Reduce fragmentation in the slab allocator

This reverts a change brought when reworking slab lists handling that made
the allocator store slabs in LIFO order, whatever their reference count.
While it's fine for free slabs, it actually increased fragmentation for
partial slabs.

* kern/slab.c (kmem_cache_alloc_from_slab): Insert slabs that become partial
at the end of the partial slabs list.

1 files changed, 5 insertions, 2 deletions

diff --git a/kern/slab.c b/kern/slab.c
index 0a4dbdf..d14d07a 100644
--- a/kern/slab.c
+++ b/kern/slab.c
@@ -955,9 +955,12 @@ static void * kmem_cache_alloc_from_slab(struct kmem_cache *cache)
         if (slab->nr_refs == 1)
             cache->nr_free_slabs--;
     } else if (slab->nr_refs == 1) {
-        /* The slab has become partial */
+        /*
+         * The slab has become partial. Insert the new slab at the end of
+         * the list to reduce fragmentation.
+         */
         list_remove(&slab->list_node);
-        list_insert_head(&cache->partial_slabs, &slab->list_node);
+        list_insert_tail(&cache->partial_slabs, &slab->list_node);
         cache->nr_free_slabs--;
     }