[OpenAFS-devel] cannot shutdown 1.3.81 on linux 2.6.12-rc[12]

[chas williams - CONTRACTOR]

In message <0dea4faeb58d085a019bd087878ce1d3@e18.physik.tu-muenchen.de>,Roland 
Kuhn writes:
>Please forgive my ignorance concerning the openafs source, but which 
>parts are synchronized by the AFS_GLOCK? I know that I can find the 
>details with grep, but what is the big picture?

the afs glock synchronizes quite a bit.

>How could I find out if it's AFS_GLOCK(), which is reducing 
>performance? Or, more generally, how can I best test AFS locking? I'm 
>running Ingo Molnar's realtime-preempt patch and lightweight usage has 
>not shown any problems yet (just the shutdown failure when running an 
>UP kernel).

i dont really have any good idea how to test this reliably.  building
two different source trees inside afs at the same time?  i know the
realtime/preempt stuff means that syscalls are no longer atomic.  afs
(without glock) relies on syscalls being atomic (as far as i know).