Leif Johansson | 1 Aug 11:18 2005
Picon
Picon

AFS hackathon 3 reminder


AFSig.se (that is to say SU and KTH) will host a Hackathon event
next week. Starting at 10:00 8/8 2005 at KTH Campus in Stockholm.
More details here: http://www.afsig.se.

If you plan to turn up please send a note to register <at> afsig.se.

	Best Regards
	Leif, Love, Harald, Jimmy, Johan and JJ
Rasmus Kaj | 1 Aug 13:39 2005
X-Face
Picon
Picon

Re: arla on dual amd 64-bit

>>>>> "MO" == Mark Ogden <ogden <at> eng.utah.edu> writes:

 MO> trying to install Arla on a new dual 64-bit AMD Opteron
 MO> machine. This is my config options:

Hi!  I've been using arla on a (single) amd64, on FreeBSD 5.4, with
some help from Haba.

Unfortunatley, my home network has been shaky the last week and seems
broken right now, so I can't give you any exact diff, but we changed
nnpfs/bsd/Makefile (I've yet to clean up the changes and make them fit
for general consumption in Makefile.in)

One change was in CFLAGS (hopefully I can give you the exact details
later today / tonight, but I'm writing now so you get at least some
response).

The ofther was in the nnpfs.ko target, changeing 

	  $(LD) -Bshareable -o $ <at>  $${pre} $(MODULE_OBJS) $${post}; \

to

	  $(LD) -r -o $ <at>  $${pre} $(MODULE_OBJS) $${post}; \

--

-- 
Rasmus Kaj --+-- rasmus <at> kaj.se --+-- http://www.stacken.kth.se/~kaj/
If God had intended Man to Smoke, He would have set him on Fire

(Continue reading)

Rasmus Kaj | 2 Aug 12:02 2005
X-Face
Picon
Picon

Re: arla on dual amd 64-bit

>>>>> "HB" == Harald Barth <haba <at> pdc.kth.se> writes:

 HB> It is not _that_ easy because configure must first detect that is is
 HB> actually a system which has the feature "kernel module-build
 HB> infrastructure" and then build the right Makefile. 

Approximatley how many systems would that be, of the BSDs?  FreeBSD
5.4 is one of them, but is all of FreeBSD 5.x ok?  I assume it isn't
broken in 6 or 7 (yes, HEAD is now 7). How about FreeBSD 4.x?  I think
we can safely ignore FreeBSD 3.x?

How about NetBSD?  And OpenBSD?, or is the OpenBSD port so specific
that we can/should ignore it?

I'm not suggesting that the configure test should just look at the
version numbers, I just want to know if the new system would be used
more often than not.

--

-- 
Rasmus Kaj --+-- rasmus <at> kaj.se --+-- http://www.stacken.kth.se/~kaj/
Proffs byggde Titanic - amatörer byggde arken

Rasmus Kaj | 1 Aug 19:59 2005
X-Face
Picon
Picon

Re: arla on dual amd 64-bit

>>>>> "RK" == Rasmus Kaj <kaj <at> kth.se> writes:
>>>>> "MO" == Mark Ogden <ogden <at> eng.utah.edu> writes:

 MO> trying to install Arla on a new dual 64-bit AMD Opteron
 MO> machine. This is my config options:

 RK> Hi!  I've been using arla on a (single) amd64, on FreeBSD 5.4, with
 RK> some help from Haba. [...]

Hi again, heres some details to my previous post.

My patch is on http://www.stacken.kth.se/~kaj/tmp/arla-amd64.patch .
Note that it is _not_ yet fit for general consumption (will break
stuff on systems other than amd64 FreeBSD 5.4) yet.

With this patch, afs seems to work for me, allthough i get loads of:

Aug  1 19:37:23 nestor kernel: kldload: unexpected relocation type 10

... when running startarla.  I don't now what that really means, so I
don't know how important it is ...

--

-- 
Rasmus Kaj --+-- rasmus <at> kaj.se --+-- http://www.stacken.kth.se/~kaj/
What garlic is to food, insanity is to art

Garrett Wollman | 2 Aug 18:56 2005
Picon
Picon

Re: arla on dual amd 64-bit

<<On Tue, 02 Aug 2005 10:47:25 +0200 (MEST), Harald Barth <haba <at> pdc.kth.se> said:

>  #include <dirent.h>
>  #if DIRENT_AND_SYS_DIR_H
> +#ifndef __FreeBSD__		/* configure test ok, but compiler warns */
>  #include <sys/dir.h>
> +#endif
>  #endif
>  #elif defined(HAVE_SYS_DIR_H)

> which should be fixed by fixing the value of DIRENT_AND_SYS_DIR_H (or
> breaking it up into two flags) instead of __FreeBSD__.

The configure test that sets that flag means "it's OK to include
dirent.h and sys/dir.h together", which is true (it only generates a
warning, not an error).  It doesn't test where there is anything
useful in sys/dir.h (there is not in any *BSD system going all the way
back, which is what the warning this conditional suppresses actually
says).

> Another alternative is to do a FreeBSD specific port (more or less
> like the OpenBSD folks do, but they are currently at 0.35.7, so is
> this really the way to go?). I think all solutions which try to
> support platform dependent code stink "pick your misery" long way.

There is a port.  It has a number of patches for 6.x/7.x which I have
not yet had a chance to try out because none of my machines are yet
running it.  See <http://cvsweb.freebsd.org/ports/net/arla/files/>.

-GAWollman
(Continue reading)

Garrett Wollman | 2 Aug 17:54 2005
Picon
Picon

Re: arla on dual amd 64-bit

<<On 02 Aug 2005 12:02:33 +0200, Rasmus Kaj <kaj <at> kth.se> said:

> Approximatley how many systems would that be, of the BSDs?  FreeBSD
> 5.4 is one of them, but is all of FreeBSD 5.x ok?  I assume it isn't
> broken in 6 or 7 (yes, HEAD is now 7). How about FreeBSD 4.x?  I think
> we can safely ignore FreeBSD 3.x?

All FreeBSD systems on which Arla is likely to work have the
module-build infrastructure, although it works somewhat differently in
4.x (now approaching EOL) than in 5+.

There needs to be some configure infrastructure to set KERNBUILDDIR to
an appropriate place, if the user is not using the GENERIC kernel.
("--with-freebsd-kernel-build-dir=/foo"?)

-GAWollman

Love | 8 Aug 10:51 2005
Picon
Picon

Re: nnpfs_lock on FreeBSD 5.4 with Arla-0.40


Marcus Crestani <crestani <at> informatik.uni-tuebingen.de> writes:

> Hi,
>
> trying to build my current XEmacs workspace on FreeBSD always leads to
> nnpfs_lock, probably caused by arla.

Can you list all process states (ps axlwww) and give backtraces force those
that are in nnpfs or nnpfs_lock (using gdb) ?

Love

Mike Hammill | 9 Aug 18:58 2005
Picon
Picon

Can't compile 0.40 kernel modules on linux 2.6.12 [resolved]

I believe I had the same problem as Friedrich Delgado Friedrichs,
26 July 2005, 
<http://www.stacken.kth.se/lists/arla-drinkers/2005-07/msg00008.html>
but with a different kernel, linux-2.6.12-gentoo-r6.

The basic error, when using no special options to configure, is:

/var/tmp/arlaobj/nnpfs/linux/kbuild-work/nnpfs_syscalls.c:41:18: kafs.h: 
No such file or directory

On the other hand, the build goes fine with configure --disable-nnpfs. 
When the patches given by Alexander Bostrom in his reply to Friedrich 
are applied, I get a working Arla.

Thanks!  Just wanted to tell you it works.
/Mike

Harald Barth | 10 Aug 11:27 2005
Picon
Picon

Re: Can't compile 0.40 kernel modules on linux 2.6.12 [resolved]


http://www.pdc.kth.se/~haba/gentoo-stuff/ You will find some portage
stuff there. I have not had the time to do a arla-0.40 ebuild file,
but there is a 0.39 you can use for inspiration.

Harald.

Rasmus Kaj | 10 Aug 19:29 2005
X-Face
Picon
Picon

Re: arla on dual amd 64-bit

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

With some help from Love I've arranged for the auto-tools make stuff
to call a freebsd makefile that includes <bsd.kmod.mk>, similar to the
one you submitted.  It's comitted to current in the arla repo.

 GW> There needs to be some configure infrastructure to set KERNBUILDDIR to
 GW> an appropriate place, if the user is not using the GENERIC kernel.
 GW> ("--with-freebsd-kernel-build-dir=/foo"?)

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.  If the make include files
are located in some strange place (other than /usr/share/mk), I don't
know if there's a way to find them.  But then, that would probably be
considered broken for a FreeBSD installation(?).

And it's FreeBSD only, since NetBSD has a <bsd.somethingelse.mk>
instead of kmod (I don't know about Dragonfly, but the "generic BSD"
way of building should continue to work there if it worked before
these changes).  On the other hand, the NetBSD configure code had a
way to get some flags out of that include file and use them while
building the kernel module.

--

-- 
Rasmus Kaj --+-- rasmus <at> kaj.se --+-- http://www.stacken.kth.se/~kaj/
Which is worse: ignorance or apathy? Who knows? Who cares?


Gmane