Since r336439 we simply take the session pointer value mod the number of
worker threads (ncpu by default). On small systems this ends up
funneling all completion work through a single thread, which becomes a
bottleneck when processing IPSec traffic using hardware crypto drivers.
(Software drivers such as aesni(4) are unaffected since they invoke
completion handlers synchonously.)
Instead, maintain an incrementing counter with a unique value per
session, and use that to distribute work to completion threads.
Reviewed by: cem, jhb
MFC after: 2 weeks
Sponsored by: Rubicon Communications, LLC ("Netgate")
Differential Revision: https://reviews.freebsd.org/D28159
98d788c867