[Intel-gfx] [PATCH 5/5] drm/i915: Tidy batch pool logic
Chris Wilson
chris at chris-wilson.co.uk
Fri Feb 13 06:57:45 PST 2015
On Fri, Feb 13, 2015 at 02:00:50PM +0000, John Harrison wrote:
> >+ list_move_tail(&obj->batch_pool_list, &pool->cache_list);
> Why is it now safe to do a move_tail instead of add_tail if the node
> has just been allocated? Was the original add_tail() wrong or am I
> not spotting some critical difference to how new pool objects are
> created?
The link is initialised in i915_gem_object_init(). It was always safe to
use list_move_tail.
> >+ i915_gem_object_pin_pages(obj);
> Is it worth updating the function description comment to add a line
> about the returned buffer now being pinned and the caller must worry
> about unpinning it?
Didn't even spot that there was a function description. The other choice
is to just push the pinning into the caller, the emphasis was on moving
get_pages() into the allocator, and so for consistency it should also
pin the pages. Will update.
Now I just want to rename it from batch pool to buffer pool...
-Chris
--
Chris Wilson, Intel Open Source Technology Centre
More information about the Intel-gfx
mailing list