On 2024-06-24 06:10, Emanuele Rocca wrote: > On 2024-06-23 12:13, Chris Hofstaedtler wrote: > > I think this also means rebuilding an existing kernel in unstable > > or testing will also break. > > That is correct, I've built Linux 6.8.12 in a sid chroot and the > resulting kernel image is 26M and unbootable.
There is an important difference in the kernel configs of a kernel affected by this issue vs a functioning one: Affected: $ grep RELR ko/config-6.8.12-arm64 CONFIG_TOOLS_SUPPORT_RELR=y CONFIG_ARCH_HAS_RELR=y CONFIG_RELR=y Working: $ grep RELR ok/config-6.8.12-arm64 CONFIG_ARCH_HAS_RELR=y TOOLS_SUPPORT_RELR is set by tools-support-relr.sh: https://sources.debian.org/src/linux/6.8.12-1/init/Kconfig/?hl=102#L102 So it seems that affected binutils version set the Kconfigs above. Binutils upstream is now looking at the issue, see: https://sourceware.org/bugzilla/show_bug.cgi?id=31924 For the time being we could explicitly set CONFIG_RELR=n in debian/config/arm64/config as a workaround, while waiting for a binutils update.
