[Intel-gfx] [PATCH 2/4] drm/i915: refcounts for forcewake

2011-04-11 Thread Ben Widawsky
The reference count provides a clean way for userspace to gain access to correct register reads and writes by initiating forcewake. The mechanism also makes it possible to limit the time struct_mutex is held on register reads and writes to the time it takes to obtain a reference. However there is

[Intel-gfx] [PATCH 2/4] drm/i915: refcounts for forcewake

2011-04-09 Thread Ben Widawsky
New refcounting mechanism for forcewake. The idea is to take away the need to hold struct_mutex for the duration of register reads, and only need to hold it to get a reference count. A possible benefit would shave off some time in actually waking the GPU if it's already awake (a couple of extra reg

[Intel-gfx] [PATCH 2/4] drm/i915: refcounts for forcewake

2011-04-09 Thread Ben Widawsky
New refcounting mechanism for forcewake. The idea is to take away the need to hold struct_mutex for the duration of register reads, and only need to hold it to get a reference count. A possible benefit would shave off some time in actually waking the GPU if it's already awake (a couple of extra reg

[Intel-gfx] [PATCH 2/4] drm/i915: refcounts for forcewake

2011-04-08 Thread Ben Widawsky
New refcounting mechanism for forcewake. The idea is to take away the need to hold struct_mutex for the duration of register reads, and only need to hold it to get a reference count. A possible benefit would shave off some time in actually waking the GPU if it's already awake (a couple of extra reg