[PATCH v19 00/27] riscv control-flow integrity for usermode

[Florian Weimer]

* Deepak Gupta:

> How will they contribute to CFI bringup without having a CFI compiled
> usersapce?

Build glibc themselves and then proceed one library at the time.

>>Another use case would be running container images with CFI on a
>>distribution kernel which supports pre-RVA23 hardware.
>
> Container image with CFI will have glibc and ld (and all other
> userspace) also compiled with shadow stack instructions in it. As soon
> as you take this container image to a pre-RVA23 hardware, you won't
> even reach vDSO. It'll break much before that, unless kernel is taking
> a trap on all sspush/sspopchk instructions in prologue/epilogue of
> functions in userspace (glibc, ld, etc)

The idea is that you can use a stock distribution kernel to run CFI
images (potentially form a different distribution or version of the
distribution).

But maybe none of this really matters.  How far out is CFI-checking
hardware?  Is it going to arrive much later than the RVA23 flag day
that people are suggesting?

Thanks,
Florian