signal/m68k: Use force_sigsegv(SIGSEGV) in fpsp040_die
authorEric W. Biederman <ebiederm@xmission.com>
Mon, 26 Jul 2021 19:23:11 +0000 (14:23 -0500)
committerEric W. Biederman <ebiederm@xmission.com>
Thu, 26 Aug 2021 14:59:15 +0000 (09:59 -0500)
commita3616a3c02722d1edb95acc7fceade242f6553ba
tree50f31503491dafdc92a75b6946970ebfdb4ea93e
parentb48c7236b13cb5ef1b5fdf744aa8841df0f7b43a
signal/m68k: Use force_sigsegv(SIGSEGV) in fpsp040_die

In the fpsp040 code when copyin or copyout fails call
force_sigsegv(SIGSEGV) instead of do_exit(SIGSEGV).

This solves a couple of problems.  Because do_exit embeds the ptrace
stop PTRACE_EVENT_EXIT a complete stack frame needs to be present for
that to work correctly.  There is always the information needed for a
ptrace stop where get_signal is called.  So exiting with a signal
solves the ptrace issue.

Further exiting with a signal ensures that all of the threads in a
process are killed not just the thread that malfunctioned.  Which
avoids confusing userspace.

To make force_sigsegv(SIGSEGV) work in fpsp040_die modify the code to
save all of the registers and jump to ret_from_exception (which
ultimately calls get_signal) after fpsp040_die returns.

v2: Updated the branches to use gas's pseudo ops that automatically
    calculate the best branch instruction to use for the purpose.

v1: https://lkml.kernel.org/r/87a6m8kgtx.fsf_-_@disp2133
Link: https://lkml.kernel.org/r/87tukghjfs.fsf_-_@disp2133
Acked-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
arch/m68k/fpsp040/skeleton.S
arch/m68k/kernel/traps.c