On Tue, 10 Jan 2023 06:57:01 GMT, David Holmes <dhol...@openjdk.org> wrote:
> > Forcing 2 reviewers to ensure @dholmes-ora can chime in before moving > > forward. > > Well I won't be able to Review as not familiar enough with the code, so > you'll need a second reviewer anyway. I don't hate this to the point of > outright rejecting it but I do have general concerns about whether we should > be directly supporting such tools in our codebase, and if we should whether > these are the right tools. So I've asked other hotspot folk to chime in. I dislike this too. I wondered whether we could hide it behind an "interface for asan-like tools", where we have a `os::poison_memory(range)` and `os::unpoison_memory(range)` function. Those functions could in turn call whatever tool is configured. At least then we don't pollute the code base with tool specifics. Granted, it sounds a bit fig leafy as long as there is only Asan. But if we wanted, we could implement a primitive poisoner tool in hotspot by mprotecting if the range spans whole pages. ------------- PR: https://git.openjdk.org/jdk/pull/11702