[OE-core] [PATCH 1/1] cross-canadian/libgcc: fix aarch64's multilib SDK

[Robert Yang]

On 11/10/2015 12:27 AM, Burton, Ross wrote:
>
> On 4 November 2015 at 13:58, Robert Yang <liezhi.yang at windriver.com
> <mailto:liezhi.yang at windriver.com>> wrote:
>
>     The arm toolchain has a "-gnueabi" suffix, but aarch64 doesn't,
>     this makes multilib sdk doesn't work, for example:
>
>     MACHINE = qemuarm64
>     require conf/multilib.conf
>     MULTILIBS = "multilib:lib32"
>     DEFAULTTUNE_virtclass-multilib-lib32 = "armv7at-neon"
>
>     $ bitbake core-image-minimal -cpopulate_sdk
>
>     Then extract SDK, the
>     environment-setup-armv7a-vfp-neon-pokymllib32-linux-gnueabi
>     doesn't work since:
>     * The CC is arm-pokymllib32-linux-gnueabi-gcc
>        which doesn't exist, the patch for cross-canadian.bbclass
>        fixes problem.
>     * Need aarch64-poky-linux/usr/lib/arm-poky-linux-linux-gnueabi
>        which doesn't exist, the patch for libgcc-common.inc fixes the
>        problem.
>
>     [YOCTO #8616]
>
>
> This breaks the allarch sstate sanity test on the autobuilder which verifies
> that allarch recipes don't have different hashes for a no-op change to the machine.

Hi Ross,

After more investigations, the different between:
MACHINE=qemux86 bitbake -Snone meta-toolchain
MACHINE=qemuarm bitbake -Snone meta-toolchain

is do_extra_symlinks' checksum, there is an ABIEXTENSION for
nativesdk-libgcc-initial when qemuarm (eabi), and no "eabi" when
qemux86.

But as the commit said, the "eabi" is a must for qemuarm since its compiler
arm-pokymllib32-linux-gnueabi-gcc contains eabi, otherwise it doesn't work.

Maybe let the test case don't compare between qemuarm and qemux86 since
arm is special ? Or do you have any other suggestions, please ?


// Robert

>
> https://autobuilder.yoctoproject.org/main/builders/nightly-oe-selftest/builds/242/steps/Running%20oe-selftest/logs/stdio
>
> Ross