11:13 a.m.
On Wed, 25 May 2022 at 16:07, Laszlo Ersek <lersek(a)redhat.com> wrote:
+ Drew & Peter
On 05/25/22 15:30, Daniel P. Berrangé wrote:
- The patch seems to do what it says in the commit message.
- QEMU commit bab52d4bba3f ("target/arm: Add "-cpu max" support",
2018-03-09) confirms what the commit message says, about both TCG and
KVM.
- To smoke-test the TCG-related change, I've edited a long-term TCG
aarch64 libvirt domain of mine, replacing "cortex-a57" with "max".
Both edk2 and the Linux guest continued working. So I guess the TCG
change is OK.
One thing to note here is that if you are using:
* TCG -cpu max
* 'virt' with no named version or with 'virt-7.0' or later
* a Linux kernel version prior to v5.12
then a bug in Linux means it won't boot. (This is because of
the LPA2 CPU feature which TCG -cpu max now emulates; older
kernels were buggy and won't boot on an LPA2 CPU, including
a real hardware one.)
You might or might not feel this is something worth noting in
release notes or equivalent.
- In additional support of the above, QEMU commit ddaebdda53fc
("target/arm: Unindent unnecessary else-clause", 2022-02-21) added a
comment saying
/* '-cpu max' for TCG: we currently do this as "A57 with extra
things" */
- Although I was more surprised by the TCG-related statement initially
(i.e. that "max" was a superset of "cortex-a57" when using TCG),
now
I'm actually more concerned about the KVM case.
The TCG part is an implementation convenience (mostly it just
means that 'max' has the A57's IMPDEF registers).
Specifically QEMU commit 0baa21be497d ("target/arm: Make KVM
-cpu max
exactly like -cpu host", 2022-02-21) eliminated a difference where
"-cpu max" had been a superset of "-cpu host", featuring the
"sve-max-vq" extra property.
The fix is part of release v7.0.0.
The difference was introduced in commits
[1] 6fa8a37949d9 ("target/arm/cpu64: max cpu: Support sve properties
with KVM", 2019-11-01)
[2] 87014c6b3660 ("target/arm/kvm: host cpu: Add support for sve<N>
properties", 2019-11-01)
and apparently *deliberately*.
No, this was unintentional. See the discussion in this thread:
https://lore.kernel.org/qemu-devel/20220203173640.shxkmatdcsfzzvtj@gator/
In particular, although KVM '-cpu max' had the sve-max-vq
property, Drew notes "sve-max-vq won't work for any of the machines that
support SVE that I know of". Which is why we removed it, rather
than adding it to '-cpu host'.
Therefore it seems that starting with qemu-4.2, but strictly
preceding
qemu-7.0, "-cpu max" and "-cpu host" are not "identical"
when KVM is
enabled; "-cpu max" has more features. Because of that, I think there
are two options:
(a) This extra feature is actually harmless, so we should only update
the commit message (i.e., generally speaking, "-cpu max" has been
a superset of "-cpu host" on KVM and of "-cpu cortex-a57" on
TCG).
(b) The feature actually presents a problem, and qemu in [v4.2.0,
v7.0.0) will not start when KVM accel and "-cpu max" are requested
simultaneously. In this case, I think the appliance needs to stick
with "-cpu host" on KVM.
I don't understand why you think these are the only two options. The
actual situation is:
(c) -cpu max and -cpu host have always been identical on KVM,
and this commit does not change that.
There happens to have been a QOM property 'sve-max-vq' on 'max'
that should not have existed there and that nobody can actually have
been usefully setting, but now there isn't.
thanks
-- PMM