[PATCH 5/5] vfio-pci: Allow to mmap MSI-X table if interrupt remapping is supported
David.Laight at ACULAB.COM
Fri May 13 19:16:18 AEST 2016
From: Alex Williamson [mailto:alex.williamson at redhat.com]
> Sent: 13 May 2016 06:33
> Simply denying direct writes to the vector table or preventing mapping
> of the vector table into the user address space does not provide any
> tangible form of protection. Many devices make use of window registers
> that allow backdoors to arbitrary device registers. Some drivers even
> use this as the primary means for configuring MSI-X, which makes them
> incompatible with device assignment without device specific quirks to
> enable virtualization of these paths.
We have one fgpa based PCIe slave where the device driver has to read
the MSI-X table and then write the value to other fpga registers so
that the logic can generate the correct PCIe write cycle when an
interrupt is requested.
The MSI-X table itself is only as a PCIe slave.
We also have host accessible DMA controllers that the device driver
uses to copy data to kernel memory.
These could easily be used to generate arbitrary MSI-X requests.
As I've said earlier it is almost certainly possible to get any
ethernet hardware to perform something similar.
So without hardware that is able to limit the memory and MSI-X
that each PCIe endpoint can access I believe that if a virtualisation
system gives a guest kernel direct access to a PCIe devices it gives
the guest kernel the ability to raise and MSI-X interrupt and read/write
any physical memory.
(I've not looked at the cpu virtualisation support, but do know what
the PCIe devices can do.)
More interestingly, probably the 'worst' thing (from a security point of view)
that changing the MSI-X table lets you do is a write to an arbitrary
physical memory address.
More information about the Linuxppc-dev