Slab: in /proc/meminfo not corrent in recent kernels?

[Dave Jones]

On Fri, Feb 04, 2005 at 12:50:42AM -0700, Wes Shull wrote:
 
 > The runlevel 5 ones aren't from separate boots, I just telinit 5'ed
 > after grabbing meminfo and slabinfo.  And the dmesgs are really from
 > /var/log/messages, but who's counting.  (And yes, I know /dev/hdg is
 > dying, no system stuff on there)  Unfortunately, I don't know the
 > significance of (m)any of the entries in the slabinfo, although I do
 > note that dentry_cache accounts for roughly 10 MB of the difference...
 > Hopefully some kernel gurus can tell if this behavior is a problem or
 > not.  Or maybe a side effect of debugging stuff?

On a quick look, it actually seems you use more slabs on the 741
kernel, the only big difference being in the number of size-32
elements allocated, but this is likely just due to the slab-debug
stuff being enabled. If you look at the size of the objects (3rd number)
you'll see that in the 1124 kernel they grow a little, and after size-64,
they're a minimum of 4kb due to the extra overhead of the checking.

This means that in the debugging kernels, we're actually running
slightly different allocation patterns than we see in normal life,
but the potential for catching more serious bugs is an acceptable
tradeoff.  Let us know if things don't go back to normal after
debugging is switched back off (usually its on up until the final
test release).

		Dave