Fatal TIMEOUT - WRITE_DMA errors return with ATA Mk.III

Since somewhere in the 5.x release cycle, my system has fallen prey to
the "TIMEOUT - WRITE_DMA" errors which result in the drive becoming
detached (which causes my geom_mirror to break and require rebuilding).
According to smartctl and disk diagnostics, there's nothing wrong with
my drives.  Plus, the problem does not manifest itself under 4-STABLE.
(I'm not the only one to have reported this problem.)

Lately, I'd had success using a patch posted to freebsd-current by Ian
Dowse.  The "TIMEOUT - WRITE_DMA" errors still occurred, but they
weren't fatal.  I updated my kernel and world recently, and, alas, the
"TIMEOUT - WRITE_DMA" problem has returned once more:

ad0: TIMEOUT - WRITE_DMA retrying (2 retries left) LBA=49981679
ad0: FAILURE - device detached
subdisk0: detached
ad0: detached
GEOM_MIRROR: Cannot update metadata on disk ad0 (error=5).
GEOM_MIRROR: Cannot update metadata on disk ad0 (error=6).
GEOM_MIRROR: Device raid1: provider ad0 disconnected.
GEOM_MIRROR: Request failed (error=6). ad0[WRITE(offset=3847741440, length=16384)]

Ian's patch was against the pre-ATA Mk.III regime.  I doubt it is
applicable to the ATA Mk.III rewrite. :-(

Here is my system (re: ATA), FWIW:

FreeBSD 6.0-CURRENT #0: Mon Apr 18 12:25:24 EDT 2005
    paul_at_zappa.Chelsea-Ct.Org:/usr/obj/usr/src/sys/ZAPPA
Timecounter "i8254" frequency 1193182 Hz quality 0
CPU: Intel Pentium III (698.39-MHz 686-class CPU)
  Origin = "GenuineIntel"  Id = 0x683  Stepping = 3
  Features=0x383f9ff<FPU,VME,DE,PSE,TSC,MSR,PAE,MCE,CX8,SEP,MTRR,PGE,MCA,CMOV,PAT,PSE36,MMX,FXSR,SSE>
real memory  = 536870912 (512 MB)
avail memory = 520253440 (496 MB)
[[...]]
atapci0: <Intel PIIX4 UDMA33 controller> port 0x1f0-0x1f7,0x3f6,0x170-0x177,0x376,0x1440-0x144f at device 7.1 on pci0
ata0: <ATA channel 0> on atapci0
ata1: <ATA channel 1> on atapci0
[[...]]
ad0: 24405MB <IBM DJNA-352500 J51OA30K> at ata0-master UDMA33
acd0: DVDR <LITE-ON DVDRW SOHW-832S/VS08> at ata0-slave UDMA33
ad2: 24405MB <IBM DJNA-352500 J51OA30K> at ata1-master UDMA33
acd1: CDRW <LG CD-RW CED-8080B/1.04> at ata1-slave PIO4


My kernel has ATAPICAM support compiled in.  Here is the pciconf -vl
output for my ATA controller:

atapci0_at_pci0:7:1:       class=0x010180 card=0x00000000 chip=0x71118086 rev=0x01 hdr=0x00
    vendor   = 'Intel Corporation'
    device   = '82371AB/EB/MB PIIX4/4E/4M IDE Controller'
    class    = mass storage
    subclass = ATA

It is in a Dell Dimension XPS T700r.

Is there any way to up the number of retries to, say, 5, to see if this
helps?

Oh, well, welcome back "gmirror rebuild..." :-)

Cheers,

Paul.
-- 
e-mail: paul_at_gromit.dlib.vt.edu

"Without music to decorate it, time is just a bunch of boring production
 deadlines or dates by which bills must be paid."
        --- Frank Vincent Zappa