DLPX-98314 delphix-kernel: control.{aws,generic}.in still Depend on linux-modules-extra, which Ubuntu dropped for those flavors in 7.0

[lyriclake]

Problem

appliance-build fails to install the appliance image once a flavor's Ubuntu base drops the linux-modules-extra package split for that kernel line:

• aws: linux-kernel-aws#72 (rebase onto Ubuntu-aws-7.0-7.0.0-1009.9_24.04.1):

  delphix-kernel-7.0.0-1009-dx2026080418-69ffacf47-aws : Depends: linux-modules-extra-7.0.0-1009-dx2026080418-69ffacf47-aws but it is not installable
  

• generic (kvm/esx): every post-push/nightly build since 2026-08-08, once linux-kernel-generic#57 (DLPX-98112) rebased onto Ubuntu-hwe-7.0-7.0.0-28.28_24.04.1:

  delphix-kernel-7.0.0-28-dx2026080722-e92af378a-generic : Depends: linux-modules-extra-7.0.0-28-dx2026080722-e92af378a-generic but it is not installable
  

Both are the same fleet-wide Canonical initiative (Launchpad #2042831) dropping the modules-extra split flavor by flavor. Each time it lands for a flavor, that flavor's debian.<flavor>/control.d/flavour-control.stub loses its linux-modules-extra-PKGVER-ABINUM-FLAVOUR stanza and rules.d/amd64.mk stops setting do_extras_package = true — but this repo's debian/control.<flavor>.in keeps hard-declaring Depends: linux-modules-extra-@@KVERS@@, so apt can never satisfy it once that flavor's base drops the split.

This PR originally only touched control.aws.in, on the reasoning that other flavors' bases still shipped the split — true when opened (2026-08-05), but no longer true for generic once linux-kernel-generic#57 landed on 2026-08-07 (DLPX-98407, folded into DLPX-98314).

Solution

Remove linux-modules-extra-@@KVERS@@ from the Depends: list in debian/control.aws.in and debian/control.generic.in.

control.oracle.in, control.gcp.in, and control.azure.in are intentionally left untouched — those flavors' Ubuntu bases still ship the modules-extra split today. Removing the dependency preemptively for those flavors would be a silent functional regression, since apt would simply stop requiring a package that still exists rather than erroring, so driver/module content would quietly stop being installed by default until something broke in the field. Each other flavor should get this same treatment only once its own base tag actually drops the split.

Testing Done

aws: Validated with git ab-pre-push using --extra-repo to combine this fix branch with the linux-kernel-aws#72 patchset and the zfs sockaddr_unsized fix (delphix/zfs#2305, DLPX-98297), rerunning the full appliance-build-orchestrator-pre-push (build #14734, PLATFORMS=aws). Compile stage SUCCESS (appliance-build » pre-push #7165). Test stage: upgrade-testing #4621 → blackbox-chained #9578: FAILURE, 2/49 failures — an SSH connectivity timeout mid-upgrade (test_upgrade_linux_system), matching a known unresolved infra flake (DLPXQA-53392) rather than a regression from this fix or the rebase.

generic: the control.generic.in change is new to this PR (added for DLPX-98407) and has not yet had its own combined ab-pre-push run with the linux-kernel-generic#57 patchset — will add that result here before merge. Note the earlier "generic: UNSTABLE, 0/97" result recorded against this PR was from linux-kernel-generic#57's own standalone pre-push run, which did not include this delphix-kernel branch (build #14734 above only built PLATFORMS=aws) — it doesn't validate the generic fix in this PR.

[david-mendez1]

Looks fine, depending on ab-pre-push

[dbjwhs-perforce]

LGTM, I see something like this was almost done back in the day PR 20

[lyriclake]

Testing status update

ab-pre-push build #14734: compile stage SUCCESS

• Build: https://selfservice-jenkins.eng-tools-prd.aws.delphixcloud.com/job/appliance-build-orchestrator-pre-push/14734/
• Tested: linux-kernel-aws@69ffacf47d20 (#72), delphix-kernel@fe74868 (this PR), zfs@e62b28d4bbc4 (zfs#2305) (against develop)
• appliance-build » develop » pre-push #7165: SUCCESS — confirms this Depends: fix resolves the linux-modules-extra apt install failure; the appliance image now installs cleanly.
• The overall orchestrator build still shows FAILURE because of the test stage (upgrade-testing/dx-integration-tests), but both failures there are infra-related test-environment issues unrelated to this fix — see linux-kernel-aws#72 for the full breakdown and in-progress reruns.

This fix is confirmed working; no further action needed on this PR pending review.

[lyriclake]

Test stage final update

• linux-kernel-aws (#72): upgrade-testing #4621 → blackbox-chained #9578: FAILURE, 2/49 failures — an SSH connectivity timeout mid-upgrade (test_upgrade_linux_system), a different signature than the prior two rounds' setup-stage timeout. Still looks like infra rather than a regression from this delphix-kernel fix or the rebase, but the signature changed rather than repeating identically; see linux-kernel-aws#72 for the full breakdown and the open call on whether to rerun once more.
• linux-kernel-generic (#57): upgrade-testing #4622 → blackbox-chained #9579: UNSTABLE, but 0/97 failures (2 unrelated pre-existing skips). Effectively a clean pass (this fix is aws-only, so generic's result is informational).

Compile stage remains confirmed SUCCESS (appliance-build » pre-push #7165: SUCCESS); no code changes needed to this fix.