Re: Freeze after disabling write cache with hdparm -W0 /dev/sda

[Simon Farnsworth <simon.farnsworth@onelan.co.uk>]

Tejun Heo wrote:
> Simon Farnsworth wrote:
>> Tejun Heo wrote:
>>> Simon Farnsworth wrote:
>>>> Tejun Heo wrote:
>>>>> Simon Farnsworth wrote:
>>>>>> Just a thought; is it possible to trigger libata EH from userspace? If
>>>>>> so, we could write a small utility to disable write cache, then force EH
>>>>>> to detect the change.
>>>>> That's automatically done.  libata snoops cache on/off and triggers
>>>>> revalidation but you can request manual rescan by echoing "- - -" to
>>>>> /sys/class/scsi_host/hostX/scan.
>>>>>
>>>> Would it be worth changing our code to do "hdparm -W1 /dev/sda && hdparm
>>>> -W0 /dev/sda", or would this not show anything. The lack of revalidation
>>>> on some drives is what's worrying me a little here.
>>> Revalidation happens after cache property is changed successfully.  What
>>> happens if you request manual rescan while the drive is not repsponding?
>>>
>> The 30 second freeze happens when we request manual rescan. We do
>> revalidate once the freeze is over, though; dmesg follows:
> 
> Can you post dmesg with printk timestamps turned on?
> 
Sorry for the delay in responding;

Turning on printk timestamps is enough to fix the problem, and stop the
stall from happening.

We'll be rebasing to the newest Fedora 7 kernel soon (next 3 months),
and I've changed our kernel build process to turn on printk timestamps.
If it still happens then, I'll get you a dmesg with timestamps.
-- 
Simon Farnsworth