Mercurial > hg > release > icedtea7-forest-2.4 > hotspot
changeset 4777:1478a623482d
8016074: NMT: assertion failed: assert(thread->thread_state() == from) failed: coming from wrong thread state
Summary: Uses os::NakedYield() on Solaris instead of os::yield_all()
Reviewed-by: acorn, coleenp, hseigel
| author | zgu |
|---|---|
| date | Thu, 04 Jul 2013 06:24:08 -0400 |
| parents | dd7d57bcd749 |
| children | 25a649574b3b 0d7106aa7e08 f4d7fb7ba02e |
| files | src/share/vm/services/memTracker.hpp |
| diffstat | 1 files changed, 14 insertions(+), 0 deletions(-) [+] |
line wrap: on
line diff
--- a/src/share/vm/services/memTracker.hpp Tue Jul 09 10:03:28 2013 +0200 +++ b/src/share/vm/services/memTracker.hpp Thu Jul 04 06:24:08 2013 -0400 @@ -402,7 +402,21 @@ static void check_NMT_load(Thread* thr) { assert(thr != NULL, "Sanity check"); if (_slowdown_calling_thread && thr != _worker_thread) { +#ifdef _WINDOWS + // On Windows, os::NakedYield() does not work as well + // as os::yield_all() os::yield_all(); +#else + // On Solaris, os::yield_all() depends on os::sleep() + // which requires JavaTherad in _thread_in_vm state. + // Transits thread to _thread_in_vm state can be dangerous + // if caller holds lock, as it may deadlock with Threads_lock. + // So use NaKedYield instead. + // + // Linux and BSD, NakedYield() and yield_all() implementations + // are the same. + os::NakedYield(); +#endif } }