On 07/08/2015 04:45 PM, David Gibson wrote:
On Wed, Jul 08, 2015 at 04:40:27PM +1000, Alexey Kardashevskiy wrote:
On 07/08/2015 03:37 PM, David Gibson wrote:
On Wed, Jul 08, 2015 at 02:10:28PM +1000, Alexey Kardashevskiy wrote:
On 07/08/2015 01:43 PM, David Gibson wrote:
From: Andrea Bolognani <abolo...@redhat.com>
Add a missing PVR value for the POWER8E v2.1 CPU. Information taken
>from the kernel cputable.
Signed-off-by: Andrea Bolognani <abolo...@redhat.com>
Signed-off-by: David Gibson <da...@gibson.dropbear.id.au>
---
target-ppc/cpu-models.c | 4 +++-
target-ppc/cpu-models.h | 1 +
2 files changed, 4 insertions(+), 1 deletion(-)
Alex,
Not sure if this counts as a bugfix which can be merged now we're in
the hard freeze. The lack of it does mean we can't work on one of our
dev machines which has this CPU.
Why is this a stopper? We stopped bothering with exact PVRs some time ago
and -cpu POWER8 or -cpu host still work.
Andrea, can you clarify?
I think it's because libvirt likes to specify a specific CPU - and if
it gets the new PVR from the host, qemu won't understand it.
A specific CPU in this case is "POWER8", I added this specifically for
libvirt (to allow migration between all versions of POWER8), it should not
use versioned CPUs and it does not in powerkvm.
Uh.. won't that make qemu attempt to set a specific PVR, though -
which will fail with recent KVM if it's not *exactly* the same as the
host PVR.
The "POWER8" CPU class is created dynamically (like the "host" CPU class)
and has the actual host PVR so setting it to KVM cannot fail.
When TCG, the class is not registered and alias is used instead.
There is actually a patch to make this PVR masking/subclassing nicer, I
replied to it with adding you in cc:, please have a look. Andreas Faerber
had objections which I did not really grasp then.
--
Alexey
