ci(x86_64): add Asterinas zone1 Jenkins test

[yydawx]

Motivation

Add Asterinas zone1 coverage to the existing x86_64 QEMU Jenkins CI while keeping x86_64/qemu as the Linux zone1 test flow.

Approach

Use a dedicated x86_64/qemu_asterinas BID for Asterinas instead of mixing Linux and Asterinas in the same BID. The existing x86_64/qemu BID remains unchanged and only tests Linux zone1.

platform_board: qemu decouples the Jenkins BID from the make/platform board:

• BID x86_64/qemu_asterinas is only a test identity and server artifact directory.
• make still uses the real platform BOARD=qemu because platform/x86_64/qemu_asterinas does not exist in the source tree.
• artifact_dir: qemu_asterinas selects the Asterinas artifact directory.

Artifact layout

qemu_asterinas provides:

• image/kernel/aster-kernel-osdk-bin
• image/virtdisk/initramfs.cpio.gz

The shared x86_64/qemu artifacts still provide zone0 files:

• image/kernel/setup.bin
• image/kernel/vmlinux.bin
• image/virtdisk/rootfs1.img

Changes

• Jenkinsfile: add platform_board parsing and the x86_64/qemu_asterinas matrix cell.
• jenkins/ci.yaml: add the dedicated BID and keep Linux zone1 in x86_64/qemu.
• jenkins/ci_config.py: expose optional platform_board.
• jenkins/ci_runner.py: map BID to the real platform board, clear BID for hvisor make, and add asterinas_zone1_regression.
• jenkins/prepare.sh: deploy Asterinas kernel/initramfs, Multiboot2 bootloader, and create zone1_disk.img with mkfs.ext2 -b 4096.
• jenkins/terminal.py: add send_one_by_one() for sensitive guest consoles.
• platform/x86_64/qemu/scripts/boot_zone1_asterinas.sh: start Asterinas zone1 with the Asterinas virtio config.

QEMU lifecycle

Each BID starts one QEMU instance and runs its own test cases. There is no Linux/Asterinas state sharing and no restart between two guest types inside one BID.

Verification

• x86_64/qemu passes Linux zone1 startup.
• x86_64/qemu_asterinas starts Asterinas zone1 and runs /test/run_regression_test.sh.
• CI polls Asterinas regression output until it sees both All regression tests passed and exit code 0; a non-zero regression exit fails immediately.
• After regression, CI sends Ctrl-A d to detach from the Asterinas screen and verifies the zone0 root@zone0 prompt.
• Jenkins build 8 is fully green, including x86_64/qemu_asterinas.

[Solicey]

I think it is OK to add a new BID for x86_64/qemu_asterinas

[yydawx]

@Solicey, could you please review the updated PR?

We decouple the Jenkins BID from the make/platform board via platform_board: qemu: the BID x86_64/qemu_asterinas is only a test identity and artifact directory, while make still uses the real platform BOARD=qemu.

If we kept the old rule that the BID board must equal the platform board, and still wanted a dedicated Asterinas BID, we would have to introduce a platform/x86_64/qemu_asterinas directory or force make BOARD=qemu_asterinas. That would turn a test scenario into a new platform, break the platform abstraction, and add maintenance cost.

[Solicey]

It would be better to add more test instructions to show that zone1 works fine. For example, screen /dev/pts/0 to enter zone1, ls, Ctrl A+D to return to zone0. You can refer to qemu/x86_64 for this.

[yydawx]

@Solicey, thanks for the review. We updated the CI flow: after entering the Asterinas zone1 console, it now runs /test/run_regression_test.sh, waits for All regression tests passed and exit code 0, then sends Ctrl-A d to detach back to zone0 and verifies the root@zone0 prompt. Jenkins build 8 is green, including x86_64/qemu_asterinas.

[yydawx]

@dallasxy, could you please review this PR?

The PR adds a dedicated x86_64/qemu_asterinas Jenkins BID while keeping x86_64/qemu as the Linux zone1 flow. It decouples the Jenkins BID from the make/platform board via platform_board: qemu, and x86_64/qemu_asterinas now runs Asterinas zone1 regression tests.

Jenkins build 8 is fully green, including x86_64/qemu_asterinas.

[Solicey]

Can you explain the reason for adding a send_one_by_one function?

[yydawx]

Thanks for the question. We found that the current Asterinas virtio-console receives one byte per IRQ, which conflicts with sending the whole command at once via send(). I also observed this in normal use: the Asterinas terminal cannot paste, and screen/tmux also require byte-by-byte input. That is why I added send_one_by_one. After removing it and using the old send(), Jenkins build 10 hung because send() could not deliver the complete command:
http://183.242.33.186:8100/job/hvisor_pr_ci_2/job/PR-376/10/

One alternative is to use the existing send path in a char-by-char loop to build the full command. I currently prefer a single helper because it keeps the terminal abstraction clean.

Which approach should we take: keep send_one_by_one, or use send() in a loop to send one character at a time?

[Solicey]

It seems that there are some conflicts with the dev branch, you can resolve these conflicts manually.