IEMAllN8veRecompiler.cpp@ 103807

Last change on this file since 103807 was 103807, checked in by vboxsync, 9 months ago
VMM/IEM: Split up the native recompiler functions (IEMNativeFunctions.cpp.h) into 4 files to speed up compilation and reduce compiler memory consumption. This involved splitting out half the content of IEMAllThrdRecompiler.cpp into IEMAllN8veRecompFuncs.h and IEMN8veRecompiler.h. bugref:10371
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 403.9 KB

Line
1	/* $Id: IEMAllN8veRecompiler.cpp 103807 2024-03-12 19:43:31Z vboxsync $ */
2	/** @file
3	* IEM - Native Recompiler
4	*
5	* Logging group IEM_RE_NATIVE assignments:
6	* - Level 1 (Log) : ...
7	* - Flow (LogFlow) : ...
8	* - Level 2 (Log2) : Details calls as they're recompiled.
9	* - Level 3 (Log3) : Disassemble native code after recompiling.
10	* - Level 4 (Log4) : ...
11	* - Level 5 (Log5) : ...
12	* - Level 6 (Log6) : ...
13	* - Level 7 (Log7) : ...
14	* - Level 8 (Log8) : ...
15	* - Level 9 (Log9) : ...
16	* - Level 10 (Log10): ...
17	* - Level 11 (Log11): Variable allocator.
18	* - Level 12 (Log12): Register allocator.
19	*/
20
21	/*
22	* Copyright (C) 2023 Oracle and/or its affiliates.
23	*
24	* This file is part of VirtualBox base platform packages, as
25	* available from https://www.virtualbox.org.
26	*
27	* This program is free software; you can redistribute it and/or
28	* modify it under the terms of the GNU General Public License
29	* as published by the Free Software Foundation, in version 3 of the
30	* License.
31	*
32	* This program is distributed in the hope that it will be useful, but
33	* WITHOUT ANY WARRANTY; without even the implied warranty of
34	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
35	* General Public License for more details.
36	*
37	* You should have received a copy of the GNU General Public License
38	* along with this program; if not, see <https://www.gnu.org/licenses>.
39	*
40	* SPDX-License-Identifier: GPL-3.0-only
41	*/
42
43
44	/*********************************************************************************************************************************
45	* Header Files *
46	*********************************************************************************************************************************/
47	#define LOG_GROUP LOG_GROUP_IEM_RE_NATIVE
48	#define IEM_WITH_OPAQUE_DECODER_STATE
49	#define VMCPU_INCL_CPUM_GST_CTX
50	#define VMM_INCLUDED_SRC_include_IEMMc_h /* block IEMMc.h inclusion. */
51	#include <VBox/vmm/iem.h>
52	#include <VBox/vmm/cpum.h>
53	#include <VBox/vmm/dbgf.h>
54	#include "IEMInternal.h"
55	#include <VBox/vmm/vmcc.h>
56	#include <VBox/log.h>
57	#include <VBox/err.h>
58	#include <VBox/dis.h>
59	#include <VBox/param.h>
60	#include <iprt/assert.h>
61	#include <iprt/heap.h>
62	#include <iprt/mem.h>
63	#include <iprt/string.h>
64	#if defined(RT_ARCH_AMD64)
65	# include <iprt/x86.h>
66	#elif defined(RT_ARCH_ARM64)
67	# include <iprt/armv8.h>
68	#endif
69
70	#ifdef RT_OS_WINDOWS
71	# include <iprt/formats/pecoff.h> /* this is incomaptible with windows.h, thus: */
72	extern "C" DECLIMPORT(uint8_t) __cdecl RtlAddFunctionTable(void *pvFunctionTable, uint32_t cEntries, uintptr_t uBaseAddress);
73	extern "C" DECLIMPORT(uint8_t) __cdecl RtlDelFunctionTable(void *pvFunctionTable);
74	#else
75	# include <iprt/formats/dwarf.h>
76	# if defined(RT_OS_DARWIN)
77	# include <libkern/OSCacheControl.h>
78	# define IEMNATIVE_USE_LIBUNWIND
79	extern "C" void __register_frame(const void *pvFde);
80	extern "C" void __deregister_frame(const void *pvFde);
81	# else
82	# ifdef DEBUG_bird /** @todo not thread safe yet */
83	# define IEMNATIVE_USE_GDB_JIT
84	# endif
85	# ifdef IEMNATIVE_USE_GDB_JIT
86	# include <iprt/critsect.h>
87	# include <iprt/once.h>
88	# include <iprt/formats/elf64.h>
89	# endif
90	extern "C" void __register_frame_info(void pvBegin, void pvObj); /* found no header for these two */
91	extern "C" void __deregister_frame_info(void pvBegin); /* (returns pvObj from __register_frame_info call) */
92	# endif
93	#endif
94	#ifdef VBOX_WITH_IEM_USING_CAPSTONE_DISASSEMBLER
95	# include "/opt/local/include/capstone/capstone.h"
96	#endif
97
98	#include "IEMInline.h"
99	#include "IEMThreadedFunctions.h"
100	#include "IEMN8veRecompiler.h"
101	#include "IEMN8veRecompilerEmit.h"
102	#include "IEMN8veRecompilerTlbLookup.h"
103	#include "IEMNativeFunctions.h"
104
105
106	/*
107	* Narrow down configs here to avoid wasting time on unused configs here.
108	* Note! Same checks in IEMAllThrdRecompiler.cpp.
109	*/
110
111	#ifndef IEM_WITH_CODE_TLB
112	# error The code TLB must be enabled for the recompiler.
113	#endif
114
115	#ifndef IEM_WITH_DATA_TLB
116	# error The data TLB must be enabled for the recompiler.
117	#endif
118
119	#ifndef IEM_WITH_SETJMP
120	# error The setjmp approach must be enabled for the recompiler.
121	#endif
122
123	/** @todo eliminate this clang build hack. */
124	#if RT_CLANG_PREREQ(4, 0)
125	# pragma GCC diagnostic ignored "-Wunused-function"
126	#endif
127
128
129	/*********************************************************************************************************************************
130	* Internal Functions *
131	*********************************************************************************************************************************/
132	#ifdef IEMNATIVE_WITH_TB_DEBUG_INFO
133	static void iemNativeDbgInfoAddLabel(PIEMRECOMPILERSTATE pReNative, IEMNATIVELABELTYPE enmType, uint16_t uData);
134	#endif
135	DECL_FORCE_INLINE(void) iemNativeRegClearGstRegShadowing(PIEMRECOMPILERSTATE pReNative, uint8_t idxHstReg, uint32_t off);
136	DECL_FORCE_INLINE(void) iemNativeRegClearGstRegShadowingOne(PIEMRECOMPILERSTATE pReNative, uint8_t idxHstReg,
137	IEMNATIVEGSTREG enmGstReg, uint32_t off);
138	DECL_INLINE_THROW(void) iemNativeVarRegisterRelease(PIEMRECOMPILERSTATE pReNative, uint8_t idxVar);
139
140
141	/*********************************************************************************************************************************
142	* Executable Memory Allocator *
143	*********************************************************************************************************************************/
144	/** @def IEMEXECMEM_USE_ALT_SUB_ALLOCATOR
145	* Use an alternative chunk sub-allocator that does store internal data
146	* in the chunk.
147	*
148	* Using the RTHeapSimple is not practial on newer darwin systems where
149	* RTMEM_PROT_WRITE and RTMEM_PROT_EXEC are mutually exclusive in process
150	* memory. We would have to change the protection of the whole chunk for
151	* every call to RTHeapSimple, which would be rather expensive.
152	*
153	* This alternative implemenation let restrict page protection modifications
154	* to the pages backing the executable memory we just allocated.
155	*/
156	#define IEMEXECMEM_USE_ALT_SUB_ALLOCATOR
157	/** The chunk sub-allocation unit size in bytes. */
158	#define IEMEXECMEM_ALT_SUB_ALLOC_UNIT_SIZE 128
159	/** The chunk sub-allocation unit size as a shift factor. */
160	#define IEMEXECMEM_ALT_SUB_ALLOC_UNIT_SHIFT 7
161
162	#if defined(IN_RING3) && !defined(RT_OS_WINDOWS)
163	# ifdef IEMNATIVE_USE_GDB_JIT
164	# define IEMNATIVE_USE_GDB_JIT_ET_DYN
165
166	/** GDB JIT: Code entry. */
167	typedef struct GDBJITCODEENTRY
168	{
169	struct GDBJITCODEENTRY *pNext;
170	struct GDBJITCODEENTRY *pPrev;
171	uint8_t *pbSymFile;
172	uint64_t cbSymFile;
173	} GDBJITCODEENTRY;
174
175	/** GDB JIT: Actions. */
176	typedef enum GDBJITACTIONS : uint32_t
177	{
178	kGdbJitaction_NoAction = 0, kGdbJitaction_Register, kGdbJitaction_Unregister
179	} GDBJITACTIONS;
180
181	/** GDB JIT: Descriptor. */
182	typedef struct GDBJITDESCRIPTOR
183	{
184	uint32_t uVersion;
185	GDBJITACTIONS enmAction;
186	GDBJITCODEENTRY *pRelevant;
187	GDBJITCODEENTRY *pHead;
188	/** Our addition: */
189	GDBJITCODEENTRY *pTail;
190	} GDBJITDESCRIPTOR;
191
192	/** GDB JIT: Our simple symbol file data. */
193	typedef struct GDBJITSYMFILE
194	{
195	Elf64_Ehdr EHdr;
196	# ifndef IEMNATIVE_USE_GDB_JIT_ET_DYN
197	Elf64_Shdr aShdrs[5];
198	# else
199	Elf64_Shdr aShdrs[7];
200	Elf64_Phdr aPhdrs[2];
201	# endif
202	/** The dwarf ehframe data for the chunk. */
203	uint8_t abEhFrame[512];
204	char szzStrTab[128];
205	Elf64_Sym aSymbols[3];
206	# ifdef IEMNATIVE_USE_GDB_JIT_ET_DYN
207	Elf64_Sym aDynSyms[2];
208	Elf64_Dyn aDyn[6];
209	# endif
210	} GDBJITSYMFILE;
211
212	extern "C" GDBJITDESCRIPTOR __jit_debug_descriptor;
213	extern "C" DECLEXPORT(void) __jit_debug_register_code(void);
214
215	/** Init once for g_IemNativeGdbJitLock. */
216	static RTONCE g_IemNativeGdbJitOnce = RTONCE_INITIALIZER;
217	/** Init once for the critical section. */
218	static RTCRITSECT g_IemNativeGdbJitLock;
219
220	/** GDB reads the info here. */
221	GDBJITDESCRIPTOR __jit_debug_descriptor = { 1, kGdbJitaction_NoAction, NULL, NULL };
222
223	/** GDB sets a breakpoint on this and checks __jit_debug_descriptor when hit. */
224	DECL_NO_INLINE(RT_NOTHING, DECLEXPORT(void)) __jit_debug_register_code(void)
225	{
226	ASMNopPause();
227	}
228
229	/** @callback_method_impl{FNRTONCE} */
230	static DECLCALLBACK(int32_t) iemNativeGdbJitInitOnce(void *pvUser)
231	{
232	RT_NOREF(pvUser);
233	return RTCritSectInit(&g_IemNativeGdbJitLock);
234	}
235
236
237	# endif /* IEMNATIVE_USE_GDB_JIT */
238
239	/**
240	* Per-chunk unwind info for non-windows hosts.
241	*/
242	typedef struct IEMEXECMEMCHUNKEHFRAME
243	{
244	# ifdef IEMNATIVE_USE_LIBUNWIND
245	/** The offset of the FDA into abEhFrame. */
246	uintptr_t offFda;
247	# else
248	/** 'struct object' storage area. */
249	uint8_t abObject[1024];
250	# endif
251	# ifdef IEMNATIVE_USE_GDB_JIT
252	# if 0
253	/** The GDB JIT 'symbol file' data. */
254	GDBJITSYMFILE GdbJitSymFile;
255	# endif
256	/** The GDB JIT list entry. */
257	GDBJITCODEENTRY GdbJitEntry;
258	# endif
259	/** The dwarf ehframe data for the chunk. */
260	uint8_t abEhFrame[512];
261	} IEMEXECMEMCHUNKEHFRAME;
262	/** Pointer to per-chunk info info for non-windows hosts. */
263	typedef IEMEXECMEMCHUNKEHFRAME *PIEMEXECMEMCHUNKEHFRAME;
264	#endif
265
266
267	/**
268	* An chunk of executable memory.
269	*/
270	typedef struct IEMEXECMEMCHUNK
271	{
272	#ifdef IEMEXECMEM_USE_ALT_SUB_ALLOCATOR
273	/** Number of free items in this chunk. */
274	uint32_t cFreeUnits;
275	/** Hint were to start searching for free space in the allocation bitmap. */
276	uint32_t idxFreeHint;
277	#else
278	/** The heap handle. */
279	RTHEAPSIMPLE hHeap;
280	#endif
281	/** Pointer to the chunk. */
282	void *pvChunk;
283	#ifdef IN_RING3
284	/**
285	* Pointer to the unwind information.
286	*
287	* This is used during C++ throw and longjmp (windows and probably most other
288	* platforms). Some debuggers (windbg) makes use of it as well.
289	*
290	* Windows: This is allocated from hHeap on windows because (at least for
291	* AMD64) the UNWIND_INFO structure address in the
292	* RUNTIME_FUNCTION entry is an RVA and the chunk is the "image".
293	*
294	* Others: Allocated from the regular heap to avoid unnecessary executable data
295	* structures. This points to an IEMEXECMEMCHUNKEHFRAME structure. */
296	void *pvUnwindInfo;
297	#elif defined(IN_RING0)
298	/** Allocation handle. */
299	RTR0MEMOBJ hMemObj;
300	#endif
301	} IEMEXECMEMCHUNK;
302	/** Pointer to a memory chunk. */
303	typedef IEMEXECMEMCHUNK *PIEMEXECMEMCHUNK;
304
305
306	/**
307	* Executable memory allocator for the native recompiler.
308	*/
309	typedef struct IEMEXECMEMALLOCATOR
310	{
311	/** Magic value (IEMEXECMEMALLOCATOR_MAGIC). */
312	uint32_t uMagic;
313
314	/** The chunk size. */
315	uint32_t cbChunk;
316	/** The maximum number of chunks. */
317	uint32_t cMaxChunks;
318	/** The current number of chunks. */
319	uint32_t cChunks;
320	/** Hint where to start looking for available memory. */
321	uint32_t idxChunkHint;
322	/** Statistics: Current number of allocations. */
323	uint32_t cAllocations;
324
325	/** The total amount of memory available. */
326	uint64_t cbTotal;
327	/** Total amount of free memory. */
328	uint64_t cbFree;
329	/** Total amount of memory allocated. */
330	uint64_t cbAllocated;
331
332	#ifdef IEMEXECMEM_USE_ALT_SUB_ALLOCATOR
333	/** Pointer to the allocation bitmaps for all the chunks (follows aChunks).
334	*
335	* Since the chunk size is a power of two and the minimum chunk size is a lot
336	* higher than the IEMEXECMEM_ALT_SUB_ALLOC_UNIT_SIZE, each chunk will always
337	* require a whole number of uint64_t elements in the allocation bitmap. So,
338	* for sake of simplicity, they are allocated as one continous chunk for
339	* simplicity/laziness. */
340	uint64_t *pbmAlloc;
341	/** Number of units (IEMEXECMEM_ALT_SUB_ALLOC_UNIT_SIZE) per chunk. */
342	uint32_t cUnitsPerChunk;
343	/** Number of bitmap elements per chunk (for quickly locating the bitmap
344	* portion corresponding to an chunk). */
345	uint32_t cBitmapElementsPerChunk;
346	#else
347	/** @name Tweaks to get 64 byte aligned allocats w/o unnecessary fragmentation.
348	* @{ */
349	/** The size of the heap internal block header. This is used to adjust the
350	* request memory size to make sure there is exacly enough room for a header at
351	* the end of the blocks we allocate before the next 64 byte alignment line. */
352	uint32_t cbHeapBlockHdr;
353	/** The size of initial heap allocation required make sure the first
354	* allocation is correctly aligned. */
355	uint32_t cbHeapAlignTweak;
356	/** The alignment tweak allocation address. */
357	void *pvAlignTweak;
358	/** @} */
359	#endif
360
361	#if defined(IN_RING3) && !defined(RT_OS_WINDOWS)
362	/** Pointer to the array of unwind info running parallel to aChunks (same
363	* allocation as this structure, located after the bitmaps).
364	* (For Windows, the structures must reside in 32-bit RVA distance to the
365	* actual chunk, so they are allocated off the chunk.) */
366	PIEMEXECMEMCHUNKEHFRAME paEhFrames;
367	#endif
368
369	/** The allocation chunks. */
370	RT_FLEXIBLE_ARRAY_EXTENSION
371	IEMEXECMEMCHUNK aChunks[RT_FLEXIBLE_ARRAY];
372	} IEMEXECMEMALLOCATOR;
373	/** Pointer to an executable memory allocator. */
374	typedef IEMEXECMEMALLOCATOR *PIEMEXECMEMALLOCATOR;
375
376	/** Magic value for IEMEXECMEMALLOCATOR::uMagic (Scott Frederick Turow). */
377	#define IEMEXECMEMALLOCATOR_MAGIC UINT32_C(0x19490412)
378
379
380	static int iemExecMemAllocatorGrow(PVMCPUCC pVCpu, PIEMEXECMEMALLOCATOR pExecMemAllocator);
381
382
383	/**
384	* Worker for iemExecMemAllocatorAlloc that returns @a pvRet after updating
385	* the heap statistics.
386	*/
387	static void * iemExecMemAllocatorAllocTailCode(PIEMEXECMEMALLOCATOR pExecMemAllocator, void *pvRet,
388	uint32_t cbReq, uint32_t idxChunk)
389	{
390	pExecMemAllocator->cAllocations += 1;
391	pExecMemAllocator->cbAllocated += cbReq;
392	#ifdef IEMEXECMEM_USE_ALT_SUB_ALLOCATOR
393	pExecMemAllocator->cbFree -= cbReq;
394	#else
395	pExecMemAllocator->cbFree -= RT_ALIGN_32(cbReq, 64);
396	#endif
397	pExecMemAllocator->idxChunkHint = idxChunk;
398
399	#ifdef RT_OS_DARWIN
400	/*
401	* Sucks, but RTMEM_PROT_EXEC and RTMEM_PROT_WRITE are mutually exclusive
402	* on darwin. So, we mark the pages returned as read+write after alloc and
403	* expect the caller to call iemExecMemAllocatorReadyForUse when done
404	* writing to the allocation.
405	*
406	* See also https://developer.apple.com/documentation/apple-silicon/porting-just-in-time-compilers-to-apple-silicon
407	* for details.
408	*/
409	/** @todo detect if this is necessary... it wasn't required on 10.15 or
410	* whatever older version it was. */
411	int rc = RTMemProtect(pvRet, cbReq, RTMEM_PROT_WRITE \| RTMEM_PROT_READ);
412	AssertRC(rc);
413	#endif
414
415	return pvRet;
416	}
417
418
419	#ifdef IEMEXECMEM_USE_ALT_SUB_ALLOCATOR
420	static void iemExecMemAllocatorAllocInChunkInt(PIEMEXECMEMALLOCATOR pExecMemAllocator, uint64_t pbmAlloc, uint32_t idxFirst,
421	uint32_t cToScan, uint32_t cReqUnits, uint32_t idxChunk)
422	{
423	/*
424	* Shift the bitmap to the idxFirst bit so we can use ASMBitFirstClear.
425	*/
426	Assert(!(cToScan & 63));
427	Assert(!(idxFirst & 63));
428	Assert(cToScan + idxFirst <= pExecMemAllocator->cUnitsPerChunk);
429	pbmAlloc += idxFirst / 64;
430
431	/*
432	* Scan the bitmap for cReqUnits of consequtive clear bits
433	*/
434	/** @todo This can probably be done more efficiently for non-x86 systems. */
435	int iBit = ASMBitFirstClear(pbmAlloc, cToScan);
436	while (iBit >= 0 && (uint32_t)iBit <= cToScan - cReqUnits)
437	{
438	uint32_t idxAddBit = 1;
439	while (idxAddBit < cReqUnits && !ASMBitTest(pbmAlloc, (uint32_t)iBit + idxAddBit))
440	idxAddBit++;
441	if (idxAddBit >= cReqUnits)
442	{
443	ASMBitSetRange(pbmAlloc, (uint32_t)iBit, (uint32_t)iBit + cReqUnits);
444
445	PIEMEXECMEMCHUNK const pChunk = &pExecMemAllocator->aChunks[idxChunk];
446	pChunk->cFreeUnits -= cReqUnits;
447	pChunk->idxFreeHint = (uint32_t)iBit + cReqUnits;
448
449	void * const pvRet = (uint8_t *)pChunk->pvChunk
450	+ ((idxFirst + (uint32_t)iBit) << IEMEXECMEM_ALT_SUB_ALLOC_UNIT_SHIFT);
451
452	return iemExecMemAllocatorAllocTailCode(pExecMemAllocator, pvRet,
453	cReqUnits << IEMEXECMEM_ALT_SUB_ALLOC_UNIT_SHIFT, idxChunk);
454	}
455
456	iBit = ASMBitNextClear(pbmAlloc, cToScan, iBit + idxAddBit - 1);
457	}
458	return NULL;
459	}
460	#endif /* IEMEXECMEM_USE_ALT_SUB_ALLOCATOR */
461
462
463	static void *iemExecMemAllocatorAllocInChunk(PIEMEXECMEMALLOCATOR pExecMemAllocator, uint32_t idxChunk, uint32_t cbReq)
464	{
465	#ifdef IEMEXECMEM_USE_ALT_SUB_ALLOCATOR
466	/*
467	* Figure out how much to allocate.
468	*/
469	uint32_t const cReqUnits = (cbReq + IEMEXECMEM_ALT_SUB_ALLOC_UNIT_SIZE - 1) >> IEMEXECMEM_ALT_SUB_ALLOC_UNIT_SHIFT;
470	if (cReqUnits <= pExecMemAllocator->aChunks[idxChunk].cFreeUnits)
471	{
472	uint64_t * const pbmAlloc = &pExecMemAllocator->pbmAlloc[pExecMemAllocator->cBitmapElementsPerChunk * idxChunk];
473	uint32_t const idxHint = pExecMemAllocator->aChunks[idxChunk].idxFreeHint & ~(uint32_t)63;
474	if (idxHint + cReqUnits <= pExecMemAllocator->cUnitsPerChunk)
475	{
476	void *pvRet = iemExecMemAllocatorAllocInChunkInt(pExecMemAllocator, pbmAlloc, idxHint,
477	pExecMemAllocator->cUnitsPerChunk - idxHint, cReqUnits, idxChunk);
478	if (pvRet)
479	return pvRet;
480	}
481	return iemExecMemAllocatorAllocInChunkInt(pExecMemAllocator, pbmAlloc, 0,
482	RT_MIN(pExecMemAllocator->cUnitsPerChunk, RT_ALIGN_32(idxHint + cReqUnits, 64)),
483	cReqUnits, idxChunk);
484	}
485	#else
486	void *pvRet = RTHeapSimpleAlloc(pExecMemAllocator->aChunks[idxChunk].hHeap, cbReq, 32);
487	if (pvRet)
488	return iemExecMemAllocatorAllocTailCode(pExecMemAllocator, pvRet, cbReq, idxChunk);
489	#endif
490	return NULL;
491
492	}
493
494
495	/**
496	* Allocates @a cbReq bytes of executable memory.
497	*
498	* @returns Pointer to the memory, NULL if out of memory or other problem
499	* encountered.
500	* @param pVCpu The cross context virtual CPU structure of the calling
501	* thread.
502	* @param cbReq How many bytes are required.
503	*/
504	static void *iemExecMemAllocatorAlloc(PVMCPU pVCpu, uint32_t cbReq)
505	{
506	PIEMEXECMEMALLOCATOR pExecMemAllocator = pVCpu->iem.s.pExecMemAllocatorR3;
507	AssertReturn(pExecMemAllocator && pExecMemAllocator->uMagic == IEMEXECMEMALLOCATOR_MAGIC, NULL);
508	AssertMsgReturn(cbReq > 32 && cbReq < _512K, ("%#x\n", cbReq), NULL);
509
510
511	for (unsigned iIteration = 0;; iIteration++)
512	{
513	/*
514	* Adjust the request size so it'll fit the allocator alignment/whatnot.
515	*
516	* For the RTHeapSimple allocator this means to follow the logic described
517	* in iemExecMemAllocatorGrow and attempt to allocate it from one of the
518	* existing chunks if we think we've got sufficient free memory around.
519	*
520	* While for the alternative one we just align it up to a whole unit size.
521	*/
522	#ifdef IEMEXECMEM_USE_ALT_SUB_ALLOCATOR
523	cbReq = RT_ALIGN_32(cbReq, IEMEXECMEM_ALT_SUB_ALLOC_UNIT_SIZE);
524	#else
525	cbReq = RT_ALIGN_32(cbReq + pExecMemAllocator->cbHeapBlockHdr, 64) - pExecMemAllocator->cbHeapBlockHdr;
526	#endif
527	if (cbReq <= pExecMemAllocator->cbFree)
528	{
529	uint32_t const cChunks = pExecMemAllocator->cChunks;
530	uint32_t const idxChunkHint = pExecMemAllocator->idxChunkHint < cChunks ? pExecMemAllocator->idxChunkHint : 0;
531	for (uint32_t idxChunk = idxChunkHint; idxChunk < cChunks; idxChunk++)
532	{
533	void *pvRet = iemExecMemAllocatorAllocInChunk(pExecMemAllocator, idxChunk, cbReq);
534	if (pvRet)
535	return pvRet;
536	}
537	for (uint32_t idxChunk = 0; idxChunk < idxChunkHint; idxChunk++)
538	{
539	void *pvRet = iemExecMemAllocatorAllocInChunk(pExecMemAllocator, idxChunk, cbReq);
540	if (pvRet)
541	return pvRet;
542	}
543	}
544
545	/*
546	* Can we grow it with another chunk?
547	*/
548	if (pExecMemAllocator->cChunks < pExecMemAllocator->cMaxChunks)
549	{
550	int rc = iemExecMemAllocatorGrow(pVCpu, pExecMemAllocator);
551	AssertLogRelRCReturn(rc, NULL);
552
553	uint32_t const idxChunk = pExecMemAllocator->cChunks - 1;
554	void *pvRet = iemExecMemAllocatorAllocInChunk(pExecMemAllocator, idxChunk, cbReq);
555	if (pvRet)
556	return pvRet;
557	AssertFailed();
558	}
559
560	/*
561	* Try prune native TBs once.
562	*/
563	if (iIteration == 0)
564	iemTbAllocatorFreeupNativeSpace(pVCpu, cbReq / sizeof(IEMNATIVEINSTR));
565	else
566	{
567	/** @todo stats... */
568	return NULL;
569	}
570	}
571
572	}
573
574
575	/** This is a hook that we may need later for changing memory protection back
576	* to readonly+exec */
577	static void iemExecMemAllocatorReadyForUse(PVMCPUCC pVCpu, void *pv, size_t cb)
578	{
579	#ifdef RT_OS_DARWIN
580	/* See iemExecMemAllocatorAllocTailCode for the explanation. */
581	int rc = RTMemProtect(pv, cb, RTMEM_PROT_EXEC \| RTMEM_PROT_READ);
582	AssertRC(rc); RT_NOREF(pVCpu);
583
584	/*
585	* Flush the instruction cache:
586	* https://developer.apple.com/documentation/apple-silicon/porting-just-in-time-compilers-to-apple-silicon
587	*/
588	/* sys_dcache_flush(pv, cb); - not necessary */
589	sys_icache_invalidate(pv, cb);
590	#else
591	RT_NOREF(pVCpu, pv, cb);
592	#endif
593	}
594
595
596	/**
597	* Frees executable memory.
598	*/
599	void iemExecMemAllocatorFree(PVMCPU pVCpu, void *pv, size_t cb)
600	{
601	PIEMEXECMEMALLOCATOR pExecMemAllocator = pVCpu->iem.s.pExecMemAllocatorR3;
602	Assert(pExecMemAllocator && pExecMemAllocator->uMagic == IEMEXECMEMALLOCATOR_MAGIC);
603	Assert(pv);
604	#ifdef IEMEXECMEM_USE_ALT_SUB_ALLOCATOR
605	Assert(!((uintptr_t)pv & (IEMEXECMEM_ALT_SUB_ALLOC_UNIT_SIZE - 1)));
606	#else
607	Assert(!((uintptr_t)pv & 63));
608	#endif
609
610	/* Align the size as we did when allocating the block. */
611	#ifdef IEMEXECMEM_USE_ALT_SUB_ALLOCATOR
612	cb = RT_ALIGN_Z(cb, IEMEXECMEM_ALT_SUB_ALLOC_UNIT_SIZE);
613	#else
614	cb = RT_ALIGN_Z(cb + pExecMemAllocator->cbHeapBlockHdr, 64) - pExecMemAllocator->cbHeapBlockHdr;
615	#endif
616
617	/* Free it / assert sanity. */
618	#if defined(VBOX_STRICT) \|\| defined(IEMEXECMEM_USE_ALT_SUB_ALLOCATOR)
619	uint32_t const cChunks = pExecMemAllocator->cChunks;
620	uint32_t const cbChunk = pExecMemAllocator->cbChunk;
621	bool fFound = false;
622	for (uint32_t idxChunk = 0; idxChunk < cChunks; idxChunk++)
623	{
624	uintptr_t const offChunk = (uintptr_t)pv - (uintptr_t)pExecMemAllocator->aChunks[idxChunk].pvChunk;
625	fFound = offChunk < cbChunk;
626	if (fFound)
627	{
628	#ifdef IEMEXECMEM_USE_ALT_SUB_ALLOCATOR
629	uint32_t const idxFirst = (uint32_t)offChunk >> IEMEXECMEM_ALT_SUB_ALLOC_UNIT_SHIFT;
630	uint32_t const cReqUnits = (uint32_t)cb >> IEMEXECMEM_ALT_SUB_ALLOC_UNIT_SHIFT;
631
632	/* Check that it's valid and free it. */
633	uint64_t * const pbmAlloc = &pExecMemAllocator->pbmAlloc[pExecMemAllocator->cBitmapElementsPerChunk * idxChunk];
634	AssertReturnVoid(ASMBitTest(pbmAlloc, idxFirst));
635	for (uint32_t i = 1; i < cReqUnits; i++)
636	AssertReturnVoid(ASMBitTest(pbmAlloc, idxFirst + i));
637	ASMBitClearRange(pbmAlloc, idxFirst, idxFirst + cReqUnits);
638
639	pExecMemAllocator->aChunks[idxChunk].cFreeUnits += cReqUnits;
640	pExecMemAllocator->aChunks[idxChunk].idxFreeHint = idxFirst;
641
642	/* Update the stats. */
643	pExecMemAllocator->cbAllocated -= cb;
644	pExecMemAllocator->cbFree += cb;
645	pExecMemAllocator->cAllocations -= 1;
646	return;
647	#else
648	Assert(RTHeapSimpleSize(pExecMemAllocator->aChunks[idxChunk].hHeap, pv) == cb);
649	break;
650	#endif
651	}
652	}
653	# ifdef IEMEXECMEM_USE_ALT_SUB_ALLOCATOR
654	AssertFailed();
655	# else
656	Assert(fFound);
657	# endif
658	#endif
659
660	#ifdef IEMEXECMEM_USE_ALT_SUB_ALLOCATOR
661	/* Update stats while cb is freshly calculated.*/
662	pExecMemAllocator->cbAllocated -= cb;
663	pExecMemAllocator->cbFree += RT_ALIGN_Z(cb, 64);
664	pExecMemAllocator->cAllocations -= 1;
665
666	/* Free it. */
667	RTHeapSimpleFree(NIL_RTHEAPSIMPLE, pv);
668	#endif
669	}
670
671
672
673	#ifdef IN_RING3
674	# ifdef RT_OS_WINDOWS
675
676	/**
677	* Initializes the unwind info structures for windows hosts.
678	*/
679	static int
680	iemExecMemAllocatorInitAndRegisterUnwindInfoForChunk(PVMCPUCC pVCpu, PIEMEXECMEMALLOCATOR pExecMemAllocator,
681	void *pvChunk, uint32_t idxChunk)
682	{
683	RT_NOREF(pVCpu);
684
685	/*
686	* The AMD64 unwind opcodes.
687	*
688	* This is a program that starts with RSP after a RET instruction that
689	* ends up in recompiled code, and the operations we describe here will
690	* restore all non-volatile registers and bring RSP back to where our
691	* RET address is. This means it's reverse order from what happens in
692	* the prologue.
693	*
694	* Note! Using a frame register approach here both because we have one
695	* and but mainly because the UWOP_ALLOC_LARGE argument values
696	* would be a pain to write initializers for. On the positive
697	* side, we're impervious to changes in the the stack variable
698	* area can can deal with dynamic stack allocations if necessary.
699	*/
700	static const IMAGE_UNWIND_CODE s_aOpcodes[] =
701	{
702	{ { 16, IMAGE_AMD64_UWOP_SET_FPREG, 0 } }, /* RSP = RBP - FrameOffset * 10 (0x60) */
703	{ { 16, IMAGE_AMD64_UWOP_ALLOC_SMALL, 0 } }, /* RSP += 8; */
704	{ { 14, IMAGE_AMD64_UWOP_PUSH_NONVOL, X86_GREG_x15 } }, /* R15 = [RSP]; RSP += 8; */
705	{ { 12, IMAGE_AMD64_UWOP_PUSH_NONVOL, X86_GREG_x14 } }, /* R14 = [RSP]; RSP += 8; */
706	{ { 10, IMAGE_AMD64_UWOP_PUSH_NONVOL, X86_GREG_x13 } }, /* R13 = [RSP]; RSP += 8; */
707	{ { 8, IMAGE_AMD64_UWOP_PUSH_NONVOL, X86_GREG_x12 } }, /* R12 = [RSP]; RSP += 8; */
708	{ { 7, IMAGE_AMD64_UWOP_PUSH_NONVOL, X86_GREG_xDI } }, /* RDI = [RSP]; RSP += 8; */
709	{ { 6, IMAGE_AMD64_UWOP_PUSH_NONVOL, X86_GREG_xSI } }, /* RSI = [RSP]; RSP += 8; */
710	{ { 5, IMAGE_AMD64_UWOP_PUSH_NONVOL, X86_GREG_xBX } }, /* RBX = [RSP]; RSP += 8; */
711	{ { 4, IMAGE_AMD64_UWOP_PUSH_NONVOL, X86_GREG_xBP } }, /* RBP = [RSP]; RSP += 8; */
712	};
713	union
714	{
715	IMAGE_UNWIND_INFO Info;
716	uint8_t abPadding[RT_UOFFSETOF(IMAGE_UNWIND_INFO, aOpcodes) + 16];
717	} s_UnwindInfo =
718	{
719	{
720	/* .Version = */ 1,
721	/* .Flags = */ 0,
722	/* .SizeOfProlog = / 16, / whatever */
723	/* .CountOfCodes = */ RT_ELEMENTS(s_aOpcodes),
724	/* .FrameRegister = */ X86_GREG_xBP,
725	/* .FrameOffset = / (-IEMNATIVE_FP_OFF_LAST_PUSH + 8) / 16 / we're off by one slot. sigh. */,
726	}
727	};
728	AssertCompile(-IEMNATIVE_FP_OFF_LAST_PUSH < 240 && -IEMNATIVE_FP_OFF_LAST_PUSH > 0);
729	AssertCompile((-IEMNATIVE_FP_OFF_LAST_PUSH & 0xf) == 8);
730
731	/*
732	* Calc how much space we need and allocate it off the exec heap.
733	*/
734	unsigned const cFunctionEntries = 1;
735	unsigned const cbUnwindInfo = sizeof(s_aOpcodes) + RT_UOFFSETOF(IMAGE_UNWIND_INFO, aOpcodes);
736	unsigned const cbNeeded = sizeof(IMAGE_RUNTIME_FUNCTION_ENTRY) * cFunctionEntries + cbUnwindInfo;
737	# ifdef IEMEXECMEM_USE_ALT_SUB_ALLOCATOR
738	unsigned const cbNeededAligned = RT_ALIGN_32(cbNeeded, IEMEXECMEM_ALT_SUB_ALLOC_UNIT_SIZE);
739	PIMAGE_RUNTIME_FUNCTION_ENTRY const paFunctions
740	= (PIMAGE_RUNTIME_FUNCTION_ENTRY)iemExecMemAllocatorAllocInChunk(pExecMemAllocator, idxChunk, cbNeededAligned);
741	# else
742	unsigned const cbNeededAligned = RT_ALIGN_32(cbNeeded + pExecMemAllocator->cbHeapBlockHdr, 64)
743	- pExecMemAllocator->cbHeapBlockHdr;
744	PIMAGE_RUNTIME_FUNCTION_ENTRY const paFunctions = (PIMAGE_RUNTIME_FUNCTION_ENTRY)RTHeapSimpleAlloc(hHeap, cbNeededAligned,
745	32 /cbAlignment/);
746	# endif
747	AssertReturn(paFunctions, VERR_INTERNAL_ERROR_5);
748	pExecMemAllocator->aChunks[idxChunk].pvUnwindInfo = paFunctions;
749
750	/*
751	* Initialize the structures.
752	*/
753	PIMAGE_UNWIND_INFO const pInfo = (PIMAGE_UNWIND_INFO)&paFunctions[cFunctionEntries];
754
755	paFunctions[0].BeginAddress = 0;
756	paFunctions[0].EndAddress = pExecMemAllocator->cbChunk;
757	paFunctions[0].UnwindInfoAddress = (uint32_t)((uintptr_t)pInfo - (uintptr_t)pvChunk);
758
759	memcpy(pInfo, &s_UnwindInfo, RT_UOFFSETOF(IMAGE_UNWIND_INFO, aOpcodes));
760	memcpy(&pInfo->aOpcodes[0], s_aOpcodes, sizeof(s_aOpcodes));
761
762	/*
763	* Register it.
764	*/
765	uint8_t fRet = RtlAddFunctionTable(paFunctions, cFunctionEntries, (uintptr_t)pvChunk);
766	AssertReturn(fRet, VERR_INTERNAL_ERROR_3); /* Nothing to clean up on failure, since its within the chunk itself. */
767
768	return VINF_SUCCESS;
769	}
770
771
772	# else /* !RT_OS_WINDOWS */
773
774	/**
775	* Emits a LEB128 encoded value between -0x2000 and 0x2000 (both exclusive).
776	*/
777	DECLINLINE(RTPTRUNION) iemDwarfPutLeb128(RTPTRUNION Ptr, int32_t iValue)
778	{
779	if (iValue >= 64)
780	{
781	Assert(iValue < 0x2000);
782	*Ptr.pb++ = ((uint8_t)iValue & 0x7f) \| 0x80;
783	*Ptr.pb++ = (uint8_t)(iValue >> 7) & 0x3f;
784	}
785	else if (iValue >= 0)
786	*Ptr.pb++ = (uint8_t)iValue;
787	else if (iValue > -64)
788	*Ptr.pb++ = ((uint8_t)iValue & 0x3f) \| 0x40;
789	else
790	{
791	Assert(iValue > -0x2000);
792	*Ptr.pb++ = ((uint8_t)iValue & 0x7f) \| 0x80;
793	*Ptr.pb++ = ((uint8_t)(iValue >> 7) & 0x3f) \| 0x40;
794	}
795	return Ptr;
796	}
797
798
799	/**
800	* Emits an ULEB128 encoded value (up to 64-bit wide).
801	*/
802	DECLINLINE(RTPTRUNION) iemDwarfPutUleb128(RTPTRUNION Ptr, uint64_t uValue)
803	{
804	while (uValue >= 0x80)
805	{
806	*Ptr.pb++ = ((uint8_t)uValue & 0x7f) \| 0x80;
807	uValue >>= 7;
808	}
809	*Ptr.pb++ = (uint8_t)uValue;
810	return Ptr;
811	}
812
813
814	/**
815	* Emits a CFA rule as register @a uReg + offset @a off.
816	*/
817	DECLINLINE(RTPTRUNION) iemDwarfPutCfaDefCfa(RTPTRUNION Ptr, uint32_t uReg, uint32_t off)
818	{
819	*Ptr.pb++ = DW_CFA_def_cfa;
820	Ptr = iemDwarfPutUleb128(Ptr, uReg);
821	Ptr = iemDwarfPutUleb128(Ptr, off);
822	return Ptr;
823	}
824
825
826	/**
827	* Emits a register (@a uReg) save location:
828	* CFA + @a off * data_alignment_factor
829	*/
830	DECLINLINE(RTPTRUNION) iemDwarfPutCfaOffset(RTPTRUNION Ptr, uint32_t uReg, uint32_t off)
831	{
832	if (uReg < 0x40)
833	*Ptr.pb++ = DW_CFA_offset \| uReg;
834	else
835	{
836	*Ptr.pb++ = DW_CFA_offset_extended;
837	Ptr = iemDwarfPutUleb128(Ptr, uReg);
838	}
839	Ptr = iemDwarfPutUleb128(Ptr, off);
840	return Ptr;
841	}
842
843
844	# if 0 /* unused */
845	/**
846	* Emits a register (@a uReg) save location, using signed offset:
847	* CFA + @a offSigned * data_alignment_factor
848	*/
849	DECLINLINE(RTPTRUNION) iemDwarfPutCfaSignedOffset(RTPTRUNION Ptr, uint32_t uReg, int32_t offSigned)
850	{
851	*Ptr.pb++ = DW_CFA_offset_extended_sf;
852	Ptr = iemDwarfPutUleb128(Ptr, uReg);
853	Ptr = iemDwarfPutLeb128(Ptr, offSigned);
854	return Ptr;
855	}
856	# endif
857
858
859	/**
860	* Initializes the unwind info section for non-windows hosts.
861	*/
862	static int
863	iemExecMemAllocatorInitAndRegisterUnwindInfoForChunk(PVMCPUCC pVCpu, PIEMEXECMEMALLOCATOR pExecMemAllocator,
864	void *pvChunk, uint32_t idxChunk)
865	{
866	PIEMEXECMEMCHUNKEHFRAME const pEhFrame = &pExecMemAllocator->paEhFrames[idxChunk];
867	pExecMemAllocator->aChunks[idxChunk].pvUnwindInfo = pEhFrame; /* not necessary, but whatever */
868
869	RTPTRUNION Ptr = { pEhFrame->abEhFrame };
870
871	/*
872	* Generate the CIE first.
873	*/
874	# ifdef IEMNATIVE_USE_LIBUNWIND /* libunwind (llvm, darwin) only supports v1 and v3. */
875	uint8_t const iDwarfVer = 3;
876	# else
877	uint8_t const iDwarfVer = 4;
878	# endif
879	RTPTRUNION const PtrCie = Ptr;
880	Ptr.pu32++ = 123; / The CIE length will be determined later. */
881	Ptr.pu32++ = 0 /UINT32_MAX/; / I'm a CIE in .eh_frame speak. */
882	Ptr.pb++ = iDwarfVer; / DwARF version */
883	Ptr.pb++ = 0; / Augmentation. */
884	if (iDwarfVer >= 4)
885	{
886	Ptr.pb++ = sizeof(uintptr_t); / Address size. */
887	Ptr.pb++ = 0; / Segment selector size. */
888	}
889	# ifdef RT_ARCH_AMD64
890	Ptr = iemDwarfPutLeb128(Ptr, 1); /* Code alignment factor (LEB128 = 1). */
891	# else
892	Ptr = iemDwarfPutLeb128(Ptr, 4); /* Code alignment factor (LEB128 = 4). */
893	# endif
894	Ptr = iemDwarfPutLeb128(Ptr, -8); /* Data alignment factor (LEB128 = -8). */
895	# ifdef RT_ARCH_AMD64
896	Ptr = iemDwarfPutUleb128(Ptr, DWREG_AMD64_RA); /* Return address column (ULEB128) */
897	# elif defined(RT_ARCH_ARM64)
898	Ptr = iemDwarfPutUleb128(Ptr, DWREG_ARM64_LR); /* Return address column (ULEB128) */
899	# else
900	# error "port me"
901	# endif
902	/* Initial instructions: */
903	# ifdef RT_ARCH_AMD64
904	Ptr = iemDwarfPutCfaDefCfa(Ptr, DWREG_AMD64_RBP, 16); /* CFA = RBP + 0x10 - first stack parameter */
905	Ptr = iemDwarfPutCfaOffset(Ptr, DWREG_AMD64_RA, 1); /* Ret RIP = [CFA + 1-8] /
906	Ptr = iemDwarfPutCfaOffset(Ptr, DWREG_AMD64_RBP, 2); /* RBP = [CFA + 2-8] /
907	Ptr = iemDwarfPutCfaOffset(Ptr, DWREG_AMD64_RBX, 3); /* RBX = [CFA + 3-8] /
908	Ptr = iemDwarfPutCfaOffset(Ptr, DWREG_AMD64_R12, 4); /* R12 = [CFA + 4-8] /
909	Ptr = iemDwarfPutCfaOffset(Ptr, DWREG_AMD64_R13, 5); /* R13 = [CFA + 5-8] /
910	Ptr = iemDwarfPutCfaOffset(Ptr, DWREG_AMD64_R14, 6); /* R14 = [CFA + 6-8] /
911	Ptr = iemDwarfPutCfaOffset(Ptr, DWREG_AMD64_R15, 7); /* R15 = [CFA + 7-8] /
912	# elif defined(RT_ARCH_ARM64)
913	# if 1
914	Ptr = iemDwarfPutCfaDefCfa(Ptr, DWREG_ARM64_BP, 16); /* CFA = BP + 0x10 - first stack parameter */
915	# else
916	Ptr = iemDwarfPutCfaDefCfa(Ptr, DWREG_ARM64_SP, IEMNATIVE_FRAME_VAR_SIZE + IEMNATIVE_FRAME_SAVE_REG_SIZE);
917	# endif
918	Ptr = iemDwarfPutCfaOffset(Ptr, DWREG_ARM64_LR, 1); /* Ret PC = [CFA + 1-8] /
919	Ptr = iemDwarfPutCfaOffset(Ptr, DWREG_ARM64_BP, 2); /* Ret BP = [CFA + 2-8] /
920	Ptr = iemDwarfPutCfaOffset(Ptr, DWREG_ARM64_X28, 3); /* X28 = [CFA + 3-8] /
921	Ptr = iemDwarfPutCfaOffset(Ptr, DWREG_ARM64_X27, 4); /* X27 = [CFA + 4-8] /
922	Ptr = iemDwarfPutCfaOffset(Ptr, DWREG_ARM64_X26, 5); /* X26 = [CFA + 5-8] /
923	Ptr = iemDwarfPutCfaOffset(Ptr, DWREG_ARM64_X25, 6); /* X25 = [CFA + 6-8] /
924	Ptr = iemDwarfPutCfaOffset(Ptr, DWREG_ARM64_X24, 7); /* X24 = [CFA + 7-8] /
925	Ptr = iemDwarfPutCfaOffset(Ptr, DWREG_ARM64_X23, 8); /* X23 = [CFA + 8-8] /
926	Ptr = iemDwarfPutCfaOffset(Ptr, DWREG_ARM64_X22, 9); /* X22 = [CFA + 9-8] /
927	Ptr = iemDwarfPutCfaOffset(Ptr, DWREG_ARM64_X21, 10); /* X21 = [CFA +10-8] /
928	Ptr = iemDwarfPutCfaOffset(Ptr, DWREG_ARM64_X20, 11); /* X20 = [CFA +11-8] /
929	Ptr = iemDwarfPutCfaOffset(Ptr, DWREG_ARM64_X19, 12); /* X19 = [CFA +12-8] /
930	AssertCompile(IEMNATIVE_FRAME_SAVE_REG_SIZE / 8 == 12);
931	/** @todo we we need to do something about clearing DWREG_ARM64_RA_SIGN_STATE or something? */
932	# else
933	# error "port me"
934	# endif
935	while ((Ptr.u - PtrCie.u) & 3)
936	*Ptr.pb++ = DW_CFA_nop;
937	/* Finalize the CIE size. */
938	*PtrCie.pu32 = Ptr.u - PtrCie.u - sizeof(uint32_t);
939
940	/*
941	* Generate an FDE for the whole chunk area.
942	*/
943	# ifdef IEMNATIVE_USE_LIBUNWIND
944	pEhFrame->offFda = Ptr.u - (uintptr_t)&pEhFrame->abEhFrame[0];
945	# endif
946	RTPTRUNION const PtrFde = Ptr;
947	Ptr.pu32++ = 123; / The CIE length will be determined later. */
948	Ptr.pu32 = Ptr.u - PtrCie.u; / Negated self relative CIE address. */
949	Ptr.pu32++;
950	Ptr.pu64++ = (uintptr_t)pvChunk; / Absolute start PC of this FDE. */
951	Ptr.pu64++ = pExecMemAllocator->cbChunk; / PC range length for this PDE. */
952	# if 0 /* not requried for recent libunwind.dylib nor recent libgcc/glib. */
953	*Ptr.pb++ = DW_CFA_nop;
954	# endif
955	while ((Ptr.u - PtrFde.u) & 3)
956	*Ptr.pb++ = DW_CFA_nop;
957	/* Finalize the FDE size. */
958	*PtrFde.pu32 = Ptr.u - PtrFde.u - sizeof(uint32_t);
959
960	/* Terminator entry. */
961	*Ptr.pu32++ = 0;
962	Ptr.pu32++ = 0; / just to be sure... */
963	Assert(Ptr.u - (uintptr_t)&pEhFrame->abEhFrame[0] <= sizeof(pEhFrame->abEhFrame));
964
965	/*
966	* Register it.
967	*/
968	# ifdef IEMNATIVE_USE_LIBUNWIND
969	__register_frame(&pEhFrame->abEhFrame[pEhFrame->offFda]);
970	# else
971	memset(pEhFrame->abObject, 0xf6, sizeof(pEhFrame->abObject)); /* color the memory to better spot usage */
972	__register_frame_info(pEhFrame->abEhFrame, pEhFrame->abObject);
973	# endif
974
975	# ifdef IEMNATIVE_USE_GDB_JIT
976	/*
977	* Now for telling GDB about this (experimental).
978	*
979	* This seems to work best with ET_DYN.
980	*/
981	unsigned const cbNeeded = sizeof(GDBJITSYMFILE);
982	# ifdef IEMEXECMEM_USE_ALT_SUB_ALLOCATOR
983	unsigned const cbNeededAligned = RT_ALIGN_32(cbNeeded, IEMEXECMEM_ALT_SUB_ALLOC_UNIT_SIZE);
984	GDBJITSYMFILE * const pSymFile = (GDBJITSYMFILE *)iemExecMemAllocatorAllocInChunk(pExecMemAllocator, idxChunk, cbNeededAligned);
985	# else
986	unsigned const cbNeededAligned = RT_ALIGN_32(cbNeeded + pExecMemAllocator->cbHeapBlockHdr, 64)
987	- pExecMemAllocator->cbHeapBlockHdr;
988	GDBJITSYMFILE * const pSymFile = (PIMAGE_RUNTIME_FUNCTION_ENTRY)RTHeapSimpleAlloc(hHeap, cbNeededAligned, 32 /cbAlignment/);
989	# endif
990	AssertReturn(pSymFile, VERR_INTERNAL_ERROR_5);
991	unsigned const offSymFileInChunk = (uintptr_t)pSymFile - (uintptr_t)pvChunk;
992
993	RT_ZERO(*pSymFile);
994
995	/*
996	* The ELF header:
997	*/
998	pSymFile->EHdr.e_ident[0] = ELFMAG0;
999	pSymFile->EHdr.e_ident[1] = ELFMAG1;
1000	pSymFile->EHdr.e_ident[2] = ELFMAG2;
1001	pSymFile->EHdr.e_ident[3] = ELFMAG3;
1002	pSymFile->EHdr.e_ident[EI_VERSION] = EV_CURRENT;
1003	pSymFile->EHdr.e_ident[EI_CLASS] = ELFCLASS64;
1004	pSymFile->EHdr.e_ident[EI_DATA] = ELFDATA2LSB;
1005	pSymFile->EHdr.e_ident[EI_OSABI] = ELFOSABI_NONE;
1006	# ifdef IEMNATIVE_USE_GDB_JIT_ET_DYN
1007	pSymFile->EHdr.e_type = ET_DYN;
1008	# else
1009	pSymFile->EHdr.e_type = ET_REL;
1010	# endif
1011	# ifdef RT_ARCH_AMD64
1012	pSymFile->EHdr.e_machine = EM_AMD64;
1013	# elif defined(RT_ARCH_ARM64)
1014	pSymFile->EHdr.e_machine = EM_AARCH64;
1015	# else
1016	# error "port me"
1017	# endif
1018	pSymFile->EHdr.e_version = 1; /?/
1019	pSymFile->EHdr.e_entry = 0;
1020	# if defined(IEMNATIVE_USE_GDB_JIT_ET_DYN)
1021	pSymFile->EHdr.e_phoff = RT_UOFFSETOF(GDBJITSYMFILE, aPhdrs);
1022	# else
1023	pSymFile->EHdr.e_phoff = 0;
1024	# endif
1025	pSymFile->EHdr.e_shoff = sizeof(pSymFile->EHdr);
1026	pSymFile->EHdr.e_flags = 0;
1027	pSymFile->EHdr.e_ehsize = sizeof(pSymFile->EHdr);
1028	# if defined(IEMNATIVE_USE_GDB_JIT_ET_DYN)
1029	pSymFile->EHdr.e_phentsize = sizeof(pSymFile->aPhdrs[0]);
1030	pSymFile->EHdr.e_phnum = RT_ELEMENTS(pSymFile->aPhdrs);
1031	# else
1032	pSymFile->EHdr.e_phentsize = 0;
1033	pSymFile->EHdr.e_phnum = 0;
1034	# endif
1035	pSymFile->EHdr.e_shentsize = sizeof(pSymFile->aShdrs[0]);
1036	pSymFile->EHdr.e_shnum = RT_ELEMENTS(pSymFile->aShdrs);
1037	pSymFile->EHdr.e_shstrndx = 0; /* set later */
1038
1039	uint32_t offStrTab = 0;
1040	#define APPEND_STR(a_szStr) do { \
1041	memcpy(&pSymFile->szzStrTab[offStrTab], a_szStr, sizeof(a_szStr)); \
1042	offStrTab += sizeof(a_szStr); \
1043	Assert(offStrTab < sizeof(pSymFile->szzStrTab)); \
1044	} while (0)
1045	#define APPEND_STR_FMT(a_szStr, ...) do { \
1046	offStrTab += RTStrPrintf(&pSymFile->szzStrTab[offStrTab], sizeof(pSymFile->szzStrTab) - offStrTab, a_szStr, __VA_ARGS__); \
1047	offStrTab++; \
1048	Assert(offStrTab < sizeof(pSymFile->szzStrTab)); \
1049	} while (0)
1050
1051	/*
1052	* Section headers.
1053	*/
1054	/* Section header #0: NULL */
1055	unsigned i = 0;
1056	APPEND_STR("");
1057	RT_ZERO(pSymFile->aShdrs[i]);
1058	i++;
1059
1060	/* Section header: .eh_frame */
1061	pSymFile->aShdrs[i].sh_name = offStrTab;
1062	APPEND_STR(".eh_frame");
1063	pSymFile->aShdrs[i].sh_type = SHT_PROGBITS;
1064	pSymFile->aShdrs[i].sh_flags = SHF_ALLOC \| SHF_EXECINSTR;
1065	# if defined(IEMNATIVE_USE_GDB_JIT_ET_DYN) \|\| defined(IEMNATIVE_USE_GDB_JIT_ELF_RVAS)
1066	pSymFile->aShdrs[i].sh_offset
1067	= pSymFile->aShdrs[i].sh_addr = RT_UOFFSETOF(GDBJITSYMFILE, abEhFrame);
1068	# else
1069	pSymFile->aShdrs[i].sh_addr = (uintptr_t)&pSymFile->abEhFrame[0];
1070	pSymFile->aShdrs[i].sh_offset = 0;
1071	# endif
1072
1073	pSymFile->aShdrs[i].sh_size = sizeof(pEhFrame->abEhFrame);
1074	pSymFile->aShdrs[i].sh_link = 0;
1075	pSymFile->aShdrs[i].sh_info = 0;
1076	pSymFile->aShdrs[i].sh_addralign = 1;
1077	pSymFile->aShdrs[i].sh_entsize = 0;
1078	memcpy(pSymFile->abEhFrame, pEhFrame->abEhFrame, sizeof(pEhFrame->abEhFrame));
1079	i++;
1080
1081	/* Section header: .shstrtab */
1082	unsigned const iShStrTab = i;
1083	pSymFile->EHdr.e_shstrndx = iShStrTab;
1084	pSymFile->aShdrs[i].sh_name = offStrTab;
1085	APPEND_STR(".shstrtab");
1086	pSymFile->aShdrs[i].sh_type = SHT_STRTAB;
1087	pSymFile->aShdrs[i].sh_flags = SHF_ALLOC;
1088	# if defined(IEMNATIVE_USE_GDB_JIT_ET_DYN) \|\| defined(IEMNATIVE_USE_GDB_JIT_ELF_RVAS)
1089	pSymFile->aShdrs[i].sh_offset
1090	= pSymFile->aShdrs[i].sh_addr = RT_UOFFSETOF(GDBJITSYMFILE, szzStrTab);
1091	# else
1092	pSymFile->aShdrs[i].sh_addr = (uintptr_t)&pSymFile->szzStrTab[0];
1093	pSymFile->aShdrs[i].sh_offset = 0;
1094	# endif
1095	pSymFile->aShdrs[i].sh_size = sizeof(pSymFile->szzStrTab);
1096	pSymFile->aShdrs[i].sh_link = 0;
1097	pSymFile->aShdrs[i].sh_info = 0;
1098	pSymFile->aShdrs[i].sh_addralign = 1;
1099	pSymFile->aShdrs[i].sh_entsize = 0;
1100	i++;
1101
1102	/* Section header: .symbols */
1103	pSymFile->aShdrs[i].sh_name = offStrTab;
1104	APPEND_STR(".symtab");
1105	pSymFile->aShdrs[i].sh_type = SHT_SYMTAB;
1106	pSymFile->aShdrs[i].sh_flags = SHF_ALLOC;
1107	pSymFile->aShdrs[i].sh_offset
1108	= pSymFile->aShdrs[i].sh_addr = RT_UOFFSETOF(GDBJITSYMFILE, aSymbols);
1109	pSymFile->aShdrs[i].sh_size = sizeof(pSymFile->aSymbols);
1110	pSymFile->aShdrs[i].sh_link = iShStrTab;
1111	pSymFile->aShdrs[i].sh_info = RT_ELEMENTS(pSymFile->aSymbols);
1112	pSymFile->aShdrs[i].sh_addralign = sizeof(pSymFile->aSymbols[0].st_value);
1113	pSymFile->aShdrs[i].sh_entsize = sizeof(pSymFile->aSymbols[0]);
1114	i++;
1115
1116	# if defined(IEMNATIVE_USE_GDB_JIT_ET_DYN)
1117	/* Section header: .symbols */
1118	pSymFile->aShdrs[i].sh_name = offStrTab;
1119	APPEND_STR(".dynsym");
1120	pSymFile->aShdrs[i].sh_type = SHT_DYNSYM;
1121	pSymFile->aShdrs[i].sh_flags = SHF_ALLOC;
1122	pSymFile->aShdrs[i].sh_offset
1123	= pSymFile->aShdrs[i].sh_addr = RT_UOFFSETOF(GDBJITSYMFILE, aDynSyms);
1124	pSymFile->aShdrs[i].sh_size = sizeof(pSymFile->aDynSyms);
1125	pSymFile->aShdrs[i].sh_link = iShStrTab;
1126	pSymFile->aShdrs[i].sh_info = RT_ELEMENTS(pSymFile->aDynSyms);
1127	pSymFile->aShdrs[i].sh_addralign = sizeof(pSymFile->aDynSyms[0].st_value);
1128	pSymFile->aShdrs[i].sh_entsize = sizeof(pSymFile->aDynSyms[0]);
1129	i++;
1130	# endif
1131
1132	# if defined(IEMNATIVE_USE_GDB_JIT_ET_DYN)
1133	/* Section header: .dynamic */
1134	pSymFile->aShdrs[i].sh_name = offStrTab;
1135	APPEND_STR(".dynamic");
1136	pSymFile->aShdrs[i].sh_type = SHT_DYNAMIC;
1137	pSymFile->aShdrs[i].sh_flags = SHF_ALLOC;
1138	pSymFile->aShdrs[i].sh_offset
1139	= pSymFile->aShdrs[i].sh_addr = RT_UOFFSETOF(GDBJITSYMFILE, aDyn);
1140	pSymFile->aShdrs[i].sh_size = sizeof(pSymFile->aDyn);
1141	pSymFile->aShdrs[i].sh_link = iShStrTab;
1142	pSymFile->aShdrs[i].sh_info = 0;
1143	pSymFile->aShdrs[i].sh_addralign = 1;
1144	pSymFile->aShdrs[i].sh_entsize = sizeof(pSymFile->aDyn[0]);
1145	i++;
1146	# endif
1147
1148	/* Section header: .text */
1149	unsigned const iShText = i;
1150	pSymFile->aShdrs[i].sh_name = offStrTab;
1151	APPEND_STR(".text");
1152	pSymFile->aShdrs[i].sh_type = SHT_PROGBITS;
1153	pSymFile->aShdrs[i].sh_flags = SHF_ALLOC \| SHF_EXECINSTR;
1154	# if defined(IEMNATIVE_USE_GDB_JIT_ET_DYN) \|\| defined(IEMNATIVE_USE_GDB_JIT_ELF_RVAS)
1155	pSymFile->aShdrs[i].sh_offset
1156	= pSymFile->aShdrs[i].sh_addr = sizeof(GDBJITSYMFILE);
1157	# else
1158	pSymFile->aShdrs[i].sh_addr = (uintptr_t)(pSymFile + 1);
1159	pSymFile->aShdrs[i].sh_offset = 0;
1160	# endif
1161	pSymFile->aShdrs[i].sh_size = pExecMemAllocator->cbChunk - offSymFileInChunk - sizeof(GDBJITSYMFILE);
1162	pSymFile->aShdrs[i].sh_link = 0;
1163	pSymFile->aShdrs[i].sh_info = 0;
1164	pSymFile->aShdrs[i].sh_addralign = 1;
1165	pSymFile->aShdrs[i].sh_entsize = 0;
1166	i++;
1167
1168	Assert(i == RT_ELEMENTS(pSymFile->aShdrs));
1169
1170	# if defined(IEMNATIVE_USE_GDB_JIT_ET_DYN)
1171	/*
1172	* The program headers:
1173	*/
1174	/* Everything in a single LOAD segment: */
1175	i = 0;
1176	pSymFile->aPhdrs[i].p_type = PT_LOAD;
1177	pSymFile->aPhdrs[i].p_flags = PF_X \| PF_R;
1178	pSymFile->aPhdrs[i].p_offset
1179	= pSymFile->aPhdrs[i].p_vaddr
1180	= pSymFile->aPhdrs[i].p_paddr = 0;
1181	pSymFile->aPhdrs[i].p_filesz /* Size of segment in file. */
1182	= pSymFile->aPhdrs[i].p_memsz = pExecMemAllocator->cbChunk - offSymFileInChunk;
1183	pSymFile->aPhdrs[i].p_align = HOST_PAGE_SIZE;
1184	i++;
1185	/* The .dynamic segment. */
1186	pSymFile->aPhdrs[i].p_type = PT_DYNAMIC;
1187	pSymFile->aPhdrs[i].p_flags = PF_R;
1188	pSymFile->aPhdrs[i].p_offset
1189	= pSymFile->aPhdrs[i].p_vaddr
1190	= pSymFile->aPhdrs[i].p_paddr = RT_UOFFSETOF(GDBJITSYMFILE, aDyn);
1191	pSymFile->aPhdrs[i].p_filesz /* Size of segment in file. */
1192	= pSymFile->aPhdrs[i].p_memsz = sizeof(pSymFile->aDyn);
1193	pSymFile->aPhdrs[i].p_align = sizeof(pSymFile->aDyn[0].d_tag);
1194	i++;
1195
1196	Assert(i == RT_ELEMENTS(pSymFile->aPhdrs));
1197
1198	/*
1199	* The dynamic section:
1200	*/
1201	i = 0;
1202	pSymFile->aDyn[i].d_tag = DT_SONAME;
1203	pSymFile->aDyn[i].d_un.d_val = offStrTab;
1204	APPEND_STR_FMT("iem-exec-chunk-%u-%u", pVCpu->idCpu, idxChunk);
1205	i++;
1206	pSymFile->aDyn[i].d_tag = DT_STRTAB;
1207	pSymFile->aDyn[i].d_un.d_ptr = RT_UOFFSETOF(GDBJITSYMFILE, szzStrTab);
1208	i++;
1209	pSymFile->aDyn[i].d_tag = DT_STRSZ;
1210	pSymFile->aDyn[i].d_un.d_val = sizeof(pSymFile->szzStrTab);
1211	i++;
1212	pSymFile->aDyn[i].d_tag = DT_SYMTAB;
1213	pSymFile->aDyn[i].d_un.d_ptr = RT_UOFFSETOF(GDBJITSYMFILE, aDynSyms);
1214	i++;
1215	pSymFile->aDyn[i].d_tag = DT_SYMENT;
1216	pSymFile->aDyn[i].d_un.d_val = sizeof(pSymFile->aDynSyms[0]);
1217	i++;
1218	pSymFile->aDyn[i].d_tag = DT_NULL;
1219	i++;
1220	Assert(i == RT_ELEMENTS(pSymFile->aDyn));
1221	# endif /* IEMNATIVE_USE_GDB_JIT_ET_DYN */
1222
1223	/*
1224	* Symbol tables:
1225	*/
1226	/** @todo gdb doesn't seem to really like this ... */
1227	i = 0;
1228	pSymFile->aSymbols[i].st_name = 0;
1229	pSymFile->aSymbols[i].st_shndx = SHN_UNDEF;
1230	pSymFile->aSymbols[i].st_value = 0;
1231	pSymFile->aSymbols[i].st_size = 0;
1232	pSymFile->aSymbols[i].st_info = ELF64_ST_INFO(STB_LOCAL, STT_NOTYPE);
1233	pSymFile->aSymbols[i].st_other = 0 /* STV_DEFAULT */;
1234	# ifdef IEMNATIVE_USE_GDB_JIT_ET_DYN
1235	pSymFile->aDynSyms[0] = pSymFile->aSymbols[i];
1236	# endif
1237	i++;
1238
1239	pSymFile->aSymbols[i].st_name = 0;
1240	pSymFile->aSymbols[i].st_shndx = SHN_ABS;
1241	pSymFile->aSymbols[i].st_value = 0;
1242	pSymFile->aSymbols[i].st_size = 0;
1243	pSymFile->aSymbols[i].st_info = ELF64_ST_INFO(STB_LOCAL, STT_FILE);
1244	pSymFile->aSymbols[i].st_other = 0 /* STV_DEFAULT */;
1245	i++;
1246
1247	pSymFile->aSymbols[i].st_name = offStrTab;
1248	APPEND_STR_FMT("iem_exec_chunk_%u_%u", pVCpu->idCpu, idxChunk);
1249	# if 0
1250	pSymFile->aSymbols[i].st_shndx = iShText;
1251	pSymFile->aSymbols[i].st_value = 0;
1252	# else
1253	pSymFile->aSymbols[i].st_shndx = SHN_ABS;
1254	pSymFile->aSymbols[i].st_value = (uintptr_t)(pSymFile + 1);
1255	# endif
1256	pSymFile->aSymbols[i].st_size = pSymFile->aShdrs[iShText].sh_size;
1257	pSymFile->aSymbols[i].st_info = ELF64_ST_INFO(STB_GLOBAL, STT_FUNC);
1258	pSymFile->aSymbols[i].st_other = 0 /* STV_DEFAULT */;
1259	# ifdef IEMNATIVE_USE_GDB_JIT_ET_DYN
1260	pSymFile->aDynSyms[1] = pSymFile->aSymbols[i];
1261	pSymFile->aDynSyms[1].st_value = (uintptr_t)(pSymFile + 1);
1262	# endif
1263	i++;
1264
1265	Assert(i == RT_ELEMENTS(pSymFile->aSymbols));
1266	Assert(offStrTab < sizeof(pSymFile->szzStrTab));
1267
1268	/*
1269	* The GDB JIT entry and informing GDB.
1270	*/
1271	pEhFrame->GdbJitEntry.pbSymFile = (uint8_t *)pSymFile;
1272	# if 1
1273	pEhFrame->GdbJitEntry.cbSymFile = pExecMemAllocator->cbChunk - ((uintptr_t)pSymFile - (uintptr_t)pvChunk);
1274	# else
1275	pEhFrame->GdbJitEntry.cbSymFile = sizeof(GDBJITSYMFILE);
1276	# endif
1277
1278	RTOnce(&g_IemNativeGdbJitOnce, iemNativeGdbJitInitOnce, NULL);
1279	RTCritSectEnter(&g_IemNativeGdbJitLock);
1280	pEhFrame->GdbJitEntry.pNext = NULL;
1281	pEhFrame->GdbJitEntry.pPrev = __jit_debug_descriptor.pTail;
1282	if (__jit_debug_descriptor.pTail)
1283	__jit_debug_descriptor.pTail->pNext = &pEhFrame->GdbJitEntry;
1284	else
1285	__jit_debug_descriptor.pHead = &pEhFrame->GdbJitEntry;
1286	__jit_debug_descriptor.pTail = &pEhFrame->GdbJitEntry;
1287	__jit_debug_descriptor.pRelevant = &pEhFrame->GdbJitEntry;
1288
1289	/* Notify GDB: */
1290	__jit_debug_descriptor.enmAction = kGdbJitaction_Register;
1291	__jit_debug_register_code();
1292	__jit_debug_descriptor.enmAction = kGdbJitaction_NoAction;
1293	RTCritSectLeave(&g_IemNativeGdbJitLock);
1294
1295	# else /* !IEMNATIVE_USE_GDB_JIT */
1296	RT_NOREF(pVCpu);
1297	# endif /* !IEMNATIVE_USE_GDB_JIT */
1298
1299	return VINF_SUCCESS;
1300	}
1301
1302	# endif /* !RT_OS_WINDOWS */
1303	#endif /* IN_RING3 */
1304
1305
1306	/**
1307	* Adds another chunk to the executable memory allocator.
1308	*
1309	* This is used by the init code for the initial allocation and later by the
1310	* regular allocator function when it's out of memory.
1311	*/
1312	static int iemExecMemAllocatorGrow(PVMCPUCC pVCpu, PIEMEXECMEMALLOCATOR pExecMemAllocator)
1313	{
1314	/* Check that we've room for growth. */
1315	uint32_t const idxChunk = pExecMemAllocator->cChunks;
1316	AssertLogRelReturn(idxChunk < pExecMemAllocator->cMaxChunks, VERR_OUT_OF_RESOURCES);
1317
1318	/* Allocate a chunk. */
1319	#ifdef RT_OS_DARWIN
1320	void *pvChunk = RTMemPageAllocEx(pExecMemAllocator->cbChunk, 0);
1321	#else
1322	void *pvChunk = RTMemPageAllocEx(pExecMemAllocator->cbChunk, RTMEMPAGEALLOC_F_EXECUTABLE);
1323	#endif
1324	AssertLogRelReturn(pvChunk, VERR_NO_EXEC_MEMORY);
1325
1326	#ifdef IEMEXECMEM_USE_ALT_SUB_ALLOCATOR
1327	int rc = VINF_SUCCESS;
1328	#else
1329	/* Initialize the heap for the chunk. */
1330	RTHEAPSIMPLE hHeap = NIL_RTHEAPSIMPLE;
1331	int rc = RTHeapSimpleInit(&hHeap, pvChunk, pExecMemAllocator->cbChunk);
1332	AssertRC(rc);
1333	if (RT_SUCCESS(rc))
1334	{
1335	/*
1336	* We want the memory to be aligned on 64 byte, so the first time thru
1337	* here we do some exploratory allocations to see how we can achieve this.
1338	* On subsequent runs we only make an initial adjustment allocation, if
1339	* necessary.
1340	*
1341	* Since we own the heap implementation, we know that the internal block
1342	* header is 32 bytes in size for 64-bit systems (see RTHEAPSIMPLEBLOCK),
1343	* so all we need to wrt allocation size adjustments is to add 32 bytes
1344	* to the size, align up by 64 bytes, and subtract 32 bytes.
1345	*
1346	* The heap anchor block is 8 * sizeof(void *) (see RTHEAPSIMPLEINTERNAL),
1347	* which mean 64 bytes on a 64-bit system, so we need to make a 64 byte
1348	* allocation to force subsequent allocations to return 64 byte aligned
1349	* user areas.
1350	*/
1351	if (!pExecMemAllocator->cbHeapBlockHdr)
1352	{
1353	pExecMemAllocator->cbHeapBlockHdr = sizeof(void ) 4; /* See RTHEAPSIMPLEBLOCK. */
1354	pExecMemAllocator->cbHeapAlignTweak = 64;
1355	pExecMemAllocator->pvAlignTweak = RTHeapSimpleAlloc(hHeap, pExecMemAllocator->cbHeapAlignTweak,
1356	32 /cbAlignment/);
1357	AssertStmt(pExecMemAllocator->pvAlignTweak, rc = VERR_INTERNAL_ERROR_2);
1358
1359	void *pvTest1 = RTHeapSimpleAlloc(hHeap,
1360	RT_ALIGN_32(256 + pExecMemAllocator->cbHeapBlockHdr, 64)
1361	- pExecMemAllocator->cbHeapBlockHdr, 32 /cbAlignment/);
1362	AssertStmt(pvTest1, rc = VERR_INTERNAL_ERROR_2);
1363	AssertStmt(!((uintptr_t)pvTest1 & 63), rc = VERR_INTERNAL_ERROR_3);
1364
1365	void *pvTest2 = RTHeapSimpleAlloc(hHeap,
1366	RT_ALIGN_32(687 + pExecMemAllocator->cbHeapBlockHdr, 64)
1367	- pExecMemAllocator->cbHeapBlockHdr, 32 /cbAlignment/);
1368	AssertStmt(pvTest2, rc = VERR_INTERNAL_ERROR_2);
1369	AssertStmt(!((uintptr_t)pvTest2 & 63), rc = VERR_INTERNAL_ERROR_3);
1370
1371	RTHeapSimpleFree(hHeap, pvTest2);
1372	RTHeapSimpleFree(hHeap, pvTest1);
1373	}
1374	else
1375	{
1376	pExecMemAllocator->pvAlignTweak = RTHeapSimpleAlloc(hHeap, pExecMemAllocator->cbHeapAlignTweak, 32 /cbAlignment/);
1377	AssertStmt(pExecMemAllocator->pvAlignTweak, rc = VERR_INTERNAL_ERROR_4);
1378	}
1379	if (RT_SUCCESS(rc))
1380	#endif /* !IEMEXECMEM_USE_ALT_SUB_ALLOCATOR */
1381	{
1382	/*
1383	* Add the chunk.
1384	*
1385	* This must be done before the unwind init so windows can allocate
1386	* memory from the chunk when using the alternative sub-allocator.
1387	*/
1388	pExecMemAllocator->aChunks[idxChunk].pvChunk = pvChunk;
1389	#ifdef IN_RING3
1390	pExecMemAllocator->aChunks[idxChunk].pvUnwindInfo = NULL;
1391	#endif
1392	#ifndef IEMEXECMEM_USE_ALT_SUB_ALLOCATOR
1393	pExecMemAllocator->aChunks[idxChunk].hHeap = hHeap;
1394	#else
1395	pExecMemAllocator->aChunks[idxChunk].cFreeUnits = pExecMemAllocator->cUnitsPerChunk;
1396	pExecMemAllocator->aChunks[idxChunk].idxFreeHint = 0;
1397	memset(&pExecMemAllocator->pbmAlloc[pExecMemAllocator->cBitmapElementsPerChunk * idxChunk],
1398	0, sizeof(pExecMemAllocator->pbmAlloc[0]) * pExecMemAllocator->cBitmapElementsPerChunk);
1399	#endif
1400
1401	pExecMemAllocator->cChunks = idxChunk + 1;
1402	pExecMemAllocator->idxChunkHint = idxChunk;
1403
1404	#ifdef IEMEXECMEM_USE_ALT_SUB_ALLOCATOR
1405	pExecMemAllocator->cbTotal += pExecMemAllocator->cbChunk;
1406	pExecMemAllocator->cbFree += pExecMemAllocator->cbChunk;
1407	#else
1408	size_t const cbFree = RTHeapSimpleGetFreeSize(hHeap);
1409	pExecMemAllocator->cbTotal += cbFree;
1410	pExecMemAllocator->cbFree += cbFree;
1411	#endif
1412
1413	#ifdef IN_RING3
1414	/*
1415	* Initialize the unwind information (this cannot really fail atm).
1416	* (This sets pvUnwindInfo.)
1417	*/
1418	rc = iemExecMemAllocatorInitAndRegisterUnwindInfoForChunk(pVCpu, pExecMemAllocator, pvChunk, idxChunk);
1419	if (RT_SUCCESS(rc))
1420	#endif
1421	{
1422	return VINF_SUCCESS;
1423	}
1424
1425	#ifdef IEMEXECMEM_USE_ALT_SUB_ALLOCATOR
1426	/* Just in case the impossible happens, undo the above up: */
1427	pExecMemAllocator->cbTotal -= pExecMemAllocator->cbChunk;
1428	pExecMemAllocator->cbFree -= pExecMemAllocator->aChunks[idxChunk].cFreeUnits << IEMEXECMEM_ALT_SUB_ALLOC_UNIT_SHIFT;
1429	pExecMemAllocator->cChunks = idxChunk;
1430	memset(&pExecMemAllocator->pbmAlloc[pExecMemAllocator->cBitmapElementsPerChunk * idxChunk],
1431	0xff, sizeof(pExecMemAllocator->pbmAlloc[0]) * pExecMemAllocator->cBitmapElementsPerChunk);
1432	pExecMemAllocator->aChunks[idxChunk].pvChunk = NULL;
1433	pExecMemAllocator->aChunks[idxChunk].cFreeUnits = 0;
1434	#endif
1435	}
1436	#ifndef IEMEXECMEM_USE_ALT_SUB_ALLOCATOR
1437	}
1438	#endif
1439	RTMemPageFree(pvChunk, pExecMemAllocator->cbChunk);
1440	RT_NOREF(pVCpu);
1441	return rc;
1442	}
1443
1444
1445	/**
1446	* Initializes the executable memory allocator for native recompilation on the
1447	* calling EMT.
1448	*
1449	* @returns VBox status code.
1450	* @param pVCpu The cross context virtual CPU structure of the calling
1451	* thread.
1452	* @param cbMax The max size of the allocator.
1453	* @param cbInitial The initial allocator size.
1454	* @param cbChunk The chunk size, 0 or UINT32_MAX for default (@a cbMax
1455	* dependent).
1456	*/
1457	int iemExecMemAllocatorInit(PVMCPU pVCpu, uint64_t cbMax, uint64_t cbInitial, uint32_t cbChunk)
1458	{
1459	/*
1460	* Validate input.
1461	*/
1462	AssertLogRelMsgReturn(cbMax >= _1M && cbMax <= _4G+_4G, ("cbMax=%RU64 (%RX64)\n", cbMax, cbMax), VERR_OUT_OF_RANGE);
1463	AssertReturn(cbInitial <= cbMax, VERR_OUT_OF_RANGE);
1464	AssertLogRelMsgReturn( cbChunk != UINT32_MAX
1465	\|\| cbChunk == 0
1466	\|\| ( RT_IS_POWER_OF_TWO(cbChunk)
1467	&& cbChunk >= _1M
1468	&& cbChunk <= _256M
1469	&& cbChunk <= cbMax),
1470	("cbChunk=%RU32 (%RX32) cbMax=%RU64\n", cbChunk, cbChunk, cbMax),
1471	VERR_OUT_OF_RANGE);
1472
1473	/*
1474	* Adjust/figure out the chunk size.
1475	*/
1476	if (cbChunk == 0 \|\| cbChunk == UINT32_MAX)
1477	{
1478	if (cbMax >= _256M)
1479	cbChunk = _64M;
1480	else
1481	{
1482	if (cbMax < _16M)
1483	cbChunk = cbMax >= _4M ? _4M : (uint32_t)cbMax;
1484	else
1485	cbChunk = (uint32_t)cbMax / 4;
1486	if (!RT_IS_POWER_OF_TWO(cbChunk))
1487	cbChunk = RT_BIT_32(ASMBitLastSetU32(cbChunk));
1488	}
1489	}
1490
1491	if (cbChunk > cbMax)
1492	cbMax = cbChunk;
1493	else
1494	cbMax = (cbMax - 1 + cbChunk) / cbChunk * cbChunk;
1495	uint32_t const cMaxChunks = (uint32_t)(cbMax / cbChunk);
1496	AssertLogRelReturn((uint64_t)cMaxChunks * cbChunk == cbMax, VERR_INTERNAL_ERROR_3);
1497
1498	/*
1499	* Allocate and initialize the allocatore instance.
1500	*/
1501	size_t cbNeeded = RT_UOFFSETOF_DYN(IEMEXECMEMALLOCATOR, aChunks[cMaxChunks]);
1502	#ifdef IEMEXECMEM_USE_ALT_SUB_ALLOCATOR
1503	size_t const offBitmaps = RT_ALIGN_Z(cbNeeded, RT_CACHELINE_SIZE);
1504	size_t const cbBitmap = cbChunk >> (IEMEXECMEM_ALT_SUB_ALLOC_UNIT_SHIFT + 3);
1505	cbNeeded += cbBitmap * cMaxChunks;
1506	AssertCompile(IEMEXECMEM_ALT_SUB_ALLOC_UNIT_SHIFT <= 10);
1507	Assert(cbChunk > RT_BIT_32(IEMEXECMEM_ALT_SUB_ALLOC_UNIT_SHIFT + 3));
1508	#endif
1509	#if defined(IN_RING3) && !defined(RT_OS_WINDOWS)
1510	size_t const offEhFrames = RT_ALIGN_Z(cbNeeded, RT_CACHELINE_SIZE);
1511	cbNeeded += sizeof(IEMEXECMEMCHUNKEHFRAME) * cMaxChunks;
1512	#endif
1513	PIEMEXECMEMALLOCATOR pExecMemAllocator = (PIEMEXECMEMALLOCATOR)RTMemAllocZ(cbNeeded);
1514	AssertLogRelMsgReturn(pExecMemAllocator, ("cbNeeded=%zx cMaxChunks=%#x cbChunk=%#x\n", cbNeeded, cMaxChunks, cbChunk),
1515	VERR_NO_MEMORY);
1516	pExecMemAllocator->uMagic = IEMEXECMEMALLOCATOR_MAGIC;
1517	pExecMemAllocator->cbChunk = cbChunk;
1518	pExecMemAllocator->cMaxChunks = cMaxChunks;
1519	pExecMemAllocator->cChunks = 0;
1520	pExecMemAllocator->idxChunkHint = 0;
1521	pExecMemAllocator->cAllocations = 0;
1522	pExecMemAllocator->cbTotal = 0;
1523	pExecMemAllocator->cbFree = 0;
1524	pExecMemAllocator->cbAllocated = 0;
1525	#ifdef IEMEXECMEM_USE_ALT_SUB_ALLOCATOR
1526	pExecMemAllocator->pbmAlloc = (uint64_t *)((uintptr_t)pExecMemAllocator + offBitmaps);
1527	pExecMemAllocator->cUnitsPerChunk = cbChunk >> IEMEXECMEM_ALT_SUB_ALLOC_UNIT_SHIFT;
1528	pExecMemAllocator->cBitmapElementsPerChunk = cbChunk >> (IEMEXECMEM_ALT_SUB_ALLOC_UNIT_SHIFT + 6);
1529	memset(pExecMemAllocator->pbmAlloc, 0xff, cbBitmap); /* Mark everything as allocated. Clear when chunks are added. */
1530	#endif
1531	#if defined(IN_RING3) && !defined(RT_OS_WINDOWS)
1532	pExecMemAllocator->paEhFrames = (PIEMEXECMEMCHUNKEHFRAME)((uintptr_t)pExecMemAllocator + offEhFrames);
1533	#endif
1534	for (uint32_t i = 0; i < cMaxChunks; i++)
1535	{
1536	#ifdef IEMEXECMEM_USE_ALT_SUB_ALLOCATOR
1537	pExecMemAllocator->aChunks[i].cFreeUnits = 0;
1538	pExecMemAllocator->aChunks[i].idxFreeHint = 0;
1539	#else
1540	pExecMemAllocator->aChunks[i].hHeap = NIL_RTHEAPSIMPLE;
1541	#endif
1542	pExecMemAllocator->aChunks[i].pvChunk = NULL;
1543	#ifdef IN_RING0
1544	pExecMemAllocator->aChunks[i].hMemObj = NIL_RTR0MEMOBJ;
1545	#else
1546	pExecMemAllocator->aChunks[i].pvUnwindInfo = NULL;
1547	#endif
1548	}
1549	pVCpu->iem.s.pExecMemAllocatorR3 = pExecMemAllocator;
1550
1551	/*
1552	* Do the initial allocations.
1553	*/
1554	while (cbInitial < (uint64_t)pExecMemAllocator->cChunks * pExecMemAllocator->cbChunk)
1555	{
1556	int rc = iemExecMemAllocatorGrow(pVCpu, pExecMemAllocator);
1557	AssertLogRelRCReturn(rc, rc);
1558	}
1559
1560	pExecMemAllocator->idxChunkHint = 0;
1561
1562	return VINF_SUCCESS;
1563	}
1564
1565
1566	/*********************************************************************************************************************************
1567	* Native Recompilation *
1568	*********************************************************************************************************************************/
1569
1570
1571	/**
1572	* Used by TB code when encountering a non-zero status or rcPassUp after a call.
1573	*/
1574	IEM_DECL_NATIVE_HLP_DEF(int, iemNativeHlpExecStatusCodeFiddling,(PVMCPUCC pVCpu, int rc, uint8_t idxInstr))
1575	{
1576	pVCpu->iem.s.cInstructions += idxInstr;
1577	return VBOXSTRICTRC_VAL(iemExecStatusCodeFiddling(pVCpu, rc == VINF_IEM_REEXEC_BREAK ? VINF_SUCCESS : rc));
1578	}
1579
1580
1581	/**
1582	* Used by TB code when it wants to raise a \#GP(0).
1583	*/
1584	IEM_DECL_NATIVE_HLP_DEF(int, iemNativeHlpExecRaiseGp0,(PVMCPUCC pVCpu))
1585	{
1586	iemRaiseGeneralProtectionFault0Jmp(pVCpu);
1587	#ifndef _MSC_VER
1588	return VINF_IEM_RAISED_XCPT; /* not reached */
1589	#endif
1590	}
1591
1592
1593	/**
1594	* Used by TB code when it wants to raise a \#NM.
1595	*/
1596	IEM_DECL_NATIVE_HLP_DEF(int, iemNativeHlpExecRaiseNm,(PVMCPUCC pVCpu))
1597	{
1598	iemRaiseDeviceNotAvailableJmp(pVCpu);
1599	#ifndef _MSC_VER
1600	return VINF_IEM_RAISED_XCPT; /* not reached */
1601	#endif
1602	}
1603
1604
1605	/**
1606	* Used by TB code when it wants to raise a \#UD.
1607	*/
1608	IEM_DECL_NATIVE_HLP_DEF(int, iemNativeHlpExecRaiseUd,(PVMCPUCC pVCpu))
1609	{
1610	iemRaiseUndefinedOpcodeJmp(pVCpu);
1611	#ifndef _MSC_VER
1612	return VINF_IEM_RAISED_XCPT; /* not reached */
1613	#endif
1614	}
1615
1616
1617	/**
1618	* Used by TB code when it wants to raise a \#MF.
1619	*/
1620	IEM_DECL_NATIVE_HLP_DEF(int, iemNativeHlpExecRaiseMf,(PVMCPUCC pVCpu))
1621	{
1622	iemRaiseMathFaultJmp(pVCpu);
1623	#ifndef _MSC_VER
1624	return VINF_IEM_RAISED_XCPT; /* not reached */
1625	#endif
1626	}
1627
1628
1629	/**
1630	* Used by TB code when it wants to raise a \#XF.
1631	*/
1632	IEM_DECL_NATIVE_HLP_DEF(int, iemNativeHlpExecRaiseXf,(PVMCPUCC pVCpu))
1633	{
1634	iemRaiseSimdFpExceptionJmp(pVCpu);
1635	#ifndef _MSC_VER
1636	return VINF_IEM_RAISED_XCPT; /* not reached */
1637	#endif
1638	}
1639
1640
1641	/**
1642	* Used by TB code when detecting opcode changes.
1643	* @see iemThreadeFuncWorkerObsoleteTb
1644	*/
1645	IEM_DECL_NATIVE_HLP_DEF(int, iemNativeHlpObsoleteTb,(PVMCPUCC pVCpu))
1646	{
1647	/* We set fSafeToFree to false where as we're being called in the context
1648	of a TB callback function, which for native TBs means we cannot release
1649	the executable memory till we've returned our way back to iemTbExec as
1650	that return path codes via the native code generated for the TB. */
1651	Log7(("TB obsolete: %p at %04x:%08RX64\n", pVCpu->iem.s.pCurTbR3, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip));
1652	iemThreadedTbObsolete(pVCpu, pVCpu->iem.s.pCurTbR3, false /fSafeToFree/);
1653	return VINF_IEM_REEXEC_BREAK;
1654	}
1655
1656
1657	/**
1658	* Used by TB code when we need to switch to a TB with CS.LIM checking.
1659	*/
1660	IEM_DECL_NATIVE_HLP_DEF(int, iemNativeHlpNeedCsLimChecking,(PVMCPUCC pVCpu))
1661	{
1662	Log7(("TB need CS.LIM: %p at %04x:%08RX64; offFromLim=%#RX64 CS.LIM=%#RX32 CS.BASE=%#RX64\n",
1663	pVCpu->iem.s.pCurTbR3, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
1664	(int64_t)pVCpu->cpum.GstCtx.cs.u32Limit - (int64_t)pVCpu->cpum.GstCtx.rip,
1665	pVCpu->cpum.GstCtx.cs.u32Limit, pVCpu->cpum.GstCtx.cs.u64Base));
1666	STAM_REL_COUNTER_INC(&pVCpu->iem.s.StatCheckNeedCsLimChecking);
1667	return VINF_IEM_REEXEC_BREAK;
1668	}
1669
1670
1671	/**
1672	* Used by TB code when we missed a PC check after a branch.
1673	*/
1674	IEM_DECL_NATIVE_HLP_DEF(int, iemNativeHlpCheckBranchMiss,(PVMCPUCC pVCpu))
1675	{
1676	Log7(("TB jmp miss: %p at %04x:%08RX64; GCPhysWithOffset=%RGp, pbInstrBuf=%p\n",
1677	pVCpu->iem.s.pCurTbR3, pVCpu->cpum.GstCtx.cs.Sel, pVCpu->cpum.GstCtx.rip,
1678	pVCpu->iem.s.GCPhysInstrBuf + pVCpu->cpum.GstCtx.rip + pVCpu->cpum.GstCtx.cs.u64Base - pVCpu->iem.s.uInstrBufPc,
1679	pVCpu->iem.s.pbInstrBuf));
1680	STAM_REL_COUNTER_INC(&pVCpu->iem.s.StatCheckBranchMisses);
1681	return VINF_IEM_REEXEC_BREAK;
1682	}
1683
1684
1685
1686	/*********************************************************************************************************************************
1687	* Helpers: Segmented memory fetches and stores. *
1688	*********************************************************************************************************************************/
1689
1690	/**
1691	* Used by TB code to load unsigned 8-bit data w/ segmentation.
1692	*/
1693	IEM_DECL_NATIVE_HLP_DEF(uint64_t, iemNativeHlpMemFetchDataU8,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg))
1694	{
1695	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_FETCH
1696	return (uint64_t)iemMemFetchDataU8SafeJmp(pVCpu, iSegReg, GCPtrMem);
1697	#else
1698	return (uint64_t)iemMemFetchDataU8Jmp(pVCpu, iSegReg, GCPtrMem);
1699	#endif
1700	}
1701
1702
1703	/**
1704	* Used by TB code to load signed 8-bit data w/ segmentation, sign extending it
1705	* to 16 bits.
1706	*/
1707	IEM_DECL_NATIVE_HLP_DEF(uint64_t, iemNativeHlpMemFetchDataU8_Sx_U16,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg))
1708	{
1709	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_FETCH
1710	return (uint64_t)(uint16_t)(int16_t)(int8_t)iemMemFetchDataU8SafeJmp(pVCpu, iSegReg, GCPtrMem);
1711	#else
1712	return (uint64_t)(uint16_t)(int16_t)(int8_t)iemMemFetchDataU8Jmp(pVCpu, iSegReg, GCPtrMem);
1713	#endif
1714	}
1715
1716
1717	/**
1718	* Used by TB code to load signed 8-bit data w/ segmentation, sign extending it
1719	* to 32 bits.
1720	*/
1721	IEM_DECL_NATIVE_HLP_DEF(uint64_t, iemNativeHlpMemFetchDataU8_Sx_U32,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg))
1722	{
1723	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_FETCH
1724	return (uint64_t)(uint32_t)(int32_t)(int8_t)iemMemFetchDataU8SafeJmp(pVCpu, iSegReg, GCPtrMem);
1725	#else
1726	return (uint64_t)(uint32_t)(int32_t)(int8_t)iemMemFetchDataU8Jmp(pVCpu, iSegReg, GCPtrMem);
1727	#endif
1728	}
1729
1730	/**
1731	* Used by TB code to load signed 8-bit data w/ segmentation, sign extending it
1732	* to 64 bits.
1733	*/
1734	IEM_DECL_NATIVE_HLP_DEF(uint64_t, iemNativeHlpMemFetchDataU8_Sx_U64,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg))
1735	{
1736	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_FETCH
1737	return (uint64_t)(int64_t)(int8_t)iemMemFetchDataU8SafeJmp(pVCpu, iSegReg, GCPtrMem);
1738	#else
1739	return (uint64_t)(int64_t)(int8_t)iemMemFetchDataU8Jmp(pVCpu, iSegReg, GCPtrMem);
1740	#endif
1741	}
1742
1743
1744	/**
1745	* Used by TB code to load unsigned 16-bit data w/ segmentation.
1746	*/
1747	IEM_DECL_NATIVE_HLP_DEF(uint64_t, iemNativeHlpMemFetchDataU16,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg))
1748	{
1749	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_FETCH
1750	return (uint64_t)iemMemFetchDataU16SafeJmp(pVCpu, iSegReg, GCPtrMem);
1751	#else
1752	return (uint64_t)iemMemFetchDataU16Jmp(pVCpu, iSegReg, GCPtrMem);
1753	#endif
1754	}
1755
1756
1757	/**
1758	* Used by TB code to load signed 16-bit data w/ segmentation, sign extending it
1759	* to 32 bits.
1760	*/
1761	IEM_DECL_NATIVE_HLP_DEF(uint64_t, iemNativeHlpMemFetchDataU16_Sx_U32,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg))
1762	{
1763	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_FETCH
1764	return (uint64_t)(uint32_t)(int32_t)(int16_t)iemMemFetchDataU16SafeJmp(pVCpu, iSegReg, GCPtrMem);
1765	#else
1766	return (uint64_t)(uint32_t)(int32_t)(int16_t)iemMemFetchDataU16Jmp(pVCpu, iSegReg, GCPtrMem);
1767	#endif
1768	}
1769
1770
1771	/**
1772	* Used by TB code to load signed 16-bit data w/ segmentation, sign extending it
1773	* to 64 bits.
1774	*/
1775	IEM_DECL_NATIVE_HLP_DEF(uint64_t, iemNativeHlpMemFetchDataU16_Sx_U64,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg))
1776	{
1777	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_FETCH
1778	return (uint64_t)(int64_t)(int16_t)iemMemFetchDataU16SafeJmp(pVCpu, iSegReg, GCPtrMem);
1779	#else
1780	return (uint64_t)(int64_t)(int16_t)iemMemFetchDataU16Jmp(pVCpu, iSegReg, GCPtrMem);
1781	#endif
1782	}
1783
1784
1785	/**
1786	* Used by TB code to load unsigned 32-bit data w/ segmentation.
1787	*/
1788	IEM_DECL_NATIVE_HLP_DEF(uint64_t, iemNativeHlpMemFetchDataU32,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg))
1789	{
1790	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_FETCH
1791	return (uint64_t)iemMemFetchDataU32SafeJmp(pVCpu, iSegReg, GCPtrMem);
1792	#else
1793	return (uint64_t)iemMemFetchDataU32Jmp(pVCpu, iSegReg, GCPtrMem);
1794	#endif
1795	}
1796
1797
1798	/**
1799	* Used by TB code to load signed 32-bit data w/ segmentation, sign extending it
1800	* to 64 bits.
1801	*/
1802	IEM_DECL_NATIVE_HLP_DEF(uint64_t, iemNativeHlpMemFetchDataU32_Sx_U64,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg))
1803	{
1804	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_FETCH
1805	return (uint64_t)(int64_t)(int32_t)iemMemFetchDataU32SafeJmp(pVCpu, iSegReg, GCPtrMem);
1806	#else
1807	return (uint64_t)(int64_t)(int32_t)iemMemFetchDataU32Jmp(pVCpu, iSegReg, GCPtrMem);
1808	#endif
1809	}
1810
1811
1812	/**
1813	* Used by TB code to load unsigned 64-bit data w/ segmentation.
1814	*/
1815	IEM_DECL_NATIVE_HLP_DEF(uint64_t, iemNativeHlpMemFetchDataU64,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg))
1816	{
1817	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_FETCH
1818	return iemMemFetchDataU64SafeJmp(pVCpu, iSegReg, GCPtrMem);
1819	#else
1820	return iemMemFetchDataU64Jmp(pVCpu, iSegReg, GCPtrMem);
1821	#endif
1822	}
1823
1824
1825	/**
1826	* Used by TB code to store unsigned 8-bit data w/ segmentation.
1827	*/
1828	IEM_DECL_NATIVE_HLP_DEF(void, iemNativeHlpMemStoreDataU8,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg, uint8_t u8Value))
1829	{
1830	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_STORE
1831	iemMemStoreDataU8SafeJmp(pVCpu, iSegReg, GCPtrMem, u8Value);
1832	#else
1833	iemMemStoreDataU8Jmp(pVCpu, iSegReg, GCPtrMem, u8Value);
1834	#endif
1835	}
1836
1837
1838	/**
1839	* Used by TB code to store unsigned 16-bit data w/ segmentation.
1840	*/
1841	IEM_DECL_NATIVE_HLP_DEF(void, iemNativeHlpMemStoreDataU16,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg, uint16_t u16Value))
1842	{
1843	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_STORE
1844	iemMemStoreDataU16SafeJmp(pVCpu, iSegReg, GCPtrMem, u16Value);
1845	#else
1846	iemMemStoreDataU16Jmp(pVCpu, iSegReg, GCPtrMem, u16Value);
1847	#endif
1848	}
1849
1850
1851	/**
1852	* Used by TB code to store unsigned 32-bit data w/ segmentation.
1853	*/
1854	IEM_DECL_NATIVE_HLP_DEF(void, iemNativeHlpMemStoreDataU32,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg, uint32_t u32Value))
1855	{
1856	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_STORE
1857	iemMemStoreDataU32SafeJmp(pVCpu, iSegReg, GCPtrMem, u32Value);
1858	#else
1859	iemMemStoreDataU32Jmp(pVCpu, iSegReg, GCPtrMem, u32Value);
1860	#endif
1861	}
1862
1863
1864	/**
1865	* Used by TB code to store unsigned 64-bit data w/ segmentation.
1866	*/
1867	IEM_DECL_NATIVE_HLP_DEF(void, iemNativeHlpMemStoreDataU64,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg, uint64_t u64Value))
1868	{
1869	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_STORE
1870	iemMemStoreDataU64SafeJmp(pVCpu, iSegReg, GCPtrMem, u64Value);
1871	#else
1872	iemMemStoreDataU64Jmp(pVCpu, iSegReg, GCPtrMem, u64Value);
1873	#endif
1874	}
1875
1876
1877
1878	/**
1879	* Used by TB code to store an unsigned 16-bit value onto a generic stack.
1880	*/
1881	IEM_DECL_NATIVE_HLP_DEF(void, iemNativeHlpStackStoreU16,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint16_t u16Value))
1882	{
1883	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_PUSH
1884	iemMemStoreStackU16SafeJmp(pVCpu, GCPtrMem, u16Value);
1885	#else
1886	iemMemStoreStackU16Jmp(pVCpu, GCPtrMem, u16Value);
1887	#endif
1888	}
1889
1890
1891	/**
1892	* Used by TB code to store an unsigned 32-bit value onto a generic stack.
1893	*/
1894	IEM_DECL_NATIVE_HLP_DEF(void, iemNativeHlpStackStoreU32,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint32_t u32Value))
1895	{
1896	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_PUSH
1897	iemMemStoreStackU32SafeJmp(pVCpu, GCPtrMem, u32Value);
1898	#else
1899	iemMemStoreStackU32Jmp(pVCpu, GCPtrMem, u32Value);
1900	#endif
1901	}
1902
1903
1904	/**
1905	* Used by TB code to store an 32-bit selector value onto a generic stack.
1906	*
1907	* Intel CPUs doesn't do write a whole dword, thus the special function.
1908	*/
1909	IEM_DECL_NATIVE_HLP_DEF(void, iemNativeHlpStackStoreU32SReg,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint32_t u32Value))
1910	{
1911	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_PUSH
1912	iemMemStoreStackU32SRegSafeJmp(pVCpu, GCPtrMem, u32Value);
1913	#else
1914	iemMemStoreStackU32SRegJmp(pVCpu, GCPtrMem, u32Value);
1915	#endif
1916	}
1917
1918
1919	/**
1920	* Used by TB code to push unsigned 64-bit value onto a generic stack.
1921	*/
1922	IEM_DECL_NATIVE_HLP_DEF(void, iemNativeHlpStackStoreU64,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint64_t u64Value))
1923	{
1924	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_PUSH
1925	iemMemStoreStackU64SafeJmp(pVCpu, GCPtrMem, u64Value);
1926	#else
1927	iemMemStoreStackU64Jmp(pVCpu, GCPtrMem, u64Value);
1928	#endif
1929	}
1930
1931
1932	/**
1933	* Used by TB code to fetch an unsigned 16-bit item off a generic stack.
1934	*/
1935	IEM_DECL_NATIVE_HLP_DEF(uint16_t, iemNativeHlpStackFetchU16,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem))
1936	{
1937	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_POP
1938	return iemMemFetchStackU16SafeJmp(pVCpu, GCPtrMem);
1939	#else
1940	return iemMemFetchStackU16Jmp(pVCpu, GCPtrMem);
1941	#endif
1942	}
1943
1944
1945	/**
1946	* Used by TB code to fetch an unsigned 32-bit item off a generic stack.
1947	*/
1948	IEM_DECL_NATIVE_HLP_DEF(uint32_t, iemNativeHlpStackFetchU32,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem))
1949	{
1950	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_POP
1951	return iemMemFetchStackU32SafeJmp(pVCpu, GCPtrMem);
1952	#else
1953	return iemMemFetchStackU32Jmp(pVCpu, GCPtrMem);
1954	#endif
1955	}
1956
1957
1958	/**
1959	* Used by TB code to fetch an unsigned 64-bit item off a generic stack.
1960	*/
1961	IEM_DECL_NATIVE_HLP_DEF(uint64_t, iemNativeHlpStackFetchU64,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem))
1962	{
1963	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_POP
1964	return iemMemFetchStackU64SafeJmp(pVCpu, GCPtrMem);
1965	#else
1966	return iemMemFetchStackU64Jmp(pVCpu, GCPtrMem);
1967	#endif
1968	}
1969
1970
1971
1972	/*********************************************************************************************************************************
1973	* Helpers: Flat memory fetches and stores. *
1974	*********************************************************************************************************************************/
1975
1976	/**
1977	* Used by TB code to load unsigned 8-bit data w/ flat address.
1978	* @note Zero extending the value to 64-bit to simplify assembly.
1979	*/
1980	IEM_DECL_NATIVE_HLP_DEF(uint64_t, iemNativeHlpMemFlatFetchDataU8,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem))
1981	{
1982	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_FETCH
1983	return (uint64_t)iemMemFetchDataU8SafeJmp(pVCpu, UINT8_MAX, GCPtrMem);
1984	#else
1985	return (uint64_t)iemMemFlatFetchDataU8Jmp(pVCpu, GCPtrMem);
1986	#endif
1987	}
1988
1989
1990	/**
1991	* Used by TB code to load signed 8-bit data w/ flat address, sign extending it
1992	* to 16 bits.
1993	* @note Zero extending the value to 64-bit to simplify assembly.
1994	*/
1995	IEM_DECL_NATIVE_HLP_DEF(uint64_t, iemNativeHlpMemFlatFetchDataU8_Sx_U16,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem))
1996	{
1997	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_FETCH
1998	return (uint64_t)(uint16_t)(int16_t)(int8_t)iemMemFetchDataU8SafeJmp(pVCpu, UINT8_MAX, GCPtrMem);
1999	#else
2000	return (uint64_t)(uint16_t)(int16_t)(int8_t)iemMemFlatFetchDataU8Jmp(pVCpu, GCPtrMem);
2001	#endif
2002	}
2003
2004
2005	/**
2006	* Used by TB code to load signed 8-bit data w/ flat address, sign extending it
2007	* to 32 bits.
2008	* @note Zero extending the value to 64-bit to simplify assembly.
2009	*/
2010	IEM_DECL_NATIVE_HLP_DEF(uint64_t, iemNativeHlpMemFlatFetchDataU8_Sx_U32,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem))
2011	{
2012	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_FETCH
2013	return (uint64_t)(uint32_t)(int32_t)(int8_t)iemMemFetchDataU8SafeJmp(pVCpu, UINT8_MAX, GCPtrMem);
2014	#else
2015	return (uint64_t)(uint32_t)(int32_t)(int8_t)iemMemFlatFetchDataU8Jmp(pVCpu, GCPtrMem);
2016	#endif
2017	}
2018
2019
2020	/**
2021	* Used by TB code to load signed 8-bit data w/ flat address, sign extending it
2022	* to 64 bits.
2023	*/
2024	IEM_DECL_NATIVE_HLP_DEF(uint64_t, iemNativeHlpMemFlatFetchDataU8_Sx_U64,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem))
2025	{
2026	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_FETCH
2027	return (uint64_t)(int64_t)(int8_t)iemMemFetchDataU8SafeJmp(pVCpu, UINT8_MAX, GCPtrMem);
2028	#else
2029	return (uint64_t)(int64_t)(int8_t)iemMemFlatFetchDataU8Jmp(pVCpu, GCPtrMem);
2030	#endif
2031	}
2032
2033
2034	/**
2035	* Used by TB code to load unsigned 16-bit data w/ flat address.
2036	* @note Zero extending the value to 64-bit to simplify assembly.
2037	*/
2038	IEM_DECL_NATIVE_HLP_DEF(uint64_t, iemNativeHlpMemFlatFetchDataU16,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem))
2039	{
2040	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_FETCH
2041	return (uint64_t)iemMemFetchDataU16SafeJmp(pVCpu, UINT8_MAX, GCPtrMem);
2042	#else
2043	return (uint64_t)iemMemFlatFetchDataU16Jmp(pVCpu, GCPtrMem);
2044	#endif
2045	}
2046
2047
2048	/**
2049	* Used by TB code to load signed 16-bit data w/ flat address, sign extending it
2050	* to 32 bits.
2051	* @note Zero extending the value to 64-bit to simplify assembly.
2052	*/
2053	IEM_DECL_NATIVE_HLP_DEF(uint64_t, iemNativeHlpMemFlatFetchDataU16_Sx_U32,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem))
2054	{
2055	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_FETCH
2056	return (uint64_t)(uint32_t)(int32_t)(int16_t)iemMemFetchDataU16SafeJmp(pVCpu, UINT8_MAX, GCPtrMem);
2057	#else
2058	return (uint64_t)(uint32_t)(int32_t)(int16_t)iemMemFlatFetchDataU16Jmp(pVCpu, GCPtrMem);
2059	#endif
2060	}
2061
2062
2063	/**
2064	* Used by TB code to load signed 16-bit data w/ flat address, sign extending it
2065	* to 64 bits.
2066	* @note Zero extending the value to 64-bit to simplify assembly.
2067	*/
2068	IEM_DECL_NATIVE_HLP_DEF(uint64_t, iemNativeHlpMemFlatFetchDataU16_Sx_U64,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem))
2069	{
2070	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_FETCH
2071	return (uint64_t)(int64_t)(int16_t)iemMemFetchDataU16SafeJmp(pVCpu, UINT8_MAX, GCPtrMem);
2072	#else
2073	return (uint64_t)(int64_t)(int16_t)iemMemFlatFetchDataU16Jmp(pVCpu, GCPtrMem);
2074	#endif
2075	}
2076
2077
2078	/**
2079	* Used by TB code to load unsigned 32-bit data w/ flat address.
2080	* @note Zero extending the value to 64-bit to simplify assembly.
2081	*/
2082	IEM_DECL_NATIVE_HLP_DEF(uint64_t, iemNativeHlpMemFlatFetchDataU32,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem))
2083	{
2084	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_FETCH
2085	return (uint64_t)iemMemFetchDataU32SafeJmp(pVCpu, UINT8_MAX, GCPtrMem);
2086	#else
2087	return (uint64_t)iemMemFlatFetchDataU32Jmp(pVCpu, GCPtrMem);
2088	#endif
2089	}
2090
2091
2092	/**
2093	* Used by TB code to load signed 32-bit data w/ flat address, sign extending it
2094	* to 64 bits.
2095	* @note Zero extending the value to 64-bit to simplify assembly.
2096	*/
2097	IEM_DECL_NATIVE_HLP_DEF(uint64_t, iemNativeHlpMemFlatFetchDataU32_Sx_U64,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem))
2098	{
2099	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_FETCH
2100	return (uint64_t)(int64_t)(int32_t)iemMemFetchDataU32SafeJmp(pVCpu, UINT8_MAX, GCPtrMem);
2101	#else
2102	return (uint64_t)(int64_t)(int32_t)iemMemFlatFetchDataU32Jmp(pVCpu, GCPtrMem);
2103	#endif
2104	}
2105
2106
2107	/**
2108	* Used by TB code to load unsigned 64-bit data w/ flat address.
2109	*/
2110	IEM_DECL_NATIVE_HLP_DEF(uint64_t, iemNativeHlpMemFlatFetchDataU64,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem))
2111	{
2112	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_FETCH
2113	return iemMemFetchDataU64SafeJmp(pVCpu, UINT8_MAX, GCPtrMem);
2114	#else
2115	return iemMemFlatFetchDataU64Jmp(pVCpu, GCPtrMem);
2116	#endif
2117	}
2118
2119
2120	/**
2121	* Used by TB code to store unsigned 8-bit data w/ flat address.
2122	*/
2123	IEM_DECL_NATIVE_HLP_DEF(void, iemNativeHlpMemFlatStoreDataU8,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t u8Value))
2124	{
2125	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_STORE
2126	iemMemStoreDataU8SafeJmp(pVCpu, UINT8_MAX, GCPtrMem, u8Value);
2127	#else
2128	iemMemFlatStoreDataU8Jmp(pVCpu, GCPtrMem, u8Value);
2129	#endif
2130	}
2131
2132
2133	/**
2134	* Used by TB code to store unsigned 16-bit data w/ flat address.
2135	*/
2136	IEM_DECL_NATIVE_HLP_DEF(void, iemNativeHlpMemFlatStoreDataU16,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint16_t u16Value))
2137	{
2138	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_STORE
2139	iemMemStoreDataU16SafeJmp(pVCpu, UINT8_MAX, GCPtrMem, u16Value);
2140	#else
2141	iemMemFlatStoreDataU16Jmp(pVCpu, GCPtrMem, u16Value);
2142	#endif
2143	}
2144
2145
2146	/**
2147	* Used by TB code to store unsigned 32-bit data w/ flat address.
2148	*/
2149	IEM_DECL_NATIVE_HLP_DEF(void, iemNativeHlpMemFlatStoreDataU32,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint32_t u32Value))
2150	{
2151	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_STORE
2152	iemMemStoreDataU32SafeJmp(pVCpu, UINT8_MAX, GCPtrMem, u32Value);
2153	#else
2154	iemMemFlatStoreDataU32Jmp(pVCpu, GCPtrMem, u32Value);
2155	#endif
2156	}
2157
2158
2159	/**
2160	* Used by TB code to store unsigned 64-bit data w/ flat address.
2161	*/
2162	IEM_DECL_NATIVE_HLP_DEF(void, iemNativeHlpMemFlatStoreDataU64,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint64_t u64Value))
2163	{
2164	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_STORE
2165	iemMemStoreDataU64SafeJmp(pVCpu, UINT8_MAX, GCPtrMem, u64Value);
2166	#else
2167	iemMemFlatStoreDataU64Jmp(pVCpu, GCPtrMem, u64Value);
2168	#endif
2169	}
2170
2171
2172
2173	/**
2174	* Used by TB code to store an unsigned 16-bit value onto a flat stack.
2175	*/
2176	IEM_DECL_NATIVE_HLP_DEF(void, iemNativeHlpStackFlatStoreU16,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint16_t u16Value))
2177	{
2178	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_PUSH
2179	iemMemStoreStackU16SafeJmp(pVCpu, GCPtrMem, u16Value);
2180	#else
2181	iemMemFlatStoreStackU16Jmp(pVCpu, GCPtrMem, u16Value);
2182	#endif
2183	}
2184
2185
2186	/**
2187	* Used by TB code to store an unsigned 32-bit value onto a flat stack.
2188	*/
2189	IEM_DECL_NATIVE_HLP_DEF(void, iemNativeHlpStackFlatStoreU32,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint32_t u32Value))
2190	{
2191	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_PUSH
2192	iemMemStoreStackU32SafeJmp(pVCpu, GCPtrMem, u32Value);
2193	#else
2194	iemMemFlatStoreStackU32Jmp(pVCpu, GCPtrMem, u32Value);
2195	#endif
2196	}
2197
2198
2199	/**
2200	* Used by TB code to store a segment selector value onto a flat stack.
2201	*
2202	* Intel CPUs doesn't do write a whole dword, thus the special function.
2203	*/
2204	IEM_DECL_NATIVE_HLP_DEF(void, iemNativeHlpStackFlatStoreU32SReg,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint32_t u32Value))
2205	{
2206	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_PUSH
2207	iemMemStoreStackU32SRegSafeJmp(pVCpu, GCPtrMem, u32Value);
2208	#else
2209	iemMemFlatStoreStackU32SRegJmp(pVCpu, GCPtrMem, u32Value);
2210	#endif
2211	}
2212
2213
2214	/**
2215	* Used by TB code to store an unsigned 64-bit value onto a flat stack.
2216	*/
2217	IEM_DECL_NATIVE_HLP_DEF(void, iemNativeHlpStackFlatStoreU64,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint64_t u64Value))
2218	{
2219	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_PUSH
2220	iemMemStoreStackU64SafeJmp(pVCpu, GCPtrMem, u64Value);
2221	#else
2222	iemMemFlatStoreStackU64Jmp(pVCpu, GCPtrMem, u64Value);
2223	#endif
2224	}
2225
2226
2227	/**
2228	* Used by TB code to fetch an unsigned 16-bit item off a generic stack.
2229	*/
2230	IEM_DECL_NATIVE_HLP_DEF(uint16_t, iemNativeHlpStackFlatFetchU16,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem))
2231	{
2232	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_POP
2233	return iemMemFetchStackU16SafeJmp(pVCpu, GCPtrMem);
2234	#else
2235	return iemMemFlatFetchStackU16Jmp(pVCpu, GCPtrMem);
2236	#endif
2237	}
2238
2239
2240	/**
2241	* Used by TB code to fetch an unsigned 32-bit item off a generic stack.
2242	*/
2243	IEM_DECL_NATIVE_HLP_DEF(uint32_t, iemNativeHlpStackFlatFetchU32,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem))
2244	{
2245	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_POP
2246	return iemMemFetchStackU32SafeJmp(pVCpu, GCPtrMem);
2247	#else
2248	return iemMemFlatFetchStackU32Jmp(pVCpu, GCPtrMem);
2249	#endif
2250	}
2251
2252
2253	/**
2254	* Used by TB code to fetch an unsigned 64-bit item off a generic stack.
2255	*/
2256	IEM_DECL_NATIVE_HLP_DEF(uint64_t, iemNativeHlpStackFlatFetchU64,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem))
2257	{
2258	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_POP
2259	return iemMemFetchStackU64SafeJmp(pVCpu, GCPtrMem);
2260	#else
2261	return iemMemFlatFetchStackU64Jmp(pVCpu, GCPtrMem);
2262	#endif
2263	}
2264
2265
2266
2267	/*********************************************************************************************************************************
2268	* Helpers: Segmented memory mapping. *
2269	*********************************************************************************************************************************/
2270
2271	/**
2272	* Used by TB code to map unsigned 8-bit data for atomic read-write w/
2273	* segmentation.
2274	*/
2275	IEM_DECL_NATIVE_HLP_DEF(uint8_t , iemNativeHlpMemMapDataU8Atomic,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo,
2276	RTGCPTR GCPtrMem, uint8_t iSegReg))
2277	{
2278	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2279	return iemMemMapDataU8AtSafeJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2280	#else
2281	return iemMemMapDataU8AtJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2282	#endif
2283	}
2284
2285
2286	/**
2287	* Used by TB code to map unsigned 8-bit data read-write w/ segmentation.
2288	*/
2289	IEM_DECL_NATIVE_HLP_DEF(uint8_t , iemNativeHlpMemMapDataU8Rw,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo,
2290	RTGCPTR GCPtrMem, uint8_t iSegReg))
2291	{
2292	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2293	return iemMemMapDataU8RwSafeJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2294	#else
2295	return iemMemMapDataU8RwJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2296	#endif
2297	}
2298
2299
2300	/**
2301	* Used by TB code to map unsigned 8-bit data writeonly w/ segmentation.
2302	*/
2303	IEM_DECL_NATIVE_HLP_DEF(uint8_t , iemNativeHlpMemMapDataU8Wo,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo,
2304	RTGCPTR GCPtrMem, uint8_t iSegReg))
2305	{
2306	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2307	return iemMemMapDataU8WoSafeJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2308	#else
2309	return iemMemMapDataU8WoJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2310	#endif
2311	}
2312
2313
2314	/**
2315	* Used by TB code to map unsigned 8-bit data readonly w/ segmentation.
2316	*/
2317	IEM_DECL_NATIVE_HLP_DEF(uint8_t const , iemNativeHlpMemMapDataU8Ro,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo,
2318	RTGCPTR GCPtrMem, uint8_t iSegReg))
2319	{
2320	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2321	return iemMemMapDataU8RoSafeJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2322	#else
2323	return iemMemMapDataU8RoJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2324	#endif
2325	}
2326
2327
2328	/**
2329	* Used by TB code to map unsigned 16-bit data for atomic read-write w/
2330	* segmentation.
2331	*/
2332	IEM_DECL_NATIVE_HLP_DEF(uint16_t , iemNativeHlpMemMapDataU16Atomic,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo,
2333	RTGCPTR GCPtrMem, uint8_t iSegReg))
2334	{
2335	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2336	return iemMemMapDataU16AtSafeJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2337	#else
2338	return iemMemMapDataU16AtJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2339	#endif
2340	}
2341
2342
2343	/**
2344	* Used by TB code to map unsigned 16-bit data read-write w/ segmentation.
2345	*/
2346	IEM_DECL_NATIVE_HLP_DEF(uint16_t , iemNativeHlpMemMapDataU16Rw,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo,
2347	RTGCPTR GCPtrMem, uint8_t iSegReg))
2348	{
2349	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2350	return iemMemMapDataU16RwSafeJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2351	#else
2352	return iemMemMapDataU16RwJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2353	#endif
2354	}
2355
2356
2357	/**
2358	* Used by TB code to map unsigned 16-bit data writeonly w/ segmentation.
2359	*/
2360	IEM_DECL_NATIVE_HLP_DEF(uint16_t , iemNativeHlpMemMapDataU16Wo,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo,
2361	RTGCPTR GCPtrMem, uint8_t iSegReg))
2362	{
2363	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2364	return iemMemMapDataU16WoSafeJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2365	#else
2366	return iemMemMapDataU16WoJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2367	#endif
2368	}
2369
2370
2371	/**
2372	* Used by TB code to map unsigned 16-bit data readonly w/ segmentation.
2373	*/
2374	IEM_DECL_NATIVE_HLP_DEF(uint16_t const , iemNativeHlpMemMapDataU16Ro,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo,
2375	RTGCPTR GCPtrMem, uint8_t iSegReg))
2376	{
2377	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2378	return iemMemMapDataU16RoSafeJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2379	#else
2380	return iemMemMapDataU16RoJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2381	#endif
2382	}
2383
2384
2385	/**
2386	* Used by TB code to map unsigned 32-bit data for atomic read-write w/
2387	* segmentation.
2388	*/
2389	IEM_DECL_NATIVE_HLP_DEF(uint32_t , iemNativeHlpMemMapDataU32Atomic,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo,
2390	RTGCPTR GCPtrMem, uint8_t iSegReg))
2391	{
2392	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2393	return iemMemMapDataU32AtSafeJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2394	#else
2395	return iemMemMapDataU32AtJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2396	#endif
2397	}
2398
2399
2400	/**
2401	* Used by TB code to map unsigned 32-bit data read-write w/ segmentation.
2402	*/
2403	IEM_DECL_NATIVE_HLP_DEF(uint32_t , iemNativeHlpMemMapDataU32Rw,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo,
2404	RTGCPTR GCPtrMem, uint8_t iSegReg))
2405	{
2406	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2407	return iemMemMapDataU32RwSafeJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2408	#else
2409	return iemMemMapDataU32RwJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2410	#endif
2411	}
2412
2413
2414	/**
2415	* Used by TB code to map unsigned 32-bit data writeonly w/ segmentation.
2416	*/
2417	IEM_DECL_NATIVE_HLP_DEF(uint32_t , iemNativeHlpMemMapDataU32Wo,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo,
2418	RTGCPTR GCPtrMem, uint8_t iSegReg))
2419	{
2420	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2421	return iemMemMapDataU32WoSafeJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2422	#else
2423	return iemMemMapDataU32WoJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2424	#endif
2425	}
2426
2427
2428	/**
2429	* Used by TB code to map unsigned 32-bit data readonly w/ segmentation.
2430	*/
2431	IEM_DECL_NATIVE_HLP_DEF(uint32_t const , iemNativeHlpMemMapDataU32Ro,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo,
2432	RTGCPTR GCPtrMem, uint8_t iSegReg))
2433	{
2434	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2435	return iemMemMapDataU32RoSafeJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2436	#else
2437	return iemMemMapDataU32RoJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2438	#endif
2439	}
2440
2441
2442	/**
2443	* Used by TB code to map unsigned 64-bit data for atomic read-write w/
2444	* segmentation.
2445	*/
2446	IEM_DECL_NATIVE_HLP_DEF(uint64_t , iemNativeHlpMemMapDataU64Atomic,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo,
2447	RTGCPTR GCPtrMem, uint8_t iSegReg))
2448	{
2449	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2450	return iemMemMapDataU64AtSafeJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2451	#else
2452	return iemMemMapDataU64AtJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2453	#endif
2454	}
2455
2456
2457	/**
2458	* Used by TB code to map unsigned 64-bit data read-write w/ segmentation.
2459	*/
2460	IEM_DECL_NATIVE_HLP_DEF(uint64_t , iemNativeHlpMemMapDataU64Rw,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo,
2461	RTGCPTR GCPtrMem, uint8_t iSegReg))
2462	{
2463	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2464	return iemMemMapDataU64RwSafeJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2465	#else
2466	return iemMemMapDataU64RwJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2467	#endif
2468	}
2469
2470
2471	/**
2472	* Used by TB code to map unsigned 64-bit data writeonly w/ segmentation.
2473	*/
2474	IEM_DECL_NATIVE_HLP_DEF(uint64_t , iemNativeHlpMemMapDataU64Wo,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo,
2475	RTGCPTR GCPtrMem, uint8_t iSegReg))
2476	{
2477	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2478	return iemMemMapDataU64WoSafeJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2479	#else
2480	return iemMemMapDataU64WoJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2481	#endif
2482	}
2483
2484
2485	/**
2486	* Used by TB code to map unsigned 64-bit data readonly w/ segmentation.
2487	*/
2488	IEM_DECL_NATIVE_HLP_DEF(uint64_t const , iemNativeHlpMemMapDataU64Ro,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo,
2489	RTGCPTR GCPtrMem, uint8_t iSegReg))
2490	{
2491	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2492	return iemMemMapDataU64RoSafeJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2493	#else
2494	return iemMemMapDataU64RoJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2495	#endif
2496	}
2497
2498
2499	/**
2500	* Used by TB code to map 80-bit float data writeonly w/ segmentation.
2501	*/
2502	IEM_DECL_NATIVE_HLP_DEF(RTFLOAT80U , iemNativeHlpMemMapDataR80Wo,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo,
2503	RTGCPTR GCPtrMem, uint8_t iSegReg))
2504	{
2505	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2506	return iemMemMapDataR80WoSafeJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2507	#else
2508	return iemMemMapDataR80WoJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2509	#endif
2510	}
2511
2512
2513	/**
2514	* Used by TB code to map 80-bit BCD data writeonly w/ segmentation.
2515	*/
2516	IEM_DECL_NATIVE_HLP_DEF(RTPBCD80U , iemNativeHlpMemMapDataD80Wo,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo,
2517	RTGCPTR GCPtrMem, uint8_t iSegReg))
2518	{
2519	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2520	return iemMemMapDataD80WoSafeJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2521	#else
2522	return iemMemMapDataD80WoJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2523	#endif
2524	}
2525
2526
2527	/**
2528	* Used by TB code to map unsigned 128-bit data for atomic read-write w/
2529	* segmentation.
2530	*/
2531	IEM_DECL_NATIVE_HLP_DEF(RTUINT128U , iemNativeHlpMemMapDataU128Atomic,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo,
2532	RTGCPTR GCPtrMem, uint8_t iSegReg))
2533	{
2534	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2535	return iemMemMapDataU128AtSafeJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2536	#else
2537	return iemMemMapDataU128AtJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2538	#endif
2539	}
2540
2541
2542	/**
2543	* Used by TB code to map unsigned 128-bit data read-write w/ segmentation.
2544	*/
2545	IEM_DECL_NATIVE_HLP_DEF(RTUINT128U , iemNativeHlpMemMapDataU128Rw,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo,
2546	RTGCPTR GCPtrMem, uint8_t iSegReg))
2547	{
2548	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2549	return iemMemMapDataU128RwSafeJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2550	#else
2551	return iemMemMapDataU128RwJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2552	#endif
2553	}
2554
2555
2556	/**
2557	* Used by TB code to map unsigned 128-bit data writeonly w/ segmentation.
2558	*/
2559	IEM_DECL_NATIVE_HLP_DEF(RTUINT128U , iemNativeHlpMemMapDataU128Wo,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo,
2560	RTGCPTR GCPtrMem, uint8_t iSegReg))
2561	{
2562	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2563	return iemMemMapDataU128WoSafeJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2564	#else
2565	return iemMemMapDataU128WoJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2566	#endif
2567	}
2568
2569
2570	/**
2571	* Used by TB code to map unsigned 128-bit data readonly w/ segmentation.
2572	*/
2573	IEM_DECL_NATIVE_HLP_DEF(RTUINT128U const , iemNativeHlpMemMapDataU128Ro,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo,
2574	RTGCPTR GCPtrMem, uint8_t iSegReg))
2575	{
2576	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2577	return iemMemMapDataU128RoSafeJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2578	#else
2579	return iemMemMapDataU128RoJmp(pVCpu, pbUnmapInfo, iSegReg, GCPtrMem);
2580	#endif
2581	}
2582
2583
2584	/*********************************************************************************************************************************
2585	* Helpers: Flat memory mapping. *
2586	*********************************************************************************************************************************/
2587
2588	/**
2589	* Used by TB code to map unsigned 8-bit data for atomic read-write w/ flat
2590	* address.
2591	*/
2592	IEM_DECL_NATIVE_HLP_DEF(uint8_t , iemNativeHlpMemFlatMapDataU8Atomic,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo, RTGCPTR GCPtrMem))
2593	{
2594	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2595	return iemMemMapDataU8AtSafeJmp(pVCpu, pbUnmapInfo, UINT8_MAX, GCPtrMem);
2596	#else
2597	return iemMemFlatMapDataU8AtJmp(pVCpu, pbUnmapInfo, GCPtrMem);
2598	#endif
2599	}
2600
2601
2602	/**
2603	* Used by TB code to map unsigned 8-bit data read-write w/ flat address.
2604	*/
2605	IEM_DECL_NATIVE_HLP_DEF(uint8_t , iemNativeHlpMemFlatMapDataU8Rw,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo, RTGCPTR GCPtrMem))
2606	{
2607	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2608	return iemMemMapDataU8RwSafeJmp(pVCpu, pbUnmapInfo, UINT8_MAX, GCPtrMem);
2609	#else
2610	return iemMemFlatMapDataU8RwJmp(pVCpu, pbUnmapInfo, GCPtrMem);
2611	#endif
2612	}
2613
2614
2615	/**
2616	* Used by TB code to map unsigned 8-bit data writeonly w/ flat address.
2617	*/
2618	IEM_DECL_NATIVE_HLP_DEF(uint8_t , iemNativeHlpMemFlatMapDataU8Wo,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo, RTGCPTR GCPtrMem))
2619	{
2620	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2621	return iemMemMapDataU8WoSafeJmp(pVCpu, pbUnmapInfo, UINT8_MAX, GCPtrMem);
2622	#else
2623	return iemMemFlatMapDataU8WoJmp(pVCpu, pbUnmapInfo, GCPtrMem);
2624	#endif
2625	}
2626
2627
2628	/**
2629	* Used by TB code to map unsigned 8-bit data readonly w/ flat address.
2630	*/
2631	IEM_DECL_NATIVE_HLP_DEF(uint8_t const , iemNativeHlpMemFlatMapDataU8Ro,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo, RTGCPTR GCPtrMem))
2632	{
2633	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2634	return iemMemMapDataU8RoSafeJmp(pVCpu, pbUnmapInfo, UINT8_MAX, GCPtrMem);
2635	#else
2636	return iemMemFlatMapDataU8RoJmp(pVCpu, pbUnmapInfo, GCPtrMem);
2637	#endif
2638	}
2639
2640
2641	/**
2642	* Used by TB code to map unsigned 16-bit data for atomic read-write w/ flat
2643	* address.
2644	*/
2645	IEM_DECL_NATIVE_HLP_DEF(uint16_t , iemNativeHlpMemFlatMapDataU16Atomic,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo, RTGCPTR GCPtrMem))
2646	{
2647	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2648	return iemMemMapDataU16AtSafeJmp(pVCpu, pbUnmapInfo, UINT8_MAX, GCPtrMem);
2649	#else
2650	return iemMemFlatMapDataU16AtJmp(pVCpu, pbUnmapInfo, GCPtrMem);
2651	#endif
2652	}
2653
2654
2655	/**
2656	* Used by TB code to map unsigned 16-bit data read-write w/ flat address.
2657	*/
2658	IEM_DECL_NATIVE_HLP_DEF(uint16_t , iemNativeHlpMemFlatMapDataU16Rw,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo, RTGCPTR GCPtrMem))
2659	{
2660	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2661	return iemMemMapDataU16RwSafeJmp(pVCpu, pbUnmapInfo, UINT8_MAX, GCPtrMem);
2662	#else
2663	return iemMemFlatMapDataU16RwJmp(pVCpu, pbUnmapInfo, GCPtrMem);
2664	#endif
2665	}
2666
2667
2668	/**
2669	* Used by TB code to map unsigned 16-bit data writeonly w/ flat address.
2670	*/
2671	IEM_DECL_NATIVE_HLP_DEF(uint16_t , iemNativeHlpMemFlatMapDataU16Wo,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo, RTGCPTR GCPtrMem))
2672	{
2673	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2674	return iemMemMapDataU16WoSafeJmp(pVCpu, pbUnmapInfo, UINT8_MAX, GCPtrMem);
2675	#else
2676	return iemMemFlatMapDataU16WoJmp(pVCpu, pbUnmapInfo, GCPtrMem);
2677	#endif
2678	}
2679
2680
2681	/**
2682	* Used by TB code to map unsigned 16-bit data readonly w/ flat address.
2683	*/
2684	IEM_DECL_NATIVE_HLP_DEF(uint16_t const , iemNativeHlpMemFlatMapDataU16Ro,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo, RTGCPTR GCPtrMem))
2685	{
2686	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2687	return iemMemMapDataU16RoSafeJmp(pVCpu, pbUnmapInfo, UINT8_MAX, GCPtrMem);
2688	#else
2689	return iemMemFlatMapDataU16RoJmp(pVCpu, pbUnmapInfo, GCPtrMem);
2690	#endif
2691	}
2692
2693
2694	/**
2695	* Used by TB code to map unsigned 32-bit data for atomic read-write w/ flat
2696	* address.
2697	*/
2698	IEM_DECL_NATIVE_HLP_DEF(uint32_t , iemNativeHlpMemFlatMapDataU32Atomic,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo, RTGCPTR GCPtrMem))
2699	{
2700	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2701	return iemMemMapDataU32AtSafeJmp(pVCpu, pbUnmapInfo, UINT8_MAX, GCPtrMem);
2702	#else
2703	return iemMemFlatMapDataU32AtJmp(pVCpu, pbUnmapInfo, GCPtrMem);
2704	#endif
2705	}
2706
2707
2708	/**
2709	* Used by TB code to map unsigned 32-bit data read-write w/ flat address.
2710	*/
2711	IEM_DECL_NATIVE_HLP_DEF(uint32_t , iemNativeHlpMemFlatMapDataU32Rw,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo, RTGCPTR GCPtrMem))
2712	{
2713	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2714	return iemMemMapDataU32RwSafeJmp(pVCpu, pbUnmapInfo, UINT8_MAX, GCPtrMem);
2715	#else
2716	return iemMemFlatMapDataU32RwJmp(pVCpu, pbUnmapInfo, GCPtrMem);
2717	#endif
2718	}
2719
2720
2721	/**
2722	* Used by TB code to map unsigned 32-bit data writeonly w/ flat address.
2723	*/
2724	IEM_DECL_NATIVE_HLP_DEF(uint32_t , iemNativeHlpMemFlatMapDataU32Wo,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo, RTGCPTR GCPtrMem))
2725	{
2726	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2727	return iemMemMapDataU32WoSafeJmp(pVCpu, pbUnmapInfo, UINT8_MAX, GCPtrMem);
2728	#else
2729	return iemMemFlatMapDataU32WoJmp(pVCpu, pbUnmapInfo, GCPtrMem);
2730	#endif
2731	}
2732
2733
2734	/**
2735	* Used by TB code to map unsigned 32-bit data readonly w/ flat address.
2736	*/
2737	IEM_DECL_NATIVE_HLP_DEF(uint32_t const , iemNativeHlpMemFlatMapDataU32Ro,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo, RTGCPTR GCPtrMem))
2738	{
2739	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2740	return iemMemMapDataU32RoSafeJmp(pVCpu, pbUnmapInfo, UINT8_MAX, GCPtrMem);
2741	#else
2742	return iemMemFlatMapDataU32RoJmp(pVCpu, pbUnmapInfo, GCPtrMem);
2743	#endif
2744	}
2745
2746
2747	/**
2748	* Used by TB code to map unsigned 64-bit data for atomic read-write w/ flat
2749	* address.
2750	*/
2751	IEM_DECL_NATIVE_HLP_DEF(uint64_t , iemNativeHlpMemFlatMapDataU64Atomic,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo, RTGCPTR GCPtrMem))
2752	{
2753	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2754	return iemMemMapDataU64AtSafeJmp(pVCpu, pbUnmapInfo, UINT8_MAX, GCPtrMem);
2755	#else
2756	return iemMemFlatMapDataU64AtJmp(pVCpu, pbUnmapInfo, GCPtrMem);
2757	#endif
2758	}
2759
2760
2761	/**
2762	* Used by TB code to map unsigned 64-bit data read-write w/ flat address.
2763	*/
2764	IEM_DECL_NATIVE_HLP_DEF(uint64_t , iemNativeHlpMemFlatMapDataU64Rw,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo, RTGCPTR GCPtrMem))
2765	{
2766	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2767	return iemMemMapDataU64RwSafeJmp(pVCpu, pbUnmapInfo, UINT8_MAX, GCPtrMem);
2768	#else
2769	return iemMemFlatMapDataU64RwJmp(pVCpu, pbUnmapInfo, GCPtrMem);
2770	#endif
2771	}
2772
2773
2774	/**
2775	* Used by TB code to map unsigned 64-bit data writeonly w/ flat address.
2776	*/
2777	IEM_DECL_NATIVE_HLP_DEF(uint64_t , iemNativeHlpMemFlatMapDataU64Wo,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo, RTGCPTR GCPtrMem))
2778	{
2779	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2780	return iemMemMapDataU64WoSafeJmp(pVCpu, pbUnmapInfo, UINT8_MAX, GCPtrMem);
2781	#else
2782	return iemMemFlatMapDataU64WoJmp(pVCpu, pbUnmapInfo, GCPtrMem);
2783	#endif
2784	}
2785
2786
2787	/**
2788	* Used by TB code to map unsigned 64-bit data readonly w/ flat address.
2789	*/
2790	IEM_DECL_NATIVE_HLP_DEF(uint64_t const , iemNativeHlpMemFlatMapDataU64Ro,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo, RTGCPTR GCPtrMem))
2791	{
2792	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2793	return iemMemMapDataU64RoSafeJmp(pVCpu, pbUnmapInfo, UINT8_MAX, GCPtrMem);
2794	#else
2795	return iemMemFlatMapDataU64RoJmp(pVCpu, pbUnmapInfo, GCPtrMem);
2796	#endif
2797	}
2798
2799
2800	/**
2801	* Used by TB code to map 80-bit float data writeonly w/ flat address.
2802	*/
2803	IEM_DECL_NATIVE_HLP_DEF(RTFLOAT80U , iemNativeHlpMemFlatMapDataR80Wo,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo, RTGCPTR GCPtrMem))
2804	{
2805	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2806	return iemMemMapDataR80WoSafeJmp(pVCpu, pbUnmapInfo, UINT8_MAX, GCPtrMem);
2807	#else
2808	return iemMemFlatMapDataR80WoJmp(pVCpu, pbUnmapInfo, GCPtrMem);
2809	#endif
2810	}
2811
2812
2813	/**
2814	* Used by TB code to map 80-bit BCD data writeonly w/ flat address.
2815	*/
2816	IEM_DECL_NATIVE_HLP_DEF(RTPBCD80U , iemNativeHlpMemFlatMapDataD80Wo,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo, RTGCPTR GCPtrMem))
2817	{
2818	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2819	return iemMemMapDataD80WoSafeJmp(pVCpu, pbUnmapInfo, UINT8_MAX, GCPtrMem);
2820	#else
2821	return iemMemFlatMapDataD80WoJmp(pVCpu, pbUnmapInfo, GCPtrMem);
2822	#endif
2823	}
2824
2825
2826	/**
2827	* Used by TB code to map unsigned 128-bit data for atomic read-write w/ flat
2828	* address.
2829	*/
2830	IEM_DECL_NATIVE_HLP_DEF(RTUINT128U , iemNativeHlpMemFlatMapDataU128Atomic,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo, RTGCPTR GCPtrMem))
2831	{
2832	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2833	return iemMemMapDataU128AtSafeJmp(pVCpu, pbUnmapInfo, UINT8_MAX, GCPtrMem);
2834	#else
2835	return iemMemFlatMapDataU128AtJmp(pVCpu, pbUnmapInfo, GCPtrMem);
2836	#endif
2837	}
2838
2839
2840	/**
2841	* Used by TB code to map unsigned 128-bit data read-write w/ flat address.
2842	*/
2843	IEM_DECL_NATIVE_HLP_DEF(RTUINT128U , iemNativeHlpMemFlatMapDataU128Rw,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo, RTGCPTR GCPtrMem))
2844	{
2845	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2846	return iemMemMapDataU128RwSafeJmp(pVCpu, pbUnmapInfo, UINT8_MAX, GCPtrMem);
2847	#else
2848	return iemMemFlatMapDataU128RwJmp(pVCpu, pbUnmapInfo, GCPtrMem);
2849	#endif
2850	}
2851
2852
2853	/**
2854	* Used by TB code to map unsigned 128-bit data writeonly w/ flat address.
2855	*/
2856	IEM_DECL_NATIVE_HLP_DEF(RTUINT128U , iemNativeHlpMemFlatMapDataU128Wo,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo, RTGCPTR GCPtrMem))
2857	{
2858	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2859	return iemMemMapDataU128WoSafeJmp(pVCpu, pbUnmapInfo, UINT8_MAX, GCPtrMem);
2860	#else
2861	return iemMemFlatMapDataU128WoJmp(pVCpu, pbUnmapInfo, GCPtrMem);
2862	#endif
2863	}
2864
2865
2866	/**
2867	* Used by TB code to map unsigned 128-bit data readonly w/ flat address.
2868	*/
2869	IEM_DECL_NATIVE_HLP_DEF(RTUINT128U const , iemNativeHlpMemFlatMapDataU128Ro,(PVMCPUCC pVCpu, uint8_t pbUnmapInfo, RTGCPTR GCPtrMem))
2870	{
2871	#ifdef IEMNATIVE_WITH_TLB_LOOKUP_MAPPED
2872	return iemMemMapDataU128RoSafeJmp(pVCpu, pbUnmapInfo, UINT8_MAX, GCPtrMem);
2873	#else
2874	return iemMemFlatMapDataU128RoJmp(pVCpu, pbUnmapInfo, GCPtrMem);
2875	#endif
2876	}
2877
2878
2879	/*********************************************************************************************************************************
2880	* Helpers: Commit, rollback & unmap *
2881	*********************************************************************************************************************************/
2882
2883	/**
2884	* Used by TB code to commit and unmap a read-write memory mapping.
2885	*/
2886	IEM_DECL_NATIVE_HLP_DEF(void, iemNativeHlpMemCommitAndUnmapAtomic,(PVMCPUCC pVCpu, uint8_t bUnmapInfo))
2887	{
2888	return iemMemCommitAndUnmapAtSafeJmp(pVCpu, bUnmapInfo);
2889	}
2890
2891
2892	/**
2893	* Used by TB code to commit and unmap a read-write memory mapping.
2894	*/
2895	IEM_DECL_NATIVE_HLP_DEF(void, iemNativeHlpMemCommitAndUnmapRw,(PVMCPUCC pVCpu, uint8_t bUnmapInfo))
2896	{
2897	return iemMemCommitAndUnmapRwSafeJmp(pVCpu, bUnmapInfo);
2898	}
2899
2900
2901	/**
2902	* Used by TB code to commit and unmap a write-only memory mapping.
2903	*/
2904	IEM_DECL_NATIVE_HLP_DEF(void, iemNativeHlpMemCommitAndUnmapWo,(PVMCPUCC pVCpu, uint8_t bUnmapInfo))
2905	{
2906	return iemMemCommitAndUnmapWoSafeJmp(pVCpu, bUnmapInfo);
2907	}
2908
2909
2910	/**
2911	* Used by TB code to commit and unmap a read-only memory mapping.
2912	*/
2913	IEM_DECL_NATIVE_HLP_DEF(void, iemNativeHlpMemCommitAndUnmapRo,(PVMCPUCC pVCpu, uint8_t bUnmapInfo))
2914	{
2915	return iemMemCommitAndUnmapRoSafeJmp(pVCpu, bUnmapInfo);
2916	}
2917
2918
2919	/**
2920	* Reinitializes the native recompiler state.
2921	*
2922	* Called before starting a new recompile job.
2923	*/
2924	static PIEMRECOMPILERSTATE iemNativeReInit(PIEMRECOMPILERSTATE pReNative, PCIEMTB pTb)
2925	{
2926	pReNative->cLabels = 0;
2927	pReNative->bmLabelTypes = 0;
2928	pReNative->cFixups = 0;
2929	#ifdef IEMNATIVE_WITH_TB_DEBUG_INFO
2930	pReNative->pDbgInfo->cEntries = 0;
2931	#endif
2932	pReNative->pTbOrg = pTb;
2933	pReNative->cCondDepth = 0;
2934	pReNative->uCondSeqNo = 0;
2935	pReNative->uCheckIrqSeqNo = 0;
2936	pReNative->uTlbSeqNo = 0;
2937
2938	#ifdef IEMNATIVE_WITH_DELAYED_PC_UPDATING
2939	pReNative->Core.offPc = 0;
2940	pReNative->Core.cInstrPcUpdateSkipped = 0;
2941	#endif
2942	pReNative->Core.bmHstRegs = IEMNATIVE_REG_FIXED_MASK
2943	#if IEMNATIVE_HST_GREG_COUNT < 32
2944	\| ~(RT_BIT(IEMNATIVE_HST_GREG_COUNT) - 1U)
2945	#endif
2946	;
2947	pReNative->Core.bmHstRegsWithGstShadow = 0;
2948	pReNative->Core.bmGstRegShadows = 0;
2949	pReNative->Core.bmVars = 0;
2950	pReNative->Core.bmStack = 0;
2951	AssertCompile(sizeof(pReNative->Core.bmStack) * 8 == IEMNATIVE_FRAME_VAR_SLOTS); /* Must set reserved slots to 1 otherwise. */
2952	pReNative->Core.u64ArgVars = UINT64_MAX;
2953
2954	AssertCompile(RT_ELEMENTS(pReNative->aidxUniqueLabels) == 13);
2955	pReNative->aidxUniqueLabels[0] = UINT32_MAX;
2956	pReNative->aidxUniqueLabels[1] = UINT32_MAX;
2957	pReNative->aidxUniqueLabels[2] = UINT32_MAX;
2958	pReNative->aidxUniqueLabels[3] = UINT32_MAX;
2959	pReNative->aidxUniqueLabels[4] = UINT32_MAX;
2960	pReNative->aidxUniqueLabels[5] = UINT32_MAX;
2961	pReNative->aidxUniqueLabels[6] = UINT32_MAX;
2962	pReNative->aidxUniqueLabels[7] = UINT32_MAX;
2963	pReNative->aidxUniqueLabels[8] = UINT32_MAX;
2964	pReNative->aidxUniqueLabels[9] = UINT32_MAX;
2965	pReNative->aidxUniqueLabels[10] = UINT32_MAX;
2966	pReNative->aidxUniqueLabels[11] = UINT32_MAX;
2967	pReNative->aidxUniqueLabels[12] = UINT32_MAX;
2968
2969	/* Full host register reinit: */
2970	for (unsigned i = 0; i < RT_ELEMENTS(pReNative->Core.aHstRegs); i++)
2971	{
2972	pReNative->Core.aHstRegs[i].fGstRegShadows = 0;
2973	pReNative->Core.aHstRegs[i].enmWhat = kIemNativeWhat_Invalid;
2974	pReNative->Core.aHstRegs[i].idxVar = UINT8_MAX;
2975	}
2976
2977	uint32_t fRegs = IEMNATIVE_REG_FIXED_MASK
2978	& ~( RT_BIT_32(IEMNATIVE_REG_FIXED_PVMCPU)
2979	#ifdef IEMNATIVE_REG_FIXED_PCPUMCTX
2980	\| RT_BIT_32(IEMNATIVE_REG_FIXED_PCPUMCTX)
2981	#endif
2982	#ifdef IEMNATIVE_REG_FIXED_PCPUMCTX
2983	\| RT_BIT_32(IEMNATIVE_REG_FIXED_TMP0)
2984	#endif
2985	#ifdef IEMNATIVE_REG_FIXED_TMP1
2986	\| RT_BIT_32(IEMNATIVE_REG_FIXED_TMP1)
2987	#endif
2988	#ifdef IEMNATIVE_REG_FIXED_PC_DBG
2989	\| RT_BIT_32(IEMNATIVE_REG_FIXED_PC_DBG)
2990	#endif
2991	);
2992	for (uint32_t idxReg = ASMBitFirstSetU32(fRegs) - 1; fRegs != 0; idxReg = ASMBitFirstSetU32(fRegs) - 1)
2993	{
2994	fRegs &= ~RT_BIT_32(idxReg);
2995	pReNative->Core.aHstRegs[IEMNATIVE_REG_FIXED_PVMCPU].enmWhat = kIemNativeWhat_FixedReserved;
2996	}
2997
2998	pReNative->Core.aHstRegs[IEMNATIVE_REG_FIXED_PVMCPU].enmWhat = kIemNativeWhat_pVCpuFixed;
2999	#ifdef IEMNATIVE_REG_FIXED_PCPUMCTX
3000	pReNative->Core.aHstRegs[IEMNATIVE_REG_FIXED_PCPUMCTX].enmWhat = kIemNativeWhat_pCtxFixed;
3001	#endif
3002	#ifdef IEMNATIVE_REG_FIXED_TMP0
3003	pReNative->Core.aHstRegs[IEMNATIVE_REG_FIXED_TMP0].enmWhat = kIemNativeWhat_FixedTmp;
3004	#endif
3005	#ifdef IEMNATIVE_REG_FIXED_TMP1
3006	pReNative->Core.aHstRegs[IEMNATIVE_REG_FIXED_TMP1].enmWhat = kIemNativeWhat_FixedTmp;
3007	#endif
3008	#ifdef IEMNATIVE_REG_FIXED_PC_DBG
3009	pReNative->Core.aHstRegs[IEMNATIVE_REG_FIXED_PC_DBG].enmWhat = kIemNativeWhat_PcShadow;
3010	#endif
3011
3012	#ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
3013	# ifdef RT_ARCH_ARM64
3014	/*
3015	* Arm64 has 32 128-bit registers only, in order to support emulating 256-bit registers we pair
3016	* two real registers statically to one virtual for now, leaving us with only 16 256-bit registers.
3017	* We always pair v0 with v1, v2 with v3, etc. so we mark the higher register as fixed here during init
3018	* and the register allocator assumes that it will be always free when the lower is picked.
3019	*/
3020	uint32_t const fFixedAdditional = UINT32_C(0xaaaaaaaa);
3021	# else
3022	uint32_t const fFixedAdditional = 0;
3023	# endif
3024
3025	pReNative->Core.bmHstSimdRegs = IEMNATIVE_SIMD_REG_FIXED_MASK
3026	\| fFixedAdditional
3027	# if IEMNATIVE_HST_SIMD_REG_COUNT < 32
3028	\| ~(RT_BIT(IEMNATIVE_HST_SIMD_REG_COUNT) - 1U)
3029	# endif
3030	;
3031	pReNative->Core.bmHstSimdRegsWithGstShadow = 0;
3032	pReNative->Core.bmGstSimdRegShadows = 0;
3033	pReNative->Core.bmGstSimdRegShadowDirtyLo128 = 0;
3034	pReNative->Core.bmGstSimdRegShadowDirtyHi128 = 0;
3035
3036	/* Full host register reinit: */
3037	for (unsigned i = 0; i < RT_ELEMENTS(pReNative->Core.aHstSimdRegs); i++)
3038	{
3039	pReNative->Core.aHstSimdRegs[i].fGstRegShadows = 0;
3040	pReNative->Core.aHstSimdRegs[i].enmWhat = kIemNativeWhat_Invalid;
3041	pReNative->Core.aHstSimdRegs[i].enmLoaded = kIemNativeGstSimdRegLdStSz_Invalid;
3042	}
3043
3044	fRegs = IEMNATIVE_SIMD_REG_FIXED_MASK \| fFixedAdditional;
3045	for (uint32_t idxReg = ASMBitFirstSetU32(fRegs) - 1; fRegs != 0; idxReg = ASMBitFirstSetU32(fRegs) - 1)
3046	{
3047	fRegs &= ~RT_BIT_32(idxReg);
3048	pReNative->Core.aHstSimdRegs[idxReg].enmWhat = kIemNativeWhat_FixedReserved;
3049	}
3050
3051	#ifdef IEMNATIVE_SIMD_REG_FIXED_TMP0
3052	pReNative->Core.aHstSimdRegs[IEMNATIVE_SIMD_REG_FIXED_TMP0].enmWhat = kIemNativeWhat_FixedTmp;
3053	#endif
3054
3055	#endif
3056
3057	return pReNative;
3058	}
3059
3060
3061	/**
3062	* Allocates and initializes the native recompiler state.
3063	*
3064	* This is called the first time an EMT wants to recompile something.
3065	*
3066	* @returns Pointer to the new recompiler state.
3067	* @param pVCpu The cross context virtual CPU structure of the calling
3068	* thread.
3069	* @param pTb The TB that's about to be recompiled.
3070	* @thread EMT(pVCpu)
3071	*/
3072	static PIEMRECOMPILERSTATE iemNativeInit(PVMCPUCC pVCpu, PCIEMTB pTb)
3073	{
3074	VMCPU_ASSERT_EMT(pVCpu);
3075
3076	PIEMRECOMPILERSTATE pReNative = (PIEMRECOMPILERSTATE)RTMemAllocZ(sizeof(*pReNative));
3077	AssertReturn(pReNative, NULL);
3078
3079	/*
3080	* Try allocate all the buffers and stuff we need.
3081	*/
3082	pReNative->pInstrBuf = (PIEMNATIVEINSTR)RTMemAllocZ(_64K);
3083	pReNative->paLabels = (PIEMNATIVELABEL)RTMemAllocZ(sizeof(IEMNATIVELABEL) * _8K);
3084	pReNative->paFixups = (PIEMNATIVEFIXUP)RTMemAllocZ(sizeof(IEMNATIVEFIXUP) * _16K);
3085	#ifdef IEMNATIVE_WITH_TB_DEBUG_INFO
3086	pReNative->pDbgInfo = (PIEMTBDBG)RTMemAllocZ(RT_UOFFSETOF_DYN(IEMTBDBG, aEntries[_16K]));
3087	#endif
3088	if (RT_LIKELY( pReNative->pInstrBuf
3089	&& pReNative->paLabels
3090	&& pReNative->paFixups)
3091	#ifdef IEMNATIVE_WITH_TB_DEBUG_INFO
3092	&& pReNative->pDbgInfo
3093	#endif
3094	)
3095	{
3096	/*
3097	* Set the buffer & array sizes on success.
3098	*/
3099	pReNative->cInstrBufAlloc = _64K / sizeof(IEMNATIVEINSTR);
3100	pReNative->cLabelsAlloc = _8K;
3101	pReNative->cFixupsAlloc = _16K;
3102	#ifdef IEMNATIVE_WITH_TB_DEBUG_INFO
3103	pReNative->cDbgInfoAlloc = _16K;
3104	#endif
3105
3106	/* Other constant stuff: */
3107	pReNative->pVCpu = pVCpu;
3108
3109	/*
3110	* Done, just need to save it and reinit it.
3111	*/
3112	pVCpu->iem.s.pNativeRecompilerStateR3 = pReNative;
3113	return iemNativeReInit(pReNative, pTb);
3114	}
3115
3116	/*
3117	* Failed. Cleanup and return.
3118	*/
3119	AssertFailed();
3120	RTMemFree(pReNative->pInstrBuf);
3121	RTMemFree(pReNative->paLabels);
3122	RTMemFree(pReNative->paFixups);
3123	#ifdef IEMNATIVE_WITH_TB_DEBUG_INFO
3124	RTMemFree(pReNative->pDbgInfo);
3125	#endif
3126	RTMemFree(pReNative);
3127	return NULL;
3128	}
3129
3130
3131	/**
3132	* Creates a label
3133	*
3134	* If the label does not yet have a defined position,
3135	* call iemNativeLabelDefine() later to set it.
3136	*
3137	* @returns Label ID. Throws VBox status code on failure, so no need to check
3138	* the return value.
3139	* @param pReNative The native recompile state.
3140	* @param enmType The label type.
3141	* @param offWhere The instruction offset of the label. UINT32_MAX if the
3142	* label is not yet defined (default).
3143	* @param uData Data associated with the lable. Only applicable to
3144	* certain type of labels. Default is zero.
3145	*/
3146	DECL_HIDDEN_THROW(uint32_t)
3147	iemNativeLabelCreate(PIEMRECOMPILERSTATE pReNative, IEMNATIVELABELTYPE enmType,
3148	uint32_t offWhere /= UINT32_MAX/, uint16_t uData /= 0/)
3149	{
3150	Assert(uData == 0 \|\| enmType >= kIemNativeLabelType_FirstWithMultipleInstances);
3151
3152	/*
3153	* Locate existing label definition.
3154	*
3155	* This is only allowed for forward declarations where offWhere=UINT32_MAX
3156	* and uData is zero.
3157	*/
3158	PIEMNATIVELABEL paLabels = pReNative->paLabels;
3159	uint32_t const cLabels = pReNative->cLabels;
3160	if ( pReNative->bmLabelTypes & RT_BIT_64(enmType)
3161	#ifndef VBOX_STRICT
3162	&& enmType < kIemNativeLabelType_FirstWithMultipleInstances
3163	&& offWhere == UINT32_MAX
3164	&& uData == 0
3165	#endif
3166	)
3167	{
3168	#ifndef VBOX_STRICT
3169	AssertStmt(enmType > kIemNativeLabelType_Invalid && enmType < kIemNativeLabelType_FirstWithMultipleInstances,
3170	IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_LABEL_IPE_1));
3171	uint32_t const idxLabel = pReNative->aidxUniqueLabels[enmType];
3172	if (idxLabel < pReNative->cLabels)
3173	return idxLabel;
3174	#else
3175	for (uint32_t i = 0; i < cLabels; i++)
3176	if ( paLabels[i].enmType == enmType
3177	&& paLabels[i].uData == uData)
3178	{
3179	AssertStmt(uData == 0, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_LABEL_IPE_1));
3180	AssertStmt(offWhere == UINT32_MAX, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_LABEL_IPE_1));
3181	AssertStmt(paLabels[i].off == UINT32_MAX, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_LABEL_IPE_2));
3182	AssertStmt(enmType < kIemNativeLabelType_FirstWithMultipleInstances && pReNative->aidxUniqueLabels[enmType] == i,
3183	IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_LABEL_IPE_1));
3184	return i;
3185	}
3186	AssertStmt( enmType >= kIemNativeLabelType_FirstWithMultipleInstances
3187	\|\| pReNative->aidxUniqueLabels[enmType] == UINT32_MAX, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_LABEL_IPE_1));
3188	#endif
3189	}
3190
3191	/*
3192	* Make sure we've got room for another label.
3193	*/
3194	if (RT_LIKELY(cLabels < pReNative->cLabelsAlloc))
3195	{ /* likely */ }
3196	else
3197	{
3198	uint32_t cNew = pReNative->cLabelsAlloc;
3199	AssertStmt(cNew, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_LABEL_IPE_3));
3200	AssertStmt(cLabels == cNew, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_LABEL_IPE_3));
3201	cNew *= 2;
3202	AssertStmt(cNew <= _64K, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_LABEL_TOO_MANY)); /* IEMNATIVEFIXUP::idxLabel type restrict this */
3203	paLabels = (PIEMNATIVELABEL)RTMemRealloc(paLabels, cNew * sizeof(paLabels[0]));
3204	AssertStmt(paLabels, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_LABEL_OUT_OF_MEMORY));
3205	pReNative->paLabels = paLabels;
3206	pReNative->cLabelsAlloc = cNew;
3207	}
3208
3209	/*
3210	* Define a new label.
3211	*/
3212	paLabels[cLabels].off = offWhere;
3213	paLabels[cLabels].enmType = enmType;
3214	paLabels[cLabels].uData = uData;
3215	pReNative->cLabels = cLabels + 1;
3216
3217	Assert((unsigned)enmType < 64);
3218	pReNative->bmLabelTypes \|= RT_BIT_64(enmType);
3219
3220	if (enmType < kIemNativeLabelType_FirstWithMultipleInstances)
3221	{
3222	Assert(uData == 0);
3223	pReNative->aidxUniqueLabels[enmType] = cLabels;
3224	}
3225
3226	if (offWhere != UINT32_MAX)
3227	{
3228	#ifdef IEMNATIVE_WITH_TB_DEBUG_INFO
3229	iemNativeDbgInfoAddNativeOffset(pReNative, offWhere);
3230	iemNativeDbgInfoAddLabel(pReNative, enmType, uData);
3231	#endif
3232	}
3233	return cLabels;
3234	}
3235
3236
3237	/**
3238	* Defines the location of an existing label.
3239	*
3240	* @param pReNative The native recompile state.
3241	* @param idxLabel The label to define.
3242	* @param offWhere The position.
3243	*/
3244	DECL_HIDDEN_THROW(void) iemNativeLabelDefine(PIEMRECOMPILERSTATE pReNative, uint32_t idxLabel, uint32_t offWhere)
3245	{
3246	AssertStmt(idxLabel < pReNative->cLabels, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_LABEL_IPE_4));
3247	PIEMNATIVELABEL const pLabel = &pReNative->paLabels[idxLabel];
3248	AssertStmt(pLabel->off == UINT32_MAX, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_LABEL_IPE_5));
3249	pLabel->off = offWhere;
3250	#ifdef IEMNATIVE_WITH_TB_DEBUG_INFO
3251	iemNativeDbgInfoAddNativeOffset(pReNative, offWhere);
3252	iemNativeDbgInfoAddLabel(pReNative, (IEMNATIVELABELTYPE)pLabel->enmType, pLabel->uData);
3253	#endif
3254	}
3255
3256
3257	/**
3258	* Looks up a lable.
3259	*
3260	* @returns Label ID if found, UINT32_MAX if not.
3261	*/
3262	static uint32_t iemNativeLabelFind(PIEMRECOMPILERSTATE pReNative, IEMNATIVELABELTYPE enmType,
3263	uint32_t offWhere = UINT32_MAX, uint16_t uData = 0) RT_NOEXCEPT
3264	{
3265	Assert((unsigned)enmType < 64);
3266	if (RT_BIT_64(enmType) & pReNative->bmLabelTypes)
3267	{
3268	if (enmType < kIemNativeLabelType_FirstWithMultipleInstances)
3269	return pReNative->aidxUniqueLabels[enmType];
3270
3271	PIEMNATIVELABEL paLabels = pReNative->paLabels;
3272	uint32_t const cLabels = pReNative->cLabels;
3273	for (uint32_t i = 0; i < cLabels; i++)
3274	if ( paLabels[i].enmType == enmType
3275	&& paLabels[i].uData == uData
3276	&& ( paLabels[i].off == offWhere
3277	\|\| offWhere == UINT32_MAX
3278	\|\| paLabels[i].off == UINT32_MAX))
3279	return i;
3280	}
3281	return UINT32_MAX;
3282	}
3283
3284
3285	/**
3286	* Adds a fixup.
3287	*
3288	* @throws VBox status code (int) on failure.
3289	* @param pReNative The native recompile state.
3290	* @param offWhere The instruction offset of the fixup location.
3291	* @param idxLabel The target label ID for the fixup.
3292	* @param enmType The fixup type.
3293	* @param offAddend Fixup addend if applicable to the type. Default is 0.
3294	*/
3295	DECL_HIDDEN_THROW(void)
3296	iemNativeAddFixup(PIEMRECOMPILERSTATE pReNative, uint32_t offWhere, uint32_t idxLabel,
3297	IEMNATIVEFIXUPTYPE enmType, int8_t offAddend /= 0/)
3298	{
3299	Assert(idxLabel <= UINT16_MAX);
3300	Assert((unsigned)enmType <= UINT8_MAX);
3301
3302	/*
3303	* Make sure we've room.
3304	*/
3305	PIEMNATIVEFIXUP paFixups = pReNative->paFixups;
3306	uint32_t const cFixups = pReNative->cFixups;
3307	if (RT_LIKELY(cFixups < pReNative->cFixupsAlloc))
3308	{ /* likely */ }
3309	else
3310	{
3311	uint32_t cNew = pReNative->cFixupsAlloc;
3312	AssertStmt(cNew, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_FIXUP_IPE_1));
3313	AssertStmt(cFixups == cNew, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_FIXUP_IPE_1));
3314	cNew *= 2;
3315	AssertStmt(cNew <= _128K, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_FIXUP_TOO_MANY));
3316	paFixups = (PIEMNATIVEFIXUP)RTMemRealloc(paFixups, cNew * sizeof(paFixups[0]));
3317	AssertStmt(paFixups, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_FIXUP_OUT_OF_MEMORY));
3318	pReNative->paFixups = paFixups;
3319	pReNative->cFixupsAlloc = cNew;
3320	}
3321
3322	/*
3323	* Add the fixup.
3324	*/
3325	paFixups[cFixups].off = offWhere;
3326	paFixups[cFixups].idxLabel = (uint16_t)idxLabel;
3327	paFixups[cFixups].enmType = enmType;
3328	paFixups[cFixups].offAddend = offAddend;
3329	pReNative->cFixups = cFixups + 1;
3330	}
3331
3332
3333	/**
3334	* Slow code path for iemNativeInstrBufEnsure.
3335	*/
3336	DECL_HIDDEN_THROW(PIEMNATIVEINSTR) iemNativeInstrBufEnsureSlow(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint32_t cInstrReq)
3337	{
3338	/* Double the buffer size till we meet the request. */
3339	uint32_t cNew = pReNative->cInstrBufAlloc;
3340	AssertStmt(cNew > 0, IEMNATIVE_DO_LONGJMP(pReNative, VERR_INTERNAL_ERROR_5)); /* impossible */
3341	do
3342	cNew *= 2;
3343	while (cNew < off + cInstrReq);
3344
3345	uint32_t const cbNew = cNew * sizeof(IEMNATIVEINSTR);
3346	#ifdef RT_ARCH_ARM64
3347	uint32_t const cbMaxInstrBuf = _1M; /* Limited by the branch instruction range (18+2 bits). */
3348	#else
3349	uint32_t const cbMaxInstrBuf = _2M;
3350	#endif
3351	AssertStmt(cbNew <= cbMaxInstrBuf, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_INSTR_BUF_TOO_LARGE));
3352
3353	void *pvNew = RTMemRealloc(pReNative->pInstrBuf, cbNew);
3354	AssertStmt(pvNew, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_INSTR_BUF_OUT_OF_MEMORY));
3355
3356	#ifdef VBOX_STRICT
3357	pReNative->offInstrBufChecked = off + cInstrReq;
3358	#endif
3359	pReNative->cInstrBufAlloc = cNew;
3360	return pReNative->pInstrBuf = (PIEMNATIVEINSTR)pvNew;
3361	}
3362
3363	#ifdef IEMNATIVE_WITH_TB_DEBUG_INFO
3364
3365	/**
3366	* Grows the static debug info array used during recompilation.
3367	*
3368	* @returns Pointer to the new debug info block; throws VBox status code on
3369	* failure, so no need to check the return value.
3370	*/
3371	DECL_NO_INLINE(static, PIEMTBDBG) iemNativeDbgInfoGrow(PIEMRECOMPILERSTATE pReNative, PIEMTBDBG pDbgInfo)
3372	{
3373	uint32_t cNew = pReNative->cDbgInfoAlloc * 2;
3374	AssertStmt(cNew < _1M && cNew != 0, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_DBGINFO_IPE_1));
3375	pDbgInfo = (PIEMTBDBG)RTMemRealloc(pDbgInfo, RT_UOFFSETOF_DYN(IEMTBDBG, aEntries[cNew]));
3376	AssertStmt(pDbgInfo, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_DBGINFO_OUT_OF_MEMORY));
3377	pReNative->pDbgInfo = pDbgInfo;
3378	pReNative->cDbgInfoAlloc = cNew;
3379	return pDbgInfo;
3380	}
3381
3382
3383	/**
3384	* Adds a new debug info uninitialized entry, returning the pointer to it.
3385	*/
3386	DECL_INLINE_THROW(PIEMTBDBGENTRY) iemNativeDbgInfoAddNewEntry(PIEMRECOMPILERSTATE pReNative, PIEMTBDBG pDbgInfo)
3387	{
3388	if (RT_LIKELY(pDbgInfo->cEntries < pReNative->cDbgInfoAlloc))
3389	{ /* likely */ }
3390	else
3391	pDbgInfo = iemNativeDbgInfoGrow(pReNative, pDbgInfo);
3392	return &pDbgInfo->aEntries[pDbgInfo->cEntries++];
3393	}
3394
3395
3396	/**
3397	* Debug Info: Adds a native offset record, if necessary.
3398	*/
3399	DECL_HIDDEN_THROW(void) iemNativeDbgInfoAddNativeOffset(PIEMRECOMPILERSTATE pReNative, uint32_t off)
3400	{
3401	PIEMTBDBG pDbgInfo = pReNative->pDbgInfo;
3402
3403	/*
3404	* Search backwards to see if we've got a similar record already.
3405	*/
3406	uint32_t idx = pDbgInfo->cEntries;
3407	uint32_t idxStop = idx > 8 ? idx - 8 : 0;
3408	while (idx-- > idxStop)
3409	if (pDbgInfo->aEntries[idx].Gen.uType == kIemTbDbgEntryType_NativeOffset)
3410	{
3411	if (pDbgInfo->aEntries[idx].NativeOffset.offNative == off)
3412	return;
3413	AssertStmt(pDbgInfo->aEntries[idx].NativeOffset.offNative < off,
3414	IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_DBGINFO_IPE_2));
3415	break;
3416	}
3417
3418	/*
3419	* Add it.
3420	*/
3421	PIEMTBDBGENTRY const pEntry = iemNativeDbgInfoAddNewEntry(pReNative, pDbgInfo);
3422	pEntry->NativeOffset.uType = kIemTbDbgEntryType_NativeOffset;
3423	pEntry->NativeOffset.offNative = off;
3424	}
3425
3426
3427	/**
3428	* Debug Info: Record info about a label.
3429	*/
3430	static void iemNativeDbgInfoAddLabel(PIEMRECOMPILERSTATE pReNative, IEMNATIVELABELTYPE enmType, uint16_t uData)
3431	{
3432	PIEMTBDBGENTRY const pEntry = iemNativeDbgInfoAddNewEntry(pReNative, pReNative->pDbgInfo);
3433	pEntry->Label.uType = kIemTbDbgEntryType_Label;
3434	pEntry->Label.uUnused = 0;
3435	pEntry->Label.enmLabel = (uint8_t)enmType;
3436	pEntry->Label.uData = uData;
3437	}
3438
3439
3440	/**
3441	* Debug Info: Record info about a threaded call.
3442	*/
3443	static void iemNativeDbgInfoAddThreadedCall(PIEMRECOMPILERSTATE pReNative, IEMTHREADEDFUNCS enmCall, bool fRecompiled)
3444	{
3445	PIEMTBDBGENTRY const pEntry = iemNativeDbgInfoAddNewEntry(pReNative, pReNative->pDbgInfo);
3446	pEntry->ThreadedCall.uType = kIemTbDbgEntryType_ThreadedCall;
3447	pEntry->ThreadedCall.fRecompiled = fRecompiled;
3448	pEntry->ThreadedCall.uUnused = 0;
3449	pEntry->ThreadedCall.enmCall = (uint16_t)enmCall;
3450	}
3451
3452
3453	/**
3454	* Debug Info: Record info about a new guest instruction.
3455	*/
3456	static void iemNativeDbgInfoAddGuestInstruction(PIEMRECOMPILERSTATE pReNative, uint32_t fExec)
3457	{
3458	PIEMTBDBGENTRY const pEntry = iemNativeDbgInfoAddNewEntry(pReNative, pReNative->pDbgInfo);
3459	pEntry->GuestInstruction.uType = kIemTbDbgEntryType_GuestInstruction;
3460	pEntry->GuestInstruction.uUnused = 0;
3461	pEntry->GuestInstruction.fExec = fExec;
3462	}
3463
3464
3465	/**
3466	* Debug Info: Record info about guest register shadowing.
3467	*/
3468	DECL_HIDDEN_THROW(void)
3469	iemNativeDbgInfoAddGuestRegShadowing(PIEMRECOMPILERSTATE pReNative, IEMNATIVEGSTREG enmGstReg,
3470	uint8_t idxHstReg /= UINT8_MAX/, uint8_t idxHstRegPrev /= UINT8_MAX/)
3471	{
3472	PIEMTBDBGENTRY const pEntry = iemNativeDbgInfoAddNewEntry(pReNative, pReNative->pDbgInfo);
3473	pEntry->GuestRegShadowing.uType = kIemTbDbgEntryType_GuestRegShadowing;
3474	pEntry->GuestRegShadowing.uUnused = 0;
3475	pEntry->GuestRegShadowing.idxGstReg = enmGstReg;
3476	pEntry->GuestRegShadowing.idxHstReg = idxHstReg;
3477	pEntry->GuestRegShadowing.idxHstRegPrev = idxHstRegPrev;
3478	}
3479
3480
3481	# ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
3482	/**
3483	* Debug Info: Record info about guest register shadowing.
3484	*/
3485	DECL_HIDDEN_THROW(void)
3486	iemNativeDbgInfoAddGuestSimdRegShadowing(PIEMRECOMPILERSTATE pReNative, IEMNATIVEGSTSIMDREG enmGstSimdReg,
3487	uint8_t idxHstSimdReg /= UINT8_MAX/, uint8_t idxHstSimdRegPrev /= UINT8_MAX/)
3488	{
3489	PIEMTBDBGENTRY const pEntry = iemNativeDbgInfoAddNewEntry(pReNative, pReNative->pDbgInfo);
3490	pEntry->GuestSimdRegShadowing.uType = kIemTbDbgEntryType_GuestSimdRegShadowing;
3491	pEntry->GuestSimdRegShadowing.uUnused = 0;
3492	pEntry->GuestSimdRegShadowing.idxGstSimdReg = enmGstSimdReg;
3493	pEntry->GuestSimdRegShadowing.idxHstSimdReg = idxHstSimdReg;
3494	pEntry->GuestSimdRegShadowing.idxHstSimdRegPrev = idxHstSimdRegPrev;
3495	}
3496	# endif
3497
3498
3499	# ifdef IEMNATIVE_WITH_DELAYED_PC_UPDATING
3500	/**
3501	* Debug Info: Record info about delayed RIP updates.
3502	*/
3503	DECL_HIDDEN_THROW(void) iemNativeDbgInfoAddDelayedPcUpdate(PIEMRECOMPILERSTATE pReNative, uint32_t offPc, uint32_t cInstrSkipped)
3504	{
3505	PIEMTBDBGENTRY const pEntry = iemNativeDbgInfoAddNewEntry(pReNative, pReNative->pDbgInfo);
3506	pEntry->DelayedPcUpdate.uType = kIemTbDbgEntryType_DelayedPcUpdate;
3507	pEntry->DelayedPcUpdate.offPc = offPc;
3508	pEntry->DelayedPcUpdate.cInstrSkipped = cInstrSkipped;
3509	}
3510	# endif
3511
3512	#endif /* IEMNATIVE_WITH_TB_DEBUG_INFO */
3513
3514
3515	/*********************************************************************************************************************************
3516	* Register Allocator *
3517	*********************************************************************************************************************************/
3518
3519	/**
3520	* Register parameter indexes (indexed by argument number).
3521	*/
3522	DECL_HIDDEN_CONST(uint8_t) const g_aidxIemNativeCallRegs[] =
3523	{
3524	IEMNATIVE_CALL_ARG0_GREG,
3525	IEMNATIVE_CALL_ARG1_GREG,
3526	IEMNATIVE_CALL_ARG2_GREG,
3527	IEMNATIVE_CALL_ARG3_GREG,
3528	#if defined(IEMNATIVE_CALL_ARG4_GREG)
3529	IEMNATIVE_CALL_ARG4_GREG,
3530	# if defined(IEMNATIVE_CALL_ARG5_GREG)
3531	IEMNATIVE_CALL_ARG5_GREG,
3532	# if defined(IEMNATIVE_CALL_ARG6_GREG)
3533	IEMNATIVE_CALL_ARG6_GREG,
3534	# if defined(IEMNATIVE_CALL_ARG7_GREG)
3535	IEMNATIVE_CALL_ARG7_GREG,
3536	# endif
3537	# endif
3538	# endif
3539	#endif
3540	};
3541	AssertCompile(RT_ELEMENTS(g_aidxIemNativeCallRegs) == IEMNATIVE_CALL_ARG_GREG_COUNT);
3542
3543	/**
3544	* Call register masks indexed by argument count.
3545	*/
3546	DECL_HIDDEN_CONST(uint32_t) const g_afIemNativeCallRegs[] =
3547	{
3548	0,
3549	RT_BIT_32(IEMNATIVE_CALL_ARG0_GREG),
3550	RT_BIT_32(IEMNATIVE_CALL_ARG0_GREG) \| RT_BIT_32(IEMNATIVE_CALL_ARG1_GREG),
3551	RT_BIT_32(IEMNATIVE_CALL_ARG0_GREG) \| RT_BIT_32(IEMNATIVE_CALL_ARG1_GREG) \| RT_BIT_32(IEMNATIVE_CALL_ARG2_GREG),
3552	RT_BIT_32(IEMNATIVE_CALL_ARG0_GREG) \| RT_BIT_32(IEMNATIVE_CALL_ARG1_GREG) \| RT_BIT_32(IEMNATIVE_CALL_ARG2_GREG)
3553	\| RT_BIT_32(IEMNATIVE_CALL_ARG3_GREG),
3554	#if defined(IEMNATIVE_CALL_ARG4_GREG)
3555	RT_BIT_32(IEMNATIVE_CALL_ARG0_GREG) \| RT_BIT_32(IEMNATIVE_CALL_ARG1_GREG) \| RT_BIT_32(IEMNATIVE_CALL_ARG2_GREG)
3556	\| RT_BIT_32(IEMNATIVE_CALL_ARG3_GREG) \| RT_BIT_32(IEMNATIVE_CALL_ARG4_GREG),
3557	# if defined(IEMNATIVE_CALL_ARG5_GREG)
3558	RT_BIT_32(IEMNATIVE_CALL_ARG0_GREG) \| RT_BIT_32(IEMNATIVE_CALL_ARG1_GREG) \| RT_BIT_32(IEMNATIVE_CALL_ARG2_GREG)
3559	\| RT_BIT_32(IEMNATIVE_CALL_ARG3_GREG) \| RT_BIT_32(IEMNATIVE_CALL_ARG4_GREG) \| RT_BIT_32(IEMNATIVE_CALL_ARG5_GREG),
3560	# if defined(IEMNATIVE_CALL_ARG6_GREG)
3561	RT_BIT_32(IEMNATIVE_CALL_ARG0_GREG) \| RT_BIT_32(IEMNATIVE_CALL_ARG1_GREG) \| RT_BIT_32(IEMNATIVE_CALL_ARG2_GREG)
3562	\| RT_BIT_32(IEMNATIVE_CALL_ARG3_GREG) \| RT_BIT_32(IEMNATIVE_CALL_ARG4_GREG) \| RT_BIT_32(IEMNATIVE_CALL_ARG5_GREG)
3563	\| RT_BIT_32(IEMNATIVE_CALL_ARG6_GREG),
3564	# if defined(IEMNATIVE_CALL_ARG7_GREG)
3565	RT_BIT_32(IEMNATIVE_CALL_ARG0_GREG) \| RT_BIT_32(IEMNATIVE_CALL_ARG1_GREG) \| RT_BIT_32(IEMNATIVE_CALL_ARG2_GREG)
3566	\| RT_BIT_32(IEMNATIVE_CALL_ARG3_GREG) \| RT_BIT_32(IEMNATIVE_CALL_ARG4_GREG) \| RT_BIT_32(IEMNATIVE_CALL_ARG5_GREG)
3567	\| RT_BIT_32(IEMNATIVE_CALL_ARG6_GREG) \| RT_BIT_32(IEMNATIVE_CALL_ARG7_GREG),
3568	# endif
3569	# endif
3570	# endif
3571	#endif
3572	};
3573
3574	#ifdef IEMNATIVE_FP_OFF_STACK_ARG0
3575	/**
3576	* BP offset of the stack argument slots.
3577	*
3578	* This array is indexed by \#argument - IEMNATIVE_CALL_ARG_GREG_COUNT and has
3579	* IEMNATIVE_FRAME_STACK_ARG_COUNT entries.
3580	*/
3581	DECL_HIDDEN_CONST(int32_t) const g_aoffIemNativeCallStackArgBpDisp[] =
3582	{
3583	IEMNATIVE_FP_OFF_STACK_ARG0,
3584	# ifdef IEMNATIVE_FP_OFF_STACK_ARG1
3585	IEMNATIVE_FP_OFF_STACK_ARG1,
3586	# endif
3587	# ifdef IEMNATIVE_FP_OFF_STACK_ARG2
3588	IEMNATIVE_FP_OFF_STACK_ARG2,
3589	# endif
3590	# ifdef IEMNATIVE_FP_OFF_STACK_ARG3
3591	IEMNATIVE_FP_OFF_STACK_ARG3,
3592	# endif
3593	};
3594	AssertCompile(RT_ELEMENTS(g_aoffIemNativeCallStackArgBpDisp) == IEMNATIVE_FRAME_STACK_ARG_COUNT);
3595	#endif /* IEMNATIVE_FP_OFF_STACK_ARG0 */
3596
3597	/**
3598	* Info about shadowed guest register values.
3599	* @see IEMNATIVEGSTREG
3600	*/
3601	DECL_HIDDEN_CONST(IEMANTIVEGSTREGINFO const) g_aGstShadowInfo[] =
3602	{
3603	#define CPUMCTX_OFF_AND_SIZE(a_Reg) (uint32_t)RT_UOFFSETOF(VMCPU, cpum.GstCtx. a_Reg), RT_SIZEOFMEMB(VMCPU, cpum.GstCtx. a_Reg)
3604	/* [kIemNativeGstReg_GprFirst + X86_GREG_xAX] = */ { CPUMCTX_OFF_AND_SIZE(rax), "rax", },
3605	/* [kIemNativeGstReg_GprFirst + X86_GREG_xCX] = */ { CPUMCTX_OFF_AND_SIZE(rcx), "rcx", },
3606	/* [kIemNativeGstReg_GprFirst + X86_GREG_xDX] = */ { CPUMCTX_OFF_AND_SIZE(rdx), "rdx", },
3607	/* [kIemNativeGstReg_GprFirst + X86_GREG_xBX] = */ { CPUMCTX_OFF_AND_SIZE(rbx), "rbx", },
3608	/* [kIemNativeGstReg_GprFirst + X86_GREG_xSP] = */ { CPUMCTX_OFF_AND_SIZE(rsp), "rsp", },
3609	/* [kIemNativeGstReg_GprFirst + X86_GREG_xBP] = */ { CPUMCTX_OFF_AND_SIZE(rbp), "rbp", },
3610	/* [kIemNativeGstReg_GprFirst + X86_GREG_xSI] = */ { CPUMCTX_OFF_AND_SIZE(rsi), "rsi", },
3611	/* [kIemNativeGstReg_GprFirst + X86_GREG_xDI] = */ { CPUMCTX_OFF_AND_SIZE(rdi), "rdi", },
3612	/* [kIemNativeGstReg_GprFirst + X86_GREG_x8 ] = */ { CPUMCTX_OFF_AND_SIZE(r8), "r8", },
3613	/* [kIemNativeGstReg_GprFirst + X86_GREG_x9 ] = */ { CPUMCTX_OFF_AND_SIZE(r9), "r9", },
3614	/* [kIemNativeGstReg_GprFirst + X86_GREG_x10] = */ { CPUMCTX_OFF_AND_SIZE(r10), "r10", },
3615	/* [kIemNativeGstReg_GprFirst + X86_GREG_x11] = */ { CPUMCTX_OFF_AND_SIZE(r11), "r11", },
3616	/* [kIemNativeGstReg_GprFirst + X86_GREG_x12] = */ { CPUMCTX_OFF_AND_SIZE(r12), "r12", },
3617	/* [kIemNativeGstReg_GprFirst + X86_GREG_x13] = */ { CPUMCTX_OFF_AND_SIZE(r13), "r13", },
3618	/* [kIemNativeGstReg_GprFirst + X86_GREG_x14] = */ { CPUMCTX_OFF_AND_SIZE(r14), "r14", },
3619	/* [kIemNativeGstReg_GprFirst + X86_GREG_x15] = */ { CPUMCTX_OFF_AND_SIZE(r15), "r15", },
3620	/* [kIemNativeGstReg_Pc] = */ { CPUMCTX_OFF_AND_SIZE(rip), "rip", },
3621	/* [kIemNativeGstReg_Cr0] = */ { CPUMCTX_OFF_AND_SIZE(cr0), "cr0", },
3622	/* [kIemNativeGstReg_FpuFcw] = */ { CPUMCTX_OFF_AND_SIZE(XState.x87.FCW), "fcw", },
3623	/* [kIemNativeGstReg_FpuFsw] = */ { CPUMCTX_OFF_AND_SIZE(XState.x87.FSW), "fsw", },
3624	/* [kIemNativeGstReg_SegBaseFirst + 0] = */ { CPUMCTX_OFF_AND_SIZE(aSRegs[0].u64Base), "es_base", },
3625	/* [kIemNativeGstReg_SegBaseFirst + 1] = */ { CPUMCTX_OFF_AND_SIZE(aSRegs[1].u64Base), "cs_base", },
3626	/* [kIemNativeGstReg_SegBaseFirst + 2] = */ { CPUMCTX_OFF_AND_SIZE(aSRegs[2].u64Base), "ss_base", },
3627	/* [kIemNativeGstReg_SegBaseFirst + 3] = */ { CPUMCTX_OFF_AND_SIZE(aSRegs[3].u64Base), "ds_base", },
3628	/* [kIemNativeGstReg_SegBaseFirst + 4] = */ { CPUMCTX_OFF_AND_SIZE(aSRegs[4].u64Base), "fs_base", },
3629	/* [kIemNativeGstReg_SegBaseFirst + 5] = */ { CPUMCTX_OFF_AND_SIZE(aSRegs[5].u64Base), "gs_base", },
3630	/* [kIemNativeGstReg_SegAttribFirst + 0] = */ { CPUMCTX_OFF_AND_SIZE(aSRegs[0].Attr.u), "es_attrib", },
3631	/* [kIemNativeGstReg_SegAttribFirst + 1] = */ { CPUMCTX_OFF_AND_SIZE(aSRegs[1].Attr.u), "cs_attrib", },
3632	/* [kIemNativeGstReg_SegAttribFirst + 2] = */ { CPUMCTX_OFF_AND_SIZE(aSRegs[2].Attr.u), "ss_attrib", },
3633	/* [kIemNativeGstReg_SegAttribFirst + 3] = */ { CPUMCTX_OFF_AND_SIZE(aSRegs[3].Attr.u), "ds_attrib", },
3634	/* [kIemNativeGstReg_SegAttribFirst + 4] = */ { CPUMCTX_OFF_AND_SIZE(aSRegs[4].Attr.u), "fs_attrib", },
3635	/* [kIemNativeGstReg_SegAttribFirst + 5] = */ { CPUMCTX_OFF_AND_SIZE(aSRegs[5].Attr.u), "gs_attrib", },
3636	/* [kIemNativeGstReg_SegLimitFirst + 0] = */ { CPUMCTX_OFF_AND_SIZE(aSRegs[0].u32Limit), "es_limit", },
3637	/* [kIemNativeGstReg_SegLimitFirst + 1] = */ { CPUMCTX_OFF_AND_SIZE(aSRegs[1].u32Limit), "cs_limit", },
3638	/* [kIemNativeGstReg_SegLimitFirst + 2] = */ { CPUMCTX_OFF_AND_SIZE(aSRegs[2].u32Limit), "ss_limit", },
3639	/* [kIemNativeGstReg_SegLimitFirst + 3] = */ { CPUMCTX_OFF_AND_SIZE(aSRegs[3].u32Limit), "ds_limit", },
3640	/* [kIemNativeGstReg_SegLimitFirst + 4] = */ { CPUMCTX_OFF_AND_SIZE(aSRegs[4].u32Limit), "fs_limit", },
3641	/* [kIemNativeGstReg_SegLimitFirst + 5] = */ { CPUMCTX_OFF_AND_SIZE(aSRegs[5].u32Limit), "gs_limit", },
3642	/* [kIemNativeGstReg_SegSelFirst + 0] = */ { CPUMCTX_OFF_AND_SIZE(aSRegs[0].Sel), "es", },
3643	/* [kIemNativeGstReg_SegSelFirst + 1] = */ { CPUMCTX_OFF_AND_SIZE(aSRegs[1].Sel), "cs", },
3644	/* [kIemNativeGstReg_SegSelFirst + 2] = */ { CPUMCTX_OFF_AND_SIZE(aSRegs[2].Sel), "ss", },
3645	/* [kIemNativeGstReg_SegSelFirst + 3] = */ { CPUMCTX_OFF_AND_SIZE(aSRegs[3].Sel), "ds", },
3646	/* [kIemNativeGstReg_SegSelFirst + 4] = */ { CPUMCTX_OFF_AND_SIZE(aSRegs[4].Sel), "fs", },
3647	/* [kIemNativeGstReg_SegSelFirst + 5] = */ { CPUMCTX_OFF_AND_SIZE(aSRegs[5].Sel), "gs", },
3648	/* [kIemNativeGstReg_Cr4] = */ { CPUMCTX_OFF_AND_SIZE(cr4), "cr4", },
3649	/* [kIemNativeGstReg_Xcr0] = */ { CPUMCTX_OFF_AND_SIZE(aXcr[0]), "xcr0", },
3650	/* [kIemNativeGstReg_MxCsr] = */ { CPUMCTX_OFF_AND_SIZE(XState.x87.MXCSR), "mxcsr", },
3651	/* [kIemNativeGstReg_EFlags] = */ { CPUMCTX_OFF_AND_SIZE(eflags), "eflags", },
3652	#undef CPUMCTX_OFF_AND_SIZE
3653	};
3654	AssertCompile(RT_ELEMENTS(g_aGstShadowInfo) == kIemNativeGstReg_End);
3655
3656
3657	/** Host CPU general purpose register names. */
3658	DECL_HIDDEN_CONST(const char * const) g_apszIemNativeHstRegNames[] =
3659	{
3660	#ifdef RT_ARCH_AMD64
3661	"rax", "rcx", "rdx", "rbx", "rsp", "rbp", "rsi", "rdi", "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
3662	#elif RT_ARCH_ARM64
3663	"x0", "x1", "x2", "x3", "x4", "x5", "x6", "x7", "x8", "x9", "x10", "x11", "x12", "x13", "x14", "x15",
3664	"x16", "x17", "x18", "x19", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "bp", "lr", "sp/xzr",
3665	#else
3666	# error "port me"
3667	#endif
3668	};
3669
3670
3671	#if 0 /* unused */
3672	/**
3673	* Tries to locate a suitable register in the given register mask.
3674	*
3675	* This ASSUMES the caller has done the minimal/optimal allocation checks and
3676	* failed.
3677	*
3678	* @returns Host register number on success, returns UINT8_MAX on failure.
3679	*/
3680	static uint8_t iemNativeRegTryAllocFree(PIEMRECOMPILERSTATE pReNative, uint32_t fRegMask)
3681	{
3682	Assert(!(fRegMask & ~IEMNATIVE_HST_GREG_MASK));
3683	uint32_t fRegs = ~pReNative->Core.bmHstRegs & fRegMask;
3684	if (fRegs)
3685	{
3686	/** @todo pick better here: */
3687	unsigned const idxReg = ASMBitFirstSetU32(fRegs) - 1;
3688
3689	Assert(pReNative->Core.aHstRegs[idxReg].fGstRegShadows != 0);
3690	Assert( (pReNative->Core.aHstRegs[idxReg].fGstRegShadows & pReNative->Core.bmGstRegShadows)
3691	== pReNative->Core.aHstRegs[idxReg].fGstRegShadows);
3692	Assert(pReNative->Core.bmHstRegsWithGstShadow & RT_BIT_32(idxReg));
3693
3694	pReNative->Core.bmGstRegShadows &= ~pReNative->Core.aHstRegs[idxReg].fGstRegShadows;
3695	pReNative->Core.bmHstRegsWithGstShadow &= ~RT_BIT_32(idxReg);
3696	pReNative->Core.aHstRegs[idxReg].fGstRegShadows = 0;
3697	return idxReg;
3698	}
3699	return UINT8_MAX;
3700	}
3701	#endif /* unused */
3702
3703
3704	/**
3705	* Locate a register, possibly freeing one up.
3706	*
3707	* This ASSUMES the caller has done the minimal/optimal allocation checks and
3708	* failed.
3709	*
3710	* @returns Host register number on success. Returns UINT8_MAX if no registers
3711	* found, the caller is supposed to deal with this and raise a
3712	* allocation type specific status code (if desired).
3713	*
3714	* @throws VBox status code if we're run into trouble spilling a variable of
3715	* recording debug info. Does NOT throw anything if we're out of
3716	* registers, though.
3717	*/
3718	static uint8_t iemNativeRegAllocFindFree(PIEMRECOMPILERSTATE pReNative, uint32_t *poff, bool fPreferVolatile,
3719	uint32_t fRegMask = IEMNATIVE_HST_GREG_MASK & ~IEMNATIVE_REG_FIXED_MASK)
3720	{
3721	STAM_COUNTER_INC(&pReNative->pVCpu->iem.s.StatNativeRegFindFree);
3722	Assert(!(fRegMask & ~IEMNATIVE_HST_GREG_MASK));
3723	Assert(!(fRegMask & IEMNATIVE_REG_FIXED_MASK));
3724
3725	/*
3726	* Try a freed register that's shadowing a guest register.
3727	*/
3728	uint32_t fRegs = ~pReNative->Core.bmHstRegs & fRegMask;
3729	if (fRegs)
3730	{
3731	STAM_COUNTER_INC(&pReNative->pVCpu->iem.s.StatNativeRegFindFreeNoVar);
3732
3733	#ifdef IEMNATIVE_WITH_LIVENESS_ANALYSIS
3734	/*
3735	* When we have livness information, we use it to kick out all shadowed
3736	* guest register that will not be needed any more in this TB. If we're
3737	* lucky, this may prevent us from ending up here again.
3738	*
3739	* Note! We must consider the previous entry here so we don't free
3740	* anything that the current threaded function requires (current
3741	* entry is produced by the next threaded function).
3742	*/
3743	uint32_t const idxCurCall = pReNative->idxCurCall;
3744	if (idxCurCall > 0)
3745	{
3746	PCIEMLIVENESSENTRY const pLivenessEntry = &pReNative->paLivenessEntries[idxCurCall - 1];
3747
3748	# ifndef IEMLIVENESS_EXTENDED_LAYOUT
3749	/* Construct a mask of the guest registers in the UNUSED and XCPT_OR_CALL state. */
3750	AssertCompile(IEMLIVENESS_STATE_UNUSED == 1 && IEMLIVENESS_STATE_XCPT_OR_CALL == 2);
3751	uint64_t fToFreeMask = pLivenessEntry->Bit0.bm64 ^ pLivenessEntry->Bit1.bm64; /* mask of regs in either UNUSED */
3752	#else
3753	/* Construct a mask of the registers not in the read or write state.
3754	Note! We could skips writes, if they aren't from us, as this is just
3755	a hack to prevent trashing registers that have just been written
3756	or will be written when we retire the current instruction. */
3757	uint64_t fToFreeMask = ~pLivenessEntry->aBits[IEMLIVENESS_BIT_READ].bm64
3758	& ~pLivenessEntry->aBits[IEMLIVENESS_BIT_WRITE].bm64
3759	& IEMLIVENESSBIT_MASK;
3760	#endif
3761	/* Merge EFLAGS. */
3762	uint64_t fTmp = fToFreeMask & (fToFreeMask >> 3); /* AF2,PF2,CF2,Other2 = AF,PF,CF,Other & OF,SF,ZF,AF */
3763	fTmp &= fTmp >> 2; /* CF3,Other3 = AF2,PF2 & CF2,Other2 */
3764	fTmp &= fTmp >> 1; /* Other4 = CF3 & Other3 */
3765	fToFreeMask &= RT_BIT_64(kIemNativeGstReg_EFlags) - 1;
3766	fToFreeMask \|= fTmp & RT_BIT_64(kIemNativeGstReg_EFlags);
3767
3768	/* If it matches any shadowed registers. */
3769	if (pReNative->Core.bmGstRegShadows & fToFreeMask)
3770	{
3771	STAM_COUNTER_INC(&pReNative->pVCpu->iem.s.StatNativeRegFindFreeLivenessUnshadowed);
3772	iemNativeRegFlushGuestShadows(pReNative, fToFreeMask);
3773	Assert(fRegs == (~pReNative->Core.bmHstRegs & fRegMask)); /* this shall not change. */
3774
3775	/* See if we've got any unshadowed registers we can return now. */
3776	uint32_t const fUnshadowedRegs = fRegs & ~pReNative->Core.bmHstRegsWithGstShadow;
3777	if (fUnshadowedRegs)
3778	{
3779	STAM_COUNTER_INC(&pReNative->pVCpu->iem.s.StatNativeRegFindFreeLivenessHelped);
3780	return (fPreferVolatile
3781	? ASMBitFirstSetU32(fUnshadowedRegs)
3782	: ASMBitLastSetU32( fUnshadowedRegs & ~IEMNATIVE_CALL_VOLATILE_GREG_MASK
3783	? fUnshadowedRegs & ~IEMNATIVE_CALL_VOLATILE_GREG_MASK : fUnshadowedRegs))
3784	- 1;
3785	}
3786	}
3787	}
3788	#endif /* IEMNATIVE_WITH_LIVENESS_ANALYSIS */
3789
3790	unsigned const idxReg = (fPreferVolatile
3791	? ASMBitFirstSetU32(fRegs)
3792	: ASMBitLastSetU32( fRegs & ~IEMNATIVE_CALL_VOLATILE_GREG_MASK
3793	? fRegs & ~IEMNATIVE_CALL_VOLATILE_GREG_MASK : fRegs))
3794	- 1;
3795
3796	Assert(pReNative->Core.aHstRegs[idxReg].fGstRegShadows != 0);
3797	Assert( (pReNative->Core.aHstRegs[idxReg].fGstRegShadows & pReNative->Core.bmGstRegShadows)
3798	== pReNative->Core.aHstRegs[idxReg].fGstRegShadows);
3799	Assert(pReNative->Core.bmHstRegsWithGstShadow & RT_BIT_32(idxReg));
3800
3801	pReNative->Core.bmHstRegsWithGstShadow &= ~RT_BIT_32(idxReg);
3802	pReNative->Core.bmGstRegShadows &= ~pReNative->Core.aHstRegs[idxReg].fGstRegShadows;
3803	pReNative->Core.aHstRegs[idxReg].fGstRegShadows = 0;
3804	return idxReg;
3805	}
3806
3807	/*
3808	* Try free up a variable that's in a register.
3809	*
3810	* We do two rounds here, first evacuating variables we don't need to be
3811	* saved on the stack, then in the second round move things to the stack.
3812	*/
3813	STAM_REL_COUNTER_INC(&pReNative->pVCpu->iem.s.StatNativeRegFindFreeVar);
3814	for (uint32_t iLoop = 0; iLoop < 2; iLoop++)
3815	{
3816	uint32_t fVars = pReNative->Core.bmVars;
3817	while (fVars)
3818	{
3819	uint32_t const idxVar = ASMBitFirstSetU32(fVars) - 1;
3820	uint8_t const idxReg = pReNative->Core.aVars[idxVar].idxReg;
3821	if ( idxReg < RT_ELEMENTS(pReNative->Core.aHstRegs)
3822	&& (RT_BIT_32(idxReg) & fRegMask)
3823	&& ( iLoop == 0
3824	? pReNative->Core.aVars[idxVar].enmKind != kIemNativeVarKind_Stack
3825	: pReNative->Core.aVars[idxVar].enmKind == kIemNativeVarKind_Stack)
3826	&& !pReNative->Core.aVars[idxVar].fRegAcquired)
3827	{
3828	Assert(pReNative->Core.bmHstRegs & RT_BIT_32(idxReg));
3829	Assert( (pReNative->Core.bmGstRegShadows & pReNative->Core.aHstRegs[idxReg].fGstRegShadows)
3830	== pReNative->Core.aHstRegs[idxReg].fGstRegShadows);
3831	Assert(pReNative->Core.bmGstRegShadows < RT_BIT_64(kIemNativeGstReg_End));
3832	Assert( RT_BOOL(pReNative->Core.bmHstRegsWithGstShadow & RT_BIT_32(idxReg))
3833	== RT_BOOL(pReNative->Core.aHstRegs[idxReg].fGstRegShadows));
3834
3835	if (pReNative->Core.aVars[idxVar].enmKind == kIemNativeVarKind_Stack)
3836	{
3837	uint8_t const idxStackSlot = iemNativeVarGetStackSlot(pReNative, IEMNATIVE_VAR_IDX_PACK(idxVar));
3838	poff = iemNativeEmitStoreGprByBp(pReNative, poff, iemNativeStackCalcBpDisp(idxStackSlot), idxReg);
3839	}
3840
3841	pReNative->Core.aVars[idxVar].idxReg = UINT8_MAX;
3842	pReNative->Core.bmHstRegs &= ~RT_BIT_32(idxReg);
3843
3844	pReNative->Core.bmHstRegsWithGstShadow &= ~RT_BIT_32(idxReg);
3845	pReNative->Core.bmGstRegShadows &= ~pReNative->Core.aHstRegs[idxReg].fGstRegShadows;
3846	pReNative->Core.aHstRegs[idxReg].fGstRegShadows = 0;
3847	return idxReg;
3848	}
3849	fVars &= ~RT_BIT_32(idxVar);
3850	}
3851	}
3852
3853	return UINT8_MAX;
3854	}
3855
3856
3857	/**
3858	* Reassigns a variable to a different register specified by the caller.
3859	*
3860	* @returns The new code buffer position.
3861	* @param pReNative The native recompile state.
3862	* @param off The current code buffer position.
3863	* @param idxVar The variable index.
3864	* @param idxRegOld The old host register number.
3865	* @param idxRegNew The new host register number.
3866	* @param pszCaller The caller for logging.
3867	*/
3868	static uint32_t iemNativeRegMoveVar(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint8_t idxVar,
3869	uint8_t idxRegOld, uint8_t idxRegNew, const char *pszCaller)
3870	{
3871	IEMNATIVE_ASSERT_VAR_IDX(pReNative, idxVar);
3872	Assert(pReNative->Core.aVars[IEMNATIVE_VAR_IDX_UNPACK(idxVar)].idxReg == idxRegOld);
3873	RT_NOREF(pszCaller);
3874
3875	iemNativeRegClearGstRegShadowing(pReNative, idxRegNew, off);
3876
3877	uint64_t fGstRegShadows = pReNative->Core.aHstRegs[idxRegOld].fGstRegShadows;
3878	Log12(("%s: moving idxVar=%#x from %s to %s (fGstRegShadows=%RX64)\n",
3879	pszCaller, idxVar, g_apszIemNativeHstRegNames[idxRegOld], g_apszIemNativeHstRegNames[idxRegNew], fGstRegShadows));
3880	off = iemNativeEmitLoadGprFromGpr(pReNative, off, idxRegNew, idxRegOld);
3881
3882	pReNative->Core.aHstRegs[idxRegNew].fGstRegShadows = fGstRegShadows;
3883	pReNative->Core.aHstRegs[idxRegNew].enmWhat = kIemNativeWhat_Var;
3884	pReNative->Core.aHstRegs[idxRegNew].idxVar = idxVar;
3885	if (fGstRegShadows)
3886	{
3887	pReNative->Core.bmHstRegsWithGstShadow = (pReNative->Core.bmHstRegsWithGstShadow & ~RT_BIT_32(idxRegOld))
3888	\| RT_BIT_32(idxRegNew);
3889	while (fGstRegShadows)
3890	{
3891	unsigned const idxGstReg = ASMBitFirstSetU64(fGstRegShadows) - 1;
3892	fGstRegShadows &= ~RT_BIT_64(idxGstReg);
3893
3894	Assert(pReNative->Core.aidxGstRegShadows[idxGstReg] == idxRegOld);
3895	pReNative->Core.aidxGstRegShadows[idxGstReg] = idxRegNew;
3896	}
3897	}
3898
3899	pReNative->Core.aVars[IEMNATIVE_VAR_IDX_UNPACK(idxVar)].idxReg = (uint8_t)idxRegNew;
3900	pReNative->Core.aHstRegs[idxRegOld].fGstRegShadows = 0;
3901	pReNative->Core.bmHstRegs = RT_BIT_32(idxRegNew) \| (pReNative->Core.bmHstRegs & ~RT_BIT_32(idxRegOld));
3902	return off;
3903	}
3904
3905
3906	/**
3907	* Moves a variable to a different register or spills it onto the stack.
3908	*
3909	* This must be a stack variable (kIemNativeVarKind_Stack) because the other
3910	* kinds can easily be recreated if needed later.
3911	*
3912	* @returns The new code buffer position.
3913	* @param pReNative The native recompile state.
3914	* @param off The current code buffer position.
3915	* @param idxVar The variable index.
3916	* @param fForbiddenRegs Mask of the forbidden registers. Defaults to
3917	* call-volatile registers.
3918	*/
3919	DECL_HIDDEN_THROW(uint32_t) iemNativeRegMoveOrSpillStackVar(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint8_t idxVar,
3920	uint32_t fForbiddenRegs /= IEMNATIVE_CALL_VOLATILE_GREG_MASK/)
3921	{
3922	IEMNATIVE_ASSERT_VAR_IDX(pReNative, idxVar);
3923	PIEMNATIVEVAR const pVar = &pReNative->Core.aVars[IEMNATIVE_VAR_IDX_UNPACK(idxVar)];
3924	Assert(pVar->enmKind == kIemNativeVarKind_Stack);
3925	Assert(!pVar->fRegAcquired);
3926
3927	uint8_t const idxRegOld = pVar->idxReg;
3928	Assert(idxRegOld < RT_ELEMENTS(pReNative->Core.aHstRegs));
3929	Assert(pReNative->Core.bmHstRegs & RT_BIT_32(idxRegOld));
3930	Assert(pReNative->Core.aHstRegs[idxRegOld].enmWhat == kIemNativeWhat_Var);
3931	Assert( (pReNative->Core.bmGstRegShadows & pReNative->Core.aHstRegs[idxRegOld].fGstRegShadows)
3932	== pReNative->Core.aHstRegs[idxRegOld].fGstRegShadows);
3933	Assert(pReNative->Core.bmGstRegShadows < RT_BIT_64(kIemNativeGstReg_End));
3934	Assert( RT_BOOL(pReNative->Core.bmHstRegsWithGstShadow & RT_BIT_32(idxRegOld))
3935	== RT_BOOL(pReNative->Core.aHstRegs[idxRegOld].fGstRegShadows));
3936
3937
3938	/** @todo Add statistics on this.*/
3939	/** @todo Implement basic variable liveness analysis (python) so variables
3940	* can be freed immediately once no longer used. This has the potential to
3941	* be trashing registers and stack for dead variables.
3942	* Update: This is mostly done. (Not IEMNATIVE_WITH_LIVENESS_ANALYSIS.) */
3943
3944	/*
3945	* First try move it to a different register, as that's cheaper.
3946	*/
3947	fForbiddenRegs \|= RT_BIT_32(idxRegOld);
3948	fForbiddenRegs \|= IEMNATIVE_REG_FIXED_MASK;
3949	uint32_t fRegs = ~pReNative->Core.bmHstRegs & ~fForbiddenRegs;
3950	if (fRegs)
3951	{
3952	/* Avoid using shadow registers, if possible. */
3953	if (fRegs & ~pReNative->Core.bmHstRegsWithGstShadow)
3954	fRegs &= ~pReNative->Core.bmHstRegsWithGstShadow;
3955	unsigned const idxRegNew = ASMBitFirstSetU32(fRegs) - 1;
3956	return iemNativeRegMoveVar(pReNative, off, idxVar, idxRegOld, idxRegNew, "iemNativeRegMoveOrSpillStackVar");
3957	}
3958
3959	/*
3960	* Otherwise we must spill the register onto the stack.
3961	*/
3962	uint8_t const idxStackSlot = iemNativeVarGetStackSlot(pReNative, idxVar);
3963	Log12(("iemNativeRegMoveOrSpillStackVar: spilling idxVar=%#x/idxReg=%d onto the stack (slot %#x bp+%d, off=%#x)\n",
3964	idxVar, idxRegOld, idxStackSlot, iemNativeStackCalcBpDisp(idxStackSlot), off));
3965	off = iemNativeEmitStoreGprByBp(pReNative, off, iemNativeStackCalcBpDisp(idxStackSlot), idxRegOld);
3966
3967	pVar->idxReg = UINT8_MAX;
3968	pReNative->Core.bmHstRegsWithGstShadow &= ~RT_BIT_32(idxRegOld);
3969	pReNative->Core.bmHstRegs &= ~RT_BIT_32(idxRegOld);
3970	pReNative->Core.bmGstRegShadows &= ~pReNative->Core.aHstRegs[idxRegOld].fGstRegShadows;
3971	pReNative->Core.aHstRegs[idxRegOld].fGstRegShadows = 0;
3972	return off;
3973	}
3974
3975
3976	/**
3977	* Allocates a temporary host general purpose register.
3978	*
3979	* This may emit code to save register content onto the stack in order to free
3980	* up a register.
3981	*
3982	* @returns The host register number; throws VBox status code on failure,
3983	* so no need to check the return value.
3984	* @param pReNative The native recompile state.
3985	* @param poff Pointer to the variable with the code buffer position.
3986	* This will be update if we need to move a variable from
3987	* register to stack in order to satisfy the request.
3988	* @param fPreferVolatile Whether to prefer volatile over non-volatile
3989	* registers (@c true, default) or the other way around
3990	* (@c false, for iemNativeRegAllocTmpForGuestReg()).
3991	*/
3992	DECL_HIDDEN_THROW(uint8_t) iemNativeRegAllocTmp(PIEMRECOMPILERSTATE pReNative, uint32_t poff, bool fPreferVolatile /= true*/)
3993	{
3994	/*
3995	* Try find a completely unused register, preferably a call-volatile one.
3996	*/
3997	uint8_t idxReg;
3998	uint32_t fRegs = ~pReNative->Core.bmHstRegs
3999	& ~pReNative->Core.bmHstRegsWithGstShadow
4000	& (~IEMNATIVE_REG_FIXED_MASK & IEMNATIVE_HST_GREG_MASK);
4001	if (fRegs)
4002	{
4003	if (fPreferVolatile)
4004	idxReg = (uint8_t)ASMBitFirstSetU32( fRegs & IEMNATIVE_CALL_VOLATILE_GREG_MASK
4005	? fRegs & IEMNATIVE_CALL_VOLATILE_GREG_MASK : fRegs) - 1;
4006	else
4007	idxReg = (uint8_t)ASMBitFirstSetU32( fRegs & ~IEMNATIVE_CALL_VOLATILE_GREG_MASK
4008	? fRegs & ~IEMNATIVE_CALL_VOLATILE_GREG_MASK : fRegs) - 1;
4009	Assert(pReNative->Core.aHstRegs[idxReg].fGstRegShadows == 0);
4010	Assert(!(pReNative->Core.bmHstRegsWithGstShadow & RT_BIT_32(idxReg)));
4011	Log12(("iemNativeRegAllocTmp: %s\n", g_apszIemNativeHstRegNames[idxReg]));
4012	}
4013	else
4014	{
4015	idxReg = iemNativeRegAllocFindFree(pReNative, poff, fPreferVolatile);
4016	AssertStmt(idxReg != UINT8_MAX, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_REG_ALLOCATOR_NO_FREE_TMP));
4017	Log12(("iemNativeRegAllocTmp: %s (slow)\n", g_apszIemNativeHstRegNames[idxReg]));
4018	}
4019	return iemNativeRegMarkAllocated(pReNative, idxReg, kIemNativeWhat_Tmp);
4020	}
4021
4022
4023	/**
4024	* Alternative version of iemNativeRegAllocTmp that takes mask with acceptable
4025	* registers.
4026	*
4027	* @returns The host register number; throws VBox status code on failure,
4028	* so no need to check the return value.
4029	* @param pReNative The native recompile state.
4030	* @param poff Pointer to the variable with the code buffer position.
4031	* This will be update if we need to move a variable from
4032	* register to stack in order to satisfy the request.
4033	* @param fRegMask Mask of acceptable registers.
4034	* @param fPreferVolatile Whether to prefer volatile over non-volatile
4035	* registers (@c true, default) or the other way around
4036	* (@c false, for iemNativeRegAllocTmpForGuestReg()).
4037	*/
4038	DECL_HIDDEN_THROW(uint8_t) iemNativeRegAllocTmpEx(PIEMRECOMPILERSTATE pReNative, uint32_t *poff, uint32_t fRegMask,
4039	bool fPreferVolatile /= true/)
4040	{
4041	Assert(!(fRegMask & ~IEMNATIVE_HST_GREG_MASK));
4042	Assert(!(fRegMask & IEMNATIVE_REG_FIXED_MASK));
4043
4044	/*
4045	* Try find a completely unused register, preferably a call-volatile one.
4046	*/
4047	uint8_t idxReg;
4048	uint32_t fRegs = ~pReNative->Core.bmHstRegs
4049	& ~pReNative->Core.bmHstRegsWithGstShadow
4050	& (~IEMNATIVE_REG_FIXED_MASK & IEMNATIVE_HST_GREG_MASK)
4051	& fRegMask;
4052	if (fRegs)
4053	{
4054	if (fPreferVolatile)
4055	idxReg = (uint8_t)ASMBitFirstSetU32( fRegs & IEMNATIVE_CALL_VOLATILE_GREG_MASK
4056	? fRegs & IEMNATIVE_CALL_VOLATILE_GREG_MASK : fRegs) - 1;
4057	else
4058	idxReg = (uint8_t)ASMBitFirstSetU32( fRegs & ~IEMNATIVE_CALL_VOLATILE_GREG_MASK
4059	? fRegs & ~IEMNATIVE_CALL_VOLATILE_GREG_MASK : fRegs) - 1;
4060	Assert(pReNative->Core.aHstRegs[idxReg].fGstRegShadows == 0);
4061	Assert(!(pReNative->Core.bmHstRegsWithGstShadow & RT_BIT_32(idxReg)));
4062	Log12(("iemNativeRegAllocTmpEx: %s\n", g_apszIemNativeHstRegNames[idxReg]));
4063	}
4064	else
4065	{
4066	idxReg = iemNativeRegAllocFindFree(pReNative, poff, fPreferVolatile, fRegMask);
4067	AssertStmt(idxReg != UINT8_MAX, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_REG_ALLOCATOR_NO_FREE_TMP));
4068	Log12(("iemNativeRegAllocTmpEx: %s (slow)\n", g_apszIemNativeHstRegNames[idxReg]));
4069	}
4070	return iemNativeRegMarkAllocated(pReNative, idxReg, kIemNativeWhat_Tmp);
4071	}
4072
4073
4074	/**
4075	* Allocates a temporary register for loading an immediate value into.
4076	*
4077	* This will emit code to load the immediate, unless there happens to be an
4078	* unused register with the value already loaded.
4079	*
4080	* The caller will not modify the returned register, it must be considered
4081	* read-only. Free using iemNativeRegFreeTmpImm.
4082	*
4083	* @returns The host register number; throws VBox status code on failure, so no
4084	* need to check the return value.
4085	* @param pReNative The native recompile state.
4086	* @param poff Pointer to the variable with the code buffer position.
4087	* @param uImm The immediate value that the register must hold upon
4088	* return.
4089	* @param fPreferVolatile Whether to prefer volatile over non-volatile
4090	* registers (@c true, default) or the other way around
4091	* (@c false).
4092	*
4093	* @note Reusing immediate values has not been implemented yet.
4094	*/
4095	DECL_HIDDEN_THROW(uint8_t)
4096	iemNativeRegAllocTmpImm(PIEMRECOMPILERSTATE pReNative, uint32_t poff, uint64_t uImm, bool fPreferVolatile /= true*/)
4097	{
4098	uint8_t const idxReg = iemNativeRegAllocTmp(pReNative, poff, fPreferVolatile);
4099	poff = iemNativeEmitLoadGprImm64(pReNative, poff, idxReg, uImm);
4100	return idxReg;
4101	}
4102
4103
4104	/**
4105	* Allocates a temporary host general purpose register for keeping a guest
4106	* register value.
4107	*
4108	* Since we may already have a register holding the guest register value,
4109	* code will be emitted to do the loading if that's not the case. Code may also
4110	* be emitted if we have to free up a register to satify the request.
4111	*
4112	* @returns The host register number; throws VBox status code on failure, so no
4113	* need to check the return value.
4114	* @param pReNative The native recompile state.
4115	* @param poff Pointer to the variable with the code buffer
4116	* position. This will be update if we need to move a
4117	* variable from register to stack in order to satisfy
4118	* the request.
4119	* @param enmGstReg The guest register that will is to be updated.
4120	* @param enmIntendedUse How the caller will be using the host register.
4121	* @param fNoVolatileRegs Set if no volatile register allowed, clear if any
4122	* register is okay (default). The ASSUMPTION here is
4123	* that the caller has already flushed all volatile
4124	* registers, so this is only applied if we allocate a
4125	* new register.
4126	* @param fSkipLivenessAssert Hack for liveness input validation of EFLAGS.
4127	* @sa iemNativeRegAllocTmpForGuestRegIfAlreadyPresent
4128	*/
4129	DECL_HIDDEN_THROW(uint8_t)
4130	iemNativeRegAllocTmpForGuestReg(PIEMRECOMPILERSTATE pReNative, uint32_t *poff, IEMNATIVEGSTREG enmGstReg,
4131	IEMNATIVEGSTREGUSE enmIntendedUse /= kIemNativeGstRegUse_ReadOnly/,
4132	bool fNoVolatileRegs /= false/, bool fSkipLivenessAssert /= false/)
4133	{
4134	Assert(enmGstReg < kIemNativeGstReg_End && g_aGstShadowInfo[enmGstReg].cb != 0);
4135	#ifdef IEMNATIVE_WITH_LIVENESS_ANALYSIS
4136	AssertMsg( fSkipLivenessAssert
4137	\|\| pReNative->idxCurCall == 0
4138	\|\| enmGstReg == kIemNativeGstReg_Pc
4139	\|\| (enmIntendedUse == kIemNativeGstRegUse_ForFullWrite
4140	? IEMLIVENESS_STATE_IS_CLOBBER_EXPECTED(iemNativeLivenessGetPrevStateByGstReg(pReNative, enmGstReg))
4141	: enmIntendedUse == kIemNativeGstRegUse_ForUpdate
4142	? IEMLIVENESS_STATE_IS_MODIFY_EXPECTED( iemNativeLivenessGetPrevStateByGstReg(pReNative, enmGstReg))
4143	: IEMLIVENESS_STATE_IS_INPUT_EXPECTED( iemNativeLivenessGetPrevStateByGstReg(pReNative, enmGstReg)) ),
4144	("%s - %u\n", g_aGstShadowInfo[enmGstReg].pszName, iemNativeLivenessGetPrevStateByGstReg(pReNative, enmGstReg)));
4145	#endif
4146	RT_NOREF(fSkipLivenessAssert);
4147	#if defined(LOG_ENABLED) \|\| defined(VBOX_STRICT)
4148	static const char * const s_pszIntendedUse[] = { "fetch", "update", "full write", "destructive calc" };
4149	#endif
4150	uint32_t const fRegMask = !fNoVolatileRegs
4151	? IEMNATIVE_HST_GREG_MASK & ~IEMNATIVE_REG_FIXED_MASK
4152	: IEMNATIVE_HST_GREG_MASK & ~IEMNATIVE_REG_FIXED_MASK & ~IEMNATIVE_CALL_VOLATILE_GREG_MASK;
4153
4154	/*
4155	* First check if the guest register value is already in a host register.
4156	*/
4157	if (pReNative->Core.bmGstRegShadows & RT_BIT_64(enmGstReg))
4158	{
4159	uint8_t idxReg = pReNative->Core.aidxGstRegShadows[enmGstReg];
4160	Assert(idxReg < RT_ELEMENTS(pReNative->Core.aHstRegs));
4161	Assert(pReNative->Core.aHstRegs[idxReg].fGstRegShadows & RT_BIT_64(enmGstReg));
4162	Assert(pReNative->Core.bmHstRegsWithGstShadow & RT_BIT_32(idxReg));
4163
4164	/* It's not supposed to be allocated... */
4165	if (!(pReNative->Core.bmHstRegs & RT_BIT_32(idxReg)))
4166	{
4167	/*
4168	* If the register will trash the guest shadow copy, try find a
4169	* completely unused register we can use instead. If that fails,
4170	* we need to disassociate the host reg from the guest reg.
4171	*/
4172	/** @todo would be nice to know if preserving the register is in any way helpful. */
4173	/* If the purpose is calculations, try duplicate the register value as
4174	we'll be clobbering the shadow. */
4175	if ( enmIntendedUse == kIemNativeGstRegUse_Calculation
4176	&& ( ~pReNative->Core.bmHstRegs
4177	& ~pReNative->Core.bmHstRegsWithGstShadow
4178	& (~IEMNATIVE_REG_FIXED_MASK & IEMNATIVE_HST_GREG_MASK)))
4179	{
4180	uint8_t const idxRegNew = iemNativeRegAllocTmpEx(pReNative, poff, fRegMask);
4181
4182	poff = iemNativeEmitLoadGprFromGpr(pReNative, poff, idxRegNew, idxReg);
4183
4184	Log12(("iemNativeRegAllocTmpForGuestReg: Duplicated %s for guest %s into %s for destructive calc\n",
4185	g_apszIemNativeHstRegNames[idxReg], g_aGstShadowInfo[enmGstReg].pszName,
4186	g_apszIemNativeHstRegNames[idxRegNew]));
4187	idxReg = idxRegNew;
4188	}
4189	/* If the current register matches the restrictions, go ahead and allocate
4190	it for the caller. */
4191	else if (fRegMask & RT_BIT_32(idxReg))
4192	{
4193	pReNative->Core.bmHstRegs \|= RT_BIT_32(idxReg);
4194	pReNative->Core.aHstRegs[idxReg].enmWhat = kIemNativeWhat_Tmp;
4195	pReNative->Core.aHstRegs[idxReg].idxVar = UINT8_MAX;
4196	if (enmIntendedUse != kIemNativeGstRegUse_Calculation)
4197	Log12(("iemNativeRegAllocTmpForGuestReg: Reusing %s for guest %s %s\n",
4198	g_apszIemNativeHstRegNames[idxReg], g_aGstShadowInfo[enmGstReg].pszName, s_pszIntendedUse[enmIntendedUse]));
4199	else
4200	{
4201	iemNativeRegClearGstRegShadowing(pReNative, idxReg, *poff);
4202	Log12(("iemNativeRegAllocTmpForGuestReg: Grabbing %s for guest %s - destructive calc\n",
4203	g_apszIemNativeHstRegNames[idxReg], g_aGstShadowInfo[enmGstReg].pszName));
4204	}
4205	}
4206	/* Otherwise, allocate a register that satisfies the caller and transfer
4207	the shadowing if compatible with the intended use. (This basically
4208	means the call wants a non-volatile register (RSP push/pop scenario).) */
4209	else
4210	{
4211	Assert(fNoVolatileRegs);
4212	uint8_t const idxRegNew = iemNativeRegAllocTmpEx(pReNative, poff, fRegMask & ~RT_BIT_32(idxReg),
4213	!fNoVolatileRegs
4214	&& enmIntendedUse == kIemNativeGstRegUse_Calculation);
4215	poff = iemNativeEmitLoadGprFromGpr(pReNative, poff, idxRegNew, idxReg);
4216	if (enmIntendedUse != kIemNativeGstRegUse_Calculation)
4217	{
4218	iemNativeRegTransferGstRegShadowing(pReNative, idxReg, idxRegNew, enmGstReg, *poff);
4219	Log12(("iemNativeRegAllocTmpForGuestReg: Transfering %s to %s for guest %s %s\n",
4220	g_apszIemNativeHstRegNames[idxReg], g_apszIemNativeHstRegNames[idxRegNew],
4221	g_aGstShadowInfo[enmGstReg].pszName, s_pszIntendedUse[enmIntendedUse]));
4222	}
4223	else
4224	Log12(("iemNativeRegAllocTmpForGuestReg: Duplicated %s for guest %s into %s for destructive calc\n",
4225	g_apszIemNativeHstRegNames[idxReg], g_aGstShadowInfo[enmGstReg].pszName,
4226	g_apszIemNativeHstRegNames[idxRegNew]));
4227	idxReg = idxRegNew;
4228	}
4229	}
4230	else
4231	{
4232	/*
4233	* Oops. Shadowed guest register already allocated!
4234	*
4235	* Allocate a new register, copy the value and, if updating, the
4236	* guest shadow copy assignment to the new register.
4237	*/
4238	AssertMsg( enmIntendedUse != kIemNativeGstRegUse_ForUpdate
4239	&& enmIntendedUse != kIemNativeGstRegUse_ForFullWrite,
4240	("This shouldn't happen: idxReg=%d enmGstReg=%d enmIntendedUse=%s\n",
4241	idxReg, enmGstReg, s_pszIntendedUse[enmIntendedUse]));
4242
4243	/** @todo share register for readonly access. */
4244	uint8_t const idxRegNew = iemNativeRegAllocTmpEx(pReNative, poff, fRegMask,
4245	enmIntendedUse == kIemNativeGstRegUse_Calculation);
4246
4247	if (enmIntendedUse != kIemNativeGstRegUse_ForFullWrite)
4248	poff = iemNativeEmitLoadGprFromGpr(pReNative, poff, idxRegNew, idxReg);
4249
4250	if ( enmIntendedUse != kIemNativeGstRegUse_ForUpdate
4251	&& enmIntendedUse != kIemNativeGstRegUse_ForFullWrite)
4252	Log12(("iemNativeRegAllocTmpForGuestReg: Duplicated %s for guest %s into %s for %s\n",
4253	g_apszIemNativeHstRegNames[idxReg], g_aGstShadowInfo[enmGstReg].pszName,
4254	g_apszIemNativeHstRegNames[idxRegNew], s_pszIntendedUse[enmIntendedUse]));
4255	else
4256	{
4257	iemNativeRegTransferGstRegShadowing(pReNative, idxReg, idxRegNew, enmGstReg, *poff);
4258	Log12(("iemNativeRegAllocTmpForGuestReg: Moved %s for guest %s into %s for %s\n",
4259	g_apszIemNativeHstRegNames[idxReg], g_aGstShadowInfo[enmGstReg].pszName,
4260	g_apszIemNativeHstRegNames[idxRegNew], s_pszIntendedUse[enmIntendedUse]));
4261	}
4262	idxReg = idxRegNew;
4263	}
4264	Assert(RT_BIT_32(idxReg) & fRegMask); /* See assumption in fNoVolatileRegs docs. */
4265
4266	#ifdef VBOX_STRICT
4267	/* Strict builds: Check that the value is correct. */
4268	poff = iemNativeEmitGuestRegValueCheck(pReNative, poff, idxReg, enmGstReg);
4269	#endif
4270
4271	return idxReg;
4272	}
4273
4274	/*
4275	* Allocate a new register, load it with the guest value and designate it as a copy of the
4276	*/
4277	uint8_t const idxRegNew = iemNativeRegAllocTmpEx(pReNative, poff, fRegMask, enmIntendedUse == kIemNativeGstRegUse_Calculation);
4278
4279	if (enmIntendedUse != kIemNativeGstRegUse_ForFullWrite)
4280	poff = iemNativeEmitLoadGprWithGstShadowReg(pReNative, poff, idxRegNew, enmGstReg);
4281
4282	if (enmIntendedUse != kIemNativeGstRegUse_Calculation)
4283	iemNativeRegMarkAsGstRegShadow(pReNative, idxRegNew, enmGstReg, *poff);
4284	Log12(("iemNativeRegAllocTmpForGuestReg: Allocated %s for guest %s %s\n",
4285	g_apszIemNativeHstRegNames[idxRegNew], g_aGstShadowInfo[enmGstReg].pszName, s_pszIntendedUse[enmIntendedUse]));
4286
4287	return idxRegNew;
4288	}
4289
4290
4291	/**
4292	* Allocates a temporary host general purpose register that already holds the
4293	* given guest register value.
4294	*
4295	* The use case for this function is places where the shadowing state cannot be
4296	* modified due to branching and such. This will fail if the we don't have a
4297	* current shadow copy handy or if it's incompatible. The only code that will
4298	* be emitted here is value checking code in strict builds.
4299	*
4300	* The intended use can only be readonly!
4301	*
4302	* @returns The host register number, UINT8_MAX if not present.
4303	* @param pReNative The native recompile state.
4304	* @param poff Pointer to the instruction buffer offset.
4305	* Will be updated in strict builds if a register is
4306	* found.
4307	* @param enmGstReg The guest register that will is to be updated.
4308	* @note In strict builds, this may throw instruction buffer growth failures.
4309	* Non-strict builds will not throw anything.
4310	* @sa iemNativeRegAllocTmpForGuestReg
4311	*/
4312	DECL_HIDDEN_THROW(uint8_t)
4313	iemNativeRegAllocTmpForGuestRegIfAlreadyPresent(PIEMRECOMPILERSTATE pReNative, uint32_t *poff, IEMNATIVEGSTREG enmGstReg)
4314	{
4315	Assert(enmGstReg < kIemNativeGstReg_End && g_aGstShadowInfo[enmGstReg].cb != 0);
4316	#ifdef IEMNATIVE_WITH_LIVENESS_ANALYSIS
4317	AssertMsg( pReNative->idxCurCall == 0
4318	\|\| IEMLIVENESS_STATE_IS_INPUT_EXPECTED(iemNativeLivenessGetPrevStateByGstReg(pReNative, enmGstReg))
4319	\|\| enmGstReg == kIemNativeGstReg_Pc,
4320	("%s - %u\n", g_aGstShadowInfo[enmGstReg].pszName, iemNativeLivenessGetPrevStateByGstReg(pReNative, enmGstReg)));
4321	#endif
4322
4323	/*
4324	* First check if the guest register value is already in a host register.
4325	*/
4326	if (pReNative->Core.bmGstRegShadows & RT_BIT_64(enmGstReg))
4327	{
4328	uint8_t idxReg = pReNative->Core.aidxGstRegShadows[enmGstReg];
4329	Assert(idxReg < RT_ELEMENTS(pReNative->Core.aHstRegs));
4330	Assert(pReNative->Core.aHstRegs[idxReg].fGstRegShadows & RT_BIT_64(enmGstReg));
4331	Assert(pReNative->Core.bmHstRegsWithGstShadow & RT_BIT_32(idxReg));
4332
4333	if (!(pReNative->Core.bmHstRegs & RT_BIT_32(idxReg)))
4334	{
4335	/*
4336	* We only do readonly use here, so easy compared to the other
4337	* variant of this code.
4338	*/
4339	pReNative->Core.bmHstRegs \|= RT_BIT_32(idxReg);
4340	pReNative->Core.aHstRegs[idxReg].enmWhat = kIemNativeWhat_Tmp;
4341	pReNative->Core.aHstRegs[idxReg].idxVar = UINT8_MAX;
4342	Log12(("iemNativeRegAllocTmpForGuestRegIfAlreadyPresent: Reusing %s for guest %s readonly\n",
4343	g_apszIemNativeHstRegNames[idxReg], g_aGstShadowInfo[enmGstReg].pszName));
4344
4345	#ifdef VBOX_STRICT
4346	/* Strict builds: Check that the value is correct. */
4347	poff = iemNativeEmitGuestRegValueCheck(pReNative, poff, idxReg, enmGstReg);
4348	#else
4349	RT_NOREF(poff);
4350	#endif
4351	return idxReg;
4352	}
4353	}
4354
4355	return UINT8_MAX;
4356	}
4357
4358
4359	/**
4360	* Allocates argument registers for a function call.
4361	*
4362	* @returns New code buffer offset on success; throws VBox status code on failure, so no
4363	* need to check the return value.
4364	* @param pReNative The native recompile state.
4365	* @param off The current code buffer offset.
4366	* @param cArgs The number of arguments the function call takes.
4367	*/
4368	DECL_HIDDEN_THROW(uint32_t) iemNativeRegAllocArgs(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint8_t cArgs)
4369	{
4370	AssertStmt(cArgs <= IEMNATIVE_CALL_ARG_GREG_COUNT + IEMNATIVE_FRAME_STACK_ARG_COUNT,
4371	IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_REG_IPE_4));
4372	Assert(RT_ELEMENTS(g_aidxIemNativeCallRegs) == IEMNATIVE_CALL_ARG_GREG_COUNT);
4373	Assert(RT_ELEMENTS(g_afIemNativeCallRegs) == IEMNATIVE_CALL_ARG_GREG_COUNT);
4374
4375	if (cArgs > RT_ELEMENTS(g_aidxIemNativeCallRegs))
4376	cArgs = RT_ELEMENTS(g_aidxIemNativeCallRegs);
4377	else if (cArgs == 0)
4378	return true;
4379
4380	/*
4381	* Do we get luck and all register are free and not shadowing anything?
4382	*/
4383	if (((pReNative->Core.bmHstRegs \| pReNative->Core.bmHstRegsWithGstShadow) & g_afIemNativeCallRegs[cArgs]) == 0)
4384	for (uint32_t i = 0; i < cArgs; i++)
4385	{
4386	uint8_t const idxReg = g_aidxIemNativeCallRegs[i];
4387	pReNative->Core.aHstRegs[idxReg].enmWhat = kIemNativeWhat_Arg;
4388	pReNative->Core.aHstRegs[idxReg].idxVar = UINT8_MAX;
4389	Assert(pReNative->Core.aHstRegs[idxReg].fGstRegShadows == 0);
4390	}
4391	/*
4392	* Okay, not lucky so we have to free up the registers.
4393	*/
4394	else
4395	for (uint32_t i = 0; i < cArgs; i++)
4396	{
4397	uint8_t const idxReg = g_aidxIemNativeCallRegs[i];
4398	if (pReNative->Core.bmHstRegs & RT_BIT_32(idxReg))
4399	{
4400	switch (pReNative->Core.aHstRegs[idxReg].enmWhat)
4401	{
4402	case kIemNativeWhat_Var:
4403	{
4404	uint8_t const idxVar = pReNative->Core.aHstRegs[idxReg].idxVar;
4405	IEMNATIVE_ASSERT_VAR_IDX(pReNative, idxVar);
4406	AssertStmt(IEMNATIVE_VAR_IDX_UNPACK(idxVar) < RT_ELEMENTS(pReNative->Core.aVars),
4407	IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_REG_IPE_5));
4408	Assert(pReNative->Core.aVars[IEMNATIVE_VAR_IDX_UNPACK(idxVar)].idxReg == idxReg);
4409
4410	if (pReNative->Core.aVars[IEMNATIVE_VAR_IDX_UNPACK(idxVar)].enmKind != kIemNativeVarKind_Stack)
4411	pReNative->Core.aVars[IEMNATIVE_VAR_IDX_UNPACK(idxVar)].idxReg = UINT8_MAX;
4412	else
4413	{
4414	off = iemNativeRegMoveOrSpillStackVar(pReNative, off, idxVar);
4415	Assert(!(pReNative->Core.bmHstRegsWithGstShadow & RT_BIT_32(idxReg)));
4416	}
4417	break;
4418	}
4419
4420	case kIemNativeWhat_Tmp:
4421	case kIemNativeWhat_Arg:
4422	case kIemNativeWhat_rc:
4423	AssertFailedStmt(IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_REG_IPE_5));
4424	default:
4425	AssertFailedStmt(IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_REG_IPE_6));
4426	}
4427
4428	}
4429	if (pReNative->Core.bmHstRegsWithGstShadow & RT_BIT_32(idxReg))
4430	{
4431	Assert(pReNative->Core.aHstRegs[idxReg].fGstRegShadows != 0);
4432	Assert( (pReNative->Core.aHstRegs[idxReg].fGstRegShadows & pReNative->Core.bmGstRegShadows)
4433	== pReNative->Core.aHstRegs[idxReg].fGstRegShadows);
4434	pReNative->Core.bmHstRegsWithGstShadow &= ~RT_BIT_32(idxReg);
4435	pReNative->Core.bmGstRegShadows &= ~pReNative->Core.aHstRegs[idxReg].fGstRegShadows;
4436	pReNative->Core.aHstRegs[idxReg].fGstRegShadows = 0;
4437	}
4438	else
4439	Assert(pReNative->Core.aHstRegs[idxReg].fGstRegShadows == 0);
4440	pReNative->Core.aHstRegs[idxReg].enmWhat = kIemNativeWhat_Arg;
4441	pReNative->Core.aHstRegs[idxReg].idxVar = UINT8_MAX;
4442	}
4443	pReNative->Core.bmHstRegs \|= g_afIemNativeCallRegs[cArgs];
4444	return true;
4445	}
4446
4447
4448	DECL_HIDDEN_THROW(uint8_t) iemNativeRegAssignRc(PIEMRECOMPILERSTATE pReNative, uint8_t idxHstReg);
4449
4450
4451	#if 0
4452	/**
4453	* Frees a register assignment of any type.
4454	*
4455	* @param pReNative The native recompile state.
4456	* @param idxHstReg The register to free.
4457	*
4458	* @note Does not update variables.
4459	*/
4460	DECLHIDDEN(void) iemNativeRegFree(PIEMRECOMPILERSTATE pReNative, uint8_t idxHstReg) RT_NOEXCEPT
4461	{
4462	Assert(idxHstReg < RT_ELEMENTS(pReNative->Core.aHstRegs));
4463	Assert(pReNative->Core.bmHstRegs & RT_BIT_32(idxHstReg));
4464	Assert(!(IEMNATIVE_REG_FIXED_MASK & RT_BIT_32(idxHstReg)));
4465	Assert( pReNative->Core.aHstRegs[idxHstReg].enmWhat == kIemNativeWhat_Var
4466	\|\| pReNative->Core.aHstRegs[idxHstReg].enmWhat == kIemNativeWhat_Tmp
4467	\|\| pReNative->Core.aHstRegs[idxHstReg].enmWhat == kIemNativeWhat_Arg
4468	\|\| pReNative->Core.aHstRegs[idxHstReg].enmWhat == kIemNativeWhat_rc);
4469	Assert( pReNative->Core.aHstRegs[idxHstReg].enmWhat != kIemNativeWhat_Var
4470	\|\| pReNative->Core.aVars[pReNative->Core.aHstRegs[idxHstReg].idxVar].idxReg == UINT8_MAX
4471	\|\| (pReNative->Core.bmVars & RT_BIT_32(pReNative->Core.aHstRegs[idxHstReg].idxVar)));
4472	Assert( (pReNative->Core.bmGstRegShadows & pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows)
4473	== pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows);
4474	Assert( RT_BOOL(pReNative->Core.bmHstRegsWithGstShadow & RT_BIT_32(idxHstReg))
4475	== RT_BOOL(pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows));
4476
4477	pReNative->Core.bmHstRegs &= ~RT_BIT_32(idxHstReg);
4478	/* no flushing, right:
4479	pReNative->Core.bmHstRegsWithGstShadow &= ~RT_BIT_32(idxHstReg);
4480	pReNative->Core.bmGstRegShadows &= ~pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows;
4481	pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows = 0;
4482	*/
4483	}
4484	#endif
4485
4486
4487	/**
4488	* Frees a temporary register.
4489	*
4490	* Any shadow copies of guest registers assigned to the host register will not
4491	* be flushed by this operation.
4492	*/
4493	DECLHIDDEN(void) iemNativeRegFreeTmp(PIEMRECOMPILERSTATE pReNative, uint8_t idxHstReg) RT_NOEXCEPT
4494	{
4495	Assert(pReNative->Core.bmHstRegs & RT_BIT_32(idxHstReg));
4496	Assert(pReNative->Core.aHstRegs[idxHstReg].enmWhat == kIemNativeWhat_Tmp);
4497	pReNative->Core.bmHstRegs &= ~RT_BIT_32(idxHstReg);
4498	Log12(("iemNativeRegFreeTmp: %s (gst: %#RX64)\n",
4499	g_apszIemNativeHstRegNames[idxHstReg], pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows));
4500	}
4501
4502
4503	/**
4504	* Frees a temporary immediate register.
4505	*
4506	* It is assumed that the call has not modified the register, so it still hold
4507	* the same value as when it was allocated via iemNativeRegAllocTmpImm().
4508	*/
4509	DECLHIDDEN(void) iemNativeRegFreeTmpImm(PIEMRECOMPILERSTATE pReNative, uint8_t idxHstReg) RT_NOEXCEPT
4510	{
4511	iemNativeRegFreeTmp(pReNative, idxHstReg);
4512	}
4513
4514
4515	/**
4516	* Frees a register assigned to a variable.
4517	*
4518	* The register will be disassociated from the variable.
4519	*/
4520	DECLHIDDEN(void) iemNativeRegFreeVar(PIEMRECOMPILERSTATE pReNative, uint8_t idxHstReg, bool fFlushShadows) RT_NOEXCEPT
4521	{
4522	Assert(pReNative->Core.bmHstRegs & RT_BIT_32(idxHstReg));
4523	Assert(pReNative->Core.aHstRegs[idxHstReg].enmWhat == kIemNativeWhat_Var);
4524	uint8_t const idxVar = pReNative->Core.aHstRegs[idxHstReg].idxVar;
4525	IEMNATIVE_ASSERT_VAR_IDX(pReNative, idxVar);
4526	Assert(pReNative->Core.aVars[idxVar].idxReg == idxHstReg);
4527
4528	pReNative->Core.aVars[IEMNATIVE_VAR_IDX_UNPACK(idxVar)].idxReg = UINT8_MAX;
4529	pReNative->Core.bmHstRegs &= ~RT_BIT_32(idxHstReg);
4530	if (!fFlushShadows)
4531	Log12(("iemNativeRegFreeVar: %s (gst: %#RX64) idxVar=%#x\n",
4532	g_apszIemNativeHstRegNames[idxHstReg], pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows, idxVar));
4533	else
4534	{
4535	pReNative->Core.bmHstRegsWithGstShadow &= ~RT_BIT_32(idxHstReg);
4536	uint64_t const fGstRegShadowsOld = pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows;
4537	pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows = 0;
4538	pReNative->Core.bmGstRegShadows &= ~fGstRegShadowsOld;
4539	uint64_t fGstRegShadows = fGstRegShadowsOld;
4540	while (fGstRegShadows)
4541	{
4542	unsigned const idxGstReg = ASMBitFirstSetU64(fGstRegShadows) - 1;
4543	fGstRegShadows &= ~RT_BIT_64(idxGstReg);
4544
4545	Assert(pReNative->Core.aidxGstRegShadows[idxGstReg] == idxHstReg);
4546	pReNative->Core.aidxGstRegShadows[idxGstReg] = UINT8_MAX;
4547	}
4548	Log12(("iemNativeRegFreeVar: %s (gst: %#RX64 -> 0) idxVar=%#x\n",
4549	g_apszIemNativeHstRegNames[idxHstReg], fGstRegShadowsOld, idxVar));
4550	}
4551	}
4552
4553
4554	/**
4555	* Called right before emitting a call instruction to move anything important
4556	* out of call-volatile registers, free and flush the call-volatile registers,
4557	* optionally freeing argument variables.
4558	*
4559	* @returns New code buffer offset, UINT32_MAX on failure.
4560	* @param pReNative The native recompile state.
4561	* @param off The code buffer offset.
4562	* @param cArgs The number of arguments the function call takes.
4563	* It is presumed that the host register part of these have
4564	* been allocated as such already and won't need moving,
4565	* just freeing.
4566	* @param fKeepVars Mask of variables that should keep their register
4567	* assignments. Caller must take care to handle these.
4568	*/
4569	DECL_HIDDEN_THROW(uint32_t)
4570	iemNativeRegMoveAndFreeAndFlushAtCall(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint8_t cArgs, uint32_t fKeepVars /= 0/)
4571	{
4572	Assert(cArgs <= IEMNATIVE_CALL_MAX_ARG_COUNT);
4573
4574	/* fKeepVars will reduce this mask. */
4575	uint32_t fRegsToFree = IEMNATIVE_CALL_VOLATILE_GREG_MASK;
4576
4577	/*
4578	* Move anything important out of volatile registers.
4579	*/
4580	if (cArgs > RT_ELEMENTS(g_aidxIemNativeCallRegs))
4581	cArgs = RT_ELEMENTS(g_aidxIemNativeCallRegs);
4582	uint32_t fRegsToMove = IEMNATIVE_CALL_VOLATILE_GREG_MASK
4583	#ifdef IEMNATIVE_REG_FIXED_TMP0
4584	& ~RT_BIT_32(IEMNATIVE_REG_FIXED_TMP0)
4585	#endif
4586	#ifdef IEMNATIVE_REG_FIXED_TMP1
4587	& ~RT_BIT_32(IEMNATIVE_REG_FIXED_TMP1)
4588	#endif
4589	#ifdef IEMNATIVE_REG_FIXED_PC_DBG
4590	& ~RT_BIT_32(IEMNATIVE_REG_FIXED_PC_DBG)
4591	#endif
4592	& ~g_afIemNativeCallRegs[cArgs];
4593
4594	fRegsToMove &= pReNative->Core.bmHstRegs;
4595	if (!fRegsToMove)
4596	{ /* likely */ }
4597	else
4598	{
4599	Log12(("iemNativeRegMoveAndFreeAndFlushAtCall: fRegsToMove=%#x\n", fRegsToMove));
4600	while (fRegsToMove != 0)
4601	{
4602	unsigned const idxReg = ASMBitFirstSetU32(fRegsToMove) - 1;
4603	fRegsToMove &= ~RT_BIT_32(idxReg);
4604
4605	switch (pReNative->Core.aHstRegs[idxReg].enmWhat)
4606	{
4607	case kIemNativeWhat_Var:
4608	{
4609	uint8_t const idxVar = pReNative->Core.aHstRegs[idxReg].idxVar;
4610	IEMNATIVE_ASSERT_VAR_IDX(pReNative, idxVar);
4611	PIEMNATIVEVAR const pVar = &pReNative->Core.aVars[IEMNATIVE_VAR_IDX_UNPACK(idxVar)];
4612	Assert(pVar->idxReg == idxReg);
4613	if (!(RT_BIT_32(idxVar) & fKeepVars))
4614	{
4615	Log12(("iemNativeRegMoveAndFreeAndFlushAtCall: idxVar=%#x enmKind=%d idxReg=%d\n",
4616	idxVar, pVar->enmKind, pVar->idxReg));
4617	if (pVar->enmKind != kIemNativeVarKind_Stack)
4618	pVar->idxReg = UINT8_MAX;
4619	else
4620	off = iemNativeRegMoveOrSpillStackVar(pReNative, off, idxVar);
4621	}
4622	else
4623	fRegsToFree &= ~RT_BIT_32(idxReg);
4624	continue;
4625	}
4626
4627	case kIemNativeWhat_Arg:
4628	AssertMsgFailed(("What?!?: %u\n", idxReg));
4629	continue;
4630
4631	case kIemNativeWhat_rc:
4632	case kIemNativeWhat_Tmp:
4633	AssertMsgFailed(("Missing free: %u\n", idxReg));
4634	continue;
4635
4636	case kIemNativeWhat_FixedTmp:
4637	case kIemNativeWhat_pVCpuFixed:
4638	case kIemNativeWhat_pCtxFixed:
4639	case kIemNativeWhat_PcShadow:
4640	case kIemNativeWhat_FixedReserved:
4641	case kIemNativeWhat_Invalid:
4642	case kIemNativeWhat_End:
4643	AssertFailedStmt(IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_REG_IPE_1));
4644	}
4645	AssertFailedStmt(IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_REG_IPE_2));
4646	}
4647	}
4648
4649	/*
4650	* Do the actual freeing.
4651	*/
4652	if (pReNative->Core.bmHstRegs & fRegsToFree)
4653	Log12(("iemNativeRegMoveAndFreeAndFlushAtCall: bmHstRegs %#x -> %#x\n",
4654	pReNative->Core.bmHstRegs, pReNative->Core.bmHstRegs & ~fRegsToFree));
4655	pReNative->Core.bmHstRegs &= ~fRegsToFree;
4656
4657	/* If there are guest register shadows in any call-volatile register, we
4658	have to clear the corrsponding guest register masks for each register. */
4659	uint32_t fHstRegsWithGstShadow = pReNative->Core.bmHstRegsWithGstShadow & fRegsToFree;
4660	if (fHstRegsWithGstShadow)
4661	{
4662	Log12(("iemNativeRegMoveAndFreeAndFlushAtCall: bmHstRegsWithGstShadow %#RX32 -> %#RX32; removed %#RX32\n",
4663	pReNative->Core.bmHstRegsWithGstShadow, pReNative->Core.bmHstRegsWithGstShadow & ~IEMNATIVE_CALL_VOLATILE_GREG_MASK, fHstRegsWithGstShadow));
4664	pReNative->Core.bmHstRegsWithGstShadow &= ~fHstRegsWithGstShadow;
4665	do
4666	{
4667	unsigned const idxReg = ASMBitFirstSetU32(fHstRegsWithGstShadow) - 1;
4668	fHstRegsWithGstShadow &= ~RT_BIT_32(idxReg);
4669
4670	AssertMsg(pReNative->Core.aHstRegs[idxReg].fGstRegShadows != 0, ("idxReg=%#x\n", idxReg));
4671	pReNative->Core.bmGstRegShadows &= ~pReNative->Core.aHstRegs[idxReg].fGstRegShadows;
4672	pReNative->Core.aHstRegs[idxReg].fGstRegShadows = 0;
4673	} while (fHstRegsWithGstShadow != 0);
4674	}
4675
4676	return off;
4677	}
4678
4679
4680	/**
4681	* Flushes a set of guest register shadow copies.
4682	*
4683	* This is usually done after calling a threaded function or a C-implementation
4684	* of an instruction.
4685	*
4686	* @param pReNative The native recompile state.
4687	* @param fGstRegs Set of guest registers to flush.
4688	*/
4689	DECLHIDDEN(void) iemNativeRegFlushGuestShadows(PIEMRECOMPILERSTATE pReNative, uint64_t fGstRegs) RT_NOEXCEPT
4690	{
4691	/*
4692	* Reduce the mask by what's currently shadowed
4693	*/
4694	uint64_t const bmGstRegShadowsOld = pReNative->Core.bmGstRegShadows;
4695	fGstRegs &= bmGstRegShadowsOld;
4696	if (fGstRegs)
4697	{
4698	uint64_t const bmGstRegShadowsNew = bmGstRegShadowsOld & ~fGstRegs;
4699	Log12(("iemNativeRegFlushGuestShadows: flushing %#RX64 (%#RX64 -> %#RX64)\n", fGstRegs, bmGstRegShadowsOld, bmGstRegShadowsNew));
4700	pReNative->Core.bmGstRegShadows = bmGstRegShadowsNew;
4701	if (bmGstRegShadowsNew)
4702	{
4703	/*
4704	* Partial.
4705	*/
4706	do
4707	{
4708	unsigned const idxGstReg = ASMBitFirstSetU64(fGstRegs) - 1;
4709	uint8_t const idxHstReg = pReNative->Core.aidxGstRegShadows[idxGstReg];
4710	Assert(idxHstReg < RT_ELEMENTS(pReNative->Core.aidxGstRegShadows));
4711	Assert(pReNative->Core.bmHstRegsWithGstShadow & RT_BIT_32(idxHstReg));
4712	Assert(pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows & RT_BIT_64(idxGstReg));
4713
4714	uint64_t const fInThisHstReg = (pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows & fGstRegs) \| RT_BIT_64(idxGstReg);
4715	fGstRegs &= ~fInThisHstReg;
4716	uint64_t const fGstRegShadowsNew = pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows & ~fInThisHstReg;
4717	pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows = fGstRegShadowsNew;
4718	if (!fGstRegShadowsNew)
4719	pReNative->Core.bmHstRegsWithGstShadow &= ~RT_BIT_32(idxHstReg);
4720	} while (fGstRegs != 0);
4721	}
4722	else
4723	{
4724	/*
4725	* Clear all.
4726	*/
4727	do
4728	{
4729	unsigned const idxGstReg = ASMBitFirstSetU64(fGstRegs) - 1;
4730	uint8_t const idxHstReg = pReNative->Core.aidxGstRegShadows[idxGstReg];
4731	Assert(idxHstReg < RT_ELEMENTS(pReNative->Core.aidxGstRegShadows));
4732	Assert(pReNative->Core.bmHstRegsWithGstShadow & RT_BIT_32(idxHstReg));
4733	Assert(pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows & RT_BIT_64(idxGstReg));
4734
4735	fGstRegs &= ~(pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows \| RT_BIT_64(idxGstReg));
4736	pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows = 0;
4737	} while (fGstRegs != 0);
4738	pReNative->Core.bmHstRegsWithGstShadow = 0;
4739	}
4740	}
4741	}
4742
4743
4744	/**
4745	* Flushes guest register shadow copies held by a set of host registers.
4746	*
4747	* This is used with the TLB lookup code for ensuring that we don't carry on
4748	* with any guest shadows in volatile registers, as these will get corrupted by
4749	* a TLB miss.
4750	*
4751	* @param pReNative The native recompile state.
4752	* @param fHstRegs Set of host registers to flush guest shadows for.
4753	*/
4754	DECLHIDDEN(void) iemNativeRegFlushGuestShadowsByHostMask(PIEMRECOMPILERSTATE pReNative, uint32_t fHstRegs) RT_NOEXCEPT
4755	{
4756	/*
4757	* Reduce the mask by what's currently shadowed.
4758	*/
4759	uint32_t const bmHstRegsWithGstShadowOld = pReNative->Core.bmHstRegsWithGstShadow;
4760	fHstRegs &= bmHstRegsWithGstShadowOld;
4761	if (fHstRegs)
4762	{
4763	uint32_t const bmHstRegsWithGstShadowNew = bmHstRegsWithGstShadowOld & ~fHstRegs;
4764	Log12(("iemNativeRegFlushGuestShadowsByHostMask: flushing %#RX32 (%#RX32 -> %#RX32)\n",
4765	fHstRegs, bmHstRegsWithGstShadowOld, bmHstRegsWithGstShadowNew));
4766	pReNative->Core.bmHstRegsWithGstShadow = bmHstRegsWithGstShadowNew;
4767	if (bmHstRegsWithGstShadowNew)
4768	{
4769	/*
4770	* Partial (likely).
4771	*/
4772	uint64_t fGstShadows = 0;
4773	do
4774	{
4775	unsigned const idxHstReg = ASMBitFirstSetU32(fHstRegs) - 1;
4776	Assert(!(pReNative->Core.bmHstRegs & RT_BIT_32(idxHstReg)));
4777	Assert( (pReNative->Core.bmGstRegShadows & pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows)
4778	== pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows);
4779
4780	fGstShadows \|= pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows;
4781	pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows = 0;
4782	fHstRegs &= ~RT_BIT_32(idxHstReg);
4783	} while (fHstRegs != 0);
4784	pReNative->Core.bmGstRegShadows &= ~fGstShadows;
4785	}
4786	else
4787	{
4788	/*
4789	* Clear all.
4790	*/
4791	do
4792	{
4793	unsigned const idxHstReg = ASMBitFirstSetU32(fHstRegs) - 1;
4794	Assert(!(pReNative->Core.bmHstRegs & RT_BIT_32(idxHstReg)));
4795	Assert( (pReNative->Core.bmGstRegShadows & pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows)
4796	== pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows);
4797
4798	pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows = 0;
4799	fHstRegs &= ~RT_BIT_32(idxHstReg);
4800	} while (fHstRegs != 0);
4801	pReNative->Core.bmGstRegShadows = 0;
4802	}
4803	}
4804	}
4805
4806
4807	/**
4808	* Restores guest shadow copies in volatile registers.
4809	*
4810	* This is used after calling a helper function (think TLB miss) to restore the
4811	* register state of volatile registers.
4812	*
4813	* @param pReNative The native recompile state.
4814	* @param off The code buffer offset.
4815	* @param fHstRegsActiveShadows Set of host registers which are allowed to
4816	* be active (allocated) w/o asserting. Hack.
4817	* @see iemNativeVarSaveVolatileRegsPreHlpCall(),
4818	* iemNativeVarRestoreVolatileRegsPostHlpCall()
4819	*/
4820	DECL_HIDDEN_THROW(uint32_t)
4821	iemNativeRegRestoreGuestShadowsInVolatileRegs(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint32_t fHstRegsActiveShadows)
4822	{
4823	uint32_t fHstRegs = pReNative->Core.bmHstRegsWithGstShadow & IEMNATIVE_CALL_VOLATILE_GREG_MASK;
4824	if (fHstRegs)
4825	{
4826	Log12(("iemNativeRegRestoreGuestShadowsInVolatileRegs: %#RX32\n", fHstRegs));
4827	do
4828	{
4829	unsigned const idxHstReg = ASMBitFirstSetU32(fHstRegs) - 1;
4830
4831	/* It's not fatal if a register is active holding a variable that
4832	shadowing a guest register, ASSUMING all pending guest register
4833	writes were flushed prior to the helper call. However, we'll be
4834	emitting duplicate restores, so it wasts code space. */
4835	Assert(!(pReNative->Core.bmHstRegs & ~fHstRegsActiveShadows & RT_BIT_32(idxHstReg)));
4836	RT_NOREF(fHstRegsActiveShadows);
4837
4838	uint64_t const fGstRegShadows = pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows;
4839	Assert((pReNative->Core.bmGstRegShadows & fGstRegShadows) == fGstRegShadows);
4840	AssertStmt(fGstRegShadows != 0 && fGstRegShadows < RT_BIT_64(kIemNativeGstReg_End),
4841	IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_REG_IPE_12));
4842
4843	unsigned const idxGstReg = ASMBitFirstSetU64(fGstRegShadows) - 1;
4844	off = iemNativeEmitLoadGprWithGstShadowReg(pReNative, off, idxHstReg, (IEMNATIVEGSTREG)idxGstReg);
4845
4846	fHstRegs &= ~RT_BIT_32(idxHstReg);
4847	} while (fHstRegs != 0);
4848	}
4849	return off;
4850	}
4851
4852
4853
4854
4855	/*********************************************************************************************************************************
4856	* SIMD register allocator (largely code duplication of the GPR allocator for now but might diverge) *
4857	*********************************************************************************************************************************/
4858	#ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
4859
4860	/**
4861	* Info about shadowed guest SIMD register values.
4862	* @see IEMNATIVEGSTSIMDREG
4863	*/
4864	static struct
4865	{
4866	/** Offset in VMCPU of XMM (low 128-bit) registers. */
4867	uint32_t offXmm;
4868	/** Offset in VMCPU of YmmHi (high 128-bit) registers. */
4869	uint32_t offYmm;
4870	/** Name (for logging). */
4871	const char *pszName;
4872	} const g_aGstSimdShadowInfo[] =
4873	{
4874	#define CPUMCTX_OFF_AND_SIZE(a_iSimdReg) (uint32_t)RT_UOFFSETOF(VMCPU, cpum.GstCtx.XState.x87.aXMM[a_iSimdReg]), \
4875	(uint32_t)RT_UOFFSETOF(VMCPU, cpum.GstCtx.XState.u.YmmHi.aYmmHi[a_iSimdReg])
4876	/* [kIemNativeGstSimdReg_SimdRegFirst + 0] = */ { CPUMCTX_OFF_AND_SIZE(0), "ymm0", },
4877	/* [kIemNativeGstSimdReg_SimdRegFirst + 1] = */ { CPUMCTX_OFF_AND_SIZE(1), "ymm1", },
4878	/* [kIemNativeGstSimdReg_SimdRegFirst + 2] = */ { CPUMCTX_OFF_AND_SIZE(2), "ymm2", },
4879	/* [kIemNativeGstSimdReg_SimdRegFirst + 3] = */ { CPUMCTX_OFF_AND_SIZE(3), "ymm3", },
4880	/* [kIemNativeGstSimdReg_SimdRegFirst + 4] = */ { CPUMCTX_OFF_AND_SIZE(4), "ymm4", },
4881	/* [kIemNativeGstSimdReg_SimdRegFirst + 5] = */ { CPUMCTX_OFF_AND_SIZE(5), "ymm5", },
4882	/* [kIemNativeGstSimdReg_SimdRegFirst + 6] = */ { CPUMCTX_OFF_AND_SIZE(6), "ymm6", },
4883	/* [kIemNativeGstSimdReg_SimdRegFirst + 7] = */ { CPUMCTX_OFF_AND_SIZE(7), "ymm7", },
4884	/* [kIemNativeGstSimdReg_SimdRegFirst + 8] = */ { CPUMCTX_OFF_AND_SIZE(8), "ymm8", },
4885	/* [kIemNativeGstSimdReg_SimdRegFirst + 9] = */ { CPUMCTX_OFF_AND_SIZE(9), "ymm9", },
4886	/* [kIemNativeGstSimdReg_SimdRegFirst + 10] = */ { CPUMCTX_OFF_AND_SIZE(10), "ymm10", },
4887	/* [kIemNativeGstSimdReg_SimdRegFirst + 11] = */ { CPUMCTX_OFF_AND_SIZE(11), "ymm11", },
4888	/* [kIemNativeGstSimdReg_SimdRegFirst + 12] = */ { CPUMCTX_OFF_AND_SIZE(12), "ymm12", },
4889	/* [kIemNativeGstSimdReg_SimdRegFirst + 13] = */ { CPUMCTX_OFF_AND_SIZE(13), "ymm13", },
4890	/* [kIemNativeGstSimdReg_SimdRegFirst + 14] = */ { CPUMCTX_OFF_AND_SIZE(14), "ymm14", },
4891	/* [kIemNativeGstSimdReg_SimdRegFirst + 15] = */ { CPUMCTX_OFF_AND_SIZE(15), "ymm15", },
4892	#undef CPUMCTX_OFF_AND_SIZE
4893	};
4894	AssertCompile(RT_ELEMENTS(g_aGstSimdShadowInfo) == kIemNativeGstSimdReg_End);
4895
4896
4897	#ifdef LOG_ENABLED
4898	/** Host CPU SIMD register names. */
4899	DECL_HIDDEN_CONST(const char * const) g_apszIemNativeHstSimdRegNames[] =
4900	{
4901	#ifdef RT_ARCH_AMD64
4902	"ymm0", "ymm1", "ymm2", "ymm3", "ymm4", "ymm5", "ymm6", "ymm7", "ymm8", "ymm9", "ymm10", "ymm11", "ymm12", "ymm13", "ymm14", "ymm15"
4903	#elif RT_ARCH_ARM64
4904	"v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7", "v8", "v9", "v10", "v11", "v12", "v13", "v14", "v15",
4905	"v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23", "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31",
4906	#else
4907	# error "port me"
4908	#endif
4909	};
4910	#endif
4911
4912
4913	/**
4914	* Frees a temporary SIMD register.
4915	*
4916	* Any shadow copies of guest registers assigned to the host register will not
4917	* be flushed by this operation.
4918	*/
4919	DECLHIDDEN(void) iemNativeSimdRegFreeTmp(PIEMRECOMPILERSTATE pReNative, uint8_t idxHstSimdReg) RT_NOEXCEPT
4920	{
4921	Assert(pReNative->Core.bmHstSimdRegs & RT_BIT_32(idxHstSimdReg));
4922	Assert(pReNative->Core.aHstSimdRegs[idxHstSimdReg].enmWhat == kIemNativeWhat_Tmp);
4923	pReNative->Core.bmHstSimdRegs &= ~RT_BIT_32(idxHstSimdReg);
4924	Log12(("iemNativeSimdRegFreeTmp: %s (gst: %#RX64)\n",
4925	g_apszIemNativeHstSimdRegNames[idxHstSimdReg], pReNative->Core.aHstSimdRegs[idxHstSimdReg].fGstRegShadows));
4926	}
4927
4928
4929	/**
4930	* Emits code to flush a pending write of the given SIMD register if any, also flushes the guest to host SIMD register association.
4931	*
4932	* @returns New code bufferoffset.
4933	* @param pReNative The native recompile state.
4934	* @param off Current code buffer position.
4935	* @param enmGstSimdReg The guest SIMD register to flush.
4936	*/
4937	DECL_HIDDEN_THROW(uint32_t)
4938	iemNativeSimdRegFlushPendingWrite(PIEMRECOMPILERSTATE pReNative, uint32_t off, IEMNATIVEGSTSIMDREG enmGstSimdReg)
4939	{
4940	uint8_t const idxHstSimdReg = pReNative->Core.aidxGstSimdRegShadows[enmGstSimdReg];
4941
4942	Log12(("iemNativeSimdRegFlushPendingWrite: Clearing guest register %s shadowed by host %s with state DirtyLo:%u DirtyHi:%u\n",
4943	g_aGstSimdShadowInfo[enmGstSimdReg].pszName, g_apszIemNativeHstSimdRegNames[idxHstSimdReg],
4944	IEMNATIVE_SIMD_REG_STATE_IS_DIRTY_LO_U128(pReNative, enmGstSimdReg),
4945	IEMNATIVE_SIMD_REG_STATE_IS_DIRTY_HI_U128(pReNative, enmGstSimdReg)));
4946
4947	if (IEMNATIVE_SIMD_REG_STATE_IS_DIRTY_LO_U128(pReNative, enmGstSimdReg))
4948	{
4949	Assert( pReNative->Core.aHstSimdRegs[idxHstSimdReg].enmLoaded == kIemNativeGstSimdRegLdStSz_256
4950	\|\| pReNative->Core.aHstSimdRegs[idxHstSimdReg].enmLoaded == kIemNativeGstSimdRegLdStSz_Low128);
4951	off = iemNativeEmitSimdStoreVecRegToVCpuLowU128(pReNative, off, idxHstSimdReg, g_aGstSimdShadowInfo[enmGstSimdReg].offXmm);
4952	}
4953
4954	if (IEMNATIVE_SIMD_REG_STATE_IS_DIRTY_HI_U128(pReNative, enmGstSimdReg))
4955	{
4956	Assert( pReNative->Core.aHstSimdRegs[idxHstSimdReg].enmLoaded == kIemNativeGstSimdRegLdStSz_256
4957	\|\| pReNative->Core.aHstSimdRegs[idxHstSimdReg].enmLoaded == kIemNativeGstSimdRegLdStSz_High128);
4958	off = iemNativeEmitSimdStoreVecRegToVCpuHighU128(pReNative, off, idxHstSimdReg, g_aGstSimdShadowInfo[enmGstSimdReg].offYmm);
4959	}
4960
4961	IEMNATIVE_SIMD_REG_STATE_CLR_DIRTY(pReNative, enmGstSimdReg);
4962	return off;
4963	}
4964
4965
4966	/**
4967	* Locate a register, possibly freeing one up.
4968	*
4969	* This ASSUMES the caller has done the minimal/optimal allocation checks and
4970	* failed.
4971	*
4972	* @returns Host register number on success. Returns UINT8_MAX if no registers
4973	* found, the caller is supposed to deal with this and raise a
4974	* allocation type specific status code (if desired).
4975	*
4976	* @throws VBox status code if we're run into trouble spilling a variable of
4977	* recording debug info. Does NOT throw anything if we're out of
4978	* registers, though.
4979	*/
4980	static uint8_t iemNativeSimdRegAllocFindFree(PIEMRECOMPILERSTATE pReNative, uint32_t *poff, bool fPreferVolatile,
4981	uint32_t fRegMask = IEMNATIVE_HST_SIMD_REG_MASK & ~IEMNATIVE_SIMD_REG_FIXED_MASK)
4982	{
4983	//STAM_COUNTER_INC(&pReNative->pVCpu->iem.s.StatNativeRegFindFree);
4984	Assert(!(fRegMask & ~IEMNATIVE_HST_SIMD_REG_MASK));
4985	Assert(!(fRegMask & IEMNATIVE_SIMD_REG_FIXED_MASK));
4986
4987	/*
4988	* Try a freed register that's shadowing a guest register.
4989	*/
4990	uint32_t fRegs = ~pReNative->Core.bmHstSimdRegs & fRegMask;
4991	if (fRegs)
4992	{
4993	//STAM_COUNTER_INC(&pReNative->pVCpu->iem.s.StatNativeRegFindFreeNoVar);
4994
4995	#if 0 /** @todo def IEMNATIVE_WITH_LIVENESS_ANALYSIS */
4996	/*
4997	* When we have livness information, we use it to kick out all shadowed
4998	* guest register that will not be needed any more in this TB. If we're
4999	* lucky, this may prevent us from ending up here again.
5000	*
5001	* Note! We must consider the previous entry here so we don't free
5002	* anything that the current threaded function requires (current
5003	* entry is produced by the next threaded function).
5004	*/
5005	uint32_t const idxCurCall = pReNative->idxCurCall;
5006	if (idxCurCall > 0)
5007	{
5008	PCIEMLIVENESSENTRY const pLivenessEntry = &pReNative->paLivenessEntries[idxCurCall - 1];
5009
5010	# ifndef IEMLIVENESS_EXTENDED_LAYOUT
5011	/* Construct a mask of the guest registers in the UNUSED and XCPT_OR_CALL state. */
5012	AssertCompile(IEMLIVENESS_STATE_UNUSED == 1 && IEMLIVENESS_STATE_XCPT_OR_CALL == 2);
5013	uint64_t fToFreeMask = pLivenessEntry->Bit0.bm64 ^ pLivenessEntry->Bit1.bm64; /* mask of regs in either UNUSED */
5014	#else
5015	/* Construct a mask of the registers not in the read or write state.
5016	Note! We could skips writes, if they aren't from us, as this is just
5017	a hack to prevent trashing registers that have just been written
5018	or will be written when we retire the current instruction. */
5019	uint64_t fToFreeMask = ~pLivenessEntry->aBits[IEMLIVENESS_BIT_READ].bm64
5020	& ~pLivenessEntry->aBits[IEMLIVENESS_BIT_WRITE].bm64
5021	& IEMLIVENESSBIT_MASK;
5022	#endif
5023	/* If it matches any shadowed registers. */
5024	if (pReNative->Core.bmGstRegShadows & fToFreeMask)
5025	{
5026	STAM_COUNTER_INC(&pReNative->pVCpu->iem.s.StatNativeRegFindFreeLivenessUnshadowed);
5027	iemNativeRegFlushGuestShadows(pReNative, fToFreeMask);
5028	Assert(fRegs == (~pReNative->Core.bmHstRegs & fRegMask)); /* this shall not change. */
5029
5030	/* See if we've got any unshadowed registers we can return now. */
5031	uint32_t const fUnshadowedRegs = fRegs & ~pReNative->Core.bmHstRegsWithGstShadow;
5032	if (fUnshadowedRegs)
5033	{
5034	STAM_COUNTER_INC(&pReNative->pVCpu->iem.s.StatNativeRegFindFreeLivenessHelped);
5035	return (fPreferVolatile
5036	? ASMBitFirstSetU32(fUnshadowedRegs)
5037	: ASMBitLastSetU32( fUnshadowedRegs & ~IEMNATIVE_CALL_VOLATILE_GREG_MASK
5038	? fUnshadowedRegs & ~IEMNATIVE_CALL_VOLATILE_GREG_MASK : fUnshadowedRegs))
5039	- 1;
5040	}
5041	}
5042	}
5043	#endif /* IEMNATIVE_WITH_LIVENESS_ANALYSIS */
5044
5045	unsigned const idxReg = (fPreferVolatile
5046	? ASMBitFirstSetU32(fRegs)
5047	: ASMBitLastSetU32( fRegs & ~IEMNATIVE_CALL_VOLATILE_SIMD_REG_MASK
5048	? fRegs & ~IEMNATIVE_CALL_VOLATILE_SIMD_REG_MASK : fRegs))
5049	- 1;
5050
5051	Assert(pReNative->Core.aHstSimdRegs[idxReg].fGstRegShadows != 0);
5052	Assert( (pReNative->Core.aHstSimdRegs[idxReg].fGstRegShadows & pReNative->Core.bmGstSimdRegShadows)
5053	== pReNative->Core.aHstSimdRegs[idxReg].fGstRegShadows);
5054	Assert(pReNative->Core.bmHstSimdRegsWithGstShadow & RT_BIT_32(idxReg));
5055	Assert(pReNative->Core.aHstSimdRegs[idxReg].enmLoaded == kIemNativeGstSimdRegLdStSz_Invalid);
5056
5057	/* We need to flush any pending guest register writes this host SIMD register shadows. */
5058	uint32_t fGstRegShadows = pReNative->Core.aHstSimdRegs[idxReg].fGstRegShadows;
5059	uint32_t idxGstSimdReg = 0;
5060	do
5061	{
5062	if (fGstRegShadows & 0x1)
5063	poff = iemNativeSimdRegFlushPendingWrite(pReNative, poff, IEMNATIVEGSTSIMDREG_SIMD(idxGstSimdReg));
5064	Assert(!IEMNATIVE_SIMD_REG_STATE_IS_DIRTY_U256(pReNative, idxGstSimdReg));
5065	idxGstSimdReg++;
5066	fGstRegShadows >>= 1;
5067	} while (fGstRegShadows);
5068
5069	pReNative->Core.bmHstSimdRegsWithGstShadow &= ~RT_BIT_32(idxReg);
5070	pReNative->Core.bmGstSimdRegShadows &= ~pReNative->Core.aHstSimdRegs[idxReg].fGstRegShadows;
5071	pReNative->Core.aHstSimdRegs[idxReg].fGstRegShadows = 0;
5072	pReNative->Core.aHstSimdRegs[idxReg].enmLoaded = kIemNativeGstSimdRegLdStSz_Invalid;
5073	return idxReg;
5074	}
5075
5076	/*
5077	* Try free up a variable that's in a register.
5078	*
5079	* We do two rounds here, first evacuating variables we don't need to be
5080	* saved on the stack, then in the second round move things to the stack.
5081	*/
5082	//STAM_REL_COUNTER_INC(&pReNative->pVCpu->iem.s.StatNativeRegFindFreeVar);
5083	AssertReleaseFailed(); /** @todo No variable support right now. */
5084	#if 0
5085	for (uint32_t iLoop = 0; iLoop < 2; iLoop++)
5086	{
5087	uint32_t fVars = pReNative->Core.bmSimdVars;
5088	while (fVars)
5089	{
5090	uint32_t const idxVar = ASMBitFirstSetU32(fVars) - 1;
5091	uint8_t const idxReg = pReNative->Core.aSimdVars[idxVar].idxReg;
5092	if ( idxReg < RT_ELEMENTS(pReNative->Core.aHstSimdRegs)
5093	&& (RT_BIT_32(idxReg) & fRegMask)
5094	&& ( iLoop == 0
5095	? pReNative->Core.aSimdVars[idxVar].enmKind != kIemNativeVarKind_Stack
5096	: pReNative->Core.aSimdVars[idxVar].enmKind == kIemNativeVarKind_Stack)
5097	&& !pReNative->Core.aSimdVars[idxVar].fRegAcquired)
5098	{
5099	Assert(pReNative->Core.bmHstRegs & RT_BIT_32(idxReg));
5100	Assert( (pReNative->Core.bmGstSimdRegShadows & pReNative->Core.aHstSimdRegs[idxReg].fGstRegShadows)
5101	== pReNative->Core.aHstSimdRegs[idxReg].fGstRegShadows);
5102	Assert(pReNative->Core.bmGstSimdRegShadows < RT_BIT_64(kIemNativeGstReg_End));
5103	Assert( RT_BOOL(pReNative->Core.bmHstRegsWithGstShadow & RT_BIT_32(idxReg))
5104	== RT_BOOL(pReNative->Core.aHstSimdRegs[idxReg].fGstRegShadows));
5105
5106	if (pReNative->Core.aSimdVars[idxVar].enmKind == kIemNativeVarKind_Stack)
5107	{
5108	uint8_t const idxStackSlot = iemNativeVarGetStackSlot(pReNative, IEMNATIVE_VAR_IDX_PACK(idxVar));
5109	poff = iemNativeEmitStoreGprByBp(pReNative, poff, iemNativeStackCalcBpDisp(idxStackSlot), idxReg);
5110	}
5111
5112	pReNative->Core.aSimdVars[idxVar].idxReg = UINT8_MAX;
5113	pReNative->Core.bmHstSimdRegs &= ~RT_BIT_32(idxReg);
5114
5115	pReNative->Core.bmHstSimdRegsWithGstShadow &= ~RT_BIT_32(idxReg);
5116	pReNative->Core.bmGstSimdRegShadows &= ~pReNative->Core.aHstSimdRegs[idxReg].fGstRegShadows;
5117	pReNative->Core.aHstSimdRegs[idxReg].fGstRegShadows = 0;
5118	return idxReg;
5119	}
5120	fVars &= ~RT_BIT_32(idxVar);
5121	}
5122	}
5123	#endif
5124
5125	AssertFailed();
5126	return UINT8_MAX;
5127	}
5128
5129
5130	/**
5131	* Flushes a set of guest register shadow copies.
5132	*
5133	* This is usually done after calling a threaded function or a C-implementation
5134	* of an instruction.
5135	*
5136	* @param pReNative The native recompile state.
5137	* @param fGstSimdRegs Set of guest SIMD registers to flush.
5138	*/
5139	DECLHIDDEN(void) iemNativeSimdRegFlushGuestShadows(PIEMRECOMPILERSTATE pReNative, uint64_t fGstSimdRegs) RT_NOEXCEPT
5140	{
5141	/*
5142	* Reduce the mask by what's currently shadowed
5143	*/
5144	uint64_t const bmGstSimdRegShadows = pReNative->Core.bmGstSimdRegShadows;
5145	fGstSimdRegs &= bmGstSimdRegShadows;
5146	if (fGstSimdRegs)
5147	{
5148	uint64_t const bmGstSimdRegShadowsNew = bmGstSimdRegShadows & ~fGstSimdRegs;
5149	Log12(("iemNativeSimdRegFlushGuestShadows: flushing %#RX64 (%#RX64 -> %#RX64)\n", fGstSimdRegs, bmGstSimdRegShadows, bmGstSimdRegShadowsNew));
5150	pReNative->Core.bmGstSimdRegShadows = bmGstSimdRegShadowsNew;
5151	if (bmGstSimdRegShadowsNew)
5152	{
5153	/*
5154	* Partial.
5155	*/
5156	do
5157	{
5158	unsigned const idxGstReg = ASMBitFirstSetU64(fGstSimdRegs) - 1;
5159	uint8_t const idxHstReg = pReNative->Core.aidxGstSimdRegShadows[idxGstReg];
5160	Assert(idxHstReg < RT_ELEMENTS(pReNative->Core.aidxGstSimdRegShadows));
5161	Assert(pReNative->Core.bmHstSimdRegsWithGstShadow & RT_BIT_32(idxHstReg));
5162	Assert(pReNative->Core.aHstSimdRegs[idxHstReg].fGstRegShadows & RT_BIT_64(idxGstReg));
5163	Assert(!IEMNATIVE_SIMD_REG_STATE_IS_DIRTY_U256(pReNative, idxGstReg));
5164
5165	uint64_t const fInThisHstReg = (pReNative->Core.aHstSimdRegs[idxHstReg].fGstRegShadows & fGstSimdRegs) \| RT_BIT_64(idxGstReg);
5166	fGstSimdRegs &= ~fInThisHstReg;
5167	uint64_t const fGstRegShadowsNew = pReNative->Core.aHstSimdRegs[idxHstReg].fGstRegShadows & ~fInThisHstReg;
5168	pReNative->Core.aHstSimdRegs[idxHstReg].fGstRegShadows = fGstRegShadowsNew;
5169	if (!fGstRegShadowsNew)
5170	{
5171	pReNative->Core.bmHstSimdRegsWithGstShadow &= ~RT_BIT_32(idxHstReg);
5172	pReNative->Core.aHstSimdRegs[idxHstReg].enmLoaded = kIemNativeGstSimdRegLdStSz_Invalid;
5173	}
5174	} while (fGstSimdRegs != 0);
5175	}
5176	else
5177	{
5178	/*
5179	* Clear all.
5180	*/
5181	do
5182	{
5183	unsigned const idxGstReg = ASMBitFirstSetU64(fGstSimdRegs) - 1;
5184	uint8_t const idxHstReg = pReNative->Core.aidxGstSimdRegShadows[idxGstReg];
5185	Assert(idxHstReg < RT_ELEMENTS(pReNative->Core.aidxGstSimdRegShadows));
5186	Assert(pReNative->Core.bmHstSimdRegsWithGstShadow & RT_BIT_32(idxHstReg));
5187	Assert(pReNative->Core.aHstSimdRegs[idxHstReg].fGstRegShadows & RT_BIT_64(idxGstReg));
5188	Assert(!IEMNATIVE_SIMD_REG_STATE_IS_DIRTY_U256(pReNative, idxGstReg));
5189
5190	fGstSimdRegs &= ~(pReNative->Core.aHstSimdRegs[idxHstReg].fGstRegShadows \| RT_BIT_64(idxGstReg));
5191	pReNative->Core.aHstSimdRegs[idxHstReg].fGstRegShadows = 0;
5192	pReNative->Core.aHstSimdRegs[idxHstReg].enmLoaded = kIemNativeGstSimdRegLdStSz_Invalid;
5193	} while (fGstSimdRegs != 0);
5194	pReNative->Core.bmHstSimdRegsWithGstShadow = 0;
5195	}
5196	}
5197	}
5198
5199
5200	/**
5201	* Allocates a temporary host SIMD register.
5202	*
5203	* This may emit code to save register content onto the stack in order to free
5204	* up a register.
5205	*
5206	* @returns The host register number; throws VBox status code on failure,
5207	* so no need to check the return value.
5208	* @param pReNative The native recompile state.
5209	* @param poff Pointer to the variable with the code buffer position.
5210	* This will be update if we need to move a variable from
5211	* register to stack in order to satisfy the request.
5212	* @param fPreferVolatile Whether to prefer volatile over non-volatile
5213	* registers (@c true, default) or the other way around
5214	* (@c false, for iemNativeRegAllocTmpForGuestReg()).
5215	*/
5216	DECL_HIDDEN_THROW(uint8_t) iemNativeSimdRegAllocTmp(PIEMRECOMPILERSTATE pReNative, uint32_t poff, bool fPreferVolatile /= true*/)
5217	{
5218	/*
5219	* Try find a completely unused register, preferably a call-volatile one.
5220	*/
5221	uint8_t idxSimdReg;
5222	uint32_t fRegs = ~pReNative->Core.bmHstRegs
5223	& ~pReNative->Core.bmHstRegsWithGstShadow
5224	& (~IEMNATIVE_SIMD_REG_FIXED_MASK & IEMNATIVE_HST_SIMD_REG_MASK);
5225	if (fRegs)
5226	{
5227	if (fPreferVolatile)
5228	idxSimdReg = (uint8_t)ASMBitFirstSetU32( fRegs & IEMNATIVE_CALL_VOLATILE_SIMD_REG_MASK
5229	? fRegs & IEMNATIVE_CALL_VOLATILE_SIMD_REG_MASK : fRegs) - 1;
5230	else
5231	idxSimdReg = (uint8_t)ASMBitFirstSetU32( fRegs & ~IEMNATIVE_CALL_VOLATILE_SIMD_REG_MASK
5232	? fRegs & ~IEMNATIVE_CALL_VOLATILE_SIMD_REG_MASK : fRegs) - 1;
5233	Assert(pReNative->Core.aHstSimdRegs[idxSimdReg].fGstRegShadows == 0);
5234	Assert(!(pReNative->Core.bmHstSimdRegsWithGstShadow & RT_BIT_32(idxSimdReg)));
5235	Log12(("iemNativeSimdRegAllocTmp: %s\n", g_apszIemNativeHstSimdRegNames[idxSimdReg]));
5236	}
5237	else
5238	{
5239	idxSimdReg = iemNativeSimdRegAllocFindFree(pReNative, poff, fPreferVolatile);
5240	AssertStmt(idxSimdReg != UINT8_MAX, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_REG_ALLOCATOR_NO_FREE_TMP));
5241	Log12(("iemNativeSimdRegAllocTmp: %s (slow)\n", g_apszIemNativeHstSimdRegNames[idxSimdReg]));
5242	}
5243
5244	Assert(pReNative->Core.aHstSimdRegs[idxSimdReg].enmLoaded == kIemNativeGstSimdRegLdStSz_Invalid);
5245	return iemNativeSimdRegMarkAllocated(pReNative, idxSimdReg, kIemNativeWhat_Tmp);
5246	}
5247
5248
5249	/**
5250	* Alternative version of iemNativeSimdRegAllocTmp that takes mask with acceptable
5251	* registers.
5252	*
5253	* @returns The host register number; throws VBox status code on failure,
5254	* so no need to check the return value.
5255	* @param pReNative The native recompile state.
5256	* @param poff Pointer to the variable with the code buffer position.
5257	* This will be update if we need to move a variable from
5258	* register to stack in order to satisfy the request.
5259	* @param fRegMask Mask of acceptable registers.
5260	* @param fPreferVolatile Whether to prefer volatile over non-volatile
5261	* registers (@c true, default) or the other way around
5262	* (@c false, for iemNativeRegAllocTmpForGuestReg()).
5263	*/
5264	DECL_HIDDEN_THROW(uint8_t) iemNativeSimdRegAllocTmpEx(PIEMRECOMPILERSTATE pReNative, uint32_t *poff, uint32_t fRegMask,
5265	bool fPreferVolatile /= true/)
5266	{
5267	Assert(!(fRegMask & ~IEMNATIVE_HST_SIMD_REG_MASK));
5268	Assert(!(fRegMask & IEMNATIVE_SIMD_REG_FIXED_MASK));
5269
5270	/*
5271	* Try find a completely unused register, preferably a call-volatile one.
5272	*/
5273	uint8_t idxSimdReg;
5274	uint32_t fRegs = ~pReNative->Core.bmHstSimdRegs
5275	& ~pReNative->Core.bmHstSimdRegsWithGstShadow
5276	& (~IEMNATIVE_SIMD_REG_FIXED_MASK & IEMNATIVE_HST_SIMD_REG_MASK)
5277	& fRegMask;
5278	if (fRegs)
5279	{
5280	if (fPreferVolatile)
5281	idxSimdReg = (uint8_t)ASMBitFirstSetU32( fRegs & IEMNATIVE_CALL_VOLATILE_SIMD_REG_MASK
5282	? fRegs & IEMNATIVE_CALL_VOLATILE_SIMD_REG_MASK : fRegs) - 1;
5283	else
5284	idxSimdReg = (uint8_t)ASMBitFirstSetU32( fRegs & ~IEMNATIVE_CALL_VOLATILE_SIMD_REG_MASK
5285	? fRegs & ~IEMNATIVE_CALL_VOLATILE_SIMD_REG_MASK : fRegs) - 1;
5286	Assert(pReNative->Core.aHstSimdRegs[idxSimdReg].fGstRegShadows == 0);
5287	Assert(!(pReNative->Core.bmHstSimdRegsWithGstShadow & RT_BIT_32(idxSimdReg)));
5288	Log12(("iemNativeSimdRegAllocTmpEx: %s\n", g_apszIemNativeHstSimdRegNames[idxSimdReg]));
5289	}
5290	else
5291	{
5292	idxSimdReg = iemNativeSimdRegAllocFindFree(pReNative, poff, fPreferVolatile, fRegMask);
5293	AssertStmt(idxSimdReg != UINT8_MAX, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_REG_ALLOCATOR_NO_FREE_TMP));
5294	Log12(("iemNativeSimdRegAllocTmpEx: %s (slow)\n", g_apszIemNativeHstSimdRegNames[idxSimdReg]));
5295	}
5296
5297	Assert(pReNative->Core.aHstSimdRegs[idxSimdReg].enmLoaded == kIemNativeGstSimdRegLdStSz_Invalid);
5298	return iemNativeSimdRegMarkAllocated(pReNative, idxSimdReg, kIemNativeWhat_Tmp);
5299	}
5300
5301
5302	/**
5303	* Sets the indiactor for which part of the given SIMD register has valid data loaded.
5304	*
5305	* @param pReNative The native recompile state.
5306	* @param idxHstSimdReg The host SIMD register to update the state for.
5307	* @param enmLoadSz The load size to set.
5308	*/
5309	DECL_FORCE_INLINE(void) iemNativeSimdRegSetValidLoadFlag(PIEMRECOMPILERSTATE pReNative, uint8_t idxHstSimdReg,
5310	IEMNATIVEGSTSIMDREGLDSTSZ enmLoadSz)
5311	{
5312	/* Everything valid already? -> nothing to do. */
5313	if (pReNative->Core.aHstSimdRegs[idxHstSimdReg].enmLoaded == kIemNativeGstSimdRegLdStSz_256)
5314	return;
5315
5316	if (pReNative->Core.aHstSimdRegs[idxHstSimdReg].enmLoaded == kIemNativeGstSimdRegLdStSz_Invalid)
5317	pReNative->Core.aHstSimdRegs[idxHstSimdReg].enmLoaded = enmLoadSz;
5318	else if (pReNative->Core.aHstSimdRegs[idxHstSimdReg].enmLoaded != enmLoadSz)
5319	{
5320	Assert( ( pReNative->Core.aHstSimdRegs[idxHstSimdReg].enmLoaded == kIemNativeGstSimdRegLdStSz_Low128
5321	&& enmLoadSz == kIemNativeGstSimdRegLdStSz_High128)
5322	\|\| ( pReNative->Core.aHstSimdRegs[idxHstSimdReg].enmLoaded == kIemNativeGstSimdRegLdStSz_High128
5323	&& enmLoadSz == kIemNativeGstSimdRegLdStSz_Low128));
5324	pReNative->Core.aHstSimdRegs[idxHstSimdReg].enmLoaded = kIemNativeGstSimdRegLdStSz_256;
5325	}
5326	}
5327
5328
5329	static uint32_t iemNativeSimdRegAllocLoadVecRegFromVecRegSz(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint8_t idxHstSimdRegDst,
5330	uint8_t idxHstSimdRegSrc, IEMNATIVEGSTSIMDREGLDSTSZ enmLoadSzDst)
5331	{
5332	/* Easy case first, either the destination loads the same range as what the source has already loaded or the source has loaded everything. */
5333	if ( pReNative->Core.aHstSimdRegs[idxHstSimdRegSrc].enmLoaded == enmLoadSzDst
5334	\|\| pReNative->Core.aHstSimdRegs[idxHstSimdRegSrc].enmLoaded == kIemNativeGstSimdRegLdStSz_256)
5335	{
5336	# ifdef RT_ARCH_ARM64
5337	/* ASSUMES that there are two adjacent 128-bit registers available for the 256-bit value. */
5338	Assert(!(idxHstSimdRegDst & 0x1)); Assert(!(idxHstSimdRegSrc & 0x1));
5339	# endif
5340
5341	if (idxHstSimdRegDst != idxHstSimdRegSrc)
5342	{
5343	switch (enmLoadSzDst)
5344	{
5345	case kIemNativeGstSimdRegLdStSz_256:
5346	off = iemNativeEmitSimdLoadVecRegFromVecRegU256(pReNative, off, idxHstSimdRegDst, idxHstSimdRegSrc);
5347	break;
5348	case kIemNativeGstSimdRegLdStSz_Low128:
5349	off = iemNativeEmitSimdLoadVecRegFromVecRegU128(pReNative, off, idxHstSimdRegDst, idxHstSimdRegSrc);
5350	break;
5351	case kIemNativeGstSimdRegLdStSz_High128:
5352	off = iemNativeEmitSimdLoadVecRegFromVecRegU128(pReNative, off, idxHstSimdRegDst + 1, idxHstSimdRegSrc + 1);
5353	break;
5354	default:
5355	AssertFailedStmt(IEMNATIVE_DO_LONGJMP(pReNative, VERR_IPE_NOT_REACHED_DEFAULT_CASE));
5356	}
5357
5358	iemNativeSimdRegSetValidLoadFlag(pReNative, idxHstSimdRegDst, enmLoadSzDst);
5359	}
5360	}
5361	else
5362	{
5363	/* Complicated stuff where the source is currently missing something, later. */
5364	AssertFailedStmt(IEMNATIVE_DO_LONGJMP(pReNative, VERR_IPE_NOT_REACHED_DEFAULT_CASE));
5365	}
5366
5367	return off;
5368	}
5369
5370
5371	/**
5372	* Allocates a temporary host SIMD register for keeping a guest
5373	* SIMD register value.
5374	*
5375	* Since we may already have a register holding the guest register value,
5376	* code will be emitted to do the loading if that's not the case. Code may also
5377	* be emitted if we have to free up a register to satify the request.
5378	*
5379	* @returns The host register number; throws VBox status code on failure, so no
5380	* need to check the return value.
5381	* @param pReNative The native recompile state.
5382	* @param poff Pointer to the variable with the code buffer
5383	* position. This will be update if we need to move a
5384	* variable from register to stack in order to satisfy
5385	* the request.
5386	* @param enmGstSimdReg The guest SIMD register that will is to be updated.
5387	* @param enmIntendedUse How the caller will be using the host register.
5388	* @param fNoVolatileRegs Set if no volatile register allowed, clear if any
5389	* register is okay (default). The ASSUMPTION here is
5390	* that the caller has already flushed all volatile
5391	* registers, so this is only applied if we allocate a
5392	* new register.
5393	* @sa iemNativeRegAllocTmpForGuestRegIfAlreadyPresent
5394	*/
5395	DECL_HIDDEN_THROW(uint8_t)
5396	iemNativeSimdRegAllocTmpForGuestSimdReg(PIEMRECOMPILERSTATE pReNative, uint32_t *poff, IEMNATIVEGSTSIMDREG enmGstSimdReg,
5397	IEMNATIVEGSTSIMDREGLDSTSZ enmLoadSz, IEMNATIVEGSTREGUSE enmIntendedUse /= kIemNativeGstRegUse_ReadOnly/,
5398	bool fNoVolatileRegs /= false/)
5399	{
5400	Assert(enmGstSimdReg < kIemNativeGstSimdReg_End);
5401	#if defined(IEMNATIVE_WITH_LIVENESS_ANALYSIS) && 0 /** @todo r=aeichner */
5402	AssertMsg( pReNative->idxCurCall == 0
5403	\|\| (enmIntendedUse == kIemNativeGstRegUse_ForFullWrite
5404	? IEMLIVENESS_STATE_IS_CLOBBER_EXPECTED(iemNativeLivenessGetPrevStateByGstReg(pReNative, enmGstSimdReg))
5405	: enmIntendedUse == kIemNativeGstRegUse_ForUpdate
5406	? IEMLIVENESS_STATE_IS_MODIFY_EXPECTED( iemNativeLivenessGetPrevStateByGstReg(pReNative, enmGstSimdReg))
5407	: IEMLIVENESS_STATE_IS_INPUT_EXPECTED( iemNativeLivenessGetPrevStateByGstReg(pReNative, enmGstSimdReg)) ),
5408	("%s - %u\n", g_aGstSimdShadowInfo[enmGstSimdReg].pszName, iemNativeLivenessGetPrevStateByGstReg(pReNative, enmGstSimdReg)));
5409	#endif
5410	#if defined(LOG_ENABLED) \|\| defined(VBOX_STRICT)
5411	static const char * const s_pszIntendedUse[] = { "fetch", "update", "full write", "destructive calc" };
5412	#endif
5413	uint32_t const fRegMask = !fNoVolatileRegs
5414	? IEMNATIVE_HST_SIMD_REG_MASK & ~IEMNATIVE_SIMD_REG_FIXED_MASK
5415	: IEMNATIVE_HST_SIMD_REG_MASK & ~IEMNATIVE_SIMD_REG_FIXED_MASK & ~IEMNATIVE_CALL_VOLATILE_SIMD_REG_MASK;
5416
5417	/*
5418	* First check if the guest register value is already in a host register.
5419	*/
5420	if (pReNative->Core.bmGstSimdRegShadows & RT_BIT_64(enmGstSimdReg))
5421	{
5422	uint8_t idxSimdReg = pReNative->Core.aidxGstSimdRegShadows[enmGstSimdReg];
5423	Assert(idxSimdReg < RT_ELEMENTS(pReNative->Core.aHstSimdRegs));
5424	Assert(pReNative->Core.aHstSimdRegs[idxSimdReg].fGstRegShadows & RT_BIT_64(enmGstSimdReg));
5425	Assert(pReNative->Core.bmHstSimdRegsWithGstShadow & RT_BIT_32(idxSimdReg));
5426
5427	/* It's not supposed to be allocated... */
5428	if (!(pReNative->Core.bmHstSimdRegs & RT_BIT_32(idxSimdReg)))
5429	{
5430	/*
5431	* If the register will trash the guest shadow copy, try find a
5432	* completely unused register we can use instead. If that fails,
5433	* we need to disassociate the host reg from the guest reg.
5434	*/
5435	/** @todo would be nice to know if preserving the register is in any way helpful. */
5436	/* If the purpose is calculations, try duplicate the register value as
5437	we'll be clobbering the shadow. */
5438	if ( enmIntendedUse == kIemNativeGstRegUse_Calculation
5439	&& ( ~pReNative->Core.bmHstSimdRegs
5440	& ~pReNative->Core.bmHstSimdRegsWithGstShadow
5441	& (~IEMNATIVE_SIMD_REG_FIXED_MASK & IEMNATIVE_HST_SIMD_REG_MASK)))
5442	{
5443	uint8_t const idxRegNew = iemNativeSimdRegAllocTmpEx(pReNative, poff, fRegMask);
5444
5445	poff = iemNativeSimdRegAllocLoadVecRegFromVecRegSz(pReNative, poff, idxRegNew, idxSimdReg, enmLoadSz);
5446
5447	Log12(("iemNativeSimdRegAllocTmpForGuestSimdReg: Duplicated %s for guest %s into %s for destructive calc\n",
5448	g_apszIemNativeHstSimdRegNames[idxSimdReg], g_aGstSimdShadowInfo[enmGstSimdReg].pszName,
5449	g_apszIemNativeHstSimdRegNames[idxRegNew]));
5450	idxSimdReg = idxRegNew;
5451	}
5452	/* If the current register matches the restrictions, go ahead and allocate
5453	it for the caller. */
5454	else if (fRegMask & RT_BIT_32(idxSimdReg))
5455	{
5456	pReNative->Core.bmHstSimdRegs \|= RT_BIT_32(idxSimdReg);
5457	pReNative->Core.aHstSimdRegs[idxSimdReg].enmWhat = kIemNativeWhat_Tmp;
5458	if (enmIntendedUse != kIemNativeGstRegUse_Calculation)
5459	{
5460	if (enmIntendedUse != kIemNativeGstRegUse_ForFullWrite)
5461	poff = iemNativeSimdRegAllocLoadVecRegFromVecRegSz(pReNative, poff, idxSimdReg, idxSimdReg, enmLoadSz);
5462	else
5463	iemNativeSimdRegSetValidLoadFlag(pReNative, idxSimdReg, enmLoadSz);
5464	Log12(("iemNativeSimdRegAllocTmpForGuestSimdReg: Reusing %s for guest %s %s\n",
5465	g_apszIemNativeHstSimdRegNames[idxSimdReg], g_aGstSimdShadowInfo[enmGstSimdReg].pszName, s_pszIntendedUse[enmIntendedUse]));
5466	}
5467	else
5468	{
5469	iemNativeSimdRegClearGstSimdRegShadowing(pReNative, idxSimdReg, *poff);
5470	Log12(("iemNativeSimdRegAllocTmpForGuestSimdReg: Grabbing %s for guest %s - destructive calc\n",
5471	g_apszIemNativeHstSimdRegNames[idxSimdReg], g_aGstSimdShadowInfo[enmGstSimdReg].pszName));
5472	}
5473	}
5474	/* Otherwise, allocate a register that satisfies the caller and transfer
5475	the shadowing if compatible with the intended use. (This basically
5476	means the call wants a non-volatile register (RSP push/pop scenario).) */
5477	else
5478	{
5479	Assert(fNoVolatileRegs);
5480	uint8_t const idxRegNew = iemNativeSimdRegAllocTmpEx(pReNative, poff, fRegMask & ~RT_BIT_32(idxSimdReg),
5481	!fNoVolatileRegs
5482	&& enmIntendedUse == kIemNativeGstRegUse_Calculation);
5483	poff = iemNativeSimdRegAllocLoadVecRegFromVecRegSz(pReNative, poff, idxRegNew, idxSimdReg, enmLoadSz);
5484	if (enmIntendedUse != kIemNativeGstRegUse_Calculation)
5485	{
5486	iemNativeSimdRegTransferGstSimdRegShadowing(pReNative, idxSimdReg, idxRegNew, enmGstSimdReg, *poff);
5487	Log12(("iemNativeSimdRegAllocTmpForGuestSimdReg: Transfering %s to %s for guest %s %s\n",
5488	g_apszIemNativeHstSimdRegNames[idxSimdReg], g_apszIemNativeHstSimdRegNames[idxRegNew],
5489	g_aGstSimdShadowInfo[enmGstSimdReg].pszName, s_pszIntendedUse[enmIntendedUse]));
5490	}
5491	else
5492	Log12(("iemNativeSimdRegAllocTmpForGuestSimdReg: Duplicated %s for guest %s into %s for destructive calc\n",
5493	g_apszIemNativeHstSimdRegNames[idxSimdReg], g_aGstSimdShadowInfo[enmGstSimdReg].pszName,
5494	g_apszIemNativeHstSimdRegNames[idxRegNew]));
5495	idxSimdReg = idxRegNew;
5496	}
5497	}
5498	else
5499	{
5500	/*
5501	* Oops. Shadowed guest register already allocated!
5502	*
5503	* Allocate a new register, copy the value and, if updating, the
5504	* guest shadow copy assignment to the new register.
5505	*/
5506	AssertMsg( enmIntendedUse != kIemNativeGstRegUse_ForUpdate
5507	&& enmIntendedUse != kIemNativeGstRegUse_ForFullWrite,
5508	("This shouldn't happen: idxSimdReg=%d enmGstSimdReg=%d enmIntendedUse=%s\n",
5509	idxSimdReg, enmGstSimdReg, s_pszIntendedUse[enmIntendedUse]));
5510
5511	/** @todo share register for readonly access. */
5512	uint8_t const idxRegNew = iemNativeSimdRegAllocTmpEx(pReNative, poff, fRegMask,
5513	enmIntendedUse == kIemNativeGstRegUse_Calculation);
5514
5515	if (enmIntendedUse != kIemNativeGstRegUse_ForFullWrite)
5516	poff = iemNativeSimdRegAllocLoadVecRegFromVecRegSz(pReNative, poff, idxRegNew, idxSimdReg, enmLoadSz);
5517	else
5518	iemNativeSimdRegSetValidLoadFlag(pReNative, idxRegNew, enmLoadSz);
5519
5520	if ( enmIntendedUse != kIemNativeGstRegUse_ForUpdate
5521	&& enmIntendedUse != kIemNativeGstRegUse_ForFullWrite)
5522	Log12(("iemNativeSimdRegAllocTmpForGuestSimdReg: Duplicated %s for guest %s into %s for %s\n",
5523	g_apszIemNativeHstSimdRegNames[idxSimdReg], g_aGstSimdShadowInfo[enmGstSimdReg].pszName,
5524	g_apszIemNativeHstSimdRegNames[idxRegNew], s_pszIntendedUse[enmIntendedUse]));
5525	else
5526	{
5527	iemNativeSimdRegTransferGstSimdRegShadowing(pReNative, idxSimdReg, idxRegNew, enmGstSimdReg, *poff);
5528	Log12(("iemNativeSimdRegAllocTmpForGuestSimdReg: Moved %s for guest %s into %s for %s\n",
5529	g_apszIemNativeHstSimdRegNames[idxSimdReg], g_aGstSimdShadowInfo[enmGstSimdReg].pszName,
5530	g_apszIemNativeHstSimdRegNames[idxRegNew], s_pszIntendedUse[enmIntendedUse]));
5531	}
5532	idxSimdReg = idxRegNew;
5533	}
5534	Assert(RT_BIT_32(idxSimdReg) & fRegMask); /* See assumption in fNoVolatileRegs docs. */
5535
5536	#ifdef VBOX_STRICT
5537	/* Strict builds: Check that the value is correct. */
5538	if (enmIntendedUse != kIemNativeGstRegUse_ForFullWrite)
5539	poff = iemNativeEmitGuestSimdRegValueCheck(pReNative, poff, idxSimdReg, enmGstSimdReg, enmLoadSz);
5540	#endif
5541
5542	return idxSimdReg;
5543	}
5544
5545	/*
5546	* Allocate a new register, load it with the guest value and designate it as a copy of the
5547	*/
5548	uint8_t const idxRegNew = iemNativeSimdRegAllocTmpEx(pReNative, poff, fRegMask, enmIntendedUse == kIemNativeGstRegUse_Calculation);
5549
5550	if (enmIntendedUse != kIemNativeGstRegUse_ForFullWrite)
5551	poff = iemNativeEmitLoadSimdRegWithGstShadowSimdReg(pReNative, poff, idxRegNew, enmGstSimdReg, enmLoadSz);
5552	else
5553	iemNativeSimdRegSetValidLoadFlag(pReNative, idxRegNew, enmLoadSz);
5554
5555	if (enmIntendedUse != kIemNativeGstRegUse_Calculation)
5556	iemNativeSimdRegMarkAsGstSimdRegShadow(pReNative, idxRegNew, enmGstSimdReg, *poff);
5557
5558	Log12(("iemNativeRegAllocTmpForGuestSimdReg: Allocated %s for guest %s %s\n",
5559	g_apszIemNativeHstSimdRegNames[idxRegNew], g_aGstSimdShadowInfo[enmGstSimdReg].pszName, s_pszIntendedUse[enmIntendedUse]));
5560
5561	return idxRegNew;
5562	}
5563
5564	#endif /* IEMNATIVE_WITH_SIMD_REG_ALLOCATOR */
5565
5566
5567
5568	/*********************************************************************************************************************************
5569	* Code emitters for flushing pending guest register writes and sanity checks *
5570	*********************************************************************************************************************************/
5571
5572	#ifdef VBOX_STRICT
5573	/**
5574	* Does internal register allocator sanity checks.
5575	*/
5576	DECLHIDDEN(void) iemNativeRegAssertSanity(PIEMRECOMPILERSTATE pReNative)
5577	{
5578	/*
5579	* Iterate host registers building a guest shadowing set.
5580	*/
5581	uint64_t bmGstRegShadows = 0;
5582	uint32_t bmHstRegsWithGstShadow = pReNative->Core.bmHstRegsWithGstShadow;
5583	AssertMsg(!(bmHstRegsWithGstShadow & IEMNATIVE_REG_FIXED_MASK), ("%#RX32\n", bmHstRegsWithGstShadow));
5584	while (bmHstRegsWithGstShadow)
5585	{
5586	unsigned const idxHstReg = ASMBitFirstSetU32(bmHstRegsWithGstShadow) - 1;
5587	Assert(idxHstReg < RT_ELEMENTS(pReNative->Core.aHstRegs));
5588	bmHstRegsWithGstShadow &= ~RT_BIT_32(idxHstReg);
5589
5590	uint64_t fThisGstRegShadows = pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows;
5591	AssertMsg(fThisGstRegShadows != 0, ("idxHstReg=%d\n", idxHstReg));
5592	AssertMsg(fThisGstRegShadows < RT_BIT_64(kIemNativeGstReg_End), ("idxHstReg=%d %#RX64\n", idxHstReg, fThisGstRegShadows));
5593	bmGstRegShadows \|= fThisGstRegShadows;
5594	while (fThisGstRegShadows)
5595	{
5596	unsigned const idxGstReg = ASMBitFirstSetU64(fThisGstRegShadows) - 1;
5597	fThisGstRegShadows &= ~RT_BIT_64(idxGstReg);
5598	AssertMsg(pReNative->Core.aidxGstRegShadows[idxGstReg] == idxHstReg,
5599	("idxHstReg=%d aidxGstRegShadows[idxGstReg=%d]=%d\n",
5600	idxHstReg, idxGstReg, pReNative->Core.aidxGstRegShadows[idxGstReg]));
5601	}
5602	}
5603	AssertMsg(bmGstRegShadows == pReNative->Core.bmGstRegShadows,
5604	("%RX64 vs %RX64; diff %RX64\n", bmGstRegShadows, pReNative->Core.bmGstRegShadows,
5605	bmGstRegShadows ^ pReNative->Core.bmGstRegShadows));
5606
5607	/*
5608	* Now the other way around, checking the guest to host index array.
5609	*/
5610	bmHstRegsWithGstShadow = 0;
5611	bmGstRegShadows = pReNative->Core.bmGstRegShadows;
5612	Assert(bmGstRegShadows < RT_BIT_64(kIemNativeGstReg_End));
5613	while (bmGstRegShadows)
5614	{
5615	unsigned const idxGstReg = ASMBitFirstSetU64(bmGstRegShadows) - 1;
5616	Assert(idxGstReg < RT_ELEMENTS(pReNative->Core.aidxGstRegShadows));
5617	bmGstRegShadows &= ~RT_BIT_64(idxGstReg);
5618
5619	uint8_t const idxHstReg = pReNative->Core.aidxGstRegShadows[idxGstReg];
5620	AssertMsg(idxHstReg < RT_ELEMENTS(pReNative->Core.aHstRegs), ("aidxGstRegShadows[%d]=%d\n", idxGstReg, idxHstReg));
5621	AssertMsg(pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows & RT_BIT_64(idxGstReg),
5622	("idxGstReg=%d idxHstReg=%d fGstRegShadows=%RX64\n",
5623	idxGstReg, idxHstReg, pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows));
5624	bmHstRegsWithGstShadow \|= RT_BIT_32(idxHstReg);
5625	}
5626	AssertMsg(bmHstRegsWithGstShadow == pReNative->Core.bmHstRegsWithGstShadow,
5627	("%RX64 vs %RX64; diff %RX64\n", bmHstRegsWithGstShadow, pReNative->Core.bmHstRegsWithGstShadow,
5628	bmHstRegsWithGstShadow ^ pReNative->Core.bmHstRegsWithGstShadow));
5629	}
5630	#endif /* VBOX_STRICT */
5631
5632
5633	/**
5634	* Flushes any delayed guest register writes.
5635	*
5636	* This must be called prior to calling CImpl functions and any helpers that use
5637	* the guest state (like raising exceptions) and such.
5638	*
5639	* This optimization has not yet been implemented. The first target would be
5640	* RIP updates, since these are the most common ones.
5641	*/
5642	DECL_HIDDEN_THROW(uint32_t)
5643	iemNativeRegFlushPendingWritesSlow(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint64_t fGstShwExcept, bool fFlushShadows)
5644	{
5645	#ifdef IEMNATIVE_WITH_DELAYED_PC_UPDATING
5646	if (!(fGstShwExcept & kIemNativeGstReg_Pc))
5647	off = iemNativeEmitPcWriteback(pReNative, off);
5648	#else
5649	RT_NOREF(pReNative, fGstShwExcept);
5650	#endif
5651
5652	#ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
5653	/** @todo r=bird: There must be a quicker way to check if anything needs
5654	* doing and then call simd function to do the flushing */
5655	/** @todo This doesn't mix well with fGstShwExcept but we ignore this for now and just flush everything. */
5656	for (uint8_t idxGstSimdReg = 0; idxGstSimdReg < RT_ELEMENTS(g_aGstSimdShadowInfo); idxGstSimdReg++)
5657	{
5658	Assert( (pReNative->Core.bmGstSimdRegShadows & RT_BIT_64(idxGstSimdReg)
5659	\|\| !IEMNATIVE_SIMD_REG_STATE_IS_DIRTY_U256(pReNative, idxGstSimdReg)));
5660
5661	if (IEMNATIVE_SIMD_REG_STATE_IS_DIRTY_U256(pReNative, idxGstSimdReg))
5662	off = iemNativeSimdRegFlushPendingWrite(pReNative, off, IEMNATIVEGSTSIMDREG_SIMD(idxGstSimdReg));
5663
5664	if ( fFlushShadows
5665	&& pReNative->Core.bmGstSimdRegShadows & RT_BIT_64(idxGstSimdReg))
5666	{
5667	uint8_t const idxHstSimdReg = pReNative->Core.aidxGstSimdRegShadows[idxGstSimdReg];
5668
5669	iemNativeSimdRegClearGstSimdRegShadowing(pReNative, idxHstSimdReg, off);
5670	iemNativeSimdRegFlushGuestShadows(pReNative, RT_BIT_64(IEMNATIVEGSTSIMDREG_SIMD(idxGstSimdReg)));
5671	}
5672	}
5673	#else
5674	RT_NOREF(pReNative, fGstShwExcept, fFlushShadows);
5675	#endif
5676
5677	return off;
5678	}
5679
5680
5681	#ifdef IEMNATIVE_WITH_DELAYED_PC_UPDATING
5682	/**
5683	* Emits code to update the guest RIP value by adding the current offset since the start of the last RIP update.
5684	*/
5685	DECL_HIDDEN_THROW(uint32_t) iemNativeEmitPcWritebackSlow(PIEMRECOMPILERSTATE pReNative, uint32_t off)
5686	{
5687	Assert(pReNative->Core.offPc);
5688	# ifdef IEMNATIVE_WITH_TB_DEBUG_INFO
5689	iemNativeDbgInfoAddNativeOffset(pReNative, off);
5690	iemNativeDbgInfoAddDelayedPcUpdate(pReNative, pReNative->Core.offPc, pReNative->Core.cInstrPcUpdateSkipped);
5691	# endif
5692
5693	# ifndef IEMNATIVE_REG_FIXED_PC_DBG
5694	/* Allocate a temporary PC register. */
5695	uint8_t const idxPcReg = iemNativeRegAllocTmpForGuestReg(pReNative, &off, kIemNativeGstReg_Pc, kIemNativeGstRegUse_ForUpdate);
5696
5697	/* Perform the addition and store the result. */
5698	off = iemNativeEmitAddGprImm(pReNative, off, idxPcReg, pReNative->Core.offPc);
5699	off = iemNativeEmitStoreGprToVCpuU64(pReNative, off, idxPcReg, RT_UOFFSETOF(VMCPU, cpum.GstCtx.rip));
5700
5701	/* Free but don't flush the PC register. */
5702	iemNativeRegFreeTmp(pReNative, idxPcReg);
5703	# else
5704	/* Compare the shadow with the context value, they should match. */
5705	off = iemNativeEmitAddGprImm(pReNative, off, IEMNATIVE_REG_FIXED_PC_DBG, pReNative->Core.offPc);
5706	off = iemNativeEmitGuestRegValueCheck(pReNative, off, IEMNATIVE_REG_FIXED_PC_DBG, kIemNativeGstReg_Pc);
5707	# endif
5708
5709	STAM_COUNTER_ADD(&pReNative->pVCpu->iem.s.StatNativePcUpdateDelayed, pReNative->Core.cInstrPcUpdateSkipped);
5710	pReNative->Core.offPc = 0;
5711	pReNative->Core.cInstrPcUpdateSkipped = 0;
5712
5713	return off;
5714	}
5715	#endif /* IEMNATIVE_WITH_DELAYED_PC_UPDATING */
5716
5717
5718	/*********************************************************************************************************************************
5719	* Code Emitters (larger snippets) *
5720	*********************************************************************************************************************************/
5721
5722	/**
5723	* Loads the guest shadow register @a enmGstReg into host reg @a idxHstReg, zero
5724	* extending to 64-bit width.
5725	*
5726	* @returns New code buffer offset on success, UINT32_MAX on failure.
5727	* @param pReNative .
5728	* @param off The current code buffer position.
5729	* @param idxHstReg The host register to load the guest register value into.
5730	* @param enmGstReg The guest register to load.
5731	*
5732	* @note This does not mark @a idxHstReg as having a shadow copy of @a enmGstReg,
5733	* that is something the caller needs to do if applicable.
5734	*/
5735	DECL_HIDDEN_THROW(uint32_t)
5736	iemNativeEmitLoadGprWithGstShadowReg(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint8_t idxHstReg, IEMNATIVEGSTREG enmGstReg)
5737	{
5738	Assert((unsigned)enmGstReg < (unsigned)kIemNativeGstReg_End);
5739	Assert(g_aGstShadowInfo[enmGstReg].cb != 0);
5740
5741	switch (g_aGstShadowInfo[enmGstReg].cb)
5742	{
5743	case sizeof(uint64_t):
5744	return iemNativeEmitLoadGprFromVCpuU64(pReNative, off, idxHstReg, g_aGstShadowInfo[enmGstReg].off);
5745	case sizeof(uint32_t):
5746	return iemNativeEmitLoadGprFromVCpuU32(pReNative, off, idxHstReg, g_aGstShadowInfo[enmGstReg].off);
5747	case sizeof(uint16_t):
5748	return iemNativeEmitLoadGprFromVCpuU16(pReNative, off, idxHstReg, g_aGstShadowInfo[enmGstReg].off);
5749	#if 0 /* not present in the table. */
5750	case sizeof(uint8_t):
5751	return iemNativeEmitLoadGprFromVCpuU8(pReNative, off, idxHstReg, g_aGstShadowInfo[enmGstReg].off);
5752	#endif
5753	default:
5754	AssertFailedStmt(IEMNATIVE_DO_LONGJMP(pReNative, VERR_IPE_NOT_REACHED_DEFAULT_CASE));
5755	}
5756	}
5757
5758
5759	#ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
5760	/**
5761	* Loads the guest shadow SIMD register @a enmGstSimdReg into host SIMD reg @a idxHstSimdReg.
5762	*
5763	* @returns New code buffer offset on success, UINT32_MAX on failure.
5764	* @param pReNative The recompiler state.
5765	* @param off The current code buffer position.
5766	* @param idxHstSimdReg The host register to load the guest register value into.
5767	* @param enmGstSimdReg The guest register to load.
5768	* @param enmLoadSz The load size of the register.
5769	*
5770	* @note This does not mark @a idxHstReg as having a shadow copy of @a enmGstReg,
5771	* that is something the caller needs to do if applicable.
5772	*/
5773	DECL_HIDDEN_THROW(uint32_t)
5774	iemNativeEmitLoadSimdRegWithGstShadowSimdReg(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint8_t idxHstSimdReg,
5775	IEMNATIVEGSTSIMDREG enmGstSimdReg, IEMNATIVEGSTSIMDREGLDSTSZ enmLoadSz)
5776	{
5777	Assert((unsigned)enmGstSimdReg < RT_ELEMENTS(g_aGstSimdShadowInfo));
5778
5779	iemNativeSimdRegSetValidLoadFlag(pReNative, idxHstSimdReg, enmLoadSz);
5780	switch (enmLoadSz)
5781	{
5782	case kIemNativeGstSimdRegLdStSz_256:
5783	off = iemNativeEmitSimdLoadVecRegFromVCpuLowU128(pReNative, off, idxHstSimdReg, g_aGstSimdShadowInfo[enmGstSimdReg].offXmm);
5784	return iemNativeEmitSimdLoadVecRegFromVCpuHighU128(pReNative, off, idxHstSimdReg, g_aGstSimdShadowInfo[enmGstSimdReg].offYmm);
5785	case kIemNativeGstSimdRegLdStSz_Low128:
5786	return iemNativeEmitSimdLoadVecRegFromVCpuLowU128(pReNative, off, idxHstSimdReg, g_aGstSimdShadowInfo[enmGstSimdReg].offXmm);
5787	case kIemNativeGstSimdRegLdStSz_High128:
5788	return iemNativeEmitSimdLoadVecRegFromVCpuHighU128(pReNative, off, idxHstSimdReg, g_aGstSimdShadowInfo[enmGstSimdReg].offYmm);
5789	default:
5790	AssertFailedStmt(IEMNATIVE_DO_LONGJMP(pReNative, VERR_IPE_NOT_REACHED_DEFAULT_CASE));
5791	}
5792	}
5793	#endif /* IEMNATIVE_WITH_SIMD_REG_ALLOCATOR */
5794
5795	#ifdef VBOX_STRICT
5796
5797	/**
5798	* Emitting code that checks that the value of @a idxReg is UINT32_MAX or less.
5799	*
5800	* @note May of course trash IEMNATIVE_REG_FIXED_TMP0.
5801	* Trashes EFLAGS on AMD64.
5802	*/
5803	DECL_HIDDEN_THROW(uint32_t)
5804	iemNativeEmitTop32BitsClearCheck(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint8_t idxReg)
5805	{
5806	# ifdef RT_ARCH_AMD64
5807	uint8_t * const pbCodeBuf = iemNativeInstrBufEnsure(pReNative, off, 20);
5808
5809	/* rol reg64, 32 */
5810	pbCodeBuf[off++] = X86_OP_REX_W \| (idxReg < 8 ? 0 : X86_OP_REX_B);
5811	pbCodeBuf[off++] = 0xc1;
5812	pbCodeBuf[off++] = X86_MODRM_MAKE(X86_MOD_REG, 0, idxReg & 7);
5813	pbCodeBuf[off++] = 32;
5814
5815	/* test reg32, ffffffffh */
5816	if (idxReg >= 8)
5817	pbCodeBuf[off++] = X86_OP_REX_B;
5818	pbCodeBuf[off++] = 0xf7;
5819	pbCodeBuf[off++] = X86_MODRM_MAKE(X86_MOD_REG, 0, idxReg & 7);
5820	pbCodeBuf[off++] = 0xff;
5821	pbCodeBuf[off++] = 0xff;
5822	pbCodeBuf[off++] = 0xff;
5823	pbCodeBuf[off++] = 0xff;
5824
5825	/* je/jz +1 */
5826	pbCodeBuf[off++] = 0x74;
5827	pbCodeBuf[off++] = 0x01;
5828
5829	/* int3 */
5830	pbCodeBuf[off++] = 0xcc;
5831
5832	/* rol reg64, 32 */
5833	pbCodeBuf[off++] = X86_OP_REX_W \| (idxReg < 8 ? 0 : X86_OP_REX_B);
5834	pbCodeBuf[off++] = 0xc1;
5835	pbCodeBuf[off++] = X86_MODRM_MAKE(X86_MOD_REG, 0, idxReg & 7);
5836	pbCodeBuf[off++] = 32;
5837
5838	# elif defined(RT_ARCH_ARM64)
5839	uint32_t * const pu32CodeBuf = iemNativeInstrBufEnsure(pReNative, off, 3);
5840	/* lsr tmp0, reg64, #32 */
5841	pu32CodeBuf[off++] = Armv8A64MkInstrLsrImm(IEMNATIVE_REG_FIXED_TMP0, idxReg, 32);
5842	/* cbz tmp0, +1 */
5843	pu32CodeBuf[off++] = Armv8A64MkInstrCbzCbnz(false /fJmpIfNotZero/, 2, IEMNATIVE_REG_FIXED_TMP0);
5844	/* brk #0x1100 */
5845	pu32CodeBuf[off++] = Armv8A64MkInstrBrk(UINT32_C(0x1100));
5846
5847	# else
5848	# error "Port me!"
5849	# endif
5850	IEMNATIVE_ASSERT_INSTR_BUF_ENSURE(pReNative, off);
5851	return off;
5852	}
5853
5854
5855	/**
5856	* Emitting code that checks that the content of register @a idxReg is the same
5857	* as what's in the guest register @a enmGstReg, resulting in a breakpoint
5858	* instruction if that's not the case.
5859	*
5860	* @note May of course trash IEMNATIVE_REG_FIXED_TMP0.
5861	* Trashes EFLAGS on AMD64.
5862	*/
5863	DECL_HIDDEN_THROW(uint32_t)
5864	iemNativeEmitGuestRegValueCheck(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint8_t idxReg, IEMNATIVEGSTREG enmGstReg)
5865	{
5866	# ifdef RT_ARCH_AMD64
5867	uint8_t * const pbCodeBuf = iemNativeInstrBufEnsure(pReNative, off, 32);
5868
5869	/* cmp reg, [mem] */
5870	if (g_aGstShadowInfo[enmGstReg].cb == sizeof(uint8_t))
5871	{
5872	if (idxReg >= 8)
5873	pbCodeBuf[off++] = X86_OP_REX_R;
5874	pbCodeBuf[off++] = 0x38;
5875	}
5876	else
5877	{
5878	if (g_aGstShadowInfo[enmGstReg].cb == sizeof(uint64_t))
5879	pbCodeBuf[off++] = X86_OP_REX_W \| (idxReg < 8 ? 0 : X86_OP_REX_R);
5880	else
5881	{
5882	if (g_aGstShadowInfo[enmGstReg].cb == sizeof(uint16_t))
5883	pbCodeBuf[off++] = X86_OP_PRF_SIZE_OP;
5884	else
5885	AssertStmt(g_aGstShadowInfo[enmGstReg].cb == sizeof(uint32_t),
5886	IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_LABEL_IPE_6));
5887	if (idxReg >= 8)
5888	pbCodeBuf[off++] = X86_OP_REX_R;
5889	}
5890	pbCodeBuf[off++] = 0x39;
5891	}
5892	off = iemNativeEmitGprByVCpuDisp(pbCodeBuf, off, idxReg, g_aGstShadowInfo[enmGstReg].off);
5893
5894	/* je/jz +1 */
5895	pbCodeBuf[off++] = 0x74;
5896	pbCodeBuf[off++] = 0x01;
5897
5898	/* int3 */
5899	pbCodeBuf[off++] = 0xcc;
5900
5901	/* For values smaller than the register size, we must check that the rest
5902	of the register is all zeros. */
5903	if (g_aGstShadowInfo[enmGstReg].cb < sizeof(uint32_t))
5904	{
5905	/* test reg64, imm32 */
5906	pbCodeBuf[off++] = X86_OP_REX_W \| (idxReg < 8 ? 0 : X86_OP_REX_B);
5907	pbCodeBuf[off++] = 0xf7;
5908	pbCodeBuf[off++] = X86_MODRM_MAKE(X86_MOD_REG, 0, idxReg & 7);
5909	pbCodeBuf[off++] = 0;
5910	pbCodeBuf[off++] = g_aGstShadowInfo[enmGstReg].cb > sizeof(uint8_t) ? 0 : 0xff;
5911	pbCodeBuf[off++] = 0xff;
5912	pbCodeBuf[off++] = 0xff;
5913
5914	/* je/jz +1 */
5915	pbCodeBuf[off++] = 0x74;
5916	pbCodeBuf[off++] = 0x01;
5917
5918	/* int3 */
5919	pbCodeBuf[off++] = 0xcc;
5920	IEMNATIVE_ASSERT_INSTR_BUF_ENSURE(pReNative, off);
5921	}
5922	else
5923	{
5924	IEMNATIVE_ASSERT_INSTR_BUF_ENSURE(pReNative, off);
5925	if (g_aGstShadowInfo[enmGstReg].cb == sizeof(uint32_t))
5926	iemNativeEmitTop32BitsClearCheck(pReNative, off, idxReg);
5927	}
5928
5929	# elif defined(RT_ARCH_ARM64)
5930	/* mov TMP0, [gstreg] */
5931	off = iemNativeEmitLoadGprWithGstShadowReg(pReNative, off, IEMNATIVE_REG_FIXED_TMP0, enmGstReg);
5932
5933	uint32_t * const pu32CodeBuf = iemNativeInstrBufEnsure(pReNative, off, 3);
5934	/* sub tmp0, tmp0, idxReg */
5935	pu32CodeBuf[off++] = Armv8A64MkInstrAddSubReg(true /fSub/, IEMNATIVE_REG_FIXED_TMP0, IEMNATIVE_REG_FIXED_TMP0, idxReg);
5936	/* cbz tmp0, +1 */
5937	pu32CodeBuf[off++] = Armv8A64MkInstrCbzCbnz(false /fJmpIfNotZero/, 2, IEMNATIVE_REG_FIXED_TMP0);
5938	/* brk #0x1000+enmGstReg */
5939	pu32CodeBuf[off++] = Armv8A64MkInstrBrk((uint32_t)enmGstReg \| UINT32_C(0x1000));
5940	IEMNATIVE_ASSERT_INSTR_BUF_ENSURE(pReNative, off);
5941
5942	# else
5943	# error "Port me!"
5944	# endif
5945	return off;
5946	}
5947
5948
5949	# ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
5950	/**
5951	* Emitting code that checks that the content of SIMD register @a idxSimdReg is the same
5952	* as what's in the guest register @a enmGstSimdReg, resulting in a breakpoint
5953	* instruction if that's not the case.
5954	*
5955	* @note May of course trash IEMNATIVE_SIMD_REG_FIXED_TMP0 and IEMNATIVE_REG_FIXED_TMP0.
5956	* Trashes EFLAGS on AMD64.
5957	*/
5958	DECL_HIDDEN_THROW(uint32_t)
5959	iemNativeEmitGuestSimdRegValueCheck(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint8_t idxSimdReg,
5960	IEMNATIVEGSTSIMDREG enmGstSimdReg, IEMNATIVEGSTSIMDREGLDSTSZ enmLoadSz)
5961	{
5962	/* We can't check the value against whats in CPUMCTX if the register is already marked as dirty, so skip the check. */
5963	if ( ( enmLoadSz == kIemNativeGstSimdRegLdStSz_256
5964	&& ( IEMNATIVE_SIMD_REG_STATE_IS_DIRTY_LO_U128(pReNative, enmGstSimdReg)
5965	\|\| IEMNATIVE_SIMD_REG_STATE_IS_DIRTY_HI_U128(pReNative, enmGstSimdReg)))
5966	\|\| ( enmLoadSz == kIemNativeGstSimdRegLdStSz_Low128
5967	&& IEMNATIVE_SIMD_REG_STATE_IS_DIRTY_LO_U128(pReNative, enmGstSimdReg))
5968	\|\| ( enmLoadSz == kIemNativeGstSimdRegLdStSz_High128
5969	&& IEMNATIVE_SIMD_REG_STATE_IS_DIRTY_HI_U128(pReNative, enmGstSimdReg)))
5970	return off;
5971
5972	# ifdef RT_ARCH_AMD64
5973	Assert(enmLoadSz == kIemNativeGstSimdRegLdStSz_Low128); /** @todo 256-bit variant. */
5974
5975	/* movdqa vectmp0, idxSimdReg */
5976	off = iemNativeEmitSimdLoadVecRegFromVecRegU128(pReNative, off, IEMNATIVE_SIMD_REG_FIXED_TMP0, idxSimdReg);
5977
5978	uint8_t * const pbCodeBuf = iemNativeInstrBufEnsure(pReNative, off, 44);
5979
5980	/* pcmpeqq vectmp0, [gstreg] (ASSUMES SSE4.1) */
5981	pbCodeBuf[off++] = X86_OP_PRF_SIZE_OP;
5982	if (idxSimdReg >= 8)
5983	pbCodeBuf[off++] = X86_OP_REX_R;
5984	pbCodeBuf[off++] = 0x0f;
5985	pbCodeBuf[off++] = 0x38;
5986	pbCodeBuf[off++] = 0x29;
5987	off = iemNativeEmitGprByVCpuDisp(pbCodeBuf, off, idxSimdReg, g_aGstSimdShadowInfo[enmGstSimdReg].offXmm);
5988
5989	/* pextrq tmp0, vectmp0, #0 (ASSUMES SSE4.1). */
5990	pbCodeBuf[off++] = X86_OP_PRF_SIZE_OP;
5991	pbCodeBuf[off++] = X86_OP_REX_W
5992	\| (IEMNATIVE_SIMD_REG_FIXED_TMP0 < 8 ? 0 : X86_OP_REX_R)
5993	\| (IEMNATIVE_REG_FIXED_TMP0 < 8 ? 0 : X86_OP_REX_B);
5994	pbCodeBuf[off++] = 0x0f;
5995	pbCodeBuf[off++] = 0x3a;
5996	pbCodeBuf[off++] = 0x16;
5997	pbCodeBuf[off++] = 0xeb;
5998	pbCodeBuf[off++] = 0x00;
5999
6000	/* test tmp0, 0xffffffff. */
6001	pbCodeBuf[off++] = X86_OP_REX_W \| (IEMNATIVE_REG_FIXED_TMP0 < 8 ? 0 : X86_OP_REX_B);
6002	pbCodeBuf[off++] = 0xf7;
6003	pbCodeBuf[off++] = X86_MODRM_MAKE(X86_MOD_REG, 0, IEMNATIVE_REG_FIXED_TMP0 & 7);
6004	pbCodeBuf[off++] = 0xff;
6005	pbCodeBuf[off++] = 0xff;
6006	pbCodeBuf[off++] = 0xff;
6007	pbCodeBuf[off++] = 0xff;
6008
6009	/* je/jz +1 */
6010	pbCodeBuf[off++] = 0x74;
6011	pbCodeBuf[off++] = 0x01;
6012
6013	/* int3 */
6014	pbCodeBuf[off++] = 0xcc;
6015
6016	/* pextrq tmp0, vectmp0, #1 (ASSUMES SSE4.1). */
6017	pbCodeBuf[off++] = X86_OP_PRF_SIZE_OP;
6018	pbCodeBuf[off++] = X86_OP_REX_W
6019	\| (IEMNATIVE_SIMD_REG_FIXED_TMP0 < 8 ? 0 : X86_OP_REX_R)
6020	\| (IEMNATIVE_REG_FIXED_TMP0 < 8 ? 0 : X86_OP_REX_B);
6021	pbCodeBuf[off++] = 0x0f;
6022	pbCodeBuf[off++] = 0x3a;
6023	pbCodeBuf[off++] = 0x16;
6024	pbCodeBuf[off++] = 0xeb;
6025	pbCodeBuf[off++] = 0x01;
6026
6027	/* test tmp0, 0xffffffff. */
6028	pbCodeBuf[off++] = X86_OP_REX_W \| (IEMNATIVE_REG_FIXED_TMP0 < 8 ? 0 : X86_OP_REX_B);
6029	pbCodeBuf[off++] = 0xf7;
6030	pbCodeBuf[off++] = X86_MODRM_MAKE(X86_MOD_REG, 0, IEMNATIVE_REG_FIXED_TMP0 & 7);
6031	pbCodeBuf[off++] = 0xff;
6032	pbCodeBuf[off++] = 0xff;
6033	pbCodeBuf[off++] = 0xff;
6034	pbCodeBuf[off++] = 0xff;
6035
6036	/* je/jz +1 */
6037	pbCodeBuf[off++] = 0x74;
6038	pbCodeBuf[off++] = 0x01;
6039
6040	/* int3 */
6041	pbCodeBuf[off++] = 0xcc;
6042
6043	# elif defined(RT_ARCH_ARM64)
6044	/* mov vectmp0, [gstreg] */
6045	off = iemNativeEmitLoadSimdRegWithGstShadowSimdReg(pReNative, off, IEMNATIVE_SIMD_REG_FIXED_TMP0, enmGstSimdReg, enmLoadSz);
6046
6047	if (enmLoadSz == kIemNativeGstSimdRegLdStSz_Low128 \|\| enmLoadSz == kIemNativeGstSimdRegLdStSz_256)
6048	{
6049	uint32_t * const pu32CodeBuf = iemNativeInstrBufEnsure(pReNative, off, 5);
6050	/* eor vectmp0, vectmp0, idxSimdReg */
6051	pu32CodeBuf[off++] = Armv8A64MkVecInstrEor(IEMNATIVE_SIMD_REG_FIXED_TMP0, IEMNATIVE_SIMD_REG_FIXED_TMP0, idxSimdReg);
6052	/* cnt vectmp0, vectmp0, #0*/
6053	pu32CodeBuf[off++] = Armv8A64MkVecInstrCnt(IEMNATIVE_SIMD_REG_FIXED_TMP0, IEMNATIVE_SIMD_REG_FIXED_TMP0);
6054	/* umov tmp0, vectmp0.D[0] */
6055	pu32CodeBuf[off++] = Armv8A64MkVecInstrUmov(IEMNATIVE_REG_FIXED_TMP0, IEMNATIVE_SIMD_REG_FIXED_TMP0,
6056	0 /idxElem/, kArmv8InstrUmovInsSz_U64);
6057	/* cbz tmp0, +1 */
6058	pu32CodeBuf[off++] = Armv8A64MkInstrCbzCbnz(false /fJmpIfNotZero/, 2, IEMNATIVE_REG_FIXED_TMP0);
6059	/* brk #0x1000+enmGstReg */
6060	pu32CodeBuf[off++] = Armv8A64MkInstrBrk((uint32_t)enmGstSimdReg \| UINT32_C(0x1000));
6061	}
6062
6063	if (enmLoadSz == kIemNativeGstSimdRegLdStSz_High128 \|\| enmLoadSz == kIemNativeGstSimdRegLdStSz_256)
6064	{
6065	uint32_t * const pu32CodeBuf = iemNativeInstrBufEnsure(pReNative, off, 5);
6066	/* eor vectmp0 + 1, vectmp0 + 1, idxSimdReg */
6067	pu32CodeBuf[off++] = Armv8A64MkVecInstrEor(IEMNATIVE_SIMD_REG_FIXED_TMP0 + 1, IEMNATIVE_SIMD_REG_FIXED_TMP0 + 1, idxSimdReg);
6068	/* cnt vectmp0 + 1, vectmp0 + 1, #0*/
6069	pu32CodeBuf[off++] = Armv8A64MkVecInstrCnt(IEMNATIVE_SIMD_REG_FIXED_TMP0 + 1, IEMNATIVE_SIMD_REG_FIXED_TMP0 + 1);
6070	/* umov tmp0, (vectmp0 + 1).D[0] */
6071	pu32CodeBuf[off++] = Armv8A64MkVecInstrUmov(IEMNATIVE_REG_FIXED_TMP0, IEMNATIVE_SIMD_REG_FIXED_TMP0 + 1,
6072	0 /idxElem/, kArmv8InstrUmovInsSz_U64);
6073	/* cbz tmp0, +1 */
6074	pu32CodeBuf[off++] = Armv8A64MkInstrCbzCbnz(false /fJmpIfNotZero/, 2, IEMNATIVE_REG_FIXED_TMP0);
6075	/* brk #0x1000+enmGstReg */
6076	pu32CodeBuf[off++] = Armv8A64MkInstrBrk((uint32_t)enmGstSimdReg \| UINT32_C(0x1000));
6077	}
6078
6079	# else
6080	# error "Port me!"
6081	# endif
6082
6083	IEMNATIVE_ASSERT_INSTR_BUF_ENSURE(pReNative, off);
6084	return off;
6085	}
6086	# endif /* IEMNATIVE_WITH_SIMD_REG_ALLOCATOR */
6087
6088
6089	/**
6090	* Emitting code that checks that IEMCPU::fExec matches @a fExec for all
6091	* important bits.
6092	*
6093	* @note May of course trash IEMNATIVE_REG_FIXED_TMP0.
6094	* Trashes EFLAGS on AMD64.
6095	*/
6096	DECL_HIDDEN_THROW(uint32_t)
6097	iemNativeEmitExecFlagsCheck(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint32_t fExec)
6098	{
6099	uint8_t const idxRegTmp = iemNativeRegAllocTmp(pReNative, &off);
6100	off = iemNativeEmitLoadGprFromVCpuU32(pReNative, off, idxRegTmp, RT_UOFFSETOF(VMCPUCC, iem.s.fExec));
6101	off = iemNativeEmitAndGpr32ByImm(pReNative, off, idxRegTmp, IEMTB_F_IEM_F_MASK & IEMTB_F_KEY_MASK);
6102	off = iemNativeEmitCmpGpr32WithImm(pReNative, off, idxRegTmp, fExec & IEMTB_F_KEY_MASK);
6103
6104	#ifdef RT_ARCH_AMD64
6105	uint8_t * const pbCodeBuf = iemNativeInstrBufEnsure(pReNative, off, 3);
6106
6107	/* je/jz +1 */
6108	pbCodeBuf[off++] = 0x74;
6109	pbCodeBuf[off++] = 0x01;
6110
6111	/* int3 */
6112	pbCodeBuf[off++] = 0xcc;
6113
6114	# elif defined(RT_ARCH_ARM64)
6115	uint32_t * const pu32CodeBuf = iemNativeInstrBufEnsure(pReNative, off, 2);
6116
6117	/* b.eq +1 */
6118	pu32CodeBuf[off++] = Armv8A64MkInstrBCond(kArmv8InstrCond_Eq, 2);
6119	/* brk #0x2000 */
6120	pu32CodeBuf[off++] = Armv8A64MkInstrBrk(UINT32_C(0x2000));
6121
6122	# else
6123	# error "Port me!"
6124	# endif
6125	IEMNATIVE_ASSERT_INSTR_BUF_ENSURE(pReNative, off);
6126
6127	iemNativeRegFreeTmp(pReNative, idxRegTmp);
6128	return off;
6129	}
6130
6131	#endif /* VBOX_STRICT */
6132
6133	/**
6134	* Emits a code for checking the return code of a call and rcPassUp, returning
6135	* from the code if either are non-zero.
6136	*/
6137	DECL_HIDDEN_THROW(uint32_t)
6138	iemNativeEmitCheckCallRetAndPassUp(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint8_t idxInstr)
6139	{
6140	#ifdef RT_ARCH_AMD64
6141	/*
6142	* AMD64: eax = call status code.
6143	*/
6144
6145	/* edx = rcPassUp */
6146	off = iemNativeEmitLoadGprFromVCpuU32(pReNative, off, X86_GREG_xDX, RT_UOFFSETOF(VMCPUCC, iem.s.rcPassUp));
6147	# ifdef IEMNATIVE_WITH_INSTRUCTION_COUNTING
6148	off = iemNativeEmitLoadGpr8Imm(pReNative, off, X86_GREG_xCX, idxInstr);
6149	# endif
6150
6151	/* edx = eax \| rcPassUp */
6152	uint8_t *pbCodeBuf = iemNativeInstrBufEnsure(pReNative, off, 2);
6153	pbCodeBuf[off++] = 0x0b; /* or edx, eax */
6154	pbCodeBuf[off++] = X86_MODRM_MAKE(X86_MOD_REG, X86_GREG_xDX, X86_GREG_xAX);
6155	IEMNATIVE_ASSERT_INSTR_BUF_ENSURE(pReNative, off);
6156
6157	/* Jump to non-zero status return path. */
6158	off = iemNativeEmitJnzToNewLabel(pReNative, off, kIemNativeLabelType_NonZeroRetOrPassUp);
6159
6160	/* done. */
6161
6162	#elif RT_ARCH_ARM64
6163	/*
6164	* ARM64: w0 = call status code.
6165	*/
6166	# ifdef IEMNATIVE_WITH_INSTRUCTION_COUNTING
6167	off = iemNativeEmitLoadGprImm64(pReNative, off, ARMV8_A64_REG_X2, idxInstr);
6168	# endif
6169	off = iemNativeEmitLoadGprFromVCpuU32(pReNative, off, ARMV8_A64_REG_X3, RT_UOFFSETOF(VMCPUCC, iem.s.rcPassUp));
6170
6171	uint32_t *pu32CodeBuf = iemNativeInstrBufEnsure(pReNative, off, 3);
6172
6173	pu32CodeBuf[off++] = Armv8A64MkInstrOrr(ARMV8_A64_REG_X4, ARMV8_A64_REG_X3, ARMV8_A64_REG_X0, false /f64Bit/);
6174
6175	uint32_t const idxLabel = iemNativeLabelCreate(pReNative, kIemNativeLabelType_NonZeroRetOrPassUp);
6176	iemNativeAddFixup(pReNative, off, idxLabel, kIemNativeFixupType_RelImm19At5);
6177	pu32CodeBuf[off++] = Armv8A64MkInstrCbzCbnz(true /fJmpIfNotZero/, 0, ARMV8_A64_REG_X4, false /f64Bit/);
6178
6179	#else
6180	# error "port me"
6181	#endif
6182	IEMNATIVE_ASSERT_INSTR_BUF_ENSURE(pReNative, off);
6183	RT_NOREF_PV(idxInstr);
6184	return off;
6185	}
6186
6187
6188	/**
6189	* Emits code to check if the content of @a idxAddrReg is a canonical address,
6190	* raising a \#GP(0) if it isn't.
6191	*
6192	* @returns New code buffer offset, UINT32_MAX on failure.
6193	* @param pReNative The native recompile state.
6194	* @param off The code buffer offset.
6195	* @param idxAddrReg The host register with the address to check.
6196	* @param idxInstr The current instruction.
6197	*/
6198	DECL_HIDDEN_THROW(uint32_t)
6199	iemNativeEmitCheckGprCanonicalMaybeRaiseGp0(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint8_t idxAddrReg, uint8_t idxInstr)
6200	{
6201	/*
6202	* Make sure we don't have any outstanding guest register writes as we may
6203	* raise an #GP(0) and all guest register must be up to date in CPUMCTX.
6204	*/
6205	off = iemNativeRegFlushPendingWrites(pReNative, off);
6206
6207	#ifdef IEMNATIVE_WITH_INSTRUCTION_COUNTING
6208	off = iemNativeEmitStoreImmToVCpuU8(pReNative, off, idxInstr, RT_UOFFSETOF(VMCPUCC, iem.s.idxTbCurInstr));
6209	#else
6210	RT_NOREF(idxInstr);
6211	#endif
6212
6213	#ifdef RT_ARCH_AMD64
6214	/*
6215	* if ((((uint32_t)(a_u64Addr >> 32) + UINT32_C(0x8000)) >> 16) != 0)
6216	* return raisexcpt();
6217	* ---- this wariant avoid loading a 64-bit immediate, but is an instruction longer.
6218	*/
6219	uint8_t const iTmpReg = iemNativeRegAllocTmp(pReNative, &off);
6220
6221	off = iemNativeEmitLoadGprFromGpr(pReNative, off, iTmpReg, idxAddrReg);
6222	off = iemNativeEmitShiftGprRight(pReNative, off, iTmpReg, 32);
6223	off = iemNativeEmitAddGpr32Imm(pReNative, off, iTmpReg, (int32_t)0x8000);
6224	off = iemNativeEmitShiftGprRight(pReNative, off, iTmpReg, 16);
6225	off = iemNativeEmitJnzToNewLabel(pReNative, off, kIemNativeLabelType_RaiseGp0);
6226
6227	iemNativeRegFreeTmp(pReNative, iTmpReg);
6228
6229	#elif defined(RT_ARCH_ARM64)
6230	/*
6231	* if ((((uint64_t)(a_u64Addr) + UINT64_C(0x800000000000)) >> 48) != 0)
6232	* return raisexcpt();
6233	* ----
6234	* mov x1, 0x800000000000
6235	* add x1, x0, x1
6236	* cmp xzr, x1, lsr 48
6237	* b.ne .Lraisexcpt
6238	*/
6239	uint8_t const iTmpReg = iemNativeRegAllocTmp(pReNative, &off);
6240
6241	off = iemNativeEmitLoadGprImm64(pReNative, off, iTmpReg, UINT64_C(0x800000000000));
6242	off = iemNativeEmitAddTwoGprs(pReNative, off, iTmpReg, idxAddrReg);
6243	off = iemNativeEmitCmpArm64(pReNative, off, ARMV8_A64_REG_XZR, iTmpReg, true /f64Bit/, 48 /cShift/, kArmv8A64InstrShift_Lsr);
6244	off = iemNativeEmitJnzToNewLabel(pReNative, off, kIemNativeLabelType_RaiseGp0);
6245
6246	iemNativeRegFreeTmp(pReNative, iTmpReg);
6247
6248	#else
6249	# error "Port me"
6250	#endif
6251	return off;
6252	}
6253
6254
6255	/**
6256	* Emits code to check if that the content of @a idxAddrReg is within the limit
6257	* of CS, raising a \#GP(0) if it isn't.
6258	*
6259	* @returns New code buffer offset; throws VBox status code on error.
6260	* @param pReNative The native recompile state.
6261	* @param off The code buffer offset.
6262	* @param idxAddrReg The host register (32-bit) with the address to
6263	* check.
6264	* @param idxInstr The current instruction.
6265	*/
6266	DECL_HIDDEN_THROW(uint32_t)
6267	iemNativeEmitCheckGpr32AgainstCsSegLimitMaybeRaiseGp0(PIEMRECOMPILERSTATE pReNative, uint32_t off,
6268	uint8_t idxAddrReg, uint8_t idxInstr)
6269	{
6270	/*
6271	* Make sure we don't have any outstanding guest register writes as we may
6272	* raise an #GP(0) and all guest register must be up to date in CPUMCTX.
6273	*/
6274	off = iemNativeRegFlushPendingWrites(pReNative, off);
6275
6276	#ifdef IEMNATIVE_WITH_INSTRUCTION_COUNTING
6277	off = iemNativeEmitStoreImmToVCpuU8(pReNative, off, idxInstr, RT_UOFFSETOF(VMCPUCC, iem.s.idxTbCurInstr));
6278	#else
6279	RT_NOREF(idxInstr);
6280	#endif
6281
6282	uint8_t const idxRegCsLim = iemNativeRegAllocTmpForGuestReg(pReNative, &off,
6283	(IEMNATIVEGSTREG)(kIemNativeGstReg_SegLimitFirst + X86_SREG_CS),
6284	kIemNativeGstRegUse_ReadOnly);
6285
6286	off = iemNativeEmitCmpGpr32WithGpr(pReNative, off, idxAddrReg, idxRegCsLim);
6287	off = iemNativeEmitJaToNewLabel(pReNative, off, kIemNativeLabelType_RaiseGp0);
6288
6289	iemNativeRegFreeTmp(pReNative, idxRegCsLim);
6290	return off;
6291	}
6292
6293
6294	/**
6295	* Emits a call to a CImpl function or something similar.
6296	*/
6297	DECL_HIDDEN_THROW(uint32_t)
6298	iemNativeEmitCImplCall(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint8_t idxInstr, uint64_t fGstShwFlush, uintptr_t pfnCImpl,
6299	uint8_t cbInstr, uint8_t cAddParams, uint64_t uParam0, uint64_t uParam1, uint64_t uParam2)
6300	{
6301	/* Writeback everything. */
6302	off = iemNativeRegFlushPendingWrites(pReNative, off);
6303
6304	/*
6305	* Flush stuff. PC and EFlags are implictly flushed, the latter because we
6306	* don't do with/without flags variants of defer-to-cimpl stuff at the moment.
6307	*/
6308	fGstShwFlush = iemNativeCImplFlagsToGuestShadowFlushMask(pReNative->fCImpl,
6309	fGstShwFlush
6310	\| RT_BIT_64(kIemNativeGstReg_Pc)
6311	\| RT_BIT_64(kIemNativeGstReg_EFlags));
6312	iemNativeRegFlushGuestShadows(pReNative, fGstShwFlush);
6313
6314	off = iemNativeRegMoveAndFreeAndFlushAtCall(pReNative, off, 4);
6315
6316	/*
6317	* Load the parameters.
6318	*/
6319	#if defined(RT_OS_WINDOWS) && defined(VBOXSTRICTRC_STRICT_ENABLED)
6320	/* Special code the hidden VBOXSTRICTRC pointer. */
6321	off = iemNativeEmitLoadGprFromGpr( pReNative, off, IEMNATIVE_CALL_ARG1_GREG, IEMNATIVE_REG_FIXED_PVMCPU);
6322	off = iemNativeEmitLoadGprImm64( pReNative, off, IEMNATIVE_CALL_ARG2_GREG, cbInstr); /** @todo 8-bit reg load opt for amd64 */
6323	if (cAddParams > 0)
6324	off = iemNativeEmitLoadGprImm64(pReNative, off, IEMNATIVE_CALL_ARG3_GREG, uParam0);
6325	if (cAddParams > 1)
6326	off = iemNativeEmitStoreImm64ByBp(pReNative, off, IEMNATIVE_FP_OFF_STACK_ARG0, uParam1);
6327	if (cAddParams > 2)
6328	off = iemNativeEmitStoreImm64ByBp(pReNative, off, IEMNATIVE_FP_OFF_STACK_ARG1, uParam2);
6329	off = iemNativeEmitLeaGprByBp(pReNative, off, X86_GREG_xCX, IEMNATIVE_FP_OFF_IN_SHADOW_ARG0); /* rcStrict */
6330
6331	#else
6332	AssertCompile(IEMNATIVE_CALL_ARG_GREG_COUNT >= 4);
6333	off = iemNativeEmitLoadGprFromGpr( pReNative, off, IEMNATIVE_CALL_ARG0_GREG, IEMNATIVE_REG_FIXED_PVMCPU);
6334	off = iemNativeEmitLoadGprImm64( pReNative, off, IEMNATIVE_CALL_ARG1_GREG, cbInstr); /** @todo 8-bit reg load opt for amd64 */
6335	if (cAddParams > 0)
6336	off = iemNativeEmitLoadGprImm64(pReNative, off, IEMNATIVE_CALL_ARG2_GREG, uParam0);
6337	if (cAddParams > 1)
6338	off = iemNativeEmitLoadGprImm64(pReNative, off, IEMNATIVE_CALL_ARG3_GREG, uParam1);
6339	if (cAddParams > 2)
6340	# if IEMNATIVE_CALL_ARG_GREG_COUNT >= 5
6341	off = iemNativeEmitLoadGprImm64(pReNative, off, IEMNATIVE_CALL_ARG4_GREG, uParam2);
6342	# else
6343	off = iemNativeEmitStoreImm64ByBp(pReNative, off, IEMNATIVE_FP_OFF_STACK_ARG0, uParam2);
6344	# endif
6345	#endif
6346
6347	/*
6348	* Make the call.
6349	*/
6350	off = iemNativeEmitCallImm(pReNative, off, pfnCImpl);
6351
6352	#if defined(RT_ARCH_AMD64) && defined(VBOXSTRICTRC_STRICT_ENABLED) && defined(RT_OS_WINDOWS)
6353	off = iemNativeEmitLoadGprByBpU32(pReNative, off, X86_GREG_xAX, IEMNATIVE_FP_OFF_IN_SHADOW_ARG0); /* rcStrict (see above) */
6354	#endif
6355
6356	/*
6357	* Check the status code.
6358	*/
6359	return iemNativeEmitCheckCallRetAndPassUp(pReNative, off, idxInstr);
6360	}
6361
6362
6363	/**
6364	* Emits a call to a threaded worker function.
6365	*/
6366	DECL_HIDDEN_THROW(uint32_t)
6367	iemNativeEmitThreadedCall(PIEMRECOMPILERSTATE pReNative, uint32_t off, PCIEMTHRDEDCALLENTRY pCallEntry)
6368	{
6369	/* We don't know what the threaded function is doing so we must flush all pending writes. */
6370	off = iemNativeRegFlushPendingWrites(pReNative, off);
6371
6372	iemNativeRegFlushGuestShadows(pReNative, UINT64_MAX); /** @todo optimize this */
6373	off = iemNativeRegMoveAndFreeAndFlushAtCall(pReNative, off, 4);
6374
6375	#ifdef IEMNATIVE_WITH_INSTRUCTION_COUNTING
6376	/* The threaded function may throw / long jmp, so set current instruction
6377	number if we're counting. */
6378	off = iemNativeEmitStoreImmToVCpuU8(pReNative, off, pCallEntry->idxInstr, RT_UOFFSETOF(VMCPUCC, iem.s.idxTbCurInstr));
6379	#endif
6380
6381	uint8_t const cParams = g_acIemThreadedFunctionUsedArgs[pCallEntry->enmFunction];
6382
6383	#ifdef RT_ARCH_AMD64
6384	/* Load the parameters and emit the call. */
6385	# ifdef RT_OS_WINDOWS
6386	# ifndef VBOXSTRICTRC_STRICT_ENABLED
6387	off = iemNativeEmitLoadGprFromGpr(pReNative, off, X86_GREG_xCX, IEMNATIVE_REG_FIXED_PVMCPU);
6388	if (cParams > 0)
6389	off = iemNativeEmitLoadGprImm64(pReNative, off, X86_GREG_xDX, pCallEntry->auParams[0]);
6390	if (cParams > 1)
6391	off = iemNativeEmitLoadGprImm64(pReNative, off, X86_GREG_x8, pCallEntry->auParams[1]);
6392	if (cParams > 2)
6393	off = iemNativeEmitLoadGprImm64(pReNative, off, X86_GREG_x9, pCallEntry->auParams[2]);
6394	# else /* VBOXSTRICTRC: Returned via hidden parameter. Sigh. */
6395	off = iemNativeEmitLoadGprFromGpr(pReNative, off, X86_GREG_xDX, IEMNATIVE_REG_FIXED_PVMCPU);
6396	if (cParams > 0)
6397	off = iemNativeEmitLoadGprImm64(pReNative, off, X86_GREG_x8, pCallEntry->auParams[0]);
6398	if (cParams > 1)
6399	off = iemNativeEmitLoadGprImm64(pReNative, off, X86_GREG_x9, pCallEntry->auParams[1]);
6400	if (cParams > 2)
6401	{
6402	off = iemNativeEmitLoadGprImm64(pReNative, off, X86_GREG_x10, pCallEntry->auParams[2]);
6403	off = iemNativeEmitStoreGprByBp(pReNative, off, IEMNATIVE_FP_OFF_STACK_ARG0, X86_GREG_x10);
6404	}
6405	off = iemNativeEmitLeaGprByBp(pReNative, off, X86_GREG_xCX, IEMNATIVE_FP_OFF_IN_SHADOW_ARG0); /* rcStrict */
6406	# endif /* VBOXSTRICTRC_STRICT_ENABLED */
6407	# else
6408	off = iemNativeEmitLoadGprFromGpr(pReNative, off, X86_GREG_xDI, IEMNATIVE_REG_FIXED_PVMCPU);
6409	if (cParams > 0)
6410	off = iemNativeEmitLoadGprImm64(pReNative, off, X86_GREG_xSI, pCallEntry->auParams[0]);
6411	if (cParams > 1)
6412	off = iemNativeEmitLoadGprImm64(pReNative, off, X86_GREG_xDX, pCallEntry->auParams[1]);
6413	if (cParams > 2)
6414	off = iemNativeEmitLoadGprImm64(pReNative, off, X86_GREG_xCX, pCallEntry->auParams[2]);
6415	# endif
6416
6417	off = iemNativeEmitCallImm(pReNative, off, (uintptr_t)g_apfnIemThreadedFunctions[pCallEntry->enmFunction]);
6418
6419	# if defined(VBOXSTRICTRC_STRICT_ENABLED) && defined(RT_OS_WINDOWS)
6420	off = iemNativeEmitLoadGprByBpU32(pReNative, off, X86_GREG_xAX, IEMNATIVE_FP_OFF_IN_SHADOW_ARG0); /* rcStrict (see above) */
6421	# endif
6422
6423	#elif RT_ARCH_ARM64
6424	/*
6425	* ARM64:
6426	*/
6427	off = iemNativeEmitLoadGprFromGpr(pReNative, off, IEMNATIVE_CALL_ARG0_GREG, IEMNATIVE_REG_FIXED_PVMCPU);
6428	if (cParams > 0)
6429	off = iemNativeEmitLoadGprImm64(pReNative, off, IEMNATIVE_CALL_ARG1_GREG, pCallEntry->auParams[0]);
6430	if (cParams > 1)
6431	off = iemNativeEmitLoadGprImm64(pReNative, off, IEMNATIVE_CALL_ARG2_GREG, pCallEntry->auParams[1]);
6432	if (cParams > 2)
6433	off = iemNativeEmitLoadGprImm64(pReNative, off, IEMNATIVE_CALL_ARG3_GREG, pCallEntry->auParams[2]);
6434
6435	off = iemNativeEmitCallImm(pReNative, off, (uintptr_t)g_apfnIemThreadedFunctions[pCallEntry->enmFunction]);
6436
6437	#else
6438	# error "port me"
6439	#endif
6440
6441	/*
6442	* Check the status code.
6443	*/
6444	off = iemNativeEmitCheckCallRetAndPassUp(pReNative, off, pCallEntry->idxInstr);
6445
6446	return off;
6447	}
6448
6449	#ifdef VBOX_WITH_STATISTICS
6450	/**
6451	* Emits code to update the thread call statistics.
6452	*/
6453	DECL_INLINE_THROW(uint32_t)
6454	iemNativeEmitThreadCallStats(PIEMRECOMPILERSTATE pReNative, uint32_t off, PCIEMTHRDEDCALLENTRY pCallEntry)
6455	{
6456	/*
6457	* Update threaded function stats.
6458	*/
6459	uint32_t const offVCpu = RT_UOFFSETOF_DYN(VMCPUCC, iem.s.acThreadedFuncStats[pCallEntry->enmFunction]);
6460	AssertCompile(sizeof(pReNative->pVCpu->iem.s.acThreadedFuncStats[pCallEntry->enmFunction]) == sizeof(uint32_t));
6461	# if defined(RT_ARCH_ARM64)
6462	uint8_t const idxTmp1 = iemNativeRegAllocTmp(pReNative, &off);
6463	uint8_t const idxTmp2 = iemNativeRegAllocTmp(pReNative, &off);
6464	off = iemNativeEmitIncU32CounterInVCpu(pReNative, off, idxTmp1, idxTmp2, offVCpu);
6465	iemNativeRegFreeTmp(pReNative, idxTmp1);
6466	iemNativeRegFreeTmp(pReNative, idxTmp2);
6467	# else
6468	off = iemNativeEmitIncU32CounterInVCpu(pReNative, off, UINT8_MAX, UINT8_MAX, offVCpu);
6469	# endif
6470	return off;
6471	}
6472	#endif /* VBOX_WITH_STATISTICS */
6473
6474
6475	/**
6476	* Emits the code at the CheckBranchMiss label.
6477	*/
6478	static uint32_t iemNativeEmitCheckBranchMiss(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint32_t idxReturnLabel)
6479	{
6480	uint32_t const idxLabel = iemNativeLabelFind(pReNative, kIemNativeLabelType_CheckBranchMiss);
6481	if (idxLabel != UINT32_MAX)
6482	{
6483	iemNativeLabelDefine(pReNative, idxLabel, off);
6484
6485	/* int iemNativeHlpCheckBranchMiss(PVMCPUCC pVCpu) */
6486	off = iemNativeEmitLoadGprFromGpr(pReNative, off, IEMNATIVE_CALL_ARG0_GREG, IEMNATIVE_REG_FIXED_PVMCPU);
6487	off = iemNativeEmitCallImm(pReNative, off, (uintptr_t)iemNativeHlpCheckBranchMiss);
6488
6489	/* jump back to the return sequence. */
6490	off = iemNativeEmitJmpToLabel(pReNative, off, idxReturnLabel);
6491	}
6492	return off;
6493	}
6494
6495
6496	/**
6497	* Emits the code at the NeedCsLimChecking label.
6498	*/
6499	static uint32_t iemNativeEmitNeedCsLimChecking(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint32_t idxReturnLabel)
6500	{
6501	uint32_t const idxLabel = iemNativeLabelFind(pReNative, kIemNativeLabelType_NeedCsLimChecking);
6502	if (idxLabel != UINT32_MAX)
6503	{
6504	iemNativeLabelDefine(pReNative, idxLabel, off);
6505
6506	/* int iemNativeHlpNeedCsLimChecking(PVMCPUCC pVCpu) */
6507	off = iemNativeEmitLoadGprFromGpr(pReNative, off, IEMNATIVE_CALL_ARG0_GREG, IEMNATIVE_REG_FIXED_PVMCPU);
6508	off = iemNativeEmitCallImm(pReNative, off, (uintptr_t)iemNativeHlpNeedCsLimChecking);
6509
6510	/* jump back to the return sequence. */
6511	off = iemNativeEmitJmpToLabel(pReNative, off, idxReturnLabel);
6512	}
6513	return off;
6514	}
6515
6516
6517	/**
6518	* Emits the code at the ObsoleteTb label.
6519	*/
6520	static uint32_t iemNativeEmitObsoleteTb(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint32_t idxReturnLabel)
6521	{
6522	uint32_t const idxLabel = iemNativeLabelFind(pReNative, kIemNativeLabelType_ObsoleteTb);
6523	if (idxLabel != UINT32_MAX)
6524	{
6525	iemNativeLabelDefine(pReNative, idxLabel, off);
6526
6527	/* int iemNativeHlpObsoleteTb(PVMCPUCC pVCpu) */
6528	off = iemNativeEmitLoadGprFromGpr(pReNative, off, IEMNATIVE_CALL_ARG0_GREG, IEMNATIVE_REG_FIXED_PVMCPU);
6529	off = iemNativeEmitCallImm(pReNative, off, (uintptr_t)iemNativeHlpObsoleteTb);
6530
6531	/* jump back to the return sequence. */
6532	off = iemNativeEmitJmpToLabel(pReNative, off, idxReturnLabel);
6533	}
6534	return off;
6535	}
6536
6537
6538	/**
6539	* Emits the code at the RaiseGP0 label.
6540	*/
6541	static uint32_t iemNativeEmitRaiseGp0(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint32_t idxReturnLabel)
6542	{
6543	uint32_t const idxLabel = iemNativeLabelFind(pReNative, kIemNativeLabelType_RaiseGp0);
6544	if (idxLabel != UINT32_MAX)
6545	{
6546	iemNativeLabelDefine(pReNative, idxLabel, off);
6547
6548	/* iemNativeHlpExecRaiseGp0(PVMCPUCC pVCpu) */
6549	off = iemNativeEmitLoadGprFromGpr(pReNative, off, IEMNATIVE_CALL_ARG0_GREG, IEMNATIVE_REG_FIXED_PVMCPU);
6550	off = iemNativeEmitCallImm(pReNative, off, (uintptr_t)iemNativeHlpExecRaiseGp0);
6551
6552	/* jump back to the return sequence. */
6553	off = iemNativeEmitJmpToLabel(pReNative, off, idxReturnLabel);
6554	}
6555	return off;
6556	}
6557
6558
6559	/**
6560	* Emits the code at the RaiseNm label.
6561	*/
6562	static uint32_t iemNativeEmitRaiseNm(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint32_t idxReturnLabel)
6563	{
6564	uint32_t const idxLabel = iemNativeLabelFind(pReNative, kIemNativeLabelType_RaiseNm);
6565	if (idxLabel != UINT32_MAX)
6566	{
6567	iemNativeLabelDefine(pReNative, idxLabel, off);
6568
6569	/* iemNativeHlpExecRaiseNm(PVMCPUCC pVCpu) */
6570	off = iemNativeEmitLoadGprFromGpr(pReNative, off, IEMNATIVE_CALL_ARG0_GREG, IEMNATIVE_REG_FIXED_PVMCPU);
6571	off = iemNativeEmitCallImm(pReNative, off, (uintptr_t)iemNativeHlpExecRaiseNm);
6572
6573	/* jump back to the return sequence. */
6574	off = iemNativeEmitJmpToLabel(pReNative, off, idxReturnLabel);
6575	}
6576	return off;
6577	}
6578
6579
6580	/**
6581	* Emits the code at the RaiseUd label.
6582	*/
6583	static uint32_t iemNativeEmitRaiseUd(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint32_t idxReturnLabel)
6584	{
6585	uint32_t const idxLabel = iemNativeLabelFind(pReNative, kIemNativeLabelType_RaiseUd);
6586	if (idxLabel != UINT32_MAX)
6587	{
6588	iemNativeLabelDefine(pReNative, idxLabel, off);
6589
6590	/* iemNativeHlpExecRaiseUd(PVMCPUCC pVCpu) */
6591	off = iemNativeEmitLoadGprFromGpr(pReNative, off, IEMNATIVE_CALL_ARG0_GREG, IEMNATIVE_REG_FIXED_PVMCPU);
6592	off = iemNativeEmitCallImm(pReNative, off, (uintptr_t)iemNativeHlpExecRaiseUd);
6593
6594	/* jump back to the return sequence. */
6595	off = iemNativeEmitJmpToLabel(pReNative, off, idxReturnLabel);
6596	}
6597	return off;
6598	}
6599
6600
6601	/**
6602	* Emits the code at the RaiseMf label.
6603	*/
6604	static uint32_t iemNativeEmitRaiseMf(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint32_t idxReturnLabel)
6605	{
6606	uint32_t const idxLabel = iemNativeLabelFind(pReNative, kIemNativeLabelType_RaiseMf);
6607	if (idxLabel != UINT32_MAX)
6608	{
6609	iemNativeLabelDefine(pReNative, idxLabel, off);
6610
6611	/* iemNativeHlpExecRaiseMf(PVMCPUCC pVCpu) */
6612	off = iemNativeEmitLoadGprFromGpr(pReNative, off, IEMNATIVE_CALL_ARG0_GREG, IEMNATIVE_REG_FIXED_PVMCPU);
6613	off = iemNativeEmitCallImm(pReNative, off, (uintptr_t)iemNativeHlpExecRaiseMf);
6614
6615	/* jump back to the return sequence. */
6616	off = iemNativeEmitJmpToLabel(pReNative, off, idxReturnLabel);
6617	}
6618	return off;
6619	}
6620
6621
6622	/**
6623	* Emits the code at the RaiseXf label.
6624	*/
6625	static uint32_t iemNativeEmitRaiseXf(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint32_t idxReturnLabel)
6626	{
6627	uint32_t const idxLabel = iemNativeLabelFind(pReNative, kIemNativeLabelType_RaiseXf);
6628	if (idxLabel != UINT32_MAX)
6629	{
6630	iemNativeLabelDefine(pReNative, idxLabel, off);
6631
6632	/* iemNativeHlpExecRaiseXf(PVMCPUCC pVCpu) */
6633	off = iemNativeEmitLoadGprFromGpr(pReNative, off, IEMNATIVE_CALL_ARG0_GREG, IEMNATIVE_REG_FIXED_PVMCPU);
6634	off = iemNativeEmitCallImm(pReNative, off, (uintptr_t)iemNativeHlpExecRaiseXf);
6635
6636	/* jump back to the return sequence. */
6637	off = iemNativeEmitJmpToLabel(pReNative, off, idxReturnLabel);
6638	}
6639	return off;
6640	}
6641
6642
6643	/**
6644	* Emits the code at the ReturnWithFlags label (returns
6645	* VINF_IEM_REEXEC_FINISH_WITH_FLAGS).
6646	*/
6647	static uint32_t iemNativeEmitReturnWithFlags(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint32_t idxReturnLabel)
6648	{
6649	uint32_t const idxLabel = iemNativeLabelFind(pReNative, kIemNativeLabelType_ReturnWithFlags);
6650	if (idxLabel != UINT32_MAX)
6651	{
6652	iemNativeLabelDefine(pReNative, idxLabel, off);
6653
6654	off = iemNativeEmitLoadGprImm64(pReNative, off, IEMNATIVE_CALL_RET_GREG, VINF_IEM_REEXEC_FINISH_WITH_FLAGS);
6655
6656	/* jump back to the return sequence. */
6657	off = iemNativeEmitJmpToLabel(pReNative, off, idxReturnLabel);
6658	}
6659	return off;
6660	}
6661
6662
6663	/**
6664	* Emits the code at the ReturnBreak label (returns VINF_IEM_REEXEC_BREAK).
6665	*/
6666	static uint32_t iemNativeEmitReturnBreak(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint32_t idxReturnLabel)
6667	{
6668	uint32_t const idxLabel = iemNativeLabelFind(pReNative, kIemNativeLabelType_ReturnBreak);
6669	if (idxLabel != UINT32_MAX)
6670	{
6671	iemNativeLabelDefine(pReNative, idxLabel, off);
6672
6673	off = iemNativeEmitLoadGprImm64(pReNative, off, IEMNATIVE_CALL_RET_GREG, VINF_IEM_REEXEC_BREAK);
6674
6675	/* jump back to the return sequence. */
6676	off = iemNativeEmitJmpToLabel(pReNative, off, idxReturnLabel);
6677	}
6678	return off;
6679	}
6680
6681
6682	/**
6683	* Emits the RC fiddling code for handling non-zero return code or rcPassUp.
6684	*/
6685	static uint32_t iemNativeEmitRcFiddling(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint32_t idxReturnLabel)
6686	{
6687	/*
6688	* Generate the rc + rcPassUp fiddling code if needed.
6689	*/
6690	uint32_t const idxLabel = iemNativeLabelFind(pReNative, kIemNativeLabelType_NonZeroRetOrPassUp);
6691	if (idxLabel != UINT32_MAX)
6692	{
6693	iemNativeLabelDefine(pReNative, idxLabel, off);
6694
6695	/* iemNativeHlpExecStatusCodeFiddling(PVMCPUCC pVCpu, int rc, uint8_t idxInstr) */
6696	#ifdef RT_ARCH_AMD64
6697	# ifdef RT_OS_WINDOWS
6698	# ifdef IEMNATIVE_WITH_INSTRUCTION_COUNTING
6699	off = iemNativeEmitLoadGprFromGpr(pReNative, off, X86_GREG_x8, X86_GREG_xCX); /* cl = instruction number */
6700	# endif
6701	off = iemNativeEmitLoadGprFromGpr(pReNative, off, X86_GREG_xCX, IEMNATIVE_REG_FIXED_PVMCPU);
6702	off = iemNativeEmitLoadGprFromGpr(pReNative, off, X86_GREG_xDX, X86_GREG_xAX);
6703	# else
6704	off = iemNativeEmitLoadGprFromGpr(pReNative, off, X86_GREG_xDI, IEMNATIVE_REG_FIXED_PVMCPU);
6705	off = iemNativeEmitLoadGprFromGpr(pReNative, off, X86_GREG_xSI, X86_GREG_xAX);
6706	# ifdef IEMNATIVE_WITH_INSTRUCTION_COUNTING
6707	off = iemNativeEmitLoadGprFromGpr(pReNative, off, X86_GREG_xDX, X86_GREG_xCX); /* cl = instruction number */
6708	# endif
6709	# endif
6710	# ifndef IEMNATIVE_WITH_INSTRUCTION_COUNTING
6711	off = iemNativeEmitLoadGpr8Imm(pReNative, off, X86_GREG_xCX, 0);
6712	# endif
6713
6714	#else
6715	off = iemNativeEmitLoadGprFromGpr(pReNative, off, IEMNATIVE_CALL_ARG1_GREG, IEMNATIVE_CALL_RET_GREG);
6716	off = iemNativeEmitLoadGprFromGpr(pReNative, off, IEMNATIVE_CALL_ARG0_GREG, IEMNATIVE_REG_FIXED_PVMCPU);
6717	/* IEMNATIVE_CALL_ARG2_GREG is already set. */
6718	#endif
6719
6720	off = iemNativeEmitCallImm(pReNative, off, (uintptr_t)iemNativeHlpExecStatusCodeFiddling);
6721	off = iemNativeEmitJmpToLabel(pReNative, off, idxReturnLabel);
6722	}
6723	return off;
6724	}
6725
6726
6727	/**
6728	* Emits a standard epilog.
6729	*/
6730	static uint32_t iemNativeEmitEpilog(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint32_t *pidxReturnLabel)
6731	{
6732	*pidxReturnLabel = UINT32_MAX;
6733
6734	/* Flush any pending writes before returning from the last instruction (RIP updates, etc.). */
6735	off = iemNativeRegFlushPendingWrites(pReNative, off);
6736
6737	/*
6738	* Successful return, so clear the return register (eax, w0).
6739	*/
6740	off = iemNativeEmitGprZero(pReNative,off, IEMNATIVE_CALL_RET_GREG);
6741
6742	/*
6743	* Define label for common return point.
6744	*/
6745	uint32_t const idxReturn = iemNativeLabelCreate(pReNative, kIemNativeLabelType_Return, off);
6746	*pidxReturnLabel = idxReturn;
6747
6748	/*
6749	* Restore registers and return.
6750	*/
6751	#ifdef RT_ARCH_AMD64
6752	uint8_t * const pbCodeBuf = iemNativeInstrBufEnsure(pReNative, off, 20);
6753
6754	/* Reposition esp at the r15 restore point. */
6755	pbCodeBuf[off++] = X86_OP_REX_W;
6756	pbCodeBuf[off++] = 0x8d; /* lea rsp, [rbp - (gcc ? 5 : 7) * 8] */
6757	pbCodeBuf[off++] = X86_MODRM_MAKE(X86_MOD_MEM1, X86_GREG_xSP, X86_GREG_xBP);
6758	pbCodeBuf[off++] = (uint8_t)IEMNATIVE_FP_OFF_LAST_PUSH;
6759
6760	/* Pop non-volatile registers and return */
6761	pbCodeBuf[off++] = X86_OP_REX_B; /* pop r15 */
6762	pbCodeBuf[off++] = 0x58 + X86_GREG_x15 - 8;
6763	pbCodeBuf[off++] = X86_OP_REX_B; /* pop r14 */
6764	pbCodeBuf[off++] = 0x58 + X86_GREG_x14 - 8;
6765	pbCodeBuf[off++] = X86_OP_REX_B; /* pop r13 */
6766	pbCodeBuf[off++] = 0x58 + X86_GREG_x13 - 8;
6767	pbCodeBuf[off++] = X86_OP_REX_B; /* pop r12 */
6768	pbCodeBuf[off++] = 0x58 + X86_GREG_x12 - 8;
6769	# ifdef RT_OS_WINDOWS
6770	pbCodeBuf[off++] = 0x58 + X86_GREG_xDI; /* pop rdi */
6771	pbCodeBuf[off++] = 0x58 + X86_GREG_xSI; /* pop rsi */
6772	# endif
6773	pbCodeBuf[off++] = 0x58 + X86_GREG_xBX; /* pop rbx */
6774	pbCodeBuf[off++] = 0xc9; /* leave */
6775	pbCodeBuf[off++] = 0xc3; /* ret */
6776	pbCodeBuf[off++] = 0xcc; /* int3 poison */
6777
6778	#elif RT_ARCH_ARM64
6779	uint32_t * const pu32CodeBuf = iemNativeInstrBufEnsure(pReNative, off, 10);
6780
6781	/* ldp x19, x20, [sp #IEMNATIVE_FRAME_VAR_SIZE]! ; Unallocate the variable space and restore x19+x20. */
6782	AssertCompile(IEMNATIVE_FRAME_VAR_SIZE < 64*8);
6783	pu32CodeBuf[off++] = Armv8A64MkInstrStLdPair(true /fLoad/, 2 /64-bit/, kArm64InstrStLdPairType_PreIndex,
6784	ARMV8_A64_REG_X19, ARMV8_A64_REG_X20, ARMV8_A64_REG_SP,
6785	IEMNATIVE_FRAME_VAR_SIZE / 8);
6786	/* Restore x21 thru x28 + BP and LR (ret address) (SP remains unchanged in the kSigned variant). */
6787	pu32CodeBuf[off++] = Armv8A64MkInstrStLdPair(true /fLoad/, 2 /64-bit/, kArm64InstrStLdPairType_Signed,
6788	ARMV8_A64_REG_X21, ARMV8_A64_REG_X22, ARMV8_A64_REG_SP, 2);
6789	pu32CodeBuf[off++] = Armv8A64MkInstrStLdPair(true /fLoad/, 2 /64-bit/, kArm64InstrStLdPairType_Signed,
6790	ARMV8_A64_REG_X23, ARMV8_A64_REG_X24, ARMV8_A64_REG_SP, 4);
6791	pu32CodeBuf[off++] = Armv8A64MkInstrStLdPair(true /fLoad/, 2 /64-bit/, kArm64InstrStLdPairType_Signed,
6792	ARMV8_A64_REG_X25, ARMV8_A64_REG_X26, ARMV8_A64_REG_SP, 6);
6793	pu32CodeBuf[off++] = Armv8A64MkInstrStLdPair(true /fLoad/, 2 /64-bit/, kArm64InstrStLdPairType_Signed,
6794	ARMV8_A64_REG_X27, ARMV8_A64_REG_X28, ARMV8_A64_REG_SP, 8);
6795	pu32CodeBuf[off++] = Armv8A64MkInstrStLdPair(true /fLoad/, 2 /64-bit/, kArm64InstrStLdPairType_Signed,
6796	ARMV8_A64_REG_BP, ARMV8_A64_REG_LR, ARMV8_A64_REG_SP, 10);
6797	AssertCompile(IEMNATIVE_FRAME_SAVE_REG_SIZE / 8 == 12);
6798
6799	/* add sp, sp, IEMNATIVE_FRAME_SAVE_REG_SIZE ; */
6800	AssertCompile(IEMNATIVE_FRAME_SAVE_REG_SIZE < 4096);
6801	pu32CodeBuf[off++] = Armv8A64MkInstrAddSubUImm12(false /fSub/, ARMV8_A64_REG_SP, ARMV8_A64_REG_SP,
6802	IEMNATIVE_FRAME_SAVE_REG_SIZE);
6803
6804	/* retab / ret */
6805	# ifdef RT_OS_DARWIN /** @todo See todo on pacibsp in the prolog. */
6806	if (1)
6807	pu32CodeBuf[off++] = ARMV8_A64_INSTR_RETAB;
6808	else
6809	# endif
6810	pu32CodeBuf[off++] = ARMV8_A64_INSTR_RET;
6811
6812	#else
6813	# error "port me"
6814	#endif
6815	IEMNATIVE_ASSERT_INSTR_BUF_ENSURE(pReNative, off);
6816
6817	return iemNativeEmitRcFiddling(pReNative, off, idxReturn);
6818	}
6819
6820
6821	/**
6822	* Emits a standard prolog.
6823	*/
6824	static uint32_t iemNativeEmitProlog(PIEMRECOMPILERSTATE pReNative, uint32_t off)
6825	{
6826	#ifdef RT_ARCH_AMD64
6827	/*
6828	* Set up a regular xBP stack frame, pushing all non-volatile GPRs,
6829	* reserving 64 bytes for stack variables plus 4 non-register argument
6830	* slots. Fixed register assignment: xBX = pReNative;
6831	*
6832	* Since we always do the same register spilling, we can use the same
6833	* unwind description for all the code.
6834	*/
6835	uint8_t *const pbCodeBuf = iemNativeInstrBufEnsure(pReNative, off, 32);
6836	pbCodeBuf[off++] = 0x50 + X86_GREG_xBP; /* push rbp */
6837	pbCodeBuf[off++] = X86_OP_REX_W; /* mov rbp, rsp */
6838	pbCodeBuf[off++] = 0x8b;
6839	pbCodeBuf[off++] = X86_MODRM_MAKE(X86_MOD_REG, X86_GREG_xBP, X86_GREG_xSP);
6840	pbCodeBuf[off++] = 0x50 + X86_GREG_xBX; /* push rbx */
6841	AssertCompile(IEMNATIVE_REG_FIXED_PVMCPU == X86_GREG_xBX);
6842	# ifdef RT_OS_WINDOWS
6843	pbCodeBuf[off++] = X86_OP_REX_W; /* mov rbx, rcx ; RBX = pVCpu */
6844	pbCodeBuf[off++] = 0x8b;
6845	pbCodeBuf[off++] = X86_MODRM_MAKE(X86_MOD_REG, X86_GREG_xBX, X86_GREG_xCX);
6846	pbCodeBuf[off++] = 0x50 + X86_GREG_xSI; /* push rsi */
6847	pbCodeBuf[off++] = 0x50 + X86_GREG_xDI; /* push rdi */
6848	# else
6849	pbCodeBuf[off++] = X86_OP_REX_W; /* mov rbx, rdi ; RBX = pVCpu */
6850	pbCodeBuf[off++] = 0x8b;
6851	pbCodeBuf[off++] = X86_MODRM_MAKE(X86_MOD_REG, X86_GREG_xBX, X86_GREG_xDI);
6852	# endif
6853	pbCodeBuf[off++] = X86_OP_REX_B; /* push r12 */
6854	pbCodeBuf[off++] = 0x50 + X86_GREG_x12 - 8;
6855	pbCodeBuf[off++] = X86_OP_REX_B; /* push r13 */
6856	pbCodeBuf[off++] = 0x50 + X86_GREG_x13 - 8;
6857	pbCodeBuf[off++] = X86_OP_REX_B; /* push r14 */
6858	pbCodeBuf[off++] = 0x50 + X86_GREG_x14 - 8;
6859	pbCodeBuf[off++] = X86_OP_REX_B; /* push r15 */
6860	pbCodeBuf[off++] = 0x50 + X86_GREG_x15 - 8;
6861
6862	# ifdef VBOX_WITH_IEM_NATIVE_RECOMPILER_LONGJMP
6863	/* Save the frame pointer. */
6864	off = iemNativeEmitStoreGprToVCpuU64Ex(pbCodeBuf, off, X86_GREG_xBP, RT_UOFFSETOF(VMCPUCC, iem.s.pvTbFramePointerR3));
6865	# endif
6866
6867	off = iemNativeEmitSubGprImm(pReNative, off, /* sub rsp, byte 28h */
6868	X86_GREG_xSP,
6869	IEMNATIVE_FRAME_ALIGN_SIZE
6870	+ IEMNATIVE_FRAME_VAR_SIZE
6871	+ IEMNATIVE_FRAME_STACK_ARG_COUNT * 8
6872	+ IEMNATIVE_FRAME_SHADOW_ARG_COUNT * 8);
6873	AssertCompile(!(IEMNATIVE_FRAME_VAR_SIZE & 0xf));
6874	AssertCompile(!(IEMNATIVE_FRAME_STACK_ARG_COUNT & 0x1));
6875	AssertCompile(!(IEMNATIVE_FRAME_SHADOW_ARG_COUNT & 0x1));
6876
6877	#elif RT_ARCH_ARM64
6878	/*
6879	* We set up a stack frame exactly like on x86, only we have to push the
6880	* return address our selves here. We save all non-volatile registers.
6881	*/
6882	uint32_t * const pu32CodeBuf = iemNativeInstrBufEnsure(pReNative, off, 16);
6883
6884	# ifdef RT_OS_DARWIN /** @todo This seems to be requirement by libunwind for JIT FDEs. Investigate further as been unable
6885	* to figure out where the BRK following AUTHB+XPACB stuff comes from in libunwind. It's
6886	* definitely the dwarf stepping code, but till found it's very tedious to figure out whether it's
6887	* in any way conditional, so just emitting this instructions now and hoping for the best... */
6888	/* pacibsp */
6889	pu32CodeBuf[off++] = ARMV8_A64_INSTR_PACIBSP;
6890	# endif
6891
6892	/* stp x19, x20, [sp, #-IEMNATIVE_FRAME_SAVE_REG_SIZE] ; Allocate space for saving registers and place x19+x20 at the bottom. */
6893	AssertCompile(IEMNATIVE_FRAME_SAVE_REG_SIZE < 64*8);
6894	pu32CodeBuf[off++] = Armv8A64MkInstrStLdPair(false /fLoad/, 2 /64-bit/, kArm64InstrStLdPairType_PreIndex,
6895	ARMV8_A64_REG_X19, ARMV8_A64_REG_X20, ARMV8_A64_REG_SP,
6896	-IEMNATIVE_FRAME_SAVE_REG_SIZE / 8);
6897	/* Save x21 thru x28 (SP remains unchanged in the kSigned variant). */
6898	pu32CodeBuf[off++] = Armv8A64MkInstrStLdPair(false /fLoad/, 2 /64-bit/, kArm64InstrStLdPairType_Signed,
6899	ARMV8_A64_REG_X21, ARMV8_A64_REG_X22, ARMV8_A64_REG_SP, 2);
6900	pu32CodeBuf[off++] = Armv8A64MkInstrStLdPair(false /fLoad/, 2 /64-bit/, kArm64InstrStLdPairType_Signed,
6901	ARMV8_A64_REG_X23, ARMV8_A64_REG_X24, ARMV8_A64_REG_SP, 4);
6902	pu32CodeBuf[off++] = Armv8A64MkInstrStLdPair(false /fLoad/, 2 /64-bit/, kArm64InstrStLdPairType_Signed,
6903	ARMV8_A64_REG_X25, ARMV8_A64_REG_X26, ARMV8_A64_REG_SP, 6);
6904	pu32CodeBuf[off++] = Armv8A64MkInstrStLdPair(false /fLoad/, 2 /64-bit/, kArm64InstrStLdPairType_Signed,
6905	ARMV8_A64_REG_X27, ARMV8_A64_REG_X28, ARMV8_A64_REG_SP, 8);
6906	/* Save the BP and LR (ret address) registers at the top of the frame. */
6907	pu32CodeBuf[off++] = Armv8A64MkInstrStLdPair(false /fLoad/, 2 /64-bit/, kArm64InstrStLdPairType_Signed,
6908	ARMV8_A64_REG_BP, ARMV8_A64_REG_LR, ARMV8_A64_REG_SP, 10);
6909	AssertCompile(IEMNATIVE_FRAME_SAVE_REG_SIZE / 8 == 12);
6910	/* add bp, sp, IEMNATIVE_FRAME_SAVE_REG_SIZE - 16 ; Set BP to point to the old BP stack address. */
6911	pu32CodeBuf[off++] = Armv8A64MkInstrAddSubUImm12(false /fSub/, ARMV8_A64_REG_BP,
6912	ARMV8_A64_REG_SP, IEMNATIVE_FRAME_SAVE_REG_SIZE - 16);
6913
6914	/* sub sp, sp, IEMNATIVE_FRAME_VAR_SIZE ; Allocate the variable area from SP. */
6915	pu32CodeBuf[off++] = Armv8A64MkInstrAddSubUImm12(true /fSub/, ARMV8_A64_REG_SP, ARMV8_A64_REG_SP, IEMNATIVE_FRAME_VAR_SIZE);
6916
6917	/* mov r28, r0 */
6918	off = iemNativeEmitLoadGprFromGprEx(pu32CodeBuf, off, IEMNATIVE_REG_FIXED_PVMCPU, IEMNATIVE_CALL_ARG0_GREG);
6919	/* mov r27, r1 */
6920	off = iemNativeEmitLoadGprFromGprEx(pu32CodeBuf, off, IEMNATIVE_REG_FIXED_PCPUMCTX, IEMNATIVE_CALL_ARG1_GREG);
6921
6922	# ifdef VBOX_WITH_IEM_NATIVE_RECOMPILER_LONGJMP
6923	/* Save the frame pointer. */
6924	off = iemNativeEmitStoreGprToVCpuU64Ex(pu32CodeBuf, off, ARMV8_A64_REG_BP, RT_UOFFSETOF(VMCPUCC, iem.s.pvTbFramePointerR3),
6925	ARMV8_A64_REG_X2);
6926	# endif
6927
6928	#else
6929	# error "port me"
6930	#endif
6931	IEMNATIVE_ASSERT_INSTR_BUF_ENSURE(pReNative, off);
6932	return off;
6933	}
6934
6935
6936	/*********************************************************************************************************************************
6937	* Emitters for IEM_MC_ARG_XXX, IEM_MC_LOCAL, IEM_MC_LOCAL_CONST, ++ *
6938	*********************************************************************************************************************************/
6939
6940	/**
6941	* Internal work that allocates a variable with kind set to
6942	* kIemNativeVarKind_Invalid and no current stack allocation.
6943	*
6944	* The kind will either be set by the caller or later when the variable is first
6945	* assigned a value.
6946	*
6947	* @returns Unpacked index.
6948	* @internal
6949	*/
6950	static uint8_t iemNativeVarAllocInt(PIEMRECOMPILERSTATE pReNative, uint8_t cbType)
6951	{
6952	Assert(cbType > 0 && cbType <= 64);
6953	unsigned const idxVar = ASMBitFirstSetU32(~pReNative->Core.bmVars) - 1;
6954	AssertStmt(idxVar < RT_ELEMENTS(pReNative->Core.aVars), IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_VAR_EXHAUSTED));
6955	pReNative->Core.bmVars \|= RT_BIT_32(idxVar);
6956	pReNative->Core.aVars[idxVar].enmKind = kIemNativeVarKind_Invalid;
6957	pReNative->Core.aVars[idxVar].cbVar = cbType;
6958	pReNative->Core.aVars[idxVar].idxStackSlot = UINT8_MAX;
6959	pReNative->Core.aVars[idxVar].idxReg = UINT8_MAX;
6960	pReNative->Core.aVars[idxVar].uArgNo = UINT8_MAX;
6961	pReNative->Core.aVars[idxVar].idxReferrerVar = UINT8_MAX;
6962	pReNative->Core.aVars[idxVar].enmGstReg = kIemNativeGstReg_End;
6963	pReNative->Core.aVars[idxVar].fRegAcquired = false;
6964	pReNative->Core.aVars[idxVar].u.uValue = 0;
6965	return idxVar;
6966	}
6967
6968
6969	/**
6970	* Internal work that allocates an argument variable w/o setting enmKind.
6971	*
6972	* @returns Unpacked index.
6973	* @internal
6974	*/
6975	static uint8_t iemNativeArgAllocInt(PIEMRECOMPILERSTATE pReNative, uint8_t iArgNo, uint8_t cbType)
6976	{
6977	iArgNo += iemNativeArgGetHiddenArgCount(pReNative);
6978	AssertStmt(iArgNo < RT_ELEMENTS(pReNative->Core.aidxArgVars), IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_VAR_IPE_1));
6979	AssertStmt(pReNative->Core.aidxArgVars[iArgNo] == UINT8_MAX, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_VAR_DUP_ARG_NO));
6980
6981	uint8_t const idxVar = iemNativeVarAllocInt(pReNative, cbType);
6982	pReNative->Core.aidxArgVars[iArgNo] = idxVar; /* (unpacked) */
6983	pReNative->Core.aVars[idxVar].uArgNo = iArgNo;
6984	return idxVar;
6985	}
6986
6987
6988	/**
6989	* Gets the stack slot for a stack variable, allocating one if necessary.
6990	*
6991	* Calling this function implies that the stack slot will contain a valid
6992	* variable value. The caller deals with any register currently assigned to the
6993	* variable, typically by spilling it into the stack slot.
6994	*
6995	* @returns The stack slot number.
6996	* @param pReNative The recompiler state.
6997	* @param idxVar The variable.
6998	* @throws VERR_IEM_VAR_OUT_OF_STACK_SLOTS
6999	*/
7000	DECL_HIDDEN_THROW(uint8_t) iemNativeVarGetStackSlot(PIEMRECOMPILERSTATE pReNative, uint8_t idxVar)
7001	{
7002	IEMNATIVE_ASSERT_VAR_IDX(pReNative, idxVar);
7003	PIEMNATIVEVAR const pVar = &pReNative->Core.aVars[IEMNATIVE_VAR_IDX_UNPACK(idxVar)];
7004	Assert(pVar->enmKind == kIemNativeVarKind_Stack);
7005
7006	/* Already got a slot? */
7007	uint8_t const idxStackSlot = pVar->idxStackSlot;
7008	if (idxStackSlot != UINT8_MAX)
7009	{
7010	Assert(idxStackSlot < IEMNATIVE_FRAME_VAR_SLOTS);
7011	return idxStackSlot;
7012	}
7013
7014	/*
7015	* A single slot is easy to allocate.
7016	* Allocate them from the top end, closest to BP, to reduce the displacement.
7017	*/
7018	if (pVar->cbVar <= sizeof(uint64_t))
7019	{
7020	unsigned const iSlot = ASMBitLastSetU32(~pReNative->Core.bmStack) - 1;
7021	AssertStmt(iSlot < IEMNATIVE_FRAME_VAR_SLOTS, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_VAR_OUT_OF_STACK_SLOTS));
7022	pReNative->Core.bmStack \|= RT_BIT_32(iSlot);
7023	pVar->idxStackSlot = (uint8_t)iSlot;
7024	Log11(("iemNativeVarSetKindToStack: idxVar=%#x iSlot=%#x\n", idxVar, iSlot));
7025	return (uint8_t)iSlot;
7026	}
7027
7028	/*
7029	* We need more than one stack slot.
7030	*
7031	* cbVar -> fBitAlignMask: 16 -> 1; 32 -> 3; 64 -> 7;
7032	*/
7033	AssertCompile(RT_IS_POWER_OF_TWO(IEMNATIVE_FRAME_VAR_SLOTS)); /* If not we have to add an overflow check. */
7034	Assert(pVar->cbVar <= 64);
7035	uint32_t const fBitAlignMask = RT_BIT_32(ASMBitLastSetU32(pVar->cbVar) - 4) - 1;
7036	uint32_t fBitAllocMask = RT_BIT_32((pVar->cbVar + 7) >> 3) - 1;
7037	uint32_t bmStack = ~pReNative->Core.bmStack;
7038	while (bmStack != UINT32_MAX)
7039	{
7040	/** @todo allocate from the top to reduce BP displacement. */
7041	unsigned const iSlot = ASMBitFirstSetU32(bmStack) - 1;
7042	AssertStmt(iSlot < IEMNATIVE_FRAME_VAR_SLOTS, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_VAR_OUT_OF_STACK_SLOTS));
7043	if (!(iSlot & fBitAlignMask))
7044	{
7045	if ((bmStack & (fBitAllocMask << iSlot)) == (fBitAllocMask << iSlot))
7046	{
7047	pReNative->Core.bmStack \|= (fBitAllocMask << iSlot);
7048	pVar->idxStackSlot = (uint8_t)iSlot;
7049	Log11(("iemNativeVarSetKindToStack: idxVar=%#x iSlot=%#x/%#x (cbVar=%#x)\n",
7050	idxVar, iSlot, fBitAllocMask, pVar->cbVar));
7051	return (uint8_t)iSlot;
7052	}
7053	}
7054	bmStack \|= fBitAlignMask << (iSlot & ~fBitAlignMask);
7055	}
7056	AssertFailedStmt(IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_VAR_OUT_OF_STACK_SLOTS));
7057	}
7058
7059
7060	/**
7061	* Changes the variable to a stack variable.
7062	*
7063	* Currently this is s only possible to do the first time the variable is used,
7064	* switching later is can be implemented but not done.
7065	*
7066	* @param pReNative The recompiler state.
7067	* @param idxVar The variable.
7068	* @throws VERR_IEM_VAR_IPE_2
7069	*/
7070	DECL_HIDDEN_THROW(void) iemNativeVarSetKindToStack(PIEMRECOMPILERSTATE pReNative, uint8_t idxVar)
7071	{
7072	IEMNATIVE_ASSERT_VAR_IDX(pReNative, idxVar);
7073	PIEMNATIVEVAR const pVar = &pReNative->Core.aVars[IEMNATIVE_VAR_IDX_UNPACK(idxVar)];
7074	if (pVar->enmKind != kIemNativeVarKind_Stack)
7075	{
7076	/* We could in theory transition from immediate to stack as well, but it
7077	would involve the caller doing work storing the value on the stack. So,
7078	till that's required we only allow transition from invalid. */
7079	AssertStmt(pVar->enmKind == kIemNativeVarKind_Invalid, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_VAR_IPE_2));
7080	AssertStmt(pVar->idxReg == UINT8_MAX, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_VAR_IPE_2));
7081	pVar->enmKind = kIemNativeVarKind_Stack;
7082
7083	/* Note! We don't allocate a stack slot here, that's only done when a
7084	slot is actually needed to hold a variable value. */
7085	}
7086	}
7087
7088
7089	/**
7090	* Sets it to a variable with a constant value.
7091	*
7092	* This does not require stack storage as we know the value and can always
7093	* reload it, unless of course it's referenced.
7094	*
7095	* @param pReNative The recompiler state.
7096	* @param idxVar The variable.
7097	* @param uValue The immediate value.
7098	* @throws VERR_IEM_VAR_OUT_OF_STACK_SLOTS, VERR_IEM_VAR_IPE_2
7099	*/
7100	DECL_HIDDEN_THROW(void) iemNativeVarSetKindToConst(PIEMRECOMPILERSTATE pReNative, uint8_t idxVar, uint64_t uValue)
7101	{
7102	IEMNATIVE_ASSERT_VAR_IDX(pReNative, idxVar);
7103	PIEMNATIVEVAR const pVar = &pReNative->Core.aVars[IEMNATIVE_VAR_IDX_UNPACK(idxVar)];
7104	if (pVar->enmKind != kIemNativeVarKind_Immediate)
7105	{
7106	/* Only simple transitions for now. */
7107	AssertStmt(pVar->enmKind == kIemNativeVarKind_Invalid, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_VAR_IPE_2));
7108	pVar->enmKind = kIemNativeVarKind_Immediate;
7109	}
7110	AssertStmt(pVar->idxReg == UINT8_MAX, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_VAR_IPE_2));
7111
7112	pVar->u.uValue = uValue;
7113	AssertMsg( pVar->cbVar >= sizeof(uint64_t)
7114	\|\| pVar->u.uValue < RT_BIT_64(pVar->cbVar * 8),
7115	("idxVar=%d cbVar=%u uValue=%#RX64\n", idxVar, pVar->cbVar, uValue));
7116	}
7117
7118
7119	/**
7120	* Sets the variable to a reference (pointer) to @a idxOtherVar.
7121	*
7122	* This does not require stack storage as we know the value and can always
7123	* reload it. Loading is postponed till needed.
7124	*
7125	* @param pReNative The recompiler state.
7126	* @param idxVar The variable. Unpacked.
7127	* @param idxOtherVar The variable to take the (stack) address of. Unpacked.
7128	*
7129	* @throws VERR_IEM_VAR_OUT_OF_STACK_SLOTS, VERR_IEM_VAR_IPE_2
7130	* @internal
7131	*/
7132	static void iemNativeVarSetKindToLocalRef(PIEMRECOMPILERSTATE pReNative, uint8_t idxVar, uint8_t idxOtherVar)
7133	{
7134	Assert(idxVar < RT_ELEMENTS(pReNative->Core.aVars) && (pReNative->Core.bmVars & RT_BIT_32(idxVar)));
7135	Assert(idxOtherVar < RT_ELEMENTS(pReNative->Core.aVars) && (pReNative->Core.bmVars & RT_BIT_32(idxOtherVar)));
7136
7137	if (pReNative->Core.aVars[idxVar].enmKind != kIemNativeVarKind_VarRef)
7138	{
7139	/* Only simple transitions for now. */
7140	AssertStmt(pReNative->Core.aVars[idxVar].enmKind == kIemNativeVarKind_Invalid,
7141	IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_VAR_IPE_2));
7142	pReNative->Core.aVars[idxVar].enmKind = kIemNativeVarKind_VarRef;
7143	}
7144	AssertStmt(pReNative->Core.aVars[idxVar].idxReg == UINT8_MAX, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_VAR_IPE_2));
7145
7146	pReNative->Core.aVars[idxVar].u.idxRefVar = idxOtherVar; /* unpacked */
7147
7148	/* Update the other variable, ensure it's a stack variable. */
7149	/** @todo handle variables with const values... that'll go boom now. */
7150	pReNative->Core.aVars[idxOtherVar].idxReferrerVar = idxVar;
7151	iemNativeVarSetKindToStack(pReNative, IEMNATIVE_VAR_IDX_PACK(idxOtherVar));
7152	}
7153
7154
7155	/**
7156	* Sets the variable to a reference (pointer) to a guest register reference.
7157	*
7158	* This does not require stack storage as we know the value and can always
7159	* reload it. Loading is postponed till needed.
7160	*
7161	* @param pReNative The recompiler state.
7162	* @param idxVar The variable.
7163	* @param enmRegClass The class guest registers to reference.
7164	* @param idxReg The register within @a enmRegClass to reference.
7165	*
7166	* @throws VERR_IEM_VAR_IPE_2
7167	*/
7168	DECL_HIDDEN_THROW(void) iemNativeVarSetKindToGstRegRef(PIEMRECOMPILERSTATE pReNative, uint8_t idxVar,
7169	IEMNATIVEGSTREGREF enmRegClass, uint8_t idxReg)
7170	{
7171	IEMNATIVE_ASSERT_VAR_IDX(pReNative, idxVar);
7172	PIEMNATIVEVAR const pVar = &pReNative->Core.aVars[IEMNATIVE_VAR_IDX_UNPACK(idxVar)];
7173
7174	if (pVar->enmKind != kIemNativeVarKind_GstRegRef)
7175	{
7176	/* Only simple transitions for now. */
7177	AssertStmt(pVar->enmKind == kIemNativeVarKind_Invalid, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_VAR_IPE_2));
7178	pVar->enmKind = kIemNativeVarKind_GstRegRef;
7179	}
7180	AssertStmt(pVar->idxReg == UINT8_MAX, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_VAR_IPE_2));
7181
7182	pVar->u.GstRegRef.enmClass = enmRegClass;
7183	pVar->u.GstRegRef.idx = idxReg;
7184	}
7185
7186
7187	DECL_HIDDEN_THROW(uint8_t) iemNativeArgAlloc(PIEMRECOMPILERSTATE pReNative, uint8_t iArgNo, uint8_t cbType)
7188	{
7189	return IEMNATIVE_VAR_IDX_PACK(iemNativeArgAllocInt(pReNative, iArgNo, cbType));
7190	}
7191
7192
7193	DECL_HIDDEN_THROW(uint8_t) iemNativeArgAllocConst(PIEMRECOMPILERSTATE pReNative, uint8_t iArgNo, uint8_t cbType, uint64_t uValue)
7194	{
7195	uint8_t const idxVar = IEMNATIVE_VAR_IDX_PACK(iemNativeArgAllocInt(pReNative, iArgNo, cbType));
7196
7197	/* Since we're using a generic uint64_t value type, we must truncate it if
7198	the variable is smaller otherwise we may end up with too large value when
7199	scaling up a imm8 w/ sign-extension.
7200
7201	This caused trouble with a "add bx, 0xffff" instruction (around f000:ac60
7202	in the bios, bx=1) when running on arm, because clang expect 16-bit
7203	register parameters to have bits 16 and up set to zero. Instead of
7204	setting x1 = 0xffff we ended up with x1 = 0xffffffffffffff and the wrong
7205	CF value in the result. */
7206	switch (cbType)
7207	{
7208	case sizeof(uint8_t): uValue &= UINT64_C(0xff); break;
7209	case sizeof(uint16_t): uValue &= UINT64_C(0xffff); break;
7210	case sizeof(uint32_t): uValue &= UINT64_C(0xffffffff); break;
7211	}
7212	iemNativeVarSetKindToConst(pReNative, idxVar, uValue);
7213	return idxVar;
7214	}
7215
7216
7217	DECL_HIDDEN_THROW(uint8_t) iemNativeArgAllocLocalRef(PIEMRECOMPILERSTATE pReNative, uint8_t iArgNo, uint8_t idxOtherVar)
7218	{
7219	IEMNATIVE_ASSERT_VAR_IDX(pReNative, idxOtherVar);
7220	idxOtherVar = IEMNATIVE_VAR_IDX_UNPACK(idxOtherVar);
7221	AssertStmt( idxOtherVar < RT_ELEMENTS(pReNative->Core.aVars)
7222	&& (pReNative->Core.bmVars & RT_BIT_32(idxOtherVar))
7223	&& pReNative->Core.aVars[idxOtherVar].uArgNo == UINT8_MAX,
7224	IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_VAR_IPE_1));
7225
7226	uint8_t const idxArgVar = iemNativeArgAlloc(pReNative, iArgNo, sizeof(uintptr_t));
7227	iemNativeVarSetKindToLocalRef(pReNative, IEMNATIVE_VAR_IDX_UNPACK(idxArgVar), idxOtherVar);
7228	return idxArgVar;
7229	}
7230
7231
7232	DECL_HIDDEN_THROW(uint8_t) iemNativeVarAlloc(PIEMRECOMPILERSTATE pReNative, uint8_t cbType)
7233	{
7234	uint8_t const idxVar = IEMNATIVE_VAR_IDX_PACK(iemNativeVarAllocInt(pReNative, cbType));
7235	/* Don't set to stack now, leave that to the first use as for instance
7236	IEM_MC_CALC_RM_EFF_ADDR may produce a const/immediate result (esp. in DOS). */
7237	return idxVar;
7238	}
7239
7240
7241	DECL_HIDDEN_THROW(uint8_t) iemNativeVarAllocConst(PIEMRECOMPILERSTATE pReNative, uint8_t cbType, uint64_t uValue)
7242	{
7243	uint8_t const idxVar = IEMNATIVE_VAR_IDX_PACK(iemNativeVarAllocInt(pReNative, cbType));
7244
7245	/* Since we're using a generic uint64_t value type, we must truncate it if
7246	the variable is smaller otherwise we may end up with too large value when
7247	scaling up a imm8 w/ sign-extension. */
7248	switch (cbType)
7249	{
7250	case sizeof(uint8_t): uValue &= UINT64_C(0xff); break;
7251	case sizeof(uint16_t): uValue &= UINT64_C(0xffff); break;
7252	case sizeof(uint32_t): uValue &= UINT64_C(0xffffffff); break;
7253	}
7254	iemNativeVarSetKindToConst(pReNative, idxVar, uValue);
7255	return idxVar;
7256	}
7257
7258
7259	/**
7260	* Makes sure variable @a idxVar has a register assigned to it and that it stays
7261	* fixed till we call iemNativeVarRegisterRelease.
7262	*
7263	* @returns The host register number.
7264	* @param pReNative The recompiler state.
7265	* @param idxVar The variable.
7266	* @param poff Pointer to the instruction buffer offset.
7267	* In case a register needs to be freed up or the value
7268	* loaded off the stack.
7269	* @param fInitialized Set if the variable must already have been initialized.
7270	* Will throw VERR_IEM_VAR_NOT_INITIALIZED if this is not
7271	* the case.
7272	* @param idxRegPref Preferred register number or UINT8_MAX.
7273	*/
7274	DECL_HIDDEN_THROW(uint8_t) iemNativeVarRegisterAcquire(PIEMRECOMPILERSTATE pReNative, uint8_t idxVar, uint32_t *poff,
7275	bool fInitialized /= false/, uint8_t idxRegPref /= UINT8_MAX/)
7276	{
7277	IEMNATIVE_ASSERT_VAR_IDX(pReNative, idxVar);
7278	PIEMNATIVEVAR const pVar = &pReNative->Core.aVars[IEMNATIVE_VAR_IDX_UNPACK(idxVar)];
7279	Assert(pVar->cbVar <= 8);
7280	Assert(!pVar->fRegAcquired);
7281
7282	uint8_t idxReg = pVar->idxReg;
7283	if (idxReg < RT_ELEMENTS(pReNative->Core.aHstRegs))
7284	{
7285	Assert( pVar->enmKind > kIemNativeVarKind_Invalid
7286	&& pVar->enmKind < kIemNativeVarKind_End);
7287	pVar->fRegAcquired = true;
7288	return idxReg;
7289	}
7290
7291	/*
7292	* If the kind of variable has not yet been set, default to 'stack'.
7293	*/
7294	Assert( pVar->enmKind >= kIemNativeVarKind_Invalid
7295	&& pVar->enmKind < kIemNativeVarKind_End);
7296	if (pVar->enmKind == kIemNativeVarKind_Invalid)
7297	iemNativeVarSetKindToStack(pReNative, idxVar);
7298
7299	/*
7300	* We have to allocate a register for the variable, even if its a stack one
7301	* as we don't know if there are modification being made to it before its
7302	* finalized (todo: analyze and insert hints about that?).
7303	*
7304	* If we can, we try get the correct register for argument variables. This
7305	* is assuming that most argument variables are fetched as close as possible
7306	* to the actual call, so that there aren't any interfering hidden calls
7307	* (memory accesses, etc) inbetween.
7308	*
7309	* If we cannot or it's a variable, we make sure no argument registers
7310	* that will be used by this MC block will be allocated here, and we always
7311	* prefer non-volatile registers to avoid needing to spill stuff for internal
7312	* call.
7313	*/
7314	/** @todo Detect too early argument value fetches and warn about hidden
7315	* calls causing less optimal code to be generated in the python script. */
7316
7317	uint8_t const uArgNo = pVar->uArgNo;
7318	if ( uArgNo < RT_ELEMENTS(g_aidxIemNativeCallRegs)
7319	&& !(pReNative->Core.bmHstRegs & RT_BIT_32(g_aidxIemNativeCallRegs[uArgNo])))
7320	{
7321	idxReg = g_aidxIemNativeCallRegs[uArgNo];
7322	iemNativeRegClearGstRegShadowing(pReNative, idxReg, *poff);
7323	Log11(("iemNativeVarRegisterAcquire: idxVar=%#x idxReg=%u (matching arg %u)\n", idxVar, idxReg, uArgNo));
7324	}
7325	else if ( idxRegPref >= RT_ELEMENTS(pReNative->Core.aHstRegs)
7326	\|\| (pReNative->Core.bmHstRegs & RT_BIT_32(idxRegPref)))
7327	{
7328	uint32_t const fNotArgsMask = ~g_afIemNativeCallRegs[RT_MIN(pReNative->cArgs, IEMNATIVE_CALL_ARG_GREG_COUNT)];
7329	uint32_t const fRegs = ~pReNative->Core.bmHstRegs
7330	& ~pReNative->Core.bmHstRegsWithGstShadow
7331	& (~IEMNATIVE_REG_FIXED_MASK & IEMNATIVE_HST_GREG_MASK)
7332	& fNotArgsMask;
7333	if (fRegs)
7334	{
7335	/* Pick from the top as that both arm64 and amd64 have a block of non-volatile registers there. */
7336	idxReg = (uint8_t)ASMBitLastSetU32( fRegs & ~IEMNATIVE_CALL_VOLATILE_GREG_MASK
7337	? fRegs & ~IEMNATIVE_CALL_VOLATILE_GREG_MASK : fRegs) - 1;
7338	Assert(pReNative->Core.aHstRegs[idxReg].fGstRegShadows == 0);
7339	Assert(!(pReNative->Core.bmHstRegsWithGstShadow & RT_BIT_32(idxReg)));
7340	Log11(("iemNativeVarRegisterAcquire: idxVar=%#x idxReg=%u (uArgNo=%u)\n", idxVar, idxReg, uArgNo));
7341	}
7342	else
7343	{
7344	idxReg = iemNativeRegAllocFindFree(pReNative, poff, false /fPreferVolatile/,
7345	IEMNATIVE_HST_GREG_MASK & ~IEMNATIVE_REG_FIXED_MASK & fNotArgsMask);
7346	AssertStmt(idxReg != UINT8_MAX, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_REG_ALLOCATOR_NO_FREE_VAR));
7347	Log11(("iemNativeVarRegisterAcquire: idxVar=%#x idxReg=%u (slow, uArgNo=%u)\n", idxVar, idxReg, uArgNo));
7348	}
7349	}
7350	else
7351	{
7352	idxReg = idxRegPref;
7353	iemNativeRegClearGstRegShadowing(pReNative, idxReg, *poff);
7354	Log11(("iemNativeVarRegisterAcquire: idxVar=%#x idxReg=%u (preferred)\n", idxVar, idxReg));
7355	}
7356	iemNativeRegMarkAllocated(pReNative, idxReg, kIemNativeWhat_Var, idxVar);
7357	pVar->idxReg = idxReg;
7358
7359	/*
7360	* Load it off the stack if we've got a stack slot.
7361	*/
7362	uint8_t const idxStackSlot = pVar->idxStackSlot;
7363	if (idxStackSlot < IEMNATIVE_FRAME_VAR_SLOTS)
7364	{
7365	Assert(fInitialized);
7366	int32_t const offDispBp = iemNativeStackCalcBpDisp(idxStackSlot);
7367	switch (pVar->cbVar)
7368	{
7369	case 1: poff = iemNativeEmitLoadGprByBpU8( pReNative, poff, idxReg, offDispBp); break;
7370	case 2: poff = iemNativeEmitLoadGprByBpU16(pReNative, poff, idxReg, offDispBp); break;
7371	case 3: AssertFailed(); RT_FALL_THRU();
7372	case 4: poff = iemNativeEmitLoadGprByBpU32(pReNative, poff, idxReg, offDispBp); break;
7373	default: AssertFailed(); RT_FALL_THRU();
7374	case 8: poff = iemNativeEmitLoadGprByBp( pReNative, poff, idxReg, offDispBp); break;
7375	}
7376	}
7377	else
7378	{
7379	Assert(idxStackSlot == UINT8_MAX);
7380	AssertStmt(!fInitialized, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_VAR_NOT_INITIALIZED));
7381	}
7382	pVar->fRegAcquired = true;
7383	return idxReg;
7384	}
7385
7386
7387	/**
7388	* The value of variable @a idxVar will be written in full to the @a enmGstReg
7389	* guest register.
7390	*
7391	* This function makes sure there is a register for it and sets it to be the
7392	* current shadow copy of @a enmGstReg.
7393	*
7394	* @returns The host register number.
7395	* @param pReNative The recompiler state.
7396	* @param idxVar The variable.
7397	* @param enmGstReg The guest register this variable will be written to
7398	* after this call.
7399	* @param poff Pointer to the instruction buffer offset.
7400	* In case a register needs to be freed up or if the
7401	* variable content needs to be loaded off the stack.
7402	*
7403	* @note We DO NOT expect @a idxVar to be an argument variable,
7404	* because we can only in the commit stage of an instruction when this
7405	* function is used.
7406	*/
7407	DECL_HIDDEN_THROW(uint8_t)
7408	iemNativeVarRegisterAcquireForGuestReg(PIEMRECOMPILERSTATE pReNative, uint8_t idxVar, IEMNATIVEGSTREG enmGstReg, uint32_t *poff)
7409	{
7410	IEMNATIVE_ASSERT_VAR_IDX(pReNative, idxVar);
7411	PIEMNATIVEVAR const pVar = &pReNative->Core.aVars[IEMNATIVE_VAR_IDX_UNPACK(idxVar)];
7412	Assert(!pVar->fRegAcquired);
7413	AssertMsgStmt( pVar->cbVar <= 8
7414	&& ( pVar->enmKind == kIemNativeVarKind_Immediate
7415	\|\| pVar->enmKind == kIemNativeVarKind_Stack),
7416	("idxVar=%#x cbVar=%d enmKind=%d enmGstReg=%s\n", idxVar, pVar->cbVar,
7417	pVar->enmKind, g_aGstShadowInfo[enmGstReg].pszName),
7418	IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_VAR_IPE_6));
7419
7420	/*
7421	* This shouldn't ever be used for arguments, unless it's in a weird else
7422	* branch that doesn't do any calling and even then it's questionable.
7423	*
7424	* However, in case someone writes crazy wrong MC code and does register
7425	* updates before making calls, just use the regular register allocator to
7426	* ensure we get a register suitable for the intended argument number.
7427	*/
7428	AssertStmt(pVar->uArgNo == UINT8_MAX, iemNativeVarRegisterAcquire(pReNative, idxVar, poff));
7429
7430	/*
7431	* If there is already a register for the variable, we transfer/set the
7432	* guest shadow copy assignment to it.
7433	*/
7434	uint8_t idxReg = pVar->idxReg;
7435	if (idxReg < RT_ELEMENTS(pReNative->Core.aHstRegs))
7436	{
7437	if (pReNative->Core.bmGstRegShadows & RT_BIT_64(enmGstReg))
7438	{
7439	uint8_t const idxRegOld = pReNative->Core.aidxGstRegShadows[enmGstReg];
7440	iemNativeRegTransferGstRegShadowing(pReNative, idxRegOld, idxReg, enmGstReg, *poff);
7441	Log12(("iemNativeVarRegisterAcquireForGuestReg: Moved %s for guest %s into %s for full write\n",
7442	g_apszIemNativeHstRegNames[idxRegOld], g_aGstShadowInfo[enmGstReg].pszName, g_apszIemNativeHstRegNames[idxReg]));
7443	}
7444	else
7445	{
7446	iemNativeRegMarkAsGstRegShadow(pReNative, idxReg, enmGstReg, *poff);
7447	Log12(("iemNativeVarRegisterAcquireForGuestReg: Marking %s as copy of guest %s (full write)\n",
7448	g_apszIemNativeHstRegNames[idxReg], g_aGstShadowInfo[enmGstReg].pszName));
7449	}
7450	/** @todo figure this one out. We need some way of making sure the register isn't
7451	* modified after this point, just in case we start writing crappy MC code. */
7452	pVar->enmGstReg = enmGstReg;
7453	pVar->fRegAcquired = true;
7454	return idxReg;
7455	}
7456	Assert(pVar->uArgNo == UINT8_MAX);
7457
7458	/*
7459	* Because this is supposed to be the commit stage, we're just tag along with the
7460	* temporary register allocator and upgrade it to a variable register.
7461	*/
7462	idxReg = iemNativeRegAllocTmpForGuestReg(pReNative, poff, enmGstReg, kIemNativeGstRegUse_ForFullWrite);
7463	Assert(pReNative->Core.aHstRegs[idxReg].enmWhat == kIemNativeWhat_Tmp);
7464	Assert(pReNative->Core.aHstRegs[idxReg].idxVar == UINT8_MAX);
7465	pReNative->Core.aHstRegs[idxReg].enmWhat = kIemNativeWhat_Var;
7466	pReNative->Core.aHstRegs[idxReg].idxVar = idxVar;
7467	pVar->idxReg = idxReg;
7468
7469	/*
7470	* Now we need to load the register value.
7471	*/
7472	if (pVar->enmKind == kIemNativeVarKind_Immediate)
7473	poff = iemNativeEmitLoadGprImm64(pReNative, poff, idxReg, pVar->u.uValue);
7474	else
7475	{
7476	uint8_t const idxStackSlot = iemNativeVarGetStackSlot(pReNative, idxVar);
7477	int32_t const offDispBp = iemNativeStackCalcBpDisp(idxStackSlot);
7478	switch (pVar->cbVar)
7479	{
7480	case sizeof(uint64_t):
7481	poff = iemNativeEmitLoadGprByBp(pReNative, poff, idxReg, offDispBp);
7482	break;
7483	case sizeof(uint32_t):
7484	poff = iemNativeEmitLoadGprByBpU32(pReNative, poff, idxReg, offDispBp);
7485	break;
7486	case sizeof(uint16_t):
7487	poff = iemNativeEmitLoadGprByBpU16(pReNative, poff, idxReg, offDispBp);
7488	break;
7489	case sizeof(uint8_t):
7490	poff = iemNativeEmitLoadGprByBpU8(pReNative, poff, idxReg, offDispBp);
7491	break;
7492	default:
7493	AssertFailedStmt(IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_VAR_IPE_6));
7494	}
7495	}
7496
7497	pVar->fRegAcquired = true;
7498	return idxReg;
7499	}
7500
7501
7502	/**
7503	* Emit code to save volatile registers prior to a call to a helper (TLB miss).
7504	*
7505	* This is used together with iemNativeVarRestoreVolatileRegsPostHlpCall() and
7506	* optionally iemNativeRegRestoreGuestShadowsInVolatileRegs() to bypass the
7507	* requirement of flushing anything in volatile host registers when making a
7508	* call.
7509	*
7510	* @returns New @a off value.
7511	* @param pReNative The recompiler state.
7512	* @param off The code buffer position.
7513	* @param fHstRegsNotToSave Set of registers not to save & restore.
7514	*/
7515	DECL_HIDDEN_THROW(uint32_t)
7516	iemNativeVarSaveVolatileRegsPreHlpCall(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint32_t fHstRegsNotToSave)
7517	{
7518	uint32_t fHstRegs = pReNative->Core.bmHstRegs & IEMNATIVE_CALL_VOLATILE_GREG_MASK & ~fHstRegsNotToSave;
7519	if (fHstRegs)
7520	{
7521	do
7522	{
7523	unsigned int const idxHstReg = ASMBitFirstSetU32(fHstRegs) - 1;
7524	fHstRegs &= ~RT_BIT_32(idxHstReg);
7525
7526	if (pReNative->Core.aHstRegs[idxHstReg].enmWhat == kIemNativeWhat_Var)
7527	{
7528	uint8_t const idxVar = pReNative->Core.aHstRegs[idxHstReg].idxVar;
7529	IEMNATIVE_ASSERT_VAR_IDX(pReNative, idxVar);
7530	AssertStmt( IEMNATIVE_VAR_IDX_UNPACK(idxVar) < RT_ELEMENTS(pReNative->Core.aVars)
7531	&& (pReNative->Core.bmVars & RT_BIT_32(IEMNATIVE_VAR_IDX_UNPACK(idxVar)))
7532	&& pReNative->Core.aVars[IEMNATIVE_VAR_IDX_UNPACK(idxVar)].idxReg == idxHstReg,
7533	IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_VAR_IPE_12));
7534	switch (pReNative->Core.aVars[IEMNATIVE_VAR_IDX_UNPACK(idxVar)].enmKind)
7535	{
7536	case kIemNativeVarKind_Stack:
7537	{
7538	/* Temporarily spill the variable register. */
7539	uint8_t const idxStackSlot = iemNativeVarGetStackSlot(pReNative, idxVar);
7540	Log12(("iemNativeVarSaveVolatileRegsPreHlpCall: spilling idxVar=%#x/idxReg=%d onto the stack (slot %#x bp+%d, off=%#x)\n",
7541	idxVar, idxHstReg, idxStackSlot, iemNativeStackCalcBpDisp(idxStackSlot), off));
7542	off = iemNativeEmitStoreGprByBp(pReNative, off, iemNativeStackCalcBpDisp(idxStackSlot), idxHstReg);
7543	continue;
7544	}
7545
7546	case kIemNativeVarKind_Immediate:
7547	case kIemNativeVarKind_VarRef:
7548	case kIemNativeVarKind_GstRegRef:
7549	/* It is weird to have any of these loaded at this point. */
7550	AssertFailedStmt(IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_VAR_IPE_13));
7551	continue;
7552
7553	case kIemNativeVarKind_End:
7554	case kIemNativeVarKind_Invalid:
7555	break;
7556	}
7557	AssertFailed();
7558	}
7559	else
7560	{
7561	/*
7562	* Allocate a temporary stack slot and spill the register to it.
7563	*/
7564	unsigned const idxStackSlot = ASMBitLastSetU32(~pReNative->Core.bmStack) - 1;
7565	AssertStmt(idxStackSlot < IEMNATIVE_FRAME_VAR_SLOTS,
7566	IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_VAR_OUT_OF_STACK_SLOTS));
7567	pReNative->Core.bmStack \|= RT_BIT_32(idxStackSlot);
7568	pReNative->Core.aHstRegs[idxHstReg].idxStackSlot = (uint8_t)idxStackSlot;
7569	Log12(("iemNativeVarSaveVolatileRegsPreHlpCall: spilling idxReg=%d onto the stack (slot %#x bp+%d, off=%#x)\n",
7570	idxHstReg, idxStackSlot, iemNativeStackCalcBpDisp(idxStackSlot), off));
7571	off = iemNativeEmitStoreGprByBp(pReNative, off, iemNativeStackCalcBpDisp(idxStackSlot), idxHstReg);
7572	}
7573	} while (fHstRegs);
7574	}
7575	return off;
7576	}
7577
7578
7579	/**
7580	* Emit code to restore volatile registers after to a call to a helper.
7581	*
7582	* @returns New @a off value.
7583	* @param pReNative The recompiler state.
7584	* @param off The code buffer position.
7585	* @param fHstRegsNotToSave Set of registers not to save & restore.
7586	* @see iemNativeVarSaveVolatileRegsPreHlpCall(),
7587	* iemNativeRegRestoreGuestShadowsInVolatileRegs()
7588	*/
7589	DECL_HIDDEN_THROW(uint32_t)
7590	iemNativeVarRestoreVolatileRegsPostHlpCall(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint32_t fHstRegsNotToSave)
7591	{
7592	uint32_t fHstRegs = pReNative->Core.bmHstRegs & IEMNATIVE_CALL_VOLATILE_GREG_MASK & ~fHstRegsNotToSave;
7593	if (fHstRegs)
7594	{
7595	do
7596	{
7597	unsigned int const idxHstReg = ASMBitFirstSetU32(fHstRegs) - 1;
7598	fHstRegs &= ~RT_BIT_32(idxHstReg);
7599
7600	if (pReNative->Core.aHstRegs[idxHstReg].enmWhat == kIemNativeWhat_Var)
7601	{
7602	uint8_t const idxVar = pReNative->Core.aHstRegs[idxHstReg].idxVar;
7603	IEMNATIVE_ASSERT_VAR_IDX(pReNative, idxVar);
7604	AssertStmt( IEMNATIVE_VAR_IDX_UNPACK(idxVar) < RT_ELEMENTS(pReNative->Core.aVars)
7605	&& (pReNative->Core.bmVars & RT_BIT_32(IEMNATIVE_VAR_IDX_UNPACK(idxVar)))
7606	&& pReNative->Core.aVars[IEMNATIVE_VAR_IDX_UNPACK(idxVar)].idxReg == idxHstReg,
7607	IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_VAR_IPE_12));
7608	switch (pReNative->Core.aVars[IEMNATIVE_VAR_IDX_UNPACK(idxVar)].enmKind)
7609	{
7610	case kIemNativeVarKind_Stack:
7611	{
7612	/* Unspill the variable register. */
7613	uint8_t const idxStackSlot = iemNativeVarGetStackSlot(pReNative, idxVar);
7614	Log12(("iemNativeVarRestoreVolatileRegsPostHlpCall: unspilling idxVar=%#x/idxReg=%d (slot %#x bp+%d, off=%#x)\n",
7615	idxVar, idxHstReg, idxStackSlot, iemNativeStackCalcBpDisp(idxStackSlot), off));
7616	off = iemNativeEmitLoadGprByBp(pReNative, off, idxHstReg, iemNativeStackCalcBpDisp(idxStackSlot));
7617	continue;
7618	}
7619
7620	case kIemNativeVarKind_Immediate:
7621	case kIemNativeVarKind_VarRef:
7622	case kIemNativeVarKind_GstRegRef:
7623	/* It is weird to have any of these loaded at this point. */
7624	AssertFailedStmt(IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_VAR_IPE_13));
7625	continue;
7626
7627	case kIemNativeVarKind_End:
7628	case kIemNativeVarKind_Invalid:
7629	break;
7630	}
7631	AssertFailed();
7632	}
7633	else
7634	{
7635	/*
7636	* Restore from temporary stack slot.
7637	*/
7638	uint8_t const idxStackSlot = pReNative->Core.aHstRegs[idxHstReg].idxStackSlot;
7639	AssertContinue(idxStackSlot < IEMNATIVE_FRAME_VAR_SLOTS && (pReNative->Core.bmStack & RT_BIT_32(idxStackSlot)));
7640	pReNative->Core.bmStack &= ~RT_BIT_32(idxStackSlot);
7641	pReNative->Core.aHstRegs[idxHstReg].idxStackSlot = UINT8_MAX;
7642
7643	off = iemNativeEmitLoadGprByBp(pReNative, off, idxHstReg, iemNativeStackCalcBpDisp(idxStackSlot));
7644	}
7645	} while (fHstRegs);
7646	}
7647	return off;
7648	}
7649
7650
7651	/**
7652	* Worker that frees the stack slots for variable @a idxVar if any allocated.
7653	*
7654	* This is used both by iemNativeVarFreeOneWorker and iemNativeEmitCallCommon.
7655	*
7656	* ASSUMES that @a idxVar is valid and unpacked.
7657	*/
7658	DECL_FORCE_INLINE(void) iemNativeVarFreeStackSlots(PIEMRECOMPILERSTATE pReNative, uint8_t idxVar)
7659	{
7660	Assert(idxVar < RT_ELEMENTS(pReNative->Core.aVars)); /* unpacked! */
7661	uint8_t const idxStackSlot = pReNative->Core.aVars[idxVar].idxStackSlot;
7662	if (idxStackSlot < IEMNATIVE_FRAME_VAR_SLOTS)
7663	{
7664	uint8_t const cbVar = pReNative->Core.aVars[idxVar].cbVar;
7665	uint8_t const cSlots = (cbVar + sizeof(uint64_t) - 1) / sizeof(uint64_t);
7666	uint32_t const fAllocMask = (uint32_t)(RT_BIT_32(cSlots) - 1U);
7667	Assert(cSlots > 0);
7668	Assert(((pReNative->Core.bmStack >> idxStackSlot) & fAllocMask) == fAllocMask);
7669	Log11(("iemNativeVarFreeStackSlots: idxVar=%d/%#x iSlot=%#x/%#x (cbVar=%#x)\n",
7670	idxVar, IEMNATIVE_VAR_IDX_PACK(idxVar), idxStackSlot, fAllocMask, cbVar));
7671	pReNative->Core.bmStack &= ~(fAllocMask << idxStackSlot);
7672	pReNative->Core.aVars[idxVar].idxStackSlot = UINT8_MAX;
7673	}
7674	else
7675	Assert(idxStackSlot == UINT8_MAX);
7676	}
7677
7678
7679	/**
7680	* Worker that frees a single variable.
7681	*
7682	* ASSUMES that @a idxVar is valid and unpacked.
7683	*/
7684	DECLHIDDEN(void) iemNativeVarFreeOneWorker(PIEMRECOMPILERSTATE pReNative, uint8_t idxVar)
7685	{
7686	Assert( pReNative->Core.aVars[idxVar].enmKind >= kIemNativeVarKind_Invalid /* Including invalid as we may have unused */
7687	&& pReNative->Core.aVars[idxVar].enmKind < kIemNativeVarKind_End); /* variables in conditional branches. */
7688	Assert(!pReNative->Core.aVars[idxVar].fRegAcquired);
7689
7690	/* Free the host register first if any assigned. */
7691	uint8_t const idxHstReg = pReNative->Core.aVars[idxVar].idxReg;
7692	if (idxHstReg < RT_ELEMENTS(pReNative->Core.aHstRegs))
7693	{
7694	Assert(pReNative->Core.aHstRegs[idxHstReg].idxVar == IEMNATIVE_VAR_IDX_PACK(idxVar));
7695	pReNative->Core.aHstRegs[idxHstReg].idxVar = UINT8_MAX;
7696	pReNative->Core.bmHstRegs &= ~RT_BIT_32(idxHstReg);
7697	}
7698
7699	/* Free argument mapping. */
7700	uint8_t const uArgNo = pReNative->Core.aVars[idxVar].uArgNo;
7701	if (uArgNo < RT_ELEMENTS(pReNative->Core.aidxArgVars))
7702	pReNative->Core.aidxArgVars[uArgNo] = UINT8_MAX;
7703
7704	/* Free the stack slots. */
7705	iemNativeVarFreeStackSlots(pReNative, idxVar);
7706
7707	/* Free the actual variable. */
7708	pReNative->Core.aVars[idxVar].enmKind = kIemNativeVarKind_Invalid;
7709	pReNative->Core.bmVars &= ~RT_BIT_32(idxVar);
7710	}
7711
7712
7713	/**
7714	* Worker for iemNativeVarFreeAll that's called when there is anything to do.
7715	*/
7716	DECLHIDDEN(void) iemNativeVarFreeAllSlow(PIEMRECOMPILERSTATE pReNative, uint32_t bmVars)
7717	{
7718	while (bmVars != 0)
7719	{
7720	uint8_t const idxVar = ASMBitFirstSetU32(bmVars) - 1;
7721	bmVars &= ~RT_BIT_32(idxVar);
7722
7723	#if 1 /** @todo optimize by simplifying this later... */
7724	iemNativeVarFreeOneWorker(pReNative, idxVar);
7725	#else
7726	/* Only need to free the host register, the rest is done as bulk updates below. */
7727	uint8_t const idxHstReg = pReNative->Core.aVars[idxVar].idxReg;
7728	if (idxHstReg < RT_ELEMENTS(pReNative->Core.aHstRegs))
7729	{
7730	Assert(pReNative->Core.aHstRegs[idxHstReg].idxVar == IEMNATIVE_VAR_IDX_PACK(idxVar));
7731	pReNative->Core.aHstRegs[idxHstReg].idxVar = UINT8_MAX;
7732	pReNative->Core.bmHstRegs &= ~RT_BIT_32(idxHstReg);
7733	}
7734	#endif
7735	}
7736	#if 0 /** @todo optimize by simplifying this later... */
7737	pReNative->Core.bmVars = 0;
7738	pReNative->Core.bmStack = 0;
7739	pReNative->Core.u64ArgVars = UINT64_MAX;
7740	#endif
7741	}
7742
7743
7744
7745	/*********************************************************************************************************************************
7746	* Emitters for IEM_MC_CALL_CIMPL_XXX *
7747	*********************************************************************************************************************************/
7748
7749	/**
7750	* Emits code to load a reference to the given guest register into @a idxGprDst.
7751	*/
7752	DECL_INLINE_THROW(uint32_t)
7753	iemNativeEmitLeaGprByGstRegRef(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint8_t idxGprDst,
7754	IEMNATIVEGSTREGREF enmClass, uint8_t idxRegInClass)
7755	{
7756	#ifdef IEMNATIVE_WITH_DELAYED_PC_UPDATING
7757	/** @todo If we ever gonna allow referencing the RIP register we need to update guest value here. */
7758	#endif
7759
7760	/*
7761	* Get the offset relative to the CPUMCTX structure.
7762	*/
7763	uint32_t offCpumCtx;
7764	switch (enmClass)
7765	{
7766	case kIemNativeGstRegRef_Gpr:
7767	Assert(idxRegInClass < 16);
7768	offCpumCtx = RT_UOFFSETOF_DYN(CPUMCTX, aGRegs[idxRegInClass]);
7769	break;
7770
7771	case kIemNativeGstRegRef_GprHighByte: /*< AH, CH, DH, BH/
7772	Assert(idxRegInClass < 4);
7773	offCpumCtx = RT_UOFFSETOF_DYN(CPUMCTX, aGRegs[0].bHi) + idxRegInClass * sizeof(CPUMCTXGREG);
7774	break;
7775
7776	case kIemNativeGstRegRef_EFlags:
7777	Assert(idxRegInClass == 0);
7778	offCpumCtx = RT_UOFFSETOF(CPUMCTX, eflags);
7779	break;
7780
7781	case kIemNativeGstRegRef_MxCsr:
7782	Assert(idxRegInClass == 0);
7783	offCpumCtx = RT_UOFFSETOF(CPUMCTX, XState.x87.MXCSR);
7784	break;
7785
7786	case kIemNativeGstRegRef_FpuReg:
7787	Assert(idxRegInClass < 8);
7788	AssertFailed(); /** @todo what kind of indexing? */
7789	offCpumCtx = RT_UOFFSETOF_DYN(CPUMCTX, XState.x87.aRegs[idxRegInClass]);
7790	break;
7791
7792	case kIemNativeGstRegRef_MReg:
7793	Assert(idxRegInClass < 8);
7794	AssertFailed(); /** @todo what kind of indexing? */
7795	offCpumCtx = RT_UOFFSETOF_DYN(CPUMCTX, XState.x87.aRegs[idxRegInClass]);
7796	break;
7797
7798	case kIemNativeGstRegRef_XReg:
7799	Assert(idxRegInClass < 16);
7800	offCpumCtx = RT_UOFFSETOF_DYN(CPUMCTX, XState.x87.aXMM[idxRegInClass]);
7801	break;
7802
7803	default:
7804	AssertFailedStmt(IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_VAR_IPE_5));
7805	}
7806
7807	/*
7808	* Load the value into the destination register.
7809	*/
7810	#ifdef RT_ARCH_AMD64
7811	off = iemNativeEmitLeaGprByVCpu(pReNative, off, idxGprDst, offCpumCtx + RT_UOFFSETOF(VMCPUCC, cpum.GstCtx));
7812
7813	#elif defined(RT_ARCH_ARM64)
7814	uint32_t * const pu32CodeBuf = iemNativeInstrBufEnsure(pReNative, off, 2);
7815	Assert(offCpumCtx < 4096);
7816	pu32CodeBuf[off++] = Armv8A64MkInstrAddSubUImm12(false /fSub/, idxGprDst, IEMNATIVE_REG_FIXED_PCPUMCTX, offCpumCtx);
7817
7818	#else
7819	# error "Port me!"
7820	#endif
7821
7822	return off;
7823	}
7824
7825
7826	/**
7827	* Common code for CIMPL and AIMPL calls.
7828	*
7829	* These are calls that uses argument variables and such. They should not be
7830	* confused with internal calls required to implement an MC operation,
7831	* like a TLB load and similar.
7832	*
7833	* Upon return all that is left to do is to load any hidden arguments and
7834	* perform the call. All argument variables are freed.
7835	*
7836	* @returns New code buffer offset; throws VBox status code on error.
7837	* @param pReNative The native recompile state.
7838	* @param off The code buffer offset.
7839	* @param cArgs The total nubmer of arguments (includes hidden
7840	* count).
7841	* @param cHiddenArgs The number of hidden arguments. The hidden
7842	* arguments must not have any variable declared for
7843	* them, whereas all the regular arguments must
7844	* (tstIEMCheckMc ensures this).
7845	*/
7846	DECL_HIDDEN_THROW(uint32_t)
7847	iemNativeEmitCallCommon(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint8_t cArgs, uint8_t cHiddenArgs)
7848	{
7849	#ifdef VBOX_STRICT
7850	/*
7851	* Assert sanity.
7852	*/
7853	Assert(cArgs <= IEMNATIVE_CALL_MAX_ARG_COUNT);
7854	Assert(cHiddenArgs < IEMNATIVE_CALL_ARG_GREG_COUNT);
7855	for (unsigned i = 0; i < cHiddenArgs; i++)
7856	Assert(pReNative->Core.aidxArgVars[i] == UINT8_MAX);
7857	for (unsigned i = cHiddenArgs; i < cArgs; i++)
7858	{
7859	Assert(pReNative->Core.aidxArgVars[i] != UINT8_MAX); /* checked by tstIEMCheckMc.cpp */
7860	Assert(pReNative->Core.bmVars & RT_BIT_32(pReNative->Core.aidxArgVars[i]));
7861	}
7862	iemNativeRegAssertSanity(pReNative);
7863	#endif
7864
7865	/* We don't know what the called function makes use of, so flush any pending register writes. */
7866	off = iemNativeRegFlushPendingWrites(pReNative, off);
7867
7868	/*
7869	* Before we do anything else, go over variables that are referenced and
7870	* make sure they are not in a register.
7871	*/
7872	uint32_t bmVars = pReNative->Core.bmVars;
7873	if (bmVars)
7874	{
7875	do
7876	{
7877	uint8_t const idxVar = ASMBitFirstSetU32(bmVars) - 1;
7878	bmVars &= ~RT_BIT_32(idxVar);
7879
7880	if (pReNative->Core.aVars[idxVar].idxReferrerVar != UINT8_MAX)
7881	{
7882	uint8_t const idxRegOld = pReNative->Core.aVars[idxVar].idxReg;
7883	if (idxRegOld < RT_ELEMENTS(pReNative->Core.aHstRegs))
7884	{
7885	uint8_t const idxStackSlot = iemNativeVarGetStackSlot(pReNative, IEMNATIVE_VAR_IDX_PACK(idxVar));
7886	Log12(("iemNativeEmitCallCommon: spilling idxVar=%d/%#x/idxReg=%d (referred to by %d) onto the stack (slot %#x bp+%d, off=%#x)\n",
7887	idxVar, IEMNATIVE_VAR_IDX_PACK(idxVar), idxRegOld, pReNative->Core.aVars[idxVar].idxReferrerVar,
7888	idxStackSlot, iemNativeStackCalcBpDisp(idxStackSlot), off));
7889	off = iemNativeEmitStoreGprByBp(pReNative, off, iemNativeStackCalcBpDisp(idxStackSlot), idxRegOld);
7890
7891	pReNative->Core.aVars[idxVar].idxReg = UINT8_MAX;
7892	pReNative->Core.bmHstRegs &= ~RT_BIT_32(idxRegOld);
7893	pReNative->Core.bmHstRegsWithGstShadow &= ~RT_BIT_32(idxRegOld);
7894	pReNative->Core.bmGstRegShadows &= ~pReNative->Core.aHstRegs[idxRegOld].fGstRegShadows;
7895	pReNative->Core.aHstRegs[idxRegOld].fGstRegShadows = 0;
7896	}
7897	}
7898	} while (bmVars != 0);
7899	#if 0 //def VBOX_STRICT
7900	iemNativeRegAssertSanity(pReNative);
7901	#endif
7902	}
7903
7904	uint8_t const cRegArgs = RT_MIN(cArgs, RT_ELEMENTS(g_aidxIemNativeCallRegs));
7905
7906	/*
7907	* First, go over the host registers that will be used for arguments and make
7908	* sure they either hold the desired argument or are free.
7909	*/
7910	if (pReNative->Core.bmHstRegs & g_afIemNativeCallRegs[cRegArgs])
7911	{
7912	for (uint32_t i = 0; i < cRegArgs; i++)
7913	{
7914	uint8_t const idxArgReg = g_aidxIemNativeCallRegs[i];
7915	if (pReNative->Core.bmHstRegs & RT_BIT_32(idxArgReg))
7916	{
7917	if (pReNative->Core.aHstRegs[idxArgReg].enmWhat == kIemNativeWhat_Var)
7918	{
7919	uint8_t const idxVar = pReNative->Core.aHstRegs[idxArgReg].idxVar;
7920	IEMNATIVE_ASSERT_VAR_IDX(pReNative, idxVar);
7921	PIEMNATIVEVAR const pVar = &pReNative->Core.aVars[IEMNATIVE_VAR_IDX_UNPACK(idxVar)];
7922	Assert(pVar->idxReg == idxArgReg);
7923	uint8_t const uArgNo = pVar->uArgNo;
7924	if (uArgNo == i)
7925	{ /* prefect */ }
7926	/* The variable allocator logic should make sure this is impossible,
7927	except for when the return register is used as a parameter (ARM,
7928	but not x86). */
7929	#if RT_BIT_32(IEMNATIVE_CALL_RET_GREG) & IEMNATIVE_CALL_ARGS_GREG_MASK
7930	else if (idxArgReg == IEMNATIVE_CALL_RET_GREG && uArgNo != UINT8_MAX)
7931	{
7932	# ifdef IEMNATIVE_FP_OFF_STACK_ARG0
7933	# error "Implement this"
7934	# endif
7935	Assert(uArgNo < IEMNATIVE_CALL_ARG_GREG_COUNT);
7936	uint8_t const idxFinalArgReg = g_aidxIemNativeCallRegs[uArgNo];
7937	AssertStmt(!(pReNative->Core.bmHstRegs & RT_BIT_32(idxFinalArgReg)),
7938	IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_REG_IPE_10));
7939	off = iemNativeRegMoveVar(pReNative, off, idxVar, idxArgReg, idxFinalArgReg, "iemNativeEmitCallCommon");
7940	}
7941	#endif
7942	else
7943	{
7944	AssertStmt(uArgNo == UINT8_MAX, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_REG_IPE_10));
7945
7946	if (pVar->enmKind == kIemNativeVarKind_Stack)
7947	off = iemNativeRegMoveOrSpillStackVar(pReNative, off, idxVar);
7948	else
7949	{
7950	/* just free it, can be reloaded if used again */
7951	pVar->idxReg = UINT8_MAX;
7952	pReNative->Core.bmHstRegs &= ~RT_BIT_32(idxArgReg);
7953	iemNativeRegClearGstRegShadowing(pReNative, idxArgReg, off);
7954	}
7955	}
7956	}
7957	else
7958	AssertStmt(pReNative->Core.aHstRegs[idxArgReg].enmWhat == kIemNativeWhat_Arg,
7959	IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_REG_IPE_8));
7960	}
7961	}
7962	#if 0 //def VBOX_STRICT
7963	iemNativeRegAssertSanity(pReNative);
7964	#endif
7965	}
7966
7967	Assert(!(pReNative->Core.bmHstRegs & g_afIemNativeCallRegs[cHiddenArgs])); /* No variables for hidden arguments. */
7968
7969	#ifdef IEMNATIVE_FP_OFF_STACK_ARG0
7970	/*
7971	* If there are any stack arguments, make sure they are in their place as well.
7972	*
7973	* We can use IEMNATIVE_CALL_ARG0_GREG as temporary register since we'll (or
7974	* the caller) be loading it later and it must be free (see first loop).
7975	*/
7976	if (cArgs > IEMNATIVE_CALL_ARG_GREG_COUNT)
7977	{
7978	for (unsigned i = IEMNATIVE_CALL_ARG_GREG_COUNT; i < cArgs; i++)
7979	{
7980	PIEMNATIVEVAR const pVar = &pReNative->Core.aVars[pReNative->Core.aidxArgVars[i]]; /* unpacked */
7981	int32_t const offBpDisp = g_aoffIemNativeCallStackArgBpDisp[i - IEMNATIVE_CALL_ARG_GREG_COUNT];
7982	if (pVar->idxReg < RT_ELEMENTS(pReNative->Core.aHstRegs))
7983	{
7984	Assert(pVar->enmKind == kIemNativeVarKind_Stack); /* Imm as well? */
7985	off = iemNativeEmitStoreGprByBp(pReNative, off, offBpDisp, pVar->idxReg);
7986	pReNative->Core.bmHstRegs &= ~RT_BIT_32(pVar->idxReg);
7987	pVar->idxReg = UINT8_MAX;
7988	}
7989	else
7990	{
7991	/* Use ARG0 as temp for stuff we need registers for. */
7992	switch (pVar->enmKind)
7993	{
7994	case kIemNativeVarKind_Stack:
7995	{
7996	uint8_t const idxStackSlot = pVar->idxStackSlot;
7997	AssertStmt(idxStackSlot != UINT8_MAX, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_VAR_NOT_INITIALIZED));
7998	off = iemNativeEmitLoadGprByBp(pReNative, off, IEMNATIVE_CALL_ARG0_GREG /* is free */,
7999	iemNativeStackCalcBpDisp(idxStackSlot));
8000	off = iemNativeEmitStoreGprByBp(pReNative, off, offBpDisp, IEMNATIVE_CALL_ARG0_GREG);
8001	continue;
8002	}
8003
8004	case kIemNativeVarKind_Immediate:
8005	off = iemNativeEmitStoreImm64ByBp(pReNative, off, offBpDisp, pVar->u.uValue);
8006	continue;
8007
8008	case kIemNativeVarKind_VarRef:
8009	{
8010	uint8_t const idxOtherVar = pVar->u.idxRefVar; /* unpacked */
8011	Assert(idxOtherVar < RT_ELEMENTS(pReNative->Core.aVars));
8012	uint8_t const idxStackSlot = iemNativeVarGetStackSlot(pReNative, IEMNATIVE_VAR_IDX_PACK(idxOtherVar));
8013	int32_t const offBpDispOther = iemNativeStackCalcBpDisp(idxStackSlot);
8014	uint8_t const idxRegOther = pReNative->Core.aVars[idxOtherVar].idxReg;
8015	if (idxRegOther < RT_ELEMENTS(pReNative->Core.aHstRegs))
8016	{
8017	off = iemNativeEmitStoreGprByBp(pReNative, off, offBpDispOther, idxRegOther);
8018	iemNativeRegFreeVar(pReNative, idxRegOther, true); /** @todo const ref? */
8019	Assert(pReNative->Core.aVars[idxOtherVar].idxReg == UINT8_MAX);
8020	}
8021	Assert( pReNative->Core.aVars[idxOtherVar].idxStackSlot != UINT8_MAX
8022	&& pReNative->Core.aVars[idxOtherVar].idxReg == UINT8_MAX);
8023	off = iemNativeEmitLeaGprByBp(pReNative, off, IEMNATIVE_CALL_ARG0_GREG, offBpDispOther);
8024	off = iemNativeEmitStoreGprByBp(pReNative, off, offBpDisp, IEMNATIVE_CALL_ARG0_GREG);
8025	continue;
8026	}
8027
8028	case kIemNativeVarKind_GstRegRef:
8029	off = iemNativeEmitLeaGprByGstRegRef(pReNative, off, IEMNATIVE_CALL_ARG0_GREG,
8030	pVar->u.GstRegRef.enmClass, pVar->u.GstRegRef.idx);
8031	off = iemNativeEmitStoreGprByBp(pReNative, off, offBpDisp, IEMNATIVE_CALL_ARG0_GREG);
8032	continue;
8033
8034	case kIemNativeVarKind_Invalid:
8035	case kIemNativeVarKind_End:
8036	break;
8037	}
8038	AssertFailedStmt(IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_VAR_IPE_3));
8039	}
8040	}
8041	# if 0 //def VBOX_STRICT
8042	iemNativeRegAssertSanity(pReNative);
8043	# endif
8044	}
8045	#else
8046	AssertCompile(IEMNATIVE_CALL_MAX_ARG_COUNT <= IEMNATIVE_CALL_ARG_GREG_COUNT);
8047	#endif
8048
8049	/*
8050	* Make sure the argument variables are loaded into their respective registers.
8051	*
8052	* We can optimize this by ASSUMING that any register allocations are for
8053	* registeres that have already been loaded and are ready. The previous step
8054	* saw to that.
8055	*/
8056	if (~pReNative->Core.bmHstRegs & (g_afIemNativeCallRegs[cRegArgs] & ~g_afIemNativeCallRegs[cHiddenArgs]))
8057	{
8058	for (unsigned i = cHiddenArgs; i < cRegArgs; i++)
8059	{
8060	uint8_t const idxArgReg = g_aidxIemNativeCallRegs[i];
8061	if (pReNative->Core.bmHstRegs & RT_BIT_32(idxArgReg))
8062	Assert( pReNative->Core.aHstRegs[idxArgReg].idxVar == IEMNATIVE_VAR_IDX_PACK(pReNative->Core.aidxArgVars[i])
8063	&& pReNative->Core.aVars[pReNative->Core.aidxArgVars[i]].uArgNo == i
8064	&& pReNative->Core.aVars[pReNative->Core.aidxArgVars[i]].idxReg == idxArgReg);
8065	else
8066	{
8067	PIEMNATIVEVAR const pVar = &pReNative->Core.aVars[pReNative->Core.aidxArgVars[i]]; /* unpacked */
8068	if (pVar->idxReg < RT_ELEMENTS(pReNative->Core.aHstRegs))
8069	{
8070	Assert(pVar->enmKind == kIemNativeVarKind_Stack);
8071	off = iemNativeEmitLoadGprFromGpr(pReNative, off, idxArgReg, pVar->idxReg);
8072	pReNative->Core.bmHstRegs = (pReNative->Core.bmHstRegs & ~RT_BIT_32(pVar->idxReg))
8073	\| RT_BIT_32(idxArgReg);
8074	pVar->idxReg = idxArgReg;
8075	}
8076	else
8077	{
8078	/* Use ARG0 as temp for stuff we need registers for. */
8079	switch (pVar->enmKind)
8080	{
8081	case kIemNativeVarKind_Stack:
8082	{
8083	uint8_t const idxStackSlot = pVar->idxStackSlot;
8084	AssertStmt(idxStackSlot != UINT8_MAX, IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_VAR_NOT_INITIALIZED));
8085	off = iemNativeEmitLoadGprByBp(pReNative, off, idxArgReg, iemNativeStackCalcBpDisp(idxStackSlot));
8086	continue;
8087	}
8088
8089	case kIemNativeVarKind_Immediate:
8090	off = iemNativeEmitLoadGprImm64(pReNative, off, idxArgReg, pVar->u.uValue);
8091	continue;
8092
8093	case kIemNativeVarKind_VarRef:
8094	{
8095	uint8_t const idxOtherVar = pVar->u.idxRefVar; /* unpacked */
8096	Assert(idxOtherVar < RT_ELEMENTS(pReNative->Core.aVars));
8097	uint8_t const idxStackSlot = iemNativeVarGetStackSlot(pReNative,
8098	IEMNATIVE_VAR_IDX_PACK(idxOtherVar));
8099	int32_t const offBpDispOther = iemNativeStackCalcBpDisp(idxStackSlot);
8100	uint8_t const idxRegOther = pReNative->Core.aVars[idxOtherVar].idxReg;
8101	if (idxRegOther < RT_ELEMENTS(pReNative->Core.aHstRegs))
8102	{
8103	off = iemNativeEmitStoreGprByBp(pReNative, off, offBpDispOther, idxRegOther);
8104	iemNativeRegFreeVar(pReNative, idxRegOther, true); /** @todo const ref? */
8105	Assert(pReNative->Core.aVars[idxOtherVar].idxReg == UINT8_MAX);
8106	}
8107	Assert( pReNative->Core.aVars[idxOtherVar].idxStackSlot != UINT8_MAX
8108	&& pReNative->Core.aVars[idxOtherVar].idxReg == UINT8_MAX);
8109	off = iemNativeEmitLeaGprByBp(pReNative, off, idxArgReg, offBpDispOther);
8110	continue;
8111	}
8112
8113	case kIemNativeVarKind_GstRegRef:
8114	off = iemNativeEmitLeaGprByGstRegRef(pReNative, off, idxArgReg,
8115	pVar->u.GstRegRef.enmClass, pVar->u.GstRegRef.idx);
8116	continue;
8117
8118	case kIemNativeVarKind_Invalid:
8119	case kIemNativeVarKind_End:
8120	break;
8121	}
8122	AssertFailedStmt(IEMNATIVE_DO_LONGJMP(pReNative, VERR_IEM_VAR_IPE_3));
8123	}
8124	}
8125	}
8126	#if 0 //def VBOX_STRICT
8127	iemNativeRegAssertSanity(pReNative);
8128	#endif
8129	}
8130	#ifdef VBOX_STRICT
8131	else
8132	for (unsigned i = cHiddenArgs; i < cRegArgs; i++)
8133	{
8134	Assert(pReNative->Core.aVars[pReNative->Core.aidxArgVars[i]].uArgNo == i);
8135	Assert(pReNative->Core.aVars[pReNative->Core.aidxArgVars[i]].idxReg == g_aidxIemNativeCallRegs[i]);
8136	}
8137	#endif
8138
8139	/*
8140	* Free all argument variables (simplified).
8141	* Their lifetime always expires with the call they are for.
8142	*/
8143	/** @todo Make the python script check that arguments aren't used after
8144	* IEM_MC_CALL_XXXX. */
8145	/** @todo There is a special with IEM_MC_MEM_MAP_U16_RW and friends requiring
8146	* a IEM_MC_MEM_COMMIT_AND_UNMAP_RW after a AIMPL call typically with
8147	* an argument value. There is also some FPU stuff. */
8148	for (uint32_t i = cHiddenArgs; i < cArgs; i++)
8149	{
8150	uint8_t const idxVar = pReNative->Core.aidxArgVars[i]; /* unpacked */
8151	Assert(idxVar < RT_ELEMENTS(pReNative->Core.aVars));
8152
8153	/* no need to free registers: */
8154	AssertMsg(i < IEMNATIVE_CALL_ARG_GREG_COUNT
8155	? pReNative->Core.aVars[idxVar].idxReg == g_aidxIemNativeCallRegs[i]
8156	\|\| pReNative->Core.aVars[idxVar].idxReg == UINT8_MAX
8157	: pReNative->Core.aVars[idxVar].idxReg == UINT8_MAX,
8158	("i=%d idxVar=%d idxReg=%d, expected %d\n", i, idxVar, pReNative->Core.aVars[idxVar].idxReg,
8159	i < IEMNATIVE_CALL_ARG_GREG_COUNT ? g_aidxIemNativeCallRegs[i] : UINT8_MAX));
8160
8161	pReNative->Core.aidxArgVars[i] = UINT8_MAX;
8162	pReNative->Core.bmVars &= ~RT_BIT_32(idxVar);
8163	iemNativeVarFreeStackSlots(pReNative, idxVar);
8164	}
8165	Assert(pReNative->Core.u64ArgVars == UINT64_MAX);
8166
8167	/*
8168	* Flush volatile registers as we make the call.
8169	*/
8170	off = iemNativeRegMoveAndFreeAndFlushAtCall(pReNative, off, cRegArgs);
8171
8172	return off;
8173	}
8174
8175
8176
8177	/*********************************************************************************************************************************
8178	* TLB Lookup. *
8179	*********************************************************************************************************************************/
8180
8181	/**
8182	* This is called via iemNativeHlpAsmSafeWrapCheckTlbLookup.
8183	*/
8184	DECLASM(void) iemNativeHlpCheckTlbLookup(PVMCPU pVCpu, uintptr_t uResult, uint64_t GCPtr, uint32_t uSegAndSizeAndAccess)
8185	{
8186	uint8_t const iSegReg = RT_BYTE1(uSegAndSizeAndAccess);
8187	uint8_t const cbMem = RT_BYTE2(uSegAndSizeAndAccess);
8188	uint32_t const fAccess = uSegAndSizeAndAccess >> 16;
8189	Log(("iemNativeHlpCheckTlbLookup: %x:%#RX64 LB %#x fAccess=%#x -> %#RX64\n", iSegReg, GCPtr, cbMem, fAccess, uResult));
8190
8191	/* Do the lookup manually. */
8192	RTGCPTR const GCPtrFlat = iSegReg == UINT8_MAX ? GCPtr : GCPtr + pVCpu->cpum.GstCtx.aSRegs[iSegReg].u64Base;
8193	uint64_t const uTag = IEMTLB_CALC_TAG( &pVCpu->iem.s.DataTlb, GCPtrFlat);
8194	PIEMTLBENTRY const pTlbe = IEMTLB_TAG_TO_ENTRY(&pVCpu->iem.s.DataTlb, uTag);
8195	if (RT_LIKELY(pTlbe->uTag == uTag))
8196	{
8197	/*
8198	* Check TLB page table level access flags.
8199	*/
8200	AssertCompile(IEMTLBE_F_PT_NO_USER == 4);
8201	uint64_t const fNoUser = (IEM_GET_CPL(pVCpu) + 1) & IEMTLBE_F_PT_NO_USER;
8202	uint64_t const fNoWriteNoDirty = !(fAccess & IEM_ACCESS_TYPE_WRITE) ? 0
8203	: IEMTLBE_F_PT_NO_WRITE \| IEMTLBE_F_PT_NO_DIRTY \| IEMTLBE_F_PG_NO_WRITE;
8204	uint64_t const fFlagsAndPhysRev = pTlbe->fFlagsAndPhysRev & ( IEMTLBE_F_PHYS_REV \| IEMTLBE_F_NO_MAPPINGR3
8205	\| IEMTLBE_F_PG_UNASSIGNED
8206	\| IEMTLBE_F_PT_NO_ACCESSED
8207	\| fNoWriteNoDirty \| fNoUser);
8208	uint64_t const uTlbPhysRev = pVCpu->iem.s.DataTlb.uTlbPhysRev;
8209	if (RT_LIKELY(fFlagsAndPhysRev == uTlbPhysRev))
8210	{
8211	/*
8212	* Return the address.
8213	*/
8214	uint8_t const * const pbAddr = &pTlbe->pbMappingR3[GCPtrFlat & GUEST_PAGE_OFFSET_MASK];
8215	if ((uintptr_t)pbAddr == uResult)
8216	return;
8217	RT_NOREF(cbMem);
8218	AssertFailed();
8219	}
8220	else
8221	AssertMsgFailed(("fFlagsAndPhysRev=%#RX64 vs uTlbPhysRev=%#RX64: %#RX64\n",
8222	fFlagsAndPhysRev, uTlbPhysRev, fFlagsAndPhysRev ^ uTlbPhysRev));
8223	}
8224	else
8225	AssertFailed();
8226	RT_BREAKPOINT();
8227	}
8228
8229	/* The rest of the code is in IEMN8veRecompilerTlbLookup.h. */
8230
8231
8232
8233	/*********************************************************************************************************************************
8234	* Recompiler Core. *
8235	*********************************************************************************************************************************/
8236
8237	/** @callback_method_impl{FNDISREADBYTES, Dummy.} */
8238	static DECLCALLBACK(int) iemNativeDisasReadBytesDummy(PDISSTATE pDis, uint8_t offInstr, uint8_t cbMinRead, uint8_t cbMaxRead)
8239	{
8240	RT_BZERO(&pDis->Instr.ab[offInstr], cbMaxRead);
8241	pDis->cbCachedInstr += cbMaxRead;
8242	RT_NOREF(cbMinRead);
8243	return VERR_NO_DATA;
8244	}
8245
8246
8247	DECLHIDDEN(const char *) iemNativeDbgVCpuOffsetToName(uint32_t off)
8248	{
8249	static struct { uint32_t off; const char *pszName; } const s_aMembers[] =
8250	{
8251	#define ENTRY(a_Member) { (uint32_t)RT_UOFFSETOF(VMCPUCC, a_Member), #a_Member } /* cast is for stupid MSC */
8252	ENTRY(fLocalForcedActions),
8253	ENTRY(iem.s.rcPassUp),
8254	ENTRY(iem.s.fExec),
8255	ENTRY(iem.s.pbInstrBuf),
8256	ENTRY(iem.s.uInstrBufPc),
8257	ENTRY(iem.s.GCPhysInstrBuf),
8258	ENTRY(iem.s.cbInstrBufTotal),
8259	ENTRY(iem.s.idxTbCurInstr),
8260	#ifdef VBOX_WITH_STATISTICS
8261	ENTRY(iem.s.StatNativeTlbHitsForFetch),
8262	ENTRY(iem.s.StatNativeTlbHitsForStore),
8263	ENTRY(iem.s.StatNativeTlbHitsForStack),
8264	ENTRY(iem.s.StatNativeTlbHitsForMapped),
8265	ENTRY(iem.s.StatNativeCodeTlbMissesNewPage),
8266	ENTRY(iem.s.StatNativeCodeTlbHitsForNewPage),
8267	ENTRY(iem.s.StatNativeCodeTlbMissesNewPageWithOffset),
8268	ENTRY(iem.s.StatNativeCodeTlbHitsForNewPageWithOffset),
8269	#endif
8270	ENTRY(iem.s.DataTlb.aEntries),
8271	ENTRY(iem.s.DataTlb.uTlbRevision),
8272	ENTRY(iem.s.DataTlb.uTlbPhysRev),
8273	ENTRY(iem.s.DataTlb.cTlbHits),
8274	ENTRY(iem.s.CodeTlb.aEntries),
8275	ENTRY(iem.s.CodeTlb.uTlbRevision),
8276	ENTRY(iem.s.CodeTlb.uTlbPhysRev),
8277	ENTRY(iem.s.CodeTlb.cTlbHits),
8278	ENTRY(pVMR3),
8279	ENTRY(cpum.GstCtx.rax),
8280	ENTRY(cpum.GstCtx.ah),
8281	ENTRY(cpum.GstCtx.rcx),
8282	ENTRY(cpum.GstCtx.ch),
8283	ENTRY(cpum.GstCtx.rdx),
8284	ENTRY(cpum.GstCtx.dh),
8285	ENTRY(cpum.GstCtx.rbx),
8286	ENTRY(cpum.GstCtx.bh),
8287	ENTRY(cpum.GstCtx.rsp),
8288	ENTRY(cpum.GstCtx.rbp),
8289	ENTRY(cpum.GstCtx.rsi),
8290	ENTRY(cpum.GstCtx.rdi),
8291	ENTRY(cpum.GstCtx.r8),
8292	ENTRY(cpum.GstCtx.r9),
8293	ENTRY(cpum.GstCtx.r10),
8294	ENTRY(cpum.GstCtx.r11),
8295	ENTRY(cpum.GstCtx.r12),
8296	ENTRY(cpum.GstCtx.r13),
8297	ENTRY(cpum.GstCtx.r14),
8298	ENTRY(cpum.GstCtx.r15),
8299	ENTRY(cpum.GstCtx.es.Sel),
8300	ENTRY(cpum.GstCtx.es.u64Base),
8301	ENTRY(cpum.GstCtx.es.u32Limit),
8302	ENTRY(cpum.GstCtx.es.Attr),
8303	ENTRY(cpum.GstCtx.cs.Sel),
8304	ENTRY(cpum.GstCtx.cs.u64Base),
8305	ENTRY(cpum.GstCtx.cs.u32Limit),
8306	ENTRY(cpum.GstCtx.cs.Attr),
8307	ENTRY(cpum.GstCtx.ss.Sel),
8308	ENTRY(cpum.GstCtx.ss.u64Base),
8309	ENTRY(cpum.GstCtx.ss.u32Limit),
8310	ENTRY(cpum.GstCtx.ss.Attr),
8311	ENTRY(cpum.GstCtx.ds.Sel),
8312	ENTRY(cpum.GstCtx.ds.u64Base),
8313	ENTRY(cpum.GstCtx.ds.u32Limit),
8314	ENTRY(cpum.GstCtx.ds.Attr),
8315	ENTRY(cpum.GstCtx.fs.Sel),
8316	ENTRY(cpum.GstCtx.fs.u64Base),
8317	ENTRY(cpum.GstCtx.fs.u32Limit),
8318	ENTRY(cpum.GstCtx.fs.Attr),
8319	ENTRY(cpum.GstCtx.gs.Sel),
8320	ENTRY(cpum.GstCtx.gs.u64Base),
8321	ENTRY(cpum.GstCtx.gs.u32Limit),
8322	ENTRY(cpum.GstCtx.gs.Attr),
8323	ENTRY(cpum.GstCtx.rip),
8324	ENTRY(cpum.GstCtx.eflags),
8325	ENTRY(cpum.GstCtx.uRipInhibitInt),
8326	#ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
8327	ENTRY(cpum.GstCtx.XState.x87.aXMM[0]),
8328	ENTRY(cpum.GstCtx.XState.x87.aXMM[1]),
8329	ENTRY(cpum.GstCtx.XState.x87.aXMM[2]),
8330	ENTRY(cpum.GstCtx.XState.x87.aXMM[3]),
8331	ENTRY(cpum.GstCtx.XState.x87.aXMM[4]),
8332	ENTRY(cpum.GstCtx.XState.x87.aXMM[5]),
8333	ENTRY(cpum.GstCtx.XState.x87.aXMM[6]),
8334	ENTRY(cpum.GstCtx.XState.x87.aXMM[7]),
8335	ENTRY(cpum.GstCtx.XState.x87.aXMM[8]),
8336	ENTRY(cpum.GstCtx.XState.x87.aXMM[9]),
8337	ENTRY(cpum.GstCtx.XState.x87.aXMM[10]),
8338	ENTRY(cpum.GstCtx.XState.x87.aXMM[11]),
8339	ENTRY(cpum.GstCtx.XState.x87.aXMM[12]),
8340	ENTRY(cpum.GstCtx.XState.x87.aXMM[13]),
8341	ENTRY(cpum.GstCtx.XState.x87.aXMM[14]),
8342	ENTRY(cpum.GstCtx.XState.x87.aXMM[15]),
8343	ENTRY(cpum.GstCtx.XState.u.YmmHi.aYmmHi[0]),
8344	ENTRY(cpum.GstCtx.XState.u.YmmHi.aYmmHi[1]),
8345	ENTRY(cpum.GstCtx.XState.u.YmmHi.aYmmHi[2]),
8346	ENTRY(cpum.GstCtx.XState.u.YmmHi.aYmmHi[3]),
8347	ENTRY(cpum.GstCtx.XState.u.YmmHi.aYmmHi[4]),
8348	ENTRY(cpum.GstCtx.XState.u.YmmHi.aYmmHi[5]),
8349	ENTRY(cpum.GstCtx.XState.u.YmmHi.aYmmHi[6]),
8350	ENTRY(cpum.GstCtx.XState.u.YmmHi.aYmmHi[7]),
8351	ENTRY(cpum.GstCtx.XState.u.YmmHi.aYmmHi[8]),
8352	ENTRY(cpum.GstCtx.XState.u.YmmHi.aYmmHi[9]),
8353	ENTRY(cpum.GstCtx.XState.u.YmmHi.aYmmHi[10]),
8354	ENTRY(cpum.GstCtx.XState.u.YmmHi.aYmmHi[11]),
8355	ENTRY(cpum.GstCtx.XState.u.YmmHi.aYmmHi[12]),
8356	ENTRY(cpum.GstCtx.XState.u.YmmHi.aYmmHi[13]),
8357	ENTRY(cpum.GstCtx.XState.u.YmmHi.aYmmHi[14]),
8358	ENTRY(cpum.GstCtx.XState.u.YmmHi.aYmmHi[15])
8359	#endif
8360	#undef ENTRY
8361	};
8362	#ifdef VBOX_STRICT
8363	static bool s_fOrderChecked = false;
8364	if (!s_fOrderChecked)
8365	{
8366	s_fOrderChecked = true;
8367	uint32_t offPrev = s_aMembers[0].off;
8368	for (unsigned i = 1; i < RT_ELEMENTS(s_aMembers); i++)
8369	{
8370	Assert(s_aMembers[i].off > offPrev);
8371	offPrev = s_aMembers[i].off;
8372	}
8373	}
8374	#endif
8375
8376	/*
8377	* Binary lookup.
8378	*/
8379	unsigned iStart = 0;
8380	unsigned iEnd = RT_ELEMENTS(s_aMembers);
8381	for (;;)
8382	{
8383	unsigned const iCur = iStart + (iEnd - iStart) / 2;
8384	uint32_t const offCur = s_aMembers[iCur].off;
8385	if (off < offCur)
8386	{
8387	if (iCur != iStart)
8388	iEnd = iCur;
8389	else
8390	break;
8391	}
8392	else if (off > offCur)
8393	{
8394	if (iCur + 1 < iEnd)
8395	iStart = iCur + 1;
8396	else
8397	break;
8398	}
8399	else
8400	return s_aMembers[iCur].pszName;
8401	}
8402	#ifdef VBOX_WITH_STATISTICS
8403	if (off - RT_UOFFSETOF(VMCPUCC, iem.s.acThreadedFuncStats) < RT_SIZEOFMEMB(VMCPUCC, iem.s.acThreadedFuncStats))
8404	return "iem.s.acThreadedFuncStats[iFn]";
8405	#endif
8406	return NULL;
8407	}
8408
8409
8410	/**
8411	* Formats TB flags (IEM_F_XXX and IEMTB_F_XXX) to string.
8412	* @returns pszBuf.
8413	* @param fFlags The flags.
8414	* @param pszBuf The output buffer.
8415	* @param cbBuf The output buffer size. At least 32 bytes.
8416	*/
8417	DECLHIDDEN(const char ) iemTbFlagsToString(uint32_t fFlags, char pszBuf, size_t cbBuf) RT_NOEXCEPT
8418	{
8419	Assert(cbBuf >= 32);
8420	static RTSTRTUPLE const s_aModes[] =
8421	{
8422	/* [00] = */ { RT_STR_TUPLE("16BIT") },
8423	/* [01] = */ { RT_STR_TUPLE("32BIT") },
8424	/* [02] = */ { RT_STR_TUPLE("!2!") },
8425	/* [03] = */ { RT_STR_TUPLE("!3!") },
8426	/* [04] = */ { RT_STR_TUPLE("16BIT_PRE_386") },
8427	/* [05] = */ { RT_STR_TUPLE("32BIT_FLAT") },
8428	/* [06] = */ { RT_STR_TUPLE("!6!") },
8429	/* [07] = */ { RT_STR_TUPLE("!7!") },
8430	/* [08] = */ { RT_STR_TUPLE("16BIT_PROT") },
8431	/* [09] = */ { RT_STR_TUPLE("32BIT_PROT") },
8432	/* [0a] = */ { RT_STR_TUPLE("64BIT") },
8433	/* [0b] = */ { RT_STR_TUPLE("!b!") },
8434	/* [0c] = */ { RT_STR_TUPLE("16BIT_PROT_PRE_386") },
8435	/* [0d] = */ { RT_STR_TUPLE("32BIT_PROT_FLAT") },
8436	/* [0e] = */ { RT_STR_TUPLE("!e!") },
8437	/* [0f] = */ { RT_STR_TUPLE("!f!") },
8438	/* [10] = */ { RT_STR_TUPLE("!10!") },
8439	/* [11] = */ { RT_STR_TUPLE("!11!") },
8440	/* [12] = */ { RT_STR_TUPLE("!12!") },
8441	/* [13] = */ { RT_STR_TUPLE("!13!") },
8442	/* [14] = */ { RT_STR_TUPLE("!14!") },
8443	/* [15] = */ { RT_STR_TUPLE("!15!") },
8444	/* [16] = */ { RT_STR_TUPLE("!16!") },
8445	/* [17] = */ { RT_STR_TUPLE("!17!") },
8446	/* [18] = */ { RT_STR_TUPLE("16BIT_PROT_V86") },
8447	/* [19] = */ { RT_STR_TUPLE("32BIT_PROT_V86") },
8448	/* [1a] = */ { RT_STR_TUPLE("!1a!") },
8449	/* [1b] = */ { RT_STR_TUPLE("!1b!") },
8450	/* [1c] = */ { RT_STR_TUPLE("!1c!") },
8451	/* [1d] = */ { RT_STR_TUPLE("!1d!") },
8452	/* [1e] = */ { RT_STR_TUPLE("!1e!") },
8453	/* [1f] = */ { RT_STR_TUPLE("!1f!") },
8454	};
8455	AssertCompile(RT_ELEMENTS(s_aModes) == IEM_F_MODE_MASK + 1);
8456	memcpy(pszBuf, s_aModes[fFlags & IEM_F_MODE_MASK].psz, s_aModes[fFlags & IEM_F_MODE_MASK].cch);
8457	size_t off = s_aModes[fFlags & IEM_F_MODE_MASK].cch;
8458
8459	pszBuf[off++] = ' ';
8460	pszBuf[off++] = 'C';
8461	pszBuf[off++] = 'P';
8462	pszBuf[off++] = 'L';
8463	pszBuf[off++] = '0' + ((fFlags >> IEM_F_X86_CPL_SHIFT) & IEM_F_X86_CPL_SMASK);
8464	Assert(off < 32);
8465
8466	fFlags &= ~(IEM_F_MODE_MASK \| IEM_F_X86_CPL_SMASK);
8467
8468	static struct { const char *pszName; uint32_t cchName; uint32_t fFlag; } const s_aFlags[] =
8469	{
8470	{ RT_STR_TUPLE("BYPASS_HANDLERS"), IEM_F_BYPASS_HANDLERS },
8471	{ RT_STR_TUPLE("PENDING_BRK_INSTR"), IEM_F_PENDING_BRK_INSTR },
8472	{ RT_STR_TUPLE("PENDING_BRK_DATA"), IEM_F_PENDING_BRK_DATA },
8473	{ RT_STR_TUPLE("PENDING_BRK_X86_IO"), IEM_F_PENDING_BRK_X86_IO },
8474	{ RT_STR_TUPLE("X86_DISREGARD_LOCK"), IEM_F_X86_DISREGARD_LOCK },
8475	{ RT_STR_TUPLE("X86_CTX_VMX"), IEM_F_X86_CTX_VMX },
8476	{ RT_STR_TUPLE("X86_CTX_SVM"), IEM_F_X86_CTX_SVM },
8477	{ RT_STR_TUPLE("X86_CTX_IN_GUEST"), IEM_F_X86_CTX_IN_GUEST },
8478	{ RT_STR_TUPLE("X86_CTX_SMM"), IEM_F_X86_CTX_SMM },
8479	{ RT_STR_TUPLE("INHIBIT_SHADOW"), IEMTB_F_INHIBIT_SHADOW },
8480	{ RT_STR_TUPLE("INHIBIT_NMI"), IEMTB_F_INHIBIT_NMI },
8481	{ RT_STR_TUPLE("CS_LIM_CHECKS"), IEMTB_F_CS_LIM_CHECKS },
8482	{ RT_STR_TUPLE("TYPE_THREADED"), IEMTB_F_TYPE_THREADED },
8483	{ RT_STR_TUPLE("TYPE_NATIVE"), IEMTB_F_TYPE_NATIVE },
8484	};
8485	if (fFlags)
8486	for (unsigned i = 0; i < RT_ELEMENTS(s_aFlags); i++)
8487	if (s_aFlags[i].fFlag & fFlags)
8488	{
8489	AssertReturnStmt(off + 1 + s_aFlags[i].cchName + 1 <= cbBuf, pszBuf[off] = '\0', pszBuf);
8490	pszBuf[off++] = ' ';
8491	memcpy(&pszBuf[off], s_aFlags[i].pszName, s_aFlags[i].cchName);
8492	off += s_aFlags[i].cchName;
8493	fFlags &= ~s_aFlags[i].fFlag;
8494	if (!fFlags)
8495	break;
8496	}
8497	pszBuf[off] = '\0';
8498
8499	return pszBuf;
8500	}
8501
8502
8503	DECLHIDDEN(void) iemNativeDisassembleTb(PCIEMTB pTb, PCDBGFINFOHLP pHlp) RT_NOEXCEPT
8504	{
8505	AssertReturnVoid((pTb->fFlags & IEMTB_F_TYPE_MASK) == IEMTB_F_TYPE_NATIVE);
8506	#if defined(RT_ARCH_AMD64)
8507	static const char * const a_apszMarkers[] =
8508	{
8509	/[0]=/ "unknown0", "CheckCsLim", "ConsiderLimChecking", "CheckOpcodes",
8510	/[4]=/ "PcAfterBranch", "LoadTlbForNewPage", "LoadTlbAfterBranch"
8511	};
8512	#endif
8513
8514	char szDisBuf[512];
8515	DISSTATE Dis;
8516	PCIEMNATIVEINSTR const paNative = pTb->Native.paInstructions;
8517	uint32_t const cNative = pTb->Native.cInstructions;
8518	uint32_t offNative = 0;
8519	#ifdef IEMNATIVE_WITH_TB_DEBUG_INFO
8520	PCIEMTBDBG const pDbgInfo = pTb->pDbgInfo;
8521	#endif
8522	DISCPUMODE enmGstCpuMode = (pTb->fFlags & IEM_F_MODE_CPUMODE_MASK) == IEMMODE_16BIT ? DISCPUMODE_16BIT
8523	: (pTb->fFlags & IEM_F_MODE_CPUMODE_MASK) == IEMMODE_32BIT ? DISCPUMODE_32BIT
8524	: DISCPUMODE_64BIT;
8525	#if defined(RT_ARCH_AMD64) && !defined(VBOX_WITH_IEM_USING_CAPSTONE_DISASSEMBLER)
8526	DISCPUMODE const enmHstCpuMode = DISCPUMODE_64BIT;
8527	#elif defined(RT_ARCH_ARM64) && !defined(VBOX_WITH_IEM_USING_CAPSTONE_DISASSEMBLER)
8528	DISCPUMODE const enmHstCpuMode = DISCPUMODE_ARMV8_A64;
8529	#elif !defined(VBOX_WITH_IEM_USING_CAPSTONE_DISASSEMBLER)
8530	# error "Port me"
8531	#else
8532	csh hDisasm = ~(size_t)0;
8533	# if defined(RT_ARCH_AMD64)
8534	cs_err rcCs = cs_open(CS_ARCH_X86, CS_MODE_LITTLE_ENDIAN \| CS_MODE_64, &hDisasm);
8535	# elif defined(RT_ARCH_ARM64)
8536	cs_err rcCs = cs_open(CS_ARCH_ARM64, CS_MODE_LITTLE_ENDIAN, &hDisasm);
8537	# else
8538	# error "Port me"
8539	# endif
8540	AssertMsgReturnVoid(rcCs == CS_ERR_OK, ("%d (%#x)\n", rcCs, rcCs));
8541
8542	//rcCs = cs_option(hDisasm, CS_OPT_DETAIL, CS_OPT_ON); - not needed as pInstr->detail doesn't provide full memory detail.
8543	//Assert(rcCs == CS_ERR_OK);
8544	#endif
8545
8546	/*
8547	* Print TB info.
8548	*/
8549	pHlp->pfnPrintf(pHlp,
8550	"pTb=%p: GCPhysPc=%RGp cInstructions=%u LB %#x cRanges=%u\n"
8551	"pTb=%p: cUsed=%u msLastUsed=%u fFlags=%#010x %s\n",
8552	pTb, pTb->GCPhysPc, pTb->cInstructions, pTb->cbOpcodes, pTb->cRanges,
8553	pTb, pTb->cUsed, pTb->msLastUsed, pTb->fFlags, iemTbFlagsToString(pTb->fFlags, szDisBuf, sizeof(szDisBuf)));
8554	#ifdef IEMNATIVE_WITH_TB_DEBUG_INFO
8555	if (pDbgInfo && pDbgInfo->cEntries > 1)
8556	{
8557	Assert(pDbgInfo->aEntries[0].Gen.uType == kIemTbDbgEntryType_NativeOffset);
8558
8559	/*
8560	* This disassembly is driven by the debug info which follows the native
8561	* code and indicates when it starts with the next guest instructions,
8562	* where labels are and such things.
8563	*/
8564	uint32_t idxThreadedCall = 0;
8565	uint32_t fExec = pTb->fFlags & UINT32_C(0x00ffffff);
8566	uint8_t idxRange = UINT8_MAX;
8567	uint8_t const cRanges = RT_MIN(pTb->cRanges, RT_ELEMENTS(pTb->aRanges));
8568	uint32_t offRange = 0;
8569	uint32_t offOpcodes = 0;
8570	uint32_t const cbOpcodes = pTb->cbOpcodes;
8571	RTGCPHYS GCPhysPc = pTb->GCPhysPc;
8572	uint32_t const cDbgEntries = pDbgInfo->cEntries;
8573	uint32_t iDbgEntry = 1;
8574	uint32_t offDbgNativeNext = pDbgInfo->aEntries[0].NativeOffset.offNative;
8575
8576	while (offNative < cNative)
8577	{
8578	/* If we're at or have passed the point where the next chunk of debug
8579	info starts, process it. */
8580	if (offDbgNativeNext <= offNative)
8581	{
8582	offDbgNativeNext = UINT32_MAX;
8583	for (; iDbgEntry < cDbgEntries; iDbgEntry++)
8584	{
8585	switch (pDbgInfo->aEntries[iDbgEntry].Gen.uType)
8586	{
8587	case kIemTbDbgEntryType_GuestInstruction:
8588	{
8589	/* Did the exec flag change? */
8590	if (fExec != pDbgInfo->aEntries[iDbgEntry].GuestInstruction.fExec)
8591	{
8592	pHlp->pfnPrintf(pHlp,
8593	" fExec change %#08x -> %#08x %s\n",
8594	fExec, pDbgInfo->aEntries[iDbgEntry].GuestInstruction.fExec,
8595	iemTbFlagsToString(pDbgInfo->aEntries[iDbgEntry].GuestInstruction.fExec,
8596	szDisBuf, sizeof(szDisBuf)));
8597	fExec = pDbgInfo->aEntries[iDbgEntry].GuestInstruction.fExec;
8598	enmGstCpuMode = (fExec & IEM_F_MODE_CPUMODE_MASK) == IEMMODE_16BIT ? DISCPUMODE_16BIT
8599	: (fExec & IEM_F_MODE_CPUMODE_MASK) == IEMMODE_32BIT ? DISCPUMODE_32BIT
8600	: DISCPUMODE_64BIT;
8601	}
8602
8603	/* New opcode range? We need to fend up a spurious debug info entry here for cases
8604	where the compilation was aborted before the opcode was recorded and the actual
8605	instruction was translated to a threaded call. This may happen when we run out
8606	of ranges, or when some complicated interrupts/FFs are found to be pending or
8607	similar. So, we just deal with it here rather than in the compiler code as it
8608	is a lot simpler to do here. */
8609	if ( idxRange == UINT8_MAX
8610	\|\| idxRange >= cRanges
8611	\|\| offRange >= pTb->aRanges[idxRange].cbOpcodes)
8612	{
8613	idxRange += 1;
8614	if (idxRange < cRanges)
8615	offRange = !idxRange ? 0 : offRange - pTb->aRanges[idxRange - 1].cbOpcodes;
8616	else
8617	continue;
8618	Assert(offOpcodes == pTb->aRanges[idxRange].offOpcodes + offRange);
8619	GCPhysPc = pTb->aRanges[idxRange].offPhysPage
8620	+ (pTb->aRanges[idxRange].idxPhysPage == 0
8621	? pTb->GCPhysPc & ~(RTGCPHYS)GUEST_PAGE_OFFSET_MASK
8622	: pTb->aGCPhysPages[pTb->aRanges[idxRange].idxPhysPage - 1]);
8623	pHlp->pfnPrintf(pHlp, " Range #%u: GCPhysPc=%RGp LB %#x [idxPg=%d]\n",
8624	idxRange, GCPhysPc, pTb->aRanges[idxRange].cbOpcodes,
8625	pTb->aRanges[idxRange].idxPhysPage);
8626	GCPhysPc += offRange;
8627	}
8628
8629	/* Disassemble the instruction. */
8630	//uint8_t const cbInstrMax = RT_MIN(pTb->aRanges[idxRange].cbOpcodes - offRange, 15);
8631	uint8_t const cbInstrMax = RT_MIN(cbOpcodes - offOpcodes, 15);
8632	uint32_t cbInstr = 1;
8633	int rc = DISInstrWithPrefetchedBytes(GCPhysPc, enmGstCpuMode, DISOPTYPE_ALL,
8634	&pTb->pabOpcodes[offOpcodes], cbInstrMax,
8635	iemNativeDisasReadBytesDummy, NULL, &Dis, &cbInstr);
8636	if (RT_SUCCESS(rc))
8637	{
8638	size_t cch = DISFormatYasmEx(&Dis, szDisBuf, sizeof(szDisBuf),
8639	DIS_FMT_FLAGS_BYTES_WIDTH_MAKE(10) \| DIS_FMT_FLAGS_BYTES_LEFT
8640	\| DIS_FMT_FLAGS_RELATIVE_BRANCH \| DIS_FMT_FLAGS_C_HEX,
8641	NULL /pfnGetSymbol/, NULL /pvUser/);
8642
8643	static unsigned const s_offMarker = 55;
8644	static char const s_szMarker[] = " ; <--- guest";
8645	if (cch < s_offMarker)
8646	{
8647	memset(&szDisBuf[cch], ' ', s_offMarker - cch);
8648	cch = s_offMarker;
8649	}
8650	if (cch + sizeof(s_szMarker) <= sizeof(szDisBuf))
8651	memcpy(&szDisBuf[cch], s_szMarker, sizeof(s_szMarker));
8652
8653	pHlp->pfnPrintf(pHlp, " %%%%%RGp: %s\n", GCPhysPc, szDisBuf);
8654	}
8655	else
8656	{
8657	pHlp->pfnPrintf(pHlp, " %%%%%RGp: %.*Rhxs - guest disassembly failure %Rrc\n",
8658	GCPhysPc, cbInstrMax, &pTb->pabOpcodes[offOpcodes], rc);
8659	cbInstr = 1;
8660	}
8661	GCPhysPc += cbInstr;
8662	offOpcodes += cbInstr;
8663	offRange += cbInstr;
8664	continue;
8665	}
8666
8667	case kIemTbDbgEntryType_ThreadedCall:
8668	pHlp->pfnPrintf(pHlp,
8669	" Call #%u to %s (%u args) - %s\n",
8670	idxThreadedCall,
8671	g_apszIemThreadedFunctions[pDbgInfo->aEntries[iDbgEntry].ThreadedCall.enmCall],
8672	g_acIemThreadedFunctionUsedArgs[pDbgInfo->aEntries[iDbgEntry].ThreadedCall.enmCall],
8673	pDbgInfo->aEntries[iDbgEntry].ThreadedCall.fRecompiled ? "recompiled" : "todo");
8674	idxThreadedCall++;
8675	continue;
8676
8677	case kIemTbDbgEntryType_GuestRegShadowing:
8678	{
8679	PCIEMTBDBGENTRY const pEntry = &pDbgInfo->aEntries[iDbgEntry];
8680	const char * const pszGstReg = g_aGstShadowInfo[pEntry->GuestRegShadowing.idxGstReg].pszName;
8681	if (pEntry->GuestRegShadowing.idxHstReg == UINT8_MAX)
8682	pHlp->pfnPrintf(pHlp, " Guest register %s != host register %s\n", pszGstReg,
8683	g_apszIemNativeHstRegNames[pEntry->GuestRegShadowing.idxHstRegPrev]);
8684	else if (pEntry->GuestRegShadowing.idxHstRegPrev == UINT8_MAX)
8685	pHlp->pfnPrintf(pHlp, " Guest register %s == host register %s\n", pszGstReg,
8686	g_apszIemNativeHstRegNames[pEntry->GuestRegShadowing.idxHstReg]);
8687	else
8688	pHlp->pfnPrintf(pHlp, " Guest register %s == host register %s (previously in %s)\n", pszGstReg,
8689	g_apszIemNativeHstRegNames[pEntry->GuestRegShadowing.idxHstReg],
8690	g_apszIemNativeHstRegNames[pEntry->GuestRegShadowing.idxHstRegPrev]);
8691	continue;
8692	}
8693
8694	#ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
8695	case kIemTbDbgEntryType_GuestSimdRegShadowing:
8696	{
8697	PCIEMTBDBGENTRY const pEntry = &pDbgInfo->aEntries[iDbgEntry];
8698	const char * const pszGstReg = g_aGstSimdShadowInfo[pEntry->GuestSimdRegShadowing.idxGstSimdReg].pszName;
8699	if (pEntry->GuestSimdRegShadowing.idxHstSimdReg == UINT8_MAX)
8700	pHlp->pfnPrintf(pHlp, " Guest SIMD register %s != host SIMD register %s\n", pszGstReg,
8701	g_apszIemNativeHstSimdRegNames[pEntry->GuestSimdRegShadowing.idxHstSimdRegPrev]);
8702	else if (pEntry->GuestSimdRegShadowing.idxHstSimdRegPrev == UINT8_MAX)
8703	pHlp->pfnPrintf(pHlp, " Guest SIMD register %s == host SIMD register %s\n", pszGstReg,
8704	g_apszIemNativeHstSimdRegNames[pEntry->GuestSimdRegShadowing.idxHstSimdReg]);
8705	else
8706	pHlp->pfnPrintf(pHlp, " Guest SIMD register %s == host SIMD register %s (previously in %s)\n", pszGstReg,
8707	g_apszIemNativeHstSimdRegNames[pEntry->GuestSimdRegShadowing.idxHstSimdReg],
8708	g_apszIemNativeHstSimdRegNames[pEntry->GuestSimdRegShadowing.idxHstSimdRegPrev]);
8709	continue;
8710	}
8711	#endif
8712
8713	case kIemTbDbgEntryType_Label:
8714	{
8715	const char *pszName = "what_the_fudge";
8716	const char *pszComment = "";
8717	bool fNumbered = pDbgInfo->aEntries[iDbgEntry].Label.uData != 0;
8718	switch ((IEMNATIVELABELTYPE)pDbgInfo->aEntries[iDbgEntry].Label.enmLabel)
8719	{
8720	case kIemNativeLabelType_Return:
8721	pszName = "Return";
8722	break;
8723	case kIemNativeLabelType_ReturnBreak:
8724	pszName = "ReturnBreak";
8725	break;
8726	case kIemNativeLabelType_ReturnWithFlags:
8727	pszName = "ReturnWithFlags";
8728	break;
8729	case kIemNativeLabelType_NonZeroRetOrPassUp:
8730	pszName = "NonZeroRetOrPassUp";
8731	break;
8732	case kIemNativeLabelType_RaiseGp0:
8733	pszName = "RaiseGp0";
8734	break;
8735	case kIemNativeLabelType_RaiseNm:
8736	pszName = "RaiseNm";
8737	break;
8738	case kIemNativeLabelType_RaiseUd:
8739	pszName = "RaiseUd";
8740	break;
8741	case kIemNativeLabelType_RaiseMf:
8742	pszName = "RaiseMf";
8743	break;
8744	case kIemNativeLabelType_RaiseXf:
8745	pszName = "RaiseXf";
8746	break;
8747	case kIemNativeLabelType_ObsoleteTb:
8748	pszName = "ObsoleteTb";
8749	break;
8750	case kIemNativeLabelType_NeedCsLimChecking:
8751	pszName = "NeedCsLimChecking";
8752	break;
8753	case kIemNativeLabelType_CheckBranchMiss:
8754	pszName = "CheckBranchMiss";
8755	break;
8756	case kIemNativeLabelType_If:
8757	pszName = "If";
8758	fNumbered = true;
8759	break;
8760	case kIemNativeLabelType_Else:
8761	pszName = "Else";
8762	fNumbered = true;
8763	pszComment = " ; regs state restored pre-if-block";
8764	break;
8765	case kIemNativeLabelType_Endif:
8766	pszName = "Endif";
8767	fNumbered = true;
8768	break;
8769	case kIemNativeLabelType_CheckIrq:
8770	pszName = "CheckIrq_CheckVM";
8771	fNumbered = true;
8772	break;
8773	case kIemNativeLabelType_TlbLookup:
8774	pszName = "TlbLookup";
8775	fNumbered = true;
8776	break;
8777	case kIemNativeLabelType_TlbMiss:
8778	pszName = "TlbMiss";
8779	fNumbered = true;
8780	break;
8781	case kIemNativeLabelType_TlbDone:
8782	pszName = "TlbDone";
8783	fNumbered = true;
8784	break;
8785	case kIemNativeLabelType_Invalid:
8786	case kIemNativeLabelType_End:
8787	break;
8788	}
8789	if (fNumbered)
8790	pHlp->pfnPrintf(pHlp, " %s_%u:%s\n", pszName, pDbgInfo->aEntries[iDbgEntry].Label.uData, pszComment);
8791	else
8792	pHlp->pfnPrintf(pHlp, " %s:\n", pszName);
8793	continue;
8794	}
8795
8796	case kIemTbDbgEntryType_NativeOffset:
8797	offDbgNativeNext = pDbgInfo->aEntries[iDbgEntry].NativeOffset.offNative;
8798	Assert(offDbgNativeNext > offNative);
8799	break;
8800
8801	#ifdef IEMNATIVE_WITH_DELAYED_PC_UPDATING
8802	case kIemTbDbgEntryType_DelayedPcUpdate:
8803	pHlp->pfnPrintf(pHlp,
8804	" Updating guest PC value by %u (cInstrSkipped=%u)\n",
8805	pDbgInfo->aEntries[iDbgEntry].DelayedPcUpdate.offPc,
8806	pDbgInfo->aEntries[iDbgEntry].DelayedPcUpdate.cInstrSkipped);
8807	continue;
8808	#endif
8809
8810	default:
8811	AssertFailed();
8812	}
8813	iDbgEntry++;
8814	break;
8815	}
8816	}
8817
8818	/*
8819	* Disassemble the next native instruction.
8820	*/
8821	PCIEMNATIVEINSTR const pNativeCur = &paNative[offNative];
8822	# ifndef VBOX_WITH_IEM_USING_CAPSTONE_DISASSEMBLER
8823	uint32_t cbInstr = sizeof(paNative[0]);
8824	int const rc = DISInstr(pNativeCur, enmHstCpuMode, &Dis, &cbInstr);
8825	if (RT_SUCCESS(rc))
8826	{
8827	# if defined(RT_ARCH_AMD64)
8828	if (Dis.pCurInstr->uOpcode == OP_NOP && cbInstr == 7) /* iemNativeEmitMarker */
8829	{
8830	uint32_t const uInfo = (uint32_t const )&Dis.Instr.ab[3];
8831	if (RT_HIWORD(uInfo) < kIemThreadedFunc_End)
8832	pHlp->pfnPrintf(pHlp, " %p: nop ; marker: call #%u to %s (%u args) - %s\n",
8833	pNativeCur, uInfo & 0x7fff, g_apszIemThreadedFunctions[RT_HIWORD(uInfo)],
8834	g_acIemThreadedFunctionUsedArgs[RT_HIWORD(uInfo)],
8835	uInfo & 0x8000 ? "recompiled" : "todo");
8836	else if ((uInfo & ~RT_BIT_32(31)) < RT_ELEMENTS(a_apszMarkers))
8837	pHlp->pfnPrintf(pHlp, " %p: nop ; marker: %s\n", pNativeCur, a_apszMarkers[uInfo & ~RT_BIT_32(31)]);
8838	else
8839	pHlp->pfnPrintf(pHlp, " %p: nop ; unknown marker: %#x (%d)\n", pNativeCur, uInfo, uInfo);
8840	}
8841	else
8842	# endif
8843	{
8844	const char *pszAnnotation = NULL;
8845	# ifdef RT_ARCH_AMD64
8846	DISFormatYasmEx(&Dis, szDisBuf, sizeof(szDisBuf),
8847	DIS_FMT_FLAGS_BYTES_WIDTH_MAKE(10) \| DIS_FMT_FLAGS_BYTES_LEFT
8848	\| DIS_FMT_FLAGS_RELATIVE_BRANCH \| DIS_FMT_FLAGS_C_HEX,
8849	NULL /pfnGetSymbol/, NULL /pvUser/);
8850	PCDISOPPARAM pMemOp;
8851	if (DISUSE_IS_EFFECTIVE_ADDR(Dis.Param1.fUse))
8852	pMemOp = &Dis.Param1;
8853	else if (DISUSE_IS_EFFECTIVE_ADDR(Dis.Param2.fUse))
8854	pMemOp = &Dis.Param2;
8855	else if (DISUSE_IS_EFFECTIVE_ADDR(Dis.Param3.fUse))
8856	pMemOp = &Dis.Param3;
8857	else
8858	pMemOp = NULL;
8859	if ( pMemOp
8860	&& pMemOp->x86.Base.idxGenReg == IEMNATIVE_REG_FIXED_PVMCPU
8861	&& (pMemOp->fUse & (DISUSE_BASE \| DISUSE_REG_GEN64)) == (DISUSE_BASE \| DISUSE_REG_GEN64))
8862	pszAnnotation = iemNativeDbgVCpuOffsetToName(pMemOp->fUse & DISUSE_DISPLACEMENT32
8863	? pMemOp->x86.uDisp.u32 : pMemOp->x86.uDisp.u8);
8864
8865	#elif defined(RT_ARCH_ARM64)
8866	DISFormatArmV8Ex(&Dis, szDisBuf, sizeof(szDisBuf),
8867	DIS_FMT_FLAGS_BYTES_LEFT \| DIS_FMT_FLAGS_RELATIVE_BRANCH \| DIS_FMT_FLAGS_C_HEX,
8868	NULL /pfnGetSymbol/, NULL /pvUser/);
8869	# else
8870	# error "Port me"
8871	# endif
8872	if (pszAnnotation)
8873	{
8874	static unsigned const s_offAnnotation = 55;
8875	size_t const cchAnnotation = strlen(pszAnnotation);
8876	size_t cchDis = strlen(szDisBuf);
8877	if (RT_MAX(cchDis, s_offAnnotation) + sizeof(" ; ") + cchAnnotation <= sizeof(szDisBuf))
8878	{
8879	if (cchDis < s_offAnnotation)
8880	{
8881	memset(&szDisBuf[cchDis], ' ', s_offAnnotation - cchDis);
8882	cchDis = s_offAnnotation;
8883	}
8884	szDisBuf[cchDis++] = ' ';
8885	szDisBuf[cchDis++] = ';';
8886	szDisBuf[cchDis++] = ' ';
8887	memcpy(&szDisBuf[cchDis], pszAnnotation, cchAnnotation + 1);
8888	}
8889	}
8890	pHlp->pfnPrintf(pHlp, " %p: %s\n", pNativeCur, szDisBuf);
8891	}
8892	}
8893	else
8894	{
8895	# if defined(RT_ARCH_AMD64)
8896	pHlp->pfnPrintf(pHlp, " %p: %.*Rhxs - disassembly failure %Rrc\n",
8897	pNativeCur, RT_MIN(cNative - offNative, 16), pNativeCur, rc);
8898	# elif defined(RT_ARCH_ARM64)
8899	pHlp->pfnPrintf(pHlp, " %p: %#010RX32 - disassembly failure %Rrc\n", pNativeCur, *pNativeCur, rc);
8900	# else
8901	# error "Port me"
8902	# endif
8903	cbInstr = sizeof(paNative[0]);
8904	}
8905	offNative += cbInstr / sizeof(paNative[0]);
8906
8907	# else /* VBOX_WITH_IEM_USING_CAPSTONE_DISASSEMBLER */
8908	cs_insn *pInstr;
8909	size_t cInstrs = cs_disasm(hDisasm, (const uint8_t )pNativeCur, (cNative - offNative) sizeof(*pNativeCur),
8910	(uintptr_t)pNativeCur, 1, &pInstr);
8911	if (cInstrs > 0)
8912	{
8913	Assert(cInstrs == 1);
8914	const char *pszAnnotation = NULL;
8915	# if defined(RT_ARCH_ARM64)
8916	if ( (pInstr->id >= ARM64_INS_LD1 && pInstr->id < ARM64_INS_LSL)
8917	\|\| (pInstr->id >= ARM64_INS_ST1 && pInstr->id < ARM64_INS_SUB))
8918	{
8919	/* This is bit crappy, but the disassembler provides incomplete addressing details. */
8920	AssertCompile(IEMNATIVE_REG_FIXED_PVMCPU == 28 && IEMNATIVE_REG_FIXED_PCPUMCTX == 27);
8921	char *psz = strchr(pInstr->op_str, '[');
8922	if (psz && psz[1] == 'x' && psz[2] == '2' && (psz[3] == '7' \|\| psz[3] == '8'))
8923	{
8924	uint32_t const offVCpu = psz[3] == '8'? 0 : RT_UOFFSETOF(VMCPU, cpum.GstCtx);
8925	int32_t off = -1;
8926	psz += 4;
8927	if (*psz == ']')
8928	off = 0;
8929	else if (*psz == ',')
8930	{
8931	psz = RTStrStripL(psz + 1);
8932	if (*psz == '#')
8933	off = RTStrToInt32(&psz[1]);
8934	/** @todo deal with index registers and LSL as well... */
8935	}
8936	if (off >= 0)
8937	pszAnnotation = iemNativeDbgVCpuOffsetToName(offVCpu + (uint32_t)off);
8938	}
8939	}
8940	# endif
8941
8942	size_t const cchOp = strlen(pInstr->op_str);
8943	# if defined(RT_ARCH_AMD64)
8944	if (pszAnnotation)
8945	pHlp->pfnPrintf(pHlp, " %p: %.Rhxs %-7s %s%s ; %s\n",
8946	pNativeCur, pInstr->size, pNativeCur, pInstr->mnemonic, pInstr->op_str,
8947	cchOp < 55 ? 55 - cchOp : 0, "", pszAnnotation);
8948	else
8949	pHlp->pfnPrintf(pHlp, " %p: %.*Rhxs %-7s %s\n",
8950	pNativeCur, pInstr->size, pNativeCur, pInstr->mnemonic, pInstr->op_str);
8951
8952	# else
8953	if (pszAnnotation)
8954	pHlp->pfnPrintf(pHlp, " %p: %#010RX32 %-7s %s%*s ; %s\n",
8955	pNativeCur, *pNativeCur, pInstr->mnemonic, pInstr->op_str,
8956	cchOp < 55 ? 55 - cchOp : 0, "", pszAnnotation);
8957	else
8958	pHlp->pfnPrintf(pHlp, " %p: %#010RX32 %-7s %s\n",
8959	pNativeCur, *pNativeCur, pInstr->mnemonic, pInstr->op_str);
8960	# endif
8961	offNative += pInstr->size / sizeof(*pNativeCur);
8962	cs_free(pInstr, cInstrs);
8963	}
8964	else
8965	{
8966	# if defined(RT_ARCH_AMD64)
8967	pHlp->pfnPrintf(pHlp, " %p: %.*Rhxs - disassembly failure %d\n",
8968	pNativeCur, RT_MIN(cNative - offNative, 16), pNativeCur, cs_errno(hDisasm)));
8969	# else
8970	pHlp->pfnPrintf(pHlp, " %p: %#010RX32 - disassembly failure %d\n", pNativeCur, *pNativeCur, cs_errno(hDisasm));
8971	# endif
8972	offNative++;
8973	}
8974	# endif /* VBOX_WITH_IEM_USING_CAPSTONE_DISASSEMBLER */
8975	}
8976	}
8977	else
8978	#endif /* IEMNATIVE_WITH_TB_DEBUG_INFO */
8979	{
8980	/*
8981	* No debug info, just disassemble the x86 code and then the native code.
8982	*
8983	* First the guest code:
8984	*/
8985	for (unsigned i = 0; i < pTb->cRanges; i++)
8986	{
8987	RTGCPHYS GCPhysPc = pTb->aRanges[i].offPhysPage
8988	+ (pTb->aRanges[i].idxPhysPage == 0
8989	? pTb->GCPhysPc & ~(RTGCPHYS)GUEST_PAGE_OFFSET_MASK
8990	: pTb->aGCPhysPages[pTb->aRanges[i].idxPhysPage - 1]);
8991	pHlp->pfnPrintf(pHlp, " Range #%u: GCPhysPc=%RGp LB %#x [idxPg=%d]\n",
8992	i, GCPhysPc, pTb->aRanges[i].cbOpcodes, pTb->aRanges[i].idxPhysPage);
8993	unsigned off = pTb->aRanges[i].offOpcodes;
8994	/** @todo this ain't working when crossing pages! */
8995	unsigned const cbOpcodes = pTb->aRanges[i].cbOpcodes + off;
8996	while (off < cbOpcodes)
8997	{
8998	uint32_t cbInstr = 1;
8999	int rc = DISInstrWithPrefetchedBytes(GCPhysPc, enmGstCpuMode, DISOPTYPE_ALL,
9000	&pTb->pabOpcodes[off], cbOpcodes - off,
9001	iemNativeDisasReadBytesDummy, NULL, &Dis, &cbInstr);
9002	if (RT_SUCCESS(rc))
9003	{
9004	DISFormatYasmEx(&Dis, szDisBuf, sizeof(szDisBuf),
9005	DIS_FMT_FLAGS_BYTES_WIDTH_MAKE(10) \| DIS_FMT_FLAGS_BYTES_LEFT
9006	\| DIS_FMT_FLAGS_RELATIVE_BRANCH \| DIS_FMT_FLAGS_C_HEX,
9007	NULL /pfnGetSymbol/, NULL /pvUser/);
9008	pHlp->pfnPrintf(pHlp, " %RGp: %s\n", GCPhysPc, szDisBuf);
9009	GCPhysPc += cbInstr;
9010	off += cbInstr;
9011	}
9012	else
9013	{
9014	pHlp->pfnPrintf(pHlp, " %RGp: %.*Rhxs - disassembly failure %Rrc\n",
9015	GCPhysPc, cbOpcodes - off, &pTb->pabOpcodes[off], rc);
9016	break;
9017	}
9018	}
9019	}
9020
9021	/*
9022	* Then the native code:
9023	*/
9024	pHlp->pfnPrintf(pHlp, " Native code %p L %#x\n", paNative, cNative);
9025	while (offNative < cNative)
9026	{
9027	PCIEMNATIVEINSTR const pNativeCur = &paNative[offNative];
9028	# ifndef VBOX_WITH_IEM_USING_CAPSTONE_DISASSEMBLER
9029	uint32_t cbInstr = sizeof(paNative[0]);
9030	int const rc = DISInstr(pNativeCur, enmHstCpuMode, &Dis, &cbInstr);
9031	if (RT_SUCCESS(rc))
9032	{
9033	# if defined(RT_ARCH_AMD64)
9034	if (Dis.pCurInstr->uOpcode == OP_NOP && cbInstr == 7) /* iemNativeEmitMarker */
9035	{
9036	uint32_t const uInfo = (uint32_t const )&Dis.Instr.ab[3];
9037	if (RT_HIWORD(uInfo) < kIemThreadedFunc_End)
9038	pHlp->pfnPrintf(pHlp, "\n %p: nop ; marker: call #%u to %s (%u args) - %s\n",
9039	pNativeCur, uInfo & 0x7fff, g_apszIemThreadedFunctions[RT_HIWORD(uInfo)],
9040	g_acIemThreadedFunctionUsedArgs[RT_HIWORD(uInfo)],
9041	uInfo & 0x8000 ? "recompiled" : "todo");
9042	else if ((uInfo & ~RT_BIT_32(31)) < RT_ELEMENTS(a_apszMarkers))
9043	pHlp->pfnPrintf(pHlp, " %p: nop ; marker: %s\n", pNativeCur, a_apszMarkers[uInfo & ~RT_BIT_32(31)]);
9044	else
9045	pHlp->pfnPrintf(pHlp, " %p: nop ; unknown marker: %#x (%d)\n", pNativeCur, uInfo, uInfo);
9046	}
9047	else
9048	# endif
9049	{
9050	# ifdef RT_ARCH_AMD64
9051	DISFormatYasmEx(&Dis, szDisBuf, sizeof(szDisBuf),
9052	DIS_FMT_FLAGS_BYTES_WIDTH_MAKE(10) \| DIS_FMT_FLAGS_BYTES_LEFT
9053	\| DIS_FMT_FLAGS_RELATIVE_BRANCH \| DIS_FMT_FLAGS_C_HEX,
9054	NULL /pfnGetSymbol/, NULL /pvUser/);
9055	# elif defined(RT_ARCH_ARM64)
9056	DISFormatArmV8Ex(&Dis, szDisBuf, sizeof(szDisBuf),
9057	DIS_FMT_FLAGS_BYTES_LEFT \| DIS_FMT_FLAGS_RELATIVE_BRANCH \| DIS_FMT_FLAGS_C_HEX,
9058	NULL /pfnGetSymbol/, NULL /pvUser/);
9059	# else
9060	# error "Port me"
9061	# endif
9062	pHlp->pfnPrintf(pHlp, " %p: %s\n", pNativeCur, szDisBuf);
9063	}
9064	}
9065	else
9066	{
9067	# if defined(RT_ARCH_AMD64)
9068	pHlp->pfnPrintf(pHlp, " %p: %.*Rhxs - disassembly failure %Rrc\n",
9069	pNativeCur, RT_MIN(cNative - offNative, 16), pNativeCur, rc);
9070	# else
9071	pHlp->pfnPrintf(pHlp, " %p: %#010RX32 - disassembly failure %Rrc\n", pNativeCur, *pNativeCur, rc);
9072	# endif
9073	cbInstr = sizeof(paNative[0]);
9074	}
9075	offNative += cbInstr / sizeof(paNative[0]);
9076
9077	# else /* VBOX_WITH_IEM_USING_CAPSTONE_DISASSEMBLER */
9078	cs_insn *pInstr;
9079	size_t cInstrs = cs_disasm(hDisasm, (const uint8_t )pNativeCur, (cNative - offNative) sizeof(*pNativeCur),
9080	(uintptr_t)pNativeCur, 1, &pInstr);
9081	if (cInstrs > 0)
9082	{
9083	Assert(cInstrs == 1);
9084	# if defined(RT_ARCH_AMD64)
9085	pHlp->pfnPrintf(pHlp, " %p: %.*Rhxs %-7s %s\n",
9086	pNativeCur, pInstr->size, pNativeCur, pInstr->mnemonic, pInstr->op_str);
9087	# else
9088	pHlp->pfnPrintf(pHlp, " %p: %#010RX32 %-7s %s\n",
9089	pNativeCur, *pNativeCur, pInstr->mnemonic, pInstr->op_str);
9090	# endif
9091	offNative += pInstr->size / sizeof(*pNativeCur);
9092	cs_free(pInstr, cInstrs);
9093	}
9094	else
9095	{
9096	# if defined(RT_ARCH_AMD64)
9097	pHlp->pfnPrintf(pHlp, " %p: %.*Rhxs - disassembly failure %d\n",
9098	pNativeCur, RT_MIN(cNative - offNative, 16), pNativeCur, cs_errno(hDisasm)));
9099	# else
9100	pHlp->pfnPrintf(pHlp, " %p: %#010RX32 - disassembly failure %d\n", pNativeCur, *pNativeCur, cs_errno(hDisasm));
9101	# endif
9102	offNative++;
9103	}
9104	# endif /* VBOX_WITH_IEM_USING_CAPSTONE_DISASSEMBLER */
9105	}
9106	}
9107
9108	#ifdef VBOX_WITH_IEM_USING_CAPSTONE_DISASSEMBLER
9109	/* Cleanup. */
9110	cs_close(&hDisasm);
9111	#endif
9112	}
9113
9114
9115	/**
9116	* Recompiles the given threaded TB into a native one.
9117	*
9118	* In case of failure the translation block will be returned as-is.
9119	*
9120	* @returns pTb.
9121	* @param pVCpu The cross context virtual CPU structure of the calling
9122	* thread.
9123	* @param pTb The threaded translation to recompile to native.
9124	*/
9125	DECLHIDDEN(PIEMTB) iemNativeRecompile(PVMCPUCC pVCpu, PIEMTB pTb) RT_NOEXCEPT
9126	{
9127	STAM_REL_PROFILE_START(&pVCpu->iem.s.StatNativeRecompilation, a);
9128
9129	/*
9130	* The first time thru, we allocate the recompiler state, the other times
9131	* we just need to reset it before using it again.
9132	*/
9133	PIEMRECOMPILERSTATE pReNative = pVCpu->iem.s.pNativeRecompilerStateR3;
9134	if (RT_LIKELY(pReNative))
9135	iemNativeReInit(pReNative, pTb);
9136	else
9137	{
9138	pReNative = iemNativeInit(pVCpu, pTb);
9139	AssertReturn(pReNative, pTb);
9140	}
9141
9142	#ifdef IEMNATIVE_WITH_LIVENESS_ANALYSIS
9143	/*
9144	* First do liveness analysis. This is done backwards.
9145	*/
9146	{
9147	uint32_t idxCall = pTb->Thrd.cCalls;
9148	if (idxCall <= pReNative->cLivenessEntriesAlloc)
9149	{ /* likely */ }
9150	else
9151	{
9152	uint32_t cAlloc = RT_MAX(pReNative->cLivenessEntriesAlloc, _4K);
9153	while (idxCall > cAlloc)
9154	cAlloc *= 2;
9155	void pvNew = RTMemRealloc(pReNative->paLivenessEntries, sizeof(pReNative->paLivenessEntries[0]) cAlloc);
9156	AssertReturn(pvNew, pTb);
9157	pReNative->paLivenessEntries = (PIEMLIVENESSENTRY)pvNew;
9158	pReNative->cLivenessEntriesAlloc = cAlloc;
9159	}
9160	AssertReturn(idxCall > 0, pTb);
9161	PIEMLIVENESSENTRY const paLivenessEntries = pReNative->paLivenessEntries;
9162
9163	/* The initial (final) entry. */
9164	idxCall--;
9165	IEM_LIVENESS_RAW_INIT_AS_UNUSED(&paLivenessEntries[idxCall]);
9166
9167	/* Loop backwards thru the calls and fill in the other entries. */
9168	PCIEMTHRDEDCALLENTRY pCallEntry = &pTb->Thrd.paCalls[idxCall];
9169	while (idxCall > 0)
9170	{
9171	PFNIEMNATIVELIVENESSFUNC const pfnLiveness = g_apfnIemNativeLivenessFunctions[pCallEntry->enmFunction];
9172	if (pfnLiveness)
9173	pfnLiveness(pCallEntry, &paLivenessEntries[idxCall], &paLivenessEntries[idxCall - 1]);
9174	else
9175	IEM_LIVENESS_RAW_INIT_WITH_XCPT_OR_CALL(&paLivenessEntries[idxCall - 1], &paLivenessEntries[idxCall]);
9176	pCallEntry--;
9177	idxCall--;
9178	}
9179
9180	# ifdef VBOX_WITH_STATISTICS
9181	/* Check if there are any EFLAGS optimization to be had here. This requires someone settings them
9182	to 'clobbered' rather that 'input'. */
9183	/** @todo */
9184	# endif
9185	}
9186	#endif
9187
9188	/*
9189	* Recompiling and emitting code is done using try/throw/catch or setjmp/longjmp
9190	* for aborting if an error happens.
9191	*/
9192	uint32_t cCallsLeft = pTb->Thrd.cCalls;
9193	#ifdef LOG_ENABLED
9194	uint32_t const cCallsOrg = cCallsLeft;
9195	#endif
9196	uint32_t off = 0;
9197	int rc = VINF_SUCCESS;
9198	IEMNATIVE_TRY_SETJMP(pReNative, rc)
9199	{
9200	/*
9201	* Emit prolog code (fixed).
9202	*/
9203	off = iemNativeEmitProlog(pReNative, off);
9204
9205	/*
9206	* Convert the calls to native code.
9207	*/
9208	#ifdef IEMNATIVE_WITH_TB_DEBUG_INFO
9209	int32_t iGstInstr = -1;
9210	#endif
9211	#ifndef VBOX_WITHOUT_RELEASE_STATISTICS
9212	uint32_t cThreadedCalls = 0;
9213	uint32_t cRecompiledCalls = 0;
9214	#endif
9215	#if defined(IEMNATIVE_WITH_LIVENESS_ANALYSIS) \|\| defined(VBOX_STRICT) \|\| defined(LOG_ENABLED)
9216	uint32_t idxCurCall = 0;
9217	#endif
9218	PCIEMTHRDEDCALLENTRY pCallEntry = pTb->Thrd.paCalls;
9219	pReNative->fExec = pTb->fFlags & IEMTB_F_IEM_F_MASK;
9220	while (cCallsLeft-- > 0)
9221	{
9222	PFNIEMNATIVERECOMPFUNC const pfnRecom = g_apfnIemNativeRecompileFunctions[pCallEntry->enmFunction];
9223	#ifdef IEMNATIVE_WITH_LIVENESS_ANALYSIS
9224	pReNative->idxCurCall = idxCurCall;
9225	#endif
9226
9227	/*
9228	* Debug info, assembly markup and statistics.
9229	*/
9230	#if defined(IEMNATIVE_WITH_TB_DEBUG_INFO) \|\| !defined(IEMNATIVE_WITH_BLTIN_CHECKMODE)
9231	if (pCallEntry->enmFunction == kIemThreadedFunc_BltIn_CheckMode)
9232	pReNative->fExec = pCallEntry->auParams[0] & IEMTB_F_IEM_F_MASK;
9233	#endif
9234	#ifdef IEMNATIVE_WITH_TB_DEBUG_INFO
9235	iemNativeDbgInfoAddNativeOffset(pReNative, off);
9236	if (iGstInstr < (int32_t)pCallEntry->idxInstr)
9237	{
9238	if (iGstInstr < (int32_t)pTb->cInstructions)
9239	iemNativeDbgInfoAddGuestInstruction(pReNative, pReNative->fExec);
9240	else
9241	Assert(iGstInstr == pTb->cInstructions);
9242	iGstInstr = pCallEntry->idxInstr;
9243	}
9244	iemNativeDbgInfoAddThreadedCall(pReNative, (IEMTHREADEDFUNCS)pCallEntry->enmFunction, pfnRecom != NULL);
9245	#endif
9246	#if defined(VBOX_STRICT)
9247	off = iemNativeEmitMarker(pReNative, off,
9248	RT_MAKE_U32(idxCurCall \| (pfnRecom ? 0x8000 : 0), pCallEntry->enmFunction));
9249	#endif
9250	#if defined(VBOX_STRICT)
9251	iemNativeRegAssertSanity(pReNative);
9252	#endif
9253	#ifdef VBOX_WITH_STATISTICS
9254	off = iemNativeEmitThreadCallStats(pReNative, off, pCallEntry);
9255	#endif
9256
9257	/*
9258	* Actual work.
9259	*/
9260	Log2(("%u[%u]: %s%s\n", idxCurCall, pCallEntry->idxInstr, g_apszIemThreadedFunctions[pCallEntry->enmFunction],
9261	pfnRecom ? "(recompiled)" : "(todo)"));
9262	if (pfnRecom) /** @todo stats on this. */
9263	{
9264	off = pfnRecom(pReNative, off, pCallEntry);
9265	STAM_REL_STATS({cRecompiledCalls++;});
9266	}
9267	else
9268	{
9269	off = iemNativeEmitThreadedCall(pReNative, off, pCallEntry);
9270	STAM_REL_STATS({cThreadedCalls++;});
9271	}
9272	Assert(off <= pReNative->cInstrBufAlloc);
9273	Assert(pReNative->cCondDepth == 0);
9274
9275	#if defined(LOG_ENABLED) && defined(IEMNATIVE_WITH_LIVENESS_ANALYSIS)
9276	if (LogIs2Enabled())
9277	{
9278	PCIEMLIVENESSENTRY pLivenessEntry = &pReNative->paLivenessEntries[idxCurCall];
9279	# ifndef IEMLIVENESS_EXTENDED_LAYOUT
9280	static const char s_achState[] = "CUXI";
9281	# else
9282	static const char s_achState[] = "UxRrWwMmCcQqKkNn";
9283	# endif
9284
9285	char szGpr[17];
9286	for (unsigned i = 0; i < 16; i++)
9287	szGpr[i] = s_achState[iemNativeLivenessGetStateByGstRegEx(pLivenessEntry, i + kIemNativeGstReg_GprFirst)];
9288	szGpr[16] = '\0';
9289
9290	char szSegBase[X86_SREG_COUNT + 1];
9291	char szSegLimit[X86_SREG_COUNT + 1];
9292	char szSegAttrib[X86_SREG_COUNT + 1];
9293	char szSegSel[X86_SREG_COUNT + 1];
9294	for (unsigned i = 0; i < X86_SREG_COUNT; i++)
9295	{
9296	szSegBase[i] = s_achState[iemNativeLivenessGetStateByGstRegEx(pLivenessEntry, i + kIemNativeGstReg_SegBaseFirst)];
9297	szSegAttrib[i] = s_achState[iemNativeLivenessGetStateByGstRegEx(pLivenessEntry, i + kIemNativeGstReg_SegAttribFirst)];
9298	szSegLimit[i] = s_achState[iemNativeLivenessGetStateByGstRegEx(pLivenessEntry, i + kIemNativeGstReg_SegLimitFirst)];
9299	szSegSel[i] = s_achState[iemNativeLivenessGetStateByGstRegEx(pLivenessEntry, i + kIemNativeGstReg_SegSelFirst)];
9300	}
9301	szSegBase[X86_SREG_COUNT] = szSegAttrib[X86_SREG_COUNT] = szSegLimit[X86_SREG_COUNT]
9302	= szSegSel[X86_SREG_COUNT] = '\0';
9303
9304	char szEFlags[8];
9305	for (unsigned i = 0; i < 7; i++)
9306	szEFlags[i] = s_achState[iemNativeLivenessGetStateByGstRegEx(pLivenessEntry, i + kIemNativeGstReg_EFlags)];
9307	szEFlags[7] = '\0';
9308
9309	Log2(("liveness: grp=%s segbase=%s segattr=%s seglim=%s segsel=%s efl=%s\n",
9310	szGpr, szSegBase, szSegAttrib, szSegLimit, szSegSel, szEFlags));
9311	}
9312	#endif
9313
9314	/*
9315	* Advance.
9316	*/
9317	pCallEntry++;
9318	#if defined(IEMNATIVE_WITH_LIVENESS_ANALYSIS) \|\| defined(VBOX_STRICT) \|\| defined(LOG_ENABLED)
9319	idxCurCall++;
9320	#endif
9321	}
9322
9323	STAM_REL_PROFILE_ADD_PERIOD(&pVCpu->iem.s.StatNativeCallsRecompiled, cRecompiledCalls);
9324	STAM_REL_PROFILE_ADD_PERIOD(&pVCpu->iem.s.StatNativeCallsThreaded, cThreadedCalls);
9325	if (!cThreadedCalls)
9326	STAM_REL_COUNTER_INC(&pVCpu->iem.s.StatNativeFullyRecompiledTbs);
9327
9328	/*
9329	* Emit the epilog code.
9330	*/
9331	uint32_t idxReturnLabel;
9332	off = iemNativeEmitEpilog(pReNative, off, &idxReturnLabel);
9333
9334	/*
9335	* Generate special jump labels.
9336	*/
9337	if (pReNative->bmLabelTypes & RT_BIT_64(kIemNativeLabelType_ReturnBreak))
9338	off = iemNativeEmitReturnBreak(pReNative, off, idxReturnLabel);
9339	if (pReNative->bmLabelTypes & RT_BIT_64(kIemNativeLabelType_ReturnWithFlags))
9340	off = iemNativeEmitReturnWithFlags(pReNative, off, idxReturnLabel);
9341	if (pReNative->bmLabelTypes & RT_BIT_64(kIemNativeLabelType_RaiseGp0))
9342	off = iemNativeEmitRaiseGp0(pReNative, off, idxReturnLabel);
9343	if (pReNative->bmLabelTypes & RT_BIT_64(kIemNativeLabelType_RaiseNm))
9344	off = iemNativeEmitRaiseNm(pReNative, off, idxReturnLabel);
9345	if (pReNative->bmLabelTypes & RT_BIT_64(kIemNativeLabelType_RaiseUd))
9346	off = iemNativeEmitRaiseUd(pReNative, off, idxReturnLabel);
9347	if (pReNative->bmLabelTypes & RT_BIT_64(kIemNativeLabelType_RaiseMf))
9348	off = iemNativeEmitRaiseMf(pReNative, off, idxReturnLabel);
9349	if (pReNative->bmLabelTypes & RT_BIT_64(kIemNativeLabelType_RaiseXf))
9350	off = iemNativeEmitRaiseXf(pReNative, off, idxReturnLabel);
9351	if (pReNative->bmLabelTypes & RT_BIT_64(kIemNativeLabelType_ObsoleteTb))
9352	off = iemNativeEmitObsoleteTb(pReNative, off, idxReturnLabel);
9353	if (pReNative->bmLabelTypes & RT_BIT_64(kIemNativeLabelType_NeedCsLimChecking))
9354	off = iemNativeEmitNeedCsLimChecking(pReNative, off, idxReturnLabel);
9355	if (pReNative->bmLabelTypes & RT_BIT_64(kIemNativeLabelType_CheckBranchMiss))
9356	off = iemNativeEmitCheckBranchMiss(pReNative, off, idxReturnLabel);
9357	}
9358	IEMNATIVE_CATCH_LONGJMP_BEGIN(pReNative, rc);
9359	{
9360	Log(("iemNativeRecompile: Caught %Rrc while recompiling!\n", rc));
9361	return pTb;
9362	}
9363	IEMNATIVE_CATCH_LONGJMP_END(pReNative);
9364	Assert(off <= pReNative->cInstrBufAlloc);
9365
9366	/*
9367	* Make sure all labels has been defined.
9368	*/
9369	PIEMNATIVELABEL const paLabels = pReNative->paLabels;
9370	#ifdef VBOX_STRICT
9371	uint32_t const cLabels = pReNative->cLabels;
9372	for (uint32_t i = 0; i < cLabels; i++)
9373	AssertMsgReturn(paLabels[i].off < off, ("i=%d enmType=%d\n", i, paLabels[i].enmType), pTb);
9374	#endif
9375
9376	/*
9377	* Allocate executable memory, copy over the code we've generated.
9378	*/
9379	PIEMTBALLOCATOR const pTbAllocator = pVCpu->iem.s.pTbAllocatorR3;
9380	if (pTbAllocator->pDelayedFreeHead)
9381	iemTbAllocatorProcessDelayedFrees(pVCpu, pVCpu->iem.s.pTbAllocatorR3);
9382
9383	PIEMNATIVEINSTR const paFinalInstrBuf = (PIEMNATIVEINSTR)iemExecMemAllocatorAlloc(pVCpu, off * sizeof(IEMNATIVEINSTR));
9384	AssertReturn(paFinalInstrBuf, pTb);
9385	memcpy(paFinalInstrBuf, pReNative->pInstrBuf, off * sizeof(paFinalInstrBuf[0]));
9386
9387	/*
9388	* Apply fixups.
9389	*/
9390	PIEMNATIVEFIXUP const paFixups = pReNative->paFixups;
9391	uint32_t const cFixups = pReNative->cFixups;
9392	for (uint32_t i = 0; i < cFixups; i++)
9393	{
9394	Assert(paFixups[i].off < off);
9395	Assert(paFixups[i].idxLabel < cLabels);
9396	AssertMsg(paLabels[paFixups[i].idxLabel].off < off,
9397	("idxLabel=%d enmType=%d off=%#x (max %#x)\n", paFixups[i].idxLabel,
9398	paLabels[paFixups[i].idxLabel].enmType, paLabels[paFixups[i].idxLabel].off, off));
9399	RTPTRUNION const Ptr = { &paFinalInstrBuf[paFixups[i].off] };
9400	switch (paFixups[i].enmType)
9401	{
9402	#if defined(RT_ARCH_AMD64) \|\| defined(RT_ARCH_X86)
9403	case kIemNativeFixupType_Rel32:
9404	Assert(paFixups[i].off + 4 <= off);
9405	*Ptr.pi32 = paLabels[paFixups[i].idxLabel].off - paFixups[i].off + paFixups[i].offAddend;
9406	continue;
9407
9408	#elif defined(RT_ARCH_ARM64)
9409	case kIemNativeFixupType_RelImm26At0:
9410	{
9411	Assert(paFixups[i].off < off);
9412	int32_t const offDisp = paLabels[paFixups[i].idxLabel].off - paFixups[i].off + paFixups[i].offAddend;
9413	Assert(offDisp >= -262144 && offDisp < 262144);
9414	Ptr.pu32 = (Ptr.pu32 & UINT32_C(0xfc000000)) \| ((uint32_t)offDisp & UINT32_C(0x03ffffff));
9415	continue;
9416	}
9417
9418	case kIemNativeFixupType_RelImm19At5:
9419	{
9420	Assert(paFixups[i].off < off);
9421	int32_t const offDisp = paLabels[paFixups[i].idxLabel].off - paFixups[i].off + paFixups[i].offAddend;
9422	Assert(offDisp >= -262144 && offDisp < 262144);
9423	Ptr.pu32 = (Ptr.pu32 & UINT32_C(0xff00001f)) \| (((uint32_t)offDisp & UINT32_C(0x0007ffff)) << 5);
9424	continue;
9425	}
9426
9427	case kIemNativeFixupType_RelImm14At5:
9428	{
9429	Assert(paFixups[i].off < off);
9430	int32_t const offDisp = paLabels[paFixups[i].idxLabel].off - paFixups[i].off + paFixups[i].offAddend;
9431	Assert(offDisp >= -8192 && offDisp < 8192);
9432	Ptr.pu32 = (Ptr.pu32 & UINT32_C(0xfff8001f)) \| (((uint32_t)offDisp & UINT32_C(0x00003fff)) << 5);
9433	continue;
9434	}
9435
9436	#endif
9437	case kIemNativeFixupType_Invalid:
9438	case kIemNativeFixupType_End:
9439	break;
9440	}
9441	AssertFailed();
9442	}
9443
9444	iemExecMemAllocatorReadyForUse(pVCpu, paFinalInstrBuf, off * sizeof(IEMNATIVEINSTR));
9445	STAM_REL_PROFILE_ADD_PERIOD(&pVCpu->iem.s.StatTbNativeCode, off * sizeof(IEMNATIVEINSTR));
9446
9447	/*
9448	* Convert the translation block.
9449	*/
9450	RTMemFree(pTb->Thrd.paCalls);
9451	pTb->Native.paInstructions = paFinalInstrBuf;
9452	pTb->Native.cInstructions = off;
9453	pTb->fFlags = (pTb->fFlags & ~IEMTB_F_TYPE_MASK) \| IEMTB_F_TYPE_NATIVE;
9454	#ifdef IEMNATIVE_WITH_TB_DEBUG_INFO
9455	pTb->pDbgInfo = (PIEMTBDBG)RTMemDup(pReNative->pDbgInfo, /* non-fatal, so not return check. */
9456	RT_UOFFSETOF_DYN(IEMTBDBG, aEntries[pReNative->pDbgInfo->cEntries]));
9457	#endif
9458
9459	Assert(pTbAllocator->cThreadedTbs > 0);
9460	pTbAllocator->cThreadedTbs -= 1;
9461	pTbAllocator->cNativeTbs += 1;
9462	Assert(pTbAllocator->cNativeTbs <= pTbAllocator->cTotalTbs);
9463
9464	#ifdef LOG_ENABLED
9465	/*
9466	* Disassemble to the log if enabled.
9467	*/
9468	if (LogIs3Enabled())
9469	{
9470	Log3(("----------------------------------------- %d calls ---------------------------------------\n", cCallsOrg));
9471	iemNativeDisassembleTb(pTb, DBGFR3InfoLogHlp());
9472	# if defined(DEBUG_bird) \|\| defined(DEBUG_aeichner)
9473	RTLogFlush(NULL);
9474	# endif
9475	}
9476	#endif
9477	/iemNativeDisassembleTb(pTb, DBGFR3InfoLogRelHlp());/
9478
9479	STAM_REL_PROFILE_STOP(&pVCpu->iem.s.StatNativeRecompilation, a);
9480	return pTb;
9481	}
9482

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

source: vbox/trunk/src/VBox/VMM/VMMAll/IEMAllN8veRecompiler.cpp@ 103807

Download in other formats: