Re: 2.6.28-rc8 big regression in VM

[Wu Fengguang <fengguang.wu@intel.com>]

On Fri, Dec 12, 2008 at 07:38:10PM +0100, Lukas Hejtmanek wrote:
> Hello,
> 
> I noticed that in the current git head, the VM subsystem contains a serious
> bug - cached memory is newer freed:
> 
> # free; echo 3 > /proc/sys/vm/drop_caches; free
>              total       used       free     shared    buffers     cached
> Mem:       2016688     669172    1347516          0       2160     335120
> -/+ buffers/cache:     331892    1684796
> Swap:      1542232          0    1542232
>              total       used       free     shared    buffers     cached
> Mem:       2016688     659336    1357352          0        320     327288
> -/+ buffers/cache:     331728    1684960
> Swap:      1542232          0    1542232
> 
> the above sequence has been done on completly idle system so there is no
> reason for non freeing the cache.

I cannot reproduce it on 2.6.28-rc7, compiling 2.6.28-rc8...

# free; echo 3 > /proc/sys/vm/drop_caches; free
             total       used       free     shared    buffers     cached
Mem:          1940        109       1831          0          0         42
-/+ buffers/cache:         66       1873
Swap:            0          0          0
             total       used       free     shared    buffers     cached
Mem:          1940         62       1877          0          0          7
-/+ buffers/cache:         55       1884
Swap:            0          0          0

> once the whole memory is eaten by the cache, the system is pretty unusable.

What's your kconfig?

Thanks,
Fengguang