Re: [Qemu-devel] [PATCH] target-arm: Implement qmp query-cpu-definitions

["Andreas Färber" <afaerber@suse.de>]

Hi,

Am 04.09.2013 17:23, schrieb Cole Robinson:
> Libvirt uses this to introspect available CPU models.
> 
> Signed-off-by: Cole Robinson <crobinso@redhat.com>
> ---
> Maybe this will be centrally handled after the QOM CPU work is done?

No, this API is about a pre-QOM command line switch, whose
implementation is target-specific. What got standardized for multiple
targets was the CPUClass::class_by_name() hook for the opposite
direction of -cpu to ObjectClass lookup (with the goal of obsoleting
cpu_init()). It is not wrong to implement query-cpus for arm, but -cpu
is considered deprecated.

And personally especially for arm I would rather welcome someone
contributing a proper, e.g., Raspberry Pi board (with clean separation
of what is on the SoC and what on the board to avoid duplication among
boards) than tweaking some other unrelated board with -cpu and then
complaining that it doesn't work as expected. ;) For machines that use
soft-core CPUs (FPGAs) or for mach-virt as non-physical machine it does
make sense, but query-cpus gives us a list irrespective of whether they
are compatible with the board. Same for ppc btw - plugging, e.g., a 440
CPU into pseries or mac99 won't work.
Short-term my view is to use fixed CPU types, ignoring -cpu, where the
CPU cannot be exchanged (e.g., DIGIC patch series) and a suitable link<>
type otherwise.

The long-term goal is to have machines be config files (thereby
user-creatable) assembling QOM objects by specifying memory mapping and
IRQ routing. The latter two parts don't work with the old qdev and
SysBusDevice APIs.

That said, patch itself looks correct,

Reviewed-by: Andreas Färber <afaerber@suse.de>

Regards,
Andreas

-- 
SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany
GF: Jeff Hawn, Jennifer Guild, Felix Imendörffer; HRB 16746 AG Nürnberg