On Tue, Apr 9, 2013 at 4:26 PM, Maarten Lankhorst <maarten.lankhorst at canonical.com> wrote: > Op 09-04-13 16:16, Jerome Glisse schreef: >> On Tue, Apr 9, 2013 at 7:37 AM, Maarten Lankhorst >> <m.b.lankhorst at gmail.com>wrote: >> >>> Signed-off-by: Maarten Lankhorst <maarten.lankhorst at canonical.com> >>> >> Can userspace pin directly ? If so then that sounds as a bad idea. >> >> Reviewed-by: Jerome Glisse <jglisse at redhat.com> >> > It's slightly better than before, it used to pin as soon as you export a > buffer for as long as the buffer exists, with the series the pinning will > only happen when importing to another device, until the device releases their > reference.
To make fully dynamic pinning work (and especially reclaiming) we first need dma_buf fences, the new ww_mutexes and all that stuff since otherwise you can trivially deadlock in buffer reservations. But I think with Maarten's stuff we have a plan to get there, and this is just the first tiny step of a long journey. -Daniel -- Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
