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

["Chen, Kenneth W" <kenneth.w.chen@intel.com>]

Adam Litke wrote on Thursday, January 12, 2006 11:49 AM
> On Thu, 2006-01-12 at 11:07 -0800, Chen, Kenneth W wrote:
> > Sorry, I don't think patch 1 by itself is functionally correct.  It opens
> > a can of worms with race window all over the place.  It does more damage
> > than what it is trying to solve.  Here is one case:
> > 
> > 1 thread fault on hugetlb page, allocate a non-zero page, insert into the
> > page cache, then proceed to zero it.  While in the middle of the zeroing,
> > 2nd thread comes along fault on the same hugetlb page.  It find it in the
> > page cache, went ahead install a pte and return to the user.  User code
> > modify some parts of the hugetlb page while the 1st thread is still
> > zeroing.  A potential silent data corruption.
> 
> I don't think the above case is possible because of find_lock_page().
> The second thread would wait on the page to be unlocked by the thread
> zeroing it before it could proceed.

I think you are correct.  Sorry for the noise.

- Ken

-
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/