diff options
author | Kartik Agaram <vc@akkartik.com> | 2018-10-13 00:21:40 -0700 |
---|---|---|
committer | Kartik Agaram <vc@akkartik.com> | 2018-10-13 00:21:40 -0700 |
commit | 861418f0dc55d50521b6276b1022c9652aef0ed8 (patch) | |
tree | 29275fad66966969289ff5d6209b6456267f87c7 /subx | |
parent | 6fc975ebcfb0b140c29927785504ffc63b0c2617 (diff) | |
download | mu-861418f0dc55d50521b6276b1022c9652aef0ed8.tar.gz |
4690
Fix a major discrepancy between the SubX VM and real x86 processors. This was responsible for the breakage identified in commit 4684. We now have failing tests, but at least they are identical running natively and on SubX.
Diffstat (limited to 'subx')
-rw-r--r-- | subx/021byte_addressing.cc | 83 |
1 files changed, 68 insertions, 15 deletions
diff --git a/subx/021byte_addressing.cc b/subx/021byte_addressing.cc index 84a51779..b49718d1 100644 --- a/subx/021byte_addressing.cc +++ b/subx/021byte_addressing.cc @@ -1,18 +1,55 @@ -//: +//: SubX mostly deals with instructions operating on 32-bit operands, but we +//: still need to deal with raw bytes for strings and so on. + +//: Unfortunately the register encodings when dealing with bytes are a mess. +//: We need a special case for them. +:(code) +string rname_8bit(uint8_t r) { + switch (r) { + case 0: return "AL"; // lowest byte of EAX + case 1: return "CL"; // lowest byte of ECX + case 2: return "DL"; // lowest byte of EDX + case 3: return "BL"; // lowest byte of EBX + case 4: return "AH"; // second lowest byte of EAX + case 5: return "CH"; // second lowest byte of ECX + case 6: return "DH"; // second lowest byte of EDX + case 7: return "BH"; // second lowest byte of EBX + default: raise << "invalid 8-bit register " << r << '\n' << end(); return ""; + } +} + +uint8_t* effective_byte_address(uint8_t modrm) { + uint8_t mod = (modrm>>6); + uint8_t rm = modrm & 0x7; + if (mod == 3) { + // select an 8-bit register + trace(90, "run") << "r/m8 is " << rname_8bit(rm) << end(); + return reg_8bit(rm); + } + // the rest is as usual + return mem_addr_u8(effective_address_number(modrm)); +} + +uint8_t* reg_8bit(uint8_t rm) { + uint8_t* result = reinterpret_cast<uint8_t*>(&Reg[rm & 0x3].i); // _L register + if (rm & 0x4) + ++result; // _H register; assumes host is little-endian + return result; +} :(before "End Initialize Op Names(name)") -put(name, "88", "copy r8 (lowermost byte of r32) to r8/m8-at-r32"); +put(name, "88", "copy r8 to r8/m8-at-r32"); :(scenario copy_r8_to_mem_at_r32) % Reg[EBX].i = 0x224488ab; % Reg[EAX].i = 0x2000; == 0x1 # op ModR/M SIB displacement immediate - 88 18 # copy just the lowermost byte of EBX to the byte at *EAX -# ModR/M in binary: 00 (indirect mode) 011 (src EBX) 000 (dest EAX) + 88 18 # copy BL to the byte at *EAX +# ModR/M in binary: 00 (indirect mode) 011 (src BL) 000 (dest EAX) == 0x2000 f0 cc bb aa # 0xf0 with more data in following bytes -+run: copy lowermost byte of EBX to r8/m8-at-r32 ++run: copy BL to r8/m8-at-r32 +run: effective address is 0x2000 (EAX) +run: storing 0xab % CHECK_EQ(0xaabbccab, read_mem_u32(0x2000)); @@ -21,10 +58,11 @@ f0 cc bb aa # 0xf0 with more data in following bytes case 0x88: { // copy r8 to r/m8 const uint8_t modrm = next(); const uint8_t rsrc = (modrm>>3)&0x7; - trace(90, "run") << "copy lowermost byte of " << rname(rsrc) << " to r8/m8-at-r32" << end(); + trace(90, "run") << "copy " << rname_8bit(rsrc) << " to r8/m8-at-r32" << end(); // use unsigned to zero-extend 8-bit value to 32 bits - uint8_t* dest = reinterpret_cast<uint8_t*>(effective_address(modrm)); - *dest = Reg[rsrc].u; + uint8_t* dest = reinterpret_cast<uint8_t*>(effective_byte_address(modrm)); + const uint8_t* src = reg_8bit(rsrc); + *dest = *src; trace(90, "run") << "storing 0x" << HEXBYTE << NUM(*dest) << end(); break; } @@ -32,18 +70,18 @@ case 0x88: { // copy r8 to r/m8 //: :(before "End Initialize Op Names(name)") -put(name, "8a", "copy r8/m8-at-r32 to r8 (lowermost byte of r32)"); +put(name, "8a", "copy r8/m8-at-r32 to r8"); :(scenario copy_mem_at_r32_to_r8) % Reg[EBX].i = 0xaabbcc0f; // one nibble each of lowest byte set to all 0s and all 1s, to maximize value of this test % Reg[EAX].i = 0x2000; == 0x1 # op ModR/M SIB displacement immediate - 8a 18 # copy just the byte at *EAX to lowermost byte of EBX (clearing remaining bytes) + 8a 18 # copy just the byte at *EAX to BL # ModR/M in binary: 00 (indirect mode) 011 (dest EBX) 000 (src EAX) == 0x2000 # data segment ab ff ff ff # 0xab with more data in following bytes -+run: copy r8/m8-at-r32 to lowermost byte of EBX ++run: copy r8/m8-at-r32 to BL +run: effective address is 0x2000 (EAX) +run: storing 0xab # remaining bytes of EBX are *not* cleared @@ -53,11 +91,26 @@ ab ff ff ff # 0xab with more data in following bytes case 0x8a: { // copy r/m8 to r8 const uint8_t modrm = next(); const uint8_t rdest = (modrm>>3)&0x7; - trace(90, "run") << "copy r8/m8-at-r32 to lowermost byte of " << rname(rdest) << end(); + trace(90, "run") << "copy r8/m8-at-r32 to " << rname_8bit(rdest) << end(); // use unsigned to zero-extend 8-bit value to 32 bits - const uint8_t* src = reinterpret_cast<uint8_t*>(effective_address(modrm)); + const uint8_t* src = reinterpret_cast<uint8_t*>(effective_byte_address(modrm)); + uint8_t* dest = reg_8bit(rdest); trace(90, "run") << "storing 0x" << HEXBYTE << NUM(*src) << end(); - *reinterpret_cast<uint8_t*>(&Reg[rdest].u) = *src; // assumes host is little-endian - trace(90, "run") << rname(rdest) << " now contains 0x" << HEXWORD << Reg[rdest].u << end(); + *dest = *src; + const uint8_t rdest_32bit = rdest & 0x3; + trace(90, "run") << rname(rdest_32bit) << " now contains 0x" << HEXWORD << Reg[rdest_32bit].u << end(); break; } + +:(scenario cannot_copy_byte_to_ESP_EBP_ESI_EDI) +% Reg[ESI].u = 0xaabbccdd; +% Reg[EBX].u = 0x11223344; +== 0x1 +# op ModR/M SIB displacement immediate + 8a f3 # copy just the byte at *EBX to 8-bit register '6' +# ModR/M in binary: 11 (direct mode) 110 (dest 8-bit 'register 6') 011 (src EBX) +# ensure 8-bit register '6' is DH, not ESI ++run: copy r8/m8-at-r32 to DH ++run: storing 0x44 +# ensure ESI is unchanged +% CHECK_EQ(Reg[ESI].u, 0xaabbccdd); |