[Lustre-devel] [wc-discuss] Important changes to libcfs primitives usage.
Ken Hornstein
kenh at cmf.nrl.navy.mil
Mon Dec 17 09:21:44 PST 2012
>Current the libcfs cleanup is mainly done by replacing kinds of
>libcfs calls with direct kernel primitives, we can implement the
>portable layer on other OS with the same name as Linux. I think
>the real concern is that on other OS there is possible some kernel
>primitives have the same name a Linux kernel but with different
>behavior/parameters, that will cause either compiling broken or running
>crashing etc. I don't have the full list of the above intersecting
>case, but it should be very rare right? We can build a separate patch(
>which should be small) for those intersecting cases for other OS, and I
>guess that patch need to be maintained separately.
Like I said, I'm fine with that; at least in the MacOS X case I don't
recall any symbol names which overlapped.
>And lustre CLIO is structured that all OS-dependent glue is supposed
>to be handled in the topmost layer (VVP for linux client). For Mac or
>Windows's native client, it is possibly just implement a new thin top
>layer if interests really raised on those platform? This could reduce
>the possibility of the above intersecting cases.
Well, it's only a "thin" layer if your VM system looks exactly like Linux;
otherwise it ain't so thin. Yeah, I did some of that, but it wasn't
easy.
--Ken
More information about the lustre-devel
mailing list