Re: [PROBLEM] c5.metal on AWS fails to kexec after "PCI: Explicitly put devices into D0 when initializing"

[Mario Limonciello <mario.limonciello@amd.com>]

On 12/4/2025 9:10 PM, Matthew Ruffell wrote:
> Sorry accidentally sent the message.
> 
> The nvme was still in state 0 / PCI_D0:
> 
> [  109.801025] mruffell: vendor: 1d0f, device: 61, state: 0
> [  109.819542] nvme 0000:90:00.0: mruffell: Current PCI device.
> 
> /sys/bus/pci/devices$ ll
> lrwxrwxrwx 1 root root 0 Dec  4 23:24 0000:90:00.0@ ->
> ../../../devices/pci0000:7a/0000:7a:02.0/0000:8d:00.0/0000:8e:01.0/0000:90:00.0
> 
> All of these devices are also state 0. Interesting.
> 
>>> I have a relatively ignorant question.  Can you reproduce with kdump and
>>> a crash too?
>>>
>>> I don't actually know if you configure kdump and then crash the kernel
>>> (say magic sys-rq key), does pci_device_shutdown() get called in order
>>> to do the kexec?  Or because the kernel is already in a crash state is
>>> there just a jump into the crash kernel image location?
>>
> 
> I did check this. I triggered a crash with magic sysrq, and
> pci_device_shutdown()
> was never called. It never printed out my debug messages from
> pci_device_shutdown(), instead it just oopsed and booted straight to the crash
> kernel.
> 
> Thanks,
> Matthew

OK so to me we have two options that you proved both work.

1) Call pci_set_master() during startup.
2) Drop pci_clear_master() for the kexec case during shutdown.

I think we need comments from Bjorn here on which direction is safer 
generally speaking.