Xen users - May 2006 - can''t find a register in class `BREG'' while reloading `asm''

Hi List-

I''m brand new to xen but an old hand with Linux.

With a fresh, clean, Gentoo installation on this box, I''m attempting to
install xen using the gentoo-standard:

emerge xen

but the process is failing with an error that''s very similar to one
reported by another user (A Streetcar Named: see
http://lists.xensource.com/archives/html/xen-devel/2004-11/msg00593.html)
in the xen-devel list back in 2004 which is why I''m posting here and
not
to the gentoo lists.  I think they would answer by saying,
"...upstream...," so I''m starting upstream.

The error is:

<snip>
gcc -O2 -fomit-frame-pointer -DNDEBUG -m32 -march=i686 -Wall
-Wstrict-prototypes -Wdeclaration-after-statement -nostdinc -fno-builtin
-fno-common -fno-strict-aliasing -iwithprefix include -Werror
-Wno-pointer-arith -pipe
-I/var/tmp/portage/xen-3.0.2/work/xen-3.0.2/xen/include
-I/var/tmp/portage/xen-3.0.2/work/xen-3.0.2/xen/include/asm-x86/mach-generic
-I/var/tmp/portage/xen-3.0.2/work/xen-3.0.2/xen/include/asm-x86/mach-default
-msoft-float -g -D__XEN__ -c multicall.c -o multicall.o
memory.c: In function `guest_remove_page'':
/var/tmp/portage/xen-3.0.2/work/xen-3.0.2/xen/include/asm/mm.h:206:
error: can''t find a register in class `BREG'' while reloading
`asm''
make[2]: *** [memory.o] Error 1
make[2]: *** Waiting for unfinished jobs....
multicall.c: In function `do_multicall'':
multicall.c:43: error: can''t find a register in class `BREG''
while
reloading `asm''
make[2]: *** [multicall.o] Error 1
grant_table.c: In function `do_grant_table_op'':
/var/tmp/portage/xen-3.0.2/work/xen-3.0.2/xen/include/asm/mm.h:206:
error: can''t find a register in class `BREG'' while reloading
`asm''
/var/tmp/portage/xen-3.0.2/work/xen-3.0.2/xen/include/asm/mm.h:206:
error: can''t find a register in class `BREG'' while reloading
`asm''
/var/tmp/portage/xen-3.0.2/work/xen-3.0.2/xen/include/asm/mm.h:206:
error: can''t find a register in class `BREG'' while reloading
`asm''
make[2]: *** [grant_table.o] Error 1
make[2]: Leaving directory
`/var/tmp/portage/xen-3.0.2/work/xen-3.0.2/xen/common''
make[1]: ***
[/var/tmp/portage/xen-3.0.2/work/xen-3.0.2/xen/common/built_in.o] Error 2
make[1]: Leaving directory
`/var/tmp/portage/xen-3.0.2/work/xen-3.0.2/xen/arch/x86''
make: *** [/var/tmp/portage/xen-3.0.2/work/xen-3.0.2/xen/xen] Error 2
make: Leaving directory `/var/tmp/portage/xen-3.0.2/work/xen-3.0.2/xen''

The gentoo USE flags I''ve chosen are as follows:

[ebuild  N    ] app-emulation/xen-3.0.2  -custom-cflags -debug -pae 0 kB

I am using a somewhat non-standard gentoo profile:

lrwxrwxrwx 1 root root 51 May  4 15:24 /etc/make.profile ->
../usr/portage/profiles/selinux/2005.1/x86/hardened

And I do have SELinux successfully installed although I''m running in
permissive mode so I don''t think SELinux has anything to do with this.
It may be the case, though, that the hardened part of the profile I''ve
chosen is a part of the problem.

The other thread that cropped up with similar error messages was also a
Gentoo hardened box and there was no resolution posted although the
fellow''s initial problem (failure to compile) was apparently resolved
by
the elimination of some (all?) of the hardened flags during compilation:

"-nopie fixed the  originally reported error, but I subsequently got
other errors that said something like "undefined reference to
`__guard''" and "undefined reference to
`__stack_smash_handler''".
Adding -fno-stack-protector seems to have solved these as well, and
I''m now compiling the kernels for the crusoe."

My gcc version is:

gcc (GCC) 3.4.6 (Gentoo Hardened 3.4.6-r1, ssp-3.4.5-1.0, pie-8.7.9)

I started off with 3.4.5 and similar or identical parameters for ssp and
pie.  The results in that case were the same so I''ve tried two
different
 gcc versions (though both are hardened with pie/ssp) to no avail.

I''m not using Pax or Grsecurity or RSBAC and like I said, SELinux is
running in permissive mode right now.

In the old thread, Keir Fraser wrote:

"I''ve checked in a patch to the unstable tree that will allow to
build
with PIE/SSP-enabled GCC."

So my hope is to be able to get xen running with these hardened flags,
but I''m wondering if something else needs to change with xen first.

I haven''t yet tried compiling under the guidance that Keir gave back in
2004 (and that worked for Streetcar), but I can if it would help.  I
suspect that like the other guy, this would get xen to compile
successfully, but I''d really like to keep the hardened nature intact so
I''m asking about that first.

Thanks in advance for any ideas.

-Kevin

PS. If I should post to xen-devel, someone please tell me and I''ll do
so.

_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

Kevin wrote:
> gcc -O2 -fomit-frame-pointer -DNDEBUG -m32 -march=i686 -Wall
> -Wstrict-prototypes -Wdeclaration-after-statement -nostdinc -fno-builtin
> -fno-common -fno-strict-aliasing -iwithprefix include -Werror
> -Wno-pointer-arith -pipe
> -I/var/tmp/portage/xen-3.0.2/work/xen-3.0.2/xen/include
>
-I/var/tmp/portage/xen-3.0.2/work/xen-3.0.2/xen/include/asm-x86/mach-generic
>
-I/var/tmp/portage/xen-3.0.2/work/xen-3.0.2/xen/include/asm-x86/mach-default
> -msoft-float -g -D__XEN__ -c multicall.c -o multicall.o
> memory.c: In function `guest_remove_page'':
> /var/tmp/portage/xen-3.0.2/work/xen-3.0.2/xen/include/asm/mm.h:206:
> error: can''t find a register in class `BREG'' while
reloading `asm''

see http://bugs.gentoo.org/show_bug.cgi?id=130167

HTH

_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

Tuan Van wrote:
> Kevin wrote:
>> gcc -O2 -fomit-frame-pointer -DNDEBUG -m32 -march=i686 -Wall
>> -Wstrict-prototypes -Wdeclaration-after-statement -nostdinc
-fno-builtin
>> -fno-common -fno-strict-aliasing -iwithprefix include -Werror
>> -Wno-pointer-arith -pipe
>> -I/var/tmp/portage/xen-3.0.2/work/xen-3.0.2/xen/include
>>
-I/var/tmp/portage/xen-3.0.2/work/xen-3.0.2/xen/include/asm-x86/mach-generic
>>
>>
-I/var/tmp/portage/xen-3.0.2/work/xen-3.0.2/xen/include/asm-x86/mach-default
>>
>> -msoft-float -g -D__XEN__ -c multicall.c -o multicall.o
>> memory.c: In function `guest_remove_page'':
>> /var/tmp/portage/xen-3.0.2/work/xen-3.0.2/xen/include/asm/mm.h:206:
>> error: can''t find a register in class `BREG'' while
reloading `asm''
> 
> 
> see http://bugs.gentoo.org/show_bug.cgi?id=130167
> 
Thanks for the pointer, Tuan.  I wasn''t aware of it and I should have
checked the gentoo & xen bugzillas before posting here.  My apologies
for not doing so.  I have now checked both gentoo and xen bugzillas and
I don''t see anything on BREG or register or pie or ssp or hardened in
the xen bugzilla.

The patch mentioned in the gentoo bug report is, however, merely an
implementation of the advice that Keir Fraser gave to A Streetcar back
in 11/2004, and it does indeed allow me to compile xen-3.0.2 (which I
suspected would be true, but it doesn''t get me what I''d really
like
which is xen code compiled with PIE/SSP for enhanced security).

Keir Fraser wrote in 11/2004:

"I''ve checked in a patch to the unstable tree that will allow to
build
with PIE/SSP-enabled GCC."

May I ask: what''s happened with this?  Was a patch applied?  If so,
does
something else need to be patched?  Or was the patch subsequently
removed?  Or do the xen-devs prefer against patching xen this way?

Are there negative implications to building xen with PIE/SSP-enabled
GCC? (or for that matter, not doing so?).  Although I''m no guru on
hardened gentoo or PIE/SSP, I had the overall impression that they were
considered by many (all?) to be a good thing for enhancing security.  Is
that not the case with xen?

TIA for any replies.

-Kevin

_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users