On Sat, 2 May 2026 11:02:48 GMT, Yasumasa Suenaga <[email protected]> wrote:
>> Kevin Walls has updated the pull request incrementally with four additional >> commits since the last revision: >> >> - Windows _lseeki64. >> - Windows _lseeki64. No timeout waiting for revivalhelper. >> - Limit tests for ZGC >> - vmError show fatal error for revived process. elffile and segment >> clarification in region checks. > > Great PR! I have some comments. @YaSuenag Yasumasa, somewhere here raised a point about MAP_FIXED_NOREPLACE which is an option: As the revivalhelper is kind of a throwaway launcher that gets libjvm and the memory loaded, and calls into the JVM, it needs to care that important parts of its own memory are not overwritten, but does not need to care about all of it. Using the current guide ranges for what we think is critical not to overwrite, gives us more freedom to overwrite memory in revivalhelper process. e.g. if some memory is not important, let it be overrwitten. If NOREPLACE is used it might see more conflicts, and more retries to get the helper launched (there are usually none...). I could try (at some point) to audit the efficienty of this to verify if the balance can be improved. ------------- PR Comment: https://git.openjdk.org/jdk/pull/31011#issuecomment-6690764592
