At 22:33 Uhr +0200 29.3.2005, Manuel Bouyer wrote:
>On Tue, Mar 29, 2005 at 09:13:13PM +0200, Hauke Fath wrote:
>> Like i386?
>>
>> There is an open issue with mpt(4) where monting disks with softdep (or
>> async, in the case of an installation) will jam the machine quickly.
>
>Is it still true ? I made some changes to scsipi to address this some
>time ago.

It ist true for netbsd-2, definitely - see kern/26825. And I tried an
install with 2.99 a while back (probably January) which gave identical
results.

I've got two machines running NetBSD with that 53c1030 controller, each
with two disks running mirrored, one acting as filtering router, the other
as mail-/webserver. "Normally", these boxes work, if you avoid high scsi
activity. From the latter, I see the occasional reboot during the
/etc/daily cron job, with messages like

mpt0: Unknown async event: 0xb
mpt0: Unknown async event: 0xb

and a stream of

sd0(mpt0:0:0:0):  Check Condition on CDB: 0x2a 00 03 69 13 bf 00 00 10 00
    SENSE KEY:  Hardware Error
    ASC/ASCQ:  Internal Target Failure

which in that case lead to

First bad, reclen=2000, DIRSIZ=216, namlen=204, flags=1000
entryoffsetinblock=0, dirblksiz = 512
/usr: bad dir ino 117616 at offset 0: mangled entry
panic: bad dir
Begin traceback...

and a coredump which I can provide, if it helps.

These two machines have intel entry-level server boards, P4, 32Bit PCI.
There are two more machines, same controllers with a later firmware
revision, boards same family, only 64Bit PCI. When I put the NetBSD disks
into one of those machines, I found that they wouldn't even boot multi-user
before locking up. The boxes run debian with a

kodos:~# uname -a
Linux kodos 2.6.8-1-386 #1 Thu Nov 11 12:18:43 EST 2004 i686 GNU/Linux

kernel just fine.

Let me know if there is anything I can test / log / try out for you.  :)

	hauke

--
/~\  The ASCII Ribbon Campaign
\ /    No HTML/RTF in email
 X     No Word docs in email
/ \  Respect for open standards