Mark Lord wrote:
> Evgeni Golov wrote:
>> On Tue, 04 Nov 2008 13:13:16 -0500 Mark Lord wrote:
>>
>>> Okay, hdparm-9.3 is now out in the wild (sourceforge),
>>> and has  --idle-immediate and --idle-unload  flags now,
>>> so it can be used to help debug/test this problem.
>>
>> Okay, got it, built it.
>> Neither --idle-immediate nor --idle-immediate brings up the reset,
>> echo 1000 > /sys/block/sda/device/unload_heads does.
> ..
> 
> Mmmm.. okay, this is new stuff in 2.6.28,
> and it appears to just issue a --idle-unload equivalent after a delay.
> 
> But it does it from within libata-eh, so I suppose there must
> be some confusion in there somewhere.
> 
> So it's up to Tejun now, I suppose.

Hmmm... maybe garbage values in unused TF regs?

-- 
tejun