On Thu, Jul 20, 2017 at 3:15 PM, Alex Williamson <alex.william...@redhat.com> wrote:
> Most of the ACS capabilities are worded as "Must be implemented by > devices that implement ..." Shouldn't a hard-wired ACS capability > sufficiently describe that, or is there something wrong with how > they're hard wired? Interesting question. I just looked at what the PCI spec says about the various bits for ACS functions in multi-function devices. Many of the functions are "must not be implemented." Of the ones that are "must be implemented if..." the key is that the if is for devices that support peer-to-peer traffic. I think the Intel NICs are compliant with the spec - they don't support any ACS mechanisms for controlling peer-to-peer traffic, but they also don't implement peer-to-peer traffic. This means that the PCI standard way of knowing that it is safe to assign individual functions does not prove it is safe - but with device-specific knowledge we do know it is safe. Hence a quirk to give that device-specific information to the kernel. - R.
