It seems that alloc_retstack_tasklist() can also take a lockless
approach for scanning the tasklist, instead of using the big global
tasklist_lock. For this we also kill another deprecated and rcu-unsafe
tsk->thread_group user replacing it with for_each_process_thread(),
maintaining semantics.
Here tasklist_lock does not protect anything other than the list
against concurrent fork/exit. And considering that the whole thing
is capped by FTRACE_RETSTACK_ALLOC_SIZE (32), it should not be a
problem to have a pontentially stale, yet stable, list. The task cannot
go away either, so we don't risk racing with ftrace_graph_exit_task()
which clears the retstack.
The tsk->ret_stack management is not protected by tasklist_lock, being
serialized with the corresponding publish/subscribe barriers against
concurrent ftrace_push_return_trace(). In addition this plays nicer
with cachelines by avoiding two atomic ops in the uncontended case.
Signed-off-by: Davidlohr Bueso <dbu...@suse.de>
---
kernel/trace/fgraph.c | 8 ++++----
1 file changed, 4 insertions(+), 4 deletions(-)
diff --git a/kernel/trace/fgraph.c b/kernel/trace/fgraph.c
index 1af321dec0f1..5658f13037b3 100644
--- a/kernel/trace/fgraph.c
+++ b/kernel/trace/fgraph.c
@@ -387,8 +387,8 @@ static int alloc_retstack_tasklist(struct ftrace_ret_stack
**ret_stack_list)
}
}
- read_lock(&tasklist_lock);
- do_each_thread(g, t) {
+ rcu_read_lock();
+ for_each_process_thread(g, t) {
if (start == end) {
ret = -EAGAIN;
goto unlock;
@@ -403,10 +403,10 @@ static int alloc_retstack_tasklist(struct
ftrace_ret_stack **ret_stack_list)
smp_wmb();
t->ret_stack = ret_stack_list[start++];
}
- } while_each_thread(g, t);
+ }
unlock:
- read_unlock(&tasklist_lock);
+ rcu_read_unlock();
free:
for (i = start; i < end; i++)
kfree(ret_stack_list[i]);
--
2.26.2
