SUPDrv.c@ 10265

Last change on this file since 10265 was 10265, checked in by vboxsync, 16 years ago
Some more IDC code.
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 153.4 KB

Line
1	/* $Revision: 10265 $ */
2	/** @file
3	* VirtualBox Support Driver - Shared code.
4	*/
5
6	/*
7	* Copyright (C) 2006-2007 Sun Microsystems, Inc.
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*
17	* The contents of this file may alternatively be used under the terms
18	* of the Common Development and Distribution License Version 1.0
19	* (CDDL) only, as it comes in the "COPYING.CDDL" file of the
20	* VirtualBox OSE distribution, in which case the provisions of the
21	* CDDL are applicable instead of those of the GPL.
22	*
23	* You may elect to license modified versions of this file under the
24	* terms and conditions of either the GPL or the CDDL or both.
25	*
26	* Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa
27	* Clara, CA 95054 USA or visit http://www.sun.com if you need
28	* additional information or have any questions.
29	*/
30
31
32	/*******************************************************************************
33	* Header Files *
34	*******************************************************************************/
35	#include "SUPDrvInternal.h"
36	#ifndef PAGE_SHIFT
37	# include <iprt/param.h>
38	#endif
39	#include <iprt/alloc.h>
40	#include <iprt/semaphore.h>
41	#include <iprt/spinlock.h>
42	#include <iprt/thread.h>
43	#include <iprt/process.h>
44	#include <iprt/mp.h>
45	#include <iprt/cpuset.h>
46	#include <iprt/log.h>
47	/* VBox/x86.h not compatible with the Linux kernel sources */
48	#ifdef RT_OS_LINUX
49	# define X86_CPUID_VENDOR_AMD_EBX 0x68747541
50	# define X86_CPUID_VENDOR_AMD_ECX 0x444d4163
51	# define X86_CPUID_VENDOR_AMD_EDX 0x69746e65
52	#else
53	# include <VBox/x86.h>
54	#endif
55
56	/*
57	* Logging assignments:
58	* Log - useful stuff, like failures.
59	* LogFlow - program flow, except the really noisy bits.
60	* Log2 - Cleanup and IDTE
61	* Log3 - Loader flow noise.
62	* Log4 - Call VMMR0 flow noise.
63	* Log5 - Native yet-to-be-defined noise.
64	* Log6 - Native ioctl flow noise.
65	*
66	* Logging requires BUILD_TYPE=debug and possibly changes to the logger
67	* instanciation in log-vbox.c(pp).
68	*/
69
70
71	/*******************************************************************************
72	* Defined Constants And Macros *
73	*******************************************************************************/
74	/* from x86.h - clashes with linux thus this duplication */
75	#undef X86_CR0_PG
76	#define X86_CR0_PG RT_BIT(31)
77	#undef X86_CR0_PE
78	#define X86_CR0_PE RT_BIT(0)
79	#undef X86_CPUID_AMD_FEATURE_EDX_NX
80	#define X86_CPUID_AMD_FEATURE_EDX_NX RT_BIT(20)
81	#undef MSR_K6_EFER
82	#define MSR_K6_EFER 0xc0000080
83	#undef MSR_K6_EFER_NXE
84	#define MSR_K6_EFER_NXE RT_BIT(11)
85	#undef MSR_K6_EFER_LMA
86	#define MSR_K6_EFER_LMA RT_BIT(10)
87	#undef X86_CR4_PGE
88	#define X86_CR4_PGE RT_BIT(7)
89	#undef X86_CR4_PAE
90	#define X86_CR4_PAE RT_BIT(5)
91	#undef X86_CPUID_AMD_FEATURE_EDX_LONG_MODE
92	#define X86_CPUID_AMD_FEATURE_EDX_LONG_MODE RT_BIT(29)
93
94
95	/** The frequency by which we recalculate the u32UpdateHz and
96	* u32UpdateIntervalNS GIP members. The value must be a power of 2. */
97	#define GIP_UPDATEHZ_RECALC_FREQ 0x800
98
99	/**
100	* Validates a session pointer.
101	*
102	* @returns true/false accordingly.
103	* @param pSession The session.
104	*/
105	#define SUP_IS_SESSION_VALID(pSession) \
106	( VALID_PTR(pSession) \
107	&& pSession->u32Cookie == BIRD_INV)
108
109
110	/*******************************************************************************
111	* Global Variables *
112	*******************************************************************************/
113	/**
114	* Array of the R0 SUP API.
115	*/
116	static SUPFUNC g_aFunctions[] =
117	{
118	/* name function */
119	{ "SUPR0ObjRegister", (void *)SUPR0ObjRegister },
120	{ "SUPR0ObjAddRef", (void *)SUPR0ObjAddRef },
121	{ "SUPR0ObjRelease", (void *)SUPR0ObjRelease },
122	{ "SUPR0ObjVerifyAccess", (void *)SUPR0ObjVerifyAccess },
123	{ "SUPR0LockMem", (void *)SUPR0LockMem },
124	{ "SUPR0UnlockMem", (void *)SUPR0UnlockMem },
125	{ "SUPR0ContAlloc", (void *)SUPR0ContAlloc },
126	{ "SUPR0ContFree", (void *)SUPR0ContFree },
127	{ "SUPR0LowAlloc", (void *)SUPR0LowAlloc },
128	{ "SUPR0LowFree", (void *)SUPR0LowFree },
129	{ "SUPR0MemAlloc", (void *)SUPR0MemAlloc },
130	{ "SUPR0MemGetPhys", (void *)SUPR0MemGetPhys },
131	{ "SUPR0MemFree", (void *)SUPR0MemFree },
132	{ "SUPR0PageAlloc", (void *)SUPR0PageAlloc },
133	{ "SUPR0PageFree", (void *)SUPR0PageFree },
134	{ "SUPR0Printf", (void *)SUPR0Printf },
135	{ "RTMemAlloc", (void *)RTMemAlloc },
136	{ "RTMemAllocZ", (void *)RTMemAllocZ },
137	{ "RTMemFree", (void *)RTMemFree },
138	/{ "RTMemDup", (void )RTMemDup },*/
139	{ "RTMemRealloc", (void *)RTMemRealloc },
140	{ "RTR0MemObjAllocLow", (void *)RTR0MemObjAllocLow },
141	{ "RTR0MemObjAllocPage", (void *)RTR0MemObjAllocPage },
142	{ "RTR0MemObjAllocPhys", (void *)RTR0MemObjAllocPhys },
143	{ "RTR0MemObjAllocPhysNC", (void *)RTR0MemObjAllocPhysNC },
144	{ "RTR0MemObjAllocCont", (void *)RTR0MemObjAllocCont },
145	{ "RTR0MemObjLockUser", (void *)RTR0MemObjLockUser },
146	{ "RTR0MemObjMapKernel", (void *)RTR0MemObjMapKernel },
147	{ "RTR0MemObjMapUser", (void *)RTR0MemObjMapUser },
148	{ "RTR0MemObjAddress", (void *)RTR0MemObjAddress },
149	{ "RTR0MemObjAddressR3", (void *)RTR0MemObjAddressR3 },
150	{ "RTR0MemObjSize", (void *)RTR0MemObjSize },
151	{ "RTR0MemObjIsMapping", (void *)RTR0MemObjIsMapping },
152	{ "RTR0MemObjGetPagePhysAddr", (void *)RTR0MemObjGetPagePhysAddr },
153	{ "RTR0MemObjFree", (void *)RTR0MemObjFree },
154	/* These don't work yet on linux - use fast mutexes!
155	{ "RTSemMutexCreate", (void *)RTSemMutexCreate },
156	{ "RTSemMutexRequest", (void *)RTSemMutexRequest },
157	{ "RTSemMutexRelease", (void *)RTSemMutexRelease },
158	{ "RTSemMutexDestroy", (void *)RTSemMutexDestroy },
159	*/
160	{ "RTProcSelf", (void *)RTProcSelf },
161	{ "RTR0ProcHandleSelf", (void *)RTR0ProcHandleSelf },
162	{ "RTSemFastMutexCreate", (void *)RTSemFastMutexCreate },
163	{ "RTSemFastMutexDestroy", (void *)RTSemFastMutexDestroy },
164	{ "RTSemFastMutexRequest", (void *)RTSemFastMutexRequest },
165	{ "RTSemFastMutexRelease", (void *)RTSemFastMutexRelease },
166	{ "RTSemEventCreate", (void *)RTSemEventCreate },
167	{ "RTSemEventSignal", (void *)RTSemEventSignal },
168	{ "RTSemEventWait", (void *)RTSemEventWait },
169	{ "RTSemEventWaitNoResume", (void *)RTSemEventWaitNoResume },
170	{ "RTSemEventDestroy", (void *)RTSemEventDestroy },
171	{ "RTSemEventMultiCreate", (void *)RTSemEventMultiCreate },
172	{ "RTSemEventMultiSignal", (void *)RTSemEventMultiSignal },
173	{ "RTSemEventMultiReset", (void *)RTSemEventMultiReset },
174	{ "RTSemEventMultiWait", (void *)RTSemEventMultiWait },
175	{ "RTSemEventMultiWaitNoResume", (void *)RTSemEventMultiWaitNoResume },
176	{ "RTSemEventMultiDestroy", (void *)RTSemEventMultiDestroy },
177	{ "RTSpinlockCreate", (void *)RTSpinlockCreate },
178	{ "RTSpinlockDestroy", (void *)RTSpinlockDestroy },
179	{ "RTSpinlockAcquire", (void *)RTSpinlockAcquire },
180	{ "RTSpinlockRelease", (void *)RTSpinlockRelease },
181	{ "RTSpinlockAcquireNoInts", (void *)RTSpinlockAcquireNoInts },
182	{ "RTSpinlockReleaseNoInts", (void *)RTSpinlockReleaseNoInts },
183	{ "RTThreadNativeSelf", (void *)RTThreadNativeSelf },
184	{ "RTThreadSleep", (void *)RTThreadSleep },
185	{ "RTThreadYield", (void *)RTThreadYield },
186	#if 0 /* Thread APIs, Part 2. */
187	{ "RTThreadSelf", (void *)RTThreadSelf },
188	{ "RTThreadCreate", (void *)RTThreadCreate },
189	{ "RTThreadGetNative", (void *)RTThreadGetNative },
190	{ "RTThreadWait", (void *)RTThreadWait },
191	{ "RTThreadWaitNoResume", (void *)RTThreadWaitNoResume },
192	{ "RTThreadGetName", (void *)RTThreadGetName },
193	{ "RTThreadSelfName", (void *)RTThreadSelfName },
194	{ "RTThreadGetType", (void *)RTThreadGetType },
195	{ "RTThreadUserSignal", (void *)RTThreadUserSignal },
196	{ "RTThreadUserReset", (void *)RTThreadUserReset },
197	{ "RTThreadUserWait", (void *)RTThreadUserWait },
198	{ "RTThreadUserWaitNoResume", (void *)RTThreadUserWaitNoResume },
199	#endif
200	{ "RTLogDefaultInstance", (void *)RTLogDefaultInstance },
201	{ "RTMpCpuId", (void *)RTMpCpuId },
202	{ "RTMpCpuIdFromSetIndex", (void *)RTMpCpuIdFromSetIndex },
203	{ "RTMpCpuIdToSetIndex", (void *)RTMpCpuIdToSetIndex },
204	{ "RTMpIsCpuPossible", (void *)RTMpIsCpuPossible },
205	{ "RTMpGetCount", (void *)RTMpGetCount },
206	{ "RTMpGetMaxCpuId", (void *)RTMpGetMaxCpuId },
207	{ "RTMpGetOnlineCount", (void *)RTMpGetOnlineCount },
208	{ "RTMpGetOnlineSet", (void *)RTMpGetOnlineSet },
209	{ "RTMpGetSet", (void *)RTMpGetSet },
210	{ "RTMpIsCpuOnline", (void *)RTMpIsCpuOnline },
211	{ "RTMpOnAll", (void *)RTMpOnAll },
212	{ "RTMpOnOthers", (void *)RTMpOnOthers },
213	{ "RTMpOnSpecific", (void *)RTMpOnSpecific },
214	{ "RTLogRelDefaultInstance", (void *)RTLogRelDefaultInstance },
215	{ "RTLogSetDefaultInstanceThread", (void *)RTLogSetDefaultInstanceThread },
216	{ "RTLogLogger", (void *)RTLogLogger },
217	{ "RTLogLoggerEx", (void *)RTLogLoggerEx },
218	{ "RTLogLoggerExV", (void *)RTLogLoggerExV },
219	{ "RTLogPrintf", (void *)RTLogPrintf },
220	{ "RTLogPrintfV", (void *)RTLogPrintfV },
221	{ "AssertMsg1", (void *)AssertMsg1 },
222	{ "AssertMsg2", (void *)AssertMsg2 },
223	};
224
225
226	/*******************************************************************************
227	* Internal Functions *
228	*******************************************************************************/
229	static int supdrvMemAdd(PSUPDRVMEMREF pMem, PSUPDRVSESSION pSession);
230	static int supdrvMemRelease(PSUPDRVSESSION pSession, RTHCUINTPTR uPtr, SUPDRVMEMREFTYPE eType);
231	#ifdef VBOX_WITH_IDT_PATCHING
232	static int supdrvIOCtl_IdtInstall(PSUPDRVDEVEXT pDevExt, PSUPDRVSESSION pSession, PSUPIDTINSTALL pReq);
233	static PSUPDRVPATCH supdrvIdtPatchOne(PSUPDRVDEVEXT pDevExt, PSUPDRVPATCH pPatch);
234	static int supdrvIOCtl_IdtRemoveAll(PSUPDRVDEVEXT pDevExt, PSUPDRVSESSION pSession);
235	static void supdrvIdtRemoveOne(PSUPDRVDEVEXT pDevExt, PSUPDRVPATCH pPatch);
236	static void supdrvIdtWrite(volatile void pvIdtEntry, const SUPDRVIDTE pNewIDTEntry);
237	#endif /* VBOX_WITH_IDT_PATCHING */
238	static int supdrvIOCtl_LdrOpen(PSUPDRVDEVEXT pDevExt, PSUPDRVSESSION pSession, PSUPLDROPEN pReq);
239	static int supdrvIOCtl_LdrLoad(PSUPDRVDEVEXT pDevExt, PSUPDRVSESSION pSession, PSUPLDRLOAD pReq);
240	static int supdrvIOCtl_LdrFree(PSUPDRVDEVEXT pDevExt, PSUPDRVSESSION pSession, PSUPLDRFREE pReq);
241	static int supdrvIOCtl_LdrGetSymbol(PSUPDRVDEVEXT pDevExt, PSUPDRVSESSION pSession, PSUPLDRGETSYMBOL pReq);
242	static int supdrvLdrSetR0EP(PSUPDRVDEVEXT pDevExt, void pvVMMR0, void pvVMMR0EntryInt, void pvVMMR0EntryFast, void pvVMMR0EntryEx);
243	static void supdrvLdrUnsetR0EP(PSUPDRVDEVEXT pDevExt);
244	static void supdrvLdrAddUsage(PSUPDRVSESSION pSession, PSUPDRVLDRIMAGE pImage);
245	static void supdrvLdrFree(PSUPDRVDEVEXT pDevExt, PSUPDRVLDRIMAGE pImage);
246	static SUPPAGINGMODE supdrvIOCtl_GetPagingMode(void);
247	static SUPGIPMODE supdrvGipDeterminTscMode(PSUPDRVDEVEXT pDevExt);
248	#ifdef RT_OS_WINDOWS
249	static int supdrvPageGetPhys(PSUPDRVSESSION pSession, RTR3PTR pvR3, uint32_t cPages, PRTHCPHYS paPages);
250	static bool supdrvPageWasLockedByPageAlloc(PSUPDRVSESSION pSession, RTR3PTR pvR3);
251	#endif /* RT_OS_WINDOWS */
252	static int supdrvGipCreate(PSUPDRVDEVEXT pDevExt);
253	static void supdrvGipDestroy(PSUPDRVDEVEXT pDevExt);
254	static DECLCALLBACK(void) supdrvGipSyncTimer(PRTTIMER pTimer, void *pvUser, uint64_t iTick);
255	static DECLCALLBACK(void) supdrvGipAsyncTimer(PRTTIMER pTimer, void *pvUser, uint64_t iTick);
256	static DECLCALLBACK(void) supdrvGipMpEvent(RTMPEVENT enmEvent, RTCPUID idCpu, void *pvUser);
257
258
259	/**
260	* Initializes the device extentsion structure.
261	*
262	* @returns IPRT status code.
263	* @param pDevExt The device extension to initialize.
264	*/
265	int VBOXCALL supdrvInitDevExt(PSUPDRVDEVEXT pDevExt)
266	{
267	/*
268	* Initialize it.
269	*/
270	int rc;
271	memset(pDevExt, 0, sizeof(*pDevExt));
272	rc = RTSpinlockCreate(&pDevExt->Spinlock);
273	if (!rc)
274	{
275	rc = RTSemFastMutexCreate(&pDevExt->mtxLdr);
276	if (!rc)
277	{
278	rc = RTSemFastMutexCreate(&pDevExt->mtxGip);
279	if (!rc)
280	{
281	rc = supdrvGipCreate(pDevExt);
282	if (RT_SUCCESS(rc))
283	{
284	pDevExt->u32Cookie = BIRD; /** @todo make this random? */
285	return VINF_SUCCESS;
286	}
287
288	RTSemFastMutexDestroy(pDevExt->mtxGip);
289	pDevExt->mtxGip = NIL_RTSEMFASTMUTEX;
290	}
291	RTSemFastMutexDestroy(pDevExt->mtxLdr);
292	pDevExt->mtxLdr = NIL_RTSEMFASTMUTEX;
293	}
294	RTSpinlockDestroy(pDevExt->Spinlock);
295	pDevExt->Spinlock = NIL_RTSPINLOCK;
296	}
297	return rc;
298	}
299
300
301	/**
302	* Delete the device extension (e.g. cleanup members).
303	*
304	* @param pDevExt The device extension to delete.
305	*/
306	void VBOXCALL supdrvDeleteDevExt(PSUPDRVDEVEXT pDevExt)
307	{
308	#ifdef VBOX_WITH_IDT_PATCHING
309	PSUPDRVPATCH pPatch;
310	#endif
311	PSUPDRVOBJ pObj;
312	PSUPDRVUSAGE pUsage;
313
314	/*
315	* Kill mutexes and spinlocks.
316	*/
317	RTSemFastMutexDestroy(pDevExt->mtxGip);
318	pDevExt->mtxGip = NIL_RTSEMFASTMUTEX;
319	RTSemFastMutexDestroy(pDevExt->mtxLdr);
320	pDevExt->mtxLdr = NIL_RTSEMFASTMUTEX;
321	RTSpinlockDestroy(pDevExt->Spinlock);
322	pDevExt->Spinlock = NIL_RTSPINLOCK;
323
324	/*
325	* Free lists.
326	*/
327	#ifdef VBOX_WITH_IDT_PATCHING
328	/* patches */
329	/** @todo make sure we don't uninstall patches which has been patched by someone else. */
330	pPatch = pDevExt->pIdtPatchesFree;
331	pDevExt->pIdtPatchesFree = NULL;
332	while (pPatch)
333	{
334	void *pvFree = pPatch;
335	pPatch = pPatch->pNext;
336	RTMemExecFree(pvFree);
337	}
338	#endif /* VBOX_WITH_IDT_PATCHING */
339
340	/* objects. */
341	pObj = pDevExt->pObjs;
342	#if !defined(DEBUG_bird) \|\| !defined(RT_OS_LINUX) /* breaks unloading, temporary, remove me! */
343	Assert(!pObj); /* (can trigger on forced unloads) */
344	#endif
345	pDevExt->pObjs = NULL;
346	while (pObj)
347	{
348	void *pvFree = pObj;
349	pObj = pObj->pNext;
350	RTMemFree(pvFree);
351	}
352
353	/* usage records. */
354	pUsage = pDevExt->pUsageFree;
355	pDevExt->pUsageFree = NULL;
356	while (pUsage)
357	{
358	void *pvFree = pUsage;
359	pUsage = pUsage->pNext;
360	RTMemFree(pvFree);
361	}
362
363	/* kill the GIP */
364	supdrvGipDestroy(pDevExt);
365	}
366
367
368	/**
369	* Create session.
370	*
371	* @returns IPRT status code.
372	* @param pDevExt Device extension.
373	* @param ppSession Where to store the pointer to the session data.
374	*/
375	int VBOXCALL supdrvCreateSession(PSUPDRVDEVEXT pDevExt, PSUPDRVSESSION *ppSession)
376	{
377	/*
378	* Allocate memory for the session data.
379	*/
380	int rc = VERR_NO_MEMORY;
381	PSUPDRVSESSION pSession = ppSession = (PSUPDRVSESSION)RTMemAllocZ(sizeof(pSession));
382	if (pSession)
383	{
384	/* Initialize session data. */
385	rc = RTSpinlockCreate(&pSession->Spinlock);
386	if (!rc)
387	{
388	Assert(pSession->Spinlock != NIL_RTSPINLOCK);
389	pSession->pDevExt = pDevExt;
390	pSession->u32Cookie = BIRD_INV;
391	/*pSession->pLdrUsage = NULL;
392	pSession->pPatchUsage = NULL;
393	pSession->pUsage = NULL;
394	pSession->pGip = NULL;
395	pSession->fGipReferenced = false;
396	pSession->Bundle.cUsed = 0 */
397
398	LogFlow(("Created session %p initial cookie=%#x\n", pSession, pSession->u32Cookie));
399	return VINF_SUCCESS;
400	}
401
402	RTMemFree(pSession);
403	*ppSession = NULL;
404	Log(("Failed to create spinlock, rc=%d!\n", rc));
405	}
406
407	return rc;
408	}
409
410
411	/**
412	* Shared code for cleaning up a session.
413	*
414	* @param pDevExt Device extension.
415	* @param pSession Session data.
416	* This data will be freed by this routine.
417	*/
418	void VBOXCALL supdrvCloseSession(PSUPDRVDEVEXT pDevExt, PSUPDRVSESSION pSession)
419	{
420	/*
421	* Cleanup the session first.
422	*/
423	supdrvCleanupSession(pDevExt, pSession);
424
425	/*
426	* Free the rest of the session stuff.
427	*/
428	RTSpinlockDestroy(pSession->Spinlock);
429	pSession->Spinlock = NIL_RTSPINLOCK;
430	pSession->pDevExt = NULL;
431	RTMemFree(pSession);
432	LogFlow(("supdrvCloseSession: returns\n"));
433	}
434
435
436	/**
437	* Shared code for cleaning up a session (but not quite freeing it).
438	*
439	* This is primarily intended for MAC OS X where we have to clean up the memory
440	* stuff before the file handle is closed.
441	*
442	* @param pDevExt Device extension.
443	* @param pSession Session data.
444	* This data will be freed by this routine.
445	*/
446	void VBOXCALL supdrvCleanupSession(PSUPDRVDEVEXT pDevExt, PSUPDRVSESSION pSession)
447	{
448	PSUPDRVBUNDLE pBundle;
449	LogFlow(("supdrvCleanupSession: pSession=%p\n", pSession));
450
451	/*
452	* Remove logger instances related to this session.
453	*/
454	RTLogSetDefaultInstanceThread(NULL, (uintptr_t)pSession);
455
456	#ifdef VBOX_WITH_IDT_PATCHING
457	/*
458	* Uninstall any IDT patches installed for this session.
459	*/
460	supdrvIOCtl_IdtRemoveAll(pDevExt, pSession);
461	#endif
462
463	/*
464	* Release object references made in this session.
465	* In theory there should be noone racing us in this session.
466	*/
467	Log2(("release objects - start\n"));
468	if (pSession->pUsage)
469	{
470	RTSPINLOCKTMP SpinlockTmp = RTSPINLOCKTMP_INITIALIZER;
471	PSUPDRVUSAGE pUsage;
472	RTSpinlockAcquire(pDevExt->Spinlock, &SpinlockTmp);
473
474	while ((pUsage = pSession->pUsage) != NULL)
475	{
476	PSUPDRVOBJ pObj = pUsage->pObj;
477	pSession->pUsage = pUsage->pNext;
478
479	AssertMsg(pUsage->cUsage >= 1 && pObj->cUsage >= pUsage->cUsage, ("glob %d; sess %d\n", pObj->cUsage, pUsage->cUsage));
480	if (pUsage->cUsage < pObj->cUsage)
481	{
482	pObj->cUsage -= pUsage->cUsage;
483	RTSpinlockRelease(pDevExt->Spinlock, &SpinlockTmp);
484	}
485	else
486	{
487	/* Destroy the object and free the record. */
488	if (pDevExt->pObjs == pObj)
489	pDevExt->pObjs = pObj->pNext;
490	else
491	{
492	PSUPDRVOBJ pObjPrev;
493	for (pObjPrev = pDevExt->pObjs; pObjPrev; pObjPrev = pObjPrev->pNext)
494	if (pObjPrev->pNext == pObj)
495	{
496	pObjPrev->pNext = pObj->pNext;
497	break;
498	}
499	Assert(pObjPrev);
500	}
501	RTSpinlockRelease(pDevExt->Spinlock, &SpinlockTmp);
502
503	Log(("supdrvCleanupSession: destroying %p/%d (%p/%p) cpid=%RTproc pid=%RTproc dtor=%p\n",
504	pObj, pObj->enmType, pObj->pvUser1, pObj->pvUser2, pObj->CreatorProcess, RTProcSelf(), pObj->pfnDestructor));
505	if (pObj->pfnDestructor)
506	pObj->pfnDestructor(pObj, pObj->pvUser1, pObj->pvUser2);
507	RTMemFree(pObj);
508	}
509
510	/* free it and continue. */
511	RTMemFree(pUsage);
512
513	RTSpinlockAcquire(pDevExt->Spinlock, &SpinlockTmp);
514	}
515
516	RTSpinlockRelease(pDevExt->Spinlock, &SpinlockTmp);
517	AssertMsg(!pSession->pUsage, ("Some buster reregistered an object during desturction!\n"));
518	}
519	Log2(("release objects - done\n"));
520
521	/*
522	* Release memory allocated in the session.
523	*
524	* We do not serialize this as we assume that the application will
525	* not allocated memory while closing the file handle object.
526	*/
527	Log2(("freeing memory:\n"));
528	pBundle = &pSession->Bundle;
529	while (pBundle)
530	{
531	PSUPDRVBUNDLE pToFree;
532	unsigned i;
533
534	/*
535	* Check and unlock all entries in the bundle.
536	*/
537	for (i = 0; i < RT_ELEMENTS(pBundle->aMem); i++)
538	{
539	if (pBundle->aMem[i].MemObj != NIL_RTR0MEMOBJ)
540	{
541	int rc;
542	Log2(("eType=%d pvR0=%p pvR3=%p cb=%ld\n", pBundle->aMem[i].eType, RTR0MemObjAddress(pBundle->aMem[i].MemObj),
543	(void *)RTR0MemObjAddressR3(pBundle->aMem[i].MapObjR3), (long)RTR0MemObjSize(pBundle->aMem[i].MemObj)));
544	if (pBundle->aMem[i].MapObjR3 != NIL_RTR0MEMOBJ)
545	{
546	rc = RTR0MemObjFree(pBundle->aMem[i].MapObjR3, false);
547	AssertRC(rc); /** @todo figure out how to handle this. */
548	pBundle->aMem[i].MapObjR3 = NIL_RTR0MEMOBJ;
549	}
550	rc = RTR0MemObjFree(pBundle->aMem[i].MemObj, false);
551	AssertRC(rc); /** @todo figure out how to handle this. */
552	pBundle->aMem[i].MemObj = NIL_RTR0MEMOBJ;
553	pBundle->aMem[i].eType = MEMREF_TYPE_UNUSED;
554	}
555	}
556
557	/*
558	* Advance and free previous bundle.
559	*/
560	pToFree = pBundle;
561	pBundle = pBundle->pNext;
562
563	pToFree->pNext = NULL;
564	pToFree->cUsed = 0;
565	if (pToFree != &pSession->Bundle)
566	RTMemFree(pToFree);
567	}
568	Log2(("freeing memory - done\n"));
569
570	/*
571	* Loaded images needs to be dereferenced and possibly freed up.
572	*/
573	RTSemFastMutexRequest(pDevExt->mtxLdr);
574	Log2(("freeing images:\n"));
575	if (pSession->pLdrUsage)
576	{
577	PSUPDRVLDRUSAGE pUsage = pSession->pLdrUsage;
578	pSession->pLdrUsage = NULL;
579	while (pUsage)
580	{
581	void *pvFree = pUsage;
582	PSUPDRVLDRIMAGE pImage = pUsage->pImage;
583	if (pImage->cUsage > pUsage->cUsage)
584	pImage->cUsage -= pUsage->cUsage;
585	else
586	supdrvLdrFree(pDevExt, pImage);
587	pUsage->pImage = NULL;
588	pUsage = pUsage->pNext;
589	RTMemFree(pvFree);
590	}
591	}
592	RTSemFastMutexRelease(pDevExt->mtxLdr);
593	Log2(("freeing images - done\n"));
594
595	/*
596	* Unmap the GIP.
597	*/
598	Log2(("umapping GIP:\n"));
599	if (pSession->GipMapObjR3 != NIL_RTR0MEMOBJ)
600	{
601	SUPR0GipUnmap(pSession);
602	pSession->fGipReferenced = 0;
603	}
604	Log2(("umapping GIP - done\n"));
605	}
606
607
608	/**
609	* Fast path I/O Control worker.
610	*
611	* @returns VBox status code that should be passed down to ring-3 unchanged.
612	* @param uIOCtl Function number.
613	* @param pDevExt Device extention.
614	* @param pSession Session data.
615	*/
616	int VBOXCALL supdrvIOCtlFast(uintptr_t uIOCtl, PSUPDRVDEVEXT pDevExt, PSUPDRVSESSION pSession)
617	{
618	int rc;
619
620	/*
621	* We check the two prereqs after doing this only to allow the compiler to optimize things better.
622	*/
623	if (RT_LIKELY(pSession->pVM && pDevExt->pfnVMMR0EntryFast))
624	{
625	switch (uIOCtl)
626	{
627	case SUP_IOCTL_FAST_DO_RAW_RUN:
628	rc = pDevExt->pfnVMMR0EntryFast(pSession->pVM, SUP_VMMR0_DO_RAW_RUN);
629	break;
630	case SUP_IOCTL_FAST_DO_HWACC_RUN:
631	rc = pDevExt->pfnVMMR0EntryFast(pSession->pVM, SUP_VMMR0_DO_HWACC_RUN);
632	break;
633	case SUP_IOCTL_FAST_DO_NOP:
634	rc = pDevExt->pfnVMMR0EntryFast(pSession->pVM, SUP_VMMR0_DO_NOP);
635	break;
636	default:
637	rc = VERR_INTERNAL_ERROR;
638	break;
639	}
640	}
641	else
642	rc = VERR_INTERNAL_ERROR;
643
644	return rc;
645	}
646
647
648	/**
649	* Helper for supdrvIOCtl. Check if pszStr contains any character of pszChars.
650	* We would use strpbrk here if this function would be contained in the RedHat kABI white
651	* list, see http://www.kerneldrivers.org/RHEL5.
652	*
653	* @return 1 if pszStr does contain any character of pszChars, 0 otherwise.
654	* @param pszStr String to check
655	* @param pszChars Character set
656	*/
657	static int supdrvCheckInvalidChar(const char pszStr, const char pszChars)
658	{
659	int chCur;
660	while ((chCur = *pszStr++) != '\0')
661	{
662	int ch;
663	const char *psz = pszChars;
664	while ((ch = *psz++) != '\0')
665	if (ch == chCur)
666	return 1;
667
668	}
669	return 0;
670	}
671
672
673	/**
674	* I/O Control worker.
675	*
676	* @returns 0 on success.
677	* @returns VERR_INVALID_PARAMETER if the request is invalid.
678	*
679	* @param uIOCtl Function number.
680	* @param pDevExt Device extention.
681	* @param pSession Session data.
682	* @param pReqHdr The request header.
683	*/
684	int VBOXCALL supdrvIOCtl(uintptr_t uIOCtl, PSUPDRVDEVEXT pDevExt, PSUPDRVSESSION pSession, PSUPREQHDR pReqHdr)
685	{
686	/*
687	* Validate the request.
688	*/
689	/* this first check could probably be omitted as its also done by the OS specific code... */
690	if (RT_UNLIKELY( (pReqHdr->fFlags & SUPREQHDR_FLAGS_MAGIC_MASK) != SUPREQHDR_FLAGS_MAGIC
691	\|\| pReqHdr->cbIn < sizeof(*pReqHdr)
692	\|\| pReqHdr->cbOut < sizeof(*pReqHdr)))
693	{
694	OSDBGPRINT(("vboxdrv: Bad ioctl request header; cbIn=%#lx cbOut=%#lx fFlags=%#lx\n",
695	(long)pReqHdr->cbIn, (long)pReqHdr->cbOut, (long)pReqHdr->fFlags));
696	return VERR_INVALID_PARAMETER;
697	}
698	if (RT_UNLIKELY(uIOCtl == SUP_IOCTL_COOKIE))
699	{
700	if (pReqHdr->u32Cookie != SUPCOOKIE_INITIAL_COOKIE)
701	{
702	OSDBGPRINT(("SUP_IOCTL_COOKIE: bad cookie %#lx\n", (long)pReqHdr->u32Cookie));
703	return VERR_INVALID_PARAMETER;
704	}
705	}
706	else if (RT_UNLIKELY( pReqHdr->u32Cookie != pDevExt->u32Cookie
707	\|\| pReqHdr->u32SessionCookie != pSession->u32Cookie))
708	{
709	OSDBGPRINT(("vboxdrv: bad cookie %#lx / %#lx.\n", (long)pReqHdr->u32Cookie, (long)pReqHdr->u32SessionCookie));
710	return VERR_INVALID_PARAMETER;
711	}
712
713	/*
714	* Validation macros
715	*/
716	#define REQ_CHECK_SIZES_EX(Name, cbInExpect, cbOutExpect) \
717	do { \
718	if (RT_UNLIKELY(pReqHdr->cbIn != (cbInExpect) \|\| pReqHdr->cbOut != (cbOutExpect))) \
719	{ \
720	OSDBGPRINT(( #Name ": Invalid input/output sizes. cbIn=%ld expected %ld. cbOut=%ld expected %ld.\n", \
721	(long)pReq->Hdr.cbIn, (long)(cbInExpect), (long)pReq->Hdr.cbOut, (long)(cbOutExpect))); \
722	return pReq->Hdr.rc = VERR_INVALID_PARAMETER; \
723	} \
724	} while (0)
725
726	#define REQ_CHECK_SIZES(Name) REQ_CHECK_SIZES_EX(Name, Name ## _SIZE_IN, Name ## _SIZE_OUT)
727
728	#define REQ_CHECK_SIZE_IN(Name, cbInExpect) \
729	do { \
730	if (RT_UNLIKELY(pReqHdr->cbIn != (cbInExpect))) \
731	{ \
732	OSDBGPRINT(( #Name ": Invalid input/output sizes. cbIn=%ld expected %ld.\n", \
733	(long)pReq->Hdr.cbIn, (long)(cbInExpect))); \
734	return pReq->Hdr.rc = VERR_INVALID_PARAMETER; \
735	} \
736	} while (0)
737
738	#define REQ_CHECK_SIZE_OUT(Name, cbOutExpect) \
739	do { \
740	if (RT_UNLIKELY(pReqHdr->cbOut != (cbOutExpect))) \
741	{ \
742	OSDBGPRINT(( #Name ": Invalid input/output sizes. cbOut=%ld expected %ld.\n", \
743	(long)pReq->Hdr.cbOut, (long)(cbOutExpect))); \
744	return pReq->Hdr.rc = VERR_INVALID_PARAMETER; \
745	} \
746	} while (0)
747
748	#define REQ_CHECK_EXPR(Name, expr) \
749	do { \
750	if (RT_UNLIKELY(!(expr))) \
751	{ \
752	OSDBGPRINT(( #Name ": %s\n", #expr)); \
753	return pReq->Hdr.rc = VERR_INVALID_PARAMETER; \
754	} \
755	} while (0)
756
757	#define REQ_CHECK_EXPR_FMT(expr, fmt) \
758	do { \
759	if (RT_UNLIKELY(!(expr))) \
760	{ \
761	OSDBGPRINT( fmt ); \
762	return pReq->Hdr.rc = VERR_INVALID_PARAMETER; \
763	} \
764	} while (0)
765
766
767	/*
768	* The switch.
769	*/
770	switch (SUP_CTL_CODE_NO_SIZE(uIOCtl))
771	{
772	case SUP_CTL_CODE_NO_SIZE(SUP_IOCTL_COOKIE):
773	{
774	PSUPCOOKIE pReq = (PSUPCOOKIE)pReqHdr;
775	REQ_CHECK_SIZES(SUP_IOCTL_COOKIE);
776	if (strncmp(pReq->u.In.szMagic, SUPCOOKIE_MAGIC, sizeof(pReq->u.In.szMagic)))
777	{
778	OSDBGPRINT(("SUP_IOCTL_COOKIE: invalid magic %.16s\n", pReq->u.In.szMagic));
779	pReq->Hdr.rc = VERR_INVALID_MAGIC;
780	return 0;
781	}
782
783	#if 0
784	/*
785	* Call out to the OS specific code and let it do permission checks on the
786	* client process.
787	*/
788	if (!supdrvOSValidateClientProcess(pDevExt, pSession))
789	{
790	pReq->u.Out.u32Cookie = 0xffffffff;
791	pReq->u.Out.u32SessionCookie = 0xffffffff;
792	pReq->u.Out.u32SessionVersion = 0xffffffff;
793	pReq->u.Out.u32DriverVersion = SUPDRVIOC_VERSION;
794	pReq->u.Out.pSession = NULL;
795	pReq->u.Out.cFunctions = 0;
796	pReq->Hdr.rc = VERR_PERMISSION_DENIED;
797	return 0;
798	}
799	#endif
800
801	/*
802	* Match the version.
803	* The current logic is very simple, match the major interface version.
804	*/
805	if ( pReq->u.In.u32MinVersion > SUPDRVIOC_VERSION
806	\|\| (pReq->u.In.u32MinVersion & 0xffff0000) != (SUPDRVIOC_VERSION & 0xffff0000))
807	{
808	OSDBGPRINT(("SUP_IOCTL_COOKIE: Version mismatch. Requested: %#x Min: %#x Current: %#x\n",
809	pReq->u.In.u32ReqVersion, pReq->u.In.u32MinVersion, SUPDRVIOC_VERSION));
810	pReq->u.Out.u32Cookie = 0xffffffff;
811	pReq->u.Out.u32SessionCookie = 0xffffffff;
812	pReq->u.Out.u32SessionVersion = 0xffffffff;
813	pReq->u.Out.u32DriverVersion = SUPDRVIOC_VERSION;
814	pReq->u.Out.pSession = NULL;
815	pReq->u.Out.cFunctions = 0;
816	pReq->Hdr.rc = VERR_VERSION_MISMATCH;
817	return 0;
818	}
819
820	/*
821	* Fill in return data and be gone.
822	* N.B. The first one to change SUPDRVIOC_VERSION shall makes sure that
823	* u32SessionVersion <= u32ReqVersion!
824	*/
825	/** @todo Somehow validate the client and negotiate a secure cookie... */
826	pReq->u.Out.u32Cookie = pDevExt->u32Cookie;
827	pReq->u.Out.u32SessionCookie = pSession->u32Cookie;
828	pReq->u.Out.u32SessionVersion = SUPDRVIOC_VERSION;
829	pReq->u.Out.u32DriverVersion = SUPDRVIOC_VERSION;
830	pReq->u.Out.pSession = pSession;
831	pReq->u.Out.cFunctions = sizeof(g_aFunctions) / sizeof(g_aFunctions[0]);
832	pReq->Hdr.rc = VINF_SUCCESS;
833	return 0;
834	}
835
836	case SUP_CTL_CODE_NO_SIZE(SUP_IOCTL_QUERY_FUNCS(0)):
837	{
838	/* validate */
839	PSUPQUERYFUNCS pReq = (PSUPQUERYFUNCS)pReqHdr;
840	REQ_CHECK_SIZES_EX(SUP_IOCTL_QUERY_FUNCS, SUP_IOCTL_QUERY_FUNCS_SIZE_IN, SUP_IOCTL_QUERY_FUNCS_SIZE_OUT(RT_ELEMENTS(g_aFunctions)));
841
842	/* execute */
843	pReq->u.Out.cFunctions = RT_ELEMENTS(g_aFunctions);
844	memcpy(&pReq->u.Out.aFunctions[0], g_aFunctions, sizeof(g_aFunctions));
845	pReq->Hdr.rc = VINF_SUCCESS;
846	return 0;
847	}
848
849	case SUP_CTL_CODE_NO_SIZE(SUP_IOCTL_IDT_INSTALL):
850	{
851	/* validate */
852	PSUPIDTINSTALL pReq = (PSUPIDTINSTALL)pReqHdr;
853	REQ_CHECK_SIZES(SUP_IOCTL_IDT_INSTALL);
854
855	/* execute */
856	#ifdef VBOX_WITH_IDT_PATCHING
857	pReq->Hdr.rc = supdrvIOCtl_IdtInstall(pDevExt, pSession, pReq);
858	#else
859	pReq->u.Out.u8Idt = 3;
860	pReq->Hdr.rc = VERR_NOT_SUPPORTED;
861	#endif
862	return 0;
863	}
864
865	case SUP_CTL_CODE_NO_SIZE(SUP_IOCTL_IDT_REMOVE):
866	{
867	/* validate */
868	PSUPIDTREMOVE pReq = (PSUPIDTREMOVE)pReqHdr;
869	REQ_CHECK_SIZES(SUP_IOCTL_IDT_REMOVE);
870
871	/* execute */
872	#ifdef VBOX_WITH_IDT_PATCHING
873	pReq->Hdr.rc = supdrvIOCtl_IdtRemoveAll(pDevExt, pSession);
874	#else
875	pReq->Hdr.rc = VERR_NOT_SUPPORTED;
876	#endif
877	return 0;
878	}
879
880	case SUP_CTL_CODE_NO_SIZE(SUP_IOCTL_PAGE_LOCK):
881	{
882	/* validate */
883	PSUPPAGELOCK pReq = (PSUPPAGELOCK)pReqHdr;
884	REQ_CHECK_SIZE_IN(SUP_IOCTL_PAGE_LOCK, SUP_IOCTL_PAGE_LOCK_SIZE_IN);
885	REQ_CHECK_SIZE_OUT(SUP_IOCTL_PAGE_LOCK, SUP_IOCTL_PAGE_LOCK_SIZE_OUT(pReq->u.In.cPages));
886	REQ_CHECK_EXPR(SUP_IOCTL_PAGE_LOCK, pReq->u.In.cPages > 0);
887	REQ_CHECK_EXPR(SUP_IOCTL_PAGE_LOCK, pReq->u.In.pvR3 >= PAGE_SIZE);
888
889	/* execute */
890	pReq->Hdr.rc = SUPR0LockMem(pSession, pReq->u.In.pvR3, pReq->u.In.cPages, &pReq->u.Out.aPages[0]);
891	if (RT_FAILURE(pReq->Hdr.rc))
892	pReq->Hdr.cbOut = sizeof(pReq->Hdr);
893	return 0;
894	}
895
896	case SUP_CTL_CODE_NO_SIZE(SUP_IOCTL_PAGE_UNLOCK):
897	{
898	/* validate */
899	PSUPPAGEUNLOCK pReq = (PSUPPAGEUNLOCK)pReqHdr;
900	REQ_CHECK_SIZES(SUP_IOCTL_PAGE_UNLOCK);
901
902	/* execute */
903	pReq->Hdr.rc = SUPR0UnlockMem(pSession, pReq->u.In.pvR3);
904	return 0;
905	}
906
907	case SUP_CTL_CODE_NO_SIZE(SUP_IOCTL_CONT_ALLOC):
908	{
909	/* validate */
910	PSUPCONTALLOC pReq = (PSUPCONTALLOC)pReqHdr;
911	REQ_CHECK_SIZES(SUP_IOCTL_CONT_ALLOC);
912
913	/* execute */
914	pReq->Hdr.rc = SUPR0ContAlloc(pSession, pReq->u.In.cPages, &pReq->u.Out.pvR0, &pReq->u.Out.pvR3, &pReq->u.Out.HCPhys);
915	if (RT_FAILURE(pReq->Hdr.rc))
916	pReq->Hdr.cbOut = sizeof(pReq->Hdr);
917	return 0;
918	}
919
920	case SUP_CTL_CODE_NO_SIZE(SUP_IOCTL_CONT_FREE):
921	{
922	/* validate */
923	PSUPCONTFREE pReq = (PSUPCONTFREE)pReqHdr;
924	REQ_CHECK_SIZES(SUP_IOCTL_CONT_FREE);
925
926	/* execute */
927	pReq->Hdr.rc = SUPR0ContFree(pSession, (RTHCUINTPTR)pReq->u.In.pvR3);
928	return 0;
929	}
930
931	case SUP_CTL_CODE_NO_SIZE(SUP_IOCTL_LDR_OPEN):
932	{
933	/* validate */
934	PSUPLDROPEN pReq = (PSUPLDROPEN)pReqHdr;
935	REQ_CHECK_SIZES(SUP_IOCTL_LDR_OPEN);
936	REQ_CHECK_EXPR(SUP_IOCTL_LDR_OPEN, pReq->u.In.cbImage > 0);
937	REQ_CHECK_EXPR(SUP_IOCTL_LDR_OPEN, pReq->u.In.cbImage < _1M*16);
938	REQ_CHECK_EXPR(SUP_IOCTL_LDR_OPEN, pReq->u.In.szName[0]);
939	REQ_CHECK_EXPR(SUP_IOCTL_LDR_OPEN, memchr(pReq->u.In.szName, '\0', sizeof(pReq->u.In.szName)));
940	REQ_CHECK_EXPR(SUP_IOCTL_LDR_OPEN, !supdrvCheckInvalidChar(pReq->u.In.szName, ";:()[]{}/\\\|&*%#@!~`\"'"));
941
942	/* execute */
943	pReq->Hdr.rc = supdrvIOCtl_LdrOpen(pDevExt, pSession, pReq);
944	return 0;
945	}
946
947	case SUP_CTL_CODE_NO_SIZE(SUP_IOCTL_LDR_LOAD):
948	{
949	/* validate */
950	PSUPLDRLOAD pReq = (PSUPLDRLOAD)pReqHdr;
951	REQ_CHECK_EXPR(Name, pReq->Hdr.cbIn >= sizeof(*pReq));
952	REQ_CHECK_SIZES_EX(SUP_IOCTL_LDR_LOAD, SUP_IOCTL_LDR_LOAD_SIZE_IN(pReq->u.In.cbImage), SUP_IOCTL_LDR_LOAD_SIZE_OUT);
953	REQ_CHECK_EXPR(SUP_IOCTL_LDR_LOAD, pReq->u.In.cSymbols <= 16384);
954	REQ_CHECK_EXPR_FMT( !pReq->u.In.cSymbols
955	\|\| ( pReq->u.In.offSymbols < pReq->u.In.cbImage
956	&& pReq->u.In.offSymbols + pReq->u.In.cSymbols * sizeof(SUPLDRSYM) <= pReq->u.In.cbImage),
957	("SUP_IOCTL_LDR_LOAD: offSymbols=%#lx cSymbols=%#lx cbImage=%#lx\n", (long)pReq->u.In.offSymbols,
958	(long)pReq->u.In.cSymbols, (long)pReq->u.In.cbImage));
959	REQ_CHECK_EXPR_FMT( !pReq->u.In.cbStrTab
960	\|\| ( pReq->u.In.offStrTab < pReq->u.In.cbImage
961	&& pReq->u.In.offStrTab + pReq->u.In.cbStrTab <= pReq->u.In.cbImage
962	&& pReq->u.In.cbStrTab <= pReq->u.In.cbImage),
963	("SUP_IOCTL_LDR_LOAD: offStrTab=%#lx cbStrTab=%#lx cbImage=%#lx\n", (long)pReq->u.In.offStrTab,
964	(long)pReq->u.In.cbStrTab, (long)pReq->u.In.cbImage));
965
966	if (pReq->u.In.cSymbols)
967	{
968	uint32_t i;
969	PSUPLDRSYM paSyms = (PSUPLDRSYM)&pReq->u.In.achImage[pReq->u.In.offSymbols];
970	for (i = 0; i < pReq->u.In.cSymbols; i++)
971	{
972	REQ_CHECK_EXPR_FMT(paSyms[i].offSymbol < pReq->u.In.cbImage,
973	("SUP_IOCTL_LDR_LOAD: sym #%ld: symb off %#lx (max=%#lx)\n", (long)i, (long)paSyms[i].offSymbol, (long)pReq->u.In.cbImage));
974	REQ_CHECK_EXPR_FMT(paSyms[i].offName < pReq->u.In.cbStrTab,
975	("SUP_IOCTL_LDR_LOAD: sym #%ld: name off %#lx (max=%#lx)\n", (long)i, (long)paSyms[i].offName, (long)pReq->u.In.cbImage));
976	REQ_CHECK_EXPR_FMT(memchr(&pReq->u.In.achImage[pReq->u.In.offStrTab + paSyms[i].offName], '\0', pReq->u.In.cbStrTab - paSyms[i].offName),
977	("SUP_IOCTL_LDR_LOAD: sym #%ld: unterminated name! (%#lx / %#lx)\n", (long)i, (long)paSyms[i].offName, (long)pReq->u.In.cbImage));
978	}
979	}
980
981	/* execute */
982	pReq->Hdr.rc = supdrvIOCtl_LdrLoad(pDevExt, pSession, pReq);
983	return 0;
984	}
985
986	case SUP_CTL_CODE_NO_SIZE(SUP_IOCTL_LDR_FREE):
987	{
988	/* validate */
989	PSUPLDRFREE pReq = (PSUPLDRFREE)pReqHdr;
990	REQ_CHECK_SIZES(SUP_IOCTL_LDR_FREE);
991
992	/* execute */
993	pReq->Hdr.rc = supdrvIOCtl_LdrFree(pDevExt, pSession, pReq);
994	return 0;
995	}
996
997	case SUP_CTL_CODE_NO_SIZE(SUP_IOCTL_LDR_GET_SYMBOL):
998	{
999	/* validate */
1000	PSUPLDRGETSYMBOL pReq = (PSUPLDRGETSYMBOL)pReqHdr;
1001	REQ_CHECK_SIZES(SUP_IOCTL_LDR_GET_SYMBOL);
1002	REQ_CHECK_EXPR(SUP_IOCTL_LDR_GET_SYMBOL, memchr(pReq->u.In.szSymbol, '\0', sizeof(pReq->u.In.szSymbol)));
1003
1004	/* execute */
1005	pReq->Hdr.rc = supdrvIOCtl_LdrGetSymbol(pDevExt, pSession, pReq);
1006	return 0;
1007	}
1008
1009	case SUP_CTL_CODE_NO_SIZE(SUP_IOCTL_CALL_VMMR0(0)):
1010	{
1011	/* validate */
1012	PSUPCALLVMMR0 pReq = (PSUPCALLVMMR0)pReqHdr;
1013	Log4(("SUP_IOCTL_CALL_VMMR0: op=%u in=%u arg=%RX64 p/t=%RTproc/%RTthrd\n",
1014	pReq->u.In.uOperation, pReq->Hdr.cbIn, pReq->u.In.u64Arg, RTProcSelf(), RTThreadNativeSelf()));
1015
1016	if (pReq->Hdr.cbIn == SUP_IOCTL_CALL_VMMR0_SIZE(0))
1017	{
1018	REQ_CHECK_SIZES_EX(SUP_IOCTL_CALL_VMMR0, SUP_IOCTL_CALL_VMMR0_SIZE_IN(0), SUP_IOCTL_CALL_VMMR0_SIZE_OUT(0));
1019
1020	/* execute */
1021	if (RT_LIKELY(pDevExt->pfnVMMR0EntryEx))
1022	pReq->Hdr.rc = pDevExt->pfnVMMR0EntryEx(pReq->u.In.pVMR0, pReq->u.In.uOperation, NULL, pReq->u.In.u64Arg);
1023	else
1024	pReq->Hdr.rc = VERR_WRONG_ORDER;
1025	}
1026	else
1027	{
1028	PSUPVMMR0REQHDR pVMMReq = (PSUPVMMR0REQHDR)&pReq->abReqPkt[0];
1029	REQ_CHECK_EXPR_FMT(pReq->Hdr.cbIn >= SUP_IOCTL_CALL_VMMR0_SIZE(sizeof(SUPVMMR0REQHDR)),
1030	("SUP_IOCTL_CALL_VMMR0: cbIn=%#x < %#x\n", pReq->Hdr.cbIn, SUP_IOCTL_CALL_VMMR0_SIZE(sizeof(SUPVMMR0REQHDR))));
1031	REQ_CHECK_EXPR(SUP_IOCTL_CALL_VMMR0, pVMMReq->u32Magic == SUPVMMR0REQHDR_MAGIC);
1032	REQ_CHECK_SIZES_EX(SUP_IOCTL_CALL_VMMR0, SUP_IOCTL_CALL_VMMR0_SIZE_IN(pVMMReq->cbReq), SUP_IOCTL_CALL_VMMR0_SIZE_OUT(pVMMReq->cbReq));
1033
1034	/* execute */
1035	if (RT_LIKELY(pDevExt->pfnVMMR0EntryEx))
1036	pReq->Hdr.rc = pDevExt->pfnVMMR0EntryEx(pReq->u.In.pVMR0, pReq->u.In.uOperation, pVMMReq, pReq->u.In.u64Arg);
1037	else
1038	pReq->Hdr.rc = VERR_WRONG_ORDER;
1039	}
1040
1041	if ( RT_FAILURE(pReq->Hdr.rc)
1042	&& pReq->Hdr.rc != VERR_INTERRUPTED
1043	&& pReq->Hdr.rc != VERR_TIMEOUT)
1044	Log(("SUP_IOCTL_CALL_VMMR0: rc=%Rrc op=%u out=%u arg=%RX64 p/t=%RTproc/%RTthrd\n",
1045	pReq->Hdr.rc, pReq->u.In.uOperation, pReq->Hdr.cbOut, pReq->u.In.u64Arg, RTProcSelf(), RTThreadNativeSelf()));
1046	else
1047	Log4(("SUP_IOCTL_CALL_VMMR0: rc=%Rrc op=%u out=%u arg=%RX64 p/t=%RTproc/%RTthrd\n",
1048	pReq->Hdr.rc, pReq->u.In.uOperation, pReq->Hdr.cbOut, pReq->u.In.u64Arg, RTProcSelf(), RTThreadNativeSelf()));
1049	return 0;
1050	}
1051
1052	case SUP_CTL_CODE_NO_SIZE(SUP_IOCTL_GET_PAGING_MODE):
1053	{
1054	/* validate */
1055	PSUPGETPAGINGMODE pReq = (PSUPGETPAGINGMODE)pReqHdr;
1056	REQ_CHECK_SIZES(SUP_IOCTL_GET_PAGING_MODE);
1057
1058	/* execute */
1059	pReq->Hdr.rc = VINF_SUCCESS;
1060	pReq->u.Out.enmMode = supdrvIOCtl_GetPagingMode();
1061	return 0;
1062	}
1063
1064	case SUP_CTL_CODE_NO_SIZE(SUP_IOCTL_LOW_ALLOC):
1065	{
1066	/* validate */
1067	PSUPLOWALLOC pReq = (PSUPLOWALLOC)pReqHdr;
1068	REQ_CHECK_EXPR(SUP_IOCTL_LOW_ALLOC, pReq->Hdr.cbIn <= SUP_IOCTL_LOW_ALLOC_SIZE_IN);
1069	REQ_CHECK_SIZES_EX(SUP_IOCTL_LOW_ALLOC, SUP_IOCTL_LOW_ALLOC_SIZE_IN, SUP_IOCTL_LOW_ALLOC_SIZE_OUT(pReq->u.In.cPages));
1070
1071	/* execute */
1072	pReq->Hdr.rc = SUPR0LowAlloc(pSession, pReq->u.In.cPages, &pReq->u.Out.pvR0, &pReq->u.Out.pvR3, &pReq->u.Out.aPages[0]);
1073	if (RT_FAILURE(pReq->Hdr.rc))
1074	pReq->Hdr.cbOut = sizeof(pReq->Hdr);
1075	return 0;
1076	}
1077
1078	case SUP_CTL_CODE_NO_SIZE(SUP_IOCTL_LOW_FREE):
1079	{
1080	/* validate */
1081	PSUPLOWFREE pReq = (PSUPLOWFREE)pReqHdr;
1082	REQ_CHECK_SIZES(SUP_IOCTL_LOW_FREE);
1083
1084	/* execute */
1085	pReq->Hdr.rc = SUPR0LowFree(pSession, (RTHCUINTPTR)pReq->u.In.pvR3);
1086	return 0;
1087	}
1088
1089	case SUP_CTL_CODE_NO_SIZE(SUP_IOCTL_GIP_MAP):
1090	{
1091	/* validate */
1092	PSUPGIPMAP pReq = (PSUPGIPMAP)pReqHdr;
1093	REQ_CHECK_SIZES(SUP_IOCTL_GIP_MAP);
1094
1095	/* execute */
1096	pReq->Hdr.rc = SUPR0GipMap(pSession, &pReq->u.Out.pGipR3, &pReq->u.Out.HCPhysGip);
1097	if (RT_SUCCESS(pReq->Hdr.rc))
1098	pReq->u.Out.pGipR0 = pDevExt->pGip;
1099	return 0;
1100	}
1101
1102	case SUP_CTL_CODE_NO_SIZE(SUP_IOCTL_GIP_UNMAP):
1103	{
1104	/* validate */
1105	PSUPGIPUNMAP pReq = (PSUPGIPUNMAP)pReqHdr;
1106	REQ_CHECK_SIZES(SUP_IOCTL_GIP_UNMAP);
1107
1108	/* execute */
1109	pReq->Hdr.rc = SUPR0GipUnmap(pSession);
1110	return 0;
1111	}
1112
1113	case SUP_CTL_CODE_NO_SIZE(SUP_IOCTL_SET_VM_FOR_FAST):
1114	{
1115	/* validate */
1116	PSUPSETVMFORFAST pReq = (PSUPSETVMFORFAST)pReqHdr;
1117	REQ_CHECK_SIZES(SUP_IOCTL_SET_VM_FOR_FAST);
1118	REQ_CHECK_EXPR_FMT( !pReq->u.In.pVMR0
1119	\|\| ( VALID_PTR(pReq->u.In.pVMR0)
1120	&& !((uintptr_t)pReq->u.In.pVMR0 & (PAGE_SIZE - 1))),
1121	("SUP_IOCTL_SET_VM_FOR_FAST: pVMR0=%p!\n", pReq->u.In.pVMR0));
1122	/* execute */
1123	pSession->pVM = pReq->u.In.pVMR0;
1124	pReq->Hdr.rc = VINF_SUCCESS;
1125	return 0;
1126	}
1127
1128	case SUP_CTL_CODE_NO_SIZE(SUP_IOCTL_PAGE_ALLOC):
1129	{
1130	/* validate */
1131	PSUPPAGEALLOC pReq = (PSUPPAGEALLOC)pReqHdr;
1132	REQ_CHECK_EXPR(SUP_IOCTL_PAGE_ALLOC, pReq->Hdr.cbIn <= SUP_IOCTL_PAGE_ALLOC_SIZE_IN);
1133	REQ_CHECK_SIZES_EX(SUP_IOCTL_PAGE_ALLOC, SUP_IOCTL_PAGE_ALLOC_SIZE_IN, SUP_IOCTL_PAGE_ALLOC_SIZE_OUT(pReq->u.In.cPages));
1134
1135	/* execute */
1136	pReq->Hdr.rc = SUPR0PageAlloc(pSession, pReq->u.In.cPages, &pReq->u.Out.pvR3, &pReq->u.Out.aPages[0]);
1137	if (RT_FAILURE(pReq->Hdr.rc))
1138	pReq->Hdr.cbOut = sizeof(pReq->Hdr);
1139	return 0;
1140	}
1141
1142	case SUP_CTL_CODE_NO_SIZE(SUP_IOCTL_PAGE_FREE):
1143	{
1144	/* validate */
1145	PSUPPAGEFREE pReq = (PSUPPAGEFREE)pReqHdr;
1146	REQ_CHECK_SIZES(SUP_IOCTL_PAGE_FREE);
1147
1148	/* execute */
1149	pReq->Hdr.rc = SUPR0PageFree(pSession, pReq->u.In.pvR3);
1150	return 0;
1151	}
1152
1153	default:
1154	Log(("Unknown IOCTL %#lx\n", (long)uIOCtl));
1155	break;
1156	}
1157	return SUPDRV_ERR_GENERAL_FAILURE;
1158	}
1159
1160
1161	/**
1162	* Inter-Driver Communcation (IDC) worker.
1163	*
1164	* @returns VBox status code.
1165	* @retval VINF_SUCCESS on success.
1166	* @retval VERR_NOT_SUPPORTED if the request isn't supported.
1167	* @retval VERR_NOT_IMPLEMENTED if during development.
1168	* @retval VERR_INVALID_PARAMETER if the request is invalid.
1169	*
1170	* @param uReq The request (function) code.
1171	* @param pDevExt Device extention.
1172	* @param pSession Session data.
1173	* @param pReqHdr The request header.
1174	*/
1175	int VBOXCALL supdrvIDC(uintptr_t uReq, PSUPDRVDEVEXT pDevExt, PSUPDRVSESSION pSession, PSUPDRVIDCREQHDR pReqHdr)
1176	{
1177	/*
1178	* The OS specific code has already validated the pSession
1179	* pointer, and the request size being greater or equal to
1180	* size of the header.
1181	*
1182	* So, just check that pSession is a kernel context session.
1183	*/
1184	if (RT_UNLIKELY( pSession
1185	&& pSession->R0Process != NIL_RTR0PROCESS))
1186	return VERR_INVALID_PARAMETER;
1187
1188	/*
1189	* Validation macro.
1190	*/
1191	#define REQ_CHECK_IDC_SIZE(Name, cbExpect) \
1192	do { \
1193	if (RT_UNLIKELY(pReqHdr->cb != (cbExpect))) \
1194	{ \
1195	OSDBGPRINT(( #Name ": Invalid input/output sizes. cb=%ld expected %ld.\n", \
1196	(long)pReqHdr->cb, (long)(cbExpect))); \
1197	return pReqHdr->rc = VERR_INVALID_PARAMETER; \
1198	} \
1199	} while (0)
1200
1201	switch (uReq)
1202	{
1203	case SUPDRV_IDC_REQ_CONNECT:
1204	{
1205	PSUPDRVIDCREQCONNECT pReq = (PSUPDRVIDCREQCONNECT)pReqHdr;
1206	REQ_CHECK_IDC_SIZE(SUPDRV_IDC_REQ_CONNECT, sizeof(*pReq));
1207
1208	return VERR_NOT_IMPLEMENTED;
1209	}
1210
1211	case SUPDRV_IDC_REQ_DISCONNECT:
1212	{
1213	REQ_CHECK_IDC_SIZE(SUPDRV_IDC_REQ_DISCONNECT, sizeof(*pReqHdr));
1214
1215	return VERR_NOT_IMPLEMENTED;
1216	}
1217
1218	case SUPDRV_IDC_REQ_GET_SYMBOL:
1219	{
1220	PSUPDRVIDCREQGETSYM pReq;
1221	REQ_CHECK_IDC_SIZE(SUPDRV_IDC_REQ_GET_SYMBOL, sizeof(*pReq));
1222
1223	return VERR_NOT_IMPLEMENTED;
1224	}
1225
1226	case SUPDRV_IDC_REQ_COMPONENT_REGISTER_FACTORY:
1227	{
1228	PSUPDRVIDCREQCOMPREGFACTORY pReq;
1229	REQ_CHECK_IDC_SIZE(SUPDRV_IDC_REQ_COMPONENT_REGISTER_FACTORY, sizeof(*pReq));
1230
1231	return VERR_NOT_IMPLEMENTED;
1232	}
1233
1234	case SUPDRV_IDC_REQ_COMPONENT_DEREGISTER_FACTORY:
1235	{
1236	PSUPDRVIDCREQCOMPDEREGFACTORY pReq;
1237	REQ_CHECK_IDC_SIZE(SUPDRV_IDC_REQ_COMPONENT_DEREGISTER_FACTORY, sizeof(*pReq));
1238
1239	return VERR_NOT_IMPLEMENTED;
1240	}
1241
1242	default:
1243	Log(("Unknown IDC %#lx\n", (long)uReq));
1244	break;
1245	}
1246
1247	#undef REQ_CHECK_IDC_SIZE
1248	return VERR_NOT_SUPPORTED;
1249	}
1250
1251
1252	/**
1253	* Register a object for reference counting.
1254	* The object is registered with one reference in the specified session.
1255	*
1256	* @returns Unique identifier on success (pointer).
1257	* All future reference must use this identifier.
1258	* @returns NULL on failure.
1259	* @param pfnDestructor The destructore function which will be called when the reference count reaches 0.
1260	* @param pvUser1 The first user argument.
1261	* @param pvUser2 The second user argument.
1262	*/
1263	SUPR0DECL(void ) SUPR0ObjRegister(PSUPDRVSESSION pSession, SUPDRVOBJTYPE enmType, PFNSUPDRVDESTRUCTOR pfnDestructor, void pvUser1, void *pvUser2)
1264	{
1265	RTSPINLOCKTMP SpinlockTmp = RTSPINLOCKTMP_INITIALIZER;
1266	PSUPDRVDEVEXT pDevExt = pSession->pDevExt;
1267	PSUPDRVOBJ pObj;
1268	PSUPDRVUSAGE pUsage;
1269
1270	/*
1271	* Validate the input.
1272	*/
1273	AssertReturn(SUP_IS_SESSION_VALID(pSession), NULL);
1274	AssertReturn(enmType > SUPDRVOBJTYPE_INVALID && enmType < SUPDRVOBJTYPE_END, NULL);
1275	AssertPtrReturn(pfnDestructor, NULL);
1276
1277	/*
1278	* Allocate and initialize the object.
1279	*/
1280	pObj = (PSUPDRVOBJ)RTMemAlloc(sizeof(*pObj));
1281	if (!pObj)
1282	return NULL;
1283	pObj->u32Magic = SUPDRVOBJ_MAGIC;
1284	pObj->enmType = enmType;
1285	pObj->pNext = NULL;
1286	pObj->cUsage = 1;
1287	pObj->pfnDestructor = pfnDestructor;
1288	pObj->pvUser1 = pvUser1;
1289	pObj->pvUser2 = pvUser2;
1290	pObj->CreatorUid = pSession->Uid;
1291	pObj->CreatorGid = pSession->Gid;
1292	pObj->CreatorProcess= pSession->Process;
1293	supdrvOSObjInitCreator(pObj, pSession);
1294
1295	/*
1296	* Allocate the usage record.
1297	* (We keep freed usage records around to simplify SUPR0ObjAddRef().)
1298	*/
1299	RTSpinlockAcquire(pDevExt->Spinlock, &SpinlockTmp);
1300
1301	pUsage = pDevExt->pUsageFree;
1302	if (pUsage)
1303	pDevExt->pUsageFree = pUsage->pNext;
1304	else
1305	{
1306	RTSpinlockRelease(pDevExt->Spinlock, &SpinlockTmp);
1307	pUsage = (PSUPDRVUSAGE)RTMemAlloc(sizeof(*pUsage));
1308	if (!pUsage)
1309	{
1310	RTMemFree(pObj);
1311	return NULL;
1312	}
1313	RTSpinlockAcquire(pDevExt->Spinlock, &SpinlockTmp);
1314	}
1315
1316	/*
1317	* Insert the object and create the session usage record.
1318	*/
1319	/* The object. */
1320	pObj->pNext = pDevExt->pObjs;
1321	pDevExt->pObjs = pObj;
1322
1323	/* The session record. */
1324	pUsage->cUsage = 1;
1325	pUsage->pObj = pObj;
1326	pUsage->pNext = pSession->pUsage;
1327	Log2(("SUPR0ObjRegister: pUsage=%p:{.pObj=%p, .pNext=%p}\n", pUsage, pUsage->pObj, pUsage->pNext));
1328	pSession->pUsage = pUsage;
1329
1330	RTSpinlockRelease(pDevExt->Spinlock, &SpinlockTmp);
1331
1332	Log(("SUPR0ObjRegister: returns %p (pvUser1=%p, pvUser=%p)\n", pObj, pvUser1, pvUser2));
1333	return pObj;
1334	}
1335
1336
1337	/**
1338	* Increment the reference counter for the object associating the reference
1339	* with the specified session.
1340	*
1341	* @returns IPRT status code.
1342	* @param pvObj The identifier returned by SUPR0ObjRegister().
1343	* @param pSession The session which is referencing the object.
1344	*/
1345	SUPR0DECL(int) SUPR0ObjAddRef(void *pvObj, PSUPDRVSESSION pSession)
1346	{
1347	RTSPINLOCKTMP SpinlockTmp = RTSPINLOCKTMP_INITIALIZER;
1348	PSUPDRVDEVEXT pDevExt = pSession->pDevExt;
1349	PSUPDRVOBJ pObj = (PSUPDRVOBJ)pvObj;
1350	PSUPDRVUSAGE pUsagePre;
1351	PSUPDRVUSAGE pUsage;
1352
1353	/*
1354	* Validate the input.
1355	*/
1356	AssertReturn(SUP_IS_SESSION_VALID(pSession), VERR_INVALID_PARAMETER);
1357	AssertMsgReturn(VALID_PTR(pObj) && pObj->u32Magic == SUPDRVOBJ_MAGIC,
1358	("Invalid pvObj=%p magic=%#x (exepcted %#x)\n", pvObj, pObj ? pObj->u32Magic : 0, SUPDRVOBJ_MAGIC),
1359	VERR_INVALID_PARAMETER);
1360
1361	/*
1362	* Preallocate the usage record.
1363	*/
1364	RTSpinlockAcquire(pDevExt->Spinlock, &SpinlockTmp);
1365
1366	pUsagePre = pDevExt->pUsageFree;
1367	if (pUsagePre)
1368	pDevExt->pUsageFree = pUsagePre->pNext;
1369	else
1370	{
1371	RTSpinlockRelease(pDevExt->Spinlock, &SpinlockTmp);
1372	pUsagePre = (PSUPDRVUSAGE)RTMemAlloc(sizeof(*pUsagePre));
1373	if (!pUsagePre)
1374	return VERR_NO_MEMORY;
1375	RTSpinlockAcquire(pDevExt->Spinlock, &SpinlockTmp);
1376	}
1377
1378	/*
1379	* Reference the object.
1380	*/
1381	pObj->cUsage++;
1382
1383	/*
1384	* Look for the session record.
1385	*/
1386	for (pUsage = pSession->pUsage; pUsage; pUsage = pUsage->pNext)
1387	{
1388	Log(("SUPR0AddRef: pUsage=%p:{.pObj=%p, .pNext=%p}\n", pUsage, pUsage->pObj, pUsage->pNext));
1389	if (pUsage->pObj == pObj)
1390	break;
1391	}
1392	if (pUsage)
1393	pUsage->cUsage++;
1394	else
1395	{
1396	/* create a new session record. */
1397	pUsagePre->cUsage = 1;
1398	pUsagePre->pObj = pObj;
1399	pUsagePre->pNext = pSession->pUsage;
1400	pSession->pUsage = pUsagePre;
1401	Log(("SUPR0AddRef: pUsagePre=%p:{.pObj=%p, .pNext=%p}\n", pUsagePre, pUsagePre->pObj, pUsagePre->pNext));
1402
1403	pUsagePre = NULL;
1404	}
1405
1406	/*
1407	* Put any unused usage record into the free list..
1408	*/
1409	if (pUsagePre)
1410	{
1411	pUsagePre->pNext = pDevExt->pUsageFree;
1412	pDevExt->pUsageFree = pUsagePre;
1413	}
1414
1415	RTSpinlockRelease(pDevExt->Spinlock, &SpinlockTmp);
1416
1417	return VINF_SUCCESS;
1418	}
1419
1420
1421	/**
1422	* Decrement / destroy a reference counter record for an object.
1423	*
1424	* The object is uniquely identified by pfnDestructor+pvUser1+pvUser2.
1425	*
1426	* @returns IPRT status code.
1427	* @param pvObj The identifier returned by SUPR0ObjRegister().
1428	* @param pSession The session which is referencing the object.
1429	*/
1430	SUPR0DECL(int) SUPR0ObjRelease(void *pvObj, PSUPDRVSESSION pSession)
1431	{
1432	RTSPINLOCKTMP SpinlockTmp = RTSPINLOCKTMP_INITIALIZER;
1433	PSUPDRVDEVEXT pDevExt = pSession->pDevExt;
1434	PSUPDRVOBJ pObj = (PSUPDRVOBJ)pvObj;
1435	bool fDestroy = false;
1436	PSUPDRVUSAGE pUsage;
1437	PSUPDRVUSAGE pUsagePrev;
1438
1439	/*
1440	* Validate the input.
1441	*/
1442	AssertReturn(SUP_IS_SESSION_VALID(pSession), VERR_INVALID_PARAMETER);
1443	AssertMsgReturn(VALID_PTR(pObj) && pObj->u32Magic == SUPDRVOBJ_MAGIC,
1444	("Invalid pvObj=%p magic=%#x (exepcted %#x)\n", pvObj, pObj ? pObj->u32Magic : 0, SUPDRVOBJ_MAGIC),
1445	VERR_INVALID_PARAMETER);
1446
1447	/*
1448	* Acquire the spinlock and look for the usage record.
1449	*/
1450	RTSpinlockAcquire(pDevExt->Spinlock, &SpinlockTmp);
1451
1452	for (pUsagePrev = NULL, pUsage = pSession->pUsage;
1453	pUsage;
1454	pUsagePrev = pUsage, pUsage = pUsage->pNext)
1455	{
1456	Log2(("SUPR0ObjRelease: pUsage=%p:{.pObj=%p, .pNext=%p}\n", pUsage, pUsage->pObj, pUsage->pNext));
1457	if (pUsage->pObj == pObj)
1458	{
1459	AssertMsg(pUsage->cUsage >= 1 && pObj->cUsage >= pUsage->cUsage, ("glob %d; sess %d\n", pObj->cUsage, pUsage->cUsage));
1460	if (pUsage->cUsage > 1)
1461	{
1462	pObj->cUsage--;
1463	pUsage->cUsage--;
1464	}
1465	else
1466	{
1467	/*
1468	* Free the session record.
1469	*/
1470	if (pUsagePrev)
1471	pUsagePrev->pNext = pUsage->pNext;
1472	else
1473	pSession->pUsage = pUsage->pNext;
1474	pUsage->pNext = pDevExt->pUsageFree;
1475	pDevExt->pUsageFree = pUsage;
1476
1477	/* What about the object? */
1478	if (pObj->cUsage > 1)
1479	pObj->cUsage--;
1480	else
1481	{
1482	/*
1483	* Object is to be destroyed, unlink it.
1484	*/
1485	pObj->u32Magic = SUPDRVOBJ_MAGIC + 1;
1486	fDestroy = true;
1487	if (pDevExt->pObjs == pObj)
1488	pDevExt->pObjs = pObj->pNext;
1489	else
1490	{
1491	PSUPDRVOBJ pObjPrev;
1492	for (pObjPrev = pDevExt->pObjs; pObjPrev; pObjPrev = pObjPrev->pNext)
1493	if (pObjPrev->pNext == pObj)
1494	{
1495	pObjPrev->pNext = pObj->pNext;
1496	break;
1497	}
1498	Assert(pObjPrev);
1499	}
1500	}
1501	}
1502	break;
1503	}
1504	}
1505
1506	RTSpinlockRelease(pDevExt->Spinlock, &SpinlockTmp);
1507
1508	/*
1509	* Call the destructor and free the object if required.
1510	*/
1511	if (fDestroy)
1512	{
1513	Log(("SUPR0ObjRelease: destroying %p/%d (%p/%p) cpid=%RTproc pid=%RTproc dtor=%p\n",
1514	pObj, pObj->enmType, pObj->pvUser1, pObj->pvUser2, pObj->CreatorProcess, RTProcSelf(), pObj->pfnDestructor));
1515	if (pObj->pfnDestructor)
1516	pObj->pfnDestructor(pObj, pObj->pvUser1, pObj->pvUser2);
1517	RTMemFree(pObj);
1518	}
1519
1520	AssertMsg(pUsage, ("pvObj=%p\n", pvObj));
1521	return pUsage ? VINF_SUCCESS : VERR_INVALID_PARAMETER;
1522	}
1523
1524	/**
1525	* Verifies that the current process can access the specified object.
1526	*
1527	* @returns The following IPRT status code:
1528	* @retval VINF_SUCCESS if access was granted.
1529	* @retval VERR_PERMISSION_DENIED if denied access.
1530	* @retval VERR_INVALID_PARAMETER if invalid parameter.
1531	*
1532	* @param pvObj The identifier returned by SUPR0ObjRegister().
1533	* @param pSession The session which wishes to access the object.
1534	* @param pszObjName Object string name. This is optional and depends on the object type.
1535	*
1536	* @remark The caller is responsible for making sure the object isn't removed while
1537	* we're inside this function. If uncertain about this, just call AddRef before calling us.
1538	*/
1539	SUPR0DECL(int) SUPR0ObjVerifyAccess(void pvObj, PSUPDRVSESSION pSession, const char pszObjName)
1540	{
1541	PSUPDRVOBJ pObj = (PSUPDRVOBJ)pvObj;
1542	int rc;
1543
1544	/*
1545	* Validate the input.
1546	*/
1547	AssertReturn(SUP_IS_SESSION_VALID(pSession), VERR_INVALID_PARAMETER);
1548	AssertMsgReturn(VALID_PTR(pObj) && pObj->u32Magic == SUPDRVOBJ_MAGIC,
1549	("Invalid pvObj=%p magic=%#x (exepcted %#x)\n", pvObj, pObj ? pObj->u32Magic : 0, SUPDRVOBJ_MAGIC),
1550	VERR_INVALID_PARAMETER);
1551
1552	/*
1553	* Check access. (returns true if a decision has been made.)
1554	*/
1555	rc = VERR_INTERNAL_ERROR;
1556	if (supdrvOSObjCanAccess(pObj, pSession, pszObjName, &rc))
1557	return rc;
1558
1559	/*
1560	* Default policy is to allow the user to access his own
1561	* stuff but nothing else.
1562	*/
1563	if (pObj->CreatorUid == pSession->Uid)
1564	return VINF_SUCCESS;
1565	return VERR_PERMISSION_DENIED;
1566	}
1567
1568
1569	/**
1570	* Lock pages.
1571	*
1572	* @returns IPRT status code.
1573	* @param pSession Session to which the locked memory should be associated.
1574	* @param pvR3 Start of the memory range to lock.
1575	* This must be page aligned.
1576	* @param cb Size of the memory range to lock.
1577	* This must be page aligned.
1578	*/
1579	SUPR0DECL(int) SUPR0LockMem(PSUPDRVSESSION pSession, RTR3PTR pvR3, uint32_t cPages, PRTHCPHYS paPages)
1580	{
1581	int rc;
1582	SUPDRVMEMREF Mem = {0};
1583	const size_t cb = (size_t)cPages << PAGE_SHIFT;
1584	LogFlow(("SUPR0LockMem: pSession=%p pvR3=%p cPages=%d paPages=%p\n", pSession, (void *)pvR3, cPages, paPages));
1585
1586	/*
1587	* Verify input.
1588	*/
1589	AssertReturn(SUP_IS_SESSION_VALID(pSession), VERR_INVALID_PARAMETER);
1590	AssertPtrReturn(paPages, VERR_INVALID_PARAMETER);
1591	if ( RT_ALIGN_R3PT(pvR3, PAGE_SIZE, RTR3PTR) != pvR3
1592	\|\| !pvR3)
1593	{
1594	Log(("pvR3 (%p) must be page aligned and not NULL!\n", (void *)pvR3));
1595	return VERR_INVALID_PARAMETER;
1596	}
1597
1598	#ifdef RT_OS_WINDOWS /* A temporary hack for windows, will be removed once all ring-3 code has been cleaned up. */
1599	/* First check if we allocated it using SUPPageAlloc; if so then we don't need to lock it again */
1600	rc = supdrvPageGetPhys(pSession, pvR3, cPages, paPages);
1601	if (RT_SUCCESS(rc))
1602	return rc;
1603	#endif
1604
1605	/*
1606	* Let IPRT do the job.
1607	*/
1608	Mem.eType = MEMREF_TYPE_LOCKED;
1609	rc = RTR0MemObjLockUser(&Mem.MemObj, pvR3, cb, RTR0ProcHandleSelf());
1610	if (RT_SUCCESS(rc))
1611	{
1612	uint32_t iPage = cPages;
1613	AssertMsg(RTR0MemObjAddressR3(Mem.MemObj) == pvR3, ("%p == %p\n", RTR0MemObjAddressR3(Mem.MemObj), pvR3));
1614	AssertMsg(RTR0MemObjSize(Mem.MemObj) == cb, ("%x == %x\n", RTR0MemObjSize(Mem.MemObj), cb));
1615
1616	while (iPage-- > 0)
1617	{
1618	paPages[iPage] = RTR0MemObjGetPagePhysAddr(Mem.MemObj, iPage);
1619	if (RT_UNLIKELY(paPages[iPage] == NIL_RTCCPHYS))
1620	{
1621	AssertMsgFailed(("iPage=%d\n", iPage));
1622	rc = VERR_INTERNAL_ERROR;
1623	break;
1624	}
1625	}
1626	if (RT_SUCCESS(rc))
1627	rc = supdrvMemAdd(&Mem, pSession);
1628	if (RT_FAILURE(rc))
1629	{
1630	int rc2 = RTR0MemObjFree(Mem.MemObj, false);
1631	AssertRC(rc2);
1632	}
1633	}
1634
1635	return rc;
1636	}
1637
1638
1639	/**
1640	* Unlocks the memory pointed to by pv.
1641	*
1642	* @returns IPRT status code.
1643	* @param pSession Session to which the memory was locked.
1644	* @param pvR3 Memory to unlock.
1645	*/
1646	SUPR0DECL(int) SUPR0UnlockMem(PSUPDRVSESSION pSession, RTR3PTR pvR3)
1647	{
1648	LogFlow(("SUPR0UnlockMem: pSession=%p pvR3=%p\n", pSession, (void *)pvR3));
1649	AssertReturn(SUP_IS_SESSION_VALID(pSession), VERR_INVALID_PARAMETER);
1650	#ifdef RT_OS_WINDOWS
1651	/*
1652	* Temporary hack for windows - SUPR0PageFree will unlock SUPR0PageAlloc
1653	* allocations; ignore this call.
1654	*/
1655	if (supdrvPageWasLockedByPageAlloc(pSession, pvR3))
1656	{
1657	LogFlow(("Page will be unlocked in SUPR0PageFree -> ignore\n"));
1658	return VINF_SUCCESS;
1659	}
1660	#endif
1661	return supdrvMemRelease(pSession, (RTHCUINTPTR)pvR3, MEMREF_TYPE_LOCKED);
1662	}
1663
1664
1665	/**
1666	* Allocates a chunk of page aligned memory with contiguous and fixed physical
1667	* backing.
1668	*
1669	* @returns IPRT status code.
1670	* @param pSession Session data.
1671	* @param cb Number of bytes to allocate.
1672	* @param ppvR0 Where to put the address of Ring-0 mapping the allocated memory.
1673	* @param ppvR3 Where to put the address of Ring-3 mapping the allocated memory.
1674	* @param pHCPhys Where to put the physical address of allocated memory.
1675	*/
1676	SUPR0DECL(int) SUPR0ContAlloc(PSUPDRVSESSION pSession, uint32_t cPages, PRTR0PTR ppvR0, PRTR3PTR ppvR3, PRTHCPHYS pHCPhys)
1677	{
1678	int rc;
1679	SUPDRVMEMREF Mem = {0};
1680	LogFlow(("SUPR0ContAlloc: pSession=%p cPages=%d ppvR0=%p ppvR3=%p pHCPhys=%p\n", pSession, cPages, ppvR0, ppvR3, pHCPhys));
1681
1682	/*
1683	* Validate input.
1684	*/
1685	AssertReturn(SUP_IS_SESSION_VALID(pSession), VERR_INVALID_PARAMETER);
1686	if (!ppvR3 \|\| !ppvR0 \|\| !pHCPhys)
1687	{
1688	Log(("Null pointer. All of these should be set: pSession=%p ppvR0=%p ppvR3=%p pHCPhys=%p\n",
1689	pSession, ppvR0, ppvR3, pHCPhys));
1690	return VERR_INVALID_PARAMETER;
1691
1692	}
1693	if (cPages < 1 \|\| cPages >= 256)
1694	{
1695	Log(("Illegal request cPages=%d, must be greater than 0 and smaller than 256\n", cPages));
1696	return VERR_INVALID_PARAMETER;
1697	}
1698
1699	/*
1700	* Let IPRT do the job.
1701	*/
1702	rc = RTR0MemObjAllocCont(&Mem.MemObj, cPages << PAGE_SHIFT, true /* executable R0 mapping */);
1703	if (RT_SUCCESS(rc))
1704	{
1705	int rc2;
1706	rc = RTR0MemObjMapUser(&Mem.MapObjR3, Mem.MemObj, (RTR3PTR)-1, 0,
1707	RTMEM_PROT_EXEC \| RTMEM_PROT_WRITE \| RTMEM_PROT_READ, RTR0ProcHandleSelf());
1708	if (RT_SUCCESS(rc))
1709	{
1710	Mem.eType = MEMREF_TYPE_CONT;
1711	rc = supdrvMemAdd(&Mem, pSession);
1712	if (!rc)
1713	{
1714	*ppvR0 = RTR0MemObjAddress(Mem.MemObj);
1715	*ppvR3 = RTR0MemObjAddressR3(Mem.MapObjR3);
1716	*pHCPhys = RTR0MemObjGetPagePhysAddr(Mem.MemObj, 0);
1717	return 0;
1718	}
1719
1720	rc2 = RTR0MemObjFree(Mem.MapObjR3, false);
1721	AssertRC(rc2);
1722	}
1723	rc2 = RTR0MemObjFree(Mem.MemObj, false);
1724	AssertRC(rc2);
1725	}
1726
1727	return rc;
1728	}
1729
1730
1731	/**
1732	* Frees memory allocated using SUPR0ContAlloc().
1733	*
1734	* @returns IPRT status code.
1735	* @param pSession The session to which the memory was allocated.
1736	* @param uPtr Pointer to the memory (ring-3 or ring-0).
1737	*/
1738	SUPR0DECL(int) SUPR0ContFree(PSUPDRVSESSION pSession, RTHCUINTPTR uPtr)
1739	{
1740	LogFlow(("SUPR0ContFree: pSession=%p uPtr=%p\n", pSession, (void *)uPtr));
1741	AssertReturn(SUP_IS_SESSION_VALID(pSession), VERR_INVALID_PARAMETER);
1742	return supdrvMemRelease(pSession, uPtr, MEMREF_TYPE_CONT);
1743	}
1744
1745
1746	/**
1747	* Allocates a chunk of page aligned memory with fixed physical backing below 4GB.
1748	*
1749	* The memory isn't zeroed.
1750	*
1751	* @returns IPRT status code.
1752	* @param pSession Session data.
1753	* @param cPages Number of pages to allocate.
1754	* @param ppvR0 Where to put the address of Ring-0 mapping of the allocated memory.
1755	* @param ppvR3 Where to put the address of Ring-3 mapping of the allocated memory.
1756	* @param paPages Where to put the physical addresses of allocated memory.
1757	*/
1758	SUPR0DECL(int) SUPR0LowAlloc(PSUPDRVSESSION pSession, uint32_t cPages, PRTR0PTR ppvR0, PRTR3PTR ppvR3, PRTHCPHYS paPages)
1759	{
1760	unsigned iPage;
1761	int rc;
1762	SUPDRVMEMREF Mem = {0};
1763	LogFlow(("SUPR0LowAlloc: pSession=%p cPages=%d ppvR3=%p ppvR0=%p paPages=%p\n", pSession, cPages, ppvR3, ppvR0, paPages));
1764
1765	/*
1766	* Validate input.
1767	*/
1768	AssertReturn(SUP_IS_SESSION_VALID(pSession), VERR_INVALID_PARAMETER);
1769	if (!ppvR3 \|\| !ppvR0 \|\| !paPages)
1770	{
1771	Log(("Null pointer. All of these should be set: pSession=%p ppvR3=%p ppvR0=%p paPages=%p\n",
1772	pSession, ppvR3, ppvR0, paPages));
1773	return VERR_INVALID_PARAMETER;
1774
1775	}
1776	if (cPages < 1 \|\| cPages > 256)
1777	{
1778	Log(("Illegal request cPages=%d, must be greater than 0 and smaller than 256.\n", cPages));
1779	return VERR_INVALID_PARAMETER;
1780	}
1781
1782	/*
1783	* Let IPRT do the work.
1784	*/
1785	rc = RTR0MemObjAllocLow(&Mem.MemObj, cPages << PAGE_SHIFT, true /* executable ring-0 mapping */);
1786	if (RT_SUCCESS(rc))
1787	{
1788	int rc2;
1789	rc = RTR0MemObjMapUser(&Mem.MapObjR3, Mem.MemObj, (RTR3PTR)-1, 0,
1790	RTMEM_PROT_EXEC \| RTMEM_PROT_WRITE \| RTMEM_PROT_READ, RTR0ProcHandleSelf());
1791	if (RT_SUCCESS(rc))
1792	{
1793	Mem.eType = MEMREF_TYPE_LOW;
1794	rc = supdrvMemAdd(&Mem, pSession);
1795	if (!rc)
1796	{
1797	for (iPage = 0; iPage < cPages; iPage++)
1798	{
1799	paPages[iPage] = RTR0MemObjGetPagePhysAddr(Mem.MemObj, iPage);
1800	AssertMsg(!(paPages[iPage] & (PAGE_SIZE - 1)), ("iPage=%d Phys=%VHp\n", paPages[iPage]));
1801	}
1802	*ppvR0 = RTR0MemObjAddress(Mem.MemObj);
1803	*ppvR3 = RTR0MemObjAddressR3(Mem.MapObjR3);
1804	return 0;
1805	}
1806
1807	rc2 = RTR0MemObjFree(Mem.MapObjR3, false);
1808	AssertRC(rc2);
1809	}
1810
1811	rc2 = RTR0MemObjFree(Mem.MemObj, false);
1812	AssertRC(rc2);
1813	}
1814
1815	return rc;
1816	}
1817
1818
1819	/**
1820	* Frees memory allocated using SUPR0LowAlloc().
1821	*
1822	* @returns IPRT status code.
1823	* @param pSession The session to which the memory was allocated.
1824	* @param uPtr Pointer to the memory (ring-3 or ring-0).
1825	*/
1826	SUPR0DECL(int) SUPR0LowFree(PSUPDRVSESSION pSession, RTHCUINTPTR uPtr)
1827	{
1828	LogFlow(("SUPR0LowFree: pSession=%p uPtr=%p\n", pSession, (void *)uPtr));
1829	AssertReturn(SUP_IS_SESSION_VALID(pSession), VERR_INVALID_PARAMETER);
1830	return supdrvMemRelease(pSession, uPtr, MEMREF_TYPE_LOW);
1831	}
1832
1833
1834
1835	/**
1836	* Allocates a chunk of memory with both R0 and R3 mappings.
1837	* The memory is fixed and it's possible to query the physical addresses using SUPR0MemGetPhys().
1838	*
1839	* @returns IPRT status code.
1840	* @param pSession The session to associated the allocation with.
1841	* @param cb Number of bytes to allocate.
1842	* @param ppvR0 Where to store the address of the Ring-0 mapping.
1843	* @param ppvR3 Where to store the address of the Ring-3 mapping.
1844	*/
1845	SUPR0DECL(int) SUPR0MemAlloc(PSUPDRVSESSION pSession, uint32_t cb, PRTR0PTR ppvR0, PRTR3PTR ppvR3)
1846	{
1847	int rc;
1848	SUPDRVMEMREF Mem = {0};
1849	LogFlow(("SUPR0MemAlloc: pSession=%p cb=%d ppvR0=%p ppvR3=%p\n", pSession, cb, ppvR0, ppvR3));
1850
1851	/*
1852	* Validate input.
1853	*/
1854	AssertReturn(SUP_IS_SESSION_VALID(pSession), VERR_INVALID_PARAMETER);
1855	AssertPtrReturn(ppvR0, VERR_INVALID_POINTER);
1856	AssertPtrReturn(ppvR3, VERR_INVALID_POINTER);
1857	if (cb < 1 \|\| cb >= _4M)
1858	{
1859	Log(("Illegal request cb=%u; must be greater than 0 and smaller than 4MB.\n", cb));
1860	return VERR_INVALID_PARAMETER;
1861	}
1862
1863	/*
1864	* Let IPRT do the work.
1865	*/
1866	rc = RTR0MemObjAllocPage(&Mem.MemObj, cb, true /* executable ring-0 mapping */);
1867	if (RT_SUCCESS(rc))
1868	{
1869	int rc2;
1870	rc = RTR0MemObjMapUser(&Mem.MapObjR3, Mem.MemObj, (RTR3PTR)-1, 0,
1871	RTMEM_PROT_EXEC \| RTMEM_PROT_WRITE \| RTMEM_PROT_READ, RTR0ProcHandleSelf());
1872	if (RT_SUCCESS(rc))
1873	{
1874	Mem.eType = MEMREF_TYPE_MEM;
1875	rc = supdrvMemAdd(&Mem, pSession);
1876	if (!rc)
1877	{
1878	*ppvR0 = RTR0MemObjAddress(Mem.MemObj);
1879	*ppvR3 = RTR0MemObjAddressR3(Mem.MapObjR3);
1880	return VINF_SUCCESS;
1881	}
1882	rc2 = RTR0MemObjFree(Mem.MapObjR3, false);
1883	AssertRC(rc2);
1884	}
1885
1886	rc2 = RTR0MemObjFree(Mem.MemObj, false);
1887	AssertRC(rc2);
1888	}
1889
1890	return rc;
1891	}
1892
1893
1894	/**
1895	* Get the physical addresses of memory allocated using SUPR0MemAlloc().
1896	*
1897	* @returns IPRT status code.
1898	* @param pSession The session to which the memory was allocated.
1899	* @param uPtr The Ring-0 or Ring-3 address returned by SUPR0MemAlloc().
1900	* @param paPages Where to store the physical addresses.
1901	*/
1902	SUPR0DECL(int) SUPR0MemGetPhys(PSUPDRVSESSION pSession, RTHCUINTPTR uPtr, PSUPPAGE paPages) /** @todo switch this bugger to RTHCPHYS */
1903	{
1904	PSUPDRVBUNDLE pBundle;
1905	RTSPINLOCKTMP SpinlockTmp = RTSPINLOCKTMP_INITIALIZER;
1906	LogFlow(("SUPR0MemGetPhys: pSession=%p uPtr=%p paPages=%p\n", pSession, (void *)uPtr, paPages));
1907
1908	/*
1909	* Validate input.
1910	*/
1911	AssertReturn(SUP_IS_SESSION_VALID(pSession), VERR_INVALID_PARAMETER);
1912	AssertPtrReturn(paPages, VERR_INVALID_POINTER);
1913	AssertReturn(uPtr, VERR_INVALID_PARAMETER);
1914
1915	/*
1916	* Search for the address.
1917	*/
1918	RTSpinlockAcquire(pSession->Spinlock, &SpinlockTmp);
1919	for (pBundle = &pSession->Bundle; pBundle; pBundle = pBundle->pNext)
1920	{
1921	if (pBundle->cUsed > 0)
1922	{
1923	unsigned i;
1924	for (i = 0; i < RT_ELEMENTS(pBundle->aMem); i++)
1925	{
1926	if ( pBundle->aMem[i].eType == MEMREF_TYPE_MEM
1927	&& pBundle->aMem[i].MemObj != NIL_RTR0MEMOBJ
1928	&& ( (RTHCUINTPTR)RTR0MemObjAddress(pBundle->aMem[i].MemObj) == uPtr
1929	\|\| ( pBundle->aMem[i].MapObjR3 != NIL_RTR0MEMOBJ
1930	&& RTR0MemObjAddressR3(pBundle->aMem[i].MapObjR3) == uPtr)
1931	)
1932	)
1933	{
1934	const unsigned cPages = RTR0MemObjSize(pBundle->aMem[i].MemObj) >> PAGE_SHIFT;
1935	unsigned iPage;
1936	for (iPage = 0; iPage < cPages; iPage++)
1937	{
1938	paPages[iPage].Phys = RTR0MemObjGetPagePhysAddr(pBundle->aMem[i].MemObj, iPage);
1939	paPages[iPage].uReserved = 0;
1940	}
1941	RTSpinlockRelease(pSession->Spinlock, &SpinlockTmp);
1942	return VINF_SUCCESS;
1943	}
1944	}
1945	}
1946	}
1947	RTSpinlockRelease(pSession->Spinlock, &SpinlockTmp);
1948	Log(("Failed to find %p!!!\n", (void *)uPtr));
1949	return VERR_INVALID_PARAMETER;
1950	}
1951
1952
1953	/**
1954	* Free memory allocated by SUPR0MemAlloc().
1955	*
1956	* @returns IPRT status code.
1957	* @param pSession The session owning the allocation.
1958	* @param uPtr The Ring-0 or Ring-3 address returned by SUPR0MemAlloc().
1959	*/
1960	SUPR0DECL(int) SUPR0MemFree(PSUPDRVSESSION pSession, RTHCUINTPTR uPtr)
1961	{
1962	LogFlow(("SUPR0MemFree: pSession=%p uPtr=%p\n", pSession, (void *)uPtr));
1963	AssertReturn(SUP_IS_SESSION_VALID(pSession), VERR_INVALID_PARAMETER);
1964	return supdrvMemRelease(pSession, uPtr, MEMREF_TYPE_MEM);
1965	}
1966
1967
1968	/**
1969	* Allocates a chunk of memory with only a R3 mappings.
1970	* The memory is fixed and it's possible to query the physical addresses using SUPR0MemGetPhys().
1971	*
1972	* @returns IPRT status code.
1973	* @param pSession The session to associated the allocation with.
1974	* @param cPages The number of pages to allocate.
1975	* @param ppvR3 Where to store the address of the Ring-3 mapping.
1976	* @param paPages Where to store the addresses of the pages. Optional.
1977	*/
1978	SUPR0DECL(int) SUPR0PageAlloc(PSUPDRVSESSION pSession, uint32_t cPages, PRTR3PTR ppvR3, PRTHCPHYS paPages)
1979	{
1980	int rc;
1981	SUPDRVMEMREF Mem = {0};
1982	LogFlow(("SUPR0PageAlloc: pSession=%p cb=%d ppvR3=%p\n", pSession, cPages, ppvR3));
1983
1984	/*
1985	* Validate input. The allowed allocation size must be at least equal to the maximum guest VRAM size.
1986	*/
1987	AssertReturn(SUP_IS_SESSION_VALID(pSession), VERR_INVALID_PARAMETER);
1988	AssertPtrReturn(ppvR3, VERR_INVALID_POINTER);
1989	if (cPages < 1 \|\| cPages > (128 * _1M)/PAGE_SIZE)
1990	{
1991	Log(("SUPR0PageAlloc: Illegal request cb=%u; must be greater than 0 and smaller than 128MB.\n", cPages));
1992	return VERR_INVALID_PARAMETER;
1993	}
1994
1995	/*
1996	* Let IPRT do the work.
1997	*/
1998	rc = RTR0MemObjAllocPhysNC(&Mem.MemObj, (size_t)cPages * PAGE_SIZE, NIL_RTHCPHYS);
1999	if (RT_SUCCESS(rc))
2000	{
2001	int rc2;
2002	rc = RTR0MemObjMapUser(&Mem.MapObjR3, Mem.MemObj, (RTR3PTR)-1, 0,
2003	RTMEM_PROT_EXEC \| RTMEM_PROT_WRITE \| RTMEM_PROT_READ, RTR0ProcHandleSelf());
2004	if (RT_SUCCESS(rc))
2005	{
2006	Mem.eType = MEMREF_TYPE_LOCKED_SUP;
2007	rc = supdrvMemAdd(&Mem, pSession);
2008	if (!rc)
2009	{
2010	*ppvR3 = RTR0MemObjAddressR3(Mem.MapObjR3);
2011	if (paPages)
2012	{
2013	uint32_t iPage = cPages;
2014	while (iPage-- > 0)
2015	{
2016	paPages[iPage] = RTR0MemObjGetPagePhysAddr(Mem.MapObjR3, iPage);
2017	Assert(paPages[iPage] != NIL_RTHCPHYS);
2018	}
2019	}
2020	return VINF_SUCCESS;
2021	}
2022	rc2 = RTR0MemObjFree(Mem.MapObjR3, false);
2023	AssertRC(rc2);
2024	}
2025
2026	rc2 = RTR0MemObjFree(Mem.MemObj, false);
2027	AssertRC(rc2);
2028	}
2029	return rc;
2030	}
2031
2032
2033	#ifdef RT_OS_WINDOWS
2034	/**
2035	* Check if the pages were locked by SUPR0PageAlloc
2036	*
2037	* This function will be removed along with the lock/unlock hacks when
2038	* we've cleaned up the ring-3 code properly.
2039	*
2040	* @returns boolean
2041	* @param pSession The session to which the memory was allocated.
2042	* @param pvR3 The Ring-3 address returned by SUPR0PageAlloc().
2043	*/
2044	static bool supdrvPageWasLockedByPageAlloc(PSUPDRVSESSION pSession, RTR3PTR pvR3)
2045	{
2046	PSUPDRVBUNDLE pBundle;
2047	RTSPINLOCKTMP SpinlockTmp = RTSPINLOCKTMP_INITIALIZER;
2048	LogFlow(("SUPR0PageIsLockedByPageAlloc: pSession=%p pvR3=%p\n", pSession, (void *)pvR3));
2049
2050	/*
2051	* Search for the address.
2052	*/
2053	RTSpinlockAcquire(pSession->Spinlock, &SpinlockTmp);
2054	for (pBundle = &pSession->Bundle; pBundle; pBundle = pBundle->pNext)
2055	{
2056	if (pBundle->cUsed > 0)
2057	{
2058	unsigned i;
2059	for (i = 0; i < RT_ELEMENTS(pBundle->aMem); i++)
2060	{
2061	if ( pBundle->aMem[i].eType == MEMREF_TYPE_LOCKED_SUP
2062	&& pBundle->aMem[i].MemObj != NIL_RTR0MEMOBJ
2063	&& pBundle->aMem[i].MapObjR3 != NIL_RTR0MEMOBJ
2064	&& RTR0MemObjAddressR3(pBundle->aMem[i].MapObjR3) == pvR3)
2065	{
2066	RTSpinlockRelease(pSession->Spinlock, &SpinlockTmp);
2067	return true;
2068	}
2069	}
2070	}
2071	}
2072	RTSpinlockRelease(pSession->Spinlock, &SpinlockTmp);
2073	return false;
2074	}
2075
2076
2077	/**
2078	* Get the physical addresses of memory allocated using SUPR0PageAlloc().
2079	*
2080	* This function will be removed along with the lock/unlock hacks when
2081	* we've cleaned up the ring-3 code properly.
2082	*
2083	* @returns IPRT status code.
2084	* @param pSession The session to which the memory was allocated.
2085	* @param pvR3 The Ring-3 address returned by SUPR0PageAlloc().
2086	* @param cPages Number of pages in paPages
2087	* @param paPages Where to store the physical addresses.
2088	*/
2089	static int supdrvPageGetPhys(PSUPDRVSESSION pSession, RTR3PTR pvR3, uint32_t cPages, PRTHCPHYS paPages)
2090	{
2091	PSUPDRVBUNDLE pBundle;
2092	RTSPINLOCKTMP SpinlockTmp = RTSPINLOCKTMP_INITIALIZER;
2093	LogFlow(("supdrvPageGetPhys: pSession=%p pvR3=%p cPages=%#lx paPages=%p\n", pSession, (void *)pvR3, (long)cPages, paPages));
2094
2095	/*
2096	* Search for the address.
2097	*/
2098	RTSpinlockAcquire(pSession->Spinlock, &SpinlockTmp);
2099	for (pBundle = &pSession->Bundle; pBundle; pBundle = pBundle->pNext)
2100	{
2101	if (pBundle->cUsed > 0)
2102	{
2103	unsigned i;
2104	for (i = 0; i < RT_ELEMENTS(pBundle->aMem); i++)
2105	{
2106	if ( pBundle->aMem[i].eType == MEMREF_TYPE_LOCKED_SUP
2107	&& pBundle->aMem[i].MemObj != NIL_RTR0MEMOBJ
2108	&& pBundle->aMem[i].MapObjR3 != NIL_RTR0MEMOBJ
2109	&& RTR0MemObjAddressR3(pBundle->aMem[i].MapObjR3) == pvR3)
2110	{
2111	uint32_t iPage = RTR0MemObjSize(pBundle->aMem[i].MemObj) >> PAGE_SHIFT;
2112	cPages = RT_MIN(iPage, cPages);
2113	for (iPage = 0; iPage < cPages; iPage++)
2114	paPages[iPage] = RTR0MemObjGetPagePhysAddr(pBundle->aMem[i].MemObj, iPage);
2115	RTSpinlockRelease(pSession->Spinlock, &SpinlockTmp);
2116	return VINF_SUCCESS;
2117	}
2118	}
2119	}
2120	}
2121	RTSpinlockRelease(pSession->Spinlock, &SpinlockTmp);
2122	return VERR_INVALID_PARAMETER;
2123	}
2124	#endif /* RT_OS_WINDOWS */
2125
2126
2127	/**
2128	* Free memory allocated by SUPR0PageAlloc().
2129	*
2130	* @returns IPRT status code.
2131	* @param pSession The session owning the allocation.
2132	* @param pvR3 The Ring-3 address returned by SUPR0PageAlloc().
2133	*/
2134	SUPR0DECL(int) SUPR0PageFree(PSUPDRVSESSION pSession, RTR3PTR pvR3)
2135	{
2136	LogFlow(("SUPR0PageFree: pSession=%p pvR3=%p\n", pSession, (void *)pvR3));
2137	AssertReturn(SUP_IS_SESSION_VALID(pSession), VERR_INVALID_PARAMETER);
2138	return supdrvMemRelease(pSession, (RTHCUINTPTR)pvR3, MEMREF_TYPE_LOCKED_SUP);
2139	}
2140
2141
2142	/**
2143	* Maps the GIP into userspace and/or get the physical address of the GIP.
2144	*
2145	* @returns IPRT status code.
2146	* @param pSession Session to which the GIP mapping should belong.
2147	* @param ppGipR3 Where to store the address of the ring-3 mapping. (optional)
2148	* @param pHCPhysGip Where to store the physical address. (optional)
2149	*
2150	* @remark There is no reference counting on the mapping, so one call to this function
2151	* count globally as one reference. One call to SUPR0GipUnmap() is will unmap GIP
2152	* and remove the session as a GIP user.
2153	*/
2154	SUPR0DECL(int) SUPR0GipMap(PSUPDRVSESSION pSession, PRTR3PTR ppGipR3, PRTHCPHYS pHCPhysGip)
2155	{
2156	int rc = 0;
2157	PSUPDRVDEVEXT pDevExt = pSession->pDevExt;
2158	RTR3PTR pGip = NIL_RTR3PTR;
2159	RTHCPHYS HCPhys = NIL_RTHCPHYS;
2160	LogFlow(("SUPR0GipMap: pSession=%p ppGipR3=%p pHCPhysGip=%p\n", pSession, ppGipR3, pHCPhysGip));
2161
2162	/*
2163	* Validate
2164	*/
2165	AssertReturn(SUP_IS_SESSION_VALID(pSession), VERR_INVALID_PARAMETER);
2166	AssertPtrNullReturn(ppGipR3, VERR_INVALID_POINTER);
2167	AssertPtrNullReturn(pHCPhysGip, VERR_INVALID_POINTER);
2168
2169	RTSemFastMutexRequest(pDevExt->mtxGip);
2170	if (pDevExt->pGip)
2171	{
2172	/*
2173	* Map it?
2174	*/
2175	if (ppGipR3)
2176	{
2177	if (pSession->GipMapObjR3 == NIL_RTR0MEMOBJ)
2178	rc = RTR0MemObjMapUser(&pSession->GipMapObjR3, pDevExt->GipMemObj, (RTR3PTR)-1, 0,
2179	RTMEM_PROT_READ, RTR0ProcHandleSelf());
2180	if (RT_SUCCESS(rc))
2181	{
2182	pGip = RTR0MemObjAddressR3(pSession->GipMapObjR3);
2183	rc = VINF_SUCCESS; /** @todo remove this and replace the !rc below with RT_SUCCESS(rc). */
2184	}
2185	}
2186
2187	/*
2188	* Get physical address.
2189	*/
2190	if (pHCPhysGip && !rc)
2191	HCPhys = pDevExt->HCPhysGip;
2192
2193	/*
2194	* Reference globally.
2195	*/
2196	if (!pSession->fGipReferenced && !rc)
2197	{
2198	pSession->fGipReferenced = 1;
2199	pDevExt->cGipUsers++;
2200	if (pDevExt->cGipUsers == 1)
2201	{
2202	PSUPGLOBALINFOPAGE pGip = pDevExt->pGip;
2203	unsigned i;
2204
2205	LogFlow(("SUPR0GipMap: Resumes GIP updating\n"));
2206
2207	for (i = 0; i < RT_ELEMENTS(pGip->aCPUs); i++)
2208	ASMAtomicXchgU32(&pGip->aCPUs[i].u32TransactionId, pGip->aCPUs[i].u32TransactionId & ~(GIP_UPDATEHZ_RECALC_FREQ * 2 - 1));
2209	ASMAtomicXchgU64(&pGip->u64NanoTSLastUpdateHz, 0);
2210
2211	rc = RTTimerStart(pDevExt->pGipTimer, 0);
2212	AssertRC(rc); rc = VINF_SUCCESS;
2213	}
2214	}
2215	}
2216	else
2217	{
2218	rc = SUPDRV_ERR_GENERAL_FAILURE;
2219	Log(("SUPR0GipMap: GIP is not available!\n"));
2220	}
2221	RTSemFastMutexRelease(pDevExt->mtxGip);
2222
2223	/*
2224	* Write returns.
2225	*/
2226	if (pHCPhysGip)
2227	*pHCPhysGip = HCPhys;
2228	if (ppGipR3)
2229	*ppGipR3 = pGip;
2230
2231	#ifdef DEBUG_DARWIN_GIP
2232	OSDBGPRINT(("SUPR0GipMap: returns %d pHCPhysGip=%lx ppGip=%p GipMapObjR3\n", rc, (unsigned long)HCPhys, pGip, pSession->GipMapObjR3));
2233	#else
2234	LogFlow(("SUPR0GipMap: returns %d pHCPhysGip=%lx ppGipR3=%p\n", rc, (unsigned long)HCPhys, (void *)(uintptr_t)pGip));
2235	#endif
2236	return rc;
2237	}
2238
2239
2240	/**
2241	* Unmaps any user mapping of the GIP and terminates all GIP access
2242	* from this session.
2243	*
2244	* @returns IPRT status code.
2245	* @param pSession Session to which the GIP mapping should belong.
2246	*/
2247	SUPR0DECL(int) SUPR0GipUnmap(PSUPDRVSESSION pSession)
2248	{
2249	int rc = VINF_SUCCESS;
2250	PSUPDRVDEVEXT pDevExt = pSession->pDevExt;
2251	#ifdef DEBUG_DARWIN_GIP
2252	OSDBGPRINT(("SUPR0GipUnmap: pSession=%p pGip=%p GipMapObjR3=%p\n",
2253	pSession,
2254	pSession->GipMapObjR3 != NIL_RTR0MEMOBJ ? RTR0MemObjAddress(pSession->GipMapObjR3) : NULL,
2255	pSession->GipMapObjR3));
2256	#else
2257	LogFlow(("SUPR0GipUnmap: pSession=%p\n", pSession));
2258	#endif
2259	AssertReturn(SUP_IS_SESSION_VALID(pSession), VERR_INVALID_PARAMETER);
2260
2261	RTSemFastMutexRequest(pDevExt->mtxGip);
2262
2263	/*
2264	* Unmap anything?
2265	*/
2266	if (pSession->GipMapObjR3 != NIL_RTR0MEMOBJ)
2267	{
2268	rc = RTR0MemObjFree(pSession->GipMapObjR3, false);
2269	AssertRC(rc);
2270	if (RT_SUCCESS(rc))
2271	pSession->GipMapObjR3 = NIL_RTR0MEMOBJ;
2272	}
2273
2274	/*
2275	* Dereference global GIP.
2276	*/
2277	if (pSession->fGipReferenced && !rc)
2278	{
2279	pSession->fGipReferenced = 0;
2280	if ( pDevExt->cGipUsers > 0
2281	&& !--pDevExt->cGipUsers)
2282	{
2283	LogFlow(("SUPR0GipUnmap: Suspends GIP updating\n"));
2284	rc = RTTimerStop(pDevExt->pGipTimer); AssertRC(rc); rc = 0;
2285	}
2286	}
2287
2288	RTSemFastMutexRelease(pDevExt->mtxGip);
2289
2290	return rc;
2291	}
2292
2293
2294	/**
2295	* Adds a memory object to the session.
2296	*
2297	* @returns IPRT status code.
2298	* @param pMem Memory tracking structure containing the
2299	* information to track.
2300	* @param pSession The session.
2301	*/
2302	static int supdrvMemAdd(PSUPDRVMEMREF pMem, PSUPDRVSESSION pSession)
2303	{
2304	PSUPDRVBUNDLE pBundle;
2305	RTSPINLOCKTMP SpinlockTmp = RTSPINLOCKTMP_INITIALIZER;
2306
2307	/*
2308	* Find free entry and record the allocation.
2309	*/
2310	RTSpinlockAcquire(pSession->Spinlock, &SpinlockTmp);
2311	for (pBundle = &pSession->Bundle; pBundle; pBundle = pBundle->pNext)
2312	{
2313	if (pBundle->cUsed < RT_ELEMENTS(pBundle->aMem))
2314	{
2315	unsigned i;
2316	for (i = 0; i < RT_ELEMENTS(pBundle->aMem); i++)
2317	{
2318	if (pBundle->aMem[i].MemObj == NIL_RTR0MEMOBJ)
2319	{
2320	pBundle->cUsed++;
2321	pBundle->aMem[i] = *pMem;
2322	RTSpinlockRelease(pSession->Spinlock, &SpinlockTmp);
2323	return VINF_SUCCESS;
2324	}
2325	}
2326	AssertFailed(); /* !!this can't be happening!!! */
2327	}
2328	}
2329	RTSpinlockRelease(pSession->Spinlock, &SpinlockTmp);
2330
2331	/*
2332	* Need to allocate a new bundle.
2333	* Insert into the last entry in the bundle.
2334	*/
2335	pBundle = (PSUPDRVBUNDLE)RTMemAllocZ(sizeof(*pBundle));
2336	if (!pBundle)
2337	return VERR_NO_MEMORY;
2338
2339	/* take last entry. */
2340	pBundle->cUsed++;
2341	pBundle->aMem[RT_ELEMENTS(pBundle->aMem) - 1] = *pMem;
2342
2343	/* insert into list. */
2344	RTSpinlockAcquire(pSession->Spinlock, &SpinlockTmp);
2345	pBundle->pNext = pSession->Bundle.pNext;
2346	pSession->Bundle.pNext = pBundle;
2347	RTSpinlockRelease(pSession->Spinlock, &SpinlockTmp);
2348
2349	return VINF_SUCCESS;
2350	}
2351
2352
2353	/**
2354	* Releases a memory object referenced by pointer and type.
2355	*
2356	* @returns IPRT status code.
2357	* @param pSession Session data.
2358	* @param uPtr Pointer to memory. This is matched against both the R0 and R3 addresses.
2359	* @param eType Memory type.
2360	*/
2361	static int supdrvMemRelease(PSUPDRVSESSION pSession, RTHCUINTPTR uPtr, SUPDRVMEMREFTYPE eType)
2362	{
2363	PSUPDRVBUNDLE pBundle;
2364	RTSPINLOCKTMP SpinlockTmp = RTSPINLOCKTMP_INITIALIZER;
2365
2366	/*
2367	* Validate input.
2368	*/
2369	if (!uPtr)
2370	{
2371	Log(("Illegal address %p\n", (void *)uPtr));
2372	return VERR_INVALID_PARAMETER;
2373	}
2374
2375	/*
2376	* Search for the address.
2377	*/
2378	RTSpinlockAcquire(pSession->Spinlock, &SpinlockTmp);
2379	for (pBundle = &pSession->Bundle; pBundle; pBundle = pBundle->pNext)
2380	{
2381	if (pBundle->cUsed > 0)
2382	{
2383	unsigned i;
2384	for (i = 0; i < RT_ELEMENTS(pBundle->aMem); i++)
2385	{
2386	if ( pBundle->aMem[i].eType == eType
2387	&& pBundle->aMem[i].MemObj != NIL_RTR0MEMOBJ
2388	&& ( (RTHCUINTPTR)RTR0MemObjAddress(pBundle->aMem[i].MemObj) == uPtr
2389	\|\| ( pBundle->aMem[i].MapObjR3 != NIL_RTR0MEMOBJ
2390	&& RTR0MemObjAddressR3(pBundle->aMem[i].MapObjR3) == uPtr))
2391	)
2392	{
2393	/* Make a copy of it and release it outside the spinlock. */
2394	SUPDRVMEMREF Mem = pBundle->aMem[i];
2395	pBundle->aMem[i].eType = MEMREF_TYPE_UNUSED;
2396	pBundle->aMem[i].MemObj = NIL_RTR0MEMOBJ;
2397	pBundle->aMem[i].MapObjR3 = NIL_RTR0MEMOBJ;
2398	RTSpinlockRelease(pSession->Spinlock, &SpinlockTmp);
2399
2400	if (Mem.MapObjR3)
2401	{
2402	int rc = RTR0MemObjFree(Mem.MapObjR3, false);
2403	AssertRC(rc); /** @todo figure out how to handle this. */
2404	}
2405	if (Mem.MemObj)
2406	{
2407	int rc = RTR0MemObjFree(Mem.MemObj, false);
2408	AssertRC(rc); /** @todo figure out how to handle this. */
2409	}
2410	return VINF_SUCCESS;
2411	}
2412	}
2413	}
2414	}
2415	RTSpinlockRelease(pSession->Spinlock, &SpinlockTmp);
2416	Log(("Failed to find %p!!! (eType=%d)\n", (void *)uPtr, eType));
2417	return VERR_INVALID_PARAMETER;
2418	}
2419
2420
2421	#ifdef VBOX_WITH_IDT_PATCHING
2422	/**
2423	* Install IDT for the current CPU.
2424	*
2425	* @returns One of the following IPRT status codes:
2426	* @retval VINF_SUCCESS on success.
2427	* @retval VERR_IDT_FAILED.
2428	* @retval VERR_NO_MEMORY.
2429	* @param pDevExt The device extension.
2430	* @param pSession The session data.
2431	* @param pReq The request.
2432	*/
2433	static int supdrvIOCtl_IdtInstall(PSUPDRVDEVEXT pDevExt, PSUPDRVSESSION pSession, PSUPIDTINSTALL pReq)
2434	{
2435	PSUPDRVPATCHUSAGE pUsagePre;
2436	PSUPDRVPATCH pPatchPre;
2437	RTIDTR Idtr;
2438	PSUPDRVPATCH pPatch;
2439	RTSPINLOCKTMP SpinlockTmp = RTSPINLOCKTMP_INITIALIZER;
2440	LogFlow(("supdrvIOCtl_IdtInstall\n"));
2441
2442	/*
2443	* Preallocate entry for this CPU cause we don't wanna do
2444	* that inside the spinlock!
2445	*/
2446	pUsagePre = (PSUPDRVPATCHUSAGE)RTMemAlloc(sizeof(*pUsagePre));
2447	if (!pUsagePre)
2448	return VERR_NO_MEMORY;
2449
2450	/*
2451	* Take the spinlock and see what we need to do.
2452	*/
2453	RTSpinlockAcquireNoInts(pDevExt->Spinlock, &SpinlockTmp);
2454
2455	/* check if we already got a free patch. */
2456	if (!pDevExt->pIdtPatchesFree)
2457	{
2458	/*
2459	* Allocate a patch - outside the spinlock of course.
2460	*/
2461	RTSpinlockReleaseNoInts(pDevExt->Spinlock, &SpinlockTmp);
2462
2463	pPatchPre = (PSUPDRVPATCH)RTMemExecAlloc(sizeof(*pPatchPre));
2464	if (!pPatchPre)
2465	return VERR_NO_MEMORY;
2466
2467	RTSpinlockAcquireNoInts(pDevExt->Spinlock, &SpinlockTmp);
2468	}
2469	else
2470	{
2471	pPatchPre = pDevExt->pIdtPatchesFree;
2472	pDevExt->pIdtPatchesFree = pPatchPre->pNext;
2473	}
2474
2475	/* look for matching patch entry */
2476	ASMGetIDTR(&Idtr);
2477	pPatch = pDevExt->pIdtPatches;
2478	while (pPatch && pPatch->pvIdt != (void *)Idtr.pIdt)
2479	pPatch = pPatch->pNext;
2480
2481	if (!pPatch)
2482	{
2483	/*
2484	* Create patch.
2485	*/
2486	pPatch = supdrvIdtPatchOne(pDevExt, pPatchPre);
2487	if (pPatch)
2488	pPatchPre = NULL; /* mark as used. */
2489	}
2490	else
2491	{
2492	/*
2493	* Simply increment patch usage.
2494	*/
2495	pPatch->cUsage++;
2496	}
2497
2498	if (pPatch)
2499	{
2500	/*
2501	* Increment and add if need be the session usage record for this patch.
2502	*/
2503	PSUPDRVPATCHUSAGE pUsage = pSession->pPatchUsage;
2504	while (pUsage && pUsage->pPatch != pPatch)
2505	pUsage = pUsage->pNext;
2506
2507	if (!pUsage)
2508	{
2509	/*
2510	* Add usage record.
2511	*/
2512	pUsagePre->cUsage = 1;
2513	pUsagePre->pPatch = pPatch;
2514	pUsagePre->pNext = pSession->pPatchUsage;
2515	pSession->pPatchUsage = pUsagePre;
2516	pUsagePre = NULL; /* mark as used. */
2517	}
2518	else
2519	{
2520	/*
2521	* Increment usage count.
2522	*/
2523	pUsage->cUsage++;
2524	}
2525	}
2526
2527	/* free patch - we accumulate them for paranoid saftly reasons. */
2528	if (pPatchPre)
2529	{
2530	pPatchPre->pNext = pDevExt->pIdtPatchesFree;
2531	pDevExt->pIdtPatchesFree = pPatchPre;
2532	}
2533
2534	RTSpinlockReleaseNoInts(pDevExt->Spinlock, &SpinlockTmp);
2535
2536	/*
2537	* Free unused preallocated buffers.
2538	*/
2539	if (pUsagePre)
2540	RTMemFree(pUsagePre);
2541
2542	pReq->u.Out.u8Idt = pDevExt->u8Idt;
2543
2544	return pPatch ? VINF_SUCCESS : VERR_IDT_FAILED;
2545	}
2546
2547
2548	/**
2549	* This creates a IDT patch entry.
2550	* If the first patch being installed it'll also determin the IDT entry
2551	* to use.
2552	*
2553	* @returns pPatch on success.
2554	* @returns NULL on failure.
2555	* @param pDevExt Pointer to globals.
2556	* @param pPatch Patch entry to use.
2557	* This will be linked into SUPDRVDEVEXT::pIdtPatches on
2558	* successful return.
2559	* @remark Call must be owning the SUPDRVDEVEXT::Spinlock!
2560	*/
2561	static PSUPDRVPATCH supdrvIdtPatchOne(PSUPDRVDEVEXT pDevExt, PSUPDRVPATCH pPatch)
2562	{
2563	RTIDTR Idtr;
2564	PSUPDRVIDTE paIdt;
2565	LogFlow(("supdrvIOCtl_IdtPatchOne: pPatch=%p\n", pPatch));
2566
2567	/*
2568	* Get IDT.
2569	*/
2570	ASMGetIDTR(&Idtr);
2571	paIdt = (PSUPDRVIDTE)Idtr.pIdt;
2572	/*
2573	* Recent Linux kernels can be configured to 1G user /3G kernel.
2574	*/
2575	if ((uintptr_t)paIdt < 0x40000000)
2576	{
2577	AssertMsgFailed(("bad paIdt=%p\n", paIdt));
2578	return NULL;
2579	}
2580
2581	if (!pDevExt->u8Idt)
2582	{
2583	/*
2584	* Test out the alternatives.
2585	*
2586	* At the moment we do not support chaining thus we ASSUME that one of
2587	* these 48 entries is unused (which is not a problem on Win32 and
2588	* Linux to my knowledge).
2589	*/
2590	/** @todo we MUST change this detection to try grab an entry which is NOT in use. This can be
2591	* combined with gathering info about which guest system call gates we can hook up directly. */
2592	unsigned i;
2593	uint8_t u8Idt = 0;
2594	static uint8_t au8Ints[] =
2595	{
2596	#ifdef RT_OS_WINDOWS /* We don't use 0xef and above because they are system stuff on linux (ef is IPI,
2597	* local apic timer, or some other frequently fireing thing). */
2598	0xef, 0xee, 0xed, 0xec,
2599	#endif
2600	0xeb, 0xea, 0xe9, 0xe8,
2601	0xdf, 0xde, 0xdd, 0xdc,
2602	0x7b, 0x7a, 0x79, 0x78,
2603	0xbf, 0xbe, 0xbd, 0xbc,
2604	};
2605	#if defined(RT_ARCH_AMD64) && defined(DEBUG)
2606	static int s_iWobble = 0;
2607	unsigned iMax = !(s_iWobble++ % 2) ? 0x80 : 0x100;
2608	Log2(("IDT: Idtr=%p:%#x\n", (void *)Idtr.pIdt, (unsigned)Idtr.cbIdt));
2609	for (i = iMax - 0x80; i*16+15 < Idtr.cbIdt && i < iMax; i++)
2610	{
2611	Log2(("%#x: %04x:%08x%04x%04x P=%d DPL=%d IST=%d Type1=%#x u32Reserved=%#x u5Reserved=%#x\n",
2612	i, paIdt[i].u16SegSel, paIdt[i].u32OffsetTop, paIdt[i].u16OffsetHigh, paIdt[i].u16OffsetLow,
2613	paIdt[i].u1Present, paIdt[i].u2DPL, paIdt[i].u3IST, paIdt[i].u5Type2,
2614	paIdt[i].u32Reserved, paIdt[i].u5Reserved));
2615	}
2616	#endif
2617	/* look for entries which are not present or otherwise unused. */
2618	for (i = 0; i < sizeof(au8Ints) / sizeof(au8Ints[0]); i++)
2619	{
2620	u8Idt = au8Ints[i];
2621	if ( u8Idt * sizeof(SUPDRVIDTE) < Idtr.cbIdt
2622	&& ( !paIdt[u8Idt].u1Present
2623	\|\| paIdt[u8Idt].u5Type2 == 0))
2624	break;
2625	u8Idt = 0;
2626	}
2627	if (!u8Idt)
2628	{
2629	/* try again, look for a compatible entry .*/
2630	for (i = 0; i < sizeof(au8Ints) / sizeof(au8Ints[0]); i++)
2631	{
2632	u8Idt = au8Ints[i];
2633	if ( u8Idt * sizeof(SUPDRVIDTE) < Idtr.cbIdt
2634	&& paIdt[u8Idt].u1Present
2635	&& paIdt[u8Idt].u5Type2 == SUPDRV_IDTE_TYPE2_INTERRUPT_GATE
2636	&& !(paIdt[u8Idt].u16SegSel & 3))
2637	break;
2638	u8Idt = 0;
2639	}
2640	if (!u8Idt)
2641	{
2642	Log(("Failed to find appropirate IDT entry!!\n"));
2643	return NULL;
2644	}
2645	}
2646	pDevExt->u8Idt = u8Idt;
2647	LogFlow(("supdrvIOCtl_IdtPatchOne: u8Idt=%x\n", u8Idt));
2648	}
2649
2650	/*
2651	* Prepare the patch
2652	*/
2653	memset(pPatch, 0, sizeof(*pPatch));
2654	pPatch->pvIdt = paIdt;
2655	pPatch->cUsage = 1;
2656	pPatch->pIdtEntry = &paIdt[pDevExt->u8Idt];
2657	pPatch->SavedIdt = paIdt[pDevExt->u8Idt];
2658	pPatch->ChangedIdt.u16OffsetLow = (uint32_t)((uintptr_t)&pPatch->auCode[0] & 0xffff);
2659	pPatch->ChangedIdt.u16OffsetHigh = (uint32_t)((uintptr_t)&pPatch->auCode[0] >> 16);
2660	#ifdef RT_ARCH_AMD64
2661	pPatch->ChangedIdt.u32OffsetTop = (uint32_t)((uintptr_t)&pPatch->auCode[0] >> 32);
2662	#endif
2663	pPatch->ChangedIdt.u16SegSel = ASMGetCS();
2664	#ifdef RT_ARCH_AMD64
2665	pPatch->ChangedIdt.u3IST = 0;
2666	pPatch->ChangedIdt.u5Reserved = 0;
2667	#else /* x86 */
2668	pPatch->ChangedIdt.u5Reserved = 0;
2669	pPatch->ChangedIdt.u3Type1 = 0;
2670	#endif /* x86 */
2671	pPatch->ChangedIdt.u5Type2 = SUPDRV_IDTE_TYPE2_INTERRUPT_GATE;
2672	pPatch->ChangedIdt.u2DPL = 3;
2673	pPatch->ChangedIdt.u1Present = 1;
2674
2675	/*
2676	* Generate the patch code.
2677	*/
2678	{
2679	#ifdef RT_ARCH_AMD64
2680	union
2681	{
2682	uint8_t *pb;
2683	uint32_t *pu32;
2684	uint64_t *pu64;
2685	} u, uFixJmp, uFixCall, uNotNested;
2686	u.pb = &pPatch->auCode[0];
2687
2688	/* check the cookie */
2689	*u.pb++ = 0x3d; // cmp eax, GLOBALCOOKIE
2690	*u.pu32++ = pDevExt->u32Cookie;
2691
2692	*u.pb++ = 0x74; // jz @VBoxCall
2693	*u.pb++ = 2;
2694
2695	/* jump to forwarder code. */
2696	*u.pb++ = 0xeb;
2697	uFixJmp = u;
2698	*u.pb++ = 0xfe;
2699
2700	// @VBoxCall:
2701	*u.pb++ = 0x0f; // swapgs
2702	*u.pb++ = 0x01;
2703	*u.pb++ = 0xf8;
2704
2705	/*
2706	* Call VMMR0Entry
2707	* We don't have to push the arguments here, but we have top
2708	* reserve some stack space for the interrupt forwarding.
2709	*/
2710	# ifdef RT_OS_WINDOWS
2711	*u.pb++ = 0x50; // push rax ; alignment filler.
2712	*u.pb++ = 0x41; // push r8 ; uArg
2713	*u.pb++ = 0x50;
2714	*u.pb++ = 0x52; // push rdx ; uOperation
2715	*u.pb++ = 0x51; // push rcx ; pVM
2716	# else
2717	*u.pb++ = 0x51; // push rcx ; alignment filler.
2718	*u.pb++ = 0x52; // push rdx ; uArg
2719	*u.pb++ = 0x56; // push rsi ; uOperation
2720	*u.pb++ = 0x57; // push rdi ; pVM
2721	# endif
2722
2723	*u.pb++ = 0xff; // call qword [pfnVMMR0EntryInt wrt rip]
2724	*u.pb++ = 0x15;
2725	uFixCall = u;
2726	*u.pu32++ = 0;
2727
2728	*u.pb++ = 0x48; // add rsp, 20h ; remove call frame.
2729	*u.pb++ = 0x81;
2730	*u.pb++ = 0xc4;
2731	*u.pu32++ = 0x20;
2732
2733	*u.pb++ = 0x0f; // swapgs
2734	*u.pb++ = 0x01;
2735	*u.pb++ = 0xf8;
2736
2737	/* Return to R3. */
2738	uNotNested = u;
2739	*u.pb++ = 0x48; // iretq
2740	*u.pb++ = 0xcf;
2741
2742	while ((uintptr_t)u.pb & 0x7) // align 8
2743	*u.pb++ = 0xcc;
2744
2745	/* Pointer to the VMMR0Entry. */ // pfnVMMR0EntryInt dq StubVMMR0Entry
2746	*uFixCall.pu32 = (uint32_t)(u.pb - uFixCall.pb - 4); uFixCall.pb = NULL;
2747	pPatch->offVMMR0EntryFixup = (uint16_t)(u.pb - &pPatch->auCode[0]);
2748	*u.pu64++ = pDevExt->pvVMMR0 ? (uint64_t)pDevExt->pfnVMMR0EntryInt : (uint64_t)u.pb + 8;
2749
2750	/* stub entry. */ // StubVMMR0Entry:
2751	pPatch->offStub = (uint16_t)(u.pb - &pPatch->auCode[0]);
2752	*u.pb++ = 0x33; // xor eax, eax
2753	*u.pb++ = 0xc0;
2754
2755	*u.pb++ = 0x48; // dec rax
2756	*u.pb++ = 0xff;
2757	*u.pb++ = 0xc8;
2758
2759	*u.pb++ = 0xc3; // ret
2760
2761	/* forward to the original handler using a retf. */
2762	*uFixJmp.pb = (uint8_t)(u.pb - uFixJmp.pb - 1); uFixJmp.pb = NULL;
2763
2764	*u.pb++ = 0x68; // push <target cs>
2765	*u.pu32++ = !pPatch->SavedIdt.u5Type2 ? ASMGetCS() : pPatch->SavedIdt.u16SegSel;
2766
2767	*u.pb++ = 0x68; // push <low target rip>
2768	*u.pu32++ = !pPatch->SavedIdt.u5Type2
2769	? (uint32_t)(uintptr_t)uNotNested.pb
2770	: (uint32_t)pPatch->SavedIdt.u16OffsetLow
2771	\| (uint32_t)pPatch->SavedIdt.u16OffsetHigh << 16;
2772
2773	*u.pb++ = 0xc7; // mov dword [rsp + 4], <high target rip>
2774	*u.pb++ = 0x44;
2775	*u.pb++ = 0x24;
2776	*u.pb++ = 0x04;
2777	*u.pu32++ = !pPatch->SavedIdt.u5Type2
2778	? (uint32_t)((uint64_t)uNotNested.pb >> 32)
2779	: pPatch->SavedIdt.u32OffsetTop;
2780
2781	*u.pb++ = 0x48; // retf ; does this require prefix?
2782	*u.pb++ = 0xcb;
2783
2784	#else /* RT_ARCH_X86 */
2785
2786	union
2787	{
2788	uint8_t *pb;
2789	uint16_t *pu16;
2790	uint32_t *pu32;
2791	} u, uFixJmpNotNested, uFixJmp, uFixCall, uNotNested;
2792	u.pb = &pPatch->auCode[0];
2793
2794	/* check the cookie */
2795	*u.pb++ = 0x81; // cmp esi, GLOBALCOOKIE
2796	*u.pb++ = 0xfe;
2797	*u.pu32++ = pDevExt->u32Cookie;
2798
2799	*u.pb++ = 0x74; // jz VBoxCall
2800	uFixJmp = u;
2801	*u.pb++ = 0;
2802
2803	/* jump (far) to the original handler / not-nested-stub. */
2804	*u.pb++ = 0xea; // jmp far NotNested
2805	uFixJmpNotNested = u;
2806	*u.pu32++ = 0;
2807	*u.pu16++ = 0;
2808
2809	/* save selector registers. */ // VBoxCall:
2810	*uFixJmp.pb = (uint8_t)(u.pb - uFixJmp.pb - 1);
2811	*u.pb++ = 0x0f; // push fs
2812	*u.pb++ = 0xa0;
2813
2814	*u.pb++ = 0x1e; // push ds
2815
2816	*u.pb++ = 0x06; // push es
2817
2818	/* call frame */
2819	*u.pb++ = 0x51; // push ecx
2820
2821	*u.pb++ = 0x52; // push edx
2822
2823	*u.pb++ = 0x50; // push eax
2824
2825	/* load ds, es and perhaps fs before call. */
2826	*u.pb++ = 0xb8; // mov eax, KernelDS
2827	*u.pu32++ = ASMGetDS();
2828
2829	*u.pb++ = 0x8e; // mov ds, eax
2830	*u.pb++ = 0xd8;
2831
2832	*u.pb++ = 0x8e; // mov es, eax
2833	*u.pb++ = 0xc0;
2834
2835	#ifdef RT_OS_WINDOWS
2836	*u.pb++ = 0xb8; // mov eax, KernelFS
2837	*u.pu32++ = ASMGetFS();
2838
2839	*u.pb++ = 0x8e; // mov fs, eax
2840	*u.pb++ = 0xe0;
2841	#endif
2842
2843	/* do the call. */
2844	*u.pb++ = 0xe8; // call _VMMR0Entry / StubVMMR0Entry
2845	uFixCall = u;
2846	pPatch->offVMMR0EntryFixup = (uint16_t)(u.pb - &pPatch->auCode[0]);
2847	*u.pu32++ = 0xfffffffb;
2848
2849	*u.pb++ = 0x83; // add esp, 0ch ; cdecl
2850	*u.pb++ = 0xc4;
2851	*u.pb++ = 0x0c;
2852
2853	/* restore selector registers. */
2854	*u.pb++ = 0x07; // pop es
2855	//
2856	*u.pb++ = 0x1f; // pop ds
2857
2858	*u.pb++ = 0x0f; // pop fs
2859	*u.pb++ = 0xa1;
2860
2861	uNotNested = u; // NotNested:
2862	*u.pb++ = 0xcf; // iretd
2863
2864	/* the stub VMMR0Entry. */ // StubVMMR0Entry:
2865	pPatch->offStub = (uint16_t)(u.pb - &pPatch->auCode[0]);
2866	*u.pb++ = 0x33; // xor eax, eax
2867	*u.pb++ = 0xc0;
2868
2869	*u.pb++ = 0x48; // dec eax
2870
2871	*u.pb++ = 0xc3; // ret
2872
2873	/* Fixup the VMMR0Entry call. */
2874	if (pDevExt->pvVMMR0)
2875	*uFixCall.pu32 = (uint32_t)pDevExt->pfnVMMR0EntryInt - (uint32_t)(uFixCall.pu32 + 1);
2876	else
2877	*uFixCall.pu32 = (uint32_t)&pPatch->auCode[pPatch->offStub] - (uint32_t)(uFixCall.pu32 + 1);
2878
2879	/* Fixup the forward / nested far jump. */
2880	if (!pPatch->SavedIdt.u5Type2)
2881	{
2882	*uFixJmpNotNested.pu32++ = (uint32_t)uNotNested.pb;
2883	*uFixJmpNotNested.pu16++ = ASMGetCS();
2884	}
2885	else
2886	{
2887	*uFixJmpNotNested.pu32++ = ((uint32_t)pPatch->SavedIdt.u16OffsetHigh << 16) \| pPatch->SavedIdt.u16OffsetLow;
2888	*uFixJmpNotNested.pu16++ = pPatch->SavedIdt.u16SegSel;
2889	}
2890	#endif /* RT_ARCH_X86 */
2891	Assert(u.pb <= &pPatch->auCode[sizeof(pPatch->auCode)]);
2892	#if 0
2893	/* dump the patch code */
2894	Log2(("patch code: %p\n", &pPatch->auCode[0]));
2895	for (uFixCall.pb = &pPatch->auCode[0]; uFixCall.pb < u.pb; uFixCall.pb++)
2896	Log2(("0x%02x,\n", *uFixCall.pb));
2897	#endif
2898	}
2899
2900	/*
2901	* Install the patch.
2902	*/
2903	supdrvIdtWrite(pPatch->pIdtEntry, &pPatch->ChangedIdt);
2904	AssertMsg(!memcmp((void *)pPatch->pIdtEntry, &pPatch->ChangedIdt, sizeof(pPatch->ChangedIdt)), ("The stupid change code didn't work!!!!!\n"));
2905
2906	/*
2907	* Link in the patch.
2908	*/
2909	pPatch->pNext = pDevExt->pIdtPatches;
2910	pDevExt->pIdtPatches = pPatch;
2911
2912	return pPatch;
2913	}
2914
2915
2916	/**
2917	* Removes the sessions IDT references.
2918	* This will uninstall our IDT patch if we left unreferenced.
2919	*
2920	* @returns VINF_SUCCESS.
2921	* @param pDevExt Device globals.
2922	* @param pSession Session data.
2923	*/
2924	static int supdrvIOCtl_IdtRemoveAll(PSUPDRVDEVEXT pDevExt, PSUPDRVSESSION pSession)
2925	{
2926	PSUPDRVPATCHUSAGE pUsage;
2927	RTSPINLOCKTMP SpinlockTmp = RTSPINLOCKTMP_INITIALIZER;
2928	LogFlow(("supdrvIOCtl_IdtRemoveAll: pSession=%p\n", pSession));
2929
2930	/*
2931	* Take the spinlock.
2932	*/
2933	RTSpinlockAcquireNoInts(pDevExt->Spinlock, &SpinlockTmp);
2934
2935	/*
2936	* Walk usage list, removing patches as their usage count reaches zero.
2937	*/
2938	pUsage = pSession->pPatchUsage;
2939	while (pUsage)
2940	{
2941	if (pUsage->pPatch->cUsage <= pUsage->cUsage)
2942	supdrvIdtRemoveOne(pDevExt, pUsage->pPatch);
2943	else
2944	pUsage->pPatch->cUsage -= pUsage->cUsage;
2945
2946	/* next */
2947	pUsage = pUsage->pNext;
2948	}
2949
2950	/*
2951	* Empty the usage chain and we're done inside the spinlock.
2952	*/
2953	pUsage = pSession->pPatchUsage;
2954	pSession->pPatchUsage = NULL;
2955
2956	RTSpinlockReleaseNoInts(pDevExt->Spinlock, &SpinlockTmp);
2957
2958	/*
2959	* Free usage entries.
2960	*/
2961	while (pUsage)
2962	{
2963	void *pvToFree = pUsage;
2964	pUsage->cUsage = 0;
2965	pUsage->pPatch = NULL;
2966	pUsage = pUsage->pNext;
2967	RTMemFree(pvToFree);
2968	}
2969
2970	return VINF_SUCCESS;
2971	}
2972
2973
2974	/**
2975	* Remove one patch.
2976	*
2977	* Worker for supdrvIOCtl_IdtRemoveAll.
2978	*
2979	* @param pDevExt Device globals.
2980	* @param pPatch Patch entry to remove.
2981	* @remark Caller must own SUPDRVDEVEXT::Spinlock!
2982	*/
2983	static void supdrvIdtRemoveOne(PSUPDRVDEVEXT pDevExt, PSUPDRVPATCH pPatch)
2984	{
2985	LogFlow(("supdrvIdtRemoveOne: pPatch=%p\n", pPatch));
2986
2987	pPatch->cUsage = 0;
2988
2989	/*
2990	* If the IDT entry was changed it have to kick around for ever!
2991	* This will be attempted freed again, perhaps next time we'll succeed :-)
2992	*/
2993	if (memcmp((void *)pPatch->pIdtEntry, &pPatch->ChangedIdt, sizeof(pPatch->ChangedIdt)))
2994	{
2995	AssertMsgFailed(("The hijacked IDT entry has CHANGED!!!\n"));
2996	return;
2997	}
2998
2999	/*
3000	* Unlink it.
3001	*/
3002	if (pDevExt->pIdtPatches != pPatch)
3003	{
3004	PSUPDRVPATCH pPatchPrev = pDevExt->pIdtPatches;
3005	while (pPatchPrev)
3006	{
3007	if (pPatchPrev->pNext == pPatch)
3008	{
3009	pPatchPrev->pNext = pPatch->pNext;
3010	break;
3011	}
3012	pPatchPrev = pPatchPrev->pNext;
3013	}
3014	Assert(!pPatchPrev);
3015	}
3016	else
3017	pDevExt->pIdtPatches = pPatch->pNext;
3018	pPatch->pNext = NULL;
3019
3020
3021	/*
3022	* Verify and restore the IDT.
3023	*/
3024	AssertMsg(!memcmp((void *)pPatch->pIdtEntry, &pPatch->ChangedIdt, sizeof(pPatch->ChangedIdt)), ("The hijacked IDT entry has CHANGED!!!\n"));
3025	supdrvIdtWrite(pPatch->pIdtEntry, &pPatch->SavedIdt);
3026	AssertMsg(!memcmp((void *)pPatch->pIdtEntry, &pPatch->SavedIdt, sizeof(pPatch->SavedIdt)), ("The hijacked IDT entry has CHANGED!!!\n"));
3027
3028	/*
3029	* Put it in the free list.
3030	* (This free list stuff is to calm my paranoia.)
3031	*/
3032	pPatch->pvIdt = NULL;
3033	pPatch->pIdtEntry = NULL;
3034
3035	pPatch->pNext = pDevExt->pIdtPatchesFree;
3036	pDevExt->pIdtPatchesFree = pPatch;
3037	}
3038
3039
3040	/**
3041	* Write to an IDT entry.
3042	*
3043	* @param pvIdtEntry Where to write.
3044	* @param pNewIDTEntry What to write.
3045	*/
3046	static void supdrvIdtWrite(volatile void pvIdtEntry, const SUPDRVIDTE pNewIDTEntry)
3047	{
3048	RTR0UINTREG uCR0;
3049	RTR0UINTREG uFlags;
3050
3051	/*
3052	* On SMP machines (P4 hyperthreading included) we must preform a
3053	* 64-bit locked write when updating the IDT entry.
3054	*
3055	* The F00F bugfix for linux (and probably other OSes) causes
3056	* the IDT to be pointing to an readonly mapping. We get around that
3057	* by temporarily turning of WP. Since we're inside a spinlock at this
3058	* point, interrupts are disabled and there isn't any way the WP bit
3059	* flipping can cause any trouble.
3060	*/
3061
3062	/* Save & Clear interrupt flag; Save & clear WP. */
3063	uFlags = ASMGetFlags();
3064	ASMSetFlags(uFlags & ~(RTR0UINTREG)(1 << 9)); /X86_EFL_IF/
3065	Assert(!(ASMGetFlags() & (1 << 9)));
3066	uCR0 = ASMGetCR0();
3067	ASMSetCR0(uCR0 & ~(RTR0UINTREG)(1 << 16)); /X86_CR0_WP/
3068
3069	/* Update IDT Entry */
3070	#ifdef RT_ARCH_AMD64
3071	ASMAtomicXchgU128((volatile uint128_t )pvIdtEntry, (uint128_t *)(uintptr_t)pNewIDTEntry);
3072	#else
3073	ASMAtomicXchgU64((volatile uint64_t )pvIdtEntry, (uint64_t *)(uintptr_t)pNewIDTEntry);
3074	#endif
3075
3076	/* Restore CR0 & Flags */
3077	ASMSetCR0(uCR0);
3078	ASMSetFlags(uFlags);
3079	}
3080	#endif /* VBOX_WITH_IDT_PATCHING */
3081
3082
3083	/**
3084	* Opens an image. If it's the first time it's opened the call must upload
3085	* the bits using the supdrvIOCtl_LdrLoad() / SUPDRV_IOCTL_LDR_LOAD function.
3086	*
3087	* This is the 1st step of the loading.
3088	*
3089	* @returns IPRT status code.
3090	* @param pDevExt Device globals.
3091	* @param pSession Session data.
3092	* @param pReq The open request.
3093	*/
3094	static int supdrvIOCtl_LdrOpen(PSUPDRVDEVEXT pDevExt, PSUPDRVSESSION pSession, PSUPLDROPEN pReq)
3095	{
3096	PSUPDRVLDRIMAGE pImage;
3097	unsigned cb;
3098	void *pv;
3099	LogFlow(("supdrvIOCtl_LdrOpen: szName=%s cbImage=%d\n", pReq->u.In.szName, pReq->u.In.cbImage));
3100
3101	/*
3102	* Check if we got an instance of the image already.
3103	*/
3104	RTSemFastMutexRequest(pDevExt->mtxLdr);
3105	for (pImage = pDevExt->pLdrImages; pImage; pImage = pImage->pNext)
3106	{
3107	if (!strcmp(pImage->szName, pReq->u.In.szName))
3108	{
3109	pImage->cUsage++;
3110	pReq->u.Out.pvImageBase = pImage->pvImage;
3111	pReq->u.Out.fNeedsLoading = pImage->uState == SUP_IOCTL_LDR_OPEN;
3112	supdrvLdrAddUsage(pSession, pImage);
3113	RTSemFastMutexRelease(pDevExt->mtxLdr);
3114	return VINF_SUCCESS;
3115	}
3116	}
3117	/* (not found - add it!) */
3118
3119	/*
3120	* Allocate memory.
3121	*/
3122	cb = pReq->u.In.cbImage + sizeof(SUPDRVLDRIMAGE) + 31;
3123	pv = RTMemExecAlloc(cb);
3124	if (!pv)
3125	{
3126	RTSemFastMutexRelease(pDevExt->mtxLdr);
3127	Log(("supdrvIOCtl_LdrOpen: RTMemExecAlloc(%u) failed\n", cb));
3128	return VERR_NO_MEMORY;
3129	}
3130
3131	/*
3132	* Setup and link in the LDR stuff.
3133	*/
3134	pImage = (PSUPDRVLDRIMAGE)pv;
3135	pImage->pvImage = RT_ALIGN_P(pImage + 1, 32);
3136	pImage->cbImage = pReq->u.In.cbImage;
3137	pImage->pfnModuleInit = NULL;
3138	pImage->pfnModuleTerm = NULL;
3139	pImage->uState = SUP_IOCTL_LDR_OPEN;
3140	pImage->cUsage = 1;
3141	strcpy(pImage->szName, pReq->u.In.szName);
3142
3143	pImage->pNext = pDevExt->pLdrImages;
3144	pDevExt->pLdrImages = pImage;
3145
3146	supdrvLdrAddUsage(pSession, pImage);
3147
3148	pReq->u.Out.pvImageBase = pImage->pvImage;
3149	pReq->u.Out.fNeedsLoading = true;
3150	RTSemFastMutexRelease(pDevExt->mtxLdr);
3151	return VINF_SUCCESS;
3152	}
3153
3154
3155	/**
3156	* Loads the image bits.
3157	*
3158	* This is the 2nd step of the loading.
3159	*
3160	* @returns IPRT status code.
3161	* @param pDevExt Device globals.
3162	* @param pSession Session data.
3163	* @param pReq The request.
3164	*/
3165	static int supdrvIOCtl_LdrLoad(PSUPDRVDEVEXT pDevExt, PSUPDRVSESSION pSession, PSUPLDRLOAD pReq)
3166	{
3167	PSUPDRVLDRUSAGE pUsage;
3168	PSUPDRVLDRIMAGE pImage;
3169	int rc;
3170	LogFlow(("supdrvIOCtl_LdrLoad: pvImageBase=%p cbImage=%d\n", pReq->u.In.pvImageBase, pReq->u.In.cbImage));
3171
3172	/*
3173	* Find the ldr image.
3174	*/
3175	RTSemFastMutexRequest(pDevExt->mtxLdr);
3176	pUsage = pSession->pLdrUsage;
3177	while (pUsage && pUsage->pImage->pvImage != pReq->u.In.pvImageBase)
3178	pUsage = pUsage->pNext;
3179	if (!pUsage)
3180	{
3181	RTSemFastMutexRelease(pDevExt->mtxLdr);
3182	Log(("SUP_IOCTL_LDR_LOAD: couldn't find image!\n"));
3183	return VERR_INVALID_HANDLE;
3184	}
3185	pImage = pUsage->pImage;
3186	if (pImage->cbImage != pReq->u.In.cbImage)
3187	{
3188	RTSemFastMutexRelease(pDevExt->mtxLdr);
3189	Log(("SUP_IOCTL_LDR_LOAD: image size mismatch!! %d(prep) != %d(load)\n", pImage->cbImage, pReq->u.In.cbImage));
3190	return VERR_INVALID_HANDLE;
3191	}
3192	if (pImage->uState != SUP_IOCTL_LDR_OPEN)
3193	{
3194	unsigned uState = pImage->uState;
3195	RTSemFastMutexRelease(pDevExt->mtxLdr);
3196	if (uState != SUP_IOCTL_LDR_LOAD)
3197	AssertMsgFailed(("SUP_IOCTL_LDR_LOAD: invalid image state %d (%#x)!\n", uState, uState));
3198	return SUPDRV_ERR_ALREADY_LOADED;
3199	}
3200	switch (pReq->u.In.eEPType)
3201	{
3202	case SUPLDRLOADEP_NOTHING:
3203	break;
3204	case SUPLDRLOADEP_VMMR0:
3205	if ( !pReq->u.In.EP.VMMR0.pvVMMR0
3206	\|\| !pReq->u.In.EP.VMMR0.pvVMMR0EntryInt
3207	\|\| !pReq->u.In.EP.VMMR0.pvVMMR0EntryFast
3208	\|\| !pReq->u.In.EP.VMMR0.pvVMMR0EntryEx)
3209	{
3210	RTSemFastMutexRelease(pDevExt->mtxLdr);
3211	Log(("NULL pointer: pvVMMR0=%p pvVMMR0EntryInt=%p pvVMMR0EntryFast=%p pvVMMR0EntryEx=%p!\n",
3212	pReq->u.In.EP.VMMR0.pvVMMR0, pReq->u.In.EP.VMMR0.pvVMMR0EntryInt,
3213	pReq->u.In.EP.VMMR0.pvVMMR0EntryFast, pReq->u.In.EP.VMMR0.pvVMMR0EntryEx));
3214	return VERR_INVALID_PARAMETER;
3215	}
3216	if ( (uintptr_t)pReq->u.In.EP.VMMR0.pvVMMR0EntryInt - (uintptr_t)pImage->pvImage >= pReq->u.In.cbImage
3217	\|\| (uintptr_t)pReq->u.In.EP.VMMR0.pvVMMR0EntryFast - (uintptr_t)pImage->pvImage >= pReq->u.In.cbImage
3218	\|\| (uintptr_t)pReq->u.In.EP.VMMR0.pvVMMR0EntryEx - (uintptr_t)pImage->pvImage >= pReq->u.In.cbImage)
3219	{
3220	RTSemFastMutexRelease(pDevExt->mtxLdr);
3221	Log(("Out of range (%p LB %#x): pvVMMR0EntryInt=%p, pvVMMR0EntryFast=%p or pvVMMR0EntryEx=%p is NULL!\n",
3222	pImage->pvImage, pReq->u.In.cbImage, pReq->u.In.EP.VMMR0.pvVMMR0EntryInt,
3223	pReq->u.In.EP.VMMR0.pvVMMR0EntryFast, pReq->u.In.EP.VMMR0.pvVMMR0EntryEx));
3224	return VERR_INVALID_PARAMETER;
3225	}
3226	break;
3227	default:
3228	RTSemFastMutexRelease(pDevExt->mtxLdr);
3229	Log(("Invalid eEPType=%d\n", pReq->u.In.eEPType));
3230	return VERR_INVALID_PARAMETER;
3231	}
3232	if ( pReq->u.In.pfnModuleInit
3233	&& (uintptr_t)pReq->u.In.pfnModuleInit - (uintptr_t)pImage->pvImage >= pReq->u.In.cbImage)
3234	{
3235	RTSemFastMutexRelease(pDevExt->mtxLdr);
3236	Log(("SUP_IOCTL_LDR_LOAD: pfnModuleInit=%p is outside the image (%p %d bytes)\n",
3237	pReq->u.In.pfnModuleInit, pImage->pvImage, pReq->u.In.cbImage));
3238	return VERR_INVALID_PARAMETER;
3239	}
3240	if ( pReq->u.In.pfnModuleTerm
3241	&& (uintptr_t)pReq->u.In.pfnModuleTerm - (uintptr_t)pImage->pvImage >= pReq->u.In.cbImage)
3242	{
3243	RTSemFastMutexRelease(pDevExt->mtxLdr);
3244	Log(("SUP_IOCTL_LDR_LOAD: pfnModuleTerm=%p is outside the image (%p %d bytes)\n",
3245	pReq->u.In.pfnModuleTerm, pImage->pvImage, pReq->u.In.cbImage));
3246	return VERR_INVALID_PARAMETER;
3247	}
3248
3249	/*
3250	* Copy the memory.
3251	*/
3252	/* no need to do try/except as this is a buffered request. */
3253	memcpy(pImage->pvImage, &pReq->u.In.achImage[0], pImage->cbImage);
3254	pImage->uState = SUP_IOCTL_LDR_LOAD;
3255	pImage->pfnModuleInit = pReq->u.In.pfnModuleInit;
3256	pImage->pfnModuleTerm = pReq->u.In.pfnModuleTerm;
3257	pImage->offSymbols = pReq->u.In.offSymbols;
3258	pImage->cSymbols = pReq->u.In.cSymbols;
3259	pImage->offStrTab = pReq->u.In.offStrTab;
3260	pImage->cbStrTab = pReq->u.In.cbStrTab;
3261
3262	/*
3263	* Update any entry points.
3264	*/
3265	switch (pReq->u.In.eEPType)
3266	{
3267	default:
3268	case SUPLDRLOADEP_NOTHING:
3269	rc = VINF_SUCCESS;
3270	break;
3271	case SUPLDRLOADEP_VMMR0:
3272	rc = supdrvLdrSetR0EP(pDevExt, pReq->u.In.EP.VMMR0.pvVMMR0, pReq->u.In.EP.VMMR0.pvVMMR0EntryInt,
3273	pReq->u.In.EP.VMMR0.pvVMMR0EntryFast, pReq->u.In.EP.VMMR0.pvVMMR0EntryEx);
3274	break;
3275	}
3276
3277	/*
3278	* On success call the module initialization.
3279	*/
3280	LogFlow(("supdrvIOCtl_LdrLoad: pfnModuleInit=%p\n", pImage->pfnModuleInit));
3281	if (RT_SUCCESS(rc) && pImage->pfnModuleInit)
3282	{
3283	Log(("supdrvIOCtl_LdrLoad: calling pfnModuleInit=%p\n", pImage->pfnModuleInit));
3284	rc = pImage->pfnModuleInit();
3285	if (rc && pDevExt->pvVMMR0 == pImage->pvImage)
3286	supdrvLdrUnsetR0EP(pDevExt);
3287	}
3288
3289	if (rc)
3290	pImage->uState = SUP_IOCTL_LDR_OPEN;
3291
3292	RTSemFastMutexRelease(pDevExt->mtxLdr);
3293	return rc;
3294	}
3295
3296
3297	/**
3298	* Frees a previously loaded (prep'ed) image.
3299	*
3300	* @returns IPRT status code.
3301	* @param pDevExt Device globals.
3302	* @param pSession Session data.
3303	* @param pReq The request.
3304	*/
3305	static int supdrvIOCtl_LdrFree(PSUPDRVDEVEXT pDevExt, PSUPDRVSESSION pSession, PSUPLDRFREE pReq)
3306	{
3307	int rc;
3308	PSUPDRVLDRUSAGE pUsagePrev;
3309	PSUPDRVLDRUSAGE pUsage;
3310	PSUPDRVLDRIMAGE pImage;
3311	LogFlow(("supdrvIOCtl_LdrFree: pvImageBase=%p\n", pReq->u.In.pvImageBase));
3312
3313	/*
3314	* Find the ldr image.
3315	*/
3316	RTSemFastMutexRequest(pDevExt->mtxLdr);
3317	pUsagePrev = NULL;
3318	pUsage = pSession->pLdrUsage;
3319	while (pUsage && pUsage->pImage->pvImage != pReq->u.In.pvImageBase)
3320	{
3321	pUsagePrev = pUsage;
3322	pUsage = pUsage->pNext;
3323	}
3324	if (!pUsage)
3325	{
3326	RTSemFastMutexRelease(pDevExt->mtxLdr);
3327	Log(("SUP_IOCTL_LDR_FREE: couldn't find image!\n"));
3328	return VERR_INVALID_HANDLE;
3329	}
3330
3331	/*
3332	* Check if we can remove anything.
3333	*/
3334	rc = VINF_SUCCESS;
3335	pImage = pUsage->pImage;
3336	if (pImage->cUsage <= 1 \|\| pUsage->cUsage <= 1)
3337	{
3338	/*
3339	* Check if there are any objects with destructors in the image, if
3340	* so leave it for the session cleanup routine so we get a chance to
3341	* clean things up in the right order and not leave them all dangling.
3342	*/
3343	RTSPINLOCKTMP SpinlockTmp = RTSPINLOCKTMP_INITIALIZER;
3344	RTSpinlockAcquire(pDevExt->Spinlock, &SpinlockTmp);
3345	if (pImage->cUsage <= 1)
3346	{
3347	PSUPDRVOBJ pObj;
3348	for (pObj = pDevExt->pObjs; pObj; pObj = pObj->pNext)
3349	if (RT_UNLIKELY((uintptr_t)pObj->pfnDestructor - (uintptr_t)pImage->pvImage < pImage->cbImage))
3350	{
3351	rc = VERR_SHARING_VIOLATION; /** @todo VERR_DANGLING_OBJECTS */
3352	break;
3353	}
3354	}
3355	else
3356	{
3357	PSUPDRVUSAGE pGenUsage;
3358	for (pGenUsage = pSession->pUsage; pGenUsage; pGenUsage = pGenUsage->pNext)
3359	if (RT_UNLIKELY((uintptr_t)pGenUsage->pObj->pfnDestructor - (uintptr_t)pImage->pvImage < pImage->cbImage))
3360	{
3361	rc = VERR_SHARING_VIOLATION; /** @todo VERR_DANGLING_OBJECTS */
3362	break;
3363	}
3364	}
3365	RTSpinlockRelease(pDevExt->Spinlock, &SpinlockTmp);
3366	if (rc == VINF_SUCCESS)
3367	{
3368	/* unlink it */
3369	if (pUsagePrev)
3370	pUsagePrev->pNext = pUsage->pNext;
3371	else
3372	pSession->pLdrUsage = pUsage->pNext;
3373
3374	/* free it */
3375	pUsage->pImage = NULL;
3376	pUsage->pNext = NULL;
3377	RTMemFree(pUsage);
3378
3379	/*
3380	* Derefrence the image.
3381	*/
3382	if (pImage->cUsage <= 1)
3383	supdrvLdrFree(pDevExt, pImage);
3384	else
3385	pImage->cUsage--;
3386	}
3387	else
3388	Log(("supdrvIOCtl_LdrFree: Dangling objects in %p/%s!\n", pImage->pvImage, pImage->szName));
3389	}
3390	else
3391	{
3392	/*
3393	* Dereference both image and usage.
3394	*/
3395	pImage->cUsage--;
3396	pUsage->cUsage--;
3397	}
3398
3399	RTSemFastMutexRelease(pDevExt->mtxLdr);
3400	return VINF_SUCCESS;
3401	}
3402
3403
3404	/**
3405	* Gets the address of a symbol in an open image.
3406	*
3407	* @returns 0 on success.
3408	* @returns SUPDRV_ERR_* on failure.
3409	* @param pDevExt Device globals.
3410	* @param pSession Session data.
3411	* @param pReq The request buffer.
3412	*/
3413	static int supdrvIOCtl_LdrGetSymbol(PSUPDRVDEVEXT pDevExt, PSUPDRVSESSION pSession, PSUPLDRGETSYMBOL pReq)
3414	{
3415	PSUPDRVLDRIMAGE pImage;
3416	PSUPDRVLDRUSAGE pUsage;
3417	uint32_t i;
3418	PSUPLDRSYM paSyms;
3419	const char *pchStrings;
3420	const size_t cbSymbol = strlen(pReq->u.In.szSymbol) + 1;
3421	void *pvSymbol = NULL;
3422	int rc = VERR_GENERAL_FAILURE;
3423	Log3(("supdrvIOCtl_LdrGetSymbol: pvImageBase=%p szSymbol=\"%s\"\n", pReq->u.In.pvImageBase, pReq->u.In.szSymbol));
3424
3425	/*
3426	* Find the ldr image.
3427	*/
3428	RTSemFastMutexRequest(pDevExt->mtxLdr);
3429	pUsage = pSession->pLdrUsage;
3430	while (pUsage && pUsage->pImage->pvImage != pReq->u.In.pvImageBase)
3431	pUsage = pUsage->pNext;
3432	if (!pUsage)
3433	{
3434	RTSemFastMutexRelease(pDevExt->mtxLdr);
3435	Log(("SUP_IOCTL_LDR_GET_SYMBOL: couldn't find image!\n"));
3436	return VERR_INVALID_HANDLE;
3437	}
3438	pImage = pUsage->pImage;
3439	if (pImage->uState != SUP_IOCTL_LDR_LOAD)
3440	{
3441	unsigned uState = pImage->uState;
3442	RTSemFastMutexRelease(pDevExt->mtxLdr);
3443	Log(("SUP_IOCTL_LDR_GET_SYMBOL: invalid image state %d (%#x)!\n", uState, uState)); NOREF(uState);
3444	return VERR_ALREADY_LOADED;
3445	}
3446
3447	/*
3448	* Search the symbol string.
3449	*/
3450	pchStrings = (const char )((uint8_t )pImage->pvImage + pImage->offStrTab);
3451	paSyms = (PSUPLDRSYM)((uint8_t *)pImage->pvImage + pImage->offSymbols);
3452	for (i = 0; i < pImage->cSymbols; i++)
3453	{
3454	if ( paSyms[i].offSymbol < pImage->cbImage /* paranoia */
3455	&& paSyms[i].offName + cbSymbol <= pImage->cbStrTab
3456	&& !memcmp(pchStrings + paSyms[i].offName, pReq->u.In.szSymbol, cbSymbol))
3457	{
3458	pvSymbol = (uint8_t *)pImage->pvImage + paSyms[i].offSymbol;
3459	rc = VINF_SUCCESS;
3460	break;
3461	}
3462	}
3463	RTSemFastMutexRelease(pDevExt->mtxLdr);
3464	pReq->u.Out.pvSymbol = pvSymbol;
3465	return rc;
3466	}
3467
3468
3469	/**
3470	* Updates the IDT patches to point to the specified VMM R0 entry
3471	* point (i.e. VMMR0Enter()).
3472	*
3473	* @returns IPRT status code.
3474	* @param pDevExt Device globals.
3475	* @param pSession Session data.
3476	* @param pVMMR0 VMMR0 image handle.
3477	* @param pvVMMR0EntryInt VMMR0EntryInt address.
3478	* @param pvVMMR0EntryFast VMMR0EntryFast address.
3479	* @param pvVMMR0EntryEx VMMR0EntryEx address.
3480	* @remark Caller must own the loader mutex.
3481	*/
3482	static int supdrvLdrSetR0EP(PSUPDRVDEVEXT pDevExt, void pvVMMR0, void pvVMMR0EntryInt, void pvVMMR0EntryFast, void pvVMMR0EntryEx)
3483	{
3484	int rc = VINF_SUCCESS;
3485	LogFlow(("supdrvLdrSetR0EP pvVMMR0=%p pvVMMR0EntryInt=%p\n", pvVMMR0, pvVMMR0EntryInt));
3486
3487
3488	/*
3489	* Check if not yet set.
3490	*/
3491	if (!pDevExt->pvVMMR0)
3492	{
3493	#ifdef VBOX_WITH_IDT_PATCHING
3494	PSUPDRVPATCH pPatch;
3495	#endif
3496
3497	/*
3498	* Set it and update IDT patch code.
3499	*/
3500	pDevExt->pvVMMR0 = pvVMMR0;
3501	pDevExt->pfnVMMR0EntryInt = pvVMMR0EntryInt;
3502	pDevExt->pfnVMMR0EntryFast = pvVMMR0EntryFast;
3503	pDevExt->pfnVMMR0EntryEx = pvVMMR0EntryEx;
3504	#ifdef VBOX_WITH_IDT_PATCHING
3505	for (pPatch = pDevExt->pIdtPatches; pPatch; pPatch = pPatch->pNext)
3506	{
3507	# ifdef RT_ARCH_AMD64
3508	ASMAtomicXchgU64((volatile uint64_t *)&pPatch->auCode[pPatch->offVMMR0EntryFixup], (uint64_t)pvVMMR0);
3509	# else /* RT_ARCH_X86 */
3510	ASMAtomicXchgU32((volatile uint32_t *)&pPatch->auCode[pPatch->offVMMR0EntryFixup],
3511	(uint32_t)pvVMMR0 - (uint32_t)&pPatch->auCode[pPatch->offVMMR0EntryFixup + 4]);
3512	# endif
3513	}
3514	#endif /* VBOX_WITH_IDT_PATCHING */
3515	}
3516	else
3517	{
3518	/*
3519	* Return failure or success depending on whether the values match or not.
3520	*/
3521	if ( pDevExt->pvVMMR0 != pvVMMR0
3522	\|\| (void *)pDevExt->pfnVMMR0EntryInt != pvVMMR0EntryInt
3523	\|\| (void *)pDevExt->pfnVMMR0EntryFast != pvVMMR0EntryFast
3524	\|\| (void *)pDevExt->pfnVMMR0EntryEx != pvVMMR0EntryEx)
3525	{
3526	AssertMsgFailed(("SUP_IOCTL_LDR_SETR0EP: Already set pointing to a different module!\n"));
3527	rc = VERR_INVALID_PARAMETER;
3528	}
3529	}
3530	return rc;
3531	}
3532
3533
3534	/**
3535	* Unsets the R0 entry point installed by supdrvLdrSetR0EP.
3536	*
3537	* @param pDevExt Device globals.
3538	*/
3539	static void supdrvLdrUnsetR0EP(PSUPDRVDEVEXT pDevExt)
3540	{
3541	#ifdef VBOX_WITH_IDT_PATCHING
3542	PSUPDRVPATCH pPatch;
3543	#endif
3544
3545	pDevExt->pvVMMR0 = NULL;
3546	pDevExt->pfnVMMR0EntryInt = NULL;
3547	pDevExt->pfnVMMR0EntryFast = NULL;
3548	pDevExt->pfnVMMR0EntryEx = NULL;
3549
3550	#ifdef VBOX_WITH_IDT_PATCHING
3551	for (pPatch = pDevExt->pIdtPatches; pPatch; pPatch = pPatch->pNext)
3552	{
3553	# ifdef RT_ARCH_AMD64
3554	ASMAtomicXchgU64((volatile uint64_t *)&pPatch->auCode[pPatch->offVMMR0EntryFixup],
3555	(uint64_t)&pPatch->auCode[pPatch->offStub]);
3556	# else /* RT_ARCH_X86 */
3557	ASMAtomicXchgU32((volatile uint32_t *)&pPatch->auCode[pPatch->offVMMR0EntryFixup],
3558	(uint32_t)&pPatch->auCode[pPatch->offStub] - (uint32_t)&pPatch->auCode[pPatch->offVMMR0EntryFixup + 4]);
3559	# endif
3560	}
3561	#endif /* VBOX_WITH_IDT_PATCHING */
3562	}
3563
3564
3565	/**
3566	* Adds a usage reference in the specified session of an image.
3567	*
3568	* @param pSession Session in question.
3569	* @param pImage Image which the session is using.
3570	*/
3571	static void supdrvLdrAddUsage(PSUPDRVSESSION pSession, PSUPDRVLDRIMAGE pImage)
3572	{
3573	PSUPDRVLDRUSAGE pUsage;
3574	LogFlow(("supdrvLdrAddUsage: pImage=%p\n", pImage));
3575
3576	/*
3577	* Referenced it already?
3578	*/
3579	pUsage = pSession->pLdrUsage;
3580	while (pUsage)
3581	{
3582	if (pUsage->pImage == pImage)
3583	{
3584	pUsage->cUsage++;
3585	return;
3586	}
3587	pUsage = pUsage->pNext;
3588	}
3589
3590	/*
3591	* Allocate new usage record.
3592	*/
3593	pUsage = (PSUPDRVLDRUSAGE)RTMemAlloc(sizeof(*pUsage));
3594	Assert(pUsage);
3595	if (pUsage)
3596	{
3597	pUsage->cUsage = 1;
3598	pUsage->pImage = pImage;
3599	pUsage->pNext = pSession->pLdrUsage;
3600	pSession->pLdrUsage = pUsage;
3601	}
3602	/* ignore errors... */
3603	}
3604
3605
3606	/**
3607	* Frees a load image.
3608	*
3609	* @param pDevExt Pointer to device extension.
3610	* @param pImage Pointer to the image we're gonna free.
3611	* This image must exit!
3612	* @remark The caller MUST own SUPDRVDEVEXT::mtxLdr!
3613	*/
3614	static void supdrvLdrFree(PSUPDRVDEVEXT pDevExt, PSUPDRVLDRIMAGE pImage)
3615	{
3616	PSUPDRVLDRIMAGE pImagePrev;
3617	LogFlow(("supdrvLdrFree: pImage=%p\n", pImage));
3618
3619	/* find it - arg. should've used doubly linked list. */
3620	Assert(pDevExt->pLdrImages);
3621	pImagePrev = NULL;
3622	if (pDevExt->pLdrImages != pImage)
3623	{
3624	pImagePrev = pDevExt->pLdrImages;
3625	while (pImagePrev->pNext != pImage)
3626	pImagePrev = pImagePrev->pNext;
3627	Assert(pImagePrev->pNext == pImage);
3628	}
3629
3630	/* unlink */
3631	if (pImagePrev)
3632	pImagePrev->pNext = pImage->pNext;
3633	else
3634	pDevExt->pLdrImages = pImage->pNext;
3635
3636	/* check if this is VMMR0.r0 and fix the Idt patches if it is. */
3637	if (pDevExt->pvVMMR0 == pImage->pvImage)
3638	supdrvLdrUnsetR0EP(pDevExt);
3639
3640	/* check for objects with destructors in this image. (Shouldn't happen.) */
3641	if (pDevExt->pObjs)
3642	{
3643	unsigned cObjs = 0;
3644	PSUPDRVOBJ pObj;
3645	RTSPINLOCKTMP SpinlockTmp = RTSPINLOCKTMP_INITIALIZER;
3646	RTSpinlockAcquire(pDevExt->Spinlock, &SpinlockTmp);
3647	for (pObj = pDevExt->pObjs; pObj; pObj = pObj->pNext)
3648	if (RT_UNLIKELY((uintptr_t)pObj->pfnDestructor - (uintptr_t)pImage->pvImage < pImage->cbImage))
3649	{
3650	pObj->pfnDestructor = NULL;
3651	cObjs++;
3652	}
3653	RTSpinlockRelease(pDevExt->Spinlock, &SpinlockTmp);
3654	if (cObjs)
3655	OSDBGPRINT(("supdrvLdrFree: Image '%s' has %d dangling objects!\n", pImage->szName, cObjs));
3656	}
3657
3658	/* call termination function if fully loaded. */
3659	if ( pImage->pfnModuleTerm
3660	&& pImage->uState == SUP_IOCTL_LDR_LOAD)
3661	{
3662	LogFlow(("supdrvIOCtl_LdrLoad: calling pfnModuleTerm=%p\n", pImage->pfnModuleTerm));
3663	pImage->pfnModuleTerm();
3664	}
3665
3666	/* free the image */
3667	pImage->cUsage = 0;
3668	pImage->pNext = 0;
3669	pImage->uState = SUP_IOCTL_LDR_FREE;
3670	RTMemExecFree(pImage);
3671	}
3672
3673
3674	/**
3675	* Gets the current paging mode of the CPU and stores in in pOut.
3676	*/
3677	static SUPPAGINGMODE supdrvIOCtl_GetPagingMode(void)
3678	{
3679	SUPPAGINGMODE enmMode;
3680
3681	RTR0UINTREG cr0 = ASMGetCR0();
3682	if ((cr0 & (X86_CR0_PG \| X86_CR0_PE)) != (X86_CR0_PG \| X86_CR0_PE))
3683	enmMode = SUPPAGINGMODE_INVALID;
3684	else
3685	{
3686	RTR0UINTREG cr4 = ASMGetCR4();
3687	uint32_t fNXEPlusLMA = 0;
3688	if (cr4 & X86_CR4_PAE)
3689	{
3690	uint32_t fAmdFeatures = ASMCpuId_EDX(0x80000001);
3691	if (fAmdFeatures & (X86_CPUID_AMD_FEATURE_EDX_NX \| X86_CPUID_AMD_FEATURE_EDX_LONG_MODE))
3692	{
3693	uint64_t efer = ASMRdMsr(MSR_K6_EFER);
3694	if ((fAmdFeatures & X86_CPUID_AMD_FEATURE_EDX_NX) && (efer & MSR_K6_EFER_NXE))
3695	fNXEPlusLMA \|= RT_BIT(0);
3696	if ((fAmdFeatures & X86_CPUID_AMD_FEATURE_EDX_LONG_MODE) && (efer & MSR_K6_EFER_LMA))
3697	fNXEPlusLMA \|= RT_BIT(1);
3698	}
3699	}
3700
3701	switch ((cr4 & (X86_CR4_PAE \| X86_CR4_PGE)) \| fNXEPlusLMA)
3702	{
3703	case 0:
3704	enmMode = SUPPAGINGMODE_32_BIT;
3705	break;
3706
3707	case X86_CR4_PGE:
3708	enmMode = SUPPAGINGMODE_32_BIT_GLOBAL;
3709	break;
3710
3711	case X86_CR4_PAE:
3712	enmMode = SUPPAGINGMODE_PAE;
3713	break;
3714
3715	case X86_CR4_PAE \| RT_BIT(0):
3716	enmMode = SUPPAGINGMODE_PAE_NX;
3717	break;
3718
3719	case X86_CR4_PAE \| X86_CR4_PGE:
3720	enmMode = SUPPAGINGMODE_PAE_GLOBAL;
3721	break;
3722
3723	case X86_CR4_PAE \| X86_CR4_PGE \| RT_BIT(0):
3724	enmMode = SUPPAGINGMODE_PAE_GLOBAL;
3725	break;
3726
3727	case RT_BIT(1) \| X86_CR4_PAE:
3728	enmMode = SUPPAGINGMODE_AMD64;
3729	break;
3730
3731	case RT_BIT(1) \| X86_CR4_PAE \| RT_BIT(0):
3732	enmMode = SUPPAGINGMODE_AMD64_NX;
3733	break;
3734
3735	case RT_BIT(1) \| X86_CR4_PAE \| X86_CR4_PGE:
3736	enmMode = SUPPAGINGMODE_AMD64_GLOBAL;
3737	break;
3738
3739	case RT_BIT(1) \| X86_CR4_PAE \| X86_CR4_PGE \| RT_BIT(0):
3740	enmMode = SUPPAGINGMODE_AMD64_GLOBAL_NX;
3741	break;
3742
3743	default:
3744	AssertMsgFailed(("Cannot happen! cr4=%#x fNXEPlusLMA=%d\n", cr4, fNXEPlusLMA));
3745	enmMode = SUPPAGINGMODE_INVALID;
3746	break;
3747	}
3748	}
3749	return enmMode;
3750	}
3751
3752
3753	/**
3754	* Creates the GIP.
3755	*
3756	* @returns negative errno.
3757	* @param pDevExt Instance data. GIP stuff may be updated.
3758	*/
3759	static int supdrvGipCreate(PSUPDRVDEVEXT pDevExt)
3760	{
3761	PSUPGLOBALINFOPAGE pGip;
3762	RTHCPHYS HCPhysGip;
3763	uint32_t u32SystemResolution;
3764	uint32_t u32Interval;
3765	int rc;
3766
3767	LogFlow(("supdrvGipCreate:\n"));
3768
3769	/* assert order */
3770	Assert(pDevExt->u32SystemTimerGranularityGrant == 0);
3771	Assert(pDevExt->GipMemObj == NIL_RTR0MEMOBJ);
3772	Assert(!pDevExt->pGipTimer);
3773
3774	/*
3775	* Allocate a suitable page with a default kernel mapping.
3776	*/
3777	rc = RTR0MemObjAllocLow(&pDevExt->GipMemObj, PAGE_SIZE, false);
3778	if (RT_FAILURE(rc))
3779	{
3780	OSDBGPRINT(("supdrvGipCreate: failed to allocate the GIP page. rc=%d\n", rc));
3781	return rc;
3782	}
3783	pGip = (PSUPGLOBALINFOPAGE)RTR0MemObjAddress(pDevExt->GipMemObj); AssertPtr(pGip);
3784	HCPhysGip = RTR0MemObjGetPagePhysAddr(pDevExt->GipMemObj, 0); Assert(HCPhysGip != NIL_RTHCPHYS);
3785
3786	#if 0 /** @todo Disabled this as we didn't used to do it before and causes unnecessary stress on laptops.
3787	* It only applies to Windows and should probably revisited later, if possible made part of the
3788	* timer code (return min granularity in RTTimerGetSystemGranularity and set it in RTTimerStart). */
3789	/*
3790	* Try bump up the system timer resolution.
3791	* The more interrupts the better...
3792	*/
3793	if ( RT_SUCCESS(RTTimerRequestSystemGranularity( 488281 /* 2048 HZ */, &u32SystemResolution))
3794	\|\| RT_SUCCESS(RTTimerRequestSystemGranularity( 500000 /* 2000 HZ */, &u32SystemResolution))
3795	\|\| RT_SUCCESS(RTTimerRequestSystemGranularity( 976563 /* 1024 HZ */, &u32SystemResolution))
3796	\|\| RT_SUCCESS(RTTimerRequestSystemGranularity( 1000000 /* 1000 HZ */, &u32SystemResolution))
3797	\|\| RT_SUCCESS(RTTimerRequestSystemGranularity( 1953125 /* 512 HZ */, &u32SystemResolution))
3798	\|\| RT_SUCCESS(RTTimerRequestSystemGranularity( 2000000 /* 500 HZ */, &u32SystemResolution))
3799	\|\| RT_SUCCESS(RTTimerRequestSystemGranularity( 3906250 /* 256 HZ */, &u32SystemResolution))
3800	\|\| RT_SUCCESS(RTTimerRequestSystemGranularity( 4000000 /* 250 HZ */, &u32SystemResolution))
3801	\|\| RT_SUCCESS(RTTimerRequestSystemGranularity( 7812500 /* 128 HZ */, &u32SystemResolution))
3802	\|\| RT_SUCCESS(RTTimerRequestSystemGranularity(10000000 /* 100 HZ */, &u32SystemResolution))
3803	\|\| RT_SUCCESS(RTTimerRequestSystemGranularity(15625000 /* 64 HZ */, &u32SystemResolution))
3804	\|\| RT_SUCCESS(RTTimerRequestSystemGranularity(31250000 /* 32 HZ */, &u32SystemResolution))
3805	)
3806	{
3807	Assert(RTTimerGetSystemGranularity() <= u32SystemResolution);
3808	pDevExt->u32SystemTimerGranularityGrant = u32SystemResolution;
3809	}
3810	#endif
3811
3812	/*
3813	* Find a reasonable update interval and initialize the structure.
3814	*/
3815	u32Interval = u32SystemResolution = RTTimerGetSystemGranularity();
3816	while (u32Interval < 10000000 /* 10 ms */)
3817	u32Interval += u32SystemResolution;
3818
3819	supdrvGipInit(pDevExt, pGip, HCPhysGip, RTTimeSystemNanoTS(), 1000000000 / u32Interval /=Hz/);
3820
3821	/*
3822	* Create the timer.
3823	* If CPU_ALL isn't supported we'll have to fall back to synchronous mode.
3824	*/
3825	if (pGip->u32Mode == SUPGIPMODE_ASYNC_TSC)
3826	{
3827	rc = RTTimerCreateEx(&pDevExt->pGipTimer, u32Interval, RTTIMER_FLAGS_CPU_ALL, supdrvGipAsyncTimer, pDevExt);
3828	if (rc == VERR_NOT_SUPPORTED)
3829	{
3830	OSDBGPRINT(("supdrvGipCreate: omni timer not supported, falling back to synchronous mode\n"));
3831	pGip->u32Mode = SUPGIPMODE_SYNC_TSC;
3832	}
3833	}
3834	if (pGip->u32Mode != SUPGIPMODE_ASYNC_TSC)
3835	rc = RTTimerCreateEx(&pDevExt->pGipTimer, u32Interval, 0, supdrvGipSyncTimer, pDevExt);
3836	if (RT_SUCCESS(rc))
3837	{
3838	if (pGip->u32Mode == SUPGIPMODE_ASYNC_TSC)
3839	rc = RTMpNotificationRegister(supdrvGipMpEvent, pDevExt);
3840	if (RT_SUCCESS(rc))
3841	{
3842	/*
3843	* We're good.
3844	*/
3845	dprintf(("supdrvGipCreate: %ld ns interval.\n", (long)u32Interval));
3846	return VINF_SUCCESS;
3847	}
3848
3849	OSDBGPRINT(("supdrvGipCreate: failed register MP event notfication. rc=%d\n", rc));
3850	}
3851	else
3852	{
3853	OSDBGPRINT(("supdrvGipCreate: failed create GIP timer at %ld ns interval. rc=%d\n", (long)u32Interval, rc));
3854	Assert(!pDevExt->pGipTimer);
3855	}
3856	supdrvGipDestroy(pDevExt);
3857	return rc;
3858	}
3859
3860
3861	/**
3862	* Terminates the GIP.
3863	*
3864	* @param pDevExt Instance data. GIP stuff may be updated.
3865	*/
3866	static void supdrvGipDestroy(PSUPDRVDEVEXT pDevExt)
3867	{
3868	int rc;
3869	#ifdef DEBUG_DARWIN_GIP
3870	OSDBGPRINT(("supdrvGipDestroy: pDevExt=%p pGip=%p pGipTimer=%p GipMemObj=%p\n", pDevExt,
3871	pDevExt->GipMemObj != NIL_RTR0MEMOBJ ? RTR0MemObjAddress(pDevExt->GipMemObj) : NULL,
3872	pDevExt->pGipTimer, pDevExt->GipMemObj));
3873	#endif
3874
3875	/*
3876	* Invalid the GIP data.
3877	*/
3878	if (pDevExt->pGip)
3879	{
3880	supdrvGipTerm(pDevExt->pGip);
3881	pDevExt->pGip = NULL;
3882	}
3883
3884	/*
3885	* Destroy the timer and free the GIP memory object.
3886	*/
3887	if (pDevExt->pGipTimer)
3888	{
3889	rc = RTTimerDestroy(pDevExt->pGipTimer); AssertRC(rc);
3890	pDevExt->pGipTimer = NULL;
3891	}
3892
3893	if (pDevExt->GipMemObj != NIL_RTR0MEMOBJ)
3894	{
3895	rc = RTR0MemObjFree(pDevExt->GipMemObj, true /* free mappings */); AssertRC(rc);
3896	pDevExt->GipMemObj = NIL_RTR0MEMOBJ;
3897	}
3898
3899	/*
3900	* Finally, release the system timer resolution request if one succeeded.
3901	*/
3902	if (pDevExt->u32SystemTimerGranularityGrant)
3903	{
3904	rc = RTTimerReleaseSystemGranularity(pDevExt->u32SystemTimerGranularityGrant); AssertRC(rc);
3905	pDevExt->u32SystemTimerGranularityGrant = 0;
3906	}
3907	}
3908
3909
3910	/**
3911	* Timer callback function sync GIP mode.
3912	* @param pTimer The timer.
3913	* @param pvUser The device extension.
3914	*/
3915	static DECLCALLBACK(void) supdrvGipSyncTimer(PRTTIMER pTimer, void *pvUser, uint64_t iTick)
3916	{
3917	PSUPDRVDEVEXT pDevExt = (PSUPDRVDEVEXT)pvUser;
3918	supdrvGipUpdate(pDevExt->pGip, RTTimeSystemNanoTS());
3919	}
3920
3921
3922	/**
3923	* Timer callback function for async GIP mode.
3924	* @param pTimer The timer.
3925	* @param pvUser The device extension.
3926	*/
3927	static DECLCALLBACK(void) supdrvGipAsyncTimer(PRTTIMER pTimer, void *pvUser, uint64_t iTick)
3928	{
3929	PSUPDRVDEVEXT pDevExt = (PSUPDRVDEVEXT)pvUser;
3930	RTCPUID idCpu = RTMpCpuId();
3931	uint64_t NanoTS = RTTimeSystemNanoTS();
3932
3933	/** @todo reset the transaction number and whatnot when iTick == 1. */
3934	if (pDevExt->idGipMaster == idCpu)
3935	supdrvGipUpdate(pDevExt->pGip, NanoTS);
3936	else
3937	supdrvGipUpdatePerCpu(pDevExt->pGip, NanoTS, ASMGetApicId());
3938	}
3939
3940
3941	/**
3942	* Multiprocessor event notification callback.
3943	*
3944	* This is used to make sue that the GIP master gets passed on to
3945	* another CPU.
3946	*
3947	* @param enmEvent The event.
3948	* @param idCpu The cpu it applies to.
3949	* @param pvUser Pointer to the device extension.
3950	*/
3951	static DECLCALLBACK(void) supdrvGipMpEvent(RTMPEVENT enmEvent, RTCPUID idCpu, void *pvUser)
3952	{
3953	PSUPDRVDEVEXT pDevExt = (PSUPDRVDEVEXT)pvUser;
3954	if (enmEvent == RTMPEVENT_OFFLINE)
3955	{
3956	RTCPUID idGipMaster;
3957	ASMAtomicReadSize(&pDevExt->idGipMaster, &idGipMaster);
3958	if (idGipMaster == idCpu)
3959	{
3960	/*
3961	* Find a new GIP master.
3962	*/
3963	bool fIgnored;
3964	unsigned i;
3965	RTCPUID idNewGipMaster = NIL_RTCPUID;
3966	RTCPUSET OnlineCpus;
3967	RTMpGetOnlineSet(&OnlineCpus);
3968
3969	for (i = 0; i < RTCPUSET_MAX_CPUS; i++)
3970	{
3971	RTCPUID idCurCpu = RTMpCpuIdFromSetIndex(i);
3972	if ( RTCpuSetIsMember(&OnlineCpus, idCurCpu)
3973	&& idCurCpu != idGipMaster)
3974	{
3975	idNewGipMaster = idCurCpu;
3976	break;
3977	}
3978	}
3979
3980	dprintf(("supdrvGipMpEvent: Gip master %#lx -> %#lx\n", (long)idGipMaster, (long)idNewGipMaster));
3981	ASMAtomicCmpXchgSize(&pDevExt->idGipMaster, idNewGipMaster, idGipMaster, fIgnored);
3982	NOREF(fIgnored);
3983	}
3984	}
3985	}
3986
3987
3988	/**
3989	* Initializes the GIP data.
3990	*
3991	* @returns IPRT status code.
3992	* @param pDevExt Pointer to the device instance data.
3993	* @param pGip Pointer to the read-write kernel mapping of the GIP.
3994	* @param HCPhys The physical address of the GIP.
3995	* @param u64NanoTS The current nanosecond timestamp.
3996	* @param uUpdateHz The update freqence.
3997	*/
3998	int VBOXCALL supdrvGipInit(PSUPDRVDEVEXT pDevExt, PSUPGLOBALINFOPAGE pGip, RTHCPHYS HCPhys, uint64_t u64NanoTS, unsigned uUpdateHz)
3999	{
4000	unsigned i;
4001	#ifdef DEBUG_DARWIN_GIP
4002	OSDBGPRINT(("supdrvGipInit: pGip=%p HCPhys=%lx u64NanoTS=%llu uUpdateHz=%d\n", pGip, (long)HCPhys, u64NanoTS, uUpdateHz));
4003	#else
4004	LogFlow(("supdrvGipInit: pGip=%p HCPhys=%lx u64NanoTS=%llu uUpdateHz=%d\n", pGip, (long)HCPhys, u64NanoTS, uUpdateHz));
4005	#endif
4006
4007	/*
4008	* Initialize the structure.
4009	*/
4010	memset(pGip, 0, PAGE_SIZE);
4011	pGip->u32Magic = SUPGLOBALINFOPAGE_MAGIC;
4012	pGip->u32Version = SUPGLOBALINFOPAGE_VERSION;
4013	pGip->u32Mode = supdrvGipDeterminTscMode(pDevExt);
4014	pGip->u32UpdateHz = uUpdateHz;
4015	pGip->u32UpdateIntervalNS = 1000000000 / uUpdateHz;
4016	pGip->u64NanoTSLastUpdateHz = u64NanoTS;
4017
4018	for (i = 0; i < RT_ELEMENTS(pGip->aCPUs); i++)
4019	{
4020	pGip->aCPUs[i].u32TransactionId = 2;
4021	pGip->aCPUs[i].u64NanoTS = u64NanoTS;
4022	pGip->aCPUs[i].u64TSC = ASMReadTSC();
4023
4024	/*
4025	* We don't know the following values until we've executed updates.
4026	* So, we'll just insert very high values.
4027	*/
4028	pGip->aCPUs[i].u64CpuHz = _4G + 1;
4029	pGip->aCPUs[i].u32UpdateIntervalTSC = _2G / 4;
4030	pGip->aCPUs[i].au32TSCHistory[0] = _2G / 4;
4031	pGip->aCPUs[i].au32TSCHistory[1] = _2G / 4;
4032	pGip->aCPUs[i].au32TSCHistory[2] = _2G / 4;
4033	pGip->aCPUs[i].au32TSCHistory[3] = _2G / 4;
4034	pGip->aCPUs[i].au32TSCHistory[4] = _2G / 4;
4035	pGip->aCPUs[i].au32TSCHistory[5] = _2G / 4;
4036	pGip->aCPUs[i].au32TSCHistory[6] = _2G / 4;
4037	pGip->aCPUs[i].au32TSCHistory[7] = _2G / 4;
4038	}
4039
4040	/*
4041	* Link it to the device extension.
4042	*/
4043	pDevExt->pGip = pGip;
4044	pDevExt->HCPhysGip = HCPhys;
4045	pDevExt->cGipUsers = 0;
4046
4047	return VINF_SUCCESS;
4048	}
4049
4050
4051	/**
4052	* Callback used by supdrvDetermineAsyncTSC to read the TSC on a CPU.
4053	*
4054	* @param idCpu Ignored.
4055	* @param pvUser1 Where to put the TSC.
4056	* @param pvUser2 Ignored.
4057	*/
4058	static DECLCALLBACK(void) supdrvDetermineAsyncTscWorker(RTCPUID idCpu, void pvUser1, void pvUser2)
4059	{
4060	#if 1
4061	ASMAtomicWriteU64((uint64_t volatile *)pvUser1, ASMReadTSC());
4062	#else
4063	(uint64_t )pvUser1 = ASMReadTSC();
4064	#endif
4065	}
4066
4067
4068	/**
4069	* Determine if Async GIP mode is required because of TSC drift.
4070	*
4071	* When using the default/normal timer code it is essential that the time stamp counter
4072	* (TSC) runs never backwards, that is, a read operation to the counter should return
4073	* a bigger value than any previous read operation. This is guaranteed by the latest
4074	* AMD CPUs and by newer Intel CPUs which never enter the C2 state (P4). In any other
4075	* case we have to choose the asynchronous timer mode.
4076	*
4077	* @param poffMin Pointer to the determined difference between different cores.
4078	* @return false if the time stamp counters appear to be synchron, true otherwise.
4079	*/
4080	bool VBOXCALL supdrvDetermineAsyncTsc(uint64_t *poffMin)
4081	{
4082	/*
4083	* Just iterate all the cpus 8 times and make sure that the TSC is
4084	* ever increasing. We don't bother taking TSC rollover into account.
4085	*/
4086	RTCPUSET CpuSet;
4087	int iLastCpu = RTCpuLastIndex(RTMpGetSet(&CpuSet));
4088	int iCpu;
4089	int cLoops = 8;
4090	bool fAsync = false;
4091	int rc;
4092	uint64_t offMax = 0;
4093	uint64_t offMin = ~(uint64_t)0;
4094	uint64_t PrevTsc = ASMReadTSC();
4095
4096	while (cLoops-- > 0)
4097	{
4098	for (iCpu = 0; iCpu <= iLastCpu; iCpu++)
4099	{
4100	uint64_t CurTsc;
4101	rc = RTMpOnSpecific(RTMpCpuIdFromSetIndex(iCpu), supdrvDetermineAsyncTscWorker, &CurTsc, NULL);
4102	if (RT_SUCCESS(rc))
4103	{
4104	if (CurTsc <= PrevTsc)
4105	{
4106	fAsync = true;
4107	offMin = offMax = PrevTsc - CurTsc;
4108	dprintf(("supdrvDetermineAsyncTsc: iCpu=%d cLoops=%d CurTsc=%llx PrevTsc=%llx\n",
4109	iCpu, cLoops, CurTsc, PrevTsc));
4110	break;
4111	}
4112
4113	/* Gather statistics (except the first time). */
4114	if (iCpu != 0 \|\| cLoops != 7)
4115	{
4116	uint64_t off = CurTsc - PrevTsc;
4117	if (off < offMin)
4118	offMin = off;
4119	if (off > offMax)
4120	offMax = off;
4121	dprintf2(("%d/%d: off=%llx\n", cLoops, iCpu, off));
4122	}
4123
4124	/* Next */
4125	PrevTsc = CurTsc;
4126	}
4127	else if (rc == VERR_NOT_SUPPORTED)
4128	break;
4129	else
4130	AssertMsg(rc == VERR_CPU_NOT_FOUND \|\| rc == VERR_CPU_OFFLINE, ("%d\n", rc));
4131	}
4132
4133	/* broke out of the loop. */
4134	if (iCpu <= iLastCpu)
4135	break;
4136	}
4137
4138	poffMin = offMin; / Almost RTMpOnSpecific profiling. */
4139	dprintf(("supdrvDetermineAsyncTsc: returns %d; iLastCpu=%d rc=%d offMin=%llx offMax=%llx\n",
4140	fAsync, iLastCpu, rc, offMin, offMax));
4141	#if !defined(RT_OS_SOLARIS) && !defined(RT_OS_OS2) && !defined(RT_OS_WINDOWS)
4142	OSDBGPRINT(("vboxdrv: fAsync=%d offMin=%#lx offMax=%#lx\n", fAsync, (long)offMin, (long)offMax));
4143	#endif
4144	return fAsync;
4145	}
4146
4147
4148	/**
4149	* Determin the GIP TSC mode.
4150	*
4151	* @returns The most suitable TSC mode.
4152	* @param pDevExt Pointer to the device instance data.
4153	*/
4154	static SUPGIPMODE supdrvGipDeterminTscMode(PSUPDRVDEVEXT pDevExt)
4155	{
4156	/*
4157	* On SMP we're faced with two problems:
4158	* (1) There might be a skew between the CPU, so that cpu0
4159	* returns a TSC that is sligtly different from cpu1.
4160	* (2) Power management (and other things) may cause the TSC
4161	* to run at a non-constant speed, and cause the speed
4162	* to be different on the cpus. This will result in (1).
4163	*
4164	* So, on SMP systems we'll have to select the ASYNC update method
4165	* if there are symphoms of these problems.
4166	*/
4167	if (RTMpGetCount() > 1)
4168	{
4169	uint32_t uEAX, uEBX, uECX, uEDX;
4170	uint64_t u64DiffCoresIgnored;
4171
4172	/* Permit the user and/or the OS specfic bits to force async mode. */
4173	if (supdrvOSGetForcedAsyncTscMode(pDevExt))
4174	return SUPGIPMODE_ASYNC_TSC;
4175
4176	/* Try check for current differences between the cpus. */
4177	if (supdrvDetermineAsyncTsc(&u64DiffCoresIgnored))
4178	return SUPGIPMODE_ASYNC_TSC;
4179
4180	/*
4181	* If the CPU supports power management and is an AMD one we
4182	* won't trust it unless it has the TscInvariant bit is set.
4183	*/
4184	/* Check for "AuthenticAMD" */
4185	ASMCpuId(0, &uEAX, &uEBX, &uECX, &uEDX);
4186	if ( uEAX >= 1
4187	&& uEBX == X86_CPUID_VENDOR_AMD_EBX
4188	&& uECX == X86_CPUID_VENDOR_AMD_ECX
4189	&& uEDX == X86_CPUID_VENDOR_AMD_EDX)
4190	{
4191	/* Check for APM support and that TscInvariant is cleared. */
4192	ASMCpuId(0x80000000, &uEAX, &uEBX, &uECX, &uEDX);
4193	if (uEAX >= 0x80000007)
4194	{
4195	ASMCpuId(0x80000007, &uEAX, &uEBX, &uECX, &uEDX);
4196	if ( !(uEDX & RT_BIT(8))/* TscInvariant */
4197	&& (uEDX & 0x3e)) /* STC\|TM\|THERMTRIP\|VID\|FID. Ignore TS. */
4198	return SUPGIPMODE_ASYNC_TSC;
4199	}
4200	}
4201	}
4202	return SUPGIPMODE_SYNC_TSC;
4203	}
4204
4205
4206	/**
4207	* Invalidates the GIP data upon termination.
4208	*
4209	* @param pGip Pointer to the read-write kernel mapping of the GIP.
4210	*/
4211	void VBOXCALL supdrvGipTerm(PSUPGLOBALINFOPAGE pGip)
4212	{
4213	unsigned i;
4214	pGip->u32Magic = 0;
4215	for (i = 0; i < RT_ELEMENTS(pGip->aCPUs); i++)
4216	{
4217	pGip->aCPUs[i].u64NanoTS = 0;
4218	pGip->aCPUs[i].u64TSC = 0;
4219	pGip->aCPUs[i].iTSCHistoryHead = 0;
4220	}
4221	}
4222
4223
4224	/**
4225	* Worker routine for supdrvGipUpdate and supdrvGipUpdatePerCpu that
4226	* updates all the per cpu data except the transaction id.
4227	*
4228	* @param pGip The GIP.
4229	* @param pGipCpu Pointer to the per cpu data.
4230	* @param u64NanoTS The current time stamp.
4231	*/
4232	static void supdrvGipDoUpdateCpu(PSUPGLOBALINFOPAGE pGip, PSUPGIPCPU pGipCpu, uint64_t u64NanoTS)
4233	{
4234	uint64_t u64TSC;
4235	uint64_t u64TSCDelta;
4236	uint32_t u32UpdateIntervalTSC;
4237	uint32_t u32UpdateIntervalTSCSlack;
4238	unsigned iTSCHistoryHead;
4239	uint64_t u64CpuHz;
4240
4241	/*
4242	* Update the NanoTS.
4243	*/
4244	ASMAtomicXchgU64(&pGipCpu->u64NanoTS, u64NanoTS);
4245
4246	/*
4247	* Calc TSC delta.
4248	*/
4249	/** @todo validate the NanoTS delta, don't trust the OS to call us when it should... */
4250	u64TSC = ASMReadTSC();
4251	u64TSCDelta = u64TSC - pGipCpu->u64TSC;
4252	ASMAtomicXchgU64(&pGipCpu->u64TSC, u64TSC);
4253
4254	if (u64TSCDelta >> 32)
4255	{
4256	u64TSCDelta = pGipCpu->u32UpdateIntervalTSC;
4257	pGipCpu->cErrors++;
4258	}
4259
4260	/*
4261	* TSC History.
4262	*/
4263	Assert(ELEMENTS(pGipCpu->au32TSCHistory) == 8);
4264
4265	iTSCHistoryHead = (pGipCpu->iTSCHistoryHead + 1) & 7;
4266	ASMAtomicXchgU32(&pGipCpu->iTSCHistoryHead, iTSCHistoryHead);
4267	ASMAtomicXchgU32(&pGipCpu->au32TSCHistory[iTSCHistoryHead], (uint32_t)u64TSCDelta);
4268
4269	/*
4270	* UpdateIntervalTSC = average of last 8,2,1 intervals depending on update HZ.
4271	*/
4272	if (pGip->u32UpdateHz >= 1000)
4273	{
4274	uint32_t u32;
4275	u32 = pGipCpu->au32TSCHistory[0];
4276	u32 += pGipCpu->au32TSCHistory[1];
4277	u32 += pGipCpu->au32TSCHistory[2];
4278	u32 += pGipCpu->au32TSCHistory[3];
4279	u32 >>= 2;
4280	u32UpdateIntervalTSC = pGipCpu->au32TSCHistory[4];
4281	u32UpdateIntervalTSC += pGipCpu->au32TSCHistory[5];
4282	u32UpdateIntervalTSC += pGipCpu->au32TSCHistory[6];
4283	u32UpdateIntervalTSC += pGipCpu->au32TSCHistory[7];
4284	u32UpdateIntervalTSC >>= 2;
4285	u32UpdateIntervalTSC += u32;
4286	u32UpdateIntervalTSC >>= 1;
4287
4288	/* Value choosen for a 2GHz Athlon64 running linux 2.6.10/11, . */
4289	u32UpdateIntervalTSCSlack = u32UpdateIntervalTSC >> 14;
4290	}
4291	else if (pGip->u32UpdateHz >= 90)
4292	{
4293	u32UpdateIntervalTSC = (uint32_t)u64TSCDelta;
4294	u32UpdateIntervalTSC += pGipCpu->au32TSCHistory[(iTSCHistoryHead - 1) & 7];
4295	u32UpdateIntervalTSC >>= 1;
4296
4297	/* value choosen on a 2GHz thinkpad running windows */
4298	u32UpdateIntervalTSCSlack = u32UpdateIntervalTSC >> 7;
4299	}
4300	else
4301	{
4302	u32UpdateIntervalTSC = (uint32_t)u64TSCDelta;
4303
4304	/* This value hasn't be checked yet.. waiting for OS/2 and 33Hz timers.. :-) */
4305	u32UpdateIntervalTSCSlack = u32UpdateIntervalTSC >> 6;
4306	}
4307	ASMAtomicXchgU32(&pGipCpu->u32UpdateIntervalTSC, u32UpdateIntervalTSC + u32UpdateIntervalTSCSlack);
4308
4309	/*
4310	* CpuHz.
4311	*/
4312	u64CpuHz = ASMMult2xU32RetU64(u32UpdateIntervalTSC, pGip->u32UpdateHz);
4313	ASMAtomicXchgU64(&pGipCpu->u64CpuHz, u64CpuHz);
4314	}
4315
4316
4317	/**
4318	* Updates the GIP.
4319	*
4320	* @param pGip Pointer to the GIP.
4321	* @param u64NanoTS The current nanosecond timesamp.
4322	*/
4323	void VBOXCALL supdrvGipUpdate(PSUPGLOBALINFOPAGE pGip, uint64_t u64NanoTS)
4324	{
4325	/*
4326	* Determin the relevant CPU data.
4327	*/
4328	PSUPGIPCPU pGipCpu;
4329	if (pGip->u32Mode != SUPGIPMODE_ASYNC_TSC)
4330	pGipCpu = &pGip->aCPUs[0];
4331	else
4332	{
4333	unsigned iCpu = ASMGetApicId();
4334	if (RT_LIKELY(iCpu >= RT_ELEMENTS(pGip->aCPUs)))
4335	return;
4336	pGipCpu = &pGip->aCPUs[iCpu];
4337	}
4338
4339	/*
4340	* Start update transaction.
4341	*/
4342	if (!(ASMAtomicIncU32(&pGipCpu->u32TransactionId) & 1))
4343	{
4344	/* this can happen on win32 if we're taking to long and there are more CPUs around. shouldn't happen though. */
4345	AssertMsgFailed(("Invalid transaction id, %#x, not odd!\n", pGipCpu->u32TransactionId));
4346	ASMAtomicIncU32(&pGipCpu->u32TransactionId);
4347	pGipCpu->cErrors++;
4348	return;
4349	}
4350
4351	/*
4352	* Recalc the update frequency every 0x800th time.
4353	*/
4354	if (!(pGipCpu->u32TransactionId & (GIP_UPDATEHZ_RECALC_FREQ * 2 - 2)))
4355	{
4356	if (pGip->u64NanoTSLastUpdateHz)
4357	{
4358	#ifdef RT_ARCH_AMD64 /** @todo fix 64-bit div here to work on x86 linux. */
4359	uint64_t u64Delta = u64NanoTS - pGip->u64NanoTSLastUpdateHz;
4360	uint32_t u32UpdateHz = (uint32_t)((UINT64_C(1000000000) * GIP_UPDATEHZ_RECALC_FREQ) / u64Delta);
4361	if (u32UpdateHz <= 2000 && u32UpdateHz >= 30)
4362	{
4363	ASMAtomicXchgU32(&pGip->u32UpdateHz, u32UpdateHz);
4364	ASMAtomicXchgU32(&pGip->u32UpdateIntervalNS, 1000000000 / u32UpdateHz);
4365	}
4366	#endif
4367	}
4368	ASMAtomicXchgU64(&pGip->u64NanoTSLastUpdateHz, u64NanoTS);
4369	}
4370
4371	/*
4372	* Update the data.
4373	*/
4374	supdrvGipDoUpdateCpu(pGip, pGipCpu, u64NanoTS);
4375
4376	/*
4377	* Complete transaction.
4378	*/
4379	ASMAtomicIncU32(&pGipCpu->u32TransactionId);
4380	}
4381
4382
4383	/**
4384	* Updates the per cpu GIP data for the calling cpu.
4385	*
4386	* @param pGip Pointer to the GIP.
4387	* @param u64NanoTS The current nanosecond timesamp.
4388	* @param iCpu The CPU index.
4389	*/
4390	void VBOXCALL supdrvGipUpdatePerCpu(PSUPGLOBALINFOPAGE pGip, uint64_t u64NanoTS, unsigned iCpu)
4391	{
4392	PSUPGIPCPU pGipCpu;
4393
4394	if (RT_LIKELY(iCpu < RT_ELEMENTS(pGip->aCPUs)))
4395	{
4396	pGipCpu = &pGip->aCPUs[iCpu];
4397
4398	/*
4399	* Start update transaction.
4400	*/
4401	if (!(ASMAtomicIncU32(&pGipCpu->u32TransactionId) & 1))
4402	{
4403	AssertMsgFailed(("Invalid transaction id, %#x, not odd!\n", pGipCpu->u32TransactionId));
4404	ASMAtomicIncU32(&pGipCpu->u32TransactionId);
4405	pGipCpu->cErrors++;
4406	return;
4407	}
4408
4409	/*
4410	* Update the data.
4411	*/
4412	supdrvGipDoUpdateCpu(pGip, pGipCpu, u64NanoTS);
4413
4414	/*
4415	* Complete transaction.
4416	*/
4417	ASMAtomicIncU32(&pGipCpu->u32TransactionId);
4418	}
4419	}
4420
4421
4422	#ifndef DEBUG /** @todo change #ifndef DEBUG -> #ifdef LOG_ENABLED */
4423	/**
4424	* Stub function for non-debug builds.
4425	*/
4426	RTDECL(PRTLOGGER) RTLogDefaultInstance(void)
4427	{
4428	return NULL;
4429	}
4430
4431	RTDECL(PRTLOGGER) RTLogRelDefaultInstance(void)
4432	{
4433	return NULL;
4434	}
4435
4436	/**
4437	* Stub function for non-debug builds.
4438	*/
4439	RTDECL(int) RTLogSetDefaultInstanceThread(PRTLOGGER pLogger, uintptr_t uKey)
4440	{
4441	return 0;
4442	}
4443
4444	/**
4445	* Stub function for non-debug builds.
4446	*/
4447	RTDECL(void) RTLogLogger(PRTLOGGER pLogger, void pvCallerRet, const char pszFormat, ...)
4448	{
4449	}
4450
4451	/**
4452	* Stub function for non-debug builds.
4453	*/
4454	RTDECL(void) RTLogLoggerEx(PRTLOGGER pLogger, unsigned fFlags, unsigned iGroup, const char *pszFormat, ...)
4455	{
4456	}
4457
4458	/**
4459	* Stub function for non-debug builds.
4460	*/
4461	RTDECL(void) RTLogLoggerExV(PRTLOGGER pLogger, unsigned fFlags, unsigned iGroup, const char *pszFormat, va_list args)
4462	{
4463	}
4464
4465	/**
4466	* Stub function for non-debug builds.
4467	*/
4468	RTDECL(void) RTLogPrintf(const char *pszFormat, ...)
4469	{
4470	}
4471
4472	/**
4473	* Stub function for non-debug builds.
4474	*/
4475	RTDECL(void) RTLogPrintfV(const char *pszFormat, va_list args)
4476	{
4477	}
4478	#endif /* !DEBUG */
4479

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

source: vbox/trunk/src/VBox/HostDrivers/Support/SUPDrv.c@ 10265

Download in other formats: