>Richard Röjfors <richard.rojfors@xxxxxxxxxxxxxx> wrote: >On 08/04/2010 12:34 PM, Pawel Osciak wrote: >>> Richard Röjfors<richard.rojfors@xxxxxxxxxxxxxx> wrote: >>> On 08/04/2010 11:50 AM, Pawel Osciak wrote: >>>>> >>>>> How do you propose to allocate the buffers? They need to be contiguous >>>>> and using uncached memory gave really bad performance. >>>> >>>> 829440 bytes is a quite a lot and one can't reliably depend on kmalloc >>>> to be able to allocate such big chunks of contiguous memory. Were you >>>> testing this on a freshly rebooted system? >>> >>> The systems have been running for a while, but not days. >>> I don't see why would dma_alloc_coherent work better than kmalloc? >>> >> >> In principle it wouldn't. It's just it's much less intensively used and >> allocates from a special area. Not really a bullet-proof solution either >> though, I agree. > >So how do we move forward? I would like to see this kind of patch go in, it >obviously makes our video driver useful. > >I could change and verify the patch using dma_alloc_noncoherent instead of >kmalloc. It would have the same "limitations" as todays' uncached buffers. By the way, I am only vaguely familiar with it, but isn't the memory always coherent on Atom? Best regards -- Pawel Osciak Linux Platform Group Samsung Poland R&D Center -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
