On 22/12/2017 11:33, Chris Wilson wrote:
We want to expose the parameters for controlling how long it takes for
us to notice and park the GPU after a stream of requests in order to try
and tune the optimal power-efficiency vs latency of a mostly idle system.
Extra space.
Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
Cc: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx>
---
drivers/gpu/drm/i915/Kconfig.profile | 20 ++++++++++++++++++++
drivers/gpu/drm/i915/i915_gem.c | 12 +++++++++++-
drivers/gpu/drm/i915/i915_gem_request.c | 2 +-
3 files changed, 32 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/i915/Kconfig.profile b/drivers/gpu/drm/i915/Kconfig.profile
index 8a230eeb98df..02253e7c18e1 100644
--- a/drivers/gpu/drm/i915/Kconfig.profile
+++ b/drivers/gpu/drm/i915/Kconfig.profile
@@ -24,3 +24,23 @@ config DRM_I915_SPIN_REQUEST_CS
spin prior to sleeping again.
May be 0 to disable spinning after being woken.
+
+config DRM_I915_GEM_RETIRE_DELAY
+ int
+ default 1000 # milliseconds
+ help
+ Before parking the engines and the GPU after the final request is
Copy-paste error.
+ We maintain a background job whose purpose is to keep cleaning up
+ after userspace, and will be the first to spot an idle system. This
+ parameter determines th interval between execution of the woker.
s/th/the/,s/woker/worker/
+
+config DRM_I915_GEM_PARK_DELAY
+ int
+ default 100 # milliseconds
+ help
+ Before parking the engines and the GPU after the final request is
+ retired, we may wait for a small delay to reduce the frequecy of
frequency
+ having to park/unpark and so the latency in executing a new request.
+
+ May be 0 to immediately start parking the engines after the last
+ request.
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index ba9f67c256f4..ba15f0edda26 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -3310,6 +3310,16 @@ bool i915_gem_unset_wedged(struct drm_i915_private *i915)
return true;
}
+static inline long retire_delay(void)
+{
+ const long delay = msecs_to_jiffies(CONFIG_DRM_I915_GEM_RETIRE_DELAY);
+
+ if (CONFIG_DRM_I915_GEM_RETIRE_DELAY >= 1000)
+ return round_jiffies_up_relative(delay);
Round to a second? I am not sure we want that if exposing a tunable, but
at least need to say in Kconfig. But I don't see why we would round as I
said.
+ else
+ return delay;
+}
+
static void
i915_gem_retire_work_handler(struct work_struct *work)
{
@@ -3331,7 +3341,7 @@ i915_gem_retire_work_handler(struct work_struct *work)
i915_queue_hangcheck(dev_priv);
queue_delayed_work(dev_priv->wq,
&dev_priv->gt.retire_work,
- round_jiffies_up_relative(HZ));
+ retire_delay());
}
i915_gem_request.c/mark_busy needs to use it as well.
}
diff --git a/drivers/gpu/drm/i915/i915_gem_request.c b/drivers/gpu/drm/i915/i915_gem_request.c
index bb8b45d4d04d..1c6d7d18e5ee 100644
--- a/drivers/gpu/drm/i915/i915_gem_request.c
+++ b/drivers/gpu/drm/i915/i915_gem_request.c
@@ -312,7 +312,7 @@ static void unreserve_engine(struct intel_engine_cs *engine)
GEM_BUG_ON(!i915->gt.awake);
mod_delayed_work(i915->wq,
&i915->gt.idle_work,
- msecs_to_jiffies(100));
+ msecs_to_jiffies(CONFIG_DRM_I915_GEM_PARK_DELAY));
}
GEM_BUG_ON(!engine->timeline->inflight_seqnos);
Regards,
Tvrtko
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
