[PATCH] kernel/softirq.c: delete 'while' looping to improve a little performance and beautify code

[PATCH] kernel/softirq.c: delete 'while' looping to improve a little performance and beautify code

[Chen Gang]

After finish the internal 'while', need not test TASKLET_STATE_SCHED
again, so looping back to outside 'while' is only for set_bit().

When use 'if' and set_bit() instead of 'while', it will save at least
one running conditional instruction, and also will be clearer for readers
(although the binary size will be a little bigger).

The related patch is "1da177e Linux-2.6.12-rc2"


Signed-off-by: Chen Gang <gang.chen@asianux.com>
---
 kernel/softirq.c |    3 ++-
 1 files changed, 2 insertions(+), 1 deletions(-)

diff --git a/kernel/softirq.c b/kernel/softirq.c
index a5f8836..52da25f 100644
--- a/kernel/softirq.c
+++ b/kernel/softirq.c
@@ -540,10 +540,11 @@ void tasklet_kill(struct tasklet_struct *t)
 	if (in_interrupt())
 		printk("Attempt to kill tasklet from interrupt\n");
 
-	while (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) {
+	if (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) {
 		do {
 			yield();
 		} while (test_bit(TASKLET_STATE_SCHED, &t->state));
+		set_bit(TASKLET_STATE_SCHED, &t->state);
 	}
 	tasklet_unlock_wait(t);
 	clear_bit(TASKLET_STATE_SCHED, &t->state);
-- 
1.7.7.6

Re: [PATCH] kernel/softirq.c: delete 'while' looping to improve a little performance and beautify code

[Paul E. McKenney]

On Sun, Jun 09, 2013 at 08:30:19PM +0800, Chen Gang wrote:
> 
> After finish the internal 'while', need not test TASKLET_STATE_SCHED
> again, so looping back to outside 'while' is only for set_bit().
> 
> When use 'if' and set_bit() instead of 'while', it will save at least
> one running conditional instruction, and also will be clearer for readers
> (although the binary size will be a little bigger).
> 
> The related patch is "1da177e Linux-2.6.12-rc2"
> 
> 
> Signed-off-by: Chen Gang <gang.chen@asianux.com>
> ---
>  kernel/softirq.c |    3 ++-
>  1 files changed, 2 insertions(+), 1 deletions(-)
> 
> diff --git a/kernel/softirq.c b/kernel/softirq.c
> index a5f8836..52da25f 100644
> --- a/kernel/softirq.c
> +++ b/kernel/softirq.c
> @@ -540,10 +540,11 @@ void tasklet_kill(struct tasklet_struct *t)
>  	if (in_interrupt())
>  		printk("Attempt to kill tasklet from interrupt\n");
> 
> -	while (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) {
> +	if (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) {
>  		do {
>  			yield();
>  		} while (test_bit(TASKLET_STATE_SCHED, &t->state));
> +		set_bit(TASKLET_STATE_SCHED, &t->state);

This replaces an atomic test-and-set with two operations, a test and
a set.  Is this safe?

							Thanx, Paul

>  	}
>  	tasklet_unlock_wait(t);
>  	clear_bit(TASKLET_STATE_SCHED, &t->state);
> -- 
> 1.7.7.6
> 

Re: [PATCH] kernel/softirq.c: delete 'while' looping to improve a little performance and beautify code

[Thomas Gleixner]

On Sun, 9 Jun 2013, Chen Gang wrote:
> After finish the internal 'while', need not test TASKLET_STATE_SCHED
> again, so looping back to outside 'while' is only for set_bit().
> 
> When use 'if' and set_bit() instead of 'while', it will save at least
> one running conditional instruction, and also will be clearer for readers
> (although the binary size will be a little bigger).

And by doing that you break the atomicity of test_and_set_bit. There
is a good reason why this is an atomic operation and why the code is
written as is.
 
> The related patch is "1da177e Linux-2.6.12-rc2"

How is that patch related to the problem?

Thanks

	tglx

Re: [PATCH] kernel/softirq.c: delete 'while' looping to improve a little performance and beautify code

[Chen Gang]

On 06/10/2013 08:25 PM, Paul E. McKenney wrote:
> On Sun, Jun 09, 2013 at 08:30:19PM +0800, Chen Gang wrote:
>> > 
>> > After finish the internal 'while', need not test TASKLET_STATE_SCHED
>> > again, so looping back to outside 'while' is only for set_bit().
>> > 
>> > When use 'if' and set_bit() instead of 'while', it will save at least
>> > one running conditional instruction, and also will be clearer for readers
>> > (although the binary size will be a little bigger).
>> > 
>> > The related patch is "1da177e Linux-2.6.12-rc2"
>> > 
>> > 
>> > Signed-off-by: Chen Gang <gang.chen@asianux.com>
>> > ---
>> >  kernel/softirq.c |    3 ++-
>> >  1 files changed, 2 insertions(+), 1 deletions(-)
>> > 
>> > diff --git a/kernel/softirq.c b/kernel/softirq.c
>> > index a5f8836..52da25f 100644
>> > --- a/kernel/softirq.c
>> > +++ b/kernel/softirq.c
>> > @@ -540,10 +540,11 @@ void tasklet_kill(struct tasklet_struct *t)
>> >  	if (in_interrupt())
>> >  		printk("Attempt to kill tasklet from interrupt\n");
>> > 
>> > -	while (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) {
>> > +	if (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) {
>> >  		do {
>> >  			yield();
>> >  		} while (test_bit(TASKLET_STATE_SCHED, &t->state));
>> > +		set_bit(TASKLET_STATE_SCHED, &t->state);
> This replaces an atomic test-and-set with two operations, a test and
> a set.  Is this safe?

Oh, it seems not safe, at least it is not the original author's willing.

It is my fault, and also sorry for replying late.


Thanks.
-- 
Chen Gang

Asianux Corporation

Re: [PATCH] kernel/softirq.c: delete 'while' looping to improve a little performance and beautify code

[Chen Gang]

On 06/10/2013 10:15 PM, Thomas Gleixner wrote:
> On Sun, 9 Jun 2013, Chen Gang wrote:
>> > After finish the internal 'while', need not test TASKLET_STATE_SCHED
>> > again, so looping back to outside 'while' is only for set_bit().
>> > 
>> > When use 'if' and set_bit() instead of 'while', it will save at least
>> > one running conditional instruction, and also will be clearer for readers
>> > (although the binary size will be a little bigger).
> And by doing that you break the atomicity of test_and_set_bit. There
> is a good reason why this is an atomic operation and why the code is
> written as is.
>  

OK, thanks. it is my fault (and also sorry for replying late).

>> > The related patch is "1da177e Linux-2.6.12-rc2"
> How is that patch related to the problem?

Because the modification since Linux-2.6.12-rc2.


Thanks.
-- 
Chen Gang

Asianux Corporation