On Fri, Sep 07, 2018 at 09:02:45AM -0700, Kees Cook wrote:

I'll continue to investigate alternatives, but I wanted to point out
that the struct change actually fills an existing padding byte (so no
change in memory usage) and marking this as an unlikely() test means
it wouldn't even be measurable due to the branch predictor (so no
change in speed). encrypt/decrypt entry is a tiny tiny fraction of the
actual work done during encryption/decryption, etc.

The point is the ON_STACK request stuff is purely for backwards
compatibility and we don't want it to proliferate and pollute the
core API.

Cheers,
-- 
Email: Herbert Xu [off-list ref]
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt