Bare-metal CI

The bare-metal scripts run on a system with gitlab-runner and Docker, connected to potentially multiple bare-metal boards that run tests of Mesa. Currently “fastboot”, “ChromeOS Servo”, and POE-powered devices are supported.

In comparison with LAVA, this doesn’t involve maintaining a separate web service with its own job scheduler and replicating jobs between the two. It also places more of the board support in Git, instead of web service configuration. On the other hand, the serial interactions and bootloader support are more primitive.

Requirements (fastboot)

This testing requires power control of the DUTs by the gitlab-runner machine, since this is what we use to reset the system and get back to a pristine state at the start of testing.

We require access to the console output from the gitlab-runner system, since that is how we get the final results back from the tests. You should probably have the console on a serial connection, so that you can see bootloader progress.

The boards need to be able to have a kernel/initramfs supplied by the gitlab-runner system, since Mesa often needs to update the kernel either for new DRM functionality, or to fix kernel bugs.

The boards must have networking, so that we can extract the dEQP XML results to artifacts on GitLab, and so that we can download traces (too large for an initramfs) for trace replay testing. Given that we need networking already, and our dEQP/Piglit/etc. payload is large, we use NFS from the x86 runner system rather than initramfs.

See src/freedreno/ci/gitlab-ci.yml for an example of fastboot on DB410c and DB820c (freedreno-a306 and freedreno-a530).

Requirements (Servo)

For Servo-connected boards, we can use the EC connection for power control to reboot the board. However, loading a kernel is not as easy as fastboot, so we assume your bootloader can do TFTP, and that your gitlab-runner mounts the runner’s tftp directory specific to the board at /tftp in the container.

Since we’re going the TFTP route, we also use NFS root. This avoids packing the rootfs and sending it to the board as a ramdisk, which means we can support larger rootfses (for Piglit testing), at the cost of needing more storage on the runner.

Telling the board about where its TFTP and NFS should come from is done using dnsmasq on the runner host. For example, this snippet in the dnsmasq.conf.d in the Google farm, with the gitlab-runner host we call “servo”:

dhcp-host=1c:69:7a:0d:a3:d3,10.42.0.10,set:servo

# Fixed dhcp addresses for my sanity, and setting a tag for
# specializing other DHCP options
dhcp-host=a0:ce:c8:c8:d9:5d,10.42.0.11,set:cheza1
dhcp-host=a0:ce:c8:c8:d8:81,10.42.0.12,set:cheza2

# Specify the next server, watch out for the double ',,'.  The
# filename didn't seem to get picked up by the bootloader, so we use
# tftp-unique-root and mount directories like
# /srv/tftp/10.42.0.11/jwerner/cheza as /tftp in the job containers.
tftp-unique-root
dhcp-boot=tag:cheza1,cheza1/vmlinuz,,10.42.0.10
dhcp-boot=tag:cheza2,cheza2/vmlinuz,,10.42.0.10

dhcp-option=tag:cheza1,option:root-path,/srv/nfs/cheza1
dhcp-option=tag:cheza2,option:root-path,/srv/nfs/cheza2

See src/freedreno/ci/gitlab-ci.yml for an example of Servo on cheza. Note that other Servo boards in CI are managed using LAVA.

Requirements (POE)

For boards with 30W or less power consumption, POE can be used for the power control. The parts list ends up looking something like (for example):

  • x86-64 gitlab-runner machine with a mid-range CPU, and 3+ GB of SSD storage per board. This can host at least 15 boards in our experience.

  • Cisco 2960S gigabit ethernet switch with POE. (Cisco 3750G, 3560G, or 2960G were also recommended as reasonable-priced HW, but make sure the name ends in G, X, or S)

  • POE splitters to power the boards (you can find ones that go to micro USB, USBC, and 5V barrel jacks at least)

  • USB serial cables (Adafruit sells pretty reliable ones)

  • A large powered USB hub for all the serial cables

  • A pile of ethernet cables

You’ll talk to the Cisco for configuration using its USB port, which provides a serial terminal at 9600 baud. You need to enable SNMP control, which we’ll do using a “mesaci” community name that the gitlab runner can access as its authentication (no password) to configure. To talk to the SNMP on the router, you need to put an IP address on the default VLAN (VLAN 1).

Setting that up looks something like:

With that set up, you should be able to power on/off a port with something like:

Note that the “1.3.6…” SNMP OID changes between switches. The last digit above is the interface id (port number). You can probably find the right OID by Google, that was easier than figuring it out from finding the switch’s MIB database. You can query the POE status from the switch serial using the show power inline command.

Other than that, find the dnsmasq/tftp/NFS setup for your boards “servo” above.

See src/broadcom/ci/gitlab-ci.yml and src/nouveau/ci/gitlab-ci.yml for an examples of POE for Raspberry Pi 3/4, and Jetson Nano.

Setup

Each board will be registered in freedesktop.org GitLab. You’ll want something like this to register a fastboot board:

sudo gitlab-runner register \
     --url https://gitlab.freedesktop.org \
     --registration-token $1 \
     --name MY_BOARD_NAME \
     --tag-list MY_BOARD_TAG \
     --executor docker \
     --docker-image "alpine:latest" \
     --docker-volumes "/dev:/dev" \
     --docker-network-mode "host" \
     --docker-privileged \
     --non-interactive

For a Servo board, you’ll need to also volume mount the board’s NFS root dir at /nfs and TFTP kernel directory at /tftp.

The registration token has to come from a freedesktop.org GitLab admin going to https://gitlab.freedesktop.org/admin/runners

The name scheme for Google’s lab is google-freedreno-boardname-n, and our tag is something like google-freedreno-db410c. The tag is what identifies a board type so that board-specific jobs can be dispatched into that pool.

We need privileged mode and the /dev bind mount in order to get at the serial console and fastboot USB devices (–device arguments don’t apply to devices that show up after container start, which is the case with fastboot, and the Servo serial devices are actually links to /dev/pts). We use host network mode so that we can spin up a nginx server to collect XML results for fastboot.

Once you’ve added your boards, you’re going to need to add a little more customization in /etc/gitlab-runner/config.toml. First, add concurrent = <number of boards> at the top (“we should have up to this many jobs running managed by this gitlab-runner”). Then for each board’s runner, set limit = 1 (“only 1 job served by this board at a time”). Finally, add the board-specific environment variables required by your bare-metal script, something like:

[[runners]]
  name = "google-freedreno-db410c-1"
  environment = ["BM_SERIAL=/dev/ttyDB410c8", "BM_POWERUP=google-power-up.sh 8", "BM_FASTBOOT_SERIAL=15e9e390", "FDO_CI_CONCURRENT=4"]

The FDO_CI_CONCURRENT variable should be set to the number of CPU threads on the board, which is used for auto-tuning of job parallelism.

Once you’ve updated your runners’ configs, restart with sudo service gitlab-runner restart

Caching downloads

To improve the runtime for downloading traces during traces job runs, you will want a pass-through HTTP cache. On your runner box, install nginx:

sudo apt install nginx libnginx-mod-http-lua

Add the server setup files:

/etc/nginx/sites-available/fdo-cache
proxy_cache_path /var/cache/nginx/ levels=1:2 keys_zone=my_cache:10m max_size=24g inactive=48h use_temp_path=off;

server {
	listen 10.42.0.1:80 default_server;
	listen 127.0.0.1:80 default_server;
	listen [::]:80 default_server;
	resolver 8.8.8.8;

	root /var/www/html;

	# Add index.php to the list if you are using PHP
	index index.html index.htm index.nginx-debian.html;

	server_name _;

        add_header X-GG-Cache-Status $upstream_cache_status;
        proxy_cache my_cache;

        location /cache_gitlab_artifacts {
                internal;
                # Gitlabs http server puts everything as no-cache even though
                # the artifacts URLS don't change. So enforce a long validity
                # time and ignore the headers that defeat caching
                proxy_cache_valid 200 48h;
                proxy_ignore_headers Cache-Control Set-Cookie;
                include snippets/uri-caching.conf;
        }

        location /cache {
                # special case gitlab artifacts
                if ($arg_uri ~*  /.*gitlab.*artifacts(\/|%2F)raw/ ) {
                        rewrite ^ /cache_gitlab_artifacts;
                }
                # Set a really low validity together with cache revalidation; Our goal
                # for caching isn't to lower the number of http requests but to
                # lower the amount of data transfer. Also for some test
                # scenarios (typical manual tests) the file at a given url
                # might get modified so avoid confusion by ensuring
                # revalidations happens often.
                proxy_cache_valid 200 10s;
                proxy_cache_revalidate on;
                include snippets/uri-caching.conf;
        }
}
/etc/nginx/snippets/uri-caching.conf
set $proxy_authorization '';

set_by_lua $proxyuri '
        unescaped =  ngx.unescape_uri(ngx.var.arg_uri);
        it, err = ngx.re.match(unescaped, "(https?://)(.*@)?([^/]*)(/.*)?");
        if not it then
                -- Hack to cause nginx to return 404
                return "http://localhost/404"
        end

        scheme = it[1];
        authstring = it[2];
        host = it[3];
        query = it[4];

        if ngx.var.http_authorization and ngx.var.http_authorization ~= "" then
                ngx.var.proxy_authorization = ngx.var.http_authorization;
        elseif authstring then
                auth = string.sub(authstring, 0, -2);
                auth64 = ngx.encode_base64(auth);
                ngx.var.proxy_authorization = "Basic " .. auth64;
        end

        -- Default to / if none is set to avoid using the request_uri query
        if not query then
                query = "/";
        end

        return scheme .. host .. query;
';

add_header X-GG-Cache-Status $upstream_cache_status;
proxy_set_header Authorization $proxy_authorization;

proxy_pass $proxyuri;
# Redirect back to ourselves on 301 replies
proxy_redirect ~^(.*)$ /cache/?uri=$1;

Edit the listener addresses in fdo-cache to suit the ethernet interface that your devices are on.

Enable the site and restart nginx:

sudo rm /etc/nginx/sites-enabled/default
sudo ln -s /etc/nginx/sites-available/fdo-cache /etc/nginx/sites-enabled/fdo-cache
sudo systemctl restart nginx

# First download will hit the internet
wget http://localhost/cache/?uri=https://s3.freedesktop.org/mesa-tracie-public/itoral-gl-terrain-demo/demo-v2.trace
# Second download should be cached.
wget http://localhost/cache/?uri=https://s3.freedesktop.org/mesa-tracie-public/itoral-gl-terrain-demo/demo-v2.trace

Now, set download-url in your traces-*.yml entry to something like http://caching-proxy/cache/?uri=https://s3.freedesktop.org/mesa-tracie-public and you should have cached downloads for traces. Add it to FDO_HTTP_CACHE_URI= in your config.toml runner environment lines and you can use it for cached artifact downloads instead of going all the way to freedesktop.org on each job.