On Tue, Jul 2, 2013 at 6:51 PM, Ben Widawsky <ben at bwidawsk.net> wrote: > On Tue, Jul 02, 2013 at 09:28:56AM +0200, Daniel Vetter wrote: >> On Mon, Jul 01, 2013 at 03:59:51PM -0700, Ben Widawsky wrote: >> > On Mon, Jul 01, 2013 at 09:08:58PM +0200, Daniel Vetter wrote: >> > > On Mon, Jul 1, 2013 at 8:43 PM, Daniel Vetter <daniel at ffwll.ch> wrote: >> > > >> All of this is addressed in future patches. As we've discussed, I think >> > > >> I'll have to respin it anyway, so I'll name it as such upfront. To me it >> > > >> felt a little weird to start calling things "ggtt" before I made the >> > > >> separation. >> > > > >> > > > I think now that we know what the end result should (more or less at >> > > > least) look like we can aim to make it right the first time we touch a >> > > > piece of code. That will reduce the churn in the patch series and so >> > > > make the beast easier to review. >> > > > >> > > > Imo foreshadowing (to keep consistent with the "a patch series should >> > > > tell a story" analogy) is perfectly fine, and in many cases helps in >> > > > understanding the big picture of a large pile of patches. >> > > >> > > I've forgotten to add one thing: If you switch these again later on >> > > (layz me didn't check for that) it's imo best to stick with those >> > > names (presuming they fit, since the gtt_size vs. obj->size >> > > disdinction is a rather important one). Again I think now that we know >> > > where to go to it's best to get there with as few intermediate steps >> > > as possible. >> > > -Daniel >> > > >> > >> > I don't recall object size being very important actually, so I don't >> > think the distinction is too important, but I'm just arguing for the >> > sake of arguing. With the sg page stuff that Imre did, I think most size >> > calculations unrelated to gtt size are there anyway, and most of our mm >> > (not page allocation) code should only ever care about the gtt. >> >> The disdinction is only important on gen2/3, which is why you don't recall >> it being important ;-) >> >> I think you have two options: >> - Trust me that it's indeed important. >> - Read up on gen2/3 fencing code and make up your own mind. >> >> Cheers, Daniel > I am not saying the distinction doesn't exist. I was saying it's not > prevalent in too many places. See the second part of my statement, I > believe it holds. > > But please be clear what you're asking for, do you want 2 getters for > vma size vs. obj size? I guess we need a few different things: - vma->node.size: Most functions can probably just access that one directly (e.g. for writing the ptes), maybe we need something for transition. I guess we could call it obj_gtt_size since while transition there's only the one gtt address space. - The obj size in the gtt. Differs from obj->size on gen2/3 for tiled objects (and has pretty funny rules at that how it changes). Mostly important for debugfs, for the gtt stats. Any access helper which just calls this obj_size is imo highly misleading, hence why I've voted for for obj_gtt_size. - I don't think we need an access helper for obj->size itself, since that is invariant and I don't think we'll move it around either. Cheers, Daniel -- Daniel Vetter Software Engineer, Intel Corporation +41 (0) 79 365 57 48 - http://blog.ffwll.ch
