LSI 1030 mpt doesn't work if I build a new kernel

[YONETANI Tomokazu]

Hi,
Recently I tried installing FreeBSD 5.4-RELEASE on an IBM X335 in my
office and tried to upgrade to 6.2-PRERELEASE, and found that although
it doesn't hang or panic during boot, running I/O intensive applications
(including buildworld) ends up in an I/O error, and eventually panicked and
rebooted.  So I tried earlier versions to find out when the problem started
to occur, and found that it started as early as 5.5-STABLE.  I noticed that
new mpt driver has been MFC after 5.4-RELEASE, so maybe that's something
to do with this issue.
Kernels other than 5.4 are all built on 5.4-RELEASE system using GENERIC file
without any optimization flags.  I haven't tried CURRENT yet because
buildworld didn't finish on 5.4-RELEASE, but is it worth trying CURRENT
and the patch posted to this thread(in other words: does my problem look like
the similar one that people are discussing in this thread)?

Here's `dmesg | grep ^mpt' output:
  mpt0: <LSILogic 1030 Ultra4 Adapter> port 0x2300-0x23ff mem 0xfbfe0000-0xfbfeffff,0xfbff0000-0xfbffffff irq 22 at device 1.0 on pci1
  mpt0: MPI Version=1.2.6.0
  mpt0: Capabilities: ( RAID-1 SAFTE )
  mpt0: 1 Active Volume (1 Max)
  mpt0: 2 Hidden Drive Members (3 Max)

I found a PR with "LSI1030" in the single-line description, but it doesn't
seem to be related to my problem: kern/96040

Any suggestions are welcome.
Regards.


5.4-RELEASE:
  OK

5.5-STABLE:

  panic after showing the following messages:
    handle_workitem_freeblocks: block count
    initiate_write_filepage: already started

6.0-RELEASE-p12, 6.2-PRERELEASE:
  panic after showing bunch of the message as follows:
    g_vfs_done():da0s1g[WRITE(offset=3287388160, length=2048)]error = 5