arla on dual amd 64-bit

[Rasmus Kaj]

>>>>> "GW" == Garrett Wollman <wollman at csail.mit.edu> writes:

 >> I don't think it matters which kernel the user uses, as long as there
 >> is a bsd.kmod.mk where make can find it.

 GW> It very much does matter, if the kernel is compiled with any of a
 GW> number of debugging options (as is likely for anyone using Arla on
 GW> FreeBSD 6 or 7 right now).  It is absolutely essential that nnpfs.ko
 GW> and the kernel be compiled with the same opt_global.h.

Ah.  That sounds like a problem we'll have to fix.

 GW> (<bsd.kmod.mk> will fake it if one can't be found, but this hack
 GW> is only correct for GENERIC kernels.)

Ok.  Regardless of the faking, how good a job does it do trying to
find one?  I can see it finding out where the kernel sources should be
(setting SYSDIR), and I can see it including opt_global.h from
KERNBUILDDIR if set, but KERNBUILDDIR doesn't seem to get set.

And, when I compile nnpfs, there is no visible mention of opt_global.h
or my kernel name.  It does seem to work, but you mean it might fail
if I had debugging options in my kernel?

If I have KERNBUILDDIR in my environment when building nnpfs,
opt_global.h gets an -include parameter.  So there should be a way for
nnpfs/freebsd/Makefile (the calling, autotools, Makefile) to set
KERNBUILDDIR when running bsd make on FreeBSD-Makefile.  I seem to be
able to get the directory from `uname -v | sed -e 's/.*://'`, but that
sounds a bit like a hack ...

-- 
Rasmus Kaj --+-- rasmus at kaj.se --+-- http://www.stacken.kth.se/~kaj/
   Just say NO to lssy cmprssn