Re: zd1211rw (2.6.26 sparc64): unaligned access (zd_mac_rx)

[Shaddy Baddah <shaddy_baddah@hotmail.com>]

Hi,

On 09/11/08 19:56, Johannes Berg wrote:
> Yes, I think you're right, it appears that the 802.11 header isn't
> aligned on a 2-byte boundary, which is extremely strange. Can you, with
> your patch applied, do something like
> 
> 	printk(KERN_DEBUG "%p\n", skb->data);
> 
> before this code in zd_mac.c:
> 
>         memcpy(skb_put(skb, length), buffer, length);
> 
>         ieee80211_rx_irqsafe(hw, skb, &stats);

I have done this... but I think that there is too much output to place 
here. What I can tell you with confidence is that every value is 4-byte 
aligned.

However, I modify filter_ack() in zd_mac.c so that the code looks like this:

         for (skb = q->next; skb != (struct sk_buff *)q; skb = skb->next) {
                 struct ieee80211_hdr *tx_hdr;

                 tx_hdr = (struct ieee80211_hdr *)skb->data;
         printk(KERN_DEBUG "%s:%u: skb %p skb->data %p tx_hdr->addr2 %p 
rx_hdr->addr1 %p\n", __FILE__, __LINE__, skb, skb->data, tx_hdr->addr2, 
rx_hdr->addr1);
                 //if (likely(!memcmp(tx_hdr->addr2, rx_hdr->addr1, 
ETH_ALEN)))
                 if (likely(!compare_ether_addr(tx_hdr->addr2, 
rx_hdr->addr1)))
                 {

(note, I re-instated the compare_ether_addr() so that I could be sure 
that the unaligned access corresponded with whatever values I was seeing 
in the print).

Here is the output of the first print from this location that I see in 
syslog:

Nov 10 00:47:17 trad kernel: [ 8239.534950] 
drivers/net/wireless/zd1211rw/zd_mac.c:639: skb fffff8003d0d8fc0 
skb->data fffff8003d874cdb tx_hdr->addr2 fffff8003d874ce5 rx_hdr->addr1 
fffff8003f3e0009
Nov 10 00:47:17 trad kernel: [ 8239.534986] Kernel unaligned access at 
TPC[100f5bd0] zd_mac_rx+0x1b0/0x398 [zd1211rw]
Nov 10 00:47:17 trad kernel: [ 8239.637697] Kernel unaligned access at 
TPC[100f5bd4] zd_mac_rx+0x1b4/0x398 [zd1211rw]
Nov 10 00:47:17 trad kernel: [ 8239.740208] Kernel unaligned access at 
TPC[100f5bd8] zd_mac_rx+0x1b8/0x398 [zd1211rw]
Nov 10 00:47:17 trad kernel: [ 8239.842417] Kernel unaligned access at 
TPC[100f5bdc] zd_mac_rx+0x1bc/0x398 [zd1211rw]
Nov 10 00:47:17 trad kernel: [ 8239.944583] Kernel unaligned access at 
TPC[100f5be0] zd_mac_rx+0x1c0/0x398 [zd1211rw]

So you can see that skb->data is not even 2-byte aligned. And my debug 
leads me to believe that the problem is this line in zd_mac_tx_to_dev():

                  skb_pull(skb, sizeof(struct zd_ctrlset));

A before and after of this line gives me this output:

Nov 10 00:47:17 trad kernel: [ 8239.534065] 
drivers/net/wireless/zd1211rw/zd_mac.c:376: skb fffff8003d0d8fc0 
skb->data fffff8003d874cd0
Nov 10 00:47:17 trad kernel: [ 8239.534092] 
drivers/net/wireless/zd1211rw/zd_mac.c:390: skb fffff8003d0d8fc0 
skb->data fffff8003d874cdb

Perhaps there needs to be some padding there?

Also, I've just had a look at the git version of zd_mac.c, and I notice 
it is all change there again. I would feel uncomfortable having studied 
the 2.6.26 code to have to start again. Is it expected that I use the 
latest available via git?

Thanks in advance,
Shaddy