On Wed, Dec 06, 2023 at 04:18:05PM +0100, Lorenzo Pieralisi wrote: > On Wed, Dec 06, 2023 at 11:05:56AM -0400, Jason Gunthorpe wrote: > > On Wed, Dec 06, 2023 at 02:49:02PM +0000, Catalin Marinas wrote: > > > On Tue, Dec 05, 2023 at 03:48:22PM -0400, Jason Gunthorpe wrote: > > > > On Tue, Dec 05, 2023 at 07:24:37PM +0000, Catalin Marinas wrote: > > > > > On Tue, Dec 05, 2023 at 12:43:18PM -0400, Jason Gunthorpe wrote: > > > > > > What if we change vfio-pci to use pgprot_device() like it already > > > > > > really should and say the pgprot_noncached() is enforced as > > > > > > DEVICE_nGnRnE and pgprot_device() may be DEVICE_nGnRE or NORMAL_NC? > > > > > > Would that be acceptable? > > > > > > > > > > pgprot_device() needs to stay as Device, otherwise you'd get speculative > > > > > reads with potential side-effects. > > > > > > > > I do not mean to change pgprot_device() I mean to detect the > > > > difference via pgprot_device() vs pgprot_noncached(). They put a > > > > different value in the PTE that we can sense. It is very hacky. > > > > > > Ah, ok, it does look hacky though (as is the alternative of coming up > > > with a new specific pgprot_*() that KVM can treat differently). > > > > > > BTW, on those Mellanox devices that require different attributes within > > > a BAR, do they have a problem with speculative reads causing > > > side-effects? > > > > Yes. We definitely have had that problem in the past on older > > devices. VFIO must map the BAR using pgprot_device/noncached() into > > the VMM, no other choice is functionally OK. > > Were those BARs tagged as prefetchable or non-prefetchable ? I assume the > latter but please let me know if I am guessing wrong. I don't know it was quite old HW. Probably. Just because a BAR is not marked as prefetchable doesn't mean that the device can't use NORMAL_NC on subsets of it. Jason
