[Qemu-devel] Re: [PATCH 1/3] NUMA: add -numa command line option

[Andre Przywara]

Anthony Liguori wrote:
> Andre Przywara wrote:
> > This adds and parses a -numa command line option to QEMU.
> > ...
> > +
> > +#define MAX_NODES 64
> > +extern int numnumanodes;
> > +extern uint64_t hostnodes[MAX_NODES];
> >   
>
> This should not be in this patch.  This patch should just contain the 
> bits needed to create a virtual guest NUMA topology.  The hostnode stuff 
> should be a separate patch.
Agreed. I now moved the firmware interface parts of patch 2 into this 
one, so that you have full functionality with this first patch already 
(except host pinning and the BIOS part). Shall I move the 'info numa' 
part of the monitor also into this one or do you prefer smaller patches?

> > @@ -3968,6 +3972,10 @@ static void help(int exitcode)
> >         "-daemonize      daemonize QEMU after initializing\n"
> >  #endif
> >         "-option-rom rom load a file, rom, into the option ROM space\n"
> > +           "-numa 
> > nrnodes[,mem:size1[;size2..]][,cpu:cpu1[;cpu2..]][,pin:node1[;node2]]\n"
> > +           "                create a multi NUMA node guest and 
> > optionally pin it to\n"
> > +           "                to the given host nodes. If mem and cpu 
> > are omitted,\n"
> > +           "                resources are split equally\n"
> >   
>
> You're whitespace damaged.
Actually the damage here is in the three lines before, which start with 
a leading tab. All other help lines are formatted like my ones.
Do you want me to send whitespace fixes for this? Before or after the 
NUMA part?

> > +    if (numnumanodes > 0) {
> > +        int i;
> > +
> > +        if (numnumanodes > smp_cpus)
> > +            numnumanodes = smp_cpus;
> >   
>
> Why have this limitation?  We would like to see CPU-less nodes supported 
> either as memory-only nodes or as IO nodes.
I introduced this after some tests, which made Linux crash. Honestly I 
don't remember the exact behavior or the error message anymore, but I 
will do some research again.

> Which leads me to the question of how to plan on describing which nodes 
> have what hardware?
Do you mean peripheral devices? Like multiple southbridges connected to 
different NUMA nodes? I don't think that this kind of emulation is 
needed for QEMU, beside that this is out of scope for the ACPI tables.
If you meant this, I can elaborate on this. If not, please explain.

Regards,
Andre.

--
Andre Przywara
AMD-Operating System Research Center (OSRC), Dresden, Germany
Tel: +49 351 277-84917
----to satisfy European Law for business letters:
AMD Saxony Limited Liability Company & Co. KG,
Wilschdorfer Landstr. 101, 01109 Dresden, Germany
Register Court Dresden: HRA 4896, General Partner authorized
to represent: AMD Saxony LLC (Wilmington, Delaware, US)
General Manager of AMD Saxony LLC: Dr. Hans-R. Deppe, Thomas McCoy