netpoll bug - kgdboe on x86_64

netpoll bug - kgdboe on x86_64

[Jim Houston]

Hi Matt,

I'm trying to get kgdboe working on x86_64.  I noticed that
netpoll_rx is calling the rx_hook with negative values for the length.
The attached patch fixes the problem. 

Jim Houston - Concurrent Computer Corp.

---

--- 2.6.1-rc1-mm2.orig/net/core/netpoll.c	2004-01-05 13:15:31.000000000 -0500
+++ 2.6.1-rc1-mm2/net/core/netpoll.c	2004-01-13 18:58:09.311479928 -0500
@@ -400,7 +400,7 @@ int netpoll_rx(struct sk_buff *skb)
 
 		if (np->rx_hook)
 			np->rx_hook(np, ntohs(uh->source),
-				    (char *)(uh+1), ulen-sizeof(uh)-4);
+				    (char *)(uh+1), ulen-sizeof(struct udphdr));
 
 		return 1;
 	}

Re: netpoll bug - kgdboe on x86_64

[Pavel Machek]

Hi!

[Hmm, nice return address, Jim]

> I'm trying to get kgdboe working on x86_64.  I noticed that
> netpoll_rx is calling the rx_hook with negative values for the length.
> The attached patch fixes the problem. 

I was able to make -mm stuff work on i386, and there were not any
strange problems. (Hope this encourages you :-)



> Jim Houston - Concurrent Computer Corp.
> 
> ---
> 
> --- 2.6.1-rc1-mm2.orig/net/core/netpoll.c	2004-01-05 13:15:31.000000000 -0500
> +++ 2.6.1-rc1-mm2/net/core/netpoll.c	2004-01-13 18:58:09.311479928 -0500
> @@ -400,7 +400,7 @@ int netpoll_rx(struct sk_buff *skb)
>  
>  		if (np->rx_hook)
>  			np->rx_hook(np, ntohs(uh->source),
> -				    (char *)(uh+1), ulen-sizeof(uh)-4);
> +				    (char *)(uh+1), ulen-sizeof(struct udphdr));
>  
>  		return 1;
>  	}

-- 
Horseback riding is like software...
...vgf orggre jura vgf serr.

Re: netpoll bug - kgdboe on x86_64

[Matt Mackall]

On Tue, Jan 13, 2004 at 07:18:30PM -0500, Jim Houston wrote:
> 
> Hi Matt,

Jim, fix your email address, you sent that as Jim Houston <jhouston@new.localdomain>.

> I'm trying to get kgdboe working on x86_64.  I noticed that
> netpoll_rx is calling the rx_hook with negative values for the length.
> The attached patch fixes the problem. 

This patch looks correct. I can't recall what I was thinking when I
tossed the -4 in there. It looks suspiciously like I did it to
mindlessly compensate for a bug where I took the sizeof a pointer
rather than a type, but I would _never_ do that. I'll send this on to
jgarzik who's queueing netpoll stuff for me.

> Jim Houston - Concurrent Computer Corp.
> 
> 
> +++ 2.6.1-rc1-mm2/net/core/netpoll.c	2004-01-13 18:58:09.311479928 -0500
> @@ -400,7 +400,7 @@ int netpoll_rx(struct sk_buff *skb)
>  
>  		if (np->rx_hook)
>  			np->rx_hook(np, ntohs(uh->source),
> -				    (char *)(uh+1), ulen-sizeof(uh)-4);
> +				    (char *)(uh+1), ulen-sizeof(struct udphdr));
>  
>  		return 1;
>  	}

-- 
Matt Mackall : http://www.selenic.com : Linux development and consulting