[oe] klibc: armv4 bx strikes again

[Andrea Adami]

On Wed, Aug 27, 2014 at 10:39 AM, Andrea Adami <andrea.adami at gmail.com> wrote:
> On Tue, Aug 26, 2014 at 6:41 PM, Khem Raj <raj.khem at gmail.com> wrote:
>> On 14-08-26 11:36:34, Andrea Adami wrote:
>>> On Mon, Aug 25, 2014 at 10:08 AM, Andrea Adami <andrea.adami at gmail.com> wrote:
>>> > I've recently discovered that the static klibc binaries produced for
>>> > armv4 contain Illegal instructions.
>>> >
>>> > Eglibc armv4 binaries are ok.
>>> >
>>> > I verified it happens to kexec-tools-klibc and kexecboot-klibc:
>>> >
>>> > andrea at mizar:/oe/oe-core/build/tmp-eglibc/sysroots/x86_64-linux/usr/bin/arm-oe-linux-gnueabi$
>>> > ./arm-oe-linux-gnueabi-objdump -S ./kexecboot-armv4 | grep bx
>>> >    110dc:       012fff1e        bxeq    lr
>>> >    11154:       e12fff1e        bx      lr
>>> >    11160:       e12fff1e        bx      lr
>>> >    11170:       e12fff1e        bx      lr
>>> >    1119c:       e12fff1e        bx      lr
>>> >    11234:       e12fff1e        bx      lr
>>> >    11240:       e12fff1e        bx      lr
>>> >    11250:       e12fff1e        bx      lr
>>> >    11268:       e12fff1e        bx      lr
>>> >    11298:       e12fff1e        bx      lr
>>> >    112b4:       e12fff1e        bx      lr
>>> >    112f0:       e12fff1e        bx      lr
>>> >    115f4:       e12fff1e        bx      lr
>>> > andrea at mizar:/oe/oe-core/build/tmp-eglibc/sysroots/x86_64-linux/usr/bin/arm-oe-linux-gnueabi$
>>> >
>>> > The binaries end-up compiled with following flags:
>>> >
>>> > arm-oe-linux-gnueabi-klcc -march=armv4 -mthumb-interwork -Wall    -o
>>> > kexecboot util.o cfgparser.o devicescan.o evdevs.o fb.o gui.o menu.o
>>> > xpm.o rgb.o tui.o kexecboot.o fstype/fstype.o machine/zaurus.o
>>
>> what is content of arm-oe-linux-gnueabi-klcc ? is it also used at link
>> step ? the option
>>
>>> >
>>> > This is rather unexpected because klibc carries one specific patch to fix that:
>>> >
>>> > http://cgit.openembedded.org/meta-openembedded/tree/meta-initramfs/recipes-devtools/klibc/klibc-2.0.4/armv4-fix-v4bx.patch
>>> >
>>> > and in fact looking at the produced klibc.config I see indeed LDFLAGS=
>>> > --fix-v4bx and that goes in  *arm-oe-linux-gnueabi-klcc.
>>> >
>>> > There have been big toolchain changes recently so I suspect something
>>> > got broken in oe-core.
>>> > Then there is a patch touching armv4/thumb but it doesn't look
>>> > suspicious at first:
>>> >
>>> > http://cgit.openembedded.org/openembedded-core/commit/meta/conf/machine/include?id=1250d3e009363d20f15bbfaced622c5912a7fb93
>>> >
>>
>> for tests
>> can you use FIX_V4BX_ARMV4 in
>> meta-initramfs/recipes-devtools/klibc/klibc.inc
>> and see if it helps
>> --
>> _______________________________________________
>> Openembedded-devel mailing list
>> Openembedded-devel at lists.openembedded.org
>> http://lists.openembedded.org/mailman/listinfo/openembedded-devel
>
>
> Hi,
>
> I've added more output for klcc in klibc.bbclass
>
> -export CC_armv4_linux-gnueabi = "${TARGET_PREFIX}klcc -march=armv4
> -mthumb-interwork"
> +export CC_armv4_linux-gnueabi = "${TARGET_PREFIX}klcc -v -march=armv4
> -mthumb-interwork"
>
> The resulting log is  http://pastebin.com/rXQ0Vy1P
>
> I'm very surprised to see that  --with-gxx-include-dir, --with-sysroot
> and  --with-build-sysroot  are absolutely wrong, peeking at another
> machine sysroot.
> It mixes armv4 and armv5...how can that happen?
>
> I'll retry a build from scratch for the single armv4 arch asap.
>

Heh, unsurprisingly building from scratch for the armv4 target is flawless.

http://pastebin.com/MxWVPhd8

How klcc gets confused on multimachine builds is still unclear to me...

> Cheers
>
> Andrea