On 10.06.2013 23:43, Thierry Reding wrote: > Can you post the corresponding wrappers to make it easier to discuss > them? If they just wrap runtime PM calls then they don't solve the > locality problems that Terje brought up. I don't think the wrappers are applicable here. They're there in downstream to fix a problem with runtime PM core: some host1x clients require host1x to be active when they're processing, and some (dc) need it only when CPU is programming DC or when a signal towards host1x is pending. Runtime PM wants to keep parent enabled when any of the clients are enabled. We solved this downstream by enabling ignore_children for host1x. As a side effect, we need to explicitly enable host1x when we're enabling one of its clients that need host1x, and that's what the wrapper does. We can fix this particular problem by moving calls to runtime PM to job.c. Mayuresh, once you've managed to test your patches, please send the v2. Terje _______________________________________________ dri-devel mailing list dri-devel@xxxxxxxxxxxxxxxxxxxxx http://lists.freedesktop.org/mailman/listinfo/dri-devel
