> > Do you know the size of the ram_addr_t space from > > VHOST_USER_SET_MEM_TABLE's user address and size fields? > > For some reason, vhost_get_log_size() also takes pc-bios region. I > think it's quite unnecessary given that the backend will not have > access to this region.
That's by design. The pc-bios is mapped into the guest memory, so it is taken into account. It shouldn't be a problem, it's just wasting memory for <4GB VMs but it's not more expensive at run time. > get_log_size() also computes the size of the rings, I wonder if it's > necessary since the dev->mem->regions should already contain the > rings, isn't it? Indeed. > > If the size isn't needed, you can reuse LOG_BASE, ignoring the content > > of the payload and adding the SCM_RIGHTS file descriptor. > > That's possible, if we assume that existing backends won't leak the > passed fds (vapp does for ex), perhaps it needs a new flag (vhost-user > features flags are common with vhost, not sure we want to add that > here, but where else) Ok, that's for mst to answer. I think the leak wouldn't be too bad because right now vhost-user live migration corrupts data so nobody must be using it. But if you want to introduce a new message, that would also be okay of course. Paolo
