> If only I knew how to use UML... See http://user-mode-linux.sourceforge.net/ for instructions and download some old kernel/fs images from there to test. These old builds are good for verifying the lack of regressions. It's also very easy to build a fresh UML kernel from your own sources: make ARCH=um defconfig; make ARCH=um; ./vmlinux root=...
Without an fs image you can test that it boots as far as the "Cannot open root device" error, and that the startup messages match what you see running on the vanilla kernel as host. > Yes, thanks. With the patches I sent the tracee stops after both > audit_syscall_entry() and audit_syscall_exit() were already called. > > If the tracer changes (say) ->orig_ax, this won't be visible to audit. Right. It's hard to claim that's not utterly useless, but still... > Oh. Then we should stop in utrace_report_syscall_entry() path and > we need the additional complications to handle SYSEMU_SINGLESTEP > correctly. Although these complications are straightforward. If we think we care, yes. > Can't we ignore this difference for now? Yes. I also think you can ask the UML people if they ever care at all and if they don't then it will be OK upstream to change these corners. > Not sure this is adequate testing, but I compiled arch/um/os-Linux/start_up.c > as user-space program (I had to remove some unnecessary stuff from *.h) and > it prints > > Checking that ptrace can change system call numbers...OK > Checking syscall emulation patch for ptrace...OK > Checking advanced syscall emulation patch for ptrace...OK That's certainly a good sign! We do need to also test at least that using some old UML kernel+fs images we get as far as booting userland and running a few processes. Thanks, Roland
