[Intel-gfx] [PATCH] drm/i915: Move global activity tracking from GEM to GT

[Chris Wilson]

Quoting Tvrtko Ursulin (2019-07-23 14:34:27)
> 
> On 22/07/2019 22:46, Chris Wilson wrote:
> > Quoting Tvrtko Ursulin (2019-07-22 13:16:38)
> >>
> >> On 18/07/2019 13:42, Chris Wilson wrote:
> >>> As our global unpark/park keep track of the number of active users, we
> >>> can simply move the accounting from the GEM layer to the base GT layer.
> >>> It was placed originally inside GEM to benefit from the 100ms extra
> >>> delay on idleness, but that has been eliminated and now there is no
> >>> substantive difference between the layers. In moving it, we move another
> >>> piece of the puzzle out from underneath struct_mutex.
> >>>
> >>> Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> >>> Cc: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
> >>> ---
> >>>    drivers/gpu/drm/i915/gem/i915_gem_pm.c | 4 ----
> >>>    drivers/gpu/drm/i915/gt/intel_gt_pm.c  | 5 +++++
> >>>    2 files changed, 5 insertions(+), 4 deletions(-)
> >>>
> >>> diff --git a/drivers/gpu/drm/i915/gem/i915_gem_pm.c b/drivers/gpu/drm/i915/gem/i915_gem_pm.c
> >>> index 8faf262278ae..1afb04660725 100644
> >>> --- a/drivers/gpu/drm/i915/gem/i915_gem_pm.c
> >>> +++ b/drivers/gpu/drm/i915/gem/i915_gem_pm.c
> >>> @@ -9,7 +9,6 @@
> >>>    #include "gt/intel_gt_pm.h"
> >>>    
> >>>    #include "i915_drv.h"
> >>> -#include "i915_globals.h"
> >>>    
> >>>    static void call_idle_barriers(struct intel_engine_cs *engine)
> >>>    {
> >>> @@ -41,8 +40,6 @@ static void i915_gem_park(struct drm_i915_private *i915)
> >>>    
> >>>        intel_timelines_park(i915);
> >>>        i915_vma_parked(i915);
> >>> -
> >>> -     i915_globals_park();
> >>
> >> Benefit of this placement was also that what will be freed on parking
> >> was freed by this point so shrinking of slabs is effective. If you move
> >> it to gt parking then that runs to early. So I think it is better that
> >> it stays where it is.
> > 
> > But there's nothing here. You are imaging things. call_idle_barriers and
> > batch_pool are in engine_park, timelines_park is gone, and vma_parked is
> > now a timer.
> 
> I was thinking about slab shrinkers. They do seem to be called from 
> i915_globals_park or I am still imagining?

No, they are. I'm trying to tie together the future strands where
i915_gem_park() is empty, and all the work we have is in engine_park /
gt_park, with the outlier being the user vma cache which I plan to put
inside a timer. From that pov, I think the vma cache itself will take a
globals reference.
-Chris