Xen devel - Sep 2007 - [Xen-3.1.0] Network transmission loss between dom0 and domU

Hi All,

I am facing a problem related to network transmission between dom0 and
PV domU in Xen 3.1.0.

here is a brief summary of what i am doing -

- I want to send some custom header from domU to network driver in
dom0.As the frontend and backend communicate using shared ring buffers
which employ use of grant table mechanism,I can only put that custom
header along with the data in the skb''s data.
- I need this custom header in network driver in dom0 and thus get
some information appended to it in the domU.


here is brief summary of how i am doing this -

- i have allocated extra space for my custom header in the alloc_skb
at the tcp layer itself. After the data is copied to the skb i
skb_push and then memcpy the header to the data part of the skb.
- This skb goes through the netfront driver in domU and is treated
like any other skb as it does not care about the data part[not even
the checksum] and sends it to the netback driver in dom0.
- Once this is received by the dom0 it is forwarded to the bridge
xenbrX in the dom0.
- bridge sends it to the correct network driver in dom0.
- In driver i am doing a skb_pull and get[copy actually] the custom
header out from the skb->data part.
- Once i have done this, i switch to another mode of operation for the
NIC depending on the custom header value i have.
- At the reception side the procedure is exactly reverse of the above
mentioned insanity. :-/.


here is what happens with a simple client and server program running
in userspace -

- One way transmission on data from either client to server is working
without any hiccups.


Problem and its cause -

- If i modify the client server program for simultaneous transmission.
Client sends data, server sends data again as part of the program.
- Running this program the server and client they both hang
indefinitely after random packets are transmitted between domU on
machine one and other client machine.
- After putting a lots of printks i came to this conclusion.
 1. netif_int in netfront is not getting invoked just before the domU
seems to be hanged. As a result the receive socket queue sleeps
indefinitely waiting for someone to come and wake it up.
2. Further inquiry helped me to find out that the EVTCHNOP_send event
is not getting transmitted to domU at all.The event is a valid one
indeed[i.e it is not zero]. It seems like either it is lost in the Xen
hypervisor or domU is ignoring the event sent through eventchannel
through hypervisor.

So, what are the cases when domU can not respond to the event passed
to it by the hypervisor?As i can see at it, either interrupts are
disabled in the domU at that unfateful moment or it is not getting
delivered at all.

Can someone point me what i am doing wrong? I bet i am messing
something somewhere. Any help?

sorry for a tedious mail. Couldn''t explain it better.

Thanks

--
pradeep singh rautela

"question = ( to ) ? be : ! be;"
                -- Wm. Shakespeare

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