Re: [RFC PATCH net-next v2 10/17] ethtool: implement GET_SETTINGS message

On Tue, Aug 21, 2018 at 11:32:46AM +0200, Michal Kubecek wrote:

On Mon, Jul 30, 2018 at 08:54:55PM +0200, Andrew Lunn wrote:

quoted

quoted

+/* Internal kernel helper to query a device ethtool_link_settings.
+ *
+ * Backward compatibility note: for compatibility with legacy drivers
+ * that implement only the ethtool_cmd API, this has to work with both
+ * drivers implementing get_link_ksettings API and drivers
+ * implementing get_settings API. When drivers implement get_settings
+ * and report ethtool_cmd deprecated fields
+ * (transceiver/maxrxpkt/maxtxpkt), these fields are silently ignored
+ * because the resulting struct ethtool_link_settings does not report them.

~/linux/drivers$ grep -r [.]get_settings *
net/ethernet/8390/etherh.c:		 .get_settings	= etherh_get_settings,

I don't think it is worth adding support for .get_settings for just
one driver. It is better to just convert that driver to the new API.

I have prepared a patch converting 8390/etherh driver to use
{g,s}et_link_ksettings and I'm going to submit it when net-next opens.
Do you think we can then drop {g,s}et_settings callbacks completely
(i.e. also from ioctl() code and ethtool_ops)?  Do we care about
unconverted out of tree drivers?

Hi Michal

We cannot break ethtool, the ABI it uses. But there is already code to
use get_link_ksettings() and only fall back to get_settings if it does
not exist. So we can clean up all the fallback code, remove the
ethtool_ops, etc.

I personally don't care about out of tree drivers. They have had over
2 years to change to the new API.

	     Andrew