On September 9, 2019 2:56 pm, Stefan Reiter wrote:
> On 9/6/19 1:42 PM, Fabian Grünbichler wrote:
>>> +# this is neither the default for 'query_supported_cpu_flags' below, nor
>>> for
>>> +# /proc/cpuinfo.
>>> +#
>>> +# To compare (or array_intersect) flags, it's a good idea to convert them
>>> all
>>> +# to a common format first (e.g. map s/\.|-/_/g).
>>> +sub query_understood_cpu_flags {
>>
>> not used anywhere in this series? also, confusingly, this returns less
>> values then the other helper (maybe because of some aliases?), and it's
>> not a strict subset :-/
>>
>
> Yes, this is for exposing via the API (for the GUI to create/edit custom
> models I have in mind). I just thought it fits with this patch well,
> since the comments refer to each other.
>
> Also, query_understood_cpu_flags returns 188 flags for me, while
> query_supported_cpu_flags only returns 104...
yes, a double check confirms this - seems I had the output files mixed
up.. sorry for the noise
>
> The discrepancies generally arise because
> a) query_understood_cpu_flags returns flags the host cannot use and
> b) query_supported_cpu_flags (rather the QMP call) doesn't actually
> return CPU flags, but CPU settings - with most of them being flags.
> Those settings (and some flags, curiously) cannot be specified as a
> "-cpu" argument however.
that makes sense - and would make a good starting point for an
informative comment ;)
>
> The intersection of those provides only flags that can be passed to
> "-cpu" without issues in my testing. Since query_understood_cpu_flags is
> not host specific though, it doesn't need broadcasting.
_______________________________________________
pve-devel mailing list
pve-devel@pve.proxmox.com
https://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
