@@ -159,10 +159,37 @@ void __weak arch_release_thread_stack(unsigned long *stack)
* kmemcache based allocator.
*/
# if THREAD_SIZE >= PAGE_SIZE || defined(CONFIG_VMAP_STACK)
+
+#ifdef CONFIG_VMAP_STACK
+/*
+ * vmalloc is a bit slow, and calling vfree enough times will force a TLB
+ * flush. Try to minimize the number of calls by caching stacks.
+ */
+#define NR_CACHED_STACKS 2
+static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
+#endif
+
static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
{
#ifdef CONFIG_VMAP_STACK
- void *stack = __vmalloc_node_range(
+ void *stack;
+ int i;
+
+ local_irq_disable();
+ for (i = 0; i < NR_CACHED_STACKS; i++) {
+ struct vm_struct *s = this_cpu_read(cached_stacks[i]);
+
+ if (!s)
+ continue;
+ this_cpu_write(cached_stacks[i], NULL);
+
+ tsk->stack_vm_area = s;
+ local_irq_enable();
+ return s->addr;
+ }
+ local_irq_enable();
+
+ stack = __vmalloc_node_range(
THREAD_SIZE, THREAD_SIZE, VMALLOC_START, VMALLOC_END,
THREADINFO_GFP | __GFP_HIGHMEM, PAGE_KERNEL,
0, node, __builtin_return_address(0));
@@ -185,10 +212,29 @@ static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
static inline void free_thread_stack(struct task_struct *tsk)
{
- if (task_stack_vm_area(tsk))
+#ifdef CONFIG_VMAP_STACK
+ if (task_stack_vm_area(tsk)) {
+ unsigned long flags;
+ int i;
+
+ local_irq_save(flags);
+ for (i = 0; i < NR_CACHED_STACKS; i++) {
+ if (this_cpu_read(cached_stacks[i]))
+ continue;
+
+ this_cpu_write(cached_stacks[i], tsk->stack_vm_area);
+ goto done;
+ }
+
vfree(tsk->stack);
- else
- free_kmem_pages((unsigned long)tsk->stack, THREAD_SIZE_ORDER);
+
+done:
+ local_irq_restore(flags);
+ return;
+ }
+#endif
+
+ free_kmem_pages((unsigned long)tsk->stack, THREAD_SIZE_ORDER);
}
# else
static struct kmem_cache *thread_stack_cache;
vmalloc is a bit slow, and pounding vmalloc/vfree will eventually force a global TLB flush. To reduce pressure on them, if CONFIG_VMAP_STACK, cache two thread stacks per cpu. This will let us quickly allocate a hopefully cache-hot, TLB-hot stack under heavy forking workloads (shell script style). On my silly pthread_create benchmark, it saves about 2 µs per pthread_create+join with CONFIG_VMAP_STACK=y. Signed-off-by: Andy Lutomirski <luto@kernel.org> --- kernel/fork.c | 54 ++++++++++++++++++++++++++++++++++++++++++++++++++---- 1 file changed, 50 insertions(+), 4 deletions(-)