On Tue, Aug 26, 2014 at 10:29:23AM -0300, Paulo Zanoni wrote: > 2014-08-26 10:18 GMT-03:00 Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>: > > On Tue, Aug 26, 2014 at 10:04:22AM -0300, Paulo Zanoni wrote: > >> Of course, we can also implement the one-shot thing on top of the > >> above, but it won't really help us reducing the amount of reads on the > >> "happy case" where we never got the error before. > > > > Actually I am tempted to dynamically patch the mmio vfuncs to avoid even > > the forcewake spinlock when we already hold it. So there won't be any > > such logic except when enabled by the user. > > Should I expect a patch from you, or should I go and write the patch > based on what we already discussed? Imo this is crazy - we have no control over what the compiler does and when exactly it loads vtable entries, so patching them at runtime would be an interesting excercise at best. Adding a special version of I915_READ/WRITE for the irq hotpath makes sense, but only if we can actually show a benefit in benchmarks. -Daniel -- Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx http://lists.freedesktop.org/mailman/listinfo/intel-gfx
