Re: Kernel memory profiling -> network memleak

Joerg Henne (joerg@cogito.de)
Fri, 07 May 1999 14:35:21 +0200


> >your memleak-Device. It clearly shows that the network driver for the RTL8139
> >is the offender (dev.c seems to be accounted for the nic driver's memory,
>
> I don't think the offender is RTL8139 but instead some packet-driver for
> the nework that seems that it's missing simply one kfree_skb() somewhere.

after some source reading I already comcluded that my judgement was a bit
hasty.

> What are you using in the network side (TCP/IP?)? Could you should be the
> configuration of the network? To get this fixed faster I should know which
> is the packet-driver that is generating the leak (arp, IP, whatever).

The system is using TCP/IP to mount most things. Besides that, some syslogging
is done and that's all.

> Please apply this patch against 2.2.7 and let me know which are the most
> used packet driver that you are using:

Here's the result; the last column shows the number of log entries during a 15
minute run:

c0170898 t packet_rcv 13099
c0159a60 T ip_rcv 2399
c0168ba0 T arp_rcv 276

One strange thing I noticed is that the number of times the packet_rcv entry
occurs in a row is steadily increasing. Initially the log showed always two
packet_rcv entries in a row. This number increased continuously up to 10 in a
row when I stopped the test. Finally it looked like this:

[...]
May 7 13:42:21 foo kernel: c0159a60
May 7 13:42:22 foo kernel: c0170898
May 7 13:42:22 foo last message repeated 9 times
May 7 13:42:22 foo kernel: c0159a60
May 7 13:42:22 foo kernel: c0170898
May 7 13:42:22 foo last message repeated 9 times

Regards
Joerg Henne

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/