[openib-general] Re: Re: Re: madvise MADV_DONTFORK/MADV_DOFORK

[Gleb Natapov]

On Wed, Feb 15, 2006 at 04:58:48PM +0200, Michael S. Tsirkin wrote:
> Quoting r. Gleb Natapov <glebn at voltaire.com>:
> > Subject: Re: [openib-general] Re: Re: Re: madvise MADV_DONTFORK/MADV_DOFORK
> > 
> > On Wed, Feb 15, 2006 at 04:52:51PM +0200, Michael S. Tsirkin wrote:
> > > Quoting r. Gleb Natapov <glebn at voltaire.com>:
> > > > > > > ibv_reg_mr works fine for sub-page regions. Doesnt it?
> > > > > >
> > > > > > Not really. It gives you the impression that it works by not returning an
> > > > > > error and aligning address and lengths for you. Same case with mmap(). You
> > > > > > can provide nonaligned length and it will not fail.
> > > > > 
> > > > > As far as I know hardware supports non-aligned regions and so does
> > > > > ibv_reg_mr: try to access outside the region and you'll get completion with
> > > > > error.  No? If not, let me know - it should be easy to fix.
> > > >
> > > > Of cause you are right about reg_mr effect in regards to infiniband protocol. 
> > > > I was talked about the effect on VM subsystem. Any real program can't
> > > > ignore this.
> > > 
> > > It cant? Why does it care?
> >
> > Because the program should be careful to not put the data it needs in the
> > same page with registered buffer.
> 
> It should? Why should it?
Because after fork it may not find it. (But somehow I think you know
that.)

--
			Gleb.