Re: [PATCH] TTM DMA pool v1.8

From: Thomas Hellstrom
Date: Mon Oct 03 2011 - 12:58:31 EST

On 10/03/2011 06:46 PM, Konrad Rzeszutek Wilk wrote:

It does now (I had a spinlock mishap).. which reminds me - how
do I test these patches with your vmwgfx driver? I've an old version
of VMWare Workstation 8, would that do?

VMware workstation 8 is OK (it's actually the latest version of workstation).

You'd need the vmwgfx kernel, driver,
latest mesa master compiled with the "svga" driver and the "dri" and "xa" state trackers.
xf86-video-vmware, the "vmwgfx" branch.

And you should be fine running 3D.

The swapping code back (so from swap to pool) does not seem to
distinguish it that much - it just allocates a new page - and
then copies from whatever was in the swap cache?

This is something you were thinking to do in the future I presume?
Yes. If / when I do that, I might be adding a new backend function
to put a ttm in an
"anonymous state", that is using only pages that can be inserted in
the swap cache or passed
around to other devices, and to put a ttm in a "device" state, that
copies it to device mappable pages.
OK, that should be no trouble - we would need to expose a function
call to "detach" the page from the TTM pool (which could mean
actually allocating a new page for the "other" device, and copying
it from the "source" to "other" and then freeing the "source).

I am thinking ... you hotplug an high-end radeon while the machine has
an ATI ES1000 in it, and want to move those pages to the new
card. The ATI ES1000 can only do up to 4GB, while the new fancy card
has no such limits (and perhaps does not want to use the TTM DMA

Is this what you had in mind?
Yes, that's a typical use-case. Or passing pages between an array of GPGPUs...



To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at
Please read the FAQ at