diff options
| author | Kent Overstreet <kent.overstreet@linux.dev> | 2026-08-31 00:49:23 -0500 |
|---|---|---|
| committer | Kent Overstreet <kent.overstreet@linux.dev> | 2026-08-31 00:53:50 -0500 |
| commit | fd0e84863fd7bb32524334b84c69f760a53b4691 (patch) | |
| tree | 85b8ef5e8f652ba0f975bac314f478598ed0a0bc /root_image.nix | |
| parent | b88d595f10d32fa54199d8873f8cfc3ae8263ed4 (diff) | |
Splash rendering was previously a question you could only answer by
rebooting a laptop and watching. Four pieces, together enough to test it:
KTEST_GUI=1 drops qemu's -nographic (-display gtk -vga std), for when a
person does need to look; ktest screendump takes the display as an image
over the monitor socket, which works headless because qemu renders
whether or not anyone is watching.
kconfig.sh gains a framebuffer, because with no DRM and no fbdev plymouth
falls back to the text plugin - whose message erase is one hardcoded line,
so a multi-line status block scrolls instead of redrawing. It decides what
plymouth renders with, not just what the pixels look like.
root_image.nix names what root_image and debootstrap actually invoke
instead of relying on the host, and buildFHSUserEnv is gone from nixpkgs.
mount.ktest reads /dev/vcsa1 - the VT layer's own character buffer - so
"what does the person booting see" has an answer in the test log.
HELD BACK, not ready to push: the file-scope kernel command line it adds
("splash plymouth.debug=stream") applies to every test in mount.ktest, and
the console tests there care about the console.
Co-Authored-By: Proof of Concept <poc@bcachefs.org>
Diffstat (limited to 'root_image.nix')
| -rw-r--r-- | root_image.nix | 32 |
1 files changed, 28 insertions, 4 deletions
diff --git a/root_image.nix b/root_image.nix index 994eea0c..7587967d 100644 --- a/root_image.nix +++ b/root_image.nix @@ -1,14 +1,38 @@ # FHS shell for running ./root_image on NixOS # execute: -# $ nix-shell root_image.nix -# $ ./root_image <params> +# $ sudo nix-shell root_image.nix +# $ ./root_image create # $ exit - +# +# sudo outside rather than inside: root_image refuses to run as anyone else, +# and sudo from within the FHS env leaves it again. +# +# targetPkgs is what ./root_image and the vendored debootstrap call by name, +# taken from what they actually invoke rather than from whatever a Debian host +# happens to have lying around: +# +# root_image - fallocate, mount, umount (util-linux), chroot (coreutils), +# mkfs.ext4 (e2fsprogs), curl, rsync +# debootstrap - wget, ar (binutils), tar, xz, gpgv (gnupg), dpkg-deb (dpkg), +# perl { pkgs ? import <nixpkgs> {} }: -(pkgs.buildFHSUserEnv { +(pkgs.buildFHSEnv { name = "root_image-env"; targetPkgs = pkgs: with pkgs; [ coreutils + util-linux + e2fsprogs + curl + wget + rsync + binutils + gnutar + xz + gzip + gnupg + dpkg + perl + debianutils ]; runScript = "bash"; }).env |