IEMAllCImplStrInstr.cpp.h@ 47681

Last change on this file since 47681 was 47444, checked in by vboxsync, 11 years ago
IEM,HM,PGM: Started on string I/O optimizations using IEM (disabled). Cleaned up confusing status code handling in hmR0VmxCheckForceFlags (involving PGM) as well as some use of incorrect doxygen groups (@name).
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 52.6 KB

Line
1	/* $Id: IEMAllCImplStrInstr.cpp.h 47444 2013-07-29 00:37:31Z vboxsync $ */
2	/** @file
3	* IEM - String Instruction Implementation Code Template.
4	*/
5
6	/*
7	* Copyright (C) 2011-2012 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*/
17
18
19	/*******************************************************************************
20	* Defined Constants And Macros *
21	*******************************************************************************/
22	#if OP_SIZE == 8
23	# define OP_rAX al
24	#elif OP_SIZE == 16
25	# define OP_rAX ax
26	#elif OP_SIZE == 32
27	# define OP_rAX eax
28	#elif OP_SIZE == 64
29	# define OP_rAX rax
30	#else
31	# error "Bad OP_SIZE."
32	#endif
33	#define OP_TYPE RT_CONCAT3(uint,OP_SIZE,_t)
34
35	#if ADDR_SIZE == 16
36	# define ADDR_rDI di
37	# define ADDR_rSI si
38	# define ADDR_rCX cx
39	# define ADDR2_TYPE uint32_t
40	#elif ADDR_SIZE == 32
41	# define ADDR_rDI edi
42	# define ADDR_rSI esi
43	# define ADDR_rCX ecx
44	# define ADDR2_TYPE uint32_t
45	#elif ADDR_SIZE == 64
46	# define ADDR_rDI rdi
47	# define ADDR_rSI rsi
48	# define ADDR_rCX rcx
49	# define ADDR2_TYPE uint64_t
50	# define IS_64_BIT_CODE(a_pIemCpu) (true)
51	#else
52	# error "Bad ADDR_SIZE."
53	#endif
54	#define ADDR_TYPE RT_CONCAT3(uint,ADDR_SIZE,_t)
55
56	#if ADDR_SIZE == 64 \|\| OP_SIZE == 64
57	# define IS_64_BIT_CODE(a_pIemCpu) (true)
58	#elif ADDR_SIZE == 32
59	# define IS_64_BIT_CODE(a_pIemCpu) ((a_pIemCpu)->enmCpuMode == IEMMODE_64BIT)
60	#else
61	# define IS_64_BIT_CODE(a_pIemCpu) (false)
62	#endif
63
64
65	/**
66	* Implements 'REPE CMPS'.
67	*/
68	IEM_CIMPL_DEF_1(RT_CONCAT4(iemCImpl_repe_cmps_op,OP_SIZE,_addr,ADDR_SIZE), uint8_t, iEffSeg)
69	{
70	PCPUMCTX pCtx = pIemCpu->CTX_SUFF(pCtx);
71
72	/*
73	* Setup.
74	*/
75	ADDR_TYPE uCounterReg = pCtx->ADDR_rCX;
76	if (uCounterReg == 0)
77	{
78	iemRegAddToRip(pIemCpu, cbInstr);
79	return VINF_SUCCESS;
80	}
81
82	PCCPUMSELREGHID pSrc1Hid = iemSRegGetHid(pIemCpu, iEffSeg);
83	uint64_t uSrc1Base;
84	VBOXSTRICTRC rcStrict = iemMemSegCheckReadAccessEx(pIemCpu, pSrc1Hid, iEffSeg, &uSrc1Base);
85	if (rcStrict != VINF_SUCCESS)
86	return rcStrict;
87
88	uint64_t uSrc2Base;
89	rcStrict = iemMemSegCheckReadAccessEx(pIemCpu, &pCtx->es, X86_SREG_ES, &uSrc2Base);
90	if (rcStrict != VINF_SUCCESS)
91	return rcStrict;
92
93	int8_t const cbIncr = pCtx->eflags.Bits.u1DF ? -(OP_SIZE / 8) : (OP_SIZE / 8);
94	ADDR_TYPE uSrc1AddrReg = pCtx->ADDR_rSI;
95	ADDR_TYPE uSrc2AddrReg = pCtx->ADDR_rDI;
96	uint32_t uEFlags = pCtx->eflags.u;
97
98	/*
99	* The loop.
100	*/
101	do
102	{
103	/*
104	* Do segmentation and virtual page stuff.
105	*/
106	ADDR2_TYPE uVirtSrc1Addr = uSrc1AddrReg + (ADDR2_TYPE)uSrc1Base;
107	ADDR2_TYPE uVirtSrc2Addr = uSrc2AddrReg + (ADDR2_TYPE)uSrc2Base;
108	uint32_t cLeftSrc1Page = (PAGE_SIZE - (uVirtSrc1Addr & PAGE_OFFSET_MASK)) / (OP_SIZE / 8);
109	if (cLeftSrc1Page > uCounterReg)
110	cLeftSrc1Page = uCounterReg;
111	uint32_t cLeftSrc2Page = (PAGE_SIZE - (uVirtSrc2Addr & PAGE_OFFSET_MASK)) / (OP_SIZE / 8);
112	uint32_t cLeftPage = RT_MIN(cLeftSrc1Page, cLeftSrc2Page);
113
114	if ( cLeftPage > 0 /* can be null if unaligned, do one fallback round. */
115	&& cbIncr > 0 /** @todo Optimize reverse direction string ops. */
116	&& ( IS_64_BIT_CODE(pIemCpu)
117	\|\| ( uSrc1AddrReg < pSrc1Hid->u32Limit
118	&& uSrc1AddrReg + (cLeftPage * (OP_SIZE / 8)) <= pSrc1Hid->u32Limit
119	&& uSrc2AddrReg < pCtx->es.u32Limit
120	&& uSrc2AddrReg + (cLeftPage * (OP_SIZE / 8)) <= pCtx->es.u32Limit)
121	)
122	)
123	{
124	RTGCPHYS GCPhysSrc1Mem;
125	rcStrict = iemMemPageTranslateAndCheckAccess(pIemCpu, uVirtSrc1Addr, IEM_ACCESS_DATA_R, &GCPhysSrc1Mem);
126	if (rcStrict != VINF_SUCCESS)
127	return rcStrict;
128
129	RTGCPHYS GCPhysSrc2Mem;
130	rcStrict = iemMemPageTranslateAndCheckAccess(pIemCpu, uVirtSrc2Addr, IEM_ACCESS_DATA_R, &GCPhysSrc2Mem);
131	if (rcStrict != VINF_SUCCESS)
132	return rcStrict;
133
134	/*
135	* If we can map the page without trouble, do a block processing
136	* until the end of the current page.
137	*/
138	PGMPAGEMAPLOCK PgLockSrc2Mem;
139	OP_TYPE const *puSrc2Mem;
140	rcStrict = iemMemPageMap(pIemCpu, GCPhysSrc2Mem, IEM_ACCESS_DATA_R, (void **)&puSrc2Mem, &PgLockSrc2Mem);
141	if (rcStrict == VINF_SUCCESS)
142	{
143	PGMPAGEMAPLOCK PgLockSrc1Mem;
144	OP_TYPE const *puSrc1Mem;
145	rcStrict = iemMemPageMap(pIemCpu, GCPhysSrc1Mem, IEM_ACCESS_DATA_R, (void **)&puSrc1Mem, &PgLockSrc1Mem);
146	if (rcStrict == VINF_SUCCESS)
147	{
148	if (!memcmp(puSrc2Mem, puSrc1Mem, cLeftPage * (OP_SIZE / 8)))
149	{
150	/* All matches, only compare the last itme to get the right eflags. */
151	RT_CONCAT(iemAImpl_cmp_u,OP_SIZE)((OP_TYPE *)&puSrc1Mem[cLeftPage-1], puSrc2Mem[cLeftPage-1], &uEFlags);
152	uSrc1AddrReg += cLeftPage * cbIncr;
153	uSrc2AddrReg += cLeftPage * cbIncr;
154	uCounterReg -= cLeftPage;
155	}
156	else
157	{
158	/* Some mismatch, compare each item (and keep volatile
159	memory in mind). */
160	uint32_t off = 0;
161	do
162	{
163	RT_CONCAT(iemAImpl_cmp_u,OP_SIZE)((OP_TYPE *)&puSrc1Mem[off], puSrc2Mem[off], &uEFlags);
164	off++;
165	} while ( off < cLeftPage
166	&& (uEFlags & X86_EFL_ZF));
167	uSrc1AddrReg += cbIncr * off;
168	uSrc2AddrReg += cbIncr * off;
169	uCounterReg -= off;
170	}
171
172	/* Update the registers before looping. */
173	pCtx->ADDR_rCX = uCounterReg;
174	pCtx->ADDR_rSI = uSrc1AddrReg;
175	pCtx->ADDR_rDI = uSrc2AddrReg;
176	pCtx->eflags.u = uEFlags;
177
178	iemMemPageUnmap(pIemCpu, GCPhysSrc1Mem, IEM_ACCESS_DATA_R, puSrc1Mem, &PgLockSrc1Mem);
179	iemMemPageUnmap(pIemCpu, GCPhysSrc2Mem, IEM_ACCESS_DATA_R, puSrc2Mem, &PgLockSrc2Mem);
180	continue;
181	}
182	}
183	iemMemPageUnmap(pIemCpu, GCPhysSrc2Mem, IEM_ACCESS_DATA_R, puSrc2Mem, &PgLockSrc2Mem);
184	}
185
186	/*
187	* Fallback - slow processing till the end of the current page.
188	* In the cross page boundrary case we will end up here with cLeftPage
189	* as 0, we execute one loop then.
190	*/
191	do
192	{
193	OP_TYPE uValue1;
194	rcStrict = RT_CONCAT(iemMemFetchDataU,OP_SIZE)(pIemCpu, &uValue1, iEffSeg, uSrc1AddrReg);
195	if (rcStrict != VINF_SUCCESS)
196	return rcStrict;
197	OP_TYPE uValue2;
198	rcStrict = RT_CONCAT(iemMemFetchDataU,OP_SIZE)(pIemCpu, &uValue2, X86_SREG_ES, uSrc2AddrReg);
199	if (rcStrict != VINF_SUCCESS)
200	return rcStrict;
201	RT_CONCAT(iemAImpl_cmp_u,OP_SIZE)(&uValue1, uValue2, &uEFlags);
202
203	pCtx->ADDR_rSI = uSrc1AddrReg += cbIncr;
204	pCtx->ADDR_rDI = uSrc2AddrReg += cbIncr;
205	pCtx->ADDR_rCX = --uCounterReg;
206	pCtx->eflags.u = uEFlags;
207	cLeftPage--;
208	} while ( (int32_t)cLeftPage > 0
209	&& (uEFlags & X86_EFL_ZF));
210	} while ( uCounterReg != 0
211	&& (uEFlags & X86_EFL_ZF));
212
213	/*
214	* Done.
215	*/
216	iemRegAddToRip(pIemCpu, cbInstr);
217	return VINF_SUCCESS;
218	}
219
220
221	/**
222	* Implements 'REPNE CMPS'.
223	*/
224	IEM_CIMPL_DEF_1(RT_CONCAT4(iemCImpl_repne_cmps_op,OP_SIZE,_addr,ADDR_SIZE), uint8_t, iEffSeg)
225	{
226	PCPUMCTX pCtx = pIemCpu->CTX_SUFF(pCtx);
227
228	/*
229	* Setup.
230	*/
231	ADDR_TYPE uCounterReg = pCtx->ADDR_rCX;
232	if (uCounterReg == 0)
233	{
234	iemRegAddToRip(pIemCpu, cbInstr);
235	return VINF_SUCCESS;
236	}
237
238	PCCPUMSELREGHID pSrc1Hid = iemSRegGetHid(pIemCpu, iEffSeg);
239	uint64_t uSrc1Base;
240	VBOXSTRICTRC rcStrict = iemMemSegCheckReadAccessEx(pIemCpu, pSrc1Hid, iEffSeg, &uSrc1Base);
241	if (rcStrict != VINF_SUCCESS)
242	return rcStrict;
243
244	uint64_t uSrc2Base;
245	rcStrict = iemMemSegCheckReadAccessEx(pIemCpu, &pCtx->es, X86_SREG_ES, &uSrc2Base);
246	if (rcStrict != VINF_SUCCESS)
247	return rcStrict;
248
249	int8_t const cbIncr = pCtx->eflags.Bits.u1DF ? -(OP_SIZE / 8) : (OP_SIZE / 8);
250	ADDR_TYPE uSrc1AddrReg = pCtx->ADDR_rSI;
251	ADDR_TYPE uSrc2AddrReg = pCtx->ADDR_rDI;
252	uint32_t uEFlags = pCtx->eflags.u;
253
254	/*
255	* The loop.
256	*/
257	do
258	{
259	/*
260	* Do segmentation and virtual page stuff.
261	*/
262	ADDR2_TYPE uVirtSrc1Addr = uSrc1AddrReg + (ADDR2_TYPE)uSrc1Base;
263	ADDR2_TYPE uVirtSrc2Addr = uSrc2AddrReg + (ADDR2_TYPE)uSrc2Base;
264	uint32_t cLeftSrc1Page = (PAGE_SIZE - (uVirtSrc1Addr & PAGE_OFFSET_MASK)) / (OP_SIZE / 8);
265	if (cLeftSrc1Page > uCounterReg)
266	cLeftSrc1Page = uCounterReg;
267	uint32_t cLeftSrc2Page = (PAGE_SIZE - (uVirtSrc2Addr & PAGE_OFFSET_MASK)) / (OP_SIZE / 8);
268	uint32_t cLeftPage = RT_MIN(cLeftSrc1Page, cLeftSrc2Page);
269
270	if ( cLeftPage > 0 /* can be null if unaligned, do one fallback round. */
271	&& cbIncr > 0 /** @todo Optimize reverse direction string ops. */
272	&& ( IS_64_BIT_CODE(pIemCpu)
273	\|\| ( uSrc1AddrReg < pSrc1Hid->u32Limit
274	&& uSrc1AddrReg + (cLeftPage * (OP_SIZE / 8)) <= pSrc1Hid->u32Limit
275	&& uSrc2AddrReg < pCtx->es.u32Limit
276	&& uSrc2AddrReg + (cLeftPage * (OP_SIZE / 8)) <= pCtx->es.u32Limit)
277	)
278	)
279	{
280	RTGCPHYS GCPhysSrc1Mem;
281	rcStrict = iemMemPageTranslateAndCheckAccess(pIemCpu, uVirtSrc1Addr, IEM_ACCESS_DATA_R, &GCPhysSrc1Mem);
282	if (rcStrict != VINF_SUCCESS)
283	return rcStrict;
284
285	RTGCPHYS GCPhysSrc2Mem;
286	rcStrict = iemMemPageTranslateAndCheckAccess(pIemCpu, uVirtSrc2Addr, IEM_ACCESS_DATA_R, &GCPhysSrc2Mem);
287	if (rcStrict != VINF_SUCCESS)
288	return rcStrict;
289
290	/*
291	* If we can map the page without trouble, do a block processing
292	* until the end of the current page.
293	*/
294	OP_TYPE const *puSrc2Mem;
295	PGMPAGEMAPLOCK PgLockSrc2Mem;
296	rcStrict = iemMemPageMap(pIemCpu, GCPhysSrc2Mem, IEM_ACCESS_DATA_R, (void **)&puSrc2Mem, &PgLockSrc2Mem);
297	if (rcStrict == VINF_SUCCESS)
298	{
299	OP_TYPE const *puSrc1Mem;
300	PGMPAGEMAPLOCK PgLockSrc1Mem;
301	rcStrict = iemMemPageMap(pIemCpu, GCPhysSrc1Mem, IEM_ACCESS_DATA_R, (void **)&puSrc1Mem, &PgLockSrc1Mem);
302	if (rcStrict == VINF_SUCCESS)
303	{
304	if (memcmp(puSrc2Mem, puSrc1Mem, cLeftPage * (OP_SIZE / 8)))
305	{
306	/* All matches, only compare the last item to get the right eflags. */
307	RT_CONCAT(iemAImpl_cmp_u,OP_SIZE)((OP_TYPE *)&puSrc1Mem[cLeftPage-1], puSrc2Mem[cLeftPage-1], &uEFlags);
308	uSrc1AddrReg += cLeftPage * cbIncr;
309	uSrc2AddrReg += cLeftPage * cbIncr;
310	uCounterReg -= cLeftPage;
311	}
312	else
313	{
314	/* Some mismatch, compare each item (and keep volatile
315	memory in mind). */
316	uint32_t off = 0;
317	do
318	{
319	RT_CONCAT(iemAImpl_cmp_u,OP_SIZE)((OP_TYPE *)&puSrc1Mem[off], puSrc2Mem[off], &uEFlags);
320	off++;
321	} while ( off < cLeftPage
322	&& !(uEFlags & X86_EFL_ZF));
323	uSrc1AddrReg += cbIncr * off;
324	uSrc2AddrReg += cbIncr * off;
325	uCounterReg -= off;
326	}
327
328	/* Update the registers before looping. */
329	pCtx->ADDR_rCX = uCounterReg;
330	pCtx->ADDR_rSI = uSrc1AddrReg;
331	pCtx->ADDR_rDI = uSrc2AddrReg;
332	pCtx->eflags.u = uEFlags;
333
334	iemMemPageUnmap(pIemCpu, GCPhysSrc1Mem, IEM_ACCESS_DATA_R, puSrc1Mem, &PgLockSrc1Mem);
335	iemMemPageUnmap(pIemCpu, GCPhysSrc2Mem, IEM_ACCESS_DATA_R, puSrc2Mem, &PgLockSrc2Mem);
336	continue;
337	}
338	iemMemPageUnmap(pIemCpu, GCPhysSrc2Mem, IEM_ACCESS_DATA_R, puSrc2Mem, &PgLockSrc2Mem);
339	}
340	}
341
342	/*
343	* Fallback - slow processing till the end of the current page.
344	* In the cross page boundrary case we will end up here with cLeftPage
345	* as 0, we execute one loop then.
346	*/
347	do
348	{
349	OP_TYPE uValue1;
350	rcStrict = RT_CONCAT(iemMemFetchDataU,OP_SIZE)(pIemCpu, &uValue1, iEffSeg, uSrc1AddrReg);
351	if (rcStrict != VINF_SUCCESS)
352	return rcStrict;
353	OP_TYPE uValue2;
354	rcStrict = RT_CONCAT(iemMemFetchDataU,OP_SIZE)(pIemCpu, &uValue2, X86_SREG_ES, uSrc2AddrReg);
355	if (rcStrict != VINF_SUCCESS)
356	return rcStrict;
357	RT_CONCAT(iemAImpl_cmp_u,OP_SIZE)(&uValue1, uValue2, &uEFlags);
358
359	pCtx->ADDR_rSI = uSrc1AddrReg += cbIncr;
360	pCtx->ADDR_rDI = uSrc2AddrReg += cbIncr;
361	pCtx->ADDR_rCX = --uCounterReg;
362	pCtx->eflags.u = uEFlags;
363	cLeftPage--;
364	} while ( (int32_t)cLeftPage > 0
365	&& !(uEFlags & X86_EFL_ZF));
366	} while ( uCounterReg != 0
367	&& !(uEFlags & X86_EFL_ZF));
368
369	/*
370	* Done.
371	*/
372	iemRegAddToRip(pIemCpu, cbInstr);
373	return VINF_SUCCESS;
374	}
375
376
377	/**
378	* Implements 'REPE SCAS'.
379	*/
380	IEM_CIMPL_DEF_0(RT_CONCAT4(iemCImpl_repe_scas_,OP_rAX,_m,ADDR_SIZE))
381	{
382	PCPUMCTX pCtx = pIemCpu->CTX_SUFF(pCtx);
383
384	/*
385	* Setup.
386	*/
387	ADDR_TYPE uCounterReg = pCtx->ADDR_rCX;
388	if (uCounterReg == 0)
389	{
390	iemRegAddToRip(pIemCpu, cbInstr);
391	return VINF_SUCCESS;
392	}
393
394	uint64_t uBaseAddr;
395	VBOXSTRICTRC rcStrict = iemMemSegCheckReadAccessEx(pIemCpu, &pCtx->es, X86_SREG_ES, &uBaseAddr);
396	if (rcStrict != VINF_SUCCESS)
397	return rcStrict;
398
399	int8_t const cbIncr = pCtx->eflags.Bits.u1DF ? -(OP_SIZE / 8) : (OP_SIZE / 8);
400	OP_TYPE const uValueReg = pCtx->OP_rAX;
401	ADDR_TYPE uAddrReg = pCtx->ADDR_rDI;
402	uint32_t uEFlags = pCtx->eflags.u;
403
404	/*
405	* The loop.
406	*/
407	do
408	{
409	/*
410	* Do segmentation and virtual page stuff.
411	*/
412	ADDR2_TYPE uVirtAddr = uAddrReg + (ADDR2_TYPE)uBaseAddr;
413	uint32_t cLeftPage = (PAGE_SIZE - (uVirtAddr & PAGE_OFFSET_MASK)) / (OP_SIZE / 8);
414	if (cLeftPage > uCounterReg)
415	cLeftPage = uCounterReg;
416	if ( cLeftPage > 0 /* can be null if unaligned, do one fallback round. */
417	&& cbIncr > 0 /** @todo Implement reverse direction string ops. */
418	&& ( IS_64_BIT_CODE(pIemCpu)
419	\|\| ( uAddrReg < pCtx->es.u32Limit
420	&& uAddrReg + (cLeftPage * (OP_SIZE / 8)) <= pCtx->es.u32Limit)
421	)
422	)
423	{
424	RTGCPHYS GCPhysMem;
425	rcStrict = iemMemPageTranslateAndCheckAccess(pIemCpu, uVirtAddr, IEM_ACCESS_DATA_R, &GCPhysMem);
426	if (rcStrict != VINF_SUCCESS)
427	return rcStrict;
428
429	/*
430	* If we can map the page without trouble, do a block processing
431	* until the end of the current page.
432	*/
433	PGMPAGEMAPLOCK PgLockMem;
434	OP_TYPE const *puMem;
435	rcStrict = iemMemPageMap(pIemCpu, GCPhysMem, IEM_ACCESS_DATA_R, (void **)&puMem, &PgLockMem);
436	if (rcStrict == VINF_SUCCESS)
437	{
438	/* Search till we find a mismatching item. */
439	OP_TYPE uTmpValue;
440	bool fQuit;
441	uint32_t i = 0;
442	do
443	{
444	uTmpValue = puMem[i++];
445	fQuit = uTmpValue != uValueReg;
446	} while (i < cLeftPage && !fQuit);
447
448	/* Update the regs. */
449	RT_CONCAT(iemAImpl_cmp_u,OP_SIZE)((OP_TYPE *)&uValueReg, uTmpValue, &uEFlags);
450	pCtx->ADDR_rCX = uCounterReg -= i;
451	pCtx->ADDR_rDI = uAddrReg += i * cbIncr;
452	pCtx->eflags.u = uEFlags;
453	Assert(!(uEFlags & X86_EFL_ZF) == fQuit);
454	iemMemPageUnmap(pIemCpu, GCPhysMem, IEM_ACCESS_DATA_R, puMem, &PgLockMem);
455	if (fQuit)
456	break;
457
458
459	/* If unaligned, we drop thru and do the page crossing access
460	below. Otherwise, do the next page. */
461	if (!(uVirtAddr & (OP_SIZE - 1)))
462	continue;
463	if (uCounterReg == 0)
464	break;
465	cLeftPage = 0;
466	}
467	}
468
469	/*
470	* Fallback - slow processing till the end of the current page.
471	* In the cross page boundrary case we will end up here with cLeftPage
472	* as 0, we execute one loop then.
473	*/
474	do
475	{
476	OP_TYPE uTmpValue;
477	rcStrict = RT_CONCAT(iemMemFetchDataU,OP_SIZE)(pIemCpu, &uTmpValue, X86_SREG_ES, uAddrReg);
478	if (rcStrict != VINF_SUCCESS)
479	return rcStrict;
480	RT_CONCAT(iemAImpl_cmp_u,OP_SIZE)((OP_TYPE *)&uValueReg, uTmpValue, &uEFlags);
481
482	pCtx->ADDR_rDI = uAddrReg += cbIncr;
483	pCtx->ADDR_rCX = --uCounterReg;
484	pCtx->eflags.u = uEFlags;
485	cLeftPage--;
486	} while ( (int32_t)cLeftPage > 0
487	&& (uEFlags & X86_EFL_ZF));
488	} while ( uCounterReg != 0
489	&& (uEFlags & X86_EFL_ZF));
490
491	/*
492	* Done.
493	*/
494	iemRegAddToRip(pIemCpu, cbInstr);
495	return VINF_SUCCESS;
496	}
497
498
499	/**
500	* Implements 'REPNE SCAS'.
501	*/
502	IEM_CIMPL_DEF_0(RT_CONCAT4(iemCImpl_repne_scas_,OP_rAX,_m,ADDR_SIZE))
503	{
504	PCPUMCTX pCtx = pIemCpu->CTX_SUFF(pCtx);
505
506	/*
507	* Setup.
508	*/
509	ADDR_TYPE uCounterReg = pCtx->ADDR_rCX;
510	if (uCounterReg == 0)
511	{
512	iemRegAddToRip(pIemCpu, cbInstr);
513	return VINF_SUCCESS;
514	}
515
516	uint64_t uBaseAddr;
517	VBOXSTRICTRC rcStrict = iemMemSegCheckReadAccessEx(pIemCpu, &pCtx->es, X86_SREG_ES, &uBaseAddr);
518	if (rcStrict != VINF_SUCCESS)
519	return rcStrict;
520
521	int8_t const cbIncr = pCtx->eflags.Bits.u1DF ? -(OP_SIZE / 8) : (OP_SIZE / 8);
522	OP_TYPE const uValueReg = pCtx->OP_rAX;
523	ADDR_TYPE uAddrReg = pCtx->ADDR_rDI;
524	uint32_t uEFlags = pCtx->eflags.u;
525
526	/*
527	* The loop.
528	*/
529	do
530	{
531	/*
532	* Do segmentation and virtual page stuff.
533	*/
534	ADDR2_TYPE uVirtAddr = uAddrReg + (ADDR2_TYPE)uBaseAddr;
535	uint32_t cLeftPage = (PAGE_SIZE - (uVirtAddr & PAGE_OFFSET_MASK)) / (OP_SIZE / 8);
536	if (cLeftPage > uCounterReg)
537	cLeftPage = uCounterReg;
538	if ( cLeftPage > 0 /* can be null if unaligned, do one fallback round. */
539	&& cbIncr > 0 /** @todo Implement reverse direction string ops. */
540	&& ( IS_64_BIT_CODE(pIemCpu)
541	\|\| ( uAddrReg < pCtx->es.u32Limit
542	&& uAddrReg + (cLeftPage * (OP_SIZE / 8)) <= pCtx->es.u32Limit)
543	)
544	)
545	{
546	RTGCPHYS GCPhysMem;
547	rcStrict = iemMemPageTranslateAndCheckAccess(pIemCpu, uVirtAddr, IEM_ACCESS_DATA_R, &GCPhysMem);
548	if (rcStrict != VINF_SUCCESS)
549	return rcStrict;
550
551	/*
552	* If we can map the page without trouble, do a block processing
553	* until the end of the current page.
554	*/
555	PGMPAGEMAPLOCK PgLockMem;
556	OP_TYPE const *puMem;
557	rcStrict = iemMemPageMap(pIemCpu, GCPhysMem, IEM_ACCESS_DATA_R, (void **)&puMem, &PgLockMem);
558	if (rcStrict == VINF_SUCCESS)
559	{
560	/* Search till we find a mismatching item. */
561	OP_TYPE uTmpValue;
562	bool fQuit;
563	uint32_t i = 0;
564	do
565	{
566	uTmpValue = puMem[i++];
567	fQuit = uTmpValue == uValueReg;
568	} while (i < cLeftPage && !fQuit);
569
570	/* Update the regs. */
571	RT_CONCAT(iemAImpl_cmp_u,OP_SIZE)((OP_TYPE *)&uValueReg, uTmpValue, &uEFlags);
572	pCtx->ADDR_rCX = uCounterReg -= i;
573	pCtx->ADDR_rDI = uAddrReg += i * cbIncr;
574	pCtx->eflags.u = uEFlags;
575	Assert(!!(uEFlags & X86_EFL_ZF) == fQuit);
576	iemMemPageUnmap(pIemCpu, GCPhysMem, IEM_ACCESS_DATA_R, puMem, &PgLockMem);
577	if (fQuit)
578	break;
579
580
581	/* If unaligned, we drop thru and do the page crossing access
582	below. Otherwise, do the next page. */
583	if (!(uVirtAddr & (OP_SIZE - 1)))
584	continue;
585	if (uCounterReg == 0)
586	break;
587	cLeftPage = 0;
588	}
589	}
590
591	/*
592	* Fallback - slow processing till the end of the current page.
593	* In the cross page boundrary case we will end up here with cLeftPage
594	* as 0, we execute one loop then.
595	*/
596	do
597	{
598	OP_TYPE uTmpValue;
599	rcStrict = RT_CONCAT(iemMemFetchDataU,OP_SIZE)(pIemCpu, &uTmpValue, X86_SREG_ES, uAddrReg);
600	if (rcStrict != VINF_SUCCESS)
601	return rcStrict;
602	RT_CONCAT(iemAImpl_cmp_u,OP_SIZE)((OP_TYPE *)&uValueReg, uTmpValue, &uEFlags);
603	pCtx->ADDR_rDI = uAddrReg += cbIncr;
604	pCtx->ADDR_rCX = --uCounterReg;
605	pCtx->eflags.u = uEFlags;
606	cLeftPage--;
607	} while ( (int32_t)cLeftPage > 0
608	&& !(uEFlags & X86_EFL_ZF));
609	} while ( uCounterReg != 0
610	&& !(uEFlags & X86_EFL_ZF));
611
612	/*
613	* Done.
614	*/
615	iemRegAddToRip(pIemCpu, cbInstr);
616	return VINF_SUCCESS;
617	}
618
619
620
621
622	/**
623	* Implements 'REP MOVS'.
624	*/
625	IEM_CIMPL_DEF_1(RT_CONCAT4(iemCImpl_rep_movs_op,OP_SIZE,_addr,ADDR_SIZE), uint8_t, iEffSeg)
626	{
627	PCPUMCTX pCtx = pIemCpu->CTX_SUFF(pCtx);
628
629	/*
630	* Setup.
631	*/
632	ADDR_TYPE uCounterReg = pCtx->ADDR_rCX;
633	if (uCounterReg == 0)
634	{
635	iemRegAddToRip(pIemCpu, cbInstr);
636	return VINF_SUCCESS;
637	}
638
639	PCCPUMSELREGHID pSrcHid = iemSRegGetHid(pIemCpu, iEffSeg);
640	uint64_t uSrcBase;
641	VBOXSTRICTRC rcStrict = iemMemSegCheckReadAccessEx(pIemCpu, pSrcHid, iEffSeg, &uSrcBase);
642	if (rcStrict != VINF_SUCCESS)
643	return rcStrict;
644
645	uint64_t uDstBase;
646	rcStrict = iemMemSegCheckWriteAccessEx(pIemCpu, &pCtx->es, X86_SREG_ES, &uDstBase);
647	if (rcStrict != VINF_SUCCESS)
648	return rcStrict;
649
650	int8_t const cbIncr = pCtx->eflags.Bits.u1DF ? -(OP_SIZE / 8) : (OP_SIZE / 8);
651	ADDR_TYPE uSrcAddrReg = pCtx->ADDR_rSI;
652	ADDR_TYPE uDstAddrReg = pCtx->ADDR_rDI;
653
654	/*
655	* Be careful with handle bypassing.
656	*/
657	if (pIemCpu->fBypassHandlers)
658	{
659	Log(("%s: declining because we're bypassing handlers\n", __FUNCTION__));
660	return VERR_IEM_ASPECT_NOT_IMPLEMENTED;
661	}
662
663	/*
664	* If we're reading back what we write, we have to let the verfication code
665	* to prevent a false positive.
666	* Note! This doesn't take aliasing or wrapping into account - lazy bird.
667	*/
668	#ifdef IEM_VERIFICATION_MODE_FULL
669	if ( IEM_VERIFICATION_ENABLED(pIemCpu)
670	&& (cbIncr > 0
671	? uSrcAddrReg <= uDstAddrReg
672	&& uSrcAddrReg + cbIncr * uCounterReg > uDstAddrReg
673	: uDstAddrReg <= uSrcAddrReg
674	&& uDstAddrReg + cbIncr * uCounterReg > uSrcAddrReg))
675	pIemCpu->fOverlappingMovs = true;
676	#endif
677
678	/*
679	* The loop.
680	*/
681	do
682	{
683	/*
684	* Do segmentation and virtual page stuff.
685	*/
686	ADDR2_TYPE uVirtSrcAddr = uSrcAddrReg + (ADDR2_TYPE)uSrcBase;
687	ADDR2_TYPE uVirtDstAddr = uDstAddrReg + (ADDR2_TYPE)uDstBase;
688	uint32_t cLeftSrcPage = (PAGE_SIZE - (uVirtSrcAddr & PAGE_OFFSET_MASK)) / (OP_SIZE / 8);
689	if (cLeftSrcPage > uCounterReg)
690	cLeftSrcPage = uCounterReg;
691	uint32_t cLeftDstPage = (PAGE_SIZE - (uVirtDstAddr & PAGE_OFFSET_MASK)) / (OP_SIZE / 8);
692	uint32_t cLeftPage = RT_MIN(cLeftSrcPage, cLeftDstPage);
693
694	if ( cLeftPage > 0 /* can be null if unaligned, do one fallback round. */
695	&& cbIncr > 0 /** @todo Implement reverse direction string ops. */
696	&& ( IS_64_BIT_CODE(pIemCpu)
697	\|\| ( uSrcAddrReg < pSrcHid->u32Limit
698	&& uSrcAddrReg + (cLeftPage * (OP_SIZE / 8)) <= pSrcHid->u32Limit
699	&& uDstAddrReg < pCtx->es.u32Limit
700	&& uDstAddrReg + (cLeftPage * (OP_SIZE / 8)) <= pCtx->es.u32Limit)
701	)
702	)
703	{
704	RTGCPHYS GCPhysSrcMem;
705	rcStrict = iemMemPageTranslateAndCheckAccess(pIemCpu, uVirtSrcAddr, IEM_ACCESS_DATA_R, &GCPhysSrcMem);
706	if (rcStrict != VINF_SUCCESS)
707	return rcStrict;
708
709	RTGCPHYS GCPhysDstMem;
710	rcStrict = iemMemPageTranslateAndCheckAccess(pIemCpu, uVirtDstAddr, IEM_ACCESS_DATA_W, &GCPhysDstMem);
711	if (rcStrict != VINF_SUCCESS)
712	return rcStrict;
713
714	/*
715	* If we can map the page without trouble, do a block processing
716	* until the end of the current page.
717	*/
718	PGMPAGEMAPLOCK PgLockDstMem;
719	OP_TYPE *puDstMem;
720	rcStrict = iemMemPageMap(pIemCpu, GCPhysDstMem, IEM_ACCESS_DATA_W, (void **)&puDstMem, &PgLockDstMem);
721	if (rcStrict == VINF_SUCCESS)
722	{
723	PGMPAGEMAPLOCK PgLockSrcMem;
724	OP_TYPE const *puSrcMem;
725	rcStrict = iemMemPageMap(pIemCpu, GCPhysSrcMem, IEM_ACCESS_DATA_R, (void **)&puSrcMem, &PgLockSrcMem);
726	if (rcStrict == VINF_SUCCESS)
727	{
728	Assert( (GCPhysSrcMem >> PAGE_SHIFT) != (GCPhysDstMem >> PAGE_SHIFT)
729	\|\| ((uintptr_t)puSrcMem >> PAGE_SHIFT) == ((uintptr_t)puDstMem >> PAGE_SHIFT));
730
731	/* Perform the operation exactly (don't use memcpy to avoid
732	having to consider how its implementation would affect
733	any overlapping source and destination area). */
734	OP_TYPE const *puSrcCur = puSrcMem;
735	OP_TYPE *puDstCur = puDstMem;
736	uint32_t cTodo = cLeftPage;
737	while (cTodo-- > 0)
738	puDstCur++ = puSrcCur++;
739
740	/* Update the registers. */
741	pCtx->ADDR_rSI = uSrcAddrReg += cLeftPage * cbIncr;
742	pCtx->ADDR_rDI = uDstAddrReg += cLeftPage * cbIncr;
743	pCtx->ADDR_rCX = uCounterReg -= cLeftPage;
744
745	iemMemPageUnmap(pIemCpu, GCPhysSrcMem, IEM_ACCESS_DATA_R, puSrcMem, &PgLockSrcMem);
746	iemMemPageUnmap(pIemCpu, GCPhysDstMem, IEM_ACCESS_DATA_W, puDstMem, &PgLockDstMem);
747	continue;
748	}
749	iemMemPageUnmap(pIemCpu, GCPhysDstMem, IEM_ACCESS_DATA_W, puDstMem, &PgLockDstMem);
750	}
751	}
752
753	/*
754	* Fallback - slow processing till the end of the current page.
755	* In the cross page boundrary case we will end up here with cLeftPage
756	* as 0, we execute one loop then.
757	*/
758	do
759	{
760	OP_TYPE uValue;
761	rcStrict = RT_CONCAT(iemMemFetchDataU,OP_SIZE)(pIemCpu, &uValue, iEffSeg, uSrcAddrReg);
762	if (rcStrict != VINF_SUCCESS)
763	return rcStrict;
764	rcStrict = RT_CONCAT(iemMemStoreDataU,OP_SIZE)(pIemCpu, X86_SREG_ES, uDstAddrReg, uValue);
765	if (rcStrict != VINF_SUCCESS)
766	return rcStrict;
767
768	pCtx->ADDR_rSI = uSrcAddrReg += cbIncr;
769	pCtx->ADDR_rDI = uDstAddrReg += cbIncr;
770	pCtx->ADDR_rCX = --uCounterReg;
771	cLeftPage--;
772	} while ((int32_t)cLeftPage > 0);
773	} while (uCounterReg != 0);
774
775	/*
776	* Done.
777	*/
778	iemRegAddToRip(pIemCpu, cbInstr);
779	return VINF_SUCCESS;
780	}
781
782
783	/**
784	* Implements 'REP STOS'.
785	*/
786	IEM_CIMPL_DEF_0(RT_CONCAT4(iemCImpl_stos_,OP_rAX,_m,ADDR_SIZE))
787	{
788	PCPUMCTX pCtx = pIemCpu->CTX_SUFF(pCtx);
789
790	/*
791	* Setup.
792	*/
793	ADDR_TYPE uCounterReg = pCtx->ADDR_rCX;
794	if (uCounterReg == 0)
795	{
796	iemRegAddToRip(pIemCpu, cbInstr);
797	return VINF_SUCCESS;
798	}
799
800	uint64_t uBaseAddr;
801	VBOXSTRICTRC rcStrict = iemMemSegCheckWriteAccessEx(pIemCpu, &pCtx->es, X86_SREG_ES, &uBaseAddr);
802	if (rcStrict != VINF_SUCCESS)
803	return rcStrict;
804
805	int8_t const cbIncr = pCtx->eflags.Bits.u1DF ? -(OP_SIZE / 8) : (OP_SIZE / 8);
806	OP_TYPE const uValue = pCtx->OP_rAX;
807	ADDR_TYPE uAddrReg = pCtx->ADDR_rDI;
808
809	/*
810	* Be careful with handle bypassing.
811	*/
812	/** @todo Permit doing a page if correctly aligned. */
813	if (pIemCpu->fBypassHandlers)
814	{
815	Log(("%s: declining because we're bypassing handlers\n", __FUNCTION__));
816	return VERR_IEM_ASPECT_NOT_IMPLEMENTED;
817	}
818
819	/*
820	* The loop.
821	*/
822	do
823	{
824	/*
825	* Do segmentation and virtual page stuff.
826	*/
827	ADDR2_TYPE uVirtAddr = uAddrReg + (ADDR2_TYPE)uBaseAddr;
828	uint32_t cLeftPage = (PAGE_SIZE - (uVirtAddr & PAGE_OFFSET_MASK)) / (OP_SIZE / 8);
829	if (cLeftPage > uCounterReg)
830	cLeftPage = uCounterReg;
831	if ( cLeftPage > 0 /* can be null if unaligned, do one fallback round. */
832	&& cbIncr > 0 /** @todo Implement reverse direction string ops. */
833	&& ( IS_64_BIT_CODE(pIemCpu)
834	\|\| ( uAddrReg < pCtx->es.u32Limit
835	&& uAddrReg + (cLeftPage * (OP_SIZE / 8)) <= pCtx->es.u32Limit)
836	)
837	)
838	{
839	RTGCPHYS GCPhysMem;
840	rcStrict = iemMemPageTranslateAndCheckAccess(pIemCpu, uVirtAddr, IEM_ACCESS_DATA_W, &GCPhysMem);
841	if (rcStrict != VINF_SUCCESS)
842	return rcStrict;
843
844	/*
845	* If we can map the page without trouble, do a block processing
846	* until the end of the current page.
847	*/
848	PGMPAGEMAPLOCK PgLockMem;
849	OP_TYPE *puMem;
850	rcStrict = iemMemPageMap(pIemCpu, GCPhysMem, IEM_ACCESS_DATA_W, (void **)&puMem, &PgLockMem);
851	if (rcStrict == VINF_SUCCESS)
852	{
853	/* Update the regs first so we can loop on cLeftPage. */
854	pCtx->ADDR_rCX = uCounterReg -= cLeftPage;
855	pCtx->ADDR_rDI = uAddrReg += cLeftPage * cbIncr;
856
857	/* Do the memsetting. */
858	#if OP_SIZE == 8
859	memset(puMem, uValue, cLeftPage);
860	/*#elif OP_SIZE == 32
861	ASMMemFill32(puMem, cLeftPage * (OP_SIZE / 8), uValue);*/
862	#else
863	while (cLeftPage-- > 0)
864	*puMem++ = uValue;
865	#endif
866
867	iemMemPageUnmap(pIemCpu, GCPhysMem, IEM_ACCESS_DATA_W, puMem, &PgLockMem);
868
869	/* If unaligned, we drop thru and do the page crossing access
870	below. Otherwise, do the next page. */
871	if (!(uVirtAddr & (OP_SIZE - 1)))
872	continue;
873	if (uCounterReg == 0)
874	break;
875	cLeftPage = 0;
876	}
877	}
878
879	/*
880	* Fallback - slow processing till the end of the current page.
881	* In the cross page boundrary case we will end up here with cLeftPage
882	* as 0, we execute one loop then.
883	*/
884	do
885	{
886	rcStrict = RT_CONCAT(iemMemStoreDataU,OP_SIZE)(pIemCpu, X86_SREG_ES, uAddrReg, uValue);
887	if (rcStrict != VINF_SUCCESS)
888	return rcStrict;
889	pCtx->ADDR_rDI = uAddrReg += cbIncr;
890	pCtx->ADDR_rCX = --uCounterReg;
891	cLeftPage--;
892	} while ((int32_t)cLeftPage > 0);
893	} while (uCounterReg != 0);
894
895	/*
896	* Done.
897	*/
898	iemRegAddToRip(pIemCpu, cbInstr);
899	return VINF_SUCCESS;
900	}
901
902
903	/**
904	* Implements 'REP LODS'.
905	*/
906	IEM_CIMPL_DEF_1(RT_CONCAT4(iemCImpl_lods_,OP_rAX,_m,ADDR_SIZE), int8_t, iEffSeg)
907	{
908	PCPUMCTX pCtx = pIemCpu->CTX_SUFF(pCtx);
909
910	/*
911	* Setup.
912	*/
913	ADDR_TYPE uCounterReg = pCtx->ADDR_rCX;
914	if (uCounterReg == 0)
915	{
916	iemRegAddToRip(pIemCpu, cbInstr);
917	return VINF_SUCCESS;
918	}
919
920	PCCPUMSELREGHID pSrcHid = iemSRegGetHid(pIemCpu, iEffSeg);
921	uint64_t uBaseAddr;
922	VBOXSTRICTRC rcStrict = iemMemSegCheckReadAccessEx(pIemCpu, pSrcHid, iEffSeg, &uBaseAddr);
923	if (rcStrict != VINF_SUCCESS)
924	return rcStrict;
925
926	int8_t const cbIncr = pCtx->eflags.Bits.u1DF ? -(OP_SIZE / 8) : (OP_SIZE / 8);
927	ADDR_TYPE uAddrReg = pCtx->ADDR_rSI;
928
929	/*
930	* The loop.
931	*/
932	do
933	{
934	/*
935	* Do segmentation and virtual page stuff.
936	*/
937	ADDR2_TYPE uVirtAddr = uAddrReg + (ADDR2_TYPE)uBaseAddr;
938	uint32_t cLeftPage = (PAGE_SIZE - (uVirtAddr & PAGE_OFFSET_MASK)) / (OP_SIZE / 8);
939	if (cLeftPage > uCounterReg)
940	cLeftPage = uCounterReg;
941	if ( cLeftPage > 0 /* can be null if unaligned, do one fallback round. */
942	&& cbIncr > 0 /** @todo Implement reverse direction string ops. */
943	&& ( IS_64_BIT_CODE(pIemCpu)
944	\|\| ( uAddrReg < pSrcHid->u32Limit
945	&& uAddrReg + (cLeftPage * (OP_SIZE / 8)) <= pSrcHid->u32Limit)
946	)
947	)
948	{
949	RTGCPHYS GCPhysMem;
950	rcStrict = iemMemPageTranslateAndCheckAccess(pIemCpu, uVirtAddr, IEM_ACCESS_DATA_R, &GCPhysMem);
951	if (rcStrict != VINF_SUCCESS)
952	return rcStrict;
953
954	/*
955	* If we can map the page without trouble, we can get away with
956	* just reading the last value on the page.
957	*/
958	PGMPAGEMAPLOCK PgLockMem;
959	OP_TYPE const *puMem;
960	rcStrict = iemMemPageMap(pIemCpu, GCPhysMem, IEM_ACCESS_DATA_R, (void **)&puMem, &PgLockMem);
961	if (rcStrict == VINF_SUCCESS)
962	{
963	/* Only get the last byte, the rest doesn't matter in direct access mode. */
964	#if OP_SIZE == 32
965	pCtx->rax = puMem[cLeftPage - 1];
966	#else
967	pCtx->OP_rAX = puMem[cLeftPage - 1];
968	#endif
969	pCtx->ADDR_rCX = uCounterReg -= cLeftPage;
970	pCtx->ADDR_rSI = uAddrReg += cLeftPage * cbIncr;
971	iemMemPageUnmap(pIemCpu, GCPhysMem, IEM_ACCESS_DATA_R, puMem, &PgLockMem);
972
973	/* If unaligned, we drop thru and do the page crossing access
974	below. Otherwise, do the next page. */
975	if (!(uVirtAddr & (OP_SIZE - 1)))
976	continue;
977	if (uCounterReg == 0)
978	break;
979	cLeftPage = 0;
980	}
981	}
982
983	/*
984	* Fallback - slow processing till the end of the current page.
985	* In the cross page boundrary case we will end up here with cLeftPage
986	* as 0, we execute one loop then.
987	*/
988	do
989	{
990	OP_TYPE uTmpValue;
991	rcStrict = RT_CONCAT(iemMemFetchDataU,OP_SIZE)(pIemCpu, &uTmpValue, iEffSeg, uAddrReg);
992	if (rcStrict != VINF_SUCCESS)
993	return rcStrict;
994	#if OP_SIZE == 32
995	pCtx->rax = uTmpValue;
996	#else
997	pCtx->OP_rAX = uTmpValue;
998	#endif
999	pCtx->ADDR_rSI = uAddrReg += cbIncr;
1000	pCtx->ADDR_rCX = --uCounterReg;
1001	cLeftPage--;
1002	} while ((int32_t)cLeftPage > 0);
1003	if (rcStrict != VINF_SUCCESS)
1004	break;
1005	} while (uCounterReg != 0);
1006
1007	/*
1008	* Done.
1009	*/
1010	iemRegAddToRip(pIemCpu, cbInstr);
1011	return VINF_SUCCESS;
1012	}
1013
1014
1015	#if OP_SIZE != 64
1016
1017	/**
1018	* Implements 'INS' (no rep)
1019	*/
1020	IEM_CIMPL_DEF_1(RT_CONCAT4(iemCImpl_ins_op,OP_SIZE,_addr,ADDR_SIZE), bool, fIoChecked)
1021	{
1022	PVM pVM = IEMCPU_TO_VM(pIemCpu);
1023	PCPUMCTX pCtx = pIemCpu->CTX_SUFF(pCtx);
1024	VBOXSTRICTRC rcStrict;
1025
1026	/*
1027	* Be careful with handle bypassing.
1028	*/
1029	if (pIemCpu->fBypassHandlers)
1030	{
1031	Log(("%s: declining because we're bypassing handlers\n", __FUNCTION__));
1032	return VERR_IEM_ASPECT_NOT_IMPLEMENTED;
1033	}
1034
1035	/*
1036	* ASSUMES the #GP for I/O permission is taken first, then any #GP for
1037	* segmentation and finally any #PF due to virtual address translation.
1038	* ASSUMES nothing is read from the I/O port before traps are taken.
1039	*/
1040	if (!fIoChecked)
1041	{
1042	rcStrict = iemHlpCheckPortIOPermission(pIemCpu, pCtx, pCtx->dx, OP_SIZE / 8);
1043	if (rcStrict != VINF_SUCCESS)
1044	return rcStrict;
1045	}
1046
1047	OP_TYPE *puMem;
1048	rcStrict = iemMemMap(pIemCpu, (void **)&puMem, OP_SIZE / 8, X86_SREG_ES, pCtx->ADDR_rDI, IEM_ACCESS_DATA_W);
1049	if (rcStrict != VINF_SUCCESS)
1050	return rcStrict;
1051
1052	uint32_t u32Value;
1053	if (!IEM_VERIFICATION_ENABLED(pIemCpu))
1054	rcStrict = IOMIOPortRead(pVM, IEMCPU_TO_VMCPU(pIemCpu), pCtx->dx, &u32Value, OP_SIZE / 8);
1055	else
1056	rcStrict = iemVerifyFakeIOPortRead(pIemCpu, pCtx->dx, &u32Value, OP_SIZE / 8);
1057	if (IOM_SUCCESS(rcStrict))
1058	{
1059	VBOXSTRICTRC rcStrict2 = iemMemCommitAndUnmap(pIemCpu, puMem, IEM_ACCESS_DATA_W);
1060	if (RT_LIKELY(rcStrict2 == VINF_SUCCESS))
1061	{
1062	if (!pCtx->eflags.Bits.u1DF)
1063	pCtx->ADDR_rDI += OP_SIZE / 8;
1064	else
1065	pCtx->ADDR_rDI -= OP_SIZE / 8;
1066	iemRegAddToRip(pIemCpu, cbInstr);
1067	}
1068	/* iemMemMap already check permissions, so this may only be real errors
1069	or access handlers medling. The access handler case is going to
1070	cause misbehavior if the instruction is re-interpreted or smth. So,
1071	we fail with an internal error here instead. */
1072	else
1073	AssertLogRelFailedReturn(VERR_IEM_IPE_1);
1074	}
1075	return rcStrict;
1076	}
1077
1078
1079	/**
1080	* Implements 'REP INS'.
1081	*/
1082	IEM_CIMPL_DEF_1(RT_CONCAT4(iemCImpl_rep_ins_op,OP_SIZE,_addr,ADDR_SIZE), bool, fIoChecked)
1083	{
1084	PVM pVM = IEMCPU_TO_VM(pIemCpu);
1085	PVMCPU pVCpu = IEMCPU_TO_VMCPU(pIemCpu);
1086	PCPUMCTX pCtx = pIemCpu->CTX_SUFF(pCtx);
1087
1088	/*
1089	* Setup.
1090	*/
1091	uint16_t const u16Port = pCtx->dx;
1092	VBOXSTRICTRC rcStrict;
1093	if (!fIoChecked)
1094	{
1095	rcStrict = iemHlpCheckPortIOPermission(pIemCpu, pCtx, u16Port, OP_SIZE / 8);
1096	if (rcStrict != VINF_SUCCESS)
1097	return rcStrict;
1098	}
1099
1100	ADDR_TYPE uCounterReg = pCtx->ADDR_rCX;
1101	if (uCounterReg == 0)
1102	{
1103	iemRegAddToRip(pIemCpu, cbInstr);
1104	return VINF_SUCCESS;
1105	}
1106
1107	uint64_t uBaseAddr;
1108	rcStrict = iemMemSegCheckWriteAccessEx(pIemCpu, &pCtx->es, X86_SREG_ES, &uBaseAddr);
1109	if (rcStrict != VINF_SUCCESS)
1110	return rcStrict;
1111
1112	int8_t const cbIncr = pCtx->eflags.Bits.u1DF ? -(OP_SIZE / 8) : (OP_SIZE / 8);
1113	ADDR_TYPE uAddrReg = pCtx->ADDR_rDI;
1114
1115	/*
1116	* Be careful with handle bypassing.
1117	*/
1118	if (pIemCpu->fBypassHandlers)
1119	{
1120	Log(("%s: declining because we're bypassing handlers\n", __FUNCTION__));
1121	return VERR_IEM_ASPECT_NOT_IMPLEMENTED;
1122	}
1123
1124	/*
1125	* The loop.
1126	*/
1127	do
1128	{
1129	/*
1130	* Do segmentation and virtual page stuff.
1131	*/
1132	ADDR2_TYPE uVirtAddr = uAddrReg + (ADDR2_TYPE)uBaseAddr;
1133	uint32_t cLeftPage = (PAGE_SIZE - (uVirtAddr & PAGE_OFFSET_MASK)) / (OP_SIZE / 8);
1134	if (cLeftPage > uCounterReg)
1135	cLeftPage = uCounterReg;
1136	if ( cLeftPage > 0 /* can be null if unaligned, do one fallback round. */
1137	&& cbIncr > 0 /** @todo Implement reverse direction string ops. */
1138	&& ( IS_64_BIT_CODE(pIemCpu)
1139	\|\| ( uAddrReg < pCtx->es.u32Limit
1140	&& uAddrReg + (cLeftPage * (OP_SIZE / 8)) <= pCtx->es.u32Limit)
1141	)
1142	)
1143	{
1144	RTGCPHYS GCPhysMem;
1145	rcStrict = iemMemPageTranslateAndCheckAccess(pIemCpu, uVirtAddr, IEM_ACCESS_DATA_W, &GCPhysMem);
1146	if (rcStrict != VINF_SUCCESS)
1147	return rcStrict;
1148
1149	/*
1150	* If we can map the page without trouble, we would've liked to use
1151	* an string I/O method to do the work, but the current IOM
1152	* interface doesn't match our current approach. So, do a regular
1153	* loop instead.
1154	*/
1155	/** @todo Change the I/O manager interface to make use of
1156	* mapped buffers instead of leaving those bits to the
1157	* device implementation! */
1158	PGMPAGEMAPLOCK PgLockMem;
1159	OP_TYPE *puMem;
1160	rcStrict = iemMemPageMap(pIemCpu, GCPhysMem, IEM_ACCESS_DATA_W, (void **)&puMem, &PgLockMem);
1161	if (rcStrict == VINF_SUCCESS)
1162	{
1163	uint32_t off = 0;
1164	while (off < cLeftPage)
1165	{
1166	uint32_t u32Value;
1167	if (!IEM_VERIFICATION_ENABLED(pIemCpu))
1168	rcStrict = IOMIOPortRead(pVM, pVCpu, u16Port, &u32Value, OP_SIZE / 8);
1169	else
1170	rcStrict = iemVerifyFakeIOPortRead(pIemCpu, u16Port, &u32Value, OP_SIZE / 8);
1171	if (IOM_SUCCESS(rcStrict))
1172	{
1173	puMem[off] = (OP_TYPE)u32Value;
1174	pCtx->ADDR_rDI = uAddrReg += cbIncr;
1175	pCtx->ADDR_rCX = --uCounterReg;
1176	}
1177	if (rcStrict != VINF_SUCCESS)
1178	{
1179	if (IOM_SUCCESS(rcStrict))
1180	{
1181	rcStrict = iemSetPassUpStatus(pIemCpu, rcStrict);
1182	if (uCounterReg == 0)
1183	iemRegAddToRip(pIemCpu, cbInstr);
1184	}
1185	iemMemPageUnmap(pIemCpu, GCPhysMem, IEM_ACCESS_DATA_W, puMem, &PgLockMem);
1186	return rcStrict;
1187	}
1188	off++;
1189	}
1190	iemMemPageUnmap(pIemCpu, GCPhysMem, IEM_ACCESS_DATA_W, puMem, &PgLockMem);
1191
1192	/* If unaligned, we drop thru and do the page crossing access
1193	below. Otherwise, do the next page. */
1194	if (!(uVirtAddr & (OP_SIZE - 1)))
1195	continue;
1196	if (uCounterReg == 0)
1197	break;
1198	cLeftPage = 0;
1199	}
1200	}
1201
1202	/*
1203	* Fallback - slow processing till the end of the current page.
1204	* In the cross page boundrary case we will end up here with cLeftPage
1205	* as 0, we execute one loop then.
1206	*
1207	* Note! We ASSUME the CPU will raise #PF or #GP before access the
1208	* I/O port, otherwise it wouldn't really be restartable.
1209	*/
1210	/** @todo investigate what the CPU actually does with \#PF/\#GP
1211	* during INS. */
1212	do
1213	{
1214	OP_TYPE *puMem;
1215	rcStrict = iemMemMap(pIemCpu, (void **)&puMem, OP_SIZE / 8, X86_SREG_ES, uAddrReg, IEM_ACCESS_DATA_W);
1216	if (rcStrict != VINF_SUCCESS)
1217	return rcStrict;
1218
1219	uint32_t u32Value;
1220	if (!IEM_VERIFICATION_ENABLED(pIemCpu))
1221	rcStrict = IOMIOPortRead(pVM, pVCpu, u16Port, &u32Value, OP_SIZE / 8);
1222	else
1223	rcStrict = iemVerifyFakeIOPortRead(pIemCpu, u16Port, &u32Value, OP_SIZE / 8);
1224	if (!IOM_SUCCESS(rcStrict))
1225	return rcStrict;
1226
1227	*puMem = (OP_TYPE)u32Value;
1228	VBOXSTRICTRC rcStrict2 = iemMemCommitAndUnmap(pIemCpu, puMem, IEM_ACCESS_DATA_W);
1229	AssertLogRelReturn(rcStrict2 == VINF_SUCCESS, VERR_IEM_IPE_1); /* See non-rep version. */
1230
1231	pCtx->ADDR_rDI = uAddrReg += cbIncr;
1232	pCtx->ADDR_rCX = --uCounterReg;
1233
1234	cLeftPage--;
1235	if (rcStrict != VINF_SUCCESS)
1236	{
1237	if (uCounterReg == 0)
1238	iemRegAddToRip(pIemCpu, cbInstr);
1239	rcStrict = iemSetPassUpStatus(pIemCpu, rcStrict);
1240	return rcStrict;
1241	}
1242	} while ((int32_t)cLeftPage > 0);
1243	} while (uCounterReg != 0);
1244
1245	/*
1246	* Done.
1247	*/
1248	iemRegAddToRip(pIemCpu, cbInstr);
1249	return VINF_SUCCESS;
1250	}
1251
1252
1253	/**
1254	* Implements 'OUTS' (no rep)
1255	*/
1256	IEM_CIMPL_DEF_2(RT_CONCAT4(iemCImpl_outs_op,OP_SIZE,_addr,ADDR_SIZE), uint8_t, iEffSeg, bool, fIoChecked)
1257	{
1258	PVM pVM = IEMCPU_TO_VM(pIemCpu);
1259	PCPUMCTX pCtx = pIemCpu->CTX_SUFF(pCtx);
1260	VBOXSTRICTRC rcStrict;
1261
1262	/*
1263	* ASSUMES the #GP for I/O permission is taken first, then any #GP for
1264	* segmentation and finally any #PF due to virtual address translation.
1265	* ASSUMES nothing is read from the I/O port before traps are taken.
1266	*/
1267	if (!fIoChecked)
1268	{
1269	rcStrict = iemHlpCheckPortIOPermission(pIemCpu, pCtx, pCtx->dx, OP_SIZE / 8);
1270	if (rcStrict != VINF_SUCCESS)
1271	return rcStrict;
1272	}
1273
1274	OP_TYPE uValue;
1275	rcStrict = RT_CONCAT(iemMemFetchDataU,OP_SIZE)(pIemCpu, &uValue, iEffSeg, pCtx->ADDR_rSI);
1276	if (rcStrict == VINF_SUCCESS)
1277	{
1278	if (!IEM_VERIFICATION_ENABLED(pIemCpu))
1279	rcStrict = IOMIOPortWrite(pVM, IEMCPU_TO_VMCPU(pIemCpu), pCtx->dx, uValue, OP_SIZE / 8);
1280	else
1281	rcStrict = iemVerifyFakeIOPortWrite(pIemCpu, pCtx->dx, uValue, OP_SIZE / 8);
1282	if (IOM_SUCCESS(rcStrict))
1283	{
1284	if (!pCtx->eflags.Bits.u1DF)
1285	pCtx->ADDR_rSI += OP_SIZE / 8;
1286	else
1287	pCtx->ADDR_rSI -= OP_SIZE / 8;
1288	iemRegAddToRip(pIemCpu, cbInstr);
1289	if (rcStrict != VINF_SUCCESS)
1290	rcStrict = iemSetPassUpStatus(pIemCpu, rcStrict);
1291	}
1292	}
1293	return rcStrict;
1294	}
1295
1296
1297	/**
1298	* Implements 'REP OUTS'.
1299	*/
1300	IEM_CIMPL_DEF_2(RT_CONCAT4(iemCImpl_rep_outs_op,OP_SIZE,_addr,ADDR_SIZE), uint8_t, iEffSeg, bool, fIoChecked)
1301	{
1302	PVM pVM = IEMCPU_TO_VM(pIemCpu);
1303	PVMCPU pVCpu = IEMCPU_TO_VMCPU(pIemCpu);
1304	PCPUMCTX pCtx = pIemCpu->CTX_SUFF(pCtx);
1305
1306	/*
1307	* Setup.
1308	*/
1309	uint16_t const u16Port = pCtx->dx;
1310	VBOXSTRICTRC rcStrict;
1311	if (!fIoChecked)
1312	{
1313	rcStrict = iemHlpCheckPortIOPermission(pIemCpu, pCtx, u16Port, OP_SIZE / 8);
1314	if (rcStrict != VINF_SUCCESS)
1315	return rcStrict;
1316	}
1317
1318	ADDR_TYPE uCounterReg = pCtx->ADDR_rCX;
1319	if (uCounterReg == 0)
1320	{
1321	iemRegAddToRip(pIemCpu, cbInstr);
1322	return VINF_SUCCESS;
1323	}
1324
1325	PCCPUMSELREGHID pHid = iemSRegGetHid(pIemCpu, iEffSeg);
1326	uint64_t uBaseAddr;
1327	rcStrict = iemMemSegCheckReadAccessEx(pIemCpu, pHid, iEffSeg, &uBaseAddr);
1328	if (rcStrict != VINF_SUCCESS)
1329	return rcStrict;
1330
1331	int8_t const cbIncr = pCtx->eflags.Bits.u1DF ? -(OP_SIZE / 8) : (OP_SIZE / 8);
1332	ADDR_TYPE uAddrReg = pCtx->ADDR_rSI;
1333
1334	/*
1335	* The loop.
1336	*/
1337	do
1338	{
1339	/*
1340	* Do segmentation and virtual page stuff.
1341	*/
1342	ADDR2_TYPE uVirtAddr = uAddrReg + (ADDR2_TYPE)uBaseAddr;
1343	uint32_t cLeftPage = (PAGE_SIZE - (uVirtAddr & PAGE_OFFSET_MASK)) / (OP_SIZE / 8);
1344	if (cLeftPage > uCounterReg)
1345	cLeftPage = uCounterReg;
1346	if ( cLeftPage > 0 /* can be null if unaligned, do one fallback round. */
1347	&& cbIncr > 0 /** @todo Implement reverse direction string ops. */
1348	&& ( IS_64_BIT_CODE(pIemCpu)
1349	\|\| ( uAddrReg < pHid->u32Limit
1350	&& uAddrReg + (cLeftPage * (OP_SIZE / 8)) <= pHid->u32Limit)
1351	)
1352	)
1353	{
1354	RTGCPHYS GCPhysMem;
1355	rcStrict = iemMemPageTranslateAndCheckAccess(pIemCpu, uVirtAddr, IEM_ACCESS_DATA_R, &GCPhysMem);
1356	if (rcStrict != VINF_SUCCESS)
1357	return rcStrict;
1358
1359	/*
1360	* If we can map the page without trouble, we would've liked to use
1361	* an string I/O method to do the work, but the current IOM
1362	* interface doesn't match our current approach. So, do a regular
1363	* loop instead.
1364	*/
1365	/** @todo Change the I/O manager interface to make use of
1366	* mapped buffers instead of leaving those bits to the
1367	* device implementation? */
1368	PGMPAGEMAPLOCK PgLockMem;
1369	OP_TYPE const *puMem;
1370	rcStrict = iemMemPageMap(pIemCpu, GCPhysMem, IEM_ACCESS_DATA_R, (void **)&puMem, &PgLockMem);
1371	if (rcStrict == VINF_SUCCESS)
1372	{
1373	uint32_t off = 0;
1374	while (off < cLeftPage)
1375	{
1376	uint32_t u32Value = *puMem++;
1377	if (!IEM_VERIFICATION_ENABLED(pIemCpu))
1378	rcStrict = IOMIOPortWrite(pVM, pVCpu, u16Port, u32Value, OP_SIZE / 8);
1379	else
1380	rcStrict = iemVerifyFakeIOPortWrite(pIemCpu, u16Port, u32Value, OP_SIZE / 8);
1381	if (IOM_SUCCESS(rcStrict))
1382	{
1383	pCtx->ADDR_rSI = uAddrReg += cbIncr;
1384	pCtx->ADDR_rCX = --uCounterReg;
1385	}
1386	if (rcStrict != VINF_SUCCESS)
1387	{
1388	if (IOM_SUCCESS(rcStrict))
1389	{
1390	rcStrict = iemSetPassUpStatus(pIemCpu, rcStrict);
1391	if (uCounterReg == 0)
1392	iemRegAddToRip(pIemCpu, cbInstr);
1393	}
1394	iemMemPageUnmap(pIemCpu, GCPhysMem, IEM_ACCESS_DATA_R, puMem, &PgLockMem);
1395	return rcStrict;
1396	}
1397	off++;
1398	}
1399	iemMemPageUnmap(pIemCpu, GCPhysMem, IEM_ACCESS_DATA_R, puMem, &PgLockMem);
1400
1401	/* If unaligned, we drop thru and do the page crossing access
1402	below. Otherwise, do the next page. */
1403	if (!(uVirtAddr & (OP_SIZE - 1)))
1404	continue;
1405	if (uCounterReg == 0)
1406	break;
1407	cLeftPage = 0;
1408	}
1409	}
1410
1411	/*
1412	* Fallback - slow processing till the end of the current page.
1413	* In the cross page boundrary case we will end up here with cLeftPage
1414	* as 0, we execute one loop then.
1415	*
1416	* Note! We ASSUME the CPU will raise #PF or #GP before access the
1417	* I/O port, otherwise it wouldn't really be restartable.
1418	*/
1419	/** @todo investigate what the CPU actually does with \#PF/\#GP
1420	* during INS. */
1421	do
1422	{
1423	OP_TYPE uValue;
1424	rcStrict = RT_CONCAT(iemMemFetchDataU,OP_SIZE)(pIemCpu, &uValue, iEffSeg, uAddrReg);
1425	if (rcStrict != VINF_SUCCESS)
1426	return rcStrict;
1427
1428	if (!IEM_VERIFICATION_ENABLED(pIemCpu))
1429	rcStrict = IOMIOPortWrite(pVM, pVCpu, u16Port, uValue, OP_SIZE / 8);
1430	else
1431	rcStrict = iemVerifyFakeIOPortWrite(pIemCpu, u16Port, uValue, OP_SIZE / 8);
1432	if (IOM_SUCCESS(rcStrict))
1433	{
1434	pCtx->ADDR_rSI = uAddrReg += cbIncr;
1435	pCtx->ADDR_rCX = --uCounterReg;
1436	cLeftPage--;
1437	}
1438	if (rcStrict != VINF_SUCCESS)
1439	{
1440	if (IOM_SUCCESS(rcStrict))
1441	{
1442	if (uCounterReg == 0)
1443	iemRegAddToRip(pIemCpu, cbInstr);
1444	rcStrict = iemSetPassUpStatus(pIemCpu, rcStrict);
1445	}
1446	return rcStrict;
1447	}
1448	} while ((int32_t)cLeftPage > 0);
1449	} while (uCounterReg != 0);
1450
1451	/*
1452	* Done.
1453	*/
1454	iemRegAddToRip(pIemCpu, cbInstr);
1455	return VINF_SUCCESS;
1456	}
1457
1458	#endif /* OP_SIZE != 64-bit */
1459
1460
1461	#undef OP_rAX
1462	#undef OP_SIZE
1463	#undef ADDR_SIZE
1464	#undef ADDR_rDI
1465	#undef ADDR_rSI
1466	#undef ADDR_rCX
1467	#undef ADDR_rIP
1468	#undef ADDR2_TYPE
1469	#undef ADDR_TYPE
1470	#undef ADDR2_TYPE
1471	#undef IS_64_BIT_CODE

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

source: vbox/trunk/src/VBox/VMM/VMMAll/IEMAllCImplStrInstr.cpp.h@ 47681

Download in other formats: