RFR: 8316694: Implement relocation of nmethod within CodeCache [v4]
Vladimir Kozlov
kvn at openjdk.org
Mon Mar 17 21:52:11 UTC 2025
On Mon, 17 Mar 2025 21:38:45 GMT, Evgeny Astigeevich <eastigeevich at openjdk.org> wrote:
>> [Fix relocation requires](https://github.com/openjdk/jdk/blob/19154f7af34bf6f13d61d7a9f05d6277964845d8/src/hotspot/share/code/relocInfo.hpp#L858-L860) that inline caches be cleared which [must be done at a safepoint](https://github.com/openjdk/jdk/blob/19154f7af34bf6f13d61d7a9f05d6277964845d8/src/hotspot/share/code/nmethod.cpp#L770).
>
> The comment to `fix_relocation_after_move` is 17 year old.
> Is it still valid?
> We make a copy of nmethod. The copy is not in use. We know call sites might have invalid inline caches. Could we go through the calls sites and clear them?
>
> Requesting a vm operation for each nmethod will be very expensive.
That is very old comment which may be not true anymore. We don't do that when we replace C1 compiled code (tier2-3) with C2 compiled code.
-------------
PR Review Comment: https://git.openjdk.org/jdk/pull/23573#discussion_r1999723570
More information about the hotspot-compiler-dev
mailing list