[aarch64-port-dev ] 8248238: Implementation of JEP: Windows AArch64 Support

[Andrew Haley]

On 26/08/2020 16:04, Bernhard Urban-Forster wrote:

> Apparently it doesn't know :-( _WriteBarrier is a nop what code
> generation is concerned, so it will emit a dmb ishst. How this
> doesn't cause any major issues is beyond my understanding.

The answer to that is probably that the actual physical
implementations of AArch64 are more conservative about their memory
ordering than they need to be.

However, someone comes along with a new implementation, and bang. This
has been seen a few times, such as recently when Felix Yang fixed
8248219: aarch64: missing memory barrier in fast_storefield and
fast_accessfield.

> FWIW the Linux implementation does turn into dmb ish with GCC 9.

Yep, that's what we need.

> I'll look into it, but this will need some more testing.

Right.

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