On Tue Jun 22 2021, Adhemerval Zanella wrote:
> Besides what Florian has explained about __ASSUME_* macros, another issue we
> have static initialization for pthread_mutex.  So the syscall probe only
> works for dynamic initialization (where caller issue a pthread_mutex_init).
>
> I view this as an inconsistent behavior and I don't have a straightforward
> solution that won't result in a performance penalization for common cases
> (it would require to probe for FUTEX_LOCK_PI2 *and* FUTEX_LOCK_PI).
>
> Instead I think we should move the possible error on the slow path and let
> the kernel advertise it any missing support.

OK, good. That makes the timedlock implementation straight forward. I
implemented it and started testing now with different Linux kernels. A
new version of patch set should be ready next week or so.

Thanks,
Kurt