Skip to contentSkip to conversation

workqueue: Use raise_softirq() in bh_pool_kick_.*()

PATCH In review Sebastian Andrzej Siewior 3 participants 6 messages Aug 19 – Aug 20 Workqueue

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 workqueue: Use raise_softirq() in bh_pool_kick_.*()

bh_pool_kick_normal() is used as irq_work callback to schedule softirq on a remote CPU.
On PREEMPT_RT the default irq_work item is initialized as IRQ_WORK_LAZY and is invoked in a thread with enabled interrupts (!RT would use softirq but interrupts would remain enabled). This triggers the warning in raise_softirq_irqoff() which expects interrupts to off while the softirq irq mask is modified.

The kick function triggers an IPI and the remote CPU wakes of irq_work/ and the callback wakes ksoftirqd/.

By initialising the irq_work as IRQ_WORK_INIT_HARD via
| *bh_pool_irq_work(pool) = IRQ_WORK_INIT_HARD(irq_work_fns[i]); we get the IPI and either (directly) a wake of ksoftirqd or "injected" the softirq into the current context if it is already in BH-disabled context which is in general undesired.
Even if ksoftirqd is woken up, the softirq work could be picked up by random task which decided to do softirqs and then "drains" the pending queue.

Scheduling a remote BH-work is undesired because of the possible context stealing. Using raise_softirq() here avoids the warning. As of today I did not see any users.

Use raise_softirq() in bh_pool_kick_.*() to avoid a warning in PREEMPT_RT.

---
 kernel/workqueue.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
kernel/workqueue.c +2 -2
index 78068ae8f28a6..0bb978df02005 100644
@@ -7923,12 +7923,12 @@ static inline void wq_watchdog_init(void) { }
 
 static void bh_pool_kick_normal(struct irq_work *irq_work)
 {
-	raise_softirq_irqoff(TASKLET_SOFTIRQ);
+	raise_softirq(TASKLET_SOFTIRQ);
 }
 
 static void bh_pool_kick_highpri(struct irq_work *irq_work)
 {
-	raise_softirq_irqoff(HI_SOFTIRQ);
+	raise_softirq(HI_SOFTIRQ);
 }
 
 static void __init restrict_unbound_cpumask(const char *name, const struct cpumask *mask)
Signature · 1 line
2.55.0
Reviewed-by on v1 1/1
#1 lore raw
Quoted 20 lines · Sebastian Andrzej Siewior

bh_pool_kick_normal() is used as irq_work callback to schedule softirq on a remote CPU.
On PREEMPT_RT the default irq_work item is initialized as IRQ_WORK_LAZY and is invoked in a thread with enabled interrupts (!RT would use softirq but interrupts would remain enabled). This triggers the warning in raise_softirq_irqoff() which expects interrupts to off while the softirq irq mask is modified.

The kick function triggers an IPI and the remote CPU wakes of irq_work/ and the callback wakes ksoftirqd/.

By initialising the irq_work as IRQ_WORK_INIT_HARD via
| *bh_pool_irq_work(pool) = IRQ_WORK_INIT_HARD(irq_work_fns[i]); we get the IPI and either (directly) a wake of ksoftirqd or "injected" the softirq into the current context if it is already in BH-disabled context which is in general undesired.
Even if ksoftirqd is woken up, the softirq work could be picked up by random task which decided to do softirqs and then "drains" the pending queue.

Quoted 7 lines

Scheduling a remote BH-work is undesired because of the possible context stealing. Using raise_softirq() here avoids the warning. As of today I did not see any users.

Use raise_softirq() in bh_pool_kick_.*() to avoid a warning in PREEMPT_RT.

Well, sure, me personally don't see any issues in this code, makes life easier

Quoted 23 lines

---
kernel/workqueue.c | 4 ++--

1 file changed, 2 insertions(+), 2 deletions(-)
kernel/workqueue.c +2 -2
index 78068ae8f28a6..0bb978df02005 100644
@@ -7923,12 +7923,12 @@ static inline void wq_watchdog_init(void) { }
 
 static void bh_pool_kick_normal(struct irq_work *irq_work)
 {
-	raise_softirq_irqoff(TASKLET_SOFTIRQ);
+	raise_softirq(TASKLET_SOFTIRQ);
 }
 
 static void bh_pool_kick_highpri(struct irq_work *irq_work)
 {
-	raise_softirq_irqoff(HI_SOFTIRQ);
+	raise_softirq(HI_SOFTIRQ);
 }
 
Quoted 3 lines

static void __init restrict_unbound_cpumask(const char *name, const struct cpumask *mask)

Thanks!

maintainer
#2 lore raw
Quoted 4 lines · Bradley Morgan

Well, sure, me personally don't see any issues in this code, makes life easier

I've been getting these review responses without much content on patches across multiple subsystems. I'm leaning towards ignoring going forward. If I'm jumping the gun prematurely, please let me know.

Signature · 1 line
tejun
#3 lore raw
Quoted 10 lines · Tejun Heo
Quoted 4 lines · Bradley Morgan

Well, sure, me personally don't see any issues in this code, makes life easier

I've been getting these review responses without much content on patches across multiple subsystems. I'm leaning towards ignoring going forward. If I'm jumping the gun prematurely, please let me know.

Hmm, hold on before u "jump the gun", I learn based on other maintainers reviews, some just do a LGTM, and then the tag, which I see as appropriate, if you don't agree with that, you could have told me..

Whatever you want me to do, I'll happily do it, wether it's more context in the review, or doing other things, please please please tell me,

I am a learning reviewer, and I am interested in kernel/ (in some cases, subdirectories), I am sorry if this looks suspicious or odd to you, just please tell me what you are finding flawed about my reviews, or just wrong,so that I could iprove

Quoted 1 line

Thanks!

maintainer
#4 lore raw
In reply to Sebastian Andrzej Siewior

Hello, Sebastian.

Quoted 2 lines · Sebastian Andrzej Siewior

Use raise_softirq() in bh_pool_kick_.*() to avoid a warning in PREEMPT_RT.

The same fix was applied last month:

df42c49ec865 ("workqueue: Use raise_softirq() to trigger softirq in irq_work handler")

but I mistakenly applied it to the already closed for-7.1-fixes branch and it never reached mainline. It's now cherry-picked to wq/for-7.3-fixes and will go to mainline and then -stable this cycle. Thanks for the fix and sorry about the confusion.

Signature · 1 line
tejun
#5 lore raw
Quoted 1 line · Tejun Heo

Hello, Sebastian.

Hi,

Quoted 6 lines · Sebastian Andrzej Siewior
Quoted 2 lines

Use raise_softirq() in bh_pool_kick_.*() to avoid a warning in PREEMPT_RT.

The same fix was applied last month:

Quoted 7 lines

df42c49ec865 ("workqueue: Use raise_softirq() to trigger softirq in irq_work handler")

but I mistakenly applied it to the already closed for-7.1-fixes branch and it never reached mainline. It's now cherry-picked to wq/for-7.3-fixes and will go to mainline and then -stable this cycle. Thanks for the fix and sorry about the confusion.

no worries.

Sebastian