On Mon, Nov 04, 2019 at 03:44:39PM -0800, Sami Tolvanen wrote: > On Mon, Nov 4, 2019 at 9:11 AM Mark Rutland <mark.rutland@xxxxxxx> wrote: > > Can you please elaborate on _how_ this is incompatible in the commit > > message? > > > > For example, it's not clear to me if you mean that's functionally > > incompatible, or if you're trying to remove return-altering gadgets. > > > > If there's a functional incompatibility, please spell that out a bit > > more clearly. Likewise if this is about minimizing the set of places > > that can mess with control-flow outside of usual function conventions. > > Sure, I'll add a better description in v5. In this case, the return > address is modified in the kernel stack, which means the changes are > ignored with SCS. Ok, that makes sense to me. I'd suggest something like: | The graph tracer hooks returns by modifying frame records on the | (regular) stack, but with SCS the return address is taken from the | shadow stack, and the value in the frame record has no effect. As we | don't currently have a mechanism to determine the corresponding slot | on the shadow stack (and to pass this through the ftrace | infrastructure), for now let's disable the graph tracer when SCS is | enabled. ... as I suspect with some rework of the trampoline and common ftrace code we'd be able to correctly manipulate the shadow stack for this. Similarly, if clang gained -fpatchable-funciton-etnry, we'd get that for free. Thanks, Mark.
