search

ofi-cray / libfabric-cray

Open Fabric Interfaces
http://ofiwg.github.io/libfabric/
Other
16 stars 9 forks source link

prov/gni: Address locking issue in initialization #1397

Closed jswaro closed 6 years ago

jswaro commented 6 years ago

If access to the ALPS LLI library is not serialized between threads, the channel can become corrupted with unexpected bytes.

Signed-off-by: James Swaro jswaro@cray.com

closes #1396

jswaro commented 6 years ago

might want to ask alps crew why the alps_app_lli_lock/unlock didn't do the job here.

The problem is based on our usage of alps LLI.

Here is what the fcntl man page says about threads and these locks which are used in libalpslli:

       The threads in a process share locks.  In other words, a
       multithreaded program can't use record locking to ensure that
       threads don't simultaneously access the same region of a file
hppritcha commented 6 years ago

check out distcheck: bot:lanl:retest