Re: [PATCH] init: use KERNEL_DS when trying to start init process

[Al Viro <viro@ZenIV.linux.org.uk>]

On Wed, Jun 08, 2011 at 02:14:56PM +0200, Mathias Krause wrote:

> SPARC, TILE and Xtensa don't call set_fs(USER_DS), albeit have
> different definitions for USER_DS and KERNEL_DS. So those might need
> fixing. I'm not familiar with those architectures, so someone else has
> to answer this.
 
sparc (both sparc32 and sparc64) does that in flush_thread() (i.e. triggered
by flush_old_exec()); the only difference is that sparc64 is trying to avoid
writing to %asi if we already had USER_DS.  Any failure exit past the call
of flush_old_exec() will send us a SIGKILL (and will not return -ENOEXEC,
so no further handlers will be called anyway).

No idea about tile and xtensa - asking on linux-arch might be a good idea.

FWIW, looking at the ->load_binary() instances...  binfmt_som does not
bother with SIGKILL, which is Not Nice(tm) - there's nowhere to return
from sys_execve() at that point.  binfmt_elf_fdpic.c has a couple of
bogosities - it sends SIGSEGV instead of SIGKILL (which is probably OK,
since signal handlers are already switched to default, and SIGSEGV would
kill just as well as SIGKILL; the only question is whether the state of
process is suitable for coredump at that point) *and* we have one case
where both SIGKILL and SIGSEGV are sent (setup_arg_pages() failure).
And binfmt_flat looks just plain weird wrt failure exits...