Ventura Boot Issues with certain GPU/Firmware combinations

[khronokernel]

An odd issue that we've seen pop up on Ventura is certain GPUs having boot issues depending on the hardware combination. Specifically stalling around PCI configuration begin stage.

We've seen this happen on the following units:

• QEMU 6.2 VM
  • Skylake-X, Nvidia GT710 (2GB HP and 1GB Zotac models tested)
  • Booting bare-metal with GT710 works
• Genuine 2013 Mac Pro (MacPro6,1)
  • Ivy Bridge-EP, Dual AMD D300
• HP Envy 15-u011dx
  • Haswell U, Intel HD 4400
  • CC. @brian105
• Genuine 2008 iMac 20"
  • Penryn, ATI HD 2600

---

When booting a normal configuration, we see the error show up extremely early:

QEMU VM	Genuine 2013 Mac Pro (MacPro6,1)

However on the QEMU VM, booting with serial=5 actually boots much further up to graphics init stage. Once arriving, we see a repeated newUserClientFailed 0x... 0x... IONVDRFramebuffer error:

---

Model	GPU	Output Protocol	Is Functional
X299 (QEMU VM)	Nvidia GT710 (2GB HP and 1GB Zotak)	GOP	NO
X299 (QEMU VM)	AMD RX 470	GOP	YES
X299 (Bare Metal)	Nvidia GT710 (2GB HP and 1GB Zotak)	GOP	YES
HP Envy	Intel HD4600	GOP	NO
2013 Mac Pro	AMD D300	GOP	NO
2008 iMac	ATI HD 2600	UGA	NO
2008 Mac Pro	Nvidia GT710 (2GB HP and 1GB Zotak)	UGA (converted from GOP)	YES

• X299 QEMU VM was tested with GOP VBIOS for all runs
  • Neither GT710 models functional in VM
  • GT710s functional bare-metal
  • AMD RX470 functional in VM
• MacPro3,1 is a UGA-only machines
  • When testing the GT710 models, used OpenCorePkg to translate from GOP to UGA
• iMac8,1 is a UGA-only system, the ATI HD 2600 is a stock unit with Apple's UGA VBIOS.

---

Additional Information:

• Currently no known work-arounds for these systems.
• Both installer and installed OS replicate issue.
• OSes Tested:
  • Oldest: 13.0 Beta 1 (22A5266r)
  • Newest: 13.0 Beta 8 (22A5352e)
• Booting without any kexts injected still replicates

[khronokernel]

For pre-XCPM machines, we found that restoring Power Management was a possible avenue to resolving IONDRVFramebuffer erroring in Ventura. Specifically the addition of modded AppleIntelCPUPowerManagement binaries (Source: build.py):

• AppleIntelCPUPowerManagement-v1.0.0.zip

• AppleIntelCPUPowerManagementClient-v1.0.0.zip

This resolved boot support on QEMU with GT710s installed as well as with iMac8,1, however Verbose Boot is still broken outright after PCI Configuration begin.

---

Additionally, the 2013 Mac Pro still fails to boot even with power management restored. No known solution available.

[ofawx]

Seeing identical issue on:

Model	GPU	Is Functional
X99 Broadwell-EP (QEMU 5.1 VM)	AMD RX 6600 XT	NO
X99 Broadwell-EP (QEMU 6.2 VM)	AMD RX 6600 XT	NO

• OC 0.8.5
• macOS build 22A380
• VMM and AVX2 present

Not yet tested:

• RX 580 8GB
• Intel Power Mgmt as above
• serial=5

[Trunkzeh]

@khronokernel I have a MacPro6,1 machine with AMD D300 graphics; keen to help out where possible in order to resolve the issue. Currently running on monterey; so will prepare a ventura install on an external drive and report findings.

[googlegot]

I have a MacPro6,1 running Ventura with AMD D500 graphics and I am encountering random boot issues that seem to be caused by this. I have a background in driver validation and debugging boot issues. I'll start digging in here to see if I can be of assistance, but if there's anything in particular you'd like tested, please don't hesitate to ask.

[bland328]

I'm also seeing this problem with:

Model	GPU	Output Protocol	Is Functional
H87 Haswell (QEMU 7.1 VM)	AMD RX 560D (2GB OEM)	GOP	NO

Removing -v boot arg allows macOS VM to boot, instead of hanging at or near PCI configuration begin.

Alternately, switching to vnc/virtio virtual GPU also allows macOS VM to boot.

I'm happy to test any configuration changes that might provide additional information.

[ExtremeXT]

I'm not 100% sure if it's related but I think it is, I saw this issue on the bugtracker and remembered that I've come across this as well on my KVM VM, it would always stall at/around [PCI Configuration Begin] with my GPU passed-through (and it got to GPU initialization in serial as well), but would boot fine with a virtual GPU (albeit only VirtIO and VNC would work, others like VGA would get stuck as well), while Monterey worked fine. I fixed this by importing a patch related to PCI Hotplugging by CaseySJ from AMD OS X: https://github.com/AMD-OSX/AMD_Vanilla/blob/master/patches.plist#L520-L548 which fixed it, and now it boots properly. My GPU is an AMD RX 6650 XT if it matters.