On Wed, 18 Mar 2026 18:59:05 GMT, Zhengyu Gu <[email protected]> wrote:

>> Please review this change that fixes crashes by jmm_GetThreadInfo() call.
>> 
>> There are several issues:
>> - ThreadIdTable::lazy_initialize() has typical double-checked locking 
>> pattern without proper memory barrier, that may result in 
>> uninitialized/partial initialized table to be observed by other threads. 
>> Added release/acquire barrier to address this issue.
>> - Query ThreadIdTable can race add/remove thread operations. In short, the 
>> thread returned from the query may be freed.  Fortunately, 
>> jmm_GetThreadInfo() acquires stable thread list before query, so we only 
>> need to make sure that returned thread is in the list (checking 
>> thread->is_exiting() does not help due to the race)
>> - I moved thread Id insertion code from ThreadSMR to Threads, to be 
>> symmetric to thread Id removal code.
>> 
>> Tests:
>> - [x] Tier1 on Linux and MacOSX (fastdebug)
>>  (`tools/javac/annotations/typeAnnotations/IncorrectCastOffsetTest.java` 
>> failure seems unrelated, it also fails in master)
>
> Zhengyu Gu has updated the pull request incrementally with one additional 
> commit since the last revision:
> 
>   @kevinjwalls' comment

I think the problem is the moving of

 jlong java_tid = java_lang_Thread::thread_id(tobj);

to after the `MutexLocker` - we now have an unhandled oop and the mutex 
acquisition could safepoint.

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

PR Comment: https://git.openjdk.org/jdk/pull/30105#issuecomment-4093946329

Reply via email to