ioremap of pci region on pSeries LPAR vs SMP

Benjamin Herrenschmidt benh at kernel.crashing.org
Sun Jan 16 09:29:21 EST 2005


On Tue, 2005-01-11 at 16:17 -0600, Linas Vepstas wrote:
> On Tue, Jan 11, 2005 at 11:08:45AM +1100, Anton Blanchard was heard to remark:
> > 
> > Ive seen HPC stuff that wants to be able to mmap a PCI cards resources into
> > userspace. Their hack on ppc64 was to look at the high nibble of the
> > address and convert it to a non EEH address if required :)
> > 
> > Im not sure how best to solve the userspace mmap issue but there are a
> > few groups wanting that.
> 
> Somewhat off-topic ... but ...
> 
> 1) If you design your hardware correctly, there are some amazing things
>    you can do (performance wise) by mmaping pci card resources into user
>    space.  If your hardwares is done right, then user corruption can't 
>    hurt the system. This was the defacto method for getting high 
>    performance graphics on IBM RS/6000, sgi, HP and Sun workstations 
>    many moons ago.

And that's exactly what X does still today on pretty much all
machines :)

> 2) There is interest in the virtual i/o community about mmaping 
>    funky stuff to userspace, but that conversation may be for a 
>    different day.  The question is (for example) how to build
>    a high-performance virtual scsi server in userspace (without
>    kernel pieces) which is a design point some people like.
>    Later...
> 
> --linas
> _______________________________________________
> Linuxppc64-dev mailing list
> Linuxppc64-dev at ozlabs.org
> https://ozlabs.org/cgi-bin/mailman/listinfo/linuxppc64-dev
-- 
Benjamin Herrenschmidt <benh at kernel.crashing.org>




More information about the Linuxppc64-dev mailing list