netdev
[Top] [All Lists]

Re: pktgen problem (skb refcount) in 2.6.12-rc1

To: Harald Welte <laforge@xxxxxxxxxxxx>
Subject: Re: pktgen problem (skb refcount) in 2.6.12-rc1
From: Ayyappan Veeraiyan <ayyappan.veeraiyan@xxxxxxxxx>
Date: Thu, 7 Apr 2005 18:15:09 -0700 (PDT)
Cc: "Veeraiyan, Ayyappan" <ayyappan.veeraiyan@xxxxxxxxx>, Robert Olsson <Robert.Olsson@xxxxxxxxxxx>, <netdev@xxxxxxxxxxx>
In-reply-to: <20050405155820.GB3261@obroa-skai.de.gnumonks.org>
Replyto: "Ayyappan Veeraiyan" <ayyappan.veeraiyan@intel.com>
Sender: netdev-bounce@xxxxxxxxxxx
On Tue, 5 Apr 2005, Harald Welte wrote:

> 
> I have tried a 2.6.11 kernel which works fine.  I have not tested
> patching e1000 in 2.6.12-rc1 to go back to whatever was shipped in
> 2.6.11.
> 
> I'll try to remember testing that once I'm back from my current business
> trip (friday), though
> 

I think one of the previos set of e1000 patches is the cause for 
this issue. TSO workaround we put in doesn't release the last TX packet 
skb if there are no more receives on the interface.
 
Could you please try the below patch to 2.6.12-rc1? Apologies if the patch 
doesn't apply cleanly.

thanks,
Ayyappan V.

--- linux-2.6.11.6/drivers/net/e1000/e1000_main.c.fix1  2005-04-04 
22:08:13.026941064 -0700
+++ linux-2.6.11.6/drivers/net/e1000/e1000_main.c       2005-04-04 
22:07:18.775188584 -0700
@@ -2426,6 +2426,14 @@
                   !(E1000_READ_REG(&adapter->hw, STATUS) & E1000_STATUS_TXOFF))
                        netif_stop_queue(netdev);
        }
+#ifdef NETIF_F_TSO
+
+       if( unlikely(!(eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
+           time_after(jiffies, adapter->previous_buffer_info.time_stamp + HZ)))
+               e1000_unmap_and_free_tx_resource(
+                   adapter, &adapter->previous_buffer_info);
+
+#endif
 
        return cleaned;
 }



<Prev in Thread] Current Thread [Next in Thread>