RE: [E1000-devel] Crash in e1000e, 3.3.8+ (tainted)

-----Original Message-----
From: Ben Greear [mailto:greearb@candelatech.com]
Sent: Tuesday, July 24, 2012 4:21 PM
To: Dave, Tushar N
Cc: e1000-devel list; netdev; Allan, Bruce W
Subject: Re: [E1000-devel] Crash in e1000e, 3.3.8+ (tainted)

On 07/24/2012 04:13 PM, Dave, Tushar N wrote:

quoted

quoted

-----Original Message-----
From: Ben Greear [mailto:greearb@candelatech.com]
Sent: Tuesday, July 24, 2012 2:46 PM
To: e1000-devel list; netdev
Subject: [E1000-devel] Crash in e1000e, 3.3.8+ (tainted)

We have a somewhat reproducible crash using a 6-port NIC with 3.3.8+
kernel.  This kernel is tainted with a proprietary module, but the
module is not in use.

The rx-all and related patches that were later accepted upstream have
been applied to this kernel.

It seems that buffer_info is NULL in the code below?


(gdb) list e1000_alloc_rx_buffers+0x5b Junk at end of line
specification.
(gdb) list *(e1000_alloc_rx_buffers+0x5b)
0x15822 is in e1000_alloc_rx_buffers (/home/greearb/git/linux-
3.3.dev.y/drivers/net/ethernet/intel/e1000e/netdev.c:611).
606
607		i = rx_ring->next_to_use;
608		buffer_info = &rx_ring->buffer_info[i];
609
610		while (cleaned_count--) {
611			skb = buffer_info->skb;
612			if (skb) {
613				skb_trim(skb, 0);
614				goto map_skb;
615			}
(gdb)

Ben,

This looks familiar to me, I believe this is due to race between adapter

reset and e1000_close.

quoted

Let me check if we have fix upstream or not.

I'm testing Bruce Allen's suggestion now:  bb9e44d0 (from 3.4).

Yep, commit bb9e44d0 the one.

It applies with fuzz to my 3.3.8+ tree.

So far, so good...but need to do some more reboots to be sure.

Thanks,
Ben

--
Ben Greear [off-list ref]
Candela Technologies Inc  http://www.candelatech.com