On Wed, Nov 20, 2024 at 12:44:54 +0000, Daniel P. Berrangé wrote:
On Tue, Nov 19, 2024 at 07:49:58PM +0100, Jiri Denemark wrote:
> Signed-off-by: Jiri Denemark <jdenemar(a)redhat.com>
> diff --git a/src/cpu_map/x86_Skylake-Client-v1.xml
b/src/cpu_map/x86_Skylake-Client-v1.xml
> new file mode 100644
> index 0000000000..31e332b371
> --- /dev/null
> +++ b/src/cpu_map/x86_Skylake-Client-v1.xml
> @@ -0,0 +1,6 @@
> +<cpus>
> + <model name='Skylake-Client-v1'>
> + <decode host='off' guest='off'/>
> + <model name='Skylake-Client'/>
> + </model>
> +</cpus>
Here you've effectively setup a perfect alias.
> diff --git a/src/cpu_map/x86_Skylake-Client-v2.xml
b/src/cpu_map/x86_Skylake-Client-v2.xml
> new file mode 100644
> index 0000000000..53318d4d9e
> --- /dev/null
> +++ b/src/cpu_map/x86_Skylake-Client-v2.xml
> @@ -0,0 +1,141 @@
> +<cpus>
> + <model name='Skylake-Client-v2'>
> + <decode host='on' guest='on'/>
> + <signature family='6' model='94'/>
> + <signature family='6' model='78'/>
> + <signature family='6' model='142'/>
> + <signature family='6' model='158'/>
> + <vendor name='Intel'/>
> + <feature name='3dnowprefetch'/>
> + <feature name='abm'/>
> + <feature name='adx'/>
snip
here though you're fully re-defining Skylake-Client-IBRS, although
not quite perfectly matching as 2 features are missing from -v2
that we did have in -IBRS:
<feature name='mpx' removed='yes'/>
<feature name='vmx-invvpid'/>
also the existing -IBRS had added='yes' for all the vmx
features.
Right, that's what I mentioned in the cover letter. The versions > 1
were not made the same as their non versioned aliases if they exist
since it made sense to me from one point of view asking for other
opinions.
IMHO the inconsistent handling of aliases between -v1 and the non-v1
models is undesirable.
OK, no problem I can go with the consistent option :-)
Jirka