fork: Only cache the VMAP stack in finish_task_switch()

The task stack could be deallocated later, but for fork()/exec() kind of
workloads (say a shell script executing several commands) it is important
that the stack is released in finish_task_switch() so that in VMAP_STACK
case it can be cached and reused in the new task.

For PREEMPT_RT it would be good if the wake-up in vfree_atomic() could
be avoided in the scheduling path. Far worse are the other
free_thread_stack() implementations which invoke __free_pages()/
kmem_cache_free() with disabled preemption.

Cache the stack in free_thread_stack() in the VMAP_STACK case and
RCU-delay the free path otherwise. Free the stack in the RCU callback.
In the VMAP_STACK case this is another opportunity to fill the cache.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Andy Lutomirski <luto@kernel.org>
Link: https://lore.kernel.org/r/20220217102406.3697941-8-bigeasy@linutronix.de

diff --git a/kernel/fork.c b/kernel/fork.c
index 2582812..177bc64 100644 (file)
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -193,6 +193,41 @@ static inline void free_task_struct(struct task_struct *tsk)
 #define NR_CACHED_STACKS 2
 static DEFINE_PER_CPU(struct vm_struct *, cached_stacks[NR_CACHED_STACKS]);
 
+struct vm_stack {
+       struct rcu_head rcu;
+       struct vm_struct *stack_vm_area;
+};
+
+static bool try_release_thread_stack_to_cache(struct vm_struct *vm)
+{
+       unsigned int i;
+
+       for (i = 0; i < NR_CACHED_STACKS; i++) {
+               if (this_cpu_cmpxchg(cached_stacks[i], NULL, vm) != NULL)
+                       continue;
+               return true;
+       }
+       return false;
+}
+
+static void thread_stack_free_rcu(struct rcu_head *rh)
+{
+       struct vm_stack *vm_stack = container_of(rh, struct vm_stack, rcu);
+
+       if (try_release_thread_stack_to_cache(vm_stack->stack_vm_area))
+               return;
+
+       vfree(vm_stack);
+}
+
+static void thread_stack_delayed_free(struct task_struct *tsk)
+{
+       struct vm_stack *vm_stack = tsk->stack;
+
+       vm_stack->stack_vm_area = tsk->stack_vm_area;
+       call_rcu(&vm_stack->rcu, thread_stack_free_rcu);
+}
+
 static int free_vm_stack_cache(unsigned int cpu)
 {
        struct vm_struct **cached_vm_stacks = per_cpu_ptr(cached_stacks, cpu);
@@ -296,24 +331,27 @@ static int alloc_thread_stack_node(struct task_struct *tsk, int node)
 
 static void free_thread_stack(struct task_struct *tsk)
 {
-       int i;
+       if (!try_release_thread_stack_to_cache(tsk->stack_vm_area))
+               thread_stack_delayed_free(tsk);
 
-       for (i = 0; i < NR_CACHED_STACKS; i++) {
-               if (this_cpu_cmpxchg(cached_stacks[i], NULL,
-                                    tsk->stack_vm_area) != NULL)
-                       continue;
-
-               tsk->stack = NULL;
-               tsk->stack_vm_area = NULL;
-               return;
-       }
-       vfree_atomic(tsk->stack);
        tsk->stack = NULL;
        tsk->stack_vm_area = NULL;
 }
 
 #  else /* !CONFIG_VMAP_STACK */
 
+static void thread_stack_free_rcu(struct rcu_head *rh)
+{
+       __free_pages(virt_to_page(rh), THREAD_SIZE_ORDER);
+}
+
+static void thread_stack_delayed_free(struct task_struct *tsk)
+{
+       struct rcu_head *rh = tsk->stack;
+
+       call_rcu(rh, thread_stack_free_rcu);
+}
+
 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
 {
        struct page *page = alloc_pages_node(node, THREADINFO_GFP,
@@ -328,7 +366,7 @@ static int alloc_thread_stack_node(struct task_struct *tsk, int node)
 
 static void free_thread_stack(struct task_struct *tsk)
 {
-       __free_pages(virt_to_page(tsk->stack), THREAD_SIZE_ORDER);
+       thread_stack_delayed_free(tsk);
        tsk->stack = NULL;
 }
 
@@ -337,6 +375,18 @@ static void free_thread_stack(struct task_struct *tsk)
 
 static struct kmem_cache *thread_stack_cache;
 
+static void thread_stack_free_rcu(struct rcu_head *rh)
+{
+       kmem_cache_free(thread_stack_cache, rh);
+}
+
+static void thread_stack_delayed_free(struct task_struct *tsk)
+{
+       struct rcu_head *rh = tsk->stack;
+
+       call_rcu(rh, thread_stack_free_rcu);
+}
+
 static int alloc_thread_stack_node(struct task_struct *tsk, int node)
 {
        unsigned long *stack;
@@ -348,7 +398,7 @@ static int alloc_thread_stack_node(struct task_struct *tsk, int node)
 
 static void free_thread_stack(struct task_struct *tsk)
 {
-       kmem_cache_free(thread_stack_cache, tsk->stack);
+       thread_stack_delayed_free(tsk);
        tsk->stack = NULL;
 }