On Thu, Jul 04, 2019 at 02:14:28PM +0200, Arnd Bergmann wrote:
> On Thu, Jul 4, 2019 at 12:31 PM Ilias Apalodimas
> <ilias.apalodi...@linaro.org> wrote:
> > > On Wed,  3 Jul 2019 12:37:50 +0200
> > > Jose Abreu <jose.ab...@synopsys.com> wrote:
> 
> > 1. page pool allocs packet. The API doesn't sync but i *think* you don't 
> > have to
> > explicitly since the CPU won't touch that buffer until the NAPI handler 
> > kicks
> > in. On the napi handler you need to dma_sync_single_for_cpu() and process 
> > the
> > packet.
> 
> > So bvottom line i *think* we can skip the dma_sync_single_for_device() on 
> > the
> > initial allocation *only*. If am terribly wrong please let me know :)
> 
> I think you have to do a sync_single_for_device /somewhere/ before the
> buffer is given to the device. On a non-cache-coherent machine with
> a write-back cache, there may be dirty cache lines that get written back
> after the device DMA's data into it (e.g. from a previous memset
> from before the buffer got freed), so you absolutely need to flush any
> dirty cache lines on it first.
Ok my bad here i forgot to add "when coherency is there", since the driver
i had in mind runs on such a device (i think this is configurable though so i'll
add the sync explicitly to make sure we won't break any configurations).

In general you are right, thanks for the explanation!
> You may also need to invalidate the cache lines in the following
> sync_single_for_cpu() to eliminate clean cache lines with stale data
> that got there when speculatively reading between the cache-invalidate
> and the DMA.
> 
>        Arnd


Thanks!
/Ilias

Reply via email to