On Sun, 20 May 2012 08:45:38 -0700, Ben Widawsky <ben at bwidawsk.net> wrote: > After Daniel split out this code, I think this makes more sense, and > looks nicer. Also added some comments to help the situation. > > v2: Make signal_mbox be all rings for symmetric-ness. > v3: submitted the wrong version of the patch before. > v2 had an issue with odd number of rings. The fix is to always emit an > even number of instructions to the ring with mbox updates. > v4: I suck. The code should send out mbox updates for NUM_RINGS-1 * > number of commands. Instead of NUM_RINGS * number of commands-1. > -ret = intel_ring_begin(ring, round_up((I915_NUM_RINGS) * 3, 2) + 4); > +ret = intel_ring_begin(ring, round_up((I915_NUM_RINGS-1) * 4, 2) + 4); I'm honestly less enthralled about v4. I'd rather see that as a separate preparatory patch so that it is put into context. -Chris -- Chris Wilson, Intel Open Source Technology Centre
