Re: [PATCH 2/2] hugetlb: synchronize alloc with page cache insert

[Adam Litke <agl@us.ibm.com>]

On Wed, 2006-01-11 at 17:05 -0800, William Lee Irwin III wrote:
> On Wed, Jan 11, 2006 at 04:40:37PM -0800, Chen, Kenneth W wrote:
> > What if two processes fault on the same page and races with find_lock_page(),
> > both find page not in the page cache.  The process won the race proceed to
> > allocate last hugetlb page.  While the other will exit with SIGBUS.
> > In theory, both processes should be OK.
> 
> This is supposed to fix the incarnation of that as a preexisting
> problem, but you're right, there is no fallback or retry for the case
> of hugepage queue exhaustion. For some reason I saw a phantom page
> allocator fallback in the hugepage allocator changes.
> 
> Looks like back to the drawing board for this pair of patches, though
> I'd be more than happy to get a solution to this.

I still think patch 1 (delayed zeroing) is a good thing to have.  It
will definitely improve performance for multi-threaded hugetlb
applications by avoiding unnecessary hugetlb page zeroing.  It also
shrinks the race window we have been talking about to a tiny fraction of
what it was.  This should ease the problem while we figure out a way to
handle the "last free page" case.

-- 
Adam Litke - (agl at us.ibm.com)
IBM Linux Technology Center

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