[for-next,3/5] IB/hfi1: Fix SDMA mmu_rb_node not being evicted in LRU order

Commit Message

Dennis Dalessandro April 7, 2023, 4:52 p.m. UTC

From: Patrick Kelsey <[e-mail]>

hfi1_mmu_rb_remove_unless_exact() did not move mmu_rb_node objects in
mmu_rb_handler->lru_list after getting a cache hit on an mmu_rb_node.

As a result, hfi1_mmu_rb_evict() was not guaranteed to evict truly
least-recently used nodes.

This could be a performance issue for an application when that
application:
- Uses some long-lived buffers frequently.
- Uses a large number of buffers once.
- Hits the mmu_rb_handler cache size or pinned-page limits, forcing
  mmu_rb_handler cache entries to be evicted.

In this case, the one-time use buffers cause the long-lived buffer
entries to eventually filter to the end of the LRU list where
hfi1_mmu_rb_evict() will consider evicting a frequently-used long-lived
entry instead of evicting one of the one-time use entries.

Fix this by inserting new mmu_rb_node at the tail of
mmu_rb_handler->lru_list and move mmu_rb_ndoe to the tail of
mmu_rb_handler->lru_list when the mmu_rb_node is a hit in
hfi1_mmu_rb_remove_unless_exact(). Change hfi1_mmu_rb_evict() to evict
from the head of mmu_rb_handler->lru_list instead of the tail.

Fixes: 0636e9ab8355 ("IB/hfi1: Add cache evict LRU list")
Signed-off-by: Brendan Cunningham <[e-mail]>
Signed-off-by: Patrick Kelsey <[e-mail]>
Signed-off-by: Dennis Dalessandro <[e-mail]>
---
 drivers/infiniband/hw/hfi1/mmu_rb.c |   13 ++++++-------
 1 file changed, 6 insertions(+), 7 deletions(-)

Patch

diff --git a/drivers/infiniband/hw/hfi1/mmu_rb.c b/drivers/infiniband/hw/hfi1/mmu_rb.c
index 7333646021bb..af46ff203342 100644
--- a/drivers/infiniband/hw/hfi1/mmu_rb.c
+++ b/drivers/infiniband/hw/hfi1/mmu_rb.c
@@ -130,7 +130,7 @@  int hfi1_mmu_rb_insert(struct mmu_rb_handler *handler,
 		goto unlock;
 	}
 	__mmu_int_rb_insert(mnode, &handler->root);
-	list_add(&mnode->list, &handler->lru_list);
+	list_add_tail(&mnode->list, &handler->lru_list);
 
 	ret = handler->ops->insert(handler->ops_arg, mnode);
 	if (ret) {
@@ -181,8 +181,10 @@  bool hfi1_mmu_rb_remove_unless_exact(struct mmu_rb_handler *handler,
 	spin_lock_irqsave(&handler->lock, flags);
 	node = __mmu_rb_search(handler, addr, len);
 	if (node) {
-		if (node->addr == addr && node->len == len)
+		if (node->addr == addr && node->len == len) {
+			list_move_tail(&node->list, &handler->lru_list);
 			goto unlock;
+		}
 		__mmu_int_rb_remove(node, &handler->root);
 		list_del(&node->list); /* remove from LRU list */
 		ret = true;
@@ -206,8 +208,7 @@  void hfi1_mmu_rb_evict(struct mmu_rb_handler *handler, void *evict_arg)
 	INIT_LIST_HEAD(&del_list);
 
 	spin_lock_irqsave(&handler->lock, flags);
-	list_for_each_entry_safe_reverse(rbnode, ptr, &handler->lru_list,
-					 list) {
+	list_for_each_entry_safe(rbnode, ptr, &handler->lru_list, list) {
 		if (handler->ops->evict(handler->ops_arg, rbnode, evict_arg,
 					&stop)) {
 			__mmu_int_rb_remove(rbnode, &handler->root);
@@ -219,9 +220,7 @@  void hfi1_mmu_rb_evict(struct mmu_rb_handler *handler, void *evict_arg)
 	}
 	spin_unlock_irqrestore(&handler->lock, flags);
 
-	while (!list_empty(&del_list)) {
-		rbnode = list_first_entry(&del_list, struct mmu_rb_node, list);
-		list_del(&rbnode->list);
+	list_for_each_entry_safe(rbnode, ptr, &del_list, list) {
 		handler->ops->remove(handler->ops_arg, rbnode);
 	}
 }