Jason, Eugenio, Apologies for a late reply; just back from the long holiday.
On Thu, Dec 21, 2023 at 09:20:40AM +0100, Eugenio Perez Martin wrote: > Si-Wei did the actual profiling as he is the one with the 128G guests, > but most of the time was spent in the memory pinning. Si-Wei, please > correct me if I'm wrong. IIUC we're talking about no-vIOMMU use case. The pinning should indeed take a lot of time if it's similar to what VFIO does. > > I didn't check VFIO, but I think it just maps at realize phase with > vfio_realize -> vfio_attach_device -> vfio_connect_container(). In > previous testings, this delayed the VM initialization by a lot, as > we're moving that 20s of blocking to every VM start. > > Investigating a way to do it only in the case of being the destination > of a live migration, I think the right place is .load_setup migration > handler. But I'm ok to move it for sure. If it's destined to map the 128G, it does sound sensible to me to do it when VM starts, rather than anytime afterwards. Could anyone help to explain what's the problem if vDPA maps 128G at VM init just like what VFIO does? Thanks, -- Peter Xu
