-CURRENT ATA woes

I've got a machine currently running RELENG_6 solidly, but trying to
put -CURRENT on it causes the ata subsystem to go out to lunch.  This
also happened on this machine with an early-March -CURRENT.

The system has a Promise TX2200 SATA controller.  dmesg from RELENG_6
(slightly rewrapped):

atapci0: <Promise PDC20571 SATA150 controller> port
    0xc000-0xc07f,0xc400-0xc4ff 
    mem 0xeb420000-0xeb420fff,0xeb400000-0xeb41ffff irq 15
    at device 13.0 on pci0
ata2: <ATA channel 0> on atapci0
ata3: <ATA channel 1> on atapci0
ata4: <ATA channel 2> on atapci0
[...]
ad4: 238475MB <HDT722525DLA380 V44OA80A> at ata2-master SATA150
ad6: 238475MB <HDT722525DLA380 V44OA80A> at ata3-master SATA150
ar0: 238475MB <Promise Fasttrak RAID1> status: READY
ar0: disk0 READY (master) using ad4 at ata2-master
ar0: disk1 READY (mirror) using ad6 at ata3-master


There is another ATA controller onboard running the DVD drive:

atapci1: <VIA 8235 UDMA133 controller> port
    0x1f0-0x1f7,0x3f6,0x170-0x177,0x376,
    0xd400-0xd40f at device 17.1 on pci0
ata0: <ATA channel 0> on atapci1
ata1: <ATA channel 1> on atapci1
acd0: DVDR <PIONEER DVD-RW DVR-110D/1.11> at ata0-master UDMA66

but I don't know if that matters.


Any significant drive access will cause the drives on the Promise
controller to wander off into la-la land.  They'll end up getting a
bunch of (hand-transcribed):

ad4: WARNING - READ_DMA UDMA ICRC error (retrying request) LBA: <various>

It may happen on ad4, or on ad6.  Once it held off long enough for me
to almost complete an installworld of the -CURRENT world (which
_royally_ screwed me, since I couldn't boot the old kernel then).
Sometimes even if I leave it very quiet, it'll eventually fall over
with very little drive activity.  I can provoke it pretty quickly with
a `find / | xargs cat >> /dev/null`.


Running RELENG_6, it's solid as a rock, running months at a time.  I
can nail the disk to the wall at will, and it won't even blink.  But
let -CURRENT at it, and it'll blow beets if you look at it hard.
Unfortunately, as it runs my mail (where I'm typing at the moment), I
can't really take it down to experiment much with.



-- 
Matthew Fuller     (MF4839)   |  fullermd_at_over-yonder.net
Systems/Network Administrator |  http://www.over-yonder.net/~fullermd/
           On the Internet, nobody can hear you scream.