|
From: | Richard Henderson |
Subject: | Re: [RFC PATCH 8/8] target/i386: Move X86XSaveArea into TCG |
Date: | Thu, 8 Jul 2021 08:22:02 -0700 |
User-agent: | Mozilla/5.0 (X11; Linux x86_64; rv:78.0) Gecko/20100101 Thunderbird/78.11.0 |
On 7/8/21 12:45 AM, David Edmondson wrote:
Actually, that's nonsense. With KVM or HVF we have to use the offsets of the host CPU, as the hardware won't do anything else, irrespective of the general CPU model chosen. To have KVM -> TCG migration work it would be necessary to pass the offsets in the migration stream and have TCG observe them, as you originally said. TCG -> KVM migration would only be possible if TCG was configured to use the same offsets as would later required by KVM (meaning, the host CPU).
And kvm -> kvm migration, with the same general cpu model chosen, but with different host cpus with different offsets?
It seems like we must migrate then and verify the offsets in that case, so that we can fail the migration.
r~
[Prev in Thread] | Current Thread | [Next in Thread] |