Asher- Posted September 11, 2019 Share Posted September 11, 2019 (edited) TL;DR: Trying to disable ASPM in BIOS fails to change pci-aspm-default from 0x2 to 0x0. Attempting to enable "PCIe ASPM Not Supported (V4)" in FACP fails to change anything. Presence of FACP.aml in Clover/ACPI/patched results in Mac iASL => File => New From ACPI containing FACP and FACP-1 FACP-1 is always the aml in Clover/ACPI/patched FACP is always a version from Apple, which is not included in Clover's ACPI dump Apple's FACP always seems to take effect, not the FACP from Clover/ACPI/patched Hardware: Asus z10pe-d8 ws 2 x E5-2683v4 (no IGPU) 2 x 64GB LRDIMM 2TB Samsung 860 Pro SSD for System + 4TB 860 EVO SSD for photos 3 x Vega AMD Frontier Edition 16GB (similar results with 1, 2 or 3) Monitor in HDMI of GFX1 Headless monitor in HDMI of GFX2 General access via Remotix over ethernet to Screen Sharing EFI/drivers64UEFI: ApfsDriverLoader-64.efi AptioMemoryFix-64.efi DataHubDxe-64.efi FSInject-64.efi VBoxHfs-64.efi VirtualSmc.efi Kexts (also in /L/E): Lilu.kext SMCProcessor.kext SMCSuperIO.kext VirtualSMC.kext I've also tried Whatevergreen.kext with various configurations, but seen no benefit along with possible unexpected changes so do not currently have it included. Details: Freezes related to Vega Frontier Edition (Windowserver crashes, DumpGPURestart, SSH still works) have led me to suspect ASPM as the culprit, particularly as the symptoms are very similar to HD related crashes due to ALPM (device suddenly disappears due to power management, OS doesn't know what happened and freaks out). IOReg reports pci-aspm-default as 0x2 for each ACPI device below PCI0/1 down to GFX0/GFX1/GFX2. Overriding pci-aspm-default via SSDT injection has not worked but has led to noticing that the injection of other SSDT-specified properties is also not working for GFX#, even though identical aml in the same file is working for HDAU (ACPI from EFI attached, ioreg link below). Overriding pci-aspm-default via Clover property injection appears to work using path from gfxutil -f gfx0 (although it sets the property type to Data rather than Number; I'm not sure if that matters). The same for gfxutil -f gfx1 does not work (no change in property value viewed via IORegistryExplorer). This error is consistent with prior attempt to use Clover property injection to specify power tables for the Vega. It appears that the reason I have not been able to enable "PCIe ASPM Not Supported (V4)" in FACP is the same reason that BIOS settings to disable ASPM are not working— Apple's FACP is somehow getting inserted in place of the Asus FACP that is configured as expected. It is not clear to me why other systems do not seem to encounter this problem disabling ASPM. Question: Can anyone assist with any of the following? Understanding how the Apple FACP relates and how to override its settings Making the property injection via _DSM work for GFX# in my SSDT, and particularly injection of the pci-aspm-default property as 0x0 Syntax for specifying device-property values explicitly in a newly created Device block rather than via return in _DSM method Understanding the relation between device-properties as shown in IOService / IODeviceTree vs. as shown in IOACPIPlane ACPI and config.plist from EFI is attached as currently configured for 2 GPUs (3rd GPU not currently connected). IOReg too big to attach, available via Dropbox: https://www.dropbox.com/s/fn1uxajkpbx23cb/One.ioreg?dl=0 Thanks! Asher config.plist efi-acpi.tar.gz Edited September 11, 2019 by Asher- Link to comment Share on other sites More sharing options...
Asher- Posted September 11, 2019 Author Share Posted September 11, 2019 IOPCIFamily.kext patch is currently in my config.plist due to testing. After further research with Hopper, my understanding is that the patch just suppresses a probe error and continues anyway. Since I'm not having any errors without it, my intuition is that it should be removed. Thoughts appreciated. Link to comment Share on other sites More sharing options...
Asher- Posted September 12, 2019 Author Share Posted September 12, 2019 Apparently _DSM is standard Apple syntax. Helpful explanation https://patchwork.kernel.org/patch/9802489/. Link to comment Share on other sites More sharing options...
Asher- Posted September 16, 2019 Author Share Posted September 16, 2019 Apparently if package size does not match actual package arg count, the whole thing is thrown out. I believe this is why my property injection was failing. Link to comment Share on other sites More sharing options...
Recommended Posts