labath wrote: > > Can we make this less brute force? I believe searching for the Build ID > > Note should be a completely deterministic process, without the need for > > heuristics. You start with the elf header, find the program headers, > > iterate to find the PT_NOTE segment (there could be more of them), and > > iterate over the notes until you find NT_GNU_BUILD_ID. All of these things > > should be in the memory (=> also in the core file) and within the first 4k > > of the mapping. There should be no need to search through a potentially > > huge file mapping just to see if it contains the build id. > > Yes, we can parse the program headers. That is a good idea as long as all the > info we need is mapped
It should all be there. In the files I looked at, both the program headers and all the note segments are in the first page of the mapping (which is the page containing the elf header). I can't find a reference for that now but I remember reading somewhere that this was intentional (to ensure the build id survives). BTW, we have some elf parsing code in `NativeProcessELF::GetELFImageInfoAddress()`, that you could draw inspiration from. https://github.com/llvm/llvm-project/pull/92078 _______________________________________________ lldb-commits mailing list lldb-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/lldb-commits