drm/i915: Trim NEWCLIENT boosting
authorChris Wilson <chris@chris-wilson.co.uk>
Mon, 4 Feb 2019 15:01:01 +0000 (15:01 +0000)
committerChris Wilson <chris@chris-wilson.co.uk>
Mon, 4 Feb 2019 21:54:38 +0000 (21:54 +0000)
commit1413b2bc0717036a5a653eef20cc3ae4cc66501a
tree56ea272d006abf74d670dbc77fcecc9c8e813596
parent3d7a64b992eab27be75c6cf734be4e87fc88e4b5
drm/i915: Trim NEWCLIENT boosting

Limit the NEWCLIENT boost to only give its small priority boost to fresh
clients only that have no dependencies.

The idea for using NEWCLIENT boosting, commit b16c765122f9 ("drm/i915:
Priority boost for new clients"), is that short-lived streams are often
interactive and require lower latency -- and that by executing those
ahead of the long running hogs, the short-lived clients do little to
interfere with the system throughput by virtue of their short-lived
nature. However, we were only considering the client's own timeline for
determining whether or not it was a fresh stream. This allowed for
compositors to wake up before their vblank and bump all of its client
streams. However, in testing with media-bench this results in chaining
all cooperating contexts together preventing us from being able to
reorder contexts to reduce bubbles (pipeline stalls), overall increasing
latency, and reducing system throughput. The exact opposite of our
intent. The compromise of applying the NEWCLIENT boost to strictly fresh
clients (that do not wait upon anything else) should maintain the
"real-time response under load" characteristics of FQ_CODEL, without
locking together the long chains of dependencies across the system.

References: b16c765122f9 ("drm/i915: Priority boost for new clients")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190204150101.30759-1-chris@chris-wilson.co.uk
drivers/gpu/drm/i915/i915_request.c