Several users reported problems in building and testing patched
kernels using the instructions in the
Kernel Handbook and the test-patches script
included in the source package for this purpose:
The test-patches script hadn't been updated to follow
the past few years' packaging changes, and produced somewhat
broken packages. It was also not robust to being interrupted
and restarted, and was needlessly slow due to running the whole
build process under fakeroot. I fixed all these
problems in the script.
I updated the Debian Kernel Handbook to cover the changes in
test-paches and to note the problems in older versions.
I revised the instructions for building without this script to
correctly cover disabling debug info, to enable parallel builds,
and to include building all required binary packages.
for security updates to the linux (4.19) and linux-5.10 packages
in Debian LTS.
I reviewed and accepted a
request updating linux to upstream version 6.3. I updated
further to stable update 6.3.1 and uploaded the package to the
Following the experimental upload, I investigated and
build failures on armel, mips64el, mipsel, and sh4 due to
increases in the kernel image size.
In cross-building linux for those architectures I found
regressions in the way we build the objtool command
that's used for post-processing and checking kernel code:
The upstream build rules for objtool always carry out a
native build so that it can be used during a cross-build of the
kernel. But we also need to be able to cross-build
objtool itself for inclusion in the
linux-kbuild-version package. Our previous hack to
do this broke.
objtool was originally introduced specifically to
handle x86 code, but now supports PowerPC as well. Since
linux-kbuild packages support cross-building kernel modules, a
single build of objtool will no longer be sufficient.
both of these issues.
I updated Debian's patch to fix reproducibility of the manual
pages for the perf tool, which was no longer working
and partly overlapped with upstream changes. The updated
version has now been applied upstream.
Unfortunately, due to reprotest's excessive memory
consumption when comparing large packages, we hadn't been
able to see that many other reproducibility issues have crept
into the linux package over the past years. I've
work on fixing those.
bug #1036019: debian-installer: Broken X display with QEMU under
UEFI with cirrus and std graphics
and found a one-line fix, but there is some reasonable concern
that my fix might cause regressions for other systems.
I reviewed and accepted several more merge requests for
linux targetting the master branch:
I made another upload of linux to the experimental suite with
all the above changes.
I reviewed a
request to update to a release candidate for 6.4 and fixed a
build regression. This isn't merged yet, but as soon as bookworm
is released the kernel team should be ready to upload packages
based on 6.3 and a 6.4 release candidate to unstable and
I updated the
branch of linux to upstream stable version 4.19.283, but
didn't make an upload this month.
I investigated Debian
bug #1036543: linux: WARNING at drivers/crypto/ccp/sev-dev.c:168
__sev_do_cmd_locked+0x31b/0x350 [ccp] and found that it was
due to an incomplete backport in a stable update. I've reported
the missing commits upstream.
I did some paid work on IPv6 support in Busybox, but I don't yet
have permission to make this public.
I started work on
IPv6 in klibc's ipconfig, as requested in Debian bug
#627164 12 years
I updated the
security tracker to add status for
branches to all active issues.
posted at: 18:50 | path: / | permanent link to this entry