On Thu, Sep 21, 2023 at 02:07:09PM -0300, Jason Gunthorpe wrote: > On Thu, Sep 21, 2023 at 01:01:12PM -0400, Michael S. Tsirkin wrote: > > On Thu, Sep 21, 2023 at 01:52:24PM -0300, Jason Gunthorpe wrote: > > > On Thu, Sep 21, 2023 at 10:43:50AM -0600, Alex Williamson wrote: > > > > > > > > With that code in place a legacy driver in the guest has the look and > > > > > feel as if having a transitional device with legacy support for both its > > > > > control and data path flows. > > > > > > > > Why do we need to enable a "legacy" driver in the guest? The very name > > > > suggests there's an alternative driver that perhaps doesn't require > > > > this I/O BAR. Why don't we just require the non-legacy driver in the > > > > guest rather than increase our maintenance burden? Thanks, > > > > > > It was my reaction also. > > > > > > Apparently there is a big deployed base of people using old guest VMs > > > with old drivers and they do not want to update their VMs. It is the > > > same basic reason why qemu supports all those weird old machine types > > > and HW emulations. The desire is to support these old devices so that > > > old VMs can work unchanged. > > > > > > Jason > > > > And you are saying all these very old VMs use such a large number of > > legacy devices that over-counting of locked memory due to vdpa not > > correctly using iommufd is a problem that urgently needs to be solved > > otherwise the solution has no value? > > No one has said that. > > iommufd is gaining alot more functions than just pinned memory > accounting. Yea it's very useful - it's also useful for vdpa whether this patchset goes in or not. At some level, if vdpa can't keep up then maybe going the vfio route is justified. I'm not sure why didn't anyone fix iommufd yet - looks like a small amount of work. I'll see if I can address it quickly because we already have virtio accelerators under vdpa and it seems confusing to people to use vdpa for some and vfio for others, with overlapping but slightly incompatible functionality. I'll get back next week, in either case. I am however genuinely curious whether all the new functionality is actually useful for these legacy guests. > > Another question I'm interested in is whether there's actually a > > performance benefit to using this as compared to just software > > vhost. I note there's a VM exit on each IO access, so ... perhaps? > > Would be nice to see some numbers. > > At least a single trap compared with an entire per-packet SW flow > undoubtably uses alot less CPU power in the hypervisor. > > Jason Something like the shadow vq thing will be more or less equivalent then? That's upstream in qemu and needs no hardware support. Worth comparing against. Anyway, there's presumably actual hardware this was tested with, so why guess? Just test and post numbers. -- MST
