Re: Kernel memory leak in bnx2x driver with vxlan tunnel

On 01/20/16 at 04:34pm, Eric Dumazet wrote:

On Wed, 2016-01-20 at 16:19 -0800, Jesse Gross wrote:

quoted

I have a patch that implements the comparison between dsts. For
packets without a dst, there isn't really a cost and if we do have a
dst then GRO is still a benefit. So it seems like it is worth doing,
even if it is more expensive than is ideal.

You guys really want to kill GRO performance.

Really the aggregation should happen at the first layer (ethernet
device), instead of doing it after tunnel decap.

This is already done for GRE, IPIP, SIT, ...

GRO having to access metadata looks wrong, if you think about trying to
do the same function in hardware (offload)

If I understand Jesse correctly then the added cost is only for
metadata enabled packets. Though I agree, what's the benefit of
doing GRO after decap?

It seems like it's way too late and we've already paid the cost
by going through the stack for each outer header packet.