FAILURE - WRITE_DMA timed out

[Mykel]

Christopher Vance wrote:

>When I use SATA disks heavily on a 5.3-R machine (Athlon) with the
>SiI3512 disk controller, I get frequent (sometimes one every 1-5
>minutes) WRITE_DMA timed out messages (with different blocks and both
>disks).
>
>It also panics with more than one kind of panic, and sometimes doesn't
>even get to finish the background fsck from the previous panic before
>dying again.
>
>Given that both disks get the WRITE_DMA messages, and the one with the
>OS on didn't have any problems when used with a VIA controller in a
>different machine, I tend to think it's either the controller or the
>software driving it.
>
>Would I have any joy upgrading along 5-STABLE, perhaps after I move
>the disk back to the machine that works, or won't that make any
>difference?  (As in, are there any relevant ata changes since then?)
>
>I've already tried setting both hw.ata.ata_dma and hw.ata.wc to 0, but
>I still get these timeouts.
>

I don't have any solutions for you, but I'm curious to know what kind of 
drives you're using... apparently SiI 3114s aren't so good with bridged 
(PATA<->SATA) connections, and Seagates are mostly native...

On my setup (SiI3112+Seagate), I get WRITE_DMA timeo's a few times a 
day. Never had panics as result of the SATA tho.

atapci1 at pci0:13:0:      class=0x018001 card=0x02509005 chip=0x31121095 
rev=0x02 hdr=0x00
    vendor   = 'Silicon Image Inc (Was: CMD Technology Inc)'
    device   = 'SiI 3112 SATARaid Controller'
    class    = mass storage

ATA channel 2:
    Master:  ad4 <ST3120026AS/3.18> ATA/ATAPI rev 6
    Slave:       no device present

I don't even use it very heavily, it's *just* for my .oggs. Nothing 
else. It sucks. We're presently looking at 3Ware and others for a heavy 
duty network setup we're doing... hoping we can make the SiIs work 
instead of dropping a few grand on 3Wares... :\ but I'm not confident 
the SiIs are good for anything important.

Myke