Hi Alexey, On Mon, Aug 28, 2000 at 08:50:41PM +0400, kuznet@ms2.inr.ac.ru wrote: > > I'll reboot the box now. Any ideas on what THAT could have been? > > It loops in kernel doing schedule(), apparently. > Just make Alt-Sysreq-p several times to see where it loops. > I know only one place, where it is theoretically possible > (when closing socket), but I cannot exclude some another hole. > All the things (lossy ping) point that you are out of memory yet. I'm sitting at the console now; everything works well until I do 'netstat' while the machine is serving its 4500 conns/s. Just did it again: test works flawlessly. After the test, 'wc -l /proc/net/tcp' gives me 28728 (TIME_WAIT) sockets. When I do the same /proc/net/tcp reading while the test runs, it delivers a result (slightly over 30000), I have a new shell prompt, but the console is dead (no num-lock toggling). Alt-Sysreq-p still shows registers, EIP is mostly at c010727d, which corresponds to default_idle. Apart from that I see two EIPs in __write_lock_failed; there, EDI is c02d1f4c, which lies in irq_stat. Hope this helps nailing it. I'll boot and go home now. Can try things tomorrow. (BTW, this time I used another machine, with a different ethernet card, e1000 instead of sk98lin, and without the VLAN code.) best regards, and congratulations for finding the previous bug so fast Patrick - : send the line "unsubscribe linux-net" in the body of a message to majordomo@vger.kernel.org
