Tons of uninitialized memory access bugs

[clbr]

Tons of uninitialized memory access bugs

[clbr]

When built with -g -fno-inline to get more accurate logs, it straight up segfaults.

==2694== Command: armips -root rspasm -sym getbitmap.sym getbitmap.s
==2694== Parent PID: 2064
==2694==
==2694==
==2694== Process terminating with default action of signal 11 (SIGSEGV)
==2694==  Bad permissions for mapped region at address 0x72FB48
==2694==    at 0x72FB48: ??? (in /tiedostot/mv/c/armips/armips)
==2694==    by 0x412835: tinyformat::detail::formatImpl(std::basic_ostream<wchar_t, std::char_traits<wchar_t> >&, wchar_t const*, tinyformat::detail::FormatArg const*, int) (tinyformat.h:815)
==2694==    by 0x4134C4: _ZN10tinyformat6formatIJmEEENSt7__cxx1112basic_stringIwSt11char_traitsIwESaIwEEEPKwDpRKT_ (in /tiedostot/mv/c/armips/armips)
==2694==    by 0x47D537: _ZN11SymbolTable18getUniqueLabelNameB5cxx11Eb (SymbolTable.cpp:172)
==2694==    by 0x423D45: _Z15preprocessMacroB5cxx11PKwR17MipsImmediateData (MipsMacros.cpp:58)
==2694==    by 0x4245B9: generateMipsMacroLoadStore(Parser&, MipsRegisterData&, MipsImmediateData&, int) (MipsMacros.cpp:215)
==2694==    by 0x42851E: MipsParser::parseMacro(Parser&) (MipsParser.cpp:1578)
==2694==    by 0x41F45A: CMipsArchitecture::parseOpcode(Parser&) (Mips.cpp:29)
==2694==    by 0x48AF4F: Parser::parseCommand() (Parser.cpp:728)
==2694==    by 0x48A192: Parser::parseCommandSequence(wchar_t, std::initializer_list<wchar_t const*>) (Parser.cpp:144)
==2694==    by 0x481217: parseDirectiveArea(Parser&, int) (DirectivesParser.cpp:459)
==2694==    by 0x48988E: Parser::parseDirective(std::unordered_multimap<std::__cxx11::basic_string<wchar_t, std::char_traits<wchar_t>, std::allocator<wchar_t> >, DirectiveEntry const, std::hash<std::__cxx11::basic_string<wchar_t, std::char_traits<wchar_t>, std::allocator<wchar_t> > >, std::equal_to<std::__cxx11::basic_string<wchar_t, std::char_traits<wchar_t>, std::allocator<wchar_t> > >, std::allocator<std::pair<std::__cxx11::basic_string<wchar_t, std::char_traits<wchar_t>, std::allocator<wchar_t> > const, DirectiveEntry const> > > const&) (Parser.cpp:242)

[Kingcom]

Can you provide an example file that produces this problem?

[clbr]

You mean it doesn't blow up for you? That many locations should effect other targets than just rsp.

Sec, will grab a test file.

[clbr]

Seeing if github upload works.

[clbr]

Nope, doesn't work. Uploaded to https://anonymousfiles.io/m1Rl3kid/

[Kingcom]

I've had no issues with it on Windows with VS 2017 using both debug and release configurations. On Ubuntu 20.04 with GCC 9 it also ran fine, though Valgrind showed some of the same uninitialized value warnings. ASan had no complaints though.

What are your build settings?

[clbr]

gcc 10.2, -O3 -march=nocona -mtune=k8 -fomit-frame-pointer

[Kingcom]

Okay, I could reproduce the crash with GCC 9 as well when using -O3 and -fno-inline (Clang 10 works). It seems to crash whenever tinyformat is used at all. Not sure what would cause it though...

[clbr]

There seem to be newer tinyformat versions, but it appears the copy here is modified to be wchar. It could also be a bug in tinyformat, even the new versions, of course.

Just a note, asan catches a small fraction of the stuff valgrind does, I believe even their own docs say so. It's meant to be fast, not complete.