RFR: 8189596: AArch64: implementation for Thread-local handshakes

[Andrew Haley]

On 24/11/17 16:19, Erik Österlund wrote:

> But I would be *very* surprised if it would get anywhere close to
> millisecond levels, which is typically when it starts being
> noticeable on the radar at all to a safepointing operation (compared
> to the time of doing the crazy things we do in safepoints).

It's very unlikely but not impossible.

> A gut feeling is that it should be comparable to the time for
> performing a dmb sy (which by contract makes preceding stores
> globally observable in the whole system).

It's not really comparable.  It can be microseconds before a store
from one core is picked up by another; I have measured times as long
as that.  On the other hand, a full barrier can be a purely local
operation: all it has to do is flush the store buffer to the local
cache and mark the entries in the invalidate queue.  That takes a
matter of nanoseconds and the core can proceed, but of course, it does
slow down subsequent load and store operations.

-- 
Andrew Haley
Java Platform Lead Engineer
Red Hat UK Ltd. <https://www.redhat.com>
EAC8 43EB D3EF DB98 CC77 2FAD A5CD 6035 332F A671