Gateworks 2348 and ath problem

[John Hay]

On Thu, Nov 30, 2006 at 08:09:53AM -0700, M. Warner Losh wrote:
> In message: <20061130142146.GA50022 at zibbi.meraka.csir.co.za>
>             John Hay <jhay at meraka.org.za> writes:
> : > 
> : > Ok, I built a kernel without all those options, but it didn't make a
> : > difference. I also tried sta mode and it still looks the same. Some
> : > new info that I have is that when I do a tcpdump on the arm box itself,
> : > adding -y IEEE802_11_RADIO makes a difference. Without it the outgoing
> : > packets looks ok, but with it 2 bytes are added between the ether (802.11)
> : > header and the ip(v6) header.
> : 
> : Ok, I found the problem. Using LLC_SNAPFRAMELEN works better than
> : sizeof(struct llc). LLC_SNAPFRAMELEN is defined as 8 and sizeof()
> : returns 10. Doing a grep in net/ if see that LLC_SNAPFRAMELEN is used
> : a lot, so I guess this is the correct way go. I nobody has anything
> : against it, I can commit it.
> 
> NetBSD marks it as __packed.  That might be a safer patch.  NetBSD
> only uses LLC_SNAPFRAMELEN in token ring code.

Well struct llc is marked as __packed. See sys/net/if_llc.h. There is
a union llc_un inside it that isn't and I have tried with it marked
as __packed, but it didn't help.

> It might be even better if we could catch these things at compile
> time, but doing so may be a bit of a pita to setup (how do you know
> all the relevant types to check).

John
-- 
John Hay -- John.Hay at meraka.csir.co.za / jhay at FreeBSD.org