mnt/reform: init #191

ehmry · 2020-09-23T15:19:02Z

Add support for the MNT Reform laptop with the initial i.MX8 CPU board - https://mntre.com/

I've used this configuration to bootstrap NixOS from the Debian SD card that comes with this laptop, but I am still working on an SD installer image.

Mic92 · 2020-09-23T15:46:19Z

mnt/reform2-nitrogen8m/uboot/default.nix

+{ buildUBoot, fetchgit }:
+
+buildUBoot rec {
+  pname = "uboot-reform2-imx8mq";


Please move the packages to nixpkgs, we cannot test/build in nixos-hardware. The module itself can remain here. If you post the PR url, I will have a look.

Depending on your motivation you can also add support here: https://github.com/Mic92/nixos-aarch64-images
This allows people without aarch64 to remaster our official aarch64 image with the custom u-boot.

cc @samueldr

~~As a complete reversal to my opinions about the kernel, U-Boot is "self-sustaining", since it's part of U-Boot mainline. Adding a build for a particular device is basically free.~~

~~We might want to stray away from including extraConfig. Sure, that means it'll have the default 3s boot delay, but that's how we build our U-Boots for now: stock as mainline does.~~

~~It's also quite cheap to build a custom U-Boot if needed by the end-user. That means it's not a huge deal if they need to customize it.~~

~~Except for: see my other comments about the patches.~~

Ugh, I ~~suck~~ didn't read carefully. There's an src here. (Leaving this comment up, but it is not relevant anymore.)

I'm not sure we want to introduce a non-mainline U-Boot into Nixpkgs. I went as far as making myself honest, by removing the last non-mainline U-Boot from Nixpkgs.

A bit like for kernels, the main reasoning behind isn't about configuration or device-specificity, but about maintainership. Adding U-Boot defconfigs for a mainline board is trivial, and pretty much free.

I guess this would be answered the same way the "extra" kernels would.

FWIW the U-Boot maintainer knows about these patches and wants this board fixed in U-Boot upstream.

Depending on your motivation you can also add support here: https://github.com/Mic92/nixos-aarch64-images This allows people with aarch64 to remaster our official aarch64 image with the custom u-boot.

Ok, I can do that when I have a working SD image.

@ehmry you basically only need to know the offsets where the chip expects uboot to be: https://github.com/Mic92/nixos-aarch64-images/blob/main/images/rockchip.nix#L19

@Mic92, be careful, that's not necessarily a given that an SoC will boot like Allwinners and Rockchips, by looking at a particular offset on the storage.

Though, here it looks like it may be, when configured appropriately.

https://community.nxp.com/t5/i-MX-Processors-Knowledge-Base/i-MX8-Boot-process-and-creating-a-bootable-image/ta-p/1101253

Mic92 · 2020-09-23T15:48:43Z

mnt/reform2-nitrogen8m/kernel.nix

+    name = "MNT-Reform-imx8mq-config";
+    patch = null;
+    extraConfig = builtins.readFile ./kernel-config;
+  }]);


@samueldr what should we do with the kernel?

This is still an open question about everything kernel related and "those kind" of platforms, where patches or outright complete kernel trees are needed.

I think we need to "sit down" and discuss that with the relevant key people: can we provide kernel builds in a useful manner, without making them part of Nixpkgs outright?

I'm thinking about a strategy similar to Mobile NixOS' own for reducing churn in Hydra, instead of tracking Nixpkgs, track channel updates. It means that it lags behind slightly, but doesn't needlessly queue builds... (Though that last part is not really an issue for kernels, few changes trigger rebuilds.)

I guess Flakes would help in this current situation, but adding a strict dependency on flakes is not great either.

Though one of the main important things about that is that managing non-mainline kernels through Nixpkgs is bound to become untenable. I already want to move the Raspberry Pi foundation kernels outside whenever possible, in addition to the few misc. probably-unused kernels still present.

Ugh, sorry to drop all that in a PR review request :/.

Oh, additionally linux_5_7 is bound to disappear. It's already marked EOL at kernel.org.

I actually would not those packages to hydra and the official binary cache but have github repositories in nix-community. We have a diverse community and only a few people with access to the nixos core infrastructure. In nix community there is a 1TB cachix cache and a hydra, which would allow to upload custom kernel builds and the whole process is a lot less bureaucratic. I think making every non-standard kernel its own flake might be actually the way to go solve this.

Ugh, sorry to drop all that in a PR review request :/.

Not a problem, the hardware hasn't shipped en masse, so we have time to do this right.

There is a level of trust around "just using a cachix repo", that I highly dislike. That's the main reason I don't publish anything for the Pinebook Pro WIP stuff, I don't want to have people either not realise they rely on an implicit trust on me, or have them trust what I do.

If the Nix community builder is transparent enough for that, it'd help, we'd probably be able to handle that for all NixOS on ARM things. I really want to move all the things in a centralized location with builds. @Mic92 care to open an issue and start the discussion about that at the right place (and tagging me in)?

@samueldr I just realized that we don't have aarch64 builds in nix-community. Apart from that adding declarative jobs is supported https://github.com/nix-community/infra/blob/0374f082455b9cce0a369bdc53cc4fa2ca9c87a0/build01/hydra-declarative-projects.nix#L5 So the only other alternative I see is to have a jobset on the official hydra. There might be also aarch64 support on travis. I don't know if this is enough to build kernel, but if it is than this is actually the better solution because than everyone can customize their built. Trust-wise it does not make a big difference if I have to trust your repo/cache or a couple of admins maintaining nix-community. Realistically most people won't understand the source code in your repository anyway and have to trust it.

samueldr · 2020-09-24T01:20:51Z

Not pertinent to reviewing the PR per se, but more about being a good citizen in the "NixOS on ARM" stuff.

I see you're patching environment variables here:

https://github.com/NixOS/nixos-hardware/blob/37d1eff124a54e8e78261acd56d3d09ea9a337dd/mnt/reform2-nitrogen8m/u-boot/env_vars.patch#L9-L30

Am I to understand that this ends up making that ~2018 era U-Boot use the Generic Distro Configuration Concept?

Well, I figure it must. Too bad about the lack of graphics in U-Boot :/.

mnt/reform2-nitrogen8m/u-boot/default.nix

samueldr · 2020-09-24T01:28:30Z

mnt/reform2-nitrogen8m/kernel.nix

+    name = "MNT-Reform-imx8mq-config";
+    patch = null;
+    extraConfig = builtins.readFile ./kernel-config;
+  }]);


Oh, additionally linux_5_7 is bound to disappear. It's already marked EOL at kernel.org.

samueldr · 2020-09-24T01:30:08Z

mnt/reform2-nitrogen8m/kernel-config

@@ -0,0 +1,2180 @@
+6LOWPAN n


Is this original work?

If it isn't, where can we get the updated recommended options to enable?

This is semi-original, I took the config of the stock MNT kernel, diffconfig'ed and trimed it a bit. Some stuff is disabled to reduce the compile time. MNT uses their patches against a 5.7 kernel, so I chose the same.

mnt/reform2-nitrogen8m/default.nix

mnt/reform2-nitrogen8m/kernel-config

ehmry · 2020-09-25T14:57:21Z

I've rebased the initial commit and pushed a second commit that adds an installer module. Not sure the best way to make it discoverable.

ehmry · 2020-09-25T17:12:00Z

FWIW I've been building the SD image by adding a nixpkgs input to the flake and a package

+    packages.aarch64-linux.reform2-nitrogen8m-sd-installer = let
+      installer = nixpkgs.lib.nixosSystem {
+        system = "aarch64-linux";
+        modules = [ ./mnt/reform2-nitrogen8m/installer.nix ];
+      };
+    in installer.config.system.build.sdImage;
+

https://mntre.com/

Update the DTS patch to disable the non-battery-backed RTC.

ehmry · 2020-12-09T11:39:41Z

I made a seprate repository here: https://github.com/nix-community/hardware-mnt-reform

This laptop needs a pinned nixpkgs to build its kernel, so I don't think its appropriate to keep here.

ehmry requested a review from samueldr September 23, 2020 15:19

ehmry force-pushed the reform branch 2 times, most recently from 6e85800 to 37d1eff Compare September 23, 2020 15:29

Mic92 reviewed Sep 23, 2020

View reviewed changes

samueldr reviewed Sep 24, 2020

View reviewed changes

mnt/reform2-nitrogen8m/u-boot/default.nix Outdated Show resolved Hide resolved

samueldr reviewed Sep 24, 2020

View reviewed changes

ehmry force-pushed the reform branch 2 times, most recently from 06a8c86 to 13e0bd7 Compare September 25, 2020 14:55

ehmry force-pushed the reform branch from 13e0bd7 to 81db742 Compare September 26, 2020 18:31

ehmry force-pushed the reform branch from 81db742 to fc33172 Compare October 6, 2020 13:34

ehmry added 2 commits October 25, 2020 12:41

mnt/reform: init

6444bda

https://mntre.com/

MNT Reform: update kernel patches

dcd3850

Update the DTS patch to disable the non-battery-backed RTC.

ehmry force-pushed the reform branch from fc33172 to 9515f1b Compare October 26, 2020 09:32

MNT Reform: trim unused kernel options

ad869a5

ehmry force-pushed the reform branch from 9515f1b to ad869a5 Compare October 26, 2020 09:37

ehmry closed this Dec 9, 2020

ehmry deleted the reform branch May 31, 2021 10:30

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

mnt/reform: init #191

mnt/reform: init #191

ehmry commented Sep 23, 2020

Mic92 Sep 23, 2020

Mic92 Sep 23, 2020 •

edited

Mic92 Sep 23, 2020

samueldr Sep 24, 2020 •

edited

samueldr Sep 24, 2020 •

edited

ehmry Sep 24, 2020

ehmry Sep 24, 2020

Mic92 Sep 25, 2020

samueldr Sep 25, 2020

Mic92 Sep 23, 2020

samueldr Sep 24, 2020

samueldr Sep 24, 2020

Mic92 Sep 24, 2020 •

edited

ehmry Sep 24, 2020

samueldr Sep 24, 2020

Mic92 Sep 26, 2020 •

edited

samueldr commented Sep 24, 2020

samueldr Sep 24, 2020

samueldr Sep 24, 2020

ehmry Sep 24, 2020

ehmry commented Sep 25, 2020 •

edited

ehmry commented Sep 25, 2020

ehmry commented Dec 9, 2020

mnt/reform: init #191

mnt/reform: init #191

Conversation

ehmry commented Sep 23, 2020

Choose a reason for hiding this comment

Mic92 Sep 23, 2020 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

samueldr Sep 24, 2020 • edited

Choose a reason for hiding this comment

samueldr Sep 24, 2020 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Mic92 Sep 24, 2020 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Mic92 Sep 26, 2020 • edited

Choose a reason for hiding this comment

samueldr commented Sep 24, 2020

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ehmry commented Sep 25, 2020 • edited

ehmry commented Sep 25, 2020

ehmry commented Dec 9, 2020

Mic92 Sep 23, 2020 •

edited

samueldr Sep 24, 2020 •

edited

samueldr Sep 24, 2020 •

edited

Mic92 Sep 24, 2020 •

edited

Mic92 Sep 26, 2020 •

edited

ehmry commented Sep 25, 2020 •

edited