NetBSD Problem Report #43021

From www@NetBSD.org  Sat Mar 20 11:21:58 2010
Return-Path: <www@NetBSD.org>
Received: from mail.netbsd.org (mail.netbsd.org [204.152.190.11])
	by www.NetBSD.org (Postfix) with ESMTP id 3D4E863C49B
	for <gnats-bugs@gnats.NetBSD.org>; Sat, 20 Mar 2010 11:21:58 +0000 (UTC)
Message-Id: <20100320112157.DAB5F63B873@www.NetBSD.org>
Date: Sat, 20 Mar 2010 11:21:57 +0000 (UTC)
From: dennis.c.ferguson@gmail.com
Reply-To: dennis.c.ferguson@gmail.com
To: gnats-bugs@NetBSD.org
Subject: librump* build procedures don't get along with powerpc64 builds
X-Send-Pr-Version: www-1.0

>Number:         43021
>Category:       port-powerpc
>Synopsis:       librump* build procedures don't get along with powerpc64 builds
>Confidential:   no
>Severity:       non-critical
>Priority:       low
>Responsible:    mrg
>State:          closed
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Sat Mar 20 11:25:00 +0000 2010
>Closed-Date:    Sun Mar 28 04:30:39 +0000 2010
>Last-Modified:  Sun Mar 28 04:30:39 +0000 2010
>Originator:     Dennis Ferguson
>Release:        very recent 5.99.24
>Organization:
>Environment:
NetBSD acer.hk.akit-ferguson.com 5.99.24 NetBSD 5.99.24 (GENERIC) #0: Sun Mar 14 17:25:30 HKT 2010  dennis@acer.hk.akit-ferguson.com:/usr/obj/sys/arch/amd64/compile/GENERIC amd64

>Description:
Doing a macppc64 build results in the following when
it gets to doing dependencies in a librump directory
(anyone will do):

#    create  librump/__cmsg_alignbytes.d
CC=/usr/src/obj/tooldir.NetBSD-5.99.24-amd64/bin/powerpc64--netbsd-gcc /usr/src/obj/tooldir.NetBSD-5.99.24-amd64/bin/nbmkdep -f __cmsg_alignbytes.d --     -I/usr/src/lib/librump/../../sys/rump/include  -D_RUMPKERNEL -D_KERNEL -DMULTIPROCESSOR -D_MODULE -DMODULAR -DMAXUSERS=32 -DDIAGNOSTIC -DCOMPAT_50 -DDEBUGPRINT -I/usr/src/lib/librump -I. -I/usr/src/lib/librump/../../sys/rump/../../common/include -I/usr/src/lib/librump/../../sys/rump/../arch -I/usr/src/lib/librump/../../sys/rump/include -I/usr/src/lib/librump/../../sys/rump/librump/rumpkern/opt -nostdinc -I/usr/src/lib/librump/../../sys/rump/.. -D__NetBSD__ -I/usr/src/lib/librump/../../sys/rump/librump/rumpkern -I/usr/src/lib/librump/../../sys/rump/librump/rumpvfs -I/usr/src/lib/librump/../../sys/rump/librump/rumpnet -I/usr/src/lib/librump/../../sys/rump/librump/rumpdev -I/usr/src/lib/librump/../../sys/rump/../lib/libkern/arch/   -I/usr/src/lib/librump/../../sys/rump/../lib/libkern/../../../common/lib/libc/quad -I/usr/src/li
 b/librump/../../sys/rump/../lib/libkern/../../../common/lib/libc/string -I/usr/src/lib/librump/../../sys/rump/../lib/libkern/../../../common/lib/libc/arch/powerpc64/string -I/usr/src/lib/librump/../../sys/rump/../lib/libkern/../../../common/include   /usr/src/lib/librump/../../sys/rump/../lib/libkern/../../../common/lib/libc/net/__cmsg_alignbytes.c
In file included from /usr/src/lib/librump/../../sys/rump/../sys/types.h:45,
        from /usr/src/lib/librump/../../sys/rump/../lib/libkern/../../../common/lib/libc/net/__cmsg_alignbytes.c:44:
        ./machine/types.h:3:27: error: powerpc/types.h: No such file or directory
 In file included from /usr/src/lib/librump/../../sys/rump/../sys/types.h:47,
          from /usr/src/lib/librump/../../sys/rump/../lib/libkern/../../../common/lib/libc/net/__cmsg_alignbytes.c:44:
          ./machine/ansi.h:3:26: error: powerpc/ansi.h: No such file or directory

and so on.  It can't find include files like <powerpc/anything.h>.

Note that a macppc64 build ends up with the following settings:

    MACHINE=macppc
    MACHINE_ARCH=powerpc64

As far as I can tell, librump doesn't want to use the target include
files in destdir.x/usr/include but does want to include stuff from
the arch-specific parts of the kernel.  To do so it appears to try
to make two symbolic links in its obj/ directories:

    machine/ -> sys/arch/$MACHINE/include/
    $MACHINE_ARCH/ -> sys/arch/$MACHINE_ARCH/include/

For powerpc64 it doesn't bother making the second link since there
is no arch/powerpc64 directory.

It then #includes stuff like <machine/types.h> and <machine/ansi.h>
via the first link.  Unfortunately, <macppc/types.h> and
<macppc/ansi.h>, along with pretty much <macppc/anything.h>,
recursively include <powerpc/anything.h>, but a link named
"powerpc" doesn't exist and the above happens.
>How-To-Repeat:
Do a

    MACHINE=macppc64 ./build.sh distribution

in a clean source tree.
>Fix:
I don't know.  I get around it now by doing a regular macppc
build in the source tree before trying macppc64; apparently
the librump makefiles don't clean up the old links before
doing the new build.

>Release-Note:

>Audit-Trail:

State-Changed-From-To: open->analyzed
State-Changed-By: pooka@NetBSD.org
State-Changed-When: Sun, 21 Mar 2010 08:53:57 +0300
State-Changed-Why:
powerpc64 fails to set MACHINE_CPU.  try to build sys/modules.
this might fix it:

Index: bsd.own.mk
===================================================================
RCS file: /cvsroot/src/share/mk/bsd.own.mk,v
retrieving revision 1.623
diff -p -u -r1.623 bsd.own.mk
--- bsd.own.mk  13 Mar 2010 19:31:51 -0000      1.623
+++ bsd.own.mk  21 Mar 2010 05:52:30 -0000
@@ -14,7 +14,7 @@ MAKECONF?=    /etc/mk.conf
 #
 # CPU model, derived from MACHINE_ARCH
 #
-MACHINE_CPU=   ${MACHINE_ARCH:C/mipse[bl]/mips/:C/mips64e[bl]/mips/:C/sh3e[bl]/sh3/:S/m68000/m68k/:S/armeb/arm/}
+MACHINE_CPU=   ${MACHINE_ARCH:C/mipse[bl]/mips/:C/mips64e[bl]/mips/:C/sh3e[bl]/sh3/:S/m68000/m68k/:S/armeb/arm/:S/powerpc64/powerpc/}

 #
 # Subdirectory used below ${RELEASEDIR} when building a release


From: Dennis Ferguson <dennis.c.ferguson@gmail.com>
To: gnats-bugs@NetBSD.org
Cc: port-powerpc-maintainer@netbsd.org,
 netbsd-bugs@netbsd.org,
 gnats-admin@netbsd.org,
 pooka@NetBSD.org
Subject: Re: port-powerpc/43021 (librump* build procedures don't get along with powerpc64 builds)
Date: Sun, 21 Mar 2010 17:20:06 +0800

 Ah, thanks so much, I can see right away that
 will fix a couple of the build problems that I
 was currently hacking around.  I'll update the
 PRs I filed if that's the case.

 It introduced another problem though: I've now
 got a powerpc64--netbsd-lint which thinks longs are
 4 bytes even though the compiler thinks they are
 8 bytes.  Will investigate that.

From: matthew green <mrg@eterna.com.au>
To: Dennis Ferguson <dennis.c.ferguson@gmail.com>
Cc: port-powerpc-maintainer@netbsd.org, netbsd-bugs@netbsd.org,
    gnats-admin@netbsd.org, pooka@NetBSD.org, gnats-bugs@NetBSD.org
Subject: re: port-powerpc/43021 (librump* build procedures don't get along with powerpc64 builds)
Date: Mon, 22 Mar 2010 00:58:39 +1100


    It introduced another problem though: I've now
    got a powerpc64--netbsd-lint which thinks longs are
    4 bytes even though the compiler thinks they are
    8 bytes.  Will investigate that.


 usr.bin/xlint/arch/powerpc64/targparam.h seems to do the right thing.
 is it not being included, but the "powerpc" one is?


 .mrg.

From: Dennis Ferguson <dennis.c.ferguson@gmail.com>
To: matthew green <mrg@eterna.com.au>
Cc: port-powerpc-maintainer@netbsd.org,
 netbsd-bugs@netbsd.org,
 gnats-admin@netbsd.org,
 pooka@NetBSD.org,
 gnats-bugs@NetBSD.org
Subject: Re: port-powerpc/43021 (librump* build procedures don't get along with powerpc64 builds)
Date: Sun, 21 Mar 2010 22:20:28 +0800

 On 21 Mar 2010, at 21:58 , matthew green wrote:
 usr.bin/xlint/arch/powerpc64/targparam.h seems to do the right thing.
 > is it not being included, but the "powerpc" one is?

 I think the problem is actually in usr.bin/xlint/Makefile.inc, which
 now uses the MACHINE* variables in the following manner:

 # On the m68000, MACHINE_CPU is "m68k", but we have different
 # target parameters (namely, long double is a different size
 # on the m68000), so we special case the directory name here.
 .if ${MACHINE_ARCH} == "m68000"
 ARCHSUBDIR=     m68000
 .elif ${MACHINE_ARCH} == "mips64eb" || ${MACHINE_ARCH} == "mips64el"
 ARCHSUBDIR=     mips64
 .else
 ARCHSUBDIR=     ${MACHINE_CPU}
 .endif

 With MACHINE_ARCH=powerpc64/MACHINE_CPU=powerpc that ends up
 with ARCHSUBDIR=powerpc, with builds a 32-bit lint.  I would
 like to change this to something like

 .if ${MACHINE_ARCH} == "mips64eb" || ${MACHINE_ARCH} == "mips64el"
 ARCHSUBDIR=     mips64
 .elif exists(${.CURDIR}/../arch/${MACHINE_ARCH})
 ARCHSUBDIR=     ${MACHINE_ARCH}
 .else
 ARCHSUBDIR=     ${MACHINE_CPU}
 .endif

 which fixes my problem and removes the m68000 special case as a
 side effect.  I'm so far out of my depth with the mechanics of
 the build system, however, that I don't know if this change
 would be fixing something that is broken, or if the thing that
 is broken is setting MACHINE_CPU=powerpc.  Note that lint isn't
 the only new problem I've got; I'm also getting unresolved symbols
 from the libgcc4 libraries when loading c++ programs (i.e. groff)
 so I think I must have hit a similar problem in there somewhere too.

 Errors in kernel and powerpc code I can deal with, but the build
 system is way scarier.

From: christos@zoulas.com (Christos Zoulas)
To: Dennis Ferguson <dennis.c.ferguson@gmail.com>, 
	matthew green <mrg@eterna.com.au>
Cc: port-powerpc-maintainer@netbsd.org, netbsd-bugs@netbsd.org, 
	gnats-admin@netbsd.org, pooka@NetBSD.org, gnats-bugs@NetBSD.org
Subject: Re: port-powerpc/43021 (librump* build procedures don't get along with powerpc64 builds)
Date: Sun, 21 Mar 2010 10:22:10 -0400

 On Mar 21, 10:20pm, dennis.c.ferguson@gmail.com (Dennis Ferguson) wrote:
 -- Subject: Re: port-powerpc/43021 (librump* build procedures don't get along

 | On 21 Mar 2010, at 21:58 , matthew green wrote:
 | usr.bin/xlint/arch/powerpc64/targparam.h seems to do the right thing.
 | > is it not being included, but the "powerpc" one is?
 | 
 | I think the problem is actually in usr.bin/xlint/Makefile.inc, which
 | now uses the MACHINE* variables in the following manner:
 | 
 | # On the m68000, MACHINE_CPU is "m68k", but we have different
 | # target parameters (namely, long double is a different size
 | # on the m68000), so we special case the directory name here.
 | .if ${MACHINE_ARCH} == "m68000"
 | ARCHSUBDIR=     m68000
 | .elif ${MACHINE_ARCH} == "mips64eb" || ${MACHINE_ARCH} == "mips64el"
 | ARCHSUBDIR=     mips64
 | .else
 | ARCHSUBDIR=     ${MACHINE_CPU}
 | .endif
 | 
 | With MACHINE_ARCH=powerpc64/MACHINE_CPU=powerpc that ends up
 | with ARCHSUBDIR=powerpc, with builds a 32-bit lint.  I would
 | like to change this to something like
 | 
 | .if ${MACHINE_ARCH} == "mips64eb" || ${MACHINE_ARCH} == "mips64el"
 | ARCHSUBDIR=     mips64
 | .elif exists(${.CURDIR}/../arch/${MACHINE_ARCH})
 | ARCHSUBDIR=     ${MACHINE_ARCH}
 | .else
 | ARCHSUBDIR=     ${MACHINE_CPU}
 | .endif
 | 
 | which fixes my problem and removes the m68000 special case as a
 | side effect.  I'm so far out of my depth with the mechanics of
 | the build system, however, that I don't know if this change
 | would be fixing something that is broken, or if the thing that
 | is broken is setting MACHINE_CPU=powerpc.  Note that lint isn't
 | the only new problem I've got; I'm also getting unresolved symbols
 | from the libgcc4 libraries when loading c++ programs (i.e. groff)
 | so I think I must have hit a similar problem in there somewhere too.
 | 
 | Errors in kernel and powerpc code I can deal with, but the build
 | system is way scarier.

 Go for it.

 christos

Responsible-Changed-From-To: port-powerpc-maintainer->mrg
Responsible-Changed-By: mrg@NetBSD.org
Responsible-Changed-When: Sun, 21 Mar 2010 23:49:26 +0000
Responsible-Changed-Why:
OK, i'll test dennis' suggested change for this.  it should work
but i'll build a few platforms and make sure it looks right too.


From: matthew green <mrg@netbsd.org>
To: gnats-bugs@gnats.NetBSD.org
Cc: 
Subject: PR/43021 CVS commit: src/share/mk
Date: Mon, 22 Mar 2010 03:35:26 +0000

 Module Name:	src
 Committed By:	mrg
 Date:		Mon Mar 22 03:35:26 UTC 2010

 Modified Files:
 	src/share/mk: bsd.own.mk

 Log Message:
 set MACHINE_CPU to powerpc for MACHINE_ARCH of powerpc64.

 from pooka via PR#43021.


 To generate a diff of this commit:
 cvs rdiff -u -r1.623 -r1.624 src/share/mk/bsd.own.mk

 Please note that diffs are not public domain; they are subject to the
 copyright notices on the relevant files.

From: matthew green <mrg@netbsd.org>
To: gnats-bugs@gnats.NetBSD.org
Cc: 
Subject: PR/43021 CVS commit: src/usr.bin/xlint
Date: Mon, 22 Mar 2010 05:12:24 +0000

 Module Name:	src
 Committed By:	mrg
 Date:		Mon Mar 22 05:12:24 UTC 2010

 Modified Files:
 	src/usr.bin/xlint: Makefile.inc

 Log Message:
 for mips64*, use mips64 subdir, otherwise use $MACHINE_ARCH if exists,
 or $MACHINE_CPU.  idea from dennis.c.ferguson@gmail.com in PR#43021.


 To generate a diff of this commit:
 cvs rdiff -u -r1.11 -r1.12 src/usr.bin/xlint/Makefile.inc

 Please note that diffs are not public domain; they are subject to the
 copyright notices on the relevant files.

State-Changed-From-To: analyzed->closed
State-Changed-By: mrg@NetBSD.org
State-Changed-When: Sun, 28 Mar 2010 04:30:39 +0000
State-Changed-Why:
this one is fixed a while ago.


>Unformatted:

Home	PR Database Search