Skip to contentSkip to conversation

Optimize synchronize_rcu_tasks_generic() on early booting

PATCH Applied Zqiang 2 participants 7 messages Jul 12, 2022 – Jul 14, 2022 RCU

lore
Share & tools
lore

Zqiang

author
#0 lore raw

cover Optimize synchronize_rcu_tasks_generic() on early booting

This commit fix warning trigger when call synchronize_rcu_tasks_generic() on early booting and make synchronize_rcu_tasks_generic() directly return when the rcu_scheduler_active variable is RCU_SCHEDULER_INACTIVE.

Zqiang (2):
rcu-tasks: Convert RCU_LOCKDEP_WARN() to WARN_ONCE()
rcu-tasks: Make synchronize_rcu_tasks_generic() no-ops on early
booting

 kernel/rcu/tasks.h | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
Signature · 1 line
2.25.1
author
#1 lore raw

PATCH 1/2 rcu-tasks: Convert RCU_LOCKDEP_WARN() to WARN_ONCE()

For kernel build with CONFIG_PROVE_RCU=y and CONFIG_DEBUG_LOCK_ALLOC=y, when the synchronize_rcu_tasks_generic() called occurs during the rcu_scheduler_active variable is RCU_SCHEDULER_INACTIVE, the warnings will be trigger, however the warnings is not actually triggered because the debug_lockdep_rcu_enabled() return false(the rcu_scheduler_active != RCU_SCHEDULER_INACTIVE condition is false).

This commit replace RCU_LOCKDEP_WARN() with WARN_ONCE() to trigger warnings.

---
 kernel/rcu/tasks.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
kernel/rcu/tasks.h +1 -1
index 83c7e6620d40..469bf2a3b505 100644
@@ -560,7 +560,7 @@ static int __noreturn rcu_tasks_kthread(void *arg)
 static void synchronize_rcu_tasks_generic(struct rcu_tasks *rtp)
 {
 	/* Complain if the scheduler has not started.  */
-	RCU_LOCKDEP_WARN(rcu_scheduler_active == RCU_SCHEDULER_INACTIVE,
+	WARN_ONCE(rcu_scheduler_active == RCU_SCHEDULER_INACTIVE,
 			 "synchronize_rcu_tasks called too soon");
 
 	// If the grace-period kthread is running, use it.
Signature · 1 line
2.25.1
maintainer Applied
#2 lore raw

Re: [PATCH 1/2] rcu-tasks: Convert RCU_LOCKDEP_WARN() to WARN_ONCE()

Quoted 11 lines · Zqiang

For kernel build with CONFIG_PROVE_RCU=y and CONFIG_DEBUG_LOCK_ALLOC=y, when the synchronize_rcu_tasks_generic() called occurs during the rcu_scheduler_active variable is RCU_SCHEDULER_INACTIVE, the warnings will be trigger, however the warnings is not actually triggered because the debug_lockdep_rcu_enabled() return false(the rcu_scheduler_active != RCU_SCHEDULER_INACTIVE condition is false).

This commit replace RCU_LOCKDEP_WARN() with WARN_ONCE() to trigger warnings.

Good catch, thank you!

Queued with wordsmithing as follows. As always, please check to see if I wordsmithed something out of existence.

Thanx, Paul

------------------------------------------------------------------------

commit 4dbfea53caf3e226e3273fc5947e9041e2ad34ab

Author: Zqiang <[email protected]>
Date:   Tue Jul 12 16:26:05 2022 +0800

rcu-tasks: Convert RCU_LOCKDEP_WARN() to WARN_ONCE()

Kernels built with CONFIG_PROVE_RCU=y and CONFIG_DEBUG_LOCK_ALLOC=y attempt to emit a warning when the synchronize_rcu_tasks_generic() function is called during early boot while the rcu_scheduler_active variable is RCU_SCHEDULER_INACTIVE. However the warnings is not actually be printed because the debug_lockdep_rcu_enabled() returns false, exactly because the rcu_scheduler_active variable is still equal to RCU_SCHEDULER_INACTIVE.

This commit therefore replaces RCU_LOCKDEP_WARN() with WARN_ONCE() to force these warnings to actually be printed.

kernel/rcu/tasks.h +1 -1
index 83c7e6620d403..469bf2a3b505e 100644
@@ -560,7 +560,7 @@ static int __noreturn rcu_tasks_kthread(void *arg)
 static void synchronize_rcu_tasks_generic(struct rcu_tasks *rtp)
 {
 	/* Complain if the scheduler has not started.  */
-	RCU_LOCKDEP_WARN(rcu_scheduler_active == RCU_SCHEDULER_INACTIVE,
+	WARN_ONCE(rcu_scheduler_active == RCU_SCHEDULER_INACTIVE,
 			 "synchronize_rcu_tasks called too soon");
 
 	// If the grace-period kthread is running, use it.
author
#3 lore raw

PATCH 2/2 rcu-tasks: Make synchronize_rcu_tasks_generic() no-ops on early booting

In reply to Zqiang

When the rcu_scheduler_active variable is RCU_SCHEDULER_INACTIVE and not yet converted to RCU_SCHEDULER_INIT, there is only idle task, any legal call synchronize_rcu_tasks_generic() is a quiescent state. this commit make synchronize_rcu_tasks_generic() no-ops when the rcu_scheduler_active variable is RCU_SCHEDULER_INACTIVE.

---
 kernel/rcu/tasks.h | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
kernel/rcu/tasks.h +3 -2
index 469bf2a3b505..0237e765c28e 100644
@@ -560,8 +560,9 @@ static int __noreturn rcu_tasks_kthread(void *arg)
 static void synchronize_rcu_tasks_generic(struct rcu_tasks *rtp)
 {
 	/* Complain if the scheduler has not started.  */
-	WARN_ONCE(rcu_scheduler_active == RCU_SCHEDULER_INACTIVE,
-			 "synchronize_rcu_tasks called too soon");
+	if (WARN_ONCE(rcu_scheduler_active == RCU_SCHEDULER_INACTIVE,
+			 "synchronize_rcu_tasks called too soon"))
+		return;
 
 	// If the grace-period kthread is running, use it.
 	if (READ_ONCE(rtp->kthread_ptr)) {
Signature · 1 line
2.25.1
maintainer
#4 lore raw

Re: [PATCH 2/2] rcu-tasks: Make synchronize_rcu_tasks_generic() no-ops on early booting

Quoted 8 lines · Zqiang

When the rcu_scheduler_active variable is RCU_SCHEDULER_INACTIVE and not yet converted to RCU_SCHEDULER_INIT, there is only idle task, any legal call synchronize_rcu_tasks_generic() is a quiescent state. this commit make synchronize_rcu_tasks_generic() no-ops when the rcu_scheduler_active variable is RCU_SCHEDULER_INACTIVE.

---

It looks like this would be a good way to provide early boot access to synchronize_rcu_tasks(), synchronize_rcu_tasks_rude(), and synchronize_rcu_tasks_trace().

But do we really need early boot access to these functions? As in has the below WARN_ON() actually triggered?

And if it has triggered, and in a context that means that these functions really are needed during early boot, how should the testing strategy change to test these at the relevant portions of the boot sequence?

From what I know, hitting these during early boot would indicate that something was removing a trace during early boot, and I know of no way to make that happen. Hence my skepticism. ;-)

But *if* this was really needed, this looks to be a reasonable way to implement it.

Thanx, Paul

Quoted 22 lines
 kernel/rcu/tasks.h | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
kernel/rcu/tasks.h +3 -2
index 469bf2a3b505..0237e765c28e 100644
@@ -560,8 +560,9 @@ static int __noreturn rcu_tasks_kthread(void *arg)
 static void synchronize_rcu_tasks_generic(struct rcu_tasks *rtp)
 {
 	/* Complain if the scheduler has not started.  */
-	WARN_ONCE(rcu_scheduler_active == RCU_SCHEDULER_INACTIVE,
-			 "synchronize_rcu_tasks called too soon");
+	if (WARN_ONCE(rcu_scheduler_active == RCU_SCHEDULER_INACTIVE,
+			 "synchronize_rcu_tasks called too soon"))
+		return;
 
 	// If the grace-period kthread is running, use it.
 	if (READ_ONCE(rtp->kthread_ptr)) {
Signature · 1 line
2.25.1
#5 lore raw

RE: [PATCH 2/2] rcu-tasks: Make synchronize_rcu_tasks_generic() no-ops on early booting

Quoted 7 lines · Zqiang

When the rcu_scheduler_active variable is RCU_SCHEDULER_INACTIVE and not yet converted to RCU_SCHEDULER_INIT, there is only idle task, any legal call synchronize_rcu_tasks_generic() is a quiescent state. this commit make synchronize_rcu_tasks_generic() no-ops when the rcu_scheduler_active variable is RCU_SCHEDULER_INACTIVE.

Quoted 8 lines

---

It looks like this would be a good way to provide early boot access to synchronize_rcu_tasks(), synchronize_rcu_tasks_rude(), and synchronize_rcu_tasks_trace().

But do we really need early boot access to these functions? As in has the below WARN_ON() actually triggered?

when the rcu_scheduler_active variable is RCU_SCHEDULER_INACTIVE, invoke synchronize_rcu_tasks_generic(), in addition to triggering a warning, also need to make it return directly, if not, the rcu_tasks_one_gp() will be called directly, but due to the rtp structure's -> pregp_func is not initialized, A null pointer bug will appear.

But like said, I don't see the need to call synchronize_rcu_tasks_generic() on early booting. maybe this change is not necessary.

Thanks
Zqiang

Quoted 36 lines

And if it has triggered, and in a context that means that these functions really are needed during early boot, how should the testing strategy change to test these at the relevant portions of the boot sequence?

From what I know, hitting these during early boot would indicate that something was removing a trace during early boot, and I know of no way to make that happen. Hence my skepticism. ;-)

But *if* this was really needed, this looks to be a reasonable way to implement it.

Thanx, Paul

 kernel/rcu/tasks.h | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
kernel/rcu/tasks.h +3 -2
index 469bf2a3b505..0237e765c28e 100644
@@ -560,8 +560,9 @@ static int __noreturn rcu_tasks_kthread(void *arg)
 static void synchronize_rcu_tasks_generic(struct rcu_tasks *rtp)
 {
 	/* Complain if the scheduler has not started.  */
-	WARN_ONCE(rcu_scheduler_active == RCU_SCHEDULER_INACTIVE,
-			 "synchronize_rcu_tasks called too soon");
+	if (WARN_ONCE(rcu_scheduler_active == RCU_SCHEDULER_INACTIVE,
+			 "synchronize_rcu_tasks called too soon"))
+		return;
 
 	// If the grace-period kthread is running, use it.
 	if (READ_ONCE(rtp->kthread_ptr)) {
Signature · 1 line
2.25.1
maintainer
#6 lore raw

Re: [PATCH 2/2] rcu-tasks: Make synchronize_rcu_tasks_generic() no-ops on early booting

Quoted 18 lines · On Thu, Jul 14, 2022 at 01:53:20AM +0000, Zhang, Qiang1 wrote:
Quoted 7 lines · Zqiang

When the rcu_scheduler_active variable is RCU_SCHEDULER_INACTIVE and not yet converted to RCU_SCHEDULER_INIT, there is only idle task, any legal call synchronize_rcu_tasks_generic() is a quiescent state. this commit make synchronize_rcu_tasks_generic() no-ops when the rcu_scheduler_active variable is RCU_SCHEDULER_INACTIVE.

Quoted 8 lines

---

It looks like this would be a good way to provide early boot access to synchronize_rcu_tasks(), synchronize_rcu_tasks_rude(), and synchronize_rcu_tasks_trace().

But do we really need early boot access to these functions? As in has the below WARN_ON() actually triggered?

Quoted 8 lines

when the rcu_scheduler_active variable is RCU_SCHEDULER_INACTIVE, invoke synchronize_rcu_tasks_generic(), in addition to triggering a warning, also need to make it return directly, if not, the rcu_tasks_one_gp() will be called directly, but due to the rtp structure's -> pregp_func is not initialized, A null pointer bug will appear.

But like said, I don't see the need to call synchronize_rcu_tasks_generic() on early booting. maybe this change is not necessary.

Not yet, anyway. And adding this would require more testing.

However, if the current warning does trigger, and the caller has a legitimate reason for invoking this function so early, please remember this patch. ;-)

Thanx, Paul

Quoted 39 lines

Thanks
Zqiang

Quoted 36 lines

And if it has triggered, and in a context that means that these functions really are needed during early boot, how should the testing strategy change to test these at the relevant portions of the boot sequence?

Quoted 1 line

From what I know, hitting these during early boot would indicate that

something was removing a trace during early boot, and I know of no way to make that happen. Hence my skepticism. ;-)

But *if* this was really needed, this looks to be a reasonable way to implement it.

Thanx, Paul

 kernel/rcu/tasks.h | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
kernel/rcu/tasks.h +3 -2
index 469bf2a3b505..0237e765c28e 100644
@@ -560,8 +560,9 @@ static int __noreturn rcu_tasks_kthread(void *arg)
 static void synchronize_rcu_tasks_generic(struct rcu_tasks *rtp)
 {
 	/* Complain if the scheduler has not started.  */
-	WARN_ONCE(rcu_scheduler_active == RCU_SCHEDULER_INACTIVE,
-			 "synchronize_rcu_tasks called too soon");
+	if (WARN_ONCE(rcu_scheduler_active == RCU_SCHEDULER_INACTIVE,
+			 "synchronize_rcu_tasks called too soon"))
+		return;
 
 	// If the grace-period kthread is running, use it.
 	if (READ_ONCE(rtp->kthread_ptr)) {
Signature · 1 line
2.25.1