reverting c37e627f9565368ed7bd1f3cf59a2d223ddba85a

reverting c37e627f9565368ed7bd1f3cf59a2d223ddba85a

[Prasad Koya]

hi Frederick

At Arista Networks, we have a service driver for AER as a loadable
module. The reasons for having Arista's own module and not enabling
in-built aer driver (CONFIG_PCIEAER) is for the reasons below:

1. We need this driver on a few of our 40+ products we sell at this
time. So we want to keep this as a module and load it only on hardware
that supports AER.

2. Reacting to the type of PCI error is handled by a user space
process, not a kernel thread. So we can change the level of logging,
how to react to certain errors, how much info to log etc in user
space.

We are upgrading our kernel software to 4.19 and we noticed that
pcieport_if.h is no longer in linux/ dir and that breaks compilation
of our AER kernel module. From the comment in commit, we are inferring
the reason for renaming this file to pci/pcie/pcieport_if.h is because
no one but code under PCI core is registering PCIe service drivers. If
you or the PCI subsystem maintainers have no objection, we'd like to
revert commit c37e627f9565368ed7bd1f3cf59a2d223ddba85a. Please let us
know your thoughts.

Thank you
Prasad

Re: reverting c37e627f9565368ed7bd1f3cf59a2d223ddba85a

[Christoph Hellwig]

On Mon, Aug 19, 2019 at 10:15:08PM -0700, Prasad Koya wrote:
> hi Frederick
> 
> At Arista Networks, we have a service driver for AER as a loadable
> module. The reasons for having Arista's own module and not enabling
> in-built aer driver (CONFIG_PCIEAER) is for the reasons below:

And as long as that isn't upstream no one is going to care for it,
and most certainly not reverting useful kernel improvements for it.