On Thu, Sep 09, 2010 at 10:10:00AM -0700, David Daney wrote: > On 09/08/2010 04:02 PM, Kevin Cernekee wrote: >> On noncoherent processors with a readahead cache, an extra platform- >> specific invalidation is required during the dma_sync_*_for_cpu() >> operations to keep drivers from seeing stale prefetched data. > > But as far as I can see, none of your plat_extra_sync_for_cpu() do anything. > > Perhaps adding this hook should be deferred until there is actually a user. looks like this is doing what the non_coherent_r10000 case does. So IMHO either which make non_coheren check more generic or could use the new plat_sync thingie for IP28 and other non coherent r10k boxes. Thomas. -- Crap can work. Given enough thrust pigs will fly, but it's not necessary a good idea. [ RFC1925, 2.3 ]
