Re: [libvirt] [PATCHv2 3/5] Allocate virtio-serial addresses when starting a domain

On Mon, Mar 23, 2015 at 05:46:23PM -0400, John Ferlan wrote: >

> > Mostly for my edification... These examples previously although > indicating they were "auto-assign" (of sorts) really weren't - they were > more force-assign for each example. Force-assign after this patch? Otherwise I don't understand.

> The way to auto-assign is by setting port='0', correct? > Yes. > However, I'm still missing something from the *auto.args output. It > seems the controller='#' is ignored and I guess I don't understand > that... I must've overlooked that. It shouldn't be a problem to take it as a hint in virDomainVirtioSerialAddrAssign.

> Sure "port='0'" (meaning first available port on the > controller), but I would have expected to see : > > kvm.port.0 use "virtio.serial.0.0,nr=1..." (which it does) > kvm.port.foo use "virtio.serial.1.0,nr=1..." (on controller 0?, but XML > has controller='1') > kvm.port.bar use "virtio.serial.1.0,nr=3..." (which it does) > kvm.port.wizz use "virtio.serial.0.0,nr=2" (incorrect due to others) > kvm.port.ooh use "virtio.serial.1.0,nr=2", (on controller 0?, but XML > has controller='1' > kvm.port.lla use "virtio.serial.2.0,nr=1" (on controller 0?, but XML has > controller='2') > > Now if been if "lla" used controller='0', then I assume would nr=4 be > chosen since 1,2 were auto-assigned, 3 was specifically assigned, thus 4 > would be the next one. > > Continuing with that same logic, the *-autoassign example could have > shown that if controller='0',port='2' and 'controller='1',port='1' were > preassigned, then the controllers/ports assigned would be 0.0,nr=1, > 0.0,nr=3, 0.0,nr=4, 1.0,nr=2 (since only 4 ports on controller='0' can > be used w/ 2 be static and controller='1' having port '1' already in use). > nr=4 is out of bounds for a controller with 4 ports. The ports are numbered from 0, but port number 0 can only be used for virtconsoles, not channels.