floating context bug still exists in Release v2

I feel too bad ,the bug which mmetz_nv mentioned in cuctxpopcurrent bug still existed in the release version Cuda2.
If there is no user’s cuda routine but cuda apis between cuCtxPushCurrent and cuctxpopcurrent in a host thread, everything seems work fine. But if I added some user cuda c routines in them , the coming cuCtxPushCurrent in this host thread will return CUDA_ERROR_INVALID_VALUE.
I wanna know if someone else here met the same problem,

Please provide a test app which reproduces this problem.

Are your synchronizing the ownership of the context with a mutex or critical section ? (only one thread can own the context at any time)

Eh,I found it’s a stupid bug of mine . I missed one cuPopCtx in one thread so cause this error.