On Wed, Sep 25, 2013 at 10:11:35AM -0700, Ben Widawsky wrote: > On Wed, Sep 25, 2013 at 05:05:15PM +0200, Daniel Vetter wrote: > > Otherwise we can die in a fire of not-yet-allocated lazy requests when > > we expect them to be there: > > And Chris accuses me of violating Rusty's rules. This is an extremely > ugly caveat to put in an interface given that active tracking and ring > dispatch should have little connection. > > > Isn't it much simpler to just call intel_ring_alloc_seqno during > move_to_active? Daniel keeps harping on about the potential allocation resulting in the shrinker stealing vma and pages. All because we cheated and dropped the pin early during reservation... -Chris -- Chris Wilson, Intel Open Source Technology Centre _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx http://lists.freedesktop.org/mailman/listinfo/intel-gfx
