[Users] held and maxheld below barrier and limit but failcnt = 4

[Kir Kolyshkin]

Franz STUMPNER wrote:
> I have the Problem that my /proc/user_beancounters displays an 
> failcount of 4 at the privvmpages section:
>
> Version: 2.5
>        uid  resource    held    maxheld barrier    limit    failcnt
>       102:  kmemsize    2827049 3378858 41840803   46024883 0
>             lockedpages 0       0       2043       2043     0
>             *privvmpages 118113  130716  173225     198547   4*
>
> How can it be that failcnt is 4 when the held and maxheld values are 
> below the barrier an limit values?
> For me it means that the privvmpages never hit the barrier or limit, 
> how can there be a failcnt!? Or I'm wrong?
That's easy -- the difference between privvmpages.maxheld and 
privvmpages.barrier is 173225-130716=42509. So if any application tries 
to grab more than 42509 pages *at once*, it will get an error (most 
probably ENOMEM), and the failcnt will be increased, but both held and 
maxheld stays the same.