On Mon, 2008-08-04 at 22:25 -0400, John Stoffel wrote:

quoted text
> What about the onboard memory of graphics cards?  Isn't that where
> Textures and such are stored as well?  So once something is loaded to
> the card, shouldn't you be able to free it in system memory?  Or swap
> it out ahead of time? =20

Right now, I'm working only on Intel integrated graphics, which doesn't
have any on-board memory. My thinking is that we'd best solve the
easiest case first before attempting the harder discrete graphics
driver. Plus, Intel pays me do do integrated graphics, so I have an
incentive.

Not that we haven't been thinking about how this plays with a discrete
card; we'll need to allocate backing store for any objects which are
placed in on-card memory in case we run out of on-card memory, and also
as a place to store objects when the system suspends. From the
perspective of shmem, it should be precisely the same; allocating
anonymous pages and handing them to the DRM driver to store video card
data.

quoted text
> I just wonder here, since GPUs are different from other drivers in
> that (I assume) they are written too much more often than they are
> read to, that they should hopefully be much more amenable to drop
> behind semantics for pages they've been sent, to free system
> resources.

Yup, I think you understand the situation fairly well. I've got a big
pile of pages sitting idle in the GTT which could be pulled out and
given back to the system. I like to keep a bunch around as the cost of
getting pages out of the CPU cache and bound to the GTT is non-trivial,
but if there's any memory pressure, the cost to free them is quite low.

--=20
keith.packard@intel.com