Re: Active Memory Defragmentation: Our implementation & problems

[Dave Hansen]

On Wed, 2004-02-04 at 10:54, Alok Mooley wrote:
> --- Dave Hansen <haveblue@xxxxxxxxxx> wrote:
> 
> > The "work until we get interrupted and restart if
> > something changes
> > state" approach is very, very common.  Can you give
> > some more examples
> > of just how a page fault would ruin the defrag
> > process?
> > 
> 
> What I mean to say is that if we have identified some
> pages for movement, & we get preempted, the pages
> identified as movable may not remain movable any more
> when we are rescheduled. We are left with the task of
> identifying new movable pages.

Depending on the quantity of work that you're trying to do at once, this
might be unavoidable.  

I know it's a difficult thing to think about, but I still don't
understand the precise cases that you're concerned about.  Page faults
to me seem like the least of your problems.  A bigger issue would be if
the page is written to by userspace after you copy, but before you
install the new pte.  Did I miss the code in your patch that invalidated
the old tlb entries?

--dave

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/