Hello,

On Friday, August 24, 2012 3:23 PM Federico Vaga wrote:

quoted

Getting back to your patch - in your approach cpu cache handling is
missing. I suspect that it worked fine only because it has been
tested on some simple platform without any cpu caches (or with very
small ones).

Is missing from the memory allocator because I do it on the device
driver. The current operations don't allow me to do that in the memory
allocator.

Memory allocator module is much more appropriate place for it. dma-sg
allocator also needs a huge cleanup in this area...

quoted

Please check the following thread:
http://www.spinics.net/lists/linux-media/msg51768.html where Tomasz
has posted his ongoing effort on updating and extending videobuf2 and
dma-contig allocator. Especially the patch
http://www.spinics.net/lists/linux-media/msg51776.html will be
interesting for you, because cpu cache synchronization
(dma_sync_single_for_device / dma_sync_single_for_cpu) should be
called from prepare/finish callbacks.

You are right, it is interesting because avoid me to use cache sync in
my driver. Can I work on these patches?

From this page I understand that these patches are not approved yet
https://patchwork.kernel.org/project/linux-media/list/?page=2

You can take the patch which adds prepare/finish methods to memory
allocators. It should not have any dependency on the other stuff from
that thread. I'm fine with merging it either together with Your patch
or via Tomasz's patchset, whatever comes first.

Best regards
-- 
Marek Szyprowski
Samsung Poland R&D Center