On Thursday 6 April 2006 12:48, Dirk Rieger wrote:
>
> any help/hints would be welcome on how to debug this...

I will do my best to help you spotting the source of this problem but I need 
your collaboration because I don't have a big-endian machine on which to do 
debugging, I have to count on your troubleshooting skills :)

> inside labd/input.d - lapd_rcv
>
> the x86 receives sk_buf->data with 48+1 byte length
> the ppc receives sk_buf->data with 512+1 byte length (the 'x86-message' is
> somewhere in the middle)

Okay, this is a good step forward, now we know that the corruption happens 
under the protocol stack.

With "length" you mean skb->len, don't you?

I would put the hfc-pci driver in debug_mode with level >= 3, you should be 
able to see the frames being received, monitor them during the path to 
visdn-netdev module and try to spot where skb->len is set to 512.

Oops... I see that this has already been done by "emotional outburst":

Apr  8 21:50:59 localhost kernel: hfc-pci: pci-0000:00:0d.0:st:chan[D] 
RX (f1=11, f2=10, z1=6001, z2=0002) len 512: 
00000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000ff0308010105a104038090a31801896c0c2183313737323734373135317009c135333238333039367d029181a9fe0002ff0308010105a104038090a31801896c0c2183313737323734373135317009c135333238333039367d029181a9fe000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000
Apr  8 21:50:59 localhost kernel: 
00000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000

Uhm... it seems that the Z1/Z2 pointers are not correctly accessed in 
memory... let me see... it appears that there is an endianity problem when 
accessing the Z(F) counters. I just committed a patch that should solve that 
problem. Of course there may be other bugs related to endianity/unaligned 
accesses but hopefully we've moved on :)

Bye,

-- 
  Daniele Orlandi
_______________________________________________
Visdn-hackers mailing list
[email protected]
https://mailman.uli.it/mailman/listinfo/visdn-hackers

Reply via email to