Hi, Stuart! Thank You! you've nailed the core of issue.
1) used /usr/local/bin/egdb for ioquake3 from packages. result: Thread 1 received signal SIGILL, Illegal instruction. 0x0000082c2633008b in ?? () (gdb) address of 'illegal instruction' seems to be a random value.. no other errors appear so far. 2) for current git build used: $ gmake LDFLAGS="-z nobtcfi" result - 100% working binaries (pass instruction to the linker) 3) i'm still not good/comfortable with ports system, so didn't check 'USE_NOBTCFI=Yes' option, sorry.. 4) for c++/eg++ (other builds with similar failure) guess that: -fcf-protection=[full|branch|return] could do the trick (? branch ?). will check later. Thanks again, best regards On Fri, Apr 19, 2024 at 08:04 +0100, Stuart Henderson wrote: > On 2024/04/19 01:46, sda wrote: > > Dwarf Error: wrong version in compilation unit header (is 4, should be > > 2) [in module /usr/libexec/ld.so] > > That's because of the ancient gdb that's in base. There's a slightly > less old one in ports. See "Debug Packages, Debuggers and Backtraces" > in https://www.openbsd.org/faq/ports/ports.html > > > Program received signal SIGILL, Illegal instruction. > > 0x000008c46184408b in ?? () > > (gdb) > > core dumped. (size 160Mb) > > Most likely this is due to the 'indirect branch tracking' cpu feature > being triggered. Building from ports with USE_NOBTCFI=Yes will likely > work around it, the proper fix would be to add 'endbr64' instructiond > at the location of jump targets in assembly code. See if the above > info on gdb will get you a backtrace. >
