[RFC][PATCH v2 10/11] livepatch: Remove klp_synchronize_transition()

From: Peter Zijlstra
Date: Wed Sep 29 2021 - 12:08:51 EST


The whole premise is broken, any trace usage outside of RCU is a BUG
and should be fixed. Notable all code prior (and a fair bit after)
ct_user_exit() is noinstr and explicitly doesn't allow any tracing.

Use regular RCU, which has the benefit of actually respecting
NOHZ_FULL.

Signed-off-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
---
kernel/livepatch/transition.c | 32 +++-----------------------------
1 file changed, 3 insertions(+), 29 deletions(-)

--- a/kernel/livepatch/transition.c
+++ b/kernel/livepatch/transition.c
@@ -42,28 +42,6 @@ static void klp_transition_work_fn(struc
static DECLARE_DELAYED_WORK(klp_transition_work, klp_transition_work_fn);

/*
- * This function is just a stub to implement a hard force
- * of synchronize_rcu(). This requires synchronizing
- * tasks even in userspace and idle.
- */
-static void klp_sync(struct work_struct *work)
-{
-}
-
-/*
- * We allow to patch also functions where RCU is not watching,
- * e.g. before ct_user_exit(). We can not rely on the RCU infrastructure
- * to do the synchronization. Instead hard force the sched synchronization.
- *
- * This approach allows to use RCU functions for manipulating func_stack
- * safely.
- */
-static void klp_synchronize_transition(void)
-{
- schedule_on_each_cpu(klp_sync);
-}
-
-/*
* The transition to the target patch state is complete. Clean up the data
* structures.
*/
@@ -96,7 +74,7 @@ static void klp_complete_transition(void
* func->transition gets cleared, the handler may choose a
* removed function.
*/
- klp_synchronize_transition();
+ synchronize_rcu();
}

klp_for_each_object(klp_transition_patch, obj)
@@ -105,7 +83,7 @@ static void klp_complete_transition(void

/* Prevent klp_ftrace_handler() from seeing KLP_UNDEFINED state */
if (klp_target_state == KLP_PATCHED)
- klp_synchronize_transition();
+ synchronize_rcu();

read_lock(&tasklist_lock);
for_each_process_thread(g, task) {
@@ -168,10 +146,6 @@ noinstr void __klp_update_patch_state(st
*/
void klp_update_patch_state(struct task_struct *task)
{
- /*
- * A variant of synchronize_rcu() is used to allow patching functions
- * where RCU is not watching, see klp_synchronize_transition().
- */
preempt_disable_notrace();

/*
@@ -626,7 +600,7 @@ void klp_reverse_transition(void)
clear_tsk_thread_flag(idle_task(cpu), TIF_PATCH_PENDING);

/* Let any remaining calls to klp_update_patch_state() complete */
- klp_synchronize_transition();
+ synchronize_rcu();

klp_start_transition();
}