Re: [Qemu-devel] [PATCH v6 01/11] machine: Don't allow CPU toplogies with partially filled cores

[Eduardo Habkost]

On Mon, Jan 25, 2016 at 02:27:25PM +0530, Bharata B Rao wrote:
> On Sat, Jan 23, 2016 at 11:47:22AM -0200, Eduardo Habkost wrote:
> > On Fri, Jan 08, 2016 at 12:25:09PM +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 don't want to enforce this can override this method.
> > > 
> > > TODO: Only sPAPR and PC changes are done in this patch, other archs
> > > will be touched after there is agreement on this approach.
> > > 
> > > Signed-off-by: Bharata B Rao <address@hidden>
> > > ---
> > >  hw/core/machine.c   | 20 ++++++++++++++++++++
> > >  hw/i386/pc_piix.c   |  7 +++++++
> > >  hw/i386/pc_q35.c    |  7 +++++++
> > >  hw/ppc/spapr.c      |  7 +++++++
> > >  include/hw/boards.h |  1 +
> > >  vl.c                |  4 ++++
> > >  6 files changed, 46 insertions(+)
> > > 
> > > diff --git a/hw/core/machine.c b/hw/core/machine.c
> > > index c46ddc7..b66c101 100644
> > > --- a/hw/core/machine.c
> > > +++ b/hw/core/machine.c
> > > @@ -336,6 +336,25 @@ static void machine_init_notify(Notifier *notifier, 
> > > void *data)
> > >      foreach_dynamic_sysbus_device(error_on_sysbus_device, NULL);
> > >  }
> > >  
> > > +static int validate_smp_config_generic(int smp_cpus, int max_cpus,
> > > +                                       int smp_threads)
> > 
> > Please make it use a Error** parameter to return error
> > information, and let the caller decide what to do with the error
> > message. One day the mc->validate_smp_config() call may be moved
> > inside a function that returns error information using Error**
> > and needs to propagate it to the caller.
> 
> Sure, will make this change.
> 
> I see that not all target archs use machine type versions and even
> if they do it will not be a one-line changer (as shown for i386 and sPAPR in
> this patch) for them to override mc->validate_smp_config() for older
> versions.
> 
> So am I expected to change .class_init routine for all these archs by
> setting mc->validate_smp_config() to a NOP validate_smp_config()
> so that the default validate_smp_config() defined for TYPE_MACHINE
> is overridden ? In order to preserve the existing behaviour with
> CPU topology for all args, it appears that such a change would be
> necessary, but I am not sure how to test all of them. Will just a compile
> test do ?

I am not sure I understand the question: if you want most machine
classes to have a NOP validate_smp_config(), why do you want to
make validate_smp_config_generic() the default in TYPE_MACHINE?
Just make a more convenient default in TYPE_MACHINE, and override
it in the pc and spapr base classes.

In other words, if you don't want to risk breaking other machine
classes, you just need explicit one-line validate_smp_config
assignments in 5 class_init/machine_options functions:

Class              | validate_smp_config pointer
-------------------+------------------------------
TYPE_MACHINE       | NULL (no explicit assignment)
TYPE_PC_MACHINE    | validate_smp_config_generic
TYPE_SPAPR_MACHINE | validate_smp_config_generic
pc-i440fx-2.5      | NULL
pc-q35-2.5         | NULL
spapr-2.5          | NULL

-- 
Eduardo