VirtualBox

source: vbox/trunk/src/VBox/VMM/VMMAll/PGMAll.cpp@ 50653

Last change on this file since 50653 was 49486, checked in by vboxsync, 11 years ago

VMM: Warnings.

  • Property svn:eol-style set to native
  • Property svn:keywords set to Id Revision
File size: 92.5 KB
Line 
1/* $Id: PGMAll.cpp 49486 2013-11-14 16:38:53Z vboxsync $ */
2/** @file
3 * PGM - Page Manager and Monitor - All context code.
4 */
5
6/*
7 * Copyright (C) 2006-2012 Oracle Corporation
8 *
9 * This file is part of VirtualBox Open Source Edition (OSE), as
10 * available from http://www.virtualbox.org. This file is free software;
11 * you can redistribute it and/or modify it under the terms of the GNU
12 * General Public License (GPL) as published by the Free Software
13 * Foundation, in version 2 as it comes in the "COPYING" file of the
14 * VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15 * hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16 */
17
18/*******************************************************************************
19* Header Files *
20*******************************************************************************/
21#define LOG_GROUP LOG_GROUP_PGM
22#include <VBox/vmm/pgm.h>
23#include <VBox/vmm/cpum.h>
24#include <VBox/vmm/selm.h>
25#include <VBox/vmm/iom.h>
26#include <VBox/sup.h>
27#include <VBox/vmm/mm.h>
28#include <VBox/vmm/stam.h>
29#include <VBox/vmm/csam.h>
30#include <VBox/vmm/patm.h>
31#include <VBox/vmm/trpm.h>
32#ifdef VBOX_WITH_REM
33# include <VBox/vmm/rem.h>
34#endif
35#include <VBox/vmm/em.h>
36#include <VBox/vmm/hm.h>
37#include <VBox/vmm/hm_vmx.h>
38#include "PGMInternal.h"
39#include <VBox/vmm/vm.h>
40#include "PGMInline.h"
41#include <iprt/assert.h>
42#include <iprt/asm-amd64-x86.h>
43#include <iprt/string.h>
44#include <VBox/log.h>
45#include <VBox/param.h>
46#include <VBox/err.h>
47
48
49/*******************************************************************************
50* Structures and Typedefs *
51*******************************************************************************/
52/**
53 * Stated structure for PGM_GST_NAME(HandlerVirtualUpdate) that's
54 * passed to PGM_GST_NAME(VirtHandlerUpdateOne) during enumeration.
55 */
56typedef struct PGMHVUSTATE
57{
58 /** Pointer to the VM. */
59 PVM pVM;
60 /** Pointer to the VMCPU. */
61 PVMCPU pVCpu;
62 /** The todo flags. */
63 RTUINT fTodo;
64 /** The CR4 register value. */
65 uint32_t cr4;
66} PGMHVUSTATE, *PPGMHVUSTATE;
67
68
69/*******************************************************************************
70* Internal Functions *
71*******************************************************************************/
72DECLINLINE(int) pgmShwGetLongModePDPtr(PVMCPU pVCpu, RTGCPTR64 GCPtr, PX86PML4E *ppPml4e, PX86PDPT *ppPdpt, PX86PDPAE *ppPD);
73DECLINLINE(int) pgmShwGetPaePoolPagePD(PVMCPU pVCpu, RTGCPTR GCPtr, PPGMPOOLPAGE *ppShwPde);
74#ifndef IN_RC
75static int pgmShwSyncLongModePDPtr(PVMCPU pVCpu, RTGCPTR64 GCPtr, X86PGPAEUINT uGstPml4e, X86PGPAEUINT uGstPdpe, PX86PDPAE *ppPD);
76static int pgmShwGetEPTPDPtr(PVMCPU pVCpu, RTGCPTR64 GCPtr, PEPTPDPT *ppPdpt, PEPTPD *ppPD);
77#endif
78
79
80/*
81 * Shadow - 32-bit mode
82 */
83#define PGM_SHW_TYPE PGM_TYPE_32BIT
84#define PGM_SHW_NAME(name) PGM_SHW_NAME_32BIT(name)
85#include "PGMAllShw.h"
86
87/* Guest - real mode */
88#define PGM_GST_TYPE PGM_TYPE_REAL
89#define PGM_GST_NAME(name) PGM_GST_NAME_REAL(name)
90#define PGM_BTH_NAME(name) PGM_BTH_NAME_32BIT_REAL(name)
91#define BTH_PGMPOOLKIND_PT_FOR_PT PGMPOOLKIND_32BIT_PT_FOR_PHYS
92#define BTH_PGMPOOLKIND_ROOT PGMPOOLKIND_32BIT_PD_PHYS
93#include "PGMGstDefs.h"
94#include "PGMAllGst.h"
95#include "PGMAllBth.h"
96#undef BTH_PGMPOOLKIND_PT_FOR_PT
97#undef BTH_PGMPOOLKIND_ROOT
98#undef PGM_BTH_NAME
99#undef PGM_GST_TYPE
100#undef PGM_GST_NAME
101
102/* Guest - protected mode */
103#define PGM_GST_TYPE PGM_TYPE_PROT
104#define PGM_GST_NAME(name) PGM_GST_NAME_PROT(name)
105#define PGM_BTH_NAME(name) PGM_BTH_NAME_32BIT_PROT(name)
106#define BTH_PGMPOOLKIND_PT_FOR_PT PGMPOOLKIND_32BIT_PT_FOR_PHYS
107#define BTH_PGMPOOLKIND_ROOT PGMPOOLKIND_32BIT_PD_PHYS
108#include "PGMGstDefs.h"
109#include "PGMAllGst.h"
110#include "PGMAllBth.h"
111#undef BTH_PGMPOOLKIND_PT_FOR_PT
112#undef BTH_PGMPOOLKIND_ROOT
113#undef PGM_BTH_NAME
114#undef PGM_GST_TYPE
115#undef PGM_GST_NAME
116
117/* Guest - 32-bit mode */
118#define PGM_GST_TYPE PGM_TYPE_32BIT
119#define PGM_GST_NAME(name) PGM_GST_NAME_32BIT(name)
120#define PGM_BTH_NAME(name) PGM_BTH_NAME_32BIT_32BIT(name)
121#define BTH_PGMPOOLKIND_PT_FOR_PT PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT
122#define BTH_PGMPOOLKIND_PT_FOR_BIG PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB
123#define BTH_PGMPOOLKIND_ROOT PGMPOOLKIND_32BIT_PD
124#include "PGMGstDefs.h"
125#include "PGMAllGst.h"
126#include "PGMAllBth.h"
127#undef BTH_PGMPOOLKIND_PT_FOR_BIG
128#undef BTH_PGMPOOLKIND_PT_FOR_PT
129#undef BTH_PGMPOOLKIND_ROOT
130#undef PGM_BTH_NAME
131#undef PGM_GST_TYPE
132#undef PGM_GST_NAME
133
134#undef PGM_SHW_TYPE
135#undef PGM_SHW_NAME
136
137
138/*
139 * Shadow - PAE mode
140 */
141#define PGM_SHW_TYPE PGM_TYPE_PAE
142#define PGM_SHW_NAME(name) PGM_SHW_NAME_PAE(name)
143#define PGM_BTH_NAME(name) PGM_BTH_NAME_PAE_REAL(name)
144#include "PGMAllShw.h"
145
146/* Guest - real mode */
147#define PGM_GST_TYPE PGM_TYPE_REAL
148#define PGM_GST_NAME(name) PGM_GST_NAME_REAL(name)
149#define PGM_BTH_NAME(name) PGM_BTH_NAME_PAE_REAL(name)
150#define BTH_PGMPOOLKIND_PT_FOR_PT PGMPOOLKIND_PAE_PT_FOR_PHYS
151#define BTH_PGMPOOLKIND_ROOT PGMPOOLKIND_PAE_PDPT_PHYS
152#include "PGMGstDefs.h"
153#include "PGMAllBth.h"
154#undef BTH_PGMPOOLKIND_PT_FOR_PT
155#undef BTH_PGMPOOLKIND_ROOT
156#undef PGM_BTH_NAME
157#undef PGM_GST_TYPE
158#undef PGM_GST_NAME
159
160/* Guest - protected mode */
161#define PGM_GST_TYPE PGM_TYPE_PROT
162#define PGM_GST_NAME(name) PGM_GST_NAME_PROT(name)
163#define PGM_BTH_NAME(name) PGM_BTH_NAME_PAE_PROT(name)
164#define BTH_PGMPOOLKIND_PT_FOR_PT PGMPOOLKIND_PAE_PT_FOR_PHYS
165#define BTH_PGMPOOLKIND_ROOT PGMPOOLKIND_PAE_PDPT_PHYS
166#include "PGMGstDefs.h"
167#include "PGMAllBth.h"
168#undef BTH_PGMPOOLKIND_PT_FOR_PT
169#undef BTH_PGMPOOLKIND_ROOT
170#undef PGM_BTH_NAME
171#undef PGM_GST_TYPE
172#undef PGM_GST_NAME
173
174/* Guest - 32-bit mode */
175#define PGM_GST_TYPE PGM_TYPE_32BIT
176#define PGM_GST_NAME(name) PGM_GST_NAME_32BIT(name)
177#define PGM_BTH_NAME(name) PGM_BTH_NAME_PAE_32BIT(name)
178#define BTH_PGMPOOLKIND_PT_FOR_PT PGMPOOLKIND_PAE_PT_FOR_32BIT_PT
179#define BTH_PGMPOOLKIND_PT_FOR_BIG PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB
180#define BTH_PGMPOOLKIND_ROOT PGMPOOLKIND_PAE_PDPT_FOR_32BIT
181#include "PGMGstDefs.h"
182#include "PGMAllBth.h"
183#undef BTH_PGMPOOLKIND_PT_FOR_BIG
184#undef BTH_PGMPOOLKIND_PT_FOR_PT
185#undef BTH_PGMPOOLKIND_ROOT
186#undef PGM_BTH_NAME
187#undef PGM_GST_TYPE
188#undef PGM_GST_NAME
189
190
191/* Guest - PAE mode */
192#define PGM_GST_TYPE PGM_TYPE_PAE
193#define PGM_GST_NAME(name) PGM_GST_NAME_PAE(name)
194#define PGM_BTH_NAME(name) PGM_BTH_NAME_PAE_PAE(name)
195#define BTH_PGMPOOLKIND_PT_FOR_PT PGMPOOLKIND_PAE_PT_FOR_PAE_PT
196#define BTH_PGMPOOLKIND_PT_FOR_BIG PGMPOOLKIND_PAE_PT_FOR_PAE_2MB
197#define BTH_PGMPOOLKIND_ROOT PGMPOOLKIND_PAE_PDPT
198#include "PGMGstDefs.h"
199#include "PGMAllGst.h"
200#include "PGMAllBth.h"
201#undef BTH_PGMPOOLKIND_PT_FOR_BIG
202#undef BTH_PGMPOOLKIND_PT_FOR_PT
203#undef BTH_PGMPOOLKIND_ROOT
204#undef PGM_BTH_NAME
205#undef PGM_GST_TYPE
206#undef PGM_GST_NAME
207
208#undef PGM_SHW_TYPE
209#undef PGM_SHW_NAME
210
211
212#ifndef IN_RC /* AMD64 implies VT-x/AMD-V */
213/*
214 * Shadow - AMD64 mode
215 */
216# define PGM_SHW_TYPE PGM_TYPE_AMD64
217# define PGM_SHW_NAME(name) PGM_SHW_NAME_AMD64(name)
218# include "PGMAllShw.h"
219
220/* Guest - protected mode (only used for AMD-V nested paging in 64 bits mode) */
221# define PGM_GST_TYPE PGM_TYPE_PROT
222# define PGM_GST_NAME(name) PGM_GST_NAME_PROT(name)
223# define PGM_BTH_NAME(name) PGM_BTH_NAME_AMD64_PROT(name)
224# define BTH_PGMPOOLKIND_PT_FOR_PT PGMPOOLKIND_PAE_PT_FOR_PHYS
225# define BTH_PGMPOOLKIND_ROOT PGMPOOLKIND_PAE_PD_PHYS
226# include "PGMGstDefs.h"
227# include "PGMAllBth.h"
228# undef BTH_PGMPOOLKIND_PT_FOR_PT
229# undef BTH_PGMPOOLKIND_ROOT
230# undef PGM_BTH_NAME
231# undef PGM_GST_TYPE
232# undef PGM_GST_NAME
233
234# ifdef VBOX_WITH_64_BITS_GUESTS
235/* Guest - AMD64 mode */
236# define PGM_GST_TYPE PGM_TYPE_AMD64
237# define PGM_GST_NAME(name) PGM_GST_NAME_AMD64(name)
238# define PGM_BTH_NAME(name) PGM_BTH_NAME_AMD64_AMD64(name)
239# define BTH_PGMPOOLKIND_PT_FOR_PT PGMPOOLKIND_PAE_PT_FOR_PAE_PT
240# define BTH_PGMPOOLKIND_PT_FOR_BIG PGMPOOLKIND_PAE_PT_FOR_PAE_2MB
241# define BTH_PGMPOOLKIND_ROOT PGMPOOLKIND_64BIT_PML4
242# include "PGMGstDefs.h"
243# include "PGMAllGst.h"
244# include "PGMAllBth.h"
245# undef BTH_PGMPOOLKIND_PT_FOR_BIG
246# undef BTH_PGMPOOLKIND_PT_FOR_PT
247# undef BTH_PGMPOOLKIND_ROOT
248# undef PGM_BTH_NAME
249# undef PGM_GST_TYPE
250# undef PGM_GST_NAME
251# endif /* VBOX_WITH_64_BITS_GUESTS */
252
253# undef PGM_SHW_TYPE
254# undef PGM_SHW_NAME
255
256
257/*
258 * Shadow - Nested paging mode
259 */
260# define PGM_SHW_TYPE PGM_TYPE_NESTED
261# define PGM_SHW_NAME(name) PGM_SHW_NAME_NESTED(name)
262# include "PGMAllShw.h"
263
264/* Guest - real mode */
265# define PGM_GST_TYPE PGM_TYPE_REAL
266# define PGM_GST_NAME(name) PGM_GST_NAME_REAL(name)
267# define PGM_BTH_NAME(name) PGM_BTH_NAME_NESTED_REAL(name)
268# include "PGMGstDefs.h"
269# include "PGMAllBth.h"
270# undef PGM_BTH_NAME
271# undef PGM_GST_TYPE
272# undef PGM_GST_NAME
273
274/* Guest - protected mode */
275# define PGM_GST_TYPE PGM_TYPE_PROT
276# define PGM_GST_NAME(name) PGM_GST_NAME_PROT(name)
277# define PGM_BTH_NAME(name) PGM_BTH_NAME_NESTED_PROT(name)
278# include "PGMGstDefs.h"
279# include "PGMAllBth.h"
280# undef PGM_BTH_NAME
281# undef PGM_GST_TYPE
282# undef PGM_GST_NAME
283
284/* Guest - 32-bit mode */
285# define PGM_GST_TYPE PGM_TYPE_32BIT
286# define PGM_GST_NAME(name) PGM_GST_NAME_32BIT(name)
287# define PGM_BTH_NAME(name) PGM_BTH_NAME_NESTED_32BIT(name)
288# include "PGMGstDefs.h"
289# include "PGMAllBth.h"
290# undef PGM_BTH_NAME
291# undef PGM_GST_TYPE
292# undef PGM_GST_NAME
293
294/* Guest - PAE mode */
295# define PGM_GST_TYPE PGM_TYPE_PAE
296# define PGM_GST_NAME(name) PGM_GST_NAME_PAE(name)
297# define PGM_BTH_NAME(name) PGM_BTH_NAME_NESTED_PAE(name)
298# include "PGMGstDefs.h"
299# include "PGMAllBth.h"
300# undef PGM_BTH_NAME
301# undef PGM_GST_TYPE
302# undef PGM_GST_NAME
303
304# ifdef VBOX_WITH_64_BITS_GUESTS
305/* Guest - AMD64 mode */
306# define PGM_GST_TYPE PGM_TYPE_AMD64
307# define PGM_GST_NAME(name) PGM_GST_NAME_AMD64(name)
308# define PGM_BTH_NAME(name) PGM_BTH_NAME_NESTED_AMD64(name)
309# include "PGMGstDefs.h"
310# include "PGMAllBth.h"
311# undef PGM_BTH_NAME
312# undef PGM_GST_TYPE
313# undef PGM_GST_NAME
314# endif /* VBOX_WITH_64_BITS_GUESTS */
315
316# undef PGM_SHW_TYPE
317# undef PGM_SHW_NAME
318
319
320/*
321 * Shadow - EPT
322 */
323# define PGM_SHW_TYPE PGM_TYPE_EPT
324# define PGM_SHW_NAME(name) PGM_SHW_NAME_EPT(name)
325# include "PGMAllShw.h"
326
327/* Guest - real mode */
328# define PGM_GST_TYPE PGM_TYPE_REAL
329# define PGM_GST_NAME(name) PGM_GST_NAME_REAL(name)
330# define PGM_BTH_NAME(name) PGM_BTH_NAME_EPT_REAL(name)
331# define BTH_PGMPOOLKIND_PT_FOR_PT PGMPOOLKIND_EPT_PT_FOR_PHYS
332# include "PGMGstDefs.h"
333# include "PGMAllBth.h"
334# undef BTH_PGMPOOLKIND_PT_FOR_PT
335# undef PGM_BTH_NAME
336# undef PGM_GST_TYPE
337# undef PGM_GST_NAME
338
339/* Guest - protected mode */
340# define PGM_GST_TYPE PGM_TYPE_PROT
341# define PGM_GST_NAME(name) PGM_GST_NAME_PROT(name)
342# define PGM_BTH_NAME(name) PGM_BTH_NAME_EPT_PROT(name)
343# define BTH_PGMPOOLKIND_PT_FOR_PT PGMPOOLKIND_EPT_PT_FOR_PHYS
344# include "PGMGstDefs.h"
345# include "PGMAllBth.h"
346# undef BTH_PGMPOOLKIND_PT_FOR_PT
347# undef PGM_BTH_NAME
348# undef PGM_GST_TYPE
349# undef PGM_GST_NAME
350
351/* Guest - 32-bit mode */
352# define PGM_GST_TYPE PGM_TYPE_32BIT
353# define PGM_GST_NAME(name) PGM_GST_NAME_32BIT(name)
354# define PGM_BTH_NAME(name) PGM_BTH_NAME_EPT_32BIT(name)
355# define BTH_PGMPOOLKIND_PT_FOR_PT PGMPOOLKIND_EPT_PT_FOR_PHYS
356# include "PGMGstDefs.h"
357# include "PGMAllBth.h"
358# undef BTH_PGMPOOLKIND_PT_FOR_PT
359# undef PGM_BTH_NAME
360# undef PGM_GST_TYPE
361# undef PGM_GST_NAME
362
363/* Guest - PAE mode */
364# define PGM_GST_TYPE PGM_TYPE_PAE
365# define PGM_GST_NAME(name) PGM_GST_NAME_PAE(name)
366# define PGM_BTH_NAME(name) PGM_BTH_NAME_EPT_PAE(name)
367# define BTH_PGMPOOLKIND_PT_FOR_PT PGMPOOLKIND_EPT_PT_FOR_PHYS
368# include "PGMGstDefs.h"
369# include "PGMAllBth.h"
370# undef BTH_PGMPOOLKIND_PT_FOR_PT
371# undef PGM_BTH_NAME
372# undef PGM_GST_TYPE
373# undef PGM_GST_NAME
374
375# ifdef VBOX_WITH_64_BITS_GUESTS
376/* Guest - AMD64 mode */
377# define PGM_GST_TYPE PGM_TYPE_AMD64
378# define PGM_GST_NAME(name) PGM_GST_NAME_AMD64(name)
379# define PGM_BTH_NAME(name) PGM_BTH_NAME_EPT_AMD64(name)
380# define BTH_PGMPOOLKIND_PT_FOR_PT PGMPOOLKIND_EPT_PT_FOR_PHYS
381# include "PGMGstDefs.h"
382# include "PGMAllBth.h"
383# undef BTH_PGMPOOLKIND_PT_FOR_PT
384# undef PGM_BTH_NAME
385# undef PGM_GST_TYPE
386# undef PGM_GST_NAME
387# endif /* VBOX_WITH_64_BITS_GUESTS */
388
389# undef PGM_SHW_TYPE
390# undef PGM_SHW_NAME
391
392#endif /* !IN_RC */
393
394
395#ifndef IN_RING3
396/**
397 * #PF Handler.
398 *
399 * @returns VBox status code (appropriate for trap handling and GC return).
400 * @param pVCpu Pointer to the VMCPU.
401 * @param uErr The trap error code.
402 * @param pRegFrame Trap register frame.
403 * @param pvFault The fault address.
404 */
405VMMDECL(int) PGMTrap0eHandler(PVMCPU pVCpu, RTGCUINT uErr, PCPUMCTXCORE pRegFrame, RTGCPTR pvFault)
406{
407 PVM pVM = pVCpu->CTX_SUFF(pVM);
408
409 Log(("PGMTrap0eHandler: uErr=%RGx pvFault=%RGv eip=%04x:%RGv cr3=%RGp\n", uErr, pvFault, pRegFrame->cs.Sel, (RTGCPTR)pRegFrame->rip, (RTGCPHYS)CPUMGetGuestCR3(pVCpu)));
410 STAM_PROFILE_START(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0e, a);
411 STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = NULL; } );
412
413
414#ifdef VBOX_WITH_STATISTICS
415 /*
416 * Error code stats.
417 */
418 if (uErr & X86_TRAP_PF_US)
419 {
420 if (!(uErr & X86_TRAP_PF_P))
421 {
422 if (uErr & X86_TRAP_PF_RW)
423 STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eUSNotPresentWrite);
424 else
425 STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eUSNotPresentRead);
426 }
427 else if (uErr & X86_TRAP_PF_RW)
428 STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eUSWrite);
429 else if (uErr & X86_TRAP_PF_RSVD)
430 STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eUSReserved);
431 else if (uErr & X86_TRAP_PF_ID)
432 STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eUSNXE);
433 else
434 STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eUSRead);
435 }
436 else
437 { /* Supervisor */
438 if (!(uErr & X86_TRAP_PF_P))
439 {
440 if (uErr & X86_TRAP_PF_RW)
441 STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eSVNotPresentWrite);
442 else
443 STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eSVNotPresentRead);
444 }
445 else if (uErr & X86_TRAP_PF_RW)
446 STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eSVWrite);
447 else if (uErr & X86_TRAP_PF_ID)
448 STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eSNXE);
449 else if (uErr & X86_TRAP_PF_RSVD)
450 STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eSVReserved);
451 }
452#endif /* VBOX_WITH_STATISTICS */
453
454 /*
455 * Call the worker.
456 */
457 bool fLockTaken = false;
458 int rc = PGM_BTH_PFN(Trap0eHandler, pVCpu)(pVCpu, uErr, pRegFrame, pvFault, &fLockTaken);
459 if (fLockTaken)
460 {
461 PGM_LOCK_ASSERT_OWNER(pVM);
462 pgmUnlock(pVM);
463 }
464 LogFlow(("PGMTrap0eHandler: uErr=%RGx pvFault=%RGv rc=%Rrc\n", uErr, pvFault, rc));
465
466 /*
467 * Return code tweaks.
468 */
469 if (rc != VINF_SUCCESS)
470 {
471 if (rc == VINF_PGM_SYNCPAGE_MODIFIED_PDE)
472 rc = VINF_SUCCESS;
473
474# ifdef IN_RING0
475 /* Note: hack alert for difficult to reproduce problem. */
476 if ( rc == VERR_PAGE_NOT_PRESENT /* SMP only ; disassembly might fail. */
477 || rc == VERR_PAGE_TABLE_NOT_PRESENT /* seen with UNI & SMP */
478 || rc == VERR_PAGE_DIRECTORY_PTR_NOT_PRESENT /* seen with SMP */
479 || rc == VERR_PAGE_MAP_LEVEL4_NOT_PRESENT) /* precaution */
480 {
481 Log(("WARNING: Unexpected VERR_PAGE_TABLE_NOT_PRESENT (%d) for page fault at %RGv error code %x (rip=%RGv)\n", rc, pvFault, uErr, pRegFrame->rip));
482 /* Some kind of inconsistency in the SMP case; it's safe to just execute the instruction again; not sure about single VCPU VMs though. */
483 rc = VINF_SUCCESS;
484 }
485# endif
486 }
487
488 STAM_STATS({ if (rc == VINF_EM_RAW_GUEST_TRAP) STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eGuestPF); });
489 STAM_STATS({ if (!pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution))
490 pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eTime2Misc; });
491 STAM_PROFILE_STOP_EX(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0e, pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution), a);
492 return rc;
493}
494#endif /* !IN_RING3 */
495
496
497/**
498 * Prefetch a page
499 *
500 * Typically used to sync commonly used pages before entering raw mode
501 * after a CR3 reload.
502 *
503 * @returns VBox status code suitable for scheduling.
504 * @retval VINF_SUCCESS on success.
505 * @retval VINF_PGM_SYNC_CR3 if we're out of shadow pages or something like that.
506 * @param pVCpu Pointer to the VMCPU.
507 * @param GCPtrPage Page to invalidate.
508 */
509VMMDECL(int) PGMPrefetchPage(PVMCPU pVCpu, RTGCPTR GCPtrPage)
510{
511 STAM_PROFILE_START(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,Prefetch), a);
512 int rc = PGM_BTH_PFN(PrefetchPage, pVCpu)(pVCpu, GCPtrPage);
513 STAM_PROFILE_STOP(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,Prefetch), a);
514 AssertMsg(rc == VINF_SUCCESS || rc == VINF_PGM_SYNC_CR3 || RT_FAILURE(rc), ("rc=%Rrc\n", rc));
515 return rc;
516}
517
518
519/**
520 * Gets the mapping corresponding to the specified address (if any).
521 *
522 * @returns Pointer to the mapping.
523 * @returns NULL if not
524 *
525 * @param pVM Pointer to the VM.
526 * @param GCPtr The guest context pointer.
527 */
528PPGMMAPPING pgmGetMapping(PVM pVM, RTGCPTR GCPtr)
529{
530 PPGMMAPPING pMapping = pVM->pgm.s.CTX_SUFF(pMappings);
531 while (pMapping)
532 {
533 if ((uintptr_t)GCPtr < (uintptr_t)pMapping->GCPtr)
534 break;
535 if ((uintptr_t)GCPtr - (uintptr_t)pMapping->GCPtr < pMapping->cb)
536 return pMapping;
537 pMapping = pMapping->CTX_SUFF(pNext);
538 }
539 return NULL;
540}
541
542
543/**
544 * Verifies a range of pages for read or write access
545 *
546 * Only checks the guest's page tables
547 *
548 * @returns VBox status code.
549 * @param pVCpu Pointer to the VMCPU.
550 * @param Addr Guest virtual address to check
551 * @param cbSize Access size
552 * @param fAccess Access type (r/w, user/supervisor (X86_PTE_*))
553 * @remarks Current not in use.
554 */
555VMMDECL(int) PGMIsValidAccess(PVMCPU pVCpu, RTGCPTR Addr, uint32_t cbSize, uint32_t fAccess)
556{
557 /*
558 * Validate input.
559 */
560 if (fAccess & ~(X86_PTE_US | X86_PTE_RW))
561 {
562 AssertMsgFailed(("PGMIsValidAccess: invalid access type %08x\n", fAccess));
563 return VERR_INVALID_PARAMETER;
564 }
565
566 uint64_t fPage;
567 int rc = PGMGstGetPage(pVCpu, (RTGCPTR)Addr, &fPage, NULL);
568 if (RT_FAILURE(rc))
569 {
570 Log(("PGMIsValidAccess: access violation for %RGv rc=%d\n", Addr, rc));
571 return VINF_EM_RAW_GUEST_TRAP;
572 }
573
574 /*
575 * Check if the access would cause a page fault
576 *
577 * Note that hypervisor page directories are not present in the guest's tables, so this check
578 * is sufficient.
579 */
580 bool fWrite = !!(fAccess & X86_PTE_RW);
581 bool fUser = !!(fAccess & X86_PTE_US);
582 if ( !(fPage & X86_PTE_P)
583 || (fWrite && !(fPage & X86_PTE_RW))
584 || (fUser && !(fPage & X86_PTE_US)) )
585 {
586 Log(("PGMIsValidAccess: access violation for %RGv attr %#llx vs %d:%d\n", Addr, fPage, fWrite, fUser));
587 return VINF_EM_RAW_GUEST_TRAP;
588 }
589 if ( RT_SUCCESS(rc)
590 && PAGE_ADDRESS(Addr) != PAGE_ADDRESS(Addr + cbSize))
591 return PGMIsValidAccess(pVCpu, Addr + PAGE_SIZE, (cbSize > PAGE_SIZE) ? cbSize - PAGE_SIZE : 1, fAccess);
592 return rc;
593}
594
595
596/**
597 * Verifies a range of pages for read or write access
598 *
599 * Supports handling of pages marked for dirty bit tracking and CSAM
600 *
601 * @returns VBox status code.
602 * @param pVCpu Pointer to the VMCPU.
603 * @param Addr Guest virtual address to check
604 * @param cbSize Access size
605 * @param fAccess Access type (r/w, user/supervisor (X86_PTE_*))
606 */
607VMMDECL(int) PGMVerifyAccess(PVMCPU pVCpu, RTGCPTR Addr, uint32_t cbSize, uint32_t fAccess)
608{
609 PVM pVM = pVCpu->CTX_SUFF(pVM);
610
611 AssertMsg(!(fAccess & ~(X86_PTE_US | X86_PTE_RW)), ("PGMVerifyAccess: invalid access type %08x\n", fAccess));
612
613 /*
614 * Get going.
615 */
616 uint64_t fPageGst;
617 int rc = PGMGstGetPage(pVCpu, (RTGCPTR)Addr, &fPageGst, NULL);
618 if (RT_FAILURE(rc))
619 {
620 Log(("PGMVerifyAccess: access violation for %RGv rc=%d\n", Addr, rc));
621 return VINF_EM_RAW_GUEST_TRAP;
622 }
623
624 /*
625 * Check if the access would cause a page fault
626 *
627 * Note that hypervisor page directories are not present in the guest's tables, so this check
628 * is sufficient.
629 */
630 const bool fWrite = !!(fAccess & X86_PTE_RW);
631 const bool fUser = !!(fAccess & X86_PTE_US);
632 if ( !(fPageGst & X86_PTE_P)
633 || (fWrite && !(fPageGst & X86_PTE_RW))
634 || (fUser && !(fPageGst & X86_PTE_US)) )
635 {
636 Log(("PGMVerifyAccess: access violation for %RGv attr %#llx vs %d:%d\n", Addr, fPageGst, fWrite, fUser));
637 return VINF_EM_RAW_GUEST_TRAP;
638 }
639
640 if (!pVM->pgm.s.fNestedPaging)
641 {
642 /*
643 * Next step is to verify if we protected this page for dirty bit tracking or for CSAM scanning
644 */
645 rc = PGMShwGetPage(pVCpu, (RTGCPTR)Addr, NULL, NULL);
646 if ( rc == VERR_PAGE_NOT_PRESENT
647 || rc == VERR_PAGE_TABLE_NOT_PRESENT)
648 {
649 /*
650 * Page is not present in our page tables.
651 * Try to sync it!
652 */
653 Assert(X86_TRAP_PF_RW == X86_PTE_RW && X86_TRAP_PF_US == X86_PTE_US);
654 uint32_t uErr = fAccess & (X86_TRAP_PF_RW | X86_TRAP_PF_US);
655 rc = PGM_BTH_PFN(VerifyAccessSyncPage, pVCpu)(pVCpu, Addr, fPageGst, uErr);
656 if (rc != VINF_SUCCESS)
657 return rc;
658 }
659 else
660 AssertMsg(rc == VINF_SUCCESS, ("PGMShwGetPage %RGv failed with %Rrc\n", Addr, rc));
661 }
662
663#if 0 /* def VBOX_STRICT; triggers too often now */
664 /*
665 * This check is a bit paranoid, but useful.
666 */
667 /* Note! This will assert when writing to monitored pages (a bit annoying actually). */
668 uint64_t fPageShw;
669 rc = PGMShwGetPage(pVCpu, (RTGCPTR)Addr, &fPageShw, NULL);
670 if ( (rc == VERR_PAGE_NOT_PRESENT || RT_FAILURE(rc))
671 || (fWrite && !(fPageShw & X86_PTE_RW))
672 || (fUser && !(fPageShw & X86_PTE_US)) )
673 {
674 AssertMsgFailed(("Unexpected access violation for %RGv! rc=%Rrc write=%d user=%d\n",
675 Addr, rc, fWrite && !(fPageShw & X86_PTE_RW), fUser && !(fPageShw & X86_PTE_US)));
676 return VINF_EM_RAW_GUEST_TRAP;
677 }
678#endif
679
680 if ( RT_SUCCESS(rc)
681 && ( PAGE_ADDRESS(Addr) != PAGE_ADDRESS(Addr + cbSize - 1)
682 || Addr + cbSize < Addr))
683 {
684 /* Don't recursively call PGMVerifyAccess as we might run out of stack. */
685 for (;;)
686 {
687 Addr += PAGE_SIZE;
688 if (cbSize > PAGE_SIZE)
689 cbSize -= PAGE_SIZE;
690 else
691 cbSize = 1;
692 rc = PGMVerifyAccess(pVCpu, Addr, 1, fAccess);
693 if (rc != VINF_SUCCESS)
694 break;
695 if (PAGE_ADDRESS(Addr) == PAGE_ADDRESS(Addr + cbSize - 1))
696 break;
697 }
698 }
699 return rc;
700}
701
702
703/**
704 * Emulation of the invlpg instruction (HC only actually).
705 *
706 * @returns Strict VBox status code, special care required.
707 * @retval VINF_PGM_SYNC_CR3 - handled.
708 * @retval VINF_EM_RAW_EMULATE_INSTR - not handled (RC only).
709 * @retval VERR_REM_FLUSHED_PAGES_OVERFLOW - not handled.
710 *
711 * @param pVCpu Pointer to the VMCPU.
712 * @param GCPtrPage Page to invalidate.
713 *
714 * @remark ASSUMES the page table entry or page directory is valid. Fairly
715 * safe, but there could be edge cases!
716 *
717 * @todo Flush page or page directory only if necessary!
718 * @todo VBOXSTRICTRC
719 */
720VMMDECL(int) PGMInvalidatePage(PVMCPU pVCpu, RTGCPTR GCPtrPage)
721{
722 PVM pVM = pVCpu->CTX_SUFF(pVM);
723 int rc;
724 Log3(("PGMInvalidatePage: GCPtrPage=%RGv\n", GCPtrPage));
725
726#if !defined(IN_RING3) && defined(VBOX_WITH_REM)
727 /*
728 * Notify the recompiler so it can record this instruction.
729 */
730 REMNotifyInvalidatePage(pVM, GCPtrPage);
731#endif /* !IN_RING3 */
732
733
734#ifdef IN_RC
735 /*
736 * Check for conflicts and pending CR3 monitoring updates.
737 */
738 if (pgmMapAreMappingsFloating(pVM))
739 {
740 if ( pgmGetMapping(pVM, GCPtrPage)
741 && PGMGstGetPage(pVCpu, GCPtrPage, NULL, NULL) != VERR_PAGE_TABLE_NOT_PRESENT)
742 {
743 LogFlow(("PGMGCInvalidatePage: Conflict!\n"));
744 VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
745 STAM_COUNTER_INC(&pVM->pgm.s.CTX_SUFF(pStats)->StatRCInvlPgConflict);
746 return VINF_PGM_SYNC_CR3;
747 }
748
749 if (pVCpu->pgm.s.fSyncFlags & PGM_SYNC_MONITOR_CR3)
750 {
751 LogFlow(("PGMGCInvalidatePage: PGM_SYNC_MONITOR_CR3 -> reinterpret instruction in R3\n"));
752 STAM_COUNTER_INC(&pVM->pgm.s.CTX_SUFF(pStats)->StatRCInvlPgSyncMonCR3);
753 return VINF_EM_RAW_EMULATE_INSTR;
754 }
755 }
756#endif /* IN_RC */
757
758 /*
759 * Call paging mode specific worker.
760 */
761 STAM_PROFILE_START(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,InvalidatePage), a);
762 pgmLock(pVM);
763 rc = PGM_BTH_PFN(InvalidatePage, pVCpu)(pVCpu, GCPtrPage);
764 pgmUnlock(pVM);
765 STAM_PROFILE_STOP(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,InvalidatePage), a);
766
767#ifdef IN_RING3
768 /*
769 * Check if we have a pending update of the CR3 monitoring.
770 */
771 if ( RT_SUCCESS(rc)
772 && (pVCpu->pgm.s.fSyncFlags & PGM_SYNC_MONITOR_CR3))
773 {
774 pVCpu->pgm.s.fSyncFlags &= ~PGM_SYNC_MONITOR_CR3;
775 Assert(!pVM->pgm.s.fMappingsFixed); Assert(pgmMapAreMappingsEnabled(pVM));
776 }
777
778# ifdef VBOX_WITH_RAW_MODE
779 /*
780 * Inform CSAM about the flush
781 *
782 * Note: This is to check if monitored pages have been changed; when we implement
783 * callbacks for virtual handlers, this is no longer required.
784 */
785 CSAMR3FlushPage(pVM, GCPtrPage);
786# endif
787#endif /* IN_RING3 */
788
789 /* Ignore all irrelevant error codes. */
790 if ( rc == VERR_PAGE_NOT_PRESENT
791 || rc == VERR_PAGE_TABLE_NOT_PRESENT
792 || rc == VERR_PAGE_DIRECTORY_PTR_NOT_PRESENT
793 || rc == VERR_PAGE_MAP_LEVEL4_NOT_PRESENT)
794 rc = VINF_SUCCESS;
795
796 return rc;
797}
798
799
800/**
801 * Executes an instruction using the interpreter.
802 *
803 * @returns VBox status code (appropriate for trap handling and GC return).
804 * @param pVM Pointer to the VM.
805 * @param pVCpu Pointer to the VMCPU.
806 * @param pRegFrame Register frame.
807 * @param pvFault Fault address.
808 */
809VMMDECL(VBOXSTRICTRC) PGMInterpretInstruction(PVM pVM, PVMCPU pVCpu, PCPUMCTXCORE pRegFrame, RTGCPTR pvFault)
810{
811 NOREF(pVM);
812 VBOXSTRICTRC rc = EMInterpretInstruction(pVCpu, pRegFrame, pvFault);
813 if (rc == VERR_EM_INTERPRETER)
814 rc = VINF_EM_RAW_EMULATE_INSTR;
815 if (rc != VINF_SUCCESS)
816 Log(("PGMInterpretInstruction: returns %Rrc (pvFault=%RGv)\n", VBOXSTRICTRC_VAL(rc), pvFault));
817 return rc;
818}
819
820
821/**
822 * Gets effective page information (from the VMM page directory).
823 *
824 * @returns VBox status.
825 * @param pVCpu Pointer to the VMCPU.
826 * @param GCPtr Guest Context virtual address of the page.
827 * @param pfFlags Where to store the flags. These are X86_PTE_*.
828 * @param pHCPhys Where to store the HC physical address of the page.
829 * This is page aligned.
830 * @remark You should use PGMMapGetPage() for pages in a mapping.
831 */
832VMMDECL(int) PGMShwGetPage(PVMCPU pVCpu, RTGCPTR GCPtr, uint64_t *pfFlags, PRTHCPHYS pHCPhys)
833{
834 pgmLock(pVCpu->CTX_SUFF(pVM));
835 int rc = PGM_SHW_PFN(GetPage, pVCpu)(pVCpu, GCPtr, pfFlags, pHCPhys);
836 pgmUnlock(pVCpu->CTX_SUFF(pVM));
837 return rc;
838}
839
840
841/**
842 * Modify page flags for a range of pages in the shadow context.
843 *
844 * The existing flags are ANDed with the fMask and ORed with the fFlags.
845 *
846 * @returns VBox status code.
847 * @param pVCpu Pointer to the VMCPU.
848 * @param GCPtr Virtual address of the first page in the range.
849 * @param fFlags The OR mask - page flags X86_PTE_*, excluding the page mask of course.
850 * @param fMask The AND mask - page flags X86_PTE_*.
851 * Be very CAREFUL when ~'ing constants which could be 32-bit!
852 * @param fOpFlags A combination of the PGM_MK_PK_XXX flags.
853 * @remark You must use PGMMapModifyPage() for pages in a mapping.
854 */
855DECLINLINE(int) pdmShwModifyPage(PVMCPU pVCpu, RTGCPTR GCPtr, uint64_t fFlags, uint64_t fMask, uint32_t fOpFlags)
856{
857 AssertMsg(!(fFlags & X86_PTE_PAE_PG_MASK), ("fFlags=%#llx\n", fFlags));
858 Assert(!(fOpFlags & ~(PGM_MK_PG_IS_MMIO2 | PGM_MK_PG_IS_WRITE_FAULT)));
859
860 GCPtr &= PAGE_BASE_GC_MASK; /** @todo this ain't necessary, right... */
861
862 PVM pVM = pVCpu->CTX_SUFF(pVM);
863 pgmLock(pVM);
864 int rc = PGM_SHW_PFN(ModifyPage, pVCpu)(pVCpu, GCPtr, PAGE_SIZE, fFlags, fMask, fOpFlags);
865 pgmUnlock(pVM);
866 return rc;
867}
868
869
870/**
871 * Changing the page flags for a single page in the shadow page tables so as to
872 * make it read-only.
873 *
874 * @returns VBox status code.
875 * @param pVCpu Pointer to the VMCPU.
876 * @param GCPtr Virtual address of the first page in the range.
877 * @param fOpFlags A combination of the PGM_MK_PK_XXX flags.
878 */
879VMMDECL(int) PGMShwMakePageReadonly(PVMCPU pVCpu, RTGCPTR GCPtr, uint32_t fOpFlags)
880{
881 return pdmShwModifyPage(pVCpu, GCPtr, 0, ~(uint64_t)X86_PTE_RW, fOpFlags);
882}
883
884
885/**
886 * Changing the page flags for a single page in the shadow page tables so as to
887 * make it writable.
888 *
889 * The call must know with 101% certainty that the guest page tables maps this
890 * as writable too. This function will deal shared, zero and write monitored
891 * pages.
892 *
893 * @returns VBox status code.
894 * @param pVCpu Pointer to the VMCPU.
895 * @param GCPtr Virtual address of the first page in the range.
896 * @param fMmio2 Set if it is an MMIO2 page.
897 * @param fOpFlags A combination of the PGM_MK_PK_XXX flags.
898 */
899VMMDECL(int) PGMShwMakePageWritable(PVMCPU pVCpu, RTGCPTR GCPtr, uint32_t fOpFlags)
900{
901 return pdmShwModifyPage(pVCpu, GCPtr, X86_PTE_RW, ~(uint64_t)0, fOpFlags);
902}
903
904
905/**
906 * Changing the page flags for a single page in the shadow page tables so as to
907 * make it not present.
908 *
909 * @returns VBox status code.
910 * @param pVCpu Pointer to the VMCPU.
911 * @param GCPtr Virtual address of the first page in the range.
912 * @param fOpFlags A combination of the PGM_MK_PG_XXX flags.
913 */
914VMMDECL(int) PGMShwMakePageNotPresent(PVMCPU pVCpu, RTGCPTR GCPtr, uint32_t fOpFlags)
915{
916 return pdmShwModifyPage(pVCpu, GCPtr, 0, 0, fOpFlags);
917}
918
919
920/**
921 * Changing the page flags for a single page in the shadow page tables so as to
922 * make it supervisor and writable.
923 *
924 * This if for dealing with CR0.WP=0 and readonly user pages.
925 *
926 * @returns VBox status code.
927 * @param pVCpu Pointer to the VMCPU.
928 * @param GCPtr Virtual address of the first page in the range.
929 * @param fBigPage Whether or not this is a big page. If it is, we have to
930 * change the shadow PDE as well. If it isn't, the caller
931 * has checked that the shadow PDE doesn't need changing.
932 * We ASSUME 4KB pages backing the big page here!
933 * @param fOpFlags A combination of the PGM_MK_PG_XXX flags.
934 */
935int pgmShwMakePageSupervisorAndWritable(PVMCPU pVCpu, RTGCPTR GCPtr, bool fBigPage, uint32_t fOpFlags)
936{
937 int rc = pdmShwModifyPage(pVCpu, GCPtr, X86_PTE_RW, ~(uint64_t)X86_PTE_US, fOpFlags);
938 if (rc == VINF_SUCCESS && fBigPage)
939 {
940 /* this is a bit ugly... */
941 switch (pVCpu->pgm.s.enmShadowMode)
942 {
943 case PGMMODE_32_BIT:
944 {
945 PX86PDE pPde = pgmShwGet32BitPDEPtr(pVCpu, GCPtr);
946 AssertReturn(pPde, VERR_INTERNAL_ERROR_3);
947 Log(("pgmShwMakePageSupervisorAndWritable: PDE=%#llx", pPde->u));
948 pPde->n.u1Write = 1;
949 Log(("-> PDE=%#llx (32)\n", pPde->u));
950 break;
951 }
952 case PGMMODE_PAE:
953 case PGMMODE_PAE_NX:
954 {
955 PX86PDEPAE pPde = pgmShwGetPaePDEPtr(pVCpu, GCPtr);
956 AssertReturn(pPde, VERR_INTERNAL_ERROR_3);
957 Log(("pgmShwMakePageSupervisorAndWritable: PDE=%#llx", pPde->u));
958 pPde->n.u1Write = 1;
959 Log(("-> PDE=%#llx (PAE)\n", pPde->u));
960 break;
961 }
962 default:
963 AssertFailedReturn(VERR_INTERNAL_ERROR_4);
964 }
965 }
966 return rc;
967}
968
969
970/**
971 * Gets the shadow page directory for the specified address, PAE.
972 *
973 * @returns Pointer to the shadow PD.
974 * @param pVCpu Pointer to the VMCPU.
975 * @param GCPtr The address.
976 * @param uGstPdpe Guest PDPT entry. Valid.
977 * @param ppPD Receives address of page directory
978 */
979int pgmShwSyncPaePDPtr(PVMCPU pVCpu, RTGCPTR GCPtr, X86PGPAEUINT uGstPdpe, PX86PDPAE *ppPD)
980{
981 const unsigned iPdPt = (GCPtr >> X86_PDPT_SHIFT) & X86_PDPT_MASK_PAE;
982 PX86PDPT pPdpt = pgmShwGetPaePDPTPtr(pVCpu);
983 PX86PDPE pPdpe = &pPdpt->a[iPdPt];
984 PVM pVM = pVCpu->CTX_SUFF(pVM);
985 PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
986 PPGMPOOLPAGE pShwPage;
987 int rc;
988
989 PGM_LOCK_ASSERT_OWNER(pVM);
990
991 /* Allocate page directory if not present. */
992 if ( !pPdpe->n.u1Present
993 && !(pPdpe->u & X86_PDPE_PG_MASK))
994 {
995 RTGCPTR64 GCPdPt;
996 PGMPOOLKIND enmKind;
997
998 if (pVM->pgm.s.fNestedPaging || !CPUMIsGuestPagingEnabled(pVCpu))
999 {
1000 /* AMD-V nested paging or real/protected mode without paging. */
1001 GCPdPt = (RTGCPTR64)iPdPt << X86_PDPT_SHIFT;
1002 enmKind = PGMPOOLKIND_PAE_PD_PHYS;
1003 }
1004 else
1005 {
1006 if (CPUMGetGuestCR4(pVCpu) & X86_CR4_PAE)
1007 {
1008 if (!(uGstPdpe & X86_PDPE_P))
1009 {
1010 /* PD not present; guest must reload CR3 to change it.
1011 * No need to monitor anything in this case.
1012 */
1013 Assert(!HMIsEnabled(pVM));
1014
1015 GCPdPt = uGstPdpe & X86_PDPE_PG_MASK;
1016 enmKind = PGMPOOLKIND_PAE_PD_PHYS;
1017 uGstPdpe |= X86_PDPE_P;
1018 }
1019 else
1020 {
1021 GCPdPt = uGstPdpe & X86_PDPE_PG_MASK;
1022 enmKind = PGMPOOLKIND_PAE_PD_FOR_PAE_PD;
1023 }
1024 }
1025 else
1026 {
1027 GCPdPt = CPUMGetGuestCR3(pVCpu);
1028 enmKind = (PGMPOOLKIND)(PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD + iPdPt);
1029 }
1030 }
1031
1032 /* Create a reference back to the PDPT by using the index in its shadow page. */
1033 rc = pgmPoolAlloc(pVM, GCPdPt, enmKind, PGMPOOLACCESS_DONTCARE, PGM_A20_IS_ENABLED(pVCpu),
1034 pVCpu->pgm.s.CTX_SUFF(pShwPageCR3)->idx, iPdPt, false /*fLockPage*/,
1035 &pShwPage);
1036 AssertRCReturn(rc, rc);
1037
1038 /* The PD was cached or created; hook it up now. */
1039 pPdpe->u |= pShwPage->Core.Key | (uGstPdpe & (X86_PDPE_P | X86_PDPE_A));
1040
1041# if defined(IN_RC)
1042 /*
1043 * In 32 bits PAE mode we *must* invalidate the TLB when changing a
1044 * PDPT entry; the CPU fetches them only during cr3 load, so any
1045 * non-present PDPT will continue to cause page faults.
1046 */
1047 ASMReloadCR3();
1048# endif
1049 PGM_DYNMAP_UNUSED_HINT(pVCpu, pPdpe);
1050 }
1051 else
1052 {
1053 pShwPage = pgmPoolGetPage(pPool, pPdpe->u & X86_PDPE_PG_MASK);
1054 AssertReturn(pShwPage, VERR_PGM_POOL_GET_PAGE_FAILED);
1055 Assert((pPdpe->u & X86_PDPE_PG_MASK) == pShwPage->Core.Key);
1056
1057 pgmPoolCacheUsed(pPool, pShwPage);
1058 }
1059 *ppPD = (PX86PDPAE)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPage);
1060 return VINF_SUCCESS;
1061}
1062
1063
1064/**
1065 * Gets the pointer to the shadow page directory entry for an address, PAE.
1066 *
1067 * @returns Pointer to the PDE.
1068 * @param pVCpu The current CPU.
1069 * @param GCPtr The address.
1070 * @param ppShwPde Receives the address of the pgm pool page for the shadow page directory
1071 */
1072DECLINLINE(int) pgmShwGetPaePoolPagePD(PVMCPU pVCpu, RTGCPTR GCPtr, PPGMPOOLPAGE *ppShwPde)
1073{
1074 const unsigned iPdPt = (GCPtr >> X86_PDPT_SHIFT) & X86_PDPT_MASK_PAE;
1075 PX86PDPT pPdpt = pgmShwGetPaePDPTPtr(pVCpu);
1076 PVM pVM = pVCpu->CTX_SUFF(pVM);
1077
1078 PGM_LOCK_ASSERT_OWNER(pVM);
1079
1080 AssertReturn(pPdpt, VERR_PAGE_DIRECTORY_PTR_NOT_PRESENT); /* can't happen */
1081 if (!pPdpt->a[iPdPt].n.u1Present)
1082 {
1083 LogFlow(("pgmShwGetPaePoolPagePD: PD %d not present (%RX64)\n", iPdPt, pPdpt->a[iPdPt].u));
1084 return VERR_PAGE_DIRECTORY_PTR_NOT_PRESENT;
1085 }
1086 AssertMsg(pPdpt->a[iPdPt].u & X86_PDPE_PG_MASK, ("GCPtr=%RGv\n", GCPtr));
1087
1088 /* Fetch the pgm pool shadow descriptor. */
1089 PPGMPOOLPAGE pShwPde = pgmPoolGetPage(pVM->pgm.s.CTX_SUFF(pPool), pPdpt->a[iPdPt].u & X86_PDPE_PG_MASK);
1090 AssertReturn(pShwPde, VERR_PGM_POOL_GET_PAGE_FAILED);
1091
1092 *ppShwPde = pShwPde;
1093 return VINF_SUCCESS;
1094}
1095
1096#ifndef IN_RC
1097
1098/**
1099 * Syncs the SHADOW page directory pointer for the specified address.
1100 *
1101 * Allocates backing pages in case the PDPT or PML4 entry is missing.
1102 *
1103 * The caller is responsible for making sure the guest has a valid PD before
1104 * calling this function.
1105 *
1106 * @returns VBox status.
1107 * @param pVCpu Pointer to the VMCPU.
1108 * @param GCPtr The address.
1109 * @param uGstPml4e Guest PML4 entry (valid).
1110 * @param uGstPdpe Guest PDPT entry (valid).
1111 * @param ppPD Receives address of page directory
1112 */
1113static int pgmShwSyncLongModePDPtr(PVMCPU pVCpu, RTGCPTR64 GCPtr, X86PGPAEUINT uGstPml4e, X86PGPAEUINT uGstPdpe, PX86PDPAE *ppPD)
1114{
1115 PVM pVM = pVCpu->CTX_SUFF(pVM);
1116 PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1117 const unsigned iPml4 = (GCPtr >> X86_PML4_SHIFT) & X86_PML4_MASK;
1118 PX86PML4E pPml4e = pgmShwGetLongModePML4EPtr(pVCpu, iPml4);
1119 bool fNestedPagingOrNoGstPaging = pVM->pgm.s.fNestedPaging || !CPUMIsGuestPagingEnabled(pVCpu);
1120 PPGMPOOLPAGE pShwPage;
1121 int rc;
1122
1123 PGM_LOCK_ASSERT_OWNER(pVM);
1124
1125 /* Allocate page directory pointer table if not present. */
1126 if ( !pPml4e->n.u1Present
1127 && !(pPml4e->u & X86_PML4E_PG_MASK))
1128 {
1129 RTGCPTR64 GCPml4;
1130 PGMPOOLKIND enmKind;
1131
1132 Assert(pVCpu->pgm.s.CTX_SUFF(pShwPageCR3));
1133
1134 if (fNestedPagingOrNoGstPaging)
1135 {
1136 /* AMD-V nested paging or real/protected mode without paging */
1137 GCPml4 = (RTGCPTR64)iPml4 << X86_PML4_SHIFT;
1138 enmKind = PGMPOOLKIND_64BIT_PDPT_FOR_PHYS;
1139 }
1140 else
1141 {
1142 GCPml4 = uGstPml4e & X86_PML4E_PG_MASK;
1143 enmKind = PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT;
1144 }
1145
1146 /* Create a reference back to the PDPT by using the index in its shadow page. */
1147 rc = pgmPoolAlloc(pVM, GCPml4, enmKind, PGMPOOLACCESS_DONTCARE, PGM_A20_IS_ENABLED(pVCpu),
1148 pVCpu->pgm.s.CTX_SUFF(pShwPageCR3)->idx, iPml4, false /*fLockPage*/,
1149 &pShwPage);
1150 AssertRCReturn(rc, rc);
1151 }
1152 else
1153 {
1154 pShwPage = pgmPoolGetPage(pPool, pPml4e->u & X86_PML4E_PG_MASK);
1155 AssertReturn(pShwPage, VERR_PGM_POOL_GET_PAGE_FAILED);
1156
1157 pgmPoolCacheUsed(pPool, pShwPage);
1158 }
1159 /* The PDPT was cached or created; hook it up now. */
1160 pPml4e->u |= pShwPage->Core.Key | (uGstPml4e & pVCpu->pgm.s.fGstAmd64ShadowedPml4eMask);
1161
1162 const unsigned iPdPt = (GCPtr >> X86_PDPT_SHIFT) & X86_PDPT_MASK_AMD64;
1163 PX86PDPT pPdpt = (PX86PDPT)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPage);
1164 PX86PDPE pPdpe = &pPdpt->a[iPdPt];
1165
1166 /* Allocate page directory if not present. */
1167 if ( !pPdpe->n.u1Present
1168 && !(pPdpe->u & X86_PDPE_PG_MASK))
1169 {
1170 RTGCPTR64 GCPdPt;
1171 PGMPOOLKIND enmKind;
1172
1173 if (fNestedPagingOrNoGstPaging)
1174 {
1175 /* AMD-V nested paging or real/protected mode without paging */
1176 GCPdPt = (RTGCPTR64)iPdPt << X86_PDPT_SHIFT;
1177 enmKind = PGMPOOLKIND_64BIT_PD_FOR_PHYS;
1178 }
1179 else
1180 {
1181 GCPdPt = uGstPdpe & X86_PDPE_PG_MASK;
1182 enmKind = PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD;
1183 }
1184
1185 /* Create a reference back to the PDPT by using the index in its shadow page. */
1186 rc = pgmPoolAlloc(pVM, GCPdPt, enmKind, PGMPOOLACCESS_DONTCARE, PGM_A20_IS_ENABLED(pVCpu),
1187 pShwPage->idx, iPdPt, false /*fLockPage*/,
1188 &pShwPage);
1189 AssertRCReturn(rc, rc);
1190 }
1191 else
1192 {
1193 pShwPage = pgmPoolGetPage(pPool, pPdpe->u & X86_PDPE_PG_MASK);
1194 AssertReturn(pShwPage, VERR_PGM_POOL_GET_PAGE_FAILED);
1195
1196 pgmPoolCacheUsed(pPool, pShwPage);
1197 }
1198 /* The PD was cached or created; hook it up now. */
1199 pPdpe->u |= pShwPage->Core.Key | (uGstPdpe & pVCpu->pgm.s.fGstAmd64ShadowedPdpeMask);
1200
1201 *ppPD = (PX86PDPAE)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPage);
1202 return VINF_SUCCESS;
1203}
1204
1205
1206/**
1207 * Gets the SHADOW page directory pointer for the specified address (long mode).
1208 *
1209 * @returns VBox status.
1210 * @param pVCpu Pointer to the VMCPU.
1211 * @param GCPtr The address.
1212 * @param ppPdpt Receives address of pdpt
1213 * @param ppPD Receives address of page directory
1214 */
1215DECLINLINE(int) pgmShwGetLongModePDPtr(PVMCPU pVCpu, RTGCPTR64 GCPtr, PX86PML4E *ppPml4e, PX86PDPT *ppPdpt, PX86PDPAE *ppPD)
1216{
1217 const unsigned iPml4 = (GCPtr >> X86_PML4_SHIFT) & X86_PML4_MASK;
1218 PCX86PML4E pPml4e = pgmShwGetLongModePML4EPtr(pVCpu, iPml4);
1219
1220 PGM_LOCK_ASSERT_OWNER(pVCpu->CTX_SUFF(pVM));
1221
1222 AssertReturn(pPml4e, VERR_PGM_PML4_MAPPING);
1223 if (ppPml4e)
1224 *ppPml4e = (PX86PML4E)pPml4e;
1225
1226 Log4(("pgmShwGetLongModePDPtr %RGv (%RHv) %RX64\n", GCPtr, pPml4e, pPml4e->u));
1227
1228 if (!pPml4e->n.u1Present)
1229 return VERR_PAGE_MAP_LEVEL4_NOT_PRESENT;
1230
1231 PVM pVM = pVCpu->CTX_SUFF(pVM);
1232 PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1233 PPGMPOOLPAGE pShwPage = pgmPoolGetPage(pPool, pPml4e->u & X86_PML4E_PG_MASK);
1234 AssertReturn(pShwPage, VERR_PGM_POOL_GET_PAGE_FAILED);
1235
1236 const unsigned iPdPt = (GCPtr >> X86_PDPT_SHIFT) & X86_PDPT_MASK_AMD64;
1237 PCX86PDPT pPdpt = *ppPdpt = (PX86PDPT)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPage);
1238 if (!pPdpt->a[iPdPt].n.u1Present)
1239 return VERR_PAGE_DIRECTORY_PTR_NOT_PRESENT;
1240
1241 pShwPage = pgmPoolGetPage(pPool, pPdpt->a[iPdPt].u & X86_PDPE_PG_MASK);
1242 AssertReturn(pShwPage, VERR_PGM_POOL_GET_PAGE_FAILED);
1243
1244 *ppPD = (PX86PDPAE)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPage);
1245 Log4(("pgmShwGetLongModePDPtr %RGv -> *ppPD=%p PDE=%p/%RX64\n", GCPtr, *ppPD, &(*ppPD)->a[(GCPtr >> X86_PD_PAE_SHIFT) & X86_PD_PAE_MASK], (*ppPD)->a[(GCPtr >> X86_PD_PAE_SHIFT) & X86_PD_PAE_MASK].u));
1246 return VINF_SUCCESS;
1247}
1248
1249
1250/**
1251 * Syncs the SHADOW EPT page directory pointer for the specified address. Allocates
1252 * backing pages in case the PDPT or PML4 entry is missing.
1253 *
1254 * @returns VBox status.
1255 * @param pVCpu Pointer to the VMCPU.
1256 * @param GCPtr The address.
1257 * @param ppPdpt Receives address of pdpt
1258 * @param ppPD Receives address of page directory
1259 */
1260static int pgmShwGetEPTPDPtr(PVMCPU pVCpu, RTGCPTR64 GCPtr, PEPTPDPT *ppPdpt, PEPTPD *ppPD)
1261{
1262 PVM pVM = pVCpu->CTX_SUFF(pVM);
1263 const unsigned iPml4 = (GCPtr >> EPT_PML4_SHIFT) & EPT_PML4_MASK;
1264 PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
1265 PEPTPML4 pPml4;
1266 PEPTPML4E pPml4e;
1267 PPGMPOOLPAGE pShwPage;
1268 int rc;
1269
1270 Assert(pVM->pgm.s.fNestedPaging);
1271 PGM_LOCK_ASSERT_OWNER(pVM);
1272
1273 pPml4 = (PEPTPML4)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pVCpu->pgm.s.CTX_SUFF(pShwPageCR3));
1274 Assert(pPml4);
1275
1276 /* Allocate page directory pointer table if not present. */
1277 pPml4e = &pPml4->a[iPml4];
1278 if ( !pPml4e->n.u1Present
1279 && !(pPml4e->u & EPT_PML4E_PG_MASK))
1280 {
1281 Assert(!(pPml4e->u & EPT_PML4E_PG_MASK));
1282 RTGCPTR64 GCPml4 = (RTGCPTR64)iPml4 << EPT_PML4_SHIFT;
1283
1284 rc = pgmPoolAlloc(pVM, GCPml4, PGMPOOLKIND_EPT_PDPT_FOR_PHYS, PGMPOOLACCESS_DONTCARE, PGM_A20_IS_ENABLED(pVCpu),
1285 pVCpu->pgm.s.CTX_SUFF(pShwPageCR3)->idx, iPml4, false /*fLockPage*/,
1286 &pShwPage);
1287 AssertRCReturn(rc, rc);
1288 }
1289 else
1290 {
1291 pShwPage = pgmPoolGetPage(pPool, pPml4e->u & EPT_PML4E_PG_MASK);
1292 AssertReturn(pShwPage, VERR_PGM_POOL_GET_PAGE_FAILED);
1293
1294 pgmPoolCacheUsed(pPool, pShwPage);
1295 }
1296 /* The PDPT was cached or created; hook it up now and fill with the default value. */
1297 pPml4e->u = pShwPage->Core.Key;
1298 pPml4e->n.u1Present = 1;
1299 pPml4e->n.u1Write = 1;
1300 pPml4e->n.u1Execute = 1;
1301
1302 const unsigned iPdPt = (GCPtr >> EPT_PDPT_SHIFT) & EPT_PDPT_MASK;
1303 PEPTPDPT pPdpt = (PEPTPDPT)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPage);
1304 PEPTPDPTE pPdpe = &pPdpt->a[iPdPt];
1305
1306 if (ppPdpt)
1307 *ppPdpt = pPdpt;
1308
1309 /* Allocate page directory if not present. */
1310 if ( !pPdpe->n.u1Present
1311 && !(pPdpe->u & EPT_PDPTE_PG_MASK))
1312 {
1313 RTGCPTR64 GCPdPt = (RTGCPTR64)iPdPt << EPT_PDPT_SHIFT;
1314 rc = pgmPoolAlloc(pVM, GCPdPt, PGMPOOLKIND_EPT_PD_FOR_PHYS, PGMPOOLACCESS_DONTCARE, PGM_A20_IS_ENABLED(pVCpu),
1315 pShwPage->idx, iPdPt, false /*fLockPage*/,
1316 &pShwPage);
1317 AssertRCReturn(rc, rc);
1318 }
1319 else
1320 {
1321 pShwPage = pgmPoolGetPage(pPool, pPdpe->u & EPT_PDPTE_PG_MASK);
1322 AssertReturn(pShwPage, VERR_PGM_POOL_GET_PAGE_FAILED);
1323
1324 pgmPoolCacheUsed(pPool, pShwPage);
1325 }
1326 /* The PD was cached or created; hook it up now and fill with the default value. */
1327 pPdpe->u = pShwPage->Core.Key;
1328 pPdpe->n.u1Present = 1;
1329 pPdpe->n.u1Write = 1;
1330 pPdpe->n.u1Execute = 1;
1331
1332 *ppPD = (PEPTPD)PGMPOOL_PAGE_2_PTR_V2(pVM, pVCpu, pShwPage);
1333 return VINF_SUCCESS;
1334}
1335
1336#endif /* IN_RC */
1337
1338#ifdef IN_RING0
1339/**
1340 * Synchronizes a range of nested page table entries.
1341 *
1342 * The caller must own the PGM lock.
1343 *
1344 * @param pVCpu The current CPU.
1345 * @param GCPhys Where to start.
1346 * @param cPages How many pages which entries should be synced.
1347 * @param enmShwPagingMode The shadow paging mode (PGMMODE_EPT for VT-x,
1348 * host paging mode for AMD-V).
1349 */
1350int pgmShwSyncNestedPageLocked(PVMCPU pVCpu, RTGCPHYS GCPhysFault, uint32_t cPages, PGMMODE enmShwPagingMode)
1351{
1352 PGM_LOCK_ASSERT_OWNER(pVCpu->CTX_SUFF(pVM));
1353
1354 int rc;
1355 switch (enmShwPagingMode)
1356 {
1357 case PGMMODE_32_BIT:
1358 {
1359 X86PDE PdeDummy = { X86_PDE_P | X86_PDE_US | X86_PDE_RW | X86_PDE_A };
1360 rc = PGM_BTH_NAME_32BIT_PROT(SyncPage)(pVCpu, PdeDummy, GCPhysFault, cPages, ~0U /*uErr*/);
1361 break;
1362 }
1363
1364 case PGMMODE_PAE:
1365 case PGMMODE_PAE_NX:
1366 {
1367 X86PDEPAE PdeDummy = { X86_PDE_P | X86_PDE_US | X86_PDE_RW | X86_PDE_A };
1368 rc = PGM_BTH_NAME_PAE_PROT(SyncPage)(pVCpu, PdeDummy, GCPhysFault, cPages, ~0U /*uErr*/);
1369 break;
1370 }
1371
1372 case PGMMODE_AMD64:
1373 case PGMMODE_AMD64_NX:
1374 {
1375 X86PDEPAE PdeDummy = { X86_PDE_P | X86_PDE_US | X86_PDE_RW | X86_PDE_A };
1376 rc = PGM_BTH_NAME_AMD64_PROT(SyncPage)(pVCpu, PdeDummy, GCPhysFault, cPages, ~0U /*uErr*/);
1377 break;
1378 }
1379
1380 case PGMMODE_EPT:
1381 {
1382 X86PDEPAE PdeDummy = { X86_PDE_P | X86_PDE_US | X86_PDE_RW | X86_PDE_A };
1383 rc = PGM_BTH_NAME_EPT_PROT(SyncPage)(pVCpu, PdeDummy, GCPhysFault, cPages, ~0U /*uErr*/);
1384 break;
1385 }
1386
1387 default:
1388 AssertMsgFailedReturn(("%d\n", enmShwPagingMode), VERR_IPE_NOT_REACHED_DEFAULT_CASE);
1389 }
1390 return rc;
1391}
1392#endif /* IN_RING0 */
1393
1394
1395/**
1396 * Gets effective Guest OS page information.
1397 *
1398 * When GCPtr is in a big page, the function will return as if it was a normal
1399 * 4KB page. If the need for distinguishing between big and normal page becomes
1400 * necessary at a later point, a PGMGstGetPage() will be created for that
1401 * purpose.
1402 *
1403 * @returns VBox status.
1404 * @param pVCpu The current CPU.
1405 * @param GCPtr Guest Context virtual address of the page.
1406 * @param pfFlags Where to store the flags. These are X86_PTE_*, even for big pages.
1407 * @param pGCPhys Where to store the GC physical address of the page.
1408 * This is page aligned. The fact that the
1409 */
1410VMMDECL(int) PGMGstGetPage(PVMCPU pVCpu, RTGCPTR GCPtr, uint64_t *pfFlags, PRTGCPHYS pGCPhys)
1411{
1412 VMCPU_ASSERT_EMT(pVCpu);
1413 return PGM_GST_PFN(GetPage, pVCpu)(pVCpu, GCPtr, pfFlags, pGCPhys);
1414}
1415
1416
1417/**
1418 * Performs a guest page table walk.
1419 *
1420 * The guest should be in paged protect mode or long mode when making a call to
1421 * this function.
1422 *
1423 * @returns VBox status code.
1424 * @retval VINF_SUCCESS on success.
1425 * @retval VERR_PAGE_TABLE_NOT_PRESENT on failure. Check pWalk for details.
1426 * @retval VERR_PGM_NOT_USED_IN_MODE if not paging isn't enabled. @a pWalk is
1427 * not valid, except enmType is PGMPTWALKGSTTYPE_INVALID.
1428 *
1429 * @param pVCpu The current CPU.
1430 * @param GCPtr The guest virtual address to walk by.
1431 * @param pWalk Where to return the walk result. This is valid on some
1432 * error codes as well.
1433 */
1434int pgmGstPtWalk(PVMCPU pVCpu, RTGCPTR GCPtr, PPGMPTWALKGST pWalk)
1435{
1436 VMCPU_ASSERT_EMT(pVCpu);
1437 switch (pVCpu->pgm.s.enmGuestMode)
1438 {
1439 case PGMMODE_32_BIT:
1440 pWalk->enmType = PGMPTWALKGSTTYPE_32BIT;
1441 return PGM_GST_NAME_32BIT(Walk)(pVCpu, GCPtr, &pWalk->u.Legacy);
1442
1443 case PGMMODE_PAE:
1444 case PGMMODE_PAE_NX:
1445 pWalk->enmType = PGMPTWALKGSTTYPE_PAE;
1446 return PGM_GST_NAME_PAE(Walk)(pVCpu, GCPtr, &pWalk->u.Pae);
1447
1448#if !defined(IN_RC)
1449 case PGMMODE_AMD64:
1450 case PGMMODE_AMD64_NX:
1451 pWalk->enmType = PGMPTWALKGSTTYPE_AMD64;
1452 return PGM_GST_NAME_AMD64(Walk)(pVCpu, GCPtr, &pWalk->u.Amd64);
1453#endif
1454
1455 case PGMMODE_REAL:
1456 case PGMMODE_PROTECTED:
1457 pWalk->enmType = PGMPTWALKGSTTYPE_INVALID;
1458 return VERR_PGM_NOT_USED_IN_MODE;
1459
1460#if defined(IN_RC)
1461 case PGMMODE_AMD64:
1462 case PGMMODE_AMD64_NX:
1463#endif
1464 case PGMMODE_NESTED:
1465 case PGMMODE_EPT:
1466 default:
1467 AssertFailed();
1468 pWalk->enmType = PGMPTWALKGSTTYPE_INVALID;
1469 return VERR_PGM_NOT_USED_IN_MODE;
1470 }
1471}
1472
1473
1474/**
1475 * Checks if the page is present.
1476 *
1477 * @returns true if the page is present.
1478 * @returns false if the page is not present.
1479 * @param pVCpu Pointer to the VMCPU.
1480 * @param GCPtr Address within the page.
1481 */
1482VMMDECL(bool) PGMGstIsPagePresent(PVMCPU pVCpu, RTGCPTR GCPtr)
1483{
1484 VMCPU_ASSERT_EMT(pVCpu);
1485 int rc = PGMGstGetPage(pVCpu, GCPtr, NULL, NULL);
1486 return RT_SUCCESS(rc);
1487}
1488
1489
1490/**
1491 * Sets (replaces) the page flags for a range of pages in the guest's tables.
1492 *
1493 * @returns VBox status.
1494 * @param pVCpu Pointer to the VMCPU.
1495 * @param GCPtr The address of the first page.
1496 * @param cb The size of the range in bytes.
1497 * @param fFlags Page flags X86_PTE_*, excluding the page mask of course.
1498 */
1499VMMDECL(int) PGMGstSetPage(PVMCPU pVCpu, RTGCPTR GCPtr, size_t cb, uint64_t fFlags)
1500{
1501 VMCPU_ASSERT_EMT(pVCpu);
1502 return PGMGstModifyPage(pVCpu, GCPtr, cb, fFlags, 0);
1503}
1504
1505
1506/**
1507 * Modify page flags for a range of pages in the guest's tables
1508 *
1509 * The existing flags are ANDed with the fMask and ORed with the fFlags.
1510 *
1511 * @returns VBox status code.
1512 * @param pVCpu Pointer to the VMCPU.
1513 * @param GCPtr Virtual address of the first page in the range.
1514 * @param cb Size (in bytes) of the range to apply the modification to.
1515 * @param fFlags The OR mask - page flags X86_PTE_*, excluding the page mask of course.
1516 * @param fMask The AND mask - page flags X86_PTE_*, excluding the page mask of course.
1517 * Be very CAREFUL when ~'ing constants which could be 32-bit!
1518 */
1519VMMDECL(int) PGMGstModifyPage(PVMCPU pVCpu, RTGCPTR GCPtr, size_t cb, uint64_t fFlags, uint64_t fMask)
1520{
1521 STAM_PROFILE_START(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,GstModifyPage), a);
1522 VMCPU_ASSERT_EMT(pVCpu);
1523
1524 /*
1525 * Validate input.
1526 */
1527 AssertMsg(!(fFlags & X86_PTE_PAE_PG_MASK), ("fFlags=%#llx\n", fFlags));
1528 Assert(cb);
1529
1530 LogFlow(("PGMGstModifyPage %RGv %d bytes fFlags=%08llx fMask=%08llx\n", GCPtr, cb, fFlags, fMask));
1531
1532 /*
1533 * Adjust input.
1534 */
1535 cb += GCPtr & PAGE_OFFSET_MASK;
1536 cb = RT_ALIGN_Z(cb, PAGE_SIZE);
1537 GCPtr = (GCPtr & PAGE_BASE_GC_MASK);
1538
1539 /*
1540 * Call worker.
1541 */
1542 int rc = PGM_GST_PFN(ModifyPage, pVCpu)(pVCpu, GCPtr, cb, fFlags, fMask);
1543
1544 STAM_PROFILE_STOP(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,GstModifyPage), a);
1545 return rc;
1546}
1547
1548
1549#ifndef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
1550
1551/**
1552 * Performs the lazy mapping of the 32-bit guest PD.
1553 *
1554 * @returns VBox status code.
1555 * @param pVCpu The current CPU.
1556 * @param ppPd Where to return the pointer to the mapping. This is
1557 * always set.
1558 */
1559int pgmGstLazyMap32BitPD(PVMCPU pVCpu, PX86PD *ppPd)
1560{
1561 PVM pVM = pVCpu->CTX_SUFF(pVM);
1562 pgmLock(pVM);
1563
1564 Assert(!pVCpu->pgm.s.CTX_SUFF(pGst32BitPd));
1565
1566 RTGCPHYS GCPhysCR3 = pVCpu->pgm.s.GCPhysCR3 & X86_CR3_PAGE_MASK;
1567 PPGMPAGE pPage;
1568 int rc = pgmPhysGetPageEx(pVM, GCPhysCR3, &pPage);
1569 if (RT_SUCCESS(rc))
1570 {
1571 RTHCPTR HCPtrGuestCR3;
1572 rc = pgmPhysGCPhys2CCPtrInternalDepr(pVM, pPage, GCPhysCR3, (void **)&HCPtrGuestCR3);
1573 if (RT_SUCCESS(rc))
1574 {
1575 pVCpu->pgm.s.pGst32BitPdR3 = (R3PTRTYPE(PX86PD))HCPtrGuestCR3;
1576# ifndef VBOX_WITH_2X_4GB_ADDR_SPACE
1577 pVCpu->pgm.s.pGst32BitPdR0 = (R0PTRTYPE(PX86PD))HCPtrGuestCR3;
1578# endif
1579 *ppPd = (PX86PD)HCPtrGuestCR3;
1580
1581 pgmUnlock(pVM);
1582 return VINF_SUCCESS;
1583 }
1584
1585 AssertRC(rc);
1586 }
1587 pgmUnlock(pVM);
1588
1589 *ppPd = NULL;
1590 return rc;
1591}
1592
1593
1594/**
1595 * Performs the lazy mapping of the PAE guest PDPT.
1596 *
1597 * @returns VBox status code.
1598 * @param pVCpu The current CPU.
1599 * @param ppPdpt Where to return the pointer to the mapping. This is
1600 * always set.
1601 */
1602int pgmGstLazyMapPaePDPT(PVMCPU pVCpu, PX86PDPT *ppPdpt)
1603{
1604 Assert(!pVCpu->pgm.s.CTX_SUFF(pGstPaePdpt));
1605 PVM pVM = pVCpu->CTX_SUFF(pVM);
1606 pgmLock(pVM);
1607
1608 RTGCPHYS GCPhysCR3 = pVCpu->pgm.s.GCPhysCR3 & X86_CR3_PAE_PAGE_MASK;
1609 PPGMPAGE pPage;
1610 int rc = pgmPhysGetPageEx(pVM, GCPhysCR3, &pPage);
1611 if (RT_SUCCESS(rc))
1612 {
1613 RTHCPTR HCPtrGuestCR3;
1614 rc = pgmPhysGCPhys2CCPtrInternalDepr(pVM, pPage, GCPhysCR3, (void **)&HCPtrGuestCR3);
1615 if (RT_SUCCESS(rc))
1616 {
1617 pVCpu->pgm.s.pGstPaePdptR3 = (R3PTRTYPE(PX86PDPT))HCPtrGuestCR3;
1618# ifndef VBOX_WITH_2X_4GB_ADDR_SPACE
1619 pVCpu->pgm.s.pGstPaePdptR0 = (R0PTRTYPE(PX86PDPT))HCPtrGuestCR3;
1620# endif
1621 *ppPdpt = (PX86PDPT)HCPtrGuestCR3;
1622
1623 pgmUnlock(pVM);
1624 return VINF_SUCCESS;
1625 }
1626
1627 AssertRC(rc);
1628 }
1629
1630 pgmUnlock(pVM);
1631 *ppPdpt = NULL;
1632 return rc;
1633}
1634
1635
1636/**
1637 * Performs the lazy mapping / updating of a PAE guest PD.
1638 *
1639 * @returns Pointer to the mapping.
1640 * @returns VBox status code.
1641 * @param pVCpu The current CPU.
1642 * @param iPdpt Which PD entry to map (0..3).
1643 * @param ppPd Where to return the pointer to the mapping. This is
1644 * always set.
1645 */
1646int pgmGstLazyMapPaePD(PVMCPU pVCpu, uint32_t iPdpt, PX86PDPAE *ppPd)
1647{
1648 PVM pVM = pVCpu->CTX_SUFF(pVM);
1649 pgmLock(pVM);
1650
1651 PX86PDPT pGuestPDPT = pVCpu->pgm.s.CTX_SUFF(pGstPaePdpt);
1652 Assert(pGuestPDPT);
1653 Assert(pGuestPDPT->a[iPdpt].n.u1Present);
1654 RTGCPHYS GCPhys = pGuestPDPT->a[iPdpt].u & X86_PDPE_PG_MASK;
1655 bool const fChanged = pVCpu->pgm.s.aGCPhysGstPaePDs[iPdpt] != GCPhys;
1656
1657 PPGMPAGE pPage;
1658 int rc = pgmPhysGetPageEx(pVM, GCPhys, &pPage);
1659 if (RT_SUCCESS(rc))
1660 {
1661 RTRCPTR RCPtr = NIL_RTRCPTR;
1662 RTHCPTR HCPtr = NIL_RTHCPTR;
1663#if !defined(IN_RC) && !defined(VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0)
1664 rc = pgmPhysGCPhys2CCPtrInternalDepr(pVM, pPage, GCPhys, &HCPtr);
1665 AssertRC(rc);
1666#endif
1667 if (RT_SUCCESS(rc) && fChanged)
1668 {
1669 RCPtr = (RTRCPTR)(RTRCUINTPTR)(pVM->pgm.s.GCPtrCR3Mapping + (1 + iPdpt) * PAGE_SIZE);
1670 rc = PGMMap(pVM, (RTRCUINTPTR)RCPtr, PGM_PAGE_GET_HCPHYS(pPage), PAGE_SIZE, 0);
1671 }
1672 if (RT_SUCCESS(rc))
1673 {
1674 pVCpu->pgm.s.apGstPaePDsR3[iPdpt] = (R3PTRTYPE(PX86PDPAE))HCPtr;
1675# ifndef VBOX_WITH_2X_4GB_ADDR_SPACE
1676 pVCpu->pgm.s.apGstPaePDsR0[iPdpt] = (R0PTRTYPE(PX86PDPAE))HCPtr;
1677# endif
1678 if (fChanged)
1679 {
1680 pVCpu->pgm.s.aGCPhysGstPaePDs[iPdpt] = GCPhys;
1681 pVCpu->pgm.s.apGstPaePDsRC[iPdpt] = (RCPTRTYPE(PX86PDPAE))RCPtr;
1682 }
1683
1684 *ppPd = pVCpu->pgm.s.CTX_SUFF(apGstPaePDs)[iPdpt];
1685 pgmUnlock(pVM);
1686 return VINF_SUCCESS;
1687 }
1688 }
1689
1690 /* Invalid page or some failure, invalidate the entry. */
1691 pVCpu->pgm.s.aGCPhysGstPaePDs[iPdpt] = NIL_RTGCPHYS;
1692 pVCpu->pgm.s.apGstPaePDsR3[iPdpt] = 0;
1693# ifndef VBOX_WITH_2X_4GB_ADDR_SPACE
1694 pVCpu->pgm.s.apGstPaePDsR0[iPdpt] = 0;
1695# endif
1696 pVCpu->pgm.s.apGstPaePDsRC[iPdpt] = 0;
1697
1698 pgmUnlock(pVM);
1699 return rc;
1700}
1701
1702#endif /* !VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0 */
1703#if !defined(IN_RC) && !defined(VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0)
1704/**
1705 * Performs the lazy mapping of the 32-bit guest PD.
1706 *
1707 * @returns VBox status code.
1708 * @param pVCpu The current CPU.
1709 * @param ppPml4 Where to return the pointer to the mapping. This will
1710 * always be set.
1711 */
1712int pgmGstLazyMapPml4(PVMCPU pVCpu, PX86PML4 *ppPml4)
1713{
1714 Assert(!pVCpu->pgm.s.CTX_SUFF(pGstAmd64Pml4));
1715 PVM pVM = pVCpu->CTX_SUFF(pVM);
1716 pgmLock(pVM);
1717
1718 RTGCPHYS GCPhysCR3 = pVCpu->pgm.s.GCPhysCR3 & X86_CR3_AMD64_PAGE_MASK;
1719 PPGMPAGE pPage;
1720 int rc = pgmPhysGetPageEx(pVM, GCPhysCR3, &pPage);
1721 if (RT_SUCCESS(rc))
1722 {
1723 RTHCPTR HCPtrGuestCR3;
1724 rc = pgmPhysGCPhys2CCPtrInternalDepr(pVM, pPage, GCPhysCR3, (void **)&HCPtrGuestCR3);
1725 if (RT_SUCCESS(rc))
1726 {
1727 pVCpu->pgm.s.pGstAmd64Pml4R3 = (R3PTRTYPE(PX86PML4))HCPtrGuestCR3;
1728# ifndef VBOX_WITH_2X_4GB_ADDR_SPACE
1729 pVCpu->pgm.s.pGstAmd64Pml4R0 = (R0PTRTYPE(PX86PML4))HCPtrGuestCR3;
1730# endif
1731 *ppPml4 = (PX86PML4)HCPtrGuestCR3;
1732
1733 pgmUnlock(pVM);
1734 return VINF_SUCCESS;
1735 }
1736 }
1737
1738 pgmUnlock(pVM);
1739 *ppPml4 = NULL;
1740 return rc;
1741}
1742#endif
1743
1744
1745/**
1746 * Gets the PAE PDPEs values cached by the CPU.
1747 *
1748 * @returns VBox status code.
1749 * @param pVCpu Pointer to the VMCPU.
1750 * @param paPdpes Where to return the four PDPEs. The array
1751 * pointed to must have 4 entries.
1752 */
1753VMM_INT_DECL(int) PGMGstGetPaePdpes(PVMCPU pVCpu, PX86PDPE paPdpes)
1754{
1755 Assert(pVCpu->pgm.s.enmShadowMode == PGMMODE_EPT);
1756
1757 paPdpes[0] = pVCpu->pgm.s.aGstPaePdpeRegs[0];
1758 paPdpes[1] = pVCpu->pgm.s.aGstPaePdpeRegs[1];
1759 paPdpes[2] = pVCpu->pgm.s.aGstPaePdpeRegs[2];
1760 paPdpes[3] = pVCpu->pgm.s.aGstPaePdpeRegs[3];
1761 return VINF_SUCCESS;
1762}
1763
1764
1765/**
1766 * Sets the PAE PDPEs values cached by the CPU.
1767 *
1768 * @remarks This must be called *AFTER* PGMUpdateCR3.
1769 *
1770 * @param pVCpu Pointer to the VMCPU.
1771 * @param paPdpes The four PDPE values. The array pointed to must
1772 * have exactly 4 entries.
1773 *
1774 * @remarks No-long-jump zone!!!
1775 */
1776VMM_INT_DECL(void) PGMGstUpdatePaePdpes(PVMCPU pVCpu, PCX86PDPE paPdpes)
1777{
1778 Assert(pVCpu->pgm.s.enmShadowMode == PGMMODE_EPT);
1779
1780 for (unsigned i = 0; i < RT_ELEMENTS(pVCpu->pgm.s.aGstPaePdpeRegs); i++)
1781 {
1782 if (pVCpu->pgm.s.aGstPaePdpeRegs[i].u != paPdpes[i].u)
1783 {
1784 pVCpu->pgm.s.aGstPaePdpeRegs[i] = paPdpes[i];
1785
1786 /* Force lazy remapping if it changed in any way. */
1787 pVCpu->pgm.s.apGstPaePDsR3[i] = 0;
1788# ifndef VBOX_WITH_2X_4GB_ADDR_SPACE
1789 pVCpu->pgm.s.apGstPaePDsR0[i] = 0;
1790# endif
1791 pVCpu->pgm.s.apGstPaePDsRC[i] = 0;
1792 pVCpu->pgm.s.aGCPhysGstPaePDs[i] = NIL_RTGCPHYS;
1793 }
1794 }
1795
1796 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES);
1797}
1798
1799
1800/**
1801 * Gets the current CR3 register value for the shadow memory context.
1802 * @returns CR3 value.
1803 * @param pVCpu Pointer to the VMCPU.
1804 */
1805VMMDECL(RTHCPHYS) PGMGetHyperCR3(PVMCPU pVCpu)
1806{
1807 PPGMPOOLPAGE pPoolPage = pVCpu->pgm.s.CTX_SUFF(pShwPageCR3);
1808 AssertPtrReturn(pPoolPage, 0);
1809 return pPoolPage->Core.Key;
1810}
1811
1812
1813/**
1814 * Gets the current CR3 register value for the nested memory context.
1815 * @returns CR3 value.
1816 * @param pVCpu Pointer to the VMCPU.
1817 */
1818VMMDECL(RTHCPHYS) PGMGetNestedCR3(PVMCPU pVCpu, PGMMODE enmShadowMode)
1819{
1820 NOREF(enmShadowMode);
1821 Assert(pVCpu->pgm.s.CTX_SUFF(pShwPageCR3));
1822 return pVCpu->pgm.s.CTX_SUFF(pShwPageCR3)->Core.Key;
1823}
1824
1825
1826/**
1827 * Gets the current CR3 register value for the HC intermediate memory context.
1828 * @returns CR3 value.
1829 * @param pVM Pointer to the VM.
1830 */
1831VMMDECL(RTHCPHYS) PGMGetInterHCCR3(PVM pVM)
1832{
1833 switch (pVM->pgm.s.enmHostMode)
1834 {
1835 case SUPPAGINGMODE_32_BIT:
1836 case SUPPAGINGMODE_32_BIT_GLOBAL:
1837 return pVM->pgm.s.HCPhysInterPD;
1838
1839 case SUPPAGINGMODE_PAE:
1840 case SUPPAGINGMODE_PAE_GLOBAL:
1841 case SUPPAGINGMODE_PAE_NX:
1842 case SUPPAGINGMODE_PAE_GLOBAL_NX:
1843 return pVM->pgm.s.HCPhysInterPaePDPT;
1844
1845 case SUPPAGINGMODE_AMD64:
1846 case SUPPAGINGMODE_AMD64_GLOBAL:
1847 case SUPPAGINGMODE_AMD64_NX:
1848 case SUPPAGINGMODE_AMD64_GLOBAL_NX:
1849 return pVM->pgm.s.HCPhysInterPaePDPT;
1850
1851 default:
1852 AssertMsgFailed(("enmHostMode=%d\n", pVM->pgm.s.enmHostMode));
1853 return NIL_RTHCPHYS;
1854 }
1855}
1856
1857
1858/**
1859 * Gets the current CR3 register value for the RC intermediate memory context.
1860 * @returns CR3 value.
1861 * @param pVM Pointer to the VM.
1862 * @param pVCpu Pointer to the VMCPU.
1863 */
1864VMMDECL(RTHCPHYS) PGMGetInterRCCR3(PVM pVM, PVMCPU pVCpu)
1865{
1866 switch (pVCpu->pgm.s.enmShadowMode)
1867 {
1868 case PGMMODE_32_BIT:
1869 return pVM->pgm.s.HCPhysInterPD;
1870
1871 case PGMMODE_PAE:
1872 case PGMMODE_PAE_NX:
1873 return pVM->pgm.s.HCPhysInterPaePDPT;
1874
1875 case PGMMODE_AMD64:
1876 case PGMMODE_AMD64_NX:
1877 return pVM->pgm.s.HCPhysInterPaePML4;
1878
1879 case PGMMODE_EPT:
1880 case PGMMODE_NESTED:
1881 return 0; /* not relevant */
1882
1883 default:
1884 AssertMsgFailed(("enmShadowMode=%d\n", pVCpu->pgm.s.enmShadowMode));
1885 return NIL_RTHCPHYS;
1886 }
1887}
1888
1889
1890/**
1891 * Gets the CR3 register value for the 32-Bit intermediate memory context.
1892 * @returns CR3 value.
1893 * @param pVM Pointer to the VM.
1894 */
1895VMMDECL(RTHCPHYS) PGMGetInter32BitCR3(PVM pVM)
1896{
1897 return pVM->pgm.s.HCPhysInterPD;
1898}
1899
1900
1901/**
1902 * Gets the CR3 register value for the PAE intermediate memory context.
1903 * @returns CR3 value.
1904 * @param pVM Pointer to the VM.
1905 */
1906VMMDECL(RTHCPHYS) PGMGetInterPaeCR3(PVM pVM)
1907{
1908 return pVM->pgm.s.HCPhysInterPaePDPT;
1909}
1910
1911
1912/**
1913 * Gets the CR3 register value for the AMD64 intermediate memory context.
1914 * @returns CR3 value.
1915 * @param pVM Pointer to the VM.
1916 */
1917VMMDECL(RTHCPHYS) PGMGetInterAmd64CR3(PVM pVM)
1918{
1919 return pVM->pgm.s.HCPhysInterPaePML4;
1920}
1921
1922
1923/**
1924 * Performs and schedules necessary updates following a CR3 load or reload.
1925 *
1926 * This will normally involve mapping the guest PD or nPDPT
1927 *
1928 * @returns VBox status code.
1929 * @retval VINF_PGM_SYNC_CR3 if monitoring requires a CR3 sync. This can
1930 * safely be ignored and overridden since the FF will be set too then.
1931 * @param pVCpu Pointer to the VMCPU.
1932 * @param cr3 The new cr3.
1933 * @param fGlobal Indicates whether this is a global flush or not.
1934 */
1935VMMDECL(int) PGMFlushTLB(PVMCPU pVCpu, uint64_t cr3, bool fGlobal)
1936{
1937 STAM_PROFILE_START(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,FlushTLB), a);
1938 PVM pVM = pVCpu->CTX_SUFF(pVM);
1939
1940 VMCPU_ASSERT_EMT(pVCpu);
1941
1942 /*
1943 * Always flag the necessary updates; necessary for hardware acceleration
1944 */
1945 /** @todo optimize this, it shouldn't always be necessary. */
1946 VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL);
1947 if (fGlobal)
1948 VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
1949 LogFlow(("PGMFlushTLB: cr3=%RX64 OldCr3=%RX64 fGlobal=%d\n", cr3, pVCpu->pgm.s.GCPhysCR3, fGlobal));
1950
1951 /*
1952 * Remap the CR3 content and adjust the monitoring if CR3 was actually changed.
1953 */
1954 int rc = VINF_SUCCESS;
1955 RTGCPHYS GCPhysCR3;
1956 switch (pVCpu->pgm.s.enmGuestMode)
1957 {
1958 case PGMMODE_PAE:
1959 case PGMMODE_PAE_NX:
1960 GCPhysCR3 = (RTGCPHYS)(cr3 & X86_CR3_PAE_PAGE_MASK);
1961 break;
1962 case PGMMODE_AMD64:
1963 case PGMMODE_AMD64_NX:
1964 GCPhysCR3 = (RTGCPHYS)(cr3 & X86_CR3_AMD64_PAGE_MASK);
1965 break;
1966 default:
1967 GCPhysCR3 = (RTGCPHYS)(cr3 & X86_CR3_PAGE_MASK);
1968 break;
1969 }
1970 PGM_A20_APPLY_TO_VAR(pVCpu, GCPhysCR3);
1971
1972 if (pVCpu->pgm.s.GCPhysCR3 != GCPhysCR3)
1973 {
1974 RTGCPHYS GCPhysOldCR3 = pVCpu->pgm.s.GCPhysCR3;
1975 pVCpu->pgm.s.GCPhysCR3 = GCPhysCR3;
1976 rc = PGM_BTH_PFN(MapCR3, pVCpu)(pVCpu, GCPhysCR3);
1977 if (RT_LIKELY(rc == VINF_SUCCESS))
1978 {
1979 if (pgmMapAreMappingsFloating(pVM))
1980 pVCpu->pgm.s.fSyncFlags &= ~PGM_SYNC_MONITOR_CR3;
1981 }
1982 else
1983 {
1984 AssertMsg(rc == VINF_PGM_SYNC_CR3, ("%Rrc\n", rc));
1985 Assert(VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL | VMCPU_FF_PGM_SYNC_CR3));
1986 pVCpu->pgm.s.GCPhysCR3 = GCPhysOldCR3;
1987 pVCpu->pgm.s.fSyncFlags |= PGM_SYNC_MAP_CR3;
1988 if (pgmMapAreMappingsFloating(pVM))
1989 pVCpu->pgm.s.fSyncFlags |= PGM_SYNC_MONITOR_CR3;
1990 }
1991
1992 if (fGlobal)
1993 STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,FlushTLBNewCR3Global));
1994 else
1995 STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,FlushTLBNewCR3));
1996 }
1997 else
1998 {
1999# ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
2000 PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
2001 if (pPool->cDirtyPages)
2002 {
2003 pgmLock(pVM);
2004 pgmPoolResetDirtyPages(pVM);
2005 pgmUnlock(pVM);
2006 }
2007# endif
2008 /*
2009 * Check if we have a pending update of the CR3 monitoring.
2010 */
2011 if (pVCpu->pgm.s.fSyncFlags & PGM_SYNC_MONITOR_CR3)
2012 {
2013 pVCpu->pgm.s.fSyncFlags &= ~PGM_SYNC_MONITOR_CR3;
2014 Assert(!pVM->pgm.s.fMappingsFixed); Assert(pgmMapAreMappingsEnabled(pVM));
2015 }
2016 if (fGlobal)
2017 STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,FlushTLBSameCR3Global));
2018 else
2019 STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,FlushTLBSameCR3));
2020 }
2021
2022 STAM_PROFILE_STOP(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,FlushTLB), a);
2023 return rc;
2024}
2025
2026
2027/**
2028 * Performs and schedules necessary updates following a CR3 load or reload when
2029 * using nested or extended paging.
2030 *
2031 * This API is an alternative to PDMFlushTLB that avoids actually flushing the
2032 * TLB and triggering a SyncCR3.
2033 *
2034 * This will normally involve mapping the guest PD or nPDPT
2035 *
2036 * @returns VBox status code.
2037 * @retval VINF_SUCCESS.
2038 * @retval VINF_PGM_SYNC_CR3 if monitoring requires a CR3 sync (not for nested
2039 * paging modes). This can safely be ignored and overridden since the
2040 * FF will be set too then.
2041 * @param pVCpu Pointer to the VMCPU.
2042 * @param cr3 The new cr3.
2043 */
2044VMMDECL(int) PGMUpdateCR3(PVMCPU pVCpu, uint64_t cr3)
2045{
2046 VMCPU_ASSERT_EMT(pVCpu);
2047 LogFlow(("PGMUpdateCR3: cr3=%RX64 OldCr3=%RX64\n", cr3, pVCpu->pgm.s.GCPhysCR3));
2048
2049 /* We assume we're only called in nested paging mode. */
2050 Assert(pVCpu->CTX_SUFF(pVM)->pgm.s.fNestedPaging || pVCpu->pgm.s.enmShadowMode == PGMMODE_EPT);
2051 Assert(!pgmMapAreMappingsEnabled(pVCpu->CTX_SUFF(pVM)));
2052 Assert(!(pVCpu->pgm.s.fSyncFlags & PGM_SYNC_MONITOR_CR3));
2053
2054 /*
2055 * Remap the CR3 content and adjust the monitoring if CR3 was actually changed.
2056 */
2057 int rc = VINF_SUCCESS;
2058 RTGCPHYS GCPhysCR3;
2059 switch (pVCpu->pgm.s.enmGuestMode)
2060 {
2061 case PGMMODE_PAE:
2062 case PGMMODE_PAE_NX:
2063 GCPhysCR3 = (RTGCPHYS)(cr3 & X86_CR3_PAE_PAGE_MASK);
2064 break;
2065 case PGMMODE_AMD64:
2066 case PGMMODE_AMD64_NX:
2067 GCPhysCR3 = (RTGCPHYS)(cr3 & X86_CR3_AMD64_PAGE_MASK);
2068 break;
2069 default:
2070 GCPhysCR3 = (RTGCPHYS)(cr3 & X86_CR3_PAGE_MASK);
2071 break;
2072 }
2073 PGM_A20_APPLY_TO_VAR(pVCpu, GCPhysCR3);
2074
2075 if (pVCpu->pgm.s.GCPhysCR3 != GCPhysCR3)
2076 {
2077 pVCpu->pgm.s.GCPhysCR3 = GCPhysCR3;
2078 rc = PGM_BTH_PFN(MapCR3, pVCpu)(pVCpu, GCPhysCR3);
2079 AssertRCSuccess(rc); /* Assumes VINF_PGM_SYNC_CR3 doesn't apply to nested paging. */ /** @todo this isn't true for the mac, but we need hw to test/fix this. */
2080 }
2081
2082 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_HM_UPDATE_CR3);
2083 return rc;
2084}
2085
2086
2087/**
2088 * Synchronize the paging structures.
2089 *
2090 * This function is called in response to the VM_FF_PGM_SYNC_CR3 and
2091 * VM_FF_PGM_SYNC_CR3_NONGLOBAL. Those two force action flags are set
2092 * in several places, most importantly whenever the CR3 is loaded.
2093 *
2094 * @returns VBox status code.
2095 * @param pVCpu Pointer to the VMCPU.
2096 * @param cr0 Guest context CR0 register
2097 * @param cr3 Guest context CR3 register
2098 * @param cr4 Guest context CR4 register
2099 * @param fGlobal Including global page directories or not
2100 */
2101VMMDECL(int) PGMSyncCR3(PVMCPU pVCpu, uint64_t cr0, uint64_t cr3, uint64_t cr4, bool fGlobal)
2102{
2103 int rc;
2104
2105 VMCPU_ASSERT_EMT(pVCpu);
2106
2107 /*
2108 * The pool may have pending stuff and even require a return to ring-3 to
2109 * clear the whole thing.
2110 */
2111 rc = pgmPoolSyncCR3(pVCpu);
2112 if (rc != VINF_SUCCESS)
2113 return rc;
2114
2115 /*
2116 * We might be called when we shouldn't.
2117 *
2118 * The mode switching will ensure that the PD is resynced after every mode
2119 * switch. So, if we find ourselves here when in protected or real mode
2120 * we can safely clear the FF and return immediately.
2121 */
2122 if (pVCpu->pgm.s.enmGuestMode <= PGMMODE_PROTECTED)
2123 {
2124 Assert((cr0 & (X86_CR0_PG | X86_CR0_PE)) != (X86_CR0_PG | X86_CR0_PE));
2125 Assert(!(pVCpu->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL));
2126 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
2127 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL);
2128 return VINF_SUCCESS;
2129 }
2130
2131 /* If global pages are not supported, then all flushes are global. */
2132 if (!(cr4 & X86_CR4_PGE))
2133 fGlobal = true;
2134 LogFlow(("PGMSyncCR3: cr0=%RX64 cr3=%RX64 cr4=%RX64 fGlobal=%d[%d,%d]\n", cr0, cr3, cr4, fGlobal,
2135 VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3), VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL)));
2136
2137 /*
2138 * Check if we need to finish an aborted MapCR3 call (see PGMFlushTLB).
2139 * This should be done before SyncCR3.
2140 */
2141 if (pVCpu->pgm.s.fSyncFlags & PGM_SYNC_MAP_CR3)
2142 {
2143 pVCpu->pgm.s.fSyncFlags &= ~PGM_SYNC_MAP_CR3;
2144
2145 RTGCPHYS GCPhysCR3Old = pVCpu->pgm.s.GCPhysCR3; NOREF(GCPhysCR3Old);
2146 RTGCPHYS GCPhysCR3;
2147 switch (pVCpu->pgm.s.enmGuestMode)
2148 {
2149 case PGMMODE_PAE:
2150 case PGMMODE_PAE_NX:
2151 GCPhysCR3 = (RTGCPHYS)(cr3 & X86_CR3_PAE_PAGE_MASK);
2152 break;
2153 case PGMMODE_AMD64:
2154 case PGMMODE_AMD64_NX:
2155 GCPhysCR3 = (RTGCPHYS)(cr3 & X86_CR3_AMD64_PAGE_MASK);
2156 break;
2157 default:
2158 GCPhysCR3 = (RTGCPHYS)(cr3 & X86_CR3_PAGE_MASK);
2159 break;
2160 }
2161 PGM_A20_APPLY_TO_VAR(pVCpu, GCPhysCR3);
2162
2163 if (pVCpu->pgm.s.GCPhysCR3 != GCPhysCR3)
2164 {
2165 pVCpu->pgm.s.GCPhysCR3 = GCPhysCR3;
2166 rc = PGM_BTH_PFN(MapCR3, pVCpu)(pVCpu, GCPhysCR3);
2167 }
2168
2169 /* Make sure we check for pending pgm pool syncs as we clear VMCPU_FF_PGM_SYNC_CR3 later on! */
2170 if ( rc == VINF_PGM_SYNC_CR3
2171 || (pVCpu->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL))
2172 {
2173 Log(("PGMSyncCR3: pending pgm pool sync after MapCR3!\n"));
2174#ifdef IN_RING3
2175 rc = pgmPoolSyncCR3(pVCpu);
2176#else
2177 if (rc == VINF_PGM_SYNC_CR3)
2178 pVCpu->pgm.s.GCPhysCR3 = GCPhysCR3Old;
2179 return VINF_PGM_SYNC_CR3;
2180#endif
2181 }
2182 AssertRCReturn(rc, rc);
2183 AssertRCSuccessReturn(rc, VERR_IPE_UNEXPECTED_INFO_STATUS);
2184 }
2185
2186 /*
2187 * Let the 'Bth' function do the work and we'll just keep track of the flags.
2188 */
2189 STAM_PROFILE_START(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,SyncCR3), a);
2190 rc = PGM_BTH_PFN(SyncCR3, pVCpu)(pVCpu, cr0, cr3, cr4, fGlobal);
2191 STAM_PROFILE_STOP(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,SyncCR3), a);
2192 AssertMsg(rc == VINF_SUCCESS || rc == VINF_PGM_SYNC_CR3 || RT_FAILURE(rc), ("rc=%Rrc\n", rc));
2193 if (rc == VINF_SUCCESS)
2194 {
2195 if (pVCpu->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL)
2196 {
2197 /* Go back to ring 3 if a pgm pool sync is again pending. */
2198 return VINF_PGM_SYNC_CR3;
2199 }
2200
2201 if (!(pVCpu->pgm.s.fSyncFlags & PGM_SYNC_ALWAYS))
2202 {
2203 Assert(!(pVCpu->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL));
2204 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
2205 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL);
2206 }
2207
2208 /*
2209 * Check if we have a pending update of the CR3 monitoring.
2210 */
2211 if (pVCpu->pgm.s.fSyncFlags & PGM_SYNC_MONITOR_CR3)
2212 {
2213 pVCpu->pgm.s.fSyncFlags &= ~PGM_SYNC_MONITOR_CR3;
2214 Assert(!pVCpu->CTX_SUFF(pVM)->pgm.s.fMappingsFixed);
2215 Assert(pgmMapAreMappingsEnabled(pVCpu->CTX_SUFF(pVM)));
2216 }
2217 }
2218
2219 /*
2220 * Now flush the CR3 (guest context).
2221 */
2222 if (rc == VINF_SUCCESS)
2223 PGM_INVL_VCPU_TLBS(pVCpu);
2224 return rc;
2225}
2226
2227
2228/**
2229 * Called whenever CR0 or CR4 in a way which may affect the paging mode.
2230 *
2231 * @returns VBox status code, with the following informational code for
2232 * VM scheduling.
2233 * @retval VINF_SUCCESS if the was no change, or it was successfully dealt with.
2234 * @retval VINF_PGM_CHANGE_MODE if we're in RC or R0 and the mode changes.
2235 * (I.e. not in R3.)
2236 * @retval VINF_EM_SUSPEND or VINF_EM_OFF on a fatal runtime error. (R3 only)
2237 *
2238 * @param pVCpu Pointer to the VMCPU.
2239 * @param cr0 The new cr0.
2240 * @param cr4 The new cr4.
2241 * @param efer The new extended feature enable register.
2242 */
2243VMMDECL(int) PGMChangeMode(PVMCPU pVCpu, uint64_t cr0, uint64_t cr4, uint64_t efer)
2244{
2245 PGMMODE enmGuestMode;
2246
2247 VMCPU_ASSERT_EMT(pVCpu);
2248
2249 /*
2250 * Calc the new guest mode.
2251 */
2252 if (!(cr0 & X86_CR0_PE))
2253 enmGuestMode = PGMMODE_REAL;
2254 else if (!(cr0 & X86_CR0_PG))
2255 enmGuestMode = PGMMODE_PROTECTED;
2256 else if (!(cr4 & X86_CR4_PAE))
2257 {
2258 bool const fPse = !!(cr4 & X86_CR4_PSE);
2259 if (pVCpu->pgm.s.fGst32BitPageSizeExtension != fPse)
2260 Log(("PGMChangeMode: CR4.PSE %d -> %d\n", pVCpu->pgm.s.fGst32BitPageSizeExtension, fPse));
2261 pVCpu->pgm.s.fGst32BitPageSizeExtension = fPse;
2262 enmGuestMode = PGMMODE_32_BIT;
2263 }
2264 else if (!(efer & MSR_K6_EFER_LME))
2265 {
2266 if (!(efer & MSR_K6_EFER_NXE))
2267 enmGuestMode = PGMMODE_PAE;
2268 else
2269 enmGuestMode = PGMMODE_PAE_NX;
2270 }
2271 else
2272 {
2273 if (!(efer & MSR_K6_EFER_NXE))
2274 enmGuestMode = PGMMODE_AMD64;
2275 else
2276 enmGuestMode = PGMMODE_AMD64_NX;
2277 }
2278
2279 /*
2280 * Did it change?
2281 */
2282 if (pVCpu->pgm.s.enmGuestMode == enmGuestMode)
2283 return VINF_SUCCESS;
2284
2285 /* Flush the TLB */
2286 PGM_INVL_VCPU_TLBS(pVCpu);
2287
2288#ifdef IN_RING3
2289 return PGMR3ChangeMode(pVCpu->CTX_SUFF(pVM), pVCpu, enmGuestMode);
2290#else
2291 LogFlow(("PGMChangeMode: returns VINF_PGM_CHANGE_MODE.\n"));
2292 return VINF_PGM_CHANGE_MODE;
2293#endif
2294}
2295
2296
2297/**
2298 * Called by CPUM or REM when CR0.WP changes to 1.
2299 *
2300 * @param pVCpu The cross context virtual CPU structure of the caller.
2301 * @thread EMT
2302 */
2303VMMDECL(void) PGMCr0WpEnabled(PVMCPU pVCpu)
2304{
2305 /*
2306 * Netware WP0+RO+US hack cleanup when WP0 -> WP1.
2307 *
2308 * Use the counter to judge whether there might be pool pages with active
2309 * hacks in them. If there are, we will be running the risk of messing up
2310 * the guest by allowing it to write to read-only pages. Thus, we have to
2311 * clear the page pool ASAP if there is the slightest chance.
2312 */
2313 if (pVCpu->pgm.s.cNetwareWp0Hacks > 0)
2314 {
2315 Assert(pVCpu->CTX_SUFF(pVM)->cCpus == 1);
2316
2317 Log(("PGMCr0WpEnabled: %llu WP0 hacks active - clearing page pool\n", pVCpu->pgm.s.cNetwareWp0Hacks));
2318 pVCpu->pgm.s.cNetwareWp0Hacks = 0;
2319 pVCpu->pgm.s.fSyncFlags |= PGM_SYNC_CLEAR_PGM_POOL;
2320 VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
2321 }
2322}
2323
2324
2325/**
2326 * Gets the current guest paging mode.
2327 *
2328 * If you just need the CPU mode (real/protected/long), use CPUMGetGuestMode().
2329 *
2330 * @returns The current paging mode.
2331 * @param pVCpu Pointer to the VMCPU.
2332 */
2333VMMDECL(PGMMODE) PGMGetGuestMode(PVMCPU pVCpu)
2334{
2335 return pVCpu->pgm.s.enmGuestMode;
2336}
2337
2338
2339/**
2340 * Gets the current shadow paging mode.
2341 *
2342 * @returns The current paging mode.
2343 * @param pVCpu Pointer to the VMCPU.
2344 */
2345VMMDECL(PGMMODE) PGMGetShadowMode(PVMCPU pVCpu)
2346{
2347 return pVCpu->pgm.s.enmShadowMode;
2348}
2349
2350
2351/**
2352 * Gets the current host paging mode.
2353 *
2354 * @returns The current paging mode.
2355 * @param pVM Pointer to the VM.
2356 */
2357VMMDECL(PGMMODE) PGMGetHostMode(PVM pVM)
2358{
2359 switch (pVM->pgm.s.enmHostMode)
2360 {
2361 case SUPPAGINGMODE_32_BIT:
2362 case SUPPAGINGMODE_32_BIT_GLOBAL:
2363 return PGMMODE_32_BIT;
2364
2365 case SUPPAGINGMODE_PAE:
2366 case SUPPAGINGMODE_PAE_GLOBAL:
2367 return PGMMODE_PAE;
2368
2369 case SUPPAGINGMODE_PAE_NX:
2370 case SUPPAGINGMODE_PAE_GLOBAL_NX:
2371 return PGMMODE_PAE_NX;
2372
2373 case SUPPAGINGMODE_AMD64:
2374 case SUPPAGINGMODE_AMD64_GLOBAL:
2375 return PGMMODE_AMD64;
2376
2377 case SUPPAGINGMODE_AMD64_NX:
2378 case SUPPAGINGMODE_AMD64_GLOBAL_NX:
2379 return PGMMODE_AMD64_NX;
2380
2381 default: AssertMsgFailed(("enmHostMode=%d\n", pVM->pgm.s.enmHostMode)); break;
2382 }
2383
2384 return PGMMODE_INVALID;
2385}
2386
2387
2388/**
2389 * Get mode name.
2390 *
2391 * @returns read-only name string.
2392 * @param enmMode The mode which name is desired.
2393 */
2394VMMDECL(const char *) PGMGetModeName(PGMMODE enmMode)
2395{
2396 switch (enmMode)
2397 {
2398 case PGMMODE_REAL: return "Real";
2399 case PGMMODE_PROTECTED: return "Protected";
2400 case PGMMODE_32_BIT: return "32-bit";
2401 case PGMMODE_PAE: return "PAE";
2402 case PGMMODE_PAE_NX: return "PAE+NX";
2403 case PGMMODE_AMD64: return "AMD64";
2404 case PGMMODE_AMD64_NX: return "AMD64+NX";
2405 case PGMMODE_NESTED: return "Nested";
2406 case PGMMODE_EPT: return "EPT";
2407 default: return "unknown mode value";
2408 }
2409}
2410
2411
2412
2413/**
2414 * Notification from CPUM that the EFER.NXE bit has changed.
2415 *
2416 * @param pVCpu The virtual CPU for which EFER changed.
2417 * @param fNxe The new NXE state.
2418 */
2419VMM_INT_DECL(void) PGMNotifyNxeChanged(PVMCPU pVCpu, bool fNxe)
2420{
2421/** @todo VMCPU_ASSERT_EMT_OR_NOT_RUNNING(pVCpu); */
2422 Log(("PGMNotifyNxeChanged: fNxe=%RTbool\n", fNxe));
2423
2424 pVCpu->pgm.s.fNoExecuteEnabled = fNxe;
2425 if (fNxe)
2426 {
2427 /*pVCpu->pgm.s.fGst32BitMbzBigPdeMask - N/A */
2428 pVCpu->pgm.s.fGstPaeMbzPteMask &= ~X86_PTE_PAE_NX;
2429 pVCpu->pgm.s.fGstPaeMbzPdeMask &= ~X86_PDE_PAE_NX;
2430 pVCpu->pgm.s.fGstPaeMbzBigPdeMask &= ~X86_PDE2M_PAE_NX;
2431 /*pVCpu->pgm.s.fGstPaeMbzPdpeMask - N/A */
2432 pVCpu->pgm.s.fGstAmd64MbzPteMask &= ~X86_PTE_PAE_NX;
2433 pVCpu->pgm.s.fGstAmd64MbzPdeMask &= ~X86_PDE_PAE_NX;
2434 pVCpu->pgm.s.fGstAmd64MbzBigPdeMask &= ~X86_PDE2M_PAE_NX;
2435 pVCpu->pgm.s.fGstAmd64MbzPdpeMask &= ~X86_PDPE_LM_NX;
2436 pVCpu->pgm.s.fGstAmd64MbzBigPdpeMask &= ~X86_PDPE_LM_NX;
2437 pVCpu->pgm.s.fGstAmd64MbzPml4eMask &= ~X86_PML4E_NX;
2438
2439 pVCpu->pgm.s.fGst64ShadowedPteMask |= X86_PTE_PAE_NX;
2440 pVCpu->pgm.s.fGst64ShadowedPdeMask |= X86_PDE_PAE_NX;
2441 pVCpu->pgm.s.fGst64ShadowedBigPdeMask |= X86_PDE2M_PAE_NX;
2442 pVCpu->pgm.s.fGst64ShadowedBigPde4PteMask |= X86_PDE2M_PAE_NX;
2443 pVCpu->pgm.s.fGstAmd64ShadowedPdpeMask |= X86_PDPE_LM_NX;
2444 pVCpu->pgm.s.fGstAmd64ShadowedPml4eMask |= X86_PML4E_NX;
2445 }
2446 else
2447 {
2448 /*pVCpu->pgm.s.fGst32BitMbzBigPdeMask - N/A */
2449 pVCpu->pgm.s.fGstPaeMbzPteMask |= X86_PTE_PAE_NX;
2450 pVCpu->pgm.s.fGstPaeMbzPdeMask |= X86_PDE_PAE_NX;
2451 pVCpu->pgm.s.fGstPaeMbzBigPdeMask |= X86_PDE2M_PAE_NX;
2452 /*pVCpu->pgm.s.fGstPaeMbzPdpeMask -N/A */
2453 pVCpu->pgm.s.fGstAmd64MbzPteMask |= X86_PTE_PAE_NX;
2454 pVCpu->pgm.s.fGstAmd64MbzPdeMask |= X86_PDE_PAE_NX;
2455 pVCpu->pgm.s.fGstAmd64MbzBigPdeMask |= X86_PDE2M_PAE_NX;
2456 pVCpu->pgm.s.fGstAmd64MbzPdpeMask |= X86_PDPE_LM_NX;
2457 pVCpu->pgm.s.fGstAmd64MbzBigPdpeMask |= X86_PDPE_LM_NX;
2458 pVCpu->pgm.s.fGstAmd64MbzPml4eMask |= X86_PML4E_NX;
2459
2460 pVCpu->pgm.s.fGst64ShadowedPteMask &= ~X86_PTE_PAE_NX;
2461 pVCpu->pgm.s.fGst64ShadowedPdeMask &= ~X86_PDE_PAE_NX;
2462 pVCpu->pgm.s.fGst64ShadowedBigPdeMask &= ~X86_PDE2M_PAE_NX;
2463 pVCpu->pgm.s.fGst64ShadowedBigPde4PteMask &= ~X86_PDE2M_PAE_NX;
2464 pVCpu->pgm.s.fGstAmd64ShadowedPdpeMask &= ~X86_PDPE_LM_NX;
2465 pVCpu->pgm.s.fGstAmd64ShadowedPml4eMask &= ~X86_PML4E_NX;
2466 }
2467}
2468
2469
2470/**
2471 * Check if any pgm pool pages are marked dirty (not monitored)
2472 *
2473 * @returns bool locked/not locked
2474 * @param pVM Pointer to the VM.
2475 */
2476VMMDECL(bool) PGMHasDirtyPages(PVM pVM)
2477{
2478 return pVM->pgm.s.CTX_SUFF(pPool)->cDirtyPages != 0;
2479}
2480
2481
2482/**
2483 * Check if this VCPU currently owns the PGM lock.
2484 *
2485 * @returns bool owner/not owner
2486 * @param pVM Pointer to the VM.
2487 */
2488VMMDECL(bool) PGMIsLockOwner(PVM pVM)
2489{
2490 return PDMCritSectIsOwner(&pVM->pgm.s.CritSectX);
2491}
2492
2493
2494/**
2495 * Enable or disable large page usage
2496 *
2497 * @returns VBox status code.
2498 * @param pVM Pointer to the VM.
2499 * @param fUseLargePages Use/not use large pages
2500 */
2501VMMDECL(int) PGMSetLargePageUsage(PVM pVM, bool fUseLargePages)
2502{
2503 VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
2504
2505 pVM->fUseLargePages = fUseLargePages;
2506 return VINF_SUCCESS;
2507}
2508
2509
2510/**
2511 * Acquire the PGM lock.
2512 *
2513 * @returns VBox status code
2514 * @param pVM Pointer to the VM.
2515 */
2516#if defined(VBOX_STRICT) && defined(IN_RING3)
2517int pgmLockDebug(PVM pVM, RT_SRC_POS_DECL)
2518#else
2519int pgmLock(PVM pVM)
2520#endif
2521{
2522#if defined(VBOX_STRICT) && defined(IN_RING3)
2523 int rc = PDMCritSectEnterDebug(&pVM->pgm.s.CritSectX, VERR_SEM_BUSY, (uintptr_t)ASMReturnAddress(), RT_SRC_POS_ARGS);
2524#else
2525 int rc = PDMCritSectEnter(&pVM->pgm.s.CritSectX, VERR_SEM_BUSY);
2526#endif
2527#if defined(IN_RC) || defined(IN_RING0)
2528 if (rc == VERR_SEM_BUSY)
2529 rc = VMMRZCallRing3NoCpu(pVM, VMMCALLRING3_PGM_LOCK, 0);
2530#endif
2531 AssertMsg(rc == VINF_SUCCESS, ("%Rrc\n", rc));
2532 return rc;
2533}
2534
2535
2536/**
2537 * Release the PGM lock.
2538 *
2539 * @returns VBox status code
2540 * @param pVM Pointer to the VM.
2541 */
2542void pgmUnlock(PVM pVM)
2543{
2544 uint32_t cDeprecatedPageLocks = pVM->pgm.s.cDeprecatedPageLocks;
2545 pVM->pgm.s.cDeprecatedPageLocks = 0;
2546 int rc = PDMCritSectLeave(&pVM->pgm.s.CritSectX);
2547 if (rc == VINF_SEM_NESTED)
2548 pVM->pgm.s.cDeprecatedPageLocks = cDeprecatedPageLocks;
2549}
2550
2551#if defined(IN_RC) || defined(VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0)
2552
2553/**
2554 * Common worker for pgmRZDynMapGCPageOffInlined and pgmRZDynMapGCPageV2Inlined.
2555 *
2556 * @returns VBox status code.
2557 * @param pVM Pointer to the VM.
2558 * @param pVCpu The current CPU.
2559 * @param GCPhys The guest physical address of the page to map. The
2560 * offset bits are not ignored.
2561 * @param ppv Where to return the address corresponding to @a GCPhys.
2562 */
2563int pgmRZDynMapGCPageCommon(PVM pVM, PVMCPU pVCpu, RTGCPHYS GCPhys, void **ppv RTLOG_COMMA_SRC_POS_DECL)
2564{
2565 pgmLock(pVM);
2566
2567 /*
2568 * Convert it to a writable page and it on to the dynamic mapper.
2569 */
2570 int rc;
2571 PPGMPAGE pPage = pgmPhysGetPage(pVM, GCPhys);
2572 if (RT_LIKELY(pPage))
2573 {
2574 rc = pgmPhysPageMakeWritable(pVM, pPage, GCPhys);
2575 if (RT_SUCCESS(rc))
2576 {
2577 void *pv;
2578 rc = pgmRZDynMapHCPageInlined(pVCpu, PGM_PAGE_GET_HCPHYS(pPage), &pv RTLOG_COMMA_SRC_POS_ARGS);
2579 if (RT_SUCCESS(rc))
2580 *ppv = (void *)((uintptr_t)pv | ((uintptr_t)GCPhys & PAGE_OFFSET_MASK));
2581 }
2582 else
2583 AssertRC(rc);
2584 }
2585 else
2586 {
2587 AssertMsgFailed(("Invalid physical address %RGp!\n", GCPhys));
2588 rc = VERR_PGM_INVALID_GC_PHYSICAL_ADDRESS;
2589 }
2590
2591 pgmUnlock(pVM);
2592 return rc;
2593}
2594
2595#endif /* IN_RC || VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0 */
2596#if !defined(IN_R0) || defined(LOG_ENABLED)
2597
2598/** Format handler for PGMPAGE.
2599 * @copydoc FNRTSTRFORMATTYPE */
2600static DECLCALLBACK(size_t) pgmFormatTypeHandlerPage(PFNRTSTROUTPUT pfnOutput, void *pvArgOutput,
2601 const char *pszType, void const *pvValue,
2602 int cchWidth, int cchPrecision, unsigned fFlags,
2603 void *pvUser)
2604{
2605 size_t cch;
2606 PCPGMPAGE pPage = (PCPGMPAGE)pvValue;
2607 if (RT_VALID_PTR(pPage))
2608 {
2609 char szTmp[64+80];
2610
2611 cch = 0;
2612
2613 /* The single char state stuff. */
2614 static const char s_achPageStates[4] = { 'Z', 'A', 'W', 'S' };
2615 szTmp[cch++] = s_achPageStates[PGM_PAGE_GET_STATE_NA(pPage)];
2616
2617#define IS_PART_INCLUDED(lvl) ( !(fFlags & RTSTR_F_PRECISION) || cchPrecision == (lvl) || cchPrecision >= (lvl)+10 )
2618 if (IS_PART_INCLUDED(5))
2619 {
2620 static const char s_achHandlerStates[4] = { '-', 't', 'w', 'a' };
2621 szTmp[cch++] = s_achHandlerStates[PGM_PAGE_GET_HNDL_PHYS_STATE(pPage)];
2622 szTmp[cch++] = s_achHandlerStates[PGM_PAGE_GET_HNDL_VIRT_STATE(pPage)];
2623 }
2624
2625 /* The type. */
2626 if (IS_PART_INCLUDED(4))
2627 {
2628 szTmp[cch++] = ':';
2629 static const char s_achPageTypes[8][4] = { "INV", "RAM", "MI2", "M2A", "SHA", "ROM", "MIO", "BAD" };
2630 szTmp[cch++] = s_achPageTypes[PGM_PAGE_GET_TYPE_NA(pPage)][0];
2631 szTmp[cch++] = s_achPageTypes[PGM_PAGE_GET_TYPE_NA(pPage)][1];
2632 szTmp[cch++] = s_achPageTypes[PGM_PAGE_GET_TYPE_NA(pPage)][2];
2633 }
2634
2635 /* The numbers. */
2636 if (IS_PART_INCLUDED(3))
2637 {
2638 szTmp[cch++] = ':';
2639 cch += RTStrFormatNumber(&szTmp[cch], PGM_PAGE_GET_HCPHYS_NA(pPage), 16, 12, 0, RTSTR_F_ZEROPAD | RTSTR_F_64BIT);
2640 }
2641
2642 if (IS_PART_INCLUDED(2))
2643 {
2644 szTmp[cch++] = ':';
2645 cch += RTStrFormatNumber(&szTmp[cch], PGM_PAGE_GET_PAGEID(pPage), 16, 7, 0, RTSTR_F_ZEROPAD | RTSTR_F_32BIT);
2646 }
2647
2648 if (IS_PART_INCLUDED(6))
2649 {
2650 szTmp[cch++] = ':';
2651 static const char s_achRefs[4] = { '-', 'U', '!', 'L' };
2652 szTmp[cch++] = s_achRefs[PGM_PAGE_GET_TD_CREFS_NA(pPage)];
2653 cch += RTStrFormatNumber(&szTmp[cch], PGM_PAGE_GET_TD_IDX_NA(pPage), 16, 4, 0, RTSTR_F_ZEROPAD | RTSTR_F_16BIT);
2654 }
2655#undef IS_PART_INCLUDED
2656
2657 cch = pfnOutput(pvArgOutput, szTmp, cch);
2658 }
2659 else
2660 cch = pfnOutput(pvArgOutput, RT_STR_TUPLE("<bad-pgmpage-ptr>"));
2661 NOREF(pszType); NOREF(cchWidth); NOREF(pvUser);
2662 return cch;
2663}
2664
2665
2666/** Format handler for PGMRAMRANGE.
2667 * @copydoc FNRTSTRFORMATTYPE */
2668static DECLCALLBACK(size_t) pgmFormatTypeHandlerRamRange(PFNRTSTROUTPUT pfnOutput, void *pvArgOutput,
2669 const char *pszType, void const *pvValue,
2670 int cchWidth, int cchPrecision, unsigned fFlags,
2671 void *pvUser)
2672{
2673 size_t cch;
2674 PGMRAMRANGE const *pRam = (PGMRAMRANGE const *)pvValue;
2675 if (VALID_PTR(pRam))
2676 {
2677 char szTmp[80];
2678 cch = RTStrPrintf(szTmp, sizeof(szTmp), "%RGp-%RGp", pRam->GCPhys, pRam->GCPhysLast);
2679 cch = pfnOutput(pvArgOutput, szTmp, cch);
2680 }
2681 else
2682 cch = pfnOutput(pvArgOutput, RT_STR_TUPLE("<bad-pgmramrange-ptr>"));
2683 NOREF(pszType); NOREF(cchWidth); NOREF(cchPrecision); NOREF(pvUser); NOREF(fFlags);
2684 return cch;
2685}
2686
2687/** Format type andlers to be registered/deregistered. */
2688static const struct
2689{
2690 char szType[24];
2691 PFNRTSTRFORMATTYPE pfnHandler;
2692} g_aPgmFormatTypes[] =
2693{
2694 { "pgmpage", pgmFormatTypeHandlerPage },
2695 { "pgmramrange", pgmFormatTypeHandlerRamRange }
2696};
2697
2698#endif /* !IN_R0 || LOG_ENABLED */
2699
2700/**
2701 * Registers the global string format types.
2702 *
2703 * This should be called at module load time or in some other manner that ensure
2704 * that it's called exactly one time.
2705 *
2706 * @returns IPRT status code on RTStrFormatTypeRegister failure.
2707 */
2708VMMDECL(int) PGMRegisterStringFormatTypes(void)
2709{
2710#if !defined(IN_R0) || defined(LOG_ENABLED)
2711 int rc = VINF_SUCCESS;
2712 unsigned i;
2713 for (i = 0; RT_SUCCESS(rc) && i < RT_ELEMENTS(g_aPgmFormatTypes); i++)
2714 {
2715 rc = RTStrFormatTypeRegister(g_aPgmFormatTypes[i].szType, g_aPgmFormatTypes[i].pfnHandler, NULL);
2716# ifdef IN_RING0
2717 if (rc == VERR_ALREADY_EXISTS)
2718 {
2719 /* in case of cleanup failure in ring-0 */
2720 RTStrFormatTypeDeregister(g_aPgmFormatTypes[i].szType);
2721 rc = RTStrFormatTypeRegister(g_aPgmFormatTypes[i].szType, g_aPgmFormatTypes[i].pfnHandler, NULL);
2722 }
2723# endif
2724 }
2725 if (RT_FAILURE(rc))
2726 while (i-- > 0)
2727 RTStrFormatTypeDeregister(g_aPgmFormatTypes[i].szType);
2728
2729 return rc;
2730#else
2731 return VINF_SUCCESS;
2732#endif
2733}
2734
2735
2736/**
2737 * Deregisters the global string format types.
2738 *
2739 * This should be called at module unload time or in some other manner that
2740 * ensure that it's called exactly one time.
2741 */
2742VMMDECL(void) PGMDeregisterStringFormatTypes(void)
2743{
2744#if !defined(IN_R0) || defined(LOG_ENABLED)
2745 for (unsigned i = 0; i < RT_ELEMENTS(g_aPgmFormatTypes); i++)
2746 RTStrFormatTypeDeregister(g_aPgmFormatTypes[i].szType);
2747#endif
2748}
2749
2750#ifdef VBOX_STRICT
2751
2752/**
2753 * Asserts that there are no mapping conflicts.
2754 *
2755 * @returns Number of conflicts.
2756 * @param pVM Pointer to the VM.
2757 */
2758VMMDECL(unsigned) PGMAssertNoMappingConflicts(PVM pVM)
2759{
2760 unsigned cErrors = 0;
2761
2762 /* Only applies to raw mode -> 1 VPCU */
2763 Assert(pVM->cCpus == 1);
2764 PVMCPU pVCpu = &pVM->aCpus[0];
2765
2766 /*
2767 * Check for mapping conflicts.
2768 */
2769 for (PPGMMAPPING pMapping = pVM->pgm.s.CTX_SUFF(pMappings);
2770 pMapping;
2771 pMapping = pMapping->CTX_SUFF(pNext))
2772 {
2773 /** @todo This is slow and should be optimized, but since it's just assertions I don't care now. */
2774 for (RTGCPTR GCPtr = pMapping->GCPtr;
2775 GCPtr <= pMapping->GCPtrLast;
2776 GCPtr += PAGE_SIZE)
2777 {
2778 int rc = PGMGstGetPage(pVCpu, (RTGCPTR)GCPtr, NULL, NULL);
2779 if (rc != VERR_PAGE_TABLE_NOT_PRESENT)
2780 {
2781 AssertMsgFailed(("Conflict at %RGv with %s\n", GCPtr, R3STRING(pMapping->pszDesc)));
2782 cErrors++;
2783 break;
2784 }
2785 }
2786 }
2787
2788 return cErrors;
2789}
2790
2791
2792/**
2793 * Asserts that everything related to the guest CR3 is correctly shadowed.
2794 *
2795 * This will call PGMAssertNoMappingConflicts() and PGMAssertHandlerAndFlagsInSync(),
2796 * and assert the correctness of the guest CR3 mapping before asserting that the
2797 * shadow page tables is in sync with the guest page tables.
2798 *
2799 * @returns Number of conflicts.
2800 * @param pVM Pointer to the VM.
2801 * @param pVCpu Pointer to the VMCPU.
2802 * @param cr3 The current guest CR3 register value.
2803 * @param cr4 The current guest CR4 register value.
2804 */
2805VMMDECL(unsigned) PGMAssertCR3(PVM pVM, PVMCPU pVCpu, uint64_t cr3, uint64_t cr4)
2806{
2807 STAM_PROFILE_START(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,SyncCR3), a);
2808 pgmLock(pVM);
2809 unsigned cErrors = PGM_BTH_PFN(AssertCR3, pVCpu)(pVCpu, cr3, cr4, 0, ~(RTGCPTR)0);
2810 pgmUnlock(pVM);
2811 STAM_PROFILE_STOP(&pVCpu->pgm.s.CTX_SUFF(pStats)->CTX_MID_Z(Stat,SyncCR3), a);
2812 return cErrors;
2813}
2814
2815#endif /* VBOX_STRICT */
Note: See TracBrowser for help on using the repository browser.

© 2023 Oracle
ContactPrivacy policyTerms of Use