memory type e820

[Sourish Mazumder]

Should there be any difference in performance while writing to a memory as
cached versus uncached mapping?

On Tue, Nov 4, 2014 at 2:16 PM, Konstantin Belousov <kostikbel at gmail.com>
wrote:

> On Mon, Nov 03, 2014 at 01:52:44PM -0500, John Baldwin wrote:
> > On Saturday 01 November 2014 18:55:53 Sourish Mazumder wrote:
> > > Hi John,
> > >
> > > I tried the pmap_mapdev() as suggested by you. Works perfectly. Thanks
> for
> > > the information.
> >
> > Sure.
> >
> > > What is required, If I want to add this nvram memory to VM pages?
> >
> > Hmm.  If this is device memory you generally don't want that.  I'm not
> > actually sure how to do this at runtime.  If you don't mind having a
> local
> > hack you can add a change in the MD startup code (e.g. in hammer_time()
> > in sys/amd64/amd64/machdep.c) to adjust the ranges added to
> > phys_avail[] and dump_avail[].
>
> The facility exists to do this.  It is OBJT_MGTDEVICE pager and
> vm_phys_fictitious_reg_range().  This is used by i915 and TTM for
> aperture, and seems XEN dom0 code uses it for mapping pages from
> other domains into dom0.
>



-- 
Sourish Mazumder
9986309755