[PATCH] RFC drm/i915: Switch to kernel context before idling at runtime

[Chris Wilson]

We can reduce our exposure to random neutrinos by resting on the kernel
context having flushed out the user contexts to system memory and
beyond. The corollary is that we then we require two passes through the
idle handler to go to sleep, which on a truly idle system involves an
extra pass through the slow and irregular retire work handler.

Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_gem.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 795ca83aed7a..f185db9327ca 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -3554,6 +3554,18 @@ i915_gem_idle_work_handler(struct work_struct *work)
 	if (!READ_ONCE(dev_priv->gt.awake))
 		return;
 
+	/*
+	 * Flush out the last user context, leaving only the pinned
+	 * kernel context resident. When we are idling on the kernel_context,
+	 * no more new requests (with a context switch) are emitted and we
+	 * can finally rest. A consequence is that the idle work handler is
+	 * always called at least twice before idling (and if the system is
+	 * idle that implies a round trip through the retire worker).
+	 */
+	mutex_lock(&dev_priv->drm.struct_mutex);
+	i915_gem_switch_to_kernel_context(dev_priv);
+	mutex_unlock(&dev_priv->drm.struct_mutex);
+
 	/*
 	 * Wait for last execlists context complete, but bail out in case a
 	 * new request is submitted. As we don't trust the hardware, we
-- 
2.17.0