Pushing image transport logic down the stack

[Keith Packard]

On Tue, 2006-09-05 at 01:18 +0100, Alan Cox wrote:

> Dear gods no, please don't go that way. TLB shootdown is expensive and
> does not scale in a pretty way. Far better simply to have a flag
> attached to the pages which the kernel can atomic operate on (or use a
> futex for that matter).

So what hardware changes would be needed to make this viable? We're
already sacrificing stability for performance, PTE mangling might give
us some semblance of security at least.

It would help me tremendously if we could start framing a piece of the
conversation around how the existing chip VM mechanisms could be fixed;
we've got piles of transistors, and a lot of interest in making them
more useful than they are today.

> The killer is the without pre-planning side of it. If you used posix
> shared memory objects and allocated those sanely instead of passing
> arbitary address space it would all be there already I think.

Yeah, I think Owen has fairly accurately tagged our needs by noting the
extremely limited utility of static object sharing. To make this viable,
we really need to be able to push a piece of our address space to
another agent. Whether that needs to be the X server is far from clear
though; any interesting performance would come only from getting DMA
straight from the application to the graphics hardware.

-- 
keith.packard at intel.com
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
URL: <http://lists.x.org/archives/xorg/attachments/20060905/fe5aa5ff/attachment.pgp>