Joe Perches wrote:
> Rather than using stack, could you use
> u32 *secrets = &tcp_secret_secondary->secrets[0];
> 
That was because I thought it needed to be copied after the unlock.
Since Eric says it can be inside the locked path, I can do away with
that entirely, and just use the result (bakery) pointer itself.  Saves a
memcpy, too.

Here's my revised attempt (untested).  Any other technical corrections?