ATA READ command timeout (and worse)

[Axel Simon]

On Wed, Jun 18, 2003 at 12:53:18AM +1000, Stephen McKay wrote:
> I've been plagued by ATA READ command timeouts on my home NFS server.
> Tonight I decided to get to the bottom of it, so I started torturing
> it. :-)
> 
> The machine is a Celeron 400 with 256MB ram in a Soyo 7IZB motherboard.
> There are 4 disks (1 on the M/B, and 3 on a SiI 0680).  There are also
> 2 NICs and a Trio64 in there.  It's pretty cramped, but I know it's not
> overheating because I've got a thermometer in there too.  (And it's
> winter here.)  It can serve up 15MB/s of NFS traffic (though 2 NICs),
> which isn't bad for an old CPU.

If your disks make some clicking noises, they are probably overheated.  
Once a disk gets too hot it can take permanent damage as the head scrapes
off some material from the surface which gets in the way of the head which
then cannot move freely.

> Jun 17 22:28:02 peon /kernel: ad6: READ command timeout tag=0 serv=0 - resetting
> Jun 17 22:28:02 peon /kernel: ata3: resetting devices .. done
> ...
> Jun 17 23:05:03 peon /kernel: ad4: READ command timeout tag=0 serv=0 - resetting
> Jun 17 23:05:03 peon /kernel: ata2: resetting devices .. ad4: DMA limited to UDMA33, non-ATA66 cable or device
> Jun 17 23:05:03 peon /kernel: done
This could be the disk getting too hot. It's IDE, so it tries to deal with 
all problems itself. AFAIK there is no way the disk can actually 
communicate more informative error messages to the driver.

> Note the cable most certainly IS ATA66 (and ATA100) capable, and correctly
> connected.  Note that at boot time ad4 is NOT limited to UDMA33.
I've heard people complaining about low-quality cables. Make sure they are 
short, preferably twist'n'flat cables (round and shielded between the 
ribbon connectors). But if...

> I have changed the cables and this made no difference.  I've tried manually
> limiting the speed (with atacontrol) to UDMA33 and that made no difference.

> The power supply has 230W written on it, and I'm assuming that (like
> nearly all power supplies) it's an optimistic value.  I really don't
> want to take the whole thing apart so I'm hoping someone finds an ata
> driver bug!
I don't think you need to. 230W = 5V * 20 A + 12V * 10A + ... Each drive 
says how much it needs which is about 1.2A maximum (i.e. when it powers 
up). If the power supply would be too weak, then it would get too hot and 
switch off for some time (also on the 5V line which will reset the PC).

Make sure you have one drive's worth of space between each pair of hard 
disks. If you can't touch the hard disk from all sides without burning 
yourself, they are too hot.

My experience,
Axel.