[PATCH v5 6/6] drm/ttm: Use common ancestor of evictor and evictee as limit pool

[Natalie Vock <natalie.vock@gmx.de>]

When checking whether to skip certain buffers because they're protected
by dmem.low, we're checking the effective protection of the evictee's
cgroup, but depending on how the evictor's cgroup relates to the
evictee's, the semantics of effective protection values change.

When testing against cgroups from different subtrees, page_counter's
recursive protection propagates memory protection afforded to a parent
down to the child cgroups, even if the children were not explicitly
protected. This prevents cgroups whose parents were afforded no
protection from stealing memory from cgroups whose parents were afforded
more protection, without users having to explicitly propagate this
protection.

However, if we always calculate protection from the root cgroup, this
breaks prioritization of sibling cgroups: If one cgroup was explicitly
protected and its siblings were not, the protected cgroup should get
higher priority, i.e. the protected cgroup should be able to steal from
unprotected siblings. This only works if we restrict the protection
calculation to the subtree shared by evictor and evictee.

Signed-off-by: Natalie Vock <natalie.vock@gmx.de>
---
 drivers/gpu/drm/ttm/ttm_bo.c | 43 ++++++++++++++++++++++++++++++++++++++++---
 1 file changed, 40 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index 86f99237f6490..53b53a2791725 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -528,11 +528,48 @@ static s64 ttm_bo_evict_cb(struct ttm_lru_walk *walk, struct ttm_buffer_object *
 {
 	struct ttm_bo_evict_walk *evict_walk =
 		container_of(walk, typeof(*evict_walk), walk);
+	struct dmem_cgroup_pool_state *limit_pool, *ancestor = NULL;
+	bool evict_valuable;
 	s64 lret;
 
-	if (!dmem_cgroup_state_evict_valuable(evict_walk->alloc_state->limit_pool,
-					      bo->resource->css, evict_walk->try_low,
-					      &evict_walk->hit_low))
+	/*
+	 * If only_evict_unprotected is set, then we're trying to evict unprotected
+	 * buffers in favor of a protected allocation for charge_pool. Explicitly skip
+	 * buffers belonging to the same cgroup here - that cgroup is definitely protected,
+	 * even though dmem_cgroup_state_evict_valuable would allow the eviction because a
+	 * cgroup is always allowed to evict from itself even if it is protected.
+	 */
+	if (evict_walk->alloc_state->only_evict_unprotected &&
+			bo->resource->css == evict_walk->alloc_state->charge_pool)
+		return 0;
+
+	limit_pool = evict_walk->alloc_state->limit_pool;
+	/*
+	 * If there is no explicit limit pool, find the root of the shared subtree between
+	 * evictor and evictee. This is important so that recursive protection rules can
+	 * apply properly: Recursive protection distributes cgroup protection afforded
+	 * to a parent cgroup but not used explicitly by a child cgroup between all child
+	 * cgroups (see docs of effective_protection in mm/page_counter.c). However, when
+	 * direct siblings compete for memory, siblings that were explicitly protected
+	 * should get prioritized over siblings that weren't. This only happens correctly
+	 * when the root of the shared subtree is passed to
+	 * dmem_cgroup_state_evict_valuable. Otherwise, the effective-protection
+	 * calculation cannot distinguish direct siblings from unrelated subtrees and the
+	 * calculated protection ends up wrong.
+	 */
+	if (!limit_pool) {
+		ancestor = dmem_cgroup_common_ancestor(bo->resource->css,
+						       evict_walk->alloc_state->charge_pool);
+		limit_pool = ancestor;
+	}
+
+	evict_valuable = dmem_cgroup_state_evict_valuable(limit_pool, bo->resource->css,
+							  evict_walk->try_low,
+							  &evict_walk->hit_low);
+	if (ancestor)
+		dmem_cgroup_pool_state_put(ancestor);
+
+	if (!evict_valuable)
 		return 0;
 
 	if (bo->pin_count || !bo->bdev->funcs->eviction_valuable(bo, evict_walk->place))

-- 
2.53.0