Re: Nick's core remove PageReserved broke vmware...

[Gleb Natapov <gleb@minantech.com>]

On Thu, Nov 10, 2005 at 01:21:25PM +0000, Hugh Dickins wrote:
> On Thu, 10 Nov 2005, Michael S. Tsirkin wrote:
> > Quoting Gleb Natapov <gleb@minantech.com>:
> > > On Thu, Nov 10, 2005 at 02:48:53PM +0200, Michael S. Tsirkin wrote:
> > > > > Also perhapse we should skip VM_SHARED VMAs?
> > > > 
> > > > Why?
> > > > 
> > > They will work correctly across fork(). 
> > 
> > So why would I call madvise on such a VMA?
> 
> To avoid the overhead of forking it e.g. if it's a large nonlinear vma,
> a lot of time may be wasted on copying its ptes for fork.  That's one
> of the reasons I came to like your DONTCOPY.
> 
> So, it may not be useful for your particular RDMA issue, but I see no
> reason to exclude VM_SHARED vmas from the madvise, and good reason to
> include them.
> 
If the scope of DONTCOPY is more broad that just RDMA then I agree.

--
			Gleb.