Re: [Patch 2/2] mlx4: add dynamic LRO disable support

[Cong Wang <amwang@redhat.com>]

On 06/09/10 18:49, Stanislaw Gruszka wrote:
> Hi Amerigo
>
> Sorry for being silent in this thread before.
>
> On Wed, Jun 09, 2010 at 05:23:35PM +0800, Cong Wang wrote:
>>>>> Is that flag test actually unsafe - and if so, how is testing num_lro
>>>>> any better?  Perhaps access to net_device::features should be wrapped
>>>>> with ACCESS_ONCE() to ensure that reads and writes are atomic.
>>>>>
>>>>
>>>> At least, I don't find there is any race with 'num_lro', thus
>>>> no lock is needed.
>>>
>>> In both cases there is a race condition but it is harmless so long as
>>> the read and the write are atomic.  There is a general assumption in
>>> networking code that this is the case for int and long.  Personally I
>>> would prefer to see this made explicit using ACCESS_ONCE(), but I don't
>>> see any specific problem in mlx4 (not that I'm familiar with this driver
>>> either).
>>
>> I read this email again.
>>
>> I think you misunderstood the race condition here. Even read and write
>> are atomic here, the race still exists. One can just set NETIF_F_LRO
>> asynchronously right before mlx4 check this flag in mlx4_en_process_rx_cq()
>> which doesn't take rtnl_lock.
>
> If so, it's better to stop device before modify LRO settings. I suggest
> something like that in mlx4_ethtool_op_set_flags:
>
> if (!!(data&  ETH_FLAG_LRO) != !!(dev->features&  NETIF_F_LRO)) {


What does this line mean? This is to ignore all other flags, right?

> 	/* Need to toggle LRO */
>
> 	if (netdev_running(dev)) {
>                 mutex_lock(&mdev->state_lock);
>                 mlx4_en_stop_port(dev);
>                 rc = mlx4_en_start_port(dev);
>                 if (rc)
>                         en_err(priv, "Failed to restart port\n");
> 	}
>
> 	dev->features ^= NETIF_F_LRO;
>
> 	if (netdev_running(dev))
>                 mutex_unlock(&mdev->state_lock);
> }
>

I don't think mdev->state_lock is used to protect dev->feature.
rtnl_lock is. I think switching to mlx4_ethtool_op_set_flags()
from the default one has already solved this.

Thanks!