amdgpu didn't start with pci=nocrs parameter, get error "Fatal error during GPU init"

[Christian König]

Am 24.02.23 um 13:29 schrieb Christian König:
> Am 24.02.23 um 09:38 schrieb Mikhail Gavrilov:
>> On Fri, Feb 24, 2023 at 12:13 PM Christian König
>> <ckoenig.leichtzumerken at gmail.com> wrote:
>>> Hi Mikhail,
>>>
>>> this is pretty clearly a problem with the system and/or it's BIOS and
>>> not the GPU hw or the driver.
>>>
>>> The option pci=nocrs makes the kernel ignore additional resource 
>>> windows
>>> the BIOS reports through ACPI. This then most likely leads to problems
>>> with amdgpu because it can't bring up its PCIe resources any more.
>>>
>>> The output of "sudo lspci -vvvv -s $BUSID_OF_AMDGPU" might help
>>> understand the problem
>> I attach both lspci for pci=nocrs and without pci=nocrs.
>>
>> The differences for Cezanne Radeon Vega Series:
>> with pci=nocrs:
>> Control: I/O- Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr-
>> Stepping- SERR- FastB2B- DisINTx-
>> Interrupt: pin A routed to IRQ 255
>> Region 4: I/O ports at e000 [disabled] [size=256]
>> Capabilities: [c0] MSI-X: Enable- Count=4 Masked-
>>
>> Without pci=nocrs:
>> Control: I/O+ Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr-
>> Stepping- SERR- FastB2B- DisINTx+
>> Interrupt: pin A routed to IRQ 44
>> Region 4: I/O ports at e000 [size=256]
>> Capabilities: [c0] MSI-X: Enable+ Count=4 Masked-
>>
>>
>> The differences for Navi 22 Radeon 6800M:
>> with pci=nocrs:
>> Control: I/O- Mem- BusMaster- SpecCycle- MemWINV- VGASnoop- ParErr-
>> Stepping- SERR- FastB2B- DisINTx-
>> Interrupt: pin A routed to IRQ 255
>> Region 0: Memory at f800000000 (64-bit, prefetchable) [disabled] 
>> [size=16G]
>> Region 2: Memory at fc00000000 (64-bit, prefetchable) [disabled] 
>> [size=256M]
>> Region 5: Memory at fca00000 (32-bit, non-prefetchable) [disabled] 
>> [size=1M]
>
> Well that explains it. When the PCI subsystem has to disable the BARs 
> of the GPU we can't access it any more.
>
> The only thing we could do is to make sure that the driver at least 
> fails gracefully.
>
> Do you still have network access to the box when amdgpu fails to load 
> and could grab whatevery is in dmesg?

Sorry I totally missed that you attached the full dmesg to your original 
mail.

Yeah, the driver did fail gracefully. But then X doesn't come up and 
then gdm just dies.

Sorry there is really nothing we can do here, maybe ping somebody with 
more ACPI background for help.

Regards,
Christian.

>
> Thanks,
> Christian.
>
>> AtomicOpsCtl: ReqEn-
>> Capabilities: [a0] MSI: Enable- Count=1/1 Maskable- 64bit+
>> Address: 0000000000000000  Data: 0000
>>
>> Without pci=nocrs:
>> Control: I/O- Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr-
>> Stepping- SERR- FastB2B- DisINTx+
>> Latency: 0, Cache Line Size: 64 bytes
>> Interrupt: pin A routed to IRQ 103
>> Region 0: Memory at f800000000 (64-bit, prefetchable) [size=16G]
>> Region 2: Memory at fc00000000 (64-bit, prefetchable) [size=256M]
>> Region 5: Memory at fca00000 (32-bit, non-prefetchable) [size=1M]
>> AtomicOpsCtl: ReqEn+
>> Capabilities: [a0] MSI: Enable+ Count=1/1 Maskable- 64bit+
>> Address: 00000000fee00000  Data: 0000
>>
>>> but I strongly suggest to try a BIOS update first.
>> This is the first thing that was done. And I am afraid no more BIOS 
>> updates.
>> https://rog.asus.com/laptops/rog-strix/2021-rog-strix-g15-advantage-edition-series/helpdesk_bios/ 
>>
>>
>> I also have experience in dealing with manufacturers' tech support.
>> Usually it ends with "we do not provide drivers for Linux".
>>
>