Xen devel - Mar 2008 - question on large send offload

Does the data for a large send offload have to occupy more than one
page? In tuning the windows drivers, if I set the minimum mss multiple
before considering offload (a windows variable) to 2, then I seem to get
hangs. If I set it to 4 then everything appears to be okay. 

If the mss is 1460 (normal for Ethernet), then 2 * 1460 = 2920, which
will often fit into one page, in which case the ring looks like:

0: tx_request - flags |= NETTXF_extra_info
1: extra_info

If the 2920 bytes were on more than one page, then the ring would look
like:

0: tx_request - flags |= NETTXF_extra_info | NETTXF_more_data
1: extra_info
2: tx_request - no change to flags

The behaviour I''m seeing is that I''ll get a bunch of packets
working
just fine, then it will pause for a bit, and I''m wondering if I could
be
getting a bunch of the latter above case, and then one of the former
which causes the hang. I haven''t yet put debug statements in to find
out
which of the above is the packet before the hang... I''ll do that
tomorrow.

Any suggestions?

Thanks

James

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On Sat, Mar 08, 2008 at 11:01:36PM +1100, James Harper
wrote:
> Does the data for a large send offload have to occupy more than one
> page? In tuning the windows drivers, if I set the minimum mss multiple
> before considering offload (a windows variable) to 2, then I seem to get
> hangs. If I set it to 4 then everything appears to be okay. 
> 
> If the mss is 1460 (normal for Ethernet), then 2 * 1460 = 2920, which
> will often fit into one page, in which case the ring looks like:
> 
> 0: tx_request - flags |= NETTXF_extra_info
> 1: extra_info
> 
> If the 2920 bytes were on more than one page, then the ring would look
> like:
> 
> 0: tx_request - flags |= NETTXF_extra_info | NETTXF_more_data
> 1: extra_info
> 2: tx_request - no change to flags
> 
> The behaviour I''m seeing is that I''ll get a bunch of
packets working
> just fine, then it will pause for a bit, and I''m wondering if I
could be
> getting a bunch of the latter above case, and then one of the former
> which causes the hang. I haven''t yet put debug statements in to
find out
> which of the above is the packet before the hang... I''ll do that
> tomorrow.
> 
> Any suggestions?
> 
Hmm.. might be a stupid suggestion/question, but have you checked the
unmodified (pv-on-hvm) drivers for Linux and how they do that? 

-- Pasi

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

> 
> Hmm.. might be a stupid suggestion/question, but have you checked the
> unmodified (pv-on-hvm) drivers for Linux and how they do that?
> 
The stupidity was on my part. I was setting feature-gso-tcpv4 to 0, not 1!

James

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel