HELP! Just 5 minutes! - kworker eating up cpu time when doing dd on 3.2.0 kernels

[Stefan Bader]

On 06.02.2012 12:06, Andy Whitcroft wrote:
> On Mon, Feb 06, 2012 at 10:53:04AM +0100, Christian Brandes wrote:
>> Hi all!
>>
>> This takes just 5 minutes!!!!!
>> Please, could someone help me!
>>
>> It looks like a serious bug in 3.2-Kernels.
>>
>> Could someone just do "dd if=/dev/zero of=/tmp/zero1.tmp bs=1024 count=13000000" on a precise system with 3.2. kernel and watch CPU usage of kworker threads going up when cache memory is filled up.
> 
> Nope could not reproduce it here on a machine with only 2GB of ram; this
> is with a 3.2.0-12 Ubuntu kernel.  I see the expected kswapd0 and flush-8:0
> using some 10% between them worst case.  Which kernel are you using,
> what platform is triggering it.
> 
> -apw
> 
Same here with 3.2.0-14.23. About 80M/s and same kswapd0 and flush-8:0
percentages. Using a dual core cpu.

One thing to verify as well: is /tmp part of / or a tmpfs mount? (though in that
case one would not expect a 13G write to succeed)...

-Stefan