5.4-RC2 freezing - ATA related?

[Thomas Hurst]

* Søren Schmidt (sos at FreeBSD.ORG) wrote:

> No, my only advise is to use the ATA mkIII patches or better yet -
> current..

In a similar vein, I'm seeing the same WRITE_DMA timeouts and system
lockups using ATA mkIII patches as I did using the standard RELENG_5
driver, on two seperate systems.

I'm getting the WRITE_DMA retries on a multi-gmirror Athlon system using
a PCI SATA card; the two PATA drives on the system are fine:

 FreeBSD 5.4-STABLE #0: Thu Apr 28 06:31:53 BST 2005
 atapci1: <SiI 3112 SATA150 controller> port
  0xcc00-0xcc0f,0xc800-0xc803,0xc400-0xc407,0xc000-0xc003,0xbc00-0xbc07
  mem 0xe7062000-0xe70621ff irq 11 at device 12.0 on pci0
 ad4: 381554MB <ST3400832AS/3.01> [775221/16/63] at ata2-master SATA150
 ad6: 381554MB <ST3400832AS/3.01> [775221/16/63] at ata3-master SATA150
 ..
 ad4: TIMEOUT - WRITE_DMA retrying (2 retries left) LBA=401743679
 ad4: TIMEOUT - WRITE_DMA retrying (2 retries left) LBA=781421759

It seems harmless, but results in writes freezing for several seconds
every couple of hundred MB (annoying with 360G of storage as you might
imagine).  It normally favours a single drive, but seems to bounce
between ad4 and 6 for no apparant reason.  Replacing the SATA card and
cables has no effect.  Attempting to drop the drives to PIO with
atacontrol doesn't seem to do anything either (they remain at SATA150).

The other system where I see the lockups (I used to get READ/WRITE_DMA
timeouts with the lockup many moons ago, which seems to have started
after a system update, but for the past 6+ months or so I just get the
lockup) is an old BP6 (dual Celeron), on two different channels on two
different drive:

 FreeBSD 5.4-STABLE #2: Tue Apr 26 17:59:25 BST 2005
 atapci1: <HighPoint HPT366 UDMA66 controller> port
   0xd800-0xd8ff,0xd400-0xd403,0xd000-0xd007 irq 18 at device 19.0 on pci0
 atapci2: <HighPoint HPT366 UDMA66 controller> port
   0xe400-0xe4ff,0xe000-0xe003,0xdc00-0xdc07 irq 18 at device 19.1 on pci0
 ad4: 76319MB <Seagate ST380011A 3.04> at ata2-master UDMA66
 ad6: 114473MB <Seagate ST3120026A 3.01> at ata3-master UDMA66

Setting these drives to PIO4 resolves the stability problems (which
again only occurs under heavy disk activity, almost always on writes),
but makes the system crawl.  I'm planning on migrating it to gmirror,
which I expect will make it behave more like the Athlon, but obviously
I'd like to be able to use DMA reliably without resorting to RAID-1
everywhere.

Save me Søren!

-- 
Thomas 'Freaky' Hurst
    http://hur.st/