Re: BUG: core dump in selftest of proc fs

[Alexey Dobriyan]

On Wed, Jan 04, 2023 at 07:27:40PM +0100, Mirsad Goran Todorovac wrote:
> Dear all,
> 
> Trying to complete `make kselftest` for the first time, so maybe I'm doing something wrong?
> 
> Or we are having a regression in 6.2-rc2 release candidate ...
> 
> However, the output of selftest run is:
> 
> make[2]: Entering directory '.../linux_torvalds/tools/testing/selftests/proc'
> TAP version 13
> 1..21
> # selftests: proc: fd-001-lookup
> ok 1 selftests: proc: fd-001-lookup
> # selftests: proc: fd-002-posix-eq
> ok 2 selftests: proc: fd-002-posix-eq
> # selftests: proc: fd-003-kthread
> ok 3 selftests: proc: fd-003-kthread
> # selftests: proc: proc-loadavg-001
> ok 4 selftests: proc: proc-loadavg-001
> # selftests: proc: proc-empty-vm
> # proc-empty-vm: proc-empty-vm.c:184: test_proc_pid_maps: Assertion `rv == 0' failed.
> # /usr/bin/timeout: the monitored command dumped core
> # Aborted
> not ok 5 selftests: proc: proc-empty-vm # exit=134
> # selftests: proc: proc-pid-vm
> # proc-pid-vm: proc-pid-vm.c:365: main: Assertion `rv == len' failed.
> # /usr/bin/timeout: the monitored command dumped core
> # Aborted
> 
> Please find attached lshw output, dmesg, config and lsmod.
> 
> I am available for further diagnostics.
> 
> The platform is Ubuntu 22.10 kinetic kudu on a Lenovo Ideapad 3 15ITL6 laptop.

The "bug" is that "call rel32" instruction testing for executable
vsyscall page which should be relocated to "call 0xffffffffff600000"
is messed up. Ubuntu 22.10 ships with "vsyscall=xonly" so there should not be
any faults when executing from it. But segfault happens with normal
randomised userspace address.

I'll change it to "call *rax" which should be more robust (and works)
and free from relocations.