Strange SCSI behavior after upgrading from 5.2.1 to 5.4 (and a panic)

[Panagiotis Christias]

Hello,

on Thurday we upgraded one of our last 5.2.1 servers to 5.4. Tonight
the server panicked, crashed and I had to power it off and on. Here
are the logs before the panic:

Jun 26 03:45:50 patroklos kernel: (da0:ahc0:0:0:0): lost device
Jun 26 03:45:50 patroklos kernel: (da0:ahc0:0:0:0): Invalidating pack
Jun 26 03:46:00 patroklos last message repeated 2 times
Jun 26 03:46:06 patroklos kernel: initiate_write_filepage: already started
Jun 26 03:46:07 patroklos last message repeated 9 times
Jun 26 03:46:07 patroklos kernel: (da0:ahc0:0:0:0): READ(10). CDB: 28
0 72 f 16 0 0 0 80 0
Jun 26 03:46:07 patroklos kernel: (da0:ahc0:0:0:0): CAM Status: SCSI
Status Error
Jun 26 03:46:07 patroklos kernel: (da0:ahc0:0:0:0): SCSI Status: Check Condition
Jun 26 03:46:07 patroklos kernel: (da0:ahc0:0:0:0): UNIT ATTENTION asc:29,0
Jun 26 03:46:07 patroklos kernel: (da0:ahc0:0:0:0): Power on, reset,
or bus device reset occurred
Jun 26 03:46:07 patroklos kernel: (da0:ahc0:0:0:0): Retries Exhausted
Jun 26 03:46:07 patroklos kernel: (da0:ahc0:0:0:0): Invalidating pack
Jun 26 03:46:31 patroklos kernel: initiate_write_filepage: already started
Jun 26 03:46:43 patroklos kernel: panic:
initiate_write_inodeblock_ufs2: already started

As I said the machine could not recover from the panic so there ais no
crashdump.

The 5.4 version of the dmesg output is available at:
http://noc.ntua.gr/~christia/tmp/dmesg-5.4.txt

The 5.2.1 version of the dmesg output is available at:
http://noc.ntua.gr/~christia/tmp/dmesg-5.2.1.txt

da0 is an 1302GB external IDE to SCSI RAID (8x200GB IDE drives in
RAID5 configuration and a SCSI U160 interface). FreeBSD 5.4 connects
to da0 at 80MB/s (40.000MHz, offset 31, 16bit, Tagged Queueing
Enabled), while FreeBSD 2.5.1 (and FreeBSD 5.3 - just tried to boot
with the 5.3-RELEASE-i386-miniinst.iso) connects happily at 160MB/s
(80.000MHz, offset 62, 16bit, Tagged Queueing Enabled) which is the
transfer rate supported by the RAID device and the SCSI card (Adaptec
3960D Ultra160 SCSI adapter/aic7899).

Any ideas what could be or where could be the problem? What has changed in 5.4?
We had preserved the 5.2.1 system disks and after the crash we moved
back to 5.2.1 until further notice. Now I'm thinking of trying 5.3
which seems to have the same behavior as 5.2.1 and will be still
supported for a year or so.

Thanks,
Panagiotis