Re: [PATCH 2/2] msync: start async writeout when MS_ASYNC

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

On Thu, 14 Jun 2012 11:02:02 +0200 Paolo Bonzini <pbonzini@redhat.com> wrote:

> Il 13/06/2012 23:29, Andrew Morton ha scritto:
> > > If the application does not want to start I/O, it can simply call msync
> > > with flags equal to MS_INVALIDATE.  This one remains a no-op, as it should
> > > be on a reasonable implementation.
> > 
> > Means that people will find that their msync(MS_ASYNC) call will newly
> > start IO.  This may well be undesirable for some.
> 
> If they knew it was a no-op, and relying on it, they might as well not
> have called it at all and save a syscall.

Nope.  Back in the day (3+ years ago?), msync(MS_ASYNC) would flush
pte-dirty bits into page->flags:PG_Dirty.  So it was used to make the
filesystem aware that userspace had modified some MAP_SHARED memory. 
The fs would then perform writeback within (typically) 30 seconds.  So
a legitimate use would be for the app to periodically run
msync(MS_ASYNC) to avoid having modified data floating about
un-written-back for arbitrarily long periods.

Nowadays we mark the pte's read-only and take a fault on first write,
and mark the page dirty at that time.

So it wasn't a no-op, and with this patch, such applications will newly
find themselves to be starting I/O within the msync() call.