Enhance measuring accuracy

[cgxu519]

In order to enhance measuring accuracy, we use tasklet
instead of timer in the case of detecting softirq off.
Meanwhile, fix a memleak in error path.

[smcdef]

In order to enhance measuring accuracy, we use tasklet
instead of timer in the case of detecting softirq off.
Meanwhile, fix a memleak in error path.

The timer is implemented base on the softirq. The
tasklet is also based on it. So there is no difference
between them. Thanks.

[smcdef]

In order to enhance measuring accuracy, we use tasklet
instead of timer in the case of detecting softirq off.
Meanwhile, fix a memleak in error path.

The timer is implemented base on the softirq. The
tasklet is also based on it. So there is no difference
between them. Thanks.

It looks like that we can make the code simply. Right?

[cgxu519]

In order to enhance measuring accuracy, we use tasklet
instead of timer in the case of detecting softirq off.
Meanwhile, fix a memleak in error path.

The timer is implemented base on the softirq. The
tasklet is also based on it. So there is no difference
between them. Thanks.

Yes, two approaches are very similar. However, I think timer depends HZ
and hard to measure the time period under 1ms in practical environment.

[cgxu519]

In order to enhance measuring accuracy, we use tasklet
instead of timer in the case of detecting softirq off.
Meanwhile, fix a memleak in error path.

The timer is implemented base on the softirq. The
tasklet is also based on it. So there is no difference
between them. Thanks.

It looks like that we can make the code simply. Right?

hmm, maybe. The only concern is that tasklet approach may sightly
change the behaviour of softirq handling so sometimes it mitigates the
softirq latency especially when softirq is handled by ksoftirqd.

[smcdef]

It seems no memleak. What I missed?

[cgxu519]

It seems no memleak. What I missed?

It seems to forget releasing percpu val in error case, Am I missing something?

[smcdef]

It seems no memleak. What I missed?

It seems to forget releasing percpu val in error case, Am I missing something?

Great. I see now. Thanks. Can you split those 2 commits to 2 mr? So that I can
merge this one firstly.

[cgxu519]

It seems no memleak. What I missed?

It seems to forget releasing percpu val in error case, Am I missing something?

Great. I see now. Thanks. Can you split those 2 commits to 2 mr? So that I can
merge this one firstly.

Sure, will update shortly.

[cgxu519]

It seems no memleak. What I missed?

It seems to forget releasing percpu val in error case, Am I missing something?

Great. I see now. Thanks. Can you split those 2 commits to 2 mr? So that I can
merge this one firstly.

I've made new PR for memory leak, I'll make another PR after that get merged.

[cgxu519]

In order to enhance measuring accuracy, we use tasklet
instead of timer in the case of detecting softirq off.
Meanwhile, fix a memleak in error path.

The timer is implemented base on the softirq. The
tasklet is also based on it. So there is no difference
between them. Thanks.

It looks like that we can make the code simply. Right?

So is there a plan to apply this change in the future?