[Intel-gfx] [RFC 3/4] drm/i915: Return immediately if trylock fails for direct-reclaim

[Tvrtko Ursulin]

From: Tvrtko Ursulin <tvrtko.ursulin at intel.com>

In an equally named patch Chris Wilson proposes skipping the attempts to
obtain struct_mutex via trylock-looping in order to improve latency for
non-i915 clients. I quote:

"""
Ignore trying to shrink from i915 if we fail to acquire the struct_mutex
in the shrinker while performing direct-reclaim. The trade-off being
(much) lower latency for non-i915 clients at an increased risk of being
unable to obtain a page from direct-reclaim without hitting the
oom-notifier. The proviso being that we still keep trying to hard
obtain the lock for oom so that we can reap under heavy memory pressure.
"""

This version of the patch does strictly what the commit message explains
and leaves other changes out.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
Cc: Chris Wilson <chris at chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_gem_shrinker.c | 30 ++++++++++++++----------
 1 file changed, 17 insertions(+), 13 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem_shrinker.c b/drivers/gpu/drm/i915/i915_gem_shrinker.c
index 586acf02727e..97deec775f40 100644
--- a/drivers/gpu/drm/i915/i915_gem_shrinker.c
+++ b/drivers/gpu/drm/i915/i915_gem_shrinker.c
@@ -36,7 +36,8 @@
 #include "i915_drv.h"
 #include "i915_trace.h"
 
-static bool shrinker_lock(struct drm_i915_private *i915, bool *unlock)
+static bool
+shrinker_lock(struct drm_i915_private *i915, unsigned int flags, bool *unlock)
 {
 	switch (mutex_trylock_recursive(&i915->drm.struct_mutex)) {
 	case MUTEX_TRYLOCK_RECURSIVE:
@@ -45,15 +46,17 @@ static bool shrinker_lock(struct drm_i915_private *i915, bool *unlock)
 
 	case MUTEX_TRYLOCK_FAILED:
 		*unlock = false;
-		preempt_disable();
-		do {
-			cpu_relax();
-			if (mutex_trylock(&i915->drm.struct_mutex)) {
-				*unlock = true;
-				break;
-			}
-		} while (!need_resched());
-		preempt_enable();
+		if (flags & I915_SHRINK_ACTIVE) {
+			preempt_disable();
+			do {
+				cpu_relax();
+				if (mutex_trylock(&i915->drm.struct_mutex)) {
+					*unlock = true;
+					break;
+				}
+			} while (!need_resched());
+			preempt_enable();
+		}
 		return *unlock;
 
 	case MUTEX_TRYLOCK_SUCCESS:
@@ -160,7 +163,8 @@ i915_gem_shrink(struct drm_i915_private *i915,
 	unsigned long scanned = 0;
 	bool unlock = false;
 
-	if (!(flags & I915_SHRINK_LOCKED) && !shrinker_lock(i915, &unlock))
+	if (!(flags & I915_SHRINK_LOCKED) &&
+	    !shrinker_lock(i915, flags, &unlock))
 		return 0;
 
 	/*
@@ -359,7 +363,7 @@ i915_gem_shrinker_scan(struct shrinker *shrinker, struct shrink_control *sc)
 
 	sc->nr_scanned = 0;
 
-	if (!shrinker_lock(i915, &unlock))
+	if (!shrinker_lock(i915, flags, &unlock))
 		return SHRINK_STOP;
 
 	freed = i915_gem_shrink(i915,
@@ -393,7 +397,7 @@ shrinker_lock_uninterruptible(struct drm_i915_private *i915, bool *unlock,
 
 	do {
 		if (i915_gem_wait_for_idle(i915, 0, timeout) == 0 &&
-		    shrinker_lock(i915, unlock))
+		    shrinker_lock(i915, 0, unlock))
 			break;
 
 		schedule_timeout_killable(1);
-- 
2.19.1