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 2/24/2026 11:21 PM, Matthew Ruffell wrote:
> Hi Mario,
> 
>> Matthew,
>>
>> Can you please try with this diff that applies my idea?
>>
>> Thanks,
>>
>> diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
>> index 1f6fd7c372932..5955a273c9ba7 100644
>> --- a/drivers/pci/pci.c
>> +++ b/drivers/pci/pci.c
>> @@ -3196,6 +3196,7 @@ void pci_pm_init(struct pci_dev *dev)
>>           }
>>
>>    poweron:
>> +       pci_clear_master(dev);
>>           pci_pm_power_up_and_verify_state(dev);
>>           pm_runtime_forbid(&dev->dev);
> 
> I applied this ontop of current head 7dff99b from 7.0-rc1.
> 
> c5.metal on AWS did not come back from a clean reboot. Looking at the serial
> console shows the exact same behaviour as attempting a kexec. It gets stuck
> trying to locate the rootfs, because the NVMe drive has no bus master bit set.
> 
> That is also maybe not a viable path to take.

OK thanks for checking it.

Could you figure out at which point it hangs?  What confuses me is why 
doesn't the call for pci_set_master() in nvme_pci_enable() help?


> 
> I'm currently working on some instrumentation, will try get my results,
> including a full dmesg capture, soon.
> 
> Thanks,
> Matthew

Could you specifically decorate the calls to nvme_pci_enable() to see if 
the hang is before or after that call to pci_set_master()?