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

[Alexander Duyck <alexander.h.duyck@redhat.com>]

On 10/15/2014 10:15 PM, Jiafei.Pan@freescale.com wrote:
>> -----Original Message-----
>> From: Eric Dumazet [mailto:eric.dumazet@gmail.com]
>> Sent: Thursday, October 16, 2014 12:15 PM
>> To: Pan Jiafei-B37022
>> Cc: David Miller; jkosina@suse.cz; netdev@vger.kernel.org; Li Yang-Leo-R58472;
>> linux-doc@vger.kernel.org
>> Subject: Re: [PATCH] net: use hardware buffer pool to allocate skb
>>
>> On Thu, 2014-10-16 at 02:17 +0000, Jiafei.Pan@freescale.com wrote:
>>
>>> Thanks for your comments and suggestion. In my case, I want to build skb
>>> from hardware block specified memory, I only can see two ways, one is modified
>>> net card driver replace common skb allocation function with my specially
>>> functions, another way is to hack common skb allocation function in which
>>> redirect to my specially functions. My patch is just for the second way.
>>> Except these two ways, would you please give me some advice for some other
>>> ways for my case? Thanks
>> I suggest you read drivers/net/ethernet numerous examples.
>>
>> No need to change anything  in net/* or include/*, really.
>>
>> For a start, look at drivers/net/ethernet/intel/igb/igb_main.c
>>
>> Mentioning 'hack' in your mails simply should hint you are doing
>> something very wrong.
>>
>> What makes you think your hardware is so special ?
>>
> In fact, I am developing a bridge driver, it can bridge between any other the
> third party net card and my own net card. My target is to let any other the
> third party net card can directly use my own net card specified buffer, then
> there will be no memory copy in the whole bridge process.
> By the way, I don’t see any similar between igb_main.c and my case. And also
> My bridge also can’t implemented with "skb frag" in order to aim at zero memory
> copy.

I think the part you are not getting is that is how buffers are 
essentially handled now.  So for example in the case if igb the only 
part we have copied out is usually the header, or the entire frame in 
the case of small packets.  This has to happen in order to allow for 
changes to the header for routing and such.  Beyond that the frags that 
are passed are the buffers that igb is still holding onto.  So 
effectively what the other device transmits in a bridging/routing 
scenario is my own net card specified buffer plus the copied/modified 
header.

For a brief period igb used build_skb but that isn't valid on most 
systems as memory mapped for a device can be overwritten if the page is 
unmapped resulting in any changes to the header for routing/bridging 
purposes being invalidated.  Thus we cannot use the buffers for both the 
skb->data header which may be changed and Rx DMA simultaneously.

Thanks,

Alex