[PATCH] Fixes to Xen pcifront and pciback (v1)

[Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>]

Hey,

These are patches that were developed for the Debian bug
810379 which san Tommi had openned.

The issue around from the two XSA fixes - which introduced
this regression.

I am the person who developed them and my explanation for this
regression oversight is that I tested for the 'exploit use-case'
and also did not have the VF automatic testing for PV guests
working at that point (only for HVM guests). When testing it
with with an PF in PV mode it worked - as the device would
fallback to legacy interrupts so things looked fine (argh)
from the outside (could ping it).

Either way it is embarrassing and I am sorry for the trouble
this caused.

Now the patches:
 [PATCH 1/4] xen/pciback: Check PF instead of VF for

Fixes regression introduced by XSA-157 for VF guests that
use MSI-X.

 [PATCH 2/4] xen/pciback: Save the number of MSI-X entries to be

Fixes regression introduced by XSA-155 - where MSI-X vectors
would not be copied back to the guest.

 [PATCH 3/4] xen/pcifront: Report the errors better.

Makes it easier to troubleshoot in the future.
 [PATCH 4/4] xen/pcifront: Fix mysterious crashes when NUMA locality

This one I think has been in the driver since the first time
the PV driver was introduced.

 arch/x86/include/asm/xen/pci.h        |  4 ++--
 arch/x86/pci/xen.c                    |  5 ++++-
 drivers/pci/xen-pcifront.c            | 11 +++++++----
 drivers/xen/xen-pciback/pciback_ops.c | 12 +++++++++---
 4 files changed, 22 insertions(+), 10 deletions(-)

Konrad Rzeszutek Wilk (4):
      xen/pciback: Check PF instead of VF for PCI_COMMAND_MEMORY
      xen/pciback: Save the number of MSI-X entries to be copied later.
      xen/pcifront: Report the errors better.
      xen/pcifront: Fix mysterious crashes when NUMA locality information was extracted.