On 3/30/15, 5:38 PM, Scott Feldman wrote:

On Mon, Mar 30, 2015 at 5:22 PM, Arad, Ronen [off-list ref] wrote:

quoted

quoted

-----Original Message-----
From: netdev-owner@vger.kernel.org [mailto:netdev-owner@vger.kernel.org] On
Behalf Of sfeldma@gmail.com
Sent: Monday, March 30, 2015 1:40 AM
To: netdev@vger.kernel.org
Cc: jiri@resnulli.us; roopa@cumulusnetworks.com; linux@roeck-us.net;
f.fainelli@gmail.com
Subject: [PATCH net-next 02/18] switchdev: flesh out get/set attr ops

From: Scott Feldman <redacted>

[cut]

quoted

int swdev_port_attr_set(struct net_device *dev, struct swdev_attr *attr)
{
-      return -EOPNOTSUPP;
+      struct swdev_attr prev = *attr;
+      int err, get_err;
+
+      get_err = swdev_port_attr_get(dev, &prev);
+
+      err = _swdev_port_attr_set(dev, attr);
+      if (err && !get_err && !(attr->flags & SWDEV_ATTR_F_NO_RECOVER))
+              /* Some err on set: revert to previous value */
+              _swdev_port_attr_set(dev, &prev);

Netlink requests could contain multiple attributes within a single request.
Reverting to the previous value applies only to the first
swdev_port_attr_set error. It does not rollback all prior changes that
were triggered by the same Netlink request.

Since attr_set scope is a single attr across ports, it can only revert
that attr for those ports.  So rewinding multiple attributes needs to
happen above attr_set.  I guess we could have a multi_attr_set that
takes an array of *attrs, but something like this can be built on top
of attr_set.  Maybe baby steps is best first so we can see how things
shape up.

Having a multi attr set/del is necessary in some cases where the hw api 
may be able to handle more than one attr set/del. I can see this 
especially in the vlan case. Having the driver get all attributes and do 
the necessary optimizations to program hardware might be better.

thanks,
Roopa