Martin Michlmayr found that userland builds fail if 'uname -m' reports
the native kernel architecture, and that's because we don't explicitly
set ARCH for the headers_install target. That's fine for most cases
of a 32-bit chroot on a 64-bit kernel because a single source
architecture covers both, but arm/arm64 is an exception.
This doesn't make actual cross-builds of userland tools work.
These packages will be taken over by src:linux-signed. Still do
everything but building the packages so we find configuration
errors before building linux-signed.
linux-kernel-headers was removed from unstable in 2007, so presumably
was last released in etch.
libkxl0-dev still depends on it, so don't remove the Provides yet.
Previously each tarball would be uploaded to .../rt/<upstream-version>/
and then later moved to .../rt/<upstream-version>/older/. Now they are
appearing in .../older/ immediately so we can construct stable URLs
under there.
Currently the postinst script distinguishes installation vs upgrade by
checking whether its second argument is empty. However, it will only
be empty if the package was *never* installed before, or was purged.
If the package has been installed, then removed, and is now being
installed again, it's seen as an upgrade.
Unfortunately dpkg only tells the preinst script whether or not this
is an upgrade. So create a flag file there, consume it in postinst
and remove it in postrm.
Use the new linux-update-symlinks command for symlink updates.
Drop support for minimal_swap, no_symlinks, use_hard_links and *_hook
parameters in /etc/kernel-img.conf.
Thanks to Aurelien Jarno and Santiago Vila for pointing out the
ischroot command and that it's in an essential package (i.e. doesn't
add a dependency).
Please enable the following options in the armmp kernel configuration,
which enables support for Exynos4, tested on Odroid-U3.
It detects all four CPUs, 2GB of ram, ethernet, and supports USB on
both 4.5.x and 4.6.x. It also supports microSD, and 4.6.x worked fine,
but 4.5.x had some outstanding issues that prevented it from working
correctly. I did not test video or audio output, though modules for
video output did appear to load.
[ Roger Shimizu: modified the comment ]
Most symbols that depend on PM_DEVFREQ - such as the newly enabled
ARM_TEGRA_DEVFREQ - do not select it. Currently it is selected by
SCSI_UFSHCD, but we should not depend on that.
This seems to have been a transitional feature for the switch
from kernel-image to linux-image. mkvmlinuz still uses it but
no hook scripts appear to depend on it.