On 2020-06-20, Christian Weisgerber <[email protected]> wrote:
> I can't get this revision of the diff to work on amd64:
> * patch source
> * build and install kernel, reboot
> * make build
> * reboot -> "Process (pid 1) got signal 11"
>
> I'm at a loss. As part of the "make build", the new libc is installed
> and dynamically linked programs should already be using the userland
> gettime calls. Clearly this works. So why does init fail on the
> next reboot?
I can recover by extracting ./sbin/init from a snapshot in the
installer. After that, the system comes up fine in multiuser mode.
Nothing else appears to be affected, apart from init.
For a while, I had a reproducible situation.
When you call init(8) as a normal user in multiuser mode, it will
just exit with "init: Operation not permitted". Instead it would
segfault! I kept tweaking lib/libc/dlfcn/init.c, rebuilding and
reinstalling libc.a, rebuilding init, and watching it segfault.
None of the debug write(2)s I inserted would produce any output,
it seemed to die before ever reaching _libc_preinit(). I finally
ktraced it:
12420 ktrace RET ktrace 0
12420 ktrace CALL execve(0x7f7ffffec412,0x7f7ffffec298,0x7f7ffffec2a8)
12420 ktrace NAMI "./obj/init"
12420 ktrace ARGS
[0] = "./obj/init"
12420 init RET execve 0
12420 init PSIG SIGSEGV SIG_DFL code SEGV_MAPERR<1> addr=0x0 trapno=6
12420 init NAMI "init.core"
There's not even a kbind(2) there.
Then I removed the clearly useless debug write()s... and since then
I have a hard time reproducing the problem.
It doesn't make any sense.
--
Christian "naddy" Weisgerber [email protected]
