bootsector2-cpu-instr-1-template.mac@ 97271

Last change on this file since 97271 was 97116, checked in by vboxsync, 2 years ago
ValKit: Also test what CMPXCHG8B does to RAX/RDX.
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 16.8 KB

Line
1	; $Id: bootsector2-cpu-instr-1-template.mac 97116 2022-10-12 15:07:25Z vboxsync $
2	;; @file
3	; Bootsector test for misc instruction - multi mode template.
4	;
5
6	;
7	; Copyright (C) 2007-2022 Oracle and/or its affiliates.
8	;
9	; This file is part of VirtualBox base platform packages, as
10	; available from https://www.virtualbox.org.
11	;
12	; This program is free software; you can redistribute it and/or
13	; modify it under the terms of the GNU General Public License
14	; as published by the Free Software Foundation, in version 3 of the
15	; License.
16	;
17	; This program is distributed in the hope that it will be useful, but
18	; WITHOUT ANY WARRANTY; without even the implied warranty of
19	; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20	; General Public License for more details.
21	;
22	; You should have received a copy of the GNU General Public License
23	; along with this program; if not, see <https://www.gnu.org/licenses>.
24	;
25	; The contents of this file may alternatively be used under the terms
26	; of the Common Development and Distribution License Version 1.0
27	; (CDDL), a copy of it is provided in the "COPYING.CDDL" file included
28	; in the VirtualBox distribution, in which case the provisions of the
29	; CDDL are applicable instead of those of the GPL.
30	;
31	; You may elect to license modified versions of this file under the
32	; terms and conditions of either the GPL or the CDDL or both.
33	;
34	; SPDX-License-Identifier: GPL-3.0-only OR CDDL-1.0
35	;
36
37
38	%include "bootsector2-template-header.mac"
39
40
41
42	;;
43	; Memory fence instructions (SSE2).
44	;
45	; @uses No registers, but BS2_SEL_SPARE0 is trashed.
46	;
47	BEGINPROC TMPL_NM(TestMemFences)
48	push xBP
49	mov xBP, xSP
50	push sAX
51	push xBX
52	push xCX
53	push xDX
54	push xDI
55	push xSI
56	sub xSP, 80h ; iret stack frame space.
57	mov xSI, xSP ; Save the stack register.
58
59	mov xAX, .s_szSubTestName
60	call TMPL_NM_CMN(TestSub)
61
62	;
63	; SSE2 supported?
64	;
65	mov eax, 1
66	xor ecx, ecx
67	cpuid
68	test edx, X86_CPUID_FEATURE_EDX_SSE2
69	jz .skip
70
71	;
72	; Check that the standard instruction encodings work.
73	;
74	mov xBX, [xSP + 10h]
75	mov [xSP], xAX
76	mfence
77	mov [xSP], xCX
78	mov xBX, [xSP + 08h]
79	sfence
80	mov [xSP], xDX
81	mov xBX, [xSP]
82	lfence
83	mov bx, [xSP + 04h]
84
85
86	;
87	; The instruction encodings in the intel manual may open the RM as well
88	; as prefixes open to interpretation. AMD sets RM=0 in their docs.
89	;
90	; lfence = 0f,ea,e8
91	; mfence = 0f,ea,f0
92	; sfence = 0f,ea,f8
93	; (RM is the lower 3 bits of the last byte.)
94
95	%assign MY_RM 0xe8
96	%rep 18h
97	db 0fh, 0aeh, MY_RM
98	db X86_OP_PRF_CS, 0fh, 0aeh, MY_RM
99	db X86_OP_PRF_DS, 0fh, 0aeh, MY_RM
100	db X86_OP_PRF_ES, 0fh, 0aeh, MY_RM
101	db X86_OP_PRF_FS, 0fh, 0aeh, MY_RM
102	db X86_OP_PRF_GS, 0fh, 0aeh, MY_RM
103	db X86_OP_PRF_SS, 0fh, 0aeh, MY_RM
104	db X86_OP_PRF_SIZE_ADDR, 0fh, 0aeh, MY_RM
105	BS2_TRAP_INSTR X86_XCPT_UD, 0, db X86_OP_PRF_SIZE_OP, 0fh, 0aeh, MY_RM ; (used in group)
106	BS2_TRAP_INSTR X86_XCPT_UD, 0, db X86_OP_PRF_LOCK, 0fh, 0aeh, MY_RM ; (used in group)
107	BS2_TRAP_INSTR X86_XCPT_UD, 0, db X86_OP_PRF_REPNZ, 0fh, 0aeh, MY_RM ; (used in group)
108	BS2_TRAP_INSTR X86_XCPT_UD, 0, db X86_OP_PRF_REPZ, 0fh, 0aeh, MY_RM ; (used in group)
109	%ifdef TMPL_64BIT
110	%assign MY_REX 0x40
111	%rep 10h
112	; Rex prefixes doesn't change anything.
113	db MY_REX, 0fh, 0aeh, MY_RM
114	db X86_OP_PRF_CS, MY_REX, 0fh, 0aeh, MY_RM
115	db X86_OP_PRF_DS, MY_REX, 0fh, 0aeh, MY_RM
116	db X86_OP_PRF_ES, MY_REX, 0fh, 0aeh, MY_RM
117	db X86_OP_PRF_FS, MY_REX, 0fh, 0aeh, MY_RM
118	db X86_OP_PRF_GS, MY_REX, 0fh, 0aeh, MY_RM
119	db X86_OP_PRF_SS, MY_REX, 0fh, 0aeh, MY_RM
120	db X86_OP_PRF_SIZE_ADDR, MY_REX, 0fh, 0aeh, MY_RM
121	BS2_TRAP_INSTR X86_XCPT_UD, 0, db X86_OP_PRF_SIZE_OP, MY_REX, 0fh, 0aeh, MY_RM ; (used in group)
122	BS2_TRAP_INSTR X86_XCPT_UD, 0, db X86_OP_PRF_LOCK, MY_REX, 0fh, 0aeh, MY_RM ; (used in group)
123	BS2_TRAP_INSTR X86_XCPT_UD, 0, db X86_OP_PRF_REPNZ, MY_REX, 0fh, 0aeh, MY_RM ; (used in group)
124	BS2_TRAP_INSTR X86_XCPT_UD, 0, db X86_OP_PRF_REPZ, MY_REX, 0fh, 0aeh, MY_RM ; (used in group)
125	%assign MY_REX (MY_REX + 1)
126	%endrep
127	%endif
128	%assign MY_RM (MY_RM + 1)
129	%endrep
130
131	;
132	; Done.
133	;
134	call TMPL_NM_CMN(TestSubDone)
135	.done:
136	mov xSP, xSI
137	add xSP, 80h
138	pop xSI
139	pop xDI
140	pop xDX
141	pop xCX
142	pop xBX
143	pop sAX
144	leave
145	ret
146
147	.skip:
148	mov xAX, .s_szSse2Missing
149	call TMPL_NM_CMN(TestSubDone)
150	jmp .done
151
152	.s_szSubTestName:
153	db TMPL_MODE_STR, ', mfence et al.', 0
154	.s_szSse2Missing:
155	db 'SSE2 is missing', 0
156	ENDPROC TMPL_NM(TestMemFences)
157
158
159	;;
160	; Floating-point to integer conversion (SSE/SSE2).
161	; Neither Intel nor AMD explicitly document what happens for the 32-bit forms
162	; of CVTxx2SI in 64-bit mode with regard to the high dword of a 64-bit
163	; destination register.
164	;
165	; @uses XMM0, and BS2_SEL_SPARE0 is trashed.
166	;
167	BEGINPROC TMPL_NM(TestCvtSize)
168	push xBP
169	mov xBP, xSP
170	push sAX
171	push xBX
172	push xCX
173	push xDX
174	push xDI
175	push xSI
176	sub xSP, 80h ; iret stack frame space.
177	mov xSI, xSP ; Save the stack register.
178
179	mov xAX, .s_szSubTestName
180	call TMPL_NM_CMN(TestSub)
181
182	;
183	; SSE2 supported?
184	;
185	mov eax, 1
186	xor ecx, ecx
187	cpuid
188	test edx, X86_CPUID_FEATURE_EDX_SSE2
189	jz .skip
190
191	%ifdef TMPL_64BIT
192
193	;
194	; Have to enable OSFXSR for SSE instructions to work.
195	;
196	mov rcx,cr4
197	mov rsi,rcx
198	or rcx,200h
199	mov cr4,rcx
200
201	;
202	; Load 32-bit float -2.75 into XMM0
203	;
204	mov eax, 0C0300000h
205	movd xmm0, eax
206	mov rbx, -1 ; make sure high dword is not zero
207	cvtss2si ebx, xmm0 ; result is -3
208	mov eax, -3 ; high dword of rax zeroed
209	TEST_ASSERT_SIMPLE rbx, rax, jz, "CVTSS2SI EBX"
210
211	mov eax, 0C0300000h
212	movd xmm0, eax
213	mov rbx, -1
214	cvttss2si ebx, xmm0 ; result is -2
215	mov eax, -2
216	TEST_ASSERT_SIMPLE rbx, rax, jz, "CVTTSS2SI EBX"
217
218	;
219	; Load 64-bit double -2.75 into XMM0
220	;
221	mov rax, 0C006000000000000h
222	movd xmm0, rax
223	mov rbx, -1
224	cvtsd2si ebx, xmm0
225	mov eax, -3
226	TEST_ASSERT_SIMPLE rbx, rax, jz, "CVTSD2SI EBX"
227
228	mov rax, 0C006000000000000h
229	mov rbx, -1
230	movd xmm0, rax
231	cvttsd2si ebx, xmm0
232	mov eax,-2
233	TEST_ASSERT_SIMPLE rbx, rax, jz, "CVTTSD2SI EBX"
234
235	;
236	; Restore prior CR4 value
237	;
238	mov cr4,rsi
239	%endif
240
241	;
242	; Done.
243	;
244	call TMPL_NM_CMN(TestSubDone)
245	.done:
246	mov xSP, xSI
247	add xSP, 80h
248	pop xSI
249	pop xDI
250	pop xDX
251	pop xCX
252	pop xBX
253	pop sAX
254	leave
255	ret
256
257	.skip:
258	mov xAX, .s_szSse2Missing
259	call TMPL_NM_CMN(TestSubDone)
260	jmp .done
261
262	.s_szSubTestName:
263	db TMPL_MODE_STR, ', cvtss2si et al.', 0
264	.s_szSse2Missing:
265	db 'SSE2 is missing', 0
266	ENDPROC TMPL_NM(TestCvtSize)
267
268
269	;;
270	; Test what CMPXCHG with 32-bit operand size does to 64-bit registers,
271	; as this is not particularly well documented by either Intel or AMD.
272	;
273	; @uses No registers, but BS2_SEL_SPARE0 is trashed.
274	;
275	BEGINPROC TMPL_NM(TestCmpxchg32)
276	push xBP
277	mov xBP, xSP
278	push sAX
279	push xBX
280	push xCX
281	push xDX
282	push xDI
283	push xSI
284	sub xSP, 80h ; iret stack frame space.
285	mov xSI, xSP ; Save the stack register.
286
287	mov xAX, .s_szSubTestName
288	call TMPL_NM_CMN(TestSub)
289
290	%ifdef TMPL_64BIT
291
292	;
293	; CMPXCHG reg, reg - values not equal, eax written
294	;
295	mov rax, -1 ; Load registers with 64-bit values
296	mov rbx, -2
297	mov rcx, -3
298	cmpxchg ebx, ecx ; Not equal, writes ebx to eax
299	mov edx, -2 ; Clears high dword
300	TEST_ASSERT_SIMPLE rax, rdx, jz, "CMPXCHG reg, unequal, rax set"
301	mov rdx, -2 ; All ones still in high dword
302	TEST_ASSERT_SIMPLE rbx, rdx, jz, "CMPXCHG reg, unequal, rbx not set"
303
304	;
305	; CMPXCHG reg, reg - values equal, first operand written
306	;
307	mov rax, -4 ; Load registers with 64-bit values
308	mov rbx, -4
309	mov rcx, -5
310	cmpxchg ebx, ecx ; Equal, writes ecx to ebx
311	mov edx, -5 ; Clears high dword
312	TEST_ASSERT_SIMPLE rbx, rdx, jz, "CMPXCHG reg, equal, rbx set"
313	mov rdx, -4 ; All ones still in high dword
314	TEST_ASSERT_SIMPLE rax, rdx, jz, "CMPXCHG reg, equal, rax not set"
315
316	;
317	; CMPXCHG mem, reg - values not equal, eax written
318	;
319	mov rax, -1 ; Load registers with 64-bit values
320	mov rbx, -2
321	push rbx
322	mov rcx, -3
323	cmpxchg [rsp], ecx ; Not equal, writes eax
324	mov edx, -2 ; Clears high dword
325	TEST_ASSERT_SIMPLE rax, rdx, jz, "CMPXCHG mem, unequal, rax set"
326	pop rbx
327
328	;
329	; CMPXCHG mem, reg - values equal, first operand written
330	;
331	mov rax, -4 ; Load registers with 64-bit values
332	mov rbx, -4
333	push rbx
334	mov rcx, -5
335	cmpxchg [rsp], ecx ; Equal, writes ecx to memory
336	mov rdx, -4 ; All ones in high dword
337	TEST_ASSERT_SIMPLE rax, rdx, jz, "CMPXCHG mem, equal, rax not set"
338	pop rbx
339
340	;
341	; CMPXCHG8B mem, reg - values equal, memory written
342	; compares edx:eax with mem64
343	;
344	mov rdx, -1 ; Load registers with 64-bit values
345	mov rax, -4
346	mov rcx, -1
347	mov rbx, -5
348	mov rsi, -4
349	push rsi
350	cmpxchg8b [rsp] ; Equal, writes ecx:ebx to memory
351	mov rsi, -4 ; All ones in high dword
352	TEST_ASSERT_SIMPLE rax, rsi, jz, "CMPXCHG8B mem, equal, rax not set"
353	mov rsi, -1 ; All ones in high dword
354	TEST_ASSERT_SIMPLE rdx, rsi, jz, "CMPXCHG8B mem, equal, rdx not set"
355	pop rsi
356
357	;
358	; CMPXCHG8B mem, reg - values unequal, edx:eax written
359	; compares edx:eax with mem64
360	;
361	mov rdx, -1 ; Load registers with 64-bit values
362	mov rax, -2
363	mov rcx, -1
364	mov rbx, -4
365	mov rsi, -3
366	push rsi
367	cmpxchg8b [rsp] ; Not equal, writes memory to edx:eax
368	mov esi, -3 ; Clears high dword
369	TEST_ASSERT_SIMPLE rax, rsi, jz, "CMPXCHG8B mem, unequal, rax set"
370	mov esi, -1 ; Clears high dword
371	TEST_ASSERT_SIMPLE rdx, rsi, jz, "CMPXCHG8B mem, unequal, rdx set"
372	pop rsi
373
374	%endif
375
376	;
377	; Done.
378	;
379	call TMPL_NM_CMN(TestSubDone)
380	.done:
381	mov xSP, xSI
382	add xSP, 80h
383	pop xSI
384	pop xDI
385	pop xDX
386	pop xCX
387	pop xBX
388	pop sAX
389	leave
390	ret
391
392	.s_szSubTestName:
393	db TMPL_MODE_STR, ', 32-bit CMPXCHG in 64-bit mode', 0
394	ENDPROC TMPL_NM(TestCmpxchg32)
395
396
397	;;
398	; Proving intel manual wrong about using REX.X for BSWAP R8-R15 on 64-bit.
399	; Checking the 'undefined' 16-bit bswap behavior.
400	;
401	; @uses No registers, but BS2_SEL_SPARE0 is trashed.
402	;
403	BEGINPROC TMPL_NM(TestBSwap)
404	push xBP
405	mov xBP, xSP
406	push sAX
407	push xBX
408	push xCX
409	push xDX
410	push xDI
411	push xSI
412	sub xSP, 80h ; iret stack frame space.
413	mov xSI, xSP ; Save the stack register.
414
415	mov xAX, .s_szSubTestName
416	call TMPL_NM_CMN(TestSub)
417
418	;
419	; Assert sanity.
420	;
421	mov eax, 11223344h
422	bswap eax
423	TEST_ASSERT_SIMPLE eax, 44332211h, jz, "32-bit BSWAP EAX"
424
425	;
426	; Buggy manual (325383-041US, December 2011).
427	;
428	%ifdef TMPL_64BIT
429	push r8
430
431	mov r8d, 55667788h
432	mov eax, 55667788h
433	db X86_OP_REX_X
434	bswap eax ; does it access r8 or eax?
435	TEST_ASSERT_SIMPLE eax, 88776655h, jz, "REX.X BSWAP EAX - Wrong EAX."
436	TEST_ASSERT_SIMPLE r8, 55667788h, jz, "REX.X BSWAP EAX - Wrong R8."
437
438	mov r8d, 55667788h
439	mov eax, 55667788h
440	db X86_OP_REX_R
441	bswap eax ; does it access r8 or eax?
442	TEST_ASSERT_SIMPLE eax, 88776655h, jz, "REX.R BSWAP EAX - Wrong EAX."
443	TEST_ASSERT_SIMPLE r8, 55667788h, jz, "REX.R BSWAP EAX - Wrong R8."
444
445	mov r8d, 55667788h
446	mov eax, 55667788h
447	db X86_OP_REX_B
448	bswap eax ; does it access r8 or eax?
449	TEST_ASSERT_SIMPLE rax, 55667788h, jz, "REX.B BSWAP R8D - Wrong RAX."
450	TEST_ASSERT_SIMPLE r8d, 88776655h, jz, "REX.B BSWAP R8D - Wrong R8D."
451
452	pop r8
453	%endif
454
455	;
456	; 'Undefined' 16-bit behavior.
457	;
458	; Zeroing of the lower 16-bits has been observed on:
459	; - Intel(R) Core(TM) i7-3960X CPU @ 3.30GHz
460	;
461	%ifndef TestBSwap16_defined
462	%define TestBSwap16_defined
463	%macro TestBSwap16 3,
464	mov %3, %2 ; save the primary register.
465	%ifdef TMPL_64BIT
466	mov %2, 0ffffffff98765432h ; Set the upper bit as well.
467	%else
468	mov %2, 98765432h
469	%endif
470	%ifndef TMPL_16BIT
471	db X86_OP_PRF_SIZE_OP
472	%endif
473	bswap %1
474	xchg %2, %3 ; Restore and save the result (xSP).
475	TEST_ASSERT_SIMPLE %3, 98760000h, jz, "Unexpected 16-bit BSWAP error."
476	%endmacro
477	%endif
478
479	TestBSwap16 eax, sAX, sSI
480	TestBSwap16 ebx, sBX, sSI
481	TestBSwap16 ecx, sCX, sSI
482	TestBSwap16 edx, sDX, sSI
483	TestBSwap16 esp, sSP, sSI
484	TestBSwap16 ebp, sBP, sSI
485	TestBSwap16 edi, sDI, sSI
486	TestBSwap16 esi, sSI, sDI
487	%ifdef TMPL_64BIT
488	TestBSwap16 r8d, r8, rax
489	TestBSwap16 r9d, r9, rax
490	TestBSwap16 r10d, r10, rax
491	TestBSwap16 r11d, r11, rax
492	TestBSwap16 r12d, r12, rax
493	TestBSwap16 r13d, r13, rax
494	TestBSwap16 r14d, r14, rax
495	TestBSwap16 r15d, r15, rax
496	%endif
497
498	;
499	; Done.
500	;
501	call TMPL_NM_CMN(TestSubDone)
502	.done:
503	mov xSP, xSI
504	add xSP, 80h
505	pop xSI
506	pop xDI
507	pop xDX
508	pop xCX
509	pop xBX
510	pop sAX
511	leave
512	ret
513
514	.s_szSubTestName:
515	db TMPL_MODE_STR, ', bswap', 0
516	ENDPROC TMPL_NM(TestBSwap)
517
518
519	;;
520	; Do the tests for this mode.
521	;
522	; @uses nothing
523	;
524	BEGINCODELOW
525	BITS 16
526	BEGINPROC TMPL_NM(DoTestsForMode_rm)
527	push bp
528	mov bp, sp
529	push ax
530
531	;
532	; Check if the mode and NX is supported, do the switch.
533	;
534	call TMPL_NM(Bs2IsModeSupported_rm)
535	jz .done
536	call TMPL_NM(Bs2EnterMode_rm)
537	BITS TMPL_BITS
538
539	;
540	; Test exception handler basics using INT3 and #BP.
541	;
542
543	call TMPL_NM(TestMemFences)
544	call TMPL_NM(TestBSwap)
545	%ifdef TMPL_64BIT
546	; Specifically tests 64-bit behavior.
547	call TMPL_NM(TestCvtSize)
548	call TMPL_NM(TestCmpxchg32)
549	%endif
550
551	;
552	; Back to real mode.
553	;
554	call TMPL_NM(Bs2ExitMode)
555	BITS 16
556	call Bs2DisableNX_r86
557
558	.done:
559	pop ax
560	leave
561	ret
562	ENDPROC TMPL_NM(DoTestsForMode_rm)
563	TMPL_BEGINCODE
564	BITS TMPL_BITS
565
566	%include "bootsector2-template-footer.mac"
567

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

Download in other formats:

Original Format