lnc interface dying under heavy I/O when running 4.6-STABLE under VMware 3.1.1

[Nuzrin Yaapar]

Guys,

The subject describes it all. I don't know if anyone else is facing this
problem, but I can reproduce this problem at will. Under VMware 3.1.1
for Windows, the lnc driver will stop responding after some time. The
weirdest part is that it will only manifest itself when doing some heavy
network I/O, such as 'make installworld' from NFS mount, or doing
netperf.

The host OS is WinXP Pro, and the guest OS is the freshly cvsup
RELENG_4. Looking at VMware log file, I see a lot these messages:

Jul 09 16:42:48: VMX|VLANCE: ethernet0 skipped 1024 time(s)
Jul 09 16:42:48: VMX|VLANCE: 254 30 24 23 26 215 11 3 2 0 0 0 0 0 0 0
Jul 09 16:42:48: VMX|VLANCE: 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 436
Jul 09 16:42:51: VMX|VLANCE: ethernet0 skipped 1280 time(s)
Jul 09 16:42:51: VMX|VLANCE: 354 44 37 35 37 313 17 3 3 1 0 0 0 0 0 0
Jul 09 16:42:51: VMX|VLANCE: 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 436

I've done some investigation, and it seems that after some heavy network
activity, the lnc driver will stop calling RX interrupt routine,
lnc_rint(). The "fix" is very simple though; in sys/i386/isa/if_lnc.c,
delete the call to outw(sc->rdp, RINT | INEA) at the very end of
lnc_rint() function. See diff below applied against the latest RELENG_4
code.

# diff -ruN sys/i386/isa/if_lnc.c.old sys/i386/isa/if_lnc.c
--- sys/i386/isa/if_lnc.c.old Tue Jul 9 21:04:31 2002
+++ sys/i386/isa/if_lnc.c Tue Jul 9 21:04:45 2002
@@ -630,7 +630,6 @@
* here have been dealt with.
*/

- outw(sc->rdp, RINT | INEA);
}

Anyone care to describe/explain/enlighten me on what is happening
actually?

- nuzrin -

__________________________________________________
Do You Yahoo!?
Sign up for SBC Yahoo! Dial - First Month Free
http://sbc.yahoo.com

To Unsubscribe: send mail to majo...@FreeBSD.org
with "unsubscribe freebsd-stable" in the body of the message

[黒岩 和行]

Hello.

I work at same environment. And I faced to same problem that lnc
stopped transmission under heavy load.

So I tried patch you posted, it works fine.

I noticed that before adapting a patch, transmission is burstly
processed and stacked. But after patched, transmission looks gently
(slowly) as Windows guest does. Is this expected behaviour?

--
kazuyuki KUROIWA Email: karu...@cam.hi-ho.ne.jp