2021-06-02 11:46:04

by Yejune Deng

Subject: [PATCH] softirq: Remove the unnecessary CONFIG_TRACE_IRQFLAGS

The local_irq_enable() and local_irq_disable() had two definitions that
include CONFIG_TRACE_IRQFLAGS or not.

Signed-off-by: Yejune Deng <[email protected]>
---
kernel/softirq.c | 4 ----
1 file changed, 4 deletions(-)

diff --git a/kernel/softirq.c b/kernel/softirq.c
index 4992853..a81d804 100644
--- a/kernel/softirq.c
+++ b/kernel/softirq.c
@@ -362,9 +362,7 @@ void __local_bh_enable_ip(unsigned long ip, unsigned int cnt)
{
WARN_ON_ONCE(in_irq());
lockdep_assert_irqs_enabled();
-#ifdef CONFIG_TRACE_IRQFLAGS
local_irq_disable();
-#endif
/*
* Are softirqs going to be turned on now:
*/
@@ -385,9 +383,7 @@ void __local_bh_enable_ip(unsigned long ip, unsigned int cnt)
}

preempt_count_dec();
-#ifdef CONFIG_TRACE_IRQFLAGS
local_irq_enable();
-#endif
preempt_check_resched();
}
EXPORT_SYMBOL(__local_bh_enable_ip);
--
2.7.4

2021-06-02 17:56:47

by Kees Cook

Subject: Re: [PATCH] softirq: Remove the unnecessary CONFIG_TRACE_IRQFLAGS

On Wed, Jun 02, 2021 at 03:34:42PM +0800, Yejune Deng wrote:
> The local_irq_enable() and local_irq_disable() had two definitions that
> include CONFIG_TRACE_IRQFLAGS or not.

Hm, this has a behavioral change. Current code only does tracing and
raw_local_irq_disable/enable when CONFIG_TRACE_IRQFLAGS. This change
causes it to always do raw_local_irq_disable/enable. Which is the right
behavior?

-Kees

>
> Signed-off-by: Yejune Deng <[email protected]>
> ---
> kernel/softirq.c | 4 ----
> 1 file changed, 4 deletions(-)
>
> diff --git a/kernel/softirq.c b/kernel/softirq.c
> index 4992853..a81d804 100644
> --- a/kernel/softirq.c
> +++ b/kernel/softirq.c
> @@ -362,9 +362,7 @@ void __local_bh_enable_ip(unsigned long ip, unsigned int cnt)
> {
> WARN_ON_ONCE(in_irq());
> lockdep_assert_irqs_enabled();
> -#ifdef CONFIG_TRACE_IRQFLAGS
> local_irq_disable();
> -#endif
> /*
> * Are softirqs going to be turned on now:
> */
> @@ -385,9 +383,7 @@ void __local_bh_enable_ip(unsigned long ip, unsigned int cnt)
> }
>
> preempt_count_dec();
> -#ifdef CONFIG_TRACE_IRQFLAGS
> local_irq_enable();
> -#endif
> preempt_check_resched();
> }
> EXPORT_SYMBOL(__local_bh_enable_ip);
> --
> 2.7.4
>

--
Kees Cook

2021-08-10 16:37:48

by Thomas Gleixner

Subject: Re: [PATCH] softirq: Remove the unnecessary CONFIG_TRACE_IRQFLAGS

On Wed, Jun 02 2021 at 10:53, Kees Cook wrote:
> On Wed, Jun 02, 2021 at 03:34:42PM +0800, Yejune Deng wrote:
>> The local_irq_enable() and local_irq_disable() had two definitions that
>> include CONFIG_TRACE_IRQFLAGS or not.
>
> Hm, this has a behavioral change. Current code only does tracing and
> raw_local_irq_disable/enable when CONFIG_TRACE_IRQFLAGS. This change
> causes it to always do raw_local_irq_disable/enable. Which is the right
> behavior?

Disabling interrupts is only required when CONFIG_TRACE_IRQFLAGS=y. So
the code is correct as is.

Thanks,

tglx