[PATCH] arm64: signal: don't force known signals to SIGKILL

[mark.rutland@arm.com (Mark Rutland)]

On Wed, Apr 18, 2018 at 02:33:32PM +0100, Dave Martin wrote:
> On Wed, Apr 18, 2018 at 12:36:52PM +0100, Mark Rutland wrote:
> > On Tue, Apr 17, 2018 at 03:46:31PM +0100, Dave Martin wrote:
> > > On Mon, Apr 16, 2018 at 04:45:01PM +0100, Mark Rutland wrote:
> > > > -	if (WARN_ON(signal != SIGKILL ||
> > > > +	if (WARN_ON(signal != SIGKILL &&
> > > >  		    siginfo_layout(signal, code) != SIL_FAULT)) {
> > > >  		signal = SIGKILL;
> > 
> > Do you have a concern with the logical change here, or can I ask for
> > your ack? ;)
> 
> No concern, I just wanted to make sure I'd understood the background
> right.  The commit message for this patch doesn't fully explain why the
> original code was wrong and why the new code is right, and my memory of
> the context for this was a bit fuzzy.
> 
> I should probably have followed the Fixes tag for context, but I was
> lazy :P
> 

> Anyway, FWIW
> 
> Reviewed-by: Dave Martin <Dave.Martin@arm.com>

Cheers!

Catalin, are you happy to queue this as a fix for v4.17-rc2?

> Out of interest, what were you seeing in userspace prior to this fix?
> Please don't tell me systemd is functionally reliant on SIGILL :(

I spotted this with a pointer authentication test case I wrote, which I
expected to receive SIGILL when using an instruction trapped and UNDEF'd
by EL2.

Thanks,
Mark.