It is no need to find the very beginning of the area within
alloc_vmap_area, which can be done by judging each node during the process
free_vmap_cache miss:
vmap_area_root
/ \
tmp_next U
/ (T1)
tmp
/
... (T2)
/
first
vmap_area_list->first->......->tmp->tmp_next->...->vmap_area_list
|-----(T3)----|
Under the scenario of free_vmap_cache miss, total time consumption of finding
the suitable hole is T = T1 + T2 + T3, while the commit decrease it to T1.
In fact, 'vmalloc' always start from the fix address(VMALLOC_START),which will
cause the 'first' to be close to the begining of the list(vmap_area_list) and
make T3 to be big.
The commit will especially help for a large and almost full vmalloc area.
Whearas, it would NOT affect current quick approach such as free_vmap_cache, for
it just take effect when free_vmap_cache miss and will reestablish it laterly.
Signed-off-by: Zhaoyang Huang <[email protected]>
---
mm/vmalloc.c | 11 +++++++++++
1 file changed, 11 insertions(+)
diff --git a/mm/vmalloc.c b/mm/vmalloc.c
index 8698c1c..f58f445 100644
--- a/mm/vmalloc.c
+++ b/mm/vmalloc.c
@@ -471,9 +471,20 @@ static struct vmap_area *alloc_vmap_area(unsigned long
size,
while (n) {
struct vmap_area *tmp;
+ struct vmap_area *tmp_next;
tmp = rb_entry(n, struct vmap_area, rb_node);
+ tmp_next = list_next_entry(tmp, list);
if (tmp->va_end >= addr) {
first = tmp;
+ if (ALIGN(tmp->va_end, align) + size
+ < tmp_next->va_start) {
+ /*
+ * free_vmap_cache miss now,don't
+ * update cached_hole_size here,
+ * as __free_vmap_area does
+ */
+ goto found;
+ }
if (tmp->va_start <= addr)
break;
n = n->rb_left;
--
1.9.1
