From: Fanael L. <fa...@gm...> - 2013-12-06 16:54:42
|
2013/12/5 Fanael Linithien <fa...@gm...>: > runs BEFORE enter_global_cs Er, before global_spin_ctor, of course. 2013/12/5 Kai Tietz <kti...@go...>: > Nice catch. This describes at least the runtime-failure we got reported. > Did you tested regressions for this patch? I have. All tests from tests_pthread pass. 2013/12/5 dw <lim...@ya...>: > 1) Shouldn't global_lock be __LONG32? Probably not. InterlockedExchange takes volatile LONG*. > 2) Would it make sense for the exchange on global_lock to be done as a > single operation (ie InterlockedCompareExchange)? It's not necessary. InterlockedExchange already provides enough synchronization. |