Re: [RFC] Tracking mlocked pages and moving them off the LRU

[Andrew Morton <akpm@linux-foundation.org>]

On Fri, 2 Feb 2007 22:20:12 -0800 (PST) Christoph Lameter <clameter@sgi.com> wrote:

> This is a new variation on the earlier RFC for tracking mlocked pages.
> We now mark a mlocked page with a bit in the page flags and remove
> them from the LRU. Pages get moved back when no vma that references
> the page has VM_LOCKED set anymore.
> 
> This means that vmscan no longer uselessly cycles over large amounts
> of mlocked memory should someone attempt to mlock large amounts of
> memory (may even result in a livelock on large systems).
> 
> Synchronization is build around state changes of the PageMlocked bit.
> The NR_MLOCK counter is incremented and decremented based on
> state transitions of PageMlocked. So the count is accurate.

I wonder if it can be simpler.  Make two changes:

a) If the scanner encounters an mlocked page on the LRU, take it off.

b) munlock() adds all affected pages to the LRU.

And that's it.  Simpler, solves the uselessly-scan-lots-of-mlocked-pages
problem (which is the sole requirement according to your description) and
doesn't consume a page flag.  Optional (and arguable) extension: scan the
vmas during munmap, don't add page to LRU if it's still mlocked.

Why _does_ your patch add a new page flag?  That info is available via a
vma scan.

> There is still some unfinished business:
> 
> 1. We use the 21st page flag and we only have 20 on 32 bit NUMA platforms.

Ow.  How were you thinking of fixing that?

> 2. Since mlocked pages are now off the LRU page migration will no longer
>    move them.

Ow.  That could be a right pain when we get around to using migration for
memory-unplug?