[Intel-gfx] [PATCH 00/10] shrink dev_priv by 300 lines

[Daniel Vetter]

On Sun, Nov 4, 2012 at 6:30 PM, Ben Widawsky <ben at bwidawsk.net> wrote:
>> Inspired by some recent patches I've figured I need to clean up and
>> put some organization behind our driver_private struct. It shrunk
>> from almost 500 lines to about 160. Things still there which could be
>> gathered:
>> - vbt and vbt-derived values
>> - shared/global modeset state
>> - random smaller bits and pieces
>>
>> Comments highly welcome.
>
> This is in reference to all your extraction of substructs ie. the l3
> parity...
>
> This really seems like moving around code mostly for the sake of moving
> around code (note that I said mostly). What I would have liked to see as a
> motivation/result of this is instead of passing dev_priv around all over
> the place, pass around the new extracted structures. If you still need
> dev_priv (for something other than reg reads or writes), then I don't
> think you've done anything useful.
>
> Changing the function prototypes to the new substructs would be a
> provably useful thing to do.

I've played around with passing the substructs to functions that
handle individual parts and notice that pretty much everywhere we need
dev_priv to handle reg I/O. So I've decided to keep things as-is to
avoid inconsistency (we already have that mess in the modeset code
with drm_* vs. intel_* ...).

Otoh if that "move register blocks around" thing becomes more
prevalent on vlv, we might want to switch to passing substructs
around, and also adding per-block reg I/O variants which implicitly
take the substruct and add the required mmio_base offset.

But even without that I very much like the new order in i915_dev_private.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch