Skip to contentSkip to conversation

make migrate disable/enable conditioned on softirq_nestcnt transition

PATCH In review Nicholas Mc Guire 3 participants 3 messages Dec 5, 2013 – Dec 15, 2013

lore
Share & tools
lore
#0 lore raw

PATCH 1/3 make migrate disable/enable conditioned on softirq_nestcnt transition

This patch removes the recursive calls to migrate_disable/enable in local_bh_disable/enable

the softirq-local-lock.patch introduces local_bh_disable/enable wich decrements/increments the current->softirq_nestcnt and disable/enables migration as well. as softirq_nestcnt (include/linux/sched.h conditioned on CONFIG_PREEMPT_RT_BASE) already is tracking the nesting level of the recursive calls to local_bh_disable/enable (all in kernel/softirq.c) - no need to do it twice.

migrate_disable/enable thus can be conditionsed on softirq_nestcnt making a transition from 0-1 to disable migration and 1-0 to re-enable it.

patch on top of 3.12.1-rt4

No change of functional behavior, this does noticably reduce the observed nesting level of migrate_disable/enable

---
 kernel/softirq.c |   14 ++++++++------
 1 files changed, 8 insertions(+), 6 deletions(-)
kernel/softirq.c +8 -6
index 9a7268f..16ebbd9 100644
@@ -569,8 +569,8 @@ static void do_current_softirqs(int need_rcu_bh_qs)
 
 void local_bh_disable(void)
 {
-	migrate_disable();
-	current->softirq_nestcnt++;
+	if (++current->softirq_nestcnt == 1)
+		migrate_disable();
 }
 EXPORT_SYMBOL(local_bh_disable);
 
@@ -584,8 +584,8 @@ void local_bh_enable(void)
 		do_current_softirqs(1);
 	local_irq_enable();
 
-	current->softirq_nestcnt--;
-	migrate_enable();
+	if (--current->softirq_nestcnt == 0)
+		migrate_enable();
 }
 EXPORT_SYMBOL(local_bh_enable);
 
@@ -597,8 +597,10 @@ EXPORT_SYMBOL(local_bh_enable_ip);
 
 void _local_bh_enable(void)
 {
-	current->softirq_nestcnt--;
-	migrate_enable();
+	if (WARN_ON(current->softirq_nestcnt == 0))
+		return;
+	if (--current->softirq_nestcnt == 0)
+		migrate_enable();
 }
 EXPORT_SYMBOL(_local_bh_enable);
 
Signature · 1 line
1.7.2.5
Reviewed-by on v1 1/3
#1 lore raw

BTW, Please Cc LKML when sending patches. It is the kernel you are modifying, not userspace.

Quoted 40 lines · On Fri, 6 Dec 2013 00:42:22 +0100 Nicholas Mc Guire <[email protected]> wrote:

This patch removes the recursive calls to migrate_disable/enable in local_bh_disable/enable

the softirq-local-lock.patch introduces local_bh_disable/enable wich decrements/increments the current->softirq_nestcnt and disable/enables migration as well. as softirq_nestcnt (include/linux/sched.h conditioned on CONFIG_PREEMPT_RT_BASE) already is tracking the nesting level of the recursive calls to local_bh_disable/enable (all in kernel/softirq.c) - no need to do it twice.

migrate_disable/enable thus can be conditionsed on softirq_nestcnt making a transition from 0-1 to disable migration and 1-0 to re-enable it.

patch on top of 3.12.1-rt4

No change of functional behavior, this does noticably reduce the observed nesting level of migrate_disable/enable

---
 kernel/softirq.c |   14 ++++++++------
 1 files changed, 8 insertions(+), 6 deletions(-)
kernel/softirq.c +2 -2
index 9a7268f..16ebbd9 100644
@@ -569,8 +569,8 @@ static void do_current_softirqs(int need_rcu_bh_qs)
 
 void local_bh_disable(void)
 {
-	migrate_disable();
-	current->softirq_nestcnt++;
+	if (++current->softirq_nestcnt == 1)
+		migrate_disable();
 }
 EXPORT_SYMBOL(local_bh_disable);
 

kernel/softirq.c void local_bh_enable(void) · line 588

Quoted 7 lines
kernel/softirq.c +2 -2
@@ -584,8 +584,8 @@ void local_bh_enable(void)
 		do_current_softirqs(1);
 	local_irq_enable();
 
-	current->softirq_nestcnt--;
-	migrate_enable();
+	if (--current->softirq_nestcnt == 0)
+		migrate_enable();

I wonder if we should add a:

	BUG_ON(current->softirq_nestcnt < 0);

As for the patch, I haven't found anything wrong with it.

-- Steve

Quoted 16 lines
kernel/softirq.c +4 -2
 }
 EXPORT_SYMBOL(local_bh_enable);
 
@@ -597,8 +597,10 @@ EXPORT_SYMBOL(local_bh_enable_ip);
 
 void _local_bh_enable(void)
 {
-	current->softirq_nestcnt--;
-	migrate_enable();
+	if (WARN_ON(current->softirq_nestcnt == 0))
+		return;
+	if (--current->softirq_nestcnt == 0)
+		migrate_enable();
 }
 EXPORT_SYMBOL(_local_bh_enable);
 
#2 lore raw
Quoted 19 lines · * Steven Rostedt | 2013-12-05 19:45:30 [-0500]:
Quoted 17 lines · On Fri, 6 Dec 2013 00:42:22 +0100 Nicholas Mc Guire <[email protected]> wrote:
kernel/softirq.c +2 -2
@@ -569,8 +569,8 @@ static void do_current_softirqs(int need_rcu_bh_qs)
 
 void local_bh_disable(void)
 {
-	migrate_disable();
-	current->softirq_nestcnt++;
+	if (++current->softirq_nestcnt == 1)
+		migrate_disable();
 }
 EXPORT_SYMBOL(local_bh_disable);
 
@@ -584,8 +584,8 @@ void local_bh_enable(void)
 		do_current_softirqs(1);
 	local_irq_enable();
 
Quoted 8 lines

kernel/softirq.c void local_bh_enable(void) · line 588

Quoted 4 lines
kernel/softirq.c +2 -2
-	current->softirq_nestcnt--;
-	migrate_enable();
+	if (--current->softirq_nestcnt == 0)
+		migrate_enable();

I wonder if we should add a:

	BUG_ON(current->softirq_nestcnt < 0);

We have a WARN_ON() in each enable path. That one in local_bh_enable() isn't part of the context here. If you want to s/WARN_/BUG_/ then I would prefer not to since there are a few people with no UART and this could break the system while the current sollution keeps the system running.

Quoted 3 lines

As for the patch, I haven't found anything wrong with it.

Thanks.

Sebastian