From cccf7638989275e57377e2e25595213caacb7da3 Mon Sep 17 00:00:00 2001 From: Mike Pall Date: Fri, 8 Jun 2012 20:54:47 +0200 Subject: Move lib/* to src/jit/*. --- Makefile | 2 +- doc/install.html | 2 +- lib/.gitignore | 1 - lib/bc.lua | 192 ------------ lib/bcsave.lua | 500 ----------------------------- lib/dis_arm.lua | 543 -------------------------------- lib/dis_mips.lua | 428 ------------------------- lib/dis_mipsel.lua | 20 -- lib/dis_ppc.lua | 591 ---------------------------------- lib/dis_x64.lua | 20 -- lib/dis_x86.lua | 836 ------------------------------------------------- lib/dump.lua | 696 ---------------------------------------- lib/v.lua | 167 ---------- src/Makefile | 2 +- src/jit/.gitignore | 1 + src/jit/bc.lua | 192 ++++++++++++ src/jit/bcsave.lua | 500 +++++++++++++++++++++++++++++ src/jit/dis_arm.lua | 543 ++++++++++++++++++++++++++++++++ src/jit/dis_mips.lua | 428 +++++++++++++++++++++++++ src/jit/dis_mipsel.lua | 20 ++ src/jit/dis_ppc.lua | 591 ++++++++++++++++++++++++++++++++++ src/jit/dis_x64.lua | 20 ++ src/jit/dis_x86.lua | 836 +++++++++++++++++++++++++++++++++++++++++++++++++ src/jit/dump.lua | 696 ++++++++++++++++++++++++++++++++++++++++ src/jit/v.lua | 167 ++++++++++ src/msvcbuild.bat | 2 +- 26 files changed, 3998 insertions(+), 3998 deletions(-) delete mode 100644 lib/.gitignore delete mode 100644 lib/bc.lua delete mode 100644 lib/bcsave.lua delete mode 100644 lib/dis_arm.lua delete mode 100644 lib/dis_mips.lua delete mode 100644 lib/dis_mipsel.lua delete mode 100644 lib/dis_ppc.lua delete mode 100644 lib/dis_x64.lua delete mode 100644 lib/dis_x86.lua delete mode 100644 lib/dump.lua delete mode 100644 lib/v.lua create mode 100644 src/jit/.gitignore create mode 100644 src/jit/bc.lua create mode 100644 src/jit/bcsave.lua create mode 100644 src/jit/dis_arm.lua create mode 100644 src/jit/dis_mips.lua create mode 100644 src/jit/dis_mipsel.lua create mode 100644 src/jit/dis_ppc.lua create mode 100644 src/jit/dis_x64.lua create mode 100644 src/jit/dis_x86.lua create mode 100644 src/jit/dump.lua create mode 100644 src/jit/v.lua diff --git a/Makefile b/Makefile index e17703ac..8117692e 100644 --- a/Makefile +++ b/Makefile @@ -120,7 +120,7 @@ install: $(INSTALL_DEP) $(INSTALL_F) $(FILE_PC).tmp $(INSTALL_PC) && \ $(RM) $(FILE_PC).tmp cd src && $(INSTALL_F) $(FILES_INC) $(INSTALL_INC) - cd lib && $(INSTALL_F) $(FILES_JITLIB) $(INSTALL_JITLIB) + cd src/jit && $(INSTALL_F) $(FILES_JITLIB) $(INSTALL_JITLIB) @echo "==== Successfully installed LuaJIT $(VERSION) to $(PREFIX) ====" @echo "" @echo "Note: the beta releases deliberately do NOT install a symlink for luajit" diff --git a/doc/install.html b/doc/install.html index 3731b305..f487958d 100644 --- a/doc/install.html +++ b/doc/install.html @@ -329,7 +329,7 @@ Then follow the installation instructions below. Copy luajit.exe and lua51.dll (built in the src directory) to a newly created directory (any location is ok). Add lua and lua\jit directories below it and copy -all Lua files from the lib directory of the distribution +all Lua files from the src\jit directory of the distribution to the latter directory.

diff --git a/lib/.gitignore b/lib/.gitignore deleted file mode 100644 index 500e2855..00000000 --- a/lib/.gitignore +++ /dev/null @@ -1 +0,0 @@ -vmdef.lua diff --git a/lib/bc.lua b/lib/bc.lua deleted file mode 100644 index 15317bcd..00000000 --- a/lib/bc.lua +++ /dev/null @@ -1,192 +0,0 @@ ----------------------------------------------------------------------------- --- LuaJIT bytecode listing module. --- --- Copyright (C) 2005-2012 Mike Pall. All rights reserved. --- Released under the MIT license. See Copyright Notice in luajit.h ----------------------------------------------------------------------------- --- --- This module lists the bytecode of a Lua function. If it's loaded by -jbc --- it hooks into the parser and lists all functions of a chunk as they --- are parsed. --- --- Example usage: --- --- luajit -jbc -e 'local x=0; for i=1,1e6 do x=x+i end; print(x)' --- luajit -jbc=- foo.lua --- luajit -jbc=foo.list foo.lua --- --- Default output is to stderr. To redirect the output to a file, pass a --- filename as an argument (use '-' for stdout) or set the environment --- variable LUAJIT_LISTFILE. The file is overwritten every time the module --- is started. --- --- This module can also be used programmatically: --- --- local bc = require("jit.bc") --- --- local function foo() print("hello") end --- --- bc.dump(foo) --> -- BYTECODE -- [...] --- print(bc.line(foo, 2)) --> 0002 KSTR 1 1 ; "hello" --- --- local out = { --- -- Do something with each line: --- write = function(t, ...) io.write(...) end, --- close = function(t) end, --- flush = function(t) end, --- } --- bc.dump(foo, out) --- ------------------------------------------------------------------------------- - --- Cache some library functions and objects. -local jit = require("jit") -assert(jit.version_num == 20000, "LuaJIT core/library version mismatch") -local jutil = require("jit.util") -local vmdef = require("jit.vmdef") -local bit = require("bit") -local sub, gsub, format = string.sub, string.gsub, string.format -local byte, band, shr = string.byte, bit.band, bit.rshift -local funcinfo, funcbc, funck = jutil.funcinfo, jutil.funcbc, jutil.funck -local funcuvname = jutil.funcuvname -local bcnames = vmdef.bcnames -local stdout, stderr = io.stdout, io.stderr - ------------------------------------------------------------------------------- - -local function ctlsub(c) - if c == "\n" then return "\\n" - elseif c == "\r" then return "\\r" - elseif c == "\t" then return "\\t" - elseif c == "\r" then return "\\r" - else return format("\\%03d", byte(c)) - end -end - --- Return one bytecode line. -local function bcline(func, pc, prefix) - local ins, m = funcbc(func, pc) - if not ins then return end - local ma, mb, mc = band(m, 7), band(m, 15*8), band(m, 15*128) - local a = band(shr(ins, 8), 0xff) - local oidx = 6*band(ins, 0xff) - local op = sub(bcnames, oidx+1, oidx+6) - local s = format("%04d %s %-6s %3s ", - pc, prefix or " ", op, ma == 0 and "" or a) - local d = shr(ins, 16) - if mc == 13*128 then -- BCMjump - return format("%s=> %04d\n", s, pc+d-0x7fff) - end - if mb ~= 0 then - d = band(d, 0xff) - elseif mc == 0 then - return s.."\n" - end - local kc - if mc == 10*128 then -- BCMstr - kc = funck(func, -d-1) - kc = format(#kc > 40 and '"%.40s"~' or '"%s"', gsub(kc, "%c", ctlsub)) - elseif mc == 9*128 then -- BCMnum - kc = funck(func, d) - if op == "TSETM " then kc = kc - 2^52 end - elseif mc == 12*128 then -- BCMfunc - local fi = funcinfo(funck(func, -d-1)) - if fi.ffid then - kc = vmdef.ffnames[fi.ffid] - else - kc = fi.loc - end - elseif mc == 5*128 then -- BCMuv - kc = funcuvname(func, d) - end - if ma == 5 then -- BCMuv - local ka = funcuvname(func, a) - if kc then kc = ka.." ; "..kc else kc = ka end - end - if mb ~= 0 then - local b = shr(ins, 24) - if kc then return format("%s%3d %3d ; %s\n", s, b, d, kc) end - return format("%s%3d %3d\n", s, b, d) - end - if kc then return format("%s%3d ; %s\n", s, d, kc) end - if mc == 7*128 and d > 32767 then d = d - 65536 end -- BCMlits - return format("%s%3d\n", s, d) -end - --- Collect branch targets of a function. -local function bctargets(func) - local target = {} - for pc=1,1000000000 do - local ins, m = funcbc(func, pc) - if not ins then break end - if band(m, 15*128) == 13*128 then target[pc+shr(ins, 16)-0x7fff] = true end - end - return target -end - --- Dump bytecode instructions of a function. -local function bcdump(func, out, all) - if not out then out = stdout end - local fi = funcinfo(func) - if all and fi.children then - for n=-1,-1000000000,-1 do - local k = funck(func, n) - if not k then break end - if type(k) == "proto" then bcdump(k, out, true) end - end - end - out:write(format("-- BYTECODE -- %s-%d\n", fi.loc, fi.lastlinedefined)) - local target = bctargets(func) - for pc=1,1000000000 do - local s = bcline(func, pc, target[pc] and "=>") - if not s then break end - out:write(s) - end - out:write("\n") - out:flush() -end - ------------------------------------------------------------------------------- - --- Active flag and output file handle. -local active, out - --- List handler. -local function h_list(func) - return bcdump(func, out) -end - --- Detach list handler. -local function bclistoff() - if active then - active = false - jit.attach(h_list) - if out and out ~= stdout and out ~= stderr then out:close() end - out = nil - end -end - --- Open the output file and attach list handler. -local function bcliston(outfile) - if active then bclistoff() end - if not outfile then outfile = os.getenv("LUAJIT_LISTFILE") end - if outfile then - out = outfile == "-" and stdout or assert(io.open(outfile, "w")) - else - out = stderr - end - jit.attach(h_list, "bc") - active = true -end - --- Public module functions. -module(...) - -line = bcline -dump = bcdump -targets = bctargets - -on = bcliston -off = bclistoff -start = bcliston -- For -j command line option. - diff --git a/lib/bcsave.lua b/lib/bcsave.lua deleted file mode 100644 index 70cb6925..00000000 --- a/lib/bcsave.lua +++ /dev/null @@ -1,500 +0,0 @@ ----------------------------------------------------------------------------- --- LuaJIT module to save/list bytecode. --- --- Copyright (C) 2005-2012 Mike Pall. All rights reserved. --- Released under the MIT license. See Copyright Notice in luajit.h ----------------------------------------------------------------------------- --- --- This module saves or lists the bytecode for an input file. --- It's run by the -b command line option. --- ------------------------------------------------------------------------------- - -local jit = require("jit") -assert(jit.version_num == 20000, "LuaJIT core/library version mismatch") - --- Symbol name prefix for LuaJIT bytecode. -local LJBC_PREFIX = "luaJIT_BC_" - ------------------------------------------------------------------------------- - -local function usage() - io.stderr:write[[ -Save LuaJIT bytecode: luajit -b[options] input output - -l Only list bytecode. - -s Strip debug info (default). - -g Keep debug info. - -n name Set module name (default: auto-detect from input name). - -t type Set output file type (default: auto-detect from output name). - -a arch Override architecture for object files (default: native). - -o os Override OS for object files (default: native). - -e chunk Use chunk string as input. - -- Stop handling options. - - Use stdin as input and/or stdout as output. - -File types: c h obj o raw (default) -]] - os.exit(1) -end - -local function check(ok, ...) - if ok then return ok, ... end - io.stderr:write("luajit: ", ...) - io.stderr:write("\n") - os.exit(1) -end - -local function readfile(input) - if type(input) == "function" then return input end - if input == "-" then input = nil end - return check(loadfile(input)) -end - -local function savefile(name, mode) - if name == "-" then return io.stdout end - return check(io.open(name, mode)) -end - ------------------------------------------------------------------------------- - -local map_type = { - raw = "raw", c = "c", h = "h", o = "obj", obj = "obj", -} - -local map_arch = { - x86 = true, x64 = true, arm = true, ppc = true, ppcspe = true, - mips = true, mipsel = true, -} - -local map_os = { - linux = true, windows = true, osx = true, freebsd = true, netbsd = true, - openbsd = true, solaris = true, -} - -local function checkarg(str, map, err) - str = string.lower(str) - local s = check(map[str], "unknown ", err) - return s == true and str or s -end - -local function detecttype(str) - local ext = string.match(string.lower(str), "%.(%a+)$") - return map_type[ext] or "raw" -end - -local function checkmodname(str) - check(string.match(str, "^[%w_.%-]+$"), "bad module name") - return string.gsub(str, "[%.%-]", "_") -end - -local function detectmodname(str) - if type(str) == "string" then - local tail = string.match(str, "[^/\\]+$") - if tail then str = tail end - local head = string.match(str, "^(.*)%.[^.]*$") - if head then str = head end - str = string.match(str, "^[%w_.%-]+") - else - str = nil - end - check(str, "cannot derive module name, use -n name") - return string.gsub(str, "[%.%-]", "_") -end - ------------------------------------------------------------------------------- - -local function bcsave_tail(fp, output, s) - local ok, err = fp:write(s) - if ok and output ~= "-" then ok, err = fp:close() end - check(ok, "cannot write ", output, ": ", err) -end - -local function bcsave_raw(output, s) - local fp = savefile(output, "wb") - bcsave_tail(fp, output, s) -end - -local function bcsave_c(ctx, output, s) - local fp = savefile(output, "w") - if ctx.type == "c" then - fp:write(string.format([[ -#ifdef _cplusplus -extern "C" -#endif -#ifdef _WIN32 -__declspec(dllexport) -#endif -const char %s%s[] = { -]], LJBC_PREFIX, ctx.modname)) - else - fp:write(string.format([[ -#define %s%s_SIZE %d -static const char %s%s[] = { -]], LJBC_PREFIX, ctx.modname, #s, LJBC_PREFIX, ctx.modname)) - end - local t, n, m = {}, 0, 0 - for i=1,#s do - local b = tostring(string.byte(s, i)) - m = m + #b + 1 - if m > 78 then - fp:write(table.concat(t, ",", 1, n), ",\n") - n, m = 0, #b + 1 - end - n = n + 1 - t[n] = b - end - bcsave_tail(fp, output, table.concat(t, ",", 1, n).."\n};\n") -end - -local function bcsave_elfobj(ctx, output, s, ffi) - ffi.cdef[[ -typedef struct { - uint8_t emagic[4], eclass, eendian, eversion, eosabi, eabiversion, epad[7]; - uint16_t type, machine; - uint32_t version; - uint32_t entry, phofs, shofs; - uint32_t flags; - uint16_t ehsize, phentsize, phnum, shentsize, shnum, shstridx; -} ELF32header; -typedef struct { - uint8_t emagic[4], eclass, eendian, eversion, eosabi, eabiversion, epad[7]; - uint16_t type, machine; - uint32_t version; - uint64_t entry, phofs, shofs; - uint32_t flags; - uint16_t ehsize, phentsize, phnum, shentsize, shnum, shstridx; -} ELF64header; -typedef struct { - uint32_t name, type, flags, addr, ofs, size, link, info, align, entsize; -} ELF32sectheader; -typedef struct { - uint32_t name, type; - uint64_t flags, addr, ofs, size; - uint32_t link, info; - uint64_t align, entsize; -} ELF64sectheader; -typedef struct { - uint32_t name, value, size; - uint8_t info, other; - uint16_t sectidx; -} ELF32symbol; -typedef struct { - uint32_t name; - uint8_t info, other; - uint16_t sectidx; - uint64_t value, size; -} ELF64symbol; -typedef struct { - ELF32header hdr; - ELF32sectheader sect[6]; - ELF32symbol sym[2]; - uint8_t space[4096]; -} ELF32obj; -typedef struct { - ELF64header hdr; - ELF64sectheader sect[6]; - ELF64symbol sym[2]; - uint8_t space[4096]; -} ELF64obj; -]] - local symname = LJBC_PREFIX..ctx.modname - local is64, isbe = false, false - if ctx.arch == "x64" then - is64 = true - elseif ctx.arch == "ppc" or ctx.arch == "ppcspe" or ctx.arch == "mips" then - isbe = true - end - - -- Handle different host/target endianess. - local function f32(x) return x end - local f16, fofs = f32, f32 - if ffi.abi("be") ~= isbe then - f32 = bit.bswap - function f16(x) return bit.rshift(bit.bswap(x), 16) end - if is64 then - function fofs(x) return bit.bswap(x)*(2ll^32) end - else - fofs = f32 - end - end - - -- Create ELF object and fill in header. - local o = ffi.new(is64 and "ELF64obj" or "ELF32obj") - local hdr = o.hdr - if ctx.os == "bsd" or ctx.os == "other" then -- Determine native hdr.eosabi. - local bf = assert(io.open("/bin/ls", "rb")) - local bs = bf:read(9) - bf:close() - ffi.copy(o, bs, 9) - check(hdr.emagic[0] == 127, "no support for writing native object files") - else - hdr.emagic = "\127ELF" - hdr.eosabi = ({ freebsd=9, netbsd=2, openbsd=12, solaris=6 })[ctx.os] or 0 - end - hdr.eclass = is64 and 2 or 1 - hdr.eendian = isbe and 2 or 1 - hdr.eversion = 1 - hdr.type = f16(1) - hdr.machine = f16(({ x86=3, x64=62, arm=40, ppc=20, ppcspe=20, mips=8, mipsel=8 })[ctx.arch]) - if ctx.arch == "mips" or ctx.arch == "mipsel" then - hdr.flags = 0x50001006 - end - hdr.version = f32(1) - hdr.shofs = fofs(ffi.offsetof(o, "sect")) - hdr.ehsize = f16(ffi.sizeof(hdr)) - hdr.shentsize = f16(ffi.sizeof(o.sect[0])) - hdr.shnum = f16(6) - hdr.shstridx = f16(2) - - -- Fill in sections and symbols. - local sofs, ofs = ffi.offsetof(o, "space"), 1 - for i,name in ipairs{ - ".symtab", ".shstrtab", ".strtab", ".rodata", ".note.GNU-stack", - } do - local sect = o.sect[i] - sect.align = fofs(1) - sect.name = f32(ofs) - ffi.copy(o.space+ofs, name) - ofs = ofs + #name+1 - end - o.sect[1].type = f32(2) -- .symtab - o.sect[1].link = f32(3) - o.sect[1].info = f32(1) - o.sect[1].align = fofs(8) - o.sect[1].ofs = fofs(ffi.offsetof(o, "sym")) - o.sect[1].entsize = fofs(ffi.sizeof(o.sym[0])) - o.sect[1].size = fofs(ffi.sizeof(o.sym)) - o.sym[1].name = f32(1) - o.sym[1].sectidx = f16(4) - o.sym[1].size = fofs(#s) - o.sym[1].info = 17 - o.sect[2].type = f32(3) -- .shstrtab - o.sect[2].ofs = fofs(sofs) - o.sect[2].size = fofs(ofs) - o.sect[3].type = f32(3) -- .strtab - o.sect[3].ofs = fofs(sofs + ofs) - o.sect[3].size = fofs(#symname+1) - ffi.copy(o.space+ofs+1, symname) - ofs = ofs + #symname + 2 - o.sect[4].type = f32(1) -- .rodata - o.sect[4].flags = fofs(2) - o.sect[4].ofs = fofs(sofs + ofs) - o.sect[4].size = fofs(#s) - o.sect[5].type = f32(1) -- .note.GNU-stack - o.sect[5].ofs = fofs(sofs + ofs + #s) - - -- Write ELF object file. - local fp = savefile(output, "wb") - fp:write(ffi.string(o, ffi.sizeof(o)-4096+ofs)) - bcsave_tail(fp, output, s) -end - -local function bcsave_peobj(ctx, output, s, ffi) - ffi.cdef[[ -typedef struct { - uint16_t arch, nsects; - uint32_t time, symtabofs, nsyms; - uint16_t opthdrsz, flags; -} PEheader; -typedef struct { - char name[8]; - uint32_t vsize, vaddr, size, ofs, relocofs, lineofs; - uint16_t nreloc, nline; - uint32_t flags; -} PEsection; -typedef struct __attribute((packed)) { - union { - char name[8]; - uint32_t nameref[2]; - }; - uint32_t value; - int16_t sect; - uint16_t type; - uint8_t scl, naux; -} PEsym; -typedef struct __attribute((packed)) { - uint32_t size; - uint16_t nreloc, nline; - uint32_t cksum; - uint16_t assoc; - uint8_t comdatsel, unused[3]; -} PEsymaux; -typedef struct { - PEheader hdr; - PEsection sect[2]; - // Must be an even number of symbol structs. - PEsym sym0; - PEsymaux sym0aux; - PEsym sym1; - PEsymaux sym1aux; - PEsym sym2; - PEsym sym3; - uint32_t strtabsize; - uint8_t space[4096]; -} PEobj; -]] - local symname = LJBC_PREFIX..ctx.modname - local is64 = false - if ctx.arch == "x86" then - symname = "_"..symname - elseif ctx.arch == "x64" then - is64 = true - end - local symexport = " /EXPORT:"..symname..",DATA " - - -- The file format is always little-endian. Swap if the host is big-endian. - local function f32(x) return x end - local f16 = f32 - if ffi.abi("be") then - f32 = bit.bswap - function f16(x) return bit.rshift(bit.bswap(x), 16) end - end - - -- Create PE object and fill in header. - local o = ffi.new("PEobj") - local hdr = o.hdr - hdr.arch = f16(({ x86=0x14c, x64=0x8664, arm=0x1c0, ppc=0x1f2, mips=0x366, mipsel=0x366 })[ctx.arch]) - hdr.nsects = f16(2) - hdr.symtabofs = f32(ffi.offsetof(o, "sym0")) - hdr.nsyms = f32(6) - - -- Fill in sections and symbols. - o.sect[0].name = ".drectve" - o.sect[0].size = f32(#symexport) - o.sect[0].flags = f32(0x00100a00) - o.sym0.sect = f16(1) - o.sym0.scl = 3 - o.sym0.name = ".drectve" - o.sym0.naux = 1 - o.sym0aux.size = f32(#symexport) - o.sect[1].name = ".rdata" - o.sect[1].size = f32(#s) - o.sect[1].flags = f32(0x40300040) - o.sym1.sect = f16(2) - o.sym1.scl = 3 - o.sym1.name = ".rdata" - o.sym1.naux = 1 - o.sym1aux.size = f32(#s) - o.sym2.sect = f16(2) - o.sym2.scl = 2 - o.sym2.nameref[1] = f32(4) - o.sym3.sect = f16(-1) - o.sym3.scl = 2 - o.sym3.value = f32(1) - o.sym3.name = "@feat.00" -- Mark as SafeSEH compliant. - ffi.copy(o.space, symname) - local ofs = #symname + 1 - o.strtabsize = f32(ofs + 4) - o.sect[0].ofs = f32(ffi.offsetof(o, "space") + ofs) - ffi.copy(o.space + ofs, symexport) - ofs = ofs + #symexport - o.sect[1].ofs = f32(ffi.offsetof(o, "space") + ofs) - - -- Write PE object file. - local fp = savefile(output, "wb") - fp:write(ffi.string(o, ffi.sizeof(o)-4096+ofs)) - bcsave_tail(fp, output, s) -end - -local function bcsave_machobj(ctx, output, s, ffi) - check(false, "NYI: no support for writing OSX object files") -end - -local function bcsave_obj(ctx, output, s) - local ok, ffi = pcall(require, "ffi") - check(ok, "FFI library required to write this file type") - if ctx.os == "windows" then - return bcsave_peobj(ctx, output, s, ffi) - elseif ctx.os == "osx" then - return bcsave_machobj(ctx, output, s, ffi) - else - return bcsave_elfobj(ctx, output, s, ffi) - end -end - ------------------------------------------------------------------------------- - -local function bclist(input, output) - local f = readfile(input) - require("jit.bc").dump(f, savefile(output, "w"), true) -end - -local function bcsave(ctx, input, output) - local f = readfile(input) - local s = string.dump(f, ctx.strip) - local t = ctx.type - if not t then - t = detecttype(output) - ctx.type = t - end - if t == "raw" then - bcsave_raw(output, s) - else - if not ctx.modname then ctx.modname = detectmodname(input) end - if t == "obj" then - bcsave_obj(ctx, output, s) - else - bcsave_c(ctx, output, s) - end - end -end - -local function docmd(...) - local arg = {...} - local n = 1 - local list = false - local ctx = { - strip = true, arch = jit.arch, os = string.lower(jit.os), - type = false, modname = false, - } - while n <= #arg do - local a = arg[n] - if type(a) == "string" and string.sub(a, 1, 1) == "-" and a ~= "-" then - table.remove(arg, n) - if a == "--" then break end - for m=2,#a do - local opt = string.sub(a, m, m) - if opt == "l" then - list = true - elseif opt == "s" then - ctx.strip = true - elseif opt == "g" then - ctx.strip = false - else - if arg[n] == nil or m ~= #a then usage() end - if opt == "e" then - if n ~= 1 then usage() end - arg[1] = check(loadstring(arg[1])) - elseif opt == "n" then - ctx.modname = checkmodname(table.remove(arg, n)) - elseif opt == "t" then - ctx.type = checkarg(table.remove(arg, n), map_type, "file type") - elseif opt == "a" then - ctx.arch = checkarg(table.remove(arg, n), map_arch, "architecture") - elseif opt == "o" then - ctx.os = checkarg(table.remove(arg, n), map_os, "OS name") - else - usage() - end - end - end - else - n = n + 1 - end - end - if list then - if #arg == 0 or #arg > 2 then usage() end - bclist(arg[1], arg[2] or "-") - else - if #arg ~= 2 then usage() end - bcsave(ctx, arg[1], arg[2]) - end -end - ------------------------------------------------------------------------------- - --- Public module functions. -module(...) - -start = docmd -- Process -b command line option. - diff --git a/lib/dis_arm.lua b/lib/dis_arm.lua deleted file mode 100644 index 0fcd1bed..00000000 --- a/lib/dis_arm.lua +++ /dev/null @@ -1,543 +0,0 @@ ----------------------------------------------------------------------------- --- LuaJIT ARM disassembler module. --- --- Copyright (C) 2005-2012 Mike Pall. All rights reserved. --- Released under the MIT license. See Copyright Notice in luajit.h ----------------------------------------------------------------------------- --- This is a helper module used by the LuaJIT machine code dumper module. --- --- It disassembles most user-mode ARMv7 instructions --- NYI: Advanced SIMD and VFP instructions. ------------------------------------------------------------------------------- - -local type = type -local sub, byte, format = string.sub, string.byte, string.format -local match, gmatch, gsub = string.match, string.gmatch, string.gsub -local concat = table.concat -local bit = require("bit") -local band, bor, ror, tohex = bit.band, bit.bor, bit.ror, bit.tohex -local lshift, rshift, arshift = bit.lshift, bit.rshift, bit.arshift - ------------------------------------------------------------------------------- --- Opcode maps ------------------------------------------------------------------------------- - -local map_loadc = { - shift = 9, mask = 7, - [5] = { - shift = 0, mask = 0 -- NYI VFP load/store. - }, - _ = { - shift = 0, mask = 0 -- NYI ldc, mcrr, mrrc. - }, -} - -local map_datac = { - shift = 24, mask = 1, - [0] = { - shift = 9, mask = 7, - [5] = { - shift = 0, mask = 0 -- NYI VFP data. - }, - _ = { - shift = 0, mask = 0 -- NYI cdp, mcr, mrc. - }, - }, - "svcT", -} - -local map_loadcu = { - shift = 0, mask = 0, -- NYI unconditional CP load/store. -} - -local map_datacu = { - shift = 0, mask = 0, -- NYI unconditional CP data. -} - -local map_simddata = { - shift = 0, mask = 0, -- NYI SIMD data. -} - -local map_simdload = { - shift = 0, mask = 0, -- NYI SIMD load/store, preload. -} - -local map_preload = { - shift = 0, mask = 0, -- NYI preload. -} - -local map_media = { - shift = 20, mask = 31, - [0] = false, - { --01 - shift = 5, mask = 7, - [0] = "sadd16DNM", "sasxDNM", "ssaxDNM", "ssub16DNM", - "sadd8DNM", false, false, "ssub8DNM", - }, - { --02 - shift = 5, mask = 7, - [0] = "qadd16DNM", "qasxDNM", "qsaxDNM", "qsub16DNM", - "qadd8DNM", false, false, "qsub8DNM", - }, - { --03 - shift = 5, mask = 7, - [0] = "shadd16DNM", "shasxDNM", "shsaxDNM", "shsub16DNM", - "shadd8DNM", false, false, "shsub8DNM", - }, - false, - { --05 - shift = 5, mask = 7, - [0] = "uadd16DNM", "uasxDNM", "usaxDNM", "usub16DNM", - "uadd8DNM", false, false, "usub8DNM", - }, - { --06 - shift = 5, mask = 7, - [0] = "uqadd16DNM", "uqasxDNM", "uqsaxDNM", "uqsub16DNM", - "uqadd8DNM", false, false, "uqsub8DNM", - }, - { --07 - shift = 5, mask = 7, - [0] = "uhadd16DNM", "uhasxDNM", "uhsaxDNM", "uhsub16DNM", - "uhadd8DNM", false, false, "uhsub8DNM", - }, - { --08 - shift = 5, mask = 7, - [0] = "pkhbtDNMU", false, "pkhtbDNMU", - { shift = 16, mask = 15, [15] = "sxtb16DMU", _ = "sxtab16DNMU", }, - "pkhbtDNMU", "selDNM", "pkhtbDNMU", - }, - false, - { --0a - shift = 5, mask = 7, - [0] = "ssatDxMu", "ssat16DxM", "ssatDxMu", - { shift = 16, mask = 15, [15] = "sxtbDMU", _ = "sxtabDNMU", }, - "ssatDxMu", false, "ssatDxMu", - }, - { --0b - shift = 5, mask = 7, - [0] = "ssatDxMu", "revDM", "ssatDxMu", - { shift = 16, mask = 15, [15] = "sxthDMU", _ = "sxtahDNMU", }, - "ssatDxMu", "rev16DM", "ssatDxMu", - }, - { --0c - shift = 5, mask = 7, - [3] = { shift = 16, mask = 15, [15] = "uxtb16DMU", _ = "uxtab16DNMU", }, - }, - false, - { --0e - shift = 5, mask = 7, - [0] = "usatDwMu", "usat16DwM", "usatDwMu", - { shift = 16, mask = 15, [15] = "uxtbDMU", _ = "uxtabDNMU", }, - "usatDwMu", false, "usatDwMu", - }, - { --0f - shift = 5, mask = 7, - [0] = "usatDwMu", "rbitDM", "usatDwMu", - { shift = 16, mask = 15, [15] = "uxthDMU", _ = "uxtahDNMU", }, - "usatDwMu", "revshDM", "usatDwMu", - }, - { --10 - shift = 12, mask = 15, - [15] = { - shift = 5, mask = 7, - "smuadNMS", "smuadxNMS", "smusdNMS", "smusdxNMS", - }, - _ = { - shift = 5, mask = 7, - [0] = "smladNMSD", "smladxNMSD", "smlsdNMSD", "smlsdxNMSD", - }, - }, - false, false, false, - { --14 - shift = 5, mask = 7, - [0] = "smlaldDNMS", "smlaldxDNMS", "smlsldDNMS", "smlsldxDNMS", - }, - { --15 - shift = 5, mask = 7, - [0] = { shift = 12, mask = 15, [15] = "smmulNMS", _ = "smmlaNMSD", }, - { shift = 12, mask = 15, [15] = "smmulrNMS", _ = "smmlarNMSD", }, - false, false, false, false, - "smmlsNMSD", "smmlsrNMSD", - }, - false, false, - { --18 - shift = 5, mask = 7, - [0] = { shift = 12, mask = 15, [15] = "usad8NMS", _ = "usada8NMSD", }, - }, - false, - { --1a - shift = 5, mask = 3, [2] = "sbfxDMvw", - }, - { --1b - shift = 5, mask = 3, [2] = "sbfxDMvw", - }, - { --1c - shift = 5, mask = 3, - [0] = { shift = 0, mask = 15, [15] = "bfcDvX", _ = "bfiDMvX", }, - }, - { --1d - shift = 5, mask = 3, - [0] = { shift = 0, mask = 15, [15] = "bfcDvX", _ = "bfiDMvX", }, - }, - { --1e - shift = 5, mask = 3, [2] = "ubfxDMvw", - }, - { --1f - shift = 5, mask = 3, [2] = "ubfxDMvw", - }, -} - -local map_load = { - shift = 21, mask = 9, - { - shift = 20, mask = 5, - [0] = "strtDL", "ldrtDL", [4] = "strbtDL", [5] = "ldrbtDL", - }, - _ = { - shift = 20, mask = 5, - [0] = "strDL", "ldrDL", [4] = "strbDL", [5] = "ldrbDL", - } -} - -local map_load1 = { - shift = 4, mask = 1, - [0] = map_load, map_media, -} - -local map_loadm = { - shift = 20, mask = 1, - [0] = { - shift = 23, mask = 3, - [0] = "stmdaNR", "stmNR", - { shift = 16, mask = 63, [45] = "pushR", _ = "stmdbNR", }, "stmibNR", - }, - { - shift = 23, mask = 3, - [0] = "ldmdaNR", { shift = 16, mask = 63, [61] = "popR", _ = "ldmNR", }, - "ldmdbNR", "ldmibNR", - }, -} - -local map_data = { - shift = 21, mask = 15, - [0] = "andDNPs", "eorDNPs", "subDNPs", "rsbDNPs", - "addDNPs", "adcDNPs", "sbcDNPs", "rscDNPs", - "tstNP", "teqNP", "cmpNP", "cmnNP", - "orrDNPs", "movDPs", "bicDNPs", "mvnDPs", -} - -local map_mul = { - shift = 21, mask = 7, - [0] = "mulNMSs", "mlaNMSDs", "umaalDNMS", "mlsDNMS", - "umullDNMSs", "umlalDNMSs", "smullDNMSs", "smlalDNMSs", -} - -local map_sync = { - shift = 20, mask = 15, -- NYI: brackets around N. R(D+1) for ldrexd/strexd. - [0] = "swpDMN", false, false, false, - "swpbDMN", false, false, false, - "strexDMN", "ldrexDN", "strexdDN", "ldrexdDN", - "strexbDMN", "ldrexbDN", "strexhDN", "ldrexhDN", -} - -local map_mulh = { - shift = 21, mask = 3, - [0] = { shift = 5, mask = 3, - [0] = "smlabbNMSD", "smlatbNMSD", "smlabtNMSD", "smlattNMSD", }, - { shift = 5, mask = 3, - [0] = "smlawbNMSD", "smulwbNMS", "smlawtNMSD", "smulwtNMS", }, - { shift = 5, mask = 3, - [0] = "smlalbbDNMS", "smlaltbDNMS", "smlalbtDNMS", "smlalttDNMS", }, - { shift = 5, mask = 3, - [0] = "smulbbNMS", "smultbNMS", "smulbtNMS", "smulttNMS", }, -} - -local map_misc = { - shift = 4, mask = 7, - -- NYI: decode PSR bits of msr. - [0] = { shift = 21, mask = 1, [0] = "mrsD", "msrM", }, - { shift = 21, mask = 3, "bxM", false, "clzDM", }, - { shift = 21, mask = 3, "bxjM", }, - { shift = 21, mask = 3, "blxM", }, - false, - { shift = 21, mask = 3, [0] = "qaddDMN", "qsubDMN", "qdaddDMN", "qdsubDMN", }, - false, - { shift = 21, mask = 3, "bkptK", }, -} - -local map_datar = { - shift = 4, mask = 9, - [9] = { - shift = 5, mask = 3, - [0] = { shift = 24, mask = 1, [0] = map_mul, map_sync, }, - { shift = 20, mask = 1, [0] = "strhDL", "ldrhDL", }, - { shift = 20, mask = 1, [0] = "ldrdDL", "ldrsbDL", }, - { shift = 20, mask = 1, [0] = "strdDL", "ldrshDL", }, - }, - _ = { - shift = 20, mask = 25, - [16] = { shift = 7, mask = 1, [0] = map_misc, map_mulh, }, - _ = { - shift = 0, mask = 0xffffffff, - [bor(0xe1a00000)] = "nop", - _ = map_data, - } - }, -} - -local map_datai = { - shift = 20, mask = 31, -- NYI: decode PSR bits of msr. Decode imm12. - [16] = "movwDW", [20] = "movtDW", - [18] = { shift = 0, mask = 0xf00ff, [0] = "nopv6", _ = "msrNW", }, - [22] = "msrNW", - _ = map_data, -} - -local map_branch = { - shift = 24, mask = 1, - [0] = "bB", "blB" -} - -local map_condins = { - [0] = map_datar, map_datai, map_load, map_load1, - map_loadm, map_branch, map_loadc, map_datac -} - --- NYI: setend. -local map_uncondins = { - [0] = false, map_simddata, map_simdload, map_preload, - false, "blxB", map_loadcu, map_datacu, -} - ------------------------------------------------------------------------------- - -local map_gpr = { - [0] = "r0", "r1", "r2", "r3", "r4", "r5", "r6", "r7", - "r8", "r9", "r10", "r11", "r12", "sp", "lr", "pc", -} - -local map_cond = { - [0] = "eq", "ne", "hs", "lo", "mi", "pl", "vs", "vc", - "hi", "ls", "ge", "lt", "gt", "le", "al", -} - -local map_shift = { [0] = "lsl", "lsr", "asr", "ror", } - ------------------------------------------------------------------------------- - --- Output a nicely formatted line with an opcode and operands. -local function putop(ctx, text, operands) - local pos = ctx.pos - local extra = "" - if ctx.rel then - local sym = ctx.symtab[ctx.rel] - if sym then - extra = "\t->"..sym - elseif band(ctx.op, 0x0e000000) ~= 0x0a000000 then - extra = "\t; 0x"..tohex(ctx.rel) - end - end - if ctx.hexdump > 0 then - ctx.out(format("%08x %s %-5s %s%s\n", - ctx.addr+pos, tohex(ctx.op), text, concat(operands, ", "), extra)) - else - ctx.out(format("%08x %-5s %s%s\n", - ctx.addr+pos, text, concat(operands, ", "), extra)) - end - ctx.pos = pos + 4 -end - --- Fallback for unknown opcodes. -local function unknown(ctx) - return putop(ctx, ".long", { "0x"..tohex(ctx.op) }) -end - --- Format operand 2 of load/store opcodes. -local function fmtload(ctx, op, pos) - local base = map_gpr[band(rshift(op, 16), 15)] - local x, ofs - local ext = (band(op, 0x04000000) == 0) - if not ext and band(op, 0x02000000) == 0 then - ofs = band(op, 4095) - if band(op, 0x00800000) == 0 then ofs = -ofs end - if base == "pc" then ctx.rel = ctx.addr + pos + 8 + ofs end - ofs = "#"..ofs - elseif ext and band(op, 0x00400000) ~= 0 then - ofs = band(op, 15) + band(rshift(op, 4), 0xf0) - if band(op, 0x00800000) == 0 then ofs = -ofs end - if base == "pc" then ctx.rel = ctx.addr + pos + 8 + ofs end - ofs = "#"..ofs - else - ofs = map_gpr[band(op, 15)] - if ext or band(op, 0xfe0) == 0 then - elseif band(op, 0xfe0) == 0x60 then - ofs = format("%s, rrx", ofs) - else - local sh = band(rshift(op, 7), 31) - if sh == 0 then sh = 32 end - ofs = format("%s, %s #%d", ofs, map_shift[band(rshift(op, 5), 3)], sh) - end - if band(op, 0x00800000) == 0 then ofs = "-"..ofs end - end - if ofs == "#0" then - x = format("[%s]", base) - elseif band(op, 0x01000000) == 0 then - x = format("[%s], %s", base, ofs) - else - x = format("[%s, %s]", base, ofs) - end - if band(op, 0x01200000) == 0x01200000 then x = x.."!" end - return x -end - --- Disassemble a single instruction. -local function disass_ins(ctx) - local pos = ctx.pos - local b0, b1, b2, b3 = byte(ctx.code, pos+1, pos+4) - local op = bor(lshift(b3, 24), lshift(b2, 16), lshift(b1, 8), b0) - local operands = {} - local suffix = "" - local last, name, pat - ctx.op = op - ctx.rel = nil - - local cond = rshift(op, 28) - local opat - if cond == 15 then - opat = map_uncondins[band(rshift(op, 25), 7)] - else - if cond ~= 14 then suffix = map_cond[cond] end - opat = map_condins[band(rshift(op, 25), 7)] - end - while type(opat) ~= "string" do - if not opat then return unknown(ctx) end - opat = opat[band(rshift(op, opat.shift), opat.mask)] or opat._ - end - name, pat = match(opat, "^([a-z0-9]*)(.*)") - - for p in gmatch(pat, ".") do - local x = nil - if p == "D" then - x = map_gpr[band(rshift(op, 12), 15)] - elseif p == "N" then - x = map_gpr[band(rshift(op, 16), 15)] - elseif p == "S" then - x = map_gpr[band(rshift(op, 8), 15)] - elseif p == "M" then - x = map_gpr[band(op, 15)] - elseif p == "P" then - if band(op, 0x02000000) ~= 0 then - x = ror(band(op, 255), 2*band(rshift(op, 8), 15)) - else - x = map_gpr[band(op, 15)] - if band(op, 0xff0) ~= 0 then - operands[#operands+1] = x - local s = map_shift[band(rshift(op, 5), 3)] - local r = nil - if band(op, 0xf90) == 0 then - if s == "ror" then s = "rrx" else r = "#32" end - elseif band(op, 0x10) == 0 then - r = "#"..band(rshift(op, 7), 31) - else - r = map_gpr[band(rshift(op, 8), 15)] - end - if name == "mov" then name = s; x = r - elseif r then x = format("%s %s", s, r) - else x = s end - end - end - elseif p == "L" then - x = fmtload(ctx, op, pos, false) - elseif p == "B" then - local addr = ctx.addr + pos + 8 + arshift(lshift(op, 8), 6) - if cond == 15 then addr = addr + band(rshift(op, 23), 2) end - ctx.rel = addr - x = "0x"..tohex(addr) - elseif p == "R" then - if band(op, 0x00200000) ~= 0 and #operands == 1 then - operands[1] = operands[1].."!" - end - local t = {} - for i=0,15 do - if band(rshift(op, i), 1) == 1 then t[#t+1] = map_gpr[i] end - end - x = "{"..concat(t, ", ").."}" - elseif p == "W" then - x = band(op, 0x0fff) + band(rshift(op, 4), 0xf000) - elseif p == "T" then - x = "#0x"..tohex(band(op, 0x00ffffff), 6) - elseif p == "U" then - x = band(rshift(op, 7), 31) - if x == 0 then x = nil end - elseif p == "u" then - x = band(rshift(op, 7), 31) - if band(op, 0x40) == 0 then - if x == 0 then x = nil else x = "lsl #"..x end - else - if x == 0 then x = "asr #32" else x = "asr #"..x end - end - elseif p == "v" then - x = band(rshift(op, 7), 31) - elseif p == "w" then - x = band(rshift(op, 16), 31) - elseif p == "x" then - x = band(rshift(op, 16), 31) + 1 - elseif p == "X" then - x = band(rshift(op, 16), 31) - last + 1 - elseif p == "K" then - x = "#0x"..tohex(band(rshift(op, 4), 0x0000fff0) + band(op, 15), 4) - elseif p == "s" then - if band(op, 0x00100000) ~= 0 then suffix = "s"..suffix end - else - assert(false) - end - if x then - last = x - if type(x) == "number" then x = "#"..x end - operands[#operands+1] = x - end - end - - return putop(ctx, name..suffix, operands) -end - ------------------------------------------------------------------------------- - --- Disassemble a block of code. -local function disass_block(ctx, ofs, len) - if not ofs then ofs = 0 end - local stop = len and ofs+len or #ctx.code - ctx.pos = ofs - ctx.rel = nil - while ctx.pos < stop do disass_ins(ctx) end -end - --- Extended API: create a disassembler context. Then call ctx:disass(ofs, len). -local function create_(code, addr, out) - local ctx = {} - ctx.code = code - ctx.addr = addr or 0 - ctx.out = out or io.write - ctx.symtab = {} - ctx.disass = disass_block - ctx.hexdump = 8 - return ctx -end - --- Simple API: disassemble code (a string) at address and output via out. -local function disass_(code, addr, out) - create_(code, addr, out):disass() -end - --- Return register name for RID. -local function regname_(r) - return map_gpr[r] -end - --- Public module functions. -module(...) - -create = create_ -disass = disass_ -regname = regname_ - diff --git a/lib/dis_mips.lua b/lib/dis_mips.lua deleted file mode 100644 index 2edfdf40..00000000 --- a/lib/dis_mips.lua +++ /dev/null @@ -1,428 +0,0 @@ ----------------------------------------------------------------------------- --- LuaJIT MIPS disassembler module. --- --- Copyright (C) 2005-2011 Mike Pall. All rights reserved. --- Released under the MIT/X license. See Copyright Notice in luajit.h ----------------------------------------------------------------------------- --- This is a helper module used by the LuaJIT machine code dumper module. --- --- It disassembles all standard MIPS32R1/R2 instructions. --- Default mode is big-endian, but see: dis_mipsel.lua ------------------------------------------------------------------------------- - -local type = type -local sub, byte, format = string.sub, string.byte, string.format -local match, gmatch, gsub = string.match, string.gmatch, string.gsub -local concat = table.concat -local bit = require("bit") -local band, bor, tohex = bit.band, bit.bor, bit.tohex -local lshift, rshift, arshift = bit.lshift, bit.rshift, bit.arshift - ------------------------------------------------------------------------------- --- Primary and extended opcode maps ------------------------------------------------------------------------------- - -local map_movci = { shift = 16, mask = 1, [0] = "movfDSC", "movtDSC", } -local map_srl = { shift = 21, mask = 1, [0] = "srlDTA", "rotrDTA", } -local map_srlv = { shift = 6, mask = 1, [0] = "srlvDTS", "rotrvDTS", } - -local map_special = { - shift = 0, mask = 63, - [0] = { shift = 0, mask = -1, [0] = "nop", _ = "sllDTA" }, - map_movci, map_srl, "sraDTA", - "sllvDTS", false, map_srlv, "sravDTS", - "jrS", "jalrD1S", "movzDST", "movnDST", - "syscallY", "breakY", false, "sync", - "mfhiD", "mthiS", "mfloD", "mtloS", - false, false, false, false, - "multST", "multuST", "divST", "divuST", - false, false, false, false, - "addDST", "addu|moveDST0", "subDST", "subu|neguDS0T", - "andDST", "orDST", "xorDST", "nor|notDST0", - false, false, "sltDST", "sltuDST", - false, false, false, false, - "tgeSTZ", "tgeuSTZ", "tltSTZ", "tltuSTZ", - "teqSTZ", false, "tneSTZ", -} - -local map_special2 = { - shift = 0, mask = 63, - [0] = "maddST", "madduST", "mulDST", false, - "msubST", "msubuST", - [32] = "clzDS", [33] = "cloDS", - [63] = "sdbbpY", -} - -local map_bshfl = { - shift = 6, mask = 31, - [2] = "wsbhDT", - [16] = "sebDT", - [24] = "sehDT", -} - -local map_special3 = { - shift = 0, mask = 63, - [0] = "extTSAK", [4] = "insTSAL", - [32] = map_bshfl, - [59] = "rdhwrTD", -} - -local map_regimm = { - shift = 16, mask = 31, - [0] = "bltzSB", "bgezSB", "bltzlSB", "bgezlSB", - false, false, false, false, - "tgeiSI", "tgeiuSI", "tltiSI", "tltiuSI", - "teqiSI", false, "tneiSI", false, - "bltzalSB", "bgezalSB", "bltzallSB", "bgezallSB", - false, false, false, false, - false, false, false, false, - false, false, false, "synciSO", -} - -local map_cop0 = { - shift = 25, mask = 1, - [0] = { - shift = 21, mask = 15, - [0] = "mfc0TDW", [4] = "mtc0TDW", - [10] = "rdpgprDT", - [11] = { shift = 5, mask = 1, [0] = "diT0", "eiT0", }, - [14] = "wrpgprDT", - }, { - shift = 0, mask = 63, - [1] = "tlbr", [2] = "tlbwi", [6] = "tlbwr", [8] = "tlbp", - [24] = "eret", [31] = "deret", - [32] = "wait", - }, -} - -local map_cop1s = { - shift = 0, mask = 63, - [0] = "add.sFGH", "sub.sFGH", "mul.sFGH", "div.sFGH", - "sqrt.sFG", "abs.sFG", "mov.sFG", "neg.sFG", - "round.l.sFG", "trunc.l.sFG", "ceil.l.sFG", "floor.l.sFG", - "round.w.sFG", "trunc.w.sFG", "ceil.w.sFG", "floor.w.sFG", - false, - { shift = 16, mask = 1, [0] = "movf.sFGC", "movt.sFGC" }, - "movz.sFGT", "movn.sFGT", - false, "recip.sFG", "rsqrt.sFG", false, - false, false, false, false, - false, false, false, false, - false, "cvt.d.sFG", false, false, - "cvt.w.sFG", "cvt.l.sFG", "cvt.ps.sFGH", false, - false, false, false, false, - false, false, false, false, - "c.f.sVGH", "c.un.sVGH", "c.eq.sVGH", "c.ueq.sVGH", - "c.olt.sVGH", "c.ult.sVGH", "c.ole.sVGH", "c.ule.sVGH", - "c.sf.sVGH", "c.ngle.sVGH", "c.seq.sVGH", "c.ngl.sVGH", - "c.lt.sVGH", "c.nge.sVGH", "c.le.sVGH", "c.ngt.sVGH", -} - -local map_cop1d = { - shift = 0, mask = 63, - [0] = "add.dFGH", "sub.dFGH", "mul.dFGH", "div.dFGH", - "sqrt.dFG", "abs.dFG", "mov.dFG", "neg.dFG", - "round.l.dFG", "trunc.l.dFG", "ceil.l.dFG", "floor.l.dFG", - "round.w.dFG", "trunc.w.dFG", "ceil.w.dFG", "floor.w.dFG", - false, - { shift = 16, mask = 1, [0] = "movf.dFGC", "movt.dFGC" }, - "movz.dFGT", "movn.dFGT", - false, "recip.dFG", "rsqrt.dFG", false, - false, false, false, false, - false, false, false, false, - "cvt.s.dFG", false, false, false, - "cvt.w.dFG", "cvt.l.dFG", false, false, - false, false, false, false, - false, false, false, false, - "c.f.dVGH", "c.un.dVGH", "c.eq.dVGH", "c.ueq.dVGH", - "c.olt.dVGH", "c.ult.dVGH", "c.ole.dVGH", "c.ule.dVGH", - "c.df.dVGH", "c.ngle.dVGH", "c.deq.dVGH", "c.ngl.dVGH", - "c.lt.dVGH", "c.nge.dVGH", "c.le.dVGH", "c.ngt.dVGH", -} - -local map_cop1ps = { - shift = 0, mask = 63, - [0] = "add.psFGH", "sub.psFGH", "mul.psFGH", false, - false, "abs.psFG", "mov.psFG", "neg.psFG", - false, false, false, false, - false, false, false, false, - false, - { shift = 16, mask = 1, [0] = "movf.psFGC", "movt.psFGC" }, - "movz.psFGT", "movn.psFGT", - false, false, false, false, - false, false, false, false, - false, false, false, false, - "cvt.s.puFG", false, false, false, - false, false, false, false, - "cvt.s.plFG", false, false, false, - "pll.psFGH", "plu.psFGH", "pul.psFGH", "puu.psFGH", - "c.f.psVGH", "c.un.psVGH", "c.eq.psVGH", "c.ueq.psVGH", - "c.olt.psVGH", "c.ult.psVGH", "c.ole.psVGH", "c.ule.psVGH", - "c.psf.psVGH", "c.ngle.psVGH", "c.pseq.psVGH", "c.ngl.psVGH", - "c.lt.psVGH", "c.nge.psVGH", "c.le.psVGH", "c.ngt.psVGH", -} - -local map_cop1w = { - shift = 0, mask = 63, - [32] = "cvt.s.wFG", [33] = "cvt.d.wFG", -} - -local map_cop1l = { - shift = 0, mask = 63, - [32] = "cvt.s.lFG", [33] = "cvt.d.lFG", -} - -local map_cop1bc = { - shift = 16, mask = 3, - [0] = "bc1fCB", "bc1tCB", "bc1flCB", "bc1tlCB", -} - -local map_cop1 = { - shift = 21, mask = 31, - [0] = "mfc1TG", false, "cfc1TG", "mfhc1TG", - "mtc1TG", false, "ctc1TG", "mthc1TG", - map_cop1bc, false, false, false, - false, false, false, false, - map_cop1s, map_cop1d, false, false, - map_cop1w, map_cop1l, map_cop1ps, -} - -local map_cop1x = { - shift = 0, mask = 63, - [0] = "lwxc1FSX", "ldxc1FSX", false, false, - false, "luxc1FSX", false, false, - "swxc1FSX", "sdxc1FSX", false, false, - false, "suxc1FSX", false, "prefxMSX", - false, false, false, false, - false, false, false, false, - false, false, false, false, - false, false, "alnv.psFGHS", false, - "madd.sFRGH", "madd.dFRGH", false, false, - false, false, "madd.psFRGH", false, - "msub.sFRGH", "msub.dFRGH", false, false, - false, false, "msub.psFRGH", false, - "nmadd.sFRGH", "nmadd.dFRGH", false, false, - false, false, "nmadd.psFRGH", false, - "nmsub.sFRGH", "nmsub.dFRGH", false, false, - false, false, "nmsub.psFRGH", false, -} - -local map_pri = { - [0] = map_special, map_regimm, "jJ", "jalJ", - "beq|beqz|bST00B", "bne|bnezST0B", "blezSB", "bgtzSB", - "addiTSI", "addiu|liTS0I", "sltiTSI", "sltiuTSI", - "andiTSU", "ori|liTS0U", "xoriTSU", "luiTU", - map_cop0, map_cop1, false, map_cop1x, - "beql|beqzlST0B", "bnel|bnezlST0B", "blezlSB", "bgtzlSB", - false, false, false, false, - map_special2, false, false, map_special3, - "lbTSO", "lhTSO", "lwlTSO", "lwTSO", - "lbuTSO", "lhuTSO", "lwrTSO", false, - "sbTSO", "shTSO", "swlTSO", "swTSO", - false, false, "swrTSO", "cacheNSO", - "llTSO", "lwc1HSO", "lwc2TSO", "prefNSO", - false, "ldc1HSO", "ldc2TSO", false, - "scTSO", "swc1HSO", "swc2TSO", false, - false, "sdc1HSO", "sdc2TSO", false, -} - ------------------------------------------------------------------------------- - -local map_gpr = { - [0] = "r0", "r1", "r2", "r3", "r4", "r5", "r6", "r7", - "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15", - "r16", "r17", "r18", "r19", "r20", "r21", "r22", "r23", - "r24", "r25", "r26", "r27", "r28", "sp", "r30", "ra", -} - ------------------------------------------------------------------------------- - --- Output a nicely formatted line with an opcode and operands. -local function putop(ctx, text, operands) - local pos = ctx.pos - local extra = "" - if ctx.rel then - local sym = ctx.symtab[ctx.rel] - if sym then extra = "\t->"..sym end - end - if ctx.hexdump > 0 then - ctx.out(format("%08x %s %-7s %s%s\n", - ctx.addr+pos, tohex(ctx.op), text, concat(operands, ", "), extra)) - else - ctx.out(format("%08x %-7s %s%s\n", - ctx.addr+pos, text, concat(operands, ", "), extra)) - end - ctx.pos = pos + 4 -end - --- Fallback for unknown opcodes. -local function unknown(ctx) - return putop(ctx, ".long", { "0x"..tohex(ctx.op) }) -end - -local function get_be(ctx) - local pos = ctx.pos - local b0, b1, b2, b3 = byte(ctx.code, pos+1, pos+4) - return bor(lshift(b0, 24), lshift(b1, 16), lshift(b2, 8), b3) -end - -local function get_le(ctx) - local pos = ctx.pos - local b0, b1, b2, b3 = byte(ctx.code, pos+1, pos+4) - return bor(lshift(b3, 24), lshift(b2, 16), lshift(b1, 8), b0) -end - --- Disassemble a single instruction. -local function disass_ins(ctx) - local op = ctx:get() - local operands = {} - local last = nil - ctx.op = op - ctx.rel = nil - - local opat = map_pri[rshift(op, 26)] - while type(opat) ~= "string" do - if not opat then return unknown(ctx) end - opat = opat[band(rshift(op, opat.shift), opat.mask)] or opat._ - end - local name, pat = match(opat, "^([a-z0-9_.]*)(.*)") - local altname, pat2 = match(pat, "|([a-z0-9_.|]*)(.*)") - if altname then pat = pat2 end - - for p in gmatch(pat, ".") do - local x = nil - if p == "S" then - x = map_gpr[band(rshift(op, 21), 31)] - elseif p == "T" then - x = map_gpr[band(rshift(op, 16), 31)] - elseif p == "D" then - x = map_gpr[band(rshift(op, 11), 31)] - elseif p == "F" then - x = "f"..band(rshift(op, 6), 31) - elseif p == "G" then - x = "f"..band(rshift(op, 11), 31) - elseif p == "H" then - x = "f"..band(rshift(op, 16), 31) - elseif p == "R" then - x = "f"..band(rshift(op, 21), 31) - elseif p == "A" then - x = band(rshift(op, 6), 31) - elseif p == "M" then - x = band(rshift(op, 11), 31) - elseif p == "N" then - x = band(rshift(op, 16), 31) - elseif p == "C" then - x = band(rshift(op, 18), 7) - if x == 0 then x = nil end - elseif p == "K" then - x = band(rshift(op, 11), 31) + 1 - elseif p == "L" then - x = band(rshift(op, 11), 31) - last + 1 - elseif p == "I" then - x = arshift(lshift(op, 16), 16) - elseif p == "U" then - x = band(op, 0xffff) - elseif p == "O" then - local disp = arshift(lshift(op, 16), 16) - operands[#operands] = format("%d(%s)", disp, last) - elseif p == "X" then - local index = map_gpr[band(rshift(op, 16), 31)] - operands[#operands] = format("%s(%s)", index, last) - elseif p == "B" then - x = ctx.addr + ctx.pos + arshift(lshift(op, 16), 16)*4 + 4 - ctx.rel = x - x = "0x"..tohex(x) - elseif p == "J" then - x = band(ctx.addr + ctx.pos, 0xf0000000) + band(op, 0x03ffffff)*4 - ctx.rel = x - x = "0x"..tohex(x) - elseif p == "V" then - x = band(rshift(op, 8), 7) - if x == 0 then x = nil end - elseif p == "W" then - x = band(op, 7) - if x == 0 then x = nil end - elseif p == "Y" then - x = band(rshift(op, 6), 0x000fffff) - if x == 0 then x = nil end - elseif p == "Z" then - x = band(rshift(op, 6), 1023) - if x == 0 then x = nil end - elseif p == "0" then - if last == "r0" or last == 0 then - local n = #operands - operands[n] = nil - last = operands[n-1] - if altname then - local a1, a2 = match(altname, "([^|]*)|(.*)") - if a1 then name, altname = a1, a2 - else name = altname end - end - end - elseif p == "1" then - if last == "ra" then - operands[#operands] = nil - end - else - assert(false) - end - if x then operands[#operands+1] = x; last = x end - end - - return putop(ctx, name, operands) -end - ------------------------------------------------------------------------------- - --- Disassemble a block of code. -local function disass_block(ctx, ofs, len) - if not ofs then ofs = 0 end - local stop = len and ofs+len or #ctx.code - stop = stop - stop % 4 - ctx.pos = ofs - ofs % 4 - ctx.rel = nil - while ctx.pos < stop do disass_ins(ctx) end -end - --- Extended API: create a disassembler context. Then call ctx:disass(ofs, len). -local function create_(code, addr, out) - local ctx = {} - ctx.code = code - ctx.addr = addr or 0 - ctx.out = out or io.write - ctx.symtab = {} - ctx.disass = disass_block - ctx.hexdump = 8 - ctx.get = get_be - return ctx -end - -local function create_el_(code, addr, out) - local ctx = create_(code, addr, out) - ctx.get = get_le - return ctx -end - --- Simple API: disassemble code (a string) at address and output via out. -local function disass_(code, addr, out) - create_(code, addr, out):disass() -end - -local function disass_el_(code, addr, out) - create_el_(code, addr, out):disass() -end - --- Return register name for RID. -local function regname_(r) - if r < 32 then return map_gpr[r] end - return "f"..(r-32) -end - --- Public module functions. -module(...) - -create = create_ -create_el = create_el_ -disass = disass_ -disass_el = disass_el_ -regname = regname_ - diff --git a/lib/dis_mipsel.lua b/lib/dis_mipsel.lua deleted file mode 100644 index 4c5a6515..00000000 --- a/lib/dis_mipsel.lua +++ /dev/null @@ -1,20 +0,0 @@ ----------------------------------------------------------------------------- --- LuaJIT MIPSEL disassembler wrapper module. --- --- Copyright (C) 2005-2011 Mike Pall. All rights reserved. --- Released under the MIT license. See Copyright Notice in luajit.h ----------------------------------------------------------------------------- --- This module just exports the little-endian functions from the --- MIPS disassembler module. All the interesting stuff is there. ------------------------------------------------------------------------------- - -local require = require - -module(...) - -local dis_mips = require(_PACKAGE.."dis_mips") - -create = dis_mips.create_el -disass = dis_mips.disass_el -regname = dis_mips.regname - diff --git a/lib/dis_ppc.lua b/lib/dis_ppc.lua deleted file mode 100644 index ad8599fc..00000000 --- a/lib/dis_ppc.lua +++ /dev/null @@ -1,591 +0,0 @@ ----------------------------------------------------------------------------- --- LuaJIT PPC disassembler module. --- --- Copyright (C) 2005-2012 Mike Pall. All rights reserved. --- Released under the MIT/X license. See Copyright Notice in luajit.h ----------------------------------------------------------------------------- --- This is a helper module used by the LuaJIT machine code dumper module. --- --- It disassembles all common, non-privileged 32/64 bit PowerPC instructions --- plus the e500 SPE instructions and some Cell/Xenon extensions. --- --- NYI: VMX, VMX128 ------------------------------------------------------------------------------- - -local type = type -local sub, byte, format = string.sub, string.byte, string.format -local match, gmatch, gsub = string.match, string.gmatch, string.gsub -local concat = table.concat -local bit = require("bit") -local band, bor, tohex = bit.band, bit.bor, bit.tohex -local lshift, rshift, arshift = bit.lshift, bit.rshift, bit.arshift - ------------------------------------------------------------------------------- --- Primary and extended opcode maps ------------------------------------------------------------------------------- - -local map_crops = { - shift = 1, mask = 1023, - [0] = "mcrfXX", - [33] = "crnor|crnotCCC=", [129] = "crandcCCC", - [193] = "crxor|crclrCCC%", [225] = "crnandCCC", - [257] = "crandCCC", [289] = "creqv|crsetCCC%", - [417] = "crorcCCC", [449] = "cror|crmoveCCC=", - [16] = "b_lrKB", [528] = "b_ctrKB", - [150] = "isync", -} - -local map_rlwinm = setmetatable({ - shift = 0, mask = -1, -}, -{ __index = function(t, x) - local rot = band(rshift(x, 11), 31) - local mb = band(rshift(x, 6), 31) - local me = band(rshift(x, 1), 31) - if mb == 0 and me == 31-rot then - return "slwiRR~A." - elseif me == 31 and mb == 32-rot then - return "srwiRR~-A." - else - return "rlwinmRR~AAA." - end - end -}) - -local map_rld = { - shift = 2, mask = 7, - [0] = "rldiclRR~HM.", "rldicrRR~HM.", "rldicRR~HM.", "rldimiRR~HM.", - { - shift = 1, mask = 1, - [0] = "rldclRR~HM.", "rldcrRR~HM.", - }, -} - -local map_ext = setmetatable({ - shift = 1, mask = 1023, - - [0] = "cmp_YLRR", [32] = "cmpl_YLRR", - [4] = "twARR", [68] = "tdARR", - - [8] = "subfcRRR.", [40] = "subfRRR.", - [104] = "negRR.", [136] = "subfeRRR.", - [200] = "subfzeRR.", [232] = "subfmeRR.", - [520] = "subfcoRRR.", [552] = "subfoRRR.", - [616] = "negoRR.", [648] = "subfeoRRR.", - [712] = "subfzeoRR.", [744] = "subfmeoRR.", - - [9] = "mulhduRRR.", [73] = "mulhdRRR.", [233] = "mulldRRR.", - [457] = "divduRRR.", [489] = "divdRRR.", - [745] = "mulldoRRR.", - [969] = "divduoRRR.", [1001] = "divdoRRR.", - - [10] = "addcRRR.", [138] = "addeRRR.", - [202] = "addzeRR.", [234] = "addmeRR.", [266] = "addRRR.", - [522] = "addcoRRR.", [650] = "addeoRRR.", - [714] = "addzeoRR.", [746] = "addmeoRR.", [778] = "addoRRR.", - - [11] = "mulhwuRRR.", [75] = "mulhwRRR.", [235] = "mullwRRR.", - [459] = "divwuRRR.", [491] = "divwRRR.", - [747] = "mullwoRRR.", - [971] = "divwouRRR.", [1003] = "divwoRRR.", - - [15] = "iselltRRR", [47] = "iselgtRRR", [79] = "iseleqRRR", - - [144] = { shift = 20, mask = 1, [0] = "mtcrfRZ~", "mtocrfRZ~", }, - [19] = { shift = 20, mask = 1, [0] = "mfcrR", "mfocrfRZ", }, - [371] = { shift = 11, mask = 1023, [392] = "mftbR", [424] = "mftbuR", }, - [339] = { - shift = 11, mask = 1023, - [32] = "mferR", [256] = "mflrR", [288] = "mfctrR", [16] = "mfspefscrR", - }, - [467] = { - shift = 11, mask = 1023, - [32] = "mtxerR", [256] = "mtlrR", [288] = "mtctrR", [16] = "mtspefscrR", - }, - - [20] = "lwarxRR0R", [84] = "ldarxRR0R", - - [21] = "ldxRR0R", [53] = "lduxRRR", - [149] = "stdxRR0R", [181] = "stduxRRR", - [341] = "lwaxRR0R", [373] = "lwauxRRR", - - [23] = "lwzxRR0R", [55] = "lwzuxRRR", - [87] = "lbzxRR0R", [119] = "lbzuxRRR", - [151] = "stwxRR0R", [183] = "stwuxRRR", - [215] = "stbxRR0R", [247] = "stbuxRRR", - [279] = "lhzxRR0R", [311] = "lhzuxRRR", - [343] = "lhaxRR0R", [375] = "lhauxRRR", - [407] = "sthxRR0R", [439] = "sthuxRRR", - - [54] = "dcbst-R0R", [86] = "dcbf-R0R", - [150] = "stwcxRR0R.", [214] = "stdcxRR0R.", - [246] = "dcbtst-R0R", [278] = "dcbt-R0R", - [310] = "eciwxRR0R", [438] = "ecowxRR0R", - [470] = "dcbi-RR", - - [598] = { - shift = 21, mask = 3, - [0] = "sync", "lwsync", "ptesync", - }, - [758] = "dcba-RR", - [854] = "eieio", [982] = "icbi-R0R", [1014] = "dcbz-R0R", - - [26] = "cntlzwRR~", [58] = "cntlzdRR~", - [122] = "popcntbRR~", - [154] = "prtywRR~", [186] = "prtydRR~", - - [28] = "andRR~R.", [60] = "andcRR~R.", [124] = "nor|notRR~R=.", - [284] = "eqvRR~R.", [316] = "xorRR~R.", - [412] = "orcRR~R.", [444] = "or|mrRR~R=.", [476] = "nandRR~R.", - [508] = "cmpbRR~R", - - [512] = "mcrxrX", - - [532] = "ldbrxRR0R", [660] = "stdbrxRR0R", - - [533] = "lswxRR0R", [597] = "lswiRR0A", - [661] = "stswxRR0R", [725] = "stswiRR0A", - - [534] = "lwbrxRR0R", [662] = "stwbrxRR0R", - [790] = "lhbrxRR0R", [918] = "sthbrxRR0R", - - [535] = "lfsxFR0R", [567] = "lfsuxFRR", - [599] = "lfdxFR0R", [631] = "lfduxFRR", - [663] = "stfsxFR0R", [695] = "stfsuxFRR", - [727] = "stfdxFR0R", [759] = "stfduxFR0R", - [855] = "lfiwaxFR0R", - [983] = "stfiwxFR0R", - - [24] = "slwRR~R.", - - [27] = "sldRR~R.", [536] = "srwRR~R.", - [792] = "srawRR~R.", [824] = "srawiRR~A.", - - [794] = "sradRR~R.", [826] = "sradiRR~H.", [827] = "sradiRR~H.", - [922] = "extshRR~.", [954] = "extsbRR~.", [986] = "extswRR~.", - - [539] = "srdRR~R.", -}, -{ __index = function(t, x) - if band(x, 31) == 15 then return "iselRRRC" end - end -}) - -local map_ld = { - shift = 0, mask = 3, - [0] = "ldRRE", "lduRRE", "lwaRRE", -} - -local map_std = { - shift = 0, mask = 3, - [0] = "stdRRE", "stduRRE", -} - -local map_fps = { - shift = 5, mask = 1, - { - shift = 1, mask = 15, - [0] = false, false, "fdivsFFF.", false, - "fsubsFFF.", "faddsFFF.", "fsqrtsF-F.", false, - "fresF-F.", "fmulsFF-F.", "frsqrtesF-F.", false, - "fmsubsFFFF~.", "fmaddsFFFF~.", "fnmsubsFFFF~.", "fnmaddsFFFF~.", - } -} - -local map_fpd = { - shift = 5, mask = 1, - [0] = { - shift = 1, mask = 1023, - [0] = "fcmpuXFF", [32] = "fcmpoXFF", [64] = "mcrfsXX", - [38] = "mtfsb1A.", [70] = "mtfsb0A.", [134] = "mtfsfiA>>-A>", - [8] = "fcpsgnFFF.", [40] = "fnegF-F.", [72] = "fmrF-F.", - [136] = "fnabsF-F.", [264] = "fabsF-F.", - [12] = "frspF-F.", - [14] = "fctiwF-F.", [15] = "fctiwzF-F.", - [583] = "mffsF.", [711] = "mtfsfZF.", - [392] = "frinF-F.", [424] = "frizF-F.", - [456] = "fripF-F.", [488] = "frimF-F.", - [814] = "fctidF-F.", [815] = "fctidzF-F.", [846] = "fcfidF-F.", - }, - { - shift = 1, mask = 15, - [0] = false, false, "fdivFFF.", false, - "fsubFFF.", "faddFFF.", "fsqrtF-F.", "fselFFFF~.", - "freF-F.", "fmulFF-F.", "frsqrteF-F.", false, - "fmsubFFFF~.", "fmaddFFFF~.", "fnmsubFFFF~.", "fnmaddFFFF~.", - } -} - -local map_spe = { - shift = 0, mask = 2047, - - [512] = "evaddwRRR", [514] = "evaddiwRAR~", - [516] = "evsubwRRR~", [518] = "evsubiwRAR~", - [520] = "evabsRR", [521] = "evnegRR", - [522] = "evextsbRR", [523] = "evextshRR", [524] = "evrndwRR", - [525] = "evcntlzwRR", [526] = "evcntlswRR", - - [527] = "brincRRR", - - [529] = "evandRRR", [530] = "evandcRRR", [534] = "evxorRRR", - [535] = "evor|evmrRRR=", [536] = "evnor|evnotRRR=", - [537] = "eveqvRRR", [539] = "evorcRRR", [542] = "evnandRRR", - - [544] = "evsrwuRRR", [545] = "evsrwsRRR", - [546] = "evsrwiuRRA", [547] = "evsrwisRRA", - [548] = "evslwRRR", [550] = "evslwiRRA", - [552] = "evrlwRRR", [553] = "evsplatiRS", - [554] = "evrlwiRRA", [555] = "evsplatfiRS", - [556] = "evmergehiRRR", [557] = "evmergeloRRR", - [558] = "evmergehiloRRR", [559] = "evmergelohiRRR", - - [560] = "evcmpgtuYRR", [561] = "evcmpgtsYRR", - [562] = "evcmpltuYRR", [563] = "evcmpltsYRR", - [564] = "evcmpeqYRR", - - [632] = "evselRRR", [633] = "evselRRRW", - [634] = "evselRRRW", [635] = "evselRRRW", - [636] = "evselRRRW", [637] = "evselRRRW", - [638] = "evselRRRW", [639] = "evselRRRW", - - [640] = "evfsaddRRR", [641] = "evfssubRRR", - [644] = "evfsabsRR", [645] = "evfsnabsRR", [646] = "evfsnegRR", - [648] = "evfsmulRRR", [649] = "evfsdivRRR", - [652] = "evfscmpgtYRR", [653] = "evfscmpltYRR", [654] = "evfscmpeqYRR", - [656] = "evfscfuiR-R", [657] = "evfscfsiR-R", - [658] = "evfscfufR-R", [659] = "evfscfsfR-R", - [660] = "evfsctuiR-R", [661] = "evfsctsiR-R", - [662] = "evfsctufR-R", [663] = "evfsctsfR-R", - [664] = "evfsctuizR-R", [666] = "evfsctsizR-R", - [668] = "evfststgtYRR", [669] = "evfststltYRR", [670] = "evfststeqYRR", - - [704] = "efsaddRRR", [705] = "efssubRRR", - [708] = "efsabsRR", [709] = "efsnabsRR", [710] = "efsnegRR", - [712] = "efsmulRRR", [713] = "efsdivRRR", - [716] = "efscmpgtYRR", [717] = "efscmpltYRR", [718] = "efscmpeqYRR", - [719] = "efscfdR-R", - [720] = "efscfuiR-R", [721] = "efscfsiR-R", - [722] = "efscfufR-R", [723] = "efscfsfR-R", - [724] = "efsctuiR-R", [725] = "efsctsiR-R", - [726] = "efsctufR-R", [727] = "efsctsfR-R", - [728] = "efsctuizR-R", [730] = "efsctsizR-R", - [732] = "efststgtYRR", [733] = "efststltYRR", [734] = "efststeqYRR", - - [736] = "efdaddRRR", [737] = "efdsubRRR", - [738] = "efdcfuidR-R", [739] = "efdcfsidR-R", - [740] = "efdabsRR", [741] = "efdnabsRR", [742] = "efdnegRR", - [744] = "efdmulRRR", [745] = "efddivRRR", - [746] = "efdctuidzR-R", [747] = "efdctsidzR-R", - [748] = "efdcmpgtYRR", [749] = "efdcmpltYRR", [750] = "efdcmpeqYRR", - [751] = "efdcfsR-R", - [752] = "efdcfuiR-R", [753] = "efdcfsiR-R", - [754] = "efdcfufR-R", [755] = "efdcfsfR-R", - [756] = "efdctuiR-R", [757] = "efdctsiR-R", - [758] = "efdctufR-R", [759] = "efdctsfR-R", - [760] = "efdctuizR-R", [762] = "efdctsizR-R", - [764] = "efdtstgtYRR", [765] = "efdtstltYRR", [766] = "efdtsteqYRR", - - [768] = "evlddxRR0R", [769] = "evlddRR8", - [770] = "evldwxRR0R", [771] = "evldwRR8", - [772] = "evldhxRR0R", [773] = "evldhRR8", - [776] = "evlhhesplatxRR0R", [777] = "evlhhesplatRR2", - [780] = "evlhhousplatxRR0R", [781] = "evlhhousplatRR2", - [782] = "evlhhossplatxRR0R", [783] = "evlhhossplatRR2", - [784] = "evlwhexRR0R", [785] = "evlwheRR4", - [788] = "evlwhouxRR0R", [789] = "evlwhouRR4", - [790] = "evlwhosxRR0R", [791] = "evlwhosRR4", - [792] = "evlwwsplatxRR0R", [793] = "evlwwsplatRR4", - [796] = "evlwhsplatxRR0R", [797] = "evlwhsplatRR4", - - [800] = "evstddxRR0R", [801] = "evstddRR8", - [802] = "evstdwxRR0R", [803] = "evstdwRR8", - [804] = "evstdhxRR0R", [805] = "evstdhRR8", - [816] = "evstwhexRR0R", [817] = "evstwheRR4", - [820] = "evstwhoxRR0R", [821] = "evstwhoRR4", - [824] = "evstwwexRR0R", [825] = "evstwweRR4", - [828] = "evstwwoxRR0R", [829] = "evstwwoRR4", - - [1027] = "evmhessfRRR", [1031] = "evmhossfRRR", [1032] = "evmheumiRRR", - [1033] = "evmhesmiRRR", [1035] = "evmhesmfRRR", [1036] = "evmhoumiRRR", - [1037] = "evmhosmiRRR", [1039] = "evmhosmfRRR", [1059] = "evmhessfaRRR", - [1063] = "evmhossfaRRR", [1064] = "evmheumiaRRR", [1065] = "evmhesmiaRRR", - [1067] = "evmhesmfaRRR", [1068] = "evmhoumiaRRR", [1069] = "evmhosmiaRRR", - [1071] = "evmhosmfaRRR", [1095] = "evmwhssfRRR", [1096] = "evmwlumiRRR", - [1100] = "evmwhumiRRR", [1101] = "evmwhsmiRRR", [1103] = "evmwhsmfRRR", - [1107] = "evmwssfRRR", [1112] = "evmwumiRRR", [1113] = "evmwsmiRRR", - [1115] = "evmwsmfRRR", [1127] = "evmwhssfaRRR", [1128] = "evmwlumiaRRR", - [1132] = "evmwhumiaRRR", [1133] = "evmwhsmiaRRR", [1135] = "evmwhsmfaRRR", - [1139] = "evmwssfaRRR", [1144] = "evmwumiaRRR", [1145] = "evmwsmiaRRR", - [1147] = "evmwsmfaRRR", - - [1216] = "evaddusiaawRR", [1217] = "evaddssiaawRR", - [1218] = "evsubfusiaawRR", [1219] = "evsubfssiaawRR", - [1220] = "evmraRR", - [1222] = "evdivwsRRR", [1223] = "evdivwuRRR", - [1224] = "evaddumiaawRR", [1225] = "evaddsmiaawRR", - [1226] = "evsubfumiaawRR", [1227] = "evsubfsmiaawRR", - - [1280] = "evmheusiaawRRR", [1281] = "evmhessiaawRRR", - [1283] = "evmhessfaawRRR", [1284] = "evmhousiaawRRR", - [1285] = "evmhossiaawRRR", [1287] = "evmhossfaawRRR", - [1288] = "evmheumiaawRRR", [1289] = "evmhesmiaawRRR", - [1291] = "evmhesmfaawRRR", [1292] = "evmhoumiaawRRR", - [1293] = "evmhosmiaawRRR", [1295] = "evmhosmfaawRRR", - [1320] = "evmhegumiaaRRR", [1321] = "evmhegsmiaaRRR", - [1323] = "evmhegsmfaaRRR", [1324] = "evmhogumiaaRRR", - [1325] = "evmhogsmiaaRRR", [1327] = "evmhogsmfaaRRR", - [1344] = "evmwlusiaawRRR", [1345] = "evmwlssiaawRRR", - [1352] = "evmwlumiaawRRR", [1353] = "evmwlsmiaawRRR", - [1363] = "evmwssfaaRRR", [1368] = "evmwumiaaRRR", - [1369] = "evmwsmiaaRRR", [1371] = "evmwsmfaaRRR", - [1408] = "evmheusianwRRR", [1409] = "evmhessianwRRR", - [1411] = "evmhessfanwRRR", [1412] = "evmhousianwRRR", - [1413] = "evmhossianwRRR", [1415] = "evmhossfanwRRR", - [1416] = "evmheumianwRRR", [1417] = "evmhesmianwRRR", - [1419] = "evmhesmfanwRRR", [1420] = "evmhoumianwRRR", - [1421] = "evmhosmianwRRR", [1423] = "evmhosmfanwRRR", - [1448] = "evmhegumianRRR", [1449] = "evmhegsmianRRR", - [1451] = "evmhegsmfanRRR", [1452] = "evmhogumianRRR", - [1453] = "evmhogsmianRRR", [1455] = "evmhogsmfanRRR", - [1472] = "evmwlusianwRRR", [1473] = "evmwlssianwRRR", - [1480] = "evmwlumianwRRR", [1481] = "evmwlsmianwRRR", - [1491] = "evmwssfanRRR", [1496] = "evmwumianRRR", - [1497] = "evmwsmianRRR", [1499] = "evmwsmfanRRR", -} - -local map_pri = { - [0] = false, false, "tdiARI", "twiARI", - map_spe, false, false, "mulliRRI", - "subficRRI", false, "cmpl_iYLRU", "cmp_iYLRI", - "addicRRI", "addic.RRI", "addi|liRR0I", "addis|lisRR0I", - "b_KBJ", "sc", "bKJ", map_crops, - "rlwimiRR~AAA.", map_rlwinm, false, "rlwnmRR~RAA.", - "oriNRR~U", "orisRR~U", "xoriRR~U", "xorisRR~U", - "andi.RR~U", "andis.RR~U", map_rld, map_ext, - "lwzRRD", "lwzuRRD", "lbzRRD", "lbzuRRD", - "stwRRD", "stwuRRD", "stbRRD", "stbuRRD", - "lhzRRD", "lhzuRRD", "lhaRRD", "lhauRRD", - "sthRRD", "sthuRRD", "lmwRRD", "stmwRRD", - "lfsFRD", "lfsuFRD", "lfdFRD", "lfduFRD", - "stfsFRD", "stfsuFRD", "stfdFRD", "stfduFRD", - false, false, map_ld, map_fps, - false, false, map_std, map_fpd, -} - ------------------------------------------------------------------------------- - -local map_gpr = { - [0] = "r0", "sp", "r2", "r3", "r4", "r5", "r6", "r7", - "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15", - "r16", "r17", "r18", "r19", "r20", "r21", "r22", "r23", - "r24", "r25", "r26", "r27", "r28", "r29", "r30", "r31", -} - -local map_cond = { [0] = "lt", "gt", "eq", "so", "ge", "le", "ne", "ns", } - --- Format a condition bit. -local function condfmt(cond) - if cond <= 3 then - return map_cond[band(cond, 3)] - else - return format("4*cr%d+%s", rshift(cond, 2), map_cond[band(cond, 3)]) - end -end - ------------------------------------------------------------------------------- - --- Output a nicely formatted line with an opcode and operands. -local function putop(ctx, text, operands) - local pos = ctx.pos - local extra = "" - if ctx.rel then - local sym = ctx.symtab[ctx.rel] - if sym then extra = "\t->"..sym end - end - if ctx.hexdump > 0 then - ctx.out(format("%08x %s %-7s %s%s\n", - ctx.addr+pos, tohex(ctx.op), text, concat(operands, ", "), extra)) - else - ctx.out(format("%08x %-7s %s%s\n", - ctx.addr+pos, text, concat(operands, ", "), extra)) - end - ctx.pos = pos + 4 -end - --- Fallback for unknown opcodes. -local function unknown(ctx) - return putop(ctx, ".long", { "0x"..tohex(ctx.op) }) -end - --- Disassemble a single instruction. -local function disass_ins(ctx) - local pos = ctx.pos - local b0, b1, b2, b3 = byte(ctx.code, pos+1, pos+4) - local op = bor(lshift(b0, 24), lshift(b1, 16), lshift(b2, 8), b3) - local operands = {} - local last = nil - local rs = 21 - ctx.op = op - ctx.rel = nil - - local opat = map_pri[rshift(b0, 2)] - while type(opat) ~= "string" do - if not opat then return unknown(ctx) end - opat = opat[band(rshift(op, opat.shift), opat.mask)] - end - local name, pat = match(opat, "^([a-z0-9_.]*)(.*)") - local altname, pat2 = match(pat, "|([a-z0-9_.]*)(.*)") - if altname then pat = pat2 end - - for p in gmatch(pat, ".") do - local x = nil - if p == "R" then - x = map_gpr[band(rshift(op, rs), 31)] - rs = rs - 5 - elseif p == "F" then - x = "f"..band(rshift(op, rs), 31) - rs = rs - 5 - elseif p == "A" then - x = band(rshift(op, rs), 31) - rs = rs - 5 - elseif p == "S" then - x = arshift(lshift(op, 27-rs), 27) - rs = rs - 5 - elseif p == "I" then - x = arshift(lshift(op, 16), 16) - elseif p == "U" then - x = band(op, 0xffff) - elseif p == "D" or p == "E" then - local disp = arshift(lshift(op, 16), 16) - if p == "E" then disp = band(disp, -4) end - if last == "r0" then last = "0" end - operands[#operands] = format("%d(%s)", disp, last) - elseif p >= "2" and p <= "8" then - local disp = band(rshift(op, rs), 31) * p - if last == "r0" then last = "0" end - operands[#operands] = format("%d(%s)", disp, last) - elseif p == "H" then - x = band(rshift(op, rs), 31) + lshift(band(op, 2), 4) - rs = rs - 5 - elseif p == "M" then - x = band(rshift(op, rs), 31) + band(op, 0x20) - elseif p == "C" then - x = condfmt(band(rshift(op, rs), 31)) - rs = rs - 5 - elseif p == "B" then - local bo = rshift(op, 21) - local cond = band(rshift(op, 16), 31) - local cn = "" - rs = rs - 10 - if band(bo, 4) == 0 then - cn = band(bo, 2) == 0 and "dnz" or "dz" - if band(bo, 0x10) == 0 then - cn = cn..(band(bo, 8) == 0 and "f" or "t") - end - if band(bo, 0x10) == 0 then x = condfmt(cond) end - name = name..(band(bo, 1) == band(rshift(op, 15), 1) and "-" or "+") - elseif band(bo, 0x10) == 0 then - cn = map_cond[band(cond, 3) + (band(bo, 8) == 0 and 4 or 0)] - if cond > 3 then x = "cr"..rshift(cond, 2) end - name = name..(band(bo, 1) == band(rshift(op, 15), 1) and "-" or "+") - end - name = gsub(name, "_", cn) - elseif p == "J" then - x = arshift(lshift(op, 27-rs), 29-rs)*4 - if band(op, 2) == 0 then x = ctx.addr + pos + x end - ctx.rel = x - x = "0x"..tohex(x) - elseif p == "K" then - if band(op, 1) ~= 0 then name = name.."l" end - if band(op, 2) ~= 0 then name = name.."a" end - elseif p == "X" or p == "Y" then - x = band(rshift(op, rs+2), 7) - if x == 0 and p == "Y" then x = nil else x = "cr"..x end - rs = rs - 5 - elseif p == "W" then - x = "cr"..band(op, 7) - elseif p == "Z" then - x = band(rshift(op, rs-4), 255) - rs = rs - 10 - elseif p == ">" then - operands[#operands] = rshift(operands[#operands], 1) - elseif p == "0" then - if last == "r0" then - operands[#operands] = nil - if altname then name = altname end - end - elseif p == "L" then - name = gsub(name, "_", band(op, 0x00200000) ~= 0 and "d" or "w") - elseif p == "." then - if band(op, 1) == 1 then name = name.."." end - elseif p == "N" then - if op == 0x60000000 then name = "nop"; break end - elseif p == "~" then - local n = #operands - operands[n-1], operands[n] = operands[n], operands[n-1] - elseif p == "=" then - local n = #operands - if last == operands[n-1] then - operands[n] = nil - name = altname - end - elseif p == "%" then - local n = #operands - if last == operands[n-1] and last == operands[n-2] then - operands[n] = nil - operands[n-1] = nil - name = altname - end - elseif p == "-" then - rs = rs - 5 - else - assert(false) - end - if x then operands[#operands+1] = x; last = x end - end - - return putop(ctx, name, operands) -end - ------------------------------------------------------------------------------- - --- Disassemble a block of code. -local function disass_block(ctx, ofs, len) - if not ofs then ofs = 0 end - local stop = len and ofs+len or #ctx.code - stop = stop - stop % 4 - ctx.pos = ofs - ofs % 4 - ctx.rel = nil - while ctx.pos < stop do disass_ins(ctx) end -end - --- Extended API: create a disassembler context. Then call ctx:disass(ofs, len). -local function create_(code, addr, out) - local ctx = {} - ctx.code = code - ctx.addr = addr or 0 - ctx.out = out or io.write - ctx.symtab = {} - ctx.disass = disass_block - ctx.hexdump = 8 - return ctx -end - --- Simple API: disassemble code (a string) at address and output via out. -local function disass_(code, addr, out) - create_(code, addr, out):disass() -end - --- Return register name for RID. -local function regname_(r) - if r < 32 then return map_gpr[r] end - return "f"..(r-32) -end - --- Public module functions. -module(...) - -create = create_ -disass = disass_ -regname = regname_ - diff --git a/lib/dis_x64.lua b/lib/dis_x64.lua deleted file mode 100644 index d95d7ded..00000000 --- a/lib/dis_x64.lua +++ /dev/null @@ -1,20 +0,0 @@ ----------------------------------------------------------------------------- --- LuaJIT x64 disassembler wrapper module. --- --- Copyright (C) 2005-2012 Mike Pall. All rights reserved. --- Released under the MIT license. See Copyright Notice in luajit.h ----------------------------------------------------------------------------- --- This module just exports the 64 bit functions from the combined --- x86/x64 disassembler module. All the interesting stuff is there. ------------------------------------------------------------------------------- - -local require = require - -module(...) - -local dis_x86 = require(_PACKAGE.."dis_x86") - -create = dis_x86.create64 -disass = dis_x86.disass64 -regname = dis_x86.regname64 - diff --git a/lib/dis_x86.lua b/lib/dis_x86.lua deleted file mode 100644 index 5aeeb449..00000000 --- a/lib/dis_x86.lua +++ /dev/null @@ -1,836 +0,0 @@ ----------------------------------------------------------------------------- --- LuaJIT x86/x64 disassembler module. --- --- Copyright (C) 2005-2012 Mike Pall. All rights reserved. --- Released under the MIT license. See Copyright Notice in luajit.h ----------------------------------------------------------------------------- --- This is a helper module used by the LuaJIT machine code dumper module. --- --- Sending small code snippets to an external disassembler and mixing the --- output with our own stuff was too fragile. So I had to bite the bullet --- and write yet another x86 disassembler. Oh well ... --- --- The output format is very similar to what ndisasm generates. But it has --- been developed independently by looking at the opcode tables from the --- Intel and AMD manuals. The supported instruction set is quite extensive --- and reflects what a current generation Intel or AMD CPU implements in --- 32 bit and 64 bit mode. Yes, this includes MMX, SSE, SSE2, SSE3, SSSE3, --- SSE4.1, SSE4.2, SSE4a and even privileged and hypervisor (VMX/SVM) --- instructions. --- --- Notes: --- * The (useless) a16 prefix, 3DNow and pre-586 opcodes are unsupported. --- * No attempt at optimization has been made -- it's fast enough for my needs. --- * The public API may change when more architectures are added. ------------------------------------------------------------------------------- - -local type = type -local sub, byte, format = string.sub, string.byte, string.format -local match, gmatch, gsub = string.match, string.gmatch, string.gsub -local lower, rep = string.lower, string.rep - --- Map for 1st opcode byte in 32 bit mode. Ugly? Well ... read on. -local map_opc1_32 = { ---0x -[0]="addBmr","addVmr","addBrm","addVrm","addBai","addVai","push es","pop es", -"orBmr","orVmr","orBrm","orVrm","orBai","orVai","push cs","opc2*", ---1x -"adcBmr","adcVmr","adcBrm","adcVrm","adcBai","adcVai","push ss","pop ss", -"sbbBmr","sbbVmr","sbbBrm","sbbVrm","sbbBai","sbbVai","push ds","pop ds", ---2x -"andBmr","andVmr","andBrm","andVrm","andBai","andVai","es:seg","daa", -"subBmr","subVmr","subBrm","subVrm","subBai","subVai","cs:seg","das", ---3x -"xorBmr","xorVmr","xorBrm","xorVrm","xorBai","xorVai","ss:seg","aaa", -"cmpBmr","cmpVmr","cmpBrm","cmpVrm","cmpBai","cmpVai","ds:seg","aas", ---4x -"incVR","incVR","incVR","incVR","incVR","incVR","incVR","incVR", -"decVR","decVR","decVR","decVR","decVR","decVR","decVR","decVR", ---5x -"pushUR","pushUR","pushUR","pushUR","pushUR","pushUR","pushUR","pushUR", -"popUR","popUR","popUR","popUR","popUR","popUR","popUR","popUR", ---6x -"sz*pushaw,pusha","sz*popaw,popa","boundVrm","arplWmr", -"fs:seg","gs:seg","o16:","a16", -"pushUi","imulVrmi","pushBs","imulVrms", -"insb","insVS","outsb","outsVS", ---7x -"joBj","jnoBj","jbBj","jnbBj","jzBj","jnzBj","jbeBj","jaBj", -"jsBj","jnsBj","jpeBj","jpoBj","jlBj","jgeBj","jleBj","jgBj", ---8x -"arith!Bmi","arith!Vmi","arith!Bmi","arith!Vms", -"testBmr","testVmr","xchgBrm","xchgVrm", -"movBmr","movVmr","movBrm","movVrm", -"movVmg","leaVrm","movWgm","popUm", ---9x -"nop*xchgVaR|pause|xchgWaR|repne nop","xchgVaR","xchgVaR","xchgVaR", -"xchgVaR","xchgVaR","xchgVaR","xchgVaR", -"sz*cbw,cwde,cdqe","sz*cwd,cdq,cqo","call farViw","wait", -"sz*pushfw,pushf","sz*popfw,popf","sahf","lahf", ---Ax -"movBao","movVao","movBoa","movVoa", -"movsb","movsVS","cmpsb","cmpsVS", -"testBai","testVai","stosb","stosVS", -"lodsb","lodsVS","scasb","scasVS", ---Bx -"movBRi","movBRi","movBRi","movBRi","movBRi","movBRi","movBRi","movBRi", -"movVRI","movVRI","movVRI","movVRI","movVRI","movVRI","movVRI","movVRI", ---Cx -"shift!Bmu","shift!Vmu","retBw","ret","$lesVrm","$ldsVrm","movBmi","movVmi", -"enterBwu","leave","retfBw","retf","int3","intBu","into","iretVS", ---Dx -"shift!Bm1","shift!Vm1","shift!Bmc","shift!Vmc","aamBu","aadBu","salc","xlatb", -"fp*0","fp*1","fp*2","fp*3","fp*4","fp*5","fp*6","fp*7", ---Ex -"loopneBj","loopeBj","loopBj","sz*jcxzBj,jecxzBj,jrcxzBj", -"inBau","inVau","outBua","outVua", -"callVj","jmpVj","jmp farViw","jmpBj","inBad","inVad","outBda","outVda", ---Fx -"lock:","int1","repne:rep","rep:","hlt","cmc","testb!Bm","testv!Vm", -"clc","stc","cli","sti","cld","std","incb!Bm","incd!Vm", -} -assert(#map_opc1_32 == 255) - --- Map for 1st opcode byte in 64 bit mode (overrides only). -local map_opc1_64 = setmetatable({ - [0x06]=false, [0x07]=false, [0x0e]=false, - [0x16]=false, [0x17]=false, [0x1e]=false, [0x1f]=false, - [0x27]=false, [0x2f]=false, [0x37]=false, [0x3f]=false, - [0x60]=false, [0x61]=false, [0x62]=false, [0x63]="movsxdVrDmt", [0x67]="a32:", - [0x40]="rex*", [0x41]="rex*b", [0x42]="rex*x", [0x43]="rex*xb", - [0x44]="rex*r", [0x45]="rex*rb", [0x46]="rex*rx", [0x47]="rex*rxb", - [0x48]="rex*w", [0x49]="rex*wb", [0x4a]="rex*wx", [0x4b]="rex*wxb", - [0x4c]="rex*wr", [0x4d]="rex*wrb", [0x4e]="rex*wrx", [0x4f]="rex*wrxb", - [0x82]=false, [0x9a]=false, [0xc4]=false, [0xc5]=false, [0xce]=false, - [0xd4]=false, [0xd5]=false, [0xd6]=false, [0xea]=false, -}, { __index = map_opc1_32 }) - --- Map for 2nd opcode byte (0F xx). True CISC hell. Hey, I told you. --- Prefix dependent MMX/SSE opcodes: (none)|rep|o16|repne, -|F3|66|F2 -local map_opc2 = { ---0x -[0]="sldt!Dmp","sgdt!Ump","larVrm","lslVrm",nil,"syscall","clts","sysret", -"invd","wbinvd",nil,"ud1",nil,"$prefetch!Bm","femms","3dnowMrmu", ---1x -"movupsXrm|movssXrm|movupdXrm|movsdXrm", -"movupsXmr|movssXmr|movupdXmr|movsdXmr", -"movhlpsXrm$movlpsXrm|movsldupXrm|movlpdXrm|movddupXrm", -"movlpsXmr||movlpdXmr", -"unpcklpsXrm||unpcklpdXrm", -"unpckhpsXrm||unpckhpdXrm", -"movlhpsXrm$movhpsXrm|movshdupXrm|movhpdXrm", -"movhpsXmr||movhpdXmr", -"$prefetcht!Bm","hintnopVm","hintnopVm","hintnopVm", -"hintnopVm","hintnopVm","hintnopVm","hintnopVm", ---2x -"movUmx$","movUmy$","movUxm$","movUym$","movUmz$",nil,"movUzm$",nil, -"movapsXrm||movapdXrm", -"movapsXmr||movapdXmr", -"cvtpi2psXrMm|cvtsi2ssXrVmt|cvtpi2pdXrMm|cvtsi2sdXrVmt", -"movntpsXmr|movntssXmr|movntpdXmr|movntsdXmr", -"cvttps2piMrXm|cvttss2siVrXm|cvttpd2piMrXm|cvttsd2siVrXm", -"cvtps2piMrXm|cvtss2siVrXm|cvtpd2piMrXm|cvtsd2siVrXm", -"ucomissXrm||ucomisdXrm", -"comissXrm||comisdXrm", ---3x -"wrmsr","rdtsc","rdmsr","rdpmc","sysenter","sysexit",nil,"getsec", -"opc3*38",nil,"opc3*3a",nil,nil,nil,nil,nil, ---4x -"cmovoVrm","cmovnoVrm","cmovbVrm","cmovnbVrm", -"cmovzVrm","cmovnzVrm","cmovbeVrm","cmovaVrm", -"cmovsVrm","cmovnsVrm","cmovpeVrm","cmovpoVrm", -"cmovlVrm","cmovgeVrm","cmovleVrm","cmovgVrm", ---5x -"movmskpsVrXm$||movmskpdVrXm$","sqrtpsXrm|sqrtssXrm|sqrtpdXrm|sqrtsdXrm", -"rsqrtpsXrm|rsqrtssXrm","rcppsXrm|rcpssXrm", -"andpsXrm||andpdXrm","andnpsXrm||andnpdXrm", -"orpsXrm||orpdXrm","xorpsXrm||xorpdXrm", -"addpsXrm|addssXrm|addpdXrm|addsdXrm","mulpsXrm|mulssXrm|mulpdXrm|mulsdXrm", -"cvtps2pdXrm|cvtss2sdXrm|cvtpd2psXrm|cvtsd2ssXrm", -"cvtdq2psXrm|cvttps2dqXrm|cvtps2dqXrm", -"subpsXrm|subssXrm|subpdXrm|subsdXrm","minpsXrm|minssXrm|minpdXrm|minsdXrm", -"divpsXrm|divssXrm|divpdXrm|divsdXrm","maxpsXrm|maxssXrm|maxpdXrm|maxsdXrm", ---6x -"punpcklbwPrm","punpcklwdPrm","punpckldqPrm","packsswbPrm", -"pcmpgtbPrm","pcmpgtwPrm","pcmpgtdPrm","packuswbPrm", -"punpckhbwPrm","punpckhwdPrm","punpckhdqPrm","packssdwPrm", -"||punpcklqdqXrm","||punpckhqdqXrm", -"movPrVSm","movqMrm|movdquXrm|movdqaXrm", ---7x -"pshufwMrmu|pshufhwXrmu|pshufdXrmu|pshuflwXrmu","pshiftw!Pmu", -"pshiftd!Pmu","pshiftq!Mmu||pshiftdq!Xmu", -"pcmpeqbPrm","pcmpeqwPrm","pcmpeqdPrm","emms|", -"vmreadUmr||extrqXmuu$|insertqXrmuu$","vmwriteUrm||extrqXrm$|insertqXrm$", -nil,nil, -"||haddpdXrm|haddpsXrm","||hsubpdXrm|hsubpsXrm", -"movVSmMr|movqXrm|movVSmXr","movqMmr|movdquXmr|movdqaXmr", ---8x -"joVj","jnoVj","jbVj","jnbVj","jzVj","jnzVj","jbeVj","jaVj", -"jsVj","jnsVj","jpeVj","jpoVj","jlVj","jgeVj","jleVj","jgVj", ---9x -"setoBm","setnoBm","setbBm","setnbBm","setzBm","setnzBm","setbeBm","setaBm", -"setsBm","setnsBm","setpeBm","setpoBm","setlBm","setgeBm","setleBm","setgBm", ---Ax -"push fs","pop fs","cpuid","btVmr","shldVmru","shldVmrc",nil,nil, -"push gs","pop gs","rsm","btsVmr","shrdVmru","shrdVmrc","fxsave!Dmp","imulVrm", ---Bx -"cmpxchgBmr","cmpxchgVmr","$lssVrm","btrVmr", -"$lfsVrm","$lgsVrm","movzxVrBmt","movzxVrWmt", -"|popcntVrm","ud2Dp","bt!Vmu","btcVmr", -"bsfVrm","bsrVrm|lzcntVrm|bsrWrm","movsxVrBmt","movsxVrWmt", ---Cx -"xaddBmr","xaddVmr", -"cmppsXrmu|cmpssXrmu|cmppdXrmu|cmpsdXrmu","$movntiVmr|", -"pinsrwPrWmu","pextrwDrPmu", -"shufpsXrmu||shufpdXrmu","$cmpxchg!Qmp", -"bswapVR","bswapVR","bswapVR","bswapVR","bswapVR","bswapVR","bswapVR","bswapVR", ---Dx -"||addsubpdXrm|addsubpsXrm","psrlwPrm","psrldPrm","psrlqPrm", -"paddqPrm","pmullwPrm", -"|movq2dqXrMm|movqXmr|movdq2qMrXm$","pmovmskbVrMm||pmovmskbVrXm", -"psubusbPrm","psubuswPrm","pminubPrm","pandPrm", -"paddusbPrm","padduswPrm","pmaxubPrm","pandnPrm", ---Ex -"pavgbPrm","psrawPrm","psradPrm","pavgwPrm", -"pmulhuwPrm","pmulhwPrm", -"|cvtdq2pdXrm|cvttpd2dqXrm|cvtpd2dqXrm","$movntqMmr||$movntdqXmr", -"psubsbPrm","psubswPrm","pminswPrm","porPrm", -"paddsbPrm","paddswPrm","pmaxswPrm","pxorPrm", ---Fx -"|||lddquXrm","psllwPrm","pslldPrm","psllqPrm", -"pmuludqPrm","pmaddwdPrm","psadbwPrm","maskmovqMrm||maskmovdquXrm$", -"psubbPrm","psubwPrm","psubdPrm","psubqPrm", -"paddbPrm","paddwPrm","padddPrm","ud", -} -assert(map_opc2[255] == "ud") - --- Map for three-byte opcodes. Can't wait for their next invention. -local map_opc3 = { -["38"] = { -- [66] 0f 38 xx ---0x -[0]="pshufbPrm","phaddwPrm","phadddPrm","phaddswPrm", -"pmaddubswPrm","phsubwPrm","phsubdPrm","phsubswPrm", -"psignbPrm","psignwPrm","psigndPrm","pmulhrswPrm", -nil,nil,nil,nil, ---1x -"||pblendvbXrma",nil,nil,nil, -"||blendvpsXrma","||blendvpdXrma",nil,"||ptestXrm", -nil,nil,nil,nil, -"pabsbPrm","pabswPrm","pabsdPrm",nil, ---2x -"||pmovsxbwXrm","||pmovsxbdXrm","||pmovsxbqXrm","||pmovsxwdXrm", -"||pmovsxwqXrm","||pmovsxdqXrm",nil,nil, -"||pmuldqXrm","||pcmpeqqXrm","||$movntdqaXrm","||packusdwXrm", -nil,nil,nil,nil, ---3x -"||pmovzxbwXrm","||pmovzxbdXrm","||pmovzxbqXrm","||pmovzxwdXrm", -"||pmovzxwqXrm","||pmovzxdqXrm",nil,"||pcmpgtqXrm", -"||pminsbXrm","||pminsdXrm","||pminuwXrm","||pminudXrm", -"||pmaxsbXrm","||pmaxsdXrm","||pmaxuwXrm","||pmaxudXrm", ---4x -"||pmulddXrm","||phminposuwXrm", ---Fx -[0xf0] = "|||crc32TrBmt",[0xf1] = "|||crc32TrVmt", -}, - -["3a"] = { -- [66] 0f 3a xx ---0x -[0x00]=nil,nil,nil,nil,nil,nil,nil,nil, -"||roundpsXrmu","||roundpdXrmu","||roundssXrmu","||roundsdXrmu", -"||blendpsXrmu","||blendpdXrmu","||pblendwXrmu","palignrPrmu", ---1x -nil,nil,nil,nil, -"||pextrbVmXru","||pextrwVmXru","||pextrVmSXru","||extractpsVmXru", -nil,nil,nil,nil,nil,nil,nil,nil, ---2x -"||pinsrbXrVmu","||insertpsXrmu","||pinsrXrVmuS",nil, ---4x -[0x40] = "||dppsXrmu", -[0x41] = "||dppdXrmu", -[0x42] = "||mpsadbwXrmu", ---6x -[0x60] = "||pcmpestrmXrmu",[0x61] = "||pcmpestriXrmu", -[0x62] = "||pcmpistrmXrmu",[0x63] = "||pcmpistriXrmu", -}, -} - --- Map for VMX/SVM opcodes 0F 01 C0-FF (sgdt group with register operands). -local map_opcvm = { -[0xc1]="vmcall",[0xc2]="vmlaunch",[0xc3]="vmresume",[0xc4]="vmxoff", -[0xc8]="monitor",[0xc9]="mwait", -[0xd8]="vmrun",[0xd9]="vmmcall",[0xda]="vmload",[0xdb]="vmsave", -[0xdc]="stgi",[0xdd]="clgi",[0xde]="skinit",[0xdf]="invlpga", -[0xf8]="swapgs",[0xf9]="rdtscp", -} - --- Map for FP opcodes. And you thought stack machines are simple? -local map_opcfp = { --- D8-DF 00-BF: opcodes with a memory operand. --- D8 -[0]="faddFm","fmulFm","fcomFm","fcompFm","fsubFm","fsubrFm","fdivFm","fdivrFm", -"fldFm",nil,"fstFm","fstpFm","fldenvVm","fldcwWm","fnstenvVm","fnstcwWm", --- DA -"fiaddDm","fimulDm","ficomDm","ficompDm", -"fisubDm","fisubrDm","fidivDm","fidivrDm", --- DB -"fildDm","fisttpDm","fistDm","fistpDm",nil,"fld twordFmp",nil,"fstp twordFmp", --- DC -"faddGm","fmulGm","fcomGm","fcompGm","fsubGm","fsubrGm","fdivGm","fdivrGm", --- DD -"fldGm","fisttpQm","fstGm","fstpGm","frstorDmp",nil,"fnsaveDmp","fnstswWm", --- DE -"fiaddWm","fimulWm","ficomWm","ficompWm", -"fisubWm","fisubrWm","fidivWm","fidivrWm", --- DF -"fildWm","fisttpWm","fistWm","fistpWm", -"fbld twordFmp","fildQm","fbstp twordFmp","fistpQm", --- xx C0-FF: opcodes with a pseudo-register operand. --- D8 -"faddFf","fmulFf","fcomFf","fcompFf","fsubFf","fsubrFf","fdivFf","fdivrFf", --- D9 -"fldFf","fxchFf",{"fnop"},nil, -{"fchs","fabs",nil,nil,"ftst","fxam"}, -{"fld1","fldl2t","fldl2e","fldpi","fldlg2","fldln2","fldz"}, -{"f2xm1","fyl2x","fptan","fpatan","fxtract","fprem1","fdecstp","fincstp"}, -{"fprem","fyl2xp1","fsqrt","fsincos","frndint","fscale","fsin","fcos"}, --- DA -"fcmovbFf","fcmoveFf","fcmovbeFf","fcmovuFf",nil,{nil,"fucompp"},nil,nil, --- DB -"fcmovnbFf","fcmovneFf","fcmovnbeFf","fcmovnuFf", -{nil,nil,"fnclex","fninit"},"fucomiFf","fcomiFf",nil, --- DC -"fadd toFf","fmul toFf",nil,nil, -"fsub toFf","fsubr toFf","fdivr toFf","fdiv toFf", --- DD -"ffreeFf",nil,"fstFf","fstpFf","fucomFf","fucompFf",nil,nil, --- DE -"faddpFf","fmulpFf",nil,{nil,"fcompp"}, -"fsubrpFf","fsubpFf","fdivrpFf","fdivpFf", --- DF -nil,nil,nil,nil,{"fnstsw ax"},"fucomipFf","fcomipFf",nil, -} -assert(map_opcfp[126] == "fcomipFf") - --- Map for opcode groups. The subkey is sp from the ModRM byte. -local map_opcgroup = { - arith = { "add", "or", "adc", "sbb", "and", "sub", "xor", "cmp" }, - shift = { "rol", "ror", "rcl", "rcr", "shl", "shr", "sal", "sar" }, - testb = { "testBmi", "testBmi", "not", "neg", "mul", "imul", "div", "idiv" }, - testv = { "testVmi", "testVmi", "not", "neg", "mul", "imul", "div", "idiv" }, - incb = { "inc", "dec" }, - incd = { "inc", "dec", "callUmp", "$call farDmp", - "jmpUmp", "$jmp farDmp", "pushUm" }, - sldt = { "sldt", "str", "lldt", "ltr", "verr", "verw" }, - sgdt = { "vm*$sgdt", "vm*$sidt", "$lgdt", "vm*$lidt", - "smsw", nil, "lmsw", "vm*$invlpg" }, - bt = { nil, nil, nil, nil, "bt", "bts", "btr", "btc" }, - cmpxchg = { nil, "sz*,cmpxchg8bQmp,cmpxchg16bXmp", nil, nil, - nil, nil, "vmptrld|vmxon|vmclear", "vmptrst" }, - pshiftw = { nil, nil, "psrlw", nil, "psraw", nil, "psllw" }, - pshiftd = { nil, nil, "psrld", nil, "psrad", nil, "pslld" }, - pshiftq = { nil, nil, "psrlq", nil, nil, nil, "psllq" }, - pshiftdq = { nil, nil, "psrlq", "psrldq", nil, nil, "psllq", "pslldq" }, - fxsave = { "$fxsave", "$fxrstor", "$ldmxcsr", "$stmxcsr", - nil, "lfenceDp$", "mfenceDp$", "sfenceDp$clflush" }, - prefetch = { "prefetch", "prefetchw" }, - prefetcht = { "prefetchnta", "prefetcht0", "prefetcht1", "prefetcht2" }, -} - ------------------------------------------------------------------------------- - --- Maps for register names. -local map_regs = { - B = { "al", "cl", "dl", "bl", "ah", "ch", "dh", "bh", - "r8b", "r9b", "r10b", "r11b", "r12b", "r13b", "r14b", "r15b" }, - B64 = { "al", "cl", "dl", "bl", "spl", "bpl", "sil", "dil", - "r8b", "r9b", "r10b", "r11b", "r12b", "r13b", "r14b", "r15b" }, - W = { "ax", "cx", "dx", "bx", "sp", "bp", "si", "di", - "r8w", "r9w", "r10w", "r11w", "r12w", "r13w", "r14w", "r15w" }, - D = { "eax", "ecx", "edx", "ebx", "esp", "ebp", "esi", "edi", - "r8d", "r9d", "r10d", "r11d", "r12d", "r13d", "r14d", "r15d" }, - Q = { "rax", "rcx", "rdx", "rbx", "rsp", "rbp", "rsi", "rdi", - "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15" }, - M = { "mm0", "mm1", "mm2", "mm3", "mm4", "mm5", "mm6", "mm7", - "mm0", "mm1", "mm2", "mm3", "mm4", "mm5", "mm6", "mm7" }, -- No x64 ext! - X = { "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7", - "xmm8", "xmm9", "xmm10", "xmm11", "xmm12", "xmm13", "xmm14", "xmm15" }, -} -local map_segregs = { "es", "cs", "ss", "ds", "fs", "gs", "segr6", "segr7" } - --- Maps for size names. -local map_sz2n = { - B = 1, W = 2, D = 4, Q = 8, M = 8, X = 16, -} -local map_sz2prefix = { - B = "byte", W = "word", D = "dword", - Q = "qword", - M = "qword", X = "xword", - F = "dword", G = "qword", -- No need for sizes/register names for these two. -} - ------------------------------------------------------------------------------- - --- Output a nicely formatted line with an opcode and operands. -local function putop(ctx, text, operands) - local code, pos, hex = ctx.code, ctx.pos, "" - local hmax = ctx.hexdump - if hmax > 0 then - for i=ctx.start,pos-1 do - hex = hex..format("%02X", byte(code, i, i)) - end - if #hex > hmax then hex = sub(hex, 1, hmax)..". " - else hex = hex..rep(" ", hmax-#hex+2) end - end - if operands then text = text.." "..operands end - if ctx.o16 then text = "o16 "..text; ctx.o16 = false end - if ctx.a32 then text = "a32 "..text; ctx.a32 = false end - if ctx.rep then text = ctx.rep.." "..text; ctx.rep = false end - if ctx.rex then - local t = (ctx.rexw and "w" or "")..(ctx.rexr and "r" or "").. - (ctx.rexx and "x" or "")..(ctx.rexb and "b" or "") - if t ~= "" then text = "rex."..t.." "..text end - ctx.rexw = false; ctx.rexr = false; ctx.rexx = false; ctx.rexb = false - ctx.rex = false - end - if ctx.seg then - local text2, n = gsub(text, "%[", "["..ctx.seg..":") - if n == 0 then text = ctx.seg.." "..text else text = text2 end - ctx.seg = false - end - if ctx.lock then text = "lock "..text; ctx.lock = false end - local imm = ctx.imm - if imm then - local sym = ctx.symtab[imm] - if sym then text = text.."\t->"..sym end - end - ctx.out(format("%08x %s%s\n", ctx.addr+ctx.start, hex, text)) - ctx.mrm = false - ctx.start = pos - ctx.imm = nil -end - --- Clear all prefix flags. -local function clearprefixes(ctx) - ctx.o16 = false; ctx.seg = false; ctx.lock = false; ctx.rep = false - ctx.rexw = false; ctx.rexr = false; ctx.rexx = false; ctx.rexb = false - ctx.rex = false; ctx.a32 = false -end - --- Fallback for incomplete opcodes at the end. -local function incomplete(ctx) - ctx.pos = ctx.stop+1 - clearprefixes(ctx) - return putop(ctx, "(incomplete)") -end - --- Fallback for unknown opcodes. -local function unknown(ctx) - clearprefixes(ctx) - return putop(ctx, "(unknown)") -end - --- Return an immediate of the specified size. -local function getimm(ctx, pos, n) - if pos+n-1 > ctx.stop then return incomplete(ctx) end - local code = ctx.code - if n == 1 then - local b1 = byte(code, pos, pos) - return b1 - elseif n == 2 then - local b1, b2 = byte(code, pos, pos+1) - return b1+b2*256 - else - local b1, b2, b3, b4 = byte(code, pos, pos+3) - local imm = b1+b2*256+b3*65536+b4*16777216 - ctx.imm = imm - return imm - end -end - --- Process pattern string and generate the operands. -local function putpat(ctx, name, pat) - local operands, regs, sz, mode, sp, rm, sc, rx, sdisp - local code, pos, stop = ctx.code, ctx.pos, ctx.stop - - -- Chars used: 1DFGIMPQRSTUVWXacdfgijmoprstuwxyz - for p in gmatch(pat, ".") do - local x = nil - if p == "V" or p == "U" then - if ctx.rexw then sz = "Q"; ctx.rexw = false - elseif ctx.o16 then sz = "W"; ctx.o16 = false - elseif p == "U" and ctx.x64 then sz = "Q" - else sz = "D" end - regs = map_regs[sz] - elseif p == "T" then - if ctx.rexw then sz = "Q"; ctx.rexw = false else sz = "D" end - regs = map_regs[sz] - elseif p == "B" then - sz = "B" - regs = ctx.rex and map_regs.B64 or map_regs.B - elseif match(p, "[WDQMXFG]") then - sz = p - regs = map_regs[sz] - elseif p == "P" then - sz = ctx.o16 and "X" or "M"; ctx.o16 = false - regs = map_regs[sz] - elseif p == "S" then - name = name..lower(sz) - elseif p == "s" then - local imm = getimm(ctx, pos, 1); if not imm then return end - x = imm <= 127 and format("+0x%02x", imm) - or format("-0x%02x", 256-imm) - pos = pos+1 - elseif p == "u" then - local imm = getimm(ctx, pos, 1); if not imm then return end - x = format("0x%02x", imm) - pos = pos+1 - elseif p == "w" then - local imm = getimm(ctx, pos, 2); if not imm then return end - x = format("0x%x", imm) - pos = pos+2 - elseif p == "o" then -- [offset] - if ctx.x64 then - local imm1 = getimm(ctx, pos, 4); if not imm1 then return end - local imm2 = getimm(ctx, pos+4, 4); if not imm2 then return end - x = format("[0x%08x%08x]", imm2, imm1) - pos = pos+8 - else - local imm = getimm(ctx, pos, 4); if not imm then return end - x = format("[0x%08x]", imm) - pos = pos+4 - end - elseif p == "i" or p == "I" then - local n = map_sz2n[sz] - if n == 8 and ctx.x64 and p == "I" then - local imm1 = getimm(ctx, pos, 4); if not imm1 then return end - local imm2 = getimm(ctx, pos+4, 4); if not imm2 then return end - x = format("0x%08x%08x", imm2, imm1) - else - if n == 8 then n = 4 end - local imm = getimm(ctx, pos, n); if not imm then return end - if sz == "Q" and (imm < 0 or imm > 0x7fffffff) then - imm = (0xffffffff+1)-imm - x = format(imm > 65535 and "-0x%08x" or "-0x%x", imm) - else - x = format(imm > 65535 and "0x%08x" or "0x%x", imm) - end - end - pos = pos+n - elseif p == "j" then - local n = map_sz2n[sz] - if n == 8 then n = 4 end - local imm = getimm(ctx, pos, n); if not imm then return end - if sz == "B" and imm > 127 then imm = imm-256 - elseif imm > 2147483647 then imm = imm-4294967296 end - pos = pos+n - imm = imm + pos + ctx.addr - if imm > 4294967295 and not ctx.x64 then imm = imm-4294967296 end - ctx.imm = imm - if sz == "W" then - x = format("word 0x%04x", imm%65536) - elseif ctx.x64 then - local lo = imm % 0x1000000 - x = format("0x%02x%06x", (imm-lo) / 0x1000000, lo) - else - x = format("0x%08x", imm) - end - elseif p == "R" then - local r = byte(code, pos-1, pos-1)%8 - if ctx.rexb then r = r + 8; ctx.rexb = false end - x = regs[r+1] - elseif p == "a" then x = regs[1] - elseif p == "c" then x = "cl" - elseif p == "d" then x = "dx" - elseif p == "1" then x = "1" - else - if not mode then - mode = ctx.mrm - if not mode then - if pos > stop then return incomplete(ctx) end - mode = byte(code, pos, pos) - pos = pos+1 - end - rm = mode%8; mode = (mode-rm)/8 - sp = mode%8; mode = (mode-sp)/8 - sdisp = "" - if mode < 3 then - if rm == 4 then - if pos > stop then return incomplete(ctx) end - sc = byte(code, pos, pos) - pos = pos+1 - rm = sc%8; sc = (sc-rm)/8 - rx = sc%8; sc = (sc-rx)/8 - if ctx.rexx then rx = rx + 8; ctx.rexx = false end - if rx == 4 then rx = nil end - end - if mode > 0 or rm == 5 then - local dsz = mode - if dsz ~= 1 then dsz = 4 end - local disp = getimm(ctx, pos, dsz); if not disp then return end - if mode == 0 then rm = nil end - if rm or rx or (not sc and ctx.x64 and not ctx.a32) then - if dsz == 1 and disp > 127 then - sdisp = format("-0x%x", 256-disp) - elseif disp >= 0 and disp <= 0x7fffffff then - sdisp = format("+0x%x", disp) - else - sdisp = format("-0x%x", (0xffffffff+1)-disp) - end - else - sdisp = format(ctx.x64 and not ctx.a32 and - not (disp >= 0 and disp <= 0x7fffffff) - and "0xffffffff%08x" or "0x%08x", disp) - end - pos = pos+dsz - end - end - if rm and ctx.rexb then rm = rm + 8; ctx.rexb = false end - if ctx.rexr then sp = sp + 8; ctx.rexr = false end - end - if p == "m" then - if mode == 3 then x = regs[rm+1] - else - local aregs = ctx.a32 and map_regs.D or ctx.aregs - local srm, srx = "", "" - if rm then srm = aregs[rm+1] - elseif not sc and ctx.x64 and not ctx.a32 then srm = "rip" end - ctx.a32 = false - if rx then - if rm then srm = srm.."+" end - srx = aregs[rx+1] - if sc > 0 then srx = srx.."*"..(2^sc) end - end - x = format("[%s%s%s]", srm, srx, sdisp) - end - if mode < 3 and - (not match(pat, "[aRrgp]") or match(pat, "t")) then -- Yuck. - x = map_sz2prefix[sz].." "..x - end - elseif p == "r" then x = regs[sp+1] - elseif p == "g" then x = map_segregs[sp+1] - elseif p == "p" then -- Suppress prefix. - elseif p == "f" then x = "st"..rm - elseif p == "x" then - if sp == 0 and ctx.lock and not ctx.x64 then - x = "CR8"; ctx.lock = false - else - x = "CR"..sp - end - elseif p == "y" then x = "DR"..sp - elseif p == "z" then x = "TR"..sp - elseif p == "t" then - else - error("bad pattern `"..pat.."'") - end - end - if x then operands = operands and operands..", "..x or x end - end - ctx.pos = pos - return putop(ctx, name, operands) -end - --- Forward declaration. -local map_act - --- Fetch and cache MRM byte. -local function getmrm(ctx) - local mrm = ctx.mrm - if not mrm then - local pos = ctx.pos - if pos > ctx.stop then return nil end - mrm = byte(ctx.code, pos, pos) - ctx.pos = pos+1 - ctx.mrm = mrm - end - return mrm -end - --- Dispatch to handler depending on pattern. -local function dispatch(ctx, opat, patgrp) - if not opat then return unknown(ctx) end - if match(opat, "%|") then -- MMX/SSE variants depending on prefix. - local p - if ctx.rep then - p = ctx.rep=="rep" and "%|([^%|]*)" or "%|[^%|]*%|[^%|]*%|([^%|]*)" - ctx.rep = false - elseif ctx.o16 then p = "%|[^%|]*%|([^%|]*)"; ctx.o16 = false - else p = "^[^%|]*" end - opat = match(opat, p) - if not opat then return unknown(ctx) end --- ctx.rep = false; ctx.o16 = false - --XXX fails for 66 f2 0f 38 f1 06 crc32 eax,WORD PTR [esi] - --XXX remove in branches? - end - if match(opat, "%$") then -- reg$mem variants. - local mrm = getmrm(ctx); if not mrm then return incomplete(ctx) end - opat = match(opat, mrm >= 192 and "^[^%$]*" or "%$(.*)") - if opat == "" then return unknown(ctx) end - end - if opat == "" then return unknown(ctx) end - local name, pat = match(opat, "^([a-z0-9 ]*)(.*)") - if pat == "" and patgrp then pat = patgrp end - return map_act[sub(pat, 1, 1)](ctx, name, pat) -end - --- Get a pattern from an opcode map and dispatch to handler. -local function dispatchmap(ctx, opcmap) - local pos = ctx.pos - local opat = opcmap[byte(ctx.code, pos, pos)] - pos = pos + 1 - ctx.pos = pos - return dispatch(ctx, opat) -end - --- Map for action codes. The key is the first char after the name. -map_act = { - -- Simple opcodes without operands. - [""] = function(ctx, name, pat) - return putop(ctx, name) - end, - - -- Operand size chars fall right through. - B = putpat, W = putpat, D = putpat, Q = putpat, - V = putpat, U = putpat, T = putpat, - M = putpat, X = putpat, P = putpat, - F = putpat, G = putpat, - - -- Collect prefixes. - [":"] = function(ctx, name, pat) - ctx[pat == ":" and name or sub(pat, 2)] = name - if ctx.pos - ctx.start > 5 then return unknown(ctx) end -- Limit #prefixes. - end, - - -- Chain to special handler specified by name. - ["*"] = function(ctx, name, pat) - return map_act[name](ctx, name, sub(pat, 2)) - end, - - -- Use named subtable for opcode group. - ["!"] = function(ctx, name, pat) - local mrm = getmrm(ctx); if not mrm then return incomplete(ctx) end - return dispatch(ctx, map_opcgroup[name][((mrm-(mrm%8))/8)%8+1], sub(pat, 2)) - end, - - -- o16,o32[,o64] variants. - sz = function(ctx, name, pat) - if ctx.o16 then ctx.o16 = false - else - pat = match(pat, ",(.*)") - if ctx.rexw then - local p = match(pat, ",(.*)") - if p then pat = p; ctx.rexw = false end - end - end - pat = match(pat, "^[^,]*") - return dispatch(ctx, pat) - end, - - -- Two-byte opcode dispatch. - opc2 = function(ctx, name, pat) - return dispatchmap(ctx, map_opc2) - end, - - -- Three-byte opcode dispatch. - opc3 = function(ctx, name, pat) - return dispatchmap(ctx, map_opc3[pat]) - end, - - -- VMX/SVM dispatch. - vm = function(ctx, name, pat) - return dispatch(ctx, map_opcvm[ctx.mrm]) - end, - - -- Floating point opcode dispatch. - fp = function(ctx, name, pat) - local mrm = getmrm(ctx); if not mrm then return incomplete(ctx) end - local rm = mrm%8 - local idx = pat*8 + ((mrm-rm)/8)%8 - if mrm >= 192 then idx = idx + 64 end - local opat = map_opcfp[idx] - if type(opat) == "table" then opat = opat[rm+1] end - return dispatch(ctx, opat) - end, - - -- REX prefix. - rex = function(ctx, name, pat) - if ctx.rex then return unknown(ctx) end -- Only 1 REX prefix allowed. - for p in gmatch(pat, ".") do ctx["rex"..p] = true end - ctx.rex = true - end, - - -- Special case for nop with REX prefix. - nop = function(ctx, name, pat) - return dispatch(ctx, ctx.rex and pat or "nop") - end, -} - ------------------------------------------------------------------------------- - --- Disassemble a block of code. -local function disass_block(ctx, ofs, len) - if not ofs then ofs = 0 end - local stop = len and ofs+len or #ctx.code - ofs = ofs + 1 - ctx.start = ofs - ctx.pos = ofs - ctx.stop = stop - ctx.imm = nil - ctx.mrm = false - clearprefixes(ctx) - while ctx.pos <= stop do dispatchmap(ctx, ctx.map1) end - if ctx.pos ~= ctx.start then incomplete(ctx) end -end - --- Extended API: create a disassembler context. Then call ctx:disass(ofs, len). -local function create_(code, addr, out) - local ctx = {} - ctx.code = code - ctx.addr = (addr or 0) - 1 - ctx.out = out or io.write - ctx.symtab = {} - ctx.disass = disass_block - ctx.hexdump = 16 - ctx.x64 = false - ctx.map1 = map_opc1_32 - ctx.aregs = map_regs.D - return ctx -end - -local function create64_(code, addr, out) - local ctx = create_(code, addr, out) - ctx.x64 = true - ctx.map1 = map_opc1_64 - ctx.aregs = map_regs.Q - return ctx -end - --- Simple API: disassemble code (a string) at address and output via out. -local function disass_(code, addr, out) - create_(code, addr, out):disass() -end - -local function disass64_(code, addr, out) - create64_(code, addr, out):disass() -end - --- Return register name for RID. -local function regname_(r) - if r < 8 then return map_regs.D[r+1] end - return map_regs.X[r-7] -end - -local function regname64_(r) - if r < 16 then return map_regs.Q[r+1] end - return map_regs.X[r-15] -end - --- Public module functions. -module(...) - -create = create_ -create64 = create64_ -disass = disass_ -disass64 = disass64_ -regname = regname_ -regname64 = regname64_ - diff --git a/lib/dump.lua b/lib/dump.lua deleted file mode 100644 index 3d62c4ea..00000000 --- a/lib/dump.lua +++ /dev/null @@ -1,696 +0,0 @@ ----------------------------------------------------------------------------- --- LuaJIT compiler dump module. --- --- Copyright (C) 2005-2012 Mike Pall. All rights reserved. --- Released under the MIT license. See Copyright Notice in luajit.h ----------------------------------------------------------------------------- --- --- This module can be used to debug the JIT compiler itself. It dumps the --- code representations and structures used in various compiler stages. --- --- Example usage: --- --- luajit -jdump -e "local x=0; for i=1,1e6 do x=x+i end; print(x)" --- luajit -jdump=im -e "for i=1,1000 do for j=1,1000 do end end" | less -R --- luajit -jdump=is myapp.lua | less -R --- luajit -jdump=-b myapp.lua --- luajit -jdump=+aH,myapp.html myapp.lua --- luajit -jdump=ixT,myapp.dump myapp.lua --- --- The first argument specifies the dump mode. The second argument gives --- the output file name. Default output is to stdout, unless the environment --- variable LUAJIT_DUMPFILE is set. The file is overwritten every time the --- module is started. --- --- Different features can be turned on or off with the dump mode. If the --- mode starts with a '+', the following features are added to the default --- set of features; a '-' removes them. Otherwise the features are replaced. --- --- The following dump features are available (* marks the default): --- --- * t Print a line for each started, ended or aborted trace (see also -jv). --- * b Dump the traced bytecode. --- * i Dump the IR (intermediate representation). --- r Augment the IR with register/stack slots. --- s Dump the snapshot map. --- * m Dump the generated machine code. --- x Print each taken trace exit. --- X Print each taken trace exit and the contents of all registers. --- --- The output format can be set with the following characters: --- --- T Plain text output. --- A ANSI-colored text output --- H Colorized HTML + CSS output. --- --- The default output format is plain text. It's set to ANSI-colored text --- if the COLORTERM variable is set. Note: this is independent of any output --- redirection, which is actually considered a feature. --- --- You probably want to use less -R to enjoy viewing ANSI-colored text from --- a pipe or a file. Add this to your ~/.bashrc: export LESS="-R" --- ------------------------------------------------------------------------------- - --- Cache some library functions and objects. -local jit = require("jit") -assert(jit.version_num == 20000, "LuaJIT core/library version mismatch") -local jutil = require("jit.util") -local vmdef = require("jit.vmdef") -local funcinfo, funcbc = jutil.funcinfo, jutil.funcbc -local traceinfo, traceir, tracek = jutil.traceinfo, jutil.traceir, jutil.tracek -local tracemc, tracesnap = jutil.tracemc, jutil.tracesnap -local traceexitstub, ircalladdr = jutil.traceexitstub, jutil.ircalladdr -local bit = require("bit") -local band, shl, shr = bit.band, bit.lshift, bit.rshift -local sub, gsub, format = string.sub, string.gsub, string.format -local byte, char, rep = string.byte, string.char, string.rep -local type, tostring = type, tostring -local stdout, stderr = io.stdout, io.stderr - --- Load other modules on-demand. -local bcline, disass - --- Active flag, output file handle and dump mode. -local active, out, dumpmode - ------------------------------------------------------------------------------- - -local symtabmt = { __index = false } -local symtab = {} -local nexitsym = 0 - --- Fill nested symbol table with per-trace exit stub addresses. -local function fillsymtab_tr(tr, nexit) - local t = {} - symtabmt.__index = t - if jit.arch == "mips" or jit.arch == "mipsel" then - t[traceexitstub(tr, 0)] = "exit" - return - end - for i=0,nexit-1 do - local addr = traceexitstub(tr, i) - t[addr] = tostring(i) - end - local addr = traceexitstub(tr, nexit) - if addr then t[addr] = "stack_check" end -end - --- Fill symbol table with trace exit stub addresses. -local function fillsymtab(tr, nexit) - local t = symtab - if nexitsym == 0 then - local ircall = vmdef.ircall - for i=0,#ircall do - local addr = ircalladdr(i) - if addr ~= 0 then t[addr] = ircall[i] end - end - end - if nexitsym == 1000000 then -- Per-trace exit stubs. - fillsymtab_tr(tr, nexit) - elseif nexit > nexitsym then -- Shared exit stubs. - for i=nexitsym,nexit-1 do - local addr = traceexitstub(i) - if addr == nil then -- Fall back to per-trace exit stubs. - fillsymtab_tr(tr, nexit) - setmetatable(symtab, symtabmt) - nexit = 1000000 - break - end - t[addr] = tostring(i) - end - nexitsym = nexit - end - return t -end - -local function dumpwrite(s) - out:write(s) -end - --- Disassemble machine code. -local function dump_mcode(tr) - local info = traceinfo(tr) - if not info then return end - local mcode, addr, loop = tracemc(tr) - if not mcode then return end - if not disass then disass = require("jit.dis_"..jit.arch) end - out:write("---- TRACE ", tr, " mcode ", #mcode, "\n") - local ctx = disass.create(mcode, addr, dumpwrite) - ctx.hexdump = 0 - ctx.symtab = fillsymtab(tr, info.nexit) - if loop ~= 0 then - symtab[addr+loop] = "LOOP" - ctx:disass(0, loop) - out:write("->LOOP:\n") - ctx:disass(loop, #mcode-loop) - symtab[addr+loop] = nil - else - ctx:disass(0, #mcode) - end -end - ------------------------------------------------------------------------------- - -local irtype_text = { - [0] = "nil", - "fal", - "tru", - "lud", - "str", - "p32", - "thr", - "pro", - "fun", - "p64", - "cdt", - "tab", - "udt", - "flt", - "num", - "i8 ", - "u8 ", - "i16", - "u16", - "int", - "u32", - "i64", - "u64", - "sfp", -} - -local colortype_ansi = { - [0] = "%s", - "%s", - "%s", - "\027[36m%s\027[m", - "\027[32m%s\027[m", - "%s", - "\027[1m%s\027[m", - "%s", - "\027[1m%s\027[m", - "%s", - "\027[33m%s\027[m", - "\027[31m%s\027[m", - "\027[36m%s\027[m", - "\027[34m%s\027[m", - "\027[34m%s\027[m", - "\027[35m%s\027[m", - "\027[35m%s\027[m", - "\027[35m%s\027[m", - "\027[35m%s\027[m", - "\027[35m%s\027[m", - "\027[35m%s\027[m", - "\027[35m%s\027[m", - "\027[35m%s\027[m", - "\027[35m%s\027[m", -} - -local function colorize_text(s, t) - return s -end - -local function colorize_ansi(s, t) - return format(colortype_ansi[t], s) -end - -local irtype_ansi = setmetatable({}, - { __index = function(tab, t) - local s = colorize_ansi(irtype_text[t], t); tab[t] = s; return s; end }) - -local html_escape = { ["<"] = "<", [">"] = ">", ["&"] = "&", } - -local function colorize_html(s, t) - s = gsub(s, "[<>&]", html_escape) - return format('%s', irtype_text[t], s) -end - -local irtype_html = setmetatable({}, - { __index = function(tab, t) - local s = colorize_html(irtype_text[t], t); tab[t] = s; return s; end }) - -local header_html = [[ - -]] - -local colorize, irtype - --- Lookup tables to convert some literals into names. -local litname = { - ["SLOAD "] = setmetatable({}, { __index = function(t, mode) - local s = "" - if band(mode, 1) ~= 0 then s = s.."P" end - if band(mode, 2) ~= 0 then s = s.."F" end - if band(mode, 4) ~= 0 then s = s.."T" end - if band(mode, 8) ~= 0 then s = s.."C" end - if band(mode, 16) ~= 0 then s = s.."R" end - if band(mode, 32) ~= 0 then s = s.."I" end - t[mode] = s - return s - end}), - ["XLOAD "] = { [0] = "", "R", "V", "RV", "U", "RU", "VU", "RVU", }, - ["CONV "] = setmetatable({}, { __index = function(t, mode) - local s = irtype[band(mode, 31)] - s = irtype[band(shr(mode, 5), 31)].."."..s - if band(mode, 0x400) ~= 0 then s = s.." trunc" - elseif band(mode, 0x800) ~= 0 then s = s.." sext" end - local c = shr(mode, 14) - if c == 2 then s = s.." index" elseif c == 3 then s = s.." check" end - t[mode] = s - return s - end}), - ["FLOAD "] = vmdef.irfield, - ["FREF "] = vmdef.irfield, - ["FPMATH"] = vmdef.irfpm, -} - -local function ctlsub(c) - if c == "\n" then return "\\n" - elseif c == "\r" then return "\\r" - elseif c == "\t" then return "\\t" - elseif c == "\r" then return "\\r" - else return format("\\%03d", byte(c)) - end -end - -local function fmtfunc(func, pc) - local fi = funcinfo(func, pc) - if fi.loc then - return fi.loc - elseif fi.ffid then - return vmdef.ffnames[fi.ffid] - elseif fi.addr then - return format("C:%x", fi.addr) - else - return "(?)" - end -end - -local function formatk(tr, idx) - local k, t, slot = tracek(tr, idx) - local tn = type(k) - local s - if tn == "number" then - if k == 2^52+2^51 then - s = "bias" - else - s = format("%+.14g", k) - end - elseif tn == "string" then - s = format(#k > 20 and '"%.20s"~' or '"%s"', gsub(k, "%c", ctlsub)) - elseif tn == "function" then - s = fmtfunc(k) - elseif tn == "table" then - s = format("{%p}", k) - elseif tn == "userdata" then - if t == 12 then - s = format("userdata:%p", k) - else - s = format("[%p]", k) - if s == "[0x00000000]" then s = "NULL" end - end - elseif t == 21 then -- int64_t - s = sub(tostring(k), 1, -3) - if sub(s, 1, 1) ~= "-" then s = "+"..s end - else - s = tostring(k) -- For primitives. - end - s = colorize(format("%-4s", s), t) - if slot then - s = format("%s @%d", s, slot) - end - return s -end - -local function printsnap(tr, snap) - local n = 2 - for s=0,snap[1]-1 do - local sn = snap[n] - if shr(sn, 24) == s then - n = n + 1 - local ref = band(sn, 0xffff) - 0x8000 -- REF_BIAS - if ref < 0 then - out:write(formatk(tr, ref)) - elseif band(sn, 0x80000) ~= 0 then -- SNAP_SOFTFPNUM - out:write(colorize(format("%04d/%04d", ref, ref+1), 14)) - else - local m, ot, op1, op2 = traceir(tr, ref) - out:write(colorize(format("%04d", ref), band(ot, 31))) - end - out:write(band(sn, 0x10000) == 0 and " " or "|") -- SNAP_FRAME - else - out:write("---- ") - end - end - out:write("]\n") -end - --- Dump snapshots (not interleaved with IR). -local function dump_snap(tr) - out:write("---- TRACE ", tr, " snapshots\n") - for i=0,1000000000 do - local snap = tracesnap(tr, i) - if not snap then break end - out:write(format("#%-3d %04d [ ", i, snap[0])) - printsnap(tr, snap) - end -end - --- Return a register name or stack slot for a rid/sp location. -local function ridsp_name(ridsp) - if not disass then disass = require("jit.dis_"..jit.arch) end - local rid = band(ridsp, 0xff) - if ridsp > 255 then return format("[%x]", shr(ridsp, 8)*4) end - if rid < 128 then return disass.regname(rid) end - return "" -end - --- Dump CALL* function ref and return optional ctype. -local function dumpcallfunc(tr, ins) - local ctype - if ins > 0 then - local m, ot, op1, op2 = traceir(tr, ins) - if band(ot, 31) == 0 then -- nil type means CARG(func, ctype). - ins = op1 - ctype = formatk(tr, op2) - end - end - if ins < 0 then - out:write(format("[0x%x](", tonumber((tracek(tr, ins))))) - else - out:write(format("%04d (", ins)) - end - return ctype -end - --- Recursively gather CALL* args and dump them. -local function dumpcallargs(tr, ins) - if ins < 0 then - out:write(formatk(tr, ins)) - else - local m, ot, op1, op2 = traceir(tr, ins) - local oidx = 6*shr(ot, 8) - local op = sub(vmdef.irnames, oidx+1, oidx+6) - if op == "CARG " then - dumpcallargs(tr, op1) - if op2 < 0 then - out:write(" ", formatk(tr, op2)) - else - out:write(" ", format("%04d", op2)) - end - else - out:write(format("%04d", ins)) - end - end -end - --- Dump IR and interleaved snapshots. -local function dump_ir(tr, dumpsnap, dumpreg) - local info = traceinfo(tr) - if not info then return end - local nins = info.nins - out:write("---- TRACE ", tr, " IR\n") - local irnames = vmdef.irnames - local snapref = 65536 - local snap, snapno - if dumpsnap then - snap = tracesnap(tr, 0) - snapref = snap[0] - snapno = 0 - end - for ins=1,nins do - if ins >= snapref then - if dumpreg then - out:write(format(".... SNAP #%-3d [ ", snapno)) - else - out:write(format(".... SNAP #%-3d [ ", snapno)) - end - printsnap(tr, snap) - snapno = snapno + 1 - snap = tracesnap(tr, snapno) - snapref = snap and snap[0] or 65536 - end - local m, ot, op1, op2, ridsp = traceir(tr, ins) - local oidx, t = 6*shr(ot, 8), band(ot, 31) - local op = sub(irnames, oidx+1, oidx+6) - if op == "LOOP " then - if dumpreg then - out:write(format("%04d ------------ LOOP ------------\n", ins)) - else - out:write(format("%04d ------ LOOP ------------\n", ins)) - end - elseif op ~= "NOP " and op ~= "CARG " and - (dumpreg or op ~= "RENAME") then - if dumpreg then - out:write(format("%04d %-5s ", ins, ridsp_name(ridsp))) - else - out:write(format("%04d ", ins)) - end - out:write(format("%s%s %s %s ", - band(ot, 128) == 0 and " " or ">", - band(ot, 64) == 0 and " " or "+", - irtype[t], op)) - local m1, m2 = band(m, 3), band(m, 3*4) - if sub(op, 1, 4) == "CALL" then - local ctype - if m2 == 1*4 then -- op2 == IRMlit - out:write(format("%-10s (", vmdef.ircall[op2])) - else - ctype = dumpcallfunc(tr, op2) - end - if op1 ~= -1 then dumpcallargs(tr, op1) end - out:write(")") - if ctype then out:write(" ctype ", ctype) end - elseif op == "CNEW " and op2 == -1 then - out:write(formatk(tr, op1)) - elseif m1 ~= 3 then -- op1 != IRMnone - if op1 < 0 then - out:write(formatk(tr, op1)) - else - out:write(format(m1 == 0 and "%04d" or "#%-3d", op1)) - end - if m2 ~= 3*4 then -- op2 != IRMnone - if m2 == 1*4 then -- op2 == IRMlit - local litn = litname[op] - if litn and litn[op2] then - out:write(" ", litn[op2]) - elseif op == "UREFO " or op == "UREFC " then - out:write(format(" #%-3d", shr(op2, 8))) - else - out:write(format(" #%-3d", op2)) - end - elseif op2 < 0 then - out:write(" ", formatk(tr, op2)) - else - out:write(format(" %04d", op2)) - end - end - end - out:write("\n") - end - end - if snap then - if dumpreg then - out:write(format(".... SNAP #%-3d [ ", snapno)) - else - out:write(format(".... SNAP #%-3d [ ", snapno)) - end - printsnap(tr, snap) - end -end - ------------------------------------------------------------------------------- - -local recprefix = "" -local recdepth = 0 - --- Format trace error message. -local function fmterr(err, info) - if type(err) == "number" then - if type(info) == "function" then info = fmtfunc(info) end - err = format(vmdef.traceerr[err], info) - end - return err -end - --- Dump trace states. -local function dump_trace(what, tr, func, pc, otr, oex) - if what == "stop" or (what == "abort" and dumpmode.a) then - if dumpmode.i then dump_ir(tr, dumpmode.s, dumpmode.r and what == "stop") - elseif dumpmode.s then dump_snap(tr) end - if dumpmode.m then dump_mcode(tr) end - end - if what == "start" then - if dumpmode.H then out:write('

\n') end
-    out:write("---- TRACE ", tr, " ", what)
-    if otr then out:write(" ", otr, "/", oex) end
-    out:write(" ", fmtfunc(func, pc), "\n")
-    recprefix = ""
-  elseif what == "stop" or what == "abort" then
-    out:write("---- TRACE ", tr, " ", what)
-    recprefix = nil
-    if what == "abort" then
-      out:write(" ", fmtfunc(func, pc), " -- ", fmterr(otr, oex), "\n")
-    else
-      local info = traceinfo(tr)
-      local link, ltype = info.link, info.linktype
-      if link == tr or link == 0 then
-	out:write(" -> ", ltype, "\n")
-      elseif ltype == "root" then
-	out:write(" -> ", link, "\n")
-      else
-	out:write(" -> ", link, " ", ltype, "\n")
-      end
-    end
-    if dumpmode.H then out:write("
\n\n") else out:write("\n") end - else - out:write("---- TRACE ", what, "\n\n") - end - out:flush() -end - --- Dump recorded bytecode. -local function dump_record(tr, func, pc, depth, callee) - if depth ~= recdepth then - recdepth = depth - recprefix = rep(" .", depth) - end - local line - if pc >= 0 then - line = bcline(func, pc, recprefix) - if dumpmode.H then line = gsub(line, "[<>&]", html_escape) end - else - line = "0000 "..recprefix.." FUNCC \n" - callee = func - end - if pc <= 0 then - out:write(sub(line, 1, -2), " ; ", fmtfunc(func), "\n") - else - out:write(line) - end - if pc >= 0 and band(funcbc(func, pc), 0xff) < 16 then -- ORDER BC - out:write(bcline(func, pc+1, recprefix)) -- Write JMP for cond. - end -end - ------------------------------------------------------------------------------- - --- Dump taken trace exits. -local function dump_texit(tr, ex, ngpr, nfpr, ...) - out:write("---- TRACE ", tr, " exit ", ex, "\n") - if dumpmode.X then - local regs = {...} - if jit.arch == "x64" then - for i=1,ngpr do - out:write(format(" %016x", regs[i])) - if i % 4 == 0 then out:write("\n") end - end - else - for i=1,ngpr do - out:write(format(" %08x", regs[i])) - if i % 8 == 0 then out:write("\n") end - end - end - if jit.arch == "mips" or jit.arch == "mipsel" then - for i=1,nfpr,2 do - out:write(format(" %+17.14g", regs[ngpr+i])) - if i % 8 == 7 then out:write("\n") end - end - else - for i=1,nfpr do - out:write(format(" %+17.14g", regs[ngpr+i])) - if i % 4 == 0 then out:write("\n") end - end - end - end -end - ------------------------------------------------------------------------------- - --- Detach dump handlers. -local function dumpoff() - if active then - active = false - jit.attach(dump_texit) - jit.attach(dump_record) - jit.attach(dump_trace) - if out and out ~= stdout and out ~= stderr then out:close() end - out = nil - end -end - --- Open the output file and attach dump handlers. -local function dumpon(opt, outfile) - if active then dumpoff() end - - local colormode = os.getenv("COLORTERM") and "A" or "T" - if opt then - opt = gsub(opt, "[TAH]", function(mode) colormode = mode; return ""; end) - end - - local m = { t=true, b=true, i=true, m=true, } - if opt and opt ~= "" then - local o = sub(opt, 1, 1) - if o ~= "+" and o ~= "-" then m = {} end - for i=1,#opt do m[sub(opt, i, i)] = (o ~= "-") end - end - dumpmode = m - - if m.t or m.b or m.i or m.s or m.m then - jit.attach(dump_trace, "trace") - end - if m.b then - jit.attach(dump_record, "record") - if not bcline then bcline = require("jit.bc").line end - end - if m.x or m.X then - jit.attach(dump_texit, "texit") - end - - if not outfile then outfile = os.getenv("LUAJIT_DUMPFILE") end - if outfile then - out = outfile == "-" and stdout or assert(io.open(outfile, "w")) - else - out = stdout - end - - m[colormode] = true - if colormode == "A" then - colorize = colorize_ansi - irtype = irtype_ansi - elseif colormode == "H" then - colorize = colorize_html - irtype = irtype_html - out:write(header_html) - else - colorize = colorize_text - irtype = irtype_text - end - - active = true -end - --- Public module functions. -module(...) - -on = dumpon -off = dumpoff -start = dumpon -- For -j command line option. - diff --git a/lib/v.lua b/lib/v.lua deleted file mode 100644 index 8f58fe3b..00000000 --- a/lib/v.lua +++ /dev/null @@ -1,167 +0,0 @@ ----------------------------------------------------------------------------- --- Verbose mode of the LuaJIT compiler. --- --- Copyright (C) 2005-2012 Mike Pall. All rights reserved. --- Released under the MIT license. See Copyright Notice in luajit.h ----------------------------------------------------------------------------- --- --- This module shows verbose information about the progress of the --- JIT compiler. It prints one line for each generated trace. This module --- is useful to see which code has been compiled or where the compiler --- punts and falls back to the interpreter. --- --- Example usage: --- --- luajit -jv -e "for i=1,1000 do for j=1,1000 do end end" --- luajit -jv=myapp.out myapp.lua --- --- Default output is to stderr. To redirect the output to a file, pass a --- filename as an argument (use '-' for stdout) or set the environment --- variable LUAJIT_VERBOSEFILE. The file is overwritten every time the --- module is started. --- --- The output from the first example should look like this: --- --- [TRACE 1 (command line):1 loop] --- [TRACE 2 (1/3) (command line):1 -> 1] --- --- The first number in each line is the internal trace number. Next are --- the file name ('(command line)') and the line number (':1') where the --- trace has started. Side traces also show the parent trace number and --- the exit number where they are attached to in parentheses ('(1/3)'). --- An arrow at the end shows where the trace links to ('-> 1'), unless --- it loops to itself. --- --- In this case the inner loop gets hot and is traced first, generating --- a root trace. Then the last exit from the 1st trace gets hot, too, --- and triggers generation of the 2nd trace. The side trace follows the --- path along the outer loop and *around* the inner loop, back to its --- start, and then links to the 1st trace. Yes, this may seem unusual, --- if you know how traditional compilers work. Trace compilers are full --- of surprises like this -- have fun! :-) --- --- Aborted traces are shown like this: --- --- [TRACE --- foo.lua:44 -- leaving loop in root trace at foo:lua:50] --- --- Don't worry -- trace aborts are quite common, even in programs which --- can be fully compiled. The compiler may retry several times until it --- finds a suitable trace. --- --- Of course this doesn't work with features that are not-yet-implemented --- (NYI error messages). The VM simply falls back to the interpreter. This --- may not matter at all if the particular trace is not very high up in --- the CPU usage profile. Oh, and the interpreter is quite fast, too. --- --- Also check out the -jdump module, which prints all the gory details. --- ------------------------------------------------------------------------------- - --- Cache some library functions and objects. -local jit = require("jit") -assert(jit.version_num == 20000, "LuaJIT core/library version mismatch") -local jutil = require("jit.util") -local vmdef = require("jit.vmdef") -local funcinfo, traceinfo = jutil.funcinfo, jutil.traceinfo -local type, format = type, string.format -local stdout, stderr = io.stdout, io.stderr - --- Active flag and output file handle. -local active, out - ------------------------------------------------------------------------------- - -local startloc, startex - -local function fmtfunc(func, pc) - local fi = funcinfo(func, pc) - if fi.loc then - return fi.loc - elseif fi.ffid then - return vmdef.ffnames[fi.ffid] - elseif fi.addr then - return format("C:%x", fi.addr) - else - return "(?)" - end -end - --- Format trace error message. -local function fmterr(err, info) - if type(err) == "number" then - if type(info) == "function" then info = fmtfunc(info) end - err = format(vmdef.traceerr[err], info) - end - return err -end - --- Dump trace states. -local function dump_trace(what, tr, func, pc, otr, oex) - if what == "start" then - startloc = fmtfunc(func, pc) - startex = otr and "("..otr.."/"..oex..") " or "" - else - if what == "abort" then - local loc = fmtfunc(func, pc) - if loc ~= startloc then - out:write(format("[TRACE --- %s%s -- %s at %s]\n", - startex, startloc, fmterr(otr, oex), loc)) - else - out:write(format("[TRACE --- %s%s -- %s]\n", - startex, startloc, fmterr(otr, oex))) - end - elseif what == "stop" then - local info = traceinfo(tr) - local link, ltype = info.link, info.linktype - if ltype == "interpreter" then - out:write(format("[TRACE %3s %s%s -- fallback to interpreter]\n", - tr, startex, startloc)) - elseif link == tr or link == 0 then - out:write(format("[TRACE %3s %s%s %s]\n", - tr, startex, startloc, ltype)) - elseif ltype == "root" then - out:write(format("[TRACE %3s %s%s -> %d]\n", - tr, startex, startloc, link)) - else - out:write(format("[TRACE %3s %s%s -> %d %s]\n", - tr, startex, startloc, link, ltype)) - end - else - out:write(format("[TRACE %s]\n", what)) - end - out:flush() - end -end - ------------------------------------------------------------------------------- - --- Detach dump handlers. -local function dumpoff() - if active then - active = false - jit.attach(dump_trace) - if out and out ~= stdout and out ~= stderr then out:close() end - out = nil - end -end - --- Open the output file and attach dump handlers. -local function dumpon(outfile) - if active then dumpoff() end - if not outfile then outfile = os.getenv("LUAJIT_VERBOSEFILE") end - if outfile then - out = outfile == "-" and stdout or assert(io.open(outfile, "w")) - else - out = stderr - end - jit.attach(dump_trace, "trace") - active = true -end - --- Public module functions. -module(...) - -on = dumpon -off = dumpoff -start = dumpon -- For -j command line option. - diff --git a/src/Makefile b/src/Makefile index a0e98068..5977c373 100644 --- a/src/Makefile +++ b/src/Makefile @@ -401,7 +401,7 @@ LJCORE_O= lj_gc.o lj_err.o lj_char.o lj_bc.o lj_obj.o \ LJVMCORE_O= $(LJVM_O) $(LJCORE_O) LJVMCORE_DYNO= $(LJVMCORE_O:.o=_dyn.o) -LIB_VMDEF= ../lib/vmdef.lua +LIB_VMDEF= jit/vmdef.lua LIB_VMDEFP= $(LIB_VMDEF) LUAJIT_O= luajit.o diff --git a/src/jit/.gitignore b/src/jit/.gitignore new file mode 100644 index 00000000..500e2855 --- /dev/null +++ b/src/jit/.gitignore @@ -0,0 +1 @@ +vmdef.lua diff --git a/src/jit/bc.lua b/src/jit/bc.lua new file mode 100644 index 00000000..15317bcd --- /dev/null +++ b/src/jit/bc.lua @@ -0,0 +1,192 @@ +---------------------------------------------------------------------------- +-- LuaJIT bytecode listing module. +-- +-- Copyright (C) 2005-2012 Mike Pall. All rights reserved. +-- Released under the MIT license. See Copyright Notice in luajit.h +---------------------------------------------------------------------------- +-- +-- This module lists the bytecode of a Lua function. If it's loaded by -jbc +-- it hooks into the parser and lists all functions of a chunk as they +-- are parsed. +-- +-- Example usage: +-- +-- luajit -jbc -e 'local x=0; for i=1,1e6 do x=x+i end; print(x)' +-- luajit -jbc=- foo.lua +-- luajit -jbc=foo.list foo.lua +-- +-- Default output is to stderr. To redirect the output to a file, pass a +-- filename as an argument (use '-' for stdout) or set the environment +-- variable LUAJIT_LISTFILE. The file is overwritten every time the module +-- is started. +-- +-- This module can also be used programmatically: +-- +-- local bc = require("jit.bc") +-- +-- local function foo() print("hello") end +-- +-- bc.dump(foo) --> -- BYTECODE -- [...] +-- print(bc.line(foo, 2)) --> 0002 KSTR 1 1 ; "hello" +-- +-- local out = { +-- -- Do something with each line: +-- write = function(t, ...) io.write(...) end, +-- close = function(t) end, +-- flush = function(t) end, +-- } +-- bc.dump(foo, out) +-- +------------------------------------------------------------------------------ + +-- Cache some library functions and objects. +local jit = require("jit") +assert(jit.version_num == 20000, "LuaJIT core/library version mismatch") +local jutil = require("jit.util") +local vmdef = require("jit.vmdef") +local bit = require("bit") +local sub, gsub, format = string.sub, string.gsub, string.format +local byte, band, shr = string.byte, bit.band, bit.rshift +local funcinfo, funcbc, funck = jutil.funcinfo, jutil.funcbc, jutil.funck +local funcuvname = jutil.funcuvname +local bcnames = vmdef.bcnames +local stdout, stderr = io.stdout, io.stderr + +------------------------------------------------------------------------------ + +local function ctlsub(c) + if c == "\n" then return "\\n" + elseif c == "\r" then return "\\r" + elseif c == "\t" then return "\\t" + elseif c == "\r" then return "\\r" + else return format("\\%03d", byte(c)) + end +end + +-- Return one bytecode line. +local function bcline(func, pc, prefix) + local ins, m = funcbc(func, pc) + if not ins then return end + local ma, mb, mc = band(m, 7), band(m, 15*8), band(m, 15*128) + local a = band(shr(ins, 8), 0xff) + local oidx = 6*band(ins, 0xff) + local op = sub(bcnames, oidx+1, oidx+6) + local s = format("%04d %s %-6s %3s ", + pc, prefix or " ", op, ma == 0 and "" or a) + local d = shr(ins, 16) + if mc == 13*128 then -- BCMjump + return format("%s=> %04d\n", s, pc+d-0x7fff) + end + if mb ~= 0 then + d = band(d, 0xff) + elseif mc == 0 then + return s.."\n" + end + local kc + if mc == 10*128 then -- BCMstr + kc = funck(func, -d-1) + kc = format(#kc > 40 and '"%.40s"~' or '"%s"', gsub(kc, "%c", ctlsub)) + elseif mc == 9*128 then -- BCMnum + kc = funck(func, d) + if op == "TSETM " then kc = kc - 2^52 end + elseif mc == 12*128 then -- BCMfunc + local fi = funcinfo(funck(func, -d-1)) + if fi.ffid then + kc = vmdef.ffnames[fi.ffid] + else + kc = fi.loc + end + elseif mc == 5*128 then -- BCMuv + kc = funcuvname(func, d) + end + if ma == 5 then -- BCMuv + local ka = funcuvname(func, a) + if kc then kc = ka.." ; "..kc else kc = ka end + end + if mb ~= 0 then + local b = shr(ins, 24) + if kc then return format("%s%3d %3d ; %s\n", s, b, d, kc) end + return format("%s%3d %3d\n", s, b, d) + end + if kc then return format("%s%3d ; %s\n", s, d, kc) end + if mc == 7*128 and d > 32767 then d = d - 65536 end -- BCMlits + return format("%s%3d\n", s, d) +end + +-- Collect branch targets of a function. +local function bctargets(func) + local target = {} + for pc=1,1000000000 do + local ins, m = funcbc(func, pc) + if not ins then break end + if band(m, 15*128) == 13*128 then target[pc+shr(ins, 16)-0x7fff] = true end + end + return target +end + +-- Dump bytecode instructions of a function. +local function bcdump(func, out, all) + if not out then out = stdout end + local fi = funcinfo(func) + if all and fi.children then + for n=-1,-1000000000,-1 do + local k = funck(func, n) + if not k then break end + if type(k) == "proto" then bcdump(k, out, true) end + end + end + out:write(format("-- BYTECODE -- %s-%d\n", fi.loc, fi.lastlinedefined)) + local target = bctargets(func) + for pc=1,1000000000 do + local s = bcline(func, pc, target[pc] and "=>") + if not s then break end + out:write(s) + end + out:write("\n") + out:flush() +end + +------------------------------------------------------------------------------ + +-- Active flag and output file handle. +local active, out + +-- List handler. +local function h_list(func) + return bcdump(func, out) +end + +-- Detach list handler. +local function bclistoff() + if active then + active = false + jit.attach(h_list) + if out and out ~= stdout and out ~= stderr then out:close() end + out = nil + end +end + +-- Open the output file and attach list handler. +local function bcliston(outfile) + if active then bclistoff() end + if not outfile then outfile = os.getenv("LUAJIT_LISTFILE") end + if outfile then + out = outfile == "-" and stdout or assert(io.open(outfile, "w")) + else + out = stderr + end + jit.attach(h_list, "bc") + active = true +end + +-- Public module functions. +module(...) + +line = bcline +dump = bcdump +targets = bctargets + +on = bcliston +off = bclistoff +start = bcliston -- For -j command line option. + diff --git a/src/jit/bcsave.lua b/src/jit/bcsave.lua new file mode 100644 index 00000000..70cb6925 --- /dev/null +++ b/src/jit/bcsave.lua @@ -0,0 +1,500 @@ +---------------------------------------------------------------------------- +-- LuaJIT module to save/list bytecode. +-- +-- Copyright (C) 2005-2012 Mike Pall. All rights reserved. +-- Released under the MIT license. See Copyright Notice in luajit.h +---------------------------------------------------------------------------- +-- +-- This module saves or lists the bytecode for an input file. +-- It's run by the -b command line option. +-- +------------------------------------------------------------------------------ + +local jit = require("jit") +assert(jit.version_num == 20000, "LuaJIT core/library version mismatch") + +-- Symbol name prefix for LuaJIT bytecode. +local LJBC_PREFIX = "luaJIT_BC_" + +------------------------------------------------------------------------------ + +local function usage() + io.stderr:write[[ +Save LuaJIT bytecode: luajit -b[options] input output + -l Only list bytecode. + -s Strip debug info (default). + -g Keep debug info. + -n name Set module name (default: auto-detect from input name). + -t type Set output file type (default: auto-detect from output name). + -a arch Override architecture for object files (default: native). + -o os Override OS for object files (default: native). + -e chunk Use chunk string as input. + -- Stop handling options. + - Use stdin as input and/or stdout as output. + +File types: c h obj o raw (default) +]] + os.exit(1) +end + +local function check(ok, ...) + if ok then return ok, ... end + io.stderr:write("luajit: ", ...) + io.stderr:write("\n") + os.exit(1) +end + +local function readfile(input) + if type(input) == "function" then return input end + if input == "-" then input = nil end + return check(loadfile(input)) +end + +local function savefile(name, mode) + if name == "-" then return io.stdout end + return check(io.open(name, mode)) +end + +------------------------------------------------------------------------------ + +local map_type = { + raw = "raw", c = "c", h = "h", o = "obj", obj = "obj", +} + +local map_arch = { + x86 = true, x64 = true, arm = true, ppc = true, ppcspe = true, + mips = true, mipsel = true, +} + +local map_os = { + linux = true, windows = true, osx = true, freebsd = true, netbsd = true, + openbsd = true, solaris = true, +} + +local function checkarg(str, map, err) + str = string.lower(str) + local s = check(map[str], "unknown ", err) + return s == true and str or s +end + +local function detecttype(str) + local ext = string.match(string.lower(str), "%.(%a+)$") + return map_type[ext] or "raw" +end + +local function checkmodname(str) + check(string.match(str, "^[%w_.%-]+$"), "bad module name") + return string.gsub(str, "[%.%-]", "_") +end + +local function detectmodname(str) + if type(str) == "string" then + local tail = string.match(str, "[^/\\]+$") + if tail then str = tail end + local head = string.match(str, "^(.*)%.[^.]*$") + if head then str = head end + str = string.match(str, "^[%w_.%-]+") + else + str = nil + end + check(str, "cannot derive module name, use -n name") + return string.gsub(str, "[%.%-]", "_") +end + +------------------------------------------------------------------------------ + +local function bcsave_tail(fp, output, s) + local ok, err = fp:write(s) + if ok and output ~= "-" then ok, err = fp:close() end + check(ok, "cannot write ", output, ": ", err) +end + +local function bcsave_raw(output, s) + local fp = savefile(output, "wb") + bcsave_tail(fp, output, s) +end + +local function bcsave_c(ctx, output, s) + local fp = savefile(output, "w") + if ctx.type == "c" then + fp:write(string.format([[ +#ifdef _cplusplus +extern "C" +#endif +#ifdef _WIN32 +__declspec(dllexport) +#endif +const char %s%s[] = { +]], LJBC_PREFIX, ctx.modname)) + else + fp:write(string.format([[ +#define %s%s_SIZE %d +static const char %s%s[] = { +]], LJBC_PREFIX, ctx.modname, #s, LJBC_PREFIX, ctx.modname)) + end + local t, n, m = {}, 0, 0 + for i=1,#s do + local b = tostring(string.byte(s, i)) + m = m + #b + 1 + if m > 78 then + fp:write(table.concat(t, ",", 1, n), ",\n") + n, m = 0, #b + 1 + end + n = n + 1 + t[n] = b + end + bcsave_tail(fp, output, table.concat(t, ",", 1, n).."\n};\n") +end + +local function bcsave_elfobj(ctx, output, s, ffi) + ffi.cdef[[ +typedef struct { + uint8_t emagic[4], eclass, eendian, eversion, eosabi, eabiversion, epad[7]; + uint16_t type, machine; + uint32_t version; + uint32_t entry, phofs, shofs; + uint32_t flags; + uint16_t ehsize, phentsize, phnum, shentsize, shnum, shstridx; +} ELF32header; +typedef struct { + uint8_t emagic[4], eclass, eendian, eversion, eosabi, eabiversion, epad[7]; + uint16_t type, machine; + uint32_t version; + uint64_t entry, phofs, shofs; + uint32_t flags; + uint16_t ehsize, phentsize, phnum, shentsize, shnum, shstridx; +} ELF64header; +typedef struct { + uint32_t name, type, flags, addr, ofs, size, link, info, align, entsize; +} ELF32sectheader; +typedef struct { + uint32_t name, type; + uint64_t flags, addr, ofs, size; + uint32_t link, info; + uint64_t align, entsize; +} ELF64sectheader; +typedef struct { + uint32_t name, value, size; + uint8_t info, other; + uint16_t sectidx; +} ELF32symbol; +typedef struct { + uint32_t name; + uint8_t info, other; + uint16_t sectidx; + uint64_t value, size; +} ELF64symbol; +typedef struct { + ELF32header hdr; + ELF32sectheader sect[6]; + ELF32symbol sym[2]; + uint8_t space[4096]; +} ELF32obj; +typedef struct { + ELF64header hdr; + ELF64sectheader sect[6]; + ELF64symbol sym[2]; + uint8_t space[4096]; +} ELF64obj; +]] + local symname = LJBC_PREFIX..ctx.modname + local is64, isbe = false, false + if ctx.arch == "x64" then + is64 = true + elseif ctx.arch == "ppc" or ctx.arch == "ppcspe" or ctx.arch == "mips" then + isbe = true + end + + -- Handle different host/target endianess. + local function f32(x) return x end + local f16, fofs = f32, f32 + if ffi.abi("be") ~= isbe then + f32 = bit.bswap + function f16(x) return bit.rshift(bit.bswap(x), 16) end + if is64 then + function fofs(x) return bit.bswap(x)*(2ll^32) end + else + fofs = f32 + end + end + + -- Create ELF object and fill in header. + local o = ffi.new(is64 and "ELF64obj" or "ELF32obj") + local hdr = o.hdr + if ctx.os == "bsd" or ctx.os == "other" then -- Determine native hdr.eosabi. + local bf = assert(io.open("/bin/ls", "rb")) + local bs = bf:read(9) + bf:close() + ffi.copy(o, bs, 9) + check(hdr.emagic[0] == 127, "no support for writing native object files") + else + hdr.emagic = "\127ELF" + hdr.eosabi = ({ freebsd=9, netbsd=2, openbsd=12, solaris=6 })[ctx.os] or 0 + end + hdr.eclass = is64 and 2 or 1 + hdr.eendian = isbe and 2 or 1 + hdr.eversion = 1 + hdr.type = f16(1) + hdr.machine = f16(({ x86=3, x64=62, arm=40, ppc=20, ppcspe=20, mips=8, mipsel=8 })[ctx.arch]) + if ctx.arch == "mips" or ctx.arch == "mipsel" then + hdr.flags = 0x50001006 + end + hdr.version = f32(1) + hdr.shofs = fofs(ffi.offsetof(o, "sect")) + hdr.ehsize = f16(ffi.sizeof(hdr)) + hdr.shentsize = f16(ffi.sizeof(o.sect[0])) + hdr.shnum = f16(6) + hdr.shstridx = f16(2) + + -- Fill in sections and symbols. + local sofs, ofs = ffi.offsetof(o, "space"), 1 + for i,name in ipairs{ + ".symtab", ".shstrtab", ".strtab", ".rodata", ".note.GNU-stack", + } do + local sect = o.sect[i] + sect.align = fofs(1) + sect.name = f32(ofs) + ffi.copy(o.space+ofs, name) + ofs = ofs + #name+1 + end + o.sect[1].type = f32(2) -- .symtab + o.sect[1].link = f32(3) + o.sect[1].info = f32(1) + o.sect[1].align = fofs(8) + o.sect[1].ofs = fofs(ffi.offsetof(o, "sym")) + o.sect[1].entsize = fofs(ffi.sizeof(o.sym[0])) + o.sect[1].size = fofs(ffi.sizeof(o.sym)) + o.sym[1].name = f32(1) + o.sym[1].sectidx = f16(4) + o.sym[1].size = fofs(#s) + o.sym[1].info = 17 + o.sect[2].type = f32(3) -- .shstrtab + o.sect[2].ofs = fofs(sofs) + o.sect[2].size = fofs(ofs) + o.sect[3].type = f32(3) -- .strtab + o.sect[3].ofs = fofs(sofs + ofs) + o.sect[3].size = fofs(#symname+1) + ffi.copy(o.space+ofs+1, symname) + ofs = ofs + #symname + 2 + o.sect[4].type = f32(1) -- .rodata + o.sect[4].flags = fofs(2) + o.sect[4].ofs = fofs(sofs + ofs) + o.sect[4].size = fofs(#s) + o.sect[5].type = f32(1) -- .note.GNU-stack + o.sect[5].ofs = fofs(sofs + ofs + #s) + + -- Write ELF object file. + local fp = savefile(output, "wb") + fp:write(ffi.string(o, ffi.sizeof(o)-4096+ofs)) + bcsave_tail(fp, output, s) +end + +local function bcsave_peobj(ctx, output, s, ffi) + ffi.cdef[[ +typedef struct { + uint16_t arch, nsects; + uint32_t time, symtabofs, nsyms; + uint16_t opthdrsz, flags; +} PEheader; +typedef struct { + char name[8]; + uint32_t vsize, vaddr, size, ofs, relocofs, lineofs; + uint16_t nreloc, nline; + uint32_t flags; +} PEsection; +typedef struct __attribute((packed)) { + union { + char name[8]; + uint32_t nameref[2]; + }; + uint32_t value; + int16_t sect; + uint16_t type; + uint8_t scl, naux; +} PEsym; +typedef struct __attribute((packed)) { + uint32_t size; + uint16_t nreloc, nline; + uint32_t cksum; + uint16_t assoc; + uint8_t comdatsel, unused[3]; +} PEsymaux; +typedef struct { + PEheader hdr; + PEsection sect[2]; + // Must be an even number of symbol structs. + PEsym sym0; + PEsymaux sym0aux; + PEsym sym1; + PEsymaux sym1aux; + PEsym sym2; + PEsym sym3; + uint32_t strtabsize; + uint8_t space[4096]; +} PEobj; +]] + local symname = LJBC_PREFIX..ctx.modname + local is64 = false + if ctx.arch == "x86" then + symname = "_"..symname + elseif ctx.arch == "x64" then + is64 = true + end + local symexport = " /EXPORT:"..symname..",DATA " + + -- The file format is always little-endian. Swap if the host is big-endian. + local function f32(x) return x end + local f16 = f32 + if ffi.abi("be") then + f32 = bit.bswap + function f16(x) return bit.rshift(bit.bswap(x), 16) end + end + + -- Create PE object and fill in header. + local o = ffi.new("PEobj") + local hdr = o.hdr + hdr.arch = f16(({ x86=0x14c, x64=0x8664, arm=0x1c0, ppc=0x1f2, mips=0x366, mipsel=0x366 })[ctx.arch]) + hdr.nsects = f16(2) + hdr.symtabofs = f32(ffi.offsetof(o, "sym0")) + hdr.nsyms = f32(6) + + -- Fill in sections and symbols. + o.sect[0].name = ".drectve" + o.sect[0].size = f32(#symexport) + o.sect[0].flags = f32(0x00100a00) + o.sym0.sect = f16(1) + o.sym0.scl = 3 + o.sym0.name = ".drectve" + o.sym0.naux = 1 + o.sym0aux.size = f32(#symexport) + o.sect[1].name = ".rdata" + o.sect[1].size = f32(#s) + o.sect[1].flags = f32(0x40300040) + o.sym1.sect = f16(2) + o.sym1.scl = 3 + o.sym1.name = ".rdata" + o.sym1.naux = 1 + o.sym1aux.size = f32(#s) + o.sym2.sect = f16(2) + o.sym2.scl = 2 + o.sym2.nameref[1] = f32(4) + o.sym3.sect = f16(-1) + o.sym3.scl = 2 + o.sym3.value = f32(1) + o.sym3.name = "@feat.00" -- Mark as SafeSEH compliant. + ffi.copy(o.space, symname) + local ofs = #symname + 1 + o.strtabsize = f32(ofs + 4) + o.sect[0].ofs = f32(ffi.offsetof(o, "space") + ofs) + ffi.copy(o.space + ofs, symexport) + ofs = ofs + #symexport + o.sect[1].ofs = f32(ffi.offsetof(o, "space") + ofs) + + -- Write PE object file. + local fp = savefile(output, "wb") + fp:write(ffi.string(o, ffi.sizeof(o)-4096+ofs)) + bcsave_tail(fp, output, s) +end + +local function bcsave_machobj(ctx, output, s, ffi) + check(false, "NYI: no support for writing OSX object files") +end + +local function bcsave_obj(ctx, output, s) + local ok, ffi = pcall(require, "ffi") + check(ok, "FFI library required to write this file type") + if ctx.os == "windows" then + return bcsave_peobj(ctx, output, s, ffi) + elseif ctx.os == "osx" then + return bcsave_machobj(ctx, output, s, ffi) + else + return bcsave_elfobj(ctx, output, s, ffi) + end +end + +------------------------------------------------------------------------------ + +local function bclist(input, output) + local f = readfile(input) + require("jit.bc").dump(f, savefile(output, "w"), true) +end + +local function bcsave(ctx, input, output) + local f = readfile(input) + local s = string.dump(f, ctx.strip) + local t = ctx.type + if not t then + t = detecttype(output) + ctx.type = t + end + if t == "raw" then + bcsave_raw(output, s) + else + if not ctx.modname then ctx.modname = detectmodname(input) end + if t == "obj" then + bcsave_obj(ctx, output, s) + else + bcsave_c(ctx, output, s) + end + end +end + +local function docmd(...) + local arg = {...} + local n = 1 + local list = false + local ctx = { + strip = true, arch = jit.arch, os = string.lower(jit.os), + type = false, modname = false, + } + while n <= #arg do + local a = arg[n] + if type(a) == "string" and string.sub(a, 1, 1) == "-" and a ~= "-" then + table.remove(arg, n) + if a == "--" then break end + for m=2,#a do + local opt = string.sub(a, m, m) + if opt == "l" then + list = true + elseif opt == "s" then + ctx.strip = true + elseif opt == "g" then + ctx.strip = false + else + if arg[n] == nil or m ~= #a then usage() end + if opt == "e" then + if n ~= 1 then usage() end + arg[1] = check(loadstring(arg[1])) + elseif opt == "n" then + ctx.modname = checkmodname(table.remove(arg, n)) + elseif opt == "t" then + ctx.type = checkarg(table.remove(arg, n), map_type, "file type") + elseif opt == "a" then + ctx.arch = checkarg(table.remove(arg, n), map_arch, "architecture") + elseif opt == "o" then + ctx.os = checkarg(table.remove(arg, n), map_os, "OS name") + else + usage() + end + end + end + else + n = n + 1 + end + end + if list then + if #arg == 0 or #arg > 2 then usage() end + bclist(arg[1], arg[2] or "-") + else + if #arg ~= 2 then usage() end + bcsave(ctx, arg[1], arg[2]) + end +end + +------------------------------------------------------------------------------ + +-- Public module functions. +module(...) + +start = docmd -- Process -b command line option. + diff --git a/src/jit/dis_arm.lua b/src/jit/dis_arm.lua new file mode 100644 index 00000000..0fcd1bed --- /dev/null +++ b/src/jit/dis_arm.lua @@ -0,0 +1,543 @@ +---------------------------------------------------------------------------- +-- LuaJIT ARM disassembler module. +-- +-- Copyright (C) 2005-2012 Mike Pall. All rights reserved. +-- Released under the MIT license. See Copyright Notice in luajit.h +---------------------------------------------------------------------------- +-- This is a helper module used by the LuaJIT machine code dumper module. +-- +-- It disassembles most user-mode ARMv7 instructions +-- NYI: Advanced SIMD and VFP instructions. +------------------------------------------------------------------------------ + +local type = type +local sub, byte, format = string.sub, string.byte, string.format +local match, gmatch, gsub = string.match, string.gmatch, string.gsub +local concat = table.concat +local bit = require("bit") +local band, bor, ror, tohex = bit.band, bit.bor, bit.ror, bit.tohex +local lshift, rshift, arshift = bit.lshift, bit.rshift, bit.arshift + +------------------------------------------------------------------------------ +-- Opcode maps +------------------------------------------------------------------------------ + +local map_loadc = { + shift = 9, mask = 7, + [5] = { + shift = 0, mask = 0 -- NYI VFP load/store. + }, + _ = { + shift = 0, mask = 0 -- NYI ldc, mcrr, mrrc. + }, +} + +local map_datac = { + shift = 24, mask = 1, + [0] = { + shift = 9, mask = 7, + [5] = { + shift = 0, mask = 0 -- NYI VFP data. + }, + _ = { + shift = 0, mask = 0 -- NYI cdp, mcr, mrc. + }, + }, + "svcT", +} + +local map_loadcu = { + shift = 0, mask = 0, -- NYI unconditional CP load/store. +} + +local map_datacu = { + shift = 0, mask = 0, -- NYI unconditional CP data. +} + +local map_simddata = { + shift = 0, mask = 0, -- NYI SIMD data. +} + +local map_simdload = { + shift = 0, mask = 0, -- NYI SIMD load/store, preload. +} + +local map_preload = { + shift = 0, mask = 0, -- NYI preload. +} + +local map_media = { + shift = 20, mask = 31, + [0] = false, + { --01 + shift = 5, mask = 7, + [0] = "sadd16DNM", "sasxDNM", "ssaxDNM", "ssub16DNM", + "sadd8DNM", false, false, "ssub8DNM", + }, + { --02 + shift = 5, mask = 7, + [0] = "qadd16DNM", "qasxDNM", "qsaxDNM", "qsub16DNM", + "qadd8DNM", false, false, "qsub8DNM", + }, + { --03 + shift = 5, mask = 7, + [0] = "shadd16DNM", "shasxDNM", "shsaxDNM", "shsub16DNM", + "shadd8DNM", false, false, "shsub8DNM", + }, + false, + { --05 + shift = 5, mask = 7, + [0] = "uadd16DNM", "uasxDNM", "usaxDNM", "usub16DNM", + "uadd8DNM", false, false, "usub8DNM", + }, + { --06 + shift = 5, mask = 7, + [0] = "uqadd16DNM", "uqasxDNM", "uqsaxDNM", "uqsub16DNM", + "uqadd8DNM", false, false, "uqsub8DNM", + }, + { --07 + shift = 5, mask = 7, + [0] = "uhadd16DNM", "uhasxDNM", "uhsaxDNM", "uhsub16DNM", + "uhadd8DNM", false, false, "uhsub8DNM", + }, + { --08 + shift = 5, mask = 7, + [0] = "pkhbtDNMU", false, "pkhtbDNMU", + { shift = 16, mask = 15, [15] = "sxtb16DMU", _ = "sxtab16DNMU", }, + "pkhbtDNMU", "selDNM", "pkhtbDNMU", + }, + false, + { --0a + shift = 5, mask = 7, + [0] = "ssatDxMu", "ssat16DxM", "ssatDxMu", + { shift = 16, mask = 15, [15] = "sxtbDMU", _ = "sxtabDNMU", }, + "ssatDxMu", false, "ssatDxMu", + }, + { --0b + shift = 5, mask = 7, + [0] = "ssatDxMu", "revDM", "ssatDxMu", + { shift = 16, mask = 15, [15] = "sxthDMU", _ = "sxtahDNMU", }, + "ssatDxMu", "rev16DM", "ssatDxMu", + }, + { --0c + shift = 5, mask = 7, + [3] = { shift = 16, mask = 15, [15] = "uxtb16DMU", _ = "uxtab16DNMU", }, + }, + false, + { --0e + shift = 5, mask = 7, + [0] = "usatDwMu", "usat16DwM", "usatDwMu", + { shift = 16, mask = 15, [15] = "uxtbDMU", _ = "uxtabDNMU", }, + "usatDwMu", false, "usatDwMu", + }, + { --0f + shift = 5, mask = 7, + [0] = "usatDwMu", "rbitDM", "usatDwMu", + { shift = 16, mask = 15, [15] = "uxthDMU", _ = "uxtahDNMU", }, + "usatDwMu", "revshDM", "usatDwMu", + }, + { --10 + shift = 12, mask = 15, + [15] = { + shift = 5, mask = 7, + "smuadNMS", "smuadxNMS", "smusdNMS", "smusdxNMS", + }, + _ = { + shift = 5, mask = 7, + [0] = "smladNMSD", "smladxNMSD", "smlsdNMSD", "smlsdxNMSD", + }, + }, + false, false, false, + { --14 + shift = 5, mask = 7, + [0] = "smlaldDNMS", "smlaldxDNMS", "smlsldDNMS", "smlsldxDNMS", + }, + { --15 + shift = 5, mask = 7, + [0] = { shift = 12, mask = 15, [15] = "smmulNMS", _ = "smmlaNMSD", }, + { shift = 12, mask = 15, [15] = "smmulrNMS", _ = "smmlarNMSD", }, + false, false, false, false, + "smmlsNMSD", "smmlsrNMSD", + }, + false, false, + { --18 + shift = 5, mask = 7, + [0] = { shift = 12, mask = 15, [15] = "usad8NMS", _ = "usada8NMSD", }, + }, + false, + { --1a + shift = 5, mask = 3, [2] = "sbfxDMvw", + }, + { --1b + shift = 5, mask = 3, [2] = "sbfxDMvw", + }, + { --1c + shift = 5, mask = 3, + [0] = { shift = 0, mask = 15, [15] = "bfcDvX", _ = "bfiDMvX", }, + }, + { --1d + shift = 5, mask = 3, + [0] = { shift = 0, mask = 15, [15] = "bfcDvX", _ = "bfiDMvX", }, + }, + { --1e + shift = 5, mask = 3, [2] = "ubfxDMvw", + }, + { --1f + shift = 5, mask = 3, [2] = "ubfxDMvw", + }, +} + +local map_load = { + shift = 21, mask = 9, + { + shift = 20, mask = 5, + [0] = "strtDL", "ldrtDL", [4] = "strbtDL", [5] = "ldrbtDL", + }, + _ = { + shift = 20, mask = 5, + [0] = "strDL", "ldrDL", [4] = "strbDL", [5] = "ldrbDL", + } +} + +local map_load1 = { + shift = 4, mask = 1, + [0] = map_load, map_media, +} + +local map_loadm = { + shift = 20, mask = 1, + [0] = { + shift = 23, mask = 3, + [0] = "stmdaNR", "stmNR", + { shift = 16, mask = 63, [45] = "pushR", _ = "stmdbNR", }, "stmibNR", + }, + { + shift = 23, mask = 3, + [0] = "ldmdaNR", { shift = 16, mask = 63, [61] = "popR", _ = "ldmNR", }, + "ldmdbNR", "ldmibNR", + }, +} + +local map_data = { + shift = 21, mask = 15, + [0] = "andDNPs", "eorDNPs", "subDNPs", "rsbDNPs", + "addDNPs", "adcDNPs", "sbcDNPs", "rscDNPs", + "tstNP", "teqNP", "cmpNP", "cmnNP", + "orrDNPs", "movDPs", "bicDNPs", "mvnDPs", +} + +local map_mul = { + shift = 21, mask = 7, + [0] = "mulNMSs", "mlaNMSDs", "umaalDNMS", "mlsDNMS", + "umullDNMSs", "umlalDNMSs", "smullDNMSs", "smlalDNMSs", +} + +local map_sync = { + shift = 20, mask = 15, -- NYI: brackets around N. R(D+1) for ldrexd/strexd. + [0] = "swpDMN", false, false, false, + "swpbDMN", false, false, false, + "strexDMN", "ldrexDN", "strexdDN", "ldrexdDN", + "strexbDMN", "ldrexbDN", "strexhDN", "ldrexhDN", +} + +local map_mulh = { + shift = 21, mask = 3, + [0] = { shift = 5, mask = 3, + [0] = "smlabbNMSD", "smlatbNMSD", "smlabtNMSD", "smlattNMSD", }, + { shift = 5, mask = 3, + [0] = "smlawbNMSD", "smulwbNMS", "smlawtNMSD", "smulwtNMS", }, + { shift = 5, mask = 3, + [0] = "smlalbbDNMS", "smlaltbDNMS", "smlalbtDNMS", "smlalttDNMS", }, + { shift = 5, mask = 3, + [0] = "smulbbNMS", "smultbNMS", "smulbtNMS", "smulttNMS", }, +} + +local map_misc = { + shift = 4, mask = 7, + -- NYI: decode PSR bits of msr. + [0] = { shift = 21, mask = 1, [0] = "mrsD", "msrM", }, + { shift = 21, mask = 3, "bxM", false, "clzDM", }, + { shift = 21, mask = 3, "bxjM", }, + { shift = 21, mask = 3, "blxM", }, + false, + { shift = 21, mask = 3, [0] = "qaddDMN", "qsubDMN", "qdaddDMN", "qdsubDMN", }, + false, + { shift = 21, mask = 3, "bkptK", }, +} + +local map_datar = { + shift = 4, mask = 9, + [9] = { + shift = 5, mask = 3, + [0] = { shift = 24, mask = 1, [0] = map_mul, map_sync, }, + { shift = 20, mask = 1, [0] = "strhDL", "ldrhDL", }, + { shift = 20, mask = 1, [0] = "ldrdDL", "ldrsbDL", }, + { shift = 20, mask = 1, [0] = "strdDL", "ldrshDL", }, + }, + _ = { + shift = 20, mask = 25, + [16] = { shift = 7, mask = 1, [0] = map_misc, map_mulh, }, + _ = { + shift = 0, mask = 0xffffffff, + [bor(0xe1a00000)] = "nop", + _ = map_data, + } + }, +} + +local map_datai = { + shift = 20, mask = 31, -- NYI: decode PSR bits of msr. Decode imm12. + [16] = "movwDW", [20] = "movtDW", + [18] = { shift = 0, mask = 0xf00ff, [0] = "nopv6", _ = "msrNW", }, + [22] = "msrNW", + _ = map_data, +} + +local map_branch = { + shift = 24, mask = 1, + [0] = "bB", "blB" +} + +local map_condins = { + [0] = map_datar, map_datai, map_load, map_load1, + map_loadm, map_branch, map_loadc, map_datac +} + +-- NYI: setend. +local map_uncondins = { + [0] = false, map_simddata, map_simdload, map_preload, + false, "blxB", map_loadcu, map_datacu, +} + +------------------------------------------------------------------------------ + +local map_gpr = { + [0] = "r0", "r1", "r2", "r3", "r4", "r5", "r6", "r7", + "r8", "r9", "r10", "r11", "r12", "sp", "lr", "pc", +} + +local map_cond = { + [0] = "eq", "ne", "hs", "lo", "mi", "pl", "vs", "vc", + "hi", "ls", "ge", "lt", "gt", "le", "al", +} + +local map_shift = { [0] = "lsl", "lsr", "asr", "ror", } + +------------------------------------------------------------------------------ + +-- Output a nicely formatted line with an opcode and operands. +local function putop(ctx, text, operands) + local pos = ctx.pos + local extra = "" + if ctx.rel then + local sym = ctx.symtab[ctx.rel] + if sym then + extra = "\t->"..sym + elseif band(ctx.op, 0x0e000000) ~= 0x0a000000 then + extra = "\t; 0x"..tohex(ctx.rel) + end + end + if ctx.hexdump > 0 then + ctx.out(format("%08x %s %-5s %s%s\n", + ctx.addr+pos, tohex(ctx.op), text, concat(operands, ", "), extra)) + else + ctx.out(format("%08x %-5s %s%s\n", + ctx.addr+pos, text, concat(operands, ", "), extra)) + end + ctx.pos = pos + 4 +end + +-- Fallback for unknown opcodes. +local function unknown(ctx) + return putop(ctx, ".long", { "0x"..tohex(ctx.op) }) +end + +-- Format operand 2 of load/store opcodes. +local function fmtload(ctx, op, pos) + local base = map_gpr[band(rshift(op, 16), 15)] + local x, ofs + local ext = (band(op, 0x04000000) == 0) + if not ext and band(op, 0x02000000) == 0 then + ofs = band(op, 4095) + if band(op, 0x00800000) == 0 then ofs = -ofs end + if base == "pc" then ctx.rel = ctx.addr + pos + 8 + ofs end + ofs = "#"..ofs + elseif ext and band(op, 0x00400000) ~= 0 then + ofs = band(op, 15) + band(rshift(op, 4), 0xf0) + if band(op, 0x00800000) == 0 then ofs = -ofs end + if base == "pc" then ctx.rel = ctx.addr + pos + 8 + ofs end + ofs = "#"..ofs + else + ofs = map_gpr[band(op, 15)] + if ext or band(op, 0xfe0) == 0 then + elseif band(op, 0xfe0) == 0x60 then + ofs = format("%s, rrx", ofs) + else + local sh = band(rshift(op, 7), 31) + if sh == 0 then sh = 32 end + ofs = format("%s, %s #%d", ofs, map_shift[band(rshift(op, 5), 3)], sh) + end + if band(op, 0x00800000) == 0 then ofs = "-"..ofs end + end + if ofs == "#0" then + x = format("[%s]", base) + elseif band(op, 0x01000000) == 0 then + x = format("[%s], %s", base, ofs) + else + x = format("[%s, %s]", base, ofs) + end + if band(op, 0x01200000) == 0x01200000 then x = x.."!" end + return x +end + +-- Disassemble a single instruction. +local function disass_ins(ctx) + local pos = ctx.pos + local b0, b1, b2, b3 = byte(ctx.code, pos+1, pos+4) + local op = bor(lshift(b3, 24), lshift(b2, 16), lshift(b1, 8), b0) + local operands = {} + local suffix = "" + local last, name, pat + ctx.op = op + ctx.rel = nil + + local cond = rshift(op, 28) + local opat + if cond == 15 then + opat = map_uncondins[band(rshift(op, 25), 7)] + else + if cond ~= 14 then suffix = map_cond[cond] end + opat = map_condins[band(rshift(op, 25), 7)] + end + while type(opat) ~= "string" do + if not opat then return unknown(ctx) end + opat = opat[band(rshift(op, opat.shift), opat.mask)] or opat._ + end + name, pat = match(opat, "^([a-z0-9]*)(.*)") + + for p in gmatch(pat, ".") do + local x = nil + if p == "D" then + x = map_gpr[band(rshift(op, 12), 15)] + elseif p == "N" then + x = map_gpr[band(rshift(op, 16), 15)] + elseif p == "S" then + x = map_gpr[band(rshift(op, 8), 15)] + elseif p == "M" then + x = map_gpr[band(op, 15)] + elseif p == "P" then + if band(op, 0x02000000) ~= 0 then + x = ror(band(op, 255), 2*band(rshift(op, 8), 15)) + else + x = map_gpr[band(op, 15)] + if band(op, 0xff0) ~= 0 then + operands[#operands+1] = x + local s = map_shift[band(rshift(op, 5), 3)] + local r = nil + if band(op, 0xf90) == 0 then + if s == "ror" then s = "rrx" else r = "#32" end + elseif band(op, 0x10) == 0 then + r = "#"..band(rshift(op, 7), 31) + else + r = map_gpr[band(rshift(op, 8), 15)] + end + if name == "mov" then name = s; x = r + elseif r then x = format("%s %s", s, r) + else x = s end + end + end + elseif p == "L" then + x = fmtload(ctx, op, pos, false) + elseif p == "B" then + local addr = ctx.addr + pos + 8 + arshift(lshift(op, 8), 6) + if cond == 15 then addr = addr + band(rshift(op, 23), 2) end + ctx.rel = addr + x = "0x"..tohex(addr) + elseif p == "R" then + if band(op, 0x00200000) ~= 0 and #operands == 1 then + operands[1] = operands[1].."!" + end + local t = {} + for i=0,15 do + if band(rshift(op, i), 1) == 1 then t[#t+1] = map_gpr[i] end + end + x = "{"..concat(t, ", ").."}" + elseif p == "W" then + x = band(op, 0x0fff) + band(rshift(op, 4), 0xf000) + elseif p == "T" then + x = "#0x"..tohex(band(op, 0x00ffffff), 6) + elseif p == "U" then + x = band(rshift(op, 7), 31) + if x == 0 then x = nil end + elseif p == "u" then + x = band(rshift(op, 7), 31) + if band(op, 0x40) == 0 then + if x == 0 then x = nil else x = "lsl #"..x end + else + if x == 0 then x = "asr #32" else x = "asr #"..x end + end + elseif p == "v" then + x = band(rshift(op, 7), 31) + elseif p == "w" then + x = band(rshift(op, 16), 31) + elseif p == "x" then + x = band(rshift(op, 16), 31) + 1 + elseif p == "X" then + x = band(rshift(op, 16), 31) - last + 1 + elseif p == "K" then + x = "#0x"..tohex(band(rshift(op, 4), 0x0000fff0) + band(op, 15), 4) + elseif p == "s" then + if band(op, 0x00100000) ~= 0 then suffix = "s"..suffix end + else + assert(false) + end + if x then + last = x + if type(x) == "number" then x = "#"..x end + operands[#operands+1] = x + end + end + + return putop(ctx, name..suffix, operands) +end + +------------------------------------------------------------------------------ + +-- Disassemble a block of code. +local function disass_block(ctx, ofs, len) + if not ofs then ofs = 0 end + local stop = len and ofs+len or #ctx.code + ctx.pos = ofs + ctx.rel = nil + while ctx.pos < stop do disass_ins(ctx) end +end + +-- Extended API: create a disassembler context. Then call ctx:disass(ofs, len). +local function create_(code, addr, out) + local ctx = {} + ctx.code = code + ctx.addr = addr or 0 + ctx.out = out or io.write + ctx.symtab = {} + ctx.disass = disass_block + ctx.hexdump = 8 + return ctx +end + +-- Simple API: disassemble code (a string) at address and output via out. +local function disass_(code, addr, out) + create_(code, addr, out):disass() +end + +-- Return register name for RID. +local function regname_(r) + return map_gpr[r] +end + +-- Public module functions. +module(...) + +create = create_ +disass = disass_ +regname = regname_ + diff --git a/src/jit/dis_mips.lua b/src/jit/dis_mips.lua new file mode 100644 index 00000000..2edfdf40 --- /dev/null +++ b/src/jit/dis_mips.lua @@ -0,0 +1,428 @@ +---------------------------------------------------------------------------- +-- LuaJIT MIPS disassembler module. +-- +-- Copyright (C) 2005-2011 Mike Pall. All rights reserved. +-- Released under the MIT/X license. See Copyright Notice in luajit.h +---------------------------------------------------------------------------- +-- This is a helper module used by the LuaJIT machine code dumper module. +-- +-- It disassembles all standard MIPS32R1/R2 instructions. +-- Default mode is big-endian, but see: dis_mipsel.lua +------------------------------------------------------------------------------ + +local type = type +local sub, byte, format = string.sub, string.byte, string.format +local match, gmatch, gsub = string.match, string.gmatch, string.gsub +local concat = table.concat +local bit = require("bit") +local band, bor, tohex = bit.band, bit.bor, bit.tohex +local lshift, rshift, arshift = bit.lshift, bit.rshift, bit.arshift + +------------------------------------------------------------------------------ +-- Primary and extended opcode maps +------------------------------------------------------------------------------ + +local map_movci = { shift = 16, mask = 1, [0] = "movfDSC", "movtDSC", } +local map_srl = { shift = 21, mask = 1, [0] = "srlDTA", "rotrDTA", } +local map_srlv = { shift = 6, mask = 1, [0] = "srlvDTS", "rotrvDTS", } + +local map_special = { + shift = 0, mask = 63, + [0] = { shift = 0, mask = -1, [0] = "nop", _ = "sllDTA" }, + map_movci, map_srl, "sraDTA", + "sllvDTS", false, map_srlv, "sravDTS", + "jrS", "jalrD1S", "movzDST", "movnDST", + "syscallY", "breakY", false, "sync", + "mfhiD", "mthiS", "mfloD", "mtloS", + false, false, false, false, + "multST", "multuST", "divST", "divuST", + false, false, false, false, + "addDST", "addu|moveDST0", "subDST", "subu|neguDS0T", + "andDST", "orDST", "xorDST", "nor|notDST0", + false, false, "sltDST", "sltuDST", + false, false, false, false, + "tgeSTZ", "tgeuSTZ", "tltSTZ", "tltuSTZ", + "teqSTZ", false, "tneSTZ", +} + +local map_special2 = { + shift = 0, mask = 63, + [0] = "maddST", "madduST", "mulDST", false, + "msubST", "msubuST", + [32] = "clzDS", [33] = "cloDS", + [63] = "sdbbpY", +} + +local map_bshfl = { + shift = 6, mask = 31, + [2] = "wsbhDT", + [16] = "sebDT", + [24] = "sehDT", +} + +local map_special3 = { + shift = 0, mask = 63, + [0] = "extTSAK", [4] = "insTSAL", + [32] = map_bshfl, + [59] = "rdhwrTD", +} + +local map_regimm = { + shift = 16, mask = 31, + [0] = "bltzSB", "bgezSB", "bltzlSB", "bgezlSB", + false, false, false, false, + "tgeiSI", "tgeiuSI", "tltiSI", "tltiuSI", + "teqiSI", false, "tneiSI", false, + "bltzalSB", "bgezalSB", "bltzallSB", "bgezallSB", + false, false, false, false, + false, false, false, false, + false, false, false, "synciSO", +} + +local map_cop0 = { + shift = 25, mask = 1, + [0] = { + shift = 21, mask = 15, + [0] = "mfc0TDW", [4] = "mtc0TDW", + [10] = "rdpgprDT", + [11] = { shift = 5, mask = 1, [0] = "diT0", "eiT0", }, + [14] = "wrpgprDT", + }, { + shift = 0, mask = 63, + [1] = "tlbr", [2] = "tlbwi", [6] = "tlbwr", [8] = "tlbp", + [24] = "eret", [31] = "deret", + [32] = "wait", + }, +} + +local map_cop1s = { + shift = 0, mask = 63, + [0] = "add.sFGH", "sub.sFGH", "mul.sFGH", "div.sFGH", + "sqrt.sFG", "abs.sFG", "mov.sFG", "neg.sFG", + "round.l.sFG", "trunc.l.sFG", "ceil.l.sFG", "floor.l.sFG", + "round.w.sFG", "trunc.w.sFG", "ceil.w.sFG", "floor.w.sFG", + false, + { shift = 16, mask = 1, [0] = "movf.sFGC", "movt.sFGC" }, + "movz.sFGT", "movn.sFGT", + false, "recip.sFG", "rsqrt.sFG", false, + false, false, false, false, + false, false, false, false, + false, "cvt.d.sFG", false, false, + "cvt.w.sFG", "cvt.l.sFG", "cvt.ps.sFGH", false, + false, false, false, false, + false, false, false, false, + "c.f.sVGH", "c.un.sVGH", "c.eq.sVGH", "c.ueq.sVGH", + "c.olt.sVGH", "c.ult.sVGH", "c.ole.sVGH", "c.ule.sVGH", + "c.sf.sVGH", "c.ngle.sVGH", "c.seq.sVGH", "c.ngl.sVGH", + "c.lt.sVGH", "c.nge.sVGH", "c.le.sVGH", "c.ngt.sVGH", +} + +local map_cop1d = { + shift = 0, mask = 63, + [0] = "add.dFGH", "sub.dFGH", "mul.dFGH", "div.dFGH", + "sqrt.dFG", "abs.dFG", "mov.dFG", "neg.dFG", + "round.l.dFG", "trunc.l.dFG", "ceil.l.dFG", "floor.l.dFG", + "round.w.dFG", "trunc.w.dFG", "ceil.w.dFG", "floor.w.dFG", + false, + { shift = 16, mask = 1, [0] = "movf.dFGC", "movt.dFGC" }, + "movz.dFGT", "movn.dFGT", + false, "recip.dFG", "rsqrt.dFG", false, + false, false, false, false, + false, false, false, false, + "cvt.s.dFG", false, false, false, + "cvt.w.dFG", "cvt.l.dFG", false, false, + false, false, false, false, + false, false, false, false, + "c.f.dVGH", "c.un.dVGH", "c.eq.dVGH", "c.ueq.dVGH", + "c.olt.dVGH", "c.ult.dVGH", "c.ole.dVGH", "c.ule.dVGH", + "c.df.dVGH", "c.ngle.dVGH", "c.deq.dVGH", "c.ngl.dVGH", + "c.lt.dVGH", "c.nge.dVGH", "c.le.dVGH", "c.ngt.dVGH", +} + +local map_cop1ps = { + shift = 0, mask = 63, + [0] = "add.psFGH", "sub.psFGH", "mul.psFGH", false, + false, "abs.psFG", "mov.psFG", "neg.psFG", + false, false, false, false, + false, false, false, false, + false, + { shift = 16, mask = 1, [0] = "movf.psFGC", "movt.psFGC" }, + "movz.psFGT", "movn.psFGT", + false, false, false, false, + false, false, false, false, + false, false, false, false, + "cvt.s.puFG", false, false, false, + false, false, false, false, + "cvt.s.plFG", false, false, false, + "pll.psFGH", "plu.psFGH", "pul.psFGH", "puu.psFGH", + "c.f.psVGH", "c.un.psVGH", "c.eq.psVGH", "c.ueq.psVGH", + "c.olt.psVGH", "c.ult.psVGH", "c.ole.psVGH", "c.ule.psVGH", + "c.psf.psVGH", "c.ngle.psVGH", "c.pseq.psVGH", "c.ngl.psVGH", + "c.lt.psVGH", "c.nge.psVGH", "c.le.psVGH", "c.ngt.psVGH", +} + +local map_cop1w = { + shift = 0, mask = 63, + [32] = "cvt.s.wFG", [33] = "cvt.d.wFG", +} + +local map_cop1l = { + shift = 0, mask = 63, + [32] = "cvt.s.lFG", [33] = "cvt.d.lFG", +} + +local map_cop1bc = { + shift = 16, mask = 3, + [0] = "bc1fCB", "bc1tCB", "bc1flCB", "bc1tlCB", +} + +local map_cop1 = { + shift = 21, mask = 31, + [0] = "mfc1TG", false, "cfc1TG", "mfhc1TG", + "mtc1TG", false, "ctc1TG", "mthc1TG", + map_cop1bc, false, false, false, + false, false, false, false, + map_cop1s, map_cop1d, false, false, + map_cop1w, map_cop1l, map_cop1ps, +} + +local map_cop1x = { + shift = 0, mask = 63, + [0] = "lwxc1FSX", "ldxc1FSX", false, false, + false, "luxc1FSX", false, false, + "swxc1FSX", "sdxc1FSX", false, false, + false, "suxc1FSX", false, "prefxMSX", + false, false, false, false, + false, false, false, false, + false, false, false, false, + false, false, "alnv.psFGHS", false, + "madd.sFRGH", "madd.dFRGH", false, false, + false, false, "madd.psFRGH", false, + "msub.sFRGH", "msub.dFRGH", false, false, + false, false, "msub.psFRGH", false, + "nmadd.sFRGH", "nmadd.dFRGH", false, false, + false, false, "nmadd.psFRGH", false, + "nmsub.sFRGH", "nmsub.dFRGH", false, false, + false, false, "nmsub.psFRGH", false, +} + +local map_pri = { + [0] = map_special, map_regimm, "jJ", "jalJ", + "beq|beqz|bST00B", "bne|bnezST0B", "blezSB", "bgtzSB", + "addiTSI", "addiu|liTS0I", "sltiTSI", "sltiuTSI", + "andiTSU", "ori|liTS0U", "xoriTSU", "luiTU", + map_cop0, map_cop1, false, map_cop1x, + "beql|beqzlST0B", "bnel|bnezlST0B", "blezlSB", "bgtzlSB", + false, false, false, false, + map_special2, false, false, map_special3, + "lbTSO", "lhTSO", "lwlTSO", "lwTSO", + "lbuTSO", "lhuTSO", "lwrTSO", false, + "sbTSO", "shTSO", "swlTSO", "swTSO", + false, false, "swrTSO", "cacheNSO", + "llTSO", "lwc1HSO", "lwc2TSO", "prefNSO", + false, "ldc1HSO", "ldc2TSO", false, + "scTSO", "swc1HSO", "swc2TSO", false, + false, "sdc1HSO", "sdc2TSO", false, +} + +------------------------------------------------------------------------------ + +local map_gpr = { + [0] = "r0", "r1", "r2", "r3", "r4", "r5", "r6", "r7", + "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15", + "r16", "r17", "r18", "r19", "r20", "r21", "r22", "r23", + "r24", "r25", "r26", "r27", "r28", "sp", "r30", "ra", +} + +------------------------------------------------------------------------------ + +-- Output a nicely formatted line with an opcode and operands. +local function putop(ctx, text, operands) + local pos = ctx.pos + local extra = "" + if ctx.rel then + local sym = ctx.symtab[ctx.rel] + if sym then extra = "\t->"..sym end + end + if ctx.hexdump > 0 then + ctx.out(format("%08x %s %-7s %s%s\n", + ctx.addr+pos, tohex(ctx.op), text, concat(operands, ", "), extra)) + else + ctx.out(format("%08x %-7s %s%s\n", + ctx.addr+pos, text, concat(operands, ", "), extra)) + end + ctx.pos = pos + 4 +end + +-- Fallback for unknown opcodes. +local function unknown(ctx) + return putop(ctx, ".long", { "0x"..tohex(ctx.op) }) +end + +local function get_be(ctx) + local pos = ctx.pos + local b0, b1, b2, b3 = byte(ctx.code, pos+1, pos+4) + return bor(lshift(b0, 24), lshift(b1, 16), lshift(b2, 8), b3) +end + +local function get_le(ctx) + local pos = ctx.pos + local b0, b1, b2, b3 = byte(ctx.code, pos+1, pos+4) + return bor(lshift(b3, 24), lshift(b2, 16), lshift(b1, 8), b0) +end + +-- Disassemble a single instruction. +local function disass_ins(ctx) + local op = ctx:get() + local operands = {} + local last = nil + ctx.op = op + ctx.rel = nil + + local opat = map_pri[rshift(op, 26)] + while type(opat) ~= "string" do + if not opat then return unknown(ctx) end + opat = opat[band(rshift(op, opat.shift), opat.mask)] or opat._ + end + local name, pat = match(opat, "^([a-z0-9_.]*)(.*)") + local altname, pat2 = match(pat, "|([a-z0-9_.|]*)(.*)") + if altname then pat = pat2 end + + for p in gmatch(pat, ".") do + local x = nil + if p == "S" then + x = map_gpr[band(rshift(op, 21), 31)] + elseif p == "T" then + x = map_gpr[band(rshift(op, 16), 31)] + elseif p == "D" then + x = map_gpr[band(rshift(op, 11), 31)] + elseif p == "F" then + x = "f"..band(rshift(op, 6), 31) + elseif p == "G" then + x = "f"..band(rshift(op, 11), 31) + elseif p == "H" then + x = "f"..band(rshift(op, 16), 31) + elseif p == "R" then + x = "f"..band(rshift(op, 21), 31) + elseif p == "A" then + x = band(rshift(op, 6), 31) + elseif p == "M" then + x = band(rshift(op, 11), 31) + elseif p == "N" then + x = band(rshift(op, 16), 31) + elseif p == "C" then + x = band(rshift(op, 18), 7) + if x == 0 then x = nil end + elseif p == "K" then + x = band(rshift(op, 11), 31) + 1 + elseif p == "L" then + x = band(rshift(op, 11), 31) - last + 1 + elseif p == "I" then + x = arshift(lshift(op, 16), 16) + elseif p == "U" then + x = band(op, 0xffff) + elseif p == "O" then + local disp = arshift(lshift(op, 16), 16) + operands[#operands] = format("%d(%s)", disp, last) + elseif p == "X" then + local index = map_gpr[band(rshift(op, 16), 31)] + operands[#operands] = format("%s(%s)", index, last) + elseif p == "B" then + x = ctx.addr + ctx.pos + arshift(lshift(op, 16), 16)*4 + 4 + ctx.rel = x + x = "0x"..tohex(x) + elseif p == "J" then + x = band(ctx.addr + ctx.pos, 0xf0000000) + band(op, 0x03ffffff)*4 + ctx.rel = x + x = "0x"..tohex(x) + elseif p == "V" then + x = band(rshift(op, 8), 7) + if x == 0 then x = nil end + elseif p == "W" then + x = band(op, 7) + if x == 0 then x = nil end + elseif p == "Y" then + x = band(rshift(op, 6), 0x000fffff) + if x == 0 then x = nil end + elseif p == "Z" then + x = band(rshift(op, 6), 1023) + if x == 0 then x = nil end + elseif p == "0" then + if last == "r0" or last == 0 then + local n = #operands + operands[n] = nil + last = operands[n-1] + if altname then + local a1, a2 = match(altname, "([^|]*)|(.*)") + if a1 then name, altname = a1, a2 + else name = altname end + end + end + elseif p == "1" then + if last == "ra" then + operands[#operands] = nil + end + else + assert(false) + end + if x then operands[#operands+1] = x; last = x end + end + + return putop(ctx, name, operands) +end + +------------------------------------------------------------------------------ + +-- Disassemble a block of code. +local function disass_block(ctx, ofs, len) + if not ofs then ofs = 0 end + local stop = len and ofs+len or #ctx.code + stop = stop - stop % 4 + ctx.pos = ofs - ofs % 4 + ctx.rel = nil + while ctx.pos < stop do disass_ins(ctx) end +end + +-- Extended API: create a disassembler context. Then call ctx:disass(ofs, len). +local function create_(code, addr, out) + local ctx = {} + ctx.code = code + ctx.addr = addr or 0 + ctx.out = out or io.write + ctx.symtab = {} + ctx.disass = disass_block + ctx.hexdump = 8 + ctx.get = get_be + return ctx +end + +local function create_el_(code, addr, out) + local ctx = create_(code, addr, out) + ctx.get = get_le + return ctx +end + +-- Simple API: disassemble code (a string) at address and output via out. +local function disass_(code, addr, out) + create_(code, addr, out):disass() +end + +local function disass_el_(code, addr, out) + create_el_(code, addr, out):disass() +end + +-- Return register name for RID. +local function regname_(r) + if r < 32 then return map_gpr[r] end + return "f"..(r-32) +end + +-- Public module functions. +module(...) + +create = create_ +create_el = create_el_ +disass = disass_ +disass_el = disass_el_ +regname = regname_ + diff --git a/src/jit/dis_mipsel.lua b/src/jit/dis_mipsel.lua new file mode 100644 index 00000000..4c5a6515 --- /dev/null +++ b/src/jit/dis_mipsel.lua @@ -0,0 +1,20 @@ +---------------------------------------------------------------------------- +-- LuaJIT MIPSEL disassembler wrapper module. +-- +-- Copyright (C) 2005-2011 Mike Pall. All rights reserved. +-- Released under the MIT license. See Copyright Notice in luajit.h +---------------------------------------------------------------------------- +-- This module just exports the little-endian functions from the +-- MIPS disassembler module. All the interesting stuff is there. +------------------------------------------------------------------------------ + +local require = require + +module(...) + +local dis_mips = require(_PACKAGE.."dis_mips") + +create = dis_mips.create_el +disass = dis_mips.disass_el +regname = dis_mips.regname + diff --git a/src/jit/dis_ppc.lua b/src/jit/dis_ppc.lua new file mode 100644 index 00000000..ad8599fc --- /dev/null +++ b/src/jit/dis_ppc.lua @@ -0,0 +1,591 @@ +---------------------------------------------------------------------------- +-- LuaJIT PPC disassembler module. +-- +-- Copyright (C) 2005-2012 Mike Pall. All rights reserved. +-- Released under the MIT/X license. See Copyright Notice in luajit.h +---------------------------------------------------------------------------- +-- This is a helper module used by the LuaJIT machine code dumper module. +-- +-- It disassembles all common, non-privileged 32/64 bit PowerPC instructions +-- plus the e500 SPE instructions and some Cell/Xenon extensions. +-- +-- NYI: VMX, VMX128 +------------------------------------------------------------------------------ + +local type = type +local sub, byte, format = string.sub, string.byte, string.format +local match, gmatch, gsub = string.match, string.gmatch, string.gsub +local concat = table.concat +local bit = require("bit") +local band, bor, tohex = bit.band, bit.bor, bit.tohex +local lshift, rshift, arshift = bit.lshift, bit.rshift, bit.arshift + +------------------------------------------------------------------------------ +-- Primary and extended opcode maps +------------------------------------------------------------------------------ + +local map_crops = { + shift = 1, mask = 1023, + [0] = "mcrfXX", + [33] = "crnor|crnotCCC=", [129] = "crandcCCC", + [193] = "crxor|crclrCCC%", [225] = "crnandCCC", + [257] = "crandCCC", [289] = "creqv|crsetCCC%", + [417] = "crorcCCC", [449] = "cror|crmoveCCC=", + [16] = "b_lrKB", [528] = "b_ctrKB", + [150] = "isync", +} + +local map_rlwinm = setmetatable({ + shift = 0, mask = -1, +}, +{ __index = function(t, x) + local rot = band(rshift(x, 11), 31) + local mb = band(rshift(x, 6), 31) + local me = band(rshift(x, 1), 31) + if mb == 0 and me == 31-rot then + return "slwiRR~A." + elseif me == 31 and mb == 32-rot then + return "srwiRR~-A." + else + return "rlwinmRR~AAA." + end + end +}) + +local map_rld = { + shift = 2, mask = 7, + [0] = "rldiclRR~HM.", "rldicrRR~HM.", "rldicRR~HM.", "rldimiRR~HM.", + { + shift = 1, mask = 1, + [0] = "rldclRR~HM.", "rldcrRR~HM.", + }, +} + +local map_ext = setmetatable({ + shift = 1, mask = 1023, + + [0] = "cmp_YLRR", [32] = "cmpl_YLRR", + [4] = "twARR", [68] = "tdARR", + + [8] = "subfcRRR.", [40] = "subfRRR.", + [104] = "negRR.", [136] = "subfeRRR.", + [200] = "subfzeRR.", [232] = "subfmeRR.", + [520] = "subfcoRRR.", [552] = "subfoRRR.", + [616] = "negoRR.", [648] = "subfeoRRR.", + [712] = "subfzeoRR.", [744] = "subfmeoRR.", + + [9] = "mulhduRRR.", [73] = "mulhdRRR.", [233] = "mulldRRR.", + [457] = "divduRRR.", [489] = "divdRRR.", + [745] = "mulldoRRR.", + [969] = "divduoRRR.", [1001] = "divdoRRR.", + + [10] = "addcRRR.", [138] = "addeRRR.", + [202] = "addzeRR.", [234] = "addmeRR.", [266] = "addRRR.", + [522] = "addcoRRR.", [650] = "addeoRRR.", + [714] = "addzeoRR.", [746] = "addmeoRR.", [778] = "addoRRR.", + + [11] = "mulhwuRRR.", [75] = "mulhwRRR.", [235] = "mullwRRR.", + [459] = "divwuRRR.", [491] = "divwRRR.", + [747] = "mullwoRRR.", + [971] = "divwouRRR.", [1003] = "divwoRRR.", + + [15] = "iselltRRR", [47] = "iselgtRRR", [79] = "iseleqRRR", + + [144] = { shift = 20, mask = 1, [0] = "mtcrfRZ~", "mtocrfRZ~", }, + [19] = { shift = 20, mask = 1, [0] = "mfcrR", "mfocrfRZ", }, + [371] = { shift = 11, mask = 1023, [392] = "mftbR", [424] = "mftbuR", }, + [339] = { + shift = 11, mask = 1023, + [32] = "mferR", [256] = "mflrR", [288] = "mfctrR", [16] = "mfspefscrR", + }, + [467] = { + shift = 11, mask = 1023, + [32] = "mtxerR", [256] = "mtlrR", [288] = "mtctrR", [16] = "mtspefscrR", + }, + + [20] = "lwarxRR0R", [84] = "ldarxRR0R", + + [21] = "ldxRR0R", [53] = "lduxRRR", + [149] = "stdxRR0R", [181] = "stduxRRR", + [341] = "lwaxRR0R", [373] = "lwauxRRR", + + [23] = "lwzxRR0R", [55] = "lwzuxRRR", + [87] = "lbzxRR0R", [119] = "lbzuxRRR", + [151] = "stwxRR0R", [183] = "stwuxRRR", + [215] = "stbxRR0R", [247] = "stbuxRRR", + [279] = "lhzxRR0R", [311] = "lhzuxRRR", + [343] = "lhaxRR0R", [375] = "lhauxRRR", + [407] = "sthxRR0R", [439] = "sthuxRRR", + + [54] = "dcbst-R0R", [86] = "dcbf-R0R", + [150] = "stwcxRR0R.", [214] = "stdcxRR0R.", + [246] = "dcbtst-R0R", [278] = "dcbt-R0R", + [310] = "eciwxRR0R", [438] = "ecowxRR0R", + [470] = "dcbi-RR", + + [598] = { + shift = 21, mask = 3, + [0] = "sync", "lwsync", "ptesync", + }, + [758] = "dcba-RR", + [854] = "eieio", [982] = "icbi-R0R", [1014] = "dcbz-R0R", + + [26] = "cntlzwRR~", [58] = "cntlzdRR~", + [122] = "popcntbRR~", + [154] = "prtywRR~", [186] = "prtydRR~", + + [28] = "andRR~R.", [60] = "andcRR~R.", [124] = "nor|notRR~R=.", + [284] = "eqvRR~R.", [316] = "xorRR~R.", + [412] = "orcRR~R.", [444] = "or|mrRR~R=.", [476] = "nandRR~R.", + [508] = "cmpbRR~R", + + [512] = "mcrxrX", + + [532] = "ldbrxRR0R", [660] = "stdbrxRR0R", + + [533] = "lswxRR0R", [597] = "lswiRR0A", + [661] = "stswxRR0R", [725] = "stswiRR0A", + + [534] = "lwbrxRR0R", [662] = "stwbrxRR0R", + [790] = "lhbrxRR0R", [918] = "sthbrxRR0R", + + [535] = "lfsxFR0R", [567] = "lfsuxFRR", + [599] = "lfdxFR0R", [631] = "lfduxFRR", + [663] = "stfsxFR0R", [695] = "stfsuxFRR", + [727] = "stfdxFR0R", [759] = "stfduxFR0R", + [855] = "lfiwaxFR0R", + [983] = "stfiwxFR0R", + + [24] = "slwRR~R.", + + [27] = "sldRR~R.", [536] = "srwRR~R.", + [792] = "srawRR~R.", [824] = "srawiRR~A.", + + [794] = "sradRR~R.", [826] = "sradiRR~H.", [827] = "sradiRR~H.", + [922] = "extshRR~.", [954] = "extsbRR~.", [986] = "extswRR~.", + + [539] = "srdRR~R.", +}, +{ __index = function(t, x) + if band(x, 31) == 15 then return "iselRRRC" end + end +}) + +local map_ld = { + shift = 0, mask = 3, + [0] = "ldRRE", "lduRRE", "lwaRRE", +} + +local map_std = { + shift = 0, mask = 3, + [0] = "stdRRE", "stduRRE", +} + +local map_fps = { + shift = 5, mask = 1, + { + shift = 1, mask = 15, + [0] = false, false, "fdivsFFF.", false, + "fsubsFFF.", "faddsFFF.", "fsqrtsF-F.", false, + "fresF-F.", "fmulsFF-F.", "frsqrtesF-F.", false, + "fmsubsFFFF~.", "fmaddsFFFF~.", "fnmsubsFFFF~.", "fnmaddsFFFF~.", + } +} + +local map_fpd = { + shift = 5, mask = 1, + [0] = { + shift = 1, mask = 1023, + [0] = "fcmpuXFF", [32] = "fcmpoXFF", [64] = "mcrfsXX", + [38] = "mtfsb1A.", [70] = "mtfsb0A.", [134] = "mtfsfiA>>-A>", + [8] = "fcpsgnFFF.", [40] = "fnegF-F.", [72] = "fmrF-F.", + [136] = "fnabsF-F.", [264] = "fabsF-F.", + [12] = "frspF-F.", + [14] = "fctiwF-F.", [15] = "fctiwzF-F.", + [583] = "mffsF.", [711] = "mtfsfZF.", + [392] = "frinF-F.", [424] = "frizF-F.", + [456] = "fripF-F.", [488] = "frimF-F.", + [814] = "fctidF-F.", [815] = "fctidzF-F.", [846] = "fcfidF-F.", + }, + { + shift = 1, mask = 15, + [0] = false, false, "fdivFFF.", false, + "fsubFFF.", "faddFFF.", "fsqrtF-F.", "fselFFFF~.", + "freF-F.", "fmulFF-F.", "frsqrteF-F.", false, + "fmsubFFFF~.", "fmaddFFFF~.", "fnmsubFFFF~.", "fnmaddFFFF~.", + } +} + +local map_spe = { + shift = 0, mask = 2047, + + [512] = "evaddwRRR", [514] = "evaddiwRAR~", + [516] = "evsubwRRR~", [518] = "evsubiwRAR~", + [520] = "evabsRR", [521] = "evnegRR", + [522] = "evextsbRR", [523] = "evextshRR", [524] = "evrndwRR", + [525] = "evcntlzwRR", [526] = "evcntlswRR", + + [527] = "brincRRR", + + [529] = "evandRRR", [530] = "evandcRRR", [534] = "evxorRRR", + [535] = "evor|evmrRRR=", [536] = "evnor|evnotRRR=", + [537] = "eveqvRRR", [539] = "evorcRRR", [542] = "evnandRRR", + + [544] = "evsrwuRRR", [545] = "evsrwsRRR", + [546] = "evsrwiuRRA", [547] = "evsrwisRRA", + [548] = "evslwRRR", [550] = "evslwiRRA", + [552] = "evrlwRRR", [553] = "evsplatiRS", + [554] = "evrlwiRRA", [555] = "evsplatfiRS", + [556] = "evmergehiRRR", [557] = "evmergeloRRR", + [558] = "evmergehiloRRR", [559] = "evmergelohiRRR", + + [560] = "evcmpgtuYRR", [561] = "evcmpgtsYRR", + [562] = "evcmpltuYRR", [563] = "evcmpltsYRR", + [564] = "evcmpeqYRR", + + [632] = "evselRRR", [633] = "evselRRRW", + [634] = "evselRRRW", [635] = "evselRRRW", + [636] = "evselRRRW", [637] = "evselRRRW", + [638] = "evselRRRW", [639] = "evselRRRW", + + [640] = "evfsaddRRR", [641] = "evfssubRRR", + [644] = "evfsabsRR", [645] = "evfsnabsRR", [646] = "evfsnegRR", + [648] = "evfsmulRRR", [649] = "evfsdivRRR", + [652] = "evfscmpgtYRR", [653] = "evfscmpltYRR", [654] = "evfscmpeqYRR", + [656] = "evfscfuiR-R", [657] = "evfscfsiR-R", + [658] = "evfscfufR-R", [659] = "evfscfsfR-R", + [660] = "evfsctuiR-R", [661] = "evfsctsiR-R", + [662] = "evfsctufR-R", [663] = "evfsctsfR-R", + [664] = "evfsctuizR-R", [666] = "evfsctsizR-R", + [668] = "evfststgtYRR", [669] = "evfststltYRR", [670] = "evfststeqYRR", + + [704] = "efsaddRRR", [705] = "efssubRRR", + [708] = "efsabsRR", [709] = "efsnabsRR", [710] = "efsnegRR", + [712] = "efsmulRRR", [713] = "efsdivRRR", + [716] = "efscmpgtYRR", [717] = "efscmpltYRR", [718] = "efscmpeqYRR", + [719] = "efscfdR-R", + [720] = "efscfuiR-R", [721] = "efscfsiR-R", + [722] = "efscfufR-R", [723] = "efscfsfR-R", + [724] = "efsctuiR-R", [725] = "efsctsiR-R", + [726] = "efsctufR-R", [727] = "efsctsfR-R", + [728] = "efsctuizR-R", [730] = "efsctsizR-R", + [732] = "efststgtYRR", [733] = "efststltYRR", [734] = "efststeqYRR", + + [736] = "efdaddRRR", [737] = "efdsubRRR", + [738] = "efdcfuidR-R", [739] = "efdcfsidR-R", + [740] = "efdabsRR", [741] = "efdnabsRR", [742] = "efdnegRR", + [744] = "efdmulRRR", [745] = "efddivRRR", + [746] = "efdctuidzR-R", [747] = "efdctsidzR-R", + [748] = "efdcmpgtYRR", [749] = "efdcmpltYRR", [750] = "efdcmpeqYRR", + [751] = "efdcfsR-R", + [752] = "efdcfuiR-R", [753] = "efdcfsiR-R", + [754] = "efdcfufR-R", [755] = "efdcfsfR-R", + [756] = "efdctuiR-R", [757] = "efdctsiR-R", + [758] = "efdctufR-R", [759] = "efdctsfR-R", + [760] = "efdctuizR-R", [762] = "efdctsizR-R", + [764] = "efdtstgtYRR", [765] = "efdtstltYRR", [766] = "efdtsteqYRR", + + [768] = "evlddxRR0R", [769] = "evlddRR8", + [770] = "evldwxRR0R", [771] = "evldwRR8", + [772] = "evldhxRR0R", [773] = "evldhRR8", + [776] = "evlhhesplatxRR0R", [777] = "evlhhesplatRR2", + [780] = "evlhhousplatxRR0R", [781] = "evlhhousplatRR2", + [782] = "evlhhossplatxRR0R", [783] = "evlhhossplatRR2", + [784] = "evlwhexRR0R", [785] = "evlwheRR4", + [788] = "evlwhouxRR0R", [789] = "evlwhouRR4", + [790] = "evlwhosxRR0R", [791] = "evlwhosRR4", + [792] = "evlwwsplatxRR0R", [793] = "evlwwsplatRR4", + [796] = "evlwhsplatxRR0R", [797] = "evlwhsplatRR4", + + [800] = "evstddxRR0R", [801] = "evstddRR8", + [802] = "evstdwxRR0R", [803] = "evstdwRR8", + [804] = "evstdhxRR0R", [805] = "evstdhRR8", + [816] = "evstwhexRR0R", [817] = "evstwheRR4", + [820] = "evstwhoxRR0R", [821] = "evstwhoRR4", + [824] = "evstwwexRR0R", [825] = "evstwweRR4", + [828] = "evstwwoxRR0R", [829] = "evstwwoRR4", + + [1027] = "evmhessfRRR", [1031] = "evmhossfRRR", [1032] = "evmheumiRRR", + [1033] = "evmhesmiRRR", [1035] = "evmhesmfRRR", [1036] = "evmhoumiRRR", + [1037] = "evmhosmiRRR", [1039] = "evmhosmfRRR", [1059] = "evmhessfaRRR", + [1063] = "evmhossfaRRR", [1064] = "evmheumiaRRR", [1065] = "evmhesmiaRRR", + [1067] = "evmhesmfaRRR", [1068] = "evmhoumiaRRR", [1069] = "evmhosmiaRRR", + [1071] = "evmhosmfaRRR", [1095] = "evmwhssfRRR", [1096] = "evmwlumiRRR", + [1100] = "evmwhumiRRR", [1101] = "evmwhsmiRRR", [1103] = "evmwhsmfRRR", + [1107] = "evmwssfRRR", [1112] = "evmwumiRRR", [1113] = "evmwsmiRRR", + [1115] = "evmwsmfRRR", [1127] = "evmwhssfaRRR", [1128] = "evmwlumiaRRR", + [1132] = "evmwhumiaRRR", [1133] = "evmwhsmiaRRR", [1135] = "evmwhsmfaRRR", + [1139] = "evmwssfaRRR", [1144] = "evmwumiaRRR", [1145] = "evmwsmiaRRR", + [1147] = "evmwsmfaRRR", + + [1216] = "evaddusiaawRR", [1217] = "evaddssiaawRR", + [1218] = "evsubfusiaawRR", [1219] = "evsubfssiaawRR", + [1220] = "evmraRR", + [1222] = "evdivwsRRR", [1223] = "evdivwuRRR", + [1224] = "evaddumiaawRR", [1225] = "evaddsmiaawRR", + [1226] = "evsubfumiaawRR", [1227] = "evsubfsmiaawRR", + + [1280] = "evmheusiaawRRR", [1281] = "evmhessiaawRRR", + [1283] = "evmhessfaawRRR", [1284] = "evmhousiaawRRR", + [1285] = "evmhossiaawRRR", [1287] = "evmhossfaawRRR", + [1288] = "evmheumiaawRRR", [1289] = "evmhesmiaawRRR", + [1291] = "evmhesmfaawRRR", [1292] = "evmhoumiaawRRR", + [1293] = "evmhosmiaawRRR", [1295] = "evmhosmfaawRRR", + [1320] = "evmhegumiaaRRR", [1321] = "evmhegsmiaaRRR", + [1323] = "evmhegsmfaaRRR", [1324] = "evmhogumiaaRRR", + [1325] = "evmhogsmiaaRRR", [1327] = "evmhogsmfaaRRR", + [1344] = "evmwlusiaawRRR", [1345] = "evmwlssiaawRRR", + [1352] = "evmwlumiaawRRR", [1353] = "evmwlsmiaawRRR", + [1363] = "evmwssfaaRRR", [1368] = "evmwumiaaRRR", + [1369] = "evmwsmiaaRRR", [1371] = "evmwsmfaaRRR", + [1408] = "evmheusianwRRR", [1409] = "evmhessianwRRR", + [1411] = "evmhessfanwRRR", [1412] = "evmhousianwRRR", + [1413] = "evmhossianwRRR", [1415] = "evmhossfanwRRR", + [1416] = "evmheumianwRRR", [1417] = "evmhesmianwRRR", + [1419] = "evmhesmfanwRRR", [1420] = "evmhoumianwRRR", + [1421] = "evmhosmianwRRR", [1423] = "evmhosmfanwRRR", + [1448] = "evmhegumianRRR", [1449] = "evmhegsmianRRR", + [1451] = "evmhegsmfanRRR", [1452] = "evmhogumianRRR", + [1453] = "evmhogsmianRRR", [1455] = "evmhogsmfanRRR", + [1472] = "evmwlusianwRRR", [1473] = "evmwlssianwRRR", + [1480] = "evmwlumianwRRR", [1481] = "evmwlsmianwRRR", + [1491] = "evmwssfanRRR", [1496] = "evmwumianRRR", + [1497] = "evmwsmianRRR", [1499] = "evmwsmfanRRR", +} + +local map_pri = { + [0] = false, false, "tdiARI", "twiARI", + map_spe, false, false, "mulliRRI", + "subficRRI", false, "cmpl_iYLRU", "cmp_iYLRI", + "addicRRI", "addic.RRI", "addi|liRR0I", "addis|lisRR0I", + "b_KBJ", "sc", "bKJ", map_crops, + "rlwimiRR~AAA.", map_rlwinm, false, "rlwnmRR~RAA.", + "oriNRR~U", "orisRR~U", "xoriRR~U", "xorisRR~U", + "andi.RR~U", "andis.RR~U", map_rld, map_ext, + "lwzRRD", "lwzuRRD", "lbzRRD", "lbzuRRD", + "stwRRD", "stwuRRD", "stbRRD", "stbuRRD", + "lhzRRD", "lhzuRRD", "lhaRRD", "lhauRRD", + "sthRRD", "sthuRRD", "lmwRRD", "stmwRRD", + "lfsFRD", "lfsuFRD", "lfdFRD", "lfduFRD", + "stfsFRD", "stfsuFRD", "stfdFRD", "stfduFRD", + false, false, map_ld, map_fps, + false, false, map_std, map_fpd, +} + +------------------------------------------------------------------------------ + +local map_gpr = { + [0] = "r0", "sp", "r2", "r3", "r4", "r5", "r6", "r7", + "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15", + "r16", "r17", "r18", "r19", "r20", "r21", "r22", "r23", + "r24", "r25", "r26", "r27", "r28", "r29", "r30", "r31", +} + +local map_cond = { [0] = "lt", "gt", "eq", "so", "ge", "le", "ne", "ns", } + +-- Format a condition bit. +local function condfmt(cond) + if cond <= 3 then + return map_cond[band(cond, 3)] + else + return format("4*cr%d+%s", rshift(cond, 2), map_cond[band(cond, 3)]) + end +end + +------------------------------------------------------------------------------ + +-- Output a nicely formatted line with an opcode and operands. +local function putop(ctx, text, operands) + local pos = ctx.pos + local extra = "" + if ctx.rel then + local sym = ctx.symtab[ctx.rel] + if sym then extra = "\t->"..sym end + end + if ctx.hexdump > 0 then + ctx.out(format("%08x %s %-7s %s%s\n", + ctx.addr+pos, tohex(ctx.op), text, concat(operands, ", "), extra)) + else + ctx.out(format("%08x %-7s %s%s\n", + ctx.addr+pos, text, concat(operands, ", "), extra)) + end + ctx.pos = pos + 4 +end + +-- Fallback for unknown opcodes. +local function unknown(ctx) + return putop(ctx, ".long", { "0x"..tohex(ctx.op) }) +end + +-- Disassemble a single instruction. +local function disass_ins(ctx) + local pos = ctx.pos + local b0, b1, b2, b3 = byte(ctx.code, pos+1, pos+4) + local op = bor(lshift(b0, 24), lshift(b1, 16), lshift(b2, 8), b3) + local operands = {} + local last = nil + local rs = 21 + ctx.op = op + ctx.rel = nil + + local opat = map_pri[rshift(b0, 2)] + while type(opat) ~= "string" do + if not opat then return unknown(ctx) end + opat = opat[band(rshift(op, opat.shift), opat.mask)] + end + local name, pat = match(opat, "^([a-z0-9_.]*)(.*)") + local altname, pat2 = match(pat, "|([a-z0-9_.]*)(.*)") + if altname then pat = pat2 end + + for p in gmatch(pat, ".") do + local x = nil + if p == "R" then + x = map_gpr[band(rshift(op, rs), 31)] + rs = rs - 5 + elseif p == "F" then + x = "f"..band(rshift(op, rs), 31) + rs = rs - 5 + elseif p == "A" then + x = band(rshift(op, rs), 31) + rs = rs - 5 + elseif p == "S" then + x = arshift(lshift(op, 27-rs), 27) + rs = rs - 5 + elseif p == "I" then + x = arshift(lshift(op, 16), 16) + elseif p == "U" then + x = band(op, 0xffff) + elseif p == "D" or p == "E" then + local disp = arshift(lshift(op, 16), 16) + if p == "E" then disp = band(disp, -4) end + if last == "r0" then last = "0" end + operands[#operands] = format("%d(%s)", disp, last) + elseif p >= "2" and p <= "8" then + local disp = band(rshift(op, rs), 31) * p + if last == "r0" then last = "0" end + operands[#operands] = format("%d(%s)", disp, last) + elseif p == "H" then + x = band(rshift(op, rs), 31) + lshift(band(op, 2), 4) + rs = rs - 5 + elseif p == "M" then + x = band(rshift(op, rs), 31) + band(op, 0x20) + elseif p == "C" then + x = condfmt(band(rshift(op, rs), 31)) + rs = rs - 5 + elseif p == "B" then + local bo = rshift(op, 21) + local cond = band(rshift(op, 16), 31) + local cn = "" + rs = rs - 10 + if band(bo, 4) == 0 then + cn = band(bo, 2) == 0 and "dnz" or "dz" + if band(bo, 0x10) == 0 then + cn = cn..(band(bo, 8) == 0 and "f" or "t") + end + if band(bo, 0x10) == 0 then x = condfmt(cond) end + name = name..(band(bo, 1) == band(rshift(op, 15), 1) and "-" or "+") + elseif band(bo, 0x10) == 0 then + cn = map_cond[band(cond, 3) + (band(bo, 8) == 0 and 4 or 0)] + if cond > 3 then x = "cr"..rshift(cond, 2) end + name = name..(band(bo, 1) == band(rshift(op, 15), 1) and "-" or "+") + end + name = gsub(name, "_", cn) + elseif p == "J" then + x = arshift(lshift(op, 27-rs), 29-rs)*4 + if band(op, 2) == 0 then x = ctx.addr + pos + x end + ctx.rel = x + x = "0x"..tohex(x) + elseif p == "K" then + if band(op, 1) ~= 0 then name = name.."l" end + if band(op, 2) ~= 0 then name = name.."a" end + elseif p == "X" or p == "Y" then + x = band(rshift(op, rs+2), 7) + if x == 0 and p == "Y" then x = nil else x = "cr"..x end + rs = rs - 5 + elseif p == "W" then + x = "cr"..band(op, 7) + elseif p == "Z" then + x = band(rshift(op, rs-4), 255) + rs = rs - 10 + elseif p == ">" then + operands[#operands] = rshift(operands[#operands], 1) + elseif p == "0" then + if last == "r0" then + operands[#operands] = nil + if altname then name = altname end + end + elseif p == "L" then + name = gsub(name, "_", band(op, 0x00200000) ~= 0 and "d" or "w") + elseif p == "." then + if band(op, 1) == 1 then name = name.."." end + elseif p == "N" then + if op == 0x60000000 then name = "nop"; break end + elseif p == "~" then + local n = #operands + operands[n-1], operands[n] = operands[n], operands[n-1] + elseif p == "=" then + local n = #operands + if last == operands[n-1] then + operands[n] = nil + name = altname + end + elseif p == "%" then + local n = #operands + if last == operands[n-1] and last == operands[n-2] then + operands[n] = nil + operands[n-1] = nil + name = altname + end + elseif p == "-" then + rs = rs - 5 + else + assert(false) + end + if x then operands[#operands+1] = x; last = x end + end + + return putop(ctx, name, operands) +end + +------------------------------------------------------------------------------ + +-- Disassemble a block of code. +local function disass_block(ctx, ofs, len) + if not ofs then ofs = 0 end + local stop = len and ofs+len or #ctx.code + stop = stop - stop % 4 + ctx.pos = ofs - ofs % 4 + ctx.rel = nil + while ctx.pos < stop do disass_ins(ctx) end +end + +-- Extended API: create a disassembler context. Then call ctx:disass(ofs, len). +local function create_(code, addr, out) + local ctx = {} + ctx.code = code + ctx.addr = addr or 0 + ctx.out = out or io.write + ctx.symtab = {} + ctx.disass = disass_block + ctx.hexdump = 8 + return ctx +end + +-- Simple API: disassemble code (a string) at address and output via out. +local function disass_(code, addr, out) + create_(code, addr, out):disass() +end + +-- Return register name for RID. +local function regname_(r) + if r < 32 then return map_gpr[r] end + return "f"..(r-32) +end + +-- Public module functions. +module(...) + +create = create_ +disass = disass_ +regname = regname_ + diff --git a/src/jit/dis_x64.lua b/src/jit/dis_x64.lua new file mode 100644 index 00000000..d95d7ded --- /dev/null +++ b/src/jit/dis_x64.lua @@ -0,0 +1,20 @@ +---------------------------------------------------------------------------- +-- LuaJIT x64 disassembler wrapper module. +-- +-- Copyright (C) 2005-2012 Mike Pall. All rights reserved. +-- Released under the MIT license. See Copyright Notice in luajit.h +---------------------------------------------------------------------------- +-- This module just exports the 64 bit functions from the combined +-- x86/x64 disassembler module. All the interesting stuff is there. +------------------------------------------------------------------------------ + +local require = require + +module(...) + +local dis_x86 = require(_PACKAGE.."dis_x86") + +create = dis_x86.create64 +disass = dis_x86.disass64 +regname = dis_x86.regname64 + diff --git a/src/jit/dis_x86.lua b/src/jit/dis_x86.lua new file mode 100644 index 00000000..5aeeb449 --- /dev/null +++ b/src/jit/dis_x86.lua @@ -0,0 +1,836 @@ +---------------------------------------------------------------------------- +-- LuaJIT x86/x64 disassembler module. +-- +-- Copyright (C) 2005-2012 Mike Pall. All rights reserved. +-- Released under the MIT license. See Copyright Notice in luajit.h +---------------------------------------------------------------------------- +-- This is a helper module used by the LuaJIT machine code dumper module. +-- +-- Sending small code snippets to an external disassembler and mixing the +-- output with our own stuff was too fragile. So I had to bite the bullet +-- and write yet another x86 disassembler. Oh well ... +-- +-- The output format is very similar to what ndisasm generates. But it has +-- been developed independently by looking at the opcode tables from the +-- Intel and AMD manuals. The supported instruction set is quite extensive +-- and reflects what a current generation Intel or AMD CPU implements in +-- 32 bit and 64 bit mode. Yes, this includes MMX, SSE, SSE2, SSE3, SSSE3, +-- SSE4.1, SSE4.2, SSE4a and even privileged and hypervisor (VMX/SVM) +-- instructions. +-- +-- Notes: +-- * The (useless) a16 prefix, 3DNow and pre-586 opcodes are unsupported. +-- * No attempt at optimization has been made -- it's fast enough for my needs. +-- * The public API may change when more architectures are added. +------------------------------------------------------------------------------ + +local type = type +local sub, byte, format = string.sub, string.byte, string.format +local match, gmatch, gsub = string.match, string.gmatch, string.gsub +local lower, rep = string.lower, string.rep + +-- Map for 1st opcode byte in 32 bit mode. Ugly? Well ... read on. +local map_opc1_32 = { +--0x +[0]="addBmr","addVmr","addBrm","addVrm","addBai","addVai","push es","pop es", +"orBmr","orVmr","orBrm","orVrm","orBai","orVai","push cs","opc2*", +--1x +"adcBmr","adcVmr","adcBrm","adcVrm","adcBai","adcVai","push ss","pop ss", +"sbbBmr","sbbVmr","sbbBrm","sbbVrm","sbbBai","sbbVai","push ds","pop ds", +--2x +"andBmr","andVmr","andBrm","andVrm","andBai","andVai","es:seg","daa", +"subBmr","subVmr","subBrm","subVrm","subBai","subVai","cs:seg","das", +--3x +"xorBmr","xorVmr","xorBrm","xorVrm","xorBai","xorVai","ss:seg","aaa", +"cmpBmr","cmpVmr","cmpBrm","cmpVrm","cmpBai","cmpVai","ds:seg","aas", +--4x +"incVR","incVR","incVR","incVR","incVR","incVR","incVR","incVR", +"decVR","decVR","decVR","decVR","decVR","decVR","decVR","decVR", +--5x +"pushUR","pushUR","pushUR","pushUR","pushUR","pushUR","pushUR","pushUR", +"popUR","popUR","popUR","popUR","popUR","popUR","popUR","popUR", +--6x +"sz*pushaw,pusha","sz*popaw,popa","boundVrm","arplWmr", +"fs:seg","gs:seg","o16:","a16", +"pushUi","imulVrmi","pushBs","imulVrms", +"insb","insVS","outsb","outsVS", +--7x +"joBj","jnoBj","jbBj","jnbBj","jzBj","jnzBj","jbeBj","jaBj", +"jsBj","jnsBj","jpeBj","jpoBj","jlBj","jgeBj","jleBj","jgBj", +--8x +"arith!Bmi","arith!Vmi","arith!Bmi","arith!Vms", +"testBmr","testVmr","xchgBrm","xchgVrm", +"movBmr","movVmr","movBrm","movVrm", +"movVmg","leaVrm","movWgm","popUm", +--9x +"nop*xchgVaR|pause|xchgWaR|repne nop","xchgVaR","xchgVaR","xchgVaR", +"xchgVaR","xchgVaR","xchgVaR","xchgVaR", +"sz*cbw,cwde,cdqe","sz*cwd,cdq,cqo","call farViw","wait", +"sz*pushfw,pushf","sz*popfw,popf","sahf","lahf", +--Ax +"movBao","movVao","movBoa","movVoa", +"movsb","movsVS","cmpsb","cmpsVS", +"testBai","testVai","stosb","stosVS", +"lodsb","lodsVS","scasb","scasVS", +--Bx +"movBRi","movBRi","movBRi","movBRi","movBRi","movBRi","movBRi","movBRi", +"movVRI","movVRI","movVRI","movVRI","movVRI","movVRI","movVRI","movVRI", +--Cx +"shift!Bmu","shift!Vmu","retBw","ret","$lesVrm","$ldsVrm","movBmi","movVmi", +"enterBwu","leave","retfBw","retf","int3","intBu","into","iretVS", +--Dx +"shift!Bm1","shift!Vm1","shift!Bmc","shift!Vmc","aamBu","aadBu","salc","xlatb", +"fp*0","fp*1","fp*2","fp*3","fp*4","fp*5","fp*6","fp*7", +--Ex +"loopneBj","loopeBj","loopBj","sz*jcxzBj,jecxzBj,jrcxzBj", +"inBau","inVau","outBua","outVua", +"callVj","jmpVj","jmp farViw","jmpBj","inBad","inVad","outBda","outVda", +--Fx +"lock:","int1","repne:rep","rep:","hlt","cmc","testb!Bm","testv!Vm", +"clc","stc","cli","sti","cld","std","incb!Bm","incd!Vm", +} +assert(#map_opc1_32 == 255) + +-- Map for 1st opcode byte in 64 bit mode (overrides only). +local map_opc1_64 = setmetatable({ + [0x06]=false, [0x07]=false, [0x0e]=false, + [0x16]=false, [0x17]=false, [0x1e]=false, [0x1f]=false, + [0x27]=false, [0x2f]=false, [0x37]=false, [0x3f]=false, + [0x60]=false, [0x61]=false, [0x62]=false, [0x63]="movsxdVrDmt", [0x67]="a32:", + [0x40]="rex*", [0x41]="rex*b", [0x42]="rex*x", [0x43]="rex*xb", + [0x44]="rex*r", [0x45]="rex*rb", [0x46]="rex*rx", [0x47]="rex*rxb", + [0x48]="rex*w", [0x49]="rex*wb", [0x4a]="rex*wx", [0x4b]="rex*wxb", + [0x4c]="rex*wr", [0x4d]="rex*wrb", [0x4e]="rex*wrx", [0x4f]="rex*wrxb", + [0x82]=false, [0x9a]=false, [0xc4]=false, [0xc5]=false, [0xce]=false, + [0xd4]=false, [0xd5]=false, [0xd6]=false, [0xea]=false, +}, { __index = map_opc1_32 }) + +-- Map for 2nd opcode byte (0F xx). True CISC hell. Hey, I told you. +-- Prefix dependent MMX/SSE opcodes: (none)|rep|o16|repne, -|F3|66|F2 +local map_opc2 = { +--0x +[0]="sldt!Dmp","sgdt!Ump","larVrm","lslVrm",nil,"syscall","clts","sysret", +"invd","wbinvd",nil,"ud1",nil,"$prefetch!Bm","femms","3dnowMrmu", +--1x +"movupsXrm|movssXrm|movupdXrm|movsdXrm", +"movupsXmr|movssXmr|movupdXmr|movsdXmr", +"movhlpsXrm$movlpsXrm|movsldupXrm|movlpdXrm|movddupXrm", +"movlpsXmr||movlpdXmr", +"unpcklpsXrm||unpcklpdXrm", +"unpckhpsXrm||unpckhpdXrm", +"movlhpsXrm$movhpsXrm|movshdupXrm|movhpdXrm", +"movhpsXmr||movhpdXmr", +"$prefetcht!Bm","hintnopVm","hintnopVm","hintnopVm", +"hintnopVm","hintnopVm","hintnopVm","hintnopVm", +--2x +"movUmx$","movUmy$","movUxm$","movUym$","movUmz$",nil,"movUzm$",nil, +"movapsXrm||movapdXrm", +"movapsXmr||movapdXmr", +"cvtpi2psXrMm|cvtsi2ssXrVmt|cvtpi2pdXrMm|cvtsi2sdXrVmt", +"movntpsXmr|movntssXmr|movntpdXmr|movntsdXmr", +"cvttps2piMrXm|cvttss2siVrXm|cvttpd2piMrXm|cvttsd2siVrXm", +"cvtps2piMrXm|cvtss2siVrXm|cvtpd2piMrXm|cvtsd2siVrXm", +"ucomissXrm||ucomisdXrm", +"comissXrm||comisdXrm", +--3x +"wrmsr","rdtsc","rdmsr","rdpmc","sysenter","sysexit",nil,"getsec", +"opc3*38",nil,"opc3*3a",nil,nil,nil,nil,nil, +--4x +"cmovoVrm","cmovnoVrm","cmovbVrm","cmovnbVrm", +"cmovzVrm","cmovnzVrm","cmovbeVrm","cmovaVrm", +"cmovsVrm","cmovnsVrm","cmovpeVrm","cmovpoVrm", +"cmovlVrm","cmovgeVrm","cmovleVrm","cmovgVrm", +--5x +"movmskpsVrXm$||movmskpdVrXm$","sqrtpsXrm|sqrtssXrm|sqrtpdXrm|sqrtsdXrm", +"rsqrtpsXrm|rsqrtssXrm","rcppsXrm|rcpssXrm", +"andpsXrm||andpdXrm","andnpsXrm||andnpdXrm", +"orpsXrm||orpdXrm","xorpsXrm||xorpdXrm", +"addpsXrm|addssXrm|addpdXrm|addsdXrm","mulpsXrm|mulssXrm|mulpdXrm|mulsdXrm", +"cvtps2pdXrm|cvtss2sdXrm|cvtpd2psXrm|cvtsd2ssXrm", +"cvtdq2psXrm|cvttps2dqXrm|cvtps2dqXrm", +"subpsXrm|subssXrm|subpdXrm|subsdXrm","minpsXrm|minssXrm|minpdXrm|minsdXrm", +"divpsXrm|divssXrm|divpdXrm|divsdXrm","maxpsXrm|maxssXrm|maxpdXrm|maxsdXrm", +--6x +"punpcklbwPrm","punpcklwdPrm","punpckldqPrm","packsswbPrm", +"pcmpgtbPrm","pcmpgtwPrm","pcmpgtdPrm","packuswbPrm", +"punpckhbwPrm","punpckhwdPrm","punpckhdqPrm","packssdwPrm", +"||punpcklqdqXrm","||punpckhqdqXrm", +"movPrVSm","movqMrm|movdquXrm|movdqaXrm", +--7x +"pshufwMrmu|pshufhwXrmu|pshufdXrmu|pshuflwXrmu","pshiftw!Pmu", +"pshiftd!Pmu","pshiftq!Mmu||pshiftdq!Xmu", +"pcmpeqbPrm","pcmpeqwPrm","pcmpeqdPrm","emms|", +"vmreadUmr||extrqXmuu$|insertqXrmuu$","vmwriteUrm||extrqXrm$|insertqXrm$", +nil,nil, +"||haddpdXrm|haddpsXrm","||hsubpdXrm|hsubpsXrm", +"movVSmMr|movqXrm|movVSmXr","movqMmr|movdquXmr|movdqaXmr", +--8x +"joVj","jnoVj","jbVj","jnbVj","jzVj","jnzVj","jbeVj","jaVj", +"jsVj","jnsVj","jpeVj","jpoVj","jlVj","jgeVj","jleVj","jgVj", +--9x +"setoBm","setnoBm","setbBm","setnbBm","setzBm","setnzBm","setbeBm","setaBm", +"setsBm","setnsBm","setpeBm","setpoBm","setlBm","setgeBm","setleBm","setgBm", +--Ax +"push fs","pop fs","cpuid","btVmr","shldVmru","shldVmrc",nil,nil, +"push gs","pop gs","rsm","btsVmr","shrdVmru","shrdVmrc","fxsave!Dmp","imulVrm", +--Bx +"cmpxchgBmr","cmpxchgVmr","$lssVrm","btrVmr", +"$lfsVrm","$lgsVrm","movzxVrBmt","movzxVrWmt", +"|popcntVrm","ud2Dp","bt!Vmu","btcVmr", +"bsfVrm","bsrVrm|lzcntVrm|bsrWrm","movsxVrBmt","movsxVrWmt", +--Cx +"xaddBmr","xaddVmr", +"cmppsXrmu|cmpssXrmu|cmppdXrmu|cmpsdXrmu","$movntiVmr|", +"pinsrwPrWmu","pextrwDrPmu", +"shufpsXrmu||shufpdXrmu","$cmpxchg!Qmp", +"bswapVR","bswapVR","bswapVR","bswapVR","bswapVR","bswapVR","bswapVR","bswapVR", +--Dx +"||addsubpdXrm|addsubpsXrm","psrlwPrm","psrldPrm","psrlqPrm", +"paddqPrm","pmullwPrm", +"|movq2dqXrMm|movqXmr|movdq2qMrXm$","pmovmskbVrMm||pmovmskbVrXm", +"psubusbPrm","psubuswPrm","pminubPrm","pandPrm", +"paddusbPrm","padduswPrm","pmaxubPrm","pandnPrm", +--Ex +"pavgbPrm","psrawPrm","psradPrm","pavgwPrm", +"pmulhuwPrm","pmulhwPrm", +"|cvtdq2pdXrm|cvttpd2dqXrm|cvtpd2dqXrm","$movntqMmr||$movntdqXmr", +"psubsbPrm","psubswPrm","pminswPrm","porPrm", +"paddsbPrm","paddswPrm","pmaxswPrm","pxorPrm", +--Fx +"|||lddquXrm","psllwPrm","pslldPrm","psllqPrm", +"pmuludqPrm","pmaddwdPrm","psadbwPrm","maskmovqMrm||maskmovdquXrm$", +"psubbPrm","psubwPrm","psubdPrm","psubqPrm", +"paddbPrm","paddwPrm","padddPrm","ud", +} +assert(map_opc2[255] == "ud") + +-- Map for three-byte opcodes. Can't wait for their next invention. +local map_opc3 = { +["38"] = { -- [66] 0f 38 xx +--0x +[0]="pshufbPrm","phaddwPrm","phadddPrm","phaddswPrm", +"pmaddubswPrm","phsubwPrm","phsubdPrm","phsubswPrm", +"psignbPrm","psignwPrm","psigndPrm","pmulhrswPrm", +nil,nil,nil,nil, +--1x +"||pblendvbXrma",nil,nil,nil, +"||blendvpsXrma","||blendvpdXrma",nil,"||ptestXrm", +nil,nil,nil,nil, +"pabsbPrm","pabswPrm","pabsdPrm",nil, +--2x +"||pmovsxbwXrm","||pmovsxbdXrm","||pmovsxbqXrm","||pmovsxwdXrm", +"||pmovsxwqXrm","||pmovsxdqXrm",nil,nil, +"||pmuldqXrm","||pcmpeqqXrm","||$movntdqaXrm","||packusdwXrm", +nil,nil,nil,nil, +--3x +"||pmovzxbwXrm","||pmovzxbdXrm","||pmovzxbqXrm","||pmovzxwdXrm", +"||pmovzxwqXrm","||pmovzxdqXrm",nil,"||pcmpgtqXrm", +"||pminsbXrm","||pminsdXrm","||pminuwXrm","||pminudXrm", +"||pmaxsbXrm","||pmaxsdXrm","||pmaxuwXrm","||pmaxudXrm", +--4x +"||pmulddXrm","||phminposuwXrm", +--Fx +[0xf0] = "|||crc32TrBmt",[0xf1] = "|||crc32TrVmt", +}, + +["3a"] = { -- [66] 0f 3a xx +--0x +[0x00]=nil,nil,nil,nil,nil,nil,nil,nil, +"||roundpsXrmu","||roundpdXrmu","||roundssXrmu","||roundsdXrmu", +"||blendpsXrmu","||blendpdXrmu","||pblendwXrmu","palignrPrmu", +--1x +nil,nil,nil,nil, +"||pextrbVmXru","||pextrwVmXru","||pextrVmSXru","||extractpsVmXru", +nil,nil,nil,nil,nil,nil,nil,nil, +--2x +"||pinsrbXrVmu","||insertpsXrmu","||pinsrXrVmuS",nil, +--4x +[0x40] = "||dppsXrmu", +[0x41] = "||dppdXrmu", +[0x42] = "||mpsadbwXrmu", +--6x +[0x60] = "||pcmpestrmXrmu",[0x61] = "||pcmpestriXrmu", +[0x62] = "||pcmpistrmXrmu",[0x63] = "||pcmpistriXrmu", +}, +} + +-- Map for VMX/SVM opcodes 0F 01 C0-FF (sgdt group with register operands). +local map_opcvm = { +[0xc1]="vmcall",[0xc2]="vmlaunch",[0xc3]="vmresume",[0xc4]="vmxoff", +[0xc8]="monitor",[0xc9]="mwait", +[0xd8]="vmrun",[0xd9]="vmmcall",[0xda]="vmload",[0xdb]="vmsave", +[0xdc]="stgi",[0xdd]="clgi",[0xde]="skinit",[0xdf]="invlpga", +[0xf8]="swapgs",[0xf9]="rdtscp", +} + +-- Map for FP opcodes. And you thought stack machines are simple? +local map_opcfp = { +-- D8-DF 00-BF: opcodes with a memory operand. +-- D8 +[0]="faddFm","fmulFm","fcomFm","fcompFm","fsubFm","fsubrFm","fdivFm","fdivrFm", +"fldFm",nil,"fstFm","fstpFm","fldenvVm","fldcwWm","fnstenvVm","fnstcwWm", +-- DA +"fiaddDm","fimulDm","ficomDm","ficompDm", +"fisubDm","fisubrDm","fidivDm","fidivrDm", +-- DB +"fildDm","fisttpDm","fistDm","fistpDm",nil,"fld twordFmp",nil,"fstp twordFmp", +-- DC +"faddGm","fmulGm","fcomGm","fcompGm","fsubGm","fsubrGm","fdivGm","fdivrGm", +-- DD +"fldGm","fisttpQm","fstGm","fstpGm","frstorDmp",nil,"fnsaveDmp","fnstswWm", +-- DE +"fiaddWm","fimulWm","ficomWm","ficompWm", +"fisubWm","fisubrWm","fidivWm","fidivrWm", +-- DF +"fildWm","fisttpWm","fistWm","fistpWm", +"fbld twordFmp","fildQm","fbstp twordFmp","fistpQm", +-- xx C0-FF: opcodes with a pseudo-register operand. +-- D8 +"faddFf","fmulFf","fcomFf","fcompFf","fsubFf","fsubrFf","fdivFf","fdivrFf", +-- D9 +"fldFf","fxchFf",{"fnop"},nil, +{"fchs","fabs",nil,nil,"ftst","fxam"}, +{"fld1","fldl2t","fldl2e","fldpi","fldlg2","fldln2","fldz"}, +{"f2xm1","fyl2x","fptan","fpatan","fxtract","fprem1","fdecstp","fincstp"}, +{"fprem","fyl2xp1","fsqrt","fsincos","frndint","fscale","fsin","fcos"}, +-- DA +"fcmovbFf","fcmoveFf","fcmovbeFf","fcmovuFf",nil,{nil,"fucompp"},nil,nil, +-- DB +"fcmovnbFf","fcmovneFf","fcmovnbeFf","fcmovnuFf", +{nil,nil,"fnclex","fninit"},"fucomiFf","fcomiFf",nil, +-- DC +"fadd toFf","fmul toFf",nil,nil, +"fsub toFf","fsubr toFf","fdivr toFf","fdiv toFf", +-- DD +"ffreeFf",nil,"fstFf","fstpFf","fucomFf","fucompFf",nil,nil, +-- DE +"faddpFf","fmulpFf",nil,{nil,"fcompp"}, +"fsubrpFf","fsubpFf","fdivrpFf","fdivpFf", +-- DF +nil,nil,nil,nil,{"fnstsw ax"},"fucomipFf","fcomipFf",nil, +} +assert(map_opcfp[126] == "fcomipFf") + +-- Map for opcode groups. The subkey is sp from the ModRM byte. +local map_opcgroup = { + arith = { "add", "or", "adc", "sbb", "and", "sub", "xor", "cmp" }, + shift = { "rol", "ror", "rcl", "rcr", "shl", "shr", "sal", "sar" }, + testb = { "testBmi", "testBmi", "not", "neg", "mul", "imul", "div", "idiv" }, + testv = { "testVmi", "testVmi", "not", "neg", "mul", "imul", "div", "idiv" }, + incb = { "inc", "dec" }, + incd = { "inc", "dec", "callUmp", "$call farDmp", + "jmpUmp", "$jmp farDmp", "pushUm" }, + sldt = { "sldt", "str", "lldt", "ltr", "verr", "verw" }, + sgdt = { "vm*$sgdt", "vm*$sidt", "$lgdt", "vm*$lidt", + "smsw", nil, "lmsw", "vm*$invlpg" }, + bt = { nil, nil, nil, nil, "bt", "bts", "btr", "btc" }, + cmpxchg = { nil, "sz*,cmpxchg8bQmp,cmpxchg16bXmp", nil, nil, + nil, nil, "vmptrld|vmxon|vmclear", "vmptrst" }, + pshiftw = { nil, nil, "psrlw", nil, "psraw", nil, "psllw" }, + pshiftd = { nil, nil, "psrld", nil, "psrad", nil, "pslld" }, + pshiftq = { nil, nil, "psrlq", nil, nil, nil, "psllq" }, + pshiftdq = { nil, nil, "psrlq", "psrldq", nil, nil, "psllq", "pslldq" }, + fxsave = { "$fxsave", "$fxrstor", "$ldmxcsr", "$stmxcsr", + nil, "lfenceDp$", "mfenceDp$", "sfenceDp$clflush" }, + prefetch = { "prefetch", "prefetchw" }, + prefetcht = { "prefetchnta", "prefetcht0", "prefetcht1", "prefetcht2" }, +} + +------------------------------------------------------------------------------ + +-- Maps for register names. +local map_regs = { + B = { "al", "cl", "dl", "bl", "ah", "ch", "dh", "bh", + "r8b", "r9b", "r10b", "r11b", "r12b", "r13b", "r14b", "r15b" }, + B64 = { "al", "cl", "dl", "bl", "spl", "bpl", "sil", "dil", + "r8b", "r9b", "r10b", "r11b", "r12b", "r13b", "r14b", "r15b" }, + W = { "ax", "cx", "dx", "bx", "sp", "bp", "si", "di", + "r8w", "r9w", "r10w", "r11w", "r12w", "r13w", "r14w", "r15w" }, + D = { "eax", "ecx", "edx", "ebx", "esp", "ebp", "esi", "edi", + "r8d", "r9d", "r10d", "r11d", "r12d", "r13d", "r14d", "r15d" }, + Q = { "rax", "rcx", "rdx", "rbx", "rsp", "rbp", "rsi", "rdi", + "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15" }, + M = { "mm0", "mm1", "mm2", "mm3", "mm4", "mm5", "mm6", "mm7", + "mm0", "mm1", "mm2", "mm3", "mm4", "mm5", "mm6", "mm7" }, -- No x64 ext! + X = { "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7", + "xmm8", "xmm9", "xmm10", "xmm11", "xmm12", "xmm13", "xmm14", "xmm15" }, +} +local map_segregs = { "es", "cs", "ss", "ds", "fs", "gs", "segr6", "segr7" } + +-- Maps for size names. +local map_sz2n = { + B = 1, W = 2, D = 4, Q = 8, M = 8, X = 16, +} +local map_sz2prefix = { + B = "byte", W = "word", D = "dword", + Q = "qword", + M = "qword", X = "xword", + F = "dword", G = "qword", -- No need for sizes/register names for these two. +} + +------------------------------------------------------------------------------ + +-- Output a nicely formatted line with an opcode and operands. +local function putop(ctx, text, operands) + local code, pos, hex = ctx.code, ctx.pos, "" + local hmax = ctx.hexdump + if hmax > 0 then + for i=ctx.start,pos-1 do + hex = hex..format("%02X", byte(code, i, i)) + end + if #hex > hmax then hex = sub(hex, 1, hmax)..". " + else hex = hex..rep(" ", hmax-#hex+2) end + end + if operands then text = text.." "..operands end + if ctx.o16 then text = "o16 "..text; ctx.o16 = false end + if ctx.a32 then text = "a32 "..text; ctx.a32 = false end + if ctx.rep then text = ctx.rep.." "..text; ctx.rep = false end + if ctx.rex then + local t = (ctx.rexw and "w" or "")..(ctx.rexr and "r" or "").. + (ctx.rexx and "x" or "")..(ctx.rexb and "b" or "") + if t ~= "" then text = "rex."..t.." "..text end + ctx.rexw = false; ctx.rexr = false; ctx.rexx = false; ctx.rexb = false + ctx.rex = false + end + if ctx.seg then + local text2, n = gsub(text, "%[", "["..ctx.seg..":") + if n == 0 then text = ctx.seg.." "..text else text = text2 end + ctx.seg = false + end + if ctx.lock then text = "lock "..text; ctx.lock = false end + local imm = ctx.imm + if imm then + local sym = ctx.symtab[imm] + if sym then text = text.."\t->"..sym end + end + ctx.out(format("%08x %s%s\n", ctx.addr+ctx.start, hex, text)) + ctx.mrm = false + ctx.start = pos + ctx.imm = nil +end + +-- Clear all prefix flags. +local function clearprefixes(ctx) + ctx.o16 = false; ctx.seg = false; ctx.lock = false; ctx.rep = false + ctx.rexw = false; ctx.rexr = false; ctx.rexx = false; ctx.rexb = false + ctx.rex = false; ctx.a32 = false +end + +-- Fallback for incomplete opcodes at the end. +local function incomplete(ctx) + ctx.pos = ctx.stop+1 + clearprefixes(ctx) + return putop(ctx, "(incomplete)") +end + +-- Fallback for unknown opcodes. +local function unknown(ctx) + clearprefixes(ctx) + return putop(ctx, "(unknown)") +end + +-- Return an immediate of the specified size. +local function getimm(ctx, pos, n) + if pos+n-1 > ctx.stop then return incomplete(ctx) end + local code = ctx.code + if n == 1 then + local b1 = byte(code, pos, pos) + return b1 + elseif n == 2 then + local b1, b2 = byte(code, pos, pos+1) + return b1+b2*256 + else + local b1, b2, b3, b4 = byte(code, pos, pos+3) + local imm = b1+b2*256+b3*65536+b4*16777216 + ctx.imm = imm + return imm + end +end + +-- Process pattern string and generate the operands. +local function putpat(ctx, name, pat) + local operands, regs, sz, mode, sp, rm, sc, rx, sdisp + local code, pos, stop = ctx.code, ctx.pos, ctx.stop + + -- Chars used: 1DFGIMPQRSTUVWXacdfgijmoprstuwxyz + for p in gmatch(pat, ".") do + local x = nil + if p == "V" or p == "U" then + if ctx.rexw then sz = "Q"; ctx.rexw = false + elseif ctx.o16 then sz = "W"; ctx.o16 = false + elseif p == "U" and ctx.x64 then sz = "Q" + else sz = "D" end + regs = map_regs[sz] + elseif p == "T" then + if ctx.rexw then sz = "Q"; ctx.rexw = false else sz = "D" end + regs = map_regs[sz] + elseif p == "B" then + sz = "B" + regs = ctx.rex and map_regs.B64 or map_regs.B + elseif match(p, "[WDQMXFG]") then + sz = p + regs = map_regs[sz] + elseif p == "P" then + sz = ctx.o16 and "X" or "M"; ctx.o16 = false + regs = map_regs[sz] + elseif p == "S" then + name = name..lower(sz) + elseif p == "s" then + local imm = getimm(ctx, pos, 1); if not imm then return end + x = imm <= 127 and format("+0x%02x", imm) + or format("-0x%02x", 256-imm) + pos = pos+1 + elseif p == "u" then + local imm = getimm(ctx, pos, 1); if not imm then return end + x = format("0x%02x", imm) + pos = pos+1 + elseif p == "w" then + local imm = getimm(ctx, pos, 2); if not imm then return end + x = format("0x%x", imm) + pos = pos+2 + elseif p == "o" then -- [offset] + if ctx.x64 then + local imm1 = getimm(ctx, pos, 4); if not imm1 then return end + local imm2 = getimm(ctx, pos+4, 4); if not imm2 then return end + x = format("[0x%08x%08x]", imm2, imm1) + pos = pos+8 + else + local imm = getimm(ctx, pos, 4); if not imm then return end + x = format("[0x%08x]", imm) + pos = pos+4 + end + elseif p == "i" or p == "I" then + local n = map_sz2n[sz] + if n == 8 and ctx.x64 and p == "I" then + local imm1 = getimm(ctx, pos, 4); if not imm1 then return end + local imm2 = getimm(ctx, pos+4, 4); if not imm2 then return end + x = format("0x%08x%08x", imm2, imm1) + else + if n == 8 then n = 4 end + local imm = getimm(ctx, pos, n); if not imm then return end + if sz == "Q" and (imm < 0 or imm > 0x7fffffff) then + imm = (0xffffffff+1)-imm + x = format(imm > 65535 and "-0x%08x" or "-0x%x", imm) + else + x = format(imm > 65535 and "0x%08x" or "0x%x", imm) + end + end + pos = pos+n + elseif p == "j" then + local n = map_sz2n[sz] + if n == 8 then n = 4 end + local imm = getimm(ctx, pos, n); if not imm then return end + if sz == "B" and imm > 127 then imm = imm-256 + elseif imm > 2147483647 then imm = imm-4294967296 end + pos = pos+n + imm = imm + pos + ctx.addr + if imm > 4294967295 and not ctx.x64 then imm = imm-4294967296 end + ctx.imm = imm + if sz == "W" then + x = format("word 0x%04x", imm%65536) + elseif ctx.x64 then + local lo = imm % 0x1000000 + x = format("0x%02x%06x", (imm-lo) / 0x1000000, lo) + else + x = format("0x%08x", imm) + end + elseif p == "R" then + local r = byte(code, pos-1, pos-1)%8 + if ctx.rexb then r = r + 8; ctx.rexb = false end + x = regs[r+1] + elseif p == "a" then x = regs[1] + elseif p == "c" then x = "cl" + elseif p == "d" then x = "dx" + elseif p == "1" then x = "1" + else + if not mode then + mode = ctx.mrm + if not mode then + if pos > stop then return incomplete(ctx) end + mode = byte(code, pos, pos) + pos = pos+1 + end + rm = mode%8; mode = (mode-rm)/8 + sp = mode%8; mode = (mode-sp)/8 + sdisp = "" + if mode < 3 then + if rm == 4 then + if pos > stop then return incomplete(ctx) end + sc = byte(code, pos, pos) + pos = pos+1 + rm = sc%8; sc = (sc-rm)/8 + rx = sc%8; sc = (sc-rx)/8 + if ctx.rexx then rx = rx + 8; ctx.rexx = false end + if rx == 4 then rx = nil end + end + if mode > 0 or rm == 5 then + local dsz = mode + if dsz ~= 1 then dsz = 4 end + local disp = getimm(ctx, pos, dsz); if not disp then return end + if mode == 0 then rm = nil end + if rm or rx or (not sc and ctx.x64 and not ctx.a32) then + if dsz == 1 and disp > 127 then + sdisp = format("-0x%x", 256-disp) + elseif disp >= 0 and disp <= 0x7fffffff then + sdisp = format("+0x%x", disp) + else + sdisp = format("-0x%x", (0xffffffff+1)-disp) + end + else + sdisp = format(ctx.x64 and not ctx.a32 and + not (disp >= 0 and disp <= 0x7fffffff) + and "0xffffffff%08x" or "0x%08x", disp) + end + pos = pos+dsz + end + end + if rm and ctx.rexb then rm = rm + 8; ctx.rexb = false end + if ctx.rexr then sp = sp + 8; ctx.rexr = false end + end + if p == "m" then + if mode == 3 then x = regs[rm+1] + else + local aregs = ctx.a32 and map_regs.D or ctx.aregs + local srm, srx = "", "" + if rm then srm = aregs[rm+1] + elseif not sc and ctx.x64 and not ctx.a32 then srm = "rip" end + ctx.a32 = false + if rx then + if rm then srm = srm.."+" end + srx = aregs[rx+1] + if sc > 0 then srx = srx.."*"..(2^sc) end + end + x = format("[%s%s%s]", srm, srx, sdisp) + end + if mode < 3 and + (not match(pat, "[aRrgp]") or match(pat, "t")) then -- Yuck. + x = map_sz2prefix[sz].." "..x + end + elseif p == "r" then x = regs[sp+1] + elseif p == "g" then x = map_segregs[sp+1] + elseif p == "p" then -- Suppress prefix. + elseif p == "f" then x = "st"..rm + elseif p == "x" then + if sp == 0 and ctx.lock and not ctx.x64 then + x = "CR8"; ctx.lock = false + else + x = "CR"..sp + end + elseif p == "y" then x = "DR"..sp + elseif p == "z" then x = "TR"..sp + elseif p == "t" then + else + error("bad pattern `"..pat.."'") + end + end + if x then operands = operands and operands..", "..x or x end + end + ctx.pos = pos + return putop(ctx, name, operands) +end + +-- Forward declaration. +local map_act + +-- Fetch and cache MRM byte. +local function getmrm(ctx) + local mrm = ctx.mrm + if not mrm then + local pos = ctx.pos + if pos > ctx.stop then return nil end + mrm = byte(ctx.code, pos, pos) + ctx.pos = pos+1 + ctx.mrm = mrm + end + return mrm +end + +-- Dispatch to handler depending on pattern. +local function dispatch(ctx, opat, patgrp) + if not opat then return unknown(ctx) end + if match(opat, "%|") then -- MMX/SSE variants depending on prefix. + local p + if ctx.rep then + p = ctx.rep=="rep" and "%|([^%|]*)" or "%|[^%|]*%|[^%|]*%|([^%|]*)" + ctx.rep = false + elseif ctx.o16 then p = "%|[^%|]*%|([^%|]*)"; ctx.o16 = false + else p = "^[^%|]*" end + opat = match(opat, p) + if not opat then return unknown(ctx) end +-- ctx.rep = false; ctx.o16 = false + --XXX fails for 66 f2 0f 38 f1 06 crc32 eax,WORD PTR [esi] + --XXX remove in branches? + end + if match(opat, "%$") then -- reg$mem variants. + local mrm = getmrm(ctx); if not mrm then return incomplete(ctx) end + opat = match(opat, mrm >= 192 and "^[^%$]*" or "%$(.*)") + if opat == "" then return unknown(ctx) end + end + if opat == "" then return unknown(ctx) end + local name, pat = match(opat, "^([a-z0-9 ]*)(.*)") + if pat == "" and patgrp then pat = patgrp end + return map_act[sub(pat, 1, 1)](ctx, name, pat) +end + +-- Get a pattern from an opcode map and dispatch to handler. +local function dispatchmap(ctx, opcmap) + local pos = ctx.pos + local opat = opcmap[byte(ctx.code, pos, pos)] + pos = pos + 1 + ctx.pos = pos + return dispatch(ctx, opat) +end + +-- Map for action codes. The key is the first char after the name. +map_act = { + -- Simple opcodes without operands. + [""] = function(ctx, name, pat) + return putop(ctx, name) + end, + + -- Operand size chars fall right through. + B = putpat, W = putpat, D = putpat, Q = putpat, + V = putpat, U = putpat, T = putpat, + M = putpat, X = putpat, P = putpat, + F = putpat, G = putpat, + + -- Collect prefixes. + [":"] = function(ctx, name, pat) + ctx[pat == ":" and name or sub(pat, 2)] = name + if ctx.pos - ctx.start > 5 then return unknown(ctx) end -- Limit #prefixes. + end, + + -- Chain to special handler specified by name. + ["*"] = function(ctx, name, pat) + return map_act[name](ctx, name, sub(pat, 2)) + end, + + -- Use named subtable for opcode group. + ["!"] = function(ctx, name, pat) + local mrm = getmrm(ctx); if not mrm then return incomplete(ctx) end + return dispatch(ctx, map_opcgroup[name][((mrm-(mrm%8))/8)%8+1], sub(pat, 2)) + end, + + -- o16,o32[,o64] variants. + sz = function(ctx, name, pat) + if ctx.o16 then ctx.o16 = false + else + pat = match(pat, ",(.*)") + if ctx.rexw then + local p = match(pat, ",(.*)") + if p then pat = p; ctx.rexw = false end + end + end + pat = match(pat, "^[^,]*") + return dispatch(ctx, pat) + end, + + -- Two-byte opcode dispatch. + opc2 = function(ctx, name, pat) + return dispatchmap(ctx, map_opc2) + end, + + -- Three-byte opcode dispatch. + opc3 = function(ctx, name, pat) + return dispatchmap(ctx, map_opc3[pat]) + end, + + -- VMX/SVM dispatch. + vm = function(ctx, name, pat) + return dispatch(ctx, map_opcvm[ctx.mrm]) + end, + + -- Floating point opcode dispatch. + fp = function(ctx, name, pat) + local mrm = getmrm(ctx); if not mrm then return incomplete(ctx) end + local rm = mrm%8 + local idx = pat*8 + ((mrm-rm)/8)%8 + if mrm >= 192 then idx = idx + 64 end + local opat = map_opcfp[idx] + if type(opat) == "table" then opat = opat[rm+1] end + return dispatch(ctx, opat) + end, + + -- REX prefix. + rex = function(ctx, name, pat) + if ctx.rex then return unknown(ctx) end -- Only 1 REX prefix allowed. + for p in gmatch(pat, ".") do ctx["rex"..p] = true end + ctx.rex = true + end, + + -- Special case for nop with REX prefix. + nop = function(ctx, name, pat) + return dispatch(ctx, ctx.rex and pat or "nop") + end, +} + +------------------------------------------------------------------------------ + +-- Disassemble a block of code. +local function disass_block(ctx, ofs, len) + if not ofs then ofs = 0 end + local stop = len and ofs+len or #ctx.code + ofs = ofs + 1 + ctx.start = ofs + ctx.pos = ofs + ctx.stop = stop + ctx.imm = nil + ctx.mrm = false + clearprefixes(ctx) + while ctx.pos <= stop do dispatchmap(ctx, ctx.map1) end + if ctx.pos ~= ctx.start then incomplete(ctx) end +end + +-- Extended API: create a disassembler context. Then call ctx:disass(ofs, len). +local function create_(code, addr, out) + local ctx = {} + ctx.code = code + ctx.addr = (addr or 0) - 1 + ctx.out = out or io.write + ctx.symtab = {} + ctx.disass = disass_block + ctx.hexdump = 16 + ctx.x64 = false + ctx.map1 = map_opc1_32 + ctx.aregs = map_regs.D + return ctx +end + +local function create64_(code, addr, out) + local ctx = create_(code, addr, out) + ctx.x64 = true + ctx.map1 = map_opc1_64 + ctx.aregs = map_regs.Q + return ctx +end + +-- Simple API: disassemble code (a string) at address and output via out. +local function disass_(code, addr, out) + create_(code, addr, out):disass() +end + +local function disass64_(code, addr, out) + create64_(code, addr, out):disass() +end + +-- Return register name for RID. +local function regname_(r) + if r < 8 then return map_regs.D[r+1] end + return map_regs.X[r-7] +end + +local function regname64_(r) + if r < 16 then return map_regs.Q[r+1] end + return map_regs.X[r-15] +end + +-- Public module functions. +module(...) + +create = create_ +create64 = create64_ +disass = disass_ +disass64 = disass64_ +regname = regname_ +regname64 = regname64_ + diff --git a/src/jit/dump.lua b/src/jit/dump.lua new file mode 100644 index 00000000..3d62c4ea --- /dev/null +++ b/src/jit/dump.lua @@ -0,0 +1,696 @@ +---------------------------------------------------------------------------- +-- LuaJIT compiler dump module. +-- +-- Copyright (C) 2005-2012 Mike Pall. All rights reserved. +-- Released under the MIT license. See Copyright Notice in luajit.h +---------------------------------------------------------------------------- +-- +-- This module can be used to debug the JIT compiler itself. It dumps the +-- code representations and structures used in various compiler stages. +-- +-- Example usage: +-- +-- luajit -jdump -e "local x=0; for i=1,1e6 do x=x+i end; print(x)" +-- luajit -jdump=im -e "for i=1,1000 do for j=1,1000 do end end" | less -R +-- luajit -jdump=is myapp.lua | less -R +-- luajit -jdump=-b myapp.lua +-- luajit -jdump=+aH,myapp.html myapp.lua +-- luajit -jdump=ixT,myapp.dump myapp.lua +-- +-- The first argument specifies the dump mode. The second argument gives +-- the output file name. Default output is to stdout, unless the environment +-- variable LUAJIT_DUMPFILE is set. The file is overwritten every time the +-- module is started. +-- +-- Different features can be turned on or off with the dump mode. If the +-- mode starts with a '+', the following features are added to the default +-- set of features; a '-' removes them. Otherwise the features are replaced. +-- +-- The following dump features are available (* marks the default): +-- +-- * t Print a line for each started, ended or aborted trace (see also -jv). +-- * b Dump the traced bytecode. +-- * i Dump the IR (intermediate representation). +-- r Augment the IR with register/stack slots. +-- s Dump the snapshot map. +-- * m Dump the generated machine code. +-- x Print each taken trace exit. +-- X Print each taken trace exit and the contents of all registers. +-- +-- The output format can be set with the following characters: +-- +-- T Plain text output. +-- A ANSI-colored text output +-- H Colorized HTML + CSS output. +-- +-- The default output format is plain text. It's set to ANSI-colored text +-- if the COLORTERM variable is set. Note: this is independent of any output +-- redirection, which is actually considered a feature. +-- +-- You probably want to use less -R to enjoy viewing ANSI-colored text from +-- a pipe or a file. Add this to your ~/.bashrc: export LESS="-R" +-- +------------------------------------------------------------------------------ + +-- Cache some library functions and objects. +local jit = require("jit") +assert(jit.version_num == 20000, "LuaJIT core/library version mismatch") +local jutil = require("jit.util") +local vmdef = require("jit.vmdef") +local funcinfo, funcbc = jutil.funcinfo, jutil.funcbc +local traceinfo, traceir, tracek = jutil.traceinfo, jutil.traceir, jutil.tracek +local tracemc, tracesnap = jutil.tracemc, jutil.tracesnap +local traceexitstub, ircalladdr = jutil.traceexitstub, jutil.ircalladdr +local bit = require("bit") +local band, shl, shr = bit.band, bit.lshift, bit.rshift +local sub, gsub, format = string.sub, string.gsub, string.format +local byte, char, rep = string.byte, string.char, string.rep +local type, tostring = type, tostring +local stdout, stderr = io.stdout, io.stderr + +-- Load other modules on-demand. +local bcline, disass + +-- Active flag, output file handle and dump mode. +local active, out, dumpmode + +------------------------------------------------------------------------------ + +local symtabmt = { __index = false } +local symtab = {} +local nexitsym = 0 + +-- Fill nested symbol table with per-trace exit stub addresses. +local function fillsymtab_tr(tr, nexit) + local t = {} + symtabmt.__index = t + if jit.arch == "mips" or jit.arch == "mipsel" then + t[traceexitstub(tr, 0)] = "exit" + return + end + for i=0,nexit-1 do + local addr = traceexitstub(tr, i) + t[addr] = tostring(i) + end + local addr = traceexitstub(tr, nexit) + if addr then t[addr] = "stack_check" end +end + +-- Fill symbol table with trace exit stub addresses. +local function fillsymtab(tr, nexit) + local t = symtab + if nexitsym == 0 then + local ircall = vmdef.ircall + for i=0,#ircall do + local addr = ircalladdr(i) + if addr ~= 0 then t[addr] = ircall[i] end + end + end + if nexitsym == 1000000 then -- Per-trace exit stubs. + fillsymtab_tr(tr, nexit) + elseif nexit > nexitsym then -- Shared exit stubs. + for i=nexitsym,nexit-1 do + local addr = traceexitstub(i) + if addr == nil then -- Fall back to per-trace exit stubs. + fillsymtab_tr(tr, nexit) + setmetatable(symtab, symtabmt) + nexit = 1000000 + break + end + t[addr] = tostring(i) + end + nexitsym = nexit + end + return t +end + +local function dumpwrite(s) + out:write(s) +end + +-- Disassemble machine code. +local function dump_mcode(tr) + local info = traceinfo(tr) + if not info then return end + local mcode, addr, loop = tracemc(tr) + if not mcode then return end + if not disass then disass = require("jit.dis_"..jit.arch) end + out:write("---- TRACE ", tr, " mcode ", #mcode, "\n") + local ctx = disass.create(mcode, addr, dumpwrite) + ctx.hexdump = 0 + ctx.symtab = fillsymtab(tr, info.nexit) + if loop ~= 0 then + symtab[addr+loop] = "LOOP" + ctx:disass(0, loop) + out:write("->LOOP:\n") + ctx:disass(loop, #mcode-loop) + symtab[addr+loop] = nil + else + ctx:disass(0, #mcode) + end +end + +------------------------------------------------------------------------------ + +local irtype_text = { + [0] = "nil", + "fal", + "tru", + "lud", + "str", + "p32", + "thr", + "pro", + "fun", + "p64", + "cdt", + "tab", + "udt", + "flt", + "num", + "i8 ", + "u8 ", + "i16", + "u16", + "int", + "u32", + "i64", + "u64", + "sfp", +} + +local colortype_ansi = { + [0] = "%s", + "%s", + "%s", + "\027[36m%s\027[m", + "\027[32m%s\027[m", + "%s", + "\027[1m%s\027[m", + "%s", + "\027[1m%s\027[m", + "%s", + "\027[33m%s\027[m", + "\027[31m%s\027[m", + "\027[36m%s\027[m", + "\027[34m%s\027[m", + "\027[34m%s\027[m", + "\027[35m%s\027[m", + "\027[35m%s\027[m", + "\027[35m%s\027[m", + "\027[35m%s\027[m", + "\027[35m%s\027[m", + "\027[35m%s\027[m", + "\027[35m%s\027[m", + "\027[35m%s\027[m", + "\027[35m%s\027[m", +} + +local function colorize_text(s, t) + return s +end + +local function colorize_ansi(s, t) + return format(colortype_ansi[t], s) +end + +local irtype_ansi = setmetatable({}, + { __index = function(tab, t) + local s = colorize_ansi(irtype_text[t], t); tab[t] = s; return s; end }) + +local html_escape = { ["<"] = "<", [">"] = ">", ["&"] = "&", } + +local function colorize_html(s, t) + s = gsub(s, "[<>&]", html_escape) + return format('%s', irtype_text[t], s) +end + +local irtype_html = setmetatable({}, + { __index = function(tab, t) + local s = colorize_html(irtype_text[t], t); tab[t] = s; return s; end }) + +local header_html = [[ + +]] + +local colorize, irtype + +-- Lookup tables to convert some literals into names. +local litname = { + ["SLOAD "] = setmetatable({}, { __index = function(t, mode) + local s = "" + if band(mode, 1) ~= 0 then s = s.."P" end + if band(mode, 2) ~= 0 then s = s.."F" end + if band(mode, 4) ~= 0 then s = s.."T" end + if band(mode, 8) ~= 0 then s = s.."C" end + if band(mode, 16) ~= 0 then s = s.."R" end + if band(mode, 32) ~= 0 then s = s.."I" end + t[mode] = s + return s + end}), + ["XLOAD "] = { [0] = "", "R", "V", "RV", "U", "RU", "VU", "RVU", }, + ["CONV "] = setmetatable({}, { __index = function(t, mode) + local s = irtype[band(mode, 31)] + s = irtype[band(shr(mode, 5), 31)].."."..s + if band(mode, 0x400) ~= 0 then s = s.." trunc" + elseif band(mode, 0x800) ~= 0 then s = s.." sext" end + local c = shr(mode, 14) + if c == 2 then s = s.." index" elseif c == 3 then s = s.." check" end + t[mode] = s + return s + end}), + ["FLOAD "] = vmdef.irfield, + ["FREF "] = vmdef.irfield, + ["FPMATH"] = vmdef.irfpm, +} + +local function ctlsub(c) + if c == "\n" then return "\\n" + elseif c == "\r" then return "\\r" + elseif c == "\t" then return "\\t" + elseif c == "\r" then return "\\r" + else return format("\\%03d", byte(c)) + end +end + +local function fmtfunc(func, pc) + local fi = funcinfo(func, pc) + if fi.loc then + return fi.loc + elseif fi.ffid then + return vmdef.ffnames[fi.ffid] + elseif fi.addr then + return format("C:%x", fi.addr) + else + return "(?)" + end +end + +local function formatk(tr, idx) + local k, t, slot = tracek(tr, idx) + local tn = type(k) + local s + if tn == "number" then + if k == 2^52+2^51 then + s = "bias" + else + s = format("%+.14g", k) + end + elseif tn == "string" then + s = format(#k > 20 and '"%.20s"~' or '"%s"', gsub(k, "%c", ctlsub)) + elseif tn == "function" then + s = fmtfunc(k) + elseif tn == "table" then + s = format("{%p}", k) + elseif tn == "userdata" then + if t == 12 then + s = format("userdata:%p", k) + else + s = format("[%p]", k) + if s == "[0x00000000]" then s = "NULL" end + end + elseif t == 21 then -- int64_t + s = sub(tostring(k), 1, -3) + if sub(s, 1, 1) ~= "-" then s = "+"..s end + else + s = tostring(k) -- For primitives. + end + s = colorize(format("%-4s", s), t) + if slot then + s = format("%s @%d", s, slot) + end + return s +end + +local function printsnap(tr, snap) + local n = 2 + for s=0,snap[1]-1 do + local sn = snap[n] + if shr(sn, 24) == s then + n = n + 1 + local ref = band(sn, 0xffff) - 0x8000 -- REF_BIAS + if ref < 0 then + out:write(formatk(tr, ref)) + elseif band(sn, 0x80000) ~= 0 then -- SNAP_SOFTFPNUM + out:write(colorize(format("%04d/%04d", ref, ref+1), 14)) + else + local m, ot, op1, op2 = traceir(tr, ref) + out:write(colorize(format("%04d", ref), band(ot, 31))) + end + out:write(band(sn, 0x10000) == 0 and " " or "|") -- SNAP_FRAME + else + out:write("---- ") + end + end + out:write("]\n") +end + +-- Dump snapshots (not interleaved with IR). +local function dump_snap(tr) + out:write("---- TRACE ", tr, " snapshots\n") + for i=0,1000000000 do + local snap = tracesnap(tr, i) + if not snap then break end + out:write(format("#%-3d %04d [ ", i, snap[0])) + printsnap(tr, snap) + end +end + +-- Return a register name or stack slot for a rid/sp location. +local function ridsp_name(ridsp) + if not disass then disass = require("jit.dis_"..jit.arch) end + local rid = band(ridsp, 0xff) + if ridsp > 255 then return format("[%x]", shr(ridsp, 8)*4) end + if rid < 128 then return disass.regname(rid) end + return "" +end + +-- Dump CALL* function ref and return optional ctype. +local function dumpcallfunc(tr, ins) + local ctype + if ins > 0 then + local m, ot, op1, op2 = traceir(tr, ins) + if band(ot, 31) == 0 then -- nil type means CARG(func, ctype). + ins = op1 + ctype = formatk(tr, op2) + end + end + if ins < 0 then + out:write(format("[0x%x](", tonumber((tracek(tr, ins))))) + else + out:write(format("%04d (", ins)) + end + return ctype +end + +-- Recursively gather CALL* args and dump them. +local function dumpcallargs(tr, ins) + if ins < 0 then + out:write(formatk(tr, ins)) + else + local m, ot, op1, op2 = traceir(tr, ins) + local oidx = 6*shr(ot, 8) + local op = sub(vmdef.irnames, oidx+1, oidx+6) + if op == "CARG " then + dumpcallargs(tr, op1) + if op2 < 0 then + out:write(" ", formatk(tr, op2)) + else + out:write(" ", format("%04d", op2)) + end + else + out:write(format("%04d", ins)) + end + end +end + +-- Dump IR and interleaved snapshots. +local function dump_ir(tr, dumpsnap, dumpreg) + local info = traceinfo(tr) + if not info then return end + local nins = info.nins + out:write("---- TRACE ", tr, " IR\n") + local irnames = vmdef.irnames + local snapref = 65536 + local snap, snapno + if dumpsnap then + snap = tracesnap(tr, 0) + snapref = snap[0] + snapno = 0 + end + for ins=1,nins do + if ins >= snapref then + if dumpreg then + out:write(format(".... SNAP #%-3d [ ", snapno)) + else + out:write(format(".... SNAP #%-3d [ ", snapno)) + end + printsnap(tr, snap) + snapno = snapno + 1 + snap = tracesnap(tr, snapno) + snapref = snap and snap[0] or 65536 + end + local m, ot, op1, op2, ridsp = traceir(tr, ins) + local oidx, t = 6*shr(ot, 8), band(ot, 31) + local op = sub(irnames, oidx+1, oidx+6) + if op == "LOOP " then + if dumpreg then + out:write(format("%04d ------------ LOOP ------------\n", ins)) + else + out:write(format("%04d ------ LOOP ------------\n", ins)) + end + elseif op ~= "NOP " and op ~= "CARG " and + (dumpreg or op ~= "RENAME") then + if dumpreg then + out:write(format("%04d %-5s ", ins, ridsp_name(ridsp))) + else + out:write(format("%04d ", ins)) + end + out:write(format("%s%s %s %s ", + band(ot, 128) == 0 and " " or ">", + band(ot, 64) == 0 and " " or "+", + irtype[t], op)) + local m1, m2 = band(m, 3), band(m, 3*4) + if sub(op, 1, 4) == "CALL" then + local ctype + if m2 == 1*4 then -- op2 == IRMlit + out:write(format("%-10s (", vmdef.ircall[op2])) + else + ctype = dumpcallfunc(tr, op2) + end + if op1 ~= -1 then dumpcallargs(tr, op1) end + out:write(")") + if ctype then out:write(" ctype ", ctype) end + elseif op == "CNEW " and op2 == -1 then + out:write(formatk(tr, op1)) + elseif m1 ~= 3 then -- op1 != IRMnone + if op1 < 0 then + out:write(formatk(tr, op1)) + else + out:write(format(m1 == 0 and "%04d" or "#%-3d", op1)) + end + if m2 ~= 3*4 then -- op2 != IRMnone + if m2 == 1*4 then -- op2 == IRMlit + local litn = litname[op] + if litn and litn[op2] then + out:write(" ", litn[op2]) + elseif op == "UREFO " or op == "UREFC " then + out:write(format(" #%-3d", shr(op2, 8))) + else + out:write(format(" #%-3d", op2)) + end + elseif op2 < 0 then + out:write(" ", formatk(tr, op2)) + else + out:write(format(" %04d", op2)) + end + end + end + out:write("\n") + end + end + if snap then + if dumpreg then + out:write(format(".... SNAP #%-3d [ ", snapno)) + else + out:write(format(".... SNAP #%-3d [ ", snapno)) + end + printsnap(tr, snap) + end +end + +------------------------------------------------------------------------------ + +local recprefix = "" +local recdepth = 0 + +-- Format trace error message. +local function fmterr(err, info) + if type(err) == "number" then + if type(info) == "function" then info = fmtfunc(info) end + err = format(vmdef.traceerr[err], info) + end + return err +end + +-- Dump trace states. +local function dump_trace(what, tr, func, pc, otr, oex) + if what == "stop" or (what == "abort" and dumpmode.a) then + if dumpmode.i then dump_ir(tr, dumpmode.s, dumpmode.r and what == "stop") + elseif dumpmode.s then dump_snap(tr) end + if dumpmode.m then dump_mcode(tr) end + end + if what == "start" then + if dumpmode.H then out:write('
\n') end
+    out:write("---- TRACE ", tr, " ", what)
+    if otr then out:write(" ", otr, "/", oex) end
+    out:write(" ", fmtfunc(func, pc), "\n")
+    recprefix = ""
+  elseif what == "stop" or what == "abort" then
+    out:write("---- TRACE ", tr, " ", what)
+    recprefix = nil
+    if what == "abort" then
+      out:write(" ", fmtfunc(func, pc), " -- ", fmterr(otr, oex), "\n")
+    else
+      local info = traceinfo(tr)
+      local link, ltype = info.link, info.linktype
+      if link == tr or link == 0 then
+	out:write(" -> ", ltype, "\n")
+      elseif ltype == "root" then
+	out:write(" -> ", link, "\n")
+      else
+	out:write(" -> ", link, " ", ltype, "\n")
+      end
+    end
+    if dumpmode.H then out:write("
\n\n") else out:write("\n") end + else + out:write("---- TRACE ", what, "\n\n") + end + out:flush() +end + +-- Dump recorded bytecode. +local function dump_record(tr, func, pc, depth, callee) + if depth ~= recdepth then + recdepth = depth + recprefix = rep(" .", depth) + end + local line + if pc >= 0 then + line = bcline(func, pc, recprefix) + if dumpmode.H then line = gsub(line, "[<>&]", html_escape) end + else + line = "0000 "..recprefix.." FUNCC \n" + callee = func + end + if pc <= 0 then + out:write(sub(line, 1, -2), " ; ", fmtfunc(func), "\n") + else + out:write(line) + end + if pc >= 0 and band(funcbc(func, pc), 0xff) < 16 then -- ORDER BC + out:write(bcline(func, pc+1, recprefix)) -- Write JMP for cond. + end +end + +------------------------------------------------------------------------------ + +-- Dump taken trace exits. +local function dump_texit(tr, ex, ngpr, nfpr, ...) + out:write("---- TRACE ", tr, " exit ", ex, "\n") + if dumpmode.X then + local regs = {...} + if jit.arch == "x64" then + for i=1,ngpr do + out:write(format(" %016x", regs[i])) + if i % 4 == 0 then out:write("\n") end + end + else + for i=1,ngpr do + out:write(format(" %08x", regs[i])) + if i % 8 == 0 then out:write("\n") end + end + end + if jit.arch == "mips" or jit.arch == "mipsel" then + for i=1,nfpr,2 do + out:write(format(" %+17.14g", regs[ngpr+i])) + if i % 8 == 7 then out:write("\n") end + end + else + for i=1,nfpr do + out:write(format(" %+17.14g", regs[ngpr+i])) + if i % 4 == 0 then out:write("\n") end + end + end + end +end + +------------------------------------------------------------------------------ + +-- Detach dump handlers. +local function dumpoff() + if active then + active = false + jit.attach(dump_texit) + jit.attach(dump_record) + jit.attach(dump_trace) + if out and out ~= stdout and out ~= stderr then out:close() end + out = nil + end +end + +-- Open the output file and attach dump handlers. +local function dumpon(opt, outfile) + if active then dumpoff() end + + local colormode = os.getenv("COLORTERM") and "A" or "T" + if opt then + opt = gsub(opt, "[TAH]", function(mode) colormode = mode; return ""; end) + end + + local m = { t=true, b=true, i=true, m=true, } + if opt and opt ~= "" then + local o = sub(opt, 1, 1) + if o ~= "+" and o ~= "-" then m = {} end + for i=1,#opt do m[sub(opt, i, i)] = (o ~= "-") end + end + dumpmode = m + + if m.t or m.b or m.i or m.s or m.m then + jit.attach(dump_trace, "trace") + end + if m.b then + jit.attach(dump_record, "record") + if not bcline then bcline = require("jit.bc").line end + end + if m.x or m.X then + jit.attach(dump_texit, "texit") + end + + if not outfile then outfile = os.getenv("LUAJIT_DUMPFILE") end + if outfile then + out = outfile == "-" and stdout or assert(io.open(outfile, "w")) + else + out = stdout + end + + m[colormode] = true + if colormode == "A" then + colorize = colorize_ansi + irtype = irtype_ansi + elseif colormode == "H" then + colorize = colorize_html + irtype = irtype_html + out:write(header_html) + else + colorize = colorize_text + irtype = irtype_text + end + + active = true +end + +-- Public module functions. +module(...) + +on = dumpon +off = dumpoff +start = dumpon -- For -j command line option. + diff --git a/src/jit/v.lua b/src/jit/v.lua new file mode 100644 index 00000000..8f58fe3b --- /dev/null +++ b/src/jit/v.lua @@ -0,0 +1,167 @@ +---------------------------------------------------------------------------- +-- Verbose mode of the LuaJIT compiler. +-- +-- Copyright (C) 2005-2012 Mike Pall. All rights reserved. +-- Released under the MIT license. See Copyright Notice in luajit.h +---------------------------------------------------------------------------- +-- +-- This module shows verbose information about the progress of the +-- JIT compiler. It prints one line for each generated trace. This module +-- is useful to see which code has been compiled or where the compiler +-- punts and falls back to the interpreter. +-- +-- Example usage: +-- +-- luajit -jv -e "for i=1,1000 do for j=1,1000 do end end" +-- luajit -jv=myapp.out myapp.lua +-- +-- Default output is to stderr. To redirect the output to a file, pass a +-- filename as an argument (use '-' for stdout) or set the environment +-- variable LUAJIT_VERBOSEFILE. The file is overwritten every time the +-- module is started. +-- +-- The output from the first example should look like this: +-- +-- [TRACE 1 (command line):1 loop] +-- [TRACE 2 (1/3) (command line):1 -> 1] +-- +-- The first number in each line is the internal trace number. Next are +-- the file name ('(command line)') and the line number (':1') where the +-- trace has started. Side traces also show the parent trace number and +-- the exit number where they are attached to in parentheses ('(1/3)'). +-- An arrow at the end shows where the trace links to ('-> 1'), unless +-- it loops to itself. +-- +-- In this case the inner loop gets hot and is traced first, generating +-- a root trace. Then the last exit from the 1st trace gets hot, too, +-- and triggers generation of the 2nd trace. The side trace follows the +-- path along the outer loop and *around* the inner loop, back to its +-- start, and then links to the 1st trace. Yes, this may seem unusual, +-- if you know how traditional compilers work. Trace compilers are full +-- of surprises like this -- have fun! :-) +-- +-- Aborted traces are shown like this: +-- +-- [TRACE --- foo.lua:44 -- leaving loop in root trace at foo:lua:50] +-- +-- Don't worry -- trace aborts are quite common, even in programs which +-- can be fully compiled. The compiler may retry several times until it +-- finds a suitable trace. +-- +-- Of course this doesn't work with features that are not-yet-implemented +-- (NYI error messages). The VM simply falls back to the interpreter. This +-- may not matter at all if the particular trace is not very high up in +-- the CPU usage profile. Oh, and the interpreter is quite fast, too. +-- +-- Also check out the -jdump module, which prints all the gory details. +-- +------------------------------------------------------------------------------ + +-- Cache some library functions and objects. +local jit = require("jit") +assert(jit.version_num == 20000, "LuaJIT core/library version mismatch") +local jutil = require("jit.util") +local vmdef = require("jit.vmdef") +local funcinfo, traceinfo = jutil.funcinfo, jutil.traceinfo +local type, format = type, string.format +local stdout, stderr = io.stdout, io.stderr + +-- Active flag and output file handle. +local active, out + +------------------------------------------------------------------------------ + +local startloc, startex + +local function fmtfunc(func, pc) + local fi = funcinfo(func, pc) + if fi.loc then + return fi.loc + elseif fi.ffid then + return vmdef.ffnames[fi.ffid] + elseif fi.addr then + return format("C:%x", fi.addr) + else + return "(?)" + end +end + +-- Format trace error message. +local function fmterr(err, info) + if type(err) == "number" then + if type(info) == "function" then info = fmtfunc(info) end + err = format(vmdef.traceerr[err], info) + end + return err +end + +-- Dump trace states. +local function dump_trace(what, tr, func, pc, otr, oex) + if what == "start" then + startloc = fmtfunc(func, pc) + startex = otr and "("..otr.."/"..oex..") " or "" + else + if what == "abort" then + local loc = fmtfunc(func, pc) + if loc ~= startloc then + out:write(format("[TRACE --- %s%s -- %s at %s]\n", + startex, startloc, fmterr(otr, oex), loc)) + else + out:write(format("[TRACE --- %s%s -- %s]\n", + startex, startloc, fmterr(otr, oex))) + end + elseif what == "stop" then + local info = traceinfo(tr) + local link, ltype = info.link, info.linktype + if ltype == "interpreter" then + out:write(format("[TRACE %3s %s%s -- fallback to interpreter]\n", + tr, startex, startloc)) + elseif link == tr or link == 0 then + out:write(format("[TRACE %3s %s%s %s]\n", + tr, startex, startloc, ltype)) + elseif ltype == "root" then + out:write(format("[TRACE %3s %s%s -> %d]\n", + tr, startex, startloc, link)) + else + out:write(format("[TRACE %3s %s%s -> %d %s]\n", + tr, startex, startloc, link, ltype)) + end + else + out:write(format("[TRACE %s]\n", what)) + end + out:flush() + end +end + +------------------------------------------------------------------------------ + +-- Detach dump handlers. +local function dumpoff() + if active then + active = false + jit.attach(dump_trace) + if out and out ~= stdout and out ~= stderr then out:close() end + out = nil + end +end + +-- Open the output file and attach dump handlers. +local function dumpon(outfile) + if active then dumpoff() end + if not outfile then outfile = os.getenv("LUAJIT_VERBOSEFILE") end + if outfile then + out = outfile == "-" and stdout or assert(io.open(outfile, "w")) + else + out = stderr + end + jit.attach(dump_trace, "trace") + active = true +end + +-- Public module functions. +module(...) + +on = dumpon +off = dumpoff +start = dumpon -- For -j command line option. + diff --git a/src/msvcbuild.bat b/src/msvcbuild.bat index 09782db4..63467308 100644 --- a/src/msvcbuild.bat +++ b/src/msvcbuild.bat @@ -46,7 +46,7 @@ buildvm -m libdef -o lj_libdef.h %ALL_LIB% @if errorlevel 1 goto :BAD buildvm -m recdef -o lj_recdef.h %ALL_LIB% @if errorlevel 1 goto :BAD -buildvm -m vmdef -o ..\lib\vmdef.lua %ALL_LIB% +buildvm -m vmdef -o jit\vmdef.lua %ALL_LIB% @if errorlevel 1 goto :BAD buildvm -m folddef -o lj_folddef.h lj_opt_fold.c @if errorlevel 1 goto :BAD -- cgit v1.2.3-55-g6feb