VMSAVE, SVM_EXIT_VMSAVE, SVM_CTRL_INTERCEPT_VMSAVE);
5755 SET_ONLY_XBM_IF_EITHER_EN( EXIT_SVM_VMSAVE, SVM_EXIT_VMSAVE);
5756 SET_INCP_XBM_IF_EITHER_EN(INSTR_SVM_STGI, SVM_EXIT_STGI, SVM_CTRL_INTERCEPT_STGI);
5757 SET_ONLY_XBM_IF_EITHER_EN( EXIT_SVM_STGI, SVM_EXIT_STGI);
5758 SET_INCP_XBM_IF_EITHER_EN(INSTR_SVM_CLGI, SVM_EXIT_CLGI, SVM_CTRL_INTERCEPT_CLGI);
5759 SET_ONLY_XBM_IF_EITHER_EN( EXIT_SVM_CLGI, SVM_EXIT_CLGI);
5760
5761 SET_INCP_XBM_IF_EITHER_EN(INSTR_CPUID, SVM_EXIT_CPUID, SVM_CTRL_INTERCEPT_CPUID);
5762 SET_ONLY_XBM_IF_EITHER_EN( EXIT_CPUID, SVM_EXIT_CPUID);
5763 SET_INCP_XBM_IF_EITHER_EN(INSTR_HALT, SVM_EXIT_HLT, SVM_CTRL_INTERCEPT_HLT);
5764 SET_ONLY_XBM_IF_EITHER_EN( EXIT_HALT, SVM_EXIT_HLT);
5765 SET_INCP_XBM_IF_EITHER_EN(INSTR_INVD, SVM_EXIT_INVD, SVM_CTRL_INTERCEPT_INVD);
5766 SET_ONLY_XBM_IF_EITHER_EN( EXIT_INVD, SVM_EXIT_INVD);
5767 SET_INCP_XBM_IF_EITHER_EN(INSTR_INVLPG, SVM_EXIT_INVLPG, SVM_CTRL_INTERCEPT_INVLPG);
5768 SET_ONLY_XBM_IF_EITHER_EN( EXIT_INVLPG, SVM_EXIT_INVLPG);
5769 SET_INCP_XBM_IF_EITHER_EN(INSTR_RDPMC, SVM_EXIT_RDPMC, SVM_CTRL_INTERCEPT_RDPMC);
5770 SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDPMC, SVM_EXIT_RDPMC);
5771 SET_INCP_XBM_IF_EITHER_EN(INSTR_RDTSC, SVM_EXIT_RDTSC, SVM_CTRL_INTERCEPT_RDTSC);
5772 SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDTSC, SVM_EXIT_RDTSC);
5773 SET_INCP_XBM_IF_EITHER_EN(INSTR_RDTSCP, SVM_EXIT_RDTSCP, SVM_CTRL_INTERCEPT_RDTSCP);
5774 SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDTSCP, SVM_EXIT_RDTSCP);
5775 SET_INCP_XBM_IF_EITHER_EN(INSTR_RSM, SVM_EXIT_RSM, SVM_CTRL_INTERCEPT_RSM);
5776 SET_ONLY_XBM_IF_EITHER_EN( EXIT_RSM, SVM_EXIT_RSM);
5777
5778 if (IS_EITHER_ENABLED(pVM, INSTR_CRX_READ))
5779 pDbgState->bmInterceptRdCRxExtra = 0xffff;
5780 if (IS_EITHER_ENABLED(pVM, INSTR_CRX_READ) || IS_EITHER_ENABLED(pVM, EXIT_CRX_READ))
5781 ASMBitSetRange(pDbgState->bmExitsToCheck, SVM_EXIT_READ_CR0, SVM_EXIT_READ_CR15 + 1);
5782
5783 if (IS_EITHER_ENABLED(pVM, INSTR_CRX_WRITE))
5784 pDbgState->bmInterceptWrCRxExtra = 0xffff;
5785 if (IS_EITHER_ENABLED(pVM, INSTR_CRX_WRITE) || IS_EITHER_ENABLED(pVM, EXIT_CRX_WRITE))
5786 {
5787 ASMBitSetRange(pDbgState->bmExitsToCheck, SVM_EXIT_WRITE_CR0, SVM_EXIT_WRITE_CR15 + 1);
5788 ASMBitSet(pDbgState->bmExitsToCheck, SVM_EXIT_CR0_SEL_WRITE);
5789 }
5790
5791 if (IS_EITHER_ENABLED(pVM, INSTR_DRX_READ))
5792 pDbgState->bmInterceptRdDRxExtra = 0xffff;
5793 if (IS_EITHER_ENABLED(pVM, INSTR_DRX_READ) || IS_EITHER_ENABLED(pVM, EXIT_DRX_READ))
5794 ASMBitSetRange(pDbgState->bmExitsToCheck, SVM_EXIT_READ_DR0, SVM_EXIT_READ_DR15 + 1);
5795
5796 if (IS_EITHER_ENABLED(pVM, INSTR_DRX_WRITE))
5797 pDbgState->bmInterceptWrDRxExtra = 0xffff;
5798 if (IS_EITHER_ENABLED(pVM, INSTR_DRX_WRITE) || IS_EITHER_ENABLED(pVM, EXIT_DRX_WRITE))
5799 ASMBitSetRange(pDbgState->bmExitsToCheck, SVM_EXIT_WRITE_DR0, SVM_EXIT_WRITE_DR15 + 1);
5800
5801 SET_ONLY_XBM_IF_EITHER_EN(INSTR_RDMSR, SVM_EXIT_MSR); /** @todo modify bitmap to intercept almost everything? (Clearing MSR_PROT just means no intercepts.) */
5802 SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDMSR, SVM_EXIT_MSR);
5803 SET_ONLY_XBM_IF_EITHER_EN(INSTR_WRMSR, SVM_EXIT_MSR); /** @todo ditto */
5804 SET_ONLY_XBM_IF_EITHER_EN( EXIT_WRMSR, SVM_EXIT_MSR);
5805 SET_INCP_XBM_IF_EITHER_EN(INSTR_MWAIT, SVM_EXIT_MWAIT, SVM_CTRL_INTERCEPT_MWAIT);
5806 SET_ONLY_XBM_IF_EITHER_EN( EXIT_MWAIT, SVM_EXIT_MWAIT);
5807 if (ASMBitTest(pDbgState->bmExitsToCheck, SVM_EXIT_MWAIT))
5808 ASMBitSet(pDbgState->bmExitsToCheck, SVM_EXIT_MWAIT_ARMED);
5809 SET_INCP_XBM_IF_EITHER_EN(INSTR_MONITOR, SVM_EXIT_MONITOR, SVM_CTRL_INTERCEPT_MONITOR);
5810 SET_ONLY_XBM_IF_EITHER_EN( EXIT_MONITOR, SVM_EXIT_MONITOR);
5811 SET_INCP_XBM_IF_EITHER_EN(INSTR_PAUSE, SVM_EXIT_PAUSE, SVM_CTRL_INTERCEPT_PAUSE);
5812 SET_ONLY_XBM_IF_EITHER_EN( EXIT_PAUSE, SVM_EXIT_PAUSE);
5813 SET_INCP_XBM_IF_EITHER_EN(INSTR_SIDT, SVM_EXIT_IDTR_READ, SVM_CTRL_INTERCEPT_IDTR_READS);
5814 SET_ONLY_XBM_IF_EITHER_EN( EXIT_SIDT, SVM_EXIT_IDTR_READ);
5815 SET_INCP_XBM_IF_EITHER_EN(INSTR_LIDT, SVM_EXIT_IDTR_WRITE, SVM_CTRL_INTERCEPT_IDTR_WRITES);
5816 SET_ONLY_XBM_IF_EITHER_EN( EXIT_LIDT, SVM_EXIT_IDTR_WRITE);
5817 SET_INCP_XBM_IF_EITHER_EN(INSTR_SGDT, SVM_EXIT_GDTR_READ, SVM_CTRL_INTERCEPT_GDTR_READS);
5818 SET_ONLY_XBM_IF_EITHER_EN( EXIT_SGDT, SVM_EXIT_GDTR_READ);
5819 SET_INCP_XBM_IF_EITHER_EN(INSTR_LGDT, SVM_EXIT_GDTR_WRITE, SVM_CTRL_INTERCEPT_GDTR_WRITES);
5820 SET_ONLY_XBM_IF_EITHER_EN( EXIT_LGDT, SVM_EXIT_GDTR_WRITE);
5821 SET_INCP_XBM_IF_EITHER_EN(INSTR_SLDT, SVM_EXIT_LDTR_READ, SVM_CTRL_INTERCEPT_LDTR_READS);
5822 SET_ONLY_XBM_IF_EITHER_EN( EXIT_SLDT, SVM_EXIT_LDTR_READ);
5823 SET_INCP_XBM_IF_EITHER_EN(INSTR_LLDT, SVM_EXIT_LDTR_WRITE, SVM_CTRL_INTERCEPT_LDTR_WRITES);
5824 SET_ONLY_XBM_IF_EITHER_EN( EXIT_LLDT, SVM_EXIT_LDTR_WRITE);
5825 SET_INCP_XBM_IF_EITHER_EN(INSTR_STR, SVM_EXIT_TR_READ, SVM_CTRL_INTERCEPT_TR_READS);
5826 SET_ONLY_XBM_IF_EITHER_EN( EXIT_STR, SVM_EXIT_TR_READ);
5827 SET_INCP_XBM_IF_EITHER_EN(INSTR_LTR, SVM_EXIT_TR_WRITE, SVM_CTRL_INTERCEPT_TR_WRITES);
5828 SET_ONLY_XBM_IF_EITHER_EN( EXIT_LTR, SVM_EXIT_TR_WRITE);
5829 SET_INCP_XBM_IF_EITHER_EN(INSTR_WBINVD, SVM_EXIT_WBINVD, SVM_CTRL_INTERCEPT_WBINVD);
5830 SET_ONLY_XBM_IF_EITHER_EN( EXIT_WBINVD, SVM_EXIT_WBINVD);
5831 SET_INCP_XBM_IF_EITHER_EN(INSTR_XSETBV, SVM_EXIT_XSETBV, SVM_CTRL_INTERCEPT_XSETBV);
5832 SET_ONLY_XBM_IF_EITHER_EN( EXIT_XSETBV, SVM_EXIT_XSETBV);
5833
5834 if (DBGF_IS_EVENT_ENABLED(pVM, DBGFEVENT_TRIPLE_FAULT))
5835 ASMBitSet(pDbgState->bmExitsToCheck, SVM_EXIT_SHUTDOWN);
5836
5837#undef IS_EITHER_ENABLED
5838#undef SET_ONLY_XBM_IF_EITHER_EN
5839#undef SET_INCP_XBM_IF_EITHER_EN
5840
5841 /*
5842 * Sanitize the control stuff.
5843 */
5844 /** @todo filter out unsupported stuff? */
5845 if ( pVCpu->hmr0.s.fDebugWantRdTscExit
5846 != RT_BOOL(pDbgState->bmInterceptExtra & (SVM_CTRL_INTERCEPT_RDTSC | SVM_CTRL_INTERCEPT_RDTSCP)))
5847 {
5848 pVCpu->hmr0.s.fDebugWantRdTscExit ^= true;
5849 /// @todo pVmxTransient->fUpdatedTscOffsettingAndPreemptTimer = false;
5850 RT_NOREF(pSvmTransient);
5851 }
5852
5853 Log6(("HM: debug state: bmInterceptExtra=%#RX64 bmXcptExtra=%#RX32%s%s%s%s bmExitsToCheck=%08RX32'%08RX32'%08RX32'%08RX32'%08RX32\n",
5854 pDbgState->bmInterceptExtra, pDbgState->bmXcptExtra,
5855 pDbgState->bmInterceptRdCRxExtra ? " rd-cr" : "",
5856 pDbgState->bmInterceptWrCRxExtra ? " wr-cr" : "",
5857 pDbgState->bmInterceptRdDRxExtra ? " rd-dr" : "",
5858 pDbgState->bmInterceptWrDRxExtra ? " wr-dr" : "",
5859 pDbgState->bmExitsToCheck[0],
5860 pDbgState->bmExitsToCheck[1],
5861 pDbgState->bmExitsToCheck[2],
5862 pDbgState->bmExitsToCheck[3],
5863 pDbgState->bmExitsToCheck[4]));
5864}
5865
5866
5867/**
5868 * Fires off DBGF events and dtrace probes for a VM-exit, when it's
5869 * appropriate.
5870 *
5871 * The caller has checked the VM-exit against the SVMRUNDBGSTATE::bmExitsToCheck
5872 * bitmap.
5873 *
5874 * @returns Strict VBox status code (i.e. informational status codes too).
5875 * @param pVCpu The cross context virtual CPU structure.
5876 * @param pSvmTransient The SVM-transient structure.
5877 * @param uExitCode The VM-exit code.
5878 *
5879 * @remarks The name of this function is displayed by dtrace, so keep it short
5880 * and to the point. No longer than 33 chars long, please.
5881 */
5882static VBOXSTRICTRC hmR0SvmHandleExitDtraceEvents(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient, uint64_t uExitCode)
5883{
5884 /*
5885 * Translate the event into a DBGF event (enmEvent + uEventArg) and at the
5886 * same time check whether any corresponding Dtrace event is enabled (fDtrace).
5887 *
5888 * Note! This is the reverse operation of what hmR0SvmPreRunGuestDebugStateUpdate
5889 * does. Must add/change/remove both places. Same ordering, please.
5890 *
5891 * Added/removed events must also be reflected in the next section
5892 * where we dispatch dtrace events.
5893 */
5894 bool fDtrace1 = false;
5895 bool fDtrace2 = false;
5896 DBGFEVENTTYPE enmEvent1 = DBGFEVENT_END;
5897 DBGFEVENTTYPE enmEvent2 = DBGFEVENT_END;
5898 uint64_t uEventArg = 0;
5899#define SET_XCPT(a_XcptName) \
5900 do { \
5901 enmEvent2 = RT_CONCAT(DBGFEVENT_XCPT_, a_XcptName); \
5902 fDtrace2 = RT_CONCAT3(VBOXVMM_XCPT_, a_XcptName, _ENABLED)(); \
5903 } while (0)
5904#define SET_EXIT(a_EventSubName) \
5905 do { \
5906 enmEvent2 = RT_CONCAT(DBGFEVENT_EXIT_, a_EventSubName); \
5907 fDtrace2 = RT_CONCAT3(VBOXVMM_EXIT_, a_EventSubName, _ENABLED)(); \
5908 } while (0)
5909#define SET_BOTH(a_EventSubName) \
5910 do { \
5911 enmEvent1 = RT_CONCAT(DBGFEVENT_INSTR_, a_EventSubName); \
5912 enmEvent2 = RT_CONCAT(DBGFEVENT_EXIT_, a_EventSubName); \
5913 fDtrace1 = RT_CONCAT3(VBOXVMM_INSTR_, a_EventSubName, _ENABLED)(); \
5914 fDtrace2 = RT_CONCAT3(VBOXVMM_EXIT_, a_EventSubName, _ENABLED)(); \
5915 } while (0)
5916 switch (uExitCode)
5917 {
5918 case SVM_EXIT_SWINT:
5919 enmEvent2 = DBGFEVENT_INTERRUPT_SOFTWARE;
5920 fDtrace2 = VBOXVMM_INT_SOFTWARE_ENABLED();
5921 uEventArg = pSvmTransient->pVmcb->ctrl.u64ExitInfo1;
5922 break;
5923
5924 case SVM_EXIT_XCPT_DE: SET_XCPT(DE); break;
5925 case SVM_EXIT_XCPT_DB: SET_XCPT(DB); break;
5926 case SVM_EXIT_XCPT_BP: SET_XCPT(BP); break;
5927 case SVM_EXIT_XCPT_OF: SET_XCPT(OF); break;
5928 case SVM_EXIT_XCPT_BR: SET_XCPT(BR); break;
5929 case SVM_EXIT_XCPT_UD: SET_XCPT(UD); break;
5930 case SVM_EXIT_XCPT_NM: SET_XCPT(NM); break;
5931 case SVM_EXIT_XCPT_DF: SET_XCPT(DF); break;
5932 case SVM_EXIT_XCPT_TS: SET_XCPT(TS); uEventArg = pSvmTransient->pVmcb->ctrl.u64ExitInfo1; break;
5933 case SVM_EXIT_XCPT_NP: SET_XCPT(NP); uEventArg = pSvmTransient->pVmcb->ctrl.u64ExitInfo1; break;
5934 case SVM_EXIT_XCPT_SS: SET_XCPT(SS); uEventArg = pSvmTransient->pVmcb->ctrl.u64ExitInfo1; break;
5935 case SVM_EXIT_XCPT_GP: SET_XCPT(GP); uEventArg = pSvmTransient->pVmcb->ctrl.u64ExitInfo1; break;
5936 case SVM_EXIT_XCPT_PF: SET_XCPT(PF); uEventArg = pSvmTransient->pVmcb->ctrl.u64ExitInfo1; break;
5937 case SVM_EXIT_XCPT_MF: SET_XCPT(MF); break;
5938 case SVM_EXIT_XCPT_AC: SET_XCPT(AC); break;
5939 case SVM_EXIT_XCPT_XF: SET_XCPT(XF); break;
5940 case SVM_EXIT_XCPT_VE: SET_XCPT(VE); break;
5941 case SVM_EXIT_XCPT_SX: SET_XCPT(SX); uEventArg = pSvmTransient->pVmcb->ctrl.u64ExitInfo1; break;
5942
5943 case SVM_EXIT_XCPT_2: enmEvent2 = DBGFEVENT_XCPT_02; break;
5944 case SVM_EXIT_XCPT_9: enmEvent2 = DBGFEVENT_XCPT_09; break;
5945 case SVM_EXIT_XCPT_15: enmEvent2 = DBGFEVENT_XCPT_0f; break;
5946 case SVM_EXIT_XCPT_18: enmEvent2 = DBGFEVENT_XCPT_MC; break;
5947 case SVM_EXIT_XCPT_21: enmEvent2 = DBGFEVENT_XCPT_15; break;
5948 case SVM_EXIT_XCPT_22: enmEvent2 = DBGFEVENT_XCPT_16; break;
5949 case SVM_EXIT_XCPT_23: enmEvent2 = DBGFEVENT_XCPT_17; break;
5950 case SVM_EXIT_XCPT_24: enmEvent2 = DBGFEVENT_XCPT_18; break;
5951 case SVM_EXIT_XCPT_25: enmEvent2 = DBGFEVENT_XCPT_19; break;
5952 case SVM_EXIT_XCPT_26: enmEvent2 = DBGFEVENT_XCPT_1a; break;
5953 case SVM_EXIT_XCPT_27: enmEvent2 = DBGFEVENT_XCPT_1b; break;
5954 case SVM_EXIT_XCPT_28: enmEvent2 = DBGFEVENT_XCPT_1c; break;
5955 case SVM_EXIT_XCPT_29: enmEvent2 = DBGFEVENT_XCPT_1d; break;
5956 case SVM_EXIT_XCPT_31: enmEvent2 = DBGFEVENT_XCPT_1f; break;
5957
5958 case SVM_EXIT_TASK_SWITCH: SET_EXIT(TASK_SWITCH); break;
5959 case SVM_EXIT_VMMCALL: SET_BOTH(VMM_CALL); break;
5960 case SVM_EXIT_VMRUN: SET_BOTH(SVM_VMRUN); break;
5961 case SVM_EXIT_VMLOAD: SET_BOTH(SVM_VMLOAD); break;
5962 case SVM_EXIT_VMSAVE: SET_BOTH(SVM_VMSAVE); break;
5963 case SVM_EXIT_STGI: SET_BOTH(SVM_STGI); break;
5964 case SVM_EXIT_CLGI: SET_BOTH(SVM_CLGI); break;
5965 case SVM_EXIT_CPUID: SET_BOTH(CPUID); break;
5966 case SVM_EXIT_HLT: SET_BOTH(HALT); break;
5967 case SVM_EXIT_INVD: SET_BOTH(INVD); break;
5968 case SVM_EXIT_INVLPG: SET_BOTH(INVLPG); break;
5969 case SVM_EXIT_RDPMC: SET_BOTH(RDPMC); break;
5970 case SVM_EXIT_RDTSC: SET_BOTH(RDTSC); break;
5971 case SVM_EXIT_RDTSCP: SET_BOTH(RDTSCP); break;
5972 case SVM_EXIT_RSM: SET_BOTH(RSM); break;
5973
5974 case SVM_EXIT_READ_CR0: case SVM_EXIT_READ_CR1: case SVM_EXIT_READ_CR2: case SVM_EXIT_READ_CR3:
5975 case SVM_EXIT_READ_CR4: case SVM_EXIT_READ_CR5: case SVM_EXIT_READ_CR6: case SVM_EXIT_READ_CR7:
5976 case SVM_EXIT_READ_CR8: case SVM_EXIT_READ_CR9: case SVM_EXIT_READ_CR10: case SVM_EXIT_READ_CR11:
5977 case SVM_EXIT_READ_CR12: case SVM_EXIT_READ_CR13: case SVM_EXIT_READ_CR14: case SVM_EXIT_READ_CR15:
5978 SET_BOTH(CRX_READ);
5979 uEventArg = uExitCode - SVM_EXIT_READ_CR0;
5980 break;
5981 case SVM_EXIT_WRITE_CR0: case SVM_EXIT_WRITE_CR1: case SVM_EXIT_WRITE_CR2: case SVM_EXIT_WRITE_CR3:
5982 case SVM_EXIT_WRITE_CR4: case SVM_EXIT_WRITE_CR5: case SVM_EXIT_WRITE_CR6: case SVM_EXIT_WRITE_CR7:
5983 case SVM_EXIT_WRITE_CR8: case SVM_EXIT_WRITE_CR9: case SVM_EXIT_WRITE_CR10: case SVM_EXIT_WRITE_CR11:
5984 case SVM_EXIT_WRITE_CR12: case SVM_EXIT_WRITE_CR13: case SVM_EXIT_WRITE_CR14: case SVM_EXIT_WRITE_CR15:
5985 case SVM_EXIT_CR0_SEL_WRITE:
5986 SET_BOTH(CRX_WRITE);
5987 uEventArg = uExitCode - SVM_EXIT_WRITE_CR0;
5988 break;
5989 case SVM_EXIT_READ_DR0: case SVM_EXIT_READ_DR1: case SVM_EXIT_READ_DR2: case SVM_EXIT_READ_DR3:
5990 case SVM_EXIT_READ_DR4: case SVM_EXIT_READ_DR5: case SVM_EXIT_READ_DR6: case SVM_EXIT_READ_DR7:
5991 case SVM_EXIT_READ_DR8: case SVM_EXIT_READ_DR9: case SVM_EXIT_READ_DR10: case SVM_EXIT_READ_DR11:
5992 case SVM_EXIT_READ_DR12: case SVM_EXIT_READ_DR13: case SVM_EXIT_READ_DR14: case SVM_EXIT_READ_DR15:
5993 SET_BOTH(DRX_READ);
5994 uEventArg = uExitCode - SVM_EXIT_READ_DR0;
5995 break;
5996 case SVM_EXIT_WRITE_DR0: case SVM_EXIT_WRITE_DR1: case SVM_EXIT_WRITE_DR2: case SVM_EXIT_WRITE_DR3:
5997 case SVM_EXIT_WRITE_DR4: case SVM_EXIT_WRITE_DR5: case SVM_EXIT_WRITE_DR6: case SVM_EXIT_WRITE_DR7:
5998 case SVM_EXIT_WRITE_DR8: case SVM_EXIT_WRITE_DR9: case SVM_EXIT_WRITE_DR10: case SVM_EXIT_WRITE_DR11:
5999 case SVM_EXIT_WRITE_DR12: case SVM_EXIT_WRITE_DR13: case SVM_EXIT_WRITE_DR14: case SVM_EXIT_WRITE_DR15:
6000 SET_BOTH(DRX_WRITE);
6001 uEventArg = uExitCode - SVM_EXIT_WRITE_DR0;
6002 break;
6003 case SVM_EXIT_MSR:
6004 if (pSvmTransient->pVmcb->ctrl.u64ExitInfo1 == SVM_EXIT1_MSR_WRITE)
6005 SET_BOTH(WRMSR);
6006 else
6007 SET_BOTH(RDMSR);
6008 break;
6009 case SVM_EXIT_MWAIT_ARMED:
6010 case SVM_EXIT_MWAIT: SET_BOTH(MWAIT); break;
6011 case SVM_EXIT_MONITOR: SET_BOTH(MONITOR); break;
6012 case SVM_EXIT_PAUSE: SET_BOTH(PAUSE); break;
6013 case SVM_EXIT_IDTR_READ: SET_BOTH(SIDT); break;
6014 case SVM_EXIT_IDTR_WRITE: SET_BOTH(LIDT); break;
6015 case SVM_EXIT_GDTR_READ: SET_BOTH(SGDT); break;
6016 case SVM_EXIT_GDTR_WRITE: SET_BOTH(LGDT); break;
6017 case SVM_EXIT_LDTR_READ: SET_BOTH(SLDT); break;
6018 case SVM_EXIT_LDTR_WRITE: SET_BOTH(LLDT); break;
6019 case SVM_EXIT_TR_READ: SET_BOTH(STR); break;
6020 case SVM_EXIT_TR_WRITE: SET_BOTH(LTR); break;
6021 case SVM_EXIT_WBINVD: SET_BOTH(WBINVD); break;
6022 case SVM_EXIT_XSETBV: SET_BOTH(XSETBV); break;
6023
6024 case SVM_EXIT_SHUTDOWN:
6025 enmEvent1 = DBGFEVENT_TRIPLE_FAULT;
6026 //fDtrace1 = VBOXVMM_EXIT_TRIPLE_FAULT_ENABLED();
6027 break;
6028
6029 default:
6030 AssertMsgFailed(("Unexpected VM-exit=%#x\n", uExitCode));
6031 break;
6032 }
6033#undef SET_BOTH
6034#undef SET_EXIT
6035
6036 /*
6037 * Dtrace tracepoints go first. We do them here at once so we don't
6038 * have to copy the guest state saving and stuff a few dozen times.
6039 * Down side is that we've got to repeat the switch, though this time
6040 * we use enmEvent since the probes are a subset of what DBGF does.
6041 */
6042 if (fDtrace1 || fDtrace2)
6043 {
6044 hmR0SvmImportGuestState(pVCpu, HMSVM_CPUMCTX_EXTRN_ALL);
6045 PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
6046 switch (enmEvent1)
6047 {
6048 /** @todo consider which extra parameters would be helpful for each probe. */
6049 case DBGFEVENT_END: break;
6050 case DBGFEVENT_INTERRUPT_SOFTWARE: VBOXVMM_INT_SOFTWARE(pVCpu, pCtx, (uint8_t)uEventArg); break;
6051 case DBGFEVENT_XCPT_DE: VBOXVMM_XCPT_DE(pVCpu, pCtx); break;
6052 case DBGFEVENT_XCPT_DB: VBOXVMM_XCPT_DB(pVCpu, pCtx, pCtx->dr[6]); break;
6053 case DBGFEVENT_XCPT_BP: VBOXVMM_XCPT_BP(pVCpu, pCtx); break;
6054 case DBGFEVENT_XCPT_OF: VBOXVMM_XCPT_OF(pVCpu, pCtx); break;
6055 case DBGFEVENT_XCPT_BR: VBOXVMM_XCPT_BR(pVCpu, pCtx); break;
6056 case DBGFEVENT_XCPT_UD: VBOXVMM_XCPT_UD(pVCpu, pCtx); break;
6057 case DBGFEVENT_XCPT_NM: VBOXVMM_XCPT_NM(pVCpu, pCtx); break;
6058 case DBGFEVENT_XCPT_DF: VBOXVMM_XCPT_DF(pVCpu, pCtx); break;
6059 case DBGFEVENT_XCPT_TS: VBOXVMM_XCPT_TS(pVCpu, pCtx, (uint32_t)uEventArg); break;
6060 case DBGFEVENT_XCPT_NP: VBOXVMM_XCPT_NP(pVCpu, pCtx, (uint32_t)uEventArg); break;
6061 case DBGFEVENT_XCPT_SS: VBOXVMM_XCPT_SS(pVCpu, pCtx, (uint32_t)uEventArg); break;
6062 case DBGFEVENT_XCPT_GP: VBOXVMM_XCPT_GP(pVCpu, pCtx, (uint32_t)uEventArg); break;
6063 case DBGFEVENT_XCPT_PF: VBOXVMM_XCPT_PF(pVCpu, pCtx, (uint32_t)uEventArg, pCtx->cr2); break;
6064 case DBGFEVENT_XCPT_MF: VBOXVMM_XCPT_MF(pVCpu, pCtx); break;
6065 case DBGFEVENT_XCPT_AC: VBOXVMM_XCPT_AC(pVCpu, pCtx); break;
6066 case DBGFEVENT_XCPT_XF: VBOXVMM_XCPT_XF(pVCpu, pCtx); break;
6067 case DBGFEVENT_XCPT_VE: VBOXVMM_XCPT_VE(pVCpu, pCtx); break;
6068 case DBGFEVENT_XCPT_SX: VBOXVMM_XCPT_SX(pVCpu, pCtx, (uint32_t)uEventArg); break;
6069 case DBGFEVENT_INSTR_CPUID: VBOXVMM_INSTR_CPUID(pVCpu, pCtx, pCtx->eax, pCtx->ecx); break;
6070 case DBGFEVENT_INSTR_HALT: VBOXVMM_INSTR_HALT(pVCpu, pCtx); break;
6071 case DBGFEVENT_INSTR_INVD: VBOXVMM_INSTR_INVD(pVCpu, pCtx); break;
6072 case DBGFEVENT_INSTR_INVLPG: VBOXVMM_INSTR_INVLPG(pVCpu, pCtx); break;
6073 case DBGFEVENT_INSTR_RDPMC: VBOXVMM_INSTR_RDPMC(pVCpu, pCtx); break;
6074 case DBGFEVENT_INSTR_RDTSC: VBOXVMM_INSTR_RDTSC(pVCpu, pCtx); break;
6075 case DBGFEVENT_INSTR_RSM: VBOXVMM_INSTR_RSM(pVCpu, pCtx); break;
6076 case DBGFEVENT_INSTR_CRX_READ: VBOXVMM_INSTR_CRX_READ(pVCpu, pCtx, (uint8_t)uEventArg); break;
6077 case DBGFEVENT_INSTR_CRX_WRITE: VBOXVMM_INSTR_CRX_WRITE(pVCpu, pCtx, (uint8_t)uEventArg); break;
6078 case DBGFEVENT_INSTR_DRX_READ: VBOXVMM_INSTR_DRX_READ(pVCpu, pCtx, (uint8_t)uEventArg); break;
6079 case DBGFEVENT_INSTR_DRX_WRITE: VBOXVMM_INSTR_DRX_WRITE(pVCpu, pCtx, (uint8_t)uEventArg); break;
6080 case DBGFEVENT_INSTR_RDMSR: VBOXVMM_INSTR_RDMSR(pVCpu, pCtx, pCtx->ecx); break;
6081 case DBGFEVENT_INSTR_WRMSR: VBOXVMM_INSTR_WRMSR(pVCpu, pCtx, pCtx->ecx,
6082 RT_MAKE_U64(pCtx->eax, pCtx->edx)); break;
6083 case DBGFEVENT_INSTR_MWAIT: VBOXVMM_INSTR_MWAIT(pVCpu, pCtx); break;
6084 case DBGFEVENT_INSTR_MONITOR: VBOXVMM_INSTR_MONITOR(pVCpu, pCtx); break;
6085 case DBGFEVENT_INSTR_PAUSE: VBOXVMM_INSTR_PAUSE(pVCpu, pCtx); break;
6086 case DBGFEVENT_INSTR_SGDT: VBOXVMM_INSTR_SGDT(pVCpu, pCtx); break;
6087 case DBGFEVENT_INSTR_SIDT: VBOXVMM_INSTR_SIDT(pVCpu, pCtx); break;
6088 case DBGFEVENT_INSTR_LGDT: VBOXVMM_INSTR_LGDT(pVCpu, pCtx); break;
6089 case DBGFEVENT_INSTR_LIDT: VBOXVMM_INSTR_LIDT(pVCpu, pCtx); break;
6090 case DBGFEVENT_INSTR_SLDT: VBOXVMM_INSTR_SLDT(pVCpu, pCtx); break;
6091 case DBGFEVENT_INSTR_STR: VBOXVMM_INSTR_STR(pVCpu, pCtx); break;
6092 case DBGFEVENT_INSTR_LLDT: VBOXVMM_INSTR_LLDT(pVCpu, pCtx); break;
6093 case DBGFEVENT_INSTR_LTR: VBOXVMM_INSTR_LTR(pVCpu, pCtx); break;
6094 case DBGFEVENT_INSTR_RDTSCP: VBOXVMM_INSTR_RDTSCP(pVCpu, pCtx); break;
6095 case DBGFEVENT_INSTR_WBINVD: VBOXVMM_INSTR_WBINVD(pVCpu, pCtx); break;
6096 case DBGFEVENT_INSTR_XSETBV: VBOXVMM_INSTR_XSETBV(pVCpu, pCtx); break;
6097 case DBGFEVENT_INSTR_VMM_CALL: VBOXVMM_INSTR_VMM_CALL(pVCpu, pCtx); break;
6098 case DBGFEVENT_INSTR_SVM_VMRUN: VBOXVMM_INSTR_SVM_VMRUN(pVCpu, pCtx); break;
6099 case DBGFEVENT_INSTR_SVM_VMLOAD: VBOXVMM_INSTR_SVM_VMLOAD(pVCpu, pCtx); break;
6100 case DBGFEVENT_INSTR_SVM_VMSAVE: VBOXVMM_INSTR_SVM_VMSAVE(pVCpu, pCtx); break;
6101 case DBGFEVENT_INSTR_SVM_STGI: VBOXVMM_INSTR_SVM_STGI(pVCpu, pCtx); break;
6102 case DBGFEVENT_INSTR_SVM_CLGI: VBOXVMM_INSTR_SVM_CLGI(pVCpu, pCtx); break;
6103 default: AssertMsgFailed(("enmEvent1=%d uExitCode=%d\n", enmEvent1, uExitCode)); break;
6104 }
6105 switch (enmEvent2)
6106 {
6107 /** @todo consider which extra parameters would be helpful for each probe. */
6108 case DBGFEVENT_END: break;
6109 case DBGFEVENT_EXIT_TASK_SWITCH: VBOXVMM_EXIT_TASK_SWITCH(pVCpu, pCtx); break;
6110 case DBGFEVENT_EXIT_CPUID: VBOXVMM_EXIT_CPUID(pVCpu, pCtx, pCtx->eax, pCtx->ecx); break;
6111 case DBGFEVENT_EXIT_HALT: VBOXVMM_EXIT_HALT(pVCpu, pCtx); break;
6112 case DBGFEVENT_EXIT_INVD: VBOXVMM_EXIT_INVD(pVCpu, pCtx); break;
6113 case DBGFEVENT_EXIT_INVLPG: VBOXVMM_EXIT_INVLPG(pVCpu, pCtx); break;
6114 case DBGFEVENT_EXIT_RDPMC: VBOXVMM_EXIT_RDPMC(pVCpu, pCtx); break;
6115 case DBGFEVENT_EXIT_RDTSC: VBOXVMM_EXIT_RDTSC(pVCpu, pCtx); break;
6116 case DBGFEVENT_EXIT_RSM: VBOXVMM_EXIT_RSM(pVCpu, pCtx); break;
6117 case DBGFEVENT_EXIT_CRX_READ: VBOXVMM_EXIT_CRX_READ(pVCpu, pCtx, (uint8_t)uEventArg); break;
6118 case DBGFEVENT_EXIT_CRX_WRITE: VBOXVMM_EXIT_CRX_WRITE(pVCpu, pCtx, (uint8_t)uEventArg); break;
6119 case DBGFEVENT_EXIT_DRX_READ: VBOXVMM_EXIT_DRX_READ(pVCpu, pCtx, (uint8_t)uEventArg); break;
6120 case DBGFEVENT_EXIT_DRX_WRITE: VBOXVMM_EXIT_DRX_WRITE(pVCpu, pCtx, (uint8_t)uEventArg); break;
6121 case DBGFEVENT_EXIT_RDMSR: VBOXVMM_EXIT_RDMSR(pVCpu, pCtx, pCtx->ecx); break;
6122 case DBGFEVENT_EXIT_WRMSR: VBOXVMM_EXIT_WRMSR(pVCpu, pCtx, pCtx->ecx,
6123 RT_MAKE_U64(pCtx->eax, pCtx->edx)); break;
6124 case DBGFEVENT_EXIT_MWAIT: VBOXVMM_EXIT_MWAIT(pVCpu, pCtx); break;
6125 case DBGFEVENT_EXIT_MONITOR: VBOXVMM_EXIT_MONITOR(pVCpu, pCtx); break;
6126 case DBGFEVENT_EXIT_PAUSE: VBOXVMM_EXIT_PAUSE(pVCpu, pCtx); break;
6127 case DBGFEVENT_EXIT_SGDT: VBOXVMM_EXIT_SGDT(pVCpu, pCtx); break;
6128 case DBGFEVENT_EXIT_SIDT: VBOXVMM_EXIT_SIDT(pVCpu, pCtx); break;
6129 case DBGFEVENT_EXIT_LGDT: VBOXVMM_EXIT_LGDT(pVCpu, pCtx); break;
6130 case DBGFEVENT_EXIT_LIDT: VBOXVMM_EXIT_LIDT(pVCpu, pCtx); break;
6131 case DBGFEVENT_EXIT_SLDT: VBOXVMM_EXIT_SLDT(pVCpu, pCtx); break;
6132 case DBGFEVENT_EXIT_STR: VBOXVMM_EXIT_STR(pVCpu, pCtx); break;
6133 case DBGFEVENT_EXIT_LLDT: VBOXVMM_EXIT_LLDT(pVCpu, pCtx); break;
6134 case DBGFEVENT_EXIT_LTR: VBOXVMM_EXIT_LTR(pVCpu, pCtx); break;
6135 case DBGFEVENT_EXIT_RDTSCP: VBOXVMM_EXIT_RDTSCP(pVCpu, pCtx); break;
6136 case DBGFEVENT_EXIT_WBINVD: VBOXVMM_EXIT_WBINVD(pVCpu, pCtx); break;
6137 case DBGFEVENT_EXIT_XSETBV: VBOXVMM_EXIT_XSETBV(pVCpu, pCtx); break;
6138 case DBGFEVENT_EXIT_VMM_CALL: VBOXVMM_EXIT_VMM_CALL(pVCpu, pCtx); break;
6139 case DBGFEVENT_EXIT_SVM_VMRUN: VBOXVMM_EXIT_SVM_VMRUN(pVCpu, pCtx); break;
6140 case DBGFEVENT_EXIT_SVM_VMLOAD: VBOXVMM_EXIT_SVM_VMLOAD(pVCpu, pCtx); break;
6141 case DBGFEVENT_EXIT_SVM_VMSAVE: VBOXVMM_EXIT_SVM_VMSAVE(pVCpu, pCtx); break;
6142 case DBGFEVENT_EXIT_SVM_STGI: VBOXVMM_EXIT_SVM_STGI(pVCpu, pCtx); break;
6143 case DBGFEVENT_EXIT_SVM_CLGI: VBOXVMM_EXIT_SVM_CLGI(pVCpu, pCtx); break;
6144 default: AssertMsgFailed(("enmEvent2=%d uExitCode=%d\n", enmEvent2, uExitCode)); break;
6145 }
6146 }
6147
6148 /*
6149 * Fire of the DBGF event, if enabled (our check here is just a quick one,
6150 * the DBGF call will do a full check).
6151 *
6152 * Note! DBGF sets DBGFEVENT_INTERRUPT_SOFTWARE in the bitmap.
6153 * Note! If we have to events, we prioritize the first, i.e. the instruction
6154 * one, in order to avoid event nesting.
6155 */
6156 PVMCC pVM = pVCpu->CTX_SUFF(pVM);
6157 VBOXSTRICTRC rcStrict;
6158 if ( enmEvent1 != DBGFEVENT_END
6159 && DBGF_IS_EVENT_ENABLED(pVM, enmEvent1))
6160 {
6161 hmR0SvmImportGuestState(pVCpu, CPUMCTX_EXTRN_CS | CPUMCTX_EXTRN_RIP);
6162 rcStrict = DBGFEventGenericWithArgs(pVM, pVCpu, enmEvent1, DBGFEVENTCTX_HM, 1, uEventArg);
6163 }
6164 else if ( enmEvent2 != DBGFEVENT_END
6165 && DBGF_IS_EVENT_ENABLED(pVM, enmEvent2))
6166 {
6167 hmR0SvmImportGuestState(pVCpu, CPUMCTX_EXTRN_CS | CPUMCTX_EXTRN_RIP);
6168 rcStrict = DBGFEventGenericWithArgs(pVM, pVCpu, enmEvent2, DBGFEVENTCTX_HM, 1, uEventArg);
6169 }
6170 else
6171 rcStrict = VINF_SUCCESS;
6172 return rcStrict;
6173}
6174
6175
6176/**
6177 * Handles a guest \#VMEXIT (for all EXITCODE values except SVM_EXIT_INVALID),
6178 * debug variant.
6179 *
6180 * @returns Strict VBox status code (informational status codes included).
6181 * @param pVCpu The cross context virtual CPU structure.
6182 * @param pSvmTransient Pointer to the SVM transient structure.
6183 * @param pDbgState The runtime debug state.
6184 */
6185static VBOXSTRICTRC hmR0SvmDebugHandleExit(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient, PSVMRUNDBGSTATE pDbgState)
6186{
6187 Assert(pSvmTransient->u64ExitCode != SVM_EXIT_INVALID);
6188 Assert(pSvmTransient->u64ExitCode <= SVM_EXIT_MAX);
6189
6190 /*
6191 * Expensive (saves context) generic dtrace VM-exit probe.
6192 */
6193 uint64_t const uExitCode = pSvmTransient->u64ExitCode;
6194 if (!VBOXVMM_R0_HMSVM_VMEXIT_ENABLED())
6195 { /* more likely */ }
6196 else
6197 {
6198 hmR0SvmImportGuestState(pVCpu, HMSVM_CPUMCTX_EXTRN_ALL);
6199 VBOXVMM_R0_HMSVM_VMEXIT(pVCpu, &pVCpu->cpum.GstCtx, uExitCode, pSvmTransient->pVmcb);
6200 }
6201
6202 /*
6203 * Check for single stepping event if we're stepping.
6204 */
6205 if (pVCpu->hm.s.fSingleInstruction)
6206 {
6207 switch (uExitCode)
6208 {
6209 /* Various events: */
6210 case SVM_EXIT_XCPT_0: case SVM_EXIT_XCPT_1: case SVM_EXIT_XCPT_2: case SVM_EXIT_XCPT_3:
6211 case SVM_EXIT_XCPT_4: case SVM_EXIT_XCPT_5: case SVM_EXIT_XCPT_6: case SVM_EXIT_XCPT_7:
6212 case SVM_EXIT_XCPT_8: case SVM_EXIT_XCPT_9: case SVM_EXIT_XCPT_10: case SVM_EXIT_XCPT_11:
6213 case SVM_EXIT_XCPT_12: case SVM_EXIT_XCPT_13: case SVM_EXIT_XCPT_14: case SVM_EXIT_XCPT_15:
6214 case SVM_EXIT_XCPT_16: case SVM_EXIT_XCPT_17: case SVM_EXIT_XCPT_18: case SVM_EXIT_XCPT_19:
6215 case SVM_EXIT_XCPT_20: case SVM_EXIT_XCPT_21: case SVM_EXIT_XCPT_22: case SVM_EXIT_XCPT_23:
6216 case SVM_EXIT_XCPT_24: case SVM_EXIT_XCPT_25: case SVM_EXIT_XCPT_26: case SVM_EXIT_XCPT_27:
6217 case SVM_EXIT_XCPT_28: case SVM_EXIT_XCPT_29: case SVM_EXIT_XCPT_30: case SVM_EXIT_XCPT_31:
6218 case SVM_EXIT_INTR:
6219 case SVM_EXIT_NMI:
6220 case SVM_EXIT_VINTR:
6221 case SVM_EXIT_NPF:
6222 case SVM_EXIT_AVIC_NOACCEL:
6223
6224 /* Instruction specific VM-exits: */
6225 case SVM_EXIT_READ_CR0: case SVM_EXIT_READ_CR1: case SVM_EXIT_READ_CR2: case SVM_EXIT_READ_CR3:
6226 case SVM_EXIT_READ_CR4: case SVM_EXIT_READ_CR5: case SVM_EXIT_READ_CR6: case SVM_EXIT_READ_CR7:
6227 case SVM_EXIT_READ_CR8: case SVM_EXIT_READ_CR9: case SVM_EXIT_READ_CR10: case SVM_EXIT_READ_CR11:
6228 case SVM_EXIT_READ_CR12: case SVM_EXIT_READ_CR13: case SVM_EXIT_READ_CR14: case SVM_EXIT_READ_CR15:
6229 case SVM_EXIT_WRITE_CR0: case SVM_EXIT_WRITE_CR1: case SVM_EXIT_WRITE_CR2: case SVM_EXIT_WRITE_CR3:
6230 case SVM_EXIT_WRITE_CR4: case SVM_EXIT_WRITE_CR5: case SVM_EXIT_WRITE_CR6: case SVM_EXIT_WRITE_CR7:
6231 case SVM_EXIT_WRITE_CR8: case SVM_EXIT_WRITE_CR9: case SVM_EXIT_WRITE_CR10: case SVM_EXIT_WRITE_CR11:
6232 case SVM_EXIT_WRITE_CR12: case SVM_EXIT_WRITE_CR13: case SVM_EXIT_WRITE_CR14: case SVM_EXIT_WRITE_CR15:
6233 case SVM_EXIT_READ_DR0: case SVM_EXIT_READ_DR1: case SVM_EXIT_READ_DR2: case SVM_EXIT_READ_DR3:
6234 case SVM_EXIT_READ_DR4: case SVM_EXIT_READ_DR5: case SVM_EXIT_READ_DR6: case SVM_EXIT_READ_DR7:
6235 case SVM_EXIT_READ_DR8: case SVM_EXIT_READ_DR9: case SVM_EXIT_READ_DR10: case SVM_EXIT_READ_DR11:
6236 case SVM_EXIT_READ_DR12: case SVM_EXIT_READ_DR13: case SVM_EXIT_READ_DR14: case SVM_EXIT_READ_DR15:
6237 case SVM_EXIT_WRITE_DR0: case SVM_EXIT_WRITE_DR1: case SVM_EXIT_WRITE_DR2: case SVM_EXIT_WRITE_DR3:
6238 case SVM_EXIT_WRITE_DR4: case SVM_EXIT_WRITE_DR5: case SVM_EXIT_WRITE_DR6: case SVM_EXIT_WRITE_DR7:
6239 case SVM_EXIT_WRITE_DR8: case SVM_EXIT_WRITE_DR9: case SVM_EXIT_WRITE_DR10: case SVM_EXIT_WRITE_DR11:
6240 case SVM_EXIT_WRITE_DR12: case SVM_EXIT_WRITE_DR13: case SVM_EXIT_WRITE_DR14: case SVM_EXIT_WRITE_DR15:
6241 case SVM_EXIT_CR0_SEL_WRITE:
6242 case SVM_EXIT_IDTR_READ:
6243 case SVM_EXIT_GDTR_READ:
6244 case SVM_EXIT_LDTR_READ:
6245 case SVM_EXIT_TR_READ:
6246 case SVM_EXIT_IDTR_WRITE:
6247 case SVM_EXIT_GDTR_WRITE:
6248 case SVM_EXIT_LDTR_WRITE:
6249 case SVM_EXIT_TR_WRITE:
6250 case SVM_EXIT_RDTSC:
6251 case SVM_EXIT_RDPMC:
6252 case SVM_EXIT_PUSHF:
6253 case SVM_EXIT_POPF:
6254 case SVM_EXIT_CPUID:
6255 case SVM_EXIT_RSM:
6256 case SVM_EXIT_IRET:
6257 case SVM_EXIT_SWINT:
6258 case SVM_EXIT_INVD:
6259 case SVM_EXIT_PAUSE:
6260 case SVM_EXIT_HLT:
6261 case SVM_EXIT_INVLPG:
6262 case SVM_EXIT_INVLPGA:
6263 case SVM_EXIT_IOIO:
6264 case SVM_EXIT_MSR:
6265 case SVM_EXIT_TASK_SWITCH:
6266 case SVM_EXIT_VMRUN:
6267 case SVM_EXIT_VMMCALL:
6268 case SVM_EXIT_VMLOAD:
6269 case SVM_EXIT_VMSAVE:
6270 case SVM_EXIT_STGI:
6271 case SVM_EXIT_CLGI:
6272 case SVM_EXIT_SKINIT:
6273 case SVM_EXIT_RDTSCP:
6274 case SVM_EXIT_ICEBP:
6275 case SVM_EXIT_WBINVD:
6276 case SVM_EXIT_MONITOR:
6277 case SVM_EXIT_MWAIT:
6278 case SVM_EXIT_MWAIT_ARMED:
6279 case SVM_EXIT_XSETBV:
6280 case SVM_EXIT_RDPRU:
6281 case SVM_EXIT_WRITE_EFER_TRAP:
6282 case SVM_EXIT_WRITE_CR0_TRAP: case SVM_EXIT_WRITE_CR1_TRAP: case SVM_EXIT_WRITE_CR2_TRAP: case SVM_EXIT_WRITE_CR3_TRAP:
6283 case SVM_EXIT_WRITE_CR4_TRAP: case SVM_EXIT_WRITE_CR5_TRAP: case SVM_EXIT_WRITE_CR6_TRAP: case SVM_EXIT_WRITE_CR7_TRAP:
6284 case SVM_EXIT_WRITE_CR8_TRAP: case SVM_EXIT_WRITE_CR9_TRAP: case SVM_EXIT_WRITE_CR10_TRAP: case SVM_EXIT_WRITE_CR11_TRAP:
6285 case SVM_EXIT_WRITE_CR12_TRAP: case SVM_EXIT_WRITE_CR13_TRAP: case SVM_EXIT_WRITE_CR14_TRAP: case SVM_EXIT_WRITE_CR15_TRAP:
6286 case SVM_EXIT_MCOMMIT:
6287 {
6288 hmR0SvmImportGuestState(pVCpu, CPUMCTX_EXTRN_CS | CPUMCTX_EXTRN_RIP);
6289 if ( pVCpu->cpum.GstCtx.rip != pDbgState->uRipStart
6290 || pVCpu->cpum.GstCtx.cs.Sel != pDbgState->uCsStart)
6291 {
6292 Log6Func(("VINF_EM_DBG_STEPPED: %04x:%08RX64 (exit %u)\n",
6293 pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip, uExitCode));
6294 return VINF_EM_DBG_STEPPED;
6295 }
6296 break;
6297 }
6298
6299 /* Errors and unexpected events: */
6300 case SVM_EXIT_FERR_FREEZE:
6301 case SVM_EXIT_SHUTDOWN:
6302 case SVM_EXIT_AVIC_INCOMPLETE_IPI:
6303 break;
6304
6305 case SVM_EXIT_SMI:
6306 case SVM_EXIT_INIT:
6307 default:
6308 AssertMsgFailed(("Unexpected VM-exit=%#x\n", uExitCode));
6309 break;
6310 }
6311 }
6312
6313 /*
6314 * Check for debugger event breakpoints and dtrace probes.
6315 */
6316 if ( uExitCode < sizeof(pDbgState->bmExitsToCheck) * 8U
6317 && ASMBitTest(pDbgState->bmExitsToCheck, uExitCode) )
6318 {
6319 VBOXSTRICTRC rcStrict = hmR0SvmHandleExitDtraceEvents(pVCpu, pSvmTransient, uExitCode);
6320 if (rcStrict != VINF_SUCCESS)
6321 {
6322 Log6Func(("%04x:%08RX64 (exit %u) -> %Rrc\n",
6323 pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip, uExitCode, VBOXSTRICTRC_VAL(rcStrict) ));
6324 return rcStrict;
6325 }
6326 }
6327
6328 /*
6329 * Normal processing.
6330 */
6331 return hmR0SvmHandleExit(pVCpu, pSvmTransient);
6332}
6333
6334
6335/**
6336 * Runs the guest code using AMD-V in single step mode.
6337 *
6338 * @returns Strict VBox status code.
6339 * @param pVCpu The cross context virtual CPU structure.
6340 * @param pcLoops Pointer to the number of executed loops.
6341 */
6342static VBOXSTRICTRC hmR0SvmRunGuestCodeDebug(PVMCPUCC pVCpu, uint32_t *pcLoops)
6343{
6344 uint32_t const cMaxResumeLoops = pVCpu->CTX_SUFF(pVM)->hmr0.s.cMaxResumeLoops;
6345 Assert(pcLoops);
6346 Assert(*pcLoops <= cMaxResumeLoops);
6347
6348 SVMTRANSIENT SvmTransient;
6349 RT_ZERO(SvmTransient);
6350 SvmTransient.fUpdateTscOffsetting = true;
6351 SvmTransient.pVmcb = pVCpu->hmr0.s.svm.pVmcb;
6352
6353 PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
6354
6355 /* Set HMCPU indicators. */
6356 bool const fSavedSingleInstruction = pVCpu->hm.s.fSingleInstruction;
6357 pVCpu->hm.s.fSingleInstruction = pVCpu->hm.s.fSingleInstruction || DBGFIsStepping(pVCpu);
6358 pVCpu->hmr0.s.fDebugWantRdTscExit = false;
6359 pVCpu->hmr0.s.fUsingDebugLoop = true;
6360
6361 /* State we keep to help modify and later restore the VMCS fields we alter, and for detecting steps. */
6362 SVMRUNDBGSTATE DbgState;
6363 hmR0SvmRunDebugStateInit(pVCpu, &SvmTransient, &DbgState);
6364 hmR0SvmPreRunGuestDebugStateUpdate(pVCpu, &SvmTransient, &DbgState);
6365
6366 /*
6367 * The loop.
6368 */
6369 VBOXSTRICTRC rc = VERR_INTERNAL_ERROR_5;
6370 for (;;)
6371 {
6372 Assert(!HMR0SuspendPending());
6373 AssertMsg(pVCpu->hmr0.s.idEnteredCpu == RTMpCpuId(),
6374 ("Illegal migration! Entered on CPU %u Current %u cLoops=%u\n", (unsigned)pVCpu->hmr0.s.idEnteredCpu,
6375 (unsigned)RTMpCpuId(), *pcLoops));
6376 bool fStepping = pVCpu->hm.s.fSingleInstruction;
6377
6378 /* Set up VM-execution controls the next two can respond to. */
6379 hmR0SvmPreRunGuestDebugStateApply(&SvmTransient, &DbgState);
6380
6381 /* Preparatory work for running nested-guest code, this may force us to return to
6382 ring-3. This bugger disables interrupts on VINF_SUCCESS! */
6383 STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
6384 rc = hmR0SvmPreRunGuest(pVCpu, &SvmTransient);
6385 if (rc != VINF_SUCCESS)
6386 break;
6387
6388 /*
6389 * No longjmps to ring-3 from this point on!!!
6390 *
6391 * Asserts() will still longjmp to ring-3 (but won't return), which is intentional,
6392 * better than a kernel panic. This also disables flushing of the R0-logger instance.
6393 */
6394 hmR0SvmPreRunGuestCommitted(pVCpu, &SvmTransient);
6395
6396 /* Override any obnoxious code in the above two calls. */
6397 hmR0SvmPreRunGuestDebugStateApply(&SvmTransient, &DbgState);
6398#if 0
6399 Log(("%04x:%08RX64 ds=%04x %04x:%08RX64 i=%#RX64\n",
6400 SvmTransient.pVmcb->guest.CS.u16Sel, SvmTransient.pVmcb->guest.u64RIP, SvmTransient.pVmcb->guest.DS.u16Sel,
6401 SvmTransient.pVmcb->guest.SS.u16Sel, SvmTransient.pVmcb->guest.u64RSP, SvmTransient.pVmcb->ctrl.EventInject.u));
6402#endif
6403
6404 /*
6405 * Finally execute guest code.
6406 */
6407 rc = hmR0SvmRunGuest(pVCpu, pVCpu->hmr0.s.svm.HCPhysVmcb);
6408
6409 /* Restore any residual host-state and save any bits shared between host and guest
6410 into the guest-CPU state. Re-enables interrupts! */
6411 hmR0SvmPostRunGuest(pVCpu, &SvmTransient, rc);
6412#if 0
6413 Log(("%04x:%08RX64 ds=%04x %04x:%08RX64 i=%#RX64 exit=%d\n",
6414 SvmTransient.pVmcb->guest.CS.u16Sel, SvmTransient.pVmcb->guest.u64RIP, SvmTransient.pVmcb->guest.DS.u16Sel,
6415 SvmTransient.pVmcb->guest.SS.u16Sel, SvmTransient.pVmcb->guest.u64RSP, SvmTransient.pVmcb->ctrl.EventInject.u, SvmTransient.u64ExitCode));
6416#endif
6417
6418 if (RT_LIKELY( rc == VINF_SUCCESS /* Check for VMRUN errors. */
6419 && SvmTransient.u64ExitCode != SVM_EXIT_INVALID)) /* Check for invalid guest-state errors. */
6420 { /* very likely*/ }
6421 else
6422 {
6423 if (rc == VINF_SUCCESS)
6424 rc = VERR_SVM_INVALID_GUEST_STATE;
6425 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatPreExit, x);
6426 hmR0SvmReportWorldSwitchError(pVCpu, VBOXSTRICTRC_VAL(rc));
6427 return rc;
6428 }
6429
6430 /* Handle the #VMEXIT. */
6431 HMSVM_DEBUG_EXITCODE_STAM_COUNTER_INC(SvmTransient.u64ExitCode);
6432 STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatPreExit, &pVCpu->hm.s.StatExitHandling, x);
6433 VBOXVMM_R0_HMSVM_VMEXIT(pVCpu, pCtx, SvmTransient.u64ExitCode, pVCpu->hmr0.s.svm.pVmcb);
6434 rc = hmR0SvmDebugHandleExit(pVCpu, &SvmTransient, &DbgState);
6435 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitHandling, x);
6436 if (rc != VINF_SUCCESS)
6437 break;
6438 if (++(*pcLoops) >= cMaxResumeLoops)
6439 {
6440 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchMaxResumeLoops);
6441 rc = VINF_EM_RAW_INTERRUPT;
6442 break;
6443 }
6444
6445 /*
6446 * Stepping: Did the RIP change, if so, consider it a single step.
6447 * Otherwise, make sure one of the TFs gets set.
6448 */
6449 if (fStepping)
6450 {
6451 hmR0SvmImportGuestState(pVCpu, CPUMCTX_EXTRN_CS | CPUMCTX_EXTRN_RIP);
6452 if ( pVCpu->cpum.GstCtx.rip != DbgState.uRipStart
6453 || pVCpu->cpum.GstCtx.cs.Sel != DbgState.uCsStart)
6454 {
6455 Log6Func(("VINF_EM_DBG_STEPPED: %04x:%08RX64 (exit %u)\n",
6456 pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip, SvmTransient.u64ExitCode));
6457 rc = VINF_EM_DBG_STEPPED;
6458 break;
6459 }
6460 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_DR7);
6461 }
6462
6463 /*
6464 * Update when dtrace settings changes (DBGF kicks us, so no need to check).
6465 * Revert the state changes afterware so we can drop intercepts no longer needed.
6466 */
6467 if (VBOXVMM_GET_SETTINGS_SEQ_NO() != DbgState.uDtraceSettingsSeqNo)
6468 {
6469 hmR0SvmPreRunGuestDebugStateUpdate(pVCpu, &SvmTransient, &DbgState);
6470 hmR0SvmRunDebugStateRevert(&SvmTransient, &DbgState);
6471 }
6472 }
6473
6474 /*
6475 * Clear the X86_EFL_TF if necessary.
6476 */
6477 if (pVCpu->hmr0.s.fClearTrapFlag)
6478 {
6479 pVCpu->hmr0.s.fClearTrapFlag = false;
6480 pCtx->eflags.Bits.u1TF = 0;
6481 }
6482
6483 /* Restore HMCPU indicators. */
6484 pVCpu->hmr0.s.fUsingDebugLoop = false;
6485 pVCpu->hmr0.s.fDebugWantRdTscExit = false;
6486 pVCpu->hm.s.fSingleInstruction = fSavedSingleInstruction;
6487
6488 /* Restore all controls applied by hmR0SvmPreRunGuestDebugStateApply above. */
6489 hmR0SvmRunDebugStateRevert(&SvmTransient, &DbgState);
6490
6491 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
6492 return rc;
6493}
6494
6495/** @} */
6496
6497#undef VMEXIT_CALL_RET
6498
6499
6500#ifdef VBOX_STRICT
6501/* Is there some generic IPRT define for this that are not in Runtime/internal/\* ?? */
6502# define HMSVM_ASSERT_PREEMPT_CPUID_VAR() \
6503 RTCPUID const idAssertCpu = RTThreadPreemptIsEnabled(NIL_RTTHREAD) ? NIL_RTCPUID : RTMpCpuId()
6504
6505# define HMSVM_ASSERT_PREEMPT_CPUID() \
6506 do \
6507 { \
6508 RTCPUID const idAssertCpuNow = RTThreadPreemptIsEnabled(NIL_RTTHREAD) ? NIL_RTCPUID : RTMpCpuId(); \
6509 AssertMsg(idAssertCpu == idAssertCpuNow, ("SVM %#x, %#x\n", idAssertCpu, idAssertCpuNow)); \
6510 } while (0)
6511
6512# define HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(a_pVCpu, a_pSvmTransient) \
6513 do { \
6514 AssertPtr((a_pVCpu)); \
6515 AssertPtr((a_pSvmTransient)); \
6516 Assert(ASMIntAreEnabled()); \
6517 HMSVM_ASSERT_PREEMPT_SAFE((a_pVCpu)); \
6518 HMSVM_ASSERT_PREEMPT_CPUID_VAR(); \
6519 Log4Func(("vcpu[%u] -v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-\n", (a_pVCpu)->idCpu)); \
6520 HMSVM_ASSERT_PREEMPT_SAFE((a_pVCpu)); \
6521 if (!VMMRZCallRing3IsEnabled((a_pVCpu))) \
6522 HMSVM_ASSERT_PREEMPT_CPUID(); \
6523 } while (0)
6524#else
6525# define HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(a_pVCpu, a_pSvmTransient) \
6526 do { \
6527 RT_NOREF2(a_pVCpu, a_pSvmTransient); \
6528 } while (0)
6529#endif
6530
6531
6532/**
6533 * Gets the IEM exception flags for the specified SVM event.
6534 *
6535 * @returns The IEM exception flags.
6536 * @param pEvent Pointer to the SVM event.
6537 *
6538 * @remarks This function currently only constructs flags required for
6539 * IEMEvaluateRecursiveXcpt and not the complete flags (e.g. error-code
6540 * and CR2 aspects of an exception are not included).
6541 */
6542static uint32_t hmR0SvmGetIemXcptFlags(PCSVMEVENT pEvent)
6543{
6544 uint8_t const uEventType = pEvent->n.u3Type;
6545 uint32_t fIemXcptFlags;
6546 switch (uEventType)
6547 {
6548 case SVM_EVENT_EXCEPTION:
6549 /*
6550 * Only INT3 and INTO instructions can raise #BP and #OF exceptions.
6551 * See AMD spec. Table 8-1. "Interrupt Vector Source and Cause".
6552 */
6553 if (pEvent->n.u8Vector == X86_XCPT_BP)
6554 {
6555 fIemXcptFlags = IEM_XCPT_FLAGS_T_SOFT_INT | IEM_XCPT_FLAGS_BP_INSTR;
6556 break;
6557 }
6558 if (pEvent->n.u8Vector == X86_XCPT_OF)
6559 {
6560 fIemXcptFlags = IEM_XCPT_FLAGS_T_SOFT_INT | IEM_XCPT_FLAGS_OF_INSTR;
6561 break;
6562 }
6563 /** @todo How do we distinguish ICEBP \#DB from the regular one? */
6564 RT_FALL_THRU();
6565 case SVM_EVENT_NMI:
6566 fIemXcptFlags = IEM_XCPT_FLAGS_T_CPU_XCPT;
6567 break;
6568
6569 case SVM_EVENT_EXTERNAL_IRQ:
6570 fIemXcptFlags = IEM_XCPT_FLAGS_T_EXT_INT;
6571 break;
6572
6573 case SVM_EVENT_SOFTWARE_INT:
6574 fIemXcptFlags = IEM_XCPT_FLAGS_T_SOFT_INT;
6575 break;
6576
6577 default:
6578 fIemXcptFlags = 0;
6579 AssertMsgFailed(("Unexpected event type! uEventType=%#x uVector=%#x", uEventType, pEvent->n.u8Vector));
6580 break;
6581 }
6582 return fIemXcptFlags;
6583}
6584
6585
6586/**
6587 * Handle a condition that occurred while delivering an event through the guest
6588 * IDT.
6589 *
6590 * @returns VBox status code (informational error codes included).
6591 * @retval VINF_SUCCESS if we should continue handling the \#VMEXIT.
6592 * @retval VINF_HM_DOUBLE_FAULT if a \#DF condition was detected and we ought to
6593 * continue execution of the guest which will delivery the \#DF.
6594 * @retval VINF_EM_RESET if we detected a triple-fault condition.
6595 * @retval VERR_EM_GUEST_CPU_HANG if we detected a guest CPU hang.
6596 *
6597 * @param pVCpu The cross context virtual CPU structure.
6598 * @param pSvmTransient Pointer to the SVM transient structure.
6599 *
6600 * @remarks No-long-jump zone!!!
6601 */
6602static int hmR0SvmCheckExitDueToEventDelivery(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
6603{
6604 /** @todo r=bird: Looks like this is called on many exits and we start by
6605 * loading CR2 on the offchance that we actually have work to do here.
6606 *
6607 * HMSVM_CHECK_EXIT_DUE_TO_EVENT_DELIVERY can surely check
6608 * pVmcb->ctrl.ExitIntInfo.n.u1Valid, can't it?
6609 *
6610 * Also, what's the deal with hmR0SvmGetCurrentVmcb() vs pSvmTransient->pVmcb?
6611 */
6612 int rc = VINF_SUCCESS;
6613 PSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
6614 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, CPUMCTX_EXTRN_CR2);
6615
6616 Log4(("EXITINTINFO: Pending vectoring event %#RX64 Valid=%RTbool ErrValid=%RTbool Err=%#RX32 Type=%u Vector=%u\n",
6617 pVmcb->ctrl.ExitIntInfo.u, !!pVmcb->ctrl.ExitIntInfo.n.u1Valid, !!pVmcb->ctrl.ExitIntInfo.n.u1ErrorCodeValid,
6618 pVmcb->ctrl.ExitIntInfo.n.u32ErrorCode, pVmcb->ctrl.ExitIntInfo.n.u3Type, pVmcb->ctrl.ExitIntInfo.n.u8Vector));
6619
6620 /*
6621 * The EXITINTINFO (if valid) contains the prior exception (IDT vector) that was trying to
6622 * be delivered to the guest which caused a #VMEXIT which was intercepted (Exit vector).
6623 *
6624 * See AMD spec. 15.7.3 "EXITINFO Pseudo-Code".
6625 */
6626 if (pVmcb->ctrl.ExitIntInfo.n.u1Valid)
6627 {
6628 IEMXCPTRAISE enmRaise;
6629 IEMXCPTRAISEINFO fRaiseInfo;
6630 bool const fExitIsHwXcpt = pSvmTransient->u64ExitCode - SVM_EXIT_XCPT_0 <= SVM_EXIT_XCPT_31;
6631 uint8_t const uIdtVector = pVmcb->ctrl.ExitIntInfo.n.u8Vector;
6632 if (fExitIsHwXcpt)
6633 {
6634 uint8_t const uExitVector = pSvmTransient->u64ExitCode - SVM_EXIT_XCPT_0;
6635 uint32_t const fIdtVectorFlags = hmR0SvmGetIemXcptFlags(&pVmcb->ctrl.ExitIntInfo);
6636 uint32_t const fExitVectorFlags = IEM_XCPT_FLAGS_T_CPU_XCPT;
6637 enmRaise = IEMEvaluateRecursiveXcpt(pVCpu, fIdtVectorFlags, uIdtVector, fExitVectorFlags, uExitVector, &fRaiseInfo);
6638 }
6639 else
6640 {
6641 /*
6642 * If delivery of an event caused a #VMEXIT that is not an exception (e.g. #NPF)
6643 * then we end up here.
6644 *
6645 * If the event was:
6646 * - a software interrupt, we can re-execute the instruction which will
6647 * regenerate the event.
6648 * - an NMI, we need to clear NMI blocking and re-inject the NMI.
6649 * - a hardware exception or external interrupt, we re-inject it.
6650 */
6651 fRaiseInfo = IEMXCPTRAISEINFO_NONE;
6652 if (pVmcb->ctrl.ExitIntInfo.n.u3Type == SVM_EVENT_SOFTWARE_INT)
6653 enmRaise = IEMXCPTRAISE_REEXEC_INSTR;
6654 else
6655 enmRaise = IEMXCPTRAISE_PREV_EVENT;
6656 }
6657
6658 switch (enmRaise)
6659 {
6660 case IEMXCPTRAISE_CURRENT_XCPT:
6661 case IEMXCPTRAISE_PREV_EVENT:
6662 {
6663 /* For software interrupts, we shall re-execute the instruction. */
6664 if (!(fRaiseInfo & IEMXCPTRAISEINFO_SOFT_INT_XCPT))
6665 {
6666 RTGCUINTPTR GCPtrFaultAddress = 0;
6667
6668 /* If we are re-injecting an NMI, clear NMI blocking. */
6669 if (pVmcb->ctrl.ExitIntInfo.n.u3Type == SVM_EVENT_NMI)
6670 CPUMClearInterruptInhibitingByNmi(&pVCpu->cpum.GstCtx);
6671
6672 /* Determine a vectoring #PF condition, see comment in hmR0SvmExitXcptPF(). */
6673 if (fRaiseInfo & (IEMXCPTRAISEINFO_EXT_INT_PF | IEMXCPTRAISEINFO_NMI_PF))
6674 {
6675 pSvmTransient->fVectoringPF = true;
6676 Log4Func(("IDT: Pending vectoring #PF due to delivery of Ext-Int/NMI. uCR2=%#RX64\n",
6677 pVCpu->cpum.GstCtx.cr2));
6678 }
6679 else if ( pVmcb->ctrl.ExitIntInfo.n.u3Type == SVM_EVENT_EXCEPTION
6680 && uIdtVector == X86_XCPT_PF)
6681 {
6682 /*
6683 * If the previous exception was a #PF, we need to recover the CR2 value.
6684 * This can't happen with shadow paging.
6685 */
6686 GCPtrFaultAddress = pVCpu->cpum.GstCtx.cr2;
6687 }
6688
6689 /*
6690 * Without nested paging, when uExitVector is #PF, CR2 value will be updated from the VMCB's
6691 * exit info. fields, if it's a guest #PF, see hmR0SvmExitXcptPF().
6692 */
6693 Assert(pVmcb->ctrl.ExitIntInfo.n.u3Type != SVM_EVENT_SOFTWARE_INT);
6694 STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectReflect);
6695 hmR0SvmSetPendingEvent(pVCpu, &pVmcb->ctrl.ExitIntInfo, GCPtrFaultAddress);
6696
6697 Log4Func(("IDT: Pending vectoring event %#RX64 ErrValid=%RTbool Err=%#RX32 GCPtrFaultAddress=%#RX64\n",
6698 pVmcb->ctrl.ExitIntInfo.u, RT_BOOL(pVmcb->ctrl.ExitIntInfo.n.u1ErrorCodeValid),
6699 pVmcb->ctrl.ExitIntInfo.n.u32ErrorCode, GCPtrFaultAddress));
6700 }
6701 break;
6702 }
6703
6704 case IEMXCPTRAISE_REEXEC_INSTR:
6705 {
6706 Assert(rc == VINF_SUCCESS);
6707 break;
6708 }
6709
6710 case IEMXCPTRAISE_DOUBLE_FAULT:
6711 {
6712 /*
6713 * Determing a vectoring double #PF condition. Used later, when PGM evaluates
6714 * the second #PF as a guest #PF (and not a shadow #PF) and needs to be
6715 * converted into a #DF.
6716 */
6717 if (fRaiseInfo & IEMXCPTRAISEINFO_PF_PF)
6718 {
6719 Log4Func(("IDT: Pending vectoring double #PF uCR2=%#RX64\n", pVCpu->cpum.GstCtx.cr2));
6720 pSvmTransient->fVectoringDoublePF = true;
6721 Assert(rc == VINF_SUCCESS);
6722 }
6723 else
6724 {
6725 STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectConvertDF);
6726 hmR0SvmSetPendingXcptDF(pVCpu);
6727 rc = VINF_HM_DOUBLE_FAULT;
6728 }
6729 break;
6730 }
6731
6732 case IEMXCPTRAISE_TRIPLE_FAULT:
6733 {
6734 rc = VINF_EM_RESET;
6735 break;
6736 }
6737
6738 case IEMXCPTRAISE_CPU_HANG:
6739 {
6740 rc = VERR_EM_GUEST_CPU_HANG;
6741 break;
6742 }
6743
6744 default:
6745 AssertMsgFailedBreakStmt(("Bogus enmRaise value: %d (%#x)\n", enmRaise, enmRaise), rc = VERR_SVM_IPE_2);
6746 }
6747 }
6748 Assert(rc == VINF_SUCCESS || rc == VINF_HM_DOUBLE_FAULT || rc == VINF_EM_RESET || rc == VERR_EM_GUEST_CPU_HANG);
6749 return rc;
6750}
6751
6752
6753/**
6754 * Advances the guest RIP by the number of bytes specified in @a cb.
6755 *
6756 * @param pVCpu The cross context virtual CPU structure.
6757 * @param cb RIP increment value in bytes.
6758 */
6759DECLINLINE(void) hmR0SvmAdvanceRip(PVMCPUCC pVCpu, uint32_t cb)
6760{
6761 pVCpu->cpum.GstCtx.rip += cb;
6762 CPUMClearInterruptShadow(&pVCpu->cpum.GstCtx);
6763 /** @todo clear RF. */
6764}
6765
6766
6767/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
6768/* -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- #VMEXIT handlers -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- */
6769/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
6770
6771/** @name \#VMEXIT handlers.
6772 * @{
6773 */
6774
6775/**
6776 * \#VMEXIT handler for external interrupts, NMIs, FPU assertion freeze and INIT
6777 * signals (SVM_EXIT_INTR, SVM_EXIT_NMI, SVM_EXIT_FERR_FREEZE, SVM_EXIT_INIT).
6778 */
6779HMSVM_EXIT_DECL hmR0SvmExitIntr(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
6780{
6781 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
6782
6783 if (pSvmTransient->u64ExitCode == SVM_EXIT_NMI)
6784 STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatExitHostNmiInGC);
6785 else if (pSvmTransient->u64ExitCode == SVM_EXIT_INTR)
6786 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitExtInt);
6787
6788 /*
6789 * AMD-V has no preemption timer and the generic periodic preemption timer has no way to
6790 * signal -before- the timer fires if the current interrupt is our own timer or a some
6791 * other host interrupt. We also cannot examine what interrupt it is until the host
6792 * actually take the interrupt.
6793 *
6794 * Going back to executing guest code here unconditionally causes random scheduling
6795 * problems (observed on an AMD Phenom 9850 Quad-Core on Windows 64-bit host).
6796 */
6797 return VINF_EM_RAW_INTERRUPT;
6798}
6799
6800
6801/**
6802 * \#VMEXIT handler for WBINVD (SVM_EXIT_WBINVD). Conditional \#VMEXIT.
6803 */
6804HMSVM_EXIT_DECL hmR0SvmExitWbinvd(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
6805{
6806 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
6807
6808 VBOXSTRICTRC rcStrict;
6809 bool const fSupportsNextRipSave = hmR0SvmSupportsNextRipSave(pVCpu);
6810 if (fSupportsNextRipSave)
6811 {
6812 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK);
6813 PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
6814 uint8_t const cbInstr = pVmcb->ctrl.u64NextRIP - pVCpu->cpum.GstCtx.rip;
6815 rcStrict = IEMExecDecodedWbinvd(pVCpu, cbInstr);
6816 }
6817 else
6818 {
6819 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_MUST_MASK);
6820 rcStrict = IEMExecOne(pVCpu);
6821 }
6822
6823 if (rcStrict == VINF_IEM_RAISED_XCPT)
6824 {
6825 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
6826 rcStrict = VINF_SUCCESS;
6827 }
6828 HMSVM_CHECK_SINGLE_STEP(pVCpu, rcStrict);
6829 return rcStrict;
6830}
6831
6832
6833/**
6834 * \#VMEXIT handler for INVD (SVM_EXIT_INVD). Unconditional \#VMEXIT.
6835 */
6836HMSVM_EXIT_DECL hmR0SvmExitInvd(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
6837{
6838 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
6839
6840 VBOXSTRICTRC rcStrict;
6841 bool const fSupportsNextRipSave = hmR0SvmSupportsNextRipSave(pVCpu);
6842 if (fSupportsNextRipSave)
6843 {
6844 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK);
6845 PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
6846 uint8_t const cbInstr = pVmcb->ctrl.u64NextRIP - pVCpu->cpum.GstCtx.rip;
6847 rcStrict = IEMExecDecodedInvd(pVCpu, cbInstr);
6848 }
6849 else
6850 {
6851 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_MUST_MASK);
6852 rcStrict = IEMExecOne(pVCpu);
6853 }
6854
6855 if (rcStrict == VINF_IEM_RAISED_XCPT)
6856 {
6857 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
6858 rcStrict = VINF_SUCCESS;
6859 }
6860 HMSVM_CHECK_SINGLE_STEP(pVCpu, rcStrict);
6861 return rcStrict;
6862}
6863
6864
6865/**
6866 * \#VMEXIT handler for INVD (SVM_EXIT_CPUID). Conditional \#VMEXIT.
6867 */
6868HMSVM_EXIT_DECL hmR0SvmExitCpuid(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
6869{
6870 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
6871
6872 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK | CPUMCTX_EXTRN_RAX | CPUMCTX_EXTRN_RCX);
6873 VBOXSTRICTRC rcStrict;
6874 PCEMEXITREC pExitRec = EMHistoryUpdateFlagsAndTypeAndPC(pVCpu,
6875 EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM | EMEXIT_F_HM, EMEXITTYPE_CPUID),
6876 pVCpu->cpum.GstCtx.rip + pVCpu->cpum.GstCtx.cs.u64Base);
6877 if (!pExitRec)
6878 {
6879 bool const fSupportsNextRipSave = hmR0SvmSupportsNextRipSave(pVCpu);
6880 if (fSupportsNextRipSave)
6881 {
6882 PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
6883 uint8_t const cbInstr = pVmcb->ctrl.u64NextRIP - pVCpu->cpum.GstCtx.rip;
6884 rcStrict = IEMExecDecodedCpuid(pVCpu, cbInstr);
6885 }
6886 else
6887 {
6888 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_MUST_MASK);
6889 rcStrict = IEMExecOne(pVCpu);
6890 }
6891
6892 if (rcStrict == VINF_IEM_RAISED_XCPT)
6893 {
6894 CPUM_ASSERT_NOT_EXTRN(pVCpu, IEM_CPUMCTX_EXTRN_XCPT_MASK);
6895 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
6896 rcStrict = VINF_SUCCESS;
6897 }
6898 HMSVM_CHECK_SINGLE_STEP(pVCpu, rcStrict);
6899 }
6900 else
6901 {
6902 /*
6903 * Frequent exit or something needing probing. Get state and call EMHistoryExec.
6904 */
6905 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_MUST_MASK);
6906
6907 Log4(("CpuIdExit/%u: %04x:%08RX64: %#x/%#x -> EMHistoryExec\n",
6908 pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip, pVCpu->cpum.GstCtx.eax, pVCpu->cpum.GstCtx.ecx));
6909
6910 rcStrict = EMHistoryExec(pVCpu, pExitRec, 0);
6911
6912 Log4(("CpuIdExit/%u: %04x:%08RX64: EMHistoryExec -> %Rrc + %04x:%08RX64\n",
6913 pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
6914 VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip));
6915 }
6916 return rcStrict;
6917}
6918
6919
6920/**
6921 * \#VMEXIT handler for RDTSC (SVM_EXIT_RDTSC). Conditional \#VMEXIT.
6922 */
6923HMSVM_EXIT_DECL hmR0SvmExitRdtsc(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
6924{
6925 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
6926
6927 VBOXSTRICTRC rcStrict;
6928 bool const fSupportsNextRipSave = hmR0SvmSupportsNextRipSave(pVCpu);
6929 if (fSupportsNextRipSave)
6930 {
6931 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK | CPUMCTX_EXTRN_CR4);
6932 PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
6933 uint8_t const cbInstr = pVmcb->ctrl.u64NextRIP - pVCpu->cpum.GstCtx.rip;
6934 rcStrict = IEMExecDecodedRdtsc(pVCpu, cbInstr);
6935 }
6936 else
6937 {
6938 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_MUST_MASK);
6939 rcStrict = IEMExecOne(pVCpu);
6940 }
6941
6942 if (rcStrict == VINF_SUCCESS)
6943 pSvmTransient->fUpdateTscOffsetting = true;
6944 else if (rcStrict == VINF_IEM_RAISED_XCPT)
6945 {
6946 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
6947 rcStrict = VINF_SUCCESS;
6948 }
6949 HMSVM_CHECK_SINGLE_STEP(pVCpu, rcStrict);
6950 return rcStrict;
6951}
6952
6953
6954/**
6955 * \#VMEXIT handler for RDTSCP (SVM_EXIT_RDTSCP). Conditional \#VMEXIT.
6956 */
6957HMSVM_EXIT_DECL hmR0SvmExitRdtscp(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
6958{
6959 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
6960
6961 VBOXSTRICTRC rcStrict;
6962 bool const fSupportsNextRipSave = hmR0SvmSupportsNextRipSave(pVCpu);
6963 if (fSupportsNextRipSave)
6964 {
6965 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK | CPUMCTX_EXTRN_CR4 | CPUMCTX_EXTRN_TSC_AUX);
6966 PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
6967 uint8_t const cbInstr = pVmcb->ctrl.u64NextRIP - pVCpu->cpum.GstCtx.rip;
6968 rcStrict = IEMExecDecodedRdtscp(pVCpu, cbInstr);
6969 }
6970 else
6971 {
6972 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_MUST_MASK);
6973 rcStrict = IEMExecOne(pVCpu);
6974 }
6975
6976 if (rcStrict == VINF_SUCCESS)
6977 pSvmTransient->fUpdateTscOffsetting = true;
6978 else if (rcStrict == VINF_IEM_RAISED_XCPT)
6979 {
6980 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
6981 rcStrict = VINF_SUCCESS;
6982 }
6983 HMSVM_CHECK_SINGLE_STEP(pVCpu, rcStrict);
6984 return rcStrict;
6985}
6986
6987
6988/**
6989 * \#VMEXIT handler for RDPMC (SVM_EXIT_RDPMC). Conditional \#VMEXIT.
6990 */
6991HMSVM_EXIT_DECL hmR0SvmExitRdpmc(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
6992{
6993 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
6994
6995 VBOXSTRICTRC rcStrict;
6996 bool const fSupportsNextRipSave = hmR0SvmSupportsNextRipSave(pVCpu);
6997 if (fSupportsNextRipSave)
6998 {
6999 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK | CPUMCTX_EXTRN_CR4);
7000 PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
7001 uint8_t const cbInstr = pVmcb->ctrl.u64NextRIP - pVCpu->cpum.GstCtx.rip;
7002 rcStrict = IEMExecDecodedRdpmc(pVCpu, cbInstr);
7003 }
7004 else
7005 {
7006 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_MUST_MASK);
7007 rcStrict = IEMExecOne(pVCpu);
7008 }
7009
7010 if (rcStrict == VINF_IEM_RAISED_XCPT)
7011 {
7012 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
7013 rcStrict = VINF_SUCCESS;
7014 }
7015 HMSVM_CHECK_SINGLE_STEP(pVCpu, rcStrict);
7016 return rcStrict;
7017}
7018
7019
7020/**
7021 * \#VMEXIT handler for INVLPG (SVM_EXIT_INVLPG). Conditional \#VMEXIT.
7022 */
7023HMSVM_EXIT_DECL hmR0SvmExitInvlpg(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
7024{
7025 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
7026 Assert(!pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging);
7027
7028 VBOXSTRICTRC rcStrict;
7029 bool const fSupportsDecodeAssists = hmR0SvmSupportsDecodeAssists(pVCpu);
7030 bool const fSupportsNextRipSave = hmR0SvmSupportsNextRipSave(pVCpu);
7031 if ( fSupportsDecodeAssists
7032 && fSupportsNextRipSave)
7033 {
7034 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK);
7035 PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
7036 uint8_t const cbInstr = pVmcb->ctrl.u64NextRIP - pVCpu->cpum.GstCtx.rip;
7037 RTGCPTR const GCPtrPage = pVmcb->ctrl.u64ExitInfo1;
7038 rcStrict = IEMExecDecodedInvlpg(pVCpu, cbInstr, GCPtrPage);
7039 }
7040 else
7041 {
7042 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_MUST_MASK);
7043 rcStrict = IEMExecOne(pVCpu);
7044 }
7045
7046 if (rcStrict == VINF_IEM_RAISED_XCPT)
7047 {
7048 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
7049 rcStrict = VINF_SUCCESS;
7050 }
7051 HMSVM_CHECK_SINGLE_STEP(pVCpu, rcStrict);
7052 return VBOXSTRICTRC_VAL(rcStrict);
7053}
7054
7055
7056/**
7057 * \#VMEXIT handler for HLT (SVM_EXIT_HLT). Conditional \#VMEXIT.
7058 */
7059HMSVM_EXIT_DECL hmR0SvmExitHlt(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
7060{
7061 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
7062
7063 VBOXSTRICTRC rcStrict;
7064 bool const fSupportsNextRipSave = hmR0SvmSupportsNextRipSave(pVCpu);
7065 if (fSupportsNextRipSave)
7066 {
7067 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK);
7068 PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
7069 uint8_t const cbInstr = pVmcb->ctrl.u64NextRIP - pVCpu->cpum.GstCtx.rip;
7070 rcStrict = IEMExecDecodedHlt(pVCpu, cbInstr);
7071 }
7072 else
7073 {
7074 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_MUST_MASK);
7075 rcStrict = IEMExecOne(pVCpu);
7076 }
7077
7078 if ( rcStrict == VINF_EM_HALT
7079 || rcStrict == VINF_SUCCESS)
7080 rcStrict = EMShouldContinueAfterHalt(pVCpu, &pVCpu->cpum.GstCtx) ? VINF_SUCCESS : VINF_EM_HALT;
7081 else if (rcStrict == VINF_IEM_RAISED_XCPT)
7082 {
7083 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
7084 rcStrict = VINF_SUCCESS;
7085 }
7086 HMSVM_CHECK_SINGLE_STEP(pVCpu, rcStrict);
7087 if (rcStrict != VINF_SUCCESS)
7088 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHltToR3);
7089 return VBOXSTRICTRC_VAL(rcStrict);;
7090}
7091
7092
7093/**
7094 * \#VMEXIT handler for MONITOR (SVM_EXIT_MONITOR). Conditional \#VMEXIT.
7095 */
7096HMSVM_EXIT_DECL hmR0SvmExitMonitor(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
7097{
7098 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
7099
7100 /*
7101 * If the instruction length is supplied by the CPU is 3 bytes, we can be certain that no
7102 * segment override prefix is present (and thus use the default segment DS). Otherwise, a
7103 * segment override prefix or other prefixes might be used, in which case we fallback to
7104 * IEMExecOne() to figure out.
7105 */
7106 VBOXSTRICTRC rcStrict;
7107 PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
7108 uint8_t const cbInstr = hmR0SvmSupportsNextRipSave(pVCpu) ? pVmcb->ctrl.u64NextRIP - pVCpu->cpum.GstCtx.rip : 0;
7109 if (cbInstr)
7110 {
7111 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK | CPUMCTX_EXTRN_DS);
7112 rcStrict = IEMExecDecodedMonitor(pVCpu, cbInstr);
7113 }
7114 else
7115 {
7116 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_MUST_MASK);
7117 rcStrict = IEMExecOne(pVCpu);
7118 }
7119
7120 if (rcStrict == VINF_IEM_RAISED_XCPT)
7121 {
7122 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
7123 rcStrict = VINF_SUCCESS;
7124 }
7125 HMSVM_CHECK_SINGLE_STEP(pVCpu, rcStrict);
7126 return rcStrict;
7127}
7128
7129
7130/**
7131 * \#VMEXIT handler for MWAIT (SVM_EXIT_MWAIT). Conditional \#VMEXIT.
7132 */
7133HMSVM_EXIT_DECL hmR0SvmExitMwait(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
7134{
7135 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
7136
7137 VBOXSTRICTRC rcStrict;
7138 bool const fSupportsNextRipSave = hmR0SvmSupportsNextRipSave(pVCpu);
7139 if (fSupportsNextRipSave)
7140 {
7141 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK);
7142 PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
7143 uint8_t const cbInstr = pVmcb->ctrl.u64NextRIP - pVCpu->cpum.GstCtx.rip;
7144 rcStrict = IEMExecDecodedMwait(pVCpu, cbInstr);
7145 }
7146 else
7147 {
7148 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_MUST_MASK);
7149 rcStrict = IEMExecOne(pVCpu);
7150 }
7151
7152 if ( rcStrict == VINF_EM_HALT
7153 && EMMonitorWaitShouldContinue(pVCpu, &pVCpu->cpum.GstCtx))
7154 rcStrict = VINF_SUCCESS;
7155 else if (rcStrict == VINF_IEM_RAISED_XCPT)
7156 {
7157 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
7158 rcStrict = VINF_SUCCESS;
7159 }
7160 HMSVM_CHECK_SINGLE_STEP(pVCpu, rcStrict);
7161 return rcStrict;
7162}
7163
7164
7165/**
7166 * \#VMEXIT handler for shutdown (triple-fault) (SVM_EXIT_SHUTDOWN). Conditional
7167 * \#VMEXIT.
7168 */
7169HMSVM_EXIT_DECL hmR0SvmExitShutdown(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
7170{
7171 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
7172 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, HMSVM_CPUMCTX_EXTRN_ALL);
7173 return VINF_EM_RESET;
7174}
7175
7176
7177/**
7178 * \#VMEXIT handler for unexpected exits. Conditional \#VMEXIT.
7179 */
7180HMSVM_EXIT_DECL hmR0SvmExitUnexpected(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
7181{
7182 PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
7183 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, HMSVM_CPUMCTX_EXTRN_ALL);
7184 AssertMsgFailed(("hmR0SvmExitUnexpected: ExitCode=%#RX64 uExitInfo1=%#RX64 uExitInfo2=%#RX64\n", pSvmTransient->u64ExitCode,
7185 pVmcb->ctrl.u64ExitInfo1, pVmcb->ctrl.u64ExitInfo2));
7186 RT_NOREF(pVmcb);
7187 pVCpu->hm.s.u32HMError = (uint32_t)pSvmTransient->u64ExitCode;
7188 return VERR_SVM_UNEXPECTED_EXIT;
7189}
7190
7191
7192/**
7193 * \#VMEXIT handler for CRx reads (SVM_EXIT_READ_CR*). Conditional \#VMEXIT.
7194 */
7195HMSVM_EXIT_DECL hmR0SvmExitReadCRx(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
7196{
7197 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
7198
7199 PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
7200 Log4Func(("CS:RIP=%04x:%RX64\n", pCtx->cs.Sel, pCtx->rip));
7201#ifdef VBOX_WITH_STATISTICS
7202 switch (pSvmTransient->u64ExitCode)
7203 {
7204 case SVM_EXIT_READ_CR0: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR0Read); break;
7205 case SVM_EXIT_READ_CR2: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR2Read); break;
7206 case SVM_EXIT_READ_CR3: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR3Read); break;
7207 case SVM_EXIT_READ_CR4: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR4Read); break;
7208 case SVM_EXIT_READ_CR8: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR8Read); break;
7209 }
7210#endif
7211
7212 bool const fSupportsDecodeAssists = hmR0SvmSupportsDecodeAssists(pVCpu);
7213 bool const fSupportsNextRipSave = hmR0SvmSupportsNextRipSave(pVCpu);
7214 if ( fSupportsDecodeAssists
7215 && fSupportsNextRipSave)
7216 {
7217 PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
7218 bool const fMovCRx = RT_BOOL(pVmcb->ctrl.u64ExitInfo1 & SVM_EXIT1_MOV_CRX_MASK);
7219 if (fMovCRx)
7220 {
7221 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK | CPUMCTX_EXTRN_CR_MASK
7222 | CPUMCTX_EXTRN_APIC_TPR);
7223 uint8_t const cbInstr = pVmcb->ctrl.u64NextRIP - pCtx->rip;
7224 uint8_t const iCrReg = pSvmTransient->u64ExitCode - SVM_EXIT_READ_CR0;
7225 uint8_t const iGReg = pVmcb->ctrl.u64ExitInfo1 & SVM_EXIT1_MOV_CRX_GPR_NUMBER;
7226 VBOXSTRICTRC rcStrict = IEMExecDecodedMovCRxRead(pVCpu, cbInstr, iGReg, iCrReg);
7227 HMSVM_CHECK_SINGLE_STEP(pVCpu, rcStrict);
7228 return VBOXSTRICTRC_VAL(rcStrict);
7229 }
7230 /* else: SMSW instruction, fall back below to IEM for this. */
7231 }
7232
7233 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_MUST_MASK);
7234 VBOXSTRICTRC rcStrict = IEMExecOne(pVCpu);
7235 AssertMsg( rcStrict == VINF_SUCCESS
7236 || rcStrict == VINF_PGM_SYNC_CR3
7237 || rcStrict == VINF_IEM_RAISED_XCPT,
7238 ("hmR0SvmExitReadCRx: IEMExecOne failed rc=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
7239 Assert((pSvmTransient->u64ExitCode - SVM_EXIT_READ_CR0) <= 15);
7240 if (rcStrict == VINF_IEM_RAISED_XCPT)
7241 {
7242 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
7243 rcStrict = VINF_SUCCESS;
7244 }
7245 HMSVM_CHECK_SINGLE_STEP(pVCpu, rcStrict);
7246 return rcStrict;
7247}
7248
7249
7250/**
7251 * \#VMEXIT handler for CRx writes (SVM_EXIT_WRITE_CR*). Conditional \#VMEXIT.
7252 */
7253HMSVM_EXIT_DECL hmR0SvmExitWriteCRx(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
7254{
7255 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
7256
7257 uint64_t const uExitCode = pSvmTransient->u64ExitCode;
7258 uint8_t const iCrReg = uExitCode == SVM_EXIT_CR0_SEL_WRITE ? 0 : (pSvmTransient->u64ExitCode - SVM_EXIT_WRITE_CR0);
7259 Assert(iCrReg <= 15);
7260
7261 VBOXSTRICTRC rcStrict = VERR_SVM_IPE_5;
7262 PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
7263 bool fDecodedInstr = false;
7264 bool const fSupportsDecodeAssists = hmR0SvmSupportsDecodeAssists(pVCpu);
7265 bool const fSupportsNextRipSave = hmR0SvmSupportsNextRipSave(pVCpu);
7266 if ( fSupportsDecodeAssists
7267 && fSupportsNextRipSave)
7268 {
7269 PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
7270 bool const fMovCRx = RT_BOOL(pVmcb->ctrl.u64ExitInfo1 & SVM_EXIT1_MOV_CRX_MASK);
7271 if (fMovCRx)
7272 {
7273 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_EXEC_DECODED_MEM_MASK | CPUMCTX_EXTRN_CR3 | CPUMCTX_EXTRN_CR4
7274 | CPUMCTX_EXTRN_APIC_TPR);
7275 uint8_t const cbInstr = pVmcb->ctrl.u64NextRIP - pCtx->rip;
7276 uint8_t const iGReg = pVmcb->ctrl.u64ExitInfo1 & SVM_EXIT1_MOV_CRX_GPR_NUMBER;
7277 Log4Func(("Mov CR%u w/ iGReg=%#x\n", iCrReg, iGReg));
7278 rcStrict = IEMExecDecodedMovCRxWrite(pVCpu, cbInstr, iCrReg, iGReg);
7279 fDecodedInstr = true;
7280 }
7281 /* else: LMSW or CLTS instruction, fall back below to IEM for this. */
7282 }
7283
7284 if (!fDecodedInstr)
7285 {
7286 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_MUST_MASK);
7287 Log4Func(("iCrReg=%#x\n", iCrReg));
7288 rcStrict = IEMExecOne(pVCpu);
7289 if (RT_UNLIKELY( rcStrict == VERR_IEM_ASPECT_NOT_IMPLEMENTED
7290 || rcStrict == VERR_IEM_INSTR_NOT_IMPLEMENTED))
7291 rcStrict = VERR_EM_INTERPRETER;
7292 }
7293
7294 if (rcStrict == VINF_SUCCESS)
7295 {
7296 switch (iCrReg)
7297 {
7298 case 0:
7299 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_CR0);
7300 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR0Write);
7301 break;
7302
7303 case 2:
7304 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_CR2);
7305 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR2Write);
7306 break;
7307
7308 case 3:
7309 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_CR3);
7310 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR3Write);
7311 break;
7312
7313 case 4:
7314 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_CR4);
7315 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR4Write);
7316 break;
7317
7318 case 8:
7319 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_APIC_TPR);
7320 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCR8Write);
7321 break;
7322
7323 default:
7324 {
7325 AssertMsgFailed(("hmR0SvmExitWriteCRx: Invalid/Unexpected Write-CRx exit. u64ExitCode=%#RX64 %#x\n",
7326 pSvmTransient->u64ExitCode, iCrReg));
7327 break;
7328 }
7329 }
7330 HMSVM_CHECK_SINGLE_STEP(pVCpu, rcStrict);
7331 }
7332 else if (rcStrict == VINF_IEM_RAISED_XCPT)
7333 {
7334 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
7335 HMSVM_CHECK_SINGLE_STEP(pVCpu, rcStrict);
7336 rcStrict = VINF_SUCCESS;
7337 }
7338 else
7339 Assert(rcStrict == VERR_EM_INTERPRETER || rcStrict == VINF_PGM_SYNC_CR3);
7340 return rcStrict;
7341}
7342
7343
7344/**
7345 * \#VMEXIT helper for read MSRs, see hmR0SvmExitMsr.
7346 *
7347 * @returns Strict VBox status code.
7348 * @param pVCpu The cross context virtual CPU structure.
7349 * @param pVmcb Pointer to the VM control block.
7350 */
7351static VBOXSTRICTRC hmR0SvmExitReadMsr(PVMCPUCC pVCpu, PSVMVMCB pVmcb)
7352{
7353 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdmsr);
7354 Log4Func(("idMsr=%#RX32\n", pVCpu->cpum.GstCtx.ecx));
7355
7356 VBOXSTRICTRC rcStrict;
7357 bool const fSupportsNextRipSave = hmR0SvmSupportsNextRipSave(pVCpu);
7358 if (fSupportsNextRipSave)
7359 {
7360 /** @todo Optimize this: Only retrieve the MSR bits we need here. CPUMAllMsrs.cpp
7361 * can ask for what it needs instead of using CPUMCTX_EXTRN_ALL_MSRS. */
7362 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK | CPUMCTX_EXTRN_ALL_MSRS);
7363 uint8_t const cbInstr = pVmcb->ctrl.u64NextRIP - pVCpu->cpum.GstCtx.rip;
7364 rcStrict = IEMExecDecodedRdmsr(pVCpu, cbInstr);
7365 }
7366 else
7367 {
7368 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_MUST_MASK | CPUMCTX_EXTRN_ALL_MSRS);
7369 rcStrict = IEMExecOne(pVCpu);
7370 }
7371
7372 AssertMsg( rcStrict == VINF_SUCCESS
7373 || rcStrict == VINF_IEM_RAISED_XCPT
7374 || rcStrict == VINF_CPUM_R3_MSR_READ,
7375 ("hmR0SvmExitReadMsr: Unexpected status %Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
7376
7377 if (rcStrict == VINF_IEM_RAISED_XCPT)
7378 {
7379 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
7380 rcStrict = VINF_SUCCESS;
7381 }
7382 HMSVM_CHECK_SINGLE_STEP(pVCpu, rcStrict);
7383 return rcStrict;
7384}
7385
7386
7387/**
7388 * \#VMEXIT helper for write MSRs, see hmR0SvmExitMsr.
7389 *
7390 * @returns Strict VBox status code.
7391 * @param pVCpu The cross context virtual CPU structure.
7392 * @param pVmcb Pointer to the VM control block.
7393 * @param pSvmTransient Pointer to the SVM-transient structure.
7394 */
7395static VBOXSTRICTRC hmR0SvmExitWriteMsr(PVMCPUCC pVCpu, PSVMVMCB pVmcb, PSVMTRANSIENT pSvmTransient)
7396{
7397 PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
7398 uint32_t const idMsr = pCtx->ecx;
7399 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitWrmsr);
7400 Log4Func(("idMsr=%#RX32\n", idMsr));
7401
7402 /*
7403 * Handle TPR patching MSR writes.
7404 * We utilitize the LSTAR MSR for patching.
7405 */
7406 bool const fSupportsNextRipSave = hmR0SvmSupportsNextRipSave(pVCpu);
7407 if ( idMsr == MSR_K8_LSTAR
7408 && pVCpu->CTX_SUFF(pVM)->hm.s.fTprPatchingActive)
7409 {
7410 unsigned cbInstr;
7411 if (fSupportsNextRipSave)
7412 cbInstr = pVmcb->ctrl.u64NextRIP - pVCpu->cpum.GstCtx.rip;
7413 else
7414 {
7415 PDISCPUSTATE pDis = &pVCpu->hmr0.s.svm.DisState;
7416 int rc = EMInterpretDisasCurrent(pVCpu, pDis, &cbInstr);
7417 if ( rc == VINF_SUCCESS
7418 && pDis->pCurInstr->uOpcode == OP_WRMSR)
7419 Assert(cbInstr > 0);
7420 else
7421 cbInstr = 0;
7422 }
7423
7424 /* Our patch code uses LSTAR for TPR caching for 32-bit guests. */
7425 if ((pCtx->eax & 0xff) != pSvmTransient->u8GuestTpr)
7426 {
7427 int rc = APICSetTpr(pVCpu, pCtx->eax & 0xff);
7428 AssertRCReturn(rc, rc);
7429 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_APIC_TPR);
7430 }
7431
7432 int rc = VINF_SUCCESS;
7433 hmR0SvmAdvanceRip(pVCpu, cbInstr);
7434 HMSVM_CHECK_SINGLE_STEP(pVCpu, rc);
7435 return rc;
7436 }
7437
7438 /*
7439 * Handle regular MSR writes.
7440 */
7441 VBOXSTRICTRC rcStrict;
7442 if (fSupportsNextRipSave)
7443 {
7444 /** @todo Optimize this: We don't need to get much of the MSR state here
7445 * since we're only updating. CPUMAllMsrs.cpp can ask for what it needs and
7446 * clear the applicable extern flags. */
7447 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK | CPUMCTX_EXTRN_ALL_MSRS);
7448 uint8_t const cbInstr = pVmcb->ctrl.u64NextRIP - pVCpu->cpum.GstCtx.rip;
7449 rcStrict = IEMExecDecodedWrmsr(pVCpu, cbInstr);
7450 }
7451 else
7452 {
7453 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_MUST_MASK | CPUMCTX_EXTRN_ALL_MSRS);
7454 rcStrict = IEMExecOne(pVCpu);
7455 }
7456
7457 AssertMsg( rcStrict == VINF_SUCCESS
7458 || rcStrict == VINF_IEM_RAISED_XCPT
7459 || rcStrict == VINF_CPUM_R3_MSR_WRITE,
7460 ("hmR0SvmExitWriteMsr: Unexpected status %Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
7461
7462 if (rcStrict == VINF_SUCCESS)
7463 {
7464 /* If this is an X2APIC WRMSR access, update the APIC TPR state. */
7465 if ( idMsr >= MSR_IA32_X2APIC_START
7466 && idMsr <= MSR_IA32_X2APIC_END)
7467 {
7468 /*
7469 * We've already saved the APIC related guest-state (TPR) in hmR0SvmPostRunGuest().
7470 * When full APIC register virtualization is implemented we'll have to make sure
7471 * APIC state is saved from the VMCB before IEM changes it.
7472 */
7473 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_APIC_TPR);
7474 }
7475 else
7476 {
7477 switch (idMsr)
7478 {
7479 case MSR_IA32_TSC: pSvmTransient->fUpdateTscOffsetting = true; break;
7480 case MSR_K6_EFER: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_EFER_MSR); break;
7481 case MSR_K8_FS_BASE: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_FS); break;
7482 case MSR_K8_GS_BASE: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_GS); break;
7483 case MSR_IA32_SYSENTER_CS: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_SYSENTER_CS_MSR); break;
7484 case MSR_IA32_SYSENTER_EIP: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_SYSENTER_EIP_MSR); break;
7485 case MSR_IA32_SYSENTER_ESP: ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_SYSENTER_ESP_MSR); break;
7486 }
7487 }
7488 }
7489 else if (rcStrict == VINF_IEM_RAISED_XCPT)
7490 {
7491 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
7492 rcStrict = VINF_SUCCESS;
7493 }
7494 HMSVM_CHECK_SINGLE_STEP(pVCpu, rcStrict);
7495 return rcStrict;
7496}
7497
7498
7499/**
7500 * \#VMEXIT handler for MSR read and writes (SVM_EXIT_MSR). Conditional
7501 * \#VMEXIT.
7502 */
7503HMSVM_EXIT_DECL hmR0SvmExitMsr(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
7504{
7505 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
7506
7507 PSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
7508 if (pVmcb->ctrl.u64ExitInfo1 == SVM_EXIT1_MSR_READ)
7509 return hmR0SvmExitReadMsr(pVCpu, pVmcb);
7510
7511 Assert(pVmcb->ctrl.u64ExitInfo1 == SVM_EXIT1_MSR_WRITE);
7512 return hmR0SvmExitWriteMsr(pVCpu, pVmcb, pSvmTransient);
7513}
7514
7515
7516/**
7517 * \#VMEXIT handler for DRx read (SVM_EXIT_READ_DRx). Conditional \#VMEXIT.
7518 */
7519HMSVM_EXIT_DECL hmR0SvmExitReadDRx(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
7520{
7521 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
7522 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, HMSVM_CPUMCTX_EXTRN_ALL);
7523
7524 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxRead);
7525
7526 /** @todo Stepping with nested-guest. */
7527 PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
7528 if (!CPUMIsGuestInSvmNestedHwVirtMode(pCtx))
7529 {
7530 /* We should -not- get this #VMEXIT if the guest's debug registers were active. */
7531 if (pSvmTransient->fWasGuestDebugStateActive)
7532 {
7533 AssertMsgFailed(("hmR0SvmExitReadDRx: Unexpected exit %#RX32\n", (uint32_t)pSvmTransient->u64ExitCode));
7534 pVCpu->hm.s.u32HMError = (uint32_t)pSvmTransient->u64ExitCode;
7535 return VERR_SVM_UNEXPECTED_EXIT;
7536 }
7537
7538 /*
7539 * Lazy DR0-3 loading.
7540 */
7541 if (!pSvmTransient->fWasHyperDebugStateActive)
7542 {
7543 Assert(!DBGFIsStepping(pVCpu)); Assert(!pVCpu->hm.s.fSingleInstruction);
7544 Log5(("hmR0SvmExitReadDRx: Lazy loading guest debug registers\n"));
7545
7546 /* Don't intercept DRx read and writes. */
7547 PSVMVMCB pVmcb = pVCpu->hmr0.s.svm.pVmcb;
7548 pVmcb->ctrl.u16InterceptRdDRx = 0;
7549 pVmcb->ctrl.u16InterceptWrDRx = 0;
7550 pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_INTERCEPTS;
7551
7552 /* We're playing with the host CPU state here, make sure we don't preempt or longjmp. */
7553 VMMRZCallRing3Disable(pVCpu);
7554 HM_DISABLE_PREEMPT(pVCpu);
7555
7556 /* Save the host & load the guest debug state, restart execution of the MOV DRx instruction. */
7557 CPUMR0LoadGuestDebugState(pVCpu, false /* include DR6 */);
7558 Assert(CPUMIsGuestDebugStateActive(pVCpu));
7559
7560 HM_RESTORE_PREEMPT();
7561 VMMRZCallRing3Enable(pVCpu);
7562
7563 STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxContextSwitch);
7564 return VINF_SUCCESS;
7565 }
7566 }
7567
7568 /*
7569 * Interpret the read/writing of DRx.
7570 */
7571 /** @todo Decode assist. */
7572 VBOXSTRICTRC rc = EMInterpretInstruction(pVCpu, CPUMCTX2CORE(pCtx), 0 /* pvFault */);
7573 Log5(("hmR0SvmExitReadDRx: Emulated DRx access: rc=%Rrc\n", VBOXSTRICTRC_VAL(rc)));
7574 if (RT_LIKELY(rc == VINF_SUCCESS))
7575 {
7576 /* Not necessary for read accesses but whatever doesn't hurt for now, will be fixed with decode assist. */
7577 /** @todo CPUM should set this flag! */
7578 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_DR_MASK);
7579 HMSVM_CHECK_SINGLE_STEP(pVCpu, rc);
7580 }
7581 else
7582 Assert(rc == VERR_EM_INTERPRETER);
7583 return rc;
7584}
7585
7586
7587/**
7588 * \#VMEXIT handler for DRx write (SVM_EXIT_WRITE_DRx). Conditional \#VMEXIT.
7589 */
7590HMSVM_EXIT_DECL hmR0SvmExitWriteDRx(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
7591{
7592 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
7593 /* For now it's the same since we interpret the instruction anyway. Will change when using of Decode Assist is implemented. */
7594 VBOXSTRICTRC rc = hmR0SvmExitReadDRx(pVCpu, pSvmTransient);
7595 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxWrite);
7596 STAM_COUNTER_DEC(&pVCpu->hm.s.StatExitDRxRead);
7597 return rc;
7598}
7599
7600
7601/**
7602 * \#VMEXIT handler for XCRx write (SVM_EXIT_XSETBV). Conditional \#VMEXIT.
7603 */
7604HMSVM_EXIT_DECL hmR0SvmExitXsetbv(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
7605{
7606 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
7607 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_MUST_MASK);
7608
7609 /** @todo decode assists... */
7610 VBOXSTRICTRC rcStrict = IEMExecOne(pVCpu);
7611 if (RT_LIKELY(rcStrict == VINF_SUCCESS))
7612 {
7613 PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
7614 bool const fLoadSaveGuestXcr0 = (pCtx->cr4 & X86_CR4_OSXSAVE) && pCtx->aXcr[0] != ASMGetXcr0();
7615 Log4Func(("New XCR0=%#RX64 fLoadSaveGuestXcr0=%RTbool (cr4=%#RX64)\n", pCtx->aXcr[0], fLoadSaveGuestXcr0, pCtx->cr4));
7616 if (fLoadSaveGuestXcr0 != pVCpu->hmr0.s.fLoadSaveGuestXcr0)
7617 {
7618 pVCpu->hmr0.s.fLoadSaveGuestXcr0 = fLoadSaveGuestXcr0;
7619 hmR0SvmUpdateVmRunFunction(pVCpu);
7620 }
7621 }
7622 else if (rcStrict == VINF_IEM_RAISED_XCPT)
7623 {
7624 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
7625 rcStrict = VINF_SUCCESS;
7626 }
7627 HMSVM_CHECK_SINGLE_STEP(pVCpu, rcStrict);
7628 return rcStrict;
7629}
7630
7631
7632/**
7633 * \#VMEXIT handler for I/O instructions (SVM_EXIT_IOIO). Conditional \#VMEXIT.
7634 */
7635HMSVM_EXIT_DECL hmR0SvmExitIOInstr(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
7636{
7637 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
7638 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_MUST_MASK | CPUMCTX_EXTRN_SREG_MASK);
7639
7640 /* I/O operation lookup arrays. */
7641 static uint32_t const s_aIOSize[8] = { 0, 1, 2, 0, 4, 0, 0, 0 }; /* Size of the I/O accesses in bytes. */
7642 static uint32_t const s_aIOOpAnd[8] = { 0, 0xff, 0xffff, 0, 0xffffffff, 0, 0, 0 }; /* AND masks for saving
7643 the result (in AL/AX/EAX). */
7644 PVMCC pVM = pVCpu->CTX_SUFF(pVM);
7645 PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
7646 PSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
7647
7648 Log4Func(("CS:RIP=%04x:%RX64\n", pCtx->cs.Sel, pCtx->rip));
7649
7650 /* Refer AMD spec. 15.10.2 "IN and OUT Behaviour" and Figure 15-2. "EXITINFO1 for IOIO Intercept" for the format. */
7651 SVMIOIOEXITINFO IoExitInfo;
7652 IoExitInfo.u = (uint32_t)pVmcb->ctrl.u64ExitInfo1;
7653 uint32_t uIOWidth = (IoExitInfo.u >> 4) & 0x7;
7654 uint32_t cbValue = s_aIOSize[uIOWidth];
7655 uint32_t uAndVal = s_aIOOpAnd[uIOWidth];
7656
7657 if (RT_UNLIKELY(!cbValue))
7658 {
7659 AssertMsgFailed(("hmR0SvmExitIOInstr: Invalid IO operation. uIOWidth=%u\n", uIOWidth));
7660 return VERR_EM_INTERPRETER;
7661 }
7662
7663 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, CPUMCTX_EXTRN_CS | CPUMCTX_EXTRN_RIP | CPUMCTX_EXTRN_RFLAGS);
7664 VBOXSTRICTRC rcStrict;
7665 PCEMEXITREC pExitRec = NULL;
7666 if ( !pVCpu->hm.s.fSingleInstruction
7667 && !pVCpu->cpum.GstCtx.eflags.Bits.u1TF)
7668 pExitRec = EMHistoryUpdateFlagsAndTypeAndPC(pVCpu,
7669 !IoExitInfo.n.u1Str
7670 ? IoExitInfo.n.u1Type == SVM_IOIO_READ
7671 ? EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM | EMEXIT_F_HM, EMEXITTYPE_IO_PORT_READ)
7672 : EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM | EMEXIT_F_HM, EMEXITTYPE_IO_PORT_WRITE)
7673 : IoExitInfo.n.u1Type == SVM_IOIO_READ
7674 ? EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM | EMEXIT_F_HM, EMEXITTYPE_IO_PORT_STR_READ)
7675 : EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM | EMEXIT_F_HM, EMEXITTYPE_IO_PORT_STR_WRITE),
7676 pVCpu->cpum.GstCtx.rip + pVCpu->cpum.GstCtx.cs.u64Base);
7677 if (!pExitRec)
7678 {
7679 bool fUpdateRipAlready = false;
7680 if (IoExitInfo.n.u1Str)
7681 {
7682 /* INS/OUTS - I/O String instruction. */
7683 /** @todo Huh? why can't we use the segment prefix information given by AMD-V
7684 * in EXITINFO1? Investigate once this thing is up and running. */
7685 Log4Func(("CS:RIP=%04x:%08RX64 %#06x/%u %c str\n", pCtx->cs.Sel, pCtx->rip, IoExitInfo.n.u16Port, cbValue,
7686 IoExitInfo.n.u1Type == SVM_IOIO_WRITE ? 'w' : 'r'));
7687 AssertReturn(pCtx->dx == IoExitInfo.n.u16Port, VERR_SVM_IPE_2);
7688 static IEMMODE const s_aenmAddrMode[8] =
7689 {
7690 (IEMMODE)-1, IEMMODE_16BIT, IEMMODE_32BIT, (IEMMODE)-1, IEMMODE_64BIT, (IEMMODE)-1, (IEMMODE)-1, (IEMMODE)-1
7691 };
7692 IEMMODE enmAddrMode = s_aenmAddrMode[(IoExitInfo.u >> 7) & 0x7];
7693 if (enmAddrMode != (IEMMODE)-1)
7694 {
7695 uint64_t cbInstr = pVmcb->ctrl.u64ExitInfo2 - pCtx->rip;
7696 if (cbInstr <= 15 && cbInstr >= 1)
7697 {
7698 Assert(cbInstr >= 1U + IoExitInfo.n.u1Rep);
7699 if (IoExitInfo.n.u1Type == SVM_IOIO_WRITE)
7700 {
7701 /* Don't know exactly how to detect whether u3Seg is valid, currently
7702 only enabling it for Bulldozer and later with NRIP. OS/2 broke on
7703 2384 Opterons when only checking NRIP. */
7704 bool const fSupportsNextRipSave = hmR0SvmSupportsNextRipSave(pVCpu);
7705 if ( fSupportsNextRipSave
7706 && pVM->cpum.ro.GuestFeatures.enmMicroarch >= kCpumMicroarch_AMD_15h_First)
7707 {
7708 AssertMsg(IoExitInfo.n.u3Seg == X86_SREG_DS || cbInstr > 1U + IoExitInfo.n.u1Rep,
7709 ("u32Seg=%d cbInstr=%d u1REP=%d", IoExitInfo.n.u3Seg, cbInstr, IoExitInfo.n.u1Rep));
7710 rcStrict = IEMExecStringIoWrite(pVCpu, cbValue, enmAddrMode, IoExitInfo.n.u1Rep, (uint8_t)cbInstr,
7711 IoExitInfo.n.u3Seg, true /*fIoChecked*/);
7712 }
7713 else if (cbInstr == 1U + IoExitInfo.n.u1Rep)
7714 rcStrict = IEMExecStringIoWrite(pVCpu, cbValue, enmAddrMode, IoExitInfo.n.u1Rep, (uint8_t)cbInstr,
7715 X86_SREG_DS, true /*fIoChecked*/);
7716 else
7717 rcStrict = IEMExecOne(pVCpu);
7718 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIOStringWrite);
7719 }
7720 else
7721 {
7722 AssertMsg(IoExitInfo.n.u3Seg == X86_SREG_ES /*=0*/, ("%#x\n", IoExitInfo.n.u3Seg));
7723 rcStrict = IEMExecStringIoRead(pVCpu, cbValue, enmAddrMode, IoExitInfo.n.u1Rep, (uint8_t)cbInstr,
7724 true /*fIoChecked*/);
7725 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIOStringRead);
7726 }
7727 }
7728 else
7729 {
7730 AssertMsgFailed(("rip=%RX64 nrip=%#RX64 cbInstr=%#RX64\n", pCtx->rip, pVmcb->ctrl.u64ExitInfo2, cbInstr));
7731 rcStrict = IEMExecOne(pVCpu);
7732 }
7733 }
7734 else
7735 {
7736 AssertMsgFailed(("IoExitInfo=%RX64\n", IoExitInfo.u));
7737 rcStrict = IEMExecOne(pVCpu);
7738 }
7739 fUpdateRipAlready = true;
7740 if (rcStrict == VINF_IEM_RAISED_XCPT)
7741 {
7742 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
7743 rcStrict = VINF_SUCCESS;
7744 }
7745 }
7746 else
7747 {
7748 /* IN/OUT - I/O instruction. */
7749 Assert(!IoExitInfo.n.u1Rep);
7750
7751 uint8_t const cbInstr = pVmcb->ctrl.u64ExitInfo2 - pCtx->rip;
7752 if (IoExitInfo.n.u1Type == SVM_IOIO_WRITE)
7753 {
7754 rcStrict = IOMIOPortWrite(pVM, pVCpu, IoExitInfo.n.u16Port, pCtx->eax & uAndVal, cbValue);
7755 if ( rcStrict == VINF_IOM_R3_IOPORT_WRITE
7756 && !pCtx->eflags.Bits.u1TF)
7757 rcStrict = EMRZSetPendingIoPortWrite(pVCpu, IoExitInfo.n.u16Port, cbInstr, cbValue, pCtx->eax & uAndVal);
7758 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIOWrite);
7759 }
7760 else
7761 {
7762 uint32_t u32Val = 0;
7763 rcStrict = IOMIOPortRead(pVM, pVCpu, IoExitInfo.n.u16Port, &u32Val, cbValue);
7764 if (IOM_SUCCESS(rcStrict))
7765 {
7766 /* Save result of I/O IN instr. in AL/AX/EAX. */
7767 /** @todo r=bird: 32-bit op size should clear high bits of rax! */
7768 pCtx->eax = (pCtx->eax & ~uAndVal) | (u32Val & uAndVal);
7769 }
7770 else if ( rcStrict == VINF_IOM_R3_IOPORT_READ
7771 && !pCtx->eflags.Bits.u1TF)
7772 rcStrict = EMRZSetPendingIoPortRead(pVCpu, IoExitInfo.n.u16Port, cbInstr, cbValue);
7773
7774 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIORead);
7775 }
7776 }
7777
7778 if (IOM_SUCCESS(rcStrict))
7779 {
7780 /* AMD-V saves the RIP of the instruction following the IO instruction in EXITINFO2. */
7781 if (!fUpdateRipAlready)
7782 pCtx->rip = pVmcb->ctrl.u64ExitInfo2;
7783
7784 /*
7785 * If any I/O breakpoints are armed, we need to check if one triggered
7786 * and take appropriate action.
7787 * Note that the I/O breakpoint type is undefined if CR4.DE is 0.
7788 */
7789 /** @todo Optimize away the DBGFBpIsHwIoArmed call by having DBGF tell the
7790 * execution engines about whether hyper BPs and such are pending. */
7791 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, CPUMCTX_EXTRN_DR7);
7792 uint32_t const uDr7 = pCtx->dr[7];
7793 if (RT_UNLIKELY( ( (uDr7 & X86_DR7_ENABLED_MASK)
7794 && X86_DR7_ANY_RW_IO(uDr7)
7795 && (pCtx->cr4 & X86_CR4_DE))
7796 || DBGFBpIsHwIoArmed(pVM)))
7797 {
7798 /* We're playing with the host CPU state here, make sure we don't preempt or longjmp. */
7799 VMMRZCallRing3Disable(pVCpu);
7800 HM_DISABLE_PREEMPT(pVCpu);
7801
7802 STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxIoCheck);
7803 CPUMR0DebugStateMaybeSaveGuest(pVCpu, false /*fDr6*/);
7804
7805 VBOXSTRICTRC rcStrict2 = DBGFBpCheckIo(pVM, pVCpu, &pVCpu->cpum.GstCtx, IoExitInfo.n.u16Port, cbValue);
7806 if (rcStrict2 == VINF_EM_RAW_GUEST_TRAP)
7807 {
7808 /* Raise #DB. */
7809 pVmcb->guest.u64DR6 = pCtx->dr[6];
7810 pVmcb->guest.u64DR7 = pCtx->dr[7];
7811 pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_DRX;
7812 hmR0SvmSetPendingXcptDB(pVCpu);
7813 }
7814 /* rcStrict is VINF_SUCCESS, VINF_IOM_R3_IOPORT_COMMIT_WRITE, or in [VINF_EM_FIRST..VINF_EM_LAST],
7815 however we can ditch VINF_IOM_R3_IOPORT_COMMIT_WRITE as it has VMCPU_FF_IOM as backup. */
7816 else if ( rcStrict2 != VINF_SUCCESS
7817 && (rcStrict == VINF_SUCCESS || rcStrict2 < rcStrict))
7818 rcStrict = rcStrict2;
7819 AssertCompile(VINF_EM_LAST < VINF_IOM_R3_IOPORT_COMMIT_WRITE);
7820
7821 HM_RESTORE_PREEMPT();
7822 VMMRZCallRing3Enable(pVCpu);
7823 }
7824
7825 HMSVM_CHECK_SINGLE_STEP(pVCpu, rcStrict);
7826 }
7827#ifdef VBOX_STRICT
7828 if ( rcStrict == VINF_IOM_R3_IOPORT_READ
7829 || rcStrict == VINF_EM_PENDING_R3_IOPORT_READ)
7830 Assert(IoExitInfo.n.u1Type == SVM_IOIO_READ);
7831 else if ( rcStrict == VINF_IOM_R3_IOPORT_WRITE
7832 || rcStrict == VINF_IOM_R3_IOPORT_COMMIT_WRITE
7833 || rcStrict == VINF_EM_PENDING_R3_IOPORT_WRITE)
7834 Assert(IoExitInfo.n.u1Type == SVM_IOIO_WRITE);
7835 else
7836 {
7837 /** @todo r=bird: This is missing a bunch of VINF_EM_FIRST..VINF_EM_LAST
7838 * statuses, that the VMM device and some others may return. See
7839 * IOM_SUCCESS() for guidance. */
7840 AssertMsg( RT_FAILURE(rcStrict)
7841 || rcStrict == VINF_SUCCESS
7842 || rcStrict == VINF_EM_RAW_EMULATE_INSTR
7843 || rcStrict == VINF_EM_DBG_BREAKPOINT
7844 || rcStrict == VINF_EM_RAW_GUEST_TRAP
7845 || rcStrict == VINF_EM_DBG_STEPPED
7846 || rcStrict == VINF_EM_RAW_TO_R3
7847 || rcStrict == VINF_EM_TRIPLE_FAULT, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
7848 }
7849#endif
7850 }
7851 else
7852 {
7853 /*
7854 * Frequent exit or something needing probing. Get state and call EMHistoryExec.
7855 */
7856 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, HMSVM_CPUMCTX_EXTRN_ALL);
7857 STAM_COUNTER_INC(!IoExitInfo.n.u1Str
7858 ? IoExitInfo.n.u1Type == SVM_IOIO_WRITE ? &pVCpu->hm.s.StatExitIOWrite : &pVCpu->hm.s.StatExitIORead
7859 : IoExitInfo.n.u1Type == SVM_IOIO_WRITE ? &pVCpu->hm.s.StatExitIOStringWrite : &pVCpu->hm.s.StatExitIOStringRead);
7860 Log4(("IOExit/%u: %04x:%08RX64: %s%s%s %#x LB %u -> EMHistoryExec\n",
7861 pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip, IoExitInfo.n.u1Rep ? "REP " : "",
7862 IoExitInfo.n.u1Type == SVM_IOIO_WRITE ? "OUT" : "IN", IoExitInfo.n.u1Str ? "S" : "", IoExitInfo.n.u16Port, uIOWidth));
7863
7864 rcStrict = EMHistoryExec(pVCpu, pExitRec, 0);
7865 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
7866
7867 Log4(("IOExit/%u: %04x:%08RX64: EMHistoryExec -> %Rrc + %04x:%08RX64\n",
7868 pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
7869 VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip));
7870 }
7871 return rcStrict;
7872}
7873
7874
7875/**
7876 * \#VMEXIT handler for Nested Page-faults (SVM_EXIT_NPF). Conditional \#VMEXIT.
7877 */
7878HMSVM_EXIT_DECL hmR0SvmExitNestedPF(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
7879{
7880 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
7881 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, HMSVM_CPUMCTX_EXTRN_ALL);
7882 HMSVM_CHECK_EXIT_DUE_TO_EVENT_DELIVERY(pVCpu, pSvmTransient);
7883
7884 PVMCC pVM = pVCpu->CTX_SUFF(pVM);
7885 PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
7886 Assert(pVM->hmr0.s.fNestedPaging);
7887
7888 /* See AMD spec. 15.25.6 "Nested versus Guest Page Faults, Fault Ordering" for VMCB details for #NPF. */
7889 PSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
7890 RTGCPHYS GCPhysFaultAddr = pVmcb->ctrl.u64ExitInfo2;
7891 uint32_t u32ErrCode = pVmcb->ctrl.u64ExitInfo1; /* Note! High bits in EXITINFO1 may contain additional info and are
7892 thus intentionally not copied into u32ErrCode. */
7893
7894 Log4Func(("#NPF at CS:RIP=%04x:%RX64 GCPhysFaultAddr=%RGp ErrCode=%#x cbInstrFetched=%u %.15Rhxs\n", pCtx->cs.Sel, pCtx->rip, GCPhysFaultAddr,
7895 u32ErrCode, pVmcb->ctrl.cbInstrFetched, pVmcb->ctrl.abInstr));
7896
7897 /*
7898 * TPR patching for 32-bit guests, using the reserved bit in the page tables for MMIO regions.
7899 */
7900 if ( pVM->hm.s.fTprPatchingAllowed
7901 && (GCPhysFaultAddr & GUEST_PAGE_OFFSET_MASK) == XAPIC_OFF_TPR
7902 && ( !(u32ErrCode & X86_TRAP_PF_P) /* Not present */
7903 || (u32ErrCode & (X86_TRAP_PF_P | X86_TRAP_PF_RSVD)) == (X86_TRAP_PF_P | X86_TRAP_PF_RSVD)) /* MMIO page. */
7904 && !CPUMIsGuestInSvmNestedHwVirtMode(pCtx)
7905 && !CPUMIsGuestInLongModeEx(pCtx)
7906 && !CPUMGetGuestCPL(pVCpu)
7907 && pVM->hm.s.cPatches < RT_ELEMENTS(pVM->hm.s.aPatches))
7908 {
7909 RTGCPHYS GCPhysApicBase = APICGetBaseMsrNoCheck(pVCpu);
7910 GCPhysApicBase &= ~(RTGCPHYS)GUEST_PAGE_OFFSET_MASK;
7911
7912 if (GCPhysFaultAddr == GCPhysApicBase + XAPIC_OFF_TPR)
7913 {
7914 /* Only attempt to patch the instruction once. */
7915 PHMTPRPATCH pPatch = (PHMTPRPATCH)RTAvloU32Get(&pVM->hm.s.PatchTree, (AVLOU32KEY)pCtx->eip);
7916 if (!pPatch)
7917 return VINF_EM_HM_PATCH_TPR_INSTR;
7918 }
7919 }
7920
7921 /*
7922 * Determine the nested paging mode.
7923 */
7924/** @todo r=bird: Gotta love this nested paging hacking we're still carrying with us... (Split PGM_TYPE_NESTED.) */
7925 PGMMODE const enmNestedPagingMode = PGMGetHostMode(pVM);
7926
7927 /*
7928 * MMIO optimization using the reserved (RSVD) bit in the guest page tables for MMIO pages.
7929 */
7930 Assert((u32ErrCode & (X86_TRAP_PF_RSVD | X86_TRAP_PF_P)) != X86_TRAP_PF_RSVD);
7931 if ((u32ErrCode & (X86_TRAP_PF_RSVD | X86_TRAP_PF_P)) == (X86_TRAP_PF_RSVD | X86_TRAP_PF_P))
7932 {
7933 /*
7934 * If event delivery causes an MMIO #NPF, go back to instruction emulation as otherwise
7935 * injecting the original pending event would most likely cause the same MMIO #NPF.
7936 */
7937 if (pVCpu->hm.s.Event.fPending)
7938 {
7939 STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectInterpret);
7940 return VINF_EM_RAW_INJECT_TRPM_EVENT;
7941 }
7942
7943 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, CPUMCTX_EXTRN_CS | CPUMCTX_EXTRN_RIP);
7944 VBOXSTRICTRC rcStrict;
7945 PCEMEXITREC pExitRec = EMHistoryUpdateFlagsAndTypeAndPC(pVCpu,
7946 EMEXIT_MAKE_FT(EMEXIT_F_KIND_EM | EMEXIT_F_HM, EMEXITTYPE_MMIO),
7947 pVCpu->cpum.GstCtx.rip + pVCpu->cpum.GstCtx.cs.u64Base);
7948 if (!pExitRec)
7949 {
7950
7951 rcStrict = PGMR0Trap0eHandlerNPMisconfig(pVM, pVCpu, enmNestedPagingMode, CPUMCTX2CORE(pCtx), GCPhysFaultAddr,
7952 u32ErrCode);
7953
7954 /*
7955 * If we succeed, resume guest execution.
7956 *
7957 * If we fail in interpreting the instruction because we couldn't get the guest
7958 * physical address of the page containing the instruction via the guest's page
7959 * tables (we would invalidate the guest page in the host TLB), resume execution
7960 * which would cause a guest page fault to let the guest handle this weird case.
7961 *
7962 * See @bugref{6043}.
7963 */
7964 if ( rcStrict == VINF_SUCCESS
7965 || rcStrict == VERR_PAGE_TABLE_NOT_PRESENT
7966 || rcStrict == VERR_PAGE_NOT_PRESENT)
7967 {
7968 /* Successfully handled MMIO operation. */
7969 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_APIC_TPR);
7970 rcStrict = VINF_SUCCESS;
7971 }
7972 }
7973 else
7974 {
7975 /*
7976 * Frequent exit or something needing probing. Get state and call EMHistoryExec.
7977 */
7978 Assert(pCtx == &pVCpu->cpum.GstCtx);
7979 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, HMSVM_CPUMCTX_EXTRN_ALL);
7980 Log4(("EptMisscfgExit/%u: %04x:%08RX64: %RGp -> EMHistoryExec\n",
7981 pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip, GCPhysFaultAddr));
7982
7983 rcStrict = EMHistoryExec(pVCpu, pExitRec, 0);
7984 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
7985
7986 Log4(("EptMisscfgExit/%u: %04x:%08RX64: EMHistoryExec -> %Rrc + %04x:%08RX64\n",
7987 pVCpu->idCpu, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
7988 VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip));
7989 }
7990 return rcStrict;
7991 }
7992
7993 /*
7994 * Nested page-fault.
7995 */
7996 TRPMAssertXcptPF(pVCpu, GCPhysFaultAddr, u32ErrCode);
7997 int rc = PGMR0Trap0eHandlerNestedPaging(pVM, pVCpu, enmNestedPagingMode, u32ErrCode, CPUMCTX2CORE(pCtx), GCPhysFaultAddr);
7998 TRPMResetTrap(pVCpu);
7999
8000 Log4Func(("#NPF: PGMR0Trap0eHandlerNestedPaging returns %Rrc CS:RIP=%04x:%RX64\n", rc, pCtx->cs.Sel, pCtx->rip));
8001
8002 /*
8003 * Same case as PGMR0Trap0eHandlerNPMisconfig(). See comment above, @bugref{6043}.
8004 */
8005 if ( rc == VINF_SUCCESS
8006 || rc == VERR_PAGE_TABLE_NOT_PRESENT
8007 || rc == VERR_PAGE_NOT_PRESENT)
8008 {
8009 /* We've successfully synced our shadow page tables. */
8010 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowPF);
8011 rc = VINF_SUCCESS;
8012 }
8013
8014 /*
8015 * If delivering an event causes an #NPF (and not MMIO), we shall resolve the fault and
8016 * re-inject the original event.
8017 */
8018 if (pVCpu->hm.s.Event.fPending)
8019 {
8020 STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectReflectNPF);
8021
8022 /*
8023 * If the #NPF handler requested emulation of the instruction, ignore it.
8024 * We need to re-inject the original event so as to not lose it.
8025 * Reproducible when booting ReactOS 0.4.12 with BTRFS (installed using BootCD,
8026 * LiveCD is broken for other reasons).
8027 */
8028 if (rc == VINF_EM_RAW_EMULATE_INSTR)
8029 rc = VINF_EM_RAW_INJECT_TRPM_EVENT;
8030 }
8031
8032 return rc;
8033}
8034
8035
8036/**
8037 * \#VMEXIT handler for virtual interrupt (SVM_EXIT_VINTR). Conditional
8038 * \#VMEXIT.
8039 */
8040HMSVM_EXIT_DECL hmR0SvmExitVIntr(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
8041{
8042 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
8043 HMSVM_ASSERT_NOT_IN_NESTED_GUEST(&pVCpu->cpum.GstCtx);
8044
8045 /* Indicate that we no longer need to #VMEXIT when the guest is ready to receive NMIs, it is now ready. */
8046 PSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
8047 hmR0SvmClearIntWindowExiting(pVCpu, pVmcb);
8048
8049 /* Deliver the pending interrupt via hmR0SvmEvaluatePendingEvent() and resume guest execution. */
8050 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIntWindow);
8051 return VINF_SUCCESS;
8052}
8053
8054
8055/**
8056 * \#VMEXIT handler for task switches (SVM_EXIT_TASK_SWITCH). Conditional
8057 * \#VMEXIT.
8058 */
8059HMSVM_EXIT_DECL hmR0SvmExitTaskSwitch(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
8060{
8061 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
8062 HMSVM_CHECK_EXIT_DUE_TO_EVENT_DELIVERY(pVCpu, pSvmTransient);
8063
8064#ifndef HMSVM_ALWAYS_TRAP_TASK_SWITCH
8065 Assert(!pVCpu->CTX_SUFF(pVM)->hmr0.s.fNestedPaging);
8066#endif
8067
8068 /* Check if this task-switch occurred while delivering an event through the guest IDT. */
8069 if (pVCpu->hm.s.Event.fPending) /* Can happen with exceptions/NMI. See @bugref{8411}. */
8070 {
8071 /*
8072 * AMD-V provides us with the exception which caused the TS; we collect
8073 * the information in the call to hmR0SvmCheckExitDueToEventDelivery().
8074 */
8075 Log4Func(("TS occurred during event delivery\n"));
8076 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTaskSwitch);
8077 return VINF_EM_RAW_INJECT_TRPM_EVENT;
8078 }
8079
8080 /** @todo Emulate task switch someday, currently just going back to ring-3 for
8081 * emulation. */
8082 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTaskSwitch);
8083 return VERR_EM_INTERPRETER;
8084}
8085
8086
8087/**
8088 * \#VMEXIT handler for VMMCALL (SVM_EXIT_VMMCALL). Conditional \#VMEXIT.
8089 */
8090HMSVM_EXIT_DECL hmR0SvmExitVmmCall(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
8091{
8092 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
8093 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, HMSVM_CPUMCTX_EXTRN_ALL);
8094
8095 PVMCC pVM = pVCpu->CTX_SUFF(pVM);
8096 if (pVM->hm.s.fTprPatchingAllowed)
8097 {
8098 int rc = hmEmulateSvmMovTpr(pVM, pVCpu);
8099 if (rc != VERR_NOT_FOUND)
8100 {
8101 Log4Func(("hmEmulateSvmMovTpr returns %Rrc\n", rc));
8102 return rc;
8103 }
8104 }
8105
8106 if (EMAreHypercallInstructionsEnabled(pVCpu))
8107 {
8108 unsigned cbInstr;
8109 if (hmR0SvmSupportsNextRipSave(pVCpu))
8110 {
8111 PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
8112 cbInstr = pVmcb->ctrl.u64NextRIP - pVCpu->cpum.GstCtx.rip;
8113 }
8114 else
8115 {
8116 PDISCPUSTATE pDis = &pVCpu->hmr0.s.svm.DisState;
8117 int rc = EMInterpretDisasCurrent(pVCpu, pDis, &cbInstr);
8118 if ( rc == VINF_SUCCESS
8119 && pDis->pCurInstr->uOpcode == OP_VMMCALL)
8120 Assert(cbInstr > 0);
8121 else
8122 cbInstr = 0;
8123 }
8124
8125 VBOXSTRICTRC rcStrict = GIMHypercall(pVCpu, &pVCpu->cpum.GstCtx);
8126 if (RT_SUCCESS(rcStrict))
8127 {
8128 /* Only update the RIP if we're continuing guest execution and not in the case
8129 of say VINF_GIM_R3_HYPERCALL. */
8130 if (rcStrict == VINF_SUCCESS)
8131 hmR0SvmAdvanceRip(pVCpu, cbInstr);
8132
8133 return VBOXSTRICTRC_VAL(rcStrict);
8134 }
8135 else
8136 Log4Func(("GIMHypercall returns %Rrc -> #UD\n", VBOXSTRICTRC_VAL(rcStrict)));
8137 }
8138
8139 hmR0SvmSetPendingXcptUD(pVCpu);
8140 return VINF_SUCCESS;
8141}
8142
8143
8144/**
8145 * \#VMEXIT handler for VMMCALL (SVM_EXIT_VMMCALL). Conditional \#VMEXIT.
8146 */
8147HMSVM_EXIT_DECL hmR0SvmExitPause(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
8148{
8149 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
8150
8151 unsigned cbInstr;
8152 bool const fSupportsNextRipSave = hmR0SvmSupportsNextRipSave(pVCpu);
8153 if (fSupportsNextRipSave)
8154 {
8155 PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
8156 cbInstr = pVmcb->ctrl.u64NextRIP - pVCpu->cpum.GstCtx.rip;
8157 }
8158 else
8159 {
8160 PDISCPUSTATE pDis = &pVCpu->hmr0.s.svm.DisState;
8161 int rc = EMInterpretDisasCurrent(pVCpu, pDis, &cbInstr);
8162 if ( rc == VINF_SUCCESS
8163 && pDis->pCurInstr->uOpcode == OP_PAUSE)
8164 Assert(cbInstr > 0);
8165 else
8166 cbInstr = 0;
8167 }
8168
8169 /** @todo The guest has likely hit a contended spinlock. We might want to
8170 * poke a schedule different guest VCPU. */
8171 hmR0SvmAdvanceRip(pVCpu, cbInstr);
8172 return VINF_EM_RAW_INTERRUPT;
8173}
8174
8175
8176/**
8177 * \#VMEXIT handler for FERR intercept (SVM_EXIT_FERR_FREEZE). Conditional
8178 * \#VMEXIT.
8179 */
8180HMSVM_EXIT_DECL hmR0SvmExitFerrFreeze(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
8181{
8182 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
8183 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, CPUMCTX_EXTRN_CR0);
8184 Assert(!(pVCpu->cpum.GstCtx.cr0 & X86_CR0_NE));
8185
8186 Log4Func(("Raising IRQ 13 in response to #FERR\n"));
8187 return PDMIsaSetIrq(pVCpu->CTX_SUFF(pVM), 13 /* u8Irq */, 1 /* u8Level */, 0 /* uTagSrc */);
8188}
8189
8190
8191/**
8192 * \#VMEXIT handler for IRET (SVM_EXIT_IRET). Conditional \#VMEXIT.
8193 */
8194HMSVM_EXIT_DECL hmR0SvmExitIret(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
8195{
8196 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
8197
8198 /* Indicate that we no longer need to #VMEXIT when the guest is ready to receive NMIs, it is now (almost) ready. */
8199 PSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
8200 hmR0SvmClearCtrlIntercept(pVCpu, pVmcb, SVM_CTRL_INTERCEPT_IRET);
8201
8202 /* Emulate the IRET. We have to execute the IRET before an NMI, but must potentially
8203 * deliver a pending NMI right after. If the IRET faults, an NMI can come before the
8204 * handler executes. Yes, x86 is ugly.
8205 */
8206 return VINF_EM_RAW_EMULATE_INSTR;
8207}
8208
8209
8210/**
8211 * \#VMEXIT handler for page-fault exceptions (SVM_EXIT_XCPT_14).
8212 * Conditional \#VMEXIT.
8213 */
8214HMSVM_EXIT_DECL hmR0SvmExitXcptPF(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
8215{
8216 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
8217 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, HMSVM_CPUMCTX_EXTRN_ALL);
8218 HMSVM_CHECK_EXIT_DUE_TO_EVENT_DELIVERY(pVCpu, pSvmTransient);
8219
8220 /* See AMD spec. 15.12.15 "#PF (Page Fault)". */
8221 PVMCC pVM = pVCpu->CTX_SUFF(pVM);
8222 PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
8223 PSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
8224 uint32_t uErrCode = pVmcb->ctrl.u64ExitInfo1;
8225 uint64_t const uFaultAddress = pVmcb->ctrl.u64ExitInfo2;
8226
8227#if defined(HMSVM_ALWAYS_TRAP_ALL_XCPTS) || defined(HMSVM_ALWAYS_TRAP_PF)
8228 if (pVM->hmr0.s.fNestedPaging)
8229 {
8230 pVCpu->hm.s.Event.fPending = false; /* In case it's a contributory or vectoring #PF. */
8231 if ( !pSvmTransient->fVectoringDoublePF
8232 || CPUMIsGuestInSvmNestedHwVirtMode(pCtx))
8233 {
8234 /* A genuine guest #PF, reflect it to the guest. */
8235 hmR0SvmSetPendingXcptPF(pVCpu, uErrCode, uFaultAddress);
8236 Log4Func(("#PF: Guest page fault at %04X:%RGv FaultAddr=%RX64 ErrCode=%#x\n", pCtx->cs.Sel, (RTGCPTR)pCtx->rip,
8237 uFaultAddress, uErrCode));
8238 }
8239 else
8240 {
8241 /* A guest page-fault occurred during delivery of a page-fault. Inject #DF. */
8242 hmR0SvmSetPendingXcptDF(pVCpu);
8243 Log4Func(("Pending #DF due to vectoring #PF. NP\n"));
8244 }
8245 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF);
8246 return VINF_SUCCESS;
8247 }
8248#endif
8249
8250 Assert(!pVM->hmr0.s.fNestedPaging);
8251
8252 /*
8253 * TPR patching shortcut for APIC TPR reads and writes; only applicable to 32-bit guests.
8254 */
8255 if ( pVM->hm.s.fTprPatchingAllowed
8256 && (uFaultAddress & 0xfff) == XAPIC_OFF_TPR
8257 && !(uErrCode & X86_TRAP_PF_P) /* Not present. */
8258 && !CPUMIsGuestInSvmNestedHwVirtMode(pCtx)
8259 && !CPUMIsGuestInLongModeEx(pCtx)
8260 && !CPUMGetGuestCPL(pVCpu)
8261 && pVM->hm.s.cPatches < RT_ELEMENTS(pVM->hm.s.aPatches))
8262 {
8263 RTGCPHYS GCPhysApicBase;
8264 GCPhysApicBase = APICGetBaseMsrNoCheck(pVCpu);
8265 GCPhysApicBase &= ~(RTGCPHYS)GUEST_PAGE_OFFSET_MASK;
8266
8267 /* Check if the page at the fault-address is the APIC base. */
8268 PGMPTWALK Walk;
8269 int rc2 = PGMGstGetPage(pVCpu, (RTGCPTR)uFaultAddress, &Walk);
8270 if ( rc2 == VINF_SUCCESS
8271 && Walk.GCPhys == GCPhysApicBase)
8272 {
8273 /* Only attempt to patch the instruction once. */
8274 PHMTPRPATCH pPatch = (PHMTPRPATCH)RTAvloU32Get(&pVM->hm.s.PatchTree, (AVLOU32KEY)pCtx->eip);
8275 if (!pPatch)
8276 return VINF_EM_HM_PATCH_TPR_INSTR;
8277 }
8278 }
8279
8280 Log4Func(("#PF: uFaultAddress=%#RX64 CS:RIP=%#04x:%#RX64 uErrCode %#RX32 cr3=%#RX64\n", uFaultAddress, pCtx->cs.Sel,
8281 pCtx->rip, uErrCode, pCtx->cr3));
8282
8283 /*
8284 * If it's a vectoring #PF, emulate injecting the original event injection as
8285 * PGMTrap0eHandler() is incapable of differentiating between instruction emulation and
8286 * event injection that caused a #PF. See @bugref{6607}.
8287 */
8288 if (pSvmTransient->fVectoringPF)
8289 {
8290 Assert(pVCpu->hm.s.Event.fPending);
8291 return VINF_EM_RAW_INJECT_TRPM_EVENT;
8292 }
8293
8294 TRPMAssertXcptPF(pVCpu, uFaultAddress, uErrCode);
8295 int rc = PGMTrap0eHandler(pVCpu, uErrCode, CPUMCTX2CORE(pCtx), (RTGCPTR)uFaultAddress);
8296
8297 Log4Func(("#PF: rc=%Rrc\n", rc));
8298
8299 if (rc == VINF_SUCCESS)
8300 {
8301 /* Successfully synced shadow pages tables or emulated an MMIO instruction. */
8302 TRPMResetTrap(pVCpu);
8303 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowPF);
8304 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_ALL_GUEST);
8305 return rc;
8306 }
8307
8308 if (rc == VINF_EM_RAW_GUEST_TRAP)
8309 {
8310 pVCpu->hm.s.Event.fPending = false; /* In case it's a contributory or vectoring #PF. */
8311
8312 /*
8313 * If a nested-guest delivers a #PF and that causes a #PF which is -not- a shadow #PF,
8314 * we should simply forward the #PF to the guest and is up to the nested-hypervisor to
8315 * determine whether it is a nested-shadow #PF or a #DF, see @bugref{7243#c121}.
8316 */
8317 if ( !pSvmTransient->fVectoringDoublePF
8318 || CPUMIsGuestInSvmNestedHwVirtMode(pCtx))
8319 {
8320 /* It's a guest (or nested-guest) page fault and needs to be reflected. */
8321 uErrCode = TRPMGetErrorCode(pVCpu); /* The error code might have been changed. */
8322 TRPMResetTrap(pVCpu);
8323
8324#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
8325 /* If the nested-guest is intercepting #PFs, cause a #PF #VMEXIT. */
8326 if ( CPUMIsGuestInSvmNestedHwVirtMode(pCtx)
8327 && CPUMIsGuestSvmXcptInterceptSet(pVCpu, pCtx, X86_XCPT_PF))
8328 return IEMExecSvmVmexit(pVCpu, SVM_EXIT_XCPT_PF, uErrCode, uFaultAddress);
8329#endif
8330
8331 hmR0SvmSetPendingXcptPF(pVCpu, uErrCode, uFaultAddress);
8332 }
8333 else
8334 {
8335 /* A guest page-fault occurred during delivery of a page-fault. Inject #DF. */
8336 TRPMResetTrap(pVCpu);
8337 hmR0SvmSetPendingXcptDF(pVCpu);
8338 Log4Func(("#PF: Pending #DF due to vectoring #PF\n"));
8339 }
8340
8341 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF);
8342 return VINF_SUCCESS;
8343 }
8344
8345 TRPMResetTrap(pVCpu);
8346 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowPFEM);
8347 return rc;
8348}
8349
8350
8351
8352/**
8353 * \#VMEXIT handler for division overflow exceptions (SVM_EXIT_XCPT_1).
8354 * Conditional \#VMEXIT.
8355 */
8356HMSVM_EXIT_DECL hmR0SvmExitXcptDE(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
8357{
8358 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
8359 HMSVM_ASSERT_NOT_IN_NESTED_GUEST(&pVCpu->cpum.GstCtx);
8360 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDE);
8361
8362 /* Paranoia; Ensure we cannot be called as a result of event delivery. */
8363 PSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
8364 Assert(!pVmcb->ctrl.ExitIntInfo.n.u1Valid); NOREF(pVmcb);
8365
8366 int rc = VERR_SVM_UNEXPECTED_XCPT_EXIT;
8367 if (pVCpu->hm.s.fGCMTrapXcptDE)
8368 {
8369 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, HMSVM_CPUMCTX_EXTRN_ALL);
8370 uint8_t cbInstr = 0;
8371 VBOXSTRICTRC rcStrict = GCMXcptDE(pVCpu, &pVCpu->cpum.GstCtx, NULL /* pDis */, &cbInstr);
8372 if (rcStrict == VINF_SUCCESS)
8373 rc = VINF_SUCCESS; /* Restart instruction with modified guest register context. */
8374 else if (rcStrict == VERR_NOT_FOUND)
8375 rc = VERR_NOT_FOUND; /* Deliver the exception. */
8376 else
8377 Assert(RT_FAILURE(VBOXSTRICTRC_VAL(rcStrict)));
8378 }
8379
8380 /* If the GCM #DE exception handler didn't succeed or wasn't needed, raise #DE. */
8381 if (RT_FAILURE(rc))
8382 {
8383 hmR0SvmSetPendingXcptDE(pVCpu);
8384 rc = VINF_SUCCESS;
8385 }
8386
8387 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDE);
8388 return rc;
8389}
8390
8391
8392/**
8393 * \#VMEXIT handler for undefined opcode (SVM_EXIT_XCPT_6).
8394 * Conditional \#VMEXIT.
8395 */
8396HMSVM_EXIT_DECL hmR0SvmExitXcptUD(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
8397{
8398 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
8399 HMSVM_ASSERT_NOT_IN_NESTED_GUEST(&pVCpu->cpum.GstCtx);
8400 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestUD);
8401
8402 /* Paranoia; Ensure we cannot be called as a result of event delivery. */
8403 PSVMVMCB pVmcb = pVCpu->hmr0.s.svm.pVmcb;
8404 Assert(!pVmcb->ctrl.ExitIntInfo.n.u1Valid); NOREF(pVmcb);
8405
8406 /** @todo if we accumulate more optional stuff here, we ought to combine the
8407 * reading of opcode bytes to avoid doing more than once. */
8408
8409 VBOXSTRICTRC rcStrict = VERR_SVM_UNEXPECTED_XCPT_EXIT;
8410 if (pVCpu->hm.s.fGIMTrapXcptUD)
8411 {
8412 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, HMSVM_CPUMCTX_EXTRN_ALL);
8413 uint8_t cbInstr = 0;
8414 rcStrict = GIMXcptUD(pVCpu, &pVCpu->cpum.GstCtx, NULL /* pDis */, &cbInstr);
8415 if (rcStrict == VINF_SUCCESS)
8416 {
8417 /* #UD #VMEXIT does not have valid NRIP information, manually advance RIP. See @bugref{7270#c170}. */
8418 hmR0SvmAdvanceRip(pVCpu, cbInstr);
8419 rcStrict = VINF_SUCCESS;
8420 HMSVM_CHECK_SINGLE_STEP(pVCpu, rcStrict);
8421 }
8422 else if (rcStrict == VINF_GIM_HYPERCALL_CONTINUING)
8423 rcStrict = VINF_SUCCESS;
8424 else if (rcStrict == VINF_GIM_R3_HYPERCALL)
8425 rcStrict = VINF_GIM_R3_HYPERCALL;
8426 else
8427 {
8428 Assert(RT_FAILURE(VBOXSTRICTRC_VAL(rcStrict)));
8429 rcStrict = VERR_SVM_UNEXPECTED_XCPT_EXIT;
8430 }
8431 }
8432
8433 if (pVCpu->hm.s.svm.fEmulateLongModeSysEnterExit)
8434 {
8435 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, CPUMCTX_EXTRN_CS | CPUMCTX_EXTRN_SS | CPUMCTX_EXTRN_RIP | CPUMCTX_EXTRN_RFLAGS
8436 | CPUMCTX_EXTRN_CR0 | CPUMCTX_EXTRN_CR3 | CPUMCTX_EXTRN_CR4 | CPUMCTX_EXTRN_EFER);
8437 if (CPUMIsGuestInLongModeEx(&pVCpu->cpum.GstCtx))
8438 {
8439 /* Ideally, IEM should just handle all these special #UD situations, but
8440 we don't quite trust things to behave optimially when doing that. So,
8441 for now we'll restrict ourselves to a handful of possible sysenter and
8442 sysexit encodings that we filter right here. */
8443 uint8_t abInstr[SVM_CTRL_GUEST_INSTR_BYTES_MAX];
8444 uint8_t cbInstr = pVmcb->ctrl.cbInstrFetched;
8445 uint32_t const uCpl = CPUMGetGuestCPL(pVCpu);
8446 uint8_t const cbMin = uCpl != 0 ? 2 : 1 + 2;
8447 RTGCPTR const GCPtrInstr = pVCpu->cpum.GstCtx.rip + pVCpu->cpum.GstCtx.cs.u64Base;
8448 if (cbInstr < cbMin || cbInstr > SVM_CTRL_GUEST_INSTR_BYTES_MAX)
8449 {
8450 cbInstr = cbMin;
8451 int rc2 = PGMPhysSimpleReadGCPtr(pVCpu, abInstr, GCPtrInstr, cbInstr);
8452 AssertRCStmt(rc2, cbInstr = 0);
8453 }
8454 else
8455 memcpy(abInstr, pVmcb->ctrl.abInstr, cbInstr); /* unlikely */
8456 if ( cbInstr == 0 /* read error */
8457 || (cbInstr >= 2 && abInstr[0] == 0x0f && abInstr[1] == 0x34) /* sysenter */
8458 || ( uCpl == 0
8459 && ( ( cbInstr >= 2 && abInstr[0] == 0x0f && abInstr[1] == 0x35) /* sysexit */
8460 || ( cbInstr >= 3 && abInstr[1] == 0x0f && abInstr[2] == 0x35 /* rex.w sysexit */
8461 && (abInstr[0] & (X86_OP_REX_W | 0xf0)) == X86_OP_REX_W))))
8462 {
8463 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_MUST_MASK
8464 | CPUMCTX_EXTRN_SREG_MASK /* without ES+DS+GS the app will #GP later - go figure */);
8465 Log6(("hmR0SvmExitXcptUD: sysenter/sysexit: %.*Rhxs at %#llx CPL=%u\n", cbInstr, abInstr, GCPtrInstr, uCpl));
8466 rcStrict = IEMExecOneWithPrefetchedByPC(pVCpu, CPUMCTX2CORE(&pVCpu->cpum.GstCtx), GCPtrInstr, abInstr, cbInstr);
8467 Log6(("hmR0SvmExitXcptUD: sysenter/sysexit: rcStrict=%Rrc %04x:%08RX64 %08RX64 %04x:%08RX64\n",
8468 VBOXSTRICTRC_VAL(rcStrict), pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip, pVCpu->cpum.GstCtx.rflags.u,
8469 pVCpu->cpum.GstCtx.ss.Sel, pVCpu->cpum.GstCtx.rsp));
8470 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestUD);
8471 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK); /** @todo Lazy bird. */
8472 if (rcStrict == VINF_IEM_RAISED_XCPT)
8473 rcStrict = VINF_SUCCESS;
8474 return rcStrict;
8475 }
8476 Log6(("hmR0SvmExitXcptUD: not sysenter/sysexit: %.*Rhxs at %#llx CPL=%u\n", cbInstr, abInstr, GCPtrInstr, uCpl));
8477 }
8478 else
8479 Log6(("hmR0SvmExitXcptUD: not in long mode at %04x:%llx\n", pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip));
8480 }
8481
8482 /* If the GIM #UD exception handler didn't succeed for some reason or wasn't needed, raise #UD. */
8483 if (RT_FAILURE(rcStrict))
8484 {
8485 hmR0SvmSetPendingXcptUD(pVCpu);
8486 rcStrict = VINF_SUCCESS;
8487 }
8488
8489 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestUD);
8490 return rcStrict;
8491}
8492
8493
8494/**
8495 * \#VMEXIT handler for math-fault exceptions (SVM_EXIT_XCPT_16).
8496 * Conditional \#VMEXIT.
8497 */
8498HMSVM_EXIT_DECL hmR0SvmExitXcptMF(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
8499{
8500 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
8501 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, HMSVM_CPUMCTX_EXTRN_ALL);
8502 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestMF);
8503
8504 PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
8505 PSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
8506
8507 /* Paranoia; Ensure we cannot be called as a result of event delivery. */
8508 Assert(!pVmcb->ctrl.ExitIntInfo.n.u1Valid); NOREF(pVmcb);
8509
8510 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestMF);
8511
8512 if (!(pCtx->cr0 & X86_CR0_NE))
8513 {
8514 PDISSTATE pDis = &pVCpu->hmr0.s.svm.DisState;
8515 unsigned cbInstr;
8516 int rc = EMInterpretDisasCurrent(pVCpu, pDis, &cbInstr);
8517 if (RT_SUCCESS(rc))
8518 {
8519 /* Convert a #MF into a FERR -> IRQ 13. See @bugref{6117}. */
8520 rc = PDMIsaSetIrq(pVCpu->CTX_SUFF(pVM), 13 /* u8Irq */, 1 /* u8Level */, 0 /* uTagSrc */);
8521 if (RT_SUCCESS(rc))
8522 hmR0SvmAdvanceRip(pVCpu, cbInstr);
8523 }
8524 else
8525 Log4Func(("EMInterpretDisasCurrent returned %Rrc uOpCode=%#x\n", rc, pDis->pCurInstr->uOpcode));
8526 return rc;
8527 }
8528
8529 hmR0SvmSetPendingXcptMF(pVCpu);
8530 return VINF_SUCCESS;
8531}
8532
8533
8534/**
8535 * \#VMEXIT handler for debug exceptions (SVM_EXIT_XCPT_1). Conditional
8536 * \#VMEXIT.
8537 */
8538HMSVM_EXIT_DECL hmR0SvmExitXcptDB(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
8539{
8540 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
8541 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, HMSVM_CPUMCTX_EXTRN_ALL);
8542 HMSVM_CHECK_EXIT_DUE_TO_EVENT_DELIVERY(pVCpu, pSvmTransient);
8543 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDB);
8544
8545 if (RT_UNLIKELY(pVCpu->hm.s.Event.fPending))
8546 {
8547 STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectInterpret);
8548 return VINF_EM_RAW_INJECT_TRPM_EVENT;
8549 }
8550
8551 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDB);
8552
8553 /*
8554 * This can be a fault-type #DB (instruction breakpoint) or a trap-type #DB (data
8555 * breakpoint). However, for both cases DR6 and DR7 are updated to what the exception
8556 * handler expects. See AMD spec. 15.12.2 "#DB (Debug)".
8557 */
8558 PVMCC pVM = pVCpu->CTX_SUFF(pVM);
8559 PSVMVMCB pVmcb = pVCpu->hmr0.s.svm.pVmcb;
8560 int rc = DBGFTrap01Handler(pVM, pVCpu, &pVCpu->cpum.GstCtx, pVmcb->guest.u64DR6, pVCpu->hm.s.fSingleInstruction);
8561 if (rc == VINF_EM_RAW_GUEST_TRAP)
8562 {
8563 Log5(("hmR0SvmExitXcptDB: DR6=%#RX64 -> guest trap\n", pVmcb->guest.u64DR6));
8564 if (CPUMIsHyperDebugStateActive(pVCpu))
8565 CPUMSetGuestDR6(pVCpu, CPUMGetGuestDR6(pVCpu) | pVmcb->guest.u64DR6);
8566
8567 /* Reflect the exception back to the guest. */
8568 hmR0SvmSetPendingXcptDB(pVCpu);
8569 rc = VINF_SUCCESS;
8570 }
8571
8572 /*
8573 * Update DR6.
8574 */
8575 if (CPUMIsHyperDebugStateActive(pVCpu))
8576 {
8577 Log5(("hmR0SvmExitXcptDB: DR6=%#RX64 -> %Rrc\n", pVmcb->guest.u64DR6, rc));
8578 pVmcb->guest.u64DR6 = X86_DR6_INIT_VAL;
8579 pVmcb->ctrl.u32VmcbCleanBits &= ~HMSVM_VMCB_CLEAN_DRX;
8580 }
8581 else
8582 {
8583 AssertMsg(rc == VINF_SUCCESS, ("rc=%Rrc\n", rc));
8584 Assert(!pVCpu->hm.s.fSingleInstruction && !DBGFIsStepping(pVCpu));
8585 }
8586
8587 return rc;
8588}
8589
8590
8591/**
8592 * \#VMEXIT handler for alignment check exceptions (SVM_EXIT_XCPT_17).
8593 * Conditional \#VMEXIT.
8594 */
8595HMSVM_EXIT_DECL hmR0SvmExitXcptAC(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
8596{
8597 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
8598 HMSVM_CHECK_EXIT_DUE_TO_EVENT_DELIVERY(pVCpu, pSvmTransient);
8599 STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatExitGuestAC);
8600
8601 SVMEVENT Event;
8602 Event.u = 0;
8603 Event.n.u1Valid = 1;
8604 Event.n.u3Type = SVM_EVENT_EXCEPTION;
8605 Event.n.u8Vector = X86_XCPT_AC;
8606 Event.n.u1ErrorCodeValid = 1;
8607 hmR0SvmSetPendingEvent(pVCpu, &Event, 0 /* GCPtrFaultAddress */);
8608 return VINF_SUCCESS;
8609}
8610
8611
8612/**
8613 * \#VMEXIT handler for breakpoint exceptions (SVM_EXIT_XCPT_3).
8614 * Conditional \#VMEXIT.
8615 */
8616HMSVM_EXIT_DECL hmR0SvmExitXcptBP(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
8617{
8618 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
8619 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, HMSVM_CPUMCTX_EXTRN_ALL);
8620 HMSVM_CHECK_EXIT_DUE_TO_EVENT_DELIVERY(pVCpu, pSvmTransient);
8621 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestBP);
8622
8623 VBOXSTRICTRC rc = DBGFTrap03Handler(pVCpu->CTX_SUFF(pVM), pVCpu, &pVCpu->cpum.GstCtx);
8624 if (rc == VINF_EM_RAW_GUEST_TRAP)
8625 {
8626 SVMEVENT Event;
8627 Event.u = 0;
8628 Event.n.u1Valid = 1;
8629 Event.n.u3Type = SVM_EVENT_EXCEPTION;
8630 Event.n.u8Vector = X86_XCPT_BP;
8631 hmR0SvmSetPendingEvent(pVCpu, &Event, 0 /* GCPtrFaultAddress */);
8632 rc = VINF_SUCCESS;
8633 }
8634
8635 Assert(rc == VINF_SUCCESS || rc == VINF_EM_DBG_BREAKPOINT);
8636 return rc;
8637}
8638
8639
8640/**
8641 * Hacks its way around the lovely mesa driver's backdoor accesses.
8642 *
8643 * @sa hmR0VmxHandleMesaDrvGp
8644 */
8645static int hmR0SvmHandleMesaDrvGp(PVMCPUCC pVCpu, PCPUMCTX pCtx, PCSVMVMCB pVmcb)
8646{
8647 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, CPUMCTX_EXTRN_CS | CPUMCTX_EXTRN_RIP | CPUMCTX_EXTRN_RFLAGS | CPUMCTX_EXTRN_GPRS_MASK);
8648 Log(("hmR0SvmHandleMesaDrvGp: at %04x:%08RX64 rcx=%RX64 rbx=%RX64\n",
8649 pVmcb->guest.CS.u16Sel, pVmcb->guest.u64RIP, pCtx->rcx, pCtx->rbx));
8650 RT_NOREF(pCtx, pVmcb);
8651
8652 /* For now we'll just skip the instruction. */
8653 hmR0SvmAdvanceRip(pVCpu, 1);
8654 return VINF_SUCCESS;
8655}
8656
8657
8658/**
8659 * Checks if the \#GP'ing instruction is the mesa driver doing it's lovely
8660 * backdoor logging w/o checking what it is running inside.
8661 *
8662 * This recognizes an "IN EAX,DX" instruction executed in flat ring-3, with the
8663 * backdoor port and magic numbers loaded in registers.
8664 *
8665 * @returns true if it is, false if it isn't.
8666 * @sa hmR0VmxIsMesaDrvGp
8667 */
8668DECLINLINE(bool) hmR0SvmIsMesaDrvGp(PVMCPUCC pVCpu, PCPUMCTX pCtx, PCSVMVMCB pVmcb)
8669{
8670 /* Check magic and port. */
8671 Assert(!(pCtx->fExtrn & (CPUMCTX_EXTRN_RDX | CPUMCTX_EXTRN_RCX)));
8672 /*Log8(("hmR0SvmIsMesaDrvGp: rax=%RX64 rdx=%RX64\n", pCtx->fExtrn & CPUMCTX_EXTRN_RAX ? pVmcb->guest.u64RAX : pCtx->rax, pCtx->rdx));*/
8673 if (pCtx->dx != UINT32_C(0x5658))
8674 return false;
8675 if ((pCtx->fExtrn & CPUMCTX_EXTRN_RAX ? pVmcb->guest.u64RAX : pCtx->rax) != UINT32_C(0x564d5868))
8676 return false;
8677
8678 /* Check that it is #GP(0). */
8679 if (pVmcb->ctrl.u64ExitInfo1 != 0)
8680 return false;
8681
8682 /* Flat ring-3 CS. */
8683 /*Log8(("hmR0SvmIsMesaDrvGp: u8CPL=%d base=%RX64\n", pVmcb->guest.u8CPL, pCtx->fExtrn & CPUMCTX_EXTRN_CS ? pVmcb->guest.CS.u64Base : pCtx->cs.u64Base));*/
8684 if (pVmcb->guest.u8CPL != 3)
8685 return false;
8686 if ((pCtx->fExtrn & CPUMCTX_EXTRN_CS ? pVmcb->guest.CS.u64Base : pCtx->cs.u64Base) != 0)
8687 return false;
8688
8689 /* 0xed: IN eAX,dx */
8690 if (pVmcb->ctrl.cbInstrFetched < 1) /* unlikely, it turns out. */
8691 {
8692 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, CPUMCTX_EXTRN_CS | CPUMCTX_EXTRN_RIP | CPUMCTX_EXTRN_GPRS_MASK
8693 | CPUMCTX_EXTRN_CR0 | CPUMCTX_EXTRN_CR3 | CPUMCTX_EXTRN_CR4 | CPUMCTX_EXTRN_EFER);
8694 uint8_t abInstr[1];
8695 int rc = PGMPhysSimpleReadGCPtr(pVCpu, abInstr, pCtx->rip, sizeof(abInstr));
8696 /*Log8(("hmR0SvmIsMesaDrvGp: PGMPhysSimpleReadGCPtr -> %Rrc %#x\n", rc, abInstr[0])); */
8697 if (RT_FAILURE(rc))
8698 return false;
8699 if (abInstr[0] != 0xed)
8700 return false;
8701 }
8702 else
8703 {
8704 /*Log8(("hmR0SvmIsMesaDrvGp: %#x\n", pVmcb->ctrl.abInstr));*/
8705 if (pVmcb->ctrl.abInstr[0] != 0xed)
8706 return false;
8707 }
8708 return true;
8709}
8710
8711
8712/**
8713 * \#VMEXIT handler for general protection faults (SVM_EXIT_XCPT_BP).
8714 * Conditional \#VMEXIT.
8715 */
8716HMSVM_EXIT_DECL hmR0SvmExitXcptGP(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
8717{
8718 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
8719 HMSVM_CHECK_EXIT_DUE_TO_EVENT_DELIVERY(pVCpu, pSvmTransient);
8720 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestGP);
8721
8722 PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
8723 Assert(pSvmTransient->u64ExitCode == pVmcb->ctrl.u64ExitCode);
8724
8725 PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
8726 if ( !pVCpu->hm.s.fTrapXcptGpForLovelyMesaDrv
8727 || !hmR0SvmIsMesaDrvGp(pVCpu, pCtx, pVmcb))
8728 {
8729 SVMEVENT Event;
8730 Event.u = 0;
8731 Event.n.u1Valid = 1;
8732 Event.n.u3Type = SVM_EVENT_EXCEPTION;
8733 Event.n.u8Vector = X86_XCPT_GP;
8734 Event.n.u1ErrorCodeValid = 1;
8735 Event.n.u32ErrorCode = (uint32_t)pVmcb->ctrl.u64ExitInfo1;
8736 hmR0SvmSetPendingEvent(pVCpu, &Event, 0 /* GCPtrFaultAddress */);
8737 return VINF_SUCCESS;
8738 }
8739 return hmR0SvmHandleMesaDrvGp(pVCpu, pCtx, pVmcb);
8740}
8741
8742
8743/**
8744 * \#VMEXIT handler for generic exceptions. Conditional \#VMEXIT.
8745 */
8746HMSVM_EXIT_DECL hmR0SvmExitXcptGeneric(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
8747{
8748 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
8749 HMSVM_CHECK_EXIT_DUE_TO_EVENT_DELIVERY(pVCpu, pSvmTransient);
8750
8751 PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
8752 uint8_t const uVector = pVmcb->ctrl.u64ExitCode - SVM_EXIT_XCPT_0;
8753 uint32_t const uErrCode = pVmcb->ctrl.u64ExitInfo1;
8754 Assert(pSvmTransient->u64ExitCode == pVmcb->ctrl.u64ExitCode);
8755 Assert(uVector <= X86_XCPT_LAST);
8756 Log4Func(("uVector=%#x uErrCode=%u\n", uVector, uErrCode));
8757
8758 SVMEVENT Event;
8759 Event.u = 0;
8760 Event.n.u1Valid = 1;
8761 Event.n.u3Type = SVM_EVENT_EXCEPTION;
8762 Event.n.u8Vector = uVector;
8763 switch (uVector)
8764 {
8765 /* Shouldn't be here for reflecting #PFs (among other things, the fault address isn't passed along). */
8766 case X86_XCPT_PF: AssertMsgFailed(("hmR0SvmExitXcptGeneric: Unexpected exception")); return VERR_SVM_IPE_5;
8767 case X86_XCPT_DF:
8768 case X86_XCPT_TS:
8769 case X86_XCPT_NP:
8770 case X86_XCPT_SS:
8771 case X86_XCPT_GP:
8772 case X86_XCPT_AC:
8773 {
8774 Event.n.u1ErrorCodeValid = 1;
8775 Event.n.u32ErrorCode = uErrCode;
8776 break;
8777 }
8778 }
8779
8780#ifdef VBOX_WITH_STATISTICS
8781 switch (uVector)
8782 {
8783 case X86_XCPT_DE: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDE); break;
8784 case X86_XCPT_DB: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDB); break;
8785 case X86_XCPT_BP: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestBP); break;
8786 case X86_XCPT_OF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestOF); break;
8787 case X86_XCPT_BR: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestBR); break;
8788 case X86_XCPT_UD: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestUD); break;
8789 case X86_XCPT_NM: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestOF); break;
8790 case X86_XCPT_DF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDF); break;
8791 case X86_XCPT_TS: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestTS); break;
8792 case X86_XCPT_NP: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestNP); break;
8793 case X86_XCPT_SS: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestSS); break;
8794 case X86_XCPT_GP: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestGP); break;
8795 case X86_XCPT_PF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF); break;
8796 case X86_XCPT_MF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestMF); break;
8797 case X86_XCPT_AC: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestAC); break;
8798 case X86_XCPT_XF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestXF); break;
8799 default:
8800 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestXcpUnk);
8801 break;
8802 }
8803#endif
8804
8805 hmR0SvmSetPendingEvent(pVCpu, &Event, 0 /* GCPtrFaultAddress */);
8806 return VINF_SUCCESS;
8807}
8808
8809
8810/**
8811 * \#VMEXIT handler for software interrupt (INTn). Conditional \#VMEXIT (debug).
8812 */
8813HMSVM_EXIT_DECL hmR0SvmExitSwInt(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
8814{
8815 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
8816 HMSVM_CHECK_EXIT_DUE_TO_EVENT_DELIVERY(pVCpu, pSvmTransient);
8817
8818 PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
8819 SVMEVENT Event;
8820 Event.u = 0;
8821 Event.n.u1Valid = 1;
8822 Event.n.u3Type = SVM_EVENT_SOFTWARE_INT;
8823 Event.n.u8Vector = pVmcb->ctrl.u64ExitInfo1 & 0xff;
8824 Log4Func(("uVector=%#x\n", Event.n.u8Vector));
8825 hmR0SvmSetPendingEvent(pVCpu, &Event, 0 /* GCPtrFaultAddress */);
8826 return VINF_SUCCESS;
8827}
8828
8829
8830/**
8831 * Generic exit handler that interprets the current instruction
8832 *
8833 * Useful exit that only gets triggered by dtrace and the debugger. Caller does
8834 * the exit logging, and this function does the rest.
8835 */
8836static VBOXSTRICTRC hmR0SvmExitInterpretInstruction(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient,
8837 uint64_t fExtraImport, uint64_t fHmChanged)
8838{
8839#if 1
8840 RT_NOREF(pSvmTransient);
8841 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_MUST_MASK | fExtraImport);
8842 VBOXSTRICTRC rcStrict = IEMExecOne(pVCpu);
8843 if (rcStrict == VINF_SUCCESS)
8844 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, fHmChanged | HM_CHANGED_GUEST_RFLAGS | HM_CHANGED_GUEST_RIP);
8845 else
8846 {
8847 Log4Func(("IEMExecOne -> %Rrc\n", VBOXSTRICTRC_VAL(rcStrict) ));
8848 if (rcStrict == VINF_IEM_RAISED_XCPT)
8849 {
8850 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK | fHmChanged);
8851 rcStrict = VINF_SUCCESS;
8852 }
8853 else
8854 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, fHmChanged);
8855 }
8856 return rcStrict;
8857#else
8858 RT_NOREF(pVCpu, pSvmTransient, fExtraImport, fHmChanged);
8859 return VINF_EM_RAW_EMULATE_INSTR;
8860#endif
8861}
8862
8863
8864/**
8865 * \#VMEXIT handler for STR. Conditional \#VMEXIT (debug).
8866 */
8867HMSVM_EXIT_DECL hmR0SvmExitTrRead(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
8868{
8869 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
8870 Log4Func(("%04x:%08RX64\n", pSvmTransient->pVmcb->guest.CS.u16Sel, pSvmTransient->pVmcb->guest.u64RIP));
8871 return hmR0SvmExitInterpretInstruction(pVCpu, pSvmTransient, CPUMCTX_EXTRN_TR, 0);
8872}
8873
8874
8875/**
8876 * \#VMEXIT handler for LTR. Conditional \#VMEXIT (OS/2 TLB workaround, debug).
8877 */
8878HMSVM_EXIT_DECL hmR0SvmExitTrWrite(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
8879{
8880 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
8881
8882 /* Workaround for lack of TLB flushing in OS/2 when returning to protected
8883 mode after a real mode call (like a BIOS call). See ticketref:20625
8884 comment 14. */
8885 PVMCC pVM = pVCpu->CTX_SUFF(pVM);
8886 if (pVM->hm.s.fMissingOS2TlbFlushWorkaround)
8887 {
8888 Log4Func(("%04x:%08RX64 TLB flush\n", pSvmTransient->pVmcb->guest.CS.u16Sel, pSvmTransient->pVmcb->guest.u64RIP));
8889 VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
8890 }
8891 else
8892 Log4Func(("%04x:%08RX64\n", pSvmTransient->pVmcb->guest.CS.u16Sel, pSvmTransient->pVmcb->guest.u64RIP));
8893
8894 return hmR0SvmExitInterpretInstruction(pVCpu, pSvmTransient, CPUMCTX_EXTRN_TR | CPUMCTX_EXTRN_GDTR, HM_CHANGED_GUEST_TR);
8895}
8896
8897
8898#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
8899/**
8900 * \#VMEXIT handler for CLGI (SVM_EXIT_CLGI). Conditional \#VMEXIT.
8901 */
8902HMSVM_EXIT_DECL hmR0SvmExitClgi(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
8903{
8904 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
8905
8906 PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
8907 Assert(pVmcb);
8908 Assert(!pVmcb->ctrl.IntCtrl.n.u1VGifEnable);
8909
8910 VBOXSTRICTRC rcStrict;
8911 bool const fSupportsNextRipSave = hmR0SvmSupportsNextRipSave(pVCpu);
8912 uint64_t const fImport = CPUMCTX_EXTRN_HWVIRT;
8913 if (fSupportsNextRipSave)
8914 {
8915 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK | fImport);
8916 uint8_t const cbInstr = pVmcb->ctrl.u64NextRIP - pVCpu->cpum.GstCtx.rip;
8917 rcStrict = IEMExecDecodedClgi(pVCpu, cbInstr);
8918 }
8919 else
8920 {
8921 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_MUST_MASK | fImport);
8922 rcStrict = IEMExecOne(pVCpu);
8923 }
8924
8925 if (rcStrict == VINF_SUCCESS)
8926 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_HWVIRT);
8927 else if (rcStrict == VINF_IEM_RAISED_XCPT)
8928 {
8929 rcStrict = VINF_SUCCESS;
8930 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
8931 }
8932 HMSVM_CHECK_SINGLE_STEP(pVCpu, rcStrict);
8933 return rcStrict;
8934}
8935
8936
8937/**
8938 * \#VMEXIT handler for STGI (SVM_EXIT_STGI). Conditional \#VMEXIT.
8939 */
8940HMSVM_EXIT_DECL hmR0SvmExitStgi(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
8941{
8942 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
8943
8944 /*
8945 * When VGIF is not used we always intercept STGI instructions. When VGIF is used,
8946 * we only intercept STGI when events are pending for GIF to become 1.
8947 */
8948 PSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
8949 if (pVmcb->ctrl.IntCtrl.n.u1VGifEnable)
8950 hmR0SvmClearCtrlIntercept(pVCpu, pVmcb, SVM_CTRL_INTERCEPT_STGI);
8951
8952 VBOXSTRICTRC rcStrict;
8953 bool const fSupportsNextRipSave = hmR0SvmSupportsNextRipSave(pVCpu);
8954 uint64_t const fImport = CPUMCTX_EXTRN_HWVIRT;
8955 if (fSupportsNextRipSave)
8956 {
8957 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK | fImport);
8958 uint8_t const cbInstr = pVmcb->ctrl.u64NextRIP - pVCpu->cpum.GstCtx.rip;
8959 rcStrict = IEMExecDecodedStgi(pVCpu, cbInstr);
8960 }
8961 else
8962 {
8963 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_MUST_MASK | fImport);
8964 rcStrict = IEMExecOne(pVCpu);
8965 }
8966
8967 if (rcStrict == VINF_SUCCESS)
8968 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_HWVIRT);
8969 else if (rcStrict == VINF_IEM_RAISED_XCPT)
8970 {
8971 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
8972 rcStrict = VINF_SUCCESS;
8973 }
8974 HMSVM_CHECK_SINGLE_STEP(pVCpu, rcStrict);
8975 return rcStrict;
8976}
8977
8978
8979/**
8980 * \#VMEXIT handler for VMLOAD (SVM_EXIT_VMLOAD). Conditional \#VMEXIT.
8981 */
8982HMSVM_EXIT_DECL hmR0SvmExitVmload(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
8983{
8984 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
8985
8986 PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
8987 Assert(pVmcb);
8988 Assert(!pVmcb->ctrl.LbrVirt.n.u1VirtVmsaveVmload);
8989
8990 VBOXSTRICTRC rcStrict;
8991 bool const fSupportsNextRipSave = hmR0SvmSupportsNextRipSave(pVCpu);
8992 uint64_t const fImport = CPUMCTX_EXTRN_FS | CPUMCTX_EXTRN_GS | CPUMCTX_EXTRN_KERNEL_GS_BASE
8993 | CPUMCTX_EXTRN_TR | CPUMCTX_EXTRN_LDTR | CPUMCTX_EXTRN_SYSCALL_MSRS
8994 | CPUMCTX_EXTRN_SYSENTER_MSRS;
8995 if (fSupportsNextRipSave)
8996 {
8997 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK | fImport);
8998 uint8_t const cbInstr = pVmcb->ctrl.u64NextRIP - pVCpu->cpum.GstCtx.rip;
8999 rcStrict = IEMExecDecodedVmload(pVCpu, cbInstr);
9000 }
9001 else
9002 {
9003 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_MUST_MASK | fImport);
9004 rcStrict = IEMExecOne(pVCpu);
9005 }
9006
9007 if (rcStrict == VINF_SUCCESS)
9008 {
9009 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_FS | HM_CHANGED_GUEST_GS
9010 | HM_CHANGED_GUEST_TR | HM_CHANGED_GUEST_LDTR
9011 | HM_CHANGED_GUEST_KERNEL_GS_BASE | HM_CHANGED_GUEST_SYSCALL_MSRS
9012 | HM_CHANGED_GUEST_SYSENTER_MSR_MASK);
9013 }
9014 else if (rcStrict == VINF_IEM_RAISED_XCPT)
9015 {
9016 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
9017 rcStrict = VINF_SUCCESS;
9018 }
9019 HMSVM_CHECK_SINGLE_STEP(pVCpu, rcStrict);
9020 return rcStrict;
9021}
9022
9023
9024/**
9025 * \#VMEXIT handler for VMSAVE (SVM_EXIT_VMSAVE). Conditional \#VMEXIT.
9026 */
9027HMSVM_EXIT_DECL hmR0SvmExitVmsave(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
9028{
9029 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
9030
9031 PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
9032 Assert(!pVmcb->ctrl.LbrVirt.n.u1VirtVmsaveVmload);
9033
9034 VBOXSTRICTRC rcStrict;
9035 bool const fSupportsNextRipSave = hmR0SvmSupportsNextRipSave(pVCpu);
9036 if (fSupportsNextRipSave)
9037 {
9038 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK);
9039 uint8_t const cbInstr = pVmcb->ctrl.u64NextRIP - pVCpu->cpum.GstCtx.rip;
9040 rcStrict = IEMExecDecodedVmsave(pVCpu, cbInstr);
9041 }
9042 else
9043 {
9044 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_MUST_MASK);
9045 rcStrict = IEMExecOne(pVCpu);
9046 }
9047
9048 if (rcStrict == VINF_IEM_RAISED_XCPT)
9049 {
9050 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
9051 rcStrict = VINF_SUCCESS;
9052 }
9053 HMSVM_CHECK_SINGLE_STEP(pVCpu, rcStrict);
9054 return rcStrict;
9055}
9056
9057
9058/**
9059 * \#VMEXIT handler for INVLPGA (SVM_EXIT_INVLPGA). Conditional \#VMEXIT.
9060 */
9061HMSVM_EXIT_DECL hmR0SvmExitInvlpga(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
9062{
9063 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
9064
9065 VBOXSTRICTRC rcStrict;
9066 bool const fSupportsNextRipSave = hmR0SvmSupportsNextRipSave(pVCpu);
9067 if (fSupportsNextRipSave)
9068 {
9069 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_EXEC_DECODED_NO_MEM_MASK);
9070 PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
9071 uint8_t const cbInstr = pVmcb->ctrl.u64NextRIP - pVCpu->cpum.GstCtx.rip;
9072 rcStrict = IEMExecDecodedInvlpga(pVCpu, cbInstr);
9073 }
9074 else
9075 {
9076 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, IEM_CPUMCTX_EXTRN_MUST_MASK);
9077 rcStrict = IEMExecOne(pVCpu);
9078 }
9079
9080 if (rcStrict == VINF_IEM_RAISED_XCPT)
9081 {
9082 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
9083 rcStrict = VINF_SUCCESS;
9084 }
9085 HMSVM_CHECK_SINGLE_STEP(pVCpu, rcStrict);
9086 return rcStrict;
9087}
9088
9089
9090/**
9091 * \#VMEXIT handler for STGI (SVM_EXIT_VMRUN). Conditional \#VMEXIT.
9092 */
9093HMSVM_EXIT_DECL hmR0SvmExitVmrun(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
9094{
9095 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
9096 /* We shall import the entire state here, just in case we enter and continue execution of
9097 the nested-guest with hardware-assisted SVM in ring-0, we would be switching VMCBs and
9098 could lose lose part of CPU state. */
9099 HMSVM_CPUMCTX_IMPORT_STATE(pVCpu, HMSVM_CPUMCTX_EXTRN_ALL);
9100
9101 VBOXSTRICTRC rcStrict;
9102 bool const fSupportsNextRipSave = hmR0SvmSupportsNextRipSave(pVCpu);
9103 STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitVmentry, z);
9104 if (fSupportsNextRipSave)
9105 {
9106 PCSVMVMCB pVmcb = hmR0SvmGetCurrentVmcb(pVCpu);
9107 uint8_t const cbInstr = pVmcb->ctrl.u64NextRIP - pVCpu->cpum.GstCtx.rip;
9108 rcStrict = IEMExecDecodedVmrun(pVCpu, cbInstr);
9109 }
9110 else
9111 {
9112 /* We use IEMExecOneBypassEx() here as it supresses attempt to continue emulating any
9113 instruction(s) when interrupt inhibition is set as part of emulating the VMRUN
9114 instruction itself, see @bugref{7243#c126} */
9115 rcStrict = IEMExecOneBypassEx(pVCpu, CPUMCTX2CORE(&pVCpu->cpum.GstCtx), NULL /* pcbWritten */);
9116 }
9117 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitVmentry, z);
9118
9119 if (rcStrict == VINF_SUCCESS)
9120 {
9121 rcStrict = VINF_SVM_VMRUN;
9122 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_SVM_VMRUN_MASK);
9123 }
9124 else if (rcStrict == VINF_IEM_RAISED_XCPT)
9125 {
9126 ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_RAISED_XCPT_MASK);
9127 rcStrict = VINF_SUCCESS;
9128 }
9129 HMSVM_CHECK_SINGLE_STEP(pVCpu, rcStrict);
9130 return rcStrict;
9131}
9132
9133
9134/**
9135 * Nested-guest \#VMEXIT handler for debug exceptions (SVM_EXIT_XCPT_1).
9136 * Unconditional \#VMEXIT.
9137 */
9138HMSVM_EXIT_DECL hmR0SvmNestedExitXcptDB(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
9139{
9140 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
9141 HMSVM_CHECK_EXIT_DUE_TO_EVENT_DELIVERY(pVCpu, pSvmTransient);
9142
9143 if (pVCpu->hm.s.Event.fPending)
9144 {
9145 STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectInterpret);
9146 return VINF_EM_RAW_INJECT_TRPM_EVENT;
9147 }
9148
9149 hmR0SvmSetPendingXcptDB(pVCpu);
9150 return VINF_SUCCESS;
9151}
9152
9153
9154/**
9155 * Nested-guest \#VMEXIT handler for breakpoint exceptions (SVM_EXIT_XCPT_3).
9156 * Conditional \#VMEXIT.
9157 */
9158HMSVM_EXIT_DECL hmR0SvmNestedExitXcptBP(PVMCPUCC pVCpu, PSVMTRANSIENT pSvmTransient)
9159{
9160 HMSVM_VALIDATE_EXIT_HANDLER_PARAMS(pVCpu, pSvmTransient);
9161 HMSVM_CHECK_EXIT_DUE_TO_EVENT_DELIVERY(pVCpu, pSvmTransient);
9162
9163 SVMEVENT Event;
9164 Event.u = 0;
9165 Event.n.u1Valid = 1;
9166 Event.n.u3Type = SVM_EVENT_EXCEPTION;
9167 Event.n.u8Vector = X86_XCPT_BP;
9168 hmR0SvmSetPendingEvent(pVCpu, &Event, 0 /* GCPtrFaultAddress */);
9169 return VINF_SUCCESS;
9170}
9171#endif /* VBOX_WITH_NESTED_HWVIRT_SVM */
9172
9173/** @} */
9174
Note: See TracBrowser for help on using the repository browser.

© 2024 Oracle Support Privacy / Do Not Sell My Info Terms of Use Trademark Policy Automated Access Etiquette