netdev - [PATCH] Re: kernel BUG in eth_alloc_tx_desc_index at drivers/net/mv643xx

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20070111104220.GA3171@ff.dom.local>
Date:	Thu, 11 Jan 2007 11:42:20 +0100
From:	Jarek Poplawski <jarkao2@...pl>
To:	Thibaut VARENE <T-Bone@...isc-linux.org>
Cc:	Dale Farnsworth <dale@...nsworth.org>, netdev@...r.kernel.org,
	mlachwani@...sta.com
Subject: [PATCH] Re: kernel BUG in eth_alloc_tx_desc_index at drivers/net/mv643xx_eth.c:1069!

On Wed, Jan 10, 2007 at 06:12:29PM +0100, Thibaut VARENE wrote:
> On 1/9/07, Thibaut VARENE <T-Bone@...isc-linux.org> wrote:
> >On 1/9/07, Dale Farnsworth <dale@...nsworth.org> wrote:
> >>
> >> Thank you Thibaut.  Please try the following patch:
> >>
> >> From: Dale Farnsworth <dale@...nsworth.org>
> >>
> >> Reserve one unused descriptor in the TX ring
> >> to facilitate testing for when the ring is full.
> >
> >Dale,
> >
> >tried it and unfortunately:
> 
> Also, I don't know if you read that bit, but everytime I reboot the
> box immediately after a crash, the NIC gets a bogus (always the same
> it seems) MAC address, and I have to reboot one more time to get back
> to the "normal" MAC address.
> 
> Dunno if that hints anything though.

There is something in the code about MAC writing and saving
some config during initialization, so probably it's possible
if reinitialization was broken.

I tried to look more into the code and here are my (maybe
wrong) conclusions:

- It looks like something could be broken during tx descs
freeing or eth_tx_timeout_task. I compared the timeout code
with e100 and tg3 and have a feeling mv643xx_eth is doing 
less but I'm not able to estimate the importance of this.

- Such errors, IMHO, could be possible with races and not
enough locking, and btw. I think suspected function isn't
properly locked: mp->tx_desc_count in while condition isn't
protected at all. Below I attach a patch proposal but I'm
not sure some irq off or spin_lock isn't also needed
elswere. If it's only locking it would be suitable to do
the test with a kernel compiled without PREEMPT and SMP,
but if irqs nothing should change... 

Regards,
Jarek P.

PS: alas I didn't even check compiling - I had no time to
find all compile dependencies of this driver
---

Signed-off-by: Jarek Poplawski <jarkao2@...pl>
---

diff -Nurp linux-2.6.20-rc4-/drivers/net/mv643xx_eth.c linux-2.6.20-rc4/drivers/net/mv643xx_eth.c
--- linux-2.6.20-rc4-/drivers/net/mv643xx_eth.c	2006-12-18 08:57:52.000000000 +0100
+++ linux-2.6.20-rc4/drivers/net/mv643xx_eth.c	2007-01-11 08:55:34.000000000 +0100
@@ -312,8 +312,8 @@ int mv643xx_eth_free_tx_descs(struct net
 	int count;
 	int released = 0;
 
+	spin_lock_irqsave(&mp->lock, flags);
 	while (mp->tx_desc_count > 0) {
-		spin_lock_irqsave(&mp->lock, flags);
 		tx_index = mp->tx_used_desc_q;
 		desc = &mp->p_tx_desc_area[tx_index];
 		cmd_sts = desc->cmd_sts;
@@ -348,8 +348,10 @@ int mv643xx_eth_free_tx_descs(struct net
 			dev_kfree_skb_irq(skb);
 
 		released = 1;
+		spin_lock_irqsave(&mp->lock, flags);
 	}
 
+	spin_unlock_irqrestore(&mp->lock, flags);
 	return released;
 }
 
-
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html