LGTM

for some reason i'm having issues now getting it to boot to system with boot_as_recovery = true. i'm 100% sure it was passing skip_initramfs before, but now it's just booting to recovery every time...

@danielfullmer can you report, whenever convenient, on whether your marlin boots in recovery mode with and without that change?

I may have a similar issue with oneplus 3, it seems all boots are recovery mode there too. Looking at the full /proc/cmdline may help here. androidboot.mode I believe is set by the bootloader.

(Related, #107)

Before this PR:
No problems with booting into nixos system. Recovery mode doesn't seem to work--it just boots into the system.

After this PR:
While booting, it would display an error mounting the vendor partition. Simply adding "ro" to the list of mount options in modules/initrd-vendor.nix was sufficient to fix this.
After this, I have no problems booting again. Booting into recovery seems to work as well.

As an aside:
I was having trouble with the aarch64 community builder (NixOS/nix#3422), and so I couldn't build a system.img. I've reflashed an android build at some point, so I didn't have an existing ext4 partition on userdata with the label NIXOS_SYSTEM. In this case, it would hang at the mobile nixos boot image. It would be nice if it would display an error message like it seemed to do for the missing vendor partition. (I was eventually able to find an old system.img I had created long ago and use that)

Thanks for taking a look!

Recovery

"Before this PR", there was no way to get in recovery mode for your device, since it uses the main boot.img as recovery. This is why you couldn't get in recovery mode.

Weird how your vendor partition requires "ro". Though it makes sense to use "ro" there, considering we should treat the vendor partition as a static opaque blob.

I wonder what makes it so @ashkitten's device continuously boots in recovery mode.

Failure in stage-1

Yep, #78 is tracking this. Right now if a task's dependencies never resolves it continues waiting forever. With a timeout the error condition will be made obvious.

Thanks for the update!