On Wed, Dec 06, 2023 at 11:16:03AM -0400, Jason Gunthorpe wrote: > On Wed, Dec 06, 2023 at 12:14:18PM +0000, Catalin Marinas wrote: > > We could do with a pgprot_maybewritecombine() or > > pgprot_writecombinenospec() (similar to Jason's idea but without > > changing the semantics of pgprot_device()). For the user mapping on > > arm64 this would be Device (even _GRE) since it can't disable > > speculation but stage 2 would leave the decision to the guest since the > > speculative loads aren't much different from committed loads done > > wrongly. > > This would be fine, as would a VMA flag. Please pick one :) > > I think a VMA flag is simpler than messing with pgprot. I guess one could write a patch and see how it goes ;). > > If we want the VMM to drive this entirely, we could add a new mmap() > > flag like MAP_WRITECOMBINE or PROT_WRITECOMBINE. They do feel a bit > > As in the other thread, we cannot unconditionally map NORMAL_NC into > the VMM. I'm not suggesting this but rather the VMM map portions of the BAR with either Device or Normal-NC, concatenate them (MAP_FIXED) and pass this range as a memory slot (or multiple if a slot doesn't allow multiple vmas). > > The latter has some benefits for DPDK but it's a lot more involved > > with > > DPDK WC support will be solved with some VFIO-only change if anyone > ever cares to make it, if that is what you mean. Yeah. Some arguments I've heard in private and public discussions is that the KVM device pass-through shouldn't be different from the DPDK case. So fixing that would cover KVM as well, though we'd need additional logic in the VMM. BenH had a short talk at Plumbers around this - https://youtu.be/QLvN3KXCn0k?t=7010. There was some statement in there that for x86, the guests are allowed to do WC without other KVM restrictions (not sure whether that's the case, not familiar with it). > > having to add device-specific knowledge into the VMM. The VMM would also > > have to present the whole BAR contiguously to the guest even if there > > are different mapping attributes within the range. So a lot of MAP_FIXED > > uses. I'd rather leaving this decision with the guest than the VMM, it > > looks like more hassle to create those mappings. The VMM or the VFIO > > could only state write-combine and speculation allowed. > > We talked about this already, the guest must decide, the VMM doesn't > have the information to pre-predict which pages the guest will want to > use WC on. Are the Device/Normal offsets within a BAR fixed, documented in e.g. the spec or this is something configurable via some MMIO that the guest does. -- Catalin
