>> core: earlier software capability detection
   

     /* properties */
+    g_object_class_install_property
+         (object_class, PROP_IS_PARTITION,
+          g_param_spec_boolean (NM_DEVICE_INFINIBAND_IS_PARTITION, "", "",
+                                FALSE,
+                                G_PARAM_READWRITE | G_PARAM_CONSTRUCT |
+                                G_PARAM_STATIC_STRINGS));
 

CONSTRUCT_ONLY




>> platform: move InfiniBand property reading into the platform


infiniband_get_info() does more then just reading sysctl. Nice. But I think you should do:

»···if (nm_rtnl_link_parse_info_data (priv->nlh,
»···                                  ifindex,
»···                                  infiniband_info_data_parser,
»···                                  &info) != 0) {
                                            ^^^^^



>> settings: shortcut nm_settings_get_connection_by_uuid() if @uuid is invalid
    
I think this patch should be dropped. When I changed nm_utils_uuid_generate() to create valid UUIDs I had patches to enforce valid UUIDs. We (rightly) agreed,
that we accept basically every string. But then this patch is wrong.



Rest LGTM

(In reply to Thomas Haller from comment #1)
> >> core: earlier software capability detection
>    
> 
>      /* properties */
> +    g_object_class_install_property
> +         (object_class, PROP_IS_PARTITION,
> +          g_param_spec_boolean (NM_DEVICE_INFINIBAND_IS_PARTITION, "", "",
> +                                FALSE,
> +                                G_PARAM_READWRITE | G_PARAM_CONSTRUCT |
> +                                G_PARAM_STATIC_STRINGS));
>  
> 
> CONSTRUCT_ONLY

Fixed.

> >> platform: move InfiniBand property reading into the platform
> 
> 
> infiniband_get_info() does more then just reading sysctl. Nice. But I think
> you should do:
> 
> »···if (nm_rtnl_link_parse_info_data (priv->nlh,
> »···                                  ifindex,
> »···                                  infiniband_info_data_parser,
> »···                                  &info) != 0) {

Fixed.

> >> settings: shortcut nm_settings_get_connection_by_uuid() if @uuid is invalid
>     
> I think this patch should be dropped. When I changed
> nm_utils_uuid_generate() to create valid UUIDs I had patches to enforce
> valid UUIDs. We (rightly) agreed,
> that we accept basically every string. But then this patch is wrong.

Dropped.

Repushed!

LGTM

Thanks, merged to master.