PDB single-step problem inline __asm Windows WinDbg

[prof79]

Hi there, I'm following exercises by https://patterndiagnostics.com which use inline assembly (__asm) in their book. They use VC++ but I tried clang as this was much easier compared to all the vcvarsall intricacies in build scripts and gave a quick win.

Stepping through the code in WinDbg I encountered misbehavior and it took me like a day to figure it out doubting flags, assembler syntax and my skills as a beginner and what not. I also had a stupid bug (erroneous line) first which didn't make things easier.

• Load the compiled EXE with symbols (PDB) in WinDbg (launch executable, maybe .reload /f).
• Break on the first lea instruction (bp main+0x12).
• Try to single-step (t)

Expected behavior:

• Step to next instruction (mov dword ptr [eax], 1)

What actually happens:

• Ending up at xor eax, eax (which is already in the prologue of clang-generated code (which is main+0x36)

This doesn't happen with the cl tooling from Visual Studio 2022.

I figured that the compiled EXE is actually fine, you can debug it without symbols and tracing will work as expected.

As a result, something must be wrong with the PDB generated by clang vs. what cl does as it causes wrong code steps.

I'll try to put a ZIP with the files somewhere.

Compilation line: clang -g -gcodeview -fasm-blocks --target=i386-windows-msvc -o PointersProjectCL86.exe PointersProject.cpp

[prof79]

Demo files - code plus binaries:

https://1drv.ms/f/s!Apk3bQT52ViAlugnQFKde1saMC1fow?e=pG3f10

[prof79]

Sorry, I forgot (Win11 23H2 22635.3570):

C:\> clang --version
clang version 18.1.4
Target: x86_64-pc-windows-msvc
Thread model: posix
InstalledDir: C:\Program Files\LLVM\bin
C:\>

[llvmbot]

@llvm/issue-subscribers-debuginfo

Author: Markus M. Egger (prof79)

Hi there, I'm following exercises by https://patterndiagnostics.com which use inline assembly (`__asm`) in their book. They use VC++ but I tried `clang` as this was much easier compared to all the `vcvarsall` intricacies in build scripts and gave a quick win.

Stepping through the code in WinDbg I encountered misbehavior and it took me like a day to figure it out doubting flags, assembler syntax and my skills as a beginner and what not. I also had a stupid bug (erroneous line) first which didn't make things easier.

• Load the compiled EXE with symbols (PDB) in WinDbg (launch executable, maybe .reload /f).
• Break on the first lea instruction (bp main+0x12).
• Try to single-step (t)

Expected behavior:

• Step to next instruction (mov dword ptr [eax], 1)

What actually happens:

• Ending up at xor eax, eax (which is already in the prologue of clang-generated code (which is main+0x36)

This doesn't happen with the cl tooling from Visual Studio 2022.

I figured that the compiled EXE is actually fine, you can debug it without symbols and tracing will work as expected.

As a result, something must be wrong with the PDB generated by clang vs. what cl does as it causes wrong code steps.

I'll try to put a ZIP with the files somewhere.

Compilation line: clang -g -gcodeview -fasm-blocks --target=i386-windows-msvc -o PointersProjectCL86.exe PointersProject.cpp

[dwblaikie]

@zmodem perhaps you can get this to some interested folks

[zmodem]

@amykhuang @ZequanWu could you take a look? It sounds like there's something wrong with line info for inline asm perhaps.