RFR: 8329103: assert(!thread->in_asgct()) failed during multi-mode profiling

[David Holmes]

On Wed, 27 Mar 2024 01:02:41 GMT, Andrei Pangin <apangin at openjdk.org> wrote:

> This fix makes `AsyncGetCallTrace` reentrant and async-signal-safe.
> Reentrancy is required in the cases when two or more profiling engines are running at the same time, e.g., when CPU and Wall clock profilers work together and therefore one profiler may interrupt another in the middle of getting a stack trace.
> 
> Tested with async-profiler:
> 
> java -agentpath:/path/to/libasyncProfiler.so=start,event=cpu,interval=1ms,wall=1ms,file=profile.jfr

I have my doubts as to whether AGCT is actually re-entrant in a general sense, but I can see that the `ThreadInAsgct` RAII object introduced a reentrancy constraint that did not exist prior, and so removing it should not make AGCT any less safe and should allow previous reentrancy cases to continue to work as before.

-------------

Marked as reviewed by dholmes (Reviewer).

PR Review: https://git.openjdk.org/jdk/pull/18504#pullrequestreview-1966173558