Re: [RFC] Remove unswappable anonymous pages off the LRU

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

On Thu, 15 Feb 2007 19:36:01 -0800 (PST) Christoph Lameter <clameter@sgi.com> wrote:

> On Thu, 15 Feb 2007, Andrew Morton wrote:
> 
> > > Yes ia64 has used the upper 32 bit. However, the lower 32 bits are fully 
> > > usable. So we have 32-20 = 12 bits to play with on 64 bit.
> > 
> > OK.  But not many things are 64-bit-only?
> 
> We could restrict some newer features to 64 bits? (ducks and runs ...)

Well.  We haven't come across many such things, and doing this would mucky
up the VM and would reduce testing coverage.  But yeah, it's always an
option if these things crop up.

> > > None of the above can occur simultaneously.
> > The actual implementation details might get messy though.  We can do a
> > non-atomic rmw of the three bits but that could corrupt a concurrent
> > modification of a different flag.  Or we could do a succession of three
> > set_bit/clear_bit operations, but that exposes intermediate invalid states.
> > 
> > It can be done I guess, but it'd be fiddly.
> 
> Right.
> 
> Maybe we could somehow splite up page->flags into 4 separate bytes?
> Updating one byte would not endanger the other bytes in the other 
> sets?

yipes.  I'm not sure that'd work?

compare-and-swap-in-a-loop could be used, I guess.  With the obvious problem..

I do think that those two swsusp flags are low-hanging-fruit.  It'd be
trivial to vmalloc a bitmap or use a radix-tree-holding-longs, but I have a
vague feeling that there were subtle issues with that.  Still, Something
Needs To Be Done.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>