On Wed, 3 Jun 2020 at 21:10, Marco Elver <el...@google.com> wrote: > > On Wed, 3 Jun 2020 at 20:16, Peter Zijlstra <pet...@infradead.org> wrote: > > > > On Wed, Jun 03, 2020 at 06:07:22PM +0200, Peter Zijlstra wrote: > > > On Wed, Jun 03, 2020 at 04:47:54PM +0200, Marco Elver wrote: > > > > > > With that in mind, you could whitelist "__ubsan_handle"-prefixed > > > > functions in objtool. Given the __always_inline+noinstr+__ubsan_handle > > > > case is quite rare, it might be reasonable. > > > > > > Yes, I think so. Let me go have dinner and then I'll try and do a patch > > > to that effect. > > > > Here's a slightly more radical patch, it unconditionally allows UBSAN. > > > > I've not actually boot tested this.. yet. > > > > --- > > Subject: x86/entry, ubsan, objtool: Whitelist __ubsan_handle_*() > > From: Peter Zijlstra <pet...@infradead.org> > > Date: Wed Jun 3 20:09:06 CEST 2020 > > > > The UBSAN instrumentation only inserts external CALLs when things go > > 'BAD', much like WARN(). So treat them similar to WARN()s for noinstr, > > that is: allow them, at the risk of taking the machine down, to get > > their message out. > > > > Suggested-by: Marco Elver <el...@google.com> > > Signed-off-by: Peter Zijlstra (Intel) <pet...@infradead.org> > > This is much cleaner, as it gets us UBSAN coverage back. Seems to work > fine for me (only lightly tested), so > > Acked-by: Marco Elver <el...@google.com> > > Thanks!
I was thinking that if we remove __no_sanitize_undefined from noinstr, we can lift the hard compiler restriction for UBSAN because __no_sanitize_undefined isn't used anywhere. Turns out, that attribute isn't broken on GCC <= 7, so I've sent v2 of my series: https://lkml.kernel.org/r/20200604055811.247298-1-el...@google.com Thanks, -- Marco
