[PATCH v2 1/1] KVM: arm64: allow the VM to select DEVICE_* and NORMAL_NC for IO memory

[Lorenzo Pieralisi]

On Thu, Dec 07, 2023 at 09:38:25AM -0400, Jason Gunthorpe wrote:
> On Thu, Dec 07, 2023 at 11:13:52AM +0100, Lorenzo Pieralisi wrote:
> > > > What about the other way around - would we have a prefetchable BAR that
> > > > has portions which are unprefetchable?
> > > 
> > > I would say possibly.
> > > 
> > > Prefetch is a dead concept in PCIe, it was obsoleted in PCI-X about 20
> > > years ago. No PCIe system has ever done prefetch.
> > > 
> > > There is a strong incentive to mark BAR's as prefetchable because it
> > > allows 64 bit addressing in configurations with bridges.
> > 
> > If by strong incentive you mean the "Additional guidance on the
> > Prefetchable Bit in Memory Space BARs" in the PCI express specifications,
> > I think it has been removed from the spec and the criteria that had to be
> > met to implement it were basically impossible to fulfill on ARM systems,
> > it did not make any sense in the first place.
> 
> No, I mean many systems don't have room to accommodate large 32 bit
> BARs and the only real way to make stuff work is to have a 64 bit BAR
> by setting prefetchable.

That's what the implementation note I mentioned referred to ;)

> Given mis-marking a read-side-effect region as prefetchable has no
> actual consequence on PCI-E I would not be surprised to learn people
> have done this.

PCIe specs 6.1, 7.5.1.2.1 "Base Address Registers"

"A function is permitted to mark a range as prefetchable if there are
no side effects on reads..."

I don't think that an OS should use the prefetchable flag to infer
anything (even though we do at the moment -> sysfs mappings, I know that
the prefetchable concept is being scrapped from the PCI specs altogether
for a reason), I don't see though how we can say that's a SW bug at
present given what I quoted above.

I'd agree that it is best not to use that flag for new code we are adding
(because it will be deprecated soon).

Thanks,
Lorenzo