On Wed, 2011-01-12 at 00:37 +0000, Ben Hutchings wrote:
> On Wed, 2011-01-12 at 00:27 +0000, Richard Mortimer wrote:
> >
> > On 09/01/2011 03:46, David Miller wrote:
> > > From: Ben Hutchings<b...@decadent.org.uk>
> > > Date: Sun, 09 Jan 2011 03:00:40 +0000
> > >
> > >> On Sun, 2011-01-09 at 01:05 +0000, Richard Mortimer wrote:
> > >>> Package: linux-2.6
> > >>> Version: 2.6.37-1~experimental.1
> > >>> Severity: normal
> > >>>
> > >>> Boot of linux-image-2.6.37-trunk-sparc64 fails to find the disks and
> > >>> drops
> > >>> to the initramfs prompt. When I try to load the sym53c8xx driver it
> > >>> fails
> > >>> as follows
> > >>>
> > >>> (initramfs) modprobe sym53c8xx
> > >>> [ 122.470284] module scsi_mod: Unknown relocation: 36
> > >>> FATAL: Error inserting sym53c8xx
> > >>> (/lib/modules/2.6.37-trunk-sparc64/kernel/drive
> > >>> rs/scsi/sym53c8xx_2/sym53c8xx.ko): Invalid module format
> > >>> (initramfs)
> > >>
> > >> David, do you have any idea how this could happen?
> > >>
> > >>> A quick web search finds a similar issue
> > >>> http://kerneltrap.org/mailarchive/linux-kernel/2010/6/16/4583942
> > >>> but I have not looked into this any further yet.
> > >> [...]
> > >>
> > >> That was apparently a bug in the build scripts for a separate module.
> > >
> > > And like that case bad build flags are causing this problem too. It
> > > means that "-mcode-model=medlow" is not making it into the module
> > > build cflags somehow.
> > >
> > > This relocation can only occur for sparc64 code models other than
> > > "medlow".
> >
> > I did a test Debian build using my Sun Fire V120 running to double check
> > the build with the 2.6.27-1@experimental.1 sources. This fails in the
> > same way that the official build fails.
> >
> > Looking at the build output it seems to end up building setup_sparc but
> > I think it should be building setup_sparc64 in the rules.
> [...]
> > I will try a package build forcing DEB_HOST_ARCH to sparc64 and see if
> > that builds the right packages.
>
> No, this is correct behaviour. $DEB_HOST_ARCH is the Debian userland
> architecture (as used in the package metadata) and may differ from the
> kernel architecture. All the defined kernel flavours for sparc are
> 64-bit.
>
Thanks Ben. I've started looking at a few other potential causes...
Looking at the relocation symbols in scsi_mod.ko as reported by objdump
it looks to me that the R_SPARC_13 and R_SPARC_UA64 are the unsupported
relocation types that are stopping the driver getting loaded.
richm@shirehall:/usr/src/linux-2.6-2.6.37/debian/build/build_sparc_none_sparc64$
objdump -r /lib/modules/2.6.37-trunk-sparc64/kernel/drivers/scsi/scsi_mod.ko |
cut -d' ' -f 2 | sort | uniq -c
51
16 RECORDS
53 R_SPARC_13
129 R_SPARC_32
2352 R_SPARC_64
825 R_SPARC_HI22
869 R_SPARC_LO10
20 R_SPARC_UA64
1011 R_SPARC_WDISP30
From that I took one of the constituent source files (scsi_error.c) as
an example and looked at what the compiler was doing with it.
debian/build/build_sparc_non_sparc64/drivers/scsi/.scsi_error.o.cmd
reports that the compile command line is as follows
gcc-4.4 -Wp,-MD,drivers/scsi/.scsi_error.o.d -nostdinc
-isystem /usr/lib/gcc/sparc-linux-gnu/4.4.5/include
-I/usr/src/linux-2.6-2.6.37/debian/build/source_sparc_none/arch/sparc/include
-Iinclude -I/usr/src/linux-2.6-2.6.37/debian/build/source_sparc_none/include
-include include/generated/autoconf.h
-I/usr/src/linux-2.6-2.6.37/debian/build/source_sparc_none/drivers/scsi
-Idrivers/scsi -D__KERNEL__ -Wall -Wundef -Wstrict-prototypes -Wno-trigraphs
-fno-strict-aliasing -fno-common -Werror-implicit-function-declaration
-Wno-format-security -fno-delete-null-pointer-checks -Os -m64 -pipe -mno-fpu
-mcpu=ultrasparc -mcmodel=medlow -ffixed-g4 -ffixed-g5 -fcall-used-g7
-Wno-sign-compare -Wa,--undeclared-regs -mtune=ultrasparc3
-Wframe-larger-than=2048 -fno-stack-protector -fomit-frame-pointer
-Wdeclaration-after-statement -Wno-pointer-sign -fno-strict-overflow
-fconserve-stack -DMODULE -D"KBUILD_STR(s)=\#s"
-D"KBUILD_BASENAME=KBUILD_STR(scsi_error)"
-D"KBUILD_MODNAME=KBUILD_STR(scsi_mod)" -c -o drivers/scsi/.tmp_scsi_error.o
/usr/src/linux-2.6-2.6.37/debian/build/source_sparc_none/drivers/scsi/scsi_error.c
That seems to have -mcmodel=medlow so I'm wondering if this is a
compiler/compiler options issue.
If I remove -Os from the gcc-4.4 commandline above the R_SPARC_13 and
R_SPARC_UA64 entries disappear from .tmp_scsi_error.o output. I also
tried changing -Os to -O2 and the relocation symbols look broken for
that too.
Regards
Richard
--
To UNSUBSCRIBE, email to debian-kernel-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Archive: http://lists.debian.org/1294961641.22576.31.camel@duncow
