kernel: enable PHY_ROCKCHIP_PCIE and PCIE_ROCKCHIP_HOST

[jakubgs]

Motivation for this change

Fixes failure to detect NVMe SSD in Rockchip devices like NanoPC-T4.

More details can be seen in:
NixOS/nixos-hardware#259
#111034

Things done

• Tested using sandboxing (nix.useSandbox on NixOS, or option sandbox in nix.conf on non-NixOS linux)
• Built on platform(s)
  • NixOS
  • macOS
  • other Linux distributions
• Tested via one or more NixOS test(s) if existing and applicable for the change (look inside nixos/tests)
• Tested compilation of all pkgs that depend on this change using nix-shell -p nixpkgs-review --run "nixpkgs-review wip"
• Tested execution of all binary files (usually in ./result/bin/)
• 21.11 Release Notes
  • (Package updates) Added a release notes entry if the change is major or breaking
  • (Module updates) Added a release notes entry if the change is significant
  • (Module addition) Added a release notes entry if adding a new NixOS module
• Fits CONTRIBUTING.md.

[samueldr]

Enabling it outright to yes for all users is not a solution. Especially if there are issues that this is working around; maybe that will cause more issues elsewhere on probe. (Though unlikely)

[samueldr]

I was made aware of this recently:

• https://patchwork.kernel.org/project/linux-rockchip/patch/[email protected]/

Try applying this patch and still use the modules. It's unclear to me whether it would help the specific issue you are facing or not.

[jakubgs]

Wow, I have not seen that. That's VERY fresh. It would fit with the timeout errors I was getting when I tried putting them initrd:

[    3.677351] rockchip-pcie f8000000.pcie: PCIe link training gen1 timeout!

It targets the same module, but it's not exactly the same part of the code that causes the timeout:
https://github.com/torvalds/linux/blob/v5.11/drivers/pci/controller/pcie-rockchip-host.c#L332-L339

I will try to use the patch with modules added to the initrd image and report back. Thanks for pointing it out!

[samueldr]

Yes, apparently it was a "known issue" for RK3399 and PCIe.

I lament the fact there is no centralized location for listing known issues.

[samueldr]

(Converted to draft to ensure no accidental merges happen... Let's see where this goes!)

[jakubgs]

Nope, I tired the following configuration:

{
  boot.kernelPackages = pkgs.linuxPackages_5_11;
  boot.kernelPatches = [{
    name = "v2-PCI-rockchip-Avoid-accessing-PCIe-registers-with-clocks-gated.diff";
    patch = pkgs.fetchurl {
      url = "https://patchwork.kernel.org/project/linux-rockchip/patch/[email protected]/raw/";
      sha256 = "0057gmjj5glbkabyi5jgnd8r9dhnsb7mkc6nbyxnv6a0rh69yx4n";
    };
  }];
  boot.initrd.kernelModules = [ "pcie-rockchip-host" "phy-rockchip-pcie" ];
}

And it fails to boot, and I see the same error in kernel logs:

[    3.694422] rockchip-pcie f8000000.pcie: PCIe link training gen1 timeout!

So no, it doesn't seem like that patch fixes it.

[jakubgs]

I've also tried the following patch:
https://patchwork.ozlabs.org/project/linux-pci/patch/[email protected]/

{
  boot.kernelPackages = pkgs.linuxPackages_5_11;
  boot.kernelPatches = [{
    name = "PCI-rockchip-Fix-timeout-in-rockchip_pcie_host_init_port.diff";
    patch = pkgs.fetchurl {
      url = "https://patchwork.ozlabs.org/project/linux-pci/patch/[email protected]/raw/";
      sha256 = "0qhggf4ikdcx36lrnlri46gmhj7rvx6j88c7jzii1pb2daxg77cy";
    };
  }];
  boot.initrd.kernelModules = [ "pcie-rockchip-host" "phy-rockchip-pcie" ];
}

Which actually touches the exact timeout for which kernel log message I've seen, but it doesn't help, which is quite confusing.

[stale]

I marked this as stale due to inactivity. → More info

[jakubgs]

Yeah, I don't think enabling that globally makes sense. If I ever find a patch that fixes this I'll open a new PR.