Re: netif/dc issues

[Eduardo Tongson <propolice@xxxxxxxxx>]

On 7/17/05, Matthew Dillon <dillon@xxxxxxxxxxxxxxxxxxxx> wrote:
> 
> :On -Preview and HEAD
> :my Davicom 9102 10/100 cards
> :are choking out these messages
> :
> :...
> :mfree: m->m_nextpkt != NULL
> :Trace beginning at frame (0xdeadbeef)
> :m_free (..) at m_free+0x5e
> :m_free (..) at m_free+0x5e
> :m_freem (..) at m_freem+0x2e
> :dc_start (..) at dc_start+0xd4
> :dc_tick (...) at dc_tick+0x148
> :softclock_handler (...) at softclock_handler+0xe2
> :lwkt_exit() lwkt_exit
> :...
> :If i get that while on my 2-port KVM switch it locks up the keyboard
> :which then needs a hard reset.
> :
> :In hopes of fixing it, I brought in changes from freebsd (see submit@)
> :but still no luck with the updates
> :
> :The card works fine on NetBSD, OpenBSD, QNX, Linux
> :
> :--ed
> :Eduardo Tongson
> 
>     I have a tentative patch set that fixes an interlock/blocking issue
>     with sockbufs.  I do not know if this will help your situation but
>     it's worth trying.
> 
>     Note that YONETANI Tomokazu reports he gets a panic when he shutsdown
>     a system with this patch and the debugging enabled, so the patch may
>     not yet solve all the issues.  I expect the patch will go through a
>     few more revisions before it goes into the tree.
> 
>     The main purpose of the patch is to fix issues that occur on SMP systems.
> 
>     The m_nextpkt field is known to not always be properly cleared, but
>     it's odd that it would occur in the dc_start path.
> 
>                                         -Matt

the cumulative patched fixed this.
no hangs anymore.

I just see this often
. ..
Interrupt 10 at 63355hz livelocked! limiting at 20khz
Interrupt 10 at 9545hz removing livelock limit
. ..
--ed 
Eduardo Tongson   <>