Follow-up Comment #4, task #5487 (project hurd):
Just a follow-up about the patch:
- Barry's tree is in goober:/devel/bdefreese/hurd-pthread3/hurd
- Note the code in urfs/pager.c: it looks inside the rwlock in an ugly
way. It looks like it is just trying to wait for the read lock to be
available, but not actually lock.
- Condition implications have to be implemented. Barry has written some
code which isn't in the patch attached to this task, just in his tree
- Note that libpthread does already provide some compatibility stuff in
cthread-compat, which provides a few cthread symbols. That is needed
for glibc that itself uses a few cthread things, like mutexes.
- As Barry mentioned, there are some linking issues: when linking
statically, the "ascii version" of libpthread.a needs to be used,
i.e. not the binary but the linker script that adds undefined
references to a few compatibility cthread functions. Else the
initialization wouldn't be done and we'd get a __pthread_threads
assertion failure. The Makefiles should be fixed so that this is
correctly done.
- Also, remembemer when testing all of this to apply the TLS patch,
e.g. the one available in the debian package, else it'll just fail.
- spinlocks should rather be just statically initialized instead of
adding constructor functions.
- calls to pthread_spin_trylock should be fixed: Barry didn't notice
that the returned value wasn't the same as in cthreads.
_______________________________________________________
Reply to this item at:
<http://savannah.gnu.org/task/?5487>
_______________________________________________
Message posté via/par Savannah
http://savannah.gnu.org/
