On Thu, 2012-09-06 at 19:05 +0530, Archit Taneja wrote: > On Thursday 06 September 2012 06:34 PM, Tomi Valkeinen wrote: > > On Wed, 2012-09-05 at 19:01 +0530, Archit Taneja wrote: > >> On Wednesday 05 September 2012 01:55 PM, Tomi Valkeinen wrote: > >>> Now that fifo merge has been removed, nobody uses the extra_info related > >>> completion code, which can be removed. > >> > >> I think this might come into use when we fix the usage of channel out > >> field. That is, since channel out is an immediate write field, when we > >> set a new manager for an overlay, we may need to wait till the overlay > >> disable kicks in, only then we should change channel out. > >> > >> For this, we would need some wait for extra_info, right? > > > > Hmm, yes, I think you are right. Previously the ovl_disable waited until > > the overlay is not used anymore, but now it doesn't. > > > > So I think I need to add wait_pending_extra_info_updates() call to the > > beginning of dss_ovl_set_manager(). Or, should it be in unset_manager... > > I think unset is better, then a "free" overlay always disabled also in > > the HW level. > > Yes, I also think it should be in unset_manager. One option could be to > leave the wait_pending_extra_info_updates() in ovl_disable itself, as it > was before. But that would force us to use mutexes there, and we'd > rather have overlay enabling and disabling as a non blocking thing. Actually, we do have mutexes there. You are thinking about the prototype API I have. (I also thought we didn't have mutex there =). So, in fact, we can have the wait at ovl_disable like it was before. The prototype API, which cannot block, will not have the wait, but there the caller (i.e. omapdrm) will have to manage the proper wait. Tomi
Attachment:
signature.asc
Description: This is a digitally signed message part
