If I create threads using clone() with the CLONE_THREAD flag, can I assume that any thread local storage the glibc runtime uses, as well as anything I have declared with __thread, is initialized properly? In Windows, _beginthreadex provides for this. If it's not the case with clone(), how do I go about doing it?

[Edit:] __thread doesn't seem to be working, as I get the same address of a static __thread variable in a function called from both threads... So how do I get __thread working when/after using clone()?

[Edited by - Prune on January 5, 2011 10:13:56 PM]

If I'm not mistaken, to have __thread work properly with clone you must use the flag CLONE_SETTLS and setup a thread local storage descriptor to pass as the sixth argument to clone. As far as I am aware this will be kernel/architecture specific. That all said, I've never done this myself.

This is all done for you if you use pthreads.

The only reason I didn't use pthreads is since pthread_create() would be the only function I'm using (I use atomic operations and FUTEXes directly for synchronization). But it's no big deal and I'll do it with pthread_create().

The second issue I'm having is: I noticed that if I setpriority() to some value and at a later point to a lower nice value (higher priority), the second call gives an error and the nice value remains the higher value. It seems to only allow a monotonically nonincreasing dynamic priority... This happens even if I try to set the priority of the thread from the main thread which remains with its originial higher priority of 0.

So how do I give a user account a permission to lower the nice value?

Looks like RLIMIT_NICE is 0 for regular users....
Adding entries for the soft and hard limits for nice for a given user/group in /etc/security/limits.conf solved the problem. I don't understand why in Linux one needs to edit a file which only root can modify in order to allow this, when in Windows SetThreadPriority() works for regular users with any priority level.