Losing Headless iGPU after sleep/display sleep (Intel HD 630 + Yeston Radeon RX550 (Polaris) )

[drmonkeyrph]

Hi All,

I am hoping I can find some help diagnosing a particularly tricky issue I am experiencing with my Hackintosh Desktop. I posted something similar to Reddit a few weeks ago without much success.  The one suggestion I received previously was to add a SSDT-SBUS-MCHC

What I am seeing is that my iGPU (HD630/Kaby Lake) is working fine, in that I am running it headless alongside my Polaris card, and Intel Power Gadget shows it getting utilized on occasion when I play an h264 movie or try some other tasks that would typically trigger QuickSync to be used.But when I either sleep the machine, or the display goes to sleep for any sort of extended period, the iGPU stops functioning properly. There are a few pieces of evidence that point to this.

1. Right after boot, VDADecoderChecker returns quickly and reports that decoding is supported. After I start experiencing the issue, VDADecoderChecker will hang until I hit Ctrl+C to kill it.

2. After sleep, I begin to experience issues with any application that uses GPU assisted rendering. Microsoft Teams, for example, is unable to be opened to anything more than a white screen after I start to experience this issue. Safari also has a difficult time loading more complicated pages.

3. Even Hackintool appears to hang while trying to read the system configuration after this issue starts, and has to be killed..

4. Intel Power Gadget shows that Prior to sleep the iGPU is being utilized(GFX AVG>0), and that the OS is requesting usage of it(GFX REQ around 0.3). After sleep, the GFX AVG is locked to 0 and the GFX REQ is now stuck to 1.10 and doesn't move. This looks like the OS still thinks the GPU should be there and is requesting cycles, but not getting them.

 

I currently have the AAPL,ig-platform-id of the iGPU set to 0x59120003 (03 00 12 59 after conversion) as seen below.

I've been through every console log I could find, but the only thing close to looking like the issue is this:

Aug  6 12:18:39 EliteDesk VTDecoderXPCService[1396]: getattrlist failed for /Library/GPUBundles/AppleIntelKBLGraphicsVADriver.bundle/Contents/MacOS/AppleIntelKBLGraphicsVADriver: #2: No such file or directory
Aug  6 12:18:39 EliteDesk VTDecoderXPCService[1396]: getattrlist failed for /Library/GPUBundles/AMDRadeonVADriver2.bundle/Contents/MacOS/AMDRadeonVADriver2: #2: No such file or directory
Aug  6 12:18:39 EliteDesk VTDecoderXPCService[1396]: getattrlist failed for /System/Library/Extensions/AMDRadeonVADriver2.bundle/Contents/MacOS/AMDRadeonVADriver2: #2: No such file or directory

Does anyone have any ideas I could use to either solve this, or debugging tools I could use to try to find an actual error message?

Thanks in Advance!

EDIT: Adding IOReg Dumps pre and post sleep

ioreg post sleep

ioreg pre sleep

 

Specs:

EliteDesk 800 G3 SFF

Processor: Intel Core i5-7500(Kaby Lake)

RAM: 32GB

Graphics: Intel HD630 (Headless Mode)

Graphics: Yeston Radeon RX550 4G LP(Polaris-based, the Hot Pink Card)

Ethernet:Intel I219-LM

Audio: Conexant CX20632

Power Supply: HP L08417-004 Power Supply (Upgraded from 180W by swapping in PSU from EliteDesk 800 G4 SFF. Fits perfectly.

[MaLd0n]

Post one Full Dump with Ioreg.
https://www.olarila.com/topic/10301-olarila-runme-dumper/
https://olarila.com/files/Utils/RunMe.app.zip
https://olarila.com/files/Utils/IORegistryExplorer.app.zip

[drmonkeyrph]

Ive uploaded the output of RunMe to Github at this link

there are also pre and post sleep ioreg files I gathered previously located below

ioreg post sleep

ioreg pre sleep

 

I really appreciate someone taking a look.  Even if nobody knows the answer,  I would appreciate any debugging tips on how to trace this one down, so I can work on it further myself.

[MaLd0n]

not a good folder

many acpi renames

very old bootloader and kexts version

 

[drmonkeyrph]

I will clear out the disabled renames.  The EC to EC0 rename is required for boot after 4+ hours of trying without it. I also attempted to upgrade Kexts and open core to 0.72 today, but I was left without Metal support on my GPU so I had to revert. 
 

While I try to clean that up a bit, do you have a way or a tool to do a diff on 2 different IOreg

files?  

[MaLd0n]

u can use ec0 to ec but others no

[drmonkeyrph]

OK, lets try this again.  I have cleaned out all patches except the EC0 rename, and upgraded OpenCore to the latest (0.72) as well as upgrading all kexts to their latest versions except WhateverGreen, because the latest WhateverGreen causes me to lose support for Metal.  I have your tool and provided the results below.  

Send Me EliteDesk-new.zip

After this cleanup, I retested my issue and the behavior has not changed, but I didn't really expect it to. 

Any thoughts on what to try next, or debugging ideas, or tools to compare multiple IOReg files would be appreciated.  I have worked in software for 17+ years so I am open to any more technical debugging steps or techniques you may have.

[MaLd0n]

test this efi folder

https://easyupload.io/1gfthi

[drmonkeyrph]

I tested your modified folder today.  Some observations:

1. It seems to have broken hardware acceleration. I see no usage in IntelPowerGadget where I did before and I now get the following error out of VDADecoderChecker

EliteDesk:Downloads dad$ ./VDADecoderChecker

VDADecoderCreate failed. err: -12473

An error was returned by the decoder layer. This may happen for example because of bitstream/data errors during a decode operation. This error may also be returned from VDADecoderCreate when hardware decoder resources are available on the system but currently in use by another process.

VDADecoderCreate failed. err: -12473

After sleep, I still see the same error, but I see the same behavior in Intel Power Gadget as I did before where the Req goes to 1.10.

2. Using the latest WhateverGreen you included breaks metal support on my GPU.  I have tested other versions and while 1.5.1. works, 1.5.2 causes my entire screen to have the colors way off and unable to rotate on my second monitor. I reverted WEG to 1.5.1 in your provided EFI so I could continue testing.

3. I see you removed all the SSDT files I manually created and replaced them with a single DSDT.aml.  Is there a performance/management benefit to this approach? both GeekBench and Heaven benchmark show no performance difference. Also, was something changed in the DSDT to allow the removal of NVMeFix.kext?

4. I see you included a new fix called RestrictEvents.kext.  Was there something specific you saw in the files provided that Indicated that I needed that file?

[MaLd0n]

in this case u need use imac smbios for hardware accel. rx550 dont provide it, we need a intel quick sync

u can remove RestrictEvents with imac smbios

here we work with full solution, our hack like one mac, removed devices, inject apple devices, etc. ssdt is pseudo solution. not good, but many people like...is easy...ctrl c + ctrl v

[drmonkeyrph]

Thanks for the reply. 
Regarding DSDT vs SSDT, I’m good either way, I was more curious than anything. I’ve done a bit of DSDT work in the past to undervolt an old dell Hackintosh so I’m reasonably comfortable in iasl.

I’m currently running iMac18,2 as it matches my processor exactly and supports a dGPU. With that, I get Intel QSV before sleep but not after. 18,3 gets the same results.

I’m guessing that I should probably be looking in DSDT for a _WAK event for the root cause, do you have any thoughts where I should start looking?