Re: [Qemu-devel] [RFC] QMP: add query-hotpluggable-cpus

[Igor Mammedov]

On Thu, 25 Feb 2016 12:03:21 +1100
David Gibson <address@hidden> wrote:

> On Wed, Feb 24, 2016 at 12:03:41PM +0100, Igor Mammedov wrote:
> > On Wed, 24 Feb 2016 21:51:19 +1100
> > David Gibson <address@hidden> wrote:
> >   
> > > On Wed, Feb 24, 2016 at 09:42:10AM +0100, Markus Armbruster wrote:  
> > > > David Gibson <address@hidden> writes:
> > > >     
> > > > > On Mon, Feb 22, 2016 at 10:05:54AM +0100, Markus Armbruster wrote:    
> > > > >> David Gibson <address@hidden> writes:
> > > > >>     
> > > > >> > On Fri, Feb 19, 2016 at 10:51:11AM +0100, Markus Armbruster wrote: 
> > > > >> >    
> > > > >> >> David Gibson <address@hidden> writes:
> > > > >> >>     
> > > > >> >> > On Thu, Feb 18, 2016 at 11:37:39AM +0100, Igor Mammedov wrote:  
> > > > >> >> >   
> > > > >> >> >> On Thu, 18 Feb 2016 14:39:52 +1100
> > > > >> >> >> David Gibson <address@hidden> wrote:
> > > > >> >> >>     
> > > > >> >> >> > On Tue, Feb 16, 2016 at 11:36:55AM +0100, Igor Mammedov 
> > > > >> >> >> > wrote:    
> > > > >> >> >> > > On Mon, 15 Feb 2016 20:43:41 +0100
> > > > >> >> >> > > Markus Armbruster <address@hidden> wrote:
> > > > >> >> >> > >       
> > > > >> >> >> > > > Igor Mammedov <address@hidden> writes:
> > > > >> >> >> > > >       
> > > > >> >> >> > > > > it will allow mgmt to query present and possible to 
> > > > >> >> >> > > > > hotplug CPUs
> > > > >> >> >> > > > > it is required from a target platform that wish to 
> > > > >> >> >> > > > > support
> > > > >> >> >> > > > > command to set board specific 
> > > > >> >> >> > > > > MachineClass.possible_cpus() hook,
> > > > >> >> >> > > > > which will return a list of possible CPUs with options
> > > > >> >> >> > > > > that would be needed for hotplugging possible CPUs.
> > > > >> >> >> > > > >
> > > > >> >> >> > > > > For RFC there are:
> > > > >> >> >> > > > >    'arch_id': 'int' - mandatory unique CPU number,
> > > > >> >> >> > > > >                       for x86 it's APIC ID for ARM 
> > > > >> >> >> > > > > it's MPIDR
> > > > >> >> >> > > > >    'type': 'str' - CPU object type for usage with 
> > > > >> >> >> > > > > device_add
> > > > >> >> >> > > > >
> > > > >> >> >> > > > > and a set of optional fields that would allows mgmt 
> > > > >> >> >> > > > > tools
> > > > >> >> >> > > > > to know at what granularity and where a new CPU could 
> > > > >> >> >> > > > > be
> > > > >> >> >> > > > > hotplugged;
> > > > >> >> >> > > > > [node],[socket],[core],[thread]
> > > > >> >> >> > > > > Hopefully that should cover needs for CPU hotplug 
> > > > >> >> >> > > > > porposes for
> > > > >> >> >> > > > > magor targets and we can extend structure in future 
> > > > >> >> >> > > > > adding
> > > > >> >> >> > > > > more fields if it will be needed.
> > > > >> >> >> > > > >
> > > > >> >> >> > > > > also for present CPUs there is a 'cpu_link' field which
> > > > >> >> >> > > > > would allow mgmt inspect whatever object/abstraction
> > > > >> >> >> > > > > the target platform considers as CPU object.
> > > > >> >> >> > > > >
> > > > >> >> >> > > > > For RFC purposes implements only for x86 target so 
> > > > >> >> >> > > > > far.        
> > > > >> >> >> > > > 
> > > > >> >> >> > > > Adding ad hoc queries as we go won't scale.  Could this 
> > > > >> >> >> > > > be solved by a
> > > > >> >> >> > > > generic introspection interface?      
> > > > >> >> >> > > Do you mean generic QOM introspection?
> > > > >> >> >> > > 
> > > > >> >> >> > > Using QOM we could have '/cpus' container and create QOM 
> > > > >> >> >> > > links
> > > > >> >> >> > > for exiting (populated links) and possible (empty links) 
> > > > >> >> >> > > CPUs.
> > > > >> >> >> > > However in that case link's name will need have a special 
> > > > >> >> >> > > format
> > > > >> >> >> > > that will convey an information necessary for mgmt to 
> > > > >> >> >> > > hotplug
> > > > >> >> >> > > a CPU object, at least:
> > > > >> >> >> > >   - where: [node],[socket],[core],[thread] options
> > > > >> >> >> > >   - optionally what CPU object to use with device_add 
> > > > >> >> >> > > command      
> > > > >> >> >> > 
> > > > >> >> >> > Hmm.. is it not enough to follow the link and get the 
> > > > >> >> >> > topology
> > > > >> >> >> > information by examining the target?    
> > > > >> >> >> One can't follow a link if it's an empty one, hence
> > > > >> >> >> CPU placement information should be provided somehow,
> > > > >> >> >> either:    
> > > > >> >> >
> > > > >> >> > Ah, right, so the issue is determining the socket/core/thread
> > > > >> >> > addresses that cpus which aren't yet present will have.
> > > > >> >> >    
> > > > >> >> >>  * by precreating cpu-package objects with properties that
> > > > >> >> >>    would describe it /could be inspected via OQM/    
> > > > >> >> >
> > > > >> >> > So, we could do this, but I think the natural way would be to 
> > > > >> >> > have the
> > > > >> >> > information for each potential thread in the package.  Just 
> > > > >> >> > putting
> > > > >> >> > say "core number" in the package itself assumes more than I'd 
> > > > >> >> > like
> > > > >> >> > about how packages sit in the heirarchy.  Plus, it means that
> > > > >> >> > management has a bunch of cases to deal with: package has all 
> > > > >> >> > the
> > > > >> >> > information, package has just a core id, package has just a 
> > > > >> >> > socket id,
> > > > >> >> > and so forth.
> > > > >> >> >
> > > > >> >> > It is a but clunky that when the package is plugged, this 
> > > > >> >> > information
> > > > >> >> > will have to sit parallel to the array of actual thread links.
> > > > >> >> >
> > > > >> >> > Markus or Andreas is there a natural way to present a list of 
> > > > >> >> > (node,
> > > > >> >> > socket, core, thread) tuples in the package object?  Preferably
> > > > >> >> > without having to create a whole bunch of "potential thread" 
> > > > >> >> > objects
> > > > >> >> > just for the purpose.    
> > > > >> >> 
> > > > >> >> I'm just a dabbler when it comes to QOM, but I can try.
> > > > >> >> 
> > > > >> >> I view a concrete cpu-package device (subtype of the abstract
> > > > >> >> cpu-package device) as a composite device containing stuff like 
> > > > >> >> actual
> > > > >> >> cores.    
> > > > >> >
> > > > >> > So.. the idea is it's a bit more abstract than that.  My intention 
> > > > >> > is
> > > > >> > that the package lists - in some manner - each of the threads
> > > > >> > (i.e. vcpus) it contains / can contain.  Depending on the platform 
> > > > >> > it
> > > > >> > *might* also have internal structure such as cores / sockets, but 
> > > > >> > it
> > > > >> > doesn't have to.  Either way, the contained threads will be listed 
> > > > >> > in
> > > > >> > a common way, as a flat array.
> > > > >> >    
> > > > >> >> To create a composite device, you start with the outer shell, 
> > > > >> >> then plug
> > > > >> >> in components one by one.  Components can be nested arbitrarily 
> > > > >> >> deep.
> > > > >> >> 
> > > > >> >> Perhaps you can define the concrete cpu-package shell in a way 
> > > > >> >> that lets
> > > > >> >> you query what you need to know from a mere shell (no components
> > > > >> >> plugged).    
> > > > >> >
> > > > >> > Right.. that's exactly what I'm suggesting, but I don't know enough
> > > > >> > about the presentation of basic data in QOM to know quite how to
> > > > >> > accomplish it.
> > > > >> >    
> > > > >> >> >> or
> > > > >> >> >>  * via QMP/HMP command that would provide the same information
> > > > >> >> >>    only without need to precreate anything. The only difference
> > > > >> >> >>    is that it allows to use -device/device_add for new CPUs.   
> > > > >> >> >>  
> > > > >> >> >
> > > > >> >> > I'd be ok with that option as well.  I'd be thinking it would be
> > > > >> >> > implemented via a class method on the package object which 
> > > > >> >> > returns the
> > > > >> >> > addresses that its contained threads will have, whether or not 
> > > > >> >> > they're
> > > > >> >> > present right now.  Does that make sense?    
> > > > >> >> 
> > > > >> >> If you model CPU packages as composite cpu-package devices, then 
> > > > >> >> you
> > > > >> >> should be able to plug and unplug these with device_add, unless 
> > > > >> >> plugging
> > > > >> >> them requires complex wiring that can't be done in qdev / 
> > > > >> >> device_add,
> > > > >> >> yet.    
> > > > >> >
> > > > >> > There's a whole bunch of issues raised by allowing device_add of
> > > > >> > cpus.  Although they're certainly interesting and probably useful, 
> > > > >> > I'd
> > > > >> > really like to punt on them for the time being, so we can get some
> > > > >> > sort of cpu hotplug working on Power (and s390 and others).    
> > > > >> 
> > > > >> If you make it a device, you can still set
> > > > >> cannot_instantiate_with_device_add_yet to disable -device / 
> > > > >> device_add
> > > > >> for now, and unset it later, when you're ready for it.    
> > > > >
> > > > > Yes, that was the plan.
> > > > >    
> > > > >> > The idea of the cpu packages is that - at least for now - the user
> > > > >> > can't control their contents apart from the single "present" bit.
> > > > >> > They already know what they can contain.    
> > > > >> 
> > > > >> Composite devices commonly do.  They're not general containers.
> > > > >> 
> > > > >> The "present" bit sounds like you propose to "pre-plug" all the 
> > > > >> possible
> > > > >> CPU packages, and thus reduce CPU hot plug/unplug to 
> > > > >> enabling/disabling
> > > > >> pre-plugged CPU packages.    
> > > > >
> > > > > Yes.    
> > > > 
> > > > I'm concerned this might suffer combinatorial explosion.
> > > > 
> > > > qemu-system-x86_64 --cpu help shows more than two dozen CPUs.  They can
> > > > be configured in numerous arrangements of sockets, cores, threads.  Many
> > > > of these wouldn't be physically possible with older CPUs.  Guest
> > > > software might work even with physically impossible configurations, but
> > > > arranging virtual models of physical hardware in physically impossible
> > > > configurations invites trouble, and should best be avoided.
> > > > 
> > > > I'm afraid I'm still in the guess-what-you-mean stage because I lack
> > > > concrete examples to go with the abstract description.  Can you
> > > > enumerate the pre-plugged CPU packages for a board of your choice to
> > > > give us a better idea of how your proposal would look like in practice?
> > > > Then describe briefly what a management application would need to know
> > > > about them, and what it would do with the knowledge?
> > > > 
> > > > Perhaps a PC board would be the most useful, because PCs are probably
> > > > second to none in random complexity :)    
> > > 
> > > Well, it may be moot at this point, since Andreas has objected
> > > strongly to Bharata's draft for reasons I have yet to really figure
> > > out.
> > > 
> > > But I think the answer below will clarify this.
> > >   
> > > > >> What if a board can take different kinds of CPU packages?  Do we
> > > > >> pre-plug all combinations?  Then some combinations are non-sensical.
> > > > >> How would we reject them?    
> > > > >
> > > > > I'm not trying to solve all cases with the present bit handling - just
> > > > > the currently common case of a machine with fixed maximum number of
> > > > > slots which are expected to contain identical processor units.
> > > > >    
> > > > >> For instance, PC machines support a wide range of CPUs in various
> > > > >> arrangements, but you generally need to use a single kind of CPU, and
> > > > >> the kind of CPU restricts the possible arrangements.  How would you
> > > > >> model that?    
> > > > >
> > > > > The idea is that the available slots are determined by the machine,
> > > > > possibly using machine or global options.  So for PC, -cpu and -smp
> > > > > would determine the number of slots and what can go into them.    
> > > > 
> > > > Do these CPU packages come with "soldered-in" CPUs?  Or do they provide
> > > > slots where CPUs can be plugged in?  From what I've read, I guess it's
> > > > the latter, together with a "thou shalt not plug in different CPUs"
> > > > commandment.  Correct?    
> > > 
> > > No, they do in fact come with "soldered in" CPUS.  Once the package is
> > > constructed it is either absent, or supplies exactly one set of cpu
> > > threads (and possibly other bits and pieces), there is no further
> > > configuration.
> > > 
> > > So:
> > >   qemu-system-x86_64 -machine pc -cpu Haswell -smp 2,maxcpus=8
> > > 
> > > Would give you 8 cpu packages. 2 would initially be present, the rest
> > > would be absent.  If you toggle an absent one to present, another
> > > single-thread Haswell would appear in the guest.
> > > 
> > >   qemu-system-x86_64 -machine pc -cpu Haswell \
> > >           -smp 2,threads=2,cores=2,sockets=2,maxcpus=8
> > >   
> > ok now lets imagine that mgmt set 'present'=on for pkg 7 and
> > that needs to be migrated, how would target QEMU be able to recreate
> > the state of source QEMU instance?  
> 
> Ugh, yeah, I'm not sure that will work.
> 
> I had just imagined that we'd migrate the present bit for the pkg, and
> it would construct the necessary threads on the far end.  But ordering
> that with the transfer of the thread state could get hairy.
That's not how migration works now so unless you'd wish to implement
this new migration behavior (I have no clue how complex it would be)
it'd be better to stick with current migration workflow where all
devices that exist on source side (including hotplugged ones) are
created on target at target's CLI in the order they were created
on source.