drm/i915: Skip switch to kernel context if already done

Some engines are never user or already sitting idle in the kernel
context and for those we can skip flushing the current context for
i915_gem_switch_to_kernel_context(). We used to perform this
optimisation but that was removed for convenience of converting over to
multiple timelines and handling the pending request queues.

From the perspective of writing selftests, reducing the number of
background operations on the engines makes defining assertions easier.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/20170114162334.10271-2-chris@chris-wilson.co.uk
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index d4c59b5..83cd2ef 100644 (file)
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -4140,7 +4140,8 @@ static void assert_kernel_context_is_current(struct drm_i915_private *dev_priv)
        enum intel_engine_id id;
 
        for_each_engine(engine, dev_priv, id)
-               GEM_BUG_ON(!i915_gem_context_is_kernel(engine->last_retired_context));
+               GEM_BUG_ON(engine->last_retired_context &&
+                          !i915_gem_context_is_kernel(engine->last_retired_context));
 }
 
 int i915_gem_suspend(struct drm_i915_private *dev_priv)

diff --git a/drivers/gpu/drm/i915/i915_gem_context.c b/drivers/gpu/drm/i915/i915_gem_context.c
index ae99c25..0a4728f 100644 (file)
--- a/drivers/gpu/drm/i915/i915_gem_context.c
+++ b/drivers/gpu/drm/i915/i915_gem_context.c
@@ -877,6 +877,26 @@ int i915_switch_context(struct drm_i915_gem_request *req)
        return do_rcs_switch(req);
 }
 
+static bool engine_has_kernel_context(struct intel_engine_cs *engine)
+{
+       struct i915_gem_timeline *timeline;
+
+       list_for_each_entry(timeline, &engine->i915->gt.timelines, link) {
+               struct intel_timeline *tl;
+
+               if (timeline == &engine->i915->gt.global_timeline)
+                       continue;
+
+               tl = &timeline->engine[engine->id];
+               if (i915_gem_active_peek(&tl->last_request,
+                                        &engine->i915->drm.struct_mutex))
+                       return false;
+       }
+
+       return (!engine->last_retired_context ||
+               i915_gem_context_is_kernel(engine->last_retired_context));
+}
+
 int i915_gem_switch_to_kernel_context(struct drm_i915_private *dev_priv)
 {
        struct intel_engine_cs *engine;
@@ -885,10 +905,15 @@ int i915_gem_switch_to_kernel_context(struct drm_i915_private *dev_priv)
 
        lockdep_assert_held(&dev_priv->drm.struct_mutex);
 
+       i915_gem_retire_requests(dev_priv);
+
        for_each_engine(engine, dev_priv, id) {
                struct drm_i915_gem_request *req;
                int ret;
 
+               if (engine_has_kernel_context(engine))
+                       continue;
+
                req = i915_gem_request_alloc(engine, dev_priv->kernel_context);
                if (IS_ERR(req))
                        return PTR_ERR(req);