[WIP] linux: translate kernel config to structured config #12158

copumpkin · 2016-01-05T14:25:23Z

Note that this isn't used yet; I'm just seeing what people think. If feedback is positive I'll actually plug it into the real config machinery. I considered doing something fancier by modeling options with the module system but it seemed like a lot of effort for not much gain when in practice most people don't tweak most options, and there's only a handful that people often turn on and off.

This is one approach to #11994.

cc @aszlig whose merging code I borrowed

copumpkin · 2016-01-05T14:33:57Z

pkgs/os-specific/linux/kernel/structured-config.nix

+      "KVM_COMPAT?"                       = whenAtLeast "4.0"  "y";
+      "KVM_DEVICE_ASSIGNMENT?"            = whenAtLeast "3.10" "y";
+      "KVM_GENERIC_DIRTYLOG_READ_PROTECT" = whenAtLeast "4.0"  "y";
+      "KVM_GUEST"                         = when (!grsecurity) "y";


Note that the logic here is slightly different because I think the previous one had a bug in it. cc @wkennington for clarification.

https://github.com/NixOS/nixpkgs/blob/master/pkgs/os-specific/linux/kernel/common-config.nix#L421-L423 for context.

edolstra · 2016-01-05T16:32:45Z

IIRC, manual-config.nix already has some support for passing a kernel config as an attribute set. At least I see lines like

    config = { CONFIG_MODULES = "y"; CONFIG_FW_LOADER = "m"; };

We should probably unify manual-config.nix and generic.nix but that's probably a separate issue.

edolstra · 2016-01-05T16:35:42Z

pkgs/os-specific/linux/kernel/structured-config.nix

+      "RCU_TORTURE_TEST"           = "n";
+      "SCHEDSTATS"                 = "n";
+      "DETECT_HUNG_TASK"           = "y";
+      "CRASH_DUMP?"                = "n";


Putting ? in the option names is probably a bad idea because it would complicate merging values. E.g. one module might set CRASH_DUMP? and another CRASH_DUMP. It might be better to move the ? to the right-hand side, e.g.

CRASH_DUMP = "?n";

or in a structured way

CRASH_DUMP = ifSupported "n";

copumpkin · 2016-01-05T16:37:58Z

@edolstra yeah, I was on the fence about that. It would also allow me to unquote all the attribute names. I'll do that, thanks.

Speaking of merging values, I should probably also warn/barf if there are multiple values for a given key. That starts looking more like the module system again 😄

Note that this isn't used yet. If feedback is positive I'll actually plug it into the real config machinery.

copumpkin · 2016-01-06T03:44:16Z

I pushed another version if anyone wants to take a look. I'd rather not make it fancy with the module system at this point, since I don't expect people to be tweaking it all that much, but I can start that process if y'all think it's worthwhile.

My main unknown about the module system is how to mark certain options as irrelevant for certain versions.

copumpkin · 2016-01-06T14:33:24Z

N.B: this is WIP because if people like the basic idea, I'll actually integrate it with other parts of the kernel config and then expand this PR to include those changes too.

This helps in some weird screens that otherwise show the console 90° turned.

aszlig · 2016-01-21T01:06:50Z

@copumpkin: 👍 I like the basic idea and also the structured approach, which I had in mind as well. More specifically for my own machines the idea was to have a basic kernel config and then use other NixOS options/modules to extend it using requiredKernelConfig, so it would be cool to tie it in there as well :-)

rasendubi · 2016-07-25T23:49:35Z

(triage) @copumpkin status?

spacekitteh · 2016-08-18T04:59:11Z

I like it! Currently trying to figure out how to stop building all these useless kernel modules I don't need has been a nightmare due to the rather hacky nature of the current system. Yours is much better!

spacekitteh · 2016-08-18T05:00:39Z

@copumpkin can you merge this?

copumpkin · 2016-08-18T12:32:33Z

It's going to need updating to account for everything that's changed since I submitted this. I don't have much time right now but might be able to fit it in next week. Would definitely welcome someone else picking it up and pushing it over the line thought :)

On Aug 18, 2016, at 01:00, Sophie Taylor notifications@github.com wrote:

@copumpkin can you merge this?

—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub, or mute the thread.

spacekitteh · 2016-08-24T07:51:26Z

@joachifm this might be helpful for #17949

spacekitteh · 2016-11-04T15:51:11Z

@grahamc kernel, security, enhancement

grahamc · 2016-11-05T00:57:34Z

@spacekitteh Not sure why security, but I added enhancement / kernel. Can you justify security?

spacekitteh · 2016-11-05T01:02:34Z

@grahamc many security-related kernel options currently require a custom config file

grahamc · 2016-11-05T01:05:12Z

Is the difference here a different / improved way to configure it, or is it enabling new functionality not available before? Sorry if I'm being dense, I'm in the middle of a move.

spacekitteh · 2016-11-05T01:10:13Z

Sort of both - it's enabling changing poorly chosen security options within a nix expression

spacekitteh · 2016-11-05T01:12:45Z

It's a basic principle of security that it should be as easy as possible to make secure; this reduces the burden (and thus likelyhood of mistakes which result in less security)

spacekitteh · 2016-11-05T16:20:05Z

@copumpkin what's needed for this to be merged?

NeQuissimus · 2017-10-19T18:44:46Z

ping :) Is this in a reasonable enough state that it just needs the new kernel configs?

danbst · 2017-11-20T23:04:28Z

The helper functions (let bindings in the top) deserve a separate file.

copumpkin · 2017-11-20T23:39:48Z

Realistically, I'm probably not going to finish this anytime soon, but would be happy if someone else picked it up. I just have too much on my plate these days.

teto · 2018-05-29T09:00:35Z

While looking at #41103, I am not fond of the "features" system which seems like an unnecessary abstraction as people/systems fiddling with the kernel will be very specific IMO.

Some questions on how to fit this with the current system:

how to merge with an extraConfig ? there should be some rules like yes wins over no
right now when a setting is set to "m" or "y" it is not overriden by preferBuiltin or autoModules. I would like to have the possibility to ultimately override every ("y" or "m") , i.e., override the "y" as "m" and vice/versa.

I might tackle this since kernel building is still painful, I've started annotating packages I use with some of their required kernel config (bcc/mininet/openvswitch...)

grahamc · 2018-05-29T10:45:23Z

On the flip side, I want to be able to override y and m with n. Graham

…

On May 29, 2018, at 5:01 AM, Matthieu Coudron ***@***.***> wrote: While looking at #41103, I am not fond of the "features" system which seems like an unnecessary abstraction as people/systems fiddling with the kernel will be very specific IMO. Some questions on how to fit this with the current system: how to merge with an extraConfig ? there should be some rules like yes wins over no right now when a setting is set to "m" or "y" it is not overriden by preferBuiltin or autoModules. I would like to have the possibility to ultimately build every ("y" or "m") , i.e., override the "y" as "m" I might tackle this since kernel building is still a lot of pain, I've started annotating packages I use with some of their required kernel config (bcc/mininet/openvswitch...) — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.

teto · 2018-06-20T04:11:16Z

I have updated this PR code and got sthg ready for review #41393.
I would like some feedback as I really would like moving on to packages/modules specifying their kernel dependencies.

Profpatsch · 2018-06-23T18:04:54Z

Thanks @teto, closing in favor of #11994

henrytill · 2018-06-23T18:38:22Z

Thanks @teto, closing in favor of #11994

@Profpatsch: bit of a circular reference here. Did you mean #41393?

Profpatsch · 2018-06-23T19:06:08Z

@henrytill yeah

copumpkin added the 9.needs: reporter feedback label Jan 5, 2016

copumpkin reviewed Jan 5, 2016
View reviewed changes

edolstra reviewed Jan 5, 2016
View reviewed changes

linux: translate config to structured config

ded0031

Note that this isn't used yet. If feedback is positive I'll actually plug it into the real config machinery.

copumpkin force-pushed the structured-linux-config branch from d837a2d to ded0031 Compare January 6, 2016 03:42

copumpkin referenced this pull request Jan 7, 2016

Adding framebuffer console rotation to kernels.

be9ad57

This helps in some weird screens that otherwise show the console 90° turned.

copumpkin mentioned this pull request Oct 3, 2016

Optionalize more kernel features #19194

Closed

7 tasks

cstrahan self-assigned this Oct 4, 2016

grahamc added 0.kind: enhancement 6.topic: kernel labels Nov 5, 2016

grahamc added the 1.severity: security label Nov 5, 2016

spacekitteh mentioned this pull request Nov 5, 2016

There should be an easy method to add extra kernel config in your configuration.nix file. #20187

Open

NeQuissimus mentioned this pull request Aug 7, 2017

linux-common-config: Refactor, clean up #27949

Merged

8 tasks

7c6f434c added the 11.by: nixpkgs-member label Mar 10, 2018

matthewbauer added the 2.status: work-in-progress label Mar 22, 2018

teto mentioned this pull request May 26, 2018

Meta issue about kernel features dependencies #41103

Closed

teto mentioned this pull request Jun 2, 2018

[RFC] linux: translate config to structured config #41393

Closed

8 tasks

Profpatsch mentioned this pull request Jun 23, 2018

Expose Linux kernel configuration in pure Nix #11994

Closed

Profpatsch closed this Jun 23, 2018

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[WIP] linux: translate kernel config to structured config #12158

[WIP] linux: translate kernel config to structured config #12158

copumpkin commented Jan 5, 2016

copumpkin Jan 5, 2016

edolstra commented Jan 5, 2016

edolstra Jan 5, 2016

copumpkin commented Jan 5, 2016

copumpkin commented Jan 6, 2016

copumpkin commented Jan 6, 2016

aszlig commented Jan 21, 2016

rasendubi commented Jul 25, 2016

spacekitteh commented Aug 18, 2016

spacekitteh commented Aug 18, 2016

copumpkin commented Aug 18, 2016

spacekitteh commented Aug 24, 2016

spacekitteh commented Nov 4, 2016

grahamc commented Nov 5, 2016

spacekitteh commented Nov 5, 2016

grahamc commented Nov 5, 2016

spacekitteh commented Nov 5, 2016

spacekitteh commented Nov 5, 2016

spacekitteh commented Nov 5, 2016

NeQuissimus commented Oct 19, 2017

danbst commented Nov 20, 2017

copumpkin commented Nov 20, 2017

teto commented May 29, 2018 •

edited

grahamc commented May 29, 2018 via email

teto commented Jun 20, 2018

Profpatsch commented Jun 23, 2018

henrytill commented Jun 23, 2018

Profpatsch commented Jun 23, 2018

[WIP] linux: translate kernel config to structured config #12158

[WIP] linux: translate kernel config to structured config #12158

Conversation

copumpkin commented Jan 5, 2016

copumpkin Jan 5, 2016

Choose a reason for hiding this comment

edolstra commented Jan 5, 2016

edolstra Jan 5, 2016

Choose a reason for hiding this comment

copumpkin commented Jan 5, 2016

copumpkin commented Jan 6, 2016

copumpkin commented Jan 6, 2016

aszlig commented Jan 21, 2016

rasendubi commented Jul 25, 2016

spacekitteh commented Aug 18, 2016

spacekitteh commented Aug 18, 2016

copumpkin commented Aug 18, 2016

spacekitteh commented Aug 24, 2016

spacekitteh commented Nov 4, 2016

grahamc commented Nov 5, 2016

spacekitteh commented Nov 5, 2016

grahamc commented Nov 5, 2016

spacekitteh commented Nov 5, 2016

spacekitteh commented Nov 5, 2016

spacekitteh commented Nov 5, 2016

NeQuissimus commented Oct 19, 2017

danbst commented Nov 20, 2017

copumpkin commented Nov 20, 2017

teto commented May 29, 2018 • edited

grahamc commented May 29, 2018 via email

teto commented Jun 20, 2018

Profpatsch commented Jun 23, 2018

henrytill commented Jun 23, 2018

Profpatsch commented Jun 23, 2018

teto commented May 29, 2018 •

edited