Re: XEN/NetBSD panic during loop in the network switch

[Anders Magnusson <ragge%ltu.se@localhost>]

Manuel Bouyer wrote:
> On Fri, Sep 26, 2008 at 05:42:37PM +0200, Fredrik Pettai wrote:
>   
>> Hi,
>>
>> We had a misconfiguration issue on one of our smaller server networks,  
>> then our personal manage to make a loop by connecting two switches  
>> together over several physical connector, without running any loop  
>> protection, like STP or equal.
>>
>> We discovered some weird behavior on one of our servers running XEN  
>> (3.1.2) / NetBSD 4.0 RELEASE.
>> It got panic directly (no other system did that), with a very long  
>> console output...
>> The only modification made to the running kernel (compared with the  
>> XEN3_DOM0) is "options    NMBCLUSTERS=65536".
>>
>> The thing that made it more interesting, is that another server  
>> running an older version of XEN (3.0.3) / NetBSD 4_BETA2, located in  
>> the same switch, didn't say a thing. It just kept running, business as  
>> usual...
>>
>> Both systems running on the exact same hardware (Dell 2950, the where  
>> purchased at the same time, but latter one hasn't been firmware  
>> updated (BIOS, raid controller etc.)). Except that the NetBSD source  
>> differ (quite much, BETA2 is from feb 2008), the kernel running on the  
>> latter system is also a XEN3_DOM0, but slightly stripped conf and with  
>> "options    NMBCLUSTERS=4096"
>>
>> Anybody else who has experienced the same thing while make loops on  
>> the network? :-)
>> Or any hints to why the newer XEN/NetBSD system went unhappy?
>>
>> Here is the console output:
>>
>> panic: bnx0: Double mbuf allocation failure!
>>     
>
> It just ran out of memory. This is a know failure mode of the bnx driver;
> I think it has been fixed recently (or maybe it was just patches floating
> around, I don't remember).
That sounds reasonable, I think that DOM0 only had 256M of memory assigned.

-- Ragge