Subject: Re: ti0: mbuf allocation failed -- packet dropped!
To: Markus W Kilbinger <kilbi@rad.rwth-aachen.de>
From: Manuel Bouyer <bouyer@antioche.eu.org>
List: current-users
Date: 09/29/2002 20:42:17
On Sun, Sep 29, 2002 at 03:33:05PM +0200, Markus W Kilbinger wrote:
> >>>>> "Manuel" == Manuel Bouyer <bouyer@antioche.eu.org> writes:
> 
>     >> Is this a 'ti' specific problem, then?
> 
>     Manuel> Not sure. It can't allocate a mbuf, I suspect because
>     Manuel> there are already too much allocated. But it may be
>     Manuel> something else eating the mbufs. [...]
> 
>     >> > Does anybody else see these messages?
>     >> 
>     >> Really nobody?
> 
>     Manuel> Well, my machine with a tigon is still running 1.5.2+,
>     Manuel> because of NFS server bugs in 1.6 ...
> 
> Hmm, serious ones? (Which PR?).

18013

> Maybe, these problems correlate?

No, these were really bugs in NFS. I've reproduced them on i386 with epic

> 
>     >> Usefull things could be vmstat -m (make sure your kernel is
>     >> compiled with options KMEMSTAT), and netstat -m.
> 
>     Manuel> Hum, sorry I just realised you already anserwed this. Can
>     Manuel> you try to get it core dump at this point ? Go to debugger
>     Manuel> (make sure to have 'options DDB' compiled in) and then
>     Manuel> reboot(0x108)
> 
> Ok, now, after the 7th hang up, I have a DDB kernel running. But: The
> system is completely software raid based (even swap)! Can I make a
> core dump in this situation?

You need to declare a dump partition on a real disk. It's easy if your swap is
on raid-1: you just need to declare this partition to match the swap partition
on disk. For example, my raid-1 raid0 uses sd0a and sd1a.
My disklabel for raid0:
3 partitions:
#        size    offset     fstype  [fsize bsize cpg/sgs]
 a:    389760         0     4.2BSD   1024  8192    16   # (Cyl.    0 - 289)
 b:   1123456    389760       swap                      # (Cyl.  290 - 1125*)
 c:   1513216         0     unused      0     0         # (Cyl.    0 - 1125*)

sd0a (and sd1a) is:
 a:   1513323         0       RAID                      # (Cyl.    0 - 470)

You have to do some maths starting from the end of the partitions
to account for the sectors "stolen" by raidframe.
So my swap partition on disk, starts at: 1513323 - 1123456 = 389867. So
sd0b is:
 b:   1123456    389867       swap                      # (Cyl.  121*- 470)

Then I have this in my fstab:
/dev/raid0b none swap sw 0 0
/dev/sd0b none dp dp 0 0

-- 
Manuel Bouyer <bouyer@antioche.eu.org>
--