Re: [PATCH] net: use hardware buffer pool to allocate skb

[Alexander Duyck <alexander.duyck@gmail.com>]

On 10/17/2014 07:05 AM, Eric Dumazet wrote:
> On Fri, 2014-10-17 at 02:35 +0000, Jiafei.Pan@freescale.com wrote:
>
>> [Pan Jiafei] Hi, Alex, thanks for your comments. I don’t confirm that
>> you have catch my concerns. For example, I want to add igb net card 
>> into my bridge, and want to igb net driver allocate skb by using
>> my specified memory address, but I don’t want to modify igb net driver
>> directly, how to do this in my bridge drivers?
> This is exactly our point : We do not want to modify all drivers so that
> your bridge is happy with them.
>
> You'll have to make your bridge using standard infra instead.
>
> IGB has no way to know in advance that a particular frame should
> eventually reach your bridge anyway.

Also why is it igb should use your buffers, is there any reason why your
device cannot use the receive buffers that are handed off to the stack
from igb?  It isn't as if there is a copy in the routing or bridging
path.  The receive buffer is normally handed off to the stack from the
ingress device, a few headers might get tweaked, and then that buffer is
transmitted by the egress interface.  Only in the case of a buffer being
handed to multiple egress devices or user space should it ever be copied.

Thanks,

Alex