Re: [PATCH] PCI/ERR: Fix run error recovery callbacks for all affected devices

[Sinan Kaya <okaya@kernel.org>]

On 1/28/2019 10:47 AM, Sinan Kaya wrote:
>> also have different PFs (device numbers are different) under the same bus.
>> This case do not need to brodcast all the devices under the same bus.
> 
> Even if OS was to call pcie_do_recovery() for other devices, nothing should
> happen because the expectation for other devices' AER status register to report
> no errors. This is the case we want you to validate. If this is not true, you
> are looking at a firmware/HW bug.

Slight clarification about the statement above. Based on quick code scan,
pcie_do_recovery() should only be called if there is an outstanding AER error
for the first AER device in non-FF (non-firmware-first) scenario.

Whereas it gets called per AER device on FF scenario and this is where the
conflict is.

I think your argument is that "if there is a non-fatal error on the parent,
should it be broadcasted to all children devices in pcie_do_recovery()?"

IMO, I think the answer is yes only if the children's AER status reports a
problem and I don't see this check in the code.

Code needs to be refactored to follow a similar pattern to FF scenario and
pci_walk_bus() calls should be removed from the NON-FATAL path.