Skip to content

[Backport wrynose] Release update: master PRs #161-#169 - #171

Merged
otavio merged 16 commits into
OSSystemsEmbeddedLinux:wrynosefrom
lucianogdittgen:backport/master-merged-2-wrynose
Nov 17, 2026
Merged

otavio merged 16 commits into
OSSystemsEmbeddedLinux:wrynosefrom
lucianogdittgen:backport/master-merged-2-wrynose

Conversation

@lucianogdittgen

Copy link
Copy Markdown
Contributor

Backport to wrynose of the master PRs merged since the last backport (#153/#154) that fit a stable branch. Each commit is a cherry-pick -x, in master merge order. #165's conf/layer.conf hunk was resolved by hand (context only: LAYERSERIES_COMPAT stays wrynose).

Left on master on purpose:

Tested on wrynose (MACHINE=fri2, DISTRO=oel and oel-tiny), before (wrynose head) versus after:

  • bitbake -S lockedsigs core-image-minimal gstreamer1.0-plugins-base zram-init net-persistent-mac parses with no errors.
  • core-image-minimal: no task signature changes. Only zram-init, net-persistent-mac and gstreamer1.0-plugins-base change; for gstreamer1.0-plugins-base the only difference is OPENGL_WINSYS " " -> "" (PACKAGECONFIG, EXTRA_OEMESON and DEPENDS are identical).
  • bitbake -e: DISTRO_FEATURES of target and native recipes identical, so distro/oel: Opt in to default DISTRO_FEATURES explicitly #169 is a no-op here too.
  • OE-Core busybox on wrynose has CONFIG_IP/CONFIG_FEATURE_IP_LINK, which the net-persistent-mac fix uses.

The runtime tests were not re-run on wrynose; the recipes and scripts they cover are identical to master, where testimage (qemux86-64, oel) passes them.

It relies on bb.persist_data, removed from BitBake (681a7516e), and on
Python 2's contextlib.nested, so every oel build crashes at ConfigParsed
once a saved_persist_data.db exists. Drop the class and its OELDIR and
DUMP_HEADREVS_DB settings in oel.inc; nothing else uses them.
Tested with bitbake -p (977 recipes, 0 errors).

Signed-off-by: Luciano Dittgen <luciano.dittgen@ossystems.com.br>
(cherry picked from commit 61a2711)
No sstate_stage_all function exists in OE-Core or BitBake, so the flag
lands on an empty variable and affects no task signature.

Signed-off-by: Luciano Dittgen <luciano.dittgen@ossystems.com.br>
(cherry picked from commit 91971fc)
OE-Core's classes-global/mirrors.bbclass shadows it, so it was never
loaded, and nothing inherits it by name. Its URLs (angstrom, nexcess,
gbxs, plain-http autobuilder) are dead anyway. Verified with bitbake -e
busybox: only the OE-Core class is in the include history and none of
these mirrors appear in MIRRORS.

Signed-off-by: Luciano Dittgen <luciano.dittgen@ossystems.com.br>
(cherry picked from commit 53c701f)
OE-Core removed cve-check (00de455f8d), whose JSON report this class
filtered, so it cannot run on blacksail or wrynose. The README now points
at OE-Core's sbom-cve-check fragment. Tested with bitbake -p (0 errors).

Signed-off-by: Luciano Dittgen <luciano.dittgen@ossystems.com.br>
(cherry picked from commit 5e0d46a)
OE-Core has carried PACKAGECONFIG[viv-fb], with the same
virtual/libgles2 virtual/libg2d dependencies, and the viv-fb entry in
OPENGL_WINSYS since 846564f1a9 and ec3a1cb771 (2021-04-15), the same
day this staging copy was added. The bbappend only duplicates it, and
choosing a Vivante window system is BSP policy that does not belong in
a distro layer anyway; a BSP layer enables it through PACKAGECONFIG.

Signed-off-by: Luciano Dittgen <luciano.dittgen@ossystems.com.br>
(cherry picked from commit f750d96)
The script once handled only $INTERFACE. The INTERFACES loop replaced it,
but a copy of the old block stayed after the loop and repeated the work
for the last interface, plus an IF_FILE built from the unset variable.
Behaviour is unchanged.

Signed-off-by: Luciano Dittgen <luciano.dittgen@ossystems.com.br>
(cherry picked from commit 7eab481)
OE-Core's busybox no longer builds ifconfig, so on oel the script stored
the MAC but failed restoring it ("ifconfig: not found"). busybox and
iproute2 both provide 'ip link set dev <if> address <mac>'. Tested with
the new ossystems_net_persistent_mac oeqa test on qemux86-64.

Signed-off-by: Luciano Dittgen <luciano.dittgen@ossystems.com.br>
(cherry picked from commit c0926e4)
Checks that the restore ran at boot (anchor and unit result) and that
removing the anchor restores the factory files and runs the hooks once.
Tested with testimage on qemux86-64.

Signed-off-by: Luciano Dittgen <luciano.dittgen@ossystems.com.br>
(cherry picked from commit a46348f)
Checks the unit result, that each configured interface has its MAC
stored, and on a dummy interface that a stored MAC is applied. Tested
with testimage on qemux86-64.

Signed-off-by: Luciano Dittgen <luciano.dittgen@ossystems.com.br>
(cherry picked from commit 766821d)
Checks that the units are enabled and, headless, that 'easysplash open'
plays the installed animation into fakesink and exits 0 after 'easysplash
client --stop'. Tested with testimage on qemux86-64.

Signed-off-by: Luciano Dittgen <luciano.dittgen@ossystems.com.br>
(cherry picked from commit cb13147)
Checks that zram_swap.service is active and a zram swap is listed in
/proc/swaps; skips on kernels without zram. It fails today on linux-yocto
with features/zram: the unit asks for zstd, and the kernel builds only
the lzo zram backend.

Signed-off-by: Luciano Dittgen <luciano.dittgen@ossystems.com.br>
(cherry picked from commit 8e9fc07)
zram_swap.service sized the device with "free -m", which BusyBox
ignores, so the size came out in KiB and zram-init failed with "cannot
set zram0 size" (110880 MB asked on a 216 MiB guest). Patch the unit to
read MemTotal instead. Tested with testimage on qemux86-64 (oel).

Signed-off-by: Luciano Dittgen <luciano.dittgen@ossystems.com.br>
(cherry picked from commit efb9190)
"zram-init 0" wrote /dev/zram0 to zram-control/hot_remove, which takes
the id, so it exited 1 after tearing the swap down and failed ExecStop
of zram_swap.service. Tested with testimage on qemux86-64 (oel), which
now stops and restarts the unit.

Signed-off-by: Luciano Dittgen <luciano.dittgen@ossystems.com.br>
(cherry picked from commit 7c6cf34)
The sysvinit branch installed upstream's openrc-run script, which needs
OpenRC, so zram-init never ran under sysvinit. Ship a small LSB script
doing what zram_swap.service does, with ZRAM_SIZE and ZRAM_ALGO
overridable in /etc/default/zram-init. Tested start/stop/restart/status
by hand on qemux86-64.

Signed-off-by: Luciano Dittgen <luciano.dittgen@ossystems.com.br>
(cherry picked from commit 0eaa700)
The test only covered start, so the broken ExecStop went unnoticed.
Tested with testimage on qemux86-64 (oel).

Signed-off-by: Luciano Dittgen <luciano.dittgen@ossystems.com.br>
(cherry picked from commit 561a5eb)
The opt-out list mirrored OE-Core's DISTRO_FEATURES_DEFAULTS, so any
feature OE-Core adds there was silently enabled in oel. oel-tiny's
native/nativesdk relied on the full defaults; list the three features
they actually take. Effective DISTRO_FEATURES and all task signatures
of core-image-minimal (oel, oel-tiny) are unchanged.

Signed-off-by: Luciano Dittgen <luciano.dittgen@ossystems.com.br>
(cherry picked from commit ff86535)
@otavio
otavio merged commit 09fbca8 into OSSystemsEmbeddedLinux:wrynose Nov 17, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants