On Tue, May 26, 2015 at 05:25:48PM +0200, Daniel Vetter wrote: > On Tue, May 26, 2015 at 03:21:22PM +0100, Michel Thierry wrote: > > There are some allocations that must be only referenced by 32bit > > offsets. > > To limit the chances of having the first 4GB already full, > > objects not requiring this workaround don't use the first 2 PDPs. This is complete tosh. Please have a later patch that uses SEARCH_BELOW for 48bit objects, and cite eviction rates and improvements. > This is the wrong way round for existing userspace (and remember, bdw is > shipping already so we can't opt out of that). Instead the w/a needs to > allow 32bit+ addresses only if the new bit is set. > > Also this needs the usual pile of userspace enabling (libdrm+mesa). Indeed, the code is very much broken as is. So I expect to see a Testcase: igt/gem_exec_48bit in the next patch. As a hint, consider reusing the object. -Chris -- Chris Wilson, Intel Open Source Technology Centre _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx http://lists.freedesktop.org/mailman/listinfo/intel-gfx
