Skip to contentSkip to conversation

preempt: add in_serving_irq() and apply to rcutiny and vsprintf

PATCH In review Changbin Du 2 participants 5 messages Aug 14, 2021 – Aug 19, 2021

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 preempt: add in_serving_irq() and apply to rcutiny and vsprintf

At some places we need to determine whether we're in nmi, hardirq or softirq context. This adds a macro in_serving_irq() as a shortcut for that.

Meanwhile, apply this new macro to existing code in rcutiny and vsprintf.

---
 include/linux/preempt.h | 4 +++-
 include/linux/rcutiny.h | 3 +--
 lib/vsprintf.c          | 2 +-
 3 files changed, 5 insertions(+), 4 deletions(-)
include/linux/preempt.h +3 -1
index 9881eac0698f..9a1c924e2c6c 100644
@@ -92,12 +92,14 @@
  * in_nmi()		- We're in NMI context
  * in_hardirq()		- We're in hard IRQ context
  * in_serving_softirq()	- We're in softirq context
+ * in_serving_irq()	- We're in nmi, hardirq or softirq context
  * in_task()		- We're in task context
  */
 #define in_nmi()		(nmi_count())
 #define in_hardirq()		(hardirq_count())
 #define in_serving_softirq()	(softirq_count() & SOFTIRQ_OFFSET)
-#define in_task()		(!(in_nmi() | in_hardirq() | in_serving_softirq()))
+#define in_serving_irq()	(in_nmi() | in_hardirq() | in_serving_softirq())
+#define in_task()		(!in_serving_irq())
 
 /*
  * The following macros are deprecated and should not be used in new code:
include/linux/rcutiny.h +1 -2
index 7fedbd33d5d2..812d42f22e9c 100644
@@ -87,8 +87,7 @@ static inline void rcu_irq_exit_irqson(void) { }
 static inline void rcu_irq_enter_irqson(void) { }
 static inline void rcu_irq_exit(void) { }
 static inline void rcu_irq_exit_check_preempt(void) { }
-#define rcu_is_idle_cpu(cpu) \
-	(is_idle_task(current) && !in_nmi() && !in_hardirq() && !in_serving_softirq())
+#define rcu_is_idle_cpu(cpu) (is_idle_task(current) && !in_serving_irq())
 static inline void exit_rcu(void) { }
 static inline bool rcu_preempt_need_deferred_qs(struct task_struct *t)
 {
lib/vsprintf.c +1 -1
index 2c5b4351330c..9324439c8543 100644
@@ -865,7 +865,7 @@ char *restricted_pointer(char *buf, char *end, const void *ptr,
 		 * kptr_restrict==1 cannot be used in IRQ context
 		 * because its test for CAP_SYSLOG would be meaningless.
 		 */
-		if (in_hardirq() || in_serving_softirq() || in_nmi()) {
+		if (in_serving_irq()) {
 			if (spec.field_width == -1)
 				spec.field_width = 2 * sizeof(ptr);
 			return error_string(buf, end, "pK-error", spec);
Signature · 1 line
2.30.2
#1 lore raw
Quoted 21 lines · Changbin Du

At some places we need to determine whether we're in nmi, hardirq or softirq context. This adds a macro in_serving_irq() as a shortcut for that.

Meanwhile, apply this new macro to existing code in rcutiny and vsprintf.

---
 include/linux/preempt.h | 4 +++-
 include/linux/rcutiny.h | 3 +--
 lib/vsprintf.c          | 2 +-
 3 files changed, 5 insertions(+), 4 deletions(-)
include/linux/preempt.h +0 -0
index 9881eac0698f..9a1c924e2c6c 100644
@@ -92,12 +92,14 @@
  * in_nmi()		- We're in NMI context
  * in_hardirq()		- We're in hard IRQ context
  * in_serving_softirq()	- We're in softirq context

include/linux/preempt.h line 103

Quoted 10 lines
include/linux/preempt.h +3 -1
+ * in_serving_irq()	- We're in nmi, hardirq or softirq context
  * in_task()		- We're in task context
  */
 #define in_nmi()		(nmi_count())
 #define in_hardirq()		(hardirq_count())
 #define in_serving_softirq()	(softirq_count() & SOFTIRQ_OFFSET)
-#define in_task()		(!(in_nmi() | in_hardirq() | in_serving_softirq()))
+#define in_serving_irq()	(in_nmi() | in_hardirq() | in_serving_softirq())
+#define in_task()		(!in_serving_irq())
 

So in_serving_irq() is !in_task(), right? If so, why not...

include/linux/rcutiny.h static inline void rcu_irq_exit_irqson(void) { } · line 90

Quoted 13 lines
include/linux/preempt.h +0 -0
 /*
  * The following macros are deprecated and should not be used in new code:
include/linux/rcutiny.h +1 -2
index 7fedbd33d5d2..812d42f22e9c 100644
@@ -87,8 +87,7 @@ static inline void rcu_irq_exit_irqson(void) { }
 static inline void rcu_irq_enter_irqson(void) { }
 static inline void rcu_irq_exit(void) { }
 static inline void rcu_irq_exit_check_preempt(void) { }
-#define rcu_is_idle_cpu(cpu) \
-	(is_idle_task(current) && !in_nmi() && !in_hardirq() && !in_serving_softirq())
+#define rcu_is_idle_cpu(cpu) (is_idle_task(current) && !in_serving_irq())

... use in_task() here, and ...

lib/vsprintf.c char *restricted_pointer(char *buf, char *end, const void *ptr, · line 868

Quoted 13 lines
include/linux/rcutiny.h +0 -0
 static inline void exit_rcu(void) { }
 static inline bool rcu_preempt_need_deferred_qs(struct task_struct *t)
 {
lib/vsprintf.c +1 -1
index 2c5b4351330c..9324439c8543 100644
@@ -865,7 +865,7 @@ char *restricted_pointer(char *buf, char *end, const void *ptr,
 		 * kptr_restrict==1 cannot be used in IRQ context
 		 * because its test for CAP_SYSLOG would be meaningless.
 		 */
-		if (in_hardirq() || in_serving_softirq() || in_nmi()) {
+		if (in_serving_irq()) {

... use !in_task() here?

And don't introduce the in_serving_irq() at all.

Regards,
Boqun

Quoted 6 lines
lib/vsprintf.c +0 -0
 			if (spec.field_width == -1)
 				spec.field_width = 2 * sizeof(ptr);
 			return error_string(buf, end, "pK-error", spec);
Signature · 1 line
2.30.2
#2 lore raw
Quoted 26 lines · Boqun Feng
Quoted 25 lines · Changbin Du

At some places we need to determine whether we're in nmi, hardirq or softirq context. This adds a macro in_serving_irq() as a shortcut for that.

Meanwhile, apply this new macro to existing code in rcutiny and vsprintf.

---
 include/linux/preempt.h | 4 +++-
 include/linux/rcutiny.h | 3 +--
 lib/vsprintf.c          | 2 +-
 3 files changed, 5 insertions(+), 4 deletions(-)
include/linux/preempt.h +1 -0
index 9881eac0698f..9a1c924e2c6c 100644
@@ -92,12 +92,14 @@
  * in_nmi()		- We're in NMI context
  * in_hardirq()		- We're in hard IRQ context
  * in_serving_softirq()	- We're in softirq context
+ * in_serving_irq()	- We're in nmi, hardirq or softirq context
  * in_task()		- We're in task context
  */
 #define in_nmi()		(nmi_count())
Quoted 9 lines

include/linux/preempt.h line 103

Quoted 6 lines
include/linux/preempt.h +2 -1
 #define in_hardirq()		(hardirq_count())
 #define in_serving_softirq()	(softirq_count() & SOFTIRQ_OFFSET)
-#define in_task()		(!(in_nmi() | in_hardirq() | in_serving_softirq()))
+#define in_serving_irq()	(in_nmi() | in_hardirq() | in_serving_softirq())
+#define in_task()		(!in_serving_irq())
 

So in_serving_irq() is !in_task(), right? If so, why not...

Adding in_serving_irq() is to reflect the real purpose so improve readability. And can we preserve that !in_task() means in serving irq context in future? I don't know.

Signature · 2 lines
Cheers,
Changbin Du
#3 lore raw

[Cc Thomas and Frederic since they contributed the clean-up to these macros recently]

Background for discussion:

	https://lore.kernel.org/lkml/[email protected]/
Quoted 36 lines · Changbin Du
Quoted 26 lines · Boqun Feng
Quoted 25 lines · Changbin Du

At some places we need to determine whether we're in nmi, hardirq or softirq context. This adds a macro in_serving_irq() as a shortcut for that.

Meanwhile, apply this new macro to existing code in rcutiny and vsprintf.

---
 include/linux/preempt.h | 4 +++-
 include/linux/rcutiny.h | 3 +--
 lib/vsprintf.c          | 2 +-
 3 files changed, 5 insertions(+), 4 deletions(-)
include/linux/preempt.h +1 -0
index 9881eac0698f..9a1c924e2c6c 100644
@@ -92,12 +92,14 @@
  * in_nmi()		- We're in NMI context
  * in_hardirq()		- We're in hard IRQ context
  * in_serving_softirq()	- We're in softirq context
+ * in_serving_irq()	- We're in nmi, hardirq or softirq context
  * in_task()		- We're in task context
  */
 #define in_nmi()		(nmi_count())
Quoted 9 lines

include/linux/preempt.h line 103

Quoted 6 lines
include/linux/preempt.h +2 -1
 #define in_hardirq()		(hardirq_count())
 #define in_serving_softirq()	(softirq_count() & SOFTIRQ_OFFSET)
-#define in_task()		(!(in_nmi() | in_hardirq() | in_serving_softirq()))
+#define in_serving_irq()	(in_nmi() | in_hardirq() | in_serving_softirq())
+#define in_task()		(!in_serving_irq())
 

So in_serving_irq() is !in_task(), right? If so, why not...

Quoted 3 lines

Adding in_serving_irq() is to reflect the real purpose so improve readability. And can we preserve that !in_task() means in serving irq context in future? I don't know.

Sure, no one could predict the future. But if a third context (other than thread context and {hard,soft}irq context) comes up, which I think is highly unlikely, we could (and should) audit all callsites of in_task() for necessary adjustment. And introducing in_serving_irq() won't help us in that case, because we will still need to audit usage of in_serving_irq(), for example, let's say rcu_is_idle_cpu() for RCU_TINY is defined as

	#define rcu_is_idle_cpu(cpu) (is_idle_task(current) && !in_serving_irq())

and we have a new type of context, and we can use in_other() to test whether we are in it. Now even with in_serving_irq() introduced, we still need to make sure the correct version of rcu_is_idle_cpu() is either

	(is_idle_task(current) && (!in_serving_irq() && !in_other()))

or

	(is_idle_task(current) && !in_serving_irq())

Therefore, I don't see the point of introducing in_serving_irq().

Regards,
Boqun

Quoted 3 lines
Signature · 2 lines
Cheers,
Changbin Du
#4 lore raw
Quoted 63 lines · Boqun Feng

[Cc Thomas and Frederic since they contributed the clean-up to these macros recently]

Background for discussion:

	https://lore.kernel.org/lkml/[email protected]/
Quoted 36 lines · Changbin Du
Quoted 26 lines · Boqun Feng
Quoted 25 lines · Changbin Du

At some places we need to determine whether we're in nmi, hardirq or softirq context. This adds a macro in_serving_irq() as a shortcut for that.

Meanwhile, apply this new macro to existing code in rcutiny and vsprintf.

---
 include/linux/preempt.h | 4 +++-
 include/linux/rcutiny.h | 3 +--
 lib/vsprintf.c          | 2 +-
 3 files changed, 5 insertions(+), 4 deletions(-)
include/linux/preempt.h +1 -0
index 9881eac0698f..9a1c924e2c6c 100644
@@ -92,12 +92,14 @@
  * in_nmi()		- We're in NMI context
  * in_hardirq()		- We're in hard IRQ context
  * in_serving_softirq()	- We're in softirq context
+ * in_serving_irq()	- We're in nmi, hardirq or softirq context
  * in_task()		- We're in task context
  */
 #define in_nmi()		(nmi_count())
Quoted 9 lines

include/linux/preempt.h line 103

Quoted 6 lines
include/linux/preempt.h +2 -1
 #define in_hardirq()		(hardirq_count())
 #define in_serving_softirq()	(softirq_count() & SOFTIRQ_OFFSET)
-#define in_task()		(!(in_nmi() | in_hardirq() | in_serving_softirq()))
+#define in_serving_irq()	(in_nmi() | in_hardirq() | in_serving_softirq())
+#define in_task()		(!in_serving_irq())
 

So in_serving_irq() is !in_task(), right? If so, why not...

Quoted 3 lines

Adding in_serving_irq() is to reflect the real purpose so improve readability. And can we preserve that !in_task() means in serving irq context in future? I don't know.

Sure, no one could predict the future. But if a third context (other than thread context and {hard,soft}irq context) comes up, which I think is highly unlikely, we could (and should) audit all callsites of in_task() for necessary adjustment. And introducing in_serving_irq() won't help us in that case, because we will still need to audit usage of in_serving_irq(), for example, let's say rcu_is_idle_cpu() for RCU_TINY is defined as

	#define rcu_is_idle_cpu(cpu) (is_idle_task(current) && !in_serving_irq())

and we have a new type of context, and we can use in_other() to test whether we are in it. Now even with in_serving_irq() introduced, we still need to make sure the correct version of rcu_is_idle_cpu() is either

Quoted 8 lines
	(is_idle_task(current) && (!in_serving_irq() && !in_other()))

or

	(is_idle_task(current) && !in_serving_irq())

Therefore, I don't see the point of introducing in_serving_irq().

ok, as in_serving_irq() is only used in two places, it is not common to judge if it is in serving irq context. So this new macro doesn't help much.

Quoted 6 lines

Regards,
Boqun

Quoted 3 lines
Signature · 2 lines
Cheers,
Changbin Du
Signature · 2 lines
Cheers,
Changbin Du