summaryrefslogtreecommitdiffstats
path: root/meta/classes
diff options
context:
space:
mode:
authorMike Crowe <mac@mcrowe.com>2019-02-24 21:00:52 +0000
committerRichard Purdie <richard.purdie@linuxfoundation.org>2019-02-26 23:04:09 +0000
commita705c0782c863ee960d65b5109168a4587a0a7b7 (patch)
tree3448f819cf537af08b4cb4a24a910ec671b279c3 /meta/classes
parente39c5ec90ebbc37064c9cd59eba12603317740cd (diff)
downloadopenembedded-core-a705c0782c863ee960d65b5109168a4587a0a7b7.tar.gz
glibc: Install AArch64 loader link correctly for usrmerge+multilib
The AArch64 little-endian ABI requires that the dynamic loader is always available at /lib/ld-linux-aarch64.so.1. Similarly, the big-endian ABI requires that the dynamic loader is always available at /lib/ld-linux-aarch64_be.so.1. glibc-package.inc contains code that tries to ensure this, but unfortunately it is defeated by the combination of multilib and usrmerge because it does not take into account that /lib is the same as /usr/lib with usrmerge when it adds the loader path to libc_baselibs and when it attempts to show that /usr/lib is empty in do_poststash_install_cleanup. This results in the symlink not being included in the package and a build failure due to rmdir failing. Richard Purdie also suggested[1] that ${nonarch_base_libdir} should not be used as a synonym for /lib in this case. This hopefully-fixed version always sets ARCH_DYNAMIC_LOADER and then uses ${root_prefix}/lib/${ARCH_DYNAMIC_LOADER} to refer to the dynamic loader which works with both multilib and usrmerge. Since ARCH_DYNAMIC_LOADER is only non-empty if the symlink is required, the code to create it can move to do_install_append. Then do_poststash_install_cleanup needs to be taught that ${exec_prefix}/lib may not be empty if the dynamic loader symlink is there. It appears not to be possible to specify the name of the loader via a variable with an override, since the _aarch64 override is applied even for _aarch64-be, so I've set the loader name using ${TARGET_ARCH} instead. Build-tested and inspected core-image-minimal rootfs with: * AArch64 no multilib (real loader in correct place) MACHINE = "qemuarm64" * AArch64 multilib (symlink in correct place) MACHINE = "qemuarm64" MULTILIBS = "multilib:lib32" DEFAULTTUNE_virtclass-multilib-lib32 = "armv7at-neon" require conf/multilib.conf * AArch64 usrmerge (real loader in correct place) DISTRO_FEATURES += "usrmerge" MACHINE = "qemuarm64" * AArch64 multilib usrmerge (symlink in correct place) DISTRO_FEATURES += "usrmerge" MACHINE = "qemuarm64" MULTILIBS = "multilib:lib32" DEFAULTTUNE_virtclass-multilib-lib32 = "armv7at-neon" require conf/multilib.conf * big-endian versions of all of the above by also setting DEFAULTTUNE = "aarch64_be". (building glibc only.) * x86_64 (real loader in /lib as before)[2] MACHINE = "qemux86" * x86_64 multilib (real loader in /lib64 as before) MACHINE="qemux86-64" MULTILIBS = "multilib:lib32" DEFAULTTUNE_virtclass-multilib-lib32 = "x86" require conf/multilib.conf I also tested leaving an unwanted file in ${exec_prefix}/lib for do_poststash_install_cleanup to detect, and I believe the detection always worked correctly. [1] http://lists.openembedded.org/pipermail/openembedded-core/2018-November/276120.html Signed-off-by: Mike Crowe <mac@mcrowe.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Diffstat (limited to 'meta/classes')
0 files changed, 0 insertions, 0 deletions