Re: [Qemu-devel] [PATCH v7 01/13] machine: Don't allow CPU toplogies with partially filled cores

[Eduardo Habkost <ehabkost@redhat.com>]

On Fri, Jan 29, 2016 at 04:10:47PM +0100, Igor Mammedov wrote:
> On Fri, 29 Jan 2016 12:24:18 -0200
> Eduardo Habkost <ehabkost@redhat.com> wrote:
> 
> > On Fri, Jan 29, 2016 at 02:52:30PM +1100, David Gibson wrote:
> > > On Thu, Jan 28, 2016 at 11:19:43AM +0530, Bharata B Rao wrote:  
> > > > Prevent guests from booting with CPU topologies that have partially
> > > > filled CPU cores or can result in partially filled CPU cores after
> > > > CPU hotplug like
> > > > 
> > > > -smp 15,sockets=1,cores=4,threads=4,maxcpus=16 or
> > > > -smp 15,sockets=1,cores=4,threads=4,maxcpus=17.
> > > > 
> > > > This is enforced by introducing MachineClass::validate_smp_config()
> > > > that gets called from generic SMP parsing code. Machine type versions
> > > > that want to enforce this can define this to the generic version
> > > > provided.
> > > > 
> > > > Only sPAPR and PC machine types starting from version 2.6 enforce this in
> > > > this patch.
> > > > 
> > > > Signed-off-by: Bharata B Rao <bharata@linux.vnet.ibm.com>  
> > > 
> > > I've been kind of lost in the back and forth about
> > > threads/cores/sockets.
> > > 
> > > What, in the end, is the rationale for allowing partially filled
> > > sockets, but not partially filled cores?  
> > 
> > I don't think there's a good reason for that (at least for PC).
> > 
> > It's easier to relax the requirements later if necessary, than
> > dealing with compatibility issues again when making the code more
> > strict. So I suggest we make validate_smp_config_generic() also
> > check if smp_cpus % (smp_threads * smp_cores) == 0.
> 
> that would break exiting setups.

Not if we do that only on newer machine classes.
validate_smp_config_generic() will be used only on *-2.6 and
newer.


> 
> Also in case of cpu hotplug this patch will break migration
> as target QEMU might refuse starting with hotplugged CPU thread.

This won't change older machine-types.

But I think you are right: it can break migration on pc-2.6, too.
But: isn't migration already broken when creating other sets of
CPUs that can't represented using -smp?

How exactly would you migrate a machine today, if you run:

  $ qemu-system-x86_64 -smp 16,sockets=2,cores=2,threads=2,maxcpus=32
  (QMP) cpu-add id=31


> 
> Perhaps this check should be enforced per target/machine if
> arch requires it.

It is. Please see the patch. It introduces a validate_smp_config
method.

But we need your input to clarify if
validate_smp_config_generic() is safe for pc-2.6 too.

-- 
Eduardo