On Wed, Jun 22, 2011 at 02:14:55PM -0700, Jesse Gross wrote:
> On Wed, Jun 22, 2011 at 10:35 AM, Ben Pfaff <[email protected]> wrote:
> > I've been thinking a bit about the Linux kernel module build. ??It
> > doesn't look all that much like how other Linux kernel modules build
> > themselves. ??The "configure" step is really odd in that context.
> >
> > I think that we could get rid of the "configure" step. ??All of the
> > Linux-related configure pieces are really simple. ??I think that we
> > could embed them all into GNU make code within a Makefile. ??And then
> > you could just run "make", possibly pointing an environment variable
> > to the right directory. ??This would be much more like how other
> > external kernel modules typically build. ??I think that proper
> > dependencies would even be possible, so that it wouldn't be necessary
> > to rerun a configuration step when KSRC changed or when some other
> > kernel got checked out in the directory that KSRC pointed to.
>
> What about all of the checks for backported functions? It seems like
> that's a fairly large chunk of code that's part of the configure step.
> Or do you mean something else?
I think that those checks could be done right in the Makefile with
proper dependencies. For example,
OVS_GREP_IFELSE([$KSRC/include/linux/skbuff.h], [raw],
[OVS_DEFINE([HAVE_MAC_RAW])])
could become something like:
CHECKS =
...other checks...
CHECKS += checks/have_mac_raw.inc
checks/have_mac_raw.inc: $(KSRC)/include/linux/skbuff.h
if grep raw $? >/dev/null; then echo '#define HAVE_MAC_RAW 1'; fi > $@
kcompat.h: $(CHECKS)
cat $(CHECKS) > $@
Use of GNU make macros and syntax would probably make it more maintainable.
> > An orthogonal change would be to move "datapath/linux-2.6" to just
> > "linux".
>
> This definitely makes sense.
OK.
_______________________________________________
dev mailing list
[email protected]
http://openvswitch.org/mailman/listinfo/dev
