Re: ath5k AP kernel panic when client uses SCP

[Tomasz Chmielewski <mangoo@wpkg.org>]

Bob Copeland wrote:

>> [  516.430000] ------------[ cut here ]------------
>> [  516.430000] WARNING: at net/core/dev.c:1566 skb_gso_segment+0x110/0x298()
>> [  516.440000] b44: caps=(0x0, 0x0) len=52 data_len=0 ip_summed=1
> 
> So this looks like the ethernet driver b44 instead of ath5k, I think.

Could be, but!

I can trigger it when doing this kind of SSH traffic:

PC >--100Mbit--> b44 (Asus WL-500gP) ath5k >--wireless--> laptop


When I transfer over wired ethernet only through this device, everything 
works properly, no page allocation failures, no warnings, panics.


>> [  516.450000] Modules linked in: configs aes_generic tun sch_sfq cls_fw sch_htb ipt_MASQUERADE iptable_nat nf_nat xt_MARK iptable_mangle ipt_ULOG xt_recent nf_conn1
>> [  516.480000] Call Trace:
>> [  516.480000] [<80013ac4>] dump_stack+0x8/0x34
>> [  516.480000] [<8002f2a0>] warn_slowpath_common+0x70/0x98
>> [  516.490000] [<8002f308>] warn_slowpath_fmt+0x24/0x30
>> [  516.490000] [<801f0398>] skb_gso_segment+0x110/0x298
> 
> ...but this is higher up the stack.  skb_gso_segment is about a year old
> so it would be odd if there were a bug here.  Can you do the following?
> 
> objdump -S net/core/dev.o, then find the address for skb_gso_segment, 
> it should look something like this (your numbers and disassembly will
> differ):
> 
> 00004190 <skb_gso_segment>:
>  *
>  *      It may return NULL if the skb requires no segmentation.  This is
>  *      only possible when GSO is used for verifying header integrity.
>  */

Hmm, I get:

00000000 <skb_gso_segment>

So I better leave the files here:

http://www1.wpkg.org/skb_gso_segment/

Let me know if you need anything else.


-- 
Tomasz Chmielewski
http://wpkg.org