PGMPhys.cpp@ 100696

Last change on this file since 100696 was 99739, checked in by vboxsync, 20 months ago
*: doxygen corrections (mostly about removing @returns from functions returning void).
Property svn:eol-style set to `native` Property svn:keywords set to `Id Revision`
File size: 225.0 KB

Line
1	/* $Id: PGMPhys.cpp 99739 2023-05-11 01:01:08Z vboxsync $ */
2	/** @file
3	* PGM - Page Manager and Monitor, Physical Memory Addressing.
4	*/
5
6	/*
7	* Copyright (C) 2006-2023 Oracle and/or its affiliates.
8	*
9	* This file is part of VirtualBox base platform packages, as
10	* available from https://www.virtualbox.org.
11	*
12	* This program is free software; you can redistribute it and/or
13	* modify it under the terms of the GNU General Public License
14	* as published by the Free Software Foundation, in version 3 of the
15	* License.
16	*
17	* This program is distributed in the hope that it will be useful, but
18	* WITHOUT ANY WARRANTY; without even the implied warranty of
19	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20	* General Public License for more details.
21	*
22	* You should have received a copy of the GNU General Public License
23	* along with this program; if not, see <https://www.gnu.org/licenses>.
24	*
25	* SPDX-License-Identifier: GPL-3.0-only
26	*/
27
28
29	/*********************************************************************************************************************************
30	* Header Files *
31	*********************************************************************************************************************************/
32	#define LOG_GROUP LOG_GROUP_PGM_PHYS
33	#define VBOX_WITHOUT_PAGING_BIT_FIELDS /* 64-bit bitfields are just asking for trouble. See @bugref{9841} and others. */
34	#include <VBox/vmm/pgm.h>
35	#include <VBox/vmm/iem.h>
36	#include <VBox/vmm/iom.h>
37	#include <VBox/vmm/mm.h>
38	#include <VBox/vmm/nem.h>
39	#include <VBox/vmm/stam.h>
40	#include <VBox/vmm/pdmdev.h>
41	#include "PGMInternal.h"
42	#include <VBox/vmm/vmcc.h>
43
44	#include "PGMInline.h"
45
46	#include <VBox/sup.h>
47	#include <VBox/param.h>
48	#include <VBox/err.h>
49	#include <VBox/log.h>
50	#include <iprt/assert.h>
51	#include <iprt/alloc.h>
52	#include <iprt/asm.h>
53	#ifdef VBOX_STRICT
54	# include <iprt/crc.h>
55	#endif
56	#include <iprt/thread.h>
57	#include <iprt/string.h>
58	#include <iprt/system.h>
59
60
61	/*********************************************************************************************************************************
62	* Defined Constants And Macros *
63	*********************************************************************************************************************************/
64	/** The number of pages to free in one batch. */
65	#define PGMPHYS_FREE_PAGE_BATCH_SIZE 128
66
67
68
69	/*********************************************************************************************************************************
70	* Reading and Writing Guest Pysical Memory *
71	*********************************************************************************************************************************/
72
73	/*
74	* PGMR3PhysReadU8-64
75	* PGMR3PhysWriteU8-64
76	*/
77	#define PGMPHYSFN_READNAME PGMR3PhysReadU8
78	#define PGMPHYSFN_WRITENAME PGMR3PhysWriteU8
79	#define PGMPHYS_DATASIZE 1
80	#define PGMPHYS_DATATYPE uint8_t
81	#include "PGMPhysRWTmpl.h"
82
83	#define PGMPHYSFN_READNAME PGMR3PhysReadU16
84	#define PGMPHYSFN_WRITENAME PGMR3PhysWriteU16
85	#define PGMPHYS_DATASIZE 2
86	#define PGMPHYS_DATATYPE uint16_t
87	#include "PGMPhysRWTmpl.h"
88
89	#define PGMPHYSFN_READNAME PGMR3PhysReadU32
90	#define PGMPHYSFN_WRITENAME PGMR3PhysWriteU32
91	#define PGMPHYS_DATASIZE 4
92	#define PGMPHYS_DATATYPE uint32_t
93	#include "PGMPhysRWTmpl.h"
94
95	#define PGMPHYSFN_READNAME PGMR3PhysReadU64
96	#define PGMPHYSFN_WRITENAME PGMR3PhysWriteU64
97	#define PGMPHYS_DATASIZE 8
98	#define PGMPHYS_DATATYPE uint64_t
99	#include "PGMPhysRWTmpl.h"
100
101
102	/**
103	* EMT worker for PGMR3PhysReadExternal.
104	*/
105	static DECLCALLBACK(int) pgmR3PhysReadExternalEMT(PVM pVM, PRTGCPHYS pGCPhys, void *pvBuf, size_t cbRead,
106	PGMACCESSORIGIN enmOrigin)
107	{
108	VBOXSTRICTRC rcStrict = PGMPhysRead(pVM, *pGCPhys, pvBuf, cbRead, enmOrigin);
109	AssertMsg(rcStrict == VINF_SUCCESS, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict))); NOREF(rcStrict);
110	return VINF_SUCCESS;
111	}
112
113
114	/**
115	* Read from physical memory, external users.
116	*
117	* @returns VBox status code.
118	* @retval VINF_SUCCESS.
119	*
120	* @param pVM The cross context VM structure.
121	* @param GCPhys Physical address to read from.
122	* @param pvBuf Where to read into.
123	* @param cbRead How many bytes to read.
124	* @param enmOrigin Who is calling.
125	*
126	* @thread Any but EMTs.
127	*/
128	VMMR3DECL(int) PGMR3PhysReadExternal(PVM pVM, RTGCPHYS GCPhys, void *pvBuf, size_t cbRead, PGMACCESSORIGIN enmOrigin)
129	{
130	VM_ASSERT_OTHER_THREAD(pVM);
131
132	AssertMsgReturn(cbRead > 0, ("don't even think about reading zero bytes!\n"), VINF_SUCCESS);
133	LogFlow(("PGMR3PhysReadExternal: %RGp %d\n", GCPhys, cbRead));
134
135	PGM_LOCK_VOID(pVM);
136
137	/*
138	* Copy loop on ram ranges.
139	*/
140	PPGMRAMRANGE pRam = pgmPhysGetRangeAtOrAbove(pVM, GCPhys);
141	for (;;)
142	{
143	/* Inside range or not? */
144	if (pRam && GCPhys >= pRam->GCPhys)
145	{
146	/*
147	* Must work our way thru this page by page.
148	*/
149	RTGCPHYS off = GCPhys - pRam->GCPhys;
150	while (off < pRam->cb)
151	{
152	unsigned iPage = off >> GUEST_PAGE_SHIFT;
153	PPGMPAGE pPage = &pRam->aPages[iPage];
154
155	/*
156	* If the page has an ALL access handler, we'll have to
157	* delegate the job to EMT.
158	*/
159	if ( PGM_PAGE_HAS_ACTIVE_ALL_HANDLERS(pPage)
160	\|\| PGM_PAGE_IS_SPECIAL_ALIAS_MMIO(pPage))
161	{
162	PGM_UNLOCK(pVM);
163
164	return VMR3ReqPriorityCallWait(pVM, VMCPUID_ANY, (PFNRT)pgmR3PhysReadExternalEMT, 5,
165	pVM, &GCPhys, pvBuf, cbRead, enmOrigin);
166	}
167	Assert(!PGM_PAGE_IS_MMIO_OR_SPECIAL_ALIAS(pPage));
168
169	/*
170	* Simple stuff, go ahead.
171	*/
172	size_t cb = GUEST_PAGE_SIZE - (off & GUEST_PAGE_OFFSET_MASK);
173	if (cb > cbRead)
174	cb = cbRead;
175	PGMPAGEMAPLOCK PgMpLck;
176	const void *pvSrc;
177	int rc = pgmPhysGCPhys2CCPtrInternalReadOnly(pVM, pPage, pRam->GCPhys + off, &pvSrc, &PgMpLck);
178	if (RT_SUCCESS(rc))
179	{
180	memcpy(pvBuf, pvSrc, cb);
181	pgmPhysReleaseInternalPageMappingLock(pVM, &PgMpLck);
182	}
183	else
184	{
185	AssertLogRelMsgFailed(("pgmPhysGCPhys2CCPtrInternalReadOnly failed on %RGp / %R[pgmpage] -> %Rrc\n",
186	pRam->GCPhys + off, pPage, rc));
187	memset(pvBuf, 0xff, cb);
188	}
189
190	/* next page */
191	if (cb >= cbRead)
192	{
193	PGM_UNLOCK(pVM);
194	return VINF_SUCCESS;
195	}
196	cbRead -= cb;
197	off += cb;
198	GCPhys += cb;
199	pvBuf = (char *)pvBuf + cb;
200	} /* walk pages in ram range. */
201	}
202	else
203	{
204	LogFlow(("PGMPhysRead: Unassigned %RGp size=%u\n", GCPhys, cbRead));
205
206	/*
207	* Unassigned address space.
208	*/
209	size_t cb = pRam ? pRam->GCPhys - GCPhys : ~(size_t)0;
210	if (cb >= cbRead)
211	{
212	memset(pvBuf, 0xff, cbRead);
213	break;
214	}
215	memset(pvBuf, 0xff, cb);
216
217	cbRead -= cb;
218	pvBuf = (char *)pvBuf + cb;
219	GCPhys += cb;
220	}
221
222	/* Advance range if necessary. */
223	while (pRam && GCPhys > pRam->GCPhysLast)
224	pRam = pRam->CTX_SUFF(pNext);
225	} /* Ram range walk */
226
227	PGM_UNLOCK(pVM);
228
229	return VINF_SUCCESS;
230	}
231
232
233	/**
234	* EMT worker for PGMR3PhysWriteExternal.
235	*/
236	static DECLCALLBACK(int) pgmR3PhysWriteExternalEMT(PVM pVM, PRTGCPHYS pGCPhys, const void *pvBuf, size_t cbWrite,
237	PGMACCESSORIGIN enmOrigin)
238	{
239	/** @todo VERR_EM_NO_MEMORY */
240	VBOXSTRICTRC rcStrict = PGMPhysWrite(pVM, *pGCPhys, pvBuf, cbWrite, enmOrigin);
241	AssertMsg(rcStrict == VINF_SUCCESS, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict))); NOREF(rcStrict);
242	return VINF_SUCCESS;
243	}
244
245
246	/**
247	* Write to physical memory, external users.
248	*
249	* @returns VBox status code.
250	* @retval VINF_SUCCESS.
251	* @retval VERR_EM_NO_MEMORY.
252	*
253	* @param pVM The cross context VM structure.
254	* @param GCPhys Physical address to write to.
255	* @param pvBuf What to write.
256	* @param cbWrite How many bytes to write.
257	* @param enmOrigin Who is calling.
258	*
259	* @thread Any but EMTs.
260	*/
261	VMMDECL(int) PGMR3PhysWriteExternal(PVM pVM, RTGCPHYS GCPhys, const void *pvBuf, size_t cbWrite, PGMACCESSORIGIN enmOrigin)
262	{
263	VM_ASSERT_OTHER_THREAD(pVM);
264
265	AssertMsg(!pVM->pgm.s.fNoMorePhysWrites,
266	("Calling PGMR3PhysWriteExternal after pgmR3Save()! GCPhys=%RGp cbWrite=%#x enmOrigin=%d\n",
267	GCPhys, cbWrite, enmOrigin));
268	AssertMsgReturn(cbWrite > 0, ("don't even think about writing zero bytes!\n"), VINF_SUCCESS);
269	LogFlow(("PGMR3PhysWriteExternal: %RGp %d\n", GCPhys, cbWrite));
270
271	PGM_LOCK_VOID(pVM);
272
273	/*
274	* Copy loop on ram ranges, stop when we hit something difficult.
275	*/
276	PPGMRAMRANGE pRam = pgmPhysGetRangeAtOrAbove(pVM, GCPhys);
277	for (;;)
278	{
279	/* Inside range or not? */
280	if (pRam && GCPhys >= pRam->GCPhys)
281	{
282	/*
283	* Must work our way thru this page by page.
284	*/
285	RTGCPTR off = GCPhys - pRam->GCPhys;
286	while (off < pRam->cb)
287	{
288	RTGCPTR iPage = off >> GUEST_PAGE_SHIFT;
289	PPGMPAGE pPage = &pRam->aPages[iPage];
290
291	/*
292	* Is the page problematic, we have to do the work on the EMT.
293	*
294	* Allocating writable pages and access handlers are
295	* problematic, write monitored pages are simple and can be
296	* dealt with here.
297	*/
298	if ( PGM_PAGE_HAS_ACTIVE_HANDLERS(pPage)
299	\|\| PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_ALLOCATED
300	\|\| PGM_PAGE_IS_SPECIAL_ALIAS_MMIO(pPage))
301	{
302	if ( PGM_PAGE_GET_STATE(pPage) == PGM_PAGE_STATE_WRITE_MONITORED
303	&& !PGM_PAGE_HAS_ACTIVE_HANDLERS(pPage))
304	pgmPhysPageMakeWriteMonitoredWritable(pVM, pPage, GCPhys);
305	else
306	{
307	PGM_UNLOCK(pVM);
308
309	return VMR3ReqPriorityCallWait(pVM, VMCPUID_ANY, (PFNRT)pgmR3PhysWriteExternalEMT, 5,
310	pVM, &GCPhys, pvBuf, cbWrite, enmOrigin);
311	}
312	}
313	Assert(!PGM_PAGE_IS_MMIO_OR_SPECIAL_ALIAS(pPage));
314
315	/*
316	* Simple stuff, go ahead.
317	*/
318	size_t cb = GUEST_PAGE_SIZE - (off & GUEST_PAGE_OFFSET_MASK);
319	if (cb > cbWrite)
320	cb = cbWrite;
321	PGMPAGEMAPLOCK PgMpLck;
322	void *pvDst;
323	int rc = pgmPhysGCPhys2CCPtrInternal(pVM, pPage, pRam->GCPhys + off, &pvDst, &PgMpLck);
324	if (RT_SUCCESS(rc))
325	{
326	memcpy(pvDst, pvBuf, cb);
327	pgmPhysReleaseInternalPageMappingLock(pVM, &PgMpLck);
328	}
329	else
330	AssertLogRelMsgFailed(("pgmPhysGCPhys2CCPtrInternal failed on %RGp / %R[pgmpage] -> %Rrc\n",
331	pRam->GCPhys + off, pPage, rc));
332
333	/* next page */
334	if (cb >= cbWrite)
335	{
336	PGM_UNLOCK(pVM);
337	return VINF_SUCCESS;
338	}
339
340	cbWrite -= cb;
341	off += cb;
342	GCPhys += cb;
343	pvBuf = (const char *)pvBuf + cb;
344	} /* walk pages in ram range */
345	}
346	else
347	{
348	/*
349	* Unassigned address space, skip it.
350	*/
351	if (!pRam)
352	break;
353	size_t cb = pRam->GCPhys - GCPhys;
354	if (cb >= cbWrite)
355	break;
356	cbWrite -= cb;
357	pvBuf = (const char *)pvBuf + cb;
358	GCPhys += cb;
359	}
360
361	/* Advance range if necessary. */
362	while (pRam && GCPhys > pRam->GCPhysLast)
363	pRam = pRam->CTX_SUFF(pNext);
364	} /* Ram range walk */
365
366	PGM_UNLOCK(pVM);
367	return VINF_SUCCESS;
368	}
369
370
371	/*********************************************************************************************************************************
372	* Mapping Guest Physical Memory *
373	*********************************************************************************************************************************/
374
375	/**
376	* VMR3ReqCall worker for PGMR3PhysGCPhys2CCPtrExternal to make pages writable.
377	*
378	* @returns see PGMR3PhysGCPhys2CCPtrExternal
379	* @param pVM The cross context VM structure.
380	* @param pGCPhys Pointer to the guest physical address.
381	* @param ppv Where to store the mapping address.
382	* @param pLock Where to store the lock.
383	*/
384	static DECLCALLBACK(int) pgmR3PhysGCPhys2CCPtrDelegated(PVM pVM, PRTGCPHYS pGCPhys, void **ppv, PPGMPAGEMAPLOCK pLock)
385	{
386	/*
387	* Just hand it to PGMPhysGCPhys2CCPtr and check that it's not a page with
388	* an access handler after it succeeds.
389	*/
390	int rc = PGM_LOCK(pVM);
391	AssertRCReturn(rc, rc);
392
393	rc = PGMPhysGCPhys2CCPtr(pVM, *pGCPhys, ppv, pLock);
394	if (RT_SUCCESS(rc))
395	{
396	PPGMPAGEMAPTLBE pTlbe;
397	int rc2 = pgmPhysPageQueryTlbe(pVM, *pGCPhys, &pTlbe);
398	AssertFatalRC(rc2);
399	PPGMPAGE pPage = pTlbe->pPage;
400	if (PGM_PAGE_IS_MMIO_OR_SPECIAL_ALIAS(pPage))
401	{
402	PGMPhysReleasePageMappingLock(pVM, pLock);
403	rc = VERR_PGM_PHYS_PAGE_RESERVED;
404	}
405	else if ( PGM_PAGE_HAS_ACTIVE_HANDLERS(pPage)
406	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
407	\|\| pgmPoolIsDirtyPage(pVM, *pGCPhys)
408	#endif
409	)
410	{
411	/* We must flush any corresponding pgm pool page here, otherwise we'll
412	* not be informed about writes and keep bogus gst->shw mappings around.
413	*/
414	pgmPoolFlushPageByGCPhys(pVM, *pGCPhys);
415	Assert(!PGM_PAGE_HAS_ACTIVE_HANDLERS(pPage));
416	/** @todo r=bird: return VERR_PGM_PHYS_PAGE_RESERVED here if it still has
417	* active handlers, see the PGMR3PhysGCPhys2CCPtrExternal docs. */
418	}
419	}
420
421	PGM_UNLOCK(pVM);
422	return rc;
423	}
424
425
426	/**
427	* Requests the mapping of a guest page into ring-3, external threads.
428	*
429	* When you're done with the page, call PGMPhysReleasePageMappingLock() ASAP to
430	* release it.
431	*
432	* This API will assume your intention is to write to the page, and will
433	* therefore replace shared and zero pages. If you do not intend to modify the
434	* page, use the PGMR3PhysGCPhys2CCPtrReadOnlyExternal() API.
435	*
436	* @returns VBox status code.
437	* @retval VINF_SUCCESS on success.
438	* @retval VERR_PGM_PHYS_PAGE_RESERVED it it's a valid page but has no physical
439	* backing or if the page has any active access handlers. The caller
440	* must fall back on using PGMR3PhysWriteExternal.
441	* @retval VERR_PGM_INVALID_GC_PHYSICAL_ADDRESS if it's not a valid physical address.
442	*
443	* @param pVM The cross context VM structure.
444	* @param GCPhys The guest physical address of the page that should be mapped.
445	* @param ppv Where to store the address corresponding to GCPhys.
446	* @param pLock Where to store the lock information that PGMPhysReleasePageMappingLock needs.
447	*
448	* @remark Avoid calling this API from within critical sections (other than the
449	* PGM one) because of the deadlock risk when we have to delegating the
450	* task to an EMT.
451	* @thread Any.
452	*/
453	VMMR3DECL(int) PGMR3PhysGCPhys2CCPtrExternal(PVM pVM, RTGCPHYS GCPhys, void **ppv, PPGMPAGEMAPLOCK pLock)
454	{
455	AssertPtr(ppv);
456	AssertPtr(pLock);
457
458	Assert(VM_IS_EMT(pVM) \|\| !PGMIsLockOwner(pVM));
459
460	int rc = PGM_LOCK(pVM);
461	AssertRCReturn(rc, rc);
462
463	/*
464	* Query the Physical TLB entry for the page (may fail).
465	*/
466	PPGMPAGEMAPTLBE pTlbe;
467	rc = pgmPhysPageQueryTlbe(pVM, GCPhys, &pTlbe);
468	if (RT_SUCCESS(rc))
469	{
470	PPGMPAGE pPage = pTlbe->pPage;
471	if (PGM_PAGE_IS_MMIO_OR_SPECIAL_ALIAS(pPage))
472	rc = VERR_PGM_PHYS_PAGE_RESERVED;
473	else
474	{
475	/*
476	* If the page is shared, the zero page, or being write monitored
477	* it must be converted to an page that's writable if possible.
478	* We can only deal with write monitored pages here, the rest have
479	* to be on an EMT.
480	*/
481	if ( PGM_PAGE_HAS_ACTIVE_HANDLERS(pPage)
482	\|\| PGM_PAGE_GET_STATE(pPage) != PGM_PAGE_STATE_ALLOCATED
483	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
484	\|\| pgmPoolIsDirtyPage(pVM, GCPhys)
485	#endif
486	)
487	{
488	if ( PGM_PAGE_GET_STATE(pPage) == PGM_PAGE_STATE_WRITE_MONITORED
489	&& !PGM_PAGE_HAS_ACTIVE_HANDLERS(pPage)
490	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
491	&& !pgmPoolIsDirtyPage(pVM, GCPhys) /** @todo we're very likely doing this twice. */
492	#endif
493	)
494	pgmPhysPageMakeWriteMonitoredWritable(pVM, pPage, GCPhys);
495	else
496	{
497	PGM_UNLOCK(pVM);
498
499	return VMR3ReqPriorityCallWait(pVM, VMCPUID_ANY, (PFNRT)pgmR3PhysGCPhys2CCPtrDelegated, 4,
500	pVM, &GCPhys, ppv, pLock);
501	}
502	}
503
504	/*
505	* Now, just perform the locking and calculate the return address.
506	*/
507	PPGMPAGEMAP pMap = pTlbe->pMap;
508	if (pMap)
509	pMap->cRefs++;
510
511	unsigned cLocks = PGM_PAGE_GET_WRITE_LOCKS(pPage);
512	if (RT_LIKELY(cLocks < PGM_PAGE_MAX_LOCKS - 1))
513	{
514	if (cLocks == 0)
515	pVM->pgm.s.cWriteLockedPages++;
516	PGM_PAGE_INC_WRITE_LOCKS(pPage);
517	}
518	else if (cLocks != PGM_PAGE_GET_WRITE_LOCKS(pPage))
519	{
520	PGM_PAGE_INC_WRITE_LOCKS(pPage);
521	AssertMsgFailed(("%RGp / %R[pgmpage] is entering permanent write locked state!\n", GCPhys, pPage));
522	if (pMap)
523	pMap->cRefs++; /* Extra ref to prevent it from going away. */
524	}
525
526	ppv = (void )((uintptr_t)pTlbe->pv \| (uintptr_t)(GCPhys & GUEST_PAGE_OFFSET_MASK));
527	pLock->uPageAndType = (uintptr_t)pPage \| PGMPAGEMAPLOCK_TYPE_WRITE;
528	pLock->pvMap = pMap;
529	}
530	}
531
532	PGM_UNLOCK(pVM);
533	return rc;
534	}
535
536
537	/**
538	* Requests the mapping of a guest page into ring-3, external threads.
539	*
540	* When you're done with the page, call PGMPhysReleasePageMappingLock() ASAP to
541	* release it.
542	*
543	* @returns VBox status code.
544	* @retval VINF_SUCCESS on success.
545	* @retval VERR_PGM_PHYS_PAGE_RESERVED it it's a valid page but has no physical
546	* backing or if the page as an active ALL access handler. The caller
547	* must fall back on using PGMPhysRead.
548	* @retval VERR_PGM_INVALID_GC_PHYSICAL_ADDRESS if it's not a valid physical address.
549	*
550	* @param pVM The cross context VM structure.
551	* @param GCPhys The guest physical address of the page that should be mapped.
552	* @param ppv Where to store the address corresponding to GCPhys.
553	* @param pLock Where to store the lock information that PGMPhysReleasePageMappingLock needs.
554	*
555	* @remark Avoid calling this API from within critical sections (other than
556	* the PGM one) because of the deadlock risk.
557	* @thread Any.
558	*/
559	VMMR3DECL(int) PGMR3PhysGCPhys2CCPtrReadOnlyExternal(PVM pVM, RTGCPHYS GCPhys, void const **ppv, PPGMPAGEMAPLOCK pLock)
560	{
561	int rc = PGM_LOCK(pVM);
562	AssertRCReturn(rc, rc);
563
564	/*
565	* Query the Physical TLB entry for the page (may fail).
566	*/
567	PPGMPAGEMAPTLBE pTlbe;
568	rc = pgmPhysPageQueryTlbe(pVM, GCPhys, &pTlbe);
569	if (RT_SUCCESS(rc))
570	{
571	PPGMPAGE pPage = pTlbe->pPage;
572	#if 1
573	/* MMIO pages doesn't have any readable backing. */
574	if (PGM_PAGE_IS_MMIO_OR_SPECIAL_ALIAS(pPage))
575	rc = VERR_PGM_PHYS_PAGE_RESERVED;
576	#else
577	if (PGM_PAGE_HAS_ACTIVE_ALL_HANDLERS(pPage))
578	rc = VERR_PGM_PHYS_PAGE_RESERVED;
579	#endif
580	else
581	{
582	/*
583	* Now, just perform the locking and calculate the return address.
584	*/
585	PPGMPAGEMAP pMap = pTlbe->pMap;
586	if (pMap)
587	pMap->cRefs++;
588
589	unsigned cLocks = PGM_PAGE_GET_READ_LOCKS(pPage);
590	if (RT_LIKELY(cLocks < PGM_PAGE_MAX_LOCKS - 1))
591	{
592	if (cLocks == 0)
593	pVM->pgm.s.cReadLockedPages++;
594	PGM_PAGE_INC_READ_LOCKS(pPage);
595	}
596	else if (cLocks != PGM_PAGE_GET_READ_LOCKS(pPage))
597	{
598	PGM_PAGE_INC_READ_LOCKS(pPage);
599	AssertMsgFailed(("%RGp / %R[pgmpage] is entering permanent readonly locked state!\n", GCPhys, pPage));
600	if (pMap)
601	pMap->cRefs++; /* Extra ref to prevent it from going away. */
602	}
603
604	ppv = (void )((uintptr_t)pTlbe->pv \| (uintptr_t)(GCPhys & GUEST_PAGE_OFFSET_MASK));
605	pLock->uPageAndType = (uintptr_t)pPage \| PGMPAGEMAPLOCK_TYPE_READ;
606	pLock->pvMap = pMap;
607	}
608	}
609
610	PGM_UNLOCK(pVM);
611	return rc;
612	}
613
614
615	/**
616	* Requests the mapping of multiple guest page into ring-3, external threads.
617	*
618	* When you're done with the pages, call PGMPhysBulkReleasePageMappingLock()
619	* ASAP to release them.
620	*
621	* This API will assume your intention is to write to the pages, and will
622	* therefore replace shared and zero pages. If you do not intend to modify the
623	* pages, use the PGMR3PhysBulkGCPhys2CCPtrReadOnlyExternal() API.
624	*
625	* @returns VBox status code.
626	* @retval VINF_SUCCESS on success.
627	* @retval VERR_PGM_PHYS_PAGE_RESERVED if any of the pages has no physical
628	* backing or if any of the pages the page has any active access
629	* handlers. The caller must fall back on using PGMR3PhysWriteExternal.
630	* @retval VERR_PGM_INVALID_GC_PHYSICAL_ADDRESS if @a paGCPhysPages contains
631	* an invalid physical address.
632	*
633	* @param pVM The cross context VM structure.
634	* @param cPages Number of pages to lock.
635	* @param paGCPhysPages The guest physical address of the pages that
636	* should be mapped (@a cPages entries).
637	* @param papvPages Where to store the ring-3 mapping addresses
638	* corresponding to @a paGCPhysPages.
639	* @param paLocks Where to store the locking information that
640	* pfnPhysBulkReleasePageMappingLock needs (@a cPages
641	* in length).
642	*
643	* @remark Avoid calling this API from within critical sections (other than the
644	* PGM one) because of the deadlock risk when we have to delegating the
645	* task to an EMT.
646	* @thread Any.
647	*/
648	VMMR3DECL(int) PGMR3PhysBulkGCPhys2CCPtrExternal(PVM pVM, uint32_t cPages, PCRTGCPHYS paGCPhysPages,
649	void **papvPages, PPGMPAGEMAPLOCK paLocks)
650	{
651	Assert(cPages > 0);
652	AssertPtr(papvPages);
653	AssertPtr(paLocks);
654
655	Assert(VM_IS_EMT(pVM) \|\| !PGMIsLockOwner(pVM));
656
657	int rc = PGM_LOCK(pVM);
658	AssertRCReturn(rc, rc);
659
660	/*
661	* Lock the pages one by one.
662	* The loop body is similar to PGMR3PhysGCPhys2CCPtrExternal.
663	*/
664	int32_t cNextYield = 128;
665	uint32_t iPage;
666	for (iPage = 0; iPage < cPages; iPage++)
667	{
668	if (--cNextYield > 0)
669	{ /* likely */ }
670	else
671	{
672	PGM_UNLOCK(pVM);
673	ASMNopPause();
674	PGM_LOCK_VOID(pVM);
675	cNextYield = 128;
676	}
677
678	/*
679	* Query the Physical TLB entry for the page (may fail).
680	*/
681	PPGMPAGEMAPTLBE pTlbe;
682	rc = pgmPhysPageQueryTlbe(pVM, paGCPhysPages[iPage], &pTlbe);
683	if (RT_SUCCESS(rc))
684	{ }
685	else
686	break;
687	PPGMPAGE pPage = pTlbe->pPage;
688
689	/*
690	* No MMIO or active access handlers.
691	*/
692	if ( !PGM_PAGE_IS_MMIO_OR_SPECIAL_ALIAS(pPage)
693	&& !PGM_PAGE_HAS_ACTIVE_HANDLERS(pPage))
694	{ }
695	else
696	{
697	rc = VERR_PGM_PHYS_PAGE_RESERVED;
698	break;
699	}
700
701	/*
702	* The page must be in the allocated state and not be a dirty pool page.
703	* We can handle converting a write monitored page to an allocated one, but
704	* anything more complicated must be delegated to an EMT.
705	*/
706	bool fDelegateToEmt = false;
707	if (PGM_PAGE_GET_STATE(pPage) == PGM_PAGE_STATE_ALLOCATED)
708	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
709	fDelegateToEmt = pgmPoolIsDirtyPage(pVM, paGCPhysPages[iPage]);
710	#else
711	fDelegateToEmt = false;
712	#endif
713	else if (PGM_PAGE_GET_STATE(pPage) == PGM_PAGE_STATE_WRITE_MONITORED)
714	{
715	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
716	if (!pgmPoolIsDirtyPage(pVM, paGCPhysPages[iPage]))
717	pgmPhysPageMakeWriteMonitoredWritable(pVM, pPage, paGCPhysPages[iPage]);
718	else
719	fDelegateToEmt = true;
720	#endif
721	}
722	else
723	fDelegateToEmt = true;
724	if (!fDelegateToEmt)
725	{ }
726	else
727	{
728	/* We could do this delegation in bulk, but considered too much work vs gain. */
729	PGM_UNLOCK(pVM);
730	rc = VMR3ReqPriorityCallWait(pVM, VMCPUID_ANY, (PFNRT)pgmR3PhysGCPhys2CCPtrDelegated, 4,
731	pVM, &paGCPhysPages[iPage], &papvPages[iPage], &paLocks[iPage]);
732	PGM_LOCK_VOID(pVM);
733	if (RT_FAILURE(rc))
734	break;
735	cNextYield = 128;
736	}
737
738	/*
739	* Now, just perform the locking and address calculation.
740	*/
741	PPGMPAGEMAP pMap = pTlbe->pMap;
742	if (pMap)
743	pMap->cRefs++;
744
745	unsigned cLocks = PGM_PAGE_GET_WRITE_LOCKS(pPage);
746	if (RT_LIKELY(cLocks < PGM_PAGE_MAX_LOCKS - 1))
747	{
748	if (cLocks == 0)
749	pVM->pgm.s.cWriteLockedPages++;
750	PGM_PAGE_INC_WRITE_LOCKS(pPage);
751	}
752	else if (cLocks != PGM_PAGE_GET_WRITE_LOCKS(pPage))
753	{
754	PGM_PAGE_INC_WRITE_LOCKS(pPage);
755	AssertMsgFailed(("%RGp / %R[pgmpage] is entering permanent write locked state!\n", paGCPhysPages[iPage], pPage));
756	if (pMap)
757	pMap->cRefs++; /* Extra ref to prevent it from going away. */
758	}
759
760	papvPages[iPage] = (void *)((uintptr_t)pTlbe->pv \| (uintptr_t)(paGCPhysPages[iPage] & GUEST_PAGE_OFFSET_MASK));
761	paLocks[iPage].uPageAndType = (uintptr_t)pPage \| PGMPAGEMAPLOCK_TYPE_WRITE;
762	paLocks[iPage].pvMap = pMap;
763	}
764
765	PGM_UNLOCK(pVM);
766
767	/*
768	* On failure we must unlock any pages we managed to get already.
769	*/
770	if (RT_FAILURE(rc) && iPage > 0)
771	PGMPhysBulkReleasePageMappingLocks(pVM, iPage, paLocks);
772
773	return rc;
774	}
775
776
777	/**
778	* Requests the mapping of multiple guest page into ring-3, for reading only,
779	* external threads.
780	*
781	* When you're done with the pages, call PGMPhysReleasePageMappingLock() ASAP
782	* to release them.
783	*
784	* @returns VBox status code.
785	* @retval VINF_SUCCESS on success.
786	* @retval VERR_PGM_PHYS_PAGE_RESERVED if any of the pages has no physical
787	* backing or if any of the pages the page has an active ALL access
788	* handler. The caller must fall back on using PGMR3PhysWriteExternal.
789	* @retval VERR_PGM_INVALID_GC_PHYSICAL_ADDRESS if @a paGCPhysPages contains
790	* an invalid physical address.
791	*
792	* @param pVM The cross context VM structure.
793	* @param cPages Number of pages to lock.
794	* @param paGCPhysPages The guest physical address of the pages that
795	* should be mapped (@a cPages entries).
796	* @param papvPages Where to store the ring-3 mapping addresses
797	* corresponding to @a paGCPhysPages.
798	* @param paLocks Where to store the lock information that
799	* pfnPhysReleasePageMappingLock needs (@a cPages
800	* in length).
801	*
802	* @remark Avoid calling this API from within critical sections (other than
803	* the PGM one) because of the deadlock risk.
804	* @thread Any.
805	*/
806	VMMR3DECL(int) PGMR3PhysBulkGCPhys2CCPtrReadOnlyExternal(PVM pVM, uint32_t cPages, PCRTGCPHYS paGCPhysPages,
807	void const **papvPages, PPGMPAGEMAPLOCK paLocks)
808	{
809	Assert(cPages > 0);
810	AssertPtr(papvPages);
811	AssertPtr(paLocks);
812
813	Assert(VM_IS_EMT(pVM) \|\| !PGMIsLockOwner(pVM));
814
815	int rc = PGM_LOCK(pVM);
816	AssertRCReturn(rc, rc);
817
818	/*
819	* Lock the pages one by one.
820	* The loop body is similar to PGMR3PhysGCPhys2CCPtrReadOnlyExternal.
821	*/
822	int32_t cNextYield = 256;
823	uint32_t iPage;
824	for (iPage = 0; iPage < cPages; iPage++)
825	{
826	if (--cNextYield > 0)
827	{ /* likely */ }
828	else
829	{
830	PGM_UNLOCK(pVM);
831	ASMNopPause();
832	PGM_LOCK_VOID(pVM);
833	cNextYield = 256;
834	}
835
836	/*
837	* Query the Physical TLB entry for the page (may fail).
838	*/
839	PPGMPAGEMAPTLBE pTlbe;
840	rc = pgmPhysPageQueryTlbe(pVM, paGCPhysPages[iPage], &pTlbe);
841	if (RT_SUCCESS(rc))
842	{ }
843	else
844	break;
845	PPGMPAGE pPage = pTlbe->pPage;
846
847	/*
848	* No MMIO or active all access handlers, everything else can be accessed.
849	*/
850	if ( !PGM_PAGE_IS_MMIO_OR_SPECIAL_ALIAS(pPage)
851	&& !PGM_PAGE_HAS_ACTIVE_ALL_HANDLERS(pPage))
852	{ }
853	else
854	{
855	rc = VERR_PGM_PHYS_PAGE_RESERVED;
856	break;
857	}
858
859	/*
860	* Now, just perform the locking and address calculation.
861	*/
862	PPGMPAGEMAP pMap = pTlbe->pMap;
863	if (pMap)
864	pMap->cRefs++;
865
866	unsigned cLocks = PGM_PAGE_GET_READ_LOCKS(pPage);
867	if (RT_LIKELY(cLocks < PGM_PAGE_MAX_LOCKS - 1))
868	{
869	if (cLocks == 0)
870	pVM->pgm.s.cReadLockedPages++;
871	PGM_PAGE_INC_READ_LOCKS(pPage);
872	}
873	else if (cLocks != PGM_PAGE_GET_READ_LOCKS(pPage))
874	{
875	PGM_PAGE_INC_READ_LOCKS(pPage);
876	AssertMsgFailed(("%RGp / %R[pgmpage] is entering permanent readonly locked state!\n", paGCPhysPages[iPage], pPage));
877	if (pMap)
878	pMap->cRefs++; /* Extra ref to prevent it from going away. */
879	}
880
881	papvPages[iPage] = (void *)((uintptr_t)pTlbe->pv \| (uintptr_t)(paGCPhysPages[iPage] & GUEST_PAGE_OFFSET_MASK));
882	paLocks[iPage].uPageAndType = (uintptr_t)pPage \| PGMPAGEMAPLOCK_TYPE_READ;
883	paLocks[iPage].pvMap = pMap;
884	}
885
886	PGM_UNLOCK(pVM);
887
888	/*
889	* On failure we must unlock any pages we managed to get already.
890	*/
891	if (RT_FAILURE(rc) && iPage > 0)
892	PGMPhysBulkReleasePageMappingLocks(pVM, iPage, paLocks);
893
894	return rc;
895	}
896
897
898	/**
899	* Converts a GC physical address to a HC ring-3 pointer, with some
900	* additional checks.
901	*
902	* @returns VBox status code.
903	* @retval VINF_SUCCESS on success.
904	* @retval VINF_PGM_PHYS_TLB_CATCH_WRITE and *ppv set if the page has a write
905	* access handler of some kind.
906	* @retval VERR_PGM_PHYS_TLB_CATCH_ALL if the page has a handler catching all
907	* accesses or is odd in any way.
908	* @retval VERR_PGM_PHYS_TLB_UNASSIGNED if the page doesn't exist.
909	*
910	* @param pVM The cross context VM structure.
911	* @param GCPhys The GC physical address to convert. Since this is only
912	* used for filling the REM TLB, the A20 mask must be
913	* applied before calling this API.
914	* @param fWritable Whether write access is required.
915	* @param ppv Where to store the pointer corresponding to GCPhys on
916	* success.
917	*/
918	VMMR3DECL(int) PGMR3PhysTlbGCPhys2Ptr(PVM pVM, RTGCPHYS GCPhys, bool fWritable, void **ppv)
919	{
920	PGM_LOCK_VOID(pVM);
921	PGM_A20_ASSERT_MASKED(VMMGetCpu(pVM), GCPhys);
922
923	PPGMRAMRANGE pRam;
924	PPGMPAGE pPage;
925	int rc = pgmPhysGetPageAndRangeEx(pVM, GCPhys, &pPage, &pRam);
926	if (RT_SUCCESS(rc))
927	{
928	if (PGM_PAGE_IS_BALLOONED(pPage))
929	rc = VINF_PGM_PHYS_TLB_CATCH_WRITE;
930	else if (!PGM_PAGE_HAS_ANY_HANDLERS(pPage))
931	rc = VINF_SUCCESS;
932	else
933	{
934	if (PGM_PAGE_HAS_ACTIVE_ALL_HANDLERS(pPage)) /* catches MMIO */
935	rc = VERR_PGM_PHYS_TLB_CATCH_ALL;
936	else if (PGM_PAGE_HAS_ACTIVE_HANDLERS(pPage))
937	{
938	/** @todo Handle TLB loads of virtual handlers so ./test.sh can be made to work
939	* in -norawr0 mode. */
940	if (fWritable)
941	rc = VINF_PGM_PHYS_TLB_CATCH_WRITE;
942	}
943	else
944	{
945	/* Temporarily disabled physical handler(s), since the recompiler
946	doesn't get notified when it's reset we'll have to pretend it's
947	operating normally. */
948	if (pgmHandlerPhysicalIsAll(pVM, GCPhys))
949	rc = VERR_PGM_PHYS_TLB_CATCH_ALL;
950	else
951	rc = VINF_PGM_PHYS_TLB_CATCH_WRITE;
952	}
953	}
954	if (RT_SUCCESS(rc))
955	{
956	int rc2;
957
958	/* Make sure what we return is writable. */
959	if (fWritable)
960	switch (PGM_PAGE_GET_STATE(pPage))
961	{
962	case PGM_PAGE_STATE_ALLOCATED:
963	break;
964	case PGM_PAGE_STATE_BALLOONED:
965	AssertFailed();
966	break;
967	case PGM_PAGE_STATE_ZERO:
968	case PGM_PAGE_STATE_SHARED:
969	if (rc == VINF_PGM_PHYS_TLB_CATCH_WRITE)
970	break;
971	RT_FALL_THRU();
972	case PGM_PAGE_STATE_WRITE_MONITORED:
973	rc2 = pgmPhysPageMakeWritable(pVM, pPage, GCPhys & ~(RTGCPHYS)GUEST_PAGE_OFFSET_MASK);
974	AssertLogRelRCReturn(rc2, rc2);
975	break;
976	}
977
978	/* Get a ring-3 mapping of the address. */
979	PPGMPAGER3MAPTLBE pTlbe;
980	rc2 = pgmPhysPageQueryTlbe(pVM, GCPhys, &pTlbe);
981	AssertLogRelRCReturn(rc2, rc2);
982	ppv = (void )((uintptr_t)pTlbe->pv \| (uintptr_t)(GCPhys & GUEST_PAGE_OFFSET_MASK));
983	/** @todo mapping/locking hell; this isn't horribly efficient since
984	* pgmPhysPageLoadIntoTlb will repeat the lookup we've done here. */
985
986	Log6(("PGMR3PhysTlbGCPhys2Ptr: GCPhys=%RGp rc=%Rrc pPage=%R[pgmpage] ppv=%p\n", GCPhys, rc, pPage, ppv));
987	}
988	else
989	Log6(("PGMR3PhysTlbGCPhys2Ptr: GCPhys=%RGp rc=%Rrc pPage=%R[pgmpage]\n", GCPhys, rc, pPage));
990
991	/* else: handler catching all access, no pointer returned. */
992	}
993	else
994	rc = VERR_PGM_PHYS_TLB_UNASSIGNED;
995
996	PGM_UNLOCK(pVM);
997	return rc;
998	}
999
1000
1001
1002	/*********************************************************************************************************************************
1003	* RAM Range Management *
1004	*********************************************************************************************************************************/
1005
1006	#define MAKE_LEAF(a_pNode) \
1007	do { \
1008	(a_pNode)->pLeftR3 = NIL_RTR3PTR; \
1009	(a_pNode)->pRightR3 = NIL_RTR3PTR; \
1010	(a_pNode)->pLeftR0 = NIL_RTR0PTR; \
1011	(a_pNode)->pRightR0 = NIL_RTR0PTR; \
1012	} while (0)
1013
1014	#define INSERT_LEFT(a_pParent, a_pNode) \
1015	do { \
1016	(a_pParent)->pLeftR3 = (a_pNode); \
1017	(a_pParent)->pLeftR0 = (a_pNode)->pSelfR0; \
1018	} while (0)
1019	#define INSERT_RIGHT(a_pParent, a_pNode) \
1020	do { \
1021	(a_pParent)->pRightR3 = (a_pNode); \
1022	(a_pParent)->pRightR0 = (a_pNode)->pSelfR0; \
1023	} while (0)
1024
1025
1026	/**
1027	* Recursive tree builder.
1028	*
1029	* @param ppRam Pointer to the iterator variable.
1030	* @param iDepth The current depth. Inserts a leaf node if 0.
1031	*/
1032	static PPGMRAMRANGE pgmR3PhysRebuildRamRangeSearchTreesRecursively(PPGMRAMRANGE *ppRam, int iDepth)
1033	{
1034	PPGMRAMRANGE pRam;
1035	if (iDepth <= 0)
1036	{
1037	/*
1038	* Leaf node.
1039	*/
1040	pRam = *ppRam;
1041	if (pRam)
1042	{
1043	*ppRam = pRam->pNextR3;
1044	MAKE_LEAF(pRam);
1045	}
1046	}
1047	else
1048	{
1049
1050	/*
1051	* Intermediate node.
1052	*/
1053	PPGMRAMRANGE pLeft = pgmR3PhysRebuildRamRangeSearchTreesRecursively(ppRam, iDepth - 1);
1054
1055	pRam = *ppRam;
1056	if (!pRam)
1057	return pLeft;
1058	*ppRam = pRam->pNextR3;
1059	MAKE_LEAF(pRam);
1060	INSERT_LEFT(pRam, pLeft);
1061
1062	PPGMRAMRANGE pRight = pgmR3PhysRebuildRamRangeSearchTreesRecursively(ppRam, iDepth - 1);
1063	if (pRight)
1064	INSERT_RIGHT(pRam, pRight);
1065	}
1066	return pRam;
1067	}
1068
1069
1070	/**
1071	* Rebuilds the RAM range search trees.
1072	*
1073	* @param pVM The cross context VM structure.
1074	*/
1075	static void pgmR3PhysRebuildRamRangeSearchTrees(PVM pVM)
1076	{
1077
1078	/*
1079	* Create the reasonably balanced tree in a sequential fashion.
1080	* For simplicity (laziness) we use standard recursion here.
1081	*/
1082	int iDepth = 0;
1083	PPGMRAMRANGE pRam = pVM->pgm.s.pRamRangesXR3;
1084	PPGMRAMRANGE pRoot = pgmR3PhysRebuildRamRangeSearchTreesRecursively(&pRam, 0);
1085	while (pRam)
1086	{
1087	PPGMRAMRANGE pLeft = pRoot;
1088
1089	pRoot = pRam;
1090	pRam = pRam->pNextR3;
1091	MAKE_LEAF(pRoot);
1092	INSERT_LEFT(pRoot, pLeft);
1093
1094	PPGMRAMRANGE pRight = pgmR3PhysRebuildRamRangeSearchTreesRecursively(&pRam, iDepth);
1095	if (pRight)
1096	INSERT_RIGHT(pRoot, pRight);
1097	/** @todo else: rotate the tree. */
1098
1099	iDepth++;
1100	}
1101
1102	pVM->pgm.s.pRamRangeTreeR3 = pRoot;
1103	pVM->pgm.s.pRamRangeTreeR0 = pRoot ? pRoot->pSelfR0 : NIL_RTR0PTR;
1104
1105	#ifdef VBOX_STRICT
1106	/*
1107	* Verify that the above code works.
1108	*/
1109	unsigned cRanges = 0;
1110	for (pRam = pVM->pgm.s.pRamRangesXR3; pRam; pRam = pRam->pNextR3)
1111	cRanges++;
1112	Assert(cRanges > 0);
1113
1114	unsigned cMaxDepth = ASMBitLastSetU32(cRanges);
1115	if ((1U << cMaxDepth) < cRanges)
1116	cMaxDepth++;
1117
1118	for (pRam = pVM->pgm.s.pRamRangesXR3; pRam; pRam = pRam->pNextR3)
1119	{
1120	unsigned cDepth = 0;
1121	PPGMRAMRANGE pRam2 = pVM->pgm.s.pRamRangeTreeR3;
1122	for (;;)
1123	{
1124	if (pRam == pRam2)
1125	break;
1126	Assert(pRam2);
1127	if (pRam->GCPhys < pRam2->GCPhys)
1128	pRam2 = pRam2->pLeftR3;
1129	else
1130	pRam2 = pRam2->pRightR3;
1131	}
1132	AssertMsg(cDepth <= cMaxDepth, ("cDepth=%d cMaxDepth=%d\n", cDepth, cMaxDepth));
1133	}
1134	#endif /* VBOX_STRICT */
1135	}
1136
1137	#undef MAKE_LEAF
1138	#undef INSERT_LEFT
1139	#undef INSERT_RIGHT
1140
1141	/**
1142	* Relinks the RAM ranges using the pSelfRC and pSelfR0 pointers.
1143	*
1144	* Called when anything was relocated.
1145	*
1146	* @param pVM The cross context VM structure.
1147	*/
1148	void pgmR3PhysRelinkRamRanges(PVM pVM)
1149	{
1150	PPGMRAMRANGE pCur;
1151
1152	#ifdef VBOX_STRICT
1153	for (pCur = pVM->pgm.s.pRamRangesXR3; pCur; pCur = pCur->pNextR3)
1154	{
1155	Assert((pCur->GCPhys & GUEST_PAGE_OFFSET_MASK) == 0);
1156	Assert((pCur->GCPhysLast & GUEST_PAGE_OFFSET_MASK) == GUEST_PAGE_OFFSET_MASK);
1157	Assert((pCur->cb & GUEST_PAGE_OFFSET_MASK) == 0);
1158	Assert(pCur->cb == pCur->GCPhysLast - pCur->GCPhys + 1);
1159	for (PPGMRAMRANGE pCur2 = pVM->pgm.s.pRamRangesXR3; pCur2; pCur2 = pCur2->pNextR3)
1160	Assert( pCur2 == pCur
1161	\|\| strcmp(pCur2->pszDesc, pCur->pszDesc)); /** @todo fix MMIO ranges!! */
1162	}
1163	#endif
1164
1165	pCur = pVM->pgm.s.pRamRangesXR3;
1166	if (pCur)
1167	{
1168	pVM->pgm.s.pRamRangesXR0 = pCur->pSelfR0;
1169
1170	for (; pCur->pNextR3; pCur = pCur->pNextR3)
1171	pCur->pNextR0 = pCur->pNextR3->pSelfR0;
1172
1173	Assert(pCur->pNextR0 == NIL_RTR0PTR);
1174	}
1175	else
1176	{
1177	Assert(pVM->pgm.s.pRamRangesXR0 == NIL_RTR0PTR);
1178	}
1179	ASMAtomicIncU32(&pVM->pgm.s.idRamRangesGen);
1180
1181	pgmR3PhysRebuildRamRangeSearchTrees(pVM);
1182	}
1183
1184
1185	/**
1186	* Links a new RAM range into the list.
1187	*
1188	* @param pVM The cross context VM structure.
1189	* @param pNew Pointer to the new list entry.
1190	* @param pPrev Pointer to the previous list entry. If NULL, insert as head.
1191	*/
1192	static void pgmR3PhysLinkRamRange(PVM pVM, PPGMRAMRANGE pNew, PPGMRAMRANGE pPrev)
1193	{
1194	AssertMsg(pNew->pszDesc, ("%RGp-%RGp\n", pNew->GCPhys, pNew->GCPhysLast));
1195
1196	PGM_LOCK_VOID(pVM);
1197
1198	PPGMRAMRANGE pRam = pPrev ? pPrev->pNextR3 : pVM->pgm.s.pRamRangesXR3;
1199	pNew->pNextR3 = pRam;
1200	pNew->pNextR0 = pRam ? pRam->pSelfR0 : NIL_RTR0PTR;
1201
1202	if (pPrev)
1203	{
1204	pPrev->pNextR3 = pNew;
1205	pPrev->pNextR0 = pNew->pSelfR0;
1206	}
1207	else
1208	{
1209	pVM->pgm.s.pRamRangesXR3 = pNew;
1210	pVM->pgm.s.pRamRangesXR0 = pNew->pSelfR0;
1211	}
1212	ASMAtomicIncU32(&pVM->pgm.s.idRamRangesGen);
1213
1214	pgmR3PhysRebuildRamRangeSearchTrees(pVM);
1215	PGM_UNLOCK(pVM);
1216	}
1217
1218
1219	/**
1220	* Unlink an existing RAM range from the list.
1221	*
1222	* @param pVM The cross context VM structure.
1223	* @param pRam Pointer to the new list entry.
1224	* @param pPrev Pointer to the previous list entry. If NULL, insert as head.
1225	*/
1226	static void pgmR3PhysUnlinkRamRange2(PVM pVM, PPGMRAMRANGE pRam, PPGMRAMRANGE pPrev)
1227	{
1228	Assert(pPrev ? pPrev->pNextR3 == pRam : pVM->pgm.s.pRamRangesXR3 == pRam);
1229
1230	PGM_LOCK_VOID(pVM);
1231
1232	PPGMRAMRANGE pNext = pRam->pNextR3;
1233	if (pPrev)
1234	{
1235	pPrev->pNextR3 = pNext;
1236	pPrev->pNextR0 = pNext ? pNext->pSelfR0 : NIL_RTR0PTR;
1237	}
1238	else
1239	{
1240	Assert(pVM->pgm.s.pRamRangesXR3 == pRam);
1241	pVM->pgm.s.pRamRangesXR3 = pNext;
1242	pVM->pgm.s.pRamRangesXR0 = pNext ? pNext->pSelfR0 : NIL_RTR0PTR;
1243	}
1244	ASMAtomicIncU32(&pVM->pgm.s.idRamRangesGen);
1245
1246	pgmR3PhysRebuildRamRangeSearchTrees(pVM);
1247	PGM_UNLOCK(pVM);
1248	}
1249
1250
1251	/**
1252	* Unlink an existing RAM range from the list.
1253	*
1254	* @param pVM The cross context VM structure.
1255	* @param pRam Pointer to the new list entry.
1256	*/
1257	static void pgmR3PhysUnlinkRamRange(PVM pVM, PPGMRAMRANGE pRam)
1258	{
1259	PGM_LOCK_VOID(pVM);
1260
1261	/* find prev. */
1262	PPGMRAMRANGE pPrev = NULL;
1263	PPGMRAMRANGE pCur = pVM->pgm.s.pRamRangesXR3;
1264	while (pCur != pRam)
1265	{
1266	pPrev = pCur;
1267	pCur = pCur->pNextR3;
1268	}
1269	AssertFatal(pCur);
1270
1271	pgmR3PhysUnlinkRamRange2(pVM, pRam, pPrev);
1272	PGM_UNLOCK(pVM);
1273	}
1274
1275
1276	/**
1277	* Gets the number of ram ranges.
1278	*
1279	* @returns Number of ram ranges. Returns UINT32_MAX if @a pVM is invalid.
1280	* @param pVM The cross context VM structure.
1281	*/
1282	VMMR3DECL(uint32_t) PGMR3PhysGetRamRangeCount(PVM pVM)
1283	{
1284	VM_ASSERT_VALID_EXT_RETURN(pVM, UINT32_MAX);
1285
1286	PGM_LOCK_VOID(pVM);
1287	uint32_t cRamRanges = 0;
1288	for (PPGMRAMRANGE pCur = pVM->pgm.s.CTX_SUFF(pRamRangesX); pCur; pCur = pCur->CTX_SUFF(pNext))
1289	cRamRanges++;
1290	PGM_UNLOCK(pVM);
1291	return cRamRanges;
1292	}
1293
1294
1295	/**
1296	* Get information about a range.
1297	*
1298	* @returns VINF_SUCCESS or VERR_OUT_OF_RANGE.
1299	* @param pVM The cross context VM structure.
1300	* @param iRange The ordinal of the range.
1301	* @param pGCPhysStart Where to return the start of the range. Optional.
1302	* @param pGCPhysLast Where to return the address of the last byte in the
1303	* range. Optional.
1304	* @param ppszDesc Where to return the range description. Optional.
1305	* @param pfIsMmio Where to indicate that this is a pure MMIO range.
1306	* Optional.
1307	*/
1308	VMMR3DECL(int) PGMR3PhysGetRange(PVM pVM, uint32_t iRange, PRTGCPHYS pGCPhysStart, PRTGCPHYS pGCPhysLast,
1309	const char *ppszDesc, bool pfIsMmio)
1310	{
1311	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
1312
1313	PGM_LOCK_VOID(pVM);
1314	uint32_t iCurRange = 0;
1315	for (PPGMRAMRANGE pCur = pVM->pgm.s.CTX_SUFF(pRamRangesX); pCur; pCur = pCur->CTX_SUFF(pNext), iCurRange++)
1316	if (iCurRange == iRange)
1317	{
1318	if (pGCPhysStart)
1319	*pGCPhysStart = pCur->GCPhys;
1320	if (pGCPhysLast)
1321	*pGCPhysLast = pCur->GCPhysLast;
1322	if (ppszDesc)
1323	*ppszDesc = pCur->pszDesc;
1324	if (pfIsMmio)
1325	*pfIsMmio = !!(pCur->fFlags & PGM_RAM_RANGE_FLAGS_AD_HOC_MMIO);
1326
1327	PGM_UNLOCK(pVM);
1328	return VINF_SUCCESS;
1329	}
1330	PGM_UNLOCK(pVM);
1331	return VERR_OUT_OF_RANGE;
1332	}
1333
1334
1335	/*********************************************************************************************************************************
1336	* RAM *
1337	*********************************************************************************************************************************/
1338
1339	/**
1340	* Frees the specified RAM page and replaces it with the ZERO page.
1341	*
1342	* This is used by ballooning, remapping MMIO2, RAM reset and state loading.
1343	*
1344	* @param pVM The cross context VM structure.
1345	* @param pReq Pointer to the request. This is NULL when doing a
1346	* bulk free in NEM memory mode.
1347	* @param pcPendingPages Where the number of pages waiting to be freed are
1348	* kept. This will normally be incremented. This is
1349	* NULL when doing a bulk free in NEM memory mode.
1350	* @param pPage Pointer to the page structure.
1351	* @param GCPhys The guest physical address of the page, if applicable.
1352	* @param enmNewType New page type for NEM notification, since several
1353	* callers will change the type upon successful return.
1354	*
1355	* @remarks The caller must own the PGM lock.
1356	*/
1357	int pgmPhysFreePage(PVM pVM, PGMMFREEPAGESREQ pReq, uint32_t *pcPendingPages, PPGMPAGE pPage, RTGCPHYS GCPhys,
1358	PGMPAGETYPE enmNewType)
1359	{
1360	/*
1361	* Assert sanity.
1362	*/
1363	PGM_LOCK_ASSERT_OWNER(pVM);
1364	if (RT_UNLIKELY( PGM_PAGE_GET_TYPE(pPage) != PGMPAGETYPE_RAM
1365	&& PGM_PAGE_GET_TYPE(pPage) != PGMPAGETYPE_ROM_SHADOW))
1366	{
1367	AssertMsgFailed(("GCPhys=%RGp pPage=%R[pgmpage]\n", GCPhys, pPage));
1368	return VMSetError(pVM, VERR_PGM_PHYS_NOT_RAM, RT_SRC_POS, "GCPhys=%RGp type=%d", GCPhys, PGM_PAGE_GET_TYPE(pPage));
1369	}
1370
1371	/** @todo What about ballooning of large pages??! */
1372	Assert( PGM_PAGE_GET_PDE_TYPE(pPage) != PGM_PAGE_PDE_TYPE_PDE
1373	&& PGM_PAGE_GET_PDE_TYPE(pPage) != PGM_PAGE_PDE_TYPE_PDE_DISABLED);
1374
1375	if ( PGM_PAGE_IS_ZERO(pPage)
1376	\|\| PGM_PAGE_IS_BALLOONED(pPage))
1377	return VINF_SUCCESS;
1378
1379	const uint32_t idPage = PGM_PAGE_GET_PAGEID(pPage);
1380	Log3(("pgmPhysFreePage: idPage=%#x GCPhys=%RGp pPage=%R[pgmpage]\n", idPage, GCPhys, pPage));
1381	if (RT_UNLIKELY(!PGM_IS_IN_NEM_MODE(pVM)
1382	? idPage == NIL_GMM_PAGEID
1383	\|\| idPage > GMM_PAGEID_LAST
1384	\|\| PGM_PAGE_GET_CHUNKID(pPage) == NIL_GMM_CHUNKID
1385	: idPage != NIL_GMM_PAGEID))
1386	{
1387	AssertMsgFailed(("GCPhys=%RGp pPage=%R[pgmpage]\n", GCPhys, pPage));
1388	return VMSetError(pVM, VERR_PGM_PHYS_INVALID_PAGE_ID, RT_SRC_POS, "GCPhys=%RGp idPage=%#x", GCPhys, pPage);
1389	}
1390	#ifdef VBOX_WITH_NATIVE_NEM
1391	const RTHCPHYS HCPhysPrev = PGM_PAGE_GET_HCPHYS(pPage);
1392	#endif
1393
1394	/* update page count stats. */
1395	if (PGM_PAGE_IS_SHARED(pPage))
1396	pVM->pgm.s.cSharedPages--;
1397	else
1398	pVM->pgm.s.cPrivatePages--;
1399	pVM->pgm.s.cZeroPages++;
1400
1401	/* Deal with write monitored pages. */
1402	if (PGM_PAGE_GET_STATE(pPage) == PGM_PAGE_STATE_WRITE_MONITORED)
1403	{
1404	PGM_PAGE_SET_WRITTEN_TO(pVM, pPage);
1405	pVM->pgm.s.cWrittenToPages++;
1406	}
1407
1408	/*
1409	* pPage = ZERO page.
1410	*/
1411	PGM_PAGE_SET_HCPHYS(pVM, pPage, pVM->pgm.s.HCPhysZeroPg);
1412	PGM_PAGE_SET_STATE(pVM, pPage, PGM_PAGE_STATE_ZERO);
1413	PGM_PAGE_SET_PAGEID(pVM, pPage, NIL_GMM_PAGEID);
1414	PGM_PAGE_SET_PDE_TYPE(pVM, pPage, PGM_PAGE_PDE_TYPE_DONTCARE);
1415	PGM_PAGE_SET_PTE_INDEX(pVM, pPage, 0);
1416	PGM_PAGE_SET_TRACKING(pVM, pPage, 0);
1417
1418	/* Flush physical page map TLB entry. */
1419	pgmPhysInvalidatePageMapTLBEntry(pVM, GCPhys);
1420	IEMTlbInvalidateAllPhysicalAllCpus(pVM, NIL_VMCPUID); /// @todo move to the perform step.
1421
1422	#ifdef VBOX_WITH_PGM_NEM_MODE
1423	/*
1424	* Skip the rest if we're doing a bulk free in NEM memory mode.
1425	*/
1426	if (!pReq)
1427	return VINF_SUCCESS;
1428	AssertLogRelReturn(!pVM->pgm.s.fNemMode, VERR_PGM_NOT_SUPPORTED_FOR_NEM_MODE);
1429	#endif
1430
1431	#ifdef VBOX_WITH_NATIVE_NEM
1432	/* Notify NEM. */
1433	/** @todo Remove this one? */
1434	if (VM_IS_NEM_ENABLED(pVM))
1435	{
1436	uint8_t u2State = PGM_PAGE_GET_NEM_STATE(pPage);
1437	NEMHCNotifyPhysPageChanged(pVM, GCPhys, HCPhysPrev, pVM->pgm.s.HCPhysZeroPg, pVM->pgm.s.abZeroPg,
1438	pgmPhysPageCalcNemProtection(pPage, enmNewType), enmNewType, &u2State);
1439	PGM_PAGE_SET_NEM_STATE(pPage, u2State);
1440	}
1441	#else
1442	RT_NOREF(enmNewType);
1443	#endif
1444
1445	/*
1446	* Make sure it's not in the handy page array.
1447	*/
1448	for (uint32_t i = pVM->pgm.s.cHandyPages; i < RT_ELEMENTS(pVM->pgm.s.aHandyPages); i++)
1449	{
1450	if (pVM->pgm.s.aHandyPages[i].idPage == idPage)
1451	{
1452	pVM->pgm.s.aHandyPages[i].HCPhysGCPhys = NIL_GMMPAGEDESC_PHYS;
1453	pVM->pgm.s.aHandyPages[i].fZeroed = false;
1454	pVM->pgm.s.aHandyPages[i].idPage = NIL_GMM_PAGEID;
1455	break;
1456	}
1457	if (pVM->pgm.s.aHandyPages[i].idSharedPage == idPage)
1458	{
1459	pVM->pgm.s.aHandyPages[i].idSharedPage = NIL_GMM_PAGEID;
1460	break;
1461	}
1462	}
1463
1464	/*
1465	* Push it onto the page array.
1466	*/
1467	uint32_t iPage = *pcPendingPages;
1468	Assert(iPage < PGMPHYS_FREE_PAGE_BATCH_SIZE);
1469	*pcPendingPages += 1;
1470
1471	pReq->aPages[iPage].idPage = idPage;
1472
1473	if (iPage + 1 < PGMPHYS_FREE_PAGE_BATCH_SIZE)
1474	return VINF_SUCCESS;
1475
1476	/*
1477	* Flush the pages.
1478	*/
1479	int rc = GMMR3FreePagesPerform(pVM, pReq, PGMPHYS_FREE_PAGE_BATCH_SIZE);
1480	if (RT_SUCCESS(rc))
1481	{
1482	GMMR3FreePagesRePrep(pVM, pReq, PGMPHYS_FREE_PAGE_BATCH_SIZE, GMMACCOUNT_BASE);
1483	*pcPendingPages = 0;
1484	}
1485	return rc;
1486	}
1487
1488
1489	/**
1490	* Frees a range of pages, replacing them with ZERO pages of the specified type.
1491	*
1492	* @returns VBox status code.
1493	* @param pVM The cross context VM structure.
1494	* @param pRam The RAM range in which the pages resides.
1495	* @param GCPhys The address of the first page.
1496	* @param GCPhysLast The address of the last page.
1497	* @param pvMmio2 Pointer to the ring-3 mapping of any MMIO2 memory that
1498	* will replace the pages we're freeing up.
1499	*/
1500	static int pgmR3PhysFreePageRange(PVM pVM, PPGMRAMRANGE pRam, RTGCPHYS GCPhys, RTGCPHYS GCPhysLast, void *pvMmio2)
1501	{
1502	PGM_LOCK_ASSERT_OWNER(pVM);
1503
1504	#ifdef VBOX_WITH_PGM_NEM_MODE
1505	/*
1506	* In simplified memory mode we don't actually free the memory,
1507	* we just unmap it and let NEM do any unlocking of it.
1508	*/
1509	if (pVM->pgm.s.fNemMode)
1510	{
1511	Assert(VM_IS_NEM_ENABLED(pVM) \|\| VM_IS_EXEC_ENGINE_IEM(pVM));
1512	uint8_t u2State = 0; /* (We don't support UINT8_MAX here.) */
1513	if (VM_IS_NEM_ENABLED(pVM))
1514	{
1515	uint32_t const fNemNotify = (pvMmio2 ? NEM_NOTIFY_PHYS_MMIO_EX_F_MMIO2 : 0) \| NEM_NOTIFY_PHYS_MMIO_EX_F_REPLACE;
1516	int rc = NEMR3NotifyPhysMmioExMapEarly(pVM, GCPhys, GCPhysLast - GCPhys + 1, fNemNotify,
1517	pRam->pvR3 ? (uint8_t *)pRam->pvR3 + GCPhys - pRam->GCPhys : NULL,
1518	pvMmio2, &u2State, NULL /puNemRange/);
1519	AssertLogRelRCReturn(rc, rc);
1520	}
1521
1522	/* Iterate the pages. */
1523	PPGMPAGE pPageDst = &pRam->aPages[(GCPhys - pRam->GCPhys) >> GUEST_PAGE_SHIFT];
1524	uint32_t cPagesLeft = ((GCPhysLast - GCPhys) >> GUEST_PAGE_SHIFT) + 1;
1525	while (cPagesLeft-- > 0)
1526	{
1527	int rc = pgmPhysFreePage(pVM, NULL, NULL, pPageDst, GCPhys, PGMPAGETYPE_MMIO);
1528	AssertLogRelRCReturn(rc, rc); /* We're done for if this goes wrong. */
1529
1530	PGM_PAGE_SET_TYPE(pVM, pPageDst, PGMPAGETYPE_MMIO);
1531	PGM_PAGE_SET_NEM_STATE(pPageDst, u2State);
1532
1533	GCPhys += GUEST_PAGE_SIZE;
1534	pPageDst++;
1535	}
1536	return VINF_SUCCESS;
1537	}
1538	#else /* !VBOX_WITH_PGM_NEM_MODE */
1539	RT_NOREF(pvMmio2);
1540	#endif /* !VBOX_WITH_PGM_NEM_MODE */
1541
1542	/*
1543	* Regular mode.
1544	*/
1545	/* Prepare. */
1546	uint32_t cPendingPages = 0;
1547	PGMMFREEPAGESREQ pReq;
1548	int rc = GMMR3FreePagesPrepare(pVM, &pReq, PGMPHYS_FREE_PAGE_BATCH_SIZE, GMMACCOUNT_BASE);
1549	AssertLogRelRCReturn(rc, rc);
1550
1551	#ifdef VBOX_WITH_NATIVE_NEM
1552	/* Tell NEM up-front. */
1553	uint8_t u2State = UINT8_MAX;
1554	if (VM_IS_NEM_ENABLED(pVM))
1555	{
1556	uint32_t const fNemNotify = (pvMmio2 ? NEM_NOTIFY_PHYS_MMIO_EX_F_MMIO2 : 0) \| NEM_NOTIFY_PHYS_MMIO_EX_F_REPLACE;
1557	rc = NEMR3NotifyPhysMmioExMapEarly(pVM, GCPhys, GCPhysLast - GCPhys + 1, fNemNotify, NULL, pvMmio2,
1558	&u2State, NULL /puNemRange/);
1559	AssertLogRelRCReturnStmt(rc, GMMR3FreePagesCleanup(pReq), rc);
1560	}
1561	#endif
1562
1563	/* Iterate the pages. */
1564	PPGMPAGE pPageDst = &pRam->aPages[(GCPhys - pRam->GCPhys) >> GUEST_PAGE_SHIFT];
1565	uint32_t cPagesLeft = ((GCPhysLast - GCPhys) >> GUEST_PAGE_SHIFT) + 1;
1566	while (cPagesLeft-- > 0)
1567	{
1568	rc = pgmPhysFreePage(pVM, pReq, &cPendingPages, pPageDst, GCPhys, PGMPAGETYPE_MMIO);
1569	AssertLogRelRCReturn(rc, rc); /* We're done for if this goes wrong. */
1570
1571	PGM_PAGE_SET_TYPE(pVM, pPageDst, PGMPAGETYPE_MMIO);
1572	#ifdef VBOX_WITH_NATIVE_NEM
1573	if (u2State != UINT8_MAX)
1574	PGM_PAGE_SET_NEM_STATE(pPageDst, u2State);
1575	#endif
1576
1577	GCPhys += GUEST_PAGE_SIZE;
1578	pPageDst++;
1579	}
1580
1581	/* Finish pending and cleanup. */
1582	if (cPendingPages)
1583	{
1584	rc = GMMR3FreePagesPerform(pVM, pReq, cPendingPages);
1585	AssertLogRelRCReturn(rc, rc);
1586	}
1587	GMMR3FreePagesCleanup(pReq);
1588
1589	return rc;
1590	}
1591
1592
1593	/**
1594	* PGMR3PhysRegisterRam worker that initializes and links a RAM range.
1595	*
1596	* In NEM mode, this will allocate the pages backing the RAM range and this may
1597	* fail. NEM registration may also fail. (In regular HM mode it won't fail.)
1598	*
1599	* @returns VBox status code.
1600	* @param pVM The cross context VM structure.
1601	* @param pNew The new RAM range.
1602	* @param GCPhys The address of the RAM range.
1603	* @param GCPhysLast The last address of the RAM range.
1604	* @param R0PtrNew Ditto for R0.
1605	* @param fFlags PGM_RAM_RANGE_FLAGS_FLOATING or zero.
1606	* @param pszDesc The description.
1607	* @param pPrev The previous RAM range (for linking).
1608	*/
1609	static int pgmR3PhysInitAndLinkRamRange(PVM pVM, PPGMRAMRANGE pNew, RTGCPHYS GCPhys, RTGCPHYS GCPhysLast,
1610	RTR0PTR R0PtrNew, uint32_t fFlags, const char *pszDesc, PPGMRAMRANGE pPrev)
1611	{
1612	/*
1613	* Initialize the range.
1614	*/
1615	pNew->pSelfR0 = R0PtrNew;
1616	pNew->GCPhys = GCPhys;
1617	pNew->GCPhysLast = GCPhysLast;
1618	pNew->cb = GCPhysLast - GCPhys + 1;
1619	pNew->pszDesc = pszDesc;
1620	pNew->fFlags = fFlags;
1621	pNew->uNemRange = UINT32_MAX;
1622	pNew->pvR3 = NULL;
1623	pNew->paLSPages = NULL;
1624
1625	uint32_t const cPages = pNew->cb >> GUEST_PAGE_SHIFT;
1626	#ifdef VBOX_WITH_PGM_NEM_MODE
1627	if (!pVM->pgm.s.fNemMode)
1628	#endif
1629	{
1630	RTGCPHYS iPage = cPages;
1631	while (iPage-- > 0)
1632	PGM_PAGE_INIT_ZERO(&pNew->aPages[iPage], pVM, PGMPAGETYPE_RAM);
1633
1634	/* Update the page count stats. */
1635	pVM->pgm.s.cZeroPages += cPages;
1636	pVM->pgm.s.cAllPages += cPages;
1637	}
1638	#ifdef VBOX_WITH_PGM_NEM_MODE
1639	else
1640	{
1641	int rc = SUPR3PageAlloc(RT_ALIGN_Z(pNew->cb, HOST_PAGE_SIZE) >> HOST_PAGE_SHIFT,
1642	pVM->pgm.s.fUseLargePages ? SUP_PAGE_ALLOC_F_LARGE_PAGES : 0, &pNew->pvR3);
1643	if (RT_FAILURE(rc))
1644	return rc;
1645
1646	RTGCPHYS iPage = cPages;
1647	while (iPage-- > 0)
1648	PGM_PAGE_INIT(&pNew->aPages[iPage], UINT64_C(0x0000fffffffff000), NIL_GMM_PAGEID,
1649	PGMPAGETYPE_RAM, PGM_PAGE_STATE_ALLOCATED);
1650
1651	/* Update the page count stats. */
1652	pVM->pgm.s.cPrivatePages += cPages;
1653	pVM->pgm.s.cAllPages += cPages;
1654	}
1655	#endif
1656
1657	/*
1658	* Link it.
1659	*/
1660	pgmR3PhysLinkRamRange(pVM, pNew, pPrev);
1661
1662	#ifdef VBOX_WITH_NATIVE_NEM
1663	/*
1664	* Notify NEM now that it has been linked.
1665	*/
1666	if (VM_IS_NEM_ENABLED(pVM))
1667	{
1668	uint8_t u2State = UINT8_MAX;
1669	int rc = NEMR3NotifyPhysRamRegister(pVM, GCPhys, pNew->cb, pNew->pvR3, &u2State, &pNew->uNemRange);
1670	if (RT_SUCCESS(rc))
1671	{
1672	if (u2State != UINT8_MAX)
1673	pgmPhysSetNemStateForPages(&pNew->aPages[0], cPages, u2State);
1674	}
1675	else
1676	pgmR3PhysUnlinkRamRange2(pVM, pNew, pPrev);
1677	return rc;
1678	}
1679	#endif
1680	return VINF_SUCCESS;
1681	}
1682
1683
1684	/**
1685	* PGMR3PhysRegisterRam worker that registers a high chunk.
1686	*
1687	* @returns VBox status code.
1688	* @param pVM The cross context VM structure.
1689	* @param GCPhys The address of the RAM.
1690	* @param cRamPages The number of RAM pages to register.
1691	* @param iChunk The chunk number.
1692	* @param pszDesc The RAM range description.
1693	* @param ppPrev Previous RAM range pointer. In/Out.
1694	*/
1695	static int pgmR3PhysRegisterHighRamChunk(PVM pVM, RTGCPHYS GCPhys, uint32_t cRamPages, uint32_t iChunk,
1696	const char pszDesc, PPGMRAMRANGE ppPrev)
1697	{
1698	const char *pszDescChunk = iChunk == 0
1699	? pszDesc
1700	: MMR3HeapAPrintf(pVM, MM_TAG_PGM_PHYS, "%s (#%u)", pszDesc, iChunk + 1);
1701	AssertReturn(pszDescChunk, VERR_NO_MEMORY);
1702
1703	/*
1704	* Allocate memory for the new chunk.
1705	*/
1706	size_t const cChunkPages = RT_ALIGN_Z(RT_UOFFSETOF_DYN(PGMRAMRANGE, aPages[cRamPages]), HOST_PAGE_SIZE) >> HOST_PAGE_SHIFT;
1707	PSUPPAGE paChunkPages = (PSUPPAGE)RTMemTmpAllocZ(sizeof(SUPPAGE) * cChunkPages);
1708	AssertReturn(paChunkPages, VERR_NO_TMP_MEMORY);
1709	RTR0PTR R0PtrChunk = NIL_RTR0PTR;
1710	void *pvChunk = NULL;
1711	int rc = SUPR3PageAllocEx(cChunkPages, 0 /fFlags/, &pvChunk, &R0PtrChunk, paChunkPages);
1712	if (RT_SUCCESS(rc))
1713	{
1714	Assert(R0PtrChunk != NIL_RTR0PTR \|\| PGM_IS_IN_NEM_MODE(pVM));
1715	memset(pvChunk, 0, cChunkPages << HOST_PAGE_SHIFT);
1716
1717	PPGMRAMRANGE pNew = (PPGMRAMRANGE)pvChunk;
1718
1719	/*
1720	* Ok, init and link the range.
1721	*/
1722	rc = pgmR3PhysInitAndLinkRamRange(pVM, pNew, GCPhys, GCPhys + ((RTGCPHYS)cRamPages << GUEST_PAGE_SHIFT) - 1,
1723	R0PtrChunk, PGM_RAM_RANGE_FLAGS_FLOATING, pszDescChunk, *ppPrev);
1724	if (RT_SUCCESS(rc))
1725	*ppPrev = pNew;
1726
1727	if (RT_FAILURE(rc))
1728	SUPR3PageFreeEx(pvChunk, cChunkPages);
1729	}
1730
1731	RTMemTmpFree(paChunkPages);
1732	return rc;
1733	}
1734
1735
1736	/**
1737	* Sets up a range RAM.
1738	*
1739	* This will check for conflicting registrations, make a resource
1740	* reservation for the memory (with GMM), and setup the per-page
1741	* tracking structures (PGMPAGE).
1742	*
1743	* @returns VBox status code.
1744	* @param pVM The cross context VM structure.
1745	* @param GCPhys The physical address of the RAM.
1746	* @param cb The size of the RAM.
1747	* @param pszDesc The description - not copied, so, don't free or change it.
1748	*/
1749	VMMR3DECL(int) PGMR3PhysRegisterRam(PVM pVM, RTGCPHYS GCPhys, RTGCPHYS cb, const char *pszDesc)
1750	{
1751	/*
1752	* Validate input.
1753	*/
1754	Log(("PGMR3PhysRegisterRam: GCPhys=%RGp cb=%RGp pszDesc=%s\n", GCPhys, cb, pszDesc));
1755	AssertReturn(RT_ALIGN_T(GCPhys, GUEST_PAGE_SIZE, RTGCPHYS) == GCPhys, VERR_INVALID_PARAMETER);
1756	AssertReturn(RT_ALIGN_T(cb, GUEST_PAGE_SIZE, RTGCPHYS) == cb, VERR_INVALID_PARAMETER);
1757	AssertReturn(cb > 0, VERR_INVALID_PARAMETER);
1758	RTGCPHYS GCPhysLast = GCPhys + (cb - 1);
1759	AssertMsgReturn(GCPhysLast > GCPhys, ("The range wraps! GCPhys=%RGp cb=%RGp\n", GCPhys, cb), VERR_INVALID_PARAMETER);
1760	AssertPtrReturn(pszDesc, VERR_INVALID_POINTER);
1761	VM_ASSERT_EMT_RETURN(pVM, VERR_VM_THREAD_NOT_EMT);
1762
1763	PGM_LOCK_VOID(pVM);
1764
1765	/*
1766	* Find range location and check for conflicts.
1767	*/
1768	PPGMRAMRANGE pPrev = NULL;
1769	PPGMRAMRANGE pRam = pVM->pgm.s.pRamRangesXR3;
1770	while (pRam && GCPhysLast >= pRam->GCPhys)
1771	{
1772	AssertLogRelMsgReturnStmt( GCPhysLast < pRam->GCPhys
1773	\|\| GCPhys > pRam->GCPhysLast,
1774	("%RGp-%RGp (%s) conflicts with existing %RGp-%RGp (%s)\n",
1775	GCPhys, GCPhysLast, pszDesc, pRam->GCPhys, pRam->GCPhysLast, pRam->pszDesc),
1776	PGM_UNLOCK(pVM), VERR_PGM_RAM_CONFLICT);
1777
1778	/* next */
1779	pPrev = pRam;
1780	pRam = pRam->pNextR3;
1781	}
1782
1783	/*
1784	* Register it with GMM (the API bitches).
1785	*/
1786	const RTGCPHYS cPages = cb >> GUEST_PAGE_SHIFT;
1787	int rc = MMR3IncreaseBaseReservation(pVM, cPages);
1788	if (RT_FAILURE(rc))
1789	{
1790	PGM_UNLOCK(pVM);
1791	return rc;
1792	}
1793
1794	if ( GCPhys >= _4G
1795	&& cPages > 256)
1796	{
1797	/*
1798	* The PGMRAMRANGE structures for the high memory can get very big.
1799	* There used to be some limitations on SUPR3PageAllocEx allocation
1800	* sizes, so traditionally we limited this to 16MB chunks. These days
1801	* we do ~64 MB chunks each covering 16GB of guest RAM, making sure
1802	* each range is a multiple of 1GB to enable eager hosts to use 1GB
1803	* pages in NEM mode.
1804	*
1805	* See also pgmR3PhysMmio2CalcChunkCount.
1806	*/
1807	uint32_t const cPagesPerChunk = _4M;
1808	Assert(RT_ALIGN_32(cPagesPerChunk, X86_PD_PAE_SHIFT - X86_PAGE_SHIFT)); /* NEM large page requirement: 1GB pages. */
1809
1810	RTGCPHYS cPagesLeft = cPages;
1811	RTGCPHYS GCPhysChunk = GCPhys;
1812	uint32_t iChunk = 0;
1813	while (cPagesLeft > 0)
1814	{
1815	uint32_t cPagesInChunk = cPagesLeft;
1816	if (cPagesInChunk > cPagesPerChunk)
1817	cPagesInChunk = cPagesPerChunk;
1818
1819	rc = pgmR3PhysRegisterHighRamChunk(pVM, GCPhysChunk, cPagesInChunk, iChunk, pszDesc, &pPrev);
1820	AssertRCReturn(rc, rc);
1821
1822	/* advance */
1823	GCPhysChunk += (RTGCPHYS)cPagesInChunk << GUEST_PAGE_SHIFT;
1824	cPagesLeft -= cPagesInChunk;
1825	iChunk++;
1826	}
1827	}
1828	else
1829	{
1830	/*
1831	* Allocate, initialize and link the new RAM range.
1832	*/
1833	const size_t cbRamRange = RT_UOFFSETOF_DYN(PGMRAMRANGE, aPages[cPages]);
1834	PPGMRAMRANGE pNew = NULL;
1835	RTR0PTR pNewR0 = NIL_RTR0PTR;
1836	rc = SUPR3PageAllocEx(RT_ALIGN_Z(cbRamRange, HOST_PAGE_SIZE) >> HOST_PAGE_SHIFT, 0 /fFlags/,
1837	(void *)&pNew, &pNewR0, NULL /paPages*/);
1838	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc cbRamRange=%zu\n", rc, cbRamRange), rc);
1839
1840	rc = pgmR3PhysInitAndLinkRamRange(pVM, pNew, GCPhys, GCPhysLast, pNewR0, 0 /fFlags/, pszDesc, pPrev);
1841	AssertLogRelMsgRCReturn(rc, ("rc=%Rrc cbRamRange=%zu\n", rc, cbRamRange), rc);
1842	}
1843	pgmPhysInvalidatePageMapTLB(pVM);
1844
1845	PGM_UNLOCK(pVM);
1846	return rc;
1847	}
1848
1849
1850	/**
1851	* Worker called by PGMR3InitFinalize if we're configured to pre-allocate RAM.
1852	*
1853	* We do this late in the init process so that all the ROM and MMIO ranges have
1854	* been registered already and we don't go wasting memory on them.
1855	*
1856	* @returns VBox status code.
1857	*
1858	* @param pVM The cross context VM structure.
1859	*/
1860	int pgmR3PhysRamPreAllocate(PVM pVM)
1861	{
1862	Assert(pVM->pgm.s.fRamPreAlloc);
1863	Log(("pgmR3PhysRamPreAllocate: enter\n"));
1864	#ifdef VBOX_WITH_PGM_NEM_MODE
1865	AssertLogRelReturn(!pVM->pgm.s.fNemMode, VERR_PGM_NOT_SUPPORTED_FOR_NEM_MODE);
1866	#endif
1867
1868	/*
1869	* Walk the RAM ranges and allocate all RAM pages, halt at
1870	* the first allocation error.
1871	*/
1872	uint64_t cPages = 0;
1873	uint64_t NanoTS = RTTimeNanoTS();
1874	PGM_LOCK_VOID(pVM);
1875	for (PPGMRAMRANGE pRam = pVM->pgm.s.pRamRangesXR3; pRam; pRam = pRam->pNextR3)
1876	{
1877	PPGMPAGE pPage = &pRam->aPages[0];
1878	RTGCPHYS GCPhys = pRam->GCPhys;
1879	uint32_t cLeft = pRam->cb >> GUEST_PAGE_SHIFT;
1880	while (cLeft-- > 0)
1881	{
1882	if (PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_RAM)
1883	{
1884	switch (PGM_PAGE_GET_STATE(pPage))
1885	{
1886	case PGM_PAGE_STATE_ZERO:
1887	{
1888	int rc = pgmPhysAllocPage(pVM, pPage, GCPhys);
1889	if (RT_FAILURE(rc))
1890	{
1891	LogRel(("PGM: RAM Pre-allocation failed at %RGp (in %s) with rc=%Rrc\n", GCPhys, pRam->pszDesc, rc));
1892	PGM_UNLOCK(pVM);
1893	return rc;
1894	}
1895	cPages++;
1896	break;
1897	}
1898
1899	case PGM_PAGE_STATE_BALLOONED:
1900	case PGM_PAGE_STATE_ALLOCATED:
1901	case PGM_PAGE_STATE_WRITE_MONITORED:
1902	case PGM_PAGE_STATE_SHARED:
1903	/* nothing to do here. */
1904	break;
1905	}
1906	}
1907
1908	/* next */
1909	pPage++;
1910	GCPhys += GUEST_PAGE_SIZE;
1911	}
1912	}
1913	PGM_UNLOCK(pVM);
1914	NanoTS = RTTimeNanoTS() - NanoTS;
1915
1916	LogRel(("PGM: Pre-allocated %llu pages in %llu ms\n", cPages, NanoTS / 1000000));
1917	Log(("pgmR3PhysRamPreAllocate: returns VINF_SUCCESS\n"));
1918	return VINF_SUCCESS;
1919	}
1920
1921
1922	/**
1923	* Checks shared page checksums.
1924	*
1925	* @param pVM The cross context VM structure.
1926	*/
1927	void pgmR3PhysAssertSharedPageChecksums(PVM pVM)
1928	{
1929	#ifdef VBOX_STRICT
1930	PGM_LOCK_VOID(pVM);
1931
1932	if (pVM->pgm.s.cSharedPages > 0)
1933	{
1934	/*
1935	* Walk the ram ranges.
1936	*/
1937	for (PPGMRAMRANGE pRam = pVM->pgm.s.pRamRangesXR3; pRam; pRam = pRam->pNextR3)
1938	{
1939	uint32_t iPage = pRam->cb >> GUEST_PAGE_SHIFT;
1940	AssertMsg(((RTGCPHYS)iPage << GUEST_PAGE_SHIFT) == pRam->cb,
1941	("%RGp %RGp\n", (RTGCPHYS)iPage << GUEST_PAGE_SHIFT, pRam->cb));
1942
1943	while (iPage-- > 0)
1944	{
1945	PPGMPAGE pPage = &pRam->aPages[iPage];
1946	if (PGM_PAGE_IS_SHARED(pPage))
1947	{
1948	uint32_t u32Checksum = pPage->s.u2Unused0/* \| ((uint32_t)pPage->s.u2Unused1 << 8)*/;
1949	if (!u32Checksum)
1950	{
1951	RTGCPHYS GCPhysPage = pRam->GCPhys + ((RTGCPHYS)iPage << GUEST_PAGE_SHIFT);
1952	void const *pvPage;
1953	int rc = pgmPhysPageMapReadOnly(pVM, pPage, GCPhysPage, &pvPage);
1954	if (RT_SUCCESS(rc))
1955	{
1956	uint32_t u32Checksum2 = RTCrc32(pvPage, GUEST_PAGE_SIZE);
1957	# if 0
1958	AssertMsg((u32Checksum2 & /UINT32_C(0x00000303)/ 0x3) == u32Checksum, ("GCPhysPage=%RGp\n", GCPhysPage));
1959	# else
1960	if ((u32Checksum2 & /UINT32_C(0x00000303)/ 0x3) == u32Checksum)
1961	LogFlow(("shpg %#x @ %RGp %#x [OK]\n", PGM_PAGE_GET_PAGEID(pPage), GCPhysPage, u32Checksum2));
1962	else
1963	AssertMsgFailed(("shpg %#x @ %RGp %#x\n", PGM_PAGE_GET_PAGEID(pPage), GCPhysPage, u32Checksum2));
1964	# endif
1965	}
1966	else
1967	AssertRC(rc);
1968	}
1969	}
1970
1971	} /* for each page */
1972
1973	} /* for each ram range */
1974	}
1975
1976	PGM_UNLOCK(pVM);
1977	#endif /* VBOX_STRICT */
1978	NOREF(pVM);
1979	}
1980
1981
1982	/**
1983	* Resets the physical memory state.
1984	*
1985	* ASSUMES that the caller owns the PGM lock.
1986	*
1987	* @returns VBox status code.
1988	* @param pVM The cross context VM structure.
1989	*/
1990	int pgmR3PhysRamReset(PVM pVM)
1991	{
1992	PGM_LOCK_ASSERT_OWNER(pVM);
1993
1994	/* Reset the memory balloon. */
1995	int rc = GMMR3BalloonedPages(pVM, GMMBALLOONACTION_RESET, 0);
1996	AssertRC(rc);
1997
1998	#ifdef VBOX_WITH_PAGE_SHARING
1999	/* Clear all registered shared modules. */
2000	pgmR3PhysAssertSharedPageChecksums(pVM);
2001	rc = GMMR3ResetSharedModules(pVM);
2002	AssertRC(rc);
2003	#endif
2004	/* Reset counters. */
2005	pVM->pgm.s.cReusedSharedPages = 0;
2006	pVM->pgm.s.cBalloonedPages = 0;
2007
2008	return VINF_SUCCESS;
2009	}
2010
2011
2012	/**
2013	* Resets (zeros) the RAM after all devices and components have been reset.
2014	*
2015	* ASSUMES that the caller owns the PGM lock.
2016	*
2017	* @returns VBox status code.
2018	* @param pVM The cross context VM structure.
2019	*/
2020	int pgmR3PhysRamZeroAll(PVM pVM)
2021	{
2022	PGM_LOCK_ASSERT_OWNER(pVM);
2023
2024	/*
2025	* We batch up pages that should be freed instead of calling GMM for
2026	* each and every one of them.
2027	*/
2028	uint32_t cPendingPages = 0;
2029	PGMMFREEPAGESREQ pReq;
2030	int rc = GMMR3FreePagesPrepare(pVM, &pReq, PGMPHYS_FREE_PAGE_BATCH_SIZE, GMMACCOUNT_BASE);
2031	AssertLogRelRCReturn(rc, rc);
2032
2033	/*
2034	* Walk the ram ranges.
2035	*/
2036	for (PPGMRAMRANGE pRam = pVM->pgm.s.pRamRangesXR3; pRam; pRam = pRam->pNextR3)
2037	{
2038	uint32_t iPage = pRam->cb >> GUEST_PAGE_SHIFT;
2039	AssertMsg(((RTGCPHYS)iPage << GUEST_PAGE_SHIFT) == pRam->cb, ("%RGp %RGp\n", (RTGCPHYS)iPage << GUEST_PAGE_SHIFT, pRam->cb));
2040
2041	if ( !pVM->pgm.s.fRamPreAlloc
2042	#ifdef VBOX_WITH_PGM_NEM_MODE
2043	&& !pVM->pgm.s.fNemMode
2044	#endif
2045	&& pVM->pgm.s.fZeroRamPagesOnReset)
2046	{
2047	/* Replace all RAM pages by ZERO pages. */
2048	while (iPage-- > 0)
2049	{
2050	PPGMPAGE pPage = &pRam->aPages[iPage];
2051	switch (PGM_PAGE_GET_TYPE(pPage))
2052	{
2053	case PGMPAGETYPE_RAM:
2054	/* Do not replace pages part of a 2 MB continuous range
2055	with zero pages, but zero them instead. */
2056	if ( PGM_PAGE_GET_PDE_TYPE(pPage) == PGM_PAGE_PDE_TYPE_PDE
2057	\|\| PGM_PAGE_GET_PDE_TYPE(pPage) == PGM_PAGE_PDE_TYPE_PDE_DISABLED)
2058	{
2059	void *pvPage;
2060	rc = pgmPhysPageMap(pVM, pPage, pRam->GCPhys + ((RTGCPHYS)iPage << GUEST_PAGE_SHIFT), &pvPage);
2061	AssertLogRelRCReturn(rc, rc);
2062	RT_BZERO(pvPage, GUEST_PAGE_SIZE);
2063	}
2064	else if (PGM_PAGE_IS_BALLOONED(pPage))
2065	{
2066	/* Turn into a zero page; the balloon status is lost when the VM reboots. */
2067	PGM_PAGE_SET_STATE(pVM, pPage, PGM_PAGE_STATE_ZERO);
2068	}
2069	else if (!PGM_PAGE_IS_ZERO(pPage))
2070	{
2071	rc = pgmPhysFreePage(pVM, pReq, &cPendingPages, pPage,
2072	pRam->GCPhys + ((RTGCPHYS)iPage << GUEST_PAGE_SHIFT), PGMPAGETYPE_RAM);
2073	AssertLogRelRCReturn(rc, rc);
2074	}
2075	break;
2076
2077	case PGMPAGETYPE_MMIO2_ALIAS_MMIO:
2078	case PGMPAGETYPE_SPECIAL_ALIAS_MMIO: /** @todo perhaps leave the special page alone? I don't think VT-x copes with this code. */
2079	pgmHandlerPhysicalResetAliasedPage(pVM, pPage, pRam->GCPhys + ((RTGCPHYS)iPage << GUEST_PAGE_SHIFT),
2080	pRam, true /fDoAccounting/, false /fFlushIemTlbs/);
2081	break;
2082
2083	case PGMPAGETYPE_MMIO2:
2084	case PGMPAGETYPE_ROM_SHADOW: /* handled by pgmR3PhysRomReset. */
2085	case PGMPAGETYPE_ROM:
2086	case PGMPAGETYPE_MMIO:
2087	break;
2088	default:
2089	AssertFailed();
2090	}
2091	} /* for each page */
2092	}
2093	else
2094	{
2095	/* Zero the memory. */
2096	while (iPage-- > 0)
2097	{
2098	PPGMPAGE pPage = &pRam->aPages[iPage];
2099	switch (PGM_PAGE_GET_TYPE(pPage))
2100	{
2101	case PGMPAGETYPE_RAM:
2102	switch (PGM_PAGE_GET_STATE(pPage))
2103	{
2104	case PGM_PAGE_STATE_ZERO:
2105	break;
2106
2107	case PGM_PAGE_STATE_BALLOONED:
2108	/* Turn into a zero page; the balloon status is lost when the VM reboots. */
2109	PGM_PAGE_SET_STATE(pVM, pPage, PGM_PAGE_STATE_ZERO);
2110	break;
2111
2112	case PGM_PAGE_STATE_SHARED:
2113	case PGM_PAGE_STATE_WRITE_MONITORED:
2114	rc = pgmPhysPageMakeWritable(pVM, pPage, pRam->GCPhys + ((RTGCPHYS)iPage << GUEST_PAGE_SHIFT));
2115	AssertLogRelRCReturn(rc, rc);
2116	RT_FALL_THRU();
2117
2118	case PGM_PAGE_STATE_ALLOCATED:
2119	if (pVM->pgm.s.fZeroRamPagesOnReset)
2120	{
2121	void *pvPage;
2122	rc = pgmPhysPageMap(pVM, pPage, pRam->GCPhys + ((RTGCPHYS)iPage << GUEST_PAGE_SHIFT), &pvPage);
2123	AssertLogRelRCReturn(rc, rc);
2124	RT_BZERO(pvPage, GUEST_PAGE_SIZE);
2125	}
2126	break;
2127	}
2128	break;
2129
2130	case PGMPAGETYPE_MMIO2_ALIAS_MMIO:
2131	case PGMPAGETYPE_SPECIAL_ALIAS_MMIO: /** @todo perhaps leave the special page alone? I don't think VT-x copes with this code. */
2132	pgmHandlerPhysicalResetAliasedPage(pVM, pPage, pRam->GCPhys + ((RTGCPHYS)iPage << GUEST_PAGE_SHIFT),
2133	pRam, true /fDoAccounting/, false /fFlushIemTlbs/);
2134	break;
2135
2136	case PGMPAGETYPE_MMIO2:
2137	case PGMPAGETYPE_ROM_SHADOW:
2138	case PGMPAGETYPE_ROM:
2139	case PGMPAGETYPE_MMIO:
2140	break;
2141	default:
2142	AssertFailed();
2143
2144	}
2145	} /* for each page */
2146	}
2147
2148	}
2149
2150	/*
2151	* Finish off any pages pending freeing.
2152	*/
2153	if (cPendingPages)
2154	{
2155	rc = GMMR3FreePagesPerform(pVM, pReq, cPendingPages);
2156	AssertLogRelRCReturn(rc, rc);
2157	}
2158	GMMR3FreePagesCleanup(pReq);
2159
2160	/*
2161	* Flush the IEM TLB, just to be sure it really is done.
2162	*/
2163	IEMTlbInvalidateAllPhysicalAllCpus(pVM, NIL_VMCPUID);
2164
2165	return VINF_SUCCESS;
2166	}
2167
2168
2169	/**
2170	* Frees all RAM during VM termination
2171	*
2172	* ASSUMES that the caller owns the PGM lock.
2173	*
2174	* @returns VBox status code.
2175	* @param pVM The cross context VM structure.
2176	*/
2177	int pgmR3PhysRamTerm(PVM pVM)
2178	{
2179	PGM_LOCK_ASSERT_OWNER(pVM);
2180
2181	/* Reset the memory balloon. */
2182	int rc = GMMR3BalloonedPages(pVM, GMMBALLOONACTION_RESET, 0);
2183	AssertRC(rc);
2184
2185	#ifdef VBOX_WITH_PAGE_SHARING
2186	/*
2187	* Clear all registered shared modules.
2188	*/
2189	pgmR3PhysAssertSharedPageChecksums(pVM);
2190	rc = GMMR3ResetSharedModules(pVM);
2191	AssertRC(rc);
2192
2193	/*
2194	* Flush the handy pages updates to make sure no shared pages are hiding
2195	* in there. (Not unlikely if the VM shuts down, apparently.)
2196	*/
2197	# ifdef VBOX_WITH_PGM_NEM_MODE
2198	if (!pVM->pgm.s.fNemMode)
2199	# endif
2200	rc = VMMR3CallR0(pVM, VMMR0_DO_PGM_FLUSH_HANDY_PAGES, 0, NULL);
2201	#endif
2202
2203	/*
2204	* We batch up pages that should be freed instead of calling GMM for
2205	* each and every one of them.
2206	*/
2207	uint32_t cPendingPages = 0;
2208	PGMMFREEPAGESREQ pReq;
2209	rc = GMMR3FreePagesPrepare(pVM, &pReq, PGMPHYS_FREE_PAGE_BATCH_SIZE, GMMACCOUNT_BASE);
2210	AssertLogRelRCReturn(rc, rc);
2211
2212	/*
2213	* Walk the ram ranges.
2214	*/
2215	for (PPGMRAMRANGE pRam = pVM->pgm.s.pRamRangesXR3; pRam; pRam = pRam->pNextR3)
2216	{
2217	uint32_t iPage = pRam->cb >> GUEST_PAGE_SHIFT;
2218	AssertMsg(((RTGCPHYS)iPage << GUEST_PAGE_SHIFT) == pRam->cb, ("%RGp %RGp\n", (RTGCPHYS)iPage << GUEST_PAGE_SHIFT, pRam->cb));
2219
2220	while (iPage-- > 0)
2221	{
2222	PPGMPAGE pPage = &pRam->aPages[iPage];
2223	switch (PGM_PAGE_GET_TYPE(pPage))
2224	{
2225	case PGMPAGETYPE_RAM:
2226	/* Free all shared pages. Private pages are automatically freed during GMM VM cleanup. */
2227	/** @todo change this to explicitly free private pages here. */
2228	if (PGM_PAGE_IS_SHARED(pPage))
2229	{
2230	rc = pgmPhysFreePage(pVM, pReq, &cPendingPages, pPage,
2231	pRam->GCPhys + ((RTGCPHYS)iPage << GUEST_PAGE_SHIFT), PGMPAGETYPE_RAM);
2232	AssertLogRelRCReturn(rc, rc);
2233	}
2234	break;
2235
2236	case PGMPAGETYPE_MMIO2_ALIAS_MMIO:
2237	case PGMPAGETYPE_SPECIAL_ALIAS_MMIO:
2238	case PGMPAGETYPE_MMIO2:
2239	case PGMPAGETYPE_ROM_SHADOW: /* handled by pgmR3PhysRomReset. */
2240	case PGMPAGETYPE_ROM:
2241	case PGMPAGETYPE_MMIO:
2242	break;
2243	default:
2244	AssertFailed();
2245	}
2246	} /* for each page */
2247	}
2248
2249	/*
2250	* Finish off any pages pending freeing.
2251	*/
2252	if (cPendingPages)
2253	{
2254	rc = GMMR3FreePagesPerform(pVM, pReq, cPendingPages);
2255	AssertLogRelRCReturn(rc, rc);
2256	}
2257	GMMR3FreePagesCleanup(pReq);
2258	return VINF_SUCCESS;
2259	}
2260
2261
2262
2263	/*********************************************************************************************************************************
2264	* MMIO *
2265	*********************************************************************************************************************************/
2266
2267	/**
2268	* This is the interface IOM is using to register an MMIO region.
2269	*
2270	* It will check for conflicts and ensure that a RAM range structure
2271	* is present before calling the PGMR3HandlerPhysicalRegister API to
2272	* register the callbacks.
2273	*
2274	* @returns VBox status code.
2275	*
2276	* @param pVM The cross context VM structure.
2277	* @param GCPhys The start of the MMIO region.
2278	* @param cb The size of the MMIO region.
2279	* @param hType The physical access handler type registration.
2280	* @param uUser The user argument.
2281	* @param pszDesc The description of the MMIO region.
2282	*/
2283	VMMR3DECL(int) PGMR3PhysMMIORegister(PVM pVM, RTGCPHYS GCPhys, RTGCPHYS cb, PGMPHYSHANDLERTYPE hType,
2284	uint64_t uUser, const char *pszDesc)
2285	{
2286	/*
2287	* Assert on some assumption.
2288	*/
2289	VM_ASSERT_EMT(pVM);
2290	AssertReturn(!(cb & GUEST_PAGE_OFFSET_MASK), VERR_INVALID_PARAMETER);
2291	AssertReturn(!(GCPhys & GUEST_PAGE_OFFSET_MASK), VERR_INVALID_PARAMETER);
2292	AssertPtrReturn(pszDesc, VERR_INVALID_POINTER);
2293	AssertReturn(*pszDesc, VERR_INVALID_PARAMETER);
2294	#ifdef VBOX_STRICT
2295	PCPGMPHYSHANDLERTYPEINT pType = pgmHandlerPhysicalTypeHandleToPtr(pVM, hType);
2296	Assert(pType);
2297	Assert(pType->enmKind == PGMPHYSHANDLERKIND_MMIO);
2298	#endif
2299
2300	int rc = PGM_LOCK(pVM);
2301	AssertRCReturn(rc, rc);
2302
2303	/*
2304	* Make sure there's a RAM range structure for the region.
2305	*/
2306	RTGCPHYS GCPhysLast = GCPhys + (cb - 1);
2307	bool fRamExists = false;
2308	PPGMRAMRANGE pRamPrev = NULL;
2309	PPGMRAMRANGE pRam = pVM->pgm.s.pRamRangesXR3;
2310	while (pRam && GCPhysLast >= pRam->GCPhys)
2311	{
2312	if ( GCPhysLast >= pRam->GCPhys
2313	&& GCPhys <= pRam->GCPhysLast)
2314	{
2315	/* Simplification: all within the same range. */
2316	AssertLogRelMsgReturnStmt( GCPhys >= pRam->GCPhys
2317	&& GCPhysLast <= pRam->GCPhysLast,
2318	("%RGp-%RGp (MMIO/%s) falls partly outside %RGp-%RGp (%s)\n",
2319	GCPhys, GCPhysLast, pszDesc,
2320	pRam->GCPhys, pRam->GCPhysLast, pRam->pszDesc),
2321	PGM_UNLOCK(pVM),
2322	VERR_PGM_RAM_CONFLICT);
2323
2324	/* Check that it's all RAM or MMIO pages. */
2325	PCPGMPAGE pPage = &pRam->aPages[(GCPhys - pRam->GCPhys) >> GUEST_PAGE_SHIFT];
2326	uint32_t cLeft = cb >> GUEST_PAGE_SHIFT;
2327	while (cLeft-- > 0)
2328	{
2329	AssertLogRelMsgReturnStmt( PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_RAM
2330	\|\| PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_MMIO,
2331	("%RGp-%RGp (MMIO/%s): %RGp is not a RAM or MMIO page - type=%d desc=%s\n",
2332	GCPhys, GCPhysLast, pszDesc, pRam->GCPhys, PGM_PAGE_GET_TYPE(pPage), pRam->pszDesc),
2333	PGM_UNLOCK(pVM),
2334	VERR_PGM_RAM_CONFLICT);
2335	pPage++;
2336	}
2337
2338	/* Looks good. */
2339	fRamExists = true;
2340	break;
2341	}
2342
2343	/* next */
2344	pRamPrev = pRam;
2345	pRam = pRam->pNextR3;
2346	}
2347	PPGMRAMRANGE pNew;
2348	if (fRamExists)
2349	{
2350	pNew = NULL;
2351
2352	/*
2353	* Make all the pages in the range MMIO/ZERO pages, freeing any
2354	* RAM pages currently mapped here. This might not be 100% correct
2355	* for PCI memory, but we're doing the same thing for MMIO2 pages.
2356	*/
2357	rc = pgmR3PhysFreePageRange(pVM, pRam, GCPhys, GCPhysLast, NULL);
2358	AssertRCReturnStmt(rc, PGM_UNLOCK(pVM), rc);
2359
2360	/* Force a PGM pool flush as guest ram references have been changed. */
2361	/** @todo not entirely SMP safe; assuming for now the guest takes
2362	* care of this internally (not touch mapped mmio while changing the
2363	* mapping). */
2364	PVMCPU pVCpu = VMMGetCpu(pVM);
2365	pVCpu->pgm.s.fSyncFlags \|= PGM_SYNC_CLEAR_PGM_POOL;
2366	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
2367	}
2368	else
2369	{
2370	/*
2371	* No RAM range, insert an ad hoc one.
2372	*
2373	* Note that we don't have to tell REM about this range because
2374	* PGMHandlerPhysicalRegisterEx will do that for us.
2375	*/
2376	Log(("PGMR3PhysMMIORegister: Adding ad hoc MMIO range for %RGp-%RGp %s\n", GCPhys, GCPhysLast, pszDesc));
2377
2378	/* Alloc. */
2379	const uint32_t cPages = cb >> GUEST_PAGE_SHIFT;
2380	const size_t cbRamRange = RT_UOFFSETOF_DYN(PGMRAMRANGE, aPages[cPages]);
2381	const size_t cRangePages = RT_ALIGN_Z(cbRamRange, HOST_PAGE_SIZE) >> HOST_PAGE_SHIFT;
2382	RTR0PTR pNewR0 = NIL_RTR0PTR;
2383	rc = SUPR3PageAllocEx(cRangePages, 0 /fFlags/, (void *)&pNew, &pNewR0, NULL /paPages*/);
2384	AssertLogRelMsgRCReturnStmt(rc, ("cbRamRange=%zu\n", cbRamRange), PGM_UNLOCK(pVM), rc);
2385
2386	#ifdef VBOX_WITH_NATIVE_NEM
2387	/* Notify NEM. */
2388	uint8_t u2State = 0; /* (must have valid state as there can't be anything to preserve) */
2389	if (VM_IS_NEM_ENABLED(pVM))
2390	{
2391	rc = NEMR3NotifyPhysMmioExMapEarly(pVM, GCPhys, cPages << GUEST_PAGE_SHIFT, 0 /fFlags/, NULL, NULL,
2392	&u2State, &pNew->uNemRange);
2393	AssertLogRelRCReturnStmt(rc, SUPR3PageFreeEx(pNew, cRangePages), rc);
2394	}
2395	#endif
2396
2397	/* Initialize the range. */
2398	pNew->pSelfR0 = pNewR0;
2399	pNew->GCPhys = GCPhys;
2400	pNew->GCPhysLast = GCPhysLast;
2401	pNew->cb = cb;
2402	pNew->pszDesc = pszDesc;
2403	pNew->fFlags = PGM_RAM_RANGE_FLAGS_AD_HOC_MMIO;
2404	pNew->pvR3 = NULL;
2405	pNew->paLSPages = NULL;
2406
2407	uint32_t iPage = cPages;
2408	while (iPage-- > 0)
2409	{
2410	PGM_PAGE_INIT_ZERO(&pNew->aPages[iPage], pVM, PGMPAGETYPE_MMIO);
2411	#ifdef VBOX_WITH_NATIVE_NEM
2412	PGM_PAGE_SET_NEM_STATE(&pNew->aPages[iPage], u2State);
2413	#endif
2414	}
2415	Assert(PGM_PAGE_GET_TYPE(&pNew->aPages[0]) == PGMPAGETYPE_MMIO);
2416
2417	/* update the page count stats. */
2418	pVM->pgm.s.cPureMmioPages += cPages;
2419	pVM->pgm.s.cAllPages += cPages;
2420
2421	/* link it */
2422	pgmR3PhysLinkRamRange(pVM, pNew, pRamPrev);
2423	}
2424
2425	/*
2426	* Register the access handler.
2427	*/
2428	rc = PGMHandlerPhysicalRegister(pVM, GCPhys, GCPhysLast, hType, uUser, pszDesc);
2429	if (RT_SUCCESS(rc))
2430	{
2431	#ifdef VBOX_WITH_NATIVE_NEM
2432	/* Late NEM notification. */
2433	if (VM_IS_NEM_ENABLED(pVM))
2434	{
2435	uint32_t const fNemNotify = (fRamExists ? NEM_NOTIFY_PHYS_MMIO_EX_F_REPLACE : 0);
2436	rc = NEMR3NotifyPhysMmioExMapLate(pVM, GCPhys, GCPhysLast - GCPhys + 1, fNemNotify,
2437	fRamExists ? (uint8_t *)pRam->pvR3 + (uintptr_t)(GCPhys - pRam->GCPhys) : NULL,
2438	NULL, !fRamExists ? &pRam->uNemRange : NULL);
2439	AssertLogRelRCReturn(rc, rc);
2440	}
2441	#endif
2442	}
2443	/** @todo the phys handler failure handling isn't complete, esp. wrt NEM. */
2444	else if (!fRamExists)
2445	{
2446	pVM->pgm.s.cPureMmioPages -= cb >> GUEST_PAGE_SHIFT;
2447	pVM->pgm.s.cAllPages -= cb >> GUEST_PAGE_SHIFT;
2448
2449	/* remove the ad hoc range. */
2450	pgmR3PhysUnlinkRamRange2(pVM, pNew, pRamPrev);
2451	pNew->cb = pNew->GCPhys = pNew->GCPhysLast = NIL_RTGCPHYS;
2452	SUPR3PageFreeEx(pRam, RT_ALIGN_Z(RT_UOFFSETOF_DYN(PGMRAMRANGE, aPages[cb >> GUEST_PAGE_SHIFT]),
2453	HOST_PAGE_SIZE) >> HOST_PAGE_SHIFT);
2454	}
2455	pgmPhysInvalidatePageMapTLB(pVM);
2456
2457	PGM_UNLOCK(pVM);
2458	return rc;
2459	}
2460
2461
2462	/**
2463	* This is the interface IOM is using to register an MMIO region.
2464	*
2465	* It will take care of calling PGMHandlerPhysicalDeregister and clean up
2466	* any ad hoc PGMRAMRANGE left behind.
2467	*
2468	* @returns VBox status code.
2469	* @param pVM The cross context VM structure.
2470	* @param GCPhys The start of the MMIO region.
2471	* @param cb The size of the MMIO region.
2472	*/
2473	VMMR3DECL(int) PGMR3PhysMMIODeregister(PVM pVM, RTGCPHYS GCPhys, RTGCPHYS cb)
2474	{
2475	VM_ASSERT_EMT(pVM);
2476
2477	int rc = PGM_LOCK(pVM);
2478	AssertRCReturn(rc, rc);
2479
2480	/*
2481	* First deregister the handler, then check if we should remove the ram range.
2482	*/
2483	rc = PGMHandlerPhysicalDeregister(pVM, GCPhys);
2484	if (RT_SUCCESS(rc))
2485	{
2486	RTGCPHYS GCPhysLast = GCPhys + (cb - 1);
2487	PPGMRAMRANGE pRamPrev = NULL;
2488	PPGMRAMRANGE pRam = pVM->pgm.s.pRamRangesXR3;
2489	while (pRam && GCPhysLast >= pRam->GCPhys)
2490	{
2491	/** @todo We're being a bit too careful here. rewrite. */
2492	if ( GCPhysLast == pRam->GCPhysLast
2493	&& GCPhys == pRam->GCPhys)
2494	{
2495	Assert(pRam->cb == cb);
2496
2497	/*
2498	* See if all the pages are dead MMIO pages.
2499	*/
2500	uint32_t const cGuestPages = cb >> GUEST_PAGE_SHIFT;
2501	bool fAllMMIO = true;
2502	uint32_t iPage = 0;
2503	uint32_t cLeft = cGuestPages;
2504	while (cLeft-- > 0)
2505	{
2506	PPGMPAGE pPage = &pRam->aPages[iPage];
2507	if ( !PGM_PAGE_IS_MMIO_OR_ALIAS(pPage)
2508	/\|\| not-out-of-action later /)
2509	{
2510	fAllMMIO = false;
2511	AssertMsgFailed(("%RGp %R[pgmpage]\n", pRam->GCPhys + ((RTGCPHYS)iPage << GUEST_PAGE_SHIFT), pPage));
2512	break;
2513	}
2514	Assert( PGM_PAGE_IS_ZERO(pPage)
2515	\|\| PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_MMIO2_ALIAS_MMIO
2516	\|\| PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_SPECIAL_ALIAS_MMIO);
2517	pPage++;
2518	}
2519	if (fAllMMIO)
2520	{
2521	/*
2522	* Ad-hoc range, unlink and free it.
2523	*/
2524	Log(("PGMR3PhysMMIODeregister: Freeing ad hoc MMIO range for %RGp-%RGp %s\n",
2525	GCPhys, GCPhysLast, pRam->pszDesc));
2526	/** @todo check the ad-hoc flags? */
2527
2528	#ifdef VBOX_WITH_NATIVE_NEM
2529	if (VM_IS_NEM_ENABLED(pVM)) /* Notify REM before we unlink the range. */
2530	{
2531	rc = NEMR3NotifyPhysMmioExUnmap(pVM, GCPhys, GCPhysLast - GCPhys + 1, 0 /fFlags/,
2532	NULL, NULL, NULL, &pRam->uNemRange);
2533	AssertLogRelRCReturn(rc, rc);
2534	}
2535	#endif
2536
2537	pVM->pgm.s.cAllPages -= cGuestPages;
2538	pVM->pgm.s.cPureMmioPages -= cGuestPages;
2539
2540	pgmR3PhysUnlinkRamRange2(pVM, pRam, pRamPrev);
2541	const uint32_t cPages = pRam->cb >> GUEST_PAGE_SHIFT;
2542	const size_t cbRamRange = RT_UOFFSETOF_DYN(PGMRAMRANGE, aPages[cPages]);
2543	pRam->cb = pRam->GCPhys = pRam->GCPhysLast = NIL_RTGCPHYS;
2544	SUPR3PageFreeEx(pRam, RT_ALIGN_Z(cbRamRange, HOST_PAGE_SIZE) >> HOST_PAGE_SHIFT);
2545	break;
2546	}
2547	}
2548
2549	/*
2550	* Range match? It will all be within one range (see PGMAllHandler.cpp).
2551	*/
2552	if ( GCPhysLast >= pRam->GCPhys
2553	&& GCPhys <= pRam->GCPhysLast)
2554	{
2555	Assert(GCPhys >= pRam->GCPhys);
2556	Assert(GCPhysLast <= pRam->GCPhysLast);
2557
2558	/*
2559	* Turn the pages back into RAM pages.
2560	*/
2561	uint32_t iPage = (GCPhys - pRam->GCPhys) >> GUEST_PAGE_SHIFT;
2562	uint32_t cLeft = cb >> GUEST_PAGE_SHIFT;
2563	while (cLeft--)
2564	{
2565	PPGMPAGE pPage = &pRam->aPages[iPage];
2566	AssertMsg( (PGM_PAGE_IS_MMIO(pPage) && PGM_PAGE_IS_ZERO(pPage))
2567	\|\| PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_MMIO2_ALIAS_MMIO
2568	\|\| PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_SPECIAL_ALIAS_MMIO,
2569	("%RGp %R[pgmpage]\n", pRam->GCPhys + ((RTGCPHYS)iPage << GUEST_PAGE_SHIFT), pPage));
2570	if (PGM_PAGE_IS_MMIO_OR_ALIAS(pPage))
2571	PGM_PAGE_SET_TYPE(pVM, pPage, PGMPAGETYPE_RAM);
2572	iPage++;
2573	}
2574
2575	#ifdef VBOX_WITH_NATIVE_NEM
2576	/* Notify REM (failure will probably leave things in a non-working state). */
2577	if (VM_IS_NEM_ENABLED(pVM))
2578	{
2579	uint8_t u2State = UINT8_MAX;
2580	rc = NEMR3NotifyPhysMmioExUnmap(pVM, GCPhys, GCPhysLast - GCPhys + 1, NEM_NOTIFY_PHYS_MMIO_EX_F_REPLACE,
2581	pRam->pvR3 ? (uint8_t *)pRam->pvR3 + GCPhys - pRam->GCPhys : NULL,
2582	NULL, &u2State, &pRam->uNemRange);
2583	AssertLogRelRCReturn(rc, rc);
2584	if (u2State != UINT8_MAX)
2585	pgmPhysSetNemStateForPages(&pRam->aPages[(GCPhys - pRam->GCPhys) >> GUEST_PAGE_SHIFT],
2586	cb >> GUEST_PAGE_SHIFT, u2State);
2587	}
2588	#endif
2589	break;
2590	}
2591
2592	/* next */
2593	pRamPrev = pRam;
2594	pRam = pRam->pNextR3;
2595	}
2596	}
2597
2598	/* Force a PGM pool flush as guest ram references have been changed. */
2599	/** @todo Not entirely SMP safe; assuming for now the guest takes care of
2600	* this internally (not touch mapped mmio while changing the mapping). */
2601	PVMCPU pVCpu = VMMGetCpu(pVM);
2602	pVCpu->pgm.s.fSyncFlags \|= PGM_SYNC_CLEAR_PGM_POOL;
2603	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
2604
2605	pgmPhysInvalidatePageMapTLB(pVM);
2606	pgmPhysInvalidRamRangeTlbs(pVM);
2607	PGM_UNLOCK(pVM);
2608	return rc;
2609	}
2610
2611
2612
2613	/*********************************************************************************************************************************
2614	* MMIO2 *
2615	*********************************************************************************************************************************/
2616
2617	/**
2618	* Locate a MMIO2 range.
2619	*
2620	* @returns Pointer to the MMIO2 range.
2621	* @param pVM The cross context VM structure.
2622	* @param pDevIns The device instance owning the region.
2623	* @param iSubDev The sub-device number.
2624	* @param iRegion The region.
2625	* @param hMmio2 Handle to look up. If NIL, use the @a iSubDev and
2626	* @a iRegion.
2627	*/
2628	DECLINLINE(PPGMREGMMIO2RANGE) pgmR3PhysMmio2Find(PVM pVM, PPDMDEVINS pDevIns, uint32_t iSubDev,
2629	uint32_t iRegion, PGMMMIO2HANDLE hMmio2)
2630	{
2631	if (hMmio2 != NIL_PGMMMIO2HANDLE)
2632	{
2633	if (hMmio2 <= RT_ELEMENTS(pVM->pgm.s.apMmio2RangesR3) && hMmio2 != 0)
2634	{
2635	PPGMREGMMIO2RANGE pCur = pVM->pgm.s.apMmio2RangesR3[hMmio2 - 1];
2636	if (pCur && pCur->pDevInsR3 == pDevIns)
2637	{
2638	Assert(pCur->idMmio2 == hMmio2);
2639	AssertReturn(pCur->fFlags & PGMREGMMIO2RANGE_F_FIRST_CHUNK, NULL);
2640	return pCur;
2641	}
2642	Assert(!pCur);
2643	}
2644	for (PPGMREGMMIO2RANGE pCur = pVM->pgm.s.pRegMmioRangesR3; pCur; pCur = pCur->pNextR3)
2645	if (pCur->idMmio2 == hMmio2)
2646	{
2647	AssertBreak(pCur->pDevInsR3 == pDevIns);
2648	AssertReturn(pCur->fFlags & PGMREGMMIO2RANGE_F_FIRST_CHUNK, NULL);
2649	return pCur;
2650	}
2651	}
2652	else
2653	{
2654	/*
2655	* Search the list. There shouldn't be many entries.
2656	*/
2657	/** @todo Optimize this lookup! There may now be many entries and it'll
2658	* become really slow when doing MMR3HyperMapMMIO2 and similar. */
2659	for (PPGMREGMMIO2RANGE pCur = pVM->pgm.s.pRegMmioRangesR3; pCur; pCur = pCur->pNextR3)
2660	if ( pCur->pDevInsR3 == pDevIns
2661	&& pCur->iRegion == iRegion
2662	&& pCur->iSubDev == iSubDev)
2663	return pCur;
2664	}
2665	return NULL;
2666	}
2667
2668
2669	/**
2670	* Worker for PGMR3PhysMmio2ControlDirtyPageTracking and PGMR3PhysMmio2Map.
2671	*/
2672	static int pgmR3PhysMmio2EnableDirtyPageTracing(PVM pVM, PPGMREGMMIO2RANGE pFirstMmio2)
2673	{
2674	int rc = VINF_SUCCESS;
2675	for (PPGMREGMMIO2RANGE pCurMmio2 = pFirstMmio2; pCurMmio2; pCurMmio2 = pCurMmio2->pNextR3)
2676	{
2677	Assert(!(pCurMmio2->fFlags & PGMREGMMIO2RANGE_F_IS_TRACKING));
2678	int rc2 = pgmHandlerPhysicalExRegister(pVM, pCurMmio2->pPhysHandlerR3, pCurMmio2->RamRange.GCPhys,
2679	pCurMmio2->RamRange.GCPhysLast);
2680	AssertLogRelMsgRC(rc2, ("%#RGp-%#RGp %s failed -> %Rrc\n", pCurMmio2->RamRange.GCPhys, pCurMmio2->RamRange.GCPhysLast,
2681	pCurMmio2->RamRange.pszDesc, rc2));
2682	if (RT_SUCCESS(rc2))
2683	pCurMmio2->fFlags \|= PGMREGMMIO2RANGE_F_IS_TRACKING;
2684	else if (RT_SUCCESS(rc))
2685	rc = rc2;
2686	if (pCurMmio2->fFlags & PGMREGMMIO2RANGE_F_LAST_CHUNK)
2687	return rc;
2688	}
2689	AssertFailed();
2690	return rc;
2691	}
2692
2693
2694	/**
2695	* Worker for PGMR3PhysMmio2ControlDirtyPageTracking and PGMR3PhysMmio2Unmap.
2696	*/
2697	static int pgmR3PhysMmio2DisableDirtyPageTracing(PVM pVM, PPGMREGMMIO2RANGE pFirstMmio2)
2698	{
2699	for (PPGMREGMMIO2RANGE pCurMmio2 = pFirstMmio2; pCurMmio2; pCurMmio2 = pCurMmio2->pNextR3)
2700	{
2701	if (pCurMmio2->fFlags & PGMREGMMIO2RANGE_F_IS_TRACKING)
2702	{
2703	int rc2 = pgmHandlerPhysicalExDeregister(pVM, pCurMmio2->pPhysHandlerR3);
2704	AssertLogRelMsgRC(rc2, ("%#RGp-%#RGp %s failed -> %Rrc\n", pCurMmio2->RamRange.GCPhys, pCurMmio2->RamRange.GCPhysLast,
2705	pCurMmio2->RamRange.pszDesc, rc2));
2706	pCurMmio2->fFlags &= ~PGMREGMMIO2RANGE_F_IS_TRACKING;
2707	}
2708	if (pCurMmio2->fFlags & PGMREGMMIO2RANGE_F_LAST_CHUNK)
2709	return VINF_SUCCESS;
2710	}
2711	AssertFailed();
2712	return VINF_SUCCESS;
2713
2714	}
2715
2716
2717	/**
2718	* Calculates the number of chunks
2719	*
2720	* @returns Number of registration chunk needed.
2721	* @param pVM The cross context VM structure.
2722	* @param cb The size of the MMIO/MMIO2 range.
2723	* @param pcPagesPerChunk Where to return the number of pages tracked by each
2724	* chunk. Optional.
2725	* @param pcbChunk Where to return the guest mapping size for a chunk.
2726	*/
2727	static uint16_t pgmR3PhysMmio2CalcChunkCount(PVM pVM, RTGCPHYS cb, uint32_t pcPagesPerChunk, uint32_t pcbChunk)
2728	{
2729	RT_NOREF_PV(pVM); /* without raw mode */
2730
2731	/*
2732	* This is the same calculation as PGMR3PhysRegisterRam does, except we'll be
2733	* needing a few bytes extra the PGMREGMMIO2RANGE structure.
2734	*
2735	* Note! In additions, we've got a 24 bit sub-page range for MMIO2 ranges, leaving
2736	* us with an absolute maximum of 16777215 pages per chunk (close to 64 GB).
2737	*/
2738	uint32_t const cPagesPerChunk = _4M;
2739	Assert(RT_ALIGN_32(cPagesPerChunk, X86_PD_PAE_SHIFT - X86_PAGE_SHIFT)); /* NEM large page requirement: 1GB pages. */
2740	uint32_t const cbChunk = RT_UOFFSETOF_DYN(PGMREGMMIO2RANGE, RamRange.aPages[cPagesPerChunk]);
2741	AssertRelease(cPagesPerChunk < _16M);
2742
2743	if (pcbChunk)
2744	*pcbChunk = cbChunk;
2745	if (pcPagesPerChunk)
2746	*pcPagesPerChunk = cPagesPerChunk;
2747
2748	/* Calc the number of chunks we need. */
2749	RTGCPHYS const cGuestPages = cb >> GUEST_PAGE_SHIFT;
2750	uint16_t cChunks = (uint16_t)((cGuestPages + cPagesPerChunk - 1) / cPagesPerChunk);
2751	AssertRelease((RTGCPHYS)cChunks * cPagesPerChunk >= cGuestPages);
2752	return cChunks;
2753	}
2754
2755
2756	/**
2757	* Worker for PGMR3PhysMMIO2Register that allocates and the PGMREGMMIO2RANGE
2758	* structures and does basic initialization.
2759	*
2760	* Caller must set type specfic members and initialize the PGMPAGE structures.
2761	*
2762	* This was previously also used by PGMR3PhysMmio2PreRegister, a function for
2763	* pre-registering MMIO that was later (6.1) replaced by a new handle based IOM
2764	* interface. The reference to caller and type above is purely historical.
2765	*
2766	* @returns VBox status code.
2767	* @param pVM The cross context VM structure.
2768	* @param pDevIns The device instance owning the region.
2769	* @param iSubDev The sub-device number (internal PCI config number).
2770	* @param iRegion The region number. If the MMIO2 memory is a PCI
2771	* I/O region this number has to be the number of that
2772	* region. Otherwise it can be any number safe
2773	* UINT8_MAX.
2774	* @param cb The size of the region. Must be page aligned.
2775	* @param fFlags PGMPHYS_MMIO2_FLAGS_XXX.
2776	* @param idMmio2 The MMIO2 ID for the first chunk.
2777	* @param pszDesc The description.
2778	* @param ppHeadRet Where to return the pointer to the first
2779	* registration chunk.
2780	*
2781	* @thread EMT
2782	*/
2783	static int pgmR3PhysMmio2Create(PVM pVM, PPDMDEVINS pDevIns, uint32_t iSubDev, uint32_t iRegion, RTGCPHYS cb, uint32_t fFlags,
2784	uint8_t idMmio2, const char pszDesc, PPGMREGMMIO2RANGE ppHeadRet)
2785	{
2786	/*
2787	* Figure out how many chunks we need and of which size.
2788	*/
2789	uint32_t cPagesPerChunk;
2790	uint16_t cChunks = pgmR3PhysMmio2CalcChunkCount(pVM, cb, &cPagesPerChunk, NULL);
2791	AssertReturn(cChunks, VERR_PGM_PHYS_MMIO_EX_IPE);
2792
2793	/*
2794	* Allocate the chunks.
2795	*/
2796	PPGMREGMMIO2RANGE *ppNext = ppHeadRet;
2797	*ppNext = NULL;
2798
2799	int rc = VINF_SUCCESS;
2800	uint32_t cPagesLeft = cb >> GUEST_PAGE_SHIFT;
2801	for (uint16_t iChunk = 0; iChunk < cChunks && RT_SUCCESS(rc); iChunk++, idMmio2++)
2802	{
2803	/*
2804	* We currently do a single RAM range for the whole thing. This will
2805	* probably have to change once someone needs really large MMIO regions,
2806	* as we will be running into SUPR3PageAllocEx limitations and such.
2807	*/
2808	const uint32_t cPagesTrackedByChunk = RT_MIN(cPagesLeft, cPagesPerChunk);
2809	const size_t cbRange = RT_UOFFSETOF_DYN(PGMREGMMIO2RANGE, RamRange.aPages[cPagesTrackedByChunk]);
2810	PPGMREGMMIO2RANGE pNew = NULL;
2811
2812	/*
2813	* Allocate memory for the registration structure.
2814	*/
2815	size_t const cChunkPages = RT_ALIGN_Z(cbRange, HOST_PAGE_SIZE) >> HOST_PAGE_SHIFT;
2816	size_t const cbChunk = (1 + cChunkPages + 1) << HOST_PAGE_SHIFT;
2817	AssertLogRelBreakStmt(cbChunk == (uint32_t)cbChunk, rc = VERR_OUT_OF_RANGE);
2818	RTR0PTR R0PtrChunk = NIL_RTR0PTR;
2819	void *pvChunk = NULL;
2820	rc = SUPR3PageAllocEx(cChunkPages, 0 /fFlags/, &pvChunk, &R0PtrChunk, NULL /paPages/);
2821	AssertLogRelMsgRCBreak(rc, ("rc=%Rrc, cChunkPages=%#zx\n", rc, cChunkPages));
2822
2823	Assert(R0PtrChunk != NIL_RTR0PTR \|\| PGM_IS_IN_NEM_MODE(pVM));
2824	RT_BZERO(pvChunk, cChunkPages << HOST_PAGE_SHIFT);
2825
2826	pNew = (PPGMREGMMIO2RANGE)pvChunk;
2827	pNew->RamRange.fFlags = PGM_RAM_RANGE_FLAGS_FLOATING;
2828	pNew->RamRange.pSelfR0 = R0PtrChunk + RT_UOFFSETOF(PGMREGMMIO2RANGE, RamRange);
2829
2830	/*
2831	* Initialize the registration structure (caller does specific bits).
2832	*/
2833	pNew->pDevInsR3 = pDevIns;
2834	//pNew->pvR3 = NULL;
2835	//pNew->pNext = NULL;
2836	if (iChunk == 0)
2837	pNew->fFlags \|= PGMREGMMIO2RANGE_F_FIRST_CHUNK;
2838	if (iChunk + 1 == cChunks)
2839	pNew->fFlags \|= PGMREGMMIO2RANGE_F_LAST_CHUNK;
2840	if (fFlags & PGMPHYS_MMIO2_FLAGS_TRACK_DIRTY_PAGES)
2841	pNew->fFlags \|= PGMREGMMIO2RANGE_F_TRACK_DIRTY_PAGES;
2842	pNew->iSubDev = iSubDev;
2843	pNew->iRegion = iRegion;
2844	pNew->idSavedState = UINT8_MAX;
2845	pNew->idMmio2 = idMmio2;
2846	//pNew->pPhysHandlerR3 = NULL;
2847	//pNew->paLSPages = NULL;
2848	pNew->RamRange.GCPhys = NIL_RTGCPHYS;
2849	pNew->RamRange.GCPhysLast = NIL_RTGCPHYS;
2850	pNew->RamRange.pszDesc = pszDesc;
2851	pNew->RamRange.cb = pNew->cbReal = (RTGCPHYS)cPagesTrackedByChunk << X86_PAGE_SHIFT;
2852	pNew->RamRange.fFlags \|= PGM_RAM_RANGE_FLAGS_AD_HOC_MMIO_EX;
2853	pNew->RamRange.uNemRange = UINT32_MAX;
2854	//pNew->RamRange.pvR3 = NULL;
2855	//pNew->RamRange.paLSPages = NULL;
2856
2857	*ppNext = pNew;
2858	ASMCompilerBarrier();
2859	cPagesLeft -= cPagesTrackedByChunk;
2860	ppNext = &pNew->pNextR3;
2861
2862	/*
2863	* Pre-allocate a handler if we're tracking dirty pages, unless NEM takes care of this.
2864	*/
2865	if ( (fFlags & PGMPHYS_MMIO2_FLAGS_TRACK_DIRTY_PAGES)
2866	#ifdef VBOX_WITH_PGM_NEM_MODE
2867	&& (!VM_IS_NEM_ENABLED(pVM) \|\| !NEMR3IsMmio2DirtyPageTrackingSupported(pVM))
2868	#endif
2869	)
2870
2871	{
2872	rc = pgmHandlerPhysicalExCreate(pVM, pVM->pgm.s.hMmio2DirtyPhysHandlerType, idMmio2, pszDesc, &pNew->pPhysHandlerR3);
2873	AssertLogRelMsgRCBreak(rc, ("idMmio2=%zu\n", idMmio2));
2874	}
2875	}
2876	Assert(cPagesLeft == 0);
2877
2878	if (RT_SUCCESS(rc))
2879	{
2880	Assert((*ppHeadRet)->fFlags & PGMREGMMIO2RANGE_F_FIRST_CHUNK);
2881	return VINF_SUCCESS;
2882	}
2883
2884	/*
2885	* Free floating ranges.
2886	*/
2887	while (*ppHeadRet)
2888	{
2889	PPGMREGMMIO2RANGE pFree = *ppHeadRet;
2890	*ppHeadRet = pFree->pNextR3;
2891
2892	if (pFree->pPhysHandlerR3)
2893	{
2894	pgmHandlerPhysicalExDestroy(pVM, pFree->pPhysHandlerR3);
2895	pFree->pPhysHandlerR3 = NULL;
2896	}
2897
2898	if (pFree->RamRange.fFlags & PGM_RAM_RANGE_FLAGS_FLOATING)
2899	{
2900	const size_t cbRange = RT_UOFFSETOF_DYN(PGMREGMMIO2RANGE,
2901	RamRange.aPages[pFree->RamRange.cb >> X86_PAGE_SHIFT]);
2902	size_t const cChunkPages = RT_ALIGN_Z(cbRange, HOST_PAGE_SIZE) >> HOST_PAGE_SHIFT;
2903	SUPR3PageFreeEx(pFree, cChunkPages);
2904	}
2905	}
2906
2907	return rc;
2908	}
2909
2910
2911	/**
2912	* Common worker PGMR3PhysMmio2PreRegister & PGMR3PhysMMIO2Register that links a
2913	* complete registration entry into the lists and lookup tables.
2914	*
2915	* @param pVM The cross context VM structure.
2916	* @param pNew The new MMIO / MMIO2 registration to link.
2917	*/
2918	static void pgmR3PhysMmio2Link(PVM pVM, PPGMREGMMIO2RANGE pNew)
2919	{
2920	Assert(pNew->idMmio2 != UINT8_MAX);
2921
2922	/*
2923	* Link it into the list (order doesn't matter, so insert it at the head).
2924	*
2925	* Note! The range we're linking may consist of multiple chunks, so we
2926	* have to find the last one.
2927	*/
2928	PPGMREGMMIO2RANGE pLast = pNew;
2929	for (pLast = pNew; ; pLast = pLast->pNextR3)
2930	{
2931	if (pLast->fFlags & PGMREGMMIO2RANGE_F_LAST_CHUNK)
2932	break;
2933	Assert(pLast->pNextR3);
2934	Assert(pLast->pNextR3->pDevInsR3 == pNew->pDevInsR3);
2935	Assert(pLast->pNextR3->iSubDev == pNew->iSubDev);
2936	Assert(pLast->pNextR3->iRegion == pNew->iRegion);
2937	Assert(pLast->pNextR3->idMmio2 == pLast->idMmio2 + 1);
2938	}
2939
2940	PGM_LOCK_VOID(pVM);
2941
2942	/* Link in the chain of ranges at the head of the list. */
2943	pLast->pNextR3 = pVM->pgm.s.pRegMmioRangesR3;
2944	pVM->pgm.s.pRegMmioRangesR3 = pNew;
2945
2946	/* Insert the MMIO2 range/page IDs. */
2947	uint8_t idMmio2 = pNew->idMmio2;
2948	for (;;)
2949	{
2950	Assert(pVM->pgm.s.apMmio2RangesR3[idMmio2 - 1] == NULL);
2951	Assert(pVM->pgm.s.apMmio2RangesR0[idMmio2 - 1] == NIL_RTR0PTR);
2952	pVM->pgm.s.apMmio2RangesR3[idMmio2 - 1] = pNew;
2953	pVM->pgm.s.apMmio2RangesR0[idMmio2 - 1] = pNew->RamRange.pSelfR0 - RT_UOFFSETOF(PGMREGMMIO2RANGE, RamRange);
2954	if (pNew->fFlags & PGMREGMMIO2RANGE_F_LAST_CHUNK)
2955	break;
2956	pNew = pNew->pNextR3;
2957	idMmio2++;
2958	}
2959
2960	pgmPhysInvalidatePageMapTLB(pVM);
2961	PGM_UNLOCK(pVM);
2962	}
2963
2964
2965	/**
2966	* Allocate and register an MMIO2 region.
2967	*
2968	* As mentioned elsewhere, MMIO2 is just RAM spelled differently. It's RAM
2969	* associated with a device. It is also non-shared memory with a permanent
2970	* ring-3 mapping and page backing (presently).
2971	*
2972	* A MMIO2 range may overlap with base memory if a lot of RAM is configured for
2973	* the VM, in which case we'll drop the base memory pages. Presently we will
2974	* make no attempt to preserve anything that happens to be present in the base
2975	* memory that is replaced, this is of course incorrect but it's too much
2976	* effort.
2977	*
2978	* @returns VBox status code.
2979	* @retval VINF_SUCCESS on success, *ppv pointing to the R3 mapping of the
2980	* memory.
2981	* @retval VERR_ALREADY_EXISTS if the region already exists.
2982	*
2983	* @param pVM The cross context VM structure.
2984	* @param pDevIns The device instance owning the region.
2985	* @param iSubDev The sub-device number.
2986	* @param iRegion The region number. If the MMIO2 memory is a PCI
2987	* I/O region this number has to be the number of that
2988	* region. Otherwise it can be any number save
2989	* UINT8_MAX.
2990	* @param cb The size of the region. Must be page aligned.
2991	* @param fFlags Reserved for future use, must be zero.
2992	* @param pszDesc The description.
2993	* @param ppv Where to store the pointer to the ring-3 mapping of
2994	* the memory.
2995	* @param phRegion Where to return the MMIO2 region handle. Optional.
2996	* @thread EMT
2997	*/
2998	VMMR3_INT_DECL(int) PGMR3PhysMmio2Register(PVM pVM, PPDMDEVINS pDevIns, uint32_t iSubDev, uint32_t iRegion, RTGCPHYS cb,
2999	uint32_t fFlags, const char pszDesc, void ppv, PGMMMIO2HANDLE phRegion)
3000	{
3001	/*
3002	* Validate input.
3003	*/
3004	AssertPtrReturn(ppv, VERR_INVALID_POINTER);
3005	*ppv = NULL;
3006	if (phRegion)
3007	{
3008	AssertPtrReturn(phRegion, VERR_INVALID_POINTER);
3009	*phRegion = NIL_PGMMMIO2HANDLE;
3010	}
3011	VM_ASSERT_EMT_RETURN(pVM, VERR_VM_THREAD_NOT_EMT);
3012	AssertPtrReturn(pDevIns, VERR_INVALID_PARAMETER);
3013	AssertReturn(iSubDev <= UINT8_MAX, VERR_INVALID_PARAMETER);
3014	AssertReturn(iRegion <= UINT8_MAX, VERR_INVALID_PARAMETER);
3015	AssertPtrReturn(pszDesc, VERR_INVALID_POINTER);
3016	AssertReturn(*pszDesc, VERR_INVALID_PARAMETER);
3017	AssertReturn(pgmR3PhysMmio2Find(pVM, pDevIns, iSubDev, iRegion, NIL_PGMMMIO2HANDLE) == NULL, VERR_ALREADY_EXISTS);
3018	AssertReturn(!(cb & GUEST_PAGE_OFFSET_MASK), VERR_INVALID_PARAMETER);
3019	AssertReturn(cb, VERR_INVALID_PARAMETER);
3020	AssertReturn(!(fFlags & ~PGMPHYS_MMIO2_FLAGS_VALID_MASK), VERR_INVALID_FLAGS);
3021
3022	const uint32_t cGuestPages = cb >> GUEST_PAGE_SHIFT;
3023	AssertLogRelReturn(((RTGCPHYS)cGuestPages << GUEST_PAGE_SHIFT) == cb, VERR_INVALID_PARAMETER);
3024	AssertLogRelReturn(cGuestPages <= (MM_MMIO_64_MAX >> X86_PAGE_SHIFT), VERR_OUT_OF_RANGE);
3025	AssertLogRelReturn(cGuestPages <= PGM_MMIO2_MAX_PAGE_COUNT, VERR_OUT_OF_RANGE);
3026
3027	/*
3028	* For the 2nd+ instance, mangle the description string so it's unique.
3029	*/
3030	if (pDevIns->iInstance > 0) /** @todo Move to PDMDevHlp.cpp and use a real string cache. */
3031	{
3032	pszDesc = MMR3HeapAPrintf(pVM, MM_TAG_PGM_PHYS, "%s [%u]", pszDesc, pDevIns->iInstance);
3033	if (!pszDesc)
3034	return VERR_NO_MEMORY;
3035	}
3036
3037	/*
3038	* Allocate an MMIO2 range ID (not freed on failure).
3039	*
3040	* The zero ID is not used as it could be confused with NIL_GMM_PAGEID, so
3041	* the IDs goes from 1 thru PGM_MMIO2_MAX_RANGES.
3042	*/
3043	unsigned cChunks = pgmR3PhysMmio2CalcChunkCount(pVM, cb, NULL, NULL);
3044
3045	PGM_LOCK_VOID(pVM);
3046	AssertCompile(PGM_MMIO2_MAX_RANGES < 255);
3047	uint8_t const idMmio2 = pVM->pgm.s.cMmio2Regions + 1;
3048	unsigned const cNewMmio2Regions = pVM->pgm.s.cMmio2Regions + cChunks;
3049	if (cNewMmio2Regions > PGM_MMIO2_MAX_RANGES)
3050	{
3051	PGM_UNLOCK(pVM);
3052	AssertLogRelFailedReturn(VERR_PGM_TOO_MANY_MMIO2_RANGES);
3053	}
3054	pVM->pgm.s.cMmio2Regions = cNewMmio2Regions;
3055	PGM_UNLOCK(pVM);
3056
3057	/*
3058	* Try reserve and allocate the backing memory first as this is what is
3059	* most likely to fail.
3060	*/
3061	int rc = MMR3AdjustFixedReservation(pVM, cGuestPages, pszDesc);
3062	if (RT_SUCCESS(rc))
3063	{
3064	const uint32_t cHostPages = RT_ALIGN_T(cb, HOST_PAGE_SIZE, RTGCPHYS) >> HOST_PAGE_SHIFT;
3065	PSUPPAGE paPages = (PSUPPAGE)RTMemTmpAlloc(cHostPages * sizeof(SUPPAGE));
3066	if (RT_SUCCESS(rc))
3067	{
3068	void *pvPages = NULL;
3069	#ifndef VBOX_WITH_LINEAR_HOST_PHYS_MEM
3070	RTR0PTR pvPagesR0 = NIL_RTR0PTR;
3071	#endif
3072	#ifdef VBOX_WITH_PGM_NEM_MODE
3073	if (PGM_IS_IN_NEM_MODE(pVM))
3074	rc = SUPR3PageAlloc(cHostPages, pVM->pgm.s.fUseLargePages ? SUP_PAGE_ALLOC_F_LARGE_PAGES : 0, &pvPages);
3075	else
3076	#endif
3077	{
3078	#ifndef VBOX_WITH_LINEAR_HOST_PHYS_MEM
3079	rc = SUPR3PageAllocEx(cHostPages, 0 /fFlags/, &pvPages, &pvPagesR0, paPages);
3080	#else
3081	rc = SUPR3PageAllocEx(cHostPages, 0 /fFlags/, &pvPages, NULL /pR0Ptr/, paPages);
3082	#endif
3083	}
3084	if (RT_SUCCESS(rc))
3085	{
3086	memset(pvPages, 0, cGuestPages * GUEST_PAGE_SIZE);
3087
3088	/*
3089	* Create the registered MMIO range record for it.
3090	*/
3091	PPGMREGMMIO2RANGE pNew;
3092	rc = pgmR3PhysMmio2Create(pVM, pDevIns, iSubDev, iRegion, cb, fFlags, idMmio2, pszDesc, &pNew);
3093	if (RT_SUCCESS(rc))
3094	{
3095	if (phRegion)
3096	phRegion = idMmio2; / The ID of the first chunk. */
3097
3098	uint32_t iSrcPage = 0;
3099	uint8_t pbCurPages = (uint8_t )pvPages;
3100	for (PPGMREGMMIO2RANGE pCur = pNew; pCur; pCur = pCur->pNextR3)
3101	{
3102	pCur->pvR3 = pbCurPages;
3103	#ifndef VBOX_WITH_LINEAR_HOST_PHYS_MEM
3104	pCur->pvR0 = pvPagesR0 + (iSrcPage << GUEST_PAGE_SHIFT);
3105	#endif
3106	pCur->RamRange.pvR3 = pbCurPages;
3107
3108	uint32_t iDstPage = pCur->RamRange.cb >> GUEST_PAGE_SHIFT;
3109	#ifdef VBOX_WITH_PGM_NEM_MODE
3110	if (PGM_IS_IN_NEM_MODE(pVM))
3111	while (iDstPage-- > 0)
3112	PGM_PAGE_INIT(&pNew->RamRange.aPages[iDstPage], UINT64_C(0x0000ffffffff0000),
3113	PGM_MMIO2_PAGEID_MAKE(idMmio2, iDstPage),
3114	PGMPAGETYPE_MMIO2, PGM_PAGE_STATE_ALLOCATED);
3115	else
3116	#endif
3117	{
3118	AssertRelease(HOST_PAGE_SHIFT == GUEST_PAGE_SHIFT);
3119	while (iDstPage-- > 0)
3120	PGM_PAGE_INIT(&pNew->RamRange.aPages[iDstPage], paPages[iDstPage + iSrcPage].Phys,
3121	PGM_MMIO2_PAGEID_MAKE(idMmio2, iDstPage),
3122	PGMPAGETYPE_MMIO2, PGM_PAGE_STATE_ALLOCATED);
3123	}
3124
3125	/* advance. */
3126	iSrcPage += pCur->RamRange.cb >> GUEST_PAGE_SHIFT;
3127	pbCurPages += pCur->RamRange.cb;
3128	}
3129
3130	RTMemTmpFree(paPages);
3131
3132	/*
3133	* Update the page count stats, link the registration and we're done.
3134	*/
3135	pVM->pgm.s.cAllPages += cGuestPages;
3136	pVM->pgm.s.cPrivatePages += cGuestPages;
3137
3138	pgmR3PhysMmio2Link(pVM, pNew);
3139
3140	*ppv = pvPages;
3141	return VINF_SUCCESS;
3142	}
3143
3144	SUPR3PageFreeEx(pvPages, cHostPages);
3145	}
3146	}
3147	RTMemTmpFree(paPages);
3148	MMR3AdjustFixedReservation(pVM, -(int32_t)cGuestPages, pszDesc);
3149	}
3150	if (pDevIns->iInstance > 0)
3151	MMR3HeapFree((void *)pszDesc);
3152	return rc;
3153	}
3154
3155
3156	/**
3157	* Deregisters and frees an MMIO2 region.
3158	*
3159	* Any physical access handlers registered for the region must be deregistered
3160	* before calling this function.
3161	*
3162	* @returns VBox status code.
3163	* @param pVM The cross context VM structure.
3164	* @param pDevIns The device instance owning the region.
3165	* @param hMmio2 The MMIO2 handle to deregister, or NIL if all
3166	* regions for the given device is to be deregistered.
3167	*/
3168	VMMR3_INT_DECL(int) PGMR3PhysMmio2Deregister(PVM pVM, PPDMDEVINS pDevIns, PGMMMIO2HANDLE hMmio2)
3169	{
3170	/*
3171	* Validate input.
3172	*/
3173	VM_ASSERT_EMT_RETURN(pVM, VERR_VM_THREAD_NOT_EMT);
3174	AssertPtrReturn(pDevIns, VERR_INVALID_PARAMETER);
3175
3176	/*
3177	* The loop here scanning all registrations will make sure that multi-chunk ranges
3178	* get properly deregistered, though it's original purpose was the wildcard iRegion.
3179	*/
3180	PGM_LOCK_VOID(pVM);
3181	int rc = VINF_SUCCESS;
3182	unsigned cFound = 0;
3183	PPGMREGMMIO2RANGE pPrev = NULL;
3184	PPGMREGMMIO2RANGE pCur = pVM->pgm.s.pRegMmioRangesR3;
3185	while (pCur)
3186	{
3187	uint32_t const fFlags = pCur->fFlags;
3188	if ( pCur->pDevInsR3 == pDevIns
3189	&& ( hMmio2 == NIL_PGMMMIO2HANDLE
3190	\|\| pCur->idMmio2 == hMmio2))
3191	{
3192	cFound++;
3193
3194	/*
3195	* Unmap it if it's mapped.
3196	*/
3197	if (fFlags & PGMREGMMIO2RANGE_F_MAPPED)
3198	{
3199	int rc2 = PGMR3PhysMmio2Unmap(pVM, pCur->pDevInsR3, pCur->idMmio2, pCur->RamRange.GCPhys);
3200	AssertRC(rc2);
3201	if (RT_FAILURE(rc2) && RT_SUCCESS(rc))
3202	rc = rc2;
3203	}
3204
3205	/*
3206	* Unlink it
3207	*/
3208	PPGMREGMMIO2RANGE pNext = pCur->pNextR3;
3209	if (pPrev)
3210	pPrev->pNextR3 = pNext;
3211	else
3212	pVM->pgm.s.pRegMmioRangesR3 = pNext;
3213	pCur->pNextR3 = NULL;
3214
3215	uint8_t idMmio2 = pCur->idMmio2;
3216	Assert(idMmio2 <= RT_ELEMENTS(pVM->pgm.s.apMmio2RangesR3));
3217	if (idMmio2 <= RT_ELEMENTS(pVM->pgm.s.apMmio2RangesR3))
3218	{
3219	Assert(pVM->pgm.s.apMmio2RangesR3[idMmio2 - 1] == pCur);
3220	pVM->pgm.s.apMmio2RangesR3[idMmio2 - 1] = NULL;
3221	pVM->pgm.s.apMmio2RangesR0[idMmio2 - 1] = NIL_RTR0PTR;
3222	}
3223
3224	/*
3225	* Free the memory.
3226	*/
3227	uint32_t const cGuestPages = pCur->cbReal >> GUEST_PAGE_SHIFT;
3228	uint32_t const cHostPages = RT_ALIGN_T(pCur->cbReal, HOST_PAGE_SIZE, RTGCPHYS) >> HOST_PAGE_SHIFT;
3229	#ifdef VBOX_WITH_PGM_NEM_MODE
3230	if (!pVM->pgm.s.fNemMode)
3231	#endif
3232	{
3233	int rc2 = SUPR3PageFreeEx(pCur->pvR3, cHostPages);
3234	AssertRC(rc2);
3235	if (RT_FAILURE(rc2) && RT_SUCCESS(rc))
3236	rc = rc2;
3237
3238	rc2 = MMR3AdjustFixedReservation(pVM, -(int32_t)cGuestPages, pCur->RamRange.pszDesc);
3239	AssertRC(rc2);
3240	if (RT_FAILURE(rc2) && RT_SUCCESS(rc))
3241	rc = rc2;
3242	}
3243	#ifdef VBOX_WITH_PGM_NEM_MODE
3244	else
3245	{
3246	int rc2 = SUPR3PageFreeEx(pCur->pvR3, cHostPages);
3247	AssertRC(rc2);
3248	if (RT_FAILURE(rc2) && RT_SUCCESS(rc))
3249	rc = rc2;
3250	}
3251	#endif
3252
3253	if (pCur->pPhysHandlerR3)
3254	{
3255	pgmHandlerPhysicalExDestroy(pVM, pCur->pPhysHandlerR3);
3256	pCur->pPhysHandlerR3 = NULL;
3257	}
3258
3259	/* we're leaking hyper memory here if done at runtime. */
3260	#ifdef VBOX_STRICT
3261	VMSTATE const enmState = VMR3GetState(pVM);
3262	AssertMsg( enmState == VMSTATE_POWERING_OFF
3263	\|\| enmState == VMSTATE_POWERING_OFF_LS
3264	\|\| enmState == VMSTATE_OFF
3265	\|\| enmState == VMSTATE_OFF_LS
3266	\|\| enmState == VMSTATE_DESTROYING
3267	\|\| enmState == VMSTATE_TERMINATED
3268	\|\| enmState == VMSTATE_CREATING
3269	, ("%s\n", VMR3GetStateName(enmState)));
3270	#endif
3271
3272	if (pCur->RamRange.fFlags & PGM_RAM_RANGE_FLAGS_FLOATING)
3273	{
3274	const size_t cbRange = RT_UOFFSETOF_DYN(PGMREGMMIO2RANGE, RamRange.aPages[cGuestPages]);
3275	size_t const cChunkPages = RT_ALIGN_Z(cbRange, HOST_PAGE_SIZE) >> HOST_PAGE_SHIFT;
3276	SUPR3PageFreeEx(pCur, cChunkPages);
3277	}
3278	/*else
3279	{
3280	rc = MMHyperFree(pVM, pCur); - does not work, see the alloc call.
3281	AssertRCReturn(rc, rc);
3282	} */
3283
3284
3285	/* update page count stats */
3286	pVM->pgm.s.cAllPages -= cGuestPages;
3287	pVM->pgm.s.cPrivatePages -= cGuestPages;
3288
3289	/* next */
3290	pCur = pNext;
3291	if (hMmio2 != NIL_PGMMMIO2HANDLE)
3292	{
3293	if (fFlags & PGMREGMMIO2RANGE_F_LAST_CHUNK)
3294	break;
3295	hMmio2++;
3296	Assert(pCur->idMmio2 == hMmio2);
3297	Assert(pCur->pDevInsR3 == pDevIns);
3298	Assert(!(pCur->fFlags & PGMREGMMIO2RANGE_F_FIRST_CHUNK));
3299	}
3300	}
3301	else
3302	{
3303	pPrev = pCur;
3304	pCur = pCur->pNextR3;
3305	}
3306	}
3307	pgmPhysInvalidatePageMapTLB(pVM);
3308	PGM_UNLOCK(pVM);
3309	return !cFound && hMmio2 != NIL_PGMMMIO2HANDLE ? VERR_NOT_FOUND : rc;
3310	}
3311
3312
3313	/**
3314	* Maps a MMIO2 region.
3315	*
3316	* This is typically done when a guest / the bios / state loading changes the
3317	* PCI config. The replacing of base memory has the same restrictions as during
3318	* registration, of course.
3319	*
3320	* @returns VBox status code.
3321	*
3322	* @param pVM The cross context VM structure.
3323	* @param pDevIns The device instance owning the region.
3324	* @param hMmio2 The handle of the region to map.
3325	* @param GCPhys The guest-physical address to be remapped.
3326	*/
3327	VMMR3_INT_DECL(int) PGMR3PhysMmio2Map(PVM pVM, PPDMDEVINS pDevIns, PGMMMIO2HANDLE hMmio2, RTGCPHYS GCPhys)
3328	{
3329	/*
3330	* Validate input.
3331	*
3332	* Note! It's safe to walk the MMIO/MMIO2 list since registrations only
3333	* happens during VM construction.
3334	*/
3335	VM_ASSERT_EMT_RETURN(pVM, VERR_VM_THREAD_NOT_EMT);
3336	AssertPtrReturn(pDevIns, VERR_INVALID_PARAMETER);
3337	AssertReturn(GCPhys != NIL_RTGCPHYS, VERR_INVALID_PARAMETER);
3338	AssertReturn(GCPhys != 0, VERR_INVALID_PARAMETER);
3339	AssertReturn(!(GCPhys & GUEST_PAGE_OFFSET_MASK), VERR_INVALID_PARAMETER);
3340	AssertReturn(hMmio2 != NIL_PGMMMIO2HANDLE, VERR_INVALID_HANDLE);
3341
3342	PPGMREGMMIO2RANGE pFirstMmio = pgmR3PhysMmio2Find(pVM, pDevIns, UINT32_MAX, UINT32_MAX, hMmio2);
3343	AssertReturn(pFirstMmio, VERR_NOT_FOUND);
3344	Assert(pFirstMmio->fFlags & PGMREGMMIO2RANGE_F_FIRST_CHUNK);
3345
3346	PPGMREGMMIO2RANGE pLastMmio = pFirstMmio;
3347	RTGCPHYS cbRange = 0;
3348	for (;;)
3349	{
3350	AssertReturn(!(pLastMmio->fFlags & PGMREGMMIO2RANGE_F_MAPPED), VERR_WRONG_ORDER);
3351	Assert(pLastMmio->RamRange.GCPhys == NIL_RTGCPHYS);
3352	Assert(pLastMmio->RamRange.GCPhysLast == NIL_RTGCPHYS);
3353	Assert(pLastMmio->pDevInsR3 == pFirstMmio->pDevInsR3);
3354	Assert(pLastMmio->iSubDev == pFirstMmio->iSubDev);
3355	Assert(pLastMmio->iRegion == pFirstMmio->iRegion);
3356	cbRange += pLastMmio->RamRange.cb;
3357	if (pLastMmio->fFlags & PGMREGMMIO2RANGE_F_LAST_CHUNK)
3358	break;
3359	pLastMmio = pLastMmio->pNextR3;
3360	}
3361
3362	RTGCPHYS GCPhysLast = GCPhys + cbRange - 1;
3363	AssertLogRelReturn(GCPhysLast > GCPhys, VERR_INVALID_PARAMETER);
3364
3365	/*
3366	* Find our location in the ram range list, checking for restriction
3367	* we don't bother implementing yet (partially overlapping, multiple
3368	* ram ranges).
3369	*/
3370	PGM_LOCK_VOID(pVM);
3371
3372	AssertReturnStmt(!(pFirstMmio->fFlags & PGMREGMMIO2RANGE_F_MAPPED), PGM_UNLOCK(pVM), VERR_WRONG_ORDER);
3373
3374	bool fRamExists = false;
3375	PPGMRAMRANGE pRamPrev = NULL;
3376	PPGMRAMRANGE pRam = pVM->pgm.s.pRamRangesXR3;
3377	while (pRam && GCPhysLast >= pRam->GCPhys)
3378	{
3379	if ( GCPhys <= pRam->GCPhysLast
3380	&& GCPhysLast >= pRam->GCPhys)
3381	{
3382	/* Completely within? */
3383	AssertLogRelMsgReturnStmt( GCPhys >= pRam->GCPhys
3384	&& GCPhysLast <= pRam->GCPhysLast,
3385	("%RGp-%RGp (MMIOEx/%s) falls partly outside %RGp-%RGp (%s)\n",
3386	GCPhys, GCPhysLast, pFirstMmio->RamRange.pszDesc,
3387	pRam->GCPhys, pRam->GCPhysLast, pRam->pszDesc),
3388	PGM_UNLOCK(pVM),
3389	VERR_PGM_RAM_CONFLICT);
3390
3391	/* Check that all the pages are RAM pages. */
3392	PPGMPAGE pPage = &pRam->aPages[(GCPhys - pRam->GCPhys) >> GUEST_PAGE_SHIFT];
3393	uint32_t cPagesLeft = cbRange >> GUEST_PAGE_SHIFT;
3394	while (cPagesLeft-- > 0)
3395	{
3396	AssertLogRelMsgReturnStmt(PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_RAM,
3397	("%RGp isn't a RAM page (%d) - mapping %RGp-%RGp (MMIO2/%s).\n",
3398	GCPhys, PGM_PAGE_GET_TYPE(pPage), GCPhys, GCPhysLast, pFirstMmio->RamRange.pszDesc),
3399	PGM_UNLOCK(pVM),
3400	VERR_PGM_RAM_CONFLICT);
3401	pPage++;
3402	}
3403
3404	/* There can only be one MMIO/MMIO2 chunk matching here! */
3405	AssertLogRelMsgReturnStmt(pFirstMmio->fFlags & PGMREGMMIO2RANGE_F_LAST_CHUNK,
3406	("%RGp-%RGp (MMIOEx/%s, flags %#X) consists of multiple chunks whereas the RAM somehow doesn't!\n",
3407	GCPhys, GCPhysLast, pFirstMmio->RamRange.pszDesc, pFirstMmio->fFlags),
3408	PGM_UNLOCK(pVM),
3409	VERR_PGM_PHYS_MMIO_EX_IPE);
3410
3411	fRamExists = true;
3412	break;
3413	}
3414
3415	/* next */
3416	pRamPrev = pRam;
3417	pRam = pRam->pNextR3;
3418	}
3419	Log(("PGMR3PhysMmio2Map: %RGp-%RGp fRamExists=%RTbool %s\n", GCPhys, GCPhysLast, fRamExists, pFirstMmio->RamRange.pszDesc));
3420
3421
3422	/*
3423	* Make the changes.
3424	*/
3425	RTGCPHYS GCPhysCur = GCPhys;
3426	for (PPGMREGMMIO2RANGE pCurMmio = pFirstMmio; ; pCurMmio = pCurMmio->pNextR3)
3427	{
3428	pCurMmio->RamRange.GCPhys = GCPhysCur;
3429	pCurMmio->RamRange.GCPhysLast = GCPhysCur + pCurMmio->RamRange.cb - 1;
3430	if (pCurMmio->fFlags & PGMREGMMIO2RANGE_F_LAST_CHUNK)
3431	{
3432	Assert(pCurMmio->RamRange.GCPhysLast == GCPhysLast);
3433	break;
3434	}
3435	GCPhysCur += pCurMmio->RamRange.cb;
3436	}
3437
3438	if (fRamExists)
3439	{
3440	/*
3441	* Make all the pages in the range MMIO/ZERO pages, freeing any
3442	* RAM pages currently mapped here. This might not be 100% correct
3443	* for PCI memory, but we're doing the same thing for MMIO2 pages.
3444	*
3445	* We replace these MMIO/ZERO pages with real pages in the MMIO2 case.
3446	*/
3447	Assert(pFirstMmio->fFlags & PGMREGMMIO2RANGE_F_LAST_CHUNK); /* Only one chunk */
3448	Assert(pFirstMmio->pvR3 == pFirstMmio->RamRange.pvR3);
3449	Assert(pFirstMmio->RamRange.pvR3 != NULL);
3450
3451	#ifdef VBOX_WITH_PGM_NEM_MODE
3452	/* We cannot mix MMIO2 into a RAM range in simplified memory mode because pRam->pvR3 can't point
3453	both at the RAM and MMIO2, so we won't ever write & read from the actual MMIO2 memory if we try. */
3454	AssertLogRelMsgReturn(!pVM->pgm.s.fNemMode, ("%s at %RGp-%RGp\n", pFirstMmio->RamRange.pszDesc, GCPhys, GCPhysLast),
3455	VERR_PGM_NOT_SUPPORTED_FOR_NEM_MODE);
3456	#endif
3457
3458	int rc = pgmR3PhysFreePageRange(pVM, pRam, GCPhys, GCPhysLast, pFirstMmio->RamRange.pvR3);
3459	AssertRCReturnStmt(rc, PGM_UNLOCK(pVM), rc);
3460
3461	/* Replace the pages, freeing all present RAM pages. */
3462	PPGMPAGE pPageSrc = &pFirstMmio->RamRange.aPages[0];
3463	PPGMPAGE pPageDst = &pRam->aPages[(GCPhys - pRam->GCPhys) >> GUEST_PAGE_SHIFT];
3464	uint32_t cPagesLeft = pFirstMmio->RamRange.cb >> GUEST_PAGE_SHIFT;
3465	while (cPagesLeft-- > 0)
3466	{
3467	Assert(PGM_PAGE_IS_MMIO(pPageDst));
3468
3469	RTHCPHYS const HCPhys = PGM_PAGE_GET_HCPHYS(pPageSrc);
3470	uint32_t const idPage = PGM_PAGE_GET_PAGEID(pPageSrc);
3471	PGM_PAGE_SET_PAGEID(pVM, pPageDst, idPage);
3472	PGM_PAGE_SET_HCPHYS(pVM, pPageDst, HCPhys);
3473	PGM_PAGE_SET_TYPE(pVM, pPageDst, PGMPAGETYPE_MMIO2);
3474	PGM_PAGE_SET_STATE(pVM, pPageDst, PGM_PAGE_STATE_ALLOCATED);
3475	PGM_PAGE_SET_PDE_TYPE(pVM, pPageDst, PGM_PAGE_PDE_TYPE_DONTCARE);
3476	PGM_PAGE_SET_PTE_INDEX(pVM, pPageDst, 0);
3477	PGM_PAGE_SET_TRACKING(pVM, pPageDst, 0);
3478	/* NEM state is set by pgmR3PhysFreePageRange. */
3479
3480	pVM->pgm.s.cZeroPages--;
3481	GCPhys += GUEST_PAGE_SIZE;
3482	pPageSrc++;
3483	pPageDst++;
3484	}
3485
3486	/* Flush physical page map TLB. */
3487	pgmPhysInvalidatePageMapTLB(pVM);
3488
3489	/* Force a PGM pool flush as guest ram references have been changed. */
3490	/** @todo not entirely SMP safe; assuming for now the guest takes care of
3491	* this internally (not touch mapped mmio while changing the mapping). */
3492	PVMCPU pVCpu = VMMGetCpu(pVM);
3493	pVCpu->pgm.s.fSyncFlags \|= PGM_SYNC_CLEAR_PGM_POOL;
3494	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
3495	}
3496	else
3497	{
3498	/*
3499	* No RAM range, insert the ones prepared during registration.
3500	*/
3501	for (PPGMREGMMIO2RANGE pCurMmio = pFirstMmio; ; pCurMmio = pCurMmio->pNextR3)
3502	{
3503	#ifdef VBOX_WITH_NATIVE_NEM
3504	/* Tell NEM and get the new NEM state for the pages. */
3505	uint8_t u2NemState = 0;
3506	if (VM_IS_NEM_ENABLED(pVM))
3507	{
3508	int rc = NEMR3NotifyPhysMmioExMapEarly(pVM, pCurMmio->RamRange.GCPhys,
3509	pCurMmio->RamRange.GCPhysLast - pCurMmio->RamRange.GCPhys + 1,
3510	NEM_NOTIFY_PHYS_MMIO_EX_F_MMIO2
3511	\| (pCurMmio->fFlags & PGMREGMMIO2RANGE_F_TRACK_DIRTY_PAGES
3512	? NEM_NOTIFY_PHYS_MMIO_EX_F_TRACK_DIRTY_PAGES : 0),
3513	NULL /pvRam/, pCurMmio->RamRange.pvR3,
3514	&u2NemState, &pCurMmio->RamRange.uNemRange);
3515	AssertLogRelRCReturnStmt(rc, PGM_UNLOCK(pVM), rc);
3516	}
3517	#endif
3518
3519	/* Clear the tracking data of pages we're going to reactivate. */
3520	PPGMPAGE pPageSrc = &pCurMmio->RamRange.aPages[0];
3521	uint32_t cPagesLeft = pCurMmio->RamRange.cb >> GUEST_PAGE_SHIFT;
3522	while (cPagesLeft-- > 0)
3523	{
3524	PGM_PAGE_SET_TRACKING(pVM, pPageSrc, 0);
3525	PGM_PAGE_SET_PTE_INDEX(pVM, pPageSrc, 0);
3526	#ifdef VBOX_WITH_NATIVE_NEM
3527	PGM_PAGE_SET_NEM_STATE(pPageSrc, u2NemState);
3528	#endif
3529	pPageSrc++;
3530	}
3531
3532	/* link in the ram range */
3533	pgmR3PhysLinkRamRange(pVM, &pCurMmio->RamRange, pRamPrev);
3534
3535	if (pCurMmio->fFlags & PGMREGMMIO2RANGE_F_LAST_CHUNK)
3536	{
3537	Assert(pCurMmio->RamRange.GCPhysLast == GCPhysLast);
3538	break;
3539	}
3540	pRamPrev = &pCurMmio->RamRange;
3541	}
3542	}
3543
3544	/*
3545	* If the range have dirty page monitoring enabled, enable that.
3546	*
3547	* We ignore failures here for now because if we fail, the whole mapping
3548	* will have to be reversed and we'll end up with nothing at all on the
3549	* screen and a grumpy guest, whereas if we just go on, we'll only have
3550	* visual distortions to gripe about. There will be something in the
3551	* release log.
3552	*/
3553	if ( pFirstMmio->pPhysHandlerR3
3554	&& (pFirstMmio->fFlags & PGMREGMMIO2RANGE_F_TRACKING_ENABLED))
3555	pgmR3PhysMmio2EnableDirtyPageTracing(pVM, pFirstMmio);
3556
3557	/*
3558	* We're good, set the flags and invalid the mapping TLB.
3559	*/
3560	for (PPGMREGMMIO2RANGE pCurMmio = pFirstMmio; ; pCurMmio = pCurMmio->pNextR3)
3561	{
3562	pCurMmio->fFlags \|= PGMREGMMIO2RANGE_F_MAPPED;
3563	if (fRamExists)
3564	pCurMmio->fFlags \|= PGMREGMMIO2RANGE_F_OVERLAPPING;
3565	else
3566	pCurMmio->fFlags &= ~PGMREGMMIO2RANGE_F_OVERLAPPING;
3567	if (pCurMmio->fFlags & PGMREGMMIO2RANGE_F_LAST_CHUNK)
3568	break;
3569	}
3570	pgmPhysInvalidatePageMapTLB(pVM);
3571
3572	#ifdef VBOX_WITH_NATIVE_NEM
3573	/*
3574	* Late NEM notification.
3575	*/
3576	if (VM_IS_NEM_ENABLED(pVM))
3577	{
3578	int rc;
3579	uint32_t fNemFlags = NEM_NOTIFY_PHYS_MMIO_EX_F_MMIO2;
3580	if (pFirstMmio->fFlags & PGMREGMMIO2RANGE_F_TRACK_DIRTY_PAGES)
3581	fNemFlags \|= NEM_NOTIFY_PHYS_MMIO_EX_F_TRACK_DIRTY_PAGES;
3582	if (fRamExists)
3583	rc = NEMR3NotifyPhysMmioExMapLate(pVM, GCPhys, GCPhysLast - GCPhys + 1, fNemFlags \| NEM_NOTIFY_PHYS_MMIO_EX_F_REPLACE,
3584	pRam->pvR3 ? (uint8_t *)pRam->pvR3 + GCPhys - pRam->GCPhys : NULL, pFirstMmio->pvR3,
3585	NULL /puNemRange/);
3586	else
3587	{
3588	rc = VINF_SUCCESS;
3589	for (PPGMREGMMIO2RANGE pCurMmio = pFirstMmio; ; pCurMmio = pCurMmio->pNextR3)
3590	{
3591	rc = NEMR3NotifyPhysMmioExMapLate(pVM, pCurMmio->RamRange.GCPhys, pCurMmio->RamRange.cb, fNemFlags,
3592	NULL, pCurMmio->RamRange.pvR3, &pCurMmio->RamRange.uNemRange);
3593	if ((pCurMmio->fFlags & PGMREGMMIO2RANGE_F_LAST_CHUNK) \|\| RT_FAILURE(rc))
3594	break;
3595	}
3596	}
3597	AssertLogRelRCReturnStmt(rc, PGMR3PhysMmio2Unmap(pVM, pDevIns, hMmio2, GCPhys); PGM_UNLOCK(pVM), rc);
3598	}
3599	#endif
3600
3601	PGM_UNLOCK(pVM);
3602
3603	return VINF_SUCCESS;
3604	}
3605
3606
3607	/**
3608	* Unmaps an MMIO2 region.
3609	*
3610	* This is typically done when a guest / the bios / state loading changes the
3611	* PCI config. The replacing of base memory has the same restrictions as during
3612	* registration, of course.
3613	*/
3614	VMMR3_INT_DECL(int) PGMR3PhysMmio2Unmap(PVM pVM, PPDMDEVINS pDevIns, PGMMMIO2HANDLE hMmio2, RTGCPHYS GCPhys)
3615	{
3616	/*
3617	* Validate input
3618	*/
3619	VM_ASSERT_EMT_RETURN(pVM, VERR_VM_THREAD_NOT_EMT);
3620	AssertPtrReturn(pDevIns, VERR_INVALID_PARAMETER);
3621	AssertReturn(hMmio2 != NIL_PGMMMIO2HANDLE, VERR_INVALID_HANDLE);
3622	if (GCPhys != NIL_RTGCPHYS)
3623	{
3624	AssertReturn(GCPhys != 0, VERR_INVALID_PARAMETER);
3625	AssertReturn(!(GCPhys & GUEST_PAGE_OFFSET_MASK), VERR_INVALID_PARAMETER);
3626	}
3627
3628	PPGMREGMMIO2RANGE pFirstMmio = pgmR3PhysMmio2Find(pVM, pDevIns, UINT32_MAX, UINT32_MAX, hMmio2);
3629	AssertReturn(pFirstMmio, VERR_NOT_FOUND);
3630	Assert(pFirstMmio->fFlags & PGMREGMMIO2RANGE_F_FIRST_CHUNK);
3631
3632	int rc = PGM_LOCK(pVM);
3633	AssertRCReturn(rc, rc);
3634
3635	PPGMREGMMIO2RANGE pLastMmio = pFirstMmio;
3636	RTGCPHYS cbRange = 0;
3637	for (;;)
3638	{
3639	AssertReturnStmt(pLastMmio->fFlags & PGMREGMMIO2RANGE_F_MAPPED, PGM_UNLOCK(pVM), VERR_WRONG_ORDER);
3640	AssertReturnStmt(pLastMmio->RamRange.GCPhys == GCPhys + cbRange \|\| GCPhys == NIL_RTGCPHYS, PGM_UNLOCK(pVM), VERR_INVALID_PARAMETER);
3641	Assert(pLastMmio->pDevInsR3 == pFirstMmio->pDevInsR3);
3642	Assert(pLastMmio->iSubDev == pFirstMmio->iSubDev);
3643	Assert(pLastMmio->iRegion == pFirstMmio->iRegion);
3644	cbRange += pLastMmio->RamRange.cb;
3645	if (pLastMmio->fFlags & PGMREGMMIO2RANGE_F_LAST_CHUNK)
3646	break;
3647	pLastMmio = pLastMmio->pNextR3;
3648	}
3649
3650	Log(("PGMR3PhysMmio2Unmap: %RGp-%RGp %s\n",
3651	pFirstMmio->RamRange.GCPhys, pLastMmio->RamRange.GCPhysLast, pFirstMmio->RamRange.pszDesc));
3652
3653	uint16_t const fOldFlags = pFirstMmio->fFlags;
3654	AssertReturnStmt(fOldFlags & PGMREGMMIO2RANGE_F_MAPPED, PGM_UNLOCK(pVM), VERR_WRONG_ORDER);
3655
3656	/*
3657	* If monitoring dirty pages, we must deregister the handlers first.
3658	*/
3659	if ( pFirstMmio->pPhysHandlerR3
3660	&& (fOldFlags & PGMREGMMIO2RANGE_F_TRACKING_ENABLED))
3661	pgmR3PhysMmio2DisableDirtyPageTracing(pVM, pFirstMmio);
3662
3663	/*
3664	* Unmap it.
3665	*/
3666	int rcRet = VINF_SUCCESS;
3667	#ifdef VBOX_WITH_NATIVE_NEM
3668	uint32_t const fNemFlags = NEM_NOTIFY_PHYS_MMIO_EX_F_MMIO2
3669	\| (fOldFlags & PGMREGMMIO2RANGE_F_TRACK_DIRTY_PAGES
3670	? NEM_NOTIFY_PHYS_MMIO_EX_F_TRACK_DIRTY_PAGES : 0);
3671	#endif
3672	if (fOldFlags & PGMREGMMIO2RANGE_F_OVERLAPPING)
3673	{
3674	/*
3675	* We've replaced RAM, replace with zero pages.
3676	*
3677	* Note! This is where we might differ a little from a real system, because
3678	* it's likely to just show the RAM pages as they were before the
3679	* MMIO/MMIO2 region was mapped here.
3680	*/
3681	/* Only one chunk allowed when overlapping! */
3682	Assert(fOldFlags & PGMREGMMIO2RANGE_F_LAST_CHUNK);
3683
3684	/* Restore the RAM pages we've replaced. */
3685	PPGMRAMRANGE pRam = pVM->pgm.s.pRamRangesXR3;
3686	while (pRam->GCPhys > pFirstMmio->RamRange.GCPhysLast)
3687	pRam = pRam->pNextR3;
3688
3689	PPGMPAGE pPageDst = &pRam->aPages[(pFirstMmio->RamRange.GCPhys - pRam->GCPhys) >> GUEST_PAGE_SHIFT];
3690	uint32_t cPagesLeft = pFirstMmio->RamRange.cb >> GUEST_PAGE_SHIFT;
3691	pVM->pgm.s.cZeroPages += cPagesLeft; /** @todo not correct for NEM mode */
3692
3693	#ifdef VBOX_WITH_NATIVE_NEM
3694	if (VM_IS_NEM_ENABLED(pVM)) /* Notify NEM. Note! we cannot be here in simple memory mode, see mapping function. */
3695	{
3696	uint8_t u2State = UINT8_MAX;
3697	rc = NEMR3NotifyPhysMmioExUnmap(pVM, pFirstMmio->RamRange.GCPhys, pFirstMmio->RamRange.cb,
3698	fNemFlags \| NEM_NOTIFY_PHYS_MMIO_EX_F_REPLACE,
3699	pRam->pvR3
3700	? (uint8_t *)pRam->pvR3 + pFirstMmio->RamRange.GCPhys - pRam->GCPhys : NULL,
3701	pFirstMmio->pvR3, &u2State, &pRam->uNemRange);
3702	AssertRCStmt(rc, rcRet = rc);
3703	if (u2State != UINT8_MAX)
3704	pgmPhysSetNemStateForPages(pPageDst, cPagesLeft, u2State);
3705	}
3706	#endif
3707
3708	while (cPagesLeft-- > 0)
3709	{
3710	PGM_PAGE_INIT_ZERO(pPageDst, pVM, PGMPAGETYPE_RAM);
3711	pPageDst++;
3712	}
3713
3714	/* Flush physical page map TLB. */
3715	pgmPhysInvalidatePageMapTLB(pVM);
3716
3717	/* Update range state. */
3718	pFirstMmio->RamRange.GCPhys = NIL_RTGCPHYS;
3719	pFirstMmio->RamRange.GCPhysLast = NIL_RTGCPHYS;
3720	pFirstMmio->fFlags &= ~(PGMREGMMIO2RANGE_F_OVERLAPPING \| PGMREGMMIO2RANGE_F_MAPPED);
3721	}
3722	else
3723	{
3724	/*
3725	* Unlink the chunks related to the MMIO/MMIO2 region.
3726	*/
3727	for (PPGMREGMMIO2RANGE pCurMmio = pFirstMmio; ; pCurMmio = pCurMmio->pNextR3)
3728	{
3729	#ifdef VBOX_WITH_NATIVE_NEM
3730	if (VM_IS_NEM_ENABLED(pVM)) /* Notify NEM. */
3731	{
3732	uint8_t u2State = UINT8_MAX;
3733	rc = NEMR3NotifyPhysMmioExUnmap(pVM, pCurMmio->RamRange.GCPhys, pCurMmio->RamRange.cb, fNemFlags,
3734	NULL, pCurMmio->pvR3, &u2State, &pCurMmio->RamRange.uNemRange);
3735	AssertRCStmt(rc, rcRet = rc);
3736	if (u2State != UINT8_MAX)
3737	pgmPhysSetNemStateForPages(pCurMmio->RamRange.aPages, pCurMmio->RamRange.cb >> GUEST_PAGE_SHIFT, u2State);
3738	}
3739	#endif
3740	pgmR3PhysUnlinkRamRange(pVM, &pCurMmio->RamRange);
3741	pCurMmio->RamRange.GCPhys = NIL_RTGCPHYS;
3742	pCurMmio->RamRange.GCPhysLast = NIL_RTGCPHYS;
3743	pCurMmio->fFlags &= ~(PGMREGMMIO2RANGE_F_OVERLAPPING \| PGMREGMMIO2RANGE_F_MAPPED);
3744	if (pCurMmio->fFlags & PGMREGMMIO2RANGE_F_LAST_CHUNK)
3745	break;
3746	}
3747	}
3748
3749	/* Force a PGM pool flush as guest ram references have been changed. */
3750	/** @todo not entirely SMP safe; assuming for now the guest takes care
3751	* of this internally (not touch mapped mmio while changing the
3752	* mapping). */
3753	PVMCPU pVCpu = VMMGetCpu(pVM);
3754	pVCpu->pgm.s.fSyncFlags \|= PGM_SYNC_CLEAR_PGM_POOL;
3755	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
3756
3757	pgmPhysInvalidatePageMapTLB(pVM);
3758	pgmPhysInvalidRamRangeTlbs(pVM);
3759
3760	PGM_UNLOCK(pVM);
3761	return rcRet;
3762	}
3763
3764
3765	/**
3766	* Reduces the mapping size of a MMIO2 region.
3767	*
3768	* This is mainly for dealing with old saved states after changing the default
3769	* size of a mapping region. See PGMDevHlpMMIOExReduce and
3770	* PDMPCIDEV::pfnRegionLoadChangeHookR3.
3771	*
3772	* The region must not currently be mapped when making this call. The VM state
3773	* must be state restore or VM construction.
3774	*
3775	* @returns VBox status code.
3776	* @param pVM The cross context VM structure.
3777	* @param pDevIns The device instance owning the region.
3778	* @param hMmio2 The handle of the region to reduce.
3779	* @param cbRegion The new mapping size.
3780	*/
3781	VMMR3_INT_DECL(int) PGMR3PhysMmio2Reduce(PVM pVM, PPDMDEVINS pDevIns, PGMMMIO2HANDLE hMmio2, RTGCPHYS cbRegion)
3782	{
3783	/*
3784	* Validate input
3785	*/
3786	VM_ASSERT_EMT_RETURN(pVM, VERR_VM_THREAD_NOT_EMT);
3787	AssertPtrReturn(pDevIns, VERR_INVALID_PARAMETER);
3788	AssertReturn(hMmio2 != NIL_PGMMMIO2HANDLE, VERR_INVALID_HANDLE);
3789	AssertReturn(cbRegion >= X86_PAGE_SIZE, VERR_INVALID_PARAMETER);
3790	AssertReturn(!(cbRegion & X86_PAGE_OFFSET_MASK), VERR_UNSUPPORTED_ALIGNMENT);
3791	VMSTATE enmVmState = VMR3GetState(pVM);
3792	AssertLogRelMsgReturn( enmVmState == VMSTATE_CREATING
3793	\|\| enmVmState == VMSTATE_LOADING,
3794	("enmVmState=%d (%s)\n", enmVmState, VMR3GetStateName(enmVmState)),
3795	VERR_VM_INVALID_VM_STATE);
3796
3797	int rc = PGM_LOCK(pVM);
3798	AssertRCReturn(rc, rc);
3799
3800	PPGMREGMMIO2RANGE pFirstMmio = pgmR3PhysMmio2Find(pVM, pDevIns, UINT32_MAX, UINT32_MAX, hMmio2);
3801	if (pFirstMmio)
3802	{
3803	Assert(pFirstMmio->fFlags & PGMREGMMIO2RANGE_F_FIRST_CHUNK);
3804	if (!(pFirstMmio->fFlags & PGMREGMMIO2RANGE_F_MAPPED))
3805	{
3806	/*
3807	* NOTE! Current implementation does not support multiple ranges.
3808	* Implement when there is a real world need and thus a testcase.
3809	*/
3810	AssertLogRelMsgStmt(pFirstMmio->fFlags & PGMREGMMIO2RANGE_F_LAST_CHUNK,
3811	("%s: %#x\n", pFirstMmio->RamRange.pszDesc, pFirstMmio->fFlags),
3812	rc = VERR_NOT_SUPPORTED);
3813	if (RT_SUCCESS(rc))
3814	{
3815	/*
3816	* Make the change.
3817	*/
3818	Log(("PGMR3PhysMmio2Reduce: %s changes from %RGp bytes (%RGp) to %RGp bytes.\n",
3819	pFirstMmio->RamRange.pszDesc, pFirstMmio->RamRange.cb, pFirstMmio->cbReal, cbRegion));
3820
3821	AssertLogRelMsgStmt(cbRegion <= pFirstMmio->cbReal,
3822	("%s: cbRegion=%#RGp cbReal=%#RGp\n", pFirstMmio->RamRange.pszDesc, cbRegion, pFirstMmio->cbReal),
3823	rc = VERR_OUT_OF_RANGE);
3824	if (RT_SUCCESS(rc))
3825	{
3826	pFirstMmio->RamRange.cb = cbRegion;
3827	}
3828	}
3829	}
3830	else
3831	rc = VERR_WRONG_ORDER;
3832	}
3833	else
3834	rc = VERR_NOT_FOUND;
3835
3836	PGM_UNLOCK(pVM);
3837	return rc;
3838	}
3839
3840
3841	/**
3842	* Validates @a hMmio2, making sure it belongs to @a pDevIns.
3843	*
3844	* @returns VBox status code.
3845	* @param pVM The cross context VM structure.
3846	* @param pDevIns The device which allegedly owns @a hMmio2.
3847	* @param hMmio2 The handle to validate.
3848	*/
3849	VMMR3_INT_DECL(int) PGMR3PhysMmio2ValidateHandle(PVM pVM, PPDMDEVINS pDevIns, PGMMMIO2HANDLE hMmio2)
3850	{
3851	/*
3852	* Validate input
3853	*/
3854	VM_ASSERT_EMT_RETURN(pVM, VERR_VM_THREAD_NOT_EMT);
3855	AssertPtrReturn(pDevIns, VERR_INVALID_POINTER);
3856
3857	/*
3858	* Just do this the simple way. No need for locking as this is only taken at
3859	*/
3860	PGM_LOCK_VOID(pVM);
3861	PPGMREGMMIO2RANGE pFirstMmio = pgmR3PhysMmio2Find(pVM, pDevIns, UINT32_MAX, UINT32_MAX, hMmio2);
3862	PGM_UNLOCK(pVM);
3863	AssertReturn(pFirstMmio, VERR_INVALID_HANDLE);
3864	AssertReturn(pFirstMmio->fFlags & PGMREGMMIO2RANGE_F_FIRST_CHUNK, VERR_INVALID_HANDLE);
3865	return VINF_SUCCESS;
3866	}
3867
3868
3869	/**
3870	* Gets the mapping address of an MMIO2 region.
3871	*
3872	* @returns Mapping address, NIL_RTGCPHYS if not mapped or invalid handle.
3873	*
3874	* @param pVM The cross context VM structure.
3875	* @param pDevIns The device owning the MMIO2 handle.
3876	* @param hMmio2 The region handle.
3877	*/
3878	VMMR3_INT_DECL(RTGCPHYS) PGMR3PhysMmio2GetMappingAddress(PVM pVM, PPDMDEVINS pDevIns, PGMMMIO2HANDLE hMmio2)
3879	{
3880	AssertPtrReturn(pDevIns, NIL_RTGCPHYS);
3881
3882	PPGMREGMMIO2RANGE pFirstRegMmio = pgmR3PhysMmio2Find(pVM, pDevIns, UINT32_MAX, UINT32_MAX, hMmio2);
3883	AssertReturn(pFirstRegMmio, NIL_RTGCPHYS);
3884
3885	if (pFirstRegMmio->fFlags & PGMREGMMIO2RANGE_F_MAPPED)
3886	return pFirstRegMmio->RamRange.GCPhys;
3887	return NIL_RTGCPHYS;
3888	}
3889
3890
3891	/**
3892	* Worker for PGMR3PhysMmio2QueryAndResetDirtyBitmap.
3893	*
3894	* Called holding the PGM lock.
3895	*/
3896	static int pgmR3PhysMmio2QueryAndResetDirtyBitmapLocked(PVM pVM, PPDMDEVINS pDevIns, PGMMMIO2HANDLE hMmio2,
3897	void *pvBitmap, size_t cbBitmap)
3898	{
3899	/*
3900	* Continue validation.
3901	*/
3902	PPGMREGMMIO2RANGE pFirstRegMmio = pgmR3PhysMmio2Find(pVM, pDevIns, UINT32_MAX, UINT32_MAX, hMmio2);
3903	AssertReturn(pFirstRegMmio, VERR_INVALID_HANDLE);
3904	AssertReturn( (pFirstRegMmio->fFlags & (PGMREGMMIO2RANGE_F_TRACK_DIRTY_PAGES \| PGMREGMMIO2RANGE_F_FIRST_CHUNK))
3905	== (PGMREGMMIO2RANGE_F_TRACK_DIRTY_PAGES \| PGMREGMMIO2RANGE_F_FIRST_CHUNK),
3906	VERR_INVALID_FUNCTION);
3907	AssertReturn(pDevIns == pFirstRegMmio->pDevInsR3, VERR_NOT_OWNER);
3908
3909	RTGCPHYS cbTotal = 0;
3910	uint16_t fTotalDirty = 0;
3911	for (PPGMREGMMIO2RANGE pCur = pFirstRegMmio;;)
3912	{
3913	cbTotal += pCur->RamRange.cb; /* Not using cbReal here, because NEM is not in on the creating, only the mapping. */
3914	fTotalDirty \|= pCur->fFlags;
3915	if (pCur->fFlags & PGMREGMMIO2RANGE_F_LAST_CHUNK)
3916	break;
3917	pCur = pCur->pNextR3;
3918	AssertPtrReturn(pCur, VERR_INTERNAL_ERROR_5);
3919	AssertReturn( (pCur->fFlags & (PGMREGMMIO2RANGE_F_TRACK_DIRTY_PAGES \| PGMREGMMIO2RANGE_F_FIRST_CHUNK))
3920	== PGMREGMMIO2RANGE_F_TRACK_DIRTY_PAGES,
3921	VERR_INTERNAL_ERROR_4);
3922	}
3923	size_t const cbTotalBitmap = RT_ALIGN_T(cbTotal, GUEST_PAGE_SIZE * 64, RTGCPHYS) / GUEST_PAGE_SIZE / 8;
3924
3925	if (cbBitmap)
3926	{
3927	AssertPtrReturn(pvBitmap, VERR_INVALID_POINTER);
3928	AssertReturn(RT_ALIGN_P(pvBitmap, sizeof(uint64_t)) == pvBitmap, VERR_INVALID_POINTER);
3929	AssertReturn(cbBitmap == cbTotalBitmap, VERR_INVALID_PARAMETER);
3930	}
3931
3932	/*
3933	* Do the work.
3934	*/
3935	int rc = VINF_SUCCESS;
3936	if (pvBitmap)
3937	{
3938	#ifdef VBOX_WITH_PGM_NEM_MODE
3939	if (pFirstRegMmio->pPhysHandlerR3 == NULL)
3940	{
3941	/** @todo This does not integrate at all with --execute-all-in-iem, leaving the
3942	* screen blank when using it together with --driverless. Fixing this won't be
3943	* entirely easy as we take the PGM_PAGE_HNDL_PHYS_STATE_DISABLED page status to
3944	* mean a dirty page. */
3945	AssertReturn(VM_IS_NEM_ENABLED(pVM), VERR_INTERNAL_ERROR_4);
3946	uint8_t pbBitmap = (uint8_t )pvBitmap;
3947	for (PPGMREGMMIO2RANGE pCur = pFirstRegMmio; pCur; pCur = pCur->pNextR3)
3948	{
3949	size_t const cbBitmapChunk = pCur->RamRange.cb / GUEST_PAGE_SIZE / 8;
3950	Assert((RTGCPHYS)cbBitmapChunk * GUEST_PAGE_SIZE * 8 == pCur->RamRange.cb);
3951	int rc2 = NEMR3PhysMmio2QueryAndResetDirtyBitmap(pVM, pCur->RamRange.GCPhys, pCur->RamRange.cb,
3952	pCur->RamRange.uNemRange, pbBitmap, cbBitmapChunk);
3953	if (RT_FAILURE(rc2) && RT_SUCCESS(rc))
3954	rc = rc2;
3955	if (pCur->fFlags & PGMREGMMIO2RANGE_F_LAST_CHUNK)
3956	break;
3957	pbBitmap += pCur->RamRange.cb / GUEST_PAGE_SIZE / 8;
3958	}
3959	}
3960	else
3961	#endif
3962	if (fTotalDirty & PGMREGMMIO2RANGE_F_IS_DIRTY)
3963	{
3964	if ( (pFirstRegMmio->fFlags & (PGMREGMMIO2RANGE_F_MAPPED \| PGMREGMMIO2RANGE_F_TRACKING_ENABLED))
3965	== (PGMREGMMIO2RANGE_F_MAPPED \| PGMREGMMIO2RANGE_F_TRACKING_ENABLED))
3966	{
3967	/*
3968	* Reset each chunk, gathering dirty bits.
3969	*/
3970	RT_BZERO(pvBitmap, cbBitmap); /* simpler for now. */
3971	uint32_t iPageNo = 0;
3972	for (PPGMREGMMIO2RANGE pCur = pFirstRegMmio; pCur; pCur = pCur->pNextR3)
3973	{
3974	if (pCur->fFlags & PGMREGMMIO2RANGE_F_IS_DIRTY)
3975	{
3976	int rc2 = pgmHandlerPhysicalResetMmio2WithBitmap(pVM, pCur->RamRange.GCPhys, pvBitmap, iPageNo);
3977	if (RT_FAILURE(rc2) && RT_SUCCESS(rc))
3978	rc = rc2;
3979	pCur->fFlags &= ~PGMREGMMIO2RANGE_F_IS_DIRTY;
3980	}
3981	if (pCur->fFlags & PGMREGMMIO2RANGE_F_LAST_CHUNK)
3982	break;
3983	iPageNo += pCur->RamRange.cb >> GUEST_PAGE_SHIFT;
3984	}
3985	}
3986	else
3987	{
3988	/*
3989	* If not mapped or tracking is disabled, we return the
3990	* PGMREGMMIO2RANGE_F_IS_DIRTY status for all pages. We cannot
3991	* get more accurate data than that after unmapping or disabling.
3992	*/
3993	RT_BZERO(pvBitmap, cbBitmap);
3994	uint32_t iPageNo = 0;
3995	for (PPGMREGMMIO2RANGE pCur = pFirstRegMmio; pCur; pCur = pCur->pNextR3)
3996	{
3997	if (pCur->fFlags & PGMREGMMIO2RANGE_F_IS_DIRTY)
3998	{
3999	ASMBitSetRange(pvBitmap, iPageNo, iPageNo + (pCur->RamRange.cb >> GUEST_PAGE_SHIFT));
4000	pCur->fFlags &= ~PGMREGMMIO2RANGE_F_IS_DIRTY;
4001	}
4002	if (pCur->fFlags & PGMREGMMIO2RANGE_F_LAST_CHUNK)
4003	break;
4004	iPageNo += pCur->RamRange.cb >> GUEST_PAGE_SHIFT;
4005	}
4006	}
4007	}
4008	/*
4009	* No dirty chunks.
4010	*/
4011	else
4012	RT_BZERO(pvBitmap, cbBitmap);
4013	}
4014	/*
4015	* No bitmap. Reset the region if tracking is currently enabled.
4016	*/
4017	else if ( (pFirstRegMmio->fFlags & (PGMREGMMIO2RANGE_F_MAPPED \| PGMREGMMIO2RANGE_F_TRACKING_ENABLED))
4018	== (PGMREGMMIO2RANGE_F_MAPPED \| PGMREGMMIO2RANGE_F_TRACKING_ENABLED))
4019	{
4020	#ifdef VBOX_WITH_PGM_NEM_MODE
4021	if (pFirstRegMmio->pPhysHandlerR3 == NULL)
4022	{
4023	AssertReturn(VM_IS_NEM_ENABLED(pVM), VERR_INTERNAL_ERROR_4);
4024	for (PPGMREGMMIO2RANGE pCur = pFirstRegMmio; pCur; pCur = pCur->pNextR3)
4025	{
4026	int rc2 = NEMR3PhysMmio2QueryAndResetDirtyBitmap(pVM, pCur->RamRange.GCPhys, pCur->RamRange.cb,
4027	pCur->RamRange.uNemRange, NULL, 0);
4028	if (RT_FAILURE(rc2) && RT_SUCCESS(rc))
4029	rc = rc2;
4030	if (pCur->fFlags & PGMREGMMIO2RANGE_F_LAST_CHUNK)
4031	break;
4032	}
4033	}
4034	else
4035	#endif
4036	{
4037	for (PPGMREGMMIO2RANGE pCur = pFirstRegMmio; pCur; pCur = pCur->pNextR3)
4038	{
4039	pCur->fFlags &= ~PGMREGMMIO2RANGE_F_IS_DIRTY;
4040	int rc2 = PGMHandlerPhysicalReset(pVM, pCur->RamRange.GCPhys);
4041	if (RT_FAILURE(rc2) && RT_SUCCESS(rc))
4042	rc = rc2;
4043	if (pCur->fFlags & PGMREGMMIO2RANGE_F_LAST_CHUNK)
4044	break;
4045	}
4046	}
4047	}
4048
4049	return rc;
4050	}
4051
4052
4053	/**
4054	* Queries the dirty page bitmap and resets the monitoring.
4055	*
4056	* The PGMPHYS_MMIO2_FLAGS_TRACK_DIRTY_PAGES flag must be specified when
4057	* creating the range for this to work.
4058	*
4059	* @returns VBox status code.
4060	* @retval VERR_INVALID_FUNCTION if not created using
4061	* PGMPHYS_MMIO2_FLAGS_TRACK_DIRTY_PAGES.
4062	* @param pVM The cross context VM structure.
4063	* @param pDevIns The device owning the MMIO2 handle.
4064	* @param hMmio2 The region handle.
4065	* @param pvBitmap The output bitmap. Must be 8-byte aligned. Ignored
4066	* when @a cbBitmap is zero.
4067	* @param cbBitmap The size of the bitmap. Must be the size of the whole
4068	* MMIO2 range, rounded up to the nearest 8 bytes.
4069	* When zero only a reset is done.
4070	*/
4071	VMMR3_INT_DECL(int) PGMR3PhysMmio2QueryAndResetDirtyBitmap(PVM pVM, PPDMDEVINS pDevIns, PGMMMIO2HANDLE hMmio2,
4072	void *pvBitmap, size_t cbBitmap)
4073	{
4074	/*
4075	* Do some basic validation before grapping the PGM lock and continuing.
4076	*/
4077	AssertPtrReturn(pDevIns, VERR_INVALID_POINTER);
4078	AssertReturn(RT_ALIGN_Z(cbBitmap, sizeof(uint64_t)) == cbBitmap, VERR_INVALID_PARAMETER);
4079	int rc = PGM_LOCK(pVM);
4080	if (RT_SUCCESS(rc))
4081	{
4082	STAM_PROFILE_START(&pVM->pgm.s.StatMmio2QueryAndResetDirtyBitmap, a);
4083	rc = pgmR3PhysMmio2QueryAndResetDirtyBitmapLocked(pVM, pDevIns, hMmio2, pvBitmap, cbBitmap);
4084	STAM_PROFILE_STOP(&pVM->pgm.s.StatMmio2QueryAndResetDirtyBitmap, a);
4085	PGM_UNLOCK(pVM);
4086	}
4087	return rc;
4088	}
4089
4090
4091	/**
4092	* Worker for PGMR3PhysMmio2ControlDirtyPageTracking
4093	*
4094	* Called owning the PGM lock.
4095	*/
4096	static int pgmR3PhysMmio2ControlDirtyPageTrackingLocked(PVM pVM, PPDMDEVINS pDevIns, PGMMMIO2HANDLE hMmio2, bool fEnabled)
4097	{
4098	/*
4099	* Continue validation.
4100	*/
4101	PPGMREGMMIO2RANGE pFirstRegMmio = pgmR3PhysMmio2Find(pVM, pDevIns, UINT32_MAX, UINT32_MAX, hMmio2);
4102	AssertReturn(pFirstRegMmio, VERR_INVALID_HANDLE);
4103	AssertReturn( (pFirstRegMmio->fFlags & (PGMREGMMIO2RANGE_F_TRACK_DIRTY_PAGES \| PGMREGMMIO2RANGE_F_FIRST_CHUNK))
4104	== (PGMREGMMIO2RANGE_F_TRACK_DIRTY_PAGES \| PGMREGMMIO2RANGE_F_FIRST_CHUNK)
4105	, VERR_INVALID_FUNCTION);
4106	AssertReturn(pDevIns == pFirstRegMmio->pDevInsR3, VERR_NOT_OWNER);
4107
4108	#ifdef VBOX_WITH_PGM_NEM_MODE
4109	/*
4110	* This is a nop if NEM is responsible for doing the tracking, we simply
4111	* leave the tracking on all the time there.
4112	*/
4113	if (pFirstRegMmio->pPhysHandlerR3 == NULL)
4114	{
4115	AssertReturn(VM_IS_NEM_ENABLED(pVM), VERR_INTERNAL_ERROR_4);
4116	return VINF_SUCCESS;
4117	}
4118	#endif
4119
4120	/*
4121	* Anyting needing doing?
4122	*/
4123	if (fEnabled != RT_BOOL(pFirstRegMmio->fFlags & PGMREGMMIO2RANGE_F_TRACKING_ENABLED))
4124	{
4125	LogFlowFunc(("fEnabled=%RTbool %s\n", fEnabled, pFirstRegMmio->RamRange.pszDesc));
4126
4127	/*
4128	* Update the PGMREGMMIO2RANGE_F_TRACKING_ENABLED flag.
4129	*/
4130	for (PPGMREGMMIO2RANGE pCur = pFirstRegMmio;;)
4131	{
4132	if (fEnabled)
4133	pCur->fFlags \|= PGMREGMMIO2RANGE_F_TRACKING_ENABLED;
4134	else
4135	pCur->fFlags &= ~PGMREGMMIO2RANGE_F_TRACKING_ENABLED;
4136	if (pCur->fFlags & PGMREGMMIO2RANGE_F_LAST_CHUNK)
4137	break;
4138	pCur = pCur->pNextR3;
4139	AssertPtrReturn(pCur, VERR_INTERNAL_ERROR_5);
4140	AssertReturn( (pCur->fFlags & (PGMREGMMIO2RANGE_F_TRACK_DIRTY_PAGES \| PGMREGMMIO2RANGE_F_FIRST_CHUNK))
4141	== PGMREGMMIO2RANGE_F_TRACK_DIRTY_PAGES
4142	, VERR_INTERNAL_ERROR_4);
4143	}
4144
4145	/*
4146	* Enable/disable handlers if currently mapped.
4147	*
4148	* We ignore status codes here as we've already changed the flags and
4149	* returning a failure status now would be confusing. Besides, the two
4150	* functions will continue past failures. As argued in the mapping code,
4151	* it's in the release log.
4152	*/
4153	if (pFirstRegMmio->fFlags & PGMREGMMIO2RANGE_F_MAPPED)
4154	{
4155	if (fEnabled)
4156	pgmR3PhysMmio2EnableDirtyPageTracing(pVM, pFirstRegMmio);
4157	else
4158	pgmR3PhysMmio2DisableDirtyPageTracing(pVM, pFirstRegMmio);
4159	}
4160	}
4161	else
4162	LogFlowFunc(("fEnabled=%RTbool %s - no change\n", fEnabled, pFirstRegMmio->RamRange.pszDesc));
4163
4164	return VINF_SUCCESS;
4165	}
4166
4167
4168	/**
4169	* Controls the dirty page tracking for an MMIO2 range.
4170	*
4171	* @returns VBox status code.
4172	* @param pVM The cross context VM structure.
4173	* @param pDevIns The device owning the MMIO2 memory.
4174	* @param hMmio2 The handle of the region.
4175	* @param fEnabled The new tracking state.
4176	*/
4177	VMMR3_INT_DECL(int) PGMR3PhysMmio2ControlDirtyPageTracking(PVM pVM, PPDMDEVINS pDevIns, PGMMMIO2HANDLE hMmio2, bool fEnabled)
4178	{
4179	/*
4180	* Do some basic validation before grapping the PGM lock and continuing.
4181	*/
4182	AssertPtrReturn(pDevIns, VERR_INVALID_POINTER);
4183	int rc = PGM_LOCK(pVM);
4184	if (RT_SUCCESS(rc))
4185	{
4186	rc = pgmR3PhysMmio2ControlDirtyPageTrackingLocked(pVM, pDevIns, hMmio2, fEnabled);
4187	PGM_UNLOCK(pVM);
4188	}
4189	return rc;
4190	}
4191
4192
4193	/**
4194	* Changes the region number of an MMIO2 region.
4195	*
4196	* This is only for dealing with save state issues, nothing else.
4197	*
4198	* @return VBox status code.
4199	*
4200	* @param pVM The cross context VM structure.
4201	* @param pDevIns The device owning the MMIO2 memory.
4202	* @param hMmio2 The handle of the region.
4203	* @param iNewRegion The new region index.
4204	*
4205	* @thread EMT(0)
4206	* @sa @bugref{9359}
4207	*/
4208	VMMR3_INT_DECL(int) PGMR3PhysMmio2ChangeRegionNo(PVM pVM, PPDMDEVINS pDevIns, PGMMMIO2HANDLE hMmio2, uint32_t iNewRegion)
4209	{
4210	/*
4211	* Validate input.
4212	*/
4213	VM_ASSERT_EMT0_RETURN(pVM, VERR_VM_THREAD_NOT_EMT);
4214	VM_ASSERT_STATE_RETURN(pVM, VMSTATE_LOADING, VERR_VM_INVALID_VM_STATE);
4215	AssertPtrReturn(pDevIns, VERR_INVALID_PARAMETER);
4216	AssertReturn(hMmio2 != NIL_PGMMMIO2HANDLE, VERR_INVALID_HANDLE);
4217	AssertReturn(iNewRegion <= UINT8_MAX, VERR_INVALID_PARAMETER);
4218
4219	AssertReturn(pVM->enmVMState == VMSTATE_LOADING, VERR_INVALID_STATE);
4220
4221	int rc = PGM_LOCK(pVM);
4222	AssertRCReturn(rc, rc);
4223
4224	PPGMREGMMIO2RANGE pFirstRegMmio = pgmR3PhysMmio2Find(pVM, pDevIns, UINT32_MAX, UINT32_MAX, hMmio2);
4225	AssertReturnStmt(pFirstRegMmio, PGM_UNLOCK(pVM), VERR_NOT_FOUND);
4226	AssertReturnStmt(pgmR3PhysMmio2Find(pVM, pDevIns, pFirstRegMmio->iSubDev, iNewRegion, NIL_PGMMMIO2HANDLE) == NULL,
4227	PGM_UNLOCK(pVM), VERR_RESOURCE_IN_USE);
4228
4229	/*
4230	* Make the change.
4231	*/
4232	pFirstRegMmio->iRegion = (uint8_t)iNewRegion;
4233
4234	PGM_UNLOCK(pVM);
4235	return VINF_SUCCESS;
4236	}
4237
4238
4239
4240	/*********************************************************************************************************************************
4241	* ROM *
4242	*********************************************************************************************************************************/
4243
4244	/**
4245	* Worker for PGMR3PhysRomRegister.
4246	*
4247	* This is here to simplify lock management, i.e. the caller does all the
4248	* locking and we can simply return without needing to remember to unlock
4249	* anything first.
4250	*
4251	* @returns VBox status code.
4252	* @param pVM The cross context VM structure.
4253	* @param pDevIns The device instance owning the ROM.
4254	* @param GCPhys First physical address in the range.
4255	* Must be page aligned!
4256	* @param cb The size of the range (in bytes).
4257	* Must be page aligned!
4258	* @param pvBinary Pointer to the binary data backing the ROM image.
4259	* @param cbBinary The size of the binary data pvBinary points to.
4260	* This must be less or equal to @a cb.
4261	* @param fFlags Mask of flags. PGMPHYS_ROM_FLAGS_SHADOWED
4262	* and/or PGMPHYS_ROM_FLAGS_PERMANENT_BINARY.
4263	* @param pszDesc Pointer to description string. This must not be freed.
4264	*/
4265	static int pgmR3PhysRomRegisterLocked(PVM pVM, PPDMDEVINS pDevIns, RTGCPHYS GCPhys, RTGCPHYS cb,
4266	const void pvBinary, uint32_t cbBinary, uint8_t fFlags, const char pszDesc)
4267	{
4268	/*
4269	* Validate input.
4270	*/
4271	AssertPtrReturn(pDevIns, VERR_INVALID_PARAMETER);
4272	AssertReturn(RT_ALIGN_T(GCPhys, GUEST_PAGE_SIZE, RTGCPHYS) == GCPhys, VERR_INVALID_PARAMETER);
4273	AssertReturn(RT_ALIGN_T(cb, GUEST_PAGE_SIZE, RTGCPHYS) == cb, VERR_INVALID_PARAMETER);
4274	RTGCPHYS GCPhysLast = GCPhys + (cb - 1);
4275	AssertReturn(GCPhysLast > GCPhys, VERR_INVALID_PARAMETER);
4276	AssertPtrReturn(pvBinary, VERR_INVALID_PARAMETER);
4277	AssertPtrReturn(pszDesc, VERR_INVALID_POINTER);
4278	AssertReturn(!(fFlags & ~PGMPHYS_ROM_FLAGS_VALID_MASK), VERR_INVALID_PARAMETER);
4279	VM_ASSERT_STATE_RETURN(pVM, VMSTATE_CREATING, VERR_VM_INVALID_VM_STATE);
4280
4281	const uint32_t cGuestPages = cb >> GUEST_PAGE_SHIFT;
4282	#ifdef VBOX_WITH_PGM_NEM_MODE
4283	const uint32_t cHostPages = RT_ALIGN_T(cb, HOST_PAGE_SIZE, RTGCPHYS) >> HOST_PAGE_SHIFT;
4284	#endif
4285
4286	/*
4287	* Find the ROM location in the ROM list first.
4288	*/
4289	PPGMROMRANGE pRomPrev = NULL;
4290	PPGMROMRANGE pRom = pVM->pgm.s.pRomRangesR3;
4291	while (pRom && GCPhysLast >= pRom->GCPhys)
4292	{
4293	if ( GCPhys <= pRom->GCPhysLast
4294	&& GCPhysLast >= pRom->GCPhys)
4295	AssertLogRelMsgFailedReturn(("%RGp-%RGp (%s) conflicts with existing %RGp-%RGp (%s)\n",
4296	GCPhys, GCPhysLast, pszDesc,
4297	pRom->GCPhys, pRom->GCPhysLast, pRom->pszDesc),
4298	VERR_PGM_RAM_CONFLICT);
4299	/* next */
4300	pRomPrev = pRom;
4301	pRom = pRom->pNextR3;
4302	}
4303
4304	/*
4305	* Find the RAM location and check for conflicts.
4306	*
4307	* Conflict detection is a bit different than for RAM registration since a
4308	* ROM can be located within a RAM range. So, what we have to check for is
4309	* other memory types (other than RAM that is) and that we don't span more
4310	* than one RAM range (lazy).
4311	*/
4312	bool fRamExists = false;
4313	PPGMRAMRANGE pRamPrev = NULL;
4314	PPGMRAMRANGE pRam = pVM->pgm.s.pRamRangesXR3;
4315	while (pRam && GCPhysLast >= pRam->GCPhys)
4316	{
4317	if ( GCPhys <= pRam->GCPhysLast
4318	&& GCPhysLast >= pRam->GCPhys)
4319	{
4320	/* completely within? */
4321	AssertLogRelMsgReturn( GCPhys >= pRam->GCPhys
4322	&& GCPhysLast <= pRam->GCPhysLast,
4323	("%RGp-%RGp (%s) falls partly outside %RGp-%RGp (%s)\n",
4324	GCPhys, GCPhysLast, pszDesc,
4325	pRam->GCPhys, pRam->GCPhysLast, pRam->pszDesc),
4326	VERR_PGM_RAM_CONFLICT);
4327	fRamExists = true;
4328	break;
4329	}
4330
4331	/* next */
4332	pRamPrev = pRam;
4333	pRam = pRam->pNextR3;
4334	}
4335	if (fRamExists)
4336	{
4337	PPGMPAGE pPage = &pRam->aPages[(GCPhys - pRam->GCPhys) >> GUEST_PAGE_SHIFT];
4338	uint32_t cPagesLeft = cGuestPages;
4339	while (cPagesLeft-- > 0)
4340	{
4341	AssertLogRelMsgReturn(PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_RAM,
4342	("%RGp (%R[pgmpage]) isn't a RAM page - registering %RGp-%RGp (%s).\n",
4343	pRam->GCPhys + ((RTGCPHYS)(uintptr_t)(pPage - &pRam->aPages[0]) << GUEST_PAGE_SHIFT),
4344	pPage, GCPhys, GCPhysLast, pszDesc), VERR_PGM_RAM_CONFLICT);
4345	Assert(PGM_PAGE_IS_ZERO(pPage) \|\| PGM_IS_IN_NEM_MODE(pVM));
4346	pPage++;
4347	}
4348	}
4349
4350	/*
4351	* Update the base memory reservation if necessary.
4352	*/
4353	uint32_t cExtraBaseCost = fRamExists ? 0 : cGuestPages;
4354	if (fFlags & PGMPHYS_ROM_FLAGS_SHADOWED)
4355	cExtraBaseCost += cGuestPages;
4356	if (cExtraBaseCost)
4357	{
4358	int rc = MMR3IncreaseBaseReservation(pVM, cExtraBaseCost);
4359	if (RT_FAILURE(rc))
4360	return rc;
4361	}
4362
4363	#ifdef VBOX_WITH_NATIVE_NEM
4364	/*
4365	* Early NEM notification before we've made any changes or anything.
4366	*/
4367	uint32_t const fNemNotify = (fRamExists ? NEM_NOTIFY_PHYS_ROM_F_REPLACE : 0)
4368	\| (fFlags & PGMPHYS_ROM_FLAGS_SHADOWED ? NEM_NOTIFY_PHYS_ROM_F_SHADOW : 0);
4369	uint8_t u2NemState = UINT8_MAX;
4370	uint32_t uNemRange = 0;
4371	if (VM_IS_NEM_ENABLED(pVM))
4372	{
4373	int rc = NEMR3NotifyPhysRomRegisterEarly(pVM, GCPhys, cGuestPages << GUEST_PAGE_SHIFT,
4374	fRamExists ? PGM_RAMRANGE_CALC_PAGE_R3PTR(pRam, GCPhys) : NULL,
4375	fNemNotify, &u2NemState, fRamExists ? &pRam->uNemRange : &uNemRange);
4376	AssertLogRelRCReturn(rc, rc);
4377	}
4378	#endif
4379
4380	/*
4381	* Allocate memory for the virgin copy of the RAM. In simplified memory mode,
4382	* we allocate memory for any ad-hoc RAM range and for shadow pages.
4383	*/
4384	PGMMALLOCATEPAGESREQ pReq = NULL;
4385	#ifdef VBOX_WITH_PGM_NEM_MODE
4386	void *pvRam = NULL;
4387	void *pvAlt = NULL;
4388	if (pVM->pgm.s.fNemMode)
4389	{
4390	if (!fRamExists)
4391	{
4392	int rc = SUPR3PageAlloc(cHostPages, 0, &pvRam);
4393	if (RT_FAILURE(rc))
4394	return rc;
4395	}
4396	if (fFlags & PGMPHYS_ROM_FLAGS_SHADOWED)
4397	{
4398	int rc = SUPR3PageAlloc(cHostPages, 0, &pvAlt);
4399	if (RT_FAILURE(rc))
4400	{
4401	if (pvRam)
4402	SUPR3PageFree(pvRam, cHostPages);
4403	return rc;
4404	}
4405	}
4406	}
4407	else
4408	#endif
4409	{
4410	int rc = GMMR3AllocatePagesPrepare(pVM, &pReq, cGuestPages, GMMACCOUNT_BASE);
4411	AssertRCReturn(rc, rc);
4412
4413	for (uint32_t iPage = 0; iPage < cGuestPages; iPage++)
4414	{
4415	pReq->aPages[iPage].HCPhysGCPhys = GCPhys + (iPage << GUEST_PAGE_SHIFT);
4416	pReq->aPages[iPage].fZeroed = false;
4417	pReq->aPages[iPage].idPage = NIL_GMM_PAGEID;
4418	pReq->aPages[iPage].idSharedPage = NIL_GMM_PAGEID;
4419	}
4420
4421	rc = GMMR3AllocatePagesPerform(pVM, pReq);
4422	if (RT_FAILURE(rc))
4423	{
4424	GMMR3AllocatePagesCleanup(pReq);
4425	return rc;
4426	}
4427	}
4428
4429	/*
4430	* Allocate the new ROM range and RAM range (if necessary).
4431	*/
4432	PPGMROMRANGE pRomNew = NULL;
4433	RTR0PTR pRomNewR0 = NIL_RTR0PTR;
4434	size_t const cbRomRange = RT_ALIGN_Z(RT_UOFFSETOF_DYN(PGMROMRANGE, aPages[cGuestPages]), 128);
4435	size_t const cbRamRange = fRamExists ? 0 : RT_UOFFSETOF_DYN(PGMROMRANGE, aPages[cGuestPages]);
4436	size_t const cRangePages = RT_ALIGN_Z(cbRomRange + cbRamRange, HOST_PAGE_SIZE) >> HOST_PAGE_SHIFT;
4437	int rc = SUPR3PageAllocEx(cRangePages, 0 /fFlags/, (void *)&pRomNew, &pRomNewR0, NULL /paPages*/);
4438	if (RT_SUCCESS(rc))
4439	{
4440
4441	/*
4442	* Initialize and insert the RAM range (if required).
4443	*/
4444	PPGMRAMRANGE pRamNew;
4445	uint32_t const idxFirstRamPage = fRamExists ? (GCPhys - pRam->GCPhys) >> GUEST_PAGE_SHIFT : 0;
4446	PPGMROMPAGE pRomPage = &pRomNew->aPages[0];
4447	if (!fRamExists)
4448	{
4449	/* New RAM range. */
4450	pRamNew = (PPGMRAMRANGE)((uintptr_t)pRomNew + cbRomRange);
4451	pRamNew->pSelfR0 = !pRomNewR0 ? NIL_RTR0PTR : pRomNewR0 + cbRomRange;
4452	pRamNew->GCPhys = GCPhys;
4453	pRamNew->GCPhysLast = GCPhysLast;
4454	pRamNew->cb = cb;
4455	pRamNew->pszDesc = pszDesc;
4456	pRamNew->fFlags = PGM_RAM_RANGE_FLAGS_AD_HOC_ROM;
4457	pRamNew->pvR3 = NULL;
4458	pRamNew->paLSPages = NULL;
4459	#ifdef VBOX_WITH_NATIVE_NEM
4460	pRamNew->uNemRange = uNemRange;
4461	#endif
4462
4463	PPGMPAGE pRamPage = &pRamNew->aPages[idxFirstRamPage];
4464	#ifdef VBOX_WITH_PGM_NEM_MODE
4465	if (pVM->pgm.s.fNemMode)
4466	{
4467	AssertPtr(pvRam); Assert(pReq == NULL);
4468	pRamNew->pvR3 = pvRam;
4469	for (uint32_t iPage = 0; iPage < cGuestPages; iPage++, pRamPage++, pRomPage++)
4470	{
4471	PGM_PAGE_INIT(pRamPage, UINT64_C(0x0000fffffffff000), NIL_GMM_PAGEID,
4472	PGMPAGETYPE_ROM, PGM_PAGE_STATE_ALLOCATED);
4473	pRomPage->Virgin = *pRamPage;
4474	}
4475	}
4476	else
4477	#endif
4478	for (uint32_t iPage = 0; iPage < cGuestPages; iPage++, pRamPage++, pRomPage++)
4479	{
4480	PGM_PAGE_INIT(pRamPage,
4481	pReq->aPages[iPage].HCPhysGCPhys,
4482	pReq->aPages[iPage].idPage,
4483	PGMPAGETYPE_ROM,
4484	PGM_PAGE_STATE_ALLOCATED);
4485
4486	pRomPage->Virgin = *pRamPage;
4487	}
4488
4489	pVM->pgm.s.cAllPages += cGuestPages;
4490	pVM->pgm.s.cPrivatePages += cGuestPages;
4491	pgmR3PhysLinkRamRange(pVM, pRamNew, pRamPrev);
4492	}
4493	else
4494	{
4495	/* Existing RAM range. */
4496	PPGMPAGE pRamPage = &pRam->aPages[idxFirstRamPage];
4497	#ifdef VBOX_WITH_PGM_NEM_MODE
4498	if (pVM->pgm.s.fNemMode)
4499	{
4500	Assert(pvRam == NULL); Assert(pReq == NULL);
4501	for (uint32_t iPage = 0; iPage < cGuestPages; iPage++, pRamPage++, pRomPage++)
4502	{
4503	Assert(PGM_PAGE_GET_HCPHYS(pRamPage) == UINT64_C(0x0000fffffffff000));
4504	Assert(PGM_PAGE_GET_PAGEID(pRamPage) == NIL_GMM_PAGEID);
4505	Assert(PGM_PAGE_GET_STATE(pRamPage) == PGM_PAGE_STATE_ALLOCATED);
4506	PGM_PAGE_SET_TYPE(pVM, pRamPage, PGMPAGETYPE_ROM);
4507	PGM_PAGE_SET_STATE(pVM, pRamPage, PGM_PAGE_STATE_ALLOCATED);
4508	PGM_PAGE_SET_PDE_TYPE(pVM, pRamPage, PGM_PAGE_PDE_TYPE_DONTCARE);
4509	PGM_PAGE_SET_PTE_INDEX(pVM, pRamPage, 0);
4510	PGM_PAGE_SET_TRACKING(pVM, pRamPage, 0);
4511
4512	pRomPage->Virgin = *pRamPage;
4513	}
4514	}
4515	else
4516	#endif
4517	{
4518	for (uint32_t iPage = 0; iPage < cGuestPages; iPage++, pRamPage++, pRomPage++)
4519	{
4520	PGM_PAGE_SET_TYPE(pVM, pRamPage, PGMPAGETYPE_ROM);
4521	PGM_PAGE_SET_HCPHYS(pVM, pRamPage, pReq->aPages[iPage].HCPhysGCPhys);
4522	PGM_PAGE_SET_STATE(pVM, pRamPage, PGM_PAGE_STATE_ALLOCATED);
4523	PGM_PAGE_SET_PAGEID(pVM, pRamPage, pReq->aPages[iPage].idPage);
4524	PGM_PAGE_SET_PDE_TYPE(pVM, pRamPage, PGM_PAGE_PDE_TYPE_DONTCARE);
4525	PGM_PAGE_SET_PTE_INDEX(pVM, pRamPage, 0);
4526	PGM_PAGE_SET_TRACKING(pVM, pRamPage, 0);
4527
4528	pRomPage->Virgin = *pRamPage;
4529	}
4530	pVM->pgm.s.cZeroPages -= cGuestPages;
4531	pVM->pgm.s.cPrivatePages += cGuestPages;
4532	}
4533	pRamNew = pRam;
4534	}
4535
4536	#ifdef VBOX_WITH_NATIVE_NEM
4537	/* Set the NEM state of the pages if needed. */
4538	if (u2NemState != UINT8_MAX)
4539	pgmPhysSetNemStateForPages(&pRamNew->aPages[idxFirstRamPage], cGuestPages, u2NemState);
4540	#endif
4541
4542	/* Flush physical page map TLB. */
4543	pgmPhysInvalidatePageMapTLB(pVM);
4544
4545	/*
4546	* Register the ROM access handler.
4547	*/
4548	rc = PGMHandlerPhysicalRegister(pVM, GCPhys, GCPhysLast, pVM->pgm.s.hRomPhysHandlerType, GCPhys, pszDesc);
4549	if (RT_SUCCESS(rc))
4550	{
4551	/*
4552	* Copy the image over to the virgin pages.
4553	* This must be done after linking in the RAM range.
4554	*/
4555	size_t cbBinaryLeft = cbBinary;
4556	PPGMPAGE pRamPage = &pRamNew->aPages[idxFirstRamPage];
4557	for (uint32_t iPage = 0; iPage < cGuestPages; iPage++, pRamPage++)
4558	{
4559	void *pvDstPage;
4560	rc = pgmPhysPageMap(pVM, pRamPage, GCPhys + (iPage << GUEST_PAGE_SHIFT), &pvDstPage);
4561	if (RT_FAILURE(rc))
4562	{
4563	VMSetError(pVM, rc, RT_SRC_POS, "Failed to map virgin ROM page at %RGp", GCPhys);
4564	break;
4565	}
4566	if (cbBinaryLeft >= GUEST_PAGE_SIZE)
4567	{
4568	memcpy(pvDstPage, (uint8_t const *)pvBinary + ((size_t)iPage << GUEST_PAGE_SHIFT), GUEST_PAGE_SIZE);
4569	cbBinaryLeft -= GUEST_PAGE_SIZE;
4570	}
4571	else
4572	{
4573	RT_BZERO(pvDstPage, GUEST_PAGE_SIZE); /* (shouldn't be necessary, but can't hurt either) */
4574	if (cbBinaryLeft > 0)
4575	{
4576	memcpy(pvDstPage, (uint8_t const *)pvBinary + ((size_t)iPage << GUEST_PAGE_SHIFT), cbBinaryLeft);
4577	cbBinaryLeft = 0;
4578	}
4579	}
4580	}
4581	if (RT_SUCCESS(rc))
4582	{
4583	/*
4584	* Initialize the ROM range.
4585	* Note that the Virgin member of the pages has already been initialized above.
4586	*/
4587	pRomNew->pSelfR0 = pRomNewR0;
4588	pRomNew->GCPhys = GCPhys;
4589	pRomNew->GCPhysLast = GCPhysLast;
4590	pRomNew->cb = cb;
4591	pRomNew->fFlags = fFlags;
4592	pRomNew->idSavedState = UINT8_MAX;
4593	pRomNew->cbOriginal = cbBinary;
4594	pRomNew->pszDesc = pszDesc;
4595	#ifdef VBOX_WITH_PGM_NEM_MODE
4596	pRomNew->pbR3Alternate = (uint8_t *)pvAlt;
4597	#endif
4598	pRomNew->pvOriginal = fFlags & PGMPHYS_ROM_FLAGS_PERMANENT_BINARY
4599	? pvBinary : RTMemDup(pvBinary, cbBinary);
4600	if (pRomNew->pvOriginal)
4601	{
4602	for (unsigned iPage = 0; iPage < cGuestPages; iPage++)
4603	{
4604	PPGMROMPAGE pPage = &pRomNew->aPages[iPage];
4605	pPage->enmProt = PGMROMPROT_READ_ROM_WRITE_IGNORE;
4606	#ifdef VBOX_WITH_PGM_NEM_MODE
4607	if (pVM->pgm.s.fNemMode)
4608	PGM_PAGE_INIT(&pPage->Shadow, UINT64_C(0x0000fffffffff000), NIL_GMM_PAGEID,
4609	PGMPAGETYPE_ROM_SHADOW, PGM_PAGE_STATE_ALLOCATED);
4610	else
4611	#endif
4612	PGM_PAGE_INIT_ZERO(&pPage->Shadow, pVM, PGMPAGETYPE_ROM_SHADOW);
4613	}
4614
4615	/* update the page count stats for the shadow pages. */
4616	if (fFlags & PGMPHYS_ROM_FLAGS_SHADOWED)
4617	{
4618	#ifdef VBOX_WITH_PGM_NEM_MODE
4619	if (pVM->pgm.s.fNemMode)
4620	pVM->pgm.s.cPrivatePages += cGuestPages;
4621	else
4622	#endif
4623	pVM->pgm.s.cZeroPages += cGuestPages;
4624	pVM->pgm.s.cAllPages += cGuestPages;
4625	}
4626
4627	/*
4628	* Insert the ROM range, tell REM and return successfully.
4629	*/
4630	pRomNew->pNextR3 = pRom;
4631	pRomNew->pNextR0 = pRom ? pRom->pSelfR0 : NIL_RTR0PTR;
4632
4633	if (pRomPrev)
4634	{
4635	pRomPrev->pNextR3 = pRomNew;
4636	pRomPrev->pNextR0 = pRomNew->pSelfR0;
4637	}
4638	else
4639	{
4640	pVM->pgm.s.pRomRangesR3 = pRomNew;
4641	pVM->pgm.s.pRomRangesR0 = pRomNew->pSelfR0;
4642	}
4643
4644	pgmPhysInvalidatePageMapTLB(pVM);
4645	#ifdef VBOX_WITH_PGM_NEM_MODE
4646	if (!pVM->pgm.s.fNemMode)
4647	#endif
4648	GMMR3AllocatePagesCleanup(pReq);
4649
4650	#ifdef VBOX_WITH_NATIVE_NEM
4651	/*
4652	* Notify NEM again.
4653	*/
4654	if (VM_IS_NEM_ENABLED(pVM))
4655	{
4656	u2NemState = UINT8_MAX;
4657	rc = NEMR3NotifyPhysRomRegisterLate(pVM, GCPhys, cb, PGM_RAMRANGE_CALC_PAGE_R3PTR(pRamNew, GCPhys),
4658	fNemNotify, &u2NemState,
4659	fRamExists ? &pRam->uNemRange : &pRamNew->uNemRange);
4660	if (u2NemState != UINT8_MAX)
4661	pgmPhysSetNemStateForPages(&pRamNew->aPages[idxFirstRamPage], cGuestPages, u2NemState);
4662	if (RT_SUCCESS(rc))
4663	return rc;
4664	}
4665	else
4666	#endif
4667	return rc;
4668
4669	/*
4670	* bail out
4671	*/
4672	#ifdef VBOX_WITH_NATIVE_NEM
4673	/* unlink */
4674	if (pRomPrev)
4675	{
4676	pRomPrev->pNextR3 = pRom;
4677	pRomPrev->pNextR0 = pRom ? pRom->pSelfR0 : NIL_RTR0PTR;
4678	}
4679	else
4680	{
4681	pVM->pgm.s.pRomRangesR3 = pRom;
4682	pVM->pgm.s.pRomRangesR0 = pRom ? pRom->pSelfR0 : NIL_RTR0PTR;
4683	}
4684
4685	if (fFlags & PGMPHYS_ROM_FLAGS_SHADOWED)
4686	{
4687	# ifdef VBOX_WITH_PGM_NEM_MODE
4688	if (pVM->pgm.s.fNemMode)
4689	pVM->pgm.s.cPrivatePages -= cGuestPages;
4690	else
4691	# endif
4692	pVM->pgm.s.cZeroPages -= cGuestPages;
4693	pVM->pgm.s.cAllPages -= cGuestPages;
4694	}
4695	#endif
4696	}
4697	else
4698	rc = VERR_NO_MEMORY;
4699	}
4700
4701	int rc2 = PGMHandlerPhysicalDeregister(pVM, GCPhys);
4702	AssertRC(rc2);
4703	}
4704
4705	if (!fRamExists)
4706	pgmR3PhysUnlinkRamRange2(pVM, pRamNew, pRamPrev);
4707	else
4708	{
4709	PPGMPAGE pRamPage = &pRam->aPages[idxFirstRamPage];
4710	#ifdef VBOX_WITH_PGM_NEM_MODE
4711	if (pVM->pgm.s.fNemMode)
4712	{
4713	Assert(pvRam == NULL); Assert(pReq == NULL);
4714	for (uint32_t iPage = 0; iPage < cGuestPages; iPage++, pRamPage++, pRomPage++)
4715	{
4716	Assert(PGM_PAGE_GET_HCPHYS(pRamPage) == UINT64_C(0x0000fffffffff000));
4717	Assert(PGM_PAGE_GET_PAGEID(pRamPage) == NIL_GMM_PAGEID);
4718	Assert(PGM_PAGE_GET_STATE(pRamPage) == PGM_PAGE_STATE_ALLOCATED);
4719	PGM_PAGE_SET_TYPE(pVM, pRamPage, PGMPAGETYPE_RAM);
4720	PGM_PAGE_SET_STATE(pVM, pRamPage, PGM_PAGE_STATE_ALLOCATED);
4721	}
4722	}
4723	else
4724	#endif
4725	{
4726	for (uint32_t iPage = 0; iPage < cGuestPages; iPage++, pRamPage++)
4727	PGM_PAGE_INIT_ZERO(pRamPage, pVM, PGMPAGETYPE_RAM);
4728	pVM->pgm.s.cZeroPages += cGuestPages;
4729	pVM->pgm.s.cPrivatePages -= cGuestPages;
4730	}
4731	}
4732
4733	SUPR3PageFreeEx(pRomNew, cRangePages);
4734	}
4735
4736	/** @todo Purge the mapping cache or something... */
4737	#ifdef VBOX_WITH_PGM_NEM_MODE
4738	if (pVM->pgm.s.fNemMode)
4739	{
4740	Assert(!pReq);
4741	if (pvRam)
4742	SUPR3PageFree(pvRam, cHostPages);
4743	if (pvAlt)
4744	SUPR3PageFree(pvAlt, cHostPages);
4745	}
4746	else
4747	#endif
4748	{
4749	GMMR3FreeAllocatedPages(pVM, pReq);
4750	GMMR3AllocatePagesCleanup(pReq);
4751	}
4752	return rc;
4753	}
4754
4755
4756	/**
4757	* Registers a ROM image.
4758	*
4759	* Shadowed ROM images requires double the amount of backing memory, so,
4760	* don't use that unless you have to. Shadowing of ROM images is process
4761	* where we can select where the reads go and where the writes go. On real
4762	* hardware the chipset provides means to configure this. We provide
4763	* PGMR3PhysProtectROM() for this purpose.
4764	*
4765	* A read-only copy of the ROM image will always be kept around while we
4766	* will allocate RAM pages for the changes on demand (unless all memory
4767	* is configured to be preallocated).
4768	*
4769	* @returns VBox status code.
4770	* @param pVM The cross context VM structure.
4771	* @param pDevIns The device instance owning the ROM.
4772	* @param GCPhys First physical address in the range.
4773	* Must be page aligned!
4774	* @param cb The size of the range (in bytes).
4775	* Must be page aligned!
4776	* @param pvBinary Pointer to the binary data backing the ROM image.
4777	* @param cbBinary The size of the binary data pvBinary points to.
4778	* This must be less or equal to @a cb.
4779	* @param fFlags Mask of flags, PGMPHYS_ROM_FLAGS_XXX.
4780	* @param pszDesc Pointer to description string. This must not be freed.
4781	*
4782	* @remark There is no way to remove the rom, automatically on device cleanup or
4783	* manually from the device yet. This isn't difficult in any way, it's
4784	* just not something we expect to be necessary for a while.
4785	*/
4786	VMMR3DECL(int) PGMR3PhysRomRegister(PVM pVM, PPDMDEVINS pDevIns, RTGCPHYS GCPhys, RTGCPHYS cb,
4787	const void pvBinary, uint32_t cbBinary, uint8_t fFlags, const char pszDesc)
4788	{
4789	Log(("PGMR3PhysRomRegister: pDevIns=%p GCPhys=%RGp(-%RGp) cb=%RGp pvBinary=%p cbBinary=%#x fFlags=%#x pszDesc=%s\n",
4790	pDevIns, GCPhys, GCPhys + cb, cb, pvBinary, cbBinary, fFlags, pszDesc));
4791	PGM_LOCK_VOID(pVM);
4792	int rc = pgmR3PhysRomRegisterLocked(pVM, pDevIns, GCPhys, cb, pvBinary, cbBinary, fFlags, pszDesc);
4793	PGM_UNLOCK(pVM);
4794	return rc;
4795	}
4796
4797
4798	/**
4799	* Called by PGMR3MemSetup to reset the shadow, switch to the virgin, and verify
4800	* that the virgin part is untouched.
4801	*
4802	* This is done after the normal memory has been cleared.
4803	*
4804	* ASSUMES that the caller owns the PGM lock.
4805	*
4806	* @param pVM The cross context VM structure.
4807	*/
4808	int pgmR3PhysRomReset(PVM pVM)
4809	{
4810	PGM_LOCK_ASSERT_OWNER(pVM);
4811	for (PPGMROMRANGE pRom = pVM->pgm.s.pRomRangesR3; pRom; pRom = pRom->pNextR3)
4812	{
4813	const uint32_t cGuestPages = pRom->cb >> GUEST_PAGE_SHIFT;
4814
4815	if (pRom->fFlags & PGMPHYS_ROM_FLAGS_SHADOWED)
4816	{
4817	/*
4818	* Reset the physical handler.
4819	*/
4820	int rc = PGMR3PhysRomProtect(pVM, pRom->GCPhys, pRom->cb, PGMROMPROT_READ_ROM_WRITE_IGNORE);
4821	AssertRCReturn(rc, rc);
4822
4823	/*
4824	* What we do with the shadow pages depends on the memory
4825	* preallocation option. If not enabled, we'll just throw
4826	* out all the dirty pages and replace them by the zero page.
4827	*/
4828	#ifdef VBOX_WITH_PGM_NEM_MODE
4829	if (pVM->pgm.s.fNemMode)
4830	{
4831	/* Clear all the shadow pages (currently using alternate backing). */
4832	RT_BZERO(pRom->pbR3Alternate, pRom->cb);
4833	}
4834	else
4835	#endif
4836	if (!pVM->pgm.s.fRamPreAlloc)
4837	{
4838	/* Free the dirty pages. */
4839	uint32_t cPendingPages = 0;
4840	PGMMFREEPAGESREQ pReq;
4841	rc = GMMR3FreePagesPrepare(pVM, &pReq, PGMPHYS_FREE_PAGE_BATCH_SIZE, GMMACCOUNT_BASE);
4842	AssertRCReturn(rc, rc);
4843
4844	for (uint32_t iPage = 0; iPage < cGuestPages; iPage++)
4845	if ( !PGM_PAGE_IS_ZERO(&pRom->aPages[iPage].Shadow)
4846	&& !PGM_PAGE_IS_BALLOONED(&pRom->aPages[iPage].Shadow))
4847	{
4848	Assert(PGM_PAGE_GET_STATE(&pRom->aPages[iPage].Shadow) == PGM_PAGE_STATE_ALLOCATED);
4849	rc = pgmPhysFreePage(pVM, pReq, &cPendingPages, &pRom->aPages[iPage].Shadow,
4850	pRom->GCPhys + (iPage << GUEST_PAGE_SHIFT),
4851	(PGMPAGETYPE)PGM_PAGE_GET_TYPE(&pRom->aPages[iPage].Shadow));
4852	AssertLogRelRCReturn(rc, rc);
4853	}
4854
4855	if (cPendingPages)
4856	{
4857	rc = GMMR3FreePagesPerform(pVM, pReq, cPendingPages);
4858	AssertLogRelRCReturn(rc, rc);
4859	}
4860	GMMR3FreePagesCleanup(pReq);
4861	}
4862	else
4863	{
4864	/* clear all the shadow pages. */
4865	for (uint32_t iPage = 0; iPage < cGuestPages; iPage++)
4866	{
4867	if (PGM_PAGE_IS_ZERO(&pRom->aPages[iPage].Shadow))
4868	continue;
4869	Assert(!PGM_PAGE_IS_BALLOONED(&pRom->aPages[iPage].Shadow));
4870	void *pvDstPage;
4871	const RTGCPHYS GCPhys = pRom->GCPhys + (iPage << GUEST_PAGE_SHIFT);
4872	rc = pgmPhysPageMakeWritableAndMap(pVM, &pRom->aPages[iPage].Shadow, GCPhys, &pvDstPage);
4873	if (RT_FAILURE(rc))
4874	break;
4875	RT_BZERO(pvDstPage, GUEST_PAGE_SIZE);
4876	}
4877	AssertRCReturn(rc, rc);
4878	}
4879	}
4880
4881	/*
4882	* Restore the original ROM pages after a saved state load.
4883	* Also, in strict builds check that ROM pages remain unmodified.
4884	*/
4885	#ifndef VBOX_STRICT
4886	if (pVM->pgm.s.fRestoreRomPagesOnReset)
4887	#endif
4888	{
4889	size_t cbSrcLeft = pRom->cbOriginal;
4890	uint8_t const pbSrcPage = (uint8_t const )pRom->pvOriginal;
4891	uint32_t cRestored = 0;
4892	for (uint32_t iPage = 0; iPage < cGuestPages && cbSrcLeft > 0; iPage++, pbSrcPage += GUEST_PAGE_SIZE)
4893	{
4894	const RTGCPHYS GCPhys = pRom->GCPhys + (iPage << GUEST_PAGE_SHIFT);
4895	PPGMPAGE const pPage = pgmPhysGetPage(pVM, GCPhys);
4896	void const *pvDstPage = NULL;
4897	int rc = pgmPhysPageMapReadOnly(pVM, pPage, GCPhys, &pvDstPage);
4898	if (RT_FAILURE(rc))
4899	break;
4900
4901	if (memcmp(pvDstPage, pbSrcPage, RT_MIN(cbSrcLeft, GUEST_PAGE_SIZE)))
4902	{
4903	if (pVM->pgm.s.fRestoreRomPagesOnReset)
4904	{
4905	void *pvDstPageW = NULL;
4906	rc = pgmPhysPageMap(pVM, pPage, GCPhys, &pvDstPageW);
4907	AssertLogRelRCReturn(rc, rc);
4908	memcpy(pvDstPageW, pbSrcPage, RT_MIN(cbSrcLeft, GUEST_PAGE_SIZE));
4909	cRestored++;
4910	}
4911	else
4912	LogRel(("pgmR3PhysRomReset: %RGp: ROM page changed (%s)\n", GCPhys, pRom->pszDesc));
4913	}
4914	cbSrcLeft -= RT_MIN(cbSrcLeft, GUEST_PAGE_SIZE);
4915	}
4916	if (cRestored > 0)
4917	LogRel(("PGM: ROM \"%s\": Reloaded %u of %u pages.\n", pRom->pszDesc, cRestored, cGuestPages));
4918	}
4919	}
4920
4921	/* Clear the ROM restore flag now as we only need to do this once after
4922	loading saved state. */
4923	pVM->pgm.s.fRestoreRomPagesOnReset = false;
4924
4925	return VINF_SUCCESS;
4926	}
4927
4928
4929	/**
4930	* Called by PGMR3Term to free resources.
4931	*
4932	* ASSUMES that the caller owns the PGM lock.
4933	*
4934	* @param pVM The cross context VM structure.
4935	*/
4936	void pgmR3PhysRomTerm(PVM pVM)
4937	{
4938	/*
4939	* Free the heap copy of the original bits.
4940	*/
4941	for (PPGMROMRANGE pRom = pVM->pgm.s.pRomRangesR3; pRom; pRom = pRom->pNextR3)
4942	{
4943	if ( pRom->pvOriginal
4944	&& !(pRom->fFlags & PGMPHYS_ROM_FLAGS_PERMANENT_BINARY))
4945	{
4946	RTMemFree((void *)pRom->pvOriginal);
4947	pRom->pvOriginal = NULL;
4948	}
4949	}
4950	}
4951
4952
4953	/**
4954	* Change the shadowing of a range of ROM pages.
4955	*
4956	* This is intended for implementing chipset specific memory registers
4957	* and will not be very strict about the input. It will silently ignore
4958	* any pages that are not the part of a shadowed ROM.
4959	*
4960	* @returns VBox status code.
4961	* @retval VINF_PGM_SYNC_CR3
4962	*
4963	* @param pVM The cross context VM structure.
4964	* @param GCPhys Where to start. Page aligned.
4965	* @param cb How much to change. Page aligned.
4966	* @param enmProt The new ROM protection.
4967	*/
4968	VMMR3DECL(int) PGMR3PhysRomProtect(PVM pVM, RTGCPHYS GCPhys, RTGCPHYS cb, PGMROMPROT enmProt)
4969	{
4970	/*
4971	* Check input
4972	*/
4973	if (!cb)
4974	return VINF_SUCCESS;
4975	AssertReturn(!(GCPhys & GUEST_PAGE_OFFSET_MASK), VERR_INVALID_PARAMETER);
4976	AssertReturn(!(cb & GUEST_PAGE_OFFSET_MASK), VERR_INVALID_PARAMETER);
4977	RTGCPHYS GCPhysLast = GCPhys + (cb - 1);
4978	AssertReturn(GCPhysLast > GCPhys, VERR_INVALID_PARAMETER);
4979	AssertReturn(enmProt >= PGMROMPROT_INVALID && enmProt <= PGMROMPROT_END, VERR_INVALID_PARAMETER);
4980
4981	/*
4982	* Process the request.
4983	*/
4984	PGM_LOCK_VOID(pVM);
4985	int rc = VINF_SUCCESS;
4986	bool fFlushTLB = false;
4987	for (PPGMROMRANGE pRom = pVM->pgm.s.pRomRangesR3; pRom; pRom = pRom->pNextR3)
4988	{
4989	if ( GCPhys <= pRom->GCPhysLast
4990	&& GCPhysLast >= pRom->GCPhys
4991	&& (pRom->fFlags & PGMPHYS_ROM_FLAGS_SHADOWED))
4992	{
4993	/*
4994	* Iterate the relevant pages and make necessary the changes.
4995	*/
4996	#ifdef VBOX_WITH_NATIVE_NEM
4997	PPGMRAMRANGE const pRam = pgmPhysGetRange(pVM, GCPhys);
4998	AssertPtrReturn(pRam, VERR_INTERNAL_ERROR_3);
4999	#endif
5000	bool fChanges = false;
5001	uint32_t const cPages = pRom->GCPhysLast <= GCPhysLast
5002	? pRom->cb >> GUEST_PAGE_SHIFT
5003	: (GCPhysLast - pRom->GCPhys + 1) >> GUEST_PAGE_SHIFT;
5004	for (uint32_t iPage = (GCPhys - pRom->GCPhys) >> GUEST_PAGE_SHIFT;
5005	iPage < cPages;
5006	iPage++)
5007	{
5008	PPGMROMPAGE pRomPage = &pRom->aPages[iPage];
5009	if (PGMROMPROT_IS_ROM(pRomPage->enmProt) != PGMROMPROT_IS_ROM(enmProt))
5010	{
5011	fChanges = true;
5012
5013	/* flush references to the page. */
5014	RTGCPHYS const GCPhysPage = pRom->GCPhys + (iPage << GUEST_PAGE_SHIFT);
5015	PPGMPAGE pRamPage = pgmPhysGetPage(pVM, GCPhysPage);
5016	int rc2 = pgmPoolTrackUpdateGCPhys(pVM, GCPhysPage, pRamPage, true /fFlushPTEs/, &fFlushTLB);
5017	if (rc2 != VINF_SUCCESS && (rc == VINF_SUCCESS \|\| RT_FAILURE(rc2)))
5018	rc = rc2;
5019	#ifdef VBOX_WITH_NATIVE_NEM
5020	uint8_t u2State = PGM_PAGE_GET_NEM_STATE(pRamPage);
5021	#endif
5022
5023	PPGMPAGE pOld = PGMROMPROT_IS_ROM(pRomPage->enmProt) ? &pRomPage->Virgin : &pRomPage->Shadow;
5024	PPGMPAGE pNew = PGMROMPROT_IS_ROM(pRomPage->enmProt) ? &pRomPage->Shadow : &pRomPage->Virgin;
5025
5026	pOld = pRamPage;
5027	pRamPage = pNew;
5028	/** @todo preserve the volatile flags (handlers) when these have been moved out of HCPhys! */
5029
5030	#ifdef VBOX_WITH_NATIVE_NEM
5031	# ifdef VBOX_WITH_PGM_NEM_MODE
5032	/* In simplified mode we have to switch the page data around too. */
5033	if (pVM->pgm.s.fNemMode)
5034	{
5035	uint8_t abPage[GUEST_PAGE_SIZE];
5036	uint8_t * const pbRamPage = PGM_RAMRANGE_CALC_PAGE_R3PTR(pRam, GCPhysPage);
5037	memcpy(abPage, &pRom->pbR3Alternate[(size_t)iPage << GUEST_PAGE_SHIFT], sizeof(abPage));
5038	memcpy(&pRom->pbR3Alternate[(size_t)iPage << GUEST_PAGE_SHIFT], pbRamPage, sizeof(abPage));
5039	memcpy(pbRamPage, abPage, sizeof(abPage));
5040	}
5041	# endif
5042	/* Tell NEM about the backing and protection change. */
5043	if (VM_IS_NEM_ENABLED(pVM))
5044	{
5045	PGMPAGETYPE enmType = (PGMPAGETYPE)PGM_PAGE_GET_TYPE(pNew);
5046	NEMHCNotifyPhysPageChanged(pVM, GCPhys, PGM_PAGE_GET_HCPHYS(pOld), PGM_PAGE_GET_HCPHYS(pNew),
5047	PGM_RAMRANGE_CALC_PAGE_R3PTR(pRam, GCPhysPage),
5048	pgmPhysPageCalcNemProtection(pRamPage, enmType), enmType, &u2State);
5049	PGM_PAGE_SET_NEM_STATE(pRamPage, u2State);
5050	}
5051	#endif
5052	}
5053	pRomPage->enmProt = enmProt;
5054	}
5055
5056	/*
5057	* Reset the access handler if we made changes, no need
5058	* to optimize this.
5059	*/
5060	if (fChanges)
5061	{
5062	int rc2 = PGMHandlerPhysicalReset(pVM, pRom->GCPhys);
5063	if (RT_FAILURE(rc2))
5064	{
5065	PGM_UNLOCK(pVM);
5066	AssertRC(rc);
5067	return rc2;
5068	}
5069	}
5070
5071	/* Advance - cb isn't updated. */
5072	GCPhys = pRom->GCPhys + (cPages << GUEST_PAGE_SHIFT);
5073	}
5074	}
5075	PGM_UNLOCK(pVM);
5076	if (fFlushTLB)
5077	PGM_INVL_ALL_VCPU_TLBS(pVM);
5078
5079	return rc;
5080	}
5081
5082
5083
5084	/*********************************************************************************************************************************
5085	* Ballooning *
5086	*********************************************************************************************************************************/
5087
5088	#if HC_ARCH_BITS == 64 && (defined(RT_OS_WINDOWS) \|\| defined(RT_OS_SOLARIS) \|\| defined(RT_OS_LINUX) \|\| defined(RT_OS_FREEBSD))
5089
5090	/**
5091	* Rendezvous callback used by PGMR3ChangeMemBalloon that changes the memory balloon size
5092	*
5093	* This is only called on one of the EMTs while the other ones are waiting for
5094	* it to complete this function.
5095	*
5096	* @returns VINF_SUCCESS (VBox strict status code).
5097	* @param pVM The cross context VM structure.
5098	* @param pVCpu The cross context virtual CPU structure of the calling EMT. Unused.
5099	* @param pvUser User parameter
5100	*/
5101	static DECLCALLBACK(VBOXSTRICTRC) pgmR3PhysChangeMemBalloonRendezvous(PVM pVM, PVMCPU pVCpu, void *pvUser)
5102	{
5103	uintptr_t paUser = (uintptr_t )pvUser;
5104	bool fInflate = !!paUser[0];
5105	unsigned cPages = paUser[1];
5106	RTGCPHYS paPhysPage = (RTGCPHYS )paUser[2];
5107	uint32_t cPendingPages = 0;
5108	PGMMFREEPAGESREQ pReq;
5109	int rc;
5110
5111	Log(("pgmR3PhysChangeMemBalloonRendezvous: %s %x pages\n", (fInflate) ? "inflate" : "deflate", cPages));
5112	PGM_LOCK_VOID(pVM);
5113
5114	if (fInflate)
5115	{
5116	/* Flush the PGM pool cache as we might have stale references to pages that we just freed. */
5117	pgmR3PoolClearAllRendezvous(pVM, pVCpu, NULL);
5118
5119	/* Replace pages with ZERO pages. */
5120	rc = GMMR3FreePagesPrepare(pVM, &pReq, PGMPHYS_FREE_PAGE_BATCH_SIZE, GMMACCOUNT_BASE);
5121	if (RT_FAILURE(rc))
5122	{
5123	PGM_UNLOCK(pVM);
5124	AssertLogRelRC(rc);
5125	return rc;
5126	}
5127
5128	/* Iterate the pages. */
5129	for (unsigned i = 0; i < cPages; i++)
5130	{
5131	PPGMPAGE pPage = pgmPhysGetPage(pVM, paPhysPage[i]);
5132	if ( pPage == NULL
5133	\|\| PGM_PAGE_GET_TYPE(pPage) != PGMPAGETYPE_RAM)
5134	{
5135	Log(("pgmR3PhysChangeMemBalloonRendezvous: invalid physical page %RGp pPage->u3Type=%d\n", paPhysPage[i], pPage ? PGM_PAGE_GET_TYPE(pPage) : 0));
5136	break;
5137	}
5138
5139	LogFlow(("balloon page: %RGp\n", paPhysPage[i]));
5140
5141	/* Flush the shadow PT if this page was previously used as a guest page table. */
5142	pgmPoolFlushPageByGCPhys(pVM, paPhysPage[i]);
5143
5144	rc = pgmPhysFreePage(pVM, pReq, &cPendingPages, pPage, paPhysPage[i], (PGMPAGETYPE)PGM_PAGE_GET_TYPE(pPage));
5145	if (RT_FAILURE(rc))
5146	{
5147	PGM_UNLOCK(pVM);
5148	AssertLogRelRC(rc);
5149	return rc;
5150	}
5151	Assert(PGM_PAGE_IS_ZERO(pPage));
5152	PGM_PAGE_SET_STATE(pVM, pPage, PGM_PAGE_STATE_BALLOONED);
5153	}
5154
5155	if (cPendingPages)
5156	{
5157	rc = GMMR3FreePagesPerform(pVM, pReq, cPendingPages);
5158	if (RT_FAILURE(rc))
5159	{
5160	PGM_UNLOCK(pVM);
5161	AssertLogRelRC(rc);
5162	return rc;
5163	}
5164	}
5165	GMMR3FreePagesCleanup(pReq);
5166	}
5167	else
5168	{
5169	/* Iterate the pages. */
5170	for (unsigned i = 0; i < cPages; i++)
5171	{
5172	PPGMPAGE pPage = pgmPhysGetPage(pVM, paPhysPage[i]);
5173	AssertBreak(pPage && PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_RAM);
5174
5175	LogFlow(("Free ballooned page: %RGp\n", paPhysPage[i]));
5176
5177	Assert(PGM_PAGE_IS_BALLOONED(pPage));
5178
5179	/* Change back to zero page. (NEM does not need to be informed.) */
5180	PGM_PAGE_SET_STATE(pVM, pPage, PGM_PAGE_STATE_ZERO);
5181	}
5182
5183	/* Note that we currently do not map any ballooned pages in our shadow page tables, so no need to flush the pgm pool. */
5184	}
5185
5186	/* Notify GMM about the balloon change. */
5187	rc = GMMR3BalloonedPages(pVM, (fInflate) ? GMMBALLOONACTION_INFLATE : GMMBALLOONACTION_DEFLATE, cPages);
5188	if (RT_SUCCESS(rc))
5189	{
5190	if (!fInflate)
5191	{
5192	Assert(pVM->pgm.s.cBalloonedPages >= cPages);
5193	pVM->pgm.s.cBalloonedPages -= cPages;
5194	}
5195	else
5196	pVM->pgm.s.cBalloonedPages += cPages;
5197	}
5198
5199	PGM_UNLOCK(pVM);
5200
5201	/* Flush the recompiler's TLB as well. */
5202	for (VMCPUID i = 0; i < pVM->cCpus; i++)
5203	CPUMSetChangedFlags(pVM->apCpusR3[i], CPUM_CHANGED_GLOBAL_TLB_FLUSH);
5204
5205	AssertLogRelRC(rc);
5206	return rc;
5207	}
5208
5209
5210	/**
5211	* Frees a range of ram pages, replacing them with ZERO pages; helper for PGMR3PhysFreeRamPages
5212	*
5213	* @param pVM The cross context VM structure.
5214	* @param fInflate Inflate or deflate memory balloon
5215	* @param cPages Number of pages to free
5216	* @param paPhysPage Array of guest physical addresses
5217	*/
5218	static DECLCALLBACK(void) pgmR3PhysChangeMemBalloonHelper(PVM pVM, bool fInflate, unsigned cPages, RTGCPHYS *paPhysPage)
5219	{
5220	uintptr_t paUser[3];
5221
5222	paUser[0] = fInflate;
5223	paUser[1] = cPages;
5224	paUser[2] = (uintptr_t)paPhysPage;
5225	int rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_ONCE, pgmR3PhysChangeMemBalloonRendezvous, (void *)paUser);
5226	AssertRC(rc);
5227
5228	/* Made a copy in PGMR3PhysFreeRamPages; free it here. */
5229	RTMemFree(paPhysPage);
5230	}
5231
5232	#endif /* 64-bit host && (Windows \|\| Solaris \|\| Linux \|\| FreeBSD) */
5233
5234	/**
5235	* Inflate or deflate a memory balloon
5236	*
5237	* @returns VBox status code.
5238	* @param pVM The cross context VM structure.
5239	* @param fInflate Inflate or deflate memory balloon
5240	* @param cPages Number of pages to free
5241	* @param paPhysPage Array of guest physical addresses
5242	*/
5243	VMMR3DECL(int) PGMR3PhysChangeMemBalloon(PVM pVM, bool fInflate, unsigned cPages, RTGCPHYS *paPhysPage)
5244	{
5245	/* This must match GMMR0Init; currently we only support memory ballooning on all 64-bit hosts except Mac OS X */
5246	#if HC_ARCH_BITS == 64 && (defined(RT_OS_WINDOWS) \|\| defined(RT_OS_SOLARIS) \|\| defined(RT_OS_LINUX) \|\| defined(RT_OS_FREEBSD))
5247	int rc;
5248
5249	/* Older additions (ancient non-functioning balloon code) pass wrong physical addresses. */
5250	AssertReturn(!(paPhysPage[0] & 0xfff), VERR_INVALID_PARAMETER);
5251
5252	/* We own the IOM lock here and could cause a deadlock by waiting for another VCPU that is blocking on the IOM lock.
5253	* In the SMP case we post a request packet to postpone the job.
5254	*/
5255	if (pVM->cCpus > 1)
5256	{
5257	unsigned cbPhysPage = cPages * sizeof(paPhysPage[0]);
5258	RTGCPHYS paPhysPageCopy = (RTGCPHYS )RTMemAlloc(cbPhysPage);
5259	AssertReturn(paPhysPageCopy, VERR_NO_MEMORY);
5260
5261	memcpy(paPhysPageCopy, paPhysPage, cbPhysPage);
5262
5263	rc = VMR3ReqCallNoWait(pVM, VMCPUID_ANY_QUEUE, (PFNRT)pgmR3PhysChangeMemBalloonHelper, 4, pVM, fInflate, cPages, paPhysPageCopy);
5264	AssertRC(rc);
5265	}
5266	else
5267	{
5268	uintptr_t paUser[3];
5269
5270	paUser[0] = fInflate;
5271	paUser[1] = cPages;
5272	paUser[2] = (uintptr_t)paPhysPage;
5273	rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_ONCE, pgmR3PhysChangeMemBalloonRendezvous, (void *)paUser);
5274	AssertRC(rc);
5275	}
5276	return rc;
5277
5278	#else
5279	NOREF(pVM); NOREF(fInflate); NOREF(cPages); NOREF(paPhysPage);
5280	return VERR_NOT_IMPLEMENTED;
5281	#endif
5282	}
5283
5284
5285	/*********************************************************************************************************************************
5286	* Write Monitoring *
5287	*********************************************************************************************************************************/
5288
5289	/**
5290	* Rendezvous callback used by PGMR3WriteProtectRAM that write protects all
5291	* physical RAM.
5292	*
5293	* This is only called on one of the EMTs while the other ones are waiting for
5294	* it to complete this function.
5295	*
5296	* @returns VINF_SUCCESS (VBox strict status code).
5297	* @param pVM The cross context VM structure.
5298	* @param pVCpu The cross context virtual CPU structure of the calling EMT. Unused.
5299	* @param pvUser User parameter, unused.
5300	*/
5301	static DECLCALLBACK(VBOXSTRICTRC) pgmR3PhysWriteProtectRAMRendezvous(PVM pVM, PVMCPU pVCpu, void *pvUser)
5302	{
5303	int rc = VINF_SUCCESS;
5304	NOREF(pvUser); NOREF(pVCpu);
5305
5306	PGM_LOCK_VOID(pVM);
5307	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
5308	pgmPoolResetDirtyPages(pVM);
5309	#endif
5310
5311	/** @todo pointless to write protect the physical page pointed to by RSP. */
5312
5313	for (PPGMRAMRANGE pRam = pVM->pgm.s.CTX_SUFF(pRamRangesX);
5314	pRam;
5315	pRam = pRam->CTX_SUFF(pNext))
5316	{
5317	uint32_t cPages = pRam->cb >> GUEST_PAGE_SHIFT;
5318	for (uint32_t iPage = 0; iPage < cPages; iPage++)
5319	{
5320	PPGMPAGE pPage = &pRam->aPages[iPage];
5321	PGMPAGETYPE enmPageType = (PGMPAGETYPE)PGM_PAGE_GET_TYPE(pPage);
5322
5323	if ( RT_LIKELY(enmPageType == PGMPAGETYPE_RAM)
5324	\|\| enmPageType == PGMPAGETYPE_MMIO2)
5325	{
5326	/*
5327	* A RAM page.
5328	*/
5329	switch (PGM_PAGE_GET_STATE(pPage))
5330	{
5331	case PGM_PAGE_STATE_ALLOCATED:
5332	/** @todo Optimize this: Don't always re-enable write
5333	* monitoring if the page is known to be very busy. */
5334	if (PGM_PAGE_IS_WRITTEN_TO(pPage))
5335	PGM_PAGE_CLEAR_WRITTEN_TO(pVM, pPage);
5336
5337	pgmPhysPageWriteMonitor(pVM, pPage, pRam->GCPhys + ((RTGCPHYS)iPage << GUEST_PAGE_SHIFT));
5338	break;
5339
5340	case PGM_PAGE_STATE_SHARED:
5341	AssertFailed();
5342	break;
5343
5344	case PGM_PAGE_STATE_WRITE_MONITORED: /* nothing to change. */
5345	default:
5346	break;
5347	}
5348	}
5349	}
5350	}
5351	pgmR3PoolWriteProtectPages(pVM);
5352	PGM_INVL_ALL_VCPU_TLBS(pVM);
5353	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
5354	CPUMSetChangedFlags(pVM->apCpusR3[idCpu], CPUM_CHANGED_GLOBAL_TLB_FLUSH);
5355
5356	PGM_UNLOCK(pVM);
5357	return rc;
5358	}
5359
5360	/**
5361	* Protect all physical RAM to monitor writes
5362	*
5363	* @returns VBox status code.
5364	* @param pVM The cross context VM structure.
5365	*/
5366	VMMR3DECL(int) PGMR3PhysWriteProtectRAM(PVM pVM)
5367	{
5368	VM_ASSERT_EMT_RETURN(pVM, VERR_VM_THREAD_NOT_EMT);
5369
5370	int rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_ONCE, pgmR3PhysWriteProtectRAMRendezvous, NULL);
5371	AssertRC(rc);
5372	return rc;
5373	}
5374
5375
5376	/*********************************************************************************************************************************
5377	* Stats. *
5378	*********************************************************************************************************************************/
5379
5380	/**
5381	* Query the amount of free memory inside VMMR0
5382	*
5383	* @returns VBox status code.
5384	* @param pUVM The user mode VM handle.
5385	* @param pcbAllocMem Where to return the amount of memory allocated
5386	* by VMs.
5387	* @param pcbFreeMem Where to return the amount of memory that is
5388	* allocated from the host but not currently used
5389	* by any VMs.
5390	* @param pcbBallonedMem Where to return the sum of memory that is
5391	* currently ballooned by the VMs.
5392	* @param pcbSharedMem Where to return the amount of memory that is
5393	* currently shared.
5394	*/
5395	VMMR3DECL(int) PGMR3QueryGlobalMemoryStats(PUVM pUVM, uint64_t pcbAllocMem, uint64_t pcbFreeMem,
5396	uint64_t pcbBallonedMem, uint64_t pcbSharedMem)
5397	{
5398	UVM_ASSERT_VALID_EXT_RETURN(pUVM, VERR_INVALID_VM_HANDLE);
5399	VM_ASSERT_VALID_EXT_RETURN(pUVM->pVM, VERR_INVALID_VM_HANDLE);
5400
5401	uint64_t cAllocPages = 0;
5402	uint64_t cFreePages = 0;
5403	uint64_t cBalloonPages = 0;
5404	uint64_t cSharedPages = 0;
5405	if (!SUPR3IsDriverless())
5406	{
5407	int rc = GMMR3QueryHypervisorMemoryStats(pUVM->pVM, &cAllocPages, &cFreePages, &cBalloonPages, &cSharedPages);
5408	AssertRCReturn(rc, rc);
5409	}
5410
5411	if (pcbAllocMem)
5412	pcbAllocMem = cAllocPages _4K;
5413
5414	if (pcbFreeMem)
5415	pcbFreeMem = cFreePages _4K;
5416
5417	if (pcbBallonedMem)
5418	pcbBallonedMem = cBalloonPages _4K;
5419
5420	if (pcbSharedMem)
5421	pcbSharedMem = cSharedPages _4K;
5422
5423	Log(("PGMR3QueryVMMMemoryStats: all=%llx free=%llx ballooned=%llx shared=%llx\n",
5424	cAllocPages, cFreePages, cBalloonPages, cSharedPages));
5425	return VINF_SUCCESS;
5426	}
5427
5428
5429	/**
5430	* Query memory stats for the VM.
5431	*
5432	* @returns VBox status code.
5433	* @param pUVM The user mode VM handle.
5434	* @param pcbTotalMem Where to return total amount memory the VM may
5435	* possibly use.
5436	* @param pcbPrivateMem Where to return the amount of private memory
5437	* currently allocated.
5438	* @param pcbSharedMem Where to return the amount of actually shared
5439	* memory currently used by the VM.
5440	* @param pcbZeroMem Where to return the amount of memory backed by
5441	* zero pages.
5442	*
5443	* @remarks The total mem is normally larger than the sum of the three
5444	* components. There are two reasons for this, first the amount of
5445	* shared memory is what we're sure is shared instead of what could
5446	* possibly be shared with someone. Secondly, because the total may
5447	* include some pure MMIO pages that doesn't go into any of the three
5448	* sub-counts.
5449	*
5450	* @todo Why do we return reused shared pages instead of anything that could
5451	* potentially be shared? Doesn't this mean the first VM gets a much
5452	* lower number of shared pages?
5453	*/
5454	VMMR3DECL(int) PGMR3QueryMemoryStats(PUVM pUVM, uint64_t pcbTotalMem, uint64_t pcbPrivateMem,
5455	uint64_t pcbSharedMem, uint64_t pcbZeroMem)
5456	{
5457	UVM_ASSERT_VALID_EXT_RETURN(pUVM, VERR_INVALID_VM_HANDLE);
5458	PVM pVM = pUVM->pVM;
5459	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
5460
5461	if (pcbTotalMem)
5462	pcbTotalMem = (uint64_t)pVM->pgm.s.cAllPages GUEST_PAGE_SIZE;
5463
5464	if (pcbPrivateMem)
5465	pcbPrivateMem = (uint64_t)pVM->pgm.s.cPrivatePages GUEST_PAGE_SIZE;
5466
5467	if (pcbSharedMem)
5468	pcbSharedMem = (uint64_t)pVM->pgm.s.cReusedSharedPages GUEST_PAGE_SIZE;
5469
5470	if (pcbZeroMem)
5471	pcbZeroMem = (uint64_t)pVM->pgm.s.cZeroPages GUEST_PAGE_SIZE;
5472
5473	Log(("PGMR3QueryMemoryStats: all=%x private=%x reused=%x zero=%x\n", pVM->pgm.s.cAllPages, pVM->pgm.s.cPrivatePages, pVM->pgm.s.cReusedSharedPages, pVM->pgm.s.cZeroPages));
5474	return VINF_SUCCESS;
5475	}
5476
5477
5478
5479	/*********************************************************************************************************************************
5480	* Chunk Mappings and Page Allocation *
5481	*********************************************************************************************************************************/
5482
5483	/**
5484	* Tree enumeration callback for dealing with age rollover.
5485	* It will perform a simple compression of the current age.
5486	*/
5487	static DECLCALLBACK(int) pgmR3PhysChunkAgeingRolloverCallback(PAVLU32NODECORE pNode, void *pvUser)
5488	{
5489	/* Age compression - ASSUMES iNow == 4. */
5490	PPGMCHUNKR3MAP pChunk = (PPGMCHUNKR3MAP)pNode;
5491	if (pChunk->iLastUsed >= UINT32_C(0xffffff00))
5492	pChunk->iLastUsed = 3;
5493	else if (pChunk->iLastUsed >= UINT32_C(0xfffff000))
5494	pChunk->iLastUsed = 2;
5495	else if (pChunk->iLastUsed)
5496	pChunk->iLastUsed = 1;
5497	else /* iLastUsed = 0 */
5498	pChunk->iLastUsed = 4;
5499
5500	NOREF(pvUser);
5501	return 0;
5502	}
5503
5504
5505	/**
5506	* The structure passed in the pvUser argument of pgmR3PhysChunkUnmapCandidateCallback().
5507	*/
5508	typedef struct PGMR3PHYSCHUNKUNMAPCB
5509	{
5510	PVM pVM; /*< Pointer to the VM. /
5511	PPGMCHUNKR3MAP pChunk; /*< The chunk to unmap. /
5512	} PGMR3PHYSCHUNKUNMAPCB, *PPGMR3PHYSCHUNKUNMAPCB;
5513
5514
5515	/**
5516	* Callback used to find the mapping that's been unused for
5517	* the longest time.
5518	*/
5519	static DECLCALLBACK(int) pgmR3PhysChunkUnmapCandidateCallback(PAVLU32NODECORE pNode, void *pvUser)
5520	{
5521	PPGMCHUNKR3MAP pChunk = (PPGMCHUNKR3MAP)pNode;
5522	PPGMR3PHYSCHUNKUNMAPCB pArg = (PPGMR3PHYSCHUNKUNMAPCB)pvUser;
5523
5524	/*
5525	* Check for locks and compare when last used.
5526	*/
5527	if (pChunk->cRefs)
5528	return 0;
5529	if (pChunk->cPermRefs)
5530	return 0;
5531	if ( pArg->pChunk
5532	&& pChunk->iLastUsed >= pArg->pChunk->iLastUsed)
5533	return 0;
5534
5535	/*
5536	* Check that it's not in any of the TLBs.
5537	*/
5538	PVM pVM = pArg->pVM;
5539	if ( pVM->pgm.s.ChunkR3Map.Tlb.aEntries[PGM_CHUNKR3MAPTLB_IDX(pChunk->Core.Key)].idChunk
5540	== pChunk->Core.Key)
5541	{
5542	pChunk = NULL;
5543	return 0;
5544	}
5545	#ifdef VBOX_STRICT
5546	for (unsigned i = 0; i < RT_ELEMENTS(pVM->pgm.s.ChunkR3Map.Tlb.aEntries); i++)
5547	{
5548	Assert(pVM->pgm.s.ChunkR3Map.Tlb.aEntries[i].pChunk != pChunk);
5549	Assert(pVM->pgm.s.ChunkR3Map.Tlb.aEntries[i].idChunk != pChunk->Core.Key);
5550	}
5551	#endif
5552
5553	for (unsigned i = 0; i < RT_ELEMENTS(pVM->pgm.s.PhysTlbR3.aEntries); i++)
5554	if (pVM->pgm.s.PhysTlbR3.aEntries[i].pMap == pChunk)
5555	return 0;
5556
5557	pArg->pChunk = pChunk;
5558	return 0;
5559	}
5560
5561
5562	/**
5563	* Finds a good candidate for unmapping when the ring-3 mapping cache is full.
5564	*
5565	* The candidate will not be part of any TLBs, so no need to flush
5566	* anything afterwards.
5567	*
5568	* @returns Chunk id.
5569	* @param pVM The cross context VM structure.
5570	*/
5571	static int32_t pgmR3PhysChunkFindUnmapCandidate(PVM pVM)
5572	{
5573	PGM_LOCK_ASSERT_OWNER(pVM);
5574
5575	/*
5576	* Enumerate the age tree starting with the left most node.
5577	*/
5578	STAM_PROFILE_START(&pVM->pgm.s.Stats.StatChunkFindCandidate, a);
5579	PGMR3PHYSCHUNKUNMAPCB Args;
5580	Args.pVM = pVM;
5581	Args.pChunk = NULL;
5582	RTAvlU32DoWithAll(&pVM->pgm.s.ChunkR3Map.pTree, true /fFromLeft/, pgmR3PhysChunkUnmapCandidateCallback, &Args);
5583	Assert(Args.pChunk);
5584	if (Args.pChunk)
5585	{
5586	Assert(Args.pChunk->cRefs == 0);
5587	Assert(Args.pChunk->cPermRefs == 0);
5588	STAM_PROFILE_STOP(&pVM->pgm.s.Stats.StatChunkFindCandidate, a);
5589	return Args.pChunk->Core.Key;
5590	}
5591
5592	STAM_PROFILE_STOP(&pVM->pgm.s.Stats.StatChunkFindCandidate, a);
5593	return INT32_MAX;
5594	}
5595
5596
5597	/**
5598	* Rendezvous callback used by pgmR3PhysUnmapChunk that unmaps a chunk
5599	*
5600	* This is only called on one of the EMTs while the other ones are waiting for
5601	* it to complete this function.
5602	*
5603	* @returns VINF_SUCCESS (VBox strict status code).
5604	* @param pVM The cross context VM structure.
5605	* @param pVCpu The cross context virtual CPU structure of the calling EMT. Unused.
5606	* @param pvUser User pointer. Unused
5607	*
5608	*/
5609	static DECLCALLBACK(VBOXSTRICTRC) pgmR3PhysUnmapChunkRendezvous(PVM pVM, PVMCPU pVCpu, void *pvUser)
5610	{
5611	int rc = VINF_SUCCESS;
5612	PGM_LOCK_VOID(pVM);
5613	NOREF(pVCpu); NOREF(pvUser);
5614
5615	if (pVM->pgm.s.ChunkR3Map.c >= pVM->pgm.s.ChunkR3Map.cMax)
5616	{
5617	/* Flush the pgm pool cache; call the internal rendezvous handler as we're already in a rendezvous handler here. */
5618	/** @todo also not really efficient to unmap a chunk that contains PD
5619	* or PT pages. */
5620	pgmR3PoolClearAllRendezvous(pVM, pVM->apCpusR3[0], NULL /* no need to flush the REM TLB as we already did that above */);
5621
5622	/*
5623	* Request the ring-0 part to unmap a chunk to make space in the mapping cache.
5624	*/
5625	GMMMAPUNMAPCHUNKREQ Req;
5626	Req.Hdr.u32Magic = SUPVMMR0REQHDR_MAGIC;
5627	Req.Hdr.cbReq = sizeof(Req);
5628	Req.pvR3 = NULL;
5629	Req.idChunkMap = NIL_GMM_CHUNKID;
5630	Req.idChunkUnmap = pgmR3PhysChunkFindUnmapCandidate(pVM);
5631	if (Req.idChunkUnmap != INT32_MAX)
5632	{
5633	STAM_PROFILE_START(&pVM->pgm.s.Stats.StatChunkUnmap, a);
5634	rc = VMMR3CallR0(pVM, VMMR0_DO_GMM_MAP_UNMAP_CHUNK, 0, &Req.Hdr);
5635	STAM_PROFILE_STOP(&pVM->pgm.s.Stats.StatChunkUnmap, a);
5636	if (RT_SUCCESS(rc))
5637	{
5638	/*
5639	* Remove the unmapped one.
5640	*/
5641	PPGMCHUNKR3MAP pUnmappedChunk = (PPGMCHUNKR3MAP)RTAvlU32Remove(&pVM->pgm.s.ChunkR3Map.pTree, Req.idChunkUnmap);
5642	AssertRelease(pUnmappedChunk);
5643	AssertRelease(!pUnmappedChunk->cRefs);
5644	AssertRelease(!pUnmappedChunk->cPermRefs);
5645	pUnmappedChunk->pv = NULL;
5646	pUnmappedChunk->Core.Key = UINT32_MAX;
5647	MMR3HeapFree(pUnmappedChunk);
5648	pVM->pgm.s.ChunkR3Map.c--;
5649	pVM->pgm.s.cUnmappedChunks++;
5650
5651	/*
5652	* Flush dangling PGM pointers (R3 & R0 ptrs to GC physical addresses).
5653	*/
5654	/** @todo We should not flush chunks which include cr3 mappings. */
5655	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
5656	{
5657	PPGMCPU pPGM = &pVM->apCpusR3[idCpu]->pgm.s;
5658
5659	pPGM->pGst32BitPdR3 = NULL;
5660	pPGM->pGstPaePdptR3 = NULL;
5661	pPGM->pGstAmd64Pml4R3 = NULL;
5662	pPGM->pGstEptPml4R3 = NULL;
5663	pPGM->pGst32BitPdR0 = NIL_RTR0PTR;
5664	pPGM->pGstPaePdptR0 = NIL_RTR0PTR;
5665	pPGM->pGstAmd64Pml4R0 = NIL_RTR0PTR;
5666	pPGM->pGstEptPml4R0 = NIL_RTR0PTR;
5667	for (unsigned i = 0; i < RT_ELEMENTS(pPGM->apGstPaePDsR3); i++)
5668	{
5669	pPGM->apGstPaePDsR3[i] = NULL;
5670	pPGM->apGstPaePDsR0[i] = NIL_RTR0PTR;
5671	}
5672
5673	/* Flush REM TLBs. */
5674	CPUMSetChangedFlags(pVM->apCpusR3[idCpu], CPUM_CHANGED_GLOBAL_TLB_FLUSH);
5675	}
5676	}
5677	}
5678	}
5679	PGM_UNLOCK(pVM);
5680	return rc;
5681	}
5682
5683	/**
5684	* Unmap a chunk to free up virtual address space (request packet handler for pgmR3PhysChunkMap)
5685	*
5686	* @param pVM The cross context VM structure.
5687	*/
5688	static DECLCALLBACK(void) pgmR3PhysUnmapChunk(PVM pVM)
5689	{
5690	int rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_ONCE, pgmR3PhysUnmapChunkRendezvous, NULL);
5691	AssertRC(rc);
5692	}
5693
5694
5695	/**
5696	* Maps the given chunk into the ring-3 mapping cache.
5697	*
5698	* This will call ring-0.
5699	*
5700	* @returns VBox status code.
5701	* @param pVM The cross context VM structure.
5702	* @param idChunk The chunk in question.
5703	* @param ppChunk Where to store the chunk tracking structure.
5704	*
5705	* @remarks Called from within the PGM critical section.
5706	* @remarks Can be called from any thread!
5707	*/
5708	int pgmR3PhysChunkMap(PVM pVM, uint32_t idChunk, PPPGMCHUNKR3MAP ppChunk)
5709	{
5710	int rc;
5711
5712	PGM_LOCK_ASSERT_OWNER(pVM);
5713
5714	/*
5715	* Move the chunk time forward.
5716	*/
5717	pVM->pgm.s.ChunkR3Map.iNow++;
5718	if (pVM->pgm.s.ChunkR3Map.iNow == 0)
5719	{
5720	pVM->pgm.s.ChunkR3Map.iNow = 4;
5721	RTAvlU32DoWithAll(&pVM->pgm.s.ChunkR3Map.pTree, true /fFromLeft/, pgmR3PhysChunkAgeingRolloverCallback, NULL);
5722	}
5723
5724	/*
5725	* Allocate a new tracking structure first.
5726	*/
5727	PPGMCHUNKR3MAP pChunk = (PPGMCHUNKR3MAP)MMR3HeapAllocZ(pVM, MM_TAG_PGM_CHUNK_MAPPING, sizeof(*pChunk));
5728	AssertReturn(pChunk, VERR_NO_MEMORY);
5729	pChunk->Core.Key = idChunk;
5730	pChunk->iLastUsed = pVM->pgm.s.ChunkR3Map.iNow;
5731
5732	/*
5733	* Request the ring-0 part to map the chunk in question.
5734	*/
5735	GMMMAPUNMAPCHUNKREQ Req;
5736	Req.Hdr.u32Magic = SUPVMMR0REQHDR_MAGIC;
5737	Req.Hdr.cbReq = sizeof(Req);
5738	Req.pvR3 = NULL;
5739	Req.idChunkMap = idChunk;
5740	Req.idChunkUnmap = NIL_GMM_CHUNKID;
5741
5742	/* Must be callable from any thread, so can't use VMMR3CallR0. */
5743	STAM_PROFILE_START(&pVM->pgm.s.Stats.StatChunkMap, a);
5744	rc = SUPR3CallVMMR0Ex(VMCC_GET_VMR0_FOR_CALL(pVM), NIL_VMCPUID, VMMR0_DO_GMM_MAP_UNMAP_CHUNK, 0, &Req.Hdr);
5745	STAM_PROFILE_STOP(&pVM->pgm.s.Stats.StatChunkMap, a);
5746	if (RT_SUCCESS(rc))
5747	{
5748	pChunk->pv = Req.pvR3;
5749
5750	/*
5751	* If we're running out of virtual address space, then we should
5752	* unmap another chunk.
5753	*
5754	* Currently, an unmap operation requires that all other virtual CPUs
5755	* are idling and not by chance making use of the memory we're
5756	* unmapping. So, we create an async unmap operation here.
5757	*
5758	* Now, when creating or restoring a saved state this wont work very
5759	* well since we may want to restore all guest RAM + a little something.
5760	* So, we have to do the unmap synchronously. Fortunately for us
5761	* though, during these operations the other virtual CPUs are inactive
5762	* and it should be safe to do this.
5763	*/
5764	/** @todo Eventually we should lock all memory when used and do
5765	* map+unmap as one kernel call without any rendezvous or
5766	* other precautions. */
5767	if (pVM->pgm.s.ChunkR3Map.c + 1 >= pVM->pgm.s.ChunkR3Map.cMax)
5768	{
5769	switch (VMR3GetState(pVM))
5770	{
5771	case VMSTATE_LOADING:
5772	case VMSTATE_SAVING:
5773	{
5774	PVMCPU pVCpu = VMMGetCpu(pVM);
5775	if ( pVCpu
5776	&& pVM->pgm.s.cDeprecatedPageLocks == 0)
5777	{
5778	pgmR3PhysUnmapChunkRendezvous(pVM, pVCpu, NULL);
5779	break;
5780	}
5781	}
5782	RT_FALL_THRU();
5783	default:
5784	rc = VMR3ReqCallNoWait(pVM, VMCPUID_ANY_QUEUE, (PFNRT)pgmR3PhysUnmapChunk, 1, pVM);
5785	AssertRC(rc);
5786	break;
5787	}
5788	}
5789
5790	/*
5791	* Update the tree. We must do this after any unmapping to make sure
5792	* the chunk we're going to return isn't unmapped by accident.
5793	*/
5794	AssertPtr(Req.pvR3);
5795	bool fRc = RTAvlU32Insert(&pVM->pgm.s.ChunkR3Map.pTree, &pChunk->Core);
5796	AssertRelease(fRc);
5797	pVM->pgm.s.ChunkR3Map.c++;
5798	pVM->pgm.s.cMappedChunks++;
5799	}
5800	else
5801	{
5802	/** @todo this may fail because of /proc/sys/vm/max_map_count, so we
5803	* should probably restrict ourselves on linux. */
5804	AssertRC(rc);
5805	MMR3HeapFree(pChunk);
5806	pChunk = NULL;
5807	}
5808
5809	*ppChunk = pChunk;
5810	return rc;
5811	}
5812
5813
5814	/**
5815	* Invalidates the TLB for the ring-3 mapping cache.
5816	*
5817	* @param pVM The cross context VM structure.
5818	*/
5819	VMMR3DECL(void) PGMR3PhysChunkInvalidateTLB(PVM pVM)
5820	{
5821	PGM_LOCK_VOID(pVM);
5822	for (unsigned i = 0; i < RT_ELEMENTS(pVM->pgm.s.ChunkR3Map.Tlb.aEntries); i++)
5823	{
5824	pVM->pgm.s.ChunkR3Map.Tlb.aEntries[i].idChunk = NIL_GMM_CHUNKID;
5825	pVM->pgm.s.ChunkR3Map.Tlb.aEntries[i].pChunk = NULL;
5826	}
5827	/* The page map TLB references chunks, so invalidate that one too. */
5828	pgmPhysInvalidatePageMapTLB(pVM);
5829	PGM_UNLOCK(pVM);
5830	}
5831
5832
5833	/**
5834	* Response to VM_FF_PGM_NEED_HANDY_PAGES and helper for pgmPhysEnsureHandyPage.
5835	*
5836	* This function will also work the VM_FF_PGM_NO_MEMORY force action flag, to
5837	* signal and clear the out of memory condition. When called, this API is used
5838	* to try clear the condition when the user wants to resume.
5839	*
5840	* @returns The following VBox status codes.
5841	* @retval VINF_SUCCESS on success. FFs cleared.
5842	* @retval VINF_EM_NO_MEMORY if we're out of memory. The FF is not cleared in
5843	* this case and it gets accompanied by VM_FF_PGM_NO_MEMORY.
5844	*
5845	* @param pVM The cross context VM structure.
5846	*
5847	* @remarks The VINF_EM_NO_MEMORY status is for the benefit of the FF processing
5848	* in EM.cpp and shouldn't be propagated outside TRPM, HM, EM and
5849	* pgmPhysEnsureHandyPage. There is one exception to this in the \#PF
5850	* handler.
5851	*/
5852	VMMR3DECL(int) PGMR3PhysAllocateHandyPages(PVM pVM)
5853	{
5854	PGM_LOCK_VOID(pVM);
5855
5856	/*
5857	* Allocate more pages, noting down the index of the first new page.
5858	*/
5859	uint32_t iClear = pVM->pgm.s.cHandyPages;
5860	AssertMsgReturn(iClear <= RT_ELEMENTS(pVM->pgm.s.aHandyPages), ("%d", iClear), VERR_PGM_HANDY_PAGE_IPE);
5861	Log(("PGMR3PhysAllocateHandyPages: %d -> %d\n", iClear, RT_ELEMENTS(pVM->pgm.s.aHandyPages)));
5862	int rc = VMMR3CallR0(pVM, VMMR0_DO_PGM_ALLOCATE_HANDY_PAGES, 0, NULL);
5863	/** @todo we should split this up into an allocate and flush operation. sometimes you want to flush and not allocate more (which will trigger the vm account limit error) */
5864	if ( rc == VERR_GMM_HIT_VM_ACCOUNT_LIMIT
5865	&& pVM->pgm.s.cHandyPages > 0)
5866	{
5867	/* Still handy pages left, so don't panic. */
5868	rc = VINF_SUCCESS;
5869	}
5870
5871	if (RT_SUCCESS(rc))
5872	{
5873	AssertMsg(rc == VINF_SUCCESS, ("%Rrc\n", rc));
5874	Assert(pVM->pgm.s.cHandyPages > 0);
5875	#ifdef VBOX_STRICT
5876	uint32_t i;
5877	for (i = iClear; i < pVM->pgm.s.cHandyPages; i++)
5878	if ( pVM->pgm.s.aHandyPages[i].idPage == NIL_GMM_PAGEID
5879	\|\| pVM->pgm.s.aHandyPages[i].idSharedPage != NIL_GMM_PAGEID
5880	\|\| (pVM->pgm.s.aHandyPages[i].HCPhysGCPhys & GUEST_PAGE_OFFSET_MASK))
5881	break;
5882	if (i != pVM->pgm.s.cHandyPages)
5883	{
5884	RTAssertMsg1Weak(NULL, __LINE__, __FILE__, __FUNCTION__);
5885	RTAssertMsg2Weak("i=%d iClear=%d cHandyPages=%d\n", i, iClear, pVM->pgm.s.cHandyPages);
5886	for (uint32_t j = iClear; j < pVM->pgm.s.cHandyPages; j++)
5887	RTAssertMsg2Add("%03d: idPage=%d HCPhysGCPhys=%RHp idSharedPage=%d%s\n", j,
5888	pVM->pgm.s.aHandyPages[j].idPage,
5889	pVM->pgm.s.aHandyPages[j].HCPhysGCPhys,
5890	pVM->pgm.s.aHandyPages[j].idSharedPage,
5891	j == i ? " <---" : "");
5892	RTAssertPanic();
5893	}
5894	#endif
5895	}
5896	else
5897	{
5898	/*
5899	* We should never get here unless there is a genuine shortage of
5900	* memory (or some internal error). Flag the error so the VM can be
5901	* suspended ASAP and the user informed. If we're totally out of
5902	* handy pages we will return failure.
5903	*/
5904	/* Report the failure. */
5905	LogRel(("PGM: Failed to procure handy pages; rc=%Rrc cHandyPages=%#x\n"
5906	" cAllPages=%#x cPrivatePages=%#x cSharedPages=%#x cZeroPages=%#x\n",
5907	rc, pVM->pgm.s.cHandyPages,
5908	pVM->pgm.s.cAllPages, pVM->pgm.s.cPrivatePages, pVM->pgm.s.cSharedPages, pVM->pgm.s.cZeroPages));
5909
5910	if ( rc != VERR_NO_MEMORY
5911	&& rc != VERR_NO_PHYS_MEMORY
5912	&& rc != VERR_LOCK_FAILED)
5913	for (uint32_t i = 0; i < RT_ELEMENTS(pVM->pgm.s.aHandyPages); i++)
5914	{
5915	LogRel(("PGM: aHandyPages[#%#04x] = {.HCPhysGCPhys=%RHp, .idPage=%#08x, .idSharedPage=%#08x}\n",
5916	i, pVM->pgm.s.aHandyPages[i].HCPhysGCPhys, pVM->pgm.s.aHandyPages[i].idPage,
5917	pVM->pgm.s.aHandyPages[i].idSharedPage));
5918	uint32_t const idPage = pVM->pgm.s.aHandyPages[i].idPage;
5919	if (idPage != NIL_GMM_PAGEID)
5920	{
5921	for (PPGMRAMRANGE pRam = pVM->pgm.s.pRamRangesXR3;
5922	pRam;
5923	pRam = pRam->pNextR3)
5924	{
5925	uint32_t const cPages = pRam->cb >> GUEST_PAGE_SHIFT;
5926	for (uint32_t iPage = 0; iPage < cPages; iPage++)
5927	if (PGM_PAGE_GET_PAGEID(&pRam->aPages[iPage]) == idPage)
5928	LogRel(("PGM: Used by %RGp %R[pgmpage] (%s)\n",
5929	pRam->GCPhys + ((RTGCPHYS)iPage << GUEST_PAGE_SHIFT), &pRam->aPages[iPage], pRam->pszDesc));
5930	}
5931	}
5932	}
5933
5934	if (rc == VERR_NO_MEMORY)
5935	{
5936	uint64_t cbHostRamAvail = 0;
5937	int rc2 = RTSystemQueryAvailableRam(&cbHostRamAvail);
5938	if (RT_SUCCESS(rc2))
5939	LogRel(("Host RAM: %RU64MB available\n", cbHostRamAvail / _1M));
5940	else
5941	LogRel(("Cannot determine the amount of available host memory\n"));
5942	}
5943
5944	/* Set the FFs and adjust rc. */
5945	VM_FF_SET(pVM, VM_FF_PGM_NEED_HANDY_PAGES);
5946	VM_FF_SET(pVM, VM_FF_PGM_NO_MEMORY);
5947	if ( rc == VERR_NO_MEMORY
5948	\|\| rc == VERR_NO_PHYS_MEMORY
5949	\|\| rc == VERR_LOCK_FAILED)
5950	rc = VINF_EM_NO_MEMORY;
5951	}
5952
5953	PGM_UNLOCK(pVM);
5954	return rc;
5955	}
5956
5957
5958	/*********************************************************************************************************************************
5959	* Other Stuff *
5960	*********************************************************************************************************************************/
5961
5962	#if !defined(VBOX_VMM_TARGET_ARMV8)
5963	/**
5964	* Sets the Address Gate 20 state.
5965	*
5966	* @param pVCpu The cross context virtual CPU structure.
5967	* @param fEnable True if the gate should be enabled.
5968	* False if the gate should be disabled.
5969	*/
5970	VMMDECL(void) PGMR3PhysSetA20(PVMCPU pVCpu, bool fEnable)
5971	{
5972	LogFlow(("PGMR3PhysSetA20 %d (was %d)\n", fEnable, pVCpu->pgm.s.fA20Enabled));
5973	if (pVCpu->pgm.s.fA20Enabled != fEnable)
5974	{
5975	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
5976	PCCPUMCTX pCtx = CPUMQueryGuestCtxPtr(pVCpu);
5977	if ( CPUMIsGuestInVmxRootMode(pCtx)
5978	&& !fEnable)
5979	{
5980	Log(("Cannot enter A20M mode while in VMX root mode\n"));
5981	return;
5982	}
5983	#endif
5984	pVCpu->pgm.s.fA20Enabled = fEnable;
5985	pVCpu->pgm.s.GCPhysA20Mask = ~((RTGCPHYS)!fEnable << 20);
5986	if (VM_IS_NEM_ENABLED(pVCpu->CTX_SUFF(pVM)))
5987	NEMR3NotifySetA20(pVCpu, fEnable);
5988	#ifdef PGM_WITH_A20
5989	VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
5990	pgmR3RefreshShadowModeAfterA20Change(pVCpu);
5991	HMFlushTlb(pVCpu);
5992	#endif
5993	#if 0 /* PGMGetPage will apply the A20 mask to the GCPhys it returns, so we must invalid both sides of the TLB. */
5994	IEMTlbInvalidateAllPhysical(pVCpu);
5995	#else
5996	IEMTlbInvalidateAll(pVCpu);
5997	#endif
5998	STAM_REL_COUNTER_INC(&pVCpu->pgm.s.cA20Changes);
5999	}
6000	}
6001	#endif
6002

Note: See TracBrowser for help on using the repository browser.

source: vbox/trunk/src/VBox/VMM/VMMR3/PGMPhys.cpp@ 100696

Download in other formats: