Re: [PATCH 1/1] PCI/AER: prevent pcie_do_fatal_recovery from using device after it is removed

[Benjamin Herrenschmidt <benh@au1.ibm.com>]

On Tue, 2018-08-14 at 14:52 +0530, poza@codeaurora.org wrote:
> > >       if (result == PCI_ERS_RESULT_RECOVERED) {
> > >               if (pcie_wait_for_link(udev, true))
> > >                       pci_rescan_bus(udev->bus);
> > > -            pci_info(dev, "Device recovery from fatal error successful\n");
> > > +            /* find the pci_dev after rescanning the bus */
> > > +            dev = pci_get_domain_bus_and_slot(domain, bus, devfn);
> > 
> > one of the motivations was to remove and re-enumerate rather then
> > going thorugh driver's recovery sequence
> > was; it might be possible that hotplug capable bridge, the device
> > might have changed.
> > hence this check will fail

Under what circumstances do you actually "unplug" the device ? We are
trying to cleanup/fix some of the PowerPC EEH code which is in a way
similar to AER, and we found that this unplug/replug, which we do if
the driver doesn't have recovery callbacks only, is causing more
problems than it solves.

We are moving toward instead unbinding the driver, resetting the
device, then re-binding the driver instead of unplug/replug.

Also why would you ever bypass the driver callbacks if the driver has
some ? The whole point is to keep the driver bound while resetting the
device (provided it has the right callbacks) so we don't lose the
linkage between stroage devices and mounted filesystems.

Cheers,
Ben.