Announcement

Collapse
No announcement yet.

Radeon RX 6600 Linux Performance Rising Even Higher With Newest Open-Source Driver

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #21
    Originally posted by chrisr View Post
    I have recently bought a Radeon RX6600 card, and it works fine with a stock Fedora kernel (5.15.4-201.fc35.x86_64), but I am having a lot of problems configuring my own kernel to use it! I can only assume that the card has some non-obvious dependencies, possibly for the PCIE bus or memory management? For reference, i am currently using:
    Code:
    CONFIG_DRM_AMDGPU=m
    CONFIG_DRM_AMDGPU_CIK=y
    CONFIG_DRM_AMDGPU_USERPTR=y
    CONFIG_DRM_AMD_DC=y
    CONFIG_DRM_AMD_DC_DCN=y
    CONFIG_DRM_AMD_DC_HDCP=y
    CONFIG_DRM_AMD_SECURE_DISPLAY=y
    CONFIG_HSA_AMD=y
    CONFIG_HSA_AMD_SVM=y
    CONFIG_DRM_AMD_ACP=y
    and I have also added:
    Code:
    CONFIG_AMD_IOMMU=y
    CONFIG_AMD_IOMMU_V2=m
    which provides me with the same kernel modules as the Fedora kernel. However, the GNOME desktop fails during login when using my own kernel, and the dmesg log fills with errors like:
    Code:
    Nov 28 12:32:38 endgame kernel: [drm] Fence fallback timer expired on ring gfx_0.0.0
    Nov 28 12:32:38 endgame kernel: [drm] Fence fallback timer expired on ring sdma1
    Nov 28 12:32:39 endgame kernel: [drm] Fence fallback timer expired on ring sdma0
    The fact that Fedora's kernel works OK must mean that my hardware is fine. Does anyone know of any kernel config options I may have missed please?

    Sounds like potentially a problem with interrupts. Maybe some problems with MSI or MSI-X support in your kernel?


    Comment


    • #22
      Originally posted by agd5f View Post

      Sounds like potentially a problem with interrupts. Maybe some problems with MSI or MSI-X support in your kernel?
      Thanks for the suggestion, I will compare all of the MSI-related CONFIG settings. I also intend to try enabling
      Code:
      CONFIG_INTEL_IOMMU_SVM=y
      to see if that makes a difference too. My motherboard supports PCIE 2.0, but the RX-6600 card also includes these devices:
      Code:
      1478 Navi 10 XL Upstream Port of PCI Express Switch
      1479 Navi 10 XL Downstream Port of PCI Express Switch
      which suggests my kernel needs PCIE functionality beyond what's on my motherboard.

      Comment


      • #23
        Originally posted by chrisr View Post
        Thanks for the suggestion, I will compare all of the MSI-related CONFIG settings. I also intend to try enabling
        Code:
        CONFIG_INTEL_IOMMU_SVM=y
        to see if that makes a difference too. My motherboard supports PCIE 2.0, but the RX-6600 card also includes these devices:
        Code:
        1478 Navi 10 XL Upstream Port of PCI Express Switch
        1479 Navi 10 XL Downstream Port of PCI Express Switch
        which suggests my kernel needs PCIE functionality beyond what's on my motherboard.
        Those are just PCI bridges built into the GPU. They shouldn't require any special configuration.

        Comment


        • #24
          Originally posted by agd5f View Post
          Those are just PCI bridges built into the GPU. They shouldn't require any special configuration.
          OK, thanks.

          FWIW, enabling
          Code:
          CONFIG_ARCH_ENABLE_MEMORY_HOTPLUG=y
          CONFIG_MEMORY_HOTPLUG=y
          CONFIG_MEMORY_HOTPLUG_SPARSE=y
          CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE=y
          CONFIG_ARCH_ENABLE_MEMORY_HOTREMOVE=y
          CONFIG_MEMORY_HOTREMOVE=y
          
          CONFIG_PCI_P2PDMA=y
          CONFIG_DEVICE_PRIVATE=y
          seem to allow GNOME to login now. The "Fence fallback timer expired" messages still happen when I launch Wine and Battle.net though.
          Last edited by chrisr; 29 November 2021, 03:16 PM.

          Comment


          • #25
            Originally posted by agd5f View Post
            Those are just PCI bridges built into the GPU. They shouldn't require any special configuration.
            Sadly, enabling CONFIG_INTEL_IOMMU_SVM did not make any difference. Nor can I locate any MSI-related options wheich are different between my and Fedora's kernels (except for CONFIG_MSI_WMI, but that's the wrong kind of MSI )

            Curious - Fedora has loaded these modules:
            Code:
            rpcrdma 331776 2
            rdma_cm 131072 1 rpcrdma
            iw_cm 57344 1 rdma_cm
            ib_cm 143360 1 rdma_cm
            ib_core 417792 4 rdma_cm,rpcrdma,iw_cm,ib_cm
            I don't recognise these. Could they be relevant please?
            Last edited by chrisr; 30 November 2021, 09:13 AM.

            Comment


            • #26
              If a stock RH kernel works, just use the config from that kernel.

              Comment


              • #27
                Originally posted by chrisr View Post
                Curious - Fedora has loaded these modules:
                Code:
                rpcrdma 331776 2
                rdma_cm 131072 1 rpcrdma
                iw_cm 57344 1 rdma_cm
                ib_cm 143360 1 rdma_cm
                ib_core 417792 4 rdma_cm,rpcrdma,iw_cm,ib_cm
                I don't recognise these. Could they be relevant please?
                Not relevant to graphics.

                Comment


                • #28
                  Originally posted by agd5f View Post
                  Not relevant to graphics.
                  Still too early to cry "Success!", but I have just managed to run World of Warcraft using a home-compiled 5.15.5 kernel. I think the key change was replacing
                  Code:
                  CONFIG_PCIEASPM_POWERSAVE=y
                  with
                  Code:
                  CONFIG_PCIEASPM_DEFAULT=y
                  although I cannot explain why.

                  Comment


                  • #29
                    Originally posted by chrisr View Post
                    Still too early to cry "Success!", but I have just managed to run World of Warcraft using a home-compiled 5.15.5 kernel. I think the key change was replacing
                    Code:
                    CONFIG_PCIEASPM_POWERSAVE=y
                    with
                    Code:
                    CONFIG_PCIEASPM_DEFAULT=y
                    although I cannot explain why.
                    APSM is known to be problematic on some platforms. Forcing it on can cause problems, while using the default settings set up by the bios is safe.

                    Comment


                    • #30
                      Originally posted by agd5f View Post
                      APSM is known to be problematic on some platforms. Forcing it on can cause problems, while using the default settings set up by the bios is safe.
                      True, but I've been enabling it on this motherboard for years with two earlier graphics cards without any problems, and so I'm inclined to think there is more to this than just ASPM.

                      I did notice that with ASPM set to "powersave", the screen would dim briefly once or twice while GNOME was logging in, as if the card were fading it out during a period of inactivity. Could an ASPM "event" interrupt something that shouldn't be interrupted?

                      In fact, ASPM is still enabled because my kernel command line includes these options:
                      Code:
                      pcie_aspm=force amdgpu.aspm=1
                      producing these messages in my dmesg log:
                      Code:
                      [ 0.055387] PCIe ASPM is forcibly enabled
                      
                      [ 1.028994] acpi PNP0A03:00: _OSC: OS supports [ExtendedConfig ASPM ClockPM Segments MSI EDR HPX-Type3]
                      
                      [ 1.171047] pci 0000:00:1c.2: ASPM: current common clock configuration is inconsistent, reconfiguring
                      
                      [ 1.185093] pci 0000:00:1c.4: ASPM: current common clock configuration is inconsistent, reconfiguring
                      Last edited by chrisr; 02 December 2021, 05:35 PM.

                      Comment

                      Working...
                      X