thinking about Haswell desktop

Lennart Sorensen lsorense-1wCw9BSqJbv44Nm34jS7GywD8/FfD2ys at public.gmane.org
Fri Sep 6 15:06:09 UTC 2013


On Fri, Sep 06, 2013 at 10:54:45AM -0400, D. Hugh Redelmeier wrote:
> Yeah.  I'm also confused about any relationship between Asus and
> ASRock (Pegatron).  Apparently ASUStek spun out Pegatron in 2010.
> Afterwards Pegatron still manufactured many Asus boards.

Asus used to own pegatron as far as I understand it (and I think still
own 25% or so).  Pegatron does manufacture a lot of boards for Asus,
although Asus does the design of course.

ASrock also used to be part of Asus but was spun of a couple of years
ago as far as I understand it, and is now on their own designing boards.
So as far as support and RMA and such is concerned, they are very much
not Asus, even if their design engineers used to work together with those
at Asus.  I think ASrock was intended as a lower end brand compared to
Asus, although now that they are on their own they are starting to aim
higher too.

> I remember when XEN was introduced (before VT-D and IOMMU), at an OLS
> presentation, XEN folks claimed:
> - XEN was a secure way of sharing a machine
> - XEN could selectively allocate PCI devices directly to guests
> 
> Of course either might be true at one time, but both cannot be true at
> one time.  There was no way to give a guest direct access to only a
> part of the PCI bus.  But they didn't say that.  A clue to me that the
> presentation was too close to marketing.

Well I think xen did it using paravirt and such, while with VT-d the
guest can run the native driver for the hardware.  I could be wrong,
since I have never dealt with any of that.

> With VT-d/IOMMU, both could be true at one time.
> 
> Giving a guest direct access to a PCI bus device could be a big
> performance boost.  Think of any high-bandwidth device.  A NIC or a
> disk controller comes to mind.  Without doing any measurements, I
> cannot be sure that the performance improvement is significant.
> 
> I really want some cards to be willing to provide virtual sub-cards.
> The greatest need is in video cards.  Allowing unpriviledged processes
> (and VM guests) constrained direct access to parts of a card would
> seem to be a big performance win.

I think some cards do have hardware to support that.  I think even some
newer video cards have some capability for allowing that.  I have seen
network cards that could supposedly do that.

> To be honest, I don't understand the security model of "direct
> rendering" or whatever it is that allows high performance graphics
> from userland.  I fear it is insecure.

Well given X used to use direct memory mapping of the video hardware to
userspace, DRI has to be more secure than that.

-- 
Len Sorensen
--
The Toronto Linux Users Group.      Meetings: http://gtalug.org/
TLUG requests: Linux topics, No HTML, wrap text below 80 columns
How to UNSUBSCRIBE: http://gtalug.org/wiki/Mailing_lists





More information about the Legacy mailing list