acks are
invoked and it is checked afterwards.

Clearing the state_saved flag in pci_restore_state() merely serves the
purpose of ensuring that the flag is cleared ahead of the next suspend
and resume cycle.

It is a fallacy to think that state_saved indicates validity of the
saved state.

Unfortunately pci_restore_state() was amended by c82f63e411f1 to
bail out if state_saved is false.  This has arguably caused more
problems than it solved, so I have prepared this development branch
which essentially reverts the commit and undoes most of the awful
workarounds that it necessitated:

https://github.com/l1k/linux/commits/aer_reset_v1

I intend to submit this after the merge window has closed.

The motivation of c82f63e411f1 was to prevent restoring state if
pci_save_state() hasn't been called before.  I am solving that by
calling pci_save_state() on device addition, hence error
recoverability is ensured at all times.

I believe this also makes patch [01/10] in your series unnecessary.

A lot of drivers call pci_save_state() in their probe hook and
that continues to be correct if they modified Config Space
vis-a-vis what was saved on device addition.

Thanks,

Lukas[PATCH v4 01/10] PCI: Avoid saving error values for config spaceLukas Wunner <lukas@wunner.de> undefinedFarhan Ali <alifm@linux.ibm.com> undefined undefined undefined undefined undefined undefined undefined undefined undefined undefined undefined%ƒp