diff options
Diffstat (limited to 'src/VBox/ValidationKit/bootsectors/bootsector2-cpu-instr-1-template.mac')
| -rw-r--r-- | src/VBox/ValidationKit/bootsectors/bootsector2-cpu-instr-1-template.mac | 567 |
1 files changed, 567 insertions, 0 deletions
diff --git a/src/VBox/ValidationKit/bootsectors/bootsector2-cpu-instr-1-template.mac b/src/VBox/ValidationKit/bootsectors/bootsector2-cpu-instr-1-template.mac new file mode 100644 index 00000000..228651be --- /dev/null +++ b/src/VBox/ValidationKit/bootsectors/bootsector2-cpu-instr-1-template.mac @@ -0,0 +1,567 @@ +; $Id: bootsector2-cpu-instr-1-template.mac $ +;; @file +; Bootsector test for misc instruction - multi mode template. +; + +; +; Copyright (C) 2007-2023 Oracle and/or its affiliates. +; +; This file is part of VirtualBox base platform packages, as +; available from https://www.virtualbox.org. +; +; This program is free software; you can redistribute it and/or +; modify it under the terms of the GNU General Public License +; as published by the Free Software Foundation, in version 3 of the +; License. +; +; This program is distributed in the hope that it will be useful, but +; WITHOUT ANY WARRANTY; without even the implied warranty of +; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +; General Public License for more details. +; +; You should have received a copy of the GNU General Public License +; along with this program; if not, see <https://www.gnu.org/licenses>. +; +; The contents of this file may alternatively be used under the terms +; of the Common Development and Distribution License Version 1.0 +; (CDDL), a copy of it is provided in the "COPYING.CDDL" file included +; in the VirtualBox distribution, in which case the provisions of the +; CDDL are applicable instead of those of the GPL. +; +; You may elect to license modified versions of this file under the +; terms and conditions of either the GPL or the CDDL or both. +; +; SPDX-License-Identifier: GPL-3.0-only OR CDDL-1.0 +; + + +%include "bootsector2-template-header.mac" + + + +;; +; Memory fence instructions (SSE2). +; +; @uses No registers, but BS2_SEL_SPARE0 is trashed. +; +BEGINPROC TMPL_NM(TestMemFences) + push xBP + mov xBP, xSP + push sAX + push xBX + push xCX + push xDX + push xDI + push xSI + sub xSP, 80h ; iret stack frame space. + mov xSI, xSP ; Save the stack register. + + mov xAX, .s_szSubTestName + call TMPL_NM_CMN(TestSub) + + ; + ; SSE2 supported? + ; + mov eax, 1 + xor ecx, ecx + cpuid + test edx, X86_CPUID_FEATURE_EDX_SSE2 + jz .skip + + ; + ; Check that the standard instruction encodings work. + ; + mov xBX, [xSP + 10h] + mov [xSP], xAX + mfence + mov [xSP], xCX + mov xBX, [xSP + 08h] + sfence + mov [xSP], xDX + mov xBX, [xSP] + lfence + mov bx, [xSP + 04h] + + + ; + ; The instruction encodings in the intel manual may open the RM as well + ; as prefixes open to interpretation. AMD sets RM=0 in their docs. + ; + ; lfence = 0f,ea,e8 + ; mfence = 0f,ea,f0 + ; sfence = 0f,ea,f8 + ; (RM is the lower 3 bits of the last byte.) + +%assign MY_RM 0xe8 +%rep 18h + db 0fh, 0aeh, MY_RM + db X86_OP_PRF_CS, 0fh, 0aeh, MY_RM + db X86_OP_PRF_DS, 0fh, 0aeh, MY_RM + db X86_OP_PRF_ES, 0fh, 0aeh, MY_RM + db X86_OP_PRF_FS, 0fh, 0aeh, MY_RM + db X86_OP_PRF_GS, 0fh, 0aeh, MY_RM + db X86_OP_PRF_SS, 0fh, 0aeh, MY_RM + db X86_OP_PRF_SIZE_ADDR, 0fh, 0aeh, MY_RM + BS2_TRAP_INSTR X86_XCPT_UD, 0, db X86_OP_PRF_SIZE_OP, 0fh, 0aeh, MY_RM ; (used in group) + BS2_TRAP_INSTR X86_XCPT_UD, 0, db X86_OP_PRF_LOCK, 0fh, 0aeh, MY_RM ; (used in group) + BS2_TRAP_INSTR X86_XCPT_UD, 0, db X86_OP_PRF_REPNZ, 0fh, 0aeh, MY_RM ; (used in group) + BS2_TRAP_INSTR X86_XCPT_UD, 0, db X86_OP_PRF_REPZ, 0fh, 0aeh, MY_RM ; (used in group) +%ifdef TMPL_64BIT + %assign MY_REX 0x40 + %rep 10h + ; Rex prefixes doesn't change anything. + db MY_REX, 0fh, 0aeh, MY_RM + db X86_OP_PRF_CS, MY_REX, 0fh, 0aeh, MY_RM + db X86_OP_PRF_DS, MY_REX, 0fh, 0aeh, MY_RM + db X86_OP_PRF_ES, MY_REX, 0fh, 0aeh, MY_RM + db X86_OP_PRF_FS, MY_REX, 0fh, 0aeh, MY_RM + db X86_OP_PRF_GS, MY_REX, 0fh, 0aeh, MY_RM + db X86_OP_PRF_SS, MY_REX, 0fh, 0aeh, MY_RM + db X86_OP_PRF_SIZE_ADDR, MY_REX, 0fh, 0aeh, MY_RM + BS2_TRAP_INSTR X86_XCPT_UD, 0, db X86_OP_PRF_SIZE_OP, MY_REX, 0fh, 0aeh, MY_RM ; (used in group) + BS2_TRAP_INSTR X86_XCPT_UD, 0, db X86_OP_PRF_LOCK, MY_REX, 0fh, 0aeh, MY_RM ; (used in group) + BS2_TRAP_INSTR X86_XCPT_UD, 0, db X86_OP_PRF_REPNZ, MY_REX, 0fh, 0aeh, MY_RM ; (used in group) + BS2_TRAP_INSTR X86_XCPT_UD, 0, db X86_OP_PRF_REPZ, MY_REX, 0fh, 0aeh, MY_RM ; (used in group) + %assign MY_REX (MY_REX + 1) + %endrep +%endif + %assign MY_RM (MY_RM + 1) +%endrep + + ; + ; Done. + ; + call TMPL_NM_CMN(TestSubDone) +.done: + mov xSP, xSI + add xSP, 80h + pop xSI + pop xDI + pop xDX + pop xCX + pop xBX + pop sAX + leave + ret + +.skip: + mov xAX, .s_szSse2Missing + call TMPL_NM_CMN(TestSubDone) + jmp .done + +.s_szSubTestName: + db TMPL_MODE_STR, ', mfence et al.', 0 +.s_szSse2Missing: + db 'SSE2 is missing', 0 +ENDPROC TMPL_NM(TestMemFences) + + +;; +; Floating-point to integer conversion (SSE/SSE2). +; Neither Intel nor AMD explicitly document what happens for the 32-bit forms +; of CVTxx2SI in 64-bit mode with regard to the high dword of a 64-bit +; destination register. +; +; @uses XMM0, and BS2_SEL_SPARE0 is trashed. +; +BEGINPROC TMPL_NM(TestCvtSize) + push xBP + mov xBP, xSP + push sAX + push xBX + push xCX + push xDX + push xDI + push xSI + sub xSP, 80h ; iret stack frame space. + mov xSI, xSP ; Save the stack register. + + mov xAX, .s_szSubTestName + call TMPL_NM_CMN(TestSub) + + ; + ; SSE2 supported? + ; + mov eax, 1 + xor ecx, ecx + cpuid + test edx, X86_CPUID_FEATURE_EDX_SSE2 + jz .skip + +%ifdef TMPL_64BIT + + ; + ; Have to enable OSFXSR for SSE instructions to work. + ; + mov rcx,cr4 + mov rsi,rcx + or rcx,200h + mov cr4,rcx + + ; + ; Load 32-bit float -2.75 into XMM0 + ; + mov eax, 0C0300000h + movd xmm0, eax + mov rbx, -1 ; make sure high dword is not zero + cvtss2si ebx, xmm0 ; result is -3 + mov eax, -3 ; high dword of rax zeroed + TEST_ASSERT_SIMPLE rbx, rax, jz, "CVTSS2SI EBX" + + mov eax, 0C0300000h + movd xmm0, eax + mov rbx, -1 + cvttss2si ebx, xmm0 ; result is -2 + mov eax, -2 + TEST_ASSERT_SIMPLE rbx, rax, jz, "CVTTSS2SI EBX" + + ; + ; Load 64-bit double -2.75 into XMM0 + ; + mov rax, 0C006000000000000h + movd xmm0, rax + mov rbx, -1 + cvtsd2si ebx, xmm0 + mov eax, -3 + TEST_ASSERT_SIMPLE rbx, rax, jz, "CVTSD2SI EBX" + + mov rax, 0C006000000000000h + mov rbx, -1 + movd xmm0, rax + cvttsd2si ebx, xmm0 + mov eax,-2 + TEST_ASSERT_SIMPLE rbx, rax, jz, "CVTTSD2SI EBX" + + ; + ; Restore prior CR4 value + ; + mov cr4,rsi +%endif + + ; + ; Done. + ; + call TMPL_NM_CMN(TestSubDone) +.done: + mov xSP, xSI + add xSP, 80h + pop xSI + pop xDI + pop xDX + pop xCX + pop xBX + pop sAX + leave + ret + +.skip: + mov xAX, .s_szSse2Missing + call TMPL_NM_CMN(TestSubDone) + jmp .done + +.s_szSubTestName: + db TMPL_MODE_STR, ', cvtss2si et al.', 0 +.s_szSse2Missing: + db 'SSE2 is missing', 0 +ENDPROC TMPL_NM(TestCvtSize) + + +;; +; Test what CMPXCHG with 32-bit operand size does to 64-bit registers, +; as this is not particularly well documented by either Intel or AMD. +; +; @uses No registers, but BS2_SEL_SPARE0 is trashed. +; +BEGINPROC TMPL_NM(TestCmpxchg32) + push xBP + mov xBP, xSP + push sAX + push xBX + push xCX + push xDX + push xDI + push xSI + sub xSP, 80h ; iret stack frame space. + mov xSI, xSP ; Save the stack register. + + mov xAX, .s_szSubTestName + call TMPL_NM_CMN(TestSub) + +%ifdef TMPL_64BIT + + ; + ; CMPXCHG reg, reg - values not equal, eax written + ; + mov rax, -1 ; Load registers with 64-bit values + mov rbx, -2 + mov rcx, -3 + cmpxchg ebx, ecx ; Not equal, writes ebx to eax + mov edx, -2 ; Clears high dword + TEST_ASSERT_SIMPLE rax, rdx, jz, "CMPXCHG reg, unequal, rax set" + mov rdx, -2 ; All ones still in high dword + TEST_ASSERT_SIMPLE rbx, rdx, jz, "CMPXCHG reg, unequal, rbx not set" + + ; + ; CMPXCHG reg, reg - values equal, first operand written + ; + mov rax, -4 ; Load registers with 64-bit values + mov rbx, -4 + mov rcx, -5 + cmpxchg ebx, ecx ; Equal, writes ecx to ebx + mov edx, -5 ; Clears high dword + TEST_ASSERT_SIMPLE rbx, rdx, jz, "CMPXCHG reg, equal, rbx set" + mov rdx, -4 ; All ones still in high dword + TEST_ASSERT_SIMPLE rax, rdx, jz, "CMPXCHG reg, equal, rax not set" + + ; + ; CMPXCHG mem, reg - values not equal, eax written + ; + mov rax, -1 ; Load registers with 64-bit values + mov rbx, -2 + push rbx + mov rcx, -3 + cmpxchg [rsp], ecx ; Not equal, writes eax + mov edx, -2 ; Clears high dword + TEST_ASSERT_SIMPLE rax, rdx, jz, "CMPXCHG mem, unequal, rax set" + pop rbx + + ; + ; CMPXCHG mem, reg - values equal, first operand written + ; + mov rax, -4 ; Load registers with 64-bit values + mov rbx, -4 + push rbx + mov rcx, -5 + cmpxchg [rsp], ecx ; Equal, writes ecx to memory + mov rdx, -4 ; All ones in high dword + TEST_ASSERT_SIMPLE rax, rdx, jz, "CMPXCHG mem, equal, rax not set" + pop rbx + + ; + ; CMPXCHG8B mem, reg - values equal, memory written + ; compares edx:eax with mem64 + ; + mov rdx, -1 ; Load registers with 64-bit values + mov rax, -4 + mov rcx, -1 + mov rbx, -5 + mov rsi, -4 + push rsi + cmpxchg8b [rsp] ; Equal, writes ecx:ebx to memory + mov rsi, -4 ; All ones in high dword + TEST_ASSERT_SIMPLE rax, rsi, jz, "CMPXCHG8B mem, equal, rax not set" + mov rsi, -1 ; All ones in high dword + TEST_ASSERT_SIMPLE rdx, rsi, jz, "CMPXCHG8B mem, equal, rdx not set" + pop rsi + + ; + ; CMPXCHG8B mem, reg - values unequal, edx:eax written + ; compares edx:eax with mem64 + ; + mov rdx, -1 ; Load registers with 64-bit values + mov rax, -2 + mov rcx, -1 + mov rbx, -4 + mov rsi, -3 + push rsi + cmpxchg8b [rsp] ; Not equal, writes memory to edx:eax + mov esi, -3 ; Clears high dword + TEST_ASSERT_SIMPLE rax, rsi, jz, "CMPXCHG8B mem, unequal, rax set" + mov esi, -1 ; Clears high dword + TEST_ASSERT_SIMPLE rdx, rsi, jz, "CMPXCHG8B mem, unequal, rdx set" + pop rsi + +%endif + + ; + ; Done. + ; + call TMPL_NM_CMN(TestSubDone) +.done: + mov xSP, xSI + add xSP, 80h + pop xSI + pop xDI + pop xDX + pop xCX + pop xBX + pop sAX + leave + ret + +.s_szSubTestName: + db TMPL_MODE_STR, ', 32-bit CMPXCHG in 64-bit mode', 0 +ENDPROC TMPL_NM(TestCmpxchg32) + + +;; +; Proving intel manual wrong about using REX.X for BSWAP R8-R15 on 64-bit. +; Checking the 'undefined' 16-bit bswap behavior. +; +; @uses No registers, but BS2_SEL_SPARE0 is trashed. +; +BEGINPROC TMPL_NM(TestBSwap) + push xBP + mov xBP, xSP + push sAX + push xBX + push xCX + push xDX + push xDI + push xSI + sub xSP, 80h ; iret stack frame space. + mov xSI, xSP ; Save the stack register. + + mov xAX, .s_szSubTestName + call TMPL_NM_CMN(TestSub) + + ; + ; Assert sanity. + ; + mov eax, 11223344h + bswap eax + TEST_ASSERT_SIMPLE eax, 44332211h, jz, "32-bit BSWAP EAX" + + ; + ; Buggy manual (325383-041US, December 2011). + ; +%ifdef TMPL_64BIT + push r8 + + mov r8d, 55667788h + mov eax, 55667788h + db X86_OP_REX_X + bswap eax ; does it access r8 or eax? + TEST_ASSERT_SIMPLE eax, 88776655h, jz, "REX.X BSWAP EAX - Wrong EAX." + TEST_ASSERT_SIMPLE r8, 55667788h, jz, "REX.X BSWAP EAX - Wrong R8." + + mov r8d, 55667788h + mov eax, 55667788h + db X86_OP_REX_R + bswap eax ; does it access r8 or eax? + TEST_ASSERT_SIMPLE eax, 88776655h, jz, "REX.R BSWAP EAX - Wrong EAX." + TEST_ASSERT_SIMPLE r8, 55667788h, jz, "REX.R BSWAP EAX - Wrong R8." + + mov r8d, 55667788h + mov eax, 55667788h + db X86_OP_REX_B + bswap eax ; does it access r8 or eax? + TEST_ASSERT_SIMPLE rax, 55667788h, jz, "REX.B BSWAP R8D - Wrong RAX." + TEST_ASSERT_SIMPLE r8d, 88776655h, jz, "REX.B BSWAP R8D - Wrong R8D." + + pop r8 +%endif + + ; + ; 'Undefined' 16-bit behavior. + ; + ; Zeroing of the lower 16-bits has been observed on: + ; - Intel(R) Core(TM) i7-3960X CPU @ 3.30GHz + ; +%ifndef TestBSwap16_defined + %define TestBSwap16_defined + %macro TestBSwap16 3, + mov %3, %2 ; save the primary register. + %ifdef TMPL_64BIT + mov %2, 0ffffffff98765432h ; Set the upper bit as well. + %else + mov %2, 98765432h + %endif + %ifndef TMPL_16BIT + db X86_OP_PRF_SIZE_OP + %endif + bswap %1 + xchg %2, %3 ; Restore and save the result (xSP). + TEST_ASSERT_SIMPLE %3, 98760000h, jz, "Unexpected 16-bit BSWAP error." + %endmacro +%endif + + TestBSwap16 eax, sAX, sSI + TestBSwap16 ebx, sBX, sSI + TestBSwap16 ecx, sCX, sSI + TestBSwap16 edx, sDX, sSI + TestBSwap16 esp, sSP, sSI + TestBSwap16 ebp, sBP, sSI + TestBSwap16 edi, sDI, sSI + TestBSwap16 esi, sSI, sDI +%ifdef TMPL_64BIT + TestBSwap16 r8d, r8, rax + TestBSwap16 r9d, r9, rax + TestBSwap16 r10d, r10, rax + TestBSwap16 r11d, r11, rax + TestBSwap16 r12d, r12, rax + TestBSwap16 r13d, r13, rax + TestBSwap16 r14d, r14, rax + TestBSwap16 r15d, r15, rax +%endif + + ; + ; Done. + ; + call TMPL_NM_CMN(TestSubDone) +.done: + mov xSP, xSI + add xSP, 80h + pop xSI + pop xDI + pop xDX + pop xCX + pop xBX + pop sAX + leave + ret + +.s_szSubTestName: + db TMPL_MODE_STR, ', bswap', 0 +ENDPROC TMPL_NM(TestBSwap) + + +;; +; Do the tests for this mode. +; +; @uses nothing +; +BEGINCODELOW +BITS 16 +BEGINPROC TMPL_NM(DoTestsForMode_rm) + push bp + mov bp, sp + push ax + + ; + ; Check if the mode and NX is supported, do the switch. + ; + call TMPL_NM(Bs2IsModeSupported_rm) + jz .done + call TMPL_NM(Bs2EnterMode_rm) +BITS TMPL_BITS + + ; + ; Test exception handler basics using INT3 and #BP. + ; + + call TMPL_NM(TestMemFences) + call TMPL_NM(TestBSwap) +%ifdef TMPL_64BIT + ; Specifically tests 64-bit behavior. + call TMPL_NM(TestCvtSize) + call TMPL_NM(TestCmpxchg32) +%endif + + ; + ; Back to real mode. + ; + call TMPL_NM(Bs2ExitMode) +BITS 16 + call Bs2DisableNX_r86 + +.done: + pop ax + leave + ret +ENDPROC TMPL_NM(DoTestsForMode_rm) +TMPL_BEGINCODE +BITS TMPL_BITS + +%include "bootsector2-template-footer.mac" + |