On Thu, 12 Jan 2023 01:56:14 -0500 "Michael S. Tsirkin" <[email protected]> wrote:
> I think fundamentally from spec POV memory is shared between devices. Right, but with the gid stuff and the corresponding rules shared between two or more arbitrary virtio-ism devices won't do. We need to find a way to express the what device can communicate with what device relationship. > How sharing is accomplished guest does not care so neither should the > spec. One of the goals of the spec is to foster interoperability. I wonder how far that goes. For example one could imagine a shared memory on the same host implementation by one vendor, and an RDMA based implementation of an other vendor both implementing the very same interface on the driver-device level. Two entities would not be able to talk to each other via virtio-ism devices that use different ways to accomplish the sharing. Is that out of scope for this spec? > Can some RDMA tricks be used for synchronisation behind the > scenes? I'm not familiar enough with RDMA. But I guess it may also depend on the "memory consistency" and coherency properties. Which are not specified for now for the ISM shared memory regions AFAIU. > Maybe, the spec does not care. But we can give an example. > At this point I'm not sure, whether the spec should care or not. > So something like: > > An ISM(Internal Shared Memory) device provides the ability to > access memory shared between multiple devices. This allows low-overhead > communication in presence of such memory. For example, memory can be > shared with guests of multiple virtual machines running on the same > host, with each virtual machine including an ISM device and with > the guests using the ISM devices to access the shared memory. > > what do others think? I agree, the spec should be as abstract as possible. As stated above, I don't have clarity on the interoperability goals. Is multiple flavors of virtio-ism devices that are not mutually interoperable a good outcome? Regards, Halil --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
