> > - The shared data structure should be modified with atomic > accesses only. (Or use a criticalsection if possible.) >
According the this website, Windows updates the times too frequently to use a critical section: http://www.dcl.hpi.uni-potsdam.de/research/WRK/2007/08/getting-os-information-the-kuser_shared_data-structure/ The values are written in the order High2, Low, High1. Applications read them in the reverse order; as long as the two High values are the same, the composite value is considered correct. This is a fairly common algorithm for detecting clock tick without atomic operations or critical sections. I could certainly use atomic operations for each of the 32-bit word writes. But since Windows apps expect the word write order, I don't think it's necessary or advisable to do 64-bit atomic ops. > > However, I do not know if we can have additional threads at all in a > Win32 process without confusing the win32 processes, or if this needs > to be solved differently (APC?). > A previous version of my patch used NtTimer/APC: http://bugs.winehq.org/attachment.cgi?id=39308 The APC code is commented out but still present in that version of the patch. It did not work, by which I mean SW:TOR did not successfully login. It was clearly the best solution, so I'd be happy to return to it if we can make the callback times more precise. cheers, Joey
