ZFS hang (system #2)
Dennis Glatting
freebsd at pki2.com
Sat Oct 20 23:37:47 UTC 2012
On Sun, 2012-10-21 at 01:17 +0300, Andriy Gapon wrote:
> on 20/10/2012 23:31 Dennis Glatting said the following:
> > The following is from a second system working on a 7TB file (started this
> > morning), which also hung. However, an important difference is this system's CPU
> > is slightly over clocked from 3.6GHz to 4.0GHz; However, prior not over clocking
> > made no difference -- it still hanged.
> >
> > This system has a Gigabyte GA-990FXA-UD7 board.
>
> To me this again looks like an issue with a stuck zio/bio, and not a deadlock.
>
> > bd3# /mnt/camcontrol tags da7 -v (** OS - RAID1 **)
> > (pass7:mps1:0:0:0): dev_openings 215
> > (pass7:mps1:0:0:0): dev_active 40
> > (pass7:mps1:0:0:0): devq_openings 215
> > (pass7:mps1:0:0:0): devq_queued 0
> > (pass7:mps1:0:0:0): held 0
> > (pass7:mps1:0:0:0): mintags 2
> > (pass7:mps1:0:0:0): maxtags 255
>
> Of all the disks this one looks the most suspicious, of course.
>
> Do you have the zio/bio debug patch there and usable kgdb?
>
In src/sys/dev/mps/mps.c are the following tunables.
/* Grab the unit-instance variables */
snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.debug_level",
device_get_unit(sc->mps_dev));
TUNABLE_INT_FETCH(tmpstr, &sc->mps_debug);
snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.disable_msix",
device_get_unit(sc->mps_dev));
TUNABLE_INT_FETCH(tmpstr, &sc->disable_msix);
snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.disable_msi",
device_get_unit(sc->mps_dev));
TUNABLE_INT_FETCH(tmpstr, &sc->disable_msi);
snprintf(tmpstr, sizeof(tmpstr), "dev.mps.%d.max_chains",
device_get_unit(sc->mps_dev));
TUNABLE_INT_FETCH(tmpstr, &sc->max_chains);
Whose after boot values are:
dev.mps.0.debug_level: 4
dev.mps.0.disable_msix: 0
dev.mps.0.disable_msi: 0
dev.mps.0.firmware_version: 14.00.00.00
dev.mps.0.driver_version: 14.00.00.01-fbsd
dev.mps.0.io_cmds_active: 0
dev.mps.0.io_cmds_highwater: 60
dev.mps.0.chain_free: 2048
dev.mps.0.chain_free_lowwater: 2047
dev.mps.0.max_chains: 2048
dev.mps.0.chain_alloc_fail: 0
Is there any value to tweaking these?
More information about the freebsd-fs
mailing list