TCP zero-copy

TCP zero-copy

[hochakhung]

Is there currently a stable implementation for zero copy on TCP stack for linux2.4? Would anyone please point me to the patch if there is any?
Thanks a lot

Steven


__________________________________________________________________
Your favorite stores, helpful shopping tools and great gift ideas. Experience the convenience of buying online with Shop@Netscape! http://shopnow.netscape.com/

Get your own FREE, personal Netscape Mail account today at http://webmail.netscape.com/

Re: TCP zero-copy

[simon]

I believe this behaviour is only possible for cards which have the 
ability to dma what appears as a list of skb data fragments and 
also has the ability to checksum the data. Namely the Alteon, now 
3com, ACENIC and the sun hme device. 

Does this kernel modification completely remove the need for a 
copy/checksum of the data between user and kernel space on both
transmit and receive ?

I have written a driver for an Intel ixf1002 chip, which has some 
surrounding HW, and is capable of checksumming and processing 
dma in fragments. Is there any information on what changes I 
would have to make to the driver to support zerocopy/checksum ?  

Many Thanks

Simon.


> 
> hochakhung@netscape.net writes:
>  > Is there currently a stable implementation for zero copy on TCP
>  > stack for linux2.4? Would anyone please point me to the patch if
>  > there is any?  Thanks a lot
> 
> It is in the standard 2.4.x kernels these days, no patch is necessary.
> 
> Later,
> David S. Miller
> davem@redhat.com
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/


__________________________

Simon Haynes - Baydel 
Phone : 44 (0) 1372 378811
Email : simon@baydel.com
__________________________

Re: TCP zero-copy

[Andrew Theurer]

On Friday 03 August 2001 04:23,  wrote:
> I believe this behaviour is only possible for cards which have the
> ability to dma what appears as a list of skb data fragments and
> also has the ability to checksum the data. Namely the Alteon, now
> 3com, ACENIC and the sun hme device.

Intel now has an e1000 driver as well.

> Does this kernel modification completely remove the need for a
> copy/checksum of the data between user and kernel space on both
> transmit and receive ?

Transmit for sure.  You should no longer see csum_partial_copy_generic() when 
using zerocopy.

> I have written a driver for an Intel ixf1002 chip, which has some
> surrounding HW, and is capable of checksumming and processing
> dma in fragments. Is there any information on what changes I
> would have to make to the driver to support zerocopy/checksum ?

Which card is this?

Andrew Theurer

Re: TCP zero-copy

[Andi Kleen]

In article <3B6A7B96.1591.241743@localhost>,
"" <simon@baydel.com> writes:

> I have written a driver for an Intel ixf1002 chip, which has some 
> surrounding HW, and is capable of checksumming and processing 
> dma in fragments. Is there any information on what changes I 
> would have to make to the driver to support zerocopy/checksum ?  

The driver has to set dev->features to its checksum  capabilities (see skbuff.h
for more details) and implement them. It should also be able to process
skbs with multiple skb fragments in hard_start_xmit for true zero copy;
in this case set NETIF_F_SG and also NETIF_F_HIGHDMA if it supports 64bit
addresses. When NETIF_F_SG and NETIF_F_IP_CSUM or NETIFI_F_HW_CSUM are set
the stack will do zero copy IO on TX with sendfile.

-Andi


-Andi