[Cygwin] lli crashes with --no-process-syms

[jeremyd2019]

lli -no-process-syms -lljit-platform=Inactive -emulated-tls -jit-kind=orc-lazy test/ExecutionEngine/OrcLazy/emulated-tls.ll

The ExecutionEngine/OrcLazy/emulated-tls.ll test fails with a segfault in

llvm-project/llvm/tools/lli/lli.cpp

Lines 1086 to 1089 in eb71fdd

	ExitOnErr(J->getProcessSymbolsJITDylib()->define(
	orc::absoluteSymbols({{J->mangleAndIntern("__main"),
	{orc::ExecutorAddr::fromPtr(mingw_noop_main),
	JITSymbolFlags::Exported}}})));

it appears that J->getProcessSymbolsJITDylib() returns NULL. Should there be a check of that return value before calling ->define on it, or is there some other place that is ensuring this is not NULL on MinGW that should also be used on Cygwin?

@mstorsjo

[mstorsjo]

it appears that J->getProcessSymbolsJITDylib() returns NULL. Should there be a check of that return value before calling ->define on it, or is there some other place that is ensuring this is not NULL on MinGW that should also be used on Cygwin?

Offhand I don't really know about this; I'd suggest just comparing with what's happening in the mingw case. The use of J->getProcessSymbolsJITDylib() here looks like it is something that really isn't supposed to return NULL...

[jeremyd2019]

Yeah, I guess I need to debug it on MinGW and see why it's not NULL there. The -no-process-syms option results in it being NULL.

[kikairoya]

I'm seeing a segmentation fault on MinGW too. Are you sure it's working as intended?
It looks like there's no code path that sets ProcessSymbols or SetupProcessSymbolsJITDylib when -no-process-syms is used.

$ lli -no-process-syms -lljit-platform=Inactive -emulated-tls -jit-kind=orc-lazy test/ExecutionEngine/OrcLazy/emulated-tls.ll
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace.
Stack dump:
0.      Program arguments: D:\\msys64\\ucrt64\\bin\\lli.exe -no-process-syms -lljit-platform=Inactive -emulated-tls -jit-kind=orc-lazy test/ExecutionEngine/OrcLazy/emulated-tls.ll
Exception Code: 0xC0000005
#0 0x00007ff7b81bc872 (D:\msys64\ucrt64\bin\lli.exe+0xc872)
#1 0x00007ff7b81cb5d6 (D:\msys64\ucrt64\bin\lli.exe+0x1b5d6)
#2 0x00007ff7b81b1318 (D:\msys64\ucrt64\bin\lli.exe+0x1318)
#3 0x00007ff7b81b1426 (D:\msys64\ucrt64\bin\lli.exe+0x1426)
#4 0x00007ffcaac97374 (C:\WINDOWS\System32\KERNEL32.DLL+0x17374)
#5 0x00007ffcab4dcc91 (C:\WINDOWS\SYSTEM32\ntdll.dll+0x4cc91)
Segmentation fault

$ uname; which lli
MINGW64_NT-10.0-19045
/ucrt64/bin/lli

test/ExecutionEngine/OrcLazy/emulated-tls.ll never runs on MinGW.

llvm-project/llvm/test/ExecutionEngine/OrcLazy/lit.local.cfg

Lines 19 to 20 in 0937233

	if sys.platform == "win32":
	config.unsupported = True

[jeremyd2019]

That's interesting. Perhaps there does need to be a guard against NULL after all