Re: [PATCH 3/3] mlock: avoid dirtying pages and triggering writeback

[Michel Lespinasse <walken@google.com>]

On Wed, Nov 17, 2010 at 04:53:09PM -0800, Andrew Morton wrote:
> On Wed, 17 Nov 2010 18:52:30 -0500
> "Ted Ts'o" <tytso@mit.edu> wrote:
> 
> > On Thu, Nov 18, 2010 at 10:11:43AM +1100, Dave Chinner wrote:
> > > I don't think ->page_mkwrite can be worked around - we need that to
> > > be called on the first write fault of any mmap()d page to ensure it
> > > is set up correctly for writeback.  If we don't get write faults
> > > after the page is mlock()d, then we need the ->page_mkwrite() call
> > > during the mlock() call.
> > 
> > OK, so I'm not an mm hacker, so maybe I'm missing something.  Could
> > part of this be fixed by simply sending the write faults for
> > mlock()'ed pages, so page_mkwrite() gets called when the page is
> > dirtied.  Seems like a real waste to have the file system pre-allocate
> > all of the blocks for a mlock()'ed region.  Why does mlock() have to
> > result in the write faults getting suppressed when the page is
> > actually dirtied?

This is actually what the patch does - by having mlock() use a read fault,
pages are loaded in memory and mlocked, but the ptes are not marked as
writable so that a later write access will be caught as a write fault at
that time (with all the usual dirtying and page_mkwrite() callbacks).

> Yup, I don't think it would be too bad to take a minor fault each time
> an mlocked page transitions from clean->dirty.
> 
> In fact we should already be doing that, after the mlocked page gets
> written back by kupdate?  Hope so!

Yes, handle_mm_fault() is careful to never create writable ptes pointing
to clean file pages, so that a later write fault will correctly dirty
the corresponding page.

-- 
Michel "Walken" Lespinasse
A program is never fully debugged until the last user dies.