Subject: kern/10438: panic: pool_get(mcpl): free list modified...
To: None <email@example.com>
From: Hal Murray <firstname.lastname@example.org>
Date: 06/25/2000 01:09:10
>Synopsis: panic: pool_get(mcpl): free list modified...
>Arrival-Date: Sun Jun 25 01:10:00 PDT 2000
>Originator: Hal Murray
>Release: NetBSD 1.5_ALPHA, 20000620-1.5
Systems Research Center, Compaq Computer Corporation
Alpha 600au with heavy traffic on the fxp driver (Intel 82558)
panic: pool_get(mcpl): free list modified...
This is problably related to kern/10359 "fxp: device timeout".
(But I don't have a smoking gun.)
I have a pair of 600au systems. One has 512 MB. The other
has 256. They are full of network cards.
I have a set of network tests. The full set runs for ~24
hours per driver/link to be tested.
I've seen at least one crash when only "nice" tests using
TCP had been run. They kept the link busy but didn't do
anything nasty like run out of mbufs.
With 1.4Z, I had ~6 crashes, all on the 512B machine.
With 1.5_ALPHA, my first try (switch) finished. On the second
try (hub), the 256 MB machine crashed.
With 1.4Z, I've run these tests on i386 systems with fxp
and tlp drivers. I've tried hub, switch, and point-point
links on the fxp driver and some of those on the tlp. I've
never seen troubles like this.
Similarly, on 1.4Z Alphas, I've run these tests on tlp, fpa
(FDDI) and ti (Alteon Gigabit Ethernet) with no troubles.
(I've got tests on 1.5_Alpha running on Alpha and i386.)
No simple recipe.
With 1.4Z, my tests would get one in 2-4 hours. So far,
it has taken much longer on 1.5_ALPHA.
Work around: Don't use fxp driver.