RFR: 8322962: Upcall stub might go undetected when freezing frames

[Jorn Vernee]

On Thu, 22 Feb 2024 14:28:18 GMT, Patricio Chilano Mateo <pchilanomate at openjdk.org> wrote:

> Please review this simple fix. There is a missing push_cont_fastpath()/pop_cont_fastpath() around the call to Java in the upcall stub. Because of this, if from that upcall a virtual thread tries to yield, and all the other methods in the stack are compiled, we will incorrectly take the freeze fast path and freeze everything in the stack including the native frames. By forcing the slow path instead we will identify the upcall stub and pin the vthread.
> 
> The issue can be easily reproduced by running the modified version of test GetStackTraceALotWhenPinned.java included in this patch. I also tested the fix by running tiers1-3 in mach5. 
> 
> Thanks,
> Patricio

x64 and aarch64 code look good to me. push/pop_cont_fastpath don't modify the register state or the stack ptr. (it looks like 32-bit x86 will use the fallback linker for now, so I'm not worried about that).

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

Marked as reviewed by jvernee (Reviewer).

PR Review: https://git.openjdk.org/jdk/pull/17964#pullrequestreview-1896090108